]>
Commit | Line | Data |
---|---|---|
20c8ccb1 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
043405e1 CO |
2 | /* |
3 | * Kernel-based Virtual Machine driver for Linux | |
4 | * | |
5 | * derived from drivers/kvm/kvm_main.c | |
6 | * | |
7 | * Copyright (C) 2006 Qumranet, Inc. | |
4d5c5d0f BAY |
8 | * Copyright (C) 2008 Qumranet, Inc. |
9 | * Copyright IBM Corporation, 2008 | |
9611c187 | 10 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
043405e1 CO |
11 | * |
12 | * Authors: | |
13 | * Avi Kivity <avi@qumranet.com> | |
14 | * Yaniv Kamay <yaniv@qumranet.com> | |
4d5c5d0f BAY |
15 | * Amit Shah <amit.shah@qumranet.com> |
16 | * Ben-Ami Yassour <benami@il.ibm.com> | |
043405e1 CO |
17 | */ |
18 | ||
edf88417 | 19 | #include <linux/kvm_host.h> |
313a3dc7 | 20 | #include "irq.h" |
88197e6a | 21 | #include "ioapic.h" |
1d737c8a | 22 | #include "mmu.h" |
7837699f | 23 | #include "i8254.h" |
37817f29 | 24 | #include "tss.h" |
5fdbf976 | 25 | #include "kvm_cache_regs.h" |
2f728d66 | 26 | #include "kvm_emulate.h" |
26eef70c | 27 | #include "x86.h" |
00b27a3e | 28 | #include "cpuid.h" |
474a5bb9 | 29 | #include "pmu.h" |
e83d5887 | 30 | #include "hyperv.h" |
8df14af4 | 31 | #include "lapic.h" |
23200b7a | 32 | #include "xen.h" |
313a3dc7 | 33 | |
18068523 | 34 | #include <linux/clocksource.h> |
4d5c5d0f | 35 | #include <linux/interrupt.h> |
313a3dc7 CO |
36 | #include <linux/kvm.h> |
37 | #include <linux/fs.h> | |
38 | #include <linux/vmalloc.h> | |
1767e931 PG |
39 | #include <linux/export.h> |
40 | #include <linux/moduleparam.h> | |
0de10343 | 41 | #include <linux/mman.h> |
2bacc55c | 42 | #include <linux/highmem.h> |
19de40a8 | 43 | #include <linux/iommu.h> |
62c476c7 | 44 | #include <linux/intel-iommu.h> |
c8076604 | 45 | #include <linux/cpufreq.h> |
18863bdd | 46 | #include <linux/user-return-notifier.h> |
a983fb23 | 47 | #include <linux/srcu.h> |
5a0e3ad6 | 48 | #include <linux/slab.h> |
ff9d07a0 | 49 | #include <linux/perf_event.h> |
7bee342a | 50 | #include <linux/uaccess.h> |
af585b92 | 51 | #include <linux/hash.h> |
a1b60c1c | 52 | #include <linux/pci.h> |
16e8d74d MT |
53 | #include <linux/timekeeper_internal.h> |
54 | #include <linux/pvclock_gtod.h> | |
87276880 FW |
55 | #include <linux/kvm_irqfd.h> |
56 | #include <linux/irqbypass.h> | |
3905f9ad | 57 | #include <linux/sched/stat.h> |
0c5f81da | 58 | #include <linux/sched/isolation.h> |
d0ec49d4 | 59 | #include <linux/mem_encrypt.h> |
72c3c0fe | 60 | #include <linux/entry-kvm.h> |
7d62874f | 61 | #include <linux/suspend.h> |
3905f9ad | 62 | |
aec51dc4 | 63 | #include <trace/events/kvm.h> |
2ed152af | 64 | |
24f1e32c | 65 | #include <asm/debugreg.h> |
d825ed0a | 66 | #include <asm/msr.h> |
a5f61300 | 67 | #include <asm/desc.h> |
890ca9ae | 68 | #include <asm/mce.h> |
f89e32e0 | 69 | #include <linux/kernel_stat.h> |
78f7f1e5 | 70 | #include <asm/fpu/internal.h> /* Ugh! */ |
1d5f066e | 71 | #include <asm/pvclock.h> |
217fc9cf | 72 | #include <asm/div64.h> |
efc64404 | 73 | #include <asm/irq_remapping.h> |
b0c39dc6 | 74 | #include <asm/mshyperv.h> |
0092e434 | 75 | #include <asm/hypervisor.h> |
9715092f | 76 | #include <asm/tlbflush.h> |
bf8c55d8 | 77 | #include <asm/intel_pt.h> |
b3dc0695 | 78 | #include <asm/emulate_prefix.h> |
fe7e9488 | 79 | #include <asm/sgx.h> |
dd2cb348 | 80 | #include <clocksource/hyperv_timer.h> |
043405e1 | 81 | |
d1898b73 DH |
82 | #define CREATE_TRACE_POINTS |
83 | #include "trace.h" | |
84 | ||
313a3dc7 | 85 | #define MAX_IO_MSRS 256 |
890ca9ae | 86 | #define KVM_MAX_MCE_BANKS 32 |
c45dcc71 AR |
87 | u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P; |
88 | EXPORT_SYMBOL_GPL(kvm_mce_cap_supported); | |
890ca9ae | 89 | |
0f65dd70 | 90 | #define emul_to_vcpu(ctxt) \ |
c9b8b07c | 91 | ((struct kvm_vcpu *)(ctxt)->vcpu) |
0f65dd70 | 92 | |
50a37eb4 JR |
93 | /* EFER defaults: |
94 | * - enable syscall per default because its emulated by KVM | |
95 | * - enable LME and LMA per default on 64 bit KVM | |
96 | */ | |
97 | #ifdef CONFIG_X86_64 | |
1260edbe LJ |
98 | static |
99 | u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA)); | |
50a37eb4 | 100 | #else |
1260edbe | 101 | static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE); |
50a37eb4 | 102 | #endif |
313a3dc7 | 103 | |
b11306b5 SC |
104 | static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS; |
105 | ||
c519265f RK |
106 | #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \ |
107 | KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK) | |
37131313 | 108 | |
cb142eb7 | 109 | static void update_cr8_intercept(struct kvm_vcpu *vcpu); |
7460fb4a | 110 | static void process_nmi(struct kvm_vcpu *vcpu); |
1f7becf1 | 111 | static void process_smi(struct kvm_vcpu *vcpu); |
ee2cd4b7 | 112 | static void enter_smm(struct kvm_vcpu *vcpu); |
6addfc42 | 113 | static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags); |
01643c51 KH |
114 | static void store_regs(struct kvm_vcpu *vcpu); |
115 | static int sync_regs(struct kvm_vcpu *vcpu); | |
674eea0f | 116 | |
afaf0b2f | 117 | struct kvm_x86_ops kvm_x86_ops __read_mostly; |
5fdbf976 | 118 | EXPORT_SYMBOL_GPL(kvm_x86_ops); |
97896d04 | 119 | |
9af5471b JB |
120 | #define KVM_X86_OP(func) \ |
121 | DEFINE_STATIC_CALL_NULL(kvm_x86_##func, \ | |
122 | *(((struct kvm_x86_ops *)0)->func)); | |
123 | #define KVM_X86_OP_NULL KVM_X86_OP | |
124 | #include <asm/kvm-x86-ops.h> | |
125 | EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits); | |
126 | EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg); | |
127 | EXPORT_STATIC_CALL_GPL(kvm_x86_tlb_flush_current); | |
128 | ||
893590c7 | 129 | static bool __read_mostly ignore_msrs = 0; |
476bc001 | 130 | module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR); |
ed85c068 | 131 | |
d855066f | 132 | bool __read_mostly report_ignored_msrs = true; |
fab0aa3b | 133 | module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR); |
d855066f | 134 | EXPORT_SYMBOL_GPL(report_ignored_msrs); |
fab0aa3b | 135 | |
4c27625b | 136 | unsigned int min_timer_period_us = 200; |
9ed96e87 MT |
137 | module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR); |
138 | ||
630994b3 MT |
139 | static bool __read_mostly kvmclock_periodic_sync = true; |
140 | module_param(kvmclock_periodic_sync, bool, S_IRUGO); | |
141 | ||
893590c7 | 142 | bool __read_mostly kvm_has_tsc_control; |
92a1f12d | 143 | EXPORT_SYMBOL_GPL(kvm_has_tsc_control); |
893590c7 | 144 | u32 __read_mostly kvm_max_guest_tsc_khz; |
92a1f12d | 145 | EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz); |
bc9b961b HZ |
146 | u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits; |
147 | EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits); | |
148 | u64 __read_mostly kvm_max_tsc_scaling_ratio; | |
149 | EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio); | |
64672c95 YJ |
150 | u64 __read_mostly kvm_default_tsc_scaling_ratio; |
151 | EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio); | |
fe6b6bc8 CQ |
152 | bool __read_mostly kvm_has_bus_lock_exit; |
153 | EXPORT_SYMBOL_GPL(kvm_has_bus_lock_exit); | |
92a1f12d | 154 | |
cc578287 | 155 | /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */ |
893590c7 | 156 | static u32 __read_mostly tsc_tolerance_ppm = 250; |
cc578287 ZA |
157 | module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR); |
158 | ||
c3941d9e SC |
159 | /* |
160 | * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables | |
d9f6e12f | 161 | * adaptive tuning starting from default advancement of 1000ns. '0' disables |
c3941d9e | 162 | * advancement entirely. Any other value is used as-is and disables adaptive |
d9f6e12f | 163 | * tuning, i.e. allows privileged userspace to set an exact advancement time. |
c3941d9e SC |
164 | */ |
165 | static int __read_mostly lapic_timer_advance_ns = -1; | |
0e6edceb | 166 | module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR); |
d0659d94 | 167 | |
52004014 FW |
168 | static bool __read_mostly vector_hashing = true; |
169 | module_param(vector_hashing, bool, S_IRUGO); | |
170 | ||
c4ae60e4 LA |
171 | bool __read_mostly enable_vmware_backdoor = false; |
172 | module_param(enable_vmware_backdoor, bool, S_IRUGO); | |
173 | EXPORT_SYMBOL_GPL(enable_vmware_backdoor); | |
174 | ||
6c86eedc WL |
175 | static bool __read_mostly force_emulation_prefix = false; |
176 | module_param(force_emulation_prefix, bool, S_IRUGO); | |
177 | ||
0c5f81da WL |
178 | int __read_mostly pi_inject_timer = -1; |
179 | module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR); | |
180 | ||
7e34fbd0 SC |
181 | /* |
182 | * Restoring the host value for MSRs that are only consumed when running in | |
183 | * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU | |
184 | * returns to userspace, i.e. the kernel can run with the guest's value. | |
185 | */ | |
186 | #define KVM_MAX_NR_USER_RETURN_MSRS 16 | |
18863bdd | 187 | |
7e34fbd0 | 188 | struct kvm_user_return_msrs { |
18863bdd AK |
189 | struct user_return_notifier urn; |
190 | bool registered; | |
7e34fbd0 | 191 | struct kvm_user_return_msr_values { |
2bf78fa7 SY |
192 | u64 host; |
193 | u64 curr; | |
7e34fbd0 | 194 | } values[KVM_MAX_NR_USER_RETURN_MSRS]; |
18863bdd AK |
195 | }; |
196 | ||
9cc39a5a SC |
197 | u32 __read_mostly kvm_nr_uret_msrs; |
198 | EXPORT_SYMBOL_GPL(kvm_nr_uret_msrs); | |
199 | static u32 __read_mostly kvm_uret_msrs_list[KVM_MAX_NR_USER_RETURN_MSRS]; | |
7e34fbd0 | 200 | static struct kvm_user_return_msrs __percpu *user_return_msrs; |
18863bdd | 201 | |
cfc48181 SC |
202 | #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \ |
203 | | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \ | |
204 | | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \ | |
205 | | XFEATURE_MASK_PKRU) | |
206 | ||
91661989 SC |
207 | u64 __read_mostly host_efer; |
208 | EXPORT_SYMBOL_GPL(host_efer); | |
209 | ||
b96e6506 | 210 | bool __read_mostly allow_smaller_maxphyaddr = 0; |
3edd6839 MG |
211 | EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr); |
212 | ||
fdf513e3 VK |
213 | bool __read_mostly enable_apicv = true; |
214 | EXPORT_SYMBOL_GPL(enable_apicv); | |
215 | ||
86137773 TL |
216 | u64 __read_mostly host_xss; |
217 | EXPORT_SYMBOL_GPL(host_xss); | |
408e9a31 PB |
218 | u64 __read_mostly supported_xss; |
219 | EXPORT_SYMBOL_GPL(supported_xss); | |
139a12cf | 220 | |
417bc304 | 221 | struct kvm_stats_debugfs_item debugfs_entries[] = { |
812756a8 EGE |
222 | VCPU_STAT("pf_fixed", pf_fixed), |
223 | VCPU_STAT("pf_guest", pf_guest), | |
224 | VCPU_STAT("tlb_flush", tlb_flush), | |
225 | VCPU_STAT("invlpg", invlpg), | |
226 | VCPU_STAT("exits", exits), | |
227 | VCPU_STAT("io_exits", io_exits), | |
228 | VCPU_STAT("mmio_exits", mmio_exits), | |
229 | VCPU_STAT("signal_exits", signal_exits), | |
230 | VCPU_STAT("irq_window", irq_window_exits), | |
231 | VCPU_STAT("nmi_window", nmi_window_exits), | |
232 | VCPU_STAT("halt_exits", halt_exits), | |
233 | VCPU_STAT("halt_successful_poll", halt_successful_poll), | |
234 | VCPU_STAT("halt_attempted_poll", halt_attempted_poll), | |
235 | VCPU_STAT("halt_poll_invalid", halt_poll_invalid), | |
236 | VCPU_STAT("halt_wakeup", halt_wakeup), | |
237 | VCPU_STAT("hypercalls", hypercalls), | |
238 | VCPU_STAT("request_irq", request_irq_exits), | |
239 | VCPU_STAT("irq_exits", irq_exits), | |
240 | VCPU_STAT("host_state_reload", host_state_reload), | |
241 | VCPU_STAT("fpu_reload", fpu_reload), | |
242 | VCPU_STAT("insn_emulation", insn_emulation), | |
243 | VCPU_STAT("insn_emulation_fail", insn_emulation_fail), | |
244 | VCPU_STAT("irq_injections", irq_injections), | |
245 | VCPU_STAT("nmi_injections", nmi_injections), | |
246 | VCPU_STAT("req_event", req_event), | |
247 | VCPU_STAT("l1d_flush", l1d_flush), | |
cb953129 DM |
248 | VCPU_STAT("halt_poll_success_ns", halt_poll_success_ns), |
249 | VCPU_STAT("halt_poll_fail_ns", halt_poll_fail_ns), | |
43c11d91 | 250 | VCPU_STAT("nested_run", nested_run), |
4a7132ef WL |
251 | VCPU_STAT("directed_yield_attempted", directed_yield_attempted), |
252 | VCPU_STAT("directed_yield_successful", directed_yield_successful), | |
812756a8 EGE |
253 | VM_STAT("mmu_shadow_zapped", mmu_shadow_zapped), |
254 | VM_STAT("mmu_pte_write", mmu_pte_write), | |
812756a8 EGE |
255 | VM_STAT("mmu_pde_zapped", mmu_pde_zapped), |
256 | VM_STAT("mmu_flooded", mmu_flooded), | |
257 | VM_STAT("mmu_recycled", mmu_recycled), | |
258 | VM_STAT("mmu_cache_miss", mmu_cache_miss), | |
259 | VM_STAT("mmu_unsync", mmu_unsync), | |
260 | VM_STAT("remote_tlb_flush", remote_tlb_flush), | |
261 | VM_STAT("largepages", lpages, .mode = 0444), | |
262 | VM_STAT("nx_largepages_splitted", nx_lpage_splits, .mode = 0444), | |
263 | VM_STAT("max_mmu_page_hash_collisions", max_mmu_page_hash_collisions), | |
417bc304 HB |
264 | { NULL } |
265 | }; | |
266 | ||
2acf923e | 267 | u64 __read_mostly host_xcr0; |
cfc48181 SC |
268 | u64 __read_mostly supported_xcr0; |
269 | EXPORT_SYMBOL_GPL(supported_xcr0); | |
2acf923e | 270 | |
80fbd280 | 271 | static struct kmem_cache *x86_fpu_cache; |
b666a4b6 | 272 | |
c9b8b07c SC |
273 | static struct kmem_cache *x86_emulator_cache; |
274 | ||
6abe9c13 PX |
275 | /* |
276 | * When called, it means the previous get/set msr reached an invalid msr. | |
cc4cb017 | 277 | * Return true if we want to ignore/silent this failed msr access. |
6abe9c13 | 278 | */ |
d632826f | 279 | static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write) |
6abe9c13 PX |
280 | { |
281 | const char *op = write ? "wrmsr" : "rdmsr"; | |
282 | ||
283 | if (ignore_msrs) { | |
284 | if (report_ignored_msrs) | |
d383b314 TI |
285 | kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n", |
286 | op, msr, data); | |
6abe9c13 | 287 | /* Mask the error */ |
cc4cb017 | 288 | return true; |
6abe9c13 | 289 | } else { |
d383b314 TI |
290 | kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n", |
291 | op, msr, data); | |
cc4cb017 | 292 | return false; |
6abe9c13 PX |
293 | } |
294 | } | |
295 | ||
c9b8b07c SC |
296 | static struct kmem_cache *kvm_alloc_emulator_cache(void) |
297 | { | |
06add254 SC |
298 | unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src); |
299 | unsigned int size = sizeof(struct x86_emulate_ctxt); | |
300 | ||
301 | return kmem_cache_create_usercopy("x86_emulator", size, | |
c9b8b07c | 302 | __alignof__(struct x86_emulate_ctxt), |
06add254 SC |
303 | SLAB_ACCOUNT, useroffset, |
304 | size - useroffset, NULL); | |
c9b8b07c SC |
305 | } |
306 | ||
b6785def | 307 | static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt); |
d6aa1000 | 308 | |
af585b92 GN |
309 | static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu) |
310 | { | |
311 | int i; | |
dd03bcaa | 312 | for (i = 0; i < ASYNC_PF_PER_VCPU; i++) |
af585b92 GN |
313 | vcpu->arch.apf.gfns[i] = ~0; |
314 | } | |
315 | ||
18863bdd AK |
316 | static void kvm_on_user_return(struct user_return_notifier *urn) |
317 | { | |
318 | unsigned slot; | |
7e34fbd0 SC |
319 | struct kvm_user_return_msrs *msrs |
320 | = container_of(urn, struct kvm_user_return_msrs, urn); | |
321 | struct kvm_user_return_msr_values *values; | |
1650b4eb IA |
322 | unsigned long flags; |
323 | ||
324 | /* | |
325 | * Disabling irqs at this point since the following code could be | |
326 | * interrupted and executed through kvm_arch_hardware_disable() | |
327 | */ | |
328 | local_irq_save(flags); | |
7e34fbd0 SC |
329 | if (msrs->registered) { |
330 | msrs->registered = false; | |
1650b4eb IA |
331 | user_return_notifier_unregister(urn); |
332 | } | |
333 | local_irq_restore(flags); | |
9cc39a5a | 334 | for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) { |
7e34fbd0 | 335 | values = &msrs->values[slot]; |
2bf78fa7 | 336 | if (values->host != values->curr) { |
9cc39a5a | 337 | wrmsrl(kvm_uret_msrs_list[slot], values->host); |
2bf78fa7 | 338 | values->curr = values->host; |
18863bdd AK |
339 | } |
340 | } | |
18863bdd AK |
341 | } |
342 | ||
e5fda4bb | 343 | static int kvm_probe_user_return_msr(u32 msr) |
5104d7ff SC |
344 | { |
345 | u64 val; | |
346 | int ret; | |
347 | ||
348 | preempt_disable(); | |
349 | ret = rdmsrl_safe(msr, &val); | |
350 | if (ret) | |
351 | goto out; | |
352 | ret = wrmsrl_safe(msr, val); | |
353 | out: | |
354 | preempt_enable(); | |
355 | return ret; | |
356 | } | |
5104d7ff | 357 | |
e5fda4bb | 358 | int kvm_add_user_return_msr(u32 msr) |
2bf78fa7 | 359 | { |
e5fda4bb SC |
360 | BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS); |
361 | ||
362 | if (kvm_probe_user_return_msr(msr)) | |
363 | return -1; | |
364 | ||
365 | kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr; | |
366 | return kvm_nr_uret_msrs++; | |
18863bdd | 367 | } |
e5fda4bb | 368 | EXPORT_SYMBOL_GPL(kvm_add_user_return_msr); |
18863bdd | 369 | |
8ea8b8d6 SC |
370 | int kvm_find_user_return_msr(u32 msr) |
371 | { | |
372 | int i; | |
373 | ||
9cc39a5a SC |
374 | for (i = 0; i < kvm_nr_uret_msrs; ++i) { |
375 | if (kvm_uret_msrs_list[i] == msr) | |
8ea8b8d6 SC |
376 | return i; |
377 | } | |
378 | return -1; | |
379 | } | |
380 | EXPORT_SYMBOL_GPL(kvm_find_user_return_msr); | |
381 | ||
7e34fbd0 | 382 | static void kvm_user_return_msr_cpu_online(void) |
18863bdd | 383 | { |
05c19c2f | 384 | unsigned int cpu = smp_processor_id(); |
7e34fbd0 | 385 | struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu); |
05c19c2f SC |
386 | u64 value; |
387 | int i; | |
18863bdd | 388 | |
9cc39a5a SC |
389 | for (i = 0; i < kvm_nr_uret_msrs; ++i) { |
390 | rdmsrl_safe(kvm_uret_msrs_list[i], &value); | |
7e34fbd0 SC |
391 | msrs->values[i].host = value; |
392 | msrs->values[i].curr = value; | |
05c19c2f | 393 | } |
18863bdd AK |
394 | } |
395 | ||
7e34fbd0 | 396 | int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask) |
18863bdd | 397 | { |
013f6a5d | 398 | unsigned int cpu = smp_processor_id(); |
7e34fbd0 | 399 | struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu); |
8b3c3104 | 400 | int err; |
18863bdd | 401 | |
7e34fbd0 SC |
402 | value = (value & mask) | (msrs->values[slot].host & ~mask); |
403 | if (value == msrs->values[slot].curr) | |
8b3c3104 | 404 | return 0; |
9cc39a5a | 405 | err = wrmsrl_safe(kvm_uret_msrs_list[slot], value); |
8b3c3104 AH |
406 | if (err) |
407 | return 1; | |
408 | ||
7e34fbd0 SC |
409 | msrs->values[slot].curr = value; |
410 | if (!msrs->registered) { | |
411 | msrs->urn.on_user_return = kvm_on_user_return; | |
412 | user_return_notifier_register(&msrs->urn); | |
413 | msrs->registered = true; | |
18863bdd | 414 | } |
8b3c3104 | 415 | return 0; |
18863bdd | 416 | } |
7e34fbd0 | 417 | EXPORT_SYMBOL_GPL(kvm_set_user_return_msr); |
18863bdd | 418 | |
13a34e06 | 419 | static void drop_user_return_notifiers(void) |
3548bab5 | 420 | { |
013f6a5d | 421 | unsigned int cpu = smp_processor_id(); |
7e34fbd0 | 422 | struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu); |
3548bab5 | 423 | |
7e34fbd0 SC |
424 | if (msrs->registered) |
425 | kvm_on_user_return(&msrs->urn); | |
3548bab5 AK |
426 | } |
427 | ||
6866b83e CO |
428 | u64 kvm_get_apic_base(struct kvm_vcpu *vcpu) |
429 | { | |
8a5a87d9 | 430 | return vcpu->arch.apic_base; |
6866b83e CO |
431 | } |
432 | EXPORT_SYMBOL_GPL(kvm_get_apic_base); | |
433 | ||
58871649 JM |
434 | enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu) |
435 | { | |
436 | return kvm_apic_mode(kvm_get_apic_base(vcpu)); | |
437 | } | |
438 | EXPORT_SYMBOL_GPL(kvm_get_apic_mode); | |
439 | ||
58cb628d JK |
440 | int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
441 | { | |
58871649 JM |
442 | enum lapic_mode old_mode = kvm_get_apic_mode(vcpu); |
443 | enum lapic_mode new_mode = kvm_apic_mode(msr_info->data); | |
a8ac864a | 444 | u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff | |
d6321d49 | 445 | (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE); |
58cb628d | 446 | |
58871649 | 447 | if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID) |
58cb628d | 448 | return 1; |
58871649 JM |
449 | if (!msr_info->host_initiated) { |
450 | if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC) | |
451 | return 1; | |
452 | if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC) | |
453 | return 1; | |
454 | } | |
58cb628d JK |
455 | |
456 | kvm_lapic_set_base(vcpu, msr_info->data); | |
4abaffce | 457 | kvm_recalculate_apic_map(vcpu->kvm); |
58cb628d | 458 | return 0; |
6866b83e CO |
459 | } |
460 | EXPORT_SYMBOL_GPL(kvm_set_apic_base); | |
461 | ||
3ebccdf3 | 462 | asmlinkage __visible noinstr void kvm_spurious_fault(void) |
e3ba45b8 GL |
463 | { |
464 | /* Fault while not rebooting. We want the trace. */ | |
b4fdcf60 | 465 | BUG_ON(!kvm_rebooting); |
e3ba45b8 GL |
466 | } |
467 | EXPORT_SYMBOL_GPL(kvm_spurious_fault); | |
468 | ||
3fd28fce ED |
469 | #define EXCPT_BENIGN 0 |
470 | #define EXCPT_CONTRIBUTORY 1 | |
471 | #define EXCPT_PF 2 | |
472 | ||
473 | static int exception_class(int vector) | |
474 | { | |
475 | switch (vector) { | |
476 | case PF_VECTOR: | |
477 | return EXCPT_PF; | |
478 | case DE_VECTOR: | |
479 | case TS_VECTOR: | |
480 | case NP_VECTOR: | |
481 | case SS_VECTOR: | |
482 | case GP_VECTOR: | |
483 | return EXCPT_CONTRIBUTORY; | |
484 | default: | |
485 | break; | |
486 | } | |
487 | return EXCPT_BENIGN; | |
488 | } | |
489 | ||
d6e8c854 NA |
490 | #define EXCPT_FAULT 0 |
491 | #define EXCPT_TRAP 1 | |
492 | #define EXCPT_ABORT 2 | |
493 | #define EXCPT_INTERRUPT 3 | |
494 | ||
495 | static int exception_type(int vector) | |
496 | { | |
497 | unsigned int mask; | |
498 | ||
499 | if (WARN_ON(vector > 31 || vector == NMI_VECTOR)) | |
500 | return EXCPT_INTERRUPT; | |
501 | ||
502 | mask = 1 << vector; | |
503 | ||
504 | /* #DB is trap, as instruction watchpoints are handled elsewhere */ | |
505 | if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR))) | |
506 | return EXCPT_TRAP; | |
507 | ||
508 | if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR))) | |
509 | return EXCPT_ABORT; | |
510 | ||
511 | /* Reserved exceptions will result in fault */ | |
512 | return EXCPT_FAULT; | |
513 | } | |
514 | ||
da998b46 JM |
515 | void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu) |
516 | { | |
517 | unsigned nr = vcpu->arch.exception.nr; | |
518 | bool has_payload = vcpu->arch.exception.has_payload; | |
519 | unsigned long payload = vcpu->arch.exception.payload; | |
520 | ||
521 | if (!has_payload) | |
522 | return; | |
523 | ||
524 | switch (nr) { | |
f10c729f JM |
525 | case DB_VECTOR: |
526 | /* | |
527 | * "Certain debug exceptions may clear bit 0-3. The | |
528 | * remaining contents of the DR6 register are never | |
529 | * cleared by the processor". | |
530 | */ | |
531 | vcpu->arch.dr6 &= ~DR_TRAP_BITS; | |
532 | /* | |
9a3ecd5e CQ |
533 | * In order to reflect the #DB exception payload in guest |
534 | * dr6, three components need to be considered: active low | |
535 | * bit, FIXED_1 bits and active high bits (e.g. DR6_BD, | |
536 | * DR6_BS and DR6_BT) | |
537 | * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits. | |
538 | * In the target guest dr6: | |
539 | * FIXED_1 bits should always be set. | |
540 | * Active low bits should be cleared if 1-setting in payload. | |
541 | * Active high bits should be set if 1-setting in payload. | |
542 | * | |
543 | * Note, the payload is compatible with the pending debug | |
544 | * exceptions/exit qualification under VMX, that active_low bits | |
545 | * are active high in payload. | |
546 | * So they need to be flipped for DR6. | |
f10c729f | 547 | */ |
9a3ecd5e | 548 | vcpu->arch.dr6 |= DR6_ACTIVE_LOW; |
f10c729f | 549 | vcpu->arch.dr6 |= payload; |
9a3ecd5e | 550 | vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW; |
307f1cfa OU |
551 | |
552 | /* | |
553 | * The #DB payload is defined as compatible with the 'pending | |
554 | * debug exceptions' field under VMX, not DR6. While bit 12 is | |
555 | * defined in the 'pending debug exceptions' field (enabled | |
556 | * breakpoint), it is reserved and must be zero in DR6. | |
557 | */ | |
558 | vcpu->arch.dr6 &= ~BIT(12); | |
f10c729f | 559 | break; |
da998b46 JM |
560 | case PF_VECTOR: |
561 | vcpu->arch.cr2 = payload; | |
562 | break; | |
563 | } | |
564 | ||
565 | vcpu->arch.exception.has_payload = false; | |
566 | vcpu->arch.exception.payload = 0; | |
567 | } | |
568 | EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload); | |
569 | ||
3fd28fce | 570 | static void kvm_multiple_exception(struct kvm_vcpu *vcpu, |
ce7ddec4 | 571 | unsigned nr, bool has_error, u32 error_code, |
91e86d22 | 572 | bool has_payload, unsigned long payload, bool reinject) |
3fd28fce ED |
573 | { |
574 | u32 prev_nr; | |
575 | int class1, class2; | |
576 | ||
3842d135 AK |
577 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
578 | ||
664f8e26 | 579 | if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) { |
3fd28fce | 580 | queue: |
664f8e26 WL |
581 | if (reinject) { |
582 | /* | |
583 | * On vmentry, vcpu->arch.exception.pending is only | |
584 | * true if an event injection was blocked by | |
585 | * nested_run_pending. In that case, however, | |
586 | * vcpu_enter_guest requests an immediate exit, | |
587 | * and the guest shouldn't proceed far enough to | |
588 | * need reinjection. | |
589 | */ | |
590 | WARN_ON_ONCE(vcpu->arch.exception.pending); | |
591 | vcpu->arch.exception.injected = true; | |
91e86d22 JM |
592 | if (WARN_ON_ONCE(has_payload)) { |
593 | /* | |
594 | * A reinjected event has already | |
595 | * delivered its payload. | |
596 | */ | |
597 | has_payload = false; | |
598 | payload = 0; | |
599 | } | |
664f8e26 WL |
600 | } else { |
601 | vcpu->arch.exception.pending = true; | |
602 | vcpu->arch.exception.injected = false; | |
603 | } | |
3fd28fce ED |
604 | vcpu->arch.exception.has_error_code = has_error; |
605 | vcpu->arch.exception.nr = nr; | |
606 | vcpu->arch.exception.error_code = error_code; | |
91e86d22 JM |
607 | vcpu->arch.exception.has_payload = has_payload; |
608 | vcpu->arch.exception.payload = payload; | |
a06230b6 | 609 | if (!is_guest_mode(vcpu)) |
da998b46 | 610 | kvm_deliver_exception_payload(vcpu); |
3fd28fce ED |
611 | return; |
612 | } | |
613 | ||
614 | /* to check exception */ | |
615 | prev_nr = vcpu->arch.exception.nr; | |
616 | if (prev_nr == DF_VECTOR) { | |
617 | /* triple fault -> shutdown */ | |
a8eeb04a | 618 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); |
3fd28fce ED |
619 | return; |
620 | } | |
621 | class1 = exception_class(prev_nr); | |
622 | class2 = exception_class(nr); | |
623 | if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY) | |
624 | || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) { | |
664f8e26 WL |
625 | /* |
626 | * Generate double fault per SDM Table 5-5. Set | |
627 | * exception.pending = true so that the double fault | |
628 | * can trigger a nested vmexit. | |
629 | */ | |
3fd28fce | 630 | vcpu->arch.exception.pending = true; |
664f8e26 | 631 | vcpu->arch.exception.injected = false; |
3fd28fce ED |
632 | vcpu->arch.exception.has_error_code = true; |
633 | vcpu->arch.exception.nr = DF_VECTOR; | |
634 | vcpu->arch.exception.error_code = 0; | |
c851436a JM |
635 | vcpu->arch.exception.has_payload = false; |
636 | vcpu->arch.exception.payload = 0; | |
3fd28fce ED |
637 | } else |
638 | /* replace previous exception with a new one in a hope | |
639 | that instruction re-execution will regenerate lost | |
640 | exception */ | |
641 | goto queue; | |
642 | } | |
643 | ||
298101da AK |
644 | void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr) |
645 | { | |
91e86d22 | 646 | kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false); |
298101da AK |
647 | } |
648 | EXPORT_SYMBOL_GPL(kvm_queue_exception); | |
649 | ||
ce7ddec4 JR |
650 | void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr) |
651 | { | |
91e86d22 | 652 | kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true); |
ce7ddec4 JR |
653 | } |
654 | EXPORT_SYMBOL_GPL(kvm_requeue_exception); | |
655 | ||
4d5523cf PB |
656 | void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr, |
657 | unsigned long payload) | |
f10c729f JM |
658 | { |
659 | kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false); | |
660 | } | |
4d5523cf | 661 | EXPORT_SYMBOL_GPL(kvm_queue_exception_p); |
f10c729f | 662 | |
da998b46 JM |
663 | static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr, |
664 | u32 error_code, unsigned long payload) | |
665 | { | |
666 | kvm_multiple_exception(vcpu, nr, true, error_code, | |
667 | true, payload, false); | |
668 | } | |
669 | ||
6affcbed | 670 | int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err) |
c3c91fee | 671 | { |
db8fcefa AP |
672 | if (err) |
673 | kvm_inject_gp(vcpu, 0); | |
674 | else | |
6affcbed KH |
675 | return kvm_skip_emulated_instruction(vcpu); |
676 | ||
677 | return 1; | |
db8fcefa AP |
678 | } |
679 | EXPORT_SYMBOL_GPL(kvm_complete_insn_gp); | |
8df25a32 | 680 | |
6389ee94 | 681 | void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault) |
c3c91fee AK |
682 | { |
683 | ++vcpu->stat.pf_guest; | |
adfe20fb WL |
684 | vcpu->arch.exception.nested_apf = |
685 | is_guest_mode(vcpu) && fault->async_page_fault; | |
da998b46 | 686 | if (vcpu->arch.exception.nested_apf) { |
adfe20fb | 687 | vcpu->arch.apf.nested_apf_token = fault->address; |
da998b46 JM |
688 | kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code); |
689 | } else { | |
690 | kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code, | |
691 | fault->address); | |
692 | } | |
c3c91fee | 693 | } |
27d6c865 | 694 | EXPORT_SYMBOL_GPL(kvm_inject_page_fault); |
c3c91fee | 695 | |
53b3d8e9 SC |
696 | bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu, |
697 | struct x86_exception *fault) | |
d4f8cf66 | 698 | { |
0cd665bd | 699 | struct kvm_mmu *fault_mmu; |
53b3d8e9 SC |
700 | WARN_ON_ONCE(fault->vector != PF_VECTOR); |
701 | ||
0cd665bd PB |
702 | fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu : |
703 | vcpu->arch.walk_mmu; | |
ef54bcfe | 704 | |
ee1fa209 JS |
705 | /* |
706 | * Invalidate the TLB entry for the faulting address, if it exists, | |
707 | * else the access will fault indefinitely (and to emulate hardware). | |
708 | */ | |
709 | if ((fault->error_code & PFERR_PRESENT_MASK) && | |
710 | !(fault->error_code & PFERR_RSVD_MASK)) | |
711 | kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address, | |
712 | fault_mmu->root_hpa); | |
713 | ||
714 | fault_mmu->inject_page_fault(vcpu, fault); | |
ef54bcfe | 715 | return fault->nested_page_fault; |
d4f8cf66 | 716 | } |
53b3d8e9 | 717 | EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault); |
d4f8cf66 | 718 | |
3419ffc8 SY |
719 | void kvm_inject_nmi(struct kvm_vcpu *vcpu) |
720 | { | |
7460fb4a AK |
721 | atomic_inc(&vcpu->arch.nmi_queued); |
722 | kvm_make_request(KVM_REQ_NMI, vcpu); | |
3419ffc8 SY |
723 | } |
724 | EXPORT_SYMBOL_GPL(kvm_inject_nmi); | |
725 | ||
298101da AK |
726 | void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code) |
727 | { | |
91e86d22 | 728 | kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false); |
298101da AK |
729 | } |
730 | EXPORT_SYMBOL_GPL(kvm_queue_exception_e); | |
731 | ||
ce7ddec4 JR |
732 | void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code) |
733 | { | |
91e86d22 | 734 | kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true); |
ce7ddec4 JR |
735 | } |
736 | EXPORT_SYMBOL_GPL(kvm_requeue_exception_e); | |
737 | ||
0a79b009 AK |
738 | /* |
739 | * Checks if cpl <= required_cpl; if true, return true. Otherwise queue | |
740 | * a #GP and return false. | |
741 | */ | |
742 | bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl) | |
298101da | 743 | { |
b3646477 | 744 | if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl) |
0a79b009 AK |
745 | return true; |
746 | kvm_queue_exception_e(vcpu, GP_VECTOR, 0); | |
747 | return false; | |
298101da | 748 | } |
0a79b009 | 749 | EXPORT_SYMBOL_GPL(kvm_require_cpl); |
298101da | 750 | |
16f8a6f9 NA |
751 | bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr) |
752 | { | |
753 | if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE)) | |
754 | return true; | |
755 | ||
756 | kvm_queue_exception(vcpu, UD_VECTOR); | |
757 | return false; | |
758 | } | |
759 | EXPORT_SYMBOL_GPL(kvm_require_dr); | |
760 | ||
ec92fe44 JR |
761 | /* |
762 | * This function will be used to read from the physical memory of the currently | |
54bf36aa | 763 | * running guest. The difference to kvm_vcpu_read_guest_page is that this function |
ec92fe44 JR |
764 | * can read from guest physical or from the guest's guest physical memory. |
765 | */ | |
766 | int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, | |
767 | gfn_t ngfn, void *data, int offset, int len, | |
768 | u32 access) | |
769 | { | |
54987b7a | 770 | struct x86_exception exception; |
ec92fe44 JR |
771 | gfn_t real_gfn; |
772 | gpa_t ngpa; | |
773 | ||
774 | ngpa = gfn_to_gpa(ngfn); | |
54987b7a | 775 | real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception); |
ec92fe44 JR |
776 | if (real_gfn == UNMAPPED_GVA) |
777 | return -EFAULT; | |
778 | ||
779 | real_gfn = gpa_to_gfn(real_gfn); | |
780 | ||
54bf36aa | 781 | return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len); |
ec92fe44 JR |
782 | } |
783 | EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu); | |
784 | ||
69b0049a | 785 | static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
3d06b8bf JR |
786 | void *data, int offset, int len, u32 access) |
787 | { | |
788 | return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn, | |
789 | data, offset, len, access); | |
790 | } | |
791 | ||
16cfacc8 SC |
792 | static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu) |
793 | { | |
5b7f575c | 794 | return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2); |
16cfacc8 SC |
795 | } |
796 | ||
a03490ed | 797 | /* |
16cfacc8 | 798 | * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise. |
a03490ed | 799 | */ |
ff03a073 | 800 | int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3) |
a03490ed CO |
801 | { |
802 | gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT; | |
803 | unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2; | |
804 | int i; | |
805 | int ret; | |
ff03a073 | 806 | u64 pdpte[ARRAY_SIZE(mmu->pdptrs)]; |
a03490ed | 807 | |
ff03a073 JR |
808 | ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte, |
809 | offset * sizeof(u64), sizeof(pdpte), | |
810 | PFERR_USER_MASK|PFERR_WRITE_MASK); | |
a03490ed CO |
811 | if (ret < 0) { |
812 | ret = 0; | |
813 | goto out; | |
814 | } | |
815 | for (i = 0; i < ARRAY_SIZE(pdpte); ++i) { | |
812f30b2 | 816 | if ((pdpte[i] & PT_PRESENT_MASK) && |
16cfacc8 | 817 | (pdpte[i] & pdptr_rsvd_bits(vcpu))) { |
a03490ed CO |
818 | ret = 0; |
819 | goto out; | |
820 | } | |
821 | } | |
822 | ret = 1; | |
823 | ||
ff03a073 | 824 | memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs)); |
cb3c1e2f SC |
825 | kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR); |
826 | ||
a03490ed | 827 | out: |
a03490ed CO |
828 | |
829 | return ret; | |
830 | } | |
cc4b6871 | 831 | EXPORT_SYMBOL_GPL(load_pdptrs); |
a03490ed | 832 | |
9ed38ffa | 833 | bool pdptrs_changed(struct kvm_vcpu *vcpu) |
d835dfec | 834 | { |
ff03a073 | 835 | u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)]; |
3d06b8bf JR |
836 | int offset; |
837 | gfn_t gfn; | |
d835dfec AK |
838 | int r; |
839 | ||
bf03d4f9 | 840 | if (!is_pae_paging(vcpu)) |
d835dfec AK |
841 | return false; |
842 | ||
cb3c1e2f | 843 | if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR)) |
6de4f3ad AK |
844 | return true; |
845 | ||
a512177e PB |
846 | gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT; |
847 | offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1); | |
3d06b8bf JR |
848 | r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte), |
849 | PFERR_USER_MASK | PFERR_WRITE_MASK); | |
d835dfec | 850 | if (r < 0) |
7f7f0d9c | 851 | return true; |
d835dfec | 852 | |
7f7f0d9c | 853 | return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0; |
d835dfec | 854 | } |
9ed38ffa | 855 | EXPORT_SYMBOL_GPL(pdptrs_changed); |
d835dfec | 856 | |
f27ad38a TL |
857 | void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0) |
858 | { | |
859 | unsigned long update_bits = X86_CR0_PG | X86_CR0_WP; | |
860 | ||
861 | if ((cr0 ^ old_cr0) & X86_CR0_PG) { | |
862 | kvm_clear_async_pf_completion_queue(vcpu); | |
863 | kvm_async_pf_hash_reset(vcpu); | |
864 | } | |
865 | ||
866 | if ((cr0 ^ old_cr0) & update_bits) | |
867 | kvm_mmu_reset_context(vcpu); | |
868 | ||
869 | if (((cr0 ^ old_cr0) & X86_CR0_CD) && | |
870 | kvm_arch_has_noncoherent_dma(vcpu->kvm) && | |
871 | !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED)) | |
872 | kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL); | |
873 | } | |
874 | EXPORT_SYMBOL_GPL(kvm_post_set_cr0); | |
875 | ||
49a9b07e | 876 | int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0) |
a03490ed | 877 | { |
aad82703 | 878 | unsigned long old_cr0 = kvm_read_cr0(vcpu); |
d42e3fae | 879 | unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG; |
aad82703 | 880 | |
f9a48e6a AK |
881 | cr0 |= X86_CR0_ET; |
882 | ||
ab344828 | 883 | #ifdef CONFIG_X86_64 |
0f12244f GN |
884 | if (cr0 & 0xffffffff00000000UL) |
885 | return 1; | |
ab344828 GN |
886 | #endif |
887 | ||
888 | cr0 &= ~CR0_RESERVED_BITS; | |
a03490ed | 889 | |
0f12244f GN |
890 | if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) |
891 | return 1; | |
a03490ed | 892 | |
0f12244f GN |
893 | if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) |
894 | return 1; | |
a03490ed | 895 | |
a03490ed | 896 | #ifdef CONFIG_X86_64 |
05487215 SC |
897 | if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) && |
898 | (cr0 & X86_CR0_PG)) { | |
899 | int cs_db, cs_l; | |
900 | ||
901 | if (!is_pae(vcpu)) | |
902 | return 1; | |
b3646477 | 903 | static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l); |
05487215 | 904 | if (cs_l) |
0f12244f | 905 | return 1; |
a03490ed | 906 | } |
05487215 SC |
907 | #endif |
908 | if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) && | |
909 | is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) && | |
910 | !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu))) | |
911 | return 1; | |
a03490ed | 912 | |
ad756a16 MJ |
913 | if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)) |
914 | return 1; | |
915 | ||
b3646477 | 916 | static_call(kvm_x86_set_cr0)(vcpu, cr0); |
a03490ed | 917 | |
f27ad38a | 918 | kvm_post_set_cr0(vcpu, old_cr0, cr0); |
b18d5431 | 919 | |
0f12244f GN |
920 | return 0; |
921 | } | |
2d3ad1f4 | 922 | EXPORT_SYMBOL_GPL(kvm_set_cr0); |
a03490ed | 923 | |
2d3ad1f4 | 924 | void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw) |
a03490ed | 925 | { |
49a9b07e | 926 | (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f)); |
a03490ed | 927 | } |
2d3ad1f4 | 928 | EXPORT_SYMBOL_GPL(kvm_lmsw); |
a03490ed | 929 | |
139a12cf | 930 | void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu) |
42bdf991 | 931 | { |
16809ecd TL |
932 | if (vcpu->arch.guest_state_protected) |
933 | return; | |
934 | ||
139a12cf AL |
935 | if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) { |
936 | ||
937 | if (vcpu->arch.xcr0 != host_xcr0) | |
938 | xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0); | |
939 | ||
940 | if (vcpu->arch.xsaves_enabled && | |
941 | vcpu->arch.ia32_xss != host_xss) | |
942 | wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss); | |
943 | } | |
37486135 BM |
944 | |
945 | if (static_cpu_has(X86_FEATURE_PKU) && | |
946 | (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) || | |
947 | (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) && | |
948 | vcpu->arch.pkru != vcpu->arch.host_pkru) | |
949 | __write_pkru(vcpu->arch.pkru); | |
42bdf991 | 950 | } |
139a12cf | 951 | EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state); |
42bdf991 | 952 | |
139a12cf | 953 | void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu) |
42bdf991 | 954 | { |
16809ecd TL |
955 | if (vcpu->arch.guest_state_protected) |
956 | return; | |
957 | ||
37486135 BM |
958 | if (static_cpu_has(X86_FEATURE_PKU) && |
959 | (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) || | |
960 | (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) { | |
961 | vcpu->arch.pkru = rdpkru(); | |
962 | if (vcpu->arch.pkru != vcpu->arch.host_pkru) | |
963 | __write_pkru(vcpu->arch.host_pkru); | |
964 | } | |
965 | ||
139a12cf AL |
966 | if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) { |
967 | ||
968 | if (vcpu->arch.xcr0 != host_xcr0) | |
969 | xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0); | |
970 | ||
971 | if (vcpu->arch.xsaves_enabled && | |
972 | vcpu->arch.ia32_xss != host_xss) | |
973 | wrmsrl(MSR_IA32_XSS, host_xss); | |
974 | } | |
975 | ||
42bdf991 | 976 | } |
139a12cf | 977 | EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state); |
42bdf991 | 978 | |
69b0049a | 979 | static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr) |
2acf923e | 980 | { |
56c103ec LJ |
981 | u64 xcr0 = xcr; |
982 | u64 old_xcr0 = vcpu->arch.xcr0; | |
46c34cb0 | 983 | u64 valid_bits; |
2acf923e DC |
984 | |
985 | /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */ | |
986 | if (index != XCR_XFEATURE_ENABLED_MASK) | |
987 | return 1; | |
d91cab78 | 988 | if (!(xcr0 & XFEATURE_MASK_FP)) |
2acf923e | 989 | return 1; |
d91cab78 | 990 | if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE)) |
2acf923e | 991 | return 1; |
46c34cb0 PB |
992 | |
993 | /* | |
994 | * Do not allow the guest to set bits that we do not support | |
995 | * saving. However, xcr0 bit 0 is always set, even if the | |
996 | * emulated CPU does not support XSAVE (see fx_init). | |
997 | */ | |
d91cab78 | 998 | valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP; |
46c34cb0 | 999 | if (xcr0 & ~valid_bits) |
2acf923e | 1000 | return 1; |
46c34cb0 | 1001 | |
d91cab78 DH |
1002 | if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) != |
1003 | (!(xcr0 & XFEATURE_MASK_BNDCSR))) | |
390bd528 LJ |
1004 | return 1; |
1005 | ||
d91cab78 DH |
1006 | if (xcr0 & XFEATURE_MASK_AVX512) { |
1007 | if (!(xcr0 & XFEATURE_MASK_YMM)) | |
612263b3 | 1008 | return 1; |
d91cab78 | 1009 | if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512) |
612263b3 CP |
1010 | return 1; |
1011 | } | |
2acf923e | 1012 | vcpu->arch.xcr0 = xcr0; |
56c103ec | 1013 | |
d91cab78 | 1014 | if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND) |
aedbaf4f | 1015 | kvm_update_cpuid_runtime(vcpu); |
2acf923e DC |
1016 | return 0; |
1017 | } | |
1018 | ||
92f9895c | 1019 | int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu) |
2acf923e | 1020 | { |
92f9895c SC |
1021 | if (static_call(kvm_x86_get_cpl)(vcpu) != 0 || |
1022 | __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) { | |
1023 | kvm_inject_gp(vcpu, 0); | |
1024 | return 1; | |
1025 | } | |
bbefd4fc | 1026 | |
92f9895c | 1027 | return kvm_skip_emulated_instruction(vcpu); |
2acf923e | 1028 | } |
92f9895c | 1029 | EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv); |
2acf923e | 1030 | |
ee69c92b | 1031 | bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4) |
a03490ed | 1032 | { |
b11306b5 | 1033 | if (cr4 & cr4_reserved_bits) |
ee69c92b | 1034 | return false; |
b9baba86 | 1035 | |
b899c132 | 1036 | if (cr4 & vcpu->arch.cr4_guest_rsvd_bits) |
ee69c92b | 1037 | return false; |
3ca94192 | 1038 | |
b3646477 | 1039 | return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4); |
3ca94192 | 1040 | } |
ee69c92b | 1041 | EXPORT_SYMBOL_GPL(kvm_is_valid_cr4); |
3ca94192 | 1042 | |
5b51cb13 TL |
1043 | void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4) |
1044 | { | |
1045 | unsigned long mmu_role_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE | | |
1046 | X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE; | |
1047 | ||
1048 | if (((cr4 ^ old_cr4) & mmu_role_bits) || | |
1049 | (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE))) | |
1050 | kvm_mmu_reset_context(vcpu); | |
3ca94192 | 1051 | } |
5b51cb13 | 1052 | EXPORT_SYMBOL_GPL(kvm_post_set_cr4); |
3ca94192 WL |
1053 | |
1054 | int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4) | |
1055 | { | |
1056 | unsigned long old_cr4 = kvm_read_cr4(vcpu); | |
1057 | unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE | | |
cb957adb | 1058 | X86_CR4_SMEP; |
3ca94192 | 1059 | |
ee69c92b | 1060 | if (!kvm_is_valid_cr4(vcpu, cr4)) |
ae3e61e1 PB |
1061 | return 1; |
1062 | ||
a03490ed | 1063 | if (is_long_mode(vcpu)) { |
0f12244f GN |
1064 | if (!(cr4 & X86_CR4_PAE)) |
1065 | return 1; | |
d74fcfc1 SC |
1066 | if ((cr4 ^ old_cr4) & X86_CR4_LA57) |
1067 | return 1; | |
a2edf57f AK |
1068 | } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE) |
1069 | && ((cr4 ^ old_cr4) & pdptr_bits) | |
9f8fe504 AK |
1070 | && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, |
1071 | kvm_read_cr3(vcpu))) | |
0f12244f GN |
1072 | return 1; |
1073 | ||
ad756a16 | 1074 | if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) { |
d6321d49 | 1075 | if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID)) |
ad756a16 MJ |
1076 | return 1; |
1077 | ||
1078 | /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */ | |
1079 | if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu)) | |
1080 | return 1; | |
1081 | } | |
1082 | ||
b3646477 | 1083 | static_call(kvm_x86_set_cr4)(vcpu, cr4); |
a03490ed | 1084 | |
5b51cb13 | 1085 | kvm_post_set_cr4(vcpu, old_cr4, cr4); |
2acf923e | 1086 | |
0f12244f GN |
1087 | return 0; |
1088 | } | |
2d3ad1f4 | 1089 | EXPORT_SYMBOL_GPL(kvm_set_cr4); |
a03490ed | 1090 | |
2390218b | 1091 | int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3) |
a03490ed | 1092 | { |
ade61e28 | 1093 | bool skip_tlb_flush = false; |
ac146235 | 1094 | #ifdef CONFIG_X86_64 |
c19986fe JS |
1095 | bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE); |
1096 | ||
ade61e28 | 1097 | if (pcid_enabled) { |
208320ba JS |
1098 | skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH; |
1099 | cr3 &= ~X86_CR3_PCID_NOFLUSH; | |
ade61e28 | 1100 | } |
ac146235 | 1101 | #endif |
9d88fca7 | 1102 | |
9f8fe504 | 1103 | if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) { |
956bf353 JS |
1104 | if (!skip_tlb_flush) { |
1105 | kvm_mmu_sync_roots(vcpu); | |
eeeb4f67 | 1106 | kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu); |
956bf353 | 1107 | } |
0f12244f | 1108 | return 0; |
d835dfec AK |
1109 | } |
1110 | ||
886bbcc7 SC |
1111 | /* |
1112 | * Do not condition the GPA check on long mode, this helper is used to | |
1113 | * stuff CR3, e.g. for RSM emulation, and there is no guarantee that | |
1114 | * the current vCPU mode is accurate. | |
1115 | */ | |
1116 | if (kvm_vcpu_is_illegal_gpa(vcpu, cr3)) | |
d1cd3ce9 | 1117 | return 1; |
886bbcc7 SC |
1118 | |
1119 | if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) | |
346874c9 | 1120 | return 1; |
a03490ed | 1121 | |
be01e8e2 | 1122 | kvm_mmu_new_pgd(vcpu, cr3, skip_tlb_flush, skip_tlb_flush); |
0f12244f | 1123 | vcpu->arch.cr3 = cr3; |
cb3c1e2f | 1124 | kvm_register_mark_available(vcpu, VCPU_EXREG_CR3); |
7c390d35 | 1125 | |
0f12244f GN |
1126 | return 0; |
1127 | } | |
2d3ad1f4 | 1128 | EXPORT_SYMBOL_GPL(kvm_set_cr3); |
a03490ed | 1129 | |
eea1cff9 | 1130 | int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8) |
a03490ed | 1131 | { |
0f12244f GN |
1132 | if (cr8 & CR8_RESERVED_BITS) |
1133 | return 1; | |
35754c98 | 1134 | if (lapic_in_kernel(vcpu)) |
a03490ed CO |
1135 | kvm_lapic_set_tpr(vcpu, cr8); |
1136 | else | |
ad312c7c | 1137 | vcpu->arch.cr8 = cr8; |
0f12244f GN |
1138 | return 0; |
1139 | } | |
2d3ad1f4 | 1140 | EXPORT_SYMBOL_GPL(kvm_set_cr8); |
a03490ed | 1141 | |
2d3ad1f4 | 1142 | unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu) |
a03490ed | 1143 | { |
35754c98 | 1144 | if (lapic_in_kernel(vcpu)) |
a03490ed CO |
1145 | return kvm_lapic_get_cr8(vcpu); |
1146 | else | |
ad312c7c | 1147 | return vcpu->arch.cr8; |
a03490ed | 1148 | } |
2d3ad1f4 | 1149 | EXPORT_SYMBOL_GPL(kvm_get_cr8); |
a03490ed | 1150 | |
ae561ede NA |
1151 | static void kvm_update_dr0123(struct kvm_vcpu *vcpu) |
1152 | { | |
1153 | int i; | |
1154 | ||
1155 | if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) { | |
1156 | for (i = 0; i < KVM_NR_DB_REGS; i++) | |
1157 | vcpu->arch.eff_db[i] = vcpu->arch.db[i]; | |
1158 | vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD; | |
1159 | } | |
1160 | } | |
1161 | ||
7c86663b | 1162 | void kvm_update_dr7(struct kvm_vcpu *vcpu) |
c8639010 JK |
1163 | { |
1164 | unsigned long dr7; | |
1165 | ||
1166 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) | |
1167 | dr7 = vcpu->arch.guest_debug_dr7; | |
1168 | else | |
1169 | dr7 = vcpu->arch.dr7; | |
b3646477 | 1170 | static_call(kvm_x86_set_dr7)(vcpu, dr7); |
360b948d PB |
1171 | vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED; |
1172 | if (dr7 & DR7_BP_EN_MASK) | |
1173 | vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED; | |
c8639010 | 1174 | } |
7c86663b | 1175 | EXPORT_SYMBOL_GPL(kvm_update_dr7); |
c8639010 | 1176 | |
6f43ed01 NA |
1177 | static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu) |
1178 | { | |
1179 | u64 fixed = DR6_FIXED_1; | |
1180 | ||
d6321d49 | 1181 | if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM)) |
6f43ed01 | 1182 | fixed |= DR6_RTM; |
e8ea85fb CQ |
1183 | |
1184 | if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT)) | |
1185 | fixed |= DR6_BUS_LOCK; | |
6f43ed01 NA |
1186 | return fixed; |
1187 | } | |
1188 | ||
996ff542 | 1189 | int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val) |
020df079 | 1190 | { |
ea740059 MP |
1191 | size_t size = ARRAY_SIZE(vcpu->arch.db); |
1192 | ||
020df079 GN |
1193 | switch (dr) { |
1194 | case 0 ... 3: | |
ea740059 | 1195 | vcpu->arch.db[array_index_nospec(dr, size)] = val; |
020df079 GN |
1196 | if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) |
1197 | vcpu->arch.eff_db[dr] = val; | |
1198 | break; | |
1199 | case 4: | |
020df079 | 1200 | case 6: |
f5f6145e | 1201 | if (!kvm_dr6_valid(val)) |
996ff542 | 1202 | return 1; /* #GP */ |
6f43ed01 | 1203 | vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu); |
020df079 GN |
1204 | break; |
1205 | case 5: | |
020df079 | 1206 | default: /* 7 */ |
b91991bf | 1207 | if (!kvm_dr7_valid(val)) |
996ff542 | 1208 | return 1; /* #GP */ |
020df079 | 1209 | vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1; |
c8639010 | 1210 | kvm_update_dr7(vcpu); |
020df079 GN |
1211 | break; |
1212 | } | |
1213 | ||
1214 | return 0; | |
1215 | } | |
1216 | EXPORT_SYMBOL_GPL(kvm_set_dr); | |
1217 | ||
29d6ca41 | 1218 | void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val) |
020df079 | 1219 | { |
ea740059 MP |
1220 | size_t size = ARRAY_SIZE(vcpu->arch.db); |
1221 | ||
020df079 GN |
1222 | switch (dr) { |
1223 | case 0 ... 3: | |
ea740059 | 1224 | *val = vcpu->arch.db[array_index_nospec(dr, size)]; |
020df079 GN |
1225 | break; |
1226 | case 4: | |
020df079 | 1227 | case 6: |
5679b803 | 1228 | *val = vcpu->arch.dr6; |
020df079 GN |
1229 | break; |
1230 | case 5: | |
020df079 GN |
1231 | default: /* 7 */ |
1232 | *val = vcpu->arch.dr7; | |
1233 | break; | |
1234 | } | |
338dbc97 | 1235 | } |
020df079 GN |
1236 | EXPORT_SYMBOL_GPL(kvm_get_dr); |
1237 | ||
c483c454 | 1238 | int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu) |
022cd0e8 | 1239 | { |
de3cd117 | 1240 | u32 ecx = kvm_rcx_read(vcpu); |
022cd0e8 | 1241 | u64 data; |
022cd0e8 | 1242 | |
c483c454 SC |
1243 | if (kvm_pmu_rdpmc(vcpu, ecx, &data)) { |
1244 | kvm_inject_gp(vcpu, 0); | |
1245 | return 1; | |
1246 | } | |
1247 | ||
de3cd117 SC |
1248 | kvm_rax_write(vcpu, (u32)data); |
1249 | kvm_rdx_write(vcpu, data >> 32); | |
c483c454 | 1250 | return kvm_skip_emulated_instruction(vcpu); |
022cd0e8 | 1251 | } |
c483c454 | 1252 | EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc); |
022cd0e8 | 1253 | |
043405e1 CO |
1254 | /* |
1255 | * List of msr numbers which we expose to userspace through KVM_GET_MSRS | |
1256 | * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST. | |
1257 | * | |
7a5ee6ed CQ |
1258 | * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features) |
1259 | * extract the supported MSRs from the related const lists. | |
1260 | * msrs_to_save is selected from the msrs_to_save_all to reflect the | |
e3267cbb | 1261 | * capabilities of the host cpu. This capabilities test skips MSRs that are |
7a5ee6ed | 1262 | * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs |
62ef68bb | 1263 | * may depend on host virtualization features rather than host cpu features. |
043405e1 | 1264 | */ |
e3267cbb | 1265 | |
7a5ee6ed | 1266 | static const u32 msrs_to_save_all[] = { |
043405e1 | 1267 | MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP, |
8c06585d | 1268 | MSR_STAR, |
043405e1 CO |
1269 | #ifdef CONFIG_X86_64 |
1270 | MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR, | |
1271 | #endif | |
b3897a49 | 1272 | MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA, |
32ad73db | 1273 | MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX, |
2bdb76c0 | 1274 | MSR_IA32_SPEC_CTRL, |
bf8c55d8 CP |
1275 | MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH, |
1276 | MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK, | |
1277 | MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B, | |
1278 | MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B, | |
1279 | MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B, | |
1280 | MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B, | |
6e3ba4ab TX |
1281 | MSR_IA32_UMWAIT_CONTROL, |
1282 | ||
e2ada66e JM |
1283 | MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1, |
1284 | MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3, | |
1285 | MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS, | |
1286 | MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL, | |
1287 | MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1, | |
1288 | MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3, | |
1289 | MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5, | |
1290 | MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7, | |
1291 | MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9, | |
1292 | MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11, | |
1293 | MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13, | |
1294 | MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15, | |
1295 | MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17, | |
e2ada66e JM |
1296 | MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1, |
1297 | MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3, | |
1298 | MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5, | |
1299 | MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7, | |
1300 | MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9, | |
1301 | MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11, | |
1302 | MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13, | |
1303 | MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15, | |
1304 | MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17, | |
043405e1 CO |
1305 | }; |
1306 | ||
7a5ee6ed | 1307 | static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)]; |
043405e1 CO |
1308 | static unsigned num_msrs_to_save; |
1309 | ||
7a5ee6ed | 1310 | static const u32 emulated_msrs_all[] = { |
62ef68bb PB |
1311 | MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK, |
1312 | MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW, | |
1313 | HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL, | |
1314 | HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC, | |
72c139ba | 1315 | HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY, |
e7d9513b AS |
1316 | HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2, |
1317 | HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL, | |
e516cebb | 1318 | HV_X64_MSR_RESET, |
11c4b1ca | 1319 | HV_X64_MSR_VP_INDEX, |
9eec50b8 | 1320 | HV_X64_MSR_VP_RUNTIME, |
5c919412 | 1321 | HV_X64_MSR_SCONTROL, |
1f4b34f8 | 1322 | HV_X64_MSR_STIMER0_CONFIG, |
d4abc577 | 1323 | HV_X64_MSR_VP_ASSIST_PAGE, |
a2e164e7 VK |
1324 | HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL, |
1325 | HV_X64_MSR_TSC_EMULATION_STATUS, | |
f97f5a56 JD |
1326 | HV_X64_MSR_SYNDBG_OPTIONS, |
1327 | HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS, | |
1328 | HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER, | |
1329 | HV_X64_MSR_SYNDBG_PENDING_BUFFER, | |
a2e164e7 VK |
1330 | |
1331 | MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME, | |
557a961a | 1332 | MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK, |
62ef68bb | 1333 | |
ba904635 | 1334 | MSR_IA32_TSC_ADJUST, |
09141ec0 | 1335 | MSR_IA32_TSC_DEADLINE, |
2bdb76c0 | 1336 | MSR_IA32_ARCH_CAPABILITIES, |
27461da3 | 1337 | MSR_IA32_PERF_CAPABILITIES, |
043405e1 | 1338 | MSR_IA32_MISC_ENABLE, |
908e75f3 AK |
1339 | MSR_IA32_MCG_STATUS, |
1340 | MSR_IA32_MCG_CTL, | |
c45dcc71 | 1341 | MSR_IA32_MCG_EXT_CTL, |
64d60670 | 1342 | MSR_IA32_SMBASE, |
52797bf9 | 1343 | MSR_SMI_COUNT, |
db2336a8 KH |
1344 | MSR_PLATFORM_INFO, |
1345 | MSR_MISC_FEATURES_ENABLES, | |
bc226f07 | 1346 | MSR_AMD64_VIRT_SPEC_CTRL, |
6c6a2ab9 | 1347 | MSR_IA32_POWER_CTL, |
99634e3e | 1348 | MSR_IA32_UCODE_REV, |
191c8137 | 1349 | |
95c5c7c7 PB |
1350 | /* |
1351 | * The following list leaves out MSRs whose values are determined | |
1352 | * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs. | |
1353 | * We always support the "true" VMX control MSRs, even if the host | |
1354 | * processor does not, so I am putting these registers here rather | |
7a5ee6ed | 1355 | * than in msrs_to_save_all. |
95c5c7c7 PB |
1356 | */ |
1357 | MSR_IA32_VMX_BASIC, | |
1358 | MSR_IA32_VMX_TRUE_PINBASED_CTLS, | |
1359 | MSR_IA32_VMX_TRUE_PROCBASED_CTLS, | |
1360 | MSR_IA32_VMX_TRUE_EXIT_CTLS, | |
1361 | MSR_IA32_VMX_TRUE_ENTRY_CTLS, | |
1362 | MSR_IA32_VMX_MISC, | |
1363 | MSR_IA32_VMX_CR0_FIXED0, | |
1364 | MSR_IA32_VMX_CR4_FIXED0, | |
1365 | MSR_IA32_VMX_VMCS_ENUM, | |
1366 | MSR_IA32_VMX_PROCBASED_CTLS2, | |
1367 | MSR_IA32_VMX_EPT_VPID_CAP, | |
1368 | MSR_IA32_VMX_VMFUNC, | |
1369 | ||
191c8137 | 1370 | MSR_K7_HWCR, |
2d5ba19b | 1371 | MSR_KVM_POLL_CONTROL, |
043405e1 CO |
1372 | }; |
1373 | ||
7a5ee6ed | 1374 | static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)]; |
62ef68bb PB |
1375 | static unsigned num_emulated_msrs; |
1376 | ||
801e459a TL |
1377 | /* |
1378 | * List of msr numbers which are used to expose MSR-based features that | |
1379 | * can be used by a hypervisor to validate requested CPU features. | |
1380 | */ | |
7a5ee6ed | 1381 | static const u32 msr_based_features_all[] = { |
1389309c PB |
1382 | MSR_IA32_VMX_BASIC, |
1383 | MSR_IA32_VMX_TRUE_PINBASED_CTLS, | |
1384 | MSR_IA32_VMX_PINBASED_CTLS, | |
1385 | MSR_IA32_VMX_TRUE_PROCBASED_CTLS, | |
1386 | MSR_IA32_VMX_PROCBASED_CTLS, | |
1387 | MSR_IA32_VMX_TRUE_EXIT_CTLS, | |
1388 | MSR_IA32_VMX_EXIT_CTLS, | |
1389 | MSR_IA32_VMX_TRUE_ENTRY_CTLS, | |
1390 | MSR_IA32_VMX_ENTRY_CTLS, | |
1391 | MSR_IA32_VMX_MISC, | |
1392 | MSR_IA32_VMX_CR0_FIXED0, | |
1393 | MSR_IA32_VMX_CR0_FIXED1, | |
1394 | MSR_IA32_VMX_CR4_FIXED0, | |
1395 | MSR_IA32_VMX_CR4_FIXED1, | |
1396 | MSR_IA32_VMX_VMCS_ENUM, | |
1397 | MSR_IA32_VMX_PROCBASED_CTLS2, | |
1398 | MSR_IA32_VMX_EPT_VPID_CAP, | |
1399 | MSR_IA32_VMX_VMFUNC, | |
1400 | ||
d1d93fa9 | 1401 | MSR_F10H_DECFG, |
518e7b94 | 1402 | MSR_IA32_UCODE_REV, |
cd283252 | 1403 | MSR_IA32_ARCH_CAPABILITIES, |
27461da3 | 1404 | MSR_IA32_PERF_CAPABILITIES, |
801e459a TL |
1405 | }; |
1406 | ||
7a5ee6ed | 1407 | static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)]; |
801e459a TL |
1408 | static unsigned int num_msr_based_features; |
1409 | ||
4d22c17c | 1410 | static u64 kvm_get_arch_capabilities(void) |
5b76a3cf | 1411 | { |
4d22c17c | 1412 | u64 data = 0; |
5b76a3cf | 1413 | |
4d22c17c XL |
1414 | if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) |
1415 | rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data); | |
5b76a3cf | 1416 | |
b8e8c830 PB |
1417 | /* |
1418 | * If nx_huge_pages is enabled, KVM's shadow paging will ensure that | |
1419 | * the nested hypervisor runs with NX huge pages. If it is not, | |
d9f6e12f | 1420 | * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other |
b8e8c830 PB |
1421 | * L1 guests, so it need not worry about its own (L2) guests. |
1422 | */ | |
1423 | data |= ARCH_CAP_PSCHANGE_MC_NO; | |
1424 | ||
5b76a3cf PB |
1425 | /* |
1426 | * If we're doing cache flushes (either "always" or "cond") | |
1427 | * we will do one whenever the guest does a vmlaunch/vmresume. | |
1428 | * If an outer hypervisor is doing the cache flush for us | |
1429 | * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that | |
1430 | * capability to the guest too, and if EPT is disabled we're not | |
1431 | * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will | |
1432 | * require a nested hypervisor to do a flush of its own. | |
1433 | */ | |
1434 | if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER) | |
1435 | data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH; | |
1436 | ||
0c54914d PB |
1437 | if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN)) |
1438 | data |= ARCH_CAP_RDCL_NO; | |
1439 | if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS)) | |
1440 | data |= ARCH_CAP_SSB_NO; | |
1441 | if (!boot_cpu_has_bug(X86_BUG_MDS)) | |
1442 | data |= ARCH_CAP_MDS_NO; | |
1443 | ||
7131636e PB |
1444 | if (!boot_cpu_has(X86_FEATURE_RTM)) { |
1445 | /* | |
1446 | * If RTM=0 because the kernel has disabled TSX, the host might | |
1447 | * have TAA_NO or TSX_CTRL. Clear TAA_NO (the guest sees RTM=0 | |
1448 | * and therefore knows that there cannot be TAA) but keep | |
1449 | * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts, | |
1450 | * and we want to allow migrating those guests to tsx=off hosts. | |
1451 | */ | |
1452 | data &= ~ARCH_CAP_TAA_NO; | |
1453 | } else if (!boot_cpu_has_bug(X86_BUG_TAA)) { | |
cbbaa272 | 1454 | data |= ARCH_CAP_TAA_NO; |
7131636e PB |
1455 | } else { |
1456 | /* | |
1457 | * Nothing to do here; we emulate TSX_CTRL if present on the | |
1458 | * host so the guest can choose between disabling TSX or | |
1459 | * using VERW to clear CPU buffers. | |
1460 | */ | |
1461 | } | |
e1d38b63 | 1462 | |
5b76a3cf PB |
1463 | return data; |
1464 | } | |
5b76a3cf | 1465 | |
66421c1e WL |
1466 | static int kvm_get_msr_feature(struct kvm_msr_entry *msr) |
1467 | { | |
1468 | switch (msr->index) { | |
cd283252 | 1469 | case MSR_IA32_ARCH_CAPABILITIES: |
5b76a3cf PB |
1470 | msr->data = kvm_get_arch_capabilities(); |
1471 | break; | |
1472 | case MSR_IA32_UCODE_REV: | |
cd283252 | 1473 | rdmsrl_safe(msr->index, &msr->data); |
518e7b94 | 1474 | break; |
66421c1e | 1475 | default: |
b3646477 | 1476 | return static_call(kvm_x86_get_msr_feature)(msr); |
66421c1e WL |
1477 | } |
1478 | return 0; | |
1479 | } | |
1480 | ||
801e459a TL |
1481 | static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data) |
1482 | { | |
1483 | struct kvm_msr_entry msr; | |
66421c1e | 1484 | int r; |
801e459a TL |
1485 | |
1486 | msr.index = index; | |
66421c1e | 1487 | r = kvm_get_msr_feature(&msr); |
12bc2132 PX |
1488 | |
1489 | if (r == KVM_MSR_RET_INVALID) { | |
1490 | /* Unconditionally clear the output for simplicity */ | |
1491 | *data = 0; | |
d632826f | 1492 | if (kvm_msr_ignored_check(index, 0, false)) |
cc4cb017 | 1493 | r = 0; |
12bc2132 PX |
1494 | } |
1495 | ||
66421c1e WL |
1496 | if (r) |
1497 | return r; | |
801e459a TL |
1498 | |
1499 | *data = msr.data; | |
1500 | ||
1501 | return 0; | |
1502 | } | |
1503 | ||
11988499 | 1504 | static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer) |
15c4a640 | 1505 | { |
1b4d56b8 | 1506 | if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT)) |
11988499 | 1507 | return false; |
1b2fd70c | 1508 | |
1b4d56b8 | 1509 | if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM)) |
11988499 | 1510 | return false; |
d8017474 | 1511 | |
0a629563 SC |
1512 | if (efer & (EFER_LME | EFER_LMA) && |
1513 | !guest_cpuid_has(vcpu, X86_FEATURE_LM)) | |
1514 | return false; | |
1515 | ||
1516 | if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX)) | |
1517 | return false; | |
d8017474 | 1518 | |
384bb783 | 1519 | return true; |
11988499 SC |
1520 | |
1521 | } | |
1522 | bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer) | |
1523 | { | |
1524 | if (efer & efer_reserved_bits) | |
1525 | return false; | |
1526 | ||
1527 | return __kvm_valid_efer(vcpu, efer); | |
384bb783 JK |
1528 | } |
1529 | EXPORT_SYMBOL_GPL(kvm_valid_efer); | |
1530 | ||
11988499 | 1531 | static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
384bb783 JK |
1532 | { |
1533 | u64 old_efer = vcpu->arch.efer; | |
11988499 | 1534 | u64 efer = msr_info->data; |
72f211ec | 1535 | int r; |
384bb783 | 1536 | |
11988499 | 1537 | if (efer & efer_reserved_bits) |
66f61c92 | 1538 | return 1; |
384bb783 | 1539 | |
11988499 SC |
1540 | if (!msr_info->host_initiated) { |
1541 | if (!__kvm_valid_efer(vcpu, efer)) | |
1542 | return 1; | |
1543 | ||
1544 | if (is_paging(vcpu) && | |
1545 | (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME)) | |
1546 | return 1; | |
1547 | } | |
384bb783 | 1548 | |
15c4a640 | 1549 | efer &= ~EFER_LMA; |
f6801dff | 1550 | efer |= vcpu->arch.efer & EFER_LMA; |
15c4a640 | 1551 | |
b3646477 | 1552 | r = static_call(kvm_x86_set_efer)(vcpu, efer); |
72f211ec ML |
1553 | if (r) { |
1554 | WARN_ON(r > 0); | |
1555 | return r; | |
1556 | } | |
a3d204e2 | 1557 | |
aad82703 SY |
1558 | /* Update reserved bits */ |
1559 | if ((efer ^ old_efer) & EFER_NX) | |
1560 | kvm_mmu_reset_context(vcpu); | |
1561 | ||
b69e8cae | 1562 | return 0; |
15c4a640 CO |
1563 | } |
1564 | ||
f2b4b7dd JR |
1565 | void kvm_enable_efer_bits(u64 mask) |
1566 | { | |
1567 | efer_reserved_bits &= ~mask; | |
1568 | } | |
1569 | EXPORT_SYMBOL_GPL(kvm_enable_efer_bits); | |
1570 | ||
51de8151 AG |
1571 | bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type) |
1572 | { | |
b318e8de SC |
1573 | struct kvm_x86_msr_filter *msr_filter; |
1574 | struct msr_bitmap_range *ranges; | |
1a155254 | 1575 | struct kvm *kvm = vcpu->kvm; |
b318e8de | 1576 | bool allowed; |
1a155254 | 1577 | int idx; |
b318e8de | 1578 | u32 i; |
1a155254 | 1579 | |
b318e8de SC |
1580 | /* x2APIC MSRs do not support filtering. */ |
1581 | if (index >= 0x800 && index <= 0x8ff) | |
1a155254 AG |
1582 | return true; |
1583 | ||
1a155254 AG |
1584 | idx = srcu_read_lock(&kvm->srcu); |
1585 | ||
b318e8de SC |
1586 | msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu); |
1587 | if (!msr_filter) { | |
1588 | allowed = true; | |
1589 | goto out; | |
1590 | } | |
1591 | ||
1592 | allowed = msr_filter->default_allow; | |
1593 | ranges = msr_filter->ranges; | |
1594 | ||
1595 | for (i = 0; i < msr_filter->count; i++) { | |
1a155254 AG |
1596 | u32 start = ranges[i].base; |
1597 | u32 end = start + ranges[i].nmsrs; | |
1598 | u32 flags = ranges[i].flags; | |
1599 | unsigned long *bitmap = ranges[i].bitmap; | |
1600 | ||
1601 | if ((index >= start) && (index < end) && (flags & type)) { | |
b318e8de | 1602 | allowed = !!test_bit(index - start, bitmap); |
1a155254 AG |
1603 | break; |
1604 | } | |
1605 | } | |
1606 | ||
b318e8de | 1607 | out: |
1a155254 AG |
1608 | srcu_read_unlock(&kvm->srcu, idx); |
1609 | ||
b318e8de | 1610 | return allowed; |
51de8151 AG |
1611 | } |
1612 | EXPORT_SYMBOL_GPL(kvm_msr_allowed); | |
1613 | ||
15c4a640 | 1614 | /* |
f20935d8 SC |
1615 | * Write @data into the MSR specified by @index. Select MSR specific fault |
1616 | * checks are bypassed if @host_initiated is %true. | |
15c4a640 CO |
1617 | * Returns 0 on success, non-0 otherwise. |
1618 | * Assumes vcpu_load() was already called. | |
1619 | */ | |
f20935d8 SC |
1620 | static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data, |
1621 | bool host_initiated) | |
15c4a640 | 1622 | { |
f20935d8 SC |
1623 | struct msr_data msr; |
1624 | ||
1a155254 | 1625 | if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE)) |
cc4cb017 | 1626 | return KVM_MSR_RET_FILTERED; |
1a155254 | 1627 | |
f20935d8 | 1628 | switch (index) { |
854e8bb1 NA |
1629 | case MSR_FS_BASE: |
1630 | case MSR_GS_BASE: | |
1631 | case MSR_KERNEL_GS_BASE: | |
1632 | case MSR_CSTAR: | |
1633 | case MSR_LSTAR: | |
f20935d8 | 1634 | if (is_noncanonical_address(data, vcpu)) |
854e8bb1 NA |
1635 | return 1; |
1636 | break; | |
1637 | case MSR_IA32_SYSENTER_EIP: | |
1638 | case MSR_IA32_SYSENTER_ESP: | |
1639 | /* | |
1640 | * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if | |
1641 | * non-canonical address is written on Intel but not on | |
1642 | * AMD (which ignores the top 32-bits, because it does | |
1643 | * not implement 64-bit SYSENTER). | |
1644 | * | |
1645 | * 64-bit code should hence be able to write a non-canonical | |
1646 | * value on AMD. Making the address canonical ensures that | |
1647 | * vmentry does not fail on Intel after writing a non-canonical | |
1648 | * value, and that something deterministic happens if the guest | |
1649 | * invokes 64-bit SYSENTER. | |
1650 | */ | |
f20935d8 | 1651 | data = get_canonical(data, vcpu_virt_addr_bits(vcpu)); |
61a05d44 SC |
1652 | break; |
1653 | case MSR_TSC_AUX: | |
1654 | if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX)) | |
1655 | return 1; | |
1656 | ||
1657 | if (!host_initiated && | |
1658 | !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) && | |
1659 | !guest_cpuid_has(vcpu, X86_FEATURE_RDPID)) | |
1660 | return 1; | |
1661 | ||
1662 | /* | |
1663 | * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has | |
1664 | * incomplete and conflicting architectural behavior. Current | |
1665 | * AMD CPUs completely ignore bits 63:32, i.e. they aren't | |
1666 | * reserved and always read as zeros. Enforce Intel's reserved | |
1667 | * bits check if and only if the guest CPU is Intel, and clear | |
1668 | * the bits in all other cases. This ensures cross-vendor | |
1669 | * migration will provide consistent behavior for the guest. | |
1670 | */ | |
1671 | if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0) | |
1672 | return 1; | |
1673 | ||
1674 | data = (u32)data; | |
1675 | break; | |
854e8bb1 | 1676 | } |
f20935d8 SC |
1677 | |
1678 | msr.data = data; | |
1679 | msr.index = index; | |
1680 | msr.host_initiated = host_initiated; | |
1681 | ||
b3646477 | 1682 | return static_call(kvm_x86_set_msr)(vcpu, &msr); |
15c4a640 CO |
1683 | } |
1684 | ||
6abe9c13 PX |
1685 | static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu, |
1686 | u32 index, u64 data, bool host_initiated) | |
1687 | { | |
1688 | int ret = __kvm_set_msr(vcpu, index, data, host_initiated); | |
1689 | ||
1690 | if (ret == KVM_MSR_RET_INVALID) | |
d632826f | 1691 | if (kvm_msr_ignored_check(index, data, true)) |
cc4cb017 | 1692 | ret = 0; |
6abe9c13 PX |
1693 | |
1694 | return ret; | |
1695 | } | |
1696 | ||
313a3dc7 | 1697 | /* |
f20935d8 SC |
1698 | * Read the MSR specified by @index into @data. Select MSR specific fault |
1699 | * checks are bypassed if @host_initiated is %true. | |
1700 | * Returns 0 on success, non-0 otherwise. | |
1701 | * Assumes vcpu_load() was already called. | |
313a3dc7 | 1702 | */ |
edef5c36 PB |
1703 | int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data, |
1704 | bool host_initiated) | |
609e36d3 PB |
1705 | { |
1706 | struct msr_data msr; | |
f20935d8 | 1707 | int ret; |
609e36d3 | 1708 | |
1a155254 | 1709 | if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ)) |
cc4cb017 | 1710 | return KVM_MSR_RET_FILTERED; |
1a155254 | 1711 | |
61a05d44 SC |
1712 | switch (index) { |
1713 | case MSR_TSC_AUX: | |
1714 | if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX)) | |
1715 | return 1; | |
1716 | ||
1717 | if (!host_initiated && | |
1718 | !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) && | |
1719 | !guest_cpuid_has(vcpu, X86_FEATURE_RDPID)) | |
1720 | return 1; | |
1721 | break; | |
1722 | } | |
1723 | ||
609e36d3 | 1724 | msr.index = index; |
f20935d8 | 1725 | msr.host_initiated = host_initiated; |
609e36d3 | 1726 | |
b3646477 | 1727 | ret = static_call(kvm_x86_get_msr)(vcpu, &msr); |
f20935d8 SC |
1728 | if (!ret) |
1729 | *data = msr.data; | |
1730 | return ret; | |
609e36d3 PB |
1731 | } |
1732 | ||
6abe9c13 PX |
1733 | static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu, |
1734 | u32 index, u64 *data, bool host_initiated) | |
1735 | { | |
1736 | int ret = __kvm_get_msr(vcpu, index, data, host_initiated); | |
1737 | ||
1738 | if (ret == KVM_MSR_RET_INVALID) { | |
1739 | /* Unconditionally clear *data for simplicity */ | |
1740 | *data = 0; | |
d632826f | 1741 | if (kvm_msr_ignored_check(index, 0, false)) |
cc4cb017 | 1742 | ret = 0; |
6abe9c13 PX |
1743 | } |
1744 | ||
1745 | return ret; | |
1746 | } | |
1747 | ||
f20935d8 | 1748 | int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data) |
313a3dc7 | 1749 | { |
6abe9c13 | 1750 | return kvm_get_msr_ignored_check(vcpu, index, data, false); |
f20935d8 SC |
1751 | } |
1752 | EXPORT_SYMBOL_GPL(kvm_get_msr); | |
8fe8ab46 | 1753 | |
f20935d8 SC |
1754 | int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data) |
1755 | { | |
6abe9c13 | 1756 | return kvm_set_msr_ignored_check(vcpu, index, data, false); |
f20935d8 SC |
1757 | } |
1758 | EXPORT_SYMBOL_GPL(kvm_set_msr); | |
1759 | ||
8b474427 | 1760 | static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu) |
1ae09954 | 1761 | { |
8b474427 PB |
1762 | int err = vcpu->run->msr.error; |
1763 | if (!err) { | |
1ae09954 AG |
1764 | kvm_rax_write(vcpu, (u32)vcpu->run->msr.data); |
1765 | kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32); | |
1766 | } | |
1767 | ||
b3646477 | 1768 | return static_call(kvm_x86_complete_emulated_msr)(vcpu, err); |
1ae09954 AG |
1769 | } |
1770 | ||
1771 | static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu) | |
1772 | { | |
b3646477 | 1773 | return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error); |
1ae09954 AG |
1774 | } |
1775 | ||
1776 | static u64 kvm_msr_reason(int r) | |
1777 | { | |
1778 | switch (r) { | |
cc4cb017 | 1779 | case KVM_MSR_RET_INVALID: |
1ae09954 | 1780 | return KVM_MSR_EXIT_REASON_UNKNOWN; |
cc4cb017 | 1781 | case KVM_MSR_RET_FILTERED: |
1a155254 | 1782 | return KVM_MSR_EXIT_REASON_FILTER; |
1ae09954 AG |
1783 | default: |
1784 | return KVM_MSR_EXIT_REASON_INVAL; | |
1785 | } | |
1786 | } | |
1787 | ||
1788 | static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index, | |
1789 | u32 exit_reason, u64 data, | |
1790 | int (*completion)(struct kvm_vcpu *vcpu), | |
1791 | int r) | |
1792 | { | |
1793 | u64 msr_reason = kvm_msr_reason(r); | |
1794 | ||
1795 | /* Check if the user wanted to know about this MSR fault */ | |
1796 | if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason)) | |
1797 | return 0; | |
1798 | ||
1799 | vcpu->run->exit_reason = exit_reason; | |
1800 | vcpu->run->msr.error = 0; | |
1801 | memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad)); | |
1802 | vcpu->run->msr.reason = msr_reason; | |
1803 | vcpu->run->msr.index = index; | |
1804 | vcpu->run->msr.data = data; | |
1805 | vcpu->arch.complete_userspace_io = completion; | |
1806 | ||
1807 | return 1; | |
1808 | } | |
1809 | ||
1810 | static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r) | |
1811 | { | |
1812 | return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0, | |
1813 | complete_emulated_rdmsr, r); | |
1814 | } | |
1815 | ||
1816 | static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r) | |
1817 | { | |
1818 | return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data, | |
1819 | complete_emulated_wrmsr, r); | |
1820 | } | |
1821 | ||
1edce0a9 SC |
1822 | int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu) |
1823 | { | |
1824 | u32 ecx = kvm_rcx_read(vcpu); | |
1825 | u64 data; | |
1ae09954 AG |
1826 | int r; |
1827 | ||
1828 | r = kvm_get_msr(vcpu, ecx, &data); | |
1edce0a9 | 1829 | |
1ae09954 AG |
1830 | /* MSR read failed? See if we should ask user space */ |
1831 | if (r && kvm_get_msr_user_space(vcpu, ecx, r)) { | |
1832 | /* Bounce to user space */ | |
1833 | return 0; | |
1834 | } | |
1835 | ||
8b474427 PB |
1836 | if (!r) { |
1837 | trace_kvm_msr_read(ecx, data); | |
1838 | ||
1839 | kvm_rax_write(vcpu, data & -1u); | |
1840 | kvm_rdx_write(vcpu, (data >> 32) & -1u); | |
1841 | } else { | |
1edce0a9 | 1842 | trace_kvm_msr_read_ex(ecx); |
1edce0a9 SC |
1843 | } |
1844 | ||
b3646477 | 1845 | return static_call(kvm_x86_complete_emulated_msr)(vcpu, r); |
1edce0a9 SC |
1846 | } |
1847 | EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr); | |
1848 | ||
1849 | int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu) | |
1850 | { | |
1851 | u32 ecx = kvm_rcx_read(vcpu); | |
1852 | u64 data = kvm_read_edx_eax(vcpu); | |
1ae09954 | 1853 | int r; |
1edce0a9 | 1854 | |
1ae09954 AG |
1855 | r = kvm_set_msr(vcpu, ecx, data); |
1856 | ||
1857 | /* MSR write failed? See if we should ask user space */ | |
7dffecaf | 1858 | if (r && kvm_set_msr_user_space(vcpu, ecx, data, r)) |
1ae09954 AG |
1859 | /* Bounce to user space */ |
1860 | return 0; | |
7dffecaf ML |
1861 | |
1862 | /* Signal all other negative errors to userspace */ | |
1863 | if (r < 0) | |
1864 | return r; | |
1ae09954 | 1865 | |
8b474427 PB |
1866 | if (!r) |
1867 | trace_kvm_msr_write(ecx, data); | |
1868 | else | |
1edce0a9 | 1869 | trace_kvm_msr_write_ex(ecx, data); |
1edce0a9 | 1870 | |
b3646477 | 1871 | return static_call(kvm_x86_complete_emulated_msr)(vcpu, r); |
1edce0a9 SC |
1872 | } |
1873 | EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr); | |
1874 | ||
5ff3a351 SC |
1875 | int kvm_emulate_as_nop(struct kvm_vcpu *vcpu) |
1876 | { | |
1877 | return kvm_skip_emulated_instruction(vcpu); | |
1878 | } | |
1879 | EXPORT_SYMBOL_GPL(kvm_emulate_as_nop); | |
1880 | ||
1881 | int kvm_emulate_invd(struct kvm_vcpu *vcpu) | |
1882 | { | |
1883 | /* Treat an INVD instruction as a NOP and just skip it. */ | |
1884 | return kvm_emulate_as_nop(vcpu); | |
1885 | } | |
1886 | EXPORT_SYMBOL_GPL(kvm_emulate_invd); | |
1887 | ||
1888 | int kvm_emulate_mwait(struct kvm_vcpu *vcpu) | |
1889 | { | |
1890 | pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n"); | |
1891 | return kvm_emulate_as_nop(vcpu); | |
1892 | } | |
1893 | EXPORT_SYMBOL_GPL(kvm_emulate_mwait); | |
1894 | ||
1895 | int kvm_handle_invalid_op(struct kvm_vcpu *vcpu) | |
1896 | { | |
1897 | kvm_queue_exception(vcpu, UD_VECTOR); | |
1898 | return 1; | |
1899 | } | |
1900 | EXPORT_SYMBOL_GPL(kvm_handle_invalid_op); | |
1901 | ||
1902 | int kvm_emulate_monitor(struct kvm_vcpu *vcpu) | |
1903 | { | |
1904 | pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n"); | |
1905 | return kvm_emulate_as_nop(vcpu); | |
1906 | } | |
1907 | EXPORT_SYMBOL_GPL(kvm_emulate_monitor); | |
1908 | ||
d89d04ab | 1909 | static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu) |
5a9f5443 | 1910 | { |
4ae7dc97 | 1911 | xfer_to_guest_mode_prepare(); |
5a9f5443 | 1912 | return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) || |
72c3c0fe | 1913 | xfer_to_guest_mode_work_pending(); |
5a9f5443 | 1914 | } |
5a9f5443 | 1915 | |
1e9e2622 WL |
1916 | /* |
1917 | * The fast path for frequent and performance sensitive wrmsr emulation, | |
1918 | * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces | |
1919 | * the latency of virtual IPI by avoiding the expensive bits of transitioning | |
1920 | * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the | |
1921 | * other cases which must be called after interrupts are enabled on the host. | |
1922 | */ | |
1923 | static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data) | |
1924 | { | |
e1be9ac8 WL |
1925 | if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic)) |
1926 | return 1; | |
1927 | ||
1928 | if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) && | |
1e9e2622 | 1929 | ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) && |
4064a4c6 WL |
1930 | ((data & APIC_MODE_MASK) == APIC_DM_FIXED) && |
1931 | ((u32)(data >> 32) != X2APIC_BROADCAST)) { | |
1e9e2622 | 1932 | |
d5361678 WL |
1933 | data &= ~(1 << 12); |
1934 | kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32)); | |
1e9e2622 | 1935 | kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32)); |
d5361678 WL |
1936 | kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data); |
1937 | trace_kvm_apic_write(APIC_ICR, (u32)data); | |
1938 | return 0; | |
1e9e2622 WL |
1939 | } |
1940 | ||
1941 | return 1; | |
1942 | } | |
1943 | ||
ae95f566 WL |
1944 | static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data) |
1945 | { | |
1946 | if (!kvm_can_use_hv_timer(vcpu)) | |
1947 | return 1; | |
1948 | ||
1949 | kvm_set_lapic_tscdeadline_msr(vcpu, data); | |
1950 | return 0; | |
1951 | } | |
1952 | ||
404d5d7b | 1953 | fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu) |
1e9e2622 WL |
1954 | { |
1955 | u32 msr = kvm_rcx_read(vcpu); | |
8a1038de | 1956 | u64 data; |
404d5d7b | 1957 | fastpath_t ret = EXIT_FASTPATH_NONE; |
1e9e2622 WL |
1958 | |
1959 | switch (msr) { | |
1960 | case APIC_BASE_MSR + (APIC_ICR >> 4): | |
8a1038de | 1961 | data = kvm_read_edx_eax(vcpu); |
404d5d7b WL |
1962 | if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) { |
1963 | kvm_skip_emulated_instruction(vcpu); | |
1964 | ret = EXIT_FASTPATH_EXIT_HANDLED; | |
80bc97f2 | 1965 | } |
1e9e2622 | 1966 | break; |
09141ec0 | 1967 | case MSR_IA32_TSC_DEADLINE: |
ae95f566 WL |
1968 | data = kvm_read_edx_eax(vcpu); |
1969 | if (!handle_fastpath_set_tscdeadline(vcpu, data)) { | |
1970 | kvm_skip_emulated_instruction(vcpu); | |
1971 | ret = EXIT_FASTPATH_REENTER_GUEST; | |
1972 | } | |
1973 | break; | |
1e9e2622 | 1974 | default: |
404d5d7b | 1975 | break; |
1e9e2622 WL |
1976 | } |
1977 | ||
404d5d7b | 1978 | if (ret != EXIT_FASTPATH_NONE) |
1e9e2622 | 1979 | trace_kvm_msr_write(msr, data); |
1e9e2622 | 1980 | |
404d5d7b | 1981 | return ret; |
1e9e2622 WL |
1982 | } |
1983 | EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff); | |
1984 | ||
f20935d8 SC |
1985 | /* |
1986 | * Adapt set_msr() to msr_io()'s calling convention | |
1987 | */ | |
1988 | static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data) | |
1989 | { | |
6abe9c13 | 1990 | return kvm_get_msr_ignored_check(vcpu, index, data, true); |
f20935d8 SC |
1991 | } |
1992 | ||
1993 | static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data) | |
1994 | { | |
6abe9c13 | 1995 | return kvm_set_msr_ignored_check(vcpu, index, *data, true); |
313a3dc7 CO |
1996 | } |
1997 | ||
16e8d74d | 1998 | #ifdef CONFIG_X86_64 |
53fafdbb MT |
1999 | struct pvclock_clock { |
2000 | int vclock_mode; | |
2001 | u64 cycle_last; | |
2002 | u64 mask; | |
2003 | u32 mult; | |
2004 | u32 shift; | |
917f9475 PB |
2005 | u64 base_cycles; |
2006 | u64 offset; | |
53fafdbb MT |
2007 | }; |
2008 | ||
16e8d74d MT |
2009 | struct pvclock_gtod_data { |
2010 | seqcount_t seq; | |
2011 | ||
53fafdbb MT |
2012 | struct pvclock_clock clock; /* extract of a clocksource struct */ |
2013 | struct pvclock_clock raw_clock; /* extract of a clocksource struct */ | |
16e8d74d | 2014 | |
917f9475 | 2015 | ktime_t offs_boot; |
55dd00a7 | 2016 | u64 wall_time_sec; |
16e8d74d MT |
2017 | }; |
2018 | ||
2019 | static struct pvclock_gtod_data pvclock_gtod_data; | |
2020 | ||
2021 | static void update_pvclock_gtod(struct timekeeper *tk) | |
2022 | { | |
2023 | struct pvclock_gtod_data *vdata = &pvclock_gtod_data; | |
2024 | ||
2025 | write_seqcount_begin(&vdata->seq); | |
2026 | ||
2027 | /* copy pvclock gtod data */ | |
b95a8a27 | 2028 | vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode; |
876e7881 PZ |
2029 | vdata->clock.cycle_last = tk->tkr_mono.cycle_last; |
2030 | vdata->clock.mask = tk->tkr_mono.mask; | |
2031 | vdata->clock.mult = tk->tkr_mono.mult; | |
2032 | vdata->clock.shift = tk->tkr_mono.shift; | |
917f9475 PB |
2033 | vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec; |
2034 | vdata->clock.offset = tk->tkr_mono.base; | |
16e8d74d | 2035 | |
b95a8a27 | 2036 | vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode; |
53fafdbb MT |
2037 | vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last; |
2038 | vdata->raw_clock.mask = tk->tkr_raw.mask; | |
2039 | vdata->raw_clock.mult = tk->tkr_raw.mult; | |
2040 | vdata->raw_clock.shift = tk->tkr_raw.shift; | |
917f9475 PB |
2041 | vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec; |
2042 | vdata->raw_clock.offset = tk->tkr_raw.base; | |
16e8d74d | 2043 | |
55dd00a7 MT |
2044 | vdata->wall_time_sec = tk->xtime_sec; |
2045 | ||
917f9475 | 2046 | vdata->offs_boot = tk->offs_boot; |
53fafdbb | 2047 | |
16e8d74d MT |
2048 | write_seqcount_end(&vdata->seq); |
2049 | } | |
8171cd68 PB |
2050 | |
2051 | static s64 get_kvmclock_base_ns(void) | |
2052 | { | |
2053 | /* Count up from boot time, but with the frequency of the raw clock. */ | |
2054 | return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot)); | |
2055 | } | |
2056 | #else | |
2057 | static s64 get_kvmclock_base_ns(void) | |
2058 | { | |
2059 | /* Master clock not used, so we can just use CLOCK_BOOTTIME. */ | |
2060 | return ktime_get_boottime_ns(); | |
2061 | } | |
16e8d74d MT |
2062 | #endif |
2063 | ||
629b5348 | 2064 | void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs) |
18068523 | 2065 | { |
9ed3c444 AK |
2066 | int version; |
2067 | int r; | |
50d0a0f9 | 2068 | struct pvclock_wall_clock wc; |
629b5348 | 2069 | u32 wc_sec_hi; |
8171cd68 | 2070 | u64 wall_nsec; |
18068523 GOC |
2071 | |
2072 | if (!wall_clock) | |
2073 | return; | |
2074 | ||
9ed3c444 AK |
2075 | r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version)); |
2076 | if (r) | |
2077 | return; | |
2078 | ||
2079 | if (version & 1) | |
2080 | ++version; /* first time write, random junk */ | |
2081 | ||
2082 | ++version; | |
18068523 | 2083 | |
1dab1345 NK |
2084 | if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version))) |
2085 | return; | |
18068523 | 2086 | |
50d0a0f9 GH |
2087 | /* |
2088 | * The guest calculates current wall clock time by adding | |
34c238a1 | 2089 | * system time (updated by kvm_guest_time_update below) to the |
8171cd68 | 2090 | * wall clock specified here. We do the reverse here. |
50d0a0f9 | 2091 | */ |
8171cd68 | 2092 | wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm); |
50d0a0f9 | 2093 | |
8171cd68 PB |
2094 | wc.nsec = do_div(wall_nsec, 1000000000); |
2095 | wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */ | |
50d0a0f9 | 2096 | wc.version = version; |
18068523 GOC |
2097 | |
2098 | kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc)); | |
2099 | ||
629b5348 JM |
2100 | if (sec_hi_ofs) { |
2101 | wc_sec_hi = wall_nsec >> 32; | |
2102 | kvm_write_guest(kvm, wall_clock + sec_hi_ofs, | |
2103 | &wc_sec_hi, sizeof(wc_sec_hi)); | |
2104 | } | |
2105 | ||
18068523 GOC |
2106 | version++; |
2107 | kvm_write_guest(kvm, wall_clock, &version, sizeof(version)); | |
18068523 GOC |
2108 | } |
2109 | ||
5b9bb0eb OU |
2110 | static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time, |
2111 | bool old_msr, bool host_initiated) | |
2112 | { | |
2113 | struct kvm_arch *ka = &vcpu->kvm->arch; | |
2114 | ||
2115 | if (vcpu->vcpu_id == 0 && !host_initiated) { | |
1e293d1a | 2116 | if (ka->boot_vcpu_runs_old_kvmclock != old_msr) |
5b9bb0eb OU |
2117 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); |
2118 | ||
2119 | ka->boot_vcpu_runs_old_kvmclock = old_msr; | |
2120 | } | |
2121 | ||
2122 | vcpu->arch.time = system_time; | |
2123 | kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu); | |
2124 | ||
2125 | /* we verify if the enable bit is set... */ | |
2126 | vcpu->arch.pv_time_enabled = false; | |
2127 | if (!(system_time & 1)) | |
2128 | return; | |
2129 | ||
2130 | if (!kvm_gfn_to_hva_cache_init(vcpu->kvm, | |
2131 | &vcpu->arch.pv_time, system_time & ~1ULL, | |
2132 | sizeof(struct pvclock_vcpu_time_info))) | |
2133 | vcpu->arch.pv_time_enabled = true; | |
2134 | ||
2135 | return; | |
2136 | } | |
2137 | ||
50d0a0f9 GH |
2138 | static uint32_t div_frac(uint32_t dividend, uint32_t divisor) |
2139 | { | |
b51012de PB |
2140 | do_shl32_div32(dividend, divisor); |
2141 | return dividend; | |
50d0a0f9 GH |
2142 | } |
2143 | ||
3ae13faa | 2144 | static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz, |
5f4e3f88 | 2145 | s8 *pshift, u32 *pmultiplier) |
50d0a0f9 | 2146 | { |
5f4e3f88 | 2147 | uint64_t scaled64; |
50d0a0f9 GH |
2148 | int32_t shift = 0; |
2149 | uint64_t tps64; | |
2150 | uint32_t tps32; | |
2151 | ||
3ae13faa PB |
2152 | tps64 = base_hz; |
2153 | scaled64 = scaled_hz; | |
50933623 | 2154 | while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) { |
50d0a0f9 GH |
2155 | tps64 >>= 1; |
2156 | shift--; | |
2157 | } | |
2158 | ||
2159 | tps32 = (uint32_t)tps64; | |
50933623 JK |
2160 | while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) { |
2161 | if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000) | |
5f4e3f88 ZA |
2162 | scaled64 >>= 1; |
2163 | else | |
2164 | tps32 <<= 1; | |
50d0a0f9 GH |
2165 | shift++; |
2166 | } | |
2167 | ||
5f4e3f88 ZA |
2168 | *pshift = shift; |
2169 | *pmultiplier = div_frac(scaled64, tps32); | |
50d0a0f9 GH |
2170 | } |
2171 | ||
d828199e | 2172 | #ifdef CONFIG_X86_64 |
16e8d74d | 2173 | static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0); |
d828199e | 2174 | #endif |
16e8d74d | 2175 | |
c8076604 | 2176 | static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz); |
69b0049a | 2177 | static unsigned long max_tsc_khz; |
c8076604 | 2178 | |
cc578287 | 2179 | static u32 adjust_tsc_khz(u32 khz, s32 ppm) |
1e993611 | 2180 | { |
cc578287 ZA |
2181 | u64 v = (u64)khz * (1000000 + ppm); |
2182 | do_div(v, 1000000); | |
2183 | return v; | |
1e993611 JR |
2184 | } |
2185 | ||
1ab9287a IS |
2186 | static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier); |
2187 | ||
381d585c HZ |
2188 | static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale) |
2189 | { | |
2190 | u64 ratio; | |
2191 | ||
2192 | /* Guest TSC same frequency as host TSC? */ | |
2193 | if (!scale) { | |
1ab9287a | 2194 | kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio); |
381d585c HZ |
2195 | return 0; |
2196 | } | |
2197 | ||
2198 | /* TSC scaling supported? */ | |
2199 | if (!kvm_has_tsc_control) { | |
2200 | if (user_tsc_khz > tsc_khz) { | |
2201 | vcpu->arch.tsc_catchup = 1; | |
2202 | vcpu->arch.tsc_always_catchup = 1; | |
2203 | return 0; | |
2204 | } else { | |
3f16a5c3 | 2205 | pr_warn_ratelimited("user requested TSC rate below hardware speed\n"); |
381d585c HZ |
2206 | return -1; |
2207 | } | |
2208 | } | |
2209 | ||
2210 | /* TSC scaling required - calculate ratio */ | |
2211 | ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits, | |
2212 | user_tsc_khz, tsc_khz); | |
2213 | ||
2214 | if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) { | |
3f16a5c3 PB |
2215 | pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n", |
2216 | user_tsc_khz); | |
381d585c HZ |
2217 | return -1; |
2218 | } | |
2219 | ||
1ab9287a | 2220 | kvm_vcpu_write_tsc_multiplier(vcpu, ratio); |
381d585c HZ |
2221 | return 0; |
2222 | } | |
2223 | ||
4941b8cb | 2224 | static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz) |
759379dd | 2225 | { |
cc578287 ZA |
2226 | u32 thresh_lo, thresh_hi; |
2227 | int use_scaling = 0; | |
217fc9cf | 2228 | |
03ba32ca | 2229 | /* tsc_khz can be zero if TSC calibration fails */ |
4941b8cb | 2230 | if (user_tsc_khz == 0) { |
ad721883 | 2231 | /* set tsc_scaling_ratio to a safe value */ |
1ab9287a | 2232 | kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio); |
381d585c | 2233 | return -1; |
ad721883 | 2234 | } |
03ba32ca | 2235 | |
c285545f | 2236 | /* Compute a scale to convert nanoseconds in TSC cycles */ |
3ae13faa | 2237 | kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC, |
cc578287 ZA |
2238 | &vcpu->arch.virtual_tsc_shift, |
2239 | &vcpu->arch.virtual_tsc_mult); | |
4941b8cb | 2240 | vcpu->arch.virtual_tsc_khz = user_tsc_khz; |
cc578287 ZA |
2241 | |
2242 | /* | |
2243 | * Compute the variation in TSC rate which is acceptable | |
2244 | * within the range of tolerance and decide if the | |
2245 | * rate being applied is within that bounds of the hardware | |
2246 | * rate. If so, no scaling or compensation need be done. | |
2247 | */ | |
2248 | thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm); | |
2249 | thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm); | |
4941b8cb PB |
2250 | if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) { |
2251 | pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi); | |
cc578287 ZA |
2252 | use_scaling = 1; |
2253 | } | |
4941b8cb | 2254 | return set_tsc_khz(vcpu, user_tsc_khz, use_scaling); |
c285545f ZA |
2255 | } |
2256 | ||
2257 | static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns) | |
2258 | { | |
e26101b1 | 2259 | u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec, |
cc578287 ZA |
2260 | vcpu->arch.virtual_tsc_mult, |
2261 | vcpu->arch.virtual_tsc_shift); | |
e26101b1 | 2262 | tsc += vcpu->arch.this_tsc_write; |
c285545f ZA |
2263 | return tsc; |
2264 | } | |
2265 | ||
b0c39dc6 VK |
2266 | static inline int gtod_is_based_on_tsc(int mode) |
2267 | { | |
b95a8a27 | 2268 | return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK; |
b0c39dc6 VK |
2269 | } |
2270 | ||
69b0049a | 2271 | static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu) |
b48aa97e MT |
2272 | { |
2273 | #ifdef CONFIG_X86_64 | |
2274 | bool vcpus_matched; | |
b48aa97e MT |
2275 | struct kvm_arch *ka = &vcpu->kvm->arch; |
2276 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; | |
2277 | ||
2278 | vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 == | |
2279 | atomic_read(&vcpu->kvm->online_vcpus)); | |
2280 | ||
7f187922 MT |
2281 | /* |
2282 | * Once the masterclock is enabled, always perform request in | |
2283 | * order to update it. | |
2284 | * | |
2285 | * In order to enable masterclock, the host clocksource must be TSC | |
2286 | * and the vcpus need to have matched TSCs. When that happens, | |
2287 | * perform request to enable masterclock. | |
2288 | */ | |
2289 | if (ka->use_master_clock || | |
b0c39dc6 | 2290 | (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched)) |
b48aa97e MT |
2291 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); |
2292 | ||
2293 | trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc, | |
2294 | atomic_read(&vcpu->kvm->online_vcpus), | |
2295 | ka->use_master_clock, gtod->clock.vclock_mode); | |
2296 | #endif | |
2297 | } | |
2298 | ||
35181e86 HZ |
2299 | /* |
2300 | * Multiply tsc by a fixed point number represented by ratio. | |
2301 | * | |
2302 | * The most significant 64-N bits (mult) of ratio represent the | |
2303 | * integral part of the fixed point number; the remaining N bits | |
2304 | * (frac) represent the fractional part, ie. ratio represents a fixed | |
2305 | * point number (mult + frac * 2^(-N)). | |
2306 | * | |
2307 | * N equals to kvm_tsc_scaling_ratio_frac_bits. | |
2308 | */ | |
2309 | static inline u64 __scale_tsc(u64 ratio, u64 tsc) | |
2310 | { | |
2311 | return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits); | |
2312 | } | |
2313 | ||
fe3eb504 | 2314 | u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc, u64 ratio) |
35181e86 HZ |
2315 | { |
2316 | u64 _tsc = tsc; | |
35181e86 HZ |
2317 | |
2318 | if (ratio != kvm_default_tsc_scaling_ratio) | |
2319 | _tsc = __scale_tsc(ratio, tsc); | |
2320 | ||
2321 | return _tsc; | |
2322 | } | |
2323 | EXPORT_SYMBOL_GPL(kvm_scale_tsc); | |
2324 | ||
9b399dfd | 2325 | static u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc) |
07c1419a HZ |
2326 | { |
2327 | u64 tsc; | |
2328 | ||
fe3eb504 | 2329 | tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio); |
07c1419a HZ |
2330 | |
2331 | return target_tsc - tsc; | |
2332 | } | |
2333 | ||
4ba76538 HZ |
2334 | u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc) |
2335 | { | |
fe3eb504 IS |
2336 | return vcpu->arch.l1_tsc_offset + |
2337 | kvm_scale_tsc(vcpu, host_tsc, vcpu->arch.l1_tsc_scaling_ratio); | |
4ba76538 HZ |
2338 | } |
2339 | EXPORT_SYMBOL_GPL(kvm_read_l1_tsc); | |
2340 | ||
83150f29 IS |
2341 | u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier) |
2342 | { | |
2343 | u64 nested_offset; | |
2344 | ||
2345 | if (l2_multiplier == kvm_default_tsc_scaling_ratio) | |
2346 | nested_offset = l1_offset; | |
2347 | else | |
2348 | nested_offset = mul_s64_u64_shr((s64) l1_offset, l2_multiplier, | |
2349 | kvm_tsc_scaling_ratio_frac_bits); | |
2350 | ||
2351 | nested_offset += l2_offset; | |
2352 | return nested_offset; | |
2353 | } | |
2354 | EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_offset); | |
2355 | ||
2356 | u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier) | |
2357 | { | |
2358 | if (l2_multiplier != kvm_default_tsc_scaling_ratio) | |
2359 | return mul_u64_u64_shr(l1_multiplier, l2_multiplier, | |
2360 | kvm_tsc_scaling_ratio_frac_bits); | |
2361 | ||
2362 | return l1_multiplier; | |
2363 | } | |
2364 | EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_multiplier); | |
2365 | ||
edcfe540 | 2366 | static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset) |
a545ab6a | 2367 | { |
edcfe540 IS |
2368 | trace_kvm_write_tsc_offset(vcpu->vcpu_id, |
2369 | vcpu->arch.l1_tsc_offset, | |
2370 | l1_offset); | |
2371 | ||
2372 | vcpu->arch.l1_tsc_offset = l1_offset; | |
2373 | ||
2374 | /* | |
2375 | * If we are here because L1 chose not to trap WRMSR to TSC then | |
2376 | * according to the spec this should set L1's TSC (as opposed to | |
2377 | * setting L1's offset for L2). | |
2378 | */ | |
2379 | if (is_guest_mode(vcpu)) | |
2380 | vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset( | |
2381 | l1_offset, | |
2382 | static_call(kvm_x86_get_l2_tsc_offset)(vcpu), | |
2383 | static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu)); | |
2384 | else | |
2385 | vcpu->arch.tsc_offset = l1_offset; | |
2386 | ||
2387 | static_call(kvm_x86_write_tsc_offset)(vcpu, vcpu->arch.tsc_offset); | |
a545ab6a LC |
2388 | } |
2389 | ||
1ab9287a IS |
2390 | static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier) |
2391 | { | |
2392 | vcpu->arch.l1_tsc_scaling_ratio = l1_multiplier; | |
2393 | ||
2394 | /* Userspace is changing the multiplier while L2 is active */ | |
2395 | if (is_guest_mode(vcpu)) | |
2396 | vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier( | |
2397 | l1_multiplier, | |
2398 | static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu)); | |
2399 | else | |
2400 | vcpu->arch.tsc_scaling_ratio = l1_multiplier; | |
2401 | ||
2402 | if (kvm_has_tsc_control) | |
2403 | static_call(kvm_x86_write_tsc_multiplier)( | |
2404 | vcpu, vcpu->arch.tsc_scaling_ratio); | |
2405 | } | |
2406 | ||
b0c39dc6 VK |
2407 | static inline bool kvm_check_tsc_unstable(void) |
2408 | { | |
2409 | #ifdef CONFIG_X86_64 | |
2410 | /* | |
2411 | * TSC is marked unstable when we're running on Hyper-V, | |
2412 | * 'TSC page' clocksource is good. | |
2413 | */ | |
b95a8a27 | 2414 | if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK) |
b0c39dc6 VK |
2415 | return false; |
2416 | #endif | |
2417 | return check_tsc_unstable(); | |
2418 | } | |
2419 | ||
0c899c25 | 2420 | static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data) |
99e3e30a ZA |
2421 | { |
2422 | struct kvm *kvm = vcpu->kvm; | |
f38e098f | 2423 | u64 offset, ns, elapsed; |
99e3e30a | 2424 | unsigned long flags; |
b48aa97e | 2425 | bool matched; |
0d3da0d2 | 2426 | bool already_matched; |
c5e8ec8e | 2427 | bool synchronizing = false; |
99e3e30a | 2428 | |
038f8c11 | 2429 | raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags); |
9b399dfd | 2430 | offset = kvm_compute_l1_tsc_offset(vcpu, data); |
8171cd68 | 2431 | ns = get_kvmclock_base_ns(); |
f38e098f | 2432 | elapsed = ns - kvm->arch.last_tsc_nsec; |
5d3cb0f6 | 2433 | |
03ba32ca | 2434 | if (vcpu->arch.virtual_tsc_khz) { |
0c899c25 | 2435 | if (data == 0) { |
bd8fab39 DP |
2436 | /* |
2437 | * detection of vcpu initialization -- need to sync | |
2438 | * with other vCPUs. This particularly helps to keep | |
2439 | * kvm_clock stable after CPU hotplug | |
2440 | */ | |
2441 | synchronizing = true; | |
2442 | } else { | |
2443 | u64 tsc_exp = kvm->arch.last_tsc_write + | |
2444 | nsec_to_cycles(vcpu, elapsed); | |
2445 | u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL; | |
2446 | /* | |
2447 | * Special case: TSC write with a small delta (1 second) | |
2448 | * of virtual cycle time against real time is | |
2449 | * interpreted as an attempt to synchronize the CPU. | |
2450 | */ | |
2451 | synchronizing = data < tsc_exp + tsc_hz && | |
2452 | data + tsc_hz > tsc_exp; | |
2453 | } | |
c5e8ec8e | 2454 | } |
f38e098f ZA |
2455 | |
2456 | /* | |
5d3cb0f6 ZA |
2457 | * For a reliable TSC, we can match TSC offsets, and for an unstable |
2458 | * TSC, we add elapsed time in this computation. We could let the | |
2459 | * compensation code attempt to catch up if we fall behind, but | |
2460 | * it's better to try to match offsets from the beginning. | |
2461 | */ | |
c5e8ec8e | 2462 | if (synchronizing && |
5d3cb0f6 | 2463 | vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) { |
b0c39dc6 | 2464 | if (!kvm_check_tsc_unstable()) { |
e26101b1 | 2465 | offset = kvm->arch.cur_tsc_offset; |
f38e098f | 2466 | } else { |
857e4099 | 2467 | u64 delta = nsec_to_cycles(vcpu, elapsed); |
5d3cb0f6 | 2468 | data += delta; |
9b399dfd | 2469 | offset = kvm_compute_l1_tsc_offset(vcpu, data); |
f38e098f | 2470 | } |
b48aa97e | 2471 | matched = true; |
0d3da0d2 | 2472 | already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation); |
e26101b1 ZA |
2473 | } else { |
2474 | /* | |
2475 | * We split periods of matched TSC writes into generations. | |
2476 | * For each generation, we track the original measured | |
2477 | * nanosecond time, offset, and write, so if TSCs are in | |
2478 | * sync, we can match exact offset, and if not, we can match | |
4a969980 | 2479 | * exact software computation in compute_guest_tsc() |
e26101b1 ZA |
2480 | * |
2481 | * These values are tracked in kvm->arch.cur_xxx variables. | |
2482 | */ | |
2483 | kvm->arch.cur_tsc_generation++; | |
2484 | kvm->arch.cur_tsc_nsec = ns; | |
2485 | kvm->arch.cur_tsc_write = data; | |
2486 | kvm->arch.cur_tsc_offset = offset; | |
b48aa97e | 2487 | matched = false; |
f38e098f | 2488 | } |
e26101b1 ZA |
2489 | |
2490 | /* | |
2491 | * We also track th most recent recorded KHZ, write and time to | |
2492 | * allow the matching interval to be extended at each write. | |
2493 | */ | |
f38e098f ZA |
2494 | kvm->arch.last_tsc_nsec = ns; |
2495 | kvm->arch.last_tsc_write = data; | |
5d3cb0f6 | 2496 | kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz; |
99e3e30a | 2497 | |
b183aa58 | 2498 | vcpu->arch.last_guest_tsc = data; |
e26101b1 ZA |
2499 | |
2500 | /* Keep track of which generation this VCPU has synchronized to */ | |
2501 | vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation; | |
2502 | vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec; | |
2503 | vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write; | |
2504 | ||
a545ab6a | 2505 | kvm_vcpu_write_tsc_offset(vcpu, offset); |
e26101b1 | 2506 | raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags); |
b48aa97e | 2507 | |
a83829f5 | 2508 | spin_lock_irqsave(&kvm->arch.pvclock_gtod_sync_lock, flags); |
0d3da0d2 | 2509 | if (!matched) { |
b48aa97e | 2510 | kvm->arch.nr_vcpus_matched_tsc = 0; |
0d3da0d2 TG |
2511 | } else if (!already_matched) { |
2512 | kvm->arch.nr_vcpus_matched_tsc++; | |
2513 | } | |
b48aa97e MT |
2514 | |
2515 | kvm_track_tsc_matching(vcpu); | |
a83829f5 | 2516 | spin_unlock_irqrestore(&kvm->arch.pvclock_gtod_sync_lock, flags); |
99e3e30a | 2517 | } |
e26101b1 | 2518 | |
58ea6767 HZ |
2519 | static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu, |
2520 | s64 adjustment) | |
2521 | { | |
56ba77a4 | 2522 | u64 tsc_offset = vcpu->arch.l1_tsc_offset; |
326e7425 | 2523 | kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment); |
58ea6767 HZ |
2524 | } |
2525 | ||
2526 | static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment) | |
2527 | { | |
805d705f | 2528 | if (vcpu->arch.l1_tsc_scaling_ratio != kvm_default_tsc_scaling_ratio) |
58ea6767 | 2529 | WARN_ON(adjustment < 0); |
fe3eb504 IS |
2530 | adjustment = kvm_scale_tsc(vcpu, (u64) adjustment, |
2531 | vcpu->arch.l1_tsc_scaling_ratio); | |
ea26e4ec | 2532 | adjust_tsc_offset_guest(vcpu, adjustment); |
58ea6767 HZ |
2533 | } |
2534 | ||
d828199e MT |
2535 | #ifdef CONFIG_X86_64 |
2536 | ||
a5a1d1c2 | 2537 | static u64 read_tsc(void) |
d828199e | 2538 | { |
a5a1d1c2 | 2539 | u64 ret = (u64)rdtsc_ordered(); |
03b9730b | 2540 | u64 last = pvclock_gtod_data.clock.cycle_last; |
d828199e MT |
2541 | |
2542 | if (likely(ret >= last)) | |
2543 | return ret; | |
2544 | ||
2545 | /* | |
2546 | * GCC likes to generate cmov here, but this branch is extremely | |
6a6256f9 | 2547 | * predictable (it's just a function of time and the likely is |
d828199e MT |
2548 | * very likely) and there's a data dependence, so force GCC |
2549 | * to generate a branch instead. I don't barrier() because | |
2550 | * we don't actually need a barrier, and if this function | |
2551 | * ever gets inlined it will generate worse code. | |
2552 | */ | |
2553 | asm volatile (""); | |
2554 | return last; | |
2555 | } | |
2556 | ||
53fafdbb MT |
2557 | static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp, |
2558 | int *mode) | |
d828199e MT |
2559 | { |
2560 | long v; | |
b0c39dc6 VK |
2561 | u64 tsc_pg_val; |
2562 | ||
53fafdbb | 2563 | switch (clock->vclock_mode) { |
b95a8a27 | 2564 | case VDSO_CLOCKMODE_HVCLOCK: |
b0c39dc6 VK |
2565 | tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(), |
2566 | tsc_timestamp); | |
2567 | if (tsc_pg_val != U64_MAX) { | |
2568 | /* TSC page valid */ | |
b95a8a27 | 2569 | *mode = VDSO_CLOCKMODE_HVCLOCK; |
53fafdbb MT |
2570 | v = (tsc_pg_val - clock->cycle_last) & |
2571 | clock->mask; | |
b0c39dc6 VK |
2572 | } else { |
2573 | /* TSC page invalid */ | |
b95a8a27 | 2574 | *mode = VDSO_CLOCKMODE_NONE; |
b0c39dc6 VK |
2575 | } |
2576 | break; | |
b95a8a27 TG |
2577 | case VDSO_CLOCKMODE_TSC: |
2578 | *mode = VDSO_CLOCKMODE_TSC; | |
b0c39dc6 | 2579 | *tsc_timestamp = read_tsc(); |
53fafdbb MT |
2580 | v = (*tsc_timestamp - clock->cycle_last) & |
2581 | clock->mask; | |
b0c39dc6 VK |
2582 | break; |
2583 | default: | |
b95a8a27 | 2584 | *mode = VDSO_CLOCKMODE_NONE; |
b0c39dc6 | 2585 | } |
d828199e | 2586 | |
b95a8a27 | 2587 | if (*mode == VDSO_CLOCKMODE_NONE) |
b0c39dc6 | 2588 | *tsc_timestamp = v = 0; |
d828199e | 2589 | |
53fafdbb | 2590 | return v * clock->mult; |
d828199e MT |
2591 | } |
2592 | ||
53fafdbb | 2593 | static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp) |
d828199e | 2594 | { |
cbcf2dd3 | 2595 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; |
d828199e | 2596 | unsigned long seq; |
d828199e | 2597 | int mode; |
cbcf2dd3 | 2598 | u64 ns; |
d828199e | 2599 | |
d828199e MT |
2600 | do { |
2601 | seq = read_seqcount_begin(>od->seq); | |
917f9475 | 2602 | ns = gtod->raw_clock.base_cycles; |
53fafdbb | 2603 | ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode); |
917f9475 PB |
2604 | ns >>= gtod->raw_clock.shift; |
2605 | ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot)); | |
d828199e | 2606 | } while (unlikely(read_seqcount_retry(>od->seq, seq))); |
cbcf2dd3 | 2607 | *t = ns; |
d828199e MT |
2608 | |
2609 | return mode; | |
2610 | } | |
2611 | ||
899a31f5 | 2612 | static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp) |
55dd00a7 MT |
2613 | { |
2614 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; | |
2615 | unsigned long seq; | |
2616 | int mode; | |
2617 | u64 ns; | |
2618 | ||
2619 | do { | |
2620 | seq = read_seqcount_begin(>od->seq); | |
55dd00a7 | 2621 | ts->tv_sec = gtod->wall_time_sec; |
917f9475 | 2622 | ns = gtod->clock.base_cycles; |
53fafdbb | 2623 | ns += vgettsc(>od->clock, tsc_timestamp, &mode); |
55dd00a7 MT |
2624 | ns >>= gtod->clock.shift; |
2625 | } while (unlikely(read_seqcount_retry(>od->seq, seq))); | |
2626 | ||
2627 | ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns); | |
2628 | ts->tv_nsec = ns; | |
2629 | ||
2630 | return mode; | |
2631 | } | |
2632 | ||
b0c39dc6 VK |
2633 | /* returns true if host is using TSC based clocksource */ |
2634 | static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp) | |
d828199e | 2635 | { |
d828199e | 2636 | /* checked again under seqlock below */ |
b0c39dc6 | 2637 | if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode)) |
d828199e MT |
2638 | return false; |
2639 | ||
53fafdbb | 2640 | return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns, |
b0c39dc6 | 2641 | tsc_timestamp)); |
d828199e | 2642 | } |
55dd00a7 | 2643 | |
b0c39dc6 | 2644 | /* returns true if host is using TSC based clocksource */ |
899a31f5 | 2645 | static bool kvm_get_walltime_and_clockread(struct timespec64 *ts, |
b0c39dc6 | 2646 | u64 *tsc_timestamp) |
55dd00a7 MT |
2647 | { |
2648 | /* checked again under seqlock below */ | |
b0c39dc6 | 2649 | if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode)) |
55dd00a7 MT |
2650 | return false; |
2651 | ||
b0c39dc6 | 2652 | return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp)); |
55dd00a7 | 2653 | } |
d828199e MT |
2654 | #endif |
2655 | ||
2656 | /* | |
2657 | * | |
b48aa97e MT |
2658 | * Assuming a stable TSC across physical CPUS, and a stable TSC |
2659 | * across virtual CPUs, the following condition is possible. | |
2660 | * Each numbered line represents an event visible to both | |
d828199e MT |
2661 | * CPUs at the next numbered event. |
2662 | * | |
2663 | * "timespecX" represents host monotonic time. "tscX" represents | |
2664 | * RDTSC value. | |
2665 | * | |
2666 | * VCPU0 on CPU0 | VCPU1 on CPU1 | |
2667 | * | |
2668 | * 1. read timespec0,tsc0 | |
2669 | * 2. | timespec1 = timespec0 + N | |
2670 | * | tsc1 = tsc0 + M | |
2671 | * 3. transition to guest | transition to guest | |
2672 | * 4. ret0 = timespec0 + (rdtsc - tsc0) | | |
2673 | * 5. | ret1 = timespec1 + (rdtsc - tsc1) | |
2674 | * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M)) | |
2675 | * | |
2676 | * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity: | |
2677 | * | |
2678 | * - ret0 < ret1 | |
2679 | * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M)) | |
2680 | * ... | |
2681 | * - 0 < N - M => M < N | |
2682 | * | |
2683 | * That is, when timespec0 != timespec1, M < N. Unfortunately that is not | |
2684 | * always the case (the difference between two distinct xtime instances | |
2685 | * might be smaller then the difference between corresponding TSC reads, | |
2686 | * when updating guest vcpus pvclock areas). | |
2687 | * | |
2688 | * To avoid that problem, do not allow visibility of distinct | |
2689 | * system_timestamp/tsc_timestamp values simultaneously: use a master | |
2690 | * copy of host monotonic time values. Update that master copy | |
2691 | * in lockstep. | |
2692 | * | |
b48aa97e | 2693 | * Rely on synchronization of host TSCs and guest TSCs for monotonicity. |
d828199e MT |
2694 | * |
2695 | */ | |
2696 | ||
2697 | static void pvclock_update_vm_gtod_copy(struct kvm *kvm) | |
2698 | { | |
2699 | #ifdef CONFIG_X86_64 | |
2700 | struct kvm_arch *ka = &kvm->arch; | |
2701 | int vclock_mode; | |
b48aa97e MT |
2702 | bool host_tsc_clocksource, vcpus_matched; |
2703 | ||
2704 | vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 == | |
2705 | atomic_read(&kvm->online_vcpus)); | |
d828199e MT |
2706 | |
2707 | /* | |
2708 | * If the host uses TSC clock, then passthrough TSC as stable | |
2709 | * to the guest. | |
2710 | */ | |
b48aa97e | 2711 | host_tsc_clocksource = kvm_get_time_and_clockread( |
d828199e MT |
2712 | &ka->master_kernel_ns, |
2713 | &ka->master_cycle_now); | |
2714 | ||
16a96021 | 2715 | ka->use_master_clock = host_tsc_clocksource && vcpus_matched |
a826faf1 | 2716 | && !ka->backwards_tsc_observed |
54750f2c | 2717 | && !ka->boot_vcpu_runs_old_kvmclock; |
b48aa97e | 2718 | |
d828199e MT |
2719 | if (ka->use_master_clock) |
2720 | atomic_set(&kvm_guest_has_master_clock, 1); | |
2721 | ||
2722 | vclock_mode = pvclock_gtod_data.clock.vclock_mode; | |
b48aa97e MT |
2723 | trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode, |
2724 | vcpus_matched); | |
d828199e MT |
2725 | #endif |
2726 | } | |
2727 | ||
2860c4b1 PB |
2728 | void kvm_make_mclock_inprogress_request(struct kvm *kvm) |
2729 | { | |
2730 | kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS); | |
2731 | } | |
2732 | ||
2e762ff7 MT |
2733 | static void kvm_gen_update_masterclock(struct kvm *kvm) |
2734 | { | |
2735 | #ifdef CONFIG_X86_64 | |
2736 | int i; | |
2737 | struct kvm_vcpu *vcpu; | |
2738 | struct kvm_arch *ka = &kvm->arch; | |
a83829f5 | 2739 | unsigned long flags; |
2e762ff7 | 2740 | |
e880c6ea VK |
2741 | kvm_hv_invalidate_tsc_page(kvm); |
2742 | ||
2e762ff7 | 2743 | kvm_make_mclock_inprogress_request(kvm); |
c2c647f9 | 2744 | |
2e762ff7 | 2745 | /* no guest entries from this point */ |
a83829f5 | 2746 | spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags); |
2e762ff7 | 2747 | pvclock_update_vm_gtod_copy(kvm); |
a83829f5 | 2748 | spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags); |
2e762ff7 MT |
2749 | |
2750 | kvm_for_each_vcpu(i, vcpu, kvm) | |
105b21bb | 2751 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
2e762ff7 MT |
2752 | |
2753 | /* guest entries allowed */ | |
2754 | kvm_for_each_vcpu(i, vcpu, kvm) | |
72875d8a | 2755 | kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu); |
2e762ff7 MT |
2756 | #endif |
2757 | } | |
2758 | ||
e891a32e | 2759 | u64 get_kvmclock_ns(struct kvm *kvm) |
108b249c | 2760 | { |
108b249c | 2761 | struct kvm_arch *ka = &kvm->arch; |
8b953440 | 2762 | struct pvclock_vcpu_time_info hv_clock; |
a83829f5 | 2763 | unsigned long flags; |
e2c2206a | 2764 | u64 ret; |
108b249c | 2765 | |
a83829f5 | 2766 | spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags); |
8b953440 | 2767 | if (!ka->use_master_clock) { |
a83829f5 | 2768 | spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags); |
8171cd68 | 2769 | return get_kvmclock_base_ns() + ka->kvmclock_offset; |
108b249c PB |
2770 | } |
2771 | ||
8b953440 PB |
2772 | hv_clock.tsc_timestamp = ka->master_cycle_now; |
2773 | hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset; | |
a83829f5 | 2774 | spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags); |
8b953440 | 2775 | |
e2c2206a WL |
2776 | /* both __this_cpu_read() and rdtsc() should be on the same cpu */ |
2777 | get_cpu(); | |
2778 | ||
e70b57a6 WL |
2779 | if (__this_cpu_read(cpu_tsc_khz)) { |
2780 | kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL, | |
2781 | &hv_clock.tsc_shift, | |
2782 | &hv_clock.tsc_to_system_mul); | |
2783 | ret = __pvclock_read_cycles(&hv_clock, rdtsc()); | |
2784 | } else | |
8171cd68 | 2785 | ret = get_kvmclock_base_ns() + ka->kvmclock_offset; |
e2c2206a WL |
2786 | |
2787 | put_cpu(); | |
2788 | ||
2789 | return ret; | |
108b249c PB |
2790 | } |
2791 | ||
aa096aa0 JM |
2792 | static void kvm_setup_pvclock_page(struct kvm_vcpu *v, |
2793 | struct gfn_to_hva_cache *cache, | |
2794 | unsigned int offset) | |
0d6dd2ff PB |
2795 | { |
2796 | struct kvm_vcpu_arch *vcpu = &v->arch; | |
2797 | struct pvclock_vcpu_time_info guest_hv_clock; | |
2798 | ||
aa096aa0 JM |
2799 | if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache, |
2800 | &guest_hv_clock, offset, sizeof(guest_hv_clock)))) | |
0d6dd2ff PB |
2801 | return; |
2802 | ||
2803 | /* This VCPU is paused, but it's legal for a guest to read another | |
2804 | * VCPU's kvmclock, so we really have to follow the specification where | |
2805 | * it says that version is odd if data is being modified, and even after | |
2806 | * it is consistent. | |
2807 | * | |
2808 | * Version field updates must be kept separate. This is because | |
2809 | * kvm_write_guest_cached might use a "rep movs" instruction, and | |
2810 | * writes within a string instruction are weakly ordered. So there | |
2811 | * are three writes overall. | |
2812 | * | |
2813 | * As a small optimization, only write the version field in the first | |
2814 | * and third write. The vcpu->pv_time cache is still valid, because the | |
2815 | * version field is the first in the struct. | |
2816 | */ | |
2817 | BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0); | |
2818 | ||
51c4b8bb LA |
2819 | if (guest_hv_clock.version & 1) |
2820 | ++guest_hv_clock.version; /* first time write, random junk */ | |
2821 | ||
0d6dd2ff | 2822 | vcpu->hv_clock.version = guest_hv_clock.version + 1; |
aa096aa0 JM |
2823 | kvm_write_guest_offset_cached(v->kvm, cache, |
2824 | &vcpu->hv_clock, offset, | |
2825 | sizeof(vcpu->hv_clock.version)); | |
0d6dd2ff PB |
2826 | |
2827 | smp_wmb(); | |
2828 | ||
2829 | /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */ | |
2830 | vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED); | |
2831 | ||
2832 | if (vcpu->pvclock_set_guest_stopped_request) { | |
2833 | vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED; | |
2834 | vcpu->pvclock_set_guest_stopped_request = false; | |
2835 | } | |
2836 | ||
2837 | trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock); | |
2838 | ||
aa096aa0 JM |
2839 | kvm_write_guest_offset_cached(v->kvm, cache, |
2840 | &vcpu->hv_clock, offset, | |
2841 | sizeof(vcpu->hv_clock)); | |
0d6dd2ff PB |
2842 | |
2843 | smp_wmb(); | |
2844 | ||
2845 | vcpu->hv_clock.version++; | |
aa096aa0 JM |
2846 | kvm_write_guest_offset_cached(v->kvm, cache, |
2847 | &vcpu->hv_clock, offset, | |
2848 | sizeof(vcpu->hv_clock.version)); | |
0d6dd2ff PB |
2849 | } |
2850 | ||
34c238a1 | 2851 | static int kvm_guest_time_update(struct kvm_vcpu *v) |
18068523 | 2852 | { |
78db6a50 | 2853 | unsigned long flags, tgt_tsc_khz; |
18068523 | 2854 | struct kvm_vcpu_arch *vcpu = &v->arch; |
d828199e | 2855 | struct kvm_arch *ka = &v->kvm->arch; |
f25e656d | 2856 | s64 kernel_ns; |
d828199e | 2857 | u64 tsc_timestamp, host_tsc; |
51d59c6b | 2858 | u8 pvclock_flags; |
d828199e MT |
2859 | bool use_master_clock; |
2860 | ||
2861 | kernel_ns = 0; | |
2862 | host_tsc = 0; | |
18068523 | 2863 | |
d828199e MT |
2864 | /* |
2865 | * If the host uses TSC clock, then passthrough TSC as stable | |
2866 | * to the guest. | |
2867 | */ | |
a83829f5 | 2868 | spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags); |
d828199e MT |
2869 | use_master_clock = ka->use_master_clock; |
2870 | if (use_master_clock) { | |
2871 | host_tsc = ka->master_cycle_now; | |
2872 | kernel_ns = ka->master_kernel_ns; | |
2873 | } | |
a83829f5 | 2874 | spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags); |
c09664bb MT |
2875 | |
2876 | /* Keep irq disabled to prevent changes to the clock */ | |
2877 | local_irq_save(flags); | |
78db6a50 PB |
2878 | tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz); |
2879 | if (unlikely(tgt_tsc_khz == 0)) { | |
c09664bb MT |
2880 | local_irq_restore(flags); |
2881 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, v); | |
2882 | return 1; | |
2883 | } | |
d828199e | 2884 | if (!use_master_clock) { |
4ea1636b | 2885 | host_tsc = rdtsc(); |
8171cd68 | 2886 | kernel_ns = get_kvmclock_base_ns(); |
d828199e MT |
2887 | } |
2888 | ||
4ba76538 | 2889 | tsc_timestamp = kvm_read_l1_tsc(v, host_tsc); |
d828199e | 2890 | |
c285545f ZA |
2891 | /* |
2892 | * We may have to catch up the TSC to match elapsed wall clock | |
2893 | * time for two reasons, even if kvmclock is used. | |
2894 | * 1) CPU could have been running below the maximum TSC rate | |
2895 | * 2) Broken TSC compensation resets the base at each VCPU | |
2896 | * entry to avoid unknown leaps of TSC even when running | |
2897 | * again on the same CPU. This may cause apparent elapsed | |
2898 | * time to disappear, and the guest to stand still or run | |
2899 | * very slowly. | |
2900 | */ | |
2901 | if (vcpu->tsc_catchup) { | |
2902 | u64 tsc = compute_guest_tsc(v, kernel_ns); | |
2903 | if (tsc > tsc_timestamp) { | |
f1e2b260 | 2904 | adjust_tsc_offset_guest(v, tsc - tsc_timestamp); |
c285545f ZA |
2905 | tsc_timestamp = tsc; |
2906 | } | |
50d0a0f9 GH |
2907 | } |
2908 | ||
18068523 GOC |
2909 | local_irq_restore(flags); |
2910 | ||
0d6dd2ff | 2911 | /* With all the info we got, fill in the values */ |
18068523 | 2912 | |
78db6a50 | 2913 | if (kvm_has_tsc_control) |
fe3eb504 IS |
2914 | tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz, |
2915 | v->arch.l1_tsc_scaling_ratio); | |
78db6a50 PB |
2916 | |
2917 | if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) { | |
3ae13faa | 2918 | kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL, |
5f4e3f88 ZA |
2919 | &vcpu->hv_clock.tsc_shift, |
2920 | &vcpu->hv_clock.tsc_to_system_mul); | |
78db6a50 | 2921 | vcpu->hw_tsc_khz = tgt_tsc_khz; |
8cfdc000 ZA |
2922 | } |
2923 | ||
1d5f066e | 2924 | vcpu->hv_clock.tsc_timestamp = tsc_timestamp; |
759379dd | 2925 | vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset; |
28e4639a | 2926 | vcpu->last_guest_tsc = tsc_timestamp; |
51d59c6b | 2927 | |
d828199e | 2928 | /* If the host uses TSC clocksource, then it is stable */ |
0d6dd2ff | 2929 | pvclock_flags = 0; |
d828199e MT |
2930 | if (use_master_clock) |
2931 | pvclock_flags |= PVCLOCK_TSC_STABLE_BIT; | |
2932 | ||
78c0337a MT |
2933 | vcpu->hv_clock.flags = pvclock_flags; |
2934 | ||
095cf55d | 2935 | if (vcpu->pv_time_enabled) |
aa096aa0 JM |
2936 | kvm_setup_pvclock_page(v, &vcpu->pv_time, 0); |
2937 | if (vcpu->xen.vcpu_info_set) | |
2938 | kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache, | |
2939 | offsetof(struct compat_vcpu_info, time)); | |
f2340cd9 JM |
2940 | if (vcpu->xen.vcpu_time_info_set) |
2941 | kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0); | |
095cf55d PB |
2942 | if (v == kvm_get_vcpu(v->kvm, 0)) |
2943 | kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock); | |
8cfdc000 | 2944 | return 0; |
c8076604 GH |
2945 | } |
2946 | ||
0061d53d MT |
2947 | /* |
2948 | * kvmclock updates which are isolated to a given vcpu, such as | |
2949 | * vcpu->cpu migration, should not allow system_timestamp from | |
2950 | * the rest of the vcpus to remain static. Otherwise ntp frequency | |
2951 | * correction applies to one vcpu's system_timestamp but not | |
2952 | * the others. | |
2953 | * | |
2954 | * So in those cases, request a kvmclock update for all vcpus. | |
7e44e449 AJ |
2955 | * We need to rate-limit these requests though, as they can |
2956 | * considerably slow guests that have a large number of vcpus. | |
2957 | * The time for a remote vcpu to update its kvmclock is bound | |
2958 | * by the delay we use to rate-limit the updates. | |
0061d53d MT |
2959 | */ |
2960 | ||
7e44e449 AJ |
2961 | #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100) |
2962 | ||
2963 | static void kvmclock_update_fn(struct work_struct *work) | |
0061d53d MT |
2964 | { |
2965 | int i; | |
7e44e449 AJ |
2966 | struct delayed_work *dwork = to_delayed_work(work); |
2967 | struct kvm_arch *ka = container_of(dwork, struct kvm_arch, | |
2968 | kvmclock_update_work); | |
2969 | struct kvm *kvm = container_of(ka, struct kvm, arch); | |
0061d53d MT |
2970 | struct kvm_vcpu *vcpu; |
2971 | ||
2972 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
105b21bb | 2973 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
0061d53d MT |
2974 | kvm_vcpu_kick(vcpu); |
2975 | } | |
2976 | } | |
2977 | ||
7e44e449 AJ |
2978 | static void kvm_gen_kvmclock_update(struct kvm_vcpu *v) |
2979 | { | |
2980 | struct kvm *kvm = v->kvm; | |
2981 | ||
105b21bb | 2982 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, v); |
7e44e449 AJ |
2983 | schedule_delayed_work(&kvm->arch.kvmclock_update_work, |
2984 | KVMCLOCK_UPDATE_DELAY); | |
2985 | } | |
2986 | ||
332967a3 AJ |
2987 | #define KVMCLOCK_SYNC_PERIOD (300 * HZ) |
2988 | ||
2989 | static void kvmclock_sync_fn(struct work_struct *work) | |
2990 | { | |
2991 | struct delayed_work *dwork = to_delayed_work(work); | |
2992 | struct kvm_arch *ka = container_of(dwork, struct kvm_arch, | |
2993 | kvmclock_sync_work); | |
2994 | struct kvm *kvm = container_of(ka, struct kvm, arch); | |
2995 | ||
630994b3 MT |
2996 | if (!kvmclock_periodic_sync) |
2997 | return; | |
2998 | ||
332967a3 AJ |
2999 | schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0); |
3000 | schedule_delayed_work(&kvm->arch.kvmclock_sync_work, | |
3001 | KVMCLOCK_SYNC_PERIOD); | |
3002 | } | |
3003 | ||
191c8137 BP |
3004 | /* |
3005 | * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP. | |
3006 | */ | |
3007 | static bool can_set_mci_status(struct kvm_vcpu *vcpu) | |
3008 | { | |
3009 | /* McStatusWrEn enabled? */ | |
23493d0a | 3010 | if (guest_cpuid_is_amd_or_hygon(vcpu)) |
191c8137 BP |
3011 | return !!(vcpu->arch.msr_hwcr & BIT_ULL(18)); |
3012 | ||
3013 | return false; | |
3014 | } | |
3015 | ||
9ffd986c | 3016 | static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
15c4a640 | 3017 | { |
890ca9ae HY |
3018 | u64 mcg_cap = vcpu->arch.mcg_cap; |
3019 | unsigned bank_num = mcg_cap & 0xff; | |
9ffd986c WL |
3020 | u32 msr = msr_info->index; |
3021 | u64 data = msr_info->data; | |
890ca9ae | 3022 | |
15c4a640 | 3023 | switch (msr) { |
15c4a640 | 3024 | case MSR_IA32_MCG_STATUS: |
890ca9ae | 3025 | vcpu->arch.mcg_status = data; |
15c4a640 | 3026 | break; |
c7ac679c | 3027 | case MSR_IA32_MCG_CTL: |
44883f01 PB |
3028 | if (!(mcg_cap & MCG_CTL_P) && |
3029 | (data || !msr_info->host_initiated)) | |
890ca9ae HY |
3030 | return 1; |
3031 | if (data != 0 && data != ~(u64)0) | |
44883f01 | 3032 | return 1; |
890ca9ae HY |
3033 | vcpu->arch.mcg_ctl = data; |
3034 | break; | |
3035 | default: | |
3036 | if (msr >= MSR_IA32_MC0_CTL && | |
81760dcc | 3037 | msr < MSR_IA32_MCx_CTL(bank_num)) { |
6ec4c5ee MP |
3038 | u32 offset = array_index_nospec( |
3039 | msr - MSR_IA32_MC0_CTL, | |
3040 | MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL); | |
3041 | ||
114be429 AP |
3042 | /* only 0 or all 1s can be written to IA32_MCi_CTL |
3043 | * some Linux kernels though clear bit 10 in bank 4 to | |
3044 | * workaround a BIOS/GART TBL issue on AMD K8s, ignore | |
3045 | * this to avoid an uncatched #GP in the guest | |
3046 | */ | |
890ca9ae | 3047 | if ((offset & 0x3) == 0 && |
114be429 | 3048 | data != 0 && (data | (1 << 10)) != ~(u64)0) |
890ca9ae | 3049 | return -1; |
191c8137 BP |
3050 | |
3051 | /* MCi_STATUS */ | |
9ffd986c | 3052 | if (!msr_info->host_initiated && |
191c8137 BP |
3053 | (offset & 0x3) == 1 && data != 0) { |
3054 | if (!can_set_mci_status(vcpu)) | |
3055 | return -1; | |
3056 | } | |
3057 | ||
890ca9ae HY |
3058 | vcpu->arch.mce_banks[offset] = data; |
3059 | break; | |
3060 | } | |
3061 | return 1; | |
3062 | } | |
3063 | return 0; | |
3064 | } | |
3065 | ||
2635b5c4 VK |
3066 | static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu) |
3067 | { | |
3068 | u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT; | |
3069 | ||
3070 | return (vcpu->arch.apf.msr_en_val & mask) == mask; | |
3071 | } | |
3072 | ||
344d9588 GN |
3073 | static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data) |
3074 | { | |
3075 | gpa_t gpa = data & ~0x3f; | |
3076 | ||
2635b5c4 VK |
3077 | /* Bits 4:5 are reserved, Should be zero */ |
3078 | if (data & 0x30) | |
344d9588 GN |
3079 | return 1; |
3080 | ||
66570e96 OU |
3081 | if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) && |
3082 | (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT)) | |
3083 | return 1; | |
3084 | ||
3085 | if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) && | |
3086 | (data & KVM_ASYNC_PF_DELIVERY_AS_INT)) | |
3087 | return 1; | |
3088 | ||
9d3c447c | 3089 | if (!lapic_in_kernel(vcpu)) |
d831de17 | 3090 | return data ? 1 : 0; |
9d3c447c | 3091 | |
2635b5c4 | 3092 | vcpu->arch.apf.msr_en_val = data; |
344d9588 | 3093 | |
2635b5c4 | 3094 | if (!kvm_pv_async_pf_enabled(vcpu)) { |
344d9588 GN |
3095 | kvm_clear_async_pf_completion_queue(vcpu); |
3096 | kvm_async_pf_hash_reset(vcpu); | |
3097 | return 0; | |
3098 | } | |
3099 | ||
4e335d9e | 3100 | if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa, |
68fd66f1 | 3101 | sizeof(u64))) |
344d9588 GN |
3102 | return 1; |
3103 | ||
6adba527 | 3104 | vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS); |
52a5c155 | 3105 | vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT; |
2635b5c4 | 3106 | |
344d9588 | 3107 | kvm_async_pf_wakeup_all(vcpu); |
2635b5c4 VK |
3108 | |
3109 | return 0; | |
3110 | } | |
3111 | ||
3112 | static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data) | |
3113 | { | |
3114 | /* Bits 8-63 are reserved */ | |
3115 | if (data >> 8) | |
3116 | return 1; | |
3117 | ||
3118 | if (!lapic_in_kernel(vcpu)) | |
3119 | return 1; | |
3120 | ||
3121 | vcpu->arch.apf.msr_int_val = data; | |
3122 | ||
3123 | vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK; | |
3124 | ||
344d9588 GN |
3125 | return 0; |
3126 | } | |
3127 | ||
12f9a48f GC |
3128 | static void kvmclock_reset(struct kvm_vcpu *vcpu) |
3129 | { | |
0b79459b | 3130 | vcpu->arch.pv_time_enabled = false; |
49dedf0d | 3131 | vcpu->arch.time = 0; |
12f9a48f GC |
3132 | } |
3133 | ||
7780938c | 3134 | static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu) |
f38a7b75 WL |
3135 | { |
3136 | ++vcpu->stat.tlb_flush; | |
b3646477 | 3137 | static_call(kvm_x86_tlb_flush_all)(vcpu); |
f38a7b75 WL |
3138 | } |
3139 | ||
0baedd79 VK |
3140 | static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu) |
3141 | { | |
3142 | ++vcpu->stat.tlb_flush; | |
b53e84ee LJ |
3143 | |
3144 | if (!tdp_enabled) { | |
3145 | /* | |
3146 | * A TLB flush on behalf of the guest is equivalent to | |
3147 | * INVPCID(all), toggling CR4.PGE, etc., which requires | |
3148 | * a forced sync of the shadow page tables. Unload the | |
3149 | * entire MMU here and the subsequent load will sync the | |
3150 | * shadow page tables, and also flush the TLB. | |
3151 | */ | |
3152 | kvm_mmu_unload(vcpu); | |
3153 | return; | |
3154 | } | |
3155 | ||
b3646477 | 3156 | static_call(kvm_x86_tlb_flush_guest)(vcpu); |
0baedd79 VK |
3157 | } |
3158 | ||
c9aaa895 GC |
3159 | static void record_steal_time(struct kvm_vcpu *vcpu) |
3160 | { | |
b0431382 BO |
3161 | struct kvm_host_map map; |
3162 | struct kvm_steal_time *st; | |
3163 | ||
30b5c851 DW |
3164 | if (kvm_xen_msr_enabled(vcpu->kvm)) { |
3165 | kvm_xen_runstate_set_running(vcpu); | |
3166 | return; | |
3167 | } | |
3168 | ||
c9aaa895 GC |
3169 | if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED)) |
3170 | return; | |
3171 | ||
b0431382 BO |
3172 | /* -EAGAIN is returned in atomic context so we can just return. */ |
3173 | if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, | |
3174 | &map, &vcpu->arch.st.cache, false)) | |
c9aaa895 GC |
3175 | return; |
3176 | ||
b0431382 BO |
3177 | st = map.hva + |
3178 | offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS); | |
3179 | ||
f38a7b75 WL |
3180 | /* |
3181 | * Doing a TLB flush here, on the guest's behalf, can avoid | |
3182 | * expensive IPIs. | |
3183 | */ | |
66570e96 | 3184 | if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) { |
af3511ff LJ |
3185 | u8 st_preempted = xchg(&st->preempted, 0); |
3186 | ||
66570e96 | 3187 | trace_kvm_pv_tlb_flush(vcpu->vcpu_id, |
af3511ff LJ |
3188 | st_preempted & KVM_VCPU_FLUSH_TLB); |
3189 | if (st_preempted & KVM_VCPU_FLUSH_TLB) | |
66570e96 | 3190 | kvm_vcpu_flush_tlb_guest(vcpu); |
1eff0ada WL |
3191 | } else { |
3192 | st->preempted = 0; | |
66570e96 | 3193 | } |
0b9f6c46 | 3194 | |
a6bd811f | 3195 | vcpu->arch.st.preempted = 0; |
35f3fae1 | 3196 | |
b0431382 BO |
3197 | if (st->version & 1) |
3198 | st->version += 1; /* first time write, random junk */ | |
35f3fae1 | 3199 | |
b0431382 | 3200 | st->version += 1; |
35f3fae1 WL |
3201 | |
3202 | smp_wmb(); | |
3203 | ||
b0431382 | 3204 | st->steal += current->sched_info.run_delay - |
c54cdf14 LC |
3205 | vcpu->arch.st.last_steal; |
3206 | vcpu->arch.st.last_steal = current->sched_info.run_delay; | |
35f3fae1 | 3207 | |
35f3fae1 WL |
3208 | smp_wmb(); |
3209 | ||
b0431382 | 3210 | st->version += 1; |
c9aaa895 | 3211 | |
b0431382 | 3212 | kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false); |
c9aaa895 GC |
3213 | } |
3214 | ||
8fe8ab46 | 3215 | int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
15c4a640 | 3216 | { |
5753785f | 3217 | bool pr = false; |
8fe8ab46 WA |
3218 | u32 msr = msr_info->index; |
3219 | u64 data = msr_info->data; | |
5753785f | 3220 | |
1232f8e6 | 3221 | if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr) |
23200b7a | 3222 | return kvm_xen_write_hypercall_page(vcpu, data); |
1232f8e6 | 3223 | |
15c4a640 | 3224 | switch (msr) { |
2e32b719 | 3225 | case MSR_AMD64_NB_CFG: |
2e32b719 BP |
3226 | case MSR_IA32_UCODE_WRITE: |
3227 | case MSR_VM_HSAVE_PA: | |
3228 | case MSR_AMD64_PATCH_LOADER: | |
3229 | case MSR_AMD64_BU_CFG2: | |
405a353a | 3230 | case MSR_AMD64_DC_CFG: |
0e1b869f | 3231 | case MSR_F15H_EX_CFG: |
2e32b719 BP |
3232 | break; |
3233 | ||
518e7b94 WL |
3234 | case MSR_IA32_UCODE_REV: |
3235 | if (msr_info->host_initiated) | |
3236 | vcpu->arch.microcode_version = data; | |
3237 | break; | |
0cf9135b SC |
3238 | case MSR_IA32_ARCH_CAPABILITIES: |
3239 | if (!msr_info->host_initiated) | |
3240 | return 1; | |
3241 | vcpu->arch.arch_capabilities = data; | |
3242 | break; | |
d574c539 VK |
3243 | case MSR_IA32_PERF_CAPABILITIES: { |
3244 | struct kvm_msr_entry msr_ent = {.index = msr, .data = 0}; | |
3245 | ||
3246 | if (!msr_info->host_initiated) | |
3247 | return 1; | |
3248 | if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent)) | |
3249 | return 1; | |
3250 | if (data & ~msr_ent.data) | |
3251 | return 1; | |
3252 | ||
3253 | vcpu->arch.perf_capabilities = data; | |
3254 | ||
3255 | return 0; | |
3256 | } | |
15c4a640 | 3257 | case MSR_EFER: |
11988499 | 3258 | return set_efer(vcpu, msr_info); |
8f1589d9 AP |
3259 | case MSR_K7_HWCR: |
3260 | data &= ~(u64)0x40; /* ignore flush filter disable */ | |
82494028 | 3261 | data &= ~(u64)0x100; /* ignore ignne emulation enable */ |
a223c313 | 3262 | data &= ~(u64)0x8; /* ignore TLB cache disable */ |
191c8137 BP |
3263 | |
3264 | /* Handle McStatusWrEn */ | |
3265 | if (data == BIT_ULL(18)) { | |
3266 | vcpu->arch.msr_hwcr = data; | |
3267 | } else if (data != 0) { | |
a737f256 CD |
3268 | vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n", |
3269 | data); | |
8f1589d9 AP |
3270 | return 1; |
3271 | } | |
15c4a640 | 3272 | break; |
f7c6d140 AP |
3273 | case MSR_FAM10H_MMIO_CONF_BASE: |
3274 | if (data != 0) { | |
a737f256 CD |
3275 | vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: " |
3276 | "0x%llx\n", data); | |
f7c6d140 AP |
3277 | return 1; |
3278 | } | |
15c4a640 | 3279 | break; |
9ba075a6 | 3280 | case 0x200 ... 0x2ff: |
ff53604b | 3281 | return kvm_mtrr_set_msr(vcpu, msr, data); |
15c4a640 | 3282 | case MSR_IA32_APICBASE: |
58cb628d | 3283 | return kvm_set_apic_base(vcpu, msr_info); |
bf10bd0b | 3284 | case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff: |
0105d1a5 | 3285 | return kvm_x2apic_msr_write(vcpu, msr, data); |
09141ec0 | 3286 | case MSR_IA32_TSC_DEADLINE: |
a3e06bbe LJ |
3287 | kvm_set_lapic_tscdeadline_msr(vcpu, data); |
3288 | break; | |
ba904635 | 3289 | case MSR_IA32_TSC_ADJUST: |
d6321d49 | 3290 | if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) { |
ba904635 | 3291 | if (!msr_info->host_initiated) { |
d913b904 | 3292 | s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr; |
d7add054 | 3293 | adjust_tsc_offset_guest(vcpu, adj); |
ba904635 WA |
3294 | } |
3295 | vcpu->arch.ia32_tsc_adjust_msr = data; | |
3296 | } | |
3297 | break; | |
15c4a640 | 3298 | case MSR_IA32_MISC_ENABLE: |
511a8556 WL |
3299 | if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) && |
3300 | ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) { | |
3301 | if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3)) | |
3302 | return 1; | |
3303 | vcpu->arch.ia32_misc_enable_msr = data; | |
aedbaf4f | 3304 | kvm_update_cpuid_runtime(vcpu); |
511a8556 WL |
3305 | } else { |
3306 | vcpu->arch.ia32_misc_enable_msr = data; | |
3307 | } | |
15c4a640 | 3308 | break; |
64d60670 PB |
3309 | case MSR_IA32_SMBASE: |
3310 | if (!msr_info->host_initiated) | |
3311 | return 1; | |
3312 | vcpu->arch.smbase = data; | |
3313 | break; | |
73f624f4 PB |
3314 | case MSR_IA32_POWER_CTL: |
3315 | vcpu->arch.msr_ia32_power_ctl = data; | |
3316 | break; | |
dd259935 | 3317 | case MSR_IA32_TSC: |
0c899c25 PB |
3318 | if (msr_info->host_initiated) { |
3319 | kvm_synchronize_tsc(vcpu, data); | |
3320 | } else { | |
9b399dfd | 3321 | u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset; |
0c899c25 PB |
3322 | adjust_tsc_offset_guest(vcpu, adj); |
3323 | vcpu->arch.ia32_tsc_adjust_msr += adj; | |
3324 | } | |
dd259935 | 3325 | break; |
864e2ab2 AL |
3326 | case MSR_IA32_XSS: |
3327 | if (!msr_info->host_initiated && | |
3328 | !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES)) | |
3329 | return 1; | |
3330 | /* | |
a1bead2a SC |
3331 | * KVM supports exposing PT to the guest, but does not support |
3332 | * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than | |
3333 | * XSAVES/XRSTORS to save/restore PT MSRs. | |
864e2ab2 | 3334 | */ |
408e9a31 | 3335 | if (data & ~supported_xss) |
864e2ab2 AL |
3336 | return 1; |
3337 | vcpu->arch.ia32_xss = data; | |
3338 | break; | |
52797bf9 LA |
3339 | case MSR_SMI_COUNT: |
3340 | if (!msr_info->host_initiated) | |
3341 | return 1; | |
3342 | vcpu->arch.smi_count = data; | |
3343 | break; | |
11c6bffa | 3344 | case MSR_KVM_WALL_CLOCK_NEW: |
66570e96 OU |
3345 | if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2)) |
3346 | return 1; | |
3347 | ||
629b5348 JM |
3348 | vcpu->kvm->arch.wall_clock = data; |
3349 | kvm_write_wall_clock(vcpu->kvm, data, 0); | |
66570e96 | 3350 | break; |
18068523 | 3351 | case MSR_KVM_WALL_CLOCK: |
66570e96 OU |
3352 | if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE)) |
3353 | return 1; | |
3354 | ||
629b5348 JM |
3355 | vcpu->kvm->arch.wall_clock = data; |
3356 | kvm_write_wall_clock(vcpu->kvm, data, 0); | |
18068523 | 3357 | break; |
11c6bffa | 3358 | case MSR_KVM_SYSTEM_TIME_NEW: |
66570e96 OU |
3359 | if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2)) |
3360 | return 1; | |
3361 | ||
5b9bb0eb OU |
3362 | kvm_write_system_time(vcpu, data, false, msr_info->host_initiated); |
3363 | break; | |
3364 | case MSR_KVM_SYSTEM_TIME: | |
66570e96 OU |
3365 | if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE)) |
3366 | return 1; | |
3367 | ||
3368 | kvm_write_system_time(vcpu, data, true, msr_info->host_initiated); | |
18068523 | 3369 | break; |
344d9588 | 3370 | case MSR_KVM_ASYNC_PF_EN: |
66570e96 OU |
3371 | if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF)) |
3372 | return 1; | |
3373 | ||
344d9588 GN |
3374 | if (kvm_pv_enable_async_pf(vcpu, data)) |
3375 | return 1; | |
3376 | break; | |
2635b5c4 | 3377 | case MSR_KVM_ASYNC_PF_INT: |
66570e96 OU |
3378 | if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT)) |
3379 | return 1; | |
3380 | ||
2635b5c4 VK |
3381 | if (kvm_pv_enable_async_pf_int(vcpu, data)) |
3382 | return 1; | |
3383 | break; | |
557a961a | 3384 | case MSR_KVM_ASYNC_PF_ACK: |
66570e96 OU |
3385 | if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF)) |
3386 | return 1; | |
557a961a VK |
3387 | if (data & 0x1) { |
3388 | vcpu->arch.apf.pageready_pending = false; | |
3389 | kvm_check_async_pf_completion(vcpu); | |
3390 | } | |
3391 | break; | |
c9aaa895 | 3392 | case MSR_KVM_STEAL_TIME: |
66570e96 OU |
3393 | if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME)) |
3394 | return 1; | |
c9aaa895 GC |
3395 | |
3396 | if (unlikely(!sched_info_on())) | |
3397 | return 1; | |
3398 | ||
3399 | if (data & KVM_STEAL_RESERVED_MASK) | |
3400 | return 1; | |
3401 | ||
c9aaa895 GC |
3402 | vcpu->arch.st.msr_val = data; |
3403 | ||
3404 | if (!(data & KVM_MSR_ENABLED)) | |
3405 | break; | |
3406 | ||
c9aaa895 GC |
3407 | kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu); |
3408 | ||
3409 | break; | |
ae7a2a3f | 3410 | case MSR_KVM_PV_EOI_EN: |
66570e96 OU |
3411 | if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI)) |
3412 | return 1; | |
3413 | ||
72bbf935 | 3414 | if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8))) |
ae7a2a3f MT |
3415 | return 1; |
3416 | break; | |
c9aaa895 | 3417 | |
2d5ba19b | 3418 | case MSR_KVM_POLL_CONTROL: |
66570e96 OU |
3419 | if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL)) |
3420 | return 1; | |
3421 | ||
2d5ba19b MT |
3422 | /* only enable bit supported */ |
3423 | if (data & (-1ULL << 1)) | |
3424 | return 1; | |
3425 | ||
3426 | vcpu->arch.msr_kvm_poll_control = data; | |
3427 | break; | |
3428 | ||
890ca9ae HY |
3429 | case MSR_IA32_MCG_CTL: |
3430 | case MSR_IA32_MCG_STATUS: | |
81760dcc | 3431 | case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1: |
9ffd986c | 3432 | return set_msr_mce(vcpu, msr_info); |
71db6023 | 3433 | |
6912ac32 WH |
3434 | case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3: |
3435 | case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1: | |
df561f66 GS |
3436 | pr = true; |
3437 | fallthrough; | |
6912ac32 WH |
3438 | case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3: |
3439 | case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1: | |
c6702c9d | 3440 | if (kvm_pmu_is_valid_msr(vcpu, msr)) |
afd80d85 | 3441 | return kvm_pmu_set_msr(vcpu, msr_info); |
5753785f GN |
3442 | |
3443 | if (pr || data != 0) | |
a737f256 CD |
3444 | vcpu_unimpl(vcpu, "disabled perfctr wrmsr: " |
3445 | "0x%x data 0x%llx\n", msr, data); | |
5753785f | 3446 | break; |
84e0cefa JS |
3447 | case MSR_K7_CLK_CTL: |
3448 | /* | |
3449 | * Ignore all writes to this no longer documented MSR. | |
3450 | * Writes are only relevant for old K7 processors, | |
3451 | * all pre-dating SVM, but a recommended workaround from | |
4a969980 | 3452 | * AMD for these chips. It is possible to specify the |
84e0cefa JS |
3453 | * affected processor models on the command line, hence |
3454 | * the need to ignore the workaround. | |
3455 | */ | |
3456 | break; | |
55cd8e5a | 3457 | case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15: |
f97f5a56 JD |
3458 | case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER: |
3459 | case HV_X64_MSR_SYNDBG_OPTIONS: | |
e7d9513b AS |
3460 | case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4: |
3461 | case HV_X64_MSR_CRASH_CTL: | |
1f4b34f8 | 3462 | case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT: |
a2e164e7 VK |
3463 | case HV_X64_MSR_REENLIGHTENMENT_CONTROL: |
3464 | case HV_X64_MSR_TSC_EMULATION_CONTROL: | |
3465 | case HV_X64_MSR_TSC_EMULATION_STATUS: | |
e7d9513b AS |
3466 | return kvm_hv_set_msr_common(vcpu, msr, data, |
3467 | msr_info->host_initiated); | |
91c9c3ed | 3468 | case MSR_IA32_BBL_CR_CTL3: |
3469 | /* Drop writes to this legacy MSR -- see rdmsr | |
3470 | * counterpart for further detail. | |
3471 | */ | |
fab0aa3b EM |
3472 | if (report_ignored_msrs) |
3473 | vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n", | |
3474 | msr, data); | |
91c9c3ed | 3475 | break; |
2b036c6b | 3476 | case MSR_AMD64_OSVW_ID_LENGTH: |
d6321d49 | 3477 | if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW)) |
2b036c6b BO |
3478 | return 1; |
3479 | vcpu->arch.osvw.length = data; | |
3480 | break; | |
3481 | case MSR_AMD64_OSVW_STATUS: | |
d6321d49 | 3482 | if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW)) |
2b036c6b BO |
3483 | return 1; |
3484 | vcpu->arch.osvw.status = data; | |
3485 | break; | |
db2336a8 KH |
3486 | case MSR_PLATFORM_INFO: |
3487 | if (!msr_info->host_initiated || | |
db2336a8 KH |
3488 | (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) && |
3489 | cpuid_fault_enabled(vcpu))) | |
3490 | return 1; | |
3491 | vcpu->arch.msr_platform_info = data; | |
3492 | break; | |
3493 | case MSR_MISC_FEATURES_ENABLES: | |
3494 | if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT || | |
3495 | (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT && | |
3496 | !supports_cpuid_fault(vcpu))) | |
3497 | return 1; | |
3498 | vcpu->arch.msr_misc_features_enables = data; | |
3499 | break; | |
15c4a640 | 3500 | default: |
c6702c9d | 3501 | if (kvm_pmu_is_valid_msr(vcpu, msr)) |
afd80d85 | 3502 | return kvm_pmu_set_msr(vcpu, msr_info); |
6abe9c13 | 3503 | return KVM_MSR_RET_INVALID; |
15c4a640 CO |
3504 | } |
3505 | return 0; | |
3506 | } | |
3507 | EXPORT_SYMBOL_GPL(kvm_set_msr_common); | |
3508 | ||
44883f01 | 3509 | static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host) |
15c4a640 CO |
3510 | { |
3511 | u64 data; | |
890ca9ae HY |
3512 | u64 mcg_cap = vcpu->arch.mcg_cap; |
3513 | unsigned bank_num = mcg_cap & 0xff; | |
15c4a640 CO |
3514 | |
3515 | switch (msr) { | |
15c4a640 CO |
3516 | case MSR_IA32_P5_MC_ADDR: |
3517 | case MSR_IA32_P5_MC_TYPE: | |
890ca9ae HY |
3518 | data = 0; |
3519 | break; | |
15c4a640 | 3520 | case MSR_IA32_MCG_CAP: |
890ca9ae HY |
3521 | data = vcpu->arch.mcg_cap; |
3522 | break; | |
c7ac679c | 3523 | case MSR_IA32_MCG_CTL: |
44883f01 | 3524 | if (!(mcg_cap & MCG_CTL_P) && !host) |
890ca9ae HY |
3525 | return 1; |
3526 | data = vcpu->arch.mcg_ctl; | |
3527 | break; | |
3528 | case MSR_IA32_MCG_STATUS: | |
3529 | data = vcpu->arch.mcg_status; | |
3530 | break; | |
3531 | default: | |
3532 | if (msr >= MSR_IA32_MC0_CTL && | |
81760dcc | 3533 | msr < MSR_IA32_MCx_CTL(bank_num)) { |
6ec4c5ee MP |
3534 | u32 offset = array_index_nospec( |
3535 | msr - MSR_IA32_MC0_CTL, | |
3536 | MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL); | |
3537 | ||
890ca9ae HY |
3538 | data = vcpu->arch.mce_banks[offset]; |
3539 | break; | |
3540 | } | |
3541 | return 1; | |
3542 | } | |
3543 | *pdata = data; | |
3544 | return 0; | |
3545 | } | |
3546 | ||
609e36d3 | 3547 | int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
890ca9ae | 3548 | { |
609e36d3 | 3549 | switch (msr_info->index) { |
890ca9ae | 3550 | case MSR_IA32_PLATFORM_ID: |
15c4a640 | 3551 | case MSR_IA32_EBL_CR_POWERON: |
b5e2fec0 AG |
3552 | case MSR_IA32_LASTBRANCHFROMIP: |
3553 | case MSR_IA32_LASTBRANCHTOIP: | |
3554 | case MSR_IA32_LASTINTFROMIP: | |
3555 | case MSR_IA32_LASTINTTOIP: | |
60af2ecd | 3556 | case MSR_K8_SYSCFG: |
3afb1121 PB |
3557 | case MSR_K8_TSEG_ADDR: |
3558 | case MSR_K8_TSEG_MASK: | |
61a6bd67 | 3559 | case MSR_VM_HSAVE_PA: |
1fdbd48c | 3560 | case MSR_K8_INT_PENDING_MSG: |
c323c0e5 | 3561 | case MSR_AMD64_NB_CFG: |
f7c6d140 | 3562 | case MSR_FAM10H_MMIO_CONF_BASE: |
2e32b719 | 3563 | case MSR_AMD64_BU_CFG2: |
0c2df2a1 | 3564 | case MSR_IA32_PERF_CTL: |
405a353a | 3565 | case MSR_AMD64_DC_CFG: |
0e1b869f | 3566 | case MSR_F15H_EX_CFG: |
2ca1a06a VS |
3567 | /* |
3568 | * Intel Sandy Bridge CPUs must support the RAPL (running average power | |
3569 | * limit) MSRs. Just return 0, as we do not want to expose the host | |
3570 | * data here. Do not conditionalize this on CPUID, as KVM does not do | |
3571 | * so for existing CPU-specific MSRs. | |
3572 | */ | |
3573 | case MSR_RAPL_POWER_UNIT: | |
3574 | case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */ | |
3575 | case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */ | |
3576 | case MSR_PKG_ENERGY_STATUS: /* Total package */ | |
3577 | case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */ | |
609e36d3 | 3578 | msr_info->data = 0; |
15c4a640 | 3579 | break; |
c51eb52b | 3580 | case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5: |
c28fa560 VK |
3581 | if (kvm_pmu_is_valid_msr(vcpu, msr_info->index)) |
3582 | return kvm_pmu_get_msr(vcpu, msr_info); | |
3583 | if (!msr_info->host_initiated) | |
3584 | return 1; | |
3585 | msr_info->data = 0; | |
3586 | break; | |
6912ac32 WH |
3587 | case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3: |
3588 | case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3: | |
3589 | case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1: | |
3590 | case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1: | |
c6702c9d | 3591 | if (kvm_pmu_is_valid_msr(vcpu, msr_info->index)) |
cbd71758 | 3592 | return kvm_pmu_get_msr(vcpu, msr_info); |
609e36d3 | 3593 | msr_info->data = 0; |
5753785f | 3594 | break; |
742bc670 | 3595 | case MSR_IA32_UCODE_REV: |
518e7b94 | 3596 | msr_info->data = vcpu->arch.microcode_version; |
742bc670 | 3597 | break; |
0cf9135b SC |
3598 | case MSR_IA32_ARCH_CAPABILITIES: |
3599 | if (!msr_info->host_initiated && | |
3600 | !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES)) | |
3601 | return 1; | |
3602 | msr_info->data = vcpu->arch.arch_capabilities; | |
3603 | break; | |
d574c539 VK |
3604 | case MSR_IA32_PERF_CAPABILITIES: |
3605 | if (!msr_info->host_initiated && | |
3606 | !guest_cpuid_has(vcpu, X86_FEATURE_PDCM)) | |
3607 | return 1; | |
3608 | msr_info->data = vcpu->arch.perf_capabilities; | |
3609 | break; | |
73f624f4 PB |
3610 | case MSR_IA32_POWER_CTL: |
3611 | msr_info->data = vcpu->arch.msr_ia32_power_ctl; | |
3612 | break; | |
cc5b54dd ML |
3613 | case MSR_IA32_TSC: { |
3614 | /* | |
3615 | * Intel SDM states that MSR_IA32_TSC read adds the TSC offset | |
3616 | * even when not intercepted. AMD manual doesn't explicitly | |
3617 | * state this but appears to behave the same. | |
3618 | * | |
ee6fa053 | 3619 | * On userspace reads and writes, however, we unconditionally |
c0623f5e | 3620 | * return L1's TSC value to ensure backwards-compatible |
ee6fa053 | 3621 | * behavior for migration. |
cc5b54dd | 3622 | */ |
fe3eb504 | 3623 | u64 offset, ratio; |
cc5b54dd | 3624 | |
fe3eb504 IS |
3625 | if (msr_info->host_initiated) { |
3626 | offset = vcpu->arch.l1_tsc_offset; | |
3627 | ratio = vcpu->arch.l1_tsc_scaling_ratio; | |
3628 | } else { | |
3629 | offset = vcpu->arch.tsc_offset; | |
3630 | ratio = vcpu->arch.tsc_scaling_ratio; | |
3631 | } | |
3632 | ||
3633 | msr_info->data = kvm_scale_tsc(vcpu, rdtsc(), ratio) + offset; | |
dd259935 | 3634 | break; |
cc5b54dd | 3635 | } |
9ba075a6 | 3636 | case MSR_MTRRcap: |
9ba075a6 | 3637 | case 0x200 ... 0x2ff: |
ff53604b | 3638 | return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data); |
15c4a640 | 3639 | case 0xcd: /* fsb frequency */ |
609e36d3 | 3640 | msr_info->data = 3; |
15c4a640 | 3641 | break; |
7b914098 JS |
3642 | /* |
3643 | * MSR_EBC_FREQUENCY_ID | |
3644 | * Conservative value valid for even the basic CPU models. | |
3645 | * Models 0,1: 000 in bits 23:21 indicating a bus speed of | |
3646 | * 100MHz, model 2 000 in bits 18:16 indicating 100MHz, | |
3647 | * and 266MHz for model 3, or 4. Set Core Clock | |
3648 | * Frequency to System Bus Frequency Ratio to 1 (bits | |
3649 | * 31:24) even though these are only valid for CPU | |
3650 | * models > 2, however guests may end up dividing or | |
3651 | * multiplying by zero otherwise. | |
3652 | */ | |
3653 | case MSR_EBC_FREQUENCY_ID: | |
609e36d3 | 3654 | msr_info->data = 1 << 24; |
7b914098 | 3655 | break; |
15c4a640 | 3656 | case MSR_IA32_APICBASE: |
609e36d3 | 3657 | msr_info->data = kvm_get_apic_base(vcpu); |
15c4a640 | 3658 | break; |
bf10bd0b | 3659 | case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff: |
609e36d3 | 3660 | return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data); |
09141ec0 | 3661 | case MSR_IA32_TSC_DEADLINE: |
609e36d3 | 3662 | msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu); |
a3e06bbe | 3663 | break; |
ba904635 | 3664 | case MSR_IA32_TSC_ADJUST: |
609e36d3 | 3665 | msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr; |
ba904635 | 3666 | break; |
15c4a640 | 3667 | case MSR_IA32_MISC_ENABLE: |
609e36d3 | 3668 | msr_info->data = vcpu->arch.ia32_misc_enable_msr; |
15c4a640 | 3669 | break; |
64d60670 PB |
3670 | case MSR_IA32_SMBASE: |
3671 | if (!msr_info->host_initiated) | |
3672 | return 1; | |
3673 | msr_info->data = vcpu->arch.smbase; | |
15c4a640 | 3674 | break; |
52797bf9 LA |
3675 | case MSR_SMI_COUNT: |
3676 | msr_info->data = vcpu->arch.smi_count; | |
3677 | break; | |
847f0ad8 AG |
3678 | case MSR_IA32_PERF_STATUS: |
3679 | /* TSC increment by tick */ | |
609e36d3 | 3680 | msr_info->data = 1000ULL; |
847f0ad8 | 3681 | /* CPU multiplier */ |
b0996ae4 | 3682 | msr_info->data |= (((uint64_t)4ULL) << 40); |
847f0ad8 | 3683 | break; |
15c4a640 | 3684 | case MSR_EFER: |
609e36d3 | 3685 | msr_info->data = vcpu->arch.efer; |
15c4a640 | 3686 | break; |
18068523 | 3687 | case MSR_KVM_WALL_CLOCK: |
1930e5dd OU |
3688 | if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE)) |
3689 | return 1; | |
3690 | ||
3691 | msr_info->data = vcpu->kvm->arch.wall_clock; | |
3692 | break; | |
11c6bffa | 3693 | case MSR_KVM_WALL_CLOCK_NEW: |
1930e5dd OU |
3694 | if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2)) |
3695 | return 1; | |
3696 | ||
609e36d3 | 3697 | msr_info->data = vcpu->kvm->arch.wall_clock; |
18068523 GOC |
3698 | break; |
3699 | case MSR_KVM_SYSTEM_TIME: | |
1930e5dd OU |
3700 | if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE)) |
3701 | return 1; | |
3702 | ||
3703 | msr_info->data = vcpu->arch.time; | |
3704 | break; | |
11c6bffa | 3705 | case MSR_KVM_SYSTEM_TIME_NEW: |
1930e5dd OU |
3706 | if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2)) |
3707 | return 1; | |
3708 | ||
609e36d3 | 3709 | msr_info->data = vcpu->arch.time; |
18068523 | 3710 | break; |
344d9588 | 3711 | case MSR_KVM_ASYNC_PF_EN: |
1930e5dd OU |
3712 | if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF)) |
3713 | return 1; | |
3714 | ||
2635b5c4 VK |
3715 | msr_info->data = vcpu->arch.apf.msr_en_val; |
3716 | break; | |
3717 | case MSR_KVM_ASYNC_PF_INT: | |
1930e5dd OU |
3718 | if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT)) |
3719 | return 1; | |
3720 | ||
2635b5c4 | 3721 | msr_info->data = vcpu->arch.apf.msr_int_val; |
344d9588 | 3722 | break; |
557a961a | 3723 | case MSR_KVM_ASYNC_PF_ACK: |
1930e5dd OU |
3724 | if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF)) |
3725 | return 1; | |
3726 | ||
557a961a VK |
3727 | msr_info->data = 0; |
3728 | break; | |
c9aaa895 | 3729 | case MSR_KVM_STEAL_TIME: |
1930e5dd OU |
3730 | if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME)) |
3731 | return 1; | |
3732 | ||
609e36d3 | 3733 | msr_info->data = vcpu->arch.st.msr_val; |
c9aaa895 | 3734 | break; |
1d92128f | 3735 | case MSR_KVM_PV_EOI_EN: |
1930e5dd OU |
3736 | if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI)) |
3737 | return 1; | |
3738 | ||
609e36d3 | 3739 | msr_info->data = vcpu->arch.pv_eoi.msr_val; |
1d92128f | 3740 | break; |
2d5ba19b | 3741 | case MSR_KVM_POLL_CONTROL: |
1930e5dd OU |
3742 | if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL)) |
3743 | return 1; | |
3744 | ||
2d5ba19b MT |
3745 | msr_info->data = vcpu->arch.msr_kvm_poll_control; |
3746 | break; | |
890ca9ae HY |
3747 | case MSR_IA32_P5_MC_ADDR: |
3748 | case MSR_IA32_P5_MC_TYPE: | |
3749 | case MSR_IA32_MCG_CAP: | |
3750 | case MSR_IA32_MCG_CTL: | |
3751 | case MSR_IA32_MCG_STATUS: | |
81760dcc | 3752 | case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1: |
44883f01 PB |
3753 | return get_msr_mce(vcpu, msr_info->index, &msr_info->data, |
3754 | msr_info->host_initiated); | |
864e2ab2 AL |
3755 | case MSR_IA32_XSS: |
3756 | if (!msr_info->host_initiated && | |
3757 | !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES)) | |
3758 | return 1; | |
3759 | msr_info->data = vcpu->arch.ia32_xss; | |
3760 | break; | |
84e0cefa JS |
3761 | case MSR_K7_CLK_CTL: |
3762 | /* | |
3763 | * Provide expected ramp-up count for K7. All other | |
3764 | * are set to zero, indicating minimum divisors for | |
3765 | * every field. | |
3766 | * | |
3767 | * This prevents guest kernels on AMD host with CPU | |
3768 | * type 6, model 8 and higher from exploding due to | |
3769 | * the rdmsr failing. | |
3770 | */ | |
609e36d3 | 3771 | msr_info->data = 0x20000000; |
84e0cefa | 3772 | break; |
55cd8e5a | 3773 | case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15: |
f97f5a56 JD |
3774 | case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER: |
3775 | case HV_X64_MSR_SYNDBG_OPTIONS: | |
e7d9513b AS |
3776 | case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4: |
3777 | case HV_X64_MSR_CRASH_CTL: | |
1f4b34f8 | 3778 | case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT: |
a2e164e7 VK |
3779 | case HV_X64_MSR_REENLIGHTENMENT_CONTROL: |
3780 | case HV_X64_MSR_TSC_EMULATION_CONTROL: | |
3781 | case HV_X64_MSR_TSC_EMULATION_STATUS: | |
e83d5887 | 3782 | return kvm_hv_get_msr_common(vcpu, |
44883f01 PB |
3783 | msr_info->index, &msr_info->data, |
3784 | msr_info->host_initiated); | |
91c9c3ed | 3785 | case MSR_IA32_BBL_CR_CTL3: |
3786 | /* This legacy MSR exists but isn't fully documented in current | |
3787 | * silicon. It is however accessed by winxp in very narrow | |
3788 | * scenarios where it sets bit #19, itself documented as | |
3789 | * a "reserved" bit. Best effort attempt to source coherent | |
3790 | * read data here should the balance of the register be | |
3791 | * interpreted by the guest: | |
3792 | * | |
3793 | * L2 cache control register 3: 64GB range, 256KB size, | |
3794 | * enabled, latency 0x1, configured | |
3795 | */ | |
609e36d3 | 3796 | msr_info->data = 0xbe702111; |
91c9c3ed | 3797 | break; |
2b036c6b | 3798 | case MSR_AMD64_OSVW_ID_LENGTH: |
d6321d49 | 3799 | if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW)) |
2b036c6b | 3800 | return 1; |
609e36d3 | 3801 | msr_info->data = vcpu->arch.osvw.length; |
2b036c6b BO |
3802 | break; |
3803 | case MSR_AMD64_OSVW_STATUS: | |
d6321d49 | 3804 | if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW)) |
2b036c6b | 3805 | return 1; |
609e36d3 | 3806 | msr_info->data = vcpu->arch.osvw.status; |
2b036c6b | 3807 | break; |
db2336a8 | 3808 | case MSR_PLATFORM_INFO: |
6fbbde9a DS |
3809 | if (!msr_info->host_initiated && |
3810 | !vcpu->kvm->arch.guest_can_read_msr_platform_info) | |
3811 | return 1; | |
db2336a8 KH |
3812 | msr_info->data = vcpu->arch.msr_platform_info; |
3813 | break; | |
3814 | case MSR_MISC_FEATURES_ENABLES: | |
3815 | msr_info->data = vcpu->arch.msr_misc_features_enables; | |
3816 | break; | |
191c8137 BP |
3817 | case MSR_K7_HWCR: |
3818 | msr_info->data = vcpu->arch.msr_hwcr; | |
3819 | break; | |
15c4a640 | 3820 | default: |
c6702c9d | 3821 | if (kvm_pmu_is_valid_msr(vcpu, msr_info->index)) |
cbd71758 | 3822 | return kvm_pmu_get_msr(vcpu, msr_info); |
6abe9c13 | 3823 | return KVM_MSR_RET_INVALID; |
15c4a640 | 3824 | } |
15c4a640 CO |
3825 | return 0; |
3826 | } | |
3827 | EXPORT_SYMBOL_GPL(kvm_get_msr_common); | |
3828 | ||
313a3dc7 CO |
3829 | /* |
3830 | * Read or write a bunch of msrs. All parameters are kernel addresses. | |
3831 | * | |
3832 | * @return number of msrs set successfully. | |
3833 | */ | |
3834 | static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs, | |
3835 | struct kvm_msr_entry *entries, | |
3836 | int (*do_msr)(struct kvm_vcpu *vcpu, | |
3837 | unsigned index, u64 *data)) | |
3838 | { | |
801e459a | 3839 | int i; |
313a3dc7 | 3840 | |
313a3dc7 CO |
3841 | for (i = 0; i < msrs->nmsrs; ++i) |
3842 | if (do_msr(vcpu, entries[i].index, &entries[i].data)) | |
3843 | break; | |
3844 | ||
313a3dc7 CO |
3845 | return i; |
3846 | } | |
3847 | ||
3848 | /* | |
3849 | * Read or write a bunch of msrs. Parameters are user addresses. | |
3850 | * | |
3851 | * @return number of msrs set successfully. | |
3852 | */ | |
3853 | static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs, | |
3854 | int (*do_msr)(struct kvm_vcpu *vcpu, | |
3855 | unsigned index, u64 *data), | |
3856 | int writeback) | |
3857 | { | |
3858 | struct kvm_msrs msrs; | |
3859 | struct kvm_msr_entry *entries; | |
3860 | int r, n; | |
3861 | unsigned size; | |
3862 | ||
3863 | r = -EFAULT; | |
0e96f31e | 3864 | if (copy_from_user(&msrs, user_msrs, sizeof(msrs))) |
313a3dc7 CO |
3865 | goto out; |
3866 | ||
3867 | r = -E2BIG; | |
3868 | if (msrs.nmsrs >= MAX_IO_MSRS) | |
3869 | goto out; | |
3870 | ||
313a3dc7 | 3871 | size = sizeof(struct kvm_msr_entry) * msrs.nmsrs; |
ff5c2c03 SL |
3872 | entries = memdup_user(user_msrs->entries, size); |
3873 | if (IS_ERR(entries)) { | |
3874 | r = PTR_ERR(entries); | |
313a3dc7 | 3875 | goto out; |
ff5c2c03 | 3876 | } |
313a3dc7 CO |
3877 | |
3878 | r = n = __msr_io(vcpu, &msrs, entries, do_msr); | |
3879 | if (r < 0) | |
3880 | goto out_free; | |
3881 | ||
3882 | r = -EFAULT; | |
3883 | if (writeback && copy_to_user(user_msrs->entries, entries, size)) | |
3884 | goto out_free; | |
3885 | ||
3886 | r = n; | |
3887 | ||
3888 | out_free: | |
7a73c028 | 3889 | kfree(entries); |
313a3dc7 CO |
3890 | out: |
3891 | return r; | |
3892 | } | |
3893 | ||
4d5422ce WL |
3894 | static inline bool kvm_can_mwait_in_guest(void) |
3895 | { | |
3896 | return boot_cpu_has(X86_FEATURE_MWAIT) && | |
8e9b29b6 KA |
3897 | !boot_cpu_has_bug(X86_BUG_MONITOR) && |
3898 | boot_cpu_has(X86_FEATURE_ARAT); | |
4d5422ce WL |
3899 | } |
3900 | ||
c21d54f0 VK |
3901 | static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu, |
3902 | struct kvm_cpuid2 __user *cpuid_arg) | |
3903 | { | |
3904 | struct kvm_cpuid2 cpuid; | |
3905 | int r; | |
3906 | ||
3907 | r = -EFAULT; | |
3908 | if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid))) | |
3909 | return r; | |
3910 | ||
3911 | r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries); | |
3912 | if (r) | |
3913 | return r; | |
3914 | ||
3915 | r = -EFAULT; | |
3916 | if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid))) | |
3917 | return r; | |
3918 | ||
3919 | return 0; | |
3920 | } | |
3921 | ||
784aa3d7 | 3922 | int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) |
018d00d2 | 3923 | { |
4d5422ce | 3924 | int r = 0; |
018d00d2 ZX |
3925 | |
3926 | switch (ext) { | |
3927 | case KVM_CAP_IRQCHIP: | |
3928 | case KVM_CAP_HLT: | |
3929 | case KVM_CAP_MMU_SHADOW_CACHE_CONTROL: | |
018d00d2 | 3930 | case KVM_CAP_SET_TSS_ADDR: |
07716717 | 3931 | case KVM_CAP_EXT_CPUID: |
9c15bb1d | 3932 | case KVM_CAP_EXT_EMUL_CPUID: |
c8076604 | 3933 | case KVM_CAP_CLOCKSOURCE: |
7837699f | 3934 | case KVM_CAP_PIT: |
a28e4f5a | 3935 | case KVM_CAP_NOP_IO_DELAY: |
62d9f0db | 3936 | case KVM_CAP_MP_STATE: |
ed848624 | 3937 | case KVM_CAP_SYNC_MMU: |
a355c85c | 3938 | case KVM_CAP_USER_NMI: |
52d939a0 | 3939 | case KVM_CAP_REINJECT_CONTROL: |
4925663a | 3940 | case KVM_CAP_IRQ_INJECT_STATUS: |
d34e6b17 | 3941 | case KVM_CAP_IOEVENTFD: |
f848a5a8 | 3942 | case KVM_CAP_IOEVENTFD_NO_LENGTH: |
c5ff41ce | 3943 | case KVM_CAP_PIT2: |
e9f42757 | 3944 | case KVM_CAP_PIT_STATE2: |
b927a3ce | 3945 | case KVM_CAP_SET_IDENTITY_MAP_ADDR: |
3cfc3092 | 3946 | case KVM_CAP_VCPU_EVENTS: |
55cd8e5a | 3947 | case KVM_CAP_HYPERV: |
10388a07 | 3948 | case KVM_CAP_HYPERV_VAPIC: |
c25bc163 | 3949 | case KVM_CAP_HYPERV_SPIN: |
5c919412 | 3950 | case KVM_CAP_HYPERV_SYNIC: |
efc479e6 | 3951 | case KVM_CAP_HYPERV_SYNIC2: |
d3457c87 | 3952 | case KVM_CAP_HYPERV_VP_INDEX: |
faeb7833 | 3953 | case KVM_CAP_HYPERV_EVENTFD: |
c1aea919 | 3954 | case KVM_CAP_HYPERV_TLBFLUSH: |
214ff83d | 3955 | case KVM_CAP_HYPERV_SEND_IPI: |
2bc39970 | 3956 | case KVM_CAP_HYPERV_CPUID: |
c21d54f0 | 3957 | case KVM_CAP_SYS_HYPERV_CPUID: |
ab9f4ecb | 3958 | case KVM_CAP_PCI_SEGMENT: |
a1efbe77 | 3959 | case KVM_CAP_DEBUGREGS: |
d2be1651 | 3960 | case KVM_CAP_X86_ROBUST_SINGLESTEP: |
2d5b5a66 | 3961 | case KVM_CAP_XSAVE: |
344d9588 | 3962 | case KVM_CAP_ASYNC_PF: |
72de5fa4 | 3963 | case KVM_CAP_ASYNC_PF_INT: |
92a1f12d | 3964 | case KVM_CAP_GET_TSC_KHZ: |
1c0b28c2 | 3965 | case KVM_CAP_KVMCLOCK_CTRL: |
4d8b81ab | 3966 | case KVM_CAP_READONLY_MEM: |
5f66b620 | 3967 | case KVM_CAP_HYPERV_TIME: |
100943c5 | 3968 | case KVM_CAP_IOAPIC_POLARITY_IGNORED: |
defcf51f | 3969 | case KVM_CAP_TSC_DEADLINE_TIMER: |
90de4a18 | 3970 | case KVM_CAP_DISABLE_QUIRKS: |
d71ba788 | 3971 | case KVM_CAP_SET_BOOT_CPU_ID: |
49df6397 | 3972 | case KVM_CAP_SPLIT_IRQCHIP: |
460df4c1 | 3973 | case KVM_CAP_IMMEDIATE_EXIT: |
66bb8a06 | 3974 | case KVM_CAP_PMU_EVENT_FILTER: |
801e459a | 3975 | case KVM_CAP_GET_MSR_FEATURES: |
6fbbde9a | 3976 | case KVM_CAP_MSR_PLATFORM_INFO: |
c4f55198 | 3977 | case KVM_CAP_EXCEPTION_PAYLOAD: |
b9b2782c | 3978 | case KVM_CAP_SET_GUEST_DEBUG: |
1aa561b1 | 3979 | case KVM_CAP_LAST_CPU: |
1ae09954 | 3980 | case KVM_CAP_X86_USER_SPACE_MSR: |
1a155254 | 3981 | case KVM_CAP_X86_MSR_FILTER: |
66570e96 | 3982 | case KVM_CAP_ENFORCE_PV_FEATURE_CPUID: |
fe7e9488 SC |
3983 | #ifdef CONFIG_X86_SGX_KVM |
3984 | case KVM_CAP_SGX_ATTRIBUTE: | |
3985 | #endif | |
54526d1f | 3986 | case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM: |
018d00d2 ZX |
3987 | r = 1; |
3988 | break; | |
7e582ccb ML |
3989 | case KVM_CAP_SET_GUEST_DEBUG2: |
3990 | return KVM_GUESTDBG_VALID_MASK; | |
b59b153d | 3991 | #ifdef CONFIG_KVM_XEN |
23200b7a JM |
3992 | case KVM_CAP_XEN_HVM: |
3993 | r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR | | |
8d4e7e80 DW |
3994 | KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL | |
3995 | KVM_XEN_HVM_CONFIG_SHARED_INFO; | |
30b5c851 DW |
3996 | if (sched_info_on()) |
3997 | r |= KVM_XEN_HVM_CONFIG_RUNSTATE; | |
23200b7a | 3998 | break; |
b59b153d | 3999 | #endif |
01643c51 KH |
4000 | case KVM_CAP_SYNC_REGS: |
4001 | r = KVM_SYNC_X86_VALID_FIELDS; | |
4002 | break; | |
e3fd9a93 PB |
4003 | case KVM_CAP_ADJUST_CLOCK: |
4004 | r = KVM_CLOCK_TSC_STABLE; | |
4005 | break; | |
4d5422ce | 4006 | case KVM_CAP_X86_DISABLE_EXITS: |
b5170063 WL |
4007 | r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE | |
4008 | KVM_X86_DISABLE_EXITS_CSTATE; | |
4d5422ce WL |
4009 | if(kvm_can_mwait_in_guest()) |
4010 | r |= KVM_X86_DISABLE_EXITS_MWAIT; | |
668fffa3 | 4011 | break; |
6d396b55 PB |
4012 | case KVM_CAP_X86_SMM: |
4013 | /* SMBASE is usually relocated above 1M on modern chipsets, | |
4014 | * and SMM handlers might indeed rely on 4G segment limits, | |
4015 | * so do not report SMM to be available if real mode is | |
4016 | * emulated via vm86 mode. Still, do not go to great lengths | |
4017 | * to avoid userspace's usage of the feature, because it is a | |
4018 | * fringe case that is not enabled except via specific settings | |
4019 | * of the module parameters. | |
4020 | */ | |
b3646477 | 4021 | r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE); |
6d396b55 | 4022 | break; |
774ead3a | 4023 | case KVM_CAP_VAPIC: |
b3646477 | 4024 | r = !static_call(kvm_x86_cpu_has_accelerated_tpr)(); |
774ead3a | 4025 | break; |
f725230a | 4026 | case KVM_CAP_NR_VCPUS: |
8c3ba334 SL |
4027 | r = KVM_SOFT_MAX_VCPUS; |
4028 | break; | |
4029 | case KVM_CAP_MAX_VCPUS: | |
f725230a AK |
4030 | r = KVM_MAX_VCPUS; |
4031 | break; | |
a86cb413 TH |
4032 | case KVM_CAP_MAX_VCPU_ID: |
4033 | r = KVM_MAX_VCPU_ID; | |
4034 | break; | |
a68a6a72 MT |
4035 | case KVM_CAP_PV_MMU: /* obsolete */ |
4036 | r = 0; | |
2f333bcb | 4037 | break; |
890ca9ae HY |
4038 | case KVM_CAP_MCE: |
4039 | r = KVM_MAX_MCE_BANKS; | |
4040 | break; | |
2d5b5a66 | 4041 | case KVM_CAP_XCRS: |
d366bf7e | 4042 | r = boot_cpu_has(X86_FEATURE_XSAVE); |
2d5b5a66 | 4043 | break; |
92a1f12d JR |
4044 | case KVM_CAP_TSC_CONTROL: |
4045 | r = kvm_has_tsc_control; | |
4046 | break; | |
37131313 RK |
4047 | case KVM_CAP_X2APIC_API: |
4048 | r = KVM_X2APIC_API_VALID_FLAGS; | |
4049 | break; | |
8fcc4b59 | 4050 | case KVM_CAP_NESTED_STATE: |
33b22172 PB |
4051 | r = kvm_x86_ops.nested_ops->get_state ? |
4052 | kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0; | |
8fcc4b59 | 4053 | break; |
344c6c80 | 4054 | case KVM_CAP_HYPERV_DIRECT_TLBFLUSH: |
afaf0b2f | 4055 | r = kvm_x86_ops.enable_direct_tlbflush != NULL; |
5a0165f6 VK |
4056 | break; |
4057 | case KVM_CAP_HYPERV_ENLIGHTENED_VMCS: | |
33b22172 | 4058 | r = kvm_x86_ops.nested_ops->enable_evmcs != NULL; |
344c6c80 | 4059 | break; |
3edd6839 MG |
4060 | case KVM_CAP_SMALLER_MAXPHYADDR: |
4061 | r = (int) allow_smaller_maxphyaddr; | |
4062 | break; | |
004a0124 AJ |
4063 | case KVM_CAP_STEAL_TIME: |
4064 | r = sched_info_on(); | |
4065 | break; | |
fe6b6bc8 CQ |
4066 | case KVM_CAP_X86_BUS_LOCK_EXIT: |
4067 | if (kvm_has_bus_lock_exit) | |
4068 | r = KVM_BUS_LOCK_DETECTION_OFF | | |
4069 | KVM_BUS_LOCK_DETECTION_EXIT; | |
4070 | else | |
4071 | r = 0; | |
4072 | break; | |
018d00d2 | 4073 | default: |
018d00d2 ZX |
4074 | break; |
4075 | } | |
4076 | return r; | |
4077 | ||
4078 | } | |
4079 | ||
043405e1 CO |
4080 | long kvm_arch_dev_ioctl(struct file *filp, |
4081 | unsigned int ioctl, unsigned long arg) | |
4082 | { | |
4083 | void __user *argp = (void __user *)arg; | |
4084 | long r; | |
4085 | ||
4086 | switch (ioctl) { | |
4087 | case KVM_GET_MSR_INDEX_LIST: { | |
4088 | struct kvm_msr_list __user *user_msr_list = argp; | |
4089 | struct kvm_msr_list msr_list; | |
4090 | unsigned n; | |
4091 | ||
4092 | r = -EFAULT; | |
0e96f31e | 4093 | if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list))) |
043405e1 CO |
4094 | goto out; |
4095 | n = msr_list.nmsrs; | |
62ef68bb | 4096 | msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs; |
0e96f31e | 4097 | if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list))) |
043405e1 CO |
4098 | goto out; |
4099 | r = -E2BIG; | |
e125e7b6 | 4100 | if (n < msr_list.nmsrs) |
043405e1 CO |
4101 | goto out; |
4102 | r = -EFAULT; | |
4103 | if (copy_to_user(user_msr_list->indices, &msrs_to_save, | |
4104 | num_msrs_to_save * sizeof(u32))) | |
4105 | goto out; | |
e125e7b6 | 4106 | if (copy_to_user(user_msr_list->indices + num_msrs_to_save, |
043405e1 | 4107 | &emulated_msrs, |
62ef68bb | 4108 | num_emulated_msrs * sizeof(u32))) |
043405e1 CO |
4109 | goto out; |
4110 | r = 0; | |
4111 | break; | |
4112 | } | |
9c15bb1d BP |
4113 | case KVM_GET_SUPPORTED_CPUID: |
4114 | case KVM_GET_EMULATED_CPUID: { | |
674eea0f AK |
4115 | struct kvm_cpuid2 __user *cpuid_arg = argp; |
4116 | struct kvm_cpuid2 cpuid; | |
4117 | ||
4118 | r = -EFAULT; | |
0e96f31e | 4119 | if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid))) |
674eea0f | 4120 | goto out; |
9c15bb1d BP |
4121 | |
4122 | r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries, | |
4123 | ioctl); | |
674eea0f AK |
4124 | if (r) |
4125 | goto out; | |
4126 | ||
4127 | r = -EFAULT; | |
0e96f31e | 4128 | if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid))) |
674eea0f AK |
4129 | goto out; |
4130 | r = 0; | |
4131 | break; | |
4132 | } | |
cf6c26ec | 4133 | case KVM_X86_GET_MCE_CAP_SUPPORTED: |
890ca9ae | 4134 | r = -EFAULT; |
c45dcc71 AR |
4135 | if (copy_to_user(argp, &kvm_mce_cap_supported, |
4136 | sizeof(kvm_mce_cap_supported))) | |
890ca9ae HY |
4137 | goto out; |
4138 | r = 0; | |
4139 | break; | |
801e459a TL |
4140 | case KVM_GET_MSR_FEATURE_INDEX_LIST: { |
4141 | struct kvm_msr_list __user *user_msr_list = argp; | |
4142 | struct kvm_msr_list msr_list; | |
4143 | unsigned int n; | |
4144 | ||
4145 | r = -EFAULT; | |
4146 | if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list))) | |
4147 | goto out; | |
4148 | n = msr_list.nmsrs; | |
4149 | msr_list.nmsrs = num_msr_based_features; | |
4150 | if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list))) | |
4151 | goto out; | |
4152 | r = -E2BIG; | |
4153 | if (n < msr_list.nmsrs) | |
4154 | goto out; | |
4155 | r = -EFAULT; | |
4156 | if (copy_to_user(user_msr_list->indices, &msr_based_features, | |
4157 | num_msr_based_features * sizeof(u32))) | |
4158 | goto out; | |
4159 | r = 0; | |
4160 | break; | |
4161 | } | |
4162 | case KVM_GET_MSRS: | |
4163 | r = msr_io(NULL, argp, do_get_msr_feature, 1); | |
4164 | break; | |
c21d54f0 VK |
4165 | case KVM_GET_SUPPORTED_HV_CPUID: |
4166 | r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp); | |
4167 | break; | |
043405e1 CO |
4168 | default: |
4169 | r = -EINVAL; | |
cf6c26ec | 4170 | break; |
043405e1 CO |
4171 | } |
4172 | out: | |
4173 | return r; | |
4174 | } | |
4175 | ||
f5f48ee1 SY |
4176 | static void wbinvd_ipi(void *garbage) |
4177 | { | |
4178 | wbinvd(); | |
4179 | } | |
4180 | ||
4181 | static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu) | |
4182 | { | |
e0f0bbc5 | 4183 | return kvm_arch_has_noncoherent_dma(vcpu->kvm); |
f5f48ee1 SY |
4184 | } |
4185 | ||
313a3dc7 CO |
4186 | void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
4187 | { | |
f5f48ee1 SY |
4188 | /* Address WBINVD may be executed by guest */ |
4189 | if (need_emulate_wbinvd(vcpu)) { | |
b3646477 | 4190 | if (static_call(kvm_x86_has_wbinvd_exit)()) |
f5f48ee1 SY |
4191 | cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask); |
4192 | else if (vcpu->cpu != -1 && vcpu->cpu != cpu) | |
4193 | smp_call_function_single(vcpu->cpu, | |
4194 | wbinvd_ipi, NULL, 1); | |
4195 | } | |
4196 | ||
b3646477 | 4197 | static_call(kvm_x86_vcpu_load)(vcpu, cpu); |
8f6055cb | 4198 | |
37486135 BM |
4199 | /* Save host pkru register if supported */ |
4200 | vcpu->arch.host_pkru = read_pkru(); | |
4201 | ||
0dd6a6ed ZA |
4202 | /* Apply any externally detected TSC adjustments (due to suspend) */ |
4203 | if (unlikely(vcpu->arch.tsc_offset_adjustment)) { | |
4204 | adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment); | |
4205 | vcpu->arch.tsc_offset_adjustment = 0; | |
105b21bb | 4206 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
0dd6a6ed | 4207 | } |
8f6055cb | 4208 | |
b0c39dc6 | 4209 | if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) { |
6f526ec5 | 4210 | s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 : |
4ea1636b | 4211 | rdtsc() - vcpu->arch.last_host_tsc; |
e48672fa ZA |
4212 | if (tsc_delta < 0) |
4213 | mark_tsc_unstable("KVM discovered backwards TSC"); | |
ce7a058a | 4214 | |
b0c39dc6 | 4215 | if (kvm_check_tsc_unstable()) { |
9b399dfd | 4216 | u64 offset = kvm_compute_l1_tsc_offset(vcpu, |
b183aa58 | 4217 | vcpu->arch.last_guest_tsc); |
a545ab6a | 4218 | kvm_vcpu_write_tsc_offset(vcpu, offset); |
c285545f | 4219 | vcpu->arch.tsc_catchup = 1; |
c285545f | 4220 | } |
a749e247 PB |
4221 | |
4222 | if (kvm_lapic_hv_timer_in_use(vcpu)) | |
4223 | kvm_lapic_restart_hv_timer(vcpu); | |
4224 | ||
d98d07ca MT |
4225 | /* |
4226 | * On a host with synchronized TSC, there is no need to update | |
4227 | * kvmclock on vcpu->cpu migration | |
4228 | */ | |
4229 | if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1) | |
0061d53d | 4230 | kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu); |
c285545f | 4231 | if (vcpu->cpu != cpu) |
1bd2009e | 4232 | kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu); |
e48672fa | 4233 | vcpu->cpu = cpu; |
6b7d7e76 | 4234 | } |
c9aaa895 | 4235 | |
c9aaa895 | 4236 | kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu); |
313a3dc7 CO |
4237 | } |
4238 | ||
0b9f6c46 PX |
4239 | static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu) |
4240 | { | |
b0431382 BO |
4241 | struct kvm_host_map map; |
4242 | struct kvm_steal_time *st; | |
4243 | ||
0b9f6c46 PX |
4244 | if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED)) |
4245 | return; | |
4246 | ||
a6bd811f | 4247 | if (vcpu->arch.st.preempted) |
8c6de56a BO |
4248 | return; |
4249 | ||
b0431382 BO |
4250 | if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map, |
4251 | &vcpu->arch.st.cache, true)) | |
9c1a0744 | 4252 | return; |
b0431382 BO |
4253 | |
4254 | st = map.hva + | |
4255 | offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS); | |
0b9f6c46 | 4256 | |
a6bd811f | 4257 | st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED; |
0b9f6c46 | 4258 | |
b0431382 | 4259 | kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true); |
0b9f6c46 PX |
4260 | } |
4261 | ||
313a3dc7 CO |
4262 | void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) |
4263 | { | |
9c1a0744 WL |
4264 | int idx; |
4265 | ||
f1c6366e | 4266 | if (vcpu->preempted && !vcpu->arch.guest_state_protected) |
b3646477 | 4267 | vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu); |
de63ad4c | 4268 | |
9c1a0744 WL |
4269 | /* |
4270 | * Take the srcu lock as memslots will be accessed to check the gfn | |
4271 | * cache generation against the memslots generation. | |
4272 | */ | |
4273 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
30b5c851 DW |
4274 | if (kvm_xen_msr_enabled(vcpu->kvm)) |
4275 | kvm_xen_runstate_set_preempted(vcpu); | |
4276 | else | |
4277 | kvm_steal_time_set_preempted(vcpu); | |
9c1a0744 | 4278 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
30b5c851 | 4279 | |
b3646477 | 4280 | static_call(kvm_x86_vcpu_put)(vcpu); |
4ea1636b | 4281 | vcpu->arch.last_host_tsc = rdtsc(); |
efdab992 | 4282 | /* |
f9dcf08e RK |
4283 | * If userspace has set any breakpoints or watchpoints, dr6 is restored |
4284 | * on every vmexit, but if not, we might have a stale dr6 from the | |
4285 | * guest. do_debug expects dr6 to be cleared after it runs, do the same. | |
efdab992 | 4286 | */ |
f9dcf08e | 4287 | set_debugreg(0, 6); |
313a3dc7 CO |
4288 | } |
4289 | ||
313a3dc7 CO |
4290 | static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu, |
4291 | struct kvm_lapic_state *s) | |
4292 | { | |
fa59cc00 | 4293 | if (vcpu->arch.apicv_active) |
b3646477 | 4294 | static_call(kvm_x86_sync_pir_to_irr)(vcpu); |
d62caabb | 4295 | |
a92e2543 | 4296 | return kvm_apic_get_state(vcpu, s); |
313a3dc7 CO |
4297 | } |
4298 | ||
4299 | static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu, | |
4300 | struct kvm_lapic_state *s) | |
4301 | { | |
a92e2543 RK |
4302 | int r; |
4303 | ||
4304 | r = kvm_apic_set_state(vcpu, s); | |
4305 | if (r) | |
4306 | return r; | |
cb142eb7 | 4307 | update_cr8_intercept(vcpu); |
313a3dc7 CO |
4308 | |
4309 | return 0; | |
4310 | } | |
4311 | ||
127a457a MG |
4312 | static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu) |
4313 | { | |
71cc849b PB |
4314 | /* |
4315 | * We can accept userspace's request for interrupt injection | |
4316 | * as long as we have a place to store the interrupt number. | |
4317 | * The actual injection will happen when the CPU is able to | |
4318 | * deliver the interrupt. | |
4319 | */ | |
4320 | if (kvm_cpu_has_extint(vcpu)) | |
4321 | return false; | |
4322 | ||
4323 | /* Acknowledging ExtINT does not happen if LINT0 is masked. */ | |
127a457a MG |
4324 | return (!lapic_in_kernel(vcpu) || |
4325 | kvm_apic_accept_pic_intr(vcpu)); | |
4326 | } | |
4327 | ||
782d422b MG |
4328 | static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu) |
4329 | { | |
4330 | return kvm_arch_interrupt_allowed(vcpu) && | |
782d422b MG |
4331 | kvm_cpu_accept_dm_intr(vcpu); |
4332 | } | |
4333 | ||
f77bc6a4 ZX |
4334 | static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, |
4335 | struct kvm_interrupt *irq) | |
4336 | { | |
02cdb50f | 4337 | if (irq->irq >= KVM_NR_INTERRUPTS) |
f77bc6a4 | 4338 | return -EINVAL; |
1c1a9ce9 SR |
4339 | |
4340 | if (!irqchip_in_kernel(vcpu->kvm)) { | |
4341 | kvm_queue_interrupt(vcpu, irq->irq, false); | |
4342 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
4343 | return 0; | |
4344 | } | |
4345 | ||
4346 | /* | |
4347 | * With in-kernel LAPIC, we only use this to inject EXTINT, so | |
4348 | * fail for in-kernel 8259. | |
4349 | */ | |
4350 | if (pic_in_kernel(vcpu->kvm)) | |
f77bc6a4 | 4351 | return -ENXIO; |
f77bc6a4 | 4352 | |
1c1a9ce9 SR |
4353 | if (vcpu->arch.pending_external_vector != -1) |
4354 | return -EEXIST; | |
f77bc6a4 | 4355 | |
1c1a9ce9 | 4356 | vcpu->arch.pending_external_vector = irq->irq; |
934bf653 | 4357 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
f77bc6a4 ZX |
4358 | return 0; |
4359 | } | |
4360 | ||
c4abb7c9 JK |
4361 | static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu) |
4362 | { | |
c4abb7c9 | 4363 | kvm_inject_nmi(vcpu); |
c4abb7c9 JK |
4364 | |
4365 | return 0; | |
4366 | } | |
4367 | ||
f077825a PB |
4368 | static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu) |
4369 | { | |
64d60670 PB |
4370 | kvm_make_request(KVM_REQ_SMI, vcpu); |
4371 | ||
f077825a PB |
4372 | return 0; |
4373 | } | |
4374 | ||
b209749f AK |
4375 | static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu, |
4376 | struct kvm_tpr_access_ctl *tac) | |
4377 | { | |
4378 | if (tac->flags) | |
4379 | return -EINVAL; | |
4380 | vcpu->arch.tpr_access_reporting = !!tac->enabled; | |
4381 | return 0; | |
4382 | } | |
4383 | ||
890ca9ae HY |
4384 | static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu, |
4385 | u64 mcg_cap) | |
4386 | { | |
4387 | int r; | |
4388 | unsigned bank_num = mcg_cap & 0xff, bank; | |
4389 | ||
4390 | r = -EINVAL; | |
c4e0e4ab | 4391 | if (!bank_num || bank_num > KVM_MAX_MCE_BANKS) |
890ca9ae | 4392 | goto out; |
c45dcc71 | 4393 | if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000)) |
890ca9ae HY |
4394 | goto out; |
4395 | r = 0; | |
4396 | vcpu->arch.mcg_cap = mcg_cap; | |
4397 | /* Init IA32_MCG_CTL to all 1s */ | |
4398 | if (mcg_cap & MCG_CTL_P) | |
4399 | vcpu->arch.mcg_ctl = ~(u64)0; | |
4400 | /* Init IA32_MCi_CTL to all 1s */ | |
4401 | for (bank = 0; bank < bank_num; bank++) | |
4402 | vcpu->arch.mce_banks[bank*4] = ~(u64)0; | |
c45dcc71 | 4403 | |
b3646477 | 4404 | static_call(kvm_x86_setup_mce)(vcpu); |
890ca9ae HY |
4405 | out: |
4406 | return r; | |
4407 | } | |
4408 | ||
4409 | static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu, | |
4410 | struct kvm_x86_mce *mce) | |
4411 | { | |
4412 | u64 mcg_cap = vcpu->arch.mcg_cap; | |
4413 | unsigned bank_num = mcg_cap & 0xff; | |
4414 | u64 *banks = vcpu->arch.mce_banks; | |
4415 | ||
4416 | if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL)) | |
4417 | return -EINVAL; | |
4418 | /* | |
4419 | * if IA32_MCG_CTL is not all 1s, the uncorrected error | |
4420 | * reporting is disabled | |
4421 | */ | |
4422 | if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) && | |
4423 | vcpu->arch.mcg_ctl != ~(u64)0) | |
4424 | return 0; | |
4425 | banks += 4 * mce->bank; | |
4426 | /* | |
4427 | * if IA32_MCi_CTL is not all 1s, the uncorrected error | |
4428 | * reporting is disabled for the bank | |
4429 | */ | |
4430 | if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0) | |
4431 | return 0; | |
4432 | if (mce->status & MCI_STATUS_UC) { | |
4433 | if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) || | |
fc78f519 | 4434 | !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) { |
a8eeb04a | 4435 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); |
890ca9ae HY |
4436 | return 0; |
4437 | } | |
4438 | if (banks[1] & MCI_STATUS_VAL) | |
4439 | mce->status |= MCI_STATUS_OVER; | |
4440 | banks[2] = mce->addr; | |
4441 | banks[3] = mce->misc; | |
4442 | vcpu->arch.mcg_status = mce->mcg_status; | |
4443 | banks[1] = mce->status; | |
4444 | kvm_queue_exception(vcpu, MC_VECTOR); | |
4445 | } else if (!(banks[1] & MCI_STATUS_VAL) | |
4446 | || !(banks[1] & MCI_STATUS_UC)) { | |
4447 | if (banks[1] & MCI_STATUS_VAL) | |
4448 | mce->status |= MCI_STATUS_OVER; | |
4449 | banks[2] = mce->addr; | |
4450 | banks[3] = mce->misc; | |
4451 | banks[1] = mce->status; | |
4452 | } else | |
4453 | banks[1] |= MCI_STATUS_OVER; | |
4454 | return 0; | |
4455 | } | |
4456 | ||
3cfc3092 JK |
4457 | static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu, |
4458 | struct kvm_vcpu_events *events) | |
4459 | { | |
7460fb4a | 4460 | process_nmi(vcpu); |
59073aaf | 4461 | |
1f7becf1 JZ |
4462 | if (kvm_check_request(KVM_REQ_SMI, vcpu)) |
4463 | process_smi(vcpu); | |
4464 | ||
a06230b6 OU |
4465 | /* |
4466 | * In guest mode, payload delivery should be deferred, | |
4467 | * so that the L1 hypervisor can intercept #PF before | |
4468 | * CR2 is modified (or intercept #DB before DR6 is | |
4469 | * modified under nVMX). Unless the per-VM capability, | |
4470 | * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of | |
4471 | * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we | |
4472 | * opportunistically defer the exception payload, deliver it if the | |
4473 | * capability hasn't been requested before processing a | |
4474 | * KVM_GET_VCPU_EVENTS. | |
4475 | */ | |
4476 | if (!vcpu->kvm->arch.exception_payload_enabled && | |
4477 | vcpu->arch.exception.pending && vcpu->arch.exception.has_payload) | |
4478 | kvm_deliver_exception_payload(vcpu); | |
4479 | ||
664f8e26 | 4480 | /* |
59073aaf JM |
4481 | * The API doesn't provide the instruction length for software |
4482 | * exceptions, so don't report them. As long as the guest RIP | |
4483 | * isn't advanced, we should expect to encounter the exception | |
4484 | * again. | |
664f8e26 | 4485 | */ |
59073aaf JM |
4486 | if (kvm_exception_is_soft(vcpu->arch.exception.nr)) { |
4487 | events->exception.injected = 0; | |
4488 | events->exception.pending = 0; | |
4489 | } else { | |
4490 | events->exception.injected = vcpu->arch.exception.injected; | |
4491 | events->exception.pending = vcpu->arch.exception.pending; | |
4492 | /* | |
4493 | * For ABI compatibility, deliberately conflate | |
4494 | * pending and injected exceptions when | |
4495 | * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled. | |
4496 | */ | |
4497 | if (!vcpu->kvm->arch.exception_payload_enabled) | |
4498 | events->exception.injected |= | |
4499 | vcpu->arch.exception.pending; | |
4500 | } | |
3cfc3092 JK |
4501 | events->exception.nr = vcpu->arch.exception.nr; |
4502 | events->exception.has_error_code = vcpu->arch.exception.has_error_code; | |
4503 | events->exception.error_code = vcpu->arch.exception.error_code; | |
59073aaf JM |
4504 | events->exception_has_payload = vcpu->arch.exception.has_payload; |
4505 | events->exception_payload = vcpu->arch.exception.payload; | |
3cfc3092 | 4506 | |
03b82a30 | 4507 | events->interrupt.injected = |
04140b41 | 4508 | vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft; |
3cfc3092 | 4509 | events->interrupt.nr = vcpu->arch.interrupt.nr; |
03b82a30 | 4510 | events->interrupt.soft = 0; |
b3646477 | 4511 | events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu); |
3cfc3092 JK |
4512 | |
4513 | events->nmi.injected = vcpu->arch.nmi_injected; | |
7460fb4a | 4514 | events->nmi.pending = vcpu->arch.nmi_pending != 0; |
b3646477 | 4515 | events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu); |
97e69aa6 | 4516 | events->nmi.pad = 0; |
3cfc3092 | 4517 | |
66450a21 | 4518 | events->sipi_vector = 0; /* never valid when reporting to user space */ |
3cfc3092 | 4519 | |
f077825a PB |
4520 | events->smi.smm = is_smm(vcpu); |
4521 | events->smi.pending = vcpu->arch.smi_pending; | |
4522 | events->smi.smm_inside_nmi = | |
4523 | !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK); | |
4524 | events->smi.latched_init = kvm_lapic_latched_init(vcpu); | |
4525 | ||
dab4b911 | 4526 | events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING |
f077825a PB |
4527 | | KVM_VCPUEVENT_VALID_SHADOW |
4528 | | KVM_VCPUEVENT_VALID_SMM); | |
59073aaf JM |
4529 | if (vcpu->kvm->arch.exception_payload_enabled) |
4530 | events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD; | |
4531 | ||
97e69aa6 | 4532 | memset(&events->reserved, 0, sizeof(events->reserved)); |
3cfc3092 JK |
4533 | } |
4534 | ||
dc87275f | 4535 | static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm); |
6ef4e07e | 4536 | |
3cfc3092 JK |
4537 | static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu, |
4538 | struct kvm_vcpu_events *events) | |
4539 | { | |
dab4b911 | 4540 | if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING |
48005f64 | 4541 | | KVM_VCPUEVENT_VALID_SIPI_VECTOR |
f077825a | 4542 | | KVM_VCPUEVENT_VALID_SHADOW |
59073aaf JM |
4543 | | KVM_VCPUEVENT_VALID_SMM |
4544 | | KVM_VCPUEVENT_VALID_PAYLOAD)) | |
3cfc3092 JK |
4545 | return -EINVAL; |
4546 | ||
59073aaf JM |
4547 | if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) { |
4548 | if (!vcpu->kvm->arch.exception_payload_enabled) | |
4549 | return -EINVAL; | |
4550 | if (events->exception.pending) | |
4551 | events->exception.injected = 0; | |
4552 | else | |
4553 | events->exception_has_payload = 0; | |
4554 | } else { | |
4555 | events->exception.pending = 0; | |
4556 | events->exception_has_payload = 0; | |
4557 | } | |
4558 | ||
4559 | if ((events->exception.injected || events->exception.pending) && | |
4560 | (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR)) | |
78e546c8 PB |
4561 | return -EINVAL; |
4562 | ||
28bf2888 DH |
4563 | /* INITs are latched while in SMM */ |
4564 | if (events->flags & KVM_VCPUEVENT_VALID_SMM && | |
4565 | (events->smi.smm || events->smi.pending) && | |
4566 | vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) | |
4567 | return -EINVAL; | |
4568 | ||
7460fb4a | 4569 | process_nmi(vcpu); |
59073aaf JM |
4570 | vcpu->arch.exception.injected = events->exception.injected; |
4571 | vcpu->arch.exception.pending = events->exception.pending; | |
3cfc3092 JK |
4572 | vcpu->arch.exception.nr = events->exception.nr; |
4573 | vcpu->arch.exception.has_error_code = events->exception.has_error_code; | |
4574 | vcpu->arch.exception.error_code = events->exception.error_code; | |
59073aaf JM |
4575 | vcpu->arch.exception.has_payload = events->exception_has_payload; |
4576 | vcpu->arch.exception.payload = events->exception_payload; | |
3cfc3092 | 4577 | |
04140b41 | 4578 | vcpu->arch.interrupt.injected = events->interrupt.injected; |
3cfc3092 JK |
4579 | vcpu->arch.interrupt.nr = events->interrupt.nr; |
4580 | vcpu->arch.interrupt.soft = events->interrupt.soft; | |
48005f64 | 4581 | if (events->flags & KVM_VCPUEVENT_VALID_SHADOW) |
b3646477 JB |
4582 | static_call(kvm_x86_set_interrupt_shadow)(vcpu, |
4583 | events->interrupt.shadow); | |
3cfc3092 JK |
4584 | |
4585 | vcpu->arch.nmi_injected = events->nmi.injected; | |
dab4b911 JK |
4586 | if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING) |
4587 | vcpu->arch.nmi_pending = events->nmi.pending; | |
b3646477 | 4588 | static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked); |
3cfc3092 | 4589 | |
66450a21 | 4590 | if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR && |
bce87cce | 4591 | lapic_in_kernel(vcpu)) |
66450a21 | 4592 | vcpu->arch.apic->sipi_vector = events->sipi_vector; |
3cfc3092 | 4593 | |
f077825a | 4594 | if (events->flags & KVM_VCPUEVENT_VALID_SMM) { |
dc87275f SC |
4595 | if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) |
4596 | kvm_smm_changed(vcpu, events->smi.smm); | |
6ef4e07e | 4597 | |
f077825a | 4598 | vcpu->arch.smi_pending = events->smi.pending; |
f4ef1910 WL |
4599 | |
4600 | if (events->smi.smm) { | |
4601 | if (events->smi.smm_inside_nmi) | |
4602 | vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK; | |
f077825a | 4603 | else |
f4ef1910 | 4604 | vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK; |
ff90afa7 LA |
4605 | } |
4606 | ||
4607 | if (lapic_in_kernel(vcpu)) { | |
4608 | if (events->smi.latched_init) | |
4609 | set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events); | |
4610 | else | |
4611 | clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events); | |
f077825a PB |
4612 | } |
4613 | } | |
4614 | ||
3842d135 AK |
4615 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
4616 | ||
3cfc3092 JK |
4617 | return 0; |
4618 | } | |
4619 | ||
a1efbe77 JK |
4620 | static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu, |
4621 | struct kvm_debugregs *dbgregs) | |
4622 | { | |
73aaf249 JK |
4623 | unsigned long val; |
4624 | ||
a1efbe77 | 4625 | memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db)); |
16f8a6f9 | 4626 | kvm_get_dr(vcpu, 6, &val); |
73aaf249 | 4627 | dbgregs->dr6 = val; |
a1efbe77 JK |
4628 | dbgregs->dr7 = vcpu->arch.dr7; |
4629 | dbgregs->flags = 0; | |
97e69aa6 | 4630 | memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved)); |
a1efbe77 JK |
4631 | } |
4632 | ||
4633 | static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu, | |
4634 | struct kvm_debugregs *dbgregs) | |
4635 | { | |
4636 | if (dbgregs->flags) | |
4637 | return -EINVAL; | |
4638 | ||
fd238002 | 4639 | if (!kvm_dr6_valid(dbgregs->dr6)) |
d14bdb55 | 4640 | return -EINVAL; |
fd238002 | 4641 | if (!kvm_dr7_valid(dbgregs->dr7)) |
d14bdb55 PB |
4642 | return -EINVAL; |
4643 | ||
a1efbe77 | 4644 | memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db)); |
ae561ede | 4645 | kvm_update_dr0123(vcpu); |
a1efbe77 JK |
4646 | vcpu->arch.dr6 = dbgregs->dr6; |
4647 | vcpu->arch.dr7 = dbgregs->dr7; | |
9926c9fd | 4648 | kvm_update_dr7(vcpu); |
a1efbe77 | 4649 | |
a1efbe77 JK |
4650 | return 0; |
4651 | } | |
4652 | ||
df1daba7 PB |
4653 | #define XSTATE_COMPACTION_ENABLED (1ULL << 63) |
4654 | ||
4655 | static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu) | |
4656 | { | |
b666a4b6 | 4657 | struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave; |
400e4b20 | 4658 | u64 xstate_bv = xsave->header.xfeatures; |
df1daba7 PB |
4659 | u64 valid; |
4660 | ||
4661 | /* | |
4662 | * Copy legacy XSAVE area, to avoid complications with CPUID | |
4663 | * leaves 0 and 1 in the loop below. | |
4664 | */ | |
4665 | memcpy(dest, xsave, XSAVE_HDR_OFFSET); | |
4666 | ||
4667 | /* Set XSTATE_BV */ | |
00c87e9a | 4668 | xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE; |
df1daba7 PB |
4669 | *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv; |
4670 | ||
4671 | /* | |
4672 | * Copy each region from the possibly compacted offset to the | |
4673 | * non-compacted offset. | |
4674 | */ | |
d91cab78 | 4675 | valid = xstate_bv & ~XFEATURE_MASK_FPSSE; |
df1daba7 | 4676 | while (valid) { |
abd16d68 SAS |
4677 | u64 xfeature_mask = valid & -valid; |
4678 | int xfeature_nr = fls64(xfeature_mask) - 1; | |
4679 | void *src = get_xsave_addr(xsave, xfeature_nr); | |
df1daba7 PB |
4680 | |
4681 | if (src) { | |
4682 | u32 size, offset, ecx, edx; | |
abd16d68 | 4683 | cpuid_count(XSTATE_CPUID, xfeature_nr, |
df1daba7 | 4684 | &size, &offset, &ecx, &edx); |
abd16d68 | 4685 | if (xfeature_nr == XFEATURE_PKRU) |
38cfd5e3 PB |
4686 | memcpy(dest + offset, &vcpu->arch.pkru, |
4687 | sizeof(vcpu->arch.pkru)); | |
4688 | else | |
4689 | memcpy(dest + offset, src, size); | |
4690 | ||
df1daba7 PB |
4691 | } |
4692 | ||
abd16d68 | 4693 | valid -= xfeature_mask; |
df1daba7 PB |
4694 | } |
4695 | } | |
4696 | ||
4697 | static void load_xsave(struct kvm_vcpu *vcpu, u8 *src) | |
4698 | { | |
b666a4b6 | 4699 | struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave; |
df1daba7 PB |
4700 | u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET); |
4701 | u64 valid; | |
4702 | ||
4703 | /* | |
4704 | * Copy legacy XSAVE area, to avoid complications with CPUID | |
4705 | * leaves 0 and 1 in the loop below. | |
4706 | */ | |
4707 | memcpy(xsave, src, XSAVE_HDR_OFFSET); | |
4708 | ||
4709 | /* Set XSTATE_BV and possibly XCOMP_BV. */ | |
400e4b20 | 4710 | xsave->header.xfeatures = xstate_bv; |
782511b0 | 4711 | if (boot_cpu_has(X86_FEATURE_XSAVES)) |
3a54450b | 4712 | xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED; |
df1daba7 PB |
4713 | |
4714 | /* | |
4715 | * Copy each region from the non-compacted offset to the | |
4716 | * possibly compacted offset. | |
4717 | */ | |
d91cab78 | 4718 | valid = xstate_bv & ~XFEATURE_MASK_FPSSE; |
df1daba7 | 4719 | while (valid) { |
abd16d68 SAS |
4720 | u64 xfeature_mask = valid & -valid; |
4721 | int xfeature_nr = fls64(xfeature_mask) - 1; | |
4722 | void *dest = get_xsave_addr(xsave, xfeature_nr); | |
df1daba7 PB |
4723 | |
4724 | if (dest) { | |
4725 | u32 size, offset, ecx, edx; | |
abd16d68 | 4726 | cpuid_count(XSTATE_CPUID, xfeature_nr, |
df1daba7 | 4727 | &size, &offset, &ecx, &edx); |
abd16d68 | 4728 | if (xfeature_nr == XFEATURE_PKRU) |
38cfd5e3 PB |
4729 | memcpy(&vcpu->arch.pkru, src + offset, |
4730 | sizeof(vcpu->arch.pkru)); | |
4731 | else | |
4732 | memcpy(dest, src + offset, size); | |
ee4100da | 4733 | } |
df1daba7 | 4734 | |
abd16d68 | 4735 | valid -= xfeature_mask; |
df1daba7 PB |
4736 | } |
4737 | } | |
4738 | ||
2d5b5a66 SY |
4739 | static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu, |
4740 | struct kvm_xsave *guest_xsave) | |
4741 | { | |
ed02b213 TL |
4742 | if (!vcpu->arch.guest_fpu) |
4743 | return; | |
4744 | ||
d366bf7e | 4745 | if (boot_cpu_has(X86_FEATURE_XSAVE)) { |
df1daba7 PB |
4746 | memset(guest_xsave, 0, sizeof(struct kvm_xsave)); |
4747 | fill_xsave((u8 *) guest_xsave->region, vcpu); | |
4344ee98 | 4748 | } else { |
2d5b5a66 | 4749 | memcpy(guest_xsave->region, |
b666a4b6 | 4750 | &vcpu->arch.guest_fpu->state.fxsave, |
c47ada30 | 4751 | sizeof(struct fxregs_state)); |
2d5b5a66 | 4752 | *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] = |
d91cab78 | 4753 | XFEATURE_MASK_FPSSE; |
2d5b5a66 SY |
4754 | } |
4755 | } | |
4756 | ||
a575813b WL |
4757 | #define XSAVE_MXCSR_OFFSET 24 |
4758 | ||
2d5b5a66 SY |
4759 | static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu, |
4760 | struct kvm_xsave *guest_xsave) | |
4761 | { | |
ed02b213 TL |
4762 | u64 xstate_bv; |
4763 | u32 mxcsr; | |
4764 | ||
4765 | if (!vcpu->arch.guest_fpu) | |
4766 | return 0; | |
4767 | ||
4768 | xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)]; | |
4769 | mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)]; | |
2d5b5a66 | 4770 | |
d366bf7e | 4771 | if (boot_cpu_has(X86_FEATURE_XSAVE)) { |
d7876f1b PB |
4772 | /* |
4773 | * Here we allow setting states that are not present in | |
4774 | * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility | |
4775 | * with old userspace. | |
4776 | */ | |
cfc48181 | 4777 | if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask) |
d7876f1b | 4778 | return -EINVAL; |
df1daba7 | 4779 | load_xsave(vcpu, (u8 *)guest_xsave->region); |
d7876f1b | 4780 | } else { |
a575813b WL |
4781 | if (xstate_bv & ~XFEATURE_MASK_FPSSE || |
4782 | mxcsr & ~mxcsr_feature_mask) | |
2d5b5a66 | 4783 | return -EINVAL; |
b666a4b6 | 4784 | memcpy(&vcpu->arch.guest_fpu->state.fxsave, |
c47ada30 | 4785 | guest_xsave->region, sizeof(struct fxregs_state)); |
2d5b5a66 SY |
4786 | } |
4787 | return 0; | |
4788 | } | |
4789 | ||
4790 | static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu, | |
4791 | struct kvm_xcrs *guest_xcrs) | |
4792 | { | |
d366bf7e | 4793 | if (!boot_cpu_has(X86_FEATURE_XSAVE)) { |
2d5b5a66 SY |
4794 | guest_xcrs->nr_xcrs = 0; |
4795 | return; | |
4796 | } | |
4797 | ||
4798 | guest_xcrs->nr_xcrs = 1; | |
4799 | guest_xcrs->flags = 0; | |
4800 | guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK; | |
4801 | guest_xcrs->xcrs[0].value = vcpu->arch.xcr0; | |
4802 | } | |
4803 | ||
4804 | static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu, | |
4805 | struct kvm_xcrs *guest_xcrs) | |
4806 | { | |
4807 | int i, r = 0; | |
4808 | ||
d366bf7e | 4809 | if (!boot_cpu_has(X86_FEATURE_XSAVE)) |
2d5b5a66 SY |
4810 | return -EINVAL; |
4811 | ||
4812 | if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags) | |
4813 | return -EINVAL; | |
4814 | ||
4815 | for (i = 0; i < guest_xcrs->nr_xcrs; i++) | |
4816 | /* Only support XCR0 currently */ | |
c67a04cb | 4817 | if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) { |
2d5b5a66 | 4818 | r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK, |
c67a04cb | 4819 | guest_xcrs->xcrs[i].value); |
2d5b5a66 SY |
4820 | break; |
4821 | } | |
4822 | if (r) | |
4823 | r = -EINVAL; | |
4824 | return r; | |
4825 | } | |
4826 | ||
1c0b28c2 EM |
4827 | /* |
4828 | * kvm_set_guest_paused() indicates to the guest kernel that it has been | |
4829 | * stopped by the hypervisor. This function will be called from the host only. | |
4830 | * EINVAL is returned when the host attempts to set the flag for a guest that | |
4831 | * does not support pv clocks. | |
4832 | */ | |
4833 | static int kvm_set_guest_paused(struct kvm_vcpu *vcpu) | |
4834 | { | |
0b79459b | 4835 | if (!vcpu->arch.pv_time_enabled) |
1c0b28c2 | 4836 | return -EINVAL; |
51d59c6b | 4837 | vcpu->arch.pvclock_set_guest_stopped_request = true; |
1c0b28c2 EM |
4838 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
4839 | return 0; | |
4840 | } | |
4841 | ||
5c919412 AS |
4842 | static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu, |
4843 | struct kvm_enable_cap *cap) | |
4844 | { | |
57b119da VK |
4845 | int r; |
4846 | uint16_t vmcs_version; | |
4847 | void __user *user_ptr; | |
4848 | ||
5c919412 AS |
4849 | if (cap->flags) |
4850 | return -EINVAL; | |
4851 | ||
4852 | switch (cap->cap) { | |
efc479e6 RK |
4853 | case KVM_CAP_HYPERV_SYNIC2: |
4854 | if (cap->args[0]) | |
4855 | return -EINVAL; | |
df561f66 | 4856 | fallthrough; |
b2869f28 | 4857 | |
5c919412 | 4858 | case KVM_CAP_HYPERV_SYNIC: |
546d87e5 WL |
4859 | if (!irqchip_in_kernel(vcpu->kvm)) |
4860 | return -EINVAL; | |
efc479e6 RK |
4861 | return kvm_hv_activate_synic(vcpu, cap->cap == |
4862 | KVM_CAP_HYPERV_SYNIC2); | |
57b119da | 4863 | case KVM_CAP_HYPERV_ENLIGHTENED_VMCS: |
33b22172 | 4864 | if (!kvm_x86_ops.nested_ops->enable_evmcs) |
5158917c | 4865 | return -ENOTTY; |
33b22172 | 4866 | r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version); |
57b119da VK |
4867 | if (!r) { |
4868 | user_ptr = (void __user *)(uintptr_t)cap->args[0]; | |
4869 | if (copy_to_user(user_ptr, &vmcs_version, | |
4870 | sizeof(vmcs_version))) | |
4871 | r = -EFAULT; | |
4872 | } | |
4873 | return r; | |
344c6c80 | 4874 | case KVM_CAP_HYPERV_DIRECT_TLBFLUSH: |
afaf0b2f | 4875 | if (!kvm_x86_ops.enable_direct_tlbflush) |
344c6c80 TL |
4876 | return -ENOTTY; |
4877 | ||
b3646477 | 4878 | return static_call(kvm_x86_enable_direct_tlbflush)(vcpu); |
57b119da | 4879 | |
66570e96 OU |
4880 | case KVM_CAP_ENFORCE_PV_FEATURE_CPUID: |
4881 | vcpu->arch.pv_cpuid.enforce = cap->args[0]; | |
01b4f510 OU |
4882 | if (vcpu->arch.pv_cpuid.enforce) |
4883 | kvm_update_pv_runtime(vcpu); | |
66570e96 OU |
4884 | |
4885 | return 0; | |
5c919412 AS |
4886 | default: |
4887 | return -EINVAL; | |
4888 | } | |
4889 | } | |
4890 | ||
313a3dc7 CO |
4891 | long kvm_arch_vcpu_ioctl(struct file *filp, |
4892 | unsigned int ioctl, unsigned long arg) | |
4893 | { | |
4894 | struct kvm_vcpu *vcpu = filp->private_data; | |
4895 | void __user *argp = (void __user *)arg; | |
4896 | int r; | |
d1ac91d8 AK |
4897 | union { |
4898 | struct kvm_lapic_state *lapic; | |
4899 | struct kvm_xsave *xsave; | |
4900 | struct kvm_xcrs *xcrs; | |
4901 | void *buffer; | |
4902 | } u; | |
4903 | ||
9b062471 CD |
4904 | vcpu_load(vcpu); |
4905 | ||
d1ac91d8 | 4906 | u.buffer = NULL; |
313a3dc7 CO |
4907 | switch (ioctl) { |
4908 | case KVM_GET_LAPIC: { | |
2204ae3c | 4909 | r = -EINVAL; |
bce87cce | 4910 | if (!lapic_in_kernel(vcpu)) |
2204ae3c | 4911 | goto out; |
254272ce BG |
4912 | u.lapic = kzalloc(sizeof(struct kvm_lapic_state), |
4913 | GFP_KERNEL_ACCOUNT); | |
313a3dc7 | 4914 | |
b772ff36 | 4915 | r = -ENOMEM; |
d1ac91d8 | 4916 | if (!u.lapic) |
b772ff36 | 4917 | goto out; |
d1ac91d8 | 4918 | r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic); |
313a3dc7 CO |
4919 | if (r) |
4920 | goto out; | |
4921 | r = -EFAULT; | |
d1ac91d8 | 4922 | if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state))) |
313a3dc7 CO |
4923 | goto out; |
4924 | r = 0; | |
4925 | break; | |
4926 | } | |
4927 | case KVM_SET_LAPIC: { | |
2204ae3c | 4928 | r = -EINVAL; |
bce87cce | 4929 | if (!lapic_in_kernel(vcpu)) |
2204ae3c | 4930 | goto out; |
ff5c2c03 | 4931 | u.lapic = memdup_user(argp, sizeof(*u.lapic)); |
9b062471 CD |
4932 | if (IS_ERR(u.lapic)) { |
4933 | r = PTR_ERR(u.lapic); | |
4934 | goto out_nofree; | |
4935 | } | |
ff5c2c03 | 4936 | |
d1ac91d8 | 4937 | r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic); |
313a3dc7 CO |
4938 | break; |
4939 | } | |
f77bc6a4 ZX |
4940 | case KVM_INTERRUPT: { |
4941 | struct kvm_interrupt irq; | |
4942 | ||
4943 | r = -EFAULT; | |
0e96f31e | 4944 | if (copy_from_user(&irq, argp, sizeof(irq))) |
f77bc6a4 ZX |
4945 | goto out; |
4946 | r = kvm_vcpu_ioctl_interrupt(vcpu, &irq); | |
f77bc6a4 ZX |
4947 | break; |
4948 | } | |
c4abb7c9 JK |
4949 | case KVM_NMI: { |
4950 | r = kvm_vcpu_ioctl_nmi(vcpu); | |
c4abb7c9 JK |
4951 | break; |
4952 | } | |
f077825a PB |
4953 | case KVM_SMI: { |
4954 | r = kvm_vcpu_ioctl_smi(vcpu); | |
4955 | break; | |
4956 | } | |
313a3dc7 CO |
4957 | case KVM_SET_CPUID: { |
4958 | struct kvm_cpuid __user *cpuid_arg = argp; | |
4959 | struct kvm_cpuid cpuid; | |
4960 | ||
4961 | r = -EFAULT; | |
0e96f31e | 4962 | if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid))) |
313a3dc7 CO |
4963 | goto out; |
4964 | r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries); | |
313a3dc7 CO |
4965 | break; |
4966 | } | |
07716717 DK |
4967 | case KVM_SET_CPUID2: { |
4968 | struct kvm_cpuid2 __user *cpuid_arg = argp; | |
4969 | struct kvm_cpuid2 cpuid; | |
4970 | ||
4971 | r = -EFAULT; | |
0e96f31e | 4972 | if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid))) |
07716717 DK |
4973 | goto out; |
4974 | r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid, | |
19355475 | 4975 | cpuid_arg->entries); |
07716717 DK |
4976 | break; |
4977 | } | |
4978 | case KVM_GET_CPUID2: { | |
4979 | struct kvm_cpuid2 __user *cpuid_arg = argp; | |
4980 | struct kvm_cpuid2 cpuid; | |
4981 | ||
4982 | r = -EFAULT; | |
0e96f31e | 4983 | if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid))) |
07716717 DK |
4984 | goto out; |
4985 | r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid, | |
19355475 | 4986 | cpuid_arg->entries); |
07716717 DK |
4987 | if (r) |
4988 | goto out; | |
4989 | r = -EFAULT; | |
0e96f31e | 4990 | if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid))) |
07716717 DK |
4991 | goto out; |
4992 | r = 0; | |
4993 | break; | |
4994 | } | |
801e459a TL |
4995 | case KVM_GET_MSRS: { |
4996 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
609e36d3 | 4997 | r = msr_io(vcpu, argp, do_get_msr, 1); |
801e459a | 4998 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
313a3dc7 | 4999 | break; |
801e459a TL |
5000 | } |
5001 | case KVM_SET_MSRS: { | |
5002 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
313a3dc7 | 5003 | r = msr_io(vcpu, argp, do_set_msr, 0); |
801e459a | 5004 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
313a3dc7 | 5005 | break; |
801e459a | 5006 | } |
b209749f AK |
5007 | case KVM_TPR_ACCESS_REPORTING: { |
5008 | struct kvm_tpr_access_ctl tac; | |
5009 | ||
5010 | r = -EFAULT; | |
0e96f31e | 5011 | if (copy_from_user(&tac, argp, sizeof(tac))) |
b209749f AK |
5012 | goto out; |
5013 | r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac); | |
5014 | if (r) | |
5015 | goto out; | |
5016 | r = -EFAULT; | |
0e96f31e | 5017 | if (copy_to_user(argp, &tac, sizeof(tac))) |
b209749f AK |
5018 | goto out; |
5019 | r = 0; | |
5020 | break; | |
5021 | }; | |
b93463aa AK |
5022 | case KVM_SET_VAPIC_ADDR: { |
5023 | struct kvm_vapic_addr va; | |
7301d6ab | 5024 | int idx; |
b93463aa AK |
5025 | |
5026 | r = -EINVAL; | |
35754c98 | 5027 | if (!lapic_in_kernel(vcpu)) |
b93463aa AK |
5028 | goto out; |
5029 | r = -EFAULT; | |
0e96f31e | 5030 | if (copy_from_user(&va, argp, sizeof(va))) |
b93463aa | 5031 | goto out; |
7301d6ab | 5032 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
fda4e2e8 | 5033 | r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr); |
7301d6ab | 5034 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
b93463aa AK |
5035 | break; |
5036 | } | |
890ca9ae HY |
5037 | case KVM_X86_SETUP_MCE: { |
5038 | u64 mcg_cap; | |
5039 | ||
5040 | r = -EFAULT; | |
0e96f31e | 5041 | if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap))) |
890ca9ae HY |
5042 | goto out; |
5043 | r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap); | |
5044 | break; | |
5045 | } | |
5046 | case KVM_X86_SET_MCE: { | |
5047 | struct kvm_x86_mce mce; | |
5048 | ||
5049 | r = -EFAULT; | |
0e96f31e | 5050 | if (copy_from_user(&mce, argp, sizeof(mce))) |
890ca9ae HY |
5051 | goto out; |
5052 | r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce); | |
5053 | break; | |
5054 | } | |
3cfc3092 JK |
5055 | case KVM_GET_VCPU_EVENTS: { |
5056 | struct kvm_vcpu_events events; | |
5057 | ||
5058 | kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events); | |
5059 | ||
5060 | r = -EFAULT; | |
5061 | if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events))) | |
5062 | break; | |
5063 | r = 0; | |
5064 | break; | |
5065 | } | |
5066 | case KVM_SET_VCPU_EVENTS: { | |
5067 | struct kvm_vcpu_events events; | |
5068 | ||
5069 | r = -EFAULT; | |
5070 | if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events))) | |
5071 | break; | |
5072 | ||
5073 | r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events); | |
5074 | break; | |
5075 | } | |
a1efbe77 JK |
5076 | case KVM_GET_DEBUGREGS: { |
5077 | struct kvm_debugregs dbgregs; | |
5078 | ||
5079 | kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs); | |
5080 | ||
5081 | r = -EFAULT; | |
5082 | if (copy_to_user(argp, &dbgregs, | |
5083 | sizeof(struct kvm_debugregs))) | |
5084 | break; | |
5085 | r = 0; | |
5086 | break; | |
5087 | } | |
5088 | case KVM_SET_DEBUGREGS: { | |
5089 | struct kvm_debugregs dbgregs; | |
5090 | ||
5091 | r = -EFAULT; | |
5092 | if (copy_from_user(&dbgregs, argp, | |
5093 | sizeof(struct kvm_debugregs))) | |
5094 | break; | |
5095 | ||
5096 | r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs); | |
5097 | break; | |
5098 | } | |
2d5b5a66 | 5099 | case KVM_GET_XSAVE: { |
254272ce | 5100 | u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT); |
2d5b5a66 | 5101 | r = -ENOMEM; |
d1ac91d8 | 5102 | if (!u.xsave) |
2d5b5a66 SY |
5103 | break; |
5104 | ||
d1ac91d8 | 5105 | kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave); |
2d5b5a66 SY |
5106 | |
5107 | r = -EFAULT; | |
d1ac91d8 | 5108 | if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave))) |
2d5b5a66 SY |
5109 | break; |
5110 | r = 0; | |
5111 | break; | |
5112 | } | |
5113 | case KVM_SET_XSAVE: { | |
ff5c2c03 | 5114 | u.xsave = memdup_user(argp, sizeof(*u.xsave)); |
9b062471 CD |
5115 | if (IS_ERR(u.xsave)) { |
5116 | r = PTR_ERR(u.xsave); | |
5117 | goto out_nofree; | |
5118 | } | |
2d5b5a66 | 5119 | |
d1ac91d8 | 5120 | r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave); |
2d5b5a66 SY |
5121 | break; |
5122 | } | |
5123 | case KVM_GET_XCRS: { | |
254272ce | 5124 | u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT); |
2d5b5a66 | 5125 | r = -ENOMEM; |
d1ac91d8 | 5126 | if (!u.xcrs) |
2d5b5a66 SY |
5127 | break; |
5128 | ||
d1ac91d8 | 5129 | kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs); |
2d5b5a66 SY |
5130 | |
5131 | r = -EFAULT; | |
d1ac91d8 | 5132 | if (copy_to_user(argp, u.xcrs, |
2d5b5a66 SY |
5133 | sizeof(struct kvm_xcrs))) |
5134 | break; | |
5135 | r = 0; | |
5136 | break; | |
5137 | } | |
5138 | case KVM_SET_XCRS: { | |
ff5c2c03 | 5139 | u.xcrs = memdup_user(argp, sizeof(*u.xcrs)); |
9b062471 CD |
5140 | if (IS_ERR(u.xcrs)) { |
5141 | r = PTR_ERR(u.xcrs); | |
5142 | goto out_nofree; | |
5143 | } | |
2d5b5a66 | 5144 | |
d1ac91d8 | 5145 | r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs); |
2d5b5a66 SY |
5146 | break; |
5147 | } | |
92a1f12d JR |
5148 | case KVM_SET_TSC_KHZ: { |
5149 | u32 user_tsc_khz; | |
5150 | ||
5151 | r = -EINVAL; | |
92a1f12d JR |
5152 | user_tsc_khz = (u32)arg; |
5153 | ||
26769f96 MT |
5154 | if (kvm_has_tsc_control && |
5155 | user_tsc_khz >= kvm_max_guest_tsc_khz) | |
92a1f12d JR |
5156 | goto out; |
5157 | ||
cc578287 ZA |
5158 | if (user_tsc_khz == 0) |
5159 | user_tsc_khz = tsc_khz; | |
5160 | ||
381d585c HZ |
5161 | if (!kvm_set_tsc_khz(vcpu, user_tsc_khz)) |
5162 | r = 0; | |
92a1f12d | 5163 | |
92a1f12d JR |
5164 | goto out; |
5165 | } | |
5166 | case KVM_GET_TSC_KHZ: { | |
cc578287 | 5167 | r = vcpu->arch.virtual_tsc_khz; |
92a1f12d JR |
5168 | goto out; |
5169 | } | |
1c0b28c2 EM |
5170 | case KVM_KVMCLOCK_CTRL: { |
5171 | r = kvm_set_guest_paused(vcpu); | |
5172 | goto out; | |
5173 | } | |
5c919412 AS |
5174 | case KVM_ENABLE_CAP: { |
5175 | struct kvm_enable_cap cap; | |
5176 | ||
5177 | r = -EFAULT; | |
5178 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
5179 | goto out; | |
5180 | r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap); | |
5181 | break; | |
5182 | } | |
8fcc4b59 JM |
5183 | case KVM_GET_NESTED_STATE: { |
5184 | struct kvm_nested_state __user *user_kvm_nested_state = argp; | |
5185 | u32 user_data_size; | |
5186 | ||
5187 | r = -EINVAL; | |
33b22172 | 5188 | if (!kvm_x86_ops.nested_ops->get_state) |
8fcc4b59 JM |
5189 | break; |
5190 | ||
5191 | BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size)); | |
26b471c7 | 5192 | r = -EFAULT; |
8fcc4b59 | 5193 | if (get_user(user_data_size, &user_kvm_nested_state->size)) |
26b471c7 | 5194 | break; |
8fcc4b59 | 5195 | |
33b22172 PB |
5196 | r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state, |
5197 | user_data_size); | |
8fcc4b59 | 5198 | if (r < 0) |
26b471c7 | 5199 | break; |
8fcc4b59 JM |
5200 | |
5201 | if (r > user_data_size) { | |
5202 | if (put_user(r, &user_kvm_nested_state->size)) | |
26b471c7 LA |
5203 | r = -EFAULT; |
5204 | else | |
5205 | r = -E2BIG; | |
5206 | break; | |
8fcc4b59 | 5207 | } |
26b471c7 | 5208 | |
8fcc4b59 JM |
5209 | r = 0; |
5210 | break; | |
5211 | } | |
5212 | case KVM_SET_NESTED_STATE: { | |
5213 | struct kvm_nested_state __user *user_kvm_nested_state = argp; | |
5214 | struct kvm_nested_state kvm_state; | |
ad5996d9 | 5215 | int idx; |
8fcc4b59 JM |
5216 | |
5217 | r = -EINVAL; | |
33b22172 | 5218 | if (!kvm_x86_ops.nested_ops->set_state) |
8fcc4b59 JM |
5219 | break; |
5220 | ||
26b471c7 | 5221 | r = -EFAULT; |
8fcc4b59 | 5222 | if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state))) |
26b471c7 | 5223 | break; |
8fcc4b59 | 5224 | |
26b471c7 | 5225 | r = -EINVAL; |
8fcc4b59 | 5226 | if (kvm_state.size < sizeof(kvm_state)) |
26b471c7 | 5227 | break; |
8fcc4b59 JM |
5228 | |
5229 | if (kvm_state.flags & | |
8cab6507 | 5230 | ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE |
cc440cda PB |
5231 | | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING |
5232 | | KVM_STATE_NESTED_GIF_SET)) | |
26b471c7 | 5233 | break; |
8fcc4b59 JM |
5234 | |
5235 | /* nested_run_pending implies guest_mode. */ | |
8cab6507 VK |
5236 | if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING) |
5237 | && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE)) | |
26b471c7 | 5238 | break; |
8fcc4b59 | 5239 | |
ad5996d9 | 5240 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
33b22172 | 5241 | r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state); |
ad5996d9 | 5242 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
8fcc4b59 JM |
5243 | break; |
5244 | } | |
c21d54f0 VK |
5245 | case KVM_GET_SUPPORTED_HV_CPUID: |
5246 | r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp); | |
2bc39970 | 5247 | break; |
b59b153d | 5248 | #ifdef CONFIG_KVM_XEN |
3e324615 DW |
5249 | case KVM_XEN_VCPU_GET_ATTR: { |
5250 | struct kvm_xen_vcpu_attr xva; | |
5251 | ||
5252 | r = -EFAULT; | |
5253 | if (copy_from_user(&xva, argp, sizeof(xva))) | |
5254 | goto out; | |
5255 | r = kvm_xen_vcpu_get_attr(vcpu, &xva); | |
5256 | if (!r && copy_to_user(argp, &xva, sizeof(xva))) | |
5257 | r = -EFAULT; | |
5258 | break; | |
5259 | } | |
5260 | case KVM_XEN_VCPU_SET_ATTR: { | |
5261 | struct kvm_xen_vcpu_attr xva; | |
5262 | ||
5263 | r = -EFAULT; | |
5264 | if (copy_from_user(&xva, argp, sizeof(xva))) | |
5265 | goto out; | |
5266 | r = kvm_xen_vcpu_set_attr(vcpu, &xva); | |
5267 | break; | |
5268 | } | |
b59b153d | 5269 | #endif |
313a3dc7 CO |
5270 | default: |
5271 | r = -EINVAL; | |
5272 | } | |
5273 | out: | |
d1ac91d8 | 5274 | kfree(u.buffer); |
9b062471 CD |
5275 | out_nofree: |
5276 | vcpu_put(vcpu); | |
313a3dc7 CO |
5277 | return r; |
5278 | } | |
5279 | ||
1499fa80 | 5280 | vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) |
5b1c1493 CO |
5281 | { |
5282 | return VM_FAULT_SIGBUS; | |
5283 | } | |
5284 | ||
1fe779f8 CO |
5285 | static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr) |
5286 | { | |
5287 | int ret; | |
5288 | ||
5289 | if (addr > (unsigned int)(-3 * PAGE_SIZE)) | |
951179ce | 5290 | return -EINVAL; |
b3646477 | 5291 | ret = static_call(kvm_x86_set_tss_addr)(kvm, addr); |
1fe779f8 CO |
5292 | return ret; |
5293 | } | |
5294 | ||
b927a3ce SY |
5295 | static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm, |
5296 | u64 ident_addr) | |
5297 | { | |
b3646477 | 5298 | return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr); |
b927a3ce SY |
5299 | } |
5300 | ||
1fe779f8 | 5301 | static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm, |
bc8a3d89 | 5302 | unsigned long kvm_nr_mmu_pages) |
1fe779f8 CO |
5303 | { |
5304 | if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES) | |
5305 | return -EINVAL; | |
5306 | ||
79fac95e | 5307 | mutex_lock(&kvm->slots_lock); |
1fe779f8 CO |
5308 | |
5309 | kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages); | |
f05e70ac | 5310 | kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages; |
1fe779f8 | 5311 | |
79fac95e | 5312 | mutex_unlock(&kvm->slots_lock); |
1fe779f8 CO |
5313 | return 0; |
5314 | } | |
5315 | ||
bc8a3d89 | 5316 | static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm) |
1fe779f8 | 5317 | { |
39de71ec | 5318 | return kvm->arch.n_max_mmu_pages; |
1fe779f8 CO |
5319 | } |
5320 | ||
1fe779f8 CO |
5321 | static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip) |
5322 | { | |
90bca052 | 5323 | struct kvm_pic *pic = kvm->arch.vpic; |
1fe779f8 CO |
5324 | int r; |
5325 | ||
5326 | r = 0; | |
5327 | switch (chip->chip_id) { | |
5328 | case KVM_IRQCHIP_PIC_MASTER: | |
90bca052 | 5329 | memcpy(&chip->chip.pic, &pic->pics[0], |
1fe779f8 CO |
5330 | sizeof(struct kvm_pic_state)); |
5331 | break; | |
5332 | case KVM_IRQCHIP_PIC_SLAVE: | |
90bca052 | 5333 | memcpy(&chip->chip.pic, &pic->pics[1], |
1fe779f8 CO |
5334 | sizeof(struct kvm_pic_state)); |
5335 | break; | |
5336 | case KVM_IRQCHIP_IOAPIC: | |
33392b49 | 5337 | kvm_get_ioapic(kvm, &chip->chip.ioapic); |
1fe779f8 CO |
5338 | break; |
5339 | default: | |
5340 | r = -EINVAL; | |
5341 | break; | |
5342 | } | |
5343 | return r; | |
5344 | } | |
5345 | ||
5346 | static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip) | |
5347 | { | |
90bca052 | 5348 | struct kvm_pic *pic = kvm->arch.vpic; |
1fe779f8 CO |
5349 | int r; |
5350 | ||
5351 | r = 0; | |
5352 | switch (chip->chip_id) { | |
5353 | case KVM_IRQCHIP_PIC_MASTER: | |
90bca052 DH |
5354 | spin_lock(&pic->lock); |
5355 | memcpy(&pic->pics[0], &chip->chip.pic, | |
1fe779f8 | 5356 | sizeof(struct kvm_pic_state)); |
90bca052 | 5357 | spin_unlock(&pic->lock); |
1fe779f8 CO |
5358 | break; |
5359 | case KVM_IRQCHIP_PIC_SLAVE: | |
90bca052 DH |
5360 | spin_lock(&pic->lock); |
5361 | memcpy(&pic->pics[1], &chip->chip.pic, | |
1fe779f8 | 5362 | sizeof(struct kvm_pic_state)); |
90bca052 | 5363 | spin_unlock(&pic->lock); |
1fe779f8 CO |
5364 | break; |
5365 | case KVM_IRQCHIP_IOAPIC: | |
33392b49 | 5366 | kvm_set_ioapic(kvm, &chip->chip.ioapic); |
1fe779f8 CO |
5367 | break; |
5368 | default: | |
5369 | r = -EINVAL; | |
5370 | break; | |
5371 | } | |
90bca052 | 5372 | kvm_pic_update_irq(pic); |
1fe779f8 CO |
5373 | return r; |
5374 | } | |
5375 | ||
e0f63cb9 SY |
5376 | static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps) |
5377 | { | |
34f3941c RK |
5378 | struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state; |
5379 | ||
5380 | BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels)); | |
5381 | ||
5382 | mutex_lock(&kps->lock); | |
5383 | memcpy(ps, &kps->channels, sizeof(*ps)); | |
5384 | mutex_unlock(&kps->lock); | |
2da29bcc | 5385 | return 0; |
e0f63cb9 SY |
5386 | } |
5387 | ||
5388 | static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps) | |
5389 | { | |
0185604c | 5390 | int i; |
09edea72 RK |
5391 | struct kvm_pit *pit = kvm->arch.vpit; |
5392 | ||
5393 | mutex_lock(&pit->pit_state.lock); | |
34f3941c | 5394 | memcpy(&pit->pit_state.channels, ps, sizeof(*ps)); |
0185604c | 5395 | for (i = 0; i < 3; i++) |
09edea72 RK |
5396 | kvm_pit_load_count(pit, i, ps->channels[i].count, 0); |
5397 | mutex_unlock(&pit->pit_state.lock); | |
2da29bcc | 5398 | return 0; |
e9f42757 BK |
5399 | } |
5400 | ||
5401 | static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps) | |
5402 | { | |
e9f42757 BK |
5403 | mutex_lock(&kvm->arch.vpit->pit_state.lock); |
5404 | memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels, | |
5405 | sizeof(ps->channels)); | |
5406 | ps->flags = kvm->arch.vpit->pit_state.flags; | |
5407 | mutex_unlock(&kvm->arch.vpit->pit_state.lock); | |
97e69aa6 | 5408 | memset(&ps->reserved, 0, sizeof(ps->reserved)); |
2da29bcc | 5409 | return 0; |
e9f42757 BK |
5410 | } |
5411 | ||
5412 | static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps) | |
5413 | { | |
2da29bcc | 5414 | int start = 0; |
0185604c | 5415 | int i; |
e9f42757 | 5416 | u32 prev_legacy, cur_legacy; |
09edea72 RK |
5417 | struct kvm_pit *pit = kvm->arch.vpit; |
5418 | ||
5419 | mutex_lock(&pit->pit_state.lock); | |
5420 | prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY; | |
e9f42757 BK |
5421 | cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY; |
5422 | if (!prev_legacy && cur_legacy) | |
5423 | start = 1; | |
09edea72 RK |
5424 | memcpy(&pit->pit_state.channels, &ps->channels, |
5425 | sizeof(pit->pit_state.channels)); | |
5426 | pit->pit_state.flags = ps->flags; | |
0185604c | 5427 | for (i = 0; i < 3; i++) |
09edea72 | 5428 | kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count, |
e5e57e7a | 5429 | start && i == 0); |
09edea72 | 5430 | mutex_unlock(&pit->pit_state.lock); |
2da29bcc | 5431 | return 0; |
e0f63cb9 SY |
5432 | } |
5433 | ||
52d939a0 MT |
5434 | static int kvm_vm_ioctl_reinject(struct kvm *kvm, |
5435 | struct kvm_reinject_control *control) | |
5436 | { | |
71474e2f RK |
5437 | struct kvm_pit *pit = kvm->arch.vpit; |
5438 | ||
71474e2f RK |
5439 | /* pit->pit_state.lock was overloaded to prevent userspace from getting |
5440 | * an inconsistent state after running multiple KVM_REINJECT_CONTROL | |
5441 | * ioctls in parallel. Use a separate lock if that ioctl isn't rare. | |
5442 | */ | |
5443 | mutex_lock(&pit->pit_state.lock); | |
5444 | kvm_pit_set_reinject(pit, control->pit_reinject); | |
5445 | mutex_unlock(&pit->pit_state.lock); | |
b39c90b6 | 5446 | |
52d939a0 MT |
5447 | return 0; |
5448 | } | |
5449 | ||
0dff0846 | 5450 | void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot) |
5bb064dc | 5451 | { |
a018eba5 | 5452 | |
88178fd4 | 5453 | /* |
a018eba5 SC |
5454 | * Flush all CPUs' dirty log buffers to the dirty_bitmap. Called |
5455 | * before reporting dirty_bitmap to userspace. KVM flushes the buffers | |
5456 | * on all VM-Exits, thus we only need to kick running vCPUs to force a | |
5457 | * VM-Exit. | |
88178fd4 | 5458 | */ |
a018eba5 SC |
5459 | struct kvm_vcpu *vcpu; |
5460 | int i; | |
5461 | ||
5462 | kvm_for_each_vcpu(i, vcpu, kvm) | |
5463 | kvm_vcpu_kick(vcpu); | |
5bb064dc ZX |
5464 | } |
5465 | ||
aa2fbe6d YZ |
5466 | int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event, |
5467 | bool line_status) | |
23d43cf9 CD |
5468 | { |
5469 | if (!irqchip_in_kernel(kvm)) | |
5470 | return -ENXIO; | |
5471 | ||
5472 | irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, | |
aa2fbe6d YZ |
5473 | irq_event->irq, irq_event->level, |
5474 | line_status); | |
23d43cf9 CD |
5475 | return 0; |
5476 | } | |
5477 | ||
e5d83c74 PB |
5478 | int kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
5479 | struct kvm_enable_cap *cap) | |
90de4a18 NA |
5480 | { |
5481 | int r; | |
5482 | ||
5483 | if (cap->flags) | |
5484 | return -EINVAL; | |
5485 | ||
5486 | switch (cap->cap) { | |
5487 | case KVM_CAP_DISABLE_QUIRKS: | |
5488 | kvm->arch.disabled_quirks = cap->args[0]; | |
5489 | r = 0; | |
5490 | break; | |
49df6397 SR |
5491 | case KVM_CAP_SPLIT_IRQCHIP: { |
5492 | mutex_lock(&kvm->lock); | |
b053b2ae SR |
5493 | r = -EINVAL; |
5494 | if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS) | |
5495 | goto split_irqchip_unlock; | |
49df6397 SR |
5496 | r = -EEXIST; |
5497 | if (irqchip_in_kernel(kvm)) | |
5498 | goto split_irqchip_unlock; | |
557abc40 | 5499 | if (kvm->created_vcpus) |
49df6397 SR |
5500 | goto split_irqchip_unlock; |
5501 | r = kvm_setup_empty_irq_routing(kvm); | |
5c0aea0e | 5502 | if (r) |
49df6397 SR |
5503 | goto split_irqchip_unlock; |
5504 | /* Pairs with irqchip_in_kernel. */ | |
5505 | smp_wmb(); | |
49776faf | 5506 | kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT; |
b053b2ae | 5507 | kvm->arch.nr_reserved_ioapic_pins = cap->args[0]; |
49df6397 SR |
5508 | r = 0; |
5509 | split_irqchip_unlock: | |
5510 | mutex_unlock(&kvm->lock); | |
5511 | break; | |
5512 | } | |
37131313 RK |
5513 | case KVM_CAP_X2APIC_API: |
5514 | r = -EINVAL; | |
5515 | if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS) | |
5516 | break; | |
5517 | ||
5518 | if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS) | |
5519 | kvm->arch.x2apic_format = true; | |
c519265f RK |
5520 | if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK) |
5521 | kvm->arch.x2apic_broadcast_quirk_disabled = true; | |
37131313 RK |
5522 | |
5523 | r = 0; | |
5524 | break; | |
4d5422ce WL |
5525 | case KVM_CAP_X86_DISABLE_EXITS: |
5526 | r = -EINVAL; | |
5527 | if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS) | |
5528 | break; | |
5529 | ||
5530 | if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) && | |
5531 | kvm_can_mwait_in_guest()) | |
5532 | kvm->arch.mwait_in_guest = true; | |
766d3571 | 5533 | if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT) |
caa057a2 | 5534 | kvm->arch.hlt_in_guest = true; |
b31c114b WL |
5535 | if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE) |
5536 | kvm->arch.pause_in_guest = true; | |
b5170063 WL |
5537 | if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE) |
5538 | kvm->arch.cstate_in_guest = true; | |
4d5422ce WL |
5539 | r = 0; |
5540 | break; | |
6fbbde9a DS |
5541 | case KVM_CAP_MSR_PLATFORM_INFO: |
5542 | kvm->arch.guest_can_read_msr_platform_info = cap->args[0]; | |
5543 | r = 0; | |
c4f55198 JM |
5544 | break; |
5545 | case KVM_CAP_EXCEPTION_PAYLOAD: | |
5546 | kvm->arch.exception_payload_enabled = cap->args[0]; | |
5547 | r = 0; | |
6fbbde9a | 5548 | break; |
1ae09954 AG |
5549 | case KVM_CAP_X86_USER_SPACE_MSR: |
5550 | kvm->arch.user_space_msr_mask = cap->args[0]; | |
5551 | r = 0; | |
5552 | break; | |
fe6b6bc8 CQ |
5553 | case KVM_CAP_X86_BUS_LOCK_EXIT: |
5554 | r = -EINVAL; | |
5555 | if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE) | |
5556 | break; | |
5557 | ||
5558 | if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) && | |
5559 | (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)) | |
5560 | break; | |
5561 | ||
5562 | if (kvm_has_bus_lock_exit && | |
5563 | cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT) | |
5564 | kvm->arch.bus_lock_detection_enabled = true; | |
5565 | r = 0; | |
5566 | break; | |
fe7e9488 SC |
5567 | #ifdef CONFIG_X86_SGX_KVM |
5568 | case KVM_CAP_SGX_ATTRIBUTE: { | |
5569 | unsigned long allowed_attributes = 0; | |
5570 | ||
5571 | r = sgx_set_attribute(&allowed_attributes, cap->args[0]); | |
5572 | if (r) | |
5573 | break; | |
5574 | ||
5575 | /* KVM only supports the PROVISIONKEY privileged attribute. */ | |
5576 | if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) && | |
5577 | !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY)) | |
5578 | kvm->arch.sgx_provisioning_allowed = true; | |
5579 | else | |
5580 | r = -EINVAL; | |
5581 | break; | |
5582 | } | |
5583 | #endif | |
54526d1f NT |
5584 | case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM: |
5585 | r = -EINVAL; | |
5586 | if (kvm_x86_ops.vm_copy_enc_context_from) | |
5587 | r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]); | |
5588 | return r; | |
90de4a18 NA |
5589 | default: |
5590 | r = -EINVAL; | |
5591 | break; | |
5592 | } | |
5593 | return r; | |
5594 | } | |
5595 | ||
b318e8de SC |
5596 | static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow) |
5597 | { | |
5598 | struct kvm_x86_msr_filter *msr_filter; | |
5599 | ||
5600 | msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT); | |
5601 | if (!msr_filter) | |
5602 | return NULL; | |
5603 | ||
5604 | msr_filter->default_allow = default_allow; | |
5605 | return msr_filter; | |
5606 | } | |
5607 | ||
5608 | static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter) | |
1a155254 AG |
5609 | { |
5610 | u32 i; | |
1a155254 | 5611 | |
b318e8de SC |
5612 | if (!msr_filter) |
5613 | return; | |
5614 | ||
5615 | for (i = 0; i < msr_filter->count; i++) | |
5616 | kfree(msr_filter->ranges[i].bitmap); | |
1a155254 | 5617 | |
b318e8de | 5618 | kfree(msr_filter); |
1a155254 AG |
5619 | } |
5620 | ||
b318e8de SC |
5621 | static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter, |
5622 | struct kvm_msr_filter_range *user_range) | |
1a155254 | 5623 | { |
1a155254 AG |
5624 | unsigned long *bitmap = NULL; |
5625 | size_t bitmap_size; | |
1a155254 AG |
5626 | |
5627 | if (!user_range->nmsrs) | |
5628 | return 0; | |
5629 | ||
aca35288 SC |
5630 | if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE)) |
5631 | return -EINVAL; | |
5632 | ||
5633 | if (!user_range->flags) | |
5634 | return -EINVAL; | |
5635 | ||
1a155254 AG |
5636 | bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long); |
5637 | if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE) | |
5638 | return -EINVAL; | |
5639 | ||
5640 | bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size); | |
5641 | if (IS_ERR(bitmap)) | |
5642 | return PTR_ERR(bitmap); | |
5643 | ||
aca35288 | 5644 | msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) { |
1a155254 AG |
5645 | .flags = user_range->flags, |
5646 | .base = user_range->base, | |
5647 | .nmsrs = user_range->nmsrs, | |
5648 | .bitmap = bitmap, | |
5649 | }; | |
5650 | ||
b318e8de | 5651 | msr_filter->count++; |
1a155254 | 5652 | return 0; |
1a155254 AG |
5653 | } |
5654 | ||
5655 | static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp) | |
5656 | { | |
5657 | struct kvm_msr_filter __user *user_msr_filter = argp; | |
b318e8de | 5658 | struct kvm_x86_msr_filter *new_filter, *old_filter; |
1a155254 AG |
5659 | struct kvm_msr_filter filter; |
5660 | bool default_allow; | |
043248b3 | 5661 | bool empty = true; |
b318e8de | 5662 | int r = 0; |
1a155254 AG |
5663 | u32 i; |
5664 | ||
5665 | if (copy_from_user(&filter, user_msr_filter, sizeof(filter))) | |
5666 | return -EFAULT; | |
5667 | ||
043248b3 PB |
5668 | for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) |
5669 | empty &= !filter.ranges[i].nmsrs; | |
1a155254 AG |
5670 | |
5671 | default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY); | |
043248b3 PB |
5672 | if (empty && !default_allow) |
5673 | return -EINVAL; | |
5674 | ||
b318e8de SC |
5675 | new_filter = kvm_alloc_msr_filter(default_allow); |
5676 | if (!new_filter) | |
5677 | return -ENOMEM; | |
1a155254 | 5678 | |
1a155254 | 5679 | for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) { |
b318e8de SC |
5680 | r = kvm_add_msr_filter(new_filter, &filter.ranges[i]); |
5681 | if (r) { | |
5682 | kvm_free_msr_filter(new_filter); | |
5683 | return r; | |
5684 | } | |
1a155254 AG |
5685 | } |
5686 | ||
b318e8de SC |
5687 | mutex_lock(&kvm->lock); |
5688 | ||
5689 | /* The per-VM filter is protected by kvm->lock... */ | |
5690 | old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1); | |
5691 | ||
5692 | rcu_assign_pointer(kvm->arch.msr_filter, new_filter); | |
5693 | synchronize_srcu(&kvm->srcu); | |
5694 | ||
5695 | kvm_free_msr_filter(old_filter); | |
5696 | ||
1a155254 AG |
5697 | kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED); |
5698 | mutex_unlock(&kvm->lock); | |
5699 | ||
b318e8de | 5700 | return 0; |
1a155254 AG |
5701 | } |
5702 | ||
7d62874f SS |
5703 | #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER |
5704 | static int kvm_arch_suspend_notifier(struct kvm *kvm) | |
5705 | { | |
5706 | struct kvm_vcpu *vcpu; | |
5707 | int i, ret = 0; | |
5708 | ||
5709 | mutex_lock(&kvm->lock); | |
5710 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
5711 | if (!vcpu->arch.pv_time_enabled) | |
5712 | continue; | |
5713 | ||
5714 | ret = kvm_set_guest_paused(vcpu); | |
5715 | if (ret) { | |
5716 | kvm_err("Failed to pause guest VCPU%d: %d\n", | |
5717 | vcpu->vcpu_id, ret); | |
5718 | break; | |
5719 | } | |
5720 | } | |
5721 | mutex_unlock(&kvm->lock); | |
5722 | ||
5723 | return ret ? NOTIFY_BAD : NOTIFY_DONE; | |
5724 | } | |
5725 | ||
5726 | int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state) | |
5727 | { | |
5728 | switch (state) { | |
5729 | case PM_HIBERNATION_PREPARE: | |
5730 | case PM_SUSPEND_PREPARE: | |
5731 | return kvm_arch_suspend_notifier(kvm); | |
5732 | } | |
5733 | ||
5734 | return NOTIFY_DONE; | |
5735 | } | |
5736 | #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */ | |
5737 | ||
1fe779f8 CO |
5738 | long kvm_arch_vm_ioctl(struct file *filp, |
5739 | unsigned int ioctl, unsigned long arg) | |
5740 | { | |
5741 | struct kvm *kvm = filp->private_data; | |
5742 | void __user *argp = (void __user *)arg; | |
367e1319 | 5743 | int r = -ENOTTY; |
f0d66275 DH |
5744 | /* |
5745 | * This union makes it completely explicit to gcc-3.x | |
5746 | * that these two variables' stack usage should be | |
5747 | * combined, not added together. | |
5748 | */ | |
5749 | union { | |
5750 | struct kvm_pit_state ps; | |
e9f42757 | 5751 | struct kvm_pit_state2 ps2; |
c5ff41ce | 5752 | struct kvm_pit_config pit_config; |
f0d66275 | 5753 | } u; |
1fe779f8 CO |
5754 | |
5755 | switch (ioctl) { | |
5756 | case KVM_SET_TSS_ADDR: | |
5757 | r = kvm_vm_ioctl_set_tss_addr(kvm, arg); | |
1fe779f8 | 5758 | break; |
b927a3ce SY |
5759 | case KVM_SET_IDENTITY_MAP_ADDR: { |
5760 | u64 ident_addr; | |
5761 | ||
1af1ac91 DH |
5762 | mutex_lock(&kvm->lock); |
5763 | r = -EINVAL; | |
5764 | if (kvm->created_vcpus) | |
5765 | goto set_identity_unlock; | |
b927a3ce | 5766 | r = -EFAULT; |
0e96f31e | 5767 | if (copy_from_user(&ident_addr, argp, sizeof(ident_addr))) |
1af1ac91 | 5768 | goto set_identity_unlock; |
b927a3ce | 5769 | r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr); |
1af1ac91 DH |
5770 | set_identity_unlock: |
5771 | mutex_unlock(&kvm->lock); | |
b927a3ce SY |
5772 | break; |
5773 | } | |
1fe779f8 CO |
5774 | case KVM_SET_NR_MMU_PAGES: |
5775 | r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg); | |
1fe779f8 CO |
5776 | break; |
5777 | case KVM_GET_NR_MMU_PAGES: | |
5778 | r = kvm_vm_ioctl_get_nr_mmu_pages(kvm); | |
5779 | break; | |
3ddea128 | 5780 | case KVM_CREATE_IRQCHIP: { |
3ddea128 | 5781 | mutex_lock(&kvm->lock); |
09941366 | 5782 | |
3ddea128 | 5783 | r = -EEXIST; |
35e6eaa3 | 5784 | if (irqchip_in_kernel(kvm)) |
3ddea128 | 5785 | goto create_irqchip_unlock; |
09941366 | 5786 | |
3e515705 | 5787 | r = -EINVAL; |
557abc40 | 5788 | if (kvm->created_vcpus) |
3e515705 | 5789 | goto create_irqchip_unlock; |
09941366 RK |
5790 | |
5791 | r = kvm_pic_init(kvm); | |
5792 | if (r) | |
3ddea128 | 5793 | goto create_irqchip_unlock; |
09941366 RK |
5794 | |
5795 | r = kvm_ioapic_init(kvm); | |
5796 | if (r) { | |
09941366 | 5797 | kvm_pic_destroy(kvm); |
3ddea128 | 5798 | goto create_irqchip_unlock; |
09941366 RK |
5799 | } |
5800 | ||
399ec807 AK |
5801 | r = kvm_setup_default_irq_routing(kvm); |
5802 | if (r) { | |
72bb2fcd | 5803 | kvm_ioapic_destroy(kvm); |
09941366 | 5804 | kvm_pic_destroy(kvm); |
71ba994c | 5805 | goto create_irqchip_unlock; |
399ec807 | 5806 | } |
49776faf | 5807 | /* Write kvm->irq_routing before enabling irqchip_in_kernel. */ |
71ba994c | 5808 | smp_wmb(); |
49776faf | 5809 | kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL; |
3ddea128 MT |
5810 | create_irqchip_unlock: |
5811 | mutex_unlock(&kvm->lock); | |
1fe779f8 | 5812 | break; |
3ddea128 | 5813 | } |
7837699f | 5814 | case KVM_CREATE_PIT: |
c5ff41ce JK |
5815 | u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY; |
5816 | goto create_pit; | |
5817 | case KVM_CREATE_PIT2: | |
5818 | r = -EFAULT; | |
5819 | if (copy_from_user(&u.pit_config, argp, | |
5820 | sizeof(struct kvm_pit_config))) | |
5821 | goto out; | |
5822 | create_pit: | |
250715a6 | 5823 | mutex_lock(&kvm->lock); |
269e05e4 AK |
5824 | r = -EEXIST; |
5825 | if (kvm->arch.vpit) | |
5826 | goto create_pit_unlock; | |
7837699f | 5827 | r = -ENOMEM; |
c5ff41ce | 5828 | kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags); |
7837699f SY |
5829 | if (kvm->arch.vpit) |
5830 | r = 0; | |
269e05e4 | 5831 | create_pit_unlock: |
250715a6 | 5832 | mutex_unlock(&kvm->lock); |
7837699f | 5833 | break; |
1fe779f8 CO |
5834 | case KVM_GET_IRQCHIP: { |
5835 | /* 0: PIC master, 1: PIC slave, 2: IOAPIC */ | |
ff5c2c03 | 5836 | struct kvm_irqchip *chip; |
1fe779f8 | 5837 | |
ff5c2c03 SL |
5838 | chip = memdup_user(argp, sizeof(*chip)); |
5839 | if (IS_ERR(chip)) { | |
5840 | r = PTR_ERR(chip); | |
1fe779f8 | 5841 | goto out; |
ff5c2c03 SL |
5842 | } |
5843 | ||
1fe779f8 | 5844 | r = -ENXIO; |
826da321 | 5845 | if (!irqchip_kernel(kvm)) |
f0d66275 DH |
5846 | goto get_irqchip_out; |
5847 | r = kvm_vm_ioctl_get_irqchip(kvm, chip); | |
1fe779f8 | 5848 | if (r) |
f0d66275 | 5849 | goto get_irqchip_out; |
1fe779f8 | 5850 | r = -EFAULT; |
0e96f31e | 5851 | if (copy_to_user(argp, chip, sizeof(*chip))) |
f0d66275 | 5852 | goto get_irqchip_out; |
1fe779f8 | 5853 | r = 0; |
f0d66275 DH |
5854 | get_irqchip_out: |
5855 | kfree(chip); | |
1fe779f8 CO |
5856 | break; |
5857 | } | |
5858 | case KVM_SET_IRQCHIP: { | |
5859 | /* 0: PIC master, 1: PIC slave, 2: IOAPIC */ | |
ff5c2c03 | 5860 | struct kvm_irqchip *chip; |
1fe779f8 | 5861 | |
ff5c2c03 SL |
5862 | chip = memdup_user(argp, sizeof(*chip)); |
5863 | if (IS_ERR(chip)) { | |
5864 | r = PTR_ERR(chip); | |
1fe779f8 | 5865 | goto out; |
ff5c2c03 SL |
5866 | } |
5867 | ||
1fe779f8 | 5868 | r = -ENXIO; |
826da321 | 5869 | if (!irqchip_kernel(kvm)) |
f0d66275 DH |
5870 | goto set_irqchip_out; |
5871 | r = kvm_vm_ioctl_set_irqchip(kvm, chip); | |
f0d66275 DH |
5872 | set_irqchip_out: |
5873 | kfree(chip); | |
1fe779f8 CO |
5874 | break; |
5875 | } | |
e0f63cb9 | 5876 | case KVM_GET_PIT: { |
e0f63cb9 | 5877 | r = -EFAULT; |
f0d66275 | 5878 | if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state))) |
e0f63cb9 SY |
5879 | goto out; |
5880 | r = -ENXIO; | |
5881 | if (!kvm->arch.vpit) | |
5882 | goto out; | |
f0d66275 | 5883 | r = kvm_vm_ioctl_get_pit(kvm, &u.ps); |
e0f63cb9 SY |
5884 | if (r) |
5885 | goto out; | |
5886 | r = -EFAULT; | |
f0d66275 | 5887 | if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state))) |
e0f63cb9 SY |
5888 | goto out; |
5889 | r = 0; | |
5890 | break; | |
5891 | } | |
5892 | case KVM_SET_PIT: { | |
e0f63cb9 | 5893 | r = -EFAULT; |
0e96f31e | 5894 | if (copy_from_user(&u.ps, argp, sizeof(u.ps))) |
e0f63cb9 | 5895 | goto out; |
7289fdb5 | 5896 | mutex_lock(&kvm->lock); |
e0f63cb9 SY |
5897 | r = -ENXIO; |
5898 | if (!kvm->arch.vpit) | |
7289fdb5 | 5899 | goto set_pit_out; |
f0d66275 | 5900 | r = kvm_vm_ioctl_set_pit(kvm, &u.ps); |
7289fdb5 SR |
5901 | set_pit_out: |
5902 | mutex_unlock(&kvm->lock); | |
e0f63cb9 SY |
5903 | break; |
5904 | } | |
e9f42757 BK |
5905 | case KVM_GET_PIT2: { |
5906 | r = -ENXIO; | |
5907 | if (!kvm->arch.vpit) | |
5908 | goto out; | |
5909 | r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2); | |
5910 | if (r) | |
5911 | goto out; | |
5912 | r = -EFAULT; | |
5913 | if (copy_to_user(argp, &u.ps2, sizeof(u.ps2))) | |
5914 | goto out; | |
5915 | r = 0; | |
5916 | break; | |
5917 | } | |
5918 | case KVM_SET_PIT2: { | |
5919 | r = -EFAULT; | |
5920 | if (copy_from_user(&u.ps2, argp, sizeof(u.ps2))) | |
5921 | goto out; | |
7289fdb5 | 5922 | mutex_lock(&kvm->lock); |
e9f42757 BK |
5923 | r = -ENXIO; |
5924 | if (!kvm->arch.vpit) | |
7289fdb5 | 5925 | goto set_pit2_out; |
e9f42757 | 5926 | r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2); |
7289fdb5 SR |
5927 | set_pit2_out: |
5928 | mutex_unlock(&kvm->lock); | |
e9f42757 BK |
5929 | break; |
5930 | } | |
52d939a0 MT |
5931 | case KVM_REINJECT_CONTROL: { |
5932 | struct kvm_reinject_control control; | |
5933 | r = -EFAULT; | |
5934 | if (copy_from_user(&control, argp, sizeof(control))) | |
5935 | goto out; | |
cad23e72 ML |
5936 | r = -ENXIO; |
5937 | if (!kvm->arch.vpit) | |
5938 | goto out; | |
52d939a0 | 5939 | r = kvm_vm_ioctl_reinject(kvm, &control); |
52d939a0 MT |
5940 | break; |
5941 | } | |
d71ba788 PB |
5942 | case KVM_SET_BOOT_CPU_ID: |
5943 | r = 0; | |
5944 | mutex_lock(&kvm->lock); | |
557abc40 | 5945 | if (kvm->created_vcpus) |
d71ba788 PB |
5946 | r = -EBUSY; |
5947 | else | |
5948 | kvm->arch.bsp_vcpu_id = arg; | |
5949 | mutex_unlock(&kvm->lock); | |
5950 | break; | |
b59b153d | 5951 | #ifdef CONFIG_KVM_XEN |
ffde22ac | 5952 | case KVM_XEN_HVM_CONFIG: { |
51776043 | 5953 | struct kvm_xen_hvm_config xhc; |
ffde22ac | 5954 | r = -EFAULT; |
51776043 | 5955 | if (copy_from_user(&xhc, argp, sizeof(xhc))) |
ffde22ac | 5956 | goto out; |
78e9878c | 5957 | r = kvm_xen_hvm_config(kvm, &xhc); |
ffde22ac ES |
5958 | break; |
5959 | } | |
a76b9641 JM |
5960 | case KVM_XEN_HVM_GET_ATTR: { |
5961 | struct kvm_xen_hvm_attr xha; | |
5962 | ||
5963 | r = -EFAULT; | |
5964 | if (copy_from_user(&xha, argp, sizeof(xha))) | |
ffde22ac | 5965 | goto out; |
a76b9641 JM |
5966 | r = kvm_xen_hvm_get_attr(kvm, &xha); |
5967 | if (!r && copy_to_user(argp, &xha, sizeof(xha))) | |
5968 | r = -EFAULT; | |
5969 | break; | |
5970 | } | |
5971 | case KVM_XEN_HVM_SET_ATTR: { | |
5972 | struct kvm_xen_hvm_attr xha; | |
5973 | ||
5974 | r = -EFAULT; | |
5975 | if (copy_from_user(&xha, argp, sizeof(xha))) | |
5976 | goto out; | |
5977 | r = kvm_xen_hvm_set_attr(kvm, &xha); | |
ffde22ac ES |
5978 | break; |
5979 | } | |
b59b153d | 5980 | #endif |
afbcf7ab | 5981 | case KVM_SET_CLOCK: { |
77fcbe82 | 5982 | struct kvm_arch *ka = &kvm->arch; |
afbcf7ab GC |
5983 | struct kvm_clock_data user_ns; |
5984 | u64 now_ns; | |
afbcf7ab GC |
5985 | |
5986 | r = -EFAULT; | |
5987 | if (copy_from_user(&user_ns, argp, sizeof(user_ns))) | |
5988 | goto out; | |
5989 | ||
5990 | r = -EINVAL; | |
5991 | if (user_ns.flags) | |
5992 | goto out; | |
5993 | ||
5994 | r = 0; | |
0bc48bea RK |
5995 | /* |
5996 | * TODO: userspace has to take care of races with VCPU_RUN, so | |
5997 | * kvm_gen_update_masterclock() can be cut down to locked | |
5998 | * pvclock_update_vm_gtod_copy(). | |
5999 | */ | |
6000 | kvm_gen_update_masterclock(kvm); | |
77fcbe82 VK |
6001 | |
6002 | /* | |
6003 | * This pairs with kvm_guest_time_update(): when masterclock is | |
6004 | * in use, we use master_kernel_ns + kvmclock_offset to set | |
6005 | * unsigned 'system_time' so if we use get_kvmclock_ns() (which | |
6006 | * is slightly ahead) here we risk going negative on unsigned | |
6007 | * 'system_time' when 'user_ns.clock' is very small. | |
6008 | */ | |
6009 | spin_lock_irq(&ka->pvclock_gtod_sync_lock); | |
6010 | if (kvm->arch.use_master_clock) | |
6011 | now_ns = ka->master_kernel_ns; | |
6012 | else | |
6013 | now_ns = get_kvmclock_base_ns(); | |
6014 | ka->kvmclock_offset = user_ns.clock - now_ns; | |
6015 | spin_unlock_irq(&ka->pvclock_gtod_sync_lock); | |
6016 | ||
0bc48bea | 6017 | kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE); |
afbcf7ab GC |
6018 | break; |
6019 | } | |
6020 | case KVM_GET_CLOCK: { | |
afbcf7ab GC |
6021 | struct kvm_clock_data user_ns; |
6022 | u64 now_ns; | |
6023 | ||
e891a32e | 6024 | now_ns = get_kvmclock_ns(kvm); |
108b249c | 6025 | user_ns.clock = now_ns; |
e3fd9a93 | 6026 | user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0; |
97e69aa6 | 6027 | memset(&user_ns.pad, 0, sizeof(user_ns.pad)); |
afbcf7ab GC |
6028 | |
6029 | r = -EFAULT; | |
6030 | if (copy_to_user(argp, &user_ns, sizeof(user_ns))) | |
6031 | goto out; | |
6032 | r = 0; | |
6033 | break; | |
6034 | } | |
5acc5c06 BS |
6035 | case KVM_MEMORY_ENCRYPT_OP: { |
6036 | r = -ENOTTY; | |
afaf0b2f | 6037 | if (kvm_x86_ops.mem_enc_op) |
b3646477 | 6038 | r = static_call(kvm_x86_mem_enc_op)(kvm, argp); |
5acc5c06 BS |
6039 | break; |
6040 | } | |
69eaedee BS |
6041 | case KVM_MEMORY_ENCRYPT_REG_REGION: { |
6042 | struct kvm_enc_region region; | |
6043 | ||
6044 | r = -EFAULT; | |
6045 | if (copy_from_user(®ion, argp, sizeof(region))) | |
6046 | goto out; | |
6047 | ||
6048 | r = -ENOTTY; | |
afaf0b2f | 6049 | if (kvm_x86_ops.mem_enc_reg_region) |
b3646477 | 6050 | r = static_call(kvm_x86_mem_enc_reg_region)(kvm, ®ion); |
69eaedee BS |
6051 | break; |
6052 | } | |
6053 | case KVM_MEMORY_ENCRYPT_UNREG_REGION: { | |
6054 | struct kvm_enc_region region; | |
6055 | ||
6056 | r = -EFAULT; | |
6057 | if (copy_from_user(®ion, argp, sizeof(region))) | |
6058 | goto out; | |
6059 | ||
6060 | r = -ENOTTY; | |
afaf0b2f | 6061 | if (kvm_x86_ops.mem_enc_unreg_region) |
b3646477 | 6062 | r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, ®ion); |
69eaedee BS |
6063 | break; |
6064 | } | |
faeb7833 RK |
6065 | case KVM_HYPERV_EVENTFD: { |
6066 | struct kvm_hyperv_eventfd hvevfd; | |
6067 | ||
6068 | r = -EFAULT; | |
6069 | if (copy_from_user(&hvevfd, argp, sizeof(hvevfd))) | |
6070 | goto out; | |
6071 | r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd); | |
6072 | break; | |
6073 | } | |
66bb8a06 EH |
6074 | case KVM_SET_PMU_EVENT_FILTER: |
6075 | r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp); | |
6076 | break; | |
1a155254 AG |
6077 | case KVM_X86_SET_MSR_FILTER: |
6078 | r = kvm_vm_ioctl_set_msr_filter(kvm, argp); | |
6079 | break; | |
1fe779f8 | 6080 | default: |
ad6260da | 6081 | r = -ENOTTY; |
1fe779f8 CO |
6082 | } |
6083 | out: | |
6084 | return r; | |
6085 | } | |
6086 | ||
a16b043c | 6087 | static void kvm_init_msr_list(void) |
043405e1 | 6088 | { |
24c29b7a | 6089 | struct x86_pmu_capability x86_pmu; |
043405e1 | 6090 | u32 dummy[2]; |
7a5ee6ed | 6091 | unsigned i; |
043405e1 | 6092 | |
e2ada66e | 6093 | BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4, |
7a5ee6ed | 6094 | "Please update the fixed PMCs in msrs_to_saved_all[]"); |
24c29b7a PB |
6095 | |
6096 | perf_get_x86_pmu_capability(&x86_pmu); | |
e2ada66e | 6097 | |
6cbee2b9 XL |
6098 | num_msrs_to_save = 0; |
6099 | num_emulated_msrs = 0; | |
6100 | num_msr_based_features = 0; | |
6101 | ||
7a5ee6ed CQ |
6102 | for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) { |
6103 | if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0) | |
043405e1 | 6104 | continue; |
93c4adc7 PB |
6105 | |
6106 | /* | |
6107 | * Even MSRs that are valid in the host may not be exposed | |
9dbe6cf9 | 6108 | * to the guests in some cases. |
93c4adc7 | 6109 | */ |
7a5ee6ed | 6110 | switch (msrs_to_save_all[i]) { |
93c4adc7 | 6111 | case MSR_IA32_BNDCFGS: |
503234b3 | 6112 | if (!kvm_mpx_supported()) |
93c4adc7 PB |
6113 | continue; |
6114 | break; | |
9dbe6cf9 | 6115 | case MSR_TSC_AUX: |
36fa06f9 SC |
6116 | if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) && |
6117 | !kvm_cpu_cap_has(X86_FEATURE_RDPID)) | |
9dbe6cf9 PB |
6118 | continue; |
6119 | break; | |
f4cfcd2d ML |
6120 | case MSR_IA32_UMWAIT_CONTROL: |
6121 | if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG)) | |
6122 | continue; | |
6123 | break; | |
bf8c55d8 CP |
6124 | case MSR_IA32_RTIT_CTL: |
6125 | case MSR_IA32_RTIT_STATUS: | |
7b874c26 | 6126 | if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT)) |
bf8c55d8 CP |
6127 | continue; |
6128 | break; | |
6129 | case MSR_IA32_RTIT_CR3_MATCH: | |
7b874c26 | 6130 | if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) || |
bf8c55d8 CP |
6131 | !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering)) |
6132 | continue; | |
6133 | break; | |
6134 | case MSR_IA32_RTIT_OUTPUT_BASE: | |
6135 | case MSR_IA32_RTIT_OUTPUT_MASK: | |
7b874c26 | 6136 | if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) || |
bf8c55d8 CP |
6137 | (!intel_pt_validate_hw_cap(PT_CAP_topa_output) && |
6138 | !intel_pt_validate_hw_cap(PT_CAP_single_range_output))) | |
6139 | continue; | |
6140 | break; | |
7cb85fc4 | 6141 | case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: |
7b874c26 | 6142 | if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) || |
7a5ee6ed | 6143 | msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >= |
bf8c55d8 CP |
6144 | intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2) |
6145 | continue; | |
6146 | break; | |
cf05a67b | 6147 | case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17: |
7a5ee6ed | 6148 | if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >= |
24c29b7a PB |
6149 | min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp)) |
6150 | continue; | |
6151 | break; | |
cf05a67b | 6152 | case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17: |
7a5ee6ed | 6153 | if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >= |
24c29b7a PB |
6154 | min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp)) |
6155 | continue; | |
7cb85fc4 | 6156 | break; |
93c4adc7 PB |
6157 | default: |
6158 | break; | |
6159 | } | |
6160 | ||
7a5ee6ed | 6161 | msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i]; |
043405e1 | 6162 | } |
62ef68bb | 6163 | |
7a5ee6ed | 6164 | for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) { |
b3646477 | 6165 | if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i])) |
bc226f07 | 6166 | continue; |
62ef68bb | 6167 | |
7a5ee6ed | 6168 | emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i]; |
62ef68bb | 6169 | } |
801e459a | 6170 | |
7a5ee6ed | 6171 | for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) { |
801e459a TL |
6172 | struct kvm_msr_entry msr; |
6173 | ||
7a5ee6ed | 6174 | msr.index = msr_based_features_all[i]; |
66421c1e | 6175 | if (kvm_get_msr_feature(&msr)) |
801e459a TL |
6176 | continue; |
6177 | ||
7a5ee6ed | 6178 | msr_based_features[num_msr_based_features++] = msr_based_features_all[i]; |
801e459a | 6179 | } |
043405e1 CO |
6180 | } |
6181 | ||
bda9020e MT |
6182 | static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len, |
6183 | const void *v) | |
bbd9b64e | 6184 | { |
70252a10 AK |
6185 | int handled = 0; |
6186 | int n; | |
6187 | ||
6188 | do { | |
6189 | n = min(len, 8); | |
bce87cce | 6190 | if (!(lapic_in_kernel(vcpu) && |
e32edf4f NN |
6191 | !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v)) |
6192 | && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v)) | |
70252a10 AK |
6193 | break; |
6194 | handled += n; | |
6195 | addr += n; | |
6196 | len -= n; | |
6197 | v += n; | |
6198 | } while (len); | |
bbd9b64e | 6199 | |
70252a10 | 6200 | return handled; |
bbd9b64e CO |
6201 | } |
6202 | ||
bda9020e | 6203 | static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v) |
bbd9b64e | 6204 | { |
70252a10 AK |
6205 | int handled = 0; |
6206 | int n; | |
6207 | ||
6208 | do { | |
6209 | n = min(len, 8); | |
bce87cce | 6210 | if (!(lapic_in_kernel(vcpu) && |
e32edf4f NN |
6211 | !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev, |
6212 | addr, n, v)) | |
6213 | && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v)) | |
70252a10 | 6214 | break; |
e39d200f | 6215 | trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v); |
70252a10 AK |
6216 | handled += n; |
6217 | addr += n; | |
6218 | len -= n; | |
6219 | v += n; | |
6220 | } while (len); | |
bbd9b64e | 6221 | |
70252a10 | 6222 | return handled; |
bbd9b64e CO |
6223 | } |
6224 | ||
2dafc6c2 GN |
6225 | static void kvm_set_segment(struct kvm_vcpu *vcpu, |
6226 | struct kvm_segment *var, int seg) | |
6227 | { | |
b3646477 | 6228 | static_call(kvm_x86_set_segment)(vcpu, var, seg); |
2dafc6c2 GN |
6229 | } |
6230 | ||
6231 | void kvm_get_segment(struct kvm_vcpu *vcpu, | |
6232 | struct kvm_segment *var, int seg) | |
6233 | { | |
b3646477 | 6234 | static_call(kvm_x86_get_segment)(vcpu, var, seg); |
2dafc6c2 GN |
6235 | } |
6236 | ||
54987b7a PB |
6237 | gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access, |
6238 | struct x86_exception *exception) | |
02f59dc9 JR |
6239 | { |
6240 | gpa_t t_gpa; | |
02f59dc9 JR |
6241 | |
6242 | BUG_ON(!mmu_is_nested(vcpu)); | |
6243 | ||
6244 | /* NPT walks are always user-walks */ | |
6245 | access |= PFERR_USER_MASK; | |
44dd3ffa | 6246 | t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception); |
02f59dc9 JR |
6247 | |
6248 | return t_gpa; | |
6249 | } | |
6250 | ||
ab9ae313 AK |
6251 | gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, |
6252 | struct x86_exception *exception) | |
1871c602 | 6253 | { |
b3646477 | 6254 | u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0; |
ab9ae313 | 6255 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
1871c602 | 6256 | } |
54f958cd | 6257 | EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read); |
1871c602 | 6258 | |
ab9ae313 AK |
6259 | gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, |
6260 | struct x86_exception *exception) | |
1871c602 | 6261 | { |
b3646477 | 6262 | u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0; |
1871c602 | 6263 | access |= PFERR_FETCH_MASK; |
ab9ae313 | 6264 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
1871c602 GN |
6265 | } |
6266 | ||
ab9ae313 AK |
6267 | gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, |
6268 | struct x86_exception *exception) | |
1871c602 | 6269 | { |
b3646477 | 6270 | u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0; |
1871c602 | 6271 | access |= PFERR_WRITE_MASK; |
ab9ae313 | 6272 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
1871c602 | 6273 | } |
54f958cd | 6274 | EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write); |
1871c602 GN |
6275 | |
6276 | /* uses this to access any guest's mapped memory without checking CPL */ | |
ab9ae313 AK |
6277 | gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, |
6278 | struct x86_exception *exception) | |
1871c602 | 6279 | { |
ab9ae313 | 6280 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception); |
1871c602 GN |
6281 | } |
6282 | ||
6283 | static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes, | |
6284 | struct kvm_vcpu *vcpu, u32 access, | |
bcc55cba | 6285 | struct x86_exception *exception) |
bbd9b64e CO |
6286 | { |
6287 | void *data = val; | |
10589a46 | 6288 | int r = X86EMUL_CONTINUE; |
bbd9b64e CO |
6289 | |
6290 | while (bytes) { | |
14dfe855 | 6291 | gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access, |
ab9ae313 | 6292 | exception); |
bbd9b64e | 6293 | unsigned offset = addr & (PAGE_SIZE-1); |
77c2002e | 6294 | unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset); |
bbd9b64e CO |
6295 | int ret; |
6296 | ||
bcc55cba | 6297 | if (gpa == UNMAPPED_GVA) |
ab9ae313 | 6298 | return X86EMUL_PROPAGATE_FAULT; |
54bf36aa PB |
6299 | ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data, |
6300 | offset, toread); | |
10589a46 | 6301 | if (ret < 0) { |
c3cd7ffa | 6302 | r = X86EMUL_IO_NEEDED; |
10589a46 MT |
6303 | goto out; |
6304 | } | |
bbd9b64e | 6305 | |
77c2002e IE |
6306 | bytes -= toread; |
6307 | data += toread; | |
6308 | addr += toread; | |
bbd9b64e | 6309 | } |
10589a46 | 6310 | out: |
10589a46 | 6311 | return r; |
bbd9b64e | 6312 | } |
77c2002e | 6313 | |
1871c602 | 6314 | /* used for instruction fetching */ |
0f65dd70 AK |
6315 | static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt, |
6316 | gva_t addr, void *val, unsigned int bytes, | |
bcc55cba | 6317 | struct x86_exception *exception) |
1871c602 | 6318 | { |
0f65dd70 | 6319 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
b3646477 | 6320 | u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0; |
44583cba PB |
6321 | unsigned offset; |
6322 | int ret; | |
0f65dd70 | 6323 | |
44583cba PB |
6324 | /* Inline kvm_read_guest_virt_helper for speed. */ |
6325 | gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK, | |
6326 | exception); | |
6327 | if (unlikely(gpa == UNMAPPED_GVA)) | |
6328 | return X86EMUL_PROPAGATE_FAULT; | |
6329 | ||
6330 | offset = addr & (PAGE_SIZE-1); | |
6331 | if (WARN_ON(offset + bytes > PAGE_SIZE)) | |
6332 | bytes = (unsigned)PAGE_SIZE - offset; | |
54bf36aa PB |
6333 | ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val, |
6334 | offset, bytes); | |
44583cba PB |
6335 | if (unlikely(ret < 0)) |
6336 | return X86EMUL_IO_NEEDED; | |
6337 | ||
6338 | return X86EMUL_CONTINUE; | |
1871c602 GN |
6339 | } |
6340 | ||
ce14e868 | 6341 | int kvm_read_guest_virt(struct kvm_vcpu *vcpu, |
0f65dd70 | 6342 | gva_t addr, void *val, unsigned int bytes, |
bcc55cba | 6343 | struct x86_exception *exception) |
1871c602 | 6344 | { |
b3646477 | 6345 | u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0; |
0f65dd70 | 6346 | |
353c0956 PB |
6347 | /* |
6348 | * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED | |
6349 | * is returned, but our callers are not ready for that and they blindly | |
6350 | * call kvm_inject_page_fault. Ensure that they at least do not leak | |
6351 | * uninitialized kernel stack memory into cr2 and error code. | |
6352 | */ | |
6353 | memset(exception, 0, sizeof(*exception)); | |
1871c602 | 6354 | return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, |
bcc55cba | 6355 | exception); |
1871c602 | 6356 | } |
064aea77 | 6357 | EXPORT_SYMBOL_GPL(kvm_read_guest_virt); |
1871c602 | 6358 | |
ce14e868 PB |
6359 | static int emulator_read_std(struct x86_emulate_ctxt *ctxt, |
6360 | gva_t addr, void *val, unsigned int bytes, | |
3c9fa24c | 6361 | struct x86_exception *exception, bool system) |
1871c602 | 6362 | { |
0f65dd70 | 6363 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
3c9fa24c PB |
6364 | u32 access = 0; |
6365 | ||
b3646477 | 6366 | if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3) |
3c9fa24c PB |
6367 | access |= PFERR_USER_MASK; |
6368 | ||
6369 | return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception); | |
1871c602 GN |
6370 | } |
6371 | ||
7a036a6f RK |
6372 | static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt, |
6373 | unsigned long addr, void *val, unsigned int bytes) | |
6374 | { | |
6375 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
6376 | int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes); | |
6377 | ||
6378 | return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE; | |
6379 | } | |
6380 | ||
ce14e868 PB |
6381 | static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes, |
6382 | struct kvm_vcpu *vcpu, u32 access, | |
6383 | struct x86_exception *exception) | |
77c2002e IE |
6384 | { |
6385 | void *data = val; | |
6386 | int r = X86EMUL_CONTINUE; | |
6387 | ||
6388 | while (bytes) { | |
14dfe855 | 6389 | gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, |
ce14e868 | 6390 | access, |
ab9ae313 | 6391 | exception); |
77c2002e IE |
6392 | unsigned offset = addr & (PAGE_SIZE-1); |
6393 | unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset); | |
6394 | int ret; | |
6395 | ||
bcc55cba | 6396 | if (gpa == UNMAPPED_GVA) |
ab9ae313 | 6397 | return X86EMUL_PROPAGATE_FAULT; |
54bf36aa | 6398 | ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite); |
77c2002e | 6399 | if (ret < 0) { |
c3cd7ffa | 6400 | r = X86EMUL_IO_NEEDED; |
77c2002e IE |
6401 | goto out; |
6402 | } | |
6403 | ||
6404 | bytes -= towrite; | |
6405 | data += towrite; | |
6406 | addr += towrite; | |
6407 | } | |
6408 | out: | |
6409 | return r; | |
6410 | } | |
ce14e868 PB |
6411 | |
6412 | static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val, | |
3c9fa24c PB |
6413 | unsigned int bytes, struct x86_exception *exception, |
6414 | bool system) | |
ce14e868 PB |
6415 | { |
6416 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
3c9fa24c PB |
6417 | u32 access = PFERR_WRITE_MASK; |
6418 | ||
b3646477 | 6419 | if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3) |
3c9fa24c | 6420 | access |= PFERR_USER_MASK; |
ce14e868 PB |
6421 | |
6422 | return kvm_write_guest_virt_helper(addr, val, bytes, vcpu, | |
3c9fa24c | 6423 | access, exception); |
ce14e868 PB |
6424 | } |
6425 | ||
6426 | int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val, | |
6427 | unsigned int bytes, struct x86_exception *exception) | |
6428 | { | |
c595ceee PB |
6429 | /* kvm_write_guest_virt_system can pull in tons of pages. */ |
6430 | vcpu->arch.l1tf_flush_l1d = true; | |
6431 | ||
ce14e868 PB |
6432 | return kvm_write_guest_virt_helper(addr, val, bytes, vcpu, |
6433 | PFERR_WRITE_MASK, exception); | |
6434 | } | |
6a4d7550 | 6435 | EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system); |
77c2002e | 6436 | |
082d06ed WL |
6437 | int handle_ud(struct kvm_vcpu *vcpu) |
6438 | { | |
b3dc0695 | 6439 | static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX }; |
6c86eedc | 6440 | int emul_type = EMULTYPE_TRAP_UD; |
6c86eedc WL |
6441 | char sig[5]; /* ud2; .ascii "kvm" */ |
6442 | struct x86_exception e; | |
6443 | ||
b3646477 | 6444 | if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0))) |
09e3e2a1 SC |
6445 | return 1; |
6446 | ||
6c86eedc | 6447 | if (force_emulation_prefix && |
3c9fa24c PB |
6448 | kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu), |
6449 | sig, sizeof(sig), &e) == 0 && | |
b3dc0695 | 6450 | memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) { |
6c86eedc | 6451 | kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig)); |
b4000606 | 6452 | emul_type = EMULTYPE_TRAP_UD_FORCED; |
6c86eedc | 6453 | } |
082d06ed | 6454 | |
60fc3d02 | 6455 | return kvm_emulate_instruction(vcpu, emul_type); |
082d06ed WL |
6456 | } |
6457 | EXPORT_SYMBOL_GPL(handle_ud); | |
6458 | ||
0f89b207 TL |
6459 | static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva, |
6460 | gpa_t gpa, bool write) | |
6461 | { | |
6462 | /* For APIC access vmexit */ | |
6463 | if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE) | |
6464 | return 1; | |
6465 | ||
6466 | if (vcpu_match_mmio_gpa(vcpu, gpa)) { | |
6467 | trace_vcpu_match_mmio(gva, gpa, write, true); | |
6468 | return 1; | |
6469 | } | |
6470 | ||
6471 | return 0; | |
6472 | } | |
6473 | ||
af7cc7d1 XG |
6474 | static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva, |
6475 | gpa_t *gpa, struct x86_exception *exception, | |
6476 | bool write) | |
6477 | { | |
b3646477 | 6478 | u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0) |
97d64b78 | 6479 | | (write ? PFERR_WRITE_MASK : 0); |
af7cc7d1 | 6480 | |
be94f6b7 HH |
6481 | /* |
6482 | * currently PKRU is only applied to ept enabled guest so | |
6483 | * there is no pkey in EPT page table for L1 guest or EPT | |
6484 | * shadow page table for L2 guest. | |
6485 | */ | |
97d64b78 | 6486 | if (vcpu_match_mmio_gva(vcpu, gva) |
97ec8c06 | 6487 | && !permission_fault(vcpu, vcpu->arch.walk_mmu, |
871bd034 | 6488 | vcpu->arch.mmio_access, 0, access)) { |
bebb106a XG |
6489 | *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT | |
6490 | (gva & (PAGE_SIZE - 1)); | |
4f022648 | 6491 | trace_vcpu_match_mmio(gva, *gpa, write, false); |
bebb106a XG |
6492 | return 1; |
6493 | } | |
6494 | ||
af7cc7d1 XG |
6495 | *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
6496 | ||
6497 | if (*gpa == UNMAPPED_GVA) | |
6498 | return -1; | |
6499 | ||
0f89b207 | 6500 | return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write); |
af7cc7d1 XG |
6501 | } |
6502 | ||
3200f405 | 6503 | int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa, |
bcc55cba | 6504 | const void *val, int bytes) |
bbd9b64e CO |
6505 | { |
6506 | int ret; | |
6507 | ||
54bf36aa | 6508 | ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes); |
9f811285 | 6509 | if (ret < 0) |
bbd9b64e | 6510 | return 0; |
0eb05bf2 | 6511 | kvm_page_track_write(vcpu, gpa, val, bytes); |
bbd9b64e CO |
6512 | return 1; |
6513 | } | |
6514 | ||
77d197b2 XG |
6515 | struct read_write_emulator_ops { |
6516 | int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val, | |
6517 | int bytes); | |
6518 | int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa, | |
6519 | void *val, int bytes); | |
6520 | int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa, | |
6521 | int bytes, void *val); | |
6522 | int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa, | |
6523 | void *val, int bytes); | |
6524 | bool write; | |
6525 | }; | |
6526 | ||
6527 | static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes) | |
6528 | { | |
6529 | if (vcpu->mmio_read_completed) { | |
77d197b2 | 6530 | trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, |
e39d200f | 6531 | vcpu->mmio_fragments[0].gpa, val); |
77d197b2 XG |
6532 | vcpu->mmio_read_completed = 0; |
6533 | return 1; | |
6534 | } | |
6535 | ||
6536 | return 0; | |
6537 | } | |
6538 | ||
6539 | static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa, | |
6540 | void *val, int bytes) | |
6541 | { | |
54bf36aa | 6542 | return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes); |
77d197b2 XG |
6543 | } |
6544 | ||
6545 | static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa, | |
6546 | void *val, int bytes) | |
6547 | { | |
6548 | return emulator_write_phys(vcpu, gpa, val, bytes); | |
6549 | } | |
6550 | ||
6551 | static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val) | |
6552 | { | |
e39d200f | 6553 | trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val); |
77d197b2 XG |
6554 | return vcpu_mmio_write(vcpu, gpa, bytes, val); |
6555 | } | |
6556 | ||
6557 | static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, | |
6558 | void *val, int bytes) | |
6559 | { | |
e39d200f | 6560 | trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL); |
77d197b2 XG |
6561 | return X86EMUL_IO_NEEDED; |
6562 | } | |
6563 | ||
6564 | static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, | |
6565 | void *val, int bytes) | |
6566 | { | |
f78146b0 AK |
6567 | struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0]; |
6568 | ||
87da7e66 | 6569 | memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len)); |
77d197b2 XG |
6570 | return X86EMUL_CONTINUE; |
6571 | } | |
6572 | ||
0fbe9b0b | 6573 | static const struct read_write_emulator_ops read_emultor = { |
77d197b2 XG |
6574 | .read_write_prepare = read_prepare, |
6575 | .read_write_emulate = read_emulate, | |
6576 | .read_write_mmio = vcpu_mmio_read, | |
6577 | .read_write_exit_mmio = read_exit_mmio, | |
6578 | }; | |
6579 | ||
0fbe9b0b | 6580 | static const struct read_write_emulator_ops write_emultor = { |
77d197b2 XG |
6581 | .read_write_emulate = write_emulate, |
6582 | .read_write_mmio = write_mmio, | |
6583 | .read_write_exit_mmio = write_exit_mmio, | |
6584 | .write = true, | |
6585 | }; | |
6586 | ||
22388a3c XG |
6587 | static int emulator_read_write_onepage(unsigned long addr, void *val, |
6588 | unsigned int bytes, | |
6589 | struct x86_exception *exception, | |
6590 | struct kvm_vcpu *vcpu, | |
0fbe9b0b | 6591 | const struct read_write_emulator_ops *ops) |
bbd9b64e | 6592 | { |
af7cc7d1 XG |
6593 | gpa_t gpa; |
6594 | int handled, ret; | |
22388a3c | 6595 | bool write = ops->write; |
f78146b0 | 6596 | struct kvm_mmio_fragment *frag; |
c9b8b07c | 6597 | struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt; |
0f89b207 TL |
6598 | |
6599 | /* | |
6600 | * If the exit was due to a NPF we may already have a GPA. | |
6601 | * If the GPA is present, use it to avoid the GVA to GPA table walk. | |
6602 | * Note, this cannot be used on string operations since string | |
6603 | * operation using rep will only have the initial GPA from the NPF | |
6604 | * occurred. | |
6605 | */ | |
744e699c SC |
6606 | if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) && |
6607 | (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) { | |
6608 | gpa = ctxt->gpa_val; | |
618232e2 BS |
6609 | ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write); |
6610 | } else { | |
6611 | ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write); | |
6612 | if (ret < 0) | |
6613 | return X86EMUL_PROPAGATE_FAULT; | |
0f89b207 | 6614 | } |
10589a46 | 6615 | |
618232e2 | 6616 | if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes)) |
bbd9b64e CO |
6617 | return X86EMUL_CONTINUE; |
6618 | ||
bbd9b64e CO |
6619 | /* |
6620 | * Is this MMIO handled locally? | |
6621 | */ | |
22388a3c | 6622 | handled = ops->read_write_mmio(vcpu, gpa, bytes, val); |
70252a10 | 6623 | if (handled == bytes) |
bbd9b64e | 6624 | return X86EMUL_CONTINUE; |
bbd9b64e | 6625 | |
70252a10 AK |
6626 | gpa += handled; |
6627 | bytes -= handled; | |
6628 | val += handled; | |
6629 | ||
87da7e66 XG |
6630 | WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS); |
6631 | frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++]; | |
6632 | frag->gpa = gpa; | |
6633 | frag->data = val; | |
6634 | frag->len = bytes; | |
f78146b0 | 6635 | return X86EMUL_CONTINUE; |
bbd9b64e CO |
6636 | } |
6637 | ||
52eb5a6d XL |
6638 | static int emulator_read_write(struct x86_emulate_ctxt *ctxt, |
6639 | unsigned long addr, | |
22388a3c XG |
6640 | void *val, unsigned int bytes, |
6641 | struct x86_exception *exception, | |
0fbe9b0b | 6642 | const struct read_write_emulator_ops *ops) |
bbd9b64e | 6643 | { |
0f65dd70 | 6644 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
f78146b0 AK |
6645 | gpa_t gpa; |
6646 | int rc; | |
6647 | ||
6648 | if (ops->read_write_prepare && | |
6649 | ops->read_write_prepare(vcpu, val, bytes)) | |
6650 | return X86EMUL_CONTINUE; | |
6651 | ||
6652 | vcpu->mmio_nr_fragments = 0; | |
0f65dd70 | 6653 | |
bbd9b64e CO |
6654 | /* Crossing a page boundary? */ |
6655 | if (((addr + bytes - 1) ^ addr) & PAGE_MASK) { | |
f78146b0 | 6656 | int now; |
bbd9b64e CO |
6657 | |
6658 | now = -addr & ~PAGE_MASK; | |
22388a3c XG |
6659 | rc = emulator_read_write_onepage(addr, val, now, exception, |
6660 | vcpu, ops); | |
6661 | ||
bbd9b64e CO |
6662 | if (rc != X86EMUL_CONTINUE) |
6663 | return rc; | |
6664 | addr += now; | |
bac15531 NA |
6665 | if (ctxt->mode != X86EMUL_MODE_PROT64) |
6666 | addr = (u32)addr; | |
bbd9b64e CO |
6667 | val += now; |
6668 | bytes -= now; | |
6669 | } | |
22388a3c | 6670 | |
f78146b0 AK |
6671 | rc = emulator_read_write_onepage(addr, val, bytes, exception, |
6672 | vcpu, ops); | |
6673 | if (rc != X86EMUL_CONTINUE) | |
6674 | return rc; | |
6675 | ||
6676 | if (!vcpu->mmio_nr_fragments) | |
6677 | return rc; | |
6678 | ||
6679 | gpa = vcpu->mmio_fragments[0].gpa; | |
6680 | ||
6681 | vcpu->mmio_needed = 1; | |
6682 | vcpu->mmio_cur_fragment = 0; | |
6683 | ||
87da7e66 | 6684 | vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len); |
f78146b0 AK |
6685 | vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write; |
6686 | vcpu->run->exit_reason = KVM_EXIT_MMIO; | |
6687 | vcpu->run->mmio.phys_addr = gpa; | |
6688 | ||
6689 | return ops->read_write_exit_mmio(vcpu, gpa, val, bytes); | |
22388a3c XG |
6690 | } |
6691 | ||
6692 | static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt, | |
6693 | unsigned long addr, | |
6694 | void *val, | |
6695 | unsigned int bytes, | |
6696 | struct x86_exception *exception) | |
6697 | { | |
6698 | return emulator_read_write(ctxt, addr, val, bytes, | |
6699 | exception, &read_emultor); | |
6700 | } | |
6701 | ||
52eb5a6d | 6702 | static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt, |
22388a3c XG |
6703 | unsigned long addr, |
6704 | const void *val, | |
6705 | unsigned int bytes, | |
6706 | struct x86_exception *exception) | |
6707 | { | |
6708 | return emulator_read_write(ctxt, addr, (void *)val, bytes, | |
6709 | exception, &write_emultor); | |
bbd9b64e | 6710 | } |
bbd9b64e | 6711 | |
daea3e73 AK |
6712 | #define CMPXCHG_TYPE(t, ptr, old, new) \ |
6713 | (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old)) | |
6714 | ||
6715 | #ifdef CONFIG_X86_64 | |
6716 | # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new) | |
6717 | #else | |
6718 | # define CMPXCHG64(ptr, old, new) \ | |
9749a6c0 | 6719 | (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old)) |
daea3e73 AK |
6720 | #endif |
6721 | ||
0f65dd70 AK |
6722 | static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt, |
6723 | unsigned long addr, | |
bbd9b64e CO |
6724 | const void *old, |
6725 | const void *new, | |
6726 | unsigned int bytes, | |
0f65dd70 | 6727 | struct x86_exception *exception) |
bbd9b64e | 6728 | { |
42e35f80 | 6729 | struct kvm_host_map map; |
0f65dd70 | 6730 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
9de6fe3c | 6731 | u64 page_line_mask; |
daea3e73 | 6732 | gpa_t gpa; |
daea3e73 AK |
6733 | char *kaddr; |
6734 | bool exchanged; | |
2bacc55c | 6735 | |
daea3e73 AK |
6736 | /* guests cmpxchg8b have to be emulated atomically */ |
6737 | if (bytes > 8 || (bytes & (bytes - 1))) | |
6738 | goto emul_write; | |
10589a46 | 6739 | |
daea3e73 | 6740 | gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL); |
2bacc55c | 6741 | |
daea3e73 AK |
6742 | if (gpa == UNMAPPED_GVA || |
6743 | (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE) | |
6744 | goto emul_write; | |
2bacc55c | 6745 | |
9de6fe3c XL |
6746 | /* |
6747 | * Emulate the atomic as a straight write to avoid #AC if SLD is | |
6748 | * enabled in the host and the access splits a cache line. | |
6749 | */ | |
6750 | if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT)) | |
6751 | page_line_mask = ~(cache_line_size() - 1); | |
6752 | else | |
6753 | page_line_mask = PAGE_MASK; | |
6754 | ||
6755 | if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask)) | |
daea3e73 | 6756 | goto emul_write; |
72dc67a6 | 6757 | |
42e35f80 | 6758 | if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map)) |
c19b8bd6 | 6759 | goto emul_write; |
72dc67a6 | 6760 | |
42e35f80 KA |
6761 | kaddr = map.hva + offset_in_page(gpa); |
6762 | ||
daea3e73 AK |
6763 | switch (bytes) { |
6764 | case 1: | |
6765 | exchanged = CMPXCHG_TYPE(u8, kaddr, old, new); | |
6766 | break; | |
6767 | case 2: | |
6768 | exchanged = CMPXCHG_TYPE(u16, kaddr, old, new); | |
6769 | break; | |
6770 | case 4: | |
6771 | exchanged = CMPXCHG_TYPE(u32, kaddr, old, new); | |
6772 | break; | |
6773 | case 8: | |
6774 | exchanged = CMPXCHG64(kaddr, old, new); | |
6775 | break; | |
6776 | default: | |
6777 | BUG(); | |
2bacc55c | 6778 | } |
42e35f80 KA |
6779 | |
6780 | kvm_vcpu_unmap(vcpu, &map, true); | |
daea3e73 AK |
6781 | |
6782 | if (!exchanged) | |
6783 | return X86EMUL_CMPXCHG_FAILED; | |
6784 | ||
0eb05bf2 | 6785 | kvm_page_track_write(vcpu, gpa, new, bytes); |
8f6abd06 GN |
6786 | |
6787 | return X86EMUL_CONTINUE; | |
4a5f48f6 | 6788 | |
3200f405 | 6789 | emul_write: |
daea3e73 | 6790 | printk_once(KERN_WARNING "kvm: emulating exchange as write\n"); |
2bacc55c | 6791 | |
0f65dd70 | 6792 | return emulator_write_emulated(ctxt, addr, new, bytes, exception); |
bbd9b64e CO |
6793 | } |
6794 | ||
cf8f70bf GN |
6795 | static int kernel_pio(struct kvm_vcpu *vcpu, void *pd) |
6796 | { | |
cbfc6c91 | 6797 | int r = 0, i; |
cf8f70bf | 6798 | |
cbfc6c91 WL |
6799 | for (i = 0; i < vcpu->arch.pio.count; i++) { |
6800 | if (vcpu->arch.pio.in) | |
6801 | r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port, | |
6802 | vcpu->arch.pio.size, pd); | |
6803 | else | |
6804 | r = kvm_io_bus_write(vcpu, KVM_PIO_BUS, | |
6805 | vcpu->arch.pio.port, vcpu->arch.pio.size, | |
6806 | pd); | |
6807 | if (r) | |
6808 | break; | |
6809 | pd += vcpu->arch.pio.size; | |
6810 | } | |
cf8f70bf GN |
6811 | return r; |
6812 | } | |
6813 | ||
6f6fbe98 XG |
6814 | static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size, |
6815 | unsigned short port, void *val, | |
6816 | unsigned int count, bool in) | |
cf8f70bf | 6817 | { |
cf8f70bf | 6818 | vcpu->arch.pio.port = port; |
6f6fbe98 | 6819 | vcpu->arch.pio.in = in; |
7972995b | 6820 | vcpu->arch.pio.count = count; |
cf8f70bf GN |
6821 | vcpu->arch.pio.size = size; |
6822 | ||
6823 | if (!kernel_pio(vcpu, vcpu->arch.pio_data)) { | |
7972995b | 6824 | vcpu->arch.pio.count = 0; |
cf8f70bf GN |
6825 | return 1; |
6826 | } | |
6827 | ||
6828 | vcpu->run->exit_reason = KVM_EXIT_IO; | |
6f6fbe98 | 6829 | vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT; |
cf8f70bf GN |
6830 | vcpu->run->io.size = size; |
6831 | vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE; | |
6832 | vcpu->run->io.count = count; | |
6833 | vcpu->run->io.port = port; | |
6834 | ||
6835 | return 0; | |
6836 | } | |
6837 | ||
2e3bb4d8 SC |
6838 | static int emulator_pio_in(struct kvm_vcpu *vcpu, int size, |
6839 | unsigned short port, void *val, unsigned int count) | |
cf8f70bf | 6840 | { |
6f6fbe98 | 6841 | int ret; |
ca1d4a9e | 6842 | |
6f6fbe98 XG |
6843 | if (vcpu->arch.pio.count) |
6844 | goto data_avail; | |
cf8f70bf | 6845 | |
cbfc6c91 WL |
6846 | memset(vcpu->arch.pio_data, 0, size * count); |
6847 | ||
6f6fbe98 XG |
6848 | ret = emulator_pio_in_out(vcpu, size, port, val, count, true); |
6849 | if (ret) { | |
6850 | data_avail: | |
6851 | memcpy(val, vcpu->arch.pio_data, size * count); | |
1171903d | 6852 | trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data); |
7972995b | 6853 | vcpu->arch.pio.count = 0; |
cf8f70bf GN |
6854 | return 1; |
6855 | } | |
6856 | ||
cf8f70bf GN |
6857 | return 0; |
6858 | } | |
6859 | ||
2e3bb4d8 SC |
6860 | static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt, |
6861 | int size, unsigned short port, void *val, | |
6862 | unsigned int count) | |
6f6fbe98 | 6863 | { |
2e3bb4d8 | 6864 | return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count); |
6f6fbe98 | 6865 | |
2e3bb4d8 | 6866 | } |
6f6fbe98 | 6867 | |
2e3bb4d8 SC |
6868 | static int emulator_pio_out(struct kvm_vcpu *vcpu, int size, |
6869 | unsigned short port, const void *val, | |
6870 | unsigned int count) | |
6871 | { | |
6f6fbe98 | 6872 | memcpy(vcpu->arch.pio_data, val, size * count); |
1171903d | 6873 | trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data); |
6f6fbe98 XG |
6874 | return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false); |
6875 | } | |
6876 | ||
2e3bb4d8 SC |
6877 | static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt, |
6878 | int size, unsigned short port, | |
6879 | const void *val, unsigned int count) | |
6880 | { | |
6881 | return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count); | |
6882 | } | |
6883 | ||
bbd9b64e CO |
6884 | static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg) |
6885 | { | |
b3646477 | 6886 | return static_call(kvm_x86_get_segment_base)(vcpu, seg); |
bbd9b64e CO |
6887 | } |
6888 | ||
3cb16fe7 | 6889 | static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address) |
bbd9b64e | 6890 | { |
3cb16fe7 | 6891 | kvm_mmu_invlpg(emul_to_vcpu(ctxt), address); |
bbd9b64e CO |
6892 | } |
6893 | ||
ae6a2375 | 6894 | static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu) |
f5f48ee1 SY |
6895 | { |
6896 | if (!need_emulate_wbinvd(vcpu)) | |
6897 | return X86EMUL_CONTINUE; | |
6898 | ||
b3646477 | 6899 | if (static_call(kvm_x86_has_wbinvd_exit)()) { |
2eec7343 JK |
6900 | int cpu = get_cpu(); |
6901 | ||
6902 | cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask); | |
c2162e13 | 6903 | on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask, |
f5f48ee1 | 6904 | wbinvd_ipi, NULL, 1); |
2eec7343 | 6905 | put_cpu(); |
f5f48ee1 | 6906 | cpumask_clear(vcpu->arch.wbinvd_dirty_mask); |
2eec7343 JK |
6907 | } else |
6908 | wbinvd(); | |
f5f48ee1 SY |
6909 | return X86EMUL_CONTINUE; |
6910 | } | |
5cb56059 JS |
6911 | |
6912 | int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu) | |
6913 | { | |
6affcbed KH |
6914 | kvm_emulate_wbinvd_noskip(vcpu); |
6915 | return kvm_skip_emulated_instruction(vcpu); | |
5cb56059 | 6916 | } |
f5f48ee1 SY |
6917 | EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd); |
6918 | ||
5cb56059 JS |
6919 | |
6920 | ||
bcaf5cc5 AK |
6921 | static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt) |
6922 | { | |
5cb56059 | 6923 | kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt)); |
bcaf5cc5 AK |
6924 | } |
6925 | ||
29d6ca41 PB |
6926 | static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, |
6927 | unsigned long *dest) | |
bbd9b64e | 6928 | { |
29d6ca41 | 6929 | kvm_get_dr(emul_to_vcpu(ctxt), dr, dest); |
bbd9b64e CO |
6930 | } |
6931 | ||
52eb5a6d XL |
6932 | static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, |
6933 | unsigned long value) | |
bbd9b64e | 6934 | { |
338dbc97 | 6935 | |
996ff542 | 6936 | return kvm_set_dr(emul_to_vcpu(ctxt), dr, value); |
bbd9b64e CO |
6937 | } |
6938 | ||
52a46617 | 6939 | static u64 mk_cr_64(u64 curr_cr, u32 new_val) |
5fdbf976 | 6940 | { |
52a46617 | 6941 | return (curr_cr & ~((1ULL << 32) - 1)) | new_val; |
5fdbf976 MT |
6942 | } |
6943 | ||
717746e3 | 6944 | static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr) |
bbd9b64e | 6945 | { |
717746e3 | 6946 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
52a46617 GN |
6947 | unsigned long value; |
6948 | ||
6949 | switch (cr) { | |
6950 | case 0: | |
6951 | value = kvm_read_cr0(vcpu); | |
6952 | break; | |
6953 | case 2: | |
6954 | value = vcpu->arch.cr2; | |
6955 | break; | |
6956 | case 3: | |
9f8fe504 | 6957 | value = kvm_read_cr3(vcpu); |
52a46617 GN |
6958 | break; |
6959 | case 4: | |
6960 | value = kvm_read_cr4(vcpu); | |
6961 | break; | |
6962 | case 8: | |
6963 | value = kvm_get_cr8(vcpu); | |
6964 | break; | |
6965 | default: | |
a737f256 | 6966 | kvm_err("%s: unexpected cr %u\n", __func__, cr); |
52a46617 GN |
6967 | return 0; |
6968 | } | |
6969 | ||
6970 | return value; | |
6971 | } | |
6972 | ||
717746e3 | 6973 | static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val) |
52a46617 | 6974 | { |
717746e3 | 6975 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
0f12244f GN |
6976 | int res = 0; |
6977 | ||
52a46617 GN |
6978 | switch (cr) { |
6979 | case 0: | |
49a9b07e | 6980 | res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val)); |
52a46617 GN |
6981 | break; |
6982 | case 2: | |
6983 | vcpu->arch.cr2 = val; | |
6984 | break; | |
6985 | case 3: | |
2390218b | 6986 | res = kvm_set_cr3(vcpu, val); |
52a46617 GN |
6987 | break; |
6988 | case 4: | |
a83b29c6 | 6989 | res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val)); |
52a46617 GN |
6990 | break; |
6991 | case 8: | |
eea1cff9 | 6992 | res = kvm_set_cr8(vcpu, val); |
52a46617 GN |
6993 | break; |
6994 | default: | |
a737f256 | 6995 | kvm_err("%s: unexpected cr %u\n", __func__, cr); |
0f12244f | 6996 | res = -1; |
52a46617 | 6997 | } |
0f12244f GN |
6998 | |
6999 | return res; | |
52a46617 GN |
7000 | } |
7001 | ||
717746e3 | 7002 | static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt) |
9c537244 | 7003 | { |
b3646477 | 7004 | return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt)); |
9c537244 GN |
7005 | } |
7006 | ||
4bff1e86 | 7007 | static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) |
2dafc6c2 | 7008 | { |
b3646477 | 7009 | static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt); |
2dafc6c2 GN |
7010 | } |
7011 | ||
4bff1e86 | 7012 | static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) |
160ce1f1 | 7013 | { |
b3646477 | 7014 | static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt); |
160ce1f1 MG |
7015 | } |
7016 | ||
1ac9d0cf AK |
7017 | static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) |
7018 | { | |
b3646477 | 7019 | static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt); |
1ac9d0cf AK |
7020 | } |
7021 | ||
7022 | static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) | |
7023 | { | |
b3646477 | 7024 | static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt); |
1ac9d0cf AK |
7025 | } |
7026 | ||
4bff1e86 AK |
7027 | static unsigned long emulator_get_cached_segment_base( |
7028 | struct x86_emulate_ctxt *ctxt, int seg) | |
5951c442 | 7029 | { |
4bff1e86 | 7030 | return get_segment_base(emul_to_vcpu(ctxt), seg); |
5951c442 GN |
7031 | } |
7032 | ||
1aa36616 AK |
7033 | static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector, |
7034 | struct desc_struct *desc, u32 *base3, | |
7035 | int seg) | |
2dafc6c2 GN |
7036 | { |
7037 | struct kvm_segment var; | |
7038 | ||
4bff1e86 | 7039 | kvm_get_segment(emul_to_vcpu(ctxt), &var, seg); |
1aa36616 | 7040 | *selector = var.selector; |
2dafc6c2 | 7041 | |
378a8b09 GN |
7042 | if (var.unusable) { |
7043 | memset(desc, 0, sizeof(*desc)); | |
f0367ee1 RK |
7044 | if (base3) |
7045 | *base3 = 0; | |
2dafc6c2 | 7046 | return false; |
378a8b09 | 7047 | } |
2dafc6c2 GN |
7048 | |
7049 | if (var.g) | |
7050 | var.limit >>= 12; | |
7051 | set_desc_limit(desc, var.limit); | |
7052 | set_desc_base(desc, (unsigned long)var.base); | |
5601d05b GN |
7053 | #ifdef CONFIG_X86_64 |
7054 | if (base3) | |
7055 | *base3 = var.base >> 32; | |
7056 | #endif | |
2dafc6c2 GN |
7057 | desc->type = var.type; |
7058 | desc->s = var.s; | |
7059 | desc->dpl = var.dpl; | |
7060 | desc->p = var.present; | |
7061 | desc->avl = var.avl; | |
7062 | desc->l = var.l; | |
7063 | desc->d = var.db; | |
7064 | desc->g = var.g; | |
7065 | ||
7066 | return true; | |
7067 | } | |
7068 | ||
1aa36616 AK |
7069 | static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector, |
7070 | struct desc_struct *desc, u32 base3, | |
7071 | int seg) | |
2dafc6c2 | 7072 | { |
4bff1e86 | 7073 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
2dafc6c2 GN |
7074 | struct kvm_segment var; |
7075 | ||
1aa36616 | 7076 | var.selector = selector; |
2dafc6c2 | 7077 | var.base = get_desc_base(desc); |
5601d05b GN |
7078 | #ifdef CONFIG_X86_64 |
7079 | var.base |= ((u64)base3) << 32; | |
7080 | #endif | |
2dafc6c2 GN |
7081 | var.limit = get_desc_limit(desc); |
7082 | if (desc->g) | |
7083 | var.limit = (var.limit << 12) | 0xfff; | |
7084 | var.type = desc->type; | |
2dafc6c2 GN |
7085 | var.dpl = desc->dpl; |
7086 | var.db = desc->d; | |
7087 | var.s = desc->s; | |
7088 | var.l = desc->l; | |
7089 | var.g = desc->g; | |
7090 | var.avl = desc->avl; | |
7091 | var.present = desc->p; | |
7092 | var.unusable = !var.present; | |
7093 | var.padding = 0; | |
7094 | ||
7095 | kvm_set_segment(vcpu, &var, seg); | |
7096 | return; | |
7097 | } | |
7098 | ||
717746e3 AK |
7099 | static int emulator_get_msr(struct x86_emulate_ctxt *ctxt, |
7100 | u32 msr_index, u64 *pdata) | |
7101 | { | |
1ae09954 AG |
7102 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
7103 | int r; | |
7104 | ||
7105 | r = kvm_get_msr(vcpu, msr_index, pdata); | |
7106 | ||
7107 | if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) { | |
7108 | /* Bounce to user space */ | |
7109 | return X86EMUL_IO_NEEDED; | |
7110 | } | |
7111 | ||
7112 | return r; | |
717746e3 AK |
7113 | } |
7114 | ||
7115 | static int emulator_set_msr(struct x86_emulate_ctxt *ctxt, | |
7116 | u32 msr_index, u64 data) | |
7117 | { | |
1ae09954 AG |
7118 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
7119 | int r; | |
7120 | ||
7121 | r = kvm_set_msr(vcpu, msr_index, data); | |
7122 | ||
7123 | if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) { | |
7124 | /* Bounce to user space */ | |
7125 | return X86EMUL_IO_NEEDED; | |
7126 | } | |
7127 | ||
7128 | return r; | |
717746e3 AK |
7129 | } |
7130 | ||
64d60670 PB |
7131 | static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt) |
7132 | { | |
7133 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
7134 | ||
7135 | return vcpu->arch.smbase; | |
7136 | } | |
7137 | ||
7138 | static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase) | |
7139 | { | |
7140 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
7141 | ||
7142 | vcpu->arch.smbase = smbase; | |
7143 | } | |
7144 | ||
67f4d428 NA |
7145 | static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt, |
7146 | u32 pmc) | |
7147 | { | |
98ff80f5 | 7148 | return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc); |
67f4d428 NA |
7149 | } |
7150 | ||
222d21aa AK |
7151 | static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt, |
7152 | u32 pmc, u64 *pdata) | |
7153 | { | |
c6702c9d | 7154 | return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata); |
222d21aa AK |
7155 | } |
7156 | ||
6c3287f7 AK |
7157 | static void emulator_halt(struct x86_emulate_ctxt *ctxt) |
7158 | { | |
7159 | emul_to_vcpu(ctxt)->arch.halt_request = 1; | |
7160 | } | |
7161 | ||
2953538e | 7162 | static int emulator_intercept(struct x86_emulate_ctxt *ctxt, |
8a76d7f2 | 7163 | struct x86_instruction_info *info, |
c4f035c6 AK |
7164 | enum x86_intercept_stage stage) |
7165 | { | |
b3646477 | 7166 | return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage, |
21f1b8f2 | 7167 | &ctxt->exception); |
c4f035c6 AK |
7168 | } |
7169 | ||
e911eb3b | 7170 | static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt, |
f91af517 SC |
7171 | u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, |
7172 | bool exact_only) | |
bdb42f5a | 7173 | { |
f91af517 | 7174 | return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only); |
bdb42f5a SB |
7175 | } |
7176 | ||
5ae78e95 SC |
7177 | static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt) |
7178 | { | |
7179 | return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM); | |
7180 | } | |
7181 | ||
7182 | static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt) | |
7183 | { | |
7184 | return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE); | |
7185 | } | |
7186 | ||
7187 | static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt) | |
7188 | { | |
7189 | return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR); | |
7190 | } | |
7191 | ||
dd856efa AK |
7192 | static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg) |
7193 | { | |
27b4a9c4 | 7194 | return kvm_register_read_raw(emul_to_vcpu(ctxt), reg); |
dd856efa AK |
7195 | } |
7196 | ||
7197 | static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val) | |
7198 | { | |
27b4a9c4 | 7199 | kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val); |
dd856efa AK |
7200 | } |
7201 | ||
801806d9 NA |
7202 | static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked) |
7203 | { | |
b3646477 | 7204 | static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked); |
801806d9 NA |
7205 | } |
7206 | ||
6ed071f0 LP |
7207 | static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt) |
7208 | { | |
7209 | return emul_to_vcpu(ctxt)->arch.hflags; | |
7210 | } | |
7211 | ||
edce4654 | 7212 | static void emulator_exiting_smm(struct x86_emulate_ctxt *ctxt) |
6ed071f0 | 7213 | { |
78fcb2c9 SC |
7214 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
7215 | ||
dc87275f | 7216 | kvm_smm_changed(vcpu, false); |
6ed071f0 LP |
7217 | } |
7218 | ||
ed19321f SC |
7219 | static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, |
7220 | const char *smstate) | |
0234bf88 | 7221 | { |
b3646477 | 7222 | return static_call(kvm_x86_pre_leave_smm)(emul_to_vcpu(ctxt), smstate); |
0234bf88 LP |
7223 | } |
7224 | ||
c5833c7a SC |
7225 | static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt) |
7226 | { | |
fa75e08b | 7227 | kvm_mmu_reset_context(emul_to_vcpu(ctxt)); |
c5833c7a SC |
7228 | } |
7229 | ||
25b17226 SC |
7230 | static void emulator_triple_fault(struct x86_emulate_ctxt *ctxt) |
7231 | { | |
7232 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, emul_to_vcpu(ctxt)); | |
7233 | } | |
7234 | ||
02d4160f VK |
7235 | static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr) |
7236 | { | |
7237 | return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr); | |
7238 | } | |
7239 | ||
0225fb50 | 7240 | static const struct x86_emulate_ops emulate_ops = { |
dd856efa AK |
7241 | .read_gpr = emulator_read_gpr, |
7242 | .write_gpr = emulator_write_gpr, | |
ce14e868 PB |
7243 | .read_std = emulator_read_std, |
7244 | .write_std = emulator_write_std, | |
7a036a6f | 7245 | .read_phys = kvm_read_guest_phys_system, |
1871c602 | 7246 | .fetch = kvm_fetch_guest_virt, |
bbd9b64e CO |
7247 | .read_emulated = emulator_read_emulated, |
7248 | .write_emulated = emulator_write_emulated, | |
7249 | .cmpxchg_emulated = emulator_cmpxchg_emulated, | |
3cb16fe7 | 7250 | .invlpg = emulator_invlpg, |
cf8f70bf GN |
7251 | .pio_in_emulated = emulator_pio_in_emulated, |
7252 | .pio_out_emulated = emulator_pio_out_emulated, | |
1aa36616 AK |
7253 | .get_segment = emulator_get_segment, |
7254 | .set_segment = emulator_set_segment, | |
5951c442 | 7255 | .get_cached_segment_base = emulator_get_cached_segment_base, |
2dafc6c2 | 7256 | .get_gdt = emulator_get_gdt, |
160ce1f1 | 7257 | .get_idt = emulator_get_idt, |
1ac9d0cf AK |
7258 | .set_gdt = emulator_set_gdt, |
7259 | .set_idt = emulator_set_idt, | |
52a46617 GN |
7260 | .get_cr = emulator_get_cr, |
7261 | .set_cr = emulator_set_cr, | |
9c537244 | 7262 | .cpl = emulator_get_cpl, |
35aa5375 GN |
7263 | .get_dr = emulator_get_dr, |
7264 | .set_dr = emulator_set_dr, | |
64d60670 PB |
7265 | .get_smbase = emulator_get_smbase, |
7266 | .set_smbase = emulator_set_smbase, | |
717746e3 AK |
7267 | .set_msr = emulator_set_msr, |
7268 | .get_msr = emulator_get_msr, | |
67f4d428 | 7269 | .check_pmc = emulator_check_pmc, |
222d21aa | 7270 | .read_pmc = emulator_read_pmc, |
6c3287f7 | 7271 | .halt = emulator_halt, |
bcaf5cc5 | 7272 | .wbinvd = emulator_wbinvd, |
d6aa1000 | 7273 | .fix_hypercall = emulator_fix_hypercall, |
c4f035c6 | 7274 | .intercept = emulator_intercept, |
bdb42f5a | 7275 | .get_cpuid = emulator_get_cpuid, |
5ae78e95 SC |
7276 | .guest_has_long_mode = emulator_guest_has_long_mode, |
7277 | .guest_has_movbe = emulator_guest_has_movbe, | |
7278 | .guest_has_fxsr = emulator_guest_has_fxsr, | |
801806d9 | 7279 | .set_nmi_mask = emulator_set_nmi_mask, |
6ed071f0 | 7280 | .get_hflags = emulator_get_hflags, |
edce4654 | 7281 | .exiting_smm = emulator_exiting_smm, |
0234bf88 | 7282 | .pre_leave_smm = emulator_pre_leave_smm, |
c5833c7a | 7283 | .post_leave_smm = emulator_post_leave_smm, |
25b17226 | 7284 | .triple_fault = emulator_triple_fault, |
02d4160f | 7285 | .set_xcr = emulator_set_xcr, |
bbd9b64e CO |
7286 | }; |
7287 | ||
95cb2295 GN |
7288 | static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask) |
7289 | { | |
b3646477 | 7290 | u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu); |
95cb2295 GN |
7291 | /* |
7292 | * an sti; sti; sequence only disable interrupts for the first | |
7293 | * instruction. So, if the last instruction, be it emulated or | |
7294 | * not, left the system with the INT_STI flag enabled, it | |
7295 | * means that the last instruction is an sti. We should not | |
7296 | * leave the flag on in this case. The same goes for mov ss | |
7297 | */ | |
37ccdcbe PB |
7298 | if (int_shadow & mask) |
7299 | mask = 0; | |
6addfc42 | 7300 | if (unlikely(int_shadow || mask)) { |
b3646477 | 7301 | static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask); |
6addfc42 PB |
7302 | if (!mask) |
7303 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
7304 | } | |
95cb2295 GN |
7305 | } |
7306 | ||
ef54bcfe | 7307 | static bool inject_emulated_exception(struct kvm_vcpu *vcpu) |
54b8486f | 7308 | { |
c9b8b07c | 7309 | struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt; |
da9cb575 | 7310 | if (ctxt->exception.vector == PF_VECTOR) |
53b3d8e9 | 7311 | return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception); |
ef54bcfe PB |
7312 | |
7313 | if (ctxt->exception.error_code_valid) | |
da9cb575 AK |
7314 | kvm_queue_exception_e(vcpu, ctxt->exception.vector, |
7315 | ctxt->exception.error_code); | |
54b8486f | 7316 | else |
da9cb575 | 7317 | kvm_queue_exception(vcpu, ctxt->exception.vector); |
ef54bcfe | 7318 | return false; |
54b8486f GN |
7319 | } |
7320 | ||
c9b8b07c SC |
7321 | static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu) |
7322 | { | |
7323 | struct x86_emulate_ctxt *ctxt; | |
7324 | ||
7325 | ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT); | |
7326 | if (!ctxt) { | |
7327 | pr_err("kvm: failed to allocate vcpu's emulator\n"); | |
7328 | return NULL; | |
7329 | } | |
7330 | ||
7331 | ctxt->vcpu = vcpu; | |
7332 | ctxt->ops = &emulate_ops; | |
7333 | vcpu->arch.emulate_ctxt = ctxt; | |
7334 | ||
7335 | return ctxt; | |
7336 | } | |
7337 | ||
8ec4722d MG |
7338 | static void init_emulate_ctxt(struct kvm_vcpu *vcpu) |
7339 | { | |
c9b8b07c | 7340 | struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt; |
8ec4722d MG |
7341 | int cs_db, cs_l; |
7342 | ||
b3646477 | 7343 | static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l); |
8ec4722d | 7344 | |
744e699c | 7345 | ctxt->gpa_available = false; |
adf52235 | 7346 | ctxt->eflags = kvm_get_rflags(vcpu); |
c8401dda PB |
7347 | ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0; |
7348 | ||
adf52235 TY |
7349 | ctxt->eip = kvm_rip_read(vcpu); |
7350 | ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL : | |
7351 | (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 : | |
42bf549f | 7352 | (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 : |
adf52235 TY |
7353 | cs_db ? X86EMUL_MODE_PROT32 : |
7354 | X86EMUL_MODE_PROT16; | |
a584539b | 7355 | BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK); |
64d60670 PB |
7356 | BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK); |
7357 | BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK); | |
adf52235 | 7358 | |
da6393cd WL |
7359 | ctxt->interruptibility = 0; |
7360 | ctxt->have_exception = false; | |
7361 | ctxt->exception.vector = -1; | |
7362 | ctxt->perm_ok = false; | |
7363 | ||
dd856efa | 7364 | init_decode_cache(ctxt); |
7ae441ea | 7365 | vcpu->arch.emulate_regs_need_sync_from_vcpu = false; |
8ec4722d MG |
7366 | } |
7367 | ||
9497e1f2 | 7368 | void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip) |
63995653 | 7369 | { |
c9b8b07c | 7370 | struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt; |
63995653 MG |
7371 | int ret; |
7372 | ||
7373 | init_emulate_ctxt(vcpu); | |
7374 | ||
9dac77fa AK |
7375 | ctxt->op_bytes = 2; |
7376 | ctxt->ad_bytes = 2; | |
7377 | ctxt->_eip = ctxt->eip + inc_eip; | |
9d74191a | 7378 | ret = emulate_int_real(ctxt, irq); |
63995653 | 7379 | |
9497e1f2 SC |
7380 | if (ret != X86EMUL_CONTINUE) { |
7381 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); | |
7382 | } else { | |
7383 | ctxt->eip = ctxt->_eip; | |
7384 | kvm_rip_write(vcpu, ctxt->eip); | |
7385 | kvm_set_rflags(vcpu, ctxt->eflags); | |
7386 | } | |
63995653 MG |
7387 | } |
7388 | EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt); | |
7389 | ||
e2366171 | 7390 | static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type) |
6d77dbfc | 7391 | { |
6d77dbfc GN |
7392 | ++vcpu->stat.insn_emulation_fail; |
7393 | trace_kvm_emulate_insn_failed(vcpu); | |
e2366171 | 7394 | |
42cbf068 SC |
7395 | if (emulation_type & EMULTYPE_VMWARE_GP) { |
7396 | kvm_queue_exception_e(vcpu, GP_VECTOR, 0); | |
60fc3d02 | 7397 | return 1; |
42cbf068 | 7398 | } |
e2366171 | 7399 | |
738fece4 SC |
7400 | if (emulation_type & EMULTYPE_SKIP) { |
7401 | vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
7402 | vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
7403 | vcpu->run->internal.ndata = 0; | |
60fc3d02 | 7404 | return 0; |
738fece4 SC |
7405 | } |
7406 | ||
22da61c9 SC |
7407 | kvm_queue_exception(vcpu, UD_VECTOR); |
7408 | ||
b3646477 | 7409 | if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) { |
fc3a9157 JR |
7410 | vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; |
7411 | vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
7412 | vcpu->run->internal.ndata = 0; | |
60fc3d02 | 7413 | return 0; |
fc3a9157 | 7414 | } |
e2366171 | 7415 | |
60fc3d02 | 7416 | return 1; |
6d77dbfc GN |
7417 | } |
7418 | ||
736c291c | 7419 | static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa, |
991eebf9 GN |
7420 | bool write_fault_to_shadow_pgtable, |
7421 | int emulation_type) | |
a6f177ef | 7422 | { |
736c291c | 7423 | gpa_t gpa = cr2_or_gpa; |
ba049e93 | 7424 | kvm_pfn_t pfn; |
a6f177ef | 7425 | |
92daa48b | 7426 | if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF)) |
991eebf9 GN |
7427 | return false; |
7428 | ||
92daa48b SC |
7429 | if (WARN_ON_ONCE(is_guest_mode(vcpu)) || |
7430 | WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF))) | |
6c3dfeb6 SC |
7431 | return false; |
7432 | ||
44dd3ffa | 7433 | if (!vcpu->arch.mmu->direct_map) { |
95b3cf69 XG |
7434 | /* |
7435 | * Write permission should be allowed since only | |
7436 | * write access need to be emulated. | |
7437 | */ | |
736c291c | 7438 | gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL); |
a6f177ef | 7439 | |
95b3cf69 XG |
7440 | /* |
7441 | * If the mapping is invalid in guest, let cpu retry | |
7442 | * it to generate fault. | |
7443 | */ | |
7444 | if (gpa == UNMAPPED_GVA) | |
7445 | return true; | |
7446 | } | |
a6f177ef | 7447 | |
8e3d9d06 XG |
7448 | /* |
7449 | * Do not retry the unhandleable instruction if it faults on the | |
7450 | * readonly host memory, otherwise it will goto a infinite loop: | |
7451 | * retry instruction -> write #PF -> emulation fail -> retry | |
7452 | * instruction -> ... | |
7453 | */ | |
7454 | pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa)); | |
95b3cf69 XG |
7455 | |
7456 | /* | |
7457 | * If the instruction failed on the error pfn, it can not be fixed, | |
7458 | * report the error to userspace. | |
7459 | */ | |
7460 | if (is_error_noslot_pfn(pfn)) | |
7461 | return false; | |
7462 | ||
7463 | kvm_release_pfn_clean(pfn); | |
7464 | ||
7465 | /* The instructions are well-emulated on direct mmu. */ | |
44dd3ffa | 7466 | if (vcpu->arch.mmu->direct_map) { |
95b3cf69 XG |
7467 | unsigned int indirect_shadow_pages; |
7468 | ||
531810ca | 7469 | write_lock(&vcpu->kvm->mmu_lock); |
95b3cf69 | 7470 | indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages; |
531810ca | 7471 | write_unlock(&vcpu->kvm->mmu_lock); |
95b3cf69 XG |
7472 | |
7473 | if (indirect_shadow_pages) | |
7474 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa)); | |
7475 | ||
a6f177ef | 7476 | return true; |
8e3d9d06 | 7477 | } |
a6f177ef | 7478 | |
95b3cf69 XG |
7479 | /* |
7480 | * if emulation was due to access to shadowed page table | |
7481 | * and it failed try to unshadow page and re-enter the | |
7482 | * guest to let CPU execute the instruction. | |
7483 | */ | |
7484 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa)); | |
93c05d3e XG |
7485 | |
7486 | /* | |
7487 | * If the access faults on its page table, it can not | |
7488 | * be fixed by unprotecting shadow page and it should | |
7489 | * be reported to userspace. | |
7490 | */ | |
7491 | return !write_fault_to_shadow_pgtable; | |
a6f177ef GN |
7492 | } |
7493 | ||
1cb3f3ae | 7494 | static bool retry_instruction(struct x86_emulate_ctxt *ctxt, |
736c291c | 7495 | gpa_t cr2_or_gpa, int emulation_type) |
1cb3f3ae XG |
7496 | { |
7497 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
736c291c | 7498 | unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa; |
1cb3f3ae XG |
7499 | |
7500 | last_retry_eip = vcpu->arch.last_retry_eip; | |
7501 | last_retry_addr = vcpu->arch.last_retry_addr; | |
7502 | ||
7503 | /* | |
7504 | * If the emulation is caused by #PF and it is non-page_table | |
7505 | * writing instruction, it means the VM-EXIT is caused by shadow | |
7506 | * page protected, we can zap the shadow page and retry this | |
7507 | * instruction directly. | |
7508 | * | |
7509 | * Note: if the guest uses a non-page-table modifying instruction | |
7510 | * on the PDE that points to the instruction, then we will unmap | |
7511 | * the instruction and go to an infinite loop. So, we cache the | |
7512 | * last retried eip and the last fault address, if we meet the eip | |
7513 | * and the address again, we can break out of the potential infinite | |
7514 | * loop. | |
7515 | */ | |
7516 | vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0; | |
7517 | ||
92daa48b | 7518 | if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF)) |
1cb3f3ae XG |
7519 | return false; |
7520 | ||
92daa48b SC |
7521 | if (WARN_ON_ONCE(is_guest_mode(vcpu)) || |
7522 | WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF))) | |
6c3dfeb6 SC |
7523 | return false; |
7524 | ||
1cb3f3ae XG |
7525 | if (x86_page_table_writing_insn(ctxt)) |
7526 | return false; | |
7527 | ||
736c291c | 7528 | if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa) |
1cb3f3ae XG |
7529 | return false; |
7530 | ||
7531 | vcpu->arch.last_retry_eip = ctxt->eip; | |
736c291c | 7532 | vcpu->arch.last_retry_addr = cr2_or_gpa; |
1cb3f3ae | 7533 | |
44dd3ffa | 7534 | if (!vcpu->arch.mmu->direct_map) |
736c291c | 7535 | gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL); |
1cb3f3ae | 7536 | |
22368028 | 7537 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa)); |
1cb3f3ae XG |
7538 | |
7539 | return true; | |
7540 | } | |
7541 | ||
716d51ab GN |
7542 | static int complete_emulated_mmio(struct kvm_vcpu *vcpu); |
7543 | static int complete_emulated_pio(struct kvm_vcpu *vcpu); | |
7544 | ||
dc87275f | 7545 | static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm) |
a584539b | 7546 | { |
1270e647 | 7547 | trace_kvm_smm_transition(vcpu->vcpu_id, vcpu->arch.smbase, entering_smm); |
0d7ee6f4 | 7548 | |
dc87275f SC |
7549 | if (entering_smm) { |
7550 | vcpu->arch.hflags |= HF_SMM_MASK; | |
7551 | } else { | |
7552 | vcpu->arch.hflags &= ~(HF_SMM_MASK | HF_SMM_INSIDE_NMI_MASK); | |
7553 | ||
c43203ca PB |
7554 | /* Process a latched INIT or SMI, if any. */ |
7555 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
64d60670 | 7556 | } |
699023e2 PB |
7557 | |
7558 | kvm_mmu_reset_context(vcpu); | |
64d60670 PB |
7559 | } |
7560 | ||
4a1e10d5 PB |
7561 | static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7, |
7562 | unsigned long *db) | |
7563 | { | |
7564 | u32 dr6 = 0; | |
7565 | int i; | |
7566 | u32 enable, rwlen; | |
7567 | ||
7568 | enable = dr7; | |
7569 | rwlen = dr7 >> 16; | |
7570 | for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4) | |
7571 | if ((enable & 3) && (rwlen & 15) == type && db[i] == addr) | |
7572 | dr6 |= (1 << i); | |
7573 | return dr6; | |
7574 | } | |
7575 | ||
120c2c4f | 7576 | static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu) |
663f4c61 PB |
7577 | { |
7578 | struct kvm_run *kvm_run = vcpu->run; | |
7579 | ||
c8401dda | 7580 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) { |
9a3ecd5e | 7581 | kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW; |
d5d260c5 | 7582 | kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu); |
c8401dda PB |
7583 | kvm_run->debug.arch.exception = DB_VECTOR; |
7584 | kvm_run->exit_reason = KVM_EXIT_DEBUG; | |
60fc3d02 | 7585 | return 0; |
663f4c61 | 7586 | } |
120c2c4f | 7587 | kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS); |
60fc3d02 | 7588 | return 1; |
663f4c61 PB |
7589 | } |
7590 | ||
6affcbed KH |
7591 | int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu) |
7592 | { | |
b3646477 | 7593 | unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu); |
f8ea7c60 | 7594 | int r; |
6affcbed | 7595 | |
b3646477 | 7596 | r = static_call(kvm_x86_skip_emulated_instruction)(vcpu); |
60fc3d02 | 7597 | if (unlikely(!r)) |
f8ea7c60 | 7598 | return 0; |
c8401dda PB |
7599 | |
7600 | /* | |
7601 | * rflags is the old, "raw" value of the flags. The new value has | |
7602 | * not been saved yet. | |
7603 | * | |
7604 | * This is correct even for TF set by the guest, because "the | |
7605 | * processor will not generate this exception after the instruction | |
7606 | * that sets the TF flag". | |
7607 | */ | |
7608 | if (unlikely(rflags & X86_EFLAGS_TF)) | |
120c2c4f | 7609 | r = kvm_vcpu_do_singlestep(vcpu); |
60fc3d02 | 7610 | return r; |
6affcbed KH |
7611 | } |
7612 | EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction); | |
7613 | ||
4a1e10d5 PB |
7614 | static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r) |
7615 | { | |
4a1e10d5 PB |
7616 | if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) && |
7617 | (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) { | |
82b32774 NA |
7618 | struct kvm_run *kvm_run = vcpu->run; |
7619 | unsigned long eip = kvm_get_linear_rip(vcpu); | |
7620 | u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0, | |
4a1e10d5 PB |
7621 | vcpu->arch.guest_debug_dr7, |
7622 | vcpu->arch.eff_db); | |
7623 | ||
7624 | if (dr6 != 0) { | |
9a3ecd5e | 7625 | kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW; |
82b32774 | 7626 | kvm_run->debug.arch.pc = eip; |
4a1e10d5 PB |
7627 | kvm_run->debug.arch.exception = DB_VECTOR; |
7628 | kvm_run->exit_reason = KVM_EXIT_DEBUG; | |
60fc3d02 | 7629 | *r = 0; |
4a1e10d5 PB |
7630 | return true; |
7631 | } | |
7632 | } | |
7633 | ||
4161a569 NA |
7634 | if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) && |
7635 | !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) { | |
82b32774 NA |
7636 | unsigned long eip = kvm_get_linear_rip(vcpu); |
7637 | u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0, | |
4a1e10d5 PB |
7638 | vcpu->arch.dr7, |
7639 | vcpu->arch.db); | |
7640 | ||
7641 | if (dr6 != 0) { | |
4d5523cf | 7642 | kvm_queue_exception_p(vcpu, DB_VECTOR, dr6); |
60fc3d02 | 7643 | *r = 1; |
4a1e10d5 PB |
7644 | return true; |
7645 | } | |
7646 | } | |
7647 | ||
7648 | return false; | |
7649 | } | |
7650 | ||
04789b66 LA |
7651 | static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt) |
7652 | { | |
2d7921c4 AM |
7653 | switch (ctxt->opcode_len) { |
7654 | case 1: | |
7655 | switch (ctxt->b) { | |
7656 | case 0xe4: /* IN */ | |
7657 | case 0xe5: | |
7658 | case 0xec: | |
7659 | case 0xed: | |
7660 | case 0xe6: /* OUT */ | |
7661 | case 0xe7: | |
7662 | case 0xee: | |
7663 | case 0xef: | |
7664 | case 0x6c: /* INS */ | |
7665 | case 0x6d: | |
7666 | case 0x6e: /* OUTS */ | |
7667 | case 0x6f: | |
7668 | return true; | |
7669 | } | |
7670 | break; | |
7671 | case 2: | |
7672 | switch (ctxt->b) { | |
7673 | case 0x33: /* RDPMC */ | |
7674 | return true; | |
7675 | } | |
7676 | break; | |
04789b66 LA |
7677 | } |
7678 | ||
7679 | return false; | |
7680 | } | |
7681 | ||
4aa2691d WH |
7682 | /* |
7683 | * Decode to be emulated instruction. Return EMULATION_OK if success. | |
7684 | */ | |
7685 | int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type, | |
7686 | void *insn, int insn_len) | |
7687 | { | |
7688 | int r = EMULATION_OK; | |
7689 | struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt; | |
7690 | ||
7691 | init_emulate_ctxt(vcpu); | |
7692 | ||
7693 | /* | |
7694 | * We will reenter on the same instruction since we do not set | |
7695 | * complete_userspace_io. This does not handle watchpoints yet, | |
7696 | * those would be handled in the emulate_ops. | |
7697 | */ | |
7698 | if (!(emulation_type & EMULTYPE_SKIP) && | |
7699 | kvm_vcpu_check_breakpoint(vcpu, &r)) | |
7700 | return r; | |
7701 | ||
b35491e6 | 7702 | r = x86_decode_insn(ctxt, insn, insn_len, emulation_type); |
4aa2691d WH |
7703 | |
7704 | trace_kvm_emulate_insn_start(vcpu); | |
7705 | ++vcpu->stat.insn_emulation; | |
7706 | ||
7707 | return r; | |
7708 | } | |
7709 | EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction); | |
7710 | ||
736c291c SC |
7711 | int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa, |
7712 | int emulation_type, void *insn, int insn_len) | |
bbd9b64e | 7713 | { |
95cb2295 | 7714 | int r; |
c9b8b07c | 7715 | struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt; |
7ae441ea | 7716 | bool writeback = true; |
09e3e2a1 SC |
7717 | bool write_fault_to_spt; |
7718 | ||
b3646477 | 7719 | if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len))) |
09e3e2a1 | 7720 | return 1; |
bbd9b64e | 7721 | |
c595ceee PB |
7722 | vcpu->arch.l1tf_flush_l1d = true; |
7723 | ||
93c05d3e XG |
7724 | /* |
7725 | * Clear write_fault_to_shadow_pgtable here to ensure it is | |
7726 | * never reused. | |
7727 | */ | |
09e3e2a1 | 7728 | write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable; |
93c05d3e | 7729 | vcpu->arch.write_fault_to_shadow_pgtable = false; |
8d7d8102 | 7730 | |
571008da | 7731 | if (!(emulation_type & EMULTYPE_NO_DECODE)) { |
4aa2691d | 7732 | kvm_clear_exception_queue(vcpu); |
4a1e10d5 | 7733 | |
4aa2691d WH |
7734 | r = x86_decode_emulated_instruction(vcpu, emulation_type, |
7735 | insn, insn_len); | |
1d2887e2 | 7736 | if (r != EMULATION_OK) { |
b4000606 | 7737 | if ((emulation_type & EMULTYPE_TRAP_UD) || |
c83fad65 SC |
7738 | (emulation_type & EMULTYPE_TRAP_UD_FORCED)) { |
7739 | kvm_queue_exception(vcpu, UD_VECTOR); | |
60fc3d02 | 7740 | return 1; |
c83fad65 | 7741 | } |
736c291c SC |
7742 | if (reexecute_instruction(vcpu, cr2_or_gpa, |
7743 | write_fault_to_spt, | |
7744 | emulation_type)) | |
60fc3d02 | 7745 | return 1; |
8530a79c | 7746 | if (ctxt->have_exception) { |
c8848cee JD |
7747 | /* |
7748 | * #UD should result in just EMULATION_FAILED, and trap-like | |
7749 | * exception should not be encountered during decode. | |
7750 | */ | |
7751 | WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR || | |
7752 | exception_type(ctxt->exception.vector) == EXCPT_TRAP); | |
8530a79c | 7753 | inject_emulated_exception(vcpu); |
60fc3d02 | 7754 | return 1; |
8530a79c | 7755 | } |
e2366171 | 7756 | return handle_emulation_failure(vcpu, emulation_type); |
bbd9b64e CO |
7757 | } |
7758 | } | |
7759 | ||
42cbf068 SC |
7760 | if ((emulation_type & EMULTYPE_VMWARE_GP) && |
7761 | !is_vmware_backdoor_opcode(ctxt)) { | |
7762 | kvm_queue_exception_e(vcpu, GP_VECTOR, 0); | |
60fc3d02 | 7763 | return 1; |
42cbf068 | 7764 | } |
04789b66 | 7765 | |
1957aa63 SC |
7766 | /* |
7767 | * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks | |
7768 | * for kvm_skip_emulated_instruction(). The caller is responsible for | |
7769 | * updating interruptibility state and injecting single-step #DBs. | |
7770 | */ | |
ba8afb6b | 7771 | if (emulation_type & EMULTYPE_SKIP) { |
9dac77fa | 7772 | kvm_rip_write(vcpu, ctxt->_eip); |
bb663c7a NA |
7773 | if (ctxt->eflags & X86_EFLAGS_RF) |
7774 | kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF); | |
60fc3d02 | 7775 | return 1; |
ba8afb6b GN |
7776 | } |
7777 | ||
736c291c | 7778 | if (retry_instruction(ctxt, cr2_or_gpa, emulation_type)) |
60fc3d02 | 7779 | return 1; |
1cb3f3ae | 7780 | |
7ae441ea | 7781 | /* this is needed for vmware backdoor interface to work since it |
4d2179e1 | 7782 | changes registers values during IO operation */ |
7ae441ea GN |
7783 | if (vcpu->arch.emulate_regs_need_sync_from_vcpu) { |
7784 | vcpu->arch.emulate_regs_need_sync_from_vcpu = false; | |
dd856efa | 7785 | emulator_invalidate_register_cache(ctxt); |
7ae441ea | 7786 | } |
4d2179e1 | 7787 | |
5cd21917 | 7788 | restart: |
92daa48b SC |
7789 | if (emulation_type & EMULTYPE_PF) { |
7790 | /* Save the faulting GPA (cr2) in the address field */ | |
7791 | ctxt->exception.address = cr2_or_gpa; | |
7792 | ||
7793 | /* With shadow page tables, cr2 contains a GVA or nGPA. */ | |
7794 | if (vcpu->arch.mmu->direct_map) { | |
744e699c SC |
7795 | ctxt->gpa_available = true; |
7796 | ctxt->gpa_val = cr2_or_gpa; | |
92daa48b SC |
7797 | } |
7798 | } else { | |
7799 | /* Sanitize the address out of an abundance of paranoia. */ | |
7800 | ctxt->exception.address = 0; | |
7801 | } | |
0f89b207 | 7802 | |
9d74191a | 7803 | r = x86_emulate_insn(ctxt); |
bbd9b64e | 7804 | |
775fde86 | 7805 | if (r == EMULATION_INTERCEPTED) |
60fc3d02 | 7806 | return 1; |
775fde86 | 7807 | |
d2ddd1c4 | 7808 | if (r == EMULATION_FAILED) { |
736c291c | 7809 | if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt, |
991eebf9 | 7810 | emulation_type)) |
60fc3d02 | 7811 | return 1; |
c3cd7ffa | 7812 | |
e2366171 | 7813 | return handle_emulation_failure(vcpu, emulation_type); |
bbd9b64e CO |
7814 | } |
7815 | ||
9d74191a | 7816 | if (ctxt->have_exception) { |
60fc3d02 | 7817 | r = 1; |
ef54bcfe PB |
7818 | if (inject_emulated_exception(vcpu)) |
7819 | return r; | |
d2ddd1c4 | 7820 | } else if (vcpu->arch.pio.count) { |
0912c977 PB |
7821 | if (!vcpu->arch.pio.in) { |
7822 | /* FIXME: return into emulator if single-stepping. */ | |
3457e419 | 7823 | vcpu->arch.pio.count = 0; |
0912c977 | 7824 | } else { |
7ae441ea | 7825 | writeback = false; |
716d51ab GN |
7826 | vcpu->arch.complete_userspace_io = complete_emulated_pio; |
7827 | } | |
60fc3d02 | 7828 | r = 0; |
7ae441ea | 7829 | } else if (vcpu->mmio_needed) { |
bc8a0aaf SC |
7830 | ++vcpu->stat.mmio_exits; |
7831 | ||
7ae441ea GN |
7832 | if (!vcpu->mmio_is_write) |
7833 | writeback = false; | |
60fc3d02 | 7834 | r = 0; |
716d51ab | 7835 | vcpu->arch.complete_userspace_io = complete_emulated_mmio; |
7ae441ea | 7836 | } else if (r == EMULATION_RESTART) |
5cd21917 | 7837 | goto restart; |
d2ddd1c4 | 7838 | else |
60fc3d02 | 7839 | r = 1; |
f850e2e6 | 7840 | |
7ae441ea | 7841 | if (writeback) { |
b3646477 | 7842 | unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu); |
9d74191a | 7843 | toggle_interruptibility(vcpu, ctxt->interruptibility); |
7ae441ea | 7844 | vcpu->arch.emulate_regs_need_sync_to_vcpu = false; |
38827dbd | 7845 | if (!ctxt->have_exception || |
75ee23b3 SC |
7846 | exception_type(ctxt->exception.vector) == EXCPT_TRAP) { |
7847 | kvm_rip_write(vcpu, ctxt->eip); | |
384dea1c | 7848 | if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP))) |
120c2c4f | 7849 | r = kvm_vcpu_do_singlestep(vcpu); |
afaf0b2f | 7850 | if (kvm_x86_ops.update_emulated_instruction) |
b3646477 | 7851 | static_call(kvm_x86_update_emulated_instruction)(vcpu); |
38827dbd | 7852 | __kvm_set_rflags(vcpu, ctxt->eflags); |
75ee23b3 | 7853 | } |
6addfc42 PB |
7854 | |
7855 | /* | |
7856 | * For STI, interrupts are shadowed; so KVM_REQ_EVENT will | |
7857 | * do nothing, and it will be requested again as soon as | |
7858 | * the shadow expires. But we still need to check here, | |
7859 | * because POPF has no interrupt shadow. | |
7860 | */ | |
7861 | if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF)) | |
7862 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
7ae441ea GN |
7863 | } else |
7864 | vcpu->arch.emulate_regs_need_sync_to_vcpu = true; | |
e85d28f8 GN |
7865 | |
7866 | return r; | |
de7d789a | 7867 | } |
c60658d1 SC |
7868 | |
7869 | int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type) | |
7870 | { | |
7871 | return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0); | |
7872 | } | |
7873 | EXPORT_SYMBOL_GPL(kvm_emulate_instruction); | |
7874 | ||
7875 | int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu, | |
7876 | void *insn, int insn_len) | |
7877 | { | |
7878 | return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len); | |
7879 | } | |
7880 | EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer); | |
de7d789a | 7881 | |
8764ed55 SC |
7882 | static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu) |
7883 | { | |
7884 | vcpu->arch.pio.count = 0; | |
7885 | return 1; | |
7886 | } | |
7887 | ||
45def77e SC |
7888 | static int complete_fast_pio_out(struct kvm_vcpu *vcpu) |
7889 | { | |
7890 | vcpu->arch.pio.count = 0; | |
7891 | ||
7892 | if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) | |
7893 | return 1; | |
7894 | ||
7895 | return kvm_skip_emulated_instruction(vcpu); | |
7896 | } | |
7897 | ||
dca7f128 SC |
7898 | static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, |
7899 | unsigned short port) | |
de7d789a | 7900 | { |
de3cd117 | 7901 | unsigned long val = kvm_rax_read(vcpu); |
2e3bb4d8 SC |
7902 | int ret = emulator_pio_out(vcpu, size, port, &val, 1); |
7903 | ||
8764ed55 SC |
7904 | if (ret) |
7905 | return ret; | |
45def77e | 7906 | |
8764ed55 SC |
7907 | /* |
7908 | * Workaround userspace that relies on old KVM behavior of %rip being | |
7909 | * incremented prior to exiting to userspace to handle "OUT 0x7e". | |
7910 | */ | |
7911 | if (port == 0x7e && | |
7912 | kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) { | |
7913 | vcpu->arch.complete_userspace_io = | |
7914 | complete_fast_pio_out_port_0x7e; | |
7915 | kvm_skip_emulated_instruction(vcpu); | |
7916 | } else { | |
45def77e SC |
7917 | vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu); |
7918 | vcpu->arch.complete_userspace_io = complete_fast_pio_out; | |
7919 | } | |
8764ed55 | 7920 | return 0; |
de7d789a | 7921 | } |
de7d789a | 7922 | |
8370c3d0 TL |
7923 | static int complete_fast_pio_in(struct kvm_vcpu *vcpu) |
7924 | { | |
7925 | unsigned long val; | |
7926 | ||
7927 | /* We should only ever be called with arch.pio.count equal to 1 */ | |
7928 | BUG_ON(vcpu->arch.pio.count != 1); | |
7929 | ||
45def77e SC |
7930 | if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) { |
7931 | vcpu->arch.pio.count = 0; | |
7932 | return 1; | |
7933 | } | |
7934 | ||
8370c3d0 | 7935 | /* For size less than 4 we merge, else we zero extend */ |
de3cd117 | 7936 | val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0; |
8370c3d0 TL |
7937 | |
7938 | /* | |
2e3bb4d8 | 7939 | * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform |
8370c3d0 TL |
7940 | * the copy and tracing |
7941 | */ | |
2e3bb4d8 | 7942 | emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1); |
de3cd117 | 7943 | kvm_rax_write(vcpu, val); |
8370c3d0 | 7944 | |
45def77e | 7945 | return kvm_skip_emulated_instruction(vcpu); |
8370c3d0 TL |
7946 | } |
7947 | ||
dca7f128 SC |
7948 | static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size, |
7949 | unsigned short port) | |
8370c3d0 TL |
7950 | { |
7951 | unsigned long val; | |
7952 | int ret; | |
7953 | ||
7954 | /* For size less than 4 we merge, else we zero extend */ | |
de3cd117 | 7955 | val = (size < 4) ? kvm_rax_read(vcpu) : 0; |
8370c3d0 | 7956 | |
2e3bb4d8 | 7957 | ret = emulator_pio_in(vcpu, size, port, &val, 1); |
8370c3d0 | 7958 | if (ret) { |
de3cd117 | 7959 | kvm_rax_write(vcpu, val); |
8370c3d0 TL |
7960 | return ret; |
7961 | } | |
7962 | ||
45def77e | 7963 | vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu); |
8370c3d0 TL |
7964 | vcpu->arch.complete_userspace_io = complete_fast_pio_in; |
7965 | ||
7966 | return 0; | |
7967 | } | |
dca7f128 SC |
7968 | |
7969 | int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in) | |
7970 | { | |
45def77e | 7971 | int ret; |
dca7f128 | 7972 | |
dca7f128 | 7973 | if (in) |
45def77e | 7974 | ret = kvm_fast_pio_in(vcpu, size, port); |
dca7f128 | 7975 | else |
45def77e SC |
7976 | ret = kvm_fast_pio_out(vcpu, size, port); |
7977 | return ret && kvm_skip_emulated_instruction(vcpu); | |
dca7f128 SC |
7978 | } |
7979 | EXPORT_SYMBOL_GPL(kvm_fast_pio); | |
8370c3d0 | 7980 | |
251a5fd6 | 7981 | static int kvmclock_cpu_down_prep(unsigned int cpu) |
8cfdc000 | 7982 | { |
0a3aee0d | 7983 | __this_cpu_write(cpu_tsc_khz, 0); |
251a5fd6 | 7984 | return 0; |
8cfdc000 ZA |
7985 | } |
7986 | ||
7987 | static void tsc_khz_changed(void *data) | |
c8076604 | 7988 | { |
8cfdc000 ZA |
7989 | struct cpufreq_freqs *freq = data; |
7990 | unsigned long khz = 0; | |
7991 | ||
7992 | if (data) | |
7993 | khz = freq->new; | |
7994 | else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) | |
7995 | khz = cpufreq_quick_get(raw_smp_processor_id()); | |
7996 | if (!khz) | |
7997 | khz = tsc_khz; | |
0a3aee0d | 7998 | __this_cpu_write(cpu_tsc_khz, khz); |
c8076604 GH |
7999 | } |
8000 | ||
5fa4ec9c | 8001 | #ifdef CONFIG_X86_64 |
0092e434 VK |
8002 | static void kvm_hyperv_tsc_notifier(void) |
8003 | { | |
0092e434 VK |
8004 | struct kvm *kvm; |
8005 | struct kvm_vcpu *vcpu; | |
8006 | int cpu; | |
a83829f5 | 8007 | unsigned long flags; |
0092e434 | 8008 | |
0d9ce162 | 8009 | mutex_lock(&kvm_lock); |
0092e434 VK |
8010 | list_for_each_entry(kvm, &vm_list, vm_list) |
8011 | kvm_make_mclock_inprogress_request(kvm); | |
8012 | ||
8013 | hyperv_stop_tsc_emulation(); | |
8014 | ||
8015 | /* TSC frequency always matches when on Hyper-V */ | |
8016 | for_each_present_cpu(cpu) | |
8017 | per_cpu(cpu_tsc_khz, cpu) = tsc_khz; | |
8018 | kvm_max_guest_tsc_khz = tsc_khz; | |
8019 | ||
8020 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
8021 | struct kvm_arch *ka = &kvm->arch; | |
8022 | ||
a83829f5 | 8023 | spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags); |
0092e434 | 8024 | pvclock_update_vm_gtod_copy(kvm); |
a83829f5 | 8025 | spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags); |
0092e434 VK |
8026 | |
8027 | kvm_for_each_vcpu(cpu, vcpu, kvm) | |
8028 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); | |
8029 | ||
8030 | kvm_for_each_vcpu(cpu, vcpu, kvm) | |
8031 | kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu); | |
0092e434 | 8032 | } |
0d9ce162 | 8033 | mutex_unlock(&kvm_lock); |
0092e434 | 8034 | } |
5fa4ec9c | 8035 | #endif |
0092e434 | 8036 | |
df24014a | 8037 | static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu) |
c8076604 | 8038 | { |
c8076604 GH |
8039 | struct kvm *kvm; |
8040 | struct kvm_vcpu *vcpu; | |
8041 | int i, send_ipi = 0; | |
8042 | ||
8cfdc000 ZA |
8043 | /* |
8044 | * We allow guests to temporarily run on slowing clocks, | |
8045 | * provided we notify them after, or to run on accelerating | |
8046 | * clocks, provided we notify them before. Thus time never | |
8047 | * goes backwards. | |
8048 | * | |
8049 | * However, we have a problem. We can't atomically update | |
8050 | * the frequency of a given CPU from this function; it is | |
8051 | * merely a notifier, which can be called from any CPU. | |
8052 | * Changing the TSC frequency at arbitrary points in time | |
8053 | * requires a recomputation of local variables related to | |
8054 | * the TSC for each VCPU. We must flag these local variables | |
8055 | * to be updated and be sure the update takes place with the | |
8056 | * new frequency before any guests proceed. | |
8057 | * | |
8058 | * Unfortunately, the combination of hotplug CPU and frequency | |
8059 | * change creates an intractable locking scenario; the order | |
8060 | * of when these callouts happen is undefined with respect to | |
8061 | * CPU hotplug, and they can race with each other. As such, | |
8062 | * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is | |
8063 | * undefined; you can actually have a CPU frequency change take | |
8064 | * place in between the computation of X and the setting of the | |
8065 | * variable. To protect against this problem, all updates of | |
8066 | * the per_cpu tsc_khz variable are done in an interrupt | |
8067 | * protected IPI, and all callers wishing to update the value | |
8068 | * must wait for a synchronous IPI to complete (which is trivial | |
8069 | * if the caller is on the CPU already). This establishes the | |
8070 | * necessary total order on variable updates. | |
8071 | * | |
8072 | * Note that because a guest time update may take place | |
8073 | * anytime after the setting of the VCPU's request bit, the | |
8074 | * correct TSC value must be set before the request. However, | |
8075 | * to ensure the update actually makes it to any guest which | |
8076 | * starts running in hardware virtualization between the set | |
8077 | * and the acquisition of the spinlock, we must also ping the | |
8078 | * CPU after setting the request bit. | |
8079 | * | |
8080 | */ | |
8081 | ||
df24014a | 8082 | smp_call_function_single(cpu, tsc_khz_changed, freq, 1); |
c8076604 | 8083 | |
0d9ce162 | 8084 | mutex_lock(&kvm_lock); |
c8076604 | 8085 | list_for_each_entry(kvm, &vm_list, vm_list) { |
988a2cae | 8086 | kvm_for_each_vcpu(i, vcpu, kvm) { |
df24014a | 8087 | if (vcpu->cpu != cpu) |
c8076604 | 8088 | continue; |
c285545f | 8089 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
0d9ce162 | 8090 | if (vcpu->cpu != raw_smp_processor_id()) |
8cfdc000 | 8091 | send_ipi = 1; |
c8076604 GH |
8092 | } |
8093 | } | |
0d9ce162 | 8094 | mutex_unlock(&kvm_lock); |
c8076604 GH |
8095 | |
8096 | if (freq->old < freq->new && send_ipi) { | |
8097 | /* | |
8098 | * We upscale the frequency. Must make the guest | |
8099 | * doesn't see old kvmclock values while running with | |
8100 | * the new frequency, otherwise we risk the guest sees | |
8101 | * time go backwards. | |
8102 | * | |
8103 | * In case we update the frequency for another cpu | |
8104 | * (which might be in guest context) send an interrupt | |
8105 | * to kick the cpu out of guest context. Next time | |
8106 | * guest context is entered kvmclock will be updated, | |
8107 | * so the guest will not see stale values. | |
8108 | */ | |
df24014a | 8109 | smp_call_function_single(cpu, tsc_khz_changed, freq, 1); |
c8076604 | 8110 | } |
df24014a VK |
8111 | } |
8112 | ||
8113 | static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val, | |
8114 | void *data) | |
8115 | { | |
8116 | struct cpufreq_freqs *freq = data; | |
8117 | int cpu; | |
8118 | ||
8119 | if (val == CPUFREQ_PRECHANGE && freq->old > freq->new) | |
8120 | return 0; | |
8121 | if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new) | |
8122 | return 0; | |
8123 | ||
8124 | for_each_cpu(cpu, freq->policy->cpus) | |
8125 | __kvmclock_cpufreq_notifier(freq, cpu); | |
8126 | ||
c8076604 GH |
8127 | return 0; |
8128 | } | |
8129 | ||
8130 | static struct notifier_block kvmclock_cpufreq_notifier_block = { | |
8cfdc000 ZA |
8131 | .notifier_call = kvmclock_cpufreq_notifier |
8132 | }; | |
8133 | ||
251a5fd6 | 8134 | static int kvmclock_cpu_online(unsigned int cpu) |
8cfdc000 | 8135 | { |
251a5fd6 SAS |
8136 | tsc_khz_changed(NULL); |
8137 | return 0; | |
8cfdc000 ZA |
8138 | } |
8139 | ||
b820cc0c ZA |
8140 | static void kvm_timer_init(void) |
8141 | { | |
c285545f | 8142 | max_tsc_khz = tsc_khz; |
460dd42e | 8143 | |
b820cc0c | 8144 | if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) { |
c285545f | 8145 | #ifdef CONFIG_CPU_FREQ |
aaec7c03 | 8146 | struct cpufreq_policy *policy; |
758f588d BP |
8147 | int cpu; |
8148 | ||
3e26f230 | 8149 | cpu = get_cpu(); |
aaec7c03 | 8150 | policy = cpufreq_cpu_get(cpu); |
9a11997e WL |
8151 | if (policy) { |
8152 | if (policy->cpuinfo.max_freq) | |
8153 | max_tsc_khz = policy->cpuinfo.max_freq; | |
8154 | cpufreq_cpu_put(policy); | |
8155 | } | |
3e26f230 | 8156 | put_cpu(); |
c285545f | 8157 | #endif |
b820cc0c ZA |
8158 | cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block, |
8159 | CPUFREQ_TRANSITION_NOTIFIER); | |
8160 | } | |
460dd42e | 8161 | |
73c1b41e | 8162 | cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online", |
251a5fd6 | 8163 | kvmclock_cpu_online, kvmclock_cpu_down_prep); |
b820cc0c ZA |
8164 | } |
8165 | ||
dd60d217 AK |
8166 | DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu); |
8167 | EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu); | |
ff9d07a0 | 8168 | |
f5132b01 | 8169 | int kvm_is_in_guest(void) |
ff9d07a0 | 8170 | { |
086c9855 | 8171 | return __this_cpu_read(current_vcpu) != NULL; |
ff9d07a0 ZY |
8172 | } |
8173 | ||
8174 | static int kvm_is_user_mode(void) | |
8175 | { | |
8176 | int user_mode = 3; | |
dcf46b94 | 8177 | |
086c9855 | 8178 | if (__this_cpu_read(current_vcpu)) |
b3646477 | 8179 | user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu)); |
dcf46b94 | 8180 | |
ff9d07a0 ZY |
8181 | return user_mode != 0; |
8182 | } | |
8183 | ||
8184 | static unsigned long kvm_get_guest_ip(void) | |
8185 | { | |
8186 | unsigned long ip = 0; | |
dcf46b94 | 8187 | |
086c9855 AS |
8188 | if (__this_cpu_read(current_vcpu)) |
8189 | ip = kvm_rip_read(__this_cpu_read(current_vcpu)); | |
dcf46b94 | 8190 | |
ff9d07a0 ZY |
8191 | return ip; |
8192 | } | |
8193 | ||
8479e04e LK |
8194 | static void kvm_handle_intel_pt_intr(void) |
8195 | { | |
8196 | struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu); | |
8197 | ||
8198 | kvm_make_request(KVM_REQ_PMI, vcpu); | |
8199 | __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT, | |
8200 | (unsigned long *)&vcpu->arch.pmu.global_status); | |
8201 | } | |
8202 | ||
ff9d07a0 ZY |
8203 | static struct perf_guest_info_callbacks kvm_guest_cbs = { |
8204 | .is_in_guest = kvm_is_in_guest, | |
8205 | .is_user_mode = kvm_is_user_mode, | |
8206 | .get_guest_ip = kvm_get_guest_ip, | |
8479e04e | 8207 | .handle_intel_pt_intr = kvm_handle_intel_pt_intr, |
ff9d07a0 ZY |
8208 | }; |
8209 | ||
16e8d74d MT |
8210 | #ifdef CONFIG_X86_64 |
8211 | static void pvclock_gtod_update_fn(struct work_struct *work) | |
8212 | { | |
d828199e MT |
8213 | struct kvm *kvm; |
8214 | ||
8215 | struct kvm_vcpu *vcpu; | |
8216 | int i; | |
8217 | ||
0d9ce162 | 8218 | mutex_lock(&kvm_lock); |
d828199e MT |
8219 | list_for_each_entry(kvm, &vm_list, vm_list) |
8220 | kvm_for_each_vcpu(i, vcpu, kvm) | |
105b21bb | 8221 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); |
d828199e | 8222 | atomic_set(&kvm_guest_has_master_clock, 0); |
0d9ce162 | 8223 | mutex_unlock(&kvm_lock); |
16e8d74d MT |
8224 | } |
8225 | ||
8226 | static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn); | |
8227 | ||
3f804f6d TG |
8228 | /* |
8229 | * Indirection to move queue_work() out of the tk_core.seq write held | |
8230 | * region to prevent possible deadlocks against time accessors which | |
8231 | * are invoked with work related locks held. | |
8232 | */ | |
8233 | static void pvclock_irq_work_fn(struct irq_work *w) | |
8234 | { | |
8235 | queue_work(system_long_wq, &pvclock_gtod_work); | |
8236 | } | |
8237 | ||
8238 | static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn); | |
8239 | ||
16e8d74d MT |
8240 | /* |
8241 | * Notification about pvclock gtod data update. | |
8242 | */ | |
8243 | static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused, | |
8244 | void *priv) | |
8245 | { | |
8246 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; | |
8247 | struct timekeeper *tk = priv; | |
8248 | ||
8249 | update_pvclock_gtod(tk); | |
8250 | ||
3f804f6d TG |
8251 | /* |
8252 | * Disable master clock if host does not trust, or does not use, | |
8253 | * TSC based clocksource. Delegate queue_work() to irq_work as | |
8254 | * this is invoked with tk_core.seq write held. | |
16e8d74d | 8255 | */ |
b0c39dc6 | 8256 | if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) && |
16e8d74d | 8257 | atomic_read(&kvm_guest_has_master_clock) != 0) |
3f804f6d | 8258 | irq_work_queue(&pvclock_irq_work); |
16e8d74d MT |
8259 | return 0; |
8260 | } | |
8261 | ||
8262 | static struct notifier_block pvclock_gtod_notifier = { | |
8263 | .notifier_call = pvclock_gtod_notify, | |
8264 | }; | |
8265 | #endif | |
8266 | ||
f8c16bba | 8267 | int kvm_arch_init(void *opaque) |
043405e1 | 8268 | { |
d008dfdb | 8269 | struct kvm_x86_init_ops *ops = opaque; |
b820cc0c | 8270 | int r; |
f8c16bba | 8271 | |
afaf0b2f | 8272 | if (kvm_x86_ops.hardware_enable) { |
f8c16bba | 8273 | printk(KERN_ERR "kvm: already loaded the other module\n"); |
56c6d28a ZX |
8274 | r = -EEXIST; |
8275 | goto out; | |
f8c16bba ZX |
8276 | } |
8277 | ||
8278 | if (!ops->cpu_has_kvm_support()) { | |
ef935c25 | 8279 | pr_err_ratelimited("kvm: no hardware support\n"); |
56c6d28a ZX |
8280 | r = -EOPNOTSUPP; |
8281 | goto out; | |
f8c16bba ZX |
8282 | } |
8283 | if (ops->disabled_by_bios()) { | |
ef935c25 | 8284 | pr_err_ratelimited("kvm: disabled by bios\n"); |
56c6d28a ZX |
8285 | r = -EOPNOTSUPP; |
8286 | goto out; | |
f8c16bba ZX |
8287 | } |
8288 | ||
b666a4b6 MO |
8289 | /* |
8290 | * KVM explicitly assumes that the guest has an FPU and | |
8291 | * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the | |
8292 | * vCPU's FPU state as a fxregs_state struct. | |
8293 | */ | |
8294 | if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) { | |
8295 | printk(KERN_ERR "kvm: inadequate fpu\n"); | |
8296 | r = -EOPNOTSUPP; | |
8297 | goto out; | |
8298 | } | |
8299 | ||
013f6a5d | 8300 | r = -ENOMEM; |
ed8e4812 | 8301 | x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu), |
b666a4b6 MO |
8302 | __alignof__(struct fpu), SLAB_ACCOUNT, |
8303 | NULL); | |
8304 | if (!x86_fpu_cache) { | |
8305 | printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n"); | |
8306 | goto out; | |
8307 | } | |
8308 | ||
c9b8b07c SC |
8309 | x86_emulator_cache = kvm_alloc_emulator_cache(); |
8310 | if (!x86_emulator_cache) { | |
8311 | pr_err("kvm: failed to allocate cache for x86 emulator\n"); | |
8312 | goto out_free_x86_fpu_cache; | |
8313 | } | |
8314 | ||
7e34fbd0 SC |
8315 | user_return_msrs = alloc_percpu(struct kvm_user_return_msrs); |
8316 | if (!user_return_msrs) { | |
8317 | printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n"); | |
c9b8b07c | 8318 | goto out_free_x86_emulator_cache; |
013f6a5d | 8319 | } |
e5fda4bb | 8320 | kvm_nr_uret_msrs = 0; |
013f6a5d | 8321 | |
97db56ce AK |
8322 | r = kvm_mmu_module_init(); |
8323 | if (r) | |
013f6a5d | 8324 | goto out_free_percpu; |
97db56ce | 8325 | |
b820cc0c | 8326 | kvm_timer_init(); |
c8076604 | 8327 | |
ff9d07a0 ZY |
8328 | perf_register_guest_info_callbacks(&kvm_guest_cbs); |
8329 | ||
cfc48181 | 8330 | if (boot_cpu_has(X86_FEATURE_XSAVE)) { |
2acf923e | 8331 | host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK); |
cfc48181 SC |
8332 | supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0; |
8333 | } | |
2acf923e | 8334 | |
0c5f81da WL |
8335 | if (pi_inject_timer == -1) |
8336 | pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER); | |
16e8d74d MT |
8337 | #ifdef CONFIG_X86_64 |
8338 | pvclock_gtod_register_notifier(&pvclock_gtod_notifier); | |
0092e434 | 8339 | |
5fa4ec9c | 8340 | if (hypervisor_is_type(X86_HYPER_MS_HYPERV)) |
0092e434 | 8341 | set_hv_tscchange_cb(kvm_hyperv_tsc_notifier); |
16e8d74d MT |
8342 | #endif |
8343 | ||
f8c16bba | 8344 | return 0; |
56c6d28a | 8345 | |
013f6a5d | 8346 | out_free_percpu: |
7e34fbd0 | 8347 | free_percpu(user_return_msrs); |
c9b8b07c SC |
8348 | out_free_x86_emulator_cache: |
8349 | kmem_cache_destroy(x86_emulator_cache); | |
b666a4b6 MO |
8350 | out_free_x86_fpu_cache: |
8351 | kmem_cache_destroy(x86_fpu_cache); | |
56c6d28a | 8352 | out: |
56c6d28a | 8353 | return r; |
043405e1 | 8354 | } |
8776e519 | 8355 | |
f8c16bba ZX |
8356 | void kvm_arch_exit(void) |
8357 | { | |
0092e434 | 8358 | #ifdef CONFIG_X86_64 |
5fa4ec9c | 8359 | if (hypervisor_is_type(X86_HYPER_MS_HYPERV)) |
0092e434 VK |
8360 | clear_hv_tscchange_cb(); |
8361 | #endif | |
cef84c30 | 8362 | kvm_lapic_exit(); |
ff9d07a0 ZY |
8363 | perf_unregister_guest_info_callbacks(&kvm_guest_cbs); |
8364 | ||
888d256e JK |
8365 | if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) |
8366 | cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block, | |
8367 | CPUFREQ_TRANSITION_NOTIFIER); | |
251a5fd6 | 8368 | cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE); |
16e8d74d MT |
8369 | #ifdef CONFIG_X86_64 |
8370 | pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier); | |
3f804f6d | 8371 | irq_work_sync(&pvclock_irq_work); |
594b27e6 | 8372 | cancel_work_sync(&pvclock_gtod_work); |
16e8d74d | 8373 | #endif |
afaf0b2f | 8374 | kvm_x86_ops.hardware_enable = NULL; |
56c6d28a | 8375 | kvm_mmu_module_exit(); |
7e34fbd0 | 8376 | free_percpu(user_return_msrs); |
dfdc0a71 | 8377 | kmem_cache_destroy(x86_emulator_cache); |
b666a4b6 | 8378 | kmem_cache_destroy(x86_fpu_cache); |
b59b153d | 8379 | #ifdef CONFIG_KVM_XEN |
c462f859 | 8380 | static_key_deferred_flush(&kvm_xen_enabled); |
7d6bbebb | 8381 | WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key)); |
b59b153d | 8382 | #endif |
56c6d28a | 8383 | } |
f8c16bba | 8384 | |
872f36eb | 8385 | static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason) |
8776e519 HB |
8386 | { |
8387 | ++vcpu->stat.halt_exits; | |
35754c98 | 8388 | if (lapic_in_kernel(vcpu)) { |
647daca2 | 8389 | vcpu->arch.mp_state = state; |
8776e519 HB |
8390 | return 1; |
8391 | } else { | |
647daca2 | 8392 | vcpu->run->exit_reason = reason; |
8776e519 HB |
8393 | return 0; |
8394 | } | |
8395 | } | |
647daca2 TL |
8396 | |
8397 | int kvm_vcpu_halt(struct kvm_vcpu *vcpu) | |
8398 | { | |
8399 | return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT); | |
8400 | } | |
5cb56059 JS |
8401 | EXPORT_SYMBOL_GPL(kvm_vcpu_halt); |
8402 | ||
8403 | int kvm_emulate_halt(struct kvm_vcpu *vcpu) | |
8404 | { | |
6affcbed KH |
8405 | int ret = kvm_skip_emulated_instruction(vcpu); |
8406 | /* | |
8407 | * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered | |
8408 | * KVM_EXIT_DEBUG here. | |
8409 | */ | |
8410 | return kvm_vcpu_halt(vcpu) && ret; | |
5cb56059 | 8411 | } |
8776e519 HB |
8412 | EXPORT_SYMBOL_GPL(kvm_emulate_halt); |
8413 | ||
647daca2 TL |
8414 | int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu) |
8415 | { | |
8416 | int ret = kvm_skip_emulated_instruction(vcpu); | |
8417 | ||
8418 | return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret; | |
8419 | } | |
8420 | EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold); | |
8421 | ||
8ef81a9a | 8422 | #ifdef CONFIG_X86_64 |
55dd00a7 MT |
8423 | static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr, |
8424 | unsigned long clock_type) | |
8425 | { | |
8426 | struct kvm_clock_pairing clock_pairing; | |
899a31f5 | 8427 | struct timespec64 ts; |
80fbd89c | 8428 | u64 cycle; |
55dd00a7 MT |
8429 | int ret; |
8430 | ||
8431 | if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK) | |
8432 | return -KVM_EOPNOTSUPP; | |
8433 | ||
7ca7f3b9 | 8434 | if (!kvm_get_walltime_and_clockread(&ts, &cycle)) |
55dd00a7 MT |
8435 | return -KVM_EOPNOTSUPP; |
8436 | ||
8437 | clock_pairing.sec = ts.tv_sec; | |
8438 | clock_pairing.nsec = ts.tv_nsec; | |
8439 | clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle); | |
8440 | clock_pairing.flags = 0; | |
bcbfbd8e | 8441 | memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad)); |
55dd00a7 MT |
8442 | |
8443 | ret = 0; | |
8444 | if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing, | |
8445 | sizeof(struct kvm_clock_pairing))) | |
8446 | ret = -KVM_EFAULT; | |
8447 | ||
8448 | return ret; | |
8449 | } | |
8ef81a9a | 8450 | #endif |
55dd00a7 | 8451 | |
6aef266c SV |
8452 | /* |
8453 | * kvm_pv_kick_cpu_op: Kick a vcpu. | |
8454 | * | |
8455 | * @apicid - apicid of vcpu to be kicked. | |
8456 | */ | |
8457 | static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid) | |
8458 | { | |
24d2166b | 8459 | struct kvm_lapic_irq lapic_irq; |
6aef266c | 8460 | |
150a84fe | 8461 | lapic_irq.shorthand = APIC_DEST_NOSHORT; |
c96001c5 | 8462 | lapic_irq.dest_mode = APIC_DEST_PHYSICAL; |
ebd28fcb | 8463 | lapic_irq.level = 0; |
24d2166b | 8464 | lapic_irq.dest_id = apicid; |
93bbf0b8 | 8465 | lapic_irq.msi_redir_hint = false; |
6aef266c | 8466 | |
24d2166b | 8467 | lapic_irq.delivery_mode = APIC_DM_REMRD; |
795a149e | 8468 | kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL); |
6aef266c SV |
8469 | } |
8470 | ||
4e19c36f SS |
8471 | bool kvm_apicv_activated(struct kvm *kvm) |
8472 | { | |
8473 | return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0); | |
8474 | } | |
8475 | EXPORT_SYMBOL_GPL(kvm_apicv_activated); | |
8476 | ||
4651fc56 | 8477 | static void kvm_apicv_init(struct kvm *kvm) |
4e19c36f | 8478 | { |
4651fc56 | 8479 | if (enable_apicv) |
4e19c36f SS |
8480 | clear_bit(APICV_INHIBIT_REASON_DISABLE, |
8481 | &kvm->arch.apicv_inhibit_reasons); | |
8482 | else | |
8483 | set_bit(APICV_INHIBIT_REASON_DISABLE, | |
8484 | &kvm->arch.apicv_inhibit_reasons); | |
8485 | } | |
4e19c36f | 8486 | |
4a7132ef | 8487 | static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id) |
71506297 WL |
8488 | { |
8489 | struct kvm_vcpu *target = NULL; | |
8490 | struct kvm_apic_map *map; | |
8491 | ||
4a7132ef WL |
8492 | vcpu->stat.directed_yield_attempted++; |
8493 | ||
72b268a8 WL |
8494 | if (single_task_running()) |
8495 | goto no_yield; | |
8496 | ||
71506297 | 8497 | rcu_read_lock(); |
4a7132ef | 8498 | map = rcu_dereference(vcpu->kvm->arch.apic_map); |
71506297 WL |
8499 | |
8500 | if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id]) | |
8501 | target = map->phys_map[dest_id]->vcpu; | |
8502 | ||
8503 | rcu_read_unlock(); | |
8504 | ||
4a7132ef WL |
8505 | if (!target || !READ_ONCE(target->ready)) |
8506 | goto no_yield; | |
8507 | ||
a1fa4cbd WL |
8508 | /* Ignore requests to yield to self */ |
8509 | if (vcpu == target) | |
8510 | goto no_yield; | |
8511 | ||
4a7132ef WL |
8512 | if (kvm_vcpu_yield_to(target) <= 0) |
8513 | goto no_yield; | |
8514 | ||
8515 | vcpu->stat.directed_yield_successful++; | |
8516 | ||
8517 | no_yield: | |
8518 | return; | |
71506297 WL |
8519 | } |
8520 | ||
8776e519 HB |
8521 | int kvm_emulate_hypercall(struct kvm_vcpu *vcpu) |
8522 | { | |
8523 | unsigned long nr, a0, a1, a2, a3, ret; | |
6356ee0c | 8524 | int op_64_bit; |
8776e519 | 8525 | |
23200b7a JM |
8526 | if (kvm_xen_hypercall_enabled(vcpu->kvm)) |
8527 | return kvm_xen_hypercall(vcpu); | |
8528 | ||
8f014550 | 8529 | if (kvm_hv_hypercall_enabled(vcpu)) |
696ca779 | 8530 | return kvm_hv_hypercall(vcpu); |
55cd8e5a | 8531 | |
de3cd117 SC |
8532 | nr = kvm_rax_read(vcpu); |
8533 | a0 = kvm_rbx_read(vcpu); | |
8534 | a1 = kvm_rcx_read(vcpu); | |
8535 | a2 = kvm_rdx_read(vcpu); | |
8536 | a3 = kvm_rsi_read(vcpu); | |
8776e519 | 8537 | |
229456fc | 8538 | trace_kvm_hypercall(nr, a0, a1, a2, a3); |
2714d1d3 | 8539 | |
a449c7aa NA |
8540 | op_64_bit = is_64_bit_mode(vcpu); |
8541 | if (!op_64_bit) { | |
8776e519 HB |
8542 | nr &= 0xFFFFFFFF; |
8543 | a0 &= 0xFFFFFFFF; | |
8544 | a1 &= 0xFFFFFFFF; | |
8545 | a2 &= 0xFFFFFFFF; | |
8546 | a3 &= 0xFFFFFFFF; | |
8547 | } | |
8548 | ||
b3646477 | 8549 | if (static_call(kvm_x86_get_cpl)(vcpu) != 0) { |
07708c4a | 8550 | ret = -KVM_EPERM; |
696ca779 | 8551 | goto out; |
07708c4a JK |
8552 | } |
8553 | ||
66570e96 OU |
8554 | ret = -KVM_ENOSYS; |
8555 | ||
8776e519 | 8556 | switch (nr) { |
b93463aa AK |
8557 | case KVM_HC_VAPIC_POLL_IRQ: |
8558 | ret = 0; | |
8559 | break; | |
6aef266c | 8560 | case KVM_HC_KICK_CPU: |
66570e96 OU |
8561 | if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT)) |
8562 | break; | |
8563 | ||
6aef266c | 8564 | kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1); |
4a7132ef | 8565 | kvm_sched_yield(vcpu, a1); |
6aef266c SV |
8566 | ret = 0; |
8567 | break; | |
8ef81a9a | 8568 | #ifdef CONFIG_X86_64 |
55dd00a7 MT |
8569 | case KVM_HC_CLOCK_PAIRING: |
8570 | ret = kvm_pv_clock_pairing(vcpu, a0, a1); | |
8571 | break; | |
1ed199a4 | 8572 | #endif |
4180bf1b | 8573 | case KVM_HC_SEND_IPI: |
66570e96 OU |
8574 | if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI)) |
8575 | break; | |
8576 | ||
4180bf1b WL |
8577 | ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit); |
8578 | break; | |
71506297 | 8579 | case KVM_HC_SCHED_YIELD: |
66570e96 OU |
8580 | if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD)) |
8581 | break; | |
8582 | ||
4a7132ef | 8583 | kvm_sched_yield(vcpu, a0); |
71506297 WL |
8584 | ret = 0; |
8585 | break; | |
8776e519 HB |
8586 | default: |
8587 | ret = -KVM_ENOSYS; | |
8588 | break; | |
8589 | } | |
696ca779 | 8590 | out: |
a449c7aa NA |
8591 | if (!op_64_bit) |
8592 | ret = (u32)ret; | |
de3cd117 | 8593 | kvm_rax_write(vcpu, ret); |
6356ee0c | 8594 | |
f11c3a8d | 8595 | ++vcpu->stat.hypercalls; |
6356ee0c | 8596 | return kvm_skip_emulated_instruction(vcpu); |
8776e519 HB |
8597 | } |
8598 | EXPORT_SYMBOL_GPL(kvm_emulate_hypercall); | |
8599 | ||
b6785def | 8600 | static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt) |
8776e519 | 8601 | { |
d6aa1000 | 8602 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
8776e519 | 8603 | char instruction[3]; |
5fdbf976 | 8604 | unsigned long rip = kvm_rip_read(vcpu); |
8776e519 | 8605 | |
b3646477 | 8606 | static_call(kvm_x86_patch_hypercall)(vcpu, instruction); |
8776e519 | 8607 | |
ce2e852e DV |
8608 | return emulator_write_emulated(ctxt, rip, instruction, 3, |
8609 | &ctxt->exception); | |
8776e519 HB |
8610 | } |
8611 | ||
851ba692 | 8612 | static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu) |
b6c7a5dc | 8613 | { |
782d422b MG |
8614 | return vcpu->run->request_interrupt_window && |
8615 | likely(!pic_in_kernel(vcpu->kvm)); | |
b6c7a5dc HB |
8616 | } |
8617 | ||
851ba692 | 8618 | static void post_kvm_run_save(struct kvm_vcpu *vcpu) |
b6c7a5dc | 8619 | { |
851ba692 AK |
8620 | struct kvm_run *kvm_run = vcpu->run; |
8621 | ||
f1c6366e TL |
8622 | /* |
8623 | * if_flag is obsolete and useless, so do not bother | |
8624 | * setting it for SEV-ES guests. Userspace can just | |
8625 | * use kvm_run->ready_for_interrupt_injection. | |
8626 | */ | |
8627 | kvm_run->if_flag = !vcpu->arch.guest_state_protected | |
8628 | && (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0; | |
8629 | ||
2d3ad1f4 | 8630 | kvm_run->cr8 = kvm_get_cr8(vcpu); |
b6c7a5dc | 8631 | kvm_run->apic_base = kvm_get_apic_base(vcpu); |
127a457a MG |
8632 | kvm_run->ready_for_interrupt_injection = |
8633 | pic_in_kernel(vcpu->kvm) || | |
782d422b | 8634 | kvm_vcpu_ready_for_interrupt_injection(vcpu); |
15aad3be CQ |
8635 | |
8636 | if (is_smm(vcpu)) | |
8637 | kvm_run->flags |= KVM_RUN_X86_SMM; | |
b6c7a5dc HB |
8638 | } |
8639 | ||
95ba8273 GN |
8640 | static void update_cr8_intercept(struct kvm_vcpu *vcpu) |
8641 | { | |
8642 | int max_irr, tpr; | |
8643 | ||
afaf0b2f | 8644 | if (!kvm_x86_ops.update_cr8_intercept) |
95ba8273 GN |
8645 | return; |
8646 | ||
bce87cce | 8647 | if (!lapic_in_kernel(vcpu)) |
88c808fd AK |
8648 | return; |
8649 | ||
d62caabb AS |
8650 | if (vcpu->arch.apicv_active) |
8651 | return; | |
8652 | ||
8db3baa2 GN |
8653 | if (!vcpu->arch.apic->vapic_addr) |
8654 | max_irr = kvm_lapic_find_highest_irr(vcpu); | |
8655 | else | |
8656 | max_irr = -1; | |
95ba8273 GN |
8657 | |
8658 | if (max_irr != -1) | |
8659 | max_irr >>= 4; | |
8660 | ||
8661 | tpr = kvm_lapic_get_cr8(vcpu); | |
8662 | ||
b3646477 | 8663 | static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr); |
95ba8273 GN |
8664 | } |
8665 | ||
b97f0745 | 8666 | |
cb6a32c2 SC |
8667 | int kvm_check_nested_events(struct kvm_vcpu *vcpu) |
8668 | { | |
cb6a32c2 SC |
8669 | if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) { |
8670 | kvm_x86_ops.nested_ops->triple_fault(vcpu); | |
8671 | return 1; | |
8672 | } | |
8673 | ||
8674 | return kvm_x86_ops.nested_ops->check_events(vcpu); | |
8675 | } | |
8676 | ||
b97f0745 ML |
8677 | static void kvm_inject_exception(struct kvm_vcpu *vcpu) |
8678 | { | |
8679 | if (vcpu->arch.exception.error_code && !is_protmode(vcpu)) | |
8680 | vcpu->arch.exception.error_code = false; | |
8681 | static_call(kvm_x86_queue_exception)(vcpu); | |
8682 | } | |
8683 | ||
a5f6909a | 8684 | static int inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit) |
95ba8273 | 8685 | { |
b6b8a145 | 8686 | int r; |
c6b22f59 | 8687 | bool can_inject = true; |
b6b8a145 | 8688 | |
95ba8273 | 8689 | /* try to reinject previous events if any */ |
664f8e26 | 8690 | |
c6b22f59 | 8691 | if (vcpu->arch.exception.injected) { |
b97f0745 | 8692 | kvm_inject_exception(vcpu); |
c6b22f59 PB |
8693 | can_inject = false; |
8694 | } | |
664f8e26 | 8695 | /* |
a042c26f LA |
8696 | * Do not inject an NMI or interrupt if there is a pending |
8697 | * exception. Exceptions and interrupts are recognized at | |
8698 | * instruction boundaries, i.e. the start of an instruction. | |
8699 | * Trap-like exceptions, e.g. #DB, have higher priority than | |
8700 | * NMIs and interrupts, i.e. traps are recognized before an | |
8701 | * NMI/interrupt that's pending on the same instruction. | |
8702 | * Fault-like exceptions, e.g. #GP and #PF, are the lowest | |
8703 | * priority, but are only generated (pended) during instruction | |
8704 | * execution, i.e. a pending fault-like exception means the | |
8705 | * fault occurred on the *previous* instruction and must be | |
8706 | * serviced prior to recognizing any new events in order to | |
8707 | * fully complete the previous instruction. | |
664f8e26 | 8708 | */ |
1a680e35 | 8709 | else if (!vcpu->arch.exception.pending) { |
c6b22f59 | 8710 | if (vcpu->arch.nmi_injected) { |
b3646477 | 8711 | static_call(kvm_x86_set_nmi)(vcpu); |
c6b22f59 PB |
8712 | can_inject = false; |
8713 | } else if (vcpu->arch.interrupt.injected) { | |
b3646477 | 8714 | static_call(kvm_x86_set_irq)(vcpu); |
c6b22f59 PB |
8715 | can_inject = false; |
8716 | } | |
664f8e26 WL |
8717 | } |
8718 | ||
3b82b8d7 SC |
8719 | WARN_ON_ONCE(vcpu->arch.exception.injected && |
8720 | vcpu->arch.exception.pending); | |
8721 | ||
1a680e35 LA |
8722 | /* |
8723 | * Call check_nested_events() even if we reinjected a previous event | |
8724 | * in order for caller to determine if it should require immediate-exit | |
8725 | * from L2 to L1 due to pending L1 events which require exit | |
8726 | * from L2 to L1. | |
8727 | */ | |
56083bdf | 8728 | if (is_guest_mode(vcpu)) { |
cb6a32c2 | 8729 | r = kvm_check_nested_events(vcpu); |
c9d40913 | 8730 | if (r < 0) |
a5f6909a | 8731 | goto out; |
664f8e26 WL |
8732 | } |
8733 | ||
8734 | /* try to inject new event if pending */ | |
b59bb7bd | 8735 | if (vcpu->arch.exception.pending) { |
5c1c85d0 AK |
8736 | trace_kvm_inj_exception(vcpu->arch.exception.nr, |
8737 | vcpu->arch.exception.has_error_code, | |
8738 | vcpu->arch.exception.error_code); | |
d6e8c854 | 8739 | |
664f8e26 WL |
8740 | vcpu->arch.exception.pending = false; |
8741 | vcpu->arch.exception.injected = true; | |
8742 | ||
d6e8c854 NA |
8743 | if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT) |
8744 | __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) | | |
8745 | X86_EFLAGS_RF); | |
8746 | ||
f10c729f | 8747 | if (vcpu->arch.exception.nr == DB_VECTOR) { |
f10c729f JM |
8748 | kvm_deliver_exception_payload(vcpu); |
8749 | if (vcpu->arch.dr7 & DR7_GD) { | |
8750 | vcpu->arch.dr7 &= ~DR7_GD; | |
8751 | kvm_update_dr7(vcpu); | |
8752 | } | |
6bdf0662 NA |
8753 | } |
8754 | ||
b97f0745 | 8755 | kvm_inject_exception(vcpu); |
c6b22f59 | 8756 | can_inject = false; |
1a680e35 LA |
8757 | } |
8758 | ||
c9d40913 PB |
8759 | /* |
8760 | * Finally, inject interrupt events. If an event cannot be injected | |
8761 | * due to architectural conditions (e.g. IF=0) a window-open exit | |
8762 | * will re-request KVM_REQ_EVENT. Sometimes however an event is pending | |
8763 | * and can architecturally be injected, but we cannot do it right now: | |
8764 | * an interrupt could have arrived just now and we have to inject it | |
8765 | * as a vmexit, or there could already an event in the queue, which is | |
8766 | * indicated by can_inject. In that case we request an immediate exit | |
8767 | * in order to make progress and get back here for another iteration. | |
8768 | * The kvm_x86_ops hooks communicate this by returning -EBUSY. | |
8769 | */ | |
8770 | if (vcpu->arch.smi_pending) { | |
b3646477 | 8771 | r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY; |
c9d40913 | 8772 | if (r < 0) |
a5f6909a | 8773 | goto out; |
c9d40913 PB |
8774 | if (r) { |
8775 | vcpu->arch.smi_pending = false; | |
8776 | ++vcpu->arch.smi_count; | |
8777 | enter_smm(vcpu); | |
8778 | can_inject = false; | |
8779 | } else | |
b3646477 | 8780 | static_call(kvm_x86_enable_smi_window)(vcpu); |
c9d40913 PB |
8781 | } |
8782 | ||
8783 | if (vcpu->arch.nmi_pending) { | |
b3646477 | 8784 | r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY; |
c9d40913 | 8785 | if (r < 0) |
a5f6909a | 8786 | goto out; |
c9d40913 PB |
8787 | if (r) { |
8788 | --vcpu->arch.nmi_pending; | |
8789 | vcpu->arch.nmi_injected = true; | |
b3646477 | 8790 | static_call(kvm_x86_set_nmi)(vcpu); |
c9d40913 | 8791 | can_inject = false; |
b3646477 | 8792 | WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0); |
c9d40913 PB |
8793 | } |
8794 | if (vcpu->arch.nmi_pending) | |
b3646477 | 8795 | static_call(kvm_x86_enable_nmi_window)(vcpu); |
c9d40913 | 8796 | } |
1a680e35 | 8797 | |
c9d40913 | 8798 | if (kvm_cpu_has_injectable_intr(vcpu)) { |
b3646477 | 8799 | r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY; |
c9d40913 | 8800 | if (r < 0) |
a5f6909a | 8801 | goto out; |
c9d40913 PB |
8802 | if (r) { |
8803 | kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false); | |
b3646477 JB |
8804 | static_call(kvm_x86_set_irq)(vcpu); |
8805 | WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0); | |
c9d40913 PB |
8806 | } |
8807 | if (kvm_cpu_has_injectable_intr(vcpu)) | |
b3646477 | 8808 | static_call(kvm_x86_enable_irq_window)(vcpu); |
95ba8273 | 8809 | } |
ee2cd4b7 | 8810 | |
c9d40913 PB |
8811 | if (is_guest_mode(vcpu) && |
8812 | kvm_x86_ops.nested_ops->hv_timer_pending && | |
8813 | kvm_x86_ops.nested_ops->hv_timer_pending(vcpu)) | |
8814 | *req_immediate_exit = true; | |
8815 | ||
8816 | WARN_ON(vcpu->arch.exception.pending); | |
a5f6909a | 8817 | return 0; |
c9d40913 | 8818 | |
a5f6909a JM |
8819 | out: |
8820 | if (r == -EBUSY) { | |
8821 | *req_immediate_exit = true; | |
8822 | r = 0; | |
8823 | } | |
8824 | return r; | |
95ba8273 GN |
8825 | } |
8826 | ||
7460fb4a AK |
8827 | static void process_nmi(struct kvm_vcpu *vcpu) |
8828 | { | |
8829 | unsigned limit = 2; | |
8830 | ||
8831 | /* | |
8832 | * x86 is limited to one NMI running, and one NMI pending after it. | |
8833 | * If an NMI is already in progress, limit further NMIs to just one. | |
8834 | * Otherwise, allow two (and we'll inject the first one immediately). | |
8835 | */ | |
b3646477 | 8836 | if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected) |
7460fb4a AK |
8837 | limit = 1; |
8838 | ||
8839 | vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0); | |
8840 | vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit); | |
8841 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
8842 | } | |
8843 | ||
ee2cd4b7 | 8844 | static u32 enter_smm_get_segment_flags(struct kvm_segment *seg) |
660a5d51 PB |
8845 | { |
8846 | u32 flags = 0; | |
8847 | flags |= seg->g << 23; | |
8848 | flags |= seg->db << 22; | |
8849 | flags |= seg->l << 21; | |
8850 | flags |= seg->avl << 20; | |
8851 | flags |= seg->present << 15; | |
8852 | flags |= seg->dpl << 13; | |
8853 | flags |= seg->s << 12; | |
8854 | flags |= seg->type << 8; | |
8855 | return flags; | |
8856 | } | |
8857 | ||
ee2cd4b7 | 8858 | static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n) |
660a5d51 PB |
8859 | { |
8860 | struct kvm_segment seg; | |
8861 | int offset; | |
8862 | ||
8863 | kvm_get_segment(vcpu, &seg, n); | |
8864 | put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector); | |
8865 | ||
8866 | if (n < 3) | |
8867 | offset = 0x7f84 + n * 12; | |
8868 | else | |
8869 | offset = 0x7f2c + (n - 3) * 12; | |
8870 | ||
8871 | put_smstate(u32, buf, offset + 8, seg.base); | |
8872 | put_smstate(u32, buf, offset + 4, seg.limit); | |
ee2cd4b7 | 8873 | put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg)); |
660a5d51 PB |
8874 | } |
8875 | ||
efbb288a | 8876 | #ifdef CONFIG_X86_64 |
ee2cd4b7 | 8877 | static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n) |
660a5d51 PB |
8878 | { |
8879 | struct kvm_segment seg; | |
8880 | int offset; | |
8881 | u16 flags; | |
8882 | ||
8883 | kvm_get_segment(vcpu, &seg, n); | |
8884 | offset = 0x7e00 + n * 16; | |
8885 | ||
ee2cd4b7 | 8886 | flags = enter_smm_get_segment_flags(&seg) >> 8; |
660a5d51 PB |
8887 | put_smstate(u16, buf, offset, seg.selector); |
8888 | put_smstate(u16, buf, offset + 2, flags); | |
8889 | put_smstate(u32, buf, offset + 4, seg.limit); | |
8890 | put_smstate(u64, buf, offset + 8, seg.base); | |
8891 | } | |
efbb288a | 8892 | #endif |
660a5d51 | 8893 | |
ee2cd4b7 | 8894 | static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf) |
660a5d51 PB |
8895 | { |
8896 | struct desc_ptr dt; | |
8897 | struct kvm_segment seg; | |
8898 | unsigned long val; | |
8899 | int i; | |
8900 | ||
8901 | put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu)); | |
8902 | put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu)); | |
8903 | put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu)); | |
8904 | put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu)); | |
8905 | ||
8906 | for (i = 0; i < 8; i++) | |
27b4a9c4 | 8907 | put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i)); |
660a5d51 PB |
8908 | |
8909 | kvm_get_dr(vcpu, 6, &val); | |
8910 | put_smstate(u32, buf, 0x7fcc, (u32)val); | |
8911 | kvm_get_dr(vcpu, 7, &val); | |
8912 | put_smstate(u32, buf, 0x7fc8, (u32)val); | |
8913 | ||
8914 | kvm_get_segment(vcpu, &seg, VCPU_SREG_TR); | |
8915 | put_smstate(u32, buf, 0x7fc4, seg.selector); | |
8916 | put_smstate(u32, buf, 0x7f64, seg.base); | |
8917 | put_smstate(u32, buf, 0x7f60, seg.limit); | |
ee2cd4b7 | 8918 | put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg)); |
660a5d51 PB |
8919 | |
8920 | kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR); | |
8921 | put_smstate(u32, buf, 0x7fc0, seg.selector); | |
8922 | put_smstate(u32, buf, 0x7f80, seg.base); | |
8923 | put_smstate(u32, buf, 0x7f7c, seg.limit); | |
ee2cd4b7 | 8924 | put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg)); |
660a5d51 | 8925 | |
b3646477 | 8926 | static_call(kvm_x86_get_gdt)(vcpu, &dt); |
660a5d51 PB |
8927 | put_smstate(u32, buf, 0x7f74, dt.address); |
8928 | put_smstate(u32, buf, 0x7f70, dt.size); | |
8929 | ||
b3646477 | 8930 | static_call(kvm_x86_get_idt)(vcpu, &dt); |
660a5d51 PB |
8931 | put_smstate(u32, buf, 0x7f58, dt.address); |
8932 | put_smstate(u32, buf, 0x7f54, dt.size); | |
8933 | ||
8934 | for (i = 0; i < 6; i++) | |
ee2cd4b7 | 8935 | enter_smm_save_seg_32(vcpu, buf, i); |
660a5d51 PB |
8936 | |
8937 | put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu)); | |
8938 | ||
8939 | /* revision id */ | |
8940 | put_smstate(u32, buf, 0x7efc, 0x00020000); | |
8941 | put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase); | |
8942 | } | |
8943 | ||
b68f3cc7 | 8944 | #ifdef CONFIG_X86_64 |
ee2cd4b7 | 8945 | static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf) |
660a5d51 | 8946 | { |
660a5d51 PB |
8947 | struct desc_ptr dt; |
8948 | struct kvm_segment seg; | |
8949 | unsigned long val; | |
8950 | int i; | |
8951 | ||
8952 | for (i = 0; i < 16; i++) | |
27b4a9c4 | 8953 | put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i)); |
660a5d51 PB |
8954 | |
8955 | put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu)); | |
8956 | put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu)); | |
8957 | ||
8958 | kvm_get_dr(vcpu, 6, &val); | |
8959 | put_smstate(u64, buf, 0x7f68, val); | |
8960 | kvm_get_dr(vcpu, 7, &val); | |
8961 | put_smstate(u64, buf, 0x7f60, val); | |
8962 | ||
8963 | put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu)); | |
8964 | put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu)); | |
8965 | put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu)); | |
8966 | ||
8967 | put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase); | |
8968 | ||
8969 | /* revision id */ | |
8970 | put_smstate(u32, buf, 0x7efc, 0x00020064); | |
8971 | ||
8972 | put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer); | |
8973 | ||
8974 | kvm_get_segment(vcpu, &seg, VCPU_SREG_TR); | |
8975 | put_smstate(u16, buf, 0x7e90, seg.selector); | |
ee2cd4b7 | 8976 | put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8); |
660a5d51 PB |
8977 | put_smstate(u32, buf, 0x7e94, seg.limit); |
8978 | put_smstate(u64, buf, 0x7e98, seg.base); | |
8979 | ||
b3646477 | 8980 | static_call(kvm_x86_get_idt)(vcpu, &dt); |
660a5d51 PB |
8981 | put_smstate(u32, buf, 0x7e84, dt.size); |
8982 | put_smstate(u64, buf, 0x7e88, dt.address); | |
8983 | ||
8984 | kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR); | |
8985 | put_smstate(u16, buf, 0x7e70, seg.selector); | |
ee2cd4b7 | 8986 | put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8); |
660a5d51 PB |
8987 | put_smstate(u32, buf, 0x7e74, seg.limit); |
8988 | put_smstate(u64, buf, 0x7e78, seg.base); | |
8989 | ||
b3646477 | 8990 | static_call(kvm_x86_get_gdt)(vcpu, &dt); |
660a5d51 PB |
8991 | put_smstate(u32, buf, 0x7e64, dt.size); |
8992 | put_smstate(u64, buf, 0x7e68, dt.address); | |
8993 | ||
8994 | for (i = 0; i < 6; i++) | |
ee2cd4b7 | 8995 | enter_smm_save_seg_64(vcpu, buf, i); |
660a5d51 | 8996 | } |
b68f3cc7 | 8997 | #endif |
660a5d51 | 8998 | |
ee2cd4b7 | 8999 | static void enter_smm(struct kvm_vcpu *vcpu) |
64d60670 | 9000 | { |
660a5d51 | 9001 | struct kvm_segment cs, ds; |
18c3626e | 9002 | struct desc_ptr dt; |
660a5d51 PB |
9003 | char buf[512]; |
9004 | u32 cr0; | |
9005 | ||
660a5d51 | 9006 | memset(buf, 0, 512); |
b68f3cc7 | 9007 | #ifdef CONFIG_X86_64 |
d6321d49 | 9008 | if (guest_cpuid_has(vcpu, X86_FEATURE_LM)) |
ee2cd4b7 | 9009 | enter_smm_save_state_64(vcpu, buf); |
660a5d51 | 9010 | else |
b68f3cc7 | 9011 | #endif |
ee2cd4b7 | 9012 | enter_smm_save_state_32(vcpu, buf); |
660a5d51 | 9013 | |
0234bf88 LP |
9014 | /* |
9015 | * Give pre_enter_smm() a chance to make ISA-specific changes to the | |
9016 | * vCPU state (e.g. leave guest mode) after we've saved the state into | |
9017 | * the SMM state-save area. | |
9018 | */ | |
b3646477 | 9019 | static_call(kvm_x86_pre_enter_smm)(vcpu, buf); |
0234bf88 | 9020 | |
dc87275f | 9021 | kvm_smm_changed(vcpu, true); |
54bf36aa | 9022 | kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf)); |
660a5d51 | 9023 | |
b3646477 | 9024 | if (static_call(kvm_x86_get_nmi_mask)(vcpu)) |
660a5d51 PB |
9025 | vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK; |
9026 | else | |
b3646477 | 9027 | static_call(kvm_x86_set_nmi_mask)(vcpu, true); |
660a5d51 PB |
9028 | |
9029 | kvm_set_rflags(vcpu, X86_EFLAGS_FIXED); | |
9030 | kvm_rip_write(vcpu, 0x8000); | |
9031 | ||
9032 | cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG); | |
b3646477 | 9033 | static_call(kvm_x86_set_cr0)(vcpu, cr0); |
660a5d51 PB |
9034 | vcpu->arch.cr0 = cr0; |
9035 | ||
b3646477 | 9036 | static_call(kvm_x86_set_cr4)(vcpu, 0); |
660a5d51 | 9037 | |
18c3626e PB |
9038 | /* Undocumented: IDT limit is set to zero on entry to SMM. */ |
9039 | dt.address = dt.size = 0; | |
b3646477 | 9040 | static_call(kvm_x86_set_idt)(vcpu, &dt); |
18c3626e | 9041 | |
996ff542 | 9042 | kvm_set_dr(vcpu, 7, DR7_FIXED_1); |
660a5d51 PB |
9043 | |
9044 | cs.selector = (vcpu->arch.smbase >> 4) & 0xffff; | |
9045 | cs.base = vcpu->arch.smbase; | |
9046 | ||
9047 | ds.selector = 0; | |
9048 | ds.base = 0; | |
9049 | ||
9050 | cs.limit = ds.limit = 0xffffffff; | |
9051 | cs.type = ds.type = 0x3; | |
9052 | cs.dpl = ds.dpl = 0; | |
9053 | cs.db = ds.db = 0; | |
9054 | cs.s = ds.s = 1; | |
9055 | cs.l = ds.l = 0; | |
9056 | cs.g = ds.g = 1; | |
9057 | cs.avl = ds.avl = 0; | |
9058 | cs.present = ds.present = 1; | |
9059 | cs.unusable = ds.unusable = 0; | |
9060 | cs.padding = ds.padding = 0; | |
9061 | ||
9062 | kvm_set_segment(vcpu, &cs, VCPU_SREG_CS); | |
9063 | kvm_set_segment(vcpu, &ds, VCPU_SREG_DS); | |
9064 | kvm_set_segment(vcpu, &ds, VCPU_SREG_ES); | |
9065 | kvm_set_segment(vcpu, &ds, VCPU_SREG_FS); | |
9066 | kvm_set_segment(vcpu, &ds, VCPU_SREG_GS); | |
9067 | kvm_set_segment(vcpu, &ds, VCPU_SREG_SS); | |
9068 | ||
b68f3cc7 | 9069 | #ifdef CONFIG_X86_64 |
d6321d49 | 9070 | if (guest_cpuid_has(vcpu, X86_FEATURE_LM)) |
b3646477 | 9071 | static_call(kvm_x86_set_efer)(vcpu, 0); |
b68f3cc7 | 9072 | #endif |
660a5d51 | 9073 | |
aedbaf4f | 9074 | kvm_update_cpuid_runtime(vcpu); |
660a5d51 | 9075 | kvm_mmu_reset_context(vcpu); |
64d60670 PB |
9076 | } |
9077 | ||
ee2cd4b7 | 9078 | static void process_smi(struct kvm_vcpu *vcpu) |
c43203ca PB |
9079 | { |
9080 | vcpu->arch.smi_pending = true; | |
9081 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
9082 | } | |
9083 | ||
7ee30bc1 NNL |
9084 | void kvm_make_scan_ioapic_request_mask(struct kvm *kvm, |
9085 | unsigned long *vcpu_bitmap) | |
9086 | { | |
9087 | cpumask_var_t cpus; | |
7ee30bc1 NNL |
9088 | |
9089 | zalloc_cpumask_var(&cpus, GFP_ATOMIC); | |
9090 | ||
db5a95ec | 9091 | kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC, |
54163a34 | 9092 | NULL, vcpu_bitmap, cpus); |
7ee30bc1 NNL |
9093 | |
9094 | free_cpumask_var(cpus); | |
9095 | } | |
9096 | ||
2860c4b1 PB |
9097 | void kvm_make_scan_ioapic_request(struct kvm *kvm) |
9098 | { | |
9099 | kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC); | |
9100 | } | |
9101 | ||
8df14af4 SS |
9102 | void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu) |
9103 | { | |
9104 | if (!lapic_in_kernel(vcpu)) | |
9105 | return; | |
9106 | ||
9107 | vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm); | |
9108 | kvm_apic_update_apicv(vcpu); | |
b3646477 | 9109 | static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu); |
8df14af4 SS |
9110 | } |
9111 | EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv); | |
9112 | ||
9113 | /* | |
9114 | * NOTE: Do not hold any lock prior to calling this. | |
9115 | * | |
9116 | * In particular, kvm_request_apicv_update() expects kvm->srcu not to be | |
9117 | * locked, because it calls __x86_set_memory_region() which does | |
9118 | * synchronize_srcu(&kvm->srcu). | |
9119 | */ | |
9120 | void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit) | |
9121 | { | |
7d611233 | 9122 | struct kvm_vcpu *except; |
8e205a6b PB |
9123 | unsigned long old, new, expected; |
9124 | ||
afaf0b2f | 9125 | if (!kvm_x86_ops.check_apicv_inhibit_reasons || |
b3646477 | 9126 | !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit)) |
ef8efd7a SS |
9127 | return; |
9128 | ||
8e205a6b PB |
9129 | old = READ_ONCE(kvm->arch.apicv_inhibit_reasons); |
9130 | do { | |
9131 | expected = new = old; | |
9132 | if (activate) | |
9133 | __clear_bit(bit, &new); | |
9134 | else | |
9135 | __set_bit(bit, &new); | |
9136 | if (new == old) | |
9137 | break; | |
9138 | old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new); | |
9139 | } while (old != expected); | |
9140 | ||
9141 | if (!!old == !!new) | |
9142 | return; | |
8df14af4 | 9143 | |
24bbf74c | 9144 | trace_kvm_apicv_update_request(activate, bit); |
afaf0b2f | 9145 | if (kvm_x86_ops.pre_update_apicv_exec_ctrl) |
b3646477 | 9146 | static_call(kvm_x86_pre_update_apicv_exec_ctrl)(kvm, activate); |
7d611233 SS |
9147 | |
9148 | /* | |
9149 | * Sending request to update APICV for all other vcpus, | |
9150 | * while update the calling vcpu immediately instead of | |
9151 | * waiting for another #VMEXIT to handle the request. | |
9152 | */ | |
9153 | except = kvm_get_running_vcpu(); | |
9154 | kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE, | |
9155 | except); | |
9156 | if (except) | |
9157 | kvm_vcpu_update_apicv(except); | |
8df14af4 SS |
9158 | } |
9159 | EXPORT_SYMBOL_GPL(kvm_request_apicv_update); | |
9160 | ||
3d81bc7e | 9161 | static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu) |
c7c9c56c | 9162 | { |
dcbd3e49 | 9163 | if (!kvm_apic_present(vcpu)) |
3d81bc7e | 9164 | return; |
c7c9c56c | 9165 | |
6308630b | 9166 | bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256); |
c7c9c56c | 9167 | |
b053b2ae | 9168 | if (irqchip_split(vcpu->kvm)) |
6308630b | 9169 | kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors); |
db2bdcbb | 9170 | else { |
fa59cc00 | 9171 | if (vcpu->arch.apicv_active) |
b3646477 | 9172 | static_call(kvm_x86_sync_pir_to_irr)(vcpu); |
e97f852f WL |
9173 | if (ioapic_in_kernel(vcpu->kvm)) |
9174 | kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors); | |
db2bdcbb | 9175 | } |
e40ff1d6 LA |
9176 | |
9177 | if (is_guest_mode(vcpu)) | |
9178 | vcpu->arch.load_eoi_exitmap_pending = true; | |
9179 | else | |
9180 | kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu); | |
9181 | } | |
9182 | ||
9183 | static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu) | |
9184 | { | |
9185 | u64 eoi_exit_bitmap[4]; | |
9186 | ||
9187 | if (!kvm_apic_hw_enabled(vcpu->arch.apic)) | |
9188 | return; | |
9189 | ||
f2bc14b6 VK |
9190 | if (to_hv_vcpu(vcpu)) |
9191 | bitmap_or((ulong *)eoi_exit_bitmap, | |
9192 | vcpu->arch.ioapic_handled_vectors, | |
9193 | to_hv_synic(vcpu)->vec_bitmap, 256); | |
9194 | ||
b3646477 | 9195 | static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap); |
c7c9c56c YZ |
9196 | } |
9197 | ||
e649b3f0 ET |
9198 | void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm, |
9199 | unsigned long start, unsigned long end) | |
b1394e74 RK |
9200 | { |
9201 | unsigned long apic_address; | |
9202 | ||
9203 | /* | |
9204 | * The physical address of apic access page is stored in the VMCS. | |
9205 | * Update it when it becomes invalid. | |
9206 | */ | |
9207 | apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT); | |
9208 | if (start <= apic_address && apic_address < end) | |
9209 | kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD); | |
9210 | } | |
9211 | ||
4256f43f TC |
9212 | void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu) |
9213 | { | |
35754c98 | 9214 | if (!lapic_in_kernel(vcpu)) |
f439ed27 PB |
9215 | return; |
9216 | ||
afaf0b2f | 9217 | if (!kvm_x86_ops.set_apic_access_page_addr) |
4256f43f TC |
9218 | return; |
9219 | ||
b3646477 | 9220 | static_call(kvm_x86_set_apic_access_page_addr)(vcpu); |
4256f43f | 9221 | } |
4256f43f | 9222 | |
d264ee0c SC |
9223 | void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu) |
9224 | { | |
9225 | smp_send_reschedule(vcpu->cpu); | |
9226 | } | |
9227 | EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit); | |
9228 | ||
9357d939 | 9229 | /* |
362c698f | 9230 | * Returns 1 to let vcpu_run() continue the guest execution loop without |
9357d939 TY |
9231 | * exiting to the userspace. Otherwise, the value will be returned to the |
9232 | * userspace. | |
9233 | */ | |
851ba692 | 9234 | static int vcpu_enter_guest(struct kvm_vcpu *vcpu) |
b6c7a5dc HB |
9235 | { |
9236 | int r; | |
62a193ed MG |
9237 | bool req_int_win = |
9238 | dm_request_for_irq_injection(vcpu) && | |
9239 | kvm_cpu_accept_dm_intr(vcpu); | |
404d5d7b | 9240 | fastpath_t exit_fastpath; |
62a193ed | 9241 | |
730dca42 | 9242 | bool req_immediate_exit = false; |
b6c7a5dc | 9243 | |
fb04a1ed PX |
9244 | /* Forbid vmenter if vcpu dirty ring is soft-full */ |
9245 | if (unlikely(vcpu->kvm->dirty_ring_size && | |
9246 | kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) { | |
9247 | vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL; | |
9248 | trace_kvm_dirty_ring_exit(vcpu); | |
9249 | r = 0; | |
9250 | goto out; | |
9251 | } | |
9252 | ||
2fa6e1e1 | 9253 | if (kvm_request_pending(vcpu)) { |
729c15c2 | 9254 | if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) { |
9a78e158 | 9255 | if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) { |
671ddc70 JM |
9256 | r = 0; |
9257 | goto out; | |
9258 | } | |
9259 | } | |
a8eeb04a | 9260 | if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) |
2e53d63a | 9261 | kvm_mmu_unload(vcpu); |
a8eeb04a | 9262 | if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu)) |
2f599714 | 9263 | __kvm_migrate_timers(vcpu); |
d828199e MT |
9264 | if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu)) |
9265 | kvm_gen_update_masterclock(vcpu->kvm); | |
0061d53d MT |
9266 | if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu)) |
9267 | kvm_gen_kvmclock_update(vcpu); | |
34c238a1 ZA |
9268 | if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) { |
9269 | r = kvm_guest_time_update(vcpu); | |
8cfdc000 ZA |
9270 | if (unlikely(r)) |
9271 | goto out; | |
9272 | } | |
a8eeb04a | 9273 | if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu)) |
4731d4c7 | 9274 | kvm_mmu_sync_roots(vcpu); |
727a7e27 PB |
9275 | if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu)) |
9276 | kvm_mmu_load_pgd(vcpu); | |
eeeb4f67 | 9277 | if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) { |
7780938c | 9278 | kvm_vcpu_flush_tlb_all(vcpu); |
eeeb4f67 SC |
9279 | |
9280 | /* Flushing all ASIDs flushes the current ASID... */ | |
9281 | kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu); | |
9282 | } | |
9283 | if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu)) | |
9284 | kvm_vcpu_flush_tlb_current(vcpu); | |
0baedd79 VK |
9285 | if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu)) |
9286 | kvm_vcpu_flush_tlb_guest(vcpu); | |
eeeb4f67 | 9287 | |
a8eeb04a | 9288 | if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) { |
851ba692 | 9289 | vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS; |
b93463aa AK |
9290 | r = 0; |
9291 | goto out; | |
9292 | } | |
a8eeb04a | 9293 | if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) { |
cb6a32c2 SC |
9294 | if (is_guest_mode(vcpu)) { |
9295 | kvm_x86_ops.nested_ops->triple_fault(vcpu); | |
9296 | } else { | |
9297 | vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN; | |
9298 | vcpu->mmio_needed = 0; | |
9299 | r = 0; | |
9300 | goto out; | |
9301 | } | |
71c4dfaf | 9302 | } |
af585b92 GN |
9303 | if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) { |
9304 | /* Page is swapped out. Do synthetic halt */ | |
9305 | vcpu->arch.apf.halted = true; | |
9306 | r = 1; | |
9307 | goto out; | |
9308 | } | |
c9aaa895 GC |
9309 | if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu)) |
9310 | record_steal_time(vcpu); | |
64d60670 PB |
9311 | if (kvm_check_request(KVM_REQ_SMI, vcpu)) |
9312 | process_smi(vcpu); | |
7460fb4a AK |
9313 | if (kvm_check_request(KVM_REQ_NMI, vcpu)) |
9314 | process_nmi(vcpu); | |
f5132b01 | 9315 | if (kvm_check_request(KVM_REQ_PMU, vcpu)) |
c6702c9d | 9316 | kvm_pmu_handle_event(vcpu); |
f5132b01 | 9317 | if (kvm_check_request(KVM_REQ_PMI, vcpu)) |
c6702c9d | 9318 | kvm_pmu_deliver_pmi(vcpu); |
7543a635 SR |
9319 | if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) { |
9320 | BUG_ON(vcpu->arch.pending_ioapic_eoi > 255); | |
9321 | if (test_bit(vcpu->arch.pending_ioapic_eoi, | |
6308630b | 9322 | vcpu->arch.ioapic_handled_vectors)) { |
7543a635 SR |
9323 | vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI; |
9324 | vcpu->run->eoi.vector = | |
9325 | vcpu->arch.pending_ioapic_eoi; | |
9326 | r = 0; | |
9327 | goto out; | |
9328 | } | |
9329 | } | |
3d81bc7e YZ |
9330 | if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu)) |
9331 | vcpu_scan_ioapic(vcpu); | |
e40ff1d6 LA |
9332 | if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu)) |
9333 | vcpu_load_eoi_exitmap(vcpu); | |
4256f43f TC |
9334 | if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu)) |
9335 | kvm_vcpu_reload_apic_access_page(vcpu); | |
2ce79189 AS |
9336 | if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) { |
9337 | vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT; | |
9338 | vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH; | |
9339 | r = 0; | |
9340 | goto out; | |
9341 | } | |
e516cebb AS |
9342 | if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) { |
9343 | vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT; | |
9344 | vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET; | |
9345 | r = 0; | |
9346 | goto out; | |
9347 | } | |
db397571 | 9348 | if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) { |
9ff5e030 VK |
9349 | struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu); |
9350 | ||
db397571 | 9351 | vcpu->run->exit_reason = KVM_EXIT_HYPERV; |
9ff5e030 | 9352 | vcpu->run->hyperv = hv_vcpu->exit; |
db397571 AS |
9353 | r = 0; |
9354 | goto out; | |
9355 | } | |
f3b138c5 AS |
9356 | |
9357 | /* | |
9358 | * KVM_REQ_HV_STIMER has to be processed after | |
9359 | * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers | |
9360 | * depend on the guest clock being up-to-date | |
9361 | */ | |
1f4b34f8 AS |
9362 | if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu)) |
9363 | kvm_hv_process_stimers(vcpu); | |
8df14af4 SS |
9364 | if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu)) |
9365 | kvm_vcpu_update_apicv(vcpu); | |
557a961a VK |
9366 | if (kvm_check_request(KVM_REQ_APF_READY, vcpu)) |
9367 | kvm_check_async_pf_completion(vcpu); | |
1a155254 | 9368 | if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu)) |
b3646477 | 9369 | static_call(kvm_x86_msr_filter_changed)(vcpu); |
a85863c2 MS |
9370 | |
9371 | if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu)) | |
9372 | static_call(kvm_x86_update_cpu_dirty_logging)(vcpu); | |
2f52d58c | 9373 | } |
b93463aa | 9374 | |
40da8ccd DW |
9375 | if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win || |
9376 | kvm_xen_has_interrupt(vcpu)) { | |
0f1e261e | 9377 | ++vcpu->stat.req_event; |
4fe09bcf JM |
9378 | r = kvm_apic_accept_events(vcpu); |
9379 | if (r < 0) { | |
9380 | r = 0; | |
9381 | goto out; | |
9382 | } | |
66450a21 JK |
9383 | if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) { |
9384 | r = 1; | |
9385 | goto out; | |
9386 | } | |
9387 | ||
a5f6909a JM |
9388 | r = inject_pending_event(vcpu, &req_immediate_exit); |
9389 | if (r < 0) { | |
9390 | r = 0; | |
9391 | goto out; | |
9392 | } | |
c9d40913 | 9393 | if (req_int_win) |
b3646477 | 9394 | static_call(kvm_x86_enable_irq_window)(vcpu); |
b463a6f7 AK |
9395 | |
9396 | if (kvm_lapic_enabled(vcpu)) { | |
9397 | update_cr8_intercept(vcpu); | |
9398 | kvm_lapic_sync_to_vapic(vcpu); | |
9399 | } | |
9400 | } | |
9401 | ||
d8368af8 AK |
9402 | r = kvm_mmu_reload(vcpu); |
9403 | if (unlikely(r)) { | |
d905c069 | 9404 | goto cancel_injection; |
d8368af8 AK |
9405 | } |
9406 | ||
b6c7a5dc HB |
9407 | preempt_disable(); |
9408 | ||
b3646477 | 9409 | static_call(kvm_x86_prepare_guest_switch)(vcpu); |
b95234c8 PB |
9410 | |
9411 | /* | |
9412 | * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt | |
9413 | * IPI are then delayed after guest entry, which ensures that they | |
9414 | * result in virtual interrupt delivery. | |
9415 | */ | |
9416 | local_irq_disable(); | |
6b7e2d09 XG |
9417 | vcpu->mode = IN_GUEST_MODE; |
9418 | ||
01b71917 MT |
9419 | srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); |
9420 | ||
0f127d12 | 9421 | /* |
b95234c8 | 9422 | * 1) We should set ->mode before checking ->requests. Please see |
cde9af6e | 9423 | * the comment in kvm_vcpu_exiting_guest_mode(). |
b95234c8 | 9424 | * |
81b01667 | 9425 | * 2) For APICv, we should set ->mode before checking PID.ON. This |
b95234c8 PB |
9426 | * pairs with the memory barrier implicit in pi_test_and_set_on |
9427 | * (see vmx_deliver_posted_interrupt). | |
9428 | * | |
9429 | * 3) This also orders the write to mode from any reads to the page | |
9430 | * tables done while the VCPU is running. Please see the comment | |
9431 | * in kvm_flush_remote_tlbs. | |
6b7e2d09 | 9432 | */ |
01b71917 | 9433 | smp_mb__after_srcu_read_unlock(); |
b6c7a5dc | 9434 | |
b95234c8 PB |
9435 | /* |
9436 | * This handles the case where a posted interrupt was | |
9437 | * notified with kvm_vcpu_kick. | |
9438 | */ | |
fa59cc00 | 9439 | if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active) |
b3646477 | 9440 | static_call(kvm_x86_sync_pir_to_irr)(vcpu); |
32f88400 | 9441 | |
5a9f5443 | 9442 | if (kvm_vcpu_exit_request(vcpu)) { |
6b7e2d09 | 9443 | vcpu->mode = OUTSIDE_GUEST_MODE; |
d94e1dc9 | 9444 | smp_wmb(); |
6c142801 AK |
9445 | local_irq_enable(); |
9446 | preempt_enable(); | |
01b71917 | 9447 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); |
6c142801 | 9448 | r = 1; |
d905c069 | 9449 | goto cancel_injection; |
6c142801 AK |
9450 | } |
9451 | ||
c43203ca PB |
9452 | if (req_immediate_exit) { |
9453 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
b3646477 | 9454 | static_call(kvm_x86_request_immediate_exit)(vcpu); |
c43203ca | 9455 | } |
d6185f20 | 9456 | |
2620fe26 SC |
9457 | fpregs_assert_state_consistent(); |
9458 | if (test_thread_flag(TIF_NEED_FPU_LOAD)) | |
9459 | switch_fpu_return(); | |
5f409e20 | 9460 | |
42dbaa5a | 9461 | if (unlikely(vcpu->arch.switch_db_regs)) { |
42dbaa5a JK |
9462 | set_debugreg(0, 7); |
9463 | set_debugreg(vcpu->arch.eff_db[0], 0); | |
9464 | set_debugreg(vcpu->arch.eff_db[1], 1); | |
9465 | set_debugreg(vcpu->arch.eff_db[2], 2); | |
9466 | set_debugreg(vcpu->arch.eff_db[3], 3); | |
c77fb5fe | 9467 | set_debugreg(vcpu->arch.dr6, 6); |
ae561ede | 9468 | vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD; |
42dbaa5a | 9469 | } |
b6c7a5dc | 9470 | |
d89d04ab PB |
9471 | for (;;) { |
9472 | exit_fastpath = static_call(kvm_x86_run)(vcpu); | |
9473 | if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST)) | |
9474 | break; | |
9475 | ||
9476 | if (unlikely(kvm_vcpu_exit_request(vcpu))) { | |
9477 | exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED; | |
9478 | break; | |
9479 | } | |
9480 | ||
9481 | if (vcpu->arch.apicv_active) | |
9482 | static_call(kvm_x86_sync_pir_to_irr)(vcpu); | |
9483 | } | |
b6c7a5dc | 9484 | |
c77fb5fe PB |
9485 | /* |
9486 | * Do this here before restoring debug registers on the host. And | |
9487 | * since we do this before handling the vmexit, a DR access vmexit | |
9488 | * can (a) read the correct value of the debug registers, (b) set | |
9489 | * KVM_DEBUGREG_WONT_EXIT again. | |
9490 | */ | |
9491 | if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) { | |
c77fb5fe | 9492 | WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP); |
b3646477 | 9493 | static_call(kvm_x86_sync_dirty_debug_regs)(vcpu); |
70e4da7a | 9494 | kvm_update_dr0123(vcpu); |
70e4da7a PB |
9495 | kvm_update_dr7(vcpu); |
9496 | vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD; | |
c77fb5fe PB |
9497 | } |
9498 | ||
24f1e32c FW |
9499 | /* |
9500 | * If the guest has used debug registers, at least dr7 | |
9501 | * will be disabled while returning to the host. | |
9502 | * If we don't have active breakpoints in the host, we don't | |
9503 | * care about the messed up debug address registers. But if | |
9504 | * we have some of them active, restore the old state. | |
9505 | */ | |
59d8eb53 | 9506 | if (hw_breakpoint_active()) |
24f1e32c | 9507 | hw_breakpoint_restore(); |
42dbaa5a | 9508 | |
c967118d | 9509 | vcpu->arch.last_vmentry_cpu = vcpu->cpu; |
4ba76538 | 9510 | vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc()); |
1d5f066e | 9511 | |
6b7e2d09 | 9512 | vcpu->mode = OUTSIDE_GUEST_MODE; |
d94e1dc9 | 9513 | smp_wmb(); |
a547c6db | 9514 | |
b3646477 | 9515 | static_call(kvm_x86_handle_exit_irqoff)(vcpu); |
b6c7a5dc | 9516 | |
d7a08882 SC |
9517 | /* |
9518 | * Consume any pending interrupts, including the possible source of | |
9519 | * VM-Exit on SVM and any ticks that occur between VM-Exit and now. | |
9520 | * An instruction is required after local_irq_enable() to fully unblock | |
9521 | * interrupts on processors that implement an interrupt shadow, the | |
9522 | * stat.exits increment will do nicely. | |
9523 | */ | |
9524 | kvm_before_interrupt(vcpu); | |
9525 | local_irq_enable(); | |
b6c7a5dc | 9526 | ++vcpu->stat.exits; |
d7a08882 SC |
9527 | local_irq_disable(); |
9528 | kvm_after_interrupt(vcpu); | |
b6c7a5dc | 9529 | |
16045714 WL |
9530 | /* |
9531 | * Wait until after servicing IRQs to account guest time so that any | |
9532 | * ticks that occurred while running the guest are properly accounted | |
9533 | * to the guest. Waiting until IRQs are enabled degrades the accuracy | |
9534 | * of accounting via context tracking, but the loss of accuracy is | |
9535 | * acceptable for all known use cases. | |
9536 | */ | |
9537 | vtime_account_guest_exit(); | |
9538 | ||
ec0671d5 WL |
9539 | if (lapic_in_kernel(vcpu)) { |
9540 | s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta; | |
9541 | if (delta != S64_MIN) { | |
9542 | trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta); | |
9543 | vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN; | |
9544 | } | |
9545 | } | |
b6c7a5dc | 9546 | |
f2485b3e | 9547 | local_irq_enable(); |
b6c7a5dc HB |
9548 | preempt_enable(); |
9549 | ||
f656ce01 | 9550 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); |
3200f405 | 9551 | |
b6c7a5dc HB |
9552 | /* |
9553 | * Profile KVM exit RIPs: | |
9554 | */ | |
9555 | if (unlikely(prof_on == KVM_PROFILING)) { | |
5fdbf976 MT |
9556 | unsigned long rip = kvm_rip_read(vcpu); |
9557 | profile_hit(KVM_PROFILING, (void *)rip); | |
b6c7a5dc HB |
9558 | } |
9559 | ||
cc578287 ZA |
9560 | if (unlikely(vcpu->arch.tsc_always_catchup)) |
9561 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); | |
298101da | 9562 | |
5cfb1d5a MT |
9563 | if (vcpu->arch.apic_attention) |
9564 | kvm_lapic_sync_from_vapic(vcpu); | |
b93463aa | 9565 | |
b3646477 | 9566 | r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath); |
d905c069 MT |
9567 | return r; |
9568 | ||
9569 | cancel_injection: | |
8081ad06 SC |
9570 | if (req_immediate_exit) |
9571 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
b3646477 | 9572 | static_call(kvm_x86_cancel_injection)(vcpu); |
ae7a2a3f MT |
9573 | if (unlikely(vcpu->arch.apic_attention)) |
9574 | kvm_lapic_sync_from_vapic(vcpu); | |
d7690175 MT |
9575 | out: |
9576 | return r; | |
9577 | } | |
b6c7a5dc | 9578 | |
362c698f PB |
9579 | static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu) |
9580 | { | |
bf9f6ac8 | 9581 | if (!kvm_arch_vcpu_runnable(vcpu) && |
b3646477 | 9582 | (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) { |
9c8fd1ba PB |
9583 | srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); |
9584 | kvm_vcpu_block(vcpu); | |
9585 | vcpu->srcu_idx = srcu_read_lock(&kvm->srcu); | |
bf9f6ac8 | 9586 | |
afaf0b2f | 9587 | if (kvm_x86_ops.post_block) |
b3646477 | 9588 | static_call(kvm_x86_post_block)(vcpu); |
bf9f6ac8 | 9589 | |
9c8fd1ba PB |
9590 | if (!kvm_check_request(KVM_REQ_UNHALT, vcpu)) |
9591 | return 1; | |
9592 | } | |
362c698f | 9593 | |
4fe09bcf JM |
9594 | if (kvm_apic_accept_events(vcpu) < 0) |
9595 | return 0; | |
362c698f PB |
9596 | switch(vcpu->arch.mp_state) { |
9597 | case KVM_MP_STATE_HALTED: | |
647daca2 | 9598 | case KVM_MP_STATE_AP_RESET_HOLD: |
362c698f PB |
9599 | vcpu->arch.pv.pv_unhalted = false; |
9600 | vcpu->arch.mp_state = | |
9601 | KVM_MP_STATE_RUNNABLE; | |
df561f66 | 9602 | fallthrough; |
362c698f PB |
9603 | case KVM_MP_STATE_RUNNABLE: |
9604 | vcpu->arch.apf.halted = false; | |
9605 | break; | |
9606 | case KVM_MP_STATE_INIT_RECEIVED: | |
9607 | break; | |
9608 | default: | |
9609 | return -EINTR; | |
362c698f PB |
9610 | } |
9611 | return 1; | |
9612 | } | |
09cec754 | 9613 | |
5d9bc648 PB |
9614 | static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu) |
9615 | { | |
56083bdf | 9616 | if (is_guest_mode(vcpu)) |
cb6a32c2 | 9617 | kvm_check_nested_events(vcpu); |
0ad3bed6 | 9618 | |
5d9bc648 PB |
9619 | return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE && |
9620 | !vcpu->arch.apf.halted); | |
9621 | } | |
9622 | ||
362c698f | 9623 | static int vcpu_run(struct kvm_vcpu *vcpu) |
d7690175 MT |
9624 | { |
9625 | int r; | |
f656ce01 | 9626 | struct kvm *kvm = vcpu->kvm; |
d7690175 | 9627 | |
f656ce01 | 9628 | vcpu->srcu_idx = srcu_read_lock(&kvm->srcu); |
c595ceee | 9629 | vcpu->arch.l1tf_flush_l1d = true; |
d7690175 | 9630 | |
362c698f | 9631 | for (;;) { |
58f800d5 | 9632 | if (kvm_vcpu_running(vcpu)) { |
851ba692 | 9633 | r = vcpu_enter_guest(vcpu); |
bf9f6ac8 | 9634 | } else { |
362c698f | 9635 | r = vcpu_block(kvm, vcpu); |
bf9f6ac8 FW |
9636 | } |
9637 | ||
09cec754 GN |
9638 | if (r <= 0) |
9639 | break; | |
9640 | ||
084071d5 | 9641 | kvm_clear_request(KVM_REQ_UNBLOCK, vcpu); |
09cec754 GN |
9642 | if (kvm_cpu_has_pending_timer(vcpu)) |
9643 | kvm_inject_pending_timer_irqs(vcpu); | |
9644 | ||
782d422b MG |
9645 | if (dm_request_for_irq_injection(vcpu) && |
9646 | kvm_vcpu_ready_for_interrupt_injection(vcpu)) { | |
4ca7dd8c PB |
9647 | r = 0; |
9648 | vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN; | |
09cec754 | 9649 | ++vcpu->stat.request_irq_exits; |
362c698f | 9650 | break; |
09cec754 | 9651 | } |
af585b92 | 9652 | |
f3020b88 | 9653 | if (__xfer_to_guest_mode_work_pending()) { |
f656ce01 | 9654 | srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); |
72c3c0fe TG |
9655 | r = xfer_to_guest_mode_handle_work(vcpu); |
9656 | if (r) | |
9657 | return r; | |
f656ce01 | 9658 | vcpu->srcu_idx = srcu_read_lock(&kvm->srcu); |
d7690175 | 9659 | } |
b6c7a5dc HB |
9660 | } |
9661 | ||
f656ce01 | 9662 | srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); |
b6c7a5dc HB |
9663 | |
9664 | return r; | |
9665 | } | |
9666 | ||
716d51ab GN |
9667 | static inline int complete_emulated_io(struct kvm_vcpu *vcpu) |
9668 | { | |
9669 | int r; | |
60fc3d02 | 9670 | |
716d51ab | 9671 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); |
0ce97a2b | 9672 | r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE); |
716d51ab | 9673 | srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); |
60fc3d02 | 9674 | return r; |
716d51ab GN |
9675 | } |
9676 | ||
9677 | static int complete_emulated_pio(struct kvm_vcpu *vcpu) | |
9678 | { | |
9679 | BUG_ON(!vcpu->arch.pio.count); | |
9680 | ||
9681 | return complete_emulated_io(vcpu); | |
9682 | } | |
9683 | ||
f78146b0 AK |
9684 | /* |
9685 | * Implements the following, as a state machine: | |
9686 | * | |
9687 | * read: | |
9688 | * for each fragment | |
87da7e66 XG |
9689 | * for each mmio piece in the fragment |
9690 | * write gpa, len | |
9691 | * exit | |
9692 | * copy data | |
f78146b0 AK |
9693 | * execute insn |
9694 | * | |
9695 | * write: | |
9696 | * for each fragment | |
87da7e66 XG |
9697 | * for each mmio piece in the fragment |
9698 | * write gpa, len | |
9699 | * copy data | |
9700 | * exit | |
f78146b0 | 9701 | */ |
716d51ab | 9702 | static int complete_emulated_mmio(struct kvm_vcpu *vcpu) |
5287f194 AK |
9703 | { |
9704 | struct kvm_run *run = vcpu->run; | |
f78146b0 | 9705 | struct kvm_mmio_fragment *frag; |
87da7e66 | 9706 | unsigned len; |
5287f194 | 9707 | |
716d51ab | 9708 | BUG_ON(!vcpu->mmio_needed); |
5287f194 | 9709 | |
716d51ab | 9710 | /* Complete previous fragment */ |
87da7e66 XG |
9711 | frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment]; |
9712 | len = min(8u, frag->len); | |
716d51ab | 9713 | if (!vcpu->mmio_is_write) |
87da7e66 XG |
9714 | memcpy(frag->data, run->mmio.data, len); |
9715 | ||
9716 | if (frag->len <= 8) { | |
9717 | /* Switch to the next fragment. */ | |
9718 | frag++; | |
9719 | vcpu->mmio_cur_fragment++; | |
9720 | } else { | |
9721 | /* Go forward to the next mmio piece. */ | |
9722 | frag->data += len; | |
9723 | frag->gpa += len; | |
9724 | frag->len -= len; | |
9725 | } | |
9726 | ||
a08d3b3b | 9727 | if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) { |
716d51ab | 9728 | vcpu->mmio_needed = 0; |
0912c977 PB |
9729 | |
9730 | /* FIXME: return into emulator if single-stepping. */ | |
cef4dea0 | 9731 | if (vcpu->mmio_is_write) |
716d51ab GN |
9732 | return 1; |
9733 | vcpu->mmio_read_completed = 1; | |
9734 | return complete_emulated_io(vcpu); | |
9735 | } | |
87da7e66 | 9736 | |
716d51ab GN |
9737 | run->exit_reason = KVM_EXIT_MMIO; |
9738 | run->mmio.phys_addr = frag->gpa; | |
9739 | if (vcpu->mmio_is_write) | |
87da7e66 XG |
9740 | memcpy(run->mmio.data, frag->data, min(8u, frag->len)); |
9741 | run->mmio.len = min(8u, frag->len); | |
716d51ab GN |
9742 | run->mmio.is_write = vcpu->mmio_is_write; |
9743 | vcpu->arch.complete_userspace_io = complete_emulated_mmio; | |
9744 | return 0; | |
5287f194 AK |
9745 | } |
9746 | ||
c9aef3b8 SC |
9747 | static void kvm_save_current_fpu(struct fpu *fpu) |
9748 | { | |
9749 | /* | |
9750 | * If the target FPU state is not resident in the CPU registers, just | |
9751 | * memcpy() from current, else save CPU state directly to the target. | |
9752 | */ | |
9753 | if (test_thread_flag(TIF_NEED_FPU_LOAD)) | |
9754 | memcpy(&fpu->state, ¤t->thread.fpu.state, | |
9755 | fpu_kernel_xstate_size); | |
9756 | else | |
9757 | copy_fpregs_to_fpstate(fpu); | |
9758 | } | |
9759 | ||
822f312d SAS |
9760 | /* Swap (qemu) user FPU context for the guest FPU context. */ |
9761 | static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu) | |
9762 | { | |
5f409e20 RR |
9763 | fpregs_lock(); |
9764 | ||
c9aef3b8 SC |
9765 | kvm_save_current_fpu(vcpu->arch.user_fpu); |
9766 | ||
ed02b213 TL |
9767 | /* |
9768 | * Guests with protected state can't have it set by the hypervisor, | |
9769 | * so skip trying to set it. | |
9770 | */ | |
9771 | if (vcpu->arch.guest_fpu) | |
9772 | /* PKRU is separately restored in kvm_x86_ops.run. */ | |
9773 | __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state, | |
9774 | ~XFEATURE_MASK_PKRU); | |
5f409e20 RR |
9775 | |
9776 | fpregs_mark_activate(); | |
9777 | fpregs_unlock(); | |
9778 | ||
822f312d SAS |
9779 | trace_kvm_fpu(1); |
9780 | } | |
9781 | ||
9782 | /* When vcpu_run ends, restore user space FPU context. */ | |
9783 | static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu) | |
9784 | { | |
5f409e20 RR |
9785 | fpregs_lock(); |
9786 | ||
ed02b213 TL |
9787 | /* |
9788 | * Guests with protected state can't have it read by the hypervisor, | |
9789 | * so skip trying to save it. | |
9790 | */ | |
9791 | if (vcpu->arch.guest_fpu) | |
9792 | kvm_save_current_fpu(vcpu->arch.guest_fpu); | |
c9aef3b8 | 9793 | |
d9a710e5 | 9794 | copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state); |
5f409e20 RR |
9795 | |
9796 | fpregs_mark_activate(); | |
9797 | fpregs_unlock(); | |
9798 | ||
822f312d SAS |
9799 | ++vcpu->stat.fpu_reload; |
9800 | trace_kvm_fpu(0); | |
9801 | } | |
9802 | ||
1b94f6f8 | 9803 | int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu) |
b6c7a5dc | 9804 | { |
1b94f6f8 | 9805 | struct kvm_run *kvm_run = vcpu->run; |
b6c7a5dc | 9806 | int r; |
b6c7a5dc | 9807 | |
accb757d | 9808 | vcpu_load(vcpu); |
20b7035c | 9809 | kvm_sigset_activate(vcpu); |
15aad3be | 9810 | kvm_run->flags = 0; |
5663d8f9 PX |
9811 | kvm_load_guest_fpu(vcpu); |
9812 | ||
a4535290 | 9813 | if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) { |
2f173d26 JS |
9814 | if (kvm_run->immediate_exit) { |
9815 | r = -EINTR; | |
9816 | goto out; | |
9817 | } | |
b6c7a5dc | 9818 | kvm_vcpu_block(vcpu); |
4fe09bcf JM |
9819 | if (kvm_apic_accept_events(vcpu) < 0) { |
9820 | r = 0; | |
9821 | goto out; | |
9822 | } | |
72875d8a | 9823 | kvm_clear_request(KVM_REQ_UNHALT, vcpu); |
ac9f6dc0 | 9824 | r = -EAGAIN; |
a0595000 JS |
9825 | if (signal_pending(current)) { |
9826 | r = -EINTR; | |
1b94f6f8 | 9827 | kvm_run->exit_reason = KVM_EXIT_INTR; |
a0595000 JS |
9828 | ++vcpu->stat.signal_exits; |
9829 | } | |
ac9f6dc0 | 9830 | goto out; |
b6c7a5dc HB |
9831 | } |
9832 | ||
1b94f6f8 | 9833 | if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) { |
01643c51 KH |
9834 | r = -EINVAL; |
9835 | goto out; | |
9836 | } | |
9837 | ||
1b94f6f8 | 9838 | if (kvm_run->kvm_dirty_regs) { |
01643c51 KH |
9839 | r = sync_regs(vcpu); |
9840 | if (r != 0) | |
9841 | goto out; | |
9842 | } | |
9843 | ||
b6c7a5dc | 9844 | /* re-sync apic's tpr */ |
35754c98 | 9845 | if (!lapic_in_kernel(vcpu)) { |
eea1cff9 AP |
9846 | if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) { |
9847 | r = -EINVAL; | |
9848 | goto out; | |
9849 | } | |
9850 | } | |
b6c7a5dc | 9851 | |
716d51ab GN |
9852 | if (unlikely(vcpu->arch.complete_userspace_io)) { |
9853 | int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io; | |
9854 | vcpu->arch.complete_userspace_io = NULL; | |
9855 | r = cui(vcpu); | |
9856 | if (r <= 0) | |
5663d8f9 | 9857 | goto out; |
716d51ab GN |
9858 | } else |
9859 | WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed); | |
5287f194 | 9860 | |
460df4c1 PB |
9861 | if (kvm_run->immediate_exit) |
9862 | r = -EINTR; | |
9863 | else | |
9864 | r = vcpu_run(vcpu); | |
b6c7a5dc HB |
9865 | |
9866 | out: | |
5663d8f9 | 9867 | kvm_put_guest_fpu(vcpu); |
1b94f6f8 | 9868 | if (kvm_run->kvm_valid_regs) |
01643c51 | 9869 | store_regs(vcpu); |
f1d86e46 | 9870 | post_kvm_run_save(vcpu); |
20b7035c | 9871 | kvm_sigset_deactivate(vcpu); |
b6c7a5dc | 9872 | |
accb757d | 9873 | vcpu_put(vcpu); |
b6c7a5dc HB |
9874 | return r; |
9875 | } | |
9876 | ||
01643c51 | 9877 | static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) |
b6c7a5dc | 9878 | { |
7ae441ea GN |
9879 | if (vcpu->arch.emulate_regs_need_sync_to_vcpu) { |
9880 | /* | |
9881 | * We are here if userspace calls get_regs() in the middle of | |
9882 | * instruction emulation. Registers state needs to be copied | |
4a969980 | 9883 | * back from emulation context to vcpu. Userspace shouldn't do |
7ae441ea GN |
9884 | * that usually, but some bad designed PV devices (vmware |
9885 | * backdoor interface) need this to work | |
9886 | */ | |
c9b8b07c | 9887 | emulator_writeback_register_cache(vcpu->arch.emulate_ctxt); |
7ae441ea GN |
9888 | vcpu->arch.emulate_regs_need_sync_to_vcpu = false; |
9889 | } | |
de3cd117 SC |
9890 | regs->rax = kvm_rax_read(vcpu); |
9891 | regs->rbx = kvm_rbx_read(vcpu); | |
9892 | regs->rcx = kvm_rcx_read(vcpu); | |
9893 | regs->rdx = kvm_rdx_read(vcpu); | |
9894 | regs->rsi = kvm_rsi_read(vcpu); | |
9895 | regs->rdi = kvm_rdi_read(vcpu); | |
e9c16c78 | 9896 | regs->rsp = kvm_rsp_read(vcpu); |
de3cd117 | 9897 | regs->rbp = kvm_rbp_read(vcpu); |
b6c7a5dc | 9898 | #ifdef CONFIG_X86_64 |
de3cd117 SC |
9899 | regs->r8 = kvm_r8_read(vcpu); |
9900 | regs->r9 = kvm_r9_read(vcpu); | |
9901 | regs->r10 = kvm_r10_read(vcpu); | |
9902 | regs->r11 = kvm_r11_read(vcpu); | |
9903 | regs->r12 = kvm_r12_read(vcpu); | |
9904 | regs->r13 = kvm_r13_read(vcpu); | |
9905 | regs->r14 = kvm_r14_read(vcpu); | |
9906 | regs->r15 = kvm_r15_read(vcpu); | |
b6c7a5dc HB |
9907 | #endif |
9908 | ||
5fdbf976 | 9909 | regs->rip = kvm_rip_read(vcpu); |
91586a3b | 9910 | regs->rflags = kvm_get_rflags(vcpu); |
01643c51 | 9911 | } |
b6c7a5dc | 9912 | |
01643c51 KH |
9913 | int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) |
9914 | { | |
9915 | vcpu_load(vcpu); | |
9916 | __get_regs(vcpu, regs); | |
1fc9b76b | 9917 | vcpu_put(vcpu); |
b6c7a5dc HB |
9918 | return 0; |
9919 | } | |
9920 | ||
01643c51 | 9921 | static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) |
b6c7a5dc | 9922 | { |
7ae441ea GN |
9923 | vcpu->arch.emulate_regs_need_sync_from_vcpu = true; |
9924 | vcpu->arch.emulate_regs_need_sync_to_vcpu = false; | |
9925 | ||
de3cd117 SC |
9926 | kvm_rax_write(vcpu, regs->rax); |
9927 | kvm_rbx_write(vcpu, regs->rbx); | |
9928 | kvm_rcx_write(vcpu, regs->rcx); | |
9929 | kvm_rdx_write(vcpu, regs->rdx); | |
9930 | kvm_rsi_write(vcpu, regs->rsi); | |
9931 | kvm_rdi_write(vcpu, regs->rdi); | |
e9c16c78 | 9932 | kvm_rsp_write(vcpu, regs->rsp); |
de3cd117 | 9933 | kvm_rbp_write(vcpu, regs->rbp); |
b6c7a5dc | 9934 | #ifdef CONFIG_X86_64 |
de3cd117 SC |
9935 | kvm_r8_write(vcpu, regs->r8); |
9936 | kvm_r9_write(vcpu, regs->r9); | |
9937 | kvm_r10_write(vcpu, regs->r10); | |
9938 | kvm_r11_write(vcpu, regs->r11); | |
9939 | kvm_r12_write(vcpu, regs->r12); | |
9940 | kvm_r13_write(vcpu, regs->r13); | |
9941 | kvm_r14_write(vcpu, regs->r14); | |
9942 | kvm_r15_write(vcpu, regs->r15); | |
b6c7a5dc HB |
9943 | #endif |
9944 | ||
5fdbf976 | 9945 | kvm_rip_write(vcpu, regs->rip); |
d73235d1 | 9946 | kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED); |
b6c7a5dc | 9947 | |
b4f14abd JK |
9948 | vcpu->arch.exception.pending = false; |
9949 | ||
3842d135 | 9950 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
01643c51 | 9951 | } |
3842d135 | 9952 | |
01643c51 KH |
9953 | int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) |
9954 | { | |
9955 | vcpu_load(vcpu); | |
9956 | __set_regs(vcpu, regs); | |
875656fe | 9957 | vcpu_put(vcpu); |
b6c7a5dc HB |
9958 | return 0; |
9959 | } | |
9960 | ||
b6c7a5dc HB |
9961 | void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l) |
9962 | { | |
9963 | struct kvm_segment cs; | |
9964 | ||
3e6e0aab | 9965 | kvm_get_segment(vcpu, &cs, VCPU_SREG_CS); |
b6c7a5dc HB |
9966 | *db = cs.db; |
9967 | *l = cs.l; | |
9968 | } | |
9969 | EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits); | |
9970 | ||
01643c51 | 9971 | static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs) |
b6c7a5dc | 9972 | { |
89a27f4d | 9973 | struct desc_ptr dt; |
b6c7a5dc | 9974 | |
5265713a TL |
9975 | if (vcpu->arch.guest_state_protected) |
9976 | goto skip_protected_regs; | |
9977 | ||
3e6e0aab GT |
9978 | kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS); |
9979 | kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS); | |
9980 | kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES); | |
9981 | kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS); | |
9982 | kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS); | |
9983 | kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS); | |
b6c7a5dc | 9984 | |
3e6e0aab GT |
9985 | kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR); |
9986 | kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR); | |
b6c7a5dc | 9987 | |
b3646477 | 9988 | static_call(kvm_x86_get_idt)(vcpu, &dt); |
89a27f4d GN |
9989 | sregs->idt.limit = dt.size; |
9990 | sregs->idt.base = dt.address; | |
b3646477 | 9991 | static_call(kvm_x86_get_gdt)(vcpu, &dt); |
89a27f4d GN |
9992 | sregs->gdt.limit = dt.size; |
9993 | sregs->gdt.base = dt.address; | |
b6c7a5dc | 9994 | |
ad312c7c | 9995 | sregs->cr2 = vcpu->arch.cr2; |
9f8fe504 | 9996 | sregs->cr3 = kvm_read_cr3(vcpu); |
5265713a TL |
9997 | |
9998 | skip_protected_regs: | |
9999 | sregs->cr0 = kvm_read_cr0(vcpu); | |
fc78f519 | 10000 | sregs->cr4 = kvm_read_cr4(vcpu); |
2d3ad1f4 | 10001 | sregs->cr8 = kvm_get_cr8(vcpu); |
f6801dff | 10002 | sregs->efer = vcpu->arch.efer; |
b6c7a5dc HB |
10003 | sregs->apic_base = kvm_get_apic_base(vcpu); |
10004 | ||
0e96f31e | 10005 | memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap)); |
b6c7a5dc | 10006 | |
04140b41 | 10007 | if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft) |
14d0bc1f GN |
10008 | set_bit(vcpu->arch.interrupt.nr, |
10009 | (unsigned long *)sregs->interrupt_bitmap); | |
01643c51 | 10010 | } |
16d7a191 | 10011 | |
01643c51 KH |
10012 | int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu, |
10013 | struct kvm_sregs *sregs) | |
10014 | { | |
10015 | vcpu_load(vcpu); | |
10016 | __get_sregs(vcpu, sregs); | |
bcdec41c | 10017 | vcpu_put(vcpu); |
b6c7a5dc HB |
10018 | return 0; |
10019 | } | |
10020 | ||
62d9f0db MT |
10021 | int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, |
10022 | struct kvm_mp_state *mp_state) | |
10023 | { | |
4fe09bcf JM |
10024 | int r; |
10025 | ||
fd232561 | 10026 | vcpu_load(vcpu); |
f958bd23 SC |
10027 | if (kvm_mpx_supported()) |
10028 | kvm_load_guest_fpu(vcpu); | |
fd232561 | 10029 | |
4fe09bcf JM |
10030 | r = kvm_apic_accept_events(vcpu); |
10031 | if (r < 0) | |
10032 | goto out; | |
10033 | r = 0; | |
10034 | ||
647daca2 TL |
10035 | if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED || |
10036 | vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) && | |
10037 | vcpu->arch.pv.pv_unhalted) | |
6aef266c SV |
10038 | mp_state->mp_state = KVM_MP_STATE_RUNNABLE; |
10039 | else | |
10040 | mp_state->mp_state = vcpu->arch.mp_state; | |
10041 | ||
4fe09bcf | 10042 | out: |
f958bd23 SC |
10043 | if (kvm_mpx_supported()) |
10044 | kvm_put_guest_fpu(vcpu); | |
fd232561 | 10045 | vcpu_put(vcpu); |
4fe09bcf | 10046 | return r; |
62d9f0db MT |
10047 | } |
10048 | ||
10049 | int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, | |
10050 | struct kvm_mp_state *mp_state) | |
10051 | { | |
e83dff5e CD |
10052 | int ret = -EINVAL; |
10053 | ||
10054 | vcpu_load(vcpu); | |
10055 | ||
bce87cce | 10056 | if (!lapic_in_kernel(vcpu) && |
66450a21 | 10057 | mp_state->mp_state != KVM_MP_STATE_RUNNABLE) |
e83dff5e | 10058 | goto out; |
66450a21 | 10059 | |
27cbe7d6 LA |
10060 | /* |
10061 | * KVM_MP_STATE_INIT_RECEIVED means the processor is in | |
10062 | * INIT state; latched init should be reported using | |
10063 | * KVM_SET_VCPU_EVENTS, so reject it here. | |
10064 | */ | |
10065 | if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) && | |
28bf2888 DH |
10066 | (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED || |
10067 | mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED)) | |
e83dff5e | 10068 | goto out; |
28bf2888 | 10069 | |
66450a21 JK |
10070 | if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) { |
10071 | vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED; | |
10072 | set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events); | |
10073 | } else | |
10074 | vcpu->arch.mp_state = mp_state->mp_state; | |
3842d135 | 10075 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
e83dff5e CD |
10076 | |
10077 | ret = 0; | |
10078 | out: | |
10079 | vcpu_put(vcpu); | |
10080 | return ret; | |
62d9f0db MT |
10081 | } |
10082 | ||
7f3d35fd KW |
10083 | int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index, |
10084 | int reason, bool has_error_code, u32 error_code) | |
b6c7a5dc | 10085 | { |
c9b8b07c | 10086 | struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt; |
8ec4722d | 10087 | int ret; |
e01c2426 | 10088 | |
8ec4722d | 10089 | init_emulate_ctxt(vcpu); |
c697518a | 10090 | |
7f3d35fd | 10091 | ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason, |
9d74191a | 10092 | has_error_code, error_code); |
1051778f SC |
10093 | if (ret) { |
10094 | vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
10095 | vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
10096 | vcpu->run->internal.ndata = 0; | |
60fc3d02 | 10097 | return 0; |
1051778f | 10098 | } |
37817f29 | 10099 | |
9d74191a TY |
10100 | kvm_rip_write(vcpu, ctxt->eip); |
10101 | kvm_set_rflags(vcpu, ctxt->eflags); | |
60fc3d02 | 10102 | return 1; |
37817f29 IE |
10103 | } |
10104 | EXPORT_SYMBOL_GPL(kvm_task_switch); | |
10105 | ||
ee69c92b | 10106 | static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs) |
f2981033 | 10107 | { |
37b95951 | 10108 | if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) { |
f2981033 LT |
10109 | /* |
10110 | * When EFER.LME and CR0.PG are set, the processor is in | |
10111 | * 64-bit mode (though maybe in a 32-bit code segment). | |
10112 | * CR4.PAE and EFER.LMA must be set. | |
10113 | */ | |
ee69c92b SC |
10114 | if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA)) |
10115 | return false; | |
ca29e145 | 10116 | if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3)) |
c1c35cf7 | 10117 | return false; |
f2981033 LT |
10118 | } else { |
10119 | /* | |
10120 | * Not in 64-bit mode: EFER.LMA is clear and the code | |
10121 | * segment cannot be 64-bit. | |
10122 | */ | |
10123 | if (sregs->efer & EFER_LMA || sregs->cs.l) | |
ee69c92b | 10124 | return false; |
f2981033 LT |
10125 | } |
10126 | ||
ee69c92b | 10127 | return kvm_is_valid_cr4(vcpu, sregs->cr4); |
f2981033 LT |
10128 | } |
10129 | ||
01643c51 | 10130 | static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs) |
b6c7a5dc | 10131 | { |
58cb628d | 10132 | struct msr_data apic_base_msr; |
b6c7a5dc | 10133 | int mmu_reset_needed = 0; |
63f42e02 | 10134 | int pending_vec, max_bits, idx; |
89a27f4d | 10135 | struct desc_ptr dt; |
b4ef9d4e CD |
10136 | int ret = -EINVAL; |
10137 | ||
ee69c92b | 10138 | if (!kvm_is_valid_sregs(vcpu, sregs)) |
8dbfb2bf | 10139 | goto out; |
f2981033 | 10140 | |
d3802286 JM |
10141 | apic_base_msr.data = sregs->apic_base; |
10142 | apic_base_msr.host_initiated = true; | |
10143 | if (kvm_set_apic_base(vcpu, &apic_base_msr)) | |
b4ef9d4e | 10144 | goto out; |
6d1068b3 | 10145 | |
5265713a TL |
10146 | if (vcpu->arch.guest_state_protected) |
10147 | goto skip_protected_regs; | |
10148 | ||
89a27f4d GN |
10149 | dt.size = sregs->idt.limit; |
10150 | dt.address = sregs->idt.base; | |
b3646477 | 10151 | static_call(kvm_x86_set_idt)(vcpu, &dt); |
89a27f4d GN |
10152 | dt.size = sregs->gdt.limit; |
10153 | dt.address = sregs->gdt.base; | |
b3646477 | 10154 | static_call(kvm_x86_set_gdt)(vcpu, &dt); |
b6c7a5dc | 10155 | |
ad312c7c | 10156 | vcpu->arch.cr2 = sregs->cr2; |
9f8fe504 | 10157 | mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3; |
dc7e795e | 10158 | vcpu->arch.cr3 = sregs->cr3; |
cb3c1e2f | 10159 | kvm_register_mark_available(vcpu, VCPU_EXREG_CR3); |
b6c7a5dc | 10160 | |
2d3ad1f4 | 10161 | kvm_set_cr8(vcpu, sregs->cr8); |
b6c7a5dc | 10162 | |
f6801dff | 10163 | mmu_reset_needed |= vcpu->arch.efer != sregs->efer; |
b3646477 | 10164 | static_call(kvm_x86_set_efer)(vcpu, sregs->efer); |
b6c7a5dc | 10165 | |
4d4ec087 | 10166 | mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0; |
b3646477 | 10167 | static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0); |
d7306163 | 10168 | vcpu->arch.cr0 = sregs->cr0; |
b6c7a5dc | 10169 | |
fc78f519 | 10170 | mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4; |
b3646477 | 10171 | static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4); |
63f42e02 XG |
10172 | |
10173 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
bf03d4f9 | 10174 | if (is_pae_paging(vcpu)) { |
9f8fe504 | 10175 | load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)); |
7c93be44 MT |
10176 | mmu_reset_needed = 1; |
10177 | } | |
63f42e02 | 10178 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
b6c7a5dc HB |
10179 | |
10180 | if (mmu_reset_needed) | |
10181 | kvm_mmu_reset_context(vcpu); | |
10182 | ||
3e6e0aab GT |
10183 | kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS); |
10184 | kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS); | |
10185 | kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES); | |
10186 | kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS); | |
10187 | kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS); | |
10188 | kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS); | |
b6c7a5dc | 10189 | |
3e6e0aab GT |
10190 | kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR); |
10191 | kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR); | |
b6c7a5dc | 10192 | |
5f0269f5 ME |
10193 | update_cr8_intercept(vcpu); |
10194 | ||
9c3e4aab | 10195 | /* Older userspace won't unhalt the vcpu on reset. */ |
c5af89b6 | 10196 | if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 && |
9c3e4aab | 10197 | sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 && |
3eeb3288 | 10198 | !is_protmode(vcpu)) |
9c3e4aab MT |
10199 | vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; |
10200 | ||
5265713a TL |
10201 | skip_protected_regs: |
10202 | max_bits = KVM_NR_INTERRUPTS; | |
10203 | pending_vec = find_first_bit( | |
10204 | (const unsigned long *)sregs->interrupt_bitmap, max_bits); | |
10205 | if (pending_vec < max_bits) { | |
10206 | kvm_queue_interrupt(vcpu, pending_vec, false); | |
10207 | pr_debug("Set back pending irq %d\n", pending_vec); | |
10208 | } | |
10209 | ||
3842d135 AK |
10210 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
10211 | ||
b4ef9d4e CD |
10212 | ret = 0; |
10213 | out: | |
01643c51 KH |
10214 | return ret; |
10215 | } | |
10216 | ||
10217 | int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu, | |
10218 | struct kvm_sregs *sregs) | |
10219 | { | |
10220 | int ret; | |
10221 | ||
10222 | vcpu_load(vcpu); | |
10223 | ret = __set_sregs(vcpu, sregs); | |
b4ef9d4e CD |
10224 | vcpu_put(vcpu); |
10225 | return ret; | |
b6c7a5dc HB |
10226 | } |
10227 | ||
d0bfb940 JK |
10228 | int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu, |
10229 | struct kvm_guest_debug *dbg) | |
b6c7a5dc | 10230 | { |
355be0b9 | 10231 | unsigned long rflags; |
ae675ef0 | 10232 | int i, r; |
b6c7a5dc | 10233 | |
8d4846b9 TL |
10234 | if (vcpu->arch.guest_state_protected) |
10235 | return -EINVAL; | |
10236 | ||
66b56562 CD |
10237 | vcpu_load(vcpu); |
10238 | ||
4f926bf2 JK |
10239 | if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) { |
10240 | r = -EBUSY; | |
10241 | if (vcpu->arch.exception.pending) | |
2122ff5e | 10242 | goto out; |
4f926bf2 JK |
10243 | if (dbg->control & KVM_GUESTDBG_INJECT_DB) |
10244 | kvm_queue_exception(vcpu, DB_VECTOR); | |
10245 | else | |
10246 | kvm_queue_exception(vcpu, BP_VECTOR); | |
10247 | } | |
10248 | ||
91586a3b JK |
10249 | /* |
10250 | * Read rflags as long as potentially injected trace flags are still | |
10251 | * filtered out. | |
10252 | */ | |
10253 | rflags = kvm_get_rflags(vcpu); | |
355be0b9 JK |
10254 | |
10255 | vcpu->guest_debug = dbg->control; | |
10256 | if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE)) | |
10257 | vcpu->guest_debug = 0; | |
10258 | ||
10259 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) { | |
ae675ef0 JK |
10260 | for (i = 0; i < KVM_NR_DB_REGS; ++i) |
10261 | vcpu->arch.eff_db[i] = dbg->arch.debugreg[i]; | |
c8639010 | 10262 | vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7]; |
ae675ef0 JK |
10263 | } else { |
10264 | for (i = 0; i < KVM_NR_DB_REGS; i++) | |
10265 | vcpu->arch.eff_db[i] = vcpu->arch.db[i]; | |
ae675ef0 | 10266 | } |
c8639010 | 10267 | kvm_update_dr7(vcpu); |
ae675ef0 | 10268 | |
f92653ee | 10269 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) |
e87e46d5 | 10270 | vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu); |
94fe45da | 10271 | |
91586a3b JK |
10272 | /* |
10273 | * Trigger an rflags update that will inject or remove the trace | |
10274 | * flags. | |
10275 | */ | |
10276 | kvm_set_rflags(vcpu, rflags); | |
b6c7a5dc | 10277 | |
b3646477 | 10278 | static_call(kvm_x86_update_exception_bitmap)(vcpu); |
b6c7a5dc | 10279 | |
4f926bf2 | 10280 | r = 0; |
d0bfb940 | 10281 | |
2122ff5e | 10282 | out: |
66b56562 | 10283 | vcpu_put(vcpu); |
b6c7a5dc HB |
10284 | return r; |
10285 | } | |
10286 | ||
8b006791 ZX |
10287 | /* |
10288 | * Translate a guest virtual address to a guest physical address. | |
10289 | */ | |
10290 | int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu, | |
10291 | struct kvm_translation *tr) | |
10292 | { | |
10293 | unsigned long vaddr = tr->linear_address; | |
10294 | gpa_t gpa; | |
f656ce01 | 10295 | int idx; |
8b006791 | 10296 | |
1da5b61d CD |
10297 | vcpu_load(vcpu); |
10298 | ||
f656ce01 | 10299 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
1871c602 | 10300 | gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL); |
f656ce01 | 10301 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
8b006791 ZX |
10302 | tr->physical_address = gpa; |
10303 | tr->valid = gpa != UNMAPPED_GVA; | |
10304 | tr->writeable = 1; | |
10305 | tr->usermode = 0; | |
8b006791 | 10306 | |
1da5b61d | 10307 | vcpu_put(vcpu); |
8b006791 ZX |
10308 | return 0; |
10309 | } | |
10310 | ||
d0752060 HB |
10311 | int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) |
10312 | { | |
1393123e | 10313 | struct fxregs_state *fxsave; |
d0752060 | 10314 | |
ed02b213 TL |
10315 | if (!vcpu->arch.guest_fpu) |
10316 | return 0; | |
10317 | ||
1393123e | 10318 | vcpu_load(vcpu); |
d0752060 | 10319 | |
b666a4b6 | 10320 | fxsave = &vcpu->arch.guest_fpu->state.fxsave; |
d0752060 HB |
10321 | memcpy(fpu->fpr, fxsave->st_space, 128); |
10322 | fpu->fcw = fxsave->cwd; | |
10323 | fpu->fsw = fxsave->swd; | |
10324 | fpu->ftwx = fxsave->twd; | |
10325 | fpu->last_opcode = fxsave->fop; | |
10326 | fpu->last_ip = fxsave->rip; | |
10327 | fpu->last_dp = fxsave->rdp; | |
0e96f31e | 10328 | memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space)); |
d0752060 | 10329 | |
1393123e | 10330 | vcpu_put(vcpu); |
d0752060 HB |
10331 | return 0; |
10332 | } | |
10333 | ||
10334 | int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) | |
10335 | { | |
6a96bc7f CD |
10336 | struct fxregs_state *fxsave; |
10337 | ||
ed02b213 TL |
10338 | if (!vcpu->arch.guest_fpu) |
10339 | return 0; | |
10340 | ||
6a96bc7f CD |
10341 | vcpu_load(vcpu); |
10342 | ||
b666a4b6 | 10343 | fxsave = &vcpu->arch.guest_fpu->state.fxsave; |
d0752060 | 10344 | |
d0752060 HB |
10345 | memcpy(fxsave->st_space, fpu->fpr, 128); |
10346 | fxsave->cwd = fpu->fcw; | |
10347 | fxsave->swd = fpu->fsw; | |
10348 | fxsave->twd = fpu->ftwx; | |
10349 | fxsave->fop = fpu->last_opcode; | |
10350 | fxsave->rip = fpu->last_ip; | |
10351 | fxsave->rdp = fpu->last_dp; | |
0e96f31e | 10352 | memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space)); |
d0752060 | 10353 | |
6a96bc7f | 10354 | vcpu_put(vcpu); |
d0752060 HB |
10355 | return 0; |
10356 | } | |
10357 | ||
01643c51 KH |
10358 | static void store_regs(struct kvm_vcpu *vcpu) |
10359 | { | |
10360 | BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES); | |
10361 | ||
10362 | if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS) | |
10363 | __get_regs(vcpu, &vcpu->run->s.regs.regs); | |
10364 | ||
10365 | if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS) | |
10366 | __get_sregs(vcpu, &vcpu->run->s.regs.sregs); | |
10367 | ||
10368 | if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS) | |
10369 | kvm_vcpu_ioctl_x86_get_vcpu_events( | |
10370 | vcpu, &vcpu->run->s.regs.events); | |
10371 | } | |
10372 | ||
10373 | static int sync_regs(struct kvm_vcpu *vcpu) | |
10374 | { | |
10375 | if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS) | |
10376 | return -EINVAL; | |
10377 | ||
10378 | if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) { | |
10379 | __set_regs(vcpu, &vcpu->run->s.regs.regs); | |
10380 | vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS; | |
10381 | } | |
10382 | if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) { | |
10383 | if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs)) | |
10384 | return -EINVAL; | |
10385 | vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS; | |
10386 | } | |
10387 | if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) { | |
10388 | if (kvm_vcpu_ioctl_x86_set_vcpu_events( | |
10389 | vcpu, &vcpu->run->s.regs.events)) | |
10390 | return -EINVAL; | |
10391 | vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS; | |
10392 | } | |
10393 | ||
10394 | return 0; | |
10395 | } | |
10396 | ||
0ee6a517 | 10397 | static void fx_init(struct kvm_vcpu *vcpu) |
d0752060 | 10398 | { |
ed02b213 TL |
10399 | if (!vcpu->arch.guest_fpu) |
10400 | return; | |
10401 | ||
b666a4b6 | 10402 | fpstate_init(&vcpu->arch.guest_fpu->state); |
782511b0 | 10403 | if (boot_cpu_has(X86_FEATURE_XSAVES)) |
b666a4b6 | 10404 | vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv = |
df1daba7 | 10405 | host_xcr0 | XSTATE_COMPACTION_ENABLED; |
d0752060 | 10406 | |
2acf923e DC |
10407 | /* |
10408 | * Ensure guest xcr0 is valid for loading | |
10409 | */ | |
d91cab78 | 10410 | vcpu->arch.xcr0 = XFEATURE_MASK_FP; |
2acf923e | 10411 | |
ad312c7c | 10412 | vcpu->arch.cr0 |= X86_CR0_ET; |
d0752060 | 10413 | } |
d0752060 | 10414 | |
ed02b213 TL |
10415 | void kvm_free_guest_fpu(struct kvm_vcpu *vcpu) |
10416 | { | |
10417 | if (vcpu->arch.guest_fpu) { | |
10418 | kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu); | |
10419 | vcpu->arch.guest_fpu = NULL; | |
10420 | } | |
10421 | } | |
10422 | EXPORT_SYMBOL_GPL(kvm_free_guest_fpu); | |
10423 | ||
897cc38e | 10424 | int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id) |
e9b11c17 | 10425 | { |
897cc38e SC |
10426 | if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0) |
10427 | pr_warn_once("kvm: SMP vm created on host with unstable TSC; " | |
10428 | "guest TSC will not be reliable\n"); | |
7f1ea208 | 10429 | |
897cc38e | 10430 | return 0; |
e9b11c17 ZX |
10431 | } |
10432 | ||
e529ef66 | 10433 | int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu) |
e9b11c17 | 10434 | { |
95a0d01e SC |
10435 | struct page *page; |
10436 | int r; | |
c447e76b | 10437 | |
95a0d01e SC |
10438 | if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu)) |
10439 | vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; | |
10440 | else | |
10441 | vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED; | |
c447e76b | 10442 | |
95a0d01e SC |
10443 | r = kvm_mmu_create(vcpu); |
10444 | if (r < 0) | |
10445 | return r; | |
10446 | ||
10447 | if (irqchip_in_kernel(vcpu->kvm)) { | |
95a0d01e SC |
10448 | r = kvm_create_lapic(vcpu, lapic_timer_advance_ns); |
10449 | if (r < 0) | |
10450 | goto fail_mmu_destroy; | |
4e19c36f SS |
10451 | if (kvm_apicv_activated(vcpu->kvm)) |
10452 | vcpu->arch.apicv_active = true; | |
95a0d01e | 10453 | } else |
6e4e3b4d | 10454 | static_branch_inc(&kvm_has_noapic_vcpu); |
95a0d01e SC |
10455 | |
10456 | r = -ENOMEM; | |
10457 | ||
93bb59ca | 10458 | page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO); |
95a0d01e SC |
10459 | if (!page) |
10460 | goto fail_free_lapic; | |
10461 | vcpu->arch.pio_data = page_address(page); | |
10462 | ||
10463 | vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4, | |
10464 | GFP_KERNEL_ACCOUNT); | |
10465 | if (!vcpu->arch.mce_banks) | |
10466 | goto fail_free_pio_data; | |
10467 | vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS; | |
10468 | ||
10469 | if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, | |
10470 | GFP_KERNEL_ACCOUNT)) | |
10471 | goto fail_free_mce_banks; | |
10472 | ||
c9b8b07c SC |
10473 | if (!alloc_emulate_ctxt(vcpu)) |
10474 | goto free_wbinvd_dirty_mask; | |
10475 | ||
95a0d01e SC |
10476 | vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache, |
10477 | GFP_KERNEL_ACCOUNT); | |
10478 | if (!vcpu->arch.user_fpu) { | |
10479 | pr_err("kvm: failed to allocate userspace's fpu\n"); | |
c9b8b07c | 10480 | goto free_emulate_ctxt; |
95a0d01e SC |
10481 | } |
10482 | ||
10483 | vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache, | |
10484 | GFP_KERNEL_ACCOUNT); | |
10485 | if (!vcpu->arch.guest_fpu) { | |
10486 | pr_err("kvm: failed to allocate vcpu's fpu\n"); | |
10487 | goto free_user_fpu; | |
10488 | } | |
10489 | fx_init(vcpu); | |
10490 | ||
95a0d01e | 10491 | vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu); |
a8ac864a | 10492 | vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu); |
95a0d01e SC |
10493 | |
10494 | vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT; | |
10495 | ||
10496 | kvm_async_pf_hash_reset(vcpu); | |
10497 | kvm_pmu_init(vcpu); | |
10498 | ||
10499 | vcpu->arch.pending_external_vector = -1; | |
10500 | vcpu->arch.preempted_in_kernel = false; | |
10501 | ||
b3646477 | 10502 | r = static_call(kvm_x86_vcpu_create)(vcpu); |
95a0d01e SC |
10503 | if (r) |
10504 | goto free_guest_fpu; | |
e9b11c17 | 10505 | |
0cf9135b | 10506 | vcpu->arch.arch_capabilities = kvm_get_arch_capabilities(); |
e53d88af | 10507 | vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT; |
19efffa2 | 10508 | kvm_vcpu_mtrr_init(vcpu); |
ec7660cc | 10509 | vcpu_load(vcpu); |
1ab9287a | 10510 | kvm_set_tsc_khz(vcpu, max_tsc_khz); |
d28bc9dd | 10511 | kvm_vcpu_reset(vcpu, false); |
e1732991 | 10512 | kvm_init_mmu(vcpu, false); |
e9b11c17 | 10513 | vcpu_put(vcpu); |
ec7660cc | 10514 | return 0; |
95a0d01e SC |
10515 | |
10516 | free_guest_fpu: | |
ed02b213 | 10517 | kvm_free_guest_fpu(vcpu); |
95a0d01e SC |
10518 | free_user_fpu: |
10519 | kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu); | |
c9b8b07c SC |
10520 | free_emulate_ctxt: |
10521 | kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt); | |
95a0d01e SC |
10522 | free_wbinvd_dirty_mask: |
10523 | free_cpumask_var(vcpu->arch.wbinvd_dirty_mask); | |
10524 | fail_free_mce_banks: | |
10525 | kfree(vcpu->arch.mce_banks); | |
10526 | fail_free_pio_data: | |
10527 | free_page((unsigned long)vcpu->arch.pio_data); | |
10528 | fail_free_lapic: | |
10529 | kvm_free_lapic(vcpu); | |
10530 | fail_mmu_destroy: | |
10531 | kvm_mmu_destroy(vcpu); | |
10532 | return r; | |
e9b11c17 ZX |
10533 | } |
10534 | ||
31928aa5 | 10535 | void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) |
42897d86 | 10536 | { |
332967a3 | 10537 | struct kvm *kvm = vcpu->kvm; |
42897d86 | 10538 | |
ec7660cc | 10539 | if (mutex_lock_killable(&vcpu->mutex)) |
31928aa5 | 10540 | return; |
ec7660cc | 10541 | vcpu_load(vcpu); |
0c899c25 | 10542 | kvm_synchronize_tsc(vcpu, 0); |
42897d86 | 10543 | vcpu_put(vcpu); |
2d5ba19b MT |
10544 | |
10545 | /* poll control enabled by default */ | |
10546 | vcpu->arch.msr_kvm_poll_control = 1; | |
10547 | ||
ec7660cc | 10548 | mutex_unlock(&vcpu->mutex); |
42897d86 | 10549 | |
b34de572 WL |
10550 | if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0) |
10551 | schedule_delayed_work(&kvm->arch.kvmclock_sync_work, | |
10552 | KVMCLOCK_SYNC_PERIOD); | |
42897d86 MT |
10553 | } |
10554 | ||
d40ccc62 | 10555 | void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) |
e9b11c17 | 10556 | { |
4cbc418a | 10557 | struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache; |
95a0d01e | 10558 | int idx; |
344d9588 | 10559 | |
4cbc418a PB |
10560 | kvm_release_pfn(cache->pfn, cache->dirty, cache); |
10561 | ||
50b143e1 | 10562 | kvmclock_reset(vcpu); |
e9b11c17 | 10563 | |
b3646477 | 10564 | static_call(kvm_x86_vcpu_free)(vcpu); |
50b143e1 | 10565 | |
c9b8b07c | 10566 | kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt); |
50b143e1 SC |
10567 | free_cpumask_var(vcpu->arch.wbinvd_dirty_mask); |
10568 | kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu); | |
ed02b213 | 10569 | kvm_free_guest_fpu(vcpu); |
95a0d01e SC |
10570 | |
10571 | kvm_hv_vcpu_uninit(vcpu); | |
10572 | kvm_pmu_destroy(vcpu); | |
10573 | kfree(vcpu->arch.mce_banks); | |
10574 | kvm_free_lapic(vcpu); | |
10575 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
10576 | kvm_mmu_destroy(vcpu); | |
10577 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
10578 | free_page((unsigned long)vcpu->arch.pio_data); | |
255cbecf | 10579 | kvfree(vcpu->arch.cpuid_entries); |
95a0d01e | 10580 | if (!lapic_in_kernel(vcpu)) |
6e4e3b4d | 10581 | static_branch_dec(&kvm_has_noapic_vcpu); |
e9b11c17 ZX |
10582 | } |
10583 | ||
d28bc9dd | 10584 | void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event) |
e9b11c17 | 10585 | { |
b7e31be3 RK |
10586 | kvm_lapic_reset(vcpu, init_event); |
10587 | ||
e69fab5d PB |
10588 | vcpu->arch.hflags = 0; |
10589 | ||
c43203ca | 10590 | vcpu->arch.smi_pending = 0; |
52797bf9 | 10591 | vcpu->arch.smi_count = 0; |
7460fb4a AK |
10592 | atomic_set(&vcpu->arch.nmi_queued, 0); |
10593 | vcpu->arch.nmi_pending = 0; | |
448fa4a9 | 10594 | vcpu->arch.nmi_injected = false; |
5f7552d4 NA |
10595 | kvm_clear_interrupt_queue(vcpu); |
10596 | kvm_clear_exception_queue(vcpu); | |
448fa4a9 | 10597 | |
42dbaa5a | 10598 | memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db)); |
ae561ede | 10599 | kvm_update_dr0123(vcpu); |
9a3ecd5e | 10600 | vcpu->arch.dr6 = DR6_ACTIVE_LOW; |
42dbaa5a | 10601 | vcpu->arch.dr7 = DR7_FIXED_1; |
c8639010 | 10602 | kvm_update_dr7(vcpu); |
42dbaa5a | 10603 | |
1119022c NA |
10604 | vcpu->arch.cr2 = 0; |
10605 | ||
3842d135 | 10606 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
2635b5c4 VK |
10607 | vcpu->arch.apf.msr_en_val = 0; |
10608 | vcpu->arch.apf.msr_int_val = 0; | |
c9aaa895 | 10609 | vcpu->arch.st.msr_val = 0; |
3842d135 | 10610 | |
12f9a48f GC |
10611 | kvmclock_reset(vcpu); |
10612 | ||
af585b92 GN |
10613 | kvm_clear_async_pf_completion_queue(vcpu); |
10614 | kvm_async_pf_hash_reset(vcpu); | |
10615 | vcpu->arch.apf.halted = false; | |
3842d135 | 10616 | |
ed02b213 | 10617 | if (vcpu->arch.guest_fpu && kvm_mpx_supported()) { |
a554d207 WL |
10618 | void *mpx_state_buffer; |
10619 | ||
10620 | /* | |
10621 | * To avoid have the INIT path from kvm_apic_has_events() that be | |
10622 | * called with loaded FPU and does not let userspace fix the state. | |
10623 | */ | |
f775b13e RR |
10624 | if (init_event) |
10625 | kvm_put_guest_fpu(vcpu); | |
b666a4b6 | 10626 | mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave, |
abd16d68 | 10627 | XFEATURE_BNDREGS); |
a554d207 WL |
10628 | if (mpx_state_buffer) |
10629 | memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state)); | |
b666a4b6 | 10630 | mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave, |
abd16d68 | 10631 | XFEATURE_BNDCSR); |
a554d207 WL |
10632 | if (mpx_state_buffer) |
10633 | memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr)); | |
f775b13e RR |
10634 | if (init_event) |
10635 | kvm_load_guest_fpu(vcpu); | |
a554d207 WL |
10636 | } |
10637 | ||
64d60670 | 10638 | if (!init_event) { |
d28bc9dd | 10639 | kvm_pmu_reset(vcpu); |
64d60670 | 10640 | vcpu->arch.smbase = 0x30000; |
db2336a8 | 10641 | |
db2336a8 | 10642 | vcpu->arch.msr_misc_features_enables = 0; |
a554d207 WL |
10643 | |
10644 | vcpu->arch.xcr0 = XFEATURE_MASK_FP; | |
64d60670 | 10645 | } |
f5132b01 | 10646 | |
66f7b72e JS |
10647 | memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs)); |
10648 | vcpu->arch.regs_avail = ~0; | |
10649 | vcpu->arch.regs_dirty = ~0; | |
10650 | ||
a554d207 WL |
10651 | vcpu->arch.ia32_xss = 0; |
10652 | ||
b3646477 | 10653 | static_call(kvm_x86_vcpu_reset)(vcpu, init_event); |
e9b11c17 ZX |
10654 | } |
10655 | ||
2b4a273b | 10656 | void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector) |
66450a21 JK |
10657 | { |
10658 | struct kvm_segment cs; | |
10659 | ||
10660 | kvm_get_segment(vcpu, &cs, VCPU_SREG_CS); | |
10661 | cs.selector = vector << 8; | |
10662 | cs.base = vector << 12; | |
10663 | kvm_set_segment(vcpu, &cs, VCPU_SREG_CS); | |
10664 | kvm_rip_write(vcpu, 0); | |
e9b11c17 | 10665 | } |
647daca2 | 10666 | EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector); |
e9b11c17 | 10667 | |
13a34e06 | 10668 | int kvm_arch_hardware_enable(void) |
e9b11c17 | 10669 | { |
ca84d1a2 ZA |
10670 | struct kvm *kvm; |
10671 | struct kvm_vcpu *vcpu; | |
10672 | int i; | |
0dd6a6ed ZA |
10673 | int ret; |
10674 | u64 local_tsc; | |
10675 | u64 max_tsc = 0; | |
10676 | bool stable, backwards_tsc = false; | |
18863bdd | 10677 | |
7e34fbd0 | 10678 | kvm_user_return_msr_cpu_online(); |
b3646477 | 10679 | ret = static_call(kvm_x86_hardware_enable)(); |
0dd6a6ed ZA |
10680 | if (ret != 0) |
10681 | return ret; | |
10682 | ||
4ea1636b | 10683 | local_tsc = rdtsc(); |
b0c39dc6 | 10684 | stable = !kvm_check_tsc_unstable(); |
0dd6a6ed ZA |
10685 | list_for_each_entry(kvm, &vm_list, vm_list) { |
10686 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
10687 | if (!stable && vcpu->cpu == smp_processor_id()) | |
105b21bb | 10688 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
0dd6a6ed ZA |
10689 | if (stable && vcpu->arch.last_host_tsc > local_tsc) { |
10690 | backwards_tsc = true; | |
10691 | if (vcpu->arch.last_host_tsc > max_tsc) | |
10692 | max_tsc = vcpu->arch.last_host_tsc; | |
10693 | } | |
10694 | } | |
10695 | } | |
10696 | ||
10697 | /* | |
10698 | * Sometimes, even reliable TSCs go backwards. This happens on | |
10699 | * platforms that reset TSC during suspend or hibernate actions, but | |
10700 | * maintain synchronization. We must compensate. Fortunately, we can | |
10701 | * detect that condition here, which happens early in CPU bringup, | |
10702 | * before any KVM threads can be running. Unfortunately, we can't | |
10703 | * bring the TSCs fully up to date with real time, as we aren't yet far | |
10704 | * enough into CPU bringup that we know how much real time has actually | |
9285ec4c | 10705 | * elapsed; our helper function, ktime_get_boottime_ns() will be using boot |
0dd6a6ed ZA |
10706 | * variables that haven't been updated yet. |
10707 | * | |
10708 | * So we simply find the maximum observed TSC above, then record the | |
10709 | * adjustment to TSC in each VCPU. When the VCPU later gets loaded, | |
10710 | * the adjustment will be applied. Note that we accumulate | |
10711 | * adjustments, in case multiple suspend cycles happen before some VCPU | |
10712 | * gets a chance to run again. In the event that no KVM threads get a | |
10713 | * chance to run, we will miss the entire elapsed period, as we'll have | |
10714 | * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may | |
10715 | * loose cycle time. This isn't too big a deal, since the loss will be | |
10716 | * uniform across all VCPUs (not to mention the scenario is extremely | |
10717 | * unlikely). It is possible that a second hibernate recovery happens | |
10718 | * much faster than a first, causing the observed TSC here to be | |
10719 | * smaller; this would require additional padding adjustment, which is | |
10720 | * why we set last_host_tsc to the local tsc observed here. | |
10721 | * | |
10722 | * N.B. - this code below runs only on platforms with reliable TSC, | |
10723 | * as that is the only way backwards_tsc is set above. Also note | |
10724 | * that this runs for ALL vcpus, which is not a bug; all VCPUs should | |
10725 | * have the same delta_cyc adjustment applied if backwards_tsc | |
10726 | * is detected. Note further, this adjustment is only done once, | |
10727 | * as we reset last_host_tsc on all VCPUs to stop this from being | |
10728 | * called multiple times (one for each physical CPU bringup). | |
10729 | * | |
4a969980 | 10730 | * Platforms with unreliable TSCs don't have to deal with this, they |
0dd6a6ed ZA |
10731 | * will be compensated by the logic in vcpu_load, which sets the TSC to |
10732 | * catchup mode. This will catchup all VCPUs to real time, but cannot | |
10733 | * guarantee that they stay in perfect synchronization. | |
10734 | */ | |
10735 | if (backwards_tsc) { | |
10736 | u64 delta_cyc = max_tsc - local_tsc; | |
10737 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
a826faf1 | 10738 | kvm->arch.backwards_tsc_observed = true; |
0dd6a6ed ZA |
10739 | kvm_for_each_vcpu(i, vcpu, kvm) { |
10740 | vcpu->arch.tsc_offset_adjustment += delta_cyc; | |
10741 | vcpu->arch.last_host_tsc = local_tsc; | |
105b21bb | 10742 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); |
0dd6a6ed ZA |
10743 | } |
10744 | ||
10745 | /* | |
10746 | * We have to disable TSC offset matching.. if you were | |
10747 | * booting a VM while issuing an S4 host suspend.... | |
10748 | * you may have some problem. Solving this issue is | |
10749 | * left as an exercise to the reader. | |
10750 | */ | |
10751 | kvm->arch.last_tsc_nsec = 0; | |
10752 | kvm->arch.last_tsc_write = 0; | |
10753 | } | |
10754 | ||
10755 | } | |
10756 | return 0; | |
e9b11c17 ZX |
10757 | } |
10758 | ||
13a34e06 | 10759 | void kvm_arch_hardware_disable(void) |
e9b11c17 | 10760 | { |
b3646477 | 10761 | static_call(kvm_x86_hardware_disable)(); |
13a34e06 | 10762 | drop_user_return_notifiers(); |
e9b11c17 ZX |
10763 | } |
10764 | ||
b9904085 | 10765 | int kvm_arch_hardware_setup(void *opaque) |
e9b11c17 | 10766 | { |
d008dfdb | 10767 | struct kvm_x86_init_ops *ops = opaque; |
9e9c3fe4 NA |
10768 | int r; |
10769 | ||
91661989 SC |
10770 | rdmsrl_safe(MSR_EFER, &host_efer); |
10771 | ||
408e9a31 PB |
10772 | if (boot_cpu_has(X86_FEATURE_XSAVES)) |
10773 | rdmsrl(MSR_IA32_XSS, host_xss); | |
10774 | ||
d008dfdb | 10775 | r = ops->hardware_setup(); |
9e9c3fe4 NA |
10776 | if (r != 0) |
10777 | return r; | |
10778 | ||
afaf0b2f | 10779 | memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops)); |
b3646477 | 10780 | kvm_ops_static_call_update(); |
69c6f69a | 10781 | |
408e9a31 PB |
10782 | if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES)) |
10783 | supported_xss = 0; | |
10784 | ||
139f7425 PB |
10785 | #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f) |
10786 | cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_); | |
10787 | #undef __kvm_cpu_cap_has | |
b11306b5 | 10788 | |
35181e86 HZ |
10789 | if (kvm_has_tsc_control) { |
10790 | /* | |
10791 | * Make sure the user can only configure tsc_khz values that | |
10792 | * fit into a signed integer. | |
273ba457 | 10793 | * A min value is not calculated because it will always |
35181e86 HZ |
10794 | * be 1 on all machines. |
10795 | */ | |
10796 | u64 max = min(0x7fffffffULL, | |
10797 | __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz)); | |
10798 | kvm_max_guest_tsc_khz = max; | |
10799 | ||
ad721883 | 10800 | kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits; |
35181e86 | 10801 | } |
ad721883 | 10802 | |
9e9c3fe4 NA |
10803 | kvm_init_msr_list(); |
10804 | return 0; | |
e9b11c17 ZX |
10805 | } |
10806 | ||
10807 | void kvm_arch_hardware_unsetup(void) | |
10808 | { | |
b3646477 | 10809 | static_call(kvm_x86_hardware_unsetup)(); |
e9b11c17 ZX |
10810 | } |
10811 | ||
b9904085 | 10812 | int kvm_arch_check_processor_compat(void *opaque) |
e9b11c17 | 10813 | { |
f1cdecf5 | 10814 | struct cpuinfo_x86 *c = &cpu_data(smp_processor_id()); |
d008dfdb | 10815 | struct kvm_x86_init_ops *ops = opaque; |
f1cdecf5 SC |
10816 | |
10817 | WARN_ON(!irqs_disabled()); | |
10818 | ||
139f7425 PB |
10819 | if (__cr4_reserved_bits(cpu_has, c) != |
10820 | __cr4_reserved_bits(cpu_has, &boot_cpu_data)) | |
f1cdecf5 SC |
10821 | return -EIO; |
10822 | ||
d008dfdb | 10823 | return ops->check_processor_compatibility(); |
d71ba788 PB |
10824 | } |
10825 | ||
10826 | bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu) | |
10827 | { | |
10828 | return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id; | |
10829 | } | |
10830 | EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp); | |
10831 | ||
10832 | bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu) | |
10833 | { | |
10834 | return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0; | |
e9b11c17 ZX |
10835 | } |
10836 | ||
6e4e3b4d CL |
10837 | __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu); |
10838 | EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu); | |
54e9818f | 10839 | |
e790d9ef RK |
10840 | void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu) |
10841 | { | |
b35e5548 LX |
10842 | struct kvm_pmu *pmu = vcpu_to_pmu(vcpu); |
10843 | ||
c595ceee | 10844 | vcpu->arch.l1tf_flush_l1d = true; |
b35e5548 LX |
10845 | if (pmu->version && unlikely(pmu->event_count)) { |
10846 | pmu->need_cleanup = true; | |
10847 | kvm_make_request(KVM_REQ_PMU, vcpu); | |
10848 | } | |
b3646477 | 10849 | static_call(kvm_x86_sched_in)(vcpu, cpu); |
e790d9ef RK |
10850 | } |
10851 | ||
562b6b08 SC |
10852 | void kvm_arch_free_vm(struct kvm *kvm) |
10853 | { | |
05f04ae4 | 10854 | kfree(to_kvm_hv(kvm)->hv_pa_pg); |
562b6b08 | 10855 | vfree(kvm); |
e790d9ef RK |
10856 | } |
10857 | ||
562b6b08 | 10858 | |
e08b9637 | 10859 | int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) |
d19a9cd2 | 10860 | { |
e08b9637 CO |
10861 | if (type) |
10862 | return -EINVAL; | |
10863 | ||
6ef768fa | 10864 | INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list); |
f05e70ac | 10865 | INIT_LIST_HEAD(&kvm->arch.active_mmu_pages); |
10605204 | 10866 | INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages); |
1aa9b957 | 10867 | INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages); |
4d5c5d0f | 10868 | INIT_LIST_HEAD(&kvm->arch.assigned_dev_head); |
e0f0bbc5 | 10869 | atomic_set(&kvm->arch.noncoherent_dma_count, 0); |
d19a9cd2 | 10870 | |
5550af4d SY |
10871 | /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */ |
10872 | set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap); | |
7a84428a AW |
10873 | /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */ |
10874 | set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID, | |
10875 | &kvm->arch.irq_sources_bitmap); | |
5550af4d | 10876 | |
038f8c11 | 10877 | raw_spin_lock_init(&kvm->arch.tsc_write_lock); |
1e08ec4a | 10878 | mutex_init(&kvm->arch.apic_map_lock); |
d828199e MT |
10879 | spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock); |
10880 | ||
8171cd68 | 10881 | kvm->arch.kvmclock_offset = -get_kvmclock_base_ns(); |
d828199e | 10882 | pvclock_update_vm_gtod_copy(kvm); |
53f658b3 | 10883 | |
6fbbde9a DS |
10884 | kvm->arch.guest_can_read_msr_platform_info = true; |
10885 | ||
7e44e449 | 10886 | INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn); |
332967a3 | 10887 | INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn); |
7e44e449 | 10888 | |
4651fc56 | 10889 | kvm_apicv_init(kvm); |
cbc0236a | 10890 | kvm_hv_init_vm(kvm); |
0eb05bf2 | 10891 | kvm_page_track_init(kvm); |
13d268ca | 10892 | kvm_mmu_init_vm(kvm); |
0eb05bf2 | 10893 | |
b3646477 | 10894 | return static_call(kvm_x86_vm_init)(kvm); |
d19a9cd2 ZX |
10895 | } |
10896 | ||
1aa9b957 JS |
10897 | int kvm_arch_post_init_vm(struct kvm *kvm) |
10898 | { | |
10899 | return kvm_mmu_post_init_vm(kvm); | |
10900 | } | |
10901 | ||
d19a9cd2 ZX |
10902 | static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu) |
10903 | { | |
ec7660cc | 10904 | vcpu_load(vcpu); |
d19a9cd2 ZX |
10905 | kvm_mmu_unload(vcpu); |
10906 | vcpu_put(vcpu); | |
10907 | } | |
10908 | ||
10909 | static void kvm_free_vcpus(struct kvm *kvm) | |
10910 | { | |
10911 | unsigned int i; | |
988a2cae | 10912 | struct kvm_vcpu *vcpu; |
d19a9cd2 ZX |
10913 | |
10914 | /* | |
10915 | * Unpin any mmu pages first. | |
10916 | */ | |
af585b92 GN |
10917 | kvm_for_each_vcpu(i, vcpu, kvm) { |
10918 | kvm_clear_async_pf_completion_queue(vcpu); | |
988a2cae | 10919 | kvm_unload_vcpu_mmu(vcpu); |
af585b92 | 10920 | } |
988a2cae | 10921 | kvm_for_each_vcpu(i, vcpu, kvm) |
4543bdc0 | 10922 | kvm_vcpu_destroy(vcpu); |
988a2cae GN |
10923 | |
10924 | mutex_lock(&kvm->lock); | |
10925 | for (i = 0; i < atomic_read(&kvm->online_vcpus); i++) | |
10926 | kvm->vcpus[i] = NULL; | |
d19a9cd2 | 10927 | |
988a2cae GN |
10928 | atomic_set(&kvm->online_vcpus, 0); |
10929 | mutex_unlock(&kvm->lock); | |
d19a9cd2 ZX |
10930 | } |
10931 | ||
ad8ba2cd SY |
10932 | void kvm_arch_sync_events(struct kvm *kvm) |
10933 | { | |
332967a3 | 10934 | cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work); |
7e44e449 | 10935 | cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work); |
aea924f6 | 10936 | kvm_free_pit(kvm); |
ad8ba2cd SY |
10937 | } |
10938 | ||
ff5a983c PX |
10939 | #define ERR_PTR_USR(e) ((void __user *)ERR_PTR(e)) |
10940 | ||
10941 | /** | |
10942 | * __x86_set_memory_region: Setup KVM internal memory slot | |
10943 | * | |
10944 | * @kvm: the kvm pointer to the VM. | |
10945 | * @id: the slot ID to setup. | |
10946 | * @gpa: the GPA to install the slot (unused when @size == 0). | |
10947 | * @size: the size of the slot. Set to zero to uninstall a slot. | |
10948 | * | |
10949 | * This function helps to setup a KVM internal memory slot. Specify | |
10950 | * @size > 0 to install a new slot, while @size == 0 to uninstall a | |
10951 | * slot. The return code can be one of the following: | |
10952 | * | |
10953 | * HVA: on success (uninstall will return a bogus HVA) | |
10954 | * -errno: on error | |
10955 | * | |
10956 | * The caller should always use IS_ERR() to check the return value | |
10957 | * before use. Note, the KVM internal memory slots are guaranteed to | |
10958 | * remain valid and unchanged until the VM is destroyed, i.e., the | |
10959 | * GPA->HVA translation will not change. However, the HVA is a user | |
10960 | * address, i.e. its accessibility is not guaranteed, and must be | |
10961 | * accessed via __copy_{to,from}_user(). | |
10962 | */ | |
10963 | void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, | |
10964 | u32 size) | |
9da0e4d5 PB |
10965 | { |
10966 | int i, r; | |
3f649ab7 | 10967 | unsigned long hva, old_npages; |
f0d648bd | 10968 | struct kvm_memslots *slots = kvm_memslots(kvm); |
0577d1ab | 10969 | struct kvm_memory_slot *slot; |
9da0e4d5 PB |
10970 | |
10971 | /* Called with kvm->slots_lock held. */ | |
1d8007bd | 10972 | if (WARN_ON(id >= KVM_MEM_SLOTS_NUM)) |
ff5a983c | 10973 | return ERR_PTR_USR(-EINVAL); |
9da0e4d5 | 10974 | |
f0d648bd PB |
10975 | slot = id_to_memslot(slots, id); |
10976 | if (size) { | |
0577d1ab | 10977 | if (slot && slot->npages) |
ff5a983c | 10978 | return ERR_PTR_USR(-EEXIST); |
f0d648bd PB |
10979 | |
10980 | /* | |
10981 | * MAP_SHARED to prevent internal slot pages from being moved | |
10982 | * by fork()/COW. | |
10983 | */ | |
10984 | hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE, | |
10985 | MAP_SHARED | MAP_ANONYMOUS, 0); | |
10986 | if (IS_ERR((void *)hva)) | |
ff5a983c | 10987 | return (void __user *)hva; |
f0d648bd | 10988 | } else { |
0577d1ab | 10989 | if (!slot || !slot->npages) |
46914534 | 10990 | return NULL; |
f0d648bd | 10991 | |
0577d1ab | 10992 | old_npages = slot->npages; |
b66f9bab | 10993 | hva = slot->userspace_addr; |
f0d648bd PB |
10994 | } |
10995 | ||
9da0e4d5 | 10996 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
1d8007bd | 10997 | struct kvm_userspace_memory_region m; |
9da0e4d5 | 10998 | |
1d8007bd PB |
10999 | m.slot = id | (i << 16); |
11000 | m.flags = 0; | |
11001 | m.guest_phys_addr = gpa; | |
f0d648bd | 11002 | m.userspace_addr = hva; |
1d8007bd | 11003 | m.memory_size = size; |
9da0e4d5 PB |
11004 | r = __kvm_set_memory_region(kvm, &m); |
11005 | if (r < 0) | |
ff5a983c | 11006 | return ERR_PTR_USR(r); |
9da0e4d5 PB |
11007 | } |
11008 | ||
103c763c | 11009 | if (!size) |
0577d1ab | 11010 | vm_munmap(hva, old_npages * PAGE_SIZE); |
f0d648bd | 11011 | |
ff5a983c | 11012 | return (void __user *)hva; |
9da0e4d5 PB |
11013 | } |
11014 | EXPORT_SYMBOL_GPL(__x86_set_memory_region); | |
11015 | ||
1aa9b957 JS |
11016 | void kvm_arch_pre_destroy_vm(struct kvm *kvm) |
11017 | { | |
11018 | kvm_mmu_pre_destroy_vm(kvm); | |
11019 | } | |
11020 | ||
d19a9cd2 ZX |
11021 | void kvm_arch_destroy_vm(struct kvm *kvm) |
11022 | { | |
27469d29 AH |
11023 | if (current->mm == kvm->mm) { |
11024 | /* | |
11025 | * Free memory regions allocated on behalf of userspace, | |
11026 | * unless the the memory map has changed due to process exit | |
11027 | * or fd copying. | |
11028 | */ | |
6a3c623b PX |
11029 | mutex_lock(&kvm->slots_lock); |
11030 | __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, | |
11031 | 0, 0); | |
11032 | __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, | |
11033 | 0, 0); | |
11034 | __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0); | |
11035 | mutex_unlock(&kvm->slots_lock); | |
27469d29 | 11036 | } |
b3646477 | 11037 | static_call_cond(kvm_x86_vm_destroy)(kvm); |
b318e8de | 11038 | kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1)); |
c761159c PX |
11039 | kvm_pic_destroy(kvm); |
11040 | kvm_ioapic_destroy(kvm); | |
d19a9cd2 | 11041 | kvm_free_vcpus(kvm); |
af1bae54 | 11042 | kvfree(rcu_dereference_check(kvm->arch.apic_map, 1)); |
66bb8a06 | 11043 | kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1)); |
13d268ca | 11044 | kvm_mmu_uninit_vm(kvm); |
2beb6dad | 11045 | kvm_page_track_cleanup(kvm); |
7d6bbebb | 11046 | kvm_xen_destroy_vm(kvm); |
cbc0236a | 11047 | kvm_hv_destroy_vm(kvm); |
d19a9cd2 | 11048 | } |
0de10343 | 11049 | |
c9b929b3 | 11050 | static void memslot_rmap_free(struct kvm_memory_slot *slot) |
db3fe4eb TY |
11051 | { |
11052 | int i; | |
11053 | ||
d89cc617 | 11054 | for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) { |
e96c81ee SC |
11055 | kvfree(slot->arch.rmap[i]); |
11056 | slot->arch.rmap[i] = NULL; | |
c9b929b3 BG |
11057 | } |
11058 | } | |
e96c81ee | 11059 | |
c9b929b3 BG |
11060 | void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot) |
11061 | { | |
11062 | int i; | |
11063 | ||
11064 | memslot_rmap_free(slot); | |
d89cc617 | 11065 | |
c9b929b3 | 11066 | for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) { |
e96c81ee SC |
11067 | kvfree(slot->arch.lpage_info[i - 1]); |
11068 | slot->arch.lpage_info[i - 1] = NULL; | |
db3fe4eb | 11069 | } |
21ebbeda | 11070 | |
e96c81ee | 11071 | kvm_page_track_free_memslot(slot); |
db3fe4eb TY |
11072 | } |
11073 | ||
56dd1019 BG |
11074 | static int memslot_rmap_alloc(struct kvm_memory_slot *slot, |
11075 | unsigned long npages) | |
11076 | { | |
11077 | const int sz = sizeof(*slot->arch.rmap[0]); | |
11078 | int i; | |
11079 | ||
11080 | for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) { | |
11081 | int level = i + 1; | |
11082 | int lpages = gfn_to_index(slot->base_gfn + npages - 1, | |
11083 | slot->base_gfn, level) + 1; | |
11084 | ||
d501f747 BG |
11085 | WARN_ON(slot->arch.rmap[i]); |
11086 | ||
56dd1019 BG |
11087 | slot->arch.rmap[i] = kvcalloc(lpages, sz, GFP_KERNEL_ACCOUNT); |
11088 | if (!slot->arch.rmap[i]) { | |
11089 | memslot_rmap_free(slot); | |
11090 | return -ENOMEM; | |
11091 | } | |
11092 | } | |
11093 | ||
11094 | return 0; | |
11095 | } | |
11096 | ||
d501f747 BG |
11097 | int alloc_all_memslots_rmaps(struct kvm *kvm) |
11098 | { | |
11099 | struct kvm_memslots *slots; | |
11100 | struct kvm_memory_slot *slot; | |
11101 | int r, i; | |
11102 | ||
11103 | /* | |
11104 | * Check if memslots alreday have rmaps early before acquiring | |
11105 | * the slots_arch_lock below. | |
11106 | */ | |
11107 | if (kvm_memslots_have_rmaps(kvm)) | |
11108 | return 0; | |
11109 | ||
11110 | mutex_lock(&kvm->slots_arch_lock); | |
11111 | ||
11112 | /* | |
11113 | * Read memslots_have_rmaps again, under the slots arch lock, | |
11114 | * before allocating the rmaps | |
11115 | */ | |
11116 | if (kvm_memslots_have_rmaps(kvm)) { | |
11117 | mutex_unlock(&kvm->slots_arch_lock); | |
11118 | return 0; | |
11119 | } | |
11120 | ||
11121 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { | |
11122 | slots = __kvm_memslots(kvm, i); | |
11123 | kvm_for_each_memslot(slot, slots) { | |
11124 | r = memslot_rmap_alloc(slot, slot->npages); | |
11125 | if (r) { | |
11126 | mutex_unlock(&kvm->slots_arch_lock); | |
11127 | return r; | |
11128 | } | |
11129 | } | |
11130 | } | |
11131 | ||
11132 | /* | |
11133 | * Ensure that memslots_have_rmaps becomes true strictly after | |
11134 | * all the rmap pointers are set. | |
11135 | */ | |
11136 | smp_store_release(&kvm->arch.memslots_have_rmaps, true); | |
11137 | mutex_unlock(&kvm->slots_arch_lock); | |
11138 | return 0; | |
11139 | } | |
11140 | ||
a2557408 BG |
11141 | static int kvm_alloc_memslot_metadata(struct kvm *kvm, |
11142 | struct kvm_memory_slot *slot, | |
0dab98b7 | 11143 | unsigned long npages) |
db3fe4eb | 11144 | { |
56dd1019 | 11145 | int i, r; |
db3fe4eb | 11146 | |
edd4fa37 SC |
11147 | /* |
11148 | * Clear out the previous array pointers for the KVM_MR_MOVE case. The | |
11149 | * old arrays will be freed by __kvm_set_memory_region() if installing | |
11150 | * the new memslot is successful. | |
11151 | */ | |
11152 | memset(&slot->arch, 0, sizeof(slot->arch)); | |
11153 | ||
e2209710 | 11154 | if (kvm_memslots_have_rmaps(kvm)) { |
a2557408 BG |
11155 | r = memslot_rmap_alloc(slot, npages); |
11156 | if (r) | |
11157 | return r; | |
11158 | } | |
56dd1019 BG |
11159 | |
11160 | for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) { | |
92f94f1e | 11161 | struct kvm_lpage_info *linfo; |
db3fe4eb TY |
11162 | unsigned long ugfn; |
11163 | int lpages; | |
d89cc617 | 11164 | int level = i + 1; |
db3fe4eb TY |
11165 | |
11166 | lpages = gfn_to_index(slot->base_gfn + npages - 1, | |
11167 | slot->base_gfn, level) + 1; | |
11168 | ||
254272ce | 11169 | linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT); |
92f94f1e | 11170 | if (!linfo) |
db3fe4eb TY |
11171 | goto out_free; |
11172 | ||
92f94f1e XG |
11173 | slot->arch.lpage_info[i - 1] = linfo; |
11174 | ||
db3fe4eb | 11175 | if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1)) |
92f94f1e | 11176 | linfo[0].disallow_lpage = 1; |
db3fe4eb | 11177 | if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1)) |
92f94f1e | 11178 | linfo[lpages - 1].disallow_lpage = 1; |
db3fe4eb TY |
11179 | ugfn = slot->userspace_addr >> PAGE_SHIFT; |
11180 | /* | |
11181 | * If the gfn and userspace address are not aligned wrt each | |
600087b6 | 11182 | * other, disable large page support for this slot. |
db3fe4eb | 11183 | */ |
600087b6 | 11184 | if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) { |
db3fe4eb TY |
11185 | unsigned long j; |
11186 | ||
11187 | for (j = 0; j < lpages; ++j) | |
92f94f1e | 11188 | linfo[j].disallow_lpage = 1; |
db3fe4eb TY |
11189 | } |
11190 | } | |
11191 | ||
21ebbeda XG |
11192 | if (kvm_page_track_create_memslot(slot, npages)) |
11193 | goto out_free; | |
11194 | ||
db3fe4eb TY |
11195 | return 0; |
11196 | ||
11197 | out_free: | |
c9b929b3 | 11198 | memslot_rmap_free(slot); |
d89cc617 | 11199 | |
c9b929b3 | 11200 | for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) { |
548ef284 | 11201 | kvfree(slot->arch.lpage_info[i - 1]); |
d89cc617 | 11202 | slot->arch.lpage_info[i - 1] = NULL; |
db3fe4eb TY |
11203 | } |
11204 | return -ENOMEM; | |
11205 | } | |
11206 | ||
15248258 | 11207 | void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen) |
e59dbe09 | 11208 | { |
91724814 BO |
11209 | struct kvm_vcpu *vcpu; |
11210 | int i; | |
11211 | ||
e6dff7d1 TY |
11212 | /* |
11213 | * memslots->generation has been incremented. | |
11214 | * mmio generation may have reached its maximum value. | |
11215 | */ | |
15248258 | 11216 | kvm_mmu_invalidate_mmio_sptes(kvm, gen); |
91724814 BO |
11217 | |
11218 | /* Force re-initialization of steal_time cache */ | |
11219 | kvm_for_each_vcpu(i, vcpu, kvm) | |
11220 | kvm_vcpu_kick(vcpu); | |
e59dbe09 TY |
11221 | } |
11222 | ||
f7784b8e MT |
11223 | int kvm_arch_prepare_memory_region(struct kvm *kvm, |
11224 | struct kvm_memory_slot *memslot, | |
09170a49 | 11225 | const struct kvm_userspace_memory_region *mem, |
7b6195a9 | 11226 | enum kvm_mr_change change) |
0de10343 | 11227 | { |
0dab98b7 | 11228 | if (change == KVM_MR_CREATE || change == KVM_MR_MOVE) |
a2557408 | 11229 | return kvm_alloc_memslot_metadata(kvm, memslot, |
0dab98b7 | 11230 | mem->memory_size >> PAGE_SHIFT); |
f7784b8e MT |
11231 | return 0; |
11232 | } | |
11233 | ||
a85863c2 MS |
11234 | |
11235 | static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable) | |
11236 | { | |
11237 | struct kvm_arch *ka = &kvm->arch; | |
11238 | ||
11239 | if (!kvm_x86_ops.cpu_dirty_log_size) | |
11240 | return; | |
11241 | ||
11242 | if ((enable && ++ka->cpu_dirty_logging_count == 1) || | |
11243 | (!enable && --ka->cpu_dirty_logging_count == 0)) | |
11244 | kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING); | |
11245 | ||
11246 | WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0); | |
11247 | } | |
11248 | ||
88178fd4 | 11249 | static void kvm_mmu_slot_apply_flags(struct kvm *kvm, |
3741679b AY |
11250 | struct kvm_memory_slot *old, |
11251 | struct kvm_memory_slot *new, | |
11252 | enum kvm_mr_change change) | |
88178fd4 | 11253 | { |
a85863c2 MS |
11254 | bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES; |
11255 | ||
3741679b | 11256 | /* |
a85863c2 MS |
11257 | * Update CPU dirty logging if dirty logging is being toggled. This |
11258 | * applies to all operations. | |
3741679b | 11259 | */ |
a85863c2 MS |
11260 | if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES) |
11261 | kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages); | |
88178fd4 KH |
11262 | |
11263 | /* | |
a85863c2 | 11264 | * Nothing more to do for RO slots (which can't be dirtied and can't be |
b6e16ae5 | 11265 | * made writable) or CREATE/MOVE/DELETE of a slot. |
88178fd4 | 11266 | * |
b6e16ae5 | 11267 | * For a memslot with dirty logging disabled: |
3741679b AY |
11268 | * CREATE: No dirty mappings will already exist. |
11269 | * MOVE/DELETE: The old mappings will already have been cleaned up by | |
11270 | * kvm_arch_flush_shadow_memslot() | |
b6e16ae5 SC |
11271 | * |
11272 | * For a memslot with dirty logging enabled: | |
11273 | * CREATE: No shadow pages exist, thus nothing to write-protect | |
11274 | * and no dirty bits to clear. | |
11275 | * MOVE/DELETE: The old mappings will already have been cleaned up by | |
11276 | * kvm_arch_flush_shadow_memslot(). | |
3741679b | 11277 | */ |
3741679b | 11278 | if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY)) |
88178fd4 | 11279 | return; |
3741679b AY |
11280 | |
11281 | /* | |
52f46079 SC |
11282 | * READONLY and non-flags changes were filtered out above, and the only |
11283 | * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty | |
11284 | * logging isn't being toggled on or off. | |
88178fd4 | 11285 | */ |
52f46079 SC |
11286 | if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES))) |
11287 | return; | |
11288 | ||
b6e16ae5 SC |
11289 | if (!log_dirty_pages) { |
11290 | /* | |
11291 | * Dirty logging tracks sptes in 4k granularity, meaning that | |
11292 | * large sptes have to be split. If live migration succeeds, | |
11293 | * the guest in the source machine will be destroyed and large | |
11294 | * sptes will be created in the destination. However, if the | |
11295 | * guest continues to run in the source machine (for example if | |
11296 | * live migration fails), small sptes will remain around and | |
11297 | * cause bad performance. | |
11298 | * | |
11299 | * Scan sptes if dirty logging has been stopped, dropping those | |
11300 | * which can be collapsed into a single large-page spte. Later | |
11301 | * page faults will create the large-page sptes. | |
11302 | */ | |
3741679b | 11303 | kvm_mmu_zap_collapsible_sptes(kvm, new); |
b6e16ae5 | 11304 | } else { |
89212919 KZ |
11305 | /* |
11306 | * Initially-all-set does not require write protecting any page, | |
11307 | * because they're all assumed to be dirty. | |
11308 | */ | |
11309 | if (kvm_dirty_log_manual_protect_and_init_set(kvm)) | |
11310 | return; | |
a1419f8b | 11311 | |
a018eba5 | 11312 | if (kvm_x86_ops.cpu_dirty_log_size) { |
89212919 KZ |
11313 | kvm_mmu_slot_leaf_clear_dirty(kvm, new); |
11314 | kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M); | |
11315 | } else { | |
11316 | kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K); | |
3c9bd400 | 11317 | } |
88178fd4 KH |
11318 | } |
11319 | } | |
11320 | ||
f7784b8e | 11321 | void kvm_arch_commit_memory_region(struct kvm *kvm, |
09170a49 | 11322 | const struct kvm_userspace_memory_region *mem, |
9d4c197c | 11323 | struct kvm_memory_slot *old, |
f36f3f28 | 11324 | const struct kvm_memory_slot *new, |
8482644a | 11325 | enum kvm_mr_change change) |
f7784b8e | 11326 | { |
48c0e4e9 | 11327 | if (!kvm->arch.n_requested_mmu_pages) |
4d66623c WY |
11328 | kvm_mmu_change_mmu_pages(kvm, |
11329 | kvm_mmu_calculate_default_mmu_pages(kvm)); | |
1c91cad4 | 11330 | |
3ea3b7fa | 11331 | /* |
f36f3f28 | 11332 | * FIXME: const-ify all uses of struct kvm_memory_slot. |
c972f3b1 | 11333 | */ |
3741679b | 11334 | kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change); |
21198846 SC |
11335 | |
11336 | /* Free the arrays associated with the old memslot. */ | |
11337 | if (change == KVM_MR_MOVE) | |
e96c81ee | 11338 | kvm_arch_free_memslot(kvm, old); |
0de10343 | 11339 | } |
1d737c8a | 11340 | |
2df72e9b | 11341 | void kvm_arch_flush_shadow_all(struct kvm *kvm) |
34d4cb8f | 11342 | { |
7390de1e | 11343 | kvm_mmu_zap_all(kvm); |
34d4cb8f MT |
11344 | } |
11345 | ||
2df72e9b MT |
11346 | void kvm_arch_flush_shadow_memslot(struct kvm *kvm, |
11347 | struct kvm_memory_slot *slot) | |
11348 | { | |
ae7cd873 | 11349 | kvm_page_track_flush_slot(kvm, slot); |
2df72e9b MT |
11350 | } |
11351 | ||
e6c67d8c LA |
11352 | static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu) |
11353 | { | |
11354 | return (is_guest_mode(vcpu) && | |
afaf0b2f | 11355 | kvm_x86_ops.guest_apic_has_interrupt && |
b3646477 | 11356 | static_call(kvm_x86_guest_apic_has_interrupt)(vcpu)); |
e6c67d8c LA |
11357 | } |
11358 | ||
5d9bc648 PB |
11359 | static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu) |
11360 | { | |
11361 | if (!list_empty_careful(&vcpu->async_pf.done)) | |
11362 | return true; | |
11363 | ||
11364 | if (kvm_apic_has_events(vcpu)) | |
11365 | return true; | |
11366 | ||
11367 | if (vcpu->arch.pv.pv_unhalted) | |
11368 | return true; | |
11369 | ||
a5f01f8e WL |
11370 | if (vcpu->arch.exception.pending) |
11371 | return true; | |
11372 | ||
47a66eed Z |
11373 | if (kvm_test_request(KVM_REQ_NMI, vcpu) || |
11374 | (vcpu->arch.nmi_pending && | |
b3646477 | 11375 | static_call(kvm_x86_nmi_allowed)(vcpu, false))) |
5d9bc648 PB |
11376 | return true; |
11377 | ||
47a66eed | 11378 | if (kvm_test_request(KVM_REQ_SMI, vcpu) || |
a9fa7cb6 | 11379 | (vcpu->arch.smi_pending && |
b3646477 | 11380 | static_call(kvm_x86_smi_allowed)(vcpu, false))) |
73917739 PB |
11381 | return true; |
11382 | ||
5d9bc648 | 11383 | if (kvm_arch_interrupt_allowed(vcpu) && |
e6c67d8c LA |
11384 | (kvm_cpu_has_interrupt(vcpu) || |
11385 | kvm_guest_apic_has_interrupt(vcpu))) | |
5d9bc648 PB |
11386 | return true; |
11387 | ||
1f4b34f8 AS |
11388 | if (kvm_hv_has_stimer_pending(vcpu)) |
11389 | return true; | |
11390 | ||
d2060bd4 SC |
11391 | if (is_guest_mode(vcpu) && |
11392 | kvm_x86_ops.nested_ops->hv_timer_pending && | |
11393 | kvm_x86_ops.nested_ops->hv_timer_pending(vcpu)) | |
11394 | return true; | |
11395 | ||
5d9bc648 PB |
11396 | return false; |
11397 | } | |
11398 | ||
1d737c8a ZX |
11399 | int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu) |
11400 | { | |
5d9bc648 | 11401 | return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu); |
1d737c8a | 11402 | } |
5736199a | 11403 | |
10dbdf98 | 11404 | bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu) |
17e433b5 | 11405 | { |
b3646477 | 11406 | if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu)) |
52acd22f WL |
11407 | return true; |
11408 | ||
11409 | return false; | |
11410 | } | |
11411 | ||
17e433b5 WL |
11412 | bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu) |
11413 | { | |
11414 | if (READ_ONCE(vcpu->arch.pv.pv_unhalted)) | |
11415 | return true; | |
11416 | ||
11417 | if (kvm_test_request(KVM_REQ_NMI, vcpu) || | |
11418 | kvm_test_request(KVM_REQ_SMI, vcpu) || | |
11419 | kvm_test_request(KVM_REQ_EVENT, vcpu)) | |
11420 | return true; | |
11421 | ||
10dbdf98 | 11422 | return kvm_arch_dy_has_pending_interrupt(vcpu); |
17e433b5 WL |
11423 | } |
11424 | ||
199b5763 LM |
11425 | bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu) |
11426 | { | |
b86bb11e WL |
11427 | if (vcpu->arch.guest_state_protected) |
11428 | return true; | |
11429 | ||
de63ad4c | 11430 | return vcpu->arch.preempted_in_kernel; |
199b5763 LM |
11431 | } |
11432 | ||
b6d33834 | 11433 | int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) |
5736199a | 11434 | { |
b6d33834 | 11435 | return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; |
5736199a | 11436 | } |
78646121 GN |
11437 | |
11438 | int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu) | |
11439 | { | |
b3646477 | 11440 | return static_call(kvm_x86_interrupt_allowed)(vcpu, false); |
78646121 | 11441 | } |
229456fc | 11442 | |
82b32774 | 11443 | unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu) |
f92653ee | 11444 | { |
7ed9abfe TL |
11445 | /* Can't read the RIP when guest state is protected, just return 0 */ |
11446 | if (vcpu->arch.guest_state_protected) | |
11447 | return 0; | |
11448 | ||
82b32774 NA |
11449 | if (is_64_bit_mode(vcpu)) |
11450 | return kvm_rip_read(vcpu); | |
11451 | return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) + | |
11452 | kvm_rip_read(vcpu)); | |
11453 | } | |
11454 | EXPORT_SYMBOL_GPL(kvm_get_linear_rip); | |
f92653ee | 11455 | |
82b32774 NA |
11456 | bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip) |
11457 | { | |
11458 | return kvm_get_linear_rip(vcpu) == linear_rip; | |
f92653ee JK |
11459 | } |
11460 | EXPORT_SYMBOL_GPL(kvm_is_linear_rip); | |
11461 | ||
94fe45da JK |
11462 | unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu) |
11463 | { | |
11464 | unsigned long rflags; | |
11465 | ||
b3646477 | 11466 | rflags = static_call(kvm_x86_get_rflags)(vcpu); |
94fe45da | 11467 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) |
c310bac5 | 11468 | rflags &= ~X86_EFLAGS_TF; |
94fe45da JK |
11469 | return rflags; |
11470 | } | |
11471 | EXPORT_SYMBOL_GPL(kvm_get_rflags); | |
11472 | ||
6addfc42 | 11473 | static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags) |
94fe45da JK |
11474 | { |
11475 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP && | |
f92653ee | 11476 | kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip)) |
c310bac5 | 11477 | rflags |= X86_EFLAGS_TF; |
b3646477 | 11478 | static_call(kvm_x86_set_rflags)(vcpu, rflags); |
6addfc42 PB |
11479 | } |
11480 | ||
11481 | void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags) | |
11482 | { | |
11483 | __kvm_set_rflags(vcpu, rflags); | |
3842d135 | 11484 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
94fe45da JK |
11485 | } |
11486 | EXPORT_SYMBOL_GPL(kvm_set_rflags); | |
11487 | ||
56028d08 GN |
11488 | void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work) |
11489 | { | |
11490 | int r; | |
11491 | ||
44dd3ffa | 11492 | if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) || |
f2e10669 | 11493 | work->wakeup_all) |
56028d08 GN |
11494 | return; |
11495 | ||
11496 | r = kvm_mmu_reload(vcpu); | |
11497 | if (unlikely(r)) | |
11498 | return; | |
11499 | ||
44dd3ffa | 11500 | if (!vcpu->arch.mmu->direct_map && |
d8dd54e0 | 11501 | work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu)) |
fb67e14f XG |
11502 | return; |
11503 | ||
7a02674d | 11504 | kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true); |
56028d08 GN |
11505 | } |
11506 | ||
af585b92 GN |
11507 | static inline u32 kvm_async_pf_hash_fn(gfn_t gfn) |
11508 | { | |
dd03bcaa PX |
11509 | BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU)); |
11510 | ||
af585b92 GN |
11511 | return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU)); |
11512 | } | |
11513 | ||
11514 | static inline u32 kvm_async_pf_next_probe(u32 key) | |
11515 | { | |
dd03bcaa | 11516 | return (key + 1) & (ASYNC_PF_PER_VCPU - 1); |
af585b92 GN |
11517 | } |
11518 | ||
11519 | static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) | |
11520 | { | |
11521 | u32 key = kvm_async_pf_hash_fn(gfn); | |
11522 | ||
11523 | while (vcpu->arch.apf.gfns[key] != ~0) | |
11524 | key = kvm_async_pf_next_probe(key); | |
11525 | ||
11526 | vcpu->arch.apf.gfns[key] = gfn; | |
11527 | } | |
11528 | ||
11529 | static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn) | |
11530 | { | |
11531 | int i; | |
11532 | u32 key = kvm_async_pf_hash_fn(gfn); | |
11533 | ||
dd03bcaa | 11534 | for (i = 0; i < ASYNC_PF_PER_VCPU && |
c7d28c24 XG |
11535 | (vcpu->arch.apf.gfns[key] != gfn && |
11536 | vcpu->arch.apf.gfns[key] != ~0); i++) | |
af585b92 GN |
11537 | key = kvm_async_pf_next_probe(key); |
11538 | ||
11539 | return key; | |
11540 | } | |
11541 | ||
11542 | bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) | |
11543 | { | |
11544 | return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn; | |
11545 | } | |
11546 | ||
11547 | static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) | |
11548 | { | |
11549 | u32 i, j, k; | |
11550 | ||
11551 | i = j = kvm_async_pf_gfn_slot(vcpu, gfn); | |
0fd46044 PX |
11552 | |
11553 | if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn)) | |
11554 | return; | |
11555 | ||
af585b92 GN |
11556 | while (true) { |
11557 | vcpu->arch.apf.gfns[i] = ~0; | |
11558 | do { | |
11559 | j = kvm_async_pf_next_probe(j); | |
11560 | if (vcpu->arch.apf.gfns[j] == ~0) | |
11561 | return; | |
11562 | k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]); | |
11563 | /* | |
11564 | * k lies cyclically in ]i,j] | |
11565 | * | i.k.j | | |
11566 | * |....j i.k.| or |.k..j i...| | |
11567 | */ | |
11568 | } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j)); | |
11569 | vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j]; | |
11570 | i = j; | |
11571 | } | |
11572 | } | |
11573 | ||
68fd66f1 | 11574 | static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu) |
7c90705b | 11575 | { |
68fd66f1 VK |
11576 | u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT; |
11577 | ||
11578 | return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason, | |
11579 | sizeof(reason)); | |
11580 | } | |
11581 | ||
11582 | static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token) | |
11583 | { | |
2635b5c4 | 11584 | unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token); |
4e335d9e | 11585 | |
2635b5c4 VK |
11586 | return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data, |
11587 | &token, offset, sizeof(token)); | |
11588 | } | |
11589 | ||
11590 | static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu) | |
11591 | { | |
11592 | unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token); | |
11593 | u32 val; | |
11594 | ||
11595 | if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data, | |
11596 | &val, offset, sizeof(val))) | |
11597 | return false; | |
11598 | ||
11599 | return !val; | |
7c90705b GN |
11600 | } |
11601 | ||
1dfdb45e PB |
11602 | static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu) |
11603 | { | |
11604 | if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu)) | |
11605 | return false; | |
11606 | ||
2635b5c4 | 11607 | if (!kvm_pv_async_pf_enabled(vcpu) || |
b3646477 | 11608 | (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0)) |
1dfdb45e PB |
11609 | return false; |
11610 | ||
11611 | return true; | |
11612 | } | |
11613 | ||
11614 | bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu) | |
11615 | { | |
11616 | if (unlikely(!lapic_in_kernel(vcpu) || | |
11617 | kvm_event_needs_reinjection(vcpu) || | |
11618 | vcpu->arch.exception.pending)) | |
11619 | return false; | |
11620 | ||
11621 | if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu)) | |
11622 | return false; | |
11623 | ||
11624 | /* | |
11625 | * If interrupts are off we cannot even use an artificial | |
11626 | * halt state. | |
11627 | */ | |
c300ab9f | 11628 | return kvm_arch_interrupt_allowed(vcpu); |
1dfdb45e PB |
11629 | } |
11630 | ||
2a18b7e7 | 11631 | bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu, |
af585b92 GN |
11632 | struct kvm_async_pf *work) |
11633 | { | |
6389ee94 AK |
11634 | struct x86_exception fault; |
11635 | ||
736c291c | 11636 | trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa); |
af585b92 | 11637 | kvm_add_async_pf_gfn(vcpu, work->arch.gfn); |
7c90705b | 11638 | |
1dfdb45e | 11639 | if (kvm_can_deliver_async_pf(vcpu) && |
68fd66f1 | 11640 | !apf_put_user_notpresent(vcpu)) { |
6389ee94 AK |
11641 | fault.vector = PF_VECTOR; |
11642 | fault.error_code_valid = true; | |
11643 | fault.error_code = 0; | |
11644 | fault.nested_page_fault = false; | |
11645 | fault.address = work->arch.token; | |
adfe20fb | 11646 | fault.async_page_fault = true; |
6389ee94 | 11647 | kvm_inject_page_fault(vcpu, &fault); |
2a18b7e7 | 11648 | return true; |
1dfdb45e PB |
11649 | } else { |
11650 | /* | |
11651 | * It is not possible to deliver a paravirtualized asynchronous | |
11652 | * page fault, but putting the guest in an artificial halt state | |
11653 | * can be beneficial nevertheless: if an interrupt arrives, we | |
11654 | * can deliver it timely and perhaps the guest will schedule | |
11655 | * another process. When the instruction that triggered a page | |
11656 | * fault is retried, hopefully the page will be ready in the host. | |
11657 | */ | |
11658 | kvm_make_request(KVM_REQ_APF_HALT, vcpu); | |
2a18b7e7 | 11659 | return false; |
7c90705b | 11660 | } |
af585b92 GN |
11661 | } |
11662 | ||
11663 | void kvm_arch_async_page_present(struct kvm_vcpu *vcpu, | |
11664 | struct kvm_async_pf *work) | |
11665 | { | |
2635b5c4 VK |
11666 | struct kvm_lapic_irq irq = { |
11667 | .delivery_mode = APIC_DM_FIXED, | |
11668 | .vector = vcpu->arch.apf.vec | |
11669 | }; | |
6389ee94 | 11670 | |
f2e10669 | 11671 | if (work->wakeup_all) |
7c90705b GN |
11672 | work->arch.token = ~0; /* broadcast wakeup */ |
11673 | else | |
11674 | kvm_del_async_pf_gfn(vcpu, work->arch.gfn); | |
736c291c | 11675 | trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa); |
7c90705b | 11676 | |
2a18b7e7 VK |
11677 | if ((work->wakeup_all || work->notpresent_injected) && |
11678 | kvm_pv_async_pf_enabled(vcpu) && | |
557a961a VK |
11679 | !apf_put_user_ready(vcpu, work->arch.token)) { |
11680 | vcpu->arch.apf.pageready_pending = true; | |
2635b5c4 | 11681 | kvm_apic_set_irq(vcpu, &irq, NULL); |
557a961a | 11682 | } |
2635b5c4 | 11683 | |
e6d53e3b | 11684 | vcpu->arch.apf.halted = false; |
a4fa1635 | 11685 | vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; |
7c90705b GN |
11686 | } |
11687 | ||
557a961a VK |
11688 | void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu) |
11689 | { | |
11690 | kvm_make_request(KVM_REQ_APF_READY, vcpu); | |
11691 | if (!vcpu->arch.apf.pageready_pending) | |
11692 | kvm_vcpu_kick(vcpu); | |
11693 | } | |
11694 | ||
7c0ade6c | 11695 | bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu) |
7c90705b | 11696 | { |
2635b5c4 | 11697 | if (!kvm_pv_async_pf_enabled(vcpu)) |
7c90705b GN |
11698 | return true; |
11699 | else | |
2f15d027 | 11700 | return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu); |
af585b92 GN |
11701 | } |
11702 | ||
5544eb9b PB |
11703 | void kvm_arch_start_assignment(struct kvm *kvm) |
11704 | { | |
57ab8794 MT |
11705 | if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1) |
11706 | static_call_cond(kvm_x86_start_assignment)(kvm); | |
5544eb9b PB |
11707 | } |
11708 | EXPORT_SYMBOL_GPL(kvm_arch_start_assignment); | |
11709 | ||
11710 | void kvm_arch_end_assignment(struct kvm *kvm) | |
11711 | { | |
11712 | atomic_dec(&kvm->arch.assigned_device_count); | |
11713 | } | |
11714 | EXPORT_SYMBOL_GPL(kvm_arch_end_assignment); | |
11715 | ||
11716 | bool kvm_arch_has_assigned_device(struct kvm *kvm) | |
11717 | { | |
11718 | return atomic_read(&kvm->arch.assigned_device_count); | |
11719 | } | |
11720 | EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device); | |
11721 | ||
e0f0bbc5 AW |
11722 | void kvm_arch_register_noncoherent_dma(struct kvm *kvm) |
11723 | { | |
11724 | atomic_inc(&kvm->arch.noncoherent_dma_count); | |
11725 | } | |
11726 | EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma); | |
11727 | ||
11728 | void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm) | |
11729 | { | |
11730 | atomic_dec(&kvm->arch.noncoherent_dma_count); | |
11731 | } | |
11732 | EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma); | |
11733 | ||
11734 | bool kvm_arch_has_noncoherent_dma(struct kvm *kvm) | |
11735 | { | |
11736 | return atomic_read(&kvm->arch.noncoherent_dma_count); | |
11737 | } | |
11738 | EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma); | |
11739 | ||
14717e20 AW |
11740 | bool kvm_arch_has_irq_bypass(void) |
11741 | { | |
92735b1b | 11742 | return true; |
14717e20 AW |
11743 | } |
11744 | ||
87276880 FW |
11745 | int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons, |
11746 | struct irq_bypass_producer *prod) | |
11747 | { | |
11748 | struct kvm_kernel_irqfd *irqfd = | |
11749 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
2edd9cb7 | 11750 | int ret; |
87276880 | 11751 | |
14717e20 | 11752 | irqfd->producer = prod; |
2edd9cb7 | 11753 | kvm_arch_start_assignment(irqfd->kvm); |
b3646477 | 11754 | ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, |
2edd9cb7 ZL |
11755 | prod->irq, irqfd->gsi, 1); |
11756 | ||
11757 | if (ret) | |
11758 | kvm_arch_end_assignment(irqfd->kvm); | |
87276880 | 11759 | |
2edd9cb7 | 11760 | return ret; |
87276880 FW |
11761 | } |
11762 | ||
11763 | void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons, | |
11764 | struct irq_bypass_producer *prod) | |
11765 | { | |
11766 | int ret; | |
11767 | struct kvm_kernel_irqfd *irqfd = | |
11768 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
11769 | ||
87276880 FW |
11770 | WARN_ON(irqfd->producer != prod); |
11771 | irqfd->producer = NULL; | |
11772 | ||
11773 | /* | |
11774 | * When producer of consumer is unregistered, we change back to | |
11775 | * remapped mode, so we can re-use the current implementation | |
bb3541f1 | 11776 | * when the irq is masked/disabled or the consumer side (KVM |
87276880 FW |
11777 | * int this case doesn't want to receive the interrupts. |
11778 | */ | |
b3646477 | 11779 | ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0); |
87276880 FW |
11780 | if (ret) |
11781 | printk(KERN_INFO "irq bypass consumer (token %p) unregistration" | |
11782 | " fails: %d\n", irqfd->consumer.token, ret); | |
2edd9cb7 ZL |
11783 | |
11784 | kvm_arch_end_assignment(irqfd->kvm); | |
87276880 FW |
11785 | } |
11786 | ||
11787 | int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq, | |
11788 | uint32_t guest_irq, bool set) | |
11789 | { | |
b3646477 | 11790 | return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set); |
87276880 FW |
11791 | } |
11792 | ||
52004014 FW |
11793 | bool kvm_vector_hashing_enabled(void) |
11794 | { | |
11795 | return vector_hashing; | |
11796 | } | |
52004014 | 11797 | |
2d5ba19b MT |
11798 | bool kvm_arch_no_poll(struct kvm_vcpu *vcpu) |
11799 | { | |
11800 | return (vcpu->arch.msr_kvm_poll_control & 1) == 0; | |
11801 | } | |
11802 | EXPORT_SYMBOL_GPL(kvm_arch_no_poll); | |
11803 | ||
841c2be0 ML |
11804 | |
11805 | int kvm_spec_ctrl_test_value(u64 value) | |
6441fa61 | 11806 | { |
841c2be0 ML |
11807 | /* |
11808 | * test that setting IA32_SPEC_CTRL to given value | |
11809 | * is allowed by the host processor | |
11810 | */ | |
6441fa61 | 11811 | |
841c2be0 ML |
11812 | u64 saved_value; |
11813 | unsigned long flags; | |
11814 | int ret = 0; | |
6441fa61 | 11815 | |
841c2be0 | 11816 | local_irq_save(flags); |
6441fa61 | 11817 | |
841c2be0 ML |
11818 | if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value)) |
11819 | ret = 1; | |
11820 | else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value)) | |
11821 | ret = 1; | |
11822 | else | |
11823 | wrmsrl(MSR_IA32_SPEC_CTRL, saved_value); | |
6441fa61 | 11824 | |
841c2be0 | 11825 | local_irq_restore(flags); |
6441fa61 | 11826 | |
841c2be0 | 11827 | return ret; |
6441fa61 | 11828 | } |
841c2be0 | 11829 | EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value); |
2d5ba19b | 11830 | |
89786147 MG |
11831 | void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code) |
11832 | { | |
11833 | struct x86_exception fault; | |
19cf4b7e PB |
11834 | u32 access = error_code & |
11835 | (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK); | |
89786147 MG |
11836 | |
11837 | if (!(error_code & PFERR_PRESENT_MASK) || | |
19cf4b7e | 11838 | vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) { |
89786147 MG |
11839 | /* |
11840 | * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page | |
11841 | * tables probably do not match the TLB. Just proceed | |
11842 | * with the error code that the processor gave. | |
11843 | */ | |
11844 | fault.vector = PF_VECTOR; | |
11845 | fault.error_code_valid = true; | |
11846 | fault.error_code = error_code; | |
11847 | fault.nested_page_fault = false; | |
11848 | fault.address = gva; | |
11849 | } | |
11850 | vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault); | |
6441fa61 | 11851 | } |
89786147 | 11852 | EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error); |
2d5ba19b | 11853 | |
3f3393b3 BM |
11854 | /* |
11855 | * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns | |
11856 | * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value | |
11857 | * indicates whether exit to userspace is needed. | |
11858 | */ | |
11859 | int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r, | |
11860 | struct x86_exception *e) | |
11861 | { | |
11862 | if (r == X86EMUL_PROPAGATE_FAULT) { | |
11863 | kvm_inject_emulated_page_fault(vcpu, e); | |
11864 | return 1; | |
11865 | } | |
11866 | ||
11867 | /* | |
11868 | * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED | |
11869 | * while handling a VMX instruction KVM could've handled the request | |
11870 | * correctly by exiting to userspace and performing I/O but there | |
11871 | * doesn't seem to be a real use-case behind such requests, just return | |
11872 | * KVM_EXIT_INTERNAL_ERROR for now. | |
11873 | */ | |
11874 | vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
11875 | vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
11876 | vcpu->run->internal.ndata = 0; | |
11877 | ||
11878 | return 0; | |
11879 | } | |
11880 | EXPORT_SYMBOL_GPL(kvm_handle_memory_failure); | |
11881 | ||
9715092f BM |
11882 | int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva) |
11883 | { | |
11884 | bool pcid_enabled; | |
11885 | struct x86_exception e; | |
11886 | unsigned i; | |
11887 | unsigned long roots_to_free = 0; | |
11888 | struct { | |
11889 | u64 pcid; | |
11890 | u64 gla; | |
11891 | } operand; | |
11892 | int r; | |
11893 | ||
11894 | r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e); | |
11895 | if (r != X86EMUL_CONTINUE) | |
11896 | return kvm_handle_memory_failure(vcpu, r, &e); | |
11897 | ||
11898 | if (operand.pcid >> 12 != 0) { | |
11899 | kvm_inject_gp(vcpu, 0); | |
11900 | return 1; | |
11901 | } | |
11902 | ||
11903 | pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE); | |
11904 | ||
11905 | switch (type) { | |
11906 | case INVPCID_TYPE_INDIV_ADDR: | |
11907 | if ((!pcid_enabled && (operand.pcid != 0)) || | |
11908 | is_noncanonical_address(operand.gla, vcpu)) { | |
11909 | kvm_inject_gp(vcpu, 0); | |
11910 | return 1; | |
11911 | } | |
11912 | kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid); | |
11913 | return kvm_skip_emulated_instruction(vcpu); | |
11914 | ||
11915 | case INVPCID_TYPE_SINGLE_CTXT: | |
11916 | if (!pcid_enabled && (operand.pcid != 0)) { | |
11917 | kvm_inject_gp(vcpu, 0); | |
11918 | return 1; | |
11919 | } | |
11920 | ||
11921 | if (kvm_get_active_pcid(vcpu) == operand.pcid) { | |
11922 | kvm_mmu_sync_roots(vcpu); | |
11923 | kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu); | |
11924 | } | |
11925 | ||
11926 | for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) | |
11927 | if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd) | |
11928 | == operand.pcid) | |
11929 | roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i); | |
11930 | ||
11931 | kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free); | |
11932 | /* | |
11933 | * If neither the current cr3 nor any of the prev_roots use the | |
11934 | * given PCID, then nothing needs to be done here because a | |
11935 | * resync will happen anyway before switching to any other CR3. | |
11936 | */ | |
11937 | ||
11938 | return kvm_skip_emulated_instruction(vcpu); | |
11939 | ||
11940 | case INVPCID_TYPE_ALL_NON_GLOBAL: | |
11941 | /* | |
11942 | * Currently, KVM doesn't mark global entries in the shadow | |
11943 | * page tables, so a non-global flush just degenerates to a | |
11944 | * global flush. If needed, we could optimize this later by | |
11945 | * keeping track of global entries in shadow page tables. | |
11946 | */ | |
11947 | ||
11948 | fallthrough; | |
11949 | case INVPCID_TYPE_ALL_INCL_GLOBAL: | |
f66c53b3 | 11950 | kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu); |
9715092f BM |
11951 | return kvm_skip_emulated_instruction(vcpu); |
11952 | ||
11953 | default: | |
11954 | BUG(); /* We have already checked above that type <= 3 */ | |
11955 | } | |
11956 | } | |
11957 | EXPORT_SYMBOL_GPL(kvm_handle_invpcid); | |
11958 | ||
8f423a80 TL |
11959 | static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu) |
11960 | { | |
11961 | struct kvm_run *run = vcpu->run; | |
11962 | struct kvm_mmio_fragment *frag; | |
11963 | unsigned int len; | |
11964 | ||
11965 | BUG_ON(!vcpu->mmio_needed); | |
11966 | ||
11967 | /* Complete previous fragment */ | |
11968 | frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment]; | |
11969 | len = min(8u, frag->len); | |
11970 | if (!vcpu->mmio_is_write) | |
11971 | memcpy(frag->data, run->mmio.data, len); | |
11972 | ||
11973 | if (frag->len <= 8) { | |
11974 | /* Switch to the next fragment. */ | |
11975 | frag++; | |
11976 | vcpu->mmio_cur_fragment++; | |
11977 | } else { | |
11978 | /* Go forward to the next mmio piece. */ | |
11979 | frag->data += len; | |
11980 | frag->gpa += len; | |
11981 | frag->len -= len; | |
11982 | } | |
11983 | ||
11984 | if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) { | |
11985 | vcpu->mmio_needed = 0; | |
11986 | ||
11987 | // VMG change, at this point, we're always done | |
11988 | // RIP has already been advanced | |
11989 | return 1; | |
11990 | } | |
11991 | ||
11992 | // More MMIO is needed | |
11993 | run->mmio.phys_addr = frag->gpa; | |
11994 | run->mmio.len = min(8u, frag->len); | |
11995 | run->mmio.is_write = vcpu->mmio_is_write; | |
11996 | if (run->mmio.is_write) | |
11997 | memcpy(run->mmio.data, frag->data, min(8u, frag->len)); | |
11998 | run->exit_reason = KVM_EXIT_MMIO; | |
11999 | ||
12000 | vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio; | |
12001 | ||
12002 | return 0; | |
12003 | } | |
12004 | ||
12005 | int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes, | |
12006 | void *data) | |
12007 | { | |
12008 | int handled; | |
12009 | struct kvm_mmio_fragment *frag; | |
12010 | ||
12011 | if (!data) | |
12012 | return -EINVAL; | |
12013 | ||
12014 | handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data); | |
12015 | if (handled == bytes) | |
12016 | return 1; | |
12017 | ||
12018 | bytes -= handled; | |
12019 | gpa += handled; | |
12020 | data += handled; | |
12021 | ||
12022 | /*TODO: Check if need to increment number of frags */ | |
12023 | frag = vcpu->mmio_fragments; | |
12024 | vcpu->mmio_nr_fragments = 1; | |
12025 | frag->len = bytes; | |
12026 | frag->gpa = gpa; | |
12027 | frag->data = data; | |
12028 | ||
12029 | vcpu->mmio_needed = 1; | |
12030 | vcpu->mmio_cur_fragment = 0; | |
12031 | ||
12032 | vcpu->run->mmio.phys_addr = gpa; | |
12033 | vcpu->run->mmio.len = min(8u, frag->len); | |
12034 | vcpu->run->mmio.is_write = 1; | |
12035 | memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len)); | |
12036 | vcpu->run->exit_reason = KVM_EXIT_MMIO; | |
12037 | ||
12038 | vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio; | |
12039 | ||
12040 | return 0; | |
12041 | } | |
12042 | EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write); | |
12043 | ||
12044 | int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes, | |
12045 | void *data) | |
12046 | { | |
12047 | int handled; | |
12048 | struct kvm_mmio_fragment *frag; | |
12049 | ||
12050 | if (!data) | |
12051 | return -EINVAL; | |
12052 | ||
12053 | handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data); | |
12054 | if (handled == bytes) | |
12055 | return 1; | |
12056 | ||
12057 | bytes -= handled; | |
12058 | gpa += handled; | |
12059 | data += handled; | |
12060 | ||
12061 | /*TODO: Check if need to increment number of frags */ | |
12062 | frag = vcpu->mmio_fragments; | |
12063 | vcpu->mmio_nr_fragments = 1; | |
12064 | frag->len = bytes; | |
12065 | frag->gpa = gpa; | |
12066 | frag->data = data; | |
12067 | ||
12068 | vcpu->mmio_needed = 1; | |
12069 | vcpu->mmio_cur_fragment = 0; | |
12070 | ||
12071 | vcpu->run->mmio.phys_addr = gpa; | |
12072 | vcpu->run->mmio.len = min(8u, frag->len); | |
12073 | vcpu->run->mmio.is_write = 0; | |
12074 | vcpu->run->exit_reason = KVM_EXIT_MMIO; | |
12075 | ||
12076 | vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio; | |
12077 | ||
12078 | return 0; | |
12079 | } | |
12080 | EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read); | |
12081 | ||
7ed9abfe TL |
12082 | static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu) |
12083 | { | |
12084 | memcpy(vcpu->arch.guest_ins_data, vcpu->arch.pio_data, | |
12085 | vcpu->arch.pio.count * vcpu->arch.pio.size); | |
12086 | vcpu->arch.pio.count = 0; | |
12087 | ||
12088 | return 1; | |
12089 | } | |
12090 | ||
12091 | static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size, | |
12092 | unsigned int port, void *data, unsigned int count) | |
12093 | { | |
12094 | int ret; | |
12095 | ||
12096 | ret = emulator_pio_out_emulated(vcpu->arch.emulate_ctxt, size, port, | |
12097 | data, count); | |
12098 | if (ret) | |
12099 | return ret; | |
12100 | ||
12101 | vcpu->arch.pio.count = 0; | |
12102 | ||
12103 | return 0; | |
12104 | } | |
12105 | ||
12106 | static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size, | |
12107 | unsigned int port, void *data, unsigned int count) | |
12108 | { | |
12109 | int ret; | |
12110 | ||
12111 | ret = emulator_pio_in_emulated(vcpu->arch.emulate_ctxt, size, port, | |
12112 | data, count); | |
12113 | if (ret) { | |
12114 | vcpu->arch.pio.count = 0; | |
12115 | } else { | |
12116 | vcpu->arch.guest_ins_data = data; | |
12117 | vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins; | |
12118 | } | |
12119 | ||
12120 | return 0; | |
12121 | } | |
12122 | ||
12123 | int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size, | |
12124 | unsigned int port, void *data, unsigned int count, | |
12125 | int in) | |
12126 | { | |
12127 | return in ? kvm_sev_es_ins(vcpu, size, port, data, count) | |
12128 | : kvm_sev_es_outs(vcpu, size, port, data, count); | |
12129 | } | |
12130 | EXPORT_SYMBOL_GPL(kvm_sev_es_string_io); | |
12131 | ||
d95df951 | 12132 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry); |
229456fc | 12133 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit); |
931c33b1 | 12134 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio); |
229456fc MT |
12135 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq); |
12136 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault); | |
12137 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr); | |
12138 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr); | |
0ac406de | 12139 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun); |
d8cabddf | 12140 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit); |
17897f36 | 12141 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject); |
236649de | 12142 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit); |
5497b955 | 12143 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed); |
ec1ff790 | 12144 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga); |
532a46b9 | 12145 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit); |
2e554e8d | 12146 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts); |
489223ed | 12147 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset); |
4f75bcc3 | 12148 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update); |
843e4330 | 12149 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full); |
efc64404 | 12150 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update); |
18f40c53 SS |
12151 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access); |
12152 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi); | |
ab56f8e6 | 12153 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log); |
24bbf74c | 12154 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request); |
d523ab6b TL |
12155 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter); |
12156 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit); | |
59e38b58 TL |
12157 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter); |
12158 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit); |