]>
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" |
1d737c8a | 21 | #include "mmu.h" |
7837699f | 22 | #include "i8254.h" |
37817f29 | 23 | #include "tss.h" |
5fdbf976 | 24 | #include "kvm_cache_regs.h" |
26eef70c | 25 | #include "x86.h" |
00b27a3e | 26 | #include "cpuid.h" |
474a5bb9 | 27 | #include "pmu.h" |
e83d5887 | 28 | #include "hyperv.h" |
313a3dc7 | 29 | |
18068523 | 30 | #include <linux/clocksource.h> |
4d5c5d0f | 31 | #include <linux/interrupt.h> |
313a3dc7 CO |
32 | #include <linux/kvm.h> |
33 | #include <linux/fs.h> | |
34 | #include <linux/vmalloc.h> | |
1767e931 PG |
35 | #include <linux/export.h> |
36 | #include <linux/moduleparam.h> | |
0de10343 | 37 | #include <linux/mman.h> |
2bacc55c | 38 | #include <linux/highmem.h> |
19de40a8 | 39 | #include <linux/iommu.h> |
62c476c7 | 40 | #include <linux/intel-iommu.h> |
c8076604 | 41 | #include <linux/cpufreq.h> |
18863bdd | 42 | #include <linux/user-return-notifier.h> |
a983fb23 | 43 | #include <linux/srcu.h> |
5a0e3ad6 | 44 | #include <linux/slab.h> |
ff9d07a0 | 45 | #include <linux/perf_event.h> |
7bee342a | 46 | #include <linux/uaccess.h> |
af585b92 | 47 | #include <linux/hash.h> |
a1b60c1c | 48 | #include <linux/pci.h> |
16e8d74d MT |
49 | #include <linux/timekeeper_internal.h> |
50 | #include <linux/pvclock_gtod.h> | |
87276880 FW |
51 | #include <linux/kvm_irqfd.h> |
52 | #include <linux/irqbypass.h> | |
3905f9ad | 53 | #include <linux/sched/stat.h> |
0c5f81da | 54 | #include <linux/sched/isolation.h> |
d0ec49d4 | 55 | #include <linux/mem_encrypt.h> |
3905f9ad | 56 | |
aec51dc4 | 57 | #include <trace/events/kvm.h> |
2ed152af | 58 | |
24f1e32c | 59 | #include <asm/debugreg.h> |
d825ed0a | 60 | #include <asm/msr.h> |
a5f61300 | 61 | #include <asm/desc.h> |
890ca9ae | 62 | #include <asm/mce.h> |
f89e32e0 | 63 | #include <linux/kernel_stat.h> |
78f7f1e5 | 64 | #include <asm/fpu/internal.h> /* Ugh! */ |
1d5f066e | 65 | #include <asm/pvclock.h> |
217fc9cf | 66 | #include <asm/div64.h> |
efc64404 | 67 | #include <asm/irq_remapping.h> |
b0c39dc6 | 68 | #include <asm/mshyperv.h> |
0092e434 | 69 | #include <asm/hypervisor.h> |
bf8c55d8 | 70 | #include <asm/intel_pt.h> |
b3dc0695 | 71 | #include <asm/emulate_prefix.h> |
dd2cb348 | 72 | #include <clocksource/hyperv_timer.h> |
043405e1 | 73 | |
d1898b73 DH |
74 | #define CREATE_TRACE_POINTS |
75 | #include "trace.h" | |
76 | ||
313a3dc7 | 77 | #define MAX_IO_MSRS 256 |
890ca9ae | 78 | #define KVM_MAX_MCE_BANKS 32 |
c45dcc71 AR |
79 | u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P; |
80 | EXPORT_SYMBOL_GPL(kvm_mce_cap_supported); | |
890ca9ae | 81 | |
0f65dd70 AK |
82 | #define emul_to_vcpu(ctxt) \ |
83 | container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt) | |
84 | ||
50a37eb4 JR |
85 | /* EFER defaults: |
86 | * - enable syscall per default because its emulated by KVM | |
87 | * - enable LME and LMA per default on 64 bit KVM | |
88 | */ | |
89 | #ifdef CONFIG_X86_64 | |
1260edbe LJ |
90 | static |
91 | u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA)); | |
50a37eb4 | 92 | #else |
1260edbe | 93 | static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE); |
50a37eb4 | 94 | #endif |
313a3dc7 | 95 | |
b11306b5 SC |
96 | static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS; |
97 | ||
833b45de PB |
98 | #define VM_STAT(x, ...) offsetof(struct kvm, stat.x), KVM_STAT_VM, ## __VA_ARGS__ |
99 | #define VCPU_STAT(x, ...) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU, ## __VA_ARGS__ | |
417bc304 | 100 | |
c519265f RK |
101 | #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \ |
102 | KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK) | |
37131313 | 103 | |
cb142eb7 | 104 | static void update_cr8_intercept(struct kvm_vcpu *vcpu); |
7460fb4a | 105 | static void process_nmi(struct kvm_vcpu *vcpu); |
ee2cd4b7 | 106 | static void enter_smm(struct kvm_vcpu *vcpu); |
6addfc42 | 107 | static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags); |
01643c51 KH |
108 | static void store_regs(struct kvm_vcpu *vcpu); |
109 | static int sync_regs(struct kvm_vcpu *vcpu); | |
674eea0f | 110 | |
893590c7 | 111 | struct kvm_x86_ops *kvm_x86_ops __read_mostly; |
5fdbf976 | 112 | EXPORT_SYMBOL_GPL(kvm_x86_ops); |
97896d04 | 113 | |
893590c7 | 114 | static bool __read_mostly ignore_msrs = 0; |
476bc001 | 115 | module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR); |
ed85c068 | 116 | |
fab0aa3b EM |
117 | static bool __read_mostly report_ignored_msrs = true; |
118 | module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR); | |
119 | ||
4c27625b | 120 | unsigned int min_timer_period_us = 200; |
9ed96e87 MT |
121 | module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR); |
122 | ||
630994b3 MT |
123 | static bool __read_mostly kvmclock_periodic_sync = true; |
124 | module_param(kvmclock_periodic_sync, bool, S_IRUGO); | |
125 | ||
893590c7 | 126 | bool __read_mostly kvm_has_tsc_control; |
92a1f12d | 127 | EXPORT_SYMBOL_GPL(kvm_has_tsc_control); |
893590c7 | 128 | u32 __read_mostly kvm_max_guest_tsc_khz; |
92a1f12d | 129 | EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz); |
bc9b961b HZ |
130 | u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits; |
131 | EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits); | |
132 | u64 __read_mostly kvm_max_tsc_scaling_ratio; | |
133 | EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio); | |
64672c95 YJ |
134 | u64 __read_mostly kvm_default_tsc_scaling_ratio; |
135 | EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio); | |
92a1f12d | 136 | |
cc578287 | 137 | /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */ |
893590c7 | 138 | static u32 __read_mostly tsc_tolerance_ppm = 250; |
cc578287 ZA |
139 | module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR); |
140 | ||
c3941d9e SC |
141 | /* |
142 | * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables | |
143 | * adaptive tuning starting from default advancment of 1000ns. '0' disables | |
144 | * advancement entirely. Any other value is used as-is and disables adaptive | |
145 | * tuning, i.e. allows priveleged userspace to set an exact advancement time. | |
146 | */ | |
147 | static int __read_mostly lapic_timer_advance_ns = -1; | |
0e6edceb | 148 | module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR); |
d0659d94 | 149 | |
52004014 FW |
150 | static bool __read_mostly vector_hashing = true; |
151 | module_param(vector_hashing, bool, S_IRUGO); | |
152 | ||
c4ae60e4 LA |
153 | bool __read_mostly enable_vmware_backdoor = false; |
154 | module_param(enable_vmware_backdoor, bool, S_IRUGO); | |
155 | EXPORT_SYMBOL_GPL(enable_vmware_backdoor); | |
156 | ||
6c86eedc WL |
157 | static bool __read_mostly force_emulation_prefix = false; |
158 | module_param(force_emulation_prefix, bool, S_IRUGO); | |
159 | ||
0c5f81da WL |
160 | int __read_mostly pi_inject_timer = -1; |
161 | module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR); | |
162 | ||
18863bdd AK |
163 | #define KVM_NR_SHARED_MSRS 16 |
164 | ||
165 | struct kvm_shared_msrs_global { | |
166 | int nr; | |
2bf78fa7 | 167 | u32 msrs[KVM_NR_SHARED_MSRS]; |
18863bdd AK |
168 | }; |
169 | ||
170 | struct kvm_shared_msrs { | |
171 | struct user_return_notifier urn; | |
172 | bool registered; | |
2bf78fa7 SY |
173 | struct kvm_shared_msr_values { |
174 | u64 host; | |
175 | u64 curr; | |
176 | } values[KVM_NR_SHARED_MSRS]; | |
18863bdd AK |
177 | }; |
178 | ||
179 | static struct kvm_shared_msrs_global __read_mostly shared_msrs_global; | |
013f6a5d | 180 | static struct kvm_shared_msrs __percpu *shared_msrs; |
18863bdd | 181 | |
139a12cf AL |
182 | static u64 __read_mostly host_xss; |
183 | ||
417bc304 | 184 | struct kvm_stats_debugfs_item debugfs_entries[] = { |
ba1389b7 AK |
185 | { "pf_fixed", VCPU_STAT(pf_fixed) }, |
186 | { "pf_guest", VCPU_STAT(pf_guest) }, | |
187 | { "tlb_flush", VCPU_STAT(tlb_flush) }, | |
188 | { "invlpg", VCPU_STAT(invlpg) }, | |
189 | { "exits", VCPU_STAT(exits) }, | |
190 | { "io_exits", VCPU_STAT(io_exits) }, | |
191 | { "mmio_exits", VCPU_STAT(mmio_exits) }, | |
192 | { "signal_exits", VCPU_STAT(signal_exits) }, | |
193 | { "irq_window", VCPU_STAT(irq_window_exits) }, | |
f08864b4 | 194 | { "nmi_window", VCPU_STAT(nmi_window_exits) }, |
ba1389b7 | 195 | { "halt_exits", VCPU_STAT(halt_exits) }, |
f7819512 | 196 | { "halt_successful_poll", VCPU_STAT(halt_successful_poll) }, |
62bea5bf | 197 | { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) }, |
3491caf2 | 198 | { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) }, |
ba1389b7 | 199 | { "halt_wakeup", VCPU_STAT(halt_wakeup) }, |
f11c3a8d | 200 | { "hypercalls", VCPU_STAT(hypercalls) }, |
ba1389b7 AK |
201 | { "request_irq", VCPU_STAT(request_irq_exits) }, |
202 | { "irq_exits", VCPU_STAT(irq_exits) }, | |
203 | { "host_state_reload", VCPU_STAT(host_state_reload) }, | |
ba1389b7 AK |
204 | { "fpu_reload", VCPU_STAT(fpu_reload) }, |
205 | { "insn_emulation", VCPU_STAT(insn_emulation) }, | |
206 | { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) }, | |
fa89a817 | 207 | { "irq_injections", VCPU_STAT(irq_injections) }, |
c4abb7c9 | 208 | { "nmi_injections", VCPU_STAT(nmi_injections) }, |
0f1e261e | 209 | { "req_event", VCPU_STAT(req_event) }, |
c595ceee | 210 | { "l1d_flush", VCPU_STAT(l1d_flush) }, |
4cee5764 AK |
211 | { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) }, |
212 | { "mmu_pte_write", VM_STAT(mmu_pte_write) }, | |
213 | { "mmu_pte_updated", VM_STAT(mmu_pte_updated) }, | |
214 | { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) }, | |
215 | { "mmu_flooded", VM_STAT(mmu_flooded) }, | |
216 | { "mmu_recycled", VM_STAT(mmu_recycled) }, | |
dfc5aa00 | 217 | { "mmu_cache_miss", VM_STAT(mmu_cache_miss) }, |
4731d4c7 | 218 | { "mmu_unsync", VM_STAT(mmu_unsync) }, |
0f74a24c | 219 | { "remote_tlb_flush", VM_STAT(remote_tlb_flush) }, |
833b45de | 220 | { "largepages", VM_STAT(lpages, .mode = 0444) }, |
b8e8c830 | 221 | { "nx_largepages_splitted", VM_STAT(nx_lpage_splits, .mode = 0444) }, |
f3414bc7 DM |
222 | { "max_mmu_page_hash_collisions", |
223 | VM_STAT(max_mmu_page_hash_collisions) }, | |
417bc304 HB |
224 | { NULL } |
225 | }; | |
226 | ||
2acf923e DC |
227 | u64 __read_mostly host_xcr0; |
228 | ||
b666a4b6 MO |
229 | struct kmem_cache *x86_fpu_cache; |
230 | EXPORT_SYMBOL_GPL(x86_fpu_cache); | |
231 | ||
b6785def | 232 | static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt); |
d6aa1000 | 233 | |
af585b92 GN |
234 | static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu) |
235 | { | |
236 | int i; | |
237 | for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++) | |
238 | vcpu->arch.apf.gfns[i] = ~0; | |
239 | } | |
240 | ||
18863bdd AK |
241 | static void kvm_on_user_return(struct user_return_notifier *urn) |
242 | { | |
243 | unsigned slot; | |
18863bdd AK |
244 | struct kvm_shared_msrs *locals |
245 | = container_of(urn, struct kvm_shared_msrs, urn); | |
2bf78fa7 | 246 | struct kvm_shared_msr_values *values; |
1650b4eb IA |
247 | unsigned long flags; |
248 | ||
249 | /* | |
250 | * Disabling irqs at this point since the following code could be | |
251 | * interrupted and executed through kvm_arch_hardware_disable() | |
252 | */ | |
253 | local_irq_save(flags); | |
254 | if (locals->registered) { | |
255 | locals->registered = false; | |
256 | user_return_notifier_unregister(urn); | |
257 | } | |
258 | local_irq_restore(flags); | |
18863bdd | 259 | for (slot = 0; slot < shared_msrs_global.nr; ++slot) { |
2bf78fa7 SY |
260 | values = &locals->values[slot]; |
261 | if (values->host != values->curr) { | |
262 | wrmsrl(shared_msrs_global.msrs[slot], values->host); | |
263 | values->curr = values->host; | |
18863bdd AK |
264 | } |
265 | } | |
18863bdd AK |
266 | } |
267 | ||
2bf78fa7 SY |
268 | void kvm_define_shared_msr(unsigned slot, u32 msr) |
269 | { | |
0123be42 | 270 | BUG_ON(slot >= KVM_NR_SHARED_MSRS); |
c847fe88 | 271 | shared_msrs_global.msrs[slot] = msr; |
18863bdd AK |
272 | if (slot >= shared_msrs_global.nr) |
273 | shared_msrs_global.nr = slot + 1; | |
18863bdd AK |
274 | } |
275 | EXPORT_SYMBOL_GPL(kvm_define_shared_msr); | |
276 | ||
277 | static void kvm_shared_msr_cpu_online(void) | |
278 | { | |
05c19c2f SC |
279 | unsigned int cpu = smp_processor_id(); |
280 | struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu); | |
281 | u64 value; | |
282 | int i; | |
18863bdd | 283 | |
05c19c2f SC |
284 | for (i = 0; i < shared_msrs_global.nr; ++i) { |
285 | rdmsrl_safe(shared_msrs_global.msrs[i], &value); | |
286 | smsr->values[i].host = value; | |
287 | smsr->values[i].curr = value; | |
288 | } | |
18863bdd AK |
289 | } |
290 | ||
8b3c3104 | 291 | int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask) |
18863bdd | 292 | { |
013f6a5d MT |
293 | unsigned int cpu = smp_processor_id(); |
294 | struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu); | |
8b3c3104 | 295 | int err; |
18863bdd | 296 | |
de1fca5d PB |
297 | value = (value & mask) | (smsr->values[slot].host & ~mask); |
298 | if (value == smsr->values[slot].curr) | |
8b3c3104 | 299 | return 0; |
8b3c3104 AH |
300 | err = wrmsrl_safe(shared_msrs_global.msrs[slot], value); |
301 | if (err) | |
302 | return 1; | |
303 | ||
de1fca5d | 304 | smsr->values[slot].curr = value; |
18863bdd AK |
305 | if (!smsr->registered) { |
306 | smsr->urn.on_user_return = kvm_on_user_return; | |
307 | user_return_notifier_register(&smsr->urn); | |
308 | smsr->registered = true; | |
309 | } | |
8b3c3104 | 310 | return 0; |
18863bdd AK |
311 | } |
312 | EXPORT_SYMBOL_GPL(kvm_set_shared_msr); | |
313 | ||
13a34e06 | 314 | static void drop_user_return_notifiers(void) |
3548bab5 | 315 | { |
013f6a5d MT |
316 | unsigned int cpu = smp_processor_id(); |
317 | struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu); | |
3548bab5 AK |
318 | |
319 | if (smsr->registered) | |
320 | kvm_on_user_return(&smsr->urn); | |
321 | } | |
322 | ||
6866b83e CO |
323 | u64 kvm_get_apic_base(struct kvm_vcpu *vcpu) |
324 | { | |
8a5a87d9 | 325 | return vcpu->arch.apic_base; |
6866b83e CO |
326 | } |
327 | EXPORT_SYMBOL_GPL(kvm_get_apic_base); | |
328 | ||
58871649 JM |
329 | enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu) |
330 | { | |
331 | return kvm_apic_mode(kvm_get_apic_base(vcpu)); | |
332 | } | |
333 | EXPORT_SYMBOL_GPL(kvm_get_apic_mode); | |
334 | ||
58cb628d JK |
335 | int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
336 | { | |
58871649 JM |
337 | enum lapic_mode old_mode = kvm_get_apic_mode(vcpu); |
338 | enum lapic_mode new_mode = kvm_apic_mode(msr_info->data); | |
d6321d49 RK |
339 | u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff | |
340 | (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE); | |
58cb628d | 341 | |
58871649 | 342 | if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID) |
58cb628d | 343 | return 1; |
58871649 JM |
344 | if (!msr_info->host_initiated) { |
345 | if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC) | |
346 | return 1; | |
347 | if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC) | |
348 | return 1; | |
349 | } | |
58cb628d JK |
350 | |
351 | kvm_lapic_set_base(vcpu, msr_info->data); | |
352 | return 0; | |
6866b83e CO |
353 | } |
354 | EXPORT_SYMBOL_GPL(kvm_set_apic_base); | |
355 | ||
2605fc21 | 356 | asmlinkage __visible void kvm_spurious_fault(void) |
e3ba45b8 GL |
357 | { |
358 | /* Fault while not rebooting. We want the trace. */ | |
b4fdcf60 | 359 | BUG_ON(!kvm_rebooting); |
e3ba45b8 GL |
360 | } |
361 | EXPORT_SYMBOL_GPL(kvm_spurious_fault); | |
362 | ||
3fd28fce ED |
363 | #define EXCPT_BENIGN 0 |
364 | #define EXCPT_CONTRIBUTORY 1 | |
365 | #define EXCPT_PF 2 | |
366 | ||
367 | static int exception_class(int vector) | |
368 | { | |
369 | switch (vector) { | |
370 | case PF_VECTOR: | |
371 | return EXCPT_PF; | |
372 | case DE_VECTOR: | |
373 | case TS_VECTOR: | |
374 | case NP_VECTOR: | |
375 | case SS_VECTOR: | |
376 | case GP_VECTOR: | |
377 | return EXCPT_CONTRIBUTORY; | |
378 | default: | |
379 | break; | |
380 | } | |
381 | return EXCPT_BENIGN; | |
382 | } | |
383 | ||
d6e8c854 NA |
384 | #define EXCPT_FAULT 0 |
385 | #define EXCPT_TRAP 1 | |
386 | #define EXCPT_ABORT 2 | |
387 | #define EXCPT_INTERRUPT 3 | |
388 | ||
389 | static int exception_type(int vector) | |
390 | { | |
391 | unsigned int mask; | |
392 | ||
393 | if (WARN_ON(vector > 31 || vector == NMI_VECTOR)) | |
394 | return EXCPT_INTERRUPT; | |
395 | ||
396 | mask = 1 << vector; | |
397 | ||
398 | /* #DB is trap, as instruction watchpoints are handled elsewhere */ | |
399 | if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR))) | |
400 | return EXCPT_TRAP; | |
401 | ||
402 | if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR))) | |
403 | return EXCPT_ABORT; | |
404 | ||
405 | /* Reserved exceptions will result in fault */ | |
406 | return EXCPT_FAULT; | |
407 | } | |
408 | ||
da998b46 JM |
409 | void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu) |
410 | { | |
411 | unsigned nr = vcpu->arch.exception.nr; | |
412 | bool has_payload = vcpu->arch.exception.has_payload; | |
413 | unsigned long payload = vcpu->arch.exception.payload; | |
414 | ||
415 | if (!has_payload) | |
416 | return; | |
417 | ||
418 | switch (nr) { | |
f10c729f JM |
419 | case DB_VECTOR: |
420 | /* | |
421 | * "Certain debug exceptions may clear bit 0-3. The | |
422 | * remaining contents of the DR6 register are never | |
423 | * cleared by the processor". | |
424 | */ | |
425 | vcpu->arch.dr6 &= ~DR_TRAP_BITS; | |
426 | /* | |
427 | * DR6.RTM is set by all #DB exceptions that don't clear it. | |
428 | */ | |
429 | vcpu->arch.dr6 |= DR6_RTM; | |
430 | vcpu->arch.dr6 |= payload; | |
431 | /* | |
432 | * Bit 16 should be set in the payload whenever the #DB | |
433 | * exception should clear DR6.RTM. This makes the payload | |
434 | * compatible with the pending debug exceptions under VMX. | |
435 | * Though not currently documented in the SDM, this also | |
436 | * makes the payload compatible with the exit qualification | |
437 | * for #DB exceptions under VMX. | |
438 | */ | |
439 | vcpu->arch.dr6 ^= payload & DR6_RTM; | |
440 | break; | |
da998b46 JM |
441 | case PF_VECTOR: |
442 | vcpu->arch.cr2 = payload; | |
443 | break; | |
444 | } | |
445 | ||
446 | vcpu->arch.exception.has_payload = false; | |
447 | vcpu->arch.exception.payload = 0; | |
448 | } | |
449 | EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload); | |
450 | ||
3fd28fce | 451 | static void kvm_multiple_exception(struct kvm_vcpu *vcpu, |
ce7ddec4 | 452 | unsigned nr, bool has_error, u32 error_code, |
91e86d22 | 453 | bool has_payload, unsigned long payload, bool reinject) |
3fd28fce ED |
454 | { |
455 | u32 prev_nr; | |
456 | int class1, class2; | |
457 | ||
3842d135 AK |
458 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
459 | ||
664f8e26 | 460 | if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) { |
3fd28fce | 461 | queue: |
3ffb2468 NA |
462 | if (has_error && !is_protmode(vcpu)) |
463 | has_error = false; | |
664f8e26 WL |
464 | if (reinject) { |
465 | /* | |
466 | * On vmentry, vcpu->arch.exception.pending is only | |
467 | * true if an event injection was blocked by | |
468 | * nested_run_pending. In that case, however, | |
469 | * vcpu_enter_guest requests an immediate exit, | |
470 | * and the guest shouldn't proceed far enough to | |
471 | * need reinjection. | |
472 | */ | |
473 | WARN_ON_ONCE(vcpu->arch.exception.pending); | |
474 | vcpu->arch.exception.injected = true; | |
91e86d22 JM |
475 | if (WARN_ON_ONCE(has_payload)) { |
476 | /* | |
477 | * A reinjected event has already | |
478 | * delivered its payload. | |
479 | */ | |
480 | has_payload = false; | |
481 | payload = 0; | |
482 | } | |
664f8e26 WL |
483 | } else { |
484 | vcpu->arch.exception.pending = true; | |
485 | vcpu->arch.exception.injected = false; | |
486 | } | |
3fd28fce ED |
487 | vcpu->arch.exception.has_error_code = has_error; |
488 | vcpu->arch.exception.nr = nr; | |
489 | vcpu->arch.exception.error_code = error_code; | |
91e86d22 JM |
490 | vcpu->arch.exception.has_payload = has_payload; |
491 | vcpu->arch.exception.payload = payload; | |
da998b46 JM |
492 | /* |
493 | * In guest mode, payload delivery should be deferred, | |
494 | * so that the L1 hypervisor can intercept #PF before | |
f10c729f JM |
495 | * CR2 is modified (or intercept #DB before DR6 is |
496 | * modified under nVMX). However, for ABI | |
497 | * compatibility with KVM_GET_VCPU_EVENTS and | |
498 | * KVM_SET_VCPU_EVENTS, we can't delay payload | |
499 | * delivery unless userspace has enabled this | |
500 | * functionality via the per-VM capability, | |
501 | * KVM_CAP_EXCEPTION_PAYLOAD. | |
da998b46 JM |
502 | */ |
503 | if (!vcpu->kvm->arch.exception_payload_enabled || | |
504 | !is_guest_mode(vcpu)) | |
505 | kvm_deliver_exception_payload(vcpu); | |
3fd28fce ED |
506 | return; |
507 | } | |
508 | ||
509 | /* to check exception */ | |
510 | prev_nr = vcpu->arch.exception.nr; | |
511 | if (prev_nr == DF_VECTOR) { | |
512 | /* triple fault -> shutdown */ | |
a8eeb04a | 513 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); |
3fd28fce ED |
514 | return; |
515 | } | |
516 | class1 = exception_class(prev_nr); | |
517 | class2 = exception_class(nr); | |
518 | if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY) | |
519 | || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) { | |
664f8e26 WL |
520 | /* |
521 | * Generate double fault per SDM Table 5-5. Set | |
522 | * exception.pending = true so that the double fault | |
523 | * can trigger a nested vmexit. | |
524 | */ | |
3fd28fce | 525 | vcpu->arch.exception.pending = true; |
664f8e26 | 526 | vcpu->arch.exception.injected = false; |
3fd28fce ED |
527 | vcpu->arch.exception.has_error_code = true; |
528 | vcpu->arch.exception.nr = DF_VECTOR; | |
529 | vcpu->arch.exception.error_code = 0; | |
c851436a JM |
530 | vcpu->arch.exception.has_payload = false; |
531 | vcpu->arch.exception.payload = 0; | |
3fd28fce ED |
532 | } else |
533 | /* replace previous exception with a new one in a hope | |
534 | that instruction re-execution will regenerate lost | |
535 | exception */ | |
536 | goto queue; | |
537 | } | |
538 | ||
298101da AK |
539 | void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr) |
540 | { | |
91e86d22 | 541 | kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false); |
298101da AK |
542 | } |
543 | EXPORT_SYMBOL_GPL(kvm_queue_exception); | |
544 | ||
ce7ddec4 JR |
545 | void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr) |
546 | { | |
91e86d22 | 547 | kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true); |
ce7ddec4 JR |
548 | } |
549 | EXPORT_SYMBOL_GPL(kvm_requeue_exception); | |
550 | ||
f10c729f JM |
551 | static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr, |
552 | unsigned long payload) | |
553 | { | |
554 | kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false); | |
555 | } | |
556 | ||
da998b46 JM |
557 | static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr, |
558 | u32 error_code, unsigned long payload) | |
559 | { | |
560 | kvm_multiple_exception(vcpu, nr, true, error_code, | |
561 | true, payload, false); | |
562 | } | |
563 | ||
6affcbed | 564 | int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err) |
c3c91fee | 565 | { |
db8fcefa AP |
566 | if (err) |
567 | kvm_inject_gp(vcpu, 0); | |
568 | else | |
6affcbed KH |
569 | return kvm_skip_emulated_instruction(vcpu); |
570 | ||
571 | return 1; | |
db8fcefa AP |
572 | } |
573 | EXPORT_SYMBOL_GPL(kvm_complete_insn_gp); | |
8df25a32 | 574 | |
6389ee94 | 575 | void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault) |
c3c91fee AK |
576 | { |
577 | ++vcpu->stat.pf_guest; | |
adfe20fb WL |
578 | vcpu->arch.exception.nested_apf = |
579 | is_guest_mode(vcpu) && fault->async_page_fault; | |
da998b46 | 580 | if (vcpu->arch.exception.nested_apf) { |
adfe20fb | 581 | vcpu->arch.apf.nested_apf_token = fault->address; |
da998b46 JM |
582 | kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code); |
583 | } else { | |
584 | kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code, | |
585 | fault->address); | |
586 | } | |
c3c91fee | 587 | } |
27d6c865 | 588 | EXPORT_SYMBOL_GPL(kvm_inject_page_fault); |
c3c91fee | 589 | |
ef54bcfe | 590 | static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault) |
d4f8cf66 | 591 | { |
6389ee94 AK |
592 | if (mmu_is_nested(vcpu) && !fault->nested_page_fault) |
593 | vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault); | |
d4f8cf66 | 594 | else |
44dd3ffa | 595 | vcpu->arch.mmu->inject_page_fault(vcpu, fault); |
ef54bcfe PB |
596 | |
597 | return fault->nested_page_fault; | |
d4f8cf66 JR |
598 | } |
599 | ||
3419ffc8 SY |
600 | void kvm_inject_nmi(struct kvm_vcpu *vcpu) |
601 | { | |
7460fb4a AK |
602 | atomic_inc(&vcpu->arch.nmi_queued); |
603 | kvm_make_request(KVM_REQ_NMI, vcpu); | |
3419ffc8 SY |
604 | } |
605 | EXPORT_SYMBOL_GPL(kvm_inject_nmi); | |
606 | ||
298101da AK |
607 | void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code) |
608 | { | |
91e86d22 | 609 | kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false); |
298101da AK |
610 | } |
611 | EXPORT_SYMBOL_GPL(kvm_queue_exception_e); | |
612 | ||
ce7ddec4 JR |
613 | void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code) |
614 | { | |
91e86d22 | 615 | kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true); |
ce7ddec4 JR |
616 | } |
617 | EXPORT_SYMBOL_GPL(kvm_requeue_exception_e); | |
618 | ||
0a79b009 AK |
619 | /* |
620 | * Checks if cpl <= required_cpl; if true, return true. Otherwise queue | |
621 | * a #GP and return false. | |
622 | */ | |
623 | bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl) | |
298101da | 624 | { |
0a79b009 AK |
625 | if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl) |
626 | return true; | |
627 | kvm_queue_exception_e(vcpu, GP_VECTOR, 0); | |
628 | return false; | |
298101da | 629 | } |
0a79b009 | 630 | EXPORT_SYMBOL_GPL(kvm_require_cpl); |
298101da | 631 | |
16f8a6f9 NA |
632 | bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr) |
633 | { | |
634 | if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE)) | |
635 | return true; | |
636 | ||
637 | kvm_queue_exception(vcpu, UD_VECTOR); | |
638 | return false; | |
639 | } | |
640 | EXPORT_SYMBOL_GPL(kvm_require_dr); | |
641 | ||
ec92fe44 JR |
642 | /* |
643 | * This function will be used to read from the physical memory of the currently | |
54bf36aa | 644 | * running guest. The difference to kvm_vcpu_read_guest_page is that this function |
ec92fe44 JR |
645 | * can read from guest physical or from the guest's guest physical memory. |
646 | */ | |
647 | int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, | |
648 | gfn_t ngfn, void *data, int offset, int len, | |
649 | u32 access) | |
650 | { | |
54987b7a | 651 | struct x86_exception exception; |
ec92fe44 JR |
652 | gfn_t real_gfn; |
653 | gpa_t ngpa; | |
654 | ||
655 | ngpa = gfn_to_gpa(ngfn); | |
54987b7a | 656 | real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception); |
ec92fe44 JR |
657 | if (real_gfn == UNMAPPED_GVA) |
658 | return -EFAULT; | |
659 | ||
660 | real_gfn = gpa_to_gfn(real_gfn); | |
661 | ||
54bf36aa | 662 | return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len); |
ec92fe44 JR |
663 | } |
664 | EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu); | |
665 | ||
69b0049a | 666 | static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
3d06b8bf JR |
667 | void *data, int offset, int len, u32 access) |
668 | { | |
669 | return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn, | |
670 | data, offset, len, access); | |
671 | } | |
672 | ||
16cfacc8 SC |
673 | static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu) |
674 | { | |
675 | return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) | | |
676 | rsvd_bits(1, 2); | |
677 | } | |
678 | ||
a03490ed | 679 | /* |
16cfacc8 | 680 | * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise. |
a03490ed | 681 | */ |
ff03a073 | 682 | int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3) |
a03490ed CO |
683 | { |
684 | gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT; | |
685 | unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2; | |
686 | int i; | |
687 | int ret; | |
ff03a073 | 688 | u64 pdpte[ARRAY_SIZE(mmu->pdptrs)]; |
a03490ed | 689 | |
ff03a073 JR |
690 | ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte, |
691 | offset * sizeof(u64), sizeof(pdpte), | |
692 | PFERR_USER_MASK|PFERR_WRITE_MASK); | |
a03490ed CO |
693 | if (ret < 0) { |
694 | ret = 0; | |
695 | goto out; | |
696 | } | |
697 | for (i = 0; i < ARRAY_SIZE(pdpte); ++i) { | |
812f30b2 | 698 | if ((pdpte[i] & PT_PRESENT_MASK) && |
16cfacc8 | 699 | (pdpte[i] & pdptr_rsvd_bits(vcpu))) { |
a03490ed CO |
700 | ret = 0; |
701 | goto out; | |
702 | } | |
703 | } | |
704 | ret = 1; | |
705 | ||
ff03a073 | 706 | memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs)); |
cb3c1e2f SC |
707 | kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR); |
708 | ||
a03490ed | 709 | out: |
a03490ed CO |
710 | |
711 | return ret; | |
712 | } | |
cc4b6871 | 713 | EXPORT_SYMBOL_GPL(load_pdptrs); |
a03490ed | 714 | |
9ed38ffa | 715 | bool pdptrs_changed(struct kvm_vcpu *vcpu) |
d835dfec | 716 | { |
ff03a073 | 717 | u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)]; |
3d06b8bf JR |
718 | int offset; |
719 | gfn_t gfn; | |
d835dfec AK |
720 | int r; |
721 | ||
bf03d4f9 | 722 | if (!is_pae_paging(vcpu)) |
d835dfec AK |
723 | return false; |
724 | ||
cb3c1e2f | 725 | if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR)) |
6de4f3ad AK |
726 | return true; |
727 | ||
a512177e PB |
728 | gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT; |
729 | offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1); | |
3d06b8bf JR |
730 | r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte), |
731 | PFERR_USER_MASK | PFERR_WRITE_MASK); | |
d835dfec | 732 | if (r < 0) |
7f7f0d9c | 733 | return true; |
d835dfec | 734 | |
7f7f0d9c | 735 | return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0; |
d835dfec | 736 | } |
9ed38ffa | 737 | EXPORT_SYMBOL_GPL(pdptrs_changed); |
d835dfec | 738 | |
49a9b07e | 739 | int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0) |
a03490ed | 740 | { |
aad82703 | 741 | unsigned long old_cr0 = kvm_read_cr0(vcpu); |
d81135a5 | 742 | unsigned long update_bits = X86_CR0_PG | X86_CR0_WP; |
aad82703 | 743 | |
f9a48e6a AK |
744 | cr0 |= X86_CR0_ET; |
745 | ||
ab344828 | 746 | #ifdef CONFIG_X86_64 |
0f12244f GN |
747 | if (cr0 & 0xffffffff00000000UL) |
748 | return 1; | |
ab344828 GN |
749 | #endif |
750 | ||
751 | cr0 &= ~CR0_RESERVED_BITS; | |
a03490ed | 752 | |
0f12244f GN |
753 | if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) |
754 | return 1; | |
a03490ed | 755 | |
0f12244f GN |
756 | if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) |
757 | return 1; | |
a03490ed CO |
758 | |
759 | if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) { | |
760 | #ifdef CONFIG_X86_64 | |
f6801dff | 761 | if ((vcpu->arch.efer & EFER_LME)) { |
a03490ed CO |
762 | int cs_db, cs_l; |
763 | ||
0f12244f GN |
764 | if (!is_pae(vcpu)) |
765 | return 1; | |
a03490ed | 766 | kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l); |
0f12244f GN |
767 | if (cs_l) |
768 | return 1; | |
a03490ed CO |
769 | } else |
770 | #endif | |
ff03a073 | 771 | if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, |
9f8fe504 | 772 | kvm_read_cr3(vcpu))) |
0f12244f | 773 | return 1; |
a03490ed CO |
774 | } |
775 | ||
ad756a16 MJ |
776 | if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)) |
777 | return 1; | |
778 | ||
a03490ed | 779 | kvm_x86_ops->set_cr0(vcpu, cr0); |
a03490ed | 780 | |
d170c419 | 781 | if ((cr0 ^ old_cr0) & X86_CR0_PG) { |
e5f3f027 | 782 | kvm_clear_async_pf_completion_queue(vcpu); |
d170c419 LJ |
783 | kvm_async_pf_hash_reset(vcpu); |
784 | } | |
e5f3f027 | 785 | |
aad82703 SY |
786 | if ((cr0 ^ old_cr0) & update_bits) |
787 | kvm_mmu_reset_context(vcpu); | |
b18d5431 | 788 | |
879ae188 LE |
789 | if (((cr0 ^ old_cr0) & X86_CR0_CD) && |
790 | kvm_arch_has_noncoherent_dma(vcpu->kvm) && | |
791 | !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED)) | |
b18d5431 XG |
792 | kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL); |
793 | ||
0f12244f GN |
794 | return 0; |
795 | } | |
2d3ad1f4 | 796 | EXPORT_SYMBOL_GPL(kvm_set_cr0); |
a03490ed | 797 | |
2d3ad1f4 | 798 | void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw) |
a03490ed | 799 | { |
49a9b07e | 800 | (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f)); |
a03490ed | 801 | } |
2d3ad1f4 | 802 | EXPORT_SYMBOL_GPL(kvm_lmsw); |
a03490ed | 803 | |
139a12cf | 804 | void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu) |
42bdf991 | 805 | { |
139a12cf AL |
806 | if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) { |
807 | ||
808 | if (vcpu->arch.xcr0 != host_xcr0) | |
809 | xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0); | |
810 | ||
811 | if (vcpu->arch.xsaves_enabled && | |
812 | vcpu->arch.ia32_xss != host_xss) | |
813 | wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss); | |
814 | } | |
42bdf991 | 815 | } |
139a12cf | 816 | EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state); |
42bdf991 | 817 | |
139a12cf | 818 | void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu) |
42bdf991 | 819 | { |
139a12cf AL |
820 | if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) { |
821 | ||
822 | if (vcpu->arch.xcr0 != host_xcr0) | |
823 | xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0); | |
824 | ||
825 | if (vcpu->arch.xsaves_enabled && | |
826 | vcpu->arch.ia32_xss != host_xss) | |
827 | wrmsrl(MSR_IA32_XSS, host_xss); | |
828 | } | |
829 | ||
42bdf991 | 830 | } |
139a12cf | 831 | EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state); |
42bdf991 | 832 | |
69b0049a | 833 | static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr) |
2acf923e | 834 | { |
56c103ec LJ |
835 | u64 xcr0 = xcr; |
836 | u64 old_xcr0 = vcpu->arch.xcr0; | |
46c34cb0 | 837 | u64 valid_bits; |
2acf923e DC |
838 | |
839 | /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */ | |
840 | if (index != XCR_XFEATURE_ENABLED_MASK) | |
841 | return 1; | |
d91cab78 | 842 | if (!(xcr0 & XFEATURE_MASK_FP)) |
2acf923e | 843 | return 1; |
d91cab78 | 844 | if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE)) |
2acf923e | 845 | return 1; |
46c34cb0 PB |
846 | |
847 | /* | |
848 | * Do not allow the guest to set bits that we do not support | |
849 | * saving. However, xcr0 bit 0 is always set, even if the | |
850 | * emulated CPU does not support XSAVE (see fx_init). | |
851 | */ | |
d91cab78 | 852 | valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP; |
46c34cb0 | 853 | if (xcr0 & ~valid_bits) |
2acf923e | 854 | return 1; |
46c34cb0 | 855 | |
d91cab78 DH |
856 | if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) != |
857 | (!(xcr0 & XFEATURE_MASK_BNDCSR))) | |
390bd528 LJ |
858 | return 1; |
859 | ||
d91cab78 DH |
860 | if (xcr0 & XFEATURE_MASK_AVX512) { |
861 | if (!(xcr0 & XFEATURE_MASK_YMM)) | |
612263b3 | 862 | return 1; |
d91cab78 | 863 | if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512) |
612263b3 CP |
864 | return 1; |
865 | } | |
2acf923e | 866 | vcpu->arch.xcr0 = xcr0; |
56c103ec | 867 | |
d91cab78 | 868 | if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND) |
56c103ec | 869 | kvm_update_cpuid(vcpu); |
2acf923e DC |
870 | return 0; |
871 | } | |
872 | ||
873 | int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr) | |
874 | { | |
764bcbc5 Z |
875 | if (kvm_x86_ops->get_cpl(vcpu) != 0 || |
876 | __kvm_set_xcr(vcpu, index, xcr)) { | |
2acf923e DC |
877 | kvm_inject_gp(vcpu, 0); |
878 | return 1; | |
879 | } | |
880 | return 0; | |
881 | } | |
882 | EXPORT_SYMBOL_GPL(kvm_set_xcr); | |
883 | ||
345599f9 SC |
884 | #define __cr4_reserved_bits(__cpu_has, __c) \ |
885 | ({ \ | |
886 | u64 __reserved_bits = CR4_RESERVED_BITS; \ | |
887 | \ | |
888 | if (!__cpu_has(__c, X86_FEATURE_XSAVE)) \ | |
889 | __reserved_bits |= X86_CR4_OSXSAVE; \ | |
890 | if (!__cpu_has(__c, X86_FEATURE_SMEP)) \ | |
891 | __reserved_bits |= X86_CR4_SMEP; \ | |
892 | if (!__cpu_has(__c, X86_FEATURE_SMAP)) \ | |
893 | __reserved_bits |= X86_CR4_SMAP; \ | |
894 | if (!__cpu_has(__c, X86_FEATURE_FSGSBASE)) \ | |
895 | __reserved_bits |= X86_CR4_FSGSBASE; \ | |
896 | if (!__cpu_has(__c, X86_FEATURE_PKU)) \ | |
897 | __reserved_bits |= X86_CR4_PKE; \ | |
898 | if (!__cpu_has(__c, X86_FEATURE_LA57)) \ | |
899 | __reserved_bits |= X86_CR4_LA57; \ | |
900 | __reserved_bits; \ | |
901 | }) | |
902 | ||
b11306b5 SC |
903 | static u64 kvm_host_cr4_reserved_bits(struct cpuinfo_x86 *c) |
904 | { | |
345599f9 | 905 | u64 reserved_bits = __cr4_reserved_bits(cpu_has, c); |
b11306b5 | 906 | |
87382003 | 907 | if (cpuid_ecx(0x7) & feature_bit(LA57)) |
345599f9 | 908 | reserved_bits &= ~X86_CR4_LA57; |
b11306b5 | 909 | |
345599f9 SC |
910 | if (kvm_x86_ops->umip_emulated()) |
911 | reserved_bits &= ~X86_CR4_UMIP; | |
b11306b5 SC |
912 | |
913 | return reserved_bits; | |
914 | } | |
915 | ||
3ca94192 | 916 | static int kvm_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4) |
a03490ed | 917 | { |
b11306b5 | 918 | if (cr4 & cr4_reserved_bits) |
3ca94192 | 919 | return -EINVAL; |
a03490ed | 920 | |
345599f9 | 921 | if (cr4 & __cr4_reserved_bits(guest_cpuid_has, vcpu)) |
3ca94192 WL |
922 | return -EINVAL; |
923 | ||
924 | return 0; | |
925 | } | |
926 | ||
927 | int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4) | |
928 | { | |
929 | unsigned long old_cr4 = kvm_read_cr4(vcpu); | |
930 | unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE | | |
931 | X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE; | |
932 | ||
933 | if (kvm_valid_cr4(vcpu, cr4)) | |
ae3e61e1 PB |
934 | return 1; |
935 | ||
a03490ed | 936 | if (is_long_mode(vcpu)) { |
0f12244f GN |
937 | if (!(cr4 & X86_CR4_PAE)) |
938 | return 1; | |
a2edf57f AK |
939 | } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE) |
940 | && ((cr4 ^ old_cr4) & pdptr_bits) | |
9f8fe504 AK |
941 | && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, |
942 | kvm_read_cr3(vcpu))) | |
0f12244f GN |
943 | return 1; |
944 | ||
ad756a16 | 945 | if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) { |
d6321d49 | 946 | if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID)) |
ad756a16 MJ |
947 | return 1; |
948 | ||
949 | /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */ | |
950 | if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu)) | |
951 | return 1; | |
952 | } | |
953 | ||
5e1746d6 | 954 | if (kvm_x86_ops->set_cr4(vcpu, cr4)) |
0f12244f | 955 | return 1; |
a03490ed | 956 | |
ad756a16 MJ |
957 | if (((cr4 ^ old_cr4) & pdptr_bits) || |
958 | (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE))) | |
aad82703 | 959 | kvm_mmu_reset_context(vcpu); |
0f12244f | 960 | |
b9baba86 | 961 | if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE)) |
00b27a3e | 962 | kvm_update_cpuid(vcpu); |
2acf923e | 963 | |
0f12244f GN |
964 | return 0; |
965 | } | |
2d3ad1f4 | 966 | EXPORT_SYMBOL_GPL(kvm_set_cr4); |
a03490ed | 967 | |
2390218b | 968 | int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3) |
a03490ed | 969 | { |
ade61e28 | 970 | bool skip_tlb_flush = false; |
ac146235 | 971 | #ifdef CONFIG_X86_64 |
c19986fe JS |
972 | bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE); |
973 | ||
ade61e28 | 974 | if (pcid_enabled) { |
208320ba JS |
975 | skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH; |
976 | cr3 &= ~X86_CR3_PCID_NOFLUSH; | |
ade61e28 | 977 | } |
ac146235 | 978 | #endif |
9d88fca7 | 979 | |
9f8fe504 | 980 | if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) { |
956bf353 JS |
981 | if (!skip_tlb_flush) { |
982 | kvm_mmu_sync_roots(vcpu); | |
ade61e28 | 983 | kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu); |
956bf353 | 984 | } |
0f12244f | 985 | return 0; |
d835dfec AK |
986 | } |
987 | ||
d1cd3ce9 | 988 | if (is_long_mode(vcpu) && |
a780a3ea | 989 | (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63))) |
d1cd3ce9 | 990 | return 1; |
bf03d4f9 PB |
991 | else if (is_pae_paging(vcpu) && |
992 | !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) | |
346874c9 | 993 | return 1; |
a03490ed | 994 | |
ade61e28 | 995 | kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush); |
0f12244f | 996 | vcpu->arch.cr3 = cr3; |
cb3c1e2f | 997 | kvm_register_mark_available(vcpu, VCPU_EXREG_CR3); |
7c390d35 | 998 | |
0f12244f GN |
999 | return 0; |
1000 | } | |
2d3ad1f4 | 1001 | EXPORT_SYMBOL_GPL(kvm_set_cr3); |
a03490ed | 1002 | |
eea1cff9 | 1003 | int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8) |
a03490ed | 1004 | { |
0f12244f GN |
1005 | if (cr8 & CR8_RESERVED_BITS) |
1006 | return 1; | |
35754c98 | 1007 | if (lapic_in_kernel(vcpu)) |
a03490ed CO |
1008 | kvm_lapic_set_tpr(vcpu, cr8); |
1009 | else | |
ad312c7c | 1010 | vcpu->arch.cr8 = cr8; |
0f12244f GN |
1011 | return 0; |
1012 | } | |
2d3ad1f4 | 1013 | EXPORT_SYMBOL_GPL(kvm_set_cr8); |
a03490ed | 1014 | |
2d3ad1f4 | 1015 | unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu) |
a03490ed | 1016 | { |
35754c98 | 1017 | if (lapic_in_kernel(vcpu)) |
a03490ed CO |
1018 | return kvm_lapic_get_cr8(vcpu); |
1019 | else | |
ad312c7c | 1020 | return vcpu->arch.cr8; |
a03490ed | 1021 | } |
2d3ad1f4 | 1022 | EXPORT_SYMBOL_GPL(kvm_get_cr8); |
a03490ed | 1023 | |
ae561ede NA |
1024 | static void kvm_update_dr0123(struct kvm_vcpu *vcpu) |
1025 | { | |
1026 | int i; | |
1027 | ||
1028 | if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) { | |
1029 | for (i = 0; i < KVM_NR_DB_REGS; i++) | |
1030 | vcpu->arch.eff_db[i] = vcpu->arch.db[i]; | |
1031 | vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD; | |
1032 | } | |
1033 | } | |
1034 | ||
73aaf249 JK |
1035 | static void kvm_update_dr6(struct kvm_vcpu *vcpu) |
1036 | { | |
1037 | if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) | |
1038 | kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6); | |
1039 | } | |
1040 | ||
c8639010 JK |
1041 | static void kvm_update_dr7(struct kvm_vcpu *vcpu) |
1042 | { | |
1043 | unsigned long dr7; | |
1044 | ||
1045 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) | |
1046 | dr7 = vcpu->arch.guest_debug_dr7; | |
1047 | else | |
1048 | dr7 = vcpu->arch.dr7; | |
1049 | kvm_x86_ops->set_dr7(vcpu, dr7); | |
360b948d PB |
1050 | vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED; |
1051 | if (dr7 & DR7_BP_EN_MASK) | |
1052 | vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED; | |
c8639010 JK |
1053 | } |
1054 | ||
6f43ed01 NA |
1055 | static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu) |
1056 | { | |
1057 | u64 fixed = DR6_FIXED_1; | |
1058 | ||
d6321d49 | 1059 | if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM)) |
6f43ed01 NA |
1060 | fixed |= DR6_RTM; |
1061 | return fixed; | |
1062 | } | |
1063 | ||
338dbc97 | 1064 | static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val) |
020df079 | 1065 | { |
ea740059 MP |
1066 | size_t size = ARRAY_SIZE(vcpu->arch.db); |
1067 | ||
020df079 GN |
1068 | switch (dr) { |
1069 | case 0 ... 3: | |
ea740059 | 1070 | vcpu->arch.db[array_index_nospec(dr, size)] = val; |
020df079 GN |
1071 | if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) |
1072 | vcpu->arch.eff_db[dr] = val; | |
1073 | break; | |
1074 | case 4: | |
020df079 GN |
1075 | /* fall through */ |
1076 | case 6: | |
338dbc97 GN |
1077 | if (val & 0xffffffff00000000ULL) |
1078 | return -1; /* #GP */ | |
6f43ed01 | 1079 | vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu); |
73aaf249 | 1080 | kvm_update_dr6(vcpu); |
020df079 GN |
1081 | break; |
1082 | case 5: | |
020df079 GN |
1083 | /* fall through */ |
1084 | default: /* 7 */ | |
338dbc97 GN |
1085 | if (val & 0xffffffff00000000ULL) |
1086 | return -1; /* #GP */ | |
020df079 | 1087 | vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1; |
c8639010 | 1088 | kvm_update_dr7(vcpu); |
020df079 GN |
1089 | break; |
1090 | } | |
1091 | ||
1092 | return 0; | |
1093 | } | |
338dbc97 GN |
1094 | |
1095 | int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val) | |
1096 | { | |
16f8a6f9 | 1097 | if (__kvm_set_dr(vcpu, dr, val)) { |
338dbc97 | 1098 | kvm_inject_gp(vcpu, 0); |
16f8a6f9 NA |
1099 | return 1; |
1100 | } | |
1101 | return 0; | |
338dbc97 | 1102 | } |
020df079 GN |
1103 | EXPORT_SYMBOL_GPL(kvm_set_dr); |
1104 | ||
16f8a6f9 | 1105 | int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val) |
020df079 | 1106 | { |
ea740059 MP |
1107 | size_t size = ARRAY_SIZE(vcpu->arch.db); |
1108 | ||
020df079 GN |
1109 | switch (dr) { |
1110 | case 0 ... 3: | |
ea740059 | 1111 | *val = vcpu->arch.db[array_index_nospec(dr, size)]; |
020df079 GN |
1112 | break; |
1113 | case 4: | |
020df079 GN |
1114 | /* fall through */ |
1115 | case 6: | |
73aaf249 JK |
1116 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) |
1117 | *val = vcpu->arch.dr6; | |
1118 | else | |
1119 | *val = kvm_x86_ops->get_dr6(vcpu); | |
020df079 GN |
1120 | break; |
1121 | case 5: | |
020df079 GN |
1122 | /* fall through */ |
1123 | default: /* 7 */ | |
1124 | *val = vcpu->arch.dr7; | |
1125 | break; | |
1126 | } | |
338dbc97 GN |
1127 | return 0; |
1128 | } | |
020df079 GN |
1129 | EXPORT_SYMBOL_GPL(kvm_get_dr); |
1130 | ||
022cd0e8 AK |
1131 | bool kvm_rdpmc(struct kvm_vcpu *vcpu) |
1132 | { | |
de3cd117 | 1133 | u32 ecx = kvm_rcx_read(vcpu); |
022cd0e8 AK |
1134 | u64 data; |
1135 | int err; | |
1136 | ||
c6702c9d | 1137 | err = kvm_pmu_rdpmc(vcpu, ecx, &data); |
022cd0e8 AK |
1138 | if (err) |
1139 | return err; | |
de3cd117 SC |
1140 | kvm_rax_write(vcpu, (u32)data); |
1141 | kvm_rdx_write(vcpu, data >> 32); | |
022cd0e8 AK |
1142 | return err; |
1143 | } | |
1144 | EXPORT_SYMBOL_GPL(kvm_rdpmc); | |
1145 | ||
043405e1 CO |
1146 | /* |
1147 | * List of msr numbers which we expose to userspace through KVM_GET_MSRS | |
1148 | * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST. | |
1149 | * | |
7a5ee6ed CQ |
1150 | * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features) |
1151 | * extract the supported MSRs from the related const lists. | |
1152 | * msrs_to_save is selected from the msrs_to_save_all to reflect the | |
e3267cbb | 1153 | * capabilities of the host cpu. This capabilities test skips MSRs that are |
7a5ee6ed | 1154 | * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs |
62ef68bb | 1155 | * may depend on host virtualization features rather than host cpu features. |
043405e1 | 1156 | */ |
e3267cbb | 1157 | |
7a5ee6ed | 1158 | static const u32 msrs_to_save_all[] = { |
043405e1 | 1159 | MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP, |
8c06585d | 1160 | MSR_STAR, |
043405e1 CO |
1161 | #ifdef CONFIG_X86_64 |
1162 | MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR, | |
1163 | #endif | |
b3897a49 | 1164 | MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA, |
9dbe6cf9 | 1165 | MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX, |
2bdb76c0 | 1166 | MSR_IA32_SPEC_CTRL, |
bf8c55d8 CP |
1167 | MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH, |
1168 | MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK, | |
1169 | MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B, | |
1170 | MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B, | |
1171 | MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B, | |
1172 | MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B, | |
6e3ba4ab TX |
1173 | MSR_IA32_UMWAIT_CONTROL, |
1174 | ||
e2ada66e JM |
1175 | MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1, |
1176 | MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3, | |
1177 | MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS, | |
1178 | MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL, | |
1179 | MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1, | |
1180 | MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3, | |
1181 | MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5, | |
1182 | MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7, | |
1183 | MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9, | |
1184 | MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11, | |
1185 | MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13, | |
1186 | MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15, | |
1187 | MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17, | |
e2ada66e JM |
1188 | MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1, |
1189 | MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3, | |
1190 | MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5, | |
1191 | MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7, | |
1192 | MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9, | |
1193 | MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11, | |
1194 | MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13, | |
1195 | MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15, | |
1196 | MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17, | |
043405e1 CO |
1197 | }; |
1198 | ||
7a5ee6ed | 1199 | static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)]; |
043405e1 CO |
1200 | static unsigned num_msrs_to_save; |
1201 | ||
7a5ee6ed | 1202 | static const u32 emulated_msrs_all[] = { |
62ef68bb PB |
1203 | MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK, |
1204 | MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW, | |
1205 | HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL, | |
1206 | HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC, | |
72c139ba | 1207 | HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY, |
e7d9513b AS |
1208 | HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2, |
1209 | HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL, | |
e516cebb | 1210 | HV_X64_MSR_RESET, |
11c4b1ca | 1211 | HV_X64_MSR_VP_INDEX, |
9eec50b8 | 1212 | HV_X64_MSR_VP_RUNTIME, |
5c919412 | 1213 | HV_X64_MSR_SCONTROL, |
1f4b34f8 | 1214 | HV_X64_MSR_STIMER0_CONFIG, |
d4abc577 | 1215 | HV_X64_MSR_VP_ASSIST_PAGE, |
a2e164e7 VK |
1216 | HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL, |
1217 | HV_X64_MSR_TSC_EMULATION_STATUS, | |
1218 | ||
1219 | MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME, | |
62ef68bb PB |
1220 | MSR_KVM_PV_EOI_EN, |
1221 | ||
ba904635 | 1222 | MSR_IA32_TSC_ADJUST, |
a3e06bbe | 1223 | MSR_IA32_TSCDEADLINE, |
2bdb76c0 | 1224 | MSR_IA32_ARCH_CAPABILITIES, |
043405e1 | 1225 | MSR_IA32_MISC_ENABLE, |
908e75f3 AK |
1226 | MSR_IA32_MCG_STATUS, |
1227 | MSR_IA32_MCG_CTL, | |
c45dcc71 | 1228 | MSR_IA32_MCG_EXT_CTL, |
64d60670 | 1229 | MSR_IA32_SMBASE, |
52797bf9 | 1230 | MSR_SMI_COUNT, |
db2336a8 KH |
1231 | MSR_PLATFORM_INFO, |
1232 | MSR_MISC_FEATURES_ENABLES, | |
bc226f07 | 1233 | MSR_AMD64_VIRT_SPEC_CTRL, |
6c6a2ab9 | 1234 | MSR_IA32_POWER_CTL, |
99634e3e | 1235 | MSR_IA32_UCODE_REV, |
191c8137 | 1236 | |
95c5c7c7 PB |
1237 | /* |
1238 | * The following list leaves out MSRs whose values are determined | |
1239 | * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs. | |
1240 | * We always support the "true" VMX control MSRs, even if the host | |
1241 | * processor does not, so I am putting these registers here rather | |
7a5ee6ed | 1242 | * than in msrs_to_save_all. |
95c5c7c7 PB |
1243 | */ |
1244 | MSR_IA32_VMX_BASIC, | |
1245 | MSR_IA32_VMX_TRUE_PINBASED_CTLS, | |
1246 | MSR_IA32_VMX_TRUE_PROCBASED_CTLS, | |
1247 | MSR_IA32_VMX_TRUE_EXIT_CTLS, | |
1248 | MSR_IA32_VMX_TRUE_ENTRY_CTLS, | |
1249 | MSR_IA32_VMX_MISC, | |
1250 | MSR_IA32_VMX_CR0_FIXED0, | |
1251 | MSR_IA32_VMX_CR4_FIXED0, | |
1252 | MSR_IA32_VMX_VMCS_ENUM, | |
1253 | MSR_IA32_VMX_PROCBASED_CTLS2, | |
1254 | MSR_IA32_VMX_EPT_VPID_CAP, | |
1255 | MSR_IA32_VMX_VMFUNC, | |
1256 | ||
191c8137 | 1257 | MSR_K7_HWCR, |
2d5ba19b | 1258 | MSR_KVM_POLL_CONTROL, |
043405e1 CO |
1259 | }; |
1260 | ||
7a5ee6ed | 1261 | static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)]; |
62ef68bb PB |
1262 | static unsigned num_emulated_msrs; |
1263 | ||
801e459a TL |
1264 | /* |
1265 | * List of msr numbers which are used to expose MSR-based features that | |
1266 | * can be used by a hypervisor to validate requested CPU features. | |
1267 | */ | |
7a5ee6ed | 1268 | static const u32 msr_based_features_all[] = { |
1389309c PB |
1269 | MSR_IA32_VMX_BASIC, |
1270 | MSR_IA32_VMX_TRUE_PINBASED_CTLS, | |
1271 | MSR_IA32_VMX_PINBASED_CTLS, | |
1272 | MSR_IA32_VMX_TRUE_PROCBASED_CTLS, | |
1273 | MSR_IA32_VMX_PROCBASED_CTLS, | |
1274 | MSR_IA32_VMX_TRUE_EXIT_CTLS, | |
1275 | MSR_IA32_VMX_EXIT_CTLS, | |
1276 | MSR_IA32_VMX_TRUE_ENTRY_CTLS, | |
1277 | MSR_IA32_VMX_ENTRY_CTLS, | |
1278 | MSR_IA32_VMX_MISC, | |
1279 | MSR_IA32_VMX_CR0_FIXED0, | |
1280 | MSR_IA32_VMX_CR0_FIXED1, | |
1281 | MSR_IA32_VMX_CR4_FIXED0, | |
1282 | MSR_IA32_VMX_CR4_FIXED1, | |
1283 | MSR_IA32_VMX_VMCS_ENUM, | |
1284 | MSR_IA32_VMX_PROCBASED_CTLS2, | |
1285 | MSR_IA32_VMX_EPT_VPID_CAP, | |
1286 | MSR_IA32_VMX_VMFUNC, | |
1287 | ||
d1d93fa9 | 1288 | MSR_F10H_DECFG, |
518e7b94 | 1289 | MSR_IA32_UCODE_REV, |
cd283252 | 1290 | MSR_IA32_ARCH_CAPABILITIES, |
801e459a TL |
1291 | }; |
1292 | ||
7a5ee6ed | 1293 | static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)]; |
801e459a TL |
1294 | static unsigned int num_msr_based_features; |
1295 | ||
4d22c17c | 1296 | static u64 kvm_get_arch_capabilities(void) |
5b76a3cf | 1297 | { |
4d22c17c | 1298 | u64 data = 0; |
5b76a3cf | 1299 | |
4d22c17c XL |
1300 | if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) |
1301 | rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data); | |
5b76a3cf | 1302 | |
b8e8c830 PB |
1303 | /* |
1304 | * If nx_huge_pages is enabled, KVM's shadow paging will ensure that | |
1305 | * the nested hypervisor runs with NX huge pages. If it is not, | |
1306 | * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other | |
1307 | * L1 guests, so it need not worry about its own (L2) guests. | |
1308 | */ | |
1309 | data |= ARCH_CAP_PSCHANGE_MC_NO; | |
1310 | ||
5b76a3cf PB |
1311 | /* |
1312 | * If we're doing cache flushes (either "always" or "cond") | |
1313 | * we will do one whenever the guest does a vmlaunch/vmresume. | |
1314 | * If an outer hypervisor is doing the cache flush for us | |
1315 | * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that | |
1316 | * capability to the guest too, and if EPT is disabled we're not | |
1317 | * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will | |
1318 | * require a nested hypervisor to do a flush of its own. | |
1319 | */ | |
1320 | if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER) | |
1321 | data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH; | |
1322 | ||
0c54914d PB |
1323 | if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN)) |
1324 | data |= ARCH_CAP_RDCL_NO; | |
1325 | if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS)) | |
1326 | data |= ARCH_CAP_SSB_NO; | |
1327 | if (!boot_cpu_has_bug(X86_BUG_MDS)) | |
1328 | data |= ARCH_CAP_MDS_NO; | |
1329 | ||
e1d38b63 | 1330 | /* |
c11f83e0 PB |
1331 | * On TAA affected systems: |
1332 | * - nothing to do if TSX is disabled on the host. | |
1333 | * - we emulate TSX_CTRL if present on the host. | |
1334 | * This lets the guest use VERW to clear CPU buffers. | |
e1d38b63 | 1335 | */ |
cbbaa272 | 1336 | if (!boot_cpu_has(X86_FEATURE_RTM)) |
c11f83e0 | 1337 | data &= ~(ARCH_CAP_TAA_NO | ARCH_CAP_TSX_CTRL_MSR); |
cbbaa272 PB |
1338 | else if (!boot_cpu_has_bug(X86_BUG_TAA)) |
1339 | data |= ARCH_CAP_TAA_NO; | |
e1d38b63 | 1340 | |
5b76a3cf PB |
1341 | return data; |
1342 | } | |
5b76a3cf | 1343 | |
66421c1e WL |
1344 | static int kvm_get_msr_feature(struct kvm_msr_entry *msr) |
1345 | { | |
1346 | switch (msr->index) { | |
cd283252 | 1347 | case MSR_IA32_ARCH_CAPABILITIES: |
5b76a3cf PB |
1348 | msr->data = kvm_get_arch_capabilities(); |
1349 | break; | |
1350 | case MSR_IA32_UCODE_REV: | |
cd283252 | 1351 | rdmsrl_safe(msr->index, &msr->data); |
518e7b94 | 1352 | break; |
66421c1e WL |
1353 | default: |
1354 | if (kvm_x86_ops->get_msr_feature(msr)) | |
1355 | return 1; | |
1356 | } | |
1357 | return 0; | |
1358 | } | |
1359 | ||
801e459a TL |
1360 | static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data) |
1361 | { | |
1362 | struct kvm_msr_entry msr; | |
66421c1e | 1363 | int r; |
801e459a TL |
1364 | |
1365 | msr.index = index; | |
66421c1e WL |
1366 | r = kvm_get_msr_feature(&msr); |
1367 | if (r) | |
1368 | return r; | |
801e459a TL |
1369 | |
1370 | *data = msr.data; | |
1371 | ||
1372 | return 0; | |
1373 | } | |
1374 | ||
11988499 | 1375 | static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer) |
15c4a640 | 1376 | { |
1b4d56b8 | 1377 | if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT)) |
11988499 | 1378 | return false; |
1b2fd70c | 1379 | |
1b4d56b8 | 1380 | if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM)) |
11988499 | 1381 | return false; |
d8017474 | 1382 | |
0a629563 SC |
1383 | if (efer & (EFER_LME | EFER_LMA) && |
1384 | !guest_cpuid_has(vcpu, X86_FEATURE_LM)) | |
1385 | return false; | |
1386 | ||
1387 | if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX)) | |
1388 | return false; | |
d8017474 | 1389 | |
384bb783 | 1390 | return true; |
11988499 SC |
1391 | |
1392 | } | |
1393 | bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer) | |
1394 | { | |
1395 | if (efer & efer_reserved_bits) | |
1396 | return false; | |
1397 | ||
1398 | return __kvm_valid_efer(vcpu, efer); | |
384bb783 JK |
1399 | } |
1400 | EXPORT_SYMBOL_GPL(kvm_valid_efer); | |
1401 | ||
11988499 | 1402 | static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
384bb783 JK |
1403 | { |
1404 | u64 old_efer = vcpu->arch.efer; | |
11988499 | 1405 | u64 efer = msr_info->data; |
384bb783 | 1406 | |
11988499 | 1407 | if (efer & efer_reserved_bits) |
66f61c92 | 1408 | return 1; |
384bb783 | 1409 | |
11988499 SC |
1410 | if (!msr_info->host_initiated) { |
1411 | if (!__kvm_valid_efer(vcpu, efer)) | |
1412 | return 1; | |
1413 | ||
1414 | if (is_paging(vcpu) && | |
1415 | (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME)) | |
1416 | return 1; | |
1417 | } | |
384bb783 | 1418 | |
15c4a640 | 1419 | efer &= ~EFER_LMA; |
f6801dff | 1420 | efer |= vcpu->arch.efer & EFER_LMA; |
15c4a640 | 1421 | |
a3d204e2 SY |
1422 | kvm_x86_ops->set_efer(vcpu, efer); |
1423 | ||
aad82703 SY |
1424 | /* Update reserved bits */ |
1425 | if ((efer ^ old_efer) & EFER_NX) | |
1426 | kvm_mmu_reset_context(vcpu); | |
1427 | ||
b69e8cae | 1428 | return 0; |
15c4a640 CO |
1429 | } |
1430 | ||
f2b4b7dd JR |
1431 | void kvm_enable_efer_bits(u64 mask) |
1432 | { | |
1433 | efer_reserved_bits &= ~mask; | |
1434 | } | |
1435 | EXPORT_SYMBOL_GPL(kvm_enable_efer_bits); | |
1436 | ||
15c4a640 | 1437 | /* |
f20935d8 SC |
1438 | * Write @data into the MSR specified by @index. Select MSR specific fault |
1439 | * checks are bypassed if @host_initiated is %true. | |
15c4a640 CO |
1440 | * Returns 0 on success, non-0 otherwise. |
1441 | * Assumes vcpu_load() was already called. | |
1442 | */ | |
f20935d8 SC |
1443 | static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data, |
1444 | bool host_initiated) | |
15c4a640 | 1445 | { |
f20935d8 SC |
1446 | struct msr_data msr; |
1447 | ||
1448 | switch (index) { | |
854e8bb1 NA |
1449 | case MSR_FS_BASE: |
1450 | case MSR_GS_BASE: | |
1451 | case MSR_KERNEL_GS_BASE: | |
1452 | case MSR_CSTAR: | |
1453 | case MSR_LSTAR: | |
f20935d8 | 1454 | if (is_noncanonical_address(data, vcpu)) |
854e8bb1 NA |
1455 | return 1; |
1456 | break; | |
1457 | case MSR_IA32_SYSENTER_EIP: | |
1458 | case MSR_IA32_SYSENTER_ESP: | |
1459 | /* | |
1460 | * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if | |
1461 | * non-canonical address is written on Intel but not on | |
1462 | * AMD (which ignores the top 32-bits, because it does | |
1463 | * not implement 64-bit SYSENTER). | |
1464 | * | |
1465 | * 64-bit code should hence be able to write a non-canonical | |
1466 | * value on AMD. Making the address canonical ensures that | |
1467 | * vmentry does not fail on Intel after writing a non-canonical | |
1468 | * value, and that something deterministic happens if the guest | |
1469 | * invokes 64-bit SYSENTER. | |
1470 | */ | |
f20935d8 | 1471 | data = get_canonical(data, vcpu_virt_addr_bits(vcpu)); |
854e8bb1 | 1472 | } |
f20935d8 SC |
1473 | |
1474 | msr.data = data; | |
1475 | msr.index = index; | |
1476 | msr.host_initiated = host_initiated; | |
1477 | ||
1478 | return kvm_x86_ops->set_msr(vcpu, &msr); | |
15c4a640 CO |
1479 | } |
1480 | ||
313a3dc7 | 1481 | /* |
f20935d8 SC |
1482 | * Read the MSR specified by @index into @data. Select MSR specific fault |
1483 | * checks are bypassed if @host_initiated is %true. | |
1484 | * Returns 0 on success, non-0 otherwise. | |
1485 | * Assumes vcpu_load() was already called. | |
313a3dc7 | 1486 | */ |
edef5c36 PB |
1487 | int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data, |
1488 | bool host_initiated) | |
609e36d3 PB |
1489 | { |
1490 | struct msr_data msr; | |
f20935d8 | 1491 | int ret; |
609e36d3 PB |
1492 | |
1493 | msr.index = index; | |
f20935d8 | 1494 | msr.host_initiated = host_initiated; |
609e36d3 | 1495 | |
f20935d8 SC |
1496 | ret = kvm_x86_ops->get_msr(vcpu, &msr); |
1497 | if (!ret) | |
1498 | *data = msr.data; | |
1499 | return ret; | |
609e36d3 PB |
1500 | } |
1501 | ||
f20935d8 | 1502 | int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data) |
313a3dc7 | 1503 | { |
f20935d8 SC |
1504 | return __kvm_get_msr(vcpu, index, data, false); |
1505 | } | |
1506 | EXPORT_SYMBOL_GPL(kvm_get_msr); | |
8fe8ab46 | 1507 | |
f20935d8 SC |
1508 | int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data) |
1509 | { | |
1510 | return __kvm_set_msr(vcpu, index, data, false); | |
1511 | } | |
1512 | EXPORT_SYMBOL_GPL(kvm_set_msr); | |
1513 | ||
1edce0a9 SC |
1514 | int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu) |
1515 | { | |
1516 | u32 ecx = kvm_rcx_read(vcpu); | |
1517 | u64 data; | |
1518 | ||
1519 | if (kvm_get_msr(vcpu, ecx, &data)) { | |
1520 | trace_kvm_msr_read_ex(ecx); | |
1521 | kvm_inject_gp(vcpu, 0); | |
1522 | return 1; | |
1523 | } | |
1524 | ||
1525 | trace_kvm_msr_read(ecx, data); | |
1526 | ||
1527 | kvm_rax_write(vcpu, data & -1u); | |
1528 | kvm_rdx_write(vcpu, (data >> 32) & -1u); | |
1529 | return kvm_skip_emulated_instruction(vcpu); | |
1530 | } | |
1531 | EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr); | |
1532 | ||
1533 | int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu) | |
1534 | { | |
1535 | u32 ecx = kvm_rcx_read(vcpu); | |
1536 | u64 data = kvm_read_edx_eax(vcpu); | |
1537 | ||
1538 | if (kvm_set_msr(vcpu, ecx, data)) { | |
1539 | trace_kvm_msr_write_ex(ecx, data); | |
1540 | kvm_inject_gp(vcpu, 0); | |
1541 | return 1; | |
1542 | } | |
1543 | ||
1544 | trace_kvm_msr_write(ecx, data); | |
1545 | return kvm_skip_emulated_instruction(vcpu); | |
1546 | } | |
1547 | EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr); | |
1548 | ||
1e9e2622 WL |
1549 | /* |
1550 | * The fast path for frequent and performance sensitive wrmsr emulation, | |
1551 | * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces | |
1552 | * the latency of virtual IPI by avoiding the expensive bits of transitioning | |
1553 | * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the | |
1554 | * other cases which must be called after interrupts are enabled on the host. | |
1555 | */ | |
1556 | static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data) | |
1557 | { | |
1558 | if (lapic_in_kernel(vcpu) && apic_x2apic_mode(vcpu->arch.apic) && | |
1559 | ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) && | |
1560 | ((data & APIC_MODE_MASK) == APIC_DM_FIXED)) { | |
1561 | ||
1562 | kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32)); | |
1563 | return kvm_lapic_reg_write(vcpu->arch.apic, APIC_ICR, (u32)data); | |
1564 | } | |
1565 | ||
1566 | return 1; | |
1567 | } | |
1568 | ||
1569 | enum exit_fastpath_completion handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu) | |
1570 | { | |
1571 | u32 msr = kvm_rcx_read(vcpu); | |
1572 | u64 data = kvm_read_edx_eax(vcpu); | |
1573 | int ret = 0; | |
1574 | ||
1575 | switch (msr) { | |
1576 | case APIC_BASE_MSR + (APIC_ICR >> 4): | |
1577 | ret = handle_fastpath_set_x2apic_icr_irqoff(vcpu, data); | |
1578 | break; | |
1579 | default: | |
1580 | return EXIT_FASTPATH_NONE; | |
1581 | } | |
1582 | ||
1583 | if (!ret) { | |
1584 | trace_kvm_msr_write(msr, data); | |
1585 | return EXIT_FASTPATH_SKIP_EMUL_INS; | |
1586 | } | |
1587 | ||
1588 | return EXIT_FASTPATH_NONE; | |
1589 | } | |
1590 | EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff); | |
1591 | ||
f20935d8 SC |
1592 | /* |
1593 | * Adapt set_msr() to msr_io()'s calling convention | |
1594 | */ | |
1595 | static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data) | |
1596 | { | |
1597 | return __kvm_get_msr(vcpu, index, data, true); | |
1598 | } | |
1599 | ||
1600 | static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data) | |
1601 | { | |
1602 | return __kvm_set_msr(vcpu, index, *data, true); | |
313a3dc7 CO |
1603 | } |
1604 | ||
16e8d74d | 1605 | #ifdef CONFIG_X86_64 |
53fafdbb MT |
1606 | struct pvclock_clock { |
1607 | int vclock_mode; | |
1608 | u64 cycle_last; | |
1609 | u64 mask; | |
1610 | u32 mult; | |
1611 | u32 shift; | |
1612 | }; | |
1613 | ||
16e8d74d MT |
1614 | struct pvclock_gtod_data { |
1615 | seqcount_t seq; | |
1616 | ||
53fafdbb MT |
1617 | struct pvclock_clock clock; /* extract of a clocksource struct */ |
1618 | struct pvclock_clock raw_clock; /* extract of a clocksource struct */ | |
16e8d74d | 1619 | |
53fafdbb | 1620 | u64 boot_ns_raw; |
cbcf2dd3 TG |
1621 | u64 boot_ns; |
1622 | u64 nsec_base; | |
55dd00a7 | 1623 | u64 wall_time_sec; |
53fafdbb | 1624 | u64 monotonic_raw_nsec; |
16e8d74d MT |
1625 | }; |
1626 | ||
1627 | static struct pvclock_gtod_data pvclock_gtod_data; | |
1628 | ||
1629 | static void update_pvclock_gtod(struct timekeeper *tk) | |
1630 | { | |
1631 | struct pvclock_gtod_data *vdata = &pvclock_gtod_data; | |
53fafdbb | 1632 | u64 boot_ns, boot_ns_raw; |
cbcf2dd3 | 1633 | |
876e7881 | 1634 | boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot)); |
53fafdbb | 1635 | boot_ns_raw = ktime_to_ns(ktime_add(tk->tkr_raw.base, tk->offs_boot)); |
16e8d74d MT |
1636 | |
1637 | write_seqcount_begin(&vdata->seq); | |
1638 | ||
1639 | /* copy pvclock gtod data */ | |
876e7881 PZ |
1640 | vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode; |
1641 | vdata->clock.cycle_last = tk->tkr_mono.cycle_last; | |
1642 | vdata->clock.mask = tk->tkr_mono.mask; | |
1643 | vdata->clock.mult = tk->tkr_mono.mult; | |
1644 | vdata->clock.shift = tk->tkr_mono.shift; | |
16e8d74d | 1645 | |
53fafdbb MT |
1646 | vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->archdata.vclock_mode; |
1647 | vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last; | |
1648 | vdata->raw_clock.mask = tk->tkr_raw.mask; | |
1649 | vdata->raw_clock.mult = tk->tkr_raw.mult; | |
1650 | vdata->raw_clock.shift = tk->tkr_raw.shift; | |
1651 | ||
cbcf2dd3 | 1652 | vdata->boot_ns = boot_ns; |
876e7881 | 1653 | vdata->nsec_base = tk->tkr_mono.xtime_nsec; |
16e8d74d | 1654 | |
55dd00a7 MT |
1655 | vdata->wall_time_sec = tk->xtime_sec; |
1656 | ||
53fafdbb MT |
1657 | vdata->boot_ns_raw = boot_ns_raw; |
1658 | vdata->monotonic_raw_nsec = tk->tkr_raw.xtime_nsec; | |
1659 | ||
16e8d74d MT |
1660 | write_seqcount_end(&vdata->seq); |
1661 | } | |
1662 | #endif | |
1663 | ||
bab5bb39 NK |
1664 | void kvm_set_pending_timer(struct kvm_vcpu *vcpu) |
1665 | { | |
bab5bb39 | 1666 | kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu); |
4d151bf3 | 1667 | kvm_vcpu_kick(vcpu); |
bab5bb39 | 1668 | } |
16e8d74d | 1669 | |
18068523 GOC |
1670 | static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock) |
1671 | { | |
9ed3c444 AK |
1672 | int version; |
1673 | int r; | |
50d0a0f9 | 1674 | struct pvclock_wall_clock wc; |
87aeb54f | 1675 | struct timespec64 boot; |
18068523 GOC |
1676 | |
1677 | if (!wall_clock) | |
1678 | return; | |
1679 | ||
9ed3c444 AK |
1680 | r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version)); |
1681 | if (r) | |
1682 | return; | |
1683 | ||
1684 | if (version & 1) | |
1685 | ++version; /* first time write, random junk */ | |
1686 | ||
1687 | ++version; | |
18068523 | 1688 | |
1dab1345 NK |
1689 | if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version))) |
1690 | return; | |
18068523 | 1691 | |
50d0a0f9 GH |
1692 | /* |
1693 | * The guest calculates current wall clock time by adding | |
34c238a1 | 1694 | * system time (updated by kvm_guest_time_update below) to the |
50d0a0f9 GH |
1695 | * wall clock specified here. guest system time equals host |
1696 | * system time for us, thus we must fill in host boot time here. | |
1697 | */ | |
87aeb54f | 1698 | getboottime64(&boot); |
50d0a0f9 | 1699 | |
4b648665 | 1700 | if (kvm->arch.kvmclock_offset) { |
87aeb54f AB |
1701 | struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset); |
1702 | boot = timespec64_sub(boot, ts); | |
4b648665 | 1703 | } |
87aeb54f | 1704 | wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */ |
50d0a0f9 GH |
1705 | wc.nsec = boot.tv_nsec; |
1706 | wc.version = version; | |
18068523 GOC |
1707 | |
1708 | kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc)); | |
1709 | ||
1710 | version++; | |
1711 | kvm_write_guest(kvm, wall_clock, &version, sizeof(version)); | |
18068523 GOC |
1712 | } |
1713 | ||
50d0a0f9 GH |
1714 | static uint32_t div_frac(uint32_t dividend, uint32_t divisor) |
1715 | { | |
b51012de PB |
1716 | do_shl32_div32(dividend, divisor); |
1717 | return dividend; | |
50d0a0f9 GH |
1718 | } |
1719 | ||
3ae13faa | 1720 | static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz, |
5f4e3f88 | 1721 | s8 *pshift, u32 *pmultiplier) |
50d0a0f9 | 1722 | { |
5f4e3f88 | 1723 | uint64_t scaled64; |
50d0a0f9 GH |
1724 | int32_t shift = 0; |
1725 | uint64_t tps64; | |
1726 | uint32_t tps32; | |
1727 | ||
3ae13faa PB |
1728 | tps64 = base_hz; |
1729 | scaled64 = scaled_hz; | |
50933623 | 1730 | while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) { |
50d0a0f9 GH |
1731 | tps64 >>= 1; |
1732 | shift--; | |
1733 | } | |
1734 | ||
1735 | tps32 = (uint32_t)tps64; | |
50933623 JK |
1736 | while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) { |
1737 | if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000) | |
5f4e3f88 ZA |
1738 | scaled64 >>= 1; |
1739 | else | |
1740 | tps32 <<= 1; | |
50d0a0f9 GH |
1741 | shift++; |
1742 | } | |
1743 | ||
5f4e3f88 ZA |
1744 | *pshift = shift; |
1745 | *pmultiplier = div_frac(scaled64, tps32); | |
50d0a0f9 GH |
1746 | } |
1747 | ||
d828199e | 1748 | #ifdef CONFIG_X86_64 |
16e8d74d | 1749 | static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0); |
d828199e | 1750 | #endif |
16e8d74d | 1751 | |
c8076604 | 1752 | static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz); |
69b0049a | 1753 | static unsigned long max_tsc_khz; |
c8076604 | 1754 | |
cc578287 | 1755 | static u32 adjust_tsc_khz(u32 khz, s32 ppm) |
1e993611 | 1756 | { |
cc578287 ZA |
1757 | u64 v = (u64)khz * (1000000 + ppm); |
1758 | do_div(v, 1000000); | |
1759 | return v; | |
1e993611 JR |
1760 | } |
1761 | ||
381d585c HZ |
1762 | static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale) |
1763 | { | |
1764 | u64 ratio; | |
1765 | ||
1766 | /* Guest TSC same frequency as host TSC? */ | |
1767 | if (!scale) { | |
1768 | vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio; | |
1769 | return 0; | |
1770 | } | |
1771 | ||
1772 | /* TSC scaling supported? */ | |
1773 | if (!kvm_has_tsc_control) { | |
1774 | if (user_tsc_khz > tsc_khz) { | |
1775 | vcpu->arch.tsc_catchup = 1; | |
1776 | vcpu->arch.tsc_always_catchup = 1; | |
1777 | return 0; | |
1778 | } else { | |
3f16a5c3 | 1779 | pr_warn_ratelimited("user requested TSC rate below hardware speed\n"); |
381d585c HZ |
1780 | return -1; |
1781 | } | |
1782 | } | |
1783 | ||
1784 | /* TSC scaling required - calculate ratio */ | |
1785 | ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits, | |
1786 | user_tsc_khz, tsc_khz); | |
1787 | ||
1788 | if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) { | |
3f16a5c3 PB |
1789 | pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n", |
1790 | user_tsc_khz); | |
381d585c HZ |
1791 | return -1; |
1792 | } | |
1793 | ||
1794 | vcpu->arch.tsc_scaling_ratio = ratio; | |
1795 | return 0; | |
1796 | } | |
1797 | ||
4941b8cb | 1798 | static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz) |
759379dd | 1799 | { |
cc578287 ZA |
1800 | u32 thresh_lo, thresh_hi; |
1801 | int use_scaling = 0; | |
217fc9cf | 1802 | |
03ba32ca | 1803 | /* tsc_khz can be zero if TSC calibration fails */ |
4941b8cb | 1804 | if (user_tsc_khz == 0) { |
ad721883 HZ |
1805 | /* set tsc_scaling_ratio to a safe value */ |
1806 | vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio; | |
381d585c | 1807 | return -1; |
ad721883 | 1808 | } |
03ba32ca | 1809 | |
c285545f | 1810 | /* Compute a scale to convert nanoseconds in TSC cycles */ |
3ae13faa | 1811 | kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC, |
cc578287 ZA |
1812 | &vcpu->arch.virtual_tsc_shift, |
1813 | &vcpu->arch.virtual_tsc_mult); | |
4941b8cb | 1814 | vcpu->arch.virtual_tsc_khz = user_tsc_khz; |
cc578287 ZA |
1815 | |
1816 | /* | |
1817 | * Compute the variation in TSC rate which is acceptable | |
1818 | * within the range of tolerance and decide if the | |
1819 | * rate being applied is within that bounds of the hardware | |
1820 | * rate. If so, no scaling or compensation need be done. | |
1821 | */ | |
1822 | thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm); | |
1823 | thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm); | |
4941b8cb PB |
1824 | if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) { |
1825 | pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi); | |
cc578287 ZA |
1826 | use_scaling = 1; |
1827 | } | |
4941b8cb | 1828 | return set_tsc_khz(vcpu, user_tsc_khz, use_scaling); |
c285545f ZA |
1829 | } |
1830 | ||
1831 | static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns) | |
1832 | { | |
e26101b1 | 1833 | u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec, |
cc578287 ZA |
1834 | vcpu->arch.virtual_tsc_mult, |
1835 | vcpu->arch.virtual_tsc_shift); | |
e26101b1 | 1836 | tsc += vcpu->arch.this_tsc_write; |
c285545f ZA |
1837 | return tsc; |
1838 | } | |
1839 | ||
b0c39dc6 VK |
1840 | static inline int gtod_is_based_on_tsc(int mode) |
1841 | { | |
1842 | return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK; | |
1843 | } | |
1844 | ||
69b0049a | 1845 | static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu) |
b48aa97e MT |
1846 | { |
1847 | #ifdef CONFIG_X86_64 | |
1848 | bool vcpus_matched; | |
b48aa97e MT |
1849 | struct kvm_arch *ka = &vcpu->kvm->arch; |
1850 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; | |
1851 | ||
1852 | vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 == | |
1853 | atomic_read(&vcpu->kvm->online_vcpus)); | |
1854 | ||
7f187922 MT |
1855 | /* |
1856 | * Once the masterclock is enabled, always perform request in | |
1857 | * order to update it. | |
1858 | * | |
1859 | * In order to enable masterclock, the host clocksource must be TSC | |
1860 | * and the vcpus need to have matched TSCs. When that happens, | |
1861 | * perform request to enable masterclock. | |
1862 | */ | |
1863 | if (ka->use_master_clock || | |
b0c39dc6 | 1864 | (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched)) |
b48aa97e MT |
1865 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); |
1866 | ||
1867 | trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc, | |
1868 | atomic_read(&vcpu->kvm->online_vcpus), | |
1869 | ka->use_master_clock, gtod->clock.vclock_mode); | |
1870 | #endif | |
1871 | } | |
1872 | ||
ba904635 WA |
1873 | static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset) |
1874 | { | |
e79f245d | 1875 | u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu); |
ba904635 WA |
1876 | vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset; |
1877 | } | |
1878 | ||
35181e86 HZ |
1879 | /* |
1880 | * Multiply tsc by a fixed point number represented by ratio. | |
1881 | * | |
1882 | * The most significant 64-N bits (mult) of ratio represent the | |
1883 | * integral part of the fixed point number; the remaining N bits | |
1884 | * (frac) represent the fractional part, ie. ratio represents a fixed | |
1885 | * point number (mult + frac * 2^(-N)). | |
1886 | * | |
1887 | * N equals to kvm_tsc_scaling_ratio_frac_bits. | |
1888 | */ | |
1889 | static inline u64 __scale_tsc(u64 ratio, u64 tsc) | |
1890 | { | |
1891 | return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits); | |
1892 | } | |
1893 | ||
1894 | u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc) | |
1895 | { | |
1896 | u64 _tsc = tsc; | |
1897 | u64 ratio = vcpu->arch.tsc_scaling_ratio; | |
1898 | ||
1899 | if (ratio != kvm_default_tsc_scaling_ratio) | |
1900 | _tsc = __scale_tsc(ratio, tsc); | |
1901 | ||
1902 | return _tsc; | |
1903 | } | |
1904 | EXPORT_SYMBOL_GPL(kvm_scale_tsc); | |
1905 | ||
07c1419a HZ |
1906 | static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc) |
1907 | { | |
1908 | u64 tsc; | |
1909 | ||
1910 | tsc = kvm_scale_tsc(vcpu, rdtsc()); | |
1911 | ||
1912 | return target_tsc - tsc; | |
1913 | } | |
1914 | ||
4ba76538 HZ |
1915 | u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc) |
1916 | { | |
e79f245d KA |
1917 | u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu); |
1918 | ||
1919 | return tsc_offset + kvm_scale_tsc(vcpu, host_tsc); | |
4ba76538 HZ |
1920 | } |
1921 | EXPORT_SYMBOL_GPL(kvm_read_l1_tsc); | |
1922 | ||
a545ab6a LC |
1923 | static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset) |
1924 | { | |
326e7425 | 1925 | vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset); |
a545ab6a LC |
1926 | } |
1927 | ||
b0c39dc6 VK |
1928 | static inline bool kvm_check_tsc_unstable(void) |
1929 | { | |
1930 | #ifdef CONFIG_X86_64 | |
1931 | /* | |
1932 | * TSC is marked unstable when we're running on Hyper-V, | |
1933 | * 'TSC page' clocksource is good. | |
1934 | */ | |
1935 | if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK) | |
1936 | return false; | |
1937 | #endif | |
1938 | return check_tsc_unstable(); | |
1939 | } | |
1940 | ||
8fe8ab46 | 1941 | void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr) |
99e3e30a ZA |
1942 | { |
1943 | struct kvm *kvm = vcpu->kvm; | |
f38e098f | 1944 | u64 offset, ns, elapsed; |
99e3e30a | 1945 | unsigned long flags; |
b48aa97e | 1946 | bool matched; |
0d3da0d2 | 1947 | bool already_matched; |
8fe8ab46 | 1948 | u64 data = msr->data; |
c5e8ec8e | 1949 | bool synchronizing = false; |
99e3e30a | 1950 | |
038f8c11 | 1951 | raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags); |
07c1419a | 1952 | offset = kvm_compute_tsc_offset(vcpu, data); |
9285ec4c | 1953 | ns = ktime_get_boottime_ns(); |
f38e098f | 1954 | elapsed = ns - kvm->arch.last_tsc_nsec; |
5d3cb0f6 | 1955 | |
03ba32ca | 1956 | if (vcpu->arch.virtual_tsc_khz) { |
bd8fab39 DP |
1957 | if (data == 0 && msr->host_initiated) { |
1958 | /* | |
1959 | * detection of vcpu initialization -- need to sync | |
1960 | * with other vCPUs. This particularly helps to keep | |
1961 | * kvm_clock stable after CPU hotplug | |
1962 | */ | |
1963 | synchronizing = true; | |
1964 | } else { | |
1965 | u64 tsc_exp = kvm->arch.last_tsc_write + | |
1966 | nsec_to_cycles(vcpu, elapsed); | |
1967 | u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL; | |
1968 | /* | |
1969 | * Special case: TSC write with a small delta (1 second) | |
1970 | * of virtual cycle time against real time is | |
1971 | * interpreted as an attempt to synchronize the CPU. | |
1972 | */ | |
1973 | synchronizing = data < tsc_exp + tsc_hz && | |
1974 | data + tsc_hz > tsc_exp; | |
1975 | } | |
c5e8ec8e | 1976 | } |
f38e098f ZA |
1977 | |
1978 | /* | |
5d3cb0f6 ZA |
1979 | * For a reliable TSC, we can match TSC offsets, and for an unstable |
1980 | * TSC, we add elapsed time in this computation. We could let the | |
1981 | * compensation code attempt to catch up if we fall behind, but | |
1982 | * it's better to try to match offsets from the beginning. | |
1983 | */ | |
c5e8ec8e | 1984 | if (synchronizing && |
5d3cb0f6 | 1985 | vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) { |
b0c39dc6 | 1986 | if (!kvm_check_tsc_unstable()) { |
e26101b1 | 1987 | offset = kvm->arch.cur_tsc_offset; |
f38e098f | 1988 | } else { |
857e4099 | 1989 | u64 delta = nsec_to_cycles(vcpu, elapsed); |
5d3cb0f6 | 1990 | data += delta; |
07c1419a | 1991 | offset = kvm_compute_tsc_offset(vcpu, data); |
f38e098f | 1992 | } |
b48aa97e | 1993 | matched = true; |
0d3da0d2 | 1994 | already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation); |
e26101b1 ZA |
1995 | } else { |
1996 | /* | |
1997 | * We split periods of matched TSC writes into generations. | |
1998 | * For each generation, we track the original measured | |
1999 | * nanosecond time, offset, and write, so if TSCs are in | |
2000 | * sync, we can match exact offset, and if not, we can match | |
4a969980 | 2001 | * exact software computation in compute_guest_tsc() |
e26101b1 ZA |
2002 | * |
2003 | * These values are tracked in kvm->arch.cur_xxx variables. | |
2004 | */ | |
2005 | kvm->arch.cur_tsc_generation++; | |
2006 | kvm->arch.cur_tsc_nsec = ns; | |
2007 | kvm->arch.cur_tsc_write = data; | |
2008 | kvm->arch.cur_tsc_offset = offset; | |
b48aa97e | 2009 | matched = false; |
f38e098f | 2010 | } |
e26101b1 ZA |
2011 | |
2012 | /* | |
2013 | * We also track th most recent recorded KHZ, write and time to | |
2014 | * allow the matching interval to be extended at each write. | |
2015 | */ | |
f38e098f ZA |
2016 | kvm->arch.last_tsc_nsec = ns; |
2017 | kvm->arch.last_tsc_write = data; | |
5d3cb0f6 | 2018 | kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz; |
99e3e30a | 2019 | |
b183aa58 | 2020 | vcpu->arch.last_guest_tsc = data; |
e26101b1 ZA |
2021 | |
2022 | /* Keep track of which generation this VCPU has synchronized to */ | |
2023 | vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation; | |
2024 | vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec; | |
2025 | vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write; | |
2026 | ||
d6321d49 | 2027 | if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) |
ba904635 | 2028 | update_ia32_tsc_adjust_msr(vcpu, offset); |
d6321d49 | 2029 | |
a545ab6a | 2030 | kvm_vcpu_write_tsc_offset(vcpu, offset); |
e26101b1 | 2031 | raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags); |
b48aa97e MT |
2032 | |
2033 | spin_lock(&kvm->arch.pvclock_gtod_sync_lock); | |
0d3da0d2 | 2034 | if (!matched) { |
b48aa97e | 2035 | kvm->arch.nr_vcpus_matched_tsc = 0; |
0d3da0d2 TG |
2036 | } else if (!already_matched) { |
2037 | kvm->arch.nr_vcpus_matched_tsc++; | |
2038 | } | |
b48aa97e MT |
2039 | |
2040 | kvm_track_tsc_matching(vcpu); | |
2041 | spin_unlock(&kvm->arch.pvclock_gtod_sync_lock); | |
99e3e30a | 2042 | } |
e26101b1 | 2043 | |
99e3e30a ZA |
2044 | EXPORT_SYMBOL_GPL(kvm_write_tsc); |
2045 | ||
58ea6767 HZ |
2046 | static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu, |
2047 | s64 adjustment) | |
2048 | { | |
326e7425 LS |
2049 | u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu); |
2050 | kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment); | |
58ea6767 HZ |
2051 | } |
2052 | ||
2053 | static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment) | |
2054 | { | |
2055 | if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio) | |
2056 | WARN_ON(adjustment < 0); | |
2057 | adjustment = kvm_scale_tsc(vcpu, (u64) adjustment); | |
ea26e4ec | 2058 | adjust_tsc_offset_guest(vcpu, adjustment); |
58ea6767 HZ |
2059 | } |
2060 | ||
d828199e MT |
2061 | #ifdef CONFIG_X86_64 |
2062 | ||
a5a1d1c2 | 2063 | static u64 read_tsc(void) |
d828199e | 2064 | { |
a5a1d1c2 | 2065 | u64 ret = (u64)rdtsc_ordered(); |
03b9730b | 2066 | u64 last = pvclock_gtod_data.clock.cycle_last; |
d828199e MT |
2067 | |
2068 | if (likely(ret >= last)) | |
2069 | return ret; | |
2070 | ||
2071 | /* | |
2072 | * GCC likes to generate cmov here, but this branch is extremely | |
6a6256f9 | 2073 | * predictable (it's just a function of time and the likely is |
d828199e MT |
2074 | * very likely) and there's a data dependence, so force GCC |
2075 | * to generate a branch instead. I don't barrier() because | |
2076 | * we don't actually need a barrier, and if this function | |
2077 | * ever gets inlined it will generate worse code. | |
2078 | */ | |
2079 | asm volatile (""); | |
2080 | return last; | |
2081 | } | |
2082 | ||
53fafdbb MT |
2083 | static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp, |
2084 | int *mode) | |
d828199e MT |
2085 | { |
2086 | long v; | |
b0c39dc6 VK |
2087 | u64 tsc_pg_val; |
2088 | ||
53fafdbb | 2089 | switch (clock->vclock_mode) { |
b0c39dc6 VK |
2090 | case VCLOCK_HVCLOCK: |
2091 | tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(), | |
2092 | tsc_timestamp); | |
2093 | if (tsc_pg_val != U64_MAX) { | |
2094 | /* TSC page valid */ | |
2095 | *mode = VCLOCK_HVCLOCK; | |
53fafdbb MT |
2096 | v = (tsc_pg_val - clock->cycle_last) & |
2097 | clock->mask; | |
b0c39dc6 VK |
2098 | } else { |
2099 | /* TSC page invalid */ | |
2100 | *mode = VCLOCK_NONE; | |
2101 | } | |
2102 | break; | |
2103 | case VCLOCK_TSC: | |
2104 | *mode = VCLOCK_TSC; | |
2105 | *tsc_timestamp = read_tsc(); | |
53fafdbb MT |
2106 | v = (*tsc_timestamp - clock->cycle_last) & |
2107 | clock->mask; | |
b0c39dc6 VK |
2108 | break; |
2109 | default: | |
2110 | *mode = VCLOCK_NONE; | |
2111 | } | |
d828199e | 2112 | |
b0c39dc6 VK |
2113 | if (*mode == VCLOCK_NONE) |
2114 | *tsc_timestamp = v = 0; | |
d828199e | 2115 | |
53fafdbb | 2116 | return v * clock->mult; |
d828199e MT |
2117 | } |
2118 | ||
53fafdbb | 2119 | static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp) |
d828199e | 2120 | { |
cbcf2dd3 | 2121 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; |
d828199e | 2122 | unsigned long seq; |
d828199e | 2123 | int mode; |
cbcf2dd3 | 2124 | u64 ns; |
d828199e | 2125 | |
d828199e MT |
2126 | do { |
2127 | seq = read_seqcount_begin(>od->seq); | |
53fafdbb MT |
2128 | ns = gtod->monotonic_raw_nsec; |
2129 | ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode); | |
d828199e | 2130 | ns >>= gtod->clock.shift; |
53fafdbb | 2131 | ns += gtod->boot_ns_raw; |
d828199e | 2132 | } while (unlikely(read_seqcount_retry(>od->seq, seq))); |
cbcf2dd3 | 2133 | *t = ns; |
d828199e MT |
2134 | |
2135 | return mode; | |
2136 | } | |
2137 | ||
899a31f5 | 2138 | static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp) |
55dd00a7 MT |
2139 | { |
2140 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; | |
2141 | unsigned long seq; | |
2142 | int mode; | |
2143 | u64 ns; | |
2144 | ||
2145 | do { | |
2146 | seq = read_seqcount_begin(>od->seq); | |
55dd00a7 MT |
2147 | ts->tv_sec = gtod->wall_time_sec; |
2148 | ns = gtod->nsec_base; | |
53fafdbb | 2149 | ns += vgettsc(>od->clock, tsc_timestamp, &mode); |
55dd00a7 MT |
2150 | ns >>= gtod->clock.shift; |
2151 | } while (unlikely(read_seqcount_retry(>od->seq, seq))); | |
2152 | ||
2153 | ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns); | |
2154 | ts->tv_nsec = ns; | |
2155 | ||
2156 | return mode; | |
2157 | } | |
2158 | ||
b0c39dc6 VK |
2159 | /* returns true if host is using TSC based clocksource */ |
2160 | static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp) | |
d828199e | 2161 | { |
d828199e | 2162 | /* checked again under seqlock below */ |
b0c39dc6 | 2163 | if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode)) |
d828199e MT |
2164 | return false; |
2165 | ||
53fafdbb | 2166 | return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns, |
b0c39dc6 | 2167 | tsc_timestamp)); |
d828199e | 2168 | } |
55dd00a7 | 2169 | |
b0c39dc6 | 2170 | /* returns true if host is using TSC based clocksource */ |
899a31f5 | 2171 | static bool kvm_get_walltime_and_clockread(struct timespec64 *ts, |
b0c39dc6 | 2172 | u64 *tsc_timestamp) |
55dd00a7 MT |
2173 | { |
2174 | /* checked again under seqlock below */ | |
b0c39dc6 | 2175 | if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode)) |
55dd00a7 MT |
2176 | return false; |
2177 | ||
b0c39dc6 | 2178 | return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp)); |
55dd00a7 | 2179 | } |
d828199e MT |
2180 | #endif |
2181 | ||
2182 | /* | |
2183 | * | |
b48aa97e MT |
2184 | * Assuming a stable TSC across physical CPUS, and a stable TSC |
2185 | * across virtual CPUs, the following condition is possible. | |
2186 | * Each numbered line represents an event visible to both | |
d828199e MT |
2187 | * CPUs at the next numbered event. |
2188 | * | |
2189 | * "timespecX" represents host monotonic time. "tscX" represents | |
2190 | * RDTSC value. | |
2191 | * | |
2192 | * VCPU0 on CPU0 | VCPU1 on CPU1 | |
2193 | * | |
2194 | * 1. read timespec0,tsc0 | |
2195 | * 2. | timespec1 = timespec0 + N | |
2196 | * | tsc1 = tsc0 + M | |
2197 | * 3. transition to guest | transition to guest | |
2198 | * 4. ret0 = timespec0 + (rdtsc - tsc0) | | |
2199 | * 5. | ret1 = timespec1 + (rdtsc - tsc1) | |
2200 | * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M)) | |
2201 | * | |
2202 | * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity: | |
2203 | * | |
2204 | * - ret0 < ret1 | |
2205 | * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M)) | |
2206 | * ... | |
2207 | * - 0 < N - M => M < N | |
2208 | * | |
2209 | * That is, when timespec0 != timespec1, M < N. Unfortunately that is not | |
2210 | * always the case (the difference between two distinct xtime instances | |
2211 | * might be smaller then the difference between corresponding TSC reads, | |
2212 | * when updating guest vcpus pvclock areas). | |
2213 | * | |
2214 | * To avoid that problem, do not allow visibility of distinct | |
2215 | * system_timestamp/tsc_timestamp values simultaneously: use a master | |
2216 | * copy of host monotonic time values. Update that master copy | |
2217 | * in lockstep. | |
2218 | * | |
b48aa97e | 2219 | * Rely on synchronization of host TSCs and guest TSCs for monotonicity. |
d828199e MT |
2220 | * |
2221 | */ | |
2222 | ||
2223 | static void pvclock_update_vm_gtod_copy(struct kvm *kvm) | |
2224 | { | |
2225 | #ifdef CONFIG_X86_64 | |
2226 | struct kvm_arch *ka = &kvm->arch; | |
2227 | int vclock_mode; | |
b48aa97e MT |
2228 | bool host_tsc_clocksource, vcpus_matched; |
2229 | ||
2230 | vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 == | |
2231 | atomic_read(&kvm->online_vcpus)); | |
d828199e MT |
2232 | |
2233 | /* | |
2234 | * If the host uses TSC clock, then passthrough TSC as stable | |
2235 | * to the guest. | |
2236 | */ | |
b48aa97e | 2237 | host_tsc_clocksource = kvm_get_time_and_clockread( |
d828199e MT |
2238 | &ka->master_kernel_ns, |
2239 | &ka->master_cycle_now); | |
2240 | ||
16a96021 | 2241 | ka->use_master_clock = host_tsc_clocksource && vcpus_matched |
a826faf1 | 2242 | && !ka->backwards_tsc_observed |
54750f2c | 2243 | && !ka->boot_vcpu_runs_old_kvmclock; |
b48aa97e | 2244 | |
d828199e MT |
2245 | if (ka->use_master_clock) |
2246 | atomic_set(&kvm_guest_has_master_clock, 1); | |
2247 | ||
2248 | vclock_mode = pvclock_gtod_data.clock.vclock_mode; | |
b48aa97e MT |
2249 | trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode, |
2250 | vcpus_matched); | |
d828199e MT |
2251 | #endif |
2252 | } | |
2253 | ||
2860c4b1 PB |
2254 | void kvm_make_mclock_inprogress_request(struct kvm *kvm) |
2255 | { | |
2256 | kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS); | |
2257 | } | |
2258 | ||
2e762ff7 MT |
2259 | static void kvm_gen_update_masterclock(struct kvm *kvm) |
2260 | { | |
2261 | #ifdef CONFIG_X86_64 | |
2262 | int i; | |
2263 | struct kvm_vcpu *vcpu; | |
2264 | struct kvm_arch *ka = &kvm->arch; | |
2265 | ||
2266 | spin_lock(&ka->pvclock_gtod_sync_lock); | |
2267 | kvm_make_mclock_inprogress_request(kvm); | |
2268 | /* no guest entries from this point */ | |
2269 | pvclock_update_vm_gtod_copy(kvm); | |
2270 | ||
2271 | kvm_for_each_vcpu(i, vcpu, kvm) | |
105b21bb | 2272 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
2e762ff7 MT |
2273 | |
2274 | /* guest entries allowed */ | |
2275 | kvm_for_each_vcpu(i, vcpu, kvm) | |
72875d8a | 2276 | kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu); |
2e762ff7 MT |
2277 | |
2278 | spin_unlock(&ka->pvclock_gtod_sync_lock); | |
2279 | #endif | |
2280 | } | |
2281 | ||
e891a32e | 2282 | u64 get_kvmclock_ns(struct kvm *kvm) |
108b249c | 2283 | { |
108b249c | 2284 | struct kvm_arch *ka = &kvm->arch; |
8b953440 | 2285 | struct pvclock_vcpu_time_info hv_clock; |
e2c2206a | 2286 | u64 ret; |
108b249c | 2287 | |
8b953440 PB |
2288 | spin_lock(&ka->pvclock_gtod_sync_lock); |
2289 | if (!ka->use_master_clock) { | |
2290 | spin_unlock(&ka->pvclock_gtod_sync_lock); | |
9285ec4c | 2291 | return ktime_get_boottime_ns() + ka->kvmclock_offset; |
108b249c PB |
2292 | } |
2293 | ||
8b953440 PB |
2294 | hv_clock.tsc_timestamp = ka->master_cycle_now; |
2295 | hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset; | |
2296 | spin_unlock(&ka->pvclock_gtod_sync_lock); | |
2297 | ||
e2c2206a WL |
2298 | /* both __this_cpu_read() and rdtsc() should be on the same cpu */ |
2299 | get_cpu(); | |
2300 | ||
e70b57a6 WL |
2301 | if (__this_cpu_read(cpu_tsc_khz)) { |
2302 | kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL, | |
2303 | &hv_clock.tsc_shift, | |
2304 | &hv_clock.tsc_to_system_mul); | |
2305 | ret = __pvclock_read_cycles(&hv_clock, rdtsc()); | |
2306 | } else | |
9285ec4c | 2307 | ret = ktime_get_boottime_ns() + ka->kvmclock_offset; |
e2c2206a WL |
2308 | |
2309 | put_cpu(); | |
2310 | ||
2311 | return ret; | |
108b249c PB |
2312 | } |
2313 | ||
0d6dd2ff PB |
2314 | static void kvm_setup_pvclock_page(struct kvm_vcpu *v) |
2315 | { | |
2316 | struct kvm_vcpu_arch *vcpu = &v->arch; | |
2317 | struct pvclock_vcpu_time_info guest_hv_clock; | |
2318 | ||
4e335d9e | 2319 | if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time, |
0d6dd2ff PB |
2320 | &guest_hv_clock, sizeof(guest_hv_clock)))) |
2321 | return; | |
2322 | ||
2323 | /* This VCPU is paused, but it's legal for a guest to read another | |
2324 | * VCPU's kvmclock, so we really have to follow the specification where | |
2325 | * it says that version is odd if data is being modified, and even after | |
2326 | * it is consistent. | |
2327 | * | |
2328 | * Version field updates must be kept separate. This is because | |
2329 | * kvm_write_guest_cached might use a "rep movs" instruction, and | |
2330 | * writes within a string instruction are weakly ordered. So there | |
2331 | * are three writes overall. | |
2332 | * | |
2333 | * As a small optimization, only write the version field in the first | |
2334 | * and third write. The vcpu->pv_time cache is still valid, because the | |
2335 | * version field is the first in the struct. | |
2336 | */ | |
2337 | BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0); | |
2338 | ||
51c4b8bb LA |
2339 | if (guest_hv_clock.version & 1) |
2340 | ++guest_hv_clock.version; /* first time write, random junk */ | |
2341 | ||
0d6dd2ff | 2342 | vcpu->hv_clock.version = guest_hv_clock.version + 1; |
4e335d9e PB |
2343 | kvm_write_guest_cached(v->kvm, &vcpu->pv_time, |
2344 | &vcpu->hv_clock, | |
2345 | sizeof(vcpu->hv_clock.version)); | |
0d6dd2ff PB |
2346 | |
2347 | smp_wmb(); | |
2348 | ||
2349 | /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */ | |
2350 | vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED); | |
2351 | ||
2352 | if (vcpu->pvclock_set_guest_stopped_request) { | |
2353 | vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED; | |
2354 | vcpu->pvclock_set_guest_stopped_request = false; | |
2355 | } | |
2356 | ||
2357 | trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock); | |
2358 | ||
4e335d9e PB |
2359 | kvm_write_guest_cached(v->kvm, &vcpu->pv_time, |
2360 | &vcpu->hv_clock, | |
2361 | sizeof(vcpu->hv_clock)); | |
0d6dd2ff PB |
2362 | |
2363 | smp_wmb(); | |
2364 | ||
2365 | vcpu->hv_clock.version++; | |
4e335d9e PB |
2366 | kvm_write_guest_cached(v->kvm, &vcpu->pv_time, |
2367 | &vcpu->hv_clock, | |
2368 | sizeof(vcpu->hv_clock.version)); | |
0d6dd2ff PB |
2369 | } |
2370 | ||
34c238a1 | 2371 | static int kvm_guest_time_update(struct kvm_vcpu *v) |
18068523 | 2372 | { |
78db6a50 | 2373 | unsigned long flags, tgt_tsc_khz; |
18068523 | 2374 | struct kvm_vcpu_arch *vcpu = &v->arch; |
d828199e | 2375 | struct kvm_arch *ka = &v->kvm->arch; |
f25e656d | 2376 | s64 kernel_ns; |
d828199e | 2377 | u64 tsc_timestamp, host_tsc; |
51d59c6b | 2378 | u8 pvclock_flags; |
d828199e MT |
2379 | bool use_master_clock; |
2380 | ||
2381 | kernel_ns = 0; | |
2382 | host_tsc = 0; | |
18068523 | 2383 | |
d828199e MT |
2384 | /* |
2385 | * If the host uses TSC clock, then passthrough TSC as stable | |
2386 | * to the guest. | |
2387 | */ | |
2388 | spin_lock(&ka->pvclock_gtod_sync_lock); | |
2389 | use_master_clock = ka->use_master_clock; | |
2390 | if (use_master_clock) { | |
2391 | host_tsc = ka->master_cycle_now; | |
2392 | kernel_ns = ka->master_kernel_ns; | |
2393 | } | |
2394 | spin_unlock(&ka->pvclock_gtod_sync_lock); | |
c09664bb MT |
2395 | |
2396 | /* Keep irq disabled to prevent changes to the clock */ | |
2397 | local_irq_save(flags); | |
78db6a50 PB |
2398 | tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz); |
2399 | if (unlikely(tgt_tsc_khz == 0)) { | |
c09664bb MT |
2400 | local_irq_restore(flags); |
2401 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, v); | |
2402 | return 1; | |
2403 | } | |
d828199e | 2404 | if (!use_master_clock) { |
4ea1636b | 2405 | host_tsc = rdtsc(); |
9285ec4c | 2406 | kernel_ns = ktime_get_boottime_ns(); |
d828199e MT |
2407 | } |
2408 | ||
4ba76538 | 2409 | tsc_timestamp = kvm_read_l1_tsc(v, host_tsc); |
d828199e | 2410 | |
c285545f ZA |
2411 | /* |
2412 | * We may have to catch up the TSC to match elapsed wall clock | |
2413 | * time for two reasons, even if kvmclock is used. | |
2414 | * 1) CPU could have been running below the maximum TSC rate | |
2415 | * 2) Broken TSC compensation resets the base at each VCPU | |
2416 | * entry to avoid unknown leaps of TSC even when running | |
2417 | * again on the same CPU. This may cause apparent elapsed | |
2418 | * time to disappear, and the guest to stand still or run | |
2419 | * very slowly. | |
2420 | */ | |
2421 | if (vcpu->tsc_catchup) { | |
2422 | u64 tsc = compute_guest_tsc(v, kernel_ns); | |
2423 | if (tsc > tsc_timestamp) { | |
f1e2b260 | 2424 | adjust_tsc_offset_guest(v, tsc - tsc_timestamp); |
c285545f ZA |
2425 | tsc_timestamp = tsc; |
2426 | } | |
50d0a0f9 GH |
2427 | } |
2428 | ||
18068523 GOC |
2429 | local_irq_restore(flags); |
2430 | ||
0d6dd2ff | 2431 | /* With all the info we got, fill in the values */ |
18068523 | 2432 | |
78db6a50 PB |
2433 | if (kvm_has_tsc_control) |
2434 | tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz); | |
2435 | ||
2436 | if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) { | |
3ae13faa | 2437 | kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL, |
5f4e3f88 ZA |
2438 | &vcpu->hv_clock.tsc_shift, |
2439 | &vcpu->hv_clock.tsc_to_system_mul); | |
78db6a50 | 2440 | vcpu->hw_tsc_khz = tgt_tsc_khz; |
8cfdc000 ZA |
2441 | } |
2442 | ||
1d5f066e | 2443 | vcpu->hv_clock.tsc_timestamp = tsc_timestamp; |
759379dd | 2444 | vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset; |
28e4639a | 2445 | vcpu->last_guest_tsc = tsc_timestamp; |
51d59c6b | 2446 | |
d828199e | 2447 | /* If the host uses TSC clocksource, then it is stable */ |
0d6dd2ff | 2448 | pvclock_flags = 0; |
d828199e MT |
2449 | if (use_master_clock) |
2450 | pvclock_flags |= PVCLOCK_TSC_STABLE_BIT; | |
2451 | ||
78c0337a MT |
2452 | vcpu->hv_clock.flags = pvclock_flags; |
2453 | ||
095cf55d PB |
2454 | if (vcpu->pv_time_enabled) |
2455 | kvm_setup_pvclock_page(v); | |
2456 | if (v == kvm_get_vcpu(v->kvm, 0)) | |
2457 | kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock); | |
8cfdc000 | 2458 | return 0; |
c8076604 GH |
2459 | } |
2460 | ||
0061d53d MT |
2461 | /* |
2462 | * kvmclock updates which are isolated to a given vcpu, such as | |
2463 | * vcpu->cpu migration, should not allow system_timestamp from | |
2464 | * the rest of the vcpus to remain static. Otherwise ntp frequency | |
2465 | * correction applies to one vcpu's system_timestamp but not | |
2466 | * the others. | |
2467 | * | |
2468 | * So in those cases, request a kvmclock update for all vcpus. | |
7e44e449 AJ |
2469 | * We need to rate-limit these requests though, as they can |
2470 | * considerably slow guests that have a large number of vcpus. | |
2471 | * The time for a remote vcpu to update its kvmclock is bound | |
2472 | * by the delay we use to rate-limit the updates. | |
0061d53d MT |
2473 | */ |
2474 | ||
7e44e449 AJ |
2475 | #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100) |
2476 | ||
2477 | static void kvmclock_update_fn(struct work_struct *work) | |
0061d53d MT |
2478 | { |
2479 | int i; | |
7e44e449 AJ |
2480 | struct delayed_work *dwork = to_delayed_work(work); |
2481 | struct kvm_arch *ka = container_of(dwork, struct kvm_arch, | |
2482 | kvmclock_update_work); | |
2483 | struct kvm *kvm = container_of(ka, struct kvm, arch); | |
0061d53d MT |
2484 | struct kvm_vcpu *vcpu; |
2485 | ||
2486 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
105b21bb | 2487 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
0061d53d MT |
2488 | kvm_vcpu_kick(vcpu); |
2489 | } | |
2490 | } | |
2491 | ||
7e44e449 AJ |
2492 | static void kvm_gen_kvmclock_update(struct kvm_vcpu *v) |
2493 | { | |
2494 | struct kvm *kvm = v->kvm; | |
2495 | ||
105b21bb | 2496 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, v); |
7e44e449 AJ |
2497 | schedule_delayed_work(&kvm->arch.kvmclock_update_work, |
2498 | KVMCLOCK_UPDATE_DELAY); | |
2499 | } | |
2500 | ||
332967a3 AJ |
2501 | #define KVMCLOCK_SYNC_PERIOD (300 * HZ) |
2502 | ||
2503 | static void kvmclock_sync_fn(struct work_struct *work) | |
2504 | { | |
2505 | struct delayed_work *dwork = to_delayed_work(work); | |
2506 | struct kvm_arch *ka = container_of(dwork, struct kvm_arch, | |
2507 | kvmclock_sync_work); | |
2508 | struct kvm *kvm = container_of(ka, struct kvm, arch); | |
2509 | ||
630994b3 MT |
2510 | if (!kvmclock_periodic_sync) |
2511 | return; | |
2512 | ||
332967a3 AJ |
2513 | schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0); |
2514 | schedule_delayed_work(&kvm->arch.kvmclock_sync_work, | |
2515 | KVMCLOCK_SYNC_PERIOD); | |
2516 | } | |
2517 | ||
191c8137 BP |
2518 | /* |
2519 | * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP. | |
2520 | */ | |
2521 | static bool can_set_mci_status(struct kvm_vcpu *vcpu) | |
2522 | { | |
2523 | /* McStatusWrEn enabled? */ | |
2524 | if (guest_cpuid_is_amd(vcpu)) | |
2525 | return !!(vcpu->arch.msr_hwcr & BIT_ULL(18)); | |
2526 | ||
2527 | return false; | |
2528 | } | |
2529 | ||
9ffd986c | 2530 | static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
15c4a640 | 2531 | { |
890ca9ae HY |
2532 | u64 mcg_cap = vcpu->arch.mcg_cap; |
2533 | unsigned bank_num = mcg_cap & 0xff; | |
9ffd986c WL |
2534 | u32 msr = msr_info->index; |
2535 | u64 data = msr_info->data; | |
890ca9ae | 2536 | |
15c4a640 | 2537 | switch (msr) { |
15c4a640 | 2538 | case MSR_IA32_MCG_STATUS: |
890ca9ae | 2539 | vcpu->arch.mcg_status = data; |
15c4a640 | 2540 | break; |
c7ac679c | 2541 | case MSR_IA32_MCG_CTL: |
44883f01 PB |
2542 | if (!(mcg_cap & MCG_CTL_P) && |
2543 | (data || !msr_info->host_initiated)) | |
890ca9ae HY |
2544 | return 1; |
2545 | if (data != 0 && data != ~(u64)0) | |
44883f01 | 2546 | return 1; |
890ca9ae HY |
2547 | vcpu->arch.mcg_ctl = data; |
2548 | break; | |
2549 | default: | |
2550 | if (msr >= MSR_IA32_MC0_CTL && | |
81760dcc | 2551 | msr < MSR_IA32_MCx_CTL(bank_num)) { |
6ec4c5ee MP |
2552 | u32 offset = array_index_nospec( |
2553 | msr - MSR_IA32_MC0_CTL, | |
2554 | MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL); | |
2555 | ||
114be429 AP |
2556 | /* only 0 or all 1s can be written to IA32_MCi_CTL |
2557 | * some Linux kernels though clear bit 10 in bank 4 to | |
2558 | * workaround a BIOS/GART TBL issue on AMD K8s, ignore | |
2559 | * this to avoid an uncatched #GP in the guest | |
2560 | */ | |
890ca9ae | 2561 | if ((offset & 0x3) == 0 && |
114be429 | 2562 | data != 0 && (data | (1 << 10)) != ~(u64)0) |
890ca9ae | 2563 | return -1; |
191c8137 BP |
2564 | |
2565 | /* MCi_STATUS */ | |
9ffd986c | 2566 | if (!msr_info->host_initiated && |
191c8137 BP |
2567 | (offset & 0x3) == 1 && data != 0) { |
2568 | if (!can_set_mci_status(vcpu)) | |
2569 | return -1; | |
2570 | } | |
2571 | ||
890ca9ae HY |
2572 | vcpu->arch.mce_banks[offset] = data; |
2573 | break; | |
2574 | } | |
2575 | return 1; | |
2576 | } | |
2577 | return 0; | |
2578 | } | |
2579 | ||
ffde22ac ES |
2580 | static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data) |
2581 | { | |
2582 | struct kvm *kvm = vcpu->kvm; | |
2583 | int lm = is_long_mode(vcpu); | |
2584 | u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64 | |
2585 | : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32; | |
2586 | u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64 | |
2587 | : kvm->arch.xen_hvm_config.blob_size_32; | |
2588 | u32 page_num = data & ~PAGE_MASK; | |
2589 | u64 page_addr = data & PAGE_MASK; | |
2590 | u8 *page; | |
2591 | int r; | |
2592 | ||
2593 | r = -E2BIG; | |
2594 | if (page_num >= blob_size) | |
2595 | goto out; | |
2596 | r = -ENOMEM; | |
ff5c2c03 SL |
2597 | page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE); |
2598 | if (IS_ERR(page)) { | |
2599 | r = PTR_ERR(page); | |
ffde22ac | 2600 | goto out; |
ff5c2c03 | 2601 | } |
54bf36aa | 2602 | if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE)) |
ffde22ac ES |
2603 | goto out_free; |
2604 | r = 0; | |
2605 | out_free: | |
2606 | kfree(page); | |
2607 | out: | |
2608 | return r; | |
2609 | } | |
2610 | ||
344d9588 GN |
2611 | static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data) |
2612 | { | |
2613 | gpa_t gpa = data & ~0x3f; | |
2614 | ||
52a5c155 WL |
2615 | /* Bits 3:5 are reserved, Should be zero */ |
2616 | if (data & 0x38) | |
344d9588 GN |
2617 | return 1; |
2618 | ||
2619 | vcpu->arch.apf.msr_val = data; | |
2620 | ||
2621 | if (!(data & KVM_ASYNC_PF_ENABLED)) { | |
2622 | kvm_clear_async_pf_completion_queue(vcpu); | |
2623 | kvm_async_pf_hash_reset(vcpu); | |
2624 | return 0; | |
2625 | } | |
2626 | ||
4e335d9e | 2627 | if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa, |
8f964525 | 2628 | sizeof(u32))) |
344d9588 GN |
2629 | return 1; |
2630 | ||
6adba527 | 2631 | vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS); |
52a5c155 | 2632 | vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT; |
344d9588 GN |
2633 | kvm_async_pf_wakeup_all(vcpu); |
2634 | return 0; | |
2635 | } | |
2636 | ||
12f9a48f GC |
2637 | static void kvmclock_reset(struct kvm_vcpu *vcpu) |
2638 | { | |
0b79459b | 2639 | vcpu->arch.pv_time_enabled = false; |
49dedf0d | 2640 | vcpu->arch.time = 0; |
12f9a48f GC |
2641 | } |
2642 | ||
f38a7b75 WL |
2643 | static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa) |
2644 | { | |
2645 | ++vcpu->stat.tlb_flush; | |
2646 | kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa); | |
2647 | } | |
2648 | ||
c9aaa895 GC |
2649 | static void record_steal_time(struct kvm_vcpu *vcpu) |
2650 | { | |
2651 | if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED)) | |
2652 | return; | |
2653 | ||
4e335d9e | 2654 | if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime, |
c9aaa895 GC |
2655 | &vcpu->arch.st.steal, sizeof(struct kvm_steal_time)))) |
2656 | return; | |
2657 | ||
f38a7b75 WL |
2658 | /* |
2659 | * Doing a TLB flush here, on the guest's behalf, can avoid | |
2660 | * expensive IPIs. | |
2661 | */ | |
b382f44e WL |
2662 | trace_kvm_pv_tlb_flush(vcpu->vcpu_id, |
2663 | vcpu->arch.st.steal.preempted & KVM_VCPU_FLUSH_TLB); | |
f38a7b75 WL |
2664 | if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB) |
2665 | kvm_vcpu_flush_tlb(vcpu, false); | |
0b9f6c46 | 2666 | |
35f3fae1 WL |
2667 | if (vcpu->arch.st.steal.version & 1) |
2668 | vcpu->arch.st.steal.version += 1; /* first time write, random junk */ | |
2669 | ||
2670 | vcpu->arch.st.steal.version += 1; | |
2671 | ||
4e335d9e | 2672 | kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime, |
35f3fae1 WL |
2673 | &vcpu->arch.st.steal, sizeof(struct kvm_steal_time)); |
2674 | ||
2675 | smp_wmb(); | |
2676 | ||
c54cdf14 LC |
2677 | vcpu->arch.st.steal.steal += current->sched_info.run_delay - |
2678 | vcpu->arch.st.last_steal; | |
2679 | vcpu->arch.st.last_steal = current->sched_info.run_delay; | |
35f3fae1 | 2680 | |
4e335d9e | 2681 | kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime, |
35f3fae1 WL |
2682 | &vcpu->arch.st.steal, sizeof(struct kvm_steal_time)); |
2683 | ||
2684 | smp_wmb(); | |
2685 | ||
2686 | vcpu->arch.st.steal.version += 1; | |
c9aaa895 | 2687 | |
4e335d9e | 2688 | kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime, |
c9aaa895 GC |
2689 | &vcpu->arch.st.steal, sizeof(struct kvm_steal_time)); |
2690 | } | |
2691 | ||
8fe8ab46 | 2692 | int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
15c4a640 | 2693 | { |
5753785f | 2694 | bool pr = false; |
8fe8ab46 WA |
2695 | u32 msr = msr_info->index; |
2696 | u64 data = msr_info->data; | |
5753785f | 2697 | |
15c4a640 | 2698 | switch (msr) { |
2e32b719 | 2699 | case MSR_AMD64_NB_CFG: |
2e32b719 BP |
2700 | case MSR_IA32_UCODE_WRITE: |
2701 | case MSR_VM_HSAVE_PA: | |
2702 | case MSR_AMD64_PATCH_LOADER: | |
2703 | case MSR_AMD64_BU_CFG2: | |
405a353a | 2704 | case MSR_AMD64_DC_CFG: |
0e1b869f | 2705 | case MSR_F15H_EX_CFG: |
2e32b719 BP |
2706 | break; |
2707 | ||
518e7b94 WL |
2708 | case MSR_IA32_UCODE_REV: |
2709 | if (msr_info->host_initiated) | |
2710 | vcpu->arch.microcode_version = data; | |
2711 | break; | |
0cf9135b SC |
2712 | case MSR_IA32_ARCH_CAPABILITIES: |
2713 | if (!msr_info->host_initiated) | |
2714 | return 1; | |
2715 | vcpu->arch.arch_capabilities = data; | |
2716 | break; | |
15c4a640 | 2717 | case MSR_EFER: |
11988499 | 2718 | return set_efer(vcpu, msr_info); |
8f1589d9 AP |
2719 | case MSR_K7_HWCR: |
2720 | data &= ~(u64)0x40; /* ignore flush filter disable */ | |
82494028 | 2721 | data &= ~(u64)0x100; /* ignore ignne emulation enable */ |
a223c313 | 2722 | data &= ~(u64)0x8; /* ignore TLB cache disable */ |
191c8137 BP |
2723 | |
2724 | /* Handle McStatusWrEn */ | |
2725 | if (data == BIT_ULL(18)) { | |
2726 | vcpu->arch.msr_hwcr = data; | |
2727 | } else if (data != 0) { | |
a737f256 CD |
2728 | vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n", |
2729 | data); | |
8f1589d9 AP |
2730 | return 1; |
2731 | } | |
15c4a640 | 2732 | break; |
f7c6d140 AP |
2733 | case MSR_FAM10H_MMIO_CONF_BASE: |
2734 | if (data != 0) { | |
a737f256 CD |
2735 | vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: " |
2736 | "0x%llx\n", data); | |
f7c6d140 AP |
2737 | return 1; |
2738 | } | |
15c4a640 | 2739 | break; |
b5e2fec0 AG |
2740 | case MSR_IA32_DEBUGCTLMSR: |
2741 | if (!data) { | |
2742 | /* We support the non-activated case already */ | |
2743 | break; | |
2744 | } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) { | |
2745 | /* Values other than LBR and BTF are vendor-specific, | |
2746 | thus reserved and should throw a #GP */ | |
2747 | return 1; | |
2748 | } | |
a737f256 CD |
2749 | vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n", |
2750 | __func__, data); | |
b5e2fec0 | 2751 | break; |
9ba075a6 | 2752 | case 0x200 ... 0x2ff: |
ff53604b | 2753 | return kvm_mtrr_set_msr(vcpu, msr, data); |
15c4a640 | 2754 | case MSR_IA32_APICBASE: |
58cb628d | 2755 | return kvm_set_apic_base(vcpu, msr_info); |
0105d1a5 GN |
2756 | case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff: |
2757 | return kvm_x2apic_msr_write(vcpu, msr, data); | |
a3e06bbe LJ |
2758 | case MSR_IA32_TSCDEADLINE: |
2759 | kvm_set_lapic_tscdeadline_msr(vcpu, data); | |
2760 | break; | |
ba904635 | 2761 | case MSR_IA32_TSC_ADJUST: |
d6321d49 | 2762 | if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) { |
ba904635 | 2763 | if (!msr_info->host_initiated) { |
d913b904 | 2764 | s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr; |
d7add054 | 2765 | adjust_tsc_offset_guest(vcpu, adj); |
ba904635 WA |
2766 | } |
2767 | vcpu->arch.ia32_tsc_adjust_msr = data; | |
2768 | } | |
2769 | break; | |
15c4a640 | 2770 | case MSR_IA32_MISC_ENABLE: |
511a8556 WL |
2771 | if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) && |
2772 | ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) { | |
2773 | if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3)) | |
2774 | return 1; | |
2775 | vcpu->arch.ia32_misc_enable_msr = data; | |
2776 | kvm_update_cpuid(vcpu); | |
2777 | } else { | |
2778 | vcpu->arch.ia32_misc_enable_msr = data; | |
2779 | } | |
15c4a640 | 2780 | break; |
64d60670 PB |
2781 | case MSR_IA32_SMBASE: |
2782 | if (!msr_info->host_initiated) | |
2783 | return 1; | |
2784 | vcpu->arch.smbase = data; | |
2785 | break; | |
73f624f4 PB |
2786 | case MSR_IA32_POWER_CTL: |
2787 | vcpu->arch.msr_ia32_power_ctl = data; | |
2788 | break; | |
dd259935 PB |
2789 | case MSR_IA32_TSC: |
2790 | kvm_write_tsc(vcpu, msr_info); | |
2791 | break; | |
864e2ab2 AL |
2792 | case MSR_IA32_XSS: |
2793 | if (!msr_info->host_initiated && | |
2794 | !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES)) | |
2795 | return 1; | |
2796 | /* | |
2797 | * We do support PT if kvm_x86_ops->pt_supported(), but we do | |
2798 | * not support IA32_XSS[bit 8]. Guests will have to use | |
2799 | * RDMSR/WRMSR rather than XSAVES/XRSTORS to save/restore PT | |
2800 | * MSRs. | |
2801 | */ | |
2802 | if (data != 0) | |
2803 | return 1; | |
2804 | vcpu->arch.ia32_xss = data; | |
2805 | break; | |
52797bf9 LA |
2806 | case MSR_SMI_COUNT: |
2807 | if (!msr_info->host_initiated) | |
2808 | return 1; | |
2809 | vcpu->arch.smi_count = data; | |
2810 | break; | |
11c6bffa | 2811 | case MSR_KVM_WALL_CLOCK_NEW: |
18068523 GOC |
2812 | case MSR_KVM_WALL_CLOCK: |
2813 | vcpu->kvm->arch.wall_clock = data; | |
2814 | kvm_write_wall_clock(vcpu->kvm, data); | |
2815 | break; | |
11c6bffa | 2816 | case MSR_KVM_SYSTEM_TIME_NEW: |
18068523 | 2817 | case MSR_KVM_SYSTEM_TIME: { |
54750f2c MT |
2818 | struct kvm_arch *ka = &vcpu->kvm->arch; |
2819 | ||
54750f2c MT |
2820 | if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) { |
2821 | bool tmp = (msr == MSR_KVM_SYSTEM_TIME); | |
2822 | ||
2823 | if (ka->boot_vcpu_runs_old_kvmclock != tmp) | |
1bd2009e | 2824 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); |
54750f2c MT |
2825 | |
2826 | ka->boot_vcpu_runs_old_kvmclock = tmp; | |
2827 | } | |
2828 | ||
18068523 | 2829 | vcpu->arch.time = data; |
0061d53d | 2830 | kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu); |
18068523 GOC |
2831 | |
2832 | /* we verify if the enable bit is set... */ | |
49dedf0d | 2833 | vcpu->arch.pv_time_enabled = false; |
18068523 GOC |
2834 | if (!(data & 1)) |
2835 | break; | |
2836 | ||
49dedf0d | 2837 | if (!kvm_gfn_to_hva_cache_init(vcpu->kvm, |
8f964525 AH |
2838 | &vcpu->arch.pv_time, data & ~1ULL, |
2839 | sizeof(struct pvclock_vcpu_time_info))) | |
0b79459b | 2840 | vcpu->arch.pv_time_enabled = true; |
32cad84f | 2841 | |
18068523 GOC |
2842 | break; |
2843 | } | |
344d9588 GN |
2844 | case MSR_KVM_ASYNC_PF_EN: |
2845 | if (kvm_pv_enable_async_pf(vcpu, data)) | |
2846 | return 1; | |
2847 | break; | |
c9aaa895 GC |
2848 | case MSR_KVM_STEAL_TIME: |
2849 | ||
2850 | if (unlikely(!sched_info_on())) | |
2851 | return 1; | |
2852 | ||
2853 | if (data & KVM_STEAL_RESERVED_MASK) | |
2854 | return 1; | |
2855 | ||
4e335d9e | 2856 | if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime, |
8f964525 AH |
2857 | data & KVM_STEAL_VALID_BITS, |
2858 | sizeof(struct kvm_steal_time))) | |
c9aaa895 GC |
2859 | return 1; |
2860 | ||
2861 | vcpu->arch.st.msr_val = data; | |
2862 | ||
2863 | if (!(data & KVM_MSR_ENABLED)) | |
2864 | break; | |
2865 | ||
c9aaa895 GC |
2866 | kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu); |
2867 | ||
2868 | break; | |
ae7a2a3f | 2869 | case MSR_KVM_PV_EOI_EN: |
72bbf935 | 2870 | if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8))) |
ae7a2a3f MT |
2871 | return 1; |
2872 | break; | |
c9aaa895 | 2873 | |
2d5ba19b MT |
2874 | case MSR_KVM_POLL_CONTROL: |
2875 | /* only enable bit supported */ | |
2876 | if (data & (-1ULL << 1)) | |
2877 | return 1; | |
2878 | ||
2879 | vcpu->arch.msr_kvm_poll_control = data; | |
2880 | break; | |
2881 | ||
890ca9ae HY |
2882 | case MSR_IA32_MCG_CTL: |
2883 | case MSR_IA32_MCG_STATUS: | |
81760dcc | 2884 | case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1: |
9ffd986c | 2885 | return set_msr_mce(vcpu, msr_info); |
71db6023 | 2886 | |
6912ac32 WH |
2887 | case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3: |
2888 | case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1: | |
2889 | pr = true; /* fall through */ | |
2890 | case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3: | |
2891 | case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1: | |
c6702c9d | 2892 | if (kvm_pmu_is_valid_msr(vcpu, msr)) |
afd80d85 | 2893 | return kvm_pmu_set_msr(vcpu, msr_info); |
5753785f GN |
2894 | |
2895 | if (pr || data != 0) | |
a737f256 CD |
2896 | vcpu_unimpl(vcpu, "disabled perfctr wrmsr: " |
2897 | "0x%x data 0x%llx\n", msr, data); | |
5753785f | 2898 | break; |
84e0cefa JS |
2899 | case MSR_K7_CLK_CTL: |
2900 | /* | |
2901 | * Ignore all writes to this no longer documented MSR. | |
2902 | * Writes are only relevant for old K7 processors, | |
2903 | * all pre-dating SVM, but a recommended workaround from | |
4a969980 | 2904 | * AMD for these chips. It is possible to specify the |
84e0cefa JS |
2905 | * affected processor models on the command line, hence |
2906 | * the need to ignore the workaround. | |
2907 | */ | |
2908 | break; | |
55cd8e5a | 2909 | case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15: |
e7d9513b AS |
2910 | case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4: |
2911 | case HV_X64_MSR_CRASH_CTL: | |
1f4b34f8 | 2912 | case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT: |
a2e164e7 VK |
2913 | case HV_X64_MSR_REENLIGHTENMENT_CONTROL: |
2914 | case HV_X64_MSR_TSC_EMULATION_CONTROL: | |
2915 | case HV_X64_MSR_TSC_EMULATION_STATUS: | |
e7d9513b AS |
2916 | return kvm_hv_set_msr_common(vcpu, msr, data, |
2917 | msr_info->host_initiated); | |
91c9c3ed | 2918 | case MSR_IA32_BBL_CR_CTL3: |
2919 | /* Drop writes to this legacy MSR -- see rdmsr | |
2920 | * counterpart for further detail. | |
2921 | */ | |
fab0aa3b EM |
2922 | if (report_ignored_msrs) |
2923 | vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n", | |
2924 | msr, data); | |
91c9c3ed | 2925 | break; |
2b036c6b | 2926 | case MSR_AMD64_OSVW_ID_LENGTH: |
d6321d49 | 2927 | if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW)) |
2b036c6b BO |
2928 | return 1; |
2929 | vcpu->arch.osvw.length = data; | |
2930 | break; | |
2931 | case MSR_AMD64_OSVW_STATUS: | |
d6321d49 | 2932 | if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW)) |
2b036c6b BO |
2933 | return 1; |
2934 | vcpu->arch.osvw.status = data; | |
2935 | break; | |
db2336a8 KH |
2936 | case MSR_PLATFORM_INFO: |
2937 | if (!msr_info->host_initiated || | |
db2336a8 KH |
2938 | (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) && |
2939 | cpuid_fault_enabled(vcpu))) | |
2940 | return 1; | |
2941 | vcpu->arch.msr_platform_info = data; | |
2942 | break; | |
2943 | case MSR_MISC_FEATURES_ENABLES: | |
2944 | if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT || | |
2945 | (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT && | |
2946 | !supports_cpuid_fault(vcpu))) | |
2947 | return 1; | |
2948 | vcpu->arch.msr_misc_features_enables = data; | |
2949 | break; | |
15c4a640 | 2950 | default: |
ffde22ac ES |
2951 | if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr)) |
2952 | return xen_hvm_config(vcpu, data); | |
c6702c9d | 2953 | if (kvm_pmu_is_valid_msr(vcpu, msr)) |
afd80d85 | 2954 | return kvm_pmu_set_msr(vcpu, msr_info); |
ed85c068 | 2955 | if (!ignore_msrs) { |
ae0f5499 | 2956 | vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n", |
a737f256 | 2957 | msr, data); |
ed85c068 AP |
2958 | return 1; |
2959 | } else { | |
fab0aa3b EM |
2960 | if (report_ignored_msrs) |
2961 | vcpu_unimpl(vcpu, | |
2962 | "ignored wrmsr: 0x%x data 0x%llx\n", | |
2963 | msr, data); | |
ed85c068 AP |
2964 | break; |
2965 | } | |
15c4a640 CO |
2966 | } |
2967 | return 0; | |
2968 | } | |
2969 | EXPORT_SYMBOL_GPL(kvm_set_msr_common); | |
2970 | ||
44883f01 | 2971 | static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host) |
15c4a640 CO |
2972 | { |
2973 | u64 data; | |
890ca9ae HY |
2974 | u64 mcg_cap = vcpu->arch.mcg_cap; |
2975 | unsigned bank_num = mcg_cap & 0xff; | |
15c4a640 CO |
2976 | |
2977 | switch (msr) { | |
15c4a640 CO |
2978 | case MSR_IA32_P5_MC_ADDR: |
2979 | case MSR_IA32_P5_MC_TYPE: | |
890ca9ae HY |
2980 | data = 0; |
2981 | break; | |
15c4a640 | 2982 | case MSR_IA32_MCG_CAP: |
890ca9ae HY |
2983 | data = vcpu->arch.mcg_cap; |
2984 | break; | |
c7ac679c | 2985 | case MSR_IA32_MCG_CTL: |
44883f01 | 2986 | if (!(mcg_cap & MCG_CTL_P) && !host) |
890ca9ae HY |
2987 | return 1; |
2988 | data = vcpu->arch.mcg_ctl; | |
2989 | break; | |
2990 | case MSR_IA32_MCG_STATUS: | |
2991 | data = vcpu->arch.mcg_status; | |
2992 | break; | |
2993 | default: | |
2994 | if (msr >= MSR_IA32_MC0_CTL && | |
81760dcc | 2995 | msr < MSR_IA32_MCx_CTL(bank_num)) { |
6ec4c5ee MP |
2996 | u32 offset = array_index_nospec( |
2997 | msr - MSR_IA32_MC0_CTL, | |
2998 | MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL); | |
2999 | ||
890ca9ae HY |
3000 | data = vcpu->arch.mce_banks[offset]; |
3001 | break; | |
3002 | } | |
3003 | return 1; | |
3004 | } | |
3005 | *pdata = data; | |
3006 | return 0; | |
3007 | } | |
3008 | ||
609e36d3 | 3009 | int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info) |
890ca9ae | 3010 | { |
609e36d3 | 3011 | switch (msr_info->index) { |
890ca9ae | 3012 | case MSR_IA32_PLATFORM_ID: |
15c4a640 | 3013 | case MSR_IA32_EBL_CR_POWERON: |
b5e2fec0 AG |
3014 | case MSR_IA32_DEBUGCTLMSR: |
3015 | case MSR_IA32_LASTBRANCHFROMIP: | |
3016 | case MSR_IA32_LASTBRANCHTOIP: | |
3017 | case MSR_IA32_LASTINTFROMIP: | |
3018 | case MSR_IA32_LASTINTTOIP: | |
60af2ecd | 3019 | case MSR_K8_SYSCFG: |
3afb1121 PB |
3020 | case MSR_K8_TSEG_ADDR: |
3021 | case MSR_K8_TSEG_MASK: | |
61a6bd67 | 3022 | case MSR_VM_HSAVE_PA: |
1fdbd48c | 3023 | case MSR_K8_INT_PENDING_MSG: |
c323c0e5 | 3024 | case MSR_AMD64_NB_CFG: |
f7c6d140 | 3025 | case MSR_FAM10H_MMIO_CONF_BASE: |
2e32b719 | 3026 | case MSR_AMD64_BU_CFG2: |
0c2df2a1 | 3027 | case MSR_IA32_PERF_CTL: |
405a353a | 3028 | case MSR_AMD64_DC_CFG: |
0e1b869f | 3029 | case MSR_F15H_EX_CFG: |
609e36d3 | 3030 | msr_info->data = 0; |
15c4a640 | 3031 | break; |
c51eb52b | 3032 | case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5: |
6912ac32 WH |
3033 | case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3: |
3034 | case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3: | |
3035 | case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1: | |
3036 | case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1: | |
c6702c9d | 3037 | if (kvm_pmu_is_valid_msr(vcpu, msr_info->index)) |
609e36d3 PB |
3038 | return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data); |
3039 | msr_info->data = 0; | |
5753785f | 3040 | break; |
742bc670 | 3041 | case MSR_IA32_UCODE_REV: |
518e7b94 | 3042 | msr_info->data = vcpu->arch.microcode_version; |
742bc670 | 3043 | break; |
0cf9135b SC |
3044 | case MSR_IA32_ARCH_CAPABILITIES: |
3045 | if (!msr_info->host_initiated && | |
3046 | !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES)) | |
3047 | return 1; | |
3048 | msr_info->data = vcpu->arch.arch_capabilities; | |
3049 | break; | |
73f624f4 PB |
3050 | case MSR_IA32_POWER_CTL: |
3051 | msr_info->data = vcpu->arch.msr_ia32_power_ctl; | |
3052 | break; | |
dd259935 PB |
3053 | case MSR_IA32_TSC: |
3054 | msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset; | |
3055 | break; | |
9ba075a6 | 3056 | case MSR_MTRRcap: |
9ba075a6 | 3057 | case 0x200 ... 0x2ff: |
ff53604b | 3058 | return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data); |
15c4a640 | 3059 | case 0xcd: /* fsb frequency */ |
609e36d3 | 3060 | msr_info->data = 3; |
15c4a640 | 3061 | break; |
7b914098 JS |
3062 | /* |
3063 | * MSR_EBC_FREQUENCY_ID | |
3064 | * Conservative value valid for even the basic CPU models. | |
3065 | * Models 0,1: 000 in bits 23:21 indicating a bus speed of | |
3066 | * 100MHz, model 2 000 in bits 18:16 indicating 100MHz, | |
3067 | * and 266MHz for model 3, or 4. Set Core Clock | |
3068 | * Frequency to System Bus Frequency Ratio to 1 (bits | |
3069 | * 31:24) even though these are only valid for CPU | |
3070 | * models > 2, however guests may end up dividing or | |
3071 | * multiplying by zero otherwise. | |
3072 | */ | |
3073 | case MSR_EBC_FREQUENCY_ID: | |
609e36d3 | 3074 | msr_info->data = 1 << 24; |
7b914098 | 3075 | break; |
15c4a640 | 3076 | case MSR_IA32_APICBASE: |
609e36d3 | 3077 | msr_info->data = kvm_get_apic_base(vcpu); |
15c4a640 | 3078 | break; |
0105d1a5 | 3079 | case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff: |
609e36d3 | 3080 | return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data); |
0105d1a5 | 3081 | break; |
a3e06bbe | 3082 | case MSR_IA32_TSCDEADLINE: |
609e36d3 | 3083 | msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu); |
a3e06bbe | 3084 | break; |
ba904635 | 3085 | case MSR_IA32_TSC_ADJUST: |
609e36d3 | 3086 | msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr; |
ba904635 | 3087 | break; |
15c4a640 | 3088 | case MSR_IA32_MISC_ENABLE: |
609e36d3 | 3089 | msr_info->data = vcpu->arch.ia32_misc_enable_msr; |
15c4a640 | 3090 | break; |
64d60670 PB |
3091 | case MSR_IA32_SMBASE: |
3092 | if (!msr_info->host_initiated) | |
3093 | return 1; | |
3094 | msr_info->data = vcpu->arch.smbase; | |
15c4a640 | 3095 | break; |
52797bf9 LA |
3096 | case MSR_SMI_COUNT: |
3097 | msr_info->data = vcpu->arch.smi_count; | |
3098 | break; | |
847f0ad8 AG |
3099 | case MSR_IA32_PERF_STATUS: |
3100 | /* TSC increment by tick */ | |
609e36d3 | 3101 | msr_info->data = 1000ULL; |
847f0ad8 | 3102 | /* CPU multiplier */ |
b0996ae4 | 3103 | msr_info->data |= (((uint64_t)4ULL) << 40); |
847f0ad8 | 3104 | break; |
15c4a640 | 3105 | case MSR_EFER: |
609e36d3 | 3106 | msr_info->data = vcpu->arch.efer; |
15c4a640 | 3107 | break; |
18068523 | 3108 | case MSR_KVM_WALL_CLOCK: |
11c6bffa | 3109 | case MSR_KVM_WALL_CLOCK_NEW: |
609e36d3 | 3110 | msr_info->data = vcpu->kvm->arch.wall_clock; |
18068523 GOC |
3111 | break; |
3112 | case MSR_KVM_SYSTEM_TIME: | |
11c6bffa | 3113 | case MSR_KVM_SYSTEM_TIME_NEW: |
609e36d3 | 3114 | msr_info->data = vcpu->arch.time; |
18068523 | 3115 | break; |
344d9588 | 3116 | case MSR_KVM_ASYNC_PF_EN: |
609e36d3 | 3117 | msr_info->data = vcpu->arch.apf.msr_val; |
344d9588 | 3118 | break; |
c9aaa895 | 3119 | case MSR_KVM_STEAL_TIME: |
609e36d3 | 3120 | msr_info->data = vcpu->arch.st.msr_val; |
c9aaa895 | 3121 | break; |
1d92128f | 3122 | case MSR_KVM_PV_EOI_EN: |
609e36d3 | 3123 | msr_info->data = vcpu->arch.pv_eoi.msr_val; |
1d92128f | 3124 | break; |
2d5ba19b MT |
3125 | case MSR_KVM_POLL_CONTROL: |
3126 | msr_info->data = vcpu->arch.msr_kvm_poll_control; | |
3127 | break; | |
890ca9ae HY |
3128 | case MSR_IA32_P5_MC_ADDR: |
3129 | case MSR_IA32_P5_MC_TYPE: | |
3130 | case MSR_IA32_MCG_CAP: | |
3131 | case MSR_IA32_MCG_CTL: | |
3132 | case MSR_IA32_MCG_STATUS: | |
81760dcc | 3133 | case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1: |
44883f01 PB |
3134 | return get_msr_mce(vcpu, msr_info->index, &msr_info->data, |
3135 | msr_info->host_initiated); | |
864e2ab2 AL |
3136 | case MSR_IA32_XSS: |
3137 | if (!msr_info->host_initiated && | |
3138 | !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES)) | |
3139 | return 1; | |
3140 | msr_info->data = vcpu->arch.ia32_xss; | |
3141 | break; | |
84e0cefa JS |
3142 | case MSR_K7_CLK_CTL: |
3143 | /* | |
3144 | * Provide expected ramp-up count for K7. All other | |
3145 | * are set to zero, indicating minimum divisors for | |
3146 | * every field. | |
3147 | * | |
3148 | * This prevents guest kernels on AMD host with CPU | |
3149 | * type 6, model 8 and higher from exploding due to | |
3150 | * the rdmsr failing. | |
3151 | */ | |
609e36d3 | 3152 | msr_info->data = 0x20000000; |
84e0cefa | 3153 | break; |
55cd8e5a | 3154 | case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15: |
e7d9513b AS |
3155 | case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4: |
3156 | case HV_X64_MSR_CRASH_CTL: | |
1f4b34f8 | 3157 | case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT: |
a2e164e7 VK |
3158 | case HV_X64_MSR_REENLIGHTENMENT_CONTROL: |
3159 | case HV_X64_MSR_TSC_EMULATION_CONTROL: | |
3160 | case HV_X64_MSR_TSC_EMULATION_STATUS: | |
e83d5887 | 3161 | return kvm_hv_get_msr_common(vcpu, |
44883f01 PB |
3162 | msr_info->index, &msr_info->data, |
3163 | msr_info->host_initiated); | |
55cd8e5a | 3164 | break; |
91c9c3ed | 3165 | case MSR_IA32_BBL_CR_CTL3: |
3166 | /* This legacy MSR exists but isn't fully documented in current | |
3167 | * silicon. It is however accessed by winxp in very narrow | |
3168 | * scenarios where it sets bit #19, itself documented as | |
3169 | * a "reserved" bit. Best effort attempt to source coherent | |
3170 | * read data here should the balance of the register be | |
3171 | * interpreted by the guest: | |
3172 | * | |
3173 | * L2 cache control register 3: 64GB range, 256KB size, | |
3174 | * enabled, latency 0x1, configured | |
3175 | */ | |
609e36d3 | 3176 | msr_info->data = 0xbe702111; |
91c9c3ed | 3177 | break; |
2b036c6b | 3178 | case MSR_AMD64_OSVW_ID_LENGTH: |
d6321d49 | 3179 | if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW)) |
2b036c6b | 3180 | return 1; |
609e36d3 | 3181 | msr_info->data = vcpu->arch.osvw.length; |
2b036c6b BO |
3182 | break; |
3183 | case MSR_AMD64_OSVW_STATUS: | |
d6321d49 | 3184 | if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW)) |
2b036c6b | 3185 | return 1; |
609e36d3 | 3186 | msr_info->data = vcpu->arch.osvw.status; |
2b036c6b | 3187 | break; |
db2336a8 | 3188 | case MSR_PLATFORM_INFO: |
6fbbde9a DS |
3189 | if (!msr_info->host_initiated && |
3190 | !vcpu->kvm->arch.guest_can_read_msr_platform_info) | |
3191 | return 1; | |
db2336a8 KH |
3192 | msr_info->data = vcpu->arch.msr_platform_info; |
3193 | break; | |
3194 | case MSR_MISC_FEATURES_ENABLES: | |
3195 | msr_info->data = vcpu->arch.msr_misc_features_enables; | |
3196 | break; | |
191c8137 BP |
3197 | case MSR_K7_HWCR: |
3198 | msr_info->data = vcpu->arch.msr_hwcr; | |
3199 | break; | |
15c4a640 | 3200 | default: |
c6702c9d | 3201 | if (kvm_pmu_is_valid_msr(vcpu, msr_info->index)) |
609e36d3 | 3202 | return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data); |
ed85c068 | 3203 | if (!ignore_msrs) { |
ae0f5499 BD |
3204 | vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n", |
3205 | msr_info->index); | |
ed85c068 AP |
3206 | return 1; |
3207 | } else { | |
fab0aa3b EM |
3208 | if (report_ignored_msrs) |
3209 | vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", | |
3210 | msr_info->index); | |
609e36d3 | 3211 | msr_info->data = 0; |
ed85c068 AP |
3212 | } |
3213 | break; | |
15c4a640 | 3214 | } |
15c4a640 CO |
3215 | return 0; |
3216 | } | |
3217 | EXPORT_SYMBOL_GPL(kvm_get_msr_common); | |
3218 | ||
313a3dc7 CO |
3219 | /* |
3220 | * Read or write a bunch of msrs. All parameters are kernel addresses. | |
3221 | * | |
3222 | * @return number of msrs set successfully. | |
3223 | */ | |
3224 | static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs, | |
3225 | struct kvm_msr_entry *entries, | |
3226 | int (*do_msr)(struct kvm_vcpu *vcpu, | |
3227 | unsigned index, u64 *data)) | |
3228 | { | |
801e459a | 3229 | int i; |
313a3dc7 | 3230 | |
313a3dc7 CO |
3231 | for (i = 0; i < msrs->nmsrs; ++i) |
3232 | if (do_msr(vcpu, entries[i].index, &entries[i].data)) | |
3233 | break; | |
3234 | ||
313a3dc7 CO |
3235 | return i; |
3236 | } | |
3237 | ||
3238 | /* | |
3239 | * Read or write a bunch of msrs. Parameters are user addresses. | |
3240 | * | |
3241 | * @return number of msrs set successfully. | |
3242 | */ | |
3243 | static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs, | |
3244 | int (*do_msr)(struct kvm_vcpu *vcpu, | |
3245 | unsigned index, u64 *data), | |
3246 | int writeback) | |
3247 | { | |
3248 | struct kvm_msrs msrs; | |
3249 | struct kvm_msr_entry *entries; | |
3250 | int r, n; | |
3251 | unsigned size; | |
3252 | ||
3253 | r = -EFAULT; | |
0e96f31e | 3254 | if (copy_from_user(&msrs, user_msrs, sizeof(msrs))) |
313a3dc7 CO |
3255 | goto out; |
3256 | ||
3257 | r = -E2BIG; | |
3258 | if (msrs.nmsrs >= MAX_IO_MSRS) | |
3259 | goto out; | |
3260 | ||
313a3dc7 | 3261 | size = sizeof(struct kvm_msr_entry) * msrs.nmsrs; |
ff5c2c03 SL |
3262 | entries = memdup_user(user_msrs->entries, size); |
3263 | if (IS_ERR(entries)) { | |
3264 | r = PTR_ERR(entries); | |
313a3dc7 | 3265 | goto out; |
ff5c2c03 | 3266 | } |
313a3dc7 CO |
3267 | |
3268 | r = n = __msr_io(vcpu, &msrs, entries, do_msr); | |
3269 | if (r < 0) | |
3270 | goto out_free; | |
3271 | ||
3272 | r = -EFAULT; | |
3273 | if (writeback && copy_to_user(user_msrs->entries, entries, size)) | |
3274 | goto out_free; | |
3275 | ||
3276 | r = n; | |
3277 | ||
3278 | out_free: | |
7a73c028 | 3279 | kfree(entries); |
313a3dc7 CO |
3280 | out: |
3281 | return r; | |
3282 | } | |
3283 | ||
4d5422ce WL |
3284 | static inline bool kvm_can_mwait_in_guest(void) |
3285 | { | |
3286 | return boot_cpu_has(X86_FEATURE_MWAIT) && | |
8e9b29b6 KA |
3287 | !boot_cpu_has_bug(X86_BUG_MONITOR) && |
3288 | boot_cpu_has(X86_FEATURE_ARAT); | |
4d5422ce WL |
3289 | } |
3290 | ||
784aa3d7 | 3291 | int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) |
018d00d2 | 3292 | { |
4d5422ce | 3293 | int r = 0; |
018d00d2 ZX |
3294 | |
3295 | switch (ext) { | |
3296 | case KVM_CAP_IRQCHIP: | |
3297 | case KVM_CAP_HLT: | |
3298 | case KVM_CAP_MMU_SHADOW_CACHE_CONTROL: | |
018d00d2 | 3299 | case KVM_CAP_SET_TSS_ADDR: |
07716717 | 3300 | case KVM_CAP_EXT_CPUID: |
9c15bb1d | 3301 | case KVM_CAP_EXT_EMUL_CPUID: |
c8076604 | 3302 | case KVM_CAP_CLOCKSOURCE: |
7837699f | 3303 | case KVM_CAP_PIT: |
a28e4f5a | 3304 | case KVM_CAP_NOP_IO_DELAY: |
62d9f0db | 3305 | case KVM_CAP_MP_STATE: |
ed848624 | 3306 | case KVM_CAP_SYNC_MMU: |
a355c85c | 3307 | case KVM_CAP_USER_NMI: |
52d939a0 | 3308 | case KVM_CAP_REINJECT_CONTROL: |
4925663a | 3309 | case KVM_CAP_IRQ_INJECT_STATUS: |
d34e6b17 | 3310 | case KVM_CAP_IOEVENTFD: |
f848a5a8 | 3311 | case KVM_CAP_IOEVENTFD_NO_LENGTH: |
c5ff41ce | 3312 | case KVM_CAP_PIT2: |
e9f42757 | 3313 | case KVM_CAP_PIT_STATE2: |
b927a3ce | 3314 | case KVM_CAP_SET_IDENTITY_MAP_ADDR: |
ffde22ac | 3315 | case KVM_CAP_XEN_HVM: |
3cfc3092 | 3316 | case KVM_CAP_VCPU_EVENTS: |
55cd8e5a | 3317 | case KVM_CAP_HYPERV: |
10388a07 | 3318 | case KVM_CAP_HYPERV_VAPIC: |
c25bc163 | 3319 | case KVM_CAP_HYPERV_SPIN: |
5c919412 | 3320 | case KVM_CAP_HYPERV_SYNIC: |
efc479e6 | 3321 | case KVM_CAP_HYPERV_SYNIC2: |
d3457c87 | 3322 | case KVM_CAP_HYPERV_VP_INDEX: |
faeb7833 | 3323 | case KVM_CAP_HYPERV_EVENTFD: |
c1aea919 | 3324 | case KVM_CAP_HYPERV_TLBFLUSH: |
214ff83d | 3325 | case KVM_CAP_HYPERV_SEND_IPI: |
2bc39970 | 3326 | case KVM_CAP_HYPERV_CPUID: |
ab9f4ecb | 3327 | case KVM_CAP_PCI_SEGMENT: |
a1efbe77 | 3328 | case KVM_CAP_DEBUGREGS: |
d2be1651 | 3329 | case KVM_CAP_X86_ROBUST_SINGLESTEP: |
2d5b5a66 | 3330 | case KVM_CAP_XSAVE: |
344d9588 | 3331 | case KVM_CAP_ASYNC_PF: |
92a1f12d | 3332 | case KVM_CAP_GET_TSC_KHZ: |
1c0b28c2 | 3333 | case KVM_CAP_KVMCLOCK_CTRL: |
4d8b81ab | 3334 | case KVM_CAP_READONLY_MEM: |
5f66b620 | 3335 | case KVM_CAP_HYPERV_TIME: |
100943c5 | 3336 | case KVM_CAP_IOAPIC_POLARITY_IGNORED: |
defcf51f | 3337 | case KVM_CAP_TSC_DEADLINE_TIMER: |
90de4a18 | 3338 | case KVM_CAP_DISABLE_QUIRKS: |
d71ba788 | 3339 | case KVM_CAP_SET_BOOT_CPU_ID: |
49df6397 | 3340 | case KVM_CAP_SPLIT_IRQCHIP: |
460df4c1 | 3341 | case KVM_CAP_IMMEDIATE_EXIT: |
66bb8a06 | 3342 | case KVM_CAP_PMU_EVENT_FILTER: |
801e459a | 3343 | case KVM_CAP_GET_MSR_FEATURES: |
6fbbde9a | 3344 | case KVM_CAP_MSR_PLATFORM_INFO: |
c4f55198 | 3345 | case KVM_CAP_EXCEPTION_PAYLOAD: |
018d00d2 ZX |
3346 | r = 1; |
3347 | break; | |
01643c51 KH |
3348 | case KVM_CAP_SYNC_REGS: |
3349 | r = KVM_SYNC_X86_VALID_FIELDS; | |
3350 | break; | |
e3fd9a93 PB |
3351 | case KVM_CAP_ADJUST_CLOCK: |
3352 | r = KVM_CLOCK_TSC_STABLE; | |
3353 | break; | |
4d5422ce | 3354 | case KVM_CAP_X86_DISABLE_EXITS: |
b5170063 WL |
3355 | r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE | |
3356 | KVM_X86_DISABLE_EXITS_CSTATE; | |
4d5422ce WL |
3357 | if(kvm_can_mwait_in_guest()) |
3358 | r |= KVM_X86_DISABLE_EXITS_MWAIT; | |
668fffa3 | 3359 | break; |
6d396b55 PB |
3360 | case KVM_CAP_X86_SMM: |
3361 | /* SMBASE is usually relocated above 1M on modern chipsets, | |
3362 | * and SMM handlers might indeed rely on 4G segment limits, | |
3363 | * so do not report SMM to be available if real mode is | |
3364 | * emulated via vm86 mode. Still, do not go to great lengths | |
3365 | * to avoid userspace's usage of the feature, because it is a | |
3366 | * fringe case that is not enabled except via specific settings | |
3367 | * of the module parameters. | |
3368 | */ | |
bc226f07 | 3369 | r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE); |
6d396b55 | 3370 | break; |
774ead3a AK |
3371 | case KVM_CAP_VAPIC: |
3372 | r = !kvm_x86_ops->cpu_has_accelerated_tpr(); | |
3373 | break; | |
f725230a | 3374 | case KVM_CAP_NR_VCPUS: |
8c3ba334 SL |
3375 | r = KVM_SOFT_MAX_VCPUS; |
3376 | break; | |
3377 | case KVM_CAP_MAX_VCPUS: | |
f725230a AK |
3378 | r = KVM_MAX_VCPUS; |
3379 | break; | |
a86cb413 TH |
3380 | case KVM_CAP_MAX_VCPU_ID: |
3381 | r = KVM_MAX_VCPU_ID; | |
3382 | break; | |
a68a6a72 MT |
3383 | case KVM_CAP_PV_MMU: /* obsolete */ |
3384 | r = 0; | |
2f333bcb | 3385 | break; |
890ca9ae HY |
3386 | case KVM_CAP_MCE: |
3387 | r = KVM_MAX_MCE_BANKS; | |
3388 | break; | |
2d5b5a66 | 3389 | case KVM_CAP_XCRS: |
d366bf7e | 3390 | r = boot_cpu_has(X86_FEATURE_XSAVE); |
2d5b5a66 | 3391 | break; |
92a1f12d JR |
3392 | case KVM_CAP_TSC_CONTROL: |
3393 | r = kvm_has_tsc_control; | |
3394 | break; | |
37131313 RK |
3395 | case KVM_CAP_X2APIC_API: |
3396 | r = KVM_X2APIC_API_VALID_FLAGS; | |
3397 | break; | |
8fcc4b59 JM |
3398 | case KVM_CAP_NESTED_STATE: |
3399 | r = kvm_x86_ops->get_nested_state ? | |
be43c440 | 3400 | kvm_x86_ops->get_nested_state(NULL, NULL, 0) : 0; |
8fcc4b59 | 3401 | break; |
344c6c80 | 3402 | case KVM_CAP_HYPERV_DIRECT_TLBFLUSH: |
5a0165f6 VK |
3403 | r = kvm_x86_ops->enable_direct_tlbflush != NULL; |
3404 | break; | |
3405 | case KVM_CAP_HYPERV_ENLIGHTENED_VMCS: | |
3406 | r = kvm_x86_ops->nested_enable_evmcs != NULL; | |
344c6c80 | 3407 | break; |
018d00d2 | 3408 | default: |
018d00d2 ZX |
3409 | break; |
3410 | } | |
3411 | return r; | |
3412 | ||
3413 | } | |
3414 | ||
043405e1 CO |
3415 | long kvm_arch_dev_ioctl(struct file *filp, |
3416 | unsigned int ioctl, unsigned long arg) | |
3417 | { | |
3418 | void __user *argp = (void __user *)arg; | |
3419 | long r; | |
3420 | ||
3421 | switch (ioctl) { | |
3422 | case KVM_GET_MSR_INDEX_LIST: { | |
3423 | struct kvm_msr_list __user *user_msr_list = argp; | |
3424 | struct kvm_msr_list msr_list; | |
3425 | unsigned n; | |
3426 | ||
3427 | r = -EFAULT; | |
0e96f31e | 3428 | if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list))) |
043405e1 CO |
3429 | goto out; |
3430 | n = msr_list.nmsrs; | |
62ef68bb | 3431 | msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs; |
0e96f31e | 3432 | if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list))) |
043405e1 CO |
3433 | goto out; |
3434 | r = -E2BIG; | |
e125e7b6 | 3435 | if (n < msr_list.nmsrs) |
043405e1 CO |
3436 | goto out; |
3437 | r = -EFAULT; | |
3438 | if (copy_to_user(user_msr_list->indices, &msrs_to_save, | |
3439 | num_msrs_to_save * sizeof(u32))) | |
3440 | goto out; | |
e125e7b6 | 3441 | if (copy_to_user(user_msr_list->indices + num_msrs_to_save, |
043405e1 | 3442 | &emulated_msrs, |
62ef68bb | 3443 | num_emulated_msrs * sizeof(u32))) |
043405e1 CO |
3444 | goto out; |
3445 | r = 0; | |
3446 | break; | |
3447 | } | |
9c15bb1d BP |
3448 | case KVM_GET_SUPPORTED_CPUID: |
3449 | case KVM_GET_EMULATED_CPUID: { | |
674eea0f AK |
3450 | struct kvm_cpuid2 __user *cpuid_arg = argp; |
3451 | struct kvm_cpuid2 cpuid; | |
3452 | ||
3453 | r = -EFAULT; | |
0e96f31e | 3454 | if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid))) |
674eea0f | 3455 | goto out; |
9c15bb1d BP |
3456 | |
3457 | r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries, | |
3458 | ioctl); | |
674eea0f AK |
3459 | if (r) |
3460 | goto out; | |
3461 | ||
3462 | r = -EFAULT; | |
0e96f31e | 3463 | if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid))) |
674eea0f AK |
3464 | goto out; |
3465 | r = 0; | |
3466 | break; | |
3467 | } | |
890ca9ae | 3468 | case KVM_X86_GET_MCE_CAP_SUPPORTED: { |
890ca9ae | 3469 | r = -EFAULT; |
c45dcc71 AR |
3470 | if (copy_to_user(argp, &kvm_mce_cap_supported, |
3471 | sizeof(kvm_mce_cap_supported))) | |
890ca9ae HY |
3472 | goto out; |
3473 | r = 0; | |
3474 | break; | |
801e459a TL |
3475 | case KVM_GET_MSR_FEATURE_INDEX_LIST: { |
3476 | struct kvm_msr_list __user *user_msr_list = argp; | |
3477 | struct kvm_msr_list msr_list; | |
3478 | unsigned int n; | |
3479 | ||
3480 | r = -EFAULT; | |
3481 | if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list))) | |
3482 | goto out; | |
3483 | n = msr_list.nmsrs; | |
3484 | msr_list.nmsrs = num_msr_based_features; | |
3485 | if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list))) | |
3486 | goto out; | |
3487 | r = -E2BIG; | |
3488 | if (n < msr_list.nmsrs) | |
3489 | goto out; | |
3490 | r = -EFAULT; | |
3491 | if (copy_to_user(user_msr_list->indices, &msr_based_features, | |
3492 | num_msr_based_features * sizeof(u32))) | |
3493 | goto out; | |
3494 | r = 0; | |
3495 | break; | |
3496 | } | |
3497 | case KVM_GET_MSRS: | |
3498 | r = msr_io(NULL, argp, do_get_msr_feature, 1); | |
3499 | break; | |
890ca9ae | 3500 | } |
043405e1 CO |
3501 | default: |
3502 | r = -EINVAL; | |
3503 | } | |
3504 | out: | |
3505 | return r; | |
3506 | } | |
3507 | ||
f5f48ee1 SY |
3508 | static void wbinvd_ipi(void *garbage) |
3509 | { | |
3510 | wbinvd(); | |
3511 | } | |
3512 | ||
3513 | static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu) | |
3514 | { | |
e0f0bbc5 | 3515 | return kvm_arch_has_noncoherent_dma(vcpu->kvm); |
f5f48ee1 SY |
3516 | } |
3517 | ||
313a3dc7 CO |
3518 | void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) |
3519 | { | |
f5f48ee1 SY |
3520 | /* Address WBINVD may be executed by guest */ |
3521 | if (need_emulate_wbinvd(vcpu)) { | |
3522 | if (kvm_x86_ops->has_wbinvd_exit()) | |
3523 | cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask); | |
3524 | else if (vcpu->cpu != -1 && vcpu->cpu != cpu) | |
3525 | smp_call_function_single(vcpu->cpu, | |
3526 | wbinvd_ipi, NULL, 1); | |
3527 | } | |
3528 | ||
313a3dc7 | 3529 | kvm_x86_ops->vcpu_load(vcpu, cpu); |
8f6055cb | 3530 | |
0dd6a6ed ZA |
3531 | /* Apply any externally detected TSC adjustments (due to suspend) */ |
3532 | if (unlikely(vcpu->arch.tsc_offset_adjustment)) { | |
3533 | adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment); | |
3534 | vcpu->arch.tsc_offset_adjustment = 0; | |
105b21bb | 3535 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
0dd6a6ed | 3536 | } |
8f6055cb | 3537 | |
b0c39dc6 | 3538 | if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) { |
6f526ec5 | 3539 | s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 : |
4ea1636b | 3540 | rdtsc() - vcpu->arch.last_host_tsc; |
e48672fa ZA |
3541 | if (tsc_delta < 0) |
3542 | mark_tsc_unstable("KVM discovered backwards TSC"); | |
ce7a058a | 3543 | |
b0c39dc6 | 3544 | if (kvm_check_tsc_unstable()) { |
07c1419a | 3545 | u64 offset = kvm_compute_tsc_offset(vcpu, |
b183aa58 | 3546 | vcpu->arch.last_guest_tsc); |
a545ab6a | 3547 | kvm_vcpu_write_tsc_offset(vcpu, offset); |
c285545f | 3548 | vcpu->arch.tsc_catchup = 1; |
c285545f | 3549 | } |
a749e247 PB |
3550 | |
3551 | if (kvm_lapic_hv_timer_in_use(vcpu)) | |
3552 | kvm_lapic_restart_hv_timer(vcpu); | |
3553 | ||
d98d07ca MT |
3554 | /* |
3555 | * On a host with synchronized TSC, there is no need to update | |
3556 | * kvmclock on vcpu->cpu migration | |
3557 | */ | |
3558 | if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1) | |
0061d53d | 3559 | kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu); |
c285545f | 3560 | if (vcpu->cpu != cpu) |
1bd2009e | 3561 | kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu); |
e48672fa | 3562 | vcpu->cpu = cpu; |
6b7d7e76 | 3563 | } |
c9aaa895 | 3564 | |
c9aaa895 | 3565 | kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu); |
313a3dc7 CO |
3566 | } |
3567 | ||
0b9f6c46 PX |
3568 | static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu) |
3569 | { | |
3570 | if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED)) | |
3571 | return; | |
3572 | ||
fa55eedd | 3573 | vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED; |
0b9f6c46 | 3574 | |
4e335d9e | 3575 | kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime, |
0b9f6c46 PX |
3576 | &vcpu->arch.st.steal.preempted, |
3577 | offsetof(struct kvm_steal_time, preempted), | |
3578 | sizeof(vcpu->arch.st.steal.preempted)); | |
3579 | } | |
3580 | ||
313a3dc7 CO |
3581 | void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) |
3582 | { | |
cc0d907c | 3583 | int idx; |
de63ad4c LM |
3584 | |
3585 | if (vcpu->preempted) | |
3586 | vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu); | |
3587 | ||
931f261b AA |
3588 | /* |
3589 | * Disable page faults because we're in atomic context here. | |
3590 | * kvm_write_guest_offset_cached() would call might_fault() | |
3591 | * that relies on pagefault_disable() to tell if there's a | |
3592 | * bug. NOTE: the write to guest memory may not go through if | |
3593 | * during postcopy live migration or if there's heavy guest | |
3594 | * paging. | |
3595 | */ | |
3596 | pagefault_disable(); | |
cc0d907c AA |
3597 | /* |
3598 | * kvm_memslots() will be called by | |
3599 | * kvm_write_guest_offset_cached() so take the srcu lock. | |
3600 | */ | |
3601 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
0b9f6c46 | 3602 | kvm_steal_time_set_preempted(vcpu); |
cc0d907c | 3603 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
931f261b | 3604 | pagefault_enable(); |
02daab21 | 3605 | kvm_x86_ops->vcpu_put(vcpu); |
4ea1636b | 3606 | vcpu->arch.last_host_tsc = rdtsc(); |
efdab992 | 3607 | /* |
f9dcf08e RK |
3608 | * If userspace has set any breakpoints or watchpoints, dr6 is restored |
3609 | * on every vmexit, but if not, we might have a stale dr6 from the | |
3610 | * guest. do_debug expects dr6 to be cleared after it runs, do the same. | |
efdab992 | 3611 | */ |
f9dcf08e | 3612 | set_debugreg(0, 6); |
313a3dc7 CO |
3613 | } |
3614 | ||
313a3dc7 CO |
3615 | static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu, |
3616 | struct kvm_lapic_state *s) | |
3617 | { | |
fa59cc00 | 3618 | if (vcpu->arch.apicv_active) |
d62caabb AS |
3619 | kvm_x86_ops->sync_pir_to_irr(vcpu); |
3620 | ||
a92e2543 | 3621 | return kvm_apic_get_state(vcpu, s); |
313a3dc7 CO |
3622 | } |
3623 | ||
3624 | static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu, | |
3625 | struct kvm_lapic_state *s) | |
3626 | { | |
a92e2543 RK |
3627 | int r; |
3628 | ||
3629 | r = kvm_apic_set_state(vcpu, s); | |
3630 | if (r) | |
3631 | return r; | |
cb142eb7 | 3632 | update_cr8_intercept(vcpu); |
313a3dc7 CO |
3633 | |
3634 | return 0; | |
3635 | } | |
3636 | ||
127a457a MG |
3637 | static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu) |
3638 | { | |
3639 | return (!lapic_in_kernel(vcpu) || | |
3640 | kvm_apic_accept_pic_intr(vcpu)); | |
3641 | } | |
3642 | ||
782d422b MG |
3643 | /* |
3644 | * if userspace requested an interrupt window, check that the | |
3645 | * interrupt window is open. | |
3646 | * | |
3647 | * No need to exit to userspace if we already have an interrupt queued. | |
3648 | */ | |
3649 | static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu) | |
3650 | { | |
3651 | return kvm_arch_interrupt_allowed(vcpu) && | |
3652 | !kvm_cpu_has_interrupt(vcpu) && | |
3653 | !kvm_event_needs_reinjection(vcpu) && | |
3654 | kvm_cpu_accept_dm_intr(vcpu); | |
3655 | } | |
3656 | ||
f77bc6a4 ZX |
3657 | static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu, |
3658 | struct kvm_interrupt *irq) | |
3659 | { | |
02cdb50f | 3660 | if (irq->irq >= KVM_NR_INTERRUPTS) |
f77bc6a4 | 3661 | return -EINVAL; |
1c1a9ce9 SR |
3662 | |
3663 | if (!irqchip_in_kernel(vcpu->kvm)) { | |
3664 | kvm_queue_interrupt(vcpu, irq->irq, false); | |
3665 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
3666 | return 0; | |
3667 | } | |
3668 | ||
3669 | /* | |
3670 | * With in-kernel LAPIC, we only use this to inject EXTINT, so | |
3671 | * fail for in-kernel 8259. | |
3672 | */ | |
3673 | if (pic_in_kernel(vcpu->kvm)) | |
f77bc6a4 | 3674 | return -ENXIO; |
f77bc6a4 | 3675 | |
1c1a9ce9 SR |
3676 | if (vcpu->arch.pending_external_vector != -1) |
3677 | return -EEXIST; | |
f77bc6a4 | 3678 | |
1c1a9ce9 | 3679 | vcpu->arch.pending_external_vector = irq->irq; |
934bf653 | 3680 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
f77bc6a4 ZX |
3681 | return 0; |
3682 | } | |
3683 | ||
c4abb7c9 JK |
3684 | static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu) |
3685 | { | |
c4abb7c9 | 3686 | kvm_inject_nmi(vcpu); |
c4abb7c9 JK |
3687 | |
3688 | return 0; | |
3689 | } | |
3690 | ||
f077825a PB |
3691 | static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu) |
3692 | { | |
64d60670 PB |
3693 | kvm_make_request(KVM_REQ_SMI, vcpu); |
3694 | ||
f077825a PB |
3695 | return 0; |
3696 | } | |
3697 | ||
b209749f AK |
3698 | static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu, |
3699 | struct kvm_tpr_access_ctl *tac) | |
3700 | { | |
3701 | if (tac->flags) | |
3702 | return -EINVAL; | |
3703 | vcpu->arch.tpr_access_reporting = !!tac->enabled; | |
3704 | return 0; | |
3705 | } | |
3706 | ||
890ca9ae HY |
3707 | static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu, |
3708 | u64 mcg_cap) | |
3709 | { | |
3710 | int r; | |
3711 | unsigned bank_num = mcg_cap & 0xff, bank; | |
3712 | ||
3713 | r = -EINVAL; | |
a9e38c3e | 3714 | if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS) |
890ca9ae | 3715 | goto out; |
c45dcc71 | 3716 | if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000)) |
890ca9ae HY |
3717 | goto out; |
3718 | r = 0; | |
3719 | vcpu->arch.mcg_cap = mcg_cap; | |
3720 | /* Init IA32_MCG_CTL to all 1s */ | |
3721 | if (mcg_cap & MCG_CTL_P) | |
3722 | vcpu->arch.mcg_ctl = ~(u64)0; | |
3723 | /* Init IA32_MCi_CTL to all 1s */ | |
3724 | for (bank = 0; bank < bank_num; bank++) | |
3725 | vcpu->arch.mce_banks[bank*4] = ~(u64)0; | |
c45dcc71 | 3726 | |
92735b1b | 3727 | kvm_x86_ops->setup_mce(vcpu); |
890ca9ae HY |
3728 | out: |
3729 | return r; | |
3730 | } | |
3731 | ||
3732 | static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu, | |
3733 | struct kvm_x86_mce *mce) | |
3734 | { | |
3735 | u64 mcg_cap = vcpu->arch.mcg_cap; | |
3736 | unsigned bank_num = mcg_cap & 0xff; | |
3737 | u64 *banks = vcpu->arch.mce_banks; | |
3738 | ||
3739 | if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL)) | |
3740 | return -EINVAL; | |
3741 | /* | |
3742 | * if IA32_MCG_CTL is not all 1s, the uncorrected error | |
3743 | * reporting is disabled | |
3744 | */ | |
3745 | if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) && | |
3746 | vcpu->arch.mcg_ctl != ~(u64)0) | |
3747 | return 0; | |
3748 | banks += 4 * mce->bank; | |
3749 | /* | |
3750 | * if IA32_MCi_CTL is not all 1s, the uncorrected error | |
3751 | * reporting is disabled for the bank | |
3752 | */ | |
3753 | if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0) | |
3754 | return 0; | |
3755 | if (mce->status & MCI_STATUS_UC) { | |
3756 | if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) || | |
fc78f519 | 3757 | !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) { |
a8eeb04a | 3758 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); |
890ca9ae HY |
3759 | return 0; |
3760 | } | |
3761 | if (banks[1] & MCI_STATUS_VAL) | |
3762 | mce->status |= MCI_STATUS_OVER; | |
3763 | banks[2] = mce->addr; | |
3764 | banks[3] = mce->misc; | |
3765 | vcpu->arch.mcg_status = mce->mcg_status; | |
3766 | banks[1] = mce->status; | |
3767 | kvm_queue_exception(vcpu, MC_VECTOR); | |
3768 | } else if (!(banks[1] & MCI_STATUS_VAL) | |
3769 | || !(banks[1] & MCI_STATUS_UC)) { | |
3770 | if (banks[1] & MCI_STATUS_VAL) | |
3771 | mce->status |= MCI_STATUS_OVER; | |
3772 | banks[2] = mce->addr; | |
3773 | banks[3] = mce->misc; | |
3774 | banks[1] = mce->status; | |
3775 | } else | |
3776 | banks[1] |= MCI_STATUS_OVER; | |
3777 | return 0; | |
3778 | } | |
3779 | ||
3cfc3092 JK |
3780 | static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu, |
3781 | struct kvm_vcpu_events *events) | |
3782 | { | |
7460fb4a | 3783 | process_nmi(vcpu); |
59073aaf | 3784 | |
664f8e26 | 3785 | /* |
59073aaf JM |
3786 | * The API doesn't provide the instruction length for software |
3787 | * exceptions, so don't report them. As long as the guest RIP | |
3788 | * isn't advanced, we should expect to encounter the exception | |
3789 | * again. | |
664f8e26 | 3790 | */ |
59073aaf JM |
3791 | if (kvm_exception_is_soft(vcpu->arch.exception.nr)) { |
3792 | events->exception.injected = 0; | |
3793 | events->exception.pending = 0; | |
3794 | } else { | |
3795 | events->exception.injected = vcpu->arch.exception.injected; | |
3796 | events->exception.pending = vcpu->arch.exception.pending; | |
3797 | /* | |
3798 | * For ABI compatibility, deliberately conflate | |
3799 | * pending and injected exceptions when | |
3800 | * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled. | |
3801 | */ | |
3802 | if (!vcpu->kvm->arch.exception_payload_enabled) | |
3803 | events->exception.injected |= | |
3804 | vcpu->arch.exception.pending; | |
3805 | } | |
3cfc3092 JK |
3806 | events->exception.nr = vcpu->arch.exception.nr; |
3807 | events->exception.has_error_code = vcpu->arch.exception.has_error_code; | |
3808 | events->exception.error_code = vcpu->arch.exception.error_code; | |
59073aaf JM |
3809 | events->exception_has_payload = vcpu->arch.exception.has_payload; |
3810 | events->exception_payload = vcpu->arch.exception.payload; | |
3cfc3092 | 3811 | |
03b82a30 | 3812 | events->interrupt.injected = |
04140b41 | 3813 | vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft; |
3cfc3092 | 3814 | events->interrupt.nr = vcpu->arch.interrupt.nr; |
03b82a30 | 3815 | events->interrupt.soft = 0; |
37ccdcbe | 3816 | events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu); |
3cfc3092 JK |
3817 | |
3818 | events->nmi.injected = vcpu->arch.nmi_injected; | |
7460fb4a | 3819 | events->nmi.pending = vcpu->arch.nmi_pending != 0; |
3cfc3092 | 3820 | events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu); |
97e69aa6 | 3821 | events->nmi.pad = 0; |
3cfc3092 | 3822 | |
66450a21 | 3823 | events->sipi_vector = 0; /* never valid when reporting to user space */ |
3cfc3092 | 3824 | |
f077825a PB |
3825 | events->smi.smm = is_smm(vcpu); |
3826 | events->smi.pending = vcpu->arch.smi_pending; | |
3827 | events->smi.smm_inside_nmi = | |
3828 | !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK); | |
3829 | events->smi.latched_init = kvm_lapic_latched_init(vcpu); | |
3830 | ||
dab4b911 | 3831 | events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING |
f077825a PB |
3832 | | KVM_VCPUEVENT_VALID_SHADOW |
3833 | | KVM_VCPUEVENT_VALID_SMM); | |
59073aaf JM |
3834 | if (vcpu->kvm->arch.exception_payload_enabled) |
3835 | events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD; | |
3836 | ||
97e69aa6 | 3837 | memset(&events->reserved, 0, sizeof(events->reserved)); |
3cfc3092 JK |
3838 | } |
3839 | ||
c5833c7a | 3840 | static void kvm_smm_changed(struct kvm_vcpu *vcpu); |
6ef4e07e | 3841 | |
3cfc3092 JK |
3842 | static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu, |
3843 | struct kvm_vcpu_events *events) | |
3844 | { | |
dab4b911 | 3845 | if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING |
48005f64 | 3846 | | KVM_VCPUEVENT_VALID_SIPI_VECTOR |
f077825a | 3847 | | KVM_VCPUEVENT_VALID_SHADOW |
59073aaf JM |
3848 | | KVM_VCPUEVENT_VALID_SMM |
3849 | | KVM_VCPUEVENT_VALID_PAYLOAD)) | |
3cfc3092 JK |
3850 | return -EINVAL; |
3851 | ||
59073aaf JM |
3852 | if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) { |
3853 | if (!vcpu->kvm->arch.exception_payload_enabled) | |
3854 | return -EINVAL; | |
3855 | if (events->exception.pending) | |
3856 | events->exception.injected = 0; | |
3857 | else | |
3858 | events->exception_has_payload = 0; | |
3859 | } else { | |
3860 | events->exception.pending = 0; | |
3861 | events->exception_has_payload = 0; | |
3862 | } | |
3863 | ||
3864 | if ((events->exception.injected || events->exception.pending) && | |
3865 | (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR)) | |
78e546c8 PB |
3866 | return -EINVAL; |
3867 | ||
28bf2888 DH |
3868 | /* INITs are latched while in SMM */ |
3869 | if (events->flags & KVM_VCPUEVENT_VALID_SMM && | |
3870 | (events->smi.smm || events->smi.pending) && | |
3871 | vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) | |
3872 | return -EINVAL; | |
3873 | ||
7460fb4a | 3874 | process_nmi(vcpu); |
59073aaf JM |
3875 | vcpu->arch.exception.injected = events->exception.injected; |
3876 | vcpu->arch.exception.pending = events->exception.pending; | |
3cfc3092 JK |
3877 | vcpu->arch.exception.nr = events->exception.nr; |
3878 | vcpu->arch.exception.has_error_code = events->exception.has_error_code; | |
3879 | vcpu->arch.exception.error_code = events->exception.error_code; | |
59073aaf JM |
3880 | vcpu->arch.exception.has_payload = events->exception_has_payload; |
3881 | vcpu->arch.exception.payload = events->exception_payload; | |
3cfc3092 | 3882 | |
04140b41 | 3883 | vcpu->arch.interrupt.injected = events->interrupt.injected; |
3cfc3092 JK |
3884 | vcpu->arch.interrupt.nr = events->interrupt.nr; |
3885 | vcpu->arch.interrupt.soft = events->interrupt.soft; | |
48005f64 JK |
3886 | if (events->flags & KVM_VCPUEVENT_VALID_SHADOW) |
3887 | kvm_x86_ops->set_interrupt_shadow(vcpu, | |
3888 | events->interrupt.shadow); | |
3cfc3092 JK |
3889 | |
3890 | vcpu->arch.nmi_injected = events->nmi.injected; | |
dab4b911 JK |
3891 | if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING) |
3892 | vcpu->arch.nmi_pending = events->nmi.pending; | |
3cfc3092 JK |
3893 | kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked); |
3894 | ||
66450a21 | 3895 | if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR && |
bce87cce | 3896 | lapic_in_kernel(vcpu)) |
66450a21 | 3897 | vcpu->arch.apic->sipi_vector = events->sipi_vector; |
3cfc3092 | 3898 | |
f077825a | 3899 | if (events->flags & KVM_VCPUEVENT_VALID_SMM) { |
c5833c7a SC |
3900 | if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) { |
3901 | if (events->smi.smm) | |
3902 | vcpu->arch.hflags |= HF_SMM_MASK; | |
3903 | else | |
3904 | vcpu->arch.hflags &= ~HF_SMM_MASK; | |
3905 | kvm_smm_changed(vcpu); | |
3906 | } | |
6ef4e07e | 3907 | |
f077825a | 3908 | vcpu->arch.smi_pending = events->smi.pending; |
f4ef1910 WL |
3909 | |
3910 | if (events->smi.smm) { | |
3911 | if (events->smi.smm_inside_nmi) | |
3912 | vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK; | |
f077825a | 3913 | else |
f4ef1910 | 3914 | vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK; |
ff90afa7 LA |
3915 | } |
3916 | ||
3917 | if (lapic_in_kernel(vcpu)) { | |
3918 | if (events->smi.latched_init) | |
3919 | set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events); | |
3920 | else | |
3921 | clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events); | |
f077825a PB |
3922 | } |
3923 | } | |
3924 | ||
3842d135 AK |
3925 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
3926 | ||
3cfc3092 JK |
3927 | return 0; |
3928 | } | |
3929 | ||
a1efbe77 JK |
3930 | static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu, |
3931 | struct kvm_debugregs *dbgregs) | |
3932 | { | |
73aaf249 JK |
3933 | unsigned long val; |
3934 | ||
a1efbe77 | 3935 | memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db)); |
16f8a6f9 | 3936 | kvm_get_dr(vcpu, 6, &val); |
73aaf249 | 3937 | dbgregs->dr6 = val; |
a1efbe77 JK |
3938 | dbgregs->dr7 = vcpu->arch.dr7; |
3939 | dbgregs->flags = 0; | |
97e69aa6 | 3940 | memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved)); |
a1efbe77 JK |
3941 | } |
3942 | ||
3943 | static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu, | |
3944 | struct kvm_debugregs *dbgregs) | |
3945 | { | |
3946 | if (dbgregs->flags) | |
3947 | return -EINVAL; | |
3948 | ||
d14bdb55 PB |
3949 | if (dbgregs->dr6 & ~0xffffffffull) |
3950 | return -EINVAL; | |
3951 | if (dbgregs->dr7 & ~0xffffffffull) | |
3952 | return -EINVAL; | |
3953 | ||
a1efbe77 | 3954 | memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db)); |
ae561ede | 3955 | kvm_update_dr0123(vcpu); |
a1efbe77 | 3956 | vcpu->arch.dr6 = dbgregs->dr6; |
73aaf249 | 3957 | kvm_update_dr6(vcpu); |
a1efbe77 | 3958 | vcpu->arch.dr7 = dbgregs->dr7; |
9926c9fd | 3959 | kvm_update_dr7(vcpu); |
a1efbe77 | 3960 | |
a1efbe77 JK |
3961 | return 0; |
3962 | } | |
3963 | ||
df1daba7 PB |
3964 | #define XSTATE_COMPACTION_ENABLED (1ULL << 63) |
3965 | ||
3966 | static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu) | |
3967 | { | |
b666a4b6 | 3968 | struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave; |
400e4b20 | 3969 | u64 xstate_bv = xsave->header.xfeatures; |
df1daba7 PB |
3970 | u64 valid; |
3971 | ||
3972 | /* | |
3973 | * Copy legacy XSAVE area, to avoid complications with CPUID | |
3974 | * leaves 0 and 1 in the loop below. | |
3975 | */ | |
3976 | memcpy(dest, xsave, XSAVE_HDR_OFFSET); | |
3977 | ||
3978 | /* Set XSTATE_BV */ | |
00c87e9a | 3979 | xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE; |
df1daba7 PB |
3980 | *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv; |
3981 | ||
3982 | /* | |
3983 | * Copy each region from the possibly compacted offset to the | |
3984 | * non-compacted offset. | |
3985 | */ | |
d91cab78 | 3986 | valid = xstate_bv & ~XFEATURE_MASK_FPSSE; |
df1daba7 | 3987 | while (valid) { |
abd16d68 SAS |
3988 | u64 xfeature_mask = valid & -valid; |
3989 | int xfeature_nr = fls64(xfeature_mask) - 1; | |
3990 | void *src = get_xsave_addr(xsave, xfeature_nr); | |
df1daba7 PB |
3991 | |
3992 | if (src) { | |
3993 | u32 size, offset, ecx, edx; | |
abd16d68 | 3994 | cpuid_count(XSTATE_CPUID, xfeature_nr, |
df1daba7 | 3995 | &size, &offset, &ecx, &edx); |
abd16d68 | 3996 | if (xfeature_nr == XFEATURE_PKRU) |
38cfd5e3 PB |
3997 | memcpy(dest + offset, &vcpu->arch.pkru, |
3998 | sizeof(vcpu->arch.pkru)); | |
3999 | else | |
4000 | memcpy(dest + offset, src, size); | |
4001 | ||
df1daba7 PB |
4002 | } |
4003 | ||
abd16d68 | 4004 | valid -= xfeature_mask; |
df1daba7 PB |
4005 | } |
4006 | } | |
4007 | ||
4008 | static void load_xsave(struct kvm_vcpu *vcpu, u8 *src) | |
4009 | { | |
b666a4b6 | 4010 | struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave; |
df1daba7 PB |
4011 | u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET); |
4012 | u64 valid; | |
4013 | ||
4014 | /* | |
4015 | * Copy legacy XSAVE area, to avoid complications with CPUID | |
4016 | * leaves 0 and 1 in the loop below. | |
4017 | */ | |
4018 | memcpy(xsave, src, XSAVE_HDR_OFFSET); | |
4019 | ||
4020 | /* Set XSTATE_BV and possibly XCOMP_BV. */ | |
400e4b20 | 4021 | xsave->header.xfeatures = xstate_bv; |
782511b0 | 4022 | if (boot_cpu_has(X86_FEATURE_XSAVES)) |
3a54450b | 4023 | xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED; |
df1daba7 PB |
4024 | |
4025 | /* | |
4026 | * Copy each region from the non-compacted offset to the | |
4027 | * possibly compacted offset. | |
4028 | */ | |
d91cab78 | 4029 | valid = xstate_bv & ~XFEATURE_MASK_FPSSE; |
df1daba7 | 4030 | while (valid) { |
abd16d68 SAS |
4031 | u64 xfeature_mask = valid & -valid; |
4032 | int xfeature_nr = fls64(xfeature_mask) - 1; | |
4033 | void *dest = get_xsave_addr(xsave, xfeature_nr); | |
df1daba7 PB |
4034 | |
4035 | if (dest) { | |
4036 | u32 size, offset, ecx, edx; | |
abd16d68 | 4037 | cpuid_count(XSTATE_CPUID, xfeature_nr, |
df1daba7 | 4038 | &size, &offset, &ecx, &edx); |
abd16d68 | 4039 | if (xfeature_nr == XFEATURE_PKRU) |
38cfd5e3 PB |
4040 | memcpy(&vcpu->arch.pkru, src + offset, |
4041 | sizeof(vcpu->arch.pkru)); | |
4042 | else | |
4043 | memcpy(dest, src + offset, size); | |
ee4100da | 4044 | } |
df1daba7 | 4045 | |
abd16d68 | 4046 | valid -= xfeature_mask; |
df1daba7 PB |
4047 | } |
4048 | } | |
4049 | ||
2d5b5a66 SY |
4050 | static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu, |
4051 | struct kvm_xsave *guest_xsave) | |
4052 | { | |
d366bf7e | 4053 | if (boot_cpu_has(X86_FEATURE_XSAVE)) { |
df1daba7 PB |
4054 | memset(guest_xsave, 0, sizeof(struct kvm_xsave)); |
4055 | fill_xsave((u8 *) guest_xsave->region, vcpu); | |
4344ee98 | 4056 | } else { |
2d5b5a66 | 4057 | memcpy(guest_xsave->region, |
b666a4b6 | 4058 | &vcpu->arch.guest_fpu->state.fxsave, |
c47ada30 | 4059 | sizeof(struct fxregs_state)); |
2d5b5a66 | 4060 | *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] = |
d91cab78 | 4061 | XFEATURE_MASK_FPSSE; |
2d5b5a66 SY |
4062 | } |
4063 | } | |
4064 | ||
a575813b WL |
4065 | #define XSAVE_MXCSR_OFFSET 24 |
4066 | ||
2d5b5a66 SY |
4067 | static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu, |
4068 | struct kvm_xsave *guest_xsave) | |
4069 | { | |
4070 | u64 xstate_bv = | |
4071 | *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)]; | |
a575813b | 4072 | u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)]; |
2d5b5a66 | 4073 | |
d366bf7e | 4074 | if (boot_cpu_has(X86_FEATURE_XSAVE)) { |
d7876f1b PB |
4075 | /* |
4076 | * Here we allow setting states that are not present in | |
4077 | * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility | |
4078 | * with old userspace. | |
4079 | */ | |
a575813b WL |
4080 | if (xstate_bv & ~kvm_supported_xcr0() || |
4081 | mxcsr & ~mxcsr_feature_mask) | |
d7876f1b | 4082 | return -EINVAL; |
df1daba7 | 4083 | load_xsave(vcpu, (u8 *)guest_xsave->region); |
d7876f1b | 4084 | } else { |
a575813b WL |
4085 | if (xstate_bv & ~XFEATURE_MASK_FPSSE || |
4086 | mxcsr & ~mxcsr_feature_mask) | |
2d5b5a66 | 4087 | return -EINVAL; |
b666a4b6 | 4088 | memcpy(&vcpu->arch.guest_fpu->state.fxsave, |
c47ada30 | 4089 | guest_xsave->region, sizeof(struct fxregs_state)); |
2d5b5a66 SY |
4090 | } |
4091 | return 0; | |
4092 | } | |
4093 | ||
4094 | static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu, | |
4095 | struct kvm_xcrs *guest_xcrs) | |
4096 | { | |
d366bf7e | 4097 | if (!boot_cpu_has(X86_FEATURE_XSAVE)) { |
2d5b5a66 SY |
4098 | guest_xcrs->nr_xcrs = 0; |
4099 | return; | |
4100 | } | |
4101 | ||
4102 | guest_xcrs->nr_xcrs = 1; | |
4103 | guest_xcrs->flags = 0; | |
4104 | guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK; | |
4105 | guest_xcrs->xcrs[0].value = vcpu->arch.xcr0; | |
4106 | } | |
4107 | ||
4108 | static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu, | |
4109 | struct kvm_xcrs *guest_xcrs) | |
4110 | { | |
4111 | int i, r = 0; | |
4112 | ||
d366bf7e | 4113 | if (!boot_cpu_has(X86_FEATURE_XSAVE)) |
2d5b5a66 SY |
4114 | return -EINVAL; |
4115 | ||
4116 | if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags) | |
4117 | return -EINVAL; | |
4118 | ||
4119 | for (i = 0; i < guest_xcrs->nr_xcrs; i++) | |
4120 | /* Only support XCR0 currently */ | |
c67a04cb | 4121 | if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) { |
2d5b5a66 | 4122 | r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK, |
c67a04cb | 4123 | guest_xcrs->xcrs[i].value); |
2d5b5a66 SY |
4124 | break; |
4125 | } | |
4126 | if (r) | |
4127 | r = -EINVAL; | |
4128 | return r; | |
4129 | } | |
4130 | ||
1c0b28c2 EM |
4131 | /* |
4132 | * kvm_set_guest_paused() indicates to the guest kernel that it has been | |
4133 | * stopped by the hypervisor. This function will be called from the host only. | |
4134 | * EINVAL is returned when the host attempts to set the flag for a guest that | |
4135 | * does not support pv clocks. | |
4136 | */ | |
4137 | static int kvm_set_guest_paused(struct kvm_vcpu *vcpu) | |
4138 | { | |
0b79459b | 4139 | if (!vcpu->arch.pv_time_enabled) |
1c0b28c2 | 4140 | return -EINVAL; |
51d59c6b | 4141 | vcpu->arch.pvclock_set_guest_stopped_request = true; |
1c0b28c2 EM |
4142 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
4143 | return 0; | |
4144 | } | |
4145 | ||
5c919412 AS |
4146 | static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu, |
4147 | struct kvm_enable_cap *cap) | |
4148 | { | |
57b119da VK |
4149 | int r; |
4150 | uint16_t vmcs_version; | |
4151 | void __user *user_ptr; | |
4152 | ||
5c919412 AS |
4153 | if (cap->flags) |
4154 | return -EINVAL; | |
4155 | ||
4156 | switch (cap->cap) { | |
efc479e6 RK |
4157 | case KVM_CAP_HYPERV_SYNIC2: |
4158 | if (cap->args[0]) | |
4159 | return -EINVAL; | |
b2869f28 GS |
4160 | /* fall through */ |
4161 | ||
5c919412 | 4162 | case KVM_CAP_HYPERV_SYNIC: |
546d87e5 WL |
4163 | if (!irqchip_in_kernel(vcpu->kvm)) |
4164 | return -EINVAL; | |
efc479e6 RK |
4165 | return kvm_hv_activate_synic(vcpu, cap->cap == |
4166 | KVM_CAP_HYPERV_SYNIC2); | |
57b119da | 4167 | case KVM_CAP_HYPERV_ENLIGHTENED_VMCS: |
5158917c SC |
4168 | if (!kvm_x86_ops->nested_enable_evmcs) |
4169 | return -ENOTTY; | |
57b119da VK |
4170 | r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version); |
4171 | if (!r) { | |
4172 | user_ptr = (void __user *)(uintptr_t)cap->args[0]; | |
4173 | if (copy_to_user(user_ptr, &vmcs_version, | |
4174 | sizeof(vmcs_version))) | |
4175 | r = -EFAULT; | |
4176 | } | |
4177 | return r; | |
344c6c80 TL |
4178 | case KVM_CAP_HYPERV_DIRECT_TLBFLUSH: |
4179 | if (!kvm_x86_ops->enable_direct_tlbflush) | |
4180 | return -ENOTTY; | |
4181 | ||
4182 | return kvm_x86_ops->enable_direct_tlbflush(vcpu); | |
57b119da | 4183 | |
5c919412 AS |
4184 | default: |
4185 | return -EINVAL; | |
4186 | } | |
4187 | } | |
4188 | ||
313a3dc7 CO |
4189 | long kvm_arch_vcpu_ioctl(struct file *filp, |
4190 | unsigned int ioctl, unsigned long arg) | |
4191 | { | |
4192 | struct kvm_vcpu *vcpu = filp->private_data; | |
4193 | void __user *argp = (void __user *)arg; | |
4194 | int r; | |
d1ac91d8 AK |
4195 | union { |
4196 | struct kvm_lapic_state *lapic; | |
4197 | struct kvm_xsave *xsave; | |
4198 | struct kvm_xcrs *xcrs; | |
4199 | void *buffer; | |
4200 | } u; | |
4201 | ||
9b062471 CD |
4202 | vcpu_load(vcpu); |
4203 | ||
d1ac91d8 | 4204 | u.buffer = NULL; |
313a3dc7 CO |
4205 | switch (ioctl) { |
4206 | case KVM_GET_LAPIC: { | |
2204ae3c | 4207 | r = -EINVAL; |
bce87cce | 4208 | if (!lapic_in_kernel(vcpu)) |
2204ae3c | 4209 | goto out; |
254272ce BG |
4210 | u.lapic = kzalloc(sizeof(struct kvm_lapic_state), |
4211 | GFP_KERNEL_ACCOUNT); | |
313a3dc7 | 4212 | |
b772ff36 | 4213 | r = -ENOMEM; |
d1ac91d8 | 4214 | if (!u.lapic) |
b772ff36 | 4215 | goto out; |
d1ac91d8 | 4216 | r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic); |
313a3dc7 CO |
4217 | if (r) |
4218 | goto out; | |
4219 | r = -EFAULT; | |
d1ac91d8 | 4220 | if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state))) |
313a3dc7 CO |
4221 | goto out; |
4222 | r = 0; | |
4223 | break; | |
4224 | } | |
4225 | case KVM_SET_LAPIC: { | |
2204ae3c | 4226 | r = -EINVAL; |
bce87cce | 4227 | if (!lapic_in_kernel(vcpu)) |
2204ae3c | 4228 | goto out; |
ff5c2c03 | 4229 | u.lapic = memdup_user(argp, sizeof(*u.lapic)); |
9b062471 CD |
4230 | if (IS_ERR(u.lapic)) { |
4231 | r = PTR_ERR(u.lapic); | |
4232 | goto out_nofree; | |
4233 | } | |
ff5c2c03 | 4234 | |
d1ac91d8 | 4235 | r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic); |
313a3dc7 CO |
4236 | break; |
4237 | } | |
f77bc6a4 ZX |
4238 | case KVM_INTERRUPT: { |
4239 | struct kvm_interrupt irq; | |
4240 | ||
4241 | r = -EFAULT; | |
0e96f31e | 4242 | if (copy_from_user(&irq, argp, sizeof(irq))) |
f77bc6a4 ZX |
4243 | goto out; |
4244 | r = kvm_vcpu_ioctl_interrupt(vcpu, &irq); | |
f77bc6a4 ZX |
4245 | break; |
4246 | } | |
c4abb7c9 JK |
4247 | case KVM_NMI: { |
4248 | r = kvm_vcpu_ioctl_nmi(vcpu); | |
c4abb7c9 JK |
4249 | break; |
4250 | } | |
f077825a PB |
4251 | case KVM_SMI: { |
4252 | r = kvm_vcpu_ioctl_smi(vcpu); | |
4253 | break; | |
4254 | } | |
313a3dc7 CO |
4255 | case KVM_SET_CPUID: { |
4256 | struct kvm_cpuid __user *cpuid_arg = argp; | |
4257 | struct kvm_cpuid cpuid; | |
4258 | ||
4259 | r = -EFAULT; | |
0e96f31e | 4260 | if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid))) |
313a3dc7 CO |
4261 | goto out; |
4262 | r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries); | |
313a3dc7 CO |
4263 | break; |
4264 | } | |
07716717 DK |
4265 | case KVM_SET_CPUID2: { |
4266 | struct kvm_cpuid2 __user *cpuid_arg = argp; | |
4267 | struct kvm_cpuid2 cpuid; | |
4268 | ||
4269 | r = -EFAULT; | |
0e96f31e | 4270 | if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid))) |
07716717 DK |
4271 | goto out; |
4272 | r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid, | |
19355475 | 4273 | cpuid_arg->entries); |
07716717 DK |
4274 | break; |
4275 | } | |
4276 | case KVM_GET_CPUID2: { | |
4277 | struct kvm_cpuid2 __user *cpuid_arg = argp; | |
4278 | struct kvm_cpuid2 cpuid; | |
4279 | ||
4280 | r = -EFAULT; | |
0e96f31e | 4281 | if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid))) |
07716717 DK |
4282 | goto out; |
4283 | r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid, | |
19355475 | 4284 | cpuid_arg->entries); |
07716717 DK |
4285 | if (r) |
4286 | goto out; | |
4287 | r = -EFAULT; | |
0e96f31e | 4288 | if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid))) |
07716717 DK |
4289 | goto out; |
4290 | r = 0; | |
4291 | break; | |
4292 | } | |
801e459a TL |
4293 | case KVM_GET_MSRS: { |
4294 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
609e36d3 | 4295 | r = msr_io(vcpu, argp, do_get_msr, 1); |
801e459a | 4296 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
313a3dc7 | 4297 | break; |
801e459a TL |
4298 | } |
4299 | case KVM_SET_MSRS: { | |
4300 | int idx = srcu_read_lock(&vcpu->kvm->srcu); | |
313a3dc7 | 4301 | r = msr_io(vcpu, argp, do_set_msr, 0); |
801e459a | 4302 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
313a3dc7 | 4303 | break; |
801e459a | 4304 | } |
b209749f AK |
4305 | case KVM_TPR_ACCESS_REPORTING: { |
4306 | struct kvm_tpr_access_ctl tac; | |
4307 | ||
4308 | r = -EFAULT; | |
0e96f31e | 4309 | if (copy_from_user(&tac, argp, sizeof(tac))) |
b209749f AK |
4310 | goto out; |
4311 | r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac); | |
4312 | if (r) | |
4313 | goto out; | |
4314 | r = -EFAULT; | |
0e96f31e | 4315 | if (copy_to_user(argp, &tac, sizeof(tac))) |
b209749f AK |
4316 | goto out; |
4317 | r = 0; | |
4318 | break; | |
4319 | }; | |
b93463aa AK |
4320 | case KVM_SET_VAPIC_ADDR: { |
4321 | struct kvm_vapic_addr va; | |
7301d6ab | 4322 | int idx; |
b93463aa AK |
4323 | |
4324 | r = -EINVAL; | |
35754c98 | 4325 | if (!lapic_in_kernel(vcpu)) |
b93463aa AK |
4326 | goto out; |
4327 | r = -EFAULT; | |
0e96f31e | 4328 | if (copy_from_user(&va, argp, sizeof(va))) |
b93463aa | 4329 | goto out; |
7301d6ab | 4330 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
fda4e2e8 | 4331 | r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr); |
7301d6ab | 4332 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
b93463aa AK |
4333 | break; |
4334 | } | |
890ca9ae HY |
4335 | case KVM_X86_SETUP_MCE: { |
4336 | u64 mcg_cap; | |
4337 | ||
4338 | r = -EFAULT; | |
0e96f31e | 4339 | if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap))) |
890ca9ae HY |
4340 | goto out; |
4341 | r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap); | |
4342 | break; | |
4343 | } | |
4344 | case KVM_X86_SET_MCE: { | |
4345 | struct kvm_x86_mce mce; | |
4346 | ||
4347 | r = -EFAULT; | |
0e96f31e | 4348 | if (copy_from_user(&mce, argp, sizeof(mce))) |
890ca9ae HY |
4349 | goto out; |
4350 | r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce); | |
4351 | break; | |
4352 | } | |
3cfc3092 JK |
4353 | case KVM_GET_VCPU_EVENTS: { |
4354 | struct kvm_vcpu_events events; | |
4355 | ||
4356 | kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events); | |
4357 | ||
4358 | r = -EFAULT; | |
4359 | if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events))) | |
4360 | break; | |
4361 | r = 0; | |
4362 | break; | |
4363 | } | |
4364 | case KVM_SET_VCPU_EVENTS: { | |
4365 | struct kvm_vcpu_events events; | |
4366 | ||
4367 | r = -EFAULT; | |
4368 | if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events))) | |
4369 | break; | |
4370 | ||
4371 | r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events); | |
4372 | break; | |
4373 | } | |
a1efbe77 JK |
4374 | case KVM_GET_DEBUGREGS: { |
4375 | struct kvm_debugregs dbgregs; | |
4376 | ||
4377 | kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs); | |
4378 | ||
4379 | r = -EFAULT; | |
4380 | if (copy_to_user(argp, &dbgregs, | |
4381 | sizeof(struct kvm_debugregs))) | |
4382 | break; | |
4383 | r = 0; | |
4384 | break; | |
4385 | } | |
4386 | case KVM_SET_DEBUGREGS: { | |
4387 | struct kvm_debugregs dbgregs; | |
4388 | ||
4389 | r = -EFAULT; | |
4390 | if (copy_from_user(&dbgregs, argp, | |
4391 | sizeof(struct kvm_debugregs))) | |
4392 | break; | |
4393 | ||
4394 | r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs); | |
4395 | break; | |
4396 | } | |
2d5b5a66 | 4397 | case KVM_GET_XSAVE: { |
254272ce | 4398 | u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT); |
2d5b5a66 | 4399 | r = -ENOMEM; |
d1ac91d8 | 4400 | if (!u.xsave) |
2d5b5a66 SY |
4401 | break; |
4402 | ||
d1ac91d8 | 4403 | kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave); |
2d5b5a66 SY |
4404 | |
4405 | r = -EFAULT; | |
d1ac91d8 | 4406 | if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave))) |
2d5b5a66 SY |
4407 | break; |
4408 | r = 0; | |
4409 | break; | |
4410 | } | |
4411 | case KVM_SET_XSAVE: { | |
ff5c2c03 | 4412 | u.xsave = memdup_user(argp, sizeof(*u.xsave)); |
9b062471 CD |
4413 | if (IS_ERR(u.xsave)) { |
4414 | r = PTR_ERR(u.xsave); | |
4415 | goto out_nofree; | |
4416 | } | |
2d5b5a66 | 4417 | |
d1ac91d8 | 4418 | r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave); |
2d5b5a66 SY |
4419 | break; |
4420 | } | |
4421 | case KVM_GET_XCRS: { | |
254272ce | 4422 | u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT); |
2d5b5a66 | 4423 | r = -ENOMEM; |
d1ac91d8 | 4424 | if (!u.xcrs) |
2d5b5a66 SY |
4425 | break; |
4426 | ||
d1ac91d8 | 4427 | kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs); |
2d5b5a66 SY |
4428 | |
4429 | r = -EFAULT; | |
d1ac91d8 | 4430 | if (copy_to_user(argp, u.xcrs, |
2d5b5a66 SY |
4431 | sizeof(struct kvm_xcrs))) |
4432 | break; | |
4433 | r = 0; | |
4434 | break; | |
4435 | } | |
4436 | case KVM_SET_XCRS: { | |
ff5c2c03 | 4437 | u.xcrs = memdup_user(argp, sizeof(*u.xcrs)); |
9b062471 CD |
4438 | if (IS_ERR(u.xcrs)) { |
4439 | r = PTR_ERR(u.xcrs); | |
4440 | goto out_nofree; | |
4441 | } | |
2d5b5a66 | 4442 | |
d1ac91d8 | 4443 | r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs); |
2d5b5a66 SY |
4444 | break; |
4445 | } | |
92a1f12d JR |
4446 | case KVM_SET_TSC_KHZ: { |
4447 | u32 user_tsc_khz; | |
4448 | ||
4449 | r = -EINVAL; | |
92a1f12d JR |
4450 | user_tsc_khz = (u32)arg; |
4451 | ||
4452 | if (user_tsc_khz >= kvm_max_guest_tsc_khz) | |
4453 | goto out; | |
4454 | ||
cc578287 ZA |
4455 | if (user_tsc_khz == 0) |
4456 | user_tsc_khz = tsc_khz; | |
4457 | ||
381d585c HZ |
4458 | if (!kvm_set_tsc_khz(vcpu, user_tsc_khz)) |
4459 | r = 0; | |
92a1f12d | 4460 | |
92a1f12d JR |
4461 | goto out; |
4462 | } | |
4463 | case KVM_GET_TSC_KHZ: { | |
cc578287 | 4464 | r = vcpu->arch.virtual_tsc_khz; |
92a1f12d JR |
4465 | goto out; |
4466 | } | |
1c0b28c2 EM |
4467 | case KVM_KVMCLOCK_CTRL: { |
4468 | r = kvm_set_guest_paused(vcpu); | |
4469 | goto out; | |
4470 | } | |
5c919412 AS |
4471 | case KVM_ENABLE_CAP: { |
4472 | struct kvm_enable_cap cap; | |
4473 | ||
4474 | r = -EFAULT; | |
4475 | if (copy_from_user(&cap, argp, sizeof(cap))) | |
4476 | goto out; | |
4477 | r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap); | |
4478 | break; | |
4479 | } | |
8fcc4b59 JM |
4480 | case KVM_GET_NESTED_STATE: { |
4481 | struct kvm_nested_state __user *user_kvm_nested_state = argp; | |
4482 | u32 user_data_size; | |
4483 | ||
4484 | r = -EINVAL; | |
4485 | if (!kvm_x86_ops->get_nested_state) | |
4486 | break; | |
4487 | ||
4488 | BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size)); | |
26b471c7 | 4489 | r = -EFAULT; |
8fcc4b59 | 4490 | if (get_user(user_data_size, &user_kvm_nested_state->size)) |
26b471c7 | 4491 | break; |
8fcc4b59 JM |
4492 | |
4493 | r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state, | |
4494 | user_data_size); | |
4495 | if (r < 0) | |
26b471c7 | 4496 | break; |
8fcc4b59 JM |
4497 | |
4498 | if (r > user_data_size) { | |
4499 | if (put_user(r, &user_kvm_nested_state->size)) | |
26b471c7 LA |
4500 | r = -EFAULT; |
4501 | else | |
4502 | r = -E2BIG; | |
4503 | break; | |
8fcc4b59 | 4504 | } |
26b471c7 | 4505 | |
8fcc4b59 JM |
4506 | r = 0; |
4507 | break; | |
4508 | } | |
4509 | case KVM_SET_NESTED_STATE: { | |
4510 | struct kvm_nested_state __user *user_kvm_nested_state = argp; | |
4511 | struct kvm_nested_state kvm_state; | |
ad5996d9 | 4512 | int idx; |
8fcc4b59 JM |
4513 | |
4514 | r = -EINVAL; | |
4515 | if (!kvm_x86_ops->set_nested_state) | |
4516 | break; | |
4517 | ||
26b471c7 | 4518 | r = -EFAULT; |
8fcc4b59 | 4519 | if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state))) |
26b471c7 | 4520 | break; |
8fcc4b59 | 4521 | |
26b471c7 | 4522 | r = -EINVAL; |
8fcc4b59 | 4523 | if (kvm_state.size < sizeof(kvm_state)) |
26b471c7 | 4524 | break; |
8fcc4b59 JM |
4525 | |
4526 | if (kvm_state.flags & | |
8cab6507 VK |
4527 | ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE |
4528 | | KVM_STATE_NESTED_EVMCS)) | |
26b471c7 | 4529 | break; |
8fcc4b59 JM |
4530 | |
4531 | /* nested_run_pending implies guest_mode. */ | |
8cab6507 VK |
4532 | if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING) |
4533 | && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE)) | |
26b471c7 | 4534 | break; |
8fcc4b59 | 4535 | |
ad5996d9 | 4536 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
8fcc4b59 | 4537 | r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state); |
ad5996d9 | 4538 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
8fcc4b59 JM |
4539 | break; |
4540 | } | |
2bc39970 VK |
4541 | case KVM_GET_SUPPORTED_HV_CPUID: { |
4542 | struct kvm_cpuid2 __user *cpuid_arg = argp; | |
4543 | struct kvm_cpuid2 cpuid; | |
4544 | ||
4545 | r = -EFAULT; | |
4546 | if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid))) | |
4547 | goto out; | |
4548 | ||
4549 | r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid, | |
4550 | cpuid_arg->entries); | |
4551 | if (r) | |
4552 | goto out; | |
4553 | ||
4554 | r = -EFAULT; | |
4555 | if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid))) | |
4556 | goto out; | |
4557 | r = 0; | |
4558 | break; | |
4559 | } | |
313a3dc7 CO |
4560 | default: |
4561 | r = -EINVAL; | |
4562 | } | |
4563 | out: | |
d1ac91d8 | 4564 | kfree(u.buffer); |
9b062471 CD |
4565 | out_nofree: |
4566 | vcpu_put(vcpu); | |
313a3dc7 CO |
4567 | return r; |
4568 | } | |
4569 | ||
1499fa80 | 4570 | vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) |
5b1c1493 CO |
4571 | { |
4572 | return VM_FAULT_SIGBUS; | |
4573 | } | |
4574 | ||
1fe779f8 CO |
4575 | static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr) |
4576 | { | |
4577 | int ret; | |
4578 | ||
4579 | if (addr > (unsigned int)(-3 * PAGE_SIZE)) | |
951179ce | 4580 | return -EINVAL; |
1fe779f8 CO |
4581 | ret = kvm_x86_ops->set_tss_addr(kvm, addr); |
4582 | return ret; | |
4583 | } | |
4584 | ||
b927a3ce SY |
4585 | static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm, |
4586 | u64 ident_addr) | |
4587 | { | |
2ac52ab8 | 4588 | return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr); |
b927a3ce SY |
4589 | } |
4590 | ||
1fe779f8 | 4591 | static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm, |
bc8a3d89 | 4592 | unsigned long kvm_nr_mmu_pages) |
1fe779f8 CO |
4593 | { |
4594 | if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES) | |
4595 | return -EINVAL; | |
4596 | ||
79fac95e | 4597 | mutex_lock(&kvm->slots_lock); |
1fe779f8 CO |
4598 | |
4599 | kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages); | |
f05e70ac | 4600 | kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages; |
1fe779f8 | 4601 | |
79fac95e | 4602 | mutex_unlock(&kvm->slots_lock); |
1fe779f8 CO |
4603 | return 0; |
4604 | } | |
4605 | ||
bc8a3d89 | 4606 | static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm) |
1fe779f8 | 4607 | { |
39de71ec | 4608 | return kvm->arch.n_max_mmu_pages; |
1fe779f8 CO |
4609 | } |
4610 | ||
1fe779f8 CO |
4611 | static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip) |
4612 | { | |
90bca052 | 4613 | struct kvm_pic *pic = kvm->arch.vpic; |
1fe779f8 CO |
4614 | int r; |
4615 | ||
4616 | r = 0; | |
4617 | switch (chip->chip_id) { | |
4618 | case KVM_IRQCHIP_PIC_MASTER: | |
90bca052 | 4619 | memcpy(&chip->chip.pic, &pic->pics[0], |
1fe779f8 CO |
4620 | sizeof(struct kvm_pic_state)); |
4621 | break; | |
4622 | case KVM_IRQCHIP_PIC_SLAVE: | |
90bca052 | 4623 | memcpy(&chip->chip.pic, &pic->pics[1], |
1fe779f8 CO |
4624 | sizeof(struct kvm_pic_state)); |
4625 | break; | |
4626 | case KVM_IRQCHIP_IOAPIC: | |
33392b49 | 4627 | kvm_get_ioapic(kvm, &chip->chip.ioapic); |
1fe779f8 CO |
4628 | break; |
4629 | default: | |
4630 | r = -EINVAL; | |
4631 | break; | |
4632 | } | |
4633 | return r; | |
4634 | } | |
4635 | ||
4636 | static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip) | |
4637 | { | |
90bca052 | 4638 | struct kvm_pic *pic = kvm->arch.vpic; |
1fe779f8 CO |
4639 | int r; |
4640 | ||
4641 | r = 0; | |
4642 | switch (chip->chip_id) { | |
4643 | case KVM_IRQCHIP_PIC_MASTER: | |
90bca052 DH |
4644 | spin_lock(&pic->lock); |
4645 | memcpy(&pic->pics[0], &chip->chip.pic, | |
1fe779f8 | 4646 | sizeof(struct kvm_pic_state)); |
90bca052 | 4647 | spin_unlock(&pic->lock); |
1fe779f8 CO |
4648 | break; |
4649 | case KVM_IRQCHIP_PIC_SLAVE: | |
90bca052 DH |
4650 | spin_lock(&pic->lock); |
4651 | memcpy(&pic->pics[1], &chip->chip.pic, | |
1fe779f8 | 4652 | sizeof(struct kvm_pic_state)); |
90bca052 | 4653 | spin_unlock(&pic->lock); |
1fe779f8 CO |
4654 | break; |
4655 | case KVM_IRQCHIP_IOAPIC: | |
33392b49 | 4656 | kvm_set_ioapic(kvm, &chip->chip.ioapic); |
1fe779f8 CO |
4657 | break; |
4658 | default: | |
4659 | r = -EINVAL; | |
4660 | break; | |
4661 | } | |
90bca052 | 4662 | kvm_pic_update_irq(pic); |
1fe779f8 CO |
4663 | return r; |
4664 | } | |
4665 | ||
e0f63cb9 SY |
4666 | static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps) |
4667 | { | |
34f3941c RK |
4668 | struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state; |
4669 | ||
4670 | BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels)); | |
4671 | ||
4672 | mutex_lock(&kps->lock); | |
4673 | memcpy(ps, &kps->channels, sizeof(*ps)); | |
4674 | mutex_unlock(&kps->lock); | |
2da29bcc | 4675 | return 0; |
e0f63cb9 SY |
4676 | } |
4677 | ||
4678 | static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps) | |
4679 | { | |
0185604c | 4680 | int i; |
09edea72 RK |
4681 | struct kvm_pit *pit = kvm->arch.vpit; |
4682 | ||
4683 | mutex_lock(&pit->pit_state.lock); | |
34f3941c | 4684 | memcpy(&pit->pit_state.channels, ps, sizeof(*ps)); |
0185604c | 4685 | for (i = 0; i < 3; i++) |
09edea72 RK |
4686 | kvm_pit_load_count(pit, i, ps->channels[i].count, 0); |
4687 | mutex_unlock(&pit->pit_state.lock); | |
2da29bcc | 4688 | return 0; |
e9f42757 BK |
4689 | } |
4690 | ||
4691 | static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps) | |
4692 | { | |
e9f42757 BK |
4693 | mutex_lock(&kvm->arch.vpit->pit_state.lock); |
4694 | memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels, | |
4695 | sizeof(ps->channels)); | |
4696 | ps->flags = kvm->arch.vpit->pit_state.flags; | |
4697 | mutex_unlock(&kvm->arch.vpit->pit_state.lock); | |
97e69aa6 | 4698 | memset(&ps->reserved, 0, sizeof(ps->reserved)); |
2da29bcc | 4699 | return 0; |
e9f42757 BK |
4700 | } |
4701 | ||
4702 | static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps) | |
4703 | { | |
2da29bcc | 4704 | int start = 0; |
0185604c | 4705 | int i; |
e9f42757 | 4706 | u32 prev_legacy, cur_legacy; |
09edea72 RK |
4707 | struct kvm_pit *pit = kvm->arch.vpit; |
4708 | ||
4709 | mutex_lock(&pit->pit_state.lock); | |
4710 | prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY; | |
e9f42757 BK |
4711 | cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY; |
4712 | if (!prev_legacy && cur_legacy) | |
4713 | start = 1; | |
09edea72 RK |
4714 | memcpy(&pit->pit_state.channels, &ps->channels, |
4715 | sizeof(pit->pit_state.channels)); | |
4716 | pit->pit_state.flags = ps->flags; | |
0185604c | 4717 | for (i = 0; i < 3; i++) |
09edea72 | 4718 | kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count, |
e5e57e7a | 4719 | start && i == 0); |
09edea72 | 4720 | mutex_unlock(&pit->pit_state.lock); |
2da29bcc | 4721 | return 0; |
e0f63cb9 SY |
4722 | } |
4723 | ||
52d939a0 MT |
4724 | static int kvm_vm_ioctl_reinject(struct kvm *kvm, |
4725 | struct kvm_reinject_control *control) | |
4726 | { | |
71474e2f RK |
4727 | struct kvm_pit *pit = kvm->arch.vpit; |
4728 | ||
71474e2f RK |
4729 | /* pit->pit_state.lock was overloaded to prevent userspace from getting |
4730 | * an inconsistent state after running multiple KVM_REINJECT_CONTROL | |
4731 | * ioctls in parallel. Use a separate lock if that ioctl isn't rare. | |
4732 | */ | |
4733 | mutex_lock(&pit->pit_state.lock); | |
4734 | kvm_pit_set_reinject(pit, control->pit_reinject); | |
4735 | mutex_unlock(&pit->pit_state.lock); | |
b39c90b6 | 4736 | |
52d939a0 MT |
4737 | return 0; |
4738 | } | |
4739 | ||
95d4c16c | 4740 | /** |
60c34612 TY |
4741 | * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot |
4742 | * @kvm: kvm instance | |
4743 | * @log: slot id and address to which we copy the log | |
95d4c16c | 4744 | * |
e108ff2f PB |
4745 | * Steps 1-4 below provide general overview of dirty page logging. See |
4746 | * kvm_get_dirty_log_protect() function description for additional details. | |
4747 | * | |
4748 | * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we | |
4749 | * always flush the TLB (step 4) even if previous step failed and the dirty | |
4750 | * bitmap may be corrupt. Regardless of previous outcome the KVM logging API | |
4751 | * does not preclude user space subsequent dirty log read. Flushing TLB ensures | |
4752 | * writes will be marked dirty for next log read. | |
95d4c16c | 4753 | * |
60c34612 TY |
4754 | * 1. Take a snapshot of the bit and clear it if needed. |
4755 | * 2. Write protect the corresponding page. | |
e108ff2f PB |
4756 | * 3. Copy the snapshot to the userspace. |
4757 | * 4. Flush TLB's if needed. | |
5bb064dc | 4758 | */ |
60c34612 | 4759 | int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log) |
5bb064dc | 4760 | { |
8fe65a82 | 4761 | bool flush = false; |
e108ff2f | 4762 | int r; |
5bb064dc | 4763 | |
79fac95e | 4764 | mutex_lock(&kvm->slots_lock); |
5bb064dc | 4765 | |
88178fd4 KH |
4766 | /* |
4767 | * Flush potentially hardware-cached dirty pages to dirty_bitmap. | |
4768 | */ | |
4769 | if (kvm_x86_ops->flush_log_dirty) | |
4770 | kvm_x86_ops->flush_log_dirty(kvm); | |
4771 | ||
8fe65a82 | 4772 | r = kvm_get_dirty_log_protect(kvm, log, &flush); |
198c74f4 XG |
4773 | |
4774 | /* | |
4775 | * All the TLBs can be flushed out of mmu lock, see the comments in | |
4776 | * kvm_mmu_slot_remove_write_access(). | |
4777 | */ | |
e108ff2f | 4778 | lockdep_assert_held(&kvm->slots_lock); |
8fe65a82 | 4779 | if (flush) |
2a31b9db PB |
4780 | kvm_flush_remote_tlbs(kvm); |
4781 | ||
4782 | mutex_unlock(&kvm->slots_lock); | |
4783 | return r; | |
4784 | } | |
4785 | ||
4786 | int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log) | |
4787 | { | |
4788 | bool flush = false; | |
4789 | int r; | |
4790 | ||
4791 | mutex_lock(&kvm->slots_lock); | |
4792 | ||
4793 | /* | |
4794 | * Flush potentially hardware-cached dirty pages to dirty_bitmap. | |
4795 | */ | |
4796 | if (kvm_x86_ops->flush_log_dirty) | |
4797 | kvm_x86_ops->flush_log_dirty(kvm); | |
4798 | ||
4799 | r = kvm_clear_dirty_log_protect(kvm, log, &flush); | |
4800 | ||
4801 | /* | |
4802 | * All the TLBs can be flushed out of mmu lock, see the comments in | |
4803 | * kvm_mmu_slot_remove_write_access(). | |
4804 | */ | |
4805 | lockdep_assert_held(&kvm->slots_lock); | |
4806 | if (flush) | |
198c74f4 XG |
4807 | kvm_flush_remote_tlbs(kvm); |
4808 | ||
79fac95e | 4809 | mutex_unlock(&kvm->slots_lock); |
5bb064dc ZX |
4810 | return r; |
4811 | } | |
4812 | ||
aa2fbe6d YZ |
4813 | int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event, |
4814 | bool line_status) | |
23d43cf9 CD |
4815 | { |
4816 | if (!irqchip_in_kernel(kvm)) | |
4817 | return -ENXIO; | |
4818 | ||
4819 | irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, | |
aa2fbe6d YZ |
4820 | irq_event->irq, irq_event->level, |
4821 | line_status); | |
23d43cf9 CD |
4822 | return 0; |
4823 | } | |
4824 | ||
e5d83c74 PB |
4825 | int kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
4826 | struct kvm_enable_cap *cap) | |
90de4a18 NA |
4827 | { |
4828 | int r; | |
4829 | ||
4830 | if (cap->flags) | |
4831 | return -EINVAL; | |
4832 | ||
4833 | switch (cap->cap) { | |
4834 | case KVM_CAP_DISABLE_QUIRKS: | |
4835 | kvm->arch.disabled_quirks = cap->args[0]; | |
4836 | r = 0; | |
4837 | break; | |
49df6397 SR |
4838 | case KVM_CAP_SPLIT_IRQCHIP: { |
4839 | mutex_lock(&kvm->lock); | |
b053b2ae SR |
4840 | r = -EINVAL; |
4841 | if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS) | |
4842 | goto split_irqchip_unlock; | |
49df6397 SR |
4843 | r = -EEXIST; |
4844 | if (irqchip_in_kernel(kvm)) | |
4845 | goto split_irqchip_unlock; | |
557abc40 | 4846 | if (kvm->created_vcpus) |
49df6397 SR |
4847 | goto split_irqchip_unlock; |
4848 | r = kvm_setup_empty_irq_routing(kvm); | |
5c0aea0e | 4849 | if (r) |
49df6397 SR |
4850 | goto split_irqchip_unlock; |
4851 | /* Pairs with irqchip_in_kernel. */ | |
4852 | smp_wmb(); | |
49776faf | 4853 | kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT; |
b053b2ae | 4854 | kvm->arch.nr_reserved_ioapic_pins = cap->args[0]; |
49df6397 SR |
4855 | r = 0; |
4856 | split_irqchip_unlock: | |
4857 | mutex_unlock(&kvm->lock); | |
4858 | break; | |
4859 | } | |
37131313 RK |
4860 | case KVM_CAP_X2APIC_API: |
4861 | r = -EINVAL; | |
4862 | if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS) | |
4863 | break; | |
4864 | ||
4865 | if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS) | |
4866 | kvm->arch.x2apic_format = true; | |
c519265f RK |
4867 | if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK) |
4868 | kvm->arch.x2apic_broadcast_quirk_disabled = true; | |
37131313 RK |
4869 | |
4870 | r = 0; | |
4871 | break; | |
4d5422ce WL |
4872 | case KVM_CAP_X86_DISABLE_EXITS: |
4873 | r = -EINVAL; | |
4874 | if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS) | |
4875 | break; | |
4876 | ||
4877 | if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) && | |
4878 | kvm_can_mwait_in_guest()) | |
4879 | kvm->arch.mwait_in_guest = true; | |
766d3571 | 4880 | if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT) |
caa057a2 | 4881 | kvm->arch.hlt_in_guest = true; |
b31c114b WL |
4882 | if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE) |
4883 | kvm->arch.pause_in_guest = true; | |
b5170063 WL |
4884 | if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE) |
4885 | kvm->arch.cstate_in_guest = true; | |
4d5422ce WL |
4886 | r = 0; |
4887 | break; | |
6fbbde9a DS |
4888 | case KVM_CAP_MSR_PLATFORM_INFO: |
4889 | kvm->arch.guest_can_read_msr_platform_info = cap->args[0]; | |
4890 | r = 0; | |
c4f55198 JM |
4891 | break; |
4892 | case KVM_CAP_EXCEPTION_PAYLOAD: | |
4893 | kvm->arch.exception_payload_enabled = cap->args[0]; | |
4894 | r = 0; | |
6fbbde9a | 4895 | break; |
90de4a18 NA |
4896 | default: |
4897 | r = -EINVAL; | |
4898 | break; | |
4899 | } | |
4900 | return r; | |
4901 | } | |
4902 | ||
1fe779f8 CO |
4903 | long kvm_arch_vm_ioctl(struct file *filp, |
4904 | unsigned int ioctl, unsigned long arg) | |
4905 | { | |
4906 | struct kvm *kvm = filp->private_data; | |
4907 | void __user *argp = (void __user *)arg; | |
367e1319 | 4908 | int r = -ENOTTY; |
f0d66275 DH |
4909 | /* |
4910 | * This union makes it completely explicit to gcc-3.x | |
4911 | * that these two variables' stack usage should be | |
4912 | * combined, not added together. | |
4913 | */ | |
4914 | union { | |
4915 | struct kvm_pit_state ps; | |
e9f42757 | 4916 | struct kvm_pit_state2 ps2; |
c5ff41ce | 4917 | struct kvm_pit_config pit_config; |
f0d66275 | 4918 | } u; |
1fe779f8 CO |
4919 | |
4920 | switch (ioctl) { | |
4921 | case KVM_SET_TSS_ADDR: | |
4922 | r = kvm_vm_ioctl_set_tss_addr(kvm, arg); | |
1fe779f8 | 4923 | break; |
b927a3ce SY |
4924 | case KVM_SET_IDENTITY_MAP_ADDR: { |
4925 | u64 ident_addr; | |
4926 | ||
1af1ac91 DH |
4927 | mutex_lock(&kvm->lock); |
4928 | r = -EINVAL; | |
4929 | if (kvm->created_vcpus) | |
4930 | goto set_identity_unlock; | |
b927a3ce | 4931 | r = -EFAULT; |
0e96f31e | 4932 | if (copy_from_user(&ident_addr, argp, sizeof(ident_addr))) |
1af1ac91 | 4933 | goto set_identity_unlock; |
b927a3ce | 4934 | r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr); |
1af1ac91 DH |
4935 | set_identity_unlock: |
4936 | mutex_unlock(&kvm->lock); | |
b927a3ce SY |
4937 | break; |
4938 | } | |
1fe779f8 CO |
4939 | case KVM_SET_NR_MMU_PAGES: |
4940 | r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg); | |
1fe779f8 CO |
4941 | break; |
4942 | case KVM_GET_NR_MMU_PAGES: | |
4943 | r = kvm_vm_ioctl_get_nr_mmu_pages(kvm); | |
4944 | break; | |
3ddea128 | 4945 | case KVM_CREATE_IRQCHIP: { |
3ddea128 | 4946 | mutex_lock(&kvm->lock); |
09941366 | 4947 | |
3ddea128 | 4948 | r = -EEXIST; |
35e6eaa3 | 4949 | if (irqchip_in_kernel(kvm)) |
3ddea128 | 4950 | goto create_irqchip_unlock; |
09941366 | 4951 | |
3e515705 | 4952 | r = -EINVAL; |
557abc40 | 4953 | if (kvm->created_vcpus) |
3e515705 | 4954 | goto create_irqchip_unlock; |
09941366 RK |
4955 | |
4956 | r = kvm_pic_init(kvm); | |
4957 | if (r) | |
3ddea128 | 4958 | goto create_irqchip_unlock; |
09941366 RK |
4959 | |
4960 | r = kvm_ioapic_init(kvm); | |
4961 | if (r) { | |
09941366 | 4962 | kvm_pic_destroy(kvm); |
3ddea128 | 4963 | goto create_irqchip_unlock; |
09941366 RK |
4964 | } |
4965 | ||
399ec807 AK |
4966 | r = kvm_setup_default_irq_routing(kvm); |
4967 | if (r) { | |
72bb2fcd | 4968 | kvm_ioapic_destroy(kvm); |
09941366 | 4969 | kvm_pic_destroy(kvm); |
71ba994c | 4970 | goto create_irqchip_unlock; |
399ec807 | 4971 | } |
49776faf | 4972 | /* Write kvm->irq_routing before enabling irqchip_in_kernel. */ |
71ba994c | 4973 | smp_wmb(); |
49776faf | 4974 | kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL; |
3ddea128 MT |
4975 | create_irqchip_unlock: |
4976 | mutex_unlock(&kvm->lock); | |
1fe779f8 | 4977 | break; |
3ddea128 | 4978 | } |
7837699f | 4979 | case KVM_CREATE_PIT: |
c5ff41ce JK |
4980 | u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY; |
4981 | goto create_pit; | |
4982 | case KVM_CREATE_PIT2: | |
4983 | r = -EFAULT; | |
4984 | if (copy_from_user(&u.pit_config, argp, | |
4985 | sizeof(struct kvm_pit_config))) | |
4986 | goto out; | |
4987 | create_pit: | |
250715a6 | 4988 | mutex_lock(&kvm->lock); |
269e05e4 AK |
4989 | r = -EEXIST; |
4990 | if (kvm->arch.vpit) | |
4991 | goto create_pit_unlock; | |
7837699f | 4992 | r = -ENOMEM; |
c5ff41ce | 4993 | kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags); |
7837699f SY |
4994 | if (kvm->arch.vpit) |
4995 | r = 0; | |
269e05e4 | 4996 | create_pit_unlock: |
250715a6 | 4997 | mutex_unlock(&kvm->lock); |
7837699f | 4998 | break; |
1fe779f8 CO |
4999 | case KVM_GET_IRQCHIP: { |
5000 | /* 0: PIC master, 1: PIC slave, 2: IOAPIC */ | |
ff5c2c03 | 5001 | struct kvm_irqchip *chip; |
1fe779f8 | 5002 | |
ff5c2c03 SL |
5003 | chip = memdup_user(argp, sizeof(*chip)); |
5004 | if (IS_ERR(chip)) { | |
5005 | r = PTR_ERR(chip); | |
1fe779f8 | 5006 | goto out; |
ff5c2c03 SL |
5007 | } |
5008 | ||
1fe779f8 | 5009 | r = -ENXIO; |
826da321 | 5010 | if (!irqchip_kernel(kvm)) |
f0d66275 DH |
5011 | goto get_irqchip_out; |
5012 | r = kvm_vm_ioctl_get_irqchip(kvm, chip); | |
1fe779f8 | 5013 | if (r) |
f0d66275 | 5014 | goto get_irqchip_out; |
1fe779f8 | 5015 | r = -EFAULT; |
0e96f31e | 5016 | if (copy_to_user(argp, chip, sizeof(*chip))) |
f0d66275 | 5017 | goto get_irqchip_out; |
1fe779f8 | 5018 | r = 0; |
f0d66275 DH |
5019 | get_irqchip_out: |
5020 | kfree(chip); | |
1fe779f8 CO |
5021 | break; |
5022 | } | |
5023 | case KVM_SET_IRQCHIP: { | |
5024 | /* 0: PIC master, 1: PIC slave, 2: IOAPIC */ | |
ff5c2c03 | 5025 | struct kvm_irqchip *chip; |
1fe779f8 | 5026 | |
ff5c2c03 SL |
5027 | chip = memdup_user(argp, sizeof(*chip)); |
5028 | if (IS_ERR(chip)) { | |
5029 | r = PTR_ERR(chip); | |
1fe779f8 | 5030 | goto out; |
ff5c2c03 SL |
5031 | } |
5032 | ||
1fe779f8 | 5033 | r = -ENXIO; |
826da321 | 5034 | if (!irqchip_kernel(kvm)) |
f0d66275 DH |
5035 | goto set_irqchip_out; |
5036 | r = kvm_vm_ioctl_set_irqchip(kvm, chip); | |
f0d66275 DH |
5037 | set_irqchip_out: |
5038 | kfree(chip); | |
1fe779f8 CO |
5039 | break; |
5040 | } | |
e0f63cb9 | 5041 | case KVM_GET_PIT: { |
e0f63cb9 | 5042 | r = -EFAULT; |
f0d66275 | 5043 | if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state))) |
e0f63cb9 SY |
5044 | goto out; |
5045 | r = -ENXIO; | |
5046 | if (!kvm->arch.vpit) | |
5047 | goto out; | |
f0d66275 | 5048 | r = kvm_vm_ioctl_get_pit(kvm, &u.ps); |
e0f63cb9 SY |
5049 | if (r) |
5050 | goto out; | |
5051 | r = -EFAULT; | |
f0d66275 | 5052 | if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state))) |
e0f63cb9 SY |
5053 | goto out; |
5054 | r = 0; | |
5055 | break; | |
5056 | } | |
5057 | case KVM_SET_PIT: { | |
e0f63cb9 | 5058 | r = -EFAULT; |
0e96f31e | 5059 | if (copy_from_user(&u.ps, argp, sizeof(u.ps))) |
e0f63cb9 SY |
5060 | goto out; |
5061 | r = -ENXIO; | |
5062 | if (!kvm->arch.vpit) | |
5063 | goto out; | |
f0d66275 | 5064 | r = kvm_vm_ioctl_set_pit(kvm, &u.ps); |
e0f63cb9 SY |
5065 | break; |
5066 | } | |
e9f42757 BK |
5067 | case KVM_GET_PIT2: { |
5068 | r = -ENXIO; | |
5069 | if (!kvm->arch.vpit) | |
5070 | goto out; | |
5071 | r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2); | |
5072 | if (r) | |
5073 | goto out; | |
5074 | r = -EFAULT; | |
5075 | if (copy_to_user(argp, &u.ps2, sizeof(u.ps2))) | |
5076 | goto out; | |
5077 | r = 0; | |
5078 | break; | |
5079 | } | |
5080 | case KVM_SET_PIT2: { | |
5081 | r = -EFAULT; | |
5082 | if (copy_from_user(&u.ps2, argp, sizeof(u.ps2))) | |
5083 | goto out; | |
5084 | r = -ENXIO; | |
5085 | if (!kvm->arch.vpit) | |
5086 | goto out; | |
5087 | r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2); | |
e9f42757 BK |
5088 | break; |
5089 | } | |
52d939a0 MT |
5090 | case KVM_REINJECT_CONTROL: { |
5091 | struct kvm_reinject_control control; | |
5092 | r = -EFAULT; | |
5093 | if (copy_from_user(&control, argp, sizeof(control))) | |
5094 | goto out; | |
cad23e72 ML |
5095 | r = -ENXIO; |
5096 | if (!kvm->arch.vpit) | |
5097 | goto out; | |
52d939a0 | 5098 | r = kvm_vm_ioctl_reinject(kvm, &control); |
52d939a0 MT |
5099 | break; |
5100 | } | |
d71ba788 PB |
5101 | case KVM_SET_BOOT_CPU_ID: |
5102 | r = 0; | |
5103 | mutex_lock(&kvm->lock); | |
557abc40 | 5104 | if (kvm->created_vcpus) |
d71ba788 PB |
5105 | r = -EBUSY; |
5106 | else | |
5107 | kvm->arch.bsp_vcpu_id = arg; | |
5108 | mutex_unlock(&kvm->lock); | |
5109 | break; | |
ffde22ac | 5110 | case KVM_XEN_HVM_CONFIG: { |
51776043 | 5111 | struct kvm_xen_hvm_config xhc; |
ffde22ac | 5112 | r = -EFAULT; |
51776043 | 5113 | if (copy_from_user(&xhc, argp, sizeof(xhc))) |
ffde22ac ES |
5114 | goto out; |
5115 | r = -EINVAL; | |
51776043 | 5116 | if (xhc.flags) |
ffde22ac | 5117 | goto out; |
51776043 | 5118 | memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc)); |
ffde22ac ES |
5119 | r = 0; |
5120 | break; | |
5121 | } | |
afbcf7ab | 5122 | case KVM_SET_CLOCK: { |
afbcf7ab GC |
5123 | struct kvm_clock_data user_ns; |
5124 | u64 now_ns; | |
afbcf7ab GC |
5125 | |
5126 | r = -EFAULT; | |
5127 | if (copy_from_user(&user_ns, argp, sizeof(user_ns))) | |
5128 | goto out; | |
5129 | ||
5130 | r = -EINVAL; | |
5131 | if (user_ns.flags) | |
5132 | goto out; | |
5133 | ||
5134 | r = 0; | |
0bc48bea RK |
5135 | /* |
5136 | * TODO: userspace has to take care of races with VCPU_RUN, so | |
5137 | * kvm_gen_update_masterclock() can be cut down to locked | |
5138 | * pvclock_update_vm_gtod_copy(). | |
5139 | */ | |
5140 | kvm_gen_update_masterclock(kvm); | |
e891a32e | 5141 | now_ns = get_kvmclock_ns(kvm); |
108b249c | 5142 | kvm->arch.kvmclock_offset += user_ns.clock - now_ns; |
0bc48bea | 5143 | kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE); |
afbcf7ab GC |
5144 | break; |
5145 | } | |
5146 | case KVM_GET_CLOCK: { | |
afbcf7ab GC |
5147 | struct kvm_clock_data user_ns; |
5148 | u64 now_ns; | |
5149 | ||
e891a32e | 5150 | now_ns = get_kvmclock_ns(kvm); |
108b249c | 5151 | user_ns.clock = now_ns; |
e3fd9a93 | 5152 | user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0; |
97e69aa6 | 5153 | memset(&user_ns.pad, 0, sizeof(user_ns.pad)); |
afbcf7ab GC |
5154 | |
5155 | r = -EFAULT; | |
5156 | if (copy_to_user(argp, &user_ns, sizeof(user_ns))) | |
5157 | goto out; | |
5158 | r = 0; | |
5159 | break; | |
5160 | } | |
5acc5c06 BS |
5161 | case KVM_MEMORY_ENCRYPT_OP: { |
5162 | r = -ENOTTY; | |
5163 | if (kvm_x86_ops->mem_enc_op) | |
5164 | r = kvm_x86_ops->mem_enc_op(kvm, argp); | |
5165 | break; | |
5166 | } | |
69eaedee BS |
5167 | case KVM_MEMORY_ENCRYPT_REG_REGION: { |
5168 | struct kvm_enc_region region; | |
5169 | ||
5170 | r = -EFAULT; | |
5171 | if (copy_from_user(®ion, argp, sizeof(region))) | |
5172 | goto out; | |
5173 | ||
5174 | r = -ENOTTY; | |
5175 | if (kvm_x86_ops->mem_enc_reg_region) | |
5176 | r = kvm_x86_ops->mem_enc_reg_region(kvm, ®ion); | |
5177 | break; | |
5178 | } | |
5179 | case KVM_MEMORY_ENCRYPT_UNREG_REGION: { | |
5180 | struct kvm_enc_region region; | |
5181 | ||
5182 | r = -EFAULT; | |
5183 | if (copy_from_user(®ion, argp, sizeof(region))) | |
5184 | goto out; | |
5185 | ||
5186 | r = -ENOTTY; | |
5187 | if (kvm_x86_ops->mem_enc_unreg_region) | |
5188 | r = kvm_x86_ops->mem_enc_unreg_region(kvm, ®ion); | |
5189 | break; | |
5190 | } | |
faeb7833 RK |
5191 | case KVM_HYPERV_EVENTFD: { |
5192 | struct kvm_hyperv_eventfd hvevfd; | |
5193 | ||
5194 | r = -EFAULT; | |
5195 | if (copy_from_user(&hvevfd, argp, sizeof(hvevfd))) | |
5196 | goto out; | |
5197 | r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd); | |
5198 | break; | |
5199 | } | |
66bb8a06 EH |
5200 | case KVM_SET_PMU_EVENT_FILTER: |
5201 | r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp); | |
5202 | break; | |
1fe779f8 | 5203 | default: |
ad6260da | 5204 | r = -ENOTTY; |
1fe779f8 CO |
5205 | } |
5206 | out: | |
5207 | return r; | |
5208 | } | |
5209 | ||
a16b043c | 5210 | static void kvm_init_msr_list(void) |
043405e1 | 5211 | { |
24c29b7a | 5212 | struct x86_pmu_capability x86_pmu; |
043405e1 | 5213 | u32 dummy[2]; |
7a5ee6ed | 5214 | unsigned i; |
043405e1 | 5215 | |
e2ada66e | 5216 | BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4, |
7a5ee6ed | 5217 | "Please update the fixed PMCs in msrs_to_saved_all[]"); |
24c29b7a PB |
5218 | |
5219 | perf_get_x86_pmu_capability(&x86_pmu); | |
e2ada66e | 5220 | |
6cbee2b9 XL |
5221 | num_msrs_to_save = 0; |
5222 | num_emulated_msrs = 0; | |
5223 | num_msr_based_features = 0; | |
5224 | ||
7a5ee6ed CQ |
5225 | for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) { |
5226 | if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0) | |
043405e1 | 5227 | continue; |
93c4adc7 PB |
5228 | |
5229 | /* | |
5230 | * Even MSRs that are valid in the host may not be exposed | |
9dbe6cf9 | 5231 | * to the guests in some cases. |
93c4adc7 | 5232 | */ |
7a5ee6ed | 5233 | switch (msrs_to_save_all[i]) { |
93c4adc7 | 5234 | case MSR_IA32_BNDCFGS: |
503234b3 | 5235 | if (!kvm_mpx_supported()) |
93c4adc7 PB |
5236 | continue; |
5237 | break; | |
9dbe6cf9 PB |
5238 | case MSR_TSC_AUX: |
5239 | if (!kvm_x86_ops->rdtscp_supported()) | |
5240 | continue; | |
5241 | break; | |
bf8c55d8 CP |
5242 | case MSR_IA32_RTIT_CTL: |
5243 | case MSR_IA32_RTIT_STATUS: | |
5244 | if (!kvm_x86_ops->pt_supported()) | |
5245 | continue; | |
5246 | break; | |
5247 | case MSR_IA32_RTIT_CR3_MATCH: | |
5248 | if (!kvm_x86_ops->pt_supported() || | |
5249 | !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering)) | |
5250 | continue; | |
5251 | break; | |
5252 | case MSR_IA32_RTIT_OUTPUT_BASE: | |
5253 | case MSR_IA32_RTIT_OUTPUT_MASK: | |
5254 | if (!kvm_x86_ops->pt_supported() || | |
5255 | (!intel_pt_validate_hw_cap(PT_CAP_topa_output) && | |
5256 | !intel_pt_validate_hw_cap(PT_CAP_single_range_output))) | |
5257 | continue; | |
5258 | break; | |
5259 | case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: { | |
5260 | if (!kvm_x86_ops->pt_supported() || | |
7a5ee6ed | 5261 | msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >= |
bf8c55d8 CP |
5262 | intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2) |
5263 | continue; | |
5264 | break; | |
cf05a67b | 5265 | case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17: |
7a5ee6ed | 5266 | if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >= |
24c29b7a PB |
5267 | min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp)) |
5268 | continue; | |
5269 | break; | |
cf05a67b | 5270 | case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17: |
7a5ee6ed | 5271 | if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >= |
24c29b7a PB |
5272 | min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp)) |
5273 | continue; | |
bf8c55d8 | 5274 | } |
93c4adc7 PB |
5275 | default: |
5276 | break; | |
5277 | } | |
5278 | ||
7a5ee6ed | 5279 | msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i]; |
043405e1 | 5280 | } |
62ef68bb | 5281 | |
7a5ee6ed CQ |
5282 | for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) { |
5283 | if (!kvm_x86_ops->has_emulated_msr(emulated_msrs_all[i])) | |
bc226f07 | 5284 | continue; |
62ef68bb | 5285 | |
7a5ee6ed | 5286 | emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i]; |
62ef68bb | 5287 | } |
801e459a | 5288 | |
7a5ee6ed | 5289 | for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) { |
801e459a TL |
5290 | struct kvm_msr_entry msr; |
5291 | ||
7a5ee6ed | 5292 | msr.index = msr_based_features_all[i]; |
66421c1e | 5293 | if (kvm_get_msr_feature(&msr)) |
801e459a TL |
5294 | continue; |
5295 | ||
7a5ee6ed | 5296 | msr_based_features[num_msr_based_features++] = msr_based_features_all[i]; |
801e459a | 5297 | } |
043405e1 CO |
5298 | } |
5299 | ||
bda9020e MT |
5300 | static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len, |
5301 | const void *v) | |
bbd9b64e | 5302 | { |
70252a10 AK |
5303 | int handled = 0; |
5304 | int n; | |
5305 | ||
5306 | do { | |
5307 | n = min(len, 8); | |
bce87cce | 5308 | if (!(lapic_in_kernel(vcpu) && |
e32edf4f NN |
5309 | !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v)) |
5310 | && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v)) | |
70252a10 AK |
5311 | break; |
5312 | handled += n; | |
5313 | addr += n; | |
5314 | len -= n; | |
5315 | v += n; | |
5316 | } while (len); | |
bbd9b64e | 5317 | |
70252a10 | 5318 | return handled; |
bbd9b64e CO |
5319 | } |
5320 | ||
bda9020e | 5321 | static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v) |
bbd9b64e | 5322 | { |
70252a10 AK |
5323 | int handled = 0; |
5324 | int n; | |
5325 | ||
5326 | do { | |
5327 | n = min(len, 8); | |
bce87cce | 5328 | if (!(lapic_in_kernel(vcpu) && |
e32edf4f NN |
5329 | !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev, |
5330 | addr, n, v)) | |
5331 | && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v)) | |
70252a10 | 5332 | break; |
e39d200f | 5333 | trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v); |
70252a10 AK |
5334 | handled += n; |
5335 | addr += n; | |
5336 | len -= n; | |
5337 | v += n; | |
5338 | } while (len); | |
bbd9b64e | 5339 | |
70252a10 | 5340 | return handled; |
bbd9b64e CO |
5341 | } |
5342 | ||
2dafc6c2 GN |
5343 | static void kvm_set_segment(struct kvm_vcpu *vcpu, |
5344 | struct kvm_segment *var, int seg) | |
5345 | { | |
5346 | kvm_x86_ops->set_segment(vcpu, var, seg); | |
5347 | } | |
5348 | ||
5349 | void kvm_get_segment(struct kvm_vcpu *vcpu, | |
5350 | struct kvm_segment *var, int seg) | |
5351 | { | |
5352 | kvm_x86_ops->get_segment(vcpu, var, seg); | |
5353 | } | |
5354 | ||
54987b7a PB |
5355 | gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access, |
5356 | struct x86_exception *exception) | |
02f59dc9 JR |
5357 | { |
5358 | gpa_t t_gpa; | |
02f59dc9 JR |
5359 | |
5360 | BUG_ON(!mmu_is_nested(vcpu)); | |
5361 | ||
5362 | /* NPT walks are always user-walks */ | |
5363 | access |= PFERR_USER_MASK; | |
44dd3ffa | 5364 | t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception); |
02f59dc9 JR |
5365 | |
5366 | return t_gpa; | |
5367 | } | |
5368 | ||
ab9ae313 AK |
5369 | gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, |
5370 | struct x86_exception *exception) | |
1871c602 GN |
5371 | { |
5372 | u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; | |
ab9ae313 | 5373 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
1871c602 GN |
5374 | } |
5375 | ||
ab9ae313 AK |
5376 | gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, |
5377 | struct x86_exception *exception) | |
1871c602 GN |
5378 | { |
5379 | u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; | |
5380 | access |= PFERR_FETCH_MASK; | |
ab9ae313 | 5381 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
1871c602 GN |
5382 | } |
5383 | ||
ab9ae313 AK |
5384 | gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, |
5385 | struct x86_exception *exception) | |
1871c602 GN |
5386 | { |
5387 | u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; | |
5388 | access |= PFERR_WRITE_MASK; | |
ab9ae313 | 5389 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
1871c602 GN |
5390 | } |
5391 | ||
5392 | /* uses this to access any guest's mapped memory without checking CPL */ | |
ab9ae313 AK |
5393 | gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, |
5394 | struct x86_exception *exception) | |
1871c602 | 5395 | { |
ab9ae313 | 5396 | return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception); |
1871c602 GN |
5397 | } |
5398 | ||
5399 | static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes, | |
5400 | struct kvm_vcpu *vcpu, u32 access, | |
bcc55cba | 5401 | struct x86_exception *exception) |
bbd9b64e CO |
5402 | { |
5403 | void *data = val; | |
10589a46 | 5404 | int r = X86EMUL_CONTINUE; |
bbd9b64e CO |
5405 | |
5406 | while (bytes) { | |
14dfe855 | 5407 | gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access, |
ab9ae313 | 5408 | exception); |
bbd9b64e | 5409 | unsigned offset = addr & (PAGE_SIZE-1); |
77c2002e | 5410 | unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset); |
bbd9b64e CO |
5411 | int ret; |
5412 | ||
bcc55cba | 5413 | if (gpa == UNMAPPED_GVA) |
ab9ae313 | 5414 | return X86EMUL_PROPAGATE_FAULT; |
54bf36aa PB |
5415 | ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data, |
5416 | offset, toread); | |
10589a46 | 5417 | if (ret < 0) { |
c3cd7ffa | 5418 | r = X86EMUL_IO_NEEDED; |
10589a46 MT |
5419 | goto out; |
5420 | } | |
bbd9b64e | 5421 | |
77c2002e IE |
5422 | bytes -= toread; |
5423 | data += toread; | |
5424 | addr += toread; | |
bbd9b64e | 5425 | } |
10589a46 | 5426 | out: |
10589a46 | 5427 | return r; |
bbd9b64e | 5428 | } |
77c2002e | 5429 | |
1871c602 | 5430 | /* used for instruction fetching */ |
0f65dd70 AK |
5431 | static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt, |
5432 | gva_t addr, void *val, unsigned int bytes, | |
bcc55cba | 5433 | struct x86_exception *exception) |
1871c602 | 5434 | { |
0f65dd70 | 5435 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
1871c602 | 5436 | u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; |
44583cba PB |
5437 | unsigned offset; |
5438 | int ret; | |
0f65dd70 | 5439 | |
44583cba PB |
5440 | /* Inline kvm_read_guest_virt_helper for speed. */ |
5441 | gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK, | |
5442 | exception); | |
5443 | if (unlikely(gpa == UNMAPPED_GVA)) | |
5444 | return X86EMUL_PROPAGATE_FAULT; | |
5445 | ||
5446 | offset = addr & (PAGE_SIZE-1); | |
5447 | if (WARN_ON(offset + bytes > PAGE_SIZE)) | |
5448 | bytes = (unsigned)PAGE_SIZE - offset; | |
54bf36aa PB |
5449 | ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val, |
5450 | offset, bytes); | |
44583cba PB |
5451 | if (unlikely(ret < 0)) |
5452 | return X86EMUL_IO_NEEDED; | |
5453 | ||
5454 | return X86EMUL_CONTINUE; | |
1871c602 GN |
5455 | } |
5456 | ||
ce14e868 | 5457 | int kvm_read_guest_virt(struct kvm_vcpu *vcpu, |
0f65dd70 | 5458 | gva_t addr, void *val, unsigned int bytes, |
bcc55cba | 5459 | struct x86_exception *exception) |
1871c602 GN |
5460 | { |
5461 | u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0; | |
0f65dd70 | 5462 | |
353c0956 PB |
5463 | /* |
5464 | * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED | |
5465 | * is returned, but our callers are not ready for that and they blindly | |
5466 | * call kvm_inject_page_fault. Ensure that they at least do not leak | |
5467 | * uninitialized kernel stack memory into cr2 and error code. | |
5468 | */ | |
5469 | memset(exception, 0, sizeof(*exception)); | |
1871c602 | 5470 | return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, |
bcc55cba | 5471 | exception); |
1871c602 | 5472 | } |
064aea77 | 5473 | EXPORT_SYMBOL_GPL(kvm_read_guest_virt); |
1871c602 | 5474 | |
ce14e868 PB |
5475 | static int emulator_read_std(struct x86_emulate_ctxt *ctxt, |
5476 | gva_t addr, void *val, unsigned int bytes, | |
3c9fa24c | 5477 | struct x86_exception *exception, bool system) |
1871c602 | 5478 | { |
0f65dd70 | 5479 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
3c9fa24c PB |
5480 | u32 access = 0; |
5481 | ||
5482 | if (!system && kvm_x86_ops->get_cpl(vcpu) == 3) | |
5483 | access |= PFERR_USER_MASK; | |
5484 | ||
5485 | return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception); | |
1871c602 GN |
5486 | } |
5487 | ||
7a036a6f RK |
5488 | static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt, |
5489 | unsigned long addr, void *val, unsigned int bytes) | |
5490 | { | |
5491 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
5492 | int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes); | |
5493 | ||
5494 | return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE; | |
5495 | } | |
5496 | ||
ce14e868 PB |
5497 | static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes, |
5498 | struct kvm_vcpu *vcpu, u32 access, | |
5499 | struct x86_exception *exception) | |
77c2002e IE |
5500 | { |
5501 | void *data = val; | |
5502 | int r = X86EMUL_CONTINUE; | |
5503 | ||
5504 | while (bytes) { | |
14dfe855 | 5505 | gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, |
ce14e868 | 5506 | access, |
ab9ae313 | 5507 | exception); |
77c2002e IE |
5508 | unsigned offset = addr & (PAGE_SIZE-1); |
5509 | unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset); | |
5510 | int ret; | |
5511 | ||
bcc55cba | 5512 | if (gpa == UNMAPPED_GVA) |
ab9ae313 | 5513 | return X86EMUL_PROPAGATE_FAULT; |
54bf36aa | 5514 | ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite); |
77c2002e | 5515 | if (ret < 0) { |
c3cd7ffa | 5516 | r = X86EMUL_IO_NEEDED; |
77c2002e IE |
5517 | goto out; |
5518 | } | |
5519 | ||
5520 | bytes -= towrite; | |
5521 | data += towrite; | |
5522 | addr += towrite; | |
5523 | } | |
5524 | out: | |
5525 | return r; | |
5526 | } | |
ce14e868 PB |
5527 | |
5528 | static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val, | |
3c9fa24c PB |
5529 | unsigned int bytes, struct x86_exception *exception, |
5530 | bool system) | |
ce14e868 PB |
5531 | { |
5532 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
3c9fa24c PB |
5533 | u32 access = PFERR_WRITE_MASK; |
5534 | ||
5535 | if (!system && kvm_x86_ops->get_cpl(vcpu) == 3) | |
5536 | access |= PFERR_USER_MASK; | |
ce14e868 PB |
5537 | |
5538 | return kvm_write_guest_virt_helper(addr, val, bytes, vcpu, | |
3c9fa24c | 5539 | access, exception); |
ce14e868 PB |
5540 | } |
5541 | ||
5542 | int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val, | |
5543 | unsigned int bytes, struct x86_exception *exception) | |
5544 | { | |
c595ceee PB |
5545 | /* kvm_write_guest_virt_system can pull in tons of pages. */ |
5546 | vcpu->arch.l1tf_flush_l1d = true; | |
5547 | ||
541ab2ae FH |
5548 | /* |
5549 | * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED | |
5550 | * is returned, but our callers are not ready for that and they blindly | |
5551 | * call kvm_inject_page_fault. Ensure that they at least do not leak | |
5552 | * uninitialized kernel stack memory into cr2 and error code. | |
5553 | */ | |
5554 | memset(exception, 0, sizeof(*exception)); | |
ce14e868 PB |
5555 | return kvm_write_guest_virt_helper(addr, val, bytes, vcpu, |
5556 | PFERR_WRITE_MASK, exception); | |
5557 | } | |
6a4d7550 | 5558 | EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system); |
77c2002e | 5559 | |
082d06ed WL |
5560 | int handle_ud(struct kvm_vcpu *vcpu) |
5561 | { | |
b3dc0695 | 5562 | static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX }; |
6c86eedc | 5563 | int emul_type = EMULTYPE_TRAP_UD; |
6c86eedc WL |
5564 | char sig[5]; /* ud2; .ascii "kvm" */ |
5565 | struct x86_exception e; | |
5566 | ||
5567 | if (force_emulation_prefix && | |
3c9fa24c PB |
5568 | kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu), |
5569 | sig, sizeof(sig), &e) == 0 && | |
b3dc0695 | 5570 | memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) { |
6c86eedc | 5571 | kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig)); |
b4000606 | 5572 | emul_type = EMULTYPE_TRAP_UD_FORCED; |
6c86eedc | 5573 | } |
082d06ed | 5574 | |
60fc3d02 | 5575 | return kvm_emulate_instruction(vcpu, emul_type); |
082d06ed WL |
5576 | } |
5577 | EXPORT_SYMBOL_GPL(handle_ud); | |
5578 | ||
0f89b207 TL |
5579 | static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva, |
5580 | gpa_t gpa, bool write) | |
5581 | { | |
5582 | /* For APIC access vmexit */ | |
5583 | if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE) | |
5584 | return 1; | |
5585 | ||
5586 | if (vcpu_match_mmio_gpa(vcpu, gpa)) { | |
5587 | trace_vcpu_match_mmio(gva, gpa, write, true); | |
5588 | return 1; | |
5589 | } | |
5590 | ||
5591 | return 0; | |
5592 | } | |
5593 | ||
af7cc7d1 XG |
5594 | static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva, |
5595 | gpa_t *gpa, struct x86_exception *exception, | |
5596 | bool write) | |
5597 | { | |
97d64b78 AK |
5598 | u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0) |
5599 | | (write ? PFERR_WRITE_MASK : 0); | |
af7cc7d1 | 5600 | |
be94f6b7 HH |
5601 | /* |
5602 | * currently PKRU is only applied to ept enabled guest so | |
5603 | * there is no pkey in EPT page table for L1 guest or EPT | |
5604 | * shadow page table for L2 guest. | |
5605 | */ | |
97d64b78 | 5606 | if (vcpu_match_mmio_gva(vcpu, gva) |
97ec8c06 | 5607 | && !permission_fault(vcpu, vcpu->arch.walk_mmu, |
871bd034 | 5608 | vcpu->arch.mmio_access, 0, access)) { |
bebb106a XG |
5609 | *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT | |
5610 | (gva & (PAGE_SIZE - 1)); | |
4f022648 | 5611 | trace_vcpu_match_mmio(gva, *gpa, write, false); |
bebb106a XG |
5612 | return 1; |
5613 | } | |
5614 | ||
af7cc7d1 XG |
5615 | *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception); |
5616 | ||
5617 | if (*gpa == UNMAPPED_GVA) | |
5618 | return -1; | |
5619 | ||
0f89b207 | 5620 | return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write); |
af7cc7d1 XG |
5621 | } |
5622 | ||
3200f405 | 5623 | int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa, |
bcc55cba | 5624 | const void *val, int bytes) |
bbd9b64e CO |
5625 | { |
5626 | int ret; | |
5627 | ||
54bf36aa | 5628 | ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes); |
9f811285 | 5629 | if (ret < 0) |
bbd9b64e | 5630 | return 0; |
0eb05bf2 | 5631 | kvm_page_track_write(vcpu, gpa, val, bytes); |
bbd9b64e CO |
5632 | return 1; |
5633 | } | |
5634 | ||
77d197b2 XG |
5635 | struct read_write_emulator_ops { |
5636 | int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val, | |
5637 | int bytes); | |
5638 | int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa, | |
5639 | void *val, int bytes); | |
5640 | int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa, | |
5641 | int bytes, void *val); | |
5642 | int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa, | |
5643 | void *val, int bytes); | |
5644 | bool write; | |
5645 | }; | |
5646 | ||
5647 | static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes) | |
5648 | { | |
5649 | if (vcpu->mmio_read_completed) { | |
77d197b2 | 5650 | trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, |
e39d200f | 5651 | vcpu->mmio_fragments[0].gpa, val); |
77d197b2 XG |
5652 | vcpu->mmio_read_completed = 0; |
5653 | return 1; | |
5654 | } | |
5655 | ||
5656 | return 0; | |
5657 | } | |
5658 | ||
5659 | static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa, | |
5660 | void *val, int bytes) | |
5661 | { | |
54bf36aa | 5662 | return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes); |
77d197b2 XG |
5663 | } |
5664 | ||
5665 | static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa, | |
5666 | void *val, int bytes) | |
5667 | { | |
5668 | return emulator_write_phys(vcpu, gpa, val, bytes); | |
5669 | } | |
5670 | ||
5671 | static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val) | |
5672 | { | |
e39d200f | 5673 | trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val); |
77d197b2 XG |
5674 | return vcpu_mmio_write(vcpu, gpa, bytes, val); |
5675 | } | |
5676 | ||
5677 | static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, | |
5678 | void *val, int bytes) | |
5679 | { | |
e39d200f | 5680 | trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL); |
77d197b2 XG |
5681 | return X86EMUL_IO_NEEDED; |
5682 | } | |
5683 | ||
5684 | static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, | |
5685 | void *val, int bytes) | |
5686 | { | |
f78146b0 AK |
5687 | struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0]; |
5688 | ||
87da7e66 | 5689 | memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len)); |
77d197b2 XG |
5690 | return X86EMUL_CONTINUE; |
5691 | } | |
5692 | ||
0fbe9b0b | 5693 | static const struct read_write_emulator_ops read_emultor = { |
77d197b2 XG |
5694 | .read_write_prepare = read_prepare, |
5695 | .read_write_emulate = read_emulate, | |
5696 | .read_write_mmio = vcpu_mmio_read, | |
5697 | .read_write_exit_mmio = read_exit_mmio, | |
5698 | }; | |
5699 | ||
0fbe9b0b | 5700 | static const struct read_write_emulator_ops write_emultor = { |
77d197b2 XG |
5701 | .read_write_emulate = write_emulate, |
5702 | .read_write_mmio = write_mmio, | |
5703 | .read_write_exit_mmio = write_exit_mmio, | |
5704 | .write = true, | |
5705 | }; | |
5706 | ||
22388a3c XG |
5707 | static int emulator_read_write_onepage(unsigned long addr, void *val, |
5708 | unsigned int bytes, | |
5709 | struct x86_exception *exception, | |
5710 | struct kvm_vcpu *vcpu, | |
0fbe9b0b | 5711 | const struct read_write_emulator_ops *ops) |
bbd9b64e | 5712 | { |
af7cc7d1 XG |
5713 | gpa_t gpa; |
5714 | int handled, ret; | |
22388a3c | 5715 | bool write = ops->write; |
f78146b0 | 5716 | struct kvm_mmio_fragment *frag; |
0f89b207 TL |
5717 | struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; |
5718 | ||
5719 | /* | |
5720 | * If the exit was due to a NPF we may already have a GPA. | |
5721 | * If the GPA is present, use it to avoid the GVA to GPA table walk. | |
5722 | * Note, this cannot be used on string operations since string | |
5723 | * operation using rep will only have the initial GPA from the NPF | |
5724 | * occurred. | |
5725 | */ | |
5726 | if (vcpu->arch.gpa_available && | |
5727 | emulator_can_use_gpa(ctxt) && | |
618232e2 BS |
5728 | (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) { |
5729 | gpa = vcpu->arch.gpa_val; | |
5730 | ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write); | |
5731 | } else { | |
5732 | ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write); | |
5733 | if (ret < 0) | |
5734 | return X86EMUL_PROPAGATE_FAULT; | |
0f89b207 | 5735 | } |
10589a46 | 5736 | |
618232e2 | 5737 | if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes)) |
bbd9b64e CO |
5738 | return X86EMUL_CONTINUE; |
5739 | ||
bbd9b64e CO |
5740 | /* |
5741 | * Is this MMIO handled locally? | |
5742 | */ | |
22388a3c | 5743 | handled = ops->read_write_mmio(vcpu, gpa, bytes, val); |
70252a10 | 5744 | if (handled == bytes) |
bbd9b64e | 5745 | return X86EMUL_CONTINUE; |
bbd9b64e | 5746 | |
70252a10 AK |
5747 | gpa += handled; |
5748 | bytes -= handled; | |
5749 | val += handled; | |
5750 | ||
87da7e66 XG |
5751 | WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS); |
5752 | frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++]; | |
5753 | frag->gpa = gpa; | |
5754 | frag->data = val; | |
5755 | frag->len = bytes; | |
f78146b0 | 5756 | return X86EMUL_CONTINUE; |
bbd9b64e CO |
5757 | } |
5758 | ||
52eb5a6d XL |
5759 | static int emulator_read_write(struct x86_emulate_ctxt *ctxt, |
5760 | unsigned long addr, | |
22388a3c XG |
5761 | void *val, unsigned int bytes, |
5762 | struct x86_exception *exception, | |
0fbe9b0b | 5763 | const struct read_write_emulator_ops *ops) |
bbd9b64e | 5764 | { |
0f65dd70 | 5765 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
f78146b0 AK |
5766 | gpa_t gpa; |
5767 | int rc; | |
5768 | ||
5769 | if (ops->read_write_prepare && | |
5770 | ops->read_write_prepare(vcpu, val, bytes)) | |
5771 | return X86EMUL_CONTINUE; | |
5772 | ||
5773 | vcpu->mmio_nr_fragments = 0; | |
0f65dd70 | 5774 | |
bbd9b64e CO |
5775 | /* Crossing a page boundary? */ |
5776 | if (((addr + bytes - 1) ^ addr) & PAGE_MASK) { | |
f78146b0 | 5777 | int now; |
bbd9b64e CO |
5778 | |
5779 | now = -addr & ~PAGE_MASK; | |
22388a3c XG |
5780 | rc = emulator_read_write_onepage(addr, val, now, exception, |
5781 | vcpu, ops); | |
5782 | ||
bbd9b64e CO |
5783 | if (rc != X86EMUL_CONTINUE) |
5784 | return rc; | |
5785 | addr += now; | |
bac15531 NA |
5786 | if (ctxt->mode != X86EMUL_MODE_PROT64) |
5787 | addr = (u32)addr; | |
bbd9b64e CO |
5788 | val += now; |
5789 | bytes -= now; | |
5790 | } | |
22388a3c | 5791 | |
f78146b0 AK |
5792 | rc = emulator_read_write_onepage(addr, val, bytes, exception, |
5793 | vcpu, ops); | |
5794 | if (rc != X86EMUL_CONTINUE) | |
5795 | return rc; | |
5796 | ||
5797 | if (!vcpu->mmio_nr_fragments) | |
5798 | return rc; | |
5799 | ||
5800 | gpa = vcpu->mmio_fragments[0].gpa; | |
5801 | ||
5802 | vcpu->mmio_needed = 1; | |
5803 | vcpu->mmio_cur_fragment = 0; | |
5804 | ||
87da7e66 | 5805 | vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len); |
f78146b0 AK |
5806 | vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write; |
5807 | vcpu->run->exit_reason = KVM_EXIT_MMIO; | |
5808 | vcpu->run->mmio.phys_addr = gpa; | |
5809 | ||
5810 | return ops->read_write_exit_mmio(vcpu, gpa, val, bytes); | |
22388a3c XG |
5811 | } |
5812 | ||
5813 | static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt, | |
5814 | unsigned long addr, | |
5815 | void *val, | |
5816 | unsigned int bytes, | |
5817 | struct x86_exception *exception) | |
5818 | { | |
5819 | return emulator_read_write(ctxt, addr, val, bytes, | |
5820 | exception, &read_emultor); | |
5821 | } | |
5822 | ||
52eb5a6d | 5823 | static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt, |
22388a3c XG |
5824 | unsigned long addr, |
5825 | const void *val, | |
5826 | unsigned int bytes, | |
5827 | struct x86_exception *exception) | |
5828 | { | |
5829 | return emulator_read_write(ctxt, addr, (void *)val, bytes, | |
5830 | exception, &write_emultor); | |
bbd9b64e | 5831 | } |
bbd9b64e | 5832 | |
daea3e73 AK |
5833 | #define CMPXCHG_TYPE(t, ptr, old, new) \ |
5834 | (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old)) | |
5835 | ||
5836 | #ifdef CONFIG_X86_64 | |
5837 | # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new) | |
5838 | #else | |
5839 | # define CMPXCHG64(ptr, old, new) \ | |
9749a6c0 | 5840 | (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old)) |
daea3e73 AK |
5841 | #endif |
5842 | ||
0f65dd70 AK |
5843 | static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt, |
5844 | unsigned long addr, | |
bbd9b64e CO |
5845 | const void *old, |
5846 | const void *new, | |
5847 | unsigned int bytes, | |
0f65dd70 | 5848 | struct x86_exception *exception) |
bbd9b64e | 5849 | { |
42e35f80 | 5850 | struct kvm_host_map map; |
0f65dd70 | 5851 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
daea3e73 | 5852 | gpa_t gpa; |
daea3e73 AK |
5853 | char *kaddr; |
5854 | bool exchanged; | |
2bacc55c | 5855 | |
daea3e73 AK |
5856 | /* guests cmpxchg8b have to be emulated atomically */ |
5857 | if (bytes > 8 || (bytes & (bytes - 1))) | |
5858 | goto emul_write; | |
10589a46 | 5859 | |
daea3e73 | 5860 | gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL); |
2bacc55c | 5861 | |
daea3e73 AK |
5862 | if (gpa == UNMAPPED_GVA || |
5863 | (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE) | |
5864 | goto emul_write; | |
2bacc55c | 5865 | |
daea3e73 AK |
5866 | if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK)) |
5867 | goto emul_write; | |
72dc67a6 | 5868 | |
42e35f80 | 5869 | if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map)) |
c19b8bd6 | 5870 | goto emul_write; |
72dc67a6 | 5871 | |
42e35f80 KA |
5872 | kaddr = map.hva + offset_in_page(gpa); |
5873 | ||
daea3e73 AK |
5874 | switch (bytes) { |
5875 | case 1: | |
5876 | exchanged = CMPXCHG_TYPE(u8, kaddr, old, new); | |
5877 | break; | |
5878 | case 2: | |
5879 | exchanged = CMPXCHG_TYPE(u16, kaddr, old, new); | |
5880 | break; | |
5881 | case 4: | |
5882 | exchanged = CMPXCHG_TYPE(u32, kaddr, old, new); | |
5883 | break; | |
5884 | case 8: | |
5885 | exchanged = CMPXCHG64(kaddr, old, new); | |
5886 | break; | |
5887 | default: | |
5888 | BUG(); | |
2bacc55c | 5889 | } |
42e35f80 KA |
5890 | |
5891 | kvm_vcpu_unmap(vcpu, &map, true); | |
daea3e73 AK |
5892 | |
5893 | if (!exchanged) | |
5894 | return X86EMUL_CMPXCHG_FAILED; | |
5895 | ||
0eb05bf2 | 5896 | kvm_page_track_write(vcpu, gpa, new, bytes); |
8f6abd06 GN |
5897 | |
5898 | return X86EMUL_CONTINUE; | |
4a5f48f6 | 5899 | |
3200f405 | 5900 | emul_write: |
daea3e73 | 5901 | printk_once(KERN_WARNING "kvm: emulating exchange as write\n"); |
2bacc55c | 5902 | |
0f65dd70 | 5903 | return emulator_write_emulated(ctxt, addr, new, bytes, exception); |
bbd9b64e CO |
5904 | } |
5905 | ||
cf8f70bf GN |
5906 | static int kernel_pio(struct kvm_vcpu *vcpu, void *pd) |
5907 | { | |
cbfc6c91 | 5908 | int r = 0, i; |
cf8f70bf | 5909 | |
cbfc6c91 WL |
5910 | for (i = 0; i < vcpu->arch.pio.count; i++) { |
5911 | if (vcpu->arch.pio.in) | |
5912 | r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port, | |
5913 | vcpu->arch.pio.size, pd); | |
5914 | else | |
5915 | r = kvm_io_bus_write(vcpu, KVM_PIO_BUS, | |
5916 | vcpu->arch.pio.port, vcpu->arch.pio.size, | |
5917 | pd); | |
5918 | if (r) | |
5919 | break; | |
5920 | pd += vcpu->arch.pio.size; | |
5921 | } | |
cf8f70bf GN |
5922 | return r; |
5923 | } | |
5924 | ||
6f6fbe98 XG |
5925 | static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size, |
5926 | unsigned short port, void *val, | |
5927 | unsigned int count, bool in) | |
cf8f70bf | 5928 | { |
cf8f70bf | 5929 | vcpu->arch.pio.port = port; |
6f6fbe98 | 5930 | vcpu->arch.pio.in = in; |
7972995b | 5931 | vcpu->arch.pio.count = count; |
cf8f70bf GN |
5932 | vcpu->arch.pio.size = size; |
5933 | ||
5934 | if (!kernel_pio(vcpu, vcpu->arch.pio_data)) { | |
7972995b | 5935 | vcpu->arch.pio.count = 0; |
cf8f70bf GN |
5936 | return 1; |
5937 | } | |
5938 | ||
5939 | vcpu->run->exit_reason = KVM_EXIT_IO; | |
6f6fbe98 | 5940 | vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT; |
cf8f70bf GN |
5941 | vcpu->run->io.size = size; |
5942 | vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE; | |
5943 | vcpu->run->io.count = count; | |
5944 | vcpu->run->io.port = port; | |
5945 | ||
5946 | return 0; | |
5947 | } | |
5948 | ||
6f6fbe98 XG |
5949 | static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt, |
5950 | int size, unsigned short port, void *val, | |
5951 | unsigned int count) | |
cf8f70bf | 5952 | { |
ca1d4a9e | 5953 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
6f6fbe98 | 5954 | int ret; |
ca1d4a9e | 5955 | |
6f6fbe98 XG |
5956 | if (vcpu->arch.pio.count) |
5957 | goto data_avail; | |
cf8f70bf | 5958 | |
cbfc6c91 WL |
5959 | memset(vcpu->arch.pio_data, 0, size * count); |
5960 | ||
6f6fbe98 XG |
5961 | ret = emulator_pio_in_out(vcpu, size, port, val, count, true); |
5962 | if (ret) { | |
5963 | data_avail: | |
5964 | memcpy(val, vcpu->arch.pio_data, size * count); | |
1171903d | 5965 | trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data); |
7972995b | 5966 | vcpu->arch.pio.count = 0; |
cf8f70bf GN |
5967 | return 1; |
5968 | } | |
5969 | ||
cf8f70bf GN |
5970 | return 0; |
5971 | } | |
5972 | ||
6f6fbe98 XG |
5973 | static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt, |
5974 | int size, unsigned short port, | |
5975 | const void *val, unsigned int count) | |
5976 | { | |
5977 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
5978 | ||
5979 | memcpy(vcpu->arch.pio_data, val, size * count); | |
1171903d | 5980 | trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data); |
6f6fbe98 XG |
5981 | return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false); |
5982 | } | |
5983 | ||
bbd9b64e CO |
5984 | static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg) |
5985 | { | |
5986 | return kvm_x86_ops->get_segment_base(vcpu, seg); | |
5987 | } | |
5988 | ||
3cb16fe7 | 5989 | static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address) |
bbd9b64e | 5990 | { |
3cb16fe7 | 5991 | kvm_mmu_invlpg(emul_to_vcpu(ctxt), address); |
bbd9b64e CO |
5992 | } |
5993 | ||
ae6a2375 | 5994 | static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu) |
f5f48ee1 SY |
5995 | { |
5996 | if (!need_emulate_wbinvd(vcpu)) | |
5997 | return X86EMUL_CONTINUE; | |
5998 | ||
5999 | if (kvm_x86_ops->has_wbinvd_exit()) { | |
2eec7343 JK |
6000 | int cpu = get_cpu(); |
6001 | ||
6002 | cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask); | |
f5f48ee1 SY |
6003 | smp_call_function_many(vcpu->arch.wbinvd_dirty_mask, |
6004 | wbinvd_ipi, NULL, 1); | |
2eec7343 | 6005 | put_cpu(); |
f5f48ee1 | 6006 | cpumask_clear(vcpu->arch.wbinvd_dirty_mask); |
2eec7343 JK |
6007 | } else |
6008 | wbinvd(); | |
f5f48ee1 SY |
6009 | return X86EMUL_CONTINUE; |
6010 | } | |
5cb56059 JS |
6011 | |
6012 | int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu) | |
6013 | { | |
6affcbed KH |
6014 | kvm_emulate_wbinvd_noskip(vcpu); |
6015 | return kvm_skip_emulated_instruction(vcpu); | |
5cb56059 | 6016 | } |
f5f48ee1 SY |
6017 | EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd); |
6018 | ||
5cb56059 JS |
6019 | |
6020 | ||
bcaf5cc5 AK |
6021 | static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt) |
6022 | { | |
5cb56059 | 6023 | kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt)); |
bcaf5cc5 AK |
6024 | } |
6025 | ||
52eb5a6d XL |
6026 | static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, |
6027 | unsigned long *dest) | |
bbd9b64e | 6028 | { |
16f8a6f9 | 6029 | return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest); |
bbd9b64e CO |
6030 | } |
6031 | ||
52eb5a6d XL |
6032 | static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, |
6033 | unsigned long value) | |
bbd9b64e | 6034 | { |
338dbc97 | 6035 | |
717746e3 | 6036 | return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value); |
bbd9b64e CO |
6037 | } |
6038 | ||
52a46617 | 6039 | static u64 mk_cr_64(u64 curr_cr, u32 new_val) |
5fdbf976 | 6040 | { |
52a46617 | 6041 | return (curr_cr & ~((1ULL << 32) - 1)) | new_val; |
5fdbf976 MT |
6042 | } |
6043 | ||
717746e3 | 6044 | static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr) |
bbd9b64e | 6045 | { |
717746e3 | 6046 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
52a46617 GN |
6047 | unsigned long value; |
6048 | ||
6049 | switch (cr) { | |
6050 | case 0: | |
6051 | value = kvm_read_cr0(vcpu); | |
6052 | break; | |
6053 | case 2: | |
6054 | value = vcpu->arch.cr2; | |
6055 | break; | |
6056 | case 3: | |
9f8fe504 | 6057 | value = kvm_read_cr3(vcpu); |
52a46617 GN |
6058 | break; |
6059 | case 4: | |
6060 | value = kvm_read_cr4(vcpu); | |
6061 | break; | |
6062 | case 8: | |
6063 | value = kvm_get_cr8(vcpu); | |
6064 | break; | |
6065 | default: | |
a737f256 | 6066 | kvm_err("%s: unexpected cr %u\n", __func__, cr); |
52a46617 GN |
6067 | return 0; |
6068 | } | |
6069 | ||
6070 | return value; | |
6071 | } | |
6072 | ||
717746e3 | 6073 | static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val) |
52a46617 | 6074 | { |
717746e3 | 6075 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
0f12244f GN |
6076 | int res = 0; |
6077 | ||
52a46617 GN |
6078 | switch (cr) { |
6079 | case 0: | |
49a9b07e | 6080 | res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val)); |
52a46617 GN |
6081 | break; |
6082 | case 2: | |
6083 | vcpu->arch.cr2 = val; | |
6084 | break; | |
6085 | case 3: | |
2390218b | 6086 | res = kvm_set_cr3(vcpu, val); |
52a46617 GN |
6087 | break; |
6088 | case 4: | |
a83b29c6 | 6089 | res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val)); |
52a46617 GN |
6090 | break; |
6091 | case 8: | |
eea1cff9 | 6092 | res = kvm_set_cr8(vcpu, val); |
52a46617 GN |
6093 | break; |
6094 | default: | |
a737f256 | 6095 | kvm_err("%s: unexpected cr %u\n", __func__, cr); |
0f12244f | 6096 | res = -1; |
52a46617 | 6097 | } |
0f12244f GN |
6098 | |
6099 | return res; | |
52a46617 GN |
6100 | } |
6101 | ||
717746e3 | 6102 | static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt) |
9c537244 | 6103 | { |
717746e3 | 6104 | return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt)); |
9c537244 GN |
6105 | } |
6106 | ||
4bff1e86 | 6107 | static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) |
2dafc6c2 | 6108 | { |
4bff1e86 | 6109 | kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt); |
2dafc6c2 GN |
6110 | } |
6111 | ||
4bff1e86 | 6112 | static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) |
160ce1f1 | 6113 | { |
4bff1e86 | 6114 | kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt); |
160ce1f1 MG |
6115 | } |
6116 | ||
1ac9d0cf AK |
6117 | static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) |
6118 | { | |
6119 | kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt); | |
6120 | } | |
6121 | ||
6122 | static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt) | |
6123 | { | |
6124 | kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt); | |
6125 | } | |
6126 | ||
4bff1e86 AK |
6127 | static unsigned long emulator_get_cached_segment_base( |
6128 | struct x86_emulate_ctxt *ctxt, int seg) | |
5951c442 | 6129 | { |
4bff1e86 | 6130 | return get_segment_base(emul_to_vcpu(ctxt), seg); |
5951c442 GN |
6131 | } |
6132 | ||
1aa36616 AK |
6133 | static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector, |
6134 | struct desc_struct *desc, u32 *base3, | |
6135 | int seg) | |
2dafc6c2 GN |
6136 | { |
6137 | struct kvm_segment var; | |
6138 | ||
4bff1e86 | 6139 | kvm_get_segment(emul_to_vcpu(ctxt), &var, seg); |
1aa36616 | 6140 | *selector = var.selector; |
2dafc6c2 | 6141 | |
378a8b09 GN |
6142 | if (var.unusable) { |
6143 | memset(desc, 0, sizeof(*desc)); | |
f0367ee1 RK |
6144 | if (base3) |
6145 | *base3 = 0; | |
2dafc6c2 | 6146 | return false; |
378a8b09 | 6147 | } |
2dafc6c2 GN |
6148 | |
6149 | if (var.g) | |
6150 | var.limit >>= 12; | |
6151 | set_desc_limit(desc, var.limit); | |
6152 | set_desc_base(desc, (unsigned long)var.base); | |
5601d05b GN |
6153 | #ifdef CONFIG_X86_64 |
6154 | if (base3) | |
6155 | *base3 = var.base >> 32; | |
6156 | #endif | |
2dafc6c2 GN |
6157 | desc->type = var.type; |
6158 | desc->s = var.s; | |
6159 | desc->dpl = var.dpl; | |
6160 | desc->p = var.present; | |
6161 | desc->avl = var.avl; | |
6162 | desc->l = var.l; | |
6163 | desc->d = var.db; | |
6164 | desc->g = var.g; | |
6165 | ||
6166 | return true; | |
6167 | } | |
6168 | ||
1aa36616 AK |
6169 | static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector, |
6170 | struct desc_struct *desc, u32 base3, | |
6171 | int seg) | |
2dafc6c2 | 6172 | { |
4bff1e86 | 6173 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
2dafc6c2 GN |
6174 | struct kvm_segment var; |
6175 | ||
1aa36616 | 6176 | var.selector = selector; |
2dafc6c2 | 6177 | var.base = get_desc_base(desc); |
5601d05b GN |
6178 | #ifdef CONFIG_X86_64 |
6179 | var.base |= ((u64)base3) << 32; | |
6180 | #endif | |
2dafc6c2 GN |
6181 | var.limit = get_desc_limit(desc); |
6182 | if (desc->g) | |
6183 | var.limit = (var.limit << 12) | 0xfff; | |
6184 | var.type = desc->type; | |
2dafc6c2 GN |
6185 | var.dpl = desc->dpl; |
6186 | var.db = desc->d; | |
6187 | var.s = desc->s; | |
6188 | var.l = desc->l; | |
6189 | var.g = desc->g; | |
6190 | var.avl = desc->avl; | |
6191 | var.present = desc->p; | |
6192 | var.unusable = !var.present; | |
6193 | var.padding = 0; | |
6194 | ||
6195 | kvm_set_segment(vcpu, &var, seg); | |
6196 | return; | |
6197 | } | |
6198 | ||
717746e3 AK |
6199 | static int emulator_get_msr(struct x86_emulate_ctxt *ctxt, |
6200 | u32 msr_index, u64 *pdata) | |
6201 | { | |
f20935d8 | 6202 | return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata); |
717746e3 AK |
6203 | } |
6204 | ||
6205 | static int emulator_set_msr(struct x86_emulate_ctxt *ctxt, | |
6206 | u32 msr_index, u64 data) | |
6207 | { | |
f20935d8 | 6208 | return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data); |
717746e3 AK |
6209 | } |
6210 | ||
64d60670 PB |
6211 | static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt) |
6212 | { | |
6213 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
6214 | ||
6215 | return vcpu->arch.smbase; | |
6216 | } | |
6217 | ||
6218 | static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase) | |
6219 | { | |
6220 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
6221 | ||
6222 | vcpu->arch.smbase = smbase; | |
6223 | } | |
6224 | ||
67f4d428 NA |
6225 | static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt, |
6226 | u32 pmc) | |
6227 | { | |
98ff80f5 | 6228 | return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc); |
67f4d428 NA |
6229 | } |
6230 | ||
222d21aa AK |
6231 | static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt, |
6232 | u32 pmc, u64 *pdata) | |
6233 | { | |
c6702c9d | 6234 | return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata); |
222d21aa AK |
6235 | } |
6236 | ||
6c3287f7 AK |
6237 | static void emulator_halt(struct x86_emulate_ctxt *ctxt) |
6238 | { | |
6239 | emul_to_vcpu(ctxt)->arch.halt_request = 1; | |
6240 | } | |
6241 | ||
2953538e | 6242 | static int emulator_intercept(struct x86_emulate_ctxt *ctxt, |
8a76d7f2 | 6243 | struct x86_instruction_info *info, |
c4f035c6 AK |
6244 | enum x86_intercept_stage stage) |
6245 | { | |
2953538e | 6246 | return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage); |
c4f035c6 AK |
6247 | } |
6248 | ||
e911eb3b YZ |
6249 | static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt, |
6250 | u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit) | |
bdb42f5a | 6251 | { |
e911eb3b | 6252 | return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit); |
bdb42f5a SB |
6253 | } |
6254 | ||
5ae78e95 SC |
6255 | static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt) |
6256 | { | |
6257 | return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM); | |
6258 | } | |
6259 | ||
6260 | static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt) | |
6261 | { | |
6262 | return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE); | |
6263 | } | |
6264 | ||
6265 | static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt) | |
6266 | { | |
6267 | return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR); | |
6268 | } | |
6269 | ||
dd856efa AK |
6270 | static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg) |
6271 | { | |
6272 | return kvm_register_read(emul_to_vcpu(ctxt), reg); | |
6273 | } | |
6274 | ||
6275 | static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val) | |
6276 | { | |
6277 | kvm_register_write(emul_to_vcpu(ctxt), reg, val); | |
6278 | } | |
6279 | ||
801806d9 NA |
6280 | static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked) |
6281 | { | |
6282 | kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked); | |
6283 | } | |
6284 | ||
6ed071f0 LP |
6285 | static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt) |
6286 | { | |
6287 | return emul_to_vcpu(ctxt)->arch.hflags; | |
6288 | } | |
6289 | ||
6290 | static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags) | |
6291 | { | |
c5833c7a | 6292 | emul_to_vcpu(ctxt)->arch.hflags = emul_flags; |
6ed071f0 LP |
6293 | } |
6294 | ||
ed19321f SC |
6295 | static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, |
6296 | const char *smstate) | |
0234bf88 | 6297 | { |
ed19321f | 6298 | return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smstate); |
0234bf88 LP |
6299 | } |
6300 | ||
c5833c7a SC |
6301 | static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt) |
6302 | { | |
6303 | kvm_smm_changed(emul_to_vcpu(ctxt)); | |
6304 | } | |
6305 | ||
02d4160f VK |
6306 | static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr) |
6307 | { | |
6308 | return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr); | |
6309 | } | |
6310 | ||
0225fb50 | 6311 | static const struct x86_emulate_ops emulate_ops = { |
dd856efa AK |
6312 | .read_gpr = emulator_read_gpr, |
6313 | .write_gpr = emulator_write_gpr, | |
ce14e868 PB |
6314 | .read_std = emulator_read_std, |
6315 | .write_std = emulator_write_std, | |
7a036a6f | 6316 | .read_phys = kvm_read_guest_phys_system, |
1871c602 | 6317 | .fetch = kvm_fetch_guest_virt, |
bbd9b64e CO |
6318 | .read_emulated = emulator_read_emulated, |
6319 | .write_emulated = emulator_write_emulated, | |
6320 | .cmpxchg_emulated = emulator_cmpxchg_emulated, | |
3cb16fe7 | 6321 | .invlpg = emulator_invlpg, |
cf8f70bf GN |
6322 | .pio_in_emulated = emulator_pio_in_emulated, |
6323 | .pio_out_emulated = emulator_pio_out_emulated, | |
1aa36616 AK |
6324 | .get_segment = emulator_get_segment, |
6325 | .set_segment = emulator_set_segment, | |
5951c442 | 6326 | .get_cached_segment_base = emulator_get_cached_segment_base, |
2dafc6c2 | 6327 | .get_gdt = emulator_get_gdt, |
160ce1f1 | 6328 | .get_idt = emulator_get_idt, |
1ac9d0cf AK |
6329 | .set_gdt = emulator_set_gdt, |
6330 | .set_idt = emulator_set_idt, | |
52a46617 GN |
6331 | .get_cr = emulator_get_cr, |
6332 | .set_cr = emulator_set_cr, | |
9c537244 | 6333 | .cpl = emulator_get_cpl, |
35aa5375 GN |
6334 | .get_dr = emulator_get_dr, |
6335 | .set_dr = emulator_set_dr, | |
64d60670 PB |
6336 | .get_smbase = emulator_get_smbase, |
6337 | .set_smbase = emulator_set_smbase, | |
717746e3 AK |
6338 | .set_msr = emulator_set_msr, |
6339 | .get_msr = emulator_get_msr, | |
67f4d428 | 6340 | .check_pmc = emulator_check_pmc, |
222d21aa | 6341 | .read_pmc = emulator_read_pmc, |
6c3287f7 | 6342 | .halt = emulator_halt, |
bcaf5cc5 | 6343 | .wbinvd = emulator_wbinvd, |
d6aa1000 | 6344 | .fix_hypercall = emulator_fix_hypercall, |
c4f035c6 | 6345 | .intercept = emulator_intercept, |
bdb42f5a | 6346 | .get_cpuid = emulator_get_cpuid, |
5ae78e95 SC |
6347 | .guest_has_long_mode = emulator_guest_has_long_mode, |
6348 | .guest_has_movbe = emulator_guest_has_movbe, | |
6349 | .guest_has_fxsr = emulator_guest_has_fxsr, | |
801806d9 | 6350 | .set_nmi_mask = emulator_set_nmi_mask, |
6ed071f0 LP |
6351 | .get_hflags = emulator_get_hflags, |
6352 | .set_hflags = emulator_set_hflags, | |
0234bf88 | 6353 | .pre_leave_smm = emulator_pre_leave_smm, |
c5833c7a | 6354 | .post_leave_smm = emulator_post_leave_smm, |
02d4160f | 6355 | .set_xcr = emulator_set_xcr, |
bbd9b64e CO |
6356 | }; |
6357 | ||
95cb2295 GN |
6358 | static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask) |
6359 | { | |
37ccdcbe | 6360 | u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu); |
95cb2295 GN |
6361 | /* |
6362 | * an sti; sti; sequence only disable interrupts for the first | |
6363 | * instruction. So, if the last instruction, be it emulated or | |
6364 | * not, left the system with the INT_STI flag enabled, it | |
6365 | * means that the last instruction is an sti. We should not | |
6366 | * leave the flag on in this case. The same goes for mov ss | |
6367 | */ | |
37ccdcbe PB |
6368 | if (int_shadow & mask) |
6369 | mask = 0; | |
6addfc42 | 6370 | if (unlikely(int_shadow || mask)) { |
95cb2295 | 6371 | kvm_x86_ops->set_interrupt_shadow(vcpu, mask); |
6addfc42 PB |
6372 | if (!mask) |
6373 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
6374 | } | |
95cb2295 GN |
6375 | } |
6376 | ||
ef54bcfe | 6377 | static bool inject_emulated_exception(struct kvm_vcpu *vcpu) |
54b8486f GN |
6378 | { |
6379 | struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; | |
da9cb575 | 6380 | if (ctxt->exception.vector == PF_VECTOR) |
ef54bcfe PB |
6381 | return kvm_propagate_fault(vcpu, &ctxt->exception); |
6382 | ||
6383 | if (ctxt->exception.error_code_valid) | |
da9cb575 AK |
6384 | kvm_queue_exception_e(vcpu, ctxt->exception.vector, |
6385 | ctxt->exception.error_code); | |
54b8486f | 6386 | else |
da9cb575 | 6387 | kvm_queue_exception(vcpu, ctxt->exception.vector); |
ef54bcfe | 6388 | return false; |
54b8486f GN |
6389 | } |
6390 | ||
8ec4722d MG |
6391 | static void init_emulate_ctxt(struct kvm_vcpu *vcpu) |
6392 | { | |
adf52235 | 6393 | struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; |
8ec4722d MG |
6394 | int cs_db, cs_l; |
6395 | ||
8ec4722d MG |
6396 | kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l); |
6397 | ||
adf52235 | 6398 | ctxt->eflags = kvm_get_rflags(vcpu); |
c8401dda PB |
6399 | ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0; |
6400 | ||
adf52235 TY |
6401 | ctxt->eip = kvm_rip_read(vcpu); |
6402 | ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL : | |
6403 | (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 : | |
42bf549f | 6404 | (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 : |
adf52235 TY |
6405 | cs_db ? X86EMUL_MODE_PROT32 : |
6406 | X86EMUL_MODE_PROT16; | |
a584539b | 6407 | BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK); |
64d60670 PB |
6408 | BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK); |
6409 | BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK); | |
adf52235 | 6410 | |
dd856efa | 6411 | init_decode_cache(ctxt); |
7ae441ea | 6412 | vcpu->arch.emulate_regs_need_sync_from_vcpu = false; |
8ec4722d MG |
6413 | } |
6414 | ||
9497e1f2 | 6415 | void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip) |
63995653 | 6416 | { |
9d74191a | 6417 | struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; |
63995653 MG |
6418 | int ret; |
6419 | ||
6420 | init_emulate_ctxt(vcpu); | |
6421 | ||
9dac77fa AK |
6422 | ctxt->op_bytes = 2; |
6423 | ctxt->ad_bytes = 2; | |
6424 | ctxt->_eip = ctxt->eip + inc_eip; | |
9d74191a | 6425 | ret = emulate_int_real(ctxt, irq); |
63995653 | 6426 | |
9497e1f2 SC |
6427 | if (ret != X86EMUL_CONTINUE) { |
6428 | kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu); | |
6429 | } else { | |
6430 | ctxt->eip = ctxt->_eip; | |
6431 | kvm_rip_write(vcpu, ctxt->eip); | |
6432 | kvm_set_rflags(vcpu, ctxt->eflags); | |
6433 | } | |
63995653 MG |
6434 | } |
6435 | EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt); | |
6436 | ||
e2366171 | 6437 | static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type) |
6d77dbfc | 6438 | { |
6d77dbfc GN |
6439 | ++vcpu->stat.insn_emulation_fail; |
6440 | trace_kvm_emulate_insn_failed(vcpu); | |
e2366171 | 6441 | |
42cbf068 SC |
6442 | if (emulation_type & EMULTYPE_VMWARE_GP) { |
6443 | kvm_queue_exception_e(vcpu, GP_VECTOR, 0); | |
60fc3d02 | 6444 | return 1; |
42cbf068 | 6445 | } |
e2366171 | 6446 | |
738fece4 SC |
6447 | if (emulation_type & EMULTYPE_SKIP) { |
6448 | vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
6449 | vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
6450 | vcpu->run->internal.ndata = 0; | |
60fc3d02 | 6451 | return 0; |
738fece4 SC |
6452 | } |
6453 | ||
22da61c9 SC |
6454 | kvm_queue_exception(vcpu, UD_VECTOR); |
6455 | ||
a2b9e6c1 | 6456 | if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) { |
fc3a9157 JR |
6457 | vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; |
6458 | vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
6459 | vcpu->run->internal.ndata = 0; | |
60fc3d02 | 6460 | return 0; |
fc3a9157 | 6461 | } |
e2366171 | 6462 | |
60fc3d02 | 6463 | return 1; |
6d77dbfc GN |
6464 | } |
6465 | ||
736c291c | 6466 | static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa, |
991eebf9 GN |
6467 | bool write_fault_to_shadow_pgtable, |
6468 | int emulation_type) | |
a6f177ef | 6469 | { |
736c291c | 6470 | gpa_t gpa = cr2_or_gpa; |
ba049e93 | 6471 | kvm_pfn_t pfn; |
a6f177ef | 6472 | |
384bf221 | 6473 | if (!(emulation_type & EMULTYPE_ALLOW_RETRY)) |
991eebf9 GN |
6474 | return false; |
6475 | ||
6c3dfeb6 SC |
6476 | if (WARN_ON_ONCE(is_guest_mode(vcpu))) |
6477 | return false; | |
6478 | ||
44dd3ffa | 6479 | if (!vcpu->arch.mmu->direct_map) { |
95b3cf69 XG |
6480 | /* |
6481 | * Write permission should be allowed since only | |
6482 | * write access need to be emulated. | |
6483 | */ | |
736c291c | 6484 | gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL); |
a6f177ef | 6485 | |
95b3cf69 XG |
6486 | /* |
6487 | * If the mapping is invalid in guest, let cpu retry | |
6488 | * it to generate fault. | |
6489 | */ | |
6490 | if (gpa == UNMAPPED_GVA) | |
6491 | return true; | |
6492 | } | |
a6f177ef | 6493 | |
8e3d9d06 XG |
6494 | /* |
6495 | * Do not retry the unhandleable instruction if it faults on the | |
6496 | * readonly host memory, otherwise it will goto a infinite loop: | |
6497 | * retry instruction -> write #PF -> emulation fail -> retry | |
6498 | * instruction -> ... | |
6499 | */ | |
6500 | pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa)); | |
95b3cf69 XG |
6501 | |
6502 | /* | |
6503 | * If the instruction failed on the error pfn, it can not be fixed, | |
6504 | * report the error to userspace. | |
6505 | */ | |
6506 | if (is_error_noslot_pfn(pfn)) | |
6507 | return false; | |
6508 | ||
6509 | kvm_release_pfn_clean(pfn); | |
6510 | ||
6511 | /* The instructions are well-emulated on direct mmu. */ | |
44dd3ffa | 6512 | if (vcpu->arch.mmu->direct_map) { |
95b3cf69 XG |
6513 | unsigned int indirect_shadow_pages; |
6514 | ||
6515 | spin_lock(&vcpu->kvm->mmu_lock); | |
6516 | indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages; | |
6517 | spin_unlock(&vcpu->kvm->mmu_lock); | |
6518 | ||
6519 | if (indirect_shadow_pages) | |
6520 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa)); | |
6521 | ||
a6f177ef | 6522 | return true; |
8e3d9d06 | 6523 | } |
a6f177ef | 6524 | |
95b3cf69 XG |
6525 | /* |
6526 | * if emulation was due to access to shadowed page table | |
6527 | * and it failed try to unshadow page and re-enter the | |
6528 | * guest to let CPU execute the instruction. | |
6529 | */ | |
6530 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa)); | |
93c05d3e XG |
6531 | |
6532 | /* | |
6533 | * If the access faults on its page table, it can not | |
6534 | * be fixed by unprotecting shadow page and it should | |
6535 | * be reported to userspace. | |
6536 | */ | |
6537 | return !write_fault_to_shadow_pgtable; | |
a6f177ef GN |
6538 | } |
6539 | ||
1cb3f3ae | 6540 | static bool retry_instruction(struct x86_emulate_ctxt *ctxt, |
736c291c | 6541 | gpa_t cr2_or_gpa, int emulation_type) |
1cb3f3ae XG |
6542 | { |
6543 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); | |
736c291c | 6544 | unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa; |
1cb3f3ae XG |
6545 | |
6546 | last_retry_eip = vcpu->arch.last_retry_eip; | |
6547 | last_retry_addr = vcpu->arch.last_retry_addr; | |
6548 | ||
6549 | /* | |
6550 | * If the emulation is caused by #PF and it is non-page_table | |
6551 | * writing instruction, it means the VM-EXIT is caused by shadow | |
6552 | * page protected, we can zap the shadow page and retry this | |
6553 | * instruction directly. | |
6554 | * | |
6555 | * Note: if the guest uses a non-page-table modifying instruction | |
6556 | * on the PDE that points to the instruction, then we will unmap | |
6557 | * the instruction and go to an infinite loop. So, we cache the | |
6558 | * last retried eip and the last fault address, if we meet the eip | |
6559 | * and the address again, we can break out of the potential infinite | |
6560 | * loop. | |
6561 | */ | |
6562 | vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0; | |
6563 | ||
384bf221 | 6564 | if (!(emulation_type & EMULTYPE_ALLOW_RETRY)) |
1cb3f3ae XG |
6565 | return false; |
6566 | ||
6c3dfeb6 SC |
6567 | if (WARN_ON_ONCE(is_guest_mode(vcpu))) |
6568 | return false; | |
6569 | ||
1cb3f3ae XG |
6570 | if (x86_page_table_writing_insn(ctxt)) |
6571 | return false; | |
6572 | ||
736c291c | 6573 | if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa) |
1cb3f3ae XG |
6574 | return false; |
6575 | ||
6576 | vcpu->arch.last_retry_eip = ctxt->eip; | |
736c291c | 6577 | vcpu->arch.last_retry_addr = cr2_or_gpa; |
1cb3f3ae | 6578 | |
44dd3ffa | 6579 | if (!vcpu->arch.mmu->direct_map) |
736c291c | 6580 | gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL); |
1cb3f3ae | 6581 | |
22368028 | 6582 | kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa)); |
1cb3f3ae XG |
6583 | |
6584 | return true; | |
6585 | } | |
6586 | ||
716d51ab GN |
6587 | static int complete_emulated_mmio(struct kvm_vcpu *vcpu); |
6588 | static int complete_emulated_pio(struct kvm_vcpu *vcpu); | |
6589 | ||
64d60670 | 6590 | static void kvm_smm_changed(struct kvm_vcpu *vcpu) |
a584539b | 6591 | { |
64d60670 | 6592 | if (!(vcpu->arch.hflags & HF_SMM_MASK)) { |
660a5d51 PB |
6593 | /* This is a good place to trace that we are exiting SMM. */ |
6594 | trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false); | |
6595 | ||
c43203ca PB |
6596 | /* Process a latched INIT or SMI, if any. */ |
6597 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
64d60670 | 6598 | } |
699023e2 PB |
6599 | |
6600 | kvm_mmu_reset_context(vcpu); | |
64d60670 PB |
6601 | } |
6602 | ||
4a1e10d5 PB |
6603 | static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7, |
6604 | unsigned long *db) | |
6605 | { | |
6606 | u32 dr6 = 0; | |
6607 | int i; | |
6608 | u32 enable, rwlen; | |
6609 | ||
6610 | enable = dr7; | |
6611 | rwlen = dr7 >> 16; | |
6612 | for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4) | |
6613 | if ((enable & 3) && (rwlen & 15) == type && db[i] == addr) | |
6614 | dr6 |= (1 << i); | |
6615 | return dr6; | |
6616 | } | |
6617 | ||
120c2c4f | 6618 | static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu) |
663f4c61 PB |
6619 | { |
6620 | struct kvm_run *kvm_run = vcpu->run; | |
6621 | ||
c8401dda PB |
6622 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) { |
6623 | kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM; | |
6624 | kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip; | |
6625 | kvm_run->debug.arch.exception = DB_VECTOR; | |
6626 | kvm_run->exit_reason = KVM_EXIT_DEBUG; | |
60fc3d02 | 6627 | return 0; |
663f4c61 | 6628 | } |
120c2c4f | 6629 | kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS); |
60fc3d02 | 6630 | return 1; |
663f4c61 PB |
6631 | } |
6632 | ||
6affcbed KH |
6633 | int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu) |
6634 | { | |
6635 | unsigned long rflags = kvm_x86_ops->get_rflags(vcpu); | |
f8ea7c60 | 6636 | int r; |
6affcbed | 6637 | |
f8ea7c60 | 6638 | r = kvm_x86_ops->skip_emulated_instruction(vcpu); |
60fc3d02 | 6639 | if (unlikely(!r)) |
f8ea7c60 | 6640 | return 0; |
c8401dda PB |
6641 | |
6642 | /* | |
6643 | * rflags is the old, "raw" value of the flags. The new value has | |
6644 | * not been saved yet. | |
6645 | * | |
6646 | * This is correct even for TF set by the guest, because "the | |
6647 | * processor will not generate this exception after the instruction | |
6648 | * that sets the TF flag". | |
6649 | */ | |
6650 | if (unlikely(rflags & X86_EFLAGS_TF)) | |
120c2c4f | 6651 | r = kvm_vcpu_do_singlestep(vcpu); |
60fc3d02 | 6652 | return r; |
6affcbed KH |
6653 | } |
6654 | EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction); | |
6655 | ||
4a1e10d5 PB |
6656 | static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r) |
6657 | { | |
4a1e10d5 PB |
6658 | if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) && |
6659 | (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) { | |
82b32774 NA |
6660 | struct kvm_run *kvm_run = vcpu->run; |
6661 | unsigned long eip = kvm_get_linear_rip(vcpu); | |
6662 | u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0, | |
4a1e10d5 PB |
6663 | vcpu->arch.guest_debug_dr7, |
6664 | vcpu->arch.eff_db); | |
6665 | ||
6666 | if (dr6 != 0) { | |
6f43ed01 | 6667 | kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM; |
82b32774 | 6668 | kvm_run->debug.arch.pc = eip; |
4a1e10d5 PB |
6669 | kvm_run->debug.arch.exception = DB_VECTOR; |
6670 | kvm_run->exit_reason = KVM_EXIT_DEBUG; | |
60fc3d02 | 6671 | *r = 0; |
4a1e10d5 PB |
6672 | return true; |
6673 | } | |
6674 | } | |
6675 | ||
4161a569 NA |
6676 | if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) && |
6677 | !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) { | |
82b32774 NA |
6678 | unsigned long eip = kvm_get_linear_rip(vcpu); |
6679 | u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0, | |
4a1e10d5 PB |
6680 | vcpu->arch.dr7, |
6681 | vcpu->arch.db); | |
6682 | ||
6683 | if (dr6 != 0) { | |
1fc5d194 | 6684 | vcpu->arch.dr6 &= ~DR_TRAP_BITS; |
6f43ed01 | 6685 | vcpu->arch.dr6 |= dr6 | DR6_RTM; |
4a1e10d5 | 6686 | kvm_queue_exception(vcpu, DB_VECTOR); |
60fc3d02 | 6687 | *r = 1; |
4a1e10d5 PB |
6688 | return true; |
6689 | } | |
6690 | } | |
6691 | ||
6692 | return false; | |
6693 | } | |
6694 | ||
04789b66 LA |
6695 | static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt) |
6696 | { | |
2d7921c4 AM |
6697 | switch (ctxt->opcode_len) { |
6698 | case 1: | |
6699 | switch (ctxt->b) { | |
6700 | case 0xe4: /* IN */ | |
6701 | case 0xe5: | |
6702 | case 0xec: | |
6703 | case 0xed: | |
6704 | case 0xe6: /* OUT */ | |
6705 | case 0xe7: | |
6706 | case 0xee: | |
6707 | case 0xef: | |
6708 | case 0x6c: /* INS */ | |
6709 | case 0x6d: | |
6710 | case 0x6e: /* OUTS */ | |
6711 | case 0x6f: | |
6712 | return true; | |
6713 | } | |
6714 | break; | |
6715 | case 2: | |
6716 | switch (ctxt->b) { | |
6717 | case 0x33: /* RDPMC */ | |
6718 | return true; | |
6719 | } | |
6720 | break; | |
04789b66 LA |
6721 | } |
6722 | ||
6723 | return false; | |
6724 | } | |
6725 | ||
736c291c SC |
6726 | int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa, |
6727 | int emulation_type, void *insn, int insn_len) | |
bbd9b64e | 6728 | { |
95cb2295 | 6729 | int r; |
9d74191a | 6730 | struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; |
7ae441ea | 6731 | bool writeback = true; |
93c05d3e | 6732 | bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable; |
bbd9b64e | 6733 | |
c595ceee PB |
6734 | vcpu->arch.l1tf_flush_l1d = true; |
6735 | ||
93c05d3e XG |
6736 | /* |
6737 | * Clear write_fault_to_shadow_pgtable here to ensure it is | |
6738 | * never reused. | |
6739 | */ | |
6740 | vcpu->arch.write_fault_to_shadow_pgtable = false; | |
26eef70c | 6741 | kvm_clear_exception_queue(vcpu); |
8d7d8102 | 6742 | |
571008da | 6743 | if (!(emulation_type & EMULTYPE_NO_DECODE)) { |
8ec4722d | 6744 | init_emulate_ctxt(vcpu); |
4a1e10d5 PB |
6745 | |
6746 | /* | |
6747 | * We will reenter on the same instruction since | |
6748 | * we do not set complete_userspace_io. This does not | |
6749 | * handle watchpoints yet, those would be handled in | |
6750 | * the emulate_ops. | |
6751 | */ | |
d391f120 VK |
6752 | if (!(emulation_type & EMULTYPE_SKIP) && |
6753 | kvm_vcpu_check_breakpoint(vcpu, &r)) | |
4a1e10d5 PB |
6754 | return r; |
6755 | ||
9d74191a TY |
6756 | ctxt->interruptibility = 0; |
6757 | ctxt->have_exception = false; | |
e0ad0b47 | 6758 | ctxt->exception.vector = -1; |
9d74191a | 6759 | ctxt->perm_ok = false; |
bbd9b64e | 6760 | |
b51e974f | 6761 | ctxt->ud = emulation_type & EMULTYPE_TRAP_UD; |
4005996e | 6762 | |
9d74191a | 6763 | r = x86_decode_insn(ctxt, insn, insn_len); |
bbd9b64e | 6764 | |
e46479f8 | 6765 | trace_kvm_emulate_insn_start(vcpu); |
f2b5756b | 6766 | ++vcpu->stat.insn_emulation; |
1d2887e2 | 6767 | if (r != EMULATION_OK) { |
b4000606 | 6768 | if ((emulation_type & EMULTYPE_TRAP_UD) || |
c83fad65 SC |
6769 | (emulation_type & EMULTYPE_TRAP_UD_FORCED)) { |
6770 | kvm_queue_exception(vcpu, UD_VECTOR); | |
60fc3d02 | 6771 | return 1; |
c83fad65 | 6772 | } |
736c291c SC |
6773 | if (reexecute_instruction(vcpu, cr2_or_gpa, |
6774 | write_fault_to_spt, | |
6775 | emulation_type)) | |
60fc3d02 | 6776 | return 1; |
8530a79c | 6777 | if (ctxt->have_exception) { |
c8848cee JD |
6778 | /* |
6779 | * #UD should result in just EMULATION_FAILED, and trap-like | |
6780 | * exception should not be encountered during decode. | |
6781 | */ | |
6782 | WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR || | |
6783 | exception_type(ctxt->exception.vector) == EXCPT_TRAP); | |
8530a79c | 6784 | inject_emulated_exception(vcpu); |
60fc3d02 | 6785 | return 1; |
8530a79c | 6786 | } |
e2366171 | 6787 | return handle_emulation_failure(vcpu, emulation_type); |
bbd9b64e CO |
6788 | } |
6789 | } | |
6790 | ||
42cbf068 SC |
6791 | if ((emulation_type & EMULTYPE_VMWARE_GP) && |
6792 | !is_vmware_backdoor_opcode(ctxt)) { | |
6793 | kvm_queue_exception_e(vcpu, GP_VECTOR, 0); | |
60fc3d02 | 6794 | return 1; |
42cbf068 | 6795 | } |
04789b66 | 6796 | |
1957aa63 SC |
6797 | /* |
6798 | * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks | |
6799 | * for kvm_skip_emulated_instruction(). The caller is responsible for | |
6800 | * updating interruptibility state and injecting single-step #DBs. | |
6801 | */ | |
ba8afb6b | 6802 | if (emulation_type & EMULTYPE_SKIP) { |
9dac77fa | 6803 | kvm_rip_write(vcpu, ctxt->_eip); |
bb663c7a NA |
6804 | if (ctxt->eflags & X86_EFLAGS_RF) |
6805 | kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF); | |
60fc3d02 | 6806 | return 1; |
ba8afb6b GN |
6807 | } |
6808 | ||
736c291c | 6809 | if (retry_instruction(ctxt, cr2_or_gpa, emulation_type)) |
60fc3d02 | 6810 | return 1; |
1cb3f3ae | 6811 | |
7ae441ea | 6812 | /* this is needed for vmware backdoor interface to work since it |
4d2179e1 | 6813 | changes registers values during IO operation */ |
7ae441ea GN |
6814 | if (vcpu->arch.emulate_regs_need_sync_from_vcpu) { |
6815 | vcpu->arch.emulate_regs_need_sync_from_vcpu = false; | |
dd856efa | 6816 | emulator_invalidate_register_cache(ctxt); |
7ae441ea | 6817 | } |
4d2179e1 | 6818 | |
5cd21917 | 6819 | restart: |
0f89b207 | 6820 | /* Save the faulting GPA (cr2) in the address field */ |
736c291c | 6821 | ctxt->exception.address = cr2_or_gpa; |
0f89b207 | 6822 | |
9d74191a | 6823 | r = x86_emulate_insn(ctxt); |
bbd9b64e | 6824 | |
775fde86 | 6825 | if (r == EMULATION_INTERCEPTED) |
60fc3d02 | 6826 | return 1; |
775fde86 | 6827 | |
d2ddd1c4 | 6828 | if (r == EMULATION_FAILED) { |
736c291c | 6829 | if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt, |
991eebf9 | 6830 | emulation_type)) |
60fc3d02 | 6831 | return 1; |
c3cd7ffa | 6832 | |
e2366171 | 6833 | return handle_emulation_failure(vcpu, emulation_type); |
bbd9b64e CO |
6834 | } |
6835 | ||
9d74191a | 6836 | if (ctxt->have_exception) { |
60fc3d02 | 6837 | r = 1; |
ef54bcfe PB |
6838 | if (inject_emulated_exception(vcpu)) |
6839 | return r; | |
d2ddd1c4 | 6840 | } else if (vcpu->arch.pio.count) { |
0912c977 PB |
6841 | if (!vcpu->arch.pio.in) { |
6842 | /* FIXME: return into emulator if single-stepping. */ | |
3457e419 | 6843 | vcpu->arch.pio.count = 0; |
0912c977 | 6844 | } else { |
7ae441ea | 6845 | writeback = false; |
716d51ab GN |
6846 | vcpu->arch.complete_userspace_io = complete_emulated_pio; |
6847 | } | |
60fc3d02 | 6848 | r = 0; |
7ae441ea | 6849 | } else if (vcpu->mmio_needed) { |
bc8a0aaf SC |
6850 | ++vcpu->stat.mmio_exits; |
6851 | ||
7ae441ea GN |
6852 | if (!vcpu->mmio_is_write) |
6853 | writeback = false; | |
60fc3d02 | 6854 | r = 0; |
716d51ab | 6855 | vcpu->arch.complete_userspace_io = complete_emulated_mmio; |
7ae441ea | 6856 | } else if (r == EMULATION_RESTART) |
5cd21917 | 6857 | goto restart; |
d2ddd1c4 | 6858 | else |
60fc3d02 | 6859 | r = 1; |
f850e2e6 | 6860 | |
7ae441ea | 6861 | if (writeback) { |
6addfc42 | 6862 | unsigned long rflags = kvm_x86_ops->get_rflags(vcpu); |
9d74191a | 6863 | toggle_interruptibility(vcpu, ctxt->interruptibility); |
7ae441ea | 6864 | vcpu->arch.emulate_regs_need_sync_to_vcpu = false; |
38827dbd | 6865 | if (!ctxt->have_exception || |
75ee23b3 SC |
6866 | exception_type(ctxt->exception.vector) == EXCPT_TRAP) { |
6867 | kvm_rip_write(vcpu, ctxt->eip); | |
60fc3d02 | 6868 | if (r && ctxt->tf) |
120c2c4f | 6869 | r = kvm_vcpu_do_singlestep(vcpu); |
38827dbd | 6870 | __kvm_set_rflags(vcpu, ctxt->eflags); |
75ee23b3 | 6871 | } |
6addfc42 PB |
6872 | |
6873 | /* | |
6874 | * For STI, interrupts are shadowed; so KVM_REQ_EVENT will | |
6875 | * do nothing, and it will be requested again as soon as | |
6876 | * the shadow expires. But we still need to check here, | |
6877 | * because POPF has no interrupt shadow. | |
6878 | */ | |
6879 | if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF)) | |
6880 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
7ae441ea GN |
6881 | } else |
6882 | vcpu->arch.emulate_regs_need_sync_to_vcpu = true; | |
e85d28f8 GN |
6883 | |
6884 | return r; | |
de7d789a | 6885 | } |
c60658d1 SC |
6886 | |
6887 | int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type) | |
6888 | { | |
6889 | return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0); | |
6890 | } | |
6891 | EXPORT_SYMBOL_GPL(kvm_emulate_instruction); | |
6892 | ||
6893 | int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu, | |
6894 | void *insn, int insn_len) | |
6895 | { | |
6896 | return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len); | |
6897 | } | |
6898 | EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer); | |
de7d789a | 6899 | |
8764ed55 SC |
6900 | static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu) |
6901 | { | |
6902 | vcpu->arch.pio.count = 0; | |
6903 | return 1; | |
6904 | } | |
6905 | ||
45def77e SC |
6906 | static int complete_fast_pio_out(struct kvm_vcpu *vcpu) |
6907 | { | |
6908 | vcpu->arch.pio.count = 0; | |
6909 | ||
6910 | if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) | |
6911 | return 1; | |
6912 | ||
6913 | return kvm_skip_emulated_instruction(vcpu); | |
6914 | } | |
6915 | ||
dca7f128 SC |
6916 | static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, |
6917 | unsigned short port) | |
de7d789a | 6918 | { |
de3cd117 | 6919 | unsigned long val = kvm_rax_read(vcpu); |
ca1d4a9e AK |
6920 | int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt, |
6921 | size, port, &val, 1); | |
8764ed55 SC |
6922 | if (ret) |
6923 | return ret; | |
45def77e | 6924 | |
8764ed55 SC |
6925 | /* |
6926 | * Workaround userspace that relies on old KVM behavior of %rip being | |
6927 | * incremented prior to exiting to userspace to handle "OUT 0x7e". | |
6928 | */ | |
6929 | if (port == 0x7e && | |
6930 | kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) { | |
6931 | vcpu->arch.complete_userspace_io = | |
6932 | complete_fast_pio_out_port_0x7e; | |
6933 | kvm_skip_emulated_instruction(vcpu); | |
6934 | } else { | |
45def77e SC |
6935 | vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu); |
6936 | vcpu->arch.complete_userspace_io = complete_fast_pio_out; | |
6937 | } | |
8764ed55 | 6938 | return 0; |
de7d789a | 6939 | } |
de7d789a | 6940 | |
8370c3d0 TL |
6941 | static int complete_fast_pio_in(struct kvm_vcpu *vcpu) |
6942 | { | |
6943 | unsigned long val; | |
6944 | ||
6945 | /* We should only ever be called with arch.pio.count equal to 1 */ | |
6946 | BUG_ON(vcpu->arch.pio.count != 1); | |
6947 | ||
45def77e SC |
6948 | if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) { |
6949 | vcpu->arch.pio.count = 0; | |
6950 | return 1; | |
6951 | } | |
6952 | ||
8370c3d0 | 6953 | /* For size less than 4 we merge, else we zero extend */ |
de3cd117 | 6954 | val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0; |
8370c3d0 TL |
6955 | |
6956 | /* | |
6957 | * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform | |
6958 | * the copy and tracing | |
6959 | */ | |
6960 | emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size, | |
6961 | vcpu->arch.pio.port, &val, 1); | |
de3cd117 | 6962 | kvm_rax_write(vcpu, val); |
8370c3d0 | 6963 | |
45def77e | 6964 | return kvm_skip_emulated_instruction(vcpu); |
8370c3d0 TL |
6965 | } |
6966 | ||
dca7f128 SC |
6967 | static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size, |
6968 | unsigned short port) | |
8370c3d0 TL |
6969 | { |
6970 | unsigned long val; | |
6971 | int ret; | |
6972 | ||
6973 | /* For size less than 4 we merge, else we zero extend */ | |
de3cd117 | 6974 | val = (size < 4) ? kvm_rax_read(vcpu) : 0; |
8370c3d0 TL |
6975 | |
6976 | ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port, | |
6977 | &val, 1); | |
6978 | if (ret) { | |
de3cd117 | 6979 | kvm_rax_write(vcpu, val); |
8370c3d0 TL |
6980 | return ret; |
6981 | } | |
6982 | ||
45def77e | 6983 | vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu); |
8370c3d0 TL |
6984 | vcpu->arch.complete_userspace_io = complete_fast_pio_in; |
6985 | ||
6986 | return 0; | |
6987 | } | |
dca7f128 SC |
6988 | |
6989 | int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in) | |
6990 | { | |
45def77e | 6991 | int ret; |
dca7f128 | 6992 | |
dca7f128 | 6993 | if (in) |
45def77e | 6994 | ret = kvm_fast_pio_in(vcpu, size, port); |
dca7f128 | 6995 | else |
45def77e SC |
6996 | ret = kvm_fast_pio_out(vcpu, size, port); |
6997 | return ret && kvm_skip_emulated_instruction(vcpu); | |
dca7f128 SC |
6998 | } |
6999 | EXPORT_SYMBOL_GPL(kvm_fast_pio); | |
8370c3d0 | 7000 | |
251a5fd6 | 7001 | static int kvmclock_cpu_down_prep(unsigned int cpu) |
8cfdc000 | 7002 | { |
0a3aee0d | 7003 | __this_cpu_write(cpu_tsc_khz, 0); |
251a5fd6 | 7004 | return 0; |
8cfdc000 ZA |
7005 | } |
7006 | ||
7007 | static void tsc_khz_changed(void *data) | |
c8076604 | 7008 | { |
8cfdc000 ZA |
7009 | struct cpufreq_freqs *freq = data; |
7010 | unsigned long khz = 0; | |
7011 | ||
7012 | if (data) | |
7013 | khz = freq->new; | |
7014 | else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) | |
7015 | khz = cpufreq_quick_get(raw_smp_processor_id()); | |
7016 | if (!khz) | |
7017 | khz = tsc_khz; | |
0a3aee0d | 7018 | __this_cpu_write(cpu_tsc_khz, khz); |
c8076604 GH |
7019 | } |
7020 | ||
5fa4ec9c | 7021 | #ifdef CONFIG_X86_64 |
0092e434 VK |
7022 | static void kvm_hyperv_tsc_notifier(void) |
7023 | { | |
0092e434 VK |
7024 | struct kvm *kvm; |
7025 | struct kvm_vcpu *vcpu; | |
7026 | int cpu; | |
7027 | ||
0d9ce162 | 7028 | mutex_lock(&kvm_lock); |
0092e434 VK |
7029 | list_for_each_entry(kvm, &vm_list, vm_list) |
7030 | kvm_make_mclock_inprogress_request(kvm); | |
7031 | ||
7032 | hyperv_stop_tsc_emulation(); | |
7033 | ||
7034 | /* TSC frequency always matches when on Hyper-V */ | |
7035 | for_each_present_cpu(cpu) | |
7036 | per_cpu(cpu_tsc_khz, cpu) = tsc_khz; | |
7037 | kvm_max_guest_tsc_khz = tsc_khz; | |
7038 | ||
7039 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
7040 | struct kvm_arch *ka = &kvm->arch; | |
7041 | ||
7042 | spin_lock(&ka->pvclock_gtod_sync_lock); | |
7043 | ||
7044 | pvclock_update_vm_gtod_copy(kvm); | |
7045 | ||
7046 | kvm_for_each_vcpu(cpu, vcpu, kvm) | |
7047 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); | |
7048 | ||
7049 | kvm_for_each_vcpu(cpu, vcpu, kvm) | |
7050 | kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu); | |
7051 | ||
7052 | spin_unlock(&ka->pvclock_gtod_sync_lock); | |
7053 | } | |
0d9ce162 | 7054 | mutex_unlock(&kvm_lock); |
0092e434 | 7055 | } |
5fa4ec9c | 7056 | #endif |
0092e434 | 7057 | |
df24014a | 7058 | static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu) |
c8076604 | 7059 | { |
c8076604 GH |
7060 | struct kvm *kvm; |
7061 | struct kvm_vcpu *vcpu; | |
7062 | int i, send_ipi = 0; | |
7063 | ||
8cfdc000 ZA |
7064 | /* |
7065 | * We allow guests to temporarily run on slowing clocks, | |
7066 | * provided we notify them after, or to run on accelerating | |
7067 | * clocks, provided we notify them before. Thus time never | |
7068 | * goes backwards. | |
7069 | * | |
7070 | * However, we have a problem. We can't atomically update | |
7071 | * the frequency of a given CPU from this function; it is | |
7072 | * merely a notifier, which can be called from any CPU. | |
7073 | * Changing the TSC frequency at arbitrary points in time | |
7074 | * requires a recomputation of local variables related to | |
7075 | * the TSC for each VCPU. We must flag these local variables | |
7076 | * to be updated and be sure the update takes place with the | |
7077 | * new frequency before any guests proceed. | |
7078 | * | |
7079 | * Unfortunately, the combination of hotplug CPU and frequency | |
7080 | * change creates an intractable locking scenario; the order | |
7081 | * of when these callouts happen is undefined with respect to | |
7082 | * CPU hotplug, and they can race with each other. As such, | |
7083 | * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is | |
7084 | * undefined; you can actually have a CPU frequency change take | |
7085 | * place in between the computation of X and the setting of the | |
7086 | * variable. To protect against this problem, all updates of | |
7087 | * the per_cpu tsc_khz variable are done in an interrupt | |
7088 | * protected IPI, and all callers wishing to update the value | |
7089 | * must wait for a synchronous IPI to complete (which is trivial | |
7090 | * if the caller is on the CPU already). This establishes the | |
7091 | * necessary total order on variable updates. | |
7092 | * | |
7093 | * Note that because a guest time update may take place | |
7094 | * anytime after the setting of the VCPU's request bit, the | |
7095 | * correct TSC value must be set before the request. However, | |
7096 | * to ensure the update actually makes it to any guest which | |
7097 | * starts running in hardware virtualization between the set | |
7098 | * and the acquisition of the spinlock, we must also ping the | |
7099 | * CPU after setting the request bit. | |
7100 | * | |
7101 | */ | |
7102 | ||
df24014a | 7103 | smp_call_function_single(cpu, tsc_khz_changed, freq, 1); |
c8076604 | 7104 | |
0d9ce162 | 7105 | mutex_lock(&kvm_lock); |
c8076604 | 7106 | list_for_each_entry(kvm, &vm_list, vm_list) { |
988a2cae | 7107 | kvm_for_each_vcpu(i, vcpu, kvm) { |
df24014a | 7108 | if (vcpu->cpu != cpu) |
c8076604 | 7109 | continue; |
c285545f | 7110 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
0d9ce162 | 7111 | if (vcpu->cpu != raw_smp_processor_id()) |
8cfdc000 | 7112 | send_ipi = 1; |
c8076604 GH |
7113 | } |
7114 | } | |
0d9ce162 | 7115 | mutex_unlock(&kvm_lock); |
c8076604 GH |
7116 | |
7117 | if (freq->old < freq->new && send_ipi) { | |
7118 | /* | |
7119 | * We upscale the frequency. Must make the guest | |
7120 | * doesn't see old kvmclock values while running with | |
7121 | * the new frequency, otherwise we risk the guest sees | |
7122 | * time go backwards. | |
7123 | * | |
7124 | * In case we update the frequency for another cpu | |
7125 | * (which might be in guest context) send an interrupt | |
7126 | * to kick the cpu out of guest context. Next time | |
7127 | * guest context is entered kvmclock will be updated, | |
7128 | * so the guest will not see stale values. | |
7129 | */ | |
df24014a | 7130 | smp_call_function_single(cpu, tsc_khz_changed, freq, 1); |
c8076604 | 7131 | } |
df24014a VK |
7132 | } |
7133 | ||
7134 | static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val, | |
7135 | void *data) | |
7136 | { | |
7137 | struct cpufreq_freqs *freq = data; | |
7138 | int cpu; | |
7139 | ||
7140 | if (val == CPUFREQ_PRECHANGE && freq->old > freq->new) | |
7141 | return 0; | |
7142 | if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new) | |
7143 | return 0; | |
7144 | ||
7145 | for_each_cpu(cpu, freq->policy->cpus) | |
7146 | __kvmclock_cpufreq_notifier(freq, cpu); | |
7147 | ||
c8076604 GH |
7148 | return 0; |
7149 | } | |
7150 | ||
7151 | static struct notifier_block kvmclock_cpufreq_notifier_block = { | |
8cfdc000 ZA |
7152 | .notifier_call = kvmclock_cpufreq_notifier |
7153 | }; | |
7154 | ||
251a5fd6 | 7155 | static int kvmclock_cpu_online(unsigned int cpu) |
8cfdc000 | 7156 | { |
251a5fd6 SAS |
7157 | tsc_khz_changed(NULL); |
7158 | return 0; | |
8cfdc000 ZA |
7159 | } |
7160 | ||
b820cc0c ZA |
7161 | static void kvm_timer_init(void) |
7162 | { | |
c285545f | 7163 | max_tsc_khz = tsc_khz; |
460dd42e | 7164 | |
b820cc0c | 7165 | if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) { |
c285545f ZA |
7166 | #ifdef CONFIG_CPU_FREQ |
7167 | struct cpufreq_policy policy; | |
758f588d BP |
7168 | int cpu; |
7169 | ||
c285545f | 7170 | memset(&policy, 0, sizeof(policy)); |
3e26f230 AK |
7171 | cpu = get_cpu(); |
7172 | cpufreq_get_policy(&policy, cpu); | |
c285545f ZA |
7173 | if (policy.cpuinfo.max_freq) |
7174 | max_tsc_khz = policy.cpuinfo.max_freq; | |
3e26f230 | 7175 | put_cpu(); |
c285545f | 7176 | #endif |
b820cc0c ZA |
7177 | cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block, |
7178 | CPUFREQ_TRANSITION_NOTIFIER); | |
7179 | } | |
460dd42e | 7180 | |
73c1b41e | 7181 | cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online", |
251a5fd6 | 7182 | kvmclock_cpu_online, kvmclock_cpu_down_prep); |
b820cc0c ZA |
7183 | } |
7184 | ||
dd60d217 AK |
7185 | DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu); |
7186 | EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu); | |
ff9d07a0 | 7187 | |
f5132b01 | 7188 | int kvm_is_in_guest(void) |
ff9d07a0 | 7189 | { |
086c9855 | 7190 | return __this_cpu_read(current_vcpu) != NULL; |
ff9d07a0 ZY |
7191 | } |
7192 | ||
7193 | static int kvm_is_user_mode(void) | |
7194 | { | |
7195 | int user_mode = 3; | |
dcf46b94 | 7196 | |
086c9855 AS |
7197 | if (__this_cpu_read(current_vcpu)) |
7198 | user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu)); | |
dcf46b94 | 7199 | |
ff9d07a0 ZY |
7200 | return user_mode != 0; |
7201 | } | |
7202 | ||
7203 | static unsigned long kvm_get_guest_ip(void) | |
7204 | { | |
7205 | unsigned long ip = 0; | |
dcf46b94 | 7206 | |
086c9855 AS |
7207 | if (__this_cpu_read(current_vcpu)) |
7208 | ip = kvm_rip_read(__this_cpu_read(current_vcpu)); | |
dcf46b94 | 7209 | |
ff9d07a0 ZY |
7210 | return ip; |
7211 | } | |
7212 | ||
8479e04e LK |
7213 | static void kvm_handle_intel_pt_intr(void) |
7214 | { | |
7215 | struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu); | |
7216 | ||
7217 | kvm_make_request(KVM_REQ_PMI, vcpu); | |
7218 | __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT, | |
7219 | (unsigned long *)&vcpu->arch.pmu.global_status); | |
7220 | } | |
7221 | ||
ff9d07a0 ZY |
7222 | static struct perf_guest_info_callbacks kvm_guest_cbs = { |
7223 | .is_in_guest = kvm_is_in_guest, | |
7224 | .is_user_mode = kvm_is_user_mode, | |
7225 | .get_guest_ip = kvm_get_guest_ip, | |
8479e04e | 7226 | .handle_intel_pt_intr = kvm_handle_intel_pt_intr, |
ff9d07a0 ZY |
7227 | }; |
7228 | ||
16e8d74d MT |
7229 | #ifdef CONFIG_X86_64 |
7230 | static void pvclock_gtod_update_fn(struct work_struct *work) | |
7231 | { | |
d828199e MT |
7232 | struct kvm *kvm; |
7233 | ||
7234 | struct kvm_vcpu *vcpu; | |
7235 | int i; | |
7236 | ||
0d9ce162 | 7237 | mutex_lock(&kvm_lock); |
d828199e MT |
7238 | list_for_each_entry(kvm, &vm_list, vm_list) |
7239 | kvm_for_each_vcpu(i, vcpu, kvm) | |
105b21bb | 7240 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); |
d828199e | 7241 | atomic_set(&kvm_guest_has_master_clock, 0); |
0d9ce162 | 7242 | mutex_unlock(&kvm_lock); |
16e8d74d MT |
7243 | } |
7244 | ||
7245 | static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn); | |
7246 | ||
7247 | /* | |
7248 | * Notification about pvclock gtod data update. | |
7249 | */ | |
7250 | static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused, | |
7251 | void *priv) | |
7252 | { | |
7253 | struct pvclock_gtod_data *gtod = &pvclock_gtod_data; | |
7254 | struct timekeeper *tk = priv; | |
7255 | ||
7256 | update_pvclock_gtod(tk); | |
7257 | ||
7258 | /* disable master clock if host does not trust, or does not | |
b0c39dc6 | 7259 | * use, TSC based clocksource. |
16e8d74d | 7260 | */ |
b0c39dc6 | 7261 | if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) && |
16e8d74d MT |
7262 | atomic_read(&kvm_guest_has_master_clock) != 0) |
7263 | queue_work(system_long_wq, &pvclock_gtod_work); | |
7264 | ||
7265 | return 0; | |
7266 | } | |
7267 | ||
7268 | static struct notifier_block pvclock_gtod_notifier = { | |
7269 | .notifier_call = pvclock_gtod_notify, | |
7270 | }; | |
7271 | #endif | |
7272 | ||
f8c16bba | 7273 | int kvm_arch_init(void *opaque) |
043405e1 | 7274 | { |
b820cc0c | 7275 | int r; |
6b61edf7 | 7276 | struct kvm_x86_ops *ops = opaque; |
f8c16bba | 7277 | |
f8c16bba ZX |
7278 | if (kvm_x86_ops) { |
7279 | printk(KERN_ERR "kvm: already loaded the other module\n"); | |
56c6d28a ZX |
7280 | r = -EEXIST; |
7281 | goto out; | |
f8c16bba ZX |
7282 | } |
7283 | ||
7284 | if (!ops->cpu_has_kvm_support()) { | |
7285 | printk(KERN_ERR "kvm: no hardware support\n"); | |
56c6d28a ZX |
7286 | r = -EOPNOTSUPP; |
7287 | goto out; | |
f8c16bba ZX |
7288 | } |
7289 | if (ops->disabled_by_bios()) { | |
7290 | printk(KERN_ERR "kvm: disabled by bios\n"); | |
56c6d28a ZX |
7291 | r = -EOPNOTSUPP; |
7292 | goto out; | |
f8c16bba ZX |
7293 | } |
7294 | ||
b666a4b6 MO |
7295 | /* |
7296 | * KVM explicitly assumes that the guest has an FPU and | |
7297 | * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the | |
7298 | * vCPU's FPU state as a fxregs_state struct. | |
7299 | */ | |
7300 | if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) { | |
7301 | printk(KERN_ERR "kvm: inadequate fpu\n"); | |
7302 | r = -EOPNOTSUPP; | |
7303 | goto out; | |
7304 | } | |
7305 | ||
013f6a5d | 7306 | r = -ENOMEM; |
ed8e4812 | 7307 | x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu), |
b666a4b6 MO |
7308 | __alignof__(struct fpu), SLAB_ACCOUNT, |
7309 | NULL); | |
7310 | if (!x86_fpu_cache) { | |
7311 | printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n"); | |
7312 | goto out; | |
7313 | } | |
7314 | ||
013f6a5d MT |
7315 | shared_msrs = alloc_percpu(struct kvm_shared_msrs); |
7316 | if (!shared_msrs) { | |
7317 | printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n"); | |
b666a4b6 | 7318 | goto out_free_x86_fpu_cache; |
013f6a5d MT |
7319 | } |
7320 | ||
97db56ce AK |
7321 | r = kvm_mmu_module_init(); |
7322 | if (r) | |
013f6a5d | 7323 | goto out_free_percpu; |
97db56ce | 7324 | |
f8c16bba | 7325 | kvm_x86_ops = ops; |
920c8377 | 7326 | |
7b52345e | 7327 | kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK, |
ffb128c8 | 7328 | PT_DIRTY_MASK, PT64_NX_MASK, 0, |
d0ec49d4 | 7329 | PT_PRESENT_MASK, 0, sme_me_mask); |
b820cc0c | 7330 | kvm_timer_init(); |
c8076604 | 7331 | |
ff9d07a0 ZY |
7332 | perf_register_guest_info_callbacks(&kvm_guest_cbs); |
7333 | ||
d366bf7e | 7334 | if (boot_cpu_has(X86_FEATURE_XSAVE)) |
2acf923e DC |
7335 | host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK); |
7336 | ||
c5cc421b | 7337 | kvm_lapic_init(); |
0c5f81da WL |
7338 | if (pi_inject_timer == -1) |
7339 | pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER); | |
16e8d74d MT |
7340 | #ifdef CONFIG_X86_64 |
7341 | pvclock_gtod_register_notifier(&pvclock_gtod_notifier); | |
0092e434 | 7342 | |
5fa4ec9c | 7343 | if (hypervisor_is_type(X86_HYPER_MS_HYPERV)) |
0092e434 | 7344 | set_hv_tscchange_cb(kvm_hyperv_tsc_notifier); |
16e8d74d MT |
7345 | #endif |
7346 | ||
f8c16bba | 7347 | return 0; |
56c6d28a | 7348 | |
013f6a5d MT |
7349 | out_free_percpu: |
7350 | free_percpu(shared_msrs); | |
b666a4b6 MO |
7351 | out_free_x86_fpu_cache: |
7352 | kmem_cache_destroy(x86_fpu_cache); | |
56c6d28a | 7353 | out: |
56c6d28a | 7354 | return r; |
043405e1 | 7355 | } |
8776e519 | 7356 | |
f8c16bba ZX |
7357 | void kvm_arch_exit(void) |
7358 | { | |
0092e434 | 7359 | #ifdef CONFIG_X86_64 |
5fa4ec9c | 7360 | if (hypervisor_is_type(X86_HYPER_MS_HYPERV)) |
0092e434 VK |
7361 | clear_hv_tscchange_cb(); |
7362 | #endif | |
cef84c30 | 7363 | kvm_lapic_exit(); |
ff9d07a0 ZY |
7364 | perf_unregister_guest_info_callbacks(&kvm_guest_cbs); |
7365 | ||
888d256e JK |
7366 | if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) |
7367 | cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block, | |
7368 | CPUFREQ_TRANSITION_NOTIFIER); | |
251a5fd6 | 7369 | cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE); |
16e8d74d MT |
7370 | #ifdef CONFIG_X86_64 |
7371 | pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier); | |
7372 | #endif | |
f8c16bba | 7373 | kvm_x86_ops = NULL; |
56c6d28a | 7374 | kvm_mmu_module_exit(); |
013f6a5d | 7375 | free_percpu(shared_msrs); |
b666a4b6 | 7376 | kmem_cache_destroy(x86_fpu_cache); |
56c6d28a | 7377 | } |
f8c16bba | 7378 | |
5cb56059 | 7379 | int kvm_vcpu_halt(struct kvm_vcpu *vcpu) |
8776e519 HB |
7380 | { |
7381 | ++vcpu->stat.halt_exits; | |
35754c98 | 7382 | if (lapic_in_kernel(vcpu)) { |
a4535290 | 7383 | vcpu->arch.mp_state = KVM_MP_STATE_HALTED; |
8776e519 HB |
7384 | return 1; |
7385 | } else { | |
7386 | vcpu->run->exit_reason = KVM_EXIT_HLT; | |
7387 | return 0; | |
7388 | } | |
7389 | } | |
5cb56059 JS |
7390 | EXPORT_SYMBOL_GPL(kvm_vcpu_halt); |
7391 | ||
7392 | int kvm_emulate_halt(struct kvm_vcpu *vcpu) | |
7393 | { | |
6affcbed KH |
7394 | int ret = kvm_skip_emulated_instruction(vcpu); |
7395 | /* | |
7396 | * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered | |
7397 | * KVM_EXIT_DEBUG here. | |
7398 | */ | |
7399 | return kvm_vcpu_halt(vcpu) && ret; | |
5cb56059 | 7400 | } |
8776e519 HB |
7401 | EXPORT_SYMBOL_GPL(kvm_emulate_halt); |
7402 | ||
8ef81a9a | 7403 | #ifdef CONFIG_X86_64 |
55dd00a7 MT |
7404 | static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr, |
7405 | unsigned long clock_type) | |
7406 | { | |
7407 | struct kvm_clock_pairing clock_pairing; | |
899a31f5 | 7408 | struct timespec64 ts; |
80fbd89c | 7409 | u64 cycle; |
55dd00a7 MT |
7410 | int ret; |
7411 | ||
7412 | if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK) | |
7413 | return -KVM_EOPNOTSUPP; | |
7414 | ||
7415 | if (kvm_get_walltime_and_clockread(&ts, &cycle) == false) | |
7416 | return -KVM_EOPNOTSUPP; | |
7417 | ||
7418 | clock_pairing.sec = ts.tv_sec; | |
7419 | clock_pairing.nsec = ts.tv_nsec; | |
7420 | clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle); | |
7421 | clock_pairing.flags = 0; | |
bcbfbd8e | 7422 | memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad)); |
55dd00a7 MT |
7423 | |
7424 | ret = 0; | |
7425 | if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing, | |
7426 | sizeof(struct kvm_clock_pairing))) | |
7427 | ret = -KVM_EFAULT; | |
7428 | ||
7429 | return ret; | |
7430 | } | |
8ef81a9a | 7431 | #endif |
55dd00a7 | 7432 | |
6aef266c SV |
7433 | /* |
7434 | * kvm_pv_kick_cpu_op: Kick a vcpu. | |
7435 | * | |
7436 | * @apicid - apicid of vcpu to be kicked. | |
7437 | */ | |
7438 | static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid) | |
7439 | { | |
24d2166b | 7440 | struct kvm_lapic_irq lapic_irq; |
6aef266c | 7441 | |
150a84fe | 7442 | lapic_irq.shorthand = APIC_DEST_NOSHORT; |
c96001c5 | 7443 | lapic_irq.dest_mode = APIC_DEST_PHYSICAL; |
ebd28fcb | 7444 | lapic_irq.level = 0; |
24d2166b | 7445 | lapic_irq.dest_id = apicid; |
93bbf0b8 | 7446 | lapic_irq.msi_redir_hint = false; |
6aef266c | 7447 | |
24d2166b | 7448 | lapic_irq.delivery_mode = APIC_DM_REMRD; |
795a149e | 7449 | kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL); |
6aef266c SV |
7450 | } |
7451 | ||
d62caabb AS |
7452 | void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu) |
7453 | { | |
f7589cca PB |
7454 | if (!lapic_in_kernel(vcpu)) { |
7455 | WARN_ON_ONCE(vcpu->arch.apicv_active); | |
7456 | return; | |
7457 | } | |
7458 | if (!vcpu->arch.apicv_active) | |
7459 | return; | |
7460 | ||
d62caabb AS |
7461 | vcpu->arch.apicv_active = false; |
7462 | kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu); | |
7463 | } | |
7464 | ||
71506297 WL |
7465 | static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id) |
7466 | { | |
7467 | struct kvm_vcpu *target = NULL; | |
7468 | struct kvm_apic_map *map; | |
7469 | ||
7470 | rcu_read_lock(); | |
7471 | map = rcu_dereference(kvm->arch.apic_map); | |
7472 | ||
7473 | if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id]) | |
7474 | target = map->phys_map[dest_id]->vcpu; | |
7475 | ||
7476 | rcu_read_unlock(); | |
7477 | ||
266e85a5 | 7478 | if (target && READ_ONCE(target->ready)) |
71506297 WL |
7479 | kvm_vcpu_yield_to(target); |
7480 | } | |
7481 | ||
8776e519 HB |
7482 | int kvm_emulate_hypercall(struct kvm_vcpu *vcpu) |
7483 | { | |
7484 | unsigned long nr, a0, a1, a2, a3, ret; | |
6356ee0c | 7485 | int op_64_bit; |
8776e519 | 7486 | |
696ca779 RK |
7487 | if (kvm_hv_hypercall_enabled(vcpu->kvm)) |
7488 | return kvm_hv_hypercall(vcpu); | |
55cd8e5a | 7489 | |
de3cd117 SC |
7490 | nr = kvm_rax_read(vcpu); |
7491 | a0 = kvm_rbx_read(vcpu); | |
7492 | a1 = kvm_rcx_read(vcpu); | |
7493 | a2 = kvm_rdx_read(vcpu); | |
7494 | a3 = kvm_rsi_read(vcpu); | |
8776e519 | 7495 | |
229456fc | 7496 | trace_kvm_hypercall(nr, a0, a1, a2, a3); |
2714d1d3 | 7497 | |
a449c7aa NA |
7498 | op_64_bit = is_64_bit_mode(vcpu); |
7499 | if (!op_64_bit) { | |
8776e519 HB |
7500 | nr &= 0xFFFFFFFF; |
7501 | a0 &= 0xFFFFFFFF; | |
7502 | a1 &= 0xFFFFFFFF; | |
7503 | a2 &= 0xFFFFFFFF; | |
7504 | a3 &= 0xFFFFFFFF; | |
7505 | } | |
7506 | ||
07708c4a JK |
7507 | if (kvm_x86_ops->get_cpl(vcpu) != 0) { |
7508 | ret = -KVM_EPERM; | |
696ca779 | 7509 | goto out; |
07708c4a JK |
7510 | } |
7511 | ||
8776e519 | 7512 | switch (nr) { |
b93463aa AK |
7513 | case KVM_HC_VAPIC_POLL_IRQ: |
7514 | ret = 0; | |
7515 | break; | |
6aef266c SV |
7516 | case KVM_HC_KICK_CPU: |
7517 | kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1); | |
266e85a5 | 7518 | kvm_sched_yield(vcpu->kvm, a1); |
6aef266c SV |
7519 | ret = 0; |
7520 | break; | |
8ef81a9a | 7521 | #ifdef CONFIG_X86_64 |
55dd00a7 MT |
7522 | case KVM_HC_CLOCK_PAIRING: |
7523 | ret = kvm_pv_clock_pairing(vcpu, a0, a1); | |
7524 | break; | |
1ed199a4 | 7525 | #endif |
4180bf1b WL |
7526 | case KVM_HC_SEND_IPI: |
7527 | ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit); | |
7528 | break; | |
71506297 WL |
7529 | case KVM_HC_SCHED_YIELD: |
7530 | kvm_sched_yield(vcpu->kvm, a0); | |
7531 | ret = 0; | |
7532 | break; | |
8776e519 HB |
7533 | default: |
7534 | ret = -KVM_ENOSYS; | |
7535 | break; | |
7536 | } | |
696ca779 | 7537 | out: |
a449c7aa NA |
7538 | if (!op_64_bit) |
7539 | ret = (u32)ret; | |
de3cd117 | 7540 | kvm_rax_write(vcpu, ret); |
6356ee0c | 7541 | |
f11c3a8d | 7542 | ++vcpu->stat.hypercalls; |
6356ee0c | 7543 | return kvm_skip_emulated_instruction(vcpu); |
8776e519 HB |
7544 | } |
7545 | EXPORT_SYMBOL_GPL(kvm_emulate_hypercall); | |
7546 | ||
b6785def | 7547 | static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt) |
8776e519 | 7548 | { |
d6aa1000 | 7549 | struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); |
8776e519 | 7550 | char instruction[3]; |
5fdbf976 | 7551 | unsigned long rip = kvm_rip_read(vcpu); |
8776e519 | 7552 | |
8776e519 | 7553 | kvm_x86_ops->patch_hypercall(vcpu, instruction); |
8776e519 | 7554 | |
ce2e852e DV |
7555 | return emulator_write_emulated(ctxt, rip, instruction, 3, |
7556 | &ctxt->exception); | |
8776e519 HB |
7557 | } |
7558 | ||
851ba692 | 7559 | static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu) |
b6c7a5dc | 7560 | { |
782d422b MG |
7561 | return vcpu->run->request_interrupt_window && |
7562 | likely(!pic_in_kernel(vcpu->kvm)); | |
b6c7a5dc HB |
7563 | } |
7564 | ||
851ba692 | 7565 | static void post_kvm_run_save(struct kvm_vcpu *vcpu) |
b6c7a5dc | 7566 | { |
851ba692 AK |
7567 | struct kvm_run *kvm_run = vcpu->run; |
7568 | ||
91586a3b | 7569 | kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0; |
f077825a | 7570 | kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0; |
2d3ad1f4 | 7571 | kvm_run->cr8 = kvm_get_cr8(vcpu); |
b6c7a5dc | 7572 | kvm_run->apic_base = kvm_get_apic_base(vcpu); |
127a457a MG |
7573 | kvm_run->ready_for_interrupt_injection = |
7574 | pic_in_kernel(vcpu->kvm) || | |
782d422b | 7575 | kvm_vcpu_ready_for_interrupt_injection(vcpu); |
b6c7a5dc HB |
7576 | } |
7577 | ||
95ba8273 GN |
7578 | static void update_cr8_intercept(struct kvm_vcpu *vcpu) |
7579 | { | |
7580 | int max_irr, tpr; | |
7581 | ||
7582 | if (!kvm_x86_ops->update_cr8_intercept) | |
7583 | return; | |
7584 | ||
bce87cce | 7585 | if (!lapic_in_kernel(vcpu)) |
88c808fd AK |
7586 | return; |
7587 | ||
d62caabb AS |
7588 | if (vcpu->arch.apicv_active) |
7589 | return; | |
7590 | ||
8db3baa2 GN |
7591 | if (!vcpu->arch.apic->vapic_addr) |
7592 | max_irr = kvm_lapic_find_highest_irr(vcpu); | |
7593 | else | |
7594 | max_irr = -1; | |
95ba8273 GN |
7595 | |
7596 | if (max_irr != -1) | |
7597 | max_irr >>= 4; | |
7598 | ||
7599 | tpr = kvm_lapic_get_cr8(vcpu); | |
7600 | ||
7601 | kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr); | |
7602 | } | |
7603 | ||
b6b8a145 | 7604 | static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win) |
95ba8273 | 7605 | { |
b6b8a145 JK |
7606 | int r; |
7607 | ||
95ba8273 | 7608 | /* try to reinject previous events if any */ |
664f8e26 | 7609 | |
1a680e35 LA |
7610 | if (vcpu->arch.exception.injected) |
7611 | kvm_x86_ops->queue_exception(vcpu); | |
664f8e26 | 7612 | /* |
a042c26f LA |
7613 | * Do not inject an NMI or interrupt if there is a pending |
7614 | * exception. Exceptions and interrupts are recognized at | |
7615 | * instruction boundaries, i.e. the start of an instruction. | |
7616 | * Trap-like exceptions, e.g. #DB, have higher priority than | |
7617 | * NMIs and interrupts, i.e. traps are recognized before an | |
7618 | * NMI/interrupt that's pending on the same instruction. | |
7619 | * Fault-like exceptions, e.g. #GP and #PF, are the lowest | |
7620 | * priority, but are only generated (pended) during instruction | |
7621 | * execution, i.e. a pending fault-like exception means the | |
7622 | * fault occurred on the *previous* instruction and must be | |
7623 | * serviced prior to recognizing any new events in order to | |
7624 | * fully complete the previous instruction. | |
664f8e26 | 7625 | */ |
1a680e35 LA |
7626 | else if (!vcpu->arch.exception.pending) { |
7627 | if (vcpu->arch.nmi_injected) | |
664f8e26 | 7628 | kvm_x86_ops->set_nmi(vcpu); |
1a680e35 | 7629 | else if (vcpu->arch.interrupt.injected) |
664f8e26 | 7630 | kvm_x86_ops->set_irq(vcpu); |
664f8e26 WL |
7631 | } |
7632 | ||
1a680e35 LA |
7633 | /* |
7634 | * Call check_nested_events() even if we reinjected a previous event | |
7635 | * in order for caller to determine if it should require immediate-exit | |
7636 | * from L2 to L1 due to pending L1 events which require exit | |
7637 | * from L2 to L1. | |
7638 | */ | |
664f8e26 WL |
7639 | if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) { |
7640 | r = kvm_x86_ops->check_nested_events(vcpu, req_int_win); | |
7641 | if (r != 0) | |
7642 | return r; | |
7643 | } | |
7644 | ||
7645 | /* try to inject new event if pending */ | |
b59bb7bd | 7646 | if (vcpu->arch.exception.pending) { |
5c1c85d0 AK |
7647 | trace_kvm_inj_exception(vcpu->arch.exception.nr, |
7648 | vcpu->arch.exception.has_error_code, | |
7649 | vcpu->arch.exception.error_code); | |
d6e8c854 | 7650 | |
1a680e35 | 7651 | WARN_ON_ONCE(vcpu->arch.exception.injected); |
664f8e26 WL |
7652 | vcpu->arch.exception.pending = false; |
7653 | vcpu->arch.exception.injected = true; | |
7654 | ||
d6e8c854 NA |
7655 | if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT) |
7656 | __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) | | |
7657 | X86_EFLAGS_RF); | |
7658 | ||
f10c729f JM |
7659 | if (vcpu->arch.exception.nr == DB_VECTOR) { |
7660 | /* | |
7661 | * This code assumes that nSVM doesn't use | |
7662 | * check_nested_events(). If it does, the | |
7663 | * DR6/DR7 changes should happen before L1 | |
7664 | * gets a #VMEXIT for an intercepted #DB in | |
7665 | * L2. (Under VMX, on the other hand, the | |
7666 | * DR6/DR7 changes should not happen in the | |
7667 | * event of a VM-exit to L1 for an intercepted | |
7668 | * #DB in L2.) | |
7669 | */ | |
7670 | kvm_deliver_exception_payload(vcpu); | |
7671 | if (vcpu->arch.dr7 & DR7_GD) { | |
7672 | vcpu->arch.dr7 &= ~DR7_GD; | |
7673 | kvm_update_dr7(vcpu); | |
7674 | } | |
6bdf0662 NA |
7675 | } |
7676 | ||
cfcd20e5 | 7677 | kvm_x86_ops->queue_exception(vcpu); |
1a680e35 LA |
7678 | } |
7679 | ||
7680 | /* Don't consider new event if we re-injected an event */ | |
7681 | if (kvm_event_needs_reinjection(vcpu)) | |
7682 | return 0; | |
7683 | ||
7684 | if (vcpu->arch.smi_pending && !is_smm(vcpu) && | |
7685 | kvm_x86_ops->smi_allowed(vcpu)) { | |
c43203ca | 7686 | vcpu->arch.smi_pending = false; |
52797bf9 | 7687 | ++vcpu->arch.smi_count; |
ee2cd4b7 | 7688 | enter_smm(vcpu); |
c43203ca | 7689 | } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) { |
321c5658 YS |
7690 | --vcpu->arch.nmi_pending; |
7691 | vcpu->arch.nmi_injected = true; | |
7692 | kvm_x86_ops->set_nmi(vcpu); | |
c7c9c56c | 7693 | } else if (kvm_cpu_has_injectable_intr(vcpu)) { |
9242b5b6 BD |
7694 | /* |
7695 | * Because interrupts can be injected asynchronously, we are | |
7696 | * calling check_nested_events again here to avoid a race condition. | |
7697 | * See https://lkml.org/lkml/2014/7/2/60 for discussion about this | |
7698 | * proposal and current concerns. Perhaps we should be setting | |
7699 | * KVM_REQ_EVENT only on certain events and not unconditionally? | |
7700 | */ | |
7701 | if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) { | |
7702 | r = kvm_x86_ops->check_nested_events(vcpu, req_int_win); | |
7703 | if (r != 0) | |
7704 | return r; | |
7705 | } | |
95ba8273 | 7706 | if (kvm_x86_ops->interrupt_allowed(vcpu)) { |
66fd3f7f GN |
7707 | kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), |
7708 | false); | |
7709 | kvm_x86_ops->set_irq(vcpu); | |
95ba8273 GN |
7710 | } |
7711 | } | |
ee2cd4b7 | 7712 | |
b6b8a145 | 7713 | return 0; |
95ba8273 GN |
7714 | } |
7715 | ||
7460fb4a AK |
7716 | static void process_nmi(struct kvm_vcpu *vcpu) |
7717 | { | |
7718 | unsigned limit = 2; | |
7719 | ||
7720 | /* | |
7721 | * x86 is limited to one NMI running, and one NMI pending after it. | |
7722 | * If an NMI is already in progress, limit further NMIs to just one. | |
7723 | * Otherwise, allow two (and we'll inject the first one immediately). | |
7724 | */ | |
7725 | if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected) | |
7726 | limit = 1; | |
7727 | ||
7728 | vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0); | |
7729 | vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit); | |
7730 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
7731 | } | |
7732 | ||
ee2cd4b7 | 7733 | static u32 enter_smm_get_segment_flags(struct kvm_segment *seg) |
660a5d51 PB |
7734 | { |
7735 | u32 flags = 0; | |
7736 | flags |= seg->g << 23; | |
7737 | flags |= seg->db << 22; | |
7738 | flags |= seg->l << 21; | |
7739 | flags |= seg->avl << 20; | |
7740 | flags |= seg->present << 15; | |
7741 | flags |= seg->dpl << 13; | |
7742 | flags |= seg->s << 12; | |
7743 | flags |= seg->type << 8; | |
7744 | return flags; | |
7745 | } | |
7746 | ||
ee2cd4b7 | 7747 | static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n) |
660a5d51 PB |
7748 | { |
7749 | struct kvm_segment seg; | |
7750 | int offset; | |
7751 | ||
7752 | kvm_get_segment(vcpu, &seg, n); | |
7753 | put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector); | |
7754 | ||
7755 | if (n < 3) | |
7756 | offset = 0x7f84 + n * 12; | |
7757 | else | |
7758 | offset = 0x7f2c + (n - 3) * 12; | |
7759 | ||
7760 | put_smstate(u32, buf, offset + 8, seg.base); | |
7761 | put_smstate(u32, buf, offset + 4, seg.limit); | |
ee2cd4b7 | 7762 | put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg)); |
660a5d51 PB |
7763 | } |
7764 | ||
efbb288a | 7765 | #ifdef CONFIG_X86_64 |
ee2cd4b7 | 7766 | static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n) |
660a5d51 PB |
7767 | { |
7768 | struct kvm_segment seg; | |
7769 | int offset; | |
7770 | u16 flags; | |
7771 | ||
7772 | kvm_get_segment(vcpu, &seg, n); | |
7773 | offset = 0x7e00 + n * 16; | |
7774 | ||
ee2cd4b7 | 7775 | flags = enter_smm_get_segment_flags(&seg) >> 8; |
660a5d51 PB |
7776 | put_smstate(u16, buf, offset, seg.selector); |
7777 | put_smstate(u16, buf, offset + 2, flags); | |
7778 | put_smstate(u32, buf, offset + 4, seg.limit); | |
7779 | put_smstate(u64, buf, offset + 8, seg.base); | |
7780 | } | |
efbb288a | 7781 | #endif |
660a5d51 | 7782 | |
ee2cd4b7 | 7783 | static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf) |
660a5d51 PB |
7784 | { |
7785 | struct desc_ptr dt; | |
7786 | struct kvm_segment seg; | |
7787 | unsigned long val; | |
7788 | int i; | |
7789 | ||
7790 | put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu)); | |
7791 | put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu)); | |
7792 | put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu)); | |
7793 | put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu)); | |
7794 | ||
7795 | for (i = 0; i < 8; i++) | |
7796 | put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i)); | |
7797 | ||
7798 | kvm_get_dr(vcpu, 6, &val); | |
7799 | put_smstate(u32, buf, 0x7fcc, (u32)val); | |
7800 | kvm_get_dr(vcpu, 7, &val); | |
7801 | put_smstate(u32, buf, 0x7fc8, (u32)val); | |
7802 | ||
7803 | kvm_get_segment(vcpu, &seg, VCPU_SREG_TR); | |
7804 | put_smstate(u32, buf, 0x7fc4, seg.selector); | |
7805 | put_smstate(u32, buf, 0x7f64, seg.base); | |
7806 | put_smstate(u32, buf, 0x7f60, seg.limit); | |
ee2cd4b7 | 7807 | put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg)); |
660a5d51 PB |
7808 | |
7809 | kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR); | |
7810 | put_smstate(u32, buf, 0x7fc0, seg.selector); | |
7811 | put_smstate(u32, buf, 0x7f80, seg.base); | |
7812 | put_smstate(u32, buf, 0x7f7c, seg.limit); | |
ee2cd4b7 | 7813 | put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg)); |
660a5d51 PB |
7814 | |
7815 | kvm_x86_ops->get_gdt(vcpu, &dt); | |
7816 | put_smstate(u32, buf, 0x7f74, dt.address); | |
7817 | put_smstate(u32, buf, 0x7f70, dt.size); | |
7818 | ||
7819 | kvm_x86_ops->get_idt(vcpu, &dt); | |
7820 | put_smstate(u32, buf, 0x7f58, dt.address); | |
7821 | put_smstate(u32, buf, 0x7f54, dt.size); | |
7822 | ||
7823 | for (i = 0; i < 6; i++) | |
ee2cd4b7 | 7824 | enter_smm_save_seg_32(vcpu, buf, i); |
660a5d51 PB |
7825 | |
7826 | put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu)); | |
7827 | ||
7828 | /* revision id */ | |
7829 | put_smstate(u32, buf, 0x7efc, 0x00020000); | |
7830 | put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase); | |
7831 | } | |
7832 | ||
b68f3cc7 | 7833 | #ifdef CONFIG_X86_64 |
ee2cd4b7 | 7834 | static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf) |
660a5d51 | 7835 | { |
660a5d51 PB |
7836 | struct desc_ptr dt; |
7837 | struct kvm_segment seg; | |
7838 | unsigned long val; | |
7839 | int i; | |
7840 | ||
7841 | for (i = 0; i < 16; i++) | |
7842 | put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i)); | |
7843 | ||
7844 | put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu)); | |
7845 | put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu)); | |
7846 | ||
7847 | kvm_get_dr(vcpu, 6, &val); | |
7848 | put_smstate(u64, buf, 0x7f68, val); | |
7849 | kvm_get_dr(vcpu, 7, &val); | |
7850 | put_smstate(u64, buf, 0x7f60, val); | |
7851 | ||
7852 | put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu)); | |
7853 | put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu)); | |
7854 | put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu)); | |
7855 | ||
7856 | put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase); | |
7857 | ||
7858 | /* revision id */ | |
7859 | put_smstate(u32, buf, 0x7efc, 0x00020064); | |
7860 | ||
7861 | put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer); | |
7862 | ||
7863 | kvm_get_segment(vcpu, &seg, VCPU_SREG_TR); | |
7864 | put_smstate(u16, buf, 0x7e90, seg.selector); | |
ee2cd4b7 | 7865 | put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8); |
660a5d51 PB |
7866 | put_smstate(u32, buf, 0x7e94, seg.limit); |
7867 | put_smstate(u64, buf, 0x7e98, seg.base); | |
7868 | ||
7869 | kvm_x86_ops->get_idt(vcpu, &dt); | |
7870 | put_smstate(u32, buf, 0x7e84, dt.size); | |
7871 | put_smstate(u64, buf, 0x7e88, dt.address); | |
7872 | ||
7873 | kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR); | |
7874 | put_smstate(u16, buf, 0x7e70, seg.selector); | |
ee2cd4b7 | 7875 | put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8); |
660a5d51 PB |
7876 | put_smstate(u32, buf, 0x7e74, seg.limit); |
7877 | put_smstate(u64, buf, 0x7e78, seg.base); | |
7878 | ||
7879 | kvm_x86_ops->get_gdt(vcpu, &dt); | |
7880 | put_smstate(u32, buf, 0x7e64, dt.size); | |
7881 | put_smstate(u64, buf, 0x7e68, dt.address); | |
7882 | ||
7883 | for (i = 0; i < 6; i++) | |
ee2cd4b7 | 7884 | enter_smm_save_seg_64(vcpu, buf, i); |
660a5d51 | 7885 | } |
b68f3cc7 | 7886 | #endif |
660a5d51 | 7887 | |
ee2cd4b7 | 7888 | static void enter_smm(struct kvm_vcpu *vcpu) |
64d60670 | 7889 | { |
660a5d51 | 7890 | struct kvm_segment cs, ds; |
18c3626e | 7891 | struct desc_ptr dt; |
660a5d51 PB |
7892 | char buf[512]; |
7893 | u32 cr0; | |
7894 | ||
660a5d51 | 7895 | trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true); |
660a5d51 | 7896 | memset(buf, 0, 512); |
b68f3cc7 | 7897 | #ifdef CONFIG_X86_64 |
d6321d49 | 7898 | if (guest_cpuid_has(vcpu, X86_FEATURE_LM)) |
ee2cd4b7 | 7899 | enter_smm_save_state_64(vcpu, buf); |
660a5d51 | 7900 | else |
b68f3cc7 | 7901 | #endif |
ee2cd4b7 | 7902 | enter_smm_save_state_32(vcpu, buf); |
660a5d51 | 7903 | |
0234bf88 LP |
7904 | /* |
7905 | * Give pre_enter_smm() a chance to make ISA-specific changes to the | |
7906 | * vCPU state (e.g. leave guest mode) after we've saved the state into | |
7907 | * the SMM state-save area. | |
7908 | */ | |
7909 | kvm_x86_ops->pre_enter_smm(vcpu, buf); | |
7910 | ||
7911 | vcpu->arch.hflags |= HF_SMM_MASK; | |
54bf36aa | 7912 | kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf)); |
660a5d51 PB |
7913 | |
7914 | if (kvm_x86_ops->get_nmi_mask(vcpu)) | |
7915 | vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK; | |
7916 | else | |
7917 | kvm_x86_ops->set_nmi_mask(vcpu, true); | |
7918 | ||
7919 | kvm_set_rflags(vcpu, X86_EFLAGS_FIXED); | |
7920 | kvm_rip_write(vcpu, 0x8000); | |
7921 | ||
7922 | cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG); | |
7923 | kvm_x86_ops->set_cr0(vcpu, cr0); | |
7924 | vcpu->arch.cr0 = cr0; | |
7925 | ||
7926 | kvm_x86_ops->set_cr4(vcpu, 0); | |
7927 | ||
18c3626e PB |
7928 | /* Undocumented: IDT limit is set to zero on entry to SMM. */ |
7929 | dt.address = dt.size = 0; | |
7930 | kvm_x86_ops->set_idt(vcpu, &dt); | |
7931 | ||
660a5d51 PB |
7932 | __kvm_set_dr(vcpu, 7, DR7_FIXED_1); |
7933 | ||
7934 | cs.selector = (vcpu->arch.smbase >> 4) & 0xffff; | |
7935 | cs.base = vcpu->arch.smbase; | |
7936 | ||
7937 | ds.selector = 0; | |
7938 | ds.base = 0; | |
7939 | ||
7940 | cs.limit = ds.limit = 0xffffffff; | |
7941 | cs.type = ds.type = 0x3; | |
7942 | cs.dpl = ds.dpl = 0; | |
7943 | cs.db = ds.db = 0; | |
7944 | cs.s = ds.s = 1; | |
7945 | cs.l = ds.l = 0; | |
7946 | cs.g = ds.g = 1; | |
7947 | cs.avl = ds.avl = 0; | |
7948 | cs.present = ds.present = 1; | |
7949 | cs.unusable = ds.unusable = 0; | |
7950 | cs.padding = ds.padding = 0; | |
7951 | ||
7952 | kvm_set_segment(vcpu, &cs, VCPU_SREG_CS); | |
7953 | kvm_set_segment(vcpu, &ds, VCPU_SREG_DS); | |
7954 | kvm_set_segment(vcpu, &ds, VCPU_SREG_ES); | |
7955 | kvm_set_segment(vcpu, &ds, VCPU_SREG_FS); | |
7956 | kvm_set_segment(vcpu, &ds, VCPU_SREG_GS); | |
7957 | kvm_set_segment(vcpu, &ds, VCPU_SREG_SS); | |
7958 | ||
b68f3cc7 | 7959 | #ifdef CONFIG_X86_64 |
d6321d49 | 7960 | if (guest_cpuid_has(vcpu, X86_FEATURE_LM)) |
660a5d51 | 7961 | kvm_x86_ops->set_efer(vcpu, 0); |
b68f3cc7 | 7962 | #endif |
660a5d51 PB |
7963 | |
7964 | kvm_update_cpuid(vcpu); | |
7965 | kvm_mmu_reset_context(vcpu); | |
64d60670 PB |
7966 | } |
7967 | ||
ee2cd4b7 | 7968 | static void process_smi(struct kvm_vcpu *vcpu) |
c43203ca PB |
7969 | { |
7970 | vcpu->arch.smi_pending = true; | |
7971 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
7972 | } | |
7973 | ||
7ee30bc1 NNL |
7974 | void kvm_make_scan_ioapic_request_mask(struct kvm *kvm, |
7975 | unsigned long *vcpu_bitmap) | |
7976 | { | |
7977 | cpumask_var_t cpus; | |
7ee30bc1 NNL |
7978 | |
7979 | zalloc_cpumask_var(&cpus, GFP_ATOMIC); | |
7980 | ||
db5a95ec MW |
7981 | kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC, |
7982 | vcpu_bitmap, cpus); | |
7ee30bc1 NNL |
7983 | |
7984 | free_cpumask_var(cpus); | |
7985 | } | |
7986 | ||
2860c4b1 PB |
7987 | void kvm_make_scan_ioapic_request(struct kvm *kvm) |
7988 | { | |
7989 | kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC); | |
7990 | } | |
7991 | ||
3d81bc7e | 7992 | static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu) |
c7c9c56c | 7993 | { |
dcbd3e49 | 7994 | if (!kvm_apic_present(vcpu)) |
3d81bc7e | 7995 | return; |
c7c9c56c | 7996 | |
6308630b | 7997 | bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256); |
c7c9c56c | 7998 | |
b053b2ae | 7999 | if (irqchip_split(vcpu->kvm)) |
6308630b | 8000 | kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors); |
db2bdcbb | 8001 | else { |
fa59cc00 | 8002 | if (vcpu->arch.apicv_active) |
d62caabb | 8003 | kvm_x86_ops->sync_pir_to_irr(vcpu); |
e97f852f WL |
8004 | if (ioapic_in_kernel(vcpu->kvm)) |
8005 | kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors); | |
db2bdcbb | 8006 | } |
e40ff1d6 LA |
8007 | |
8008 | if (is_guest_mode(vcpu)) | |
8009 | vcpu->arch.load_eoi_exitmap_pending = true; | |
8010 | else | |
8011 | kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu); | |
8012 | } | |
8013 | ||
8014 | static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu) | |
8015 | { | |
8016 | u64 eoi_exit_bitmap[4]; | |
8017 | ||
8018 | if (!kvm_apic_hw_enabled(vcpu->arch.apic)) | |
8019 | return; | |
8020 | ||
5c919412 AS |
8021 | bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors, |
8022 | vcpu_to_synic(vcpu)->vec_bitmap, 256); | |
8023 | kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap); | |
c7c9c56c YZ |
8024 | } |
8025 | ||
93065ac7 MH |
8026 | int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm, |
8027 | unsigned long start, unsigned long end, | |
8028 | bool blockable) | |
b1394e74 RK |
8029 | { |
8030 | unsigned long apic_address; | |
8031 | ||
8032 | /* | |
8033 | * The physical address of apic access page is stored in the VMCS. | |
8034 | * Update it when it becomes invalid. | |
8035 | */ | |
8036 | apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT); | |
8037 | if (start <= apic_address && apic_address < end) | |
8038 | kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD); | |
93065ac7 MH |
8039 | |
8040 | return 0; | |
b1394e74 RK |
8041 | } |
8042 | ||
4256f43f TC |
8043 | void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu) |
8044 | { | |
c24ae0dc TC |
8045 | struct page *page = NULL; |
8046 | ||
35754c98 | 8047 | if (!lapic_in_kernel(vcpu)) |
f439ed27 PB |
8048 | return; |
8049 | ||
4256f43f TC |
8050 | if (!kvm_x86_ops->set_apic_access_page_addr) |
8051 | return; | |
8052 | ||
c24ae0dc | 8053 | page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT); |
e8fd5e9e AA |
8054 | if (is_error_page(page)) |
8055 | return; | |
c24ae0dc TC |
8056 | kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page)); |
8057 | ||
8058 | /* | |
8059 | * Do not pin apic access page in memory, the MMU notifier | |
8060 | * will call us again if it is migrated or swapped out. | |
8061 | */ | |
8062 | put_page(page); | |
4256f43f | 8063 | } |
4256f43f | 8064 | |
d264ee0c SC |
8065 | void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu) |
8066 | { | |
8067 | smp_send_reschedule(vcpu->cpu); | |
8068 | } | |
8069 | EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit); | |
8070 | ||
9357d939 | 8071 | /* |
362c698f | 8072 | * Returns 1 to let vcpu_run() continue the guest execution loop without |
9357d939 TY |
8073 | * exiting to the userspace. Otherwise, the value will be returned to the |
8074 | * userspace. | |
8075 | */ | |
851ba692 | 8076 | static int vcpu_enter_guest(struct kvm_vcpu *vcpu) |
b6c7a5dc HB |
8077 | { |
8078 | int r; | |
62a193ed MG |
8079 | bool req_int_win = |
8080 | dm_request_for_irq_injection(vcpu) && | |
8081 | kvm_cpu_accept_dm_intr(vcpu); | |
1e9e2622 | 8082 | enum exit_fastpath_completion exit_fastpath = EXIT_FASTPATH_NONE; |
62a193ed | 8083 | |
730dca42 | 8084 | bool req_immediate_exit = false; |
b6c7a5dc | 8085 | |
2fa6e1e1 | 8086 | if (kvm_request_pending(vcpu)) { |
671ddc70 JM |
8087 | if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu)) { |
8088 | if (unlikely(!kvm_x86_ops->get_vmcs12_pages(vcpu))) { | |
8089 | r = 0; | |
8090 | goto out; | |
8091 | } | |
8092 | } | |
a8eeb04a | 8093 | if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) |
2e53d63a | 8094 | kvm_mmu_unload(vcpu); |
a8eeb04a | 8095 | if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu)) |
2f599714 | 8096 | __kvm_migrate_timers(vcpu); |
d828199e MT |
8097 | if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu)) |
8098 | kvm_gen_update_masterclock(vcpu->kvm); | |
0061d53d MT |
8099 | if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu)) |
8100 | kvm_gen_kvmclock_update(vcpu); | |
34c238a1 ZA |
8101 | if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) { |
8102 | r = kvm_guest_time_update(vcpu); | |
8cfdc000 ZA |
8103 | if (unlikely(r)) |
8104 | goto out; | |
8105 | } | |
a8eeb04a | 8106 | if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu)) |
4731d4c7 | 8107 | kvm_mmu_sync_roots(vcpu); |
6e42782f JS |
8108 | if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu)) |
8109 | kvm_mmu_load_cr3(vcpu); | |
a8eeb04a | 8110 | if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) |
c2ba05cc | 8111 | kvm_vcpu_flush_tlb(vcpu, true); |
a8eeb04a | 8112 | if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) { |
851ba692 | 8113 | vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS; |
b93463aa AK |
8114 | r = 0; |
8115 | goto out; | |
8116 | } | |
a8eeb04a | 8117 | if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) { |
851ba692 | 8118 | vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN; |
bbeac283 | 8119 | vcpu->mmio_needed = 0; |
71c4dfaf JR |
8120 | r = 0; |
8121 | goto out; | |
8122 | } | |
af585b92 GN |
8123 | if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) { |
8124 | /* Page is swapped out. Do synthetic halt */ | |
8125 | vcpu->arch.apf.halted = true; | |
8126 | r = 1; | |
8127 | goto out; | |
8128 | } | |
c9aaa895 GC |
8129 | if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu)) |
8130 | record_steal_time(vcpu); | |
64d60670 PB |
8131 | if (kvm_check_request(KVM_REQ_SMI, vcpu)) |
8132 | process_smi(vcpu); | |
7460fb4a AK |
8133 | if (kvm_check_request(KVM_REQ_NMI, vcpu)) |
8134 | process_nmi(vcpu); | |
f5132b01 | 8135 | if (kvm_check_request(KVM_REQ_PMU, vcpu)) |
c6702c9d | 8136 | kvm_pmu_handle_event(vcpu); |
f5132b01 | 8137 | if (kvm_check_request(KVM_REQ_PMI, vcpu)) |
c6702c9d | 8138 | kvm_pmu_deliver_pmi(vcpu); |
7543a635 SR |
8139 | if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) { |
8140 | BUG_ON(vcpu->arch.pending_ioapic_eoi > 255); | |
8141 | if (test_bit(vcpu->arch.pending_ioapic_eoi, | |
6308630b | 8142 | vcpu->arch.ioapic_handled_vectors)) { |
7543a635 SR |
8143 | vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI; |
8144 | vcpu->run->eoi.vector = | |
8145 | vcpu->arch.pending_ioapic_eoi; | |
8146 | r = 0; | |
8147 | goto out; | |
8148 | } | |
8149 | } | |
3d81bc7e YZ |
8150 | if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu)) |
8151 | vcpu_scan_ioapic(vcpu); | |
e40ff1d6 LA |
8152 | if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu)) |
8153 | vcpu_load_eoi_exitmap(vcpu); | |
4256f43f TC |
8154 | if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu)) |
8155 | kvm_vcpu_reload_apic_access_page(vcpu); | |
2ce79189 AS |
8156 | if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) { |
8157 | vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT; | |
8158 | vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH; | |
8159 | r = 0; | |
8160 | goto out; | |
8161 | } | |
e516cebb AS |
8162 | if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) { |
8163 | vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT; | |
8164 | vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET; | |
8165 | r = 0; | |
8166 | goto out; | |
8167 | } | |
db397571 AS |
8168 | if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) { |
8169 | vcpu->run->exit_reason = KVM_EXIT_HYPERV; | |
8170 | vcpu->run->hyperv = vcpu->arch.hyperv.exit; | |
8171 | r = 0; | |
8172 | goto out; | |
8173 | } | |
f3b138c5 AS |
8174 | |
8175 | /* | |
8176 | * KVM_REQ_HV_STIMER has to be processed after | |
8177 | * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers | |
8178 | * depend on the guest clock being up-to-date | |
8179 | */ | |
1f4b34f8 AS |
8180 | if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu)) |
8181 | kvm_hv_process_stimers(vcpu); | |
2f52d58c | 8182 | } |
b93463aa | 8183 | |
b463a6f7 | 8184 | if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) { |
0f1e261e | 8185 | ++vcpu->stat.req_event; |
66450a21 JK |
8186 | kvm_apic_accept_events(vcpu); |
8187 | if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) { | |
8188 | r = 1; | |
8189 | goto out; | |
8190 | } | |
8191 | ||
b6b8a145 JK |
8192 | if (inject_pending_event(vcpu, req_int_win) != 0) |
8193 | req_immediate_exit = true; | |
321c5658 | 8194 | else { |
cc3d967f | 8195 | /* Enable SMI/NMI/IRQ window open exits if needed. |
c43203ca | 8196 | * |
cc3d967f LP |
8197 | * SMIs have three cases: |
8198 | * 1) They can be nested, and then there is nothing to | |
8199 | * do here because RSM will cause a vmexit anyway. | |
8200 | * 2) There is an ISA-specific reason why SMI cannot be | |
8201 | * injected, and the moment when this changes can be | |
8202 | * intercepted. | |
8203 | * 3) Or the SMI can be pending because | |
8204 | * inject_pending_event has completed the injection | |
8205 | * of an IRQ or NMI from the previous vmexit, and | |
8206 | * then we request an immediate exit to inject the | |
8207 | * SMI. | |
c43203ca PB |
8208 | */ |
8209 | if (vcpu->arch.smi_pending && !is_smm(vcpu)) | |
cc3d967f LP |
8210 | if (!kvm_x86_ops->enable_smi_window(vcpu)) |
8211 | req_immediate_exit = true; | |
321c5658 YS |
8212 | if (vcpu->arch.nmi_pending) |
8213 | kvm_x86_ops->enable_nmi_window(vcpu); | |
8214 | if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win) | |
8215 | kvm_x86_ops->enable_irq_window(vcpu); | |
664f8e26 | 8216 | WARN_ON(vcpu->arch.exception.pending); |
321c5658 | 8217 | } |
b463a6f7 AK |
8218 | |
8219 | if (kvm_lapic_enabled(vcpu)) { | |
8220 | update_cr8_intercept(vcpu); | |
8221 | kvm_lapic_sync_to_vapic(vcpu); | |
8222 | } | |
8223 | } | |
8224 | ||
d8368af8 AK |
8225 | r = kvm_mmu_reload(vcpu); |
8226 | if (unlikely(r)) { | |
d905c069 | 8227 | goto cancel_injection; |
d8368af8 AK |
8228 | } |
8229 | ||
b6c7a5dc HB |
8230 | preempt_disable(); |
8231 | ||
8232 | kvm_x86_ops->prepare_guest_switch(vcpu); | |
b95234c8 PB |
8233 | |
8234 | /* | |
8235 | * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt | |
8236 | * IPI are then delayed after guest entry, which ensures that they | |
8237 | * result in virtual interrupt delivery. | |
8238 | */ | |
8239 | local_irq_disable(); | |
6b7e2d09 XG |
8240 | vcpu->mode = IN_GUEST_MODE; |
8241 | ||
01b71917 MT |
8242 | srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); |
8243 | ||
0f127d12 | 8244 | /* |
b95234c8 | 8245 | * 1) We should set ->mode before checking ->requests. Please see |
cde9af6e | 8246 | * the comment in kvm_vcpu_exiting_guest_mode(). |
b95234c8 | 8247 | * |
81b01667 | 8248 | * 2) For APICv, we should set ->mode before checking PID.ON. This |
b95234c8 PB |
8249 | * pairs with the memory barrier implicit in pi_test_and_set_on |
8250 | * (see vmx_deliver_posted_interrupt). | |
8251 | * | |
8252 | * 3) This also orders the write to mode from any reads to the page | |
8253 | * tables done while the VCPU is running. Please see the comment | |
8254 | * in kvm_flush_remote_tlbs. | |
6b7e2d09 | 8255 | */ |
01b71917 | 8256 | smp_mb__after_srcu_read_unlock(); |
b6c7a5dc | 8257 | |
b95234c8 PB |
8258 | /* |
8259 | * This handles the case where a posted interrupt was | |
8260 | * notified with kvm_vcpu_kick. | |
8261 | */ | |
fa59cc00 LA |
8262 | if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active) |
8263 | kvm_x86_ops->sync_pir_to_irr(vcpu); | |
32f88400 | 8264 | |
2fa6e1e1 | 8265 | if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) |
d94e1dc9 | 8266 | || need_resched() || signal_pending(current)) { |
6b7e2d09 | 8267 | vcpu->mode = OUTSIDE_GUEST_MODE; |
d94e1dc9 | 8268 | smp_wmb(); |
6c142801 AK |
8269 | local_irq_enable(); |
8270 | preempt_enable(); | |
01b71917 | 8271 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); |
6c142801 | 8272 | r = 1; |
d905c069 | 8273 | goto cancel_injection; |
6c142801 AK |
8274 | } |
8275 | ||
c43203ca PB |
8276 | if (req_immediate_exit) { |
8277 | kvm_make_request(KVM_REQ_EVENT, vcpu); | |
d264ee0c | 8278 | kvm_x86_ops->request_immediate_exit(vcpu); |
c43203ca | 8279 | } |
d6185f20 | 8280 | |
8b89fe1f | 8281 | trace_kvm_entry(vcpu->vcpu_id); |
6edaa530 | 8282 | guest_enter_irqoff(); |
b6c7a5dc | 8283 | |
2620fe26 SC |
8284 | fpregs_assert_state_consistent(); |
8285 | if (test_thread_flag(TIF_NEED_FPU_LOAD)) | |
8286 | switch_fpu_return(); | |
5f409e20 | 8287 | |
42dbaa5a | 8288 | if (unlikely(vcpu->arch.switch_db_regs)) { |
42dbaa5a JK |
8289 | set_debugreg(0, 7); |
8290 | set_debugreg(vcpu->arch.eff_db[0], 0); | |
8291 | set_debugreg(vcpu->arch.eff_db[1], 1); | |
8292 | set_debugreg(vcpu->arch.eff_db[2], 2); | |
8293 | set_debugreg(vcpu->arch.eff_db[3], 3); | |
c77fb5fe | 8294 | set_debugreg(vcpu->arch.dr6, 6); |
ae561ede | 8295 | vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD; |
42dbaa5a | 8296 | } |
b6c7a5dc | 8297 | |
851ba692 | 8298 | kvm_x86_ops->run(vcpu); |
b6c7a5dc | 8299 | |
c77fb5fe PB |
8300 | /* |
8301 | * Do this here before restoring debug registers on the host. And | |
8302 | * since we do this before handling the vmexit, a DR access vmexit | |
8303 | * can (a) read the correct value of the debug registers, (b) set | |
8304 | * KVM_DEBUGREG_WONT_EXIT again. | |
8305 | */ | |
8306 | if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) { | |
c77fb5fe PB |
8307 | WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP); |
8308 | kvm_x86_ops->sync_dirty_debug_regs(vcpu); | |
70e4da7a PB |
8309 | kvm_update_dr0123(vcpu); |
8310 | kvm_update_dr6(vcpu); | |
8311 | kvm_update_dr7(vcpu); | |
8312 | vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD; | |
c77fb5fe PB |
8313 | } |
8314 | ||
24f1e32c FW |
8315 | /* |
8316 | * If the guest has used debug registers, at least dr7 | |
8317 | * will be disabled while returning to the host. | |
8318 | * If we don't have active breakpoints in the host, we don't | |
8319 | * care about the messed up debug address registers. But if | |
8320 | * we have some of them active, restore the old state. | |
8321 | */ | |
59d8eb53 | 8322 | if (hw_breakpoint_active()) |
24f1e32c | 8323 | hw_breakpoint_restore(); |
42dbaa5a | 8324 | |
4ba76538 | 8325 | vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc()); |
1d5f066e | 8326 | |
6b7e2d09 | 8327 | vcpu->mode = OUTSIDE_GUEST_MODE; |
d94e1dc9 | 8328 | smp_wmb(); |
a547c6db | 8329 | |
1e9e2622 | 8330 | kvm_x86_ops->handle_exit_irqoff(vcpu, &exit_fastpath); |
b6c7a5dc | 8331 | |
d7a08882 SC |
8332 | /* |
8333 | * Consume any pending interrupts, including the possible source of | |
8334 | * VM-Exit on SVM and any ticks that occur between VM-Exit and now. | |
8335 | * An instruction is required after local_irq_enable() to fully unblock | |
8336 | * interrupts on processors that implement an interrupt shadow, the | |
8337 | * stat.exits increment will do nicely. | |
8338 | */ | |
8339 | kvm_before_interrupt(vcpu); | |
8340 | local_irq_enable(); | |
b6c7a5dc | 8341 | ++vcpu->stat.exits; |
d7a08882 SC |
8342 | local_irq_disable(); |
8343 | kvm_after_interrupt(vcpu); | |
b6c7a5dc | 8344 | |
f2485b3e | 8345 | guest_exit_irqoff(); |
ec0671d5 WL |
8346 | if (lapic_in_kernel(vcpu)) { |
8347 | s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta; | |
8348 | if (delta != S64_MIN) { | |
8349 | trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta); | |
8350 | vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN; | |
8351 | } | |
8352 | } | |
b6c7a5dc | 8353 | |
f2485b3e | 8354 | local_irq_enable(); |
b6c7a5dc HB |
8355 | preempt_enable(); |
8356 | ||
f656ce01 | 8357 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); |
3200f405 | 8358 | |
b6c7a5dc HB |
8359 | /* |
8360 | * Profile KVM exit RIPs: | |
8361 | */ | |
8362 | if (unlikely(prof_on == KVM_PROFILING)) { | |
5fdbf976 MT |
8363 | unsigned long rip = kvm_rip_read(vcpu); |
8364 | profile_hit(KVM_PROFILING, (void *)rip); | |
b6c7a5dc HB |
8365 | } |
8366 | ||
cc578287 ZA |
8367 | if (unlikely(vcpu->arch.tsc_always_catchup)) |
8368 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); | |
298101da | 8369 | |
5cfb1d5a MT |
8370 | if (vcpu->arch.apic_attention) |
8371 | kvm_lapic_sync_from_vapic(vcpu); | |
b93463aa | 8372 | |
618232e2 | 8373 | vcpu->arch.gpa_available = false; |
1e9e2622 | 8374 | r = kvm_x86_ops->handle_exit(vcpu, exit_fastpath); |
d905c069 MT |
8375 | return r; |
8376 | ||
8377 | cancel_injection: | |
8378 | kvm_x86_ops->cancel_injection(vcpu); | |
ae7a2a3f MT |
8379 | if (unlikely(vcpu->arch.apic_attention)) |
8380 | kvm_lapic_sync_from_vapic(vcpu); | |
d7690175 MT |
8381 | out: |
8382 | return r; | |
8383 | } | |
b6c7a5dc | 8384 | |
362c698f PB |
8385 | static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu) |
8386 | { | |
bf9f6ac8 FW |
8387 | if (!kvm_arch_vcpu_runnable(vcpu) && |
8388 | (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) { | |
9c8fd1ba PB |
8389 | srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); |
8390 | kvm_vcpu_block(vcpu); | |
8391 | vcpu->srcu_idx = srcu_read_lock(&kvm->srcu); | |
bf9f6ac8 FW |
8392 | |
8393 | if (kvm_x86_ops->post_block) | |
8394 | kvm_x86_ops->post_block(vcpu); | |
8395 | ||
9c8fd1ba PB |
8396 | if (!kvm_check_request(KVM_REQ_UNHALT, vcpu)) |
8397 | return 1; | |
8398 | } | |
362c698f PB |
8399 | |
8400 | kvm_apic_accept_events(vcpu); | |
8401 | switch(vcpu->arch.mp_state) { | |
8402 | case KVM_MP_STATE_HALTED: | |
8403 | vcpu->arch.pv.pv_unhalted = false; | |
8404 | vcpu->arch.mp_state = | |
8405 | KVM_MP_STATE_RUNNABLE; | |
b2869f28 | 8406 | /* fall through */ |
362c698f PB |
8407 | case KVM_MP_STATE_RUNNABLE: |
8408 | vcpu->arch.apf.halted = false; | |
8409 | break; | |
8410 | case KVM_MP_STATE_INIT_RECEIVED: | |
8411 | break; | |
8412 | default: | |
8413 | return -EINTR; | |
8414 | break; | |
8415 | } | |
8416 | return 1; | |
8417 | } | |
09cec754 | 8418 | |
5d9bc648 PB |
8419 | static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu) |
8420 | { | |
0ad3bed6 PB |
8421 | if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) |
8422 | kvm_x86_ops->check_nested_events(vcpu, false); | |
8423 | ||
5d9bc648 PB |
8424 | return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE && |
8425 | !vcpu->arch.apf.halted); | |
8426 | } | |
8427 | ||
362c698f | 8428 | static int vcpu_run(struct kvm_vcpu *vcpu) |
d7690175 MT |
8429 | { |
8430 | int r; | |
f656ce01 | 8431 | struct kvm *kvm = vcpu->kvm; |
d7690175 | 8432 | |
f656ce01 | 8433 | vcpu->srcu_idx = srcu_read_lock(&kvm->srcu); |
c595ceee | 8434 | vcpu->arch.l1tf_flush_l1d = true; |
d7690175 | 8435 | |
362c698f | 8436 | for (;;) { |
58f800d5 | 8437 | if (kvm_vcpu_running(vcpu)) { |
851ba692 | 8438 | r = vcpu_enter_guest(vcpu); |
bf9f6ac8 | 8439 | } else { |
362c698f | 8440 | r = vcpu_block(kvm, vcpu); |
bf9f6ac8 FW |
8441 | } |
8442 | ||
09cec754 GN |
8443 | if (r <= 0) |
8444 | break; | |
8445 | ||
72875d8a | 8446 | kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu); |
09cec754 GN |
8447 | if (kvm_cpu_has_pending_timer(vcpu)) |
8448 | kvm_inject_pending_timer_irqs(vcpu); | |
8449 | ||
782d422b MG |
8450 | if (dm_request_for_irq_injection(vcpu) && |
8451 | kvm_vcpu_ready_for_interrupt_injection(vcpu)) { | |
4ca7dd8c PB |
8452 | r = 0; |
8453 | vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN; | |
09cec754 | 8454 | ++vcpu->stat.request_irq_exits; |
362c698f | 8455 | break; |
09cec754 | 8456 | } |
af585b92 GN |
8457 | |
8458 | kvm_check_async_pf_completion(vcpu); | |
8459 | ||
09cec754 GN |
8460 | if (signal_pending(current)) { |
8461 | r = -EINTR; | |
851ba692 | 8462 | vcpu->run->exit_reason = KVM_EXIT_INTR; |
09cec754 | 8463 | ++vcpu->stat.signal_exits; |
362c698f | 8464 | break; |
09cec754 GN |
8465 | } |
8466 | if (need_resched()) { | |
f656ce01 | 8467 | srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); |
c08ac06a | 8468 | cond_resched(); |
f656ce01 | 8469 | vcpu->srcu_idx = srcu_read_lock(&kvm->srcu); |
d7690175 | 8470 | } |
b6c7a5dc HB |
8471 | } |
8472 | ||
f656ce01 | 8473 | srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx); |
b6c7a5dc HB |
8474 | |
8475 | return r; | |
8476 | } | |
8477 | ||
716d51ab GN |
8478 | static inline int complete_emulated_io(struct kvm_vcpu *vcpu) |
8479 | { | |
8480 | int r; | |
60fc3d02 | 8481 | |
716d51ab | 8482 | vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu); |
0ce97a2b | 8483 | r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE); |
716d51ab | 8484 | srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); |
60fc3d02 | 8485 | return r; |
716d51ab GN |
8486 | } |
8487 | ||
8488 | static int complete_emulated_pio(struct kvm_vcpu *vcpu) | |
8489 | { | |
8490 | BUG_ON(!vcpu->arch.pio.count); | |
8491 | ||
8492 | return complete_emulated_io(vcpu); | |
8493 | } | |
8494 | ||
f78146b0 AK |
8495 | /* |
8496 | * Implements the following, as a state machine: | |
8497 | * | |
8498 | * read: | |
8499 | * for each fragment | |
87da7e66 XG |
8500 | * for each mmio piece in the fragment |
8501 | * write gpa, len | |
8502 | * exit | |
8503 | * copy data | |
f78146b0 AK |
8504 | * execute insn |
8505 | * | |
8506 | * write: | |
8507 | * for each fragment | |
87da7e66 XG |
8508 | * for each mmio piece in the fragment |
8509 | * write gpa, len | |
8510 | * copy data | |
8511 | * exit | |
f78146b0 | 8512 | */ |
716d51ab | 8513 | static int complete_emulated_mmio(struct kvm_vcpu *vcpu) |
5287f194 AK |
8514 | { |
8515 | struct kvm_run *run = vcpu->run; | |
f78146b0 | 8516 | struct kvm_mmio_fragment *frag; |
87da7e66 | 8517 | unsigned len; |
5287f194 | 8518 | |
716d51ab | 8519 | BUG_ON(!vcpu->mmio_needed); |
5287f194 | 8520 | |
716d51ab | 8521 | /* Complete previous fragment */ |
87da7e66 XG |
8522 | frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment]; |
8523 | len = min(8u, frag->len); | |
716d51ab | 8524 | if (!vcpu->mmio_is_write) |
87da7e66 XG |
8525 | memcpy(frag->data, run->mmio.data, len); |
8526 | ||
8527 | if (frag->len <= 8) { | |
8528 | /* Switch to the next fragment. */ | |
8529 | frag++; | |
8530 | vcpu->mmio_cur_fragment++; | |
8531 | } else { | |
8532 | /* Go forward to the next mmio piece. */ | |
8533 | frag->data += len; | |
8534 | frag->gpa += len; | |
8535 | frag->len -= len; | |
8536 | } | |
8537 | ||
a08d3b3b | 8538 | if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) { |
716d51ab | 8539 | vcpu->mmio_needed = 0; |
0912c977 PB |
8540 | |
8541 | /* FIXME: return into emulator if single-stepping. */ | |
cef4dea0 | 8542 | if (vcpu->mmio_is_write) |
716d51ab GN |
8543 | return 1; |
8544 | vcpu->mmio_read_completed = 1; | |
8545 | return complete_emulated_io(vcpu); | |
8546 | } | |
87da7e66 | 8547 | |
716d51ab GN |
8548 | run->exit_reason = KVM_EXIT_MMIO; |
8549 | run->mmio.phys_addr = frag->gpa; | |
8550 | if (vcpu->mmio_is_write) | |
87da7e66 XG |
8551 | memcpy(run->mmio.data, frag->data, min(8u, frag->len)); |
8552 | run->mmio.len = min(8u, frag->len); | |
716d51ab GN |
8553 | run->mmio.is_write = vcpu->mmio_is_write; |
8554 | vcpu->arch.complete_userspace_io = complete_emulated_mmio; | |
8555 | return 0; | |
5287f194 AK |
8556 | } |
8557 | ||
c9aef3b8 SC |
8558 | static void kvm_save_current_fpu(struct fpu *fpu) |
8559 | { | |
8560 | /* | |
8561 | * If the target FPU state is not resident in the CPU registers, just | |
8562 | * memcpy() from current, else save CPU state directly to the target. | |
8563 | */ | |
8564 | if (test_thread_flag(TIF_NEED_FPU_LOAD)) | |
8565 | memcpy(&fpu->state, ¤t->thread.fpu.state, | |
8566 | fpu_kernel_xstate_size); | |
8567 | else | |
8568 | copy_fpregs_to_fpstate(fpu); | |
8569 | } | |
8570 | ||
822f312d SAS |
8571 | /* Swap (qemu) user FPU context for the guest FPU context. */ |
8572 | static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu) | |
8573 | { | |
5f409e20 RR |
8574 | fpregs_lock(); |
8575 | ||
c9aef3b8 SC |
8576 | kvm_save_current_fpu(vcpu->arch.user_fpu); |
8577 | ||
822f312d | 8578 | /* PKRU is separately restored in kvm_x86_ops->run. */ |
b666a4b6 | 8579 | __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state, |
822f312d | 8580 | ~XFEATURE_MASK_PKRU); |
5f409e20 RR |
8581 | |
8582 | fpregs_mark_activate(); | |
8583 | fpregs_unlock(); | |
8584 | ||
822f312d SAS |
8585 | trace_kvm_fpu(1); |
8586 | } | |
8587 | ||
8588 | /* When vcpu_run ends, restore user space FPU context. */ | |
8589 | static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu) | |
8590 | { | |
5f409e20 RR |
8591 | fpregs_lock(); |
8592 | ||
c9aef3b8 SC |
8593 | kvm_save_current_fpu(vcpu->arch.guest_fpu); |
8594 | ||
d9a710e5 | 8595 | copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state); |
5f409e20 RR |
8596 | |
8597 | fpregs_mark_activate(); | |
8598 | fpregs_unlock(); | |
8599 | ||
822f312d SAS |
8600 | ++vcpu->stat.fpu_reload; |
8601 | trace_kvm_fpu(0); | |
8602 | } | |
8603 | ||
b6c7a5dc HB |
8604 | int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run) |
8605 | { | |
8606 | int r; | |
b6c7a5dc | 8607 | |
accb757d | 8608 | vcpu_load(vcpu); |
20b7035c | 8609 | kvm_sigset_activate(vcpu); |
5663d8f9 PX |
8610 | kvm_load_guest_fpu(vcpu); |
8611 | ||
a4535290 | 8612 | if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) { |
2f173d26 JS |
8613 | if (kvm_run->immediate_exit) { |
8614 | r = -EINTR; | |
8615 | goto out; | |
8616 | } | |
b6c7a5dc | 8617 | kvm_vcpu_block(vcpu); |
66450a21 | 8618 | kvm_apic_accept_events(vcpu); |
72875d8a | 8619 | kvm_clear_request(KVM_REQ_UNHALT, vcpu); |
ac9f6dc0 | 8620 | r = -EAGAIN; |
a0595000 JS |
8621 | if (signal_pending(current)) { |
8622 | r = -EINTR; | |
8623 | vcpu->run->exit_reason = KVM_EXIT_INTR; | |
8624 | ++vcpu->stat.signal_exits; | |
8625 | } | |
ac9f6dc0 | 8626 | goto out; |
b6c7a5dc HB |
8627 | } |
8628 | ||
01643c51 KH |
8629 | if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) { |
8630 | r = -EINVAL; | |
8631 | goto out; | |
8632 | } | |
8633 | ||
8634 | if (vcpu->run->kvm_dirty_regs) { | |
8635 | r = sync_regs(vcpu); | |
8636 | if (r != 0) | |
8637 | goto out; | |
8638 | } | |
8639 | ||
b6c7a5dc | 8640 | /* re-sync apic's tpr */ |
35754c98 | 8641 | if (!lapic_in_kernel(vcpu)) { |
eea1cff9 AP |
8642 | if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) { |
8643 | r = -EINVAL; | |
8644 | goto out; | |
8645 | } | |
8646 | } | |
b6c7a5dc | 8647 | |
716d51ab GN |
8648 | if (unlikely(vcpu->arch.complete_userspace_io)) { |
8649 | int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io; | |
8650 | vcpu->arch.complete_userspace_io = NULL; | |
8651 | r = cui(vcpu); | |
8652 | if (r <= 0) | |
5663d8f9 | 8653 | goto out; |
716d51ab GN |
8654 | } else |
8655 | WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed); | |
5287f194 | 8656 | |
460df4c1 PB |
8657 | if (kvm_run->immediate_exit) |
8658 | r = -EINTR; | |
8659 | else | |
8660 | r = vcpu_run(vcpu); | |
b6c7a5dc HB |
8661 | |
8662 | out: | |
5663d8f9 | 8663 | kvm_put_guest_fpu(vcpu); |
01643c51 KH |
8664 | if (vcpu->run->kvm_valid_regs) |
8665 | store_regs(vcpu); | |
f1d86e46 | 8666 | post_kvm_run_save(vcpu); |
20b7035c | 8667 | kvm_sigset_deactivate(vcpu); |
b6c7a5dc | 8668 | |
accb757d | 8669 | vcpu_put(vcpu); |
b6c7a5dc HB |
8670 | return r; |
8671 | } | |
8672 | ||
01643c51 | 8673 | static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) |
b6c7a5dc | 8674 | { |
7ae441ea GN |
8675 | if (vcpu->arch.emulate_regs_need_sync_to_vcpu) { |
8676 | /* | |
8677 | * We are here if userspace calls get_regs() in the middle of | |
8678 | * instruction emulation. Registers state needs to be copied | |
4a969980 | 8679 | * back from emulation context to vcpu. Userspace shouldn't do |
7ae441ea GN |
8680 | * that usually, but some bad designed PV devices (vmware |
8681 | * backdoor interface) need this to work | |
8682 | */ | |
dd856efa | 8683 | emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt); |
7ae441ea GN |
8684 | vcpu->arch.emulate_regs_need_sync_to_vcpu = false; |
8685 | } | |
de3cd117 SC |
8686 | regs->rax = kvm_rax_read(vcpu); |
8687 | regs->rbx = kvm_rbx_read(vcpu); | |
8688 | regs->rcx = kvm_rcx_read(vcpu); | |
8689 | regs->rdx = kvm_rdx_read(vcpu); | |
8690 | regs->rsi = kvm_rsi_read(vcpu); | |
8691 | regs->rdi = kvm_rdi_read(vcpu); | |
e9c16c78 | 8692 | regs->rsp = kvm_rsp_read(vcpu); |
de3cd117 | 8693 | regs->rbp = kvm_rbp_read(vcpu); |
b6c7a5dc | 8694 | #ifdef CONFIG_X86_64 |
de3cd117 SC |
8695 | regs->r8 = kvm_r8_read(vcpu); |
8696 | regs->r9 = kvm_r9_read(vcpu); | |
8697 | regs->r10 = kvm_r10_read(vcpu); | |
8698 | regs->r11 = kvm_r11_read(vcpu); | |
8699 | regs->r12 = kvm_r12_read(vcpu); | |
8700 | regs->r13 = kvm_r13_read(vcpu); | |
8701 | regs->r14 = kvm_r14_read(vcpu); | |
8702 | regs->r15 = kvm_r15_read(vcpu); | |
b6c7a5dc HB |
8703 | #endif |
8704 | ||
5fdbf976 | 8705 | regs->rip = kvm_rip_read(vcpu); |
91586a3b | 8706 | regs->rflags = kvm_get_rflags(vcpu); |
01643c51 | 8707 | } |
b6c7a5dc | 8708 | |
01643c51 KH |
8709 | int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) |
8710 | { | |
8711 | vcpu_load(vcpu); | |
8712 | __get_regs(vcpu, regs); | |
1fc9b76b | 8713 | vcpu_put(vcpu); |
b6c7a5dc HB |
8714 | return 0; |
8715 | } | |
8716 | ||
01643c51 | 8717 | static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) |
b6c7a5dc | 8718 | { |
7ae441ea GN |
8719 | vcpu->arch.emulate_regs_need_sync_from_vcpu = true; |
8720 | vcpu->arch.emulate_regs_need_sync_to_vcpu = false; | |
8721 | ||
de3cd117 SC |
8722 | kvm_rax_write(vcpu, regs->rax); |
8723 | kvm_rbx_write(vcpu, regs->rbx); | |
8724 | kvm_rcx_write(vcpu, regs->rcx); | |
8725 | kvm_rdx_write(vcpu, regs->rdx); | |
8726 | kvm_rsi_write(vcpu, regs->rsi); | |
8727 | kvm_rdi_write(vcpu, regs->rdi); | |
e9c16c78 | 8728 | kvm_rsp_write(vcpu, regs->rsp); |
de3cd117 | 8729 | kvm_rbp_write(vcpu, regs->rbp); |
b6c7a5dc | 8730 | #ifdef CONFIG_X86_64 |
de3cd117 SC |
8731 | kvm_r8_write(vcpu, regs->r8); |
8732 | kvm_r9_write(vcpu, regs->r9); | |
8733 | kvm_r10_write(vcpu, regs->r10); | |
8734 | kvm_r11_write(vcpu, regs->r11); | |
8735 | kvm_r12_write(vcpu, regs->r12); | |
8736 | kvm_r13_write(vcpu, regs->r13); | |
8737 | kvm_r14_write(vcpu, regs->r14); | |
8738 | kvm_r15_write(vcpu, regs->r15); | |
b6c7a5dc HB |
8739 | #endif |
8740 | ||
5fdbf976 | 8741 | kvm_rip_write(vcpu, regs->rip); |
d73235d1 | 8742 | kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED); |
b6c7a5dc | 8743 | |
b4f14abd JK |
8744 | vcpu->arch.exception.pending = false; |
8745 | ||
3842d135 | 8746 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
01643c51 | 8747 | } |
3842d135 | 8748 | |
01643c51 KH |
8749 | int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs) |
8750 | { | |
8751 | vcpu_load(vcpu); | |
8752 | __set_regs(vcpu, regs); | |
875656fe | 8753 | vcpu_put(vcpu); |
b6c7a5dc HB |
8754 | return 0; |
8755 | } | |
8756 | ||
b6c7a5dc HB |
8757 | void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l) |
8758 | { | |
8759 | struct kvm_segment cs; | |
8760 | ||
3e6e0aab | 8761 | kvm_get_segment(vcpu, &cs, VCPU_SREG_CS); |
b6c7a5dc HB |
8762 | *db = cs.db; |
8763 | *l = cs.l; | |
8764 | } | |
8765 | EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits); | |
8766 | ||
01643c51 | 8767 | static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs) |
b6c7a5dc | 8768 | { |
89a27f4d | 8769 | struct desc_ptr dt; |
b6c7a5dc | 8770 | |
3e6e0aab GT |
8771 | kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS); |
8772 | kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS); | |
8773 | kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES); | |
8774 | kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS); | |
8775 | kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS); | |
8776 | kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS); | |
b6c7a5dc | 8777 | |
3e6e0aab GT |
8778 | kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR); |
8779 | kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR); | |
b6c7a5dc HB |
8780 | |
8781 | kvm_x86_ops->get_idt(vcpu, &dt); | |
89a27f4d GN |
8782 | sregs->idt.limit = dt.size; |
8783 | sregs->idt.base = dt.address; | |
b6c7a5dc | 8784 | kvm_x86_ops->get_gdt(vcpu, &dt); |
89a27f4d GN |
8785 | sregs->gdt.limit = dt.size; |
8786 | sregs->gdt.base = dt.address; | |
b6c7a5dc | 8787 | |
4d4ec087 | 8788 | sregs->cr0 = kvm_read_cr0(vcpu); |
ad312c7c | 8789 | sregs->cr2 = vcpu->arch.cr2; |
9f8fe504 | 8790 | sregs->cr3 = kvm_read_cr3(vcpu); |
fc78f519 | 8791 | sregs->cr4 = kvm_read_cr4(vcpu); |
2d3ad1f4 | 8792 | sregs->cr8 = kvm_get_cr8(vcpu); |
f6801dff | 8793 | sregs->efer = vcpu->arch.efer; |
b6c7a5dc HB |
8794 | sregs->apic_base = kvm_get_apic_base(vcpu); |
8795 | ||
0e96f31e | 8796 | memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap)); |
b6c7a5dc | 8797 | |
04140b41 | 8798 | if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft) |
14d0bc1f GN |
8799 | set_bit(vcpu->arch.interrupt.nr, |
8800 | (unsigned long *)sregs->interrupt_bitmap); | |
01643c51 | 8801 | } |
16d7a191 | 8802 | |
01643c51 KH |
8803 | int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu, |
8804 | struct kvm_sregs *sregs) | |
8805 | { | |
8806 | vcpu_load(vcpu); | |
8807 | __get_sregs(vcpu, sregs); | |
bcdec41c | 8808 | vcpu_put(vcpu); |
b6c7a5dc HB |
8809 | return 0; |
8810 | } | |
8811 | ||
62d9f0db MT |
8812 | int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, |
8813 | struct kvm_mp_state *mp_state) | |
8814 | { | |
fd232561 | 8815 | vcpu_load(vcpu); |
f958bd23 SC |
8816 | if (kvm_mpx_supported()) |
8817 | kvm_load_guest_fpu(vcpu); | |
fd232561 | 8818 | |
66450a21 | 8819 | kvm_apic_accept_events(vcpu); |
6aef266c SV |
8820 | if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED && |
8821 | vcpu->arch.pv.pv_unhalted) | |
8822 | mp_state->mp_state = KVM_MP_STATE_RUNNABLE; | |
8823 | else | |
8824 | mp_state->mp_state = vcpu->arch.mp_state; | |
8825 | ||
f958bd23 SC |
8826 | if (kvm_mpx_supported()) |
8827 | kvm_put_guest_fpu(vcpu); | |
fd232561 | 8828 | vcpu_put(vcpu); |
62d9f0db MT |
8829 | return 0; |
8830 | } | |
8831 | ||
8832 | int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, | |
8833 | struct kvm_mp_state *mp_state) | |
8834 | { | |
e83dff5e CD |
8835 | int ret = -EINVAL; |
8836 | ||
8837 | vcpu_load(vcpu); | |
8838 | ||
bce87cce | 8839 | if (!lapic_in_kernel(vcpu) && |
66450a21 | 8840 | mp_state->mp_state != KVM_MP_STATE_RUNNABLE) |
e83dff5e | 8841 | goto out; |
66450a21 | 8842 | |
27cbe7d6 LA |
8843 | /* |
8844 | * KVM_MP_STATE_INIT_RECEIVED means the processor is in | |
8845 | * INIT state; latched init should be reported using | |
8846 | * KVM_SET_VCPU_EVENTS, so reject it here. | |
8847 | */ | |
8848 | if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) && | |
28bf2888 DH |
8849 | (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED || |
8850 | mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED)) | |
e83dff5e | 8851 | goto out; |
28bf2888 | 8852 | |
66450a21 JK |
8853 | if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) { |
8854 | vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED; | |
8855 | set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events); | |
8856 | } else | |
8857 | vcpu->arch.mp_state = mp_state->mp_state; | |
3842d135 | 8858 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
e83dff5e CD |
8859 | |
8860 | ret = 0; | |
8861 | out: | |
8862 | vcpu_put(vcpu); | |
8863 | return ret; | |
62d9f0db MT |
8864 | } |
8865 | ||
7f3d35fd KW |
8866 | int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index, |
8867 | int reason, bool has_error_code, u32 error_code) | |
b6c7a5dc | 8868 | { |
9d74191a | 8869 | struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt; |
8ec4722d | 8870 | int ret; |
e01c2426 | 8871 | |
8ec4722d | 8872 | init_emulate_ctxt(vcpu); |
c697518a | 8873 | |
7f3d35fd | 8874 | ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason, |
9d74191a | 8875 | has_error_code, error_code); |
1051778f SC |
8876 | if (ret) { |
8877 | vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR; | |
8878 | vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION; | |
8879 | vcpu->run->internal.ndata = 0; | |
60fc3d02 | 8880 | return 0; |
1051778f | 8881 | } |
37817f29 | 8882 | |
9d74191a TY |
8883 | kvm_rip_write(vcpu, ctxt->eip); |
8884 | kvm_set_rflags(vcpu, ctxt->eflags); | |
3842d135 | 8885 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
60fc3d02 | 8886 | return 1; |
37817f29 IE |
8887 | } |
8888 | EXPORT_SYMBOL_GPL(kvm_task_switch); | |
8889 | ||
3140c156 | 8890 | static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs) |
f2981033 | 8891 | { |
37b95951 | 8892 | if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) { |
f2981033 LT |
8893 | /* |
8894 | * When EFER.LME and CR0.PG are set, the processor is in | |
8895 | * 64-bit mode (though maybe in a 32-bit code segment). | |
8896 | * CR4.PAE and EFER.LMA must be set. | |
8897 | */ | |
37b95951 | 8898 | if (!(sregs->cr4 & X86_CR4_PAE) |
f2981033 LT |
8899 | || !(sregs->efer & EFER_LMA)) |
8900 | return -EINVAL; | |
8901 | } else { | |
8902 | /* | |
8903 | * Not in 64-bit mode: EFER.LMA is clear and the code | |
8904 | * segment cannot be 64-bit. | |
8905 | */ | |
8906 | if (sregs->efer & EFER_LMA || sregs->cs.l) | |
8907 | return -EINVAL; | |
8908 | } | |
8909 | ||
3ca94192 | 8910 | return kvm_valid_cr4(vcpu, sregs->cr4); |
f2981033 LT |
8911 | } |
8912 | ||
01643c51 | 8913 | static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs) |
b6c7a5dc | 8914 | { |
58cb628d | 8915 | struct msr_data apic_base_msr; |
b6c7a5dc | 8916 | int mmu_reset_needed = 0; |
c4d21882 | 8917 | int cpuid_update_needed = 0; |
63f42e02 | 8918 | int pending_vec, max_bits, idx; |
89a27f4d | 8919 | struct desc_ptr dt; |
b4ef9d4e CD |
8920 | int ret = -EINVAL; |
8921 | ||
f2981033 | 8922 | if (kvm_valid_sregs(vcpu, sregs)) |
8dbfb2bf | 8923 | goto out; |
f2981033 | 8924 | |
d3802286 JM |
8925 | apic_base_msr.data = sregs->apic_base; |
8926 | apic_base_msr.host_initiated = true; | |
8927 | if (kvm_set_apic_base(vcpu, &apic_base_msr)) | |
b4ef9d4e | 8928 | goto out; |
6d1068b3 | 8929 | |
89a27f4d GN |
8930 | dt.size = sregs->idt.limit; |
8931 | dt.address = sregs->idt.base; | |
b6c7a5dc | 8932 | kvm_x86_ops->set_idt(vcpu, &dt); |
89a27f4d GN |
8933 | dt.size = sregs->gdt.limit; |
8934 | dt.address = sregs->gdt.base; | |
b6c7a5dc HB |
8935 | kvm_x86_ops->set_gdt(vcpu, &dt); |
8936 | ||
ad312c7c | 8937 | vcpu->arch.cr2 = sregs->cr2; |
9f8fe504 | 8938 | mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3; |
dc7e795e | 8939 | vcpu->arch.cr3 = sregs->cr3; |
cb3c1e2f | 8940 | kvm_register_mark_available(vcpu, VCPU_EXREG_CR3); |
b6c7a5dc | 8941 | |
2d3ad1f4 | 8942 | kvm_set_cr8(vcpu, sregs->cr8); |
b6c7a5dc | 8943 | |
f6801dff | 8944 | mmu_reset_needed |= vcpu->arch.efer != sregs->efer; |
b6c7a5dc | 8945 | kvm_x86_ops->set_efer(vcpu, sregs->efer); |
b6c7a5dc | 8946 | |
4d4ec087 | 8947 | mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0; |
b6c7a5dc | 8948 | kvm_x86_ops->set_cr0(vcpu, sregs->cr0); |
d7306163 | 8949 | vcpu->arch.cr0 = sregs->cr0; |
b6c7a5dc | 8950 | |
fc78f519 | 8951 | mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4; |
c4d21882 WH |
8952 | cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) & |
8953 | (X86_CR4_OSXSAVE | X86_CR4_PKE)); | |
b6c7a5dc | 8954 | kvm_x86_ops->set_cr4(vcpu, sregs->cr4); |
c4d21882 | 8955 | if (cpuid_update_needed) |
00b27a3e | 8956 | kvm_update_cpuid(vcpu); |
63f42e02 XG |
8957 | |
8958 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
bf03d4f9 | 8959 | if (is_pae_paging(vcpu)) { |
9f8fe504 | 8960 | load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)); |
7c93be44 MT |
8961 | mmu_reset_needed = 1; |
8962 | } | |
63f42e02 | 8963 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
b6c7a5dc HB |
8964 | |
8965 | if (mmu_reset_needed) | |
8966 | kvm_mmu_reset_context(vcpu); | |
8967 | ||
a50abc3b | 8968 | max_bits = KVM_NR_INTERRUPTS; |
923c61bb GN |
8969 | pending_vec = find_first_bit( |
8970 | (const unsigned long *)sregs->interrupt_bitmap, max_bits); | |
8971 | if (pending_vec < max_bits) { | |
66fd3f7f | 8972 | kvm_queue_interrupt(vcpu, pending_vec, false); |
923c61bb | 8973 | pr_debug("Set back pending irq %d\n", pending_vec); |
b6c7a5dc HB |
8974 | } |
8975 | ||
3e6e0aab GT |
8976 | kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS); |
8977 | kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS); | |
8978 | kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES); | |
8979 | kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS); | |
8980 | kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS); | |
8981 | kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS); | |
b6c7a5dc | 8982 | |
3e6e0aab GT |
8983 | kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR); |
8984 | kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR); | |
b6c7a5dc | 8985 | |
5f0269f5 ME |
8986 | update_cr8_intercept(vcpu); |
8987 | ||
9c3e4aab | 8988 | /* Older userspace won't unhalt the vcpu on reset. */ |
c5af89b6 | 8989 | if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 && |
9c3e4aab | 8990 | sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 && |
3eeb3288 | 8991 | !is_protmode(vcpu)) |
9c3e4aab MT |
8992 | vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; |
8993 | ||
3842d135 AK |
8994 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
8995 | ||
b4ef9d4e CD |
8996 | ret = 0; |
8997 | out: | |
01643c51 KH |
8998 | return ret; |
8999 | } | |
9000 | ||
9001 | int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu, | |
9002 | struct kvm_sregs *sregs) | |
9003 | { | |
9004 | int ret; | |
9005 | ||
9006 | vcpu_load(vcpu); | |
9007 | ret = __set_sregs(vcpu, sregs); | |
b4ef9d4e CD |
9008 | vcpu_put(vcpu); |
9009 | return ret; | |
b6c7a5dc HB |
9010 | } |
9011 | ||
d0bfb940 JK |
9012 | int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu, |
9013 | struct kvm_guest_debug *dbg) | |
b6c7a5dc | 9014 | { |
355be0b9 | 9015 | unsigned long rflags; |
ae675ef0 | 9016 | int i, r; |
b6c7a5dc | 9017 | |
66b56562 CD |
9018 | vcpu_load(vcpu); |
9019 | ||
4f926bf2 JK |
9020 | if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) { |
9021 | r = -EBUSY; | |
9022 | if (vcpu->arch.exception.pending) | |
2122ff5e | 9023 | goto out; |
4f926bf2 JK |
9024 | if (dbg->control & KVM_GUESTDBG_INJECT_DB) |
9025 | kvm_queue_exception(vcpu, DB_VECTOR); | |
9026 | else | |
9027 | kvm_queue_exception(vcpu, BP_VECTOR); | |
9028 | } | |
9029 | ||
91586a3b JK |
9030 | /* |
9031 | * Read rflags as long as potentially injected trace flags are still | |
9032 | * filtered out. | |
9033 | */ | |
9034 | rflags = kvm_get_rflags(vcpu); | |
355be0b9 JK |
9035 | |
9036 | vcpu->guest_debug = dbg->control; | |
9037 | if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE)) | |
9038 | vcpu->guest_debug = 0; | |
9039 | ||
9040 | if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) { | |
ae675ef0 JK |
9041 | for (i = 0; i < KVM_NR_DB_REGS; ++i) |
9042 | vcpu->arch.eff_db[i] = dbg->arch.debugreg[i]; | |
c8639010 | 9043 | vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7]; |
ae675ef0 JK |
9044 | } else { |
9045 | for (i = 0; i < KVM_NR_DB_REGS; i++) | |
9046 | vcpu->arch.eff_db[i] = vcpu->arch.db[i]; | |
ae675ef0 | 9047 | } |
c8639010 | 9048 | kvm_update_dr7(vcpu); |
ae675ef0 | 9049 | |
f92653ee JK |
9050 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) |
9051 | vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) + | |
9052 | get_segment_base(vcpu, VCPU_SREG_CS); | |
94fe45da | 9053 | |
91586a3b JK |
9054 | /* |
9055 | * Trigger an rflags update that will inject or remove the trace | |
9056 | * flags. | |
9057 | */ | |
9058 | kvm_set_rflags(vcpu, rflags); | |
b6c7a5dc | 9059 | |
a96036b8 | 9060 | kvm_x86_ops->update_bp_intercept(vcpu); |
b6c7a5dc | 9061 | |
4f926bf2 | 9062 | r = 0; |
d0bfb940 | 9063 | |
2122ff5e | 9064 | out: |
66b56562 | 9065 | vcpu_put(vcpu); |
b6c7a5dc HB |
9066 | return r; |
9067 | } | |
9068 | ||
8b006791 ZX |
9069 | /* |
9070 | * Translate a guest virtual address to a guest physical address. | |
9071 | */ | |
9072 | int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu, | |
9073 | struct kvm_translation *tr) | |
9074 | { | |
9075 | unsigned long vaddr = tr->linear_address; | |
9076 | gpa_t gpa; | |
f656ce01 | 9077 | int idx; |
8b006791 | 9078 | |
1da5b61d CD |
9079 | vcpu_load(vcpu); |
9080 | ||
f656ce01 | 9081 | idx = srcu_read_lock(&vcpu->kvm->srcu); |
1871c602 | 9082 | gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL); |
f656ce01 | 9083 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
8b006791 ZX |
9084 | tr->physical_address = gpa; |
9085 | tr->valid = gpa != UNMAPPED_GVA; | |
9086 | tr->writeable = 1; | |
9087 | tr->usermode = 0; | |
8b006791 | 9088 | |
1da5b61d | 9089 | vcpu_put(vcpu); |
8b006791 ZX |
9090 | return 0; |
9091 | } | |
9092 | ||
d0752060 HB |
9093 | int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) |
9094 | { | |
1393123e | 9095 | struct fxregs_state *fxsave; |
d0752060 | 9096 | |
1393123e | 9097 | vcpu_load(vcpu); |
d0752060 | 9098 | |
b666a4b6 | 9099 | fxsave = &vcpu->arch.guest_fpu->state.fxsave; |
d0752060 HB |
9100 | memcpy(fpu->fpr, fxsave->st_space, 128); |
9101 | fpu->fcw = fxsave->cwd; | |
9102 | fpu->fsw = fxsave->swd; | |
9103 | fpu->ftwx = fxsave->twd; | |
9104 | fpu->last_opcode = fxsave->fop; | |
9105 | fpu->last_ip = fxsave->rip; | |
9106 | fpu->last_dp = fxsave->rdp; | |
0e96f31e | 9107 | memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space)); |
d0752060 | 9108 | |
1393123e | 9109 | vcpu_put(vcpu); |
d0752060 HB |
9110 | return 0; |
9111 | } | |
9112 | ||
9113 | int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu) | |
9114 | { | |
6a96bc7f CD |
9115 | struct fxregs_state *fxsave; |
9116 | ||
9117 | vcpu_load(vcpu); | |
9118 | ||
b666a4b6 | 9119 | fxsave = &vcpu->arch.guest_fpu->state.fxsave; |
d0752060 | 9120 | |
d0752060 HB |
9121 | memcpy(fxsave->st_space, fpu->fpr, 128); |
9122 | fxsave->cwd = fpu->fcw; | |
9123 | fxsave->swd = fpu->fsw; | |
9124 | fxsave->twd = fpu->ftwx; | |
9125 | fxsave->fop = fpu->last_opcode; | |
9126 | fxsave->rip = fpu->last_ip; | |
9127 | fxsave->rdp = fpu->last_dp; | |
0e96f31e | 9128 | memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space)); |
d0752060 | 9129 | |
6a96bc7f | 9130 | vcpu_put(vcpu); |
d0752060 HB |
9131 | return 0; |
9132 | } | |
9133 | ||
01643c51 KH |
9134 | static void store_regs(struct kvm_vcpu *vcpu) |
9135 | { | |
9136 | BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES); | |
9137 | ||
9138 | if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS) | |
9139 | __get_regs(vcpu, &vcpu->run->s.regs.regs); | |
9140 | ||
9141 | if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS) | |
9142 | __get_sregs(vcpu, &vcpu->run->s.regs.sregs); | |
9143 | ||
9144 | if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS) | |
9145 | kvm_vcpu_ioctl_x86_get_vcpu_events( | |
9146 | vcpu, &vcpu->run->s.regs.events); | |
9147 | } | |
9148 | ||
9149 | static int sync_regs(struct kvm_vcpu *vcpu) | |
9150 | { | |
9151 | if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS) | |
9152 | return -EINVAL; | |
9153 | ||
9154 | if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) { | |
9155 | __set_regs(vcpu, &vcpu->run->s.regs.regs); | |
9156 | vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS; | |
9157 | } | |
9158 | if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) { | |
9159 | if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs)) | |
9160 | return -EINVAL; | |
9161 | vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS; | |
9162 | } | |
9163 | if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) { | |
9164 | if (kvm_vcpu_ioctl_x86_set_vcpu_events( | |
9165 | vcpu, &vcpu->run->s.regs.events)) | |
9166 | return -EINVAL; | |
9167 | vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS; | |
9168 | } | |
9169 | ||
9170 | return 0; | |
9171 | } | |
9172 | ||
0ee6a517 | 9173 | static void fx_init(struct kvm_vcpu *vcpu) |
d0752060 | 9174 | { |
b666a4b6 | 9175 | fpstate_init(&vcpu->arch.guest_fpu->state); |
782511b0 | 9176 | if (boot_cpu_has(X86_FEATURE_XSAVES)) |
b666a4b6 | 9177 | vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv = |
df1daba7 | 9178 | host_xcr0 | XSTATE_COMPACTION_ENABLED; |
d0752060 | 9179 | |
2acf923e DC |
9180 | /* |
9181 | * Ensure guest xcr0 is valid for loading | |
9182 | */ | |
d91cab78 | 9183 | vcpu->arch.xcr0 = XFEATURE_MASK_FP; |
2acf923e | 9184 | |
ad312c7c | 9185 | vcpu->arch.cr0 |= X86_CR0_ET; |
d0752060 | 9186 | } |
d0752060 | 9187 | |
897cc38e SC |
9188 | int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id) |
9189 | { | |
9190 | if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0) | |
9191 | pr_warn_once("kvm: SMP vm created on host with unstable TSC; " | |
9192 | "guest TSC will not be reliable\n"); | |
9193 | ||
9194 | return 0; | |
9195 | } | |
9196 | ||
e529ef66 | 9197 | int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu) |
e9b11c17 | 9198 | { |
95a0d01e SC |
9199 | struct page *page; |
9200 | int r; | |
5f73db11 | 9201 | |
95a0d01e SC |
9202 | vcpu->arch.emulate_ctxt.ops = &emulate_ops; |
9203 | if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu)) | |
9204 | vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; | |
9205 | else | |
9206 | vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED; | |
9207 | ||
9208 | kvm_set_tsc_khz(vcpu, max_tsc_khz); | |
9209 | ||
9210 | r = kvm_mmu_create(vcpu); | |
9211 | if (r < 0) | |
9212 | return r; | |
9213 | ||
9214 | if (irqchip_in_kernel(vcpu->kvm)) { | |
9215 | vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu->kvm); | |
9216 | r = kvm_create_lapic(vcpu, lapic_timer_advance_ns); | |
9217 | if (r < 0) | |
9218 | goto fail_mmu_destroy; | |
9219 | } else | |
9220 | static_key_slow_inc(&kvm_no_apic_vcpu); | |
9221 | ||
9222 | r = -ENOMEM; | |
9223 | ||
9224 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); | |
9225 | if (!page) | |
9226 | goto fail_free_lapic; | |
9227 | vcpu->arch.pio_data = page_address(page); | |
9228 | ||
9229 | vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4, | |
9230 | GFP_KERNEL_ACCOUNT); | |
9231 | if (!vcpu->arch.mce_banks) | |
9232 | goto fail_free_pio_data; | |
9233 | vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS; | |
9234 | ||
9235 | if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, | |
9236 | GFP_KERNEL_ACCOUNT)) | |
9237 | goto fail_free_mce_banks; | |
9238 | ||
9239 | vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache, | |
9240 | GFP_KERNEL_ACCOUNT); | |
9241 | if (!vcpu->arch.user_fpu) { | |
9242 | pr_err("kvm: failed to allocate userspace's fpu\n"); | |
9243 | goto free_wbinvd_dirty_mask; | |
9244 | } | |
9245 | ||
9246 | vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache, | |
9247 | GFP_KERNEL_ACCOUNT); | |
9248 | if (!vcpu->arch.guest_fpu) { | |
9249 | pr_err("kvm: failed to allocate vcpu's fpu\n"); | |
9250 | goto free_user_fpu; | |
9251 | } | |
9252 | fx_init(vcpu); | |
9253 | ||
9254 | vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET; | |
9255 | ||
9256 | vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu); | |
9257 | ||
9258 | vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT; | |
9259 | ||
9260 | kvm_async_pf_hash_reset(vcpu); | |
9261 | kvm_pmu_init(vcpu); | |
9262 | ||
9263 | vcpu->arch.pending_external_vector = -1; | |
9264 | vcpu->arch.preempted_in_kernel = false; | |
9265 | ||
9266 | kvm_hv_vcpu_init(vcpu); | |
9267 | ||
9268 | r = kvm_x86_ops->vcpu_create(vcpu); | |
9269 | if (r) | |
9270 | goto free_guest_fpu; | |
e9b11c17 | 9271 | |
0cf9135b | 9272 | vcpu->arch.arch_capabilities = kvm_get_arch_capabilities(); |
e53d88af | 9273 | vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT; |
19efffa2 | 9274 | kvm_vcpu_mtrr_init(vcpu); |
ec7660cc | 9275 | vcpu_load(vcpu); |
d28bc9dd | 9276 | kvm_vcpu_reset(vcpu, false); |
e1732991 | 9277 | kvm_init_mmu(vcpu, false); |
e9b11c17 | 9278 | vcpu_put(vcpu); |
ec7660cc | 9279 | return 0; |
95a0d01e SC |
9280 | |
9281 | free_guest_fpu: | |
9282 | kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu); | |
9283 | free_user_fpu: | |
9284 | kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu); | |
9285 | free_wbinvd_dirty_mask: | |
9286 | free_cpumask_var(vcpu->arch.wbinvd_dirty_mask); | |
9287 | fail_free_mce_banks: | |
9288 | kfree(vcpu->arch.mce_banks); | |
9289 | fail_free_pio_data: | |
9290 | free_page((unsigned long)vcpu->arch.pio_data); | |
9291 | fail_free_lapic: | |
9292 | kvm_free_lapic(vcpu); | |
9293 | fail_mmu_destroy: | |
9294 | kvm_mmu_destroy(vcpu); | |
9295 | return r; | |
e9b11c17 ZX |
9296 | } |
9297 | ||
31928aa5 | 9298 | void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) |
42897d86 | 9299 | { |
8fe8ab46 | 9300 | struct msr_data msr; |
332967a3 | 9301 | struct kvm *kvm = vcpu->kvm; |
42897d86 | 9302 | |
d3457c87 RK |
9303 | kvm_hv_vcpu_postcreate(vcpu); |
9304 | ||
ec7660cc | 9305 | if (mutex_lock_killable(&vcpu->mutex)) |
31928aa5 | 9306 | return; |
ec7660cc | 9307 | vcpu_load(vcpu); |
8fe8ab46 WA |
9308 | msr.data = 0x0; |
9309 | msr.index = MSR_IA32_TSC; | |
9310 | msr.host_initiated = true; | |
9311 | kvm_write_tsc(vcpu, &msr); | |
42897d86 | 9312 | vcpu_put(vcpu); |
2d5ba19b MT |
9313 | |
9314 | /* poll control enabled by default */ | |
9315 | vcpu->arch.msr_kvm_poll_control = 1; | |
9316 | ||
ec7660cc | 9317 | mutex_unlock(&vcpu->mutex); |
42897d86 | 9318 | |
630994b3 MT |
9319 | if (!kvmclock_periodic_sync) |
9320 | return; | |
9321 | ||
332967a3 AJ |
9322 | schedule_delayed_work(&kvm->arch.kvmclock_sync_work, |
9323 | KVMCLOCK_SYNC_PERIOD); | |
42897d86 MT |
9324 | } |
9325 | ||
d40ccc62 | 9326 | void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) |
e9b11c17 | 9327 | { |
95a0d01e SC |
9328 | int idx; |
9329 | ||
50b143e1 SC |
9330 | kvmclock_reset(vcpu); |
9331 | ||
9332 | kvm_x86_ops->vcpu_free(vcpu); | |
9333 | ||
50b143e1 SC |
9334 | free_cpumask_var(vcpu->arch.wbinvd_dirty_mask); |
9335 | kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu); | |
9336 | kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu); | |
95a0d01e SC |
9337 | |
9338 | kvm_hv_vcpu_uninit(vcpu); | |
9339 | kvm_pmu_destroy(vcpu); | |
9340 | kfree(vcpu->arch.mce_banks); | |
9341 | kvm_free_lapic(vcpu); | |
9342 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
9343 | kvm_mmu_destroy(vcpu); | |
9344 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
9345 | free_page((unsigned long)vcpu->arch.pio_data); | |
9346 | if (!lapic_in_kernel(vcpu)) | |
9347 | static_key_slow_dec(&kvm_no_apic_vcpu); | |
e9b11c17 ZX |
9348 | } |
9349 | ||
d28bc9dd | 9350 | void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event) |
e9b11c17 | 9351 | { |
b7e31be3 RK |
9352 | kvm_lapic_reset(vcpu, init_event); |
9353 | ||
e69fab5d PB |
9354 | vcpu->arch.hflags = 0; |
9355 | ||
c43203ca | 9356 | vcpu->arch.smi_pending = 0; |
52797bf9 | 9357 | vcpu->arch.smi_count = 0; |
7460fb4a AK |
9358 | atomic_set(&vcpu->arch.nmi_queued, 0); |
9359 | vcpu->arch.nmi_pending = 0; | |
448fa4a9 | 9360 | vcpu->arch.nmi_injected = false; |
5f7552d4 NA |
9361 | kvm_clear_interrupt_queue(vcpu); |
9362 | kvm_clear_exception_queue(vcpu); | |
448fa4a9 | 9363 | |
42dbaa5a | 9364 | memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db)); |
ae561ede | 9365 | kvm_update_dr0123(vcpu); |
6f43ed01 | 9366 | vcpu->arch.dr6 = DR6_INIT; |
73aaf249 | 9367 | kvm_update_dr6(vcpu); |
42dbaa5a | 9368 | vcpu->arch.dr7 = DR7_FIXED_1; |
c8639010 | 9369 | kvm_update_dr7(vcpu); |
42dbaa5a | 9370 | |
1119022c NA |
9371 | vcpu->arch.cr2 = 0; |
9372 | ||
3842d135 | 9373 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
344d9588 | 9374 | vcpu->arch.apf.msr_val = 0; |
c9aaa895 | 9375 | vcpu->arch.st.msr_val = 0; |
3842d135 | 9376 | |
12f9a48f GC |
9377 | kvmclock_reset(vcpu); |
9378 | ||
af585b92 GN |
9379 | kvm_clear_async_pf_completion_queue(vcpu); |
9380 | kvm_async_pf_hash_reset(vcpu); | |
9381 | vcpu->arch.apf.halted = false; | |
3842d135 | 9382 | |
a554d207 WL |
9383 | if (kvm_mpx_supported()) { |
9384 | void *mpx_state_buffer; | |
9385 | ||
9386 | /* | |
9387 | * To avoid have the INIT path from kvm_apic_has_events() that be | |
9388 | * called with loaded FPU and does not let userspace fix the state. | |
9389 | */ | |
f775b13e RR |
9390 | if (init_event) |
9391 | kvm_put_guest_fpu(vcpu); | |
b666a4b6 | 9392 | mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave, |
abd16d68 | 9393 | XFEATURE_BNDREGS); |
a554d207 WL |
9394 | if (mpx_state_buffer) |
9395 | memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state)); | |
b666a4b6 | 9396 | mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave, |
abd16d68 | 9397 | XFEATURE_BNDCSR); |
a554d207 WL |
9398 | if (mpx_state_buffer) |
9399 | memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr)); | |
f775b13e RR |
9400 | if (init_event) |
9401 | kvm_load_guest_fpu(vcpu); | |
a554d207 WL |
9402 | } |
9403 | ||
64d60670 | 9404 | if (!init_event) { |
d28bc9dd | 9405 | kvm_pmu_reset(vcpu); |
64d60670 | 9406 | vcpu->arch.smbase = 0x30000; |
db2336a8 | 9407 | |
db2336a8 | 9408 | vcpu->arch.msr_misc_features_enables = 0; |
a554d207 WL |
9409 | |
9410 | vcpu->arch.xcr0 = XFEATURE_MASK_FP; | |
64d60670 | 9411 | } |
f5132b01 | 9412 | |
66f7b72e JS |
9413 | memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs)); |
9414 | vcpu->arch.regs_avail = ~0; | |
9415 | vcpu->arch.regs_dirty = ~0; | |
9416 | ||
a554d207 WL |
9417 | vcpu->arch.ia32_xss = 0; |
9418 | ||
d28bc9dd | 9419 | kvm_x86_ops->vcpu_reset(vcpu, init_event); |
e9b11c17 ZX |
9420 | } |
9421 | ||
2b4a273b | 9422 | void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector) |
66450a21 JK |
9423 | { |
9424 | struct kvm_segment cs; | |
9425 | ||
9426 | kvm_get_segment(vcpu, &cs, VCPU_SREG_CS); | |
9427 | cs.selector = vector << 8; | |
9428 | cs.base = vector << 12; | |
9429 | kvm_set_segment(vcpu, &cs, VCPU_SREG_CS); | |
9430 | kvm_rip_write(vcpu, 0); | |
e9b11c17 ZX |
9431 | } |
9432 | ||
13a34e06 | 9433 | int kvm_arch_hardware_enable(void) |
e9b11c17 | 9434 | { |
ca84d1a2 ZA |
9435 | struct kvm *kvm; |
9436 | struct kvm_vcpu *vcpu; | |
9437 | int i; | |
0dd6a6ed ZA |
9438 | int ret; |
9439 | u64 local_tsc; | |
9440 | u64 max_tsc = 0; | |
9441 | bool stable, backwards_tsc = false; | |
18863bdd AK |
9442 | |
9443 | kvm_shared_msr_cpu_online(); | |
13a34e06 | 9444 | ret = kvm_x86_ops->hardware_enable(); |
0dd6a6ed ZA |
9445 | if (ret != 0) |
9446 | return ret; | |
9447 | ||
4ea1636b | 9448 | local_tsc = rdtsc(); |
b0c39dc6 | 9449 | stable = !kvm_check_tsc_unstable(); |
0dd6a6ed ZA |
9450 | list_for_each_entry(kvm, &vm_list, vm_list) { |
9451 | kvm_for_each_vcpu(i, vcpu, kvm) { | |
9452 | if (!stable && vcpu->cpu == smp_processor_id()) | |
105b21bb | 9453 | kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu); |
0dd6a6ed ZA |
9454 | if (stable && vcpu->arch.last_host_tsc > local_tsc) { |
9455 | backwards_tsc = true; | |
9456 | if (vcpu->arch.last_host_tsc > max_tsc) | |
9457 | max_tsc = vcpu->arch.last_host_tsc; | |
9458 | } | |
9459 | } | |
9460 | } | |
9461 | ||
9462 | /* | |
9463 | * Sometimes, even reliable TSCs go backwards. This happens on | |
9464 | * platforms that reset TSC during suspend or hibernate actions, but | |
9465 | * maintain synchronization. We must compensate. Fortunately, we can | |
9466 | * detect that condition here, which happens early in CPU bringup, | |
9467 | * before any KVM threads can be running. Unfortunately, we can't | |
9468 | * bring the TSCs fully up to date with real time, as we aren't yet far | |
9469 | * enough into CPU bringup that we know how much real time has actually | |
9285ec4c | 9470 | * elapsed; our helper function, ktime_get_boottime_ns() will be using boot |
0dd6a6ed ZA |
9471 | * variables that haven't been updated yet. |
9472 | * | |
9473 | * So we simply find the maximum observed TSC above, then record the | |
9474 | * adjustment to TSC in each VCPU. When the VCPU later gets loaded, | |
9475 | * the adjustment will be applied. Note that we accumulate | |
9476 | * adjustments, in case multiple suspend cycles happen before some VCPU | |
9477 | * gets a chance to run again. In the event that no KVM threads get a | |
9478 | * chance to run, we will miss the entire elapsed period, as we'll have | |
9479 | * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may | |
9480 | * loose cycle time. This isn't too big a deal, since the loss will be | |
9481 | * uniform across all VCPUs (not to mention the scenario is extremely | |
9482 | * unlikely). It is possible that a second hibernate recovery happens | |
9483 | * much faster than a first, causing the observed TSC here to be | |
9484 | * smaller; this would require additional padding adjustment, which is | |
9485 | * why we set last_host_tsc to the local tsc observed here. | |
9486 | * | |
9487 | * N.B. - this code below runs only on platforms with reliable TSC, | |
9488 | * as that is the only way backwards_tsc is set above. Also note | |
9489 | * that this runs for ALL vcpus, which is not a bug; all VCPUs should | |
9490 | * have the same delta_cyc adjustment applied if backwards_tsc | |
9491 | * is detected. Note further, this adjustment is only done once, | |
9492 | * as we reset last_host_tsc on all VCPUs to stop this from being | |
9493 | * called multiple times (one for each physical CPU bringup). | |
9494 | * | |
4a969980 | 9495 | * Platforms with unreliable TSCs don't have to deal with this, they |
0dd6a6ed ZA |
9496 | * will be compensated by the logic in vcpu_load, which sets the TSC to |
9497 | * catchup mode. This will catchup all VCPUs to real time, but cannot | |
9498 | * guarantee that they stay in perfect synchronization. | |
9499 | */ | |
9500 | if (backwards_tsc) { | |
9501 | u64 delta_cyc = max_tsc - local_tsc; | |
9502 | list_for_each_entry(kvm, &vm_list, vm_list) { | |
a826faf1 | 9503 | kvm->arch.backwards_tsc_observed = true; |
0dd6a6ed ZA |
9504 | kvm_for_each_vcpu(i, vcpu, kvm) { |
9505 | vcpu->arch.tsc_offset_adjustment += delta_cyc; | |
9506 | vcpu->arch.last_host_tsc = local_tsc; | |
105b21bb | 9507 | kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu); |
0dd6a6ed ZA |
9508 | } |
9509 | ||
9510 | /* | |
9511 | * We have to disable TSC offset matching.. if you were | |
9512 | * booting a VM while issuing an S4 host suspend.... | |
9513 | * you may have some problem. Solving this issue is | |
9514 | * left as an exercise to the reader. | |
9515 | */ | |
9516 | kvm->arch.last_tsc_nsec = 0; | |
9517 | kvm->arch.last_tsc_write = 0; | |
9518 | } | |
9519 | ||
9520 | } | |
9521 | return 0; | |
e9b11c17 ZX |
9522 | } |
9523 | ||
13a34e06 | 9524 | void kvm_arch_hardware_disable(void) |
e9b11c17 | 9525 | { |
13a34e06 RK |
9526 | kvm_x86_ops->hardware_disable(); |
9527 | drop_user_return_notifiers(); | |
e9b11c17 ZX |
9528 | } |
9529 | ||
9530 | int kvm_arch_hardware_setup(void) | |
9531 | { | |
9e9c3fe4 NA |
9532 | int r; |
9533 | ||
9534 | r = kvm_x86_ops->hardware_setup(); | |
9535 | if (r != 0) | |
9536 | return r; | |
9537 | ||
b11306b5 SC |
9538 | cr4_reserved_bits = kvm_host_cr4_reserved_bits(&boot_cpu_data); |
9539 | ||
35181e86 HZ |
9540 | if (kvm_has_tsc_control) { |
9541 | /* | |
9542 | * Make sure the user can only configure tsc_khz values that | |
9543 | * fit into a signed integer. | |
273ba457 | 9544 | * A min value is not calculated because it will always |
35181e86 HZ |
9545 | * be 1 on all machines. |
9546 | */ | |
9547 | u64 max = min(0x7fffffffULL, | |
9548 | __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz)); | |
9549 | kvm_max_guest_tsc_khz = max; | |
9550 | ||
ad721883 | 9551 | kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits; |
35181e86 | 9552 | } |
ad721883 | 9553 | |
139a12cf AL |
9554 | if (boot_cpu_has(X86_FEATURE_XSAVES)) |
9555 | rdmsrl(MSR_IA32_XSS, host_xss); | |
9556 | ||
9e9c3fe4 NA |
9557 | kvm_init_msr_list(); |
9558 | return 0; | |
e9b11c17 ZX |
9559 | } |
9560 | ||
9561 | void kvm_arch_hardware_unsetup(void) | |
9562 | { | |
9563 | kvm_x86_ops->hardware_unsetup(); | |
9564 | } | |
9565 | ||
f257d6dc | 9566 | int kvm_arch_check_processor_compat(void) |
e9b11c17 | 9567 | { |
f1cdecf5 SC |
9568 | struct cpuinfo_x86 *c = &cpu_data(smp_processor_id()); |
9569 | ||
9570 | WARN_ON(!irqs_disabled()); | |
9571 | ||
9572 | if (kvm_host_cr4_reserved_bits(c) != cr4_reserved_bits) | |
9573 | return -EIO; | |
9574 | ||
f257d6dc | 9575 | return kvm_x86_ops->check_processor_compatibility(); |
d71ba788 PB |
9576 | } |
9577 | ||
9578 | bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu) | |
9579 | { | |
9580 | return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id; | |
9581 | } | |
9582 | EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp); | |
9583 | ||
9584 | bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu) | |
9585 | { | |
9586 | return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0; | |
e9b11c17 ZX |
9587 | } |
9588 | ||
54e9818f | 9589 | struct static_key kvm_no_apic_vcpu __read_mostly; |
bce87cce | 9590 | EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu); |
54e9818f | 9591 | |
e790d9ef RK |
9592 | void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu) |
9593 | { | |
b35e5548 LX |
9594 | struct kvm_pmu *pmu = vcpu_to_pmu(vcpu); |
9595 | ||
c595ceee | 9596 | vcpu->arch.l1tf_flush_l1d = true; |
b35e5548 LX |
9597 | if (pmu->version && unlikely(pmu->event_count)) { |
9598 | pmu->need_cleanup = true; | |
9599 | kvm_make_request(KVM_REQ_PMU, vcpu); | |
9600 | } | |
ae97a3b8 | 9601 | kvm_x86_ops->sched_in(vcpu, cpu); |
e790d9ef RK |
9602 | } |
9603 | ||
e08b9637 | 9604 | int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) |
d19a9cd2 | 9605 | { |
e08b9637 CO |
9606 | if (type) |
9607 | return -EINVAL; | |
9608 | ||
6ef768fa | 9609 | INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list); |
f05e70ac | 9610 | INIT_LIST_HEAD(&kvm->arch.active_mmu_pages); |
10605204 | 9611 | INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages); |
1aa9b957 | 9612 | INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages); |
4d5c5d0f | 9613 | INIT_LIST_HEAD(&kvm->arch.assigned_dev_head); |
e0f0bbc5 | 9614 | atomic_set(&kvm->arch.noncoherent_dma_count, 0); |
d19a9cd2 | 9615 | |
5550af4d SY |
9616 | /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */ |
9617 | set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap); | |
7a84428a AW |
9618 | /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */ |
9619 | set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID, | |
9620 | &kvm->arch.irq_sources_bitmap); | |
5550af4d | 9621 | |
038f8c11 | 9622 | raw_spin_lock_init(&kvm->arch.tsc_write_lock); |
1e08ec4a | 9623 | mutex_init(&kvm->arch.apic_map_lock); |
d828199e MT |
9624 | spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock); |
9625 | ||
9285ec4c | 9626 | kvm->arch.kvmclock_offset = -ktime_get_boottime_ns(); |
d828199e | 9627 | pvclock_update_vm_gtod_copy(kvm); |
53f658b3 | 9628 | |
6fbbde9a DS |
9629 | kvm->arch.guest_can_read_msr_platform_info = true; |
9630 | ||
7e44e449 | 9631 | INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn); |
332967a3 | 9632 | INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn); |
7e44e449 | 9633 | |
cbc0236a | 9634 | kvm_hv_init_vm(kvm); |
0eb05bf2 | 9635 | kvm_page_track_init(kvm); |
13d268ca | 9636 | kvm_mmu_init_vm(kvm); |
0eb05bf2 | 9637 | |
92735b1b | 9638 | return kvm_x86_ops->vm_init(kvm); |
d19a9cd2 ZX |
9639 | } |
9640 | ||
1aa9b957 JS |
9641 | int kvm_arch_post_init_vm(struct kvm *kvm) |
9642 | { | |
9643 | return kvm_mmu_post_init_vm(kvm); | |
9644 | } | |
9645 | ||
d19a9cd2 ZX |
9646 | static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu) |
9647 | { | |
ec7660cc | 9648 | vcpu_load(vcpu); |
d19a9cd2 ZX |
9649 | kvm_mmu_unload(vcpu); |
9650 | vcpu_put(vcpu); | |
9651 | } | |
9652 | ||
9653 | static void kvm_free_vcpus(struct kvm *kvm) | |
9654 | { | |
9655 | unsigned int i; | |
988a2cae | 9656 | struct kvm_vcpu *vcpu; |
d19a9cd2 ZX |
9657 | |
9658 | /* | |
9659 | * Unpin any mmu pages first. | |
9660 | */ | |
af585b92 GN |
9661 | kvm_for_each_vcpu(i, vcpu, kvm) { |
9662 | kvm_clear_async_pf_completion_queue(vcpu); | |
988a2cae | 9663 | kvm_unload_vcpu_mmu(vcpu); |
af585b92 | 9664 | } |
988a2cae | 9665 | kvm_for_each_vcpu(i, vcpu, kvm) |
4543bdc0 | 9666 | kvm_vcpu_destroy(vcpu); |
988a2cae GN |
9667 | |
9668 | mutex_lock(&kvm->lock); | |
9669 | for (i = 0; i < atomic_read(&kvm->online_vcpus); i++) | |
9670 | kvm->vcpus[i] = NULL; | |
d19a9cd2 | 9671 | |
988a2cae GN |
9672 | atomic_set(&kvm->online_vcpus, 0); |
9673 | mutex_unlock(&kvm->lock); | |
d19a9cd2 ZX |
9674 | } |
9675 | ||
ad8ba2cd SY |
9676 | void kvm_arch_sync_events(struct kvm *kvm) |
9677 | { | |
332967a3 | 9678 | cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work); |
7e44e449 | 9679 | cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work); |
aea924f6 | 9680 | kvm_free_pit(kvm); |
ad8ba2cd SY |
9681 | } |
9682 | ||
1d8007bd | 9683 | int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size) |
9da0e4d5 PB |
9684 | { |
9685 | int i, r; | |
25188b99 | 9686 | unsigned long hva; |
f0d648bd PB |
9687 | struct kvm_memslots *slots = kvm_memslots(kvm); |
9688 | struct kvm_memory_slot *slot, old; | |
9da0e4d5 PB |
9689 | |
9690 | /* Called with kvm->slots_lock held. */ | |
1d8007bd PB |
9691 | if (WARN_ON(id >= KVM_MEM_SLOTS_NUM)) |
9692 | return -EINVAL; | |
9da0e4d5 | 9693 | |
f0d648bd PB |
9694 | slot = id_to_memslot(slots, id); |
9695 | if (size) { | |
b21629da | 9696 | if (slot->npages) |
f0d648bd PB |
9697 | return -EEXIST; |
9698 | ||
9699 | /* | |
9700 | * MAP_SHARED to prevent internal slot pages from being moved | |
9701 | * by fork()/COW. | |
9702 | */ | |
9703 | hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE, | |
9704 | MAP_SHARED | MAP_ANONYMOUS, 0); | |
9705 | if (IS_ERR((void *)hva)) | |
9706 | return PTR_ERR((void *)hva); | |
9707 | } else { | |
9708 | if (!slot->npages) | |
9709 | return 0; | |
9710 | ||
9711 | hva = 0; | |
9712 | } | |
9713 | ||
9714 | old = *slot; | |
9da0e4d5 | 9715 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
1d8007bd | 9716 | struct kvm_userspace_memory_region m; |
9da0e4d5 | 9717 | |
1d8007bd PB |
9718 | m.slot = id | (i << 16); |
9719 | m.flags = 0; | |
9720 | m.guest_phys_addr = gpa; | |
f0d648bd | 9721 | m.userspace_addr = hva; |
1d8007bd | 9722 | m.memory_size = size; |
9da0e4d5 PB |
9723 | r = __kvm_set_memory_region(kvm, &m); |
9724 | if (r < 0) | |
9725 | return r; | |
9726 | } | |
9727 | ||
103c763c EB |
9728 | if (!size) |
9729 | vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE); | |
f0d648bd | 9730 | |
9da0e4d5 PB |
9731 | return 0; |
9732 | } | |
9733 | EXPORT_SYMBOL_GPL(__x86_set_memory_region); | |
9734 | ||
1d8007bd | 9735 | int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size) |
9da0e4d5 PB |
9736 | { |
9737 | int r; | |
9738 | ||
9739 | mutex_lock(&kvm->slots_lock); | |
1d8007bd | 9740 | r = __x86_set_memory_region(kvm, id, gpa, size); |
9da0e4d5 PB |
9741 | mutex_unlock(&kvm->slots_lock); |
9742 | ||
9743 | return r; | |
9744 | } | |
9745 | EXPORT_SYMBOL_GPL(x86_set_memory_region); | |
9746 | ||
1aa9b957 JS |
9747 | void kvm_arch_pre_destroy_vm(struct kvm *kvm) |
9748 | { | |
9749 | kvm_mmu_pre_destroy_vm(kvm); | |
9750 | } | |
9751 | ||
d19a9cd2 ZX |
9752 | void kvm_arch_destroy_vm(struct kvm *kvm) |
9753 | { | |
27469d29 AH |
9754 | if (current->mm == kvm->mm) { |
9755 | /* | |
9756 | * Free memory regions allocated on behalf of userspace, | |
9757 | * unless the the memory map has changed due to process exit | |
9758 | * or fd copying. | |
9759 | */ | |
1d8007bd PB |
9760 | x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0); |
9761 | x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0); | |
9762 | x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0); | |
27469d29 | 9763 | } |
03543133 SS |
9764 | if (kvm_x86_ops->vm_destroy) |
9765 | kvm_x86_ops->vm_destroy(kvm); | |
c761159c PX |
9766 | kvm_pic_destroy(kvm); |
9767 | kvm_ioapic_destroy(kvm); | |
d19a9cd2 | 9768 | kvm_free_vcpus(kvm); |
af1bae54 | 9769 | kvfree(rcu_dereference_check(kvm->arch.apic_map, 1)); |
66bb8a06 | 9770 | kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1)); |
13d268ca | 9771 | kvm_mmu_uninit_vm(kvm); |
2beb6dad | 9772 | kvm_page_track_cleanup(kvm); |
cbc0236a | 9773 | kvm_hv_destroy_vm(kvm); |
d19a9cd2 | 9774 | } |
0de10343 | 9775 | |
5587027c | 9776 | void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free, |
db3fe4eb TY |
9777 | struct kvm_memory_slot *dont) |
9778 | { | |
9779 | int i; | |
9780 | ||
d89cc617 TY |
9781 | for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) { |
9782 | if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) { | |
548ef284 | 9783 | kvfree(free->arch.rmap[i]); |
d89cc617 | 9784 | free->arch.rmap[i] = NULL; |
77d11309 | 9785 | } |
d89cc617 TY |
9786 | if (i == 0) |
9787 | continue; | |
9788 | ||
9789 | if (!dont || free->arch.lpage_info[i - 1] != | |
9790 | dont->arch.lpage_info[i - 1]) { | |
548ef284 | 9791 | kvfree(free->arch.lpage_info[i - 1]); |
d89cc617 | 9792 | free->arch.lpage_info[i - 1] = NULL; |
db3fe4eb TY |
9793 | } |
9794 | } | |
21ebbeda XG |
9795 | |
9796 | kvm_page_track_free_memslot(free, dont); | |
db3fe4eb TY |
9797 | } |
9798 | ||
5587027c AK |
9799 | int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot, |
9800 | unsigned long npages) | |
db3fe4eb TY |
9801 | { |
9802 | int i; | |
9803 | ||
d89cc617 | 9804 | for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) { |
92f94f1e | 9805 | struct kvm_lpage_info *linfo; |
db3fe4eb TY |
9806 | unsigned long ugfn; |
9807 | int lpages; | |
d89cc617 | 9808 | int level = i + 1; |
db3fe4eb TY |
9809 | |
9810 | lpages = gfn_to_index(slot->base_gfn + npages - 1, | |
9811 | slot->base_gfn, level) + 1; | |
9812 | ||
d89cc617 | 9813 | slot->arch.rmap[i] = |
778e1cdd | 9814 | kvcalloc(lpages, sizeof(*slot->arch.rmap[i]), |
254272ce | 9815 | GFP_KERNEL_ACCOUNT); |
d89cc617 | 9816 | if (!slot->arch.rmap[i]) |
77d11309 | 9817 | goto out_free; |
d89cc617 TY |
9818 | if (i == 0) |
9819 | continue; | |
77d11309 | 9820 | |
254272ce | 9821 | linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT); |
92f94f1e | 9822 | if (!linfo) |
db3fe4eb TY |
9823 | goto out_free; |
9824 | ||
92f94f1e XG |
9825 | slot->arch.lpage_info[i - 1] = linfo; |
9826 | ||
db3fe4eb | 9827 | if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1)) |
92f94f1e | 9828 | linfo[0].disallow_lpage = 1; |
db3fe4eb | 9829 | if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1)) |
92f94f1e | 9830 | linfo[lpages - 1].disallow_lpage = 1; |
db3fe4eb TY |
9831 | ugfn = slot->userspace_addr >> PAGE_SHIFT; |
9832 | /* | |
9833 | * If the gfn and userspace address are not aligned wrt each | |
9834 | * other, or if explicitly asked to, disable large page | |
9835 | * support for this slot | |
9836 | */ | |
9837 | if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) || | |
9838 | !kvm_largepages_enabled()) { | |
9839 | unsigned long j; | |
9840 | ||
9841 | for (j = 0; j < lpages; ++j) | |
92f94f1e | 9842 | linfo[j].disallow_lpage = 1; |
db3fe4eb TY |
9843 | } |
9844 | } | |
9845 | ||
21ebbeda XG |
9846 | if (kvm_page_track_create_memslot(slot, npages)) |
9847 | goto out_free; | |
9848 | ||
db3fe4eb TY |
9849 | return 0; |
9850 | ||
9851 | out_free: | |
d89cc617 | 9852 | for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) { |
548ef284 | 9853 | kvfree(slot->arch.rmap[i]); |
d89cc617 TY |
9854 | slot->arch.rmap[i] = NULL; |
9855 | if (i == 0) | |
9856 | continue; | |
9857 | ||
548ef284 | 9858 | kvfree(slot->arch.lpage_info[i - 1]); |
d89cc617 | 9859 | slot->arch.lpage_info[i - 1] = NULL; |
db3fe4eb TY |
9860 | } |
9861 | return -ENOMEM; | |
9862 | } | |
9863 | ||
15248258 | 9864 | void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen) |
e59dbe09 | 9865 | { |
e6dff7d1 TY |
9866 | /* |
9867 | * memslots->generation has been incremented. | |
9868 | * mmio generation may have reached its maximum value. | |
9869 | */ | |
15248258 | 9870 | kvm_mmu_invalidate_mmio_sptes(kvm, gen); |
e59dbe09 TY |
9871 | } |
9872 | ||
f7784b8e MT |
9873 | int kvm_arch_prepare_memory_region(struct kvm *kvm, |
9874 | struct kvm_memory_slot *memslot, | |
09170a49 | 9875 | const struct kvm_userspace_memory_region *mem, |
7b6195a9 | 9876 | enum kvm_mr_change change) |
0de10343 | 9877 | { |
f7784b8e MT |
9878 | return 0; |
9879 | } | |
9880 | ||
88178fd4 KH |
9881 | static void kvm_mmu_slot_apply_flags(struct kvm *kvm, |
9882 | struct kvm_memory_slot *new) | |
9883 | { | |
9884 | /* Still write protect RO slot */ | |
9885 | if (new->flags & KVM_MEM_READONLY) { | |
9886 | kvm_mmu_slot_remove_write_access(kvm, new); | |
9887 | return; | |
9888 | } | |
9889 | ||
9890 | /* | |
9891 | * Call kvm_x86_ops dirty logging hooks when they are valid. | |
9892 | * | |
9893 | * kvm_x86_ops->slot_disable_log_dirty is called when: | |
9894 | * | |
9895 | * - KVM_MR_CREATE with dirty logging is disabled | |
9896 | * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag | |
9897 | * | |
9898 | * The reason is, in case of PML, we need to set D-bit for any slots | |
9899 | * with dirty logging disabled in order to eliminate unnecessary GPA | |
0a03cbda | 9900 | * logging in PML buffer (and potential PML buffer full VMEXIT). This |
88178fd4 | 9901 | * guarantees leaving PML enabled during guest's lifetime won't have |
bdd303cb | 9902 | * any additional overhead from PML when guest is running with dirty |
88178fd4 KH |
9903 | * logging disabled for memory slots. |
9904 | * | |
9905 | * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot | |
9906 | * to dirty logging mode. | |
9907 | * | |
9908 | * If kvm_x86_ops dirty logging hooks are invalid, use write protect. | |
9909 | * | |
9910 | * In case of write protect: | |
9911 | * | |
9912 | * Write protect all pages for dirty logging. | |
9913 | * | |
9914 | * All the sptes including the large sptes which point to this | |
9915 | * slot are set to readonly. We can not create any new large | |
9916 | * spte on this slot until the end of the logging. | |
9917 | * | |
9918 | * See the comments in fast_page_fault(). | |
9919 | */ | |
9920 | if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) { | |
9921 | if (kvm_x86_ops->slot_enable_log_dirty) | |
9922 | kvm_x86_ops->slot_enable_log_dirty(kvm, new); | |
9923 | else | |
9924 | kvm_mmu_slot_remove_write_access(kvm, new); | |
9925 | } else { | |
9926 | if (kvm_x86_ops->slot_disable_log_dirty) | |
9927 | kvm_x86_ops->slot_disable_log_dirty(kvm, new); | |
9928 | } | |
9929 | } | |
9930 | ||
f7784b8e | 9931 | void kvm_arch_commit_memory_region(struct kvm *kvm, |
09170a49 | 9932 | const struct kvm_userspace_memory_region *mem, |
8482644a | 9933 | const struct kvm_memory_slot *old, |
f36f3f28 | 9934 | const struct kvm_memory_slot *new, |
8482644a | 9935 | enum kvm_mr_change change) |
f7784b8e | 9936 | { |
48c0e4e9 | 9937 | if (!kvm->arch.n_requested_mmu_pages) |
4d66623c WY |
9938 | kvm_mmu_change_mmu_pages(kvm, |
9939 | kvm_mmu_calculate_default_mmu_pages(kvm)); | |
1c91cad4 | 9940 | |
3ea3b7fa WL |
9941 | /* |
9942 | * Dirty logging tracks sptes in 4k granularity, meaning that large | |
9943 | * sptes have to be split. If live migration is successful, the guest | |
9944 | * in the source machine will be destroyed and large sptes will be | |
9945 | * created in the destination. However, if the guest continues to run | |
9946 | * in the source machine (for example if live migration fails), small | |
9947 | * sptes will remain around and cause bad performance. | |
9948 | * | |
9949 | * Scan sptes if dirty logging has been stopped, dropping those | |
9950 | * which can be collapsed into a single large-page spte. Later | |
9951 | * page faults will create the large-page sptes. | |
319109a2 SC |
9952 | * |
9953 | * There is no need to do this in any of the following cases: | |
9954 | * CREATE: No dirty mappings will already exist. | |
9955 | * MOVE/DELETE: The old mappings will already have been cleaned up by | |
9956 | * kvm_arch_flush_shadow_memslot() | |
3ea3b7fa | 9957 | */ |
319109a2 | 9958 | if (change == KVM_MR_FLAGS_ONLY && |
3ea3b7fa WL |
9959 | (old->flags & KVM_MEM_LOG_DIRTY_PAGES) && |
9960 | !(new->flags & KVM_MEM_LOG_DIRTY_PAGES)) | |
9961 | kvm_mmu_zap_collapsible_sptes(kvm, new); | |
9962 | ||
c972f3b1 | 9963 | /* |
88178fd4 | 9964 | * Set up write protection and/or dirty logging for the new slot. |
c126d94f | 9965 | * |
88178fd4 KH |
9966 | * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have |
9967 | * been zapped so no dirty logging staff is needed for old slot. For | |
9968 | * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the | |
9969 | * new and it's also covered when dealing with the new slot. | |
f36f3f28 PB |
9970 | * |
9971 | * FIXME: const-ify all uses of struct kvm_memory_slot. | |
c972f3b1 | 9972 | */ |
88178fd4 | 9973 | if (change != KVM_MR_DELETE) |
f36f3f28 | 9974 | kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new); |
0de10343 | 9975 | } |
1d737c8a | 9976 | |
2df72e9b | 9977 | void kvm_arch_flush_shadow_all(struct kvm *kvm) |
34d4cb8f | 9978 | { |
7390de1e | 9979 | kvm_mmu_zap_all(kvm); |
34d4cb8f MT |
9980 | } |
9981 | ||
2df72e9b MT |
9982 | void kvm_arch_flush_shadow_memslot(struct kvm *kvm, |
9983 | struct kvm_memory_slot *slot) | |
9984 | { | |
ae7cd873 | 9985 | kvm_page_track_flush_slot(kvm, slot); |
2df72e9b MT |
9986 | } |
9987 | ||
e6c67d8c LA |
9988 | static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu) |
9989 | { | |
9990 | return (is_guest_mode(vcpu) && | |
9991 | kvm_x86_ops->guest_apic_has_interrupt && | |
9992 | kvm_x86_ops->guest_apic_has_interrupt(vcpu)); | |
9993 | } | |
9994 | ||
5d9bc648 PB |
9995 | static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu) |
9996 | { | |
9997 | if (!list_empty_careful(&vcpu->async_pf.done)) | |
9998 | return true; | |
9999 | ||
10000 | if (kvm_apic_has_events(vcpu)) | |
10001 | return true; | |
10002 | ||
10003 | if (vcpu->arch.pv.pv_unhalted) | |
10004 | return true; | |
10005 | ||
a5f01f8e WL |
10006 | if (vcpu->arch.exception.pending) |
10007 | return true; | |
10008 | ||
47a66eed Z |
10009 | if (kvm_test_request(KVM_REQ_NMI, vcpu) || |
10010 | (vcpu->arch.nmi_pending && | |
10011 | kvm_x86_ops->nmi_allowed(vcpu))) | |
5d9bc648 PB |
10012 | return true; |
10013 | ||
47a66eed Z |
10014 | if (kvm_test_request(KVM_REQ_SMI, vcpu) || |
10015 | (vcpu->arch.smi_pending && !is_smm(vcpu))) | |
73917739 PB |
10016 | return true; |
10017 | ||
5d9bc648 | 10018 | if (kvm_arch_interrupt_allowed(vcpu) && |
e6c67d8c LA |
10019 | (kvm_cpu_has_interrupt(vcpu) || |
10020 | kvm_guest_apic_has_interrupt(vcpu))) | |
5d9bc648 PB |
10021 | return true; |
10022 | ||
1f4b34f8 AS |
10023 | if (kvm_hv_has_stimer_pending(vcpu)) |
10024 | return true; | |
10025 | ||
5d9bc648 PB |
10026 | return false; |
10027 | } | |
10028 | ||
1d737c8a ZX |
10029 | int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu) |
10030 | { | |
5d9bc648 | 10031 | return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu); |
1d737c8a | 10032 | } |
5736199a | 10033 | |
17e433b5 WL |
10034 | bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu) |
10035 | { | |
10036 | if (READ_ONCE(vcpu->arch.pv.pv_unhalted)) | |
10037 | return true; | |
10038 | ||
10039 | if (kvm_test_request(KVM_REQ_NMI, vcpu) || | |
10040 | kvm_test_request(KVM_REQ_SMI, vcpu) || | |
10041 | kvm_test_request(KVM_REQ_EVENT, vcpu)) | |
10042 | return true; | |
10043 | ||
10044 | if (vcpu->arch.apicv_active && kvm_x86_ops->dy_apicv_has_pending_interrupt(vcpu)) | |
10045 | return true; | |
10046 | ||
10047 | return false; | |
10048 | } | |
10049 | ||
199b5763 LM |
10050 | bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu) |
10051 | { | |
de63ad4c | 10052 | return vcpu->arch.preempted_in_kernel; |
199b5763 LM |
10053 | } |
10054 | ||
b6d33834 | 10055 | int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) |
5736199a | 10056 | { |
b6d33834 | 10057 | return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; |
5736199a | 10058 | } |
78646121 GN |
10059 | |
10060 | int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu) | |
10061 | { | |
10062 | return kvm_x86_ops->interrupt_allowed(vcpu); | |
10063 | } | |
229456fc | 10064 | |
82b32774 | 10065 | unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu) |
f92653ee | 10066 | { |
82b32774 NA |
10067 | if (is_64_bit_mode(vcpu)) |
10068 | return kvm_rip_read(vcpu); | |
10069 | return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) + | |
10070 | kvm_rip_read(vcpu)); | |
10071 | } | |
10072 | EXPORT_SYMBOL_GPL(kvm_get_linear_rip); | |
f92653ee | 10073 | |
82b32774 NA |
10074 | bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip) |
10075 | { | |
10076 | return kvm_get_linear_rip(vcpu) == linear_rip; | |
f92653ee JK |
10077 | } |
10078 | EXPORT_SYMBOL_GPL(kvm_is_linear_rip); | |
10079 | ||
94fe45da JK |
10080 | unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu) |
10081 | { | |
10082 | unsigned long rflags; | |
10083 | ||
10084 | rflags = kvm_x86_ops->get_rflags(vcpu); | |
10085 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) | |
c310bac5 | 10086 | rflags &= ~X86_EFLAGS_TF; |
94fe45da JK |
10087 | return rflags; |
10088 | } | |
10089 | EXPORT_SYMBOL_GPL(kvm_get_rflags); | |
10090 | ||
6addfc42 | 10091 | static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags) |
94fe45da JK |
10092 | { |
10093 | if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP && | |
f92653ee | 10094 | kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip)) |
c310bac5 | 10095 | rflags |= X86_EFLAGS_TF; |
94fe45da | 10096 | kvm_x86_ops->set_rflags(vcpu, rflags); |
6addfc42 PB |
10097 | } |
10098 | ||
10099 | void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags) | |
10100 | { | |
10101 | __kvm_set_rflags(vcpu, rflags); | |
3842d135 | 10102 | kvm_make_request(KVM_REQ_EVENT, vcpu); |
94fe45da JK |
10103 | } |
10104 | EXPORT_SYMBOL_GPL(kvm_set_rflags); | |
10105 | ||
56028d08 GN |
10106 | void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work) |
10107 | { | |
10108 | int r; | |
10109 | ||
44dd3ffa | 10110 | if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) || |
f2e10669 | 10111 | work->wakeup_all) |
56028d08 GN |
10112 | return; |
10113 | ||
10114 | r = kvm_mmu_reload(vcpu); | |
10115 | if (unlikely(r)) | |
10116 | return; | |
10117 | ||
44dd3ffa VK |
10118 | if (!vcpu->arch.mmu->direct_map && |
10119 | work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu)) | |
fb67e14f XG |
10120 | return; |
10121 | ||
736c291c | 10122 | vcpu->arch.mmu->page_fault(vcpu, work->cr2_or_gpa, 0, true); |
56028d08 GN |
10123 | } |
10124 | ||
af585b92 GN |
10125 | static inline u32 kvm_async_pf_hash_fn(gfn_t gfn) |
10126 | { | |
10127 | return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU)); | |
10128 | } | |
10129 | ||
10130 | static inline u32 kvm_async_pf_next_probe(u32 key) | |
10131 | { | |
10132 | return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1); | |
10133 | } | |
10134 | ||
10135 | static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) | |
10136 | { | |
10137 | u32 key = kvm_async_pf_hash_fn(gfn); | |
10138 | ||
10139 | while (vcpu->arch.apf.gfns[key] != ~0) | |
10140 | key = kvm_async_pf_next_probe(key); | |
10141 | ||
10142 | vcpu->arch.apf.gfns[key] = gfn; | |
10143 | } | |
10144 | ||
10145 | static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn) | |
10146 | { | |
10147 | int i; | |
10148 | u32 key = kvm_async_pf_hash_fn(gfn); | |
10149 | ||
10150 | for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) && | |
c7d28c24 XG |
10151 | (vcpu->arch.apf.gfns[key] != gfn && |
10152 | vcpu->arch.apf.gfns[key] != ~0); i++) | |
af585b92 GN |
10153 | key = kvm_async_pf_next_probe(key); |
10154 | ||
10155 | return key; | |
10156 | } | |
10157 | ||
10158 | bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) | |
10159 | { | |
10160 | return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn; | |
10161 | } | |
10162 | ||
10163 | static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn) | |
10164 | { | |
10165 | u32 i, j, k; | |
10166 | ||
10167 | i = j = kvm_async_pf_gfn_slot(vcpu, gfn); | |
10168 | while (true) { | |
10169 | vcpu->arch.apf.gfns[i] = ~0; | |
10170 | do { | |
10171 | j = kvm_async_pf_next_probe(j); | |
10172 | if (vcpu->arch.apf.gfns[j] == ~0) | |
10173 | return; | |
10174 | k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]); | |
10175 | /* | |
10176 | * k lies cyclically in ]i,j] | |
10177 | * | i.k.j | | |
10178 | * |....j i.k.| or |.k..j i...| | |
10179 | */ | |
10180 | } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j)); | |
10181 | vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j]; | |
10182 | i = j; | |
10183 | } | |
10184 | } | |
10185 | ||
7c90705b GN |
10186 | static int apf_put_user(struct kvm_vcpu *vcpu, u32 val) |
10187 | { | |
4e335d9e PB |
10188 | |
10189 | return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val, | |
10190 | sizeof(val)); | |
7c90705b GN |
10191 | } |
10192 | ||
9a6e7c39 WL |
10193 | static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val) |
10194 | { | |
10195 | ||
10196 | return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val, | |
10197 | sizeof(u32)); | |
10198 | } | |
10199 | ||
1dfdb45e PB |
10200 | static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu) |
10201 | { | |
10202 | if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu)) | |
10203 | return false; | |
10204 | ||
10205 | if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) || | |
10206 | (vcpu->arch.apf.send_user_only && | |
10207 | kvm_x86_ops->get_cpl(vcpu) == 0)) | |
10208 | return false; | |
10209 | ||
10210 | return true; | |
10211 | } | |
10212 | ||
10213 | bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu) | |
10214 | { | |
10215 | if (unlikely(!lapic_in_kernel(vcpu) || | |
10216 | kvm_event_needs_reinjection(vcpu) || | |
10217 | vcpu->arch.exception.pending)) | |
10218 | return false; | |
10219 | ||
10220 | if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu)) | |
10221 | return false; | |
10222 | ||
10223 | /* | |
10224 | * If interrupts are off we cannot even use an artificial | |
10225 | * halt state. | |
10226 | */ | |
10227 | return kvm_x86_ops->interrupt_allowed(vcpu); | |
10228 | } | |
10229 | ||
af585b92 GN |
10230 | void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu, |
10231 | struct kvm_async_pf *work) | |
10232 | { | |
6389ee94 AK |
10233 | struct x86_exception fault; |
10234 | ||
736c291c | 10235 | trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa); |
af585b92 | 10236 | kvm_add_async_pf_gfn(vcpu, work->arch.gfn); |
7c90705b | 10237 | |
1dfdb45e PB |
10238 | if (kvm_can_deliver_async_pf(vcpu) && |
10239 | !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) { | |
6389ee94 AK |
10240 | fault.vector = PF_VECTOR; |
10241 | fault.error_code_valid = true; | |
10242 | fault.error_code = 0; | |
10243 | fault.nested_page_fault = false; | |
10244 | fault.address = work->arch.token; | |
adfe20fb | 10245 | fault.async_page_fault = true; |
6389ee94 | 10246 | kvm_inject_page_fault(vcpu, &fault); |
1dfdb45e PB |
10247 | } else { |
10248 | /* | |
10249 | * It is not possible to deliver a paravirtualized asynchronous | |
10250 | * page fault, but putting the guest in an artificial halt state | |
10251 | * can be beneficial nevertheless: if an interrupt arrives, we | |
10252 | * can deliver it timely and perhaps the guest will schedule | |
10253 | * another process. When the instruction that triggered a page | |
10254 | * fault is retried, hopefully the page will be ready in the host. | |
10255 | */ | |
10256 | kvm_make_request(KVM_REQ_APF_HALT, vcpu); | |
7c90705b | 10257 | } |
af585b92 GN |
10258 | } |
10259 | ||
10260 | void kvm_arch_async_page_present(struct kvm_vcpu *vcpu, | |
10261 | struct kvm_async_pf *work) | |
10262 | { | |
6389ee94 | 10263 | struct x86_exception fault; |
9a6e7c39 | 10264 | u32 val; |
6389ee94 | 10265 | |
f2e10669 | 10266 | if (work->wakeup_all) |
7c90705b GN |
10267 | work->arch.token = ~0; /* broadcast wakeup */ |
10268 | else | |
10269 | kvm_del_async_pf_gfn(vcpu, work->arch.gfn); | |
736c291c | 10270 | trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa); |
7c90705b | 10271 | |
9a6e7c39 WL |
10272 | if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED && |
10273 | !apf_get_user(vcpu, &val)) { | |
10274 | if (val == KVM_PV_REASON_PAGE_NOT_PRESENT && | |
10275 | vcpu->arch.exception.pending && | |
10276 | vcpu->arch.exception.nr == PF_VECTOR && | |
10277 | !apf_put_user(vcpu, 0)) { | |
10278 | vcpu->arch.exception.injected = false; | |
10279 | vcpu->arch.exception.pending = false; | |
10280 | vcpu->arch.exception.nr = 0; | |
10281 | vcpu->arch.exception.has_error_code = false; | |
10282 | vcpu->arch.exception.error_code = 0; | |
c851436a JM |
10283 | vcpu->arch.exception.has_payload = false; |
10284 | vcpu->arch.exception.payload = 0; | |
9a6e7c39 WL |
10285 | } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) { |
10286 | fault.vector = PF_VECTOR; | |
10287 | fault.error_code_valid = true; | |
10288 | fault.error_code = 0; | |
10289 | fault.nested_page_fault = false; | |
10290 | fault.address = work->arch.token; | |
10291 | fault.async_page_fault = true; | |
10292 | kvm_inject_page_fault(vcpu, &fault); | |
10293 | } | |
7c90705b | 10294 | } |
e6d53e3b | 10295 | vcpu->arch.apf.halted = false; |
a4fa1635 | 10296 | vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; |
7c90705b GN |
10297 | } |
10298 | ||
10299 | bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu) | |
10300 | { | |
10301 | if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED)) | |
10302 | return true; | |
10303 | else | |
9bc1f09f | 10304 | return kvm_can_do_async_pf(vcpu); |
af585b92 GN |
10305 | } |
10306 | ||
5544eb9b PB |
10307 | void kvm_arch_start_assignment(struct kvm *kvm) |
10308 | { | |
10309 | atomic_inc(&kvm->arch.assigned_device_count); | |
10310 | } | |
10311 | EXPORT_SYMBOL_GPL(kvm_arch_start_assignment); | |
10312 | ||
10313 | void kvm_arch_end_assignment(struct kvm *kvm) | |
10314 | { | |
10315 | atomic_dec(&kvm->arch.assigned_device_count); | |
10316 | } | |
10317 | EXPORT_SYMBOL_GPL(kvm_arch_end_assignment); | |
10318 | ||
10319 | bool kvm_arch_has_assigned_device(struct kvm *kvm) | |
10320 | { | |
10321 | return atomic_read(&kvm->arch.assigned_device_count); | |
10322 | } | |
10323 | EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device); | |
10324 | ||
e0f0bbc5 AW |
10325 | void kvm_arch_register_noncoherent_dma(struct kvm *kvm) |
10326 | { | |
10327 | atomic_inc(&kvm->arch.noncoherent_dma_count); | |
10328 | } | |
10329 | EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma); | |
10330 | ||
10331 | void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm) | |
10332 | { | |
10333 | atomic_dec(&kvm->arch.noncoherent_dma_count); | |
10334 | } | |
10335 | EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma); | |
10336 | ||
10337 | bool kvm_arch_has_noncoherent_dma(struct kvm *kvm) | |
10338 | { | |
10339 | return atomic_read(&kvm->arch.noncoherent_dma_count); | |
10340 | } | |
10341 | EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma); | |
10342 | ||
14717e20 AW |
10343 | bool kvm_arch_has_irq_bypass(void) |
10344 | { | |
92735b1b | 10345 | return true; |
14717e20 AW |
10346 | } |
10347 | ||
87276880 FW |
10348 | int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons, |
10349 | struct irq_bypass_producer *prod) | |
10350 | { | |
10351 | struct kvm_kernel_irqfd *irqfd = | |
10352 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
10353 | ||
14717e20 | 10354 | irqfd->producer = prod; |
87276880 | 10355 | |
14717e20 AW |
10356 | return kvm_x86_ops->update_pi_irte(irqfd->kvm, |
10357 | prod->irq, irqfd->gsi, 1); | |
87276880 FW |
10358 | } |
10359 | ||
10360 | void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons, | |
10361 | struct irq_bypass_producer *prod) | |
10362 | { | |
10363 | int ret; | |
10364 | struct kvm_kernel_irqfd *irqfd = | |
10365 | container_of(cons, struct kvm_kernel_irqfd, consumer); | |
10366 | ||
87276880 FW |
10367 | WARN_ON(irqfd->producer != prod); |
10368 | irqfd->producer = NULL; | |
10369 | ||
10370 | /* | |
10371 | * When producer of consumer is unregistered, we change back to | |
10372 | * remapped mode, so we can re-use the current implementation | |
bb3541f1 | 10373 | * when the irq is masked/disabled or the consumer side (KVM |
87276880 FW |
10374 | * int this case doesn't want to receive the interrupts. |
10375 | */ | |
10376 | ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0); | |
10377 | if (ret) | |
10378 | printk(KERN_INFO "irq bypass consumer (token %p) unregistration" | |
10379 | " fails: %d\n", irqfd->consumer.token, ret); | |
10380 | } | |
10381 | ||
10382 | int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq, | |
10383 | uint32_t guest_irq, bool set) | |
10384 | { | |
87276880 FW |
10385 | return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set); |
10386 | } | |
10387 | ||
52004014 FW |
10388 | bool kvm_vector_hashing_enabled(void) |
10389 | { | |
10390 | return vector_hashing; | |
10391 | } | |
10392 | EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled); | |
10393 | ||
2d5ba19b MT |
10394 | bool kvm_arch_no_poll(struct kvm_vcpu *vcpu) |
10395 | { | |
10396 | return (vcpu->arch.msr_kvm_poll_control & 1) == 0; | |
10397 | } | |
10398 | EXPORT_SYMBOL_GPL(kvm_arch_no_poll); | |
10399 | ||
6441fa61 PB |
10400 | u64 kvm_spec_ctrl_valid_bits(struct kvm_vcpu *vcpu) |
10401 | { | |
10402 | uint64_t bits = SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD; | |
10403 | ||
10404 | /* The STIBP bit doesn't fault even if it's not advertised */ | |
10405 | if (!guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL) && | |
10406 | !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS)) | |
10407 | bits &= ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP); | |
10408 | if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL) && | |
10409 | !boot_cpu_has(X86_FEATURE_AMD_IBRS)) | |
10410 | bits &= ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP); | |
10411 | ||
10412 | if (!guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL_SSBD) && | |
10413 | !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD)) | |
10414 | bits &= ~SPEC_CTRL_SSBD; | |
10415 | if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) && | |
10416 | !boot_cpu_has(X86_FEATURE_AMD_SSBD)) | |
10417 | bits &= ~SPEC_CTRL_SSBD; | |
10418 | ||
10419 | return bits; | |
10420 | } | |
10421 | EXPORT_SYMBOL_GPL(kvm_spec_ctrl_valid_bits); | |
2d5ba19b | 10422 | |
229456fc | 10423 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit); |
931c33b1 | 10424 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio); |
229456fc MT |
10425 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq); |
10426 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault); | |
10427 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr); | |
10428 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr); | |
0ac406de | 10429 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun); |
d8cabddf | 10430 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit); |
17897f36 | 10431 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject); |
236649de | 10432 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit); |
5497b955 | 10433 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed); |
ec1ff790 | 10434 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga); |
532a46b9 | 10435 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit); |
2e554e8d | 10436 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts); |
489223ed | 10437 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset); |
4f75bcc3 | 10438 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update); |
843e4330 | 10439 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full); |
efc64404 | 10440 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update); |
18f40c53 SS |
10441 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access); |
10442 | EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi); |