]>
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" |
313a3dc7 | 32 | |
18068523 | 33 | #include <linux/clocksource.h> |
4d5c5d0f | 34 | #include <linux/interrupt.h> |
313a3dc7 CO |
35 | #include <linux/kvm.h> |
36 | #include <linux/fs.h> | |
37 | #include <linux/vmalloc.h> | |
1767e931 PG |
38 | #include <linux/export.h> |
39 | #include <linux/moduleparam.h> | |
0de10343 | 40 | #include <linux/mman.h> |
2bacc55c | 41 | #include <linux/highmem.h> |
19de40a8 | 42 | #include <linux/iommu.h> |
62c476c7 | 43 | #include <linux/intel-iommu.h> |
c8076604 | 44 | #include <linux/cpufreq.h> |
18863bdd | 45 | #include <linux/user-return-notifier.h> |
a983fb23 | 46 | #include <linux/srcu.h> |
5a0e3ad6 | 47 | #include <linux/slab.h> |
ff9d07a0 | 48 | #include <linux/perf_event.h> |
7bee342a | 49 | #include <linux/uaccess.h> |
af585b92 | 50 | #include <linux/hash.h> |
a1b60c1c | 51 | #include <linux/pci.h> |
16e8d74d MT |
52 | #include <linux/timekeeper_internal.h> |
53 | #include <linux/pvclock_gtod.h> | |
87276880 FW |
54 | #include <linux/kvm_irqfd.h> |
55 | #include <linux/irqbypass.h> | |
3905f9ad | 56 | #include <linux/sched/stat.h> |
0c5f81da | 57 | #include <linux/sched/isolation.h> |
d0ec49d4 | 58 | #include <linux/mem_encrypt.h> |
72c3c0fe | 59 | #include <linux/entry-kvm.h> |
3905f9ad | 60 | |
aec51dc4 | 61 | #include <trace/events/kvm.h> |
2ed152af | 62 | |
24f1e32c | 63 | #include <asm/debugreg.h> |
d825ed0a | 64 | #include <asm/msr.h> |
a5f61300 | 65 | #include <asm/desc.h> |
890ca9ae | 66 | #include <asm/mce.h> |
f89e32e0 | 67 | #include <linux/kernel_stat.h> |
78f7f1e5 | 68 | #include <asm/fpu/internal.h> /* Ugh! */ |
1d5f066e | 69 | #include <asm/pvclock.h> |
217fc9cf | 70 | #include <asm/div64.h> |
efc64404 | 71 | #include <asm/irq_remapping.h> |
b0c39dc6 | 72 | #include <asm/mshyperv.h> |
0092e434 | 73 | #include <asm/hypervisor.h> |
9715092f | 74 | #include <asm/tlbflush.h> |
bf8c55d8 | 75 | #include <asm/intel_pt.h> |
b3dc0695 | 76 | #include <asm/emulate_prefix.h> |
dd2cb348 | 77 | #include <clocksource/hyperv_timer.h> |
043405e1 | 78 | |
d1898b73 DH |
79 | #define CREATE_TRACE_POINTS |
80 | #include "trace.h" | |
81 | ||
313a3dc7 | 82 | #define MAX_IO_MSRS 256 |
890ca9ae | 83 | #define KVM_MAX_MCE_BANKS 32 |
c45dcc71 AR |
84 | u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P; |
85 | EXPORT_SYMBOL_GPL(kvm_mce_cap_supported); | |
890ca9ae | 86 | |
0f65dd70 | 87 | #define emul_to_vcpu(ctxt) \ |
c9b8b07c | 88 | ((struct kvm_vcpu *)(ctxt)->vcpu) |
0f65dd70 | 89 | |
50a37eb4 JR |
90 | /* EFER defaults: |
91 | * - enable syscall per default because its emulated by KVM | |
92 | * - enable LME and LMA per default on 64 bit KVM | |
93 | */ | |
94 | #ifdef CONFIG_X86_64 | |
1260edbe LJ |
95 | static |
96 | u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA)); | |
50a37eb4 | 97 | #else |
1260edbe | 98 | static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE); |
50a37eb4 | 99 | #endif |
313a3dc7 | 100 | |
b11306b5 SC |
101 | static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS; |
102 | ||
c519265f RK |
103 | #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \ |
104 | KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK) | |
37131313 | 105 | |
cb142eb7 | 106 | static void update_cr8_intercept(struct kvm_vcpu *vcpu); |
7460fb4a | 107 | static void process_nmi(struct kvm_vcpu *vcpu); |
ee2cd4b7 | 108 | static void enter_smm(struct kvm_vcpu *vcpu); |
6addfc42 | 109 | static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags); |
01643c51 KH |
110 | static void store_regs(struct kvm_vcpu *vcpu); |
111 | static int sync_regs(struct kvm_vcpu *vcpu); | |
674eea0f | 112 | |
afaf0b2f | 113 | struct kvm_x86_ops kvm_x86_ops __read_mostly; |
5fdbf976 | 114 | EXPORT_SYMBOL_GPL(kvm_x86_ops); |
97896d04 | 115 | |
893590c7 | 116 | static bool __read_mostly ignore_msrs = 0; |
476bc001 | 117 | module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR); |
ed85c068 | 118 | |
fab0aa3b EM |
119 | static bool __read_mostly report_ignored_msrs = true; |
120 | module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR); | |
121 | ||
4c27625b | 122 | unsigned int min_timer_period_us = 200; |
9ed96e87 MT |
123 | module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR); |
124 | ||
630994b3 MT |
125 | static bool __read_mostly kvmclock_periodic_sync = true; |
126 | module_param(kvmclock_periodic_sync, bool, S_IRUGO); | |
127 | ||
893590c7 | 128 | bool __read_mostly kvm_has_tsc_control; |
92a1f12d | 129 | EXPORT_SYMBOL_GPL(kvm_has_tsc_control); |
893590c7 | 130 | u32 __read_mostly kvm_max_guest_tsc_khz; |
92a1f12d | 131 | EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz); |
bc9b961b HZ |
132 | u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits; |
133 | EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits); | |
134 | u64 __read_mostly kvm_max_tsc_scaling_ratio; | |
135 | EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio); | |
64672c95 YJ |
136 | u64 __read_mostly kvm_default_tsc_scaling_ratio; |
137 | EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio); | |
92a1f12d | 138 | |
cc578287 | 139 | /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */ |
893590c7 | 140 | static u32 __read_mostly tsc_tolerance_ppm = 250; |
cc578287 ZA |
141 | module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR); |
142 | ||
c3941d9e SC |
143 | /* |
144 | * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables | |
145 | * adaptive tuning starting from default advancment of 1000ns. '0' disables | |
146 | * advancement entirely. Any other value is used as-is and disables adaptive | |
147 | * tuning, i.e. allows priveleged userspace to set an exact advancement time. | |
148 | */ | |
149 | static int __read_mostly lapic_timer_advance_ns = -1; | |
0e6edceb | 150 | module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR); |
d0659d94 | 151 | |
52004014 FW |
152 | static bool __read_mostly vector_hashing = true; |
153 | module_param(vector_hashing, bool, S_IRUGO); | |
154 | ||
c4ae60e4 LA |
155 | bool __read_mostly enable_vmware_backdoor = false; |
156 | module_param(enable_vmware_backdoor, bool, S_IRUGO); | |
157 | EXPORT_SYMBOL_GPL(enable_vmware_backdoor); | |
158 | ||
6c86eedc WL |
159 | static bool __read_mostly force_emulation_prefix = false; |
160 | module_param(force_emulation_prefix, bool, S_IRUGO); | |
161 | ||
0c5f81da WL |
162 | int __read_mostly pi_inject_timer = -1; |
163 | module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR); | |
164 | ||
7e34fbd0 SC |
165 | /* |
166 | * Restoring the host value for MSRs that are only consumed when running in | |
167 | * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU | |
168 | * returns to userspace, i.e. the kernel can run with the guest's value. | |
169 | */ | |
170 | #define KVM_MAX_NR_USER_RETURN_MSRS 16 | |
18863bdd | 171 | |
7e34fbd0 | 172 | struct kvm_user_return_msrs_global { |
18863bdd | 173 | int nr; |
7e34fbd0 | 174 | u32 msrs[KVM_MAX_NR_USER_RETURN_MSRS]; |
18863bdd AK |
175 | }; |
176 | ||
7e34fbd0 | 177 | struct kvm_user_return_msrs { |
18863bdd AK |
178 | struct user_return_notifier urn; |
179 | bool registered; | |
7e34fbd0 | 180 | struct kvm_user_return_msr_values { |
2bf78fa7 SY |
181 | u64 host; |
182 | u64 curr; | |
7e34fbd0 | 183 | } values[KVM_MAX_NR_USER_RETURN_MSRS]; |
18863bdd AK |
184 | }; |
185 | ||
7e34fbd0 SC |
186 | static struct kvm_user_return_msrs_global __read_mostly user_return_msrs_global; |
187 | static struct kvm_user_return_msrs __percpu *user_return_msrs; | |
18863bdd | 188 | |
cfc48181 SC |
189 | #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \ |
190 | | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \ | |
191 | | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \ | |
192 | | XFEATURE_MASK_PKRU) | |
193 | ||
91661989 SC |
194 | u64 __read_mostly host_efer; |
195 | EXPORT_SYMBOL_GPL(host_efer); | |
196 | ||
3edd6839 MG |
197 | bool __read_mostly allow_smaller_maxphyaddr; |
198 | EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr); | |
199 | ||
139a12cf | 200 | static u64 __read_mostly host_xss; |
408e9a31 PB |
201 | u64 __read_mostly supported_xss; |
202 | EXPORT_SYMBOL_GPL(supported_xss); | |
139a12cf | 203 | |
417bc304 | 204 | struct kvm_stats_debugfs_item debugfs_entries[] = { |
812756a8 EGE |
205 | VCPU_STAT("pf_fixed", pf_fixed), |
206 | VCPU_STAT("pf_guest", pf_guest), | |
207 | VCPU_STAT("tlb_flush", tlb_flush), | |
208 | VCPU_STAT("invlpg", invlpg), | |
209 | VCPU_STAT("exits", exits), | |
210 | VCPU_STAT("io_exits", io_exits), | |
211 | VCPU_STAT("mmio_exits", mmio_exits), | |
212 | VCPU_STAT("signal_exits", signal_exits), | |
213 | VCPU_STAT("irq_window", irq_window_exits), | |
214 | VCPU_STAT("nmi_window", nmi_window_exits), | |
215 | VCPU_STAT("halt_exits", halt_exits), | |
216 | VCPU_STAT("halt_successful_poll", halt_successful_poll), | |
217 | VCPU_STAT("halt_attempted_poll", halt_attempted_poll), | |
218 | VCPU_STAT("halt_poll_invalid", halt_poll_invalid), | |
219 | VCPU_STAT("halt_wakeup", halt_wakeup), | |
220 | VCPU_STAT("hypercalls", hypercalls), | |
221 | VCPU_STAT("request_irq", request_irq_exits), | |
222 | VCPU_STAT("irq_exits", irq_exits), | |
223 | VCPU_STAT("host_state_reload", host_state_reload), | |
224 | VCPU_STAT("fpu_reload", fpu_reload), | |
225 | VCPU_STAT("insn_emulation", insn_emulation), | |
226 | VCPU_STAT("insn_emulation_fail", insn_emulation_fail), | |
227 | VCPU_STAT("irq_injections", irq_injections), | |
228 | VCPU_STAT("nmi_injections", nmi_injections), | |
229 | VCPU_STAT("req_event", req_event), | |
230 | VCPU_STAT("l1d_flush", l1d_flush), | |
cb953129 DM |
231 | VCPU_STAT("halt_poll_success_ns", halt_poll_success_ns), |
232 | VCPU_STAT("halt_poll_fail_ns", halt_poll_fail_ns), | |
812756a8 EGE |
233 | VM_STAT("mmu_shadow_zapped", mmu_shadow_zapped), |
234 | VM_STAT("mmu_pte_write", mmu_pte_write), | |
235 | VM_STAT("mmu_pte_updated", mmu_pte_updated), | |
236 | VM_STAT("mmu_pde_zapped", mmu_pde_zapped), | |
237 | VM_STAT("mmu_flooded", mmu_flooded), | |
238 | VM_STAT("mmu_recycled", mmu_recycled), | |
239 | VM_STAT("mmu_cache_miss", mmu_cache_miss), | |
240 | VM_STAT("mmu_unsync", mmu_unsync), | |
241 | VM_STAT("remote_tlb_flush", remote_tlb_flush), | |
242 | VM_STAT("largepages", lpages, .mode = 0444), | |
243 | VM_STAT("nx_largepages_splitted", nx_lpage_splits, .mode = 0444), | |
244 | VM_STAT("max_mmu_page_hash_collisions", max_mmu_page_hash_collisions), | |
417bc304 HB |
245 | { NULL } |
246 | }; | |
247 | ||
2acf923e | 248 | u64 __read_mostly host_xcr0; |
cfc48181 SC |
249 | u64 __read_mostly supported_xcr0; |
250 | EXPORT_SYMBOL_GPL(supported_xcr0); | |
2acf923e | 251 | |
80fbd280 | 252 | static struct kmem_cache *x86_fpu_cache; |
b666a4b6 | 253 | |
c9b8b07c SC |
254 | static struct kmem_cache *x86_emulator_cache; |
255 | ||
6abe9c13 PX |
256 | /* |
257 | * When called, it means the previous get/set msr reached an invalid msr. | |
258 | * Return 0 if we want to ignore/silent this failed msr access, or 1 if we want | |
259 | * to fail the caller. | |
260 | */ | |
261 | static int kvm_msr_ignored_check(struct kvm_vcpu *vcpu, u32 msr, | |
262 | u64 data, bool write) | |
263 | { | |
264 | const char *op = write ? "wrmsr" : "rdmsr"; | |
265 | ||
266 | if (ignore_msrs) { | |
267 | if (report_ignored_msrs) | |
268 | vcpu_unimpl(vcpu, "ignored %s: 0x%x data 0x%llx\n", | |
269 | op, msr, data); | |
270 | /* Mask the error */ | |
271 | return 0; | |
272 | } else { | |
273 | vcpu_debug_ratelimited(vcpu, "unhandled %s: 0x%x data 0x%llx\n", | |
274 | op, msr, data); | |
90218e43 | 275 | return -ENOENT; |
6abe9c13 PX |
276 | } |
277 | } | |
278 | ||
c9b8b07c SC |
279 | static struct kmem_cache *kvm_alloc_emulator_cache(void) |
280 | { | |
06add254 SC |
281 | unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src); |
282 | unsigned int size = sizeof(struct x86_emulate_ctxt); | |
283 | ||
284 | return kmem_cache_create_usercopy("x86_emulator", size, | |
c9b8b07c | 285 | __alignof__(struct x86_emulate_ctxt), |
06add254 SC |
286 | SLAB_ACCOUNT, useroffset, |
287 | size - useroffset, NULL); | |
c9b8b07c SC |
288 | } |
289 | ||
b6785def | 290 | static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt); |
d6aa1000 | 291 | |
af585b92 GN |
292 | static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu) |
293 | { | |
294 | int i; | |
dd03bcaa | 295 | for (i = 0; i < ASYNC_PF_PER_VCPU; i++) |
af585b92 GN |
296 | vcpu->arch.apf.gfns[i] = ~0; |
297 | } | |
298 | ||
18863bdd AK |
299 | static void kvm_on_user_return(struct user_return_notifier *urn) |
300 | { | |
301 | unsigned slot; | |
7e34fbd0 SC |
302 | struct kvm_user_return_msrs *msrs |
303 | = container_of(urn, struct kvm_user_return_msrs, urn); | |
304 | struct kvm_user_return_msr_values *values; | |
1650b4eb IA |
305 | unsigned long flags; |
306 | ||
307 | /* | |
308 | * Disabling irqs at this point since the following code could be | |
309 | * interrupted and executed through kvm_arch_hardware_disable() | |
310 | */ | |
311 | local_irq_save(flags); | |
7e34fbd0 SC |
312 | if (msrs->registered) { |
313 | msrs->registered = false; | |
1650b4eb IA |
314 | user_return_notifier_unregister(urn); |
315 | } | |
316 | local_irq_restore(flags); | |
7e34fbd0 SC |
317 | for (slot = 0; slot < user_return_msrs_global.nr; ++slot) { |
318 | values = &msrs->values[slot]; | |
2bf78fa7 | 319 | if (values->host != values->curr) { |
7e34fbd0 | 320 | wrmsrl(user_return_msrs_global.msrs[slot], values->host); |
2bf78fa7 | 321 | values->curr = values->host; |
18863bdd AK |
322 | } |
323 | } | |
18863bdd AK |
324 | } |
325 | ||
7e34fbd0 | 326 | void kvm_define_user_return_msr(unsigned slot, u32 msr) |
2bf78fa7 | 327 | { |
7e34fbd0 SC |
328 | BUG_ON(slot >= KVM_MAX_NR_USER_RETURN_MSRS); |
329 | user_return_msrs_global.msrs[slot] = msr; | |
330 | if (slot >= user_return_msrs_global.nr) | |
331 | user_return_msrs_global.nr = slot + 1; | |
18863bdd | 332 | } |
7e34fbd0 | 333 | EXPORT_SYMBOL_GPL(kvm_define_user_return_msr); |
18863bdd | 334 | |
7e34fbd0 | 335 | static void kvm_user_return_msr_cpu_online(void) |
18863bdd | 336 | { |
05c19c2f | 337 | unsigned int cpu = smp_processor_id(); |
7e34fbd0 | 338 | struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu); |
05c19c2f SC |
339 | u64 value; |
340 | int i; | |
18863bdd | 341 | |
7e34fbd0 SC |
342 | for (i = 0; i < user_return_msrs_global.nr; ++i) { |
343 | rdmsrl_safe(user_return_msrs_global.msrs[i], &value); | |
344 | msrs->values[i].host = value; | |
345 | msrs->values[i].curr = value; | |
05c19c2f | 346 | } |
18863bdd AK |
347 | } |
348 | ||
7e34fbd0 | 349 | int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask) |
18863bdd | 350 | { |
013f6a5d | 351 | unsigned int cpu = smp_processor_id(); |
7e34fbd0 | 352 | struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu); |
8b3c3104 | 353 | int err; |
18863bdd | 354 | |
7e34fbd0 SC |
355 | value = (value & mask) | (msrs->values[slot].host & ~mask); |
356 | if (value == msrs->values[slot].curr) | |
8b3c3104 | 357 | return 0; |
7e34fbd0 | 358 | err = wrmsrl_safe(user_return_msrs_global.msrs[slot], value); |
8b3c3104 AH |
359 | if (err) |
360 | return 1; | |
361 | ||
7e34fbd0 SC |
362 | msrs->values[slot].curr = value; |
363 | if (!msrs->registered) { | |
364 | msrs->urn.on_user_return = kvm_on_user_return; | |
365 | user_return_notifier_register(&msrs->urn); | |
366 | msrs->registered = true; | |
18863bdd | 367 | } |
8b3c3104 | 368 | return 0; |
18863bdd | 369 | } |
7e34fbd0 | 370 | EXPORT_SYMBOL_GPL(kvm_set_user_return_msr); |
18863bdd | 371 | |
13a34e06 | 372 | static void drop_user_return_notifiers(void) |
3548bab5 | 373 | { |
013f6a5d | 374 | unsigned int cpu = smp_processor_id(); |
7e34fbd0 | 375 | struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu); |
3548bab5 | 376 | |
7e34fbd0 SC |
377 | if (msrs->registered) |
378 | kvm_on_user_return(&msrs->urn); | |
3548bab5 AK |
379 | } |
380 | ||
6866b83e CO |
381 | u64 kvm_get_apic_base(struct kvm_vcpu *vcpu) |
382 | { | |
8a5a87d9 | 383 | return vcpu->arch.apic_base; |
6866b83e CO |
384 | } |
385 | EXPORT_SYMBOL_GPL(kvm_get_apic_base); | |
386 | ||
58871649 JM |
387 | enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu) |
388 | { | |
389 | return kvm_apic_mode(kvm_get_apic_base(vcpu)); | |
390 | } | |
391 | EXPORT_SYMBOL_GPL(kvm_get_apic_mode); | |
392 | ||
58cb628d JK |
393 | int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
394 | { | |
58871649 JM |
395 | enum lapic_mode old_mode = kvm_get_apic_mode(vcpu); |
396 | enum lapic_mode new_mode = kvm_apic_mode(msr_info->data); | |
d6321d49 RK |
397 | u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff | |
398 | (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE); | |
58cb628d | 399 | |
58871649 | 400 | if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID) |
58cb628d | 401 | return 1; |
58871649 JM |
402 | if (!msr_info->host_initiated) { |
403 | if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC) | |
404 | return 1; | |
405 | if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC) | |
406 | return 1; | |
407 | } | |
58cb628d JK |
408 | |
409 | kvm_lapic_set_base(vcpu, msr_info->data); | |
4abaffce | 410 | kvm_recalculate_apic_map(vcpu->kvm); |
58cb628d | 411 | return 0; |
6866b83e CO |
412 | } |
413 | EXPORT_SYMBOL_GPL(kvm_set_apic_base); | |
414 | ||
3ebccdf3 | 415 | asmlinkage __visible noinstr void kvm_spurious_fault(void) |
e3ba45b8 GL |
416 | { |
417 | /* Fault while not rebooting. We want the trace. */ | |
b4fdcf60 | 418 | BUG_ON(!kvm_rebooting); |
e3ba45b8 GL |
419 | } |
420 | EXPORT_SYMBOL_GPL(kvm_spurious_fault); | |
421 | ||
3fd28fce ED |
422 | #define EXCPT_BENIGN 0 |
423 | #define EXCPT_CONTRIBUTORY 1 | |
424 | #define EXCPT_PF 2 | |
425 | ||
426 | static int exception_class(int vector) | |
427 | { | |
428 | switch (vector) { | |
429 | case PF_VECTOR: | |
430 | return EXCPT_PF; | |
431 | case DE_VECTOR: | |
432 | case TS_VECTOR: | |
433 | case NP_VECTOR: | |
434 | case SS_VECTOR: | |
435 | case GP_VECTOR: | |
436 | return EXCPT_CONTRIBUTORY; | |
437 | default: | |
438 | break; | |
439 | } | |
440 | return EXCPT_BENIGN; | |
441 | } | |
442 | ||
d6e8c854 NA |
443 | #define EXCPT_FAULT 0 |
444 | #define EXCPT_TRAP 1 | |
445 | #define EXCPT_ABORT 2 | |
446 | #define EXCPT_INTERRUPT 3 | |
447 | ||
448 | static int exception_type(int vector) | |
449 | { | |
450 | unsigned int mask; | |
451 | ||
452 | if (WARN_ON(vector > 31 || vector == NMI_VECTOR)) | |
453 | return EXCPT_INTERRUPT; | |
454 | ||
455 | mask = 1 << vector; | |
456 | ||
457 | /* #DB is trap, as instruction watchpoints are handled elsewhere */ | |
458 | if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR))) | |
459 | return EXCPT_TRAP; | |
460 | ||
461 | if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR))) | |
462 | return EXCPT_ABORT; | |
463 | ||
464 | /* Reserved exceptions will result in fault */ | |
465 | return EXCPT_FAULT; | |
466 | } | |
467 | ||
da998b46 JM |
468 | void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu) |
469 | { | |
470 | unsigned nr = vcpu->arch.exception.nr; | |
471 | bool has_payload = vcpu->arch.exception.has_payload; | |
472 | unsigned long payload = vcpu->arch.exception.payload; | |
473 | ||
474 | if (!has_payload) | |
475 | return; | |
476 | ||
477 | switch (nr) { | |
f10c729f JM |
478 | case DB_VECTOR: |
479 | /* | |
480 | * "Certain debug exceptions may clear bit 0-3. The | |
481 | * remaining contents of the DR6 register are never | |
482 | * cleared by the processor". | |
483 | */ | |
484 | vcpu->arch.dr6 &= ~DR_TRAP_BITS; | |
485 | /* | |
486 | * DR6.RTM is set by all #DB exceptions that don't clear it. | |
487 | */ | |
488 | vcpu->arch.dr6 |= DR6_RTM; | |
489 | vcpu->arch.dr6 |= payload; | |
490 | /* | |
491 | * Bit 16 should be set in the payload whenever the #DB | |
492 | * exception should clear DR6.RTM. This makes the payload | |
493 | * compatible with the pending debug exceptions under VMX. | |
494 | * Though not currently documented in the SDM, this also | |
495 | * makes the payload compatible with the exit qualification | |
496 | * for #DB exceptions under VMX. | |
497 | */ | |
498 | vcpu->arch.dr6 ^= payload & DR6_RTM; | |
307f1cfa OU |
499 | |
500 | /* | |
501 | * The #DB payload is defined as compatible with the 'pending | |
502 | * debug exceptions' field under VMX, not DR6. While bit 12 is | |
503 | * defined in the 'pending debug exceptions' field (enabled | |
504 | * breakpoint), it is reserved and must be zero in DR6. | |
505 | */ | |
506 | vcpu->arch.dr6 &= ~BIT(12); | |
f10c729f | 507 | break; |
da998b46 JM |
508 | case PF_VECTOR: |
509 | vcpu->arch.cr2 = payload; | |
510 | break; | |
511 | } | |
512 | ||
513 | vcpu->arch.exception.has_payload = false; | |
514 | vcpu->arch.exception.payload = 0; | |
515 | } | |
516 | EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload); | |
517 | ||
3fd28fce | 518 | static void kvm_multiple_exception(struct kvm_vcpu *vcpu, |
ce7ddec4 | 519 | unsigned nr, bool has_error, u32 error_code, |
91e86d22 | 520 | bool has_payload, unsigned long payload, bool reinject) |
3fd28fce ED |
521 | { |
522 | u32 prev_nr; | |
523 | int class1, class2; | |
524 | ||
3842d135 AK |
525 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
526 | ||
664f8e26 | 527 | if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) { |
3fd28fce | 528 | queue: |
3ffb2468 NA |
529 | if (has_error && !is_protmode(vcpu)) |
530 | has_error = false; | |
664f8e26 WL |
531 | if (reinject) { |
532 | /* | |
533 | * On vmentry, vcpu->arch.exception.pending is only | |
534 | * true if an event injection was blocked by | |
535 | * nested_run_pending. In that case, however, | |
536 | * vcpu_enter_guest requests an immediate exit, | |
537 | * and the guest shouldn't proceed far enough to | |
538 | * need reinjection. | |
539 | */ | |
540 | WARN_ON_ONCE(vcpu->arch.exception.pending); | |
541 | vcpu->arch.exception.injected = true; | |
91e86d22 JM |
542 | if (WARN_ON_ONCE(has_payload)) { |
543 | /* | |
544 | * A reinjected event has already | |
545 | * delivered its payload. | |
546 | */ | |
547 | has_payload = false; | |
548 | payload = 0; | |
549 | } | |
664f8e26 WL |
550 | } else { |
551 | vcpu->arch.exception.pending = true; | |
552 | vcpu->arch.exception.injected = false; | |
553 | } | |
3fd28fce ED |
554 | vcpu->arch.exception.has_error_code = has_error; |
555 | vcpu->arch.exception.nr = nr; | |
556 | vcpu->arch.exception.error_code = error_code; | |
91e86d22 JM |
557 | vcpu->arch.exception.has_payload = has_payload; |
558 | vcpu->arch.exception.payload = payload; | |
a06230b6 | 559 | if (!is_guest_mode(vcpu)) |
da998b46 | 560 | kvm_deliver_exception_payload(vcpu); |
3fd28fce ED |
561 | return; |
562 | } | |
563 | ||
564 | /* to check exception */ | |
565 | prev_nr = vcpu->arch.exception.nr; | |
566 | if (prev_nr == DF_VECTOR) { | |
567 | /* triple fault -> shutdown */ | |
a8eeb04a | 568 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); |
3fd28fce ED |
569 | return; |
570 | } | |
571 | class1 = exception_class(prev_nr); | |
572 | class2 = exception_class(nr); | |
573 | if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY) | |
574 | || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) { | |
664f8e26 WL |
575 | /* |
576 | * Generate double fault per SDM Table 5-5. Set | |
577 | * exception.pending = true so that the double fault | |
578 | * can trigger a nested vmexit. | |
579 | */ | |
3fd28fce | 580 | vcpu->arch.exception.pending = true; |
664f8e26 | 581 | vcpu->arch.exception.injected = false; |
3fd28fce ED |
582 | vcpu->arch.exception.has_error_code = true; |
583 | vcpu->arch.exception.nr = DF_VECTOR; | |
584 | vcpu->arch.exception.error_code = 0; | |
c851436a JM |
585 | vcpu->arch.exception.has_payload = false; |
586 | vcpu->arch.exception.payload = 0; | |
3fd28fce ED |
587 | } else |
588 | /* replace previous exception with a new one in a hope | |
589 | that instruction re-execution will regenerate lost | |
590 | exception */ | |
591 | goto queue; | |
592 | } | |
593 | ||
298101da AK |
594 | void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr) |
595 | { | |
91e86d22 | 596 | kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false); |
298101da AK |
597 | } |
598 | EXPORT_SYMBOL_GPL(kvm_queue_exception); | |
599 | ||
ce7ddec4 JR |
600 | void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr) |
601 | { | |
91e86d22 | 602 | kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true); |
ce7ddec4 JR |
603 | } |
604 | EXPORT_SYMBOL_GPL(kvm_requeue_exception); | |
605 | ||
4d5523cf PB |
606 | void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr, |
607 | unsigned long payload) | |
f10c729f JM |
608 | { |
609 | kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false); | |
610 | } | |
4d5523cf | 611 | EXPORT_SYMBOL_GPL(kvm_queue_exception_p); |
f10c729f | 612 | |
da998b46 JM |
613 | static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr, |
614 | u32 error_code, unsigned long payload) | |
615 | { | |
616 | kvm_multiple_exception(vcpu, nr, true, error_code, | |
617 | true, payload, false); | |
618 | } | |
619 | ||
6affcbed | 620 | int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err) |
c3c91fee | 621 | { |
db8fcefa AP |
622 | if (err) |
623 | kvm_inject_gp(vcpu, 0); | |
624 | else | |
6affcbed KH |
625 | return kvm_skip_emulated_instruction(vcpu); |
626 | ||
627 | return 1; | |
db8fcefa AP |
628 | } |
629 | EXPORT_SYMBOL_GPL(kvm_complete_insn_gp); | |
8df25a32 | 630 | |
6389ee94 | 631 | void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault) |
c3c91fee AK |
632 | { |
633 | ++vcpu->stat.pf_guest; | |
adfe20fb WL |
634 | vcpu->arch.exception.nested_apf = |
635 | is_guest_mode(vcpu) && fault->async_page_fault; | |
da998b46 | 636 | if (vcpu->arch.exception.nested_apf) { |
adfe20fb | 637 | vcpu->arch.apf.nested_apf_token = fault->address; |
da998b46 JM |
638 | kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code); |
639 | } else { | |
640 | kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code, | |
641 | fault->address); | |
642 | } | |
c3c91fee | 643 | } |
27d6c865 | 644 | EXPORT_SYMBOL_GPL(kvm_inject_page_fault); |
c3c91fee | 645 | |
53b3d8e9 SC |
646 | bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu, |
647 | struct x86_exception *fault) | |
d4f8cf66 | 648 | { |
0cd665bd | 649 | struct kvm_mmu *fault_mmu; |
53b3d8e9 SC |
650 | WARN_ON_ONCE(fault->vector != PF_VECTOR); |
651 | ||
0cd665bd PB |
652 | fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu : |
653 | vcpu->arch.walk_mmu; | |
ef54bcfe | 654 | |
ee1fa209 JS |
655 | /* |
656 | * Invalidate the TLB entry for the faulting address, if it exists, | |
657 | * else the access will fault indefinitely (and to emulate hardware). | |
658 | */ | |
659 | if ((fault->error_code & PFERR_PRESENT_MASK) && | |
660 | !(fault->error_code & PFERR_RSVD_MASK)) | |
661 | kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address, | |
662 | fault_mmu->root_hpa); | |
663 | ||
664 | fault_mmu->inject_page_fault(vcpu, fault); | |
ef54bcfe | 665 | return fault->nested_page_fault; |
d4f8cf66 | 666 | } |
53b3d8e9 | 667 | EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault); |
d4f8cf66 | 668 | |
3419ffc8 SY |
669 | void kvm_inject_nmi(struct kvm_vcpu *vcpu) |
670 | { | |
7460fb4a AK |
671 | atomic_inc(&vcpu->arch.nmi_queued); |
672 | kvm_make_request(KVM_REQ_NMI, vcpu); | |
3419ffc8 SY |
673 | } |
674 | EXPORT_SYMBOL_GPL(kvm_inject_nmi); | |
675 | ||
298101da AK |
676 | void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code) |
677 | { | |
91e86d22 | 678 | kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false); |
298101da AK |
679 | } |
680 | EXPORT_SYMBOL_GPL(kvm_queue_exception_e); | |
681 | ||
ce7ddec4 JR |
682 | void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code) |
683 | { | |
91e86d22 | 684 | kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true); |
ce7ddec4 JR |
685 | } |
686 | EXPORT_SYMBOL_GPL(kvm_requeue_exception_e); | |
687 | ||
0a79b009 AK |
688 | /* |
689 | * Checks if cpl <= required_cpl; if true, return true. Otherwise queue | |
690 | * a #GP and return false. | |
691 | */ | |
692 | bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl) | |
298101da | 693 | { |
afaf0b2f | 694 | if (kvm_x86_ops.get_cpl(vcpu) <= required_cpl) |
0a79b009 AK |
695 | return true; |
696 | kvm_queue_exception_e(vcpu, GP_VECTOR, 0); | |
697 | return false; | |
298101da | 698 | } |
0a79b009 | 699 | EXPORT_SYMBOL_GPL(kvm_require_cpl); |
298101da | 700 | |
16f8a6f9 NA |
701 | bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr) |
702 | { | |
703 | if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE)) | |
704 | return true; | |
705 | ||
706 | kvm_queue_exception(vcpu, UD_VECTOR); | |
707 | return false; | |
708 | } | |
709 | EXPORT_SYMBOL_GPL(kvm_require_dr); | |
710 | ||
ec92fe44 JR |
711 | /* |
712 | * This function will be used to read from the physical memory of the currently | |
54bf36aa | 713 | * running guest. The difference to kvm_vcpu_read_guest_page is that this function |
ec92fe44 JR |
714 | * can read from guest physical or from the guest's guest physical memory. |
715 | */ | |
716 | int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, | |
717 | gfn_t ngfn, void *data, int offset, int len, | |
718 | u32 access) | |
719 | { | |
54987b7a | 720 | struct x86_exception exception; |
ec92fe44 JR |
721 | gfn_t real_gfn; |
722 | gpa_t ngpa; | |
723 | ||
724 | ngpa = gfn_to_gpa(ngfn); | |
54987b7a | 725 | real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception); |
ec92fe44 JR |
726 | if (real_gfn == UNMAPPED_GVA) |
727 | return -EFAULT; | |
728 | ||
729 | real_gfn = gpa_to_gfn(real_gfn); | |
730 | ||
54bf36aa | 731 | return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len); |
ec92fe44 JR |
732 | } |
733 | EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu); | |
734 | ||
69b0049a | 735 | static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
3d06b8bf JR |
736 | void *data, int offset, int len, u32 access) |
737 | { | |
738 | return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn, | |
739 | data, offset, len, access); | |
740 | } | |
741 | ||
16cfacc8 SC |
742 | static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu) |
743 | { | |
744 | return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) | | |
745 | rsvd_bits(1, 2); | |
746 | } | |
747 | ||
a03490ed | 748 | /* |
16cfacc8 | 749 | * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise. |
a03490ed | 750 | */ |
ff03a073 | 751 | int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3) |
a03490ed CO |
752 | { |
753 | gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT; | |
754 | unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2; | |
755 | int i; | |
756 | int ret; | |
ff03a073 | 757 | u64 pdpte[ARRAY_SIZE(mmu->pdptrs)]; |
a03490ed | 758 | |
ff03a073 JR |
759 | ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte, |
760 | offset * sizeof(u64), sizeof(pdpte), | |
761 | PFERR_USER_MASK|PFERR_WRITE_MASK); | |
a03490ed CO |
762 | if (ret < 0) { |
763 | ret = 0; | |
764 | goto out; | |
765 | } | |
766 | for (i = 0; i < ARRAY_SIZE(pdpte); ++i) { | |
812f30b2 | 767 | if ((pdpte[i] & PT_PRESENT_MASK) && |
16cfacc8 | 768 | (pdpte[i] & pdptr_rsvd_bits(vcpu))) { |
a03490ed CO |
769 | ret = 0; |
770 | goto out; | |
771 | } | |
772 | } | |
773 | ret = 1; | |
774 | ||
ff03a073 | 775 | memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs)); |
cb3c1e2f SC |
776 | kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR); |
777 | ||
a03490ed | 778 | out: |
a03490ed CO |
779 | |
780 | return ret; | |
781 | } | |
cc4b6871 | 782 | EXPORT_SYMBOL_GPL(load_pdptrs); |
a03490ed | 783 | |
9ed38ffa | 784 | bool pdptrs_changed(struct kvm_vcpu *vcpu) |
d835dfec | 785 | { |
ff03a073 | 786 | u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)]; |
3d06b8bf JR |
787 | int offset; |
788 | gfn_t gfn; | |
d835dfec AK |
789 | int r; |
790 | ||
bf03d4f9 | 791 | if (!is_pae_paging(vcpu)) |
d835dfec AK |
792 | return false; |
793 | ||
cb3c1e2f | 794 | if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR)) |
6de4f3ad AK |
795 | return true; |
796 | ||
a512177e PB |
797 | gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT; |
798 | offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1); | |
3d06b8bf JR |
799 | r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte), |
800 | PFERR_USER_MASK | PFERR_WRITE_MASK); | |
d835dfec | 801 | if (r < 0) |
7f7f0d9c | 802 | return true; |
d835dfec | 803 | |
7f7f0d9c | 804 | return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0; |
d835dfec | 805 | } |
9ed38ffa | 806 | EXPORT_SYMBOL_GPL(pdptrs_changed); |
d835dfec | 807 | |
49a9b07e | 808 | int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0) |
a03490ed | 809 | { |
aad82703 | 810 | unsigned long old_cr0 = kvm_read_cr0(vcpu); |
d42e3fae | 811 | unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG; |
d81135a5 | 812 | unsigned long update_bits = X86_CR0_PG | X86_CR0_WP; |
aad82703 | 813 | |
f9a48e6a AK |
814 | cr0 |= X86_CR0_ET; |
815 | ||
ab344828 | 816 | #ifdef CONFIG_X86_64 |
0f12244f GN |
817 | if (cr0 & 0xffffffff00000000UL) |
818 | return 1; | |
ab344828 GN |
819 | #endif |
820 | ||
821 | cr0 &= ~CR0_RESERVED_BITS; | |
a03490ed | 822 | |
0f12244f GN |
823 | if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) |
824 | return 1; | |
a03490ed | 825 | |
0f12244f GN |
826 | if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) |
827 | return 1; | |
a03490ed | 828 | |
a03490ed | 829 | #ifdef CONFIG_X86_64 |
05487215 SC |
830 | if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) && |
831 | (cr0 & X86_CR0_PG)) { | |
832 | int cs_db, cs_l; | |
833 | ||
834 | if (!is_pae(vcpu)) | |
835 | return 1; | |
836 | kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l); | |
837 | if (cs_l) | |
0f12244f | 838 | return 1; |
a03490ed | 839 | } |
05487215 SC |
840 | #endif |
841 | if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) && | |
842 | is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) && | |
843 | !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu))) | |
844 | return 1; | |
a03490ed | 845 | |
ad756a16 MJ |
846 | if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)) |
847 | return 1; | |
848 | ||
afaf0b2f | 849 | kvm_x86_ops.set_cr0(vcpu, cr0); |
a03490ed | 850 | |
d170c419 | 851 | if ((cr0 ^ old_cr0) & X86_CR0_PG) { |
e5f3f027 | 852 | kvm_clear_async_pf_completion_queue(vcpu); |
d170c419 LJ |
853 | kvm_async_pf_hash_reset(vcpu); |
854 | } | |
e5f3f027 | 855 | |
aad82703 SY |
856 | if ((cr0 ^ old_cr0) & update_bits) |
857 | kvm_mmu_reset_context(vcpu); | |
b18d5431 | 858 | |
879ae188 LE |
859 | if (((cr0 ^ old_cr0) & X86_CR0_CD) && |
860 | kvm_arch_has_noncoherent_dma(vcpu->kvm) && | |
861 | !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED)) | |
b18d5431 XG |
862 | kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL); |
863 | ||
0f12244f GN |
864 | return 0; |
865 | } | |
2d3ad1f4 | 866 | EXPORT_SYMBOL_GPL(kvm_set_cr0); |
a03490ed | 867 | |
2d3ad1f4 | 868 | void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw) |
a03490ed | 869 | { |
49a9b07e | 870 | (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f)); |
a03490ed | 871 | } |
2d3ad1f4 | 872 | EXPORT_SYMBOL_GPL(kvm_lmsw); |
a03490ed | 873 | |
139a12cf | 874 | void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu) |
42bdf991 | 875 | { |
139a12cf AL |
876 | if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) { |
877 | ||
878 | if (vcpu->arch.xcr0 != host_xcr0) | |
879 | xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0); | |
880 | ||
881 | if (vcpu->arch.xsaves_enabled && | |
882 | vcpu->arch.ia32_xss != host_xss) | |
883 | wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss); | |
884 | } | |
37486135 BM |
885 | |
886 | if (static_cpu_has(X86_FEATURE_PKU) && | |
887 | (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) || | |
888 | (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) && | |
889 | vcpu->arch.pkru != vcpu->arch.host_pkru) | |
890 | __write_pkru(vcpu->arch.pkru); | |
42bdf991 | 891 | } |
139a12cf | 892 | EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state); |
42bdf991 | 893 | |
139a12cf | 894 | void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu) |
42bdf991 | 895 | { |
37486135 BM |
896 | if (static_cpu_has(X86_FEATURE_PKU) && |
897 | (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) || | |
898 | (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) { | |
899 | vcpu->arch.pkru = rdpkru(); | |
900 | if (vcpu->arch.pkru != vcpu->arch.host_pkru) | |
901 | __write_pkru(vcpu->arch.host_pkru); | |
902 | } | |
903 | ||
139a12cf AL |
904 | if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) { |
905 | ||
906 | if (vcpu->arch.xcr0 != host_xcr0) | |
907 | xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0); | |
908 | ||
909 | if (vcpu->arch.xsaves_enabled && | |
910 | vcpu->arch.ia32_xss != host_xss) | |
911 | wrmsrl(MSR_IA32_XSS, host_xss); | |
912 | } | |
913 | ||
42bdf991 | 914 | } |
139a12cf | 915 | EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state); |
42bdf991 | 916 | |
69b0049a | 917 | static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr) |
2acf923e | 918 | { |
56c103ec LJ |
919 | u64 xcr0 = xcr; |
920 | u64 old_xcr0 = vcpu->arch.xcr0; | |
46c34cb0 | 921 | u64 valid_bits; |
2acf923e DC |
922 | |
923 | /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */ | |
924 | if (index != XCR_XFEATURE_ENABLED_MASK) | |
925 | return 1; | |
d91cab78 | 926 | if (!(xcr0 & XFEATURE_MASK_FP)) |
2acf923e | 927 | return 1; |
d91cab78 | 928 | if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE)) |
2acf923e | 929 | return 1; |
46c34cb0 PB |
930 | |
931 | /* | |
932 | * Do not allow the guest to set bits that we do not support | |
933 | * saving. However, xcr0 bit 0 is always set, even if the | |
934 | * emulated CPU does not support XSAVE (see fx_init). | |
935 | */ | |
d91cab78 | 936 | valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP; |
46c34cb0 | 937 | if (xcr0 & ~valid_bits) |
2acf923e | 938 | return 1; |
46c34cb0 | 939 | |
d91cab78 DH |
940 | if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) != |
941 | (!(xcr0 & XFEATURE_MASK_BNDCSR))) | |
390bd528 LJ |
942 | return 1; |
943 | ||
d91cab78 DH |
944 | if (xcr0 & XFEATURE_MASK_AVX512) { |
945 | if (!(xcr0 & XFEATURE_MASK_YMM)) | |
612263b3 | 946 | return 1; |
d91cab78 | 947 | if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512) |
612263b3 CP |
948 | return 1; |
949 | } | |
2acf923e | 950 | vcpu->arch.xcr0 = xcr0; |
56c103ec | 951 | |
d91cab78 | 952 | if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND) |
aedbaf4f | 953 | kvm_update_cpuid_runtime(vcpu); |
2acf923e DC |
954 | return 0; |
955 | } | |
956 | ||
957 | int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr) | |
958 | { | |
afaf0b2f | 959 | if (kvm_x86_ops.get_cpl(vcpu) != 0 || |
764bcbc5 | 960 | __kvm_set_xcr(vcpu, index, xcr)) { |
2acf923e DC |
961 | kvm_inject_gp(vcpu, 0); |
962 | return 1; | |
963 | } | |
964 | return 0; | |
965 | } | |
966 | EXPORT_SYMBOL_GPL(kvm_set_xcr); | |
967 | ||
761e4169 | 968 | int kvm_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4) |
a03490ed | 969 | { |
b11306b5 | 970 | if (cr4 & cr4_reserved_bits) |
3ca94192 | 971 | return -EINVAL; |
b9baba86 | 972 | |
b899c132 | 973 | if (cr4 & vcpu->arch.cr4_guest_rsvd_bits) |
3ca94192 WL |
974 | return -EINVAL; |
975 | ||
976 | return 0; | |
977 | } | |
761e4169 | 978 | EXPORT_SYMBOL_GPL(kvm_valid_cr4); |
3ca94192 WL |
979 | |
980 | int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4) | |
981 | { | |
982 | unsigned long old_cr4 = kvm_read_cr4(vcpu); | |
983 | unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE | | |
cb957adb | 984 | X86_CR4_SMEP; |
3ca94192 WL |
985 | |
986 | if (kvm_valid_cr4(vcpu, cr4)) | |
ae3e61e1 PB |
987 | return 1; |
988 | ||
a03490ed | 989 | if (is_long_mode(vcpu)) { |
0f12244f GN |
990 | if (!(cr4 & X86_CR4_PAE)) |
991 | return 1; | |
d74fcfc1 SC |
992 | if ((cr4 ^ old_cr4) & X86_CR4_LA57) |
993 | return 1; | |
a2edf57f AK |
994 | } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE) |
995 | && ((cr4 ^ old_cr4) & pdptr_bits) | |
9f8fe504 AK |
996 | && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, |
997 | kvm_read_cr3(vcpu))) | |
0f12244f GN |
998 | return 1; |
999 | ||
ad756a16 | 1000 | if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) { |
d6321d49 | 1001 | if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID)) |
ad756a16 MJ |
1002 | return 1; |
1003 | ||
1004 | /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */ | |
1005 | if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu)) | |
1006 | return 1; | |
1007 | } | |
1008 | ||
afaf0b2f | 1009 | if (kvm_x86_ops.set_cr4(vcpu, cr4)) |
0f12244f | 1010 | return 1; |
a03490ed | 1011 | |
ad756a16 MJ |
1012 | if (((cr4 ^ old_cr4) & pdptr_bits) || |
1013 | (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE))) | |
aad82703 | 1014 | kvm_mmu_reset_context(vcpu); |
0f12244f | 1015 | |
b9baba86 | 1016 | if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE)) |
aedbaf4f | 1017 | kvm_update_cpuid_runtime(vcpu); |
2acf923e | 1018 | |
0f12244f GN |
1019 | return 0; |
1020 | } | |
2d3ad1f4 | 1021 | EXPORT_SYMBOL_GPL(kvm_set_cr4); |
a03490ed | 1022 | |
2390218b | 1023 | int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3) |
a03490ed | 1024 | { |
ade61e28 | 1025 | bool skip_tlb_flush = false; |
ac146235 | 1026 | #ifdef CONFIG_X86_64 |
c19986fe JS |
1027 | bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE); |
1028 | ||
ade61e28 | 1029 | if (pcid_enabled) { |
208320ba JS |
1030 | skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH; |
1031 | cr3 &= ~X86_CR3_PCID_NOFLUSH; | |
ade61e28 | 1032 | } |
ac146235 | 1033 | #endif |
9d88fca7 | 1034 | |
9f8fe504 | 1035 | if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) { |
956bf353 JS |
1036 | if (!skip_tlb_flush) { |
1037 | kvm_mmu_sync_roots(vcpu); | |
eeeb4f67 | 1038 | kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu); |
956bf353 | 1039 | } |
0f12244f | 1040 | return 0; |
d835dfec AK |
1041 | } |
1042 | ||
d1cd3ce9 | 1043 | if (is_long_mode(vcpu) && |
a780a3ea | 1044 | (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63))) |
d1cd3ce9 | 1045 | return 1; |
bf03d4f9 PB |
1046 | else if (is_pae_paging(vcpu) && |
1047 | !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) | |
346874c9 | 1048 | return 1; |
a03490ed | 1049 | |
be01e8e2 | 1050 | kvm_mmu_new_pgd(vcpu, cr3, skip_tlb_flush, skip_tlb_flush); |
0f12244f | 1051 | vcpu->arch.cr3 = cr3; |
cb3c1e2f | 1052 | kvm_register_mark_available(vcpu, VCPU_EXREG_CR3); |
7c390d35 | 1053 | |
0f12244f GN |
1054 | return 0; |
1055 | } | |
2d3ad1f4 | 1056 | EXPORT_SYMBOL_GPL(kvm_set_cr3); |
a03490ed | 1057 | |
eea1cff9 | 1058 | int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8) |
a03490ed | 1059 | { |
0f12244f GN |
1060 | if (cr8 & CR8_RESERVED_BITS) |
1061 | return 1; | |
35754c98 | 1062 | if (lapic_in_kernel(vcpu)) |
a03490ed CO |
1063 | kvm_lapic_set_tpr(vcpu, cr8); |
1064 | else | |
ad312c7c | 1065 | vcpu->arch.cr8 = cr8; |
0f12244f GN |
1066 | return 0; |
1067 | } | |
2d3ad1f4 | 1068 | EXPORT_SYMBOL_GPL(kvm_set_cr8); |
a03490ed | 1069 | |
2d3ad1f4 | 1070 | unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu) |
a03490ed | 1071 | { |
35754c98 | 1072 | if (lapic_in_kernel(vcpu)) |
a03490ed CO |
1073 | return kvm_lapic_get_cr8(vcpu); |
1074 | else | |
ad312c7c | 1075 | return vcpu->arch.cr8; |
a03490ed | 1076 | } |
2d3ad1f4 | 1077 | EXPORT_SYMBOL_GPL(kvm_get_cr8); |
a03490ed | 1078 | |
ae561ede NA |
1079 | static void kvm_update_dr0123(struct kvm_vcpu *vcpu) |
1080 | { | |
1081 | int i; | |
1082 | ||
1083 | if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) { | |
1084 | for (i = 0; i < KVM_NR_DB_REGS; i++) | |
1085 | vcpu->arch.eff_db[i] = vcpu->arch.db[i]; | |
1086 | vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD; | |
1087 | } | |
1088 | } | |
1089 | ||
7c86663b | 1090 | void kvm_update_dr7(struct kvm_vcpu *vcpu) |
c8639010 JK |
1091 | { |
1092 | unsigned long dr7; | |
1093 | ||
1094 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) | |
1095 | dr7 = vcpu->arch.guest_debug_dr7; | |
1096 | else | |
1097 | dr7 = vcpu->arch.dr7; | |
afaf0b2f | 1098 | kvm_x86_ops.set_dr7(vcpu, dr7); |
360b948d PB |
1099 | vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED; |
1100 | if (dr7 & DR7_BP_EN_MASK) | |
1101 | vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED; | |
c8639010 | 1102 | } |
7c86663b | 1103 | EXPORT_SYMBOL_GPL(kvm_update_dr7); |
c8639010 | 1104 | |
6f43ed01 NA |
1105 | static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu) |
1106 | { | |
1107 | u64 fixed = DR6_FIXED_1; | |
1108 | ||
d6321d49 | 1109 | if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM)) |
6f43ed01 NA |
1110 | fixed |= DR6_RTM; |
1111 | return fixed; | |
1112 | } | |
1113 | ||
338dbc97 | 1114 | static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val) |
020df079 | 1115 | { |
ea740059 MP |
1116 | size_t size = ARRAY_SIZE(vcpu->arch.db); |
1117 | ||
020df079 GN |
1118 | switch (dr) { |
1119 | case 0 ... 3: | |
ea740059 | 1120 | vcpu->arch.db[array_index_nospec(dr, size)] = val; |
020df079 GN |
1121 | if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) |
1122 | vcpu->arch.eff_db[dr] = val; | |
1123 | break; | |
1124 | case 4: | |
020df079 | 1125 | case 6: |
f5f6145e | 1126 | if (!kvm_dr6_valid(val)) |
338dbc97 | 1127 | return -1; /* #GP */ |
6f43ed01 | 1128 | vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu); |
020df079 GN |
1129 | break; |
1130 | case 5: | |
020df079 | 1131 | default: /* 7 */ |
b91991bf | 1132 | if (!kvm_dr7_valid(val)) |
338dbc97 | 1133 | return -1; /* #GP */ |
020df079 | 1134 | vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1; |
c8639010 | 1135 | kvm_update_dr7(vcpu); |
020df079 GN |
1136 | break; |
1137 | } | |
1138 | ||
1139 | return 0; | |
1140 | } | |
338dbc97 GN |
1141 | |
1142 | int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val) | |
1143 | { | |
16f8a6f9 | 1144 | if (__kvm_set_dr(vcpu, dr, val)) { |
338dbc97 | 1145 | kvm_inject_gp(vcpu, 0); |
16f8a6f9 NA |
1146 | return 1; |
1147 | } | |
1148 | return 0; | |
338dbc97 | 1149 | } |
020df079 GN |
1150 | EXPORT_SYMBOL_GPL(kvm_set_dr); |
1151 | ||
16f8a6f9 | 1152 | int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val) |
020df079 | 1153 | { |
ea740059 MP |
1154 | size_t size = ARRAY_SIZE(vcpu->arch.db); |
1155 | ||
020df079 GN |
1156 | switch (dr) { |
1157 | case 0 ... 3: | |
ea740059 | 1158 | *val = vcpu->arch.db[array_index_nospec(dr, size)]; |
020df079 GN |
1159 | break; |
1160 | case 4: | |
020df079 | 1161 | case 6: |
5679b803 | 1162 | *val = vcpu->arch.dr6; |
020df079 GN |
1163 | break; |
1164 | case 5: | |
020df079 GN |
1165 | default: /* 7 */ |
1166 | *val = vcpu->arch.dr7; | |
1167 | break; | |
1168 | } | |
338dbc97 GN |
1169 | return 0; |
1170 | } | |
020df079 GN |
1171 | EXPORT_SYMBOL_GPL(kvm_get_dr); |
1172 | ||
022cd0e8 AK |
1173 | bool kvm_rdpmc(struct kvm_vcpu *vcpu) |
1174 | { | |
de3cd117 | 1175 | u32 ecx = kvm_rcx_read(vcpu); |
022cd0e8 AK |
1176 | u64 data; |
1177 | int err; | |
1178 | ||
c6702c9d | 1179 | err = kvm_pmu_rdpmc(vcpu, ecx, &data); |
022cd0e8 AK |
1180 | if (err) |
1181 | return err; | |
de3cd117 SC |
1182 | kvm_rax_write(vcpu, (u32)data); |
1183 | kvm_rdx_write(vcpu, data >> 32); | |
022cd0e8 AK |
1184 | return err; |
1185 | } | |
1186 | EXPORT_SYMBOL_GPL(kvm_rdpmc); | |
1187 | ||
043405e1 CO |
1188 | /* |
1189 | * List of msr numbers which we expose to userspace through KVM_GET_MSRS | |
1190 | * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST. | |
1191 | * | |
7a5ee6ed CQ |
1192 | * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features) |
1193 | * extract the supported MSRs from the related const lists. | |
1194 | * msrs_to_save is selected from the msrs_to_save_all to reflect the | |
e3267cbb | 1195 | * capabilities of the host cpu. This capabilities test skips MSRs that are |
7a5ee6ed | 1196 | * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs |
62ef68bb | 1197 | * may depend on host virtualization features rather than host cpu features. |
043405e1 | 1198 | */ |
e3267cbb | 1199 | |
7a5ee6ed | 1200 | static const u32 msrs_to_save_all[] = { |
043405e1 | 1201 | MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP, |
8c06585d | 1202 | MSR_STAR, |
043405e1 CO |
1203 | #ifdef CONFIG_X86_64 |
1204 | MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR, | |
1205 | #endif | |
b3897a49 | 1206 | MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA, |
32ad73db | 1207 | MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX, |
2bdb76c0 | 1208 | MSR_IA32_SPEC_CTRL, |
bf8c55d8 CP |
1209 | MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH, |
1210 | MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK, | |
1211 | MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B, | |
1212 | MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B, | |
1213 | MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B, | |
1214 | MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B, | |
6e3ba4ab TX |
1215 | MSR_IA32_UMWAIT_CONTROL, |
1216 | ||
e2ada66e JM |
1217 | MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1, |
1218 | MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3, | |
1219 | MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS, | |
1220 | MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL, | |
1221 | MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1, | |
1222 | MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3, | |
1223 | MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5, | |
1224 | MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7, | |
1225 | MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9, | |
1226 | MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11, | |
1227 | MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13, | |
1228 | MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15, | |
1229 | MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17, | |
e2ada66e JM |
1230 | MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1, |
1231 | MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3, | |
1232 | MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5, | |
1233 | MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7, | |
1234 | MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9, | |
1235 | MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11, | |
1236 | MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13, | |
1237 | MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15, | |
1238 | MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17, | |
043405e1 CO |
1239 | }; |
1240 | ||
7a5ee6ed | 1241 | static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)]; |
043405e1 CO |
1242 | static unsigned num_msrs_to_save; |
1243 | ||
7a5ee6ed | 1244 | static const u32 emulated_msrs_all[] = { |
62ef68bb PB |
1245 | MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK, |
1246 | MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW, | |
1247 | HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL, | |
1248 | HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC, | |
72c139ba | 1249 | HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY, |
e7d9513b AS |
1250 | HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2, |
1251 | HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL, | |
e516cebb | 1252 | HV_X64_MSR_RESET, |
11c4b1ca | 1253 | HV_X64_MSR_VP_INDEX, |
9eec50b8 | 1254 | HV_X64_MSR_VP_RUNTIME, |
5c919412 | 1255 | HV_X64_MSR_SCONTROL, |
1f4b34f8 | 1256 | HV_X64_MSR_STIMER0_CONFIG, |
d4abc577 | 1257 | HV_X64_MSR_VP_ASSIST_PAGE, |
a2e164e7 VK |
1258 | HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL, |
1259 | HV_X64_MSR_TSC_EMULATION_STATUS, | |
f97f5a56 JD |
1260 | HV_X64_MSR_SYNDBG_OPTIONS, |
1261 | HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS, | |
1262 | HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER, | |
1263 | HV_X64_MSR_SYNDBG_PENDING_BUFFER, | |
a2e164e7 VK |
1264 | |
1265 | MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME, | |
557a961a | 1266 | MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK, |
62ef68bb | 1267 | |
ba904635 | 1268 | MSR_IA32_TSC_ADJUST, |
a3e06bbe | 1269 | MSR_IA32_TSCDEADLINE, |
2bdb76c0 | 1270 | MSR_IA32_ARCH_CAPABILITIES, |
27461da3 | 1271 | MSR_IA32_PERF_CAPABILITIES, |
043405e1 | 1272 | MSR_IA32_MISC_ENABLE, |
908e75f3 AK |
1273 | MSR_IA32_MCG_STATUS, |
1274 | MSR_IA32_MCG_CTL, | |
c45dcc71 | 1275 | MSR_IA32_MCG_EXT_CTL, |
64d60670 | 1276 | MSR_IA32_SMBASE, |
52797bf9 | 1277 | MSR_SMI_COUNT, |
db2336a8 KH |
1278 | MSR_PLATFORM_INFO, |
1279 | MSR_MISC_FEATURES_ENABLES, | |
bc226f07 | 1280 | MSR_AMD64_VIRT_SPEC_CTRL, |
6c6a2ab9 | 1281 | MSR_IA32_POWER_CTL, |
99634e3e | 1282 | MSR_IA32_UCODE_REV, |
191c8137 | 1283 | |
95c5c7c7 PB |
1284 | /* |
1285 | * The following list leaves out MSRs whose values are determined | |
1286 | * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs. | |
1287 | * We always support the "true" VMX control MSRs, even if the host | |
1288 | * processor does not, so I am putting these registers here rather | |
7a5ee6ed | 1289 | * than in msrs_to_save_all. |
95c5c7c7 PB |
1290 | */ |
1291 | MSR_IA32_VMX_BASIC, | |
1292 | MSR_IA32_VMX_TRUE_PINBASED_CTLS, | |
1293 | MSR_IA32_VMX_TRUE_PROCBASED_CTLS, | |
1294 | MSR_IA32_VMX_TRUE_EXIT_CTLS, | |
1295 | MSR_IA32_VMX_TRUE_ENTRY_CTLS, | |
1296 | MSR_IA32_VMX_MISC, | |
1297 | MSR_IA32_VMX_CR0_FIXED0, | |
1298 | MSR_IA32_VMX_CR4_FIXED0, | |
1299 | MSR_IA32_VMX_VMCS_ENUM, | |
1300 | MSR_IA32_VMX_PROCBASED_CTLS2, | |
1301 | MSR_IA32_VMX_EPT_VPID_CAP, | |
1302 | MSR_IA32_VMX_VMFUNC, | |
1303 | ||
191c8137 | 1304 | MSR_K7_HWCR, |
2d5ba19b | 1305 | MSR_KVM_POLL_CONTROL, |
043405e1 CO |
1306 | }; |
1307 | ||
7a5ee6ed | 1308 | static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)]; |
62ef68bb PB |
1309 | static unsigned num_emulated_msrs; |
1310 | ||
801e459a TL |
1311 | /* |
1312 | * List of msr numbers which are used to expose MSR-based features that | |
1313 | * can be used by a hypervisor to validate requested CPU features. | |
1314 | */ | |
7a5ee6ed | 1315 | static const u32 msr_based_features_all[] = { |
1389309c PB |
1316 | MSR_IA32_VMX_BASIC, |
1317 | MSR_IA32_VMX_TRUE_PINBASED_CTLS, | |
1318 | MSR_IA32_VMX_PINBASED_CTLS, | |
1319 | MSR_IA32_VMX_TRUE_PROCBASED_CTLS, | |
1320 | MSR_IA32_VMX_PROCBASED_CTLS, | |
1321 | MSR_IA32_VMX_TRUE_EXIT_CTLS, | |
1322 | MSR_IA32_VMX_EXIT_CTLS, | |
1323 | MSR_IA32_VMX_TRUE_ENTRY_CTLS, | |
1324 | MSR_IA32_VMX_ENTRY_CTLS, | |
1325 | MSR_IA32_VMX_MISC, | |
1326 | MSR_IA32_VMX_CR0_FIXED0, | |
1327 | MSR_IA32_VMX_CR0_FIXED1, | |
1328 | MSR_IA32_VMX_CR4_FIXED0, | |
1329 | MSR_IA32_VMX_CR4_FIXED1, | |
1330 | MSR_IA32_VMX_VMCS_ENUM, | |
1331 | MSR_IA32_VMX_PROCBASED_CTLS2, | |
1332 | MSR_IA32_VMX_EPT_VPID_CAP, | |
1333 | MSR_IA32_VMX_VMFUNC, | |
1334 | ||
d1d93fa9 | 1335 | MSR_F10H_DECFG, |
518e7b94 | 1336 | MSR_IA32_UCODE_REV, |
cd283252 | 1337 | MSR_IA32_ARCH_CAPABILITIES, |
27461da3 | 1338 | MSR_IA32_PERF_CAPABILITIES, |
801e459a TL |
1339 | }; |
1340 | ||
7a5ee6ed | 1341 | static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)]; |
801e459a TL |
1342 | static unsigned int num_msr_based_features; |
1343 | ||
4d22c17c | 1344 | static u64 kvm_get_arch_capabilities(void) |
5b76a3cf | 1345 | { |
4d22c17c | 1346 | u64 data = 0; |
5b76a3cf | 1347 | |
4d22c17c XL |
1348 | if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) |
1349 | rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data); | |
5b76a3cf | 1350 | |
b8e8c830 PB |
1351 | /* |
1352 | * If nx_huge_pages is enabled, KVM's shadow paging will ensure that | |
1353 | * the nested hypervisor runs with NX huge pages. If it is not, | |
1354 | * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other | |
1355 | * L1 guests, so it need not worry about its own (L2) guests. | |
1356 | */ | |
1357 | data |= ARCH_CAP_PSCHANGE_MC_NO; | |
1358 | ||
5b76a3cf PB |
1359 | /* |
1360 | * If we're doing cache flushes (either "always" or "cond") | |
1361 | * we will do one whenever the guest does a vmlaunch/vmresume. | |
1362 | * If an outer hypervisor is doing the cache flush for us | |
1363 | * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that | |
1364 | * capability to the guest too, and if EPT is disabled we're not | |
1365 | * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will | |
1366 | * require a nested hypervisor to do a flush of its own. | |
1367 | */ | |
1368 | if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER) | |
1369 | data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH; | |
1370 | ||
0c54914d PB |
1371 | if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN)) |
1372 | data |= ARCH_CAP_RDCL_NO; | |
1373 | if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS)) | |
1374 | data |= ARCH_CAP_SSB_NO; | |
1375 | if (!boot_cpu_has_bug(X86_BUG_MDS)) | |
1376 | data |= ARCH_CAP_MDS_NO; | |
1377 | ||
e1d38b63 | 1378 | /* |
c11f83e0 PB |
1379 | * On TAA affected systems: |
1380 | * - nothing to do if TSX is disabled on the host. | |
1381 | * - we emulate TSX_CTRL if present on the host. | |
1382 | * This lets the guest use VERW to clear CPU buffers. | |
e1d38b63 | 1383 | */ |
cbbaa272 | 1384 | if (!boot_cpu_has(X86_FEATURE_RTM)) |
c11f83e0 | 1385 | data &= ~(ARCH_CAP_TAA_NO | ARCH_CAP_TSX_CTRL_MSR); |
cbbaa272 PB |
1386 | else if (!boot_cpu_has_bug(X86_BUG_TAA)) |
1387 | data |= ARCH_CAP_TAA_NO; | |
e1d38b63 | 1388 | |
5b76a3cf PB |
1389 | return data; |
1390 | } | |
5b76a3cf | 1391 | |
66421c1e WL |
1392 | static int kvm_get_msr_feature(struct kvm_msr_entry *msr) |
1393 | { | |
1394 | switch (msr->index) { | |
cd283252 | 1395 | case MSR_IA32_ARCH_CAPABILITIES: |
5b76a3cf PB |
1396 | msr->data = kvm_get_arch_capabilities(); |
1397 | break; | |
1398 | case MSR_IA32_UCODE_REV: | |
cd283252 | 1399 | rdmsrl_safe(msr->index, &msr->data); |
518e7b94 | 1400 | break; |
66421c1e | 1401 | default: |
12bc2132 | 1402 | return kvm_x86_ops.get_msr_feature(msr); |
66421c1e WL |
1403 | } |
1404 | return 0; | |
1405 | } | |
1406 | ||
801e459a TL |
1407 | static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data) |
1408 | { | |
1409 | struct kvm_msr_entry msr; | |
66421c1e | 1410 | int r; |
801e459a TL |
1411 | |
1412 | msr.index = index; | |
66421c1e | 1413 | r = kvm_get_msr_feature(&msr); |
12bc2132 PX |
1414 | |
1415 | if (r == KVM_MSR_RET_INVALID) { | |
1416 | /* Unconditionally clear the output for simplicity */ | |
1417 | *data = 0; | |
1418 | r = kvm_msr_ignored_check(vcpu, index, 0, false); | |
1419 | } | |
1420 | ||
66421c1e WL |
1421 | if (r) |
1422 | return r; | |
801e459a TL |
1423 | |
1424 | *data = msr.data; | |
1425 | ||
1426 | return 0; | |
1427 | } | |
1428 | ||
11988499 | 1429 | static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer) |
15c4a640 | 1430 | { |
1b4d56b8 | 1431 | if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT)) |
11988499 | 1432 | return false; |
1b2fd70c | 1433 | |
1b4d56b8 | 1434 | if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM)) |
11988499 | 1435 | return false; |
d8017474 | 1436 | |
0a629563 SC |
1437 | if (efer & (EFER_LME | EFER_LMA) && |
1438 | !guest_cpuid_has(vcpu, X86_FEATURE_LM)) | |
1439 | return false; | |
1440 | ||
1441 | if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX)) | |
1442 | return false; | |
d8017474 | 1443 | |
384bb783 | 1444 | return true; |
11988499 SC |
1445 | |
1446 | } | |
1447 | bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer) | |
1448 | { | |
1449 | if (efer & efer_reserved_bits) | |
1450 | return false; | |
1451 | ||
1452 | return __kvm_valid_efer(vcpu, efer); | |
384bb783 JK |
1453 | } |
1454 | EXPORT_SYMBOL_GPL(kvm_valid_efer); | |
1455 | ||
11988499 | 1456 | static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
384bb783 JK |
1457 | { |
1458 | u64 old_efer = vcpu->arch.efer; | |
11988499 | 1459 | u64 efer = msr_info->data; |
384bb783 | 1460 | |
11988499 | 1461 | if (efer & efer_reserved_bits) |
66f61c92 | 1462 | return 1; |
384bb783 | 1463 | |
11988499 SC |
1464 | if (!msr_info->host_initiated) { |
1465 | if (!__kvm_valid_efer(vcpu, efer)) | |
1466 | return 1; | |
1467 | ||
1468 | if (is_paging(vcpu) && | |
1469 | (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME)) | |
1470 | return 1; | |
1471 | } | |
384bb783 | 1472 | |
15c4a640 | 1473 | efer &= ~EFER_LMA; |
f6801dff | 1474 | efer |= vcpu->arch.efer & EFER_LMA; |
15c4a640 | 1475 | |
afaf0b2f | 1476 | kvm_x86_ops.set_efer(vcpu, efer); |
a3d204e2 | 1477 | |
aad82703 SY |
1478 | /* Update reserved bits */ |
1479 | if ((efer ^ old_efer) & EFER_NX) | |
1480 | kvm_mmu_reset_context(vcpu); | |
1481 | ||
b69e8cae | 1482 | return 0; |
15c4a640 CO |
1483 | } |
1484 | ||
f2b4b7dd JR |
1485 | void kvm_enable_efer_bits(u64 mask) |
1486 | { | |
1487 | efer_reserved_bits &= ~mask; | |
1488 | } | |
1489 | EXPORT_SYMBOL_GPL(kvm_enable_efer_bits); | |
1490 | ||
51de8151 AG |
1491 | bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type) |
1492 | { | |
1a155254 AG |
1493 | struct kvm *kvm = vcpu->kvm; |
1494 | struct msr_bitmap_range *ranges = kvm->arch.msr_filter.ranges; | |
1495 | u32 count = kvm->arch.msr_filter.count; | |
1496 | u32 i; | |
1497 | bool r = kvm->arch.msr_filter.default_allow; | |
1498 | int idx; | |
1499 | ||
9389b9d5 SC |
1500 | /* MSR filtering not set up or x2APIC enabled, allow everything */ |
1501 | if (!count || (index >= 0x800 && index <= 0x8ff)) | |
1a155254 AG |
1502 | return true; |
1503 | ||
1504 | /* Prevent collision with set_msr_filter */ | |
1505 | idx = srcu_read_lock(&kvm->srcu); | |
1506 | ||
1507 | for (i = 0; i < count; i++) { | |
1508 | u32 start = ranges[i].base; | |
1509 | u32 end = start + ranges[i].nmsrs; | |
1510 | u32 flags = ranges[i].flags; | |
1511 | unsigned long *bitmap = ranges[i].bitmap; | |
1512 | ||
1513 | if ((index >= start) && (index < end) && (flags & type)) { | |
1514 | r = !!test_bit(index - start, bitmap); | |
1515 | break; | |
1516 | } | |
1517 | } | |
1518 | ||
1519 | srcu_read_unlock(&kvm->srcu, idx); | |
1520 | ||
1521 | return r; | |
51de8151 AG |
1522 | } |
1523 | EXPORT_SYMBOL_GPL(kvm_msr_allowed); | |
1524 | ||
15c4a640 | 1525 | /* |
f20935d8 SC |
1526 | * Write @data into the MSR specified by @index. Select MSR specific fault |
1527 | * checks are bypassed if @host_initiated is %true. | |
15c4a640 CO |
1528 | * Returns 0 on success, non-0 otherwise. |
1529 | * Assumes vcpu_load() was already called. | |
1530 | */ | |
f20935d8 SC |
1531 | static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data, |
1532 | bool host_initiated) | |
15c4a640 | 1533 | { |
f20935d8 SC |
1534 | struct msr_data msr; |
1535 | ||
1a155254 AG |
1536 | if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE)) |
1537 | return -EPERM; | |
1538 | ||
f20935d8 | 1539 | switch (index) { |
854e8bb1 NA |
1540 | case MSR_FS_BASE: |
1541 | case MSR_GS_BASE: | |
1542 | case MSR_KERNEL_GS_BASE: | |
1543 | case MSR_CSTAR: | |
1544 | case MSR_LSTAR: | |
f20935d8 | 1545 | if (is_noncanonical_address(data, vcpu)) |
854e8bb1 NA |
1546 | return 1; |
1547 | break; | |
1548 | case MSR_IA32_SYSENTER_EIP: | |
1549 | case MSR_IA32_SYSENTER_ESP: | |
1550 | /* | |
1551 | * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if | |
1552 | * non-canonical address is written on Intel but not on | |
1553 | * AMD (which ignores the top 32-bits, because it does | |
1554 | * not implement 64-bit SYSENTER). | |
1555 | * | |
1556 | * 64-bit code should hence be able to write a non-canonical | |
1557 | * value on AMD. Making the address canonical ensures that | |
1558 | * vmentry does not fail on Intel after writing a non-canonical | |
1559 | * value, and that something deterministic happens if the guest | |
1560 | * invokes 64-bit SYSENTER. | |
1561 | */ | |
f20935d8 | 1562 | data = get_canonical(data, vcpu_virt_addr_bits(vcpu)); |
854e8bb1 | 1563 | } |
f20935d8 SC |
1564 | |
1565 | msr.data = data; | |
1566 | msr.index = index; | |
1567 | msr.host_initiated = host_initiated; | |
1568 | ||
afaf0b2f | 1569 | return kvm_x86_ops.set_msr(vcpu, &msr); |
15c4a640 CO |
1570 | } |
1571 | ||
6abe9c13 PX |
1572 | static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu, |
1573 | u32 index, u64 data, bool host_initiated) | |
1574 | { | |
1575 | int ret = __kvm_set_msr(vcpu, index, data, host_initiated); | |
1576 | ||
1577 | if (ret == KVM_MSR_RET_INVALID) | |
1578 | ret = kvm_msr_ignored_check(vcpu, index, data, true); | |
1579 | ||
1580 | return ret; | |
1581 | } | |
1582 | ||
313a3dc7 | 1583 | /* |
f20935d8 SC |
1584 | * Read the MSR specified by @index into @data. Select MSR specific fault |
1585 | * checks are bypassed if @host_initiated is %true. | |
1586 | * Returns 0 on success, non-0 otherwise. | |
1587 | * Assumes vcpu_load() was already called. | |
313a3dc7 | 1588 | */ |
edef5c36 PB |
1589 | int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data, |
1590 | bool host_initiated) | |
609e36d3 PB |
1591 | { |
1592 | struct msr_data msr; | |
f20935d8 | 1593 | int ret; |
609e36d3 | 1594 | |
1a155254 AG |
1595 | if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ)) |
1596 | return -EPERM; | |
1597 | ||
609e36d3 | 1598 | msr.index = index; |
f20935d8 | 1599 | msr.host_initiated = host_initiated; |
609e36d3 | 1600 | |
afaf0b2f | 1601 | ret = kvm_x86_ops.get_msr(vcpu, &msr); |
f20935d8 SC |
1602 | if (!ret) |
1603 | *data = msr.data; | |
1604 | return ret; | |
609e36d3 PB |
1605 | } |
1606 | ||
6abe9c13 PX |
1607 | static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu, |
1608 | u32 index, u64 *data, bool host_initiated) | |
1609 | { | |
1610 | int ret = __kvm_get_msr(vcpu, index, data, host_initiated); | |
1611 | ||
1612 | if (ret == KVM_MSR_RET_INVALID) { | |
1613 | /* Unconditionally clear *data for simplicity */ | |
1614 | *data = 0; | |
1615 | ret = kvm_msr_ignored_check(vcpu, index, 0, false); | |
1616 | } | |
1617 | ||
1618 | return ret; | |
1619 | } | |
1620 | ||
f20935d8 | 1621 | int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data) |
313a3dc7 | 1622 | { |
6abe9c13 | 1623 | return kvm_get_msr_ignored_check(vcpu, index, data, false); |
f20935d8 SC |
1624 | } |
1625 | EXPORT_SYMBOL_GPL(kvm_get_msr); | |
8fe8ab46 | 1626 | |
f20935d8 SC |
1627 | int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data) |
1628 | { | |
6abe9c13 | 1629 | return kvm_set_msr_ignored_check(vcpu, index, data, false); |
f20935d8 SC |
1630 | } |
1631 | EXPORT_SYMBOL_GPL(kvm_set_msr); | |
1632 | ||
1ae09954 AG |
1633 | static int complete_emulated_msr(struct kvm_vcpu *vcpu, bool is_read) |
1634 | { | |
1635 | if (vcpu->run->msr.error) { | |
1636 | kvm_inject_gp(vcpu, 0); | |
1637 | return 1; | |
1638 | } else if (is_read) { | |
1639 | kvm_rax_write(vcpu, (u32)vcpu->run->msr.data); | |
1640 | kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32); | |
1641 | } | |
1642 | ||
1643 | return kvm_skip_emulated_instruction(vcpu); | |
1644 | } | |
1645 | ||
1646 | static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu) | |
1647 | { | |
1648 | return complete_emulated_msr(vcpu, true); | |
1649 | } | |
1650 | ||
1651 | static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu) | |
1652 | { | |
1653 | return complete_emulated_msr(vcpu, false); | |
1654 | } | |
1655 | ||
1656 | static u64 kvm_msr_reason(int r) | |
1657 | { | |
1658 | switch (r) { | |
1659 | case -ENOENT: | |
1660 | return KVM_MSR_EXIT_REASON_UNKNOWN; | |
1a155254 AG |
1661 | case -EPERM: |
1662 | return KVM_MSR_EXIT_REASON_FILTER; | |
1ae09954 AG |
1663 | default: |
1664 | return KVM_MSR_EXIT_REASON_INVAL; | |
1665 | } | |
1666 | } | |
1667 | ||
1668 | static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index, | |
1669 | u32 exit_reason, u64 data, | |
1670 | int (*completion)(struct kvm_vcpu *vcpu), | |
1671 | int r) | |
1672 | { | |
1673 | u64 msr_reason = kvm_msr_reason(r); | |
1674 | ||
1675 | /* Check if the user wanted to know about this MSR fault */ | |
1676 | if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason)) | |
1677 | return 0; | |
1678 | ||
1679 | vcpu->run->exit_reason = exit_reason; | |
1680 | vcpu->run->msr.error = 0; | |
1681 | memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad)); | |
1682 | vcpu->run->msr.reason = msr_reason; | |
1683 | vcpu->run->msr.index = index; | |
1684 | vcpu->run->msr.data = data; | |
1685 | vcpu->arch.complete_userspace_io = completion; | |
1686 | ||
1687 | return 1; | |
1688 | } | |
1689 | ||
1690 | static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r) | |
1691 | { | |
1692 | return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0, | |
1693 | complete_emulated_rdmsr, r); | |
1694 | } | |
1695 | ||
1696 | static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r) | |
1697 | { | |
1698 | return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data, | |
1699 | complete_emulated_wrmsr, r); | |
1700 | } | |
1701 | ||
1edce0a9 SC |
1702 | int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu) |
1703 | { | |
1704 | u32 ecx = kvm_rcx_read(vcpu); | |
1705 | u64 data; | |
1ae09954 AG |
1706 | int r; |
1707 | ||
1708 | r = kvm_get_msr(vcpu, ecx, &data); | |
1edce0a9 | 1709 | |
1ae09954 AG |
1710 | /* MSR read failed? See if we should ask user space */ |
1711 | if (r && kvm_get_msr_user_space(vcpu, ecx, r)) { | |
1712 | /* Bounce to user space */ | |
1713 | return 0; | |
1714 | } | |
1715 | ||
1716 | /* MSR read failed? Inject a #GP */ | |
1717 | if (r) { | |
1edce0a9 SC |
1718 | trace_kvm_msr_read_ex(ecx); |
1719 | kvm_inject_gp(vcpu, 0); | |
1720 | return 1; | |
1721 | } | |
1722 | ||
1723 | trace_kvm_msr_read(ecx, data); | |
1724 | ||
1725 | kvm_rax_write(vcpu, data & -1u); | |
1726 | kvm_rdx_write(vcpu, (data >> 32) & -1u); | |
1727 | return kvm_skip_emulated_instruction(vcpu); | |
1728 | } | |
1729 | EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr); | |
1730 | ||
1731 | int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu) | |
1732 | { | |
1733 | u32 ecx = kvm_rcx_read(vcpu); | |
1734 | u64 data = kvm_read_edx_eax(vcpu); | |
1ae09954 | 1735 | int r; |
1edce0a9 | 1736 | |
1ae09954 AG |
1737 | r = kvm_set_msr(vcpu, ecx, data); |
1738 | ||
1739 | /* MSR write failed? See if we should ask user space */ | |
7dffecaf | 1740 | if (r && kvm_set_msr_user_space(vcpu, ecx, data, r)) |
1ae09954 AG |
1741 | /* Bounce to user space */ |
1742 | return 0; | |
7dffecaf ML |
1743 | |
1744 | /* Signal all other negative errors to userspace */ | |
1745 | if (r < 0) | |
1746 | return r; | |
1ae09954 AG |
1747 | |
1748 | /* MSR write failed? Inject a #GP */ | |
7dffecaf | 1749 | if (r > 0) { |
1edce0a9 SC |
1750 | trace_kvm_msr_write_ex(ecx, data); |
1751 | kvm_inject_gp(vcpu, 0); | |
1752 | return 1; | |
1753 | } | |
1754 | ||
1755 | trace_kvm_msr_write(ecx, data); | |
1756 | return kvm_skip_emulated_instruction(vcpu); | |
1757 | } | |
1758 | EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr); | |
1759 | ||
5a9f5443 WL |
1760 | bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu) |
1761 | { | |
1762 | return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) || | |
72c3c0fe | 1763 | xfer_to_guest_mode_work_pending(); |
5a9f5443 WL |
1764 | } |
1765 | EXPORT_SYMBOL_GPL(kvm_vcpu_exit_request); | |
1766 | ||
1e9e2622 WL |
1767 | /* |
1768 | * The fast path for frequent and performance sensitive wrmsr emulation, | |
1769 | * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces | |
1770 | * the latency of virtual IPI by avoiding the expensive bits of transitioning | |
1771 | * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the | |
1772 | * other cases which must be called after interrupts are enabled on the host. | |
1773 | */ | |
1774 | static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data) | |
1775 | { | |
e1be9ac8 WL |
1776 | if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic)) |
1777 | return 1; | |
1778 | ||
1779 | if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) && | |
1e9e2622 | 1780 | ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) && |
4064a4c6 WL |
1781 | ((data & APIC_MODE_MASK) == APIC_DM_FIXED) && |
1782 | ((u32)(data >> 32) != X2APIC_BROADCAST)) { | |
1e9e2622 | 1783 | |
d5361678 WL |
1784 | data &= ~(1 << 12); |
1785 | kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32)); | |
1e9e2622 | 1786 | kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32)); |
d5361678 WL |
1787 | kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data); |
1788 | trace_kvm_apic_write(APIC_ICR, (u32)data); | |
1789 | return 0; | |
1e9e2622 WL |
1790 | } |
1791 | ||
1792 | return 1; | |
1793 | } | |
1794 | ||
ae95f566 WL |
1795 | static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data) |
1796 | { | |
1797 | if (!kvm_can_use_hv_timer(vcpu)) | |
1798 | return 1; | |
1799 | ||
1800 | kvm_set_lapic_tscdeadline_msr(vcpu, data); | |
1801 | return 0; | |
1802 | } | |
1803 | ||
404d5d7b | 1804 | fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu) |
1e9e2622 WL |
1805 | { |
1806 | u32 msr = kvm_rcx_read(vcpu); | |
8a1038de | 1807 | u64 data; |
404d5d7b | 1808 | fastpath_t ret = EXIT_FASTPATH_NONE; |
1e9e2622 WL |
1809 | |
1810 | switch (msr) { | |
1811 | case APIC_BASE_MSR + (APIC_ICR >> 4): | |
8a1038de | 1812 | data = kvm_read_edx_eax(vcpu); |
404d5d7b WL |
1813 | if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) { |
1814 | kvm_skip_emulated_instruction(vcpu); | |
1815 | ret = EXIT_FASTPATH_EXIT_HANDLED; | |
80bc97f2 | 1816 | } |
1e9e2622 | 1817 | break; |
ae95f566 WL |
1818 | case MSR_IA32_TSCDEADLINE: |
1819 | data = kvm_read_edx_eax(vcpu); | |
1820 | if (!handle_fastpath_set_tscdeadline(vcpu, data)) { | |
1821 | kvm_skip_emulated_instruction(vcpu); | |
1822 | ret = EXIT_FASTPATH_REENTER_GUEST; | |
1823 | } | |
1824 | break; | |
1e9e2622 | 1825 | default: |
404d5d7b | 1826 | break; |
1e9e2622 WL |
1827 | } |
1828 | ||
404d5d7b | 1829 | if (ret != EXIT_FASTPATH_NONE) |
1e9e2622 | 1830 | trace_kvm_msr_write(msr, data); |
1e9e2622 | 1831 | |
404d5d7b | 1832 | return ret; |
1e9e2622 WL |
1833 | } |
1834 | EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff); | |
1835 | ||
f20935d8 SC |
1836 | /* |
1837 | * Adapt set_msr() to msr_io()'s calling convention | |
1838 | */ | |
1839 | static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data) | |
1840 | { | |
6abe9c13 | 1841 | return kvm_get_msr_ignored_check(vcpu, index, data, true); |
f20935d8 SC |
1842 | } |
1843 | ||
1844 | static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data) | |
1845 | { | |
6abe9c13 | 1846 | return kvm_set_msr_ignored_check(vcpu, index, *data, true); |
313a3dc7 CO |
1847 | } |
1848 | ||
16e8d74d | 1849 | #ifdef CONFIG_X86_64 |
53fafdbb MT |
1850 | struct pvclock_clock { |
1851 | int vclock_mode; | |
1852 | u64 cycle_last; | |
1853 | u64 mask; | |
1854 | u32 mult; | |
1855 | u32 shift; | |
917f9475 PB |
1856 | u64 base_cycles; |
1857 | u64 offset; | |
53fafdbb MT |
1858 | }; |
1859 | ||
16e8d74d MT |
1860 | struct pvclock_gtod_data { |
1861 | seqcount_t seq; | |
1862 | ||
53fafdbb MT |
1863 | struct pvclock_clock clock; /* extract of a clocksource struct */ |
1864 | struct pvclock_clock raw_clock; /* extract of a clocksource struct */ | |
16e8d74d | 1865 | |
917f9475 | 1866 | ktime_t offs_boot; |
55dd00a7 | 1867 | u64 wall_time_sec; |
16e8d74d MT |
1868 | }; |
1869 | ||
1870 | static struct pvclock_gtod_data pvclock_gtod_data; | |
1871 | ||
1872 | static void update_pvclock_gtod(struct timekeeper *tk) | |
1873 | { | |
1874 | struct pvclock_gtod_data *vdata = &pvclock_gtod_data; | |
1875 | ||
1876 | write_seqcount_begin(&vdata->seq); | |
1877 | ||
1878 | /* copy pvclock gtod data */ | |
b95a8a27 | 1879 | vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode; |
876e7881 PZ |
1880 | vdata->clock.cycle_last = tk->tkr_mono.cycle_last; |
1881 | vdata->clock.mask = tk->tkr_mono.mask; | |
1882 | vdata->clock.mult = tk->tkr_mono.mult; | |
1883 | vdata->clock.shift = tk->tkr_mono.shift; | |
917f9475 PB |
1884 | vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec; |
1885 | vdata->clock.offset = tk->tkr_mono.base; | |
16e8d74d | 1886 | |
b95a8a27 | 1887 | vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode; |
53fafdbb MT |
1888 | vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last; |
1889 | vdata->raw_clock.mask = tk->tkr_raw.mask; | |
1890 | vdata->raw_clock.mult = tk->tkr_raw.mult; | |
1891 | vdata->raw_clock.shift = tk->tkr_raw.shift; | |
917f9475 PB |
1892 | vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec; |
1893 | vdata->raw_clock.offset = tk->tkr_raw.base; | |
16e8d74d | 1894 | |
55dd00a7 MT |
1895 | vdata->wall_time_sec = tk->xtime_sec; |
1896 | ||
917f9475 | 1897 | vdata->offs_boot = tk->offs_boot; |
53fafdbb | 1898 | |
16e8d74d MT |
1899 | write_seqcount_end(&vdata->seq); |
1900 | } | |
8171cd68 PB |
1901 | |
1902 | static s64 get_kvmclock_base_ns(void) | |
1903 | { | |
1904 | /* Count up from boot time, but with the frequency of the raw clock. */ | |
1905 | return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot)); | |
1906 | } | |
1907 | #else | |
1908 | static s64 get_kvmclock_base_ns(void) | |
1909 | { | |
1910 | /* Master clock not used, so we can just use CLOCK_BOOTTIME. */ | |
1911 | return ktime_get_boottime_ns(); | |
1912 | } | |
16e8d74d MT |
1913 | #endif |
1914 | ||
18068523 GOC |
1915 | static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock) |
1916 | { | |
9ed3c444 AK |
1917 | int version; |
1918 | int r; | |
50d0a0f9 | 1919 | struct pvclock_wall_clock wc; |
8171cd68 | 1920 | u64 wall_nsec; |
18068523 | 1921 | |
210dfd93 OU |
1922 | kvm->arch.wall_clock = wall_clock; |
1923 | ||
18068523 GOC |
1924 | if (!wall_clock) |
1925 | return; | |
1926 | ||
9ed3c444 AK |
1927 | r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version)); |
1928 | if (r) | |
1929 | return; | |
1930 | ||
1931 | if (version & 1) | |
1932 | ++version; /* first time write, random junk */ | |
1933 | ||
1934 | ++version; | |
18068523 | 1935 | |
1dab1345 NK |
1936 | if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version))) |
1937 | return; | |
18068523 | 1938 | |
50d0a0f9 GH |
1939 | /* |
1940 | * The guest calculates current wall clock time by adding | |
34c238a1 | 1941 | * system time (updated by kvm_guest_time_update below) to the |
8171cd68 | 1942 | * wall clock specified here. We do the reverse here. |
50d0a0f9 | 1943 | */ |
8171cd68 | 1944 | wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm); |
50d0a0f9 | 1945 | |
8171cd68 PB |
1946 | wc.nsec = do_div(wall_nsec, 1000000000); |
1947 | wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */ | |
50d0a0f9 | 1948 | wc.version = version; |
18068523 GOC |
1949 | |
1950 | kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc)); | |
1951 | ||
1952 | version++; | |
1953 | kvm_write_guest(kvm, wall_clock, &version, sizeof(version)); | |
18068523 GOC |
1954 | } |
1955 | ||
5b9bb0eb OU |
1956 | static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time, |
1957 | bool old_msr, bool host_initiated) | |
1958 | { | |
1959 | struct kvm_arch *ka = &vcpu->kvm->arch; | |
1960 | ||
1961 | if (vcpu->vcpu_id == 0 && !host_initiated) { | |
1962 | if (ka->boot_vcpu_runs_old_kvmclock && old_msr) | |
1963 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); | |
1964 | ||
1965 | ka->boot_vcpu_runs_old_kvmclock = old_msr; | |
1966 | } | |
1967 | ||
1968 | vcpu->arch.time = system_time; | |
1969 | kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu); | |
1970 | ||
1971 | /* we verify if the enable bit is set... */ | |
1972 | vcpu->arch.pv_time_enabled = false; | |
1973 | if (!(system_time & 1)) | |
1974 | return; | |
1975 | ||
1976 | if (!kvm_gfn_to_hva_cache_init(vcpu->kvm, | |
1977 | &vcpu->arch.pv_time, system_time & ~1ULL, | |
1978 | sizeof(struct pvclock_vcpu_time_info))) | |
1979 | vcpu->arch.pv_time_enabled = true; | |
1980 | ||
1981 | return; | |
1982 | } | |
1983 | ||
50d0a0f9 GH |
1984 | static uint32_t div_frac(uint32_t dividend, uint32_t divisor) |
1985 | { | |
b51012de PB |
1986 | do_shl32_div32(dividend, divisor); |
1987 | return dividend; | |
50d0a0f9 GH |
1988 | } |
1989 | ||
3ae13faa | 1990 | static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz, |
5f4e3f88 | 1991 | s8 *pshift, u32 *pmultiplier) |
50d0a0f9 | 1992 | { |
5f4e3f88 | 1993 | uint64_t scaled64; |
50d0a0f9 GH |
1994 | int32_t shift = 0; |
1995 | uint64_t tps64; | |
1996 | uint32_t tps32; | |
1997 | ||
3ae13faa PB |
1998 | tps64 = base_hz; |
1999 | scaled64 = scaled_hz; | |
50933623 | 2000 | while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) { |
50d0a0f9 GH |
2001 | tps64 >>= 1; |
2002 | shift--; | |
2003 | } | |
2004 | ||
2005 | tps32 = (uint32_t)tps64; | |
50933623 JK |
2006 | while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) { |
2007 | if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000) | |
5f4e3f88 ZA |
2008 | scaled64 >>= 1; |
2009 | else | |
2010 | tps32 <<= 1; | |
50d0a0f9 GH |
2011 | shift++; |
2012 | } | |
2013 | ||
5f4e3f88 ZA |
2014 | *pshift = shift; |
2015 | *pmultiplier = div_frac(scaled64, tps32); | |
50d0a0f9 GH |
2016 | } |
2017 | ||
d828199e | 2018 | #ifdef CONFIG_X86_64 |
16e8d74d | 2019 | static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0); |
d828199e | 2020 | #endif |
16e8d74d | 2021 | |
c8076604 | 2022 | static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz); |
69b0049a | 2023 | static unsigned long max_tsc_khz; |
c8076604 | 2024 | |
cc578287 | 2025 | static u32 adjust_tsc_khz(u32 khz, s32 ppm) |
1e993611 | 2026 | { |
cc578287 ZA |
2027 | u64 v = (u64)khz * (1000000 + ppm); |
2028 | do_div(v, 1000000); | |
2029 | return v; | |
1e993611 JR |
2030 | } |
2031 | ||
381d585c HZ |
2032 | static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale) |
2033 | { | |
2034 | u64 ratio; | |
2035 | ||
2036 | /* Guest TSC same frequency as host TSC? */ | |
2037 | if (!scale) { | |
2038 | vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio; | |
2039 | return 0; | |
2040 | } | |
2041 | ||
2042 | /* TSC scaling supported? */ | |
2043 | if (!kvm_has_tsc_control) { | |
2044 | if (user_tsc_khz > tsc_khz) { | |
2045 | vcpu->arch.tsc_catchup = 1; | |
2046 | vcpu->arch.tsc_always_catchup = 1; | |
2047 | return 0; | |
2048 | } else { | |
3f16a5c3 | 2049 | pr_warn_ratelimited("user requested TSC rate below hardware speed\n"); |
381d585c HZ |
2050 | return -1; |
2051 | } | |
2052 | } | |
2053 | ||
2054 | /* TSC scaling required - calculate ratio */ | |
2055 | ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits, | |
2056 | user_tsc_khz, tsc_khz); | |
2057 | ||
2058 | if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) { | |
3f16a5c3 PB |
2059 | pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n", |
2060 | user_tsc_khz); | |
381d585c HZ |
2061 | return -1; |
2062 | } | |
2063 | ||
2064 | vcpu->arch.tsc_scaling_ratio = ratio; | |
2065 | return 0; | |
2066 | } | |
2067 | ||
4941b8cb | 2068 | static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz) |
759379dd | 2069 | { |
cc578287 ZA |
2070 | u32 thresh_lo, thresh_hi; |
2071 | int use_scaling = 0; | |
217fc9cf | 2072 | |
03ba32ca | 2073 | /* tsc_khz can be zero if TSC calibration fails */ |
4941b8cb | 2074 | if (user_tsc_khz == 0) { |
ad721883 HZ |
2075 | /* set tsc_scaling_ratio to a safe value */ |
2076 | vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio; | |
381d585c | 2077 | return -1; |
ad721883 | 2078 | } |
03ba32ca | 2079 | |
c285545f | 2080 | /* Compute a scale to convert nanoseconds in TSC cycles */ |
3ae13faa | 2081 | kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC, |
cc578287 ZA |
2082 | &vcpu->arch.virtual_tsc_shift, |
2083 | &vcpu->arch.virtual_tsc_mult); | |
4941b8cb | 2084 | vcpu->arch.virtual_tsc_khz = user_tsc_khz; |
cc578287 ZA |
2085 | |
2086 | /* | |
2087 | * Compute the variation in TSC rate which is acceptable | |
2088 | * within the range of tolerance and decide if the | |
2089 | * rate being applied is within that bounds of the hardware | |
2090 | * rate. If so, no scaling or compensation need be done. | |
2091 | */ | |
2092 | thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm); | |
2093 | thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm); | |
4941b8cb PB |
2094 | if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) { |
2095 | pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi); | |
cc578287 ZA |
2096 | use_scaling = 1; |
2097 | } | |
4941b8cb | 2098 | return set_tsc_khz(vcpu, user_tsc_khz, use_scaling); |
c285545f ZA |
2099 | } |
2100 | ||
2101 | static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns) | |
2102 | { | |
e26101b1 | 2103 | u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec, |
cc578287 ZA |
2104 | vcpu->arch.virtual_tsc_mult, |
2105 | vcpu->arch.virtual_tsc_shift); | |
e26101b1 | 2106 | tsc += vcpu->arch.this_tsc_write; |
c285545f ZA |
2107 | return tsc; |
2108 | } | |
2109 | ||
b0c39dc6 VK |
2110 | static inline int gtod_is_based_on_tsc(int mode) |
2111 | { | |
b95a8a27 | 2112 | return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK; |
b0c39dc6 VK |
2113 | } |
2114 | ||
69b0049a | 2115 | static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu) |
b48aa97e MT |
2116 | { |
2117 | #ifdef CONFIG_X86_64 | |
2118 | bool vcpus_matched; | |
b48aa97e MT |
2119 | struct kvm_arch *ka = &vcpu->kvm->arch; |
2120 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; | |
2121 | ||
2122 | vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 == | |
2123 | atomic_read(&vcpu->kvm->online_vcpus)); | |
2124 | ||
7f187922 MT |
2125 | /* |
2126 | * Once the masterclock is enabled, always perform request in | |
2127 | * order to update it. | |
2128 | * | |
2129 | * In order to enable masterclock, the host clocksource must be TSC | |
2130 | * and the vcpus need to have matched TSCs. When that happens, | |
2131 | * perform request to enable masterclock. | |
2132 | */ | |
2133 | if (ka->use_master_clock || | |
b0c39dc6 | 2134 | (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched)) |
b48aa97e MT |
2135 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); |
2136 | ||
2137 | trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc, | |
2138 | atomic_read(&vcpu->kvm->online_vcpus), | |
2139 | ka->use_master_clock, gtod->clock.vclock_mode); | |
2140 | #endif | |
2141 | } | |
2142 | ||
35181e86 HZ |
2143 | /* |
2144 | * Multiply tsc by a fixed point number represented by ratio. | |
2145 | * | |
2146 | * The most significant 64-N bits (mult) of ratio represent the | |
2147 | * integral part of the fixed point number; the remaining N bits | |
2148 | * (frac) represent the fractional part, ie. ratio represents a fixed | |
2149 | * point number (mult + frac * 2^(-N)). | |
2150 | * | |
2151 | * N equals to kvm_tsc_scaling_ratio_frac_bits. | |
2152 | */ | |
2153 | static inline u64 __scale_tsc(u64 ratio, u64 tsc) | |
2154 | { | |
2155 | return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits); | |
2156 | } | |
2157 | ||
2158 | u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc) | |
2159 | { | |
2160 | u64 _tsc = tsc; | |
2161 | u64 ratio = vcpu->arch.tsc_scaling_ratio; | |
2162 | ||
2163 | if (ratio != kvm_default_tsc_scaling_ratio) | |
2164 | _tsc = __scale_tsc(ratio, tsc); | |
2165 | ||
2166 | return _tsc; | |
2167 | } | |
2168 | EXPORT_SYMBOL_GPL(kvm_scale_tsc); | |
2169 | ||
07c1419a HZ |
2170 | static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc) |
2171 | { | |
2172 | u64 tsc; | |
2173 | ||
2174 | tsc = kvm_scale_tsc(vcpu, rdtsc()); | |
2175 | ||
2176 | return target_tsc - tsc; | |
2177 | } | |
2178 | ||
4ba76538 HZ |
2179 | u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc) |
2180 | { | |
56ba77a4 | 2181 | return vcpu->arch.l1_tsc_offset + kvm_scale_tsc(vcpu, host_tsc); |
4ba76538 HZ |
2182 | } |
2183 | EXPORT_SYMBOL_GPL(kvm_read_l1_tsc); | |
2184 | ||
a545ab6a LC |
2185 | static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset) |
2186 | { | |
56ba77a4 | 2187 | vcpu->arch.l1_tsc_offset = offset; |
afaf0b2f | 2188 | vcpu->arch.tsc_offset = kvm_x86_ops.write_l1_tsc_offset(vcpu, offset); |
a545ab6a LC |
2189 | } |
2190 | ||
b0c39dc6 VK |
2191 | static inline bool kvm_check_tsc_unstable(void) |
2192 | { | |
2193 | #ifdef CONFIG_X86_64 | |
2194 | /* | |
2195 | * TSC is marked unstable when we're running on Hyper-V, | |
2196 | * 'TSC page' clocksource is good. | |
2197 | */ | |
b95a8a27 | 2198 | if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK) |
b0c39dc6 VK |
2199 | return false; |
2200 | #endif | |
2201 | return check_tsc_unstable(); | |
2202 | } | |
2203 | ||
0c899c25 | 2204 | static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data) |
99e3e30a ZA |
2205 | { |
2206 | struct kvm *kvm = vcpu->kvm; | |
f38e098f | 2207 | u64 offset, ns, elapsed; |
99e3e30a | 2208 | unsigned long flags; |
b48aa97e | 2209 | bool matched; |
0d3da0d2 | 2210 | bool already_matched; |
c5e8ec8e | 2211 | bool synchronizing = false; |
99e3e30a | 2212 | |
038f8c11 | 2213 | raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags); |
07c1419a | 2214 | offset = kvm_compute_tsc_offset(vcpu, data); |
8171cd68 | 2215 | ns = get_kvmclock_base_ns(); |
f38e098f | 2216 | elapsed = ns - kvm->arch.last_tsc_nsec; |
5d3cb0f6 | 2217 | |
03ba32ca | 2218 | if (vcpu->arch.virtual_tsc_khz) { |
0c899c25 | 2219 | if (data == 0) { |
bd8fab39 DP |
2220 | /* |
2221 | * detection of vcpu initialization -- need to sync | |
2222 | * with other vCPUs. This particularly helps to keep | |
2223 | * kvm_clock stable after CPU hotplug | |
2224 | */ | |
2225 | synchronizing = true; | |
2226 | } else { | |
2227 | u64 tsc_exp = kvm->arch.last_tsc_write + | |
2228 | nsec_to_cycles(vcpu, elapsed); | |
2229 | u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL; | |
2230 | /* | |
2231 | * Special case: TSC write with a small delta (1 second) | |
2232 | * of virtual cycle time against real time is | |
2233 | * interpreted as an attempt to synchronize the CPU. | |
2234 | */ | |
2235 | synchronizing = data < tsc_exp + tsc_hz && | |
2236 | data + tsc_hz > tsc_exp; | |
2237 | } | |
c5e8ec8e | 2238 | } |
f38e098f ZA |
2239 | |
2240 | /* | |
5d3cb0f6 ZA |
2241 | * For a reliable TSC, we can match TSC offsets, and for an unstable |
2242 | * TSC, we add elapsed time in this computation. We could let the | |
2243 | * compensation code attempt to catch up if we fall behind, but | |
2244 | * it's better to try to match offsets from the beginning. | |
2245 | */ | |
c5e8ec8e | 2246 | if (synchronizing && |
5d3cb0f6 | 2247 | vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) { |
b0c39dc6 | 2248 | if (!kvm_check_tsc_unstable()) { |
e26101b1 | 2249 | offset = kvm->arch.cur_tsc_offset; |
f38e098f | 2250 | } else { |
857e4099 | 2251 | u64 delta = nsec_to_cycles(vcpu, elapsed); |
5d3cb0f6 | 2252 | data += delta; |
07c1419a | 2253 | offset = kvm_compute_tsc_offset(vcpu, data); |
f38e098f | 2254 | } |
b48aa97e | 2255 | matched = true; |
0d3da0d2 | 2256 | already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation); |
e26101b1 ZA |
2257 | } else { |
2258 | /* | |
2259 | * We split periods of matched TSC writes into generations. | |
2260 | * For each generation, we track the original measured | |
2261 | * nanosecond time, offset, and write, so if TSCs are in | |
2262 | * sync, we can match exact offset, and if not, we can match | |
4a969980 | 2263 | * exact software computation in compute_guest_tsc() |
e26101b1 ZA |
2264 | * |
2265 | * These values are tracked in kvm->arch.cur_xxx variables. | |
2266 | */ | |
2267 | kvm->arch.cur_tsc_generation++; | |
2268 | kvm->arch.cur_tsc_nsec = ns; | |
2269 | kvm->arch.cur_tsc_write = data; | |
2270 | kvm->arch.cur_tsc_offset = offset; | |
b48aa97e | 2271 | matched = false; |
f38e098f | 2272 | } |
e26101b1 ZA |
2273 | |
2274 | /* | |
2275 | * We also track th most recent recorded KHZ, write and time to | |
2276 | * allow the matching interval to be extended at each write. | |
2277 | */ | |
f38e098f ZA |
2278 | kvm->arch.last_tsc_nsec = ns; |
2279 | kvm->arch.last_tsc_write = data; | |
5d3cb0f6 | 2280 | kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz; |
99e3e30a | 2281 | |
b183aa58 | 2282 | vcpu->arch.last_guest_tsc = data; |
e26101b1 ZA |
2283 | |
2284 | /* Keep track of which generation this VCPU has synchronized to */ | |
2285 | vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation; | |
2286 | vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec; | |
2287 | vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write; | |
2288 | ||
a545ab6a | 2289 | kvm_vcpu_write_tsc_offset(vcpu, offset); |
e26101b1 | 2290 | raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags); |
b48aa97e MT |
2291 | |
2292 | spin_lock(&kvm->arch.pvclock_gtod_sync_lock); | |
0d3da0d2 | 2293 | if (!matched) { |
b48aa97e | 2294 | kvm->arch.nr_vcpus_matched_tsc = 0; |
0d3da0d2 TG |
2295 | } else if (!already_matched) { |
2296 | kvm->arch.nr_vcpus_matched_tsc++; | |
2297 | } | |
b48aa97e MT |
2298 | |
2299 | kvm_track_tsc_matching(vcpu); | |
2300 | spin_unlock(&kvm->arch.pvclock_gtod_sync_lock); | |
99e3e30a | 2301 | } |
e26101b1 | 2302 | |
58ea6767 HZ |
2303 | static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu, |
2304 | s64 adjustment) | |
2305 | { | |
56ba77a4 | 2306 | u64 tsc_offset = vcpu->arch.l1_tsc_offset; |
326e7425 | 2307 | kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment); |
58ea6767 HZ |
2308 | } |
2309 | ||
2310 | static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment) | |
2311 | { | |
2312 | if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio) | |
2313 | WARN_ON(adjustment < 0); | |
2314 | adjustment = kvm_scale_tsc(vcpu, (u64) adjustment); | |
ea26e4ec | 2315 | adjust_tsc_offset_guest(vcpu, adjustment); |
58ea6767 HZ |
2316 | } |
2317 | ||
d828199e MT |
2318 | #ifdef CONFIG_X86_64 |
2319 | ||
a5a1d1c2 | 2320 | static u64 read_tsc(void) |
d828199e | 2321 | { |
a5a1d1c2 | 2322 | u64 ret = (u64)rdtsc_ordered(); |
03b9730b | 2323 | u64 last = pvclock_gtod_data.clock.cycle_last; |
d828199e MT |
2324 | |
2325 | if (likely(ret >= last)) | |
2326 | return ret; | |
2327 | ||
2328 | /* | |
2329 | * GCC likes to generate cmov here, but this branch is extremely | |
6a6256f9 | 2330 | * predictable (it's just a function of time and the likely is |
d828199e MT |
2331 | * very likely) and there's a data dependence, so force GCC |
2332 | * to generate a branch instead. I don't barrier() because | |
2333 | * we don't actually need a barrier, and if this function | |
2334 | * ever gets inlined it will generate worse code. | |
2335 | */ | |
2336 | asm volatile (""); | |
2337 | return last; | |
2338 | } | |
2339 | ||
53fafdbb MT |
2340 | static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp, |
2341 | int *mode) | |
d828199e MT |
2342 | { |
2343 | long v; | |
b0c39dc6 VK |
2344 | u64 tsc_pg_val; |
2345 | ||
53fafdbb | 2346 | switch (clock->vclock_mode) { |
b95a8a27 | 2347 | case VDSO_CLOCKMODE_HVCLOCK: |
b0c39dc6 VK |
2348 | tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(), |
2349 | tsc_timestamp); | |
2350 | if (tsc_pg_val != U64_MAX) { | |
2351 | /* TSC page valid */ | |
b95a8a27 | 2352 | *mode = VDSO_CLOCKMODE_HVCLOCK; |
53fafdbb MT |
2353 | v = (tsc_pg_val - clock->cycle_last) & |
2354 | clock->mask; | |
b0c39dc6 VK |
2355 | } else { |
2356 | /* TSC page invalid */ | |
b95a8a27 | 2357 | *mode = VDSO_CLOCKMODE_NONE; |
b0c39dc6 VK |
2358 | } |
2359 | break; | |
b95a8a27 TG |
2360 | case VDSO_CLOCKMODE_TSC: |
2361 | *mode = VDSO_CLOCKMODE_TSC; | |
b0c39dc6 | 2362 | *tsc_timestamp = read_tsc(); |
53fafdbb MT |
2363 | v = (*tsc_timestamp - clock->cycle_last) & |
2364 | clock->mask; | |
b0c39dc6 VK |
2365 | break; |
2366 | default: | |
b95a8a27 | 2367 | *mode = VDSO_CLOCKMODE_NONE; |
b0c39dc6 | 2368 | } |
d828199e | 2369 | |
b95a8a27 | 2370 | if (*mode == VDSO_CLOCKMODE_NONE) |
b0c39dc6 | 2371 | *tsc_timestamp = v = 0; |
d828199e | 2372 | |
53fafdbb | 2373 | return v * clock->mult; |
d828199e MT |
2374 | } |
2375 | ||
53fafdbb | 2376 | static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp) |
d828199e | 2377 | { |
cbcf2dd3 | 2378 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; |
d828199e | 2379 | unsigned long seq; |
d828199e | 2380 | int mode; |
cbcf2dd3 | 2381 | u64 ns; |
d828199e | 2382 | |
d828199e MT |
2383 | do { |
2384 | seq = read_seqcount_begin(>od->seq); | |
917f9475 | 2385 | ns = gtod->raw_clock.base_cycles; |
53fafdbb | 2386 | ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode); |
917f9475 PB |
2387 | ns >>= gtod->raw_clock.shift; |
2388 | ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot)); | |
d828199e | 2389 | } while (unlikely(read_seqcount_retry(>od->seq, seq))); |
cbcf2dd3 | 2390 | *t = ns; |
d828199e MT |
2391 | |
2392 | return mode; | |
2393 | } | |
2394 | ||
899a31f5 | 2395 | static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp) |
55dd00a7 MT |
2396 | { |
2397 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; | |
2398 | unsigned long seq; | |
2399 | int mode; | |
2400 | u64 ns; | |
2401 | ||
2402 | do { | |
2403 | seq = read_seqcount_begin(>od->seq); | |
55dd00a7 | 2404 | ts->tv_sec = gtod->wall_time_sec; |
917f9475 | 2405 | ns = gtod->clock.base_cycles; |
53fafdbb | 2406 | ns += vgettsc(>od->clock, tsc_timestamp, &mode); |
55dd00a7 MT |
2407 | ns >>= gtod->clock.shift; |
2408 | } while (unlikely(read_seqcount_retry(>od->seq, seq))); | |
2409 | ||
2410 | ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns); | |
2411 | ts->tv_nsec = ns; | |
2412 | ||
2413 | return mode; | |
2414 | } | |
2415 | ||
b0c39dc6 VK |
2416 | /* returns true if host is using TSC based clocksource */ |
2417 | static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp) | |
d828199e | 2418 | { |
d828199e | 2419 | /* checked again under seqlock below */ |
b0c39dc6 | 2420 | if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode)) |
d828199e MT |
2421 | return false; |
2422 | ||
53fafdbb | 2423 | return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns, |
b0c39dc6 | 2424 | tsc_timestamp)); |
d828199e | 2425 | } |
55dd00a7 | 2426 | |
b0c39dc6 | 2427 | /* returns true if host is using TSC based clocksource */ |
899a31f5 | 2428 | static bool kvm_get_walltime_and_clockread(struct timespec64 *ts, |
b0c39dc6 | 2429 | u64 *tsc_timestamp) |
55dd00a7 MT |
2430 | { |
2431 | /* checked again under seqlock below */ | |
b0c39dc6 | 2432 | if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode)) |
55dd00a7 MT |
2433 | return false; |
2434 | ||
b0c39dc6 | 2435 | return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp)); |
55dd00a7 | 2436 | } |
d828199e MT |
2437 | #endif |
2438 | ||
2439 | /* | |
2440 | * | |
b48aa97e MT |
2441 | * Assuming a stable TSC across physical CPUS, and a stable TSC |
2442 | * across virtual CPUs, the following condition is possible. | |
2443 | * Each numbered line represents an event visible to both | |
d828199e MT |
2444 | * CPUs at the next numbered event. |
2445 | * | |
2446 | * "timespecX" represents host monotonic time. "tscX" represents | |
2447 | * RDTSC value. | |
2448 | * | |
2449 | * VCPU0 on CPU0 | VCPU1 on CPU1 | |
2450 | * | |
2451 | * 1. read timespec0,tsc0 | |
2452 | * 2. | timespec1 = timespec0 + N | |
2453 | * | tsc1 = tsc0 + M | |
2454 | * 3. transition to guest | transition to guest | |
2455 | * 4. ret0 = timespec0 + (rdtsc - tsc0) | | |
2456 | * 5. | ret1 = timespec1 + (rdtsc - tsc1) | |
2457 | * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M)) | |
2458 | * | |
2459 | * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity: | |
2460 | * | |
2461 | * - ret0 < ret1 | |
2462 | * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M)) | |
2463 | * ... | |
2464 | * - 0 < N - M => M < N | |
2465 | * | |
2466 | * That is, when timespec0 != timespec1, M < N. Unfortunately that is not | |
2467 | * always the case (the difference between two distinct xtime instances | |
2468 | * might be smaller then the difference between corresponding TSC reads, | |
2469 | * when updating guest vcpus pvclock areas). | |
2470 | * | |
2471 | * To avoid that problem, do not allow visibility of distinct | |
2472 | * system_timestamp/tsc_timestamp values simultaneously: use a master | |
2473 | * copy of host monotonic time values. Update that master copy | |
2474 | * in lockstep. | |
2475 | * | |
b48aa97e | 2476 | * Rely on synchronization of host TSCs and guest TSCs for monotonicity. |
d828199e MT |
2477 | * |
2478 | */ | |
2479 | ||
2480 | static void pvclock_update_vm_gtod_copy(struct kvm *kvm) | |
2481 | { | |
2482 | #ifdef CONFIG_X86_64 | |
2483 | struct kvm_arch *ka = &kvm->arch; | |
2484 | int vclock_mode; | |
b48aa97e MT |
2485 | bool host_tsc_clocksource, vcpus_matched; |
2486 | ||
2487 | vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 == | |
2488 | atomic_read(&kvm->online_vcpus)); | |
d828199e MT |
2489 | |
2490 | /* | |
2491 | * If the host uses TSC clock, then passthrough TSC as stable | |
2492 | * to the guest. | |
2493 | */ | |
b48aa97e | 2494 | host_tsc_clocksource = kvm_get_time_and_clockread( |
d828199e MT |
2495 | &ka->master_kernel_ns, |
2496 | &ka->master_cycle_now); | |
2497 | ||
16a96021 | 2498 | ka->use_master_clock = host_tsc_clocksource && vcpus_matched |
a826faf1 | 2499 | && !ka->backwards_tsc_observed |
54750f2c | 2500 | && !ka->boot_vcpu_runs_old_kvmclock; |
b48aa97e | 2501 | |
d828199e MT |
2502 | if (ka->use_master_clock) |
2503 | atomic_set(&kvm_guest_has_master_clock, 1); | |
2504 | ||
2505 | vclock_mode = pvclock_gtod_data.clock.vclock_mode; | |
b48aa97e MT |
2506 | trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode, |
2507 | vcpus_matched); | |
d828199e MT |
2508 | #endif |
2509 | } | |
2510 | ||
2860c4b1 PB |
2511 | void kvm_make_mclock_inprogress_request(struct kvm *kvm) |
2512 | { | |
2513 | kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS); | |
2514 | } | |
2515 | ||
2e762ff7 MT |
2516 | static void kvm_gen_update_masterclock(struct kvm *kvm) |
2517 | { | |
2518 | #ifdef CONFIG_X86_64 | |
2519 | int i; | |
2520 | struct kvm_vcpu *vcpu; | |
2521 | struct kvm_arch *ka = &kvm->arch; | |
2522 | ||
2523 | spin_lock(&ka->pvclock_gtod_sync_lock); | |
2524 | kvm_make_mclock_inprogress_request(kvm); | |
2525 | /* no guest entries from this point */ | |
2526 | pvclock_update_vm_gtod_copy(kvm); | |
2527 | ||
2528 | kvm_for_each_vcpu(i, vcpu, kvm) | |
105b21bb | 2529 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
2e762ff7 MT |
2530 | |
2531 | /* guest entries allowed */ | |
2532 | kvm_for_each_vcpu(i, vcpu, kvm) | |
72875d8a | 2533 | kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu); |
2e762ff7 MT |
2534 | |
2535 | spin_unlock(&ka->pvclock_gtod_sync_lock); | |
2536 | #endif | |
2537 | } | |
2538 | ||
e891a32e | 2539 | u64 get_kvmclock_ns(struct kvm *kvm) |
108b249c | 2540 | { |
108b249c | 2541 | struct kvm_arch *ka = &kvm->arch; |
8b953440 | 2542 | struct pvclock_vcpu_time_info hv_clock; |
e2c2206a | 2543 | u64 ret; |
108b249c | 2544 | |
8b953440 PB |
2545 | spin_lock(&ka->pvclock_gtod_sync_lock); |
2546 | if (!ka->use_master_clock) { | |
2547 | spin_unlock(&ka->pvclock_gtod_sync_lock); | |
8171cd68 | 2548 | return get_kvmclock_base_ns() + ka->kvmclock_offset; |
108b249c PB |
2549 | } |
2550 | ||
8b953440 PB |
2551 | hv_clock.tsc_timestamp = ka->master_cycle_now; |
2552 | hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset; | |
2553 | spin_unlock(&ka->pvclock_gtod_sync_lock); | |
2554 | ||
e2c2206a WL |
2555 | /* both __this_cpu_read() and rdtsc() should be on the same cpu */ |
2556 | get_cpu(); | |
2557 | ||
e70b57a6 WL |
2558 | if (__this_cpu_read(cpu_tsc_khz)) { |
2559 | kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL, | |
2560 | &hv_clock.tsc_shift, | |
2561 | &hv_clock.tsc_to_system_mul); | |
2562 | ret = __pvclock_read_cycles(&hv_clock, rdtsc()); | |
2563 | } else | |
8171cd68 | 2564 | ret = get_kvmclock_base_ns() + ka->kvmclock_offset; |
e2c2206a WL |
2565 | |
2566 | put_cpu(); | |
2567 | ||
2568 | return ret; | |
108b249c PB |
2569 | } |
2570 | ||
0d6dd2ff PB |
2571 | static void kvm_setup_pvclock_page(struct kvm_vcpu *v) |
2572 | { | |
2573 | struct kvm_vcpu_arch *vcpu = &v->arch; | |
2574 | struct pvclock_vcpu_time_info guest_hv_clock; | |
2575 | ||
4e335d9e | 2576 | if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time, |
0d6dd2ff PB |
2577 | &guest_hv_clock, sizeof(guest_hv_clock)))) |
2578 | return; | |
2579 | ||
2580 | /* This VCPU is paused, but it's legal for a guest to read another | |
2581 | * VCPU's kvmclock, so we really have to follow the specification where | |
2582 | * it says that version is odd if data is being modified, and even after | |
2583 | * it is consistent. | |
2584 | * | |
2585 | * Version field updates must be kept separate. This is because | |
2586 | * kvm_write_guest_cached might use a "rep movs" instruction, and | |
2587 | * writes within a string instruction are weakly ordered. So there | |
2588 | * are three writes overall. | |
2589 | * | |
2590 | * As a small optimization, only write the version field in the first | |
2591 | * and third write. The vcpu->pv_time cache is still valid, because the | |
2592 | * version field is the first in the struct. | |
2593 | */ | |
2594 | BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0); | |
2595 | ||
51c4b8bb LA |
2596 | if (guest_hv_clock.version & 1) |
2597 | ++guest_hv_clock.version; /* first time write, random junk */ | |
2598 | ||
0d6dd2ff | 2599 | vcpu->hv_clock.version = guest_hv_clock.version + 1; |
4e335d9e PB |
2600 | kvm_write_guest_cached(v->kvm, &vcpu->pv_time, |
2601 | &vcpu->hv_clock, | |
2602 | sizeof(vcpu->hv_clock.version)); | |
0d6dd2ff PB |
2603 | |
2604 | smp_wmb(); | |
2605 | ||
2606 | /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */ | |
2607 | vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED); | |
2608 | ||
2609 | if (vcpu->pvclock_set_guest_stopped_request) { | |
2610 | vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED; | |
2611 | vcpu->pvclock_set_guest_stopped_request = false; | |
2612 | } | |
2613 | ||
2614 | trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock); | |
2615 | ||
4e335d9e PB |
2616 | kvm_write_guest_cached(v->kvm, &vcpu->pv_time, |
2617 | &vcpu->hv_clock, | |
2618 | sizeof(vcpu->hv_clock)); | |
0d6dd2ff PB |
2619 | |
2620 | smp_wmb(); | |
2621 | ||
2622 | vcpu->hv_clock.version++; | |
4e335d9e PB |
2623 | kvm_write_guest_cached(v->kvm, &vcpu->pv_time, |
2624 | &vcpu->hv_clock, | |
2625 | sizeof(vcpu->hv_clock.version)); | |
0d6dd2ff PB |
2626 | } |
2627 | ||
34c238a1 | 2628 | static int kvm_guest_time_update(struct kvm_vcpu *v) |
18068523 | 2629 | { |
78db6a50 | 2630 | unsigned long flags, tgt_tsc_khz; |
18068523 | 2631 | struct kvm_vcpu_arch *vcpu = &v->arch; |
d828199e | 2632 | struct kvm_arch *ka = &v->kvm->arch; |
f25e656d | 2633 | s64 kernel_ns; |
d828199e | 2634 | u64 tsc_timestamp, host_tsc; |
51d59c6b | 2635 | u8 pvclock_flags; |
d828199e MT |
2636 | bool use_master_clock; |
2637 | ||
2638 | kernel_ns = 0; | |
2639 | host_tsc = 0; | |
18068523 | 2640 | |
d828199e MT |
2641 | /* |
2642 | * If the host uses TSC clock, then passthrough TSC as stable | |
2643 | * to the guest. | |
2644 | */ | |
2645 | spin_lock(&ka->pvclock_gtod_sync_lock); | |
2646 | use_master_clock = ka->use_master_clock; | |
2647 | if (use_master_clock) { | |
2648 | host_tsc = ka->master_cycle_now; | |
2649 | kernel_ns = ka->master_kernel_ns; | |
2650 | } | |
2651 | spin_unlock(&ka->pvclock_gtod_sync_lock); | |
c09664bb MT |
2652 | |
2653 | /* Keep irq disabled to prevent changes to the clock */ | |
2654 | local_irq_save(flags); | |
78db6a50 PB |
2655 | tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz); |
2656 | if (unlikely(tgt_tsc_khz == 0)) { | |
c09664bb MT |
2657 | local_irq_restore(flags); |
2658 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, v); | |
2659 | return 1; | |
2660 | } | |
d828199e | 2661 | if (!use_master_clock) { |
4ea1636b | 2662 | host_tsc = rdtsc(); |
8171cd68 | 2663 | kernel_ns = get_kvmclock_base_ns(); |
d828199e MT |
2664 | } |
2665 | ||
4ba76538 | 2666 | tsc_timestamp = kvm_read_l1_tsc(v, host_tsc); |
d828199e | 2667 | |
c285545f ZA |
2668 | /* |
2669 | * We may have to catch up the TSC to match elapsed wall clock | |
2670 | * time for two reasons, even if kvmclock is used. | |
2671 | * 1) CPU could have been running below the maximum TSC rate | |
2672 | * 2) Broken TSC compensation resets the base at each VCPU | |
2673 | * entry to avoid unknown leaps of TSC even when running | |
2674 | * again on the same CPU. This may cause apparent elapsed | |
2675 | * time to disappear, and the guest to stand still or run | |
2676 | * very slowly. | |
2677 | */ | |
2678 | if (vcpu->tsc_catchup) { | |
2679 | u64 tsc = compute_guest_tsc(v, kernel_ns); | |
2680 | if (tsc > tsc_timestamp) { | |
f1e2b260 | 2681 | adjust_tsc_offset_guest(v, tsc - tsc_timestamp); |
c285545f ZA |
2682 | tsc_timestamp = tsc; |
2683 | } | |
50d0a0f9 GH |
2684 | } |
2685 | ||
18068523 GOC |
2686 | local_irq_restore(flags); |
2687 | ||
0d6dd2ff | 2688 | /* With all the info we got, fill in the values */ |
18068523 | 2689 | |
78db6a50 PB |
2690 | if (kvm_has_tsc_control) |
2691 | tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz); | |
2692 | ||
2693 | if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) { | |
3ae13faa | 2694 | kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL, |
5f4e3f88 ZA |
2695 | &vcpu->hv_clock.tsc_shift, |
2696 | &vcpu->hv_clock.tsc_to_system_mul); | |
78db6a50 | 2697 | vcpu->hw_tsc_khz = tgt_tsc_khz; |
8cfdc000 ZA |
2698 | } |
2699 | ||
1d5f066e | 2700 | vcpu->hv_clock.tsc_timestamp = tsc_timestamp; |
759379dd | 2701 | vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset; |
28e4639a | 2702 | vcpu->last_guest_tsc = tsc_timestamp; |
51d59c6b | 2703 | |
d828199e | 2704 | /* If the host uses TSC clocksource, then it is stable */ |
0d6dd2ff | 2705 | pvclock_flags = 0; |
d828199e MT |
2706 | if (use_master_clock) |
2707 | pvclock_flags |= PVCLOCK_TSC_STABLE_BIT; | |
2708 | ||
78c0337a MT |
2709 | vcpu->hv_clock.flags = pvclock_flags; |
2710 | ||
095cf55d PB |
2711 | if (vcpu->pv_time_enabled) |
2712 | kvm_setup_pvclock_page(v); | |
2713 | if (v == kvm_get_vcpu(v->kvm, 0)) | |
2714 | kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock); | |
8cfdc000 | 2715 | return 0; |
c8076604 GH |
2716 | } |
2717 | ||
0061d53d MT |
2718 | /* |
2719 | * kvmclock updates which are isolated to a given vcpu, such as | |
2720 | * vcpu->cpu migration, should not allow system_timestamp from | |
2721 | * the rest of the vcpus to remain static. Otherwise ntp frequency | |
2722 | * correction applies to one vcpu's system_timestamp but not | |
2723 | * the others. | |
2724 | * | |
2725 | * So in those cases, request a kvmclock update for all vcpus. | |
7e44e449 AJ |
2726 | * We need to rate-limit these requests though, as they can |
2727 | * considerably slow guests that have a large number of vcpus. | |
2728 | * The time for a remote vcpu to update its kvmclock is bound | |
2729 | * by the delay we use to rate-limit the updates. | |
0061d53d MT |
2730 | */ |
2731 | ||
7e44e449 AJ |
2732 | #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100) |
2733 | ||
2734 | static void kvmclock_update_fn(struct work_struct *work) | |
0061d53d MT |
2735 | { |
2736 | int i; | |
7e44e449 AJ |
2737 | struct delayed_work *dwork = to_delayed_work(work); |
2738 | struct kvm_arch *ka = container_of(dwork, struct kvm_arch, | |
2739 | kvmclock_update_work); | |
2740 | struct kvm *kvm = container_of(ka, struct kvm, arch); | |
0061d53d MT |
2741 | struct kvm_vcpu *vcpu; |
2742 | ||
2743 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
105b21bb | 2744 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
0061d53d MT |
2745 | kvm_vcpu_kick(vcpu); |
2746 | } | |
2747 | } | |
2748 | ||
7e44e449 AJ |
2749 | static void kvm_gen_kvmclock_update(struct kvm_vcpu *v) |
2750 | { | |
2751 | struct kvm *kvm = v->kvm; | |
2752 | ||
105b21bb | 2753 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, v); |
7e44e449 AJ |
2754 | schedule_delayed_work(&kvm->arch.kvmclock_update_work, |
2755 | KVMCLOCK_UPDATE_DELAY); | |
2756 | } | |
2757 | ||
332967a3 AJ |
2758 | #define KVMCLOCK_SYNC_PERIOD (300 * HZ) |
2759 | ||
2760 | static void kvmclock_sync_fn(struct work_struct *work) | |
2761 | { | |
2762 | struct delayed_work *dwork = to_delayed_work(work); | |
2763 | struct kvm_arch *ka = container_of(dwork, struct kvm_arch, | |
2764 | kvmclock_sync_work); | |
2765 | struct kvm *kvm = container_of(ka, struct kvm, arch); | |
2766 | ||
630994b3 MT |
2767 | if (!kvmclock_periodic_sync) |
2768 | return; | |
2769 | ||
332967a3 AJ |
2770 | schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0); |
2771 | schedule_delayed_work(&kvm->arch.kvmclock_sync_work, | |
2772 | KVMCLOCK_SYNC_PERIOD); | |
2773 | } | |
2774 | ||
191c8137 BP |
2775 | /* |
2776 | * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP. | |
2777 | */ | |
2778 | static bool can_set_mci_status(struct kvm_vcpu *vcpu) | |
2779 | { | |
2780 | /* McStatusWrEn enabled? */ | |
23493d0a | 2781 | if (guest_cpuid_is_amd_or_hygon(vcpu)) |
191c8137 BP |
2782 | return !!(vcpu->arch.msr_hwcr & BIT_ULL(18)); |
2783 | ||
2784 | return false; | |
2785 | } | |
2786 | ||
9ffd986c | 2787 | static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
15c4a640 | 2788 | { |
890ca9ae HY |
2789 | u64 mcg_cap = vcpu->arch.mcg_cap; |
2790 | unsigned bank_num = mcg_cap & 0xff; | |
9ffd986c WL |
2791 | u32 msr = msr_info->index; |
2792 | u64 data = msr_info->data; | |
890ca9ae | 2793 | |
15c4a640 | 2794 | switch (msr) { |
15c4a640 | 2795 | case MSR_IA32_MCG_STATUS: |
890ca9ae | 2796 | vcpu->arch.mcg_status = data; |
15c4a640 | 2797 | break; |
c7ac679c | 2798 | case MSR_IA32_MCG_CTL: |
44883f01 PB |
2799 | if (!(mcg_cap & MCG_CTL_P) && |
2800 | (data || !msr_info->host_initiated)) | |
890ca9ae HY |
2801 | return 1; |
2802 | if (data != 0 && data != ~(u64)0) | |
44883f01 | 2803 | return 1; |
890ca9ae HY |
2804 | vcpu->arch.mcg_ctl = data; |
2805 | break; | |
2806 | default: | |
2807 | if (msr >= MSR_IA32_MC0_CTL && | |
81760dcc | 2808 | msr < MSR_IA32_MCx_CTL(bank_num)) { |
6ec4c5ee MP |
2809 | u32 offset = array_index_nospec( |
2810 | msr - MSR_IA32_MC0_CTL, | |
2811 | MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL); | |
2812 | ||
114be429 AP |
2813 | /* only 0 or all 1s can be written to IA32_MCi_CTL |
2814 | * some Linux kernels though clear bit 10 in bank 4 to | |
2815 | * workaround a BIOS/GART TBL issue on AMD K8s, ignore | |
2816 | * this to avoid an uncatched #GP in the guest | |
2817 | */ | |
890ca9ae | 2818 | if ((offset & 0x3) == 0 && |
114be429 | 2819 | data != 0 && (data | (1 << 10)) != ~(u64)0) |
890ca9ae | 2820 | return -1; |
191c8137 BP |
2821 | |
2822 | /* MCi_STATUS */ | |
9ffd986c | 2823 | if (!msr_info->host_initiated && |
191c8137 BP |
2824 | (offset & 0x3) == 1 && data != 0) { |
2825 | if (!can_set_mci_status(vcpu)) | |
2826 | return -1; | |
2827 | } | |
2828 | ||
890ca9ae HY |
2829 | vcpu->arch.mce_banks[offset] = data; |
2830 | break; | |
2831 | } | |
2832 | return 1; | |
2833 | } | |
2834 | return 0; | |
2835 | } | |
2836 | ||
ffde22ac ES |
2837 | static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data) |
2838 | { | |
2839 | struct kvm *kvm = vcpu->kvm; | |
2840 | int lm = is_long_mode(vcpu); | |
2841 | u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64 | |
2842 | : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32; | |
2843 | u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64 | |
2844 | : kvm->arch.xen_hvm_config.blob_size_32; | |
2845 | u32 page_num = data & ~PAGE_MASK; | |
2846 | u64 page_addr = data & PAGE_MASK; | |
2847 | u8 *page; | |
ffde22ac | 2848 | |
ffde22ac | 2849 | if (page_num >= blob_size) |
36385ccc ML |
2850 | return 1; |
2851 | ||
ff5c2c03 | 2852 | page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE); |
36385ccc ML |
2853 | if (IS_ERR(page)) |
2854 | return PTR_ERR(page); | |
2855 | ||
2856 | if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE)) { | |
2857 | kfree(page); | |
2858 | return 1; | |
ff5c2c03 | 2859 | } |
36385ccc | 2860 | return 0; |
ffde22ac ES |
2861 | } |
2862 | ||
2635b5c4 VK |
2863 | static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu) |
2864 | { | |
2865 | u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT; | |
2866 | ||
2867 | return (vcpu->arch.apf.msr_en_val & mask) == mask; | |
2868 | } | |
2869 | ||
344d9588 GN |
2870 | static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data) |
2871 | { | |
2872 | gpa_t gpa = data & ~0x3f; | |
2873 | ||
2635b5c4 VK |
2874 | /* Bits 4:5 are reserved, Should be zero */ |
2875 | if (data & 0x30) | |
344d9588 GN |
2876 | return 1; |
2877 | ||
66570e96 OU |
2878 | if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) && |
2879 | (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT)) | |
2880 | return 1; | |
2881 | ||
2882 | if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) && | |
2883 | (data & KVM_ASYNC_PF_DELIVERY_AS_INT)) | |
2884 | return 1; | |
2885 | ||
9d3c447c | 2886 | if (!lapic_in_kernel(vcpu)) |
d831de17 | 2887 | return data ? 1 : 0; |
9d3c447c | 2888 | |
2635b5c4 | 2889 | vcpu->arch.apf.msr_en_val = data; |
344d9588 | 2890 | |
2635b5c4 | 2891 | if (!kvm_pv_async_pf_enabled(vcpu)) { |
344d9588 GN |
2892 | kvm_clear_async_pf_completion_queue(vcpu); |
2893 | kvm_async_pf_hash_reset(vcpu); | |
2894 | return 0; | |
2895 | } | |
2896 | ||
4e335d9e | 2897 | if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa, |
68fd66f1 | 2898 | sizeof(u64))) |
344d9588 GN |
2899 | return 1; |
2900 | ||
6adba527 | 2901 | vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS); |
52a5c155 | 2902 | vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT; |
2635b5c4 | 2903 | |
344d9588 | 2904 | kvm_async_pf_wakeup_all(vcpu); |
2635b5c4 VK |
2905 | |
2906 | return 0; | |
2907 | } | |
2908 | ||
2909 | static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data) | |
2910 | { | |
2911 | /* Bits 8-63 are reserved */ | |
2912 | if (data >> 8) | |
2913 | return 1; | |
2914 | ||
2915 | if (!lapic_in_kernel(vcpu)) | |
2916 | return 1; | |
2917 | ||
2918 | vcpu->arch.apf.msr_int_val = data; | |
2919 | ||
2920 | vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK; | |
2921 | ||
344d9588 GN |
2922 | return 0; |
2923 | } | |
2924 | ||
12f9a48f GC |
2925 | static void kvmclock_reset(struct kvm_vcpu *vcpu) |
2926 | { | |
0b79459b | 2927 | vcpu->arch.pv_time_enabled = false; |
49dedf0d | 2928 | vcpu->arch.time = 0; |
12f9a48f GC |
2929 | } |
2930 | ||
7780938c | 2931 | static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu) |
f38a7b75 WL |
2932 | { |
2933 | ++vcpu->stat.tlb_flush; | |
7780938c | 2934 | kvm_x86_ops.tlb_flush_all(vcpu); |
f38a7b75 WL |
2935 | } |
2936 | ||
0baedd79 VK |
2937 | static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu) |
2938 | { | |
2939 | ++vcpu->stat.tlb_flush; | |
2940 | kvm_x86_ops.tlb_flush_guest(vcpu); | |
2941 | } | |
2942 | ||
c9aaa895 GC |
2943 | static void record_steal_time(struct kvm_vcpu *vcpu) |
2944 | { | |
b0431382 BO |
2945 | struct kvm_host_map map; |
2946 | struct kvm_steal_time *st; | |
2947 | ||
c9aaa895 GC |
2948 | if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED)) |
2949 | return; | |
2950 | ||
b0431382 BO |
2951 | /* -EAGAIN is returned in atomic context so we can just return. */ |
2952 | if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, | |
2953 | &map, &vcpu->arch.st.cache, false)) | |
c9aaa895 GC |
2954 | return; |
2955 | ||
b0431382 BO |
2956 | st = map.hva + |
2957 | offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS); | |
2958 | ||
f38a7b75 WL |
2959 | /* |
2960 | * Doing a TLB flush here, on the guest's behalf, can avoid | |
2961 | * expensive IPIs. | |
2962 | */ | |
66570e96 OU |
2963 | if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) { |
2964 | trace_kvm_pv_tlb_flush(vcpu->vcpu_id, | |
2965 | st->preempted & KVM_VCPU_FLUSH_TLB); | |
2966 | if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB) | |
2967 | kvm_vcpu_flush_tlb_guest(vcpu); | |
2968 | } | |
0b9f6c46 | 2969 | |
a6bd811f | 2970 | vcpu->arch.st.preempted = 0; |
35f3fae1 | 2971 | |
b0431382 BO |
2972 | if (st->version & 1) |
2973 | st->version += 1; /* first time write, random junk */ | |
35f3fae1 | 2974 | |
b0431382 | 2975 | st->version += 1; |
35f3fae1 WL |
2976 | |
2977 | smp_wmb(); | |
2978 | ||
b0431382 | 2979 | st->steal += current->sched_info.run_delay - |
c54cdf14 LC |
2980 | vcpu->arch.st.last_steal; |
2981 | vcpu->arch.st.last_steal = current->sched_info.run_delay; | |
35f3fae1 | 2982 | |
35f3fae1 WL |
2983 | smp_wmb(); |
2984 | ||
b0431382 | 2985 | st->version += 1; |
c9aaa895 | 2986 | |
b0431382 | 2987 | kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false); |
c9aaa895 GC |
2988 | } |
2989 | ||
8fe8ab46 | 2990 | int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
15c4a640 | 2991 | { |
5753785f | 2992 | bool pr = false; |
8fe8ab46 WA |
2993 | u32 msr = msr_info->index; |
2994 | u64 data = msr_info->data; | |
5753785f | 2995 | |
15c4a640 | 2996 | switch (msr) { |
2e32b719 | 2997 | case MSR_AMD64_NB_CFG: |
2e32b719 BP |
2998 | case MSR_IA32_UCODE_WRITE: |
2999 | case MSR_VM_HSAVE_PA: | |
3000 | case MSR_AMD64_PATCH_LOADER: | |
3001 | case MSR_AMD64_BU_CFG2: | |
405a353a | 3002 | case MSR_AMD64_DC_CFG: |
0e1b869f | 3003 | case MSR_F15H_EX_CFG: |
2e32b719 BP |
3004 | break; |
3005 | ||
518e7b94 WL |
3006 | case MSR_IA32_UCODE_REV: |
3007 | if (msr_info->host_initiated) | |
3008 | vcpu->arch.microcode_version = data; | |
3009 | break; | |
0cf9135b SC |
3010 | case MSR_IA32_ARCH_CAPABILITIES: |
3011 | if (!msr_info->host_initiated) | |
3012 | return 1; | |
3013 | vcpu->arch.arch_capabilities = data; | |
3014 | break; | |
d574c539 VK |
3015 | case MSR_IA32_PERF_CAPABILITIES: { |
3016 | struct kvm_msr_entry msr_ent = {.index = msr, .data = 0}; | |
3017 | ||
3018 | if (!msr_info->host_initiated) | |
3019 | return 1; | |
3020 | if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent)) | |
3021 | return 1; | |
3022 | if (data & ~msr_ent.data) | |
3023 | return 1; | |
3024 | ||
3025 | vcpu->arch.perf_capabilities = data; | |
3026 | ||
3027 | return 0; | |
3028 | } | |
15c4a640 | 3029 | case MSR_EFER: |
11988499 | 3030 | return set_efer(vcpu, msr_info); |
8f1589d9 AP |
3031 | case MSR_K7_HWCR: |
3032 | data &= ~(u64)0x40; /* ignore flush filter disable */ | |
82494028 | 3033 | data &= ~(u64)0x100; /* ignore ignne emulation enable */ |
a223c313 | 3034 | data &= ~(u64)0x8; /* ignore TLB cache disable */ |
191c8137 BP |
3035 | |
3036 | /* Handle McStatusWrEn */ | |
3037 | if (data == BIT_ULL(18)) { | |
3038 | vcpu->arch.msr_hwcr = data; | |
3039 | } else if (data != 0) { | |
a737f256 CD |
3040 | vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n", |
3041 | data); | |
8f1589d9 AP |
3042 | return 1; |
3043 | } | |
15c4a640 | 3044 | break; |
f7c6d140 AP |
3045 | case MSR_FAM10H_MMIO_CONF_BASE: |
3046 | if (data != 0) { | |
a737f256 CD |
3047 | vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: " |
3048 | "0x%llx\n", data); | |
f7c6d140 AP |
3049 | return 1; |
3050 | } | |
15c4a640 | 3051 | break; |
b5e2fec0 AG |
3052 | case MSR_IA32_DEBUGCTLMSR: |
3053 | if (!data) { | |
3054 | /* We support the non-activated case already */ | |
3055 | break; | |
3056 | } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) { | |
3057 | /* Values other than LBR and BTF are vendor-specific, | |
3058 | thus reserved and should throw a #GP */ | |
3059 | return 1; | |
3060 | } | |
a737f256 CD |
3061 | vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n", |
3062 | __func__, data); | |
b5e2fec0 | 3063 | break; |
9ba075a6 | 3064 | case 0x200 ... 0x2ff: |
ff53604b | 3065 | return kvm_mtrr_set_msr(vcpu, msr, data); |
15c4a640 | 3066 | case MSR_IA32_APICBASE: |
58cb628d | 3067 | return kvm_set_apic_base(vcpu, msr_info); |
bf10bd0b | 3068 | case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff: |
0105d1a5 | 3069 | return kvm_x2apic_msr_write(vcpu, msr, data); |
a3e06bbe LJ |
3070 | case MSR_IA32_TSCDEADLINE: |
3071 | kvm_set_lapic_tscdeadline_msr(vcpu, data); | |
3072 | break; | |
ba904635 | 3073 | case MSR_IA32_TSC_ADJUST: |
d6321d49 | 3074 | if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) { |
ba904635 | 3075 | if (!msr_info->host_initiated) { |
d913b904 | 3076 | s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr; |
d7add054 | 3077 | adjust_tsc_offset_guest(vcpu, adj); |
ba904635 WA |
3078 | } |
3079 | vcpu->arch.ia32_tsc_adjust_msr = data; | |
3080 | } | |
3081 | break; | |
15c4a640 | 3082 | case MSR_IA32_MISC_ENABLE: |
511a8556 WL |
3083 | if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) && |
3084 | ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) { | |
3085 | if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3)) | |
3086 | return 1; | |
3087 | vcpu->arch.ia32_misc_enable_msr = data; | |
aedbaf4f | 3088 | kvm_update_cpuid_runtime(vcpu); |
511a8556 WL |
3089 | } else { |
3090 | vcpu->arch.ia32_misc_enable_msr = data; | |
3091 | } | |
15c4a640 | 3092 | break; |
64d60670 PB |
3093 | case MSR_IA32_SMBASE: |
3094 | if (!msr_info->host_initiated) | |
3095 | return 1; | |
3096 | vcpu->arch.smbase = data; | |
3097 | break; | |
73f624f4 PB |
3098 | case MSR_IA32_POWER_CTL: |
3099 | vcpu->arch.msr_ia32_power_ctl = data; | |
3100 | break; | |
dd259935 | 3101 | case MSR_IA32_TSC: |
0c899c25 PB |
3102 | if (msr_info->host_initiated) { |
3103 | kvm_synchronize_tsc(vcpu, data); | |
3104 | } else { | |
3105 | u64 adj = kvm_compute_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset; | |
3106 | adjust_tsc_offset_guest(vcpu, adj); | |
3107 | vcpu->arch.ia32_tsc_adjust_msr += adj; | |
3108 | } | |
dd259935 | 3109 | break; |
864e2ab2 AL |
3110 | case MSR_IA32_XSS: |
3111 | if (!msr_info->host_initiated && | |
3112 | !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES)) | |
3113 | return 1; | |
3114 | /* | |
a1bead2a SC |
3115 | * KVM supports exposing PT to the guest, but does not support |
3116 | * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than | |
3117 | * XSAVES/XRSTORS to save/restore PT MSRs. | |
864e2ab2 | 3118 | */ |
408e9a31 | 3119 | if (data & ~supported_xss) |
864e2ab2 AL |
3120 | return 1; |
3121 | vcpu->arch.ia32_xss = data; | |
3122 | break; | |
52797bf9 LA |
3123 | case MSR_SMI_COUNT: |
3124 | if (!msr_info->host_initiated) | |
3125 | return 1; | |
3126 | vcpu->arch.smi_count = data; | |
3127 | break; | |
11c6bffa | 3128 | case MSR_KVM_WALL_CLOCK_NEW: |
66570e96 OU |
3129 | if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2)) |
3130 | return 1; | |
3131 | ||
3132 | kvm_write_wall_clock(vcpu->kvm, data); | |
3133 | break; | |
18068523 | 3134 | case MSR_KVM_WALL_CLOCK: |
66570e96 OU |
3135 | if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE)) |
3136 | return 1; | |
3137 | ||
18068523 GOC |
3138 | kvm_write_wall_clock(vcpu->kvm, data); |
3139 | break; | |
11c6bffa | 3140 | case MSR_KVM_SYSTEM_TIME_NEW: |
66570e96 OU |
3141 | if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2)) |
3142 | return 1; | |
3143 | ||
5b9bb0eb OU |
3144 | kvm_write_system_time(vcpu, data, false, msr_info->host_initiated); |
3145 | break; | |
3146 | case MSR_KVM_SYSTEM_TIME: | |
66570e96 OU |
3147 | if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE)) |
3148 | return 1; | |
3149 | ||
3150 | kvm_write_system_time(vcpu, data, true, msr_info->host_initiated); | |
18068523 | 3151 | break; |
344d9588 | 3152 | case MSR_KVM_ASYNC_PF_EN: |
66570e96 OU |
3153 | if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF)) |
3154 | return 1; | |
3155 | ||
344d9588 GN |
3156 | if (kvm_pv_enable_async_pf(vcpu, data)) |
3157 | return 1; | |
3158 | break; | |
2635b5c4 | 3159 | case MSR_KVM_ASYNC_PF_INT: |
66570e96 OU |
3160 | if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT)) |
3161 | return 1; | |
3162 | ||
2635b5c4 VK |
3163 | if (kvm_pv_enable_async_pf_int(vcpu, data)) |
3164 | return 1; | |
3165 | break; | |
557a961a | 3166 | case MSR_KVM_ASYNC_PF_ACK: |
66570e96 OU |
3167 | if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF)) |
3168 | return 1; | |
557a961a VK |
3169 | if (data & 0x1) { |
3170 | vcpu->arch.apf.pageready_pending = false; | |
3171 | kvm_check_async_pf_completion(vcpu); | |
3172 | } | |
3173 | break; | |
c9aaa895 | 3174 | case MSR_KVM_STEAL_TIME: |
66570e96 OU |
3175 | if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME)) |
3176 | return 1; | |
c9aaa895 GC |
3177 | |
3178 | if (unlikely(!sched_info_on())) | |
3179 | return 1; | |
3180 | ||
3181 | if (data & KVM_STEAL_RESERVED_MASK) | |
3182 | return 1; | |
3183 | ||
c9aaa895 GC |
3184 | vcpu->arch.st.msr_val = data; |
3185 | ||
3186 | if (!(data & KVM_MSR_ENABLED)) | |
3187 | break; | |
3188 | ||
c9aaa895 GC |
3189 | kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu); |
3190 | ||
3191 | break; | |
ae7a2a3f | 3192 | case MSR_KVM_PV_EOI_EN: |
66570e96 OU |
3193 | if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI)) |
3194 | return 1; | |
3195 | ||
72bbf935 | 3196 | if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8))) |
ae7a2a3f MT |
3197 | return 1; |
3198 | break; | |
c9aaa895 | 3199 | |
2d5ba19b | 3200 | case MSR_KVM_POLL_CONTROL: |
66570e96 OU |
3201 | if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL)) |
3202 | return 1; | |
3203 | ||
2d5ba19b MT |
3204 | /* only enable bit supported */ |
3205 | if (data & (-1ULL << 1)) | |
3206 | return 1; | |
3207 | ||
3208 | vcpu->arch.msr_kvm_poll_control = data; | |
3209 | break; | |
3210 | ||
890ca9ae HY |
3211 | case MSR_IA32_MCG_CTL: |
3212 | case MSR_IA32_MCG_STATUS: | |
81760dcc | 3213 | case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1: |
9ffd986c | 3214 | return set_msr_mce(vcpu, msr_info); |
71db6023 | 3215 | |
6912ac32 WH |
3216 | case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3: |
3217 | case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1: | |
df561f66 GS |
3218 | pr = true; |
3219 | fallthrough; | |
6912ac32 WH |
3220 | case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3: |
3221 | case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1: | |
c6702c9d | 3222 | if (kvm_pmu_is_valid_msr(vcpu, msr)) |
afd80d85 | 3223 | return kvm_pmu_set_msr(vcpu, msr_info); |
5753785f GN |
3224 | |
3225 | if (pr || data != 0) | |
a737f256 CD |
3226 | vcpu_unimpl(vcpu, "disabled perfctr wrmsr: " |
3227 | "0x%x data 0x%llx\n", msr, data); | |
5753785f | 3228 | break; |
84e0cefa JS |
3229 | case MSR_K7_CLK_CTL: |
3230 | /* | |
3231 | * Ignore all writes to this no longer documented MSR. | |
3232 | * Writes are only relevant for old K7 processors, | |
3233 | * all pre-dating SVM, but a recommended workaround from | |
4a969980 | 3234 | * AMD for these chips. It is possible to specify the |
84e0cefa JS |
3235 | * affected processor models on the command line, hence |
3236 | * the need to ignore the workaround. | |
3237 | */ | |
3238 | break; | |
55cd8e5a | 3239 | case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15: |
f97f5a56 JD |
3240 | case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER: |
3241 | case HV_X64_MSR_SYNDBG_OPTIONS: | |
e7d9513b AS |
3242 | case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4: |
3243 | case HV_X64_MSR_CRASH_CTL: | |
1f4b34f8 | 3244 | case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT: |
a2e164e7 VK |
3245 | case HV_X64_MSR_REENLIGHTENMENT_CONTROL: |
3246 | case HV_X64_MSR_TSC_EMULATION_CONTROL: | |
3247 | case HV_X64_MSR_TSC_EMULATION_STATUS: | |
e7d9513b AS |
3248 | return kvm_hv_set_msr_common(vcpu, msr, data, |
3249 | msr_info->host_initiated); | |
91c9c3ed | 3250 | case MSR_IA32_BBL_CR_CTL3: |
3251 | /* Drop writes to this legacy MSR -- see rdmsr | |
3252 | * counterpart for further detail. | |
3253 | */ | |
fab0aa3b EM |
3254 | if (report_ignored_msrs) |
3255 | vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n", | |
3256 | msr, data); | |
91c9c3ed | 3257 | break; |
2b036c6b | 3258 | case MSR_AMD64_OSVW_ID_LENGTH: |
d6321d49 | 3259 | if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW)) |
2b036c6b BO |
3260 | return 1; |
3261 | vcpu->arch.osvw.length = data; | |
3262 | break; | |
3263 | case MSR_AMD64_OSVW_STATUS: | |
d6321d49 | 3264 | if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW)) |
2b036c6b BO |
3265 | return 1; |
3266 | vcpu->arch.osvw.status = data; | |
3267 | break; | |
db2336a8 KH |
3268 | case MSR_PLATFORM_INFO: |
3269 | if (!msr_info->host_initiated || | |
db2336a8 KH |
3270 | (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) && |
3271 | cpuid_fault_enabled(vcpu))) | |
3272 | return 1; | |
3273 | vcpu->arch.msr_platform_info = data; | |
3274 | break; | |
3275 | case MSR_MISC_FEATURES_ENABLES: | |
3276 | if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT || | |
3277 | (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT && | |
3278 | !supports_cpuid_fault(vcpu))) | |
3279 | return 1; | |
3280 | vcpu->arch.msr_misc_features_enables = data; | |
3281 | break; | |
15c4a640 | 3282 | default: |
ffde22ac ES |
3283 | if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr)) |
3284 | return xen_hvm_config(vcpu, data); | |
c6702c9d | 3285 | if (kvm_pmu_is_valid_msr(vcpu, msr)) |
afd80d85 | 3286 | return kvm_pmu_set_msr(vcpu, msr_info); |
6abe9c13 | 3287 | return KVM_MSR_RET_INVALID; |
15c4a640 CO |
3288 | } |
3289 | return 0; | |
3290 | } | |
3291 | EXPORT_SYMBOL_GPL(kvm_set_msr_common); | |
3292 | ||
44883f01 | 3293 | static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host) |
15c4a640 CO |
3294 | { |
3295 | u64 data; | |
890ca9ae HY |
3296 | u64 mcg_cap = vcpu->arch.mcg_cap; |
3297 | unsigned bank_num = mcg_cap & 0xff; | |
15c4a640 CO |
3298 | |
3299 | switch (msr) { | |
15c4a640 CO |
3300 | case MSR_IA32_P5_MC_ADDR: |
3301 | case MSR_IA32_P5_MC_TYPE: | |
890ca9ae HY |
3302 | data = 0; |
3303 | break; | |
15c4a640 | 3304 | case MSR_IA32_MCG_CAP: |
890ca9ae HY |
3305 | data = vcpu->arch.mcg_cap; |
3306 | break; | |
c7ac679c | 3307 | case MSR_IA32_MCG_CTL: |
44883f01 | 3308 | if (!(mcg_cap & MCG_CTL_P) && !host) |
890ca9ae HY |
3309 | return 1; |
3310 | data = vcpu->arch.mcg_ctl; | |
3311 | break; | |
3312 | case MSR_IA32_MCG_STATUS: | |
3313 | data = vcpu->arch.mcg_status; | |
3314 | break; | |
3315 | default: | |
3316 | if (msr >= MSR_IA32_MC0_CTL && | |
81760dcc | 3317 | msr < MSR_IA32_MCx_CTL(bank_num)) { |
6ec4c5ee MP |
3318 | u32 offset = array_index_nospec( |
3319 | msr - MSR_IA32_MC0_CTL, | |
3320 | MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL); | |
3321 | ||
890ca9ae HY |
3322 | data = vcpu->arch.mce_banks[offset]; |
3323 | break; | |
3324 | } | |
3325 | return 1; | |
3326 | } | |
3327 | *pdata = data; | |
3328 | return 0; | |
3329 | } | |
3330 | ||
609e36d3 | 3331 | int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
890ca9ae | 3332 | { |
609e36d3 | 3333 | switch (msr_info->index) { |
890ca9ae | 3334 | case MSR_IA32_PLATFORM_ID: |
15c4a640 | 3335 | case MSR_IA32_EBL_CR_POWERON: |
b5e2fec0 AG |
3336 | case MSR_IA32_DEBUGCTLMSR: |
3337 | case MSR_IA32_LASTBRANCHFROMIP: | |
3338 | case MSR_IA32_LASTBRANCHTOIP: | |
3339 | case MSR_IA32_LASTINTFROMIP: | |
3340 | case MSR_IA32_LASTINTTOIP: | |
60af2ecd | 3341 | case MSR_K8_SYSCFG: |
3afb1121 PB |
3342 | case MSR_K8_TSEG_ADDR: |
3343 | case MSR_K8_TSEG_MASK: | |
61a6bd67 | 3344 | case MSR_VM_HSAVE_PA: |
1fdbd48c | 3345 | case MSR_K8_INT_PENDING_MSG: |
c323c0e5 | 3346 | case MSR_AMD64_NB_CFG: |
f7c6d140 | 3347 | case MSR_FAM10H_MMIO_CONF_BASE: |
2e32b719 | 3348 | case MSR_AMD64_BU_CFG2: |
0c2df2a1 | 3349 | case MSR_IA32_PERF_CTL: |
405a353a | 3350 | case MSR_AMD64_DC_CFG: |
0e1b869f | 3351 | case MSR_F15H_EX_CFG: |
2ca1a06a VS |
3352 | /* |
3353 | * Intel Sandy Bridge CPUs must support the RAPL (running average power | |
3354 | * limit) MSRs. Just return 0, as we do not want to expose the host | |
3355 | * data here. Do not conditionalize this on CPUID, as KVM does not do | |
3356 | * so for existing CPU-specific MSRs. | |
3357 | */ | |
3358 | case MSR_RAPL_POWER_UNIT: | |
3359 | case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */ | |
3360 | case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */ | |
3361 | case MSR_PKG_ENERGY_STATUS: /* Total package */ | |
3362 | case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */ | |
609e36d3 | 3363 | msr_info->data = 0; |
15c4a640 | 3364 | break; |
c51eb52b | 3365 | case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5: |
6912ac32 WH |
3366 | case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3: |
3367 | case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3: | |
3368 | case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1: | |
3369 | case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1: | |
c6702c9d | 3370 | if (kvm_pmu_is_valid_msr(vcpu, msr_info->index)) |
cbd71758 | 3371 | return kvm_pmu_get_msr(vcpu, msr_info); |
609e36d3 | 3372 | msr_info->data = 0; |
5753785f | 3373 | break; |
742bc670 | 3374 | case MSR_IA32_UCODE_REV: |
518e7b94 | 3375 | msr_info->data = vcpu->arch.microcode_version; |
742bc670 | 3376 | break; |
0cf9135b SC |
3377 | case MSR_IA32_ARCH_CAPABILITIES: |
3378 | if (!msr_info->host_initiated && | |
3379 | !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES)) | |
3380 | return 1; | |
3381 | msr_info->data = vcpu->arch.arch_capabilities; | |
3382 | break; | |
d574c539 VK |
3383 | case MSR_IA32_PERF_CAPABILITIES: |
3384 | if (!msr_info->host_initiated && | |
3385 | !guest_cpuid_has(vcpu, X86_FEATURE_PDCM)) | |
3386 | return 1; | |
3387 | msr_info->data = vcpu->arch.perf_capabilities; | |
3388 | break; | |
73f624f4 PB |
3389 | case MSR_IA32_POWER_CTL: |
3390 | msr_info->data = vcpu->arch.msr_ia32_power_ctl; | |
3391 | break; | |
cc5b54dd ML |
3392 | case MSR_IA32_TSC: { |
3393 | /* | |
3394 | * Intel SDM states that MSR_IA32_TSC read adds the TSC offset | |
3395 | * even when not intercepted. AMD manual doesn't explicitly | |
3396 | * state this but appears to behave the same. | |
3397 | * | |
09e3e2a1 SC |
3398 | * Unconditionally return L1's TSC offset on userspace reads |
3399 | * so that userspace reads and writes always operate on L1's | |
3400 | * offset, e.g. to ensure deterministic behavior for migration. | |
cc5b54dd ML |
3401 | */ |
3402 | u64 tsc_offset = msr_info->host_initiated ? vcpu->arch.l1_tsc_offset : | |
3403 | vcpu->arch.tsc_offset; | |
3404 | ||
3405 | msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + tsc_offset; | |
dd259935 | 3406 | break; |
cc5b54dd | 3407 | } |
9ba075a6 | 3408 | case MSR_MTRRcap: |
9ba075a6 | 3409 | case 0x200 ... 0x2ff: |
ff53604b | 3410 | return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data); |
15c4a640 | 3411 | case 0xcd: /* fsb frequency */ |
609e36d3 | 3412 | msr_info->data = 3; |
15c4a640 | 3413 | break; |
7b914098 JS |
3414 | /* |
3415 | * MSR_EBC_FREQUENCY_ID | |
3416 | * Conservative value valid for even the basic CPU models. | |
3417 | * Models 0,1: 000 in bits 23:21 indicating a bus speed of | |
3418 | * 100MHz, model 2 000 in bits 18:16 indicating 100MHz, | |
3419 | * and 266MHz for model 3, or 4. Set Core Clock | |
3420 | * Frequency to System Bus Frequency Ratio to 1 (bits | |
3421 | * 31:24) even though these are only valid for CPU | |
3422 | * models > 2, however guests may end up dividing or | |
3423 | * multiplying by zero otherwise. | |
3424 | */ | |
3425 | case MSR_EBC_FREQUENCY_ID: | |
609e36d3 | 3426 | msr_info->data = 1 << 24; |
7b914098 | 3427 | break; |
15c4a640 | 3428 | case MSR_IA32_APICBASE: |
609e36d3 | 3429 | msr_info->data = kvm_get_apic_base(vcpu); |
15c4a640 | 3430 | break; |
bf10bd0b | 3431 | case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff: |
609e36d3 | 3432 | return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data); |
a3e06bbe | 3433 | case MSR_IA32_TSCDEADLINE: |
609e36d3 | 3434 | msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu); |
a3e06bbe | 3435 | break; |
ba904635 | 3436 | case MSR_IA32_TSC_ADJUST: |
609e36d3 | 3437 | msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr; |
ba904635 | 3438 | break; |
15c4a640 | 3439 | case MSR_IA32_MISC_ENABLE: |
609e36d3 | 3440 | msr_info->data = vcpu->arch.ia32_misc_enable_msr; |
15c4a640 | 3441 | break; |
64d60670 PB |
3442 | case MSR_IA32_SMBASE: |
3443 | if (!msr_info->host_initiated) | |
3444 | return 1; | |
3445 | msr_info->data = vcpu->arch.smbase; | |
15c4a640 | 3446 | break; |
52797bf9 LA |
3447 | case MSR_SMI_COUNT: |
3448 | msr_info->data = vcpu->arch.smi_count; | |
3449 | break; | |
847f0ad8 AG |
3450 | case MSR_IA32_PERF_STATUS: |
3451 | /* TSC increment by tick */ | |
609e36d3 | 3452 | msr_info->data = 1000ULL; |
847f0ad8 | 3453 | /* CPU multiplier */ |
b0996ae4 | 3454 | msr_info->data |= (((uint64_t)4ULL) << 40); |
847f0ad8 | 3455 | break; |
15c4a640 | 3456 | case MSR_EFER: |
609e36d3 | 3457 | msr_info->data = vcpu->arch.efer; |
15c4a640 | 3458 | break; |
18068523 | 3459 | case MSR_KVM_WALL_CLOCK: |
11c6bffa | 3460 | case MSR_KVM_WALL_CLOCK_NEW: |
609e36d3 | 3461 | msr_info->data = vcpu->kvm->arch.wall_clock; |
18068523 GOC |
3462 | break; |
3463 | case MSR_KVM_SYSTEM_TIME: | |
11c6bffa | 3464 | case MSR_KVM_SYSTEM_TIME_NEW: |
609e36d3 | 3465 | msr_info->data = vcpu->arch.time; |
18068523 | 3466 | break; |
344d9588 | 3467 | case MSR_KVM_ASYNC_PF_EN: |
2635b5c4 VK |
3468 | msr_info->data = vcpu->arch.apf.msr_en_val; |
3469 | break; | |
3470 | case MSR_KVM_ASYNC_PF_INT: | |
3471 | msr_info->data = vcpu->arch.apf.msr_int_val; | |
344d9588 | 3472 | break; |
557a961a VK |
3473 | case MSR_KVM_ASYNC_PF_ACK: |
3474 | msr_info->data = 0; | |
3475 | break; | |
c9aaa895 | 3476 | case MSR_KVM_STEAL_TIME: |
609e36d3 | 3477 | msr_info->data = vcpu->arch.st.msr_val; |
c9aaa895 | 3478 | break; |
1d92128f | 3479 | case MSR_KVM_PV_EOI_EN: |
609e36d3 | 3480 | msr_info->data = vcpu->arch.pv_eoi.msr_val; |
1d92128f | 3481 | break; |
2d5ba19b MT |
3482 | case MSR_KVM_POLL_CONTROL: |
3483 | msr_info->data = vcpu->arch.msr_kvm_poll_control; | |
3484 | break; | |
890ca9ae HY |
3485 | case MSR_IA32_P5_MC_ADDR: |
3486 | case MSR_IA32_P5_MC_TYPE: | |
3487 | case MSR_IA32_MCG_CAP: | |
3488 | case MSR_IA32_MCG_CTL: | |
3489 | case MSR_IA32_MCG_STATUS: | |
81760dcc | 3490 | case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1: |
44883f01 PB |
3491 | return get_msr_mce(vcpu, msr_info->index, &msr_info->data, |
3492 | msr_info->host_initiated); | |
864e2ab2 AL |
3493 | case MSR_IA32_XSS: |
3494 | if (!msr_info->host_initiated && | |
3495 | !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES)) | |
3496 | return 1; | |
3497 | msr_info->data = vcpu->arch.ia32_xss; | |
3498 | break; | |
84e0cefa JS |
3499 | case MSR_K7_CLK_CTL: |
3500 | /* | |
3501 | * Provide expected ramp-up count for K7. All other | |
3502 | * are set to zero, indicating minimum divisors for | |
3503 | * every field. | |
3504 | * | |
3505 | * This prevents guest kernels on AMD host with CPU | |
3506 | * type 6, model 8 and higher from exploding due to | |
3507 | * the rdmsr failing. | |
3508 | */ | |
609e36d3 | 3509 | msr_info->data = 0x20000000; |
84e0cefa | 3510 | break; |
55cd8e5a | 3511 | case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15: |
f97f5a56 JD |
3512 | case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER: |
3513 | case HV_X64_MSR_SYNDBG_OPTIONS: | |
e7d9513b AS |
3514 | case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4: |
3515 | case HV_X64_MSR_CRASH_CTL: | |
1f4b34f8 | 3516 | case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT: |
a2e164e7 VK |
3517 | case HV_X64_MSR_REENLIGHTENMENT_CONTROL: |
3518 | case HV_X64_MSR_TSC_EMULATION_CONTROL: | |
3519 | case HV_X64_MSR_TSC_EMULATION_STATUS: | |
e83d5887 | 3520 | return kvm_hv_get_msr_common(vcpu, |
44883f01 PB |
3521 | msr_info->index, &msr_info->data, |
3522 | msr_info->host_initiated); | |
91c9c3ed | 3523 | case MSR_IA32_BBL_CR_CTL3: |
3524 | /* This legacy MSR exists but isn't fully documented in current | |
3525 | * silicon. It is however accessed by winxp in very narrow | |
3526 | * scenarios where it sets bit #19, itself documented as | |
3527 | * a "reserved" bit. Best effort attempt to source coherent | |
3528 | * read data here should the balance of the register be | |
3529 | * interpreted by the guest: | |
3530 | * | |
3531 | * L2 cache control register 3: 64GB range, 256KB size, | |
3532 | * enabled, latency 0x1, configured | |
3533 | */ | |
609e36d3 | 3534 | msr_info->data = 0xbe702111; |
91c9c3ed | 3535 | break; |
2b036c6b | 3536 | case MSR_AMD64_OSVW_ID_LENGTH: |
d6321d49 | 3537 | if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW)) |
2b036c6b | 3538 | return 1; |
609e36d3 | 3539 | msr_info->data = vcpu->arch.osvw.length; |
2b036c6b BO |
3540 | break; |
3541 | case MSR_AMD64_OSVW_STATUS: | |
d6321d49 | 3542 | if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW)) |
2b036c6b | 3543 | return 1; |
609e36d3 | 3544 | msr_info->data = vcpu->arch.osvw.status; |
2b036c6b | 3545 | break; |
db2336a8 | 3546 | case MSR_PLATFORM_INFO: |
6fbbde9a DS |
3547 | if (!msr_info->host_initiated && |
3548 | !vcpu->kvm->arch.guest_can_read_msr_platform_info) | |
3549 | return 1; | |
db2336a8 KH |
3550 | msr_info->data = vcpu->arch.msr_platform_info; |
3551 | break; | |
3552 | case MSR_MISC_FEATURES_ENABLES: | |
3553 | msr_info->data = vcpu->arch.msr_misc_features_enables; | |
3554 | break; | |
191c8137 BP |
3555 | case MSR_K7_HWCR: |
3556 | msr_info->data = vcpu->arch.msr_hwcr; | |
3557 | break; | |
15c4a640 | 3558 | default: |
c6702c9d | 3559 | if (kvm_pmu_is_valid_msr(vcpu, msr_info->index)) |
cbd71758 | 3560 | return kvm_pmu_get_msr(vcpu, msr_info); |
6abe9c13 | 3561 | return KVM_MSR_RET_INVALID; |
15c4a640 | 3562 | } |
15c4a640 CO |
3563 | return 0; |
3564 | } | |
3565 | EXPORT_SYMBOL_GPL(kvm_get_msr_common); | |
3566 | ||
313a3dc7 CO |
3567 | /* |
3568 | * Read or write a bunch of msrs. All parameters are kernel addresses. | |
3569 | * | |
3570 | * @return number of msrs set successfully. | |
3571 | */ | |
3572 | static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs, | |
3573 | struct kvm_msr_entry *entries, | |
3574 | int (*do_msr)(struct kvm_vcpu *vcpu, | |
3575 | unsigned index, u64 *data)) | |
3576 | { | |
801e459a | 3577 | int i; |
313a3dc7 | 3578 | |
313a3dc7 CO |
3579 | for (i = 0; i < msrs->nmsrs; ++i) |
3580 | if (do_msr(vcpu, entries[i].index, &entries[i].data)) | |
3581 | break; | |
3582 | ||
313a3dc7 CO |
3583 | return i; |
3584 | } | |
3585 | ||
3586 | /* | |
3587 | * Read or write a bunch of msrs. Parameters are user addresses. | |
3588 | * | |
3589 | * @return number of msrs set successfully. | |
3590 | */ | |
3591 | static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs, | |
3592 | int (*do_msr)(struct kvm_vcpu *vcpu, | |
3593 | unsigned index, u64 *data), | |
3594 | int writeback) | |
3595 | { | |
3596 | struct kvm_msrs msrs; | |
3597 | struct kvm_msr_entry *entries; | |
3598 | int r, n; | |
3599 | unsigned size; | |
3600 | ||
3601 | r = -EFAULT; | |
0e96f31e | 3602 | if (copy_from_user(&msrs, user_msrs, sizeof(msrs))) |
313a3dc7 CO |
3603 | goto out; |
3604 | ||
3605 | r = -E2BIG; | |
3606 | if (msrs.nmsrs >= MAX_IO_MSRS) | |
3607 | goto out; | |
3608 | ||
313a3dc7 | 3609 | size = sizeof(struct kvm_msr_entry) * msrs.nmsrs; |
ff5c2c03 SL |
3610 | entries = memdup_user(user_msrs->entries, size); |
3611 | if (IS_ERR(entries)) { | |
3612 | r = PTR_ERR(entries); | |
313a3dc7 | 3613 | goto out; |
ff5c2c03 | 3614 | } |
313a3dc7 CO |
3615 | |
3616 | r = n = __msr_io(vcpu, &msrs, entries, do_msr); | |
3617 | if (r < 0) | |
3618 | goto out_free; | |
3619 | ||
3620 | r = -EFAULT; | |
3621 | if (writeback && copy_to_user(user_msrs->entries, entries, size)) | |
3622 | goto out_free; | |
3623 | ||
3624 | r = n; | |
3625 | ||
3626 | out_free: | |
7a73c028 | 3627 | kfree(entries); |
313a3dc7 CO |
3628 | out: |
3629 | return r; | |
3630 | } | |
3631 | ||
4d5422ce WL |
3632 | static inline bool kvm_can_mwait_in_guest(void) |
3633 | { | |
3634 | return boot_cpu_has(X86_FEATURE_MWAIT) && | |
8e9b29b6 KA |
3635 | !boot_cpu_has_bug(X86_BUG_MONITOR) && |
3636 | boot_cpu_has(X86_FEATURE_ARAT); | |
4d5422ce WL |
3637 | } |
3638 | ||
784aa3d7 | 3639 | int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) |
018d00d2 | 3640 | { |
4d5422ce | 3641 | int r = 0; |
018d00d2 ZX |
3642 | |
3643 | switch (ext) { | |
3644 | case KVM_CAP_IRQCHIP: | |
3645 | case KVM_CAP_HLT: | |
3646 | case KVM_CAP_MMU_SHADOW_CACHE_CONTROL: | |
018d00d2 | 3647 | case KVM_CAP_SET_TSS_ADDR: |
07716717 | 3648 | case KVM_CAP_EXT_CPUID: |
9c15bb1d | 3649 | case KVM_CAP_EXT_EMUL_CPUID: |
c8076604 | 3650 | case KVM_CAP_CLOCKSOURCE: |
7837699f | 3651 | case KVM_CAP_PIT: |
a28e4f5a | 3652 | case KVM_CAP_NOP_IO_DELAY: |
62d9f0db | 3653 | case KVM_CAP_MP_STATE: |
ed848624 | 3654 | case KVM_CAP_SYNC_MMU: |
a355c85c | 3655 | case KVM_CAP_USER_NMI: |
52d939a0 | 3656 | case KVM_CAP_REINJECT_CONTROL: |
4925663a | 3657 | case KVM_CAP_IRQ_INJECT_STATUS: |
d34e6b17 | 3658 | case KVM_CAP_IOEVENTFD: |
f848a5a8 | 3659 | case KVM_CAP_IOEVENTFD_NO_LENGTH: |
c5ff41ce | 3660 | case KVM_CAP_PIT2: |
e9f42757 | 3661 | case KVM_CAP_PIT_STATE2: |
b927a3ce | 3662 | case KVM_CAP_SET_IDENTITY_MAP_ADDR: |
ffde22ac | 3663 | case KVM_CAP_XEN_HVM: |
3cfc3092 | 3664 | case KVM_CAP_VCPU_EVENTS: |
55cd8e5a | 3665 | case KVM_CAP_HYPERV: |
10388a07 | 3666 | case KVM_CAP_HYPERV_VAPIC: |
c25bc163 | 3667 | case KVM_CAP_HYPERV_SPIN: |
5c919412 | 3668 | case KVM_CAP_HYPERV_SYNIC: |
efc479e6 | 3669 | case KVM_CAP_HYPERV_SYNIC2: |
d3457c87 | 3670 | case KVM_CAP_HYPERV_VP_INDEX: |
faeb7833 | 3671 | case KVM_CAP_HYPERV_EVENTFD: |
c1aea919 | 3672 | case KVM_CAP_HYPERV_TLBFLUSH: |
214ff83d | 3673 | case KVM_CAP_HYPERV_SEND_IPI: |
2bc39970 | 3674 | case KVM_CAP_HYPERV_CPUID: |
ab9f4ecb | 3675 | case KVM_CAP_PCI_SEGMENT: |
a1efbe77 | 3676 | case KVM_CAP_DEBUGREGS: |
d2be1651 | 3677 | case KVM_CAP_X86_ROBUST_SINGLESTEP: |
2d5b5a66 | 3678 | case KVM_CAP_XSAVE: |
344d9588 | 3679 | case KVM_CAP_ASYNC_PF: |
72de5fa4 | 3680 | case KVM_CAP_ASYNC_PF_INT: |
92a1f12d | 3681 | case KVM_CAP_GET_TSC_KHZ: |
1c0b28c2 | 3682 | case KVM_CAP_KVMCLOCK_CTRL: |
4d8b81ab | 3683 | case KVM_CAP_READONLY_MEM: |
5f66b620 | 3684 | case KVM_CAP_HYPERV_TIME: |
100943c5 | 3685 | case KVM_CAP_IOAPIC_POLARITY_IGNORED: |
defcf51f | 3686 | case KVM_CAP_TSC_DEADLINE_TIMER: |
90de4a18 | 3687 | case KVM_CAP_DISABLE_QUIRKS: |
d71ba788 | 3688 | case KVM_CAP_SET_BOOT_CPU_ID: |
49df6397 | 3689 | case KVM_CAP_SPLIT_IRQCHIP: |
460df4c1 | 3690 | case KVM_CAP_IMMEDIATE_EXIT: |
66bb8a06 | 3691 | case KVM_CAP_PMU_EVENT_FILTER: |
801e459a | 3692 | case KVM_CAP_GET_MSR_FEATURES: |
6fbbde9a | 3693 | case KVM_CAP_MSR_PLATFORM_INFO: |
c4f55198 | 3694 | case KVM_CAP_EXCEPTION_PAYLOAD: |
b9b2782c | 3695 | case KVM_CAP_SET_GUEST_DEBUG: |
1aa561b1 | 3696 | case KVM_CAP_LAST_CPU: |
1ae09954 | 3697 | case KVM_CAP_X86_USER_SPACE_MSR: |
1a155254 | 3698 | case KVM_CAP_X86_MSR_FILTER: |
66570e96 | 3699 | case KVM_CAP_ENFORCE_PV_FEATURE_CPUID: |
018d00d2 ZX |
3700 | r = 1; |
3701 | break; | |
01643c51 KH |
3702 | case KVM_CAP_SYNC_REGS: |
3703 | r = KVM_SYNC_X86_VALID_FIELDS; | |
3704 | break; | |
e3fd9a93 PB |
3705 | case KVM_CAP_ADJUST_CLOCK: |
3706 | r = KVM_CLOCK_TSC_STABLE; | |
3707 | break; | |
4d5422ce | 3708 | case KVM_CAP_X86_DISABLE_EXITS: |
b5170063 WL |
3709 | r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE | |
3710 | KVM_X86_DISABLE_EXITS_CSTATE; | |
4d5422ce WL |
3711 | if(kvm_can_mwait_in_guest()) |
3712 | r |= KVM_X86_DISABLE_EXITS_MWAIT; | |
668fffa3 | 3713 | break; |
6d396b55 PB |
3714 | case KVM_CAP_X86_SMM: |
3715 | /* SMBASE is usually relocated above 1M on modern chipsets, | |
3716 | * and SMM handlers might indeed rely on 4G segment limits, | |
3717 | * so do not report SMM to be available if real mode is | |
3718 | * emulated via vm86 mode. Still, do not go to great lengths | |
3719 | * to avoid userspace's usage of the feature, because it is a | |
3720 | * fringe case that is not enabled except via specific settings | |
3721 | * of the module parameters. | |
3722 | */ | |
afaf0b2f | 3723 | r = kvm_x86_ops.has_emulated_msr(MSR_IA32_SMBASE); |
6d396b55 | 3724 | break; |
774ead3a | 3725 | case KVM_CAP_VAPIC: |
afaf0b2f | 3726 | r = !kvm_x86_ops.cpu_has_accelerated_tpr(); |
774ead3a | 3727 | break; |
f725230a | 3728 | case KVM_CAP_NR_VCPUS: |
8c3ba334 SL |
3729 | r = KVM_SOFT_MAX_VCPUS; |
3730 | break; | |
3731 | case KVM_CAP_MAX_VCPUS: | |
f725230a AK |
3732 | r = KVM_MAX_VCPUS; |
3733 | break; | |
a86cb413 TH |
3734 | case KVM_CAP_MAX_VCPU_ID: |
3735 | r = KVM_MAX_VCPU_ID; | |
3736 | break; | |
a68a6a72 MT |
3737 | case KVM_CAP_PV_MMU: /* obsolete */ |
3738 | r = 0; | |
2f333bcb | 3739 | break; |
890ca9ae HY |
3740 | case KVM_CAP_MCE: |
3741 | r = KVM_MAX_MCE_BANKS; | |
3742 | break; | |
2d5b5a66 | 3743 | case KVM_CAP_XCRS: |
d366bf7e | 3744 | r = boot_cpu_has(X86_FEATURE_XSAVE); |
2d5b5a66 | 3745 | break; |
92a1f12d JR |
3746 | case KVM_CAP_TSC_CONTROL: |
3747 | r = kvm_has_tsc_control; | |
3748 | break; | |
37131313 RK |
3749 | case KVM_CAP_X2APIC_API: |
3750 | r = KVM_X2APIC_API_VALID_FLAGS; | |
3751 | break; | |
8fcc4b59 | 3752 | case KVM_CAP_NESTED_STATE: |
33b22172 PB |
3753 | r = kvm_x86_ops.nested_ops->get_state ? |
3754 | kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0; | |
8fcc4b59 | 3755 | break; |
344c6c80 | 3756 | case KVM_CAP_HYPERV_DIRECT_TLBFLUSH: |
afaf0b2f | 3757 | r = kvm_x86_ops.enable_direct_tlbflush != NULL; |
5a0165f6 VK |
3758 | break; |
3759 | case KVM_CAP_HYPERV_ENLIGHTENED_VMCS: | |
33b22172 | 3760 | r = kvm_x86_ops.nested_ops->enable_evmcs != NULL; |
344c6c80 | 3761 | break; |
3edd6839 MG |
3762 | case KVM_CAP_SMALLER_MAXPHYADDR: |
3763 | r = (int) allow_smaller_maxphyaddr; | |
3764 | break; | |
004a0124 AJ |
3765 | case KVM_CAP_STEAL_TIME: |
3766 | r = sched_info_on(); | |
3767 | break; | |
018d00d2 | 3768 | default: |
018d00d2 ZX |
3769 | break; |
3770 | } | |
3771 | return r; | |
3772 | ||
3773 | } | |
3774 | ||
043405e1 CO |
3775 | long kvm_arch_dev_ioctl(struct file *filp, |
3776 | unsigned int ioctl, unsigned long arg) | |
3777 | { | |
3778 | void __user *argp = (void __user *)arg; | |
3779 | long r; | |
3780 | ||
3781 | switch (ioctl) { | |
3782 | case KVM_GET_MSR_INDEX_LIST: { | |
3783 | struct kvm_msr_list __user *user_msr_list = argp; | |
3784 | struct kvm_msr_list msr_list; | |
3785 | unsigned n; | |
3786 | ||
3787 | r = -EFAULT; | |
0e96f31e | 3788 | if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list))) |
043405e1 CO |
3789 | goto out; |
3790 | n = msr_list.nmsrs; | |
62ef68bb | 3791 | msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs; |
0e96f31e | 3792 | if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list))) |
043405e1 CO |
3793 | goto out; |
3794 | r = -E2BIG; | |
e125e7b6 | 3795 | if (n < msr_list.nmsrs) |
043405e1 CO |
3796 | goto out; |
3797 | r = -EFAULT; | |
3798 | if (copy_to_user(user_msr_list->indices, &msrs_to_save, | |
3799 | num_msrs_to_save * sizeof(u32))) | |
3800 | goto out; | |
e125e7b6 | 3801 | if (copy_to_user(user_msr_list->indices + num_msrs_to_save, |
043405e1 | 3802 | &emulated_msrs, |
62ef68bb | 3803 | num_emulated_msrs * sizeof(u32))) |
043405e1 CO |
3804 | goto out; |
3805 | r = 0; | |
3806 | break; | |
3807 | } | |
9c15bb1d BP |
3808 | case KVM_GET_SUPPORTED_CPUID: |
3809 | case KVM_GET_EMULATED_CPUID: { | |
674eea0f AK |
3810 | struct kvm_cpuid2 __user *cpuid_arg = argp; |
3811 | struct kvm_cpuid2 cpuid; | |
3812 | ||
3813 | r = -EFAULT; | |
0e96f31e | 3814 | if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid))) |
674eea0f | 3815 | goto out; |
9c15bb1d BP |
3816 | |
3817 | r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries, | |
3818 | ioctl); | |
674eea0f AK |
3819 | if (r) |
3820 | goto out; | |
3821 | ||
3822 | r = -EFAULT; | |
0e96f31e | 3823 | if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid))) |
674eea0f AK |
3824 | goto out; |
3825 | r = 0; | |
3826 | break; | |
3827 | } | |
cf6c26ec | 3828 | case KVM_X86_GET_MCE_CAP_SUPPORTED: |
890ca9ae | 3829 | r = -EFAULT; |
c45dcc71 AR |
3830 | if (copy_to_user(argp, &kvm_mce_cap_supported, |
3831 | sizeof(kvm_mce_cap_supported))) | |
890ca9ae HY |
3832 | goto out; |
3833 | r = 0; | |
3834 | break; | |
801e459a TL |
3835 | case KVM_GET_MSR_FEATURE_INDEX_LIST: { |
3836 | struct kvm_msr_list __user *user_msr_list = argp; | |
3837 | struct kvm_msr_list msr_list; | |
3838 | unsigned int n; | |
3839 | ||
3840 | r = -EFAULT; | |
3841 | if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list))) | |
3842 | goto out; | |
3843 | n = msr_list.nmsrs; | |
3844 | msr_list.nmsrs = num_msr_based_features; | |
3845 | if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list))) | |
3846 | goto out; | |
3847 | r = -E2BIG; | |
3848 | if (n < msr_list.nmsrs) | |
3849 | goto out; | |
3850 | r = -EFAULT; | |
3851 | if (copy_to_user(user_msr_list->indices, &msr_based_features, | |
3852 | num_msr_based_features * sizeof(u32))) | |
3853 | goto out; | |
3854 | r = 0; | |
3855 | break; | |
3856 | } | |
3857 | case KVM_GET_MSRS: | |
3858 | r = msr_io(NULL, argp, do_get_msr_feature, 1); | |
3859 | break; | |
043405e1 CO |
3860 | default: |
3861 | r = -EINVAL; | |
cf6c26ec | 3862 | break; |
043405e1 CO |
3863 | } |
3864 | out: | |
3865 | return r; | |
3866 | } | |
3867 | ||
f5f48ee1 SY |
3868 | static void wbinvd_ipi(void *garbage) |
3869 | { | |
3870 | wbinvd(); | |
3871 | } | |
3872 | ||
3873 | static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu) | |
3874 | { | |
e0f0bbc5 | 3875 | return kvm_arch_has_noncoherent_dma(vcpu->kvm); |
f5f48ee1 SY |
3876 | } |
3877 | ||
313a3dc7 CO |
3878 | void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
3879 | { | |
f5f48ee1 SY |
3880 | /* Address WBINVD may be executed by guest */ |
3881 | if (need_emulate_wbinvd(vcpu)) { | |
afaf0b2f | 3882 | if (kvm_x86_ops.has_wbinvd_exit()) |
f5f48ee1 SY |
3883 | cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask); |
3884 | else if (vcpu->cpu != -1 && vcpu->cpu != cpu) | |
3885 | smp_call_function_single(vcpu->cpu, | |
3886 | wbinvd_ipi, NULL, 1); | |
3887 | } | |
3888 | ||
afaf0b2f | 3889 | kvm_x86_ops.vcpu_load(vcpu, cpu); |
8f6055cb | 3890 | |
37486135 BM |
3891 | /* Save host pkru register if supported */ |
3892 | vcpu->arch.host_pkru = read_pkru(); | |
3893 | ||
0dd6a6ed ZA |
3894 | /* Apply any externally detected TSC adjustments (due to suspend) */ |
3895 | if (unlikely(vcpu->arch.tsc_offset_adjustment)) { | |
3896 | adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment); | |
3897 | vcpu->arch.tsc_offset_adjustment = 0; | |
105b21bb | 3898 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
0dd6a6ed | 3899 | } |
8f6055cb | 3900 | |
b0c39dc6 | 3901 | if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) { |
6f526ec5 | 3902 | s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 : |
4ea1636b | 3903 | rdtsc() - vcpu->arch.last_host_tsc; |
e48672fa ZA |
3904 | if (tsc_delta < 0) |
3905 | mark_tsc_unstable("KVM discovered backwards TSC"); | |
ce7a058a | 3906 | |
b0c39dc6 | 3907 | if (kvm_check_tsc_unstable()) { |
07c1419a | 3908 | u64 offset = kvm_compute_tsc_offset(vcpu, |
b183aa58 | 3909 | vcpu->arch.last_guest_tsc); |
a545ab6a | 3910 | kvm_vcpu_write_tsc_offset(vcpu, offset); |
c285545f | 3911 | vcpu->arch.tsc_catchup = 1; |
c285545f | 3912 | } |
a749e247 PB |
3913 | |
3914 | if (kvm_lapic_hv_timer_in_use(vcpu)) | |
3915 | kvm_lapic_restart_hv_timer(vcpu); | |
3916 | ||
d98d07ca MT |
3917 | /* |
3918 | * On a host with synchronized TSC, there is no need to update | |
3919 | * kvmclock on vcpu->cpu migration | |
3920 | */ | |
3921 | if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1) | |
0061d53d | 3922 | kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu); |
c285545f | 3923 | if (vcpu->cpu != cpu) |
1bd2009e | 3924 | kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu); |
e48672fa | 3925 | vcpu->cpu = cpu; |
6b7d7e76 | 3926 | } |
c9aaa895 | 3927 | |
c9aaa895 | 3928 | kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu); |
313a3dc7 CO |
3929 | } |
3930 | ||
0b9f6c46 PX |
3931 | static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu) |
3932 | { | |
b0431382 BO |
3933 | struct kvm_host_map map; |
3934 | struct kvm_steal_time *st; | |
3935 | ||
0b9f6c46 PX |
3936 | if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED)) |
3937 | return; | |
3938 | ||
a6bd811f | 3939 | if (vcpu->arch.st.preempted) |
8c6de56a BO |
3940 | return; |
3941 | ||
b0431382 BO |
3942 | if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map, |
3943 | &vcpu->arch.st.cache, true)) | |
3944 | return; | |
3945 | ||
3946 | st = map.hva + | |
3947 | offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS); | |
0b9f6c46 | 3948 | |
a6bd811f | 3949 | st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED; |
0b9f6c46 | 3950 | |
b0431382 | 3951 | kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true); |
0b9f6c46 PX |
3952 | } |
3953 | ||
313a3dc7 CO |
3954 | void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) |
3955 | { | |
cc0d907c | 3956 | int idx; |
de63ad4c LM |
3957 | |
3958 | if (vcpu->preempted) | |
afaf0b2f | 3959 | vcpu->arch.preempted_in_kernel = !kvm_x86_ops.get_cpl(vcpu); |
de63ad4c | 3960 | |
931f261b AA |
3961 | /* |
3962 | * Disable page faults because we're in atomic context here. | |
3963 | * kvm_write_guest_offset_cached() would call might_fault() | |
3964 | * that relies on pagefault_disable() to tell if there's a | |
3965 | * bug. NOTE: the write to guest memory may not go through if | |
3966 | * during postcopy live migration or if there's heavy guest | |
3967 | * paging. | |
3968 | */ | |
3969 | pagefault_disable(); | |
cc0d907c AA |
3970 | /* |
3971 | * kvm_memslots() will be called by | |
3972 | * kvm_write_guest_offset_cached() so take the srcu lock. | |
3973 | */ | |
3974 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
0b9f6c46 | 3975 | kvm_steal_time_set_preempted(vcpu); |
cc0d907c | 3976 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
931f261b | 3977 | pagefault_enable(); |
afaf0b2f | 3978 | kvm_x86_ops.vcpu_put(vcpu); |
4ea1636b | 3979 | vcpu->arch.last_host_tsc = rdtsc(); |
efdab992 | 3980 | /* |
f9dcf08e RK |
3981 | * If userspace has set any breakpoints or watchpoints, dr6 is restored |
3982 | * on every vmexit, but if not, we might have a stale dr6 from the | |
3983 | * guest. do_debug expects dr6 to be cleared after it runs, do the same. | |
efdab992 | 3984 | */ |
f9dcf08e | 3985 | set_debugreg(0, 6); |
313a3dc7 CO |
3986 | } |
3987 | ||
313a3dc7 CO |
3988 | static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu, |
3989 | struct kvm_lapic_state *s) | |
3990 | { | |
fa59cc00 | 3991 | if (vcpu->arch.apicv_active) |
afaf0b2f | 3992 | kvm_x86_ops.sync_pir_to_irr(vcpu); |
d62caabb | 3993 | |
a92e2543 | 3994 | return kvm_apic_get_state(vcpu, s); |
313a3dc7 CO |
3995 | } |
3996 | ||
3997 | static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu, | |
3998 | struct kvm_lapic_state *s) | |
3999 | { | |
a92e2543 RK |
4000 | int r; |
4001 | ||
4002 | r = kvm_apic_set_state(vcpu, s); | |
4003 | if (r) | |
4004 | return r; | |
cb142eb7 | 4005 | update_cr8_intercept(vcpu); |
313a3dc7 CO |
4006 | |
4007 | return 0; | |
4008 | } | |
4009 | ||
127a457a MG |
4010 | static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu) |
4011 | { | |
4012 | return (!lapic_in_kernel(vcpu) || | |
4013 | kvm_apic_accept_pic_intr(vcpu)); | |
4014 | } | |
4015 | ||
782d422b MG |
4016 | /* |
4017 | * if userspace requested an interrupt window, check that the | |
4018 | * interrupt window is open. | |
4019 | * | |
4020 | * No need to exit to userspace if we already have an interrupt queued. | |
4021 | */ | |
4022 | static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu) | |
4023 | { | |
4024 | return kvm_arch_interrupt_allowed(vcpu) && | |
4025 | !kvm_cpu_has_interrupt(vcpu) && | |
4026 | !kvm_event_needs_reinjection(vcpu) && | |
4027 | kvm_cpu_accept_dm_intr(vcpu); | |
4028 | } | |
4029 | ||
f77bc6a4 ZX |
4030 | static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, |
4031 | struct kvm_interrupt *irq) | |
4032 | { | |
02cdb50f | 4033 | if (irq->irq >= KVM_NR_INTERRUPTS) |
f77bc6a4 | 4034 | return -EINVAL; |
1c1a9ce9 SR |
4035 | |
4036 | if (!irqchip_in_kernel(vcpu->kvm)) { | |
4037 | kvm_queue_interrupt(vcpu, irq->irq, false); | |
4038 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
4039 | return 0; | |
4040 | } | |
4041 | ||
4042 | /* | |
4043 | * With in-kernel LAPIC, we only use this to inject EXTINT, so | |
4044 | * fail for in-kernel 8259. | |
4045 | */ | |
4046 | if (pic_in_kernel(vcpu->kvm)) | |
f77bc6a4 | 4047 | return -ENXIO; |
f77bc6a4 | 4048 | |
1c1a9ce9 SR |
4049 | if (vcpu->arch.pending_external_vector != -1) |
4050 | return -EEXIST; | |
f77bc6a4 | 4051 | |
1c1a9ce9 | 4052 | vcpu->arch.pending_external_vector = irq->irq; |
934bf653 | 4053 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
f77bc6a4 ZX |
4054 | return 0; |
4055 | } | |
4056 | ||
c4abb7c9 JK |
4057 | static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu) |
4058 | { | |
c4abb7c9 | 4059 | kvm_inject_nmi(vcpu); |
c4abb7c9 JK |
4060 | |
4061 | return 0; | |
4062 | } | |
4063 | ||
f077825a PB |
4064 | static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu) |
4065 | { | |
64d60670 PB |
4066 | kvm_make_request(KVM_REQ_SMI, vcpu); |
4067 | ||
f077825a PB |
4068 | return 0; |
4069 | } | |
4070 | ||
b209749f AK |
4071 | static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu, |
4072 | struct kvm_tpr_access_ctl *tac) | |
4073 | { | |
4074 | if (tac->flags) | |
4075 | return -EINVAL; | |
4076 | vcpu->arch.tpr_access_reporting = !!tac->enabled; | |
4077 | return 0; | |
4078 | } | |
4079 | ||
890ca9ae HY |
4080 | static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu, |
4081 | u64 mcg_cap) | |
4082 | { | |
4083 | int r; | |
4084 | unsigned bank_num = mcg_cap & 0xff, bank; | |
4085 | ||
4086 | r = -EINVAL; | |
c4e0e4ab | 4087 | if (!bank_num || bank_num > KVM_MAX_MCE_BANKS) |
890ca9ae | 4088 | goto out; |
c45dcc71 | 4089 | if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000)) |
890ca9ae HY |
4090 | goto out; |
4091 | r = 0; | |
4092 | vcpu->arch.mcg_cap = mcg_cap; | |
4093 | /* Init IA32_MCG_CTL to all 1s */ | |
4094 | if (mcg_cap & MCG_CTL_P) | |
4095 | vcpu->arch.mcg_ctl = ~(u64)0; | |
4096 | /* Init IA32_MCi_CTL to all 1s */ | |
4097 | for (bank = 0; bank < bank_num; bank++) | |
4098 | vcpu->arch.mce_banks[bank*4] = ~(u64)0; | |
c45dcc71 | 4099 | |
afaf0b2f | 4100 | kvm_x86_ops.setup_mce(vcpu); |
890ca9ae HY |
4101 | out: |
4102 | return r; | |
4103 | } | |
4104 | ||
4105 | static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu, | |
4106 | struct kvm_x86_mce *mce) | |
4107 | { | |
4108 | u64 mcg_cap = vcpu->arch.mcg_cap; | |
4109 | unsigned bank_num = mcg_cap & 0xff; | |
4110 | u64 *banks = vcpu->arch.mce_banks; | |
4111 | ||
4112 | if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL)) | |
4113 | return -EINVAL; | |
4114 | /* | |
4115 | * if IA32_MCG_CTL is not all 1s, the uncorrected error | |
4116 | * reporting is disabled | |
4117 | */ | |
4118 | if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) && | |
4119 | vcpu->arch.mcg_ctl != ~(u64)0) | |
4120 | return 0; | |
4121 | banks += 4 * mce->bank; | |
4122 | /* | |
4123 | * if IA32_MCi_CTL is not all 1s, the uncorrected error | |
4124 | * reporting is disabled for the bank | |
4125 | */ | |
4126 | if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0) | |
4127 | return 0; | |
4128 | if (mce->status & MCI_STATUS_UC) { | |
4129 | if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) || | |
fc78f519 | 4130 | !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) { |
a8eeb04a | 4131 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); |
890ca9ae HY |
4132 | return 0; |
4133 | } | |
4134 | if (banks[1] & MCI_STATUS_VAL) | |
4135 | mce->status |= MCI_STATUS_OVER; | |
4136 | banks[2] = mce->addr; | |
4137 | banks[3] = mce->misc; | |
4138 | vcpu->arch.mcg_status = mce->mcg_status; | |
4139 | banks[1] = mce->status; | |
4140 | kvm_queue_exception(vcpu, MC_VECTOR); | |
4141 | } else if (!(banks[1] & MCI_STATUS_VAL) | |
4142 | || !(banks[1] & MCI_STATUS_UC)) { | |
4143 | if (banks[1] & MCI_STATUS_VAL) | |
4144 | mce->status |= MCI_STATUS_OVER; | |
4145 | banks[2] = mce->addr; | |
4146 | banks[3] = mce->misc; | |
4147 | banks[1] = mce->status; | |
4148 | } else | |
4149 | banks[1] |= MCI_STATUS_OVER; | |
4150 | return 0; | |
4151 | } | |
4152 | ||
3cfc3092 JK |
4153 | static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu, |
4154 | struct kvm_vcpu_events *events) | |
4155 | { | |
7460fb4a | 4156 | process_nmi(vcpu); |
59073aaf | 4157 | |
a06230b6 OU |
4158 | /* |
4159 | * In guest mode, payload delivery should be deferred, | |
4160 | * so that the L1 hypervisor can intercept #PF before | |
4161 | * CR2 is modified (or intercept #DB before DR6 is | |
4162 | * modified under nVMX). Unless the per-VM capability, | |
4163 | * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of | |
4164 | * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we | |
4165 | * opportunistically defer the exception payload, deliver it if the | |
4166 | * capability hasn't been requested before processing a | |
4167 | * KVM_GET_VCPU_EVENTS. | |
4168 | */ | |
4169 | if (!vcpu->kvm->arch.exception_payload_enabled && | |
4170 | vcpu->arch.exception.pending && vcpu->arch.exception.has_payload) | |
4171 | kvm_deliver_exception_payload(vcpu); | |
4172 | ||
664f8e26 | 4173 | /* |
59073aaf JM |
4174 | * The API doesn't provide the instruction length for software |
4175 | * exceptions, so don't report them. As long as the guest RIP | |
4176 | * isn't advanced, we should expect to encounter the exception | |
4177 | * again. | |
664f8e26 | 4178 | */ |
59073aaf JM |
4179 | if (kvm_exception_is_soft(vcpu->arch.exception.nr)) { |
4180 | events->exception.injected = 0; | |
4181 | events->exception.pending = 0; | |
4182 | } else { | |
4183 | events->exception.injected = vcpu->arch.exception.injected; | |
4184 | events->exception.pending = vcpu->arch.exception.pending; | |
4185 | /* | |
4186 | * For ABI compatibility, deliberately conflate | |
4187 | * pending and injected exceptions when | |
4188 | * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled. | |
4189 | */ | |
4190 | if (!vcpu->kvm->arch.exception_payload_enabled) | |
4191 | events->exception.injected |= | |
4192 | vcpu->arch.exception.pending; | |
4193 | } | |
3cfc3092 JK |
4194 | events->exception.nr = vcpu->arch.exception.nr; |
4195 | events->exception.has_error_code = vcpu->arch.exception.has_error_code; | |
4196 | events->exception.error_code = vcpu->arch.exception.error_code; | |
59073aaf JM |
4197 | events->exception_has_payload = vcpu->arch.exception.has_payload; |
4198 | events->exception_payload = vcpu->arch.exception.payload; | |
3cfc3092 | 4199 | |
03b82a30 | 4200 | events->interrupt.injected = |
04140b41 | 4201 | vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft; |
3cfc3092 | 4202 | events->interrupt.nr = vcpu->arch.interrupt.nr; |
03b82a30 | 4203 | events->interrupt.soft = 0; |
afaf0b2f | 4204 | events->interrupt.shadow = kvm_x86_ops.get_interrupt_shadow(vcpu); |
3cfc3092 JK |
4205 | |
4206 | events->nmi.injected = vcpu->arch.nmi_injected; | |
7460fb4a | 4207 | events->nmi.pending = vcpu->arch.nmi_pending != 0; |
afaf0b2f | 4208 | events->nmi.masked = kvm_x86_ops.get_nmi_mask(vcpu); |
97e69aa6 | 4209 | events->nmi.pad = 0; |
3cfc3092 | 4210 | |
66450a21 | 4211 | events->sipi_vector = 0; /* never valid when reporting to user space */ |
3cfc3092 | 4212 | |
f077825a PB |
4213 | events->smi.smm = is_smm(vcpu); |
4214 | events->smi.pending = vcpu->arch.smi_pending; | |
4215 | events->smi.smm_inside_nmi = | |
4216 | !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK); | |
4217 | events->smi.latched_init = kvm_lapic_latched_init(vcpu); | |
4218 | ||
dab4b911 | 4219 | events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING |
f077825a PB |
4220 | | KVM_VCPUEVENT_VALID_SHADOW |
4221 | | KVM_VCPUEVENT_VALID_SMM); | |
59073aaf JM |
4222 | if (vcpu->kvm->arch.exception_payload_enabled) |
4223 | events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD; | |
4224 | ||
97e69aa6 | 4225 | memset(&events->reserved, 0, sizeof(events->reserved)); |
3cfc3092 JK |
4226 | } |
4227 | ||
c5833c7a | 4228 | static void kvm_smm_changed(struct kvm_vcpu *vcpu); |
6ef4e07e | 4229 | |
3cfc3092 JK |
4230 | static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu, |
4231 | struct kvm_vcpu_events *events) | |
4232 | { | |
dab4b911 | 4233 | if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING |
48005f64 | 4234 | | KVM_VCPUEVENT_VALID_SIPI_VECTOR |
f077825a | 4235 | | KVM_VCPUEVENT_VALID_SHADOW |
59073aaf JM |
4236 | | KVM_VCPUEVENT_VALID_SMM |
4237 | | KVM_VCPUEVENT_VALID_PAYLOAD)) | |
3cfc3092 JK |
4238 | return -EINVAL; |
4239 | ||
59073aaf JM |
4240 | if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) { |
4241 | if (!vcpu->kvm->arch.exception_payload_enabled) | |
4242 | return -EINVAL; | |
4243 | if (events->exception.pending) | |
4244 | events->exception.injected = 0; | |
4245 | else | |
4246 | events->exception_has_payload = 0; | |
4247 | } else { | |
4248 | events->exception.pending = 0; | |
4249 | events->exception_has_payload = 0; | |
4250 | } | |
4251 | ||
4252 | if ((events->exception.injected || events->exception.pending) && | |
4253 | (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR)) | |
78e546c8 PB |
4254 | return -EINVAL; |
4255 | ||
28bf2888 DH |
4256 | /* INITs are latched while in SMM */ |
4257 | if (events->flags & KVM_VCPUEVENT_VALID_SMM && | |
4258 | (events->smi.smm || events->smi.pending) && | |
4259 | vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) | |
4260 | return -EINVAL; | |
4261 | ||
7460fb4a | 4262 | process_nmi(vcpu); |
59073aaf JM |
4263 | vcpu->arch.exception.injected = events->exception.injected; |
4264 | vcpu->arch.exception.pending = events->exception.pending; | |
3cfc3092 JK |
4265 | vcpu->arch.exception.nr = events->exception.nr; |
4266 | vcpu->arch.exception.has_error_code = events->exception.has_error_code; | |
4267 | vcpu->arch.exception.error_code = events->exception.error_code; | |
59073aaf JM |
4268 | vcpu->arch.exception.has_payload = events->exception_has_payload; |
4269 | vcpu->arch.exception.payload = events->exception_payload; | |
3cfc3092 | 4270 | |
04140b41 | 4271 | vcpu->arch.interrupt.injected = events->interrupt.injected; |
3cfc3092 JK |
4272 | vcpu->arch.interrupt.nr = events->interrupt.nr; |
4273 | vcpu->arch.interrupt.soft = events->interrupt.soft; | |
48005f64 | 4274 | if (events->flags & KVM_VCPUEVENT_VALID_SHADOW) |
afaf0b2f | 4275 | kvm_x86_ops.set_interrupt_shadow(vcpu, |
48005f64 | 4276 | events->interrupt.shadow); |
3cfc3092 JK |
4277 | |
4278 | vcpu->arch.nmi_injected = events->nmi.injected; | |
dab4b911 JK |
4279 | if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING) |
4280 | vcpu->arch.nmi_pending = events->nmi.pending; | |
afaf0b2f | 4281 | kvm_x86_ops.set_nmi_mask(vcpu, events->nmi.masked); |
3cfc3092 | 4282 | |
66450a21 | 4283 | if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR && |
bce87cce | 4284 | lapic_in_kernel(vcpu)) |
66450a21 | 4285 | vcpu->arch.apic->sipi_vector = events->sipi_vector; |
3cfc3092 | 4286 | |
f077825a | 4287 | if (events->flags & KVM_VCPUEVENT_VALID_SMM) { |
c5833c7a SC |
4288 | if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) { |
4289 | if (events->smi.smm) | |
4290 | vcpu->arch.hflags |= HF_SMM_MASK; | |
4291 | else | |
4292 | vcpu->arch.hflags &= ~HF_SMM_MASK; | |
4293 | kvm_smm_changed(vcpu); | |
4294 | } | |
6ef4e07e | 4295 | |
f077825a | 4296 | vcpu->arch.smi_pending = events->smi.pending; |
f4ef1910 WL |
4297 | |
4298 | if (events->smi.smm) { | |
4299 | if (events->smi.smm_inside_nmi) | |
4300 | vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK; | |
f077825a | 4301 | else |
f4ef1910 | 4302 | vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK; |
ff90afa7 LA |
4303 | } |
4304 | ||
4305 | if (lapic_in_kernel(vcpu)) { | |
4306 | if (events->smi.latched_init) | |
4307 | set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events); | |
4308 | else | |
4309 | clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events); | |
f077825a PB |
4310 | } |
4311 | } | |
4312 | ||
3842d135 AK |
4313 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
4314 | ||
3cfc3092 JK |
4315 | return 0; |
4316 | } | |
4317 | ||
a1efbe77 JK |
4318 | static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu, |
4319 | struct kvm_debugregs *dbgregs) | |
4320 | { | |
73aaf249 JK |
4321 | unsigned long val; |
4322 | ||
a1efbe77 | 4323 | memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db)); |
16f8a6f9 | 4324 | kvm_get_dr(vcpu, 6, &val); |
73aaf249 | 4325 | dbgregs->dr6 = val; |
a1efbe77 JK |
4326 | dbgregs->dr7 = vcpu->arch.dr7; |
4327 | dbgregs->flags = 0; | |
97e69aa6 | 4328 | memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved)); |
a1efbe77 JK |
4329 | } |
4330 | ||
4331 | static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu, | |
4332 | struct kvm_debugregs *dbgregs) | |
4333 | { | |
4334 | if (dbgregs->flags) | |
4335 | return -EINVAL; | |
4336 | ||
d14bdb55 PB |
4337 | if (dbgregs->dr6 & ~0xffffffffull) |
4338 | return -EINVAL; | |
4339 | if (dbgregs->dr7 & ~0xffffffffull) | |
4340 | return -EINVAL; | |
4341 | ||
a1efbe77 | 4342 | memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db)); |
ae561ede | 4343 | kvm_update_dr0123(vcpu); |
a1efbe77 JK |
4344 | vcpu->arch.dr6 = dbgregs->dr6; |
4345 | vcpu->arch.dr7 = dbgregs->dr7; | |
9926c9fd | 4346 | kvm_update_dr7(vcpu); |
a1efbe77 | 4347 | |
a1efbe77 JK |
4348 | return 0; |
4349 | } | |
4350 | ||
df1daba7 PB |
4351 | #define XSTATE_COMPACTION_ENABLED (1ULL << 63) |
4352 | ||
4353 | static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu) | |
4354 | { | |
b666a4b6 | 4355 | struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave; |
400e4b20 | 4356 | u64 xstate_bv = xsave->header.xfeatures; |
df1daba7 PB |
4357 | u64 valid; |
4358 | ||
4359 | /* | |
4360 | * Copy legacy XSAVE area, to avoid complications with CPUID | |
4361 | * leaves 0 and 1 in the loop below. | |
4362 | */ | |
4363 | memcpy(dest, xsave, XSAVE_HDR_OFFSET); | |
4364 | ||
4365 | /* Set XSTATE_BV */ | |
00c87e9a | 4366 | xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE; |
df1daba7 PB |
4367 | *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv; |
4368 | ||
4369 | /* | |
4370 | * Copy each region from the possibly compacted offset to the | |
4371 | * non-compacted offset. | |
4372 | */ | |
d91cab78 | 4373 | valid = xstate_bv & ~XFEATURE_MASK_FPSSE; |
df1daba7 | 4374 | while (valid) { |
abd16d68 SAS |
4375 | u64 xfeature_mask = valid & -valid; |
4376 | int xfeature_nr = fls64(xfeature_mask) - 1; | |
4377 | void *src = get_xsave_addr(xsave, xfeature_nr); | |
df1daba7 PB |
4378 | |
4379 | if (src) { | |
4380 | u32 size, offset, ecx, edx; | |
abd16d68 | 4381 | cpuid_count(XSTATE_CPUID, xfeature_nr, |
df1daba7 | 4382 | &size, &offset, &ecx, &edx); |
abd16d68 | 4383 | if (xfeature_nr == XFEATURE_PKRU) |
38cfd5e3 PB |
4384 | memcpy(dest + offset, &vcpu->arch.pkru, |
4385 | sizeof(vcpu->arch.pkru)); | |
4386 | else | |
4387 | memcpy(dest + offset, src, size); | |
4388 | ||
df1daba7 PB |
4389 | } |
4390 | ||
abd16d68 | 4391 | valid -= xfeature_mask; |
df1daba7 PB |
4392 | } |
4393 | } | |
4394 | ||
4395 | static void load_xsave(struct kvm_vcpu *vcpu, u8 *src) | |
4396 | { | |
b666a4b6 | 4397 | struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave; |
df1daba7 PB |
4398 | u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET); |
4399 | u64 valid; | |
4400 | ||
4401 | /* | |
4402 | * Copy legacy XSAVE area, to avoid complications with CPUID | |
4403 | * leaves 0 and 1 in the loop below. | |
4404 | */ | |
4405 | memcpy(xsave, src, XSAVE_HDR_OFFSET); | |
4406 | ||
4407 | /* Set XSTATE_BV and possibly XCOMP_BV. */ | |
400e4b20 | 4408 | xsave->header.xfeatures = xstate_bv; |
782511b0 | 4409 | if (boot_cpu_has(X86_FEATURE_XSAVES)) |
3a54450b | 4410 | xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED; |
df1daba7 PB |
4411 | |
4412 | /* | |
4413 | * Copy each region from the non-compacted offset to the | |
4414 | * possibly compacted offset. | |
4415 | */ | |
d91cab78 | 4416 | valid = xstate_bv & ~XFEATURE_MASK_FPSSE; |
df1daba7 | 4417 | while (valid) { |
abd16d68 SAS |
4418 | u64 xfeature_mask = valid & -valid; |
4419 | int xfeature_nr = fls64(xfeature_mask) - 1; | |
4420 | void *dest = get_xsave_addr(xsave, xfeature_nr); | |
df1daba7 PB |
4421 | |
4422 | if (dest) { | |
4423 | u32 size, offset, ecx, edx; | |
abd16d68 | 4424 | cpuid_count(XSTATE_CPUID, xfeature_nr, |
df1daba7 | 4425 | &size, &offset, &ecx, &edx); |
abd16d68 | 4426 | if (xfeature_nr == XFEATURE_PKRU) |
38cfd5e3 PB |
4427 | memcpy(&vcpu->arch.pkru, src + offset, |
4428 | sizeof(vcpu->arch.pkru)); | |
4429 | else | |
4430 | memcpy(dest, src + offset, size); | |
ee4100da | 4431 | } |
df1daba7 | 4432 | |
abd16d68 | 4433 | valid -= xfeature_mask; |
df1daba7 PB |
4434 | } |
4435 | } | |
4436 | ||
2d5b5a66 SY |
4437 | static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu, |
4438 | struct kvm_xsave *guest_xsave) | |
4439 | { | |
d366bf7e | 4440 | if (boot_cpu_has(X86_FEATURE_XSAVE)) { |
df1daba7 PB |
4441 | memset(guest_xsave, 0, sizeof(struct kvm_xsave)); |
4442 | fill_xsave((u8 *) guest_xsave->region, vcpu); | |
4344ee98 | 4443 | } else { |
2d5b5a66 | 4444 | memcpy(guest_xsave->region, |
b666a4b6 | 4445 | &vcpu->arch.guest_fpu->state.fxsave, |
c47ada30 | 4446 | sizeof(struct fxregs_state)); |
2d5b5a66 | 4447 | *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] = |
d91cab78 | 4448 | XFEATURE_MASK_FPSSE; |
2d5b5a66 SY |
4449 | } |
4450 | } | |
4451 | ||
a575813b WL |
4452 | #define XSAVE_MXCSR_OFFSET 24 |
4453 | ||
2d5b5a66 SY |
4454 | static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu, |
4455 | struct kvm_xsave *guest_xsave) | |
4456 | { | |
4457 | u64 xstate_bv = | |
4458 | *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)]; | |
a575813b | 4459 | u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)]; |
2d5b5a66 | 4460 | |
d366bf7e | 4461 | if (boot_cpu_has(X86_FEATURE_XSAVE)) { |
d7876f1b PB |
4462 | /* |
4463 | * Here we allow setting states that are not present in | |
4464 | * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility | |
4465 | * with old userspace. | |
4466 | */ | |
cfc48181 | 4467 | if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask) |
d7876f1b | 4468 | return -EINVAL; |
df1daba7 | 4469 | load_xsave(vcpu, (u8 *)guest_xsave->region); |
d7876f1b | 4470 | } else { |
a575813b WL |
4471 | if (xstate_bv & ~XFEATURE_MASK_FPSSE || |
4472 | mxcsr & ~mxcsr_feature_mask) | |
2d5b5a66 | 4473 | return -EINVAL; |
b666a4b6 | 4474 | memcpy(&vcpu->arch.guest_fpu->state.fxsave, |
c47ada30 | 4475 | guest_xsave->region, sizeof(struct fxregs_state)); |
2d5b5a66 SY |
4476 | } |
4477 | return 0; | |
4478 | } | |
4479 | ||
4480 | static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu, | |
4481 | struct kvm_xcrs *guest_xcrs) | |
4482 | { | |
d366bf7e | 4483 | if (!boot_cpu_has(X86_FEATURE_XSAVE)) { |
2d5b5a66 SY |
4484 | guest_xcrs->nr_xcrs = 0; |
4485 | return; | |
4486 | } | |
4487 | ||
4488 | guest_xcrs->nr_xcrs = 1; | |
4489 | guest_xcrs->flags = 0; | |
4490 | guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK; | |
4491 | guest_xcrs->xcrs[0].value = vcpu->arch.xcr0; | |
4492 | } | |
4493 | ||
4494 | static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu, | |
4495 | struct kvm_xcrs *guest_xcrs) | |
4496 | { | |
4497 | int i, r = 0; | |
4498 | ||
d366bf7e | 4499 | if (!boot_cpu_has(X86_FEATURE_XSAVE)) |
2d5b5a66 SY |
4500 | return -EINVAL; |
4501 | ||
4502 | if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags) | |
4503 | return -EINVAL; | |
4504 | ||
4505 | for (i = 0; i < guest_xcrs->nr_xcrs; i++) | |
4506 | /* Only support XCR0 currently */ | |
c67a04cb | 4507 | if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) { |
2d5b5a66 | 4508 | r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK, |
c67a04cb | 4509 | guest_xcrs->xcrs[i].value); |
2d5b5a66 SY |
4510 | break; |
4511 | } | |
4512 | if (r) | |
4513 | r = -EINVAL; | |
4514 | return r; | |
4515 | } | |
4516 | ||
1c0b28c2 EM |
4517 | /* |
4518 | * kvm_set_guest_paused() indicates to the guest kernel that it has been | |
4519 | * stopped by the hypervisor. This function will be called from the host only. | |
4520 | * EINVAL is returned when the host attempts to set the flag for a guest that | |
4521 | * does not support pv clocks. | |
4522 | */ | |
4523 | static int kvm_set_guest_paused(struct kvm_vcpu *vcpu) | |
4524 | { | |
0b79459b | 4525 | if (!vcpu->arch.pv_time_enabled) |
1c0b28c2 | 4526 | return -EINVAL; |
51d59c6b | 4527 | vcpu->arch.pvclock_set_guest_stopped_request = true; |
1c0b28c2 EM |
4528 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
4529 | return 0; | |
4530 | } | |
4531 | ||
5c919412 AS |
4532 | static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu, |
4533 | struct kvm_enable_cap *cap) | |
4534 | { | |
57b119da VK |
4535 | int r; |
4536 | uint16_t vmcs_version; | |
4537 | void __user *user_ptr; | |
4538 | ||
5c919412 AS |
4539 | if (cap->flags) |
4540 | return -EINVAL; | |
4541 | ||
4542 | switch (cap->cap) { | |
efc479e6 RK |
4543 | case KVM_CAP_HYPERV_SYNIC2: |
4544 | if (cap->args[0]) | |
4545 | return -EINVAL; | |
df561f66 | 4546 | fallthrough; |
b2869f28 | 4547 | |
5c919412 | 4548 | case KVM_CAP_HYPERV_SYNIC: |
546d87e5 WL |
4549 | if (!irqchip_in_kernel(vcpu->kvm)) |
4550 | return -EINVAL; | |
efc479e6 RK |
4551 | return kvm_hv_activate_synic(vcpu, cap->cap == |
4552 | KVM_CAP_HYPERV_SYNIC2); | |
57b119da | 4553 | case KVM_CAP_HYPERV_ENLIGHTENED_VMCS: |
33b22172 | 4554 | if (!kvm_x86_ops.nested_ops->enable_evmcs) |
5158917c | 4555 | return -ENOTTY; |
33b22172 | 4556 | r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version); |
57b119da VK |
4557 | if (!r) { |
4558 | user_ptr = (void __user *)(uintptr_t)cap->args[0]; | |
4559 | if (copy_to_user(user_ptr, &vmcs_version, | |
4560 | sizeof(vmcs_version))) | |
4561 | r = -EFAULT; | |
4562 | } | |
4563 | return r; | |
344c6c80 | 4564 | case KVM_CAP_HYPERV_DIRECT_TLBFLUSH: |
afaf0b2f | 4565 | if (!kvm_x86_ops.enable_direct_tlbflush) |
344c6c80 TL |
4566 | return -ENOTTY; |
4567 | ||
afaf0b2f | 4568 | return kvm_x86_ops.enable_direct_tlbflush(vcpu); |
57b119da | 4569 | |
66570e96 OU |
4570 | case KVM_CAP_ENFORCE_PV_FEATURE_CPUID: |
4571 | vcpu->arch.pv_cpuid.enforce = cap->args[0]; | |
4572 | ||
4573 | return 0; | |
4574 | ||
5c919412 AS |
4575 | default: |
4576 | return -EINVAL; | |
4577 | } | |
4578 | } | |
4579 | ||
313a3dc7 CO |
4580 | long kvm_arch_vcpu_ioctl(struct file *filp, |
4581 | unsigned int ioctl, unsigned long arg) | |
4582 | { | |
4583 | struct kvm_vcpu *vcpu = filp->private_data; | |
4584 | void __user *argp = (void __user *)arg; | |
4585 | int r; | |
d1ac91d8 AK |
4586 | union { |
4587 | struct kvm_lapic_state *lapic; | |
4588 | struct kvm_xsave *xsave; | |
4589 | struct kvm_xcrs *xcrs; | |
4590 | void *buffer; | |
4591 | } u; | |
4592 | ||
9b062471 CD |
4593 | vcpu_load(vcpu); |
4594 | ||
d1ac91d8 | 4595 | u.buffer = NULL; |
313a3dc7 CO |
4596 | switch (ioctl) { |
4597 | case KVM_GET_LAPIC: { | |
2204ae3c | 4598 | r = -EINVAL; |
bce87cce | 4599 | if (!lapic_in_kernel(vcpu)) |
2204ae3c | 4600 | goto out; |
254272ce BG |
4601 | u.lapic = kzalloc(sizeof(struct kvm_lapic_state), |
4602 | GFP_KERNEL_ACCOUNT); | |
313a3dc7 | 4603 | |
b772ff36 | 4604 | r = -ENOMEM; |
d1ac91d8 | 4605 | if (!u.lapic) |
b772ff36 | 4606 | goto out; |
d1ac91d8 | 4607 | r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic); |
313a3dc7 CO |
4608 | if (r) |
4609 | goto out; | |
4610 | r = -EFAULT; | |
d1ac91d8 | 4611 | if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state))) |
313a3dc7 CO |
4612 | goto out; |
4613 | r = 0; | |
4614 | break; | |
4615 | } | |
4616 | case KVM_SET_LAPIC: { | |
2204ae3c | 4617 | r = -EINVAL; |
bce87cce | 4618 | if (!lapic_in_kernel(vcpu)) |
2204ae3c | 4619 | goto out; |
ff5c2c03 | 4620 | u.lapic = memdup_user(argp, sizeof(*u.lapic)); |
9b062471 CD |
4621 | if (IS_ERR(u.lapic)) { |
4622 | r = PTR_ERR(u.lapic); | |
4623 | goto out_nofree; | |
4624 | } | |
ff5c2c03 | 4625 | |
d1ac91d8 | 4626 | r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic); |
313a3dc7 CO |
4627 | break; |
4628 | } | |
f77bc6a4 ZX |
4629 | case KVM_INTERRUPT: { |
4630 | struct kvm_interrupt irq; | |
4631 | ||
4632 | r = -EFAULT; | |
0e96f31e | 4633 | if (copy_from_user(&irq, argp, sizeof(irq))) |
f77bc6a4 ZX |
4634 | goto out; |
4635 | r = kvm_vcpu_ioctl_interrupt(vcpu, &irq); | |
f77bc6a4 ZX |
4636 | break; |
4637 | } | |
c4abb7c9 JK |
4638 | case KVM_NMI: { |
4639 | r = kvm_vcpu_ioctl_nmi(vcpu); | |
c4abb7c9 JK |
4640 | break; |
4641 | } | |
f077825a PB |
4642 | case KVM_SMI: { |
4643 | r = kvm_vcpu_ioctl_smi(vcpu); | |
4644 | break; | |
4645 | } | |
313a3dc7 CO |
4646 | case KVM_SET_CPUID: { |
4647 | struct kvm_cpuid __user *cpuid_arg = argp; | |
4648 | struct kvm_cpuid cpuid; | |
4649 | ||
4650 | r = -EFAULT; | |
0e96f31e | 4651 | if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid))) |
313a3dc7 CO |
4652 | goto out; |
4653 | r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries); | |
313a3dc7 CO |
4654 | break; |
4655 | } | |
07716717 DK |
4656 | case KVM_SET_CPUID2: { |
4657 | struct kvm_cpuid2 __user *cpuid_arg = argp; | |
4658 | struct kvm_cpuid2 cpuid; | |
4659 | ||
4660 | r = -EFAULT; | |
0e96f31e | 4661 | if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid))) |
07716717 DK |
4662 | goto out; |
4663 | r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid, | |
19355475 | 4664 | cpuid_arg->entries); |
07716717 DK |
4665 | break; |
4666 | } | |
4667 | case KVM_GET_CPUID2: { | |
4668 | struct kvm_cpuid2 __user *cpuid_arg = argp; | |
4669 | struct kvm_cpuid2 cpuid; | |
4670 | ||
4671 | r = -EFAULT; | |
0e96f31e | 4672 | if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid))) |
07716717 DK |
4673 | goto out; |
4674 | r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid, | |
19355475 | 4675 | cpuid_arg->entries); |
07716717 DK |
4676 | if (r) |
4677 | goto out; | |
4678 | r = -EFAULT; | |
0e96f31e | 4679 | if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid))) |
07716717 DK |
4680 | goto out; |
4681 | r = 0; | |
4682 | break; | |
4683 | } | |
801e459a TL |
4684 | case KVM_GET_MSRS: { |
4685 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
609e36d3 | 4686 | r = msr_io(vcpu, argp, do_get_msr, 1); |
801e459a | 4687 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
313a3dc7 | 4688 | break; |
801e459a TL |
4689 | } |
4690 | case KVM_SET_MSRS: { | |
4691 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
313a3dc7 | 4692 | r = msr_io(vcpu, argp, do_set_msr, 0); |
801e459a | 4693 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
313a3dc7 | 4694 | break; |
801e459a | 4695 | } |
b209749f AK |
4696 | case KVM_TPR_ACCESS_REPORTING: { |
4697 | struct kvm_tpr_access_ctl tac; | |
4698 | ||
4699 | r = -EFAULT; | |
0e96f31e | 4700 | if (copy_from_user(&tac, argp, sizeof(tac))) |
b209749f AK |
4701 | goto out; |
4702 | r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac); | |
4703 | if (r) | |
4704 | goto out; | |
4705 | r = -EFAULT; | |
0e96f31e | 4706 | if (copy_to_user(argp, &tac, sizeof(tac))) |
b209749f AK |
4707 | goto out; |
4708 | r = 0; | |
4709 | break; | |
4710 | }; | |
b93463aa AK |
4711 | case KVM_SET_VAPIC_ADDR: { |
4712 | struct kvm_vapic_addr va; | |
7301d6ab | 4713 | int idx; |
b93463aa AK |
4714 | |
4715 | r = -EINVAL; | |
35754c98 | 4716 | if (!lapic_in_kernel(vcpu)) |
b93463aa AK |
4717 | goto out; |
4718 | r = -EFAULT; | |
0e96f31e | 4719 | if (copy_from_user(&va, argp, sizeof(va))) |
b93463aa | 4720 | goto out; |
7301d6ab | 4721 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
fda4e2e8 | 4722 | r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr); |
7301d6ab | 4723 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
b93463aa AK |
4724 | break; |
4725 | } | |
890ca9ae HY |
4726 | case KVM_X86_SETUP_MCE: { |
4727 | u64 mcg_cap; | |
4728 | ||
4729 | r = -EFAULT; | |
0e96f31e | 4730 | if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap))) |
890ca9ae HY |
4731 | goto out; |
4732 | r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap); | |
4733 | break; | |
4734 | } | |
4735 | case KVM_X86_SET_MCE: { | |
4736 | struct kvm_x86_mce mce; | |
4737 | ||
4738 | r = -EFAULT; | |
0e96f31e | 4739 | if (copy_from_user(&mce, argp, sizeof(mce))) |
890ca9ae HY |
4740 | goto out; |
4741 | r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce); | |
4742 | break; | |
4743 | } | |
3cfc3092 JK |
4744 | case KVM_GET_VCPU_EVENTS: { |
4745 | struct kvm_vcpu_events events; | |
4746 | ||
4747 | kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events); | |
4748 | ||
4749 | r = -EFAULT; | |
4750 | if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events))) | |
4751 | break; | |
4752 | r = 0; | |
4753 | break; | |
4754 | } | |
4755 | case KVM_SET_VCPU_EVENTS: { | |
4756 | struct kvm_vcpu_events events; | |
4757 | ||
4758 | r = -EFAULT; | |
4759 | if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events))) | |
4760 | break; | |
4761 | ||
4762 | r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events); | |
4763 | break; | |
4764 | } | |
a1efbe77 JK |
4765 | case KVM_GET_DEBUGREGS: { |
4766 | struct kvm_debugregs dbgregs; | |
4767 | ||
4768 | kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs); | |
4769 | ||
4770 | r = -EFAULT; | |
4771 | if (copy_to_user(argp, &dbgregs, | |
4772 | sizeof(struct kvm_debugregs))) | |
4773 | break; | |
4774 | r = 0; | |
4775 | break; | |
4776 | } | |
4777 | case KVM_SET_DEBUGREGS: { | |
4778 | struct kvm_debugregs dbgregs; | |
4779 | ||
4780 | r = -EFAULT; | |
4781 | if (copy_from_user(&dbgregs, argp, | |
4782 | sizeof(struct kvm_debugregs))) | |
4783 | break; | |
4784 | ||
4785 | r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs); | |
4786 | break; | |
4787 | } | |
2d5b5a66 | 4788 | case KVM_GET_XSAVE: { |
254272ce | 4789 | u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT); |
2d5b5a66 | 4790 | r = -ENOMEM; |
d1ac91d8 | 4791 | if (!u.xsave) |
2d5b5a66 SY |
4792 | break; |
4793 | ||
d1ac91d8 | 4794 | kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave); |
2d5b5a66 SY |
4795 | |
4796 | r = -EFAULT; | |
d1ac91d8 | 4797 | if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave))) |
2d5b5a66 SY |
4798 | break; |
4799 | r = 0; | |
4800 | break; | |
4801 | } | |
4802 | case KVM_SET_XSAVE: { | |
ff5c2c03 | 4803 | u.xsave = memdup_user(argp, sizeof(*u.xsave)); |
9b062471 CD |
4804 | if (IS_ERR(u.xsave)) { |
4805 | r = PTR_ERR(u.xsave); | |
4806 | goto out_nofree; | |
4807 | } | |
2d5b5a66 | 4808 | |
d1ac91d8 | 4809 | r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave); |
2d5b5a66 SY |
4810 | break; |
4811 | } | |
4812 | case KVM_GET_XCRS: { | |
254272ce | 4813 | u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT); |
2d5b5a66 | 4814 | r = -ENOMEM; |
d1ac91d8 | 4815 | if (!u.xcrs) |
2d5b5a66 SY |
4816 | break; |
4817 | ||
d1ac91d8 | 4818 | kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs); |
2d5b5a66 SY |
4819 | |
4820 | r = -EFAULT; | |
d1ac91d8 | 4821 | if (copy_to_user(argp, u.xcrs, |
2d5b5a66 SY |
4822 | sizeof(struct kvm_xcrs))) |
4823 | break; | |
4824 | r = 0; | |
4825 | break; | |
4826 | } | |
4827 | case KVM_SET_XCRS: { | |
ff5c2c03 | 4828 | u.xcrs = memdup_user(argp, sizeof(*u.xcrs)); |
9b062471 CD |
4829 | if (IS_ERR(u.xcrs)) { |
4830 | r = PTR_ERR(u.xcrs); | |
4831 | goto out_nofree; | |
4832 | } | |
2d5b5a66 | 4833 | |
d1ac91d8 | 4834 | r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs); |
2d5b5a66 SY |
4835 | break; |
4836 | } | |
92a1f12d JR |
4837 | case KVM_SET_TSC_KHZ: { |
4838 | u32 user_tsc_khz; | |
4839 | ||
4840 | r = -EINVAL; | |
92a1f12d JR |
4841 | user_tsc_khz = (u32)arg; |
4842 | ||
26769f96 MT |
4843 | if (kvm_has_tsc_control && |
4844 | user_tsc_khz >= kvm_max_guest_tsc_khz) | |
92a1f12d JR |
4845 | goto out; |
4846 | ||
cc578287 ZA |
4847 | if (user_tsc_khz == 0) |
4848 | user_tsc_khz = tsc_khz; | |
4849 | ||
381d585c HZ |
4850 | if (!kvm_set_tsc_khz(vcpu, user_tsc_khz)) |
4851 | r = 0; | |
92a1f12d | 4852 | |
92a1f12d JR |
4853 | goto out; |
4854 | } | |
4855 | case KVM_GET_TSC_KHZ: { | |
cc578287 | 4856 | r = vcpu->arch.virtual_tsc_khz; |
92a1f12d JR |
4857 | goto out; |
4858 | } | |
1c0b28c2 EM |
4859 | case KVM_KVMCLOCK_CTRL: { |
4860 | r = kvm_set_guest_paused(vcpu); | |
4861 | goto out; | |
4862 | } | |
5c919412 AS |
4863 | case KVM_ENABLE_CAP: { |
4864 | struct kvm_enable_cap cap; | |
4865 | ||
4866 | r = -EFAULT; | |
4867 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
4868 | goto out; | |
4869 | r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap); | |
4870 | break; | |
4871 | } | |
8fcc4b59 JM |
4872 | case KVM_GET_NESTED_STATE: { |
4873 | struct kvm_nested_state __user *user_kvm_nested_state = argp; | |
4874 | u32 user_data_size; | |
4875 | ||
4876 | r = -EINVAL; | |
33b22172 | 4877 | if (!kvm_x86_ops.nested_ops->get_state) |
8fcc4b59 JM |
4878 | break; |
4879 | ||
4880 | BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size)); | |
26b471c7 | 4881 | r = -EFAULT; |
8fcc4b59 | 4882 | if (get_user(user_data_size, &user_kvm_nested_state->size)) |
26b471c7 | 4883 | break; |
8fcc4b59 | 4884 | |
33b22172 PB |
4885 | r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state, |
4886 | user_data_size); | |
8fcc4b59 | 4887 | if (r < 0) |
26b471c7 | 4888 | break; |
8fcc4b59 JM |
4889 | |
4890 | if (r > user_data_size) { | |
4891 | if (put_user(r, &user_kvm_nested_state->size)) | |
26b471c7 LA |
4892 | r = -EFAULT; |
4893 | else | |
4894 | r = -E2BIG; | |
4895 | break; | |
8fcc4b59 | 4896 | } |
26b471c7 | 4897 | |
8fcc4b59 JM |
4898 | r = 0; |
4899 | break; | |
4900 | } | |
4901 | case KVM_SET_NESTED_STATE: { | |
4902 | struct kvm_nested_state __user *user_kvm_nested_state = argp; | |
4903 | struct kvm_nested_state kvm_state; | |
ad5996d9 | 4904 | int idx; |
8fcc4b59 JM |
4905 | |
4906 | r = -EINVAL; | |
33b22172 | 4907 | if (!kvm_x86_ops.nested_ops->set_state) |
8fcc4b59 JM |
4908 | break; |
4909 | ||
26b471c7 | 4910 | r = -EFAULT; |
8fcc4b59 | 4911 | if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state))) |
26b471c7 | 4912 | break; |
8fcc4b59 | 4913 | |
26b471c7 | 4914 | r = -EINVAL; |
8fcc4b59 | 4915 | if (kvm_state.size < sizeof(kvm_state)) |
26b471c7 | 4916 | break; |
8fcc4b59 JM |
4917 | |
4918 | if (kvm_state.flags & | |
8cab6507 | 4919 | ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE |
cc440cda PB |
4920 | | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING |
4921 | | KVM_STATE_NESTED_GIF_SET)) | |
26b471c7 | 4922 | break; |
8fcc4b59 JM |
4923 | |
4924 | /* nested_run_pending implies guest_mode. */ | |
8cab6507 VK |
4925 | if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING) |
4926 | && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE)) | |
26b471c7 | 4927 | break; |
8fcc4b59 | 4928 | |
ad5996d9 | 4929 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
33b22172 | 4930 | r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state); |
ad5996d9 | 4931 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
8fcc4b59 JM |
4932 | break; |
4933 | } | |
2bc39970 VK |
4934 | case KVM_GET_SUPPORTED_HV_CPUID: { |
4935 | struct kvm_cpuid2 __user *cpuid_arg = argp; | |
4936 | struct kvm_cpuid2 cpuid; | |
4937 | ||
4938 | r = -EFAULT; | |
4939 | if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid))) | |
4940 | goto out; | |
4941 | ||
4942 | r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid, | |
4943 | cpuid_arg->entries); | |
4944 | if (r) | |
4945 | goto out; | |
4946 | ||
4947 | r = -EFAULT; | |
4948 | if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid))) | |
4949 | goto out; | |
4950 | r = 0; | |
4951 | break; | |
4952 | } | |
313a3dc7 CO |
4953 | default: |
4954 | r = -EINVAL; | |
4955 | } | |
4956 | out: | |
d1ac91d8 | 4957 | kfree(u.buffer); |
9b062471 CD |
4958 | out_nofree: |
4959 | vcpu_put(vcpu); | |
313a3dc7 CO |
4960 | return r; |
4961 | } | |
4962 | ||
1499fa80 | 4963 | vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) |
5b1c1493 CO |
4964 | { |
4965 | return VM_FAULT_SIGBUS; | |
4966 | } | |
4967 | ||
1fe779f8 CO |
4968 | static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr) |
4969 | { | |
4970 | int ret; | |
4971 | ||
4972 | if (addr > (unsigned int)(-3 * PAGE_SIZE)) | |
951179ce | 4973 | return -EINVAL; |
afaf0b2f | 4974 | ret = kvm_x86_ops.set_tss_addr(kvm, addr); |
1fe779f8 CO |
4975 | return ret; |
4976 | } | |
4977 | ||
b927a3ce SY |
4978 | static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm, |
4979 | u64 ident_addr) | |
4980 | { | |
afaf0b2f | 4981 | return kvm_x86_ops.set_identity_map_addr(kvm, ident_addr); |
b927a3ce SY |
4982 | } |
4983 | ||
1fe779f8 | 4984 | static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm, |
bc8a3d89 | 4985 | unsigned long kvm_nr_mmu_pages) |
1fe779f8 CO |
4986 | { |
4987 | if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES) | |
4988 | return -EINVAL; | |
4989 | ||
79fac95e | 4990 | mutex_lock(&kvm->slots_lock); |
1fe779f8 CO |
4991 | |
4992 | kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages); | |
f05e70ac | 4993 | kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages; |
1fe779f8 | 4994 | |
79fac95e | 4995 | mutex_unlock(&kvm->slots_lock); |
1fe779f8 CO |
4996 | return 0; |
4997 | } | |
4998 | ||
bc8a3d89 | 4999 | static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm) |
1fe779f8 | 5000 | { |
39de71ec | 5001 | return kvm->arch.n_max_mmu_pages; |
1fe779f8 CO |
5002 | } |
5003 | ||
1fe779f8 CO |
5004 | static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip) |
5005 | { | |
90bca052 | 5006 | struct kvm_pic *pic = kvm->arch.vpic; |
1fe779f8 CO |
5007 | int r; |
5008 | ||
5009 | r = 0; | |
5010 | switch (chip->chip_id) { | |
5011 | case KVM_IRQCHIP_PIC_MASTER: | |
90bca052 | 5012 | memcpy(&chip->chip.pic, &pic->pics[0], |
1fe779f8 CO |
5013 | sizeof(struct kvm_pic_state)); |
5014 | break; | |
5015 | case KVM_IRQCHIP_PIC_SLAVE: | |
90bca052 | 5016 | memcpy(&chip->chip.pic, &pic->pics[1], |
1fe779f8 CO |
5017 | sizeof(struct kvm_pic_state)); |
5018 | break; | |
5019 | case KVM_IRQCHIP_IOAPIC: | |
33392b49 | 5020 | kvm_get_ioapic(kvm, &chip->chip.ioapic); |
1fe779f8 CO |
5021 | break; |
5022 | default: | |
5023 | r = -EINVAL; | |
5024 | break; | |
5025 | } | |
5026 | return r; | |
5027 | } | |
5028 | ||
5029 | static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip) | |
5030 | { | |
90bca052 | 5031 | struct kvm_pic *pic = kvm->arch.vpic; |
1fe779f8 CO |
5032 | int r; |
5033 | ||
5034 | r = 0; | |
5035 | switch (chip->chip_id) { | |
5036 | case KVM_IRQCHIP_PIC_MASTER: | |
90bca052 DH |
5037 | spin_lock(&pic->lock); |
5038 | memcpy(&pic->pics[0], &chip->chip.pic, | |
1fe779f8 | 5039 | sizeof(struct kvm_pic_state)); |
90bca052 | 5040 | spin_unlock(&pic->lock); |
1fe779f8 CO |
5041 | break; |
5042 | case KVM_IRQCHIP_PIC_SLAVE: | |
90bca052 DH |
5043 | spin_lock(&pic->lock); |
5044 | memcpy(&pic->pics[1], &chip->chip.pic, | |
1fe779f8 | 5045 | sizeof(struct kvm_pic_state)); |
90bca052 | 5046 | spin_unlock(&pic->lock); |
1fe779f8 CO |
5047 | break; |
5048 | case KVM_IRQCHIP_IOAPIC: | |
33392b49 | 5049 | kvm_set_ioapic(kvm, &chip->chip.ioapic); |
1fe779f8 CO |
5050 | break; |
5051 | default: | |
5052 | r = -EINVAL; | |
5053 | break; | |
5054 | } | |
90bca052 | 5055 | kvm_pic_update_irq(pic); |
1fe779f8 CO |
5056 | return r; |
5057 | } | |
5058 | ||
e0f63cb9 SY |
5059 | static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps) |
5060 | { | |
34f3941c RK |
5061 | struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state; |
5062 | ||
5063 | BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels)); | |
5064 | ||
5065 | mutex_lock(&kps->lock); | |
5066 | memcpy(ps, &kps->channels, sizeof(*ps)); | |
5067 | mutex_unlock(&kps->lock); | |
2da29bcc | 5068 | return 0; |
e0f63cb9 SY |
5069 | } |
5070 | ||
5071 | static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps) | |
5072 | { | |
0185604c | 5073 | int i; |
09edea72 RK |
5074 | struct kvm_pit *pit = kvm->arch.vpit; |
5075 | ||
5076 | mutex_lock(&pit->pit_state.lock); | |
34f3941c | 5077 | memcpy(&pit->pit_state.channels, ps, sizeof(*ps)); |
0185604c | 5078 | for (i = 0; i < 3; i++) |
09edea72 RK |
5079 | kvm_pit_load_count(pit, i, ps->channels[i].count, 0); |
5080 | mutex_unlock(&pit->pit_state.lock); | |
2da29bcc | 5081 | return 0; |
e9f42757 BK |
5082 | } |
5083 | ||
5084 | static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps) | |
5085 | { | |
e9f42757 BK |
5086 | mutex_lock(&kvm->arch.vpit->pit_state.lock); |
5087 | memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels, | |
5088 | sizeof(ps->channels)); | |
5089 | ps->flags = kvm->arch.vpit->pit_state.flags; | |
5090 | mutex_unlock(&kvm->arch.vpit->pit_state.lock); | |
97e69aa6 | 5091 | memset(&ps->reserved, 0, sizeof(ps->reserved)); |
2da29bcc | 5092 | return 0; |
e9f42757 BK |
5093 | } |
5094 | ||
5095 | static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps) | |
5096 | { | |
2da29bcc | 5097 | int start = 0; |
0185604c | 5098 | int i; |
e9f42757 | 5099 | u32 prev_legacy, cur_legacy; |
09edea72 RK |
5100 | struct kvm_pit *pit = kvm->arch.vpit; |
5101 | ||
5102 | mutex_lock(&pit->pit_state.lock); | |
5103 | prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY; | |
e9f42757 BK |
5104 | cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY; |
5105 | if (!prev_legacy && cur_legacy) | |
5106 | start = 1; | |
09edea72 RK |
5107 | memcpy(&pit->pit_state.channels, &ps->channels, |
5108 | sizeof(pit->pit_state.channels)); | |
5109 | pit->pit_state.flags = ps->flags; | |
0185604c | 5110 | for (i = 0; i < 3; i++) |
09edea72 | 5111 | kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count, |
e5e57e7a | 5112 | start && i == 0); |
09edea72 | 5113 | mutex_unlock(&pit->pit_state.lock); |
2da29bcc | 5114 | return 0; |
e0f63cb9 SY |
5115 | } |
5116 | ||
52d939a0 MT |
5117 | static int kvm_vm_ioctl_reinject(struct kvm *kvm, |
5118 | struct kvm_reinject_control *control) | |
5119 | { | |
71474e2f RK |
5120 | struct kvm_pit *pit = kvm->arch.vpit; |
5121 | ||
71474e2f RK |
5122 | /* pit->pit_state.lock was overloaded to prevent userspace from getting |
5123 | * an inconsistent state after running multiple KVM_REINJECT_CONTROL | |
5124 | * ioctls in parallel. Use a separate lock if that ioctl isn't rare. | |
5125 | */ | |
5126 | mutex_lock(&pit->pit_state.lock); | |
5127 | kvm_pit_set_reinject(pit, control->pit_reinject); | |
5128 | mutex_unlock(&pit->pit_state.lock); | |
b39c90b6 | 5129 | |
52d939a0 MT |
5130 | return 0; |
5131 | } | |
5132 | ||
0dff0846 | 5133 | void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot) |
5bb064dc | 5134 | { |
88178fd4 KH |
5135 | /* |
5136 | * Flush potentially hardware-cached dirty pages to dirty_bitmap. | |
5137 | */ | |
afaf0b2f SC |
5138 | if (kvm_x86_ops.flush_log_dirty) |
5139 | kvm_x86_ops.flush_log_dirty(kvm); | |
5bb064dc ZX |
5140 | } |
5141 | ||
aa2fbe6d YZ |
5142 | int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event, |
5143 | bool line_status) | |
23d43cf9 CD |
5144 | { |
5145 | if (!irqchip_in_kernel(kvm)) | |
5146 | return -ENXIO; | |
5147 | ||
5148 | irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, | |
aa2fbe6d YZ |
5149 | irq_event->irq, irq_event->level, |
5150 | line_status); | |
23d43cf9 CD |
5151 | return 0; |
5152 | } | |
5153 | ||
e5d83c74 PB |
5154 | int kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
5155 | struct kvm_enable_cap *cap) | |
90de4a18 NA |
5156 | { |
5157 | int r; | |
5158 | ||
5159 | if (cap->flags) | |
5160 | return -EINVAL; | |
5161 | ||
5162 | switch (cap->cap) { | |
5163 | case KVM_CAP_DISABLE_QUIRKS: | |
5164 | kvm->arch.disabled_quirks = cap->args[0]; | |
5165 | r = 0; | |
5166 | break; | |
49df6397 SR |
5167 | case KVM_CAP_SPLIT_IRQCHIP: { |
5168 | mutex_lock(&kvm->lock); | |
b053b2ae SR |
5169 | r = -EINVAL; |
5170 | if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS) | |
5171 | goto split_irqchip_unlock; | |
49df6397 SR |
5172 | r = -EEXIST; |
5173 | if (irqchip_in_kernel(kvm)) | |
5174 | goto split_irqchip_unlock; | |
557abc40 | 5175 | if (kvm->created_vcpus) |
49df6397 SR |
5176 | goto split_irqchip_unlock; |
5177 | r = kvm_setup_empty_irq_routing(kvm); | |
5c0aea0e | 5178 | if (r) |
49df6397 SR |
5179 | goto split_irqchip_unlock; |
5180 | /* Pairs with irqchip_in_kernel. */ | |
5181 | smp_wmb(); | |
49776faf | 5182 | kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT; |
b053b2ae | 5183 | kvm->arch.nr_reserved_ioapic_pins = cap->args[0]; |
49df6397 SR |
5184 | r = 0; |
5185 | split_irqchip_unlock: | |
5186 | mutex_unlock(&kvm->lock); | |
5187 | break; | |
5188 | } | |
37131313 RK |
5189 | case KVM_CAP_X2APIC_API: |
5190 | r = -EINVAL; | |
5191 | if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS) | |
5192 | break; | |
5193 | ||
5194 | if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS) | |
5195 | kvm->arch.x2apic_format = true; | |
c519265f RK |
5196 | if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK) |
5197 | kvm->arch.x2apic_broadcast_quirk_disabled = true; | |
37131313 RK |
5198 | |
5199 | r = 0; | |
5200 | break; | |
4d5422ce WL |
5201 | case KVM_CAP_X86_DISABLE_EXITS: |
5202 | r = -EINVAL; | |
5203 | if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS) | |
5204 | break; | |
5205 | ||
5206 | if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) && | |
5207 | kvm_can_mwait_in_guest()) | |
5208 | kvm->arch.mwait_in_guest = true; | |
766d3571 | 5209 | if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT) |
caa057a2 | 5210 | kvm->arch.hlt_in_guest = true; |
b31c114b WL |
5211 | if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE) |
5212 | kvm->arch.pause_in_guest = true; | |
b5170063 WL |
5213 | if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE) |
5214 | kvm->arch.cstate_in_guest = true; | |
4d5422ce WL |
5215 | r = 0; |
5216 | break; | |
6fbbde9a DS |
5217 | case KVM_CAP_MSR_PLATFORM_INFO: |
5218 | kvm->arch.guest_can_read_msr_platform_info = cap->args[0]; | |
5219 | r = 0; | |
c4f55198 JM |
5220 | break; |
5221 | case KVM_CAP_EXCEPTION_PAYLOAD: | |
5222 | kvm->arch.exception_payload_enabled = cap->args[0]; | |
5223 | r = 0; | |
6fbbde9a | 5224 | break; |
1ae09954 AG |
5225 | case KVM_CAP_X86_USER_SPACE_MSR: |
5226 | kvm->arch.user_space_msr_mask = cap->args[0]; | |
5227 | r = 0; | |
5228 | break; | |
90de4a18 NA |
5229 | default: |
5230 | r = -EINVAL; | |
5231 | break; | |
5232 | } | |
5233 | return r; | |
5234 | } | |
5235 | ||
1a155254 AG |
5236 | static void kvm_clear_msr_filter(struct kvm *kvm) |
5237 | { | |
5238 | u32 i; | |
5239 | u32 count = kvm->arch.msr_filter.count; | |
5240 | struct msr_bitmap_range ranges[16]; | |
5241 | ||
5242 | mutex_lock(&kvm->lock); | |
5243 | kvm->arch.msr_filter.count = 0; | |
5244 | memcpy(ranges, kvm->arch.msr_filter.ranges, count * sizeof(ranges[0])); | |
5245 | mutex_unlock(&kvm->lock); | |
5246 | synchronize_srcu(&kvm->srcu); | |
5247 | ||
5248 | for (i = 0; i < count; i++) | |
5249 | kfree(ranges[i].bitmap); | |
5250 | } | |
5251 | ||
5252 | static int kvm_add_msr_filter(struct kvm *kvm, struct kvm_msr_filter_range *user_range) | |
5253 | { | |
5254 | struct msr_bitmap_range *ranges = kvm->arch.msr_filter.ranges; | |
5255 | struct msr_bitmap_range range; | |
5256 | unsigned long *bitmap = NULL; | |
5257 | size_t bitmap_size; | |
5258 | int r; | |
5259 | ||
5260 | if (!user_range->nmsrs) | |
5261 | return 0; | |
5262 | ||
5263 | bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long); | |
5264 | if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE) | |
5265 | return -EINVAL; | |
5266 | ||
5267 | bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size); | |
5268 | if (IS_ERR(bitmap)) | |
5269 | return PTR_ERR(bitmap); | |
5270 | ||
5271 | range = (struct msr_bitmap_range) { | |
5272 | .flags = user_range->flags, | |
5273 | .base = user_range->base, | |
5274 | .nmsrs = user_range->nmsrs, | |
5275 | .bitmap = bitmap, | |
5276 | }; | |
5277 | ||
5278 | if (range.flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE)) { | |
5279 | r = -EINVAL; | |
5280 | goto err; | |
5281 | } | |
5282 | ||
5283 | if (!range.flags) { | |
5284 | r = -EINVAL; | |
5285 | goto err; | |
5286 | } | |
5287 | ||
5288 | /* Everything ok, add this range identifier to our global pool */ | |
5289 | ranges[kvm->arch.msr_filter.count] = range; | |
5290 | /* Make sure we filled the array before we tell anyone to walk it */ | |
5291 | smp_wmb(); | |
5292 | kvm->arch.msr_filter.count++; | |
5293 | ||
5294 | return 0; | |
5295 | err: | |
5296 | kfree(bitmap); | |
5297 | return r; | |
5298 | } | |
5299 | ||
5300 | static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp) | |
5301 | { | |
5302 | struct kvm_msr_filter __user *user_msr_filter = argp; | |
5303 | struct kvm_msr_filter filter; | |
5304 | bool default_allow; | |
5305 | int r = 0; | |
043248b3 | 5306 | bool empty = true; |
1a155254 AG |
5307 | u32 i; |
5308 | ||
5309 | if (copy_from_user(&filter, user_msr_filter, sizeof(filter))) | |
5310 | return -EFAULT; | |
5311 | ||
043248b3 PB |
5312 | for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) |
5313 | empty &= !filter.ranges[i].nmsrs; | |
1a155254 AG |
5314 | |
5315 | default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY); | |
043248b3 PB |
5316 | if (empty && !default_allow) |
5317 | return -EINVAL; | |
5318 | ||
5319 | kvm_clear_msr_filter(kvm); | |
5320 | ||
1a155254 AG |
5321 | kvm->arch.msr_filter.default_allow = default_allow; |
5322 | ||
5323 | /* | |
5324 | * Protect from concurrent calls to this function that could trigger | |
5325 | * a TOCTOU violation on kvm->arch.msr_filter.count. | |
5326 | */ | |
5327 | mutex_lock(&kvm->lock); | |
5328 | for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) { | |
5329 | r = kvm_add_msr_filter(kvm, &filter.ranges[i]); | |
5330 | if (r) | |
5331 | break; | |
5332 | } | |
5333 | ||
5334 | kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED); | |
5335 | mutex_unlock(&kvm->lock); | |
5336 | ||
5337 | return r; | |
5338 | } | |
5339 | ||
1fe779f8 CO |
5340 | long kvm_arch_vm_ioctl(struct file *filp, |
5341 | unsigned int ioctl, unsigned long arg) | |
5342 | { | |
5343 | struct kvm *kvm = filp->private_data; | |
5344 | void __user *argp = (void __user *)arg; | |
367e1319 | 5345 | int r = -ENOTTY; |
f0d66275 DH |
5346 | /* |
5347 | * This union makes it completely explicit to gcc-3.x | |
5348 | * that these two variables' stack usage should be | |
5349 | * combined, not added together. | |
5350 | */ | |
5351 | union { | |
5352 | struct kvm_pit_state ps; | |
e9f42757 | 5353 | struct kvm_pit_state2 ps2; |
c5ff41ce | 5354 | struct kvm_pit_config pit_config; |
f0d66275 | 5355 | } u; |
1fe779f8 CO |
5356 | |
5357 | switch (ioctl) { | |
5358 | case KVM_SET_TSS_ADDR: | |
5359 | r = kvm_vm_ioctl_set_tss_addr(kvm, arg); | |
1fe779f8 | 5360 | break; |
b927a3ce SY |
5361 | case KVM_SET_IDENTITY_MAP_ADDR: { |
5362 | u64 ident_addr; | |
5363 | ||
1af1ac91 DH |
5364 | mutex_lock(&kvm->lock); |
5365 | r = -EINVAL; | |
5366 | if (kvm->created_vcpus) | |
5367 | goto set_identity_unlock; | |
b927a3ce | 5368 | r = -EFAULT; |
0e96f31e | 5369 | if (copy_from_user(&ident_addr, argp, sizeof(ident_addr))) |
1af1ac91 | 5370 | goto set_identity_unlock; |
b927a3ce | 5371 | r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr); |
1af1ac91 DH |
5372 | set_identity_unlock: |
5373 | mutex_unlock(&kvm->lock); | |
b927a3ce SY |
5374 | break; |
5375 | } | |
1fe779f8 CO |
5376 | case KVM_SET_NR_MMU_PAGES: |
5377 | r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg); | |
1fe779f8 CO |
5378 | break; |
5379 | case KVM_GET_NR_MMU_PAGES: | |
5380 | r = kvm_vm_ioctl_get_nr_mmu_pages(kvm); | |
5381 | break; | |
3ddea128 | 5382 | case KVM_CREATE_IRQCHIP: { |
3ddea128 | 5383 | mutex_lock(&kvm->lock); |
09941366 | 5384 | |
3ddea128 | 5385 | r = -EEXIST; |
35e6eaa3 | 5386 | if (irqchip_in_kernel(kvm)) |
3ddea128 | 5387 | goto create_irqchip_unlock; |
09941366 | 5388 | |
3e515705 | 5389 | r = -EINVAL; |
557abc40 | 5390 | if (kvm->created_vcpus) |
3e515705 | 5391 | goto create_irqchip_unlock; |
09941366 RK |
5392 | |
5393 | r = kvm_pic_init(kvm); | |
5394 | if (r) | |
3ddea128 | 5395 | goto create_irqchip_unlock; |
09941366 RK |
5396 | |
5397 | r = kvm_ioapic_init(kvm); | |
5398 | if (r) { | |
09941366 | 5399 | kvm_pic_destroy(kvm); |
3ddea128 | 5400 | goto create_irqchip_unlock; |
09941366 RK |
5401 | } |
5402 | ||
399ec807 AK |
5403 | r = kvm_setup_default_irq_routing(kvm); |
5404 | if (r) { | |
72bb2fcd | 5405 | kvm_ioapic_destroy(kvm); |
09941366 | 5406 | kvm_pic_destroy(kvm); |
71ba994c | 5407 | goto create_irqchip_unlock; |
399ec807 | 5408 | } |
49776faf | 5409 | /* Write kvm->irq_routing before enabling irqchip_in_kernel. */ |
71ba994c | 5410 | smp_wmb(); |
49776faf | 5411 | kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL; |
3ddea128 MT |
5412 | create_irqchip_unlock: |
5413 | mutex_unlock(&kvm->lock); | |
1fe779f8 | 5414 | break; |
3ddea128 | 5415 | } |
7837699f | 5416 | case KVM_CREATE_PIT: |
c5ff41ce JK |
5417 | u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY; |
5418 | goto create_pit; | |
5419 | case KVM_CREATE_PIT2: | |
5420 | r = -EFAULT; | |
5421 | if (copy_from_user(&u.pit_config, argp, | |
5422 | sizeof(struct kvm_pit_config))) | |
5423 | goto out; | |
5424 | create_pit: | |
250715a6 | 5425 | mutex_lock(&kvm->lock); |
269e05e4 AK |
5426 | r = -EEXIST; |
5427 | if (kvm->arch.vpit) | |
5428 | goto create_pit_unlock; | |
7837699f | 5429 | r = -ENOMEM; |
c5ff41ce | 5430 | kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags); |
7837699f SY |
5431 | if (kvm->arch.vpit) |
5432 | r = 0; | |
269e05e4 | 5433 | create_pit_unlock: |
250715a6 | 5434 | mutex_unlock(&kvm->lock); |
7837699f | 5435 | break; |
1fe779f8 CO |
5436 | case KVM_GET_IRQCHIP: { |
5437 | /* 0: PIC master, 1: PIC slave, 2: IOAPIC */ | |
ff5c2c03 | 5438 | struct kvm_irqchip *chip; |
1fe779f8 | 5439 | |
ff5c2c03 SL |
5440 | chip = memdup_user(argp, sizeof(*chip)); |
5441 | if (IS_ERR(chip)) { | |
5442 | r = PTR_ERR(chip); | |
1fe779f8 | 5443 | goto out; |
ff5c2c03 SL |
5444 | } |
5445 | ||
1fe779f8 | 5446 | r = -ENXIO; |
826da321 | 5447 | if (!irqchip_kernel(kvm)) |
f0d66275 DH |
5448 | goto get_irqchip_out; |
5449 | r = kvm_vm_ioctl_get_irqchip(kvm, chip); | |
1fe779f8 | 5450 | if (r) |
f0d66275 | 5451 | goto get_irqchip_out; |
1fe779f8 | 5452 | r = -EFAULT; |
0e96f31e | 5453 | if (copy_to_user(argp, chip, sizeof(*chip))) |
f0d66275 | 5454 | goto get_irqchip_out; |
1fe779f8 | 5455 | r = 0; |
f0d66275 DH |
5456 | get_irqchip_out: |
5457 | kfree(chip); | |
1fe779f8 CO |
5458 | break; |
5459 | } | |
5460 | case KVM_SET_IRQCHIP: { | |
5461 | /* 0: PIC master, 1: PIC slave, 2: IOAPIC */ | |
ff5c2c03 | 5462 | struct kvm_irqchip *chip; |
1fe779f8 | 5463 | |
ff5c2c03 SL |
5464 | chip = memdup_user(argp, sizeof(*chip)); |
5465 | if (IS_ERR(chip)) { | |
5466 | r = PTR_ERR(chip); | |
1fe779f8 | 5467 | goto out; |
ff5c2c03 SL |
5468 | } |
5469 | ||
1fe779f8 | 5470 | r = -ENXIO; |
826da321 | 5471 | if (!irqchip_kernel(kvm)) |
f0d66275 DH |
5472 | goto set_irqchip_out; |
5473 | r = kvm_vm_ioctl_set_irqchip(kvm, chip); | |
f0d66275 DH |
5474 | set_irqchip_out: |
5475 | kfree(chip); | |
1fe779f8 CO |
5476 | break; |
5477 | } | |
e0f63cb9 | 5478 | case KVM_GET_PIT: { |
e0f63cb9 | 5479 | r = -EFAULT; |
f0d66275 | 5480 | if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state))) |
e0f63cb9 SY |
5481 | goto out; |
5482 | r = -ENXIO; | |
5483 | if (!kvm->arch.vpit) | |
5484 | goto out; | |
f0d66275 | 5485 | r = kvm_vm_ioctl_get_pit(kvm, &u.ps); |
e0f63cb9 SY |
5486 | if (r) |
5487 | goto out; | |
5488 | r = -EFAULT; | |
f0d66275 | 5489 | if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state))) |
e0f63cb9 SY |
5490 | goto out; |
5491 | r = 0; | |
5492 | break; | |
5493 | } | |
5494 | case KVM_SET_PIT: { | |
e0f63cb9 | 5495 | r = -EFAULT; |
0e96f31e | 5496 | if (copy_from_user(&u.ps, argp, sizeof(u.ps))) |
e0f63cb9 | 5497 | goto out; |
7289fdb5 | 5498 | mutex_lock(&kvm->lock); |
e0f63cb9 SY |
5499 | r = -ENXIO; |
5500 | if (!kvm->arch.vpit) | |
7289fdb5 | 5501 | goto set_pit_out; |
f0d66275 | 5502 | r = kvm_vm_ioctl_set_pit(kvm, &u.ps); |
7289fdb5 SR |
5503 | set_pit_out: |
5504 | mutex_unlock(&kvm->lock); | |
e0f63cb9 SY |
5505 | break; |
5506 | } | |
e9f42757 BK |
5507 | case KVM_GET_PIT2: { |
5508 | r = -ENXIO; | |
5509 | if (!kvm->arch.vpit) | |
5510 | goto out; | |
5511 | r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2); | |
5512 | if (r) | |
5513 | goto out; | |
5514 | r = -EFAULT; | |
5515 | if (copy_to_user(argp, &u.ps2, sizeof(u.ps2))) | |
5516 | goto out; | |
5517 | r = 0; | |
5518 | break; | |
5519 | } | |
5520 | case KVM_SET_PIT2: { | |
5521 | r = -EFAULT; | |
5522 | if (copy_from_user(&u.ps2, argp, sizeof(u.ps2))) | |
5523 | goto out; | |
7289fdb5 | 5524 | mutex_lock(&kvm->lock); |
e9f42757 BK |
5525 | r = -ENXIO; |
5526 | if (!kvm->arch.vpit) | |
7289fdb5 | 5527 | goto set_pit2_out; |
e9f42757 | 5528 | r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2); |
7289fdb5 SR |
5529 | set_pit2_out: |
5530 | mutex_unlock(&kvm->lock); | |
e9f42757 BK |
5531 | break; |
5532 | } | |
52d939a0 MT |
5533 | case KVM_REINJECT_CONTROL: { |
5534 | struct kvm_reinject_control control; | |
5535 | r = -EFAULT; | |
5536 | if (copy_from_user(&control, argp, sizeof(control))) | |
5537 | goto out; | |
cad23e72 ML |
5538 | r = -ENXIO; |
5539 | if (!kvm->arch.vpit) | |
5540 | goto out; | |
52d939a0 | 5541 | r = kvm_vm_ioctl_reinject(kvm, &control); |
52d939a0 MT |
5542 | break; |
5543 | } | |
d71ba788 PB |
5544 | case KVM_SET_BOOT_CPU_ID: |
5545 | r = 0; | |
5546 | mutex_lock(&kvm->lock); | |
557abc40 | 5547 | if (kvm->created_vcpus) |
d71ba788 PB |
5548 | r = -EBUSY; |
5549 | else | |
5550 | kvm->arch.bsp_vcpu_id = arg; | |
5551 | mutex_unlock(&kvm->lock); | |
5552 | break; | |
ffde22ac | 5553 | case KVM_XEN_HVM_CONFIG: { |
51776043 | 5554 | struct kvm_xen_hvm_config xhc; |
ffde22ac | 5555 | r = -EFAULT; |
51776043 | 5556 | if (copy_from_user(&xhc, argp, sizeof(xhc))) |
ffde22ac ES |
5557 | goto out; |
5558 | r = -EINVAL; | |
51776043 | 5559 | if (xhc.flags) |
ffde22ac | 5560 | goto out; |
51776043 | 5561 | memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc)); |
ffde22ac ES |
5562 | r = 0; |
5563 | break; | |
5564 | } | |
afbcf7ab | 5565 | case KVM_SET_CLOCK: { |
afbcf7ab GC |
5566 | struct kvm_clock_data user_ns; |
5567 | u64 now_ns; | |
afbcf7ab GC |
5568 | |
5569 | r = -EFAULT; | |
5570 | if (copy_from_user(&user_ns, argp, sizeof(user_ns))) | |
5571 | goto out; | |
5572 | ||
5573 | r = -EINVAL; | |
5574 | if (user_ns.flags) | |
5575 | goto out; | |
5576 | ||
5577 | r = 0; | |
0bc48bea RK |
5578 | /* |
5579 | * TODO: userspace has to take care of races with VCPU_RUN, so | |
5580 | * kvm_gen_update_masterclock() can be cut down to locked | |
5581 | * pvclock_update_vm_gtod_copy(). | |
5582 | */ | |
5583 | kvm_gen_update_masterclock(kvm); | |
e891a32e | 5584 | now_ns = get_kvmclock_ns(kvm); |
108b249c | 5585 | kvm->arch.kvmclock_offset += user_ns.clock - now_ns; |
0bc48bea | 5586 | kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE); |
afbcf7ab GC |
5587 | break; |
5588 | } | |
5589 | case KVM_GET_CLOCK: { | |
afbcf7ab GC |
5590 | struct kvm_clock_data user_ns; |
5591 | u64 now_ns; | |
5592 | ||
e891a32e | 5593 | now_ns = get_kvmclock_ns(kvm); |
108b249c | 5594 | user_ns.clock = now_ns; |
e3fd9a93 | 5595 | user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0; |
97e69aa6 | 5596 | memset(&user_ns.pad, 0, sizeof(user_ns.pad)); |
afbcf7ab GC |
5597 | |
5598 | r = -EFAULT; | |
5599 | if (copy_to_user(argp, &user_ns, sizeof(user_ns))) | |
5600 | goto out; | |
5601 | r = 0; | |
5602 | break; | |
5603 | } | |
5acc5c06 BS |
5604 | case KVM_MEMORY_ENCRYPT_OP: { |
5605 | r = -ENOTTY; | |
afaf0b2f SC |
5606 | if (kvm_x86_ops.mem_enc_op) |
5607 | r = kvm_x86_ops.mem_enc_op(kvm, argp); | |
5acc5c06 BS |
5608 | break; |
5609 | } | |
69eaedee BS |
5610 | case KVM_MEMORY_ENCRYPT_REG_REGION: { |
5611 | struct kvm_enc_region region; | |
5612 | ||
5613 | r = -EFAULT; | |
5614 | if (copy_from_user(®ion, argp, sizeof(region))) | |
5615 | goto out; | |
5616 | ||
5617 | r = -ENOTTY; | |
afaf0b2f SC |
5618 | if (kvm_x86_ops.mem_enc_reg_region) |
5619 | r = kvm_x86_ops.mem_enc_reg_region(kvm, ®ion); | |
69eaedee BS |
5620 | break; |
5621 | } | |
5622 | case KVM_MEMORY_ENCRYPT_UNREG_REGION: { | |
5623 | struct kvm_enc_region region; | |
5624 | ||
5625 | r = -EFAULT; | |
5626 | if (copy_from_user(®ion, argp, sizeof(region))) | |
5627 | goto out; | |
5628 | ||
5629 | r = -ENOTTY; | |
afaf0b2f SC |
5630 | if (kvm_x86_ops.mem_enc_unreg_region) |
5631 | r = kvm_x86_ops.mem_enc_unreg_region(kvm, ®ion); | |
69eaedee BS |
5632 | break; |
5633 | } | |
faeb7833 RK |
5634 | case KVM_HYPERV_EVENTFD: { |
5635 | struct kvm_hyperv_eventfd hvevfd; | |
5636 | ||
5637 | r = -EFAULT; | |
5638 | if (copy_from_user(&hvevfd, argp, sizeof(hvevfd))) | |
5639 | goto out; | |
5640 | r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd); | |
5641 | break; | |
5642 | } | |
66bb8a06 EH |
5643 | case KVM_SET_PMU_EVENT_FILTER: |
5644 | r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp); | |
5645 | break; | |
1a155254 AG |
5646 | case KVM_X86_SET_MSR_FILTER: |
5647 | r = kvm_vm_ioctl_set_msr_filter(kvm, argp); | |
5648 | break; | |
1fe779f8 | 5649 | default: |
ad6260da | 5650 | r = -ENOTTY; |
1fe779f8 CO |
5651 | } |
5652 | out: | |
5653 | return r; | |
5654 | } | |
5655 | ||
a16b043c | 5656 | static void kvm_init_msr_list(void) |
043405e1 | 5657 | { |
24c29b7a | 5658 | struct x86_pmu_capability x86_pmu; |
043405e1 | 5659 | u32 dummy[2]; |
7a5ee6ed | 5660 | unsigned i; |
043405e1 | 5661 | |
e2ada66e | 5662 | BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4, |
7a5ee6ed | 5663 | "Please update the fixed PMCs in msrs_to_saved_all[]"); |
24c29b7a PB |
5664 | |
5665 | perf_get_x86_pmu_capability(&x86_pmu); | |
e2ada66e | 5666 | |
6cbee2b9 XL |
5667 | num_msrs_to_save = 0; |
5668 | num_emulated_msrs = 0; | |
5669 | num_msr_based_features = 0; | |
5670 | ||
7a5ee6ed CQ |
5671 | for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) { |
5672 | if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0) | |
043405e1 | 5673 | continue; |
93c4adc7 PB |
5674 | |
5675 | /* | |
5676 | * Even MSRs that are valid in the host may not be exposed | |
9dbe6cf9 | 5677 | * to the guests in some cases. |
93c4adc7 | 5678 | */ |
7a5ee6ed | 5679 | switch (msrs_to_save_all[i]) { |
93c4adc7 | 5680 | case MSR_IA32_BNDCFGS: |
503234b3 | 5681 | if (!kvm_mpx_supported()) |
93c4adc7 PB |
5682 | continue; |
5683 | break; | |
9dbe6cf9 | 5684 | case MSR_TSC_AUX: |
13908510 | 5685 | if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP)) |
9dbe6cf9 PB |
5686 | continue; |
5687 | break; | |
f4cfcd2d ML |
5688 | case MSR_IA32_UMWAIT_CONTROL: |
5689 | if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG)) | |
5690 | continue; | |
5691 | break; | |
bf8c55d8 CP |
5692 | case MSR_IA32_RTIT_CTL: |
5693 | case MSR_IA32_RTIT_STATUS: | |
7b874c26 | 5694 | if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT)) |
bf8c55d8 CP |
5695 | continue; |
5696 | break; | |
5697 | case MSR_IA32_RTIT_CR3_MATCH: | |
7b874c26 | 5698 | if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) || |
bf8c55d8 CP |
5699 | !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering)) |
5700 | continue; | |
5701 | break; | |
5702 | case MSR_IA32_RTIT_OUTPUT_BASE: | |
5703 | case MSR_IA32_RTIT_OUTPUT_MASK: | |
7b874c26 | 5704 | if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) || |
bf8c55d8 CP |
5705 | (!intel_pt_validate_hw_cap(PT_CAP_topa_output) && |
5706 | !intel_pt_validate_hw_cap(PT_CAP_single_range_output))) | |
5707 | continue; | |
5708 | break; | |
7cb85fc4 | 5709 | case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: |
7b874c26 | 5710 | if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) || |
7a5ee6ed | 5711 | msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >= |
bf8c55d8 CP |
5712 | intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2) |
5713 | continue; | |
5714 | break; | |
cf05a67b | 5715 | case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17: |
7a5ee6ed | 5716 | if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >= |
24c29b7a PB |
5717 | min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp)) |
5718 | continue; | |
5719 | break; | |
cf05a67b | 5720 | case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17: |
7a5ee6ed | 5721 | if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >= |
24c29b7a PB |
5722 | min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp)) |
5723 | continue; | |
7cb85fc4 | 5724 | break; |
93c4adc7 PB |
5725 | default: |
5726 | break; | |
5727 | } | |
5728 | ||
7a5ee6ed | 5729 | msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i]; |
043405e1 | 5730 | } |
62ef68bb | 5731 | |
7a5ee6ed | 5732 | for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) { |
afaf0b2f | 5733 | if (!kvm_x86_ops.has_emulated_msr(emulated_msrs_all[i])) |
bc226f07 | 5734 | continue; |
62ef68bb | 5735 | |
7a5ee6ed | 5736 | emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i]; |
62ef68bb | 5737 | } |
801e459a | 5738 | |
7a5ee6ed | 5739 | for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) { |
801e459a TL |
5740 | struct kvm_msr_entry msr; |
5741 | ||
7a5ee6ed | 5742 | msr.index = msr_based_features_all[i]; |
66421c1e | 5743 | if (kvm_get_msr_feature(&msr)) |
801e459a TL |
5744 | continue; |
5745 | ||
7a5ee6ed | 5746 | msr_based_features[num_msr_based_features++] = msr_based_features_all[i]; |
801e459a | 5747 | } |
043405e1 CO |
5748 | } |
5749 | ||
bda9020e MT |
5750 | static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len, |
5751 | const void *v) | |
bbd9b64e | 5752 | { |
70252a10 AK |
5753 | int handled = 0; |
5754 | int n; | |
5755 | ||
5756 | do { | |
5757 | n = min(len, 8); | |
bce87cce | 5758 | if (!(lapic_in_kernel(vcpu) && |
e32edf4f NN |
5759 | !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v)) |
5760 | && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v)) | |
70252a10 AK |
5761 | break; |
5762 | handled += n; | |
5763 | addr += n; | |
5764 | len -= n; | |
5765 | v += n; | |
5766 | } while (len); | |
bbd9b64e | 5767 | |
70252a10 | 5768 | return handled; |
bbd9b64e CO |
5769 | } |
5770 | ||
bda9020e | 5771 | static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v) |
bbd9b64e | 5772 | { |
70252a10 AK |
5773 | int handled = 0; |
5774 | int n; | |
5775 | ||
5776 | do { | |
5777 | n = min(len, 8); | |
bce87cce | 5778 | if (!(lapic_in_kernel(vcpu) && |
e32edf4f NN |
5779 | !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev, |
5780 | addr, n, v)) | |
5781 | && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v)) | |
70252a10 | 5782 | break; |
e39d200f | 5783 | trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v); |
70252a10 AK |
5784 | handled += n; |
5785 | addr += n; | |
5786 | len -= n; | |
5787 | v += n; | |
5788 | } while (len); | |
bbd9b64e | 5789 | |
70252a10 | 5790 | return handled; |
bbd9b64e CO |
5791 | } |
5792 | ||
2dafc6c2 GN |
5793 | static void kvm_set_segment(struct kvm_vcpu *vcpu, |
5794 | struct kvm_segment *var, int seg) | |
5795 | { | |
afaf0b2f | 5796 | kvm_x86_ops.set_segment(vcpu, var, seg); |
2dafc6c2 GN |
5797 | } |
5798 | ||
5799 | void kvm_get_segment(struct kvm_vcpu *vcpu, | |
5800 | struct kvm_segment *var, int seg) | |
5801 | { | |
afaf0b2f | 5802 | kvm_x86_ops.get_segment(vcpu, var, seg); |
2dafc6c2 GN |
5803 | } |
5804 | ||
54987b7a PB |
5805 | gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access, |
5806 | struct x86_exception *exception) | |
02f59dc9 JR |
5807 | { |
5808 | gpa_t t_gpa; | |
02f59dc9 JR |
5809 | |
5810 | BUG_ON(!mmu_is_nested(vcpu)); | |
5811 | ||
5812 | /* NPT walks are always user-walks */ | |
5813 | access |= PFERR_USER_MASK; | |
44dd3ffa | 5814 | t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception); |
02f59dc9 JR |
5815 | |
5816 | return t_gpa; | |
5817 | } | |
5818 | ||
ab9ae313 AK |
5819 | gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, |
5820 | struct x86_exception *exception) | |
1871c602 | 5821 | { |
afaf0b2f | 5822 | u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; |
ab9ae313 | 5823 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
1871c602 GN |
5824 | } |
5825 | ||
ab9ae313 AK |
5826 | gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, |
5827 | struct x86_exception *exception) | |
1871c602 | 5828 | { |
afaf0b2f | 5829 | u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; |
1871c602 | 5830 | access |= PFERR_FETCH_MASK; |
ab9ae313 | 5831 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
1871c602 GN |
5832 | } |
5833 | ||
ab9ae313 AK |
5834 | gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, |
5835 | struct x86_exception *exception) | |
1871c602 | 5836 | { |
afaf0b2f | 5837 | u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; |
1871c602 | 5838 | access |= PFERR_WRITE_MASK; |
ab9ae313 | 5839 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
1871c602 GN |
5840 | } |
5841 | ||
5842 | /* uses this to access any guest's mapped memory without checking CPL */ | |
ab9ae313 AK |
5843 | gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, |
5844 | struct x86_exception *exception) | |
1871c602 | 5845 | { |
ab9ae313 | 5846 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception); |
1871c602 GN |
5847 | } |
5848 | ||
5849 | static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes, | |
5850 | struct kvm_vcpu *vcpu, u32 access, | |
bcc55cba | 5851 | struct x86_exception *exception) |
bbd9b64e CO |
5852 | { |
5853 | void *data = val; | |
10589a46 | 5854 | int r = X86EMUL_CONTINUE; |
bbd9b64e CO |
5855 | |
5856 | while (bytes) { | |
14dfe855 | 5857 | gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access, |
ab9ae313 | 5858 | exception); |
bbd9b64e | 5859 | unsigned offset = addr & (PAGE_SIZE-1); |
77c2002e | 5860 | unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset); |
bbd9b64e CO |
5861 | int ret; |
5862 | ||
bcc55cba | 5863 | if (gpa == UNMAPPED_GVA) |
ab9ae313 | 5864 | return X86EMUL_PROPAGATE_FAULT; |
54bf36aa PB |
5865 | ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data, |
5866 | offset, toread); | |
10589a46 | 5867 | if (ret < 0) { |
c3cd7ffa | 5868 | r = X86EMUL_IO_NEEDED; |
10589a46 MT |
5869 | goto out; |
5870 | } | |
bbd9b64e | 5871 | |
77c2002e IE |
5872 | bytes -= toread; |
5873 | data += toread; | |
5874 | addr += toread; | |
bbd9b64e | 5875 | } |
10589a46 | 5876 | out: |
10589a46 | 5877 | return r; |
bbd9b64e | 5878 | } |
77c2002e | 5879 | |
1871c602 | 5880 | /* used for instruction fetching */ |
0f65dd70 AK |
5881 | static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt, |
5882 | gva_t addr, void *val, unsigned int bytes, | |
bcc55cba | 5883 | struct x86_exception *exception) |
1871c602 | 5884 | { |
0f65dd70 | 5885 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
afaf0b2f | 5886 | u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; |
44583cba PB |
5887 | unsigned offset; |
5888 | int ret; | |
0f65dd70 | 5889 | |
44583cba PB |
5890 | /* Inline kvm_read_guest_virt_helper for speed. */ |
5891 | gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK, | |
5892 | exception); | |
5893 | if (unlikely(gpa == UNMAPPED_GVA)) | |
5894 | return X86EMUL_PROPAGATE_FAULT; | |
5895 | ||
5896 | offset = addr & (PAGE_SIZE-1); | |
5897 | if (WARN_ON(offset + bytes > PAGE_SIZE)) | |
5898 | bytes = (unsigned)PAGE_SIZE - offset; | |
54bf36aa PB |
5899 | ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val, |
5900 | offset, bytes); | |
44583cba PB |
5901 | if (unlikely(ret < 0)) |
5902 | return X86EMUL_IO_NEEDED; | |
5903 | ||
5904 | return X86EMUL_CONTINUE; | |
1871c602 GN |
5905 | } |
5906 | ||
ce14e868 | 5907 | int kvm_read_guest_virt(struct kvm_vcpu *vcpu, |
0f65dd70 | 5908 | gva_t addr, void *val, unsigned int bytes, |
bcc55cba | 5909 | struct x86_exception *exception) |
1871c602 | 5910 | { |
afaf0b2f | 5911 | u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; |
0f65dd70 | 5912 | |
353c0956 PB |
5913 | /* |
5914 | * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED | |
5915 | * is returned, but our callers are not ready for that and they blindly | |
5916 | * call kvm_inject_page_fault. Ensure that they at least do not leak | |
5917 | * uninitialized kernel stack memory into cr2 and error code. | |
5918 | */ | |
5919 | memset(exception, 0, sizeof(*exception)); | |
1871c602 | 5920 | return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, |
bcc55cba | 5921 | exception); |
1871c602 | 5922 | } |
064aea77 | 5923 | EXPORT_SYMBOL_GPL(kvm_read_guest_virt); |
1871c602 | 5924 | |
ce14e868 PB |
5925 | static int emulator_read_std(struct x86_emulate_ctxt *ctxt, |
5926 | gva_t addr, void *val, unsigned int bytes, | |
3c9fa24c | 5927 | struct x86_exception *exception, bool system) |
1871c602 | 5928 | { |
0f65dd70 | 5929 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
3c9fa24c PB |
5930 | u32 access = 0; |
5931 | ||
afaf0b2f | 5932 | if (!system && kvm_x86_ops.get_cpl(vcpu) == 3) |
3c9fa24c PB |
5933 | access |= PFERR_USER_MASK; |
5934 | ||
5935 | return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception); | |
1871c602 GN |
5936 | } |
5937 | ||
7a036a6f RK |
5938 | static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt, |
5939 | unsigned long addr, void *val, unsigned int bytes) | |
5940 | { | |
5941 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
5942 | int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes); | |
5943 | ||
5944 | return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE; | |
5945 | } | |
5946 | ||
ce14e868 PB |
5947 | static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes, |
5948 | struct kvm_vcpu *vcpu, u32 access, | |
5949 | struct x86_exception *exception) | |
77c2002e IE |
5950 | { |
5951 | void *data = val; | |
5952 | int r = X86EMUL_CONTINUE; | |
5953 | ||
5954 | while (bytes) { | |
14dfe855 | 5955 | gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, |
ce14e868 | 5956 | access, |
ab9ae313 | 5957 | exception); |
77c2002e IE |
5958 | unsigned offset = addr & (PAGE_SIZE-1); |
5959 | unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset); | |
5960 | int ret; | |
5961 | ||
bcc55cba | 5962 | if (gpa == UNMAPPED_GVA) |
ab9ae313 | 5963 | return X86EMUL_PROPAGATE_FAULT; |
54bf36aa | 5964 | ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite); |
77c2002e | 5965 | if (ret < 0) { |
c3cd7ffa | 5966 | r = X86EMUL_IO_NEEDED; |
77c2002e IE |
5967 | goto out; |
5968 | } | |
5969 | ||
5970 | bytes -= towrite; | |
5971 | data += towrite; | |
5972 | addr += towrite; | |
5973 | } | |
5974 | out: | |
5975 | return r; | |
5976 | } | |
ce14e868 PB |
5977 | |
5978 | static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val, | |
3c9fa24c PB |
5979 | unsigned int bytes, struct x86_exception *exception, |
5980 | bool system) | |
ce14e868 PB |
5981 | { |
5982 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
3c9fa24c PB |
5983 | u32 access = PFERR_WRITE_MASK; |
5984 | ||
afaf0b2f | 5985 | if (!system && kvm_x86_ops.get_cpl(vcpu) == 3) |
3c9fa24c | 5986 | access |= PFERR_USER_MASK; |
ce14e868 PB |
5987 | |
5988 | return kvm_write_guest_virt_helper(addr, val, bytes, vcpu, | |
3c9fa24c | 5989 | access, exception); |
ce14e868 PB |
5990 | } |
5991 | ||
5992 | int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val, | |
5993 | unsigned int bytes, struct x86_exception *exception) | |
5994 | { | |
c595ceee PB |
5995 | /* kvm_write_guest_virt_system can pull in tons of pages. */ |
5996 | vcpu->arch.l1tf_flush_l1d = true; | |
5997 | ||
ce14e868 PB |
5998 | return kvm_write_guest_virt_helper(addr, val, bytes, vcpu, |
5999 | PFERR_WRITE_MASK, exception); | |
6000 | } | |
6a4d7550 | 6001 | EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system); |
77c2002e | 6002 | |
082d06ed WL |
6003 | int handle_ud(struct kvm_vcpu *vcpu) |
6004 | { | |
b3dc0695 | 6005 | static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX }; |
6c86eedc | 6006 | int emul_type = EMULTYPE_TRAP_UD; |
6c86eedc WL |
6007 | char sig[5]; /* ud2; .ascii "kvm" */ |
6008 | struct x86_exception e; | |
6009 | ||
09e3e2a1 SC |
6010 | if (unlikely(!kvm_x86_ops.can_emulate_instruction(vcpu, NULL, 0))) |
6011 | return 1; | |
6012 | ||
6c86eedc | 6013 | if (force_emulation_prefix && |
3c9fa24c PB |
6014 | kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu), |
6015 | sig, sizeof(sig), &e) == 0 && | |
b3dc0695 | 6016 | memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) { |
6c86eedc | 6017 | kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig)); |
b4000606 | 6018 | emul_type = EMULTYPE_TRAP_UD_FORCED; |
6c86eedc | 6019 | } |
082d06ed | 6020 | |
60fc3d02 | 6021 | return kvm_emulate_instruction(vcpu, emul_type); |
082d06ed WL |
6022 | } |
6023 | EXPORT_SYMBOL_GPL(handle_ud); | |
6024 | ||
0f89b207 TL |
6025 | static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva, |
6026 | gpa_t gpa, bool write) | |
6027 | { | |
6028 | /* For APIC access vmexit */ | |
6029 | if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE) | |
6030 | return 1; | |
6031 | ||
6032 | if (vcpu_match_mmio_gpa(vcpu, gpa)) { | |
6033 | trace_vcpu_match_mmio(gva, gpa, write, true); | |
6034 | return 1; | |
6035 | } | |
6036 | ||
6037 | return 0; | |
6038 | } | |
6039 | ||
af7cc7d1 XG |
6040 | static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva, |
6041 | gpa_t *gpa, struct x86_exception *exception, | |
6042 | bool write) | |
6043 | { | |
afaf0b2f | 6044 | u32 access = ((kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0) |
97d64b78 | 6045 | | (write ? PFERR_WRITE_MASK : 0); |
af7cc7d1 | 6046 | |
be94f6b7 HH |
6047 | /* |
6048 | * currently PKRU is only applied to ept enabled guest so | |
6049 | * there is no pkey in EPT page table for L1 guest or EPT | |
6050 | * shadow page table for L2 guest. | |
6051 | */ | |
97d64b78 | 6052 | if (vcpu_match_mmio_gva(vcpu, gva) |
97ec8c06 | 6053 | && !permission_fault(vcpu, vcpu->arch.walk_mmu, |
871bd034 | 6054 | vcpu->arch.mmio_access, 0, access)) { |
bebb106a XG |
6055 | *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT | |
6056 | (gva & (PAGE_SIZE - 1)); | |
4f022648 | 6057 | trace_vcpu_match_mmio(gva, *gpa, write, false); |
bebb106a XG |
6058 | return 1; |
6059 | } | |
6060 | ||
af7cc7d1 XG |
6061 | *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
6062 | ||
6063 | if (*gpa == UNMAPPED_GVA) | |
6064 | return -1; | |
6065 | ||
0f89b207 | 6066 | return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write); |
af7cc7d1 XG |
6067 | } |
6068 | ||
3200f405 | 6069 | int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa, |
bcc55cba | 6070 | const void *val, int bytes) |
bbd9b64e CO |
6071 | { |
6072 | int ret; | |
6073 | ||
54bf36aa | 6074 | ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes); |
9f811285 | 6075 | if (ret < 0) |
bbd9b64e | 6076 | return 0; |
0eb05bf2 | 6077 | kvm_page_track_write(vcpu, gpa, val, bytes); |
bbd9b64e CO |
6078 | return 1; |
6079 | } | |
6080 | ||
77d197b2 XG |
6081 | struct read_write_emulator_ops { |
6082 | int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val, | |
6083 | int bytes); | |
6084 | int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa, | |
6085 | void *val, int bytes); | |
6086 | int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa, | |
6087 | int bytes, void *val); | |
6088 | int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa, | |
6089 | void *val, int bytes); | |
6090 | bool write; | |
6091 | }; | |
6092 | ||
6093 | static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes) | |
6094 | { | |
6095 | if (vcpu->mmio_read_completed) { | |
77d197b2 | 6096 | trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, |
e39d200f | 6097 | vcpu->mmio_fragments[0].gpa, val); |
77d197b2 XG |
6098 | vcpu->mmio_read_completed = 0; |
6099 | return 1; | |
6100 | } | |
6101 | ||
6102 | return 0; | |
6103 | } | |
6104 | ||
6105 | static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa, | |
6106 | void *val, int bytes) | |
6107 | { | |
54bf36aa | 6108 | return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes); |
77d197b2 XG |
6109 | } |
6110 | ||
6111 | static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa, | |
6112 | void *val, int bytes) | |
6113 | { | |
6114 | return emulator_write_phys(vcpu, gpa, val, bytes); | |
6115 | } | |
6116 | ||
6117 | static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val) | |
6118 | { | |
e39d200f | 6119 | trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val); |
77d197b2 XG |
6120 | return vcpu_mmio_write(vcpu, gpa, bytes, val); |
6121 | } | |
6122 | ||
6123 | static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, | |
6124 | void *val, int bytes) | |
6125 | { | |
e39d200f | 6126 | trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL); |
77d197b2 XG |
6127 | return X86EMUL_IO_NEEDED; |
6128 | } | |
6129 | ||
6130 | static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, | |
6131 | void *val, int bytes) | |
6132 | { | |
f78146b0 AK |
6133 | struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0]; |
6134 | ||
87da7e66 | 6135 | memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len)); |
77d197b2 XG |
6136 | return X86EMUL_CONTINUE; |
6137 | } | |
6138 | ||
0fbe9b0b | 6139 | static const struct read_write_emulator_ops read_emultor = { |
77d197b2 XG |
6140 | .read_write_prepare = read_prepare, |
6141 | .read_write_emulate = read_emulate, | |
6142 | .read_write_mmio = vcpu_mmio_read, | |
6143 | .read_write_exit_mmio = read_exit_mmio, | |
6144 | }; | |
6145 | ||
0fbe9b0b | 6146 | static const struct read_write_emulator_ops write_emultor = { |
77d197b2 XG |
6147 | .read_write_emulate = write_emulate, |
6148 | .read_write_mmio = write_mmio, | |
6149 | .read_write_exit_mmio = write_exit_mmio, | |
6150 | .write = true, | |
6151 | }; | |
6152 | ||
22388a3c XG |
6153 | static int emulator_read_write_onepage(unsigned long addr, void *val, |
6154 | unsigned int bytes, | |
6155 | struct x86_exception *exception, | |
6156 | struct kvm_vcpu *vcpu, | |
0fbe9b0b | 6157 | const struct read_write_emulator_ops *ops) |
bbd9b64e | 6158 | { |
af7cc7d1 XG |
6159 | gpa_t gpa; |
6160 | int handled, ret; | |
22388a3c | 6161 | bool write = ops->write; |
f78146b0 | 6162 | struct kvm_mmio_fragment *frag; |
c9b8b07c | 6163 | struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt; |
0f89b207 TL |
6164 | |
6165 | /* | |
6166 | * If the exit was due to a NPF we may already have a GPA. | |
6167 | * If the GPA is present, use it to avoid the GVA to GPA table walk. | |
6168 | * Note, this cannot be used on string operations since string | |
6169 | * operation using rep will only have the initial GPA from the NPF | |
6170 | * occurred. | |
6171 | */ | |
744e699c SC |
6172 | if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) && |
6173 | (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) { | |
6174 | gpa = ctxt->gpa_val; | |
618232e2 BS |
6175 | ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write); |
6176 | } else { | |
6177 | ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write); | |
6178 | if (ret < 0) | |
6179 | return X86EMUL_PROPAGATE_FAULT; | |
0f89b207 | 6180 | } |
10589a46 | 6181 | |
618232e2 | 6182 | if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes)) |
bbd9b64e CO |
6183 | return X86EMUL_CONTINUE; |
6184 | ||
bbd9b64e CO |
6185 | /* |
6186 | * Is this MMIO handled locally? | |
6187 | */ | |
22388a3c | 6188 | handled = ops->read_write_mmio(vcpu, gpa, bytes, val); |
70252a10 | 6189 | if (handled == bytes) |
bbd9b64e | 6190 | return X86EMUL_CONTINUE; |
bbd9b64e | 6191 | |
70252a10 AK |
6192 | gpa += handled; |
6193 | bytes -= handled; | |
6194 | val += handled; | |
6195 | ||
87da7e66 XG |
6196 | WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS); |
6197 | frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++]; | |
6198 | frag->gpa = gpa; | |
6199 | frag->data = val; | |
6200 | frag->len = bytes; | |
f78146b0 | 6201 | return X86EMUL_CONTINUE; |
bbd9b64e CO |
6202 | } |
6203 | ||
52eb5a6d XL |
6204 | static int emulator_read_write(struct x86_emulate_ctxt *ctxt, |
6205 | unsigned long addr, | |
22388a3c XG |
6206 | void *val, unsigned int bytes, |
6207 | struct x86_exception *exception, | |
0fbe9b0b | 6208 | const struct read_write_emulator_ops *ops) |
bbd9b64e | 6209 | { |
0f65dd70 | 6210 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
f78146b0 AK |
6211 | gpa_t gpa; |
6212 | int rc; | |
6213 | ||
6214 | if (ops->read_write_prepare && | |
6215 | ops->read_write_prepare(vcpu, val, bytes)) | |
6216 | return X86EMUL_CONTINUE; | |
6217 | ||
6218 | vcpu->mmio_nr_fragments = 0; | |
0f65dd70 | 6219 | |
bbd9b64e CO |
6220 | /* Crossing a page boundary? */ |
6221 | if (((addr + bytes - 1) ^ addr) & PAGE_MASK) { | |
f78146b0 | 6222 | int now; |
bbd9b64e CO |
6223 | |
6224 | now = -addr & ~PAGE_MASK; | |
22388a3c XG |
6225 | rc = emulator_read_write_onepage(addr, val, now, exception, |
6226 | vcpu, ops); | |
6227 | ||
bbd9b64e CO |
6228 | if (rc != X86EMUL_CONTINUE) |
6229 | return rc; | |
6230 | addr += now; | |
bac15531 NA |
6231 | if (ctxt->mode != X86EMUL_MODE_PROT64) |
6232 | addr = (u32)addr; | |
bbd9b64e CO |
6233 | val += now; |
6234 | bytes -= now; | |
6235 | } | |
22388a3c | 6236 | |
f78146b0 AK |
6237 | rc = emulator_read_write_onepage(addr, val, bytes, exception, |
6238 | vcpu, ops); | |
6239 | if (rc != X86EMUL_CONTINUE) | |
6240 | return rc; | |
6241 | ||
6242 | if (!vcpu->mmio_nr_fragments) | |
6243 | return rc; | |
6244 | ||
6245 | gpa = vcpu->mmio_fragments[0].gpa; | |
6246 | ||
6247 | vcpu->mmio_needed = 1; | |
6248 | vcpu->mmio_cur_fragment = 0; | |
6249 | ||
87da7e66 | 6250 | vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len); |
f78146b0 AK |
6251 | vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write; |
6252 | vcpu->run->exit_reason = KVM_EXIT_MMIO; | |
6253 | vcpu->run->mmio.phys_addr = gpa; | |
6254 | ||
6255 | return ops->read_write_exit_mmio(vcpu, gpa, val, bytes); | |
22388a3c XG |
6256 | } |
6257 | ||
6258 | static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt, | |
6259 | unsigned long addr, | |
6260 | void *val, | |
6261 | unsigned int bytes, | |
6262 | struct x86_exception *exception) | |
6263 | { | |
6264 | return emulator_read_write(ctxt, addr, val, bytes, | |
6265 | exception, &read_emultor); | |
6266 | } | |
6267 | ||
52eb5a6d | 6268 | static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt, |
22388a3c XG |
6269 | unsigned long addr, |
6270 | const void *val, | |
6271 | unsigned int bytes, | |
6272 | struct x86_exception *exception) | |
6273 | { | |
6274 | return emulator_read_write(ctxt, addr, (void *)val, bytes, | |
6275 | exception, &write_emultor); | |
bbd9b64e | 6276 | } |
bbd9b64e | 6277 | |
daea3e73 AK |
6278 | #define CMPXCHG_TYPE(t, ptr, old, new) \ |
6279 | (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old)) | |
6280 | ||
6281 | #ifdef CONFIG_X86_64 | |
6282 | # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new) | |
6283 | #else | |
6284 | # define CMPXCHG64(ptr, old, new) \ | |
9749a6c0 | 6285 | (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old)) |
daea3e73 AK |
6286 | #endif |
6287 | ||
0f65dd70 AK |
6288 | static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt, |
6289 | unsigned long addr, | |
bbd9b64e CO |
6290 | const void *old, |
6291 | const void *new, | |
6292 | unsigned int bytes, | |
0f65dd70 | 6293 | struct x86_exception *exception) |
bbd9b64e | 6294 | { |
42e35f80 | 6295 | struct kvm_host_map map; |
0f65dd70 | 6296 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
9de6fe3c | 6297 | u64 page_line_mask; |
daea3e73 | 6298 | gpa_t gpa; |
daea3e73 AK |
6299 | char *kaddr; |
6300 | bool exchanged; | |
2bacc55c | 6301 | |
daea3e73 AK |
6302 | /* guests cmpxchg8b have to be emulated atomically */ |
6303 | if (bytes > 8 || (bytes & (bytes - 1))) | |
6304 | goto emul_write; | |
10589a46 | 6305 | |
daea3e73 | 6306 | gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL); |
2bacc55c | 6307 | |
daea3e73 AK |
6308 | if (gpa == UNMAPPED_GVA || |
6309 | (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE) | |
6310 | goto emul_write; | |
2bacc55c | 6311 | |
9de6fe3c XL |
6312 | /* |
6313 | * Emulate the atomic as a straight write to avoid #AC if SLD is | |
6314 | * enabled in the host and the access splits a cache line. | |
6315 | */ | |
6316 | if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT)) | |
6317 | page_line_mask = ~(cache_line_size() - 1); | |
6318 | else | |
6319 | page_line_mask = PAGE_MASK; | |
6320 | ||
6321 | if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask)) | |
daea3e73 | 6322 | goto emul_write; |
72dc67a6 | 6323 | |
42e35f80 | 6324 | if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map)) |
c19b8bd6 | 6325 | goto emul_write; |
72dc67a6 | 6326 | |
42e35f80 KA |
6327 | kaddr = map.hva + offset_in_page(gpa); |
6328 | ||
daea3e73 AK |
6329 | switch (bytes) { |
6330 | case 1: | |
6331 | exchanged = CMPXCHG_TYPE(u8, kaddr, old, new); | |
6332 | break; | |
6333 | case 2: | |
6334 | exchanged = CMPXCHG_TYPE(u16, kaddr, old, new); | |
6335 | break; | |
6336 | case 4: | |
6337 | exchanged = CMPXCHG_TYPE(u32, kaddr, old, new); | |
6338 | break; | |
6339 | case 8: | |
6340 | exchanged = CMPXCHG64(kaddr, old, new); | |
6341 | break; | |
6342 | default: | |
6343 | BUG(); | |
2bacc55c | 6344 | } |
42e35f80 KA |
6345 | |
6346 | kvm_vcpu_unmap(vcpu, &map, true); | |
daea3e73 AK |
6347 | |
6348 | if (!exchanged) | |
6349 | return X86EMUL_CMPXCHG_FAILED; | |
6350 | ||
0eb05bf2 | 6351 | kvm_page_track_write(vcpu, gpa, new, bytes); |
8f6abd06 GN |
6352 | |
6353 | return X86EMUL_CONTINUE; | |
4a5f48f6 | 6354 | |
3200f405 | 6355 | emul_write: |
daea3e73 | 6356 | printk_once(KERN_WARNING "kvm: emulating exchange as write\n"); |
2bacc55c | 6357 | |
0f65dd70 | 6358 | return emulator_write_emulated(ctxt, addr, new, bytes, exception); |
bbd9b64e CO |
6359 | } |
6360 | ||
cf8f70bf GN |
6361 | static int kernel_pio(struct kvm_vcpu *vcpu, void *pd) |
6362 | { | |
cbfc6c91 | 6363 | int r = 0, i; |
cf8f70bf | 6364 | |
cbfc6c91 WL |
6365 | for (i = 0; i < vcpu->arch.pio.count; i++) { |
6366 | if (vcpu->arch.pio.in) | |
6367 | r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port, | |
6368 | vcpu->arch.pio.size, pd); | |
6369 | else | |
6370 | r = kvm_io_bus_write(vcpu, KVM_PIO_BUS, | |
6371 | vcpu->arch.pio.port, vcpu->arch.pio.size, | |
6372 | pd); | |
6373 | if (r) | |
6374 | break; | |
6375 | pd += vcpu->arch.pio.size; | |
6376 | } | |
cf8f70bf GN |
6377 | return r; |
6378 | } | |
6379 | ||
6f6fbe98 XG |
6380 | static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size, |
6381 | unsigned short port, void *val, | |
6382 | unsigned int count, bool in) | |
cf8f70bf | 6383 | { |
cf8f70bf | 6384 | vcpu->arch.pio.port = port; |
6f6fbe98 | 6385 | vcpu->arch.pio.in = in; |
7972995b | 6386 | vcpu->arch.pio.count = count; |
cf8f70bf GN |
6387 | vcpu->arch.pio.size = size; |
6388 | ||
6389 | if (!kernel_pio(vcpu, vcpu->arch.pio_data)) { | |
7972995b | 6390 | vcpu->arch.pio.count = 0; |
cf8f70bf GN |
6391 | return 1; |
6392 | } | |
6393 | ||
6394 | vcpu->run->exit_reason = KVM_EXIT_IO; | |
6f6fbe98 | 6395 | vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT; |
cf8f70bf GN |
6396 | vcpu->run->io.size = size; |
6397 | vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE; | |
6398 | vcpu->run->io.count = count; | |
6399 | vcpu->run->io.port = port; | |
6400 | ||
6401 | return 0; | |
6402 | } | |
6403 | ||
2e3bb4d8 SC |
6404 | static int emulator_pio_in(struct kvm_vcpu *vcpu, int size, |
6405 | unsigned short port, void *val, unsigned int count) | |
cf8f70bf | 6406 | { |
6f6fbe98 | 6407 | int ret; |
ca1d4a9e | 6408 | |
6f6fbe98 XG |
6409 | if (vcpu->arch.pio.count) |
6410 | goto data_avail; | |
cf8f70bf | 6411 | |
cbfc6c91 WL |
6412 | memset(vcpu->arch.pio_data, 0, size * count); |
6413 | ||
6f6fbe98 XG |
6414 | ret = emulator_pio_in_out(vcpu, size, port, val, count, true); |
6415 | if (ret) { | |
6416 | data_avail: | |
6417 | memcpy(val, vcpu->arch.pio_data, size * count); | |
1171903d | 6418 | trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data); |
7972995b | 6419 | vcpu->arch.pio.count = 0; |
cf8f70bf GN |
6420 | return 1; |
6421 | } | |
6422 | ||
cf8f70bf GN |
6423 | return 0; |
6424 | } | |
6425 | ||
2e3bb4d8 SC |
6426 | static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt, |
6427 | int size, unsigned short port, void *val, | |
6428 | unsigned int count) | |
6f6fbe98 | 6429 | { |
2e3bb4d8 | 6430 | return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count); |
6f6fbe98 | 6431 | |
2e3bb4d8 | 6432 | } |
6f6fbe98 | 6433 | |
2e3bb4d8 SC |
6434 | static int emulator_pio_out(struct kvm_vcpu *vcpu, int size, |
6435 | unsigned short port, const void *val, | |
6436 | unsigned int count) | |
6437 | { | |
6f6fbe98 | 6438 | memcpy(vcpu->arch.pio_data, val, size * count); |
1171903d | 6439 | trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data); |
6f6fbe98 XG |
6440 | return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false); |
6441 | } | |
6442 | ||
2e3bb4d8 SC |
6443 | static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt, |
6444 | int size, unsigned short port, | |
6445 | const void *val, unsigned int count) | |
6446 | { | |
6447 | return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count); | |
6448 | } | |
6449 | ||
bbd9b64e CO |
6450 | static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg) |
6451 | { | |
afaf0b2f | 6452 | return kvm_x86_ops.get_segment_base(vcpu, seg); |
bbd9b64e CO |
6453 | } |
6454 | ||
3cb16fe7 | 6455 | static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address) |
bbd9b64e | 6456 | { |
3cb16fe7 | 6457 | kvm_mmu_invlpg(emul_to_vcpu(ctxt), address); |
bbd9b64e CO |
6458 | } |
6459 | ||
ae6a2375 | 6460 | static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu) |
f5f48ee1 SY |
6461 | { |
6462 | if (!need_emulate_wbinvd(vcpu)) | |
6463 | return X86EMUL_CONTINUE; | |
6464 | ||
afaf0b2f | 6465 | if (kvm_x86_ops.has_wbinvd_exit()) { |
2eec7343 JK |
6466 | int cpu = get_cpu(); |
6467 | ||
6468 | cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask); | |
f5f48ee1 SY |
6469 | smp_call_function_many(vcpu->arch.wbinvd_dirty_mask, |
6470 | wbinvd_ipi, NULL, 1); | |
2eec7343 | 6471 | put_cpu(); |
f5f48ee1 | 6472 | cpumask_clear(vcpu->arch.wbinvd_dirty_mask); |
2eec7343 JK |
6473 | } else |
6474 | wbinvd(); | |
f5f48ee1 SY |
6475 | return X86EMUL_CONTINUE; |
6476 | } | |
5cb56059 JS |
6477 | |
6478 | int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu) | |
6479 | { | |
6affcbed KH |
6480 | kvm_emulate_wbinvd_noskip(vcpu); |
6481 | return kvm_skip_emulated_instruction(vcpu); | |
5cb56059 | 6482 | } |
f5f48ee1 SY |
6483 | EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd); |
6484 | ||
5cb56059 JS |
6485 | |
6486 | ||
bcaf5cc5 AK |
6487 | static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt) |
6488 | { | |
5cb56059 | 6489 | kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt)); |
bcaf5cc5 AK |
6490 | } |
6491 | ||
52eb5a6d XL |
6492 | static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, |
6493 | unsigned long *dest) | |
bbd9b64e | 6494 | { |
16f8a6f9 | 6495 | return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest); |
bbd9b64e CO |
6496 | } |
6497 | ||
52eb5a6d XL |
6498 | static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, |
6499 | unsigned long value) | |
bbd9b64e | 6500 | { |
338dbc97 | 6501 | |
717746e3 | 6502 | return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value); |
bbd9b64e CO |
6503 | } |
6504 | ||
52a46617 | 6505 | static u64 mk_cr_64(u64 curr_cr, u32 new_val) |
5fdbf976 | 6506 | { |
52a46617 | 6507 | return (curr_cr & ~((1ULL << 32) - 1)) | new_val; |
5fdbf976 MT |
6508 | } |
6509 | ||
717746e3 | 6510 | static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr) |
bbd9b64e | 6511 | { |
717746e3 | 6512 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
52a46617 GN |
6513 | unsigned long value; |
6514 | ||
6515 | switch (cr) { | |
6516 | case 0: | |
6517 | value = kvm_read_cr0(vcpu); | |
6518 | break; | |
6519 | case 2: | |
6520 | value = vcpu->arch.cr2; | |
6521 | break; | |
6522 | case 3: | |
9f8fe504 | 6523 | value = kvm_read_cr3(vcpu); |
52a46617 GN |
6524 | break; |
6525 | case 4: | |
6526 | value = kvm_read_cr4(vcpu); | |
6527 | break; | |
6528 | case 8: | |
6529 | value = kvm_get_cr8(vcpu); | |
6530 | break; | |
6531 | default: | |
a737f256 | 6532 | kvm_err("%s: unexpected cr %u\n", __func__, cr); |
52a46617 GN |
6533 | return 0; |
6534 | } | |
6535 | ||
6536 | return value; | |
6537 | } | |
6538 | ||
717746e3 | 6539 | static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val) |
52a46617 | 6540 | { |
717746e3 | 6541 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
0f12244f GN |
6542 | int res = 0; |
6543 | ||
52a46617 GN |
6544 | switch (cr) { |
6545 | case 0: | |
49a9b07e | 6546 | res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val)); |
52a46617 GN |
6547 | break; |
6548 | case 2: | |
6549 | vcpu->arch.cr2 = val; | |
6550 | break; | |
6551 | case 3: | |
2390218b | 6552 | res = kvm_set_cr3(vcpu, val); |
52a46617 GN |
6553 | break; |
6554 | case 4: | |
a83b29c6 | 6555 | res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val)); |
52a46617 GN |
6556 | break; |
6557 | case 8: | |
eea1cff9 | 6558 | res = kvm_set_cr8(vcpu, val); |
52a46617 GN |
6559 | break; |
6560 | default: | |
a737f256 | 6561 | kvm_err("%s: unexpected cr %u\n", __func__, cr); |
0f12244f | 6562 | res = -1; |
52a46617 | 6563 | } |
0f12244f GN |
6564 | |
6565 | return res; | |
52a46617 GN |
6566 | } |
6567 | ||
717746e3 | 6568 | static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt) |
9c537244 | 6569 | { |
afaf0b2f | 6570 | return kvm_x86_ops.get_cpl(emul_to_vcpu(ctxt)); |
9c537244 GN |
6571 | } |
6572 | ||
4bff1e86 | 6573 | static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) |
2dafc6c2 | 6574 | { |
afaf0b2f | 6575 | kvm_x86_ops.get_gdt(emul_to_vcpu(ctxt), dt); |
2dafc6c2 GN |
6576 | } |
6577 | ||
4bff1e86 | 6578 | static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) |
160ce1f1 | 6579 | { |
afaf0b2f | 6580 | kvm_x86_ops.get_idt(emul_to_vcpu(ctxt), dt); |
160ce1f1 MG |
6581 | } |
6582 | ||
1ac9d0cf AK |
6583 | static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) |
6584 | { | |
afaf0b2f | 6585 | kvm_x86_ops.set_gdt(emul_to_vcpu(ctxt), dt); |
1ac9d0cf AK |
6586 | } |
6587 | ||
6588 | static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) | |
6589 | { | |
afaf0b2f | 6590 | kvm_x86_ops.set_idt(emul_to_vcpu(ctxt), dt); |
1ac9d0cf AK |
6591 | } |
6592 | ||
4bff1e86 AK |
6593 | static unsigned long emulator_get_cached_segment_base( |
6594 | struct x86_emulate_ctxt *ctxt, int seg) | |
5951c442 | 6595 | { |
4bff1e86 | 6596 | return get_segment_base(emul_to_vcpu(ctxt), seg); |
5951c442 GN |
6597 | } |
6598 | ||
1aa36616 AK |
6599 | static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector, |
6600 | struct desc_struct *desc, u32 *base3, | |
6601 | int seg) | |
2dafc6c2 GN |
6602 | { |
6603 | struct kvm_segment var; | |
6604 | ||
4bff1e86 | 6605 | kvm_get_segment(emul_to_vcpu(ctxt), &var, seg); |
1aa36616 | 6606 | *selector = var.selector; |
2dafc6c2 | 6607 | |
378a8b09 GN |
6608 | if (var.unusable) { |
6609 | memset(desc, 0, sizeof(*desc)); | |
f0367ee1 RK |
6610 | if (base3) |
6611 | *base3 = 0; | |
2dafc6c2 | 6612 | return false; |
378a8b09 | 6613 | } |
2dafc6c2 GN |
6614 | |
6615 | if (var.g) | |
6616 | var.limit >>= 12; | |
6617 | set_desc_limit(desc, var.limit); | |
6618 | set_desc_base(desc, (unsigned long)var.base); | |
5601d05b GN |
6619 | #ifdef CONFIG_X86_64 |
6620 | if (base3) | |
6621 | *base3 = var.base >> 32; | |
6622 | #endif | |
2dafc6c2 GN |
6623 | desc->type = var.type; |
6624 | desc->s = var.s; | |
6625 | desc->dpl = var.dpl; | |
6626 | desc->p = var.present; | |
6627 | desc->avl = var.avl; | |
6628 | desc->l = var.l; | |
6629 | desc->d = var.db; | |
6630 | desc->g = var.g; | |
6631 | ||
6632 | return true; | |
6633 | } | |
6634 | ||
1aa36616 AK |
6635 | static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector, |
6636 | struct desc_struct *desc, u32 base3, | |
6637 | int seg) | |
2dafc6c2 | 6638 | { |
4bff1e86 | 6639 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
2dafc6c2 GN |
6640 | struct kvm_segment var; |
6641 | ||
1aa36616 | 6642 | var.selector = selector; |
2dafc6c2 | 6643 | var.base = get_desc_base(desc); |
5601d05b GN |
6644 | #ifdef CONFIG_X86_64 |
6645 | var.base |= ((u64)base3) << 32; | |
6646 | #endif | |
2dafc6c2 GN |
6647 | var.limit = get_desc_limit(desc); |
6648 | if (desc->g) | |
6649 | var.limit = (var.limit << 12) | 0xfff; | |
6650 | var.type = desc->type; | |
2dafc6c2 GN |
6651 | var.dpl = desc->dpl; |
6652 | var.db = desc->d; | |
6653 | var.s = desc->s; | |
6654 | var.l = desc->l; | |
6655 | var.g = desc->g; | |
6656 | var.avl = desc->avl; | |
6657 | var.present = desc->p; | |
6658 | var.unusable = !var.present; | |
6659 | var.padding = 0; | |
6660 | ||
6661 | kvm_set_segment(vcpu, &var, seg); | |
6662 | return; | |
6663 | } | |
6664 | ||
717746e3 AK |
6665 | static int emulator_get_msr(struct x86_emulate_ctxt *ctxt, |
6666 | u32 msr_index, u64 *pdata) | |
6667 | { | |
1ae09954 AG |
6668 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
6669 | int r; | |
6670 | ||
6671 | r = kvm_get_msr(vcpu, msr_index, pdata); | |
6672 | ||
6673 | if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) { | |
6674 | /* Bounce to user space */ | |
6675 | return X86EMUL_IO_NEEDED; | |
6676 | } | |
6677 | ||
6678 | return r; | |
717746e3 AK |
6679 | } |
6680 | ||
6681 | static int emulator_set_msr(struct x86_emulate_ctxt *ctxt, | |
6682 | u32 msr_index, u64 data) | |
6683 | { | |
1ae09954 AG |
6684 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
6685 | int r; | |
6686 | ||
6687 | r = kvm_set_msr(vcpu, msr_index, data); | |
6688 | ||
6689 | if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) { | |
6690 | /* Bounce to user space */ | |
6691 | return X86EMUL_IO_NEEDED; | |
6692 | } | |
6693 | ||
6694 | return r; | |
717746e3 AK |
6695 | } |
6696 | ||
64d60670 PB |
6697 | static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt) |
6698 | { | |
6699 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
6700 | ||
6701 | return vcpu->arch.smbase; | |
6702 | } | |
6703 | ||
6704 | static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase) | |
6705 | { | |
6706 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
6707 | ||
6708 | vcpu->arch.smbase = smbase; | |
6709 | } | |
6710 | ||
67f4d428 NA |
6711 | static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt, |
6712 | u32 pmc) | |
6713 | { | |
98ff80f5 | 6714 | return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc); |
67f4d428 NA |
6715 | } |
6716 | ||
222d21aa AK |
6717 | static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt, |
6718 | u32 pmc, u64 *pdata) | |
6719 | { | |
c6702c9d | 6720 | return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata); |
222d21aa AK |
6721 | } |
6722 | ||
6c3287f7 AK |
6723 | static void emulator_halt(struct x86_emulate_ctxt *ctxt) |
6724 | { | |
6725 | emul_to_vcpu(ctxt)->arch.halt_request = 1; | |
6726 | } | |
6727 | ||
2953538e | 6728 | static int emulator_intercept(struct x86_emulate_ctxt *ctxt, |
8a76d7f2 | 6729 | struct x86_instruction_info *info, |
c4f035c6 AK |
6730 | enum x86_intercept_stage stage) |
6731 | { | |
afaf0b2f | 6732 | return kvm_x86_ops.check_intercept(emul_to_vcpu(ctxt), info, stage, |
21f1b8f2 | 6733 | &ctxt->exception); |
c4f035c6 AK |
6734 | } |
6735 | ||
e911eb3b | 6736 | static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt, |
f91af517 SC |
6737 | u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, |
6738 | bool exact_only) | |
bdb42f5a | 6739 | { |
f91af517 | 6740 | return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only); |
bdb42f5a SB |
6741 | } |
6742 | ||
5ae78e95 SC |
6743 | static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt) |
6744 | { | |
6745 | return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM); | |
6746 | } | |
6747 | ||
6748 | static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt) | |
6749 | { | |
6750 | return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE); | |
6751 | } | |
6752 | ||
6753 | static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt) | |
6754 | { | |
6755 | return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR); | |
6756 | } | |
6757 | ||
dd856efa AK |
6758 | static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg) |
6759 | { | |
6760 | return kvm_register_read(emul_to_vcpu(ctxt), reg); | |
6761 | } | |
6762 | ||
6763 | static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val) | |
6764 | { | |
6765 | kvm_register_write(emul_to_vcpu(ctxt), reg, val); | |
6766 | } | |
6767 | ||
801806d9 NA |
6768 | static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked) |
6769 | { | |
afaf0b2f | 6770 | kvm_x86_ops.set_nmi_mask(emul_to_vcpu(ctxt), masked); |
801806d9 NA |
6771 | } |
6772 | ||
6ed071f0 LP |
6773 | static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt) |
6774 | { | |
6775 | return emul_to_vcpu(ctxt)->arch.hflags; | |
6776 | } | |
6777 | ||
6778 | static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags) | |
6779 | { | |
c5833c7a | 6780 | emul_to_vcpu(ctxt)->arch.hflags = emul_flags; |
6ed071f0 LP |
6781 | } |
6782 | ||
ed19321f SC |
6783 | static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, |
6784 | const char *smstate) | |
0234bf88 | 6785 | { |
afaf0b2f | 6786 | return kvm_x86_ops.pre_leave_smm(emul_to_vcpu(ctxt), smstate); |
0234bf88 LP |
6787 | } |
6788 | ||
c5833c7a SC |
6789 | static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt) |
6790 | { | |
6791 | kvm_smm_changed(emul_to_vcpu(ctxt)); | |
6792 | } | |
6793 | ||
02d4160f VK |
6794 | static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr) |
6795 | { | |
6796 | return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr); | |
6797 | } | |
6798 | ||
0225fb50 | 6799 | static const struct x86_emulate_ops emulate_ops = { |
dd856efa AK |
6800 | .read_gpr = emulator_read_gpr, |
6801 | .write_gpr = emulator_write_gpr, | |
ce14e868 PB |
6802 | .read_std = emulator_read_std, |
6803 | .write_std = emulator_write_std, | |
7a036a6f | 6804 | .read_phys = kvm_read_guest_phys_system, |
1871c602 | 6805 | .fetch = kvm_fetch_guest_virt, |
bbd9b64e CO |
6806 | .read_emulated = emulator_read_emulated, |
6807 | .write_emulated = emulator_write_emulated, | |
6808 | .cmpxchg_emulated = emulator_cmpxchg_emulated, | |
3cb16fe7 | 6809 | .invlpg = emulator_invlpg, |
cf8f70bf GN |
6810 | .pio_in_emulated = emulator_pio_in_emulated, |
6811 | .pio_out_emulated = emulator_pio_out_emulated, | |
1aa36616 AK |
6812 | .get_segment = emulator_get_segment, |
6813 | .set_segment = emulator_set_segment, | |
5951c442 | 6814 | .get_cached_segment_base = emulator_get_cached_segment_base, |
2dafc6c2 | 6815 | .get_gdt = emulator_get_gdt, |
160ce1f1 | 6816 | .get_idt = emulator_get_idt, |
1ac9d0cf AK |
6817 | .set_gdt = emulator_set_gdt, |
6818 | .set_idt = emulator_set_idt, | |
52a46617 GN |
6819 | .get_cr = emulator_get_cr, |
6820 | .set_cr = emulator_set_cr, | |
9c537244 | 6821 | .cpl = emulator_get_cpl, |
35aa5375 GN |
6822 | .get_dr = emulator_get_dr, |
6823 | .set_dr = emulator_set_dr, | |
64d60670 PB |
6824 | .get_smbase = emulator_get_smbase, |
6825 | .set_smbase = emulator_set_smbase, | |
717746e3 AK |
6826 | .set_msr = emulator_set_msr, |
6827 | .get_msr = emulator_get_msr, | |
67f4d428 | 6828 | .check_pmc = emulator_check_pmc, |
222d21aa | 6829 | .read_pmc = emulator_read_pmc, |
6c3287f7 | 6830 | .halt = emulator_halt, |
bcaf5cc5 | 6831 | .wbinvd = emulator_wbinvd, |
d6aa1000 | 6832 | .fix_hypercall = emulator_fix_hypercall, |
c4f035c6 | 6833 | .intercept = emulator_intercept, |
bdb42f5a | 6834 | .get_cpuid = emulator_get_cpuid, |
5ae78e95 SC |
6835 | .guest_has_long_mode = emulator_guest_has_long_mode, |
6836 | .guest_has_movbe = emulator_guest_has_movbe, | |
6837 | .guest_has_fxsr = emulator_guest_has_fxsr, | |
801806d9 | 6838 | .set_nmi_mask = emulator_set_nmi_mask, |
6ed071f0 LP |
6839 | .get_hflags = emulator_get_hflags, |
6840 | .set_hflags = emulator_set_hflags, | |
0234bf88 | 6841 | .pre_leave_smm = emulator_pre_leave_smm, |
c5833c7a | 6842 | .post_leave_smm = emulator_post_leave_smm, |
02d4160f | 6843 | .set_xcr = emulator_set_xcr, |
bbd9b64e CO |
6844 | }; |
6845 | ||
95cb2295 GN |
6846 | static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask) |
6847 | { | |
afaf0b2f | 6848 | u32 int_shadow = kvm_x86_ops.get_interrupt_shadow(vcpu); |
95cb2295 GN |
6849 | /* |
6850 | * an sti; sti; sequence only disable interrupts for the first | |
6851 | * instruction. So, if the last instruction, be it emulated or | |
6852 | * not, left the system with the INT_STI flag enabled, it | |
6853 | * means that the last instruction is an sti. We should not | |
6854 | * leave the flag on in this case. The same goes for mov ss | |
6855 | */ | |
37ccdcbe PB |
6856 | if (int_shadow & mask) |
6857 | mask = 0; | |
6addfc42 | 6858 | if (unlikely(int_shadow || mask)) { |
afaf0b2f | 6859 | kvm_x86_ops.set_interrupt_shadow(vcpu, mask); |
6addfc42 PB |
6860 | if (!mask) |
6861 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
6862 | } | |
95cb2295 GN |
6863 | } |
6864 | ||
ef54bcfe | 6865 | static bool inject_emulated_exception(struct kvm_vcpu *vcpu) |
54b8486f | 6866 | { |
c9b8b07c | 6867 | struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt; |
da9cb575 | 6868 | if (ctxt->exception.vector == PF_VECTOR) |
53b3d8e9 | 6869 | return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception); |
ef54bcfe PB |
6870 | |
6871 | if (ctxt->exception.error_code_valid) | |
da9cb575 AK |
6872 | kvm_queue_exception_e(vcpu, ctxt->exception.vector, |
6873 | ctxt->exception.error_code); | |
54b8486f | 6874 | else |
da9cb575 | 6875 | kvm_queue_exception(vcpu, ctxt->exception.vector); |
ef54bcfe | 6876 | return false; |
54b8486f GN |
6877 | } |
6878 | ||
c9b8b07c SC |
6879 | static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu) |
6880 | { | |
6881 | struct x86_emulate_ctxt *ctxt; | |
6882 | ||
6883 | ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT); | |
6884 | if (!ctxt) { | |
6885 | pr_err("kvm: failed to allocate vcpu's emulator\n"); | |
6886 | return NULL; | |
6887 | } | |
6888 | ||
6889 | ctxt->vcpu = vcpu; | |
6890 | ctxt->ops = &emulate_ops; | |
6891 | vcpu->arch.emulate_ctxt = ctxt; | |
6892 | ||
6893 | return ctxt; | |
6894 | } | |
6895 | ||
8ec4722d MG |
6896 | static void init_emulate_ctxt(struct kvm_vcpu *vcpu) |
6897 | { | |
c9b8b07c | 6898 | struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt; |
8ec4722d MG |
6899 | int cs_db, cs_l; |
6900 | ||
afaf0b2f | 6901 | kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l); |
8ec4722d | 6902 | |
744e699c | 6903 | ctxt->gpa_available = false; |
adf52235 | 6904 | ctxt->eflags = kvm_get_rflags(vcpu); |
c8401dda PB |
6905 | ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0; |
6906 | ||
adf52235 TY |
6907 | ctxt->eip = kvm_rip_read(vcpu); |
6908 | ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL : | |
6909 | (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 : | |
42bf549f | 6910 | (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 : |
adf52235 TY |
6911 | cs_db ? X86EMUL_MODE_PROT32 : |
6912 | X86EMUL_MODE_PROT16; | |
a584539b | 6913 | BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK); |
64d60670 PB |
6914 | BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK); |
6915 | BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK); | |
adf52235 | 6916 | |
dd856efa | 6917 | init_decode_cache(ctxt); |
7ae441ea | 6918 | vcpu->arch.emulate_regs_need_sync_from_vcpu = false; |
8ec4722d MG |
6919 | } |
6920 | ||
9497e1f2 | 6921 | void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip) |
63995653 | 6922 | { |
c9b8b07c | 6923 | struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt; |
63995653 MG |
6924 | int ret; |
6925 | ||
6926 | init_emulate_ctxt(vcpu); | |
6927 | ||
9dac77fa AK |
6928 | ctxt->op_bytes = 2; |
6929 | ctxt->ad_bytes = 2; | |
6930 | ctxt->_eip = ctxt->eip + inc_eip; | |
9d74191a | 6931 | ret = emulate_int_real(ctxt, irq); |
63995653 | 6932 | |
9497e1f2 SC |
6933 | if (ret != X86EMUL_CONTINUE) { |
6934 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); | |
6935 | } else { | |
6936 | ctxt->eip = ctxt->_eip; | |
6937 | kvm_rip_write(vcpu, ctxt->eip); | |
6938 | kvm_set_rflags(vcpu, ctxt->eflags); | |
6939 | } | |
63995653 MG |
6940 | } |
6941 | EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt); | |
6942 | ||
e2366171 | 6943 | static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type) |
6d77dbfc | 6944 | { |
6d77dbfc GN |
6945 | ++vcpu->stat.insn_emulation_fail; |
6946 | trace_kvm_emulate_insn_failed(vcpu); | |
e2366171 | 6947 | |
42cbf068 SC |
6948 | if (emulation_type & EMULTYPE_VMWARE_GP) { |
6949 | kvm_queue_exception_e(vcpu, GP_VECTOR, 0); | |
60fc3d02 | 6950 | return 1; |
42cbf068 | 6951 | } |
e2366171 | 6952 | |
738fece4 SC |
6953 | if (emulation_type & EMULTYPE_SKIP) { |
6954 | vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
6955 | vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
6956 | vcpu->run->internal.ndata = 0; | |
60fc3d02 | 6957 | return 0; |
738fece4 SC |
6958 | } |
6959 | ||
22da61c9 SC |
6960 | kvm_queue_exception(vcpu, UD_VECTOR); |
6961 | ||
afaf0b2f | 6962 | if (!is_guest_mode(vcpu) && kvm_x86_ops.get_cpl(vcpu) == 0) { |
fc3a9157 JR |
6963 | vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; |
6964 | vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
6965 | vcpu->run->internal.ndata = 0; | |
60fc3d02 | 6966 | return 0; |
fc3a9157 | 6967 | } |
e2366171 | 6968 | |
60fc3d02 | 6969 | return 1; |
6d77dbfc GN |
6970 | } |
6971 | ||
736c291c | 6972 | static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa, |
991eebf9 GN |
6973 | bool write_fault_to_shadow_pgtable, |
6974 | int emulation_type) | |
a6f177ef | 6975 | { |
736c291c | 6976 | gpa_t gpa = cr2_or_gpa; |
ba049e93 | 6977 | kvm_pfn_t pfn; |
a6f177ef | 6978 | |
92daa48b | 6979 | if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF)) |
991eebf9 GN |
6980 | return false; |
6981 | ||
92daa48b SC |
6982 | if (WARN_ON_ONCE(is_guest_mode(vcpu)) || |
6983 | WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF))) | |
6c3dfeb6 SC |
6984 | return false; |
6985 | ||
44dd3ffa | 6986 | if (!vcpu->arch.mmu->direct_map) { |
95b3cf69 XG |
6987 | /* |
6988 | * Write permission should be allowed since only | |
6989 | * write access need to be emulated. | |
6990 | */ | |
736c291c | 6991 | gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL); |
a6f177ef | 6992 | |
95b3cf69 XG |
6993 | /* |
6994 | * If the mapping is invalid in guest, let cpu retry | |
6995 | * it to generate fault. | |
6996 | */ | |
6997 | if (gpa == UNMAPPED_GVA) | |
6998 | return true; | |
6999 | } | |
a6f177ef | 7000 | |
8e3d9d06 XG |
7001 | /* |
7002 | * Do not retry the unhandleable instruction if it faults on the | |
7003 | * readonly host memory, otherwise it will goto a infinite loop: | |
7004 | * retry instruction -> write #PF -> emulation fail -> retry | |
7005 | * instruction -> ... | |
7006 | */ | |
7007 | pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa)); | |
95b3cf69 XG |
7008 | |
7009 | /* | |
7010 | * If the instruction failed on the error pfn, it can not be fixed, | |
7011 | * report the error to userspace. | |
7012 | */ | |
7013 | if (is_error_noslot_pfn(pfn)) | |
7014 | return false; | |
7015 | ||
7016 | kvm_release_pfn_clean(pfn); | |
7017 | ||
7018 | /* The instructions are well-emulated on direct mmu. */ | |
44dd3ffa | 7019 | if (vcpu->arch.mmu->direct_map) { |
95b3cf69 XG |
7020 | unsigned int indirect_shadow_pages; |
7021 | ||
7022 | spin_lock(&vcpu->kvm->mmu_lock); | |
7023 | indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages; | |
7024 | spin_unlock(&vcpu->kvm->mmu_lock); | |
7025 | ||
7026 | if (indirect_shadow_pages) | |
7027 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa)); | |
7028 | ||
a6f177ef | 7029 | return true; |
8e3d9d06 | 7030 | } |
a6f177ef | 7031 | |
95b3cf69 XG |
7032 | /* |
7033 | * if emulation was due to access to shadowed page table | |
7034 | * and it failed try to unshadow page and re-enter the | |
7035 | * guest to let CPU execute the instruction. | |
7036 | */ | |
7037 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa)); | |
93c05d3e XG |
7038 | |
7039 | /* | |
7040 | * If the access faults on its page table, it can not | |
7041 | * be fixed by unprotecting shadow page and it should | |
7042 | * be reported to userspace. | |
7043 | */ | |
7044 | return !write_fault_to_shadow_pgtable; | |
a6f177ef GN |
7045 | } |
7046 | ||
1cb3f3ae | 7047 | static bool retry_instruction(struct x86_emulate_ctxt *ctxt, |
736c291c | 7048 | gpa_t cr2_or_gpa, int emulation_type) |
1cb3f3ae XG |
7049 | { |
7050 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
736c291c | 7051 | unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa; |
1cb3f3ae XG |
7052 | |
7053 | last_retry_eip = vcpu->arch.last_retry_eip; | |
7054 | last_retry_addr = vcpu->arch.last_retry_addr; | |
7055 | ||
7056 | /* | |
7057 | * If the emulation is caused by #PF and it is non-page_table | |
7058 | * writing instruction, it means the VM-EXIT is caused by shadow | |
7059 | * page protected, we can zap the shadow page and retry this | |
7060 | * instruction directly. | |
7061 | * | |
7062 | * Note: if the guest uses a non-page-table modifying instruction | |
7063 | * on the PDE that points to the instruction, then we will unmap | |
7064 | * the instruction and go to an infinite loop. So, we cache the | |
7065 | * last retried eip and the last fault address, if we meet the eip | |
7066 | * and the address again, we can break out of the potential infinite | |
7067 | * loop. | |
7068 | */ | |
7069 | vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0; | |
7070 | ||
92daa48b | 7071 | if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF)) |
1cb3f3ae XG |
7072 | return false; |
7073 | ||
92daa48b SC |
7074 | if (WARN_ON_ONCE(is_guest_mode(vcpu)) || |
7075 | WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF))) | |
6c3dfeb6 SC |
7076 | return false; |
7077 | ||
1cb3f3ae XG |
7078 | if (x86_page_table_writing_insn(ctxt)) |
7079 | return false; | |
7080 | ||
736c291c | 7081 | if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa) |
1cb3f3ae XG |
7082 | return false; |
7083 | ||
7084 | vcpu->arch.last_retry_eip = ctxt->eip; | |
736c291c | 7085 | vcpu->arch.last_retry_addr = cr2_or_gpa; |
1cb3f3ae | 7086 | |
44dd3ffa | 7087 | if (!vcpu->arch.mmu->direct_map) |
736c291c | 7088 | gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL); |
1cb3f3ae | 7089 | |
22368028 | 7090 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa)); |
1cb3f3ae XG |
7091 | |
7092 | return true; | |
7093 | } | |
7094 | ||
716d51ab GN |
7095 | static int complete_emulated_mmio(struct kvm_vcpu *vcpu); |
7096 | static int complete_emulated_pio(struct kvm_vcpu *vcpu); | |
7097 | ||
64d60670 | 7098 | static void kvm_smm_changed(struct kvm_vcpu *vcpu) |
a584539b | 7099 | { |
64d60670 | 7100 | if (!(vcpu->arch.hflags & HF_SMM_MASK)) { |
660a5d51 PB |
7101 | /* This is a good place to trace that we are exiting SMM. */ |
7102 | trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false); | |
7103 | ||
c43203ca PB |
7104 | /* Process a latched INIT or SMI, if any. */ |
7105 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
64d60670 | 7106 | } |
699023e2 PB |
7107 | |
7108 | kvm_mmu_reset_context(vcpu); | |
64d60670 PB |
7109 | } |
7110 | ||
4a1e10d5 PB |
7111 | static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7, |
7112 | unsigned long *db) | |
7113 | { | |
7114 | u32 dr6 = 0; | |
7115 | int i; | |
7116 | u32 enable, rwlen; | |
7117 | ||
7118 | enable = dr7; | |
7119 | rwlen = dr7 >> 16; | |
7120 | for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4) | |
7121 | if ((enable & 3) && (rwlen & 15) == type && db[i] == addr) | |
7122 | dr6 |= (1 << i); | |
7123 | return dr6; | |
7124 | } | |
7125 | ||
120c2c4f | 7126 | static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu) |
663f4c61 PB |
7127 | { |
7128 | struct kvm_run *kvm_run = vcpu->run; | |
7129 | ||
c8401dda PB |
7130 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) { |
7131 | kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM; | |
d5d260c5 | 7132 | kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu); |
c8401dda PB |
7133 | kvm_run->debug.arch.exception = DB_VECTOR; |
7134 | kvm_run->exit_reason = KVM_EXIT_DEBUG; | |
60fc3d02 | 7135 | return 0; |
663f4c61 | 7136 | } |
120c2c4f | 7137 | kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS); |
60fc3d02 | 7138 | return 1; |
663f4c61 PB |
7139 | } |
7140 | ||
6affcbed KH |
7141 | int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu) |
7142 | { | |
afaf0b2f | 7143 | unsigned long rflags = kvm_x86_ops.get_rflags(vcpu); |
f8ea7c60 | 7144 | int r; |
6affcbed | 7145 | |
afaf0b2f | 7146 | r = kvm_x86_ops.skip_emulated_instruction(vcpu); |
60fc3d02 | 7147 | if (unlikely(!r)) |
f8ea7c60 | 7148 | return 0; |
c8401dda PB |
7149 | |
7150 | /* | |
7151 | * rflags is the old, "raw" value of the flags. The new value has | |
7152 | * not been saved yet. | |
7153 | * | |
7154 | * This is correct even for TF set by the guest, because "the | |
7155 | * processor will not generate this exception after the instruction | |
7156 | * that sets the TF flag". | |
7157 | */ | |
7158 | if (unlikely(rflags & X86_EFLAGS_TF)) | |
120c2c4f | 7159 | r = kvm_vcpu_do_singlestep(vcpu); |
60fc3d02 | 7160 | return r; |
6affcbed KH |
7161 | } |
7162 | EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction); | |
7163 | ||
4a1e10d5 PB |
7164 | static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r) |
7165 | { | |
4a1e10d5 PB |
7166 | if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) && |
7167 | (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) { | |
82b32774 NA |
7168 | struct kvm_run *kvm_run = vcpu->run; |
7169 | unsigned long eip = kvm_get_linear_rip(vcpu); | |
7170 | u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0, | |
4a1e10d5 PB |
7171 | vcpu->arch.guest_debug_dr7, |
7172 | vcpu->arch.eff_db); | |
7173 | ||
7174 | if (dr6 != 0) { | |
6f43ed01 | 7175 | kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM; |
82b32774 | 7176 | kvm_run->debug.arch.pc = eip; |
4a1e10d5 PB |
7177 | kvm_run->debug.arch.exception = DB_VECTOR; |
7178 | kvm_run->exit_reason = KVM_EXIT_DEBUG; | |
60fc3d02 | 7179 | *r = 0; |
4a1e10d5 PB |
7180 | return true; |
7181 | } | |
7182 | } | |
7183 | ||
4161a569 NA |
7184 | if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) && |
7185 | !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) { | |
82b32774 NA |
7186 | unsigned long eip = kvm_get_linear_rip(vcpu); |
7187 | u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0, | |
4a1e10d5 PB |
7188 | vcpu->arch.dr7, |
7189 | vcpu->arch.db); | |
7190 | ||
7191 | if (dr6 != 0) { | |
4d5523cf | 7192 | kvm_queue_exception_p(vcpu, DB_VECTOR, dr6); |
60fc3d02 | 7193 | *r = 1; |
4a1e10d5 PB |
7194 | return true; |
7195 | } | |
7196 | } | |
7197 | ||
7198 | return false; | |
7199 | } | |
7200 | ||
04789b66 LA |
7201 | static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt) |
7202 | { | |
2d7921c4 AM |
7203 | switch (ctxt->opcode_len) { |
7204 | case 1: | |
7205 | switch (ctxt->b) { | |
7206 | case 0xe4: /* IN */ | |
7207 | case 0xe5: | |
7208 | case 0xec: | |
7209 | case 0xed: | |
7210 | case 0xe6: /* OUT */ | |
7211 | case 0xe7: | |
7212 | case 0xee: | |
7213 | case 0xef: | |
7214 | case 0x6c: /* INS */ | |
7215 | case 0x6d: | |
7216 | case 0x6e: /* OUTS */ | |
7217 | case 0x6f: | |
7218 | return true; | |
7219 | } | |
7220 | break; | |
7221 | case 2: | |
7222 | switch (ctxt->b) { | |
7223 | case 0x33: /* RDPMC */ | |
7224 | return true; | |
7225 | } | |
7226 | break; | |
04789b66 LA |
7227 | } |
7228 | ||
7229 | return false; | |
7230 | } | |
7231 | ||
736c291c SC |
7232 | int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa, |
7233 | int emulation_type, void *insn, int insn_len) | |
bbd9b64e | 7234 | { |
95cb2295 | 7235 | int r; |
c9b8b07c | 7236 | struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt; |
7ae441ea | 7237 | bool writeback = true; |
09e3e2a1 SC |
7238 | bool write_fault_to_spt; |
7239 | ||
7240 | if (unlikely(!kvm_x86_ops.can_emulate_instruction(vcpu, insn, insn_len))) | |
7241 | return 1; | |
bbd9b64e | 7242 | |
c595ceee PB |
7243 | vcpu->arch.l1tf_flush_l1d = true; |
7244 | ||
93c05d3e XG |
7245 | /* |
7246 | * Clear write_fault_to_shadow_pgtable here to ensure it is | |
7247 | * never reused. | |
7248 | */ | |
09e3e2a1 | 7249 | write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable; |
93c05d3e | 7250 | vcpu->arch.write_fault_to_shadow_pgtable = false; |
26eef70c | 7251 | kvm_clear_exception_queue(vcpu); |
8d7d8102 | 7252 | |
571008da | 7253 | if (!(emulation_type & EMULTYPE_NO_DECODE)) { |
8ec4722d | 7254 | init_emulate_ctxt(vcpu); |
4a1e10d5 PB |
7255 | |
7256 | /* | |
7257 | * We will reenter on the same instruction since | |
7258 | * we do not set complete_userspace_io. This does not | |
7259 | * handle watchpoints yet, those would be handled in | |
7260 | * the emulate_ops. | |
7261 | */ | |
d391f120 VK |
7262 | if (!(emulation_type & EMULTYPE_SKIP) && |
7263 | kvm_vcpu_check_breakpoint(vcpu, &r)) | |
4a1e10d5 PB |
7264 | return r; |
7265 | ||
9d74191a TY |
7266 | ctxt->interruptibility = 0; |
7267 | ctxt->have_exception = false; | |
e0ad0b47 | 7268 | ctxt->exception.vector = -1; |
9d74191a | 7269 | ctxt->perm_ok = false; |
bbd9b64e | 7270 | |
b51e974f | 7271 | ctxt->ud = emulation_type & EMULTYPE_TRAP_UD; |
4005996e | 7272 | |
9d74191a | 7273 | r = x86_decode_insn(ctxt, insn, insn_len); |
bbd9b64e | 7274 | |
e46479f8 | 7275 | trace_kvm_emulate_insn_start(vcpu); |
f2b5756b | 7276 | ++vcpu->stat.insn_emulation; |
1d2887e2 | 7277 | if (r != EMULATION_OK) { |
b4000606 | 7278 | if ((emulation_type & EMULTYPE_TRAP_UD) || |
c83fad65 SC |
7279 | (emulation_type & EMULTYPE_TRAP_UD_FORCED)) { |
7280 | kvm_queue_exception(vcpu, UD_VECTOR); | |
60fc3d02 | 7281 | return 1; |
c83fad65 | 7282 | } |
736c291c SC |
7283 | if (reexecute_instruction(vcpu, cr2_or_gpa, |
7284 | write_fault_to_spt, | |
7285 | emulation_type)) | |
60fc3d02 | 7286 | return 1; |
8530a79c | 7287 | if (ctxt->have_exception) { |
c8848cee JD |
7288 | /* |
7289 | * #UD should result in just EMULATION_FAILED, and trap-like | |
7290 | * exception should not be encountered during decode. | |
7291 | */ | |
7292 | WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR || | |
7293 | exception_type(ctxt->exception.vector) == EXCPT_TRAP); | |
8530a79c | 7294 | inject_emulated_exception(vcpu); |
60fc3d02 | 7295 | return 1; |
8530a79c | 7296 | } |
e2366171 | 7297 | return handle_emulation_failure(vcpu, emulation_type); |
bbd9b64e CO |
7298 | } |
7299 | } | |
7300 | ||
42cbf068 SC |
7301 | if ((emulation_type & EMULTYPE_VMWARE_GP) && |
7302 | !is_vmware_backdoor_opcode(ctxt)) { | |
7303 | kvm_queue_exception_e(vcpu, GP_VECTOR, 0); | |
60fc3d02 | 7304 | return 1; |
42cbf068 | 7305 | } |
04789b66 | 7306 | |
1957aa63 SC |
7307 | /* |
7308 | * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks | |
7309 | * for kvm_skip_emulated_instruction(). The caller is responsible for | |
7310 | * updating interruptibility state and injecting single-step #DBs. | |
7311 | */ | |
ba8afb6b | 7312 | if (emulation_type & EMULTYPE_SKIP) { |
9dac77fa | 7313 | kvm_rip_write(vcpu, ctxt->_eip); |
bb663c7a NA |
7314 | if (ctxt->eflags & X86_EFLAGS_RF) |
7315 | kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF); | |
60fc3d02 | 7316 | return 1; |
ba8afb6b GN |
7317 | } |
7318 | ||
736c291c | 7319 | if (retry_instruction(ctxt, cr2_or_gpa, emulation_type)) |
60fc3d02 | 7320 | return 1; |
1cb3f3ae | 7321 | |
7ae441ea | 7322 | /* this is needed for vmware backdoor interface to work since it |
4d2179e1 | 7323 | changes registers values during IO operation */ |
7ae441ea GN |
7324 | if (vcpu->arch.emulate_regs_need_sync_from_vcpu) { |
7325 | vcpu->arch.emulate_regs_need_sync_from_vcpu = false; | |
dd856efa | 7326 | emulator_invalidate_register_cache(ctxt); |
7ae441ea | 7327 | } |
4d2179e1 | 7328 | |
5cd21917 | 7329 | restart: |
92daa48b SC |
7330 | if (emulation_type & EMULTYPE_PF) { |
7331 | /* Save the faulting GPA (cr2) in the address field */ | |
7332 | ctxt->exception.address = cr2_or_gpa; | |
7333 | ||
7334 | /* With shadow page tables, cr2 contains a GVA or nGPA. */ | |
7335 | if (vcpu->arch.mmu->direct_map) { | |
744e699c SC |
7336 | ctxt->gpa_available = true; |
7337 | ctxt->gpa_val = cr2_or_gpa; | |
92daa48b SC |
7338 | } |
7339 | } else { | |
7340 | /* Sanitize the address out of an abundance of paranoia. */ | |
7341 | ctxt->exception.address = 0; | |
7342 | } | |
0f89b207 | 7343 | |
9d74191a | 7344 | r = x86_emulate_insn(ctxt); |
bbd9b64e | 7345 | |
775fde86 | 7346 | if (r == EMULATION_INTERCEPTED) |
60fc3d02 | 7347 | return 1; |
775fde86 | 7348 | |
d2ddd1c4 | 7349 | if (r == EMULATION_FAILED) { |
736c291c | 7350 | if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt, |
991eebf9 | 7351 | emulation_type)) |
60fc3d02 | 7352 | return 1; |
c3cd7ffa | 7353 | |
e2366171 | 7354 | return handle_emulation_failure(vcpu, emulation_type); |
bbd9b64e CO |
7355 | } |
7356 | ||
9d74191a | 7357 | if (ctxt->have_exception) { |
60fc3d02 | 7358 | r = 1; |
ef54bcfe PB |
7359 | if (inject_emulated_exception(vcpu)) |
7360 | return r; | |
d2ddd1c4 | 7361 | } else if (vcpu->arch.pio.count) { |
0912c977 PB |
7362 | if (!vcpu->arch.pio.in) { |
7363 | /* FIXME: return into emulator if single-stepping. */ | |
3457e419 | 7364 | vcpu->arch.pio.count = 0; |
0912c977 | 7365 | } else { |
7ae441ea | 7366 | writeback = false; |
716d51ab GN |
7367 | vcpu->arch.complete_userspace_io = complete_emulated_pio; |
7368 | } | |
60fc3d02 | 7369 | r = 0; |
7ae441ea | 7370 | } else if (vcpu->mmio_needed) { |
bc8a0aaf SC |
7371 | ++vcpu->stat.mmio_exits; |
7372 | ||
7ae441ea GN |
7373 | if (!vcpu->mmio_is_write) |
7374 | writeback = false; | |
60fc3d02 | 7375 | r = 0; |
716d51ab | 7376 | vcpu->arch.complete_userspace_io = complete_emulated_mmio; |
7ae441ea | 7377 | } else if (r == EMULATION_RESTART) |
5cd21917 | 7378 | goto restart; |
d2ddd1c4 | 7379 | else |
60fc3d02 | 7380 | r = 1; |
f850e2e6 | 7381 | |
7ae441ea | 7382 | if (writeback) { |
afaf0b2f | 7383 | unsigned long rflags = kvm_x86_ops.get_rflags(vcpu); |
9d74191a | 7384 | toggle_interruptibility(vcpu, ctxt->interruptibility); |
7ae441ea | 7385 | vcpu->arch.emulate_regs_need_sync_to_vcpu = false; |
38827dbd | 7386 | if (!ctxt->have_exception || |
75ee23b3 SC |
7387 | exception_type(ctxt->exception.vector) == EXCPT_TRAP) { |
7388 | kvm_rip_write(vcpu, ctxt->eip); | |
384dea1c | 7389 | if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP))) |
120c2c4f | 7390 | r = kvm_vcpu_do_singlestep(vcpu); |
afaf0b2f SC |
7391 | if (kvm_x86_ops.update_emulated_instruction) |
7392 | kvm_x86_ops.update_emulated_instruction(vcpu); | |
38827dbd | 7393 | __kvm_set_rflags(vcpu, ctxt->eflags); |
75ee23b3 | 7394 | } |
6addfc42 PB |
7395 | |
7396 | /* | |
7397 | * For STI, interrupts are shadowed; so KVM_REQ_EVENT will | |
7398 | * do nothing, and it will be requested again as soon as | |
7399 | * the shadow expires. But we still need to check here, | |
7400 | * because POPF has no interrupt shadow. | |
7401 | */ | |
7402 | if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF)) | |
7403 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
7ae441ea GN |
7404 | } else |
7405 | vcpu->arch.emulate_regs_need_sync_to_vcpu = true; | |
e85d28f8 GN |
7406 | |
7407 | return r; | |
de7d789a | 7408 | } |
c60658d1 SC |
7409 | |
7410 | int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type) | |
7411 | { | |
7412 | return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0); | |
7413 | } | |
7414 | EXPORT_SYMBOL_GPL(kvm_emulate_instruction); | |
7415 | ||
7416 | int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu, | |
7417 | void *insn, int insn_len) | |
7418 | { | |
7419 | return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len); | |
7420 | } | |
7421 | EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer); | |
de7d789a | 7422 | |
8764ed55 SC |
7423 | static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu) |
7424 | { | |
7425 | vcpu->arch.pio.count = 0; | |
7426 | return 1; | |
7427 | } | |
7428 | ||
45def77e SC |
7429 | static int complete_fast_pio_out(struct kvm_vcpu *vcpu) |
7430 | { | |
7431 | vcpu->arch.pio.count = 0; | |
7432 | ||
7433 | if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) | |
7434 | return 1; | |
7435 | ||
7436 | return kvm_skip_emulated_instruction(vcpu); | |
7437 | } | |
7438 | ||
dca7f128 SC |
7439 | static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, |
7440 | unsigned short port) | |
de7d789a | 7441 | { |
de3cd117 | 7442 | unsigned long val = kvm_rax_read(vcpu); |
2e3bb4d8 SC |
7443 | int ret = emulator_pio_out(vcpu, size, port, &val, 1); |
7444 | ||
8764ed55 SC |
7445 | if (ret) |
7446 | return ret; | |
45def77e | 7447 | |
8764ed55 SC |
7448 | /* |
7449 | * Workaround userspace that relies on old KVM behavior of %rip being | |
7450 | * incremented prior to exiting to userspace to handle "OUT 0x7e". | |
7451 | */ | |
7452 | if (port == 0x7e && | |
7453 | kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) { | |
7454 | vcpu->arch.complete_userspace_io = | |
7455 | complete_fast_pio_out_port_0x7e; | |
7456 | kvm_skip_emulated_instruction(vcpu); | |
7457 | } else { | |
45def77e SC |
7458 | vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu); |
7459 | vcpu->arch.complete_userspace_io = complete_fast_pio_out; | |
7460 | } | |
8764ed55 | 7461 | return 0; |
de7d789a | 7462 | } |
de7d789a | 7463 | |
8370c3d0 TL |
7464 | static int complete_fast_pio_in(struct kvm_vcpu *vcpu) |
7465 | { | |
7466 | unsigned long val; | |
7467 | ||
7468 | /* We should only ever be called with arch.pio.count equal to 1 */ | |
7469 | BUG_ON(vcpu->arch.pio.count != 1); | |
7470 | ||
45def77e SC |
7471 | if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) { |
7472 | vcpu->arch.pio.count = 0; | |
7473 | return 1; | |
7474 | } | |
7475 | ||
8370c3d0 | 7476 | /* For size less than 4 we merge, else we zero extend */ |
de3cd117 | 7477 | val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0; |
8370c3d0 TL |
7478 | |
7479 | /* | |
2e3bb4d8 | 7480 | * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform |
8370c3d0 TL |
7481 | * the copy and tracing |
7482 | */ | |
2e3bb4d8 | 7483 | emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1); |
de3cd117 | 7484 | kvm_rax_write(vcpu, val); |
8370c3d0 | 7485 | |
45def77e | 7486 | return kvm_skip_emulated_instruction(vcpu); |
8370c3d0 TL |
7487 | } |
7488 | ||
dca7f128 SC |
7489 | static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size, |
7490 | unsigned short port) | |
8370c3d0 TL |
7491 | { |
7492 | unsigned long val; | |
7493 | int ret; | |
7494 | ||
7495 | /* For size less than 4 we merge, else we zero extend */ | |
de3cd117 | 7496 | val = (size < 4) ? kvm_rax_read(vcpu) : 0; |
8370c3d0 | 7497 | |
2e3bb4d8 | 7498 | ret = emulator_pio_in(vcpu, size, port, &val, 1); |
8370c3d0 | 7499 | if (ret) { |
de3cd117 | 7500 | kvm_rax_write(vcpu, val); |
8370c3d0 TL |
7501 | return ret; |
7502 | } | |
7503 | ||
45def77e | 7504 | vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu); |
8370c3d0 TL |
7505 | vcpu->arch.complete_userspace_io = complete_fast_pio_in; |
7506 | ||
7507 | return 0; | |
7508 | } | |
dca7f128 SC |
7509 | |
7510 | int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in) | |
7511 | { | |
45def77e | 7512 | int ret; |
dca7f128 | 7513 | |
dca7f128 | 7514 | if (in) |
45def77e | 7515 | ret = kvm_fast_pio_in(vcpu, size, port); |
dca7f128 | 7516 | else |
45def77e SC |
7517 | ret = kvm_fast_pio_out(vcpu, size, port); |
7518 | return ret && kvm_skip_emulated_instruction(vcpu); | |
dca7f128 SC |
7519 | } |
7520 | EXPORT_SYMBOL_GPL(kvm_fast_pio); | |
8370c3d0 | 7521 | |
251a5fd6 | 7522 | static int kvmclock_cpu_down_prep(unsigned int cpu) |
8cfdc000 | 7523 | { |
0a3aee0d | 7524 | __this_cpu_write(cpu_tsc_khz, 0); |
251a5fd6 | 7525 | return 0; |
8cfdc000 ZA |
7526 | } |
7527 | ||
7528 | static void tsc_khz_changed(void *data) | |
c8076604 | 7529 | { |
8cfdc000 ZA |
7530 | struct cpufreq_freqs *freq = data; |
7531 | unsigned long khz = 0; | |
7532 | ||
7533 | if (data) | |
7534 | khz = freq->new; | |
7535 | else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) | |
7536 | khz = cpufreq_quick_get(raw_smp_processor_id()); | |
7537 | if (!khz) | |
7538 | khz = tsc_khz; | |
0a3aee0d | 7539 | __this_cpu_write(cpu_tsc_khz, khz); |
c8076604 GH |
7540 | } |
7541 | ||
5fa4ec9c | 7542 | #ifdef CONFIG_X86_64 |
0092e434 VK |
7543 | static void kvm_hyperv_tsc_notifier(void) |
7544 | { | |
0092e434 VK |
7545 | struct kvm *kvm; |
7546 | struct kvm_vcpu *vcpu; | |
7547 | int cpu; | |
7548 | ||
0d9ce162 | 7549 | mutex_lock(&kvm_lock); |
0092e434 VK |
7550 | list_for_each_entry(kvm, &vm_list, vm_list) |
7551 | kvm_make_mclock_inprogress_request(kvm); | |
7552 | ||
7553 | hyperv_stop_tsc_emulation(); | |
7554 | ||
7555 | /* TSC frequency always matches when on Hyper-V */ | |
7556 | for_each_present_cpu(cpu) | |
7557 | per_cpu(cpu_tsc_khz, cpu) = tsc_khz; | |
7558 | kvm_max_guest_tsc_khz = tsc_khz; | |
7559 | ||
7560 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
7561 | struct kvm_arch *ka = &kvm->arch; | |
7562 | ||
7563 | spin_lock(&ka->pvclock_gtod_sync_lock); | |
7564 | ||
7565 | pvclock_update_vm_gtod_copy(kvm); | |
7566 | ||
7567 | kvm_for_each_vcpu(cpu, vcpu, kvm) | |
7568 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); | |
7569 | ||
7570 | kvm_for_each_vcpu(cpu, vcpu, kvm) | |
7571 | kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu); | |
7572 | ||
7573 | spin_unlock(&ka->pvclock_gtod_sync_lock); | |
7574 | } | |
0d9ce162 | 7575 | mutex_unlock(&kvm_lock); |
0092e434 | 7576 | } |
5fa4ec9c | 7577 | #endif |
0092e434 | 7578 | |
df24014a | 7579 | static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu) |
c8076604 | 7580 | { |
c8076604 GH |
7581 | struct kvm *kvm; |
7582 | struct kvm_vcpu *vcpu; | |
7583 | int i, send_ipi = 0; | |
7584 | ||
8cfdc000 ZA |
7585 | /* |
7586 | * We allow guests to temporarily run on slowing clocks, | |
7587 | * provided we notify them after, or to run on accelerating | |
7588 | * clocks, provided we notify them before. Thus time never | |
7589 | * goes backwards. | |
7590 | * | |
7591 | * However, we have a problem. We can't atomically update | |
7592 | * the frequency of a given CPU from this function; it is | |
7593 | * merely a notifier, which can be called from any CPU. | |
7594 | * Changing the TSC frequency at arbitrary points in time | |
7595 | * requires a recomputation of local variables related to | |
7596 | * the TSC for each VCPU. We must flag these local variables | |
7597 | * to be updated and be sure the update takes place with the | |
7598 | * new frequency before any guests proceed. | |
7599 | * | |
7600 | * Unfortunately, the combination of hotplug CPU and frequency | |
7601 | * change creates an intractable locking scenario; the order | |
7602 | * of when these callouts happen is undefined with respect to | |
7603 | * CPU hotplug, and they can race with each other. As such, | |
7604 | * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is | |
7605 | * undefined; you can actually have a CPU frequency change take | |
7606 | * place in between the computation of X and the setting of the | |
7607 | * variable. To protect against this problem, all updates of | |
7608 | * the per_cpu tsc_khz variable are done in an interrupt | |
7609 | * protected IPI, and all callers wishing to update the value | |
7610 | * must wait for a synchronous IPI to complete (which is trivial | |
7611 | * if the caller is on the CPU already). This establishes the | |
7612 | * necessary total order on variable updates. | |
7613 | * | |
7614 | * Note that because a guest time update may take place | |
7615 | * anytime after the setting of the VCPU's request bit, the | |
7616 | * correct TSC value must be set before the request. However, | |
7617 | * to ensure the update actually makes it to any guest which | |
7618 | * starts running in hardware virtualization between the set | |
7619 | * and the acquisition of the spinlock, we must also ping the | |
7620 | * CPU after setting the request bit. | |
7621 | * | |
7622 | */ | |
7623 | ||
df24014a | 7624 | smp_call_function_single(cpu, tsc_khz_changed, freq, 1); |
c8076604 | 7625 | |
0d9ce162 | 7626 | mutex_lock(&kvm_lock); |
c8076604 | 7627 | list_for_each_entry(kvm, &vm_list, vm_list) { |
988a2cae | 7628 | kvm_for_each_vcpu(i, vcpu, kvm) { |
df24014a | 7629 | if (vcpu->cpu != cpu) |
c8076604 | 7630 | continue; |
c285545f | 7631 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
0d9ce162 | 7632 | if (vcpu->cpu != raw_smp_processor_id()) |
8cfdc000 | 7633 | send_ipi = 1; |
c8076604 GH |
7634 | } |
7635 | } | |
0d9ce162 | 7636 | mutex_unlock(&kvm_lock); |
c8076604 GH |
7637 | |
7638 | if (freq->old < freq->new && send_ipi) { | |
7639 | /* | |
7640 | * We upscale the frequency. Must make the guest | |
7641 | * doesn't see old kvmclock values while running with | |
7642 | * the new frequency, otherwise we risk the guest sees | |
7643 | * time go backwards. | |
7644 | * | |
7645 | * In case we update the frequency for another cpu | |
7646 | * (which might be in guest context) send an interrupt | |
7647 | * to kick the cpu out of guest context. Next time | |
7648 | * guest context is entered kvmclock will be updated, | |
7649 | * so the guest will not see stale values. | |
7650 | */ | |
df24014a | 7651 | smp_call_function_single(cpu, tsc_khz_changed, freq, 1); |
c8076604 | 7652 | } |
df24014a VK |
7653 | } |
7654 | ||
7655 | static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val, | |
7656 | void *data) | |
7657 | { | |
7658 | struct cpufreq_freqs *freq = data; | |
7659 | int cpu; | |
7660 | ||
7661 | if (val == CPUFREQ_PRECHANGE && freq->old > freq->new) | |
7662 | return 0; | |
7663 | if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new) | |
7664 | return 0; | |
7665 | ||
7666 | for_each_cpu(cpu, freq->policy->cpus) | |
7667 | __kvmclock_cpufreq_notifier(freq, cpu); | |
7668 | ||
c8076604 GH |
7669 | return 0; |
7670 | } | |
7671 | ||
7672 | static struct notifier_block kvmclock_cpufreq_notifier_block = { | |
8cfdc000 ZA |
7673 | .notifier_call = kvmclock_cpufreq_notifier |
7674 | }; | |
7675 | ||
251a5fd6 | 7676 | static int kvmclock_cpu_online(unsigned int cpu) |
8cfdc000 | 7677 | { |
251a5fd6 SAS |
7678 | tsc_khz_changed(NULL); |
7679 | return 0; | |
8cfdc000 ZA |
7680 | } |
7681 | ||
b820cc0c ZA |
7682 | static void kvm_timer_init(void) |
7683 | { | |
c285545f | 7684 | max_tsc_khz = tsc_khz; |
460dd42e | 7685 | |
b820cc0c | 7686 | if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) { |
c285545f | 7687 | #ifdef CONFIG_CPU_FREQ |
aaec7c03 | 7688 | struct cpufreq_policy *policy; |
758f588d BP |
7689 | int cpu; |
7690 | ||
3e26f230 | 7691 | cpu = get_cpu(); |
aaec7c03 | 7692 | policy = cpufreq_cpu_get(cpu); |
9a11997e WL |
7693 | if (policy) { |
7694 | if (policy->cpuinfo.max_freq) | |
7695 | max_tsc_khz = policy->cpuinfo.max_freq; | |
7696 | cpufreq_cpu_put(policy); | |
7697 | } | |
3e26f230 | 7698 | put_cpu(); |
c285545f | 7699 | #endif |
b820cc0c ZA |
7700 | cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block, |
7701 | CPUFREQ_TRANSITION_NOTIFIER); | |
7702 | } | |
460dd42e | 7703 | |
73c1b41e | 7704 | cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online", |
251a5fd6 | 7705 | kvmclock_cpu_online, kvmclock_cpu_down_prep); |
b820cc0c ZA |
7706 | } |
7707 | ||
dd60d217 AK |
7708 | DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu); |
7709 | EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu); | |
ff9d07a0 | 7710 | |
f5132b01 | 7711 | int kvm_is_in_guest(void) |
ff9d07a0 | 7712 | { |
086c9855 | 7713 | return __this_cpu_read(current_vcpu) != NULL; |
ff9d07a0 ZY |
7714 | } |
7715 | ||
7716 | static int kvm_is_user_mode(void) | |
7717 | { | |
7718 | int user_mode = 3; | |
dcf46b94 | 7719 | |
086c9855 | 7720 | if (__this_cpu_read(current_vcpu)) |
afaf0b2f | 7721 | user_mode = kvm_x86_ops.get_cpl(__this_cpu_read(current_vcpu)); |
dcf46b94 | 7722 | |
ff9d07a0 ZY |
7723 | return user_mode != 0; |
7724 | } | |
7725 | ||
7726 | static unsigned long kvm_get_guest_ip(void) | |
7727 | { | |
7728 | unsigned long ip = 0; | |
dcf46b94 | 7729 | |
086c9855 AS |
7730 | if (__this_cpu_read(current_vcpu)) |
7731 | ip = kvm_rip_read(__this_cpu_read(current_vcpu)); | |
dcf46b94 | 7732 | |
ff9d07a0 ZY |
7733 | return ip; |
7734 | } | |
7735 | ||
8479e04e LK |
7736 | static void kvm_handle_intel_pt_intr(void) |
7737 | { | |
7738 | struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu); | |
7739 | ||
7740 | kvm_make_request(KVM_REQ_PMI, vcpu); | |
7741 | __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT, | |
7742 | (unsigned long *)&vcpu->arch.pmu.global_status); | |
7743 | } | |
7744 | ||
ff9d07a0 ZY |
7745 | static struct perf_guest_info_callbacks kvm_guest_cbs = { |
7746 | .is_in_guest = kvm_is_in_guest, | |
7747 | .is_user_mode = kvm_is_user_mode, | |
7748 | .get_guest_ip = kvm_get_guest_ip, | |
8479e04e | 7749 | .handle_intel_pt_intr = kvm_handle_intel_pt_intr, |
ff9d07a0 ZY |
7750 | }; |
7751 | ||
16e8d74d MT |
7752 | #ifdef CONFIG_X86_64 |
7753 | static void pvclock_gtod_update_fn(struct work_struct *work) | |
7754 | { | |
d828199e MT |
7755 | struct kvm *kvm; |
7756 | ||
7757 | struct kvm_vcpu *vcpu; | |
7758 | int i; | |
7759 | ||
0d9ce162 | 7760 | mutex_lock(&kvm_lock); |
d828199e MT |
7761 | list_for_each_entry(kvm, &vm_list, vm_list) |
7762 | kvm_for_each_vcpu(i, vcpu, kvm) | |
105b21bb | 7763 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); |
d828199e | 7764 | atomic_set(&kvm_guest_has_master_clock, 0); |
0d9ce162 | 7765 | mutex_unlock(&kvm_lock); |
16e8d74d MT |
7766 | } |
7767 | ||
7768 | static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn); | |
7769 | ||
7770 | /* | |
7771 | * Notification about pvclock gtod data update. | |
7772 | */ | |
7773 | static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused, | |
7774 | void *priv) | |
7775 | { | |
7776 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; | |
7777 | struct timekeeper *tk = priv; | |
7778 | ||
7779 | update_pvclock_gtod(tk); | |
7780 | ||
7781 | /* disable master clock if host does not trust, or does not | |
b0c39dc6 | 7782 | * use, TSC based clocksource. |
16e8d74d | 7783 | */ |
b0c39dc6 | 7784 | if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) && |
16e8d74d MT |
7785 | atomic_read(&kvm_guest_has_master_clock) != 0) |
7786 | queue_work(system_long_wq, &pvclock_gtod_work); | |
7787 | ||
7788 | return 0; | |
7789 | } | |
7790 | ||
7791 | static struct notifier_block pvclock_gtod_notifier = { | |
7792 | .notifier_call = pvclock_gtod_notify, | |
7793 | }; | |
7794 | #endif | |
7795 | ||
f8c16bba | 7796 | int kvm_arch_init(void *opaque) |
043405e1 | 7797 | { |
d008dfdb | 7798 | struct kvm_x86_init_ops *ops = opaque; |
b820cc0c | 7799 | int r; |
f8c16bba | 7800 | |
afaf0b2f | 7801 | if (kvm_x86_ops.hardware_enable) { |
f8c16bba | 7802 | printk(KERN_ERR "kvm: already loaded the other module\n"); |
56c6d28a ZX |
7803 | r = -EEXIST; |
7804 | goto out; | |
f8c16bba ZX |
7805 | } |
7806 | ||
7807 | if (!ops->cpu_has_kvm_support()) { | |
ef935c25 | 7808 | pr_err_ratelimited("kvm: no hardware support\n"); |
56c6d28a ZX |
7809 | r = -EOPNOTSUPP; |
7810 | goto out; | |
f8c16bba ZX |
7811 | } |
7812 | if (ops->disabled_by_bios()) { | |
ef935c25 | 7813 | pr_err_ratelimited("kvm: disabled by bios\n"); |
56c6d28a ZX |
7814 | r = -EOPNOTSUPP; |
7815 | goto out; | |
f8c16bba ZX |
7816 | } |
7817 | ||
b666a4b6 MO |
7818 | /* |
7819 | * KVM explicitly assumes that the guest has an FPU and | |
7820 | * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the | |
7821 | * vCPU's FPU state as a fxregs_state struct. | |
7822 | */ | |
7823 | if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) { | |
7824 | printk(KERN_ERR "kvm: inadequate fpu\n"); | |
7825 | r = -EOPNOTSUPP; | |
7826 | goto out; | |
7827 | } | |
7828 | ||
013f6a5d | 7829 | r = -ENOMEM; |
ed8e4812 | 7830 | x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu), |
b666a4b6 MO |
7831 | __alignof__(struct fpu), SLAB_ACCOUNT, |
7832 | NULL); | |
7833 | if (!x86_fpu_cache) { | |
7834 | printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n"); | |
7835 | goto out; | |
7836 | } | |
7837 | ||
c9b8b07c SC |
7838 | x86_emulator_cache = kvm_alloc_emulator_cache(); |
7839 | if (!x86_emulator_cache) { | |
7840 | pr_err("kvm: failed to allocate cache for x86 emulator\n"); | |
7841 | goto out_free_x86_fpu_cache; | |
7842 | } | |
7843 | ||
7e34fbd0 SC |
7844 | user_return_msrs = alloc_percpu(struct kvm_user_return_msrs); |
7845 | if (!user_return_msrs) { | |
7846 | printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n"); | |
c9b8b07c | 7847 | goto out_free_x86_emulator_cache; |
013f6a5d MT |
7848 | } |
7849 | ||
97db56ce AK |
7850 | r = kvm_mmu_module_init(); |
7851 | if (r) | |
013f6a5d | 7852 | goto out_free_percpu; |
97db56ce | 7853 | |
7b52345e | 7854 | kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK, |
ffb128c8 | 7855 | PT_DIRTY_MASK, PT64_NX_MASK, 0, |
d0ec49d4 | 7856 | PT_PRESENT_MASK, 0, sme_me_mask); |
b820cc0c | 7857 | kvm_timer_init(); |
c8076604 | 7858 | |
ff9d07a0 ZY |
7859 | perf_register_guest_info_callbacks(&kvm_guest_cbs); |
7860 | ||
cfc48181 | 7861 | if (boot_cpu_has(X86_FEATURE_XSAVE)) { |
2acf923e | 7862 | host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK); |
cfc48181 SC |
7863 | supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0; |
7864 | } | |
2acf923e | 7865 | |
c5cc421b | 7866 | kvm_lapic_init(); |
0c5f81da WL |
7867 | if (pi_inject_timer == -1) |
7868 | pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER); | |
16e8d74d MT |
7869 | #ifdef CONFIG_X86_64 |
7870 | pvclock_gtod_register_notifier(&pvclock_gtod_notifier); | |
0092e434 | 7871 | |
5fa4ec9c | 7872 | if (hypervisor_is_type(X86_HYPER_MS_HYPERV)) |
0092e434 | 7873 | set_hv_tscchange_cb(kvm_hyperv_tsc_notifier); |
16e8d74d MT |
7874 | #endif |
7875 | ||
f8c16bba | 7876 | return 0; |
56c6d28a | 7877 | |
013f6a5d | 7878 | out_free_percpu: |
7e34fbd0 | 7879 | free_percpu(user_return_msrs); |
c9b8b07c SC |
7880 | out_free_x86_emulator_cache: |
7881 | kmem_cache_destroy(x86_emulator_cache); | |
b666a4b6 MO |
7882 | out_free_x86_fpu_cache: |
7883 | kmem_cache_destroy(x86_fpu_cache); | |
56c6d28a | 7884 | out: |
56c6d28a | 7885 | return r; |
043405e1 | 7886 | } |
8776e519 | 7887 | |
f8c16bba ZX |
7888 | void kvm_arch_exit(void) |
7889 | { | |
0092e434 | 7890 | #ifdef CONFIG_X86_64 |
5fa4ec9c | 7891 | if (hypervisor_is_type(X86_HYPER_MS_HYPERV)) |
0092e434 VK |
7892 | clear_hv_tscchange_cb(); |
7893 | #endif | |
cef84c30 | 7894 | kvm_lapic_exit(); |
ff9d07a0 ZY |
7895 | perf_unregister_guest_info_callbacks(&kvm_guest_cbs); |
7896 | ||
888d256e JK |
7897 | if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) |
7898 | cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block, | |
7899 | CPUFREQ_TRANSITION_NOTIFIER); | |
251a5fd6 | 7900 | cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE); |
16e8d74d MT |
7901 | #ifdef CONFIG_X86_64 |
7902 | pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier); | |
7903 | #endif | |
afaf0b2f | 7904 | kvm_x86_ops.hardware_enable = NULL; |
56c6d28a | 7905 | kvm_mmu_module_exit(); |
7e34fbd0 | 7906 | free_percpu(user_return_msrs); |
b666a4b6 | 7907 | kmem_cache_destroy(x86_fpu_cache); |
56c6d28a | 7908 | } |
f8c16bba | 7909 | |
5cb56059 | 7910 | int kvm_vcpu_halt(struct kvm_vcpu *vcpu) |
8776e519 HB |
7911 | { |
7912 | ++vcpu->stat.halt_exits; | |
35754c98 | 7913 | if (lapic_in_kernel(vcpu)) { |
a4535290 | 7914 | vcpu->arch.mp_state = KVM_MP_STATE_HALTED; |
8776e519 HB |
7915 | return 1; |
7916 | } else { | |
7917 | vcpu->run->exit_reason = KVM_EXIT_HLT; | |
7918 | return 0; | |
7919 | } | |
7920 | } | |
5cb56059 JS |
7921 | EXPORT_SYMBOL_GPL(kvm_vcpu_halt); |
7922 | ||
7923 | int kvm_emulate_halt(struct kvm_vcpu *vcpu) | |
7924 | { | |
6affcbed KH |
7925 | int ret = kvm_skip_emulated_instruction(vcpu); |
7926 | /* | |
7927 | * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered | |
7928 | * KVM_EXIT_DEBUG here. | |
7929 | */ | |
7930 | return kvm_vcpu_halt(vcpu) && ret; | |
5cb56059 | 7931 | } |
8776e519 HB |
7932 | EXPORT_SYMBOL_GPL(kvm_emulate_halt); |
7933 | ||
8ef81a9a | 7934 | #ifdef CONFIG_X86_64 |
55dd00a7 MT |
7935 | static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr, |
7936 | unsigned long clock_type) | |
7937 | { | |
7938 | struct kvm_clock_pairing clock_pairing; | |
899a31f5 | 7939 | struct timespec64 ts; |
80fbd89c | 7940 | u64 cycle; |
55dd00a7 MT |
7941 | int ret; |
7942 | ||
7943 | if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK) | |
7944 | return -KVM_EOPNOTSUPP; | |
7945 | ||
7946 | if (kvm_get_walltime_and_clockread(&ts, &cycle) == false) | |
7947 | return -KVM_EOPNOTSUPP; | |
7948 | ||
7949 | clock_pairing.sec = ts.tv_sec; | |
7950 | clock_pairing.nsec = ts.tv_nsec; | |
7951 | clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle); | |
7952 | clock_pairing.flags = 0; | |
bcbfbd8e | 7953 | memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad)); |
55dd00a7 MT |
7954 | |
7955 | ret = 0; | |
7956 | if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing, | |
7957 | sizeof(struct kvm_clock_pairing))) | |
7958 | ret = -KVM_EFAULT; | |
7959 | ||
7960 | return ret; | |
7961 | } | |
8ef81a9a | 7962 | #endif |
55dd00a7 | 7963 | |
6aef266c SV |
7964 | /* |
7965 | * kvm_pv_kick_cpu_op: Kick a vcpu. | |
7966 | * | |
7967 | * @apicid - apicid of vcpu to be kicked. | |
7968 | */ | |
7969 | static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid) | |
7970 | { | |
24d2166b | 7971 | struct kvm_lapic_irq lapic_irq; |
6aef266c | 7972 | |
150a84fe | 7973 | lapic_irq.shorthand = APIC_DEST_NOSHORT; |
c96001c5 | 7974 | lapic_irq.dest_mode = APIC_DEST_PHYSICAL; |
ebd28fcb | 7975 | lapic_irq.level = 0; |
24d2166b | 7976 | lapic_irq.dest_id = apicid; |
93bbf0b8 | 7977 | lapic_irq.msi_redir_hint = false; |
6aef266c | 7978 | |
24d2166b | 7979 | lapic_irq.delivery_mode = APIC_DM_REMRD; |
795a149e | 7980 | kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL); |
6aef266c SV |
7981 | } |
7982 | ||
4e19c36f SS |
7983 | bool kvm_apicv_activated(struct kvm *kvm) |
7984 | { | |
7985 | return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0); | |
7986 | } | |
7987 | EXPORT_SYMBOL_GPL(kvm_apicv_activated); | |
7988 | ||
7989 | void kvm_apicv_init(struct kvm *kvm, bool enable) | |
7990 | { | |
7991 | if (enable) | |
7992 | clear_bit(APICV_INHIBIT_REASON_DISABLE, | |
7993 | &kvm->arch.apicv_inhibit_reasons); | |
7994 | else | |
7995 | set_bit(APICV_INHIBIT_REASON_DISABLE, | |
7996 | &kvm->arch.apicv_inhibit_reasons); | |
7997 | } | |
7998 | EXPORT_SYMBOL_GPL(kvm_apicv_init); | |
7999 | ||
71506297 WL |
8000 | static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id) |
8001 | { | |
8002 | struct kvm_vcpu *target = NULL; | |
8003 | struct kvm_apic_map *map; | |
8004 | ||
8005 | rcu_read_lock(); | |
8006 | map = rcu_dereference(kvm->arch.apic_map); | |
8007 | ||
8008 | if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id]) | |
8009 | target = map->phys_map[dest_id]->vcpu; | |
8010 | ||
8011 | rcu_read_unlock(); | |
8012 | ||
266e85a5 | 8013 | if (target && READ_ONCE(target->ready)) |
71506297 WL |
8014 | kvm_vcpu_yield_to(target); |
8015 | } | |
8016 | ||
8776e519 HB |
8017 | int kvm_emulate_hypercall(struct kvm_vcpu *vcpu) |
8018 | { | |
8019 | unsigned long nr, a0, a1, a2, a3, ret; | |
6356ee0c | 8020 | int op_64_bit; |
8776e519 | 8021 | |
696ca779 RK |
8022 | if (kvm_hv_hypercall_enabled(vcpu->kvm)) |
8023 | return kvm_hv_hypercall(vcpu); | |
55cd8e5a | 8024 | |
de3cd117 SC |
8025 | nr = kvm_rax_read(vcpu); |
8026 | a0 = kvm_rbx_read(vcpu); | |
8027 | a1 = kvm_rcx_read(vcpu); | |
8028 | a2 = kvm_rdx_read(vcpu); | |
8029 | a3 = kvm_rsi_read(vcpu); | |
8776e519 | 8030 | |
229456fc | 8031 | trace_kvm_hypercall(nr, a0, a1, a2, a3); |
2714d1d3 | 8032 | |
a449c7aa NA |
8033 | op_64_bit = is_64_bit_mode(vcpu); |
8034 | if (!op_64_bit) { | |
8776e519 HB |
8035 | nr &= 0xFFFFFFFF; |
8036 | a0 &= 0xFFFFFFFF; | |
8037 | a1 &= 0xFFFFFFFF; | |
8038 | a2 &= 0xFFFFFFFF; | |
8039 | a3 &= 0xFFFFFFFF; | |
8040 | } | |
8041 | ||
afaf0b2f | 8042 | if (kvm_x86_ops.get_cpl(vcpu) != 0) { |
07708c4a | 8043 | ret = -KVM_EPERM; |
696ca779 | 8044 | goto out; |
07708c4a JK |
8045 | } |
8046 | ||
66570e96 OU |
8047 | ret = -KVM_ENOSYS; |
8048 | ||
8776e519 | 8049 | switch (nr) { |
b93463aa AK |
8050 | case KVM_HC_VAPIC_POLL_IRQ: |
8051 | ret = 0; | |
8052 | break; | |
6aef266c | 8053 | case KVM_HC_KICK_CPU: |
66570e96 OU |
8054 | if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT)) |
8055 | break; | |
8056 | ||
6aef266c | 8057 | kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1); |
266e85a5 | 8058 | kvm_sched_yield(vcpu->kvm, a1); |
6aef266c SV |
8059 | ret = 0; |
8060 | break; | |
8ef81a9a | 8061 | #ifdef CONFIG_X86_64 |
55dd00a7 MT |
8062 | case KVM_HC_CLOCK_PAIRING: |
8063 | ret = kvm_pv_clock_pairing(vcpu, a0, a1); | |
8064 | break; | |
1ed199a4 | 8065 | #endif |
4180bf1b | 8066 | case KVM_HC_SEND_IPI: |
66570e96 OU |
8067 | if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI)) |
8068 | break; | |
8069 | ||
4180bf1b WL |
8070 | ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit); |
8071 | break; | |
71506297 | 8072 | case KVM_HC_SCHED_YIELD: |
66570e96 OU |
8073 | if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD)) |
8074 | break; | |
8075 | ||
71506297 WL |
8076 | kvm_sched_yield(vcpu->kvm, a0); |
8077 | ret = 0; | |
8078 | break; | |
8776e519 HB |
8079 | default: |
8080 | ret = -KVM_ENOSYS; | |
8081 | break; | |
8082 | } | |
696ca779 | 8083 | out: |
a449c7aa NA |
8084 | if (!op_64_bit) |
8085 | ret = (u32)ret; | |
de3cd117 | 8086 | kvm_rax_write(vcpu, ret); |
6356ee0c | 8087 | |
f11c3a8d | 8088 | ++vcpu->stat.hypercalls; |
6356ee0c | 8089 | return kvm_skip_emulated_instruction(vcpu); |
8776e519 HB |
8090 | } |
8091 | EXPORT_SYMBOL_GPL(kvm_emulate_hypercall); | |
8092 | ||
b6785def | 8093 | static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt) |
8776e519 | 8094 | { |
d6aa1000 | 8095 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
8776e519 | 8096 | char instruction[3]; |
5fdbf976 | 8097 | unsigned long rip = kvm_rip_read(vcpu); |
8776e519 | 8098 | |
afaf0b2f | 8099 | kvm_x86_ops.patch_hypercall(vcpu, instruction); |
8776e519 | 8100 | |
ce2e852e DV |
8101 | return emulator_write_emulated(ctxt, rip, instruction, 3, |
8102 | &ctxt->exception); | |
8776e519 HB |
8103 | } |
8104 | ||
851ba692 | 8105 | static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu) |
b6c7a5dc | 8106 | { |
782d422b MG |
8107 | return vcpu->run->request_interrupt_window && |
8108 | likely(!pic_in_kernel(vcpu->kvm)); | |
b6c7a5dc HB |
8109 | } |
8110 | ||
851ba692 | 8111 | static void post_kvm_run_save(struct kvm_vcpu *vcpu) |
b6c7a5dc | 8112 | { |
851ba692 AK |
8113 | struct kvm_run *kvm_run = vcpu->run; |
8114 | ||
91586a3b | 8115 | kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0; |
f077825a | 8116 | kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0; |
2d3ad1f4 | 8117 | kvm_run->cr8 = kvm_get_cr8(vcpu); |
b6c7a5dc | 8118 | kvm_run->apic_base = kvm_get_apic_base(vcpu); |
127a457a MG |
8119 | kvm_run->ready_for_interrupt_injection = |
8120 | pic_in_kernel(vcpu->kvm) || | |
782d422b | 8121 | kvm_vcpu_ready_for_interrupt_injection(vcpu); |
b6c7a5dc HB |
8122 | } |
8123 | ||
95ba8273 GN |
8124 | static void update_cr8_intercept(struct kvm_vcpu *vcpu) |
8125 | { | |
8126 | int max_irr, tpr; | |
8127 | ||
afaf0b2f | 8128 | if (!kvm_x86_ops.update_cr8_intercept) |
95ba8273 GN |
8129 | return; |
8130 | ||
bce87cce | 8131 | if (!lapic_in_kernel(vcpu)) |
88c808fd AK |
8132 | return; |
8133 | ||
d62caabb AS |
8134 | if (vcpu->arch.apicv_active) |
8135 | return; | |
8136 | ||
8db3baa2 GN |
8137 | if (!vcpu->arch.apic->vapic_addr) |
8138 | max_irr = kvm_lapic_find_highest_irr(vcpu); | |
8139 | else | |
8140 | max_irr = -1; | |
95ba8273 GN |
8141 | |
8142 | if (max_irr != -1) | |
8143 | max_irr >>= 4; | |
8144 | ||
8145 | tpr = kvm_lapic_get_cr8(vcpu); | |
8146 | ||
afaf0b2f | 8147 | kvm_x86_ops.update_cr8_intercept(vcpu, tpr, max_irr); |
95ba8273 GN |
8148 | } |
8149 | ||
c9d40913 | 8150 | static void inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit) |
95ba8273 | 8151 | { |
b6b8a145 | 8152 | int r; |
c6b22f59 | 8153 | bool can_inject = true; |
b6b8a145 | 8154 | |
95ba8273 | 8155 | /* try to reinject previous events if any */ |
664f8e26 | 8156 | |
c6b22f59 | 8157 | if (vcpu->arch.exception.injected) { |
afaf0b2f | 8158 | kvm_x86_ops.queue_exception(vcpu); |
c6b22f59 PB |
8159 | can_inject = false; |
8160 | } | |
664f8e26 | 8161 | /* |
a042c26f LA |
8162 | * Do not inject an NMI or interrupt if there is a pending |
8163 | * exception. Exceptions and interrupts are recognized at | |
8164 | * instruction boundaries, i.e. the start of an instruction. | |
8165 | * Trap-like exceptions, e.g. #DB, have higher priority than | |
8166 | * NMIs and interrupts, i.e. traps are recognized before an | |
8167 | * NMI/interrupt that's pending on the same instruction. | |
8168 | * Fault-like exceptions, e.g. #GP and #PF, are the lowest | |
8169 | * priority, but are only generated (pended) during instruction | |
8170 | * execution, i.e. a pending fault-like exception means the | |
8171 | * fault occurred on the *previous* instruction and must be | |
8172 | * serviced prior to recognizing any new events in order to | |
8173 | * fully complete the previous instruction. | |
664f8e26 | 8174 | */ |
1a680e35 | 8175 | else if (!vcpu->arch.exception.pending) { |
c6b22f59 | 8176 | if (vcpu->arch.nmi_injected) { |
afaf0b2f | 8177 | kvm_x86_ops.set_nmi(vcpu); |
c6b22f59 PB |
8178 | can_inject = false; |
8179 | } else if (vcpu->arch.interrupt.injected) { | |
afaf0b2f | 8180 | kvm_x86_ops.set_irq(vcpu); |
c6b22f59 PB |
8181 | can_inject = false; |
8182 | } | |
664f8e26 WL |
8183 | } |
8184 | ||
3b82b8d7 SC |
8185 | WARN_ON_ONCE(vcpu->arch.exception.injected && |
8186 | vcpu->arch.exception.pending); | |
8187 | ||
1a680e35 LA |
8188 | /* |
8189 | * Call check_nested_events() even if we reinjected a previous event | |
8190 | * in order for caller to determine if it should require immediate-exit | |
8191 | * from L2 to L1 due to pending L1 events which require exit | |
8192 | * from L2 to L1. | |
8193 | */ | |
56083bdf | 8194 | if (is_guest_mode(vcpu)) { |
33b22172 | 8195 | r = kvm_x86_ops.nested_ops->check_events(vcpu); |
c9d40913 PB |
8196 | if (r < 0) |
8197 | goto busy; | |
664f8e26 WL |
8198 | } |
8199 | ||
8200 | /* try to inject new event if pending */ | |
b59bb7bd | 8201 | if (vcpu->arch.exception.pending) { |
5c1c85d0 AK |
8202 | trace_kvm_inj_exception(vcpu->arch.exception.nr, |
8203 | vcpu->arch.exception.has_error_code, | |
8204 | vcpu->arch.exception.error_code); | |
d6e8c854 | 8205 | |
664f8e26 WL |
8206 | vcpu->arch.exception.pending = false; |
8207 | vcpu->arch.exception.injected = true; | |
8208 | ||
d6e8c854 NA |
8209 | if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT) |
8210 | __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) | | |
8211 | X86_EFLAGS_RF); | |
8212 | ||
f10c729f | 8213 | if (vcpu->arch.exception.nr == DB_VECTOR) { |
f10c729f JM |
8214 | kvm_deliver_exception_payload(vcpu); |
8215 | if (vcpu->arch.dr7 & DR7_GD) { | |
8216 | vcpu->arch.dr7 &= ~DR7_GD; | |
8217 | kvm_update_dr7(vcpu); | |
8218 | } | |
6bdf0662 NA |
8219 | } |
8220 | ||
afaf0b2f | 8221 | kvm_x86_ops.queue_exception(vcpu); |
c6b22f59 | 8222 | can_inject = false; |
1a680e35 LA |
8223 | } |
8224 | ||
c9d40913 PB |
8225 | /* |
8226 | * Finally, inject interrupt events. If an event cannot be injected | |
8227 | * due to architectural conditions (e.g. IF=0) a window-open exit | |
8228 | * will re-request KVM_REQ_EVENT. Sometimes however an event is pending | |
8229 | * and can architecturally be injected, but we cannot do it right now: | |
8230 | * an interrupt could have arrived just now and we have to inject it | |
8231 | * as a vmexit, or there could already an event in the queue, which is | |
8232 | * indicated by can_inject. In that case we request an immediate exit | |
8233 | * in order to make progress and get back here for another iteration. | |
8234 | * The kvm_x86_ops hooks communicate this by returning -EBUSY. | |
8235 | */ | |
8236 | if (vcpu->arch.smi_pending) { | |
8237 | r = can_inject ? kvm_x86_ops.smi_allowed(vcpu, true) : -EBUSY; | |
8238 | if (r < 0) | |
8239 | goto busy; | |
8240 | if (r) { | |
8241 | vcpu->arch.smi_pending = false; | |
8242 | ++vcpu->arch.smi_count; | |
8243 | enter_smm(vcpu); | |
8244 | can_inject = false; | |
8245 | } else | |
8246 | kvm_x86_ops.enable_smi_window(vcpu); | |
8247 | } | |
8248 | ||
8249 | if (vcpu->arch.nmi_pending) { | |
8250 | r = can_inject ? kvm_x86_ops.nmi_allowed(vcpu, true) : -EBUSY; | |
8251 | if (r < 0) | |
8252 | goto busy; | |
8253 | if (r) { | |
8254 | --vcpu->arch.nmi_pending; | |
8255 | vcpu->arch.nmi_injected = true; | |
8256 | kvm_x86_ops.set_nmi(vcpu); | |
8257 | can_inject = false; | |
8258 | WARN_ON(kvm_x86_ops.nmi_allowed(vcpu, true) < 0); | |
8259 | } | |
8260 | if (vcpu->arch.nmi_pending) | |
8261 | kvm_x86_ops.enable_nmi_window(vcpu); | |
8262 | } | |
1a680e35 | 8263 | |
c9d40913 PB |
8264 | if (kvm_cpu_has_injectable_intr(vcpu)) { |
8265 | r = can_inject ? kvm_x86_ops.interrupt_allowed(vcpu, true) : -EBUSY; | |
8266 | if (r < 0) | |
8267 | goto busy; | |
8268 | if (r) { | |
8269 | kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false); | |
8270 | kvm_x86_ops.set_irq(vcpu); | |
8271 | WARN_ON(kvm_x86_ops.interrupt_allowed(vcpu, true) < 0); | |
8272 | } | |
8273 | if (kvm_cpu_has_injectable_intr(vcpu)) | |
8274 | kvm_x86_ops.enable_irq_window(vcpu); | |
95ba8273 | 8275 | } |
ee2cd4b7 | 8276 | |
c9d40913 PB |
8277 | if (is_guest_mode(vcpu) && |
8278 | kvm_x86_ops.nested_ops->hv_timer_pending && | |
8279 | kvm_x86_ops.nested_ops->hv_timer_pending(vcpu)) | |
8280 | *req_immediate_exit = true; | |
8281 | ||
8282 | WARN_ON(vcpu->arch.exception.pending); | |
8283 | return; | |
8284 | ||
8285 | busy: | |
8286 | *req_immediate_exit = true; | |
8287 | return; | |
95ba8273 GN |
8288 | } |
8289 | ||
7460fb4a AK |
8290 | static void process_nmi(struct kvm_vcpu *vcpu) |
8291 | { | |
8292 | unsigned limit = 2; | |
8293 | ||
8294 | /* | |
8295 | * x86 is limited to one NMI running, and one NMI pending after it. | |
8296 | * If an NMI is already in progress, limit further NMIs to just one. | |
8297 | * Otherwise, allow two (and we'll inject the first one immediately). | |
8298 | */ | |
afaf0b2f | 8299 | if (kvm_x86_ops.get_nmi_mask(vcpu) || vcpu->arch.nmi_injected) |
7460fb4a AK |
8300 | limit = 1; |
8301 | ||
8302 | vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0); | |
8303 | vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit); | |
8304 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
8305 | } | |
8306 | ||
ee2cd4b7 | 8307 | static u32 enter_smm_get_segment_flags(struct kvm_segment *seg) |
660a5d51 PB |
8308 | { |
8309 | u32 flags = 0; | |
8310 | flags |= seg->g << 23; | |
8311 | flags |= seg->db << 22; | |
8312 | flags |= seg->l << 21; | |
8313 | flags |= seg->avl << 20; | |
8314 | flags |= seg->present << 15; | |
8315 | flags |= seg->dpl << 13; | |
8316 | flags |= seg->s << 12; | |
8317 | flags |= seg->type << 8; | |
8318 | return flags; | |
8319 | } | |
8320 | ||
ee2cd4b7 | 8321 | static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n) |
660a5d51 PB |
8322 | { |
8323 | struct kvm_segment seg; | |
8324 | int offset; | |
8325 | ||
8326 | kvm_get_segment(vcpu, &seg, n); | |
8327 | put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector); | |
8328 | ||
8329 | if (n < 3) | |
8330 | offset = 0x7f84 + n * 12; | |
8331 | else | |
8332 | offset = 0x7f2c + (n - 3) * 12; | |
8333 | ||
8334 | put_smstate(u32, buf, offset + 8, seg.base); | |
8335 | put_smstate(u32, buf, offset + 4, seg.limit); | |
ee2cd4b7 | 8336 | put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg)); |
660a5d51 PB |
8337 | } |
8338 | ||
efbb288a | 8339 | #ifdef CONFIG_X86_64 |
ee2cd4b7 | 8340 | static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n) |
660a5d51 PB |
8341 | { |
8342 | struct kvm_segment seg; | |
8343 | int offset; | |
8344 | u16 flags; | |
8345 | ||
8346 | kvm_get_segment(vcpu, &seg, n); | |
8347 | offset = 0x7e00 + n * 16; | |
8348 | ||
ee2cd4b7 | 8349 | flags = enter_smm_get_segment_flags(&seg) >> 8; |
660a5d51 PB |
8350 | put_smstate(u16, buf, offset, seg.selector); |
8351 | put_smstate(u16, buf, offset + 2, flags); | |
8352 | put_smstate(u32, buf, offset + 4, seg.limit); | |
8353 | put_smstate(u64, buf, offset + 8, seg.base); | |
8354 | } | |
efbb288a | 8355 | #endif |
660a5d51 | 8356 | |
ee2cd4b7 | 8357 | static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf) |
660a5d51 PB |
8358 | { |
8359 | struct desc_ptr dt; | |
8360 | struct kvm_segment seg; | |
8361 | unsigned long val; | |
8362 | int i; | |
8363 | ||
8364 | put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu)); | |
8365 | put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu)); | |
8366 | put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu)); | |
8367 | put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu)); | |
8368 | ||
8369 | for (i = 0; i < 8; i++) | |
8370 | put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i)); | |
8371 | ||
8372 | kvm_get_dr(vcpu, 6, &val); | |
8373 | put_smstate(u32, buf, 0x7fcc, (u32)val); | |
8374 | kvm_get_dr(vcpu, 7, &val); | |
8375 | put_smstate(u32, buf, 0x7fc8, (u32)val); | |
8376 | ||
8377 | kvm_get_segment(vcpu, &seg, VCPU_SREG_TR); | |
8378 | put_smstate(u32, buf, 0x7fc4, seg.selector); | |
8379 | put_smstate(u32, buf, 0x7f64, seg.base); | |
8380 | put_smstate(u32, buf, 0x7f60, seg.limit); | |
ee2cd4b7 | 8381 | put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg)); |
660a5d51 PB |
8382 | |
8383 | kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR); | |
8384 | put_smstate(u32, buf, 0x7fc0, seg.selector); | |
8385 | put_smstate(u32, buf, 0x7f80, seg.base); | |
8386 | put_smstate(u32, buf, 0x7f7c, seg.limit); | |
ee2cd4b7 | 8387 | put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg)); |
660a5d51 | 8388 | |
afaf0b2f | 8389 | kvm_x86_ops.get_gdt(vcpu, &dt); |
660a5d51 PB |
8390 | put_smstate(u32, buf, 0x7f74, dt.address); |
8391 | put_smstate(u32, buf, 0x7f70, dt.size); | |
8392 | ||
afaf0b2f | 8393 | kvm_x86_ops.get_idt(vcpu, &dt); |
660a5d51 PB |
8394 | put_smstate(u32, buf, 0x7f58, dt.address); |
8395 | put_smstate(u32, buf, 0x7f54, dt.size); | |
8396 | ||
8397 | for (i = 0; i < 6; i++) | |
ee2cd4b7 | 8398 | enter_smm_save_seg_32(vcpu, buf, i); |
660a5d51 PB |
8399 | |
8400 | put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu)); | |
8401 | ||
8402 | /* revision id */ | |
8403 | put_smstate(u32, buf, 0x7efc, 0x00020000); | |
8404 | put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase); | |
8405 | } | |
8406 | ||
b68f3cc7 | 8407 | #ifdef CONFIG_X86_64 |
ee2cd4b7 | 8408 | static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf) |
660a5d51 | 8409 | { |
660a5d51 PB |
8410 | struct desc_ptr dt; |
8411 | struct kvm_segment seg; | |
8412 | unsigned long val; | |
8413 | int i; | |
8414 | ||
8415 | for (i = 0; i < 16; i++) | |
8416 | put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i)); | |
8417 | ||
8418 | put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu)); | |
8419 | put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu)); | |
8420 | ||
8421 | kvm_get_dr(vcpu, 6, &val); | |
8422 | put_smstate(u64, buf, 0x7f68, val); | |
8423 | kvm_get_dr(vcpu, 7, &val); | |
8424 | put_smstate(u64, buf, 0x7f60, val); | |
8425 | ||
8426 | put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu)); | |
8427 | put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu)); | |
8428 | put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu)); | |
8429 | ||
8430 | put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase); | |
8431 | ||
8432 | /* revision id */ | |
8433 | put_smstate(u32, buf, 0x7efc, 0x00020064); | |
8434 | ||
8435 | put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer); | |
8436 | ||
8437 | kvm_get_segment(vcpu, &seg, VCPU_SREG_TR); | |
8438 | put_smstate(u16, buf, 0x7e90, seg.selector); | |
ee2cd4b7 | 8439 | put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8); |
660a5d51 PB |
8440 | put_smstate(u32, buf, 0x7e94, seg.limit); |
8441 | put_smstate(u64, buf, 0x7e98, seg.base); | |
8442 | ||
afaf0b2f | 8443 | kvm_x86_ops.get_idt(vcpu, &dt); |
660a5d51 PB |
8444 | put_smstate(u32, buf, 0x7e84, dt.size); |
8445 | put_smstate(u64, buf, 0x7e88, dt.address); | |
8446 | ||
8447 | kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR); | |
8448 | put_smstate(u16, buf, 0x7e70, seg.selector); | |
ee2cd4b7 | 8449 | put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8); |
660a5d51 PB |
8450 | put_smstate(u32, buf, 0x7e74, seg.limit); |
8451 | put_smstate(u64, buf, 0x7e78, seg.base); | |
8452 | ||
afaf0b2f | 8453 | kvm_x86_ops.get_gdt(vcpu, &dt); |
660a5d51 PB |
8454 | put_smstate(u32, buf, 0x7e64, dt.size); |
8455 | put_smstate(u64, buf, 0x7e68, dt.address); | |
8456 | ||
8457 | for (i = 0; i < 6; i++) | |
ee2cd4b7 | 8458 | enter_smm_save_seg_64(vcpu, buf, i); |
660a5d51 | 8459 | } |
b68f3cc7 | 8460 | #endif |
660a5d51 | 8461 | |
ee2cd4b7 | 8462 | static void enter_smm(struct kvm_vcpu *vcpu) |
64d60670 | 8463 | { |
660a5d51 | 8464 | struct kvm_segment cs, ds; |
18c3626e | 8465 | struct desc_ptr dt; |
660a5d51 PB |
8466 | char buf[512]; |
8467 | u32 cr0; | |
8468 | ||
660a5d51 | 8469 | trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true); |
660a5d51 | 8470 | memset(buf, 0, 512); |
b68f3cc7 | 8471 | #ifdef CONFIG_X86_64 |
d6321d49 | 8472 | if (guest_cpuid_has(vcpu, X86_FEATURE_LM)) |
ee2cd4b7 | 8473 | enter_smm_save_state_64(vcpu, buf); |
660a5d51 | 8474 | else |
b68f3cc7 | 8475 | #endif |
ee2cd4b7 | 8476 | enter_smm_save_state_32(vcpu, buf); |
660a5d51 | 8477 | |
0234bf88 LP |
8478 | /* |
8479 | * Give pre_enter_smm() a chance to make ISA-specific changes to the | |
8480 | * vCPU state (e.g. leave guest mode) after we've saved the state into | |
8481 | * the SMM state-save area. | |
8482 | */ | |
afaf0b2f | 8483 | kvm_x86_ops.pre_enter_smm(vcpu, buf); |
0234bf88 LP |
8484 | |
8485 | vcpu->arch.hflags |= HF_SMM_MASK; | |
54bf36aa | 8486 | kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf)); |
660a5d51 | 8487 | |
afaf0b2f | 8488 | if (kvm_x86_ops.get_nmi_mask(vcpu)) |
660a5d51 PB |
8489 | vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK; |
8490 | else | |
afaf0b2f | 8491 | kvm_x86_ops.set_nmi_mask(vcpu, true); |
660a5d51 PB |
8492 | |
8493 | kvm_set_rflags(vcpu, X86_EFLAGS_FIXED); | |
8494 | kvm_rip_write(vcpu, 0x8000); | |
8495 | ||
8496 | cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG); | |
afaf0b2f | 8497 | kvm_x86_ops.set_cr0(vcpu, cr0); |
660a5d51 PB |
8498 | vcpu->arch.cr0 = cr0; |
8499 | ||
afaf0b2f | 8500 | kvm_x86_ops.set_cr4(vcpu, 0); |
660a5d51 | 8501 | |
18c3626e PB |
8502 | /* Undocumented: IDT limit is set to zero on entry to SMM. */ |
8503 | dt.address = dt.size = 0; | |
afaf0b2f | 8504 | kvm_x86_ops.set_idt(vcpu, &dt); |
18c3626e | 8505 | |
660a5d51 PB |
8506 | __kvm_set_dr(vcpu, 7, DR7_FIXED_1); |
8507 | ||
8508 | cs.selector = (vcpu->arch.smbase >> 4) & 0xffff; | |
8509 | cs.base = vcpu->arch.smbase; | |
8510 | ||
8511 | ds.selector = 0; | |
8512 | ds.base = 0; | |
8513 | ||
8514 | cs.limit = ds.limit = 0xffffffff; | |
8515 | cs.type = ds.type = 0x3; | |
8516 | cs.dpl = ds.dpl = 0; | |
8517 | cs.db = ds.db = 0; | |
8518 | cs.s = ds.s = 1; | |
8519 | cs.l = ds.l = 0; | |
8520 | cs.g = ds.g = 1; | |
8521 | cs.avl = ds.avl = 0; | |
8522 | cs.present = ds.present = 1; | |
8523 | cs.unusable = ds.unusable = 0; | |
8524 | cs.padding = ds.padding = 0; | |
8525 | ||
8526 | kvm_set_segment(vcpu, &cs, VCPU_SREG_CS); | |
8527 | kvm_set_segment(vcpu, &ds, VCPU_SREG_DS); | |
8528 | kvm_set_segment(vcpu, &ds, VCPU_SREG_ES); | |
8529 | kvm_set_segment(vcpu, &ds, VCPU_SREG_FS); | |
8530 | kvm_set_segment(vcpu, &ds, VCPU_SREG_GS); | |
8531 | kvm_set_segment(vcpu, &ds, VCPU_SREG_SS); | |
8532 | ||
b68f3cc7 | 8533 | #ifdef CONFIG_X86_64 |
d6321d49 | 8534 | if (guest_cpuid_has(vcpu, X86_FEATURE_LM)) |
afaf0b2f | 8535 | kvm_x86_ops.set_efer(vcpu, 0); |
b68f3cc7 | 8536 | #endif |
660a5d51 | 8537 | |
aedbaf4f | 8538 | kvm_update_cpuid_runtime(vcpu); |
660a5d51 | 8539 | kvm_mmu_reset_context(vcpu); |
64d60670 PB |
8540 | } |
8541 | ||
ee2cd4b7 | 8542 | static void process_smi(struct kvm_vcpu *vcpu) |
c43203ca PB |
8543 | { |
8544 | vcpu->arch.smi_pending = true; | |
8545 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
8546 | } | |
8547 | ||
7ee30bc1 NNL |
8548 | void kvm_make_scan_ioapic_request_mask(struct kvm *kvm, |
8549 | unsigned long *vcpu_bitmap) | |
8550 | { | |
8551 | cpumask_var_t cpus; | |
7ee30bc1 NNL |
8552 | |
8553 | zalloc_cpumask_var(&cpus, GFP_ATOMIC); | |
8554 | ||
db5a95ec | 8555 | kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC, |
54163a34 | 8556 | NULL, vcpu_bitmap, cpus); |
7ee30bc1 NNL |
8557 | |
8558 | free_cpumask_var(cpus); | |
8559 | } | |
8560 | ||
2860c4b1 PB |
8561 | void kvm_make_scan_ioapic_request(struct kvm *kvm) |
8562 | { | |
8563 | kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC); | |
8564 | } | |
8565 | ||
8df14af4 SS |
8566 | void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu) |
8567 | { | |
8568 | if (!lapic_in_kernel(vcpu)) | |
8569 | return; | |
8570 | ||
8571 | vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm); | |
8572 | kvm_apic_update_apicv(vcpu); | |
afaf0b2f | 8573 | kvm_x86_ops.refresh_apicv_exec_ctrl(vcpu); |
8df14af4 SS |
8574 | } |
8575 | EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv); | |
8576 | ||
8577 | /* | |
8578 | * NOTE: Do not hold any lock prior to calling this. | |
8579 | * | |
8580 | * In particular, kvm_request_apicv_update() expects kvm->srcu not to be | |
8581 | * locked, because it calls __x86_set_memory_region() which does | |
8582 | * synchronize_srcu(&kvm->srcu). | |
8583 | */ | |
8584 | void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit) | |
8585 | { | |
7d611233 | 8586 | struct kvm_vcpu *except; |
8e205a6b PB |
8587 | unsigned long old, new, expected; |
8588 | ||
afaf0b2f SC |
8589 | if (!kvm_x86_ops.check_apicv_inhibit_reasons || |
8590 | !kvm_x86_ops.check_apicv_inhibit_reasons(bit)) | |
ef8efd7a SS |
8591 | return; |
8592 | ||
8e205a6b PB |
8593 | old = READ_ONCE(kvm->arch.apicv_inhibit_reasons); |
8594 | do { | |
8595 | expected = new = old; | |
8596 | if (activate) | |
8597 | __clear_bit(bit, &new); | |
8598 | else | |
8599 | __set_bit(bit, &new); | |
8600 | if (new == old) | |
8601 | break; | |
8602 | old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new); | |
8603 | } while (old != expected); | |
8604 | ||
8605 | if (!!old == !!new) | |
8606 | return; | |
8df14af4 | 8607 | |
24bbf74c | 8608 | trace_kvm_apicv_update_request(activate, bit); |
afaf0b2f SC |
8609 | if (kvm_x86_ops.pre_update_apicv_exec_ctrl) |
8610 | kvm_x86_ops.pre_update_apicv_exec_ctrl(kvm, activate); | |
7d611233 SS |
8611 | |
8612 | /* | |
8613 | * Sending request to update APICV for all other vcpus, | |
8614 | * while update the calling vcpu immediately instead of | |
8615 | * waiting for another #VMEXIT to handle the request. | |
8616 | */ | |
8617 | except = kvm_get_running_vcpu(); | |
8618 | kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE, | |
8619 | except); | |
8620 | if (except) | |
8621 | kvm_vcpu_update_apicv(except); | |
8df14af4 SS |
8622 | } |
8623 | EXPORT_SYMBOL_GPL(kvm_request_apicv_update); | |
8624 | ||
3d81bc7e | 8625 | static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu) |
c7c9c56c | 8626 | { |
dcbd3e49 | 8627 | if (!kvm_apic_present(vcpu)) |
3d81bc7e | 8628 | return; |
c7c9c56c | 8629 | |
6308630b | 8630 | bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256); |
c7c9c56c | 8631 | |
b053b2ae | 8632 | if (irqchip_split(vcpu->kvm)) |
6308630b | 8633 | kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors); |
db2bdcbb | 8634 | else { |
fa59cc00 | 8635 | if (vcpu->arch.apicv_active) |
afaf0b2f | 8636 | kvm_x86_ops.sync_pir_to_irr(vcpu); |
e97f852f WL |
8637 | if (ioapic_in_kernel(vcpu->kvm)) |
8638 | kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors); | |
db2bdcbb | 8639 | } |
e40ff1d6 LA |
8640 | |
8641 | if (is_guest_mode(vcpu)) | |
8642 | vcpu->arch.load_eoi_exitmap_pending = true; | |
8643 | else | |
8644 | kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu); | |
8645 | } | |
8646 | ||
8647 | static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu) | |
8648 | { | |
8649 | u64 eoi_exit_bitmap[4]; | |
8650 | ||
8651 | if (!kvm_apic_hw_enabled(vcpu->arch.apic)) | |
8652 | return; | |
8653 | ||
5c919412 AS |
8654 | bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors, |
8655 | vcpu_to_synic(vcpu)->vec_bitmap, 256); | |
afaf0b2f | 8656 | kvm_x86_ops.load_eoi_exitmap(vcpu, eoi_exit_bitmap); |
c7c9c56c YZ |
8657 | } |
8658 | ||
e649b3f0 ET |
8659 | void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm, |
8660 | unsigned long start, unsigned long end) | |
b1394e74 RK |
8661 | { |
8662 | unsigned long apic_address; | |
8663 | ||
8664 | /* | |
8665 | * The physical address of apic access page is stored in the VMCS. | |
8666 | * Update it when it becomes invalid. | |
8667 | */ | |
8668 | apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT); | |
8669 | if (start <= apic_address && apic_address < end) | |
8670 | kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD); | |
8671 | } | |
8672 | ||
4256f43f TC |
8673 | void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu) |
8674 | { | |
35754c98 | 8675 | if (!lapic_in_kernel(vcpu)) |
f439ed27 PB |
8676 | return; |
8677 | ||
afaf0b2f | 8678 | if (!kvm_x86_ops.set_apic_access_page_addr) |
4256f43f TC |
8679 | return; |
8680 | ||
a4148b7c | 8681 | kvm_x86_ops.set_apic_access_page_addr(vcpu); |
4256f43f | 8682 | } |
4256f43f | 8683 | |
d264ee0c SC |
8684 | void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu) |
8685 | { | |
8686 | smp_send_reschedule(vcpu->cpu); | |
8687 | } | |
8688 | EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit); | |
8689 | ||
9357d939 | 8690 | /* |
362c698f | 8691 | * Returns 1 to let vcpu_run() continue the guest execution loop without |
9357d939 TY |
8692 | * exiting to the userspace. Otherwise, the value will be returned to the |
8693 | * userspace. | |
8694 | */ | |
851ba692 | 8695 | static int vcpu_enter_guest(struct kvm_vcpu *vcpu) |
b6c7a5dc HB |
8696 | { |
8697 | int r; | |
62a193ed MG |
8698 | bool req_int_win = |
8699 | dm_request_for_irq_injection(vcpu) && | |
8700 | kvm_cpu_accept_dm_intr(vcpu); | |
404d5d7b | 8701 | fastpath_t exit_fastpath; |
62a193ed | 8702 | |
730dca42 | 8703 | bool req_immediate_exit = false; |
b6c7a5dc | 8704 | |
2fa6e1e1 | 8705 | if (kvm_request_pending(vcpu)) { |
729c15c2 PB |
8706 | if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) { |
8707 | if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) { | |
671ddc70 JM |
8708 | r = 0; |
8709 | goto out; | |
8710 | } | |
8711 | } | |
a8eeb04a | 8712 | if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) |
2e53d63a | 8713 | kvm_mmu_unload(vcpu); |
a8eeb04a | 8714 | if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu)) |
2f599714 | 8715 | __kvm_migrate_timers(vcpu); |
d828199e MT |
8716 | if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu)) |
8717 | kvm_gen_update_masterclock(vcpu->kvm); | |
0061d53d MT |
8718 | if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu)) |
8719 | kvm_gen_kvmclock_update(vcpu); | |
34c238a1 ZA |
8720 | if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) { |
8721 | r = kvm_guest_time_update(vcpu); | |
8cfdc000 ZA |
8722 | if (unlikely(r)) |
8723 | goto out; | |
8724 | } | |
a8eeb04a | 8725 | if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu)) |
4731d4c7 | 8726 | kvm_mmu_sync_roots(vcpu); |
727a7e27 PB |
8727 | if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu)) |
8728 | kvm_mmu_load_pgd(vcpu); | |
eeeb4f67 | 8729 | if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) { |
7780938c | 8730 | kvm_vcpu_flush_tlb_all(vcpu); |
eeeb4f67 SC |
8731 | |
8732 | /* Flushing all ASIDs flushes the current ASID... */ | |
8733 | kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu); | |
8734 | } | |
8735 | if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu)) | |
8736 | kvm_vcpu_flush_tlb_current(vcpu); | |
0baedd79 VK |
8737 | if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu)) |
8738 | kvm_vcpu_flush_tlb_guest(vcpu); | |
eeeb4f67 | 8739 | |
a8eeb04a | 8740 | if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) { |
851ba692 | 8741 | vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS; |
b93463aa AK |
8742 | r = 0; |
8743 | goto out; | |
8744 | } | |
a8eeb04a | 8745 | if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) { |
851ba692 | 8746 | vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN; |
bbeac283 | 8747 | vcpu->mmio_needed = 0; |
71c4dfaf JR |
8748 | r = 0; |
8749 | goto out; | |
8750 | } | |
af585b92 GN |
8751 | if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) { |
8752 | /* Page is swapped out. Do synthetic halt */ | |
8753 | vcpu->arch.apf.halted = true; | |
8754 | r = 1; | |
8755 | goto out; | |
8756 | } | |
c9aaa895 GC |
8757 | if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu)) |
8758 | record_steal_time(vcpu); | |
64d60670 PB |
8759 | if (kvm_check_request(KVM_REQ_SMI, vcpu)) |
8760 | process_smi(vcpu); | |
7460fb4a AK |
8761 | if (kvm_check_request(KVM_REQ_NMI, vcpu)) |
8762 | process_nmi(vcpu); | |
f5132b01 | 8763 | if (kvm_check_request(KVM_REQ_PMU, vcpu)) |
c6702c9d | 8764 | kvm_pmu_handle_event(vcpu); |
f5132b01 | 8765 | if (kvm_check_request(KVM_REQ_PMI, vcpu)) |
c6702c9d | 8766 | kvm_pmu_deliver_pmi(vcpu); |
7543a635 SR |
8767 | if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) { |
8768 | BUG_ON(vcpu->arch.pending_ioapic_eoi > 255); | |
8769 | if (test_bit(vcpu->arch.pending_ioapic_eoi, | |
6308630b | 8770 | vcpu->arch.ioapic_handled_vectors)) { |
7543a635 SR |
8771 | vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI; |
8772 | vcpu->run->eoi.vector = | |
8773 | vcpu->arch.pending_ioapic_eoi; | |
8774 | r = 0; | |
8775 | goto out; | |
8776 | } | |
8777 | } | |
3d81bc7e YZ |
8778 | if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu)) |
8779 | vcpu_scan_ioapic(vcpu); | |
e40ff1d6 LA |
8780 | if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu)) |
8781 | vcpu_load_eoi_exitmap(vcpu); | |
4256f43f TC |
8782 | if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu)) |
8783 | kvm_vcpu_reload_apic_access_page(vcpu); | |
2ce79189 AS |
8784 | if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) { |
8785 | vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT; | |
8786 | vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH; | |
8787 | r = 0; | |
8788 | goto out; | |
8789 | } | |
e516cebb AS |
8790 | if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) { |
8791 | vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT; | |
8792 | vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET; | |
8793 | r = 0; | |
8794 | goto out; | |
8795 | } | |
db397571 AS |
8796 | if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) { |
8797 | vcpu->run->exit_reason = KVM_EXIT_HYPERV; | |
8798 | vcpu->run->hyperv = vcpu->arch.hyperv.exit; | |
8799 | r = 0; | |
8800 | goto out; | |
8801 | } | |
f3b138c5 AS |
8802 | |
8803 | /* | |
8804 | * KVM_REQ_HV_STIMER has to be processed after | |
8805 | * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers | |
8806 | * depend on the guest clock being up-to-date | |
8807 | */ | |
1f4b34f8 AS |
8808 | if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu)) |
8809 | kvm_hv_process_stimers(vcpu); | |
8df14af4 SS |
8810 | if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu)) |
8811 | kvm_vcpu_update_apicv(vcpu); | |
557a961a VK |
8812 | if (kvm_check_request(KVM_REQ_APF_READY, vcpu)) |
8813 | kvm_check_async_pf_completion(vcpu); | |
1a155254 AG |
8814 | if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu)) |
8815 | kvm_x86_ops.msr_filter_changed(vcpu); | |
2f52d58c | 8816 | } |
b93463aa | 8817 | |
b463a6f7 | 8818 | if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) { |
0f1e261e | 8819 | ++vcpu->stat.req_event; |
66450a21 JK |
8820 | kvm_apic_accept_events(vcpu); |
8821 | if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) { | |
8822 | r = 1; | |
8823 | goto out; | |
8824 | } | |
8825 | ||
c9d40913 PB |
8826 | inject_pending_event(vcpu, &req_immediate_exit); |
8827 | if (req_int_win) | |
8828 | kvm_x86_ops.enable_irq_window(vcpu); | |
b463a6f7 AK |
8829 | |
8830 | if (kvm_lapic_enabled(vcpu)) { | |
8831 | update_cr8_intercept(vcpu); | |
8832 | kvm_lapic_sync_to_vapic(vcpu); | |
8833 | } | |
8834 | } | |
8835 | ||
d8368af8 AK |
8836 | r = kvm_mmu_reload(vcpu); |
8837 | if (unlikely(r)) { | |
d905c069 | 8838 | goto cancel_injection; |
d8368af8 AK |
8839 | } |
8840 | ||
b6c7a5dc HB |
8841 | preempt_disable(); |
8842 | ||
afaf0b2f | 8843 | kvm_x86_ops.prepare_guest_switch(vcpu); |
b95234c8 PB |
8844 | |
8845 | /* | |
8846 | * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt | |
8847 | * IPI are then delayed after guest entry, which ensures that they | |
8848 | * result in virtual interrupt delivery. | |
8849 | */ | |
8850 | local_irq_disable(); | |
6b7e2d09 XG |
8851 | vcpu->mode = IN_GUEST_MODE; |
8852 | ||
01b71917 MT |
8853 | srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); |
8854 | ||
0f127d12 | 8855 | /* |
b95234c8 | 8856 | * 1) We should set ->mode before checking ->requests. Please see |
cde9af6e | 8857 | * the comment in kvm_vcpu_exiting_guest_mode(). |
b95234c8 | 8858 | * |
81b01667 | 8859 | * 2) For APICv, we should set ->mode before checking PID.ON. This |
b95234c8 PB |
8860 | * pairs with the memory barrier implicit in pi_test_and_set_on |
8861 | * (see vmx_deliver_posted_interrupt). | |
8862 | * | |
8863 | * 3) This also orders the write to mode from any reads to the page | |
8864 | * tables done while the VCPU is running. Please see the comment | |
8865 | * in kvm_flush_remote_tlbs. | |
6b7e2d09 | 8866 | */ |
01b71917 | 8867 | smp_mb__after_srcu_read_unlock(); |
b6c7a5dc | 8868 | |
b95234c8 PB |
8869 | /* |
8870 | * This handles the case where a posted interrupt was | |
8871 | * notified with kvm_vcpu_kick. | |
8872 | */ | |
fa59cc00 | 8873 | if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active) |
afaf0b2f | 8874 | kvm_x86_ops.sync_pir_to_irr(vcpu); |
32f88400 | 8875 | |
5a9f5443 | 8876 | if (kvm_vcpu_exit_request(vcpu)) { |
6b7e2d09 | 8877 | vcpu->mode = OUTSIDE_GUEST_MODE; |
d94e1dc9 | 8878 | smp_wmb(); |
6c142801 AK |
8879 | local_irq_enable(); |
8880 | preempt_enable(); | |
01b71917 | 8881 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); |
6c142801 | 8882 | r = 1; |
d905c069 | 8883 | goto cancel_injection; |
6c142801 AK |
8884 | } |
8885 | ||
c43203ca PB |
8886 | if (req_immediate_exit) { |
8887 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
afaf0b2f | 8888 | kvm_x86_ops.request_immediate_exit(vcpu); |
c43203ca | 8889 | } |
d6185f20 | 8890 | |
b2d52255 | 8891 | trace_kvm_entry(vcpu); |
b6c7a5dc | 8892 | |
2620fe26 SC |
8893 | fpregs_assert_state_consistent(); |
8894 | if (test_thread_flag(TIF_NEED_FPU_LOAD)) | |
8895 | switch_fpu_return(); | |
5f409e20 | 8896 | |
42dbaa5a | 8897 | if (unlikely(vcpu->arch.switch_db_regs)) { |
42dbaa5a JK |
8898 | set_debugreg(0, 7); |
8899 | set_debugreg(vcpu->arch.eff_db[0], 0); | |
8900 | set_debugreg(vcpu->arch.eff_db[1], 1); | |
8901 | set_debugreg(vcpu->arch.eff_db[2], 2); | |
8902 | set_debugreg(vcpu->arch.eff_db[3], 3); | |
c77fb5fe | 8903 | set_debugreg(vcpu->arch.dr6, 6); |
ae561ede | 8904 | vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD; |
42dbaa5a | 8905 | } |
b6c7a5dc | 8906 | |
a9ab13ff | 8907 | exit_fastpath = kvm_x86_ops.run(vcpu); |
b6c7a5dc | 8908 | |
c77fb5fe PB |
8909 | /* |
8910 | * Do this here before restoring debug registers on the host. And | |
8911 | * since we do this before handling the vmexit, a DR access vmexit | |
8912 | * can (a) read the correct value of the debug registers, (b) set | |
8913 | * KVM_DEBUGREG_WONT_EXIT again. | |
8914 | */ | |
8915 | if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) { | |
c77fb5fe | 8916 | WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP); |
afaf0b2f | 8917 | kvm_x86_ops.sync_dirty_debug_regs(vcpu); |
70e4da7a | 8918 | kvm_update_dr0123(vcpu); |
70e4da7a PB |
8919 | kvm_update_dr7(vcpu); |
8920 | vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD; | |
c77fb5fe PB |
8921 | } |
8922 | ||
24f1e32c FW |
8923 | /* |
8924 | * If the guest has used debug registers, at least dr7 | |
8925 | * will be disabled while returning to the host. | |
8926 | * If we don't have active breakpoints in the host, we don't | |
8927 | * care about the messed up debug address registers. But if | |
8928 | * we have some of them active, restore the old state. | |
8929 | */ | |
59d8eb53 | 8930 | if (hw_breakpoint_active()) |
24f1e32c | 8931 | hw_breakpoint_restore(); |
42dbaa5a | 8932 | |
c967118d | 8933 | vcpu->arch.last_vmentry_cpu = vcpu->cpu; |
4ba76538 | 8934 | vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc()); |
1d5f066e | 8935 | |
6b7e2d09 | 8936 | vcpu->mode = OUTSIDE_GUEST_MODE; |
d94e1dc9 | 8937 | smp_wmb(); |
a547c6db | 8938 | |
a9ab13ff | 8939 | kvm_x86_ops.handle_exit_irqoff(vcpu); |
b6c7a5dc | 8940 | |
d7a08882 SC |
8941 | /* |
8942 | * Consume any pending interrupts, including the possible source of | |
8943 | * VM-Exit on SVM and any ticks that occur between VM-Exit and now. | |
8944 | * An instruction is required after local_irq_enable() to fully unblock | |
8945 | * interrupts on processors that implement an interrupt shadow, the | |
8946 | * stat.exits increment will do nicely. | |
8947 | */ | |
8948 | kvm_before_interrupt(vcpu); | |
8949 | local_irq_enable(); | |
b6c7a5dc | 8950 | ++vcpu->stat.exits; |
d7a08882 SC |
8951 | local_irq_disable(); |
8952 | kvm_after_interrupt(vcpu); | |
b6c7a5dc | 8953 | |
ec0671d5 WL |
8954 | if (lapic_in_kernel(vcpu)) { |
8955 | s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta; | |
8956 | if (delta != S64_MIN) { | |
8957 | trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta); | |
8958 | vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN; | |
8959 | } | |
8960 | } | |
b6c7a5dc | 8961 | |
f2485b3e | 8962 | local_irq_enable(); |
b6c7a5dc HB |
8963 | preempt_enable(); |
8964 | ||
f656ce01 | 8965 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); |
3200f405 | 8966 | |
b6c7a5dc HB |
8967 | /* |
8968 | * Profile KVM exit RIPs: | |
8969 | */ | |
8970 | if (unlikely(prof_on == KVM_PROFILING)) { | |
5fdbf976 MT |
8971 | unsigned long rip = kvm_rip_read(vcpu); |
8972 | profile_hit(KVM_PROFILING, (void *)rip); | |
b6c7a5dc HB |
8973 | } |
8974 | ||
cc578287 ZA |
8975 | if (unlikely(vcpu->arch.tsc_always_catchup)) |
8976 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); | |
298101da | 8977 | |
5cfb1d5a MT |
8978 | if (vcpu->arch.apic_attention) |
8979 | kvm_lapic_sync_from_vapic(vcpu); | |
b93463aa | 8980 | |
afaf0b2f | 8981 | r = kvm_x86_ops.handle_exit(vcpu, exit_fastpath); |
d905c069 MT |
8982 | return r; |
8983 | ||
8984 | cancel_injection: | |
8081ad06 SC |
8985 | if (req_immediate_exit) |
8986 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
afaf0b2f | 8987 | kvm_x86_ops.cancel_injection(vcpu); |
ae7a2a3f MT |
8988 | if (unlikely(vcpu->arch.apic_attention)) |
8989 | kvm_lapic_sync_from_vapic(vcpu); | |
d7690175 MT |
8990 | out: |
8991 | return r; | |
8992 | } | |
b6c7a5dc | 8993 | |
362c698f PB |
8994 | static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu) |
8995 | { | |
bf9f6ac8 | 8996 | if (!kvm_arch_vcpu_runnable(vcpu) && |
afaf0b2f | 8997 | (!kvm_x86_ops.pre_block || kvm_x86_ops.pre_block(vcpu) == 0)) { |
9c8fd1ba PB |
8998 | srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); |
8999 | kvm_vcpu_block(vcpu); | |
9000 | vcpu->srcu_idx = srcu_read_lock(&kvm->srcu); | |
bf9f6ac8 | 9001 | |
afaf0b2f SC |
9002 | if (kvm_x86_ops.post_block) |
9003 | kvm_x86_ops.post_block(vcpu); | |
bf9f6ac8 | 9004 | |
9c8fd1ba PB |
9005 | if (!kvm_check_request(KVM_REQ_UNHALT, vcpu)) |
9006 | return 1; | |
9007 | } | |
362c698f PB |
9008 | |
9009 | kvm_apic_accept_events(vcpu); | |
9010 | switch(vcpu->arch.mp_state) { | |
9011 | case KVM_MP_STATE_HALTED: | |
9012 | vcpu->arch.pv.pv_unhalted = false; | |
9013 | vcpu->arch.mp_state = | |
9014 | KVM_MP_STATE_RUNNABLE; | |
df561f66 | 9015 | fallthrough; |
362c698f PB |
9016 | case KVM_MP_STATE_RUNNABLE: |
9017 | vcpu->arch.apf.halted = false; | |
9018 | break; | |
9019 | case KVM_MP_STATE_INIT_RECEIVED: | |
9020 | break; | |
9021 | default: | |
9022 | return -EINTR; | |
362c698f PB |
9023 | } |
9024 | return 1; | |
9025 | } | |
09cec754 | 9026 | |
5d9bc648 PB |
9027 | static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu) |
9028 | { | |
56083bdf | 9029 | if (is_guest_mode(vcpu)) |
33b22172 | 9030 | kvm_x86_ops.nested_ops->check_events(vcpu); |
0ad3bed6 | 9031 | |
5d9bc648 PB |
9032 | return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE && |
9033 | !vcpu->arch.apf.halted); | |
9034 | } | |
9035 | ||
362c698f | 9036 | static int vcpu_run(struct kvm_vcpu *vcpu) |
d7690175 MT |
9037 | { |
9038 | int r; | |
f656ce01 | 9039 | struct kvm *kvm = vcpu->kvm; |
d7690175 | 9040 | |
f656ce01 | 9041 | vcpu->srcu_idx = srcu_read_lock(&kvm->srcu); |
c595ceee | 9042 | vcpu->arch.l1tf_flush_l1d = true; |
d7690175 | 9043 | |
362c698f | 9044 | for (;;) { |
58f800d5 | 9045 | if (kvm_vcpu_running(vcpu)) { |
851ba692 | 9046 | r = vcpu_enter_guest(vcpu); |
bf9f6ac8 | 9047 | } else { |
362c698f | 9048 | r = vcpu_block(kvm, vcpu); |
bf9f6ac8 FW |
9049 | } |
9050 | ||
09cec754 GN |
9051 | if (r <= 0) |
9052 | break; | |
9053 | ||
72875d8a | 9054 | kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu); |
09cec754 GN |
9055 | if (kvm_cpu_has_pending_timer(vcpu)) |
9056 | kvm_inject_pending_timer_irqs(vcpu); | |
9057 | ||
782d422b MG |
9058 | if (dm_request_for_irq_injection(vcpu) && |
9059 | kvm_vcpu_ready_for_interrupt_injection(vcpu)) { | |
4ca7dd8c PB |
9060 | r = 0; |
9061 | vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN; | |
09cec754 | 9062 | ++vcpu->stat.request_irq_exits; |
362c698f | 9063 | break; |
09cec754 | 9064 | } |
af585b92 | 9065 | |
f3020b88 | 9066 | if (__xfer_to_guest_mode_work_pending()) { |
f656ce01 | 9067 | srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); |
72c3c0fe TG |
9068 | r = xfer_to_guest_mode_handle_work(vcpu); |
9069 | if (r) | |
9070 | return r; | |
f656ce01 | 9071 | vcpu->srcu_idx = srcu_read_lock(&kvm->srcu); |
d7690175 | 9072 | } |
b6c7a5dc HB |
9073 | } |
9074 | ||
f656ce01 | 9075 | srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); |
b6c7a5dc HB |
9076 | |
9077 | return r; | |
9078 | } | |
9079 | ||
716d51ab GN |
9080 | static inline int complete_emulated_io(struct kvm_vcpu *vcpu) |
9081 | { | |
9082 | int r; | |
60fc3d02 | 9083 | |
716d51ab | 9084 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); |
0ce97a2b | 9085 | r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE); |
716d51ab | 9086 | srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); |
60fc3d02 | 9087 | return r; |
716d51ab GN |
9088 | } |
9089 | ||
9090 | static int complete_emulated_pio(struct kvm_vcpu *vcpu) | |
9091 | { | |
9092 | BUG_ON(!vcpu->arch.pio.count); | |
9093 | ||
9094 | return complete_emulated_io(vcpu); | |
9095 | } | |
9096 | ||
f78146b0 AK |
9097 | /* |
9098 | * Implements the following, as a state machine: | |
9099 | * | |
9100 | * read: | |
9101 | * for each fragment | |
87da7e66 XG |
9102 | * for each mmio piece in the fragment |
9103 | * write gpa, len | |
9104 | * exit | |
9105 | * copy data | |
f78146b0 AK |
9106 | * execute insn |
9107 | * | |
9108 | * write: | |
9109 | * for each fragment | |
87da7e66 XG |
9110 | * for each mmio piece in the fragment |
9111 | * write gpa, len | |
9112 | * copy data | |
9113 | * exit | |
f78146b0 | 9114 | */ |
716d51ab | 9115 | static int complete_emulated_mmio(struct kvm_vcpu *vcpu) |
5287f194 AK |
9116 | { |
9117 | struct kvm_run *run = vcpu->run; | |
f78146b0 | 9118 | struct kvm_mmio_fragment *frag; |
87da7e66 | 9119 | unsigned len; |
5287f194 | 9120 | |
716d51ab | 9121 | BUG_ON(!vcpu->mmio_needed); |
5287f194 | 9122 | |
716d51ab | 9123 | /* Complete previous fragment */ |
87da7e66 XG |
9124 | frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment]; |
9125 | len = min(8u, frag->len); | |
716d51ab | 9126 | if (!vcpu->mmio_is_write) |
87da7e66 XG |
9127 | memcpy(frag->data, run->mmio.data, len); |
9128 | ||
9129 | if (frag->len <= 8) { | |
9130 | /* Switch to the next fragment. */ | |
9131 | frag++; | |
9132 | vcpu->mmio_cur_fragment++; | |
9133 | } else { | |
9134 | /* Go forward to the next mmio piece. */ | |
9135 | frag->data += len; | |
9136 | frag->gpa += len; | |
9137 | frag->len -= len; | |
9138 | } | |
9139 | ||
a08d3b3b | 9140 | if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) { |
716d51ab | 9141 | vcpu->mmio_needed = 0; |
0912c977 PB |
9142 | |
9143 | /* FIXME: return into emulator if single-stepping. */ | |
cef4dea0 | 9144 | if (vcpu->mmio_is_write) |
716d51ab GN |
9145 | return 1; |
9146 | vcpu->mmio_read_completed = 1; | |
9147 | return complete_emulated_io(vcpu); | |
9148 | } | |
87da7e66 | 9149 | |
716d51ab GN |
9150 | run->exit_reason = KVM_EXIT_MMIO; |
9151 | run->mmio.phys_addr = frag->gpa; | |
9152 | if (vcpu->mmio_is_write) | |
87da7e66 XG |
9153 | memcpy(run->mmio.data, frag->data, min(8u, frag->len)); |
9154 | run->mmio.len = min(8u, frag->len); | |
716d51ab GN |
9155 | run->mmio.is_write = vcpu->mmio_is_write; |
9156 | vcpu->arch.complete_userspace_io = complete_emulated_mmio; | |
9157 | return 0; | |
5287f194 AK |
9158 | } |
9159 | ||
c9aef3b8 SC |
9160 | static void kvm_save_current_fpu(struct fpu *fpu) |
9161 | { | |
9162 | /* | |
9163 | * If the target FPU state is not resident in the CPU registers, just | |
9164 | * memcpy() from current, else save CPU state directly to the target. | |
9165 | */ | |
9166 | if (test_thread_flag(TIF_NEED_FPU_LOAD)) | |
9167 | memcpy(&fpu->state, ¤t->thread.fpu.state, | |
9168 | fpu_kernel_xstate_size); | |
9169 | else | |
9170 | copy_fpregs_to_fpstate(fpu); | |
9171 | } | |
9172 | ||
822f312d SAS |
9173 | /* Swap (qemu) user FPU context for the guest FPU context. */ |
9174 | static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu) | |
9175 | { | |
5f409e20 RR |
9176 | fpregs_lock(); |
9177 | ||
c9aef3b8 SC |
9178 | kvm_save_current_fpu(vcpu->arch.user_fpu); |
9179 | ||
afaf0b2f | 9180 | /* PKRU is separately restored in kvm_x86_ops.run. */ |
b666a4b6 | 9181 | __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state, |
822f312d | 9182 | ~XFEATURE_MASK_PKRU); |
5f409e20 RR |
9183 | |
9184 | fpregs_mark_activate(); | |
9185 | fpregs_unlock(); | |
9186 | ||
822f312d SAS |
9187 | trace_kvm_fpu(1); |
9188 | } | |
9189 | ||
9190 | /* When vcpu_run ends, restore user space FPU context. */ | |
9191 | static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu) | |
9192 | { | |
5f409e20 RR |
9193 | fpregs_lock(); |
9194 | ||
c9aef3b8 SC |
9195 | kvm_save_current_fpu(vcpu->arch.guest_fpu); |
9196 | ||
d9a710e5 | 9197 | copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state); |
5f409e20 RR |
9198 | |
9199 | fpregs_mark_activate(); | |
9200 | fpregs_unlock(); | |
9201 | ||
822f312d SAS |
9202 | ++vcpu->stat.fpu_reload; |
9203 | trace_kvm_fpu(0); | |
9204 | } | |
9205 | ||
1b94f6f8 | 9206 | int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu) |
b6c7a5dc | 9207 | { |
1b94f6f8 | 9208 | struct kvm_run *kvm_run = vcpu->run; |
b6c7a5dc | 9209 | int r; |
b6c7a5dc | 9210 | |
accb757d | 9211 | vcpu_load(vcpu); |
20b7035c | 9212 | kvm_sigset_activate(vcpu); |
5663d8f9 PX |
9213 | kvm_load_guest_fpu(vcpu); |
9214 | ||
a4535290 | 9215 | if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) { |
2f173d26 JS |
9216 | if (kvm_run->immediate_exit) { |
9217 | r = -EINTR; | |
9218 | goto out; | |
9219 | } | |
b6c7a5dc | 9220 | kvm_vcpu_block(vcpu); |
66450a21 | 9221 | kvm_apic_accept_events(vcpu); |
72875d8a | 9222 | kvm_clear_request(KVM_REQ_UNHALT, vcpu); |
ac9f6dc0 | 9223 | r = -EAGAIN; |
a0595000 JS |
9224 | if (signal_pending(current)) { |
9225 | r = -EINTR; | |
1b94f6f8 | 9226 | kvm_run->exit_reason = KVM_EXIT_INTR; |
a0595000 JS |
9227 | ++vcpu->stat.signal_exits; |
9228 | } | |
ac9f6dc0 | 9229 | goto out; |
b6c7a5dc HB |
9230 | } |
9231 | ||
1b94f6f8 | 9232 | if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) { |
01643c51 KH |
9233 | r = -EINVAL; |
9234 | goto out; | |
9235 | } | |
9236 | ||
1b94f6f8 | 9237 | if (kvm_run->kvm_dirty_regs) { |
01643c51 KH |
9238 | r = sync_regs(vcpu); |
9239 | if (r != 0) | |
9240 | goto out; | |
9241 | } | |
9242 | ||
b6c7a5dc | 9243 | /* re-sync apic's tpr */ |
35754c98 | 9244 | if (!lapic_in_kernel(vcpu)) { |
eea1cff9 AP |
9245 | if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) { |
9246 | r = -EINVAL; | |
9247 | goto out; | |
9248 | } | |
9249 | } | |
b6c7a5dc | 9250 | |
716d51ab GN |
9251 | if (unlikely(vcpu->arch.complete_userspace_io)) { |
9252 | int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io; | |
9253 | vcpu->arch.complete_userspace_io = NULL; | |
9254 | r = cui(vcpu); | |
9255 | if (r <= 0) | |
5663d8f9 | 9256 | goto out; |
716d51ab GN |
9257 | } else |
9258 | WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed); | |
5287f194 | 9259 | |
460df4c1 PB |
9260 | if (kvm_run->immediate_exit) |
9261 | r = -EINTR; | |
9262 | else | |
9263 | r = vcpu_run(vcpu); | |
b6c7a5dc HB |
9264 | |
9265 | out: | |
5663d8f9 | 9266 | kvm_put_guest_fpu(vcpu); |
1b94f6f8 | 9267 | if (kvm_run->kvm_valid_regs) |
01643c51 | 9268 | store_regs(vcpu); |
f1d86e46 | 9269 | post_kvm_run_save(vcpu); |
20b7035c | 9270 | kvm_sigset_deactivate(vcpu); |
b6c7a5dc | 9271 | |
accb757d | 9272 | vcpu_put(vcpu); |
b6c7a5dc HB |
9273 | return r; |
9274 | } | |
9275 | ||
01643c51 | 9276 | static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) |
b6c7a5dc | 9277 | { |
7ae441ea GN |
9278 | if (vcpu->arch.emulate_regs_need_sync_to_vcpu) { |
9279 | /* | |
9280 | * We are here if userspace calls get_regs() in the middle of | |
9281 | * instruction emulation. Registers state needs to be copied | |
4a969980 | 9282 | * back from emulation context to vcpu. Userspace shouldn't do |
7ae441ea GN |
9283 | * that usually, but some bad designed PV devices (vmware |
9284 | * backdoor interface) need this to work | |
9285 | */ | |
c9b8b07c | 9286 | emulator_writeback_register_cache(vcpu->arch.emulate_ctxt); |
7ae441ea GN |
9287 | vcpu->arch.emulate_regs_need_sync_to_vcpu = false; |
9288 | } | |
de3cd117 SC |
9289 | regs->rax = kvm_rax_read(vcpu); |
9290 | regs->rbx = kvm_rbx_read(vcpu); | |
9291 | regs->rcx = kvm_rcx_read(vcpu); | |
9292 | regs->rdx = kvm_rdx_read(vcpu); | |
9293 | regs->rsi = kvm_rsi_read(vcpu); | |
9294 | regs->rdi = kvm_rdi_read(vcpu); | |
e9c16c78 | 9295 | regs->rsp = kvm_rsp_read(vcpu); |
de3cd117 | 9296 | regs->rbp = kvm_rbp_read(vcpu); |
b6c7a5dc | 9297 | #ifdef CONFIG_X86_64 |
de3cd117 SC |
9298 | regs->r8 = kvm_r8_read(vcpu); |
9299 | regs->r9 = kvm_r9_read(vcpu); | |
9300 | regs->r10 = kvm_r10_read(vcpu); | |
9301 | regs->r11 = kvm_r11_read(vcpu); | |
9302 | regs->r12 = kvm_r12_read(vcpu); | |
9303 | regs->r13 = kvm_r13_read(vcpu); | |
9304 | regs->r14 = kvm_r14_read(vcpu); | |
9305 | regs->r15 = kvm_r15_read(vcpu); | |
b6c7a5dc HB |
9306 | #endif |
9307 | ||
5fdbf976 | 9308 | regs->rip = kvm_rip_read(vcpu); |
91586a3b | 9309 | regs->rflags = kvm_get_rflags(vcpu); |
01643c51 | 9310 | } |
b6c7a5dc | 9311 | |
01643c51 KH |
9312 | int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) |
9313 | { | |
9314 | vcpu_load(vcpu); | |
9315 | __get_regs(vcpu, regs); | |
1fc9b76b | 9316 | vcpu_put(vcpu); |
b6c7a5dc HB |
9317 | return 0; |
9318 | } | |
9319 | ||
01643c51 | 9320 | static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) |
b6c7a5dc | 9321 | { |
7ae441ea GN |
9322 | vcpu->arch.emulate_regs_need_sync_from_vcpu = true; |
9323 | vcpu->arch.emulate_regs_need_sync_to_vcpu = false; | |
9324 | ||
de3cd117 SC |
9325 | kvm_rax_write(vcpu, regs->rax); |
9326 | kvm_rbx_write(vcpu, regs->rbx); | |
9327 | kvm_rcx_write(vcpu, regs->rcx); | |
9328 | kvm_rdx_write(vcpu, regs->rdx); | |
9329 | kvm_rsi_write(vcpu, regs->rsi); | |
9330 | kvm_rdi_write(vcpu, regs->rdi); | |
e9c16c78 | 9331 | kvm_rsp_write(vcpu, regs->rsp); |
de3cd117 | 9332 | kvm_rbp_write(vcpu, regs->rbp); |
b6c7a5dc | 9333 | #ifdef CONFIG_X86_64 |
de3cd117 SC |
9334 | kvm_r8_write(vcpu, regs->r8); |
9335 | kvm_r9_write(vcpu, regs->r9); | |
9336 | kvm_r10_write(vcpu, regs->r10); | |
9337 | kvm_r11_write(vcpu, regs->r11); | |
9338 | kvm_r12_write(vcpu, regs->r12); | |
9339 | kvm_r13_write(vcpu, regs->r13); | |
9340 | kvm_r14_write(vcpu, regs->r14); | |
9341 | kvm_r15_write(vcpu, regs->r15); | |
b6c7a5dc HB |
9342 | #endif |
9343 | ||
5fdbf976 | 9344 | kvm_rip_write(vcpu, regs->rip); |
d73235d1 | 9345 | kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED); |
b6c7a5dc | 9346 | |
b4f14abd JK |
9347 | vcpu->arch.exception.pending = false; |
9348 | ||
3842d135 | 9349 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
01643c51 | 9350 | } |
3842d135 | 9351 | |
01643c51 KH |
9352 | int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) |
9353 | { | |
9354 | vcpu_load(vcpu); | |
9355 | __set_regs(vcpu, regs); | |
875656fe | 9356 | vcpu_put(vcpu); |
b6c7a5dc HB |
9357 | return 0; |
9358 | } | |
9359 | ||
b6c7a5dc HB |
9360 | void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l) |
9361 | { | |
9362 | struct kvm_segment cs; | |
9363 | ||
3e6e0aab | 9364 | kvm_get_segment(vcpu, &cs, VCPU_SREG_CS); |
b6c7a5dc HB |
9365 | *db = cs.db; |
9366 | *l = cs.l; | |
9367 | } | |
9368 | EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits); | |
9369 | ||
01643c51 | 9370 | static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs) |
b6c7a5dc | 9371 | { |
89a27f4d | 9372 | struct desc_ptr dt; |
b6c7a5dc | 9373 | |
3e6e0aab GT |
9374 | kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS); |
9375 | kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS); | |
9376 | kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES); | |
9377 | kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS); | |
9378 | kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS); | |
9379 | kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS); | |
b6c7a5dc | 9380 | |
3e6e0aab GT |
9381 | kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR); |
9382 | kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR); | |
b6c7a5dc | 9383 | |
afaf0b2f | 9384 | kvm_x86_ops.get_idt(vcpu, &dt); |
89a27f4d GN |
9385 | sregs->idt.limit = dt.size; |
9386 | sregs->idt.base = dt.address; | |
afaf0b2f | 9387 | kvm_x86_ops.get_gdt(vcpu, &dt); |
89a27f4d GN |
9388 | sregs->gdt.limit = dt.size; |
9389 | sregs->gdt.base = dt.address; | |
b6c7a5dc | 9390 | |
4d4ec087 | 9391 | sregs->cr0 = kvm_read_cr0(vcpu); |
ad312c7c | 9392 | sregs->cr2 = vcpu->arch.cr2; |
9f8fe504 | 9393 | sregs->cr3 = kvm_read_cr3(vcpu); |
fc78f519 | 9394 | sregs->cr4 = kvm_read_cr4(vcpu); |
2d3ad1f4 | 9395 | sregs->cr8 = kvm_get_cr8(vcpu); |
f6801dff | 9396 | sregs->efer = vcpu->arch.efer; |
b6c7a5dc HB |
9397 | sregs->apic_base = kvm_get_apic_base(vcpu); |
9398 | ||
0e96f31e | 9399 | memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap)); |
b6c7a5dc | 9400 | |
04140b41 | 9401 | if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft) |
14d0bc1f GN |
9402 | set_bit(vcpu->arch.interrupt.nr, |
9403 | (unsigned long *)sregs->interrupt_bitmap); | |
01643c51 | 9404 | } |
16d7a191 | 9405 | |
01643c51 KH |
9406 | int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu, |
9407 | struct kvm_sregs *sregs) | |
9408 | { | |
9409 | vcpu_load(vcpu); | |
9410 | __get_sregs(vcpu, sregs); | |
bcdec41c | 9411 | vcpu_put(vcpu); |
b6c7a5dc HB |
9412 | return 0; |
9413 | } | |
9414 | ||
62d9f0db MT |
9415 | int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, |
9416 | struct kvm_mp_state *mp_state) | |
9417 | { | |
fd232561 | 9418 | vcpu_load(vcpu); |
f958bd23 SC |
9419 | if (kvm_mpx_supported()) |
9420 | kvm_load_guest_fpu(vcpu); | |
fd232561 | 9421 | |
66450a21 | 9422 | kvm_apic_accept_events(vcpu); |
6aef266c SV |
9423 | if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED && |
9424 | vcpu->arch.pv.pv_unhalted) | |
9425 | mp_state->mp_state = KVM_MP_STATE_RUNNABLE; | |
9426 | else | |
9427 | mp_state->mp_state = vcpu->arch.mp_state; | |
9428 | ||
f958bd23 SC |
9429 | if (kvm_mpx_supported()) |
9430 | kvm_put_guest_fpu(vcpu); | |
fd232561 | 9431 | vcpu_put(vcpu); |
62d9f0db MT |
9432 | return 0; |
9433 | } | |
9434 | ||
9435 | int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, | |
9436 | struct kvm_mp_state *mp_state) | |
9437 | { | |
e83dff5e CD |
9438 | int ret = -EINVAL; |
9439 | ||
9440 | vcpu_load(vcpu); | |
9441 | ||
bce87cce | 9442 | if (!lapic_in_kernel(vcpu) && |
66450a21 | 9443 | mp_state->mp_state != KVM_MP_STATE_RUNNABLE) |
e83dff5e | 9444 | goto out; |
66450a21 | 9445 | |
27cbe7d6 LA |
9446 | /* |
9447 | * KVM_MP_STATE_INIT_RECEIVED means the processor is in | |
9448 | * INIT state; latched init should be reported using | |
9449 | * KVM_SET_VCPU_EVENTS, so reject it here. | |
9450 | */ | |
9451 | if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) && | |
28bf2888 DH |
9452 | (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED || |
9453 | mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED)) | |
e83dff5e | 9454 | goto out; |
28bf2888 | 9455 | |
66450a21 JK |
9456 | if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) { |
9457 | vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED; | |
9458 | set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events); | |
9459 | } else | |
9460 | vcpu->arch.mp_state = mp_state->mp_state; | |
3842d135 | 9461 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
e83dff5e CD |
9462 | |
9463 | ret = 0; | |
9464 | out: | |
9465 | vcpu_put(vcpu); | |
9466 | return ret; | |
62d9f0db MT |
9467 | } |
9468 | ||
7f3d35fd KW |
9469 | int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index, |
9470 | int reason, bool has_error_code, u32 error_code) | |
b6c7a5dc | 9471 | { |
c9b8b07c | 9472 | struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt; |
8ec4722d | 9473 | int ret; |
e01c2426 | 9474 | |
8ec4722d | 9475 | init_emulate_ctxt(vcpu); |
c697518a | 9476 | |
7f3d35fd | 9477 | ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason, |
9d74191a | 9478 | has_error_code, error_code); |
1051778f SC |
9479 | if (ret) { |
9480 | vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
9481 | vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
9482 | vcpu->run->internal.ndata = 0; | |
60fc3d02 | 9483 | return 0; |
1051778f | 9484 | } |
37817f29 | 9485 | |
9d74191a TY |
9486 | kvm_rip_write(vcpu, ctxt->eip); |
9487 | kvm_set_rflags(vcpu, ctxt->eflags); | |
60fc3d02 | 9488 | return 1; |
37817f29 IE |
9489 | } |
9490 | EXPORT_SYMBOL_GPL(kvm_task_switch); | |
9491 | ||
3140c156 | 9492 | static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs) |
f2981033 | 9493 | { |
37b95951 | 9494 | if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) { |
f2981033 LT |
9495 | /* |
9496 | * When EFER.LME and CR0.PG are set, the processor is in | |
9497 | * 64-bit mode (though maybe in a 32-bit code segment). | |
9498 | * CR4.PAE and EFER.LMA must be set. | |
9499 | */ | |
37b95951 | 9500 | if (!(sregs->cr4 & X86_CR4_PAE) |
f2981033 LT |
9501 | || !(sregs->efer & EFER_LMA)) |
9502 | return -EINVAL; | |
9503 | } else { | |
9504 | /* | |
9505 | * Not in 64-bit mode: EFER.LMA is clear and the code | |
9506 | * segment cannot be 64-bit. | |
9507 | */ | |
9508 | if (sregs->efer & EFER_LMA || sregs->cs.l) | |
9509 | return -EINVAL; | |
9510 | } | |
9511 | ||
3ca94192 | 9512 | return kvm_valid_cr4(vcpu, sregs->cr4); |
f2981033 LT |
9513 | } |
9514 | ||
01643c51 | 9515 | static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs) |
b6c7a5dc | 9516 | { |
58cb628d | 9517 | struct msr_data apic_base_msr; |
b6c7a5dc | 9518 | int mmu_reset_needed = 0; |
c4d21882 | 9519 | int cpuid_update_needed = 0; |
63f42e02 | 9520 | int pending_vec, max_bits, idx; |
89a27f4d | 9521 | struct desc_ptr dt; |
b4ef9d4e CD |
9522 | int ret = -EINVAL; |
9523 | ||
f2981033 | 9524 | if (kvm_valid_sregs(vcpu, sregs)) |
8dbfb2bf | 9525 | goto out; |
f2981033 | 9526 | |
d3802286 JM |
9527 | apic_base_msr.data = sregs->apic_base; |
9528 | apic_base_msr.host_initiated = true; | |
9529 | if (kvm_set_apic_base(vcpu, &apic_base_msr)) | |
b4ef9d4e | 9530 | goto out; |
6d1068b3 | 9531 | |
89a27f4d GN |
9532 | dt.size = sregs->idt.limit; |
9533 | dt.address = sregs->idt.base; | |
afaf0b2f | 9534 | kvm_x86_ops.set_idt(vcpu, &dt); |
89a27f4d GN |
9535 | dt.size = sregs->gdt.limit; |
9536 | dt.address = sregs->gdt.base; | |
afaf0b2f | 9537 | kvm_x86_ops.set_gdt(vcpu, &dt); |
b6c7a5dc | 9538 | |
ad312c7c | 9539 | vcpu->arch.cr2 = sregs->cr2; |
9f8fe504 | 9540 | mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3; |
dc7e795e | 9541 | vcpu->arch.cr3 = sregs->cr3; |
cb3c1e2f | 9542 | kvm_register_mark_available(vcpu, VCPU_EXREG_CR3); |
b6c7a5dc | 9543 | |
2d3ad1f4 | 9544 | kvm_set_cr8(vcpu, sregs->cr8); |
b6c7a5dc | 9545 | |
f6801dff | 9546 | mmu_reset_needed |= vcpu->arch.efer != sregs->efer; |
afaf0b2f | 9547 | kvm_x86_ops.set_efer(vcpu, sregs->efer); |
b6c7a5dc | 9548 | |
4d4ec087 | 9549 | mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0; |
afaf0b2f | 9550 | kvm_x86_ops.set_cr0(vcpu, sregs->cr0); |
d7306163 | 9551 | vcpu->arch.cr0 = sregs->cr0; |
b6c7a5dc | 9552 | |
fc78f519 | 9553 | mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4; |
c4d21882 WH |
9554 | cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) & |
9555 | (X86_CR4_OSXSAVE | X86_CR4_PKE)); | |
afaf0b2f | 9556 | kvm_x86_ops.set_cr4(vcpu, sregs->cr4); |
c4d21882 | 9557 | if (cpuid_update_needed) |
aedbaf4f | 9558 | kvm_update_cpuid_runtime(vcpu); |
63f42e02 XG |
9559 | |
9560 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
bf03d4f9 | 9561 | if (is_pae_paging(vcpu)) { |
9f8fe504 | 9562 | load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)); |
7c93be44 MT |
9563 | mmu_reset_needed = 1; |
9564 | } | |
63f42e02 | 9565 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
b6c7a5dc HB |
9566 | |
9567 | if (mmu_reset_needed) | |
9568 | kvm_mmu_reset_context(vcpu); | |
9569 | ||
a50abc3b | 9570 | max_bits = KVM_NR_INTERRUPTS; |
923c61bb GN |
9571 | pending_vec = find_first_bit( |
9572 | (const unsigned long *)sregs->interrupt_bitmap, max_bits); | |
9573 | if (pending_vec < max_bits) { | |
66fd3f7f | 9574 | kvm_queue_interrupt(vcpu, pending_vec, false); |
923c61bb | 9575 | pr_debug("Set back pending irq %d\n", pending_vec); |
b6c7a5dc HB |
9576 | } |
9577 | ||
3e6e0aab GT |
9578 | kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS); |
9579 | kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS); | |
9580 | kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES); | |
9581 | kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS); | |
9582 | kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS); | |
9583 | kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS); | |
b6c7a5dc | 9584 | |
3e6e0aab GT |
9585 | kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR); |
9586 | kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR); | |
b6c7a5dc | 9587 | |
5f0269f5 ME |
9588 | update_cr8_intercept(vcpu); |
9589 | ||
9c3e4aab | 9590 | /* Older userspace won't unhalt the vcpu on reset. */ |
c5af89b6 | 9591 | if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 && |
9c3e4aab | 9592 | sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 && |
3eeb3288 | 9593 | !is_protmode(vcpu)) |
9c3e4aab MT |
9594 | vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; |
9595 | ||
3842d135 AK |
9596 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
9597 | ||
b4ef9d4e CD |
9598 | ret = 0; |
9599 | out: | |
01643c51 KH |
9600 | return ret; |
9601 | } | |
9602 | ||
9603 | int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu, | |
9604 | struct kvm_sregs *sregs) | |
9605 | { | |
9606 | int ret; | |
9607 | ||
9608 | vcpu_load(vcpu); | |
9609 | ret = __set_sregs(vcpu, sregs); | |
b4ef9d4e CD |
9610 | vcpu_put(vcpu); |
9611 | return ret; | |
b6c7a5dc HB |
9612 | } |
9613 | ||
d0bfb940 JK |
9614 | int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu, |
9615 | struct kvm_guest_debug *dbg) | |
b6c7a5dc | 9616 | { |
355be0b9 | 9617 | unsigned long rflags; |
ae675ef0 | 9618 | int i, r; |
b6c7a5dc | 9619 | |
66b56562 CD |
9620 | vcpu_load(vcpu); |
9621 | ||
4f926bf2 JK |
9622 | if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) { |
9623 | r = -EBUSY; | |
9624 | if (vcpu->arch.exception.pending) | |
2122ff5e | 9625 | goto out; |
4f926bf2 JK |
9626 | if (dbg->control & KVM_GUESTDBG_INJECT_DB) |
9627 | kvm_queue_exception(vcpu, DB_VECTOR); | |
9628 | else | |
9629 | kvm_queue_exception(vcpu, BP_VECTOR); | |
9630 | } | |
9631 | ||
91586a3b JK |
9632 | /* |
9633 | * Read rflags as long as potentially injected trace flags are still | |
9634 | * filtered out. | |
9635 | */ | |
9636 | rflags = kvm_get_rflags(vcpu); | |
355be0b9 JK |
9637 | |
9638 | vcpu->guest_debug = dbg->control; | |
9639 | if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE)) | |
9640 | vcpu->guest_debug = 0; | |
9641 | ||
9642 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) { | |
ae675ef0 JK |
9643 | for (i = 0; i < KVM_NR_DB_REGS; ++i) |
9644 | vcpu->arch.eff_db[i] = dbg->arch.debugreg[i]; | |
c8639010 | 9645 | vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7]; |
ae675ef0 JK |
9646 | } else { |
9647 | for (i = 0; i < KVM_NR_DB_REGS; i++) | |
9648 | vcpu->arch.eff_db[i] = vcpu->arch.db[i]; | |
ae675ef0 | 9649 | } |
c8639010 | 9650 | kvm_update_dr7(vcpu); |
ae675ef0 | 9651 | |
f92653ee JK |
9652 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) |
9653 | vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) + | |
9654 | get_segment_base(vcpu, VCPU_SREG_CS); | |
94fe45da | 9655 | |
91586a3b JK |
9656 | /* |
9657 | * Trigger an rflags update that will inject or remove the trace | |
9658 | * flags. | |
9659 | */ | |
9660 | kvm_set_rflags(vcpu, rflags); | |
b6c7a5dc | 9661 | |
6986982f | 9662 | kvm_x86_ops.update_exception_bitmap(vcpu); |
b6c7a5dc | 9663 | |
4f926bf2 | 9664 | r = 0; |
d0bfb940 | 9665 | |
2122ff5e | 9666 | out: |
66b56562 | 9667 | vcpu_put(vcpu); |
b6c7a5dc HB |
9668 | return r; |
9669 | } | |
9670 | ||
8b006791 ZX |
9671 | /* |
9672 | * Translate a guest virtual address to a guest physical address. | |
9673 | */ | |
9674 | int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu, | |
9675 | struct kvm_translation *tr) | |
9676 | { | |
9677 | unsigned long vaddr = tr->linear_address; | |
9678 | gpa_t gpa; | |
f656ce01 | 9679 | int idx; |
8b006791 | 9680 | |
1da5b61d CD |
9681 | vcpu_load(vcpu); |
9682 | ||
f656ce01 | 9683 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
1871c602 | 9684 | gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL); |
f656ce01 | 9685 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
8b006791 ZX |
9686 | tr->physical_address = gpa; |
9687 | tr->valid = gpa != UNMAPPED_GVA; | |
9688 | tr->writeable = 1; | |
9689 | tr->usermode = 0; | |
8b006791 | 9690 | |
1da5b61d | 9691 | vcpu_put(vcpu); |
8b006791 ZX |
9692 | return 0; |
9693 | } | |
9694 | ||
d0752060 HB |
9695 | int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) |
9696 | { | |
1393123e | 9697 | struct fxregs_state *fxsave; |
d0752060 | 9698 | |
1393123e | 9699 | vcpu_load(vcpu); |
d0752060 | 9700 | |
b666a4b6 | 9701 | fxsave = &vcpu->arch.guest_fpu->state.fxsave; |
d0752060 HB |
9702 | memcpy(fpu->fpr, fxsave->st_space, 128); |
9703 | fpu->fcw = fxsave->cwd; | |
9704 | fpu->fsw = fxsave->swd; | |
9705 | fpu->ftwx = fxsave->twd; | |
9706 | fpu->last_opcode = fxsave->fop; | |
9707 | fpu->last_ip = fxsave->rip; | |
9708 | fpu->last_dp = fxsave->rdp; | |
0e96f31e | 9709 | memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space)); |
d0752060 | 9710 | |
1393123e | 9711 | vcpu_put(vcpu); |
d0752060 HB |
9712 | return 0; |
9713 | } | |
9714 | ||
9715 | int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) | |
9716 | { | |
6a96bc7f CD |
9717 | struct fxregs_state *fxsave; |
9718 | ||
9719 | vcpu_load(vcpu); | |
9720 | ||
b666a4b6 | 9721 | fxsave = &vcpu->arch.guest_fpu->state.fxsave; |
d0752060 | 9722 | |
d0752060 HB |
9723 | memcpy(fxsave->st_space, fpu->fpr, 128); |
9724 | fxsave->cwd = fpu->fcw; | |
9725 | fxsave->swd = fpu->fsw; | |
9726 | fxsave->twd = fpu->ftwx; | |
9727 | fxsave->fop = fpu->last_opcode; | |
9728 | fxsave->rip = fpu->last_ip; | |
9729 | fxsave->rdp = fpu->last_dp; | |
0e96f31e | 9730 | memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space)); |
d0752060 | 9731 | |
6a96bc7f | 9732 | vcpu_put(vcpu); |
d0752060 HB |
9733 | return 0; |
9734 | } | |
9735 | ||
01643c51 KH |
9736 | static void store_regs(struct kvm_vcpu *vcpu) |
9737 | { | |
9738 | BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES); | |
9739 | ||
9740 | if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS) | |
9741 | __get_regs(vcpu, &vcpu->run->s.regs.regs); | |
9742 | ||
9743 | if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS) | |
9744 | __get_sregs(vcpu, &vcpu->run->s.regs.sregs); | |
9745 | ||
9746 | if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS) | |
9747 | kvm_vcpu_ioctl_x86_get_vcpu_events( | |
9748 | vcpu, &vcpu->run->s.regs.events); | |
9749 | } | |
9750 | ||
9751 | static int sync_regs(struct kvm_vcpu *vcpu) | |
9752 | { | |
9753 | if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS) | |
9754 | return -EINVAL; | |
9755 | ||
9756 | if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) { | |
9757 | __set_regs(vcpu, &vcpu->run->s.regs.regs); | |
9758 | vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS; | |
9759 | } | |
9760 | if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) { | |
9761 | if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs)) | |
9762 | return -EINVAL; | |
9763 | vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS; | |
9764 | } | |
9765 | if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) { | |
9766 | if (kvm_vcpu_ioctl_x86_set_vcpu_events( | |
9767 | vcpu, &vcpu->run->s.regs.events)) | |
9768 | return -EINVAL; | |
9769 | vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS; | |
9770 | } | |
9771 | ||
9772 | return 0; | |
9773 | } | |
9774 | ||
0ee6a517 | 9775 | static void fx_init(struct kvm_vcpu *vcpu) |
d0752060 | 9776 | { |
b666a4b6 | 9777 | fpstate_init(&vcpu->arch.guest_fpu->state); |
782511b0 | 9778 | if (boot_cpu_has(X86_FEATURE_XSAVES)) |
b666a4b6 | 9779 | vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv = |
df1daba7 | 9780 | host_xcr0 | XSTATE_COMPACTION_ENABLED; |
d0752060 | 9781 | |
2acf923e DC |
9782 | /* |
9783 | * Ensure guest xcr0 is valid for loading | |
9784 | */ | |
d91cab78 | 9785 | vcpu->arch.xcr0 = XFEATURE_MASK_FP; |
2acf923e | 9786 | |
ad312c7c | 9787 | vcpu->arch.cr0 |= X86_CR0_ET; |
d0752060 | 9788 | } |
d0752060 | 9789 | |
897cc38e | 9790 | int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id) |
e9b11c17 | 9791 | { |
897cc38e SC |
9792 | if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0) |
9793 | pr_warn_once("kvm: SMP vm created on host with unstable TSC; " | |
9794 | "guest TSC will not be reliable\n"); | |
7f1ea208 | 9795 | |
897cc38e | 9796 | return 0; |
e9b11c17 ZX |
9797 | } |
9798 | ||
e529ef66 | 9799 | int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu) |
e9b11c17 | 9800 | { |
95a0d01e SC |
9801 | struct page *page; |
9802 | int r; | |
c447e76b | 9803 | |
95a0d01e SC |
9804 | if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu)) |
9805 | vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; | |
9806 | else | |
9807 | vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED; | |
c447e76b | 9808 | |
95a0d01e | 9809 | kvm_set_tsc_khz(vcpu, max_tsc_khz); |
c447e76b | 9810 | |
95a0d01e SC |
9811 | r = kvm_mmu_create(vcpu); |
9812 | if (r < 0) | |
9813 | return r; | |
9814 | ||
9815 | if (irqchip_in_kernel(vcpu->kvm)) { | |
95a0d01e SC |
9816 | r = kvm_create_lapic(vcpu, lapic_timer_advance_ns); |
9817 | if (r < 0) | |
9818 | goto fail_mmu_destroy; | |
4e19c36f SS |
9819 | if (kvm_apicv_activated(vcpu->kvm)) |
9820 | vcpu->arch.apicv_active = true; | |
95a0d01e SC |
9821 | } else |
9822 | static_key_slow_inc(&kvm_no_apic_vcpu); | |
9823 | ||
9824 | r = -ENOMEM; | |
9825 | ||
9826 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); | |
9827 | if (!page) | |
9828 | goto fail_free_lapic; | |
9829 | vcpu->arch.pio_data = page_address(page); | |
9830 | ||
9831 | vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4, | |
9832 | GFP_KERNEL_ACCOUNT); | |
9833 | if (!vcpu->arch.mce_banks) | |
9834 | goto fail_free_pio_data; | |
9835 | vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS; | |
9836 | ||
9837 | if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, | |
9838 | GFP_KERNEL_ACCOUNT)) | |
9839 | goto fail_free_mce_banks; | |
9840 | ||
c9b8b07c SC |
9841 | if (!alloc_emulate_ctxt(vcpu)) |
9842 | goto free_wbinvd_dirty_mask; | |
9843 | ||
95a0d01e SC |
9844 | vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache, |
9845 | GFP_KERNEL_ACCOUNT); | |
9846 | if (!vcpu->arch.user_fpu) { | |
9847 | pr_err("kvm: failed to allocate userspace's fpu\n"); | |
c9b8b07c | 9848 | goto free_emulate_ctxt; |
95a0d01e SC |
9849 | } |
9850 | ||
9851 | vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache, | |
9852 | GFP_KERNEL_ACCOUNT); | |
9853 | if (!vcpu->arch.guest_fpu) { | |
9854 | pr_err("kvm: failed to allocate vcpu's fpu\n"); | |
9855 | goto free_user_fpu; | |
9856 | } | |
9857 | fx_init(vcpu); | |
9858 | ||
95a0d01e SC |
9859 | vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu); |
9860 | ||
9861 | vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT; | |
9862 | ||
9863 | kvm_async_pf_hash_reset(vcpu); | |
9864 | kvm_pmu_init(vcpu); | |
9865 | ||
9866 | vcpu->arch.pending_external_vector = -1; | |
9867 | vcpu->arch.preempted_in_kernel = false; | |
9868 | ||
9869 | kvm_hv_vcpu_init(vcpu); | |
9870 | ||
afaf0b2f | 9871 | r = kvm_x86_ops.vcpu_create(vcpu); |
95a0d01e SC |
9872 | if (r) |
9873 | goto free_guest_fpu; | |
e9b11c17 | 9874 | |
0cf9135b | 9875 | vcpu->arch.arch_capabilities = kvm_get_arch_capabilities(); |
e53d88af | 9876 | vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT; |
19efffa2 | 9877 | kvm_vcpu_mtrr_init(vcpu); |
ec7660cc | 9878 | vcpu_load(vcpu); |
d28bc9dd | 9879 | kvm_vcpu_reset(vcpu, false); |
e1732991 | 9880 | kvm_init_mmu(vcpu, false); |
e9b11c17 | 9881 | vcpu_put(vcpu); |
ec7660cc | 9882 | return 0; |
95a0d01e SC |
9883 | |
9884 | free_guest_fpu: | |
9885 | kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu); | |
9886 | free_user_fpu: | |
9887 | kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu); | |
c9b8b07c SC |
9888 | free_emulate_ctxt: |
9889 | kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt); | |
95a0d01e SC |
9890 | free_wbinvd_dirty_mask: |
9891 | free_cpumask_var(vcpu->arch.wbinvd_dirty_mask); | |
9892 | fail_free_mce_banks: | |
9893 | kfree(vcpu->arch.mce_banks); | |
9894 | fail_free_pio_data: | |
9895 | free_page((unsigned long)vcpu->arch.pio_data); | |
9896 | fail_free_lapic: | |
9897 | kvm_free_lapic(vcpu); | |
9898 | fail_mmu_destroy: | |
9899 | kvm_mmu_destroy(vcpu); | |
9900 | return r; | |
e9b11c17 ZX |
9901 | } |
9902 | ||
31928aa5 | 9903 | void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) |
42897d86 | 9904 | { |
332967a3 | 9905 | struct kvm *kvm = vcpu->kvm; |
42897d86 | 9906 | |
d3457c87 RK |
9907 | kvm_hv_vcpu_postcreate(vcpu); |
9908 | ||
ec7660cc | 9909 | if (mutex_lock_killable(&vcpu->mutex)) |
31928aa5 | 9910 | return; |
ec7660cc | 9911 | vcpu_load(vcpu); |
0c899c25 | 9912 | kvm_synchronize_tsc(vcpu, 0); |
42897d86 | 9913 | vcpu_put(vcpu); |
2d5ba19b MT |
9914 | |
9915 | /* poll control enabled by default */ | |
9916 | vcpu->arch.msr_kvm_poll_control = 1; | |
9917 | ||
ec7660cc | 9918 | mutex_unlock(&vcpu->mutex); |
42897d86 | 9919 | |
b34de572 WL |
9920 | if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0) |
9921 | schedule_delayed_work(&kvm->arch.kvmclock_sync_work, | |
9922 | KVMCLOCK_SYNC_PERIOD); | |
42897d86 MT |
9923 | } |
9924 | ||
d40ccc62 | 9925 | void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) |
e9b11c17 | 9926 | { |
4cbc418a | 9927 | struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache; |
95a0d01e | 9928 | int idx; |
344d9588 | 9929 | |
4cbc418a PB |
9930 | kvm_release_pfn(cache->pfn, cache->dirty, cache); |
9931 | ||
50b143e1 | 9932 | kvmclock_reset(vcpu); |
e9b11c17 | 9933 | |
afaf0b2f | 9934 | kvm_x86_ops.vcpu_free(vcpu); |
50b143e1 | 9935 | |
c9b8b07c | 9936 | kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt); |
50b143e1 SC |
9937 | free_cpumask_var(vcpu->arch.wbinvd_dirty_mask); |
9938 | kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu); | |
9939 | kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu); | |
95a0d01e SC |
9940 | |
9941 | kvm_hv_vcpu_uninit(vcpu); | |
9942 | kvm_pmu_destroy(vcpu); | |
9943 | kfree(vcpu->arch.mce_banks); | |
9944 | kvm_free_lapic(vcpu); | |
9945 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
9946 | kvm_mmu_destroy(vcpu); | |
9947 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
9948 | free_page((unsigned long)vcpu->arch.pio_data); | |
255cbecf | 9949 | kvfree(vcpu->arch.cpuid_entries); |
95a0d01e SC |
9950 | if (!lapic_in_kernel(vcpu)) |
9951 | static_key_slow_dec(&kvm_no_apic_vcpu); | |
e9b11c17 ZX |
9952 | } |
9953 | ||
d28bc9dd | 9954 | void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event) |
e9b11c17 | 9955 | { |
b7e31be3 RK |
9956 | kvm_lapic_reset(vcpu, init_event); |
9957 | ||
e69fab5d PB |
9958 | vcpu->arch.hflags = 0; |
9959 | ||
c43203ca | 9960 | vcpu->arch.smi_pending = 0; |
52797bf9 | 9961 | vcpu->arch.smi_count = 0; |
7460fb4a AK |
9962 | atomic_set(&vcpu->arch.nmi_queued, 0); |
9963 | vcpu->arch.nmi_pending = 0; | |
448fa4a9 | 9964 | vcpu->arch.nmi_injected = false; |
5f7552d4 NA |
9965 | kvm_clear_interrupt_queue(vcpu); |
9966 | kvm_clear_exception_queue(vcpu); | |
448fa4a9 | 9967 | |
42dbaa5a | 9968 | memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db)); |
ae561ede | 9969 | kvm_update_dr0123(vcpu); |
6f43ed01 | 9970 | vcpu->arch.dr6 = DR6_INIT; |
42dbaa5a | 9971 | vcpu->arch.dr7 = DR7_FIXED_1; |
c8639010 | 9972 | kvm_update_dr7(vcpu); |
42dbaa5a | 9973 | |
1119022c NA |
9974 | vcpu->arch.cr2 = 0; |
9975 | ||
3842d135 | 9976 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
2635b5c4 VK |
9977 | vcpu->arch.apf.msr_en_val = 0; |
9978 | vcpu->arch.apf.msr_int_val = 0; | |
c9aaa895 | 9979 | vcpu->arch.st.msr_val = 0; |
3842d135 | 9980 | |
12f9a48f GC |
9981 | kvmclock_reset(vcpu); |
9982 | ||
af585b92 GN |
9983 | kvm_clear_async_pf_completion_queue(vcpu); |
9984 | kvm_async_pf_hash_reset(vcpu); | |
9985 | vcpu->arch.apf.halted = false; | |
3842d135 | 9986 | |
a554d207 WL |
9987 | if (kvm_mpx_supported()) { |
9988 | void *mpx_state_buffer; | |
9989 | ||
9990 | /* | |
9991 | * To avoid have the INIT path from kvm_apic_has_events() that be | |
9992 | * called with loaded FPU and does not let userspace fix the state. | |
9993 | */ | |
f775b13e RR |
9994 | if (init_event) |
9995 | kvm_put_guest_fpu(vcpu); | |
b666a4b6 | 9996 | mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave, |
abd16d68 | 9997 | XFEATURE_BNDREGS); |
a554d207 WL |
9998 | if (mpx_state_buffer) |
9999 | memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state)); | |
b666a4b6 | 10000 | mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave, |
abd16d68 | 10001 | XFEATURE_BNDCSR); |
a554d207 WL |
10002 | if (mpx_state_buffer) |
10003 | memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr)); | |
f775b13e RR |
10004 | if (init_event) |
10005 | kvm_load_guest_fpu(vcpu); | |
a554d207 WL |
10006 | } |
10007 | ||
64d60670 | 10008 | if (!init_event) { |
d28bc9dd | 10009 | kvm_pmu_reset(vcpu); |
64d60670 | 10010 | vcpu->arch.smbase = 0x30000; |
db2336a8 | 10011 | |
db2336a8 | 10012 | vcpu->arch.msr_misc_features_enables = 0; |
a554d207 WL |
10013 | |
10014 | vcpu->arch.xcr0 = XFEATURE_MASK_FP; | |
64d60670 | 10015 | } |
f5132b01 | 10016 | |
66f7b72e JS |
10017 | memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs)); |
10018 | vcpu->arch.regs_avail = ~0; | |
10019 | vcpu->arch.regs_dirty = ~0; | |
10020 | ||
a554d207 WL |
10021 | vcpu->arch.ia32_xss = 0; |
10022 | ||
afaf0b2f | 10023 | kvm_x86_ops.vcpu_reset(vcpu, init_event); |
e9b11c17 ZX |
10024 | } |
10025 | ||
2b4a273b | 10026 | void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector) |
66450a21 JK |
10027 | { |
10028 | struct kvm_segment cs; | |
10029 | ||
10030 | kvm_get_segment(vcpu, &cs, VCPU_SREG_CS); | |
10031 | cs.selector = vector << 8; | |
10032 | cs.base = vector << 12; | |
10033 | kvm_set_segment(vcpu, &cs, VCPU_SREG_CS); | |
10034 | kvm_rip_write(vcpu, 0); | |
e9b11c17 ZX |
10035 | } |
10036 | ||
13a34e06 | 10037 | int kvm_arch_hardware_enable(void) |
e9b11c17 | 10038 | { |
ca84d1a2 ZA |
10039 | struct kvm *kvm; |
10040 | struct kvm_vcpu *vcpu; | |
10041 | int i; | |
0dd6a6ed ZA |
10042 | int ret; |
10043 | u64 local_tsc; | |
10044 | u64 max_tsc = 0; | |
10045 | bool stable, backwards_tsc = false; | |
18863bdd | 10046 | |
7e34fbd0 | 10047 | kvm_user_return_msr_cpu_online(); |
afaf0b2f | 10048 | ret = kvm_x86_ops.hardware_enable(); |
0dd6a6ed ZA |
10049 | if (ret != 0) |
10050 | return ret; | |
10051 | ||
4ea1636b | 10052 | local_tsc = rdtsc(); |
b0c39dc6 | 10053 | stable = !kvm_check_tsc_unstable(); |
0dd6a6ed ZA |
10054 | list_for_each_entry(kvm, &vm_list, vm_list) { |
10055 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
10056 | if (!stable && vcpu->cpu == smp_processor_id()) | |
105b21bb | 10057 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
0dd6a6ed ZA |
10058 | if (stable && vcpu->arch.last_host_tsc > local_tsc) { |
10059 | backwards_tsc = true; | |
10060 | if (vcpu->arch.last_host_tsc > max_tsc) | |
10061 | max_tsc = vcpu->arch.last_host_tsc; | |
10062 | } | |
10063 | } | |
10064 | } | |
10065 | ||
10066 | /* | |
10067 | * Sometimes, even reliable TSCs go backwards. This happens on | |
10068 | * platforms that reset TSC during suspend or hibernate actions, but | |
10069 | * maintain synchronization. We must compensate. Fortunately, we can | |
10070 | * detect that condition here, which happens early in CPU bringup, | |
10071 | * before any KVM threads can be running. Unfortunately, we can't | |
10072 | * bring the TSCs fully up to date with real time, as we aren't yet far | |
10073 | * enough into CPU bringup that we know how much real time has actually | |
9285ec4c | 10074 | * elapsed; our helper function, ktime_get_boottime_ns() will be using boot |
0dd6a6ed ZA |
10075 | * variables that haven't been updated yet. |
10076 | * | |
10077 | * So we simply find the maximum observed TSC above, then record the | |
10078 | * adjustment to TSC in each VCPU. When the VCPU later gets loaded, | |
10079 | * the adjustment will be applied. Note that we accumulate | |
10080 | * adjustments, in case multiple suspend cycles happen before some VCPU | |
10081 | * gets a chance to run again. In the event that no KVM threads get a | |
10082 | * chance to run, we will miss the entire elapsed period, as we'll have | |
10083 | * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may | |
10084 | * loose cycle time. This isn't too big a deal, since the loss will be | |
10085 | * uniform across all VCPUs (not to mention the scenario is extremely | |
10086 | * unlikely). It is possible that a second hibernate recovery happens | |
10087 | * much faster than a first, causing the observed TSC here to be | |
10088 | * smaller; this would require additional padding adjustment, which is | |
10089 | * why we set last_host_tsc to the local tsc observed here. | |
10090 | * | |
10091 | * N.B. - this code below runs only on platforms with reliable TSC, | |
10092 | * as that is the only way backwards_tsc is set above. Also note | |
10093 | * that this runs for ALL vcpus, which is not a bug; all VCPUs should | |
10094 | * have the same delta_cyc adjustment applied if backwards_tsc | |
10095 | * is detected. Note further, this adjustment is only done once, | |
10096 | * as we reset last_host_tsc on all VCPUs to stop this from being | |
10097 | * called multiple times (one for each physical CPU bringup). | |
10098 | * | |
4a969980 | 10099 | * Platforms with unreliable TSCs don't have to deal with this, they |
0dd6a6ed ZA |
10100 | * will be compensated by the logic in vcpu_load, which sets the TSC to |
10101 | * catchup mode. This will catchup all VCPUs to real time, but cannot | |
10102 | * guarantee that they stay in perfect synchronization. | |
10103 | */ | |
10104 | if (backwards_tsc) { | |
10105 | u64 delta_cyc = max_tsc - local_tsc; | |
10106 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
a826faf1 | 10107 | kvm->arch.backwards_tsc_observed = true; |
0dd6a6ed ZA |
10108 | kvm_for_each_vcpu(i, vcpu, kvm) { |
10109 | vcpu->arch.tsc_offset_adjustment += delta_cyc; | |
10110 | vcpu->arch.last_host_tsc = local_tsc; | |
105b21bb | 10111 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); |
0dd6a6ed ZA |
10112 | } |
10113 | ||
10114 | /* | |
10115 | * We have to disable TSC offset matching.. if you were | |
10116 | * booting a VM while issuing an S4 host suspend.... | |
10117 | * you may have some problem. Solving this issue is | |
10118 | * left as an exercise to the reader. | |
10119 | */ | |
10120 | kvm->arch.last_tsc_nsec = 0; | |
10121 | kvm->arch.last_tsc_write = 0; | |
10122 | } | |
10123 | ||
10124 | } | |
10125 | return 0; | |
e9b11c17 ZX |
10126 | } |
10127 | ||
13a34e06 | 10128 | void kvm_arch_hardware_disable(void) |
e9b11c17 | 10129 | { |
afaf0b2f | 10130 | kvm_x86_ops.hardware_disable(); |
13a34e06 | 10131 | drop_user_return_notifiers(); |
e9b11c17 ZX |
10132 | } |
10133 | ||
b9904085 | 10134 | int kvm_arch_hardware_setup(void *opaque) |
e9b11c17 | 10135 | { |
d008dfdb | 10136 | struct kvm_x86_init_ops *ops = opaque; |
9e9c3fe4 NA |
10137 | int r; |
10138 | ||
91661989 SC |
10139 | rdmsrl_safe(MSR_EFER, &host_efer); |
10140 | ||
408e9a31 PB |
10141 | if (boot_cpu_has(X86_FEATURE_XSAVES)) |
10142 | rdmsrl(MSR_IA32_XSS, host_xss); | |
10143 | ||
d008dfdb | 10144 | r = ops->hardware_setup(); |
9e9c3fe4 NA |
10145 | if (r != 0) |
10146 | return r; | |
10147 | ||
afaf0b2f | 10148 | memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops)); |
69c6f69a | 10149 | |
408e9a31 PB |
10150 | if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES)) |
10151 | supported_xss = 0; | |
10152 | ||
139f7425 PB |
10153 | #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f) |
10154 | cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_); | |
10155 | #undef __kvm_cpu_cap_has | |
b11306b5 | 10156 | |
35181e86 HZ |
10157 | if (kvm_has_tsc_control) { |
10158 | /* | |
10159 | * Make sure the user can only configure tsc_khz values that | |
10160 | * fit into a signed integer. | |
273ba457 | 10161 | * A min value is not calculated because it will always |
35181e86 HZ |
10162 | * be 1 on all machines. |
10163 | */ | |
10164 | u64 max = min(0x7fffffffULL, | |
10165 | __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz)); | |
10166 | kvm_max_guest_tsc_khz = max; | |
10167 | ||
ad721883 | 10168 | kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits; |
35181e86 | 10169 | } |
ad721883 | 10170 | |
9e9c3fe4 NA |
10171 | kvm_init_msr_list(); |
10172 | return 0; | |
e9b11c17 ZX |
10173 | } |
10174 | ||
10175 | void kvm_arch_hardware_unsetup(void) | |
10176 | { | |
afaf0b2f | 10177 | kvm_x86_ops.hardware_unsetup(); |
e9b11c17 ZX |
10178 | } |
10179 | ||
b9904085 | 10180 | int kvm_arch_check_processor_compat(void *opaque) |
e9b11c17 | 10181 | { |
f1cdecf5 | 10182 | struct cpuinfo_x86 *c = &cpu_data(smp_processor_id()); |
d008dfdb | 10183 | struct kvm_x86_init_ops *ops = opaque; |
f1cdecf5 SC |
10184 | |
10185 | WARN_ON(!irqs_disabled()); | |
10186 | ||
139f7425 PB |
10187 | if (__cr4_reserved_bits(cpu_has, c) != |
10188 | __cr4_reserved_bits(cpu_has, &boot_cpu_data)) | |
f1cdecf5 SC |
10189 | return -EIO; |
10190 | ||
d008dfdb | 10191 | return ops->check_processor_compatibility(); |
d71ba788 PB |
10192 | } |
10193 | ||
10194 | bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu) | |
10195 | { | |
10196 | return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id; | |
10197 | } | |
10198 | EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp); | |
10199 | ||
10200 | bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu) | |
10201 | { | |
10202 | return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0; | |
e9b11c17 ZX |
10203 | } |
10204 | ||
54e9818f | 10205 | struct static_key kvm_no_apic_vcpu __read_mostly; |
bce87cce | 10206 | EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu); |
54e9818f | 10207 | |
e790d9ef RK |
10208 | void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu) |
10209 | { | |
b35e5548 LX |
10210 | struct kvm_pmu *pmu = vcpu_to_pmu(vcpu); |
10211 | ||
c595ceee | 10212 | vcpu->arch.l1tf_flush_l1d = true; |
b35e5548 LX |
10213 | if (pmu->version && unlikely(pmu->event_count)) { |
10214 | pmu->need_cleanup = true; | |
10215 | kvm_make_request(KVM_REQ_PMU, vcpu); | |
10216 | } | |
afaf0b2f | 10217 | kvm_x86_ops.sched_in(vcpu, cpu); |
e790d9ef RK |
10218 | } |
10219 | ||
562b6b08 SC |
10220 | void kvm_arch_free_vm(struct kvm *kvm) |
10221 | { | |
10222 | kfree(kvm->arch.hyperv.hv_pa_pg); | |
10223 | vfree(kvm); | |
e790d9ef RK |
10224 | } |
10225 | ||
562b6b08 | 10226 | |
e08b9637 | 10227 | int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) |
d19a9cd2 | 10228 | { |
e08b9637 CO |
10229 | if (type) |
10230 | return -EINVAL; | |
10231 | ||
6ef768fa | 10232 | INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list); |
f05e70ac | 10233 | INIT_LIST_HEAD(&kvm->arch.active_mmu_pages); |
10605204 | 10234 | INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages); |
1aa9b957 | 10235 | INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages); |
4d5c5d0f | 10236 | INIT_LIST_HEAD(&kvm->arch.assigned_dev_head); |
e0f0bbc5 | 10237 | atomic_set(&kvm->arch.noncoherent_dma_count, 0); |
d19a9cd2 | 10238 | |
5550af4d SY |
10239 | /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */ |
10240 | set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap); | |
7a84428a AW |
10241 | /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */ |
10242 | set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID, | |
10243 | &kvm->arch.irq_sources_bitmap); | |
5550af4d | 10244 | |
038f8c11 | 10245 | raw_spin_lock_init(&kvm->arch.tsc_write_lock); |
1e08ec4a | 10246 | mutex_init(&kvm->arch.apic_map_lock); |
d828199e MT |
10247 | spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock); |
10248 | ||
8171cd68 | 10249 | kvm->arch.kvmclock_offset = -get_kvmclock_base_ns(); |
d828199e | 10250 | pvclock_update_vm_gtod_copy(kvm); |
53f658b3 | 10251 | |
6fbbde9a DS |
10252 | kvm->arch.guest_can_read_msr_platform_info = true; |
10253 | ||
7e44e449 | 10254 | INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn); |
332967a3 | 10255 | INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn); |
7e44e449 | 10256 | |
cbc0236a | 10257 | kvm_hv_init_vm(kvm); |
0eb05bf2 | 10258 | kvm_page_track_init(kvm); |
13d268ca | 10259 | kvm_mmu_init_vm(kvm); |
0eb05bf2 | 10260 | |
afaf0b2f | 10261 | return kvm_x86_ops.vm_init(kvm); |
d19a9cd2 ZX |
10262 | } |
10263 | ||
1aa9b957 JS |
10264 | int kvm_arch_post_init_vm(struct kvm *kvm) |
10265 | { | |
10266 | return kvm_mmu_post_init_vm(kvm); | |
10267 | } | |
10268 | ||
d19a9cd2 ZX |
10269 | static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu) |
10270 | { | |
ec7660cc | 10271 | vcpu_load(vcpu); |
d19a9cd2 ZX |
10272 | kvm_mmu_unload(vcpu); |
10273 | vcpu_put(vcpu); | |
10274 | } | |
10275 | ||
10276 | static void kvm_free_vcpus(struct kvm *kvm) | |
10277 | { | |
10278 | unsigned int i; | |
988a2cae | 10279 | struct kvm_vcpu *vcpu; |
d19a9cd2 ZX |
10280 | |
10281 | /* | |
10282 | * Unpin any mmu pages first. | |
10283 | */ | |
af585b92 GN |
10284 | kvm_for_each_vcpu(i, vcpu, kvm) { |
10285 | kvm_clear_async_pf_completion_queue(vcpu); | |
988a2cae | 10286 | kvm_unload_vcpu_mmu(vcpu); |
af585b92 | 10287 | } |
988a2cae | 10288 | kvm_for_each_vcpu(i, vcpu, kvm) |
4543bdc0 | 10289 | kvm_vcpu_destroy(vcpu); |
988a2cae GN |
10290 | |
10291 | mutex_lock(&kvm->lock); | |
10292 | for (i = 0; i < atomic_read(&kvm->online_vcpus); i++) | |
10293 | kvm->vcpus[i] = NULL; | |
d19a9cd2 | 10294 | |
988a2cae GN |
10295 | atomic_set(&kvm->online_vcpus, 0); |
10296 | mutex_unlock(&kvm->lock); | |
d19a9cd2 ZX |
10297 | } |
10298 | ||
ad8ba2cd SY |
10299 | void kvm_arch_sync_events(struct kvm *kvm) |
10300 | { | |
332967a3 | 10301 | cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work); |
7e44e449 | 10302 | cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work); |
aea924f6 | 10303 | kvm_free_pit(kvm); |
ad8ba2cd SY |
10304 | } |
10305 | ||
1d8007bd | 10306 | int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size) |
9da0e4d5 PB |
10307 | { |
10308 | int i, r; | |
3f649ab7 | 10309 | unsigned long hva, old_npages; |
f0d648bd | 10310 | struct kvm_memslots *slots = kvm_memslots(kvm); |
0577d1ab | 10311 | struct kvm_memory_slot *slot; |
9da0e4d5 PB |
10312 | |
10313 | /* Called with kvm->slots_lock held. */ | |
1d8007bd PB |
10314 | if (WARN_ON(id >= KVM_MEM_SLOTS_NUM)) |
10315 | return -EINVAL; | |
9da0e4d5 | 10316 | |
f0d648bd PB |
10317 | slot = id_to_memslot(slots, id); |
10318 | if (size) { | |
0577d1ab | 10319 | if (slot && slot->npages) |
f0d648bd PB |
10320 | return -EEXIST; |
10321 | ||
10322 | /* | |
10323 | * MAP_SHARED to prevent internal slot pages from being moved | |
10324 | * by fork()/COW. | |
10325 | */ | |
10326 | hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE, | |
10327 | MAP_SHARED | MAP_ANONYMOUS, 0); | |
10328 | if (IS_ERR((void *)hva)) | |
10329 | return PTR_ERR((void *)hva); | |
10330 | } else { | |
0577d1ab | 10331 | if (!slot || !slot->npages) |
f0d648bd PB |
10332 | return 0; |
10333 | ||
0577d1ab | 10334 | old_npages = slot->npages; |
e0135a10 | 10335 | hva = 0; |
f0d648bd PB |
10336 | } |
10337 | ||
9da0e4d5 | 10338 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
1d8007bd | 10339 | struct kvm_userspace_memory_region m; |
9da0e4d5 | 10340 | |
1d8007bd PB |
10341 | m.slot = id | (i << 16); |
10342 | m.flags = 0; | |
10343 | m.guest_phys_addr = gpa; | |
f0d648bd | 10344 | m.userspace_addr = hva; |
1d8007bd | 10345 | m.memory_size = size; |
9da0e4d5 PB |
10346 | r = __kvm_set_memory_region(kvm, &m); |
10347 | if (r < 0) | |
10348 | return r; | |
10349 | } | |
10350 | ||
103c763c | 10351 | if (!size) |
0577d1ab | 10352 | vm_munmap(hva, old_npages * PAGE_SIZE); |
f0d648bd | 10353 | |
9da0e4d5 PB |
10354 | return 0; |
10355 | } | |
10356 | EXPORT_SYMBOL_GPL(__x86_set_memory_region); | |
10357 | ||
1aa9b957 JS |
10358 | void kvm_arch_pre_destroy_vm(struct kvm *kvm) |
10359 | { | |
10360 | kvm_mmu_pre_destroy_vm(kvm); | |
10361 | } | |
10362 | ||
d19a9cd2 ZX |
10363 | void kvm_arch_destroy_vm(struct kvm *kvm) |
10364 | { | |
1a155254 AG |
10365 | u32 i; |
10366 | ||
27469d29 AH |
10367 | if (current->mm == kvm->mm) { |
10368 | /* | |
10369 | * Free memory regions allocated on behalf of userspace, | |
10370 | * unless the the memory map has changed due to process exit | |
10371 | * or fd copying. | |
10372 | */ | |
6a3c623b PX |
10373 | mutex_lock(&kvm->slots_lock); |
10374 | __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, | |
10375 | 0, 0); | |
10376 | __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, | |
10377 | 0, 0); | |
10378 | __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0); | |
10379 | mutex_unlock(&kvm->slots_lock); | |
27469d29 | 10380 | } |
afaf0b2f SC |
10381 | if (kvm_x86_ops.vm_destroy) |
10382 | kvm_x86_ops.vm_destroy(kvm); | |
1a155254 AG |
10383 | for (i = 0; i < kvm->arch.msr_filter.count; i++) |
10384 | kfree(kvm->arch.msr_filter.ranges[i].bitmap); | |
c761159c PX |
10385 | kvm_pic_destroy(kvm); |
10386 | kvm_ioapic_destroy(kvm); | |
d19a9cd2 | 10387 | kvm_free_vcpus(kvm); |
af1bae54 | 10388 | kvfree(rcu_dereference_check(kvm->arch.apic_map, 1)); |
66bb8a06 | 10389 | kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1)); |
13d268ca | 10390 | kvm_mmu_uninit_vm(kvm); |
2beb6dad | 10391 | kvm_page_track_cleanup(kvm); |
cbc0236a | 10392 | kvm_hv_destroy_vm(kvm); |
d19a9cd2 | 10393 | } |
0de10343 | 10394 | |
e96c81ee | 10395 | void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot) |
db3fe4eb TY |
10396 | { |
10397 | int i; | |
10398 | ||
d89cc617 | 10399 | for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) { |
e96c81ee SC |
10400 | kvfree(slot->arch.rmap[i]); |
10401 | slot->arch.rmap[i] = NULL; | |
10402 | ||
d89cc617 TY |
10403 | if (i == 0) |
10404 | continue; | |
10405 | ||
e96c81ee SC |
10406 | kvfree(slot->arch.lpage_info[i - 1]); |
10407 | slot->arch.lpage_info[i - 1] = NULL; | |
db3fe4eb | 10408 | } |
21ebbeda | 10409 | |
e96c81ee | 10410 | kvm_page_track_free_memslot(slot); |
db3fe4eb TY |
10411 | } |
10412 | ||
0dab98b7 SC |
10413 | static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot, |
10414 | unsigned long npages) | |
db3fe4eb TY |
10415 | { |
10416 | int i; | |
10417 | ||
edd4fa37 SC |
10418 | /* |
10419 | * Clear out the previous array pointers for the KVM_MR_MOVE case. The | |
10420 | * old arrays will be freed by __kvm_set_memory_region() if installing | |
10421 | * the new memslot is successful. | |
10422 | */ | |
10423 | memset(&slot->arch, 0, sizeof(slot->arch)); | |
10424 | ||
d89cc617 | 10425 | for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) { |
92f94f1e | 10426 | struct kvm_lpage_info *linfo; |
db3fe4eb TY |
10427 | unsigned long ugfn; |
10428 | int lpages; | |
d89cc617 | 10429 | int level = i + 1; |
db3fe4eb TY |
10430 | |
10431 | lpages = gfn_to_index(slot->base_gfn + npages - 1, | |
10432 | slot->base_gfn, level) + 1; | |
10433 | ||
d89cc617 | 10434 | slot->arch.rmap[i] = |
778e1cdd | 10435 | kvcalloc(lpages, sizeof(*slot->arch.rmap[i]), |
254272ce | 10436 | GFP_KERNEL_ACCOUNT); |
d89cc617 | 10437 | if (!slot->arch.rmap[i]) |
77d11309 | 10438 | goto out_free; |
d89cc617 TY |
10439 | if (i == 0) |
10440 | continue; | |
77d11309 | 10441 | |
254272ce | 10442 | linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT); |
92f94f1e | 10443 | if (!linfo) |
db3fe4eb TY |
10444 | goto out_free; |
10445 | ||
92f94f1e XG |
10446 | slot->arch.lpage_info[i - 1] = linfo; |
10447 | ||
db3fe4eb | 10448 | if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1)) |
92f94f1e | 10449 | linfo[0].disallow_lpage = 1; |
db3fe4eb | 10450 | if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1)) |
92f94f1e | 10451 | linfo[lpages - 1].disallow_lpage = 1; |
db3fe4eb TY |
10452 | ugfn = slot->userspace_addr >> PAGE_SHIFT; |
10453 | /* | |
10454 | * If the gfn and userspace address are not aligned wrt each | |
600087b6 | 10455 | * other, disable large page support for this slot. |
db3fe4eb | 10456 | */ |
600087b6 | 10457 | if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) { |
db3fe4eb TY |
10458 | unsigned long j; |
10459 | ||
10460 | for (j = 0; j < lpages; ++j) | |
92f94f1e | 10461 | linfo[j].disallow_lpage = 1; |
db3fe4eb TY |
10462 | } |
10463 | } | |
10464 | ||
21ebbeda XG |
10465 | if (kvm_page_track_create_memslot(slot, npages)) |
10466 | goto out_free; | |
10467 | ||
db3fe4eb TY |
10468 | return 0; |
10469 | ||
10470 | out_free: | |
d89cc617 | 10471 | for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) { |
548ef284 | 10472 | kvfree(slot->arch.rmap[i]); |
d89cc617 TY |
10473 | slot->arch.rmap[i] = NULL; |
10474 | if (i == 0) | |
10475 | continue; | |
10476 | ||
548ef284 | 10477 | kvfree(slot->arch.lpage_info[i - 1]); |
d89cc617 | 10478 | slot->arch.lpage_info[i - 1] = NULL; |
db3fe4eb TY |
10479 | } |
10480 | return -ENOMEM; | |
10481 | } | |
10482 | ||
15248258 | 10483 | void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen) |
e59dbe09 | 10484 | { |
91724814 BO |
10485 | struct kvm_vcpu *vcpu; |
10486 | int i; | |
10487 | ||
e6dff7d1 TY |
10488 | /* |
10489 | * memslots->generation has been incremented. | |
10490 | * mmio generation may have reached its maximum value. | |
10491 | */ | |
15248258 | 10492 | kvm_mmu_invalidate_mmio_sptes(kvm, gen); |
91724814 BO |
10493 | |
10494 | /* Force re-initialization of steal_time cache */ | |
10495 | kvm_for_each_vcpu(i, vcpu, kvm) | |
10496 | kvm_vcpu_kick(vcpu); | |
e59dbe09 TY |
10497 | } |
10498 | ||
f7784b8e MT |
10499 | int kvm_arch_prepare_memory_region(struct kvm *kvm, |
10500 | struct kvm_memory_slot *memslot, | |
09170a49 | 10501 | const struct kvm_userspace_memory_region *mem, |
7b6195a9 | 10502 | enum kvm_mr_change change) |
0de10343 | 10503 | { |
0dab98b7 SC |
10504 | if (change == KVM_MR_CREATE || change == KVM_MR_MOVE) |
10505 | return kvm_alloc_memslot_metadata(memslot, | |
10506 | mem->memory_size >> PAGE_SHIFT); | |
f7784b8e MT |
10507 | return 0; |
10508 | } | |
10509 | ||
88178fd4 | 10510 | static void kvm_mmu_slot_apply_flags(struct kvm *kvm, |
3741679b AY |
10511 | struct kvm_memory_slot *old, |
10512 | struct kvm_memory_slot *new, | |
10513 | enum kvm_mr_change change) | |
88178fd4 | 10514 | { |
3741679b AY |
10515 | /* |
10516 | * Nothing to do for RO slots or CREATE/MOVE/DELETE of a slot. | |
10517 | * See comments below. | |
10518 | */ | |
10519 | if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY)) | |
88178fd4 | 10520 | return; |
88178fd4 KH |
10521 | |
10522 | /* | |
3741679b AY |
10523 | * Dirty logging tracks sptes in 4k granularity, meaning that large |
10524 | * sptes have to be split. If live migration is successful, the guest | |
10525 | * in the source machine will be destroyed and large sptes will be | |
10526 | * created in the destination. However, if the guest continues to run | |
10527 | * in the source machine (for example if live migration fails), small | |
10528 | * sptes will remain around and cause bad performance. | |
88178fd4 | 10529 | * |
3741679b AY |
10530 | * Scan sptes if dirty logging has been stopped, dropping those |
10531 | * which can be collapsed into a single large-page spte. Later | |
10532 | * page faults will create the large-page sptes. | |
88178fd4 | 10533 | * |
3741679b AY |
10534 | * There is no need to do this in any of the following cases: |
10535 | * CREATE: No dirty mappings will already exist. | |
10536 | * MOVE/DELETE: The old mappings will already have been cleaned up by | |
10537 | * kvm_arch_flush_shadow_memslot() | |
10538 | */ | |
10539 | if ((old->flags & KVM_MEM_LOG_DIRTY_PAGES) && | |
10540 | !(new->flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
10541 | kvm_mmu_zap_collapsible_sptes(kvm, new); | |
10542 | ||
10543 | /* | |
10544 | * Enable or disable dirty logging for the slot. | |
88178fd4 | 10545 | * |
3741679b AY |
10546 | * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of the old |
10547 | * slot have been zapped so no dirty logging updates are needed for | |
10548 | * the old slot. | |
10549 | * For KVM_MR_CREATE and KVM_MR_MOVE, once the new slot is visible | |
10550 | * any mappings that might be created in it will consume the | |
10551 | * properties of the new slot and do not need to be updated here. | |
88178fd4 | 10552 | * |
3741679b AY |
10553 | * When PML is enabled, the kvm_x86_ops dirty logging hooks are |
10554 | * called to enable/disable dirty logging. | |
88178fd4 | 10555 | * |
3741679b AY |
10556 | * When disabling dirty logging with PML enabled, the D-bit is set |
10557 | * for sptes in the slot in order to prevent unnecessary GPA | |
10558 | * logging in the PML buffer (and potential PML buffer full VMEXIT). | |
10559 | * This guarantees leaving PML enabled for the guest's lifetime | |
10560 | * won't have any additional overhead from PML when the guest is | |
10561 | * running with dirty logging disabled. | |
88178fd4 | 10562 | * |
3741679b AY |
10563 | * When enabling dirty logging, large sptes are write-protected |
10564 | * so they can be split on first write. New large sptes cannot | |
10565 | * be created for this slot until the end of the logging. | |
88178fd4 | 10566 | * See the comments in fast_page_fault(). |
3741679b AY |
10567 | * For small sptes, nothing is done if the dirty log is in the |
10568 | * initial-all-set state. Otherwise, depending on whether pml | |
10569 | * is enabled the D-bit or the W-bit will be cleared. | |
88178fd4 KH |
10570 | */ |
10571 | if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) { | |
afaf0b2f SC |
10572 | if (kvm_x86_ops.slot_enable_log_dirty) { |
10573 | kvm_x86_ops.slot_enable_log_dirty(kvm, new); | |
3c9bd400 JZ |
10574 | } else { |
10575 | int level = | |
10576 | kvm_dirty_log_manual_protect_and_init_set(kvm) ? | |
3bae0459 | 10577 | PG_LEVEL_2M : PG_LEVEL_4K; |
3c9bd400 JZ |
10578 | |
10579 | /* | |
10580 | * If we're with initial-all-set, we don't need | |
10581 | * to write protect any small page because | |
10582 | * they're reported as dirty already. However | |
10583 | * we still need to write-protect huge pages | |
10584 | * so that the page split can happen lazily on | |
10585 | * the first write to the huge page. | |
10586 | */ | |
10587 | kvm_mmu_slot_remove_write_access(kvm, new, level); | |
10588 | } | |
88178fd4 | 10589 | } else { |
afaf0b2f SC |
10590 | if (kvm_x86_ops.slot_disable_log_dirty) |
10591 | kvm_x86_ops.slot_disable_log_dirty(kvm, new); | |
88178fd4 KH |
10592 | } |
10593 | } | |
10594 | ||
f7784b8e | 10595 | void kvm_arch_commit_memory_region(struct kvm *kvm, |
09170a49 | 10596 | const struct kvm_userspace_memory_region *mem, |
9d4c197c | 10597 | struct kvm_memory_slot *old, |
f36f3f28 | 10598 | const struct kvm_memory_slot *new, |
8482644a | 10599 | enum kvm_mr_change change) |
f7784b8e | 10600 | { |
48c0e4e9 | 10601 | if (!kvm->arch.n_requested_mmu_pages) |
4d66623c WY |
10602 | kvm_mmu_change_mmu_pages(kvm, |
10603 | kvm_mmu_calculate_default_mmu_pages(kvm)); | |
1c91cad4 | 10604 | |
3ea3b7fa | 10605 | /* |
f36f3f28 | 10606 | * FIXME: const-ify all uses of struct kvm_memory_slot. |
c972f3b1 | 10607 | */ |
3741679b | 10608 | kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change); |
21198846 SC |
10609 | |
10610 | /* Free the arrays associated with the old memslot. */ | |
10611 | if (change == KVM_MR_MOVE) | |
e96c81ee | 10612 | kvm_arch_free_memslot(kvm, old); |
0de10343 | 10613 | } |
1d737c8a | 10614 | |
2df72e9b | 10615 | void kvm_arch_flush_shadow_all(struct kvm *kvm) |
34d4cb8f | 10616 | { |
7390de1e | 10617 | kvm_mmu_zap_all(kvm); |
34d4cb8f MT |
10618 | } |
10619 | ||
2df72e9b MT |
10620 | void kvm_arch_flush_shadow_memslot(struct kvm *kvm, |
10621 | struct kvm_memory_slot *slot) | |
10622 | { | |
ae7cd873 | 10623 | kvm_page_track_flush_slot(kvm, slot); |
2df72e9b MT |
10624 | } |
10625 | ||
e6c67d8c LA |
10626 | static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu) |
10627 | { | |
10628 | return (is_guest_mode(vcpu) && | |
afaf0b2f SC |
10629 | kvm_x86_ops.guest_apic_has_interrupt && |
10630 | kvm_x86_ops.guest_apic_has_interrupt(vcpu)); | |
e6c67d8c LA |
10631 | } |
10632 | ||
5d9bc648 PB |
10633 | static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu) |
10634 | { | |
10635 | if (!list_empty_careful(&vcpu->async_pf.done)) | |
10636 | return true; | |
10637 | ||
10638 | if (kvm_apic_has_events(vcpu)) | |
10639 | return true; | |
10640 | ||
10641 | if (vcpu->arch.pv.pv_unhalted) | |
10642 | return true; | |
10643 | ||
a5f01f8e WL |
10644 | if (vcpu->arch.exception.pending) |
10645 | return true; | |
10646 | ||
47a66eed Z |
10647 | if (kvm_test_request(KVM_REQ_NMI, vcpu) || |
10648 | (vcpu->arch.nmi_pending && | |
c300ab9f | 10649 | kvm_x86_ops.nmi_allowed(vcpu, false))) |
5d9bc648 PB |
10650 | return true; |
10651 | ||
47a66eed | 10652 | if (kvm_test_request(KVM_REQ_SMI, vcpu) || |
a9fa7cb6 | 10653 | (vcpu->arch.smi_pending && |
c300ab9f | 10654 | kvm_x86_ops.smi_allowed(vcpu, false))) |
73917739 PB |
10655 | return true; |
10656 | ||
5d9bc648 | 10657 | if (kvm_arch_interrupt_allowed(vcpu) && |
e6c67d8c LA |
10658 | (kvm_cpu_has_interrupt(vcpu) || |
10659 | kvm_guest_apic_has_interrupt(vcpu))) | |
5d9bc648 PB |
10660 | return true; |
10661 | ||
1f4b34f8 AS |
10662 | if (kvm_hv_has_stimer_pending(vcpu)) |
10663 | return true; | |
10664 | ||
d2060bd4 SC |
10665 | if (is_guest_mode(vcpu) && |
10666 | kvm_x86_ops.nested_ops->hv_timer_pending && | |
10667 | kvm_x86_ops.nested_ops->hv_timer_pending(vcpu)) | |
10668 | return true; | |
10669 | ||
5d9bc648 PB |
10670 | return false; |
10671 | } | |
10672 | ||
1d737c8a ZX |
10673 | int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu) |
10674 | { | |
5d9bc648 | 10675 | return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu); |
1d737c8a | 10676 | } |
5736199a | 10677 | |
17e433b5 WL |
10678 | bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu) |
10679 | { | |
10680 | if (READ_ONCE(vcpu->arch.pv.pv_unhalted)) | |
10681 | return true; | |
10682 | ||
10683 | if (kvm_test_request(KVM_REQ_NMI, vcpu) || | |
10684 | kvm_test_request(KVM_REQ_SMI, vcpu) || | |
10685 | kvm_test_request(KVM_REQ_EVENT, vcpu)) | |
10686 | return true; | |
10687 | ||
afaf0b2f | 10688 | if (vcpu->arch.apicv_active && kvm_x86_ops.dy_apicv_has_pending_interrupt(vcpu)) |
17e433b5 WL |
10689 | return true; |
10690 | ||
10691 | return false; | |
10692 | } | |
10693 | ||
199b5763 LM |
10694 | bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu) |
10695 | { | |
de63ad4c | 10696 | return vcpu->arch.preempted_in_kernel; |
199b5763 LM |
10697 | } |
10698 | ||
b6d33834 | 10699 | int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) |
5736199a | 10700 | { |
b6d33834 | 10701 | return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; |
5736199a | 10702 | } |
78646121 GN |
10703 | |
10704 | int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu) | |
10705 | { | |
c300ab9f | 10706 | return kvm_x86_ops.interrupt_allowed(vcpu, false); |
78646121 | 10707 | } |
229456fc | 10708 | |
82b32774 | 10709 | unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu) |
f92653ee | 10710 | { |
82b32774 NA |
10711 | if (is_64_bit_mode(vcpu)) |
10712 | return kvm_rip_read(vcpu); | |
10713 | return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) + | |
10714 | kvm_rip_read(vcpu)); | |
10715 | } | |
10716 | EXPORT_SYMBOL_GPL(kvm_get_linear_rip); | |
f92653ee | 10717 | |
82b32774 NA |
10718 | bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip) |
10719 | { | |
10720 | return kvm_get_linear_rip(vcpu) == linear_rip; | |
f92653ee JK |
10721 | } |
10722 | EXPORT_SYMBOL_GPL(kvm_is_linear_rip); | |
10723 | ||
94fe45da JK |
10724 | unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu) |
10725 | { | |
10726 | unsigned long rflags; | |
10727 | ||
afaf0b2f | 10728 | rflags = kvm_x86_ops.get_rflags(vcpu); |
94fe45da | 10729 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) |
c310bac5 | 10730 | rflags &= ~X86_EFLAGS_TF; |
94fe45da JK |
10731 | return rflags; |
10732 | } | |
10733 | EXPORT_SYMBOL_GPL(kvm_get_rflags); | |
10734 | ||
6addfc42 | 10735 | static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags) |
94fe45da JK |
10736 | { |
10737 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP && | |
f92653ee | 10738 | kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip)) |
c310bac5 | 10739 | rflags |= X86_EFLAGS_TF; |
afaf0b2f | 10740 | kvm_x86_ops.set_rflags(vcpu, rflags); |
6addfc42 PB |
10741 | } |
10742 | ||
10743 | void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags) | |
10744 | { | |
10745 | __kvm_set_rflags(vcpu, rflags); | |
3842d135 | 10746 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
94fe45da JK |
10747 | } |
10748 | EXPORT_SYMBOL_GPL(kvm_set_rflags); | |
10749 | ||
56028d08 GN |
10750 | void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work) |
10751 | { | |
10752 | int r; | |
10753 | ||
44dd3ffa | 10754 | if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) || |
f2e10669 | 10755 | work->wakeup_all) |
56028d08 GN |
10756 | return; |
10757 | ||
10758 | r = kvm_mmu_reload(vcpu); | |
10759 | if (unlikely(r)) | |
10760 | return; | |
10761 | ||
44dd3ffa | 10762 | if (!vcpu->arch.mmu->direct_map && |
d8dd54e0 | 10763 | work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu)) |
fb67e14f XG |
10764 | return; |
10765 | ||
7a02674d | 10766 | kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true); |
56028d08 GN |
10767 | } |
10768 | ||
af585b92 GN |
10769 | static inline u32 kvm_async_pf_hash_fn(gfn_t gfn) |
10770 | { | |
dd03bcaa PX |
10771 | BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU)); |
10772 | ||
af585b92 GN |
10773 | return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU)); |
10774 | } | |
10775 | ||
10776 | static inline u32 kvm_async_pf_next_probe(u32 key) | |
10777 | { | |
dd03bcaa | 10778 | return (key + 1) & (ASYNC_PF_PER_VCPU - 1); |
af585b92 GN |
10779 | } |
10780 | ||
10781 | static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) | |
10782 | { | |
10783 | u32 key = kvm_async_pf_hash_fn(gfn); | |
10784 | ||
10785 | while (vcpu->arch.apf.gfns[key] != ~0) | |
10786 | key = kvm_async_pf_next_probe(key); | |
10787 | ||
10788 | vcpu->arch.apf.gfns[key] = gfn; | |
10789 | } | |
10790 | ||
10791 | static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn) | |
10792 | { | |
10793 | int i; | |
10794 | u32 key = kvm_async_pf_hash_fn(gfn); | |
10795 | ||
dd03bcaa | 10796 | for (i = 0; i < ASYNC_PF_PER_VCPU && |
c7d28c24 XG |
10797 | (vcpu->arch.apf.gfns[key] != gfn && |
10798 | vcpu->arch.apf.gfns[key] != ~0); i++) | |
af585b92 GN |
10799 | key = kvm_async_pf_next_probe(key); |
10800 | ||
10801 | return key; | |
10802 | } | |
10803 | ||
10804 | bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) | |
10805 | { | |
10806 | return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn; | |
10807 | } | |
10808 | ||
10809 | static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) | |
10810 | { | |
10811 | u32 i, j, k; | |
10812 | ||
10813 | i = j = kvm_async_pf_gfn_slot(vcpu, gfn); | |
0fd46044 PX |
10814 | |
10815 | if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn)) | |
10816 | return; | |
10817 | ||
af585b92 GN |
10818 | while (true) { |
10819 | vcpu->arch.apf.gfns[i] = ~0; | |
10820 | do { | |
10821 | j = kvm_async_pf_next_probe(j); | |
10822 | if (vcpu->arch.apf.gfns[j] == ~0) | |
10823 | return; | |
10824 | k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]); | |
10825 | /* | |
10826 | * k lies cyclically in ]i,j] | |
10827 | * | i.k.j | | |
10828 | * |....j i.k.| or |.k..j i...| | |
10829 | */ | |
10830 | } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j)); | |
10831 | vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j]; | |
10832 | i = j; | |
10833 | } | |
10834 | } | |
10835 | ||
68fd66f1 | 10836 | static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu) |
7c90705b | 10837 | { |
68fd66f1 VK |
10838 | u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT; |
10839 | ||
10840 | return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason, | |
10841 | sizeof(reason)); | |
10842 | } | |
10843 | ||
10844 | static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token) | |
10845 | { | |
2635b5c4 | 10846 | unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token); |
4e335d9e | 10847 | |
2635b5c4 VK |
10848 | return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data, |
10849 | &token, offset, sizeof(token)); | |
10850 | } | |
10851 | ||
10852 | static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu) | |
10853 | { | |
10854 | unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token); | |
10855 | u32 val; | |
10856 | ||
10857 | if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data, | |
10858 | &val, offset, sizeof(val))) | |
10859 | return false; | |
10860 | ||
10861 | return !val; | |
7c90705b GN |
10862 | } |
10863 | ||
1dfdb45e PB |
10864 | static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu) |
10865 | { | |
10866 | if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu)) | |
10867 | return false; | |
10868 | ||
2635b5c4 VK |
10869 | if (!kvm_pv_async_pf_enabled(vcpu) || |
10870 | (vcpu->arch.apf.send_user_only && kvm_x86_ops.get_cpl(vcpu) == 0)) | |
1dfdb45e PB |
10871 | return false; |
10872 | ||
10873 | return true; | |
10874 | } | |
10875 | ||
10876 | bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu) | |
10877 | { | |
10878 | if (unlikely(!lapic_in_kernel(vcpu) || | |
10879 | kvm_event_needs_reinjection(vcpu) || | |
10880 | vcpu->arch.exception.pending)) | |
10881 | return false; | |
10882 | ||
10883 | if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu)) | |
10884 | return false; | |
10885 | ||
10886 | /* | |
10887 | * If interrupts are off we cannot even use an artificial | |
10888 | * halt state. | |
10889 | */ | |
c300ab9f | 10890 | return kvm_arch_interrupt_allowed(vcpu); |
1dfdb45e PB |
10891 | } |
10892 | ||
2a18b7e7 | 10893 | bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu, |
af585b92 GN |
10894 | struct kvm_async_pf *work) |
10895 | { | |
6389ee94 AK |
10896 | struct x86_exception fault; |
10897 | ||
736c291c | 10898 | trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa); |
af585b92 | 10899 | kvm_add_async_pf_gfn(vcpu, work->arch.gfn); |
7c90705b | 10900 | |
1dfdb45e | 10901 | if (kvm_can_deliver_async_pf(vcpu) && |
68fd66f1 | 10902 | !apf_put_user_notpresent(vcpu)) { |
6389ee94 AK |
10903 | fault.vector = PF_VECTOR; |
10904 | fault.error_code_valid = true; | |
10905 | fault.error_code = 0; | |
10906 | fault.nested_page_fault = false; | |
10907 | fault.address = work->arch.token; | |
adfe20fb | 10908 | fault.async_page_fault = true; |
6389ee94 | 10909 | kvm_inject_page_fault(vcpu, &fault); |
2a18b7e7 | 10910 | return true; |
1dfdb45e PB |
10911 | } else { |
10912 | /* | |
10913 | * It is not possible to deliver a paravirtualized asynchronous | |
10914 | * page fault, but putting the guest in an artificial halt state | |
10915 | * can be beneficial nevertheless: if an interrupt arrives, we | |
10916 | * can deliver it timely and perhaps the guest will schedule | |
10917 | * another process. When the instruction that triggered a page | |
10918 | * fault is retried, hopefully the page will be ready in the host. | |
10919 | */ | |
10920 | kvm_make_request(KVM_REQ_APF_HALT, vcpu); | |
2a18b7e7 | 10921 | return false; |
7c90705b | 10922 | } |
af585b92 GN |
10923 | } |
10924 | ||
10925 | void kvm_arch_async_page_present(struct kvm_vcpu *vcpu, | |
10926 | struct kvm_async_pf *work) | |
10927 | { | |
2635b5c4 VK |
10928 | struct kvm_lapic_irq irq = { |
10929 | .delivery_mode = APIC_DM_FIXED, | |
10930 | .vector = vcpu->arch.apf.vec | |
10931 | }; | |
6389ee94 | 10932 | |
f2e10669 | 10933 | if (work->wakeup_all) |
7c90705b GN |
10934 | work->arch.token = ~0; /* broadcast wakeup */ |
10935 | else | |
10936 | kvm_del_async_pf_gfn(vcpu, work->arch.gfn); | |
736c291c | 10937 | trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa); |
7c90705b | 10938 | |
2a18b7e7 VK |
10939 | if ((work->wakeup_all || work->notpresent_injected) && |
10940 | kvm_pv_async_pf_enabled(vcpu) && | |
557a961a VK |
10941 | !apf_put_user_ready(vcpu, work->arch.token)) { |
10942 | vcpu->arch.apf.pageready_pending = true; | |
2635b5c4 | 10943 | kvm_apic_set_irq(vcpu, &irq, NULL); |
557a961a | 10944 | } |
2635b5c4 | 10945 | |
e6d53e3b | 10946 | vcpu->arch.apf.halted = false; |
a4fa1635 | 10947 | vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; |
7c90705b GN |
10948 | } |
10949 | ||
557a961a VK |
10950 | void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu) |
10951 | { | |
10952 | kvm_make_request(KVM_REQ_APF_READY, vcpu); | |
10953 | if (!vcpu->arch.apf.pageready_pending) | |
10954 | kvm_vcpu_kick(vcpu); | |
10955 | } | |
10956 | ||
7c0ade6c | 10957 | bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu) |
7c90705b | 10958 | { |
2635b5c4 | 10959 | if (!kvm_pv_async_pf_enabled(vcpu)) |
7c90705b GN |
10960 | return true; |
10961 | else | |
2635b5c4 | 10962 | return apf_pageready_slot_free(vcpu); |
af585b92 GN |
10963 | } |
10964 | ||
5544eb9b PB |
10965 | void kvm_arch_start_assignment(struct kvm *kvm) |
10966 | { | |
10967 | atomic_inc(&kvm->arch.assigned_device_count); | |
10968 | } | |
10969 | EXPORT_SYMBOL_GPL(kvm_arch_start_assignment); | |
10970 | ||
10971 | void kvm_arch_end_assignment(struct kvm *kvm) | |
10972 | { | |
10973 | atomic_dec(&kvm->arch.assigned_device_count); | |
10974 | } | |
10975 | EXPORT_SYMBOL_GPL(kvm_arch_end_assignment); | |
10976 | ||
10977 | bool kvm_arch_has_assigned_device(struct kvm *kvm) | |
10978 | { | |
10979 | return atomic_read(&kvm->arch.assigned_device_count); | |
10980 | } | |
10981 | EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device); | |
10982 | ||
e0f0bbc5 AW |
10983 | void kvm_arch_register_noncoherent_dma(struct kvm *kvm) |
10984 | { | |
10985 | atomic_inc(&kvm->arch.noncoherent_dma_count); | |
10986 | } | |
10987 | EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma); | |
10988 | ||
10989 | void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm) | |
10990 | { | |
10991 | atomic_dec(&kvm->arch.noncoherent_dma_count); | |
10992 | } | |
10993 | EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma); | |
10994 | ||
10995 | bool kvm_arch_has_noncoherent_dma(struct kvm *kvm) | |
10996 | { | |
10997 | return atomic_read(&kvm->arch.noncoherent_dma_count); | |
10998 | } | |
10999 | EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma); | |
11000 | ||
14717e20 AW |
11001 | bool kvm_arch_has_irq_bypass(void) |
11002 | { | |
92735b1b | 11003 | return true; |
14717e20 AW |
11004 | } |
11005 | ||
87276880 FW |
11006 | int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons, |
11007 | struct irq_bypass_producer *prod) | |
11008 | { | |
11009 | struct kvm_kernel_irqfd *irqfd = | |
11010 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
2edd9cb7 | 11011 | int ret; |
87276880 | 11012 | |
14717e20 | 11013 | irqfd->producer = prod; |
2edd9cb7 ZL |
11014 | kvm_arch_start_assignment(irqfd->kvm); |
11015 | ret = kvm_x86_ops.update_pi_irte(irqfd->kvm, | |
11016 | prod->irq, irqfd->gsi, 1); | |
11017 | ||
11018 | if (ret) | |
11019 | kvm_arch_end_assignment(irqfd->kvm); | |
87276880 | 11020 | |
2edd9cb7 | 11021 | return ret; |
87276880 FW |
11022 | } |
11023 | ||
11024 | void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons, | |
11025 | struct irq_bypass_producer *prod) | |
11026 | { | |
11027 | int ret; | |
11028 | struct kvm_kernel_irqfd *irqfd = | |
11029 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
11030 | ||
87276880 FW |
11031 | WARN_ON(irqfd->producer != prod); |
11032 | irqfd->producer = NULL; | |
11033 | ||
11034 | /* | |
11035 | * When producer of consumer is unregistered, we change back to | |
11036 | * remapped mode, so we can re-use the current implementation | |
bb3541f1 | 11037 | * when the irq is masked/disabled or the consumer side (KVM |
87276880 FW |
11038 | * int this case doesn't want to receive the interrupts. |
11039 | */ | |
afaf0b2f | 11040 | ret = kvm_x86_ops.update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0); |
87276880 FW |
11041 | if (ret) |
11042 | printk(KERN_INFO "irq bypass consumer (token %p) unregistration" | |
11043 | " fails: %d\n", irqfd->consumer.token, ret); | |
2edd9cb7 ZL |
11044 | |
11045 | kvm_arch_end_assignment(irqfd->kvm); | |
87276880 FW |
11046 | } |
11047 | ||
11048 | int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq, | |
11049 | uint32_t guest_irq, bool set) | |
11050 | { | |
afaf0b2f | 11051 | return kvm_x86_ops.update_pi_irte(kvm, host_irq, guest_irq, set); |
87276880 FW |
11052 | } |
11053 | ||
52004014 FW |
11054 | bool kvm_vector_hashing_enabled(void) |
11055 | { | |
11056 | return vector_hashing; | |
11057 | } | |
52004014 | 11058 | |
2d5ba19b MT |
11059 | bool kvm_arch_no_poll(struct kvm_vcpu *vcpu) |
11060 | { | |
11061 | return (vcpu->arch.msr_kvm_poll_control & 1) == 0; | |
11062 | } | |
11063 | EXPORT_SYMBOL_GPL(kvm_arch_no_poll); | |
11064 | ||
841c2be0 ML |
11065 | |
11066 | int kvm_spec_ctrl_test_value(u64 value) | |
6441fa61 | 11067 | { |
841c2be0 ML |
11068 | /* |
11069 | * test that setting IA32_SPEC_CTRL to given value | |
11070 | * is allowed by the host processor | |
11071 | */ | |
6441fa61 | 11072 | |
841c2be0 ML |
11073 | u64 saved_value; |
11074 | unsigned long flags; | |
11075 | int ret = 0; | |
6441fa61 | 11076 | |
841c2be0 | 11077 | local_irq_save(flags); |
6441fa61 | 11078 | |
841c2be0 ML |
11079 | if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value)) |
11080 | ret = 1; | |
11081 | else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value)) | |
11082 | ret = 1; | |
11083 | else | |
11084 | wrmsrl(MSR_IA32_SPEC_CTRL, saved_value); | |
6441fa61 | 11085 | |
841c2be0 | 11086 | local_irq_restore(flags); |
6441fa61 | 11087 | |
841c2be0 | 11088 | return ret; |
6441fa61 | 11089 | } |
841c2be0 | 11090 | EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value); |
2d5ba19b | 11091 | |
89786147 MG |
11092 | void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code) |
11093 | { | |
11094 | struct x86_exception fault; | |
19cf4b7e PB |
11095 | u32 access = error_code & |
11096 | (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK); | |
89786147 MG |
11097 | |
11098 | if (!(error_code & PFERR_PRESENT_MASK) || | |
19cf4b7e | 11099 | vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) { |
89786147 MG |
11100 | /* |
11101 | * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page | |
11102 | * tables probably do not match the TLB. Just proceed | |
11103 | * with the error code that the processor gave. | |
11104 | */ | |
11105 | fault.vector = PF_VECTOR; | |
11106 | fault.error_code_valid = true; | |
11107 | fault.error_code = error_code; | |
11108 | fault.nested_page_fault = false; | |
11109 | fault.address = gva; | |
11110 | } | |
11111 | vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault); | |
6441fa61 | 11112 | } |
89786147 | 11113 | EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error); |
2d5ba19b | 11114 | |
3f3393b3 BM |
11115 | /* |
11116 | * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns | |
11117 | * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value | |
11118 | * indicates whether exit to userspace is needed. | |
11119 | */ | |
11120 | int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r, | |
11121 | struct x86_exception *e) | |
11122 | { | |
11123 | if (r == X86EMUL_PROPAGATE_FAULT) { | |
11124 | kvm_inject_emulated_page_fault(vcpu, e); | |
11125 | return 1; | |
11126 | } | |
11127 | ||
11128 | /* | |
11129 | * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED | |
11130 | * while handling a VMX instruction KVM could've handled the request | |
11131 | * correctly by exiting to userspace and performing I/O but there | |
11132 | * doesn't seem to be a real use-case behind such requests, just return | |
11133 | * KVM_EXIT_INTERNAL_ERROR for now. | |
11134 | */ | |
11135 | vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
11136 | vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
11137 | vcpu->run->internal.ndata = 0; | |
11138 | ||
11139 | return 0; | |
11140 | } | |
11141 | EXPORT_SYMBOL_GPL(kvm_handle_memory_failure); | |
11142 | ||
9715092f BM |
11143 | int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva) |
11144 | { | |
11145 | bool pcid_enabled; | |
11146 | struct x86_exception e; | |
11147 | unsigned i; | |
11148 | unsigned long roots_to_free = 0; | |
11149 | struct { | |
11150 | u64 pcid; | |
11151 | u64 gla; | |
11152 | } operand; | |
11153 | int r; | |
11154 | ||
11155 | r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e); | |
11156 | if (r != X86EMUL_CONTINUE) | |
11157 | return kvm_handle_memory_failure(vcpu, r, &e); | |
11158 | ||
11159 | if (operand.pcid >> 12 != 0) { | |
11160 | kvm_inject_gp(vcpu, 0); | |
11161 | return 1; | |
11162 | } | |
11163 | ||
11164 | pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE); | |
11165 | ||
11166 | switch (type) { | |
11167 | case INVPCID_TYPE_INDIV_ADDR: | |
11168 | if ((!pcid_enabled && (operand.pcid != 0)) || | |
11169 | is_noncanonical_address(operand.gla, vcpu)) { | |
11170 | kvm_inject_gp(vcpu, 0); | |
11171 | return 1; | |
11172 | } | |
11173 | kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid); | |
11174 | return kvm_skip_emulated_instruction(vcpu); | |
11175 | ||
11176 | case INVPCID_TYPE_SINGLE_CTXT: | |
11177 | if (!pcid_enabled && (operand.pcid != 0)) { | |
11178 | kvm_inject_gp(vcpu, 0); | |
11179 | return 1; | |
11180 | } | |
11181 | ||
11182 | if (kvm_get_active_pcid(vcpu) == operand.pcid) { | |
11183 | kvm_mmu_sync_roots(vcpu); | |
11184 | kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu); | |
11185 | } | |
11186 | ||
11187 | for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++) | |
11188 | if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd) | |
11189 | == operand.pcid) | |
11190 | roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i); | |
11191 | ||
11192 | kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free); | |
11193 | /* | |
11194 | * If neither the current cr3 nor any of the prev_roots use the | |
11195 | * given PCID, then nothing needs to be done here because a | |
11196 | * resync will happen anyway before switching to any other CR3. | |
11197 | */ | |
11198 | ||
11199 | return kvm_skip_emulated_instruction(vcpu); | |
11200 | ||
11201 | case INVPCID_TYPE_ALL_NON_GLOBAL: | |
11202 | /* | |
11203 | * Currently, KVM doesn't mark global entries in the shadow | |
11204 | * page tables, so a non-global flush just degenerates to a | |
11205 | * global flush. If needed, we could optimize this later by | |
11206 | * keeping track of global entries in shadow page tables. | |
11207 | */ | |
11208 | ||
11209 | fallthrough; | |
11210 | case INVPCID_TYPE_ALL_INCL_GLOBAL: | |
11211 | kvm_mmu_unload(vcpu); | |
11212 | return kvm_skip_emulated_instruction(vcpu); | |
11213 | ||
11214 | default: | |
11215 | BUG(); /* We have already checked above that type <= 3 */ | |
11216 | } | |
11217 | } | |
11218 | EXPORT_SYMBOL_GPL(kvm_handle_invpcid); | |
11219 | ||
229456fc | 11220 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit); |
931c33b1 | 11221 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio); |
229456fc MT |
11222 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq); |
11223 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault); | |
11224 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr); | |
11225 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr); | |
0ac406de | 11226 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun); |
d8cabddf | 11227 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit); |
17897f36 | 11228 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject); |
236649de | 11229 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit); |
5497b955 | 11230 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed); |
ec1ff790 | 11231 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga); |
532a46b9 | 11232 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit); |
2e554e8d | 11233 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts); |
489223ed | 11234 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset); |
4f75bcc3 | 11235 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update); |
843e4330 | 11236 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full); |
efc64404 | 11237 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update); |
18f40c53 SS |
11238 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access); |
11239 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi); | |
ab56f8e6 | 11240 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log); |
24bbf74c | 11241 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request); |