]> git.proxmox.com Git - mirror_ubuntu-focal-kernel.git/blame - arch/x86/kvm/x86.c
projects / mirror_ubuntu-focal-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge remote-tracking branch 'asoc/topic/pcm512x' into asoc-next
[mirror_ubuntu-focal-kernel.git] / arch / x86 / kvm / x86.c
CommitLineData
043405e1
CO		 1/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * derived from drivers/kvm/kvm_main.c
5 *
6 * Copyright (C) 2006 Qumranet, Inc.
4d5c5d0f
BAY		 7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9611c187 9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
043405e1
CO		 10 *
11 * Authors:
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
4d5c5d0f
BAY		 14 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
043405e1
CO		 16 *
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
19 *
20 */
21
edf88417 22#include <linux/kvm_host.h>
313a3dc7 23#include "irq.h"
1d737c8a 24#include "mmu.h"
7837699f 25#include "i8254.h"
37817f29 26#include "tss.h"
5fdbf976 27#include "kvm_cache_regs.h"
26eef70c 28#include "x86.h"
00b27a3e 29#include "cpuid.h"
474a5bb9 30#include "pmu.h"
e83d5887 31#include "hyperv.h"
313a3dc7 32
18068523 33#include <linux/clocksource.h>
4d5c5d0f 34#include <linux/interrupt.h>
313a3dc7
CO		 35#include <linux/kvm.h>
36#include <linux/fs.h>
37#include <linux/vmalloc.h>
1767e931
PG		 38#include <linux/export.h>
39#include <linux/moduleparam.h>
0de10343 40#include <linux/mman.h>
2bacc55c 41#include <linux/highmem.h>
19de40a8 42#include <linux/iommu.h>
62c476c7 43#include <linux/intel-iommu.h>
c8076604 44#include <linux/cpufreq.h>
18863bdd 45#include <linux/user-return-notifier.h>
a983fb23 46#include <linux/srcu.h>
5a0e3ad6 47#include <linux/slab.h>
ff9d07a0 48#include <linux/perf_event.h>
7bee342a 49#include <linux/uaccess.h>
af585b92 50#include <linux/hash.h>
a1b60c1c 51#include <linux/pci.h>
16e8d74d
MT		 52#include <linux/timekeeper_internal.h>
53#include <linux/pvclock_gtod.h>
87276880
FW		 54#include <linux/kvm_irqfd.h>
55#include <linux/irqbypass.h>
3905f9ad 56#include <linux/sched/stat.h>
d0ec49d4 57#include <linux/mem_encrypt.h>
3905f9ad 58
aec51dc4 59#include <trace/events/kvm.h>
2ed152af 60
24f1e32c 61#include <asm/debugreg.h>
d825ed0a 62#include <asm/msr.h>
a5f61300 63#include <asm/desc.h>
890ca9ae 64#include <asm/mce.h>
f89e32e0 65#include <linux/kernel_stat.h>
78f7f1e5 66#include <asm/fpu/internal.h> /* Ugh! */
1d5f066e 67#include <asm/pvclock.h>
217fc9cf 68#include <asm/div64.h>
efc64404 69#include <asm/irq_remapping.h>
043405e1 70
d1898b73
DH		 71#define CREATE_TRACE_POINTS
72#include "trace.h"
73
313a3dc7 74#define MAX_IO_MSRS 256
890ca9ae 75#define KVM_MAX_MCE_BANKS 32
c45dcc71
AR		 76u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
77EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
890ca9ae 78
0f65dd70
AK		 79#define emul_to_vcpu(ctxt) \
80 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
81
50a37eb4
JR		 82/* EFER defaults:
83 * - enable syscall per default because its emulated by KVM
84 * - enable LME and LMA per default on 64 bit KVM
85 */
86#ifdef CONFIG_X86_64
1260edbe
LJ		 87static
88u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
50a37eb4 89#else
1260edbe 90static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
50a37eb4 91#endif
313a3dc7 92
ba1389b7
AK		 93#define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
94#define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
417bc304 95
c519265f
RK		 96#define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
97 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
37131313 98
cb142eb7 99static void update_cr8_intercept(struct kvm_vcpu *vcpu);
7460fb4a 100static void process_nmi(struct kvm_vcpu *vcpu);
ee2cd4b7 101static void enter_smm(struct kvm_vcpu *vcpu);
6addfc42 102static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
674eea0f 103
893590c7 104struct kvm_x86_ops *kvm_x86_ops __read_mostly;
5fdbf976 105EXPORT_SYMBOL_GPL(kvm_x86_ops);
97896d04 106
893590c7 107static bool __read_mostly ignore_msrs = 0;
476bc001 108module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
ed85c068 109
fab0aa3b
EM		 110static bool __read_mostly report_ignored_msrs = true;
111module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
112
9ed96e87
MT		 113unsigned int min_timer_period_us = 500;
114module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
115
630994b3
MT		 116static bool __read_mostly kvmclock_periodic_sync = true;
117module_param(kvmclock_periodic_sync, bool, S_IRUGO);
118
893590c7 119bool __read_mostly kvm_has_tsc_control;
92a1f12d 120EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
893590c7 121u32 __read_mostly kvm_max_guest_tsc_khz;
92a1f12d 122EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
bc9b961b
HZ		 123u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
124EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
125u64 __read_mostly kvm_max_tsc_scaling_ratio;
126EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
64672c95
YJ		 127u64 __read_mostly kvm_default_tsc_scaling_ratio;
128EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
92a1f12d 129
cc578287 130/* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
893590c7 131static u32 __read_mostly tsc_tolerance_ppm = 250;
cc578287
ZA		 132module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
133
d0659d94 134/* lapic timer advance (tscdeadline mode only) in nanoseconds */
893590c7 135unsigned int __read_mostly lapic_timer_advance_ns = 0;
d0659d94
MT		 136module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
137
52004014
FW		 138static bool __read_mostly vector_hashing = true;
139module_param(vector_hashing, bool, S_IRUGO);
140
18863bdd
AK		 141#define KVM_NR_SHARED_MSRS 16
142
143struct kvm_shared_msrs_global {
144 int nr;
2bf78fa7 145 u32 msrs[KVM_NR_SHARED_MSRS];
18863bdd
AK		 146};
147
148struct kvm_shared_msrs {
149 struct user_return_notifier urn;
150 bool registered;
2bf78fa7
SY		 151 struct kvm_shared_msr_values {
152 u64 host;
153 u64 curr;
154 } values[KVM_NR_SHARED_MSRS];
18863bdd
AK		 155};
156
157static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
013f6a5d 158static struct kvm_shared_msrs __percpu *shared_msrs;
18863bdd 159
417bc304 160struct kvm_stats_debugfs_item debugfs_entries[] = {
ba1389b7
AK		 161 { "pf_fixed", VCPU_STAT(pf_fixed) },
162 { "pf_guest", VCPU_STAT(pf_guest) },
163 { "tlb_flush", VCPU_STAT(tlb_flush) },
164 { "invlpg", VCPU_STAT(invlpg) },
165 { "exits", VCPU_STAT(exits) },
166 { "io_exits", VCPU_STAT(io_exits) },
167 { "mmio_exits", VCPU_STAT(mmio_exits) },
168 { "signal_exits", VCPU_STAT(signal_exits) },
169 { "irq_window", VCPU_STAT(irq_window_exits) },
f08864b4 170 { "nmi_window", VCPU_STAT(nmi_window_exits) },
ba1389b7 171 { "halt_exits", VCPU_STAT(halt_exits) },
f7819512 172 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
62bea5bf 173 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
3491caf2 174 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
ba1389b7 175 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
f11c3a8d 176 { "hypercalls", VCPU_STAT(hypercalls) },
ba1389b7
AK		 177 { "request_irq", VCPU_STAT(request_irq_exits) },
178 { "irq_exits", VCPU_STAT(irq_exits) },
179 { "host_state_reload", VCPU_STAT(host_state_reload) },
180 { "efer_reload", VCPU_STAT(efer_reload) },
181 { "fpu_reload", VCPU_STAT(fpu_reload) },
182 { "insn_emulation", VCPU_STAT(insn_emulation) },
183 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
fa89a817 184 { "irq_injections", VCPU_STAT(irq_injections) },
c4abb7c9 185 { "nmi_injections", VCPU_STAT(nmi_injections) },
0f1e261e 186 { "req_event", VCPU_STAT(req_event) },
4cee5764
AK		 187 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
188 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
189 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
190 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
191 { "mmu_flooded", VM_STAT(mmu_flooded) },
192 { "mmu_recycled", VM_STAT(mmu_recycled) },
dfc5aa00 193 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
4731d4c7 194 { "mmu_unsync", VM_STAT(mmu_unsync) },
0f74a24c 195 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
05da4558 196 { "largepages", VM_STAT(lpages) },
f3414bc7
DM		 197 { "max_mmu_page_hash_collisions",
198 VM_STAT(max_mmu_page_hash_collisions) },
417bc304
HB		 199 { NULL }
200};
201
2acf923e
DC		 202u64 __read_mostly host_xcr0;
203
b6785def 204static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
d6aa1000 205
af585b92
GN		 206static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
207{
208 int i;
209 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
210 vcpu->arch.apf.gfns[i] = ~0;
211}
212
18863bdd
AK		 213static void kvm_on_user_return(struct user_return_notifier *urn)
214{
215 unsigned slot;
18863bdd
AK		 216 struct kvm_shared_msrs *locals
217 = container_of(urn, struct kvm_shared_msrs, urn);
2bf78fa7 218 struct kvm_shared_msr_values *values;
1650b4eb
IA		 219 unsigned long flags;
220
221 /*
222 * Disabling irqs at this point since the following code could be
223 * interrupted and executed through kvm_arch_hardware_disable()
224 */
225 local_irq_save(flags);
226 if (locals->registered) {
227 locals->registered = false;
228 user_return_notifier_unregister(urn);
229 }
230 local_irq_restore(flags);
18863bdd 231 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
2bf78fa7
SY		 232 values = &locals->values[slot];
233 if (values->host != values->curr) {
234 wrmsrl(shared_msrs_global.msrs[slot], values->host);
235 values->curr = values->host;
18863bdd
AK		 236 }
237 }
18863bdd
AK		 238}
239
2bf78fa7 240static void shared_msr_update(unsigned slot, u32 msr)
18863bdd 241{
18863bdd 242 u64 value;
013f6a5d
MT		 243 unsigned int cpu = smp_processor_id();
244 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
18863bdd 245
2bf78fa7
SY		 246 /* only read, and nobody should modify it at this time,
247 * so don't need lock */
248 if (slot >= shared_msrs_global.nr) {
249 printk(KERN_ERR "kvm: invalid MSR slot!");
250 return;
251 }
252 rdmsrl_safe(msr, &value);
253 smsr->values[slot].host = value;
254 smsr->values[slot].curr = value;
255}
256
257void kvm_define_shared_msr(unsigned slot, u32 msr)
258{
0123be42 259 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
c847fe88 260 shared_msrs_global.msrs[slot] = msr;
18863bdd
AK		 261 if (slot >= shared_msrs_global.nr)
262 shared_msrs_global.nr = slot + 1;
18863bdd
AK		 263}
264EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
265
266static void kvm_shared_msr_cpu_online(void)
267{
268 unsigned i;
18863bdd
AK		 269
270 for (i = 0; i < shared_msrs_global.nr; ++i)
2bf78fa7 271 shared_msr_update(i, shared_msrs_global.msrs[i]);
18863bdd
AK		 272}
273
8b3c3104 274int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
18863bdd 275{
013f6a5d
MT		 276 unsigned int cpu = smp_processor_id();
277 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
8b3c3104 278 int err;
18863bdd 279
2bf78fa7 280 if (((value ^ smsr->values[slot].curr) & mask) == 0)
8b3c3104 281 return 0;
2bf78fa7 282 smsr->values[slot].curr = value;
8b3c3104
AH		 283 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
284 if (err)
285 return 1;
286
18863bdd
AK		 287 if (!smsr->registered) {
288 smsr->urn.on_user_return = kvm_on_user_return;
289 user_return_notifier_register(&smsr->urn);
290 smsr->registered = true;
291 }
8b3c3104 292 return 0;
18863bdd
AK		 293}
294EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
295
13a34e06 296static void drop_user_return_notifiers(void)
3548bab5 297{
013f6a5d
MT		 298 unsigned int cpu = smp_processor_id();
299 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
3548bab5
AK		 300
301 if (smsr->registered)
302 kvm_on_user_return(&smsr->urn);
303}
304
6866b83e
CO		 305u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
306{
8a5a87d9 307 return vcpu->arch.apic_base;
6866b83e
CO		 308}
309EXPORT_SYMBOL_GPL(kvm_get_apic_base);
310
58cb628d
JK		 311int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
312{
313 u64 old_state = vcpu->arch.apic_base &
314 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
315 u64 new_state = msr_info->data &
316 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
d6321d49
RK		 317 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
318 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
58cb628d 319
d3802286
JM		 320 if ((msr_info->data & reserved_bits) || new_state == X2APIC_ENABLE)
321 return 1;
58cb628d 322 if (!msr_info->host_initiated &&
d3802286 323 ((new_state == MSR_IA32_APICBASE_ENABLE &&
58cb628d
JK		 324 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
325 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
326 old_state == 0)))
327 return 1;
328
329 kvm_lapic_set_base(vcpu, msr_info->data);
330 return 0;
6866b83e
CO		 331}
332EXPORT_SYMBOL_GPL(kvm_set_apic_base);
333
2605fc21 334asmlinkage __visible void kvm_spurious_fault(void)
e3ba45b8
GL		 335{
336 /* Fault while not rebooting. We want the trace. */
337 BUG();
338}
339EXPORT_SYMBOL_GPL(kvm_spurious_fault);
340
3fd28fce
ED		 341#define EXCPT_BENIGN 0
342#define EXCPT_CONTRIBUTORY 1
343#define EXCPT_PF 2
344
345static int exception_class(int vector)
346{
347 switch (vector) {
348 case PF_VECTOR:
349 return EXCPT_PF;
350 case DE_VECTOR:
351 case TS_VECTOR:
352 case NP_VECTOR:
353 case SS_VECTOR:
354 case GP_VECTOR:
355 return EXCPT_CONTRIBUTORY;
356 default:
357 break;
358 }
359 return EXCPT_BENIGN;
360}
361
d6e8c854
NA		 362#define EXCPT_FAULT 0
363#define EXCPT_TRAP 1
364#define EXCPT_ABORT 2
365#define EXCPT_INTERRUPT 3
366
367static int exception_type(int vector)
368{
369 unsigned int mask;
370
371 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
372 return EXCPT_INTERRUPT;
373
374 mask = 1 << vector;
375
376 /* #DB is trap, as instruction watchpoints are handled elsewhere */
377 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
378 return EXCPT_TRAP;
379
380 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
381 return EXCPT_ABORT;
382
383 /* Reserved exceptions will result in fault */
384 return EXCPT_FAULT;
385}
386
3fd28fce 387static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
ce7ddec4
JR		 388 unsigned nr, bool has_error, u32 error_code,
389 bool reinject)
3fd28fce
ED		 390{
391 u32 prev_nr;
392 int class1, class2;
393
3842d135
AK		 394 kvm_make_request(KVM_REQ_EVENT, vcpu);
395
664f8e26 396 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
3fd28fce 397 queue:
3ffb2468
NA		 398 if (has_error && !is_protmode(vcpu))
399 has_error = false;
664f8e26
WL		 400 if (reinject) {
401 /*
402 * On vmentry, vcpu->arch.exception.pending is only
403 * true if an event injection was blocked by
404 * nested_run_pending. In that case, however,
405 * vcpu_enter_guest requests an immediate exit,
406 * and the guest shouldn't proceed far enough to
407 * need reinjection.
408 */
409 WARN_ON_ONCE(vcpu->arch.exception.pending);
410 vcpu->arch.exception.injected = true;
411 } else {
412 vcpu->arch.exception.pending = true;
413 vcpu->arch.exception.injected = false;
414 }
3fd28fce
ED		 415 vcpu->arch.exception.has_error_code = has_error;
416 vcpu->arch.exception.nr = nr;
417 vcpu->arch.exception.error_code = error_code;
418 return;
419 }
420
421 /* to check exception */
422 prev_nr = vcpu->arch.exception.nr;
423 if (prev_nr == DF_VECTOR) {
424 /* triple fault -> shutdown */
a8eeb04a 425 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3fd28fce
ED		 426 return;
427 }
428 class1 = exception_class(prev_nr);
429 class2 = exception_class(nr);
430 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
431 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
664f8e26
WL		 432 /*
433 * Generate double fault per SDM Table 5-5. Set
434 * exception.pending = true so that the double fault
435 * can trigger a nested vmexit.
436 */
3fd28fce 437 vcpu->arch.exception.pending = true;
664f8e26 438 vcpu->arch.exception.injected = false;
3fd28fce
ED		 439 vcpu->arch.exception.has_error_code = true;
440 vcpu->arch.exception.nr = DF_VECTOR;
441 vcpu->arch.exception.error_code = 0;
442 } else
443 /* replace previous exception with a new one in a hope
444 that instruction re-execution will regenerate lost
445 exception */
446 goto queue;
447}
448
298101da
AK		 449void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
450{
ce7ddec4 451 kvm_multiple_exception(vcpu, nr, false, 0, false);
298101da
AK		 452}
453EXPORT_SYMBOL_GPL(kvm_queue_exception);
454
ce7ddec4
JR		 455void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
456{
457 kvm_multiple_exception(vcpu, nr, false, 0, true);
458}
459EXPORT_SYMBOL_GPL(kvm_requeue_exception);
460
6affcbed 461int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
c3c91fee 462{
db8fcefa
AP		 463 if (err)
464 kvm_inject_gp(vcpu, 0);
465 else
6affcbed
KH		 466 return kvm_skip_emulated_instruction(vcpu);
467
468 return 1;
db8fcefa
AP		 469}
470EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
8df25a32 471
6389ee94 472void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
c3c91fee
AK		 473{
474 ++vcpu->stat.pf_guest;
adfe20fb
WL		 475 vcpu->arch.exception.nested_apf =
476 is_guest_mode(vcpu) && fault->async_page_fault;
477 if (vcpu->arch.exception.nested_apf)
478 vcpu->arch.apf.nested_apf_token = fault->address;
479 else
480 vcpu->arch.cr2 = fault->address;
6389ee94 481 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
c3c91fee 482}
27d6c865 483EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
c3c91fee 484
ef54bcfe 485static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
d4f8cf66 486{
6389ee94
AK		 487 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
488 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
d4f8cf66 489 else
6389ee94 490 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
ef54bcfe
PB		 491
492 return fault->nested_page_fault;
d4f8cf66
JR		 493}
494
3419ffc8
SY		 495void kvm_inject_nmi(struct kvm_vcpu *vcpu)
496{
7460fb4a
AK		 497 atomic_inc(&vcpu->arch.nmi_queued);
498 kvm_make_request(KVM_REQ_NMI, vcpu);
3419ffc8
SY		 499}
500EXPORT_SYMBOL_GPL(kvm_inject_nmi);
501
298101da
AK		 502void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
503{
ce7ddec4 504 kvm_multiple_exception(vcpu, nr, true, error_code, false);
298101da
AK		 505}
506EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
507
ce7ddec4
JR		 508void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
509{
510 kvm_multiple_exception(vcpu, nr, true, error_code, true);
511}
512EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
513
0a79b009
AK		 514/*
515 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
516 * a #GP and return false.
517 */
518bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
298101da 519{
0a79b009
AK		 520 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
521 return true;
522 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
523 return false;
298101da 524}
0a79b009 525EXPORT_SYMBOL_GPL(kvm_require_cpl);
298101da 526
16f8a6f9
NA		 527bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
528{
529 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
530 return true;
531
532 kvm_queue_exception(vcpu, UD_VECTOR);
533 return false;
534}
535EXPORT_SYMBOL_GPL(kvm_require_dr);
536
ec92fe44
JR		 537/*
538 * This function will be used to read from the physical memory of the currently
54bf36aa 539 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
ec92fe44
JR		 540 * can read from guest physical or from the guest's guest physical memory.
541 */
542int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
543 gfn_t ngfn, void *data, int offset, int len,
544 u32 access)
545{
54987b7a 546 struct x86_exception exception;
ec92fe44
JR		 547 gfn_t real_gfn;
548 gpa_t ngpa;
549
550 ngpa = gfn_to_gpa(ngfn);
54987b7a 551 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
ec92fe44
JR		 552 if (real_gfn == UNMAPPED_GVA)
553 return -EFAULT;
554
555 real_gfn = gpa_to_gfn(real_gfn);
556
54bf36aa 557 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
ec92fe44
JR		 558}
559EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
560
69b0049a 561static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
3d06b8bf
JR		 562 void *data, int offset, int len, u32 access)
563{
564 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
565 data, offset, len, access);
566}
567
a03490ed
CO		 568/*
569 * Load the pae pdptrs. Return true is they are all valid.
570 */
ff03a073 571int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
a03490ed
CO		 572{
573 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
574 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
575 int i;
576 int ret;
ff03a073 577 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
a03490ed 578
ff03a073
JR		 579 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
580 offset * sizeof(u64), sizeof(pdpte),
581 PFERR_USER_MASK|PFERR_WRITE_MASK);
a03490ed
CO		 582 if (ret < 0) {
583 ret = 0;
584 goto out;
585 }
586 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
812f30b2 587 if ((pdpte[i] & PT_PRESENT_MASK) &&
a0a64f50
XG		 588 (pdpte[i] &
589 vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
a03490ed
CO		 590 ret = 0;
591 goto out;
592 }
593 }
594 ret = 1;
595
ff03a073 596 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
6de4f3ad
AK		 597 __set_bit(VCPU_EXREG_PDPTR,
598 (unsigned long *)&vcpu->arch.regs_avail);
599 __set_bit(VCPU_EXREG_PDPTR,
600 (unsigned long *)&vcpu->arch.regs_dirty);
a03490ed 601out:
a03490ed
CO		 602
603 return ret;
604}
cc4b6871 605EXPORT_SYMBOL_GPL(load_pdptrs);
a03490ed 606
9ed38ffa 607bool pdptrs_changed(struct kvm_vcpu *vcpu)
d835dfec 608{
ff03a073 609 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
d835dfec 610 bool changed = true;
3d06b8bf
JR		 611 int offset;
612 gfn_t gfn;
d835dfec
AK		 613 int r;
614
615 if (is_long_mode(vcpu) || !is_pae(vcpu))
616 return false;
617
6de4f3ad
AK		 618 if (!test_bit(VCPU_EXREG_PDPTR,
619 (unsigned long *)&vcpu->arch.regs_avail))
620 return true;
621
a512177e
PB		 622 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
623 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
3d06b8bf
JR		 624 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
625 PFERR_USER_MASK | PFERR_WRITE_MASK);
d835dfec
AK		 626 if (r < 0)
627 goto out;
ff03a073 628 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
d835dfec 629out:
d835dfec
AK		 630
631 return changed;
632}
9ed38ffa 633EXPORT_SYMBOL_GPL(pdptrs_changed);
d835dfec 634
49a9b07e 635int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
a03490ed 636{
aad82703 637 unsigned long old_cr0 = kvm_read_cr0(vcpu);
d81135a5 638 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
aad82703 639
f9a48e6a
AK		 640 cr0 |= X86_CR0_ET;
641
ab344828 642#ifdef CONFIG_X86_64
0f12244f
GN		 643 if (cr0 & 0xffffffff00000000UL)
644 return 1;
ab344828
GN		 645#endif
646
647 cr0 &= ~CR0_RESERVED_BITS;
a03490ed 648
0f12244f
GN		 649 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
650 return 1;
a03490ed 651
0f12244f
GN		 652 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
653 return 1;
a03490ed
CO		 654
655 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
656#ifdef CONFIG_X86_64
f6801dff 657 if ((vcpu->arch.efer & EFER_LME)) {
a03490ed
CO		 658 int cs_db, cs_l;
659
0f12244f
GN		 660 if (!is_pae(vcpu))
661 return 1;
a03490ed 662 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
0f12244f
GN		 663 if (cs_l)
664 return 1;
a03490ed
CO		 665 } else
666#endif
ff03a073 667 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
9f8fe504 668 kvm_read_cr3(vcpu)))
0f12244f 669 return 1;
a03490ed
CO		 670 }
671
ad756a16
MJ		 672 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
673 return 1;
674
a03490ed 675 kvm_x86_ops->set_cr0(vcpu, cr0);
a03490ed 676
d170c419 677 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
e5f3f027 678 kvm_clear_async_pf_completion_queue(vcpu);
d170c419
LJ		 679 kvm_async_pf_hash_reset(vcpu);
680 }
e5f3f027 681
aad82703
SY		 682 if ((cr0 ^ old_cr0) & update_bits)
683 kvm_mmu_reset_context(vcpu);
b18d5431 684
879ae188
LE		 685 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
686 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
687 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
b18d5431
XG		 688 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
689
0f12244f
GN		 690 return 0;
691}
2d3ad1f4 692EXPORT_SYMBOL_GPL(kvm_set_cr0);
a03490ed 693
2d3ad1f4 694void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
a03490ed 695{
49a9b07e 696 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
a03490ed 697}
2d3ad1f4 698EXPORT_SYMBOL_GPL(kvm_lmsw);
a03490ed 699
42bdf991
MT		 700static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
701{
702 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
703 !vcpu->guest_xcr0_loaded) {
704 /* kvm_set_xcr() also depends on this */
705 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
706 vcpu->guest_xcr0_loaded = 1;
707 }
708}
709
710static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
711{
712 if (vcpu->guest_xcr0_loaded) {
713 if (vcpu->arch.xcr0 != host_xcr0)
714 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
715 vcpu->guest_xcr0_loaded = 0;
716 }
717}
718
69b0049a 719static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
2acf923e 720{
56c103ec
LJ		 721 u64 xcr0 = xcr;
722 u64 old_xcr0 = vcpu->arch.xcr0;
46c34cb0 723 u64 valid_bits;
2acf923e
DC		 724
725 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
726 if (index != XCR_XFEATURE_ENABLED_MASK)
727 return 1;
d91cab78 728 if (!(xcr0 & XFEATURE_MASK_FP))
2acf923e 729 return 1;
d91cab78 730 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
2acf923e 731 return 1;
46c34cb0
PB		 732
733 /*
734 * Do not allow the guest to set bits that we do not support
735 * saving. However, xcr0 bit 0 is always set, even if the
736 * emulated CPU does not support XSAVE (see fx_init).
737 */
d91cab78 738 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
46c34cb0 739 if (xcr0 & ~valid_bits)
2acf923e 740 return 1;
46c34cb0 741
d91cab78
DH		 742 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
743 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
390bd528
LJ		 744 return 1;
745
d91cab78
DH		 746 if (xcr0 & XFEATURE_MASK_AVX512) {
747 if (!(xcr0 & XFEATURE_MASK_YMM))
612263b3 748 return 1;
d91cab78 749 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
612263b3
CP		 750 return 1;
751 }
2acf923e 752 vcpu->arch.xcr0 = xcr0;
56c103ec 753
d91cab78 754 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
56c103ec 755 kvm_update_cpuid(vcpu);
2acf923e
DC		 756 return 0;
757}
758
759int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
760{
764bcbc5
Z		 761 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
762 __kvm_set_xcr(vcpu, index, xcr)) {
2acf923e
DC		 763 kvm_inject_gp(vcpu, 0);
764 return 1;
765 }
766 return 0;
767}
768EXPORT_SYMBOL_GPL(kvm_set_xcr);
769
a83b29c6 770int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
a03490ed 771{
fc78f519 772 unsigned long old_cr4 = kvm_read_cr4(vcpu);
0be0226f 773 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
b9baba86 774 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
0be0226f 775
0f12244f
GN		 776 if (cr4 & CR4_RESERVED_BITS)
777 return 1;
a03490ed 778
d6321d49 779 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
2acf923e
DC		 780 return 1;
781
d6321d49 782 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
2acf923e
DC		 783 return 1;
784
d6321d49 785 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
c68b734f
YW		 786 return 1;
787
d6321d49 788 if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
97ec8c06
FW		 789 return 1;
790
d6321d49 791 if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
74dc2b4f
YW		 792 return 1;
793
fd8cb433 794 if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
b9baba86
HH		 795 return 1;
796
a03490ed 797 if (is_long_mode(vcpu)) {
0f12244f
GN		 798 if (!(cr4 & X86_CR4_PAE))
799 return 1;
a2edf57f
AK		 800 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
801 && ((cr4 ^ old_cr4) & pdptr_bits)
9f8fe504
AK		 802 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
803 kvm_read_cr3(vcpu)))
0f12244f
GN		 804 return 1;
805
ad756a16 806 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
d6321d49 807 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
ad756a16
MJ		 808 return 1;
809
810 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
811 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
812 return 1;
813 }
814
5e1746d6 815 if (kvm_x86_ops->set_cr4(vcpu, cr4))
0f12244f 816 return 1;
a03490ed 817
ad756a16
MJ		 818 if (((cr4 ^ old_cr4) & pdptr_bits) ||
819 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
aad82703 820 kvm_mmu_reset_context(vcpu);
0f12244f 821
b9baba86 822 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
00b27a3e 823 kvm_update_cpuid(vcpu);
2acf923e 824
0f12244f
GN		 825 return 0;
826}
2d3ad1f4 827EXPORT_SYMBOL_GPL(kvm_set_cr4);
a03490ed 828
2390218b 829int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
a03490ed 830{
ac146235 831#ifdef CONFIG_X86_64
9d88fca7 832 cr3 &= ~CR3_PCID_INVD;
ac146235 833#endif
9d88fca7 834
9f8fe504 835 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
0ba73cda 836 kvm_mmu_sync_roots(vcpu);
77c3913b 837 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
0f12244f 838 return 0;
d835dfec
AK		 839 }
840
d1cd3ce9
YZ		 841 if (is_long_mode(vcpu) &&
842 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 62)))
843 return 1;
844 else if (is_pae(vcpu) && is_paging(vcpu) &&
d9f89b88 845 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
346874c9 846 return 1;
a03490ed 847
0f12244f 848 vcpu->arch.cr3 = cr3;
aff48baa 849 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
d8d173da 850 kvm_mmu_new_cr3(vcpu);
0f12244f
GN		 851 return 0;
852}
2d3ad1f4 853EXPORT_SYMBOL_GPL(kvm_set_cr3);
a03490ed 854
eea1cff9 855int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
a03490ed 856{
0f12244f
GN		 857 if (cr8 & CR8_RESERVED_BITS)
858 return 1;
35754c98 859 if (lapic_in_kernel(vcpu))
a03490ed
CO		 860 kvm_lapic_set_tpr(vcpu, cr8);
861 else
ad312c7c 862 vcpu->arch.cr8 = cr8;
0f12244f
GN		 863 return 0;
864}
2d3ad1f4 865EXPORT_SYMBOL_GPL(kvm_set_cr8);
a03490ed 866
2d3ad1f4 867unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
a03490ed 868{
35754c98 869 if (lapic_in_kernel(vcpu))
a03490ed
CO		 870 return kvm_lapic_get_cr8(vcpu);
871 else
ad312c7c 872 return vcpu->arch.cr8;
a03490ed 873}
2d3ad1f4 874EXPORT_SYMBOL_GPL(kvm_get_cr8);
a03490ed 875
ae561ede
NA		 876static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
877{
878 int i;
879
880 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
881 for (i = 0; i < KVM_NR_DB_REGS; i++)
882 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
883 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
884 }
885}
886
73aaf249
JK		 887static void kvm_update_dr6(struct kvm_vcpu *vcpu)
888{
889 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
890 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
891}
892
c8639010
JK		 893static void kvm_update_dr7(struct kvm_vcpu *vcpu)
894{
895 unsigned long dr7;
896
897 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
898 dr7 = vcpu->arch.guest_debug_dr7;
899 else
900 dr7 = vcpu->arch.dr7;
901 kvm_x86_ops->set_dr7(vcpu, dr7);
360b948d
PB		 902 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
903 if (dr7 & DR7_BP_EN_MASK)
904 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
c8639010
JK		 905}
906
6f43ed01
NA		 907static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
908{
909 u64 fixed = DR6_FIXED_1;
910
d6321d49 911 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
6f43ed01
NA		 912 fixed |= DR6_RTM;
913 return fixed;
914}
915
338dbc97 916static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
020df079
GN		 917{
918 switch (dr) {
919 case 0 ... 3:
920 vcpu->arch.db[dr] = val;
921 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
922 vcpu->arch.eff_db[dr] = val;
923 break;
924 case 4:
020df079
GN		 925 /* fall through */
926 case 6:
338dbc97
GN		 927 if (val & 0xffffffff00000000ULL)
928 return -1; /* #GP */
6f43ed01 929 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
73aaf249 930 kvm_update_dr6(vcpu);
020df079
GN		 931 break;
932 case 5:
020df079
GN		 933 /* fall through */
934 default: /* 7 */
338dbc97
GN		 935 if (val & 0xffffffff00000000ULL)
936 return -1; /* #GP */
020df079 937 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
c8639010 938 kvm_update_dr7(vcpu);
020df079
GN		 939 break;
940 }
941
942 return 0;
943}
338dbc97
GN		 944
945int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
946{
16f8a6f9 947 if (__kvm_set_dr(vcpu, dr, val)) {
338dbc97 948 kvm_inject_gp(vcpu, 0);
16f8a6f9
NA		 949 return 1;
950 }
951 return 0;
338dbc97 952}
020df079
GN		 953EXPORT_SYMBOL_GPL(kvm_set_dr);
954
16f8a6f9 955int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
020df079
GN		 956{
957 switch (dr) {
958 case 0 ... 3:
959 *val = vcpu->arch.db[dr];
960 break;
961 case 4:
020df079
GN		 962 /* fall through */
963 case 6:
73aaf249
JK		 964 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
965 *val = vcpu->arch.dr6;
966 else
967 *val = kvm_x86_ops->get_dr6(vcpu);
020df079
GN		 968 break;
969 case 5:
020df079
GN		 970 /* fall through */
971 default: /* 7 */
972 *val = vcpu->arch.dr7;
973 break;
974 }
338dbc97
GN		 975 return 0;
976}
020df079
GN		 977EXPORT_SYMBOL_GPL(kvm_get_dr);
978
022cd0e8
AK		 979bool kvm_rdpmc(struct kvm_vcpu *vcpu)
980{
981 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
982 u64 data;
983 int err;
984
c6702c9d 985 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
022cd0e8
AK		 986 if (err)
987 return err;
988 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
989 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
990 return err;
991}
992EXPORT_SYMBOL_GPL(kvm_rdpmc);
993
043405e1
CO		 994/*
995 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
996 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
997 *
998 * This list is modified at module load time to reflect the
e3267cbb 999 * capabilities of the host cpu. This capabilities test skips MSRs that are
62ef68bb
PB		 1000 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1001 * may depend on host virtualization features rather than host cpu features.
043405e1 1002 */
e3267cbb 1003
043405e1
CO		 1004static u32 msrs_to_save[] = {
1005 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
8c06585d 1006 MSR_STAR,
043405e1
CO		 1007#ifdef CONFIG_X86_64
1008 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1009#endif
b3897a49 1010 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
9dbe6cf9 1011 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
043405e1
CO		 1012};
1013
1014static unsigned num_msrs_to_save;
1015
62ef68bb
PB		 1016static u32 emulated_msrs[] = {
1017 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1018 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1019 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1020 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
72c139ba 1021 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
e7d9513b
AS		 1022 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1023 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
e516cebb 1024 HV_X64_MSR_RESET,
11c4b1ca 1025 HV_X64_MSR_VP_INDEX,
9eec50b8 1026 HV_X64_MSR_VP_RUNTIME,
5c919412 1027 HV_X64_MSR_SCONTROL,
1f4b34f8 1028 HV_X64_MSR_STIMER0_CONFIG,
62ef68bb
PB		 1029 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1030 MSR_KVM_PV_EOI_EN,
1031
ba904635 1032 MSR_IA32_TSC_ADJUST,
a3e06bbe 1033 MSR_IA32_TSCDEADLINE,
043405e1 1034 MSR_IA32_MISC_ENABLE,
908e75f3
AK		 1035 MSR_IA32_MCG_STATUS,
1036 MSR_IA32_MCG_CTL,
c45dcc71 1037 MSR_IA32_MCG_EXT_CTL,
64d60670 1038 MSR_IA32_SMBASE,
db2336a8
KH		 1039 MSR_PLATFORM_INFO,
1040 MSR_MISC_FEATURES_ENABLES,
043405e1
CO		 1041};
1042
62ef68bb
PB		 1043static unsigned num_emulated_msrs;
1044
384bb783 1045bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
15c4a640 1046{
b69e8cae 1047 if (efer & efer_reserved_bits)
384bb783 1048 return false;
15c4a640 1049
1b4d56b8 1050 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
384bb783 1051 return false;
1b2fd70c 1052
1b4d56b8 1053 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
384bb783 1054 return false;
d8017474 1055
384bb783
JK		 1056 return true;
1057}
1058EXPORT_SYMBOL_GPL(kvm_valid_efer);
1059
1060static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1061{
1062 u64 old_efer = vcpu->arch.efer;
1063
1064 if (!kvm_valid_efer(vcpu, efer))
1065 return 1;
1066
1067 if (is_paging(vcpu)
1068 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1069 return 1;
1070
15c4a640 1071 efer &= ~EFER_LMA;
f6801dff 1072 efer |= vcpu->arch.efer & EFER_LMA;
15c4a640 1073
a3d204e2
SY		 1074 kvm_x86_ops->set_efer(vcpu, efer);
1075
aad82703
SY		 1076 /* Update reserved bits */
1077 if ((efer ^ old_efer) & EFER_NX)
1078 kvm_mmu_reset_context(vcpu);
1079
b69e8cae 1080 return 0;
15c4a640
CO		 1081}
1082
f2b4b7dd
JR		 1083void kvm_enable_efer_bits(u64 mask)
1084{
1085 efer_reserved_bits &= ~mask;
1086}
1087EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1088
15c4a640
CO		 1089/*
1090 * Writes msr value into into the appropriate "register".
1091 * Returns 0 on success, non-0 otherwise.
1092 * Assumes vcpu_load() was already called.
1093 */
8fe8ab46 1094int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
15c4a640 1095{
854e8bb1
NA		 1096 switch (msr->index) {
1097 case MSR_FS_BASE:
1098 case MSR_GS_BASE:
1099 case MSR_KERNEL_GS_BASE:
1100 case MSR_CSTAR:
1101 case MSR_LSTAR:
fd8cb433 1102 if (is_noncanonical_address(msr->data, vcpu))
854e8bb1
NA		 1103 return 1;
1104 break;
1105 case MSR_IA32_SYSENTER_EIP:
1106 case MSR_IA32_SYSENTER_ESP:
1107 /*
1108 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1109 * non-canonical address is written on Intel but not on
1110 * AMD (which ignores the top 32-bits, because it does
1111 * not implement 64-bit SYSENTER).
1112 *
1113 * 64-bit code should hence be able to write a non-canonical
1114 * value on AMD. Making the address canonical ensures that
1115 * vmentry does not fail on Intel after writing a non-canonical
1116 * value, and that something deterministic happens if the guest
1117 * invokes 64-bit SYSENTER.
1118 */
fd8cb433 1119 msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
854e8bb1 1120 }
8fe8ab46 1121 return kvm_x86_ops->set_msr(vcpu, msr);
15c4a640 1122}
854e8bb1 1123EXPORT_SYMBOL_GPL(kvm_set_msr);
15c4a640 1124
313a3dc7
CO		 1125/*
1126 * Adapt set_msr() to msr_io()'s calling convention
1127 */
609e36d3
PB		 1128static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1129{
1130 struct msr_data msr;
1131 int r;
1132
1133 msr.index = index;
1134 msr.host_initiated = true;
1135 r = kvm_get_msr(vcpu, &msr);
1136 if (r)
1137 return r;
1138
1139 *data = msr.data;
1140 return 0;
1141}
1142
313a3dc7
CO		 1143static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1144{
8fe8ab46
WA		 1145 struct msr_data msr;
1146
1147 msr.data = *data;
1148 msr.index = index;
1149 msr.host_initiated = true;
1150 return kvm_set_msr(vcpu, &msr);
313a3dc7
CO		 1151}
1152
16e8d74d
MT		 1153#ifdef CONFIG_X86_64
1154struct pvclock_gtod_data {
1155 seqcount_t seq;
1156
1157 struct { /* extract of a clocksource struct */
1158 int vclock_mode;
a5a1d1c2
TG		 1159 u64 cycle_last;
1160 u64 mask;
16e8d74d
MT		 1161 u32 mult;
1162 u32 shift;
1163 } clock;
1164
cbcf2dd3
TG		 1165 u64 boot_ns;
1166 u64 nsec_base;
55dd00a7 1167 u64 wall_time_sec;
16e8d74d
MT		 1168};
1169
1170static struct pvclock_gtod_data pvclock_gtod_data;
1171
1172static void update_pvclock_gtod(struct timekeeper *tk)
1173{
1174 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
cbcf2dd3
TG		 1175 u64 boot_ns;
1176
876e7881 1177 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
16e8d74d
MT		 1178
1179 write_seqcount_begin(&vdata->seq);
1180
1181 /* copy pvclock gtod data */
876e7881
PZ		 1182 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1183 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1184 vdata->clock.mask = tk->tkr_mono.mask;
1185 vdata->clock.mult = tk->tkr_mono.mult;
1186 vdata->clock.shift = tk->tkr_mono.shift;
16e8d74d 1187
cbcf2dd3 1188 vdata->boot_ns = boot_ns;
876e7881 1189 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
16e8d74d 1190
55dd00a7
MT		 1191 vdata->wall_time_sec = tk->xtime_sec;
1192
16e8d74d
MT		 1193 write_seqcount_end(&vdata->seq);
1194}
1195#endif
1196
bab5bb39
NK		 1197void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1198{
1199 /*
1200 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1201 * vcpu_enter_guest. This function is only called from
1202 * the physical CPU that is running vcpu.
1203 */
1204 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1205}
16e8d74d 1206
18068523
GOC		 1207static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1208{
9ed3c444
AK		 1209 int version;
1210 int r;
50d0a0f9 1211 struct pvclock_wall_clock wc;
87aeb54f 1212 struct timespec64 boot;
18068523
GOC		 1213
1214 if (!wall_clock)
1215 return;
1216
9ed3c444
AK		 1217 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1218 if (r)
1219 return;
1220
1221 if (version & 1)
1222 ++version; /* first time write, random junk */
1223
1224 ++version;
18068523 1225
1dab1345
NK		 1226 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1227 return;
18068523 1228
50d0a0f9
GH		 1229 /*
1230 * The guest calculates current wall clock time by adding
34c238a1 1231 * system time (updated by kvm_guest_time_update below) to the
50d0a0f9
GH		 1232 * wall clock specified here. guest system time equals host
1233 * system time for us, thus we must fill in host boot time here.
1234 */
87aeb54f 1235 getboottime64(&boot);
50d0a0f9 1236
4b648665 1237 if (kvm->arch.kvmclock_offset) {
87aeb54f
AB		 1238 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1239 boot = timespec64_sub(boot, ts);
4b648665 1240 }
87aeb54f 1241 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
50d0a0f9
GH		 1242 wc.nsec = boot.tv_nsec;
1243 wc.version = version;
18068523
GOC		 1244
1245 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1246
1247 version++;
1248 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
18068523
GOC		 1249}
1250
50d0a0f9
GH		 1251static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1252{
b51012de
PB		 1253 do_shl32_div32(dividend, divisor);
1254 return dividend;
50d0a0f9
GH		 1255}
1256
3ae13faa 1257static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
5f4e3f88 1258 s8 *pshift, u32 *pmultiplier)
50d0a0f9 1259{
5f4e3f88 1260 uint64_t scaled64;
50d0a0f9
GH		 1261 int32_t shift = 0;
1262 uint64_t tps64;
1263 uint32_t tps32;
1264
3ae13faa
PB		 1265 tps64 = base_hz;
1266 scaled64 = scaled_hz;
50933623 1267 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
50d0a0f9
GH		 1268 tps64 >>= 1;
1269 shift--;
1270 }
1271
1272 tps32 = (uint32_t)tps64;
50933623
JK		 1273 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1274 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
5f4e3f88
ZA		 1275 scaled64 >>= 1;
1276 else
1277 tps32 <<= 1;
50d0a0f9
GH		 1278 shift++;
1279 }
1280
5f4e3f88
ZA		 1281 *pshift = shift;
1282 *pmultiplier = div_frac(scaled64, tps32);
50d0a0f9 1283
3ae13faa
PB		 1284 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1285 __func__, base_hz, scaled_hz, shift, *pmultiplier);
50d0a0f9
GH		 1286}
1287
d828199e 1288#ifdef CONFIG_X86_64
16e8d74d 1289static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
d828199e 1290#endif
16e8d74d 1291
c8076604 1292static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
69b0049a 1293static unsigned long max_tsc_khz;
c8076604 1294
cc578287 1295static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1e993611 1296{
cc578287
ZA		 1297 u64 v = (u64)khz * (1000000 + ppm);
1298 do_div(v, 1000000);
1299 return v;
1e993611
JR		 1300}
1301
381d585c
HZ		 1302static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1303{
1304 u64 ratio;
1305
1306 /* Guest TSC same frequency as host TSC? */
1307 if (!scale) {
1308 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1309 return 0;
1310 }
1311
1312 /* TSC scaling supported? */
1313 if (!kvm_has_tsc_control) {
1314 if (user_tsc_khz > tsc_khz) {
1315 vcpu->arch.tsc_catchup = 1;
1316 vcpu->arch.tsc_always_catchup = 1;
1317 return 0;
1318 } else {
1319 WARN(1, "user requested TSC rate below hardware speed\n");
1320 return -1;
1321 }
1322 }
1323
1324 /* TSC scaling required - calculate ratio */
1325 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1326 user_tsc_khz, tsc_khz);
1327
1328 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1329 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1330 user_tsc_khz);
1331 return -1;
1332 }
1333
1334 vcpu->arch.tsc_scaling_ratio = ratio;
1335 return 0;
1336}
1337
4941b8cb 1338static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
759379dd 1339{
cc578287
ZA		 1340 u32 thresh_lo, thresh_hi;
1341 int use_scaling = 0;
217fc9cf 1342
03ba32ca 1343 /* tsc_khz can be zero if TSC calibration fails */
4941b8cb 1344 if (user_tsc_khz == 0) {
ad721883
HZ		 1345 /* set tsc_scaling_ratio to a safe value */
1346 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
381d585c 1347 return -1;
ad721883 1348 }
03ba32ca 1349
c285545f 1350 /* Compute a scale to convert nanoseconds in TSC cycles */
3ae13faa 1351 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
cc578287
ZA		 1352 &vcpu->arch.virtual_tsc_shift,
1353 &vcpu->arch.virtual_tsc_mult);
4941b8cb 1354 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
cc578287
ZA		 1355
1356 /*
1357 * Compute the variation in TSC rate which is acceptable
1358 * within the range of tolerance and decide if the
1359 * rate being applied is within that bounds of the hardware
1360 * rate. If so, no scaling or compensation need be done.
1361 */
1362 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1363 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
4941b8cb
PB		 1364 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1365 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
cc578287
ZA		 1366 use_scaling = 1;
1367 }
4941b8cb 1368 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
c285545f
ZA		 1369}
1370
1371static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1372{
e26101b1 1373 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
cc578287
ZA		 1374 vcpu->arch.virtual_tsc_mult,
1375 vcpu->arch.virtual_tsc_shift);
e26101b1 1376 tsc += vcpu->arch.this_tsc_write;
c285545f
ZA		 1377 return tsc;
1378}
1379
69b0049a 1380static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
b48aa97e
MT		 1381{
1382#ifdef CONFIG_X86_64
1383 bool vcpus_matched;
b48aa97e
MT		 1384 struct kvm_arch *ka = &vcpu->kvm->arch;
1385 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1386
1387 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1388 atomic_read(&vcpu->kvm->online_vcpus));
1389
7f187922
MT		 1390 /*
1391 * Once the masterclock is enabled, always perform request in
1392 * order to update it.
1393 *
1394 * In order to enable masterclock, the host clocksource must be TSC
1395 * and the vcpus need to have matched TSCs. When that happens,
1396 * perform request to enable masterclock.
1397 */
1398 if (ka->use_master_clock ||
1399 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
b48aa97e
MT		 1400 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1401
1402 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1403 atomic_read(&vcpu->kvm->online_vcpus),
1404 ka->use_master_clock, gtod->clock.vclock_mode);
1405#endif
1406}
1407
ba904635
WA		 1408static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1409{
3e3f5026 1410 u64 curr_offset = vcpu->arch.tsc_offset;
ba904635
WA		 1411 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1412}
1413
35181e86
HZ		 1414/*
1415 * Multiply tsc by a fixed point number represented by ratio.
1416 *
1417 * The most significant 64-N bits (mult) of ratio represent the
1418 * integral part of the fixed point number; the remaining N bits
1419 * (frac) represent the fractional part, ie. ratio represents a fixed
1420 * point number (mult + frac * 2^(-N)).
1421 *
1422 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1423 */
1424static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1425{
1426 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1427}
1428
1429u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1430{
1431 u64 _tsc = tsc;
1432 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1433
1434 if (ratio != kvm_default_tsc_scaling_ratio)
1435 _tsc = __scale_tsc(ratio, tsc);
1436
1437 return _tsc;
1438}
1439EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1440
07c1419a
HZ		 1441static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1442{
1443 u64 tsc;
1444
1445 tsc = kvm_scale_tsc(vcpu, rdtsc());
1446
1447 return target_tsc - tsc;
1448}
1449
4ba76538
HZ		 1450u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1451{
ea26e4ec 1452 return vcpu->arch.tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
4ba76538
HZ		 1453}
1454EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1455
a545ab6a
LC		 1456static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1457{
1458 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1459 vcpu->arch.tsc_offset = offset;
1460}
1461
8fe8ab46 1462void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
99e3e30a
ZA		 1463{
1464 struct kvm *kvm = vcpu->kvm;
f38e098f 1465 u64 offset, ns, elapsed;
99e3e30a 1466 unsigned long flags;
b48aa97e 1467 bool matched;
0d3da0d2 1468 bool already_matched;
8fe8ab46 1469 u64 data = msr->data;
c5e8ec8e 1470 bool synchronizing = false;
99e3e30a 1471
038f8c11 1472 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
07c1419a 1473 offset = kvm_compute_tsc_offset(vcpu, data);
108b249c 1474 ns = ktime_get_boot_ns();
f38e098f 1475 elapsed = ns - kvm->arch.last_tsc_nsec;
5d3cb0f6 1476
03ba32ca 1477 if (vcpu->arch.virtual_tsc_khz) {
bd8fab39
DP		 1478 if (data == 0 && msr->host_initiated) {
1479 /*
1480 * detection of vcpu initialization -- need to sync
1481 * with other vCPUs. This particularly helps to keep
1482 * kvm_clock stable after CPU hotplug
1483 */
1484 synchronizing = true;
1485 } else {
1486 u64 tsc_exp = kvm->arch.last_tsc_write +
1487 nsec_to_cycles(vcpu, elapsed);
1488 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1489 /*
1490 * Special case: TSC write with a small delta (1 second)
1491 * of virtual cycle time against real time is
1492 * interpreted as an attempt to synchronize the CPU.
1493 */
1494 synchronizing = data < tsc_exp + tsc_hz &&
1495 data + tsc_hz > tsc_exp;
1496 }
c5e8ec8e 1497 }
f38e098f
ZA		 1498
1499 /*
5d3cb0f6
ZA		 1500 * For a reliable TSC, we can match TSC offsets, and for an unstable
1501 * TSC, we add elapsed time in this computation. We could let the
1502 * compensation code attempt to catch up if we fall behind, but
1503 * it's better to try to match offsets from the beginning.
1504 */
c5e8ec8e 1505 if (synchronizing &&
5d3cb0f6 1506 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
f38e098f 1507 if (!check_tsc_unstable()) {
e26101b1 1508 offset = kvm->arch.cur_tsc_offset;
f38e098f
ZA		 1509 pr_debug("kvm: matched tsc offset for %llu\n", data);
1510 } else {
857e4099 1511 u64 delta = nsec_to_cycles(vcpu, elapsed);
5d3cb0f6 1512 data += delta;
07c1419a 1513 offset = kvm_compute_tsc_offset(vcpu, data);
759379dd 1514 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
f38e098f 1515 }
b48aa97e 1516 matched = true;
0d3da0d2 1517 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
e26101b1
ZA		 1518 } else {
1519 /*
1520 * We split periods of matched TSC writes into generations.
1521 * For each generation, we track the original measured
1522 * nanosecond time, offset, and write, so if TSCs are in
1523 * sync, we can match exact offset, and if not, we can match
4a969980 1524 * exact software computation in compute_guest_tsc()
e26101b1
ZA		 1525 *
1526 * These values are tracked in kvm->arch.cur_xxx variables.
1527 */
1528 kvm->arch.cur_tsc_generation++;
1529 kvm->arch.cur_tsc_nsec = ns;
1530 kvm->arch.cur_tsc_write = data;
1531 kvm->arch.cur_tsc_offset = offset;
b48aa97e 1532 matched = false;
0d3da0d2 1533 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
e26101b1 1534 kvm->arch.cur_tsc_generation, data);
f38e098f 1535 }
e26101b1
ZA		 1536
1537 /*
1538 * We also track th most recent recorded KHZ, write and time to
1539 * allow the matching interval to be extended at each write.
1540 */
f38e098f
ZA		 1541 kvm->arch.last_tsc_nsec = ns;
1542 kvm->arch.last_tsc_write = data;
5d3cb0f6 1543 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
99e3e30a 1544
b183aa58 1545 vcpu->arch.last_guest_tsc = data;
e26101b1
ZA		 1546
1547 /* Keep track of which generation this VCPU has synchronized to */
1548 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1549 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1550 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1551
d6321d49 1552 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
ba904635 1553 update_ia32_tsc_adjust_msr(vcpu, offset);
d6321d49 1554
a545ab6a 1555 kvm_vcpu_write_tsc_offset(vcpu, offset);
e26101b1 1556 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
b48aa97e
MT		 1557
1558 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
0d3da0d2 1559 if (!matched) {
b48aa97e 1560 kvm->arch.nr_vcpus_matched_tsc = 0;
0d3da0d2
TG		 1561 } else if (!already_matched) {
1562 kvm->arch.nr_vcpus_matched_tsc++;
1563 }
b48aa97e
MT		 1564
1565 kvm_track_tsc_matching(vcpu);
1566 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
99e3e30a 1567}
e26101b1 1568
99e3e30a
ZA		 1569EXPORT_SYMBOL_GPL(kvm_write_tsc);
1570
58ea6767
HZ		 1571static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1572 s64 adjustment)
1573{
ea26e4ec 1574 kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.tsc_offset + adjustment);
58ea6767
HZ		 1575}
1576
1577static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1578{
1579 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1580 WARN_ON(adjustment < 0);
1581 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
ea26e4ec 1582 adjust_tsc_offset_guest(vcpu, adjustment);
58ea6767
HZ		 1583}
1584
d828199e
MT		 1585#ifdef CONFIG_X86_64
1586
a5a1d1c2 1587static u64 read_tsc(void)
d828199e 1588{
a5a1d1c2 1589 u64 ret = (u64)rdtsc_ordered();
03b9730b 1590 u64 last = pvclock_gtod_data.clock.cycle_last;
d828199e
MT		 1591
1592 if (likely(ret >= last))
1593 return ret;
1594
1595 /*
1596 * GCC likes to generate cmov here, but this branch is extremely
6a6256f9 1597 * predictable (it's just a function of time and the likely is
d828199e
MT		 1598 * very likely) and there's a data dependence, so force GCC
1599 * to generate a branch instead. I don't barrier() because
1600 * we don't actually need a barrier, and if this function
1601 * ever gets inlined it will generate worse code.
1602 */
1603 asm volatile ("");
1604 return last;
1605}
1606
a5a1d1c2 1607static inline u64 vgettsc(u64 *cycle_now)
d828199e
MT		 1608{
1609 long v;
1610 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1611
1612 *cycle_now = read_tsc();
1613
1614 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1615 return v * gtod->clock.mult;
1616}
1617
a5a1d1c2 1618static int do_monotonic_boot(s64 *t, u64 *cycle_now)
d828199e 1619{
cbcf2dd3 1620 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
d828199e 1621 unsigned long seq;
d828199e 1622 int mode;
cbcf2dd3 1623 u64 ns;
d828199e 1624
d828199e
MT		 1625 do {
1626 seq = read_seqcount_begin(&gtod->seq);
1627 mode = gtod->clock.vclock_mode;
cbcf2dd3 1628 ns = gtod->nsec_base;
d828199e
MT		 1629 ns += vgettsc(cycle_now);
1630 ns >>= gtod->clock.shift;
cbcf2dd3 1631 ns += gtod->boot_ns;
d828199e 1632 } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
cbcf2dd3 1633 *t = ns;
d828199e
MT		 1634
1635 return mode;
1636}
1637
55dd00a7
MT		 1638static int do_realtime(struct timespec *ts, u64 *cycle_now)
1639{
1640 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1641 unsigned long seq;
1642 int mode;
1643 u64 ns;
1644
1645 do {
1646 seq = read_seqcount_begin(&gtod->seq);
1647 mode = gtod->clock.vclock_mode;
1648 ts->tv_sec = gtod->wall_time_sec;
1649 ns = gtod->nsec_base;
1650 ns += vgettsc(cycle_now);
1651 ns >>= gtod->clock.shift;
1652 } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1653
1654 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1655 ts->tv_nsec = ns;
1656
1657 return mode;
1658}
1659
d828199e 1660/* returns true if host is using tsc clocksource */
a5a1d1c2 1661static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *cycle_now)
d828199e 1662{
d828199e
MT		 1663 /* checked again under seqlock below */
1664 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1665 return false;
1666
cbcf2dd3 1667 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
d828199e 1668}
55dd00a7
MT		 1669
1670/* returns true if host is using tsc clocksource */
1671static bool kvm_get_walltime_and_clockread(struct timespec *ts,
1672 u64 *cycle_now)
1673{
1674 /* checked again under seqlock below */
1675 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1676 return false;
1677
1678 return do_realtime(ts, cycle_now) == VCLOCK_TSC;
1679}
d828199e
MT		 1680#endif
1681
1682/*
1683 *
b48aa97e
MT		 1684 * Assuming a stable TSC across physical CPUS, and a stable TSC
1685 * across virtual CPUs, the following condition is possible.
1686 * Each numbered line represents an event visible to both
d828199e
MT		 1687 * CPUs at the next numbered event.
1688 *
1689 * "timespecX" represents host monotonic time. "tscX" represents
1690 * RDTSC value.
1691 *
1692 * VCPU0 on CPU0 | VCPU1 on CPU1
1693 *
1694 * 1. read timespec0,tsc0
1695 * 2. | timespec1 = timespec0 + N
1696 * | tsc1 = tsc0 + M
1697 * 3. transition to guest | transition to guest
1698 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1699 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1700 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1701 *
1702 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1703 *
1704 * - ret0 < ret1
1705 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1706 * ...
1707 * - 0 < N - M => M < N
1708 *
1709 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1710 * always the case (the difference between two distinct xtime instances
1711 * might be smaller then the difference between corresponding TSC reads,
1712 * when updating guest vcpus pvclock areas).
1713 *
1714 * To avoid that problem, do not allow visibility of distinct
1715 * system_timestamp/tsc_timestamp values simultaneously: use a master
1716 * copy of host monotonic time values. Update that master copy
1717 * in lockstep.
1718 *
b48aa97e 1719 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
d828199e
MT		 1720 *
1721 */
1722
1723static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1724{
1725#ifdef CONFIG_X86_64
1726 struct kvm_arch *ka = &kvm->arch;
1727 int vclock_mode;
b48aa97e
MT		 1728 bool host_tsc_clocksource, vcpus_matched;
1729
1730 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1731 atomic_read(&kvm->online_vcpus));
d828199e
MT		 1732
1733 /*
1734 * If the host uses TSC clock, then passthrough TSC as stable
1735 * to the guest.
1736 */
b48aa97e 1737 host_tsc_clocksource = kvm_get_time_and_clockread(
d828199e
MT		 1738 &ka->master_kernel_ns,
1739 &ka->master_cycle_now);
1740
16a96021 1741 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
a826faf1 1742 && !ka->backwards_tsc_observed
54750f2c 1743 && !ka->boot_vcpu_runs_old_kvmclock;
b48aa97e 1744
d828199e
MT		 1745 if (ka->use_master_clock)
1746 atomic_set(&kvm_guest_has_master_clock, 1);
1747
1748 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
b48aa97e
MT		 1749 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1750 vcpus_matched);
d828199e
MT		 1751#endif
1752}
1753
2860c4b1
PB		 1754void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1755{
1756 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1757}
1758
2e762ff7
MT		 1759static void kvm_gen_update_masterclock(struct kvm *kvm)
1760{
1761#ifdef CONFIG_X86_64
1762 int i;
1763 struct kvm_vcpu *vcpu;
1764 struct kvm_arch *ka = &kvm->arch;
1765
1766 spin_lock(&ka->pvclock_gtod_sync_lock);
1767 kvm_make_mclock_inprogress_request(kvm);
1768 /* no guest entries from this point */
1769 pvclock_update_vm_gtod_copy(kvm);
1770
1771 kvm_for_each_vcpu(i, vcpu, kvm)
105b21bb 1772 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2e762ff7
MT		 1773
1774 /* guest entries allowed */
1775 kvm_for_each_vcpu(i, vcpu, kvm)
72875d8a 1776 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2e762ff7
MT		 1777
1778 spin_unlock(&ka->pvclock_gtod_sync_lock);
1779#endif
1780}
1781
e891a32e 1782u64 get_kvmclock_ns(struct kvm *kvm)
108b249c 1783{
108b249c 1784 struct kvm_arch *ka = &kvm->arch;
8b953440 1785 struct pvclock_vcpu_time_info hv_clock;
e2c2206a 1786 u64 ret;
108b249c 1787
8b953440
PB		 1788 spin_lock(&ka->pvclock_gtod_sync_lock);
1789 if (!ka->use_master_clock) {
1790 spin_unlock(&ka->pvclock_gtod_sync_lock);
1791 return ktime_get_boot_ns() + ka->kvmclock_offset;
108b249c
PB		 1792 }
1793
8b953440
PB		 1794 hv_clock.tsc_timestamp = ka->master_cycle_now;
1795 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
1796 spin_unlock(&ka->pvclock_gtod_sync_lock);
1797
e2c2206a
WL		 1798 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
1799 get_cpu();
1800
e70b57a6
WL		 1801 if (__this_cpu_read(cpu_tsc_khz)) {
1802 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
1803 &hv_clock.tsc_shift,
1804 &hv_clock.tsc_to_system_mul);
1805 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
1806 } else
1807 ret = ktime_get_boot_ns() + ka->kvmclock_offset;
e2c2206a
WL		 1808
1809 put_cpu();
1810
1811 return ret;
108b249c
PB		 1812}
1813
0d6dd2ff
PB		 1814static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
1815{
1816 struct kvm_vcpu_arch *vcpu = &v->arch;
1817 struct pvclock_vcpu_time_info guest_hv_clock;
1818
4e335d9e 1819 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
0d6dd2ff
PB		 1820 &guest_hv_clock, sizeof(guest_hv_clock))))
1821 return;
1822
1823 /* This VCPU is paused, but it's legal for a guest to read another
1824 * VCPU's kvmclock, so we really have to follow the specification where
1825 * it says that version is odd if data is being modified, and even after
1826 * it is consistent.
1827 *
1828 * Version field updates must be kept separate. This is because
1829 * kvm_write_guest_cached might use a "rep movs" instruction, and
1830 * writes within a string instruction are weakly ordered. So there
1831 * are three writes overall.
1832 *
1833 * As a small optimization, only write the version field in the first
1834 * and third write. The vcpu->pv_time cache is still valid, because the
1835 * version field is the first in the struct.
1836 */
1837 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1838
51c4b8bb
LA		 1839 if (guest_hv_clock.version & 1)
1840 ++guest_hv_clock.version; /* first time write, random junk */
1841
0d6dd2ff 1842 vcpu->hv_clock.version = guest_hv_clock.version + 1;
4e335d9e
PB		 1843 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1844 &vcpu->hv_clock,
1845 sizeof(vcpu->hv_clock.version));
0d6dd2ff
PB		 1846
1847 smp_wmb();
1848
1849 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1850 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1851
1852 if (vcpu->pvclock_set_guest_stopped_request) {
1853 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
1854 vcpu->pvclock_set_guest_stopped_request = false;
1855 }
1856
1857 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1858
4e335d9e
PB		 1859 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1860 &vcpu->hv_clock,
1861 sizeof(vcpu->hv_clock));
0d6dd2ff
PB		 1862
1863 smp_wmb();
1864
1865 vcpu->hv_clock.version++;
4e335d9e
PB		 1866 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1867 &vcpu->hv_clock,
1868 sizeof(vcpu->hv_clock.version));
0d6dd2ff
PB		 1869}
1870
34c238a1 1871static int kvm_guest_time_update(struct kvm_vcpu *v)
18068523 1872{
78db6a50 1873 unsigned long flags, tgt_tsc_khz;
18068523 1874 struct kvm_vcpu_arch *vcpu = &v->arch;
d828199e 1875 struct kvm_arch *ka = &v->kvm->arch;
f25e656d 1876 s64 kernel_ns;
d828199e 1877 u64 tsc_timestamp, host_tsc;
51d59c6b 1878 u8 pvclock_flags;
d828199e
MT		 1879 bool use_master_clock;
1880
1881 kernel_ns = 0;
1882 host_tsc = 0;
18068523 1883
d828199e
MT		 1884 /*
1885 * If the host uses TSC clock, then passthrough TSC as stable
1886 * to the guest.
1887 */
1888 spin_lock(&ka->pvclock_gtod_sync_lock);
1889 use_master_clock = ka->use_master_clock;
1890 if (use_master_clock) {
1891 host_tsc = ka->master_cycle_now;
1892 kernel_ns = ka->master_kernel_ns;
1893 }
1894 spin_unlock(&ka->pvclock_gtod_sync_lock);
c09664bb
MT		 1895
1896 /* Keep irq disabled to prevent changes to the clock */
1897 local_irq_save(flags);
78db6a50
PB		 1898 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1899 if (unlikely(tgt_tsc_khz == 0)) {
c09664bb
MT		 1900 local_irq_restore(flags);
1901 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1902 return 1;
1903 }
d828199e 1904 if (!use_master_clock) {
4ea1636b 1905 host_tsc = rdtsc();
108b249c 1906 kernel_ns = ktime_get_boot_ns();
d828199e
MT		 1907 }
1908
4ba76538 1909 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
d828199e 1910
c285545f
ZA		 1911 /*
1912 * We may have to catch up the TSC to match elapsed wall clock
1913 * time for two reasons, even if kvmclock is used.
1914 * 1) CPU could have been running below the maximum TSC rate
1915 * 2) Broken TSC compensation resets the base at each VCPU
1916 * entry to avoid unknown leaps of TSC even when running
1917 * again on the same CPU. This may cause apparent elapsed
1918 * time to disappear, and the guest to stand still or run
1919 * very slowly.
1920 */
1921 if (vcpu->tsc_catchup) {
1922 u64 tsc = compute_guest_tsc(v, kernel_ns);
1923 if (tsc > tsc_timestamp) {
f1e2b260 1924 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
c285545f
ZA		 1925 tsc_timestamp = tsc;
1926 }
50d0a0f9
GH		 1927 }
1928
18068523
GOC		 1929 local_irq_restore(flags);
1930
0d6dd2ff 1931 /* With all the info we got, fill in the values */
18068523 1932
78db6a50
PB		 1933 if (kvm_has_tsc_control)
1934 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
1935
1936 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
3ae13faa 1937 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
5f4e3f88
ZA		 1938 &vcpu->hv_clock.tsc_shift,
1939 &vcpu->hv_clock.tsc_to_system_mul);
78db6a50 1940 vcpu->hw_tsc_khz = tgt_tsc_khz;
8cfdc000
ZA		 1941 }
1942
1d5f066e 1943 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
759379dd 1944 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
28e4639a 1945 vcpu->last_guest_tsc = tsc_timestamp;
51d59c6b 1946
d828199e 1947 /* If the host uses TSC clocksource, then it is stable */
0d6dd2ff 1948 pvclock_flags = 0;
d828199e
MT		 1949 if (use_master_clock)
1950 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1951
78c0337a
MT		 1952 vcpu->hv_clock.flags = pvclock_flags;
1953
095cf55d
PB		 1954 if (vcpu->pv_time_enabled)
1955 kvm_setup_pvclock_page(v);
1956 if (v == kvm_get_vcpu(v->kvm, 0))
1957 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
8cfdc000 1958 return 0;
c8076604
GH		 1959}
1960
0061d53d
MT		 1961/*
1962 * kvmclock updates which are isolated to a given vcpu, such as
1963 * vcpu->cpu migration, should not allow system_timestamp from
1964 * the rest of the vcpus to remain static. Otherwise ntp frequency
1965 * correction applies to one vcpu's system_timestamp but not
1966 * the others.
1967 *
1968 * So in those cases, request a kvmclock update for all vcpus.
7e44e449
AJ		 1969 * We need to rate-limit these requests though, as they can
1970 * considerably slow guests that have a large number of vcpus.
1971 * The time for a remote vcpu to update its kvmclock is bound
1972 * by the delay we use to rate-limit the updates.
0061d53d
MT		 1973 */
1974
7e44e449
AJ		 1975#define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1976
1977static void kvmclock_update_fn(struct work_struct *work)
0061d53d
MT		 1978{
1979 int i;
7e44e449
AJ		 1980 struct delayed_work *dwork = to_delayed_work(work);
1981 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1982 kvmclock_update_work);
1983 struct kvm *kvm = container_of(ka, struct kvm, arch);
0061d53d
MT		 1984 struct kvm_vcpu *vcpu;
1985
1986 kvm_for_each_vcpu(i, vcpu, kvm) {
105b21bb 1987 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
0061d53d
MT		 1988 kvm_vcpu_kick(vcpu);
1989 }
1990}
1991
7e44e449
AJ		 1992static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1993{
1994 struct kvm *kvm = v->kvm;
1995
105b21bb 1996 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
7e44e449
AJ		 1997 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1998 KVMCLOCK_UPDATE_DELAY);
1999}
2000
332967a3
AJ		 2001#define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2002
2003static void kvmclock_sync_fn(struct work_struct *work)
2004{
2005 struct delayed_work *dwork = to_delayed_work(work);
2006 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2007 kvmclock_sync_work);
2008 struct kvm *kvm = container_of(ka, struct kvm, arch);
2009
630994b3
MT		 2010 if (!kvmclock_periodic_sync)
2011 return;
2012
332967a3
AJ		 2013 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2014 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2015 KVMCLOCK_SYNC_PERIOD);
2016}
2017
9ffd986c 2018static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
15c4a640 2019{
890ca9ae
HY		 2020 u64 mcg_cap = vcpu->arch.mcg_cap;
2021 unsigned bank_num = mcg_cap & 0xff;
9ffd986c
WL		 2022 u32 msr = msr_info->index;
2023 u64 data = msr_info->data;
890ca9ae 2024
15c4a640 2025 switch (msr) {
15c4a640 2026 case MSR_IA32_MCG_STATUS:
890ca9ae 2027 vcpu->arch.mcg_status = data;
15c4a640 2028 break;
c7ac679c 2029 case MSR_IA32_MCG_CTL:
890ca9ae
HY		 2030 if (!(mcg_cap & MCG_CTL_P))
2031 return 1;
2032 if (data != 0 && data != ~(u64)0)
2033 return -1;
2034 vcpu->arch.mcg_ctl = data;
2035 break;
2036 default:
2037 if (msr >= MSR_IA32_MC0_CTL &&
81760dcc 2038 msr < MSR_IA32_MCx_CTL(bank_num)) {
890ca9ae 2039 u32 offset = msr - MSR_IA32_MC0_CTL;
114be429
AP		 2040 /* only 0 or all 1s can be written to IA32_MCi_CTL
2041 * some Linux kernels though clear bit 10 in bank 4 to
2042 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2043 * this to avoid an uncatched #GP in the guest
2044 */
890ca9ae 2045 if ((offset & 0x3) == 0 &&
114be429 2046 data != 0 && (data | (1 << 10)) != ~(u64)0)
890ca9ae 2047 return -1;
9ffd986c
WL		 2048 if (!msr_info->host_initiated &&
2049 (offset & 0x3) == 1 && data != 0)
2050 return -1;
890ca9ae
HY		 2051 vcpu->arch.mce_banks[offset] = data;
2052 break;
2053 }
2054 return 1;
2055 }
2056 return 0;
2057}
2058
ffde22ac
ES		 2059static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2060{
2061 struct kvm *kvm = vcpu->kvm;
2062 int lm = is_long_mode(vcpu);
2063 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2064 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2065 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2066 : kvm->arch.xen_hvm_config.blob_size_32;
2067 u32 page_num = data & ~PAGE_MASK;
2068 u64 page_addr = data & PAGE_MASK;
2069 u8 *page;
2070 int r;
2071
2072 r = -E2BIG;
2073 if (page_num >= blob_size)
2074 goto out;
2075 r = -ENOMEM;
ff5c2c03
SL		 2076 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2077 if (IS_ERR(page)) {
2078 r = PTR_ERR(page);
ffde22ac 2079 goto out;
ff5c2c03 2080 }
54bf36aa 2081 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
ffde22ac
ES		 2082 goto out_free;
2083 r = 0;
2084out_free:
2085 kfree(page);
2086out:
2087 return r;
2088}
2089
344d9588
GN		 2090static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2091{
2092 gpa_t gpa = data & ~0x3f;
2093
52a5c155
WL		 2094 /* Bits 3:5 are reserved, Should be zero */
2095 if (data & 0x38)
344d9588
GN		 2096 return 1;
2097
2098 vcpu->arch.apf.msr_val = data;
2099
2100 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2101 kvm_clear_async_pf_completion_queue(vcpu);
2102 kvm_async_pf_hash_reset(vcpu);
2103 return 0;
2104 }
2105
4e335d9e 2106 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
8f964525 2107 sizeof(u32)))
344d9588
GN		 2108 return 1;
2109
6adba527 2110 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
52a5c155 2111 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
344d9588
GN		 2112 kvm_async_pf_wakeup_all(vcpu);
2113 return 0;
2114}
2115
12f9a48f
GC		 2116static void kvmclock_reset(struct kvm_vcpu *vcpu)
2117{
0b79459b 2118 vcpu->arch.pv_time_enabled = false;
12f9a48f
GC		 2119}
2120
c9aaa895
GC		 2121static void record_steal_time(struct kvm_vcpu *vcpu)
2122{
2123 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2124 return;
2125
4e335d9e 2126 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
c9aaa895
GC		 2127 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2128 return;
2129
0b9f6c46
PX		 2130 vcpu->arch.st.steal.preempted = 0;
2131
35f3fae1
WL		 2132 if (vcpu->arch.st.steal.version & 1)
2133 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2134
2135 vcpu->arch.st.steal.version += 1;
2136
4e335d9e 2137 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
35f3fae1
WL		 2138 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2139
2140 smp_wmb();
2141
c54cdf14
LC		 2142 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2143 vcpu->arch.st.last_steal;
2144 vcpu->arch.st.last_steal = current->sched_info.run_delay;
35f3fae1 2145
4e335d9e 2146 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
35f3fae1
WL		 2147 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2148
2149 smp_wmb();
2150
2151 vcpu->arch.st.steal.version += 1;
c9aaa895 2152
4e335d9e 2153 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
c9aaa895
GC		 2154 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2155}
2156
8fe8ab46 2157int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
15c4a640 2158{
5753785f 2159 bool pr = false;
8fe8ab46
WA		 2160 u32 msr = msr_info->index;
2161 u64 data = msr_info->data;
5753785f 2162
15c4a640 2163 switch (msr) {
2e32b719
BP		 2164 case MSR_AMD64_NB_CFG:
2165 case MSR_IA32_UCODE_REV:
2166 case MSR_IA32_UCODE_WRITE:
2167 case MSR_VM_HSAVE_PA:
2168 case MSR_AMD64_PATCH_LOADER:
2169 case MSR_AMD64_BU_CFG2:
405a353a 2170 case MSR_AMD64_DC_CFG:
2e32b719
BP		 2171 break;
2172
15c4a640 2173 case MSR_EFER:
b69e8cae 2174 return set_efer(vcpu, data);
8f1589d9
AP		 2175 case MSR_K7_HWCR:
2176 data &= ~(u64)0x40; /* ignore flush filter disable */
82494028 2177 data &= ~(u64)0x100; /* ignore ignne emulation enable */
a223c313 2178 data &= ~(u64)0x8; /* ignore TLB cache disable */
22d48b2d 2179 data &= ~(u64)0x40000; /* ignore Mc status write enable */
8f1589d9 2180 if (data != 0) {
a737f256
CD		 2181 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2182 data);
8f1589d9
AP		 2183 return 1;
2184 }
15c4a640 2185 break;
f7c6d140
AP		 2186 case MSR_FAM10H_MMIO_CONF_BASE:
2187 if (data != 0) {
a737f256
CD		 2188 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2189 "0x%llx\n", data);
f7c6d140
AP		 2190 return 1;
2191 }
15c4a640 2192 break;
b5e2fec0
AG		 2193 case MSR_IA32_DEBUGCTLMSR:
2194 if (!data) {
2195 /* We support the non-activated case already */
2196 break;
2197 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2198 /* Values other than LBR and BTF are vendor-specific,
2199 thus reserved and should throw a #GP */
2200 return 1;
2201 }
a737f256
CD		 2202 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2203 __func__, data);
b5e2fec0 2204 break;
9ba075a6 2205 case 0x200 ... 0x2ff:
ff53604b 2206 return kvm_mtrr_set_msr(vcpu, msr, data);
15c4a640 2207 case MSR_IA32_APICBASE:
58cb628d 2208 return kvm_set_apic_base(vcpu, msr_info);
0105d1a5
GN		 2209 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2210 return kvm_x2apic_msr_write(vcpu, msr, data);
a3e06bbe
LJ		 2211 case MSR_IA32_TSCDEADLINE:
2212 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2213 break;
ba904635 2214 case MSR_IA32_TSC_ADJUST:
d6321d49 2215 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
ba904635 2216 if (!msr_info->host_initiated) {
d913b904 2217 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
d7add054 2218 adjust_tsc_offset_guest(vcpu, adj);
ba904635
WA		 2219 }
2220 vcpu->arch.ia32_tsc_adjust_msr = data;
2221 }
2222 break;
15c4a640 2223 case MSR_IA32_MISC_ENABLE:
ad312c7c 2224 vcpu->arch.ia32_misc_enable_msr = data;
15c4a640 2225 break;
64d60670
PB		 2226 case MSR_IA32_SMBASE:
2227 if (!msr_info->host_initiated)
2228 return 1;
2229 vcpu->arch.smbase = data;
2230 break;
11c6bffa 2231 case MSR_KVM_WALL_CLOCK_NEW:
18068523
GOC		 2232 case MSR_KVM_WALL_CLOCK:
2233 vcpu->kvm->arch.wall_clock = data;
2234 kvm_write_wall_clock(vcpu->kvm, data);
2235 break;
11c6bffa 2236 case MSR_KVM_SYSTEM_TIME_NEW:
18068523 2237 case MSR_KVM_SYSTEM_TIME: {
54750f2c
MT		 2238 struct kvm_arch *ka = &vcpu->kvm->arch;
2239
12f9a48f 2240 kvmclock_reset(vcpu);
18068523 2241
54750f2c
MT		 2242 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2243 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2244
2245 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
1bd2009e 2246 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
54750f2c
MT		 2247
2248 ka->boot_vcpu_runs_old_kvmclock = tmp;
2249 }
2250
18068523 2251 vcpu->arch.time = data;
0061d53d 2252 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
18068523
GOC		 2253
2254 /* we verify if the enable bit is set... */
2255 if (!(data & 1))
2256 break;
2257
4e335d9e 2258 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
8f964525
AH		 2259 &vcpu->arch.pv_time, data & ~1ULL,
2260 sizeof(struct pvclock_vcpu_time_info)))
0b79459b
AH		 2261 vcpu->arch.pv_time_enabled = false;
2262 else
2263 vcpu->arch.pv_time_enabled = true;
32cad84f 2264
18068523
GOC		 2265 break;
2266 }
344d9588
GN		 2267 case MSR_KVM_ASYNC_PF_EN:
2268 if (kvm_pv_enable_async_pf(vcpu, data))
2269 return 1;
2270 break;
c9aaa895
GC		 2271 case MSR_KVM_STEAL_TIME:
2272
2273 if (unlikely(!sched_info_on()))
2274 return 1;
2275
2276 if (data & KVM_STEAL_RESERVED_MASK)
2277 return 1;
2278
4e335d9e 2279 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
8f964525
AH		 2280 data & KVM_STEAL_VALID_BITS,
2281 sizeof(struct kvm_steal_time)))
c9aaa895
GC		 2282 return 1;
2283
2284 vcpu->arch.st.msr_val = data;
2285
2286 if (!(data & KVM_MSR_ENABLED))
2287 break;
2288
c9aaa895
GC		 2289 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2290
2291 break;
ae7a2a3f
MT		 2292 case MSR_KVM_PV_EOI_EN:
2293 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2294 return 1;
2295 break;
c9aaa895 2296
890ca9ae
HY		 2297 case MSR_IA32_MCG_CTL:
2298 case MSR_IA32_MCG_STATUS:
81760dcc 2299 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
9ffd986c 2300 return set_msr_mce(vcpu, msr_info);
71db6023 2301
6912ac32
WH		 2302 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2303 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2304 pr = true; /* fall through */
2305 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2306 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
c6702c9d 2307 if (kvm_pmu_is_valid_msr(vcpu, msr))
afd80d85 2308 return kvm_pmu_set_msr(vcpu, msr_info);
5753785f
GN		 2309
2310 if (pr || data != 0)
a737f256
CD		 2311 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2312 "0x%x data 0x%llx\n", msr, data);
5753785f 2313 break;
84e0cefa
JS		 2314 case MSR_K7_CLK_CTL:
2315 /*
2316 * Ignore all writes to this no longer documented MSR.
2317 * Writes are only relevant for old K7 processors,
2318 * all pre-dating SVM, but a recommended workaround from
4a969980 2319 * AMD for these chips. It is possible to specify the
84e0cefa
JS		 2320 * affected processor models on the command line, hence
2321 * the need to ignore the workaround.
2322 */
2323 break;
55cd8e5a 2324 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
e7d9513b
AS		 2325 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2326 case HV_X64_MSR_CRASH_CTL:
1f4b34f8 2327 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
e7d9513b
AS		 2328 return kvm_hv_set_msr_common(vcpu, msr, data,
2329 msr_info->host_initiated);
91c9c3ed 2330 case MSR_IA32_BBL_CR_CTL3:
2331 /* Drop writes to this legacy MSR -- see rdmsr
2332 * counterpart for further detail.
2333 */
fab0aa3b
EM		 2334 if (report_ignored_msrs)
2335 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2336 msr, data);
91c9c3ed 2337 break;
2b036c6b 2338 case MSR_AMD64_OSVW_ID_LENGTH:
d6321d49 2339 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2b036c6b
BO		 2340 return 1;
2341 vcpu->arch.osvw.length = data;
2342 break;
2343 case MSR_AMD64_OSVW_STATUS:
d6321d49 2344 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2b036c6b
BO		 2345 return 1;
2346 vcpu->arch.osvw.status = data;
2347 break;
db2336a8
KH		 2348 case MSR_PLATFORM_INFO:
2349 if (!msr_info->host_initiated ||
2350 data & ~MSR_PLATFORM_INFO_CPUID_FAULT ||
2351 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2352 cpuid_fault_enabled(vcpu)))
2353 return 1;
2354 vcpu->arch.msr_platform_info = data;
2355 break;
2356 case MSR_MISC_FEATURES_ENABLES:
2357 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2358 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2359 !supports_cpuid_fault(vcpu)))
2360 return 1;
2361 vcpu->arch.msr_misc_features_enables = data;
2362 break;
15c4a640 2363 default:
ffde22ac
ES		 2364 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2365 return xen_hvm_config(vcpu, data);
c6702c9d 2366 if (kvm_pmu_is_valid_msr(vcpu, msr))
afd80d85 2367 return kvm_pmu_set_msr(vcpu, msr_info);
ed85c068 2368 if (!ignore_msrs) {
ae0f5499 2369 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
a737f256 2370 msr, data);
ed85c068
AP		 2371 return 1;
2372 } else {
fab0aa3b
EM		 2373 if (report_ignored_msrs)
2374 vcpu_unimpl(vcpu,
2375 "ignored wrmsr: 0x%x data 0x%llx\n",
2376 msr, data);
ed85c068
AP		 2377 break;
2378 }
15c4a640
CO		 2379 }
2380 return 0;
2381}
2382EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2383
2384
2385/*
2386 * Reads an msr value (of 'msr_index') into 'pdata'.
2387 * Returns 0 on success, non-0 otherwise.
2388 * Assumes vcpu_load() was already called.
2389 */
609e36d3 2390int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
15c4a640 2391{
609e36d3 2392 return kvm_x86_ops->get_msr(vcpu, msr);
15c4a640 2393}
ff651cb6 2394EXPORT_SYMBOL_GPL(kvm_get_msr);
15c4a640 2395
890ca9ae 2396static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
15c4a640
CO		 2397{
2398 u64 data;
890ca9ae
HY		 2399 u64 mcg_cap = vcpu->arch.mcg_cap;
2400 unsigned bank_num = mcg_cap & 0xff;
15c4a640
CO		 2401
2402 switch (msr) {
15c4a640
CO		 2403 case MSR_IA32_P5_MC_ADDR:
2404 case MSR_IA32_P5_MC_TYPE:
890ca9ae
HY		 2405 data = 0;
2406 break;
15c4a640 2407 case MSR_IA32_MCG_CAP:
890ca9ae
HY		 2408 data = vcpu->arch.mcg_cap;
2409 break;
c7ac679c 2410 case MSR_IA32_MCG_CTL:
890ca9ae
HY		 2411 if (!(mcg_cap & MCG_CTL_P))
2412 return 1;
2413 data = vcpu->arch.mcg_ctl;
2414 break;
2415 case MSR_IA32_MCG_STATUS:
2416 data = vcpu->arch.mcg_status;
2417 break;
2418 default:
2419 if (msr >= MSR_IA32_MC0_CTL &&
81760dcc 2420 msr < MSR_IA32_MCx_CTL(bank_num)) {
890ca9ae
HY		 2421 u32 offset = msr - MSR_IA32_MC0_CTL;
2422 data = vcpu->arch.mce_banks[offset];
2423 break;
2424 }
2425 return 1;
2426 }
2427 *pdata = data;
2428 return 0;
2429}
2430
609e36d3 2431int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
890ca9ae 2432{
609e36d3 2433 switch (msr_info->index) {
890ca9ae 2434 case MSR_IA32_PLATFORM_ID:
15c4a640 2435 case MSR_IA32_EBL_CR_POWERON:
b5e2fec0
AG		 2436 case MSR_IA32_DEBUGCTLMSR:
2437 case MSR_IA32_LASTBRANCHFROMIP:
2438 case MSR_IA32_LASTBRANCHTOIP:
2439 case MSR_IA32_LASTINTFROMIP:
2440 case MSR_IA32_LASTINTTOIP:
60af2ecd 2441 case MSR_K8_SYSCFG:
3afb1121
PB		 2442 case MSR_K8_TSEG_ADDR:
2443 case MSR_K8_TSEG_MASK:
60af2ecd 2444 case MSR_K7_HWCR:
61a6bd67 2445 case MSR_VM_HSAVE_PA:
1fdbd48c 2446 case MSR_K8_INT_PENDING_MSG:
c323c0e5 2447 case MSR_AMD64_NB_CFG:
f7c6d140 2448 case MSR_FAM10H_MMIO_CONF_BASE:
2e32b719 2449 case MSR_AMD64_BU_CFG2:
0c2df2a1 2450 case MSR_IA32_PERF_CTL:
405a353a 2451 case MSR_AMD64_DC_CFG:
609e36d3 2452 msr_info->data = 0;
15c4a640 2453 break;
6912ac32
WH		 2454 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2455 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2456 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2457 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
c6702c9d 2458 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
609e36d3
PB		 2459 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2460 msr_info->data = 0;
5753785f 2461 break;
742bc670 2462 case MSR_IA32_UCODE_REV:
609e36d3 2463 msr_info->data = 0x100000000ULL;
742bc670 2464 break;
9ba075a6 2465 case MSR_MTRRcap:
9ba075a6 2466 case 0x200 ... 0x2ff:
ff53604b 2467 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
15c4a640 2468 case 0xcd: /* fsb frequency */
609e36d3 2469 msr_info->data = 3;
15c4a640 2470 break;
7b914098
JS		 2471 /*
2472 * MSR_EBC_FREQUENCY_ID
2473 * Conservative value valid for even the basic CPU models.
2474 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2475 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2476 * and 266MHz for model 3, or 4. Set Core Clock
2477 * Frequency to System Bus Frequency Ratio to 1 (bits
2478 * 31:24) even though these are only valid for CPU
2479 * models > 2, however guests may end up dividing or
2480 * multiplying by zero otherwise.
2481 */
2482 case MSR_EBC_FREQUENCY_ID:
609e36d3 2483 msr_info->data = 1 << 24;
7b914098 2484 break;
15c4a640 2485 case MSR_IA32_APICBASE:
609e36d3 2486 msr_info->data = kvm_get_apic_base(vcpu);
15c4a640 2487 break;
0105d1a5 2488 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
609e36d3 2489 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
0105d1a5 2490 break;
a3e06bbe 2491 case MSR_IA32_TSCDEADLINE:
609e36d3 2492 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
a3e06bbe 2493 break;
ba904635 2494 case MSR_IA32_TSC_ADJUST:
609e36d3 2495 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
ba904635 2496 break;
15c4a640 2497 case MSR_IA32_MISC_ENABLE:
609e36d3 2498 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
15c4a640 2499 break;
64d60670
PB		 2500 case MSR_IA32_SMBASE:
2501 if (!msr_info->host_initiated)
2502 return 1;
2503 msr_info->data = vcpu->arch.smbase;
15c4a640 2504 break;
847f0ad8
AG		 2505 case MSR_IA32_PERF_STATUS:
2506 /* TSC increment by tick */
609e36d3 2507 msr_info->data = 1000ULL;
847f0ad8 2508 /* CPU multiplier */
b0996ae4 2509 msr_info->data |= (((uint64_t)4ULL) << 40);
847f0ad8 2510 break;
15c4a640 2511 case MSR_EFER:
609e36d3 2512 msr_info->data = vcpu->arch.efer;
15c4a640 2513 break;
18068523 2514 case MSR_KVM_WALL_CLOCK:
11c6bffa 2515 case MSR_KVM_WALL_CLOCK_NEW:
609e36d3 2516 msr_info->data = vcpu->kvm->arch.wall_clock;
18068523
GOC		 2517 break;
2518 case MSR_KVM_SYSTEM_TIME:
11c6bffa 2519 case MSR_KVM_SYSTEM_TIME_NEW:
609e36d3 2520 msr_info->data = vcpu->arch.time;
18068523 2521 break;
344d9588 2522 case MSR_KVM_ASYNC_PF_EN:
609e36d3 2523 msr_info->data = vcpu->arch.apf.msr_val;
344d9588 2524 break;
c9aaa895 2525 case MSR_KVM_STEAL_TIME:
609e36d3 2526 msr_info->data = vcpu->arch.st.msr_val;
c9aaa895 2527 break;
1d92128f 2528 case MSR_KVM_PV_EOI_EN:
609e36d3 2529 msr_info->data = vcpu->arch.pv_eoi.msr_val;
1d92128f 2530 break;
890ca9ae
HY		 2531 case MSR_IA32_P5_MC_ADDR:
2532 case MSR_IA32_P5_MC_TYPE:
2533 case MSR_IA32_MCG_CAP:
2534 case MSR_IA32_MCG_CTL:
2535 case MSR_IA32_MCG_STATUS:
81760dcc 2536 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
609e36d3 2537 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
84e0cefa
JS		 2538 case MSR_K7_CLK_CTL:
2539 /*
2540 * Provide expected ramp-up count for K7. All other
2541 * are set to zero, indicating minimum divisors for
2542 * every field.
2543 *
2544 * This prevents guest kernels on AMD host with CPU
2545 * type 6, model 8 and higher from exploding due to
2546 * the rdmsr failing.
2547 */
609e36d3 2548 msr_info->data = 0x20000000;
84e0cefa 2549 break;
55cd8e5a 2550 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
e7d9513b
AS		 2551 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2552 case HV_X64_MSR_CRASH_CTL:
1f4b34f8 2553 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
e83d5887
AS		 2554 return kvm_hv_get_msr_common(vcpu,
2555 msr_info->index, &msr_info->data);
55cd8e5a 2556 break;
91c9c3ed 2557 case MSR_IA32_BBL_CR_CTL3:
2558 /* This legacy MSR exists but isn't fully documented in current
2559 * silicon. It is however accessed by winxp in very narrow
2560 * scenarios where it sets bit #19, itself documented as
2561 * a "reserved" bit. Best effort attempt to source coherent
2562 * read data here should the balance of the register be
2563 * interpreted by the guest:
2564 *
2565 * L2 cache control register 3: 64GB range, 256KB size,
2566 * enabled, latency 0x1, configured
2567 */
609e36d3 2568 msr_info->data = 0xbe702111;
91c9c3ed 2569 break;
2b036c6b 2570 case MSR_AMD64_OSVW_ID_LENGTH:
d6321d49 2571 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2b036c6b 2572 return 1;
609e36d3 2573 msr_info->data = vcpu->arch.osvw.length;
2b036c6b
BO		 2574 break;
2575 case MSR_AMD64_OSVW_STATUS:
d6321d49 2576 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2b036c6b 2577 return 1;
609e36d3 2578 msr_info->data = vcpu->arch.osvw.status;
2b036c6b 2579 break;
db2336a8
KH		 2580 case MSR_PLATFORM_INFO:
2581 msr_info->data = vcpu->arch.msr_platform_info;
2582 break;
2583 case MSR_MISC_FEATURES_ENABLES:
2584 msr_info->data = vcpu->arch.msr_misc_features_enables;
2585 break;
15c4a640 2586 default:
c6702c9d 2587 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
609e36d3 2588 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
ed85c068 2589 if (!ignore_msrs) {
ae0f5499
BD		 2590 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2591 msr_info->index);
ed85c068
AP		 2592 return 1;
2593 } else {
fab0aa3b
EM		 2594 if (report_ignored_msrs)
2595 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
2596 msr_info->index);
609e36d3 2597 msr_info->data = 0;
ed85c068
AP		 2598 }
2599 break;
15c4a640 2600 }
15c4a640
CO		 2601 return 0;
2602}
2603EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2604
313a3dc7
CO		 2605/*
2606 * Read or write a bunch of msrs. All parameters are kernel addresses.
2607 *
2608 * @return number of msrs set successfully.
2609 */
2610static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2611 struct kvm_msr_entry *entries,
2612 int (*do_msr)(struct kvm_vcpu *vcpu,
2613 unsigned index, u64 *data))
2614{
f656ce01 2615 int i, idx;
313a3dc7 2616
f656ce01 2617 idx = srcu_read_lock(&vcpu->kvm->srcu);
313a3dc7
CO		 2618 for (i = 0; i < msrs->nmsrs; ++i)
2619 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2620 break;
f656ce01 2621 srcu_read_unlock(&vcpu->kvm->srcu, idx);
313a3dc7 2622
313a3dc7
CO		 2623 return i;
2624}
2625
2626/*
2627 * Read or write a bunch of msrs. Parameters are user addresses.
2628 *
2629 * @return number of msrs set successfully.
2630 */
2631static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2632 int (*do_msr)(struct kvm_vcpu *vcpu,
2633 unsigned index, u64 *data),
2634 int writeback)
2635{
2636 struct kvm_msrs msrs;
2637 struct kvm_msr_entry *entries;
2638 int r, n;
2639 unsigned size;
2640
2641 r = -EFAULT;
2642 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2643 goto out;
2644
2645 r = -E2BIG;
2646 if (msrs.nmsrs >= MAX_IO_MSRS)
2647 goto out;
2648
313a3dc7 2649 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
ff5c2c03
SL		 2650 entries = memdup_user(user_msrs->entries, size);
2651 if (IS_ERR(entries)) {
2652 r = PTR_ERR(entries);
313a3dc7 2653 goto out;
ff5c2c03 2654 }
313a3dc7
CO		 2655
2656 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2657 if (r < 0)
2658 goto out_free;
2659
2660 r = -EFAULT;
2661 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2662 goto out_free;
2663
2664 r = n;
2665
2666out_free:
7a73c028 2667 kfree(entries);
313a3dc7
CO		 2668out:
2669 return r;
2670}
2671
784aa3d7 2672int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
018d00d2
ZX		 2673{
2674 int r;
2675
2676 switch (ext) {
2677 case KVM_CAP_IRQCHIP:
2678 case KVM_CAP_HLT:
2679 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
018d00d2 2680 case KVM_CAP_SET_TSS_ADDR:
07716717 2681 case KVM_CAP_EXT_CPUID:
9c15bb1d 2682 case KVM_CAP_EXT_EMUL_CPUID:
c8076604 2683 case KVM_CAP_CLOCKSOURCE:
7837699f 2684 case KVM_CAP_PIT:
a28e4f5a 2685 case KVM_CAP_NOP_IO_DELAY:
62d9f0db 2686 case KVM_CAP_MP_STATE:
ed848624 2687 case KVM_CAP_SYNC_MMU:
a355c85c 2688 case KVM_CAP_USER_NMI:
52d939a0 2689 case KVM_CAP_REINJECT_CONTROL:
4925663a 2690 case KVM_CAP_IRQ_INJECT_STATUS:
d34e6b17 2691 case KVM_CAP_IOEVENTFD:
f848a5a8 2692 case KVM_CAP_IOEVENTFD_NO_LENGTH:
c5ff41ce 2693 case KVM_CAP_PIT2:
e9f42757 2694 case KVM_CAP_PIT_STATE2:
b927a3ce 2695 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
ffde22ac 2696 case KVM_CAP_XEN_HVM:
3cfc3092 2697 case KVM_CAP_VCPU_EVENTS:
55cd8e5a 2698 case KVM_CAP_HYPERV:
10388a07 2699 case KVM_CAP_HYPERV_VAPIC:
c25bc163 2700 case KVM_CAP_HYPERV_SPIN:
5c919412 2701 case KVM_CAP_HYPERV_SYNIC:
efc479e6 2702 case KVM_CAP_HYPERV_SYNIC2:
d3457c87 2703 case KVM_CAP_HYPERV_VP_INDEX:
ab9f4ecb 2704 case KVM_CAP_PCI_SEGMENT:
a1efbe77 2705 case KVM_CAP_DEBUGREGS:
d2be1651 2706 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2d5b5a66 2707 case KVM_CAP_XSAVE:
344d9588 2708 case KVM_CAP_ASYNC_PF:
92a1f12d 2709 case KVM_CAP_GET_TSC_KHZ:
1c0b28c2 2710 case KVM_CAP_KVMCLOCK_CTRL:
4d8b81ab 2711 case KVM_CAP_READONLY_MEM:
5f66b620 2712 case KVM_CAP_HYPERV_TIME:
100943c5 2713 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
defcf51f 2714 case KVM_CAP_TSC_DEADLINE_TIMER:
90de4a18
NA		 2715 case KVM_CAP_ENABLE_CAP_VM:
2716 case KVM_CAP_DISABLE_QUIRKS:
d71ba788 2717 case KVM_CAP_SET_BOOT_CPU_ID:
49df6397 2718 case KVM_CAP_SPLIT_IRQCHIP:
460df4c1 2719 case KVM_CAP_IMMEDIATE_EXIT:
018d00d2
ZX		 2720 r = 1;
2721 break;
e3fd9a93
PB		 2722 case KVM_CAP_ADJUST_CLOCK:
2723 r = KVM_CLOCK_TSC_STABLE;
2724 break;
668fffa3
MT		 2725 case KVM_CAP_X86_GUEST_MWAIT:
2726 r = kvm_mwait_in_guest();
2727 break;
6d396b55
PB		 2728 case KVM_CAP_X86_SMM:
2729 /* SMBASE is usually relocated above 1M on modern chipsets,
2730 * and SMM handlers might indeed rely on 4G segment limits,
2731 * so do not report SMM to be available if real mode is
2732 * emulated via vm86 mode. Still, do not go to great lengths
2733 * to avoid userspace's usage of the feature, because it is a
2734 * fringe case that is not enabled except via specific settings
2735 * of the module parameters.
2736 */
2737 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2738 break;
774ead3a
AK		 2739 case KVM_CAP_VAPIC:
2740 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2741 break;
f725230a 2742 case KVM_CAP_NR_VCPUS:
8c3ba334
SL		 2743 r = KVM_SOFT_MAX_VCPUS;
2744 break;
2745 case KVM_CAP_MAX_VCPUS:
f725230a
AK		 2746 r = KVM_MAX_VCPUS;
2747 break;
a988b910 2748 case KVM_CAP_NR_MEMSLOTS:
bbacc0c1 2749 r = KVM_USER_MEM_SLOTS;
a988b910 2750 break;
a68a6a72
MT		 2751 case KVM_CAP_PV_MMU: /* obsolete */
2752 r = 0;
2f333bcb 2753 break;
890ca9ae
HY		 2754 case KVM_CAP_MCE:
2755 r = KVM_MAX_MCE_BANKS;
2756 break;
2d5b5a66 2757 case KVM_CAP_XCRS:
d366bf7e 2758 r = boot_cpu_has(X86_FEATURE_XSAVE);
2d5b5a66 2759 break;
92a1f12d
JR		 2760 case KVM_CAP_TSC_CONTROL:
2761 r = kvm_has_tsc_control;
2762 break;
37131313
RK		 2763 case KVM_CAP_X2APIC_API:
2764 r = KVM_X2APIC_API_VALID_FLAGS;
2765 break;
018d00d2
ZX		 2766 default:
2767 r = 0;
2768 break;
2769 }
2770 return r;
2771
2772}
2773
043405e1
CO		 2774long kvm_arch_dev_ioctl(struct file *filp,
2775 unsigned int ioctl, unsigned long arg)
2776{
2777 void __user *argp = (void __user *)arg;
2778 long r;
2779
2780 switch (ioctl) {
2781 case KVM_GET_MSR_INDEX_LIST: {
2782 struct kvm_msr_list __user *user_msr_list = argp;
2783 struct kvm_msr_list msr_list;
2784 unsigned n;
2785
2786 r = -EFAULT;
2787 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2788 goto out;
2789 n = msr_list.nmsrs;
62ef68bb 2790 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
043405e1
CO		 2791 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2792 goto out;
2793 r = -E2BIG;
e125e7b6 2794 if (n < msr_list.nmsrs)
043405e1
CO		 2795 goto out;
2796 r = -EFAULT;
2797 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2798 num_msrs_to_save * sizeof(u32)))
2799 goto out;
e125e7b6 2800 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
043405e1 2801 &emulated_msrs,
62ef68bb 2802 num_emulated_msrs * sizeof(u32)))
043405e1
CO		 2803 goto out;
2804 r = 0;
2805 break;
2806 }
9c15bb1d
BP		 2807 case KVM_GET_SUPPORTED_CPUID:
2808 case KVM_GET_EMULATED_CPUID: {
674eea0f
AK		 2809 struct kvm_cpuid2 __user *cpuid_arg = argp;
2810 struct kvm_cpuid2 cpuid;
2811
2812 r = -EFAULT;
2813 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2814 goto out;
9c15bb1d
BP		 2815
2816 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2817 ioctl);
674eea0f
AK		 2818 if (r)
2819 goto out;
2820
2821 r = -EFAULT;
2822 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2823 goto out;
2824 r = 0;
2825 break;
2826 }
890ca9ae 2827 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
890ca9ae 2828 r = -EFAULT;
c45dcc71
AR		 2829 if (copy_to_user(argp, &kvm_mce_cap_supported,
2830 sizeof(kvm_mce_cap_supported)))
890ca9ae
HY		 2831 goto out;
2832 r = 0;
2833 break;
2834 }
043405e1
CO		 2835 default:
2836 r = -EINVAL;
2837 }
2838out:
2839 return r;
2840}
2841
f5f48ee1
SY		 2842static void wbinvd_ipi(void *garbage)
2843{
2844 wbinvd();
2845}
2846
2847static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2848{
e0f0bbc5 2849 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
f5f48ee1
SY		 2850}
2851
313a3dc7
CO		 2852void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2853{
f5f48ee1
SY		 2854 /* Address WBINVD may be executed by guest */
2855 if (need_emulate_wbinvd(vcpu)) {
2856 if (kvm_x86_ops->has_wbinvd_exit())
2857 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2858 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2859 smp_call_function_single(vcpu->cpu,
2860 wbinvd_ipi, NULL, 1);
2861 }
2862
313a3dc7 2863 kvm_x86_ops->vcpu_load(vcpu, cpu);
8f6055cb 2864
0dd6a6ed
ZA		 2865 /* Apply any externally detected TSC adjustments (due to suspend) */
2866 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2867 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2868 vcpu->arch.tsc_offset_adjustment = 0;
105b21bb 2869 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
0dd6a6ed 2870 }
8f6055cb 2871
48434c20 2872 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
6f526ec5 2873 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4ea1636b 2874 rdtsc() - vcpu->arch.last_host_tsc;
e48672fa
ZA		 2875 if (tsc_delta < 0)
2876 mark_tsc_unstable("KVM discovered backwards TSC");
ce7a058a 2877
c285545f 2878 if (check_tsc_unstable()) {
07c1419a 2879 u64 offset = kvm_compute_tsc_offset(vcpu,
b183aa58 2880 vcpu->arch.last_guest_tsc);
a545ab6a 2881 kvm_vcpu_write_tsc_offset(vcpu, offset);
c285545f 2882 vcpu->arch.tsc_catchup = 1;
c285545f 2883 }
a749e247
PB		 2884
2885 if (kvm_lapic_hv_timer_in_use(vcpu))
2886 kvm_lapic_restart_hv_timer(vcpu);
2887
d98d07ca
MT		 2888 /*
2889 * On a host with synchronized TSC, there is no need to update
2890 * kvmclock on vcpu->cpu migration
2891 */
2892 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
0061d53d 2893 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
c285545f 2894 if (vcpu->cpu != cpu)
1bd2009e 2895 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
e48672fa 2896 vcpu->cpu = cpu;
6b7d7e76 2897 }
c9aaa895 2898
c9aaa895 2899 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
313a3dc7
CO		 2900}
2901
0b9f6c46
PX		 2902static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
2903{
2904 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2905 return;
2906
2907 vcpu->arch.st.steal.preempted = 1;
2908
4e335d9e 2909 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
0b9f6c46
PX		 2910 &vcpu->arch.st.steal.preempted,
2911 offsetof(struct kvm_steal_time, preempted),
2912 sizeof(vcpu->arch.st.steal.preempted));
2913}
2914
313a3dc7
CO		 2915void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2916{
cc0d907c 2917 int idx;
de63ad4c
LM		 2918
2919 if (vcpu->preempted)
2920 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
2921
931f261b
AA		 2922 /*
2923 * Disable page faults because we're in atomic context here.
2924 * kvm_write_guest_offset_cached() would call might_fault()
2925 * that relies on pagefault_disable() to tell if there's a
2926 * bug. NOTE: the write to guest memory may not go through if
2927 * during postcopy live migration or if there's heavy guest
2928 * paging.
2929 */
2930 pagefault_disable();
cc0d907c
AA		 2931 /*
2932 * kvm_memslots() will be called by
2933 * kvm_write_guest_offset_cached() so take the srcu lock.
2934 */
2935 idx = srcu_read_lock(&vcpu->kvm->srcu);
0b9f6c46 2936 kvm_steal_time_set_preempted(vcpu);
cc0d907c 2937 srcu_read_unlock(&vcpu->kvm->srcu, idx);
931f261b 2938 pagefault_enable();
02daab21 2939 kvm_x86_ops->vcpu_put(vcpu);
4ea1636b 2940 vcpu->arch.last_host_tsc = rdtsc();
313a3dc7
CO		 2941}
2942
313a3dc7
CO		 2943static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2944 struct kvm_lapic_state *s)
2945{
76dfafd5 2946 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
d62caabb
AS		 2947 kvm_x86_ops->sync_pir_to_irr(vcpu);
2948
a92e2543 2949 return kvm_apic_get_state(vcpu, s);
313a3dc7
CO		 2950}
2951
2952static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2953 struct kvm_lapic_state *s)
2954{
a92e2543
RK		 2955 int r;
2956
2957 r = kvm_apic_set_state(vcpu, s);
2958 if (r)
2959 return r;
cb142eb7 2960 update_cr8_intercept(vcpu);
313a3dc7
CO		 2961
2962 return 0;
2963}
2964
127a457a
MG		 2965static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2966{
2967 return (!lapic_in_kernel(vcpu) ||
2968 kvm_apic_accept_pic_intr(vcpu));
2969}
2970
782d422b
MG		 2971/*
2972 * if userspace requested an interrupt window, check that the
2973 * interrupt window is open.
2974 *
2975 * No need to exit to userspace if we already have an interrupt queued.
2976 */
2977static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2978{
2979 return kvm_arch_interrupt_allowed(vcpu) &&
2980 !kvm_cpu_has_interrupt(vcpu) &&
2981 !kvm_event_needs_reinjection(vcpu) &&
2982 kvm_cpu_accept_dm_intr(vcpu);
2983}
2984
f77bc6a4
ZX		 2985static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2986 struct kvm_interrupt *irq)
2987{
02cdb50f 2988 if (irq->irq >= KVM_NR_INTERRUPTS)
f77bc6a4 2989 return -EINVAL;
1c1a9ce9
SR		 2990
2991 if (!irqchip_in_kernel(vcpu->kvm)) {
2992 kvm_queue_interrupt(vcpu, irq->irq, false);
2993 kvm_make_request(KVM_REQ_EVENT, vcpu);
2994 return 0;
2995 }
2996
2997 /*
2998 * With in-kernel LAPIC, we only use this to inject EXTINT, so
2999 * fail for in-kernel 8259.
3000 */
3001 if (pic_in_kernel(vcpu->kvm))
f77bc6a4 3002 return -ENXIO;
f77bc6a4 3003
1c1a9ce9
SR		 3004 if (vcpu->arch.pending_external_vector != -1)
3005 return -EEXIST;
f77bc6a4 3006
1c1a9ce9 3007 vcpu->arch.pending_external_vector = irq->irq;
934bf653 3008 kvm_make_request(KVM_REQ_EVENT, vcpu);
f77bc6a4
ZX		 3009 return 0;
3010}
3011
c4abb7c9
JK		 3012static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3013{
c4abb7c9 3014 kvm_inject_nmi(vcpu);
c4abb7c9
JK		 3015
3016 return 0;
3017}
3018
f077825a
PB		 3019static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3020{
64d60670
PB		 3021 kvm_make_request(KVM_REQ_SMI, vcpu);
3022
f077825a
PB		 3023 return 0;
3024}
3025
b209749f
AK		 3026static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3027 struct kvm_tpr_access_ctl *tac)
3028{
3029 if (tac->flags)
3030 return -EINVAL;
3031 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3032 return 0;
3033}
3034
890ca9ae
HY		 3035static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3036 u64 mcg_cap)
3037{
3038 int r;
3039 unsigned bank_num = mcg_cap & 0xff, bank;
3040
3041 r = -EINVAL;
a9e38c3e 3042 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
890ca9ae 3043 goto out;
c45dcc71 3044 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
890ca9ae
HY		 3045 goto out;
3046 r = 0;
3047 vcpu->arch.mcg_cap = mcg_cap;
3048 /* Init IA32_MCG_CTL to all 1s */
3049 if (mcg_cap & MCG_CTL_P)
3050 vcpu->arch.mcg_ctl = ~(u64)0;
3051 /* Init IA32_MCi_CTL to all 1s */
3052 for (bank = 0; bank < bank_num; bank++)
3053 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
c45dcc71
AR		 3054
3055 if (kvm_x86_ops->setup_mce)
3056 kvm_x86_ops->setup_mce(vcpu);
890ca9ae
HY		 3057out:
3058 return r;
3059}
3060
3061static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3062 struct kvm_x86_mce *mce)
3063{
3064 u64 mcg_cap = vcpu->arch.mcg_cap;
3065 unsigned bank_num = mcg_cap & 0xff;
3066 u64 *banks = vcpu->arch.mce_banks;
3067
3068 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3069 return -EINVAL;
3070 /*
3071 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3072 * reporting is disabled
3073 */
3074 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3075 vcpu->arch.mcg_ctl != ~(u64)0)
3076 return 0;
3077 banks += 4 * mce->bank;
3078 /*
3079 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3080 * reporting is disabled for the bank
3081 */
3082 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3083 return 0;
3084 if (mce->status & MCI_STATUS_UC) {
3085 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
fc78f519 3086 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
a8eeb04a 3087 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
890ca9ae
HY		 3088 return 0;
3089 }
3090 if (banks[1] & MCI_STATUS_VAL)
3091 mce->status |= MCI_STATUS_OVER;
3092 banks[2] = mce->addr;
3093 banks[3] = mce->misc;
3094 vcpu->arch.mcg_status = mce->mcg_status;
3095 banks[1] = mce->status;
3096 kvm_queue_exception(vcpu, MC_VECTOR);
3097 } else if (!(banks[1] & MCI_STATUS_VAL)
3098 || !(banks[1] & MCI_STATUS_UC)) {
3099 if (banks[1] & MCI_STATUS_VAL)
3100 mce->status |= MCI_STATUS_OVER;
3101 banks[2] = mce->addr;
3102 banks[3] = mce->misc;
3103 banks[1] = mce->status;
3104 } else
3105 banks[1] |= MCI_STATUS_OVER;
3106 return 0;
3107}
3108
3cfc3092
JK		 3109static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3110 struct kvm_vcpu_events *events)
3111{
7460fb4a 3112 process_nmi(vcpu);
664f8e26
WL		 3113 /*
3114 * FIXME: pass injected and pending separately. This is only
3115 * needed for nested virtualization, whose state cannot be
3116 * migrated yet. For now we can combine them.
3117 */
03b82a30 3118 events->exception.injected =
664f8e26
WL		 3119 (vcpu->arch.exception.pending ||
3120 vcpu->arch.exception.injected) &&
03b82a30 3121 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3cfc3092
JK		 3122 events->exception.nr = vcpu->arch.exception.nr;
3123 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
97e69aa6 3124 events->exception.pad = 0;
3cfc3092
JK		 3125 events->exception.error_code = vcpu->arch.exception.error_code;
3126
03b82a30
JK		 3127 events->interrupt.injected =
3128 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3cfc3092 3129 events->interrupt.nr = vcpu->arch.interrupt.nr;
03b82a30 3130 events->interrupt.soft = 0;
37ccdcbe 3131 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3cfc3092
JK		 3132
3133 events->nmi.injected = vcpu->arch.nmi_injected;
7460fb4a 3134 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3cfc3092 3135 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
97e69aa6 3136 events->nmi.pad = 0;
3cfc3092 3137
66450a21 3138 events->sipi_vector = 0; /* never valid when reporting to user space */
3cfc3092 3139
f077825a
PB		 3140 events->smi.smm = is_smm(vcpu);
3141 events->smi.pending = vcpu->arch.smi_pending;
3142 events->smi.smm_inside_nmi =
3143 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3144 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3145
dab4b911 3146 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
f077825a
PB		 3147 | KVM_VCPUEVENT_VALID_SHADOW
3148 | KVM_VCPUEVENT_VALID_SMM);
97e69aa6 3149 memset(&events->reserved, 0, sizeof(events->reserved));
3cfc3092
JK		 3150}
3151
6ef4e07e
XG		 3152static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3153
3cfc3092
JK		 3154static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3155 struct kvm_vcpu_events *events)
3156{
dab4b911 3157 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
48005f64 3158 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
f077825a
PB		 3159 | KVM_VCPUEVENT_VALID_SHADOW
3160 | KVM_VCPUEVENT_VALID_SMM))
3cfc3092
JK		 3161 return -EINVAL;
3162
78e546c8 3163 if (events->exception.injected &&
28d06353
JM		 3164 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR ||
3165 is_guest_mode(vcpu)))
78e546c8
PB		 3166 return -EINVAL;
3167
28bf2888
DH		 3168 /* INITs are latched while in SMM */
3169 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3170 (events->smi.smm || events->smi.pending) &&
3171 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3172 return -EINVAL;
3173
7460fb4a 3174 process_nmi(vcpu);
664f8e26 3175 vcpu->arch.exception.injected = false;
3cfc3092
JK		 3176 vcpu->arch.exception.pending = events->exception.injected;
3177 vcpu->arch.exception.nr = events->exception.nr;
3178 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3179 vcpu->arch.exception.error_code = events->exception.error_code;
3180
3181 vcpu->arch.interrupt.pending = events->interrupt.injected;
3182 vcpu->arch.interrupt.nr = events->interrupt.nr;
3183 vcpu->arch.interrupt.soft = events->interrupt.soft;
48005f64
JK		 3184 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3185 kvm_x86_ops->set_interrupt_shadow(vcpu,
3186 events->interrupt.shadow);
3cfc3092
JK		 3187
3188 vcpu->arch.nmi_injected = events->nmi.injected;
dab4b911
JK		 3189 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3190 vcpu->arch.nmi_pending = events->nmi.pending;
3cfc3092
JK		 3191 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3192
66450a21 3193 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
bce87cce 3194 lapic_in_kernel(vcpu))
66450a21 3195 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3cfc3092 3196
f077825a 3197 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
6ef4e07e 3198 u32 hflags = vcpu->arch.hflags;
f077825a 3199 if (events->smi.smm)
6ef4e07e 3200 hflags |= HF_SMM_MASK;
f077825a 3201 else
6ef4e07e
XG		 3202 hflags &= ~HF_SMM_MASK;
3203 kvm_set_hflags(vcpu, hflags);
3204
f077825a 3205 vcpu->arch.smi_pending = events->smi.pending;
f4ef1910
WL		 3206
3207 if (events->smi.smm) {
3208 if (events->smi.smm_inside_nmi)
3209 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
f077825a 3210 else
f4ef1910
WL		 3211 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3212 if (lapic_in_kernel(vcpu)) {
3213 if (events->smi.latched_init)
3214 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3215 else
3216 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3217 }
f077825a
PB		 3218 }
3219 }
3220
3842d135
AK		 3221 kvm_make_request(KVM_REQ_EVENT, vcpu);
3222
3cfc3092
JK		 3223 return 0;
3224}
3225
a1efbe77
JK		 3226static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3227 struct kvm_debugregs *dbgregs)
3228{
73aaf249
JK		 3229 unsigned long val;
3230
a1efbe77 3231 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
16f8a6f9 3232 kvm_get_dr(vcpu, 6, &val);
73aaf249 3233 dbgregs->dr6 = val;
a1efbe77
JK		 3234 dbgregs->dr7 = vcpu->arch.dr7;
3235 dbgregs->flags = 0;
97e69aa6 3236 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
a1efbe77
JK		 3237}
3238
3239static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3240 struct kvm_debugregs *dbgregs)
3241{
3242 if (dbgregs->flags)
3243 return -EINVAL;
3244
d14bdb55
PB		 3245 if (dbgregs->dr6 & ~0xffffffffull)
3246 return -EINVAL;
3247 if (dbgregs->dr7 & ~0xffffffffull)
3248 return -EINVAL;
3249
a1efbe77 3250 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
ae561ede 3251 kvm_update_dr0123(vcpu);
a1efbe77 3252 vcpu->arch.dr6 = dbgregs->dr6;
73aaf249 3253 kvm_update_dr6(vcpu);
a1efbe77 3254 vcpu->arch.dr7 = dbgregs->dr7;
9926c9fd 3255 kvm_update_dr7(vcpu);
a1efbe77 3256
a1efbe77
JK		 3257 return 0;
3258}
3259
df1daba7
PB		 3260#define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3261
3262static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3263{
c47ada30 3264 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
400e4b20 3265 u64 xstate_bv = xsave->header.xfeatures;
df1daba7
PB		 3266 u64 valid;
3267
3268 /*
3269 * Copy legacy XSAVE area, to avoid complications with CPUID
3270 * leaves 0 and 1 in the loop below.
3271 */
3272 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3273
3274 /* Set XSTATE_BV */
00c87e9a 3275 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
df1daba7
PB		 3276 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3277
3278 /*
3279 * Copy each region from the possibly compacted offset to the
3280 * non-compacted offset.
3281 */
d91cab78 3282 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
df1daba7
PB		 3283 while (valid) {
3284 u64 feature = valid & -valid;
3285 int index = fls64(feature) - 1;
3286 void *src = get_xsave_addr(xsave, feature);
3287
3288 if (src) {
3289 u32 size, offset, ecx, edx;
3290 cpuid_count(XSTATE_CPUID, index,
3291 &size, &offset, &ecx, &edx);
38cfd5e3
PB		 3292 if (feature == XFEATURE_MASK_PKRU)
3293 memcpy(dest + offset, &vcpu->arch.pkru,
3294 sizeof(vcpu->arch.pkru));
3295 else
3296 memcpy(dest + offset, src, size);
3297
df1daba7
PB		 3298 }
3299
3300 valid -= feature;
3301 }
3302}
3303
3304static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3305{
c47ada30 3306 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
df1daba7
PB		 3307 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3308 u64 valid;
3309
3310 /*
3311 * Copy legacy XSAVE area, to avoid complications with CPUID
3312 * leaves 0 and 1 in the loop below.
3313 */
3314 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3315
3316 /* Set XSTATE_BV and possibly XCOMP_BV. */
400e4b20 3317 xsave->header.xfeatures = xstate_bv;
782511b0 3318 if (boot_cpu_has(X86_FEATURE_XSAVES))
3a54450b 3319 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
df1daba7
PB		 3320
3321 /*
3322 * Copy each region from the non-compacted offset to the
3323 * possibly compacted offset.
3324 */
d91cab78 3325 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
df1daba7
PB		 3326 while (valid) {
3327 u64 feature = valid & -valid;
3328 int index = fls64(feature) - 1;
3329 void *dest = get_xsave_addr(xsave, feature);
3330
3331 if (dest) {
3332 u32 size, offset, ecx, edx;
3333 cpuid_count(XSTATE_CPUID, index,
3334 &size, &offset, &ecx, &edx);
38cfd5e3
PB		 3335 if (feature == XFEATURE_MASK_PKRU)
3336 memcpy(&vcpu->arch.pkru, src + offset,
3337 sizeof(vcpu->arch.pkru));
3338 else
3339 memcpy(dest, src + offset, size);
ee4100da 3340 }
df1daba7
PB		 3341
3342 valid -= feature;
3343 }
3344}
3345
2d5b5a66
SY		 3346static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3347 struct kvm_xsave *guest_xsave)
3348{
d366bf7e 3349 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
df1daba7
PB		 3350 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3351 fill_xsave((u8 *) guest_xsave->region, vcpu);
4344ee98 3352 } else {
2d5b5a66 3353 memcpy(guest_xsave->region,
7366ed77 3354 &vcpu->arch.guest_fpu.state.fxsave,
c47ada30 3355 sizeof(struct fxregs_state));
2d5b5a66 3356 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
d91cab78 3357 XFEATURE_MASK_FPSSE;
2d5b5a66
SY		 3358 }
3359}
3360
a575813b
WL		 3361#define XSAVE_MXCSR_OFFSET 24
3362
2d5b5a66
SY		 3363static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3364 struct kvm_xsave *guest_xsave)
3365{
3366 u64 xstate_bv =
3367 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
a575813b 3368 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
2d5b5a66 3369
d366bf7e 3370 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
d7876f1b
PB		 3371 /*
3372 * Here we allow setting states that are not present in
3373 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3374 * with old userspace.
3375 */
a575813b
WL		 3376 if (xstate_bv & ~kvm_supported_xcr0() ||
3377 mxcsr & ~mxcsr_feature_mask)
d7876f1b 3378 return -EINVAL;
df1daba7 3379 load_xsave(vcpu, (u8 *)guest_xsave->region);
d7876f1b 3380 } else {
a575813b
WL		 3381 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3382 mxcsr & ~mxcsr_feature_mask)
2d5b5a66 3383 return -EINVAL;
7366ed77 3384 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
c47ada30 3385 guest_xsave->region, sizeof(struct fxregs_state));
2d5b5a66
SY		 3386 }
3387 return 0;
3388}
3389
3390static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3391 struct kvm_xcrs *guest_xcrs)
3392{
d366bf7e 3393 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
2d5b5a66
SY		 3394 guest_xcrs->nr_xcrs = 0;
3395 return;
3396 }
3397
3398 guest_xcrs->nr_xcrs = 1;
3399 guest_xcrs->flags = 0;
3400 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3401 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3402}
3403
3404static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3405 struct kvm_xcrs *guest_xcrs)
3406{
3407 int i, r = 0;
3408
d366bf7e 3409 if (!boot_cpu_has(X86_FEATURE_XSAVE))
2d5b5a66
SY		 3410 return -EINVAL;
3411
3412 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3413 return -EINVAL;
3414
3415 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3416 /* Only support XCR0 currently */
c67a04cb 3417 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
2d5b5a66 3418 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
c67a04cb 3419 guest_xcrs->xcrs[i].value);
2d5b5a66
SY		 3420 break;
3421 }
3422 if (r)
3423 r = -EINVAL;
3424 return r;
3425}
3426
1c0b28c2
EM		 3427/*
3428 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3429 * stopped by the hypervisor. This function will be called from the host only.
3430 * EINVAL is returned when the host attempts to set the flag for a guest that
3431 * does not support pv clocks.
3432 */
3433static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3434{
0b79459b 3435 if (!vcpu->arch.pv_time_enabled)
1c0b28c2 3436 return -EINVAL;
51d59c6b 3437 vcpu->arch.pvclock_set_guest_stopped_request = true;
1c0b28c2
EM		 3438 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3439 return 0;
3440}
3441
5c919412
AS		 3442static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3443 struct kvm_enable_cap *cap)
3444{
3445 if (cap->flags)
3446 return -EINVAL;
3447
3448 switch (cap->cap) {
efc479e6
RK		 3449 case KVM_CAP_HYPERV_SYNIC2:
3450 if (cap->args[0])
3451 return -EINVAL;
5c919412 3452 case KVM_CAP_HYPERV_SYNIC:
546d87e5
WL		 3453 if (!irqchip_in_kernel(vcpu->kvm))
3454 return -EINVAL;
efc479e6
RK		 3455 return kvm_hv_activate_synic(vcpu, cap->cap ==
3456 KVM_CAP_HYPERV_SYNIC2);
5c919412
AS		 3457 default:
3458 return -EINVAL;
3459 }
3460}
3461
313a3dc7
CO		 3462long kvm_arch_vcpu_ioctl(struct file *filp,
3463 unsigned int ioctl, unsigned long arg)
3464{
3465 struct kvm_vcpu *vcpu = filp->private_data;
3466 void __user *argp = (void __user *)arg;
3467 int r;
d1ac91d8
AK		 3468 union {
3469 struct kvm_lapic_state *lapic;
3470 struct kvm_xsave *xsave;
3471 struct kvm_xcrs *xcrs;
3472 void *buffer;
3473 } u;
3474
3475 u.buffer = NULL;
313a3dc7
CO		 3476 switch (ioctl) {
3477 case KVM_GET_LAPIC: {
2204ae3c 3478 r = -EINVAL;
bce87cce 3479 if (!lapic_in_kernel(vcpu))
2204ae3c 3480 goto out;
d1ac91d8 3481 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
313a3dc7 3482
b772ff36 3483 r = -ENOMEM;
d1ac91d8 3484 if (!u.lapic)
b772ff36 3485 goto out;
d1ac91d8 3486 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
313a3dc7
CO		 3487 if (r)
3488 goto out;
3489 r = -EFAULT;
d1ac91d8 3490 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
313a3dc7
CO		 3491 goto out;
3492 r = 0;
3493 break;
3494 }
3495 case KVM_SET_LAPIC: {
2204ae3c 3496 r = -EINVAL;
bce87cce 3497 if (!lapic_in_kernel(vcpu))
2204ae3c 3498 goto out;
ff5c2c03 3499 u.lapic = memdup_user(argp, sizeof(*u.lapic));
18595411
GC		 3500 if (IS_ERR(u.lapic))
3501 return PTR_ERR(u.lapic);
ff5c2c03 3502
d1ac91d8 3503 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
313a3dc7
CO		 3504 break;
3505 }
f77bc6a4
ZX		 3506 case KVM_INTERRUPT: {
3507 struct kvm_interrupt irq;
3508
3509 r = -EFAULT;
3510 if (copy_from_user(&irq, argp, sizeof irq))
3511 goto out;
3512 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
f77bc6a4
ZX		 3513 break;
3514 }
c4abb7c9
JK		 3515 case KVM_NMI: {
3516 r = kvm_vcpu_ioctl_nmi(vcpu);
c4abb7c9
JK		 3517 break;
3518 }
f077825a
PB		 3519 case KVM_SMI: {
3520 r = kvm_vcpu_ioctl_smi(vcpu);
3521 break;
3522 }
313a3dc7
CO		 3523 case KVM_SET_CPUID: {
3524 struct kvm_cpuid __user *cpuid_arg = argp;
3525 struct kvm_cpuid cpuid;
3526
3527 r = -EFAULT;
3528 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3529 goto out;
3530 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
313a3dc7
CO		 3531 break;
3532 }
07716717
DK		 3533 case KVM_SET_CPUID2: {
3534 struct kvm_cpuid2 __user *cpuid_arg = argp;
3535 struct kvm_cpuid2 cpuid;
3536
3537 r = -EFAULT;
3538 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3539 goto out;
3540 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
19355475 3541 cpuid_arg->entries);
07716717
DK		 3542 break;
3543 }
3544 case KVM_GET_CPUID2: {
3545 struct kvm_cpuid2 __user *cpuid_arg = argp;
3546 struct kvm_cpuid2 cpuid;
3547
3548 r = -EFAULT;
3549 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3550 goto out;
3551 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
19355475 3552 cpuid_arg->entries);
07716717
DK		 3553 if (r)
3554 goto out;
3555 r = -EFAULT;
3556 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3557 goto out;
3558 r = 0;
3559 break;
3560 }
313a3dc7 3561 case KVM_GET_MSRS:
609e36d3 3562 r = msr_io(vcpu, argp, do_get_msr, 1);
313a3dc7
CO		 3563 break;
3564 case KVM_SET_MSRS:
3565 r = msr_io(vcpu, argp, do_set_msr, 0);
3566 break;
b209749f
AK		 3567 case KVM_TPR_ACCESS_REPORTING: {
3568 struct kvm_tpr_access_ctl tac;
3569
3570 r = -EFAULT;
3571 if (copy_from_user(&tac, argp, sizeof tac))
3572 goto out;
3573 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3574 if (r)
3575 goto out;
3576 r = -EFAULT;
3577 if (copy_to_user(argp, &tac, sizeof tac))
3578 goto out;
3579 r = 0;
3580 break;
3581 };
b93463aa
AK		 3582 case KVM_SET_VAPIC_ADDR: {
3583 struct kvm_vapic_addr va;
7301d6ab 3584 int idx;
b93463aa
AK		 3585
3586 r = -EINVAL;
35754c98 3587 if (!lapic_in_kernel(vcpu))
b93463aa
AK		 3588 goto out;
3589 r = -EFAULT;
3590 if (copy_from_user(&va, argp, sizeof va))
3591 goto out;
7301d6ab 3592 idx = srcu_read_lock(&vcpu->kvm->srcu);
fda4e2e8 3593 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
7301d6ab 3594 srcu_read_unlock(&vcpu->kvm->srcu, idx);
b93463aa
AK		 3595 break;
3596 }
890ca9ae
HY		 3597 case KVM_X86_SETUP_MCE: {
3598 u64 mcg_cap;
3599
3600 r = -EFAULT;
3601 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3602 goto out;
3603 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3604 break;
3605 }
3606 case KVM_X86_SET_MCE: {
3607 struct kvm_x86_mce mce;
3608
3609 r = -EFAULT;
3610 if (copy_from_user(&mce, argp, sizeof mce))
3611 goto out;
3612 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3613 break;
3614 }
3cfc3092
JK		 3615 case KVM_GET_VCPU_EVENTS: {
3616 struct kvm_vcpu_events events;
3617
3618 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3619
3620 r = -EFAULT;
3621 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3622 break;
3623 r = 0;
3624 break;
3625 }
3626 case KVM_SET_VCPU_EVENTS: {
3627 struct kvm_vcpu_events events;
3628
3629 r = -EFAULT;
3630 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3631 break;
3632
3633 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3634 break;
3635 }
a1efbe77
JK		 3636 case KVM_GET_DEBUGREGS: {
3637 struct kvm_debugregs dbgregs;
3638
3639 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3640
3641 r = -EFAULT;
3642 if (copy_to_user(argp, &dbgregs,
3643 sizeof(struct kvm_debugregs)))
3644 break;
3645 r = 0;
3646 break;
3647 }
3648 case KVM_SET_DEBUGREGS: {
3649 struct kvm_debugregs dbgregs;
3650
3651 r = -EFAULT;
3652 if (copy_from_user(&dbgregs, argp,
3653 sizeof(struct kvm_debugregs)))
3654 break;
3655
3656 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3657 break;
3658 }
2d5b5a66 3659 case KVM_GET_XSAVE: {
d1ac91d8 3660 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2d5b5a66 3661 r = -ENOMEM;
d1ac91d8 3662 if (!u.xsave)
2d5b5a66
SY		 3663 break;
3664
d1ac91d8 3665 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
2d5b5a66
SY		 3666
3667 r = -EFAULT;
d1ac91d8 3668 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
2d5b5a66
SY		 3669 break;
3670 r = 0;
3671 break;
3672 }
3673 case KVM_SET_XSAVE: {
ff5c2c03 3674 u.xsave = memdup_user(argp, sizeof(*u.xsave));
18595411
GC		 3675 if (IS_ERR(u.xsave))
3676 return PTR_ERR(u.xsave);
2d5b5a66 3677
d1ac91d8 3678 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
2d5b5a66
SY		 3679 break;
3680 }
3681 case KVM_GET_XCRS: {
d1ac91d8 3682 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2d5b5a66 3683 r = -ENOMEM;
d1ac91d8 3684 if (!u.xcrs)
2d5b5a66
SY		 3685 break;
3686
d1ac91d8 3687 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
2d5b5a66
SY		 3688
3689 r = -EFAULT;
d1ac91d8 3690 if (copy_to_user(argp, u.xcrs,
2d5b5a66
SY		 3691 sizeof(struct kvm_xcrs)))
3692 break;
3693 r = 0;
3694 break;
3695 }
3696 case KVM_SET_XCRS: {
ff5c2c03 3697 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
18595411
GC		 3698 if (IS_ERR(u.xcrs))
3699 return PTR_ERR(u.xcrs);
2d5b5a66 3700
d1ac91d8 3701 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
2d5b5a66
SY		 3702 break;
3703 }
92a1f12d
JR		 3704 case KVM_SET_TSC_KHZ: {
3705 u32 user_tsc_khz;
3706
3707 r = -EINVAL;
92a1f12d
JR		 3708 user_tsc_khz = (u32)arg;
3709
3710 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3711 goto out;
3712
cc578287
ZA		 3713 if (user_tsc_khz == 0)
3714 user_tsc_khz = tsc_khz;
3715
381d585c
HZ		 3716 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3717 r = 0;
92a1f12d 3718
92a1f12d
JR		 3719 goto out;
3720 }
3721 case KVM_GET_TSC_KHZ: {
cc578287 3722 r = vcpu->arch.virtual_tsc_khz;
92a1f12d
JR		 3723 goto out;
3724 }
1c0b28c2
EM		 3725 case KVM_KVMCLOCK_CTRL: {
3726 r = kvm_set_guest_paused(vcpu);
3727 goto out;
3728 }
5c919412
AS		 3729 case KVM_ENABLE_CAP: {
3730 struct kvm_enable_cap cap;
3731
3732 r = -EFAULT;
3733 if (copy_from_user(&cap, argp, sizeof(cap)))
3734 goto out;
3735 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3736 break;
3737 }
313a3dc7
CO		 3738 default:
3739 r = -EINVAL;
3740 }
3741out:
d1ac91d8 3742 kfree(u.buffer);
313a3dc7
CO		 3743 return r;
3744}
3745
5b1c1493
CO		 3746int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3747{
3748 return VM_FAULT_SIGBUS;
3749}
3750
1fe779f8
CO		 3751static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3752{
3753 int ret;
3754
3755 if (addr > (unsigned int)(-3 * PAGE_SIZE))
951179ce 3756 return -EINVAL;
1fe779f8
CO		 3757 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3758 return ret;
3759}
3760
b927a3ce
SY		 3761static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3762 u64 ident_addr)
3763{
3764 kvm->arch.ept_identity_map_addr = ident_addr;
3765 return 0;
3766}
3767
1fe779f8
CO		 3768static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3769 u32 kvm_nr_mmu_pages)
3770{
3771 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3772 return -EINVAL;
3773
79fac95e 3774 mutex_lock(&kvm->slots_lock);
1fe779f8
CO		 3775
3776 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
f05e70ac 3777 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1fe779f8 3778
79fac95e 3779 mutex_unlock(&kvm->slots_lock);
1fe779f8
CO		 3780 return 0;
3781}
3782
3783static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3784{
39de71ec 3785 return kvm->arch.n_max_mmu_pages;
1fe779f8
CO		 3786}
3787
1fe779f8
CO		 3788static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3789{
90bca052 3790 struct kvm_pic *pic = kvm->arch.vpic;
1fe779f8
CO		 3791 int r;
3792
3793 r = 0;
3794 switch (chip->chip_id) {
3795 case KVM_IRQCHIP_PIC_MASTER:
90bca052 3796 memcpy(&chip->chip.pic, &pic->pics[0],
1fe779f8
CO		 3797 sizeof(struct kvm_pic_state));
3798 break;
3799 case KVM_IRQCHIP_PIC_SLAVE:
90bca052 3800 memcpy(&chip->chip.pic, &pic->pics[1],
1fe779f8
CO		 3801 sizeof(struct kvm_pic_state));
3802 break;
3803 case KVM_IRQCHIP_IOAPIC:
33392b49 3804 kvm_get_ioapic(kvm, &chip->chip.ioapic);
1fe779f8
CO		 3805 break;
3806 default:
3807 r = -EINVAL;
3808 break;
3809 }
3810 return r;
3811}
3812
3813static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3814{
90bca052 3815 struct kvm_pic *pic = kvm->arch.vpic;
1fe779f8
CO		 3816 int r;
3817
3818 r = 0;
3819 switch (chip->chip_id) {
3820 case KVM_IRQCHIP_PIC_MASTER:
90bca052
DH		 3821 spin_lock(&pic->lock);
3822 memcpy(&pic->pics[0], &chip->chip.pic,
1fe779f8 3823 sizeof(struct kvm_pic_state));
90bca052 3824 spin_unlock(&pic->lock);
1fe779f8
CO		 3825 break;
3826 case KVM_IRQCHIP_PIC_SLAVE:
90bca052
DH		 3827 spin_lock(&pic->lock);
3828 memcpy(&pic->pics[1], &chip->chip.pic,
1fe779f8 3829 sizeof(struct kvm_pic_state));
90bca052 3830 spin_unlock(&pic->lock);
1fe779f8
CO		 3831 break;
3832 case KVM_IRQCHIP_IOAPIC:
33392b49 3833 kvm_set_ioapic(kvm, &chip->chip.ioapic);
1fe779f8
CO		 3834 break;
3835 default:
3836 r = -EINVAL;
3837 break;
3838 }
90bca052 3839 kvm_pic_update_irq(pic);
1fe779f8
CO		 3840 return r;
3841}
3842
e0f63cb9
SY		 3843static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3844{
34f3941c
RK		 3845 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3846
3847 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3848
3849 mutex_lock(&kps->lock);
3850 memcpy(ps, &kps->channels, sizeof(*ps));
3851 mutex_unlock(&kps->lock);
2da29bcc 3852 return 0;
e0f63cb9
SY		 3853}
3854
3855static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3856{
0185604c 3857 int i;
09edea72
RK		 3858 struct kvm_pit *pit = kvm->arch.vpit;
3859
3860 mutex_lock(&pit->pit_state.lock);
34f3941c 3861 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
0185604c 3862 for (i = 0; i < 3; i++)
09edea72
RK		 3863 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3864 mutex_unlock(&pit->pit_state.lock);
2da29bcc 3865 return 0;
e9f42757
BK		 3866}
3867
3868static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3869{
e9f42757
BK		 3870 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3871 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3872 sizeof(ps->channels));
3873 ps->flags = kvm->arch.vpit->pit_state.flags;
3874 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
97e69aa6 3875 memset(&ps->reserved, 0, sizeof(ps->reserved));
2da29bcc 3876 return 0;
e9f42757
BK		 3877}
3878
3879static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3880{
2da29bcc 3881 int start = 0;
0185604c 3882 int i;
e9f42757 3883 u32 prev_legacy, cur_legacy;
09edea72
RK		 3884 struct kvm_pit *pit = kvm->arch.vpit;
3885
3886 mutex_lock(&pit->pit_state.lock);
3887 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
e9f42757
BK		 3888 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3889 if (!prev_legacy && cur_legacy)
3890 start = 1;
09edea72
RK		 3891 memcpy(&pit->pit_state.channels, &ps->channels,
3892 sizeof(pit->pit_state.channels));
3893 pit->pit_state.flags = ps->flags;
0185604c 3894 for (i = 0; i < 3; i++)
09edea72 3895 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
e5e57e7a 3896 start && i == 0);
09edea72 3897 mutex_unlock(&pit->pit_state.lock);
2da29bcc 3898 return 0;
e0f63cb9
SY		 3899}
3900
52d939a0
MT		 3901static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3902 struct kvm_reinject_control *control)
3903{
71474e2f
RK		 3904 struct kvm_pit *pit = kvm->arch.vpit;
3905
3906 if (!pit)
52d939a0 3907 return -ENXIO;
b39c90b6 3908
71474e2f
RK		 3909 /* pit->pit_state.lock was overloaded to prevent userspace from getting
3910 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3911 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
3912 */
3913 mutex_lock(&pit->pit_state.lock);
3914 kvm_pit_set_reinject(pit, control->pit_reinject);
3915 mutex_unlock(&pit->pit_state.lock);
b39c90b6 3916
52d939a0
MT		 3917 return 0;
3918}
3919
95d4c16c 3920/**
60c34612
TY		 3921 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3922 * @kvm: kvm instance
3923 * @log: slot id and address to which we copy the log
95d4c16c 3924 *
e108ff2f
PB		 3925 * Steps 1-4 below provide general overview of dirty page logging. See
3926 * kvm_get_dirty_log_protect() function description for additional details.
3927 *
3928 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3929 * always flush the TLB (step 4) even if previous step failed and the dirty
3930 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3931 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3932 * writes will be marked dirty for next log read.
95d4c16c 3933 *
60c34612
TY		 3934 * 1. Take a snapshot of the bit and clear it if needed.
3935 * 2. Write protect the corresponding page.
e108ff2f
PB		 3936 * 3. Copy the snapshot to the userspace.
3937 * 4. Flush TLB's if needed.
5bb064dc 3938 */
60c34612 3939int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
5bb064dc 3940{
60c34612 3941 bool is_dirty = false;
e108ff2f 3942 int r;
5bb064dc 3943
79fac95e 3944 mutex_lock(&kvm->slots_lock);
5bb064dc 3945
88178fd4
KH		 3946 /*
3947 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3948 */
3949 if (kvm_x86_ops->flush_log_dirty)
3950 kvm_x86_ops->flush_log_dirty(kvm);
3951
e108ff2f 3952 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
198c74f4
XG		 3953
3954 /*
3955 * All the TLBs can be flushed out of mmu lock, see the comments in
3956 * kvm_mmu_slot_remove_write_access().
3957 */
e108ff2f 3958 lockdep_assert_held(&kvm->slots_lock);
198c74f4
XG		 3959 if (is_dirty)
3960 kvm_flush_remote_tlbs(kvm);
3961
79fac95e 3962 mutex_unlock(&kvm->slots_lock);
5bb064dc
ZX		 3963 return r;
3964}
3965
aa2fbe6d
YZ		 3966int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3967 bool line_status)
23d43cf9
CD		 3968{
3969 if (!irqchip_in_kernel(kvm))
3970 return -ENXIO;
3971
3972 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
aa2fbe6d
YZ		 3973 irq_event->irq, irq_event->level,
3974 line_status);
23d43cf9
CD		 3975 return 0;
3976}
3977
90de4a18
NA		 3978static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3979 struct kvm_enable_cap *cap)
3980{
3981 int r;
3982
3983 if (cap->flags)
3984 return -EINVAL;
3985
3986 switch (cap->cap) {
3987 case KVM_CAP_DISABLE_QUIRKS:
3988 kvm->arch.disabled_quirks = cap->args[0];
3989 r = 0;
3990 break;
49df6397
SR		 3991 case KVM_CAP_SPLIT_IRQCHIP: {
3992 mutex_lock(&kvm->lock);
b053b2ae
SR		 3993 r = -EINVAL;
3994 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3995 goto split_irqchip_unlock;
49df6397
SR		 3996 r = -EEXIST;
3997 if (irqchip_in_kernel(kvm))
3998 goto split_irqchip_unlock;
557abc40 3999 if (kvm->created_vcpus)
49df6397
SR		 4000 goto split_irqchip_unlock;
4001 r = kvm_setup_empty_irq_routing(kvm);
5c0aea0e 4002 if (r)
49df6397
SR		 4003 goto split_irqchip_unlock;
4004 /* Pairs with irqchip_in_kernel. */
4005 smp_wmb();
49776faf 4006 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
b053b2ae 4007 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
49df6397
SR		 4008 r = 0;
4009split_irqchip_unlock:
4010 mutex_unlock(&kvm->lock);
4011 break;
4012 }
37131313
RK		 4013 case KVM_CAP_X2APIC_API:
4014 r = -EINVAL;
4015 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4016 break;
4017
4018 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4019 kvm->arch.x2apic_format = true;
c519265f
RK		 4020 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4021 kvm->arch.x2apic_broadcast_quirk_disabled = true;
37131313
RK		 4022
4023 r = 0;
4024 break;
90de4a18
NA		 4025 default:
4026 r = -EINVAL;
4027 break;
4028 }
4029 return r;
4030}
4031
1fe779f8
CO		 4032long kvm_arch_vm_ioctl(struct file *filp,
4033 unsigned int ioctl, unsigned long arg)
4034{
4035 struct kvm *kvm = filp->private_data;
4036 void __user *argp = (void __user *)arg;
367e1319 4037 int r = -ENOTTY;
f0d66275
DH		 4038 /*
4039 * This union makes it completely explicit to gcc-3.x
4040 * that these two variables' stack usage should be
4041 * combined, not added together.
4042 */
4043 union {
4044 struct kvm_pit_state ps;
e9f42757 4045 struct kvm_pit_state2 ps2;
c5ff41ce 4046 struct kvm_pit_config pit_config;
f0d66275 4047 } u;
1fe779f8
CO		 4048
4049 switch (ioctl) {
4050 case KVM_SET_TSS_ADDR:
4051 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1fe779f8 4052 break;
b927a3ce
SY		 4053 case KVM_SET_IDENTITY_MAP_ADDR: {
4054 u64 ident_addr;
4055
1af1ac91
DH		 4056 mutex_lock(&kvm->lock);
4057 r = -EINVAL;
4058 if (kvm->created_vcpus)
4059 goto set_identity_unlock;
b927a3ce
SY		 4060 r = -EFAULT;
4061 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
1af1ac91 4062 goto set_identity_unlock;
b927a3ce 4063 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
1af1ac91
DH		 4064set_identity_unlock:
4065 mutex_unlock(&kvm->lock);
b927a3ce
SY		 4066 break;
4067 }
1fe779f8
CO		 4068 case KVM_SET_NR_MMU_PAGES:
4069 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1fe779f8
CO		 4070 break;
4071 case KVM_GET_NR_MMU_PAGES:
4072 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4073 break;
3ddea128 4074 case KVM_CREATE_IRQCHIP: {
3ddea128 4075 mutex_lock(&kvm->lock);
09941366 4076
3ddea128 4077 r = -EEXIST;
35e6eaa3 4078 if (irqchip_in_kernel(kvm))
3ddea128 4079 goto create_irqchip_unlock;
09941366 4080
3e515705 4081 r = -EINVAL;
557abc40 4082 if (kvm->created_vcpus)
3e515705 4083 goto create_irqchip_unlock;
09941366
RK		 4084
4085 r = kvm_pic_init(kvm);
4086 if (r)
3ddea128 4087 goto create_irqchip_unlock;
09941366
RK		 4088
4089 r = kvm_ioapic_init(kvm);
4090 if (r) {
09941366 4091 kvm_pic_destroy(kvm);
3ddea128 4092 goto create_irqchip_unlock;
09941366
RK		 4093 }
4094
399ec807
AK		 4095 r = kvm_setup_default_irq_routing(kvm);
4096 if (r) {
72bb2fcd 4097 kvm_ioapic_destroy(kvm);
09941366 4098 kvm_pic_destroy(kvm);
71ba994c 4099 goto create_irqchip_unlock;
399ec807 4100 }
49776faf 4101 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
71ba994c 4102 smp_wmb();
49776faf 4103 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
3ddea128
MT		 4104 create_irqchip_unlock:
4105 mutex_unlock(&kvm->lock);
1fe779f8 4106 break;
3ddea128 4107 }
7837699f 4108 case KVM_CREATE_PIT:
c5ff41ce
JK		 4109 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4110 goto create_pit;
4111 case KVM_CREATE_PIT2:
4112 r = -EFAULT;
4113 if (copy_from_user(&u.pit_config, argp,
4114 sizeof(struct kvm_pit_config)))
4115 goto out;
4116 create_pit:
250715a6 4117 mutex_lock(&kvm->lock);
269e05e4
AK		 4118 r = -EEXIST;
4119 if (kvm->arch.vpit)
4120 goto create_pit_unlock;
7837699f 4121 r = -ENOMEM;
c5ff41ce 4122 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
7837699f
SY		 4123 if (kvm->arch.vpit)
4124 r = 0;
269e05e4 4125 create_pit_unlock:
250715a6 4126 mutex_unlock(&kvm->lock);
7837699f 4127 break;
1fe779f8
CO		 4128 case KVM_GET_IRQCHIP: {
4129 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
ff5c2c03 4130 struct kvm_irqchip *chip;
1fe779f8 4131
ff5c2c03
SL		 4132 chip = memdup_user(argp, sizeof(*chip));
4133 if (IS_ERR(chip)) {
4134 r = PTR_ERR(chip);
1fe779f8 4135 goto out;
ff5c2c03
SL		 4136 }
4137
1fe779f8 4138 r = -ENXIO;
826da321 4139 if (!irqchip_kernel(kvm))
f0d66275
DH		 4140 goto get_irqchip_out;
4141 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1fe779f8 4142 if (r)
f0d66275 4143 goto get_irqchip_out;
1fe779f8 4144 r = -EFAULT;
f0d66275
DH		 4145 if (copy_to_user(argp, chip, sizeof *chip))
4146 goto get_irqchip_out;
1fe779f8 4147 r = 0;
f0d66275
DH		 4148 get_irqchip_out:
4149 kfree(chip);
1fe779f8
CO		 4150 break;
4151 }
4152 case KVM_SET_IRQCHIP: {
4153 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
ff5c2c03 4154 struct kvm_irqchip *chip;
1fe779f8 4155
ff5c2c03
SL		 4156 chip = memdup_user(argp, sizeof(*chip));
4157 if (IS_ERR(chip)) {
4158 r = PTR_ERR(chip);
1fe779f8 4159 goto out;
ff5c2c03
SL		 4160 }
4161
1fe779f8 4162 r = -ENXIO;
826da321 4163 if (!irqchip_kernel(kvm))
f0d66275
DH		 4164 goto set_irqchip_out;
4165 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1fe779f8 4166 if (r)
f0d66275 4167 goto set_irqchip_out;
1fe779f8 4168 r = 0;
f0d66275
DH		 4169 set_irqchip_out:
4170 kfree(chip);
1fe779f8
CO		 4171 break;
4172 }
e0f63cb9 4173 case KVM_GET_PIT: {
e0f63cb9 4174 r = -EFAULT;
f0d66275 4175 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
e0f63cb9
SY		 4176 goto out;
4177 r = -ENXIO;
4178 if (!kvm->arch.vpit)
4179 goto out;
f0d66275 4180 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
e0f63cb9
SY		 4181 if (r)
4182 goto out;
4183 r = -EFAULT;
f0d66275 4184 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
e0f63cb9
SY		 4185 goto out;
4186 r = 0;
4187 break;
4188 }
4189 case KVM_SET_PIT: {
e0f63cb9 4190 r = -EFAULT;
f0d66275 4191 if (copy_from_user(&u.ps, argp, sizeof u.ps))
e0f63cb9
SY		 4192 goto out;
4193 r = -ENXIO;
4194 if (!kvm->arch.vpit)
4195 goto out;
f0d66275 4196 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
e0f63cb9
SY		 4197 break;
4198 }
e9f42757
BK		 4199 case KVM_GET_PIT2: {
4200 r = -ENXIO;
4201 if (!kvm->arch.vpit)
4202 goto out;
4203 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4204 if (r)
4205 goto out;
4206 r = -EFAULT;
4207 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4208 goto out;
4209 r = 0;
4210 break;
4211 }
4212 case KVM_SET_PIT2: {
4213 r = -EFAULT;
4214 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4215 goto out;
4216 r = -ENXIO;
4217 if (!kvm->arch.vpit)
4218 goto out;
4219 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
e9f42757
BK		 4220 break;
4221 }
52d939a0
MT		 4222 case KVM_REINJECT_CONTROL: {
4223 struct kvm_reinject_control control;
4224 r = -EFAULT;
4225 if (copy_from_user(&control, argp, sizeof(control)))
4226 goto out;
4227 r = kvm_vm_ioctl_reinject(kvm, &control);
52d939a0
MT		 4228 break;
4229 }
d71ba788
PB		 4230 case KVM_SET_BOOT_CPU_ID:
4231 r = 0;
4232 mutex_lock(&kvm->lock);
557abc40 4233 if (kvm->created_vcpus)
d71ba788
PB		 4234 r = -EBUSY;
4235 else
4236 kvm->arch.bsp_vcpu_id = arg;
4237 mutex_unlock(&kvm->lock);
4238 break;
ffde22ac
ES		 4239 case KVM_XEN_HVM_CONFIG: {
4240 r = -EFAULT;
4241 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4242 sizeof(struct kvm_xen_hvm_config)))
4243 goto out;
4244 r = -EINVAL;
4245 if (kvm->arch.xen_hvm_config.flags)
4246 goto out;
4247 r = 0;
4248 break;
4249 }
afbcf7ab 4250 case KVM_SET_CLOCK: {
afbcf7ab
GC		 4251 struct kvm_clock_data user_ns;
4252 u64 now_ns;
afbcf7ab
GC		 4253
4254 r = -EFAULT;
4255 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4256 goto out;
4257
4258 r = -EINVAL;
4259 if (user_ns.flags)
4260 goto out;
4261
4262 r = 0;
0bc48bea
RK		 4263 /*
4264 * TODO: userspace has to take care of races with VCPU_RUN, so
4265 * kvm_gen_update_masterclock() can be cut down to locked
4266 * pvclock_update_vm_gtod_copy().
4267 */
4268 kvm_gen_update_masterclock(kvm);
e891a32e 4269 now_ns = get_kvmclock_ns(kvm);
108b249c 4270 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
0bc48bea 4271 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
afbcf7ab
GC		 4272 break;
4273 }
4274 case KVM_GET_CLOCK: {
afbcf7ab
GC		 4275 struct kvm_clock_data user_ns;
4276 u64 now_ns;
4277
e891a32e 4278 now_ns = get_kvmclock_ns(kvm);
108b249c 4279 user_ns.clock = now_ns;
e3fd9a93 4280 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
97e69aa6 4281 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
afbcf7ab
GC		 4282
4283 r = -EFAULT;
4284 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4285 goto out;
4286 r = 0;
4287 break;
4288 }
90de4a18
NA		 4289 case KVM_ENABLE_CAP: {
4290 struct kvm_enable_cap cap;
afbcf7ab 4291
90de4a18
NA		 4292 r = -EFAULT;
4293 if (copy_from_user(&cap, argp, sizeof(cap)))
4294 goto out;
4295 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4296 break;
4297 }
1fe779f8 4298 default:
ad6260da 4299 r = -ENOTTY;
1fe779f8
CO		 4300 }
4301out:
4302 return r;
4303}
4304
a16b043c 4305static void kvm_init_msr_list(void)
043405e1
CO		 4306{
4307 u32 dummy[2];
4308 unsigned i, j;
4309
62ef68bb 4310 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
043405e1
CO		 4311 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4312 continue;
93c4adc7
PB		 4313
4314 /*
4315 * Even MSRs that are valid in the host may not be exposed
9dbe6cf9 4316 * to the guests in some cases.
93c4adc7
PB		 4317 */
4318 switch (msrs_to_save[i]) {
4319 case MSR_IA32_BNDCFGS:
4320 if (!kvm_x86_ops->mpx_supported())
4321 continue;
4322 break;
9dbe6cf9
PB		 4323 case MSR_TSC_AUX:
4324 if (!kvm_x86_ops->rdtscp_supported())
4325 continue;
4326 break;
93c4adc7
PB		 4327 default:
4328 break;
4329 }
4330
043405e1
CO		 4331 if (j < i)
4332 msrs_to_save[j] = msrs_to_save[i];
4333 j++;
4334 }
4335 num_msrs_to_save = j;
62ef68bb
PB		 4336
4337 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4338 switch (emulated_msrs[i]) {
6d396b55
PB		 4339 case MSR_IA32_SMBASE:
4340 if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4341 continue;
4342 break;
62ef68bb
PB		 4343 default:
4344 break;
4345 }
4346
4347 if (j < i)
4348 emulated_msrs[j] = emulated_msrs[i];
4349 j++;
4350 }
4351 num_emulated_msrs = j;
043405e1
CO		 4352}
4353
bda9020e
MT		 4354static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4355 const void *v)
bbd9b64e 4356{
70252a10
AK		 4357 int handled = 0;
4358 int n;
4359
4360 do {
4361 n = min(len, 8);
bce87cce 4362 if (!(lapic_in_kernel(vcpu) &&
e32edf4f
NN		 4363 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4364 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
70252a10
AK		 4365 break;
4366 handled += n;
4367 addr += n;
4368 len -= n;
4369 v += n;
4370 } while (len);
bbd9b64e 4371
70252a10 4372 return handled;
bbd9b64e
CO		 4373}
4374
bda9020e 4375static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
bbd9b64e 4376{
70252a10
AK		 4377 int handled = 0;
4378 int n;
4379
4380 do {
4381 n = min(len, 8);
bce87cce 4382 if (!(lapic_in_kernel(vcpu) &&
e32edf4f
NN		 4383 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4384 addr, n, v))
4385 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
70252a10 4386 break;
e39d200f 4387 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
70252a10
AK		 4388 handled += n;
4389 addr += n;
4390 len -= n;
4391 v += n;
4392 } while (len);
bbd9b64e 4393
70252a10 4394 return handled;
bbd9b64e
CO		 4395}
4396
2dafc6c2
GN		 4397static void kvm_set_segment(struct kvm_vcpu *vcpu,
4398 struct kvm_segment *var, int seg)
4399{
4400 kvm_x86_ops->set_segment(vcpu, var, seg);
4401}
4402
4403void kvm_get_segment(struct kvm_vcpu *vcpu,
4404 struct kvm_segment *var, int seg)
4405{
4406 kvm_x86_ops->get_segment(vcpu, var, seg);
4407}
4408
54987b7a
PB		 4409gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4410 struct x86_exception *exception)
02f59dc9
JR		 4411{
4412 gpa_t t_gpa;
02f59dc9
JR		 4413
4414 BUG_ON(!mmu_is_nested(vcpu));
4415
4416 /* NPT walks are always user-walks */
4417 access |= PFERR_USER_MASK;
54987b7a 4418 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
02f59dc9
JR		 4419
4420 return t_gpa;
4421}
4422
ab9ae313
AK		 4423gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4424 struct x86_exception *exception)
1871c602
GN		 4425{
4426 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
ab9ae313 4427 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 4428}
4429
ab9ae313
AK		 4430 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4431 struct x86_exception *exception)
1871c602
GN		 4432{
4433 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4434 access |= PFERR_FETCH_MASK;
ab9ae313 4435 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 4436}
4437
ab9ae313
AK		 4438gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4439 struct x86_exception *exception)
1871c602
GN		 4440{
4441 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4442 access |= PFERR_WRITE_MASK;
ab9ae313 4443 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 4444}
4445
4446/* uses this to access any guest's mapped memory without checking CPL */
ab9ae313
AK		 4447gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4448 struct x86_exception *exception)
1871c602 4449{
ab9ae313 4450 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
1871c602
GN		 4451}
4452
4453static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4454 struct kvm_vcpu *vcpu, u32 access,
bcc55cba 4455 struct x86_exception *exception)
bbd9b64e
CO		 4456{
4457 void *data = val;
10589a46 4458 int r = X86EMUL_CONTINUE;
bbd9b64e
CO		 4459
4460 while (bytes) {
14dfe855 4461 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
ab9ae313 4462 exception);
bbd9b64e 4463 unsigned offset = addr & (PAGE_SIZE-1);
77c2002e 4464 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
bbd9b64e
CO		 4465 int ret;
4466
bcc55cba 4467 if (gpa == UNMAPPED_GVA)
ab9ae313 4468 return X86EMUL_PROPAGATE_FAULT;
54bf36aa
PB		 4469 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4470 offset, toread);
10589a46 4471 if (ret < 0) {
c3cd7ffa 4472 r = X86EMUL_IO_NEEDED;
10589a46
MT		 4473 goto out;
4474 }
bbd9b64e 4475
77c2002e
IE		 4476 bytes -= toread;
4477 data += toread;
4478 addr += toread;
bbd9b64e 4479 }
10589a46 4480out:
10589a46 4481 return r;
bbd9b64e 4482}
77c2002e 4483
1871c602 4484/* used for instruction fetching */
0f65dd70
AK		 4485static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4486 gva_t addr, void *val, unsigned int bytes,
bcc55cba 4487 struct x86_exception *exception)
1871c602 4488{
0f65dd70 4489 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
1871c602 4490 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
44583cba
PB		 4491 unsigned offset;
4492 int ret;
0f65dd70 4493
44583cba
PB		 4494 /* Inline kvm_read_guest_virt_helper for speed. */
4495 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4496 exception);
4497 if (unlikely(gpa == UNMAPPED_GVA))
4498 return X86EMUL_PROPAGATE_FAULT;
4499
4500 offset = addr & (PAGE_SIZE-1);
4501 if (WARN_ON(offset + bytes > PAGE_SIZE))
4502 bytes = (unsigned)PAGE_SIZE - offset;
54bf36aa
PB		 4503 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4504 offset, bytes);
44583cba
PB		 4505 if (unlikely(ret < 0))
4506 return X86EMUL_IO_NEEDED;
4507
4508 return X86EMUL_CONTINUE;
1871c602
GN		 4509}
4510
064aea77 4511int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
0f65dd70 4512 gva_t addr, void *val, unsigned int bytes,
bcc55cba 4513 struct x86_exception *exception)
1871c602 4514{
0f65dd70 4515 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
1871c602 4516 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
0f65dd70 4517
1871c602 4518 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
bcc55cba 4519 exception);
1871c602 4520}
064aea77 4521EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
1871c602 4522
0f65dd70
AK		 4523static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4524 gva_t addr, void *val, unsigned int bytes,
bcc55cba 4525 struct x86_exception *exception)
1871c602 4526{
0f65dd70 4527 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
bcc55cba 4528 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
1871c602
GN		 4529}
4530
7a036a6f
RK		 4531static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4532 unsigned long addr, void *val, unsigned int bytes)
4533{
4534 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4535 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4536
4537 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4538}
4539
6a4d7550 4540int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
0f65dd70 4541 gva_t addr, void *val,
2dafc6c2 4542 unsigned int bytes,
bcc55cba 4543 struct x86_exception *exception)
77c2002e 4544{
0f65dd70 4545 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
77c2002e
IE		 4546 void *data = val;
4547 int r = X86EMUL_CONTINUE;
4548
4549 while (bytes) {
14dfe855
JR		 4550 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4551 PFERR_WRITE_MASK,
ab9ae313 4552 exception);
77c2002e
IE		 4553 unsigned offset = addr & (PAGE_SIZE-1);
4554 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4555 int ret;
4556
bcc55cba 4557 if (gpa == UNMAPPED_GVA)
ab9ae313 4558 return X86EMUL_PROPAGATE_FAULT;
54bf36aa 4559 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
77c2002e 4560 if (ret < 0) {
c3cd7ffa 4561 r = X86EMUL_IO_NEEDED;
77c2002e
IE		 4562 goto out;
4563 }
4564
4565 bytes -= towrite;
4566 data += towrite;
4567 addr += towrite;
4568 }
4569out:
4570 return r;
4571}
6a4d7550 4572EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
77c2002e 4573
0f89b207
TL		 4574static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4575 gpa_t gpa, bool write)
4576{
4577 /* For APIC access vmexit */
4578 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4579 return 1;
4580
4581 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
4582 trace_vcpu_match_mmio(gva, gpa, write, true);
4583 return 1;
4584 }
4585
4586 return 0;
4587}
4588
af7cc7d1
XG		 4589static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4590 gpa_t *gpa, struct x86_exception *exception,
4591 bool write)
4592{
97d64b78
AK		 4593 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4594 | (write ? PFERR_WRITE_MASK : 0);
af7cc7d1 4595
be94f6b7
HH		 4596 /*
4597 * currently PKRU is only applied to ept enabled guest so
4598 * there is no pkey in EPT page table for L1 guest or EPT
4599 * shadow page table for L2 guest.
4600 */
97d64b78 4601 if (vcpu_match_mmio_gva(vcpu, gva)
97ec8c06 4602 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
be94f6b7 4603 vcpu->arch.access, 0, access)) {
bebb106a
XG		 4604 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4605 (gva & (PAGE_SIZE - 1));
4f022648 4606 trace_vcpu_match_mmio(gva, *gpa, write, false);
bebb106a
XG		 4607 return 1;
4608 }
4609
af7cc7d1
XG		 4610 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4611
4612 if (*gpa == UNMAPPED_GVA)
4613 return -1;
4614
0f89b207 4615 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
af7cc7d1
XG		 4616}
4617
3200f405 4618int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
bcc55cba 4619 const void *val, int bytes)
bbd9b64e
CO		 4620{
4621 int ret;
4622
54bf36aa 4623 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
9f811285 4624 if (ret < 0)
bbd9b64e 4625 return 0;
0eb05bf2 4626 kvm_page_track_write(vcpu, gpa, val, bytes);
bbd9b64e
CO		 4627 return 1;
4628}
4629
77d197b2
XG		 4630struct read_write_emulator_ops {
4631 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4632 int bytes);
4633 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4634 void *val, int bytes);
4635 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4636 int bytes, void *val);
4637 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4638 void *val, int bytes);
4639 bool write;
4640};
4641
4642static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4643{
4644 if (vcpu->mmio_read_completed) {
77d197b2 4645 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
e39d200f 4646 vcpu->mmio_fragments[0].gpa, val);
77d197b2
XG		 4647 vcpu->mmio_read_completed = 0;
4648 return 1;
4649 }
4650
4651 return 0;
4652}
4653
4654static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4655 void *val, int bytes)
4656{
54bf36aa 4657 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
77d197b2
XG		 4658}
4659
4660static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4661 void *val, int bytes)
4662{
4663 return emulator_write_phys(vcpu, gpa, val, bytes);
4664}
4665
4666static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4667{
e39d200f 4668 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
77d197b2
XG		 4669 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4670}
4671
4672static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4673 void *val, int bytes)
4674{
e39d200f 4675 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
77d197b2
XG		 4676 return X86EMUL_IO_NEEDED;
4677}
4678
4679static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4680 void *val, int bytes)
4681{
f78146b0
AK		 4682 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4683
87da7e66 4684 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
77d197b2
XG		 4685 return X86EMUL_CONTINUE;
4686}
4687
0fbe9b0b 4688static const struct read_write_emulator_ops read_emultor = {
77d197b2
XG		 4689 .read_write_prepare = read_prepare,
4690 .read_write_emulate = read_emulate,
4691 .read_write_mmio = vcpu_mmio_read,
4692 .read_write_exit_mmio = read_exit_mmio,
4693};
4694
0fbe9b0b 4695static const struct read_write_emulator_ops write_emultor = {
77d197b2
XG		 4696 .read_write_emulate = write_emulate,
4697 .read_write_mmio = write_mmio,
4698 .read_write_exit_mmio = write_exit_mmio,
4699 .write = true,
4700};
4701
22388a3c
XG		 4702static int emulator_read_write_onepage(unsigned long addr, void *val,
4703 unsigned int bytes,
4704 struct x86_exception *exception,
4705 struct kvm_vcpu *vcpu,
0fbe9b0b 4706 const struct read_write_emulator_ops *ops)
bbd9b64e 4707{
af7cc7d1
XG		 4708 gpa_t gpa;
4709 int handled, ret;
22388a3c 4710 bool write = ops->write;
f78146b0 4711 struct kvm_mmio_fragment *frag;
0f89b207
TL		 4712 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4713
4714 /*
4715 * If the exit was due to a NPF we may already have a GPA.
4716 * If the GPA is present, use it to avoid the GVA to GPA table walk.
4717 * Note, this cannot be used on string operations since string
4718 * operation using rep will only have the initial GPA from the NPF
4719 * occurred.
4720 */
4721 if (vcpu->arch.gpa_available &&
4722 emulator_can_use_gpa(ctxt) &&
618232e2
BS		 4723 (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
4724 gpa = vcpu->arch.gpa_val;
4725 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
4726 } else {
4727 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4728 if (ret < 0)
4729 return X86EMUL_PROPAGATE_FAULT;
0f89b207 4730 }
10589a46 4731
618232e2 4732 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
bbd9b64e
CO		 4733 return X86EMUL_CONTINUE;
4734
bbd9b64e
CO		 4735 /*
4736 * Is this MMIO handled locally?
4737 */
22388a3c 4738 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
70252a10 4739 if (handled == bytes)
bbd9b64e 4740 return X86EMUL_CONTINUE;
bbd9b64e 4741
70252a10
AK		 4742 gpa += handled;
4743 bytes -= handled;
4744 val += handled;
4745
87da7e66
XG		 4746 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4747 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4748 frag->gpa = gpa;
4749 frag->data = val;
4750 frag->len = bytes;
f78146b0 4751 return X86EMUL_CONTINUE;
bbd9b64e
CO		 4752}
4753
52eb5a6d
XL		 4754static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4755 unsigned long addr,
22388a3c
XG		 4756 void *val, unsigned int bytes,
4757 struct x86_exception *exception,
0fbe9b0b 4758 const struct read_write_emulator_ops *ops)
bbd9b64e 4759{
0f65dd70 4760 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
f78146b0
AK		 4761 gpa_t gpa;
4762 int rc;
4763
4764 if (ops->read_write_prepare &&
4765 ops->read_write_prepare(vcpu, val, bytes))
4766 return X86EMUL_CONTINUE;
4767
4768 vcpu->mmio_nr_fragments = 0;
0f65dd70 4769
bbd9b64e
CO		 4770 /* Crossing a page boundary? */
4771 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
f78146b0 4772 int now;
bbd9b64e
CO		 4773
4774 now = -addr & ~PAGE_MASK;
22388a3c
XG		 4775 rc = emulator_read_write_onepage(addr, val, now, exception,
4776 vcpu, ops);
4777
bbd9b64e
CO		 4778 if (rc != X86EMUL_CONTINUE)
4779 return rc;
4780 addr += now;
bac15531
NA		 4781 if (ctxt->mode != X86EMUL_MODE_PROT64)
4782 addr = (u32)addr;
bbd9b64e
CO		 4783 val += now;
4784 bytes -= now;
4785 }
22388a3c 4786
f78146b0
AK		 4787 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4788 vcpu, ops);
4789 if (rc != X86EMUL_CONTINUE)
4790 return rc;
4791
4792 if (!vcpu->mmio_nr_fragments)
4793 return rc;
4794
4795 gpa = vcpu->mmio_fragments[0].gpa;
4796
4797 vcpu->mmio_needed = 1;
4798 vcpu->mmio_cur_fragment = 0;
4799
87da7e66 4800 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
f78146b0
AK		 4801 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4802 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4803 vcpu->run->mmio.phys_addr = gpa;
4804
4805 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
22388a3c
XG		 4806}
4807
4808static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4809 unsigned long addr,
4810 void *val,
4811 unsigned int bytes,
4812 struct x86_exception *exception)
4813{
4814 return emulator_read_write(ctxt, addr, val, bytes,
4815 exception, &read_emultor);
4816}
4817
52eb5a6d 4818static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
22388a3c
XG		 4819 unsigned long addr,
4820 const void *val,
4821 unsigned int bytes,
4822 struct x86_exception *exception)
4823{
4824 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4825 exception, &write_emultor);
bbd9b64e 4826}
bbd9b64e 4827
daea3e73
AK		 4828#define CMPXCHG_TYPE(t, ptr, old, new) \
4829 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4830
4831#ifdef CONFIG_X86_64
4832# define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4833#else
4834# define CMPXCHG64(ptr, old, new) \
9749a6c0 4835 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
daea3e73
AK		 4836#endif
4837
0f65dd70
AK		 4838static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4839 unsigned long addr,
bbd9b64e
CO		 4840 const void *old,
4841 const void *new,
4842 unsigned int bytes,
0f65dd70 4843 struct x86_exception *exception)
bbd9b64e 4844{
0f65dd70 4845 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
daea3e73
AK		 4846 gpa_t gpa;
4847 struct page *page;
4848 char *kaddr;
4849 bool exchanged;
2bacc55c 4850
daea3e73
AK		 4851 /* guests cmpxchg8b have to be emulated atomically */
4852 if (bytes > 8 || (bytes & (bytes - 1)))
4853 goto emul_write;
10589a46 4854
daea3e73 4855 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
2bacc55c 4856
daea3e73
AK		 4857 if (gpa == UNMAPPED_GVA ||
4858 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4859 goto emul_write;
2bacc55c 4860
daea3e73
AK		 4861 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4862 goto emul_write;
72dc67a6 4863
54bf36aa 4864 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
32cad84f 4865 if (is_error_page(page))
c19b8bd6 4866 goto emul_write;
72dc67a6 4867
8fd75e12 4868 kaddr = kmap_atomic(page);
daea3e73
AK		 4869 kaddr += offset_in_page(gpa);
4870 switch (bytes) {
4871 case 1:
4872 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4873 break;
4874 case 2:
4875 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4876 break;
4877 case 4:
4878 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4879 break;
4880 case 8:
4881 exchanged = CMPXCHG64(kaddr, old, new);
4882 break;
4883 default:
4884 BUG();
2bacc55c 4885 }
8fd75e12 4886 kunmap_atomic(kaddr);
daea3e73
AK		 4887 kvm_release_page_dirty(page);
4888
4889 if (!exchanged)
4890 return X86EMUL_CMPXCHG_FAILED;
4891
54bf36aa 4892 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
0eb05bf2 4893 kvm_page_track_write(vcpu, gpa, new, bytes);
8f6abd06
GN		 4894
4895 return X86EMUL_CONTINUE;
4a5f48f6 4896
3200f405 4897emul_write:
daea3e73 4898 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
2bacc55c 4899
0f65dd70 4900 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
bbd9b64e
CO		 4901}
4902
cf8f70bf
GN		 4903static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4904{
cbfc6c91 4905 int r = 0, i;
cf8f70bf 4906
cbfc6c91
WL		 4907 for (i = 0; i < vcpu->arch.pio.count; i++) {
4908 if (vcpu->arch.pio.in)
4909 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4910 vcpu->arch.pio.size, pd);
4911 else
4912 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4913 vcpu->arch.pio.port, vcpu->arch.pio.size,
4914 pd);
4915 if (r)
4916 break;
4917 pd += vcpu->arch.pio.size;
4918 }
cf8f70bf
GN		 4919 return r;
4920}
4921
6f6fbe98
XG		 4922static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4923 unsigned short port, void *val,
4924 unsigned int count, bool in)
cf8f70bf 4925{
cf8f70bf 4926 vcpu->arch.pio.port = port;
6f6fbe98 4927 vcpu->arch.pio.in = in;
7972995b 4928 vcpu->arch.pio.count = count;
cf8f70bf
GN		 4929 vcpu->arch.pio.size = size;
4930
4931 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
7972995b 4932 vcpu->arch.pio.count = 0;
cf8f70bf
GN		 4933 return 1;
4934 }
4935
4936 vcpu->run->exit_reason = KVM_EXIT_IO;
6f6fbe98 4937 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
cf8f70bf
GN		 4938 vcpu->run->io.size = size;
4939 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4940 vcpu->run->io.count = count;
4941 vcpu->run->io.port = port;
4942
4943 return 0;
4944}
4945
6f6fbe98
XG		 4946static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4947 int size, unsigned short port, void *val,
4948 unsigned int count)
cf8f70bf 4949{
ca1d4a9e 4950 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6f6fbe98 4951 int ret;
ca1d4a9e 4952
6f6fbe98
XG		 4953 if (vcpu->arch.pio.count)
4954 goto data_avail;
cf8f70bf 4955
cbfc6c91
WL		 4956 memset(vcpu->arch.pio_data, 0, size * count);
4957
6f6fbe98
XG		 4958 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4959 if (ret) {
4960data_avail:
4961 memcpy(val, vcpu->arch.pio_data, size * count);
1171903d 4962 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
7972995b 4963 vcpu->arch.pio.count = 0;
cf8f70bf
GN		 4964 return 1;
4965 }
4966
cf8f70bf
GN		 4967 return 0;
4968}
4969
6f6fbe98
XG		 4970static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4971 int size, unsigned short port,
4972 const void *val, unsigned int count)
4973{
4974 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4975
4976 memcpy(vcpu->arch.pio_data, val, size * count);
1171903d 4977 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6f6fbe98
XG		 4978 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4979}
4980
bbd9b64e
CO		 4981static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4982{
4983 return kvm_x86_ops->get_segment_base(vcpu, seg);
4984}
4985
3cb16fe7 4986static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
bbd9b64e 4987{
3cb16fe7 4988 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
bbd9b64e
CO		 4989}
4990
ae6a2375 4991static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
f5f48ee1
SY		 4992{
4993 if (!need_emulate_wbinvd(vcpu))
4994 return X86EMUL_CONTINUE;
4995
4996 if (kvm_x86_ops->has_wbinvd_exit()) {
2eec7343
JK		 4997 int cpu = get_cpu();
4998
4999 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
f5f48ee1
SY		 5000 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5001 wbinvd_ipi, NULL, 1);
2eec7343 5002 put_cpu();
f5f48ee1 5003 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
2eec7343
JK		 5004 } else
5005 wbinvd();
f5f48ee1
SY		 5006 return X86EMUL_CONTINUE;
5007}
5cb56059
JS		 5008
5009int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5010{
6affcbed
KH		 5011 kvm_emulate_wbinvd_noskip(vcpu);
5012 return kvm_skip_emulated_instruction(vcpu);
5cb56059 5013}
f5f48ee1
SY		 5014EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5015
5cb56059
JS		 5016
5017
bcaf5cc5
AK		 5018static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5019{
5cb56059 5020 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
bcaf5cc5
AK		 5021}
5022
52eb5a6d
XL		 5023static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5024 unsigned long *dest)
bbd9b64e 5025{
16f8a6f9 5026 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
bbd9b64e
CO		 5027}
5028
52eb5a6d
XL		 5029static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5030 unsigned long value)
bbd9b64e 5031{
338dbc97 5032
717746e3 5033 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
bbd9b64e
CO		 5034}
5035
52a46617 5036static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5fdbf976 5037{
52a46617 5038 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5fdbf976
MT		 5039}
5040
717746e3 5041static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
bbd9b64e 5042{
717746e3 5043 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
52a46617
GN		 5044 unsigned long value;
5045
5046 switch (cr) {
5047 case 0:
5048 value = kvm_read_cr0(vcpu);
5049 break;
5050 case 2:
5051 value = vcpu->arch.cr2;
5052 break;
5053 case 3:
9f8fe504 5054 value = kvm_read_cr3(vcpu);
52a46617
GN		 5055 break;
5056 case 4:
5057 value = kvm_read_cr4(vcpu);
5058 break;
5059 case 8:
5060 value = kvm_get_cr8(vcpu);
5061 break;
5062 default:
a737f256 5063 kvm_err("%s: unexpected cr %u\n", __func__, cr);
52a46617
GN		 5064 return 0;
5065 }
5066
5067 return value;
5068}
5069
717746e3 5070static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
52a46617 5071{
717746e3 5072 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
0f12244f
GN		 5073 int res = 0;
5074
52a46617
GN		 5075 switch (cr) {
5076 case 0:
49a9b07e 5077 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
52a46617
GN		 5078 break;
5079 case 2:
5080 vcpu->arch.cr2 = val;
5081 break;
5082 case 3:
2390218b 5083 res = kvm_set_cr3(vcpu, val);
52a46617
GN		 5084 break;
5085 case 4:
a83b29c6 5086 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
52a46617
GN		 5087 break;
5088 case 8:
eea1cff9 5089 res = kvm_set_cr8(vcpu, val);
52a46617
GN		 5090 break;
5091 default:
a737f256 5092 kvm_err("%s: unexpected cr %u\n", __func__, cr);
0f12244f 5093 res = -1;
52a46617 5094 }
0f12244f
GN		 5095
5096 return res;
52a46617
GN		 5097}
5098
717746e3 5099static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
9c537244 5100{
717746e3 5101 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
9c537244
GN		 5102}
5103
4bff1e86 5104static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
2dafc6c2 5105{
4bff1e86 5106 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
2dafc6c2
GN		 5107}
5108
4bff1e86 5109static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
160ce1f1 5110{
4bff1e86 5111 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
160ce1f1
MG		 5112}
5113
1ac9d0cf
AK		 5114static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5115{
5116 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5117}
5118
5119static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5120{
5121 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5122}
5123
4bff1e86
AK		 5124static unsigned long emulator_get_cached_segment_base(
5125 struct x86_emulate_ctxt *ctxt, int seg)
5951c442 5126{
4bff1e86 5127 return get_segment_base(emul_to_vcpu(ctxt), seg);
5951c442
GN		 5128}
5129
1aa36616
AK		 5130static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5131 struct desc_struct *desc, u32 *base3,
5132 int seg)
2dafc6c2
GN		 5133{
5134 struct kvm_segment var;
5135
4bff1e86 5136 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
1aa36616 5137 *selector = var.selector;
2dafc6c2 5138
378a8b09
GN		 5139 if (var.unusable) {
5140 memset(desc, 0, sizeof(*desc));
f0367ee1
RK		 5141 if (base3)
5142 *base3 = 0;
2dafc6c2 5143 return false;
378a8b09 5144 }
2dafc6c2
GN		 5145
5146 if (var.g)
5147 var.limit >>= 12;
5148 set_desc_limit(desc, var.limit);
5149 set_desc_base(desc, (unsigned long)var.base);
5601d05b
GN		 5150#ifdef CONFIG_X86_64
5151 if (base3)
5152 *base3 = var.base >> 32;
5153#endif
2dafc6c2
GN		 5154 desc->type = var.type;
5155 desc->s = var.s;
5156 desc->dpl = var.dpl;
5157 desc->p = var.present;
5158 desc->avl = var.avl;
5159 desc->l = var.l;
5160 desc->d = var.db;
5161 desc->g = var.g;
5162
5163 return true;
5164}
5165
1aa36616
AK		 5166static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5167 struct desc_struct *desc, u32 base3,
5168 int seg)
2dafc6c2 5169{
4bff1e86 5170 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
2dafc6c2
GN		 5171 struct kvm_segment var;
5172
1aa36616 5173 var.selector = selector;
2dafc6c2 5174 var.base = get_desc_base(desc);
5601d05b
GN		 5175#ifdef CONFIG_X86_64
5176 var.base |= ((u64)base3) << 32;
5177#endif
2dafc6c2
GN		 5178 var.limit = get_desc_limit(desc);
5179 if (desc->g)
5180 var.limit = (var.limit << 12) | 0xfff;
5181 var.type = desc->type;
2dafc6c2
GN		 5182 var.dpl = desc->dpl;
5183 var.db = desc->d;
5184 var.s = desc->s;
5185 var.l = desc->l;
5186 var.g = desc->g;
5187 var.avl = desc->avl;
5188 var.present = desc->p;
5189 var.unusable = !var.present;
5190 var.padding = 0;
5191
5192 kvm_set_segment(vcpu, &var, seg);
5193 return;
5194}
5195
717746e3
AK		 5196static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5197 u32 msr_index, u64 *pdata)
5198{
609e36d3
PB		 5199 struct msr_data msr;
5200 int r;
5201
5202 msr.index = msr_index;
5203 msr.host_initiated = false;
5204 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5205 if (r)
5206 return r;
5207
5208 *pdata = msr.data;
5209 return 0;
717746e3
AK		 5210}
5211
5212static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5213 u32 msr_index, u64 data)
5214{
8fe8ab46
WA		 5215 struct msr_data msr;
5216
5217 msr.data = data;
5218 msr.index = msr_index;
5219 msr.host_initiated = false;
5220 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
717746e3
AK		 5221}
5222
64d60670
PB		 5223static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5224{
5225 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5226
5227 return vcpu->arch.smbase;
5228}
5229
5230static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5231{
5232 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5233
5234 vcpu->arch.smbase = smbase;
5235}
5236
67f4d428
NA		 5237static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5238 u32 pmc)
5239{
c6702c9d 5240 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
67f4d428
NA		 5241}
5242
222d21aa
AK		 5243static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5244 u32 pmc, u64 *pdata)
5245{
c6702c9d 5246 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
222d21aa
AK		 5247}
5248
6c3287f7
AK		 5249static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5250{
5251 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5252}
5253
2953538e 5254static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
8a76d7f2 5255 struct x86_instruction_info *info,
c4f035c6
AK		 5256 enum x86_intercept_stage stage)
5257{
2953538e 5258 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
c4f035c6
AK		 5259}
5260
e911eb3b
YZ		 5261static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5262 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
bdb42f5a 5263{
e911eb3b 5264 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
bdb42f5a
SB		 5265}
5266
dd856efa
AK		 5267static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5268{
5269 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5270}
5271
5272static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5273{
5274 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5275}
5276
801806d9
NA		 5277static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5278{
5279 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5280}
5281
6ed071f0
LP		 5282static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5283{
5284 return emul_to_vcpu(ctxt)->arch.hflags;
5285}
5286
5287static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5288{
5289 kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5290}
5291
0234bf88
LP		 5292static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, u64 smbase)
5293{
5294 return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smbase);
5295}
5296
0225fb50 5297static const struct x86_emulate_ops emulate_ops = {
dd856efa
AK		 5298 .read_gpr = emulator_read_gpr,
5299 .write_gpr = emulator_write_gpr,
1871c602 5300 .read_std = kvm_read_guest_virt_system,
2dafc6c2 5301 .write_std = kvm_write_guest_virt_system,
7a036a6f 5302 .read_phys = kvm_read_guest_phys_system,
1871c602 5303 .fetch = kvm_fetch_guest_virt,
bbd9b64e
CO		 5304 .read_emulated = emulator_read_emulated,
5305 .write_emulated = emulator_write_emulated,
5306 .cmpxchg_emulated = emulator_cmpxchg_emulated,
3cb16fe7 5307 .invlpg = emulator_invlpg,
cf8f70bf
GN		 5308 .pio_in_emulated = emulator_pio_in_emulated,
5309 .pio_out_emulated = emulator_pio_out_emulated,
1aa36616
AK		 5310 .get_segment = emulator_get_segment,
5311 .set_segment = emulator_set_segment,
5951c442 5312 .get_cached_segment_base = emulator_get_cached_segment_base,
2dafc6c2 5313 .get_gdt = emulator_get_gdt,
160ce1f1 5314 .get_idt = emulator_get_idt,
1ac9d0cf
AK		 5315 .set_gdt = emulator_set_gdt,
5316 .set_idt = emulator_set_idt,
52a46617
GN		 5317 .get_cr = emulator_get_cr,
5318 .set_cr = emulator_set_cr,
9c537244 5319 .cpl = emulator_get_cpl,
35aa5375
GN		 5320 .get_dr = emulator_get_dr,
5321 .set_dr = emulator_set_dr,
64d60670
PB		 5322 .get_smbase = emulator_get_smbase,
5323 .set_smbase = emulator_set_smbase,
717746e3
AK		 5324 .set_msr = emulator_set_msr,
5325 .get_msr = emulator_get_msr,
67f4d428 5326 .check_pmc = emulator_check_pmc,
222d21aa 5327 .read_pmc = emulator_read_pmc,
6c3287f7 5328 .halt = emulator_halt,
bcaf5cc5 5329 .wbinvd = emulator_wbinvd,
d6aa1000 5330 .fix_hypercall = emulator_fix_hypercall,
c4f035c6 5331 .intercept = emulator_intercept,
bdb42f5a 5332 .get_cpuid = emulator_get_cpuid,
801806d9 5333 .set_nmi_mask = emulator_set_nmi_mask,
6ed071f0
LP		 5334 .get_hflags = emulator_get_hflags,
5335 .set_hflags = emulator_set_hflags,
0234bf88 5336 .pre_leave_smm = emulator_pre_leave_smm,
bbd9b64e
CO		 5337};
5338
95cb2295
GN		 5339static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5340{
37ccdcbe 5341 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
95cb2295
GN		 5342 /*
5343 * an sti; sti; sequence only disable interrupts for the first
5344 * instruction. So, if the last instruction, be it emulated or
5345 * not, left the system with the INT_STI flag enabled, it
5346 * means that the last instruction is an sti. We should not
5347 * leave the flag on in this case. The same goes for mov ss
5348 */
37ccdcbe
PB		 5349 if (int_shadow & mask)
5350 mask = 0;
6addfc42 5351 if (unlikely(int_shadow || mask)) {
95cb2295 5352 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6addfc42
PB		 5353 if (!mask)
5354 kvm_make_request(KVM_REQ_EVENT, vcpu);
5355 }
95cb2295
GN		 5356}
5357
ef54bcfe 5358static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
54b8486f
GN		 5359{
5360 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
da9cb575 5361 if (ctxt->exception.vector == PF_VECTOR)
ef54bcfe
PB		 5362 return kvm_propagate_fault(vcpu, &ctxt->exception);
5363
5364 if (ctxt->exception.error_code_valid)
da9cb575
AK		 5365 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5366 ctxt->exception.error_code);
54b8486f 5367 else
da9cb575 5368 kvm_queue_exception(vcpu, ctxt->exception.vector);
ef54bcfe 5369 return false;
54b8486f
GN		 5370}
5371
8ec4722d
MG		 5372static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5373{
adf52235 5374 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8ec4722d
MG		 5375 int cs_db, cs_l;
5376
8ec4722d
MG		 5377 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5378
adf52235 5379 ctxt->eflags = kvm_get_rflags(vcpu);
c8401dda
PB		 5380 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
5381
adf52235
TY		 5382 ctxt->eip = kvm_rip_read(vcpu);
5383 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5384 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
42bf549f 5385 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
adf52235
TY		 5386 cs_db ? X86EMUL_MODE_PROT32 :
5387 X86EMUL_MODE_PROT16;
a584539b 5388 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
64d60670
PB		 5389 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5390 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
adf52235 5391
dd856efa 5392 init_decode_cache(ctxt);
7ae441ea 5393 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8ec4722d
MG		 5394}
5395
71f9833b 5396int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
63995653 5397{
9d74191a 5398 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
63995653
MG		 5399 int ret;
5400
5401 init_emulate_ctxt(vcpu);
5402
9dac77fa
AK		 5403 ctxt->op_bytes = 2;
5404 ctxt->ad_bytes = 2;
5405 ctxt->_eip = ctxt->eip + inc_eip;
9d74191a 5406 ret = emulate_int_real(ctxt, irq);
63995653
MG		 5407
5408 if (ret != X86EMUL_CONTINUE)
5409 return EMULATE_FAIL;
5410
9dac77fa 5411 ctxt->eip = ctxt->_eip;
9d74191a
TY		 5412 kvm_rip_write(vcpu, ctxt->eip);
5413 kvm_set_rflags(vcpu, ctxt->eflags);
63995653
MG		 5414
5415 if (irq == NMI_VECTOR)
7460fb4a 5416 vcpu->arch.nmi_pending = 0;
63995653
MG		 5417 else
5418 vcpu->arch.interrupt.pending = false;
5419
5420 return EMULATE_DONE;
5421}
5422EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5423
6d77dbfc
GN		 5424static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5425{
fc3a9157
JR		 5426 int r = EMULATE_DONE;
5427
6d77dbfc
GN		 5428 ++vcpu->stat.insn_emulation_fail;
5429 trace_kvm_emulate_insn_failed(vcpu);
a2b9e6c1 5430 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
fc3a9157
JR		 5431 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5432 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5433 vcpu->run->internal.ndata = 0;
1f4dcb3b 5434 r = EMULATE_USER_EXIT;
fc3a9157 5435 }
6d77dbfc 5436 kvm_queue_exception(vcpu, UD_VECTOR);
fc3a9157
JR		 5437
5438 return r;
6d77dbfc
GN		 5439}
5440
93c05d3e 5441static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
991eebf9
GN		 5442 bool write_fault_to_shadow_pgtable,
5443 int emulation_type)
a6f177ef 5444{
95b3cf69 5445 gpa_t gpa = cr2;
ba049e93 5446 kvm_pfn_t pfn;
a6f177ef 5447
991eebf9
GN		 5448 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5449 return false;
5450
95b3cf69
XG		 5451 if (!vcpu->arch.mmu.direct_map) {
5452 /*
5453 * Write permission should be allowed since only
5454 * write access need to be emulated.
5455 */
5456 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
a6f177ef 5457
95b3cf69
XG		 5458 /*
5459 * If the mapping is invalid in guest, let cpu retry
5460 * it to generate fault.
5461 */
5462 if (gpa == UNMAPPED_GVA)
5463 return true;
5464 }
a6f177ef 5465
8e3d9d06
XG		 5466 /*
5467 * Do not retry the unhandleable instruction if it faults on the
5468 * readonly host memory, otherwise it will goto a infinite loop:
5469 * retry instruction -> write #PF -> emulation fail -> retry
5470 * instruction -> ...
5471 */
5472 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
95b3cf69
XG		 5473
5474 /*
5475 * If the instruction failed on the error pfn, it can not be fixed,
5476 * report the error to userspace.
5477 */
5478 if (is_error_noslot_pfn(pfn))
5479 return false;
5480
5481 kvm_release_pfn_clean(pfn);
5482
5483 /* The instructions are well-emulated on direct mmu. */
5484 if (vcpu->arch.mmu.direct_map) {
5485 unsigned int indirect_shadow_pages;
5486
5487 spin_lock(&vcpu->kvm->mmu_lock);
5488 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5489 spin_unlock(&vcpu->kvm->mmu_lock);
5490
5491 if (indirect_shadow_pages)
5492 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5493
a6f177ef 5494 return true;
8e3d9d06 5495 }
a6f177ef 5496
95b3cf69
XG		 5497 /*
5498 * if emulation was due to access to shadowed page table
5499 * and it failed try to unshadow page and re-enter the
5500 * guest to let CPU execute the instruction.
5501 */
5502 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
93c05d3e
XG		 5503
5504 /*
5505 * If the access faults on its page table, it can not
5506 * be fixed by unprotecting shadow page and it should
5507 * be reported to userspace.
5508 */
5509 return !write_fault_to_shadow_pgtable;
a6f177ef
GN		 5510}
5511
1cb3f3ae
XG		 5512static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5513 unsigned long cr2, int emulation_type)
5514{
5515 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5516 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5517
5518 last_retry_eip = vcpu->arch.last_retry_eip;
5519 last_retry_addr = vcpu->arch.last_retry_addr;
5520
5521 /*
5522 * If the emulation is caused by #PF and it is non-page_table
5523 * writing instruction, it means the VM-EXIT is caused by shadow
5524 * page protected, we can zap the shadow page and retry this
5525 * instruction directly.
5526 *
5527 * Note: if the guest uses a non-page-table modifying instruction
5528 * on the PDE that points to the instruction, then we will unmap
5529 * the instruction and go to an infinite loop. So, we cache the
5530 * last retried eip and the last fault address, if we meet the eip
5531 * and the address again, we can break out of the potential infinite
5532 * loop.
5533 */
5534 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5535
5536 if (!(emulation_type & EMULTYPE_RETRY))
5537 return false;
5538
5539 if (x86_page_table_writing_insn(ctxt))
5540 return false;
5541
5542 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5543 return false;
5544
5545 vcpu->arch.last_retry_eip = ctxt->eip;
5546 vcpu->arch.last_retry_addr = cr2;
5547
5548 if (!vcpu->arch.mmu.direct_map)
5549 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5550
22368028 5551 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
1cb3f3ae
XG		 5552
5553 return true;
5554}
5555
716d51ab
GN		 5556static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5557static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5558
64d60670 5559static void kvm_smm_changed(struct kvm_vcpu *vcpu)
a584539b 5560{
64d60670 5561 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
660a5d51
PB		 5562 /* This is a good place to trace that we are exiting SMM. */
5563 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5564
c43203ca
PB		 5565 /* Process a latched INIT or SMI, if any. */
5566 kvm_make_request(KVM_REQ_EVENT, vcpu);
64d60670 5567 }
699023e2
PB		 5568
5569 kvm_mmu_reset_context(vcpu);
64d60670
PB		 5570}
5571
5572static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5573{
5574 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5575
a584539b 5576 vcpu->arch.hflags = emul_flags;
64d60670
PB		 5577
5578 if (changed & HF_SMM_MASK)
5579 kvm_smm_changed(vcpu);
a584539b
PB		 5580}
5581
4a1e10d5
PB		 5582static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5583 unsigned long *db)
5584{
5585 u32 dr6 = 0;
5586 int i;
5587 u32 enable, rwlen;
5588
5589 enable = dr7;
5590 rwlen = dr7 >> 16;
5591 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5592 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5593 dr6 |= (1 << i);
5594 return dr6;
5595}
5596
c8401dda 5597static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
663f4c61
PB		 5598{
5599 struct kvm_run *kvm_run = vcpu->run;
5600
c8401dda
PB		 5601 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5602 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
5603 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5604 kvm_run->debug.arch.exception = DB_VECTOR;
5605 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5606 *r = EMULATE_USER_EXIT;
5607 } else {
5608 /*
5609 * "Certain debug exceptions may clear bit 0-3. The
5610 * remaining contents of the DR6 register are never
5611 * cleared by the processor".
5612 */
5613 vcpu->arch.dr6 &= ~15;
5614 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5615 kvm_queue_exception(vcpu, DB_VECTOR);
663f4c61
PB		 5616 }
5617}
5618
6affcbed
KH		 5619int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
5620{
5621 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5622 int r = EMULATE_DONE;
5623
5624 kvm_x86_ops->skip_emulated_instruction(vcpu);
c8401dda
PB		 5625
5626 /*
5627 * rflags is the old, "raw" value of the flags. The new value has
5628 * not been saved yet.
5629 *
5630 * This is correct even for TF set by the guest, because "the
5631 * processor will not generate this exception after the instruction
5632 * that sets the TF flag".
5633 */
5634 if (unlikely(rflags & X86_EFLAGS_TF))
5635 kvm_vcpu_do_singlestep(vcpu, &r);
6affcbed
KH		 5636 return r == EMULATE_DONE;
5637}
5638EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
5639
4a1e10d5
PB		 5640static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5641{
4a1e10d5
PB		 5642 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5643 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
82b32774
NA		 5644 struct kvm_run *kvm_run = vcpu->run;
5645 unsigned long eip = kvm_get_linear_rip(vcpu);
5646 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
4a1e10d5
PB		 5647 vcpu->arch.guest_debug_dr7,
5648 vcpu->arch.eff_db);
5649
5650 if (dr6 != 0) {
6f43ed01 5651 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
82b32774 5652 kvm_run->debug.arch.pc = eip;
4a1e10d5
PB		 5653 kvm_run->debug.arch.exception = DB_VECTOR;
5654 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5655 *r = EMULATE_USER_EXIT;
5656 return true;
5657 }
5658 }
5659
4161a569
NA		 5660 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5661 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
82b32774
NA		 5662 unsigned long eip = kvm_get_linear_rip(vcpu);
5663 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
4a1e10d5
PB		 5664 vcpu->arch.dr7,
5665 vcpu->arch.db);
5666
5667 if (dr6 != 0) {
5668 vcpu->arch.dr6 &= ~15;
6f43ed01 5669 vcpu->arch.dr6 |= dr6 | DR6_RTM;
4a1e10d5
PB		 5670 kvm_queue_exception(vcpu, DB_VECTOR);
5671 *r = EMULATE_DONE;
5672 return true;
5673 }
5674 }
5675
5676 return false;
5677}
5678
51d8b661
AP		 5679int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5680 unsigned long cr2,
dc25e89e
AP		 5681 int emulation_type,
5682 void *insn,
5683 int insn_len)
bbd9b64e 5684{
95cb2295 5685 int r;
9d74191a 5686 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7ae441ea 5687 bool writeback = true;
93c05d3e 5688 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
bbd9b64e 5689
93c05d3e
XG		 5690 /*
5691 * Clear write_fault_to_shadow_pgtable here to ensure it is
5692 * never reused.
5693 */
5694 vcpu->arch.write_fault_to_shadow_pgtable = false;
26eef70c 5695 kvm_clear_exception_queue(vcpu);
8d7d8102 5696
571008da 5697 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
8ec4722d 5698 init_emulate_ctxt(vcpu);
4a1e10d5
PB		 5699
5700 /*
5701 * We will reenter on the same instruction since
5702 * we do not set complete_userspace_io. This does not
5703 * handle watchpoints yet, those would be handled in
5704 * the emulate_ops.
5705 */
5706 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5707 return r;
5708
9d74191a
TY		 5709 ctxt->interruptibility = 0;
5710 ctxt->have_exception = false;
e0ad0b47 5711 ctxt->exception.vector = -1;
9d74191a 5712 ctxt->perm_ok = false;
bbd9b64e 5713
b51e974f 5714 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
4005996e 5715
9d74191a 5716 r = x86_decode_insn(ctxt, insn, insn_len);
bbd9b64e 5717
e46479f8 5718 trace_kvm_emulate_insn_start(vcpu);
f2b5756b 5719 ++vcpu->stat.insn_emulation;
1d2887e2 5720 if (r != EMULATION_OK) {
4005996e
AK		 5721 if (emulation_type & EMULTYPE_TRAP_UD)
5722 return EMULATE_FAIL;
991eebf9
GN		 5723 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5724 emulation_type))
bbd9b64e 5725 return EMULATE_DONE;
6ea6e843
PB		 5726 if (ctxt->have_exception && inject_emulated_exception(vcpu))
5727 return EMULATE_DONE;
6d77dbfc
GN		 5728 if (emulation_type & EMULTYPE_SKIP)
5729 return EMULATE_FAIL;
5730 return handle_emulation_failure(vcpu);
bbd9b64e
CO		 5731 }
5732 }
5733
ba8afb6b 5734 if (emulation_type & EMULTYPE_SKIP) {
9dac77fa 5735 kvm_rip_write(vcpu, ctxt->_eip);
bb663c7a
NA		 5736 if (ctxt->eflags & X86_EFLAGS_RF)
5737 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
ba8afb6b
GN		 5738 return EMULATE_DONE;
5739 }
5740
1cb3f3ae
XG		 5741 if (retry_instruction(ctxt, cr2, emulation_type))
5742 return EMULATE_DONE;
5743
7ae441ea 5744 /* this is needed for vmware backdoor interface to work since it
4d2179e1 5745 changes registers values during IO operation */
7ae441ea
GN		 5746 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5747 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
dd856efa 5748 emulator_invalidate_register_cache(ctxt);
7ae441ea 5749 }
4d2179e1 5750
5cd21917 5751restart:
0f89b207
TL		 5752 /* Save the faulting GPA (cr2) in the address field */
5753 ctxt->exception.address = cr2;
5754
9d74191a 5755 r = x86_emulate_insn(ctxt);
bbd9b64e 5756
775fde86
JR		 5757 if (r == EMULATION_INTERCEPTED)
5758 return EMULATE_DONE;
5759
d2ddd1c4 5760 if (r == EMULATION_FAILED) {
991eebf9
GN		 5761 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5762 emulation_type))
c3cd7ffa
GN		 5763 return EMULATE_DONE;
5764
6d77dbfc 5765 return handle_emulation_failure(vcpu);
bbd9b64e
CO		 5766 }
5767
9d74191a 5768 if (ctxt->have_exception) {
d2ddd1c4 5769 r = EMULATE_DONE;
ef54bcfe
PB		 5770 if (inject_emulated_exception(vcpu))
5771 return r;
d2ddd1c4 5772 } else if (vcpu->arch.pio.count) {
0912c977
PB		 5773 if (!vcpu->arch.pio.in) {
5774 /* FIXME: return into emulator if single-stepping. */
3457e419 5775 vcpu->arch.pio.count = 0;
0912c977 5776 } else {
7ae441ea 5777 writeback = false;
716d51ab
GN		 5778 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5779 }
ac0a48c3 5780 r = EMULATE_USER_EXIT;
7ae441ea
GN		 5781 } else if (vcpu->mmio_needed) {
5782 if (!vcpu->mmio_is_write)
5783 writeback = false;
ac0a48c3 5784 r = EMULATE_USER_EXIT;
716d51ab 5785 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7ae441ea 5786 } else if (r == EMULATION_RESTART)
5cd21917 5787 goto restart;
d2ddd1c4
GN		 5788 else
5789 r = EMULATE_DONE;
f850e2e6 5790
7ae441ea 5791 if (writeback) {
6addfc42 5792 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
9d74191a 5793 toggle_interruptibility(vcpu, ctxt->interruptibility);
7ae441ea 5794 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9d74191a 5795 kvm_rip_write(vcpu, ctxt->eip);
c8401dda
PB		 5796 if (r == EMULATE_DONE &&
5797 (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
5798 kvm_vcpu_do_singlestep(vcpu, &r);
38827dbd
NA		 5799 if (!ctxt->have_exception ||
5800 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5801 __kvm_set_rflags(vcpu, ctxt->eflags);
6addfc42
PB		 5802
5803 /*
5804 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5805 * do nothing, and it will be requested again as soon as
5806 * the shadow expires. But we still need to check here,
5807 * because POPF has no interrupt shadow.
5808 */
5809 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5810 kvm_make_request(KVM_REQ_EVENT, vcpu);
7ae441ea
GN		 5811 } else
5812 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
e85d28f8
GN		 5813
5814 return r;
de7d789a 5815}
51d8b661 5816EXPORT_SYMBOL_GPL(x86_emulate_instruction);
de7d789a 5817
cf8f70bf 5818int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
de7d789a 5819{
cf8f70bf 5820 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
ca1d4a9e
AK		 5821 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5822 size, port, &val, 1);
cf8f70bf 5823 /* do not return to emulator after return from userspace */
7972995b 5824 vcpu->arch.pio.count = 0;
de7d789a
CO		 5825 return ret;
5826}
cf8f70bf 5827EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
de7d789a 5828
8370c3d0
TL		 5829static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
5830{
5831 unsigned long val;
5832
5833 /* We should only ever be called with arch.pio.count equal to 1 */
5834 BUG_ON(vcpu->arch.pio.count != 1);
5835
5836 /* For size less than 4 we merge, else we zero extend */
5837 val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
5838 : 0;
5839
5840 /*
5841 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
5842 * the copy and tracing
5843 */
5844 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
5845 vcpu->arch.pio.port, &val, 1);
5846 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5847
5848 return 1;
5849}
5850
5851int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size, unsigned short port)
5852{
5853 unsigned long val;
5854 int ret;
5855
5856 /* For size less than 4 we merge, else we zero extend */
5857 val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
5858
5859 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
5860 &val, 1);
5861 if (ret) {
5862 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5863 return ret;
5864 }
5865
5866 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
5867
5868 return 0;
5869}
5870EXPORT_SYMBOL_GPL(kvm_fast_pio_in);
5871
251a5fd6 5872static int kvmclock_cpu_down_prep(unsigned int cpu)
8cfdc000 5873{
0a3aee0d 5874 __this_cpu_write(cpu_tsc_khz, 0);
251a5fd6 5875 return 0;
8cfdc000
ZA		 5876}
5877
5878static void tsc_khz_changed(void *data)
c8076604 5879{
8cfdc000
ZA		 5880 struct cpufreq_freqs *freq = data;
5881 unsigned long khz = 0;
5882
5883 if (data)
5884 khz = freq->new;
5885 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5886 khz = cpufreq_quick_get(raw_smp_processor_id());
5887 if (!khz)
5888 khz = tsc_khz;
0a3aee0d 5889 __this_cpu_write(cpu_tsc_khz, khz);
c8076604
GH		 5890}
5891
c8076604
GH		 5892static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5893 void *data)
5894{
5895 struct cpufreq_freqs *freq = data;
5896 struct kvm *kvm;
5897 struct kvm_vcpu *vcpu;
5898 int i, send_ipi = 0;
5899
8cfdc000
ZA		 5900 /*
5901 * We allow guests to temporarily run on slowing clocks,
5902 * provided we notify them after, or to run on accelerating
5903 * clocks, provided we notify them before. Thus time never
5904 * goes backwards.
5905 *
5906 * However, we have a problem. We can't atomically update
5907 * the frequency of a given CPU from this function; it is
5908 * merely a notifier, which can be called from any CPU.
5909 * Changing the TSC frequency at arbitrary points in time
5910 * requires a recomputation of local variables related to
5911 * the TSC for each VCPU. We must flag these local variables
5912 * to be updated and be sure the update takes place with the
5913 * new frequency before any guests proceed.
5914 *
5915 * Unfortunately, the combination of hotplug CPU and frequency
5916 * change creates an intractable locking scenario; the order
5917 * of when these callouts happen is undefined with respect to
5918 * CPU hotplug, and they can race with each other. As such,
5919 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5920 * undefined; you can actually have a CPU frequency change take
5921 * place in between the computation of X and the setting of the
5922 * variable. To protect against this problem, all updates of
5923 * the per_cpu tsc_khz variable are done in an interrupt
5924 * protected IPI, and all callers wishing to update the value
5925 * must wait for a synchronous IPI to complete (which is trivial
5926 * if the caller is on the CPU already). This establishes the
5927 * necessary total order on variable updates.
5928 *
5929 * Note that because a guest time update may take place
5930 * anytime after the setting of the VCPU's request bit, the
5931 * correct TSC value must be set before the request. However,
5932 * to ensure the update actually makes it to any guest which
5933 * starts running in hardware virtualization between the set
5934 * and the acquisition of the spinlock, we must also ping the
5935 * CPU after setting the request bit.
5936 *
5937 */
5938
c8076604
GH		 5939 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5940 return 0;
5941 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5942 return 0;
8cfdc000
ZA		 5943
5944 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
c8076604 5945
2f303b74 5946 spin_lock(&kvm_lock);
c8076604 5947 list_for_each_entry(kvm, &vm_list, vm_list) {
988a2cae 5948 kvm_for_each_vcpu(i, vcpu, kvm) {
c8076604
GH		 5949 if (vcpu->cpu != freq->cpu)
5950 continue;
c285545f 5951 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
c8076604 5952 if (vcpu->cpu != smp_processor_id())
8cfdc000 5953 send_ipi = 1;
c8076604
GH		 5954 }
5955 }
2f303b74 5956 spin_unlock(&kvm_lock);
c8076604
GH		 5957
5958 if (freq->old < freq->new && send_ipi) {
5959 /*
5960 * We upscale the frequency. Must make the guest
5961 * doesn't see old kvmclock values while running with
5962 * the new frequency, otherwise we risk the guest sees
5963 * time go backwards.
5964 *
5965 * In case we update the frequency for another cpu
5966 * (which might be in guest context) send an interrupt
5967 * to kick the cpu out of guest context. Next time
5968 * guest context is entered kvmclock will be updated,
5969 * so the guest will not see stale values.
5970 */
8cfdc000 5971 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
c8076604
GH		 5972 }
5973 return 0;
5974}
5975
5976static struct notifier_block kvmclock_cpufreq_notifier_block = {
8cfdc000
ZA		 5977 .notifier_call = kvmclock_cpufreq_notifier
5978};
5979
251a5fd6 5980static int kvmclock_cpu_online(unsigned int cpu)
8cfdc000 5981{
251a5fd6
SAS		 5982 tsc_khz_changed(NULL);
5983 return 0;
8cfdc000
ZA		 5984}
5985
b820cc0c
ZA		 5986static void kvm_timer_init(void)
5987{
c285545f 5988 max_tsc_khz = tsc_khz;
460dd42e 5989
b820cc0c 5990 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
c285545f
ZA		 5991#ifdef CONFIG_CPU_FREQ
5992 struct cpufreq_policy policy;
758f588d
BP		 5993 int cpu;
5994
c285545f 5995 memset(&policy, 0, sizeof(policy));
3e26f230
AK		 5996 cpu = get_cpu();
5997 cpufreq_get_policy(&policy, cpu);
c285545f
ZA		 5998 if (policy.cpuinfo.max_freq)
5999 max_tsc_khz = policy.cpuinfo.max_freq;
3e26f230 6000 put_cpu();
c285545f 6001#endif
b820cc0c
ZA		 6002 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6003 CPUFREQ_TRANSITION_NOTIFIER);
6004 }
c285545f 6005 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
460dd42e 6006
73c1b41e 6007 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
251a5fd6 6008 kvmclock_cpu_online, kvmclock_cpu_down_prep);
b820cc0c
ZA		 6009}
6010
ff9d07a0
ZY		 6011static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6012
f5132b01 6013int kvm_is_in_guest(void)
ff9d07a0 6014{
086c9855 6015 return __this_cpu_read(current_vcpu) != NULL;
ff9d07a0
ZY		 6016}
6017
6018static int kvm_is_user_mode(void)
6019{
6020 int user_mode = 3;
dcf46b94 6021
086c9855
AS		 6022 if (__this_cpu_read(current_vcpu))
6023 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
dcf46b94 6024
ff9d07a0
ZY		 6025 return user_mode != 0;
6026}
6027
6028static unsigned long kvm_get_guest_ip(void)
6029{
6030 unsigned long ip = 0;
dcf46b94 6031
086c9855
AS		 6032 if (__this_cpu_read(current_vcpu))
6033 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
dcf46b94 6034
ff9d07a0
ZY		 6035 return ip;
6036}
6037
6038static struct perf_guest_info_callbacks kvm_guest_cbs = {
6039 .is_in_guest = kvm_is_in_guest,
6040 .is_user_mode = kvm_is_user_mode,
6041 .get_guest_ip = kvm_get_guest_ip,
6042};
6043
6044void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
6045{
086c9855 6046 __this_cpu_write(current_vcpu, vcpu);
ff9d07a0
ZY		 6047}
6048EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
6049
6050void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
6051{
086c9855 6052 __this_cpu_write(current_vcpu, NULL);
ff9d07a0
ZY		 6053}
6054EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
6055
ce88decf
XG		 6056static void kvm_set_mmio_spte_mask(void)
6057{
6058 u64 mask;
6059 int maxphyaddr = boot_cpu_data.x86_phys_bits;
6060
6061 /*
6062 * Set the reserved bits and the present bit of an paging-structure
6063 * entry to generate page fault with PFER.RSV = 1.
6064 */
885032b9 6065 /* Mask the reserved physical address bits. */
d1431483 6066 mask = rsvd_bits(maxphyaddr, 51);
885032b9 6067
885032b9 6068 /* Set the present bit. */
ce88decf
XG		 6069 mask |= 1ull;
6070
6071#ifdef CONFIG_X86_64
6072 /*
6073 * If reserved bit is not supported, clear the present bit to disable
6074 * mmio page fault.
6075 */
6076 if (maxphyaddr == 52)
6077 mask &= ~1ull;
6078#endif
6079
dcdca5fe 6080 kvm_mmu_set_mmio_spte_mask(mask, mask);
ce88decf
XG		 6081}
6082
16e8d74d
MT		 6083#ifdef CONFIG_X86_64
6084static void pvclock_gtod_update_fn(struct work_struct *work)
6085{
d828199e
MT		 6086 struct kvm *kvm;
6087
6088 struct kvm_vcpu *vcpu;
6089 int i;
6090
2f303b74 6091 spin_lock(&kvm_lock);
d828199e
MT		 6092 list_for_each_entry(kvm, &vm_list, vm_list)
6093 kvm_for_each_vcpu(i, vcpu, kvm)
105b21bb 6094 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
d828199e 6095 atomic_set(&kvm_guest_has_master_clock, 0);
2f303b74 6096 spin_unlock(&kvm_lock);
16e8d74d
MT		 6097}
6098
6099static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6100
6101/*
6102 * Notification about pvclock gtod data update.
6103 */
6104static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6105 void *priv)
6106{
6107 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6108 struct timekeeper *tk = priv;
6109
6110 update_pvclock_gtod(tk);
6111
6112 /* disable master clock if host does not trust, or does not
6113 * use, TSC clocksource
6114 */
6115 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
6116 atomic_read(&kvm_guest_has_master_clock) != 0)
6117 queue_work(system_long_wq, &pvclock_gtod_work);
6118
6119 return 0;
6120}
6121
6122static struct notifier_block pvclock_gtod_notifier = {
6123 .notifier_call = pvclock_gtod_notify,
6124};
6125#endif
6126
f8c16bba 6127int kvm_arch_init(void *opaque)
043405e1 6128{
b820cc0c 6129 int r;
6b61edf7 6130 struct kvm_x86_ops *ops = opaque;
f8c16bba 6131
f8c16bba
ZX		 6132 if (kvm_x86_ops) {
6133 printk(KERN_ERR "kvm: already loaded the other module\n");
56c6d28a
ZX		 6134 r = -EEXIST;
6135 goto out;
f8c16bba
ZX		 6136 }
6137
6138 if (!ops->cpu_has_kvm_support()) {
6139 printk(KERN_ERR "kvm: no hardware support\n");
56c6d28a
ZX		 6140 r = -EOPNOTSUPP;
6141 goto out;
f8c16bba
ZX		 6142 }
6143 if (ops->disabled_by_bios()) {
6144 printk(KERN_ERR "kvm: disabled by bios\n");
56c6d28a
ZX		 6145 r = -EOPNOTSUPP;
6146 goto out;
f8c16bba
ZX		 6147 }
6148
013f6a5d
MT		 6149 r = -ENOMEM;
6150 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6151 if (!shared_msrs) {
6152 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6153 goto out;
6154 }
6155
97db56ce
AK		 6156 r = kvm_mmu_module_init();
6157 if (r)
013f6a5d 6158 goto out_free_percpu;
97db56ce 6159
ce88decf 6160 kvm_set_mmio_spte_mask();
97db56ce 6161
f8c16bba 6162 kvm_x86_ops = ops;
920c8377 6163
7b52345e 6164 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
ffb128c8 6165 PT_DIRTY_MASK, PT64_NX_MASK, 0,
d0ec49d4 6166 PT_PRESENT_MASK, 0, sme_me_mask);
b820cc0c 6167 kvm_timer_init();
c8076604 6168
ff9d07a0
ZY		 6169 perf_register_guest_info_callbacks(&kvm_guest_cbs);
6170
d366bf7e 6171 if (boot_cpu_has(X86_FEATURE_XSAVE))
2acf923e
DC		 6172 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6173
c5cc421b 6174 kvm_lapic_init();
16e8d74d
MT		 6175#ifdef CONFIG_X86_64
6176 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6177#endif
6178
f8c16bba 6179 return 0;
56c6d28a 6180
013f6a5d
MT		 6181out_free_percpu:
6182 free_percpu(shared_msrs);
56c6d28a 6183out:
56c6d28a 6184 return r;
043405e1 6185}
8776e519 6186
f8c16bba
ZX		 6187void kvm_arch_exit(void)
6188{
cef84c30 6189 kvm_lapic_exit();
ff9d07a0
ZY		 6190 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6191
888d256e
JK		 6192 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6193 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6194 CPUFREQ_TRANSITION_NOTIFIER);
251a5fd6 6195 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
16e8d74d
MT		 6196#ifdef CONFIG_X86_64
6197 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6198#endif
f8c16bba 6199 kvm_x86_ops = NULL;
56c6d28a 6200 kvm_mmu_module_exit();
013f6a5d 6201 free_percpu(shared_msrs);
56c6d28a 6202}
f8c16bba 6203
5cb56059 6204int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8776e519
HB		 6205{
6206 ++vcpu->stat.halt_exits;
35754c98 6207 if (lapic_in_kernel(vcpu)) {
a4535290 6208 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
8776e519
HB		 6209 return 1;
6210 } else {
6211 vcpu->run->exit_reason = KVM_EXIT_HLT;
6212 return 0;
6213 }
6214}
5cb56059
JS		 6215EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6216
6217int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6218{
6affcbed
KH		 6219 int ret = kvm_skip_emulated_instruction(vcpu);
6220 /*
6221 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6222 * KVM_EXIT_DEBUG here.
6223 */
6224 return kvm_vcpu_halt(vcpu) && ret;
5cb56059 6225}
8776e519
HB		 6226EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6227
8ef81a9a 6228#ifdef CONFIG_X86_64
55dd00a7
MT		 6229static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6230 unsigned long clock_type)
6231{
6232 struct kvm_clock_pairing clock_pairing;
6233 struct timespec ts;
80fbd89c 6234 u64 cycle;
55dd00a7
MT		 6235 int ret;
6236
6237 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6238 return -KVM_EOPNOTSUPP;
6239
6240 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6241 return -KVM_EOPNOTSUPP;
6242
6243 clock_pairing.sec = ts.tv_sec;
6244 clock_pairing.nsec = ts.tv_nsec;
6245 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6246 clock_pairing.flags = 0;
6247
6248 ret = 0;
6249 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6250 sizeof(struct kvm_clock_pairing)))
6251 ret = -KVM_EFAULT;
6252
6253 return ret;
6254}
8ef81a9a 6255#endif
55dd00a7 6256
6aef266c
SV		 6257/*
6258 * kvm_pv_kick_cpu_op: Kick a vcpu.
6259 *
6260 * @apicid - apicid of vcpu to be kicked.
6261 */
6262static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6263{
24d2166b 6264 struct kvm_lapic_irq lapic_irq;
6aef266c 6265
24d2166b
R		 6266 lapic_irq.shorthand = 0;
6267 lapic_irq.dest_mode = 0;
ebd28fcb 6268 lapic_irq.level = 0;
24d2166b 6269 lapic_irq.dest_id = apicid;
93bbf0b8 6270 lapic_irq.msi_redir_hint = false;
6aef266c 6271
24d2166b 6272 lapic_irq.delivery_mode = APIC_DM_REMRD;
795a149e 6273 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6aef266c
SV		 6274}
6275
d62caabb
AS		 6276void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6277{
6278 vcpu->arch.apicv_active = false;
6279 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6280}
6281
8776e519
HB		 6282int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6283{
6284 unsigned long nr, a0, a1, a2, a3, ret;
6affcbed 6285 int op_64_bit, r;
8776e519 6286
6affcbed 6287 r = kvm_skip_emulated_instruction(vcpu);
5cb56059 6288
55cd8e5a
GN		 6289 if (kvm_hv_hypercall_enabled(vcpu->kvm))
6290 return kvm_hv_hypercall(vcpu);
6291
5fdbf976
MT		 6292 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6293 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6294 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6295 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6296 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
8776e519 6297
229456fc 6298 trace_kvm_hypercall(nr, a0, a1, a2, a3);
2714d1d3 6299
a449c7aa
NA		 6300 op_64_bit = is_64_bit_mode(vcpu);
6301 if (!op_64_bit) {
8776e519
HB		 6302 nr &= 0xFFFFFFFF;
6303 a0 &= 0xFFFFFFFF;
6304 a1 &= 0xFFFFFFFF;
6305 a2 &= 0xFFFFFFFF;
6306 a3 &= 0xFFFFFFFF;
6307 }
6308
07708c4a
JK		 6309 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6310 ret = -KVM_EPERM;
6311 goto out;
6312 }
6313
8776e519 6314 switch (nr) {
b93463aa
AK		 6315 case KVM_HC_VAPIC_POLL_IRQ:
6316 ret = 0;
6317 break;
6aef266c
SV		 6318 case KVM_HC_KICK_CPU:
6319 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6320 ret = 0;
6321 break;
8ef81a9a 6322#ifdef CONFIG_X86_64
55dd00a7
MT		 6323 case KVM_HC_CLOCK_PAIRING:
6324 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
6325 break;
8ef81a9a 6326#endif
8776e519
HB		 6327 default:
6328 ret = -KVM_ENOSYS;
6329 break;
6330 }
07708c4a 6331out:
a449c7aa
NA		 6332 if (!op_64_bit)
6333 ret = (u32)ret;
5fdbf976 6334 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
f11c3a8d 6335 ++vcpu->stat.hypercalls;
2f333bcb 6336 return r;
8776e519
HB		 6337}
6338EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6339
b6785def 6340static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8776e519 6341{
d6aa1000 6342 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8776e519 6343 char instruction[3];
5fdbf976 6344 unsigned long rip = kvm_rip_read(vcpu);
8776e519 6345
8776e519 6346 kvm_x86_ops->patch_hypercall(vcpu, instruction);
8776e519 6347
ce2e852e
DV		 6348 return emulator_write_emulated(ctxt, rip, instruction, 3,
6349 &ctxt->exception);
8776e519
HB		 6350}
6351
851ba692 6352static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
b6c7a5dc 6353{
782d422b
MG		 6354 return vcpu->run->request_interrupt_window &&
6355 likely(!pic_in_kernel(vcpu->kvm));
b6c7a5dc
HB		 6356}
6357
851ba692 6358static void post_kvm_run_save(struct kvm_vcpu *vcpu)
b6c7a5dc 6359{
851ba692
AK		 6360 struct kvm_run *kvm_run = vcpu->run;
6361
91586a3b 6362 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
f077825a 6363 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
2d3ad1f4 6364 kvm_run->cr8 = kvm_get_cr8(vcpu);
b6c7a5dc 6365 kvm_run->apic_base = kvm_get_apic_base(vcpu);
127a457a
MG		 6366 kvm_run->ready_for_interrupt_injection =
6367 pic_in_kernel(vcpu->kvm) ||
782d422b 6368 kvm_vcpu_ready_for_interrupt_injection(vcpu);
b6c7a5dc
HB		 6369}
6370
95ba8273
GN		 6371static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6372{
6373 int max_irr, tpr;
6374
6375 if (!kvm_x86_ops->update_cr8_intercept)
6376 return;
6377
bce87cce 6378 if (!lapic_in_kernel(vcpu))
88c808fd
AK		 6379 return;
6380
d62caabb
AS		 6381 if (vcpu->arch.apicv_active)
6382 return;
6383
8db3baa2
GN		 6384 if (!vcpu->arch.apic->vapic_addr)
6385 max_irr = kvm_lapic_find_highest_irr(vcpu);
6386 else
6387 max_irr = -1;
95ba8273
GN		 6388
6389 if (max_irr != -1)
6390 max_irr >>= 4;
6391
6392 tpr = kvm_lapic_get_cr8(vcpu);
6393
6394 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6395}
6396
b6b8a145 6397static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
95ba8273 6398{
b6b8a145
JK		 6399 int r;
6400
95ba8273 6401 /* try to reinject previous events if any */
664f8e26
WL		 6402 if (vcpu->arch.exception.injected) {
6403 kvm_x86_ops->queue_exception(vcpu);
6404 return 0;
6405 }
6406
6407 /*
6408 * Exceptions must be injected immediately, or the exception
6409 * frame will have the address of the NMI or interrupt handler.
6410 */
6411 if (!vcpu->arch.exception.pending) {
6412 if (vcpu->arch.nmi_injected) {
6413 kvm_x86_ops->set_nmi(vcpu);
6414 return 0;
6415 }
6416
6417 if (vcpu->arch.interrupt.pending) {
6418 kvm_x86_ops->set_irq(vcpu);
6419 return 0;
6420 }
6421 }
6422
6423 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6424 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6425 if (r != 0)
6426 return r;
6427 }
6428
6429 /* try to inject new event if pending */
b59bb7bd 6430 if (vcpu->arch.exception.pending) {
5c1c85d0
AK		 6431 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6432 vcpu->arch.exception.has_error_code,
6433 vcpu->arch.exception.error_code);
d6e8c854 6434
664f8e26
WL		 6435 vcpu->arch.exception.pending = false;
6436 vcpu->arch.exception.injected = true;
6437
d6e8c854
NA		 6438 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6439 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6440 X86_EFLAGS_RF);
6441
6bdf0662
NA		 6442 if (vcpu->arch.exception.nr == DB_VECTOR &&
6443 (vcpu->arch.dr7 & DR7_GD)) {
6444 vcpu->arch.dr7 &= ~DR7_GD;
6445 kvm_update_dr7(vcpu);
6446 }
6447
cfcd20e5 6448 kvm_x86_ops->queue_exception(vcpu);
72d7b374 6449 } else if (vcpu->arch.smi_pending && !is_smm(vcpu) && kvm_x86_ops->smi_allowed(vcpu)) {
c43203ca 6450 vcpu->arch.smi_pending = false;
ee2cd4b7 6451 enter_smm(vcpu);
c43203ca 6452 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
321c5658
YS		 6453 --vcpu->arch.nmi_pending;
6454 vcpu->arch.nmi_injected = true;
6455 kvm_x86_ops->set_nmi(vcpu);
c7c9c56c 6456 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
9242b5b6
BD		 6457 /*
6458 * Because interrupts can be injected asynchronously, we are
6459 * calling check_nested_events again here to avoid a race condition.
6460 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6461 * proposal and current concerns. Perhaps we should be setting
6462 * KVM_REQ_EVENT only on certain events and not unconditionally?
6463 */
6464 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6465 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6466 if (r != 0)
6467 return r;
6468 }
95ba8273 6469 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
66fd3f7f
GN		 6470 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6471 false);
6472 kvm_x86_ops->set_irq(vcpu);
95ba8273
GN		 6473 }
6474 }
ee2cd4b7 6475
b6b8a145 6476 return 0;
95ba8273
GN		 6477}
6478
7460fb4a
AK		 6479static void process_nmi(struct kvm_vcpu *vcpu)
6480{
6481 unsigned limit = 2;
6482
6483 /*
6484 * x86 is limited to one NMI running, and one NMI pending after it.
6485 * If an NMI is already in progress, limit further NMIs to just one.
6486 * Otherwise, allow two (and we'll inject the first one immediately).
6487 */
6488 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6489 limit = 1;
6490
6491 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6492 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6493 kvm_make_request(KVM_REQ_EVENT, vcpu);
6494}
6495
ee2cd4b7 6496static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
660a5d51
PB		 6497{
6498 u32 flags = 0;
6499 flags |= seg->g << 23;
6500 flags |= seg->db << 22;
6501 flags |= seg->l << 21;
6502 flags |= seg->avl << 20;
6503 flags |= seg->present << 15;
6504 flags |= seg->dpl << 13;
6505 flags |= seg->s << 12;
6506 flags |= seg->type << 8;
6507 return flags;
6508}
6509
ee2cd4b7 6510static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
660a5d51
PB		 6511{
6512 struct kvm_segment seg;
6513 int offset;
6514
6515 kvm_get_segment(vcpu, &seg, n);
6516 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6517
6518 if (n < 3)
6519 offset = 0x7f84 + n * 12;
6520 else
6521 offset = 0x7f2c + (n - 3) * 12;
6522
6523 put_smstate(u32, buf, offset + 8, seg.base);
6524 put_smstate(u32, buf, offset + 4, seg.limit);
ee2cd4b7 6525 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
660a5d51
PB		 6526}
6527
efbb288a 6528#ifdef CONFIG_X86_64
ee2cd4b7 6529static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
660a5d51
PB		 6530{
6531 struct kvm_segment seg;
6532 int offset;
6533 u16 flags;
6534
6535 kvm_get_segment(vcpu, &seg, n);
6536 offset = 0x7e00 + n * 16;
6537
ee2cd4b7 6538 flags = enter_smm_get_segment_flags(&seg) >> 8;
660a5d51
PB		 6539 put_smstate(u16, buf, offset, seg.selector);
6540 put_smstate(u16, buf, offset + 2, flags);
6541 put_smstate(u32, buf, offset + 4, seg.limit);
6542 put_smstate(u64, buf, offset + 8, seg.base);
6543}
efbb288a 6544#endif
660a5d51 6545
ee2cd4b7 6546static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
660a5d51
PB		 6547{
6548 struct desc_ptr dt;
6549 struct kvm_segment seg;
6550 unsigned long val;
6551 int i;
6552
6553 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6554 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6555 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6556 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6557
6558 for (i = 0; i < 8; i++)
6559 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6560
6561 kvm_get_dr(vcpu, 6, &val);
6562 put_smstate(u32, buf, 0x7fcc, (u32)val);
6563 kvm_get_dr(vcpu, 7, &val);
6564 put_smstate(u32, buf, 0x7fc8, (u32)val);
6565
6566 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6567 put_smstate(u32, buf, 0x7fc4, seg.selector);
6568 put_smstate(u32, buf, 0x7f64, seg.base);
6569 put_smstate(u32, buf, 0x7f60, seg.limit);
ee2cd4b7 6570 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
660a5d51
PB		 6571
6572 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6573 put_smstate(u32, buf, 0x7fc0, seg.selector);
6574 put_smstate(u32, buf, 0x7f80, seg.base);
6575 put_smstate(u32, buf, 0x7f7c, seg.limit);
ee2cd4b7 6576 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
660a5d51
PB		 6577
6578 kvm_x86_ops->get_gdt(vcpu, &dt);
6579 put_smstate(u32, buf, 0x7f74, dt.address);
6580 put_smstate(u32, buf, 0x7f70, dt.size);
6581
6582 kvm_x86_ops->get_idt(vcpu, &dt);
6583 put_smstate(u32, buf, 0x7f58, dt.address);
6584 put_smstate(u32, buf, 0x7f54, dt.size);
6585
6586 for (i = 0; i < 6; i++)
ee2cd4b7 6587 enter_smm_save_seg_32(vcpu, buf, i);
660a5d51
PB		 6588
6589 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6590
6591 /* revision id */
6592 put_smstate(u32, buf, 0x7efc, 0x00020000);
6593 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6594}
6595
ee2cd4b7 6596static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
660a5d51
PB		 6597{
6598#ifdef CONFIG_X86_64
6599 struct desc_ptr dt;
6600 struct kvm_segment seg;
6601 unsigned long val;
6602 int i;
6603
6604 for (i = 0; i < 16; i++)
6605 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6606
6607 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6608 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6609
6610 kvm_get_dr(vcpu, 6, &val);
6611 put_smstate(u64, buf, 0x7f68, val);
6612 kvm_get_dr(vcpu, 7, &val);
6613 put_smstate(u64, buf, 0x7f60, val);
6614
6615 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6616 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6617 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6618
6619 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6620
6621 /* revision id */
6622 put_smstate(u32, buf, 0x7efc, 0x00020064);
6623
6624 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6625
6626 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6627 put_smstate(u16, buf, 0x7e90, seg.selector);
ee2cd4b7 6628 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
660a5d51
PB		 6629 put_smstate(u32, buf, 0x7e94, seg.limit);
6630 put_smstate(u64, buf, 0x7e98, seg.base);
6631
6632 kvm_x86_ops->get_idt(vcpu, &dt);
6633 put_smstate(u32, buf, 0x7e84, dt.size);
6634 put_smstate(u64, buf, 0x7e88, dt.address);
6635
6636 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6637 put_smstate(u16, buf, 0x7e70, seg.selector);
ee2cd4b7 6638 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
660a5d51
PB		 6639 put_smstate(u32, buf, 0x7e74, seg.limit);
6640 put_smstate(u64, buf, 0x7e78, seg.base);
6641
6642 kvm_x86_ops->get_gdt(vcpu, &dt);
6643 put_smstate(u32, buf, 0x7e64, dt.size);
6644 put_smstate(u64, buf, 0x7e68, dt.address);
6645
6646 for (i = 0; i < 6; i++)
ee2cd4b7 6647 enter_smm_save_seg_64(vcpu, buf, i);
660a5d51
PB		 6648#else
6649 WARN_ON_ONCE(1);
6650#endif
6651}
6652
ee2cd4b7 6653static void enter_smm(struct kvm_vcpu *vcpu)
64d60670 6654{
660a5d51 6655 struct kvm_segment cs, ds;
18c3626e 6656 struct desc_ptr dt;
660a5d51
PB		 6657 char buf[512];
6658 u32 cr0;
6659
660a5d51 6660 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
660a5d51 6661 memset(buf, 0, 512);
d6321d49 6662 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
ee2cd4b7 6663 enter_smm_save_state_64(vcpu, buf);
660a5d51 6664 else
ee2cd4b7 6665 enter_smm_save_state_32(vcpu, buf);
660a5d51 6666
0234bf88
LP		 6667 /*
6668 * Give pre_enter_smm() a chance to make ISA-specific changes to the
6669 * vCPU state (e.g. leave guest mode) after we've saved the state into
6670 * the SMM state-save area.
6671 */
6672 kvm_x86_ops->pre_enter_smm(vcpu, buf);
6673
6674 vcpu->arch.hflags |= HF_SMM_MASK;
54bf36aa 6675 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
660a5d51
PB		 6676
6677 if (kvm_x86_ops->get_nmi_mask(vcpu))
6678 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6679 else
6680 kvm_x86_ops->set_nmi_mask(vcpu, true);
6681
6682 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6683 kvm_rip_write(vcpu, 0x8000);
6684
6685 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6686 kvm_x86_ops->set_cr0(vcpu, cr0);
6687 vcpu->arch.cr0 = cr0;
6688
6689 kvm_x86_ops->set_cr4(vcpu, 0);
6690
18c3626e
PB		 6691 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6692 dt.address = dt.size = 0;
6693 kvm_x86_ops->set_idt(vcpu, &dt);
6694
660a5d51
PB		 6695 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6696
6697 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6698 cs.base = vcpu->arch.smbase;
6699
6700 ds.selector = 0;
6701 ds.base = 0;
6702
6703 cs.limit = ds.limit = 0xffffffff;
6704 cs.type = ds.type = 0x3;
6705 cs.dpl = ds.dpl = 0;
6706 cs.db = ds.db = 0;
6707 cs.s = ds.s = 1;
6708 cs.l = ds.l = 0;
6709 cs.g = ds.g = 1;
6710 cs.avl = ds.avl = 0;
6711 cs.present = ds.present = 1;
6712 cs.unusable = ds.unusable = 0;
6713 cs.padding = ds.padding = 0;
6714
6715 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6716 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6717 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6718 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6719 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6720 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6721
d6321d49 6722 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
660a5d51
PB		 6723 kvm_x86_ops->set_efer(vcpu, 0);
6724
6725 kvm_update_cpuid(vcpu);
6726 kvm_mmu_reset_context(vcpu);
64d60670
PB		 6727}
6728
ee2cd4b7 6729static void process_smi(struct kvm_vcpu *vcpu)
c43203ca
PB		 6730{
6731 vcpu->arch.smi_pending = true;
6732 kvm_make_request(KVM_REQ_EVENT, vcpu);
6733}
6734
2860c4b1
PB		 6735void kvm_make_scan_ioapic_request(struct kvm *kvm)
6736{
6737 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6738}
6739
3d81bc7e 6740static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
c7c9c56c 6741{
5c919412
AS		 6742 u64 eoi_exit_bitmap[4];
6743
3d81bc7e
YZ		 6744 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6745 return;
c7c9c56c 6746
6308630b 6747 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
c7c9c56c 6748
b053b2ae 6749 if (irqchip_split(vcpu->kvm))
6308630b 6750 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
db2bdcbb 6751 else {
76dfafd5 6752 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
d62caabb 6753 kvm_x86_ops->sync_pir_to_irr(vcpu);
6308630b 6754 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
db2bdcbb 6755 }
5c919412
AS		 6756 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6757 vcpu_to_synic(vcpu)->vec_bitmap, 256);
6758 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
c7c9c56c
YZ		 6759}
6760
a70656b6
RK		 6761static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6762{
6763 ++vcpu->stat.tlb_flush;
6764 kvm_x86_ops->tlb_flush(vcpu);
6765}
6766
b1394e74
RK		 6767void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
6768 unsigned long start, unsigned long end)
6769{
6770 unsigned long apic_address;
6771
6772 /*
6773 * The physical address of apic access page is stored in the VMCS.
6774 * Update it when it becomes invalid.
6775 */
6776 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6777 if (start <= apic_address && apic_address < end)
6778 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6779}
6780
4256f43f
TC		 6781void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6782{
c24ae0dc
TC		 6783 struct page *page = NULL;
6784
35754c98 6785 if (!lapic_in_kernel(vcpu))
f439ed27
PB		 6786 return;
6787
4256f43f
TC		 6788 if (!kvm_x86_ops->set_apic_access_page_addr)
6789 return;
6790
c24ae0dc 6791 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
e8fd5e9e
AA		 6792 if (is_error_page(page))
6793 return;
c24ae0dc
TC		 6794 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6795
6796 /*
6797 * Do not pin apic access page in memory, the MMU notifier
6798 * will call us again if it is migrated or swapped out.
6799 */
6800 put_page(page);
4256f43f
TC		 6801}
6802EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6803
9357d939 6804/*
362c698f 6805 * Returns 1 to let vcpu_run() continue the guest execution loop without
9357d939
TY		 6806 * exiting to the userspace. Otherwise, the value will be returned to the
6807 * userspace.
6808 */
851ba692 6809static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
b6c7a5dc
HB		 6810{
6811 int r;
62a193ed
MG		 6812 bool req_int_win =
6813 dm_request_for_irq_injection(vcpu) &&
6814 kvm_cpu_accept_dm_intr(vcpu);
6815
730dca42 6816 bool req_immediate_exit = false;
b6c7a5dc 6817
2fa6e1e1 6818 if (kvm_request_pending(vcpu)) {
a8eeb04a 6819 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
2e53d63a 6820 kvm_mmu_unload(vcpu);
a8eeb04a 6821 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
2f599714 6822 __kvm_migrate_timers(vcpu);
d828199e
MT		 6823 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6824 kvm_gen_update_masterclock(vcpu->kvm);
0061d53d
MT		 6825 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6826 kvm_gen_kvmclock_update(vcpu);
34c238a1
ZA		 6827 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6828 r = kvm_guest_time_update(vcpu);
8cfdc000
ZA		 6829 if (unlikely(r))
6830 goto out;
6831 }
a8eeb04a 6832 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
4731d4c7 6833 kvm_mmu_sync_roots(vcpu);
a8eeb04a 6834 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
a70656b6 6835 kvm_vcpu_flush_tlb(vcpu);
a8eeb04a 6836 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
851ba692 6837 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
b93463aa
AK		 6838 r = 0;
6839 goto out;
6840 }
a8eeb04a 6841 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
851ba692 6842 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
bbeac283 6843 vcpu->mmio_needed = 0;
71c4dfaf
JR		 6844 r = 0;
6845 goto out;
6846 }
af585b92
GN		 6847 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6848 /* Page is swapped out. Do synthetic halt */
6849 vcpu->arch.apf.halted = true;
6850 r = 1;
6851 goto out;
6852 }
c9aaa895
GC		 6853 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6854 record_steal_time(vcpu);
64d60670
PB		 6855 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6856 process_smi(vcpu);
7460fb4a
AK		 6857 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6858 process_nmi(vcpu);
f5132b01 6859 if (kvm_check_request(KVM_REQ_PMU, vcpu))
c6702c9d 6860 kvm_pmu_handle_event(vcpu);
f5132b01 6861 if (kvm_check_request(KVM_REQ_PMI, vcpu))
c6702c9d 6862 kvm_pmu_deliver_pmi(vcpu);
7543a635
SR		 6863 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6864 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6865 if (test_bit(vcpu->arch.pending_ioapic_eoi,
6308630b 6866 vcpu->arch.ioapic_handled_vectors)) {
7543a635
SR		 6867 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6868 vcpu->run->eoi.vector =
6869 vcpu->arch.pending_ioapic_eoi;
6870 r = 0;
6871 goto out;
6872 }
6873 }
3d81bc7e
YZ		 6874 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6875 vcpu_scan_ioapic(vcpu);
4256f43f
TC		 6876 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6877 kvm_vcpu_reload_apic_access_page(vcpu);
2ce79189
AS		 6878 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6879 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6880 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6881 r = 0;
6882 goto out;
6883 }
e516cebb
AS		 6884 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6885 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6886 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6887 r = 0;
6888 goto out;
6889 }
db397571
AS		 6890 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6891 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6892 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6893 r = 0;
6894 goto out;
6895 }
f3b138c5
AS		 6896
6897 /*
6898 * KVM_REQ_HV_STIMER has to be processed after
6899 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6900 * depend on the guest clock being up-to-date
6901 */
1f4b34f8
AS		 6902 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6903 kvm_hv_process_stimers(vcpu);
2f52d58c 6904 }
b93463aa 6905
b463a6f7 6906 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
0f1e261e 6907 ++vcpu->stat.req_event;
66450a21
JK		 6908 kvm_apic_accept_events(vcpu);
6909 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6910 r = 1;
6911 goto out;
6912 }
6913
b6b8a145
JK		 6914 if (inject_pending_event(vcpu, req_int_win) != 0)
6915 req_immediate_exit = true;
321c5658 6916 else {
cc3d967f 6917 /* Enable SMI/NMI/IRQ window open exits if needed.
c43203ca 6918 *
cc3d967f
LP		 6919 * SMIs have three cases:
6920 * 1) They can be nested, and then there is nothing to
6921 * do here because RSM will cause a vmexit anyway.
6922 * 2) There is an ISA-specific reason why SMI cannot be
6923 * injected, and the moment when this changes can be
6924 * intercepted.
6925 * 3) Or the SMI can be pending because
6926 * inject_pending_event has completed the injection
6927 * of an IRQ or NMI from the previous vmexit, and
6928 * then we request an immediate exit to inject the
6929 * SMI.
c43203ca
PB		 6930 */
6931 if (vcpu->arch.smi_pending && !is_smm(vcpu))
cc3d967f
LP		 6932 if (!kvm_x86_ops->enable_smi_window(vcpu))
6933 req_immediate_exit = true;
321c5658
YS		 6934 if (vcpu->arch.nmi_pending)
6935 kvm_x86_ops->enable_nmi_window(vcpu);
6936 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6937 kvm_x86_ops->enable_irq_window(vcpu);
664f8e26 6938 WARN_ON(vcpu->arch.exception.pending);
321c5658 6939 }
b463a6f7
AK		 6940
6941 if (kvm_lapic_enabled(vcpu)) {
6942 update_cr8_intercept(vcpu);
6943 kvm_lapic_sync_to_vapic(vcpu);
6944 }
6945 }
6946
d8368af8
AK		 6947 r = kvm_mmu_reload(vcpu);
6948 if (unlikely(r)) {
d905c069 6949 goto cancel_injection;
d8368af8
AK		 6950 }
6951
b6c7a5dc
HB		 6952 preempt_disable();
6953
6954 kvm_x86_ops->prepare_guest_switch(vcpu);
b95234c8
PB		 6955
6956 /*
6957 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
6958 * IPI are then delayed after guest entry, which ensures that they
6959 * result in virtual interrupt delivery.
6960 */
6961 local_irq_disable();
6b7e2d09
XG		 6962 vcpu->mode = IN_GUEST_MODE;
6963
01b71917
MT		 6964 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6965
0f127d12 6966 /*
b95234c8 6967 * 1) We should set ->mode before checking ->requests. Please see
cde9af6e 6968 * the comment in kvm_vcpu_exiting_guest_mode().
b95234c8
PB		 6969 *
6970 * 2) For APICv, we should set ->mode before checking PIR.ON. This
6971 * pairs with the memory barrier implicit in pi_test_and_set_on
6972 * (see vmx_deliver_posted_interrupt).
6973 *
6974 * 3) This also orders the write to mode from any reads to the page
6975 * tables done while the VCPU is running. Please see the comment
6976 * in kvm_flush_remote_tlbs.
6b7e2d09 6977 */
01b71917 6978 smp_mb__after_srcu_read_unlock();
b6c7a5dc 6979
b95234c8
PB		 6980 /*
6981 * This handles the case where a posted interrupt was
6982 * notified with kvm_vcpu_kick.
6983 */
6984 if (kvm_lapic_enabled(vcpu)) {
6985 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
6986 kvm_x86_ops->sync_pir_to_irr(vcpu);
6987 }
32f88400 6988
2fa6e1e1 6989 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
d94e1dc9 6990 || need_resched() || signal_pending(current)) {
6b7e2d09 6991 vcpu->mode = OUTSIDE_GUEST_MODE;
d94e1dc9 6992 smp_wmb();
6c142801
AK		 6993 local_irq_enable();
6994 preempt_enable();
01b71917 6995 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6c142801 6996 r = 1;
d905c069 6997 goto cancel_injection;
6c142801
AK		 6998 }
6999
fc5b7f3b
DM		 7000 kvm_load_guest_xcr0(vcpu);
7001
c43203ca
PB		 7002 if (req_immediate_exit) {
7003 kvm_make_request(KVM_REQ_EVENT, vcpu);
d6185f20 7004 smp_send_reschedule(vcpu->cpu);
c43203ca 7005 }
d6185f20 7006
8b89fe1f
PB		 7007 trace_kvm_entry(vcpu->vcpu_id);
7008 wait_lapic_expire(vcpu);
6edaa530 7009 guest_enter_irqoff();
b6c7a5dc 7010
42dbaa5a 7011 if (unlikely(vcpu->arch.switch_db_regs)) {
42dbaa5a
JK		 7012 set_debugreg(0, 7);
7013 set_debugreg(vcpu->arch.eff_db[0], 0);
7014 set_debugreg(vcpu->arch.eff_db[1], 1);
7015 set_debugreg(vcpu->arch.eff_db[2], 2);
7016 set_debugreg(vcpu->arch.eff_db[3], 3);
c77fb5fe 7017 set_debugreg(vcpu->arch.dr6, 6);
ae561ede 7018 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
42dbaa5a 7019 }
b6c7a5dc 7020
851ba692 7021 kvm_x86_ops->run(vcpu);
b6c7a5dc 7022
c77fb5fe
PB		 7023 /*
7024 * Do this here before restoring debug registers on the host. And
7025 * since we do this before handling the vmexit, a DR access vmexit
7026 * can (a) read the correct value of the debug registers, (b) set
7027 * KVM_DEBUGREG_WONT_EXIT again.
7028 */
7029 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
c77fb5fe
PB		 7030 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
7031 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
70e4da7a
PB		 7032 kvm_update_dr0123(vcpu);
7033 kvm_update_dr6(vcpu);
7034 kvm_update_dr7(vcpu);
7035 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
c77fb5fe
PB		 7036 }
7037
24f1e32c
FW		 7038 /*
7039 * If the guest has used debug registers, at least dr7
7040 * will be disabled while returning to the host.
7041 * If we don't have active breakpoints in the host, we don't
7042 * care about the messed up debug address registers. But if
7043 * we have some of them active, restore the old state.
7044 */
59d8eb53 7045 if (hw_breakpoint_active())
24f1e32c 7046 hw_breakpoint_restore();
42dbaa5a 7047
4ba76538 7048 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1d5f066e 7049
6b7e2d09 7050 vcpu->mode = OUTSIDE_GUEST_MODE;
d94e1dc9 7051 smp_wmb();
a547c6db 7052
fc5b7f3b
DM		 7053 kvm_put_guest_xcr0(vcpu);
7054
a547c6db 7055 kvm_x86_ops->handle_external_intr(vcpu);
b6c7a5dc
HB		 7056
7057 ++vcpu->stat.exits;
7058
f2485b3e 7059 guest_exit_irqoff();
b6c7a5dc 7060
f2485b3e 7061 local_irq_enable();
b6c7a5dc
HB		 7062 preempt_enable();
7063
f656ce01 7064 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3200f405 7065
b6c7a5dc
HB		 7066 /*
7067 * Profile KVM exit RIPs:
7068 */
7069 if (unlikely(prof_on == KVM_PROFILING)) {
5fdbf976
MT		 7070 unsigned long rip = kvm_rip_read(vcpu);
7071 profile_hit(KVM_PROFILING, (void *)rip);
b6c7a5dc
HB		 7072 }
7073
cc578287
ZA		 7074 if (unlikely(vcpu->arch.tsc_always_catchup))
7075 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
298101da 7076
5cfb1d5a
MT		 7077 if (vcpu->arch.apic_attention)
7078 kvm_lapic_sync_from_vapic(vcpu);
b93463aa 7079
618232e2 7080 vcpu->arch.gpa_available = false;
851ba692 7081 r = kvm_x86_ops->handle_exit(vcpu);
d905c069
MT		 7082 return r;
7083
7084cancel_injection:
7085 kvm_x86_ops->cancel_injection(vcpu);
ae7a2a3f
MT		 7086 if (unlikely(vcpu->arch.apic_attention))
7087 kvm_lapic_sync_from_vapic(vcpu);
d7690175
MT		 7088out:
7089 return r;
7090}
b6c7a5dc 7091
362c698f
PB		 7092static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7093{
bf9f6ac8
FW		 7094 if (!kvm_arch_vcpu_runnable(vcpu) &&
7095 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
9c8fd1ba
PB		 7096 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7097 kvm_vcpu_block(vcpu);
7098 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
bf9f6ac8
FW		 7099
7100 if (kvm_x86_ops->post_block)
7101 kvm_x86_ops->post_block(vcpu);
7102
9c8fd1ba
PB		 7103 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7104 return 1;
7105 }
362c698f
PB		 7106
7107 kvm_apic_accept_events(vcpu);
7108 switch(vcpu->arch.mp_state) {
7109 case KVM_MP_STATE_HALTED:
7110 vcpu->arch.pv.pv_unhalted = false;
7111 vcpu->arch.mp_state =
7112 KVM_MP_STATE_RUNNABLE;
7113 case KVM_MP_STATE_RUNNABLE:
7114 vcpu->arch.apf.halted = false;
7115 break;
7116 case KVM_MP_STATE_INIT_RECEIVED:
7117 break;
7118 default:
7119 return -EINTR;
7120 break;
7121 }
7122 return 1;
7123}
09cec754 7124
5d9bc648
PB		 7125static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7126{
0ad3bed6
PB		 7127 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7128 kvm_x86_ops->check_nested_events(vcpu, false);
7129
5d9bc648
PB		 7130 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7131 !vcpu->arch.apf.halted);
7132}
7133
362c698f 7134static int vcpu_run(struct kvm_vcpu *vcpu)
d7690175
MT		 7135{
7136 int r;
f656ce01 7137 struct kvm *kvm = vcpu->kvm;
d7690175 7138
f656ce01 7139 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
d7690175 7140
362c698f 7141 for (;;) {
58f800d5 7142 if (kvm_vcpu_running(vcpu)) {
851ba692 7143 r = vcpu_enter_guest(vcpu);
bf9f6ac8 7144 } else {
362c698f 7145 r = vcpu_block(kvm, vcpu);
bf9f6ac8
FW		 7146 }
7147
09cec754
GN		 7148 if (r <= 0)
7149 break;
7150
72875d8a 7151 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
09cec754
GN		 7152 if (kvm_cpu_has_pending_timer(vcpu))
7153 kvm_inject_pending_timer_irqs(vcpu);
7154
782d422b
MG		 7155 if (dm_request_for_irq_injection(vcpu) &&
7156 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
4ca7dd8c
PB		 7157 r = 0;
7158 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
09cec754 7159 ++vcpu->stat.request_irq_exits;
362c698f 7160 break;
09cec754 7161 }
af585b92
GN		 7162
7163 kvm_check_async_pf_completion(vcpu);
7164
09cec754
GN		 7165 if (signal_pending(current)) {
7166 r = -EINTR;
851ba692 7167 vcpu->run->exit_reason = KVM_EXIT_INTR;
09cec754 7168 ++vcpu->stat.signal_exits;
362c698f 7169 break;
09cec754
GN		 7170 }
7171 if (need_resched()) {
f656ce01 7172 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
c08ac06a 7173 cond_resched();
f656ce01 7174 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
d7690175 7175 }
b6c7a5dc
HB		 7176 }
7177
f656ce01 7178 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
b6c7a5dc
HB		 7179
7180 return r;
7181}
7182
716d51ab
GN		 7183static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7184{
7185 int r;
7186 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7187 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7188 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7189 if (r != EMULATE_DONE)
7190 return 0;
7191 return 1;
7192}
7193
7194static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7195{
7196 BUG_ON(!vcpu->arch.pio.count);
7197
7198 return complete_emulated_io(vcpu);
7199}
7200
f78146b0
AK		 7201/*
7202 * Implements the following, as a state machine:
7203 *
7204 * read:
7205 * for each fragment
87da7e66
XG		 7206 * for each mmio piece in the fragment
7207 * write gpa, len
7208 * exit
7209 * copy data
f78146b0
AK		 7210 * execute insn
7211 *
7212 * write:
7213 * for each fragment
87da7e66
XG		 7214 * for each mmio piece in the fragment
7215 * write gpa, len
7216 * copy data
7217 * exit
f78146b0 7218 */
716d51ab 7219static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
5287f194
AK		 7220{
7221 struct kvm_run *run = vcpu->run;
f78146b0 7222 struct kvm_mmio_fragment *frag;
87da7e66 7223 unsigned len;
5287f194 7224
716d51ab 7225 BUG_ON(!vcpu->mmio_needed);
5287f194 7226
716d51ab 7227 /* Complete previous fragment */
87da7e66
XG		 7228 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7229 len = min(8u, frag->len);
716d51ab 7230 if (!vcpu->mmio_is_write)
87da7e66
XG		 7231 memcpy(frag->data, run->mmio.data, len);
7232
7233 if (frag->len <= 8) {
7234 /* Switch to the next fragment. */
7235 frag++;
7236 vcpu->mmio_cur_fragment++;
7237 } else {
7238 /* Go forward to the next mmio piece. */
7239 frag->data += len;
7240 frag->gpa += len;
7241 frag->len -= len;
7242 }
7243
a08d3b3b 7244 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
716d51ab 7245 vcpu->mmio_needed = 0;
0912c977
PB		 7246
7247 /* FIXME: return into emulator if single-stepping. */
cef4dea0 7248 if (vcpu->mmio_is_write)
716d51ab
GN		 7249 return 1;
7250 vcpu->mmio_read_completed = 1;
7251 return complete_emulated_io(vcpu);
7252 }
87da7e66 7253
716d51ab
GN		 7254 run->exit_reason = KVM_EXIT_MMIO;
7255 run->mmio.phys_addr = frag->gpa;
7256 if (vcpu->mmio_is_write)
87da7e66
XG		 7257 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7258 run->mmio.len = min(8u, frag->len);
716d51ab
GN		 7259 run->mmio.is_write = vcpu->mmio_is_write;
7260 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7261 return 0;
5287f194
AK		 7262}
7263
716d51ab 7264
b6c7a5dc
HB		 7265int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7266{
7267 int r;
b6c7a5dc 7268
20b7035c 7269 kvm_sigset_activate(vcpu);
ac9f6dc0 7270
5663d8f9
PX		 7271 kvm_load_guest_fpu(vcpu);
7272
a4535290 7273 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
2f173d26
JS		 7274 if (kvm_run->immediate_exit) {
7275 r = -EINTR;
7276 goto out;
7277 }
b6c7a5dc 7278 kvm_vcpu_block(vcpu);
66450a21 7279 kvm_apic_accept_events(vcpu);
72875d8a 7280 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
ac9f6dc0 7281 r = -EAGAIN;
a0595000
JS		 7282 if (signal_pending(current)) {
7283 r = -EINTR;
7284 vcpu->run->exit_reason = KVM_EXIT_INTR;
7285 ++vcpu->stat.signal_exits;
7286 }
ac9f6dc0 7287 goto out;
b6c7a5dc
HB		 7288 }
7289
b6c7a5dc 7290 /* re-sync apic's tpr */
35754c98 7291 if (!lapic_in_kernel(vcpu)) {
eea1cff9
AP		 7292 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7293 r = -EINVAL;
7294 goto out;
7295 }
7296 }
b6c7a5dc 7297
716d51ab
GN		 7298 if (unlikely(vcpu->arch.complete_userspace_io)) {
7299 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7300 vcpu->arch.complete_userspace_io = NULL;
7301 r = cui(vcpu);
7302 if (r <= 0)
5663d8f9 7303 goto out;
716d51ab
GN		 7304 } else
7305 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
5287f194 7306
460df4c1
PB		 7307 if (kvm_run->immediate_exit)
7308 r = -EINTR;
7309 else
7310 r = vcpu_run(vcpu);
b6c7a5dc
HB		 7311
7312out:
5663d8f9 7313 kvm_put_guest_fpu(vcpu);
f1d86e46 7314 post_kvm_run_save(vcpu);
20b7035c 7315 kvm_sigset_deactivate(vcpu);
b6c7a5dc 7316
b6c7a5dc
HB		 7317 return r;
7318}
7319
7320int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7321{
7ae441ea
GN		 7322 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
7323 /*
7324 * We are here if userspace calls get_regs() in the middle of
7325 * instruction emulation. Registers state needs to be copied
4a969980 7326 * back from emulation context to vcpu. Userspace shouldn't do
7ae441ea
GN		 7327 * that usually, but some bad designed PV devices (vmware
7328 * backdoor interface) need this to work
7329 */
dd856efa 7330 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7ae441ea
GN		 7331 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7332 }
5fdbf976
MT		 7333 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
7334 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
7335 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
7336 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
7337 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
7338 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
7339 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7340 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
b6c7a5dc 7341#ifdef CONFIG_X86_64
5fdbf976
MT		 7342 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7343 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7344 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7345 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7346 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7347 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7348 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7349 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
b6c7a5dc
HB		 7350#endif
7351
5fdbf976 7352 regs->rip = kvm_rip_read(vcpu);
91586a3b 7353 regs->rflags = kvm_get_rflags(vcpu);
b6c7a5dc 7354
b6c7a5dc
HB		 7355 return 0;
7356}
7357
7358int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7359{
7ae441ea
GN		 7360 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7361 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7362
5fdbf976
MT		 7363 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7364 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7365 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7366 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7367 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7368 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7369 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7370 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
b6c7a5dc 7371#ifdef CONFIG_X86_64
5fdbf976
MT		 7372 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7373 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7374 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7375 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7376 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7377 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7378 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7379 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
b6c7a5dc
HB		 7380#endif
7381
5fdbf976 7382 kvm_rip_write(vcpu, regs->rip);
d73235d1 7383 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
b6c7a5dc 7384
b4f14abd
JK		 7385 vcpu->arch.exception.pending = false;
7386
3842d135
AK		 7387 kvm_make_request(KVM_REQ_EVENT, vcpu);
7388
b6c7a5dc
HB		 7389 return 0;
7390}
7391
b6c7a5dc
HB		 7392void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7393{
7394 struct kvm_segment cs;
7395
3e6e0aab 7396 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
b6c7a5dc
HB		 7397 *db = cs.db;
7398 *l = cs.l;
7399}
7400EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7401
7402int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7403 struct kvm_sregs *sregs)
7404{
89a27f4d 7405 struct desc_ptr dt;
b6c7a5dc 7406
3e6e0aab
GT		 7407 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7408 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7409 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7410 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7411 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7412 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
b6c7a5dc 7413
3e6e0aab
GT		 7414 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7415 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
b6c7a5dc
HB		 7416
7417 kvm_x86_ops->get_idt(vcpu, &dt);
89a27f4d
GN		 7418 sregs->idt.limit = dt.size;
7419 sregs->idt.base = dt.address;
b6c7a5dc 7420 kvm_x86_ops->get_gdt(vcpu, &dt);
89a27f4d
GN		 7421 sregs->gdt.limit = dt.size;
7422 sregs->gdt.base = dt.address;
b6c7a5dc 7423
4d4ec087 7424 sregs->cr0 = kvm_read_cr0(vcpu);
ad312c7c 7425 sregs->cr2 = vcpu->arch.cr2;
9f8fe504 7426 sregs->cr3 = kvm_read_cr3(vcpu);
fc78f519 7427 sregs->cr4 = kvm_read_cr4(vcpu);
2d3ad1f4 7428 sregs->cr8 = kvm_get_cr8(vcpu);
f6801dff 7429 sregs->efer = vcpu->arch.efer;
b6c7a5dc
HB		 7430 sregs->apic_base = kvm_get_apic_base(vcpu);
7431
923c61bb 7432 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
b6c7a5dc 7433
36752c9b 7434 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
14d0bc1f
GN		 7435 set_bit(vcpu->arch.interrupt.nr,
7436 (unsigned long *)sregs->interrupt_bitmap);
16d7a191 7437
b6c7a5dc
HB		 7438 return 0;
7439}
7440
62d9f0db
MT		 7441int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7442 struct kvm_mp_state *mp_state)
7443{
66450a21 7444 kvm_apic_accept_events(vcpu);
6aef266c
SV		 7445 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7446 vcpu->arch.pv.pv_unhalted)
7447 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7448 else
7449 mp_state->mp_state = vcpu->arch.mp_state;
7450
62d9f0db
MT		 7451 return 0;
7452}
7453
7454int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7455 struct kvm_mp_state *mp_state)
7456{
bce87cce 7457 if (!lapic_in_kernel(vcpu) &&
66450a21
JK		 7458 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7459 return -EINVAL;
7460
28bf2888
DH		 7461 /* INITs are latched while in SMM */
7462 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
7463 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
7464 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
7465 return -EINVAL;
7466
66450a21
JK		 7467 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7468 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7469 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7470 } else
7471 vcpu->arch.mp_state = mp_state->mp_state;
3842d135 7472 kvm_make_request(KVM_REQ_EVENT, vcpu);
62d9f0db
MT		 7473 return 0;
7474}
7475
7f3d35fd
KW		 7476int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7477 int reason, bool has_error_code, u32 error_code)
b6c7a5dc 7478{
9d74191a 7479 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8ec4722d 7480 int ret;
e01c2426 7481
8ec4722d 7482 init_emulate_ctxt(vcpu);
c697518a 7483
7f3d35fd 7484 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9d74191a 7485 has_error_code, error_code);
c697518a 7486
c697518a 7487 if (ret)
19d04437 7488 return EMULATE_FAIL;
37817f29 7489
9d74191a
TY		 7490 kvm_rip_write(vcpu, ctxt->eip);
7491 kvm_set_rflags(vcpu, ctxt->eflags);
3842d135 7492 kvm_make_request(KVM_REQ_EVENT, vcpu);
19d04437 7493 return EMULATE_DONE;
37817f29
IE		 7494}
7495EXPORT_SYMBOL_GPL(kvm_task_switch);
7496
f2981033
LT		 7497int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
7498{
7499 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG_BIT)) {
7500 /*
7501 * When EFER.LME and CR0.PG are set, the processor is in
7502 * 64-bit mode (though maybe in a 32-bit code segment).
7503 * CR4.PAE and EFER.LMA must be set.
7504 */
7505 if (!(sregs->cr4 & X86_CR4_PAE_BIT)
7506 || !(sregs->efer & EFER_LMA))
7507 return -EINVAL;
7508 } else {
7509 /*
7510 * Not in 64-bit mode: EFER.LMA is clear and the code
7511 * segment cannot be 64-bit.
7512 */
7513 if (sregs->efer & EFER_LMA || sregs->cs.l)
7514 return -EINVAL;
7515 }
7516
7517 return 0;
7518}
7519
b6c7a5dc
HB		 7520int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7521 struct kvm_sregs *sregs)
7522{
58cb628d 7523 struct msr_data apic_base_msr;
b6c7a5dc 7524 int mmu_reset_needed = 0;
63f42e02 7525 int pending_vec, max_bits, idx;
89a27f4d 7526 struct desc_ptr dt;
b6c7a5dc 7527
d6321d49
RK		 7528 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
7529 (sregs->cr4 & X86_CR4_OSXSAVE))
6d1068b3
PM		 7530 return -EINVAL;
7531
f2981033
LT		 7532 if (kvm_valid_sregs(vcpu, sregs))
7533 return -EINVAL;
7534
d3802286
JM		 7535 apic_base_msr.data = sregs->apic_base;
7536 apic_base_msr.host_initiated = true;
7537 if (kvm_set_apic_base(vcpu, &apic_base_msr))
6d1068b3
PM		 7538 return -EINVAL;
7539
89a27f4d
GN		 7540 dt.size = sregs->idt.limit;
7541 dt.address = sregs->idt.base;
b6c7a5dc 7542 kvm_x86_ops->set_idt(vcpu, &dt);
89a27f4d
GN		 7543 dt.size = sregs->gdt.limit;
7544 dt.address = sregs->gdt.base;
b6c7a5dc
HB		 7545 kvm_x86_ops->set_gdt(vcpu, &dt);
7546
ad312c7c 7547 vcpu->arch.cr2 = sregs->cr2;
9f8fe504 7548 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
dc7e795e 7549 vcpu->arch.cr3 = sregs->cr3;
aff48baa 7550 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
b6c7a5dc 7551
2d3ad1f4 7552 kvm_set_cr8(vcpu, sregs->cr8);
b6c7a5dc 7553
f6801dff 7554 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
b6c7a5dc 7555 kvm_x86_ops->set_efer(vcpu, sregs->efer);
b6c7a5dc 7556
4d4ec087 7557 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
b6c7a5dc 7558 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
d7306163 7559 vcpu->arch.cr0 = sregs->cr0;
b6c7a5dc 7560
fc78f519 7561 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
b6c7a5dc 7562 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
b9baba86 7563 if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
00b27a3e 7564 kvm_update_cpuid(vcpu);
63f42e02
XG		 7565
7566 idx = srcu_read_lock(&vcpu->kvm->srcu);
7c93be44 7567 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
9f8fe504 7568 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7c93be44
MT		 7569 mmu_reset_needed = 1;
7570 }
63f42e02 7571 srcu_read_unlock(&vcpu->kvm->srcu, idx);
b6c7a5dc
HB		 7572
7573 if (mmu_reset_needed)
7574 kvm_mmu_reset_context(vcpu);
7575
a50abc3b 7576 max_bits = KVM_NR_INTERRUPTS;
923c61bb
GN		 7577 pending_vec = find_first_bit(
7578 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7579 if (pending_vec < max_bits) {
66fd3f7f 7580 kvm_queue_interrupt(vcpu, pending_vec, false);
923c61bb 7581 pr_debug("Set back pending irq %d\n", pending_vec);
b6c7a5dc
HB		 7582 }
7583
3e6e0aab
GT		 7584 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7585 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7586 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7587 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7588 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7589 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
b6c7a5dc 7590
3e6e0aab
GT		 7591 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7592 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
b6c7a5dc 7593
5f0269f5
ME		 7594 update_cr8_intercept(vcpu);
7595
9c3e4aab 7596 /* Older userspace won't unhalt the vcpu on reset. */
c5af89b6 7597 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9c3e4aab 7598 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
3eeb3288 7599 !is_protmode(vcpu))
9c3e4aab
MT		 7600 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7601
3842d135
AK		 7602 kvm_make_request(KVM_REQ_EVENT, vcpu);
7603
b6c7a5dc
HB		 7604 return 0;
7605}
7606
d0bfb940
JK		 7607int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7608 struct kvm_guest_debug *dbg)
b6c7a5dc 7609{
355be0b9 7610 unsigned long rflags;
ae675ef0 7611 int i, r;
b6c7a5dc 7612
4f926bf2
JK		 7613 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7614 r = -EBUSY;
7615 if (vcpu->arch.exception.pending)
2122ff5e 7616 goto out;
4f926bf2
JK		 7617 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7618 kvm_queue_exception(vcpu, DB_VECTOR);
7619 else
7620 kvm_queue_exception(vcpu, BP_VECTOR);
7621 }
7622
91586a3b
JK		 7623 /*
7624 * Read rflags as long as potentially injected trace flags are still
7625 * filtered out.
7626 */
7627 rflags = kvm_get_rflags(vcpu);
355be0b9
JK		 7628
7629 vcpu->guest_debug = dbg->control;
7630 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7631 vcpu->guest_debug = 0;
7632
7633 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
ae675ef0
JK		 7634 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7635 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
c8639010 7636 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
ae675ef0
JK		 7637 } else {
7638 for (i = 0; i < KVM_NR_DB_REGS; i++)
7639 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
ae675ef0 7640 }
c8639010 7641 kvm_update_dr7(vcpu);
ae675ef0 7642
f92653ee
JK		 7643 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7644 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7645 get_segment_base(vcpu, VCPU_SREG_CS);
94fe45da 7646
91586a3b
JK		 7647 /*
7648 * Trigger an rflags update that will inject or remove the trace
7649 * flags.
7650 */
7651 kvm_set_rflags(vcpu, rflags);
b6c7a5dc 7652
a96036b8 7653 kvm_x86_ops->update_bp_intercept(vcpu);
b6c7a5dc 7654
4f926bf2 7655 r = 0;
d0bfb940 7656
2122ff5e 7657out:
b6c7a5dc
HB		 7658
7659 return r;
7660}
7661
8b006791
ZX		 7662/*
7663 * Translate a guest virtual address to a guest physical address.
7664 */
7665int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7666 struct kvm_translation *tr)
7667{
7668 unsigned long vaddr = tr->linear_address;
7669 gpa_t gpa;
f656ce01 7670 int idx;
8b006791 7671
f656ce01 7672 idx = srcu_read_lock(&vcpu->kvm->srcu);
1871c602 7673 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
f656ce01 7674 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8b006791
ZX		 7675 tr->physical_address = gpa;
7676 tr->valid = gpa != UNMAPPED_GVA;
7677 tr->writeable = 1;
7678 tr->usermode = 0;
8b006791
ZX		 7679
7680 return 0;
7681}
7682
d0752060
HB		 7683int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7684{
c47ada30 7685 struct fxregs_state *fxsave =
7366ed77 7686 &vcpu->arch.guest_fpu.state.fxsave;
d0752060 7687
d0752060
HB		 7688 memcpy(fpu->fpr, fxsave->st_space, 128);
7689 fpu->fcw = fxsave->cwd;
7690 fpu->fsw = fxsave->swd;
7691 fpu->ftwx = fxsave->twd;
7692 fpu->last_opcode = fxsave->fop;
7693 fpu->last_ip = fxsave->rip;
7694 fpu->last_dp = fxsave->rdp;
7695 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7696
d0752060
HB		 7697 return 0;
7698}
7699
7700int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7701{
c47ada30 7702 struct fxregs_state *fxsave =
7366ed77 7703 &vcpu->arch.guest_fpu.state.fxsave;
d0752060 7704
d0752060
HB		 7705 memcpy(fxsave->st_space, fpu->fpr, 128);
7706 fxsave->cwd = fpu->fcw;
7707 fxsave->swd = fpu->fsw;
7708 fxsave->twd = fpu->ftwx;
7709 fxsave->fop = fpu->last_opcode;
7710 fxsave->rip = fpu->last_ip;
7711 fxsave->rdp = fpu->last_dp;
7712 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7713
d0752060
HB		 7714 return 0;
7715}
7716
0ee6a517 7717static void fx_init(struct kvm_vcpu *vcpu)
d0752060 7718{
bf935b0b 7719 fpstate_init(&vcpu->arch.guest_fpu.state);
782511b0 7720 if (boot_cpu_has(X86_FEATURE_XSAVES))
7366ed77 7721 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
df1daba7 7722 host_xcr0 | XSTATE_COMPACTION_ENABLED;
d0752060 7723
2acf923e
DC		 7724 /*
7725 * Ensure guest xcr0 is valid for loading
7726 */
d91cab78 7727 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
2acf923e 7728
ad312c7c 7729 vcpu->arch.cr0 |= X86_CR0_ET;
d0752060 7730}
d0752060 7731
f775b13e 7732/* Swap (qemu) user FPU context for the guest FPU context. */
d0752060
HB		 7733void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7734{
f775b13e
RR		 7735 preempt_disable();
7736 copy_fpregs_to_fpstate(&vcpu->arch.user_fpu);
38cfd5e3
PB		 7737 /* PKRU is separately restored in kvm_x86_ops->run. */
7738 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state,
7739 ~XFEATURE_MASK_PKRU);
f775b13e 7740 preempt_enable();
0c04851c 7741 trace_kvm_fpu(1);
d0752060 7742}
d0752060 7743
f775b13e 7744/* When vcpu_run ends, restore user space FPU context. */
d0752060
HB		 7745void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7746{
f775b13e 7747 preempt_disable();
4f836347 7748 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
f775b13e
RR		 7749 copy_kernel_to_fpregs(&vcpu->arch.user_fpu.state);
7750 preempt_enable();
f096ed85 7751 ++vcpu->stat.fpu_reload;
0c04851c 7752 trace_kvm_fpu(0);
d0752060 7753}
e9b11c17
ZX		 7754
7755void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7756{
bd768e14
IY		 7757 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
7758
12f9a48f 7759 kvmclock_reset(vcpu);
7f1ea208 7760
e9b11c17 7761 kvm_x86_ops->vcpu_free(vcpu);
bd768e14 7762 free_cpumask_var(wbinvd_dirty_mask);
e9b11c17
ZX		 7763}
7764
7765struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7766 unsigned int id)
7767{
c447e76b
LL		 7768 struct kvm_vcpu *vcpu;
7769
6755bae8
ZA		 7770 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7771 printk_once(KERN_WARNING
7772 "kvm: SMP vm created on host with unstable TSC; "
7773 "guest TSC will not be reliable\n");
c447e76b
LL		 7774
7775 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7776
c447e76b 7777 return vcpu;
26e5215f 7778}
e9b11c17 7779
26e5215f
AK		 7780int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7781{
7782 int r;
e9b11c17 7783
19efffa2 7784 kvm_vcpu_mtrr_init(vcpu);
9fc77441
MT		 7785 r = vcpu_load(vcpu);
7786 if (r)
7787 return r;
d28bc9dd 7788 kvm_vcpu_reset(vcpu, false);
8a3c1a33 7789 kvm_mmu_setup(vcpu);
e9b11c17 7790 vcpu_put(vcpu);
26e5215f 7791 return r;
e9b11c17
ZX		 7792}
7793
31928aa5 7794void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
42897d86 7795{
8fe8ab46 7796 struct msr_data msr;
332967a3 7797 struct kvm *kvm = vcpu->kvm;
42897d86 7798
d3457c87
RK		 7799 kvm_hv_vcpu_postcreate(vcpu);
7800
31928aa5
DD		 7801 if (vcpu_load(vcpu))
7802 return;
8fe8ab46
WA		 7803 msr.data = 0x0;
7804 msr.index = MSR_IA32_TSC;
7805 msr.host_initiated = true;
7806 kvm_write_tsc(vcpu, &msr);
42897d86
MT		 7807 vcpu_put(vcpu);
7808
630994b3
MT		 7809 if (!kvmclock_periodic_sync)
7810 return;
7811
332967a3
AJ		 7812 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7813 KVMCLOCK_SYNC_PERIOD);
42897d86
MT		 7814}
7815
d40ccc62 7816void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
e9b11c17 7817{
9fc77441 7818 int r;
344d9588
GN		 7819 vcpu->arch.apf.msr_val = 0;
7820
9fc77441
MT		 7821 r = vcpu_load(vcpu);
7822 BUG_ON(r);
e9b11c17
ZX		 7823 kvm_mmu_unload(vcpu);
7824 vcpu_put(vcpu);
7825
7826 kvm_x86_ops->vcpu_free(vcpu);
7827}
7828
d28bc9dd 7829void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
e9b11c17 7830{
e69fab5d
PB		 7831 vcpu->arch.hflags = 0;
7832
c43203ca 7833 vcpu->arch.smi_pending = 0;
7460fb4a
AK		 7834 atomic_set(&vcpu->arch.nmi_queued, 0);
7835 vcpu->arch.nmi_pending = 0;
448fa4a9 7836 vcpu->arch.nmi_injected = false;
5f7552d4
NA		 7837 kvm_clear_interrupt_queue(vcpu);
7838 kvm_clear_exception_queue(vcpu);
664f8e26 7839 vcpu->arch.exception.pending = false;
448fa4a9 7840
42dbaa5a 7841 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
ae561ede 7842 kvm_update_dr0123(vcpu);
6f43ed01 7843 vcpu->arch.dr6 = DR6_INIT;
73aaf249 7844 kvm_update_dr6(vcpu);
42dbaa5a 7845 vcpu->arch.dr7 = DR7_FIXED_1;
c8639010 7846 kvm_update_dr7(vcpu);
42dbaa5a 7847
1119022c
NA		 7848 vcpu->arch.cr2 = 0;
7849
3842d135 7850 kvm_make_request(KVM_REQ_EVENT, vcpu);
344d9588 7851 vcpu->arch.apf.msr_val = 0;
c9aaa895 7852 vcpu->arch.st.msr_val = 0;
3842d135 7853
12f9a48f
GC		 7854 kvmclock_reset(vcpu);
7855
af585b92
GN		 7856 kvm_clear_async_pf_completion_queue(vcpu);
7857 kvm_async_pf_hash_reset(vcpu);
7858 vcpu->arch.apf.halted = false;
3842d135 7859
a554d207
WL		 7860 if (kvm_mpx_supported()) {
7861 void *mpx_state_buffer;
7862
7863 /*
7864 * To avoid have the INIT path from kvm_apic_has_events() that be
7865 * called with loaded FPU and does not let userspace fix the state.
7866 */
f775b13e
RR		 7867 if (init_event)
7868 kvm_put_guest_fpu(vcpu);
a554d207
WL		 7869 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
7870 XFEATURE_MASK_BNDREGS);
7871 if (mpx_state_buffer)
7872 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
7873 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
7874 XFEATURE_MASK_BNDCSR);
7875 if (mpx_state_buffer)
7876 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
f775b13e
RR		 7877 if (init_event)
7878 kvm_load_guest_fpu(vcpu);
a554d207
WL		 7879 }
7880
64d60670 7881 if (!init_event) {
d28bc9dd 7882 kvm_pmu_reset(vcpu);
64d60670 7883 vcpu->arch.smbase = 0x30000;
db2336a8
KH		 7884
7885 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
7886 vcpu->arch.msr_misc_features_enables = 0;
a554d207
WL		 7887
7888 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
64d60670 7889 }
f5132b01 7890
66f7b72e
JS		 7891 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7892 vcpu->arch.regs_avail = ~0;
7893 vcpu->arch.regs_dirty = ~0;
7894
a554d207
WL		 7895 vcpu->arch.ia32_xss = 0;
7896
d28bc9dd 7897 kvm_x86_ops->vcpu_reset(vcpu, init_event);
e9b11c17
ZX		 7898}
7899
2b4a273b 7900void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
66450a21
JK		 7901{
7902 struct kvm_segment cs;
7903
7904 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7905 cs.selector = vector << 8;
7906 cs.base = vector << 12;
7907 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7908 kvm_rip_write(vcpu, 0);
e9b11c17
ZX		 7909}
7910
13a34e06 7911int kvm_arch_hardware_enable(void)
e9b11c17 7912{
ca84d1a2
ZA		 7913 struct kvm *kvm;
7914 struct kvm_vcpu *vcpu;
7915 int i;
0dd6a6ed
ZA		 7916 int ret;
7917 u64 local_tsc;
7918 u64 max_tsc = 0;
7919 bool stable, backwards_tsc = false;
18863bdd
AK		 7920
7921 kvm_shared_msr_cpu_online();
13a34e06 7922 ret = kvm_x86_ops->hardware_enable();
0dd6a6ed
ZA		 7923 if (ret != 0)
7924 return ret;
7925
4ea1636b 7926 local_tsc = rdtsc();
0dd6a6ed
ZA		 7927 stable = !check_tsc_unstable();
7928 list_for_each_entry(kvm, &vm_list, vm_list) {
7929 kvm_for_each_vcpu(i, vcpu, kvm) {
7930 if (!stable && vcpu->cpu == smp_processor_id())
105b21bb 7931 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
0dd6a6ed
ZA		 7932 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7933 backwards_tsc = true;
7934 if (vcpu->arch.last_host_tsc > max_tsc)
7935 max_tsc = vcpu->arch.last_host_tsc;
7936 }
7937 }
7938 }
7939
7940 /*
7941 * Sometimes, even reliable TSCs go backwards. This happens on
7942 * platforms that reset TSC during suspend or hibernate actions, but
7943 * maintain synchronization. We must compensate. Fortunately, we can
7944 * detect that condition here, which happens early in CPU bringup,
7945 * before any KVM threads can be running. Unfortunately, we can't
7946 * bring the TSCs fully up to date with real time, as we aren't yet far
7947 * enough into CPU bringup that we know how much real time has actually
108b249c 7948 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
0dd6a6ed
ZA		 7949 * variables that haven't been updated yet.
7950 *
7951 * So we simply find the maximum observed TSC above, then record the
7952 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7953 * the adjustment will be applied. Note that we accumulate
7954 * adjustments, in case multiple suspend cycles happen before some VCPU
7955 * gets a chance to run again. In the event that no KVM threads get a
7956 * chance to run, we will miss the entire elapsed period, as we'll have
7957 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7958 * loose cycle time. This isn't too big a deal, since the loss will be
7959 * uniform across all VCPUs (not to mention the scenario is extremely
7960 * unlikely). It is possible that a second hibernate recovery happens
7961 * much faster than a first, causing the observed TSC here to be
7962 * smaller; this would require additional padding adjustment, which is
7963 * why we set last_host_tsc to the local tsc observed here.
7964 *
7965 * N.B. - this code below runs only on platforms with reliable TSC,
7966 * as that is the only way backwards_tsc is set above. Also note
7967 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7968 * have the same delta_cyc adjustment applied if backwards_tsc
7969 * is detected. Note further, this adjustment is only done once,
7970 * as we reset last_host_tsc on all VCPUs to stop this from being
7971 * called multiple times (one for each physical CPU bringup).
7972 *
4a969980 7973 * Platforms with unreliable TSCs don't have to deal with this, they
0dd6a6ed
ZA		 7974 * will be compensated by the logic in vcpu_load, which sets the TSC to
7975 * catchup mode. This will catchup all VCPUs to real time, but cannot
7976 * guarantee that they stay in perfect synchronization.
7977 */
7978 if (backwards_tsc) {
7979 u64 delta_cyc = max_tsc - local_tsc;
7980 list_for_each_entry(kvm, &vm_list, vm_list) {
a826faf1 7981 kvm->arch.backwards_tsc_observed = true;
0dd6a6ed
ZA		 7982 kvm_for_each_vcpu(i, vcpu, kvm) {
7983 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7984 vcpu->arch.last_host_tsc = local_tsc;
105b21bb 7985 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
0dd6a6ed
ZA		 7986 }
7987
7988 /*
7989 * We have to disable TSC offset matching.. if you were
7990 * booting a VM while issuing an S4 host suspend....
7991 * you may have some problem. Solving this issue is
7992 * left as an exercise to the reader.
7993 */
7994 kvm->arch.last_tsc_nsec = 0;
7995 kvm->arch.last_tsc_write = 0;
7996 }
7997
7998 }
7999 return 0;
e9b11c17
ZX		 8000}
8001
13a34e06 8002void kvm_arch_hardware_disable(void)
e9b11c17 8003{
13a34e06
RK		 8004 kvm_x86_ops->hardware_disable();
8005 drop_user_return_notifiers();
e9b11c17
ZX		 8006}
8007
8008int kvm_arch_hardware_setup(void)
8009{
9e9c3fe4
NA		 8010 int r;
8011
8012 r = kvm_x86_ops->hardware_setup();
8013 if (r != 0)
8014 return r;
8015
35181e86
HZ		 8016 if (kvm_has_tsc_control) {
8017 /*
8018 * Make sure the user can only configure tsc_khz values that
8019 * fit into a signed integer.
8020 * A min value is not calculated needed because it will always
8021 * be 1 on all machines.
8022 */
8023 u64 max = min(0x7fffffffULL,
8024 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
8025 kvm_max_guest_tsc_khz = max;
8026
ad721883 8027 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
35181e86 8028 }
ad721883 8029
9e9c3fe4
NA		 8030 kvm_init_msr_list();
8031 return 0;
e9b11c17
ZX		 8032}
8033
8034void kvm_arch_hardware_unsetup(void)
8035{
8036 kvm_x86_ops->hardware_unsetup();
8037}
8038
8039void kvm_arch_check_processor_compat(void *rtn)
8040{
8041 kvm_x86_ops->check_processor_compatibility(rtn);
d71ba788
PB		 8042}
8043
8044bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
8045{
8046 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
8047}
8048EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
8049
8050bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
8051{
8052 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
e9b11c17
ZX		 8053}
8054
54e9818f 8055struct static_key kvm_no_apic_vcpu __read_mostly;
bce87cce 8056EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
54e9818f 8057
e9b11c17
ZX		 8058int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
8059{
8060 struct page *page;
e9b11c17
ZX		 8061 int r;
8062
b2a05fef 8063 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
9aabc88f 8064 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
26de7988 8065 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
a4535290 8066 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
e9b11c17 8067 else
a4535290 8068 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
e9b11c17
ZX		 8069
8070 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
8071 if (!page) {
8072 r = -ENOMEM;
8073 goto fail;
8074 }
ad312c7c 8075 vcpu->arch.pio_data = page_address(page);
e9b11c17 8076
cc578287 8077 kvm_set_tsc_khz(vcpu, max_tsc_khz);
c285545f 8078
e9b11c17
ZX		 8079 r = kvm_mmu_create(vcpu);
8080 if (r < 0)
8081 goto fail_free_pio_data;
8082
26de7988 8083 if (irqchip_in_kernel(vcpu->kvm)) {
e9b11c17
ZX		 8084 r = kvm_create_lapic(vcpu);
8085 if (r < 0)
8086 goto fail_mmu_destroy;
54e9818f
GN		 8087 } else
8088 static_key_slow_inc(&kvm_no_apic_vcpu);
e9b11c17 8089
890ca9ae
HY		 8090 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
8091 GFP_KERNEL);
8092 if (!vcpu->arch.mce_banks) {
8093 r = -ENOMEM;
443c39bc 8094 goto fail_free_lapic;
890ca9ae
HY		 8095 }
8096 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
8097
f1797359
WY		 8098 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
8099 r = -ENOMEM;
f5f48ee1 8100 goto fail_free_mce_banks;
f1797359 8101 }
f5f48ee1 8102
0ee6a517 8103 fx_init(vcpu);
66f7b72e 8104
4344ee98 8105 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
d7876f1b 8106
5a4f55cd
EK		 8107 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
8108
74545705
RK		 8109 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
8110
af585b92 8111 kvm_async_pf_hash_reset(vcpu);
f5132b01 8112 kvm_pmu_init(vcpu);
af585b92 8113
1c1a9ce9 8114 vcpu->arch.pending_external_vector = -1;
de63ad4c 8115 vcpu->arch.preempted_in_kernel = false;
1c1a9ce9 8116
5c919412
AS		 8117 kvm_hv_vcpu_init(vcpu);
8118
e9b11c17 8119 return 0;
0ee6a517 8120
f5f48ee1
SY		 8121fail_free_mce_banks:
8122 kfree(vcpu->arch.mce_banks);
443c39bc
WY		 8123fail_free_lapic:
8124 kvm_free_lapic(vcpu);
e9b11c17
ZX		 8125fail_mmu_destroy:
8126 kvm_mmu_destroy(vcpu);
8127fail_free_pio_data:
ad312c7c 8128 free_page((unsigned long)vcpu->arch.pio_data);
e9b11c17
ZX		 8129fail:
8130 return r;
8131}
8132
8133void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
8134{
f656ce01
MT		 8135 int idx;
8136
1f4b34f8 8137 kvm_hv_vcpu_uninit(vcpu);
f5132b01 8138 kvm_pmu_destroy(vcpu);
36cb93fd 8139 kfree(vcpu->arch.mce_banks);
e9b11c17 8140 kvm_free_lapic(vcpu);
f656ce01 8141 idx = srcu_read_lock(&vcpu->kvm->srcu);
e9b11c17 8142 kvm_mmu_destroy(vcpu);
f656ce01 8143 srcu_read_unlock(&vcpu->kvm->srcu, idx);
ad312c7c 8144 free_page((unsigned long)vcpu->arch.pio_data);
35754c98 8145 if (!lapic_in_kernel(vcpu))
54e9818f 8146 static_key_slow_dec(&kvm_no_apic_vcpu);
e9b11c17 8147}
d19a9cd2 8148
e790d9ef
RK		 8149void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
8150{
ae97a3b8 8151 kvm_x86_ops->sched_in(vcpu, cpu);
e790d9ef
RK		 8152}
8153
e08b9637 8154int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
d19a9cd2 8155{
e08b9637
CO		 8156 if (type)
8157 return -EINVAL;
8158
6ef768fa 8159 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
f05e70ac 8160 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
365c8868 8161 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
4d5c5d0f 8162 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
e0f0bbc5 8163 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
d19a9cd2 8164
5550af4d
SY		 8165 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
8166 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7a84428a
AW		 8167 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
8168 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
8169 &kvm->arch.irq_sources_bitmap);
5550af4d 8170
038f8c11 8171 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
1e08ec4a 8172 mutex_init(&kvm->arch.apic_map_lock);
3f5ad8be 8173 mutex_init(&kvm->arch.hyperv.hv_lock);
d828199e
MT		 8174 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
8175
108b249c 8176 kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
d828199e 8177 pvclock_update_vm_gtod_copy(kvm);
53f658b3 8178
7e44e449 8179 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
332967a3 8180 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7e44e449 8181
0eb05bf2 8182 kvm_page_track_init(kvm);
13d268ca 8183 kvm_mmu_init_vm(kvm);
0eb05bf2 8184
03543133
SS		 8185 if (kvm_x86_ops->vm_init)
8186 return kvm_x86_ops->vm_init(kvm);
8187
d89f5eff 8188 return 0;
d19a9cd2
ZX		 8189}
8190
8191static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
8192{
9fc77441
MT		 8193 int r;
8194 r = vcpu_load(vcpu);
8195 BUG_ON(r);
d19a9cd2
ZX		 8196 kvm_mmu_unload(vcpu);
8197 vcpu_put(vcpu);
8198}
8199
8200static void kvm_free_vcpus(struct kvm *kvm)
8201{
8202 unsigned int i;
988a2cae 8203 struct kvm_vcpu *vcpu;
d19a9cd2
ZX		 8204
8205 /*
8206 * Unpin any mmu pages first.
8207 */
af585b92
GN		 8208 kvm_for_each_vcpu(i, vcpu, kvm) {
8209 kvm_clear_async_pf_completion_queue(vcpu);
988a2cae 8210 kvm_unload_vcpu_mmu(vcpu);
af585b92 8211 }
988a2cae
GN		 8212 kvm_for_each_vcpu(i, vcpu, kvm)
8213 kvm_arch_vcpu_free(vcpu);
8214
8215 mutex_lock(&kvm->lock);
8216 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
8217 kvm->vcpus[i] = NULL;
d19a9cd2 8218
988a2cae
GN		 8219 atomic_set(&kvm->online_vcpus, 0);
8220 mutex_unlock(&kvm->lock);
d19a9cd2
ZX		 8221}
8222
ad8ba2cd
SY		 8223void kvm_arch_sync_events(struct kvm *kvm)
8224{
332967a3 8225 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7e44e449 8226 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
aea924f6 8227 kvm_free_pit(kvm);
ad8ba2cd
SY		 8228}
8229
1d8007bd 8230int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9da0e4d5
PB		 8231{
8232 int i, r;
25188b99 8233 unsigned long hva;
f0d648bd
PB		 8234 struct kvm_memslots *slots = kvm_memslots(kvm);
8235 struct kvm_memory_slot *slot, old;
9da0e4d5
PB		 8236
8237 /* Called with kvm->slots_lock held. */
1d8007bd
PB		 8238 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
8239 return -EINVAL;
9da0e4d5 8240
f0d648bd
PB		 8241 slot = id_to_memslot(slots, id);
8242 if (size) {
b21629da 8243 if (slot->npages)
f0d648bd
PB		 8244 return -EEXIST;
8245
8246 /*
8247 * MAP_SHARED to prevent internal slot pages from being moved
8248 * by fork()/COW.
8249 */
8250 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
8251 MAP_SHARED | MAP_ANONYMOUS, 0);
8252 if (IS_ERR((void *)hva))
8253 return PTR_ERR((void *)hva);
8254 } else {
8255 if (!slot->npages)
8256 return 0;
8257
8258 hva = 0;
8259 }
8260
8261 old = *slot;
9da0e4d5 8262 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1d8007bd 8263 struct kvm_userspace_memory_region m;
9da0e4d5 8264
1d8007bd
PB		 8265 m.slot = id | (i << 16);
8266 m.flags = 0;
8267 m.guest_phys_addr = gpa;
f0d648bd 8268 m.userspace_addr = hva;
1d8007bd 8269 m.memory_size = size;
9da0e4d5
PB		 8270 r = __kvm_set_memory_region(kvm, &m);
8271 if (r < 0)
8272 return r;
8273 }
8274
f0d648bd
PB		 8275 if (!size) {
8276 r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
8277 WARN_ON(r < 0);
8278 }
8279
9da0e4d5
PB		 8280 return 0;
8281}
8282EXPORT_SYMBOL_GPL(__x86_set_memory_region);
8283
1d8007bd 8284int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9da0e4d5
PB		 8285{
8286 int r;
8287
8288 mutex_lock(&kvm->slots_lock);
1d8007bd 8289 r = __x86_set_memory_region(kvm, id, gpa, size);
9da0e4d5
PB		 8290 mutex_unlock(&kvm->slots_lock);
8291
8292 return r;
8293}
8294EXPORT_SYMBOL_GPL(x86_set_memory_region);
8295
d19a9cd2
ZX		 8296void kvm_arch_destroy_vm(struct kvm *kvm)
8297{
27469d29
AH		 8298 if (current->mm == kvm->mm) {
8299 /*
8300 * Free memory regions allocated on behalf of userspace,
8301 * unless the the memory map has changed due to process exit
8302 * or fd copying.
8303 */
1d8007bd
PB		 8304 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
8305 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
8306 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
27469d29 8307 }
03543133
SS		 8308 if (kvm_x86_ops->vm_destroy)
8309 kvm_x86_ops->vm_destroy(kvm);
c761159c
PX		 8310 kvm_pic_destroy(kvm);
8311 kvm_ioapic_destroy(kvm);
d19a9cd2 8312 kvm_free_vcpus(kvm);
af1bae54 8313 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
13d268ca 8314 kvm_mmu_uninit_vm(kvm);
2beb6dad 8315 kvm_page_track_cleanup(kvm);
d19a9cd2 8316}
0de10343 8317
5587027c 8318void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
db3fe4eb
TY		 8319 struct kvm_memory_slot *dont)
8320{
8321 int i;
8322
d89cc617
TY		 8323 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8324 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
548ef284 8325 kvfree(free->arch.rmap[i]);
d89cc617 8326 free->arch.rmap[i] = NULL;
77d11309 8327 }
d89cc617
TY		 8328 if (i == 0)
8329 continue;
8330
8331 if (!dont || free->arch.lpage_info[i - 1] !=
8332 dont->arch.lpage_info[i - 1]) {
548ef284 8333 kvfree(free->arch.lpage_info[i - 1]);
d89cc617 8334 free->arch.lpage_info[i - 1] = NULL;
db3fe4eb
TY		 8335 }
8336 }
21ebbeda
XG		 8337
8338 kvm_page_track_free_memslot(free, dont);
db3fe4eb
TY		 8339}
8340
5587027c
AK		 8341int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
8342 unsigned long npages)
db3fe4eb
TY		 8343{
8344 int i;
8345
d89cc617 8346 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
92f94f1e 8347 struct kvm_lpage_info *linfo;
db3fe4eb
TY		 8348 unsigned long ugfn;
8349 int lpages;
d89cc617 8350 int level = i + 1;
db3fe4eb
TY		 8351
8352 lpages = gfn_to_index(slot->base_gfn + npages - 1,
8353 slot->base_gfn, level) + 1;
8354
d89cc617 8355 slot->arch.rmap[i] =
a7c3e901 8356 kvzalloc(lpages * sizeof(*slot->arch.rmap[i]), GFP_KERNEL);
d89cc617 8357 if (!slot->arch.rmap[i])
77d11309 8358 goto out_free;
d89cc617
TY		 8359 if (i == 0)
8360 continue;
77d11309 8361
a7c3e901 8362 linfo = kvzalloc(lpages * sizeof(*linfo), GFP_KERNEL);
92f94f1e 8363 if (!linfo)
db3fe4eb
TY		 8364 goto out_free;
8365
92f94f1e
XG		 8366 slot->arch.lpage_info[i - 1] = linfo;
8367
db3fe4eb 8368 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
92f94f1e 8369 linfo[0].disallow_lpage = 1;
db3fe4eb 8370 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
92f94f1e 8371 linfo[lpages - 1].disallow_lpage = 1;
db3fe4eb
TY		 8372 ugfn = slot->userspace_addr >> PAGE_SHIFT;
8373 /*
8374 * If the gfn and userspace address are not aligned wrt each
8375 * other, or if explicitly asked to, disable large page
8376 * support for this slot
8377 */
8378 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
8379 !kvm_largepages_enabled()) {
8380 unsigned long j;
8381
8382 for (j = 0; j < lpages; ++j)
92f94f1e 8383 linfo[j].disallow_lpage = 1;
db3fe4eb
TY		 8384 }
8385 }
8386
21ebbeda
XG		 8387 if (kvm_page_track_create_memslot(slot, npages))
8388 goto out_free;
8389
db3fe4eb
TY		 8390 return 0;
8391
8392out_free:
d89cc617 8393 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
548ef284 8394 kvfree(slot->arch.rmap[i]);
d89cc617
TY		 8395 slot->arch.rmap[i] = NULL;
8396 if (i == 0)
8397 continue;
8398
548ef284 8399 kvfree(slot->arch.lpage_info[i - 1]);
d89cc617 8400 slot->arch.lpage_info[i - 1] = NULL;
db3fe4eb
TY		 8401 }
8402 return -ENOMEM;
8403}
8404
15f46015 8405void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
e59dbe09 8406{
e6dff7d1
TY		 8407 /*
8408 * memslots->generation has been incremented.
8409 * mmio generation may have reached its maximum value.
8410 */
54bf36aa 8411 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
e59dbe09
TY		 8412}
8413
f7784b8e
MT		 8414int kvm_arch_prepare_memory_region(struct kvm *kvm,
8415 struct kvm_memory_slot *memslot,
09170a49 8416 const struct kvm_userspace_memory_region *mem,
7b6195a9 8417 enum kvm_mr_change change)
0de10343 8418{
f7784b8e
MT		 8419 return 0;
8420}
8421
88178fd4
KH		 8422static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8423 struct kvm_memory_slot *new)
8424{
8425 /* Still write protect RO slot */
8426 if (new->flags & KVM_MEM_READONLY) {
8427 kvm_mmu_slot_remove_write_access(kvm, new);
8428 return;
8429 }
8430
8431 /*
8432 * Call kvm_x86_ops dirty logging hooks when they are valid.
8433 *
8434 * kvm_x86_ops->slot_disable_log_dirty is called when:
8435 *
8436 * - KVM_MR_CREATE with dirty logging is disabled
8437 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8438 *
8439 * The reason is, in case of PML, we need to set D-bit for any slots
8440 * with dirty logging disabled in order to eliminate unnecessary GPA
8441 * logging in PML buffer (and potential PML buffer full VMEXT). This
8442 * guarantees leaving PML enabled during guest's lifetime won't have
8443 * any additonal overhead from PML when guest is running with dirty
8444 * logging disabled for memory slots.
8445 *
8446 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8447 * to dirty logging mode.
8448 *
8449 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8450 *
8451 * In case of write protect:
8452 *
8453 * Write protect all pages for dirty logging.
8454 *
8455 * All the sptes including the large sptes which point to this
8456 * slot are set to readonly. We can not create any new large
8457 * spte on this slot until the end of the logging.
8458 *
8459 * See the comments in fast_page_fault().
8460 */
8461 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8462 if (kvm_x86_ops->slot_enable_log_dirty)
8463 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8464 else
8465 kvm_mmu_slot_remove_write_access(kvm, new);
8466 } else {
8467 if (kvm_x86_ops->slot_disable_log_dirty)
8468 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8469 }
8470}
8471
f7784b8e 8472void kvm_arch_commit_memory_region(struct kvm *kvm,
09170a49 8473 const struct kvm_userspace_memory_region *mem,
8482644a 8474 const struct kvm_memory_slot *old,
f36f3f28 8475 const struct kvm_memory_slot *new,
8482644a 8476 enum kvm_mr_change change)
f7784b8e 8477{
8482644a 8478 int nr_mmu_pages = 0;
f7784b8e 8479
48c0e4e9
XG		 8480 if (!kvm->arch.n_requested_mmu_pages)
8481 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8482
48c0e4e9 8483 if (nr_mmu_pages)
0de10343 8484 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
1c91cad4 8485
3ea3b7fa
WL		 8486 /*
8487 * Dirty logging tracks sptes in 4k granularity, meaning that large
8488 * sptes have to be split. If live migration is successful, the guest
8489 * in the source machine will be destroyed and large sptes will be
8490 * created in the destination. However, if the guest continues to run
8491 * in the source machine (for example if live migration fails), small
8492 * sptes will remain around and cause bad performance.
8493 *
8494 * Scan sptes if dirty logging has been stopped, dropping those
8495 * which can be collapsed into a single large-page spte. Later
8496 * page faults will create the large-page sptes.
8497 */
8498 if ((change != KVM_MR_DELETE) &&
8499 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8500 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8501 kvm_mmu_zap_collapsible_sptes(kvm, new);
8502
c972f3b1 8503 /*
88178fd4 8504 * Set up write protection and/or dirty logging for the new slot.
c126d94f 8505 *
88178fd4
KH		 8506 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8507 * been zapped so no dirty logging staff is needed for old slot. For
8508 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8509 * new and it's also covered when dealing with the new slot.
f36f3f28
PB		 8510 *
8511 * FIXME: const-ify all uses of struct kvm_memory_slot.
c972f3b1 8512 */
88178fd4 8513 if (change != KVM_MR_DELETE)
f36f3f28 8514 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
0de10343 8515}
1d737c8a 8516
2df72e9b 8517void kvm_arch_flush_shadow_all(struct kvm *kvm)
34d4cb8f 8518{
6ca18b69 8519 kvm_mmu_invalidate_zap_all_pages(kvm);
34d4cb8f
MT		 8520}
8521
2df72e9b
MT		 8522void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8523 struct kvm_memory_slot *slot)
8524{
ae7cd873 8525 kvm_page_track_flush_slot(kvm, slot);
2df72e9b
MT		 8526}
8527
5d9bc648
PB		 8528static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8529{
8530 if (!list_empty_careful(&vcpu->async_pf.done))
8531 return true;
8532
8533 if (kvm_apic_has_events(vcpu))
8534 return true;
8535
8536 if (vcpu->arch.pv.pv_unhalted)
8537 return true;
8538
a5f01f8e
WL		 8539 if (vcpu->arch.exception.pending)
8540 return true;
8541
47a66eed
Z		 8542 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
8543 (vcpu->arch.nmi_pending &&
8544 kvm_x86_ops->nmi_allowed(vcpu)))
5d9bc648
PB		 8545 return true;
8546
47a66eed
Z		 8547 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
8548 (vcpu->arch.smi_pending && !is_smm(vcpu)))
73917739
PB		 8549 return true;
8550
5d9bc648
PB		 8551 if (kvm_arch_interrupt_allowed(vcpu) &&
8552 kvm_cpu_has_interrupt(vcpu))
8553 return true;
8554
1f4b34f8
AS		 8555 if (kvm_hv_has_stimer_pending(vcpu))
8556 return true;
8557
5d9bc648
PB		 8558 return false;
8559}
8560
1d737c8a
ZX		 8561int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8562{
5d9bc648 8563 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
1d737c8a 8564}
5736199a 8565
199b5763
LM		 8566bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
8567{
de63ad4c 8568 return vcpu->arch.preempted_in_kernel;
199b5763
LM		 8569}
8570
b6d33834 8571int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
5736199a 8572{
b6d33834 8573 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
5736199a 8574}
78646121
GN		 8575
8576int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8577{
8578 return kvm_x86_ops->interrupt_allowed(vcpu);
8579}
229456fc 8580
82b32774 8581unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
f92653ee 8582{
82b32774
NA		 8583 if (is_64_bit_mode(vcpu))
8584 return kvm_rip_read(vcpu);
8585 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8586 kvm_rip_read(vcpu));
8587}
8588EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
f92653ee 8589
82b32774
NA		 8590bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8591{
8592 return kvm_get_linear_rip(vcpu) == linear_rip;
f92653ee
JK		 8593}
8594EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8595
94fe45da
JK		 8596unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8597{
8598 unsigned long rflags;
8599
8600 rflags = kvm_x86_ops->get_rflags(vcpu);
8601 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
c310bac5 8602 rflags &= ~X86_EFLAGS_TF;
94fe45da
JK		 8603 return rflags;
8604}
8605EXPORT_SYMBOL_GPL(kvm_get_rflags);
8606
6addfc42 8607static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
94fe45da
JK		 8608{
8609 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
f92653ee 8610 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
c310bac5 8611 rflags |= X86_EFLAGS_TF;
94fe45da 8612 kvm_x86_ops->set_rflags(vcpu, rflags);
6addfc42
PB		 8613}
8614
8615void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8616{
8617 __kvm_set_rflags(vcpu, rflags);
3842d135 8618 kvm_make_request(KVM_REQ_EVENT, vcpu);
94fe45da
JK		 8619}
8620EXPORT_SYMBOL_GPL(kvm_set_rflags);
8621
56028d08
GN		 8622void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8623{
8624 int r;
8625
fb67e14f 8626 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
f2e10669 8627 work->wakeup_all)
56028d08
GN		 8628 return;
8629
8630 r = kvm_mmu_reload(vcpu);
8631 if (unlikely(r))
8632 return;
8633
fb67e14f
XG		 8634 if (!vcpu->arch.mmu.direct_map &&
8635 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8636 return;
8637
56028d08
GN		 8638 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8639}
8640
af585b92
GN		 8641static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8642{
8643 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8644}
8645
8646static inline u32 kvm_async_pf_next_probe(u32 key)
8647{
8648 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8649}
8650
8651static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8652{
8653 u32 key = kvm_async_pf_hash_fn(gfn);
8654
8655 while (vcpu->arch.apf.gfns[key] != ~0)
8656 key = kvm_async_pf_next_probe(key);
8657
8658 vcpu->arch.apf.gfns[key] = gfn;
8659}
8660
8661static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8662{
8663 int i;
8664 u32 key = kvm_async_pf_hash_fn(gfn);
8665
8666 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
c7d28c24
XG		 8667 (vcpu->arch.apf.gfns[key] != gfn &&
8668 vcpu->arch.apf.gfns[key] != ~0); i++)
af585b92
GN		 8669 key = kvm_async_pf_next_probe(key);
8670
8671 return key;
8672}
8673
8674bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8675{
8676 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8677}
8678
8679static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8680{
8681 u32 i, j, k;
8682
8683 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8684 while (true) {
8685 vcpu->arch.apf.gfns[i] = ~0;
8686 do {
8687 j = kvm_async_pf_next_probe(j);
8688 if (vcpu->arch.apf.gfns[j] == ~0)
8689 return;
8690 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8691 /*
8692 * k lies cyclically in ]i,j]
8693 * | i.k.j |
8694 * |....j i.k.| or |.k..j i...|
8695 */
8696 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8697 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8698 i = j;
8699 }
8700}
8701
7c90705b
GN		 8702static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8703{
4e335d9e
PB		 8704
8705 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8706 sizeof(val));
7c90705b
GN		 8707}
8708
9a6e7c39
WL		 8709static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
8710{
8711
8712 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
8713 sizeof(u32));
8714}
8715
af585b92
GN		 8716void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8717 struct kvm_async_pf *work)
8718{
6389ee94
AK		 8719 struct x86_exception fault;
8720
7c90705b 8721 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
af585b92 8722 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7c90705b
GN		 8723
8724 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
fc5f06fa
GN		 8725 (vcpu->arch.apf.send_user_only &&
8726 kvm_x86_ops->get_cpl(vcpu) == 0))
7c90705b
GN		 8727 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8728 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6389ee94
AK		 8729 fault.vector = PF_VECTOR;
8730 fault.error_code_valid = true;
8731 fault.error_code = 0;
8732 fault.nested_page_fault = false;
8733 fault.address = work->arch.token;
adfe20fb 8734 fault.async_page_fault = true;
6389ee94 8735 kvm_inject_page_fault(vcpu, &fault);
7c90705b 8736 }
af585b92
GN		 8737}
8738
8739void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8740 struct kvm_async_pf *work)
8741{
6389ee94 8742 struct x86_exception fault;
9a6e7c39 8743 u32 val;
6389ee94 8744
f2e10669 8745 if (work->wakeup_all)
7c90705b
GN		 8746 work->arch.token = ~0; /* broadcast wakeup */
8747 else
8748 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
24dccf83 8749 trace_kvm_async_pf_ready(work->arch.token, work->gva);
7c90705b 8750
9a6e7c39
WL		 8751 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
8752 !apf_get_user(vcpu, &val)) {
8753 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
8754 vcpu->arch.exception.pending &&
8755 vcpu->arch.exception.nr == PF_VECTOR &&
8756 !apf_put_user(vcpu, 0)) {
8757 vcpu->arch.exception.injected = false;
8758 vcpu->arch.exception.pending = false;
8759 vcpu->arch.exception.nr = 0;
8760 vcpu->arch.exception.has_error_code = false;
8761 vcpu->arch.exception.error_code = 0;
8762 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8763 fault.vector = PF_VECTOR;
8764 fault.error_code_valid = true;
8765 fault.error_code = 0;
8766 fault.nested_page_fault = false;
8767 fault.address = work->arch.token;
8768 fault.async_page_fault = true;
8769 kvm_inject_page_fault(vcpu, &fault);
8770 }
7c90705b 8771 }
e6d53e3b 8772 vcpu->arch.apf.halted = false;
a4fa1635 8773 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7c90705b
GN		 8774}
8775
8776bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8777{
8778 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8779 return true;
8780 else
9bc1f09f 8781 return kvm_can_do_async_pf(vcpu);
af585b92
GN		 8782}
8783
5544eb9b
PB		 8784void kvm_arch_start_assignment(struct kvm *kvm)
8785{
8786 atomic_inc(&kvm->arch.assigned_device_count);
8787}
8788EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8789
8790void kvm_arch_end_assignment(struct kvm *kvm)
8791{
8792 atomic_dec(&kvm->arch.assigned_device_count);
8793}
8794EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8795
8796bool kvm_arch_has_assigned_device(struct kvm *kvm)
8797{
8798 return atomic_read(&kvm->arch.assigned_device_count);
8799}
8800EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8801
e0f0bbc5
AW		 8802void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8803{
8804 atomic_inc(&kvm->arch.noncoherent_dma_count);
8805}
8806EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8807
8808void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8809{
8810 atomic_dec(&kvm->arch.noncoherent_dma_count);
8811}
8812EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8813
8814bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8815{
8816 return atomic_read(&kvm->arch.noncoherent_dma_count);
8817}
8818EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8819
14717e20
AW		 8820bool kvm_arch_has_irq_bypass(void)
8821{
8822 return kvm_x86_ops->update_pi_irte != NULL;
8823}
8824
87276880
FW		 8825int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8826 struct irq_bypass_producer *prod)
8827{
8828 struct kvm_kernel_irqfd *irqfd =
8829 container_of(cons, struct kvm_kernel_irqfd, consumer);
8830
14717e20 8831 irqfd->producer = prod;
87276880 8832
14717e20
AW		 8833 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8834 prod->irq, irqfd->gsi, 1);
87276880
FW		 8835}
8836
8837void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8838 struct irq_bypass_producer *prod)
8839{
8840 int ret;
8841 struct kvm_kernel_irqfd *irqfd =
8842 container_of(cons, struct kvm_kernel_irqfd, consumer);
8843
87276880
FW		 8844 WARN_ON(irqfd->producer != prod);
8845 irqfd->producer = NULL;
8846
8847 /*
8848 * When producer of consumer is unregistered, we change back to
8849 * remapped mode, so we can re-use the current implementation
bb3541f1 8850 * when the irq is masked/disabled or the consumer side (KVM
87276880
FW		 8851 * int this case doesn't want to receive the interrupts.
8852 */
8853 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8854 if (ret)
8855 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8856 " fails: %d\n", irqfd->consumer.token, ret);
8857}
8858
8859int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8860 uint32_t guest_irq, bool set)
8861{
8862 if (!kvm_x86_ops->update_pi_irte)
8863 return -EINVAL;
8864
8865 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8866}
8867
52004014
FW		 8868bool kvm_vector_hashing_enabled(void)
8869{
8870 return vector_hashing;
8871}
8872EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
8873
229456fc 8874EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
931c33b1 8875EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
229456fc
MT		 8876EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8877EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8878EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8879EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
0ac406de 8880EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
d8cabddf 8881EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
17897f36 8882EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
236649de 8883EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
ec1ff790 8884EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
532a46b9 8885EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
2e554e8d 8886EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
489223ed 8887EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
7b46268d 8888EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
843e4330 8889EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
efc64404 8890EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
18f40c53
SS		 8891EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
8892EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
Ubuntu Focal Linux kernel mirror