projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
kvm: fix schedule in atomic in kvm_steal_time_set_preempted()
[linux-2.6-block.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"
c9eab58f 30#include "assigned-dev.h"
474a5bb9 31#include "pmu.h"
e83d5887 32#include "hyperv.h"
313a3dc7 33
18068523 34#include <linux/clocksource.h>
4d5c5d0f 35#include <linux/interrupt.h>
313a3dc7
CO		 36#include <linux/kvm.h>
37#include <linux/fs.h>
38#include <linux/vmalloc.h>
1767e931
PG		 39#include <linux/export.h>
40#include <linux/moduleparam.h>
0de10343 41#include <linux/mman.h>
2bacc55c 42#include <linux/highmem.h>
19de40a8 43#include <linux/iommu.h>
62c476c7 44#include <linux/intel-iommu.h>
c8076604 45#include <linux/cpufreq.h>
18863bdd 46#include <linux/user-return-notifier.h>
a983fb23 47#include <linux/srcu.h>
5a0e3ad6 48#include <linux/slab.h>
ff9d07a0 49#include <linux/perf_event.h>
7bee342a 50#include <linux/uaccess.h>
af585b92 51#include <linux/hash.h>
a1b60c1c 52#include <linux/pci.h>
16e8d74d
MT		 53#include <linux/timekeeper_internal.h>
54#include <linux/pvclock_gtod.h>
87276880
FW		 55#include <linux/kvm_irqfd.h>
56#include <linux/irqbypass.h>
aec51dc4 57#include <trace/events/kvm.h>
2ed152af 58
24f1e32c 59#include <asm/debugreg.h>
d825ed0a 60#include <asm/msr.h>
a5f61300 61#include <asm/desc.h>
890ca9ae 62#include <asm/mce.h>
f89e32e0 63#include <linux/kernel_stat.h>
78f7f1e5 64#include <asm/fpu/internal.h> /* Ugh! */
1d5f066e 65#include <asm/pvclock.h>
217fc9cf 66#include <asm/div64.h>
efc64404 67#include <asm/irq_remapping.h>
043405e1 68
d1898b73
DH		 69#define CREATE_TRACE_POINTS
70#include "trace.h"
71
313a3dc7 72#define MAX_IO_MSRS 256
890ca9ae 73#define KVM_MAX_MCE_BANKS 32
c45dcc71
AR		 74u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
75EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
890ca9ae 76
0f65dd70
AK		 77#define emul_to_vcpu(ctxt) \
78 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
79
50a37eb4
JR		 80/* EFER defaults:
81 * - enable syscall per default because its emulated by KVM
82 * - enable LME and LMA per default on 64 bit KVM
83 */
84#ifdef CONFIG_X86_64
1260edbe
LJ		 85static
86u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
50a37eb4 87#else
1260edbe 88static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
50a37eb4 89#endif
313a3dc7 90
ba1389b7
AK		 91#define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
92#define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
417bc304 93
c519265f
RK		 94#define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
95 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
37131313 96
cb142eb7 97static void update_cr8_intercept(struct kvm_vcpu *vcpu);
7460fb4a 98static void process_nmi(struct kvm_vcpu *vcpu);
ee2cd4b7 99static void enter_smm(struct kvm_vcpu *vcpu);
6addfc42 100static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
674eea0f 101
893590c7 102struct kvm_x86_ops *kvm_x86_ops __read_mostly;
5fdbf976 103EXPORT_SYMBOL_GPL(kvm_x86_ops);
97896d04 104
893590c7 105static bool __read_mostly ignore_msrs = 0;
476bc001 106module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
ed85c068 107
9ed96e87
MT		 108unsigned int min_timer_period_us = 500;
109module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
110
630994b3
MT		 111static bool __read_mostly kvmclock_periodic_sync = true;
112module_param(kvmclock_periodic_sync, bool, S_IRUGO);
113
893590c7 114bool __read_mostly kvm_has_tsc_control;
92a1f12d 115EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
893590c7 116u32 __read_mostly kvm_max_guest_tsc_khz;
92a1f12d 117EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
bc9b961b
HZ		 118u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
119EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
120u64 __read_mostly kvm_max_tsc_scaling_ratio;
121EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
64672c95
YJ		 122u64 __read_mostly kvm_default_tsc_scaling_ratio;
123EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
92a1f12d 124
cc578287 125/* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
893590c7 126static u32 __read_mostly tsc_tolerance_ppm = 250;
cc578287
ZA		 127module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
128
d0659d94 129/* lapic timer advance (tscdeadline mode only) in nanoseconds */
893590c7 130unsigned int __read_mostly lapic_timer_advance_ns = 0;
d0659d94
MT		 131module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
132
52004014
FW		 133static bool __read_mostly vector_hashing = true;
134module_param(vector_hashing, bool, S_IRUGO);
135
893590c7 136static bool __read_mostly backwards_tsc_observed = false;
16a96021 137
18863bdd
AK		 138#define KVM_NR_SHARED_MSRS 16
139
140struct kvm_shared_msrs_global {
141 int nr;
2bf78fa7 142 u32 msrs[KVM_NR_SHARED_MSRS];
18863bdd
AK		 143};
144
145struct kvm_shared_msrs {
146 struct user_return_notifier urn;
147 bool registered;
2bf78fa7
SY		 148 struct kvm_shared_msr_values {
149 u64 host;
150 u64 curr;
151 } values[KVM_NR_SHARED_MSRS];
18863bdd
AK		 152};
153
154static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
013f6a5d 155static struct kvm_shared_msrs __percpu *shared_msrs;
18863bdd 156
417bc304 157struct kvm_stats_debugfs_item debugfs_entries[] = {
ba1389b7
AK		 158 { "pf_fixed", VCPU_STAT(pf_fixed) },
159 { "pf_guest", VCPU_STAT(pf_guest) },
160 { "tlb_flush", VCPU_STAT(tlb_flush) },
161 { "invlpg", VCPU_STAT(invlpg) },
162 { "exits", VCPU_STAT(exits) },
163 { "io_exits", VCPU_STAT(io_exits) },
164 { "mmio_exits", VCPU_STAT(mmio_exits) },
165 { "signal_exits", VCPU_STAT(signal_exits) },
166 { "irq_window", VCPU_STAT(irq_window_exits) },
f08864b4 167 { "nmi_window", VCPU_STAT(nmi_window_exits) },
ba1389b7 168 { "halt_exits", VCPU_STAT(halt_exits) },
f7819512 169 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
62bea5bf 170 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
3491caf2 171 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
ba1389b7 172 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
f11c3a8d 173 { "hypercalls", VCPU_STAT(hypercalls) },
ba1389b7
AK		 174 { "request_irq", VCPU_STAT(request_irq_exits) },
175 { "irq_exits", VCPU_STAT(irq_exits) },
176 { "host_state_reload", VCPU_STAT(host_state_reload) },
177 { "efer_reload", VCPU_STAT(efer_reload) },
178 { "fpu_reload", VCPU_STAT(fpu_reload) },
179 { "insn_emulation", VCPU_STAT(insn_emulation) },
180 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
fa89a817 181 { "irq_injections", VCPU_STAT(irq_injections) },
c4abb7c9 182 { "nmi_injections", VCPU_STAT(nmi_injections) },
4cee5764
AK		 183 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
184 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
185 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
186 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
187 { "mmu_flooded", VM_STAT(mmu_flooded) },
188 { "mmu_recycled", VM_STAT(mmu_recycled) },
dfc5aa00 189 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
4731d4c7 190 { "mmu_unsync", VM_STAT(mmu_unsync) },
0f74a24c 191 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
05da4558 192 { "largepages", VM_STAT(lpages) },
417bc304
HB		 193 { NULL }
194};
195
2acf923e
DC		 196u64 __read_mostly host_xcr0;
197
b6785def 198static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
d6aa1000 199
af585b92
GN		 200static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
201{
202 int i;
203 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
204 vcpu->arch.apf.gfns[i] = ~0;
205}
206
18863bdd
AK		 207static void kvm_on_user_return(struct user_return_notifier *urn)
208{
209 unsigned slot;
18863bdd
AK		 210 struct kvm_shared_msrs *locals
211 = container_of(urn, struct kvm_shared_msrs, urn);
2bf78fa7 212 struct kvm_shared_msr_values *values;
1650b4eb
IA		 213 unsigned long flags;
214
215 /*
216 * Disabling irqs at this point since the following code could be
217 * interrupted and executed through kvm_arch_hardware_disable()
218 */
219 local_irq_save(flags);
220 if (locals->registered) {
221 locals->registered = false;
222 user_return_notifier_unregister(urn);
223 }
224 local_irq_restore(flags);
18863bdd 225 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
2bf78fa7
SY		 226 values = &locals->values[slot];
227 if (values->host != values->curr) {
228 wrmsrl(shared_msrs_global.msrs[slot], values->host);
229 values->curr = values->host;
18863bdd
AK		 230 }
231 }
18863bdd
AK		 232}
233
2bf78fa7 234static void shared_msr_update(unsigned slot, u32 msr)
18863bdd 235{
18863bdd 236 u64 value;
013f6a5d
MT		 237 unsigned int cpu = smp_processor_id();
238 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
18863bdd 239
2bf78fa7
SY		 240 /* only read, and nobody should modify it at this time,
241 * so don't need lock */
242 if (slot >= shared_msrs_global.nr) {
243 printk(KERN_ERR "kvm: invalid MSR slot!");
244 return;
245 }
246 rdmsrl_safe(msr, &value);
247 smsr->values[slot].host = value;
248 smsr->values[slot].curr = value;
249}
250
251void kvm_define_shared_msr(unsigned slot, u32 msr)
252{
0123be42 253 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
c847fe88 254 shared_msrs_global.msrs[slot] = msr;
18863bdd
AK		 255 if (slot >= shared_msrs_global.nr)
256 shared_msrs_global.nr = slot + 1;
18863bdd
AK		 257}
258EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
259
260static void kvm_shared_msr_cpu_online(void)
261{
262 unsigned i;
18863bdd
AK		 263
264 for (i = 0; i < shared_msrs_global.nr; ++i)
2bf78fa7 265 shared_msr_update(i, shared_msrs_global.msrs[i]);
18863bdd
AK		 266}
267
8b3c3104 268int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
18863bdd 269{
013f6a5d
MT		 270 unsigned int cpu = smp_processor_id();
271 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
8b3c3104 272 int err;
18863bdd 273
2bf78fa7 274 if (((value ^ smsr->values[slot].curr) & mask) == 0)
8b3c3104 275 return 0;
2bf78fa7 276 smsr->values[slot].curr = value;
8b3c3104
AH		 277 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
278 if (err)
279 return 1;
280
18863bdd
AK		 281 if (!smsr->registered) {
282 smsr->urn.on_user_return = kvm_on_user_return;
283 user_return_notifier_register(&smsr->urn);
284 smsr->registered = true;
285 }
8b3c3104 286 return 0;
18863bdd
AK		 287}
288EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
289
13a34e06 290static void drop_user_return_notifiers(void)
3548bab5 291{
013f6a5d
MT		 292 unsigned int cpu = smp_processor_id();
293 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
3548bab5
AK		 294
295 if (smsr->registered)
296 kvm_on_user_return(&smsr->urn);
297}
298
6866b83e
CO		 299u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
300{
8a5a87d9 301 return vcpu->arch.apic_base;
6866b83e
CO		 302}
303EXPORT_SYMBOL_GPL(kvm_get_apic_base);
304
58cb628d
JK		 305int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
306{
307 u64 old_state = vcpu->arch.apic_base &
308 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
309 u64 new_state = msr_info->data &
310 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
311 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
312 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
313
314 if (!msr_info->host_initiated &&
315 ((msr_info->data & reserved_bits) != 0 ||
316 new_state == X2APIC_ENABLE ||
317 (new_state == MSR_IA32_APICBASE_ENABLE &&
318 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
319 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
320 old_state == 0)))
321 return 1;
322
323 kvm_lapic_set_base(vcpu, msr_info->data);
324 return 0;
6866b83e
CO		 325}
326EXPORT_SYMBOL_GPL(kvm_set_apic_base);
327
2605fc21 328asmlinkage __visible void kvm_spurious_fault(void)
e3ba45b8
GL		 329{
330 /* Fault while not rebooting. We want the trace. */
331 BUG();
332}
333EXPORT_SYMBOL_GPL(kvm_spurious_fault);
334
3fd28fce
ED		 335#define EXCPT_BENIGN 0
336#define EXCPT_CONTRIBUTORY 1
337#define EXCPT_PF 2
338
339static int exception_class(int vector)
340{
341 switch (vector) {
342 case PF_VECTOR:
343 return EXCPT_PF;
344 case DE_VECTOR:
345 case TS_VECTOR:
346 case NP_VECTOR:
347 case SS_VECTOR:
348 case GP_VECTOR:
349 return EXCPT_CONTRIBUTORY;
350 default:
351 break;
352 }
353 return EXCPT_BENIGN;
354}
355
d6e8c854
NA		 356#define EXCPT_FAULT 0
357#define EXCPT_TRAP 1
358#define EXCPT_ABORT 2
359#define EXCPT_INTERRUPT 3
360
361static int exception_type(int vector)
362{
363 unsigned int mask;
364
365 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
366 return EXCPT_INTERRUPT;
367
368 mask = 1 << vector;
369
370 /* #DB is trap, as instruction watchpoints are handled elsewhere */
371 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
372 return EXCPT_TRAP;
373
374 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
375 return EXCPT_ABORT;
376
377 /* Reserved exceptions will result in fault */
378 return EXCPT_FAULT;
379}
380
3fd28fce 381static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
ce7ddec4
JR		 382 unsigned nr, bool has_error, u32 error_code,
383 bool reinject)
3fd28fce
ED		 384{
385 u32 prev_nr;
386 int class1, class2;
387
3842d135
AK		 388 kvm_make_request(KVM_REQ_EVENT, vcpu);
389
3fd28fce
ED		 390 if (!vcpu->arch.exception.pending) {
391 queue:
3ffb2468
NA		 392 if (has_error && !is_protmode(vcpu))
393 has_error = false;
3fd28fce
ED		 394 vcpu->arch.exception.pending = true;
395 vcpu->arch.exception.has_error_code = has_error;
396 vcpu->arch.exception.nr = nr;
397 vcpu->arch.exception.error_code = error_code;
3f0fd292 398 vcpu->arch.exception.reinject = reinject;
3fd28fce
ED		 399 return;
400 }
401
402 /* to check exception */
403 prev_nr = vcpu->arch.exception.nr;
404 if (prev_nr == DF_VECTOR) {
405 /* triple fault -> shutdown */
a8eeb04a 406 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3fd28fce
ED		 407 return;
408 }
409 class1 = exception_class(prev_nr);
410 class2 = exception_class(nr);
411 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
412 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
413 /* generate double fault per SDM Table 5-5 */
414 vcpu->arch.exception.pending = true;
415 vcpu->arch.exception.has_error_code = true;
416 vcpu->arch.exception.nr = DF_VECTOR;
417 vcpu->arch.exception.error_code = 0;
418 } else
419 /* replace previous exception with a new one in a hope
420 that instruction re-execution will regenerate lost
421 exception */
422 goto queue;
423}
424
298101da
AK		 425void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
426{
ce7ddec4 427 kvm_multiple_exception(vcpu, nr, false, 0, false);
298101da
AK		 428}
429EXPORT_SYMBOL_GPL(kvm_queue_exception);
430
ce7ddec4
JR		 431void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
432{
433 kvm_multiple_exception(vcpu, nr, false, 0, true);
434}
435EXPORT_SYMBOL_GPL(kvm_requeue_exception);
436
6affcbed 437int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
c3c91fee 438{
db8fcefa
AP		 439 if (err)
440 kvm_inject_gp(vcpu, 0);
441 else
6affcbed
KH		 442 return kvm_skip_emulated_instruction(vcpu);
443
444 return 1;
db8fcefa
AP		 445}
446EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
8df25a32 447
6389ee94 448void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
c3c91fee
AK		 449{
450 ++vcpu->stat.pf_guest;
6389ee94
AK		 451 vcpu->arch.cr2 = fault->address;
452 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
c3c91fee 453}
27d6c865 454EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
c3c91fee 455
ef54bcfe 456static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
d4f8cf66 457{
6389ee94
AK		 458 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
459 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
d4f8cf66 460 else
6389ee94 461 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
ef54bcfe
PB		 462
463 return fault->nested_page_fault;
d4f8cf66
JR		 464}
465
3419ffc8
SY		 466void kvm_inject_nmi(struct kvm_vcpu *vcpu)
467{
7460fb4a
AK		 468 atomic_inc(&vcpu->arch.nmi_queued);
469 kvm_make_request(KVM_REQ_NMI, vcpu);
3419ffc8
SY		 470}
471EXPORT_SYMBOL_GPL(kvm_inject_nmi);
472
298101da
AK		 473void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
474{
ce7ddec4 475 kvm_multiple_exception(vcpu, nr, true, error_code, false);
298101da
AK		 476}
477EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
478
ce7ddec4
JR		 479void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
480{
481 kvm_multiple_exception(vcpu, nr, true, error_code, true);
482}
483EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
484
0a79b009
AK		 485/*
486 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
487 * a #GP and return false.
488 */
489bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
298101da 490{
0a79b009
AK		 491 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
492 return true;
493 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
494 return false;
298101da 495}
0a79b009 496EXPORT_SYMBOL_GPL(kvm_require_cpl);
298101da 497
16f8a6f9
NA		 498bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
499{
500 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
501 return true;
502
503 kvm_queue_exception(vcpu, UD_VECTOR);
504 return false;
505}
506EXPORT_SYMBOL_GPL(kvm_require_dr);
507
ec92fe44
JR		 508/*
509 * This function will be used to read from the physical memory of the currently
54bf36aa 510 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
ec92fe44
JR		 511 * can read from guest physical or from the guest's guest physical memory.
512 */
513int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
514 gfn_t ngfn, void *data, int offset, int len,
515 u32 access)
516{
54987b7a 517 struct x86_exception exception;
ec92fe44
JR		 518 gfn_t real_gfn;
519 gpa_t ngpa;
520
521 ngpa = gfn_to_gpa(ngfn);
54987b7a 522 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
ec92fe44
JR		 523 if (real_gfn == UNMAPPED_GVA)
524 return -EFAULT;
525
526 real_gfn = gpa_to_gfn(real_gfn);
527
54bf36aa 528 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
ec92fe44
JR		 529}
530EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
531
69b0049a 532static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
3d06b8bf
JR		 533 void *data, int offset, int len, u32 access)
534{
535 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
536 data, offset, len, access);
537}
538
a03490ed
CO		 539/*
540 * Load the pae pdptrs. Return true is they are all valid.
541 */
ff03a073 542int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
a03490ed
CO		 543{
544 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
545 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
546 int i;
547 int ret;
ff03a073 548 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
a03490ed 549
ff03a073
JR		 550 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
551 offset * sizeof(u64), sizeof(pdpte),
552 PFERR_USER_MASK|PFERR_WRITE_MASK);
a03490ed
CO		 553 if (ret < 0) {
554 ret = 0;
555 goto out;
556 }
557 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
812f30b2 558 if ((pdpte[i] & PT_PRESENT_MASK) &&
a0a64f50
XG		 559 (pdpte[i] &
560 vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
a03490ed
CO		 561 ret = 0;
562 goto out;
563 }
564 }
565 ret = 1;
566
ff03a073 567 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
6de4f3ad
AK		 568 __set_bit(VCPU_EXREG_PDPTR,
569 (unsigned long *)&vcpu->arch.regs_avail);
570 __set_bit(VCPU_EXREG_PDPTR,
571 (unsigned long *)&vcpu->arch.regs_dirty);
a03490ed 572out:
a03490ed
CO		 573
574 return ret;
575}
cc4b6871 576EXPORT_SYMBOL_GPL(load_pdptrs);
a03490ed 577
9ed38ffa 578bool pdptrs_changed(struct kvm_vcpu *vcpu)
d835dfec 579{
ff03a073 580 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
d835dfec 581 bool changed = true;
3d06b8bf
JR		 582 int offset;
583 gfn_t gfn;
d835dfec
AK		 584 int r;
585
586 if (is_long_mode(vcpu) || !is_pae(vcpu))
587 return false;
588
6de4f3ad
AK		 589 if (!test_bit(VCPU_EXREG_PDPTR,
590 (unsigned long *)&vcpu->arch.regs_avail))
591 return true;
592
9f8fe504
AK		 593 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
594 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
3d06b8bf
JR		 595 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
596 PFERR_USER_MASK | PFERR_WRITE_MASK);
d835dfec
AK		 597 if (r < 0)
598 goto out;
ff03a073 599 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
d835dfec 600out:
d835dfec
AK		 601
602 return changed;
603}
9ed38ffa 604EXPORT_SYMBOL_GPL(pdptrs_changed);
d835dfec 605
49a9b07e 606int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
a03490ed 607{
aad82703 608 unsigned long old_cr0 = kvm_read_cr0(vcpu);
d81135a5 609 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
aad82703 610
f9a48e6a
AK		 611 cr0 |= X86_CR0_ET;
612
ab344828 613#ifdef CONFIG_X86_64
0f12244f
GN		 614 if (cr0 & 0xffffffff00000000UL)
615 return 1;
ab344828
GN		 616#endif
617
618 cr0 &= ~CR0_RESERVED_BITS;
a03490ed 619
0f12244f
GN		 620 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
621 return 1;
a03490ed 622
0f12244f
GN		 623 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
624 return 1;
a03490ed
CO		 625
626 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
627#ifdef CONFIG_X86_64
f6801dff 628 if ((vcpu->arch.efer & EFER_LME)) {
a03490ed
CO		 629 int cs_db, cs_l;
630
0f12244f
GN		 631 if (!is_pae(vcpu))
632 return 1;
a03490ed 633 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
0f12244f
GN		 634 if (cs_l)
635 return 1;
a03490ed
CO		 636 } else
637#endif
ff03a073 638 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
9f8fe504 639 kvm_read_cr3(vcpu)))
0f12244f 640 return 1;
a03490ed
CO		 641 }
642
ad756a16
MJ		 643 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
644 return 1;
645
a03490ed 646 kvm_x86_ops->set_cr0(vcpu, cr0);
a03490ed 647
d170c419 648 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
e5f3f027 649 kvm_clear_async_pf_completion_queue(vcpu);
d170c419
LJ		 650 kvm_async_pf_hash_reset(vcpu);
651 }
e5f3f027 652
aad82703
SY		 653 if ((cr0 ^ old_cr0) & update_bits)
654 kvm_mmu_reset_context(vcpu);
b18d5431 655
879ae188
LE		 656 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
657 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
658 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
b18d5431
XG		 659 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
660
0f12244f
GN		 661 return 0;
662}
2d3ad1f4 663EXPORT_SYMBOL_GPL(kvm_set_cr0);
a03490ed 664
2d3ad1f4 665void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
a03490ed 666{
49a9b07e 667 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
a03490ed 668}
2d3ad1f4 669EXPORT_SYMBOL_GPL(kvm_lmsw);
a03490ed 670
42bdf991
MT		 671static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
672{
673 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
674 !vcpu->guest_xcr0_loaded) {
675 /* kvm_set_xcr() also depends on this */
676 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
677 vcpu->guest_xcr0_loaded = 1;
678 }
679}
680
681static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
682{
683 if (vcpu->guest_xcr0_loaded) {
684 if (vcpu->arch.xcr0 != host_xcr0)
685 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
686 vcpu->guest_xcr0_loaded = 0;
687 }
688}
689
69b0049a 690static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
2acf923e 691{
56c103ec
LJ		 692 u64 xcr0 = xcr;
693 u64 old_xcr0 = vcpu->arch.xcr0;
46c34cb0 694 u64 valid_bits;
2acf923e
DC		 695
696 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
697 if (index != XCR_XFEATURE_ENABLED_MASK)
698 return 1;
d91cab78 699 if (!(xcr0 & XFEATURE_MASK_FP))
2acf923e 700 return 1;
d91cab78 701 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
2acf923e 702 return 1;
46c34cb0
PB		 703
704 /*
705 * Do not allow the guest to set bits that we do not support
706 * saving. However, xcr0 bit 0 is always set, even if the
707 * emulated CPU does not support XSAVE (see fx_init).
708 */
d91cab78 709 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
46c34cb0 710 if (xcr0 & ~valid_bits)
2acf923e 711 return 1;
46c34cb0 712
d91cab78
DH		 713 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
714 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
390bd528
LJ		 715 return 1;
716
d91cab78
DH		 717 if (xcr0 & XFEATURE_MASK_AVX512) {
718 if (!(xcr0 & XFEATURE_MASK_YMM))
612263b3 719 return 1;
d91cab78 720 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
612263b3
CP		 721 return 1;
722 }
2acf923e 723 vcpu->arch.xcr0 = xcr0;
56c103ec 724
d91cab78 725 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
56c103ec 726 kvm_update_cpuid(vcpu);
2acf923e
DC		 727 return 0;
728}
729
730int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
731{
764bcbc5
Z		 732 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
733 __kvm_set_xcr(vcpu, index, xcr)) {
2acf923e
DC		 734 kvm_inject_gp(vcpu, 0);
735 return 1;
736 }
737 return 0;
738}
739EXPORT_SYMBOL_GPL(kvm_set_xcr);
740
a83b29c6 741int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
a03490ed 742{
fc78f519 743 unsigned long old_cr4 = kvm_read_cr4(vcpu);
0be0226f 744 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
b9baba86 745 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
0be0226f 746
0f12244f
GN		 747 if (cr4 & CR4_RESERVED_BITS)
748 return 1;
a03490ed 749
2acf923e
DC		 750 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
751 return 1;
752
c68b734f
YW		 753 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
754 return 1;
755
97ec8c06
FW		 756 if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
757 return 1;
758
afcbf13f 759 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
74dc2b4f
YW		 760 return 1;
761
b9baba86
HH		 762 if (!guest_cpuid_has_pku(vcpu) && (cr4 & X86_CR4_PKE))
763 return 1;
764
a03490ed 765 if (is_long_mode(vcpu)) {
0f12244f
GN		 766 if (!(cr4 & X86_CR4_PAE))
767 return 1;
a2edf57f
AK		 768 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
769 && ((cr4 ^ old_cr4) & pdptr_bits)
9f8fe504
AK		 770 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
771 kvm_read_cr3(vcpu)))
0f12244f
GN		 772 return 1;
773
ad756a16
MJ		 774 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
775 if (!guest_cpuid_has_pcid(vcpu))
776 return 1;
777
778 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
779 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
780 return 1;
781 }
782
5e1746d6 783 if (kvm_x86_ops->set_cr4(vcpu, cr4))
0f12244f 784 return 1;
a03490ed 785
ad756a16
MJ		 786 if (((cr4 ^ old_cr4) & pdptr_bits) ||
787 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
aad82703 788 kvm_mmu_reset_context(vcpu);
0f12244f 789
b9baba86 790 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
00b27a3e 791 kvm_update_cpuid(vcpu);
2acf923e 792
0f12244f
GN		 793 return 0;
794}
2d3ad1f4 795EXPORT_SYMBOL_GPL(kvm_set_cr4);
a03490ed 796
2390218b 797int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
a03490ed 798{
ac146235 799#ifdef CONFIG_X86_64
9d88fca7 800 cr3 &= ~CR3_PCID_INVD;
ac146235 801#endif
9d88fca7 802
9f8fe504 803 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
0ba73cda 804 kvm_mmu_sync_roots(vcpu);
77c3913b 805 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
0f12244f 806 return 0;
d835dfec
AK		 807 }
808
a03490ed 809 if (is_long_mode(vcpu)) {
d9f89b88
JK		 810 if (cr3 & CR3_L_MODE_RESERVED_BITS)
811 return 1;
812 } else if (is_pae(vcpu) && is_paging(vcpu) &&
813 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
346874c9 814 return 1;
a03490ed 815
0f12244f 816 vcpu->arch.cr3 = cr3;
aff48baa 817 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
d8d173da 818 kvm_mmu_new_cr3(vcpu);
0f12244f
GN		 819 return 0;
820}
2d3ad1f4 821EXPORT_SYMBOL_GPL(kvm_set_cr3);
a03490ed 822
eea1cff9 823int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
a03490ed 824{
0f12244f
GN		 825 if (cr8 & CR8_RESERVED_BITS)
826 return 1;
35754c98 827 if (lapic_in_kernel(vcpu))
a03490ed
CO		 828 kvm_lapic_set_tpr(vcpu, cr8);
829 else
ad312c7c 830 vcpu->arch.cr8 = cr8;
0f12244f
GN		 831 return 0;
832}
2d3ad1f4 833EXPORT_SYMBOL_GPL(kvm_set_cr8);
a03490ed 834
2d3ad1f4 835unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
a03490ed 836{
35754c98 837 if (lapic_in_kernel(vcpu))
a03490ed
CO		 838 return kvm_lapic_get_cr8(vcpu);
839 else
ad312c7c 840 return vcpu->arch.cr8;
a03490ed 841}
2d3ad1f4 842EXPORT_SYMBOL_GPL(kvm_get_cr8);
a03490ed 843
ae561ede
NA		 844static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
845{
846 int i;
847
848 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
849 for (i = 0; i < KVM_NR_DB_REGS; i++)
850 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
851 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
852 }
853}
854
73aaf249
JK		 855static void kvm_update_dr6(struct kvm_vcpu *vcpu)
856{
857 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
858 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
859}
860
c8639010
JK		 861static void kvm_update_dr7(struct kvm_vcpu *vcpu)
862{
863 unsigned long dr7;
864
865 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
866 dr7 = vcpu->arch.guest_debug_dr7;
867 else
868 dr7 = vcpu->arch.dr7;
869 kvm_x86_ops->set_dr7(vcpu, dr7);
360b948d
PB		 870 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
871 if (dr7 & DR7_BP_EN_MASK)
872 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
c8639010
JK		 873}
874
6f43ed01
NA		 875static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
876{
877 u64 fixed = DR6_FIXED_1;
878
879 if (!guest_cpuid_has_rtm(vcpu))
880 fixed |= DR6_RTM;
881 return fixed;
882}
883
338dbc97 884static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
020df079
GN		 885{
886 switch (dr) {
887 case 0 ... 3:
888 vcpu->arch.db[dr] = val;
889 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
890 vcpu->arch.eff_db[dr] = val;
891 break;
892 case 4:
020df079
GN		 893 /* fall through */
894 case 6:
338dbc97
GN		 895 if (val & 0xffffffff00000000ULL)
896 return -1; /* #GP */
6f43ed01 897 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
73aaf249 898 kvm_update_dr6(vcpu);
020df079
GN		 899 break;
900 case 5:
020df079
GN		 901 /* fall through */
902 default: /* 7 */
338dbc97
GN		 903 if (val & 0xffffffff00000000ULL)
904 return -1; /* #GP */
020df079 905 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
c8639010 906 kvm_update_dr7(vcpu);
020df079
GN		 907 break;
908 }
909
910 return 0;
911}
338dbc97
GN		 912
913int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
914{
16f8a6f9 915 if (__kvm_set_dr(vcpu, dr, val)) {
338dbc97 916 kvm_inject_gp(vcpu, 0);
16f8a6f9
NA		 917 return 1;
918 }
919 return 0;
338dbc97 920}
020df079
GN		 921EXPORT_SYMBOL_GPL(kvm_set_dr);
922
16f8a6f9 923int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
020df079
GN		 924{
925 switch (dr) {
926 case 0 ... 3:
927 *val = vcpu->arch.db[dr];
928 break;
929 case 4:
020df079
GN		 930 /* fall through */
931 case 6:
73aaf249
JK		 932 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
933 *val = vcpu->arch.dr6;
934 else
935 *val = kvm_x86_ops->get_dr6(vcpu);
020df079
GN		 936 break;
937 case 5:
020df079
GN		 938 /* fall through */
939 default: /* 7 */
940 *val = vcpu->arch.dr7;
941 break;
942 }
338dbc97
GN		 943 return 0;
944}
020df079
GN		 945EXPORT_SYMBOL_GPL(kvm_get_dr);
946
022cd0e8
AK		 947bool kvm_rdpmc(struct kvm_vcpu *vcpu)
948{
949 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
950 u64 data;
951 int err;
952
c6702c9d 953 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
022cd0e8
AK		 954 if (err)
955 return err;
956 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
957 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
958 return err;
959}
960EXPORT_SYMBOL_GPL(kvm_rdpmc);
961
043405e1
CO		 962/*
963 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
964 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
965 *
966 * This list is modified at module load time to reflect the
e3267cbb 967 * capabilities of the host cpu. This capabilities test skips MSRs that are
62ef68bb
PB		 968 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
969 * may depend on host virtualization features rather than host cpu features.
043405e1 970 */
e3267cbb 971
043405e1
CO		 972static u32 msrs_to_save[] = {
973 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
8c06585d 974 MSR_STAR,
043405e1
CO		 975#ifdef CONFIG_X86_64
976 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
977#endif
b3897a49 978 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
9dbe6cf9 979 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
043405e1
CO		 980};
981
982static unsigned num_msrs_to_save;
983
62ef68bb
PB		 984static u32 emulated_msrs[] = {
985 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
986 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
987 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
988 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
e7d9513b
AS		 989 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
990 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
e516cebb 991 HV_X64_MSR_RESET,
11c4b1ca 992 HV_X64_MSR_VP_INDEX,
9eec50b8 993 HV_X64_MSR_VP_RUNTIME,
5c919412 994 HV_X64_MSR_SCONTROL,
1f4b34f8 995 HV_X64_MSR_STIMER0_CONFIG,
62ef68bb
PB		 996 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
997 MSR_KVM_PV_EOI_EN,
998
ba904635 999 MSR_IA32_TSC_ADJUST,
a3e06bbe 1000 MSR_IA32_TSCDEADLINE,
043405e1 1001 MSR_IA32_MISC_ENABLE,
908e75f3
AK		 1002 MSR_IA32_MCG_STATUS,
1003 MSR_IA32_MCG_CTL,
c45dcc71 1004 MSR_IA32_MCG_EXT_CTL,
64d60670 1005 MSR_IA32_SMBASE,
043405e1
CO		 1006};
1007
62ef68bb
PB		 1008static unsigned num_emulated_msrs;
1009
384bb783 1010bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
15c4a640 1011{
b69e8cae 1012 if (efer & efer_reserved_bits)
384bb783 1013 return false;
15c4a640 1014
1b2fd70c
AG		 1015 if (efer & EFER_FFXSR) {
1016 struct kvm_cpuid_entry2 *feat;
1017
1018 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
b69e8cae 1019 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
384bb783 1020 return false;
1b2fd70c
AG		 1021 }
1022
d8017474
AG		 1023 if (efer & EFER_SVME) {
1024 struct kvm_cpuid_entry2 *feat;
1025
1026 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
b69e8cae 1027 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
384bb783 1028 return false;
d8017474
AG		 1029 }
1030
384bb783
JK		 1031 return true;
1032}
1033EXPORT_SYMBOL_GPL(kvm_valid_efer);
1034
1035static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1036{
1037 u64 old_efer = vcpu->arch.efer;
1038
1039 if (!kvm_valid_efer(vcpu, efer))
1040 return 1;
1041
1042 if (is_paging(vcpu)
1043 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1044 return 1;
1045
15c4a640 1046 efer &= ~EFER_LMA;
f6801dff 1047 efer |= vcpu->arch.efer & EFER_LMA;
15c4a640 1048
a3d204e2
SY		 1049 kvm_x86_ops->set_efer(vcpu, efer);
1050
aad82703
SY		 1051 /* Update reserved bits */
1052 if ((efer ^ old_efer) & EFER_NX)
1053 kvm_mmu_reset_context(vcpu);
1054
b69e8cae 1055 return 0;
15c4a640
CO		 1056}
1057
f2b4b7dd
JR		 1058void kvm_enable_efer_bits(u64 mask)
1059{
1060 efer_reserved_bits &= ~mask;
1061}
1062EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1063
15c4a640
CO		 1064/*
1065 * Writes msr value into into the appropriate "register".
1066 * Returns 0 on success, non-0 otherwise.
1067 * Assumes vcpu_load() was already called.
1068 */
8fe8ab46 1069int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
15c4a640 1070{
854e8bb1
NA		 1071 switch (msr->index) {
1072 case MSR_FS_BASE:
1073 case MSR_GS_BASE:
1074 case MSR_KERNEL_GS_BASE:
1075 case MSR_CSTAR:
1076 case MSR_LSTAR:
1077 if (is_noncanonical_address(msr->data))
1078 return 1;
1079 break;
1080 case MSR_IA32_SYSENTER_EIP:
1081 case MSR_IA32_SYSENTER_ESP:
1082 /*
1083 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1084 * non-canonical address is written on Intel but not on
1085 * AMD (which ignores the top 32-bits, because it does
1086 * not implement 64-bit SYSENTER).
1087 *
1088 * 64-bit code should hence be able to write a non-canonical
1089 * value on AMD. Making the address canonical ensures that
1090 * vmentry does not fail on Intel after writing a non-canonical
1091 * value, and that something deterministic happens if the guest
1092 * invokes 64-bit SYSENTER.
1093 */
1094 msr->data = get_canonical(msr->data);
1095 }
8fe8ab46 1096 return kvm_x86_ops->set_msr(vcpu, msr);
15c4a640 1097}
854e8bb1 1098EXPORT_SYMBOL_GPL(kvm_set_msr);
15c4a640 1099
313a3dc7
CO		 1100/*
1101 * Adapt set_msr() to msr_io()'s calling convention
1102 */
609e36d3
PB		 1103static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1104{
1105 struct msr_data msr;
1106 int r;
1107
1108 msr.index = index;
1109 msr.host_initiated = true;
1110 r = kvm_get_msr(vcpu, &msr);
1111 if (r)
1112 return r;
1113
1114 *data = msr.data;
1115 return 0;
1116}
1117
313a3dc7
CO		 1118static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1119{
8fe8ab46
WA		 1120 struct msr_data msr;
1121
1122 msr.data = *data;
1123 msr.index = index;
1124 msr.host_initiated = true;
1125 return kvm_set_msr(vcpu, &msr);
313a3dc7
CO		 1126}
1127
16e8d74d
MT		 1128#ifdef CONFIG_X86_64
1129struct pvclock_gtod_data {
1130 seqcount_t seq;
1131
1132 struct { /* extract of a clocksource struct */
1133 int vclock_mode;
1134 cycle_t cycle_last;
1135 cycle_t mask;
1136 u32 mult;
1137 u32 shift;
1138 } clock;
1139
cbcf2dd3
TG		 1140 u64 boot_ns;
1141 u64 nsec_base;
16e8d74d
MT		 1142};
1143
1144static struct pvclock_gtod_data pvclock_gtod_data;
1145
1146static void update_pvclock_gtod(struct timekeeper *tk)
1147{
1148 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
cbcf2dd3
TG		 1149 u64 boot_ns;
1150
876e7881 1151 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
16e8d74d
MT		 1152
1153 write_seqcount_begin(&vdata->seq);
1154
1155 /* copy pvclock gtod data */
876e7881
PZ		 1156 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1157 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1158 vdata->clock.mask = tk->tkr_mono.mask;
1159 vdata->clock.mult = tk->tkr_mono.mult;
1160 vdata->clock.shift = tk->tkr_mono.shift;
16e8d74d 1161
cbcf2dd3 1162 vdata->boot_ns = boot_ns;
876e7881 1163 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
16e8d74d
MT		 1164
1165 write_seqcount_end(&vdata->seq);
1166}
1167#endif
1168
bab5bb39
NK		 1169void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1170{
1171 /*
1172 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1173 * vcpu_enter_guest. This function is only called from
1174 * the physical CPU that is running vcpu.
1175 */
1176 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1177}
16e8d74d 1178
18068523
GOC		 1179static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1180{
9ed3c444
AK		 1181 int version;
1182 int r;
50d0a0f9 1183 struct pvclock_wall_clock wc;
87aeb54f 1184 struct timespec64 boot;
18068523
GOC		 1185
1186 if (!wall_clock)
1187 return;
1188
9ed3c444
AK		 1189 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1190 if (r)
1191 return;
1192
1193 if (version & 1)
1194 ++version; /* first time write, random junk */
1195
1196 ++version;
18068523 1197
1dab1345
NK		 1198 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1199 return;
18068523 1200
50d0a0f9
GH		 1201 /*
1202 * The guest calculates current wall clock time by adding
34c238a1 1203 * system time (updated by kvm_guest_time_update below) to the
50d0a0f9
GH		 1204 * wall clock specified here. guest system time equals host
1205 * system time for us, thus we must fill in host boot time here.
1206 */
87aeb54f 1207 getboottime64(&boot);
50d0a0f9 1208
4b648665 1209 if (kvm->arch.kvmclock_offset) {
87aeb54f
AB		 1210 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1211 boot = timespec64_sub(boot, ts);
4b648665 1212 }
87aeb54f 1213 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
50d0a0f9
GH		 1214 wc.nsec = boot.tv_nsec;
1215 wc.version = version;
18068523
GOC		 1216
1217 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1218
1219 version++;
1220 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
18068523
GOC		 1221}
1222
50d0a0f9
GH		 1223static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1224{
b51012de
PB		 1225 do_shl32_div32(dividend, divisor);
1226 return dividend;
50d0a0f9
GH		 1227}
1228
3ae13faa 1229static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
5f4e3f88 1230 s8 *pshift, u32 *pmultiplier)
50d0a0f9 1231{
5f4e3f88 1232 uint64_t scaled64;
50d0a0f9
GH		 1233 int32_t shift = 0;
1234 uint64_t tps64;
1235 uint32_t tps32;
1236
3ae13faa
PB		 1237 tps64 = base_hz;
1238 scaled64 = scaled_hz;
50933623 1239 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
50d0a0f9
GH		 1240 tps64 >>= 1;
1241 shift--;
1242 }
1243
1244 tps32 = (uint32_t)tps64;
50933623
JK		 1245 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1246 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
5f4e3f88
ZA		 1247 scaled64 >>= 1;
1248 else
1249 tps32 <<= 1;
50d0a0f9
GH		 1250 shift++;
1251 }
1252
5f4e3f88
ZA		 1253 *pshift = shift;
1254 *pmultiplier = div_frac(scaled64, tps32);
50d0a0f9 1255
3ae13faa
PB		 1256 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1257 __func__, base_hz, scaled_hz, shift, *pmultiplier);
50d0a0f9
GH		 1258}
1259
d828199e 1260#ifdef CONFIG_X86_64
16e8d74d 1261static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
d828199e 1262#endif
16e8d74d 1263
c8076604 1264static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
69b0049a 1265static unsigned long max_tsc_khz;
c8076604 1266
cc578287 1267static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1e993611 1268{
cc578287
ZA		 1269 u64 v = (u64)khz * (1000000 + ppm);
1270 do_div(v, 1000000);
1271 return v;
1e993611
JR		 1272}
1273
381d585c
HZ		 1274static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1275{
1276 u64 ratio;
1277
1278 /* Guest TSC same frequency as host TSC? */
1279 if (!scale) {
1280 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1281 return 0;
1282 }
1283
1284 /* TSC scaling supported? */
1285 if (!kvm_has_tsc_control) {
1286 if (user_tsc_khz > tsc_khz) {
1287 vcpu->arch.tsc_catchup = 1;
1288 vcpu->arch.tsc_always_catchup = 1;
1289 return 0;
1290 } else {
1291 WARN(1, "user requested TSC rate below hardware speed\n");
1292 return -1;
1293 }
1294 }
1295
1296 /* TSC scaling required - calculate ratio */
1297 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1298 user_tsc_khz, tsc_khz);
1299
1300 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1301 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1302 user_tsc_khz);
1303 return -1;
1304 }
1305
1306 vcpu->arch.tsc_scaling_ratio = ratio;
1307 return 0;
1308}
1309
4941b8cb 1310static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
759379dd 1311{
cc578287
ZA		 1312 u32 thresh_lo, thresh_hi;
1313 int use_scaling = 0;
217fc9cf 1314
03ba32ca 1315 /* tsc_khz can be zero if TSC calibration fails */
4941b8cb 1316 if (user_tsc_khz == 0) {
ad721883
HZ		 1317 /* set tsc_scaling_ratio to a safe value */
1318 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
381d585c 1319 return -1;
ad721883 1320 }
03ba32ca 1321
c285545f 1322 /* Compute a scale to convert nanoseconds in TSC cycles */
3ae13faa 1323 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
cc578287
ZA		 1324 &vcpu->arch.virtual_tsc_shift,
1325 &vcpu->arch.virtual_tsc_mult);
4941b8cb 1326 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
cc578287
ZA		 1327
1328 /*
1329 * Compute the variation in TSC rate which is acceptable
1330 * within the range of tolerance and decide if the
1331 * rate being applied is within that bounds of the hardware
1332 * rate. If so, no scaling or compensation need be done.
1333 */
1334 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1335 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
4941b8cb
PB		 1336 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1337 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
cc578287
ZA		 1338 use_scaling = 1;
1339 }
4941b8cb 1340 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
c285545f
ZA		 1341}
1342
1343static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1344{
e26101b1 1345 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
cc578287
ZA		 1346 vcpu->arch.virtual_tsc_mult,
1347 vcpu->arch.virtual_tsc_shift);
e26101b1 1348 tsc += vcpu->arch.this_tsc_write;
c285545f
ZA		 1349 return tsc;
1350}
1351
69b0049a 1352static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
b48aa97e
MT		 1353{
1354#ifdef CONFIG_X86_64
1355 bool vcpus_matched;
b48aa97e
MT		 1356 struct kvm_arch *ka = &vcpu->kvm->arch;
1357 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1358
1359 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1360 atomic_read(&vcpu->kvm->online_vcpus));
1361
7f187922
MT		 1362 /*
1363 * Once the masterclock is enabled, always perform request in
1364 * order to update it.
1365 *
1366 * In order to enable masterclock, the host clocksource must be TSC
1367 * and the vcpus need to have matched TSCs. When that happens,
1368 * perform request to enable masterclock.
1369 */
1370 if (ka->use_master_clock ||
1371 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
b48aa97e
MT		 1372 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1373
1374 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1375 atomic_read(&vcpu->kvm->online_vcpus),
1376 ka->use_master_clock, gtod->clock.vclock_mode);
1377#endif
1378}
1379
ba904635
WA		 1380static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1381{
3e3f5026 1382 u64 curr_offset = vcpu->arch.tsc_offset;
ba904635
WA		 1383 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1384}
1385
35181e86
HZ		 1386/*
1387 * Multiply tsc by a fixed point number represented by ratio.
1388 *
1389 * The most significant 64-N bits (mult) of ratio represent the
1390 * integral part of the fixed point number; the remaining N bits
1391 * (frac) represent the fractional part, ie. ratio represents a fixed
1392 * point number (mult + frac * 2^(-N)).
1393 *
1394 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1395 */
1396static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1397{
1398 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1399}
1400
1401u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1402{
1403 u64 _tsc = tsc;
1404 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1405
1406 if (ratio != kvm_default_tsc_scaling_ratio)
1407 _tsc = __scale_tsc(ratio, tsc);
1408
1409 return _tsc;
1410}
1411EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1412
07c1419a
HZ		 1413static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1414{
1415 u64 tsc;
1416
1417 tsc = kvm_scale_tsc(vcpu, rdtsc());
1418
1419 return target_tsc - tsc;
1420}
1421
4ba76538
HZ		 1422u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1423{
ea26e4ec 1424 return vcpu->arch.tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
4ba76538
HZ		 1425}
1426EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1427
a545ab6a
LC		 1428static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1429{
1430 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1431 vcpu->arch.tsc_offset = offset;
1432}
1433
8fe8ab46 1434void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
99e3e30a
ZA		 1435{
1436 struct kvm *kvm = vcpu->kvm;
f38e098f 1437 u64 offset, ns, elapsed;
99e3e30a 1438 unsigned long flags;
02626b6a 1439 s64 usdiff;
b48aa97e 1440 bool matched;
0d3da0d2 1441 bool already_matched;
8fe8ab46 1442 u64 data = msr->data;
99e3e30a 1443
038f8c11 1444 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
07c1419a 1445 offset = kvm_compute_tsc_offset(vcpu, data);
108b249c 1446 ns = ktime_get_boot_ns();
f38e098f 1447 elapsed = ns - kvm->arch.last_tsc_nsec;
5d3cb0f6 1448
03ba32ca 1449 if (vcpu->arch.virtual_tsc_khz) {
8915aa27
MT		 1450 int faulted = 0;
1451
03ba32ca
MT		 1452 /* n.b - signed multiplication and division required */
1453 usdiff = data - kvm->arch.last_tsc_write;
5d3cb0f6 1454#ifdef CONFIG_X86_64
03ba32ca 1455 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
5d3cb0f6 1456#else
03ba32ca 1457 /* do_div() only does unsigned */
8915aa27
MT		 1458 asm("1: idivl %[divisor]\n"
1459 "2: xor %%edx, %%edx\n"
1460 " movl $0, %[faulted]\n"
1461 "3:\n"
1462 ".section .fixup,\"ax\"\n"
1463 "4: movl $1, %[faulted]\n"
1464 " jmp 3b\n"
1465 ".previous\n"
1466
1467 _ASM_EXTABLE(1b, 4b)
1468
1469 : "=A"(usdiff), [faulted] "=r" (faulted)
1470 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1471
5d3cb0f6 1472#endif
03ba32ca
MT		 1473 do_div(elapsed, 1000);
1474 usdiff -= elapsed;
1475 if (usdiff < 0)
1476 usdiff = -usdiff;
8915aa27
MT		 1477
1478 /* idivl overflow => difference is larger than USEC_PER_SEC */
1479 if (faulted)
1480 usdiff = USEC_PER_SEC;
03ba32ca
MT		 1481 } else
1482 usdiff = USEC_PER_SEC; /* disable TSC match window below */
f38e098f
ZA		 1483
1484 /*
5d3cb0f6
ZA		 1485 * Special case: TSC write with a small delta (1 second) of virtual
1486 * cycle time against real time is interpreted as an attempt to
1487 * synchronize the CPU.
1488 *
1489 * For a reliable TSC, we can match TSC offsets, and for an unstable
1490 * TSC, we add elapsed time in this computation. We could let the
1491 * compensation code attempt to catch up if we fall behind, but
1492 * it's better to try to match offsets from the beginning.
1493 */
02626b6a 1494 if (usdiff < USEC_PER_SEC &&
5d3cb0f6 1495 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
f38e098f 1496 if (!check_tsc_unstable()) {
e26101b1 1497 offset = kvm->arch.cur_tsc_offset;
f38e098f
ZA		 1498 pr_debug("kvm: matched tsc offset for %llu\n", data);
1499 } else {
857e4099 1500 u64 delta = nsec_to_cycles(vcpu, elapsed);
5d3cb0f6 1501 data += delta;
07c1419a 1502 offset = kvm_compute_tsc_offset(vcpu, data);
759379dd 1503 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
f38e098f 1504 }
b48aa97e 1505 matched = true;
0d3da0d2 1506 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
e26101b1
ZA		 1507 } else {
1508 /*
1509 * We split periods of matched TSC writes into generations.
1510 * For each generation, we track the original measured
1511 * nanosecond time, offset, and write, so if TSCs are in
1512 * sync, we can match exact offset, and if not, we can match
4a969980 1513 * exact software computation in compute_guest_tsc()
e26101b1
ZA		 1514 *
1515 * These values are tracked in kvm->arch.cur_xxx variables.
1516 */
1517 kvm->arch.cur_tsc_generation++;
1518 kvm->arch.cur_tsc_nsec = ns;
1519 kvm->arch.cur_tsc_write = data;
1520 kvm->arch.cur_tsc_offset = offset;
b48aa97e 1521 matched = false;
0d3da0d2 1522 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
e26101b1 1523 kvm->arch.cur_tsc_generation, data);
f38e098f 1524 }
e26101b1
ZA		 1525
1526 /*
1527 * We also track th most recent recorded KHZ, write and time to
1528 * allow the matching interval to be extended at each write.
1529 */
f38e098f
ZA		 1530 kvm->arch.last_tsc_nsec = ns;
1531 kvm->arch.last_tsc_write = data;
5d3cb0f6 1532 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
99e3e30a 1533
b183aa58 1534 vcpu->arch.last_guest_tsc = data;
e26101b1
ZA		 1535
1536 /* Keep track of which generation this VCPU has synchronized to */
1537 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1538 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1539 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1540
ba904635
WA		 1541 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1542 update_ia32_tsc_adjust_msr(vcpu, offset);
a545ab6a 1543 kvm_vcpu_write_tsc_offset(vcpu, offset);
e26101b1 1544 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
b48aa97e
MT		 1545
1546 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
0d3da0d2 1547 if (!matched) {
b48aa97e 1548 kvm->arch.nr_vcpus_matched_tsc = 0;
0d3da0d2
TG		 1549 } else if (!already_matched) {
1550 kvm->arch.nr_vcpus_matched_tsc++;
1551 }
b48aa97e
MT		 1552
1553 kvm_track_tsc_matching(vcpu);
1554 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
99e3e30a 1555}
e26101b1 1556
99e3e30a
ZA		 1557EXPORT_SYMBOL_GPL(kvm_write_tsc);
1558
58ea6767
HZ		 1559static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1560 s64 adjustment)
1561{
ea26e4ec 1562 kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.tsc_offset + adjustment);
58ea6767
HZ		 1563}
1564
1565static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1566{
1567 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1568 WARN_ON(adjustment < 0);
1569 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
ea26e4ec 1570 adjust_tsc_offset_guest(vcpu, adjustment);
58ea6767
HZ		 1571}
1572
d828199e
MT		 1573#ifdef CONFIG_X86_64
1574
1575static cycle_t read_tsc(void)
1576{
03b9730b
AL		 1577 cycle_t ret = (cycle_t)rdtsc_ordered();
1578 u64 last = pvclock_gtod_data.clock.cycle_last;
d828199e
MT		 1579
1580 if (likely(ret >= last))
1581 return ret;
1582
1583 /*
1584 * GCC likes to generate cmov here, but this branch is extremely
6a6256f9 1585 * predictable (it's just a function of time and the likely is
d828199e
MT		 1586 * very likely) and there's a data dependence, so force GCC
1587 * to generate a branch instead. I don't barrier() because
1588 * we don't actually need a barrier, and if this function
1589 * ever gets inlined it will generate worse code.
1590 */
1591 asm volatile ("");
1592 return last;
1593}
1594
1595static inline u64 vgettsc(cycle_t *cycle_now)
1596{
1597 long v;
1598 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1599
1600 *cycle_now = read_tsc();
1601
1602 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1603 return v * gtod->clock.mult;
1604}
1605
cbcf2dd3 1606static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
d828199e 1607{
cbcf2dd3 1608 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
d828199e 1609 unsigned long seq;
d828199e 1610 int mode;
cbcf2dd3 1611 u64 ns;
d828199e 1612
d828199e
MT		 1613 do {
1614 seq = read_seqcount_begin(&gtod->seq);
1615 mode = gtod->clock.vclock_mode;
cbcf2dd3 1616 ns = gtod->nsec_base;
d828199e
MT		 1617 ns += vgettsc(cycle_now);
1618 ns >>= gtod->clock.shift;
cbcf2dd3 1619 ns += gtod->boot_ns;
d828199e 1620 } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
cbcf2dd3 1621 *t = ns;
d828199e
MT		 1622
1623 return mode;
1624}
1625
1626/* returns true if host is using tsc clocksource */
1627static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1628{
d828199e
MT		 1629 /* checked again under seqlock below */
1630 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1631 return false;
1632
cbcf2dd3 1633 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
d828199e
MT		 1634}
1635#endif
1636
1637/*
1638 *
b48aa97e
MT		 1639 * Assuming a stable TSC across physical CPUS, and a stable TSC
1640 * across virtual CPUs, the following condition is possible.
1641 * Each numbered line represents an event visible to both
d828199e
MT		 1642 * CPUs at the next numbered event.
1643 *
1644 * "timespecX" represents host monotonic time. "tscX" represents
1645 * RDTSC value.
1646 *
1647 * VCPU0 on CPU0 | VCPU1 on CPU1
1648 *
1649 * 1. read timespec0,tsc0
1650 * 2. | timespec1 = timespec0 + N
1651 * | tsc1 = tsc0 + M
1652 * 3. transition to guest | transition to guest
1653 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1654 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1655 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1656 *
1657 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1658 *
1659 * - ret0 < ret1
1660 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1661 * ...
1662 * - 0 < N - M => M < N
1663 *
1664 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1665 * always the case (the difference between two distinct xtime instances
1666 * might be smaller then the difference between corresponding TSC reads,
1667 * when updating guest vcpus pvclock areas).
1668 *
1669 * To avoid that problem, do not allow visibility of distinct
1670 * system_timestamp/tsc_timestamp values simultaneously: use a master
1671 * copy of host monotonic time values. Update that master copy
1672 * in lockstep.
1673 *
b48aa97e 1674 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
d828199e
MT		 1675 *
1676 */
1677
1678static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1679{
1680#ifdef CONFIG_X86_64
1681 struct kvm_arch *ka = &kvm->arch;
1682 int vclock_mode;
b48aa97e
MT		 1683 bool host_tsc_clocksource, vcpus_matched;
1684
1685 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1686 atomic_read(&kvm->online_vcpus));
d828199e
MT		 1687
1688 /*
1689 * If the host uses TSC clock, then passthrough TSC as stable
1690 * to the guest.
1691 */
b48aa97e 1692 host_tsc_clocksource = kvm_get_time_and_clockread(
d828199e
MT		 1693 &ka->master_kernel_ns,
1694 &ka->master_cycle_now);
1695
16a96021 1696 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
54750f2c
MT		 1697 && !backwards_tsc_observed
1698 && !ka->boot_vcpu_runs_old_kvmclock;
b48aa97e 1699
d828199e
MT		 1700 if (ka->use_master_clock)
1701 atomic_set(&kvm_guest_has_master_clock, 1);
1702
1703 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
b48aa97e
MT		 1704 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1705 vcpus_matched);
d828199e
MT		 1706#endif
1707}
1708
2860c4b1
PB		 1709void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1710{
1711 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1712}
1713
2e762ff7
MT		 1714static void kvm_gen_update_masterclock(struct kvm *kvm)
1715{
1716#ifdef CONFIG_X86_64
1717 int i;
1718 struct kvm_vcpu *vcpu;
1719 struct kvm_arch *ka = &kvm->arch;
1720
1721 spin_lock(&ka->pvclock_gtod_sync_lock);
1722 kvm_make_mclock_inprogress_request(kvm);
1723 /* no guest entries from this point */
1724 pvclock_update_vm_gtod_copy(kvm);
1725
1726 kvm_for_each_vcpu(i, vcpu, kvm)
105b21bb 1727 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2e762ff7
MT		 1728
1729 /* guest entries allowed */
1730 kvm_for_each_vcpu(i, vcpu, kvm)
1731 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1732
1733 spin_unlock(&ka->pvclock_gtod_sync_lock);
1734#endif
1735}
1736
108b249c
PB		 1737static u64 __get_kvmclock_ns(struct kvm *kvm)
1738{
108b249c 1739 struct kvm_arch *ka = &kvm->arch;
8b953440 1740 struct pvclock_vcpu_time_info hv_clock;
108b249c 1741
8b953440
PB		 1742 spin_lock(&ka->pvclock_gtod_sync_lock);
1743 if (!ka->use_master_clock) {
1744 spin_unlock(&ka->pvclock_gtod_sync_lock);
1745 return ktime_get_boot_ns() + ka->kvmclock_offset;
108b249c
PB		 1746 }
1747
8b953440
PB		 1748 hv_clock.tsc_timestamp = ka->master_cycle_now;
1749 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
1750 spin_unlock(&ka->pvclock_gtod_sync_lock);
1751
1752 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
1753 &hv_clock.tsc_shift,
1754 &hv_clock.tsc_to_system_mul);
1755 return __pvclock_read_cycles(&hv_clock, rdtsc());
108b249c
PB		 1756}
1757
1758u64 get_kvmclock_ns(struct kvm *kvm)
1759{
1760 unsigned long flags;
1761 s64 ns;
1762
1763 local_irq_save(flags);
1764 ns = __get_kvmclock_ns(kvm);
1765 local_irq_restore(flags);
1766
1767 return ns;
1768}
1769
0d6dd2ff
PB		 1770static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
1771{
1772 struct kvm_vcpu_arch *vcpu = &v->arch;
1773 struct pvclock_vcpu_time_info guest_hv_clock;
1774
1775 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1776 &guest_hv_clock, sizeof(guest_hv_clock))))
1777 return;
1778
1779 /* This VCPU is paused, but it's legal for a guest to read another
1780 * VCPU's kvmclock, so we really have to follow the specification where
1781 * it says that version is odd if data is being modified, and even after
1782 * it is consistent.
1783 *
1784 * Version field updates must be kept separate. This is because
1785 * kvm_write_guest_cached might use a "rep movs" instruction, and
1786 * writes within a string instruction are weakly ordered. So there
1787 * are three writes overall.
1788 *
1789 * As a small optimization, only write the version field in the first
1790 * and third write. The vcpu->pv_time cache is still valid, because the
1791 * version field is the first in the struct.
1792 */
1793 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1794
1795 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1796 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1797 &vcpu->hv_clock,
1798 sizeof(vcpu->hv_clock.version));
1799
1800 smp_wmb();
1801
1802 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1803 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1804
1805 if (vcpu->pvclock_set_guest_stopped_request) {
1806 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
1807 vcpu->pvclock_set_guest_stopped_request = false;
1808 }
1809
1810 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1811
1812 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1813 &vcpu->hv_clock,
1814 sizeof(vcpu->hv_clock));
1815
1816 smp_wmb();
1817
1818 vcpu->hv_clock.version++;
1819 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1820 &vcpu->hv_clock,
1821 sizeof(vcpu->hv_clock.version));
1822}
1823
34c238a1 1824static int kvm_guest_time_update(struct kvm_vcpu *v)
18068523 1825{
78db6a50 1826 unsigned long flags, tgt_tsc_khz;
18068523 1827 struct kvm_vcpu_arch *vcpu = &v->arch;
d828199e 1828 struct kvm_arch *ka = &v->kvm->arch;
f25e656d 1829 s64 kernel_ns;
d828199e 1830 u64 tsc_timestamp, host_tsc;
51d59c6b 1831 u8 pvclock_flags;
d828199e
MT		 1832 bool use_master_clock;
1833
1834 kernel_ns = 0;
1835 host_tsc = 0;
18068523 1836
d828199e
MT		 1837 /*
1838 * If the host uses TSC clock, then passthrough TSC as stable
1839 * to the guest.
1840 */
1841 spin_lock(&ka->pvclock_gtod_sync_lock);
1842 use_master_clock = ka->use_master_clock;
1843 if (use_master_clock) {
1844 host_tsc = ka->master_cycle_now;
1845 kernel_ns = ka->master_kernel_ns;
1846 }
1847 spin_unlock(&ka->pvclock_gtod_sync_lock);
c09664bb
MT		 1848
1849 /* Keep irq disabled to prevent changes to the clock */
1850 local_irq_save(flags);
78db6a50
PB		 1851 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1852 if (unlikely(tgt_tsc_khz == 0)) {
c09664bb
MT		 1853 local_irq_restore(flags);
1854 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1855 return 1;
1856 }
d828199e 1857 if (!use_master_clock) {
4ea1636b 1858 host_tsc = rdtsc();
108b249c 1859 kernel_ns = ktime_get_boot_ns();
d828199e
MT		 1860 }
1861
4ba76538 1862 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
d828199e 1863
c285545f
ZA		 1864 /*
1865 * We may have to catch up the TSC to match elapsed wall clock
1866 * time for two reasons, even if kvmclock is used.
1867 * 1) CPU could have been running below the maximum TSC rate
1868 * 2) Broken TSC compensation resets the base at each VCPU
1869 * entry to avoid unknown leaps of TSC even when running
1870 * again on the same CPU. This may cause apparent elapsed
1871 * time to disappear, and the guest to stand still or run
1872 * very slowly.
1873 */
1874 if (vcpu->tsc_catchup) {
1875 u64 tsc = compute_guest_tsc(v, kernel_ns);
1876 if (tsc > tsc_timestamp) {
f1e2b260 1877 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
c285545f
ZA		 1878 tsc_timestamp = tsc;
1879 }
50d0a0f9
GH		 1880 }
1881
18068523
GOC		 1882 local_irq_restore(flags);
1883
0d6dd2ff 1884 /* With all the info we got, fill in the values */
18068523 1885
78db6a50
PB		 1886 if (kvm_has_tsc_control)
1887 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
1888
1889 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
3ae13faa 1890 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
5f4e3f88
ZA		 1891 &vcpu->hv_clock.tsc_shift,
1892 &vcpu->hv_clock.tsc_to_system_mul);
78db6a50 1893 vcpu->hw_tsc_khz = tgt_tsc_khz;
8cfdc000
ZA		 1894 }
1895
1d5f066e 1896 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
759379dd 1897 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
28e4639a 1898 vcpu->last_guest_tsc = tsc_timestamp;
51d59c6b 1899
d828199e 1900 /* If the host uses TSC clocksource, then it is stable */
0d6dd2ff 1901 pvclock_flags = 0;
d828199e
MT		 1902 if (use_master_clock)
1903 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1904
78c0337a
MT		 1905 vcpu->hv_clock.flags = pvclock_flags;
1906
095cf55d
PB		 1907 if (vcpu->pv_time_enabled)
1908 kvm_setup_pvclock_page(v);
1909 if (v == kvm_get_vcpu(v->kvm, 0))
1910 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
8cfdc000 1911 return 0;
c8076604
GH		 1912}
1913
0061d53d
MT		 1914/*
1915 * kvmclock updates which are isolated to a given vcpu, such as
1916 * vcpu->cpu migration, should not allow system_timestamp from
1917 * the rest of the vcpus to remain static. Otherwise ntp frequency
1918 * correction applies to one vcpu's system_timestamp but not
1919 * the others.
1920 *
1921 * So in those cases, request a kvmclock update for all vcpus.
7e44e449
AJ		 1922 * We need to rate-limit these requests though, as they can
1923 * considerably slow guests that have a large number of vcpus.
1924 * The time for a remote vcpu to update its kvmclock is bound
1925 * by the delay we use to rate-limit the updates.
0061d53d
MT		 1926 */
1927
7e44e449
AJ		 1928#define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1929
1930static void kvmclock_update_fn(struct work_struct *work)
0061d53d
MT		 1931{
1932 int i;
7e44e449
AJ		 1933 struct delayed_work *dwork = to_delayed_work(work);
1934 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1935 kvmclock_update_work);
1936 struct kvm *kvm = container_of(ka, struct kvm, arch);
0061d53d
MT		 1937 struct kvm_vcpu *vcpu;
1938
1939 kvm_for_each_vcpu(i, vcpu, kvm) {
105b21bb 1940 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
0061d53d
MT		 1941 kvm_vcpu_kick(vcpu);
1942 }
1943}
1944
7e44e449
AJ		 1945static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1946{
1947 struct kvm *kvm = v->kvm;
1948
105b21bb 1949 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
7e44e449
AJ		 1950 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1951 KVMCLOCK_UPDATE_DELAY);
1952}
1953
332967a3
AJ		 1954#define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1955
1956static void kvmclock_sync_fn(struct work_struct *work)
1957{
1958 struct delayed_work *dwork = to_delayed_work(work);
1959 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1960 kvmclock_sync_work);
1961 struct kvm *kvm = container_of(ka, struct kvm, arch);
1962
630994b3
MT		 1963 if (!kvmclock_periodic_sync)
1964 return;
1965
332967a3
AJ		 1966 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1967 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1968 KVMCLOCK_SYNC_PERIOD);
1969}
1970
890ca9ae 1971static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
15c4a640 1972{
890ca9ae
HY		 1973 u64 mcg_cap = vcpu->arch.mcg_cap;
1974 unsigned bank_num = mcg_cap & 0xff;
1975
15c4a640 1976 switch (msr) {
15c4a640 1977 case MSR_IA32_MCG_STATUS:
890ca9ae 1978 vcpu->arch.mcg_status = data;
15c4a640 1979 break;
c7ac679c 1980 case MSR_IA32_MCG_CTL:
890ca9ae
HY		 1981 if (!(mcg_cap & MCG_CTL_P))
1982 return 1;
1983 if (data != 0 && data != ~(u64)0)
1984 return -1;
1985 vcpu->arch.mcg_ctl = data;
1986 break;
1987 default:
1988 if (msr >= MSR_IA32_MC0_CTL &&
81760dcc 1989 msr < MSR_IA32_MCx_CTL(bank_num)) {
890ca9ae 1990 u32 offset = msr - MSR_IA32_MC0_CTL;
114be429
AP		 1991 /* only 0 or all 1s can be written to IA32_MCi_CTL
1992 * some Linux kernels though clear bit 10 in bank 4 to
1993 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1994 * this to avoid an uncatched #GP in the guest
1995 */
890ca9ae 1996 if ((offset & 0x3) == 0 &&
114be429 1997 data != 0 && (data | (1 << 10)) != ~(u64)0)
890ca9ae
HY		 1998 return -1;
1999 vcpu->arch.mce_banks[offset] = data;
2000 break;
2001 }
2002 return 1;
2003 }
2004 return 0;
2005}
2006
ffde22ac
ES		 2007static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2008{
2009 struct kvm *kvm = vcpu->kvm;
2010 int lm = is_long_mode(vcpu);
2011 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2012 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2013 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2014 : kvm->arch.xen_hvm_config.blob_size_32;
2015 u32 page_num = data & ~PAGE_MASK;
2016 u64 page_addr = data & PAGE_MASK;
2017 u8 *page;
2018 int r;
2019
2020 r = -E2BIG;
2021 if (page_num >= blob_size)
2022 goto out;
2023 r = -ENOMEM;
ff5c2c03
SL		 2024 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2025 if (IS_ERR(page)) {
2026 r = PTR_ERR(page);
ffde22ac 2027 goto out;
ff5c2c03 2028 }
54bf36aa 2029 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
ffde22ac
ES		 2030 goto out_free;
2031 r = 0;
2032out_free:
2033 kfree(page);
2034out:
2035 return r;
2036}
2037
344d9588
GN		 2038static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2039{
2040 gpa_t gpa = data & ~0x3f;
2041
4a969980 2042 /* Bits 2:5 are reserved, Should be zero */
6adba527 2043 if (data & 0x3c)
344d9588
GN		 2044 return 1;
2045
2046 vcpu->arch.apf.msr_val = data;
2047
2048 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2049 kvm_clear_async_pf_completion_queue(vcpu);
2050 kvm_async_pf_hash_reset(vcpu);
2051 return 0;
2052 }
2053
8f964525
AH		 2054 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2055 sizeof(u32)))
344d9588
GN		 2056 return 1;
2057
6adba527 2058 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
344d9588
GN		 2059 kvm_async_pf_wakeup_all(vcpu);
2060 return 0;
2061}
2062
12f9a48f
GC		 2063static void kvmclock_reset(struct kvm_vcpu *vcpu)
2064{
0b79459b 2065 vcpu->arch.pv_time_enabled = false;
12f9a48f
GC		 2066}
2067
c9aaa895
GC		 2068static void record_steal_time(struct kvm_vcpu *vcpu)
2069{
2070 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2071 return;
2072
2073 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2074 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2075 return;
2076
0b9f6c46
PX		 2077 vcpu->arch.st.steal.preempted = 0;
2078
35f3fae1
WL		 2079 if (vcpu->arch.st.steal.version & 1)
2080 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2081
2082 vcpu->arch.st.steal.version += 1;
2083
2084 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2085 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2086
2087 smp_wmb();
2088
c54cdf14
LC		 2089 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2090 vcpu->arch.st.last_steal;
2091 vcpu->arch.st.last_steal = current->sched_info.run_delay;
35f3fae1
WL		 2092
2093 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2094 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2095
2096 smp_wmb();
2097
2098 vcpu->arch.st.steal.version += 1;
c9aaa895
GC		 2099
2100 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2101 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2102}
2103
8fe8ab46 2104int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
15c4a640 2105{
5753785f 2106 bool pr = false;
8fe8ab46
WA		 2107 u32 msr = msr_info->index;
2108 u64 data = msr_info->data;
5753785f 2109
15c4a640 2110 switch (msr) {
2e32b719
BP		 2111 case MSR_AMD64_NB_CFG:
2112 case MSR_IA32_UCODE_REV:
2113 case MSR_IA32_UCODE_WRITE:
2114 case MSR_VM_HSAVE_PA:
2115 case MSR_AMD64_PATCH_LOADER:
2116 case MSR_AMD64_BU_CFG2:
2117 break;
2118
15c4a640 2119 case MSR_EFER:
b69e8cae 2120 return set_efer(vcpu, data);
8f1589d9
AP		 2121 case MSR_K7_HWCR:
2122 data &= ~(u64)0x40; /* ignore flush filter disable */
82494028 2123 data &= ~(u64)0x100; /* ignore ignne emulation enable */
a223c313 2124 data &= ~(u64)0x8; /* ignore TLB cache disable */
22d48b2d 2125 data &= ~(u64)0x40000; /* ignore Mc status write enable */
8f1589d9 2126 if (data != 0) {
a737f256
CD		 2127 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2128 data);
8f1589d9
AP		 2129 return 1;
2130 }
15c4a640 2131 break;
f7c6d140
AP		 2132 case MSR_FAM10H_MMIO_CONF_BASE:
2133 if (data != 0) {
a737f256
CD		 2134 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2135 "0x%llx\n", data);
f7c6d140
AP		 2136 return 1;
2137 }
15c4a640 2138 break;
b5e2fec0
AG		 2139 case MSR_IA32_DEBUGCTLMSR:
2140 if (!data) {
2141 /* We support the non-activated case already */
2142 break;
2143 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2144 /* Values other than LBR and BTF are vendor-specific,
2145 thus reserved and should throw a #GP */
2146 return 1;
2147 }
a737f256
CD		 2148 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2149 __func__, data);
b5e2fec0 2150 break;
9ba075a6 2151 case 0x200 ... 0x2ff:
ff53604b 2152 return kvm_mtrr_set_msr(vcpu, msr, data);
15c4a640 2153 case MSR_IA32_APICBASE:
58cb628d 2154 return kvm_set_apic_base(vcpu, msr_info);
0105d1a5
GN		 2155 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2156 return kvm_x2apic_msr_write(vcpu, msr, data);
a3e06bbe
LJ		 2157 case MSR_IA32_TSCDEADLINE:
2158 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2159 break;
ba904635
WA		 2160 case MSR_IA32_TSC_ADJUST:
2161 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2162 if (!msr_info->host_initiated) {
d913b904 2163 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
d7add054 2164 adjust_tsc_offset_guest(vcpu, adj);
ba904635
WA		 2165 }
2166 vcpu->arch.ia32_tsc_adjust_msr = data;
2167 }
2168 break;
15c4a640 2169 case MSR_IA32_MISC_ENABLE:
ad312c7c 2170 vcpu->arch.ia32_misc_enable_msr = data;
15c4a640 2171 break;
64d60670
PB		 2172 case MSR_IA32_SMBASE:
2173 if (!msr_info->host_initiated)
2174 return 1;
2175 vcpu->arch.smbase = data;
2176 break;
11c6bffa 2177 case MSR_KVM_WALL_CLOCK_NEW:
18068523
GOC		 2178 case MSR_KVM_WALL_CLOCK:
2179 vcpu->kvm->arch.wall_clock = data;
2180 kvm_write_wall_clock(vcpu->kvm, data);
2181 break;
11c6bffa 2182 case MSR_KVM_SYSTEM_TIME_NEW:
18068523 2183 case MSR_KVM_SYSTEM_TIME: {
54750f2c
MT		 2184 struct kvm_arch *ka = &vcpu->kvm->arch;
2185
12f9a48f 2186 kvmclock_reset(vcpu);
18068523 2187
54750f2c
MT		 2188 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2189 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2190
2191 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2192 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2193 &vcpu->requests);
2194
2195 ka->boot_vcpu_runs_old_kvmclock = tmp;
2196 }
2197
18068523 2198 vcpu->arch.time = data;
0061d53d 2199 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
18068523
GOC		 2200
2201 /* we verify if the enable bit is set... */
2202 if (!(data & 1))
2203 break;
2204
0b79459b 2205 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
8f964525
AH		 2206 &vcpu->arch.pv_time, data & ~1ULL,
2207 sizeof(struct pvclock_vcpu_time_info)))
0b79459b
AH		 2208 vcpu->arch.pv_time_enabled = false;
2209 else
2210 vcpu->arch.pv_time_enabled = true;
32cad84f 2211
18068523
GOC		 2212 break;
2213 }
344d9588
GN		 2214 case MSR_KVM_ASYNC_PF_EN:
2215 if (kvm_pv_enable_async_pf(vcpu, data))
2216 return 1;
2217 break;
c9aaa895
GC		 2218 case MSR_KVM_STEAL_TIME:
2219
2220 if (unlikely(!sched_info_on()))
2221 return 1;
2222
2223 if (data & KVM_STEAL_RESERVED_MASK)
2224 return 1;
2225
2226 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
8f964525
AH		 2227 data & KVM_STEAL_VALID_BITS,
2228 sizeof(struct kvm_steal_time)))
c9aaa895
GC		 2229 return 1;
2230
2231 vcpu->arch.st.msr_val = data;
2232
2233 if (!(data & KVM_MSR_ENABLED))
2234 break;
2235
c9aaa895
GC		 2236 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2237
2238 break;
ae7a2a3f
MT		 2239 case MSR_KVM_PV_EOI_EN:
2240 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2241 return 1;
2242 break;
c9aaa895 2243
890ca9ae
HY		 2244 case MSR_IA32_MCG_CTL:
2245 case MSR_IA32_MCG_STATUS:
81760dcc 2246 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
890ca9ae 2247 return set_msr_mce(vcpu, msr, data);
71db6023 2248
6912ac32
WH		 2249 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2250 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2251 pr = true; /* fall through */
2252 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2253 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
c6702c9d 2254 if (kvm_pmu_is_valid_msr(vcpu, msr))
afd80d85 2255 return kvm_pmu_set_msr(vcpu, msr_info);
5753785f
GN		 2256
2257 if (pr || data != 0)
a737f256
CD		 2258 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2259 "0x%x data 0x%llx\n", msr, data);
5753785f 2260 break;
84e0cefa
JS		 2261 case MSR_K7_CLK_CTL:
2262 /*
2263 * Ignore all writes to this no longer documented MSR.
2264 * Writes are only relevant for old K7 processors,
2265 * all pre-dating SVM, but a recommended workaround from
4a969980 2266 * AMD for these chips. It is possible to specify the
84e0cefa
JS		 2267 * affected processor models on the command line, hence
2268 * the need to ignore the workaround.
2269 */
2270 break;
55cd8e5a 2271 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
e7d9513b
AS		 2272 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2273 case HV_X64_MSR_CRASH_CTL:
1f4b34f8 2274 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
e7d9513b
AS		 2275 return kvm_hv_set_msr_common(vcpu, msr, data,
2276 msr_info->host_initiated);
91c9c3ed 2277 case MSR_IA32_BBL_CR_CTL3:
2278 /* Drop writes to this legacy MSR -- see rdmsr
2279 * counterpart for further detail.
2280 */
796f4687 2281 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n", msr, data);
91c9c3ed 2282 break;
2b036c6b
BO		 2283 case MSR_AMD64_OSVW_ID_LENGTH:
2284 if (!guest_cpuid_has_osvw(vcpu))
2285 return 1;
2286 vcpu->arch.osvw.length = data;
2287 break;
2288 case MSR_AMD64_OSVW_STATUS:
2289 if (!guest_cpuid_has_osvw(vcpu))
2290 return 1;
2291 vcpu->arch.osvw.status = data;
2292 break;
15c4a640 2293 default:
ffde22ac
ES		 2294 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2295 return xen_hvm_config(vcpu, data);
c6702c9d 2296 if (kvm_pmu_is_valid_msr(vcpu, msr))
afd80d85 2297 return kvm_pmu_set_msr(vcpu, msr_info);
ed85c068 2298 if (!ignore_msrs) {
ae0f5499 2299 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
a737f256 2300 msr, data);
ed85c068
AP		 2301 return 1;
2302 } else {
796f4687 2303 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
a737f256 2304 msr, data);
ed85c068
AP		 2305 break;
2306 }
15c4a640
CO		 2307 }
2308 return 0;
2309}
2310EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2311
2312
2313/*
2314 * Reads an msr value (of 'msr_index') into 'pdata'.
2315 * Returns 0 on success, non-0 otherwise.
2316 * Assumes vcpu_load() was already called.
2317 */
609e36d3 2318int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
15c4a640 2319{
609e36d3 2320 return kvm_x86_ops->get_msr(vcpu, msr);
15c4a640 2321}
ff651cb6 2322EXPORT_SYMBOL_GPL(kvm_get_msr);
15c4a640 2323
890ca9ae 2324static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
15c4a640
CO		 2325{
2326 u64 data;
890ca9ae
HY		 2327 u64 mcg_cap = vcpu->arch.mcg_cap;
2328 unsigned bank_num = mcg_cap & 0xff;
15c4a640
CO		 2329
2330 switch (msr) {
15c4a640
CO		 2331 case MSR_IA32_P5_MC_ADDR:
2332 case MSR_IA32_P5_MC_TYPE:
890ca9ae
HY		 2333 data = 0;
2334 break;
15c4a640 2335 case MSR_IA32_MCG_CAP:
890ca9ae
HY		 2336 data = vcpu->arch.mcg_cap;
2337 break;
c7ac679c 2338 case MSR_IA32_MCG_CTL:
890ca9ae
HY		 2339 if (!(mcg_cap & MCG_CTL_P))
2340 return 1;
2341 data = vcpu->arch.mcg_ctl;
2342 break;
2343 case MSR_IA32_MCG_STATUS:
2344 data = vcpu->arch.mcg_status;
2345 break;
2346 default:
2347 if (msr >= MSR_IA32_MC0_CTL &&
81760dcc 2348 msr < MSR_IA32_MCx_CTL(bank_num)) {
890ca9ae
HY		 2349 u32 offset = msr - MSR_IA32_MC0_CTL;
2350 data = vcpu->arch.mce_banks[offset];
2351 break;
2352 }
2353 return 1;
2354 }
2355 *pdata = data;
2356 return 0;
2357}
2358
609e36d3 2359int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
890ca9ae 2360{
609e36d3 2361 switch (msr_info->index) {
890ca9ae 2362 case MSR_IA32_PLATFORM_ID:
15c4a640 2363 case MSR_IA32_EBL_CR_POWERON:
b5e2fec0
AG		 2364 case MSR_IA32_DEBUGCTLMSR:
2365 case MSR_IA32_LASTBRANCHFROMIP:
2366 case MSR_IA32_LASTBRANCHTOIP:
2367 case MSR_IA32_LASTINTFROMIP:
2368 case MSR_IA32_LASTINTTOIP:
60af2ecd 2369 case MSR_K8_SYSCFG:
3afb1121
PB		 2370 case MSR_K8_TSEG_ADDR:
2371 case MSR_K8_TSEG_MASK:
60af2ecd 2372 case MSR_K7_HWCR:
61a6bd67 2373 case MSR_VM_HSAVE_PA:
1fdbd48c 2374 case MSR_K8_INT_PENDING_MSG:
c323c0e5 2375 case MSR_AMD64_NB_CFG:
f7c6d140 2376 case MSR_FAM10H_MMIO_CONF_BASE:
2e32b719 2377 case MSR_AMD64_BU_CFG2:
0c2df2a1 2378 case MSR_IA32_PERF_CTL:
609e36d3 2379 msr_info->data = 0;
15c4a640 2380 break;
6912ac32
WH		 2381 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2382 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2383 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2384 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
c6702c9d 2385 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
609e36d3
PB		 2386 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2387 msr_info->data = 0;
5753785f 2388 break;
742bc670 2389 case MSR_IA32_UCODE_REV:
609e36d3 2390 msr_info->data = 0x100000000ULL;
742bc670 2391 break;
9ba075a6 2392 case MSR_MTRRcap:
9ba075a6 2393 case 0x200 ... 0x2ff:
ff53604b 2394 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
15c4a640 2395 case 0xcd: /* fsb frequency */
609e36d3 2396 msr_info->data = 3;
15c4a640 2397 break;
7b914098
JS		 2398 /*
2399 * MSR_EBC_FREQUENCY_ID
2400 * Conservative value valid for even the basic CPU models.
2401 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2402 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2403 * and 266MHz for model 3, or 4. Set Core Clock
2404 * Frequency to System Bus Frequency Ratio to 1 (bits
2405 * 31:24) even though these are only valid for CPU
2406 * models > 2, however guests may end up dividing or
2407 * multiplying by zero otherwise.
2408 */
2409 case MSR_EBC_FREQUENCY_ID:
609e36d3 2410 msr_info->data = 1 << 24;
7b914098 2411 break;
15c4a640 2412 case MSR_IA32_APICBASE:
609e36d3 2413 msr_info->data = kvm_get_apic_base(vcpu);
15c4a640 2414 break;
0105d1a5 2415 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
609e36d3 2416 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
0105d1a5 2417 break;
a3e06bbe 2418 case MSR_IA32_TSCDEADLINE:
609e36d3 2419 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
a3e06bbe 2420 break;
ba904635 2421 case MSR_IA32_TSC_ADJUST:
609e36d3 2422 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
ba904635 2423 break;
15c4a640 2424 case MSR_IA32_MISC_ENABLE:
609e36d3 2425 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
15c4a640 2426 break;
64d60670
PB		 2427 case MSR_IA32_SMBASE:
2428 if (!msr_info->host_initiated)
2429 return 1;
2430 msr_info->data = vcpu->arch.smbase;
15c4a640 2431 break;
847f0ad8
AG		 2432 case MSR_IA32_PERF_STATUS:
2433 /* TSC increment by tick */
609e36d3 2434 msr_info->data = 1000ULL;
847f0ad8 2435 /* CPU multiplier */
b0996ae4 2436 msr_info->data |= (((uint64_t)4ULL) << 40);
847f0ad8 2437 break;
15c4a640 2438 case MSR_EFER:
609e36d3 2439 msr_info->data = vcpu->arch.efer;
15c4a640 2440 break;
18068523 2441 case MSR_KVM_WALL_CLOCK:
11c6bffa 2442 case MSR_KVM_WALL_CLOCK_NEW:
609e36d3 2443 msr_info->data = vcpu->kvm->arch.wall_clock;
18068523
GOC		 2444 break;
2445 case MSR_KVM_SYSTEM_TIME:
11c6bffa 2446 case MSR_KVM_SYSTEM_TIME_NEW:
609e36d3 2447 msr_info->data = vcpu->arch.time;
18068523 2448 break;
344d9588 2449 case MSR_KVM_ASYNC_PF_EN:
609e36d3 2450 msr_info->data = vcpu->arch.apf.msr_val;
344d9588 2451 break;
c9aaa895 2452 case MSR_KVM_STEAL_TIME:
609e36d3 2453 msr_info->data = vcpu->arch.st.msr_val;
c9aaa895 2454 break;
1d92128f 2455 case MSR_KVM_PV_EOI_EN:
609e36d3 2456 msr_info->data = vcpu->arch.pv_eoi.msr_val;
1d92128f 2457 break;
890ca9ae
HY		 2458 case MSR_IA32_P5_MC_ADDR:
2459 case MSR_IA32_P5_MC_TYPE:
2460 case MSR_IA32_MCG_CAP:
2461 case MSR_IA32_MCG_CTL:
2462 case MSR_IA32_MCG_STATUS:
81760dcc 2463 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
609e36d3 2464 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
84e0cefa
JS		 2465 case MSR_K7_CLK_CTL:
2466 /*
2467 * Provide expected ramp-up count for K7. All other
2468 * are set to zero, indicating minimum divisors for
2469 * every field.
2470 *
2471 * This prevents guest kernels on AMD host with CPU
2472 * type 6, model 8 and higher from exploding due to
2473 * the rdmsr failing.
2474 */
609e36d3 2475 msr_info->data = 0x20000000;
84e0cefa 2476 break;
55cd8e5a 2477 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
e7d9513b
AS		 2478 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2479 case HV_X64_MSR_CRASH_CTL:
1f4b34f8 2480 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
e83d5887
AS		 2481 return kvm_hv_get_msr_common(vcpu,
2482 msr_info->index, &msr_info->data);
55cd8e5a 2483 break;
91c9c3ed 2484 case MSR_IA32_BBL_CR_CTL3:
2485 /* This legacy MSR exists but isn't fully documented in current
2486 * silicon. It is however accessed by winxp in very narrow
2487 * scenarios where it sets bit #19, itself documented as
2488 * a "reserved" bit. Best effort attempt to source coherent
2489 * read data here should the balance of the register be
2490 * interpreted by the guest:
2491 *
2492 * L2 cache control register 3: 64GB range, 256KB size,
2493 * enabled, latency 0x1, configured
2494 */
609e36d3 2495 msr_info->data = 0xbe702111;
91c9c3ed 2496 break;
2b036c6b
BO		 2497 case MSR_AMD64_OSVW_ID_LENGTH:
2498 if (!guest_cpuid_has_osvw(vcpu))
2499 return 1;
609e36d3 2500 msr_info->data = vcpu->arch.osvw.length;
2b036c6b
BO		 2501 break;
2502 case MSR_AMD64_OSVW_STATUS:
2503 if (!guest_cpuid_has_osvw(vcpu))
2504 return 1;
609e36d3 2505 msr_info->data = vcpu->arch.osvw.status;
2b036c6b 2506 break;
15c4a640 2507 default:
c6702c9d 2508 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
609e36d3 2509 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
ed85c068 2510 if (!ignore_msrs) {
ae0f5499
BD		 2511 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2512 msr_info->index);
ed85c068
AP		 2513 return 1;
2514 } else {
609e36d3
PB		 2515 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2516 msr_info->data = 0;
ed85c068
AP		 2517 }
2518 break;
15c4a640 2519 }
15c4a640
CO		 2520 return 0;
2521}
2522EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2523
313a3dc7
CO		 2524/*
2525 * Read or write a bunch of msrs. All parameters are kernel addresses.
2526 *
2527 * @return number of msrs set successfully.
2528 */
2529static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2530 struct kvm_msr_entry *entries,
2531 int (*do_msr)(struct kvm_vcpu *vcpu,
2532 unsigned index, u64 *data))
2533{
f656ce01 2534 int i, idx;
313a3dc7 2535
f656ce01 2536 idx = srcu_read_lock(&vcpu->kvm->srcu);
313a3dc7
CO		 2537 for (i = 0; i < msrs->nmsrs; ++i)
2538 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2539 break;
f656ce01 2540 srcu_read_unlock(&vcpu->kvm->srcu, idx);
313a3dc7 2541
313a3dc7
CO		 2542 return i;
2543}
2544
2545/*
2546 * Read or write a bunch of msrs. Parameters are user addresses.
2547 *
2548 * @return number of msrs set successfully.
2549 */
2550static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2551 int (*do_msr)(struct kvm_vcpu *vcpu,
2552 unsigned index, u64 *data),
2553 int writeback)
2554{
2555 struct kvm_msrs msrs;
2556 struct kvm_msr_entry *entries;
2557 int r, n;
2558 unsigned size;
2559
2560 r = -EFAULT;
2561 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2562 goto out;
2563
2564 r = -E2BIG;
2565 if (msrs.nmsrs >= MAX_IO_MSRS)
2566 goto out;
2567
313a3dc7 2568 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
ff5c2c03
SL		 2569 entries = memdup_user(user_msrs->entries, size);
2570 if (IS_ERR(entries)) {
2571 r = PTR_ERR(entries);
313a3dc7 2572 goto out;
ff5c2c03 2573 }
313a3dc7
CO		 2574
2575 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2576 if (r < 0)
2577 goto out_free;
2578
2579 r = -EFAULT;
2580 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2581 goto out_free;
2582
2583 r = n;
2584
2585out_free:
7a73c028 2586 kfree(entries);
313a3dc7
CO		 2587out:
2588 return r;
2589}
2590
784aa3d7 2591int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
018d00d2
ZX		 2592{
2593 int r;
2594
2595 switch (ext) {
2596 case KVM_CAP_IRQCHIP:
2597 case KVM_CAP_HLT:
2598 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
018d00d2 2599 case KVM_CAP_SET_TSS_ADDR:
07716717 2600 case KVM_CAP_EXT_CPUID:
9c15bb1d 2601 case KVM_CAP_EXT_EMUL_CPUID:
c8076604 2602 case KVM_CAP_CLOCKSOURCE:
7837699f 2603 case KVM_CAP_PIT:
a28e4f5a 2604 case KVM_CAP_NOP_IO_DELAY:
62d9f0db 2605 case KVM_CAP_MP_STATE:
ed848624 2606 case KVM_CAP_SYNC_MMU:
a355c85c 2607 case KVM_CAP_USER_NMI:
52d939a0 2608 case KVM_CAP_REINJECT_CONTROL:
4925663a 2609 case KVM_CAP_IRQ_INJECT_STATUS:
d34e6b17 2610 case KVM_CAP_IOEVENTFD:
f848a5a8 2611 case KVM_CAP_IOEVENTFD_NO_LENGTH:
c5ff41ce 2612 case KVM_CAP_PIT2:
e9f42757 2613 case KVM_CAP_PIT_STATE2:
b927a3ce 2614 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
ffde22ac 2615 case KVM_CAP_XEN_HVM:
3cfc3092 2616 case KVM_CAP_VCPU_EVENTS:
55cd8e5a 2617 case KVM_CAP_HYPERV:
10388a07 2618 case KVM_CAP_HYPERV_VAPIC:
c25bc163 2619 case KVM_CAP_HYPERV_SPIN:
5c919412 2620 case KVM_CAP_HYPERV_SYNIC:
ab9f4ecb 2621 case KVM_CAP_PCI_SEGMENT:
a1efbe77 2622 case KVM_CAP_DEBUGREGS:
d2be1651 2623 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2d5b5a66 2624 case KVM_CAP_XSAVE:
344d9588 2625 case KVM_CAP_ASYNC_PF:
92a1f12d 2626 case KVM_CAP_GET_TSC_KHZ:
1c0b28c2 2627 case KVM_CAP_KVMCLOCK_CTRL:
4d8b81ab 2628 case KVM_CAP_READONLY_MEM:
5f66b620 2629 case KVM_CAP_HYPERV_TIME:
100943c5 2630 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
defcf51f 2631 case KVM_CAP_TSC_DEADLINE_TIMER:
90de4a18
NA		 2632 case KVM_CAP_ENABLE_CAP_VM:
2633 case KVM_CAP_DISABLE_QUIRKS:
d71ba788 2634 case KVM_CAP_SET_BOOT_CPU_ID:
49df6397 2635 case KVM_CAP_SPLIT_IRQCHIP:
2a5bab10
AW		 2636#ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2637 case KVM_CAP_ASSIGN_DEV_IRQ:
2638 case KVM_CAP_PCI_2_3:
2639#endif
018d00d2
ZX		 2640 r = 1;
2641 break;
e3fd9a93
PB		 2642 case KVM_CAP_ADJUST_CLOCK:
2643 r = KVM_CLOCK_TSC_STABLE;
2644 break;
6d396b55
PB		 2645 case KVM_CAP_X86_SMM:
2646 /* SMBASE is usually relocated above 1M on modern chipsets,
2647 * and SMM handlers might indeed rely on 4G segment limits,
2648 * so do not report SMM to be available if real mode is
2649 * emulated via vm86 mode. Still, do not go to great lengths
2650 * to avoid userspace's usage of the feature, because it is a
2651 * fringe case that is not enabled except via specific settings
2652 * of the module parameters.
2653 */
2654 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2655 break;
542472b5
LV		 2656 case KVM_CAP_COALESCED_MMIO:
2657 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2658 break;
774ead3a
AK		 2659 case KVM_CAP_VAPIC:
2660 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2661 break;
f725230a 2662 case KVM_CAP_NR_VCPUS:
8c3ba334
SL		 2663 r = KVM_SOFT_MAX_VCPUS;
2664 break;
2665 case KVM_CAP_MAX_VCPUS:
f725230a
AK		 2666 r = KVM_MAX_VCPUS;
2667 break;
a988b910 2668 case KVM_CAP_NR_MEMSLOTS:
bbacc0c1 2669 r = KVM_USER_MEM_SLOTS;
a988b910 2670 break;
a68a6a72
MT		 2671 case KVM_CAP_PV_MMU: /* obsolete */
2672 r = 0;
2f333bcb 2673 break;
4cee4b72 2674#ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
62c476c7 2675 case KVM_CAP_IOMMU:
a1b60c1c 2676 r = iommu_present(&pci_bus_type);
62c476c7 2677 break;
4cee4b72 2678#endif
890ca9ae
HY		 2679 case KVM_CAP_MCE:
2680 r = KVM_MAX_MCE_BANKS;
2681 break;
2d5b5a66 2682 case KVM_CAP_XCRS:
d366bf7e 2683 r = boot_cpu_has(X86_FEATURE_XSAVE);
2d5b5a66 2684 break;
92a1f12d
JR		 2685 case KVM_CAP_TSC_CONTROL:
2686 r = kvm_has_tsc_control;
2687 break;
37131313
RK		 2688 case KVM_CAP_X2APIC_API:
2689 r = KVM_X2APIC_API_VALID_FLAGS;
2690 break;
018d00d2
ZX		 2691 default:
2692 r = 0;
2693 break;
2694 }
2695 return r;
2696
2697}
2698
043405e1
CO		 2699long kvm_arch_dev_ioctl(struct file *filp,
2700 unsigned int ioctl, unsigned long arg)
2701{
2702 void __user *argp = (void __user *)arg;
2703 long r;
2704
2705 switch (ioctl) {
2706 case KVM_GET_MSR_INDEX_LIST: {
2707 struct kvm_msr_list __user *user_msr_list = argp;
2708 struct kvm_msr_list msr_list;
2709 unsigned n;
2710
2711 r = -EFAULT;
2712 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2713 goto out;
2714 n = msr_list.nmsrs;
62ef68bb 2715 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
043405e1
CO		 2716 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2717 goto out;
2718 r = -E2BIG;
e125e7b6 2719 if (n < msr_list.nmsrs)
043405e1
CO		 2720 goto out;
2721 r = -EFAULT;
2722 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2723 num_msrs_to_save * sizeof(u32)))
2724 goto out;
e125e7b6 2725 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
043405e1 2726 &emulated_msrs,
62ef68bb 2727 num_emulated_msrs * sizeof(u32)))
043405e1
CO		 2728 goto out;
2729 r = 0;
2730 break;
2731 }
9c15bb1d
BP		 2732 case KVM_GET_SUPPORTED_CPUID:
2733 case KVM_GET_EMULATED_CPUID: {
674eea0f
AK		 2734 struct kvm_cpuid2 __user *cpuid_arg = argp;
2735 struct kvm_cpuid2 cpuid;
2736
2737 r = -EFAULT;
2738 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2739 goto out;
9c15bb1d
BP		 2740
2741 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2742 ioctl);
674eea0f
AK		 2743 if (r)
2744 goto out;
2745
2746 r = -EFAULT;
2747 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2748 goto out;
2749 r = 0;
2750 break;
2751 }
890ca9ae 2752 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
890ca9ae 2753 r = -EFAULT;
c45dcc71
AR		 2754 if (copy_to_user(argp, &kvm_mce_cap_supported,
2755 sizeof(kvm_mce_cap_supported)))
890ca9ae
HY		 2756 goto out;
2757 r = 0;
2758 break;
2759 }
043405e1
CO		 2760 default:
2761 r = -EINVAL;
2762 }
2763out:
2764 return r;
2765}
2766
f5f48ee1
SY		 2767static void wbinvd_ipi(void *garbage)
2768{
2769 wbinvd();
2770}
2771
2772static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2773{
e0f0bbc5 2774 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
f5f48ee1
SY		 2775}
2776
2860c4b1
PB		 2777static inline void kvm_migrate_timers(struct kvm_vcpu *vcpu)
2778{
2779 set_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests);
2780}
2781
313a3dc7
CO		 2782void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2783{
f5f48ee1
SY		 2784 /* Address WBINVD may be executed by guest */
2785 if (need_emulate_wbinvd(vcpu)) {
2786 if (kvm_x86_ops->has_wbinvd_exit())
2787 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2788 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2789 smp_call_function_single(vcpu->cpu,
2790 wbinvd_ipi, NULL, 1);
2791 }
2792
313a3dc7 2793 kvm_x86_ops->vcpu_load(vcpu, cpu);
8f6055cb 2794
0dd6a6ed
ZA		 2795 /* Apply any externally detected TSC adjustments (due to suspend) */
2796 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2797 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2798 vcpu->arch.tsc_offset_adjustment = 0;
105b21bb 2799 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
0dd6a6ed 2800 }
8f6055cb 2801
48434c20 2802 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
6f526ec5 2803 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4ea1636b 2804 rdtsc() - vcpu->arch.last_host_tsc;
e48672fa
ZA		 2805 if (tsc_delta < 0)
2806 mark_tsc_unstable("KVM discovered backwards TSC");
ce7a058a 2807
c285545f 2808 if (check_tsc_unstable()) {
07c1419a 2809 u64 offset = kvm_compute_tsc_offset(vcpu,
b183aa58 2810 vcpu->arch.last_guest_tsc);
a545ab6a 2811 kvm_vcpu_write_tsc_offset(vcpu, offset);
c285545f 2812 vcpu->arch.tsc_catchup = 1;
c285545f 2813 }
e12c8f36
WL		 2814 if (kvm_lapic_hv_timer_in_use(vcpu) &&
2815 kvm_x86_ops->set_hv_timer(vcpu,
498f8162 2816 kvm_get_lapic_target_expiration_tsc(vcpu)))
e12c8f36 2817 kvm_lapic_switch_to_sw_timer(vcpu);
d98d07ca
MT		 2818 /*
2819 * On a host with synchronized TSC, there is no need to update
2820 * kvmclock on vcpu->cpu migration
2821 */
2822 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
0061d53d 2823 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
c285545f
ZA		 2824 if (vcpu->cpu != cpu)
2825 kvm_migrate_timers(vcpu);
e48672fa 2826 vcpu->cpu = cpu;
6b7d7e76 2827 }
c9aaa895 2828
c9aaa895 2829 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
313a3dc7
CO		 2830}
2831
0b9f6c46
PX		 2832static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
2833{
2834 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2835 return;
2836
2837 vcpu->arch.st.steal.preempted = 1;
2838
2839 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
2840 &vcpu->arch.st.steal.preempted,
2841 offsetof(struct kvm_steal_time, preempted),
2842 sizeof(vcpu->arch.st.steal.preempted));
2843}
2844
313a3dc7
CO		 2845void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2846{
931f261b
AA		 2847 /*
2848 * Disable page faults because we're in atomic context here.
2849 * kvm_write_guest_offset_cached() would call might_fault()
2850 * that relies on pagefault_disable() to tell if there's a
2851 * bug. NOTE: the write to guest memory may not go through if
2852 * during postcopy live migration or if there's heavy guest
2853 * paging.
2854 */
2855 pagefault_disable();
0b9f6c46 2856 kvm_steal_time_set_preempted(vcpu);
931f261b 2857 pagefault_enable();
02daab21 2858 kvm_x86_ops->vcpu_put(vcpu);
1c11e713 2859 kvm_put_guest_fpu(vcpu);
4ea1636b 2860 vcpu->arch.last_host_tsc = rdtsc();
313a3dc7
CO		 2861}
2862
313a3dc7
CO		 2863static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2864 struct kvm_lapic_state *s)
2865{
d62caabb
AS		 2866 if (vcpu->arch.apicv_active)
2867 kvm_x86_ops->sync_pir_to_irr(vcpu);
2868
a92e2543 2869 return kvm_apic_get_state(vcpu, s);
313a3dc7
CO		 2870}
2871
2872static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2873 struct kvm_lapic_state *s)
2874{
a92e2543
RK		 2875 int r;
2876
2877 r = kvm_apic_set_state(vcpu, s);
2878 if (r)
2879 return r;
cb142eb7 2880 update_cr8_intercept(vcpu);
313a3dc7
CO		 2881
2882 return 0;
2883}
2884
127a457a
MG		 2885static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2886{
2887 return (!lapic_in_kernel(vcpu) ||
2888 kvm_apic_accept_pic_intr(vcpu));
2889}
2890
782d422b
MG		 2891/*
2892 * if userspace requested an interrupt window, check that the
2893 * interrupt window is open.
2894 *
2895 * No need to exit to userspace if we already have an interrupt queued.
2896 */
2897static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2898{
2899 return kvm_arch_interrupt_allowed(vcpu) &&
2900 !kvm_cpu_has_interrupt(vcpu) &&
2901 !kvm_event_needs_reinjection(vcpu) &&
2902 kvm_cpu_accept_dm_intr(vcpu);
2903}
2904
f77bc6a4
ZX		 2905static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2906 struct kvm_interrupt *irq)
2907{
02cdb50f 2908 if (irq->irq >= KVM_NR_INTERRUPTS)
f77bc6a4 2909 return -EINVAL;
1c1a9ce9
SR		 2910
2911 if (!irqchip_in_kernel(vcpu->kvm)) {
2912 kvm_queue_interrupt(vcpu, irq->irq, false);
2913 kvm_make_request(KVM_REQ_EVENT, vcpu);
2914 return 0;
2915 }
2916
2917 /*
2918 * With in-kernel LAPIC, we only use this to inject EXTINT, so
2919 * fail for in-kernel 8259.
2920 */
2921 if (pic_in_kernel(vcpu->kvm))
f77bc6a4 2922 return -ENXIO;
f77bc6a4 2923
1c1a9ce9
SR		 2924 if (vcpu->arch.pending_external_vector != -1)
2925 return -EEXIST;
f77bc6a4 2926
1c1a9ce9 2927 vcpu->arch.pending_external_vector = irq->irq;
934bf653 2928 kvm_make_request(KVM_REQ_EVENT, vcpu);
f77bc6a4
ZX		 2929 return 0;
2930}
2931
c4abb7c9
JK		 2932static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2933{
c4abb7c9 2934 kvm_inject_nmi(vcpu);
c4abb7c9
JK		 2935
2936 return 0;
2937}
2938
f077825a
PB		 2939static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2940{
64d60670
PB		 2941 kvm_make_request(KVM_REQ_SMI, vcpu);
2942
f077825a
PB		 2943 return 0;
2944}
2945
b209749f
AK		 2946static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2947 struct kvm_tpr_access_ctl *tac)
2948{
2949 if (tac->flags)
2950 return -EINVAL;
2951 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2952 return 0;
2953}
2954
890ca9ae
HY		 2955static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2956 u64 mcg_cap)
2957{
2958 int r;
2959 unsigned bank_num = mcg_cap & 0xff, bank;
2960
2961 r = -EINVAL;
a9e38c3e 2962 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
890ca9ae 2963 goto out;
c45dcc71 2964 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
890ca9ae
HY		 2965 goto out;
2966 r = 0;
2967 vcpu->arch.mcg_cap = mcg_cap;
2968 /* Init IA32_MCG_CTL to all 1s */
2969 if (mcg_cap & MCG_CTL_P)
2970 vcpu->arch.mcg_ctl = ~(u64)0;
2971 /* Init IA32_MCi_CTL to all 1s */
2972 for (bank = 0; bank < bank_num; bank++)
2973 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
c45dcc71
AR		 2974
2975 if (kvm_x86_ops->setup_mce)
2976 kvm_x86_ops->setup_mce(vcpu);
890ca9ae
HY		 2977out:
2978 return r;
2979}
2980
2981static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2982 struct kvm_x86_mce *mce)
2983{
2984 u64 mcg_cap = vcpu->arch.mcg_cap;
2985 unsigned bank_num = mcg_cap & 0xff;
2986 u64 *banks = vcpu->arch.mce_banks;
2987
2988 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2989 return -EINVAL;
2990 /*
2991 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2992 * reporting is disabled
2993 */
2994 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2995 vcpu->arch.mcg_ctl != ~(u64)0)
2996 return 0;
2997 banks += 4 * mce->bank;
2998 /*
2999 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3000 * reporting is disabled for the bank
3001 */
3002 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3003 return 0;
3004 if (mce->status & MCI_STATUS_UC) {
3005 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
fc78f519 3006 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
a8eeb04a 3007 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
890ca9ae
HY		 3008 return 0;
3009 }
3010 if (banks[1] & MCI_STATUS_VAL)
3011 mce->status |= MCI_STATUS_OVER;
3012 banks[2] = mce->addr;
3013 banks[3] = mce->misc;
3014 vcpu->arch.mcg_status = mce->mcg_status;
3015 banks[1] = mce->status;
3016 kvm_queue_exception(vcpu, MC_VECTOR);
3017 } else if (!(banks[1] & MCI_STATUS_VAL)
3018 || !(banks[1] & MCI_STATUS_UC)) {
3019 if (banks[1] & MCI_STATUS_VAL)
3020 mce->status |= MCI_STATUS_OVER;
3021 banks[2] = mce->addr;
3022 banks[3] = mce->misc;
3023 banks[1] = mce->status;
3024 } else
3025 banks[1] |= MCI_STATUS_OVER;
3026 return 0;
3027}
3028
3cfc3092
JK		 3029static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3030 struct kvm_vcpu_events *events)
3031{
7460fb4a 3032 process_nmi(vcpu);
03b82a30
JK		 3033 events->exception.injected =
3034 vcpu->arch.exception.pending &&
3035 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3cfc3092
JK		 3036 events->exception.nr = vcpu->arch.exception.nr;
3037 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
97e69aa6 3038 events->exception.pad = 0;
3cfc3092
JK		 3039 events->exception.error_code = vcpu->arch.exception.error_code;
3040
03b82a30
JK		 3041 events->interrupt.injected =
3042 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3cfc3092 3043 events->interrupt.nr = vcpu->arch.interrupt.nr;
03b82a30 3044 events->interrupt.soft = 0;
37ccdcbe 3045 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3cfc3092
JK		 3046
3047 events->nmi.injected = vcpu->arch.nmi_injected;
7460fb4a 3048 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3cfc3092 3049 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
97e69aa6 3050 events->nmi.pad = 0;
3cfc3092 3051
66450a21 3052 events->sipi_vector = 0; /* never valid when reporting to user space */
3cfc3092 3053
f077825a
PB		 3054 events->smi.smm = is_smm(vcpu);
3055 events->smi.pending = vcpu->arch.smi_pending;
3056 events->smi.smm_inside_nmi =
3057 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3058 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3059
dab4b911 3060 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
f077825a
PB		 3061 | KVM_VCPUEVENT_VALID_SHADOW
3062 | KVM_VCPUEVENT_VALID_SMM);
97e69aa6 3063 memset(&events->reserved, 0, sizeof(events->reserved));
3cfc3092
JK		 3064}
3065
3066static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3067 struct kvm_vcpu_events *events)
3068{
dab4b911 3069 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
48005f64 3070 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
f077825a
PB		 3071 | KVM_VCPUEVENT_VALID_SHADOW
3072 | KVM_VCPUEVENT_VALID_SMM))
3cfc3092
JK		 3073 return -EINVAL;
3074
78e546c8
PB		 3075 if (events->exception.injected &&
3076 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3077 return -EINVAL;
3078
7460fb4a 3079 process_nmi(vcpu);
3cfc3092
JK		 3080 vcpu->arch.exception.pending = events->exception.injected;
3081 vcpu->arch.exception.nr = events->exception.nr;
3082 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3083 vcpu->arch.exception.error_code = events->exception.error_code;
3084
3085 vcpu->arch.interrupt.pending = events->interrupt.injected;
3086 vcpu->arch.interrupt.nr = events->interrupt.nr;
3087 vcpu->arch.interrupt.soft = events->interrupt.soft;
48005f64
JK		 3088 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3089 kvm_x86_ops->set_interrupt_shadow(vcpu,
3090 events->interrupt.shadow);
3cfc3092
JK		 3091
3092 vcpu->arch.nmi_injected = events->nmi.injected;
dab4b911
JK		 3093 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3094 vcpu->arch.nmi_pending = events->nmi.pending;
3cfc3092
JK		 3095 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3096
66450a21 3097 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
bce87cce 3098 lapic_in_kernel(vcpu))
66450a21 3099 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3cfc3092 3100
f077825a
PB		 3101 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3102 if (events->smi.smm)
3103 vcpu->arch.hflags |= HF_SMM_MASK;
3104 else
3105 vcpu->arch.hflags &= ~HF_SMM_MASK;
3106 vcpu->arch.smi_pending = events->smi.pending;
3107 if (events->smi.smm_inside_nmi)
3108 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3109 else
3110 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
bce87cce 3111 if (lapic_in_kernel(vcpu)) {
f077825a
PB		 3112 if (events->smi.latched_init)
3113 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3114 else
3115 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3116 }
3117 }
3118
3842d135
AK		 3119 kvm_make_request(KVM_REQ_EVENT, vcpu);
3120
3cfc3092
JK		 3121 return 0;
3122}
3123
a1efbe77
JK		 3124static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3125 struct kvm_debugregs *dbgregs)
3126{
73aaf249
JK		 3127 unsigned long val;
3128
a1efbe77 3129 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
16f8a6f9 3130 kvm_get_dr(vcpu, 6, &val);
73aaf249 3131 dbgregs->dr6 = val;
a1efbe77
JK		 3132 dbgregs->dr7 = vcpu->arch.dr7;
3133 dbgregs->flags = 0;
97e69aa6 3134 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
a1efbe77
JK		 3135}
3136
3137static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3138 struct kvm_debugregs *dbgregs)
3139{
3140 if (dbgregs->flags)
3141 return -EINVAL;
3142
d14bdb55
PB		 3143 if (dbgregs->dr6 & ~0xffffffffull)
3144 return -EINVAL;
3145 if (dbgregs->dr7 & ~0xffffffffull)
3146 return -EINVAL;
3147
a1efbe77 3148 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
ae561ede 3149 kvm_update_dr0123(vcpu);
a1efbe77 3150 vcpu->arch.dr6 = dbgregs->dr6;
73aaf249 3151 kvm_update_dr6(vcpu);
a1efbe77 3152 vcpu->arch.dr7 = dbgregs->dr7;
9926c9fd 3153 kvm_update_dr7(vcpu);
a1efbe77 3154
a1efbe77
JK		 3155 return 0;
3156}
3157
df1daba7
PB		 3158#define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3159
3160static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3161{
c47ada30 3162 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
400e4b20 3163 u64 xstate_bv = xsave->header.xfeatures;
df1daba7
PB		 3164 u64 valid;
3165
3166 /*
3167 * Copy legacy XSAVE area, to avoid complications with CPUID
3168 * leaves 0 and 1 in the loop below.
3169 */
3170 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3171
3172 /* Set XSTATE_BV */
3173 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3174
3175 /*
3176 * Copy each region from the possibly compacted offset to the
3177 * non-compacted offset.
3178 */
d91cab78 3179 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
df1daba7
PB		 3180 while (valid) {
3181 u64 feature = valid & -valid;
3182 int index = fls64(feature) - 1;
3183 void *src = get_xsave_addr(xsave, feature);
3184
3185 if (src) {
3186 u32 size, offset, ecx, edx;
3187 cpuid_count(XSTATE_CPUID, index,
3188 &size, &offset, &ecx, &edx);
3189 memcpy(dest + offset, src, size);
3190 }
3191
3192 valid -= feature;
3193 }
3194}
3195
3196static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3197{
c47ada30 3198 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
df1daba7
PB		 3199 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3200 u64 valid;
3201
3202 /*
3203 * Copy legacy XSAVE area, to avoid complications with CPUID
3204 * leaves 0 and 1 in the loop below.
3205 */
3206 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3207
3208 /* Set XSTATE_BV and possibly XCOMP_BV. */
400e4b20 3209 xsave->header.xfeatures = xstate_bv;
782511b0 3210 if (boot_cpu_has(X86_FEATURE_XSAVES))
3a54450b 3211 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
df1daba7
PB		 3212
3213 /*
3214 * Copy each region from the non-compacted offset to the
3215 * possibly compacted offset.
3216 */
d91cab78 3217 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
df1daba7
PB		 3218 while (valid) {
3219 u64 feature = valid & -valid;
3220 int index = fls64(feature) - 1;
3221 void *dest = get_xsave_addr(xsave, feature);
3222
3223 if (dest) {
3224 u32 size, offset, ecx, edx;
3225 cpuid_count(XSTATE_CPUID, index,
3226 &size, &offset, &ecx, &edx);
3227 memcpy(dest, src + offset, size);
ee4100da 3228 }
df1daba7
PB		 3229
3230 valid -= feature;
3231 }
3232}
3233
2d5b5a66
SY		 3234static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3235 struct kvm_xsave *guest_xsave)
3236{
d366bf7e 3237 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
df1daba7
PB		 3238 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3239 fill_xsave((u8 *) guest_xsave->region, vcpu);
4344ee98 3240 } else {
2d5b5a66 3241 memcpy(guest_xsave->region,
7366ed77 3242 &vcpu->arch.guest_fpu.state.fxsave,
c47ada30 3243 sizeof(struct fxregs_state));
2d5b5a66 3244 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
d91cab78 3245 XFEATURE_MASK_FPSSE;
2d5b5a66
SY		 3246 }
3247}
3248
3249static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3250 struct kvm_xsave *guest_xsave)
3251{
3252 u64 xstate_bv =
3253 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3254
d366bf7e 3255 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
d7876f1b
PB		 3256 /*
3257 * Here we allow setting states that are not present in
3258 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3259 * with old userspace.
3260 */
4ff41732 3261 if (xstate_bv & ~kvm_supported_xcr0())
d7876f1b 3262 return -EINVAL;
df1daba7 3263 load_xsave(vcpu, (u8 *)guest_xsave->region);
d7876f1b 3264 } else {
d91cab78 3265 if (xstate_bv & ~XFEATURE_MASK_FPSSE)
2d5b5a66 3266 return -EINVAL;
7366ed77 3267 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
c47ada30 3268 guest_xsave->region, sizeof(struct fxregs_state));
2d5b5a66
SY		 3269 }
3270 return 0;
3271}
3272
3273static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3274 struct kvm_xcrs *guest_xcrs)
3275{
d366bf7e 3276 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
2d5b5a66
SY		 3277 guest_xcrs->nr_xcrs = 0;
3278 return;
3279 }
3280
3281 guest_xcrs->nr_xcrs = 1;
3282 guest_xcrs->flags = 0;
3283 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3284 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3285}
3286
3287static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3288 struct kvm_xcrs *guest_xcrs)
3289{
3290 int i, r = 0;
3291
d366bf7e 3292 if (!boot_cpu_has(X86_FEATURE_XSAVE))
2d5b5a66
SY		 3293 return -EINVAL;
3294
3295 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3296 return -EINVAL;
3297
3298 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3299 /* Only support XCR0 currently */
c67a04cb 3300 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
2d5b5a66 3301 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
c67a04cb 3302 guest_xcrs->xcrs[i].value);
2d5b5a66
SY		 3303 break;
3304 }
3305 if (r)
3306 r = -EINVAL;
3307 return r;
3308}
3309
1c0b28c2
EM		 3310/*
3311 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3312 * stopped by the hypervisor. This function will be called from the host only.
3313 * EINVAL is returned when the host attempts to set the flag for a guest that
3314 * does not support pv clocks.
3315 */
3316static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3317{
0b79459b 3318 if (!vcpu->arch.pv_time_enabled)
1c0b28c2 3319 return -EINVAL;
51d59c6b 3320 vcpu->arch.pvclock_set_guest_stopped_request = true;
1c0b28c2
EM		 3321 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3322 return 0;
3323}
3324
5c919412
AS		 3325static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3326 struct kvm_enable_cap *cap)
3327{
3328 if (cap->flags)
3329 return -EINVAL;
3330
3331 switch (cap->cap) {
3332 case KVM_CAP_HYPERV_SYNIC:
3333 return kvm_hv_activate_synic(vcpu);
3334 default:
3335 return -EINVAL;
3336 }
3337}
3338
313a3dc7
CO		 3339long kvm_arch_vcpu_ioctl(struct file *filp,
3340 unsigned int ioctl, unsigned long arg)
3341{
3342 struct kvm_vcpu *vcpu = filp->private_data;
3343 void __user *argp = (void __user *)arg;
3344 int r;
d1ac91d8
AK		 3345 union {
3346 struct kvm_lapic_state *lapic;
3347 struct kvm_xsave *xsave;
3348 struct kvm_xcrs *xcrs;
3349 void *buffer;
3350 } u;
3351
3352 u.buffer = NULL;
313a3dc7
CO		 3353 switch (ioctl) {
3354 case KVM_GET_LAPIC: {
2204ae3c 3355 r = -EINVAL;
bce87cce 3356 if (!lapic_in_kernel(vcpu))
2204ae3c 3357 goto out;
d1ac91d8 3358 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
313a3dc7 3359
b772ff36 3360 r = -ENOMEM;
d1ac91d8 3361 if (!u.lapic)
b772ff36 3362 goto out;
d1ac91d8 3363 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
313a3dc7
CO		 3364 if (r)
3365 goto out;
3366 r = -EFAULT;
d1ac91d8 3367 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
313a3dc7
CO		 3368 goto out;
3369 r = 0;
3370 break;
3371 }
3372 case KVM_SET_LAPIC: {
2204ae3c 3373 r = -EINVAL;
bce87cce 3374 if (!lapic_in_kernel(vcpu))
2204ae3c 3375 goto out;
ff5c2c03 3376 u.lapic = memdup_user(argp, sizeof(*u.lapic));
18595411
GC		 3377 if (IS_ERR(u.lapic))
3378 return PTR_ERR(u.lapic);
ff5c2c03 3379
d1ac91d8 3380 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
313a3dc7
CO		 3381 break;
3382 }
f77bc6a4
ZX		 3383 case KVM_INTERRUPT: {
3384 struct kvm_interrupt irq;
3385
3386 r = -EFAULT;
3387 if (copy_from_user(&irq, argp, sizeof irq))
3388 goto out;
3389 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
f77bc6a4
ZX		 3390 break;
3391 }
c4abb7c9
JK		 3392 case KVM_NMI: {
3393 r = kvm_vcpu_ioctl_nmi(vcpu);
c4abb7c9
JK		 3394 break;
3395 }
f077825a
PB		 3396 case KVM_SMI: {
3397 r = kvm_vcpu_ioctl_smi(vcpu);
3398 break;
3399 }
313a3dc7
CO		 3400 case KVM_SET_CPUID: {
3401 struct kvm_cpuid __user *cpuid_arg = argp;
3402 struct kvm_cpuid cpuid;
3403
3404 r = -EFAULT;
3405 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3406 goto out;
3407 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
313a3dc7
CO		 3408 break;
3409 }
07716717
DK		 3410 case KVM_SET_CPUID2: {
3411 struct kvm_cpuid2 __user *cpuid_arg = argp;
3412 struct kvm_cpuid2 cpuid;
3413
3414 r = -EFAULT;
3415 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3416 goto out;
3417 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
19355475 3418 cpuid_arg->entries);
07716717
DK		 3419 break;
3420 }
3421 case KVM_GET_CPUID2: {
3422 struct kvm_cpuid2 __user *cpuid_arg = argp;
3423 struct kvm_cpuid2 cpuid;
3424
3425 r = -EFAULT;
3426 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3427 goto out;
3428 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
19355475 3429 cpuid_arg->entries);
07716717
DK		 3430 if (r)
3431 goto out;
3432 r = -EFAULT;
3433 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3434 goto out;
3435 r = 0;
3436 break;
3437 }
313a3dc7 3438 case KVM_GET_MSRS:
609e36d3 3439 r = msr_io(vcpu, argp, do_get_msr, 1);
313a3dc7
CO		 3440 break;
3441 case KVM_SET_MSRS:
3442 r = msr_io(vcpu, argp, do_set_msr, 0);
3443 break;
b209749f
AK		 3444 case KVM_TPR_ACCESS_REPORTING: {
3445 struct kvm_tpr_access_ctl tac;
3446
3447 r = -EFAULT;
3448 if (copy_from_user(&tac, argp, sizeof tac))
3449 goto out;
3450 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3451 if (r)
3452 goto out;
3453 r = -EFAULT;
3454 if (copy_to_user(argp, &tac, sizeof tac))
3455 goto out;
3456 r = 0;
3457 break;
3458 };
b93463aa
AK		 3459 case KVM_SET_VAPIC_ADDR: {
3460 struct kvm_vapic_addr va;
7301d6ab 3461 int idx;
b93463aa
AK		 3462
3463 r = -EINVAL;
35754c98 3464 if (!lapic_in_kernel(vcpu))
b93463aa
AK		 3465 goto out;
3466 r = -EFAULT;
3467 if (copy_from_user(&va, argp, sizeof va))
3468 goto out;
7301d6ab 3469 idx = srcu_read_lock(&vcpu->kvm->srcu);
fda4e2e8 3470 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
7301d6ab 3471 srcu_read_unlock(&vcpu->kvm->srcu, idx);
b93463aa
AK		 3472 break;
3473 }
890ca9ae
HY		 3474 case KVM_X86_SETUP_MCE: {
3475 u64 mcg_cap;
3476
3477 r = -EFAULT;
3478 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3479 goto out;
3480 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3481 break;
3482 }
3483 case KVM_X86_SET_MCE: {
3484 struct kvm_x86_mce mce;
3485
3486 r = -EFAULT;
3487 if (copy_from_user(&mce, argp, sizeof mce))
3488 goto out;
3489 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3490 break;
3491 }
3cfc3092
JK		 3492 case KVM_GET_VCPU_EVENTS: {
3493 struct kvm_vcpu_events events;
3494
3495 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3496
3497 r = -EFAULT;
3498 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3499 break;
3500 r = 0;
3501 break;
3502 }
3503 case KVM_SET_VCPU_EVENTS: {
3504 struct kvm_vcpu_events events;
3505
3506 r = -EFAULT;
3507 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3508 break;
3509
3510 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3511 break;
3512 }
a1efbe77
JK		 3513 case KVM_GET_DEBUGREGS: {
3514 struct kvm_debugregs dbgregs;
3515
3516 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3517
3518 r = -EFAULT;
3519 if (copy_to_user(argp, &dbgregs,
3520 sizeof(struct kvm_debugregs)))
3521 break;
3522 r = 0;
3523 break;
3524 }
3525 case KVM_SET_DEBUGREGS: {
3526 struct kvm_debugregs dbgregs;
3527
3528 r = -EFAULT;
3529 if (copy_from_user(&dbgregs, argp,
3530 sizeof(struct kvm_debugregs)))
3531 break;
3532
3533 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3534 break;
3535 }
2d5b5a66 3536 case KVM_GET_XSAVE: {
d1ac91d8 3537 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2d5b5a66 3538 r = -ENOMEM;
d1ac91d8 3539 if (!u.xsave)
2d5b5a66
SY		 3540 break;
3541
d1ac91d8 3542 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
2d5b5a66
SY		 3543
3544 r = -EFAULT;
d1ac91d8 3545 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
2d5b5a66
SY		 3546 break;
3547 r = 0;
3548 break;
3549 }
3550 case KVM_SET_XSAVE: {
ff5c2c03 3551 u.xsave = memdup_user(argp, sizeof(*u.xsave));
18595411
GC		 3552 if (IS_ERR(u.xsave))
3553 return PTR_ERR(u.xsave);
2d5b5a66 3554
d1ac91d8 3555 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
2d5b5a66
SY		 3556 break;
3557 }
3558 case KVM_GET_XCRS: {
d1ac91d8 3559 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2d5b5a66 3560 r = -ENOMEM;
d1ac91d8 3561 if (!u.xcrs)
2d5b5a66
SY		 3562 break;
3563
d1ac91d8 3564 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
2d5b5a66
SY		 3565
3566 r = -EFAULT;
d1ac91d8 3567 if (copy_to_user(argp, u.xcrs,
2d5b5a66
SY		 3568 sizeof(struct kvm_xcrs)))
3569 break;
3570 r = 0;
3571 break;
3572 }
3573 case KVM_SET_XCRS: {
ff5c2c03 3574 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
18595411
GC		 3575 if (IS_ERR(u.xcrs))
3576 return PTR_ERR(u.xcrs);
2d5b5a66 3577
d1ac91d8 3578 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
2d5b5a66
SY		 3579 break;
3580 }
92a1f12d
JR		 3581 case KVM_SET_TSC_KHZ: {
3582 u32 user_tsc_khz;
3583
3584 r = -EINVAL;
92a1f12d
JR		 3585 user_tsc_khz = (u32)arg;
3586
3587 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3588 goto out;
3589
cc578287
ZA		 3590 if (user_tsc_khz == 0)
3591 user_tsc_khz = tsc_khz;
3592
381d585c
HZ		 3593 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3594 r = 0;
92a1f12d 3595
92a1f12d
JR		 3596 goto out;
3597 }
3598 case KVM_GET_TSC_KHZ: {
cc578287 3599 r = vcpu->arch.virtual_tsc_khz;
92a1f12d
JR		 3600 goto out;
3601 }
1c0b28c2
EM		 3602 case KVM_KVMCLOCK_CTRL: {
3603 r = kvm_set_guest_paused(vcpu);
3604 goto out;
3605 }
5c919412
AS		 3606 case KVM_ENABLE_CAP: {
3607 struct kvm_enable_cap cap;
3608
3609 r = -EFAULT;
3610 if (copy_from_user(&cap, argp, sizeof(cap)))
3611 goto out;
3612 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3613 break;
3614 }
313a3dc7
CO		 3615 default:
3616 r = -EINVAL;
3617 }
3618out:
d1ac91d8 3619 kfree(u.buffer);
313a3dc7
CO		 3620 return r;
3621}
3622
5b1c1493
CO		 3623int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3624{
3625 return VM_FAULT_SIGBUS;
3626}
3627
1fe779f8
CO		 3628static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3629{
3630 int ret;
3631
3632 if (addr > (unsigned int)(-3 * PAGE_SIZE))
951179ce 3633 return -EINVAL;
1fe779f8
CO		 3634 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3635 return ret;
3636}
3637
b927a3ce
SY		 3638static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3639 u64 ident_addr)
3640{
3641 kvm->arch.ept_identity_map_addr = ident_addr;
3642 return 0;
3643}
3644
1fe779f8
CO		 3645static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3646 u32 kvm_nr_mmu_pages)
3647{
3648 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3649 return -EINVAL;
3650
79fac95e 3651 mutex_lock(&kvm->slots_lock);
1fe779f8
CO		 3652
3653 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
f05e70ac 3654 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1fe779f8 3655
79fac95e 3656 mutex_unlock(&kvm->slots_lock);
1fe779f8
CO		 3657 return 0;
3658}
3659
3660static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3661{
39de71ec 3662 return kvm->arch.n_max_mmu_pages;
1fe779f8
CO		 3663}
3664
1fe779f8
CO		 3665static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3666{
3667 int r;
3668
3669 r = 0;
3670 switch (chip->chip_id) {
3671 case KVM_IRQCHIP_PIC_MASTER:
3672 memcpy(&chip->chip.pic,
3673 &pic_irqchip(kvm)->pics[0],
3674 sizeof(struct kvm_pic_state));
3675 break;
3676 case KVM_IRQCHIP_PIC_SLAVE:
3677 memcpy(&chip->chip.pic,
3678 &pic_irqchip(kvm)->pics[1],
3679 sizeof(struct kvm_pic_state));
3680 break;
3681 case KVM_IRQCHIP_IOAPIC:
eba0226b 3682 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
1fe779f8
CO		 3683 break;
3684 default:
3685 r = -EINVAL;
3686 break;
3687 }
3688 return r;
3689}
3690
3691static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3692{
3693 int r;
3694
3695 r = 0;
3696 switch (chip->chip_id) {
3697 case KVM_IRQCHIP_PIC_MASTER:
f4f51050 3698 spin_lock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3699 memcpy(&pic_irqchip(kvm)->pics[0],
3700 &chip->chip.pic,
3701 sizeof(struct kvm_pic_state));
f4f51050 3702 spin_unlock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3703 break;
3704 case KVM_IRQCHIP_PIC_SLAVE:
f4f51050 3705 spin_lock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3706 memcpy(&pic_irqchip(kvm)->pics[1],
3707 &chip->chip.pic,
3708 sizeof(struct kvm_pic_state));
f4f51050 3709 spin_unlock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3710 break;
3711 case KVM_IRQCHIP_IOAPIC:
eba0226b 3712 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
1fe779f8
CO		 3713 break;
3714 default:
3715 r = -EINVAL;
3716 break;
3717 }
3718 kvm_pic_update_irq(pic_irqchip(kvm));
3719 return r;
3720}
3721
e0f63cb9
SY		 3722static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3723{
34f3941c
RK		 3724 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3725
3726 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3727
3728 mutex_lock(&kps->lock);
3729 memcpy(ps, &kps->channels, sizeof(*ps));
3730 mutex_unlock(&kps->lock);
2da29bcc 3731 return 0;
e0f63cb9
SY		 3732}
3733
3734static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3735{
0185604c 3736 int i;
09edea72
RK		 3737 struct kvm_pit *pit = kvm->arch.vpit;
3738
3739 mutex_lock(&pit->pit_state.lock);
34f3941c 3740 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
0185604c 3741 for (i = 0; i < 3; i++)
09edea72
RK		 3742 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3743 mutex_unlock(&pit->pit_state.lock);
2da29bcc 3744 return 0;
e9f42757
BK		 3745}
3746
3747static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3748{
e9f42757
BK		 3749 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3750 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3751 sizeof(ps->channels));
3752 ps->flags = kvm->arch.vpit->pit_state.flags;
3753 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
97e69aa6 3754 memset(&ps->reserved, 0, sizeof(ps->reserved));
2da29bcc 3755 return 0;
e9f42757
BK		 3756}
3757
3758static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3759{
2da29bcc 3760 int start = 0;
0185604c 3761 int i;
e9f42757 3762 u32 prev_legacy, cur_legacy;
09edea72
RK		 3763 struct kvm_pit *pit = kvm->arch.vpit;
3764
3765 mutex_lock(&pit->pit_state.lock);
3766 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
e9f42757
BK		 3767 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3768 if (!prev_legacy && cur_legacy)
3769 start = 1;
09edea72
RK		 3770 memcpy(&pit->pit_state.channels, &ps->channels,
3771 sizeof(pit->pit_state.channels));
3772 pit->pit_state.flags = ps->flags;
0185604c 3773 for (i = 0; i < 3; i++)
09edea72 3774 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
e5e57e7a 3775 start && i == 0);
09edea72 3776 mutex_unlock(&pit->pit_state.lock);
2da29bcc 3777 return 0;
e0f63cb9
SY		 3778}
3779
52d939a0
MT		 3780static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3781 struct kvm_reinject_control *control)
3782{
71474e2f
RK		 3783 struct kvm_pit *pit = kvm->arch.vpit;
3784
3785 if (!pit)
52d939a0 3786 return -ENXIO;
b39c90b6 3787
71474e2f
RK		 3788 /* pit->pit_state.lock was overloaded to prevent userspace from getting
3789 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3790 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
3791 */
3792 mutex_lock(&pit->pit_state.lock);
3793 kvm_pit_set_reinject(pit, control->pit_reinject);
3794 mutex_unlock(&pit->pit_state.lock);
b39c90b6 3795
52d939a0
MT		 3796 return 0;
3797}
3798
95d4c16c 3799/**
60c34612
TY		 3800 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3801 * @kvm: kvm instance
3802 * @log: slot id and address to which we copy the log
95d4c16c 3803 *
e108ff2f
PB		 3804 * Steps 1-4 below provide general overview of dirty page logging. See
3805 * kvm_get_dirty_log_protect() function description for additional details.
3806 *
3807 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3808 * always flush the TLB (step 4) even if previous step failed and the dirty
3809 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3810 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3811 * writes will be marked dirty for next log read.
95d4c16c 3812 *
60c34612
TY		 3813 * 1. Take a snapshot of the bit and clear it if needed.
3814 * 2. Write protect the corresponding page.
e108ff2f
PB		 3815 * 3. Copy the snapshot to the userspace.
3816 * 4. Flush TLB's if needed.
5bb064dc 3817 */
60c34612 3818int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
5bb064dc 3819{
60c34612 3820 bool is_dirty = false;
e108ff2f 3821 int r;
5bb064dc 3822
79fac95e 3823 mutex_lock(&kvm->slots_lock);
5bb064dc 3824
88178fd4
KH		 3825 /*
3826 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3827 */
3828 if (kvm_x86_ops->flush_log_dirty)
3829 kvm_x86_ops->flush_log_dirty(kvm);
3830
e108ff2f 3831 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
198c74f4
XG		 3832
3833 /*
3834 * All the TLBs can be flushed out of mmu lock, see the comments in
3835 * kvm_mmu_slot_remove_write_access().
3836 */
e108ff2f 3837 lockdep_assert_held(&kvm->slots_lock);
198c74f4
XG		 3838 if (is_dirty)
3839 kvm_flush_remote_tlbs(kvm);
3840
79fac95e 3841 mutex_unlock(&kvm->slots_lock);
5bb064dc
ZX		 3842 return r;
3843}
3844
aa2fbe6d
YZ		 3845int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3846 bool line_status)
23d43cf9
CD		 3847{
3848 if (!irqchip_in_kernel(kvm))
3849 return -ENXIO;
3850
3851 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
aa2fbe6d
YZ		 3852 irq_event->irq, irq_event->level,
3853 line_status);
23d43cf9
CD		 3854 return 0;
3855}
3856
90de4a18
NA		 3857static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3858 struct kvm_enable_cap *cap)
3859{
3860 int r;
3861
3862 if (cap->flags)
3863 return -EINVAL;
3864
3865 switch (cap->cap) {
3866 case KVM_CAP_DISABLE_QUIRKS:
3867 kvm->arch.disabled_quirks = cap->args[0];
3868 r = 0;
3869 break;
49df6397
SR		 3870 case KVM_CAP_SPLIT_IRQCHIP: {
3871 mutex_lock(&kvm->lock);
b053b2ae
SR		 3872 r = -EINVAL;
3873 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3874 goto split_irqchip_unlock;
49df6397
SR		 3875 r = -EEXIST;
3876 if (irqchip_in_kernel(kvm))
3877 goto split_irqchip_unlock;
557abc40 3878 if (kvm->created_vcpus)
49df6397
SR		 3879 goto split_irqchip_unlock;
3880 r = kvm_setup_empty_irq_routing(kvm);
3881 if (r)
3882 goto split_irqchip_unlock;
3883 /* Pairs with irqchip_in_kernel. */
3884 smp_wmb();
3885 kvm->arch.irqchip_split = true;
b053b2ae 3886 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
49df6397
SR		 3887 r = 0;
3888split_irqchip_unlock:
3889 mutex_unlock(&kvm->lock);
3890 break;
3891 }
37131313
RK		 3892 case KVM_CAP_X2APIC_API:
3893 r = -EINVAL;
3894 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
3895 break;
3896
3897 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
3898 kvm->arch.x2apic_format = true;
c519265f
RK		 3899 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
3900 kvm->arch.x2apic_broadcast_quirk_disabled = true;
37131313
RK		 3901
3902 r = 0;
3903 break;
90de4a18
NA		 3904 default:
3905 r = -EINVAL;
3906 break;
3907 }
3908 return r;
3909}
3910
1fe779f8
CO		 3911long kvm_arch_vm_ioctl(struct file *filp,
3912 unsigned int ioctl, unsigned long arg)
3913{
3914 struct kvm *kvm = filp->private_data;
3915 void __user *argp = (void __user *)arg;
367e1319 3916 int r = -ENOTTY;
f0d66275
DH		 3917 /*
3918 * This union makes it completely explicit to gcc-3.x
3919 * that these two variables' stack usage should be
3920 * combined, not added together.
3921 */
3922 union {
3923 struct kvm_pit_state ps;
e9f42757 3924 struct kvm_pit_state2 ps2;
c5ff41ce 3925 struct kvm_pit_config pit_config;
f0d66275 3926 } u;
1fe779f8
CO		 3927
3928 switch (ioctl) {
3929 case KVM_SET_TSS_ADDR:
3930 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1fe779f8 3931 break;
b927a3ce
SY		 3932 case KVM_SET_IDENTITY_MAP_ADDR: {
3933 u64 ident_addr;
3934
3935 r = -EFAULT;
3936 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3937 goto out;
3938 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
b927a3ce
SY		 3939 break;
3940 }
1fe779f8
CO		 3941 case KVM_SET_NR_MMU_PAGES:
3942 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1fe779f8
CO		 3943 break;
3944 case KVM_GET_NR_MMU_PAGES:
3945 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3946 break;
3ddea128
MT		 3947 case KVM_CREATE_IRQCHIP: {
3948 struct kvm_pic *vpic;
3949
3950 mutex_lock(&kvm->lock);
3951 r = -EEXIST;
3952 if (kvm->arch.vpic)
3953 goto create_irqchip_unlock;
3e515705 3954 r = -EINVAL;
557abc40 3955 if (kvm->created_vcpus)
3e515705 3956 goto create_irqchip_unlock;
1fe779f8 3957 r = -ENOMEM;
3ddea128
MT		 3958 vpic = kvm_create_pic(kvm);
3959 if (vpic) {
1fe779f8
CO		 3960 r = kvm_ioapic_init(kvm);
3961 if (r) {
175504cd 3962 mutex_lock(&kvm->slots_lock);
71ba994c 3963 kvm_destroy_pic(vpic);
175504cd 3964 mutex_unlock(&kvm->slots_lock);
3ddea128 3965 goto create_irqchip_unlock;
1fe779f8
CO		 3966 }
3967 } else
3ddea128 3968 goto create_irqchip_unlock;
399ec807
AK		 3969 r = kvm_setup_default_irq_routing(kvm);
3970 if (r) {
175504cd 3971 mutex_lock(&kvm->slots_lock);
3ddea128 3972 mutex_lock(&kvm->irq_lock);
72bb2fcd 3973 kvm_ioapic_destroy(kvm);
71ba994c 3974 kvm_destroy_pic(vpic);
3ddea128 3975 mutex_unlock(&kvm->irq_lock);
175504cd 3976 mutex_unlock(&kvm->slots_lock);
71ba994c 3977 goto create_irqchip_unlock;
399ec807 3978 }
71ba994c
PB		 3979 /* Write kvm->irq_routing before kvm->arch.vpic. */
3980 smp_wmb();
3981 kvm->arch.vpic = vpic;
3ddea128
MT		 3982 create_irqchip_unlock:
3983 mutex_unlock(&kvm->lock);
1fe779f8 3984 break;
3ddea128 3985 }
7837699f 3986 case KVM_CREATE_PIT:
c5ff41ce
JK		 3987 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3988 goto create_pit;
3989 case KVM_CREATE_PIT2:
3990 r = -EFAULT;
3991 if (copy_from_user(&u.pit_config, argp,
3992 sizeof(struct kvm_pit_config)))
3993 goto out;
3994 create_pit:
250715a6 3995 mutex_lock(&kvm->lock);
269e05e4
AK		 3996 r = -EEXIST;
3997 if (kvm->arch.vpit)
3998 goto create_pit_unlock;
7837699f 3999 r = -ENOMEM;
c5ff41ce 4000 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
7837699f
SY		 4001 if (kvm->arch.vpit)
4002 r = 0;
269e05e4 4003 create_pit_unlock:
250715a6 4004 mutex_unlock(&kvm->lock);
7837699f 4005 break;
1fe779f8
CO		 4006 case KVM_GET_IRQCHIP: {
4007 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
ff5c2c03 4008 struct kvm_irqchip *chip;
1fe779f8 4009
ff5c2c03
SL		 4010 chip = memdup_user(argp, sizeof(*chip));
4011 if (IS_ERR(chip)) {
4012 r = PTR_ERR(chip);
1fe779f8 4013 goto out;
ff5c2c03
SL		 4014 }
4015
1fe779f8 4016 r = -ENXIO;
49df6397 4017 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
f0d66275
DH		 4018 goto get_irqchip_out;
4019 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1fe779f8 4020 if (r)
f0d66275 4021 goto get_irqchip_out;
1fe779f8 4022 r = -EFAULT;
f0d66275
DH		 4023 if (copy_to_user(argp, chip, sizeof *chip))
4024 goto get_irqchip_out;
1fe779f8 4025 r = 0;
f0d66275
DH		 4026 get_irqchip_out:
4027 kfree(chip);
1fe779f8
CO		 4028 break;
4029 }
4030 case KVM_SET_IRQCHIP: {
4031 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
ff5c2c03 4032 struct kvm_irqchip *chip;
1fe779f8 4033
ff5c2c03
SL		 4034 chip = memdup_user(argp, sizeof(*chip));
4035 if (IS_ERR(chip)) {
4036 r = PTR_ERR(chip);
1fe779f8 4037 goto out;
ff5c2c03
SL		 4038 }
4039
1fe779f8 4040 r = -ENXIO;
49df6397 4041 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
f0d66275
DH		 4042 goto set_irqchip_out;
4043 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1fe779f8 4044 if (r)
f0d66275 4045 goto set_irqchip_out;
1fe779f8 4046 r = 0;
f0d66275
DH		 4047 set_irqchip_out:
4048 kfree(chip);
1fe779f8
CO		 4049 break;
4050 }
e0f63cb9 4051 case KVM_GET_PIT: {
e0f63cb9 4052 r = -EFAULT;
f0d66275 4053 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
e0f63cb9
SY		 4054 goto out;
4055 r = -ENXIO;
4056 if (!kvm->arch.vpit)
4057 goto out;
f0d66275 4058 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
e0f63cb9
SY		 4059 if (r)
4060 goto out;
4061 r = -EFAULT;
f0d66275 4062 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
e0f63cb9
SY		 4063 goto out;
4064 r = 0;
4065 break;
4066 }
4067 case KVM_SET_PIT: {
e0f63cb9 4068 r = -EFAULT;
f0d66275 4069 if (copy_from_user(&u.ps, argp, sizeof u.ps))
e0f63cb9
SY		 4070 goto out;
4071 r = -ENXIO;
4072 if (!kvm->arch.vpit)
4073 goto out;
f0d66275 4074 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
e0f63cb9
SY		 4075 break;
4076 }
e9f42757
BK		 4077 case KVM_GET_PIT2: {
4078 r = -ENXIO;
4079 if (!kvm->arch.vpit)
4080 goto out;
4081 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4082 if (r)
4083 goto out;
4084 r = -EFAULT;
4085 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4086 goto out;
4087 r = 0;
4088 break;
4089 }
4090 case KVM_SET_PIT2: {
4091 r = -EFAULT;
4092 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4093 goto out;
4094 r = -ENXIO;
4095 if (!kvm->arch.vpit)
4096 goto out;
4097 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
e9f42757
BK		 4098 break;
4099 }
52d939a0
MT		 4100 case KVM_REINJECT_CONTROL: {
4101 struct kvm_reinject_control control;
4102 r = -EFAULT;
4103 if (copy_from_user(&control, argp, sizeof(control)))
4104 goto out;
4105 r = kvm_vm_ioctl_reinject(kvm, &control);
52d939a0
MT		 4106 break;
4107 }
d71ba788
PB		 4108 case KVM_SET_BOOT_CPU_ID:
4109 r = 0;
4110 mutex_lock(&kvm->lock);
557abc40 4111 if (kvm->created_vcpus)
d71ba788
PB		 4112 r = -EBUSY;
4113 else
4114 kvm->arch.bsp_vcpu_id = arg;
4115 mutex_unlock(&kvm->lock);
4116 break;
ffde22ac
ES		 4117 case KVM_XEN_HVM_CONFIG: {
4118 r = -EFAULT;
4119 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4120 sizeof(struct kvm_xen_hvm_config)))
4121 goto out;
4122 r = -EINVAL;
4123 if (kvm->arch.xen_hvm_config.flags)
4124 goto out;
4125 r = 0;
4126 break;
4127 }
afbcf7ab 4128 case KVM_SET_CLOCK: {
afbcf7ab
GC		 4129 struct kvm_clock_data user_ns;
4130 u64 now_ns;
afbcf7ab
GC		 4131
4132 r = -EFAULT;
4133 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4134 goto out;
4135
4136 r = -EINVAL;
4137 if (user_ns.flags)
4138 goto out;
4139
4140 r = 0;
395c6b0a 4141 local_irq_disable();
108b249c
PB		 4142 now_ns = __get_kvmclock_ns(kvm);
4143 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
395c6b0a 4144 local_irq_enable();
2e762ff7 4145 kvm_gen_update_masterclock(kvm);
afbcf7ab
GC		 4146 break;
4147 }
4148 case KVM_GET_CLOCK: {
afbcf7ab
GC		 4149 struct kvm_clock_data user_ns;
4150 u64 now_ns;
4151
e3fd9a93
PB		 4152 local_irq_disable();
4153 now_ns = __get_kvmclock_ns(kvm);
108b249c 4154 user_ns.clock = now_ns;
e3fd9a93
PB		 4155 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4156 local_irq_enable();
97e69aa6 4157 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
afbcf7ab
GC		 4158
4159 r = -EFAULT;
4160 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4161 goto out;
4162 r = 0;
4163 break;
4164 }
90de4a18
NA		 4165 case KVM_ENABLE_CAP: {
4166 struct kvm_enable_cap cap;
afbcf7ab 4167
90de4a18
NA		 4168 r = -EFAULT;
4169 if (copy_from_user(&cap, argp, sizeof(cap)))
4170 goto out;
4171 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4172 break;
4173 }
1fe779f8 4174 default:
c274e03a 4175 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1fe779f8
CO		 4176 }
4177out:
4178 return r;
4179}
4180
a16b043c 4181static void kvm_init_msr_list(void)
043405e1
CO		 4182{
4183 u32 dummy[2];
4184 unsigned i, j;
4185
62ef68bb 4186 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
043405e1
CO		 4187 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4188 continue;
93c4adc7
PB		 4189
4190 /*
4191 * Even MSRs that are valid in the host may not be exposed
9dbe6cf9 4192 * to the guests in some cases.
93c4adc7
PB		 4193 */
4194 switch (msrs_to_save[i]) {
4195 case MSR_IA32_BNDCFGS:
4196 if (!kvm_x86_ops->mpx_supported())
4197 continue;
4198 break;
9dbe6cf9
PB		 4199 case MSR_TSC_AUX:
4200 if (!kvm_x86_ops->rdtscp_supported())
4201 continue;
4202 break;
93c4adc7
PB		 4203 default:
4204 break;
4205 }
4206
043405e1
CO		 4207 if (j < i)
4208 msrs_to_save[j] = msrs_to_save[i];
4209 j++;
4210 }
4211 num_msrs_to_save = j;
62ef68bb
PB		 4212
4213 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4214 switch (emulated_msrs[i]) {
6d396b55
PB		 4215 case MSR_IA32_SMBASE:
4216 if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4217 continue;
4218 break;
62ef68bb
PB		 4219 default:
4220 break;
4221 }
4222
4223 if (j < i)
4224 emulated_msrs[j] = emulated_msrs[i];
4225 j++;
4226 }
4227 num_emulated_msrs = j;
043405e1
CO		 4228}
4229
bda9020e
MT		 4230static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4231 const void *v)
bbd9b64e 4232{
70252a10
AK		 4233 int handled = 0;
4234 int n;
4235
4236 do {
4237 n = min(len, 8);
bce87cce 4238 if (!(lapic_in_kernel(vcpu) &&
e32edf4f
NN		 4239 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4240 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
70252a10
AK		 4241 break;
4242 handled += n;
4243 addr += n;
4244 len -= n;
4245 v += n;
4246 } while (len);
bbd9b64e 4247
70252a10 4248 return handled;
bbd9b64e
CO		 4249}
4250
bda9020e 4251static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
bbd9b64e 4252{
70252a10
AK		 4253 int handled = 0;
4254 int n;
4255
4256 do {
4257 n = min(len, 8);
bce87cce 4258 if (!(lapic_in_kernel(vcpu) &&
e32edf4f
NN		 4259 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4260 addr, n, v))
4261 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
70252a10
AK		 4262 break;
4263 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4264 handled += n;
4265 addr += n;
4266 len -= n;
4267 v += n;
4268 } while (len);
bbd9b64e 4269
70252a10 4270 return handled;
bbd9b64e
CO		 4271}
4272
2dafc6c2
GN		 4273static void kvm_set_segment(struct kvm_vcpu *vcpu,
4274 struct kvm_segment *var, int seg)
4275{
4276 kvm_x86_ops->set_segment(vcpu, var, seg);
4277}
4278
4279void kvm_get_segment(struct kvm_vcpu *vcpu,
4280 struct kvm_segment *var, int seg)
4281{
4282 kvm_x86_ops->get_segment(vcpu, var, seg);
4283}
4284
54987b7a
PB		 4285gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4286 struct x86_exception *exception)
02f59dc9
JR		 4287{
4288 gpa_t t_gpa;
02f59dc9
JR		 4289
4290 BUG_ON(!mmu_is_nested(vcpu));
4291
4292 /* NPT walks are always user-walks */
4293 access |= PFERR_USER_MASK;
54987b7a 4294 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
02f59dc9
JR		 4295
4296 return t_gpa;
4297}
4298
ab9ae313
AK		 4299gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4300 struct x86_exception *exception)
1871c602
GN		 4301{
4302 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
ab9ae313 4303 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 4304}
4305
ab9ae313
AK		 4306 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4307 struct x86_exception *exception)
1871c602
GN		 4308{
4309 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4310 access |= PFERR_FETCH_MASK;
ab9ae313 4311 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 4312}
4313
ab9ae313
AK		 4314gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4315 struct x86_exception *exception)
1871c602
GN		 4316{
4317 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4318 access |= PFERR_WRITE_MASK;
ab9ae313 4319 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 4320}
4321
4322/* uses this to access any guest's mapped memory without checking CPL */
ab9ae313
AK		 4323gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4324 struct x86_exception *exception)
1871c602 4325{
ab9ae313 4326 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
1871c602
GN		 4327}
4328
4329static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4330 struct kvm_vcpu *vcpu, u32 access,
bcc55cba 4331 struct x86_exception *exception)
bbd9b64e
CO		 4332{
4333 void *data = val;
10589a46 4334 int r = X86EMUL_CONTINUE;
bbd9b64e
CO		 4335
4336 while (bytes) {
14dfe855 4337 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
ab9ae313 4338 exception);
bbd9b64e 4339 unsigned offset = addr & (PAGE_SIZE-1);
77c2002e 4340 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
bbd9b64e
CO		 4341 int ret;
4342
bcc55cba 4343 if (gpa == UNMAPPED_GVA)
ab9ae313 4344 return X86EMUL_PROPAGATE_FAULT;
54bf36aa
PB		 4345 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4346 offset, toread);
10589a46 4347 if (ret < 0) {
c3cd7ffa 4348 r = X86EMUL_IO_NEEDED;
10589a46
MT		 4349 goto out;
4350 }
bbd9b64e 4351
77c2002e
IE		 4352 bytes -= toread;
4353 data += toread;
4354 addr += toread;
bbd9b64e 4355 }
10589a46 4356out:
10589a46 4357 return r;
bbd9b64e 4358}
77c2002e 4359
1871c602 4360/* used for instruction fetching */
0f65dd70
AK		 4361static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4362 gva_t addr, void *val, unsigned int bytes,
bcc55cba 4363 struct x86_exception *exception)
1871c602 4364{
0f65dd70 4365 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
1871c602 4366 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
44583cba
PB		 4367 unsigned offset;
4368 int ret;
0f65dd70 4369
44583cba
PB		 4370 /* Inline kvm_read_guest_virt_helper for speed. */
4371 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4372 exception);
4373 if (unlikely(gpa == UNMAPPED_GVA))
4374 return X86EMUL_PROPAGATE_FAULT;
4375
4376 offset = addr & (PAGE_SIZE-1);
4377 if (WARN_ON(offset + bytes > PAGE_SIZE))
4378 bytes = (unsigned)PAGE_SIZE - offset;
54bf36aa
PB		 4379 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4380 offset, bytes);
44583cba
PB		 4381 if (unlikely(ret < 0))
4382 return X86EMUL_IO_NEEDED;
4383
4384 return X86EMUL_CONTINUE;
1871c602
GN		 4385}
4386
064aea77 4387int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
0f65dd70 4388 gva_t addr, void *val, unsigned int bytes,
bcc55cba 4389 struct x86_exception *exception)
1871c602 4390{
0f65dd70 4391 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
1871c602 4392 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
0f65dd70 4393
1871c602 4394 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
bcc55cba 4395 exception);
1871c602 4396}
064aea77 4397EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
1871c602 4398
0f65dd70
AK		 4399static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4400 gva_t addr, void *val, unsigned int bytes,
bcc55cba 4401 struct x86_exception *exception)
1871c602 4402{
0f65dd70 4403 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
bcc55cba 4404 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
1871c602
GN		 4405}
4406
7a036a6f
RK		 4407static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4408 unsigned long addr, void *val, unsigned int bytes)
4409{
4410 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4411 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4412
4413 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4414}
4415
6a4d7550 4416int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
0f65dd70 4417 gva_t addr, void *val,
2dafc6c2 4418 unsigned int bytes,
bcc55cba 4419 struct x86_exception *exception)
77c2002e 4420{
0f65dd70 4421 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
77c2002e
IE		 4422 void *data = val;
4423 int r = X86EMUL_CONTINUE;
4424
4425 while (bytes) {
14dfe855
JR		 4426 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4427 PFERR_WRITE_MASK,
ab9ae313 4428 exception);
77c2002e
IE		 4429 unsigned offset = addr & (PAGE_SIZE-1);
4430 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4431 int ret;
4432
bcc55cba 4433 if (gpa == UNMAPPED_GVA)
ab9ae313 4434 return X86EMUL_PROPAGATE_FAULT;
54bf36aa 4435 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
77c2002e 4436 if (ret < 0) {
c3cd7ffa 4437 r = X86EMUL_IO_NEEDED;
77c2002e
IE		 4438 goto out;
4439 }
4440
4441 bytes -= towrite;
4442 data += towrite;
4443 addr += towrite;
4444 }
4445out:
4446 return r;
4447}
6a4d7550 4448EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
77c2002e 4449
af7cc7d1
XG		 4450static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4451 gpa_t *gpa, struct x86_exception *exception,
4452 bool write)
4453{
97d64b78
AK		 4454 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4455 | (write ? PFERR_WRITE_MASK : 0);
af7cc7d1 4456
be94f6b7
HH		 4457 /*
4458 * currently PKRU is only applied to ept enabled guest so
4459 * there is no pkey in EPT page table for L1 guest or EPT
4460 * shadow page table for L2 guest.
4461 */
97d64b78 4462 if (vcpu_match_mmio_gva(vcpu, gva)
97ec8c06 4463 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
be94f6b7 4464 vcpu->arch.access, 0, access)) {
bebb106a
XG		 4465 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4466 (gva & (PAGE_SIZE - 1));
4f022648 4467 trace_vcpu_match_mmio(gva, *gpa, write, false);
bebb106a
XG		 4468 return 1;
4469 }
4470
af7cc7d1
XG		 4471 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4472
4473 if (*gpa == UNMAPPED_GVA)
4474 return -1;
4475
4476 /* For APIC access vmexit */
4477 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4478 return 1;
4479
4f022648
XG		 4480 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4481 trace_vcpu_match_mmio(gva, *gpa, write, true);
bebb106a 4482 return 1;
4f022648 4483 }
bebb106a 4484
af7cc7d1
XG		 4485 return 0;
4486}
4487
3200f405 4488int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
bcc55cba 4489 const void *val, int bytes)
bbd9b64e
CO		 4490{
4491 int ret;
4492
54bf36aa 4493 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
9f811285 4494 if (ret < 0)
bbd9b64e 4495 return 0;
0eb05bf2 4496 kvm_page_track_write(vcpu, gpa, val, bytes);
bbd9b64e
CO		 4497 return 1;
4498}
4499
77d197b2
XG		 4500struct read_write_emulator_ops {
4501 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4502 int bytes);
4503 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4504 void *val, int bytes);
4505 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4506 int bytes, void *val);
4507 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4508 void *val, int bytes);
4509 bool write;
4510};
4511
4512static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4513{
4514 if (vcpu->mmio_read_completed) {
77d197b2 4515 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
f78146b0 4516 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
77d197b2
XG		 4517 vcpu->mmio_read_completed = 0;
4518 return 1;
4519 }
4520
4521 return 0;
4522}
4523
4524static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4525 void *val, int bytes)
4526{
54bf36aa 4527 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
77d197b2
XG		 4528}
4529
4530static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4531 void *val, int bytes)
4532{
4533 return emulator_write_phys(vcpu, gpa, val, bytes);
4534}
4535
4536static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4537{
4538 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4539 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4540}
4541
4542static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4543 void *val, int bytes)
4544{
4545 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4546 return X86EMUL_IO_NEEDED;
4547}
4548
4549static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4550 void *val, int bytes)
4551{
f78146b0
AK		 4552 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4553
87da7e66 4554 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
77d197b2
XG		 4555 return X86EMUL_CONTINUE;
4556}
4557
0fbe9b0b 4558static const struct read_write_emulator_ops read_emultor = {
77d197b2
XG		 4559 .read_write_prepare = read_prepare,
4560 .read_write_emulate = read_emulate,
4561 .read_write_mmio = vcpu_mmio_read,
4562 .read_write_exit_mmio = read_exit_mmio,
4563};
4564
0fbe9b0b 4565static const struct read_write_emulator_ops write_emultor = {
77d197b2
XG		 4566 .read_write_emulate = write_emulate,
4567 .read_write_mmio = write_mmio,
4568 .read_write_exit_mmio = write_exit_mmio,
4569 .write = true,
4570};
4571
22388a3c
XG		 4572static int emulator_read_write_onepage(unsigned long addr, void *val,
4573 unsigned int bytes,
4574 struct x86_exception *exception,
4575 struct kvm_vcpu *vcpu,
0fbe9b0b 4576 const struct read_write_emulator_ops *ops)
bbd9b64e 4577{
af7cc7d1
XG		 4578 gpa_t gpa;
4579 int handled, ret;
22388a3c 4580 bool write = ops->write;
f78146b0 4581 struct kvm_mmio_fragment *frag;
10589a46 4582
22388a3c 4583 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
bbd9b64e 4584
af7cc7d1 4585 if (ret < 0)
bbd9b64e 4586 return X86EMUL_PROPAGATE_FAULT;
bbd9b64e
CO		 4587
4588 /* For APIC access vmexit */
af7cc7d1 4589 if (ret)
bbd9b64e
CO		 4590 goto mmio;
4591
22388a3c 4592 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
bbd9b64e
CO		 4593 return X86EMUL_CONTINUE;
4594
4595mmio:
4596 /*
4597 * Is this MMIO handled locally?
4598 */
22388a3c 4599 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
70252a10 4600 if (handled == bytes)
bbd9b64e 4601 return X86EMUL_CONTINUE;
bbd9b64e 4602
70252a10
AK		 4603 gpa += handled;
4604 bytes -= handled;
4605 val += handled;
4606
87da7e66
XG		 4607 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4608 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4609 frag->gpa = gpa;
4610 frag->data = val;
4611 frag->len = bytes;
f78146b0 4612 return X86EMUL_CONTINUE;
bbd9b64e
CO		 4613}
4614
52eb5a6d
XL		 4615static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4616 unsigned long addr,
22388a3c
XG		 4617 void *val, unsigned int bytes,
4618 struct x86_exception *exception,
0fbe9b0b 4619 const struct read_write_emulator_ops *ops)
bbd9b64e 4620{
0f65dd70 4621 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
f78146b0
AK		 4622 gpa_t gpa;
4623 int rc;
4624
4625 if (ops->read_write_prepare &&
4626 ops->read_write_prepare(vcpu, val, bytes))
4627 return X86EMUL_CONTINUE;
4628
4629 vcpu->mmio_nr_fragments = 0;
0f65dd70 4630
bbd9b64e
CO		 4631 /* Crossing a page boundary? */
4632 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
f78146b0 4633 int now;
bbd9b64e
CO		 4634
4635 now = -addr & ~PAGE_MASK;
22388a3c
XG		 4636 rc = emulator_read_write_onepage(addr, val, now, exception,
4637 vcpu, ops);
4638
bbd9b64e
CO		 4639 if (rc != X86EMUL_CONTINUE)
4640 return rc;
4641 addr += now;
bac15531
NA		 4642 if (ctxt->mode != X86EMUL_MODE_PROT64)
4643 addr = (u32)addr;
bbd9b64e
CO		 4644 val += now;
4645 bytes -= now;
4646 }
22388a3c 4647
f78146b0
AK		 4648 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4649 vcpu, ops);
4650 if (rc != X86EMUL_CONTINUE)
4651 return rc;
4652
4653 if (!vcpu->mmio_nr_fragments)
4654 return rc;
4655
4656 gpa = vcpu->mmio_fragments[0].gpa;
4657
4658 vcpu->mmio_needed = 1;
4659 vcpu->mmio_cur_fragment = 0;
4660
87da7e66 4661 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
f78146b0
AK		 4662 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4663 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4664 vcpu->run->mmio.phys_addr = gpa;
4665
4666 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
22388a3c
XG		 4667}
4668
4669static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4670 unsigned long addr,
4671 void *val,
4672 unsigned int bytes,
4673 struct x86_exception *exception)
4674{
4675 return emulator_read_write(ctxt, addr, val, bytes,
4676 exception, &read_emultor);
4677}
4678
52eb5a6d 4679static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
22388a3c
XG		 4680 unsigned long addr,
4681 const void *val,
4682 unsigned int bytes,
4683 struct x86_exception *exception)
4684{
4685 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4686 exception, &write_emultor);
bbd9b64e 4687}
bbd9b64e 4688
daea3e73
AK		 4689#define CMPXCHG_TYPE(t, ptr, old, new) \
4690 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4691
4692#ifdef CONFIG_X86_64
4693# define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4694#else
4695# define CMPXCHG64(ptr, old, new) \
9749a6c0 4696 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
daea3e73
AK		 4697#endif
4698
0f65dd70
AK		 4699static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4700 unsigned long addr,
bbd9b64e
CO		 4701 const void *old,
4702 const void *new,
4703 unsigned int bytes,
0f65dd70 4704 struct x86_exception *exception)
bbd9b64e 4705{
0f65dd70 4706 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
daea3e73
AK		 4707 gpa_t gpa;
4708 struct page *page;
4709 char *kaddr;
4710 bool exchanged;
2bacc55c 4711
daea3e73
AK		 4712 /* guests cmpxchg8b have to be emulated atomically */
4713 if (bytes > 8 || (bytes & (bytes - 1)))
4714 goto emul_write;
10589a46 4715
daea3e73 4716 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
2bacc55c 4717
daea3e73
AK		 4718 if (gpa == UNMAPPED_GVA ||
4719 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4720 goto emul_write;
2bacc55c 4721
daea3e73
AK		 4722 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4723 goto emul_write;
72dc67a6 4724
54bf36aa 4725 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
32cad84f 4726 if (is_error_page(page))
c19b8bd6 4727 goto emul_write;
72dc67a6 4728
8fd75e12 4729 kaddr = kmap_atomic(page);
daea3e73
AK		 4730 kaddr += offset_in_page(gpa);
4731 switch (bytes) {
4732 case 1:
4733 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4734 break;
4735 case 2:
4736 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4737 break;
4738 case 4:
4739 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4740 break;
4741 case 8:
4742 exchanged = CMPXCHG64(kaddr, old, new);
4743 break;
4744 default:
4745 BUG();
2bacc55c 4746 }
8fd75e12 4747 kunmap_atomic(kaddr);
daea3e73
AK		 4748 kvm_release_page_dirty(page);
4749
4750 if (!exchanged)
4751 return X86EMUL_CMPXCHG_FAILED;
4752
54bf36aa 4753 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
0eb05bf2 4754 kvm_page_track_write(vcpu, gpa, new, bytes);
8f6abd06
GN		 4755
4756 return X86EMUL_CONTINUE;
4a5f48f6 4757
3200f405 4758emul_write:
daea3e73 4759 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
2bacc55c 4760
0f65dd70 4761 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
bbd9b64e
CO		 4762}
4763
cf8f70bf
GN		 4764static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4765{
4766 /* TODO: String I/O for in kernel device */
4767 int r;
4768
4769 if (vcpu->arch.pio.in)
e32edf4f 4770 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
cf8f70bf
GN		 4771 vcpu->arch.pio.size, pd);
4772 else
e32edf4f 4773 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
cf8f70bf
GN		 4774 vcpu->arch.pio.port, vcpu->arch.pio.size,
4775 pd);
4776 return r;
4777}
4778
6f6fbe98
XG		 4779static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4780 unsigned short port, void *val,
4781 unsigned int count, bool in)
cf8f70bf 4782{
cf8f70bf 4783 vcpu->arch.pio.port = port;
6f6fbe98 4784 vcpu->arch.pio.in = in;
7972995b 4785 vcpu->arch.pio.count = count;
cf8f70bf
GN		 4786 vcpu->arch.pio.size = size;
4787
4788 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
7972995b 4789 vcpu->arch.pio.count = 0;
cf8f70bf
GN		 4790 return 1;
4791 }
4792
4793 vcpu->run->exit_reason = KVM_EXIT_IO;
6f6fbe98 4794 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
cf8f70bf
GN		 4795 vcpu->run->io.size = size;
4796 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4797 vcpu->run->io.count = count;
4798 vcpu->run->io.port = port;
4799
4800 return 0;
4801}
4802
6f6fbe98
XG		 4803static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4804 int size, unsigned short port, void *val,
4805 unsigned int count)
cf8f70bf 4806{
ca1d4a9e 4807 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6f6fbe98 4808 int ret;
ca1d4a9e 4809
6f6fbe98
XG		 4810 if (vcpu->arch.pio.count)
4811 goto data_avail;
cf8f70bf 4812
6f6fbe98
XG		 4813 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4814 if (ret) {
4815data_avail:
4816 memcpy(val, vcpu->arch.pio_data, size * count);
1171903d 4817 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
7972995b 4818 vcpu->arch.pio.count = 0;
cf8f70bf
GN		 4819 return 1;
4820 }
4821
cf8f70bf
GN		 4822 return 0;
4823}
4824
6f6fbe98
XG		 4825static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4826 int size, unsigned short port,
4827 const void *val, unsigned int count)
4828{
4829 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4830
4831 memcpy(vcpu->arch.pio_data, val, size * count);
1171903d 4832 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6f6fbe98
XG		 4833 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4834}
4835
bbd9b64e
CO		 4836static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4837{
4838 return kvm_x86_ops->get_segment_base(vcpu, seg);
4839}
4840
3cb16fe7 4841static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
bbd9b64e 4842{
3cb16fe7 4843 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
bbd9b64e
CO		 4844}
4845
ae6a2375 4846static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
f5f48ee1
SY		 4847{
4848 if (!need_emulate_wbinvd(vcpu))
4849 return X86EMUL_CONTINUE;
4850
4851 if (kvm_x86_ops->has_wbinvd_exit()) {
2eec7343
JK		 4852 int cpu = get_cpu();
4853
4854 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
f5f48ee1
SY		 4855 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4856 wbinvd_ipi, NULL, 1);
2eec7343 4857 put_cpu();
f5f48ee1 4858 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
2eec7343
JK		 4859 } else
4860 wbinvd();
f5f48ee1
SY		 4861 return X86EMUL_CONTINUE;
4862}
5cb56059
JS		 4863
4864int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4865{
6affcbed
KH		 4866 kvm_emulate_wbinvd_noskip(vcpu);
4867 return kvm_skip_emulated_instruction(vcpu);
5cb56059 4868}
f5f48ee1
SY		 4869EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4870
5cb56059
JS		 4871
4872
bcaf5cc5
AK		 4873static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4874{
5cb56059 4875 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
bcaf5cc5
AK		 4876}
4877
52eb5a6d
XL		 4878static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4879 unsigned long *dest)
bbd9b64e 4880{
16f8a6f9 4881 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
bbd9b64e
CO		 4882}
4883
52eb5a6d
XL		 4884static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4885 unsigned long value)
bbd9b64e 4886{
338dbc97 4887
717746e3 4888 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
bbd9b64e
CO		 4889}
4890
52a46617 4891static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5fdbf976 4892{
52a46617 4893 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5fdbf976
MT		 4894}
4895
717746e3 4896static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
bbd9b64e 4897{
717746e3 4898 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
52a46617
GN		 4899 unsigned long value;
4900
4901 switch (cr) {
4902 case 0:
4903 value = kvm_read_cr0(vcpu);
4904 break;
4905 case 2:
4906 value = vcpu->arch.cr2;
4907 break;
4908 case 3:
9f8fe504 4909 value = kvm_read_cr3(vcpu);
52a46617
GN		 4910 break;
4911 case 4:
4912 value = kvm_read_cr4(vcpu);
4913 break;
4914 case 8:
4915 value = kvm_get_cr8(vcpu);
4916 break;
4917 default:
a737f256 4918 kvm_err("%s: unexpected cr %u\n", __func__, cr);
52a46617
GN		 4919 return 0;
4920 }
4921
4922 return value;
4923}
4924
717746e3 4925static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
52a46617 4926{
717746e3 4927 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
0f12244f
GN		 4928 int res = 0;
4929
52a46617
GN		 4930 switch (cr) {
4931 case 0:
49a9b07e 4932 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
52a46617
GN		 4933 break;
4934 case 2:
4935 vcpu->arch.cr2 = val;
4936 break;
4937 case 3:
2390218b 4938 res = kvm_set_cr3(vcpu, val);
52a46617
GN		 4939 break;
4940 case 4:
a83b29c6 4941 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
52a46617
GN		 4942 break;
4943 case 8:
eea1cff9 4944 res = kvm_set_cr8(vcpu, val);
52a46617
GN		 4945 break;
4946 default:
a737f256 4947 kvm_err("%s: unexpected cr %u\n", __func__, cr);
0f12244f 4948 res = -1;
52a46617 4949 }
0f12244f
GN		 4950
4951 return res;
52a46617
GN		 4952}
4953
717746e3 4954static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
9c537244 4955{
717746e3 4956 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
9c537244
GN		 4957}
4958
4bff1e86 4959static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
2dafc6c2 4960{
4bff1e86 4961 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
2dafc6c2
GN		 4962}
4963
4bff1e86 4964static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
160ce1f1 4965{
4bff1e86 4966 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
160ce1f1
MG		 4967}
4968
1ac9d0cf
AK		 4969static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4970{
4971 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4972}
4973
4974static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4975{
4976 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4977}
4978
4bff1e86
AK		 4979static unsigned long emulator_get_cached_segment_base(
4980 struct x86_emulate_ctxt *ctxt, int seg)
5951c442 4981{
4bff1e86 4982 return get_segment_base(emul_to_vcpu(ctxt), seg);
5951c442
GN		 4983}
4984
1aa36616
AK		 4985static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4986 struct desc_struct *desc, u32 *base3,
4987 int seg)
2dafc6c2
GN		 4988{
4989 struct kvm_segment var;
4990
4bff1e86 4991 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
1aa36616 4992 *selector = var.selector;
2dafc6c2 4993
378a8b09
GN		 4994 if (var.unusable) {
4995 memset(desc, 0, sizeof(*desc));
2dafc6c2 4996 return false;
378a8b09 4997 }
2dafc6c2
GN		 4998
4999 if (var.g)
5000 var.limit >>= 12;
5001 set_desc_limit(desc, var.limit);
5002 set_desc_base(desc, (unsigned long)var.base);
5601d05b
GN		 5003#ifdef CONFIG_X86_64
5004 if (base3)
5005 *base3 = var.base >> 32;
5006#endif
2dafc6c2
GN		 5007 desc->type = var.type;
5008 desc->s = var.s;
5009 desc->dpl = var.dpl;
5010 desc->p = var.present;
5011 desc->avl = var.avl;
5012 desc->l = var.l;
5013 desc->d = var.db;
5014 desc->g = var.g;
5015
5016 return true;
5017}
5018
1aa36616
AK		 5019static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5020 struct desc_struct *desc, u32 base3,
5021 int seg)
2dafc6c2 5022{
4bff1e86 5023 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
2dafc6c2
GN		 5024 struct kvm_segment var;
5025
1aa36616 5026 var.selector = selector;
2dafc6c2 5027 var.base = get_desc_base(desc);
5601d05b
GN		 5028#ifdef CONFIG_X86_64
5029 var.base |= ((u64)base3) << 32;
5030#endif
2dafc6c2
GN		 5031 var.limit = get_desc_limit(desc);
5032 if (desc->g)
5033 var.limit = (var.limit << 12) | 0xfff;
5034 var.type = desc->type;
2dafc6c2
GN		 5035 var.dpl = desc->dpl;
5036 var.db = desc->d;
5037 var.s = desc->s;
5038 var.l = desc->l;
5039 var.g = desc->g;
5040 var.avl = desc->avl;
5041 var.present = desc->p;
5042 var.unusable = !var.present;
5043 var.padding = 0;
5044
5045 kvm_set_segment(vcpu, &var, seg);
5046 return;
5047}
5048
717746e3
AK		 5049static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5050 u32 msr_index, u64 *pdata)
5051{
609e36d3
PB		 5052 struct msr_data msr;
5053 int r;
5054
5055 msr.index = msr_index;
5056 msr.host_initiated = false;
5057 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5058 if (r)
5059 return r;
5060
5061 *pdata = msr.data;
5062 return 0;
717746e3
AK		 5063}
5064
5065static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5066 u32 msr_index, u64 data)
5067{
8fe8ab46
WA		 5068 struct msr_data msr;
5069
5070 msr.data = data;
5071 msr.index = msr_index;
5072 msr.host_initiated = false;
5073 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
717746e3
AK		 5074}
5075
64d60670
PB		 5076static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5077{
5078 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5079
5080 return vcpu->arch.smbase;
5081}
5082
5083static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5084{
5085 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5086
5087 vcpu->arch.smbase = smbase;
5088}
5089
67f4d428
NA		 5090static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5091 u32 pmc)
5092{
c6702c9d 5093 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
67f4d428
NA		 5094}
5095
222d21aa
AK		 5096static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5097 u32 pmc, u64 *pdata)
5098{
c6702c9d 5099 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
222d21aa
AK		 5100}
5101
6c3287f7
AK		 5102static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5103{
5104 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5105}
5106
5037f6f3
AK		 5107static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
5108{
5109 preempt_disable();
5197b808 5110 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
5037f6f3
AK		 5111}
5112
5113static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
5114{
5115 preempt_enable();
5116}
5117
2953538e 5118static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
8a76d7f2 5119 struct x86_instruction_info *info,
c4f035c6
AK		 5120 enum x86_intercept_stage stage)
5121{
2953538e 5122 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
c4f035c6
AK		 5123}
5124
0017f93a 5125static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
bdb42f5a
SB		 5126 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
5127{
0017f93a 5128 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
bdb42f5a
SB		 5129}
5130
dd856efa
AK		 5131static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5132{
5133 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5134}
5135
5136static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5137{
5138 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5139}
5140
801806d9
NA		 5141static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5142{
5143 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5144}
5145
0225fb50 5146static const struct x86_emulate_ops emulate_ops = {
dd856efa
AK		 5147 .read_gpr = emulator_read_gpr,
5148 .write_gpr = emulator_write_gpr,
1871c602 5149 .read_std = kvm_read_guest_virt_system,
2dafc6c2 5150 .write_std = kvm_write_guest_virt_system,
7a036a6f 5151 .read_phys = kvm_read_guest_phys_system,
1871c602 5152 .fetch = kvm_fetch_guest_virt,
bbd9b64e
CO		 5153 .read_emulated = emulator_read_emulated,
5154 .write_emulated = emulator_write_emulated,
5155 .cmpxchg_emulated = emulator_cmpxchg_emulated,
3cb16fe7 5156 .invlpg = emulator_invlpg,
cf8f70bf
GN		 5157 .pio_in_emulated = emulator_pio_in_emulated,
5158 .pio_out_emulated = emulator_pio_out_emulated,
1aa36616
AK		 5159 .get_segment = emulator_get_segment,
5160 .set_segment = emulator_set_segment,
5951c442 5161 .get_cached_segment_base = emulator_get_cached_segment_base,
2dafc6c2 5162 .get_gdt = emulator_get_gdt,
160ce1f1 5163 .get_idt = emulator_get_idt,
1ac9d0cf
AK		 5164 .set_gdt = emulator_set_gdt,
5165 .set_idt = emulator_set_idt,
52a46617
GN		 5166 .get_cr = emulator_get_cr,
5167 .set_cr = emulator_set_cr,
9c537244 5168 .cpl = emulator_get_cpl,
35aa5375
GN		 5169 .get_dr = emulator_get_dr,
5170 .set_dr = emulator_set_dr,
64d60670
PB		 5171 .get_smbase = emulator_get_smbase,
5172 .set_smbase = emulator_set_smbase,
717746e3
AK		 5173 .set_msr = emulator_set_msr,
5174 .get_msr = emulator_get_msr,
67f4d428 5175 .check_pmc = emulator_check_pmc,
222d21aa 5176 .read_pmc = emulator_read_pmc,
6c3287f7 5177 .halt = emulator_halt,
bcaf5cc5 5178 .wbinvd = emulator_wbinvd,
d6aa1000 5179 .fix_hypercall = emulator_fix_hypercall,
5037f6f3
AK		 5180 .get_fpu = emulator_get_fpu,
5181 .put_fpu = emulator_put_fpu,
c4f035c6 5182 .intercept = emulator_intercept,
bdb42f5a 5183 .get_cpuid = emulator_get_cpuid,
801806d9 5184 .set_nmi_mask = emulator_set_nmi_mask,
bbd9b64e
CO		 5185};
5186
95cb2295
GN		 5187static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5188{
37ccdcbe 5189 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
95cb2295
GN		 5190 /*
5191 * an sti; sti; sequence only disable interrupts for the first
5192 * instruction. So, if the last instruction, be it emulated or
5193 * not, left the system with the INT_STI flag enabled, it
5194 * means that the last instruction is an sti. We should not
5195 * leave the flag on in this case. The same goes for mov ss
5196 */
37ccdcbe
PB		 5197 if (int_shadow & mask)
5198 mask = 0;
6addfc42 5199 if (unlikely(int_shadow || mask)) {
95cb2295 5200 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6addfc42
PB		 5201 if (!mask)
5202 kvm_make_request(KVM_REQ_EVENT, vcpu);
5203 }
95cb2295
GN		 5204}
5205
ef54bcfe 5206static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
54b8486f
GN		 5207{
5208 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
da9cb575 5209 if (ctxt->exception.vector == PF_VECTOR)
ef54bcfe
PB		 5210 return kvm_propagate_fault(vcpu, &ctxt->exception);
5211
5212 if (ctxt->exception.error_code_valid)
da9cb575
AK		 5213 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5214 ctxt->exception.error_code);
54b8486f 5215 else
da9cb575 5216 kvm_queue_exception(vcpu, ctxt->exception.vector);
ef54bcfe 5217 return false;
54b8486f
GN		 5218}
5219
8ec4722d
MG		 5220static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5221{
adf52235 5222 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8ec4722d
MG		 5223 int cs_db, cs_l;
5224
8ec4722d
MG		 5225 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5226
adf52235
TY		 5227 ctxt->eflags = kvm_get_rflags(vcpu);
5228 ctxt->eip = kvm_rip_read(vcpu);
5229 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5230 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
42bf549f 5231 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
adf52235
TY		 5232 cs_db ? X86EMUL_MODE_PROT32 :
5233 X86EMUL_MODE_PROT16;
a584539b 5234 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
64d60670
PB		 5235 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5236 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
a584539b 5237 ctxt->emul_flags = vcpu->arch.hflags;
adf52235 5238
dd856efa 5239 init_decode_cache(ctxt);
7ae441ea 5240 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8ec4722d
MG		 5241}
5242
71f9833b 5243int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
63995653 5244{
9d74191a 5245 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
63995653
MG		 5246 int ret;
5247
5248 init_emulate_ctxt(vcpu);
5249
9dac77fa
AK		 5250 ctxt->op_bytes = 2;
5251 ctxt->ad_bytes = 2;
5252 ctxt->_eip = ctxt->eip + inc_eip;
9d74191a 5253 ret = emulate_int_real(ctxt, irq);
63995653
MG		 5254
5255 if (ret != X86EMUL_CONTINUE)
5256 return EMULATE_FAIL;
5257
9dac77fa 5258 ctxt->eip = ctxt->_eip;
9d74191a
TY		 5259 kvm_rip_write(vcpu, ctxt->eip);
5260 kvm_set_rflags(vcpu, ctxt->eflags);
63995653
MG		 5261
5262 if (irq == NMI_VECTOR)
7460fb4a 5263 vcpu->arch.nmi_pending = 0;
63995653
MG		 5264 else
5265 vcpu->arch.interrupt.pending = false;
5266
5267 return EMULATE_DONE;
5268}
5269EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5270
6d77dbfc
GN		 5271static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5272{
fc3a9157
JR		 5273 int r = EMULATE_DONE;
5274
6d77dbfc
GN		 5275 ++vcpu->stat.insn_emulation_fail;
5276 trace_kvm_emulate_insn_failed(vcpu);
a2b9e6c1 5277 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
fc3a9157
JR		 5278 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5279 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5280 vcpu->run->internal.ndata = 0;
5281 r = EMULATE_FAIL;
5282 }
6d77dbfc 5283 kvm_queue_exception(vcpu, UD_VECTOR);
fc3a9157
JR		 5284
5285 return r;
6d77dbfc
GN		 5286}
5287
93c05d3e 5288static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
991eebf9
GN		 5289 bool write_fault_to_shadow_pgtable,
5290 int emulation_type)
a6f177ef 5291{
95b3cf69 5292 gpa_t gpa = cr2;
ba049e93 5293 kvm_pfn_t pfn;
a6f177ef 5294
991eebf9
GN		 5295 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5296 return false;
5297
95b3cf69
XG		 5298 if (!vcpu->arch.mmu.direct_map) {
5299 /*
5300 * Write permission should be allowed since only
5301 * write access need to be emulated.
5302 */
5303 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
a6f177ef 5304
95b3cf69
XG		 5305 /*
5306 * If the mapping is invalid in guest, let cpu retry
5307 * it to generate fault.
5308 */
5309 if (gpa == UNMAPPED_GVA)
5310 return true;
5311 }
a6f177ef 5312
8e3d9d06
XG		 5313 /*
5314 * Do not retry the unhandleable instruction if it faults on the
5315 * readonly host memory, otherwise it will goto a infinite loop:
5316 * retry instruction -> write #PF -> emulation fail -> retry
5317 * instruction -> ...
5318 */
5319 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
95b3cf69
XG		 5320
5321 /*
5322 * If the instruction failed on the error pfn, it can not be fixed,
5323 * report the error to userspace.
5324 */
5325 if (is_error_noslot_pfn(pfn))
5326 return false;
5327
5328 kvm_release_pfn_clean(pfn);
5329
5330 /* The instructions are well-emulated on direct mmu. */
5331 if (vcpu->arch.mmu.direct_map) {
5332 unsigned int indirect_shadow_pages;
5333
5334 spin_lock(&vcpu->kvm->mmu_lock);
5335 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5336 spin_unlock(&vcpu->kvm->mmu_lock);
5337
5338 if (indirect_shadow_pages)
5339 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5340
a6f177ef 5341 return true;
8e3d9d06 5342 }
a6f177ef 5343
95b3cf69
XG		 5344 /*
5345 * if emulation was due to access to shadowed page table
5346 * and it failed try to unshadow page and re-enter the
5347 * guest to let CPU execute the instruction.
5348 */
5349 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
93c05d3e
XG		 5350
5351 /*
5352 * If the access faults on its page table, it can not
5353 * be fixed by unprotecting shadow page and it should
5354 * be reported to userspace.
5355 */
5356 return !write_fault_to_shadow_pgtable;
a6f177ef
GN		 5357}
5358
1cb3f3ae
XG		 5359static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5360 unsigned long cr2, int emulation_type)
5361{
5362 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5363 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5364
5365 last_retry_eip = vcpu->arch.last_retry_eip;
5366 last_retry_addr = vcpu->arch.last_retry_addr;
5367
5368 /*
5369 * If the emulation is caused by #PF and it is non-page_table
5370 * writing instruction, it means the VM-EXIT is caused by shadow
5371 * page protected, we can zap the shadow page and retry this
5372 * instruction directly.
5373 *
5374 * Note: if the guest uses a non-page-table modifying instruction
5375 * on the PDE that points to the instruction, then we will unmap
5376 * the instruction and go to an infinite loop. So, we cache the
5377 * last retried eip and the last fault address, if we meet the eip
5378 * and the address again, we can break out of the potential infinite
5379 * loop.
5380 */
5381 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5382
5383 if (!(emulation_type & EMULTYPE_RETRY))
5384 return false;
5385
5386 if (x86_page_table_writing_insn(ctxt))
5387 return false;
5388
5389 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5390 return false;
5391
5392 vcpu->arch.last_retry_eip = ctxt->eip;
5393 vcpu->arch.last_retry_addr = cr2;
5394
5395 if (!vcpu->arch.mmu.direct_map)
5396 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5397
22368028 5398 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
1cb3f3ae
XG		 5399
5400 return true;
5401}
5402
716d51ab
GN		 5403static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5404static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5405
64d60670 5406static void kvm_smm_changed(struct kvm_vcpu *vcpu)
a584539b 5407{
64d60670 5408 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
660a5d51
PB		 5409 /* This is a good place to trace that we are exiting SMM. */
5410 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5411
c43203ca
PB		 5412 /* Process a latched INIT or SMI, if any. */
5413 kvm_make_request(KVM_REQ_EVENT, vcpu);
64d60670 5414 }
699023e2
PB		 5415
5416 kvm_mmu_reset_context(vcpu);
64d60670
PB		 5417}
5418
5419static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5420{
5421 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5422
a584539b 5423 vcpu->arch.hflags = emul_flags;
64d60670
PB		 5424
5425 if (changed & HF_SMM_MASK)
5426 kvm_smm_changed(vcpu);
a584539b
PB		 5427}
5428
4a1e10d5
PB		 5429static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5430 unsigned long *db)
5431{
5432 u32 dr6 = 0;
5433 int i;
5434 u32 enable, rwlen;
5435
5436 enable = dr7;
5437 rwlen = dr7 >> 16;
5438 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5439 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5440 dr6 |= (1 << i);
5441 return dr6;
5442}
5443
6addfc42 5444static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
663f4c61
PB		 5445{
5446 struct kvm_run *kvm_run = vcpu->run;
5447
5448 /*
6addfc42
PB		 5449 * rflags is the old, "raw" value of the flags. The new value has
5450 * not been saved yet.
663f4c61
PB		 5451 *
5452 * This is correct even for TF set by the guest, because "the
5453 * processor will not generate this exception after the instruction
5454 * that sets the TF flag".
5455 */
663f4c61
PB		 5456 if (unlikely(rflags & X86_EFLAGS_TF)) {
5457 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6f43ed01
NA		 5458 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5459 DR6_RTM;
663f4c61
PB		 5460 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5461 kvm_run->debug.arch.exception = DB_VECTOR;
5462 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5463 *r = EMULATE_USER_EXIT;
5464 } else {
663f4c61
PB		 5465 /*
5466 * "Certain debug exceptions may clear bit 0-3. The
5467 * remaining contents of the DR6 register are never
5468 * cleared by the processor".
5469 */
5470 vcpu->arch.dr6 &= ~15;
6f43ed01 5471 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
663f4c61
PB		 5472 kvm_queue_exception(vcpu, DB_VECTOR);
5473 }
5474 }
5475}
5476
6affcbed
KH		 5477int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
5478{
5479 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5480 int r = EMULATE_DONE;
5481
5482 kvm_x86_ops->skip_emulated_instruction(vcpu);
5483 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5484 return r == EMULATE_DONE;
5485}
5486EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
5487
4a1e10d5
PB		 5488static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5489{
4a1e10d5
PB		 5490 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5491 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
82b32774
NA		 5492 struct kvm_run *kvm_run = vcpu->run;
5493 unsigned long eip = kvm_get_linear_rip(vcpu);
5494 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
4a1e10d5
PB		 5495 vcpu->arch.guest_debug_dr7,
5496 vcpu->arch.eff_db);
5497
5498 if (dr6 != 0) {
6f43ed01 5499 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
82b32774 5500 kvm_run->debug.arch.pc = eip;
4a1e10d5
PB		 5501 kvm_run->debug.arch.exception = DB_VECTOR;
5502 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5503 *r = EMULATE_USER_EXIT;
5504 return true;
5505 }
5506 }
5507
4161a569
NA		 5508 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5509 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
82b32774
NA		 5510 unsigned long eip = kvm_get_linear_rip(vcpu);
5511 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
4a1e10d5
PB		 5512 vcpu->arch.dr7,
5513 vcpu->arch.db);
5514
5515 if (dr6 != 0) {
5516 vcpu->arch.dr6 &= ~15;
6f43ed01 5517 vcpu->arch.dr6 |= dr6 | DR6_RTM;
4a1e10d5
PB		 5518 kvm_queue_exception(vcpu, DB_VECTOR);
5519 *r = EMULATE_DONE;
5520 return true;
5521 }
5522 }
5523
5524 return false;
5525}
5526
51d8b661
AP		 5527int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5528 unsigned long cr2,
dc25e89e
AP		 5529 int emulation_type,
5530 void *insn,
5531 int insn_len)
bbd9b64e 5532{
95cb2295 5533 int r;
9d74191a 5534 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7ae441ea 5535 bool writeback = true;
93c05d3e 5536 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
bbd9b64e 5537
93c05d3e
XG		 5538 /*
5539 * Clear write_fault_to_shadow_pgtable here to ensure it is
5540 * never reused.
5541 */
5542 vcpu->arch.write_fault_to_shadow_pgtable = false;
26eef70c 5543 kvm_clear_exception_queue(vcpu);
8d7d8102 5544
571008da 5545 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
8ec4722d 5546 init_emulate_ctxt(vcpu);
4a1e10d5
PB		 5547
5548 /*
5549 * We will reenter on the same instruction since
5550 * we do not set complete_userspace_io. This does not
5551 * handle watchpoints yet, those would be handled in
5552 * the emulate_ops.
5553 */
5554 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5555 return r;
5556
9d74191a
TY		 5557 ctxt->interruptibility = 0;
5558 ctxt->have_exception = false;
e0ad0b47 5559 ctxt->exception.vector = -1;
9d74191a 5560 ctxt->perm_ok = false;
bbd9b64e 5561
b51e974f 5562 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
4005996e 5563
9d74191a 5564 r = x86_decode_insn(ctxt, insn, insn_len);
bbd9b64e 5565
e46479f8 5566 trace_kvm_emulate_insn_start(vcpu);
f2b5756b 5567 ++vcpu->stat.insn_emulation;
1d2887e2 5568 if (r != EMULATION_OK) {
4005996e
AK		 5569 if (emulation_type & EMULTYPE_TRAP_UD)
5570 return EMULATE_FAIL;
991eebf9
GN		 5571 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5572 emulation_type))
bbd9b64e 5573 return EMULATE_DONE;
6d77dbfc
GN		 5574 if (emulation_type & EMULTYPE_SKIP)
5575 return EMULATE_FAIL;
5576 return handle_emulation_failure(vcpu);
bbd9b64e
CO		 5577 }
5578 }
5579
ba8afb6b 5580 if (emulation_type & EMULTYPE_SKIP) {
9dac77fa 5581 kvm_rip_write(vcpu, ctxt->_eip);
bb663c7a
NA		 5582 if (ctxt->eflags & X86_EFLAGS_RF)
5583 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
ba8afb6b
GN		 5584 return EMULATE_DONE;
5585 }
5586
1cb3f3ae
XG		 5587 if (retry_instruction(ctxt, cr2, emulation_type))
5588 return EMULATE_DONE;
5589
7ae441ea 5590 /* this is needed for vmware backdoor interface to work since it
4d2179e1 5591 changes registers values during IO operation */
7ae441ea
GN		 5592 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5593 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
dd856efa 5594 emulator_invalidate_register_cache(ctxt);
7ae441ea 5595 }
4d2179e1 5596
5cd21917 5597restart:
9d74191a 5598 r = x86_emulate_insn(ctxt);
bbd9b64e 5599
775fde86
JR		 5600 if (r == EMULATION_INTERCEPTED)
5601 return EMULATE_DONE;
5602
d2ddd1c4 5603 if (r == EMULATION_FAILED) {
991eebf9
GN		 5604 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5605 emulation_type))
c3cd7ffa
GN		 5606 return EMULATE_DONE;
5607
6d77dbfc 5608 return handle_emulation_failure(vcpu);
bbd9b64e
CO		 5609 }
5610
9d74191a 5611 if (ctxt->have_exception) {
d2ddd1c4 5612 r = EMULATE_DONE;
ef54bcfe
PB		 5613 if (inject_emulated_exception(vcpu))
5614 return r;
d2ddd1c4 5615 } else if (vcpu->arch.pio.count) {
0912c977
PB		 5616 if (!vcpu->arch.pio.in) {
5617 /* FIXME: return into emulator if single-stepping. */
3457e419 5618 vcpu->arch.pio.count = 0;
0912c977 5619 } else {
7ae441ea 5620 writeback = false;
716d51ab
GN		 5621 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5622 }
ac0a48c3 5623 r = EMULATE_USER_EXIT;
7ae441ea
GN		 5624 } else if (vcpu->mmio_needed) {
5625 if (!vcpu->mmio_is_write)
5626 writeback = false;
ac0a48c3 5627 r = EMULATE_USER_EXIT;
716d51ab 5628 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7ae441ea 5629 } else if (r == EMULATION_RESTART)
5cd21917 5630 goto restart;
d2ddd1c4
GN		 5631 else
5632 r = EMULATE_DONE;
f850e2e6 5633
7ae441ea 5634 if (writeback) {
6addfc42 5635 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
9d74191a 5636 toggle_interruptibility(vcpu, ctxt->interruptibility);
7ae441ea 5637 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
a584539b
PB		 5638 if (vcpu->arch.hflags != ctxt->emul_flags)
5639 kvm_set_hflags(vcpu, ctxt->emul_flags);
9d74191a 5640 kvm_rip_write(vcpu, ctxt->eip);
663f4c61 5641 if (r == EMULATE_DONE)
6addfc42 5642 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
38827dbd
NA		 5643 if (!ctxt->have_exception ||
5644 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5645 __kvm_set_rflags(vcpu, ctxt->eflags);
6addfc42
PB		 5646
5647 /*
5648 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5649 * do nothing, and it will be requested again as soon as
5650 * the shadow expires. But we still need to check here,
5651 * because POPF has no interrupt shadow.
5652 */
5653 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5654 kvm_make_request(KVM_REQ_EVENT, vcpu);
7ae441ea
GN		 5655 } else
5656 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
e85d28f8
GN		 5657
5658 return r;
de7d789a 5659}
51d8b661 5660EXPORT_SYMBOL_GPL(x86_emulate_instruction);
de7d789a 5661
cf8f70bf 5662int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
de7d789a 5663{
cf8f70bf 5664 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
ca1d4a9e
AK		 5665 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5666 size, port, &val, 1);
cf8f70bf 5667 /* do not return to emulator after return from userspace */
7972995b 5668 vcpu->arch.pio.count = 0;
de7d789a
CO		 5669 return ret;
5670}
cf8f70bf 5671EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
de7d789a 5672
8370c3d0
TL		 5673static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
5674{
5675 unsigned long val;
5676
5677 /* We should only ever be called with arch.pio.count equal to 1 */
5678 BUG_ON(vcpu->arch.pio.count != 1);
5679
5680 /* For size less than 4 we merge, else we zero extend */
5681 val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
5682 : 0;
5683
5684 /*
5685 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
5686 * the copy and tracing
5687 */
5688 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
5689 vcpu->arch.pio.port, &val, 1);
5690 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5691
5692 return 1;
5693}
5694
5695int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size, unsigned short port)
5696{
5697 unsigned long val;
5698 int ret;
5699
5700 /* For size less than 4 we merge, else we zero extend */
5701 val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
5702
5703 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
5704 &val, 1);
5705 if (ret) {
5706 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5707 return ret;
5708 }
5709
5710 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
5711
5712 return 0;
5713}
5714EXPORT_SYMBOL_GPL(kvm_fast_pio_in);
5715
251a5fd6 5716static int kvmclock_cpu_down_prep(unsigned int cpu)
8cfdc000 5717{
0a3aee0d 5718 __this_cpu_write(cpu_tsc_khz, 0);
251a5fd6 5719 return 0;
8cfdc000
ZA		 5720}
5721
5722static void tsc_khz_changed(void *data)
c8076604 5723{
8cfdc000
ZA		 5724 struct cpufreq_freqs *freq = data;
5725 unsigned long khz = 0;
5726
5727 if (data)
5728 khz = freq->new;
5729 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5730 khz = cpufreq_quick_get(raw_smp_processor_id());
5731 if (!khz)
5732 khz = tsc_khz;
0a3aee0d 5733 __this_cpu_write(cpu_tsc_khz, khz);
c8076604
GH		 5734}
5735
c8076604
GH		 5736static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5737 void *data)
5738{
5739 struct cpufreq_freqs *freq = data;
5740 struct kvm *kvm;
5741 struct kvm_vcpu *vcpu;
5742 int i, send_ipi = 0;
5743
8cfdc000
ZA		 5744 /*
5745 * We allow guests to temporarily run on slowing clocks,
5746 * provided we notify them after, or to run on accelerating
5747 * clocks, provided we notify them before. Thus time never
5748 * goes backwards.
5749 *
5750 * However, we have a problem. We can't atomically update
5751 * the frequency of a given CPU from this function; it is
5752 * merely a notifier, which can be called from any CPU.
5753 * Changing the TSC frequency at arbitrary points in time
5754 * requires a recomputation of local variables related to
5755 * the TSC for each VCPU. We must flag these local variables
5756 * to be updated and be sure the update takes place with the
5757 * new frequency before any guests proceed.
5758 *
5759 * Unfortunately, the combination of hotplug CPU and frequency
5760 * change creates an intractable locking scenario; the order
5761 * of when these callouts happen is undefined with respect to
5762 * CPU hotplug, and they can race with each other. As such,
5763 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5764 * undefined; you can actually have a CPU frequency change take
5765 * place in between the computation of X and the setting of the
5766 * variable. To protect against this problem, all updates of
5767 * the per_cpu tsc_khz variable are done in an interrupt
5768 * protected IPI, and all callers wishing to update the value
5769 * must wait for a synchronous IPI to complete (which is trivial
5770 * if the caller is on the CPU already). This establishes the
5771 * necessary total order on variable updates.
5772 *
5773 * Note that because a guest time update may take place
5774 * anytime after the setting of the VCPU's request bit, the
5775 * correct TSC value must be set before the request. However,
5776 * to ensure the update actually makes it to any guest which
5777 * starts running in hardware virtualization between the set
5778 * and the acquisition of the spinlock, we must also ping the
5779 * CPU after setting the request bit.
5780 *
5781 */
5782
c8076604
GH		 5783 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5784 return 0;
5785 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5786 return 0;
8cfdc000
ZA		 5787
5788 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
c8076604 5789
2f303b74 5790 spin_lock(&kvm_lock);
c8076604 5791 list_for_each_entry(kvm, &vm_list, vm_list) {
988a2cae 5792 kvm_for_each_vcpu(i, vcpu, kvm) {
c8076604
GH		 5793 if (vcpu->cpu != freq->cpu)
5794 continue;
c285545f 5795 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
c8076604 5796 if (vcpu->cpu != smp_processor_id())
8cfdc000 5797 send_ipi = 1;
c8076604
GH		 5798 }
5799 }
2f303b74 5800 spin_unlock(&kvm_lock);
c8076604
GH		 5801
5802 if (freq->old < freq->new && send_ipi) {
5803 /*
5804 * We upscale the frequency. Must make the guest
5805 * doesn't see old kvmclock values while running with
5806 * the new frequency, otherwise we risk the guest sees
5807 * time go backwards.
5808 *
5809 * In case we update the frequency for another cpu
5810 * (which might be in guest context) send an interrupt
5811 * to kick the cpu out of guest context. Next time
5812 * guest context is entered kvmclock will be updated,
5813 * so the guest will not see stale values.
5814 */
8cfdc000 5815 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
c8076604
GH		 5816 }
5817 return 0;
5818}
5819
5820static struct notifier_block kvmclock_cpufreq_notifier_block = {
8cfdc000
ZA		 5821 .notifier_call = kvmclock_cpufreq_notifier
5822};
5823
251a5fd6 5824static int kvmclock_cpu_online(unsigned int cpu)
8cfdc000 5825{
251a5fd6
SAS		 5826 tsc_khz_changed(NULL);
5827 return 0;
8cfdc000
ZA		 5828}
5829
b820cc0c
ZA		 5830static void kvm_timer_init(void)
5831{
c285545f 5832 max_tsc_khz = tsc_khz;
460dd42e 5833
b820cc0c 5834 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
c285545f
ZA		 5835#ifdef CONFIG_CPU_FREQ
5836 struct cpufreq_policy policy;
758f588d
BP		 5837 int cpu;
5838
c285545f 5839 memset(&policy, 0, sizeof(policy));
3e26f230
AK		 5840 cpu = get_cpu();
5841 cpufreq_get_policy(&policy, cpu);
c285545f
ZA		 5842 if (policy.cpuinfo.max_freq)
5843 max_tsc_khz = policy.cpuinfo.max_freq;
3e26f230 5844 put_cpu();
c285545f 5845#endif
b820cc0c
ZA		 5846 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5847 CPUFREQ_TRANSITION_NOTIFIER);
5848 }
c285545f 5849 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
460dd42e 5850
251a5fd6
SAS		 5851 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "AP_X86_KVM_CLK_ONLINE",
5852 kvmclock_cpu_online, kvmclock_cpu_down_prep);
b820cc0c
ZA		 5853}
5854
ff9d07a0
ZY		 5855static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5856
f5132b01 5857int kvm_is_in_guest(void)
ff9d07a0 5858{
086c9855 5859 return __this_cpu_read(current_vcpu) != NULL;
ff9d07a0
ZY		 5860}
5861
5862static int kvm_is_user_mode(void)
5863{
5864 int user_mode = 3;
dcf46b94 5865
086c9855
AS		 5866 if (__this_cpu_read(current_vcpu))
5867 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
dcf46b94 5868
ff9d07a0
ZY		 5869 return user_mode != 0;
5870}
5871
5872static unsigned long kvm_get_guest_ip(void)
5873{
5874 unsigned long ip = 0;
dcf46b94 5875
086c9855
AS		 5876 if (__this_cpu_read(current_vcpu))
5877 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
dcf46b94 5878
ff9d07a0
ZY		 5879 return ip;
5880}
5881
5882static struct perf_guest_info_callbacks kvm_guest_cbs = {
5883 .is_in_guest = kvm_is_in_guest,
5884 .is_user_mode = kvm_is_user_mode,
5885 .get_guest_ip = kvm_get_guest_ip,
5886};
5887
5888void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5889{
086c9855 5890 __this_cpu_write(current_vcpu, vcpu);
ff9d07a0
ZY		 5891}
5892EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5893
5894void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5895{
086c9855 5896 __this_cpu_write(current_vcpu, NULL);
ff9d07a0
ZY		 5897}
5898EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5899
ce88decf
XG		 5900static void kvm_set_mmio_spte_mask(void)
5901{
5902 u64 mask;
5903 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5904
5905 /*
5906 * Set the reserved bits and the present bit of an paging-structure
5907 * entry to generate page fault with PFER.RSV = 1.
5908 */
885032b9 5909 /* Mask the reserved physical address bits. */
d1431483 5910 mask = rsvd_bits(maxphyaddr, 51);
885032b9
XG		 5911
5912 /* Bit 62 is always reserved for 32bit host. */
5913 mask |= 0x3ull << 62;
5914
5915 /* Set the present bit. */
ce88decf
XG		 5916 mask |= 1ull;
5917
5918#ifdef CONFIG_X86_64
5919 /*
5920 * If reserved bit is not supported, clear the present bit to disable
5921 * mmio page fault.
5922 */
5923 if (maxphyaddr == 52)
5924 mask &= ~1ull;
5925#endif
5926
5927 kvm_mmu_set_mmio_spte_mask(mask);
5928}
5929
16e8d74d
MT		 5930#ifdef CONFIG_X86_64
5931static void pvclock_gtod_update_fn(struct work_struct *work)
5932{
d828199e
MT		 5933 struct kvm *kvm;
5934
5935 struct kvm_vcpu *vcpu;
5936 int i;
5937
2f303b74 5938 spin_lock(&kvm_lock);
d828199e
MT		 5939 list_for_each_entry(kvm, &vm_list, vm_list)
5940 kvm_for_each_vcpu(i, vcpu, kvm)
105b21bb 5941 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
d828199e 5942 atomic_set(&kvm_guest_has_master_clock, 0);
2f303b74 5943 spin_unlock(&kvm_lock);
16e8d74d
MT		 5944}
5945
5946static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5947
5948/*
5949 * Notification about pvclock gtod data update.
5950 */
5951static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5952 void *priv)
5953{
5954 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5955 struct timekeeper *tk = priv;
5956
5957 update_pvclock_gtod(tk);
5958
5959 /* disable master clock if host does not trust, or does not
5960 * use, TSC clocksource
5961 */
5962 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5963 atomic_read(&kvm_guest_has_master_clock) != 0)
5964 queue_work(system_long_wq, &pvclock_gtod_work);
5965
5966 return 0;
5967}
5968
5969static struct notifier_block pvclock_gtod_notifier = {
5970 .notifier_call = pvclock_gtod_notify,
5971};
5972#endif
5973
f8c16bba 5974int kvm_arch_init(void *opaque)
043405e1 5975{
b820cc0c 5976 int r;
6b61edf7 5977 struct kvm_x86_ops *ops = opaque;
f8c16bba 5978
f8c16bba
ZX		 5979 if (kvm_x86_ops) {
5980 printk(KERN_ERR "kvm: already loaded the other module\n");
56c6d28a
ZX		 5981 r = -EEXIST;
5982 goto out;
f8c16bba
ZX		 5983 }
5984
5985 if (!ops->cpu_has_kvm_support()) {
5986 printk(KERN_ERR "kvm: no hardware support\n");
56c6d28a
ZX		 5987 r = -EOPNOTSUPP;
5988 goto out;
f8c16bba
ZX		 5989 }
5990 if (ops->disabled_by_bios()) {
5991 printk(KERN_ERR "kvm: disabled by bios\n");
56c6d28a
ZX		 5992 r = -EOPNOTSUPP;
5993 goto out;
f8c16bba
ZX		 5994 }
5995
013f6a5d
MT		 5996 r = -ENOMEM;
5997 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5998 if (!shared_msrs) {
5999 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6000 goto out;
6001 }
6002
97db56ce
AK		 6003 r = kvm_mmu_module_init();
6004 if (r)
013f6a5d 6005 goto out_free_percpu;
97db56ce 6006
ce88decf 6007 kvm_set_mmio_spte_mask();
97db56ce 6008
f8c16bba 6009 kvm_x86_ops = ops;
920c8377 6010
7b52345e 6011 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
ffb128c8
BD		 6012 PT_DIRTY_MASK, PT64_NX_MASK, 0,
6013 PT_PRESENT_MASK);
b820cc0c 6014 kvm_timer_init();
c8076604 6015
ff9d07a0
ZY		 6016 perf_register_guest_info_callbacks(&kvm_guest_cbs);
6017
d366bf7e 6018 if (boot_cpu_has(X86_FEATURE_XSAVE))
2acf923e
DC		 6019 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6020
c5cc421b 6021 kvm_lapic_init();
16e8d74d
MT		 6022#ifdef CONFIG_X86_64
6023 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6024#endif
6025
f8c16bba 6026 return 0;
56c6d28a 6027
013f6a5d
MT		 6028out_free_percpu:
6029 free_percpu(shared_msrs);
56c6d28a 6030out:
56c6d28a 6031 return r;
043405e1 6032}
8776e519 6033
f8c16bba
ZX		 6034void kvm_arch_exit(void)
6035{
ff9d07a0
ZY		 6036 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6037
888d256e
JK		 6038 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6039 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6040 CPUFREQ_TRANSITION_NOTIFIER);
251a5fd6 6041 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
16e8d74d
MT		 6042#ifdef CONFIG_X86_64
6043 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6044#endif
f8c16bba 6045 kvm_x86_ops = NULL;
56c6d28a 6046 kvm_mmu_module_exit();
013f6a5d 6047 free_percpu(shared_msrs);
56c6d28a 6048}
f8c16bba 6049
5cb56059 6050int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8776e519
HB		 6051{
6052 ++vcpu->stat.halt_exits;
35754c98 6053 if (lapic_in_kernel(vcpu)) {
a4535290 6054 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
8776e519
HB		 6055 return 1;
6056 } else {
6057 vcpu->run->exit_reason = KVM_EXIT_HLT;
6058 return 0;
6059 }
6060}
5cb56059
JS		 6061EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6062
6063int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6064{
6affcbed
KH		 6065 int ret = kvm_skip_emulated_instruction(vcpu);
6066 /*
6067 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6068 * KVM_EXIT_DEBUG here.
6069 */
6070 return kvm_vcpu_halt(vcpu) && ret;
5cb56059 6071}
8776e519
HB		 6072EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6073
6aef266c
SV		 6074/*
6075 * kvm_pv_kick_cpu_op: Kick a vcpu.
6076 *
6077 * @apicid - apicid of vcpu to be kicked.
6078 */
6079static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6080{
24d2166b 6081 struct kvm_lapic_irq lapic_irq;
6aef266c 6082
24d2166b
R		 6083 lapic_irq.shorthand = 0;
6084 lapic_irq.dest_mode = 0;
6085 lapic_irq.dest_id = apicid;
93bbf0b8 6086 lapic_irq.msi_redir_hint = false;
6aef266c 6087
24d2166b 6088 lapic_irq.delivery_mode = APIC_DM_REMRD;
795a149e 6089 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6aef266c
SV		 6090}
6091
d62caabb
AS		 6092void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6093{
6094 vcpu->arch.apicv_active = false;
6095 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6096}
6097
8776e519
HB		 6098int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6099{
6100 unsigned long nr, a0, a1, a2, a3, ret;
6affcbed 6101 int op_64_bit, r;
8776e519 6102
6affcbed 6103 r = kvm_skip_emulated_instruction(vcpu);
5cb56059 6104
55cd8e5a
GN		 6105 if (kvm_hv_hypercall_enabled(vcpu->kvm))
6106 return kvm_hv_hypercall(vcpu);
6107
5fdbf976
MT		 6108 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6109 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6110 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6111 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6112 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
8776e519 6113
229456fc 6114 trace_kvm_hypercall(nr, a0, a1, a2, a3);
2714d1d3 6115
a449c7aa
NA		 6116 op_64_bit = is_64_bit_mode(vcpu);
6117 if (!op_64_bit) {
8776e519
HB		 6118 nr &= 0xFFFFFFFF;
6119 a0 &= 0xFFFFFFFF;
6120 a1 &= 0xFFFFFFFF;
6121 a2 &= 0xFFFFFFFF;
6122 a3 &= 0xFFFFFFFF;
6123 }
6124
07708c4a
JK		 6125 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6126 ret = -KVM_EPERM;
6127 goto out;
6128 }
6129
8776e519 6130 switch (nr) {
b93463aa
AK		 6131 case KVM_HC_VAPIC_POLL_IRQ:
6132 ret = 0;
6133 break;
6aef266c
SV		 6134 case KVM_HC_KICK_CPU:
6135 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6136 ret = 0;
6137 break;
8776e519
HB		 6138 default:
6139 ret = -KVM_ENOSYS;
6140 break;
6141 }
07708c4a 6142out:
a449c7aa
NA		 6143 if (!op_64_bit)
6144 ret = (u32)ret;
5fdbf976 6145 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
f11c3a8d 6146 ++vcpu->stat.hypercalls;
2f333bcb 6147 return r;
8776e519
HB		 6148}
6149EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6150
b6785def 6151static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8776e519 6152{
d6aa1000 6153 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8776e519 6154 char instruction[3];
5fdbf976 6155 unsigned long rip = kvm_rip_read(vcpu);
8776e519 6156
8776e519 6157 kvm_x86_ops->patch_hypercall(vcpu, instruction);
8776e519 6158
9d74191a 6159 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
8776e519
HB		 6160}
6161
851ba692 6162static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
b6c7a5dc 6163{
782d422b
MG		 6164 return vcpu->run->request_interrupt_window &&
6165 likely(!pic_in_kernel(vcpu->kvm));
b6c7a5dc
HB		 6166}
6167
851ba692 6168static void post_kvm_run_save(struct kvm_vcpu *vcpu)
b6c7a5dc 6169{
851ba692
AK		 6170 struct kvm_run *kvm_run = vcpu->run;
6171
91586a3b 6172 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
f077825a 6173 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
2d3ad1f4 6174 kvm_run->cr8 = kvm_get_cr8(vcpu);
b6c7a5dc 6175 kvm_run->apic_base = kvm_get_apic_base(vcpu);
127a457a
MG		 6176 kvm_run->ready_for_interrupt_injection =
6177 pic_in_kernel(vcpu->kvm) ||
782d422b 6178 kvm_vcpu_ready_for_interrupt_injection(vcpu);
b6c7a5dc
HB		 6179}
6180
95ba8273
GN		 6181static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6182{
6183 int max_irr, tpr;
6184
6185 if (!kvm_x86_ops->update_cr8_intercept)
6186 return;
6187
bce87cce 6188 if (!lapic_in_kernel(vcpu))
88c808fd
AK		 6189 return;
6190
d62caabb
AS		 6191 if (vcpu->arch.apicv_active)
6192 return;
6193
8db3baa2
GN		 6194 if (!vcpu->arch.apic->vapic_addr)
6195 max_irr = kvm_lapic_find_highest_irr(vcpu);
6196 else
6197 max_irr = -1;
95ba8273
GN		 6198
6199 if (max_irr != -1)
6200 max_irr >>= 4;
6201
6202 tpr = kvm_lapic_get_cr8(vcpu);
6203
6204 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6205}
6206
b6b8a145 6207static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
95ba8273 6208{
b6b8a145
JK		 6209 int r;
6210
95ba8273 6211 /* try to reinject previous events if any */
b59bb7bd 6212 if (vcpu->arch.exception.pending) {
5c1c85d0
AK		 6213 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6214 vcpu->arch.exception.has_error_code,
6215 vcpu->arch.exception.error_code);
d6e8c854
NA		 6216
6217 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6218 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6219 X86_EFLAGS_RF);
6220
6bdf0662
NA		 6221 if (vcpu->arch.exception.nr == DB_VECTOR &&
6222 (vcpu->arch.dr7 & DR7_GD)) {
6223 vcpu->arch.dr7 &= ~DR7_GD;
6224 kvm_update_dr7(vcpu);
6225 }
6226
b59bb7bd
GN		 6227 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6228 vcpu->arch.exception.has_error_code,
ce7ddec4
JR		 6229 vcpu->arch.exception.error_code,
6230 vcpu->arch.exception.reinject);
b6b8a145 6231 return 0;
b59bb7bd
GN		 6232 }
6233
95ba8273
GN		 6234 if (vcpu->arch.nmi_injected) {
6235 kvm_x86_ops->set_nmi(vcpu);
b6b8a145 6236 return 0;
95ba8273
GN		 6237 }
6238
6239 if (vcpu->arch.interrupt.pending) {
66fd3f7f 6240 kvm_x86_ops->set_irq(vcpu);
b6b8a145
JK		 6241 return 0;
6242 }
6243
6244 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6245 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6246 if (r != 0)
6247 return r;
95ba8273
GN		 6248 }
6249
6250 /* try to inject new event if pending */
c43203ca
PB		 6251 if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
6252 vcpu->arch.smi_pending = false;
ee2cd4b7 6253 enter_smm(vcpu);
c43203ca 6254 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
321c5658
YS		 6255 --vcpu->arch.nmi_pending;
6256 vcpu->arch.nmi_injected = true;
6257 kvm_x86_ops->set_nmi(vcpu);
c7c9c56c 6258 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
9242b5b6
BD		 6259 /*
6260 * Because interrupts can be injected asynchronously, we are
6261 * calling check_nested_events again here to avoid a race condition.
6262 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6263 * proposal and current concerns. Perhaps we should be setting
6264 * KVM_REQ_EVENT only on certain events and not unconditionally?
6265 */
6266 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6267 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6268 if (r != 0)
6269 return r;
6270 }
95ba8273 6271 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
66fd3f7f
GN		 6272 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6273 false);
6274 kvm_x86_ops->set_irq(vcpu);
95ba8273
GN		 6275 }
6276 }
ee2cd4b7 6277
b6b8a145 6278 return 0;
95ba8273
GN		 6279}
6280
7460fb4a
AK		 6281static void process_nmi(struct kvm_vcpu *vcpu)
6282{
6283 unsigned limit = 2;
6284
6285 /*
6286 * x86 is limited to one NMI running, and one NMI pending after it.
6287 * If an NMI is already in progress, limit further NMIs to just one.
6288 * Otherwise, allow two (and we'll inject the first one immediately).
6289 */
6290 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6291 limit = 1;
6292
6293 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6294 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6295 kvm_make_request(KVM_REQ_EVENT, vcpu);
6296}
6297
660a5d51
PB		 6298#define put_smstate(type, buf, offset, val) \
6299 *(type *)((buf) + (offset) - 0x7e00) = val
6300
ee2cd4b7 6301static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
660a5d51
PB		 6302{
6303 u32 flags = 0;
6304 flags |= seg->g << 23;
6305 flags |= seg->db << 22;
6306 flags |= seg->l << 21;
6307 flags |= seg->avl << 20;
6308 flags |= seg->present << 15;
6309 flags |= seg->dpl << 13;
6310 flags |= seg->s << 12;
6311 flags |= seg->type << 8;
6312 return flags;
6313}
6314
ee2cd4b7 6315static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
660a5d51
PB		 6316{
6317 struct kvm_segment seg;
6318 int offset;
6319
6320 kvm_get_segment(vcpu, &seg, n);
6321 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6322
6323 if (n < 3)
6324 offset = 0x7f84 + n * 12;
6325 else
6326 offset = 0x7f2c + (n - 3) * 12;
6327
6328 put_smstate(u32, buf, offset + 8, seg.base);
6329 put_smstate(u32, buf, offset + 4, seg.limit);
ee2cd4b7 6330 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
660a5d51
PB		 6331}
6332
efbb288a 6333#ifdef CONFIG_X86_64
ee2cd4b7 6334static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
660a5d51
PB		 6335{
6336 struct kvm_segment seg;
6337 int offset;
6338 u16 flags;
6339
6340 kvm_get_segment(vcpu, &seg, n);
6341 offset = 0x7e00 + n * 16;
6342
ee2cd4b7 6343 flags = enter_smm_get_segment_flags(&seg) >> 8;
660a5d51
PB		 6344 put_smstate(u16, buf, offset, seg.selector);
6345 put_smstate(u16, buf, offset + 2, flags);
6346 put_smstate(u32, buf, offset + 4, seg.limit);
6347 put_smstate(u64, buf, offset + 8, seg.base);
6348}
efbb288a 6349#endif
660a5d51 6350
ee2cd4b7 6351static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
660a5d51
PB		 6352{
6353 struct desc_ptr dt;
6354 struct kvm_segment seg;
6355 unsigned long val;
6356 int i;
6357
6358 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6359 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6360 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6361 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6362
6363 for (i = 0; i < 8; i++)
6364 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6365
6366 kvm_get_dr(vcpu, 6, &val);
6367 put_smstate(u32, buf, 0x7fcc, (u32)val);
6368 kvm_get_dr(vcpu, 7, &val);
6369 put_smstate(u32, buf, 0x7fc8, (u32)val);
6370
6371 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6372 put_smstate(u32, buf, 0x7fc4, seg.selector);
6373 put_smstate(u32, buf, 0x7f64, seg.base);
6374 put_smstate(u32, buf, 0x7f60, seg.limit);
ee2cd4b7 6375 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
660a5d51
PB		 6376
6377 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6378 put_smstate(u32, buf, 0x7fc0, seg.selector);
6379 put_smstate(u32, buf, 0x7f80, seg.base);
6380 put_smstate(u32, buf, 0x7f7c, seg.limit);
ee2cd4b7 6381 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
660a5d51
PB		 6382
6383 kvm_x86_ops->get_gdt(vcpu, &dt);
6384 put_smstate(u32, buf, 0x7f74, dt.address);
6385 put_smstate(u32, buf, 0x7f70, dt.size);
6386
6387 kvm_x86_ops->get_idt(vcpu, &dt);
6388 put_smstate(u32, buf, 0x7f58, dt.address);
6389 put_smstate(u32, buf, 0x7f54, dt.size);
6390
6391 for (i = 0; i < 6; i++)
ee2cd4b7 6392 enter_smm_save_seg_32(vcpu, buf, i);
660a5d51
PB		 6393
6394 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6395
6396 /* revision id */
6397 put_smstate(u32, buf, 0x7efc, 0x00020000);
6398 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6399}
6400
ee2cd4b7 6401static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
660a5d51
PB		 6402{
6403#ifdef CONFIG_X86_64
6404 struct desc_ptr dt;
6405 struct kvm_segment seg;
6406 unsigned long val;
6407 int i;
6408
6409 for (i = 0; i < 16; i++)
6410 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6411
6412 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6413 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6414
6415 kvm_get_dr(vcpu, 6, &val);
6416 put_smstate(u64, buf, 0x7f68, val);
6417 kvm_get_dr(vcpu, 7, &val);
6418 put_smstate(u64, buf, 0x7f60, val);
6419
6420 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6421 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6422 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6423
6424 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6425
6426 /* revision id */
6427 put_smstate(u32, buf, 0x7efc, 0x00020064);
6428
6429 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6430
6431 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6432 put_smstate(u16, buf, 0x7e90, seg.selector);
ee2cd4b7 6433 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
660a5d51
PB		 6434 put_smstate(u32, buf, 0x7e94, seg.limit);
6435 put_smstate(u64, buf, 0x7e98, seg.base);
6436
6437 kvm_x86_ops->get_idt(vcpu, &dt);
6438 put_smstate(u32, buf, 0x7e84, dt.size);
6439 put_smstate(u64, buf, 0x7e88, dt.address);
6440
6441 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6442 put_smstate(u16, buf, 0x7e70, seg.selector);
ee2cd4b7 6443 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
660a5d51
PB		 6444 put_smstate(u32, buf, 0x7e74, seg.limit);
6445 put_smstate(u64, buf, 0x7e78, seg.base);
6446
6447 kvm_x86_ops->get_gdt(vcpu, &dt);
6448 put_smstate(u32, buf, 0x7e64, dt.size);
6449 put_smstate(u64, buf, 0x7e68, dt.address);
6450
6451 for (i = 0; i < 6; i++)
ee2cd4b7 6452 enter_smm_save_seg_64(vcpu, buf, i);
660a5d51
PB		 6453#else
6454 WARN_ON_ONCE(1);
6455#endif
6456}
6457
ee2cd4b7 6458static void enter_smm(struct kvm_vcpu *vcpu)
64d60670 6459{
660a5d51 6460 struct kvm_segment cs, ds;
18c3626e 6461 struct desc_ptr dt;
660a5d51
PB		 6462 char buf[512];
6463 u32 cr0;
6464
660a5d51
PB		 6465 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6466 vcpu->arch.hflags |= HF_SMM_MASK;
6467 memset(buf, 0, 512);
6468 if (guest_cpuid_has_longmode(vcpu))
ee2cd4b7 6469 enter_smm_save_state_64(vcpu, buf);
660a5d51 6470 else
ee2cd4b7 6471 enter_smm_save_state_32(vcpu, buf);
660a5d51 6472
54bf36aa 6473 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
660a5d51
PB		 6474
6475 if (kvm_x86_ops->get_nmi_mask(vcpu))
6476 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6477 else
6478 kvm_x86_ops->set_nmi_mask(vcpu, true);
6479
6480 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6481 kvm_rip_write(vcpu, 0x8000);
6482
6483 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6484 kvm_x86_ops->set_cr0(vcpu, cr0);
6485 vcpu->arch.cr0 = cr0;
6486
6487 kvm_x86_ops->set_cr4(vcpu, 0);
6488
18c3626e
PB		 6489 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6490 dt.address = dt.size = 0;
6491 kvm_x86_ops->set_idt(vcpu, &dt);
6492
660a5d51
PB		 6493 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6494
6495 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6496 cs.base = vcpu->arch.smbase;
6497
6498 ds.selector = 0;
6499 ds.base = 0;
6500
6501 cs.limit = ds.limit = 0xffffffff;
6502 cs.type = ds.type = 0x3;
6503 cs.dpl = ds.dpl = 0;
6504 cs.db = ds.db = 0;
6505 cs.s = ds.s = 1;
6506 cs.l = ds.l = 0;
6507 cs.g = ds.g = 1;
6508 cs.avl = ds.avl = 0;
6509 cs.present = ds.present = 1;
6510 cs.unusable = ds.unusable = 0;
6511 cs.padding = ds.padding = 0;
6512
6513 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6514 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6515 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6516 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6517 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6518 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6519
6520 if (guest_cpuid_has_longmode(vcpu))
6521 kvm_x86_ops->set_efer(vcpu, 0);
6522
6523 kvm_update_cpuid(vcpu);
6524 kvm_mmu_reset_context(vcpu);
64d60670
PB		 6525}
6526
ee2cd4b7 6527static void process_smi(struct kvm_vcpu *vcpu)
c43203ca
PB		 6528{
6529 vcpu->arch.smi_pending = true;
6530 kvm_make_request(KVM_REQ_EVENT, vcpu);
6531}
6532
2860c4b1
PB		 6533void kvm_make_scan_ioapic_request(struct kvm *kvm)
6534{
6535 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6536}
6537
3d81bc7e 6538static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
c7c9c56c 6539{
5c919412
AS		 6540 u64 eoi_exit_bitmap[4];
6541
3d81bc7e
YZ		 6542 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6543 return;
c7c9c56c 6544
6308630b 6545 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
c7c9c56c 6546
b053b2ae 6547 if (irqchip_split(vcpu->kvm))
6308630b 6548 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
db2bdcbb 6549 else {
d62caabb
AS		 6550 if (vcpu->arch.apicv_active)
6551 kvm_x86_ops->sync_pir_to_irr(vcpu);
6308630b 6552 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
db2bdcbb 6553 }
5c919412
AS		 6554 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6555 vcpu_to_synic(vcpu)->vec_bitmap, 256);
6556 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
c7c9c56c
YZ		 6557}
6558
a70656b6
RK		 6559static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6560{
6561 ++vcpu->stat.tlb_flush;
6562 kvm_x86_ops->tlb_flush(vcpu);
6563}
6564
4256f43f
TC		 6565void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6566{
c24ae0dc
TC		 6567 struct page *page = NULL;
6568
35754c98 6569 if (!lapic_in_kernel(vcpu))
f439ed27
PB		 6570 return;
6571
4256f43f
TC		 6572 if (!kvm_x86_ops->set_apic_access_page_addr)
6573 return;
6574
c24ae0dc 6575 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
e8fd5e9e
AA		 6576 if (is_error_page(page))
6577 return;
c24ae0dc
TC		 6578 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6579
6580 /*
6581 * Do not pin apic access page in memory, the MMU notifier
6582 * will call us again if it is migrated or swapped out.
6583 */
6584 put_page(page);
4256f43f
TC		 6585}
6586EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6587
fe71557a
TC		 6588void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6589 unsigned long address)
6590{
c24ae0dc
TC		 6591 /*
6592 * The physical address of apic access page is stored in the VMCS.
6593 * Update it when it becomes invalid.
6594 */
6595 if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6596 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
fe71557a
TC		 6597}
6598
9357d939 6599/*
362c698f 6600 * Returns 1 to let vcpu_run() continue the guest execution loop without
9357d939
TY		 6601 * exiting to the userspace. Otherwise, the value will be returned to the
6602 * userspace.
6603 */
851ba692 6604static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
b6c7a5dc
HB		 6605{
6606 int r;
62a193ed
MG		 6607 bool req_int_win =
6608 dm_request_for_irq_injection(vcpu) &&
6609 kvm_cpu_accept_dm_intr(vcpu);
6610
730dca42 6611 bool req_immediate_exit = false;
b6c7a5dc 6612
3e007509 6613 if (vcpu->requests) {
a8eeb04a 6614 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
2e53d63a 6615 kvm_mmu_unload(vcpu);
a8eeb04a 6616 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
2f599714 6617 __kvm_migrate_timers(vcpu);
d828199e
MT		 6618 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6619 kvm_gen_update_masterclock(vcpu->kvm);
0061d53d
MT		 6620 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6621 kvm_gen_kvmclock_update(vcpu);
34c238a1
ZA		 6622 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6623 r = kvm_guest_time_update(vcpu);
8cfdc000
ZA		 6624 if (unlikely(r))
6625 goto out;
6626 }
a8eeb04a 6627 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
4731d4c7 6628 kvm_mmu_sync_roots(vcpu);
a8eeb04a 6629 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
a70656b6 6630 kvm_vcpu_flush_tlb(vcpu);
a8eeb04a 6631 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
851ba692 6632 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
b93463aa
AK		 6633 r = 0;
6634 goto out;
6635 }
a8eeb04a 6636 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
851ba692 6637 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
71c4dfaf
JR		 6638 r = 0;
6639 goto out;
6640 }
a8eeb04a 6641 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
02daab21
AK		 6642 vcpu->fpu_active = 0;
6643 kvm_x86_ops->fpu_deactivate(vcpu);
6644 }
af585b92
GN		 6645 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6646 /* Page is swapped out. Do synthetic halt */
6647 vcpu->arch.apf.halted = true;
6648 r = 1;
6649 goto out;
6650 }
c9aaa895
GC		 6651 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6652 record_steal_time(vcpu);
64d60670
PB		 6653 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6654 process_smi(vcpu);
7460fb4a
AK		 6655 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6656 process_nmi(vcpu);
f5132b01 6657 if (kvm_check_request(KVM_REQ_PMU, vcpu))
c6702c9d 6658 kvm_pmu_handle_event(vcpu);
f5132b01 6659 if (kvm_check_request(KVM_REQ_PMI, vcpu))
c6702c9d 6660 kvm_pmu_deliver_pmi(vcpu);
7543a635
SR		 6661 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6662 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6663 if (test_bit(vcpu->arch.pending_ioapic_eoi,
6308630b 6664 vcpu->arch.ioapic_handled_vectors)) {
7543a635
SR		 6665 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6666 vcpu->run->eoi.vector =
6667 vcpu->arch.pending_ioapic_eoi;
6668 r = 0;
6669 goto out;
6670 }
6671 }
3d81bc7e
YZ		 6672 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6673 vcpu_scan_ioapic(vcpu);
4256f43f
TC		 6674 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6675 kvm_vcpu_reload_apic_access_page(vcpu);
2ce79189
AS		 6676 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6677 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6678 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6679 r = 0;
6680 goto out;
6681 }
e516cebb
AS		 6682 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6683 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6684 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6685 r = 0;
6686 goto out;
6687 }
db397571
AS		 6688 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6689 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6690 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6691 r = 0;
6692 goto out;
6693 }
f3b138c5
AS		 6694
6695 /*
6696 * KVM_REQ_HV_STIMER has to be processed after
6697 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6698 * depend on the guest clock being up-to-date
6699 */
1f4b34f8
AS		 6700 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6701 kvm_hv_process_stimers(vcpu);
2f52d58c 6702 }
b93463aa 6703
bf9f6ac8
FW		 6704 /*
6705 * KVM_REQ_EVENT is not set when posted interrupts are set by
6706 * VT-d hardware, so we have to update RVI unconditionally.
6707 */
6708 if (kvm_lapic_enabled(vcpu)) {
6709 /*
6710 * Update architecture specific hints for APIC
6711 * virtual interrupt delivery.
6712 */
d62caabb 6713 if (vcpu->arch.apicv_active)
bf9f6ac8
FW		 6714 kvm_x86_ops->hwapic_irr_update(vcpu,
6715 kvm_lapic_find_highest_irr(vcpu));
2f52d58c 6716 }
b93463aa 6717
b463a6f7 6718 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
66450a21
JK		 6719 kvm_apic_accept_events(vcpu);
6720 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6721 r = 1;
6722 goto out;
6723 }
6724
b6b8a145
JK		 6725 if (inject_pending_event(vcpu, req_int_win) != 0)
6726 req_immediate_exit = true;
321c5658 6727 else {
c43203ca
PB		 6728 /* Enable NMI/IRQ window open exits if needed.
6729 *
6730 * SMIs have two cases: 1) they can be nested, and
6731 * then there is nothing to do here because RSM will
6732 * cause a vmexit anyway; 2) or the SMI can be pending
6733 * because inject_pending_event has completed the
6734 * injection of an IRQ or NMI from the previous vmexit,
6735 * and then we request an immediate exit to inject the SMI.
6736 */
6737 if (vcpu->arch.smi_pending && !is_smm(vcpu))
6738 req_immediate_exit = true;
321c5658
YS		 6739 if (vcpu->arch.nmi_pending)
6740 kvm_x86_ops->enable_nmi_window(vcpu);
6741 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6742 kvm_x86_ops->enable_irq_window(vcpu);
6743 }
b463a6f7
AK		 6744
6745 if (kvm_lapic_enabled(vcpu)) {
6746 update_cr8_intercept(vcpu);
6747 kvm_lapic_sync_to_vapic(vcpu);
6748 }
6749 }
6750
d8368af8
AK		 6751 r = kvm_mmu_reload(vcpu);
6752 if (unlikely(r)) {
d905c069 6753 goto cancel_injection;
d8368af8
AK		 6754 }
6755
b6c7a5dc
HB		 6756 preempt_disable();
6757
6758 kvm_x86_ops->prepare_guest_switch(vcpu);
2608d7a1
AK		 6759 if (vcpu->fpu_active)
6760 kvm_load_guest_fpu(vcpu);
6b7e2d09
XG		 6761 vcpu->mode = IN_GUEST_MODE;
6762
01b71917
MT		 6763 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6764
0f127d12
LT		 6765 /*
6766 * We should set ->mode before check ->requests,
6767 * Please see the comment in kvm_make_all_cpus_request.
6768 * This also orders the write to mode from any reads
6769 * to the page tables done while the VCPU is running.
6770 * Please see the comment in kvm_flush_remote_tlbs.
6b7e2d09 6771 */
01b71917 6772 smp_mb__after_srcu_read_unlock();
b6c7a5dc 6773
d94e1dc9 6774 local_irq_disable();
32f88400 6775
6b7e2d09 6776 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
d94e1dc9 6777 || need_resched() || signal_pending(current)) {
6b7e2d09 6778 vcpu->mode = OUTSIDE_GUEST_MODE;
d94e1dc9 6779 smp_wmb();
6c142801
AK		 6780 local_irq_enable();
6781 preempt_enable();
01b71917 6782 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6c142801 6783 r = 1;
d905c069 6784 goto cancel_injection;
6c142801
AK		 6785 }
6786
fc5b7f3b
DM		 6787 kvm_load_guest_xcr0(vcpu);
6788
c43203ca
PB		 6789 if (req_immediate_exit) {
6790 kvm_make_request(KVM_REQ_EVENT, vcpu);
d6185f20 6791 smp_send_reschedule(vcpu->cpu);
c43203ca 6792 }
d6185f20 6793
8b89fe1f
PB		 6794 trace_kvm_entry(vcpu->vcpu_id);
6795 wait_lapic_expire(vcpu);
6edaa530 6796 guest_enter_irqoff();
b6c7a5dc 6797
42dbaa5a 6798 if (unlikely(vcpu->arch.switch_db_regs)) {
42dbaa5a
JK		 6799 set_debugreg(0, 7);
6800 set_debugreg(vcpu->arch.eff_db[0], 0);
6801 set_debugreg(vcpu->arch.eff_db[1], 1);
6802 set_debugreg(vcpu->arch.eff_db[2], 2);
6803 set_debugreg(vcpu->arch.eff_db[3], 3);
c77fb5fe 6804 set_debugreg(vcpu->arch.dr6, 6);
ae561ede 6805 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
42dbaa5a 6806 }
b6c7a5dc 6807
851ba692 6808 kvm_x86_ops->run(vcpu);
b6c7a5dc 6809
c77fb5fe
PB		 6810 /*
6811 * Do this here before restoring debug registers on the host. And
6812 * since we do this before handling the vmexit, a DR access vmexit
6813 * can (a) read the correct value of the debug registers, (b) set
6814 * KVM_DEBUGREG_WONT_EXIT again.
6815 */
6816 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
c77fb5fe
PB		 6817 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6818 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
70e4da7a
PB		 6819 kvm_update_dr0123(vcpu);
6820 kvm_update_dr6(vcpu);
6821 kvm_update_dr7(vcpu);
6822 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
c77fb5fe
PB		 6823 }
6824
24f1e32c
FW		 6825 /*
6826 * If the guest has used debug registers, at least dr7
6827 * will be disabled while returning to the host.
6828 * If we don't have active breakpoints in the host, we don't
6829 * care about the messed up debug address registers. But if
6830 * we have some of them active, restore the old state.
6831 */
59d8eb53 6832 if (hw_breakpoint_active())
24f1e32c 6833 hw_breakpoint_restore();
42dbaa5a 6834
4ba76538 6835 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1d5f066e 6836
6b7e2d09 6837 vcpu->mode = OUTSIDE_GUEST_MODE;
d94e1dc9 6838 smp_wmb();
a547c6db 6839
fc5b7f3b
DM		 6840 kvm_put_guest_xcr0(vcpu);
6841
a547c6db 6842 kvm_x86_ops->handle_external_intr(vcpu);
b6c7a5dc
HB		 6843
6844 ++vcpu->stat.exits;
6845
f2485b3e 6846 guest_exit_irqoff();
b6c7a5dc 6847
f2485b3e 6848 local_irq_enable();
b6c7a5dc
HB		 6849 preempt_enable();
6850
f656ce01 6851 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3200f405 6852
b6c7a5dc
HB		 6853 /*
6854 * Profile KVM exit RIPs:
6855 */
6856 if (unlikely(prof_on == KVM_PROFILING)) {
5fdbf976
MT		 6857 unsigned long rip = kvm_rip_read(vcpu);
6858 profile_hit(KVM_PROFILING, (void *)rip);
b6c7a5dc
HB		 6859 }
6860
cc578287
ZA		 6861 if (unlikely(vcpu->arch.tsc_always_catchup))
6862 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
298101da 6863
5cfb1d5a
MT		 6864 if (vcpu->arch.apic_attention)
6865 kvm_lapic_sync_from_vapic(vcpu);
b93463aa 6866
851ba692 6867 r = kvm_x86_ops->handle_exit(vcpu);
d905c069
MT		 6868 return r;
6869
6870cancel_injection:
6871 kvm_x86_ops->cancel_injection(vcpu);
ae7a2a3f
MT		 6872 if (unlikely(vcpu->arch.apic_attention))
6873 kvm_lapic_sync_from_vapic(vcpu);
d7690175
MT		 6874out:
6875 return r;
6876}
b6c7a5dc 6877
362c698f
PB		 6878static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6879{
bf9f6ac8
FW		 6880 if (!kvm_arch_vcpu_runnable(vcpu) &&
6881 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
9c8fd1ba
PB		 6882 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6883 kvm_vcpu_block(vcpu);
6884 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
bf9f6ac8
FW		 6885
6886 if (kvm_x86_ops->post_block)
6887 kvm_x86_ops->post_block(vcpu);
6888
9c8fd1ba
PB		 6889 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6890 return 1;
6891 }
362c698f
PB		 6892
6893 kvm_apic_accept_events(vcpu);
6894 switch(vcpu->arch.mp_state) {
6895 case KVM_MP_STATE_HALTED:
6896 vcpu->arch.pv.pv_unhalted = false;
6897 vcpu->arch.mp_state =
6898 KVM_MP_STATE_RUNNABLE;
6899 case KVM_MP_STATE_RUNNABLE:
6900 vcpu->arch.apf.halted = false;
6901 break;
6902 case KVM_MP_STATE_INIT_RECEIVED:
6903 break;
6904 default:
6905 return -EINTR;
6906 break;
6907 }
6908 return 1;
6909}
09cec754 6910
5d9bc648
PB		 6911static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
6912{
6913 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6914 !vcpu->arch.apf.halted);
6915}
6916
362c698f 6917static int vcpu_run(struct kvm_vcpu *vcpu)
d7690175
MT		 6918{
6919 int r;
f656ce01 6920 struct kvm *kvm = vcpu->kvm;
d7690175 6921
f656ce01 6922 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
d7690175 6923
362c698f 6924 for (;;) {
58f800d5 6925 if (kvm_vcpu_running(vcpu)) {
851ba692 6926 r = vcpu_enter_guest(vcpu);
bf9f6ac8 6927 } else {
362c698f 6928 r = vcpu_block(kvm, vcpu);
bf9f6ac8
FW		 6929 }
6930
09cec754
GN		 6931 if (r <= 0)
6932 break;
6933
6934 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6935 if (kvm_cpu_has_pending_timer(vcpu))
6936 kvm_inject_pending_timer_irqs(vcpu);
6937
782d422b
MG		 6938 if (dm_request_for_irq_injection(vcpu) &&
6939 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
4ca7dd8c
PB		 6940 r = 0;
6941 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
09cec754 6942 ++vcpu->stat.request_irq_exits;
362c698f 6943 break;
09cec754 6944 }
af585b92
GN		 6945
6946 kvm_check_async_pf_completion(vcpu);
6947
09cec754
GN		 6948 if (signal_pending(current)) {
6949 r = -EINTR;
851ba692 6950 vcpu->run->exit_reason = KVM_EXIT_INTR;
09cec754 6951 ++vcpu->stat.signal_exits;
362c698f 6952 break;
09cec754
GN		 6953 }
6954 if (need_resched()) {
f656ce01 6955 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
c08ac06a 6956 cond_resched();
f656ce01 6957 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
d7690175 6958 }
b6c7a5dc
HB		 6959 }
6960
f656ce01 6961 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
b6c7a5dc
HB		 6962
6963 return r;
6964}
6965
716d51ab
GN		 6966static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6967{
6968 int r;
6969 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6970 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6971 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6972 if (r != EMULATE_DONE)
6973 return 0;
6974 return 1;
6975}
6976
6977static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6978{
6979 BUG_ON(!vcpu->arch.pio.count);
6980
6981 return complete_emulated_io(vcpu);
6982}
6983
f78146b0
AK		 6984/*
6985 * Implements the following, as a state machine:
6986 *
6987 * read:
6988 * for each fragment
87da7e66
XG		 6989 * for each mmio piece in the fragment
6990 * write gpa, len
6991 * exit
6992 * copy data
f78146b0
AK		 6993 * execute insn
6994 *
6995 * write:
6996 * for each fragment
87da7e66
XG		 6997 * for each mmio piece in the fragment
6998 * write gpa, len
6999 * copy data
7000 * exit
f78146b0 7001 */
716d51ab 7002static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
5287f194
AK		 7003{
7004 struct kvm_run *run = vcpu->run;
f78146b0 7005 struct kvm_mmio_fragment *frag;
87da7e66 7006 unsigned len;
5287f194 7007
716d51ab 7008 BUG_ON(!vcpu->mmio_needed);
5287f194 7009
716d51ab 7010 /* Complete previous fragment */
87da7e66
XG		 7011 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7012 len = min(8u, frag->len);
716d51ab 7013 if (!vcpu->mmio_is_write)
87da7e66
XG		 7014 memcpy(frag->data, run->mmio.data, len);
7015
7016 if (frag->len <= 8) {
7017 /* Switch to the next fragment. */
7018 frag++;
7019 vcpu->mmio_cur_fragment++;
7020 } else {
7021 /* Go forward to the next mmio piece. */
7022 frag->data += len;
7023 frag->gpa += len;
7024 frag->len -= len;
7025 }
7026
a08d3b3b 7027 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
716d51ab 7028 vcpu->mmio_needed = 0;
0912c977
PB		 7029
7030 /* FIXME: return into emulator if single-stepping. */
cef4dea0 7031 if (vcpu->mmio_is_write)
716d51ab
GN		 7032 return 1;
7033 vcpu->mmio_read_completed = 1;
7034 return complete_emulated_io(vcpu);
7035 }
87da7e66 7036
716d51ab
GN		 7037 run->exit_reason = KVM_EXIT_MMIO;
7038 run->mmio.phys_addr = frag->gpa;
7039 if (vcpu->mmio_is_write)
87da7e66
XG		 7040 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7041 run->mmio.len = min(8u, frag->len);
716d51ab
GN		 7042 run->mmio.is_write = vcpu->mmio_is_write;
7043 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7044 return 0;
5287f194
AK		 7045}
7046
716d51ab 7047
b6c7a5dc
HB		 7048int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7049{
c5bedc68 7050 struct fpu *fpu = &current->thread.fpu;
b6c7a5dc
HB		 7051 int r;
7052 sigset_t sigsaved;
7053
c4d72e2d 7054 fpu__activate_curr(fpu);
e5c30142 7055
ac9f6dc0
AK		 7056 if (vcpu->sigset_active)
7057 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
7058
a4535290 7059 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
b6c7a5dc 7060 kvm_vcpu_block(vcpu);
66450a21 7061 kvm_apic_accept_events(vcpu);
d7690175 7062 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
ac9f6dc0
AK		 7063 r = -EAGAIN;
7064 goto out;
b6c7a5dc
HB		 7065 }
7066
b6c7a5dc 7067 /* re-sync apic's tpr */
35754c98 7068 if (!lapic_in_kernel(vcpu)) {
eea1cff9
AP		 7069 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7070 r = -EINVAL;
7071 goto out;
7072 }
7073 }
b6c7a5dc 7074
716d51ab
GN		 7075 if (unlikely(vcpu->arch.complete_userspace_io)) {
7076 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7077 vcpu->arch.complete_userspace_io = NULL;
7078 r = cui(vcpu);
7079 if (r <= 0)
7080 goto out;
7081 } else
7082 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
5287f194 7083
362c698f 7084 r = vcpu_run(vcpu);
b6c7a5dc
HB		 7085
7086out:
f1d86e46 7087 post_kvm_run_save(vcpu);
b6c7a5dc
HB		 7088 if (vcpu->sigset_active)
7089 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
7090
b6c7a5dc
HB		 7091 return r;
7092}
7093
7094int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7095{
7ae441ea
GN		 7096 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
7097 /*
7098 * We are here if userspace calls get_regs() in the middle of
7099 * instruction emulation. Registers state needs to be copied
4a969980 7100 * back from emulation context to vcpu. Userspace shouldn't do
7ae441ea
GN		 7101 * that usually, but some bad designed PV devices (vmware
7102 * backdoor interface) need this to work
7103 */
dd856efa 7104 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7ae441ea
GN		 7105 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7106 }
5fdbf976
MT		 7107 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
7108 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
7109 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
7110 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
7111 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
7112 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
7113 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7114 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
b6c7a5dc 7115#ifdef CONFIG_X86_64
5fdbf976
MT		 7116 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7117 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7118 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7119 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7120 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7121 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7122 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7123 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
b6c7a5dc
HB		 7124#endif
7125
5fdbf976 7126 regs->rip = kvm_rip_read(vcpu);
91586a3b 7127 regs->rflags = kvm_get_rflags(vcpu);
b6c7a5dc 7128
b6c7a5dc
HB		 7129 return 0;
7130}
7131
7132int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7133{
7ae441ea
GN		 7134 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7135 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7136
5fdbf976
MT		 7137 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7138 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7139 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7140 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7141 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7142 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7143 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7144 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
b6c7a5dc 7145#ifdef CONFIG_X86_64
5fdbf976
MT		 7146 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7147 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7148 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7149 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7150 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7151 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7152 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7153 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
b6c7a5dc
HB		 7154#endif
7155
5fdbf976 7156 kvm_rip_write(vcpu, regs->rip);
91586a3b 7157 kvm_set_rflags(vcpu, regs->rflags);
b6c7a5dc 7158
b4f14abd
JK		 7159 vcpu->arch.exception.pending = false;
7160
3842d135
AK		 7161 kvm_make_request(KVM_REQ_EVENT, vcpu);
7162
b6c7a5dc
HB		 7163 return 0;
7164}
7165
b6c7a5dc
HB		 7166void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7167{
7168 struct kvm_segment cs;
7169
3e6e0aab 7170 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
b6c7a5dc
HB		 7171 *db = cs.db;
7172 *l = cs.l;
7173}
7174EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7175
7176int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7177 struct kvm_sregs *sregs)
7178{
89a27f4d 7179 struct desc_ptr dt;
b6c7a5dc 7180
3e6e0aab
GT		 7181 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7182 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7183 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7184 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7185 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7186 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
b6c7a5dc 7187
3e6e0aab
GT		 7188 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7189 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
b6c7a5dc
HB		 7190
7191 kvm_x86_ops->get_idt(vcpu, &dt);
89a27f4d
GN		 7192 sregs->idt.limit = dt.size;
7193 sregs->idt.base = dt.address;
b6c7a5dc 7194 kvm_x86_ops->get_gdt(vcpu, &dt);
89a27f4d
GN		 7195 sregs->gdt.limit = dt.size;
7196 sregs->gdt.base = dt.address;
b6c7a5dc 7197
4d4ec087 7198 sregs->cr0 = kvm_read_cr0(vcpu);
ad312c7c 7199 sregs->cr2 = vcpu->arch.cr2;
9f8fe504 7200 sregs->cr3 = kvm_read_cr3(vcpu);
fc78f519 7201 sregs->cr4 = kvm_read_cr4(vcpu);
2d3ad1f4 7202 sregs->cr8 = kvm_get_cr8(vcpu);
f6801dff 7203 sregs->efer = vcpu->arch.efer;
b6c7a5dc
HB		 7204 sregs->apic_base = kvm_get_apic_base(vcpu);
7205
923c61bb 7206 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
b6c7a5dc 7207
36752c9b 7208 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
14d0bc1f
GN		 7209 set_bit(vcpu->arch.interrupt.nr,
7210 (unsigned long *)sregs->interrupt_bitmap);
16d7a191 7211
b6c7a5dc
HB		 7212 return 0;
7213}
7214
62d9f0db
MT		 7215int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7216 struct kvm_mp_state *mp_state)
7217{
66450a21 7218 kvm_apic_accept_events(vcpu);
6aef266c
SV		 7219 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7220 vcpu->arch.pv.pv_unhalted)
7221 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7222 else
7223 mp_state->mp_state = vcpu->arch.mp_state;
7224
62d9f0db
MT		 7225 return 0;
7226}
7227
7228int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7229 struct kvm_mp_state *mp_state)
7230{
bce87cce 7231 if (!lapic_in_kernel(vcpu) &&
66450a21
JK		 7232 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7233 return -EINVAL;
7234
7235 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7236 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7237 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7238 } else
7239 vcpu->arch.mp_state = mp_state->mp_state;
3842d135 7240 kvm_make_request(KVM_REQ_EVENT, vcpu);
62d9f0db
MT		 7241 return 0;
7242}
7243
7f3d35fd
KW		 7244int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7245 int reason, bool has_error_code, u32 error_code)
b6c7a5dc 7246{
9d74191a 7247 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8ec4722d 7248 int ret;
e01c2426 7249
8ec4722d 7250 init_emulate_ctxt(vcpu);
c697518a 7251
7f3d35fd 7252 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9d74191a 7253 has_error_code, error_code);
c697518a 7254
c697518a 7255 if (ret)
19d04437 7256 return EMULATE_FAIL;
37817f29 7257
9d74191a
TY		 7258 kvm_rip_write(vcpu, ctxt->eip);
7259 kvm_set_rflags(vcpu, ctxt->eflags);
3842d135 7260 kvm_make_request(KVM_REQ_EVENT, vcpu);
19d04437 7261 return EMULATE_DONE;
37817f29
IE		 7262}
7263EXPORT_SYMBOL_GPL(kvm_task_switch);
7264
b6c7a5dc
HB		 7265int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7266 struct kvm_sregs *sregs)
7267{
58cb628d 7268 struct msr_data apic_base_msr;
b6c7a5dc 7269 int mmu_reset_needed = 0;
63f42e02 7270 int pending_vec, max_bits, idx;
89a27f4d 7271 struct desc_ptr dt;
b6c7a5dc 7272
6d1068b3
PM		 7273 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7274 return -EINVAL;
7275
89a27f4d
GN		 7276 dt.size = sregs->idt.limit;
7277 dt.address = sregs->idt.base;
b6c7a5dc 7278 kvm_x86_ops->set_idt(vcpu, &dt);
89a27f4d
GN		 7279 dt.size = sregs->gdt.limit;
7280 dt.address = sregs->gdt.base;
b6c7a5dc
HB		 7281 kvm_x86_ops->set_gdt(vcpu, &dt);
7282
ad312c7c 7283 vcpu->arch.cr2 = sregs->cr2;
9f8fe504 7284 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
dc7e795e 7285 vcpu->arch.cr3 = sregs->cr3;
aff48baa 7286 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
b6c7a5dc 7287
2d3ad1f4 7288 kvm_set_cr8(vcpu, sregs->cr8);
b6c7a5dc 7289
f6801dff 7290 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
b6c7a5dc 7291 kvm_x86_ops->set_efer(vcpu, sregs->efer);
58cb628d
JK		 7292 apic_base_msr.data = sregs->apic_base;
7293 apic_base_msr.host_initiated = true;
7294 kvm_set_apic_base(vcpu, &apic_base_msr);
b6c7a5dc 7295
4d4ec087 7296 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
b6c7a5dc 7297 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
d7306163 7298 vcpu->arch.cr0 = sregs->cr0;
b6c7a5dc 7299
fc78f519 7300 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
b6c7a5dc 7301 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
b9baba86 7302 if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
00b27a3e 7303 kvm_update_cpuid(vcpu);
63f42e02
XG		 7304
7305 idx = srcu_read_lock(&vcpu->kvm->srcu);
7c93be44 7306 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
9f8fe504 7307 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7c93be44
MT		 7308 mmu_reset_needed = 1;
7309 }
63f42e02 7310 srcu_read_unlock(&vcpu->kvm->srcu, idx);
b6c7a5dc
HB		 7311
7312 if (mmu_reset_needed)
7313 kvm_mmu_reset_context(vcpu);
7314
a50abc3b 7315 max_bits = KVM_NR_INTERRUPTS;
923c61bb
GN		 7316 pending_vec = find_first_bit(
7317 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7318 if (pending_vec < max_bits) {
66fd3f7f 7319 kvm_queue_interrupt(vcpu, pending_vec, false);
923c61bb 7320 pr_debug("Set back pending irq %d\n", pending_vec);
b6c7a5dc
HB		 7321 }
7322
3e6e0aab
GT		 7323 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7324 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7325 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7326 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7327 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7328 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
b6c7a5dc 7329
3e6e0aab
GT		 7330 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7331 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
b6c7a5dc 7332
5f0269f5
ME		 7333 update_cr8_intercept(vcpu);
7334
9c3e4aab 7335 /* Older userspace won't unhalt the vcpu on reset. */
c5af89b6 7336 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9c3e4aab 7337 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
3eeb3288 7338 !is_protmode(vcpu))
9c3e4aab
MT		 7339 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7340
3842d135
AK		 7341 kvm_make_request(KVM_REQ_EVENT, vcpu);
7342
b6c7a5dc
HB		 7343 return 0;
7344}
7345
d0bfb940
JK		 7346int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7347 struct kvm_guest_debug *dbg)
b6c7a5dc 7348{
355be0b9 7349 unsigned long rflags;
ae675ef0 7350 int i, r;
b6c7a5dc 7351
4f926bf2
JK		 7352 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7353 r = -EBUSY;
7354 if (vcpu->arch.exception.pending)
2122ff5e 7355 goto out;
4f926bf2
JK		 7356 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7357 kvm_queue_exception(vcpu, DB_VECTOR);
7358 else
7359 kvm_queue_exception(vcpu, BP_VECTOR);
7360 }
7361
91586a3b
JK		 7362 /*
7363 * Read rflags as long as potentially injected trace flags are still
7364 * filtered out.
7365 */
7366 rflags = kvm_get_rflags(vcpu);
355be0b9
JK		 7367
7368 vcpu->guest_debug = dbg->control;
7369 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7370 vcpu->guest_debug = 0;
7371
7372 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
ae675ef0
JK		 7373 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7374 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
c8639010 7375 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
ae675ef0
JK		 7376 } else {
7377 for (i = 0; i < KVM_NR_DB_REGS; i++)
7378 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
ae675ef0 7379 }
c8639010 7380 kvm_update_dr7(vcpu);
ae675ef0 7381
f92653ee
JK		 7382 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7383 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7384 get_segment_base(vcpu, VCPU_SREG_CS);
94fe45da 7385
91586a3b
JK		 7386 /*
7387 * Trigger an rflags update that will inject or remove the trace
7388 * flags.
7389 */
7390 kvm_set_rflags(vcpu, rflags);
b6c7a5dc 7391
a96036b8 7392 kvm_x86_ops->update_bp_intercept(vcpu);
b6c7a5dc 7393
4f926bf2 7394 r = 0;
d0bfb940 7395
2122ff5e 7396out:
b6c7a5dc
HB		 7397
7398 return r;
7399}
7400
8b006791
ZX		 7401/*
7402 * Translate a guest virtual address to a guest physical address.
7403 */
7404int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7405 struct kvm_translation *tr)
7406{
7407 unsigned long vaddr = tr->linear_address;
7408 gpa_t gpa;
f656ce01 7409 int idx;
8b006791 7410
f656ce01 7411 idx = srcu_read_lock(&vcpu->kvm->srcu);
1871c602 7412 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
f656ce01 7413 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8b006791
ZX		 7414 tr->physical_address = gpa;
7415 tr->valid = gpa != UNMAPPED_GVA;
7416 tr->writeable = 1;
7417 tr->usermode = 0;
8b006791
ZX		 7418
7419 return 0;
7420}
7421
d0752060
HB		 7422int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7423{
c47ada30 7424 struct fxregs_state *fxsave =
7366ed77 7425 &vcpu->arch.guest_fpu.state.fxsave;
d0752060 7426
d0752060
HB		 7427 memcpy(fpu->fpr, fxsave->st_space, 128);
7428 fpu->fcw = fxsave->cwd;
7429 fpu->fsw = fxsave->swd;
7430 fpu->ftwx = fxsave->twd;
7431 fpu->last_opcode = fxsave->fop;
7432 fpu->last_ip = fxsave->rip;
7433 fpu->last_dp = fxsave->rdp;
7434 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7435
d0752060
HB		 7436 return 0;
7437}
7438
7439int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7440{
c47ada30 7441 struct fxregs_state *fxsave =
7366ed77 7442 &vcpu->arch.guest_fpu.state.fxsave;
d0752060 7443
d0752060
HB		 7444 memcpy(fxsave->st_space, fpu->fpr, 128);
7445 fxsave->cwd = fpu->fcw;
7446 fxsave->swd = fpu->fsw;
7447 fxsave->twd = fpu->ftwx;
7448 fxsave->fop = fpu->last_opcode;
7449 fxsave->rip = fpu->last_ip;
7450 fxsave->rdp = fpu->last_dp;
7451 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7452
d0752060
HB		 7453 return 0;
7454}
7455
0ee6a517 7456static void fx_init(struct kvm_vcpu *vcpu)
d0752060 7457{
bf935b0b 7458 fpstate_init(&vcpu->arch.guest_fpu.state);
782511b0 7459 if (boot_cpu_has(X86_FEATURE_XSAVES))
7366ed77 7460 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
df1daba7 7461 host_xcr0 | XSTATE_COMPACTION_ENABLED;
d0752060 7462
2acf923e
DC		 7463 /*
7464 * Ensure guest xcr0 is valid for loading
7465 */
d91cab78 7466 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
2acf923e 7467
ad312c7c 7468 vcpu->arch.cr0 |= X86_CR0_ET;
d0752060 7469}
d0752060
HB		 7470
7471void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7472{
2608d7a1 7473 if (vcpu->guest_fpu_loaded)
d0752060
HB		 7474 return;
7475
2acf923e
DC		 7476 /*
7477 * Restore all possible states in the guest,
7478 * and assume host would use all available bits.
7479 * Guest xcr0 would be loaded later.
7480 */
d0752060 7481 vcpu->guest_fpu_loaded = 1;
b1a74bf8 7482 __kernel_fpu_begin();
003e2e8b 7483 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
0c04851c 7484 trace_kvm_fpu(1);
d0752060 7485}
d0752060
HB		 7486
7487void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7488{
3d42de25 7489 if (!vcpu->guest_fpu_loaded)
d0752060
HB		 7490 return;
7491
7492 vcpu->guest_fpu_loaded = 0;
4f836347 7493 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
b1a74bf8 7494 __kernel_fpu_end();
f096ed85 7495 ++vcpu->stat.fpu_reload;
0c04851c 7496 trace_kvm_fpu(0);
d0752060 7497}
e9b11c17
ZX		 7498
7499void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7500{
bd768e14
IY		 7501 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
7502
12f9a48f 7503 kvmclock_reset(vcpu);
7f1ea208 7504
e9b11c17 7505 kvm_x86_ops->vcpu_free(vcpu);
bd768e14 7506 free_cpumask_var(wbinvd_dirty_mask);
e9b11c17
ZX		 7507}
7508
7509struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7510 unsigned int id)
7511{
c447e76b
LL		 7512 struct kvm_vcpu *vcpu;
7513
6755bae8
ZA		 7514 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7515 printk_once(KERN_WARNING
7516 "kvm: SMP vm created on host with unstable TSC; "
7517 "guest TSC will not be reliable\n");
c447e76b
LL		 7518
7519 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7520
c447e76b 7521 return vcpu;
26e5215f 7522}
e9b11c17 7523
26e5215f
AK		 7524int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7525{
7526 int r;
e9b11c17 7527
19efffa2 7528 kvm_vcpu_mtrr_init(vcpu);
9fc77441
MT		 7529 r = vcpu_load(vcpu);
7530 if (r)
7531 return r;
d28bc9dd 7532 kvm_vcpu_reset(vcpu, false);
8a3c1a33 7533 kvm_mmu_setup(vcpu);
e9b11c17 7534 vcpu_put(vcpu);
26e5215f 7535 return r;
e9b11c17
ZX		 7536}
7537
31928aa5 7538void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
42897d86 7539{
8fe8ab46 7540 struct msr_data msr;
332967a3 7541 struct kvm *kvm = vcpu->kvm;
42897d86 7542
31928aa5
DD		 7543 if (vcpu_load(vcpu))
7544 return;
8fe8ab46
WA		 7545 msr.data = 0x0;
7546 msr.index = MSR_IA32_TSC;
7547 msr.host_initiated = true;
7548 kvm_write_tsc(vcpu, &msr);
42897d86
MT		 7549 vcpu_put(vcpu);
7550
630994b3
MT		 7551 if (!kvmclock_periodic_sync)
7552 return;
7553
332967a3
AJ		 7554 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7555 KVMCLOCK_SYNC_PERIOD);
42897d86
MT		 7556}
7557
d40ccc62 7558void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
e9b11c17 7559{
9fc77441 7560 int r;
344d9588
GN		 7561 vcpu->arch.apf.msr_val = 0;
7562
9fc77441
MT		 7563 r = vcpu_load(vcpu);
7564 BUG_ON(r);
e9b11c17
ZX		 7565 kvm_mmu_unload(vcpu);
7566 vcpu_put(vcpu);
7567
7568 kvm_x86_ops->vcpu_free(vcpu);
7569}
7570
d28bc9dd 7571void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
e9b11c17 7572{
e69fab5d
PB		 7573 vcpu->arch.hflags = 0;
7574
c43203ca 7575 vcpu->arch.smi_pending = 0;
7460fb4a
AK		 7576 atomic_set(&vcpu->arch.nmi_queued, 0);
7577 vcpu->arch.nmi_pending = 0;
448fa4a9 7578 vcpu->arch.nmi_injected = false;
5f7552d4
NA		 7579 kvm_clear_interrupt_queue(vcpu);
7580 kvm_clear_exception_queue(vcpu);
448fa4a9 7581
42dbaa5a 7582 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
ae561ede 7583 kvm_update_dr0123(vcpu);
6f43ed01 7584 vcpu->arch.dr6 = DR6_INIT;
73aaf249 7585 kvm_update_dr6(vcpu);
42dbaa5a 7586 vcpu->arch.dr7 = DR7_FIXED_1;
c8639010 7587 kvm_update_dr7(vcpu);
42dbaa5a 7588
1119022c
NA		 7589 vcpu->arch.cr2 = 0;
7590
3842d135 7591 kvm_make_request(KVM_REQ_EVENT, vcpu);
344d9588 7592 vcpu->arch.apf.msr_val = 0;
c9aaa895 7593 vcpu->arch.st.msr_val = 0;
3842d135 7594
12f9a48f
GC		 7595 kvmclock_reset(vcpu);
7596
af585b92
GN		 7597 kvm_clear_async_pf_completion_queue(vcpu);
7598 kvm_async_pf_hash_reset(vcpu);
7599 vcpu->arch.apf.halted = false;
3842d135 7600
64d60670 7601 if (!init_event) {
d28bc9dd 7602 kvm_pmu_reset(vcpu);
64d60670
PB		 7603 vcpu->arch.smbase = 0x30000;
7604 }
f5132b01 7605
66f7b72e
JS		 7606 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7607 vcpu->arch.regs_avail = ~0;
7608 vcpu->arch.regs_dirty = ~0;
7609
d28bc9dd 7610 kvm_x86_ops->vcpu_reset(vcpu, init_event);
e9b11c17
ZX		 7611}
7612
2b4a273b 7613void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
66450a21
JK		 7614{
7615 struct kvm_segment cs;
7616
7617 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7618 cs.selector = vector << 8;
7619 cs.base = vector << 12;
7620 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7621 kvm_rip_write(vcpu, 0);
e9b11c17
ZX		 7622}
7623
13a34e06 7624int kvm_arch_hardware_enable(void)
e9b11c17 7625{
ca84d1a2
ZA		 7626 struct kvm *kvm;
7627 struct kvm_vcpu *vcpu;
7628 int i;
0dd6a6ed
ZA		 7629 int ret;
7630 u64 local_tsc;
7631 u64 max_tsc = 0;
7632 bool stable, backwards_tsc = false;
18863bdd
AK		 7633
7634 kvm_shared_msr_cpu_online();
13a34e06 7635 ret = kvm_x86_ops->hardware_enable();
0dd6a6ed
ZA		 7636 if (ret != 0)
7637 return ret;
7638
4ea1636b 7639 local_tsc = rdtsc();
0dd6a6ed
ZA		 7640 stable = !check_tsc_unstable();
7641 list_for_each_entry(kvm, &vm_list, vm_list) {
7642 kvm_for_each_vcpu(i, vcpu, kvm) {
7643 if (!stable && vcpu->cpu == smp_processor_id())
105b21bb 7644 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
0dd6a6ed
ZA		 7645 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7646 backwards_tsc = true;
7647 if (vcpu->arch.last_host_tsc > max_tsc)
7648 max_tsc = vcpu->arch.last_host_tsc;
7649 }
7650 }
7651 }
7652
7653 /*
7654 * Sometimes, even reliable TSCs go backwards. This happens on
7655 * platforms that reset TSC during suspend or hibernate actions, but
7656 * maintain synchronization. We must compensate. Fortunately, we can
7657 * detect that condition here, which happens early in CPU bringup,
7658 * before any KVM threads can be running. Unfortunately, we can't
7659 * bring the TSCs fully up to date with real time, as we aren't yet far
7660 * enough into CPU bringup that we know how much real time has actually
108b249c 7661 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
0dd6a6ed
ZA		 7662 * variables that haven't been updated yet.
7663 *
7664 * So we simply find the maximum observed TSC above, then record the
7665 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7666 * the adjustment will be applied. Note that we accumulate
7667 * adjustments, in case multiple suspend cycles happen before some VCPU
7668 * gets a chance to run again. In the event that no KVM threads get a
7669 * chance to run, we will miss the entire elapsed period, as we'll have
7670 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7671 * loose cycle time. This isn't too big a deal, since the loss will be
7672 * uniform across all VCPUs (not to mention the scenario is extremely
7673 * unlikely). It is possible that a second hibernate recovery happens
7674 * much faster than a first, causing the observed TSC here to be
7675 * smaller; this would require additional padding adjustment, which is
7676 * why we set last_host_tsc to the local tsc observed here.
7677 *
7678 * N.B. - this code below runs only on platforms with reliable TSC,
7679 * as that is the only way backwards_tsc is set above. Also note
7680 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7681 * have the same delta_cyc adjustment applied if backwards_tsc
7682 * is detected. Note further, this adjustment is only done once,
7683 * as we reset last_host_tsc on all VCPUs to stop this from being
7684 * called multiple times (one for each physical CPU bringup).
7685 *
4a969980 7686 * Platforms with unreliable TSCs don't have to deal with this, they
0dd6a6ed
ZA		 7687 * will be compensated by the logic in vcpu_load, which sets the TSC to
7688 * catchup mode. This will catchup all VCPUs to real time, but cannot
7689 * guarantee that they stay in perfect synchronization.
7690 */
7691 if (backwards_tsc) {
7692 u64 delta_cyc = max_tsc - local_tsc;
16a96021 7693 backwards_tsc_observed = true;
0dd6a6ed
ZA		 7694 list_for_each_entry(kvm, &vm_list, vm_list) {
7695 kvm_for_each_vcpu(i, vcpu, kvm) {
7696 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7697 vcpu->arch.last_host_tsc = local_tsc;
105b21bb 7698 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
0dd6a6ed
ZA		 7699 }
7700
7701 /*
7702 * We have to disable TSC offset matching.. if you were
7703 * booting a VM while issuing an S4 host suspend....
7704 * you may have some problem. Solving this issue is
7705 * left as an exercise to the reader.
7706 */
7707 kvm->arch.last_tsc_nsec = 0;
7708 kvm->arch.last_tsc_write = 0;
7709 }
7710
7711 }
7712 return 0;
e9b11c17
ZX		 7713}
7714
13a34e06 7715void kvm_arch_hardware_disable(void)
e9b11c17 7716{
13a34e06
RK		 7717 kvm_x86_ops->hardware_disable();
7718 drop_user_return_notifiers();
e9b11c17
ZX		 7719}
7720
7721int kvm_arch_hardware_setup(void)
7722{
9e9c3fe4
NA		 7723 int r;
7724
7725 r = kvm_x86_ops->hardware_setup();
7726 if (r != 0)
7727 return r;
7728
35181e86
HZ		 7729 if (kvm_has_tsc_control) {
7730 /*
7731 * Make sure the user can only configure tsc_khz values that
7732 * fit into a signed integer.
7733 * A min value is not calculated needed because it will always
7734 * be 1 on all machines.
7735 */
7736 u64 max = min(0x7fffffffULL,
7737 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
7738 kvm_max_guest_tsc_khz = max;
7739
ad721883 7740 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
35181e86 7741 }
ad721883 7742
9e9c3fe4
NA		 7743 kvm_init_msr_list();
7744 return 0;
e9b11c17
ZX		 7745}
7746
7747void kvm_arch_hardware_unsetup(void)
7748{
7749 kvm_x86_ops->hardware_unsetup();
7750}
7751
7752void kvm_arch_check_processor_compat(void *rtn)
7753{
7754 kvm_x86_ops->check_processor_compatibility(rtn);
d71ba788
PB		 7755}
7756
7757bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7758{
7759 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7760}
7761EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7762
7763bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7764{
7765 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
e9b11c17
ZX		 7766}
7767
54e9818f 7768struct static_key kvm_no_apic_vcpu __read_mostly;
bce87cce 7769EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
54e9818f 7770
e9b11c17
ZX		 7771int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7772{
7773 struct page *page;
7774 struct kvm *kvm;
7775 int r;
7776
7777 BUG_ON(vcpu->kvm == NULL);
7778 kvm = vcpu->kvm;
7779
d62caabb 7780 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv();
6aef266c 7781 vcpu->arch.pv.pv_unhalted = false;
9aabc88f 7782 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
58d269d8 7783 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
a4535290 7784 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
e9b11c17 7785 else
a4535290 7786 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
e9b11c17
ZX		 7787
7788 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7789 if (!page) {
7790 r = -ENOMEM;
7791 goto fail;
7792 }
ad312c7c 7793 vcpu->arch.pio_data = page_address(page);
e9b11c17 7794
cc578287 7795 kvm_set_tsc_khz(vcpu, max_tsc_khz);
c285545f 7796
e9b11c17
ZX		 7797 r = kvm_mmu_create(vcpu);
7798 if (r < 0)
7799 goto fail_free_pio_data;
7800
7801 if (irqchip_in_kernel(kvm)) {
7802 r = kvm_create_lapic(vcpu);
7803 if (r < 0)
7804 goto fail_mmu_destroy;
54e9818f
GN		 7805 } else
7806 static_key_slow_inc(&kvm_no_apic_vcpu);
e9b11c17 7807
890ca9ae
HY		 7808 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7809 GFP_KERNEL);
7810 if (!vcpu->arch.mce_banks) {
7811 r = -ENOMEM;
443c39bc 7812 goto fail_free_lapic;
890ca9ae
HY		 7813 }
7814 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7815
f1797359
WY		 7816 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7817 r = -ENOMEM;
f5f48ee1 7818 goto fail_free_mce_banks;
f1797359 7819 }
f5f48ee1 7820
0ee6a517 7821 fx_init(vcpu);
66f7b72e 7822
ba904635 7823 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
0b79459b 7824 vcpu->arch.pv_time_enabled = false;
d7876f1b
PB		 7825
7826 vcpu->arch.guest_supported_xcr0 = 0;
4344ee98 7827 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
d7876f1b 7828
5a4f55cd
EK		 7829 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7830
74545705
RK		 7831 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7832
af585b92 7833 kvm_async_pf_hash_reset(vcpu);
f5132b01 7834 kvm_pmu_init(vcpu);
af585b92 7835
1c1a9ce9
SR		 7836 vcpu->arch.pending_external_vector = -1;
7837
5c919412
AS		 7838 kvm_hv_vcpu_init(vcpu);
7839
e9b11c17 7840 return 0;
0ee6a517 7841
f5f48ee1
SY		 7842fail_free_mce_banks:
7843 kfree(vcpu->arch.mce_banks);
443c39bc
WY		 7844fail_free_lapic:
7845 kvm_free_lapic(vcpu);
e9b11c17
ZX		 7846fail_mmu_destroy:
7847 kvm_mmu_destroy(vcpu);
7848fail_free_pio_data:
ad312c7c 7849 free_page((unsigned long)vcpu->arch.pio_data);
e9b11c17
ZX		 7850fail:
7851 return r;
7852}
7853
7854void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7855{
f656ce01
MT		 7856 int idx;
7857
1f4b34f8 7858 kvm_hv_vcpu_uninit(vcpu);
f5132b01 7859 kvm_pmu_destroy(vcpu);
36cb93fd 7860 kfree(vcpu->arch.mce_banks);
e9b11c17 7861 kvm_free_lapic(vcpu);
f656ce01 7862 idx = srcu_read_lock(&vcpu->kvm->srcu);
e9b11c17 7863 kvm_mmu_destroy(vcpu);
f656ce01 7864 srcu_read_unlock(&vcpu->kvm->srcu, idx);
ad312c7c 7865 free_page((unsigned long)vcpu->arch.pio_data);
35754c98 7866 if (!lapic_in_kernel(vcpu))
54e9818f 7867 static_key_slow_dec(&kvm_no_apic_vcpu);
e9b11c17 7868}
d19a9cd2 7869
e790d9ef
RK		 7870void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7871{
ae97a3b8 7872 kvm_x86_ops->sched_in(vcpu, cpu);
e790d9ef
RK		 7873}
7874
e08b9637 7875int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
d19a9cd2 7876{
e08b9637
CO		 7877 if (type)
7878 return -EINVAL;
7879
6ef768fa 7880 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
f05e70ac 7881 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
365c8868 7882 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
4d5c5d0f 7883 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
e0f0bbc5 7884 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
d19a9cd2 7885
5550af4d
SY		 7886 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7887 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7a84428a
AW		 7888 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7889 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7890 &kvm->arch.irq_sources_bitmap);
5550af4d 7891
038f8c11 7892 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
1e08ec4a 7893 mutex_init(&kvm->arch.apic_map_lock);
3f5ad8be 7894 mutex_init(&kvm->arch.hyperv.hv_lock);
d828199e
MT		 7895 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7896
108b249c 7897 kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
d828199e 7898 pvclock_update_vm_gtod_copy(kvm);
53f658b3 7899
7e44e449 7900 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
332967a3 7901 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7e44e449 7902
0eb05bf2 7903 kvm_page_track_init(kvm);
13d268ca 7904 kvm_mmu_init_vm(kvm);
0eb05bf2 7905
03543133
SS		 7906 if (kvm_x86_ops->vm_init)
7907 return kvm_x86_ops->vm_init(kvm);
7908
d89f5eff 7909 return 0;
d19a9cd2
ZX		 7910}
7911
7912static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7913{
9fc77441
MT		 7914 int r;
7915 r = vcpu_load(vcpu);
7916 BUG_ON(r);
d19a9cd2
ZX		 7917 kvm_mmu_unload(vcpu);
7918 vcpu_put(vcpu);
7919}
7920
7921static void kvm_free_vcpus(struct kvm *kvm)
7922{
7923 unsigned int i;
988a2cae 7924 struct kvm_vcpu *vcpu;
d19a9cd2
ZX		 7925
7926 /*
7927 * Unpin any mmu pages first.
7928 */
af585b92
GN		 7929 kvm_for_each_vcpu(i, vcpu, kvm) {
7930 kvm_clear_async_pf_completion_queue(vcpu);
988a2cae 7931 kvm_unload_vcpu_mmu(vcpu);
af585b92 7932 }
988a2cae
GN		 7933 kvm_for_each_vcpu(i, vcpu, kvm)
7934 kvm_arch_vcpu_free(vcpu);
7935
7936 mutex_lock(&kvm->lock);
7937 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7938 kvm->vcpus[i] = NULL;
d19a9cd2 7939
988a2cae
GN		 7940 atomic_set(&kvm->online_vcpus, 0);
7941 mutex_unlock(&kvm->lock);
d19a9cd2
ZX		 7942}
7943
ad8ba2cd
SY		 7944void kvm_arch_sync_events(struct kvm *kvm)
7945{
332967a3 7946 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7e44e449 7947 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
ba4cef31 7948 kvm_free_all_assigned_devices(kvm);
aea924f6 7949 kvm_free_pit(kvm);
ad8ba2cd
SY		 7950}
7951
1d8007bd 7952int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9da0e4d5
PB		 7953{
7954 int i, r;
25188b99 7955 unsigned long hva;
f0d648bd
PB		 7956 struct kvm_memslots *slots = kvm_memslots(kvm);
7957 struct kvm_memory_slot *slot, old;
9da0e4d5
PB		 7958
7959 /* Called with kvm->slots_lock held. */
1d8007bd
PB		 7960 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
7961 return -EINVAL;
9da0e4d5 7962
f0d648bd
PB		 7963 slot = id_to_memslot(slots, id);
7964 if (size) {
b21629da 7965 if (slot->npages)
f0d648bd
PB		 7966 return -EEXIST;
7967
7968 /*
7969 * MAP_SHARED to prevent internal slot pages from being moved
7970 * by fork()/COW.
7971 */
7972 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
7973 MAP_SHARED | MAP_ANONYMOUS, 0);
7974 if (IS_ERR((void *)hva))
7975 return PTR_ERR((void *)hva);
7976 } else {
7977 if (!slot->npages)
7978 return 0;
7979
7980 hva = 0;
7981 }
7982
7983 old = *slot;
9da0e4d5 7984 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1d8007bd 7985 struct kvm_userspace_memory_region m;
9da0e4d5 7986
1d8007bd
PB		 7987 m.slot = id | (i << 16);
7988 m.flags = 0;
7989 m.guest_phys_addr = gpa;
f0d648bd 7990 m.userspace_addr = hva;
1d8007bd 7991 m.memory_size = size;
9da0e4d5
PB		 7992 r = __kvm_set_memory_region(kvm, &m);
7993 if (r < 0)
7994 return r;
7995 }
7996
f0d648bd
PB		 7997 if (!size) {
7998 r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
7999 WARN_ON(r < 0);
8000 }
8001
9da0e4d5
PB		 8002 return 0;
8003}
8004EXPORT_SYMBOL_GPL(__x86_set_memory_region);
8005
1d8007bd 8006int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9da0e4d5
PB		 8007{
8008 int r;
8009
8010 mutex_lock(&kvm->slots_lock);
1d8007bd 8011 r = __x86_set_memory_region(kvm, id, gpa, size);
9da0e4d5
PB		 8012 mutex_unlock(&kvm->slots_lock);
8013
8014 return r;
8015}
8016EXPORT_SYMBOL_GPL(x86_set_memory_region);
8017
d19a9cd2
ZX		 8018void kvm_arch_destroy_vm(struct kvm *kvm)
8019{
27469d29
AH		 8020 if (current->mm == kvm->mm) {
8021 /*
8022 * Free memory regions allocated on behalf of userspace,
8023 * unless the the memory map has changed due to process exit
8024 * or fd copying.
8025 */
1d8007bd
PB		 8026 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
8027 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
8028 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
27469d29 8029 }
03543133
SS		 8030 if (kvm_x86_ops->vm_destroy)
8031 kvm_x86_ops->vm_destroy(kvm);
6eb55818 8032 kvm_iommu_unmap_guest(kvm);
d7deeeb0
ZX		 8033 kfree(kvm->arch.vpic);
8034 kfree(kvm->arch.vioapic);
d19a9cd2 8035 kvm_free_vcpus(kvm);
af1bae54 8036 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
13d268ca 8037 kvm_mmu_uninit_vm(kvm);
d19a9cd2 8038}
0de10343 8039
5587027c 8040void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
db3fe4eb
TY		 8041 struct kvm_memory_slot *dont)
8042{
8043 int i;
8044
d89cc617
TY		 8045 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8046 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
548ef284 8047 kvfree(free->arch.rmap[i]);
d89cc617 8048 free->arch.rmap[i] = NULL;
77d11309 8049 }
d89cc617
TY		 8050 if (i == 0)
8051 continue;
8052
8053 if (!dont || free->arch.lpage_info[i - 1] !=
8054 dont->arch.lpage_info[i - 1]) {
548ef284 8055 kvfree(free->arch.lpage_info[i - 1]);
d89cc617 8056 free->arch.lpage_info[i - 1] = NULL;
db3fe4eb
TY		 8057 }
8058 }
21ebbeda
XG		 8059
8060 kvm_page_track_free_memslot(free, dont);
db3fe4eb
TY		 8061}
8062
5587027c
AK		 8063int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
8064 unsigned long npages)
db3fe4eb
TY		 8065{
8066 int i;
8067
d89cc617 8068 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
92f94f1e 8069 struct kvm_lpage_info *linfo;
db3fe4eb
TY		 8070 unsigned long ugfn;
8071 int lpages;
d89cc617 8072 int level = i + 1;
db3fe4eb
TY		 8073
8074 lpages = gfn_to_index(slot->base_gfn + npages - 1,
8075 slot->base_gfn, level) + 1;
8076
d89cc617
TY		 8077 slot->arch.rmap[i] =
8078 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
8079 if (!slot->arch.rmap[i])
77d11309 8080 goto out_free;
d89cc617
TY		 8081 if (i == 0)
8082 continue;
77d11309 8083
92f94f1e
XG		 8084 linfo = kvm_kvzalloc(lpages * sizeof(*linfo));
8085 if (!linfo)
db3fe4eb
TY		 8086 goto out_free;
8087
92f94f1e
XG		 8088 slot->arch.lpage_info[i - 1] = linfo;
8089
db3fe4eb 8090 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
92f94f1e 8091 linfo[0].disallow_lpage = 1;
db3fe4eb 8092 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
92f94f1e 8093 linfo[lpages - 1].disallow_lpage = 1;
db3fe4eb
TY		 8094 ugfn = slot->userspace_addr >> PAGE_SHIFT;
8095 /*
8096 * If the gfn and userspace address are not aligned wrt each
8097 * other, or if explicitly asked to, disable large page
8098 * support for this slot
8099 */
8100 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
8101 !kvm_largepages_enabled()) {
8102 unsigned long j;
8103
8104 for (j = 0; j < lpages; ++j)
92f94f1e 8105 linfo[j].disallow_lpage = 1;
db3fe4eb
TY		 8106 }
8107 }
8108
21ebbeda
XG		 8109 if (kvm_page_track_create_memslot(slot, npages))
8110 goto out_free;
8111
db3fe4eb
TY		 8112 return 0;
8113
8114out_free:
d89cc617 8115 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
548ef284 8116 kvfree(slot->arch.rmap[i]);
d89cc617
TY		 8117 slot->arch.rmap[i] = NULL;
8118 if (i == 0)
8119 continue;
8120
548ef284 8121 kvfree(slot->arch.lpage_info[i - 1]);
d89cc617 8122 slot->arch.lpage_info[i - 1] = NULL;
db3fe4eb
TY		 8123 }
8124 return -ENOMEM;
8125}
8126
15f46015 8127void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
e59dbe09 8128{
e6dff7d1
TY		 8129 /*
8130 * memslots->generation has been incremented.
8131 * mmio generation may have reached its maximum value.
8132 */
54bf36aa 8133 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
e59dbe09
TY		 8134}
8135
f7784b8e
MT		 8136int kvm_arch_prepare_memory_region(struct kvm *kvm,
8137 struct kvm_memory_slot *memslot,
09170a49 8138 const struct kvm_userspace_memory_region *mem,
7b6195a9 8139 enum kvm_mr_change change)
0de10343 8140{
f7784b8e
MT		 8141 return 0;
8142}
8143
88178fd4
KH		 8144static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8145 struct kvm_memory_slot *new)
8146{
8147 /* Still write protect RO slot */
8148 if (new->flags & KVM_MEM_READONLY) {
8149 kvm_mmu_slot_remove_write_access(kvm, new);
8150 return;
8151 }
8152
8153 /*
8154 * Call kvm_x86_ops dirty logging hooks when they are valid.
8155 *
8156 * kvm_x86_ops->slot_disable_log_dirty is called when:
8157 *
8158 * - KVM_MR_CREATE with dirty logging is disabled
8159 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8160 *
8161 * The reason is, in case of PML, we need to set D-bit for any slots
8162 * with dirty logging disabled in order to eliminate unnecessary GPA
8163 * logging in PML buffer (and potential PML buffer full VMEXT). This
8164 * guarantees leaving PML enabled during guest's lifetime won't have
8165 * any additonal overhead from PML when guest is running with dirty
8166 * logging disabled for memory slots.
8167 *
8168 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8169 * to dirty logging mode.
8170 *
8171 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8172 *
8173 * In case of write protect:
8174 *
8175 * Write protect all pages for dirty logging.
8176 *
8177 * All the sptes including the large sptes which point to this
8178 * slot are set to readonly. We can not create any new large
8179 * spte on this slot until the end of the logging.
8180 *
8181 * See the comments in fast_page_fault().
8182 */
8183 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8184 if (kvm_x86_ops->slot_enable_log_dirty)
8185 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8186 else
8187 kvm_mmu_slot_remove_write_access(kvm, new);
8188 } else {
8189 if (kvm_x86_ops->slot_disable_log_dirty)
8190 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8191 }
8192}
8193
f7784b8e 8194void kvm_arch_commit_memory_region(struct kvm *kvm,
09170a49 8195 const struct kvm_userspace_memory_region *mem,
8482644a 8196 const struct kvm_memory_slot *old,
f36f3f28 8197 const struct kvm_memory_slot *new,
8482644a 8198 enum kvm_mr_change change)
f7784b8e 8199{
8482644a 8200 int nr_mmu_pages = 0;
f7784b8e 8201
48c0e4e9
XG		 8202 if (!kvm->arch.n_requested_mmu_pages)
8203 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8204
48c0e4e9 8205 if (nr_mmu_pages)
0de10343 8206 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
1c91cad4 8207
3ea3b7fa
WL		 8208 /*
8209 * Dirty logging tracks sptes in 4k granularity, meaning that large
8210 * sptes have to be split. If live migration is successful, the guest
8211 * in the source machine will be destroyed and large sptes will be
8212 * created in the destination. However, if the guest continues to run
8213 * in the source machine (for example if live migration fails), small
8214 * sptes will remain around and cause bad performance.
8215 *
8216 * Scan sptes if dirty logging has been stopped, dropping those
8217 * which can be collapsed into a single large-page spte. Later
8218 * page faults will create the large-page sptes.
8219 */
8220 if ((change != KVM_MR_DELETE) &&
8221 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8222 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8223 kvm_mmu_zap_collapsible_sptes(kvm, new);
8224
c972f3b1 8225 /*
88178fd4 8226 * Set up write protection and/or dirty logging for the new slot.
c126d94f 8227 *
88178fd4
KH		 8228 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8229 * been zapped so no dirty logging staff is needed for old slot. For
8230 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8231 * new and it's also covered when dealing with the new slot.
f36f3f28
PB		 8232 *
8233 * FIXME: const-ify all uses of struct kvm_memory_slot.
c972f3b1 8234 */
88178fd4 8235 if (change != KVM_MR_DELETE)
f36f3f28 8236 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
0de10343 8237}
1d737c8a 8238
2df72e9b 8239void kvm_arch_flush_shadow_all(struct kvm *kvm)
34d4cb8f 8240{
6ca18b69 8241 kvm_mmu_invalidate_zap_all_pages(kvm);
34d4cb8f
MT		 8242}
8243
2df72e9b
MT		 8244void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8245 struct kvm_memory_slot *slot)
8246{
ae7cd873 8247 kvm_page_track_flush_slot(kvm, slot);
2df72e9b
MT		 8248}
8249
5d9bc648
PB		 8250static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8251{
8252 if (!list_empty_careful(&vcpu->async_pf.done))
8253 return true;
8254
8255 if (kvm_apic_has_events(vcpu))
8256 return true;
8257
8258 if (vcpu->arch.pv.pv_unhalted)
8259 return true;
8260
8261 if (atomic_read(&vcpu->arch.nmi_queued))
8262 return true;
8263
73917739
PB		 8264 if (test_bit(KVM_REQ_SMI, &vcpu->requests))
8265 return true;
8266
5d9bc648
PB		 8267 if (kvm_arch_interrupt_allowed(vcpu) &&
8268 kvm_cpu_has_interrupt(vcpu))
8269 return true;
8270
1f4b34f8
AS		 8271 if (kvm_hv_has_stimer_pending(vcpu))
8272 return true;
8273
5d9bc648
PB		 8274 return false;
8275}
8276
1d737c8a
ZX		 8277int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8278{
b6b8a145
JK		 8279 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8280 kvm_x86_ops->check_nested_events(vcpu, false);
8281
5d9bc648 8282 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
1d737c8a 8283}
5736199a 8284
b6d33834 8285int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
5736199a 8286{
b6d33834 8287 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
5736199a 8288}
78646121
GN		 8289
8290int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8291{
8292 return kvm_x86_ops->interrupt_allowed(vcpu);
8293}
229456fc 8294
82b32774 8295unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
f92653ee 8296{
82b32774
NA		 8297 if (is_64_bit_mode(vcpu))
8298 return kvm_rip_read(vcpu);
8299 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8300 kvm_rip_read(vcpu));
8301}
8302EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
f92653ee 8303
82b32774
NA		 8304bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8305{
8306 return kvm_get_linear_rip(vcpu) == linear_rip;
f92653ee
JK		 8307}
8308EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8309
94fe45da
JK		 8310unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8311{
8312 unsigned long rflags;
8313
8314 rflags = kvm_x86_ops->get_rflags(vcpu);
8315 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
c310bac5 8316 rflags &= ~X86_EFLAGS_TF;
94fe45da
JK		 8317 return rflags;
8318}
8319EXPORT_SYMBOL_GPL(kvm_get_rflags);
8320
6addfc42 8321static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
94fe45da
JK		 8322{
8323 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
f92653ee 8324 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
c310bac5 8325 rflags |= X86_EFLAGS_TF;
94fe45da 8326 kvm_x86_ops->set_rflags(vcpu, rflags);
6addfc42
PB		 8327}
8328
8329void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8330{
8331 __kvm_set_rflags(vcpu, rflags);
3842d135 8332 kvm_make_request(KVM_REQ_EVENT, vcpu);
94fe45da
JK		 8333}
8334EXPORT_SYMBOL_GPL(kvm_set_rflags);
8335
56028d08
GN		 8336void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8337{
8338 int r;
8339
fb67e14f 8340 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
f2e10669 8341 work->wakeup_all)
56028d08
GN		 8342 return;
8343
8344 r = kvm_mmu_reload(vcpu);
8345 if (unlikely(r))
8346 return;
8347
fb67e14f
XG		 8348 if (!vcpu->arch.mmu.direct_map &&
8349 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8350 return;
8351
56028d08
GN		 8352 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8353}
8354
af585b92
GN		 8355static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8356{
8357 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8358}
8359
8360static inline u32 kvm_async_pf_next_probe(u32 key)
8361{
8362 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8363}
8364
8365static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8366{
8367 u32 key = kvm_async_pf_hash_fn(gfn);
8368
8369 while (vcpu->arch.apf.gfns[key] != ~0)
8370 key = kvm_async_pf_next_probe(key);
8371
8372 vcpu->arch.apf.gfns[key] = gfn;
8373}
8374
8375static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8376{
8377 int i;
8378 u32 key = kvm_async_pf_hash_fn(gfn);
8379
8380 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
c7d28c24
XG		 8381 (vcpu->arch.apf.gfns[key] != gfn &&
8382 vcpu->arch.apf.gfns[key] != ~0); i++)
af585b92
GN		 8383 key = kvm_async_pf_next_probe(key);
8384
8385 return key;
8386}
8387
8388bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8389{
8390 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8391}
8392
8393static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8394{
8395 u32 i, j, k;
8396
8397 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8398 while (true) {
8399 vcpu->arch.apf.gfns[i] = ~0;
8400 do {
8401 j = kvm_async_pf_next_probe(j);
8402 if (vcpu->arch.apf.gfns[j] == ~0)
8403 return;
8404 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8405 /*
8406 * k lies cyclically in ]i,j]
8407 * | i.k.j |
8408 * |....j i.k.| or |.k..j i...|
8409 */
8410 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8411 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8412 i = j;
8413 }
8414}
8415
7c90705b
GN		 8416static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8417{
8418
8419 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8420 sizeof(val));
8421}
8422
af585b92
GN		 8423void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8424 struct kvm_async_pf *work)
8425{
6389ee94
AK		 8426 struct x86_exception fault;
8427
7c90705b 8428 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
af585b92 8429 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7c90705b
GN		 8430
8431 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
fc5f06fa
GN		 8432 (vcpu->arch.apf.send_user_only &&
8433 kvm_x86_ops->get_cpl(vcpu) == 0))
7c90705b
GN		 8434 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8435 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6389ee94
AK		 8436 fault.vector = PF_VECTOR;
8437 fault.error_code_valid = true;
8438 fault.error_code = 0;
8439 fault.nested_page_fault = false;
8440 fault.address = work->arch.token;
8441 kvm_inject_page_fault(vcpu, &fault);
7c90705b 8442 }
af585b92
GN		 8443}
8444
8445void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8446 struct kvm_async_pf *work)
8447{
6389ee94
AK		 8448 struct x86_exception fault;
8449
7c90705b 8450 trace_kvm_async_pf_ready(work->arch.token, work->gva);
f2e10669 8451 if (work->wakeup_all)
7c90705b
GN		 8452 work->arch.token = ~0; /* broadcast wakeup */
8453 else
8454 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8455
8456 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8457 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6389ee94
AK		 8458 fault.vector = PF_VECTOR;
8459 fault.error_code_valid = true;
8460 fault.error_code = 0;
8461 fault.nested_page_fault = false;
8462 fault.address = work->arch.token;
8463 kvm_inject_page_fault(vcpu, &fault);
7c90705b 8464 }
e6d53e3b 8465 vcpu->arch.apf.halted = false;
a4fa1635 8466 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7c90705b
GN		 8467}
8468
8469bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8470{
8471 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8472 return true;
8473 else
8474 return !kvm_event_needs_reinjection(vcpu) &&
8475 kvm_x86_ops->interrupt_allowed(vcpu);
af585b92
GN		 8476}
8477
5544eb9b
PB		 8478void kvm_arch_start_assignment(struct kvm *kvm)
8479{
8480 atomic_inc(&kvm->arch.assigned_device_count);
8481}
8482EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8483
8484void kvm_arch_end_assignment(struct kvm *kvm)
8485{
8486 atomic_dec(&kvm->arch.assigned_device_count);
8487}
8488EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8489
8490bool kvm_arch_has_assigned_device(struct kvm *kvm)
8491{
8492 return atomic_read(&kvm->arch.assigned_device_count);
8493}
8494EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8495
e0f0bbc5
AW		 8496void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8497{
8498 atomic_inc(&kvm->arch.noncoherent_dma_count);
8499}
8500EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8501
8502void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8503{
8504 atomic_dec(&kvm->arch.noncoherent_dma_count);
8505}
8506EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8507
8508bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8509{
8510 return atomic_read(&kvm->arch.noncoherent_dma_count);
8511}
8512EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8513
14717e20
AW		 8514bool kvm_arch_has_irq_bypass(void)
8515{
8516 return kvm_x86_ops->update_pi_irte != NULL;
8517}
8518
87276880
FW		 8519int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8520 struct irq_bypass_producer *prod)
8521{
8522 struct kvm_kernel_irqfd *irqfd =
8523 container_of(cons, struct kvm_kernel_irqfd, consumer);
8524
14717e20 8525 irqfd->producer = prod;
87276880 8526
14717e20
AW		 8527 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8528 prod->irq, irqfd->gsi, 1);
87276880
FW		 8529}
8530
8531void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8532 struct irq_bypass_producer *prod)
8533{
8534 int ret;
8535 struct kvm_kernel_irqfd *irqfd =
8536 container_of(cons, struct kvm_kernel_irqfd, consumer);
8537
87276880
FW		 8538 WARN_ON(irqfd->producer != prod);
8539 irqfd->producer = NULL;
8540
8541 /*
8542 * When producer of consumer is unregistered, we change back to
8543 * remapped mode, so we can re-use the current implementation
bb3541f1 8544 * when the irq is masked/disabled or the consumer side (KVM
87276880
FW		 8545 * int this case doesn't want to receive the interrupts.
8546 */
8547 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8548 if (ret)
8549 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8550 " fails: %d\n", irqfd->consumer.token, ret);
8551}
8552
8553int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8554 uint32_t guest_irq, bool set)
8555{
8556 if (!kvm_x86_ops->update_pi_irte)
8557 return -EINVAL;
8558
8559 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8560}
8561
52004014
FW		 8562bool kvm_vector_hashing_enabled(void)
8563{
8564 return vector_hashing;
8565}
8566EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
8567
229456fc 8568EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
931c33b1 8569EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
229456fc
MT		 8570EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8571EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8572EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8573EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
0ac406de 8574EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
d8cabddf 8575EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
17897f36 8576EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
236649de 8577EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
ec1ff790 8578EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
532a46b9 8579EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
2e554e8d 8580EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
489223ed 8581EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
7b46268d 8582EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
843e4330 8583EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
efc64404 8584EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
18f40c53
SS		 8585EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
8586EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
Linux 6.x block layer and io_uring tree(s)