]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - arch/x86/kvm/x86.c
projects / mirror_ubuntu-zesty-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: x86: fix user triggerable warning in kvm_apic_accept_events()
[mirror_ubuntu-zesty-kernel.git] / arch / x86 / kvm / x86.c
CommitLineData
043405e1
CO		 1/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * derived from drivers/kvm/kvm_main.c
5 *
6 * Copyright (C) 2006 Qumranet, Inc.
4d5c5d0f
BAY		 7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9611c187 9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
043405e1
CO		 10 *
11 * Authors:
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
4d5c5d0f
BAY		 14 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
043405e1
CO		 16 *
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
19 *
20 */
21
edf88417 22#include <linux/kvm_host.h>
313a3dc7 23#include "irq.h"
1d737c8a 24#include "mmu.h"
7837699f 25#include "i8254.h"
37817f29 26#include "tss.h"
5fdbf976 27#include "kvm_cache_regs.h"
26eef70c 28#include "x86.h"
00b27a3e 29#include "cpuid.h"
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;
a5a1d1c2
TG		 1134 u64 cycle_last;
1135 u64 mask;
16e8d74d
MT		 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
a5a1d1c2 1575static u64 read_tsc(void)
d828199e 1576{
a5a1d1c2 1577 u64 ret = (u64)rdtsc_ordered();
03b9730b 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
a5a1d1c2 1595static inline u64 vgettsc(u64 *cycle_now)
d828199e
MT		 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
a5a1d1c2 1606static int do_monotonic_boot(s64 *t, u64 *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 */
a5a1d1c2 1627static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *cycle_now)
d828199e 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{
cc0d907c 2847 int idx;
931f261b
AA		 2848 /*
2849 * Disable page faults because we're in atomic context here.
2850 * kvm_write_guest_offset_cached() would call might_fault()
2851 * that relies on pagefault_disable() to tell if there's a
2852 * bug. NOTE: the write to guest memory may not go through if
2853 * during postcopy live migration or if there's heavy guest
2854 * paging.
2855 */
2856 pagefault_disable();
cc0d907c
AA		 2857 /*
2858 * kvm_memslots() will be called by
2859 * kvm_write_guest_offset_cached() so take the srcu lock.
2860 */
2861 idx = srcu_read_lock(&vcpu->kvm->srcu);
0b9f6c46 2862 kvm_steal_time_set_preempted(vcpu);
cc0d907c 2863 srcu_read_unlock(&vcpu->kvm->srcu, idx);
931f261b 2864 pagefault_enable();
02daab21 2865 kvm_x86_ops->vcpu_put(vcpu);
1c11e713 2866 kvm_put_guest_fpu(vcpu);
4ea1636b 2867 vcpu->arch.last_host_tsc = rdtsc();
313a3dc7
CO		 2868}
2869
313a3dc7
CO		 2870static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2871 struct kvm_lapic_state *s)
2872{
d62caabb
AS		 2873 if (vcpu->arch.apicv_active)
2874 kvm_x86_ops->sync_pir_to_irr(vcpu);
2875
a92e2543 2876 return kvm_apic_get_state(vcpu, s);
313a3dc7
CO		 2877}
2878
2879static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2880 struct kvm_lapic_state *s)
2881{
a92e2543
RK		 2882 int r;
2883
2884 r = kvm_apic_set_state(vcpu, s);
2885 if (r)
2886 return r;
cb142eb7 2887 update_cr8_intercept(vcpu);
313a3dc7
CO		 2888
2889 return 0;
2890}
2891
127a457a
MG		 2892static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2893{
2894 return (!lapic_in_kernel(vcpu) ||
2895 kvm_apic_accept_pic_intr(vcpu));
2896}
2897
782d422b
MG		 2898/*
2899 * if userspace requested an interrupt window, check that the
2900 * interrupt window is open.
2901 *
2902 * No need to exit to userspace if we already have an interrupt queued.
2903 */
2904static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2905{
2906 return kvm_arch_interrupt_allowed(vcpu) &&
2907 !kvm_cpu_has_interrupt(vcpu) &&
2908 !kvm_event_needs_reinjection(vcpu) &&
2909 kvm_cpu_accept_dm_intr(vcpu);
2910}
2911
f77bc6a4
ZX		 2912static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2913 struct kvm_interrupt *irq)
2914{
02cdb50f 2915 if (irq->irq >= KVM_NR_INTERRUPTS)
f77bc6a4 2916 return -EINVAL;
1c1a9ce9
SR		 2917
2918 if (!irqchip_in_kernel(vcpu->kvm)) {
2919 kvm_queue_interrupt(vcpu, irq->irq, false);
2920 kvm_make_request(KVM_REQ_EVENT, vcpu);
2921 return 0;
2922 }
2923
2924 /*
2925 * With in-kernel LAPIC, we only use this to inject EXTINT, so
2926 * fail for in-kernel 8259.
2927 */
2928 if (pic_in_kernel(vcpu->kvm))
f77bc6a4 2929 return -ENXIO;
f77bc6a4 2930
1c1a9ce9
SR		 2931 if (vcpu->arch.pending_external_vector != -1)
2932 return -EEXIST;
f77bc6a4 2933
1c1a9ce9 2934 vcpu->arch.pending_external_vector = irq->irq;
934bf653 2935 kvm_make_request(KVM_REQ_EVENT, vcpu);
f77bc6a4
ZX		 2936 return 0;
2937}
2938
c4abb7c9
JK		 2939static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2940{
c4abb7c9 2941 kvm_inject_nmi(vcpu);
c4abb7c9
JK		 2942
2943 return 0;
2944}
2945
f077825a
PB		 2946static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2947{
64d60670
PB		 2948 kvm_make_request(KVM_REQ_SMI, vcpu);
2949
f077825a
PB		 2950 return 0;
2951}
2952
b209749f
AK		 2953static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2954 struct kvm_tpr_access_ctl *tac)
2955{
2956 if (tac->flags)
2957 return -EINVAL;
2958 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2959 return 0;
2960}
2961
890ca9ae
HY		 2962static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2963 u64 mcg_cap)
2964{
2965 int r;
2966 unsigned bank_num = mcg_cap & 0xff, bank;
2967
2968 r = -EINVAL;
a9e38c3e 2969 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
890ca9ae 2970 goto out;
c45dcc71 2971 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
890ca9ae
HY		 2972 goto out;
2973 r = 0;
2974 vcpu->arch.mcg_cap = mcg_cap;
2975 /* Init IA32_MCG_CTL to all 1s */
2976 if (mcg_cap & MCG_CTL_P)
2977 vcpu->arch.mcg_ctl = ~(u64)0;
2978 /* Init IA32_MCi_CTL to all 1s */
2979 for (bank = 0; bank < bank_num; bank++)
2980 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
c45dcc71
AR		 2981
2982 if (kvm_x86_ops->setup_mce)
2983 kvm_x86_ops->setup_mce(vcpu);
890ca9ae
HY		 2984out:
2985 return r;
2986}
2987
2988static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2989 struct kvm_x86_mce *mce)
2990{
2991 u64 mcg_cap = vcpu->arch.mcg_cap;
2992 unsigned bank_num = mcg_cap & 0xff;
2993 u64 *banks = vcpu->arch.mce_banks;
2994
2995 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2996 return -EINVAL;
2997 /*
2998 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2999 * reporting is disabled
3000 */
3001 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3002 vcpu->arch.mcg_ctl != ~(u64)0)
3003 return 0;
3004 banks += 4 * mce->bank;
3005 /*
3006 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3007 * reporting is disabled for the bank
3008 */
3009 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3010 return 0;
3011 if (mce->status & MCI_STATUS_UC) {
3012 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
fc78f519 3013 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
a8eeb04a 3014 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
890ca9ae
HY		 3015 return 0;
3016 }
3017 if (banks[1] & MCI_STATUS_VAL)
3018 mce->status |= MCI_STATUS_OVER;
3019 banks[2] = mce->addr;
3020 banks[3] = mce->misc;
3021 vcpu->arch.mcg_status = mce->mcg_status;
3022 banks[1] = mce->status;
3023 kvm_queue_exception(vcpu, MC_VECTOR);
3024 } else if (!(banks[1] & MCI_STATUS_VAL)
3025 || !(banks[1] & MCI_STATUS_UC)) {
3026 if (banks[1] & MCI_STATUS_VAL)
3027 mce->status |= MCI_STATUS_OVER;
3028 banks[2] = mce->addr;
3029 banks[3] = mce->misc;
3030 banks[1] = mce->status;
3031 } else
3032 banks[1] |= MCI_STATUS_OVER;
3033 return 0;
3034}
3035
3cfc3092
JK		 3036static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3037 struct kvm_vcpu_events *events)
3038{
7460fb4a 3039 process_nmi(vcpu);
03b82a30
JK		 3040 events->exception.injected =
3041 vcpu->arch.exception.pending &&
3042 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3cfc3092
JK		 3043 events->exception.nr = vcpu->arch.exception.nr;
3044 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
97e69aa6 3045 events->exception.pad = 0;
3cfc3092
JK		 3046 events->exception.error_code = vcpu->arch.exception.error_code;
3047
03b82a30
JK		 3048 events->interrupt.injected =
3049 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3cfc3092 3050 events->interrupt.nr = vcpu->arch.interrupt.nr;
03b82a30 3051 events->interrupt.soft = 0;
37ccdcbe 3052 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3cfc3092
JK		 3053
3054 events->nmi.injected = vcpu->arch.nmi_injected;
7460fb4a 3055 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3cfc3092 3056 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
97e69aa6 3057 events->nmi.pad = 0;
3cfc3092 3058
66450a21 3059 events->sipi_vector = 0; /* never valid when reporting to user space */
3cfc3092 3060
f077825a
PB		 3061 events->smi.smm = is_smm(vcpu);
3062 events->smi.pending = vcpu->arch.smi_pending;
3063 events->smi.smm_inside_nmi =
3064 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3065 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3066
dab4b911 3067 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
f077825a
PB		 3068 | KVM_VCPUEVENT_VALID_SHADOW
3069 | KVM_VCPUEVENT_VALID_SMM);
97e69aa6 3070 memset(&events->reserved, 0, sizeof(events->reserved));
3cfc3092
JK		 3071}
3072
6ef4e07e
XG		 3073static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3074
3cfc3092
JK		 3075static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3076 struct kvm_vcpu_events *events)
3077{
dab4b911 3078 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
48005f64 3079 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
f077825a
PB		 3080 | KVM_VCPUEVENT_VALID_SHADOW
3081 | KVM_VCPUEVENT_VALID_SMM))
3cfc3092
JK		 3082 return -EINVAL;
3083
78e546c8
PB		 3084 if (events->exception.injected &&
3085 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3086 return -EINVAL;
3087
767cefdd
DH		 3088 /* INITs are latched while in SMM */
3089 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3090 (events->smi.smm || events->smi.pending) &&
3091 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3092 return -EINVAL;
3093
7460fb4a 3094 process_nmi(vcpu);
3cfc3092
JK		 3095 vcpu->arch.exception.pending = events->exception.injected;
3096 vcpu->arch.exception.nr = events->exception.nr;
3097 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3098 vcpu->arch.exception.error_code = events->exception.error_code;
3099
3100 vcpu->arch.interrupt.pending = events->interrupt.injected;
3101 vcpu->arch.interrupt.nr = events->interrupt.nr;
3102 vcpu->arch.interrupt.soft = events->interrupt.soft;
48005f64
JK		 3103 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3104 kvm_x86_ops->set_interrupt_shadow(vcpu,
3105 events->interrupt.shadow);
3cfc3092
JK		 3106
3107 vcpu->arch.nmi_injected = events->nmi.injected;
dab4b911
JK		 3108 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3109 vcpu->arch.nmi_pending = events->nmi.pending;
3cfc3092
JK		 3110 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3111
66450a21 3112 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
bce87cce 3113 lapic_in_kernel(vcpu))
66450a21 3114 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3cfc3092 3115
f077825a 3116 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
6ef4e07e 3117 u32 hflags = vcpu->arch.hflags;
f077825a 3118 if (events->smi.smm)
6ef4e07e 3119 hflags |= HF_SMM_MASK;
f077825a 3120 else
6ef4e07e
XG		 3121 hflags &= ~HF_SMM_MASK;
3122 kvm_set_hflags(vcpu, hflags);
3123
f077825a
PB		 3124 vcpu->arch.smi_pending = events->smi.pending;
3125 if (events->smi.smm_inside_nmi)
3126 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3127 else
3128 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
bce87cce 3129 if (lapic_in_kernel(vcpu)) {
f077825a
PB		 3130 if (events->smi.latched_init)
3131 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3132 else
3133 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3134 }
3135 }
3136
3842d135
AK		 3137 kvm_make_request(KVM_REQ_EVENT, vcpu);
3138
3cfc3092
JK		 3139 return 0;
3140}
3141
a1efbe77
JK		 3142static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3143 struct kvm_debugregs *dbgregs)
3144{
73aaf249
JK		 3145 unsigned long val;
3146
a1efbe77 3147 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
16f8a6f9 3148 kvm_get_dr(vcpu, 6, &val);
73aaf249 3149 dbgregs->dr6 = val;
a1efbe77
JK		 3150 dbgregs->dr7 = vcpu->arch.dr7;
3151 dbgregs->flags = 0;
97e69aa6 3152 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
a1efbe77
JK		 3153}
3154
3155static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3156 struct kvm_debugregs *dbgregs)
3157{
3158 if (dbgregs->flags)
3159 return -EINVAL;
3160
d14bdb55
PB		 3161 if (dbgregs->dr6 & ~0xffffffffull)
3162 return -EINVAL;
3163 if (dbgregs->dr7 & ~0xffffffffull)
3164 return -EINVAL;
3165
a1efbe77 3166 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
ae561ede 3167 kvm_update_dr0123(vcpu);
a1efbe77 3168 vcpu->arch.dr6 = dbgregs->dr6;
73aaf249 3169 kvm_update_dr6(vcpu);
a1efbe77 3170 vcpu->arch.dr7 = dbgregs->dr7;
9926c9fd 3171 kvm_update_dr7(vcpu);
a1efbe77 3172
a1efbe77
JK		 3173 return 0;
3174}
3175
df1daba7
PB		 3176#define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3177
3178static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3179{
c47ada30 3180 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
400e4b20 3181 u64 xstate_bv = xsave->header.xfeatures;
df1daba7
PB		 3182 u64 valid;
3183
3184 /*
3185 * Copy legacy XSAVE area, to avoid complications with CPUID
3186 * leaves 0 and 1 in the loop below.
3187 */
3188 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3189
3190 /* Set XSTATE_BV */
00c87e9a 3191 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
df1daba7
PB		 3192 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3193
3194 /*
3195 * Copy each region from the possibly compacted offset to the
3196 * non-compacted offset.
3197 */
d91cab78 3198 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
df1daba7
PB		 3199 while (valid) {
3200 u64 feature = valid & -valid;
3201 int index = fls64(feature) - 1;
3202 void *src = get_xsave_addr(xsave, feature);
3203
3204 if (src) {
3205 u32 size, offset, ecx, edx;
3206 cpuid_count(XSTATE_CPUID, index,
3207 &size, &offset, &ecx, &edx);
3208 memcpy(dest + offset, src, size);
3209 }
3210
3211 valid -= feature;
3212 }
3213}
3214
3215static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3216{
c47ada30 3217 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
df1daba7
PB		 3218 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3219 u64 valid;
3220
3221 /*
3222 * Copy legacy XSAVE area, to avoid complications with CPUID
3223 * leaves 0 and 1 in the loop below.
3224 */
3225 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3226
3227 /* Set XSTATE_BV and possibly XCOMP_BV. */
400e4b20 3228 xsave->header.xfeatures = xstate_bv;
782511b0 3229 if (boot_cpu_has(X86_FEATURE_XSAVES))
3a54450b 3230 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
df1daba7
PB		 3231
3232 /*
3233 * Copy each region from the non-compacted offset to the
3234 * possibly compacted offset.
3235 */
d91cab78 3236 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
df1daba7
PB		 3237 while (valid) {
3238 u64 feature = valid & -valid;
3239 int index = fls64(feature) - 1;
3240 void *dest = get_xsave_addr(xsave, feature);
3241
3242 if (dest) {
3243 u32 size, offset, ecx, edx;
3244 cpuid_count(XSTATE_CPUID, index,
3245 &size, &offset, &ecx, &edx);
3246 memcpy(dest, src + offset, size);
ee4100da 3247 }
df1daba7
PB		 3248
3249 valid -= feature;
3250 }
3251}
3252
2d5b5a66
SY		 3253static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3254 struct kvm_xsave *guest_xsave)
3255{
d366bf7e 3256 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
df1daba7
PB		 3257 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3258 fill_xsave((u8 *) guest_xsave->region, vcpu);
4344ee98 3259 } else {
2d5b5a66 3260 memcpy(guest_xsave->region,
7366ed77 3261 &vcpu->arch.guest_fpu.state.fxsave,
c47ada30 3262 sizeof(struct fxregs_state));
2d5b5a66 3263 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
d91cab78 3264 XFEATURE_MASK_FPSSE;
2d5b5a66
SY		 3265 }
3266}
3267
3268static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3269 struct kvm_xsave *guest_xsave)
3270{
3271 u64 xstate_bv =
3272 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3273
d366bf7e 3274 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
d7876f1b
PB		 3275 /*
3276 * Here we allow setting states that are not present in
3277 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3278 * with old userspace.
3279 */
4ff41732 3280 if (xstate_bv & ~kvm_supported_xcr0())
d7876f1b 3281 return -EINVAL;
df1daba7 3282 load_xsave(vcpu, (u8 *)guest_xsave->region);
d7876f1b 3283 } else {
d91cab78 3284 if (xstate_bv & ~XFEATURE_MASK_FPSSE)
2d5b5a66 3285 return -EINVAL;
7366ed77 3286 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
c47ada30 3287 guest_xsave->region, sizeof(struct fxregs_state));
2d5b5a66
SY		 3288 }
3289 return 0;
3290}
3291
3292static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3293 struct kvm_xcrs *guest_xcrs)
3294{
d366bf7e 3295 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
2d5b5a66
SY		 3296 guest_xcrs->nr_xcrs = 0;
3297 return;
3298 }
3299
3300 guest_xcrs->nr_xcrs = 1;
3301 guest_xcrs->flags = 0;
3302 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3303 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3304}
3305
3306static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3307 struct kvm_xcrs *guest_xcrs)
3308{
3309 int i, r = 0;
3310
d366bf7e 3311 if (!boot_cpu_has(X86_FEATURE_XSAVE))
2d5b5a66
SY		 3312 return -EINVAL;
3313
3314 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3315 return -EINVAL;
3316
3317 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3318 /* Only support XCR0 currently */
c67a04cb 3319 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
2d5b5a66 3320 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
c67a04cb 3321 guest_xcrs->xcrs[i].value);
2d5b5a66
SY		 3322 break;
3323 }
3324 if (r)
3325 r = -EINVAL;
3326 return r;
3327}
3328
1c0b28c2
EM		 3329/*
3330 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3331 * stopped by the hypervisor. This function will be called from the host only.
3332 * EINVAL is returned when the host attempts to set the flag for a guest that
3333 * does not support pv clocks.
3334 */
3335static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3336{
0b79459b 3337 if (!vcpu->arch.pv_time_enabled)
1c0b28c2 3338 return -EINVAL;
51d59c6b 3339 vcpu->arch.pvclock_set_guest_stopped_request = true;
1c0b28c2
EM		 3340 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3341 return 0;
3342}
3343
5c919412
AS		 3344static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3345 struct kvm_enable_cap *cap)
3346{
3347 if (cap->flags)
3348 return -EINVAL;
3349
3350 switch (cap->cap) {
3351 case KVM_CAP_HYPERV_SYNIC:
546d87e5
WL		 3352 if (!irqchip_in_kernel(vcpu->kvm))
3353 return -EINVAL;
5c919412
AS		 3354 return kvm_hv_activate_synic(vcpu);
3355 default:
3356 return -EINVAL;
3357 }
3358}
3359
313a3dc7
CO		 3360long kvm_arch_vcpu_ioctl(struct file *filp,
3361 unsigned int ioctl, unsigned long arg)
3362{
3363 struct kvm_vcpu *vcpu = filp->private_data;
3364 void __user *argp = (void __user *)arg;
3365 int r;
d1ac91d8
AK		 3366 union {
3367 struct kvm_lapic_state *lapic;
3368 struct kvm_xsave *xsave;
3369 struct kvm_xcrs *xcrs;
3370 void *buffer;
3371 } u;
3372
3373 u.buffer = NULL;
313a3dc7
CO		 3374 switch (ioctl) {
3375 case KVM_GET_LAPIC: {
2204ae3c 3376 r = -EINVAL;
bce87cce 3377 if (!lapic_in_kernel(vcpu))
2204ae3c 3378 goto out;
d1ac91d8 3379 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
313a3dc7 3380
b772ff36 3381 r = -ENOMEM;
d1ac91d8 3382 if (!u.lapic)
b772ff36 3383 goto out;
d1ac91d8 3384 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
313a3dc7
CO		 3385 if (r)
3386 goto out;
3387 r = -EFAULT;
d1ac91d8 3388 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
313a3dc7
CO		 3389 goto out;
3390 r = 0;
3391 break;
3392 }
3393 case KVM_SET_LAPIC: {
2204ae3c 3394 r = -EINVAL;
bce87cce 3395 if (!lapic_in_kernel(vcpu))
2204ae3c 3396 goto out;
ff5c2c03 3397 u.lapic = memdup_user(argp, sizeof(*u.lapic));
18595411
GC		 3398 if (IS_ERR(u.lapic))
3399 return PTR_ERR(u.lapic);
ff5c2c03 3400
d1ac91d8 3401 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
313a3dc7
CO		 3402 break;
3403 }
f77bc6a4
ZX		 3404 case KVM_INTERRUPT: {
3405 struct kvm_interrupt irq;
3406
3407 r = -EFAULT;
3408 if (copy_from_user(&irq, argp, sizeof irq))
3409 goto out;
3410 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
f77bc6a4
ZX		 3411 break;
3412 }
c4abb7c9
JK		 3413 case KVM_NMI: {
3414 r = kvm_vcpu_ioctl_nmi(vcpu);
c4abb7c9
JK		 3415 break;
3416 }
f077825a
PB		 3417 case KVM_SMI: {
3418 r = kvm_vcpu_ioctl_smi(vcpu);
3419 break;
3420 }
313a3dc7
CO		 3421 case KVM_SET_CPUID: {
3422 struct kvm_cpuid __user *cpuid_arg = argp;
3423 struct kvm_cpuid cpuid;
3424
3425 r = -EFAULT;
3426 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3427 goto out;
3428 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
313a3dc7
CO		 3429 break;
3430 }
07716717
DK		 3431 case KVM_SET_CPUID2: {
3432 struct kvm_cpuid2 __user *cpuid_arg = argp;
3433 struct kvm_cpuid2 cpuid;
3434
3435 r = -EFAULT;
3436 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3437 goto out;
3438 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
19355475 3439 cpuid_arg->entries);
07716717
DK		 3440 break;
3441 }
3442 case KVM_GET_CPUID2: {
3443 struct kvm_cpuid2 __user *cpuid_arg = argp;
3444 struct kvm_cpuid2 cpuid;
3445
3446 r = -EFAULT;
3447 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3448 goto out;
3449 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
19355475 3450 cpuid_arg->entries);
07716717
DK		 3451 if (r)
3452 goto out;
3453 r = -EFAULT;
3454 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3455 goto out;
3456 r = 0;
3457 break;
3458 }
313a3dc7 3459 case KVM_GET_MSRS:
609e36d3 3460 r = msr_io(vcpu, argp, do_get_msr, 1);
313a3dc7
CO		 3461 break;
3462 case KVM_SET_MSRS:
3463 r = msr_io(vcpu, argp, do_set_msr, 0);
3464 break;
b209749f
AK		 3465 case KVM_TPR_ACCESS_REPORTING: {
3466 struct kvm_tpr_access_ctl tac;
3467
3468 r = -EFAULT;
3469 if (copy_from_user(&tac, argp, sizeof tac))
3470 goto out;
3471 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3472 if (r)
3473 goto out;
3474 r = -EFAULT;
3475 if (copy_to_user(argp, &tac, sizeof tac))
3476 goto out;
3477 r = 0;
3478 break;
3479 };
b93463aa
AK		 3480 case KVM_SET_VAPIC_ADDR: {
3481 struct kvm_vapic_addr va;
7301d6ab 3482 int idx;
b93463aa
AK		 3483
3484 r = -EINVAL;
35754c98 3485 if (!lapic_in_kernel(vcpu))
b93463aa
AK		 3486 goto out;
3487 r = -EFAULT;
3488 if (copy_from_user(&va, argp, sizeof va))
3489 goto out;
7301d6ab 3490 idx = srcu_read_lock(&vcpu->kvm->srcu);
fda4e2e8 3491 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
7301d6ab 3492 srcu_read_unlock(&vcpu->kvm->srcu, idx);
b93463aa
AK		 3493 break;
3494 }
890ca9ae
HY		 3495 case KVM_X86_SETUP_MCE: {
3496 u64 mcg_cap;
3497
3498 r = -EFAULT;
3499 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3500 goto out;
3501 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3502 break;
3503 }
3504 case KVM_X86_SET_MCE: {
3505 struct kvm_x86_mce mce;
3506
3507 r = -EFAULT;
3508 if (copy_from_user(&mce, argp, sizeof mce))
3509 goto out;
3510 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3511 break;
3512 }
3cfc3092
JK		 3513 case KVM_GET_VCPU_EVENTS: {
3514 struct kvm_vcpu_events events;
3515
3516 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3517
3518 r = -EFAULT;
3519 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3520 break;
3521 r = 0;
3522 break;
3523 }
3524 case KVM_SET_VCPU_EVENTS: {
3525 struct kvm_vcpu_events events;
3526
3527 r = -EFAULT;
3528 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3529 break;
3530
3531 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3532 break;
3533 }
a1efbe77
JK		 3534 case KVM_GET_DEBUGREGS: {
3535 struct kvm_debugregs dbgregs;
3536
3537 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3538
3539 r = -EFAULT;
3540 if (copy_to_user(argp, &dbgregs,
3541 sizeof(struct kvm_debugregs)))
3542 break;
3543 r = 0;
3544 break;
3545 }
3546 case KVM_SET_DEBUGREGS: {
3547 struct kvm_debugregs dbgregs;
3548
3549 r = -EFAULT;
3550 if (copy_from_user(&dbgregs, argp,
3551 sizeof(struct kvm_debugregs)))
3552 break;
3553
3554 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3555 break;
3556 }
2d5b5a66 3557 case KVM_GET_XSAVE: {
d1ac91d8 3558 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2d5b5a66 3559 r = -ENOMEM;
d1ac91d8 3560 if (!u.xsave)
2d5b5a66
SY		 3561 break;
3562
d1ac91d8 3563 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
2d5b5a66
SY		 3564
3565 r = -EFAULT;
d1ac91d8 3566 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
2d5b5a66
SY		 3567 break;
3568 r = 0;
3569 break;
3570 }
3571 case KVM_SET_XSAVE: {
ff5c2c03 3572 u.xsave = memdup_user(argp, sizeof(*u.xsave));
18595411
GC		 3573 if (IS_ERR(u.xsave))
3574 return PTR_ERR(u.xsave);
2d5b5a66 3575
d1ac91d8 3576 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
2d5b5a66
SY		 3577 break;
3578 }
3579 case KVM_GET_XCRS: {
d1ac91d8 3580 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2d5b5a66 3581 r = -ENOMEM;
d1ac91d8 3582 if (!u.xcrs)
2d5b5a66
SY		 3583 break;
3584
d1ac91d8 3585 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
2d5b5a66
SY		 3586
3587 r = -EFAULT;
d1ac91d8 3588 if (copy_to_user(argp, u.xcrs,
2d5b5a66
SY		 3589 sizeof(struct kvm_xcrs)))
3590 break;
3591 r = 0;
3592 break;
3593 }
3594 case KVM_SET_XCRS: {
ff5c2c03 3595 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
18595411
GC		 3596 if (IS_ERR(u.xcrs))
3597 return PTR_ERR(u.xcrs);
2d5b5a66 3598
d1ac91d8 3599 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
2d5b5a66
SY		 3600 break;
3601 }
92a1f12d
JR		 3602 case KVM_SET_TSC_KHZ: {
3603 u32 user_tsc_khz;
3604
3605 r = -EINVAL;
92a1f12d
JR		 3606 user_tsc_khz = (u32)arg;
3607
3608 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3609 goto out;
3610
cc578287
ZA		 3611 if (user_tsc_khz == 0)
3612 user_tsc_khz = tsc_khz;
3613
381d585c
HZ		 3614 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3615 r = 0;
92a1f12d 3616
92a1f12d
JR		 3617 goto out;
3618 }
3619 case KVM_GET_TSC_KHZ: {
cc578287 3620 r = vcpu->arch.virtual_tsc_khz;
92a1f12d
JR		 3621 goto out;
3622 }
1c0b28c2
EM		 3623 case KVM_KVMCLOCK_CTRL: {
3624 r = kvm_set_guest_paused(vcpu);
3625 goto out;
3626 }
5c919412
AS		 3627 case KVM_ENABLE_CAP: {
3628 struct kvm_enable_cap cap;
3629
3630 r = -EFAULT;
3631 if (copy_from_user(&cap, argp, sizeof(cap)))
3632 goto out;
3633 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3634 break;
3635 }
313a3dc7
CO		 3636 default:
3637 r = -EINVAL;
3638 }
3639out:
d1ac91d8 3640 kfree(u.buffer);
313a3dc7
CO		 3641 return r;
3642}
3643
5b1c1493
CO		 3644int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3645{
3646 return VM_FAULT_SIGBUS;
3647}
3648
1fe779f8
CO		 3649static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3650{
3651 int ret;
3652
3653 if (addr > (unsigned int)(-3 * PAGE_SIZE))
951179ce 3654 return -EINVAL;
1fe779f8
CO		 3655 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3656 return ret;
3657}
3658
b927a3ce
SY		 3659static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3660 u64 ident_addr)
3661{
3662 kvm->arch.ept_identity_map_addr = ident_addr;
3663 return 0;
3664}
3665
1fe779f8
CO		 3666static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3667 u32 kvm_nr_mmu_pages)
3668{
3669 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3670 return -EINVAL;
3671
79fac95e 3672 mutex_lock(&kvm->slots_lock);
1fe779f8
CO		 3673
3674 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
f05e70ac 3675 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1fe779f8 3676
79fac95e 3677 mutex_unlock(&kvm->slots_lock);
1fe779f8
CO		 3678 return 0;
3679}
3680
3681static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3682{
39de71ec 3683 return kvm->arch.n_max_mmu_pages;
1fe779f8
CO		 3684}
3685
1fe779f8
CO		 3686static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3687{
3688 int r;
3689
3690 r = 0;
3691 switch (chip->chip_id) {
3692 case KVM_IRQCHIP_PIC_MASTER:
3693 memcpy(&chip->chip.pic,
3694 &pic_irqchip(kvm)->pics[0],
3695 sizeof(struct kvm_pic_state));
3696 break;
3697 case KVM_IRQCHIP_PIC_SLAVE:
3698 memcpy(&chip->chip.pic,
3699 &pic_irqchip(kvm)->pics[1],
3700 sizeof(struct kvm_pic_state));
3701 break;
3702 case KVM_IRQCHIP_IOAPIC:
eba0226b 3703 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
1fe779f8
CO		 3704 break;
3705 default:
3706 r = -EINVAL;
3707 break;
3708 }
3709 return r;
3710}
3711
3712static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3713{
3714 int r;
3715
3716 r = 0;
3717 switch (chip->chip_id) {
3718 case KVM_IRQCHIP_PIC_MASTER:
f4f51050 3719 spin_lock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3720 memcpy(&pic_irqchip(kvm)->pics[0],
3721 &chip->chip.pic,
3722 sizeof(struct kvm_pic_state));
f4f51050 3723 spin_unlock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3724 break;
3725 case KVM_IRQCHIP_PIC_SLAVE:
f4f51050 3726 spin_lock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3727 memcpy(&pic_irqchip(kvm)->pics[1],
3728 &chip->chip.pic,
3729 sizeof(struct kvm_pic_state));
f4f51050 3730 spin_unlock(&pic_irqchip(kvm)->lock);
1fe779f8
CO		 3731 break;
3732 case KVM_IRQCHIP_IOAPIC:
eba0226b 3733 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
1fe779f8
CO		 3734 break;
3735 default:
3736 r = -EINVAL;
3737 break;
3738 }
3739 kvm_pic_update_irq(pic_irqchip(kvm));
3740 return r;
3741}
3742
e0f63cb9
SY		 3743static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3744{
34f3941c
RK		 3745 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3746
3747 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3748
3749 mutex_lock(&kps->lock);
3750 memcpy(ps, &kps->channels, sizeof(*ps));
3751 mutex_unlock(&kps->lock);
2da29bcc 3752 return 0;
e0f63cb9
SY		 3753}
3754
3755static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3756{
0185604c 3757 int i;
09edea72
RK		 3758 struct kvm_pit *pit = kvm->arch.vpit;
3759
3760 mutex_lock(&pit->pit_state.lock);
34f3941c 3761 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
0185604c 3762 for (i = 0; i < 3; i++)
09edea72
RK		 3763 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3764 mutex_unlock(&pit->pit_state.lock);
2da29bcc 3765 return 0;
e9f42757
BK		 3766}
3767
3768static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3769{
e9f42757
BK		 3770 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3771 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3772 sizeof(ps->channels));
3773 ps->flags = kvm->arch.vpit->pit_state.flags;
3774 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
97e69aa6 3775 memset(&ps->reserved, 0, sizeof(ps->reserved));
2da29bcc 3776 return 0;
e9f42757
BK		 3777}
3778
3779static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3780{
2da29bcc 3781 int start = 0;
0185604c 3782 int i;
e9f42757 3783 u32 prev_legacy, cur_legacy;
09edea72
RK		 3784 struct kvm_pit *pit = kvm->arch.vpit;
3785
3786 mutex_lock(&pit->pit_state.lock);
3787 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
e9f42757
BK		 3788 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3789 if (!prev_legacy && cur_legacy)
3790 start = 1;
09edea72
RK		 3791 memcpy(&pit->pit_state.channels, &ps->channels,
3792 sizeof(pit->pit_state.channels));
3793 pit->pit_state.flags = ps->flags;
0185604c 3794 for (i = 0; i < 3; i++)
09edea72 3795 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
e5e57e7a 3796 start && i == 0);
09edea72 3797 mutex_unlock(&pit->pit_state.lock);
2da29bcc 3798 return 0;
e0f63cb9
SY		 3799}
3800
52d939a0
MT		 3801static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3802 struct kvm_reinject_control *control)
3803{
71474e2f
RK		 3804 struct kvm_pit *pit = kvm->arch.vpit;
3805
3806 if (!pit)
52d939a0 3807 return -ENXIO;
b39c90b6 3808
71474e2f
RK		 3809 /* pit->pit_state.lock was overloaded to prevent userspace from getting
3810 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3811 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
3812 */
3813 mutex_lock(&pit->pit_state.lock);
3814 kvm_pit_set_reinject(pit, control->pit_reinject);
3815 mutex_unlock(&pit->pit_state.lock);
b39c90b6 3816
52d939a0
MT		 3817 return 0;
3818}
3819
95d4c16c 3820/**
60c34612
TY		 3821 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3822 * @kvm: kvm instance
3823 * @log: slot id and address to which we copy the log
95d4c16c 3824 *
e108ff2f
PB		 3825 * Steps 1-4 below provide general overview of dirty page logging. See
3826 * kvm_get_dirty_log_protect() function description for additional details.
3827 *
3828 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3829 * always flush the TLB (step 4) even if previous step failed and the dirty
3830 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3831 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3832 * writes will be marked dirty for next log read.
95d4c16c 3833 *
60c34612
TY		 3834 * 1. Take a snapshot of the bit and clear it if needed.
3835 * 2. Write protect the corresponding page.
e108ff2f
PB		 3836 * 3. Copy the snapshot to the userspace.
3837 * 4. Flush TLB's if needed.
5bb064dc 3838 */
60c34612 3839int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
5bb064dc 3840{
60c34612 3841 bool is_dirty = false;
e108ff2f 3842 int r;
5bb064dc 3843
79fac95e 3844 mutex_lock(&kvm->slots_lock);
5bb064dc 3845
88178fd4
KH		 3846 /*
3847 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3848 */
3849 if (kvm_x86_ops->flush_log_dirty)
3850 kvm_x86_ops->flush_log_dirty(kvm);
3851
e108ff2f 3852 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
198c74f4
XG		 3853
3854 /*
3855 * All the TLBs can be flushed out of mmu lock, see the comments in
3856 * kvm_mmu_slot_remove_write_access().
3857 */
e108ff2f 3858 lockdep_assert_held(&kvm->slots_lock);
198c74f4
XG		 3859 if (is_dirty)
3860 kvm_flush_remote_tlbs(kvm);
3861
79fac95e 3862 mutex_unlock(&kvm->slots_lock);
5bb064dc
ZX		 3863 return r;
3864}
3865
aa2fbe6d
YZ		 3866int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3867 bool line_status)
23d43cf9
CD		 3868{
3869 if (!irqchip_in_kernel(kvm))
3870 return -ENXIO;
3871
3872 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
aa2fbe6d
YZ		 3873 irq_event->irq, irq_event->level,
3874 line_status);
23d43cf9
CD		 3875 return 0;
3876}
3877
90de4a18
NA		 3878static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3879 struct kvm_enable_cap *cap)
3880{
3881 int r;
3882
3883 if (cap->flags)
3884 return -EINVAL;
3885
3886 switch (cap->cap) {
3887 case KVM_CAP_DISABLE_QUIRKS:
3888 kvm->arch.disabled_quirks = cap->args[0];
3889 r = 0;
3890 break;
49df6397
SR		 3891 case KVM_CAP_SPLIT_IRQCHIP: {
3892 mutex_lock(&kvm->lock);
b053b2ae
SR		 3893 r = -EINVAL;
3894 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3895 goto split_irqchip_unlock;
49df6397
SR		 3896 r = -EEXIST;
3897 if (irqchip_in_kernel(kvm))
3898 goto split_irqchip_unlock;
557abc40 3899 if (kvm->created_vcpus)
49df6397
SR		 3900 goto split_irqchip_unlock;
3901 r = kvm_setup_empty_irq_routing(kvm);
3902 if (r)
3903 goto split_irqchip_unlock;
3904 /* Pairs with irqchip_in_kernel. */
3905 smp_wmb();
3906 kvm->arch.irqchip_split = true;
b053b2ae 3907 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
49df6397
SR		 3908 r = 0;
3909split_irqchip_unlock:
3910 mutex_unlock(&kvm->lock);
3911 break;
3912 }
37131313
RK		 3913 case KVM_CAP_X2APIC_API:
3914 r = -EINVAL;
3915 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
3916 break;
3917
3918 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
3919 kvm->arch.x2apic_format = true;
c519265f
RK		 3920 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
3921 kvm->arch.x2apic_broadcast_quirk_disabled = true;
37131313
RK		 3922
3923 r = 0;
3924 break;
90de4a18
NA		 3925 default:
3926 r = -EINVAL;
3927 break;
3928 }
3929 return r;
3930}
3931
1fe779f8
CO		 3932long kvm_arch_vm_ioctl(struct file *filp,
3933 unsigned int ioctl, unsigned long arg)
3934{
3935 struct kvm *kvm = filp->private_data;
3936 void __user *argp = (void __user *)arg;
367e1319 3937 int r = -ENOTTY;
f0d66275
DH		 3938 /*
3939 * This union makes it completely explicit to gcc-3.x
3940 * that these two variables' stack usage should be
3941 * combined, not added together.
3942 */
3943 union {
3944 struct kvm_pit_state ps;
e9f42757 3945 struct kvm_pit_state2 ps2;
c5ff41ce 3946 struct kvm_pit_config pit_config;
f0d66275 3947 } u;
1fe779f8
CO		 3948
3949 switch (ioctl) {
3950 case KVM_SET_TSS_ADDR:
3951 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1fe779f8 3952 break;
b927a3ce
SY		 3953 case KVM_SET_IDENTITY_MAP_ADDR: {
3954 u64 ident_addr;
3955
3956 r = -EFAULT;
3957 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3958 goto out;
3959 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
b927a3ce
SY		 3960 break;
3961 }
1fe779f8
CO		 3962 case KVM_SET_NR_MMU_PAGES:
3963 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1fe779f8
CO		 3964 break;
3965 case KVM_GET_NR_MMU_PAGES:
3966 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3967 break;
3ddea128
MT		 3968 case KVM_CREATE_IRQCHIP: {
3969 struct kvm_pic *vpic;
3970
3971 mutex_lock(&kvm->lock);
3972 r = -EEXIST;
3973 if (kvm->arch.vpic)
3974 goto create_irqchip_unlock;
3e515705 3975 r = -EINVAL;
557abc40 3976 if (kvm->created_vcpus)
3e515705 3977 goto create_irqchip_unlock;
1fe779f8 3978 r = -ENOMEM;
3ddea128
MT		 3979 vpic = kvm_create_pic(kvm);
3980 if (vpic) {
1fe779f8
CO		 3981 r = kvm_ioapic_init(kvm);
3982 if (r) {
175504cd 3983 mutex_lock(&kvm->slots_lock);
71ba994c 3984 kvm_destroy_pic(vpic);
175504cd 3985 mutex_unlock(&kvm->slots_lock);
3ddea128 3986 goto create_irqchip_unlock;
1fe779f8
CO		 3987 }
3988 } else
3ddea128 3989 goto create_irqchip_unlock;
399ec807
AK		 3990 r = kvm_setup_default_irq_routing(kvm);
3991 if (r) {
175504cd 3992 mutex_lock(&kvm->slots_lock);
3ddea128 3993 mutex_lock(&kvm->irq_lock);
72bb2fcd 3994 kvm_ioapic_destroy(kvm);
71ba994c 3995 kvm_destroy_pic(vpic);
3ddea128 3996 mutex_unlock(&kvm->irq_lock);
175504cd 3997 mutex_unlock(&kvm->slots_lock);
71ba994c 3998 goto create_irqchip_unlock;
399ec807 3999 }
71ba994c
PB		 4000 /* Write kvm->irq_routing before kvm->arch.vpic. */
4001 smp_wmb();
4002 kvm->arch.vpic = vpic;
3ddea128
MT		 4003 create_irqchip_unlock:
4004 mutex_unlock(&kvm->lock);
1fe779f8 4005 break;
3ddea128 4006 }
7837699f 4007 case KVM_CREATE_PIT:
c5ff41ce
JK		 4008 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4009 goto create_pit;
4010 case KVM_CREATE_PIT2:
4011 r = -EFAULT;
4012 if (copy_from_user(&u.pit_config, argp,
4013 sizeof(struct kvm_pit_config)))
4014 goto out;
4015 create_pit:
250715a6 4016 mutex_lock(&kvm->lock);
269e05e4
AK		 4017 r = -EEXIST;
4018 if (kvm->arch.vpit)
4019 goto create_pit_unlock;
7837699f 4020 r = -ENOMEM;
c5ff41ce 4021 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
7837699f
SY		 4022 if (kvm->arch.vpit)
4023 r = 0;
269e05e4 4024 create_pit_unlock:
250715a6 4025 mutex_unlock(&kvm->lock);
7837699f 4026 break;
1fe779f8
CO		 4027 case KVM_GET_IRQCHIP: {
4028 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
ff5c2c03 4029 struct kvm_irqchip *chip;
1fe779f8 4030
ff5c2c03
SL		 4031 chip = memdup_user(argp, sizeof(*chip));
4032 if (IS_ERR(chip)) {
4033 r = PTR_ERR(chip);
1fe779f8 4034 goto out;
ff5c2c03
SL		 4035 }
4036
1fe779f8 4037 r = -ENXIO;
49df6397 4038 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
f0d66275
DH		 4039 goto get_irqchip_out;
4040 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1fe779f8 4041 if (r)
f0d66275 4042 goto get_irqchip_out;
1fe779f8 4043 r = -EFAULT;
f0d66275
DH		 4044 if (copy_to_user(argp, chip, sizeof *chip))
4045 goto get_irqchip_out;
1fe779f8 4046 r = 0;
f0d66275
DH		 4047 get_irqchip_out:
4048 kfree(chip);
1fe779f8
CO		 4049 break;
4050 }
4051 case KVM_SET_IRQCHIP: {
4052 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
ff5c2c03 4053 struct kvm_irqchip *chip;
1fe779f8 4054
ff5c2c03
SL		 4055 chip = memdup_user(argp, sizeof(*chip));
4056 if (IS_ERR(chip)) {
4057 r = PTR_ERR(chip);
1fe779f8 4058 goto out;
ff5c2c03
SL		 4059 }
4060
1fe779f8 4061 r = -ENXIO;
49df6397 4062 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
f0d66275
DH		 4063 goto set_irqchip_out;
4064 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1fe779f8 4065 if (r)
f0d66275 4066 goto set_irqchip_out;
1fe779f8 4067 r = 0;
f0d66275
DH		 4068 set_irqchip_out:
4069 kfree(chip);
1fe779f8
CO		 4070 break;
4071 }
e0f63cb9 4072 case KVM_GET_PIT: {
e0f63cb9 4073 r = -EFAULT;
f0d66275 4074 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
e0f63cb9
SY		 4075 goto out;
4076 r = -ENXIO;
4077 if (!kvm->arch.vpit)
4078 goto out;
f0d66275 4079 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
e0f63cb9
SY		 4080 if (r)
4081 goto out;
4082 r = -EFAULT;
f0d66275 4083 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
e0f63cb9
SY		 4084 goto out;
4085 r = 0;
4086 break;
4087 }
4088 case KVM_SET_PIT: {
e0f63cb9 4089 r = -EFAULT;
f0d66275 4090 if (copy_from_user(&u.ps, argp, sizeof u.ps))
e0f63cb9
SY		 4091 goto out;
4092 r = -ENXIO;
4093 if (!kvm->arch.vpit)
4094 goto out;
f0d66275 4095 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
e0f63cb9
SY		 4096 break;
4097 }
e9f42757
BK		 4098 case KVM_GET_PIT2: {
4099 r = -ENXIO;
4100 if (!kvm->arch.vpit)
4101 goto out;
4102 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4103 if (r)
4104 goto out;
4105 r = -EFAULT;
4106 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4107 goto out;
4108 r = 0;
4109 break;
4110 }
4111 case KVM_SET_PIT2: {
4112 r = -EFAULT;
4113 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4114 goto out;
4115 r = -ENXIO;
4116 if (!kvm->arch.vpit)
4117 goto out;
4118 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
e9f42757
BK		 4119 break;
4120 }
52d939a0
MT		 4121 case KVM_REINJECT_CONTROL: {
4122 struct kvm_reinject_control control;
4123 r = -EFAULT;
4124 if (copy_from_user(&control, argp, sizeof(control)))
4125 goto out;
4126 r = kvm_vm_ioctl_reinject(kvm, &control);
52d939a0
MT		 4127 break;
4128 }
d71ba788
PB		 4129 case KVM_SET_BOOT_CPU_ID:
4130 r = 0;
4131 mutex_lock(&kvm->lock);
557abc40 4132 if (kvm->created_vcpus)
d71ba788
PB		 4133 r = -EBUSY;
4134 else
4135 kvm->arch.bsp_vcpu_id = arg;
4136 mutex_unlock(&kvm->lock);
4137 break;
ffde22ac
ES		 4138 case KVM_XEN_HVM_CONFIG: {
4139 r = -EFAULT;
4140 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4141 sizeof(struct kvm_xen_hvm_config)))
4142 goto out;
4143 r = -EINVAL;
4144 if (kvm->arch.xen_hvm_config.flags)
4145 goto out;
4146 r = 0;
4147 break;
4148 }
afbcf7ab 4149 case KVM_SET_CLOCK: {
afbcf7ab
GC		 4150 struct kvm_clock_data user_ns;
4151 u64 now_ns;
afbcf7ab
GC		 4152
4153 r = -EFAULT;
4154 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4155 goto out;
4156
4157 r = -EINVAL;
4158 if (user_ns.flags)
4159 goto out;
4160
4161 r = 0;
395c6b0a 4162 local_irq_disable();
108b249c
PB		 4163 now_ns = __get_kvmclock_ns(kvm);
4164 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
395c6b0a 4165 local_irq_enable();
2e762ff7 4166 kvm_gen_update_masterclock(kvm);
afbcf7ab
GC		 4167 break;
4168 }
4169 case KVM_GET_CLOCK: {
afbcf7ab
GC		 4170 struct kvm_clock_data user_ns;
4171 u64 now_ns;
4172
e3fd9a93
PB		 4173 local_irq_disable();
4174 now_ns = __get_kvmclock_ns(kvm);
108b249c 4175 user_ns.clock = now_ns;
e3fd9a93
PB		 4176 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4177 local_irq_enable();
97e69aa6 4178 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
afbcf7ab
GC		 4179
4180 r = -EFAULT;
4181 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4182 goto out;
4183 r = 0;
4184 break;
4185 }
90de4a18
NA		 4186 case KVM_ENABLE_CAP: {
4187 struct kvm_enable_cap cap;
afbcf7ab 4188
90de4a18
NA		 4189 r = -EFAULT;
4190 if (copy_from_user(&cap, argp, sizeof(cap)))
4191 goto out;
4192 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4193 break;
4194 }
1fe779f8 4195 default:
c274e03a 4196 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1fe779f8
CO		 4197 }
4198out:
4199 return r;
4200}
4201
a16b043c 4202static void kvm_init_msr_list(void)
043405e1
CO		 4203{
4204 u32 dummy[2];
4205 unsigned i, j;
4206
62ef68bb 4207 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
043405e1
CO		 4208 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4209 continue;
93c4adc7
PB		 4210
4211 /*
4212 * Even MSRs that are valid in the host may not be exposed
9dbe6cf9 4213 * to the guests in some cases.
93c4adc7
PB		 4214 */
4215 switch (msrs_to_save[i]) {
4216 case MSR_IA32_BNDCFGS:
4217 if (!kvm_x86_ops->mpx_supported())
4218 continue;
4219 break;
9dbe6cf9
PB		 4220 case MSR_TSC_AUX:
4221 if (!kvm_x86_ops->rdtscp_supported())
4222 continue;
4223 break;
93c4adc7
PB		 4224 default:
4225 break;
4226 }
4227
043405e1
CO		 4228 if (j < i)
4229 msrs_to_save[j] = msrs_to_save[i];
4230 j++;
4231 }
4232 num_msrs_to_save = j;
62ef68bb
PB		 4233
4234 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4235 switch (emulated_msrs[i]) {
6d396b55
PB		 4236 case MSR_IA32_SMBASE:
4237 if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4238 continue;
4239 break;
62ef68bb
PB		 4240 default:
4241 break;
4242 }
4243
4244 if (j < i)
4245 emulated_msrs[j] = emulated_msrs[i];
4246 j++;
4247 }
4248 num_emulated_msrs = j;
043405e1
CO		 4249}
4250
bda9020e
MT		 4251static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4252 const void *v)
bbd9b64e 4253{
70252a10
AK		 4254 int handled = 0;
4255 int n;
4256
4257 do {
4258 n = min(len, 8);
bce87cce 4259 if (!(lapic_in_kernel(vcpu) &&
e32edf4f
NN		 4260 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4261 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
70252a10
AK		 4262 break;
4263 handled += n;
4264 addr += n;
4265 len -= n;
4266 v += n;
4267 } while (len);
bbd9b64e 4268
70252a10 4269 return handled;
bbd9b64e
CO		 4270}
4271
bda9020e 4272static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
bbd9b64e 4273{
70252a10
AK		 4274 int handled = 0;
4275 int n;
4276
4277 do {
4278 n = min(len, 8);
bce87cce 4279 if (!(lapic_in_kernel(vcpu) &&
e32edf4f
NN		 4280 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4281 addr, n, v))
4282 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
70252a10
AK		 4283 break;
4284 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4285 handled += n;
4286 addr += n;
4287 len -= n;
4288 v += n;
4289 } while (len);
bbd9b64e 4290
70252a10 4291 return handled;
bbd9b64e
CO		 4292}
4293
2dafc6c2
GN		 4294static void kvm_set_segment(struct kvm_vcpu *vcpu,
4295 struct kvm_segment *var, int seg)
4296{
4297 kvm_x86_ops->set_segment(vcpu, var, seg);
4298}
4299
4300void kvm_get_segment(struct kvm_vcpu *vcpu,
4301 struct kvm_segment *var, int seg)
4302{
4303 kvm_x86_ops->get_segment(vcpu, var, seg);
4304}
4305
54987b7a
PB		 4306gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4307 struct x86_exception *exception)
02f59dc9
JR		 4308{
4309 gpa_t t_gpa;
02f59dc9
JR		 4310
4311 BUG_ON(!mmu_is_nested(vcpu));
4312
4313 /* NPT walks are always user-walks */
4314 access |= PFERR_USER_MASK;
54987b7a 4315 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
02f59dc9
JR		 4316
4317 return t_gpa;
4318}
4319
ab9ae313
AK		 4320gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4321 struct x86_exception *exception)
1871c602
GN		 4322{
4323 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
ab9ae313 4324 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 4325}
4326
ab9ae313
AK		 4327 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4328 struct x86_exception *exception)
1871c602
GN		 4329{
4330 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4331 access |= PFERR_FETCH_MASK;
ab9ae313 4332 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 4333}
4334
ab9ae313
AK		 4335gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4336 struct x86_exception *exception)
1871c602
GN		 4337{
4338 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4339 access |= PFERR_WRITE_MASK;
ab9ae313 4340 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 4341}
4342
4343/* uses this to access any guest's mapped memory without checking CPL */
ab9ae313
AK		 4344gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4345 struct x86_exception *exception)
1871c602 4346{
ab9ae313 4347 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
1871c602
GN		 4348}
4349
4350static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4351 struct kvm_vcpu *vcpu, u32 access,
bcc55cba 4352 struct x86_exception *exception)
bbd9b64e
CO		 4353{
4354 void *data = val;
10589a46 4355 int r = X86EMUL_CONTINUE;
bbd9b64e
CO		 4356
4357 while (bytes) {
14dfe855 4358 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
ab9ae313 4359 exception);
bbd9b64e 4360 unsigned offset = addr & (PAGE_SIZE-1);
77c2002e 4361 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
bbd9b64e
CO		 4362 int ret;
4363
bcc55cba 4364 if (gpa == UNMAPPED_GVA)
ab9ae313 4365 return X86EMUL_PROPAGATE_FAULT;
54bf36aa
PB		 4366 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4367 offset, toread);
10589a46 4368 if (ret < 0) {
c3cd7ffa 4369 r = X86EMUL_IO_NEEDED;
10589a46
MT		 4370 goto out;
4371 }
bbd9b64e 4372
77c2002e
IE		 4373 bytes -= toread;
4374 data += toread;
4375 addr += toread;
bbd9b64e 4376 }
10589a46 4377out:
10589a46 4378 return r;
bbd9b64e 4379}
77c2002e 4380
1871c602 4381/* used for instruction fetching */
0f65dd70
AK		 4382static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4383 gva_t addr, void *val, unsigned int bytes,
bcc55cba 4384 struct x86_exception *exception)
1871c602 4385{
0f65dd70 4386 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
1871c602 4387 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
44583cba
PB		 4388 unsigned offset;
4389 int ret;
0f65dd70 4390
44583cba
PB		 4391 /* Inline kvm_read_guest_virt_helper for speed. */
4392 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4393 exception);
4394 if (unlikely(gpa == UNMAPPED_GVA))
4395 return X86EMUL_PROPAGATE_FAULT;
4396
4397 offset = addr & (PAGE_SIZE-1);
4398 if (WARN_ON(offset + bytes > PAGE_SIZE))
4399 bytes = (unsigned)PAGE_SIZE - offset;
54bf36aa
PB		 4400 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4401 offset, bytes);
44583cba
PB		 4402 if (unlikely(ret < 0))
4403 return X86EMUL_IO_NEEDED;
4404
4405 return X86EMUL_CONTINUE;
1871c602
GN		 4406}
4407
064aea77 4408int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
0f65dd70 4409 gva_t addr, void *val, unsigned int bytes,
bcc55cba 4410 struct x86_exception *exception)
1871c602 4411{
0f65dd70 4412 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
1871c602 4413 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
0f65dd70 4414
1871c602 4415 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
bcc55cba 4416 exception);
1871c602 4417}
064aea77 4418EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
1871c602 4419
0f65dd70
AK		 4420static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4421 gva_t addr, void *val, unsigned int bytes,
bcc55cba 4422 struct x86_exception *exception)
1871c602 4423{
0f65dd70 4424 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
bcc55cba 4425 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
1871c602
GN		 4426}
4427
7a036a6f
RK		 4428static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4429 unsigned long addr, void *val, unsigned int bytes)
4430{
4431 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4432 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4433
4434 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4435}
4436
6a4d7550 4437int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
0f65dd70 4438 gva_t addr, void *val,
2dafc6c2 4439 unsigned int bytes,
bcc55cba 4440 struct x86_exception *exception)
77c2002e 4441{
0f65dd70 4442 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
77c2002e
IE		 4443 void *data = val;
4444 int r = X86EMUL_CONTINUE;
4445
4446 while (bytes) {
14dfe855
JR		 4447 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4448 PFERR_WRITE_MASK,
ab9ae313 4449 exception);
77c2002e
IE		 4450 unsigned offset = addr & (PAGE_SIZE-1);
4451 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4452 int ret;
4453
bcc55cba 4454 if (gpa == UNMAPPED_GVA)
ab9ae313 4455 return X86EMUL_PROPAGATE_FAULT;
54bf36aa 4456 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
77c2002e 4457 if (ret < 0) {
c3cd7ffa 4458 r = X86EMUL_IO_NEEDED;
77c2002e
IE		 4459 goto out;
4460 }
4461
4462 bytes -= towrite;
4463 data += towrite;
4464 addr += towrite;
4465 }
4466out:
4467 return r;
4468}
6a4d7550 4469EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
77c2002e 4470
af7cc7d1
XG		 4471static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4472 gpa_t *gpa, struct x86_exception *exception,
4473 bool write)
4474{
97d64b78
AK		 4475 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4476 | (write ? PFERR_WRITE_MASK : 0);
af7cc7d1 4477
be94f6b7
HH		 4478 /*
4479 * currently PKRU is only applied to ept enabled guest so
4480 * there is no pkey in EPT page table for L1 guest or EPT
4481 * shadow page table for L2 guest.
4482 */
97d64b78 4483 if (vcpu_match_mmio_gva(vcpu, gva)
97ec8c06 4484 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
be94f6b7 4485 vcpu->arch.access, 0, access)) {
bebb106a
XG		 4486 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4487 (gva & (PAGE_SIZE - 1));
4f022648 4488 trace_vcpu_match_mmio(gva, *gpa, write, false);
bebb106a
XG		 4489 return 1;
4490 }
4491
af7cc7d1
XG		 4492 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4493
4494 if (*gpa == UNMAPPED_GVA)
4495 return -1;
4496
4497 /* For APIC access vmexit */
4498 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4499 return 1;
4500
4f022648
XG		 4501 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4502 trace_vcpu_match_mmio(gva, *gpa, write, true);
bebb106a 4503 return 1;
4f022648 4504 }
bebb106a 4505
af7cc7d1
XG		 4506 return 0;
4507}
4508
3200f405 4509int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
bcc55cba 4510 const void *val, int bytes)
bbd9b64e
CO		 4511{
4512 int ret;
4513
54bf36aa 4514 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
9f811285 4515 if (ret < 0)
bbd9b64e 4516 return 0;
0eb05bf2 4517 kvm_page_track_write(vcpu, gpa, val, bytes);
bbd9b64e
CO		 4518 return 1;
4519}
4520
77d197b2
XG		 4521struct read_write_emulator_ops {
4522 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4523 int bytes);
4524 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4525 void *val, int bytes);
4526 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4527 int bytes, void *val);
4528 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4529 void *val, int bytes);
4530 bool write;
4531};
4532
4533static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4534{
4535 if (vcpu->mmio_read_completed) {
77d197b2 4536 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
f78146b0 4537 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
77d197b2
XG		 4538 vcpu->mmio_read_completed = 0;
4539 return 1;
4540 }
4541
4542 return 0;
4543}
4544
4545static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4546 void *val, int bytes)
4547{
54bf36aa 4548 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
77d197b2
XG		 4549}
4550
4551static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4552 void *val, int bytes)
4553{
4554 return emulator_write_phys(vcpu, gpa, val, bytes);
4555}
4556
4557static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4558{
4559 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4560 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4561}
4562
4563static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4564 void *val, int bytes)
4565{
4566 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4567 return X86EMUL_IO_NEEDED;
4568}
4569
4570static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4571 void *val, int bytes)
4572{
f78146b0
AK		 4573 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4574
87da7e66 4575 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
77d197b2
XG		 4576 return X86EMUL_CONTINUE;
4577}
4578
0fbe9b0b 4579static const struct read_write_emulator_ops read_emultor = {
77d197b2
XG		 4580 .read_write_prepare = read_prepare,
4581 .read_write_emulate = read_emulate,
4582 .read_write_mmio = vcpu_mmio_read,
4583 .read_write_exit_mmio = read_exit_mmio,
4584};
4585
0fbe9b0b 4586static const struct read_write_emulator_ops write_emultor = {
77d197b2
XG		 4587 .read_write_emulate = write_emulate,
4588 .read_write_mmio = write_mmio,
4589 .read_write_exit_mmio = write_exit_mmio,
4590 .write = true,
4591};
4592
22388a3c
XG		 4593static int emulator_read_write_onepage(unsigned long addr, void *val,
4594 unsigned int bytes,
4595 struct x86_exception *exception,
4596 struct kvm_vcpu *vcpu,
0fbe9b0b 4597 const struct read_write_emulator_ops *ops)
bbd9b64e 4598{
af7cc7d1
XG		 4599 gpa_t gpa;
4600 int handled, ret;
22388a3c 4601 bool write = ops->write;
f78146b0 4602 struct kvm_mmio_fragment *frag;
10589a46 4603
22388a3c 4604 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
bbd9b64e 4605
af7cc7d1 4606 if (ret < 0)
bbd9b64e 4607 return X86EMUL_PROPAGATE_FAULT;
bbd9b64e
CO		 4608
4609 /* For APIC access vmexit */
af7cc7d1 4610 if (ret)
bbd9b64e
CO		 4611 goto mmio;
4612
22388a3c 4613 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
bbd9b64e
CO		 4614 return X86EMUL_CONTINUE;
4615
4616mmio:
4617 /*
4618 * Is this MMIO handled locally?
4619 */
22388a3c 4620 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
70252a10 4621 if (handled == bytes)
bbd9b64e 4622 return X86EMUL_CONTINUE;
bbd9b64e 4623
70252a10
AK		 4624 gpa += handled;
4625 bytes -= handled;
4626 val += handled;
4627
87da7e66
XG		 4628 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4629 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4630 frag->gpa = gpa;
4631 frag->data = val;
4632 frag->len = bytes;
f78146b0 4633 return X86EMUL_CONTINUE;
bbd9b64e
CO		 4634}
4635
52eb5a6d
XL		 4636static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4637 unsigned long addr,
22388a3c
XG		 4638 void *val, unsigned int bytes,
4639 struct x86_exception *exception,
0fbe9b0b 4640 const struct read_write_emulator_ops *ops)
bbd9b64e 4641{
0f65dd70 4642 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
f78146b0
AK		 4643 gpa_t gpa;
4644 int rc;
4645
4646 if (ops->read_write_prepare &&
4647 ops->read_write_prepare(vcpu, val, bytes))
4648 return X86EMUL_CONTINUE;
4649
4650 vcpu->mmio_nr_fragments = 0;
0f65dd70 4651
bbd9b64e
CO		 4652 /* Crossing a page boundary? */
4653 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
f78146b0 4654 int now;
bbd9b64e
CO		 4655
4656 now = -addr & ~PAGE_MASK;
22388a3c
XG		 4657 rc = emulator_read_write_onepage(addr, val, now, exception,
4658 vcpu, ops);
4659
bbd9b64e
CO		 4660 if (rc != X86EMUL_CONTINUE)
4661 return rc;
4662 addr += now;
bac15531
NA		 4663 if (ctxt->mode != X86EMUL_MODE_PROT64)
4664 addr = (u32)addr;
bbd9b64e
CO		 4665 val += now;
4666 bytes -= now;
4667 }
22388a3c 4668
f78146b0
AK		 4669 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4670 vcpu, ops);
4671 if (rc != X86EMUL_CONTINUE)
4672 return rc;
4673
4674 if (!vcpu->mmio_nr_fragments)
4675 return rc;
4676
4677 gpa = vcpu->mmio_fragments[0].gpa;
4678
4679 vcpu->mmio_needed = 1;
4680 vcpu->mmio_cur_fragment = 0;
4681
87da7e66 4682 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
f78146b0
AK		 4683 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4684 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4685 vcpu->run->mmio.phys_addr = gpa;
4686
4687 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
22388a3c
XG		 4688}
4689
4690static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4691 unsigned long addr,
4692 void *val,
4693 unsigned int bytes,
4694 struct x86_exception *exception)
4695{
4696 return emulator_read_write(ctxt, addr, val, bytes,
4697 exception, &read_emultor);
4698}
4699
52eb5a6d 4700static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
22388a3c
XG		 4701 unsigned long addr,
4702 const void *val,
4703 unsigned int bytes,
4704 struct x86_exception *exception)
4705{
4706 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4707 exception, &write_emultor);
bbd9b64e 4708}
bbd9b64e 4709
daea3e73
AK		 4710#define CMPXCHG_TYPE(t, ptr, old, new) \
4711 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4712
4713#ifdef CONFIG_X86_64
4714# define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4715#else
4716# define CMPXCHG64(ptr, old, new) \
9749a6c0 4717 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
daea3e73
AK		 4718#endif
4719
0f65dd70
AK		 4720static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4721 unsigned long addr,
bbd9b64e
CO		 4722 const void *old,
4723 const void *new,
4724 unsigned int bytes,
0f65dd70 4725 struct x86_exception *exception)
bbd9b64e 4726{
0f65dd70 4727 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
daea3e73
AK		 4728 gpa_t gpa;
4729 struct page *page;
4730 char *kaddr;
4731 bool exchanged;
2bacc55c 4732
daea3e73
AK		 4733 /* guests cmpxchg8b have to be emulated atomically */
4734 if (bytes > 8 || (bytes & (bytes - 1)))
4735 goto emul_write;
10589a46 4736
daea3e73 4737 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
2bacc55c 4738
daea3e73
AK		 4739 if (gpa == UNMAPPED_GVA ||
4740 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4741 goto emul_write;
2bacc55c 4742
daea3e73
AK		 4743 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4744 goto emul_write;
72dc67a6 4745
54bf36aa 4746 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
32cad84f 4747 if (is_error_page(page))
c19b8bd6 4748 goto emul_write;
72dc67a6 4749
8fd75e12 4750 kaddr = kmap_atomic(page);
daea3e73
AK		 4751 kaddr += offset_in_page(gpa);
4752 switch (bytes) {
4753 case 1:
4754 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4755 break;
4756 case 2:
4757 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4758 break;
4759 case 4:
4760 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4761 break;
4762 case 8:
4763 exchanged = CMPXCHG64(kaddr, old, new);
4764 break;
4765 default:
4766 BUG();
2bacc55c 4767 }
8fd75e12 4768 kunmap_atomic(kaddr);
daea3e73
AK		 4769 kvm_release_page_dirty(page);
4770
4771 if (!exchanged)
4772 return X86EMUL_CMPXCHG_FAILED;
4773
54bf36aa 4774 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
0eb05bf2 4775 kvm_page_track_write(vcpu, gpa, new, bytes);
8f6abd06
GN		 4776
4777 return X86EMUL_CONTINUE;
4a5f48f6 4778
3200f405 4779emul_write:
daea3e73 4780 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
2bacc55c 4781
0f65dd70 4782 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
bbd9b64e
CO		 4783}
4784
cf8f70bf
GN		 4785static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4786{
4787 /* TODO: String I/O for in kernel device */
4788 int r;
4789
4790 if (vcpu->arch.pio.in)
e32edf4f 4791 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
cf8f70bf
GN		 4792 vcpu->arch.pio.size, pd);
4793 else
e32edf4f 4794 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
cf8f70bf
GN		 4795 vcpu->arch.pio.port, vcpu->arch.pio.size,
4796 pd);
4797 return r;
4798}
4799
6f6fbe98
XG		 4800static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4801 unsigned short port, void *val,
4802 unsigned int count, bool in)
cf8f70bf 4803{
cf8f70bf 4804 vcpu->arch.pio.port = port;
6f6fbe98 4805 vcpu->arch.pio.in = in;
7972995b 4806 vcpu->arch.pio.count = count;
cf8f70bf
GN		 4807 vcpu->arch.pio.size = size;
4808
4809 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
7972995b 4810 vcpu->arch.pio.count = 0;
cf8f70bf
GN		 4811 return 1;
4812 }
4813
4814 vcpu->run->exit_reason = KVM_EXIT_IO;
6f6fbe98 4815 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
cf8f70bf
GN		 4816 vcpu->run->io.size = size;
4817 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4818 vcpu->run->io.count = count;
4819 vcpu->run->io.port = port;
4820
4821 return 0;
4822}
4823
6f6fbe98
XG		 4824static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4825 int size, unsigned short port, void *val,
4826 unsigned int count)
cf8f70bf 4827{
ca1d4a9e 4828 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6f6fbe98 4829 int ret;
ca1d4a9e 4830
6f6fbe98
XG		 4831 if (vcpu->arch.pio.count)
4832 goto data_avail;
cf8f70bf 4833
6f6fbe98
XG		 4834 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4835 if (ret) {
4836data_avail:
4837 memcpy(val, vcpu->arch.pio_data, size * count);
1171903d 4838 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
7972995b 4839 vcpu->arch.pio.count = 0;
cf8f70bf
GN		 4840 return 1;
4841 }
4842
cf8f70bf
GN		 4843 return 0;
4844}
4845
6f6fbe98
XG		 4846static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4847 int size, unsigned short port,
4848 const void *val, unsigned int count)
4849{
4850 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4851
4852 memcpy(vcpu->arch.pio_data, val, size * count);
1171903d 4853 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6f6fbe98
XG		 4854 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4855}
4856
bbd9b64e
CO		 4857static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4858{
4859 return kvm_x86_ops->get_segment_base(vcpu, seg);
4860}
4861
3cb16fe7 4862static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
bbd9b64e 4863{
3cb16fe7 4864 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
bbd9b64e
CO		 4865}
4866
ae6a2375 4867static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
f5f48ee1
SY		 4868{
4869 if (!need_emulate_wbinvd(vcpu))
4870 return X86EMUL_CONTINUE;
4871
4872 if (kvm_x86_ops->has_wbinvd_exit()) {
2eec7343
JK		 4873 int cpu = get_cpu();
4874
4875 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
f5f48ee1
SY		 4876 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4877 wbinvd_ipi, NULL, 1);
2eec7343 4878 put_cpu();
f5f48ee1 4879 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
2eec7343
JK		 4880 } else
4881 wbinvd();
f5f48ee1
SY		 4882 return X86EMUL_CONTINUE;
4883}
5cb56059
JS		 4884
4885int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4886{
6affcbed
KH		 4887 kvm_emulate_wbinvd_noskip(vcpu);
4888 return kvm_skip_emulated_instruction(vcpu);
5cb56059 4889}
f5f48ee1
SY		 4890EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4891
5cb56059
JS		 4892
4893
bcaf5cc5
AK		 4894static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4895{
5cb56059 4896 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
bcaf5cc5
AK		 4897}
4898
52eb5a6d
XL		 4899static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4900 unsigned long *dest)
bbd9b64e 4901{
16f8a6f9 4902 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
bbd9b64e
CO		 4903}
4904
52eb5a6d
XL		 4905static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4906 unsigned long value)
bbd9b64e 4907{
338dbc97 4908
717746e3 4909 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
bbd9b64e
CO		 4910}
4911
52a46617 4912static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5fdbf976 4913{
52a46617 4914 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5fdbf976
MT		 4915}
4916
717746e3 4917static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
bbd9b64e 4918{
717746e3 4919 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
52a46617
GN		 4920 unsigned long value;
4921
4922 switch (cr) {
4923 case 0:
4924 value = kvm_read_cr0(vcpu);
4925 break;
4926 case 2:
4927 value = vcpu->arch.cr2;
4928 break;
4929 case 3:
9f8fe504 4930 value = kvm_read_cr3(vcpu);
52a46617
GN		 4931 break;
4932 case 4:
4933 value = kvm_read_cr4(vcpu);
4934 break;
4935 case 8:
4936 value = kvm_get_cr8(vcpu);
4937 break;
4938 default:
a737f256 4939 kvm_err("%s: unexpected cr %u\n", __func__, cr);
52a46617
GN		 4940 return 0;
4941 }
4942
4943 return value;
4944}
4945
717746e3 4946static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
52a46617 4947{
717746e3 4948 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
0f12244f
GN		 4949 int res = 0;
4950
52a46617
GN		 4951 switch (cr) {
4952 case 0:
49a9b07e 4953 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
52a46617
GN		 4954 break;
4955 case 2:
4956 vcpu->arch.cr2 = val;
4957 break;
4958 case 3:
2390218b 4959 res = kvm_set_cr3(vcpu, val);
52a46617
GN		 4960 break;
4961 case 4:
a83b29c6 4962 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
52a46617
GN		 4963 break;
4964 case 8:
eea1cff9 4965 res = kvm_set_cr8(vcpu, val);
52a46617
GN		 4966 break;
4967 default:
a737f256 4968 kvm_err("%s: unexpected cr %u\n", __func__, cr);
0f12244f 4969 res = -1;
52a46617 4970 }
0f12244f
GN		 4971
4972 return res;
52a46617
GN		 4973}
4974
717746e3 4975static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
9c537244 4976{
717746e3 4977 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
9c537244
GN		 4978}
4979
4bff1e86 4980static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
2dafc6c2 4981{
4bff1e86 4982 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
2dafc6c2
GN		 4983}
4984
4bff1e86 4985static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
160ce1f1 4986{
4bff1e86 4987 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
160ce1f1
MG		 4988}
4989
1ac9d0cf
AK		 4990static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4991{
4992 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4993}
4994
4995static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4996{
4997 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4998}
4999
4bff1e86
AK		 5000static unsigned long emulator_get_cached_segment_base(
5001 struct x86_emulate_ctxt *ctxt, int seg)
5951c442 5002{
4bff1e86 5003 return get_segment_base(emul_to_vcpu(ctxt), seg);
5951c442
GN		 5004}
5005
1aa36616
AK		 5006static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5007 struct desc_struct *desc, u32 *base3,
5008 int seg)
2dafc6c2
GN		 5009{
5010 struct kvm_segment var;
5011
4bff1e86 5012 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
1aa36616 5013 *selector = var.selector;
2dafc6c2 5014
378a8b09
GN		 5015 if (var.unusable) {
5016 memset(desc, 0, sizeof(*desc));
2dafc6c2 5017 return false;
378a8b09 5018 }
2dafc6c2
GN		 5019
5020 if (var.g)
5021 var.limit >>= 12;
5022 set_desc_limit(desc, var.limit);
5023 set_desc_base(desc, (unsigned long)var.base);
5601d05b
GN		 5024#ifdef CONFIG_X86_64
5025 if (base3)
5026 *base3 = var.base >> 32;
5027#endif
2dafc6c2
GN		 5028 desc->type = var.type;
5029 desc->s = var.s;
5030 desc->dpl = var.dpl;
5031 desc->p = var.present;
5032 desc->avl = var.avl;
5033 desc->l = var.l;
5034 desc->d = var.db;
5035 desc->g = var.g;
5036
5037 return true;
5038}
5039
1aa36616
AK		 5040static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5041 struct desc_struct *desc, u32 base3,
5042 int seg)
2dafc6c2 5043{
4bff1e86 5044 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
2dafc6c2
GN		 5045 struct kvm_segment var;
5046
1aa36616 5047 var.selector = selector;
2dafc6c2 5048 var.base = get_desc_base(desc);
5601d05b
GN		 5049#ifdef CONFIG_X86_64
5050 var.base |= ((u64)base3) << 32;
5051#endif
2dafc6c2
GN		 5052 var.limit = get_desc_limit(desc);
5053 if (desc->g)
5054 var.limit = (var.limit << 12) | 0xfff;
5055 var.type = desc->type;
2dafc6c2
GN		 5056 var.dpl = desc->dpl;
5057 var.db = desc->d;
5058 var.s = desc->s;
5059 var.l = desc->l;
5060 var.g = desc->g;
5061 var.avl = desc->avl;
5062 var.present = desc->p;
5063 var.unusable = !var.present;
5064 var.padding = 0;
5065
5066 kvm_set_segment(vcpu, &var, seg);
5067 return;
5068}
5069
717746e3
AK		 5070static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5071 u32 msr_index, u64 *pdata)
5072{
609e36d3
PB		 5073 struct msr_data msr;
5074 int r;
5075
5076 msr.index = msr_index;
5077 msr.host_initiated = false;
5078 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5079 if (r)
5080 return r;
5081
5082 *pdata = msr.data;
5083 return 0;
717746e3
AK		 5084}
5085
5086static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5087 u32 msr_index, u64 data)
5088{
8fe8ab46
WA		 5089 struct msr_data msr;
5090
5091 msr.data = data;
5092 msr.index = msr_index;
5093 msr.host_initiated = false;
5094 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
717746e3
AK		 5095}
5096
64d60670
PB		 5097static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5098{
5099 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5100
5101 return vcpu->arch.smbase;
5102}
5103
5104static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5105{
5106 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5107
5108 vcpu->arch.smbase = smbase;
5109}
5110
67f4d428
NA		 5111static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5112 u32 pmc)
5113{
c6702c9d 5114 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
67f4d428
NA		 5115}
5116
222d21aa
AK		 5117static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5118 u32 pmc, u64 *pdata)
5119{
c6702c9d 5120 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
222d21aa
AK		 5121}
5122
6c3287f7
AK		 5123static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5124{
5125 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5126}
5127
5037f6f3
AK		 5128static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
5129{
5130 preempt_disable();
5197b808 5131 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
5037f6f3
AK		 5132}
5133
5134static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
5135{
5136 preempt_enable();
5137}
5138
2953538e 5139static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
8a76d7f2 5140 struct x86_instruction_info *info,
c4f035c6
AK		 5141 enum x86_intercept_stage stage)
5142{
2953538e 5143 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
c4f035c6
AK		 5144}
5145
0017f93a 5146static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
bdb42f5a
SB		 5147 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
5148{
0017f93a 5149 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
bdb42f5a
SB		 5150}
5151
dd856efa
AK		 5152static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5153{
5154 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5155}
5156
5157static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5158{
5159 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5160}
5161
801806d9
NA		 5162static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5163{
5164 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5165}
5166
0225fb50 5167static const struct x86_emulate_ops emulate_ops = {
dd856efa
AK		 5168 .read_gpr = emulator_read_gpr,
5169 .write_gpr = emulator_write_gpr,
1871c602 5170 .read_std = kvm_read_guest_virt_system,
2dafc6c2 5171 .write_std = kvm_write_guest_virt_system,
7a036a6f 5172 .read_phys = kvm_read_guest_phys_system,
1871c602 5173 .fetch = kvm_fetch_guest_virt,
bbd9b64e
CO		 5174 .read_emulated = emulator_read_emulated,
5175 .write_emulated = emulator_write_emulated,
5176 .cmpxchg_emulated = emulator_cmpxchg_emulated,
3cb16fe7 5177 .invlpg = emulator_invlpg,
cf8f70bf
GN		 5178 .pio_in_emulated = emulator_pio_in_emulated,
5179 .pio_out_emulated = emulator_pio_out_emulated,
1aa36616
AK		 5180 .get_segment = emulator_get_segment,
5181 .set_segment = emulator_set_segment,
5951c442 5182 .get_cached_segment_base = emulator_get_cached_segment_base,
2dafc6c2 5183 .get_gdt = emulator_get_gdt,
160ce1f1 5184 .get_idt = emulator_get_idt,
1ac9d0cf
AK		 5185 .set_gdt = emulator_set_gdt,
5186 .set_idt = emulator_set_idt,
52a46617
GN		 5187 .get_cr = emulator_get_cr,
5188 .set_cr = emulator_set_cr,
9c537244 5189 .cpl = emulator_get_cpl,
35aa5375
GN		 5190 .get_dr = emulator_get_dr,
5191 .set_dr = emulator_set_dr,
64d60670
PB		 5192 .get_smbase = emulator_get_smbase,
5193 .set_smbase = emulator_set_smbase,
717746e3
AK		 5194 .set_msr = emulator_set_msr,
5195 .get_msr = emulator_get_msr,
67f4d428 5196 .check_pmc = emulator_check_pmc,
222d21aa 5197 .read_pmc = emulator_read_pmc,
6c3287f7 5198 .halt = emulator_halt,
bcaf5cc5 5199 .wbinvd = emulator_wbinvd,
d6aa1000 5200 .fix_hypercall = emulator_fix_hypercall,
5037f6f3
AK		 5201 .get_fpu = emulator_get_fpu,
5202 .put_fpu = emulator_put_fpu,
c4f035c6 5203 .intercept = emulator_intercept,
bdb42f5a 5204 .get_cpuid = emulator_get_cpuid,
801806d9 5205 .set_nmi_mask = emulator_set_nmi_mask,
bbd9b64e
CO		 5206};
5207
95cb2295
GN		 5208static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5209{
37ccdcbe 5210 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
95cb2295
GN		 5211 /*
5212 * an sti; sti; sequence only disable interrupts for the first
5213 * instruction. So, if the last instruction, be it emulated or
5214 * not, left the system with the INT_STI flag enabled, it
5215 * means that the last instruction is an sti. We should not
5216 * leave the flag on in this case. The same goes for mov ss
5217 */
37ccdcbe
PB		 5218 if (int_shadow & mask)
5219 mask = 0;
6addfc42 5220 if (unlikely(int_shadow || mask)) {
95cb2295 5221 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6addfc42
PB		 5222 if (!mask)
5223 kvm_make_request(KVM_REQ_EVENT, vcpu);
5224 }
95cb2295
GN		 5225}
5226
ef54bcfe 5227static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
54b8486f
GN		 5228{
5229 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
da9cb575 5230 if (ctxt->exception.vector == PF_VECTOR)
ef54bcfe
PB		 5231 return kvm_propagate_fault(vcpu, &ctxt->exception);
5232
5233 if (ctxt->exception.error_code_valid)
da9cb575
AK		 5234 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5235 ctxt->exception.error_code);
54b8486f 5236 else
da9cb575 5237 kvm_queue_exception(vcpu, ctxt->exception.vector);
ef54bcfe 5238 return false;
54b8486f
GN		 5239}
5240
8ec4722d
MG		 5241static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5242{
adf52235 5243 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8ec4722d
MG		 5244 int cs_db, cs_l;
5245
8ec4722d
MG		 5246 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5247
adf52235
TY		 5248 ctxt->eflags = kvm_get_rflags(vcpu);
5249 ctxt->eip = kvm_rip_read(vcpu);
5250 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5251 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
42bf549f 5252 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
adf52235
TY		 5253 cs_db ? X86EMUL_MODE_PROT32 :
5254 X86EMUL_MODE_PROT16;
a584539b 5255 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
64d60670
PB		 5256 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5257 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
a584539b 5258 ctxt->emul_flags = vcpu->arch.hflags;
adf52235 5259
dd856efa 5260 init_decode_cache(ctxt);
7ae441ea 5261 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8ec4722d
MG		 5262}
5263
71f9833b 5264int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
63995653 5265{
9d74191a 5266 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
63995653
MG		 5267 int ret;
5268
5269 init_emulate_ctxt(vcpu);
5270
9dac77fa
AK		 5271 ctxt->op_bytes = 2;
5272 ctxt->ad_bytes = 2;
5273 ctxt->_eip = ctxt->eip + inc_eip;
9d74191a 5274 ret = emulate_int_real(ctxt, irq);
63995653
MG		 5275
5276 if (ret != X86EMUL_CONTINUE)
5277 return EMULATE_FAIL;
5278
9dac77fa 5279 ctxt->eip = ctxt->_eip;
9d74191a
TY		 5280 kvm_rip_write(vcpu, ctxt->eip);
5281 kvm_set_rflags(vcpu, ctxt->eflags);
63995653
MG		 5282
5283 if (irq == NMI_VECTOR)
7460fb4a 5284 vcpu->arch.nmi_pending = 0;
63995653
MG		 5285 else
5286 vcpu->arch.interrupt.pending = false;
5287
5288 return EMULATE_DONE;
5289}
5290EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5291
6d77dbfc
GN		 5292static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5293{
fc3a9157
JR		 5294 int r = EMULATE_DONE;
5295
6d77dbfc
GN		 5296 ++vcpu->stat.insn_emulation_fail;
5297 trace_kvm_emulate_insn_failed(vcpu);
a2b9e6c1 5298 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
fc3a9157
JR		 5299 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5300 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5301 vcpu->run->internal.ndata = 0;
5302 r = EMULATE_FAIL;
5303 }
6d77dbfc 5304 kvm_queue_exception(vcpu, UD_VECTOR);
fc3a9157
JR		 5305
5306 return r;
6d77dbfc
GN		 5307}
5308
93c05d3e 5309static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
991eebf9
GN		 5310 bool write_fault_to_shadow_pgtable,
5311 int emulation_type)
a6f177ef 5312{
95b3cf69 5313 gpa_t gpa = cr2;
ba049e93 5314 kvm_pfn_t pfn;
a6f177ef 5315
991eebf9
GN		 5316 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5317 return false;
5318
95b3cf69
XG		 5319 if (!vcpu->arch.mmu.direct_map) {
5320 /*
5321 * Write permission should be allowed since only
5322 * write access need to be emulated.
5323 */
5324 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
a6f177ef 5325
95b3cf69
XG		 5326 /*
5327 * If the mapping is invalid in guest, let cpu retry
5328 * it to generate fault.
5329 */
5330 if (gpa == UNMAPPED_GVA)
5331 return true;
5332 }
a6f177ef 5333
8e3d9d06
XG		 5334 /*
5335 * Do not retry the unhandleable instruction if it faults on the
5336 * readonly host memory, otherwise it will goto a infinite loop:
5337 * retry instruction -> write #PF -> emulation fail -> retry
5338 * instruction -> ...
5339 */
5340 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
95b3cf69
XG		 5341
5342 /*
5343 * If the instruction failed on the error pfn, it can not be fixed,
5344 * report the error to userspace.
5345 */
5346 if (is_error_noslot_pfn(pfn))
5347 return false;
5348
5349 kvm_release_pfn_clean(pfn);
5350
5351 /* The instructions are well-emulated on direct mmu. */
5352 if (vcpu->arch.mmu.direct_map) {
5353 unsigned int indirect_shadow_pages;
5354
5355 spin_lock(&vcpu->kvm->mmu_lock);
5356 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5357 spin_unlock(&vcpu->kvm->mmu_lock);
5358
5359 if (indirect_shadow_pages)
5360 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5361
a6f177ef 5362 return true;
8e3d9d06 5363 }
a6f177ef 5364
95b3cf69
XG		 5365 /*
5366 * if emulation was due to access to shadowed page table
5367 * and it failed try to unshadow page and re-enter the
5368 * guest to let CPU execute the instruction.
5369 */
5370 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
93c05d3e
XG		 5371
5372 /*
5373 * If the access faults on its page table, it can not
5374 * be fixed by unprotecting shadow page and it should
5375 * be reported to userspace.
5376 */
5377 return !write_fault_to_shadow_pgtable;
a6f177ef
GN		 5378}
5379
1cb3f3ae
XG		 5380static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5381 unsigned long cr2, int emulation_type)
5382{
5383 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5384 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5385
5386 last_retry_eip = vcpu->arch.last_retry_eip;
5387 last_retry_addr = vcpu->arch.last_retry_addr;
5388
5389 /*
5390 * If the emulation is caused by #PF and it is non-page_table
5391 * writing instruction, it means the VM-EXIT is caused by shadow
5392 * page protected, we can zap the shadow page and retry this
5393 * instruction directly.
5394 *
5395 * Note: if the guest uses a non-page-table modifying instruction
5396 * on the PDE that points to the instruction, then we will unmap
5397 * the instruction and go to an infinite loop. So, we cache the
5398 * last retried eip and the last fault address, if we meet the eip
5399 * and the address again, we can break out of the potential infinite
5400 * loop.
5401 */
5402 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5403
5404 if (!(emulation_type & EMULTYPE_RETRY))
5405 return false;
5406
5407 if (x86_page_table_writing_insn(ctxt))
5408 return false;
5409
5410 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5411 return false;
5412
5413 vcpu->arch.last_retry_eip = ctxt->eip;
5414 vcpu->arch.last_retry_addr = cr2;
5415
5416 if (!vcpu->arch.mmu.direct_map)
5417 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5418
22368028 5419 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
1cb3f3ae
XG		 5420
5421 return true;
5422}
5423
716d51ab
GN		 5424static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5425static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5426
64d60670 5427static void kvm_smm_changed(struct kvm_vcpu *vcpu)
a584539b 5428{
64d60670 5429 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
660a5d51
PB		 5430 /* This is a good place to trace that we are exiting SMM. */
5431 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5432
c43203ca
PB		 5433 /* Process a latched INIT or SMI, if any. */
5434 kvm_make_request(KVM_REQ_EVENT, vcpu);
64d60670 5435 }
699023e2
PB		 5436
5437 kvm_mmu_reset_context(vcpu);
64d60670
PB		 5438}
5439
5440static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5441{
5442 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5443
a584539b 5444 vcpu->arch.hflags = emul_flags;
64d60670
PB		 5445
5446 if (changed & HF_SMM_MASK)
5447 kvm_smm_changed(vcpu);
a584539b
PB		 5448}
5449
4a1e10d5
PB		 5450static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5451 unsigned long *db)
5452{
5453 u32 dr6 = 0;
5454 int i;
5455 u32 enable, rwlen;
5456
5457 enable = dr7;
5458 rwlen = dr7 >> 16;
5459 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5460 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5461 dr6 |= (1 << i);
5462 return dr6;
5463}
5464
6addfc42 5465static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
663f4c61
PB		 5466{
5467 struct kvm_run *kvm_run = vcpu->run;
5468
5469 /*
6addfc42
PB		 5470 * rflags is the old, "raw" value of the flags. The new value has
5471 * not been saved yet.
663f4c61
PB		 5472 *
5473 * This is correct even for TF set by the guest, because "the
5474 * processor will not generate this exception after the instruction
5475 * that sets the TF flag".
5476 */
663f4c61
PB		 5477 if (unlikely(rflags & X86_EFLAGS_TF)) {
5478 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6f43ed01
NA		 5479 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5480 DR6_RTM;
663f4c61
PB		 5481 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5482 kvm_run->debug.arch.exception = DB_VECTOR;
5483 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5484 *r = EMULATE_USER_EXIT;
5485 } else {
663f4c61
PB		 5486 /*
5487 * "Certain debug exceptions may clear bit 0-3. The
5488 * remaining contents of the DR6 register are never
5489 * cleared by the processor".
5490 */
5491 vcpu->arch.dr6 &= ~15;
6f43ed01 5492 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
663f4c61
PB		 5493 kvm_queue_exception(vcpu, DB_VECTOR);
5494 }
5495 }
5496}
5497
6affcbed
KH		 5498int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
5499{
5500 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5501 int r = EMULATE_DONE;
5502
5503 kvm_x86_ops->skip_emulated_instruction(vcpu);
5504 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5505 return r == EMULATE_DONE;
5506}
5507EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
5508
4a1e10d5
PB		 5509static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5510{
4a1e10d5
PB		 5511 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5512 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
82b32774
NA		 5513 struct kvm_run *kvm_run = vcpu->run;
5514 unsigned long eip = kvm_get_linear_rip(vcpu);
5515 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
4a1e10d5
PB		 5516 vcpu->arch.guest_debug_dr7,
5517 vcpu->arch.eff_db);
5518
5519 if (dr6 != 0) {
6f43ed01 5520 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
82b32774 5521 kvm_run->debug.arch.pc = eip;
4a1e10d5
PB		 5522 kvm_run->debug.arch.exception = DB_VECTOR;
5523 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5524 *r = EMULATE_USER_EXIT;
5525 return true;
5526 }
5527 }
5528
4161a569
NA		 5529 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5530 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
82b32774
NA		 5531 unsigned long eip = kvm_get_linear_rip(vcpu);
5532 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
4a1e10d5
PB		 5533 vcpu->arch.dr7,
5534 vcpu->arch.db);
5535
5536 if (dr6 != 0) {
5537 vcpu->arch.dr6 &= ~15;
6f43ed01 5538 vcpu->arch.dr6 |= dr6 | DR6_RTM;
4a1e10d5
PB		 5539 kvm_queue_exception(vcpu, DB_VECTOR);
5540 *r = EMULATE_DONE;
5541 return true;
5542 }
5543 }
5544
5545 return false;
5546}
5547
51d8b661
AP		 5548int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5549 unsigned long cr2,
dc25e89e
AP		 5550 int emulation_type,
5551 void *insn,
5552 int insn_len)
bbd9b64e 5553{
95cb2295 5554 int r;
9d74191a 5555 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7ae441ea 5556 bool writeback = true;
93c05d3e 5557 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
bbd9b64e 5558
93c05d3e
XG		 5559 /*
5560 * Clear write_fault_to_shadow_pgtable here to ensure it is
5561 * never reused.
5562 */
5563 vcpu->arch.write_fault_to_shadow_pgtable = false;
26eef70c 5564 kvm_clear_exception_queue(vcpu);
8d7d8102 5565
571008da 5566 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
8ec4722d 5567 init_emulate_ctxt(vcpu);
4a1e10d5
PB		 5568
5569 /*
5570 * We will reenter on the same instruction since
5571 * we do not set complete_userspace_io. This does not
5572 * handle watchpoints yet, those would be handled in
5573 * the emulate_ops.
5574 */
5575 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5576 return r;
5577
9d74191a
TY		 5578 ctxt->interruptibility = 0;
5579 ctxt->have_exception = false;
e0ad0b47 5580 ctxt->exception.vector = -1;
9d74191a 5581 ctxt->perm_ok = false;
bbd9b64e 5582
b51e974f 5583 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
4005996e 5584
9d74191a 5585 r = x86_decode_insn(ctxt, insn, insn_len);
bbd9b64e 5586
e46479f8 5587 trace_kvm_emulate_insn_start(vcpu);
f2b5756b 5588 ++vcpu->stat.insn_emulation;
1d2887e2 5589 if (r != EMULATION_OK) {
4005996e
AK		 5590 if (emulation_type & EMULTYPE_TRAP_UD)
5591 return EMULATE_FAIL;
991eebf9
GN		 5592 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5593 emulation_type))
bbd9b64e 5594 return EMULATE_DONE;
6d77dbfc
GN		 5595 if (emulation_type & EMULTYPE_SKIP)
5596 return EMULATE_FAIL;
5597 return handle_emulation_failure(vcpu);
bbd9b64e
CO		 5598 }
5599 }
5600
ba8afb6b 5601 if (emulation_type & EMULTYPE_SKIP) {
9dac77fa 5602 kvm_rip_write(vcpu, ctxt->_eip);
bb663c7a
NA		 5603 if (ctxt->eflags & X86_EFLAGS_RF)
5604 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
ba8afb6b
GN		 5605 return EMULATE_DONE;
5606 }
5607
1cb3f3ae
XG		 5608 if (retry_instruction(ctxt, cr2, emulation_type))
5609 return EMULATE_DONE;
5610
7ae441ea 5611 /* this is needed for vmware backdoor interface to work since it
4d2179e1 5612 changes registers values during IO operation */
7ae441ea
GN		 5613 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5614 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
dd856efa 5615 emulator_invalidate_register_cache(ctxt);
7ae441ea 5616 }
4d2179e1 5617
5cd21917 5618restart:
9d74191a 5619 r = x86_emulate_insn(ctxt);
bbd9b64e 5620
775fde86
JR		 5621 if (r == EMULATION_INTERCEPTED)
5622 return EMULATE_DONE;
5623
d2ddd1c4 5624 if (r == EMULATION_FAILED) {
991eebf9
GN		 5625 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5626 emulation_type))
c3cd7ffa
GN		 5627 return EMULATE_DONE;
5628
6d77dbfc 5629 return handle_emulation_failure(vcpu);
bbd9b64e
CO		 5630 }
5631
9d74191a 5632 if (ctxt->have_exception) {
d2ddd1c4 5633 r = EMULATE_DONE;
ef54bcfe
PB		 5634 if (inject_emulated_exception(vcpu))
5635 return r;
d2ddd1c4 5636 } else if (vcpu->arch.pio.count) {
0912c977
PB		 5637 if (!vcpu->arch.pio.in) {
5638 /* FIXME: return into emulator if single-stepping. */
3457e419 5639 vcpu->arch.pio.count = 0;
0912c977 5640 } else {
7ae441ea 5641 writeback = false;
716d51ab
GN		 5642 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5643 }
ac0a48c3 5644 r = EMULATE_USER_EXIT;
7ae441ea
GN		 5645 } else if (vcpu->mmio_needed) {
5646 if (!vcpu->mmio_is_write)
5647 writeback = false;
ac0a48c3 5648 r = EMULATE_USER_EXIT;
716d51ab 5649 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7ae441ea 5650 } else if (r == EMULATION_RESTART)
5cd21917 5651 goto restart;
d2ddd1c4
GN		 5652 else
5653 r = EMULATE_DONE;
f850e2e6 5654
7ae441ea 5655 if (writeback) {
6addfc42 5656 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
9d74191a 5657 toggle_interruptibility(vcpu, ctxt->interruptibility);
7ae441ea 5658 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
a584539b
PB		 5659 if (vcpu->arch.hflags != ctxt->emul_flags)
5660 kvm_set_hflags(vcpu, ctxt->emul_flags);
9d74191a 5661 kvm_rip_write(vcpu, ctxt->eip);
663f4c61 5662 if (r == EMULATE_DONE)
6addfc42 5663 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
38827dbd
NA		 5664 if (!ctxt->have_exception ||
5665 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5666 __kvm_set_rflags(vcpu, ctxt->eflags);
6addfc42
PB		 5667
5668 /*
5669 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5670 * do nothing, and it will be requested again as soon as
5671 * the shadow expires. But we still need to check here,
5672 * because POPF has no interrupt shadow.
5673 */
5674 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5675 kvm_make_request(KVM_REQ_EVENT, vcpu);
7ae441ea
GN		 5676 } else
5677 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
e85d28f8
GN		 5678
5679 return r;
de7d789a 5680}
51d8b661 5681EXPORT_SYMBOL_GPL(x86_emulate_instruction);
de7d789a 5682
cf8f70bf 5683int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
de7d789a 5684{
cf8f70bf 5685 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
ca1d4a9e
AK		 5686 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5687 size, port, &val, 1);
cf8f70bf 5688 /* do not return to emulator after return from userspace */
7972995b 5689 vcpu->arch.pio.count = 0;
de7d789a
CO		 5690 return ret;
5691}
cf8f70bf 5692EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
de7d789a 5693
8370c3d0
TL		 5694static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
5695{
5696 unsigned long val;
5697
5698 /* We should only ever be called with arch.pio.count equal to 1 */
5699 BUG_ON(vcpu->arch.pio.count != 1);
5700
5701 /* For size less than 4 we merge, else we zero extend */
5702 val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
5703 : 0;
5704
5705 /*
5706 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
5707 * the copy and tracing
5708 */
5709 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
5710 vcpu->arch.pio.port, &val, 1);
5711 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5712
5713 return 1;
5714}
5715
5716int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size, unsigned short port)
5717{
5718 unsigned long val;
5719 int ret;
5720
5721 /* For size less than 4 we merge, else we zero extend */
5722 val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
5723
5724 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
5725 &val, 1);
5726 if (ret) {
5727 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5728 return ret;
5729 }
5730
5731 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
5732
5733 return 0;
5734}
5735EXPORT_SYMBOL_GPL(kvm_fast_pio_in);
5736
251a5fd6 5737static int kvmclock_cpu_down_prep(unsigned int cpu)
8cfdc000 5738{
0a3aee0d 5739 __this_cpu_write(cpu_tsc_khz, 0);
251a5fd6 5740 return 0;
8cfdc000
ZA		 5741}
5742
5743static void tsc_khz_changed(void *data)
c8076604 5744{
8cfdc000
ZA		 5745 struct cpufreq_freqs *freq = data;
5746 unsigned long khz = 0;
5747
5748 if (data)
5749 khz = freq->new;
5750 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5751 khz = cpufreq_quick_get(raw_smp_processor_id());
5752 if (!khz)
5753 khz = tsc_khz;
0a3aee0d 5754 __this_cpu_write(cpu_tsc_khz, khz);
c8076604
GH		 5755}
5756
c8076604
GH		 5757static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5758 void *data)
5759{
5760 struct cpufreq_freqs *freq = data;
5761 struct kvm *kvm;
5762 struct kvm_vcpu *vcpu;
5763 int i, send_ipi = 0;
5764
8cfdc000
ZA		 5765 /*
5766 * We allow guests to temporarily run on slowing clocks,
5767 * provided we notify them after, or to run on accelerating
5768 * clocks, provided we notify them before. Thus time never
5769 * goes backwards.
5770 *
5771 * However, we have a problem. We can't atomically update
5772 * the frequency of a given CPU from this function; it is
5773 * merely a notifier, which can be called from any CPU.
5774 * Changing the TSC frequency at arbitrary points in time
5775 * requires a recomputation of local variables related to
5776 * the TSC for each VCPU. We must flag these local variables
5777 * to be updated and be sure the update takes place with the
5778 * new frequency before any guests proceed.
5779 *
5780 * Unfortunately, the combination of hotplug CPU and frequency
5781 * change creates an intractable locking scenario; the order
5782 * of when these callouts happen is undefined with respect to
5783 * CPU hotplug, and they can race with each other. As such,
5784 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5785 * undefined; you can actually have a CPU frequency change take
5786 * place in between the computation of X and the setting of the
5787 * variable. To protect against this problem, all updates of
5788 * the per_cpu tsc_khz variable are done in an interrupt
5789 * protected IPI, and all callers wishing to update the value
5790 * must wait for a synchronous IPI to complete (which is trivial
5791 * if the caller is on the CPU already). This establishes the
5792 * necessary total order on variable updates.
5793 *
5794 * Note that because a guest time update may take place
5795 * anytime after the setting of the VCPU's request bit, the
5796 * correct TSC value must be set before the request. However,
5797 * to ensure the update actually makes it to any guest which
5798 * starts running in hardware virtualization between the set
5799 * and the acquisition of the spinlock, we must also ping the
5800 * CPU after setting the request bit.
5801 *
5802 */
5803
c8076604
GH		 5804 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5805 return 0;
5806 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5807 return 0;
8cfdc000
ZA		 5808
5809 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
c8076604 5810
2f303b74 5811 spin_lock(&kvm_lock);
c8076604 5812 list_for_each_entry(kvm, &vm_list, vm_list) {
988a2cae 5813 kvm_for_each_vcpu(i, vcpu, kvm) {
c8076604
GH		 5814 if (vcpu->cpu != freq->cpu)
5815 continue;
c285545f 5816 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
c8076604 5817 if (vcpu->cpu != smp_processor_id())
8cfdc000 5818 send_ipi = 1;
c8076604
GH		 5819 }
5820 }
2f303b74 5821 spin_unlock(&kvm_lock);
c8076604
GH		 5822
5823 if (freq->old < freq->new && send_ipi) {
5824 /*
5825 * We upscale the frequency. Must make the guest
5826 * doesn't see old kvmclock values while running with
5827 * the new frequency, otherwise we risk the guest sees
5828 * time go backwards.
5829 *
5830 * In case we update the frequency for another cpu
5831 * (which might be in guest context) send an interrupt
5832 * to kick the cpu out of guest context. Next time
5833 * guest context is entered kvmclock will be updated,
5834 * so the guest will not see stale values.
5835 */
8cfdc000 5836 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
c8076604
GH		 5837 }
5838 return 0;
5839}
5840
5841static struct notifier_block kvmclock_cpufreq_notifier_block = {
8cfdc000
ZA		 5842 .notifier_call = kvmclock_cpufreq_notifier
5843};
5844
251a5fd6 5845static int kvmclock_cpu_online(unsigned int cpu)
8cfdc000 5846{
251a5fd6
SAS		 5847 tsc_khz_changed(NULL);
5848 return 0;
8cfdc000
ZA		 5849}
5850
b820cc0c
ZA		 5851static void kvm_timer_init(void)
5852{
c285545f 5853 max_tsc_khz = tsc_khz;
460dd42e 5854
b820cc0c 5855 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
c285545f
ZA		 5856#ifdef CONFIG_CPU_FREQ
5857 struct cpufreq_policy policy;
758f588d
BP		 5858 int cpu;
5859
c285545f 5860 memset(&policy, 0, sizeof(policy));
3e26f230
AK		 5861 cpu = get_cpu();
5862 cpufreq_get_policy(&policy, cpu);
c285545f
ZA		 5863 if (policy.cpuinfo.max_freq)
5864 max_tsc_khz = policy.cpuinfo.max_freq;
3e26f230 5865 put_cpu();
c285545f 5866#endif
b820cc0c
ZA		 5867 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5868 CPUFREQ_TRANSITION_NOTIFIER);
5869 }
c285545f 5870 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
460dd42e 5871
73c1b41e 5872 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
251a5fd6 5873 kvmclock_cpu_online, kvmclock_cpu_down_prep);
b820cc0c
ZA		 5874}
5875
ff9d07a0
ZY		 5876static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5877
f5132b01 5878int kvm_is_in_guest(void)
ff9d07a0 5879{
086c9855 5880 return __this_cpu_read(current_vcpu) != NULL;
ff9d07a0
ZY		 5881}
5882
5883static int kvm_is_user_mode(void)
5884{
5885 int user_mode = 3;
dcf46b94 5886
086c9855
AS		 5887 if (__this_cpu_read(current_vcpu))
5888 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
dcf46b94 5889
ff9d07a0
ZY		 5890 return user_mode != 0;
5891}
5892
5893static unsigned long kvm_get_guest_ip(void)
5894{
5895 unsigned long ip = 0;
dcf46b94 5896
086c9855
AS		 5897 if (__this_cpu_read(current_vcpu))
5898 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
dcf46b94 5899
ff9d07a0
ZY		 5900 return ip;
5901}
5902
5903static struct perf_guest_info_callbacks kvm_guest_cbs = {
5904 .is_in_guest = kvm_is_in_guest,
5905 .is_user_mode = kvm_is_user_mode,
5906 .get_guest_ip = kvm_get_guest_ip,
5907};
5908
5909void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5910{
086c9855 5911 __this_cpu_write(current_vcpu, vcpu);
ff9d07a0
ZY		 5912}
5913EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5914
5915void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5916{
086c9855 5917 __this_cpu_write(current_vcpu, NULL);
ff9d07a0
ZY		 5918}
5919EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5920
ce88decf
XG		 5921static void kvm_set_mmio_spte_mask(void)
5922{
5923 u64 mask;
5924 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5925
5926 /*
5927 * Set the reserved bits and the present bit of an paging-structure
5928 * entry to generate page fault with PFER.RSV = 1.
5929 */
885032b9 5930 /* Mask the reserved physical address bits. */
d1431483 5931 mask = rsvd_bits(maxphyaddr, 51);
885032b9
XG		 5932
5933 /* Bit 62 is always reserved for 32bit host. */
5934 mask |= 0x3ull << 62;
5935
5936 /* Set the present bit. */
ce88decf
XG		 5937 mask |= 1ull;
5938
5939#ifdef CONFIG_X86_64
5940 /*
5941 * If reserved bit is not supported, clear the present bit to disable
5942 * mmio page fault.
5943 */
5944 if (maxphyaddr == 52)
5945 mask &= ~1ull;
5946#endif
5947
5948 kvm_mmu_set_mmio_spte_mask(mask);
5949}
5950
16e8d74d
MT		 5951#ifdef CONFIG_X86_64
5952static void pvclock_gtod_update_fn(struct work_struct *work)
5953{
d828199e
MT		 5954 struct kvm *kvm;
5955
5956 struct kvm_vcpu *vcpu;
5957 int i;
5958
2f303b74 5959 spin_lock(&kvm_lock);
d828199e
MT		 5960 list_for_each_entry(kvm, &vm_list, vm_list)
5961 kvm_for_each_vcpu(i, vcpu, kvm)
105b21bb 5962 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
d828199e 5963 atomic_set(&kvm_guest_has_master_clock, 0);
2f303b74 5964 spin_unlock(&kvm_lock);
16e8d74d
MT		 5965}
5966
5967static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5968
5969/*
5970 * Notification about pvclock gtod data update.
5971 */
5972static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5973 void *priv)
5974{
5975 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5976 struct timekeeper *tk = priv;
5977
5978 update_pvclock_gtod(tk);
5979
5980 /* disable master clock if host does not trust, or does not
5981 * use, TSC clocksource
5982 */
5983 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5984 atomic_read(&kvm_guest_has_master_clock) != 0)
5985 queue_work(system_long_wq, &pvclock_gtod_work);
5986
5987 return 0;
5988}
5989
5990static struct notifier_block pvclock_gtod_notifier = {
5991 .notifier_call = pvclock_gtod_notify,
5992};
5993#endif
5994
f8c16bba 5995int kvm_arch_init(void *opaque)
043405e1 5996{
b820cc0c 5997 int r;
6b61edf7 5998 struct kvm_x86_ops *ops = opaque;
f8c16bba 5999
f8c16bba
ZX		 6000 if (kvm_x86_ops) {
6001 printk(KERN_ERR "kvm: already loaded the other module\n");
56c6d28a
ZX		 6002 r = -EEXIST;
6003 goto out;
f8c16bba
ZX		 6004 }
6005
6006 if (!ops->cpu_has_kvm_support()) {
6007 printk(KERN_ERR "kvm: no hardware support\n");
56c6d28a
ZX		 6008 r = -EOPNOTSUPP;
6009 goto out;
f8c16bba
ZX		 6010 }
6011 if (ops->disabled_by_bios()) {
86a210ae 6012 printk(KERN_WARNING "kvm: disabled by bios\n");
56c6d28a
ZX		 6013 r = -EOPNOTSUPP;
6014 goto out;
f8c16bba
ZX		 6015 }
6016
013f6a5d
MT		 6017 r = -ENOMEM;
6018 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6019 if (!shared_msrs) {
6020 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6021 goto out;
6022 }
6023
97db56ce
AK		 6024 r = kvm_mmu_module_init();
6025 if (r)
013f6a5d 6026 goto out_free_percpu;
97db56ce 6027
ce88decf 6028 kvm_set_mmio_spte_mask();
97db56ce 6029
f8c16bba 6030 kvm_x86_ops = ops;
920c8377 6031
7b52345e 6032 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
ffb128c8
BD		 6033 PT_DIRTY_MASK, PT64_NX_MASK, 0,
6034 PT_PRESENT_MASK);
b820cc0c 6035 kvm_timer_init();
c8076604 6036
ff9d07a0
ZY		 6037 perf_register_guest_info_callbacks(&kvm_guest_cbs);
6038
d366bf7e 6039 if (boot_cpu_has(X86_FEATURE_XSAVE))
2acf923e
DC		 6040 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6041
c5cc421b 6042 kvm_lapic_init();
16e8d74d
MT		 6043#ifdef CONFIG_X86_64
6044 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6045#endif
6046
f8c16bba 6047 return 0;
56c6d28a 6048
013f6a5d
MT		 6049out_free_percpu:
6050 free_percpu(shared_msrs);
56c6d28a 6051out:
56c6d28a 6052 return r;
043405e1 6053}
8776e519 6054
f8c16bba
ZX		 6055void kvm_arch_exit(void)
6056{
cef84c30 6057 kvm_lapic_exit();
ff9d07a0
ZY		 6058 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6059
888d256e
JK		 6060 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6061 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6062 CPUFREQ_TRANSITION_NOTIFIER);
251a5fd6 6063 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
16e8d74d
MT		 6064#ifdef CONFIG_X86_64
6065 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6066#endif
f8c16bba 6067 kvm_x86_ops = NULL;
56c6d28a 6068 kvm_mmu_module_exit();
013f6a5d 6069 free_percpu(shared_msrs);
56c6d28a 6070}
f8c16bba 6071
5cb56059 6072int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8776e519
HB		 6073{
6074 ++vcpu->stat.halt_exits;
35754c98 6075 if (lapic_in_kernel(vcpu)) {
a4535290 6076 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
8776e519
HB		 6077 return 1;
6078 } else {
6079 vcpu->run->exit_reason = KVM_EXIT_HLT;
6080 return 0;
6081 }
6082}
5cb56059
JS		 6083EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6084
6085int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6086{
6affcbed
KH		 6087 int ret = kvm_skip_emulated_instruction(vcpu);
6088 /*
6089 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6090 * KVM_EXIT_DEBUG here.
6091 */
6092 return kvm_vcpu_halt(vcpu) && ret;
5cb56059 6093}
8776e519
HB		 6094EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6095
6aef266c
SV		 6096/*
6097 * kvm_pv_kick_cpu_op: Kick a vcpu.
6098 *
6099 * @apicid - apicid of vcpu to be kicked.
6100 */
6101static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6102{
24d2166b 6103 struct kvm_lapic_irq lapic_irq;
6aef266c 6104
24d2166b
R		 6105 lapic_irq.shorthand = 0;
6106 lapic_irq.dest_mode = 0;
6107 lapic_irq.dest_id = apicid;
93bbf0b8 6108 lapic_irq.msi_redir_hint = false;
6aef266c 6109
24d2166b 6110 lapic_irq.delivery_mode = APIC_DM_REMRD;
795a149e 6111 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6aef266c
SV		 6112}
6113
d62caabb
AS		 6114void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6115{
6116 vcpu->arch.apicv_active = false;
6117 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6118}
6119
8776e519
HB		 6120int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6121{
6122 unsigned long nr, a0, a1, a2, a3, ret;
6affcbed 6123 int op_64_bit, r;
8776e519 6124
6affcbed 6125 r = kvm_skip_emulated_instruction(vcpu);
5cb56059 6126
55cd8e5a
GN		 6127 if (kvm_hv_hypercall_enabled(vcpu->kvm))
6128 return kvm_hv_hypercall(vcpu);
6129
5fdbf976
MT		 6130 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6131 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6132 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6133 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6134 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
8776e519 6135
229456fc 6136 trace_kvm_hypercall(nr, a0, a1, a2, a3);
2714d1d3 6137
a449c7aa
NA		 6138 op_64_bit = is_64_bit_mode(vcpu);
6139 if (!op_64_bit) {
8776e519
HB		 6140 nr &= 0xFFFFFFFF;
6141 a0 &= 0xFFFFFFFF;
6142 a1 &= 0xFFFFFFFF;
6143 a2 &= 0xFFFFFFFF;
6144 a3 &= 0xFFFFFFFF;
6145 }
6146
07708c4a
JK		 6147 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6148 ret = -KVM_EPERM;
6149 goto out;
6150 }
6151
8776e519 6152 switch (nr) {
b93463aa
AK		 6153 case KVM_HC_VAPIC_POLL_IRQ:
6154 ret = 0;
6155 break;
6aef266c
SV		 6156 case KVM_HC_KICK_CPU:
6157 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6158 ret = 0;
6159 break;
8776e519
HB		 6160 default:
6161 ret = -KVM_ENOSYS;
6162 break;
6163 }
07708c4a 6164out:
a449c7aa
NA		 6165 if (!op_64_bit)
6166 ret = (u32)ret;
5fdbf976 6167 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
f11c3a8d 6168 ++vcpu->stat.hypercalls;
2f333bcb 6169 return r;
8776e519
HB		 6170}
6171EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6172
b6785def 6173static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8776e519 6174{
d6aa1000 6175 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8776e519 6176 char instruction[3];
5fdbf976 6177 unsigned long rip = kvm_rip_read(vcpu);
8776e519 6178
8776e519 6179 kvm_x86_ops->patch_hypercall(vcpu, instruction);
8776e519 6180
ce2e852e
DV		 6181 return emulator_write_emulated(ctxt, rip, instruction, 3,
6182 &ctxt->exception);
8776e519
HB		 6183}
6184
851ba692 6185static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
b6c7a5dc 6186{
782d422b
MG		 6187 return vcpu->run->request_interrupt_window &&
6188 likely(!pic_in_kernel(vcpu->kvm));
b6c7a5dc
HB		 6189}
6190
851ba692 6191static void post_kvm_run_save(struct kvm_vcpu *vcpu)
b6c7a5dc 6192{
851ba692
AK		 6193 struct kvm_run *kvm_run = vcpu->run;
6194
91586a3b 6195 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
f077825a 6196 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
2d3ad1f4 6197 kvm_run->cr8 = kvm_get_cr8(vcpu);
b6c7a5dc 6198 kvm_run->apic_base = kvm_get_apic_base(vcpu);
127a457a
MG		 6199 kvm_run->ready_for_interrupt_injection =
6200 pic_in_kernel(vcpu->kvm) ||
782d422b 6201 kvm_vcpu_ready_for_interrupt_injection(vcpu);
b6c7a5dc
HB		 6202}
6203
95ba8273
GN		 6204static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6205{
6206 int max_irr, tpr;
6207
6208 if (!kvm_x86_ops->update_cr8_intercept)
6209 return;
6210
bce87cce 6211 if (!lapic_in_kernel(vcpu))
88c808fd
AK		 6212 return;
6213
d62caabb
AS		 6214 if (vcpu->arch.apicv_active)
6215 return;
6216
8db3baa2
GN		 6217 if (!vcpu->arch.apic->vapic_addr)
6218 max_irr = kvm_lapic_find_highest_irr(vcpu);
6219 else
6220 max_irr = -1;
95ba8273
GN		 6221
6222 if (max_irr != -1)
6223 max_irr >>= 4;
6224
6225 tpr = kvm_lapic_get_cr8(vcpu);
6226
6227 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6228}
6229
b6b8a145 6230static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
95ba8273 6231{
b6b8a145
JK		 6232 int r;
6233
95ba8273 6234 /* try to reinject previous events if any */
b59bb7bd 6235 if (vcpu->arch.exception.pending) {
5c1c85d0
AK		 6236 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6237 vcpu->arch.exception.has_error_code,
6238 vcpu->arch.exception.error_code);
d6e8c854
NA		 6239
6240 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6241 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6242 X86_EFLAGS_RF);
6243
6bdf0662
NA		 6244 if (vcpu->arch.exception.nr == DB_VECTOR &&
6245 (vcpu->arch.dr7 & DR7_GD)) {
6246 vcpu->arch.dr7 &= ~DR7_GD;
6247 kvm_update_dr7(vcpu);
6248 }
6249
b59bb7bd
GN		 6250 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6251 vcpu->arch.exception.has_error_code,
ce7ddec4
JR		 6252 vcpu->arch.exception.error_code,
6253 vcpu->arch.exception.reinject);
b6b8a145 6254 return 0;
b59bb7bd
GN		 6255 }
6256
95ba8273
GN		 6257 if (vcpu->arch.nmi_injected) {
6258 kvm_x86_ops->set_nmi(vcpu);
b6b8a145 6259 return 0;
95ba8273
GN		 6260 }
6261
6262 if (vcpu->arch.interrupt.pending) {
66fd3f7f 6263 kvm_x86_ops->set_irq(vcpu);
b6b8a145
JK		 6264 return 0;
6265 }
6266
6267 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6268 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6269 if (r != 0)
6270 return r;
95ba8273
GN		 6271 }
6272
6273 /* try to inject new event if pending */
c43203ca
PB		 6274 if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
6275 vcpu->arch.smi_pending = false;
ee2cd4b7 6276 enter_smm(vcpu);
c43203ca 6277 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
321c5658
YS		 6278 --vcpu->arch.nmi_pending;
6279 vcpu->arch.nmi_injected = true;
6280 kvm_x86_ops->set_nmi(vcpu);
c7c9c56c 6281 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
9242b5b6
BD		 6282 /*
6283 * Because interrupts can be injected asynchronously, we are
6284 * calling check_nested_events again here to avoid a race condition.
6285 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6286 * proposal and current concerns. Perhaps we should be setting
6287 * KVM_REQ_EVENT only on certain events and not unconditionally?
6288 */
6289 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6290 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6291 if (r != 0)
6292 return r;
6293 }
95ba8273 6294 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
66fd3f7f
GN		 6295 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6296 false);
6297 kvm_x86_ops->set_irq(vcpu);
95ba8273
GN		 6298 }
6299 }
ee2cd4b7 6300
b6b8a145 6301 return 0;
95ba8273
GN		 6302}
6303
7460fb4a
AK		 6304static void process_nmi(struct kvm_vcpu *vcpu)
6305{
6306 unsigned limit = 2;
6307
6308 /*
6309 * x86 is limited to one NMI running, and one NMI pending after it.
6310 * If an NMI is already in progress, limit further NMIs to just one.
6311 * Otherwise, allow two (and we'll inject the first one immediately).
6312 */
6313 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6314 limit = 1;
6315
6316 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6317 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6318 kvm_make_request(KVM_REQ_EVENT, vcpu);
6319}
6320
660a5d51
PB		 6321#define put_smstate(type, buf, offset, val) \
6322 *(type *)((buf) + (offset) - 0x7e00) = val
6323
ee2cd4b7 6324static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
660a5d51
PB		 6325{
6326 u32 flags = 0;
6327 flags |= seg->g << 23;
6328 flags |= seg->db << 22;
6329 flags |= seg->l << 21;
6330 flags |= seg->avl << 20;
6331 flags |= seg->present << 15;
6332 flags |= seg->dpl << 13;
6333 flags |= seg->s << 12;
6334 flags |= seg->type << 8;
6335 return flags;
6336}
6337
ee2cd4b7 6338static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
660a5d51
PB		 6339{
6340 struct kvm_segment seg;
6341 int offset;
6342
6343 kvm_get_segment(vcpu, &seg, n);
6344 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6345
6346 if (n < 3)
6347 offset = 0x7f84 + n * 12;
6348 else
6349 offset = 0x7f2c + (n - 3) * 12;
6350
6351 put_smstate(u32, buf, offset + 8, seg.base);
6352 put_smstate(u32, buf, offset + 4, seg.limit);
ee2cd4b7 6353 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
660a5d51
PB		 6354}
6355
efbb288a 6356#ifdef CONFIG_X86_64
ee2cd4b7 6357static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
660a5d51
PB		 6358{
6359 struct kvm_segment seg;
6360 int offset;
6361 u16 flags;
6362
6363 kvm_get_segment(vcpu, &seg, n);
6364 offset = 0x7e00 + n * 16;
6365
ee2cd4b7 6366 flags = enter_smm_get_segment_flags(&seg) >> 8;
660a5d51
PB		 6367 put_smstate(u16, buf, offset, seg.selector);
6368 put_smstate(u16, buf, offset + 2, flags);
6369 put_smstate(u32, buf, offset + 4, seg.limit);
6370 put_smstate(u64, buf, offset + 8, seg.base);
6371}
efbb288a 6372#endif
660a5d51 6373
ee2cd4b7 6374static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
660a5d51
PB		 6375{
6376 struct desc_ptr dt;
6377 struct kvm_segment seg;
6378 unsigned long val;
6379 int i;
6380
6381 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6382 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6383 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6384 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6385
6386 for (i = 0; i < 8; i++)
6387 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6388
6389 kvm_get_dr(vcpu, 6, &val);
6390 put_smstate(u32, buf, 0x7fcc, (u32)val);
6391 kvm_get_dr(vcpu, 7, &val);
6392 put_smstate(u32, buf, 0x7fc8, (u32)val);
6393
6394 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6395 put_smstate(u32, buf, 0x7fc4, seg.selector);
6396 put_smstate(u32, buf, 0x7f64, seg.base);
6397 put_smstate(u32, buf, 0x7f60, seg.limit);
ee2cd4b7 6398 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
660a5d51
PB		 6399
6400 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6401 put_smstate(u32, buf, 0x7fc0, seg.selector);
6402 put_smstate(u32, buf, 0x7f80, seg.base);
6403 put_smstate(u32, buf, 0x7f7c, seg.limit);
ee2cd4b7 6404 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
660a5d51
PB		 6405
6406 kvm_x86_ops->get_gdt(vcpu, &dt);
6407 put_smstate(u32, buf, 0x7f74, dt.address);
6408 put_smstate(u32, buf, 0x7f70, dt.size);
6409
6410 kvm_x86_ops->get_idt(vcpu, &dt);
6411 put_smstate(u32, buf, 0x7f58, dt.address);
6412 put_smstate(u32, buf, 0x7f54, dt.size);
6413
6414 for (i = 0; i < 6; i++)
ee2cd4b7 6415 enter_smm_save_seg_32(vcpu, buf, i);
660a5d51
PB		 6416
6417 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6418
6419 /* revision id */
6420 put_smstate(u32, buf, 0x7efc, 0x00020000);
6421 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6422}
6423
ee2cd4b7 6424static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
660a5d51
PB		 6425{
6426#ifdef CONFIG_X86_64
6427 struct desc_ptr dt;
6428 struct kvm_segment seg;
6429 unsigned long val;
6430 int i;
6431
6432 for (i = 0; i < 16; i++)
6433 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6434
6435 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6436 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6437
6438 kvm_get_dr(vcpu, 6, &val);
6439 put_smstate(u64, buf, 0x7f68, val);
6440 kvm_get_dr(vcpu, 7, &val);
6441 put_smstate(u64, buf, 0x7f60, val);
6442
6443 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6444 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6445 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6446
6447 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6448
6449 /* revision id */
6450 put_smstate(u32, buf, 0x7efc, 0x00020064);
6451
6452 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6453
6454 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6455 put_smstate(u16, buf, 0x7e90, seg.selector);
ee2cd4b7 6456 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
660a5d51
PB		 6457 put_smstate(u32, buf, 0x7e94, seg.limit);
6458 put_smstate(u64, buf, 0x7e98, seg.base);
6459
6460 kvm_x86_ops->get_idt(vcpu, &dt);
6461 put_smstate(u32, buf, 0x7e84, dt.size);
6462 put_smstate(u64, buf, 0x7e88, dt.address);
6463
6464 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6465 put_smstate(u16, buf, 0x7e70, seg.selector);
ee2cd4b7 6466 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
660a5d51
PB		 6467 put_smstate(u32, buf, 0x7e74, seg.limit);
6468 put_smstate(u64, buf, 0x7e78, seg.base);
6469
6470 kvm_x86_ops->get_gdt(vcpu, &dt);
6471 put_smstate(u32, buf, 0x7e64, dt.size);
6472 put_smstate(u64, buf, 0x7e68, dt.address);
6473
6474 for (i = 0; i < 6; i++)
ee2cd4b7 6475 enter_smm_save_seg_64(vcpu, buf, i);
660a5d51
PB		 6476#else
6477 WARN_ON_ONCE(1);
6478#endif
6479}
6480
ee2cd4b7 6481static void enter_smm(struct kvm_vcpu *vcpu)
64d60670 6482{
660a5d51 6483 struct kvm_segment cs, ds;
18c3626e 6484 struct desc_ptr dt;
660a5d51
PB		 6485 char buf[512];
6486 u32 cr0;
6487
660a5d51
PB		 6488 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6489 vcpu->arch.hflags |= HF_SMM_MASK;
6490 memset(buf, 0, 512);
6491 if (guest_cpuid_has_longmode(vcpu))
ee2cd4b7 6492 enter_smm_save_state_64(vcpu, buf);
660a5d51 6493 else
ee2cd4b7 6494 enter_smm_save_state_32(vcpu, buf);
660a5d51 6495
54bf36aa 6496 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
660a5d51
PB		 6497
6498 if (kvm_x86_ops->get_nmi_mask(vcpu))
6499 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6500 else
6501 kvm_x86_ops->set_nmi_mask(vcpu, true);
6502
6503 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6504 kvm_rip_write(vcpu, 0x8000);
6505
6506 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6507 kvm_x86_ops->set_cr0(vcpu, cr0);
6508 vcpu->arch.cr0 = cr0;
6509
6510 kvm_x86_ops->set_cr4(vcpu, 0);
6511
18c3626e
PB		 6512 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6513 dt.address = dt.size = 0;
6514 kvm_x86_ops->set_idt(vcpu, &dt);
6515
660a5d51
PB		 6516 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6517
6518 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6519 cs.base = vcpu->arch.smbase;
6520
6521 ds.selector = 0;
6522 ds.base = 0;
6523
6524 cs.limit = ds.limit = 0xffffffff;
6525 cs.type = ds.type = 0x3;
6526 cs.dpl = ds.dpl = 0;
6527 cs.db = ds.db = 0;
6528 cs.s = ds.s = 1;
6529 cs.l = ds.l = 0;
6530 cs.g = ds.g = 1;
6531 cs.avl = ds.avl = 0;
6532 cs.present = ds.present = 1;
6533 cs.unusable = ds.unusable = 0;
6534 cs.padding = ds.padding = 0;
6535
6536 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6537 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6538 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6539 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6540 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6541 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6542
6543 if (guest_cpuid_has_longmode(vcpu))
6544 kvm_x86_ops->set_efer(vcpu, 0);
6545
6546 kvm_update_cpuid(vcpu);
6547 kvm_mmu_reset_context(vcpu);
64d60670
PB		 6548}
6549
ee2cd4b7 6550static void process_smi(struct kvm_vcpu *vcpu)
c43203ca
PB		 6551{
6552 vcpu->arch.smi_pending = true;
6553 kvm_make_request(KVM_REQ_EVENT, vcpu);
6554}
6555
2860c4b1
PB		 6556void kvm_make_scan_ioapic_request(struct kvm *kvm)
6557{
6558 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6559}
6560
3d81bc7e 6561static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
c7c9c56c 6562{
5c919412
AS		 6563 u64 eoi_exit_bitmap[4];
6564
3d81bc7e
YZ		 6565 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6566 return;
c7c9c56c 6567
6308630b 6568 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
c7c9c56c 6569
b053b2ae 6570 if (irqchip_split(vcpu->kvm))
6308630b 6571 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
db2bdcbb 6572 else {
d62caabb
AS		 6573 if (vcpu->arch.apicv_active)
6574 kvm_x86_ops->sync_pir_to_irr(vcpu);
6308630b 6575 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
db2bdcbb 6576 }
5c919412
AS		 6577 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6578 vcpu_to_synic(vcpu)->vec_bitmap, 256);
6579 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
c7c9c56c
YZ		 6580}
6581
a70656b6
RK		 6582static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6583{
6584 ++vcpu->stat.tlb_flush;
6585 kvm_x86_ops->tlb_flush(vcpu);
6586}
6587
4256f43f
TC		 6588void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6589{
c24ae0dc
TC		 6590 struct page *page = NULL;
6591
35754c98 6592 if (!lapic_in_kernel(vcpu))
f439ed27
PB		 6593 return;
6594
4256f43f
TC		 6595 if (!kvm_x86_ops->set_apic_access_page_addr)
6596 return;
6597
c24ae0dc 6598 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
e8fd5e9e
AA		 6599 if (is_error_page(page))
6600 return;
c24ae0dc
TC		 6601 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6602
6603 /*
6604 * Do not pin apic access page in memory, the MMU notifier
6605 * will call us again if it is migrated or swapped out.
6606 */
6607 put_page(page);
4256f43f
TC		 6608}
6609EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6610
fe71557a
TC		 6611void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6612 unsigned long address)
6613{
c24ae0dc
TC		 6614 /*
6615 * The physical address of apic access page is stored in the VMCS.
6616 * Update it when it becomes invalid.
6617 */
6618 if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6619 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
fe71557a
TC		 6620}
6621
9357d939 6622/*
362c698f 6623 * Returns 1 to let vcpu_run() continue the guest execution loop without
9357d939
TY		 6624 * exiting to the userspace. Otherwise, the value will be returned to the
6625 * userspace.
6626 */
851ba692 6627static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
b6c7a5dc
HB		 6628{
6629 int r;
62a193ed
MG		 6630 bool req_int_win =
6631 dm_request_for_irq_injection(vcpu) &&
6632 kvm_cpu_accept_dm_intr(vcpu);
6633
730dca42 6634 bool req_immediate_exit = false;
b6c7a5dc 6635
3e007509 6636 if (vcpu->requests) {
a8eeb04a 6637 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
2e53d63a 6638 kvm_mmu_unload(vcpu);
a8eeb04a 6639 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
2f599714 6640 __kvm_migrate_timers(vcpu);
d828199e
MT		 6641 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6642 kvm_gen_update_masterclock(vcpu->kvm);
0061d53d
MT		 6643 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6644 kvm_gen_kvmclock_update(vcpu);
34c238a1
ZA		 6645 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6646 r = kvm_guest_time_update(vcpu);
8cfdc000
ZA		 6647 if (unlikely(r))
6648 goto out;
6649 }
a8eeb04a 6650 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
4731d4c7 6651 kvm_mmu_sync_roots(vcpu);
a8eeb04a 6652 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
a70656b6 6653 kvm_vcpu_flush_tlb(vcpu);
a8eeb04a 6654 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
851ba692 6655 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
b93463aa
AK		 6656 r = 0;
6657 goto out;
6658 }
a8eeb04a 6659 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
851ba692 6660 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
71c4dfaf
JR		 6661 r = 0;
6662 goto out;
6663 }
a8eeb04a 6664 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
02daab21
AK		 6665 vcpu->fpu_active = 0;
6666 kvm_x86_ops->fpu_deactivate(vcpu);
6667 }
af585b92
GN		 6668 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6669 /* Page is swapped out. Do synthetic halt */
6670 vcpu->arch.apf.halted = true;
6671 r = 1;
6672 goto out;
6673 }
c9aaa895
GC		 6674 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6675 record_steal_time(vcpu);
64d60670
PB		 6676 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6677 process_smi(vcpu);
7460fb4a
AK		 6678 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6679 process_nmi(vcpu);
f5132b01 6680 if (kvm_check_request(KVM_REQ_PMU, vcpu))
c6702c9d 6681 kvm_pmu_handle_event(vcpu);
f5132b01 6682 if (kvm_check_request(KVM_REQ_PMI, vcpu))
c6702c9d 6683 kvm_pmu_deliver_pmi(vcpu);
7543a635
SR		 6684 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6685 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6686 if (test_bit(vcpu->arch.pending_ioapic_eoi,
6308630b 6687 vcpu->arch.ioapic_handled_vectors)) {
7543a635
SR		 6688 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6689 vcpu->run->eoi.vector =
6690 vcpu->arch.pending_ioapic_eoi;
6691 r = 0;
6692 goto out;
6693 }
6694 }
3d81bc7e
YZ		 6695 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6696 vcpu_scan_ioapic(vcpu);
4256f43f
TC		 6697 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6698 kvm_vcpu_reload_apic_access_page(vcpu);
2ce79189
AS		 6699 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6700 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6701 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6702 r = 0;
6703 goto out;
6704 }
e516cebb
AS		 6705 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6706 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6707 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6708 r = 0;
6709 goto out;
6710 }
db397571
AS		 6711 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6712 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6713 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6714 r = 0;
6715 goto out;
6716 }
f3b138c5
AS		 6717
6718 /*
6719 * KVM_REQ_HV_STIMER has to be processed after
6720 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6721 * depend on the guest clock being up-to-date
6722 */
1f4b34f8
AS		 6723 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6724 kvm_hv_process_stimers(vcpu);
2f52d58c 6725 }
b93463aa 6726
bf9f6ac8
FW		 6727 /*
6728 * KVM_REQ_EVENT is not set when posted interrupts are set by
6729 * VT-d hardware, so we have to update RVI unconditionally.
6730 */
6731 if (kvm_lapic_enabled(vcpu)) {
6732 /*
6733 * Update architecture specific hints for APIC
6734 * virtual interrupt delivery.
6735 */
d62caabb 6736 if (vcpu->arch.apicv_active)
bf9f6ac8
FW		 6737 kvm_x86_ops->hwapic_irr_update(vcpu,
6738 kvm_lapic_find_highest_irr(vcpu));
2f52d58c 6739 }
b93463aa 6740
b463a6f7 6741 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
66450a21
JK		 6742 kvm_apic_accept_events(vcpu);
6743 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6744 r = 1;
6745 goto out;
6746 }
6747
b6b8a145
JK		 6748 if (inject_pending_event(vcpu, req_int_win) != 0)
6749 req_immediate_exit = true;
321c5658 6750 else {
c43203ca
PB		 6751 /* Enable NMI/IRQ window open exits if needed.
6752 *
6753 * SMIs have two cases: 1) they can be nested, and
6754 * then there is nothing to do here because RSM will
6755 * cause a vmexit anyway; 2) or the SMI can be pending
6756 * because inject_pending_event has completed the
6757 * injection of an IRQ or NMI from the previous vmexit,
6758 * and then we request an immediate exit to inject the SMI.
6759 */
6760 if (vcpu->arch.smi_pending && !is_smm(vcpu))
6761 req_immediate_exit = true;
321c5658
YS		 6762 if (vcpu->arch.nmi_pending)
6763 kvm_x86_ops->enable_nmi_window(vcpu);
6764 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6765 kvm_x86_ops->enable_irq_window(vcpu);
6766 }
b463a6f7
AK		 6767
6768 if (kvm_lapic_enabled(vcpu)) {
6769 update_cr8_intercept(vcpu);
6770 kvm_lapic_sync_to_vapic(vcpu);
6771 }
6772 }
6773
d8368af8
AK		 6774 r = kvm_mmu_reload(vcpu);
6775 if (unlikely(r)) {
d905c069 6776 goto cancel_injection;
d8368af8
AK		 6777 }
6778
b6c7a5dc
HB		 6779 preempt_disable();
6780
6781 kvm_x86_ops->prepare_guest_switch(vcpu);
2608d7a1
AK		 6782 if (vcpu->fpu_active)
6783 kvm_load_guest_fpu(vcpu);
6b7e2d09
XG		 6784 vcpu->mode = IN_GUEST_MODE;
6785
01b71917
MT		 6786 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6787
0f127d12
LT		 6788 /*
6789 * We should set ->mode before check ->requests,
6790 * Please see the comment in kvm_make_all_cpus_request.
6791 * This also orders the write to mode from any reads
6792 * to the page tables done while the VCPU is running.
6793 * Please see the comment in kvm_flush_remote_tlbs.
6b7e2d09 6794 */
01b71917 6795 smp_mb__after_srcu_read_unlock();
b6c7a5dc 6796
d94e1dc9 6797 local_irq_disable();
32f88400 6798
6b7e2d09 6799 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
d94e1dc9 6800 || need_resched() || signal_pending(current)) {
6b7e2d09 6801 vcpu->mode = OUTSIDE_GUEST_MODE;
d94e1dc9 6802 smp_wmb();
6c142801
AK		 6803 local_irq_enable();
6804 preempt_enable();
01b71917 6805 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6c142801 6806 r = 1;
d905c069 6807 goto cancel_injection;
6c142801
AK		 6808 }
6809
fc5b7f3b
DM		 6810 kvm_load_guest_xcr0(vcpu);
6811
c43203ca
PB		 6812 if (req_immediate_exit) {
6813 kvm_make_request(KVM_REQ_EVENT, vcpu);
d6185f20 6814 smp_send_reschedule(vcpu->cpu);
c43203ca 6815 }
d6185f20 6816
8b89fe1f
PB		 6817 trace_kvm_entry(vcpu->vcpu_id);
6818 wait_lapic_expire(vcpu);
6edaa530 6819 guest_enter_irqoff();
b6c7a5dc 6820
42dbaa5a 6821 if (unlikely(vcpu->arch.switch_db_regs)) {
42dbaa5a
JK		 6822 set_debugreg(0, 7);
6823 set_debugreg(vcpu->arch.eff_db[0], 0);
6824 set_debugreg(vcpu->arch.eff_db[1], 1);
6825 set_debugreg(vcpu->arch.eff_db[2], 2);
6826 set_debugreg(vcpu->arch.eff_db[3], 3);
c77fb5fe 6827 set_debugreg(vcpu->arch.dr6, 6);
ae561ede 6828 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
42dbaa5a 6829 }
b6c7a5dc 6830
851ba692 6831 kvm_x86_ops->run(vcpu);
b6c7a5dc 6832
c77fb5fe
PB		 6833 /*
6834 * Do this here before restoring debug registers on the host. And
6835 * since we do this before handling the vmexit, a DR access vmexit
6836 * can (a) read the correct value of the debug registers, (b) set
6837 * KVM_DEBUGREG_WONT_EXIT again.
6838 */
6839 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
c77fb5fe
PB		 6840 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6841 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
70e4da7a
PB		 6842 kvm_update_dr0123(vcpu);
6843 kvm_update_dr6(vcpu);
6844 kvm_update_dr7(vcpu);
6845 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
c77fb5fe
PB		 6846 }
6847
24f1e32c
FW		 6848 /*
6849 * If the guest has used debug registers, at least dr7
6850 * will be disabled while returning to the host.
6851 * If we don't have active breakpoints in the host, we don't
6852 * care about the messed up debug address registers. But if
6853 * we have some of them active, restore the old state.
6854 */
59d8eb53 6855 if (hw_breakpoint_active())
24f1e32c 6856 hw_breakpoint_restore();
42dbaa5a 6857
4ba76538 6858 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1d5f066e 6859
6b7e2d09 6860 vcpu->mode = OUTSIDE_GUEST_MODE;
d94e1dc9 6861 smp_wmb();
a547c6db 6862
fc5b7f3b
DM		 6863 kvm_put_guest_xcr0(vcpu);
6864
a547c6db 6865 kvm_x86_ops->handle_external_intr(vcpu);
b6c7a5dc
HB		 6866
6867 ++vcpu->stat.exits;
6868
f2485b3e 6869 guest_exit_irqoff();
b6c7a5dc 6870
f2485b3e 6871 local_irq_enable();
b6c7a5dc
HB		 6872 preempt_enable();
6873
f656ce01 6874 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3200f405 6875
b6c7a5dc
HB		 6876 /*
6877 * Profile KVM exit RIPs:
6878 */
6879 if (unlikely(prof_on == KVM_PROFILING)) {
5fdbf976
MT		 6880 unsigned long rip = kvm_rip_read(vcpu);
6881 profile_hit(KVM_PROFILING, (void *)rip);
b6c7a5dc
HB		 6882 }
6883
cc578287
ZA		 6884 if (unlikely(vcpu->arch.tsc_always_catchup))
6885 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
298101da 6886
5cfb1d5a
MT		 6887 if (vcpu->arch.apic_attention)
6888 kvm_lapic_sync_from_vapic(vcpu);
b93463aa 6889
851ba692 6890 r = kvm_x86_ops->handle_exit(vcpu);
d905c069
MT		 6891 return r;
6892
6893cancel_injection:
6894 kvm_x86_ops->cancel_injection(vcpu);
ae7a2a3f
MT		 6895 if (unlikely(vcpu->arch.apic_attention))
6896 kvm_lapic_sync_from_vapic(vcpu);
d7690175
MT		 6897out:
6898 return r;
6899}
b6c7a5dc 6900
362c698f
PB		 6901static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6902{
bf9f6ac8
FW		 6903 if (!kvm_arch_vcpu_runnable(vcpu) &&
6904 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
9c8fd1ba
PB		 6905 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6906 kvm_vcpu_block(vcpu);
6907 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
bf9f6ac8
FW		 6908
6909 if (kvm_x86_ops->post_block)
6910 kvm_x86_ops->post_block(vcpu);
6911
9c8fd1ba
PB		 6912 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6913 return 1;
6914 }
362c698f
PB		 6915
6916 kvm_apic_accept_events(vcpu);
6917 switch(vcpu->arch.mp_state) {
6918 case KVM_MP_STATE_HALTED:
6919 vcpu->arch.pv.pv_unhalted = false;
6920 vcpu->arch.mp_state =
6921 KVM_MP_STATE_RUNNABLE;
6922 case KVM_MP_STATE_RUNNABLE:
6923 vcpu->arch.apf.halted = false;
6924 break;
6925 case KVM_MP_STATE_INIT_RECEIVED:
6926 break;
6927 default:
6928 return -EINTR;
6929 break;
6930 }
6931 return 1;
6932}
09cec754 6933
5d9bc648
PB		 6934static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
6935{
6936 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6937 !vcpu->arch.apf.halted);
6938}
6939
362c698f 6940static int vcpu_run(struct kvm_vcpu *vcpu)
d7690175
MT		 6941{
6942 int r;
f656ce01 6943 struct kvm *kvm = vcpu->kvm;
d7690175 6944
f656ce01 6945 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
d7690175 6946
362c698f 6947 for (;;) {
58f800d5 6948 if (kvm_vcpu_running(vcpu)) {
851ba692 6949 r = vcpu_enter_guest(vcpu);
bf9f6ac8 6950 } else {
362c698f 6951 r = vcpu_block(kvm, vcpu);
bf9f6ac8
FW		 6952 }
6953
09cec754
GN		 6954 if (r <= 0)
6955 break;
6956
6957 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6958 if (kvm_cpu_has_pending_timer(vcpu))
6959 kvm_inject_pending_timer_irqs(vcpu);
6960
782d422b
MG		 6961 if (dm_request_for_irq_injection(vcpu) &&
6962 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
4ca7dd8c
PB		 6963 r = 0;
6964 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
09cec754 6965 ++vcpu->stat.request_irq_exits;
362c698f 6966 break;
09cec754 6967 }
af585b92
GN		 6968
6969 kvm_check_async_pf_completion(vcpu);
6970
09cec754
GN		 6971 if (signal_pending(current)) {
6972 r = -EINTR;
851ba692 6973 vcpu->run->exit_reason = KVM_EXIT_INTR;
09cec754 6974 ++vcpu->stat.signal_exits;
362c698f 6975 break;
09cec754
GN		 6976 }
6977 if (need_resched()) {
f656ce01 6978 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
c08ac06a 6979 cond_resched();
f656ce01 6980 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
d7690175 6981 }
b6c7a5dc
HB		 6982 }
6983
f656ce01 6984 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
b6c7a5dc
HB		 6985
6986 return r;
6987}
6988
716d51ab
GN		 6989static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6990{
6991 int r;
6992 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6993 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6994 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6995 if (r != EMULATE_DONE)
6996 return 0;
6997 return 1;
6998}
6999
7000static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7001{
7002 BUG_ON(!vcpu->arch.pio.count);
7003
7004 return complete_emulated_io(vcpu);
7005}
7006
f78146b0
AK		 7007/*
7008 * Implements the following, as a state machine:
7009 *
7010 * read:
7011 * for each fragment
87da7e66
XG		 7012 * for each mmio piece in the fragment
7013 * write gpa, len
7014 * exit
7015 * copy data
f78146b0
AK		 7016 * execute insn
7017 *
7018 * write:
7019 * for each fragment
87da7e66
XG		 7020 * for each mmio piece in the fragment
7021 * write gpa, len
7022 * copy data
7023 * exit
f78146b0 7024 */
716d51ab 7025static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
5287f194
AK		 7026{
7027 struct kvm_run *run = vcpu->run;
f78146b0 7028 struct kvm_mmio_fragment *frag;
87da7e66 7029 unsigned len;
5287f194 7030
716d51ab 7031 BUG_ON(!vcpu->mmio_needed);
5287f194 7032
716d51ab 7033 /* Complete previous fragment */
87da7e66
XG		 7034 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7035 len = min(8u, frag->len);
716d51ab 7036 if (!vcpu->mmio_is_write)
87da7e66
XG		 7037 memcpy(frag->data, run->mmio.data, len);
7038
7039 if (frag->len <= 8) {
7040 /* Switch to the next fragment. */
7041 frag++;
7042 vcpu->mmio_cur_fragment++;
7043 } else {
7044 /* Go forward to the next mmio piece. */
7045 frag->data += len;
7046 frag->gpa += len;
7047 frag->len -= len;
7048 }
7049
a08d3b3b 7050 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
716d51ab 7051 vcpu->mmio_needed = 0;
0912c977
PB		 7052
7053 /* FIXME: return into emulator if single-stepping. */
cef4dea0 7054 if (vcpu->mmio_is_write)
716d51ab
GN		 7055 return 1;
7056 vcpu->mmio_read_completed = 1;
7057 return complete_emulated_io(vcpu);
7058 }
87da7e66 7059
716d51ab
GN		 7060 run->exit_reason = KVM_EXIT_MMIO;
7061 run->mmio.phys_addr = frag->gpa;
7062 if (vcpu->mmio_is_write)
87da7e66
XG		 7063 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7064 run->mmio.len = min(8u, frag->len);
716d51ab
GN		 7065 run->mmio.is_write = vcpu->mmio_is_write;
7066 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7067 return 0;
5287f194
AK		 7068}
7069
716d51ab 7070
b6c7a5dc
HB		 7071int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7072{
c5bedc68 7073 struct fpu *fpu = &current->thread.fpu;
b6c7a5dc
HB		 7074 int r;
7075 sigset_t sigsaved;
7076
c4d72e2d 7077 fpu__activate_curr(fpu);
e5c30142 7078
ac9f6dc0
AK		 7079 if (vcpu->sigset_active)
7080 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
7081
a4535290 7082 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
b6c7a5dc 7083 kvm_vcpu_block(vcpu);
66450a21 7084 kvm_apic_accept_events(vcpu);
d7690175 7085 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
ac9f6dc0
AK		 7086 r = -EAGAIN;
7087 goto out;
b6c7a5dc
HB		 7088 }
7089
b6c7a5dc 7090 /* re-sync apic's tpr */
35754c98 7091 if (!lapic_in_kernel(vcpu)) {
eea1cff9
AP		 7092 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7093 r = -EINVAL;
7094 goto out;
7095 }
7096 }
b6c7a5dc 7097
716d51ab
GN		 7098 if (unlikely(vcpu->arch.complete_userspace_io)) {
7099 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7100 vcpu->arch.complete_userspace_io = NULL;
7101 r = cui(vcpu);
7102 if (r <= 0)
7103 goto out;
7104 } else
7105 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
5287f194 7106
362c698f 7107 r = vcpu_run(vcpu);
b6c7a5dc
HB		 7108
7109out:
f1d86e46 7110 post_kvm_run_save(vcpu);
b6c7a5dc
HB		 7111 if (vcpu->sigset_active)
7112 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
7113
b6c7a5dc
HB		 7114 return r;
7115}
7116
7117int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7118{
7ae441ea
GN		 7119 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
7120 /*
7121 * We are here if userspace calls get_regs() in the middle of
7122 * instruction emulation. Registers state needs to be copied
4a969980 7123 * back from emulation context to vcpu. Userspace shouldn't do
7ae441ea
GN		 7124 * that usually, but some bad designed PV devices (vmware
7125 * backdoor interface) need this to work
7126 */
dd856efa 7127 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7ae441ea
GN		 7128 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7129 }
5fdbf976
MT		 7130 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
7131 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
7132 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
7133 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
7134 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
7135 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
7136 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7137 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
b6c7a5dc 7138#ifdef CONFIG_X86_64
5fdbf976
MT		 7139 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7140 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7141 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7142 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7143 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7144 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7145 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7146 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
b6c7a5dc
HB		 7147#endif
7148
5fdbf976 7149 regs->rip = kvm_rip_read(vcpu);
91586a3b 7150 regs->rflags = kvm_get_rflags(vcpu);
b6c7a5dc 7151
b6c7a5dc
HB		 7152 return 0;
7153}
7154
7155int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7156{
7ae441ea
GN		 7157 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7158 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7159
5fdbf976
MT		 7160 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7161 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7162 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7163 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7164 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7165 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7166 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7167 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
b6c7a5dc 7168#ifdef CONFIG_X86_64
5fdbf976
MT		 7169 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7170 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7171 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7172 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7173 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7174 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7175 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7176 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
b6c7a5dc
HB		 7177#endif
7178
5fdbf976 7179 kvm_rip_write(vcpu, regs->rip);
91586a3b 7180 kvm_set_rflags(vcpu, regs->rflags);
b6c7a5dc 7181
b4f14abd
JK		 7182 vcpu->arch.exception.pending = false;
7183
3842d135
AK		 7184 kvm_make_request(KVM_REQ_EVENT, vcpu);
7185
b6c7a5dc
HB		 7186 return 0;
7187}
7188
b6c7a5dc
HB		 7189void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7190{
7191 struct kvm_segment cs;
7192
3e6e0aab 7193 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
b6c7a5dc
HB		 7194 *db = cs.db;
7195 *l = cs.l;
7196}
7197EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7198
7199int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7200 struct kvm_sregs *sregs)
7201{
89a27f4d 7202 struct desc_ptr dt;
b6c7a5dc 7203
3e6e0aab
GT		 7204 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7205 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7206 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7207 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7208 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7209 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
b6c7a5dc 7210
3e6e0aab
GT		 7211 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7212 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
b6c7a5dc
HB		 7213
7214 kvm_x86_ops->get_idt(vcpu, &dt);
89a27f4d
GN		 7215 sregs->idt.limit = dt.size;
7216 sregs->idt.base = dt.address;
b6c7a5dc 7217 kvm_x86_ops->get_gdt(vcpu, &dt);
89a27f4d
GN		 7218 sregs->gdt.limit = dt.size;
7219 sregs->gdt.base = dt.address;
b6c7a5dc 7220
4d4ec087 7221 sregs->cr0 = kvm_read_cr0(vcpu);
ad312c7c 7222 sregs->cr2 = vcpu->arch.cr2;
9f8fe504 7223 sregs->cr3 = kvm_read_cr3(vcpu);
fc78f519 7224 sregs->cr4 = kvm_read_cr4(vcpu);
2d3ad1f4 7225 sregs->cr8 = kvm_get_cr8(vcpu);
f6801dff 7226 sregs->efer = vcpu->arch.efer;
b6c7a5dc
HB		 7227 sregs->apic_base = kvm_get_apic_base(vcpu);
7228
923c61bb 7229 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
b6c7a5dc 7230
36752c9b 7231 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
14d0bc1f
GN		 7232 set_bit(vcpu->arch.interrupt.nr,
7233 (unsigned long *)sregs->interrupt_bitmap);
16d7a191 7234
b6c7a5dc
HB		 7235 return 0;
7236}
7237
62d9f0db
MT		 7238int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7239 struct kvm_mp_state *mp_state)
7240{
66450a21 7241 kvm_apic_accept_events(vcpu);
6aef266c
SV		 7242 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7243 vcpu->arch.pv.pv_unhalted)
7244 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7245 else
7246 mp_state->mp_state = vcpu->arch.mp_state;
7247
62d9f0db
MT		 7248 return 0;
7249}
7250
7251int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7252 struct kvm_mp_state *mp_state)
7253{
bce87cce 7254 if (!lapic_in_kernel(vcpu) &&
66450a21
JK		 7255 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7256 return -EINVAL;
7257
767cefdd
DH		 7258 /* INITs are latched while in SMM */
7259 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
7260 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
7261 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
7262 return -EINVAL;
7263
66450a21
JK		 7264 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7265 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7266 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7267 } else
7268 vcpu->arch.mp_state = mp_state->mp_state;
3842d135 7269 kvm_make_request(KVM_REQ_EVENT, vcpu);
62d9f0db
MT		 7270 return 0;
7271}
7272
7f3d35fd
KW		 7273int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7274 int reason, bool has_error_code, u32 error_code)
b6c7a5dc 7275{
9d74191a 7276 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8ec4722d 7277 int ret;
e01c2426 7278
8ec4722d 7279 init_emulate_ctxt(vcpu);
c697518a 7280
7f3d35fd 7281 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9d74191a 7282 has_error_code, error_code);
c697518a 7283
c697518a 7284 if (ret)
19d04437 7285 return EMULATE_FAIL;
37817f29 7286
9d74191a
TY		 7287 kvm_rip_write(vcpu, ctxt->eip);
7288 kvm_set_rflags(vcpu, ctxt->eflags);
3842d135 7289 kvm_make_request(KVM_REQ_EVENT, vcpu);
19d04437 7290 return EMULATE_DONE;
37817f29
IE		 7291}
7292EXPORT_SYMBOL_GPL(kvm_task_switch);
7293
b6c7a5dc
HB		 7294int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7295 struct kvm_sregs *sregs)
7296{
58cb628d 7297 struct msr_data apic_base_msr;
b6c7a5dc 7298 int mmu_reset_needed = 0;
63f42e02 7299 int pending_vec, max_bits, idx;
89a27f4d 7300 struct desc_ptr dt;
b6c7a5dc 7301
6d1068b3
PM		 7302 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7303 return -EINVAL;
7304
89a27f4d
GN		 7305 dt.size = sregs->idt.limit;
7306 dt.address = sregs->idt.base;
b6c7a5dc 7307 kvm_x86_ops->set_idt(vcpu, &dt);
89a27f4d
GN		 7308 dt.size = sregs->gdt.limit;
7309 dt.address = sregs->gdt.base;
b6c7a5dc
HB		 7310 kvm_x86_ops->set_gdt(vcpu, &dt);
7311
ad312c7c 7312 vcpu->arch.cr2 = sregs->cr2;
9f8fe504 7313 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
dc7e795e 7314 vcpu->arch.cr3 = sregs->cr3;
aff48baa 7315 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
b6c7a5dc 7316
2d3ad1f4 7317 kvm_set_cr8(vcpu, sregs->cr8);
b6c7a5dc 7318
f6801dff 7319 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
b6c7a5dc 7320 kvm_x86_ops->set_efer(vcpu, sregs->efer);
58cb628d
JK		 7321 apic_base_msr.data = sregs->apic_base;
7322 apic_base_msr.host_initiated = true;
7323 kvm_set_apic_base(vcpu, &apic_base_msr);
b6c7a5dc 7324
4d4ec087 7325 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
b6c7a5dc 7326 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
d7306163 7327 vcpu->arch.cr0 = sregs->cr0;
b6c7a5dc 7328
fc78f519 7329 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
b6c7a5dc 7330 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
b9baba86 7331 if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
00b27a3e 7332 kvm_update_cpuid(vcpu);
63f42e02
XG		 7333
7334 idx = srcu_read_lock(&vcpu->kvm->srcu);
7c93be44 7335 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
9f8fe504 7336 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7c93be44
MT		 7337 mmu_reset_needed = 1;
7338 }
63f42e02 7339 srcu_read_unlock(&vcpu->kvm->srcu, idx);
b6c7a5dc
HB		 7340
7341 if (mmu_reset_needed)
7342 kvm_mmu_reset_context(vcpu);
7343
a50abc3b 7344 max_bits = KVM_NR_INTERRUPTS;
923c61bb
GN		 7345 pending_vec = find_first_bit(
7346 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7347 if (pending_vec < max_bits) {
66fd3f7f 7348 kvm_queue_interrupt(vcpu, pending_vec, false);
923c61bb 7349 pr_debug("Set back pending irq %d\n", pending_vec);
b6c7a5dc
HB		 7350 }
7351
3e6e0aab
GT		 7352 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7353 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7354 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7355 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7356 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7357 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
b6c7a5dc 7358
3e6e0aab
GT		 7359 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7360 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
b6c7a5dc 7361
5f0269f5
ME		 7362 update_cr8_intercept(vcpu);
7363
9c3e4aab 7364 /* Older userspace won't unhalt the vcpu on reset. */
c5af89b6 7365 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9c3e4aab 7366 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
3eeb3288 7367 !is_protmode(vcpu))
9c3e4aab
MT		 7368 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7369
3842d135
AK		 7370 kvm_make_request(KVM_REQ_EVENT, vcpu);
7371
b6c7a5dc
HB		 7372 return 0;
7373}
7374
d0bfb940
JK		 7375int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7376 struct kvm_guest_debug *dbg)
b6c7a5dc 7377{
355be0b9 7378 unsigned long rflags;
ae675ef0 7379 int i, r;
b6c7a5dc 7380
4f926bf2
JK		 7381 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7382 r = -EBUSY;
7383 if (vcpu->arch.exception.pending)
2122ff5e 7384 goto out;
4f926bf2
JK		 7385 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7386 kvm_queue_exception(vcpu, DB_VECTOR);
7387 else
7388 kvm_queue_exception(vcpu, BP_VECTOR);
7389 }
7390
91586a3b
JK		 7391 /*
7392 * Read rflags as long as potentially injected trace flags are still
7393 * filtered out.
7394 */
7395 rflags = kvm_get_rflags(vcpu);
355be0b9
JK		 7396
7397 vcpu->guest_debug = dbg->control;
7398 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7399 vcpu->guest_debug = 0;
7400
7401 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
ae675ef0
JK		 7402 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7403 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
c8639010 7404 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
ae675ef0
JK		 7405 } else {
7406 for (i = 0; i < KVM_NR_DB_REGS; i++)
7407 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
ae675ef0 7408 }
c8639010 7409 kvm_update_dr7(vcpu);
ae675ef0 7410
f92653ee
JK		 7411 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7412 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7413 get_segment_base(vcpu, VCPU_SREG_CS);
94fe45da 7414
91586a3b
JK		 7415 /*
7416 * Trigger an rflags update that will inject or remove the trace
7417 * flags.
7418 */
7419 kvm_set_rflags(vcpu, rflags);
b6c7a5dc 7420
a96036b8 7421 kvm_x86_ops->update_bp_intercept(vcpu);
b6c7a5dc 7422
4f926bf2 7423 r = 0;
d0bfb940 7424
2122ff5e 7425out:
b6c7a5dc
HB		 7426
7427 return r;
7428}
7429
8b006791
ZX		 7430/*
7431 * Translate a guest virtual address to a guest physical address.
7432 */
7433int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7434 struct kvm_translation *tr)
7435{
7436 unsigned long vaddr = tr->linear_address;
7437 gpa_t gpa;
f656ce01 7438 int idx;
8b006791 7439
f656ce01 7440 idx = srcu_read_lock(&vcpu->kvm->srcu);
1871c602 7441 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
f656ce01 7442 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8b006791
ZX		 7443 tr->physical_address = gpa;
7444 tr->valid = gpa != UNMAPPED_GVA;
7445 tr->writeable = 1;
7446 tr->usermode = 0;
8b006791
ZX		 7447
7448 return 0;
7449}
7450
d0752060
HB		 7451int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7452{
c47ada30 7453 struct fxregs_state *fxsave =
7366ed77 7454 &vcpu->arch.guest_fpu.state.fxsave;
d0752060 7455
d0752060
HB		 7456 memcpy(fpu->fpr, fxsave->st_space, 128);
7457 fpu->fcw = fxsave->cwd;
7458 fpu->fsw = fxsave->swd;
7459 fpu->ftwx = fxsave->twd;
7460 fpu->last_opcode = fxsave->fop;
7461 fpu->last_ip = fxsave->rip;
7462 fpu->last_dp = fxsave->rdp;
7463 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7464
d0752060
HB		 7465 return 0;
7466}
7467
7468int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7469{
c47ada30 7470 struct fxregs_state *fxsave =
7366ed77 7471 &vcpu->arch.guest_fpu.state.fxsave;
d0752060 7472
d0752060
HB		 7473 memcpy(fxsave->st_space, fpu->fpr, 128);
7474 fxsave->cwd = fpu->fcw;
7475 fxsave->swd = fpu->fsw;
7476 fxsave->twd = fpu->ftwx;
7477 fxsave->fop = fpu->last_opcode;
7478 fxsave->rip = fpu->last_ip;
7479 fxsave->rdp = fpu->last_dp;
7480 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7481
d0752060
HB		 7482 return 0;
7483}
7484
0ee6a517 7485static void fx_init(struct kvm_vcpu *vcpu)
d0752060 7486{
bf935b0b 7487 fpstate_init(&vcpu->arch.guest_fpu.state);
782511b0 7488 if (boot_cpu_has(X86_FEATURE_XSAVES))
7366ed77 7489 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
df1daba7 7490 host_xcr0 | XSTATE_COMPACTION_ENABLED;
d0752060 7491
2acf923e
DC		 7492 /*
7493 * Ensure guest xcr0 is valid for loading
7494 */
d91cab78 7495 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
2acf923e 7496
ad312c7c 7497 vcpu->arch.cr0 |= X86_CR0_ET;
d0752060 7498}
d0752060
HB		 7499
7500void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7501{
2608d7a1 7502 if (vcpu->guest_fpu_loaded)
d0752060
HB		 7503 return;
7504
2acf923e
DC		 7505 /*
7506 * Restore all possible states in the guest,
7507 * and assume host would use all available bits.
7508 * Guest xcr0 would be loaded later.
7509 */
d0752060 7510 vcpu->guest_fpu_loaded = 1;
b1a74bf8 7511 __kernel_fpu_begin();
003e2e8b 7512 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
0c04851c 7513 trace_kvm_fpu(1);
d0752060 7514}
d0752060
HB		 7515
7516void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7517{
3d42de25 7518 if (!vcpu->guest_fpu_loaded)
d0752060
HB		 7519 return;
7520
7521 vcpu->guest_fpu_loaded = 0;
4f836347 7522 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
b1a74bf8 7523 __kernel_fpu_end();
f096ed85 7524 ++vcpu->stat.fpu_reload;
0c04851c 7525 trace_kvm_fpu(0);
d0752060 7526}
e9b11c17
ZX		 7527
7528void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7529{
bd768e14
IY		 7530 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
7531
12f9a48f 7532 kvmclock_reset(vcpu);
7f1ea208 7533
e9b11c17 7534 kvm_x86_ops->vcpu_free(vcpu);
bd768e14 7535 free_cpumask_var(wbinvd_dirty_mask);
e9b11c17
ZX		 7536}
7537
7538struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7539 unsigned int id)
7540{
c447e76b
LL		 7541 struct kvm_vcpu *vcpu;
7542
6755bae8
ZA		 7543 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7544 printk_once(KERN_WARNING
7545 "kvm: SMP vm created on host with unstable TSC; "
7546 "guest TSC will not be reliable\n");
c447e76b
LL		 7547
7548 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7549
c447e76b 7550 return vcpu;
26e5215f 7551}
e9b11c17 7552
26e5215f
AK		 7553int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7554{
7555 int r;
e9b11c17 7556
19efffa2 7557 kvm_vcpu_mtrr_init(vcpu);
9fc77441
MT		 7558 r = vcpu_load(vcpu);
7559 if (r)
7560 return r;
d28bc9dd 7561 kvm_vcpu_reset(vcpu, false);
8a3c1a33 7562 kvm_mmu_setup(vcpu);
e9b11c17 7563 vcpu_put(vcpu);
26e5215f 7564 return r;
e9b11c17
ZX		 7565}
7566
31928aa5 7567void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
42897d86 7568{
8fe8ab46 7569 struct msr_data msr;
332967a3 7570 struct kvm *kvm = vcpu->kvm;
42897d86 7571
31928aa5
DD		 7572 if (vcpu_load(vcpu))
7573 return;
8fe8ab46
WA		 7574 msr.data = 0x0;
7575 msr.index = MSR_IA32_TSC;
7576 msr.host_initiated = true;
7577 kvm_write_tsc(vcpu, &msr);
42897d86
MT		 7578 vcpu_put(vcpu);
7579
630994b3
MT		 7580 if (!kvmclock_periodic_sync)
7581 return;
7582
332967a3
AJ		 7583 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7584 KVMCLOCK_SYNC_PERIOD);
42897d86
MT		 7585}
7586
d40ccc62 7587void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
e9b11c17 7588{
9fc77441 7589 int r;
344d9588
GN		 7590 vcpu->arch.apf.msr_val = 0;
7591
9fc77441
MT		 7592 r = vcpu_load(vcpu);
7593 BUG_ON(r);
e9b11c17
ZX		 7594 kvm_mmu_unload(vcpu);
7595 vcpu_put(vcpu);
7596
7597 kvm_x86_ops->vcpu_free(vcpu);
7598}
7599
d28bc9dd 7600void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
e9b11c17 7601{
e69fab5d
PB		 7602 vcpu->arch.hflags = 0;
7603
c43203ca 7604 vcpu->arch.smi_pending = 0;
7460fb4a
AK		 7605 atomic_set(&vcpu->arch.nmi_queued, 0);
7606 vcpu->arch.nmi_pending = 0;
448fa4a9 7607 vcpu->arch.nmi_injected = false;
5f7552d4
NA		 7608 kvm_clear_interrupt_queue(vcpu);
7609 kvm_clear_exception_queue(vcpu);
448fa4a9 7610
42dbaa5a 7611 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
ae561ede 7612 kvm_update_dr0123(vcpu);
6f43ed01 7613 vcpu->arch.dr6 = DR6_INIT;
73aaf249 7614 kvm_update_dr6(vcpu);
42dbaa5a 7615 vcpu->arch.dr7 = DR7_FIXED_1;
c8639010 7616 kvm_update_dr7(vcpu);
42dbaa5a 7617
1119022c
NA		 7618 vcpu->arch.cr2 = 0;
7619
3842d135 7620 kvm_make_request(KVM_REQ_EVENT, vcpu);
344d9588 7621 vcpu->arch.apf.msr_val = 0;
c9aaa895 7622 vcpu->arch.st.msr_val = 0;
3842d135 7623
12f9a48f
GC		 7624 kvmclock_reset(vcpu);
7625
af585b92
GN		 7626 kvm_clear_async_pf_completion_queue(vcpu);
7627 kvm_async_pf_hash_reset(vcpu);
7628 vcpu->arch.apf.halted = false;
3842d135 7629
64d60670 7630 if (!init_event) {
d28bc9dd 7631 kvm_pmu_reset(vcpu);
64d60670
PB		 7632 vcpu->arch.smbase = 0x30000;
7633 }
f5132b01 7634
66f7b72e
JS		 7635 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7636 vcpu->arch.regs_avail = ~0;
7637 vcpu->arch.regs_dirty = ~0;
7638
d28bc9dd 7639 kvm_x86_ops->vcpu_reset(vcpu, init_event);
e9b11c17
ZX		 7640}
7641
2b4a273b 7642void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
66450a21
JK		 7643{
7644 struct kvm_segment cs;
7645
7646 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7647 cs.selector = vector << 8;
7648 cs.base = vector << 12;
7649 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7650 kvm_rip_write(vcpu, 0);
e9b11c17
ZX		 7651}
7652
13a34e06 7653int kvm_arch_hardware_enable(void)
e9b11c17 7654{
ca84d1a2
ZA		 7655 struct kvm *kvm;
7656 struct kvm_vcpu *vcpu;
7657 int i;
0dd6a6ed
ZA		 7658 int ret;
7659 u64 local_tsc;
7660 u64 max_tsc = 0;
7661 bool stable, backwards_tsc = false;
18863bdd
AK		 7662
7663 kvm_shared_msr_cpu_online();
13a34e06 7664 ret = kvm_x86_ops->hardware_enable();
0dd6a6ed
ZA		 7665 if (ret != 0)
7666 return ret;
7667
4ea1636b 7668 local_tsc = rdtsc();
0dd6a6ed
ZA		 7669 stable = !check_tsc_unstable();
7670 list_for_each_entry(kvm, &vm_list, vm_list) {
7671 kvm_for_each_vcpu(i, vcpu, kvm) {
7672 if (!stable && vcpu->cpu == smp_processor_id())
105b21bb 7673 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
0dd6a6ed
ZA		 7674 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7675 backwards_tsc = true;
7676 if (vcpu->arch.last_host_tsc > max_tsc)
7677 max_tsc = vcpu->arch.last_host_tsc;
7678 }
7679 }
7680 }
7681
7682 /*
7683 * Sometimes, even reliable TSCs go backwards. This happens on
7684 * platforms that reset TSC during suspend or hibernate actions, but
7685 * maintain synchronization. We must compensate. Fortunately, we can
7686 * detect that condition here, which happens early in CPU bringup,
7687 * before any KVM threads can be running. Unfortunately, we can't
7688 * bring the TSCs fully up to date with real time, as we aren't yet far
7689 * enough into CPU bringup that we know how much real time has actually
108b249c 7690 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
0dd6a6ed
ZA		 7691 * variables that haven't been updated yet.
7692 *
7693 * So we simply find the maximum observed TSC above, then record the
7694 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7695 * the adjustment will be applied. Note that we accumulate
7696 * adjustments, in case multiple suspend cycles happen before some VCPU
7697 * gets a chance to run again. In the event that no KVM threads get a
7698 * chance to run, we will miss the entire elapsed period, as we'll have
7699 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7700 * loose cycle time. This isn't too big a deal, since the loss will be
7701 * uniform across all VCPUs (not to mention the scenario is extremely
7702 * unlikely). It is possible that a second hibernate recovery happens
7703 * much faster than a first, causing the observed TSC here to be
7704 * smaller; this would require additional padding adjustment, which is
7705 * why we set last_host_tsc to the local tsc observed here.
7706 *
7707 * N.B. - this code below runs only on platforms with reliable TSC,
7708 * as that is the only way backwards_tsc is set above. Also note
7709 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7710 * have the same delta_cyc adjustment applied if backwards_tsc
7711 * is detected. Note further, this adjustment is only done once,
7712 * as we reset last_host_tsc on all VCPUs to stop this from being
7713 * called multiple times (one for each physical CPU bringup).
7714 *
4a969980 7715 * Platforms with unreliable TSCs don't have to deal with this, they
0dd6a6ed
ZA		 7716 * will be compensated by the logic in vcpu_load, which sets the TSC to
7717 * catchup mode. This will catchup all VCPUs to real time, but cannot
7718 * guarantee that they stay in perfect synchronization.
7719 */
7720 if (backwards_tsc) {
7721 u64 delta_cyc = max_tsc - local_tsc;
16a96021 7722 backwards_tsc_observed = true;
0dd6a6ed
ZA		 7723 list_for_each_entry(kvm, &vm_list, vm_list) {
7724 kvm_for_each_vcpu(i, vcpu, kvm) {
7725 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7726 vcpu->arch.last_host_tsc = local_tsc;
105b21bb 7727 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
0dd6a6ed
ZA		 7728 }
7729
7730 /*
7731 * We have to disable TSC offset matching.. if you were
7732 * booting a VM while issuing an S4 host suspend....
7733 * you may have some problem. Solving this issue is
7734 * left as an exercise to the reader.
7735 */
7736 kvm->arch.last_tsc_nsec = 0;
7737 kvm->arch.last_tsc_write = 0;
7738 }
7739
7740 }
7741 return 0;
e9b11c17
ZX		 7742}
7743
13a34e06 7744void kvm_arch_hardware_disable(void)
e9b11c17 7745{
13a34e06
RK		 7746 kvm_x86_ops->hardware_disable();
7747 drop_user_return_notifiers();
e9b11c17
ZX		 7748}
7749
7750int kvm_arch_hardware_setup(void)
7751{
9e9c3fe4
NA		 7752 int r;
7753
7754 r = kvm_x86_ops->hardware_setup();
7755 if (r != 0)
7756 return r;
7757
35181e86
HZ		 7758 if (kvm_has_tsc_control) {
7759 /*
7760 * Make sure the user can only configure tsc_khz values that
7761 * fit into a signed integer.
7762 * A min value is not calculated needed because it will always
7763 * be 1 on all machines.
7764 */
7765 u64 max = min(0x7fffffffULL,
7766 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
7767 kvm_max_guest_tsc_khz = max;
7768
ad721883 7769 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
35181e86 7770 }
ad721883 7771
9e9c3fe4
NA		 7772 kvm_init_msr_list();
7773 return 0;
e9b11c17
ZX		 7774}
7775
7776void kvm_arch_hardware_unsetup(void)
7777{
7778 kvm_x86_ops->hardware_unsetup();
7779}
7780
7781void kvm_arch_check_processor_compat(void *rtn)
7782{
7783 kvm_x86_ops->check_processor_compatibility(rtn);
d71ba788
PB		 7784}
7785
7786bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7787{
7788 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7789}
7790EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7791
7792bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7793{
7794 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
e9b11c17
ZX		 7795}
7796
54e9818f 7797struct static_key kvm_no_apic_vcpu __read_mostly;
bce87cce 7798EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
54e9818f 7799
e9b11c17
ZX		 7800int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7801{
7802 struct page *page;
7803 struct kvm *kvm;
7804 int r;
7805
7806 BUG_ON(vcpu->kvm == NULL);
7807 kvm = vcpu->kvm;
7808
d62caabb 7809 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv();
6aef266c 7810 vcpu->arch.pv.pv_unhalted = false;
9aabc88f 7811 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
58d269d8 7812 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
a4535290 7813 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
e9b11c17 7814 else
a4535290 7815 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
e9b11c17
ZX		 7816
7817 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7818 if (!page) {
7819 r = -ENOMEM;
7820 goto fail;
7821 }
ad312c7c 7822 vcpu->arch.pio_data = page_address(page);
e9b11c17 7823
cc578287 7824 kvm_set_tsc_khz(vcpu, max_tsc_khz);
c285545f 7825
e9b11c17
ZX		 7826 r = kvm_mmu_create(vcpu);
7827 if (r < 0)
7828 goto fail_free_pio_data;
7829
7830 if (irqchip_in_kernel(kvm)) {
7831 r = kvm_create_lapic(vcpu);
7832 if (r < 0)
7833 goto fail_mmu_destroy;
54e9818f
GN		 7834 } else
7835 static_key_slow_inc(&kvm_no_apic_vcpu);
e9b11c17 7836
890ca9ae
HY		 7837 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7838 GFP_KERNEL);
7839 if (!vcpu->arch.mce_banks) {
7840 r = -ENOMEM;
443c39bc 7841 goto fail_free_lapic;
890ca9ae
HY		 7842 }
7843 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7844
f1797359
WY		 7845 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7846 r = -ENOMEM;
f5f48ee1 7847 goto fail_free_mce_banks;
f1797359 7848 }
f5f48ee1 7849
0ee6a517 7850 fx_init(vcpu);
66f7b72e 7851
ba904635 7852 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
0b79459b 7853 vcpu->arch.pv_time_enabled = false;
d7876f1b
PB		 7854
7855 vcpu->arch.guest_supported_xcr0 = 0;
4344ee98 7856 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
d7876f1b 7857
5a4f55cd
EK		 7858 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7859
74545705
RK		 7860 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7861
af585b92 7862 kvm_async_pf_hash_reset(vcpu);
f5132b01 7863 kvm_pmu_init(vcpu);
af585b92 7864
1c1a9ce9
SR		 7865 vcpu->arch.pending_external_vector = -1;
7866
5c919412
AS		 7867 kvm_hv_vcpu_init(vcpu);
7868
e9b11c17 7869 return 0;
0ee6a517 7870
f5f48ee1
SY		 7871fail_free_mce_banks:
7872 kfree(vcpu->arch.mce_banks);
443c39bc
WY		 7873fail_free_lapic:
7874 kvm_free_lapic(vcpu);
e9b11c17
ZX		 7875fail_mmu_destroy:
7876 kvm_mmu_destroy(vcpu);
7877fail_free_pio_data:
ad312c7c 7878 free_page((unsigned long)vcpu->arch.pio_data);
e9b11c17
ZX		 7879fail:
7880 return r;
7881}
7882
7883void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7884{
f656ce01
MT		 7885 int idx;
7886
1f4b34f8 7887 kvm_hv_vcpu_uninit(vcpu);
f5132b01 7888 kvm_pmu_destroy(vcpu);
36cb93fd 7889 kfree(vcpu->arch.mce_banks);
e9b11c17 7890 kvm_free_lapic(vcpu);
f656ce01 7891 idx = srcu_read_lock(&vcpu->kvm->srcu);
e9b11c17 7892 kvm_mmu_destroy(vcpu);
f656ce01 7893 srcu_read_unlock(&vcpu->kvm->srcu, idx);
ad312c7c 7894 free_page((unsigned long)vcpu->arch.pio_data);
35754c98 7895 if (!lapic_in_kernel(vcpu))
54e9818f 7896 static_key_slow_dec(&kvm_no_apic_vcpu);
e9b11c17 7897}
d19a9cd2 7898
e790d9ef
RK		 7899void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7900{
ae97a3b8 7901 kvm_x86_ops->sched_in(vcpu, cpu);
e790d9ef
RK		 7902}
7903
e08b9637 7904int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
d19a9cd2 7905{
e08b9637
CO		 7906 if (type)
7907 return -EINVAL;
7908
6ef768fa 7909 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
f05e70ac 7910 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
365c8868 7911 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
4d5c5d0f 7912 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
e0f0bbc5 7913 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
d19a9cd2 7914
5550af4d
SY		 7915 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7916 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7a84428a
AW		 7917 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7918 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7919 &kvm->arch.irq_sources_bitmap);
5550af4d 7920
038f8c11 7921 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
1e08ec4a 7922 mutex_init(&kvm->arch.apic_map_lock);
3f5ad8be 7923 mutex_init(&kvm->arch.hyperv.hv_lock);
d828199e
MT		 7924 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7925
108b249c 7926 kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
d828199e 7927 pvclock_update_vm_gtod_copy(kvm);
53f658b3 7928
7e44e449 7929 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
332967a3 7930 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7e44e449 7931
0eb05bf2 7932 kvm_page_track_init(kvm);
13d268ca 7933 kvm_mmu_init_vm(kvm);
0eb05bf2 7934
03543133
SS		 7935 if (kvm_x86_ops->vm_init)
7936 return kvm_x86_ops->vm_init(kvm);
7937
d89f5eff 7938 return 0;
d19a9cd2
ZX		 7939}
7940
7941static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7942{
9fc77441
MT		 7943 int r;
7944 r = vcpu_load(vcpu);
7945 BUG_ON(r);
d19a9cd2
ZX		 7946 kvm_mmu_unload(vcpu);
7947 vcpu_put(vcpu);
7948}
7949
7950static void kvm_free_vcpus(struct kvm *kvm)
7951{
7952 unsigned int i;
988a2cae 7953 struct kvm_vcpu *vcpu;
d19a9cd2
ZX		 7954
7955 /*
7956 * Unpin any mmu pages first.
7957 */
af585b92
GN		 7958 kvm_for_each_vcpu(i, vcpu, kvm) {
7959 kvm_clear_async_pf_completion_queue(vcpu);
988a2cae 7960 kvm_unload_vcpu_mmu(vcpu);
af585b92 7961 }
988a2cae
GN		 7962 kvm_for_each_vcpu(i, vcpu, kvm)
7963 kvm_arch_vcpu_free(vcpu);
7964
7965 mutex_lock(&kvm->lock);
7966 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7967 kvm->vcpus[i] = NULL;
d19a9cd2 7968
988a2cae
GN		 7969 atomic_set(&kvm->online_vcpus, 0);
7970 mutex_unlock(&kvm->lock);
d19a9cd2
ZX		 7971}
7972
ad8ba2cd
SY		 7973void kvm_arch_sync_events(struct kvm *kvm)
7974{
332967a3 7975 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7e44e449 7976 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
ba4cef31 7977 kvm_free_all_assigned_devices(kvm);
aea924f6 7978 kvm_free_pit(kvm);
ad8ba2cd
SY		 7979}
7980
1d8007bd 7981int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9da0e4d5
PB		 7982{
7983 int i, r;
25188b99 7984 unsigned long hva;
f0d648bd
PB		 7985 struct kvm_memslots *slots = kvm_memslots(kvm);
7986 struct kvm_memory_slot *slot, old;
9da0e4d5
PB		 7987
7988 /* Called with kvm->slots_lock held. */
1d8007bd
PB		 7989 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
7990 return -EINVAL;
9da0e4d5 7991
f0d648bd
PB		 7992 slot = id_to_memslot(slots, id);
7993 if (size) {
b21629da 7994 if (slot->npages)
f0d648bd
PB		 7995 return -EEXIST;
7996
7997 /*
7998 * MAP_SHARED to prevent internal slot pages from being moved
7999 * by fork()/COW.
8000 */
8001 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
8002 MAP_SHARED | MAP_ANONYMOUS, 0);
8003 if (IS_ERR((void *)hva))
8004 return PTR_ERR((void *)hva);
8005 } else {
8006 if (!slot->npages)
8007 return 0;
8008
8009 hva = 0;
8010 }
8011
8012 old = *slot;
9da0e4d5 8013 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1d8007bd 8014 struct kvm_userspace_memory_region m;
9da0e4d5 8015
1d8007bd
PB		 8016 m.slot = id | (i << 16);
8017 m.flags = 0;
8018 m.guest_phys_addr = gpa;
f0d648bd 8019 m.userspace_addr = hva;
1d8007bd 8020 m.memory_size = size;
9da0e4d5
PB		 8021 r = __kvm_set_memory_region(kvm, &m);
8022 if (r < 0)
8023 return r;
8024 }
8025
f0d648bd
PB		 8026 if (!size) {
8027 r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
8028 WARN_ON(r < 0);
8029 }
8030
9da0e4d5
PB		 8031 return 0;
8032}
8033EXPORT_SYMBOL_GPL(__x86_set_memory_region);
8034
1d8007bd 8035int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9da0e4d5
PB		 8036{
8037 int r;
8038
8039 mutex_lock(&kvm->slots_lock);
1d8007bd 8040 r = __x86_set_memory_region(kvm, id, gpa, size);
9da0e4d5
PB		 8041 mutex_unlock(&kvm->slots_lock);
8042
8043 return r;
8044}
8045EXPORT_SYMBOL_GPL(x86_set_memory_region);
8046
d19a9cd2
ZX		 8047void kvm_arch_destroy_vm(struct kvm *kvm)
8048{
27469d29
AH		 8049 if (current->mm == kvm->mm) {
8050 /*
8051 * Free memory regions allocated on behalf of userspace,
8052 * unless the the memory map has changed due to process exit
8053 * or fd copying.
8054 */
1d8007bd
PB		 8055 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
8056 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
8057 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
27469d29 8058 }
03543133
SS		 8059 if (kvm_x86_ops->vm_destroy)
8060 kvm_x86_ops->vm_destroy(kvm);
6eb55818 8061 kvm_iommu_unmap_guest(kvm);
d7deeeb0
ZX		 8062 kfree(kvm->arch.vpic);
8063 kfree(kvm->arch.vioapic);
d19a9cd2 8064 kvm_free_vcpus(kvm);
af1bae54 8065 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
13d268ca 8066 kvm_mmu_uninit_vm(kvm);
10fa66bd 8067 kvm_page_track_cleanup(kvm);
d19a9cd2 8068}
0de10343 8069
5587027c 8070void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
db3fe4eb
TY		 8071 struct kvm_memory_slot *dont)
8072{
8073 int i;
8074
d89cc617
TY		 8075 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8076 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
548ef284 8077 kvfree(free->arch.rmap[i]);
d89cc617 8078 free->arch.rmap[i] = NULL;
77d11309 8079 }
d89cc617
TY		 8080 if (i == 0)
8081 continue;
8082
8083 if (!dont || free->arch.lpage_info[i - 1] !=
8084 dont->arch.lpage_info[i - 1]) {
548ef284 8085 kvfree(free->arch.lpage_info[i - 1]);
d89cc617 8086 free->arch.lpage_info[i - 1] = NULL;
db3fe4eb
TY		 8087 }
8088 }
21ebbeda
XG		 8089
8090 kvm_page_track_free_memslot(free, dont);
db3fe4eb
TY		 8091}
8092
5587027c
AK		 8093int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
8094 unsigned long npages)
db3fe4eb
TY		 8095{
8096 int i;
8097
d89cc617 8098 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
92f94f1e 8099 struct kvm_lpage_info *linfo;
db3fe4eb
TY		 8100 unsigned long ugfn;
8101 int lpages;
d89cc617 8102 int level = i + 1;
db3fe4eb
TY		 8103
8104 lpages = gfn_to_index(slot->base_gfn + npages - 1,
8105 slot->base_gfn, level) + 1;
8106
d89cc617
TY		 8107 slot->arch.rmap[i] =
8108 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
8109 if (!slot->arch.rmap[i])
77d11309 8110 goto out_free;
d89cc617
TY		 8111 if (i == 0)
8112 continue;
77d11309 8113
92f94f1e
XG		 8114 linfo = kvm_kvzalloc(lpages * sizeof(*linfo));
8115 if (!linfo)
db3fe4eb
TY		 8116 goto out_free;
8117
92f94f1e
XG		 8118 slot->arch.lpage_info[i - 1] = linfo;
8119
db3fe4eb 8120 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
92f94f1e 8121 linfo[0].disallow_lpage = 1;
db3fe4eb 8122 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
92f94f1e 8123 linfo[lpages - 1].disallow_lpage = 1;
db3fe4eb
TY		 8124 ugfn = slot->userspace_addr >> PAGE_SHIFT;
8125 /*
8126 * If the gfn and userspace address are not aligned wrt each
8127 * other, or if explicitly asked to, disable large page
8128 * support for this slot
8129 */
8130 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
8131 !kvm_largepages_enabled()) {
8132 unsigned long j;
8133
8134 for (j = 0; j < lpages; ++j)
92f94f1e 8135 linfo[j].disallow_lpage = 1;
db3fe4eb
TY		 8136 }
8137 }
8138
21ebbeda
XG		 8139 if (kvm_page_track_create_memslot(slot, npages))
8140 goto out_free;
8141
db3fe4eb
TY		 8142 return 0;
8143
8144out_free:
d89cc617 8145 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
548ef284 8146 kvfree(slot->arch.rmap[i]);
d89cc617
TY		 8147 slot->arch.rmap[i] = NULL;
8148 if (i == 0)
8149 continue;
8150
548ef284 8151 kvfree(slot->arch.lpage_info[i - 1]);
d89cc617 8152 slot->arch.lpage_info[i - 1] = NULL;
db3fe4eb
TY		 8153 }
8154 return -ENOMEM;
8155}
8156
15f46015 8157void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
e59dbe09 8158{
e6dff7d1
TY		 8159 /*
8160 * memslots->generation has been incremented.
8161 * mmio generation may have reached its maximum value.
8162 */
54bf36aa 8163 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
e59dbe09
TY		 8164}
8165
f7784b8e
MT		 8166int kvm_arch_prepare_memory_region(struct kvm *kvm,
8167 struct kvm_memory_slot *memslot,
09170a49 8168 const struct kvm_userspace_memory_region *mem,
7b6195a9 8169 enum kvm_mr_change change)
0de10343 8170{
f7784b8e
MT		 8171 return 0;
8172}
8173
88178fd4
KH		 8174static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8175 struct kvm_memory_slot *new)
8176{
8177 /* Still write protect RO slot */
8178 if (new->flags & KVM_MEM_READONLY) {
8179 kvm_mmu_slot_remove_write_access(kvm, new);
8180 return;
8181 }
8182
8183 /*
8184 * Call kvm_x86_ops dirty logging hooks when they are valid.
8185 *
8186 * kvm_x86_ops->slot_disable_log_dirty is called when:
8187 *
8188 * - KVM_MR_CREATE with dirty logging is disabled
8189 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8190 *
8191 * The reason is, in case of PML, we need to set D-bit for any slots
8192 * with dirty logging disabled in order to eliminate unnecessary GPA
8193 * logging in PML buffer (and potential PML buffer full VMEXT). This
8194 * guarantees leaving PML enabled during guest's lifetime won't have
8195 * any additonal overhead from PML when guest is running with dirty
8196 * logging disabled for memory slots.
8197 *
8198 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8199 * to dirty logging mode.
8200 *
8201 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8202 *
8203 * In case of write protect:
8204 *
8205 * Write protect all pages for dirty logging.
8206 *
8207 * All the sptes including the large sptes which point to this
8208 * slot are set to readonly. We can not create any new large
8209 * spte on this slot until the end of the logging.
8210 *
8211 * See the comments in fast_page_fault().
8212 */
8213 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8214 if (kvm_x86_ops->slot_enable_log_dirty)
8215 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8216 else
8217 kvm_mmu_slot_remove_write_access(kvm, new);
8218 } else {
8219 if (kvm_x86_ops->slot_disable_log_dirty)
8220 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8221 }
8222}
8223
f7784b8e 8224void kvm_arch_commit_memory_region(struct kvm *kvm,
09170a49 8225 const struct kvm_userspace_memory_region *mem,
8482644a 8226 const struct kvm_memory_slot *old,
f36f3f28 8227 const struct kvm_memory_slot *new,
8482644a 8228 enum kvm_mr_change change)
f7784b8e 8229{
8482644a 8230 int nr_mmu_pages = 0;
f7784b8e 8231
48c0e4e9
XG		 8232 if (!kvm->arch.n_requested_mmu_pages)
8233 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8234
48c0e4e9 8235 if (nr_mmu_pages)
0de10343 8236 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
1c91cad4 8237
3ea3b7fa
WL		 8238 /*
8239 * Dirty logging tracks sptes in 4k granularity, meaning that large
8240 * sptes have to be split. If live migration is successful, the guest
8241 * in the source machine will be destroyed and large sptes will be
8242 * created in the destination. However, if the guest continues to run
8243 * in the source machine (for example if live migration fails), small
8244 * sptes will remain around and cause bad performance.
8245 *
8246 * Scan sptes if dirty logging has been stopped, dropping those
8247 * which can be collapsed into a single large-page spte. Later
8248 * page faults will create the large-page sptes.
8249 */
8250 if ((change != KVM_MR_DELETE) &&
8251 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8252 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8253 kvm_mmu_zap_collapsible_sptes(kvm, new);
8254
c972f3b1 8255 /*
88178fd4 8256 * Set up write protection and/or dirty logging for the new slot.
c126d94f 8257 *
88178fd4
KH		 8258 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8259 * been zapped so no dirty logging staff is needed for old slot. For
8260 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8261 * new and it's also covered when dealing with the new slot.
f36f3f28
PB		 8262 *
8263 * FIXME: const-ify all uses of struct kvm_memory_slot.
c972f3b1 8264 */
88178fd4 8265 if (change != KVM_MR_DELETE)
f36f3f28 8266 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
0de10343 8267}
1d737c8a 8268
2df72e9b 8269void kvm_arch_flush_shadow_all(struct kvm *kvm)
34d4cb8f 8270{
6ca18b69 8271 kvm_mmu_invalidate_zap_all_pages(kvm);
34d4cb8f
MT		 8272}
8273
2df72e9b
MT		 8274void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8275 struct kvm_memory_slot *slot)
8276{
ae7cd873 8277 kvm_page_track_flush_slot(kvm, slot);
2df72e9b
MT		 8278}
8279
5d9bc648
PB		 8280static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8281{
8282 if (!list_empty_careful(&vcpu->async_pf.done))
8283 return true;
8284
8285 if (kvm_apic_has_events(vcpu))
8286 return true;
8287
8288 if (vcpu->arch.pv.pv_unhalted)
8289 return true;
8290
8291 if (atomic_read(&vcpu->arch.nmi_queued))
8292 return true;
8293
73917739
PB		 8294 if (test_bit(KVM_REQ_SMI, &vcpu->requests))
8295 return true;
8296
5d9bc648
PB		 8297 if (kvm_arch_interrupt_allowed(vcpu) &&
8298 kvm_cpu_has_interrupt(vcpu))
8299 return true;
8300
1f4b34f8
AS		 8301 if (kvm_hv_has_stimer_pending(vcpu))
8302 return true;
8303
5d9bc648
PB		 8304 return false;
8305}
8306
1d737c8a
ZX		 8307int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8308{
b6b8a145
JK		 8309 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8310 kvm_x86_ops->check_nested_events(vcpu, false);
8311
5d9bc648 8312 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
1d737c8a 8313}
5736199a 8314
b6d33834 8315int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
5736199a 8316{
b6d33834 8317 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
5736199a 8318}
78646121
GN		 8319
8320int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8321{
8322 return kvm_x86_ops->interrupt_allowed(vcpu);
8323}
229456fc 8324
82b32774 8325unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
f92653ee 8326{
82b32774
NA		 8327 if (is_64_bit_mode(vcpu))
8328 return kvm_rip_read(vcpu);
8329 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8330 kvm_rip_read(vcpu));
8331}
8332EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
f92653ee 8333
82b32774
NA		 8334bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8335{
8336 return kvm_get_linear_rip(vcpu) == linear_rip;
f92653ee
JK		 8337}
8338EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8339
94fe45da
JK		 8340unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8341{
8342 unsigned long rflags;
8343
8344 rflags = kvm_x86_ops->get_rflags(vcpu);
8345 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
c310bac5 8346 rflags &= ~X86_EFLAGS_TF;
94fe45da
JK		 8347 return rflags;
8348}
8349EXPORT_SYMBOL_GPL(kvm_get_rflags);
8350
6addfc42 8351static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
94fe45da
JK		 8352{
8353 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
f92653ee 8354 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
c310bac5 8355 rflags |= X86_EFLAGS_TF;
94fe45da 8356 kvm_x86_ops->set_rflags(vcpu, rflags);
6addfc42
PB		 8357}
8358
8359void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8360{
8361 __kvm_set_rflags(vcpu, rflags);
3842d135 8362 kvm_make_request(KVM_REQ_EVENT, vcpu);
94fe45da
JK		 8363}
8364EXPORT_SYMBOL_GPL(kvm_set_rflags);
8365
56028d08
GN		 8366void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8367{
8368 int r;
8369
fb67e14f 8370 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
f2e10669 8371 work->wakeup_all)
56028d08
GN		 8372 return;
8373
8374 r = kvm_mmu_reload(vcpu);
8375 if (unlikely(r))
8376 return;
8377
fb67e14f
XG		 8378 if (!vcpu->arch.mmu.direct_map &&
8379 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8380 return;
8381
56028d08
GN		 8382 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8383}
8384
af585b92
GN		 8385static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8386{
8387 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8388}
8389
8390static inline u32 kvm_async_pf_next_probe(u32 key)
8391{
8392 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8393}
8394
8395static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8396{
8397 u32 key = kvm_async_pf_hash_fn(gfn);
8398
8399 while (vcpu->arch.apf.gfns[key] != ~0)
8400 key = kvm_async_pf_next_probe(key);
8401
8402 vcpu->arch.apf.gfns[key] = gfn;
8403}
8404
8405static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8406{
8407 int i;
8408 u32 key = kvm_async_pf_hash_fn(gfn);
8409
8410 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
c7d28c24
XG		 8411 (vcpu->arch.apf.gfns[key] != gfn &&
8412 vcpu->arch.apf.gfns[key] != ~0); i++)
af585b92
GN		 8413 key = kvm_async_pf_next_probe(key);
8414
8415 return key;
8416}
8417
8418bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8419{
8420 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8421}
8422
8423static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8424{
8425 u32 i, j, k;
8426
8427 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8428 while (true) {
8429 vcpu->arch.apf.gfns[i] = ~0;
8430 do {
8431 j = kvm_async_pf_next_probe(j);
8432 if (vcpu->arch.apf.gfns[j] == ~0)
8433 return;
8434 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8435 /*
8436 * k lies cyclically in ]i,j]
8437 * | i.k.j |
8438 * |....j i.k.| or |.k..j i...|
8439 */
8440 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8441 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8442 i = j;
8443 }
8444}
8445
7c90705b
GN		 8446static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8447{
8448
8449 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8450 sizeof(val));
8451}
8452
af585b92
GN		 8453void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8454 struct kvm_async_pf *work)
8455{
6389ee94
AK		 8456 struct x86_exception fault;
8457
7c90705b 8458 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
af585b92 8459 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7c90705b
GN		 8460
8461 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
fc5f06fa
GN		 8462 (vcpu->arch.apf.send_user_only &&
8463 kvm_x86_ops->get_cpl(vcpu) == 0))
7c90705b
GN		 8464 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8465 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6389ee94
AK		 8466 fault.vector = PF_VECTOR;
8467 fault.error_code_valid = true;
8468 fault.error_code = 0;
8469 fault.nested_page_fault = false;
8470 fault.address = work->arch.token;
8471 kvm_inject_page_fault(vcpu, &fault);
7c90705b 8472 }
af585b92
GN		 8473}
8474
8475void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8476 struct kvm_async_pf *work)
8477{
6389ee94
AK		 8478 struct x86_exception fault;
8479
7c90705b 8480 trace_kvm_async_pf_ready(work->arch.token, work->gva);
f2e10669 8481 if (work->wakeup_all)
7c90705b
GN		 8482 work->arch.token = ~0; /* broadcast wakeup */
8483 else
8484 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8485
8486 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8487 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6389ee94
AK		 8488 fault.vector = PF_VECTOR;
8489 fault.error_code_valid = true;
8490 fault.error_code = 0;
8491 fault.nested_page_fault = false;
8492 fault.address = work->arch.token;
8493 kvm_inject_page_fault(vcpu, &fault);
7c90705b 8494 }
e6d53e3b 8495 vcpu->arch.apf.halted = false;
a4fa1635 8496 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7c90705b
GN		 8497}
8498
8499bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8500{
8501 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8502 return true;
8503 else
8504 return !kvm_event_needs_reinjection(vcpu) &&
8505 kvm_x86_ops->interrupt_allowed(vcpu);
af585b92
GN		 8506}
8507
5544eb9b
PB		 8508void kvm_arch_start_assignment(struct kvm *kvm)
8509{
8510 atomic_inc(&kvm->arch.assigned_device_count);
8511}
8512EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8513
8514void kvm_arch_end_assignment(struct kvm *kvm)
8515{
8516 atomic_dec(&kvm->arch.assigned_device_count);
8517}
8518EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8519
8520bool kvm_arch_has_assigned_device(struct kvm *kvm)
8521{
8522 return atomic_read(&kvm->arch.assigned_device_count);
8523}
8524EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8525
e0f0bbc5
AW		 8526void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8527{
8528 atomic_inc(&kvm->arch.noncoherent_dma_count);
8529}
8530EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8531
8532void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8533{
8534 atomic_dec(&kvm->arch.noncoherent_dma_count);
8535}
8536EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8537
8538bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8539{
8540 return atomic_read(&kvm->arch.noncoherent_dma_count);
8541}
8542EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8543
14717e20
AW		 8544bool kvm_arch_has_irq_bypass(void)
8545{
8546 return kvm_x86_ops->update_pi_irte != NULL;
8547}
8548
87276880
FW		 8549int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8550 struct irq_bypass_producer *prod)
8551{
8552 struct kvm_kernel_irqfd *irqfd =
8553 container_of(cons, struct kvm_kernel_irqfd, consumer);
8554
14717e20 8555 irqfd->producer = prod;
87276880 8556
14717e20
AW		 8557 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8558 prod->irq, irqfd->gsi, 1);
87276880
FW		 8559}
8560
8561void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8562 struct irq_bypass_producer *prod)
8563{
8564 int ret;
8565 struct kvm_kernel_irqfd *irqfd =
8566 container_of(cons, struct kvm_kernel_irqfd, consumer);
8567
87276880
FW		 8568 WARN_ON(irqfd->producer != prod);
8569 irqfd->producer = NULL;
8570
8571 /*
8572 * When producer of consumer is unregistered, we change back to
8573 * remapped mode, so we can re-use the current implementation
bb3541f1 8574 * when the irq is masked/disabled or the consumer side (KVM
87276880
FW		 8575 * int this case doesn't want to receive the interrupts.
8576 */
8577 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8578 if (ret)
8579 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8580 " fails: %d\n", irqfd->consumer.token, ret);
8581}
8582
8583int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8584 uint32_t guest_irq, bool set)
8585{
8586 if (!kvm_x86_ops->update_pi_irte)
8587 return -EINVAL;
8588
8589 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8590}
8591
52004014
FW		 8592bool kvm_vector_hashing_enabled(void)
8593{
8594 return vector_hashing;
8595}
8596EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
8597
229456fc 8598EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
931c33b1 8599EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
229456fc
MT		 8600EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8601EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8602EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8603EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
0ac406de 8604EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
d8cabddf 8605EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
17897f36 8606EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
236649de 8607EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
ec1ff790 8608EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
532a46b9 8609EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
2e554e8d 8610EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
489223ed 8611EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
7b46268d 8612EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
843e4330 8613EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
efc64404 8614EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
18f40c53
SS		 8615EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
8616EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
ubuntu-zesty-kernel mirror