projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: x86: Rename interrupt.pending to interrupt.injected
[linux-2.6-block.git] / arch / x86 / kvm / x86.c
CommitLineData
043405e1
CO		 1/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * derived from drivers/kvm/kvm_main.c
5 *
6 * Copyright (C) 2006 Qumranet, Inc.
4d5c5d0f
BAY		 7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9611c187 9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
043405e1
CO		 10 *
11 * Authors:
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
4d5c5d0f
BAY		 14 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
043405e1
CO		 16 *
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
19 *
20 */
21
edf88417 22#include <linux/kvm_host.h>
313a3dc7 23#include "irq.h"
1d737c8a 24#include "mmu.h"
7837699f 25#include "i8254.h"
37817f29 26#include "tss.h"
5fdbf976 27#include "kvm_cache_regs.h"
26eef70c 28#include "x86.h"
00b27a3e 29#include "cpuid.h"
474a5bb9 30#include "pmu.h"
e83d5887 31#include "hyperv.h"
313a3dc7 32
18068523 33#include <linux/clocksource.h>
4d5c5d0f 34#include <linux/interrupt.h>
313a3dc7
CO		 35#include <linux/kvm.h>
36#include <linux/fs.h>
37#include <linux/vmalloc.h>
1767e931
PG		 38#include <linux/export.h>
39#include <linux/moduleparam.h>
0de10343 40#include <linux/mman.h>
2bacc55c 41#include <linux/highmem.h>
19de40a8 42#include <linux/iommu.h>
62c476c7 43#include <linux/intel-iommu.h>
c8076604 44#include <linux/cpufreq.h>
18863bdd 45#include <linux/user-return-notifier.h>
a983fb23 46#include <linux/srcu.h>
5a0e3ad6 47#include <linux/slab.h>
ff9d07a0 48#include <linux/perf_event.h>
7bee342a 49#include <linux/uaccess.h>
af585b92 50#include <linux/hash.h>
a1b60c1c 51#include <linux/pci.h>
16e8d74d
MT		 52#include <linux/timekeeper_internal.h>
53#include <linux/pvclock_gtod.h>
87276880
FW		 54#include <linux/kvm_irqfd.h>
55#include <linux/irqbypass.h>
3905f9ad 56#include <linux/sched/stat.h>
d0ec49d4 57#include <linux/mem_encrypt.h>
3905f9ad 58
aec51dc4 59#include <trace/events/kvm.h>
2ed152af 60
24f1e32c 61#include <asm/debugreg.h>
d825ed0a 62#include <asm/msr.h>
a5f61300 63#include <asm/desc.h>
890ca9ae 64#include <asm/mce.h>
f89e32e0 65#include <linux/kernel_stat.h>
78f7f1e5 66#include <asm/fpu/internal.h> /* Ugh! */
1d5f066e 67#include <asm/pvclock.h>
217fc9cf 68#include <asm/div64.h>
efc64404 69#include <asm/irq_remapping.h>
b0c39dc6 70#include <asm/mshyperv.h>
0092e434 71#include <asm/hypervisor.h>
043405e1 72
d1898b73
DH		 73#define CREATE_TRACE_POINTS
74#include "trace.h"
75
313a3dc7 76#define MAX_IO_MSRS 256
890ca9ae 77#define KVM_MAX_MCE_BANKS 32
c45dcc71
AR		 78u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
79EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
890ca9ae 80
0f65dd70
AK		 81#define emul_to_vcpu(ctxt) \
82 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
83
50a37eb4
JR		 84/* EFER defaults:
85 * - enable syscall per default because its emulated by KVM
86 * - enable LME and LMA per default on 64 bit KVM
87 */
88#ifdef CONFIG_X86_64
1260edbe
LJ		 89static
90u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
50a37eb4 91#else
1260edbe 92static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
50a37eb4 93#endif
313a3dc7 94
ba1389b7
AK		 95#define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
96#define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
417bc304 97
c519265f
RK		 98#define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
99 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
37131313 100
cb142eb7 101static void update_cr8_intercept(struct kvm_vcpu *vcpu);
7460fb4a 102static void process_nmi(struct kvm_vcpu *vcpu);
ee2cd4b7 103static void enter_smm(struct kvm_vcpu *vcpu);
6addfc42 104static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
01643c51
KH		 105static void store_regs(struct kvm_vcpu *vcpu);
106static int sync_regs(struct kvm_vcpu *vcpu);
674eea0f 107
893590c7 108struct kvm_x86_ops *kvm_x86_ops __read_mostly;
5fdbf976 109EXPORT_SYMBOL_GPL(kvm_x86_ops);
97896d04 110
893590c7 111static bool __read_mostly ignore_msrs = 0;
476bc001 112module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
ed85c068 113
fab0aa3b
EM		 114static bool __read_mostly report_ignored_msrs = true;
115module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
116
9ed96e87
MT		 117unsigned int min_timer_period_us = 500;
118module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
119
630994b3
MT		 120static bool __read_mostly kvmclock_periodic_sync = true;
121module_param(kvmclock_periodic_sync, bool, S_IRUGO);
122
893590c7 123bool __read_mostly kvm_has_tsc_control;
92a1f12d 124EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
893590c7 125u32 __read_mostly kvm_max_guest_tsc_khz;
92a1f12d 126EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
bc9b961b
HZ		 127u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
128EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
129u64 __read_mostly kvm_max_tsc_scaling_ratio;
130EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
64672c95
YJ		 131u64 __read_mostly kvm_default_tsc_scaling_ratio;
132EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
92a1f12d 133
cc578287 134/* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
893590c7 135static u32 __read_mostly tsc_tolerance_ppm = 250;
cc578287
ZA		 136module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
137
d0659d94 138/* lapic timer advance (tscdeadline mode only) in nanoseconds */
893590c7 139unsigned int __read_mostly lapic_timer_advance_ns = 0;
d0659d94
MT		 140module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
141
52004014
FW		 142static bool __read_mostly vector_hashing = true;
143module_param(vector_hashing, bool, S_IRUGO);
144
c4ae60e4
LA		 145bool __read_mostly enable_vmware_backdoor = false;
146module_param(enable_vmware_backdoor, bool, S_IRUGO);
147EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
148
18863bdd
AK		 149#define KVM_NR_SHARED_MSRS 16
150
151struct kvm_shared_msrs_global {
152 int nr;
2bf78fa7 153 u32 msrs[KVM_NR_SHARED_MSRS];
18863bdd
AK		 154};
155
156struct kvm_shared_msrs {
157 struct user_return_notifier urn;
158 bool registered;
2bf78fa7
SY		 159 struct kvm_shared_msr_values {
160 u64 host;
161 u64 curr;
162 } values[KVM_NR_SHARED_MSRS];
18863bdd
AK		 163};
164
165static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
013f6a5d 166static struct kvm_shared_msrs __percpu *shared_msrs;
18863bdd 167
417bc304 168struct kvm_stats_debugfs_item debugfs_entries[] = {
ba1389b7
AK		 169 { "pf_fixed", VCPU_STAT(pf_fixed) },
170 { "pf_guest", VCPU_STAT(pf_guest) },
171 { "tlb_flush", VCPU_STAT(tlb_flush) },
172 { "invlpg", VCPU_STAT(invlpg) },
173 { "exits", VCPU_STAT(exits) },
174 { "io_exits", VCPU_STAT(io_exits) },
175 { "mmio_exits", VCPU_STAT(mmio_exits) },
176 { "signal_exits", VCPU_STAT(signal_exits) },
177 { "irq_window", VCPU_STAT(irq_window_exits) },
f08864b4 178 { "nmi_window", VCPU_STAT(nmi_window_exits) },
ba1389b7 179 { "halt_exits", VCPU_STAT(halt_exits) },
f7819512 180 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
62bea5bf 181 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
3491caf2 182 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
ba1389b7 183 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
f11c3a8d 184 { "hypercalls", VCPU_STAT(hypercalls) },
ba1389b7
AK		 185 { "request_irq", VCPU_STAT(request_irq_exits) },
186 { "irq_exits", VCPU_STAT(irq_exits) },
187 { "host_state_reload", VCPU_STAT(host_state_reload) },
ba1389b7
AK		 188 { "fpu_reload", VCPU_STAT(fpu_reload) },
189 { "insn_emulation", VCPU_STAT(insn_emulation) },
190 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
fa89a817 191 { "irq_injections", VCPU_STAT(irq_injections) },
c4abb7c9 192 { "nmi_injections", VCPU_STAT(nmi_injections) },
0f1e261e 193 { "req_event", VCPU_STAT(req_event) },
4cee5764
AK		 194 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
195 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
196 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
197 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
198 { "mmu_flooded", VM_STAT(mmu_flooded) },
199 { "mmu_recycled", VM_STAT(mmu_recycled) },
dfc5aa00 200 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
4731d4c7 201 { "mmu_unsync", VM_STAT(mmu_unsync) },
0f74a24c 202 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
05da4558 203 { "largepages", VM_STAT(lpages) },
f3414bc7
DM		 204 { "max_mmu_page_hash_collisions",
205 VM_STAT(max_mmu_page_hash_collisions) },
417bc304
HB		 206 { NULL }
207};
208
2acf923e
DC		 209u64 __read_mostly host_xcr0;
210
b6785def 211static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
d6aa1000 212
af585b92
GN		 213static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
214{
215 int i;
216 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
217 vcpu->arch.apf.gfns[i] = ~0;
218}
219
18863bdd
AK		 220static void kvm_on_user_return(struct user_return_notifier *urn)
221{
222 unsigned slot;
18863bdd
AK		 223 struct kvm_shared_msrs *locals
224 = container_of(urn, struct kvm_shared_msrs, urn);
2bf78fa7 225 struct kvm_shared_msr_values *values;
1650b4eb
IA		 226 unsigned long flags;
227
228 /*
229 * Disabling irqs at this point since the following code could be
230 * interrupted and executed through kvm_arch_hardware_disable()
231 */
232 local_irq_save(flags);
233 if (locals->registered) {
234 locals->registered = false;
235 user_return_notifier_unregister(urn);
236 }
237 local_irq_restore(flags);
18863bdd 238 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
2bf78fa7
SY		 239 values = &locals->values[slot];
240 if (values->host != values->curr) {
241 wrmsrl(shared_msrs_global.msrs[slot], values->host);
242 values->curr = values->host;
18863bdd
AK		 243 }
244 }
18863bdd
AK		 245}
246
2bf78fa7 247static void shared_msr_update(unsigned slot, u32 msr)
18863bdd 248{
18863bdd 249 u64 value;
013f6a5d
MT		 250 unsigned int cpu = smp_processor_id();
251 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
18863bdd 252
2bf78fa7
SY		 253 /* only read, and nobody should modify it at this time,
254 * so don't need lock */
255 if (slot >= shared_msrs_global.nr) {
256 printk(KERN_ERR "kvm: invalid MSR slot!");
257 return;
258 }
259 rdmsrl_safe(msr, &value);
260 smsr->values[slot].host = value;
261 smsr->values[slot].curr = value;
262}
263
264void kvm_define_shared_msr(unsigned slot, u32 msr)
265{
0123be42 266 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
c847fe88 267 shared_msrs_global.msrs[slot] = msr;
18863bdd
AK		 268 if (slot >= shared_msrs_global.nr)
269 shared_msrs_global.nr = slot + 1;
18863bdd
AK		 270}
271EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
272
273static void kvm_shared_msr_cpu_online(void)
274{
275 unsigned i;
18863bdd
AK		 276
277 for (i = 0; i < shared_msrs_global.nr; ++i)
2bf78fa7 278 shared_msr_update(i, shared_msrs_global.msrs[i]);
18863bdd
AK		 279}
280
8b3c3104 281int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
18863bdd 282{
013f6a5d
MT		 283 unsigned int cpu = smp_processor_id();
284 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
8b3c3104 285 int err;
18863bdd 286
2bf78fa7 287 if (((value ^ smsr->values[slot].curr) & mask) == 0)
8b3c3104 288 return 0;
2bf78fa7 289 smsr->values[slot].curr = value;
8b3c3104
AH		 290 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
291 if (err)
292 return 1;
293
18863bdd
AK		 294 if (!smsr->registered) {
295 smsr->urn.on_user_return = kvm_on_user_return;
296 user_return_notifier_register(&smsr->urn);
297 smsr->registered = true;
298 }
8b3c3104 299 return 0;
18863bdd
AK		 300}
301EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
302
13a34e06 303static void drop_user_return_notifiers(void)
3548bab5 304{
013f6a5d
MT		 305 unsigned int cpu = smp_processor_id();
306 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
3548bab5
AK		 307
308 if (smsr->registered)
309 kvm_on_user_return(&smsr->urn);
310}
311
6866b83e
CO		 312u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
313{
8a5a87d9 314 return vcpu->arch.apic_base;
6866b83e
CO		 315}
316EXPORT_SYMBOL_GPL(kvm_get_apic_base);
317
58cb628d
JK		 318int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
319{
320 u64 old_state = vcpu->arch.apic_base &
321 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
322 u64 new_state = msr_info->data &
323 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
d6321d49
RK		 324 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
325 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
58cb628d 326
d3802286
JM		 327 if ((msr_info->data & reserved_bits) || new_state == X2APIC_ENABLE)
328 return 1;
58cb628d 329 if (!msr_info->host_initiated &&
d3802286 330 ((new_state == MSR_IA32_APICBASE_ENABLE &&
58cb628d
JK		 331 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
332 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
333 old_state == 0)))
334 return 1;
335
336 kvm_lapic_set_base(vcpu, msr_info->data);
337 return 0;
6866b83e
CO		 338}
339EXPORT_SYMBOL_GPL(kvm_set_apic_base);
340
2605fc21 341asmlinkage __visible void kvm_spurious_fault(void)
e3ba45b8
GL		 342{
343 /* Fault while not rebooting. We want the trace. */
344 BUG();
345}
346EXPORT_SYMBOL_GPL(kvm_spurious_fault);
347
3fd28fce
ED		 348#define EXCPT_BENIGN 0
349#define EXCPT_CONTRIBUTORY 1
350#define EXCPT_PF 2
351
352static int exception_class(int vector)
353{
354 switch (vector) {
355 case PF_VECTOR:
356 return EXCPT_PF;
357 case DE_VECTOR:
358 case TS_VECTOR:
359 case NP_VECTOR:
360 case SS_VECTOR:
361 case GP_VECTOR:
362 return EXCPT_CONTRIBUTORY;
363 default:
364 break;
365 }
366 return EXCPT_BENIGN;
367}
368
d6e8c854
NA		 369#define EXCPT_FAULT 0
370#define EXCPT_TRAP 1
371#define EXCPT_ABORT 2
372#define EXCPT_INTERRUPT 3
373
374static int exception_type(int vector)
375{
376 unsigned int mask;
377
378 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
379 return EXCPT_INTERRUPT;
380
381 mask = 1 << vector;
382
383 /* #DB is trap, as instruction watchpoints are handled elsewhere */
384 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
385 return EXCPT_TRAP;
386
387 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
388 return EXCPT_ABORT;
389
390 /* Reserved exceptions will result in fault */
391 return EXCPT_FAULT;
392}
393
3fd28fce 394static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
ce7ddec4
JR		 395 unsigned nr, bool has_error, u32 error_code,
396 bool reinject)
3fd28fce
ED		 397{
398 u32 prev_nr;
399 int class1, class2;
400
3842d135
AK		 401 kvm_make_request(KVM_REQ_EVENT, vcpu);
402
664f8e26 403 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
3fd28fce 404 queue:
3ffb2468
NA		 405 if (has_error && !is_protmode(vcpu))
406 has_error = false;
664f8e26
WL		 407 if (reinject) {
408 /*
409 * On vmentry, vcpu->arch.exception.pending is only
410 * true if an event injection was blocked by
411 * nested_run_pending. In that case, however,
412 * vcpu_enter_guest requests an immediate exit,
413 * and the guest shouldn't proceed far enough to
414 * need reinjection.
415 */
416 WARN_ON_ONCE(vcpu->arch.exception.pending);
417 vcpu->arch.exception.injected = true;
418 } else {
419 vcpu->arch.exception.pending = true;
420 vcpu->arch.exception.injected = false;
421 }
3fd28fce
ED		 422 vcpu->arch.exception.has_error_code = has_error;
423 vcpu->arch.exception.nr = nr;
424 vcpu->arch.exception.error_code = error_code;
425 return;
426 }
427
428 /* to check exception */
429 prev_nr = vcpu->arch.exception.nr;
430 if (prev_nr == DF_VECTOR) {
431 /* triple fault -> shutdown */
a8eeb04a 432 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3fd28fce
ED		 433 return;
434 }
435 class1 = exception_class(prev_nr);
436 class2 = exception_class(nr);
437 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
438 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
664f8e26
WL		 439 /*
440 * Generate double fault per SDM Table 5-5. Set
441 * exception.pending = true so that the double fault
442 * can trigger a nested vmexit.
443 */
3fd28fce 444 vcpu->arch.exception.pending = true;
664f8e26 445 vcpu->arch.exception.injected = false;
3fd28fce
ED		 446 vcpu->arch.exception.has_error_code = true;
447 vcpu->arch.exception.nr = DF_VECTOR;
448 vcpu->arch.exception.error_code = 0;
449 } else
450 /* replace previous exception with a new one in a hope
451 that instruction re-execution will regenerate lost
452 exception */
453 goto queue;
454}
455
298101da
AK		 456void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
457{
ce7ddec4 458 kvm_multiple_exception(vcpu, nr, false, 0, false);
298101da
AK		 459}
460EXPORT_SYMBOL_GPL(kvm_queue_exception);
461
ce7ddec4
JR		 462void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
463{
464 kvm_multiple_exception(vcpu, nr, false, 0, true);
465}
466EXPORT_SYMBOL_GPL(kvm_requeue_exception);
467
6affcbed 468int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
c3c91fee 469{
db8fcefa
AP		 470 if (err)
471 kvm_inject_gp(vcpu, 0);
472 else
6affcbed
KH		 473 return kvm_skip_emulated_instruction(vcpu);
474
475 return 1;
db8fcefa
AP		 476}
477EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
8df25a32 478
6389ee94 479void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
c3c91fee
AK		 480{
481 ++vcpu->stat.pf_guest;
adfe20fb
WL		 482 vcpu->arch.exception.nested_apf =
483 is_guest_mode(vcpu) && fault->async_page_fault;
484 if (vcpu->arch.exception.nested_apf)
485 vcpu->arch.apf.nested_apf_token = fault->address;
486 else
487 vcpu->arch.cr2 = fault->address;
6389ee94 488 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
c3c91fee 489}
27d6c865 490EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
c3c91fee 491
ef54bcfe 492static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
d4f8cf66 493{
6389ee94
AK		 494 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
495 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
d4f8cf66 496 else
6389ee94 497 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
ef54bcfe
PB		 498
499 return fault->nested_page_fault;
d4f8cf66
JR		 500}
501
3419ffc8
SY		 502void kvm_inject_nmi(struct kvm_vcpu *vcpu)
503{
7460fb4a
AK		 504 atomic_inc(&vcpu->arch.nmi_queued);
505 kvm_make_request(KVM_REQ_NMI, vcpu);
3419ffc8
SY		 506}
507EXPORT_SYMBOL_GPL(kvm_inject_nmi);
508
298101da
AK		 509void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
510{
ce7ddec4 511 kvm_multiple_exception(vcpu, nr, true, error_code, false);
298101da
AK		 512}
513EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
514
ce7ddec4
JR		 515void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
516{
517 kvm_multiple_exception(vcpu, nr, true, error_code, true);
518}
519EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
520
0a79b009
AK		 521/*
522 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
523 * a #GP and return false.
524 */
525bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
298101da 526{
0a79b009
AK		 527 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
528 return true;
529 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
530 return false;
298101da 531}
0a79b009 532EXPORT_SYMBOL_GPL(kvm_require_cpl);
298101da 533
16f8a6f9
NA		 534bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
535{
536 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
537 return true;
538
539 kvm_queue_exception(vcpu, UD_VECTOR);
540 return false;
541}
542EXPORT_SYMBOL_GPL(kvm_require_dr);
543
ec92fe44
JR		 544/*
545 * This function will be used to read from the physical memory of the currently
54bf36aa 546 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
ec92fe44
JR		 547 * can read from guest physical or from the guest's guest physical memory.
548 */
549int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
550 gfn_t ngfn, void *data, int offset, int len,
551 u32 access)
552{
54987b7a 553 struct x86_exception exception;
ec92fe44
JR		 554 gfn_t real_gfn;
555 gpa_t ngpa;
556
557 ngpa = gfn_to_gpa(ngfn);
54987b7a 558 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
ec92fe44
JR		 559 if (real_gfn == UNMAPPED_GVA)
560 return -EFAULT;
561
562 real_gfn = gpa_to_gfn(real_gfn);
563
54bf36aa 564 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
ec92fe44
JR		 565}
566EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
567
69b0049a 568static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
3d06b8bf
JR		 569 void *data, int offset, int len, u32 access)
570{
571 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
572 data, offset, len, access);
573}
574
a03490ed
CO		 575/*
576 * Load the pae pdptrs. Return true is they are all valid.
577 */
ff03a073 578int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
a03490ed
CO		 579{
580 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
581 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
582 int i;
583 int ret;
ff03a073 584 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
a03490ed 585
ff03a073
JR		 586 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
587 offset * sizeof(u64), sizeof(pdpte),
588 PFERR_USER_MASK|PFERR_WRITE_MASK);
a03490ed
CO		 589 if (ret < 0) {
590 ret = 0;
591 goto out;
592 }
593 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
812f30b2 594 if ((pdpte[i] & PT_PRESENT_MASK) &&
a0a64f50
XG		 595 (pdpte[i] &
596 vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
a03490ed
CO		 597 ret = 0;
598 goto out;
599 }
600 }
601 ret = 1;
602
ff03a073 603 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
6de4f3ad
AK		 604 __set_bit(VCPU_EXREG_PDPTR,
605 (unsigned long *)&vcpu->arch.regs_avail);
606 __set_bit(VCPU_EXREG_PDPTR,
607 (unsigned long *)&vcpu->arch.regs_dirty);
a03490ed 608out:
a03490ed
CO		 609
610 return ret;
611}
cc4b6871 612EXPORT_SYMBOL_GPL(load_pdptrs);
a03490ed 613
9ed38ffa 614bool pdptrs_changed(struct kvm_vcpu *vcpu)
d835dfec 615{
ff03a073 616 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
d835dfec 617 bool changed = true;
3d06b8bf
JR		 618 int offset;
619 gfn_t gfn;
d835dfec
AK		 620 int r;
621
622 if (is_long_mode(vcpu) || !is_pae(vcpu))
623 return false;
624
6de4f3ad
AK		 625 if (!test_bit(VCPU_EXREG_PDPTR,
626 (unsigned long *)&vcpu->arch.regs_avail))
627 return true;
628
a512177e
PB		 629 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
630 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
3d06b8bf
JR		 631 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
632 PFERR_USER_MASK | PFERR_WRITE_MASK);
d835dfec
AK		 633 if (r < 0)
634 goto out;
ff03a073 635 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
d835dfec 636out:
d835dfec
AK		 637
638 return changed;
639}
9ed38ffa 640EXPORT_SYMBOL_GPL(pdptrs_changed);
d835dfec 641
49a9b07e 642int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
a03490ed 643{
aad82703 644 unsigned long old_cr0 = kvm_read_cr0(vcpu);
d81135a5 645 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
aad82703 646
f9a48e6a
AK		 647 cr0 |= X86_CR0_ET;
648
ab344828 649#ifdef CONFIG_X86_64
0f12244f
GN		 650 if (cr0 & 0xffffffff00000000UL)
651 return 1;
ab344828
GN		 652#endif
653
654 cr0 &= ~CR0_RESERVED_BITS;
a03490ed 655
0f12244f
GN		 656 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
657 return 1;
a03490ed 658
0f12244f
GN		 659 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
660 return 1;
a03490ed
CO		 661
662 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
663#ifdef CONFIG_X86_64
f6801dff 664 if ((vcpu->arch.efer & EFER_LME)) {
a03490ed
CO		 665 int cs_db, cs_l;
666
0f12244f
GN		 667 if (!is_pae(vcpu))
668 return 1;
a03490ed 669 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
0f12244f
GN		 670 if (cs_l)
671 return 1;
a03490ed
CO		 672 } else
673#endif
ff03a073 674 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
9f8fe504 675 kvm_read_cr3(vcpu)))
0f12244f 676 return 1;
a03490ed
CO		 677 }
678
ad756a16
MJ		 679 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
680 return 1;
681
a03490ed 682 kvm_x86_ops->set_cr0(vcpu, cr0);
a03490ed 683
d170c419 684 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
e5f3f027 685 kvm_clear_async_pf_completion_queue(vcpu);
d170c419
LJ		 686 kvm_async_pf_hash_reset(vcpu);
687 }
e5f3f027 688
aad82703
SY		 689 if ((cr0 ^ old_cr0) & update_bits)
690 kvm_mmu_reset_context(vcpu);
b18d5431 691
879ae188
LE		 692 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
693 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
694 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
b18d5431
XG		 695 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
696
0f12244f
GN		 697 return 0;
698}
2d3ad1f4 699EXPORT_SYMBOL_GPL(kvm_set_cr0);
a03490ed 700
2d3ad1f4 701void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
a03490ed 702{
49a9b07e 703 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
a03490ed 704}
2d3ad1f4 705EXPORT_SYMBOL_GPL(kvm_lmsw);
a03490ed 706
42bdf991
MT		 707static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
708{
709 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
710 !vcpu->guest_xcr0_loaded) {
711 /* kvm_set_xcr() also depends on this */
476b7ada
PB		 712 if (vcpu->arch.xcr0 != host_xcr0)
713 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
42bdf991
MT		 714 vcpu->guest_xcr0_loaded = 1;
715 }
716}
717
718static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
719{
720 if (vcpu->guest_xcr0_loaded) {
721 if (vcpu->arch.xcr0 != host_xcr0)
722 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
723 vcpu->guest_xcr0_loaded = 0;
724 }
725}
726
69b0049a 727static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
2acf923e 728{
56c103ec
LJ		 729 u64 xcr0 = xcr;
730 u64 old_xcr0 = vcpu->arch.xcr0;
46c34cb0 731 u64 valid_bits;
2acf923e
DC		 732
733 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
734 if (index != XCR_XFEATURE_ENABLED_MASK)
735 return 1;
d91cab78 736 if (!(xcr0 & XFEATURE_MASK_FP))
2acf923e 737 return 1;
d91cab78 738 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
2acf923e 739 return 1;
46c34cb0
PB		 740
741 /*
742 * Do not allow the guest to set bits that we do not support
743 * saving. However, xcr0 bit 0 is always set, even if the
744 * emulated CPU does not support XSAVE (see fx_init).
745 */
d91cab78 746 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
46c34cb0 747 if (xcr0 & ~valid_bits)
2acf923e 748 return 1;
46c34cb0 749
d91cab78
DH		 750 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
751 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
390bd528
LJ		 752 return 1;
753
d91cab78
DH		 754 if (xcr0 & XFEATURE_MASK_AVX512) {
755 if (!(xcr0 & XFEATURE_MASK_YMM))
612263b3 756 return 1;
d91cab78 757 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
612263b3
CP		 758 return 1;
759 }
2acf923e 760 vcpu->arch.xcr0 = xcr0;
56c103ec 761
d91cab78 762 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
56c103ec 763 kvm_update_cpuid(vcpu);
2acf923e
DC		 764 return 0;
765}
766
767int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
768{
764bcbc5
Z		 769 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
770 __kvm_set_xcr(vcpu, index, xcr)) {
2acf923e
DC		 771 kvm_inject_gp(vcpu, 0);
772 return 1;
773 }
774 return 0;
775}
776EXPORT_SYMBOL_GPL(kvm_set_xcr);
777
a83b29c6 778int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
a03490ed 779{
fc78f519 780 unsigned long old_cr4 = kvm_read_cr4(vcpu);
0be0226f 781 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
b9baba86 782 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
0be0226f 783
0f12244f
GN		 784 if (cr4 & CR4_RESERVED_BITS)
785 return 1;
a03490ed 786
d6321d49 787 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
2acf923e
DC		 788 return 1;
789
d6321d49 790 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
2acf923e
DC		 791 return 1;
792
d6321d49 793 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
c68b734f
YW		 794 return 1;
795
d6321d49 796 if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
97ec8c06
FW		 797 return 1;
798
d6321d49 799 if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
74dc2b4f
YW		 800 return 1;
801
fd8cb433 802 if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
b9baba86
HH		 803 return 1;
804
ae3e61e1
PB		 805 if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
806 return 1;
807
a03490ed 808 if (is_long_mode(vcpu)) {
0f12244f
GN		 809 if (!(cr4 & X86_CR4_PAE))
810 return 1;
a2edf57f
AK		 811 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
812 && ((cr4 ^ old_cr4) & pdptr_bits)
9f8fe504
AK		 813 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
814 kvm_read_cr3(vcpu)))
0f12244f
GN		 815 return 1;
816
ad756a16 817 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
d6321d49 818 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
ad756a16
MJ		 819 return 1;
820
821 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
822 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
823 return 1;
824 }
825
5e1746d6 826 if (kvm_x86_ops->set_cr4(vcpu, cr4))
0f12244f 827 return 1;
a03490ed 828
ad756a16
MJ		 829 if (((cr4 ^ old_cr4) & pdptr_bits) ||
830 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
aad82703 831 kvm_mmu_reset_context(vcpu);
0f12244f 832
b9baba86 833 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
00b27a3e 834 kvm_update_cpuid(vcpu);
2acf923e 835
0f12244f
GN		 836 return 0;
837}
2d3ad1f4 838EXPORT_SYMBOL_GPL(kvm_set_cr4);
a03490ed 839
2390218b 840int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
a03490ed 841{
ac146235 842#ifdef CONFIG_X86_64
9d88fca7 843 cr3 &= ~CR3_PCID_INVD;
ac146235 844#endif
9d88fca7 845
9f8fe504 846 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
0ba73cda 847 kvm_mmu_sync_roots(vcpu);
77c3913b 848 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
0f12244f 849 return 0;
d835dfec
AK		 850 }
851
d1cd3ce9
YZ		 852 if (is_long_mode(vcpu) &&
853 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 62)))
854 return 1;
855 else if (is_pae(vcpu) && is_paging(vcpu) &&
d9f89b88 856 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
346874c9 857 return 1;
a03490ed 858
0f12244f 859 vcpu->arch.cr3 = cr3;
aff48baa 860 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
d8d173da 861 kvm_mmu_new_cr3(vcpu);
0f12244f
GN		 862 return 0;
863}
2d3ad1f4 864EXPORT_SYMBOL_GPL(kvm_set_cr3);
a03490ed 865
eea1cff9 866int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
a03490ed 867{
0f12244f
GN		 868 if (cr8 & CR8_RESERVED_BITS)
869 return 1;
35754c98 870 if (lapic_in_kernel(vcpu))
a03490ed
CO		 871 kvm_lapic_set_tpr(vcpu, cr8);
872 else
ad312c7c 873 vcpu->arch.cr8 = cr8;
0f12244f
GN		 874 return 0;
875}
2d3ad1f4 876EXPORT_SYMBOL_GPL(kvm_set_cr8);
a03490ed 877
2d3ad1f4 878unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
a03490ed 879{
35754c98 880 if (lapic_in_kernel(vcpu))
a03490ed
CO		 881 return kvm_lapic_get_cr8(vcpu);
882 else
ad312c7c 883 return vcpu->arch.cr8;
a03490ed 884}
2d3ad1f4 885EXPORT_SYMBOL_GPL(kvm_get_cr8);
a03490ed 886
ae561ede
NA		 887static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
888{
889 int i;
890
891 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
892 for (i = 0; i < KVM_NR_DB_REGS; i++)
893 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
894 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
895 }
896}
897
73aaf249
JK		 898static void kvm_update_dr6(struct kvm_vcpu *vcpu)
899{
900 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
901 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
902}
903
c8639010
JK		 904static void kvm_update_dr7(struct kvm_vcpu *vcpu)
905{
906 unsigned long dr7;
907
908 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
909 dr7 = vcpu->arch.guest_debug_dr7;
910 else
911 dr7 = vcpu->arch.dr7;
912 kvm_x86_ops->set_dr7(vcpu, dr7);
360b948d
PB		 913 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
914 if (dr7 & DR7_BP_EN_MASK)
915 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
c8639010
JK		 916}
917
6f43ed01
NA		 918static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
919{
920 u64 fixed = DR6_FIXED_1;
921
d6321d49 922 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
6f43ed01
NA		 923 fixed |= DR6_RTM;
924 return fixed;
925}
926
338dbc97 927static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
020df079
GN		 928{
929 switch (dr) {
930 case 0 ... 3:
931 vcpu->arch.db[dr] = val;
932 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
933 vcpu->arch.eff_db[dr] = val;
934 break;
935 case 4:
020df079
GN		 936 /* fall through */
937 case 6:
338dbc97
GN		 938 if (val & 0xffffffff00000000ULL)
939 return -1; /* #GP */
6f43ed01 940 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
73aaf249 941 kvm_update_dr6(vcpu);
020df079
GN		 942 break;
943 case 5:
020df079
GN		 944 /* fall through */
945 default: /* 7 */
338dbc97
GN		 946 if (val & 0xffffffff00000000ULL)
947 return -1; /* #GP */
020df079 948 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
c8639010 949 kvm_update_dr7(vcpu);
020df079
GN		 950 break;
951 }
952
953 return 0;
954}
338dbc97
GN		 955
956int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
957{
16f8a6f9 958 if (__kvm_set_dr(vcpu, dr, val)) {
338dbc97 959 kvm_inject_gp(vcpu, 0);
16f8a6f9
NA		 960 return 1;
961 }
962 return 0;
338dbc97 963}
020df079
GN		 964EXPORT_SYMBOL_GPL(kvm_set_dr);
965
16f8a6f9 966int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
020df079
GN		 967{
968 switch (dr) {
969 case 0 ... 3:
970 *val = vcpu->arch.db[dr];
971 break;
972 case 4:
020df079
GN		 973 /* fall through */
974 case 6:
73aaf249
JK		 975 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
976 *val = vcpu->arch.dr6;
977 else
978 *val = kvm_x86_ops->get_dr6(vcpu);
020df079
GN		 979 break;
980 case 5:
020df079
GN		 981 /* fall through */
982 default: /* 7 */
983 *val = vcpu->arch.dr7;
984 break;
985 }
338dbc97
GN		 986 return 0;
987}
020df079
GN		 988EXPORT_SYMBOL_GPL(kvm_get_dr);
989
022cd0e8
AK		 990bool kvm_rdpmc(struct kvm_vcpu *vcpu)
991{
992 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
993 u64 data;
994 int err;
995
c6702c9d 996 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
022cd0e8
AK		 997 if (err)
998 return err;
999 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
1000 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
1001 return err;
1002}
1003EXPORT_SYMBOL_GPL(kvm_rdpmc);
1004
043405e1
CO		 1005/*
1006 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1007 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1008 *
1009 * This list is modified at module load time to reflect the
e3267cbb 1010 * capabilities of the host cpu. This capabilities test skips MSRs that are
62ef68bb
PB		 1011 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1012 * may depend on host virtualization features rather than host cpu features.
043405e1 1013 */
e3267cbb 1014
043405e1
CO		 1015static u32 msrs_to_save[] = {
1016 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
8c06585d 1017 MSR_STAR,
043405e1
CO		 1018#ifdef CONFIG_X86_64
1019 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1020#endif
b3897a49 1021 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
9dbe6cf9 1022 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
d28b387f 1023 MSR_IA32_SPEC_CTRL, MSR_IA32_ARCH_CAPABILITIES
043405e1
CO		 1024};
1025
1026static unsigned num_msrs_to_save;
1027
62ef68bb
PB		 1028static u32 emulated_msrs[] = {
1029 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1030 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1031 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1032 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
72c139ba 1033 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
e7d9513b
AS		 1034 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1035 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
e516cebb 1036 HV_X64_MSR_RESET,
11c4b1ca 1037 HV_X64_MSR_VP_INDEX,
9eec50b8 1038 HV_X64_MSR_VP_RUNTIME,
5c919412 1039 HV_X64_MSR_SCONTROL,
1f4b34f8 1040 HV_X64_MSR_STIMER0_CONFIG,
d4abc577 1041 HV_X64_MSR_VP_ASSIST_PAGE,
a2e164e7
VK		 1042 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1043 HV_X64_MSR_TSC_EMULATION_STATUS,
1044
1045 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
62ef68bb
PB		 1046 MSR_KVM_PV_EOI_EN,
1047
ba904635 1048 MSR_IA32_TSC_ADJUST,
a3e06bbe 1049 MSR_IA32_TSCDEADLINE,
043405e1 1050 MSR_IA32_MISC_ENABLE,
908e75f3
AK		 1051 MSR_IA32_MCG_STATUS,
1052 MSR_IA32_MCG_CTL,
c45dcc71 1053 MSR_IA32_MCG_EXT_CTL,
64d60670 1054 MSR_IA32_SMBASE,
52797bf9 1055 MSR_SMI_COUNT,
db2336a8
KH		 1056 MSR_PLATFORM_INFO,
1057 MSR_MISC_FEATURES_ENABLES,
043405e1
CO		 1058};
1059
62ef68bb
PB		 1060static unsigned num_emulated_msrs;
1061
801e459a
TL		 1062/*
1063 * List of msr numbers which are used to expose MSR-based features that
1064 * can be used by a hypervisor to validate requested CPU features.
1065 */
1066static u32 msr_based_features[] = {
1389309c
PB		 1067 MSR_IA32_VMX_BASIC,
1068 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1069 MSR_IA32_VMX_PINBASED_CTLS,
1070 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1071 MSR_IA32_VMX_PROCBASED_CTLS,
1072 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1073 MSR_IA32_VMX_EXIT_CTLS,
1074 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1075 MSR_IA32_VMX_ENTRY_CTLS,
1076 MSR_IA32_VMX_MISC,
1077 MSR_IA32_VMX_CR0_FIXED0,
1078 MSR_IA32_VMX_CR0_FIXED1,
1079 MSR_IA32_VMX_CR4_FIXED0,
1080 MSR_IA32_VMX_CR4_FIXED1,
1081 MSR_IA32_VMX_VMCS_ENUM,
1082 MSR_IA32_VMX_PROCBASED_CTLS2,
1083 MSR_IA32_VMX_EPT_VPID_CAP,
1084 MSR_IA32_VMX_VMFUNC,
1085
d1d93fa9 1086 MSR_F10H_DECFG,
518e7b94 1087 MSR_IA32_UCODE_REV,
801e459a
TL		 1088};
1089
1090static unsigned int num_msr_based_features;
1091
66421c1e
WL		 1092static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1093{
1094 switch (msr->index) {
518e7b94
WL		 1095 case MSR_IA32_UCODE_REV:
1096 rdmsrl(msr->index, msr->data);
1097 break;
66421c1e
WL		 1098 default:
1099 if (kvm_x86_ops->get_msr_feature(msr))
1100 return 1;
1101 }
1102 return 0;
1103}
1104
801e459a
TL		 1105static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1106{
1107 struct kvm_msr_entry msr;
66421c1e 1108 int r;
801e459a
TL		 1109
1110 msr.index = index;
66421c1e
WL		 1111 r = kvm_get_msr_feature(&msr);
1112 if (r)
1113 return r;
801e459a
TL		 1114
1115 *data = msr.data;
1116
1117 return 0;
1118}
1119
384bb783 1120bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
15c4a640 1121{
b69e8cae 1122 if (efer & efer_reserved_bits)
384bb783 1123 return false;
15c4a640 1124
1b4d56b8 1125 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
384bb783 1126 return false;
1b2fd70c 1127
1b4d56b8 1128 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
384bb783 1129 return false;
d8017474 1130
384bb783
JK		 1131 return true;
1132}
1133EXPORT_SYMBOL_GPL(kvm_valid_efer);
1134
1135static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1136{
1137 u64 old_efer = vcpu->arch.efer;
1138
1139 if (!kvm_valid_efer(vcpu, efer))
1140 return 1;
1141
1142 if (is_paging(vcpu)
1143 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1144 return 1;
1145
15c4a640 1146 efer &= ~EFER_LMA;
f6801dff 1147 efer |= vcpu->arch.efer & EFER_LMA;
15c4a640 1148
a3d204e2
SY		 1149 kvm_x86_ops->set_efer(vcpu, efer);
1150
aad82703
SY		 1151 /* Update reserved bits */
1152 if ((efer ^ old_efer) & EFER_NX)
1153 kvm_mmu_reset_context(vcpu);
1154
b69e8cae 1155 return 0;
15c4a640
CO		 1156}
1157
f2b4b7dd
JR		 1158void kvm_enable_efer_bits(u64 mask)
1159{
1160 efer_reserved_bits &= ~mask;
1161}
1162EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1163
15c4a640
CO		 1164/*
1165 * Writes msr value into into the appropriate "register".
1166 * Returns 0 on success, non-0 otherwise.
1167 * Assumes vcpu_load() was already called.
1168 */
8fe8ab46 1169int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
15c4a640 1170{
854e8bb1
NA		 1171 switch (msr->index) {
1172 case MSR_FS_BASE:
1173 case MSR_GS_BASE:
1174 case MSR_KERNEL_GS_BASE:
1175 case MSR_CSTAR:
1176 case MSR_LSTAR:
fd8cb433 1177 if (is_noncanonical_address(msr->data, vcpu))
854e8bb1
NA		 1178 return 1;
1179 break;
1180 case MSR_IA32_SYSENTER_EIP:
1181 case MSR_IA32_SYSENTER_ESP:
1182 /*
1183 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1184 * non-canonical address is written on Intel but not on
1185 * AMD (which ignores the top 32-bits, because it does
1186 * not implement 64-bit SYSENTER).
1187 *
1188 * 64-bit code should hence be able to write a non-canonical
1189 * value on AMD. Making the address canonical ensures that
1190 * vmentry does not fail on Intel after writing a non-canonical
1191 * value, and that something deterministic happens if the guest
1192 * invokes 64-bit SYSENTER.
1193 */
fd8cb433 1194 msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
854e8bb1 1195 }
8fe8ab46 1196 return kvm_x86_ops->set_msr(vcpu, msr);
15c4a640 1197}
854e8bb1 1198EXPORT_SYMBOL_GPL(kvm_set_msr);
15c4a640 1199
313a3dc7
CO		 1200/*
1201 * Adapt set_msr() to msr_io()'s calling convention
1202 */
609e36d3
PB		 1203static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1204{
1205 struct msr_data msr;
1206 int r;
1207
1208 msr.index = index;
1209 msr.host_initiated = true;
1210 r = kvm_get_msr(vcpu, &msr);
1211 if (r)
1212 return r;
1213
1214 *data = msr.data;
1215 return 0;
1216}
1217
313a3dc7
CO		 1218static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1219{
8fe8ab46
WA		 1220 struct msr_data msr;
1221
1222 msr.data = *data;
1223 msr.index = index;
1224 msr.host_initiated = true;
1225 return kvm_set_msr(vcpu, &msr);
313a3dc7
CO		 1226}
1227
16e8d74d
MT		 1228#ifdef CONFIG_X86_64
1229struct pvclock_gtod_data {
1230 seqcount_t seq;
1231
1232 struct { /* extract of a clocksource struct */
1233 int vclock_mode;
a5a1d1c2
TG		 1234 u64 cycle_last;
1235 u64 mask;
16e8d74d
MT		 1236 u32 mult;
1237 u32 shift;
1238 } clock;
1239
cbcf2dd3
TG		 1240 u64 boot_ns;
1241 u64 nsec_base;
55dd00a7 1242 u64 wall_time_sec;
16e8d74d
MT		 1243};
1244
1245static struct pvclock_gtod_data pvclock_gtod_data;
1246
1247static void update_pvclock_gtod(struct timekeeper *tk)
1248{
1249 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
cbcf2dd3
TG		 1250 u64 boot_ns;
1251
876e7881 1252 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
16e8d74d
MT		 1253
1254 write_seqcount_begin(&vdata->seq);
1255
1256 /* copy pvclock gtod data */
876e7881
PZ		 1257 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1258 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1259 vdata->clock.mask = tk->tkr_mono.mask;
1260 vdata->clock.mult = tk->tkr_mono.mult;
1261 vdata->clock.shift = tk->tkr_mono.shift;
16e8d74d 1262
cbcf2dd3 1263 vdata->boot_ns = boot_ns;
876e7881 1264 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
16e8d74d 1265
55dd00a7
MT		 1266 vdata->wall_time_sec = tk->xtime_sec;
1267
16e8d74d
MT		 1268 write_seqcount_end(&vdata->seq);
1269}
1270#endif
1271
bab5bb39
NK		 1272void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1273{
1274 /*
1275 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1276 * vcpu_enter_guest. This function is only called from
1277 * the physical CPU that is running vcpu.
1278 */
1279 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1280}
16e8d74d 1281
18068523
GOC		 1282static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1283{
9ed3c444
AK		 1284 int version;
1285 int r;
50d0a0f9 1286 struct pvclock_wall_clock wc;
87aeb54f 1287 struct timespec64 boot;
18068523
GOC		 1288
1289 if (!wall_clock)
1290 return;
1291
9ed3c444
AK		 1292 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1293 if (r)
1294 return;
1295
1296 if (version & 1)
1297 ++version; /* first time write, random junk */
1298
1299 ++version;
18068523 1300
1dab1345
NK		 1301 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1302 return;
18068523 1303
50d0a0f9
GH		 1304 /*
1305 * The guest calculates current wall clock time by adding
34c238a1 1306 * system time (updated by kvm_guest_time_update below) to the
50d0a0f9
GH		 1307 * wall clock specified here. guest system time equals host
1308 * system time for us, thus we must fill in host boot time here.
1309 */
87aeb54f 1310 getboottime64(&boot);
50d0a0f9 1311
4b648665 1312 if (kvm->arch.kvmclock_offset) {
87aeb54f
AB		 1313 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1314 boot = timespec64_sub(boot, ts);
4b648665 1315 }
87aeb54f 1316 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
50d0a0f9
GH		 1317 wc.nsec = boot.tv_nsec;
1318 wc.version = version;
18068523
GOC		 1319
1320 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1321
1322 version++;
1323 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
18068523
GOC		 1324}
1325
50d0a0f9
GH		 1326static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1327{
b51012de
PB		 1328 do_shl32_div32(dividend, divisor);
1329 return dividend;
50d0a0f9
GH		 1330}
1331
3ae13faa 1332static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
5f4e3f88 1333 s8 *pshift, u32 *pmultiplier)
50d0a0f9 1334{
5f4e3f88 1335 uint64_t scaled64;
50d0a0f9
GH		 1336 int32_t shift = 0;
1337 uint64_t tps64;
1338 uint32_t tps32;
1339
3ae13faa
PB		 1340 tps64 = base_hz;
1341 scaled64 = scaled_hz;
50933623 1342 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
50d0a0f9
GH		 1343 tps64 >>= 1;
1344 shift--;
1345 }
1346
1347 tps32 = (uint32_t)tps64;
50933623
JK		 1348 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1349 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
5f4e3f88
ZA		 1350 scaled64 >>= 1;
1351 else
1352 tps32 <<= 1;
50d0a0f9
GH		 1353 shift++;
1354 }
1355
5f4e3f88
ZA		 1356 *pshift = shift;
1357 *pmultiplier = div_frac(scaled64, tps32);
50d0a0f9 1358
3ae13faa
PB		 1359 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1360 __func__, base_hz, scaled_hz, shift, *pmultiplier);
50d0a0f9
GH		 1361}
1362
d828199e 1363#ifdef CONFIG_X86_64
16e8d74d 1364static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
d828199e 1365#endif
16e8d74d 1366
c8076604 1367static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
69b0049a 1368static unsigned long max_tsc_khz;
c8076604 1369
cc578287 1370static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1e993611 1371{
cc578287
ZA		 1372 u64 v = (u64)khz * (1000000 + ppm);
1373 do_div(v, 1000000);
1374 return v;
1e993611
JR		 1375}
1376
381d585c
HZ		 1377static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1378{
1379 u64 ratio;
1380
1381 /* Guest TSC same frequency as host TSC? */
1382 if (!scale) {
1383 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1384 return 0;
1385 }
1386
1387 /* TSC scaling supported? */
1388 if (!kvm_has_tsc_control) {
1389 if (user_tsc_khz > tsc_khz) {
1390 vcpu->arch.tsc_catchup = 1;
1391 vcpu->arch.tsc_always_catchup = 1;
1392 return 0;
1393 } else {
1394 WARN(1, "user requested TSC rate below hardware speed\n");
1395 return -1;
1396 }
1397 }
1398
1399 /* TSC scaling required - calculate ratio */
1400 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1401 user_tsc_khz, tsc_khz);
1402
1403 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1404 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1405 user_tsc_khz);
1406 return -1;
1407 }
1408
1409 vcpu->arch.tsc_scaling_ratio = ratio;
1410 return 0;
1411}
1412
4941b8cb 1413static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
759379dd 1414{
cc578287
ZA		 1415 u32 thresh_lo, thresh_hi;
1416 int use_scaling = 0;
217fc9cf 1417
03ba32ca 1418 /* tsc_khz can be zero if TSC calibration fails */
4941b8cb 1419 if (user_tsc_khz == 0) {
ad721883
HZ		 1420 /* set tsc_scaling_ratio to a safe value */
1421 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
381d585c 1422 return -1;
ad721883 1423 }
03ba32ca 1424
c285545f 1425 /* Compute a scale to convert nanoseconds in TSC cycles */
3ae13faa 1426 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
cc578287
ZA		 1427 &vcpu->arch.virtual_tsc_shift,
1428 &vcpu->arch.virtual_tsc_mult);
4941b8cb 1429 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
cc578287
ZA		 1430
1431 /*
1432 * Compute the variation in TSC rate which is acceptable
1433 * within the range of tolerance and decide if the
1434 * rate being applied is within that bounds of the hardware
1435 * rate. If so, no scaling or compensation need be done.
1436 */
1437 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1438 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
4941b8cb
PB		 1439 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1440 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
cc578287
ZA		 1441 use_scaling = 1;
1442 }
4941b8cb 1443 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
c285545f
ZA		 1444}
1445
1446static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1447{
e26101b1 1448 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
cc578287
ZA		 1449 vcpu->arch.virtual_tsc_mult,
1450 vcpu->arch.virtual_tsc_shift);
e26101b1 1451 tsc += vcpu->arch.this_tsc_write;
c285545f
ZA		 1452 return tsc;
1453}
1454
b0c39dc6
VK		 1455static inline int gtod_is_based_on_tsc(int mode)
1456{
1457 return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1458}
1459
69b0049a 1460static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
b48aa97e
MT		 1461{
1462#ifdef CONFIG_X86_64
1463 bool vcpus_matched;
b48aa97e
MT		 1464 struct kvm_arch *ka = &vcpu->kvm->arch;
1465 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1466
1467 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1468 atomic_read(&vcpu->kvm->online_vcpus));
1469
7f187922
MT		 1470 /*
1471 * Once the masterclock is enabled, always perform request in
1472 * order to update it.
1473 *
1474 * In order to enable masterclock, the host clocksource must be TSC
1475 * and the vcpus need to have matched TSCs. When that happens,
1476 * perform request to enable masterclock.
1477 */
1478 if (ka->use_master_clock ||
b0c39dc6 1479 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
b48aa97e
MT		 1480 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1481
1482 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1483 atomic_read(&vcpu->kvm->online_vcpus),
1484 ka->use_master_clock, gtod->clock.vclock_mode);
1485#endif
1486}
1487
ba904635
WA		 1488static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1489{
3e3f5026 1490 u64 curr_offset = vcpu->arch.tsc_offset;
ba904635
WA		 1491 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1492}
1493
35181e86
HZ		 1494/*
1495 * Multiply tsc by a fixed point number represented by ratio.
1496 *
1497 * The most significant 64-N bits (mult) of ratio represent the
1498 * integral part of the fixed point number; the remaining N bits
1499 * (frac) represent the fractional part, ie. ratio represents a fixed
1500 * point number (mult + frac * 2^(-N)).
1501 *
1502 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1503 */
1504static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1505{
1506 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1507}
1508
1509u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1510{
1511 u64 _tsc = tsc;
1512 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1513
1514 if (ratio != kvm_default_tsc_scaling_ratio)
1515 _tsc = __scale_tsc(ratio, tsc);
1516
1517 return _tsc;
1518}
1519EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1520
07c1419a
HZ		 1521static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1522{
1523 u64 tsc;
1524
1525 tsc = kvm_scale_tsc(vcpu, rdtsc());
1526
1527 return target_tsc - tsc;
1528}
1529
4ba76538
HZ		 1530u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1531{
ea26e4ec 1532 return vcpu->arch.tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
4ba76538
HZ		 1533}
1534EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1535
a545ab6a
LC		 1536static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1537{
1538 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1539 vcpu->arch.tsc_offset = offset;
1540}
1541
b0c39dc6
VK		 1542static inline bool kvm_check_tsc_unstable(void)
1543{
1544#ifdef CONFIG_X86_64
1545 /*
1546 * TSC is marked unstable when we're running on Hyper-V,
1547 * 'TSC page' clocksource is good.
1548 */
1549 if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1550 return false;
1551#endif
1552 return check_tsc_unstable();
1553}
1554
8fe8ab46 1555void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
99e3e30a
ZA		 1556{
1557 struct kvm *kvm = vcpu->kvm;
f38e098f 1558 u64 offset, ns, elapsed;
99e3e30a 1559 unsigned long flags;
b48aa97e 1560 bool matched;
0d3da0d2 1561 bool already_matched;
8fe8ab46 1562 u64 data = msr->data;
c5e8ec8e 1563 bool synchronizing = false;
99e3e30a 1564
038f8c11 1565 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
07c1419a 1566 offset = kvm_compute_tsc_offset(vcpu, data);
108b249c 1567 ns = ktime_get_boot_ns();
f38e098f 1568 elapsed = ns - kvm->arch.last_tsc_nsec;
5d3cb0f6 1569
03ba32ca 1570 if (vcpu->arch.virtual_tsc_khz) {
bd8fab39
DP		 1571 if (data == 0 && msr->host_initiated) {
1572 /*
1573 * detection of vcpu initialization -- need to sync
1574 * with other vCPUs. This particularly helps to keep
1575 * kvm_clock stable after CPU hotplug
1576 */
1577 synchronizing = true;
1578 } else {
1579 u64 tsc_exp = kvm->arch.last_tsc_write +
1580 nsec_to_cycles(vcpu, elapsed);
1581 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1582 /*
1583 * Special case: TSC write with a small delta (1 second)
1584 * of virtual cycle time against real time is
1585 * interpreted as an attempt to synchronize the CPU.
1586 */
1587 synchronizing = data < tsc_exp + tsc_hz &&
1588 data + tsc_hz > tsc_exp;
1589 }
c5e8ec8e 1590 }
f38e098f
ZA		 1591
1592 /*
5d3cb0f6
ZA		 1593 * For a reliable TSC, we can match TSC offsets, and for an unstable
1594 * TSC, we add elapsed time in this computation. We could let the
1595 * compensation code attempt to catch up if we fall behind, but
1596 * it's better to try to match offsets from the beginning.
1597 */
c5e8ec8e 1598 if (synchronizing &&
5d3cb0f6 1599 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
b0c39dc6 1600 if (!kvm_check_tsc_unstable()) {
e26101b1 1601 offset = kvm->arch.cur_tsc_offset;
f38e098f
ZA		 1602 pr_debug("kvm: matched tsc offset for %llu\n", data);
1603 } else {
857e4099 1604 u64 delta = nsec_to_cycles(vcpu, elapsed);
5d3cb0f6 1605 data += delta;
07c1419a 1606 offset = kvm_compute_tsc_offset(vcpu, data);
759379dd 1607 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
f38e098f 1608 }
b48aa97e 1609 matched = true;
0d3da0d2 1610 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
e26101b1
ZA		 1611 } else {
1612 /*
1613 * We split periods of matched TSC writes into generations.
1614 * For each generation, we track the original measured
1615 * nanosecond time, offset, and write, so if TSCs are in
1616 * sync, we can match exact offset, and if not, we can match
4a969980 1617 * exact software computation in compute_guest_tsc()
e26101b1
ZA		 1618 *
1619 * These values are tracked in kvm->arch.cur_xxx variables.
1620 */
1621 kvm->arch.cur_tsc_generation++;
1622 kvm->arch.cur_tsc_nsec = ns;
1623 kvm->arch.cur_tsc_write = data;
1624 kvm->arch.cur_tsc_offset = offset;
b48aa97e 1625 matched = false;
0d3da0d2 1626 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
e26101b1 1627 kvm->arch.cur_tsc_generation, data);
f38e098f 1628 }
e26101b1
ZA		 1629
1630 /*
1631 * We also track th most recent recorded KHZ, write and time to
1632 * allow the matching interval to be extended at each write.
1633 */
f38e098f
ZA		 1634 kvm->arch.last_tsc_nsec = ns;
1635 kvm->arch.last_tsc_write = data;
5d3cb0f6 1636 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
99e3e30a 1637
b183aa58 1638 vcpu->arch.last_guest_tsc = data;
e26101b1
ZA		 1639
1640 /* Keep track of which generation this VCPU has synchronized to */
1641 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1642 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1643 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1644
d6321d49 1645 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
ba904635 1646 update_ia32_tsc_adjust_msr(vcpu, offset);
d6321d49 1647
a545ab6a 1648 kvm_vcpu_write_tsc_offset(vcpu, offset);
e26101b1 1649 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
b48aa97e
MT		 1650
1651 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
0d3da0d2 1652 if (!matched) {
b48aa97e 1653 kvm->arch.nr_vcpus_matched_tsc = 0;
0d3da0d2
TG		 1654 } else if (!already_matched) {
1655 kvm->arch.nr_vcpus_matched_tsc++;
1656 }
b48aa97e
MT		 1657
1658 kvm_track_tsc_matching(vcpu);
1659 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
99e3e30a 1660}
e26101b1 1661
99e3e30a
ZA		 1662EXPORT_SYMBOL_GPL(kvm_write_tsc);
1663
58ea6767
HZ		 1664static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1665 s64 adjustment)
1666{
ea26e4ec 1667 kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.tsc_offset + adjustment);
58ea6767
HZ		 1668}
1669
1670static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1671{
1672 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1673 WARN_ON(adjustment < 0);
1674 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
ea26e4ec 1675 adjust_tsc_offset_guest(vcpu, adjustment);
58ea6767
HZ		 1676}
1677
d828199e
MT		 1678#ifdef CONFIG_X86_64
1679
a5a1d1c2 1680static u64 read_tsc(void)
d828199e 1681{
a5a1d1c2 1682 u64 ret = (u64)rdtsc_ordered();
03b9730b 1683 u64 last = pvclock_gtod_data.clock.cycle_last;
d828199e
MT		 1684
1685 if (likely(ret >= last))
1686 return ret;
1687
1688 /*
1689 * GCC likes to generate cmov here, but this branch is extremely
6a6256f9 1690 * predictable (it's just a function of time and the likely is
d828199e
MT		 1691 * very likely) and there's a data dependence, so force GCC
1692 * to generate a branch instead. I don't barrier() because
1693 * we don't actually need a barrier, and if this function
1694 * ever gets inlined it will generate worse code.
1695 */
1696 asm volatile ("");
1697 return last;
1698}
1699
b0c39dc6 1700static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
d828199e
MT		 1701{
1702 long v;
1703 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
b0c39dc6
VK		 1704 u64 tsc_pg_val;
1705
1706 switch (gtod->clock.vclock_mode) {
1707 case VCLOCK_HVCLOCK:
1708 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
1709 tsc_timestamp);
1710 if (tsc_pg_val != U64_MAX) {
1711 /* TSC page valid */
1712 *mode = VCLOCK_HVCLOCK;
1713 v = (tsc_pg_val - gtod->clock.cycle_last) &
1714 gtod->clock.mask;
1715 } else {
1716 /* TSC page invalid */
1717 *mode = VCLOCK_NONE;
1718 }
1719 break;
1720 case VCLOCK_TSC:
1721 *mode = VCLOCK_TSC;
1722 *tsc_timestamp = read_tsc();
1723 v = (*tsc_timestamp - gtod->clock.cycle_last) &
1724 gtod->clock.mask;
1725 break;
1726 default:
1727 *mode = VCLOCK_NONE;
1728 }
d828199e 1729
b0c39dc6
VK		 1730 if (*mode == VCLOCK_NONE)
1731 *tsc_timestamp = v = 0;
d828199e 1732
d828199e
MT		 1733 return v * gtod->clock.mult;
1734}
1735
b0c39dc6 1736static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
d828199e 1737{
cbcf2dd3 1738 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
d828199e 1739 unsigned long seq;
d828199e 1740 int mode;
cbcf2dd3 1741 u64 ns;
d828199e 1742
d828199e
MT		 1743 do {
1744 seq = read_seqcount_begin(&gtod->seq);
cbcf2dd3 1745 ns = gtod->nsec_base;
b0c39dc6 1746 ns += vgettsc(tsc_timestamp, &mode);
d828199e 1747 ns >>= gtod->clock.shift;
cbcf2dd3 1748 ns += gtod->boot_ns;
d828199e 1749 } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
cbcf2dd3 1750 *t = ns;
d828199e
MT		 1751
1752 return mode;
1753}
1754
b0c39dc6 1755static int do_realtime(struct timespec *ts, u64 *tsc_timestamp)
55dd00a7
MT		 1756{
1757 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1758 unsigned long seq;
1759 int mode;
1760 u64 ns;
1761
1762 do {
1763 seq = read_seqcount_begin(&gtod->seq);
55dd00a7
MT		 1764 ts->tv_sec = gtod->wall_time_sec;
1765 ns = gtod->nsec_base;
b0c39dc6 1766 ns += vgettsc(tsc_timestamp, &mode);
55dd00a7
MT		 1767 ns >>= gtod->clock.shift;
1768 } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1769
1770 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1771 ts->tv_nsec = ns;
1772
1773 return mode;
1774}
1775
b0c39dc6
VK		 1776/* returns true if host is using TSC based clocksource */
1777static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
d828199e 1778{
d828199e 1779 /* checked again under seqlock below */
b0c39dc6 1780 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
d828199e
MT		 1781 return false;
1782
b0c39dc6
VK		 1783 return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
1784 tsc_timestamp));
d828199e 1785}
55dd00a7 1786
b0c39dc6 1787/* returns true if host is using TSC based clocksource */
55dd00a7 1788static bool kvm_get_walltime_and_clockread(struct timespec *ts,
b0c39dc6 1789 u64 *tsc_timestamp)
55dd00a7
MT		 1790{
1791 /* checked again under seqlock below */
b0c39dc6 1792 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
55dd00a7
MT		 1793 return false;
1794
b0c39dc6 1795 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
55dd00a7 1796}
d828199e
MT		 1797#endif
1798
1799/*
1800 *
b48aa97e
MT		 1801 * Assuming a stable TSC across physical CPUS, and a stable TSC
1802 * across virtual CPUs, the following condition is possible.
1803 * Each numbered line represents an event visible to both
d828199e
MT		 1804 * CPUs at the next numbered event.
1805 *
1806 * "timespecX" represents host monotonic time. "tscX" represents
1807 * RDTSC value.
1808 *
1809 * VCPU0 on CPU0 | VCPU1 on CPU1
1810 *
1811 * 1. read timespec0,tsc0
1812 * 2. | timespec1 = timespec0 + N
1813 * | tsc1 = tsc0 + M
1814 * 3. transition to guest | transition to guest
1815 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1816 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1817 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1818 *
1819 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1820 *
1821 * - ret0 < ret1
1822 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1823 * ...
1824 * - 0 < N - M => M < N
1825 *
1826 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1827 * always the case (the difference between two distinct xtime instances
1828 * might be smaller then the difference between corresponding TSC reads,
1829 * when updating guest vcpus pvclock areas).
1830 *
1831 * To avoid that problem, do not allow visibility of distinct
1832 * system_timestamp/tsc_timestamp values simultaneously: use a master
1833 * copy of host monotonic time values. Update that master copy
1834 * in lockstep.
1835 *
b48aa97e 1836 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
d828199e
MT		 1837 *
1838 */
1839
1840static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1841{
1842#ifdef CONFIG_X86_64
1843 struct kvm_arch *ka = &kvm->arch;
1844 int vclock_mode;
b48aa97e
MT		 1845 bool host_tsc_clocksource, vcpus_matched;
1846
1847 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1848 atomic_read(&kvm->online_vcpus));
d828199e
MT		 1849
1850 /*
1851 * If the host uses TSC clock, then passthrough TSC as stable
1852 * to the guest.
1853 */
b48aa97e 1854 host_tsc_clocksource = kvm_get_time_and_clockread(
d828199e
MT		 1855 &ka->master_kernel_ns,
1856 &ka->master_cycle_now);
1857
16a96021 1858 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
a826faf1 1859 && !ka->backwards_tsc_observed
54750f2c 1860 && !ka->boot_vcpu_runs_old_kvmclock;
b48aa97e 1861
d828199e
MT		 1862 if (ka->use_master_clock)
1863 atomic_set(&kvm_guest_has_master_clock, 1);
1864
1865 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
b48aa97e
MT		 1866 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1867 vcpus_matched);
d828199e
MT		 1868#endif
1869}
1870
2860c4b1
PB		 1871void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1872{
1873 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1874}
1875
2e762ff7
MT		 1876static void kvm_gen_update_masterclock(struct kvm *kvm)
1877{
1878#ifdef CONFIG_X86_64
1879 int i;
1880 struct kvm_vcpu *vcpu;
1881 struct kvm_arch *ka = &kvm->arch;
1882
1883 spin_lock(&ka->pvclock_gtod_sync_lock);
1884 kvm_make_mclock_inprogress_request(kvm);
1885 /* no guest entries from this point */
1886 pvclock_update_vm_gtod_copy(kvm);
1887
1888 kvm_for_each_vcpu(i, vcpu, kvm)
105b21bb 1889 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2e762ff7
MT		 1890
1891 /* guest entries allowed */
1892 kvm_for_each_vcpu(i, vcpu, kvm)
72875d8a 1893 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2e762ff7
MT		 1894
1895 spin_unlock(&ka->pvclock_gtod_sync_lock);
1896#endif
1897}
1898
e891a32e 1899u64 get_kvmclock_ns(struct kvm *kvm)
108b249c 1900{
108b249c 1901 struct kvm_arch *ka = &kvm->arch;
8b953440 1902 struct pvclock_vcpu_time_info hv_clock;
e2c2206a 1903 u64 ret;
108b249c 1904
8b953440
PB		 1905 spin_lock(&ka->pvclock_gtod_sync_lock);
1906 if (!ka->use_master_clock) {
1907 spin_unlock(&ka->pvclock_gtod_sync_lock);
1908 return ktime_get_boot_ns() + ka->kvmclock_offset;
108b249c
PB		 1909 }
1910
8b953440
PB		 1911 hv_clock.tsc_timestamp = ka->master_cycle_now;
1912 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
1913 spin_unlock(&ka->pvclock_gtod_sync_lock);
1914
e2c2206a
WL		 1915 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
1916 get_cpu();
1917
e70b57a6
WL		 1918 if (__this_cpu_read(cpu_tsc_khz)) {
1919 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
1920 &hv_clock.tsc_shift,
1921 &hv_clock.tsc_to_system_mul);
1922 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
1923 } else
1924 ret = ktime_get_boot_ns() + ka->kvmclock_offset;
e2c2206a
WL		 1925
1926 put_cpu();
1927
1928 return ret;
108b249c
PB		 1929}
1930
0d6dd2ff
PB		 1931static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
1932{
1933 struct kvm_vcpu_arch *vcpu = &v->arch;
1934 struct pvclock_vcpu_time_info guest_hv_clock;
1935
4e335d9e 1936 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
0d6dd2ff
PB		 1937 &guest_hv_clock, sizeof(guest_hv_clock))))
1938 return;
1939
1940 /* This VCPU is paused, but it's legal for a guest to read another
1941 * VCPU's kvmclock, so we really have to follow the specification where
1942 * it says that version is odd if data is being modified, and even after
1943 * it is consistent.
1944 *
1945 * Version field updates must be kept separate. This is because
1946 * kvm_write_guest_cached might use a "rep movs" instruction, and
1947 * writes within a string instruction are weakly ordered. So there
1948 * are three writes overall.
1949 *
1950 * As a small optimization, only write the version field in the first
1951 * and third write. The vcpu->pv_time cache is still valid, because the
1952 * version field is the first in the struct.
1953 */
1954 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1955
51c4b8bb
LA		 1956 if (guest_hv_clock.version & 1)
1957 ++guest_hv_clock.version; /* first time write, random junk */
1958
0d6dd2ff 1959 vcpu->hv_clock.version = guest_hv_clock.version + 1;
4e335d9e
PB		 1960 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1961 &vcpu->hv_clock,
1962 sizeof(vcpu->hv_clock.version));
0d6dd2ff
PB		 1963
1964 smp_wmb();
1965
1966 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1967 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1968
1969 if (vcpu->pvclock_set_guest_stopped_request) {
1970 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
1971 vcpu->pvclock_set_guest_stopped_request = false;
1972 }
1973
1974 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1975
4e335d9e
PB		 1976 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1977 &vcpu->hv_clock,
1978 sizeof(vcpu->hv_clock));
0d6dd2ff
PB		 1979
1980 smp_wmb();
1981
1982 vcpu->hv_clock.version++;
4e335d9e
PB		 1983 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1984 &vcpu->hv_clock,
1985 sizeof(vcpu->hv_clock.version));
0d6dd2ff
PB		 1986}
1987
34c238a1 1988static int kvm_guest_time_update(struct kvm_vcpu *v)
18068523 1989{
78db6a50 1990 unsigned long flags, tgt_tsc_khz;
18068523 1991 struct kvm_vcpu_arch *vcpu = &v->arch;
d828199e 1992 struct kvm_arch *ka = &v->kvm->arch;
f25e656d 1993 s64 kernel_ns;
d828199e 1994 u64 tsc_timestamp, host_tsc;
51d59c6b 1995 u8 pvclock_flags;
d828199e
MT		 1996 bool use_master_clock;
1997
1998 kernel_ns = 0;
1999 host_tsc = 0;
18068523 2000
d828199e
MT		 2001 /*
2002 * If the host uses TSC clock, then passthrough TSC as stable
2003 * to the guest.
2004 */
2005 spin_lock(&ka->pvclock_gtod_sync_lock);
2006 use_master_clock = ka->use_master_clock;
2007 if (use_master_clock) {
2008 host_tsc = ka->master_cycle_now;
2009 kernel_ns = ka->master_kernel_ns;
2010 }
2011 spin_unlock(&ka->pvclock_gtod_sync_lock);
c09664bb
MT		 2012
2013 /* Keep irq disabled to prevent changes to the clock */
2014 local_irq_save(flags);
78db6a50
PB		 2015 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2016 if (unlikely(tgt_tsc_khz == 0)) {
c09664bb
MT		 2017 local_irq_restore(flags);
2018 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2019 return 1;
2020 }
d828199e 2021 if (!use_master_clock) {
4ea1636b 2022 host_tsc = rdtsc();
108b249c 2023 kernel_ns = ktime_get_boot_ns();
d828199e
MT		 2024 }
2025
4ba76538 2026 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
d828199e 2027
c285545f
ZA		 2028 /*
2029 * We may have to catch up the TSC to match elapsed wall clock
2030 * time for two reasons, even if kvmclock is used.
2031 * 1) CPU could have been running below the maximum TSC rate
2032 * 2) Broken TSC compensation resets the base at each VCPU
2033 * entry to avoid unknown leaps of TSC even when running
2034 * again on the same CPU. This may cause apparent elapsed
2035 * time to disappear, and the guest to stand still or run
2036 * very slowly.
2037 */
2038 if (vcpu->tsc_catchup) {
2039 u64 tsc = compute_guest_tsc(v, kernel_ns);
2040 if (tsc > tsc_timestamp) {
f1e2b260 2041 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
c285545f
ZA		 2042 tsc_timestamp = tsc;
2043 }
50d0a0f9
GH		 2044 }
2045
18068523
GOC		 2046 local_irq_restore(flags);
2047
0d6dd2ff 2048 /* With all the info we got, fill in the values */
18068523 2049
78db6a50
PB		 2050 if (kvm_has_tsc_control)
2051 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2052
2053 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
3ae13faa 2054 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
5f4e3f88
ZA		 2055 &vcpu->hv_clock.tsc_shift,
2056 &vcpu->hv_clock.tsc_to_system_mul);
78db6a50 2057 vcpu->hw_tsc_khz = tgt_tsc_khz;
8cfdc000
ZA		 2058 }
2059
1d5f066e 2060 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
759379dd 2061 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
28e4639a 2062 vcpu->last_guest_tsc = tsc_timestamp;
51d59c6b 2063
d828199e 2064 /* If the host uses TSC clocksource, then it is stable */
0d6dd2ff 2065 pvclock_flags = 0;
d828199e
MT		 2066 if (use_master_clock)
2067 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2068
78c0337a
MT		 2069 vcpu->hv_clock.flags = pvclock_flags;
2070
095cf55d
PB		 2071 if (vcpu->pv_time_enabled)
2072 kvm_setup_pvclock_page(v);
2073 if (v == kvm_get_vcpu(v->kvm, 0))
2074 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
8cfdc000 2075 return 0;
c8076604
GH		 2076}
2077
0061d53d
MT		 2078/*
2079 * kvmclock updates which are isolated to a given vcpu, such as
2080 * vcpu->cpu migration, should not allow system_timestamp from
2081 * the rest of the vcpus to remain static. Otherwise ntp frequency
2082 * correction applies to one vcpu's system_timestamp but not
2083 * the others.
2084 *
2085 * So in those cases, request a kvmclock update for all vcpus.
7e44e449
AJ		 2086 * We need to rate-limit these requests though, as they can
2087 * considerably slow guests that have a large number of vcpus.
2088 * The time for a remote vcpu to update its kvmclock is bound
2089 * by the delay we use to rate-limit the updates.
0061d53d
MT		 2090 */
2091
7e44e449
AJ		 2092#define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2093
2094static void kvmclock_update_fn(struct work_struct *work)
0061d53d
MT		 2095{
2096 int i;
7e44e449
AJ		 2097 struct delayed_work *dwork = to_delayed_work(work);
2098 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2099 kvmclock_update_work);
2100 struct kvm *kvm = container_of(ka, struct kvm, arch);
0061d53d
MT		 2101 struct kvm_vcpu *vcpu;
2102
2103 kvm_for_each_vcpu(i, vcpu, kvm) {
105b21bb 2104 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
0061d53d
MT		 2105 kvm_vcpu_kick(vcpu);
2106 }
2107}
2108
7e44e449
AJ		 2109static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2110{
2111 struct kvm *kvm = v->kvm;
2112
105b21bb 2113 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
7e44e449
AJ		 2114 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2115 KVMCLOCK_UPDATE_DELAY);
2116}
2117
332967a3
AJ		 2118#define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2119
2120static void kvmclock_sync_fn(struct work_struct *work)
2121{
2122 struct delayed_work *dwork = to_delayed_work(work);
2123 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2124 kvmclock_sync_work);
2125 struct kvm *kvm = container_of(ka, struct kvm, arch);
2126
630994b3
MT		 2127 if (!kvmclock_periodic_sync)
2128 return;
2129
332967a3
AJ		 2130 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2131 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2132 KVMCLOCK_SYNC_PERIOD);
2133}
2134
9ffd986c 2135static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
15c4a640 2136{
890ca9ae
HY		 2137 u64 mcg_cap = vcpu->arch.mcg_cap;
2138 unsigned bank_num = mcg_cap & 0xff;
9ffd986c
WL		 2139 u32 msr = msr_info->index;
2140 u64 data = msr_info->data;
890ca9ae 2141
15c4a640 2142 switch (msr) {
15c4a640 2143 case MSR_IA32_MCG_STATUS:
890ca9ae 2144 vcpu->arch.mcg_status = data;
15c4a640 2145 break;
c7ac679c 2146 case MSR_IA32_MCG_CTL:
890ca9ae
HY		 2147 if (!(mcg_cap & MCG_CTL_P))
2148 return 1;
2149 if (data != 0 && data != ~(u64)0)
2150 return -1;
2151 vcpu->arch.mcg_ctl = data;
2152 break;
2153 default:
2154 if (msr >= MSR_IA32_MC0_CTL &&
81760dcc 2155 msr < MSR_IA32_MCx_CTL(bank_num)) {
890ca9ae 2156 u32 offset = msr - MSR_IA32_MC0_CTL;
114be429
AP		 2157 /* only 0 or all 1s can be written to IA32_MCi_CTL
2158 * some Linux kernels though clear bit 10 in bank 4 to
2159 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2160 * this to avoid an uncatched #GP in the guest
2161 */
890ca9ae 2162 if ((offset & 0x3) == 0 &&
114be429 2163 data != 0 && (data | (1 << 10)) != ~(u64)0)
890ca9ae 2164 return -1;
9ffd986c
WL		 2165 if (!msr_info->host_initiated &&
2166 (offset & 0x3) == 1 && data != 0)
2167 return -1;
890ca9ae
HY		 2168 vcpu->arch.mce_banks[offset] = data;
2169 break;
2170 }
2171 return 1;
2172 }
2173 return 0;
2174}
2175
ffde22ac
ES		 2176static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2177{
2178 struct kvm *kvm = vcpu->kvm;
2179 int lm = is_long_mode(vcpu);
2180 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2181 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2182 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2183 : kvm->arch.xen_hvm_config.blob_size_32;
2184 u32 page_num = data & ~PAGE_MASK;
2185 u64 page_addr = data & PAGE_MASK;
2186 u8 *page;
2187 int r;
2188
2189 r = -E2BIG;
2190 if (page_num >= blob_size)
2191 goto out;
2192 r = -ENOMEM;
ff5c2c03
SL		 2193 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2194 if (IS_ERR(page)) {
2195 r = PTR_ERR(page);
ffde22ac 2196 goto out;
ff5c2c03 2197 }
54bf36aa 2198 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
ffde22ac
ES		 2199 goto out_free;
2200 r = 0;
2201out_free:
2202 kfree(page);
2203out:
2204 return r;
2205}
2206
344d9588
GN		 2207static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2208{
2209 gpa_t gpa = data & ~0x3f;
2210
52a5c155
WL		 2211 /* Bits 3:5 are reserved, Should be zero */
2212 if (data & 0x38)
344d9588
GN		 2213 return 1;
2214
2215 vcpu->arch.apf.msr_val = data;
2216
2217 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2218 kvm_clear_async_pf_completion_queue(vcpu);
2219 kvm_async_pf_hash_reset(vcpu);
2220 return 0;
2221 }
2222
4e335d9e 2223 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
8f964525 2224 sizeof(u32)))
344d9588
GN		 2225 return 1;
2226
6adba527 2227 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
52a5c155 2228 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
344d9588
GN		 2229 kvm_async_pf_wakeup_all(vcpu);
2230 return 0;
2231}
2232
12f9a48f
GC		 2233static void kvmclock_reset(struct kvm_vcpu *vcpu)
2234{
0b79459b 2235 vcpu->arch.pv_time_enabled = false;
12f9a48f
GC		 2236}
2237
f38a7b75
WL		 2238static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2239{
2240 ++vcpu->stat.tlb_flush;
2241 kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2242}
2243
c9aaa895
GC		 2244static void record_steal_time(struct kvm_vcpu *vcpu)
2245{
2246 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2247 return;
2248
4e335d9e 2249 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
c9aaa895
GC		 2250 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2251 return;
2252
f38a7b75
WL		 2253 /*
2254 * Doing a TLB flush here, on the guest's behalf, can avoid
2255 * expensive IPIs.
2256 */
2257 if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2258 kvm_vcpu_flush_tlb(vcpu, false);
0b9f6c46 2259
35f3fae1
WL		 2260 if (vcpu->arch.st.steal.version & 1)
2261 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2262
2263 vcpu->arch.st.steal.version += 1;
2264
4e335d9e 2265 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
35f3fae1
WL		 2266 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2267
2268 smp_wmb();
2269
c54cdf14
LC		 2270 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2271 vcpu->arch.st.last_steal;
2272 vcpu->arch.st.last_steal = current->sched_info.run_delay;
35f3fae1 2273
4e335d9e 2274 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
35f3fae1
WL		 2275 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2276
2277 smp_wmb();
2278
2279 vcpu->arch.st.steal.version += 1;
c9aaa895 2280
4e335d9e 2281 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
c9aaa895
GC		 2282 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2283}
2284
8fe8ab46 2285int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
15c4a640 2286{
5753785f 2287 bool pr = false;
8fe8ab46
WA		 2288 u32 msr = msr_info->index;
2289 u64 data = msr_info->data;
5753785f 2290
15c4a640 2291 switch (msr) {
2e32b719 2292 case MSR_AMD64_NB_CFG:
2e32b719
BP		 2293 case MSR_IA32_UCODE_WRITE:
2294 case MSR_VM_HSAVE_PA:
2295 case MSR_AMD64_PATCH_LOADER:
2296 case MSR_AMD64_BU_CFG2:
405a353a 2297 case MSR_AMD64_DC_CFG:
2e32b719
BP		 2298 break;
2299
518e7b94
WL		 2300 case MSR_IA32_UCODE_REV:
2301 if (msr_info->host_initiated)
2302 vcpu->arch.microcode_version = data;
2303 break;
15c4a640 2304 case MSR_EFER:
b69e8cae 2305 return set_efer(vcpu, data);
8f1589d9
AP		 2306 case MSR_K7_HWCR:
2307 data &= ~(u64)0x40; /* ignore flush filter disable */
82494028 2308 data &= ~(u64)0x100; /* ignore ignne emulation enable */
a223c313 2309 data &= ~(u64)0x8; /* ignore TLB cache disable */
22d48b2d 2310 data &= ~(u64)0x40000; /* ignore Mc status write enable */
8f1589d9 2311 if (data != 0) {
a737f256
CD		 2312 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2313 data);
8f1589d9
AP		 2314 return 1;
2315 }
15c4a640 2316 break;
f7c6d140
AP		 2317 case MSR_FAM10H_MMIO_CONF_BASE:
2318 if (data != 0) {
a737f256
CD		 2319 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2320 "0x%llx\n", data);
f7c6d140
AP		 2321 return 1;
2322 }
15c4a640 2323 break;
b5e2fec0
AG		 2324 case MSR_IA32_DEBUGCTLMSR:
2325 if (!data) {
2326 /* We support the non-activated case already */
2327 break;
2328 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2329 /* Values other than LBR and BTF are vendor-specific,
2330 thus reserved and should throw a #GP */
2331 return 1;
2332 }
a737f256
CD		 2333 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2334 __func__, data);
b5e2fec0 2335 break;
9ba075a6 2336 case 0x200 ... 0x2ff:
ff53604b 2337 return kvm_mtrr_set_msr(vcpu, msr, data);
15c4a640 2338 case MSR_IA32_APICBASE:
58cb628d 2339 return kvm_set_apic_base(vcpu, msr_info);
0105d1a5
GN		 2340 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2341 return kvm_x2apic_msr_write(vcpu, msr, data);
a3e06bbe
LJ		 2342 case MSR_IA32_TSCDEADLINE:
2343 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2344 break;
ba904635 2345 case MSR_IA32_TSC_ADJUST:
d6321d49 2346 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
ba904635 2347 if (!msr_info->host_initiated) {
d913b904 2348 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
d7add054 2349 adjust_tsc_offset_guest(vcpu, adj);
ba904635
WA		 2350 }
2351 vcpu->arch.ia32_tsc_adjust_msr = data;
2352 }
2353 break;
15c4a640 2354 case MSR_IA32_MISC_ENABLE:
ad312c7c 2355 vcpu->arch.ia32_misc_enable_msr = data;
15c4a640 2356 break;
64d60670
PB		 2357 case MSR_IA32_SMBASE:
2358 if (!msr_info->host_initiated)
2359 return 1;
2360 vcpu->arch.smbase = data;
2361 break;
52797bf9
LA		 2362 case MSR_SMI_COUNT:
2363 if (!msr_info->host_initiated)
2364 return 1;
2365 vcpu->arch.smi_count = data;
2366 break;
11c6bffa 2367 case MSR_KVM_WALL_CLOCK_NEW:
18068523
GOC		 2368 case MSR_KVM_WALL_CLOCK:
2369 vcpu->kvm->arch.wall_clock = data;
2370 kvm_write_wall_clock(vcpu->kvm, data);
2371 break;
11c6bffa 2372 case MSR_KVM_SYSTEM_TIME_NEW:
18068523 2373 case MSR_KVM_SYSTEM_TIME: {
54750f2c
MT		 2374 struct kvm_arch *ka = &vcpu->kvm->arch;
2375
12f9a48f 2376 kvmclock_reset(vcpu);
18068523 2377
54750f2c
MT		 2378 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2379 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2380
2381 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
1bd2009e 2382 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
54750f2c
MT		 2383
2384 ka->boot_vcpu_runs_old_kvmclock = tmp;
2385 }
2386
18068523 2387 vcpu->arch.time = data;
0061d53d 2388 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
18068523
GOC		 2389
2390 /* we verify if the enable bit is set... */
2391 if (!(data & 1))
2392 break;
2393
4e335d9e 2394 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
8f964525
AH		 2395 &vcpu->arch.pv_time, data & ~1ULL,
2396 sizeof(struct pvclock_vcpu_time_info)))
0b79459b
AH		 2397 vcpu->arch.pv_time_enabled = false;
2398 else
2399 vcpu->arch.pv_time_enabled = true;
32cad84f 2400
18068523
GOC		 2401 break;
2402 }
344d9588
GN		 2403 case MSR_KVM_ASYNC_PF_EN:
2404 if (kvm_pv_enable_async_pf(vcpu, data))
2405 return 1;
2406 break;
c9aaa895
GC		 2407 case MSR_KVM_STEAL_TIME:
2408
2409 if (unlikely(!sched_info_on()))
2410 return 1;
2411
2412 if (data & KVM_STEAL_RESERVED_MASK)
2413 return 1;
2414
4e335d9e 2415 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
8f964525
AH		 2416 data & KVM_STEAL_VALID_BITS,
2417 sizeof(struct kvm_steal_time)))
c9aaa895
GC		 2418 return 1;
2419
2420 vcpu->arch.st.msr_val = data;
2421
2422 if (!(data & KVM_MSR_ENABLED))
2423 break;
2424
c9aaa895
GC		 2425 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2426
2427 break;
ae7a2a3f
MT		 2428 case MSR_KVM_PV_EOI_EN:
2429 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2430 return 1;
2431 break;
c9aaa895 2432
890ca9ae
HY		 2433 case MSR_IA32_MCG_CTL:
2434 case MSR_IA32_MCG_STATUS:
81760dcc 2435 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
9ffd986c 2436 return set_msr_mce(vcpu, msr_info);
71db6023 2437
6912ac32
WH		 2438 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2439 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2440 pr = true; /* fall through */
2441 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2442 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
c6702c9d 2443 if (kvm_pmu_is_valid_msr(vcpu, msr))
afd80d85 2444 return kvm_pmu_set_msr(vcpu, msr_info);
5753785f
GN		 2445
2446 if (pr || data != 0)
a737f256
CD		 2447 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2448 "0x%x data 0x%llx\n", msr, data);
5753785f 2449 break;
84e0cefa
JS		 2450 case MSR_K7_CLK_CTL:
2451 /*
2452 * Ignore all writes to this no longer documented MSR.
2453 * Writes are only relevant for old K7 processors,
2454 * all pre-dating SVM, but a recommended workaround from
4a969980 2455 * AMD for these chips. It is possible to specify the
84e0cefa
JS		 2456 * affected processor models on the command line, hence
2457 * the need to ignore the workaround.
2458 */
2459 break;
55cd8e5a 2460 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
e7d9513b
AS		 2461 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2462 case HV_X64_MSR_CRASH_CTL:
1f4b34f8 2463 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
a2e164e7
VK		 2464 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2465 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2466 case HV_X64_MSR_TSC_EMULATION_STATUS:
e7d9513b
AS		 2467 return kvm_hv_set_msr_common(vcpu, msr, data,
2468 msr_info->host_initiated);
91c9c3ed 2469 case MSR_IA32_BBL_CR_CTL3:
2470 /* Drop writes to this legacy MSR -- see rdmsr
2471 * counterpart for further detail.
2472 */
fab0aa3b
EM		 2473 if (report_ignored_msrs)
2474 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2475 msr, data);
91c9c3ed 2476 break;
2b036c6b 2477 case MSR_AMD64_OSVW_ID_LENGTH:
d6321d49 2478 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2b036c6b
BO		 2479 return 1;
2480 vcpu->arch.osvw.length = data;
2481 break;
2482 case MSR_AMD64_OSVW_STATUS:
d6321d49 2483 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2b036c6b
BO		 2484 return 1;
2485 vcpu->arch.osvw.status = data;
2486 break;
db2336a8
KH		 2487 case MSR_PLATFORM_INFO:
2488 if (!msr_info->host_initiated ||
2489 data & ~MSR_PLATFORM_INFO_CPUID_FAULT ||
2490 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2491 cpuid_fault_enabled(vcpu)))
2492 return 1;
2493 vcpu->arch.msr_platform_info = data;
2494 break;
2495 case MSR_MISC_FEATURES_ENABLES:
2496 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2497 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2498 !supports_cpuid_fault(vcpu)))
2499 return 1;
2500 vcpu->arch.msr_misc_features_enables = data;
2501 break;
15c4a640 2502 default:
ffde22ac
ES		 2503 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2504 return xen_hvm_config(vcpu, data);
c6702c9d 2505 if (kvm_pmu_is_valid_msr(vcpu, msr))
afd80d85 2506 return kvm_pmu_set_msr(vcpu, msr_info);
ed85c068 2507 if (!ignore_msrs) {
ae0f5499 2508 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
a737f256 2509 msr, data);
ed85c068
AP		 2510 return 1;
2511 } else {
fab0aa3b
EM		 2512 if (report_ignored_msrs)
2513 vcpu_unimpl(vcpu,
2514 "ignored wrmsr: 0x%x data 0x%llx\n",
2515 msr, data);
ed85c068
AP		 2516 break;
2517 }
15c4a640
CO		 2518 }
2519 return 0;
2520}
2521EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2522
2523
2524/*
2525 * Reads an msr value (of 'msr_index') into 'pdata'.
2526 * Returns 0 on success, non-0 otherwise.
2527 * Assumes vcpu_load() was already called.
2528 */
609e36d3 2529int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
15c4a640 2530{
609e36d3 2531 return kvm_x86_ops->get_msr(vcpu, msr);
15c4a640 2532}
ff651cb6 2533EXPORT_SYMBOL_GPL(kvm_get_msr);
15c4a640 2534
890ca9ae 2535static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
15c4a640
CO		 2536{
2537 u64 data;
890ca9ae
HY		 2538 u64 mcg_cap = vcpu->arch.mcg_cap;
2539 unsigned bank_num = mcg_cap & 0xff;
15c4a640
CO		 2540
2541 switch (msr) {
15c4a640
CO		 2542 case MSR_IA32_P5_MC_ADDR:
2543 case MSR_IA32_P5_MC_TYPE:
890ca9ae
HY		 2544 data = 0;
2545 break;
15c4a640 2546 case MSR_IA32_MCG_CAP:
890ca9ae
HY		 2547 data = vcpu->arch.mcg_cap;
2548 break;
c7ac679c 2549 case MSR_IA32_MCG_CTL:
890ca9ae
HY		 2550 if (!(mcg_cap & MCG_CTL_P))
2551 return 1;
2552 data = vcpu->arch.mcg_ctl;
2553 break;
2554 case MSR_IA32_MCG_STATUS:
2555 data = vcpu->arch.mcg_status;
2556 break;
2557 default:
2558 if (msr >= MSR_IA32_MC0_CTL &&
81760dcc 2559 msr < MSR_IA32_MCx_CTL(bank_num)) {
890ca9ae
HY		 2560 u32 offset = msr - MSR_IA32_MC0_CTL;
2561 data = vcpu->arch.mce_banks[offset];
2562 break;
2563 }
2564 return 1;
2565 }
2566 *pdata = data;
2567 return 0;
2568}
2569
609e36d3 2570int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
890ca9ae 2571{
609e36d3 2572 switch (msr_info->index) {
890ca9ae 2573 case MSR_IA32_PLATFORM_ID:
15c4a640 2574 case MSR_IA32_EBL_CR_POWERON:
b5e2fec0
AG		 2575 case MSR_IA32_DEBUGCTLMSR:
2576 case MSR_IA32_LASTBRANCHFROMIP:
2577 case MSR_IA32_LASTBRANCHTOIP:
2578 case MSR_IA32_LASTINTFROMIP:
2579 case MSR_IA32_LASTINTTOIP:
60af2ecd 2580 case MSR_K8_SYSCFG:
3afb1121
PB		 2581 case MSR_K8_TSEG_ADDR:
2582 case MSR_K8_TSEG_MASK:
60af2ecd 2583 case MSR_K7_HWCR:
61a6bd67 2584 case MSR_VM_HSAVE_PA:
1fdbd48c 2585 case MSR_K8_INT_PENDING_MSG:
c323c0e5 2586 case MSR_AMD64_NB_CFG:
f7c6d140 2587 case MSR_FAM10H_MMIO_CONF_BASE:
2e32b719 2588 case MSR_AMD64_BU_CFG2:
0c2df2a1 2589 case MSR_IA32_PERF_CTL:
405a353a 2590 case MSR_AMD64_DC_CFG:
609e36d3 2591 msr_info->data = 0;
15c4a640 2592 break;
c51eb52b 2593 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
6912ac32
WH		 2594 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2595 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2596 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2597 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
c6702c9d 2598 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
609e36d3
PB		 2599 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2600 msr_info->data = 0;
5753785f 2601 break;
742bc670 2602 case MSR_IA32_UCODE_REV:
518e7b94 2603 msr_info->data = vcpu->arch.microcode_version;
742bc670 2604 break;
9ba075a6 2605 case MSR_MTRRcap:
9ba075a6 2606 case 0x200 ... 0x2ff:
ff53604b 2607 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
15c4a640 2608 case 0xcd: /* fsb frequency */
609e36d3 2609 msr_info->data = 3;
15c4a640 2610 break;
7b914098
JS		 2611 /*
2612 * MSR_EBC_FREQUENCY_ID
2613 * Conservative value valid for even the basic CPU models.
2614 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2615 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2616 * and 266MHz for model 3, or 4. Set Core Clock
2617 * Frequency to System Bus Frequency Ratio to 1 (bits
2618 * 31:24) even though these are only valid for CPU
2619 * models > 2, however guests may end up dividing or
2620 * multiplying by zero otherwise.
2621 */
2622 case MSR_EBC_FREQUENCY_ID:
609e36d3 2623 msr_info->data = 1 << 24;
7b914098 2624 break;
15c4a640 2625 case MSR_IA32_APICBASE:
609e36d3 2626 msr_info->data = kvm_get_apic_base(vcpu);
15c4a640 2627 break;
0105d1a5 2628 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
609e36d3 2629 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
0105d1a5 2630 break;
a3e06bbe 2631 case MSR_IA32_TSCDEADLINE:
609e36d3 2632 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
a3e06bbe 2633 break;
ba904635 2634 case MSR_IA32_TSC_ADJUST:
609e36d3 2635 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
ba904635 2636 break;
15c4a640 2637 case MSR_IA32_MISC_ENABLE:
609e36d3 2638 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
15c4a640 2639 break;
64d60670
PB		 2640 case MSR_IA32_SMBASE:
2641 if (!msr_info->host_initiated)
2642 return 1;
2643 msr_info->data = vcpu->arch.smbase;
15c4a640 2644 break;
52797bf9
LA		 2645 case MSR_SMI_COUNT:
2646 msr_info->data = vcpu->arch.smi_count;
2647 break;
847f0ad8
AG		 2648 case MSR_IA32_PERF_STATUS:
2649 /* TSC increment by tick */
609e36d3 2650 msr_info->data = 1000ULL;
847f0ad8 2651 /* CPU multiplier */
b0996ae4 2652 msr_info->data |= (((uint64_t)4ULL) << 40);
847f0ad8 2653 break;
15c4a640 2654 case MSR_EFER:
609e36d3 2655 msr_info->data = vcpu->arch.efer;
15c4a640 2656 break;
18068523 2657 case MSR_KVM_WALL_CLOCK:
11c6bffa 2658 case MSR_KVM_WALL_CLOCK_NEW:
609e36d3 2659 msr_info->data = vcpu->kvm->arch.wall_clock;
18068523
GOC		 2660 break;
2661 case MSR_KVM_SYSTEM_TIME:
11c6bffa 2662 case MSR_KVM_SYSTEM_TIME_NEW:
609e36d3 2663 msr_info->data = vcpu->arch.time;
18068523 2664 break;
344d9588 2665 case MSR_KVM_ASYNC_PF_EN:
609e36d3 2666 msr_info->data = vcpu->arch.apf.msr_val;
344d9588 2667 break;
c9aaa895 2668 case MSR_KVM_STEAL_TIME:
609e36d3 2669 msr_info->data = vcpu->arch.st.msr_val;
c9aaa895 2670 break;
1d92128f 2671 case MSR_KVM_PV_EOI_EN:
609e36d3 2672 msr_info->data = vcpu->arch.pv_eoi.msr_val;
1d92128f 2673 break;
890ca9ae
HY		 2674 case MSR_IA32_P5_MC_ADDR:
2675 case MSR_IA32_P5_MC_TYPE:
2676 case MSR_IA32_MCG_CAP:
2677 case MSR_IA32_MCG_CTL:
2678 case MSR_IA32_MCG_STATUS:
81760dcc 2679 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
609e36d3 2680 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
84e0cefa
JS		 2681 case MSR_K7_CLK_CTL:
2682 /*
2683 * Provide expected ramp-up count for K7. All other
2684 * are set to zero, indicating minimum divisors for
2685 * every field.
2686 *
2687 * This prevents guest kernels on AMD host with CPU
2688 * type 6, model 8 and higher from exploding due to
2689 * the rdmsr failing.
2690 */
609e36d3 2691 msr_info->data = 0x20000000;
84e0cefa 2692 break;
55cd8e5a 2693 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
e7d9513b
AS		 2694 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2695 case HV_X64_MSR_CRASH_CTL:
1f4b34f8 2696 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
a2e164e7
VK		 2697 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2698 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2699 case HV_X64_MSR_TSC_EMULATION_STATUS:
e83d5887
AS		 2700 return kvm_hv_get_msr_common(vcpu,
2701 msr_info->index, &msr_info->data);
55cd8e5a 2702 break;
91c9c3ed 2703 case MSR_IA32_BBL_CR_CTL3:
2704 /* This legacy MSR exists but isn't fully documented in current
2705 * silicon. It is however accessed by winxp in very narrow
2706 * scenarios where it sets bit #19, itself documented as
2707 * a "reserved" bit. Best effort attempt to source coherent
2708 * read data here should the balance of the register be
2709 * interpreted by the guest:
2710 *
2711 * L2 cache control register 3: 64GB range, 256KB size,
2712 * enabled, latency 0x1, configured
2713 */
609e36d3 2714 msr_info->data = 0xbe702111;
91c9c3ed 2715 break;
2b036c6b 2716 case MSR_AMD64_OSVW_ID_LENGTH:
d6321d49 2717 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2b036c6b 2718 return 1;
609e36d3 2719 msr_info->data = vcpu->arch.osvw.length;
2b036c6b
BO		 2720 break;
2721 case MSR_AMD64_OSVW_STATUS:
d6321d49 2722 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2b036c6b 2723 return 1;
609e36d3 2724 msr_info->data = vcpu->arch.osvw.status;
2b036c6b 2725 break;
db2336a8
KH		 2726 case MSR_PLATFORM_INFO:
2727 msr_info->data = vcpu->arch.msr_platform_info;
2728 break;
2729 case MSR_MISC_FEATURES_ENABLES:
2730 msr_info->data = vcpu->arch.msr_misc_features_enables;
2731 break;
15c4a640 2732 default:
c6702c9d 2733 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
609e36d3 2734 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
ed85c068 2735 if (!ignore_msrs) {
ae0f5499
BD		 2736 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2737 msr_info->index);
ed85c068
AP		 2738 return 1;
2739 } else {
fab0aa3b
EM		 2740 if (report_ignored_msrs)
2741 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
2742 msr_info->index);
609e36d3 2743 msr_info->data = 0;
ed85c068
AP		 2744 }
2745 break;
15c4a640 2746 }
15c4a640
CO		 2747 return 0;
2748}
2749EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2750
313a3dc7
CO		 2751/*
2752 * Read or write a bunch of msrs. All parameters are kernel addresses.
2753 *
2754 * @return number of msrs set successfully.
2755 */
2756static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2757 struct kvm_msr_entry *entries,
2758 int (*do_msr)(struct kvm_vcpu *vcpu,
2759 unsigned index, u64 *data))
2760{
801e459a 2761 int i;
313a3dc7 2762
313a3dc7
CO		 2763 for (i = 0; i < msrs->nmsrs; ++i)
2764 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2765 break;
2766
313a3dc7
CO		 2767 return i;
2768}
2769
2770/*
2771 * Read or write a bunch of msrs. Parameters are user addresses.
2772 *
2773 * @return number of msrs set successfully.
2774 */
2775static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2776 int (*do_msr)(struct kvm_vcpu *vcpu,
2777 unsigned index, u64 *data),
2778 int writeback)
2779{
2780 struct kvm_msrs msrs;
2781 struct kvm_msr_entry *entries;
2782 int r, n;
2783 unsigned size;
2784
2785 r = -EFAULT;
2786 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2787 goto out;
2788
2789 r = -E2BIG;
2790 if (msrs.nmsrs >= MAX_IO_MSRS)
2791 goto out;
2792
313a3dc7 2793 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
ff5c2c03
SL		 2794 entries = memdup_user(user_msrs->entries, size);
2795 if (IS_ERR(entries)) {
2796 r = PTR_ERR(entries);
313a3dc7 2797 goto out;
ff5c2c03 2798 }
313a3dc7
CO		 2799
2800 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2801 if (r < 0)
2802 goto out_free;
2803
2804 r = -EFAULT;
2805 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2806 goto out_free;
2807
2808 r = n;
2809
2810out_free:
7a73c028 2811 kfree(entries);
313a3dc7
CO		 2812out:
2813 return r;
2814}
2815
4d5422ce
WL		 2816static inline bool kvm_can_mwait_in_guest(void)
2817{
2818 return boot_cpu_has(X86_FEATURE_MWAIT) &&
2819 !boot_cpu_has_bug(X86_BUG_MONITOR);
2820}
2821
784aa3d7 2822int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
018d00d2 2823{
4d5422ce 2824 int r = 0;
018d00d2
ZX		 2825
2826 switch (ext) {
2827 case KVM_CAP_IRQCHIP:
2828 case KVM_CAP_HLT:
2829 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
018d00d2 2830 case KVM_CAP_SET_TSS_ADDR:
07716717 2831 case KVM_CAP_EXT_CPUID:
9c15bb1d 2832 case KVM_CAP_EXT_EMUL_CPUID:
c8076604 2833 case KVM_CAP_CLOCKSOURCE:
7837699f 2834 case KVM_CAP_PIT:
a28e4f5a 2835 case KVM_CAP_NOP_IO_DELAY:
62d9f0db 2836 case KVM_CAP_MP_STATE:
ed848624 2837 case KVM_CAP_SYNC_MMU:
a355c85c 2838 case KVM_CAP_USER_NMI:
52d939a0 2839 case KVM_CAP_REINJECT_CONTROL:
4925663a 2840 case KVM_CAP_IRQ_INJECT_STATUS:
d34e6b17 2841 case KVM_CAP_IOEVENTFD:
f848a5a8 2842 case KVM_CAP_IOEVENTFD_NO_LENGTH:
c5ff41ce 2843 case KVM_CAP_PIT2:
e9f42757 2844 case KVM_CAP_PIT_STATE2:
b927a3ce 2845 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
ffde22ac 2846 case KVM_CAP_XEN_HVM:
3cfc3092 2847 case KVM_CAP_VCPU_EVENTS:
55cd8e5a 2848 case KVM_CAP_HYPERV:
10388a07 2849 case KVM_CAP_HYPERV_VAPIC:
c25bc163 2850 case KVM_CAP_HYPERV_SPIN:
5c919412 2851 case KVM_CAP_HYPERV_SYNIC:
efc479e6 2852 case KVM_CAP_HYPERV_SYNIC2:
d3457c87 2853 case KVM_CAP_HYPERV_VP_INDEX:
faeb7833 2854 case KVM_CAP_HYPERV_EVENTFD:
ab9f4ecb 2855 case KVM_CAP_PCI_SEGMENT:
a1efbe77 2856 case KVM_CAP_DEBUGREGS:
d2be1651 2857 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2d5b5a66 2858 case KVM_CAP_XSAVE:
344d9588 2859 case KVM_CAP_ASYNC_PF:
92a1f12d 2860 case KVM_CAP_GET_TSC_KHZ:
1c0b28c2 2861 case KVM_CAP_KVMCLOCK_CTRL:
4d8b81ab 2862 case KVM_CAP_READONLY_MEM:
5f66b620 2863 case KVM_CAP_HYPERV_TIME:
100943c5 2864 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
defcf51f 2865 case KVM_CAP_TSC_DEADLINE_TIMER:
90de4a18
NA		 2866 case KVM_CAP_ENABLE_CAP_VM:
2867 case KVM_CAP_DISABLE_QUIRKS:
d71ba788 2868 case KVM_CAP_SET_BOOT_CPU_ID:
49df6397 2869 case KVM_CAP_SPLIT_IRQCHIP:
460df4c1 2870 case KVM_CAP_IMMEDIATE_EXIT:
801e459a 2871 case KVM_CAP_GET_MSR_FEATURES:
018d00d2
ZX		 2872 r = 1;
2873 break;
01643c51
KH		 2874 case KVM_CAP_SYNC_REGS:
2875 r = KVM_SYNC_X86_VALID_FIELDS;
2876 break;
e3fd9a93
PB		 2877 case KVM_CAP_ADJUST_CLOCK:
2878 r = KVM_CLOCK_TSC_STABLE;
2879 break;
4d5422ce 2880 case KVM_CAP_X86_DISABLE_EXITS:
b31c114b 2881 r |= KVM_X86_DISABLE_EXITS_HTL | KVM_X86_DISABLE_EXITS_PAUSE;
4d5422ce
WL		 2882 if(kvm_can_mwait_in_guest())
2883 r |= KVM_X86_DISABLE_EXITS_MWAIT;
668fffa3 2884 break;
6d396b55
PB		 2885 case KVM_CAP_X86_SMM:
2886 /* SMBASE is usually relocated above 1M on modern chipsets,
2887 * and SMM handlers might indeed rely on 4G segment limits,
2888 * so do not report SMM to be available if real mode is
2889 * emulated via vm86 mode. Still, do not go to great lengths
2890 * to avoid userspace's usage of the feature, because it is a
2891 * fringe case that is not enabled except via specific settings
2892 * of the module parameters.
2893 */
2894 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2895 break;
774ead3a
AK		 2896 case KVM_CAP_VAPIC:
2897 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2898 break;
f725230a 2899 case KVM_CAP_NR_VCPUS:
8c3ba334
SL		 2900 r = KVM_SOFT_MAX_VCPUS;
2901 break;
2902 case KVM_CAP_MAX_VCPUS:
f725230a
AK		 2903 r = KVM_MAX_VCPUS;
2904 break;
a988b910 2905 case KVM_CAP_NR_MEMSLOTS:
bbacc0c1 2906 r = KVM_USER_MEM_SLOTS;
a988b910 2907 break;
a68a6a72
MT		 2908 case KVM_CAP_PV_MMU: /* obsolete */
2909 r = 0;
2f333bcb 2910 break;
890ca9ae
HY		 2911 case KVM_CAP_MCE:
2912 r = KVM_MAX_MCE_BANKS;
2913 break;
2d5b5a66 2914 case KVM_CAP_XCRS:
d366bf7e 2915 r = boot_cpu_has(X86_FEATURE_XSAVE);
2d5b5a66 2916 break;
92a1f12d
JR		 2917 case KVM_CAP_TSC_CONTROL:
2918 r = kvm_has_tsc_control;
2919 break;
37131313
RK		 2920 case KVM_CAP_X2APIC_API:
2921 r = KVM_X2APIC_API_VALID_FLAGS;
2922 break;
018d00d2 2923 default:
018d00d2
ZX		 2924 break;
2925 }
2926 return r;
2927
2928}
2929
043405e1
CO		 2930long kvm_arch_dev_ioctl(struct file *filp,
2931 unsigned int ioctl, unsigned long arg)
2932{
2933 void __user *argp = (void __user *)arg;
2934 long r;
2935
2936 switch (ioctl) {
2937 case KVM_GET_MSR_INDEX_LIST: {
2938 struct kvm_msr_list __user *user_msr_list = argp;
2939 struct kvm_msr_list msr_list;
2940 unsigned n;
2941
2942 r = -EFAULT;
2943 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2944 goto out;
2945 n = msr_list.nmsrs;
62ef68bb 2946 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
043405e1
CO		 2947 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2948 goto out;
2949 r = -E2BIG;
e125e7b6 2950 if (n < msr_list.nmsrs)
043405e1
CO		 2951 goto out;
2952 r = -EFAULT;
2953 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2954 num_msrs_to_save * sizeof(u32)))
2955 goto out;
e125e7b6 2956 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
043405e1 2957 &emulated_msrs,
62ef68bb 2958 num_emulated_msrs * sizeof(u32)))
043405e1
CO		 2959 goto out;
2960 r = 0;
2961 break;
2962 }
9c15bb1d
BP		 2963 case KVM_GET_SUPPORTED_CPUID:
2964 case KVM_GET_EMULATED_CPUID: {
674eea0f
AK		 2965 struct kvm_cpuid2 __user *cpuid_arg = argp;
2966 struct kvm_cpuid2 cpuid;
2967
2968 r = -EFAULT;
2969 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2970 goto out;
9c15bb1d
BP		 2971
2972 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2973 ioctl);
674eea0f
AK		 2974 if (r)
2975 goto out;
2976
2977 r = -EFAULT;
2978 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2979 goto out;
2980 r = 0;
2981 break;
2982 }
890ca9ae 2983 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
890ca9ae 2984 r = -EFAULT;
c45dcc71
AR		 2985 if (copy_to_user(argp, &kvm_mce_cap_supported,
2986 sizeof(kvm_mce_cap_supported)))
890ca9ae
HY		 2987 goto out;
2988 r = 0;
2989 break;
801e459a
TL		 2990 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
2991 struct kvm_msr_list __user *user_msr_list = argp;
2992 struct kvm_msr_list msr_list;
2993 unsigned int n;
2994
2995 r = -EFAULT;
2996 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
2997 goto out;
2998 n = msr_list.nmsrs;
2999 msr_list.nmsrs = num_msr_based_features;
3000 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3001 goto out;
3002 r = -E2BIG;
3003 if (n < msr_list.nmsrs)
3004 goto out;
3005 r = -EFAULT;
3006 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3007 num_msr_based_features * sizeof(u32)))
3008 goto out;
3009 r = 0;
3010 break;
3011 }
3012 case KVM_GET_MSRS:
3013 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3014 break;
890ca9ae 3015 }
043405e1
CO		 3016 default:
3017 r = -EINVAL;
3018 }
3019out:
3020 return r;
3021}
3022
f5f48ee1
SY		 3023static void wbinvd_ipi(void *garbage)
3024{
3025 wbinvd();
3026}
3027
3028static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3029{
e0f0bbc5 3030 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
f5f48ee1
SY		 3031}
3032
313a3dc7
CO		 3033void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3034{
f5f48ee1
SY		 3035 /* Address WBINVD may be executed by guest */
3036 if (need_emulate_wbinvd(vcpu)) {
3037 if (kvm_x86_ops->has_wbinvd_exit())
3038 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3039 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3040 smp_call_function_single(vcpu->cpu,
3041 wbinvd_ipi, NULL, 1);
3042 }
3043
313a3dc7 3044 kvm_x86_ops->vcpu_load(vcpu, cpu);
8f6055cb 3045
0dd6a6ed
ZA		 3046 /* Apply any externally detected TSC adjustments (due to suspend) */
3047 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3048 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3049 vcpu->arch.tsc_offset_adjustment = 0;
105b21bb 3050 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
0dd6a6ed 3051 }
8f6055cb 3052
b0c39dc6 3053 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
6f526ec5 3054 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4ea1636b 3055 rdtsc() - vcpu->arch.last_host_tsc;
e48672fa
ZA		 3056 if (tsc_delta < 0)
3057 mark_tsc_unstable("KVM discovered backwards TSC");
ce7a058a 3058
b0c39dc6 3059 if (kvm_check_tsc_unstable()) {
07c1419a 3060 u64 offset = kvm_compute_tsc_offset(vcpu,
b183aa58 3061 vcpu->arch.last_guest_tsc);
a545ab6a 3062 kvm_vcpu_write_tsc_offset(vcpu, offset);
c285545f 3063 vcpu->arch.tsc_catchup = 1;
c285545f 3064 }
a749e247
PB		 3065
3066 if (kvm_lapic_hv_timer_in_use(vcpu))
3067 kvm_lapic_restart_hv_timer(vcpu);
3068
d98d07ca
MT		 3069 /*
3070 * On a host with synchronized TSC, there is no need to update
3071 * kvmclock on vcpu->cpu migration
3072 */
3073 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
0061d53d 3074 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
c285545f 3075 if (vcpu->cpu != cpu)
1bd2009e 3076 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
e48672fa 3077 vcpu->cpu = cpu;
6b7d7e76 3078 }
c9aaa895 3079
c9aaa895 3080 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
313a3dc7
CO		 3081}
3082
0b9f6c46
PX		 3083static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3084{
3085 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3086 return;
3087
fa55eedd 3088 vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
0b9f6c46 3089
4e335d9e 3090 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
0b9f6c46
PX		 3091 &vcpu->arch.st.steal.preempted,
3092 offsetof(struct kvm_steal_time, preempted),
3093 sizeof(vcpu->arch.st.steal.preempted));
3094}
3095
313a3dc7
CO		 3096void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3097{
cc0d907c 3098 int idx;
de63ad4c
LM		 3099
3100 if (vcpu->preempted)
3101 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3102
931f261b
AA		 3103 /*
3104 * Disable page faults because we're in atomic context here.
3105 * kvm_write_guest_offset_cached() would call might_fault()
3106 * that relies on pagefault_disable() to tell if there's a
3107 * bug. NOTE: the write to guest memory may not go through if
3108 * during postcopy live migration or if there's heavy guest
3109 * paging.
3110 */
3111 pagefault_disable();
cc0d907c
AA		 3112 /*
3113 * kvm_memslots() will be called by
3114 * kvm_write_guest_offset_cached() so take the srcu lock.
3115 */
3116 idx = srcu_read_lock(&vcpu->kvm->srcu);
0b9f6c46 3117 kvm_steal_time_set_preempted(vcpu);
cc0d907c 3118 srcu_read_unlock(&vcpu->kvm->srcu, idx);
931f261b 3119 pagefault_enable();
02daab21 3120 kvm_x86_ops->vcpu_put(vcpu);
4ea1636b 3121 vcpu->arch.last_host_tsc = rdtsc();
efdab992
WL		 3122 /*
3123 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3124 * on every vmexit, but if not, we might have a stale dr6 from the
3125 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3126 */
3127 set_debugreg(0, 6);
313a3dc7
CO		 3128}
3129
313a3dc7
CO		 3130static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3131 struct kvm_lapic_state *s)
3132{
fa59cc00 3133 if (vcpu->arch.apicv_active)
d62caabb
AS		 3134 kvm_x86_ops->sync_pir_to_irr(vcpu);
3135
a92e2543 3136 return kvm_apic_get_state(vcpu, s);
313a3dc7
CO		 3137}
3138
3139static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3140 struct kvm_lapic_state *s)
3141{
a92e2543
RK		 3142 int r;
3143
3144 r = kvm_apic_set_state(vcpu, s);
3145 if (r)
3146 return r;
cb142eb7 3147 update_cr8_intercept(vcpu);
313a3dc7
CO		 3148
3149 return 0;
3150}
3151
127a457a
MG		 3152static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3153{
3154 return (!lapic_in_kernel(vcpu) ||
3155 kvm_apic_accept_pic_intr(vcpu));
3156}
3157
782d422b
MG		 3158/*
3159 * if userspace requested an interrupt window, check that the
3160 * interrupt window is open.
3161 *
3162 * No need to exit to userspace if we already have an interrupt queued.
3163 */
3164static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3165{
3166 return kvm_arch_interrupt_allowed(vcpu) &&
3167 !kvm_cpu_has_interrupt(vcpu) &&
3168 !kvm_event_needs_reinjection(vcpu) &&
3169 kvm_cpu_accept_dm_intr(vcpu);
3170}
3171
f77bc6a4
ZX		 3172static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3173 struct kvm_interrupt *irq)
3174{
02cdb50f 3175 if (irq->irq >= KVM_NR_INTERRUPTS)
f77bc6a4 3176 return -EINVAL;
1c1a9ce9
SR		 3177
3178 if (!irqchip_in_kernel(vcpu->kvm)) {
3179 kvm_queue_interrupt(vcpu, irq->irq, false);
3180 kvm_make_request(KVM_REQ_EVENT, vcpu);
3181 return 0;
3182 }
3183
3184 /*
3185 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3186 * fail for in-kernel 8259.
3187 */
3188 if (pic_in_kernel(vcpu->kvm))
f77bc6a4 3189 return -ENXIO;
f77bc6a4 3190
1c1a9ce9
SR		 3191 if (vcpu->arch.pending_external_vector != -1)
3192 return -EEXIST;
f77bc6a4 3193
1c1a9ce9 3194 vcpu->arch.pending_external_vector = irq->irq;
934bf653 3195 kvm_make_request(KVM_REQ_EVENT, vcpu);
f77bc6a4
ZX		 3196 return 0;
3197}
3198
c4abb7c9
JK		 3199static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3200{
c4abb7c9 3201 kvm_inject_nmi(vcpu);
c4abb7c9
JK		 3202
3203 return 0;
3204}
3205
f077825a
PB		 3206static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3207{
64d60670
PB		 3208 kvm_make_request(KVM_REQ_SMI, vcpu);
3209
f077825a
PB		 3210 return 0;
3211}
3212
b209749f
AK		 3213static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3214 struct kvm_tpr_access_ctl *tac)
3215{
3216 if (tac->flags)
3217 return -EINVAL;
3218 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3219 return 0;
3220}
3221
890ca9ae
HY		 3222static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3223 u64 mcg_cap)
3224{
3225 int r;
3226 unsigned bank_num = mcg_cap & 0xff, bank;
3227
3228 r = -EINVAL;
a9e38c3e 3229 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
890ca9ae 3230 goto out;
c45dcc71 3231 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
890ca9ae
HY		 3232 goto out;
3233 r = 0;
3234 vcpu->arch.mcg_cap = mcg_cap;
3235 /* Init IA32_MCG_CTL to all 1s */
3236 if (mcg_cap & MCG_CTL_P)
3237 vcpu->arch.mcg_ctl = ~(u64)0;
3238 /* Init IA32_MCi_CTL to all 1s */
3239 for (bank = 0; bank < bank_num; bank++)
3240 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
c45dcc71
AR		 3241
3242 if (kvm_x86_ops->setup_mce)
3243 kvm_x86_ops->setup_mce(vcpu);
890ca9ae
HY		 3244out:
3245 return r;
3246}
3247
3248static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3249 struct kvm_x86_mce *mce)
3250{
3251 u64 mcg_cap = vcpu->arch.mcg_cap;
3252 unsigned bank_num = mcg_cap & 0xff;
3253 u64 *banks = vcpu->arch.mce_banks;
3254
3255 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3256 return -EINVAL;
3257 /*
3258 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3259 * reporting is disabled
3260 */
3261 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3262 vcpu->arch.mcg_ctl != ~(u64)0)
3263 return 0;
3264 banks += 4 * mce->bank;
3265 /*
3266 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3267 * reporting is disabled for the bank
3268 */
3269 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3270 return 0;
3271 if (mce->status & MCI_STATUS_UC) {
3272 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
fc78f519 3273 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
a8eeb04a 3274 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
890ca9ae
HY		 3275 return 0;
3276 }
3277 if (banks[1] & MCI_STATUS_VAL)
3278 mce->status |= MCI_STATUS_OVER;
3279 banks[2] = mce->addr;
3280 banks[3] = mce->misc;
3281 vcpu->arch.mcg_status = mce->mcg_status;
3282 banks[1] = mce->status;
3283 kvm_queue_exception(vcpu, MC_VECTOR);
3284 } else if (!(banks[1] & MCI_STATUS_VAL)
3285 || !(banks[1] & MCI_STATUS_UC)) {
3286 if (banks[1] & MCI_STATUS_VAL)
3287 mce->status |= MCI_STATUS_OVER;
3288 banks[2] = mce->addr;
3289 banks[3] = mce->misc;
3290 banks[1] = mce->status;
3291 } else
3292 banks[1] |= MCI_STATUS_OVER;
3293 return 0;
3294}
3295
3cfc3092
JK		 3296static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3297 struct kvm_vcpu_events *events)
3298{
7460fb4a 3299 process_nmi(vcpu);
664f8e26
WL		 3300 /*
3301 * FIXME: pass injected and pending separately. This is only
3302 * needed for nested virtualization, whose state cannot be
3303 * migrated yet. For now we can combine them.
3304 */
03b82a30 3305 events->exception.injected =
664f8e26
WL		 3306 (vcpu->arch.exception.pending ||
3307 vcpu->arch.exception.injected) &&
03b82a30 3308 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3cfc3092
JK		 3309 events->exception.nr = vcpu->arch.exception.nr;
3310 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
97e69aa6 3311 events->exception.pad = 0;
3cfc3092
JK		 3312 events->exception.error_code = vcpu->arch.exception.error_code;
3313
03b82a30 3314 events->interrupt.injected =
04140b41 3315 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3cfc3092 3316 events->interrupt.nr = vcpu->arch.interrupt.nr;
03b82a30 3317 events->interrupt.soft = 0;
37ccdcbe 3318 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3cfc3092
JK		 3319
3320 events->nmi.injected = vcpu->arch.nmi_injected;
7460fb4a 3321 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3cfc3092 3322 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
97e69aa6 3323 events->nmi.pad = 0;
3cfc3092 3324
66450a21 3325 events->sipi_vector = 0; /* never valid when reporting to user space */
3cfc3092 3326
f077825a
PB		 3327 events->smi.smm = is_smm(vcpu);
3328 events->smi.pending = vcpu->arch.smi_pending;
3329 events->smi.smm_inside_nmi =
3330 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3331 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3332
dab4b911 3333 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
f077825a
PB		 3334 | KVM_VCPUEVENT_VALID_SHADOW
3335 | KVM_VCPUEVENT_VALID_SMM);
97e69aa6 3336 memset(&events->reserved, 0, sizeof(events->reserved));
3cfc3092
JK		 3337}
3338
6ef4e07e
XG		 3339static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3340
3cfc3092
JK		 3341static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3342 struct kvm_vcpu_events *events)
3343{
dab4b911 3344 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
48005f64 3345 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
f077825a
PB		 3346 | KVM_VCPUEVENT_VALID_SHADOW
3347 | KVM_VCPUEVENT_VALID_SMM))
3cfc3092
JK		 3348 return -EINVAL;
3349
78e546c8 3350 if (events->exception.injected &&
28d06353
JM		 3351 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR ||
3352 is_guest_mode(vcpu)))
78e546c8
PB		 3353 return -EINVAL;
3354
28bf2888
DH		 3355 /* INITs are latched while in SMM */
3356 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3357 (events->smi.smm || events->smi.pending) &&
3358 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3359 return -EINVAL;
3360
7460fb4a 3361 process_nmi(vcpu);
664f8e26 3362 vcpu->arch.exception.injected = false;
3cfc3092
JK		 3363 vcpu->arch.exception.pending = events->exception.injected;
3364 vcpu->arch.exception.nr = events->exception.nr;
3365 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3366 vcpu->arch.exception.error_code = events->exception.error_code;
3367
04140b41 3368 vcpu->arch.interrupt.injected = events->interrupt.injected;
3cfc3092
JK		 3369 vcpu->arch.interrupt.nr = events->interrupt.nr;
3370 vcpu->arch.interrupt.soft = events->interrupt.soft;
48005f64
JK		 3371 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3372 kvm_x86_ops->set_interrupt_shadow(vcpu,
3373 events->interrupt.shadow);
3cfc3092
JK		 3374
3375 vcpu->arch.nmi_injected = events->nmi.injected;
dab4b911
JK		 3376 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3377 vcpu->arch.nmi_pending = events->nmi.pending;
3cfc3092
JK		 3378 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3379
66450a21 3380 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
bce87cce 3381 lapic_in_kernel(vcpu))
66450a21 3382 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3cfc3092 3383
f077825a 3384 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
6ef4e07e 3385 u32 hflags = vcpu->arch.hflags;
f077825a 3386 if (events->smi.smm)
6ef4e07e 3387 hflags |= HF_SMM_MASK;
f077825a 3388 else
6ef4e07e
XG		 3389 hflags &= ~HF_SMM_MASK;
3390 kvm_set_hflags(vcpu, hflags);
3391
f077825a 3392 vcpu->arch.smi_pending = events->smi.pending;
f4ef1910
WL		 3393
3394 if (events->smi.smm) {
3395 if (events->smi.smm_inside_nmi)
3396 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
f077825a 3397 else
f4ef1910
WL		 3398 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3399 if (lapic_in_kernel(vcpu)) {
3400 if (events->smi.latched_init)
3401 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3402 else
3403 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3404 }
f077825a
PB		 3405 }
3406 }
3407
3842d135
AK		 3408 kvm_make_request(KVM_REQ_EVENT, vcpu);
3409
3cfc3092
JK		 3410 return 0;
3411}
3412
a1efbe77
JK		 3413static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3414 struct kvm_debugregs *dbgregs)
3415{
73aaf249
JK		 3416 unsigned long val;
3417
a1efbe77 3418 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
16f8a6f9 3419 kvm_get_dr(vcpu, 6, &val);
73aaf249 3420 dbgregs->dr6 = val;
a1efbe77
JK		 3421 dbgregs->dr7 = vcpu->arch.dr7;
3422 dbgregs->flags = 0;
97e69aa6 3423 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
a1efbe77
JK		 3424}
3425
3426static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3427 struct kvm_debugregs *dbgregs)
3428{
3429 if (dbgregs->flags)
3430 return -EINVAL;
3431
d14bdb55
PB		 3432 if (dbgregs->dr6 & ~0xffffffffull)
3433 return -EINVAL;
3434 if (dbgregs->dr7 & ~0xffffffffull)
3435 return -EINVAL;
3436
a1efbe77 3437 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
ae561ede 3438 kvm_update_dr0123(vcpu);
a1efbe77 3439 vcpu->arch.dr6 = dbgregs->dr6;
73aaf249 3440 kvm_update_dr6(vcpu);
a1efbe77 3441 vcpu->arch.dr7 = dbgregs->dr7;
9926c9fd 3442 kvm_update_dr7(vcpu);
a1efbe77 3443
a1efbe77
JK		 3444 return 0;
3445}
3446
df1daba7
PB		 3447#define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3448
3449static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3450{
c47ada30 3451 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
400e4b20 3452 u64 xstate_bv = xsave->header.xfeatures;
df1daba7
PB		 3453 u64 valid;
3454
3455 /*
3456 * Copy legacy XSAVE area, to avoid complications with CPUID
3457 * leaves 0 and 1 in the loop below.
3458 */
3459 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3460
3461 /* Set XSTATE_BV */
00c87e9a 3462 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
df1daba7
PB		 3463 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3464
3465 /*
3466 * Copy each region from the possibly compacted offset to the
3467 * non-compacted offset.
3468 */
d91cab78 3469 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
df1daba7
PB		 3470 while (valid) {
3471 u64 feature = valid & -valid;
3472 int index = fls64(feature) - 1;
3473 void *src = get_xsave_addr(xsave, feature);
3474
3475 if (src) {
3476 u32 size, offset, ecx, edx;
3477 cpuid_count(XSTATE_CPUID, index,
3478 &size, &offset, &ecx, &edx);
38cfd5e3
PB		 3479 if (feature == XFEATURE_MASK_PKRU)
3480 memcpy(dest + offset, &vcpu->arch.pkru,
3481 sizeof(vcpu->arch.pkru));
3482 else
3483 memcpy(dest + offset, src, size);
3484
df1daba7
PB		 3485 }
3486
3487 valid -= feature;
3488 }
3489}
3490
3491static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3492{
c47ada30 3493 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
df1daba7
PB		 3494 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3495 u64 valid;
3496
3497 /*
3498 * Copy legacy XSAVE area, to avoid complications with CPUID
3499 * leaves 0 and 1 in the loop below.
3500 */
3501 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3502
3503 /* Set XSTATE_BV and possibly XCOMP_BV. */
400e4b20 3504 xsave->header.xfeatures = xstate_bv;
782511b0 3505 if (boot_cpu_has(X86_FEATURE_XSAVES))
3a54450b 3506 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
df1daba7
PB		 3507
3508 /*
3509 * Copy each region from the non-compacted offset to the
3510 * possibly compacted offset.
3511 */
d91cab78 3512 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
df1daba7
PB		 3513 while (valid) {
3514 u64 feature = valid & -valid;
3515 int index = fls64(feature) - 1;
3516 void *dest = get_xsave_addr(xsave, feature);
3517
3518 if (dest) {
3519 u32 size, offset, ecx, edx;
3520 cpuid_count(XSTATE_CPUID, index,
3521 &size, &offset, &ecx, &edx);
38cfd5e3
PB		 3522 if (feature == XFEATURE_MASK_PKRU)
3523 memcpy(&vcpu->arch.pkru, src + offset,
3524 sizeof(vcpu->arch.pkru));
3525 else
3526 memcpy(dest, src + offset, size);
ee4100da 3527 }
df1daba7
PB		 3528
3529 valid -= feature;
3530 }
3531}
3532
2d5b5a66
SY		 3533static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3534 struct kvm_xsave *guest_xsave)
3535{
d366bf7e 3536 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
df1daba7
PB		 3537 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3538 fill_xsave((u8 *) guest_xsave->region, vcpu);
4344ee98 3539 } else {
2d5b5a66 3540 memcpy(guest_xsave->region,
7366ed77 3541 &vcpu->arch.guest_fpu.state.fxsave,
c47ada30 3542 sizeof(struct fxregs_state));
2d5b5a66 3543 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
d91cab78 3544 XFEATURE_MASK_FPSSE;
2d5b5a66
SY		 3545 }
3546}
3547
a575813b
WL		 3548#define XSAVE_MXCSR_OFFSET 24
3549
2d5b5a66
SY		 3550static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3551 struct kvm_xsave *guest_xsave)
3552{
3553 u64 xstate_bv =
3554 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
a575813b 3555 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
2d5b5a66 3556
d366bf7e 3557 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
d7876f1b
PB		 3558 /*
3559 * Here we allow setting states that are not present in
3560 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3561 * with old userspace.
3562 */
a575813b
WL		 3563 if (xstate_bv & ~kvm_supported_xcr0() ||
3564 mxcsr & ~mxcsr_feature_mask)
d7876f1b 3565 return -EINVAL;
df1daba7 3566 load_xsave(vcpu, (u8 *)guest_xsave->region);
d7876f1b 3567 } else {
a575813b
WL		 3568 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3569 mxcsr & ~mxcsr_feature_mask)
2d5b5a66 3570 return -EINVAL;
7366ed77 3571 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
c47ada30 3572 guest_xsave->region, sizeof(struct fxregs_state));
2d5b5a66
SY		 3573 }
3574 return 0;
3575}
3576
3577static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3578 struct kvm_xcrs *guest_xcrs)
3579{
d366bf7e 3580 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
2d5b5a66
SY		 3581 guest_xcrs->nr_xcrs = 0;
3582 return;
3583 }
3584
3585 guest_xcrs->nr_xcrs = 1;
3586 guest_xcrs->flags = 0;
3587 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3588 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3589}
3590
3591static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3592 struct kvm_xcrs *guest_xcrs)
3593{
3594 int i, r = 0;
3595
d366bf7e 3596 if (!boot_cpu_has(X86_FEATURE_XSAVE))
2d5b5a66
SY		 3597 return -EINVAL;
3598
3599 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3600 return -EINVAL;
3601
3602 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3603 /* Only support XCR0 currently */
c67a04cb 3604 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
2d5b5a66 3605 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
c67a04cb 3606 guest_xcrs->xcrs[i].value);
2d5b5a66
SY		 3607 break;
3608 }
3609 if (r)
3610 r = -EINVAL;
3611 return r;
3612}
3613
1c0b28c2
EM		 3614/*
3615 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3616 * stopped by the hypervisor. This function will be called from the host only.
3617 * EINVAL is returned when the host attempts to set the flag for a guest that
3618 * does not support pv clocks.
3619 */
3620static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3621{
0b79459b 3622 if (!vcpu->arch.pv_time_enabled)
1c0b28c2 3623 return -EINVAL;
51d59c6b 3624 vcpu->arch.pvclock_set_guest_stopped_request = true;
1c0b28c2
EM		 3625 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3626 return 0;
3627}
3628
5c919412
AS		 3629static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3630 struct kvm_enable_cap *cap)
3631{
3632 if (cap->flags)
3633 return -EINVAL;
3634
3635 switch (cap->cap) {
efc479e6
RK		 3636 case KVM_CAP_HYPERV_SYNIC2:
3637 if (cap->args[0])
3638 return -EINVAL;
5c919412 3639 case KVM_CAP_HYPERV_SYNIC:
546d87e5
WL		 3640 if (!irqchip_in_kernel(vcpu->kvm))
3641 return -EINVAL;
efc479e6
RK		 3642 return kvm_hv_activate_synic(vcpu, cap->cap ==
3643 KVM_CAP_HYPERV_SYNIC2);
5c919412
AS		 3644 default:
3645 return -EINVAL;
3646 }
3647}
3648
313a3dc7
CO		 3649long kvm_arch_vcpu_ioctl(struct file *filp,
3650 unsigned int ioctl, unsigned long arg)
3651{
3652 struct kvm_vcpu *vcpu = filp->private_data;
3653 void __user *argp = (void __user *)arg;
3654 int r;
d1ac91d8
AK		 3655 union {
3656 struct kvm_lapic_state *lapic;
3657 struct kvm_xsave *xsave;
3658 struct kvm_xcrs *xcrs;
3659 void *buffer;
3660 } u;
3661
9b062471
CD		 3662 vcpu_load(vcpu);
3663
d1ac91d8 3664 u.buffer = NULL;
313a3dc7
CO		 3665 switch (ioctl) {
3666 case KVM_GET_LAPIC: {
2204ae3c 3667 r = -EINVAL;
bce87cce 3668 if (!lapic_in_kernel(vcpu))
2204ae3c 3669 goto out;
d1ac91d8 3670 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
313a3dc7 3671
b772ff36 3672 r = -ENOMEM;
d1ac91d8 3673 if (!u.lapic)
b772ff36 3674 goto out;
d1ac91d8 3675 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
313a3dc7
CO		 3676 if (r)
3677 goto out;
3678 r = -EFAULT;
d1ac91d8 3679 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
313a3dc7
CO		 3680 goto out;
3681 r = 0;
3682 break;
3683 }
3684 case KVM_SET_LAPIC: {
2204ae3c 3685 r = -EINVAL;
bce87cce 3686 if (!lapic_in_kernel(vcpu))
2204ae3c 3687 goto out;
ff5c2c03 3688 u.lapic = memdup_user(argp, sizeof(*u.lapic));
9b062471
CD		 3689 if (IS_ERR(u.lapic)) {
3690 r = PTR_ERR(u.lapic);
3691 goto out_nofree;
3692 }
ff5c2c03 3693
d1ac91d8 3694 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
313a3dc7
CO		 3695 break;
3696 }
f77bc6a4
ZX		 3697 case KVM_INTERRUPT: {
3698 struct kvm_interrupt irq;
3699
3700 r = -EFAULT;
3701 if (copy_from_user(&irq, argp, sizeof irq))
3702 goto out;
3703 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
f77bc6a4
ZX		 3704 break;
3705 }
c4abb7c9
JK		 3706 case KVM_NMI: {
3707 r = kvm_vcpu_ioctl_nmi(vcpu);
c4abb7c9
JK		 3708 break;
3709 }
f077825a
PB		 3710 case KVM_SMI: {
3711 r = kvm_vcpu_ioctl_smi(vcpu);
3712 break;
3713 }
313a3dc7
CO		 3714 case KVM_SET_CPUID: {
3715 struct kvm_cpuid __user *cpuid_arg = argp;
3716 struct kvm_cpuid cpuid;
3717
3718 r = -EFAULT;
3719 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3720 goto out;
3721 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
313a3dc7
CO		 3722 break;
3723 }
07716717
DK		 3724 case KVM_SET_CPUID2: {
3725 struct kvm_cpuid2 __user *cpuid_arg = argp;
3726 struct kvm_cpuid2 cpuid;
3727
3728 r = -EFAULT;
3729 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3730 goto out;
3731 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
19355475 3732 cpuid_arg->entries);
07716717
DK		 3733 break;
3734 }
3735 case KVM_GET_CPUID2: {
3736 struct kvm_cpuid2 __user *cpuid_arg = argp;
3737 struct kvm_cpuid2 cpuid;
3738
3739 r = -EFAULT;
3740 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3741 goto out;
3742 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
19355475 3743 cpuid_arg->entries);
07716717
DK		 3744 if (r)
3745 goto out;
3746 r = -EFAULT;
3747 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3748 goto out;
3749 r = 0;
3750 break;
3751 }
801e459a
TL		 3752 case KVM_GET_MSRS: {
3753 int idx = srcu_read_lock(&vcpu->kvm->srcu);
609e36d3 3754 r = msr_io(vcpu, argp, do_get_msr, 1);
801e459a 3755 srcu_read_unlock(&vcpu->kvm->srcu, idx);
313a3dc7 3756 break;
801e459a
TL		 3757 }
3758 case KVM_SET_MSRS: {
3759 int idx = srcu_read_lock(&vcpu->kvm->srcu);
313a3dc7 3760 r = msr_io(vcpu, argp, do_set_msr, 0);
801e459a 3761 srcu_read_unlock(&vcpu->kvm->srcu, idx);
313a3dc7 3762 break;
801e459a 3763 }
b209749f
AK		 3764 case KVM_TPR_ACCESS_REPORTING: {
3765 struct kvm_tpr_access_ctl tac;
3766
3767 r = -EFAULT;
3768 if (copy_from_user(&tac, argp, sizeof tac))
3769 goto out;
3770 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3771 if (r)
3772 goto out;
3773 r = -EFAULT;
3774 if (copy_to_user(argp, &tac, sizeof tac))
3775 goto out;
3776 r = 0;
3777 break;
3778 };
b93463aa
AK		 3779 case KVM_SET_VAPIC_ADDR: {
3780 struct kvm_vapic_addr va;
7301d6ab 3781 int idx;
b93463aa
AK		 3782
3783 r = -EINVAL;
35754c98 3784 if (!lapic_in_kernel(vcpu))
b93463aa
AK		 3785 goto out;
3786 r = -EFAULT;
3787 if (copy_from_user(&va, argp, sizeof va))
3788 goto out;
7301d6ab 3789 idx = srcu_read_lock(&vcpu->kvm->srcu);
fda4e2e8 3790 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
7301d6ab 3791 srcu_read_unlock(&vcpu->kvm->srcu, idx);
b93463aa
AK		 3792 break;
3793 }
890ca9ae
HY		 3794 case KVM_X86_SETUP_MCE: {
3795 u64 mcg_cap;
3796
3797 r = -EFAULT;
3798 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3799 goto out;
3800 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3801 break;
3802 }
3803 case KVM_X86_SET_MCE: {
3804 struct kvm_x86_mce mce;
3805
3806 r = -EFAULT;
3807 if (copy_from_user(&mce, argp, sizeof mce))
3808 goto out;
3809 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3810 break;
3811 }
3cfc3092
JK		 3812 case KVM_GET_VCPU_EVENTS: {
3813 struct kvm_vcpu_events events;
3814
3815 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3816
3817 r = -EFAULT;
3818 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3819 break;
3820 r = 0;
3821 break;
3822 }
3823 case KVM_SET_VCPU_EVENTS: {
3824 struct kvm_vcpu_events events;
3825
3826 r = -EFAULT;
3827 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3828 break;
3829
3830 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3831 break;
3832 }
a1efbe77
JK		 3833 case KVM_GET_DEBUGREGS: {
3834 struct kvm_debugregs dbgregs;
3835
3836 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3837
3838 r = -EFAULT;
3839 if (copy_to_user(argp, &dbgregs,
3840 sizeof(struct kvm_debugregs)))
3841 break;
3842 r = 0;
3843 break;
3844 }
3845 case KVM_SET_DEBUGREGS: {
3846 struct kvm_debugregs dbgregs;
3847
3848 r = -EFAULT;
3849 if (copy_from_user(&dbgregs, argp,
3850 sizeof(struct kvm_debugregs)))
3851 break;
3852
3853 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3854 break;
3855 }
2d5b5a66 3856 case KVM_GET_XSAVE: {
d1ac91d8 3857 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2d5b5a66 3858 r = -ENOMEM;
d1ac91d8 3859 if (!u.xsave)
2d5b5a66
SY		 3860 break;
3861
d1ac91d8 3862 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
2d5b5a66
SY		 3863
3864 r = -EFAULT;
d1ac91d8 3865 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
2d5b5a66
SY		 3866 break;
3867 r = 0;
3868 break;
3869 }
3870 case KVM_SET_XSAVE: {
ff5c2c03 3871 u.xsave = memdup_user(argp, sizeof(*u.xsave));
9b062471
CD		 3872 if (IS_ERR(u.xsave)) {
3873 r = PTR_ERR(u.xsave);
3874 goto out_nofree;
3875 }
2d5b5a66 3876
d1ac91d8 3877 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
2d5b5a66
SY		 3878 break;
3879 }
3880 case KVM_GET_XCRS: {
d1ac91d8 3881 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2d5b5a66 3882 r = -ENOMEM;
d1ac91d8 3883 if (!u.xcrs)
2d5b5a66
SY		 3884 break;
3885
d1ac91d8 3886 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
2d5b5a66
SY		 3887
3888 r = -EFAULT;
d1ac91d8 3889 if (copy_to_user(argp, u.xcrs,
2d5b5a66
SY		 3890 sizeof(struct kvm_xcrs)))
3891 break;
3892 r = 0;
3893 break;
3894 }
3895 case KVM_SET_XCRS: {
ff5c2c03 3896 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
9b062471
CD		 3897 if (IS_ERR(u.xcrs)) {
3898 r = PTR_ERR(u.xcrs);
3899 goto out_nofree;
3900 }
2d5b5a66 3901
d1ac91d8 3902 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
2d5b5a66
SY		 3903 break;
3904 }
92a1f12d
JR		 3905 case KVM_SET_TSC_KHZ: {
3906 u32 user_tsc_khz;
3907
3908 r = -EINVAL;
92a1f12d
JR		 3909 user_tsc_khz = (u32)arg;
3910
3911 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3912 goto out;
3913
cc578287
ZA		 3914 if (user_tsc_khz == 0)
3915 user_tsc_khz = tsc_khz;
3916
381d585c
HZ		 3917 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3918 r = 0;
92a1f12d 3919
92a1f12d
JR		 3920 goto out;
3921 }
3922 case KVM_GET_TSC_KHZ: {
cc578287 3923 r = vcpu->arch.virtual_tsc_khz;
92a1f12d
JR		 3924 goto out;
3925 }
1c0b28c2
EM		 3926 case KVM_KVMCLOCK_CTRL: {
3927 r = kvm_set_guest_paused(vcpu);
3928 goto out;
3929 }
5c919412
AS		 3930 case KVM_ENABLE_CAP: {
3931 struct kvm_enable_cap cap;
3932
3933 r = -EFAULT;
3934 if (copy_from_user(&cap, argp, sizeof(cap)))
3935 goto out;
3936 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3937 break;
3938 }
313a3dc7
CO		 3939 default:
3940 r = -EINVAL;
3941 }
3942out:
d1ac91d8 3943 kfree(u.buffer);
9b062471
CD		 3944out_nofree:
3945 vcpu_put(vcpu);
313a3dc7
CO		 3946 return r;
3947}
3948
5b1c1493
CO		 3949int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3950{
3951 return VM_FAULT_SIGBUS;
3952}
3953
1fe779f8
CO		 3954static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3955{
3956 int ret;
3957
3958 if (addr > (unsigned int)(-3 * PAGE_SIZE))
951179ce 3959 return -EINVAL;
1fe779f8
CO		 3960 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3961 return ret;
3962}
3963
b927a3ce
SY		 3964static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3965 u64 ident_addr)
3966{
2ac52ab8 3967 return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
b927a3ce
SY		 3968}
3969
1fe779f8
CO		 3970static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3971 u32 kvm_nr_mmu_pages)
3972{
3973 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3974 return -EINVAL;
3975
79fac95e 3976 mutex_lock(&kvm->slots_lock);
1fe779f8
CO		 3977
3978 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
f05e70ac 3979 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1fe779f8 3980
79fac95e 3981 mutex_unlock(&kvm->slots_lock);
1fe779f8
CO		 3982 return 0;
3983}
3984
3985static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3986{
39de71ec 3987 return kvm->arch.n_max_mmu_pages;
1fe779f8
CO		 3988}
3989
1fe779f8
CO		 3990static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3991{
90bca052 3992 struct kvm_pic *pic = kvm->arch.vpic;
1fe779f8
CO		 3993 int r;
3994
3995 r = 0;
3996 switch (chip->chip_id) {
3997 case KVM_IRQCHIP_PIC_MASTER:
90bca052 3998 memcpy(&chip->chip.pic, &pic->pics[0],
1fe779f8
CO		 3999 sizeof(struct kvm_pic_state));
4000 break;
4001 case KVM_IRQCHIP_PIC_SLAVE:
90bca052 4002 memcpy(&chip->chip.pic, &pic->pics[1],
1fe779f8
CO		 4003 sizeof(struct kvm_pic_state));
4004 break;
4005 case KVM_IRQCHIP_IOAPIC:
33392b49 4006 kvm_get_ioapic(kvm, &chip->chip.ioapic);
1fe779f8
CO		 4007 break;
4008 default:
4009 r = -EINVAL;
4010 break;
4011 }
4012 return r;
4013}
4014
4015static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4016{
90bca052 4017 struct kvm_pic *pic = kvm->arch.vpic;
1fe779f8
CO		 4018 int r;
4019
4020 r = 0;
4021 switch (chip->chip_id) {
4022 case KVM_IRQCHIP_PIC_MASTER:
90bca052
DH		 4023 spin_lock(&pic->lock);
4024 memcpy(&pic->pics[0], &chip->chip.pic,
1fe779f8 4025 sizeof(struct kvm_pic_state));
90bca052 4026 spin_unlock(&pic->lock);
1fe779f8
CO		 4027 break;
4028 case KVM_IRQCHIP_PIC_SLAVE:
90bca052
DH		 4029 spin_lock(&pic->lock);
4030 memcpy(&pic->pics[1], &chip->chip.pic,
1fe779f8 4031 sizeof(struct kvm_pic_state));
90bca052 4032 spin_unlock(&pic->lock);
1fe779f8
CO		 4033 break;
4034 case KVM_IRQCHIP_IOAPIC:
33392b49 4035 kvm_set_ioapic(kvm, &chip->chip.ioapic);
1fe779f8
CO		 4036 break;
4037 default:
4038 r = -EINVAL;
4039 break;
4040 }
90bca052 4041 kvm_pic_update_irq(pic);
1fe779f8
CO		 4042 return r;
4043}
4044
e0f63cb9
SY		 4045static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4046{
34f3941c
RK		 4047 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4048
4049 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4050
4051 mutex_lock(&kps->lock);
4052 memcpy(ps, &kps->channels, sizeof(*ps));
4053 mutex_unlock(&kps->lock);
2da29bcc 4054 return 0;
e0f63cb9
SY		 4055}
4056
4057static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4058{
0185604c 4059 int i;
09edea72
RK		 4060 struct kvm_pit *pit = kvm->arch.vpit;
4061
4062 mutex_lock(&pit->pit_state.lock);
34f3941c 4063 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
0185604c 4064 for (i = 0; i < 3; i++)
09edea72
RK		 4065 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4066 mutex_unlock(&pit->pit_state.lock);
2da29bcc 4067 return 0;
e9f42757
BK		 4068}
4069
4070static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4071{
e9f42757
BK		 4072 mutex_lock(&kvm->arch.vpit->pit_state.lock);
4073 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4074 sizeof(ps->channels));
4075 ps->flags = kvm->arch.vpit->pit_state.flags;
4076 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
97e69aa6 4077 memset(&ps->reserved, 0, sizeof(ps->reserved));
2da29bcc 4078 return 0;
e9f42757
BK		 4079}
4080
4081static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4082{
2da29bcc 4083 int start = 0;
0185604c 4084 int i;
e9f42757 4085 u32 prev_legacy, cur_legacy;
09edea72
RK		 4086 struct kvm_pit *pit = kvm->arch.vpit;
4087
4088 mutex_lock(&pit->pit_state.lock);
4089 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
e9f42757
BK		 4090 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4091 if (!prev_legacy && cur_legacy)
4092 start = 1;
09edea72
RK		 4093 memcpy(&pit->pit_state.channels, &ps->channels,
4094 sizeof(pit->pit_state.channels));
4095 pit->pit_state.flags = ps->flags;
0185604c 4096 for (i = 0; i < 3; i++)
09edea72 4097 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
e5e57e7a 4098 start && i == 0);
09edea72 4099 mutex_unlock(&pit->pit_state.lock);
2da29bcc 4100 return 0;
e0f63cb9
SY		 4101}
4102
52d939a0
MT		 4103static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4104 struct kvm_reinject_control *control)
4105{
71474e2f
RK		 4106 struct kvm_pit *pit = kvm->arch.vpit;
4107
4108 if (!pit)
52d939a0 4109 return -ENXIO;
b39c90b6 4110
71474e2f
RK		 4111 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4112 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4113 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4114 */
4115 mutex_lock(&pit->pit_state.lock);
4116 kvm_pit_set_reinject(pit, control->pit_reinject);
4117 mutex_unlock(&pit->pit_state.lock);
b39c90b6 4118
52d939a0
MT		 4119 return 0;
4120}
4121
95d4c16c 4122/**
60c34612
TY		 4123 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4124 * @kvm: kvm instance
4125 * @log: slot id and address to which we copy the log
95d4c16c 4126 *
e108ff2f
PB		 4127 * Steps 1-4 below provide general overview of dirty page logging. See
4128 * kvm_get_dirty_log_protect() function description for additional details.
4129 *
4130 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4131 * always flush the TLB (step 4) even if previous step failed and the dirty
4132 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4133 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4134 * writes will be marked dirty for next log read.
95d4c16c 4135 *
60c34612
TY		 4136 * 1. Take a snapshot of the bit and clear it if needed.
4137 * 2. Write protect the corresponding page.
e108ff2f
PB		 4138 * 3. Copy the snapshot to the userspace.
4139 * 4. Flush TLB's if needed.
5bb064dc 4140 */
60c34612 4141int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
5bb064dc 4142{
60c34612 4143 bool is_dirty = false;
e108ff2f 4144 int r;
5bb064dc 4145
79fac95e 4146 mutex_lock(&kvm->slots_lock);
5bb064dc 4147
88178fd4
KH		 4148 /*
4149 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4150 */
4151 if (kvm_x86_ops->flush_log_dirty)
4152 kvm_x86_ops->flush_log_dirty(kvm);
4153
e108ff2f 4154 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
198c74f4
XG		 4155
4156 /*
4157 * All the TLBs can be flushed out of mmu lock, see the comments in
4158 * kvm_mmu_slot_remove_write_access().
4159 */
e108ff2f 4160 lockdep_assert_held(&kvm->slots_lock);
198c74f4
XG		 4161 if (is_dirty)
4162 kvm_flush_remote_tlbs(kvm);
4163
79fac95e 4164 mutex_unlock(&kvm->slots_lock);
5bb064dc
ZX		 4165 return r;
4166}
4167
aa2fbe6d
YZ		 4168int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4169 bool line_status)
23d43cf9
CD		 4170{
4171 if (!irqchip_in_kernel(kvm))
4172 return -ENXIO;
4173
4174 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
aa2fbe6d
YZ		 4175 irq_event->irq, irq_event->level,
4176 line_status);
23d43cf9
CD		 4177 return 0;
4178}
4179
90de4a18
NA		 4180static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4181 struct kvm_enable_cap *cap)
4182{
4183 int r;
4184
4185 if (cap->flags)
4186 return -EINVAL;
4187
4188 switch (cap->cap) {
4189 case KVM_CAP_DISABLE_QUIRKS:
4190 kvm->arch.disabled_quirks = cap->args[0];
4191 r = 0;
4192 break;
49df6397
SR		 4193 case KVM_CAP_SPLIT_IRQCHIP: {
4194 mutex_lock(&kvm->lock);
b053b2ae
SR		 4195 r = -EINVAL;
4196 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4197 goto split_irqchip_unlock;
49df6397
SR		 4198 r = -EEXIST;
4199 if (irqchip_in_kernel(kvm))
4200 goto split_irqchip_unlock;
557abc40 4201 if (kvm->created_vcpus)
49df6397
SR		 4202 goto split_irqchip_unlock;
4203 r = kvm_setup_empty_irq_routing(kvm);
5c0aea0e 4204 if (r)
49df6397
SR		 4205 goto split_irqchip_unlock;
4206 /* Pairs with irqchip_in_kernel. */
4207 smp_wmb();
49776faf 4208 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
b053b2ae 4209 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
49df6397
SR		 4210 r = 0;
4211split_irqchip_unlock:
4212 mutex_unlock(&kvm->lock);
4213 break;
4214 }
37131313
RK		 4215 case KVM_CAP_X2APIC_API:
4216 r = -EINVAL;
4217 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4218 break;
4219
4220 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4221 kvm->arch.x2apic_format = true;
c519265f
RK		 4222 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4223 kvm->arch.x2apic_broadcast_quirk_disabled = true;
37131313
RK		 4224
4225 r = 0;
4226 break;
4d5422ce
WL		 4227 case KVM_CAP_X86_DISABLE_EXITS:
4228 r = -EINVAL;
4229 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4230 break;
4231
4232 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4233 kvm_can_mwait_in_guest())
4234 kvm->arch.mwait_in_guest = true;
caa057a2
WL		 4235 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HTL)
4236 kvm->arch.hlt_in_guest = true;
b31c114b
WL		 4237 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4238 kvm->arch.pause_in_guest = true;
4d5422ce
WL		 4239 r = 0;
4240 break;
90de4a18
NA		 4241 default:
4242 r = -EINVAL;
4243 break;
4244 }
4245 return r;
4246}
4247
1fe779f8
CO		 4248long kvm_arch_vm_ioctl(struct file *filp,
4249 unsigned int ioctl, unsigned long arg)
4250{
4251 struct kvm *kvm = filp->private_data;
4252 void __user *argp = (void __user *)arg;
367e1319 4253 int r = -ENOTTY;
f0d66275
DH		 4254 /*
4255 * This union makes it completely explicit to gcc-3.x
4256 * that these two variables' stack usage should be
4257 * combined, not added together.
4258 */
4259 union {
4260 struct kvm_pit_state ps;
e9f42757 4261 struct kvm_pit_state2 ps2;
c5ff41ce 4262 struct kvm_pit_config pit_config;
f0d66275 4263 } u;
1fe779f8
CO		 4264
4265 switch (ioctl) {
4266 case KVM_SET_TSS_ADDR:
4267 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
1fe779f8 4268 break;
b927a3ce
SY		 4269 case KVM_SET_IDENTITY_MAP_ADDR: {
4270 u64 ident_addr;
4271
1af1ac91
DH		 4272 mutex_lock(&kvm->lock);
4273 r = -EINVAL;
4274 if (kvm->created_vcpus)
4275 goto set_identity_unlock;
b927a3ce
SY		 4276 r = -EFAULT;
4277 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
1af1ac91 4278 goto set_identity_unlock;
b927a3ce 4279 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
1af1ac91
DH		 4280set_identity_unlock:
4281 mutex_unlock(&kvm->lock);
b927a3ce
SY		 4282 break;
4283 }
1fe779f8
CO		 4284 case KVM_SET_NR_MMU_PAGES:
4285 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
1fe779f8
CO		 4286 break;
4287 case KVM_GET_NR_MMU_PAGES:
4288 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4289 break;
3ddea128 4290 case KVM_CREATE_IRQCHIP: {
3ddea128 4291 mutex_lock(&kvm->lock);
09941366 4292
3ddea128 4293 r = -EEXIST;
35e6eaa3 4294 if (irqchip_in_kernel(kvm))
3ddea128 4295 goto create_irqchip_unlock;
09941366 4296
3e515705 4297 r = -EINVAL;
557abc40 4298 if (kvm->created_vcpus)
3e515705 4299 goto create_irqchip_unlock;
09941366
RK		 4300
4301 r = kvm_pic_init(kvm);
4302 if (r)
3ddea128 4303 goto create_irqchip_unlock;
09941366
RK		 4304
4305 r = kvm_ioapic_init(kvm);
4306 if (r) {
09941366 4307 kvm_pic_destroy(kvm);
3ddea128 4308 goto create_irqchip_unlock;
09941366
RK		 4309 }
4310
399ec807
AK		 4311 r = kvm_setup_default_irq_routing(kvm);
4312 if (r) {
72bb2fcd 4313 kvm_ioapic_destroy(kvm);
09941366 4314 kvm_pic_destroy(kvm);
71ba994c 4315 goto create_irqchip_unlock;
399ec807 4316 }
49776faf 4317 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
71ba994c 4318 smp_wmb();
49776faf 4319 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
3ddea128
MT		 4320 create_irqchip_unlock:
4321 mutex_unlock(&kvm->lock);
1fe779f8 4322 break;
3ddea128 4323 }
7837699f 4324 case KVM_CREATE_PIT:
c5ff41ce
JK		 4325 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4326 goto create_pit;
4327 case KVM_CREATE_PIT2:
4328 r = -EFAULT;
4329 if (copy_from_user(&u.pit_config, argp,
4330 sizeof(struct kvm_pit_config)))
4331 goto out;
4332 create_pit:
250715a6 4333 mutex_lock(&kvm->lock);
269e05e4
AK		 4334 r = -EEXIST;
4335 if (kvm->arch.vpit)
4336 goto create_pit_unlock;
7837699f 4337 r = -ENOMEM;
c5ff41ce 4338 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
7837699f
SY		 4339 if (kvm->arch.vpit)
4340 r = 0;
269e05e4 4341 create_pit_unlock:
250715a6 4342 mutex_unlock(&kvm->lock);
7837699f 4343 break;
1fe779f8
CO		 4344 case KVM_GET_IRQCHIP: {
4345 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
ff5c2c03 4346 struct kvm_irqchip *chip;
1fe779f8 4347
ff5c2c03
SL		 4348 chip = memdup_user(argp, sizeof(*chip));
4349 if (IS_ERR(chip)) {
4350 r = PTR_ERR(chip);
1fe779f8 4351 goto out;
ff5c2c03
SL		 4352 }
4353
1fe779f8 4354 r = -ENXIO;
826da321 4355 if (!irqchip_kernel(kvm))
f0d66275
DH		 4356 goto get_irqchip_out;
4357 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
1fe779f8 4358 if (r)
f0d66275 4359 goto get_irqchip_out;
1fe779f8 4360 r = -EFAULT;
f0d66275
DH		 4361 if (copy_to_user(argp, chip, sizeof *chip))
4362 goto get_irqchip_out;
1fe779f8 4363 r = 0;
f0d66275
DH		 4364 get_irqchip_out:
4365 kfree(chip);
1fe779f8
CO		 4366 break;
4367 }
4368 case KVM_SET_IRQCHIP: {
4369 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
ff5c2c03 4370 struct kvm_irqchip *chip;
1fe779f8 4371
ff5c2c03
SL		 4372 chip = memdup_user(argp, sizeof(*chip));
4373 if (IS_ERR(chip)) {
4374 r = PTR_ERR(chip);
1fe779f8 4375 goto out;
ff5c2c03
SL		 4376 }
4377
1fe779f8 4378 r = -ENXIO;
826da321 4379 if (!irqchip_kernel(kvm))
f0d66275
DH		 4380 goto set_irqchip_out;
4381 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
1fe779f8 4382 if (r)
f0d66275 4383 goto set_irqchip_out;
1fe779f8 4384 r = 0;
f0d66275
DH		 4385 set_irqchip_out:
4386 kfree(chip);
1fe779f8
CO		 4387 break;
4388 }
e0f63cb9 4389 case KVM_GET_PIT: {
e0f63cb9 4390 r = -EFAULT;
f0d66275 4391 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
e0f63cb9
SY		 4392 goto out;
4393 r = -ENXIO;
4394 if (!kvm->arch.vpit)
4395 goto out;
f0d66275 4396 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
e0f63cb9
SY		 4397 if (r)
4398 goto out;
4399 r = -EFAULT;
f0d66275 4400 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
e0f63cb9
SY		 4401 goto out;
4402 r = 0;
4403 break;
4404 }
4405 case KVM_SET_PIT: {
e0f63cb9 4406 r = -EFAULT;
f0d66275 4407 if (copy_from_user(&u.ps, argp, sizeof u.ps))
e0f63cb9
SY		 4408 goto out;
4409 r = -ENXIO;
4410 if (!kvm->arch.vpit)
4411 goto out;
f0d66275 4412 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
e0f63cb9
SY		 4413 break;
4414 }
e9f42757
BK		 4415 case KVM_GET_PIT2: {
4416 r = -ENXIO;
4417 if (!kvm->arch.vpit)
4418 goto out;
4419 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4420 if (r)
4421 goto out;
4422 r = -EFAULT;
4423 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4424 goto out;
4425 r = 0;
4426 break;
4427 }
4428 case KVM_SET_PIT2: {
4429 r = -EFAULT;
4430 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4431 goto out;
4432 r = -ENXIO;
4433 if (!kvm->arch.vpit)
4434 goto out;
4435 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
e9f42757
BK		 4436 break;
4437 }
52d939a0
MT		 4438 case KVM_REINJECT_CONTROL: {
4439 struct kvm_reinject_control control;
4440 r = -EFAULT;
4441 if (copy_from_user(&control, argp, sizeof(control)))
4442 goto out;
4443 r = kvm_vm_ioctl_reinject(kvm, &control);
52d939a0
MT		 4444 break;
4445 }
d71ba788
PB		 4446 case KVM_SET_BOOT_CPU_ID:
4447 r = 0;
4448 mutex_lock(&kvm->lock);
557abc40 4449 if (kvm->created_vcpus)
d71ba788
PB		 4450 r = -EBUSY;
4451 else
4452 kvm->arch.bsp_vcpu_id = arg;
4453 mutex_unlock(&kvm->lock);
4454 break;
ffde22ac 4455 case KVM_XEN_HVM_CONFIG: {
51776043 4456 struct kvm_xen_hvm_config xhc;
ffde22ac 4457 r = -EFAULT;
51776043 4458 if (copy_from_user(&xhc, argp, sizeof(xhc)))
ffde22ac
ES		 4459 goto out;
4460 r = -EINVAL;
51776043 4461 if (xhc.flags)
ffde22ac 4462 goto out;
51776043 4463 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
ffde22ac
ES		 4464 r = 0;
4465 break;
4466 }
afbcf7ab 4467 case KVM_SET_CLOCK: {
afbcf7ab
GC		 4468 struct kvm_clock_data user_ns;
4469 u64 now_ns;
afbcf7ab
GC		 4470
4471 r = -EFAULT;
4472 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4473 goto out;
4474
4475 r = -EINVAL;
4476 if (user_ns.flags)
4477 goto out;
4478
4479 r = 0;
0bc48bea
RK		 4480 /*
4481 * TODO: userspace has to take care of races with VCPU_RUN, so
4482 * kvm_gen_update_masterclock() can be cut down to locked
4483 * pvclock_update_vm_gtod_copy().
4484 */
4485 kvm_gen_update_masterclock(kvm);
e891a32e 4486 now_ns = get_kvmclock_ns(kvm);
108b249c 4487 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
0bc48bea 4488 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
afbcf7ab
GC		 4489 break;
4490 }
4491 case KVM_GET_CLOCK: {
afbcf7ab
GC		 4492 struct kvm_clock_data user_ns;
4493 u64 now_ns;
4494
e891a32e 4495 now_ns = get_kvmclock_ns(kvm);
108b249c 4496 user_ns.clock = now_ns;
e3fd9a93 4497 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
97e69aa6 4498 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
afbcf7ab
GC		 4499
4500 r = -EFAULT;
4501 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4502 goto out;
4503 r = 0;
4504 break;
4505 }
90de4a18
NA		 4506 case KVM_ENABLE_CAP: {
4507 struct kvm_enable_cap cap;
afbcf7ab 4508
90de4a18
NA		 4509 r = -EFAULT;
4510 if (copy_from_user(&cap, argp, sizeof(cap)))
4511 goto out;
4512 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4513 break;
4514 }
5acc5c06
BS		 4515 case KVM_MEMORY_ENCRYPT_OP: {
4516 r = -ENOTTY;
4517 if (kvm_x86_ops->mem_enc_op)
4518 r = kvm_x86_ops->mem_enc_op(kvm, argp);
4519 break;
4520 }
69eaedee
BS		 4521 case KVM_MEMORY_ENCRYPT_REG_REGION: {
4522 struct kvm_enc_region region;
4523
4524 r = -EFAULT;
4525 if (copy_from_user(&region, argp, sizeof(region)))
4526 goto out;
4527
4528 r = -ENOTTY;
4529 if (kvm_x86_ops->mem_enc_reg_region)
4530 r = kvm_x86_ops->mem_enc_reg_region(kvm, &region);
4531 break;
4532 }
4533 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
4534 struct kvm_enc_region region;
4535
4536 r = -EFAULT;
4537 if (copy_from_user(&region, argp, sizeof(region)))
4538 goto out;
4539
4540 r = -ENOTTY;
4541 if (kvm_x86_ops->mem_enc_unreg_region)
4542 r = kvm_x86_ops->mem_enc_unreg_region(kvm, &region);
4543 break;
4544 }
faeb7833
RK		 4545 case KVM_HYPERV_EVENTFD: {
4546 struct kvm_hyperv_eventfd hvevfd;
4547
4548 r = -EFAULT;
4549 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
4550 goto out;
4551 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
4552 break;
4553 }
1fe779f8 4554 default:
ad6260da 4555 r = -ENOTTY;
1fe779f8
CO		 4556 }
4557out:
4558 return r;
4559}
4560
a16b043c 4561static void kvm_init_msr_list(void)
043405e1
CO		 4562{
4563 u32 dummy[2];
4564 unsigned i, j;
4565
62ef68bb 4566 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
043405e1
CO		 4567 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4568 continue;
93c4adc7
PB		 4569
4570 /*
4571 * Even MSRs that are valid in the host may not be exposed
9dbe6cf9 4572 * to the guests in some cases.
93c4adc7
PB		 4573 */
4574 switch (msrs_to_save[i]) {
4575 case MSR_IA32_BNDCFGS:
4576 if (!kvm_x86_ops->mpx_supported())
4577 continue;
4578 break;
9dbe6cf9
PB		 4579 case MSR_TSC_AUX:
4580 if (!kvm_x86_ops->rdtscp_supported())
4581 continue;
4582 break;
93c4adc7
PB		 4583 default:
4584 break;
4585 }
4586
043405e1
CO		 4587 if (j < i)
4588 msrs_to_save[j] = msrs_to_save[i];
4589 j++;
4590 }
4591 num_msrs_to_save = j;
62ef68bb
PB		 4592
4593 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4594 switch (emulated_msrs[i]) {
6d396b55
PB		 4595 case MSR_IA32_SMBASE:
4596 if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4597 continue;
4598 break;
62ef68bb
PB		 4599 default:
4600 break;
4601 }
4602
4603 if (j < i)
4604 emulated_msrs[j] = emulated_msrs[i];
4605 j++;
4606 }
4607 num_emulated_msrs = j;
801e459a
TL		 4608
4609 for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
4610 struct kvm_msr_entry msr;
4611
4612 msr.index = msr_based_features[i];
66421c1e 4613 if (kvm_get_msr_feature(&msr))
801e459a
TL		 4614 continue;
4615
4616 if (j < i)
4617 msr_based_features[j] = msr_based_features[i];
4618 j++;
4619 }
4620 num_msr_based_features = j;
043405e1
CO		 4621}
4622
bda9020e
MT		 4623static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4624 const void *v)
bbd9b64e 4625{
70252a10
AK		 4626 int handled = 0;
4627 int n;
4628
4629 do {
4630 n = min(len, 8);
bce87cce 4631 if (!(lapic_in_kernel(vcpu) &&
e32edf4f
NN		 4632 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4633 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
70252a10
AK		 4634 break;
4635 handled += n;
4636 addr += n;
4637 len -= n;
4638 v += n;
4639 } while (len);
bbd9b64e 4640
70252a10 4641 return handled;
bbd9b64e
CO		 4642}
4643
bda9020e 4644static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
bbd9b64e 4645{
70252a10
AK		 4646 int handled = 0;
4647 int n;
4648
4649 do {
4650 n = min(len, 8);
bce87cce 4651 if (!(lapic_in_kernel(vcpu) &&
e32edf4f
NN		 4652 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4653 addr, n, v))
4654 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
70252a10 4655 break;
e39d200f 4656 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
70252a10
AK		 4657 handled += n;
4658 addr += n;
4659 len -= n;
4660 v += n;
4661 } while (len);
bbd9b64e 4662
70252a10 4663 return handled;
bbd9b64e
CO		 4664}
4665
2dafc6c2
GN		 4666static void kvm_set_segment(struct kvm_vcpu *vcpu,
4667 struct kvm_segment *var, int seg)
4668{
4669 kvm_x86_ops->set_segment(vcpu, var, seg);
4670}
4671
4672void kvm_get_segment(struct kvm_vcpu *vcpu,
4673 struct kvm_segment *var, int seg)
4674{
4675 kvm_x86_ops->get_segment(vcpu, var, seg);
4676}
4677
54987b7a
PB		 4678gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4679 struct x86_exception *exception)
02f59dc9
JR		 4680{
4681 gpa_t t_gpa;
02f59dc9
JR		 4682
4683 BUG_ON(!mmu_is_nested(vcpu));
4684
4685 /* NPT walks are always user-walks */
4686 access |= PFERR_USER_MASK;
54987b7a 4687 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
02f59dc9
JR		 4688
4689 return t_gpa;
4690}
4691
ab9ae313
AK		 4692gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4693 struct x86_exception *exception)
1871c602
GN		 4694{
4695 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
ab9ae313 4696 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 4697}
4698
ab9ae313
AK		 4699 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4700 struct x86_exception *exception)
1871c602
GN		 4701{
4702 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4703 access |= PFERR_FETCH_MASK;
ab9ae313 4704 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 4705}
4706
ab9ae313
AK		 4707gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4708 struct x86_exception *exception)
1871c602
GN		 4709{
4710 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4711 access |= PFERR_WRITE_MASK;
ab9ae313 4712 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
1871c602
GN		 4713}
4714
4715/* uses this to access any guest's mapped memory without checking CPL */
ab9ae313
AK		 4716gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4717 struct x86_exception *exception)
1871c602 4718{
ab9ae313 4719 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
1871c602
GN		 4720}
4721
4722static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4723 struct kvm_vcpu *vcpu, u32 access,
bcc55cba 4724 struct x86_exception *exception)
bbd9b64e
CO		 4725{
4726 void *data = val;
10589a46 4727 int r = X86EMUL_CONTINUE;
bbd9b64e
CO		 4728
4729 while (bytes) {
14dfe855 4730 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
ab9ae313 4731 exception);
bbd9b64e 4732 unsigned offset = addr & (PAGE_SIZE-1);
77c2002e 4733 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
bbd9b64e
CO		 4734 int ret;
4735
bcc55cba 4736 if (gpa == UNMAPPED_GVA)
ab9ae313 4737 return X86EMUL_PROPAGATE_FAULT;
54bf36aa
PB		 4738 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4739 offset, toread);
10589a46 4740 if (ret < 0) {
c3cd7ffa 4741 r = X86EMUL_IO_NEEDED;
10589a46
MT		 4742 goto out;
4743 }
bbd9b64e 4744
77c2002e
IE		 4745 bytes -= toread;
4746 data += toread;
4747 addr += toread;
bbd9b64e 4748 }
10589a46 4749out:
10589a46 4750 return r;
bbd9b64e 4751}
77c2002e 4752
1871c602 4753/* used for instruction fetching */
0f65dd70
AK		 4754static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4755 gva_t addr, void *val, unsigned int bytes,
bcc55cba 4756 struct x86_exception *exception)
1871c602 4757{
0f65dd70 4758 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
1871c602 4759 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
44583cba
PB		 4760 unsigned offset;
4761 int ret;
0f65dd70 4762
44583cba
PB		 4763 /* Inline kvm_read_guest_virt_helper for speed. */
4764 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4765 exception);
4766 if (unlikely(gpa == UNMAPPED_GVA))
4767 return X86EMUL_PROPAGATE_FAULT;
4768
4769 offset = addr & (PAGE_SIZE-1);
4770 if (WARN_ON(offset + bytes > PAGE_SIZE))
4771 bytes = (unsigned)PAGE_SIZE - offset;
54bf36aa
PB		 4772 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4773 offset, bytes);
44583cba
PB		 4774 if (unlikely(ret < 0))
4775 return X86EMUL_IO_NEEDED;
4776
4777 return X86EMUL_CONTINUE;
1871c602
GN		 4778}
4779
064aea77 4780int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
0f65dd70 4781 gva_t addr, void *val, unsigned int bytes,
bcc55cba 4782 struct x86_exception *exception)
1871c602 4783{
0f65dd70 4784 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
1871c602 4785 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
0f65dd70 4786
1871c602 4787 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
bcc55cba 4788 exception);
1871c602 4789}
064aea77 4790EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
1871c602 4791
0f65dd70
AK		 4792static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4793 gva_t addr, void *val, unsigned int bytes,
bcc55cba 4794 struct x86_exception *exception)
1871c602 4795{
0f65dd70 4796 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
bcc55cba 4797 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
1871c602
GN		 4798}
4799
7a036a6f
RK		 4800static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4801 unsigned long addr, void *val, unsigned int bytes)
4802{
4803 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4804 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4805
4806 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4807}
4808
6a4d7550 4809int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
0f65dd70 4810 gva_t addr, void *val,
2dafc6c2 4811 unsigned int bytes,
bcc55cba 4812 struct x86_exception *exception)
77c2002e 4813{
0f65dd70 4814 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
77c2002e
IE		 4815 void *data = val;
4816 int r = X86EMUL_CONTINUE;
4817
4818 while (bytes) {
14dfe855
JR		 4819 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4820 PFERR_WRITE_MASK,
ab9ae313 4821 exception);
77c2002e
IE		 4822 unsigned offset = addr & (PAGE_SIZE-1);
4823 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4824 int ret;
4825
bcc55cba 4826 if (gpa == UNMAPPED_GVA)
ab9ae313 4827 return X86EMUL_PROPAGATE_FAULT;
54bf36aa 4828 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
77c2002e 4829 if (ret < 0) {
c3cd7ffa 4830 r = X86EMUL_IO_NEEDED;
77c2002e
IE		 4831 goto out;
4832 }
4833
4834 bytes -= towrite;
4835 data += towrite;
4836 addr += towrite;
4837 }
4838out:
4839 return r;
4840}
6a4d7550 4841EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
77c2002e 4842
0f89b207
TL		 4843static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4844 gpa_t gpa, bool write)
4845{
4846 /* For APIC access vmexit */
4847 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4848 return 1;
4849
4850 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
4851 trace_vcpu_match_mmio(gva, gpa, write, true);
4852 return 1;
4853 }
4854
4855 return 0;
4856}
4857
af7cc7d1
XG		 4858static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4859 gpa_t *gpa, struct x86_exception *exception,
4860 bool write)
4861{
97d64b78
AK		 4862 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4863 | (write ? PFERR_WRITE_MASK : 0);
af7cc7d1 4864
be94f6b7
HH		 4865 /*
4866 * currently PKRU is only applied to ept enabled guest so
4867 * there is no pkey in EPT page table for L1 guest or EPT
4868 * shadow page table for L2 guest.
4869 */
97d64b78 4870 if (vcpu_match_mmio_gva(vcpu, gva)
97ec8c06 4871 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
be94f6b7 4872 vcpu->arch.access, 0, access)) {
bebb106a
XG		 4873 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4874 (gva & (PAGE_SIZE - 1));
4f022648 4875 trace_vcpu_match_mmio(gva, *gpa, write, false);
bebb106a
XG		 4876 return 1;
4877 }
4878
af7cc7d1
XG		 4879 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4880
4881 if (*gpa == UNMAPPED_GVA)
4882 return -1;
4883
0f89b207 4884 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
af7cc7d1
XG		 4885}
4886
3200f405 4887int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
bcc55cba 4888 const void *val, int bytes)
bbd9b64e
CO		 4889{
4890 int ret;
4891
54bf36aa 4892 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
9f811285 4893 if (ret < 0)
bbd9b64e 4894 return 0;
0eb05bf2 4895 kvm_page_track_write(vcpu, gpa, val, bytes);
bbd9b64e
CO		 4896 return 1;
4897}
4898
77d197b2
XG		 4899struct read_write_emulator_ops {
4900 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4901 int bytes);
4902 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4903 void *val, int bytes);
4904 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4905 int bytes, void *val);
4906 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4907 void *val, int bytes);
4908 bool write;
4909};
4910
4911static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4912{
4913 if (vcpu->mmio_read_completed) {
77d197b2 4914 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
e39d200f 4915 vcpu->mmio_fragments[0].gpa, val);
77d197b2
XG		 4916 vcpu->mmio_read_completed = 0;
4917 return 1;
4918 }
4919
4920 return 0;
4921}
4922
4923static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4924 void *val, int bytes)
4925{
54bf36aa 4926 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
77d197b2
XG		 4927}
4928
4929static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4930 void *val, int bytes)
4931{
4932 return emulator_write_phys(vcpu, gpa, val, bytes);
4933}
4934
4935static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4936{
e39d200f 4937 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
77d197b2
XG		 4938 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4939}
4940
4941static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4942 void *val, int bytes)
4943{
e39d200f 4944 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
77d197b2
XG		 4945 return X86EMUL_IO_NEEDED;
4946}
4947
4948static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4949 void *val, int bytes)
4950{
f78146b0
AK		 4951 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4952
87da7e66 4953 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
77d197b2
XG		 4954 return X86EMUL_CONTINUE;
4955}
4956
0fbe9b0b 4957static const struct read_write_emulator_ops read_emultor = {
77d197b2
XG		 4958 .read_write_prepare = read_prepare,
4959 .read_write_emulate = read_emulate,
4960 .read_write_mmio = vcpu_mmio_read,
4961 .read_write_exit_mmio = read_exit_mmio,
4962};
4963
0fbe9b0b 4964static const struct read_write_emulator_ops write_emultor = {
77d197b2
XG		 4965 .read_write_emulate = write_emulate,
4966 .read_write_mmio = write_mmio,
4967 .read_write_exit_mmio = write_exit_mmio,
4968 .write = true,
4969};
4970
22388a3c
XG		 4971static int emulator_read_write_onepage(unsigned long addr, void *val,
4972 unsigned int bytes,
4973 struct x86_exception *exception,
4974 struct kvm_vcpu *vcpu,
0fbe9b0b 4975 const struct read_write_emulator_ops *ops)
bbd9b64e 4976{
af7cc7d1
XG		 4977 gpa_t gpa;
4978 int handled, ret;
22388a3c 4979 bool write = ops->write;
f78146b0 4980 struct kvm_mmio_fragment *frag;
0f89b207
TL		 4981 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4982
4983 /*
4984 * If the exit was due to a NPF we may already have a GPA.
4985 * If the GPA is present, use it to avoid the GVA to GPA table walk.
4986 * Note, this cannot be used on string operations since string
4987 * operation using rep will only have the initial GPA from the NPF
4988 * occurred.
4989 */
4990 if (vcpu->arch.gpa_available &&
4991 emulator_can_use_gpa(ctxt) &&
618232e2
BS		 4992 (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
4993 gpa = vcpu->arch.gpa_val;
4994 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
4995 } else {
4996 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4997 if (ret < 0)
4998 return X86EMUL_PROPAGATE_FAULT;
0f89b207 4999 }
10589a46 5000
618232e2 5001 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
bbd9b64e
CO		 5002 return X86EMUL_CONTINUE;
5003
bbd9b64e
CO		 5004 /*
5005 * Is this MMIO handled locally?
5006 */
22388a3c 5007 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
70252a10 5008 if (handled == bytes)
bbd9b64e 5009 return X86EMUL_CONTINUE;
bbd9b64e 5010
70252a10
AK		 5011 gpa += handled;
5012 bytes -= handled;
5013 val += handled;
5014
87da7e66
XG		 5015 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5016 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5017 frag->gpa = gpa;
5018 frag->data = val;
5019 frag->len = bytes;
f78146b0 5020 return X86EMUL_CONTINUE;
bbd9b64e
CO		 5021}
5022
52eb5a6d
XL		 5023static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5024 unsigned long addr,
22388a3c
XG		 5025 void *val, unsigned int bytes,
5026 struct x86_exception *exception,
0fbe9b0b 5027 const struct read_write_emulator_ops *ops)
bbd9b64e 5028{
0f65dd70 5029 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
f78146b0
AK		 5030 gpa_t gpa;
5031 int rc;
5032
5033 if (ops->read_write_prepare &&
5034 ops->read_write_prepare(vcpu, val, bytes))
5035 return X86EMUL_CONTINUE;
5036
5037 vcpu->mmio_nr_fragments = 0;
0f65dd70 5038
bbd9b64e
CO		 5039 /* Crossing a page boundary? */
5040 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
f78146b0 5041 int now;
bbd9b64e
CO		 5042
5043 now = -addr & ~PAGE_MASK;
22388a3c
XG		 5044 rc = emulator_read_write_onepage(addr, val, now, exception,
5045 vcpu, ops);
5046
bbd9b64e
CO		 5047 if (rc != X86EMUL_CONTINUE)
5048 return rc;
5049 addr += now;
bac15531
NA		 5050 if (ctxt->mode != X86EMUL_MODE_PROT64)
5051 addr = (u32)addr;
bbd9b64e
CO		 5052 val += now;
5053 bytes -= now;
5054 }
22388a3c 5055
f78146b0
AK		 5056 rc = emulator_read_write_onepage(addr, val, bytes, exception,
5057 vcpu, ops);
5058 if (rc != X86EMUL_CONTINUE)
5059 return rc;
5060
5061 if (!vcpu->mmio_nr_fragments)
5062 return rc;
5063
5064 gpa = vcpu->mmio_fragments[0].gpa;
5065
5066 vcpu->mmio_needed = 1;
5067 vcpu->mmio_cur_fragment = 0;
5068
87da7e66 5069 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
f78146b0
AK		 5070 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5071 vcpu->run->exit_reason = KVM_EXIT_MMIO;
5072 vcpu->run->mmio.phys_addr = gpa;
5073
5074 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
22388a3c
XG		 5075}
5076
5077static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5078 unsigned long addr,
5079 void *val,
5080 unsigned int bytes,
5081 struct x86_exception *exception)
5082{
5083 return emulator_read_write(ctxt, addr, val, bytes,
5084 exception, &read_emultor);
5085}
5086
52eb5a6d 5087static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
22388a3c
XG		 5088 unsigned long addr,
5089 const void *val,
5090 unsigned int bytes,
5091 struct x86_exception *exception)
5092{
5093 return emulator_read_write(ctxt, addr, (void *)val, bytes,
5094 exception, &write_emultor);
bbd9b64e 5095}
bbd9b64e 5096
daea3e73
AK		 5097#define CMPXCHG_TYPE(t, ptr, old, new) \
5098 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5099
5100#ifdef CONFIG_X86_64
5101# define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5102#else
5103# define CMPXCHG64(ptr, old, new) \
9749a6c0 5104 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
daea3e73
AK		 5105#endif
5106
0f65dd70
AK		 5107static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5108 unsigned long addr,
bbd9b64e
CO		 5109 const void *old,
5110 const void *new,
5111 unsigned int bytes,
0f65dd70 5112 struct x86_exception *exception)
bbd9b64e 5113{
0f65dd70 5114 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
daea3e73
AK		 5115 gpa_t gpa;
5116 struct page *page;
5117 char *kaddr;
5118 bool exchanged;
2bacc55c 5119
daea3e73
AK		 5120 /* guests cmpxchg8b have to be emulated atomically */
5121 if (bytes > 8 || (bytes & (bytes - 1)))
5122 goto emul_write;
10589a46 5123
daea3e73 5124 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
2bacc55c 5125
daea3e73
AK		 5126 if (gpa == UNMAPPED_GVA ||
5127 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5128 goto emul_write;
2bacc55c 5129
daea3e73
AK		 5130 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5131 goto emul_write;
72dc67a6 5132
54bf36aa 5133 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
32cad84f 5134 if (is_error_page(page))
c19b8bd6 5135 goto emul_write;
72dc67a6 5136
8fd75e12 5137 kaddr = kmap_atomic(page);
daea3e73
AK		 5138 kaddr += offset_in_page(gpa);
5139 switch (bytes) {
5140 case 1:
5141 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5142 break;
5143 case 2:
5144 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5145 break;
5146 case 4:
5147 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5148 break;
5149 case 8:
5150 exchanged = CMPXCHG64(kaddr, old, new);
5151 break;
5152 default:
5153 BUG();
2bacc55c 5154 }
8fd75e12 5155 kunmap_atomic(kaddr);
daea3e73
AK		 5156 kvm_release_page_dirty(page);
5157
5158 if (!exchanged)
5159 return X86EMUL_CMPXCHG_FAILED;
5160
54bf36aa 5161 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
0eb05bf2 5162 kvm_page_track_write(vcpu, gpa, new, bytes);
8f6abd06
GN		 5163
5164 return X86EMUL_CONTINUE;
4a5f48f6 5165
3200f405 5166emul_write:
daea3e73 5167 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
2bacc55c 5168
0f65dd70 5169 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
bbd9b64e
CO		 5170}
5171
cf8f70bf
GN		 5172static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5173{
cbfc6c91 5174 int r = 0, i;
cf8f70bf 5175
cbfc6c91
WL		 5176 for (i = 0; i < vcpu->arch.pio.count; i++) {
5177 if (vcpu->arch.pio.in)
5178 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5179 vcpu->arch.pio.size, pd);
5180 else
5181 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5182 vcpu->arch.pio.port, vcpu->arch.pio.size,
5183 pd);
5184 if (r)
5185 break;
5186 pd += vcpu->arch.pio.size;
5187 }
cf8f70bf
GN		 5188 return r;
5189}
5190
6f6fbe98
XG		 5191static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5192 unsigned short port, void *val,
5193 unsigned int count, bool in)
cf8f70bf 5194{
cf8f70bf 5195 vcpu->arch.pio.port = port;
6f6fbe98 5196 vcpu->arch.pio.in = in;
7972995b 5197 vcpu->arch.pio.count = count;
cf8f70bf
GN		 5198 vcpu->arch.pio.size = size;
5199
5200 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
7972995b 5201 vcpu->arch.pio.count = 0;
cf8f70bf
GN		 5202 return 1;
5203 }
5204
5205 vcpu->run->exit_reason = KVM_EXIT_IO;
6f6fbe98 5206 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
cf8f70bf
GN		 5207 vcpu->run->io.size = size;
5208 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5209 vcpu->run->io.count = count;
5210 vcpu->run->io.port = port;
5211
5212 return 0;
5213}
5214
6f6fbe98
XG		 5215static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5216 int size, unsigned short port, void *val,
5217 unsigned int count)
cf8f70bf 5218{
ca1d4a9e 5219 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6f6fbe98 5220 int ret;
ca1d4a9e 5221
6f6fbe98
XG		 5222 if (vcpu->arch.pio.count)
5223 goto data_avail;
cf8f70bf 5224
cbfc6c91
WL		 5225 memset(vcpu->arch.pio_data, 0, size * count);
5226
6f6fbe98
XG		 5227 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5228 if (ret) {
5229data_avail:
5230 memcpy(val, vcpu->arch.pio_data, size * count);
1171903d 5231 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
7972995b 5232 vcpu->arch.pio.count = 0;
cf8f70bf
GN		 5233 return 1;
5234 }
5235
cf8f70bf
GN		 5236 return 0;
5237}
5238
6f6fbe98
XG		 5239static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5240 int size, unsigned short port,
5241 const void *val, unsigned int count)
5242{
5243 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5244
5245 memcpy(vcpu->arch.pio_data, val, size * count);
1171903d 5246 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6f6fbe98
XG		 5247 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5248}
5249
bbd9b64e
CO		 5250static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5251{
5252 return kvm_x86_ops->get_segment_base(vcpu, seg);
5253}
5254
3cb16fe7 5255static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
bbd9b64e 5256{
3cb16fe7 5257 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
bbd9b64e
CO		 5258}
5259
ae6a2375 5260static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
f5f48ee1
SY		 5261{
5262 if (!need_emulate_wbinvd(vcpu))
5263 return X86EMUL_CONTINUE;
5264
5265 if (kvm_x86_ops->has_wbinvd_exit()) {
2eec7343
JK		 5266 int cpu = get_cpu();
5267
5268 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
f5f48ee1
SY		 5269 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5270 wbinvd_ipi, NULL, 1);
2eec7343 5271 put_cpu();
f5f48ee1 5272 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
2eec7343
JK		 5273 } else
5274 wbinvd();
f5f48ee1
SY		 5275 return X86EMUL_CONTINUE;
5276}
5cb56059
JS		 5277
5278int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5279{
6affcbed
KH		 5280 kvm_emulate_wbinvd_noskip(vcpu);
5281 return kvm_skip_emulated_instruction(vcpu);
5cb56059 5282}
f5f48ee1
SY		 5283EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5284
5cb56059
JS		 5285
5286
bcaf5cc5
AK		 5287static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5288{
5cb56059 5289 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
bcaf5cc5
AK		 5290}
5291
52eb5a6d
XL		 5292static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5293 unsigned long *dest)
bbd9b64e 5294{
16f8a6f9 5295 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
bbd9b64e
CO		 5296}
5297
52eb5a6d
XL		 5298static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5299 unsigned long value)
bbd9b64e 5300{
338dbc97 5301
717746e3 5302 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
bbd9b64e
CO		 5303}
5304
52a46617 5305static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5fdbf976 5306{
52a46617 5307 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5fdbf976
MT		 5308}
5309
717746e3 5310static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
bbd9b64e 5311{
717746e3 5312 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
52a46617
GN		 5313 unsigned long value;
5314
5315 switch (cr) {
5316 case 0:
5317 value = kvm_read_cr0(vcpu);
5318 break;
5319 case 2:
5320 value = vcpu->arch.cr2;
5321 break;
5322 case 3:
9f8fe504 5323 value = kvm_read_cr3(vcpu);
52a46617
GN		 5324 break;
5325 case 4:
5326 value = kvm_read_cr4(vcpu);
5327 break;
5328 case 8:
5329 value = kvm_get_cr8(vcpu);
5330 break;
5331 default:
a737f256 5332 kvm_err("%s: unexpected cr %u\n", __func__, cr);
52a46617
GN		 5333 return 0;
5334 }
5335
5336 return value;
5337}
5338
717746e3 5339static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
52a46617 5340{
717746e3 5341 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
0f12244f
GN		 5342 int res = 0;
5343
52a46617
GN		 5344 switch (cr) {
5345 case 0:
49a9b07e 5346 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
52a46617
GN		 5347 break;
5348 case 2:
5349 vcpu->arch.cr2 = val;
5350 break;
5351 case 3:
2390218b 5352 res = kvm_set_cr3(vcpu, val);
52a46617
GN		 5353 break;
5354 case 4:
a83b29c6 5355 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
52a46617
GN		 5356 break;
5357 case 8:
eea1cff9 5358 res = kvm_set_cr8(vcpu, val);
52a46617
GN		 5359 break;
5360 default:
a737f256 5361 kvm_err("%s: unexpected cr %u\n", __func__, cr);
0f12244f 5362 res = -1;
52a46617 5363 }
0f12244f
GN		 5364
5365 return res;
52a46617
GN		 5366}
5367
717746e3 5368static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
9c537244 5369{
717746e3 5370 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
9c537244
GN		 5371}
5372
4bff1e86 5373static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
2dafc6c2 5374{
4bff1e86 5375 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
2dafc6c2
GN		 5376}
5377
4bff1e86 5378static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
160ce1f1 5379{
4bff1e86 5380 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
160ce1f1
MG		 5381}
5382
1ac9d0cf
AK		 5383static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5384{
5385 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5386}
5387
5388static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5389{
5390 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5391}
5392
4bff1e86
AK		 5393static unsigned long emulator_get_cached_segment_base(
5394 struct x86_emulate_ctxt *ctxt, int seg)
5951c442 5395{
4bff1e86 5396 return get_segment_base(emul_to_vcpu(ctxt), seg);
5951c442
GN		 5397}
5398
1aa36616
AK		 5399static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5400 struct desc_struct *desc, u32 *base3,
5401 int seg)
2dafc6c2
GN		 5402{
5403 struct kvm_segment var;
5404
4bff1e86 5405 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
1aa36616 5406 *selector = var.selector;
2dafc6c2 5407
378a8b09
GN		 5408 if (var.unusable) {
5409 memset(desc, 0, sizeof(*desc));
f0367ee1
RK		 5410 if (base3)
5411 *base3 = 0;
2dafc6c2 5412 return false;
378a8b09 5413 }
2dafc6c2
GN		 5414
5415 if (var.g)
5416 var.limit >>= 12;
5417 set_desc_limit(desc, var.limit);
5418 set_desc_base(desc, (unsigned long)var.base);
5601d05b
GN		 5419#ifdef CONFIG_X86_64
5420 if (base3)
5421 *base3 = var.base >> 32;
5422#endif
2dafc6c2
GN		 5423 desc->type = var.type;
5424 desc->s = var.s;
5425 desc->dpl = var.dpl;
5426 desc->p = var.present;
5427 desc->avl = var.avl;
5428 desc->l = var.l;
5429 desc->d = var.db;
5430 desc->g = var.g;
5431
5432 return true;
5433}
5434
1aa36616
AK		 5435static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5436 struct desc_struct *desc, u32 base3,
5437 int seg)
2dafc6c2 5438{
4bff1e86 5439 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
2dafc6c2
GN		 5440 struct kvm_segment var;
5441
1aa36616 5442 var.selector = selector;
2dafc6c2 5443 var.base = get_desc_base(desc);
5601d05b
GN		 5444#ifdef CONFIG_X86_64
5445 var.base |= ((u64)base3) << 32;
5446#endif
2dafc6c2
GN		 5447 var.limit = get_desc_limit(desc);
5448 if (desc->g)
5449 var.limit = (var.limit << 12) | 0xfff;
5450 var.type = desc->type;
2dafc6c2
GN		 5451 var.dpl = desc->dpl;
5452 var.db = desc->d;
5453 var.s = desc->s;
5454 var.l = desc->l;
5455 var.g = desc->g;
5456 var.avl = desc->avl;
5457 var.present = desc->p;
5458 var.unusable = !var.present;
5459 var.padding = 0;
5460
5461 kvm_set_segment(vcpu, &var, seg);
5462 return;
5463}
5464
717746e3
AK		 5465static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5466 u32 msr_index, u64 *pdata)
5467{
609e36d3
PB		 5468 struct msr_data msr;
5469 int r;
5470
5471 msr.index = msr_index;
5472 msr.host_initiated = false;
5473 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5474 if (r)
5475 return r;
5476
5477 *pdata = msr.data;
5478 return 0;
717746e3
AK		 5479}
5480
5481static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5482 u32 msr_index, u64 data)
5483{
8fe8ab46
WA		 5484 struct msr_data msr;
5485
5486 msr.data = data;
5487 msr.index = msr_index;
5488 msr.host_initiated = false;
5489 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
717746e3
AK		 5490}
5491
64d60670
PB		 5492static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5493{
5494 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5495
5496 return vcpu->arch.smbase;
5497}
5498
5499static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5500{
5501 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5502
5503 vcpu->arch.smbase = smbase;
5504}
5505
67f4d428
NA		 5506static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5507 u32 pmc)
5508{
c6702c9d 5509 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
67f4d428
NA		 5510}
5511
222d21aa
AK		 5512static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5513 u32 pmc, u64 *pdata)
5514{
c6702c9d 5515 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
222d21aa
AK		 5516}
5517
6c3287f7
AK		 5518static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5519{
5520 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5521}
5522
2953538e 5523static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
8a76d7f2 5524 struct x86_instruction_info *info,
c4f035c6
AK		 5525 enum x86_intercept_stage stage)
5526{
2953538e 5527 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
c4f035c6
AK		 5528}
5529
e911eb3b
YZ		 5530static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5531 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
bdb42f5a 5532{
e911eb3b 5533 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
bdb42f5a
SB		 5534}
5535
dd856efa
AK		 5536static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5537{
5538 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5539}
5540
5541static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5542{
5543 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5544}
5545
801806d9
NA		 5546static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5547{
5548 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5549}
5550
6ed071f0
LP		 5551static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5552{
5553 return emul_to_vcpu(ctxt)->arch.hflags;
5554}
5555
5556static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5557{
5558 kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5559}
5560
0234bf88
LP		 5561static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, u64 smbase)
5562{
5563 return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smbase);
5564}
5565
0225fb50 5566static const struct x86_emulate_ops emulate_ops = {
dd856efa
AK		 5567 .read_gpr = emulator_read_gpr,
5568 .write_gpr = emulator_write_gpr,
1871c602 5569 .read_std = kvm_read_guest_virt_system,
2dafc6c2 5570 .write_std = kvm_write_guest_virt_system,
7a036a6f 5571 .read_phys = kvm_read_guest_phys_system,
1871c602 5572 .fetch = kvm_fetch_guest_virt,
bbd9b64e
CO		 5573 .read_emulated = emulator_read_emulated,
5574 .write_emulated = emulator_write_emulated,
5575 .cmpxchg_emulated = emulator_cmpxchg_emulated,
3cb16fe7 5576 .invlpg = emulator_invlpg,
cf8f70bf
GN		 5577 .pio_in_emulated = emulator_pio_in_emulated,
5578 .pio_out_emulated = emulator_pio_out_emulated,
1aa36616
AK		 5579 .get_segment = emulator_get_segment,
5580 .set_segment = emulator_set_segment,
5951c442 5581 .get_cached_segment_base = emulator_get_cached_segment_base,
2dafc6c2 5582 .get_gdt = emulator_get_gdt,
160ce1f1 5583 .get_idt = emulator_get_idt,
1ac9d0cf
AK		 5584 .set_gdt = emulator_set_gdt,
5585 .set_idt = emulator_set_idt,
52a46617
GN		 5586 .get_cr = emulator_get_cr,
5587 .set_cr = emulator_set_cr,
9c537244 5588 .cpl = emulator_get_cpl,
35aa5375
GN		 5589 .get_dr = emulator_get_dr,
5590 .set_dr = emulator_set_dr,
64d60670
PB		 5591 .get_smbase = emulator_get_smbase,
5592 .set_smbase = emulator_set_smbase,
717746e3
AK		 5593 .set_msr = emulator_set_msr,
5594 .get_msr = emulator_get_msr,
67f4d428 5595 .check_pmc = emulator_check_pmc,
222d21aa 5596 .read_pmc = emulator_read_pmc,
6c3287f7 5597 .halt = emulator_halt,
bcaf5cc5 5598 .wbinvd = emulator_wbinvd,
d6aa1000 5599 .fix_hypercall = emulator_fix_hypercall,
c4f035c6 5600 .intercept = emulator_intercept,
bdb42f5a 5601 .get_cpuid = emulator_get_cpuid,
801806d9 5602 .set_nmi_mask = emulator_set_nmi_mask,
6ed071f0
LP		 5603 .get_hflags = emulator_get_hflags,
5604 .set_hflags = emulator_set_hflags,
0234bf88 5605 .pre_leave_smm = emulator_pre_leave_smm,
bbd9b64e
CO		 5606};
5607
95cb2295
GN		 5608static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5609{
37ccdcbe 5610 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
95cb2295
GN		 5611 /*
5612 * an sti; sti; sequence only disable interrupts for the first
5613 * instruction. So, if the last instruction, be it emulated or
5614 * not, left the system with the INT_STI flag enabled, it
5615 * means that the last instruction is an sti. We should not
5616 * leave the flag on in this case. The same goes for mov ss
5617 */
37ccdcbe
PB		 5618 if (int_shadow & mask)
5619 mask = 0;
6addfc42 5620 if (unlikely(int_shadow || mask)) {
95cb2295 5621 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6addfc42
PB		 5622 if (!mask)
5623 kvm_make_request(KVM_REQ_EVENT, vcpu);
5624 }
95cb2295
GN		 5625}
5626
ef54bcfe 5627static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
54b8486f
GN		 5628{
5629 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
da9cb575 5630 if (ctxt->exception.vector == PF_VECTOR)
ef54bcfe
PB		 5631 return kvm_propagate_fault(vcpu, &ctxt->exception);
5632
5633 if (ctxt->exception.error_code_valid)
da9cb575
AK		 5634 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5635 ctxt->exception.error_code);
54b8486f 5636 else
da9cb575 5637 kvm_queue_exception(vcpu, ctxt->exception.vector);
ef54bcfe 5638 return false;
54b8486f
GN		 5639}
5640
8ec4722d
MG		 5641static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5642{
adf52235 5643 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8ec4722d
MG		 5644 int cs_db, cs_l;
5645
8ec4722d
MG		 5646 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5647
adf52235 5648 ctxt->eflags = kvm_get_rflags(vcpu);
c8401dda
PB		 5649 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
5650
adf52235
TY		 5651 ctxt->eip = kvm_rip_read(vcpu);
5652 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5653 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
42bf549f 5654 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
adf52235
TY		 5655 cs_db ? X86EMUL_MODE_PROT32 :
5656 X86EMUL_MODE_PROT16;
a584539b 5657 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
64d60670
PB		 5658 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5659 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
adf52235 5660
dd856efa 5661 init_decode_cache(ctxt);
7ae441ea 5662 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8ec4722d
MG		 5663}
5664
71f9833b 5665int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
63995653 5666{
9d74191a 5667 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
63995653
MG		 5668 int ret;
5669
5670 init_emulate_ctxt(vcpu);
5671
9dac77fa
AK		 5672 ctxt->op_bytes = 2;
5673 ctxt->ad_bytes = 2;
5674 ctxt->_eip = ctxt->eip + inc_eip;
9d74191a 5675 ret = emulate_int_real(ctxt, irq);
63995653
MG		 5676
5677 if (ret != X86EMUL_CONTINUE)
5678 return EMULATE_FAIL;
5679
9dac77fa 5680 ctxt->eip = ctxt->_eip;
9d74191a
TY		 5681 kvm_rip_write(vcpu, ctxt->eip);
5682 kvm_set_rflags(vcpu, ctxt->eflags);
63995653 5683
63995653
MG		 5684 return EMULATE_DONE;
5685}
5686EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5687
e2366171 5688static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6d77dbfc 5689{
fc3a9157
JR		 5690 int r = EMULATE_DONE;
5691
6d77dbfc
GN		 5692 ++vcpu->stat.insn_emulation_fail;
5693 trace_kvm_emulate_insn_failed(vcpu);
e2366171
LA		 5694
5695 if (emulation_type & EMULTYPE_NO_UD_ON_FAIL)
5696 return EMULATE_FAIL;
5697
a2b9e6c1 5698 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
fc3a9157
JR		 5699 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5700 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5701 vcpu->run->internal.ndata = 0;
1f4dcb3b 5702 r = EMULATE_USER_EXIT;
fc3a9157 5703 }
e2366171 5704
6d77dbfc 5705 kvm_queue_exception(vcpu, UD_VECTOR);
fc3a9157
JR		 5706
5707 return r;
6d77dbfc
GN		 5708}
5709
93c05d3e 5710static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
991eebf9
GN		 5711 bool write_fault_to_shadow_pgtable,
5712 int emulation_type)
a6f177ef 5713{
95b3cf69 5714 gpa_t gpa = cr2;
ba049e93 5715 kvm_pfn_t pfn;
a6f177ef 5716
991eebf9
GN		 5717 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5718 return false;
5719
95b3cf69
XG		 5720 if (!vcpu->arch.mmu.direct_map) {
5721 /*
5722 * Write permission should be allowed since only
5723 * write access need to be emulated.
5724 */
5725 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
a6f177ef 5726
95b3cf69
XG		 5727 /*
5728 * If the mapping is invalid in guest, let cpu retry
5729 * it to generate fault.
5730 */
5731 if (gpa == UNMAPPED_GVA)
5732 return true;
5733 }
a6f177ef 5734
8e3d9d06
XG		 5735 /*
5736 * Do not retry the unhandleable instruction if it faults on the
5737 * readonly host memory, otherwise it will goto a infinite loop:
5738 * retry instruction -> write #PF -> emulation fail -> retry
5739 * instruction -> ...
5740 */
5741 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
95b3cf69
XG		 5742
5743 /*
5744 * If the instruction failed on the error pfn, it can not be fixed,
5745 * report the error to userspace.
5746 */
5747 if (is_error_noslot_pfn(pfn))
5748 return false;
5749
5750 kvm_release_pfn_clean(pfn);
5751
5752 /* The instructions are well-emulated on direct mmu. */
5753 if (vcpu->arch.mmu.direct_map) {
5754 unsigned int indirect_shadow_pages;
5755
5756 spin_lock(&vcpu->kvm->mmu_lock);
5757 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5758 spin_unlock(&vcpu->kvm->mmu_lock);
5759
5760 if (indirect_shadow_pages)
5761 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5762
a6f177ef 5763 return true;
8e3d9d06 5764 }
a6f177ef 5765
95b3cf69
XG		 5766 /*
5767 * if emulation was due to access to shadowed page table
5768 * and it failed try to unshadow page and re-enter the
5769 * guest to let CPU execute the instruction.
5770 */
5771 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
93c05d3e
XG		 5772
5773 /*
5774 * If the access faults on its page table, it can not
5775 * be fixed by unprotecting shadow page and it should
5776 * be reported to userspace.
5777 */
5778 return !write_fault_to_shadow_pgtable;
a6f177ef
GN		 5779}
5780
1cb3f3ae
XG		 5781static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5782 unsigned long cr2, int emulation_type)
5783{
5784 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5785 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5786
5787 last_retry_eip = vcpu->arch.last_retry_eip;
5788 last_retry_addr = vcpu->arch.last_retry_addr;
5789
5790 /*
5791 * If the emulation is caused by #PF and it is non-page_table
5792 * writing instruction, it means the VM-EXIT is caused by shadow
5793 * page protected, we can zap the shadow page and retry this
5794 * instruction directly.
5795 *
5796 * Note: if the guest uses a non-page-table modifying instruction
5797 * on the PDE that points to the instruction, then we will unmap
5798 * the instruction and go to an infinite loop. So, we cache the
5799 * last retried eip and the last fault address, if we meet the eip
5800 * and the address again, we can break out of the potential infinite
5801 * loop.
5802 */
5803 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5804
5805 if (!(emulation_type & EMULTYPE_RETRY))
5806 return false;
5807
5808 if (x86_page_table_writing_insn(ctxt))
5809 return false;
5810
5811 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5812 return false;
5813
5814 vcpu->arch.last_retry_eip = ctxt->eip;
5815 vcpu->arch.last_retry_addr = cr2;
5816
5817 if (!vcpu->arch.mmu.direct_map)
5818 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5819
22368028 5820 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
1cb3f3ae
XG		 5821
5822 return true;
5823}
5824
716d51ab
GN		 5825static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5826static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5827
64d60670 5828static void kvm_smm_changed(struct kvm_vcpu *vcpu)
a584539b 5829{
64d60670 5830 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
660a5d51
PB		 5831 /* This is a good place to trace that we are exiting SMM. */
5832 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5833
c43203ca
PB		 5834 /* Process a latched INIT or SMI, if any. */
5835 kvm_make_request(KVM_REQ_EVENT, vcpu);
64d60670 5836 }
699023e2
PB		 5837
5838 kvm_mmu_reset_context(vcpu);
64d60670
PB		 5839}
5840
5841static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5842{
5843 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5844
a584539b 5845 vcpu->arch.hflags = emul_flags;
64d60670
PB		 5846
5847 if (changed & HF_SMM_MASK)
5848 kvm_smm_changed(vcpu);
a584539b
PB		 5849}
5850
4a1e10d5
PB		 5851static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5852 unsigned long *db)
5853{
5854 u32 dr6 = 0;
5855 int i;
5856 u32 enable, rwlen;
5857
5858 enable = dr7;
5859 rwlen = dr7 >> 16;
5860 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5861 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5862 dr6 |= (1 << i);
5863 return dr6;
5864}
5865
c8401dda 5866static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
663f4c61
PB		 5867{
5868 struct kvm_run *kvm_run = vcpu->run;
5869
c8401dda
PB		 5870 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5871 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
5872 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5873 kvm_run->debug.arch.exception = DB_VECTOR;
5874 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5875 *r = EMULATE_USER_EXIT;
5876 } else {
5877 /*
5878 * "Certain debug exceptions may clear bit 0-3. The
5879 * remaining contents of the DR6 register are never
5880 * cleared by the processor".
5881 */
5882 vcpu->arch.dr6 &= ~15;
5883 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5884 kvm_queue_exception(vcpu, DB_VECTOR);
663f4c61
PB		 5885 }
5886}
5887
6affcbed
KH		 5888int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
5889{
5890 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5891 int r = EMULATE_DONE;
5892
5893 kvm_x86_ops->skip_emulated_instruction(vcpu);
c8401dda
PB		 5894
5895 /*
5896 * rflags is the old, "raw" value of the flags. The new value has
5897 * not been saved yet.
5898 *
5899 * This is correct even for TF set by the guest, because "the
5900 * processor will not generate this exception after the instruction
5901 * that sets the TF flag".
5902 */
5903 if (unlikely(rflags & X86_EFLAGS_TF))
5904 kvm_vcpu_do_singlestep(vcpu, &r);
6affcbed
KH		 5905 return r == EMULATE_DONE;
5906}
5907EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
5908
4a1e10d5
PB		 5909static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5910{
4a1e10d5
PB		 5911 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5912 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
82b32774
NA		 5913 struct kvm_run *kvm_run = vcpu->run;
5914 unsigned long eip = kvm_get_linear_rip(vcpu);
5915 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
4a1e10d5
PB		 5916 vcpu->arch.guest_debug_dr7,
5917 vcpu->arch.eff_db);
5918
5919 if (dr6 != 0) {
6f43ed01 5920 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
82b32774 5921 kvm_run->debug.arch.pc = eip;
4a1e10d5
PB		 5922 kvm_run->debug.arch.exception = DB_VECTOR;
5923 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5924 *r = EMULATE_USER_EXIT;
5925 return true;
5926 }
5927 }
5928
4161a569
NA		 5929 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5930 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
82b32774
NA		 5931 unsigned long eip = kvm_get_linear_rip(vcpu);
5932 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
4a1e10d5
PB		 5933 vcpu->arch.dr7,
5934 vcpu->arch.db);
5935
5936 if (dr6 != 0) {
5937 vcpu->arch.dr6 &= ~15;
6f43ed01 5938 vcpu->arch.dr6 |= dr6 | DR6_RTM;
4a1e10d5
PB		 5939 kvm_queue_exception(vcpu, DB_VECTOR);
5940 *r = EMULATE_DONE;
5941 return true;
5942 }
5943 }
5944
5945 return false;
5946}
5947
04789b66
LA		 5948static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
5949{
2d7921c4
AM		 5950 switch (ctxt->opcode_len) {
5951 case 1:
5952 switch (ctxt->b) {
5953 case 0xe4: /* IN */
5954 case 0xe5:
5955 case 0xec:
5956 case 0xed:
5957 case 0xe6: /* OUT */
5958 case 0xe7:
5959 case 0xee:
5960 case 0xef:
5961 case 0x6c: /* INS */
5962 case 0x6d:
5963 case 0x6e: /* OUTS */
5964 case 0x6f:
5965 return true;
5966 }
5967 break;
5968 case 2:
5969 switch (ctxt->b) {
5970 case 0x33: /* RDPMC */
5971 return true;
5972 }
5973 break;
04789b66
LA		 5974 }
5975
5976 return false;
5977}
5978
51d8b661
AP		 5979int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5980 unsigned long cr2,
dc25e89e
AP		 5981 int emulation_type,
5982 void *insn,
5983 int insn_len)
bbd9b64e 5984{
95cb2295 5985 int r;
9d74191a 5986 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7ae441ea 5987 bool writeback = true;
93c05d3e 5988 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
bbd9b64e 5989
93c05d3e
XG		 5990 /*
5991 * Clear write_fault_to_shadow_pgtable here to ensure it is
5992 * never reused.
5993 */
5994 vcpu->arch.write_fault_to_shadow_pgtable = false;
26eef70c 5995 kvm_clear_exception_queue(vcpu);
8d7d8102 5996
571008da 5997 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
8ec4722d 5998 init_emulate_ctxt(vcpu);
4a1e10d5
PB		 5999
6000 /*
6001 * We will reenter on the same instruction since
6002 * we do not set complete_userspace_io. This does not
6003 * handle watchpoints yet, those would be handled in
6004 * the emulate_ops.
6005 */
d391f120
VK		 6006 if (!(emulation_type & EMULTYPE_SKIP) &&
6007 kvm_vcpu_check_breakpoint(vcpu, &r))
4a1e10d5
PB		 6008 return r;
6009
9d74191a
TY		 6010 ctxt->interruptibility = 0;
6011 ctxt->have_exception = false;
e0ad0b47 6012 ctxt->exception.vector = -1;
9d74191a 6013 ctxt->perm_ok = false;
bbd9b64e 6014
b51e974f 6015 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
4005996e 6016
9d74191a 6017 r = x86_decode_insn(ctxt, insn, insn_len);
bbd9b64e 6018
e46479f8 6019 trace_kvm_emulate_insn_start(vcpu);
f2b5756b 6020 ++vcpu->stat.insn_emulation;
1d2887e2 6021 if (r != EMULATION_OK) {
4005996e
AK		 6022 if (emulation_type & EMULTYPE_TRAP_UD)
6023 return EMULATE_FAIL;
991eebf9
GN		 6024 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6025 emulation_type))
bbd9b64e 6026 return EMULATE_DONE;
6ea6e843
PB		 6027 if (ctxt->have_exception && inject_emulated_exception(vcpu))
6028 return EMULATE_DONE;
6d77dbfc
GN		 6029 if (emulation_type & EMULTYPE_SKIP)
6030 return EMULATE_FAIL;
e2366171 6031 return handle_emulation_failure(vcpu, emulation_type);
bbd9b64e
CO		 6032 }
6033 }
6034
04789b66
LA		 6035 if ((emulation_type & EMULTYPE_VMWARE) &&
6036 !is_vmware_backdoor_opcode(ctxt))
6037 return EMULATE_FAIL;
6038
ba8afb6b 6039 if (emulation_type & EMULTYPE_SKIP) {
9dac77fa 6040 kvm_rip_write(vcpu, ctxt->_eip);
bb663c7a
NA		 6041 if (ctxt->eflags & X86_EFLAGS_RF)
6042 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
ba8afb6b
GN		 6043 return EMULATE_DONE;
6044 }
6045
1cb3f3ae
XG		 6046 if (retry_instruction(ctxt, cr2, emulation_type))
6047 return EMULATE_DONE;
6048
7ae441ea 6049 /* this is needed for vmware backdoor interface to work since it
4d2179e1 6050 changes registers values during IO operation */
7ae441ea
GN		 6051 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6052 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
dd856efa 6053 emulator_invalidate_register_cache(ctxt);
7ae441ea 6054 }
4d2179e1 6055
5cd21917 6056restart:
0f89b207
TL		 6057 /* Save the faulting GPA (cr2) in the address field */
6058 ctxt->exception.address = cr2;
6059
9d74191a 6060 r = x86_emulate_insn(ctxt);
bbd9b64e 6061
775fde86
JR		 6062 if (r == EMULATION_INTERCEPTED)
6063 return EMULATE_DONE;
6064
d2ddd1c4 6065 if (r == EMULATION_FAILED) {
991eebf9
GN		 6066 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6067 emulation_type))
c3cd7ffa
GN		 6068 return EMULATE_DONE;
6069
e2366171 6070 return handle_emulation_failure(vcpu, emulation_type);
bbd9b64e
CO		 6071 }
6072
9d74191a 6073 if (ctxt->have_exception) {
d2ddd1c4 6074 r = EMULATE_DONE;
ef54bcfe
PB		 6075 if (inject_emulated_exception(vcpu))
6076 return r;
d2ddd1c4 6077 } else if (vcpu->arch.pio.count) {
0912c977
PB		 6078 if (!vcpu->arch.pio.in) {
6079 /* FIXME: return into emulator if single-stepping. */
3457e419 6080 vcpu->arch.pio.count = 0;
0912c977 6081 } else {
7ae441ea 6082 writeback = false;
716d51ab
GN		 6083 vcpu->arch.complete_userspace_io = complete_emulated_pio;
6084 }
ac0a48c3 6085 r = EMULATE_USER_EXIT;
7ae441ea
GN		 6086 } else if (vcpu->mmio_needed) {
6087 if (!vcpu->mmio_is_write)
6088 writeback = false;
ac0a48c3 6089 r = EMULATE_USER_EXIT;
716d51ab 6090 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7ae441ea 6091 } else if (r == EMULATION_RESTART)
5cd21917 6092 goto restart;
d2ddd1c4
GN		 6093 else
6094 r = EMULATE_DONE;
f850e2e6 6095
7ae441ea 6096 if (writeback) {
6addfc42 6097 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
9d74191a 6098 toggle_interruptibility(vcpu, ctxt->interruptibility);
7ae441ea 6099 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9d74191a 6100 kvm_rip_write(vcpu, ctxt->eip);
c8401dda
PB		 6101 if (r == EMULATE_DONE &&
6102 (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
6103 kvm_vcpu_do_singlestep(vcpu, &r);
38827dbd
NA		 6104 if (!ctxt->have_exception ||
6105 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
6106 __kvm_set_rflags(vcpu, ctxt->eflags);
6addfc42
PB		 6107
6108 /*
6109 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6110 * do nothing, and it will be requested again as soon as
6111 * the shadow expires. But we still need to check here,
6112 * because POPF has no interrupt shadow.
6113 */
6114 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6115 kvm_make_request(KVM_REQ_EVENT, vcpu);
7ae441ea
GN		 6116 } else
6117 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
e85d28f8
GN		 6118
6119 return r;
de7d789a 6120}
51d8b661 6121EXPORT_SYMBOL_GPL(x86_emulate_instruction);
de7d789a 6122
dca7f128
SC		 6123static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6124 unsigned short port)
de7d789a 6125{
cf8f70bf 6126 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
ca1d4a9e
AK		 6127 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6128 size, port, &val, 1);
cf8f70bf 6129 /* do not return to emulator after return from userspace */
7972995b 6130 vcpu->arch.pio.count = 0;
de7d789a
CO		 6131 return ret;
6132}
de7d789a 6133
8370c3d0
TL		 6134static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6135{
6136 unsigned long val;
6137
6138 /* We should only ever be called with arch.pio.count equal to 1 */
6139 BUG_ON(vcpu->arch.pio.count != 1);
6140
6141 /* For size less than 4 we merge, else we zero extend */
6142 val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
6143 : 0;
6144
6145 /*
6146 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6147 * the copy and tracing
6148 */
6149 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6150 vcpu->arch.pio.port, &val, 1);
6151 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6152
6153 return 1;
6154}
6155
dca7f128
SC		 6156static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6157 unsigned short port)
8370c3d0
TL		 6158{
6159 unsigned long val;
6160 int ret;
6161
6162 /* For size less than 4 we merge, else we zero extend */
6163 val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
6164
6165 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6166 &val, 1);
6167 if (ret) {
6168 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6169 return ret;
6170 }
6171
6172 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6173
6174 return 0;
6175}
dca7f128
SC		 6176
6177int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6178{
6179 int ret = kvm_skip_emulated_instruction(vcpu);
6180
6181 /*
6182 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
6183 * KVM_EXIT_DEBUG here.
6184 */
6185 if (in)
6186 return kvm_fast_pio_in(vcpu, size, port) && ret;
6187 else
6188 return kvm_fast_pio_out(vcpu, size, port) && ret;
6189}
6190EXPORT_SYMBOL_GPL(kvm_fast_pio);
8370c3d0 6191
251a5fd6 6192static int kvmclock_cpu_down_prep(unsigned int cpu)
8cfdc000 6193{
0a3aee0d 6194 __this_cpu_write(cpu_tsc_khz, 0);
251a5fd6 6195 return 0;
8cfdc000
ZA		 6196}
6197
6198static void tsc_khz_changed(void *data)
c8076604 6199{
8cfdc000
ZA		 6200 struct cpufreq_freqs *freq = data;
6201 unsigned long khz = 0;
6202
6203 if (data)
6204 khz = freq->new;
6205 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6206 khz = cpufreq_quick_get(raw_smp_processor_id());
6207 if (!khz)
6208 khz = tsc_khz;
0a3aee0d 6209 __this_cpu_write(cpu_tsc_khz, khz);
c8076604
GH		 6210}
6211
5fa4ec9c 6212#ifdef CONFIG_X86_64
0092e434
VK		 6213static void kvm_hyperv_tsc_notifier(void)
6214{
0092e434
VK		 6215 struct kvm *kvm;
6216 struct kvm_vcpu *vcpu;
6217 int cpu;
6218
6219 spin_lock(&kvm_lock);
6220 list_for_each_entry(kvm, &vm_list, vm_list)
6221 kvm_make_mclock_inprogress_request(kvm);
6222
6223 hyperv_stop_tsc_emulation();
6224
6225 /* TSC frequency always matches when on Hyper-V */
6226 for_each_present_cpu(cpu)
6227 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6228 kvm_max_guest_tsc_khz = tsc_khz;
6229
6230 list_for_each_entry(kvm, &vm_list, vm_list) {
6231 struct kvm_arch *ka = &kvm->arch;
6232
6233 spin_lock(&ka->pvclock_gtod_sync_lock);
6234
6235 pvclock_update_vm_gtod_copy(kvm);
6236
6237 kvm_for_each_vcpu(cpu, vcpu, kvm)
6238 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6239
6240 kvm_for_each_vcpu(cpu, vcpu, kvm)
6241 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6242
6243 spin_unlock(&ka->pvclock_gtod_sync_lock);
6244 }
6245 spin_unlock(&kvm_lock);
0092e434 6246}
5fa4ec9c 6247#endif
0092e434 6248
c8076604
GH		 6249static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
6250 void *data)
6251{
6252 struct cpufreq_freqs *freq = data;
6253 struct kvm *kvm;
6254 struct kvm_vcpu *vcpu;
6255 int i, send_ipi = 0;
6256
8cfdc000
ZA		 6257 /*
6258 * We allow guests to temporarily run on slowing clocks,
6259 * provided we notify them after, or to run on accelerating
6260 * clocks, provided we notify them before. Thus time never
6261 * goes backwards.
6262 *
6263 * However, we have a problem. We can't atomically update
6264 * the frequency of a given CPU from this function; it is
6265 * merely a notifier, which can be called from any CPU.
6266 * Changing the TSC frequency at arbitrary points in time
6267 * requires a recomputation of local variables related to
6268 * the TSC for each VCPU. We must flag these local variables
6269 * to be updated and be sure the update takes place with the
6270 * new frequency before any guests proceed.
6271 *
6272 * Unfortunately, the combination of hotplug CPU and frequency
6273 * change creates an intractable locking scenario; the order
6274 * of when these callouts happen is undefined with respect to
6275 * CPU hotplug, and they can race with each other. As such,
6276 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6277 * undefined; you can actually have a CPU frequency change take
6278 * place in between the computation of X and the setting of the
6279 * variable. To protect against this problem, all updates of
6280 * the per_cpu tsc_khz variable are done in an interrupt
6281 * protected IPI, and all callers wishing to update the value
6282 * must wait for a synchronous IPI to complete (which is trivial
6283 * if the caller is on the CPU already). This establishes the
6284 * necessary total order on variable updates.
6285 *
6286 * Note that because a guest time update may take place
6287 * anytime after the setting of the VCPU's request bit, the
6288 * correct TSC value must be set before the request. However,
6289 * to ensure the update actually makes it to any guest which
6290 * starts running in hardware virtualization between the set
6291 * and the acquisition of the spinlock, we must also ping the
6292 * CPU after setting the request bit.
6293 *
6294 */
6295
c8076604
GH		 6296 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
6297 return 0;
6298 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
6299 return 0;
8cfdc000
ZA		 6300
6301 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
c8076604 6302
2f303b74 6303 spin_lock(&kvm_lock);
c8076604 6304 list_for_each_entry(kvm, &vm_list, vm_list) {
988a2cae 6305 kvm_for_each_vcpu(i, vcpu, kvm) {
c8076604
GH		 6306 if (vcpu->cpu != freq->cpu)
6307 continue;
c285545f 6308 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
c8076604 6309 if (vcpu->cpu != smp_processor_id())
8cfdc000 6310 send_ipi = 1;
c8076604
GH		 6311 }
6312 }
2f303b74 6313 spin_unlock(&kvm_lock);
c8076604
GH		 6314
6315 if (freq->old < freq->new && send_ipi) {
6316 /*
6317 * We upscale the frequency. Must make the guest
6318 * doesn't see old kvmclock values while running with
6319 * the new frequency, otherwise we risk the guest sees
6320 * time go backwards.
6321 *
6322 * In case we update the frequency for another cpu
6323 * (which might be in guest context) send an interrupt
6324 * to kick the cpu out of guest context. Next time
6325 * guest context is entered kvmclock will be updated,
6326 * so the guest will not see stale values.
6327 */
8cfdc000 6328 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
c8076604
GH		 6329 }
6330 return 0;
6331}
6332
6333static struct notifier_block kvmclock_cpufreq_notifier_block = {
8cfdc000
ZA		 6334 .notifier_call = kvmclock_cpufreq_notifier
6335};
6336
251a5fd6 6337static int kvmclock_cpu_online(unsigned int cpu)
8cfdc000 6338{
251a5fd6
SAS		 6339 tsc_khz_changed(NULL);
6340 return 0;
8cfdc000
ZA		 6341}
6342
b820cc0c
ZA		 6343static void kvm_timer_init(void)
6344{
c285545f 6345 max_tsc_khz = tsc_khz;
460dd42e 6346
b820cc0c 6347 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
c285545f
ZA		 6348#ifdef CONFIG_CPU_FREQ
6349 struct cpufreq_policy policy;
758f588d
BP		 6350 int cpu;
6351
c285545f 6352 memset(&policy, 0, sizeof(policy));
3e26f230
AK		 6353 cpu = get_cpu();
6354 cpufreq_get_policy(&policy, cpu);
c285545f
ZA		 6355 if (policy.cpuinfo.max_freq)
6356 max_tsc_khz = policy.cpuinfo.max_freq;
3e26f230 6357 put_cpu();
c285545f 6358#endif
b820cc0c
ZA		 6359 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6360 CPUFREQ_TRANSITION_NOTIFIER);
6361 }
c285545f 6362 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
460dd42e 6363
73c1b41e 6364 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
251a5fd6 6365 kvmclock_cpu_online, kvmclock_cpu_down_prep);
b820cc0c
ZA		 6366}
6367
dd60d217
AK		 6368DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6369EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
ff9d07a0 6370
f5132b01 6371int kvm_is_in_guest(void)
ff9d07a0 6372{
086c9855 6373 return __this_cpu_read(current_vcpu) != NULL;
ff9d07a0
ZY		 6374}
6375
6376static int kvm_is_user_mode(void)
6377{
6378 int user_mode = 3;
dcf46b94 6379
086c9855
AS		 6380 if (__this_cpu_read(current_vcpu))
6381 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
dcf46b94 6382
ff9d07a0
ZY		 6383 return user_mode != 0;
6384}
6385
6386static unsigned long kvm_get_guest_ip(void)
6387{
6388 unsigned long ip = 0;
dcf46b94 6389
086c9855
AS		 6390 if (__this_cpu_read(current_vcpu))
6391 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
dcf46b94 6392
ff9d07a0
ZY		 6393 return ip;
6394}
6395
6396static struct perf_guest_info_callbacks kvm_guest_cbs = {
6397 .is_in_guest = kvm_is_in_guest,
6398 .is_user_mode = kvm_is_user_mode,
6399 .get_guest_ip = kvm_get_guest_ip,
6400};
6401
ce88decf
XG		 6402static void kvm_set_mmio_spte_mask(void)
6403{
6404 u64 mask;
6405 int maxphyaddr = boot_cpu_data.x86_phys_bits;
6406
6407 /*
6408 * Set the reserved bits and the present bit of an paging-structure
6409 * entry to generate page fault with PFER.RSV = 1.
6410 */
885032b9 6411 /* Mask the reserved physical address bits. */
d1431483 6412 mask = rsvd_bits(maxphyaddr, 51);
885032b9 6413
885032b9 6414 /* Set the present bit. */
ce88decf
XG		 6415 mask |= 1ull;
6416
6417#ifdef CONFIG_X86_64
6418 /*
6419 * If reserved bit is not supported, clear the present bit to disable
6420 * mmio page fault.
6421 */
6422 if (maxphyaddr == 52)
6423 mask &= ~1ull;
6424#endif
6425
dcdca5fe 6426 kvm_mmu_set_mmio_spte_mask(mask, mask);
ce88decf
XG		 6427}
6428
16e8d74d
MT		 6429#ifdef CONFIG_X86_64
6430static void pvclock_gtod_update_fn(struct work_struct *work)
6431{
d828199e
MT		 6432 struct kvm *kvm;
6433
6434 struct kvm_vcpu *vcpu;
6435 int i;
6436
2f303b74 6437 spin_lock(&kvm_lock);
d828199e
MT		 6438 list_for_each_entry(kvm, &vm_list, vm_list)
6439 kvm_for_each_vcpu(i, vcpu, kvm)
105b21bb 6440 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
d828199e 6441 atomic_set(&kvm_guest_has_master_clock, 0);
2f303b74 6442 spin_unlock(&kvm_lock);
16e8d74d
MT		 6443}
6444
6445static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6446
6447/*
6448 * Notification about pvclock gtod data update.
6449 */
6450static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6451 void *priv)
6452{
6453 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6454 struct timekeeper *tk = priv;
6455
6456 update_pvclock_gtod(tk);
6457
6458 /* disable master clock if host does not trust, or does not
b0c39dc6 6459 * use, TSC based clocksource.
16e8d74d 6460 */
b0c39dc6 6461 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
16e8d74d
MT		 6462 atomic_read(&kvm_guest_has_master_clock) != 0)
6463 queue_work(system_long_wq, &pvclock_gtod_work);
6464
6465 return 0;
6466}
6467
6468static struct notifier_block pvclock_gtod_notifier = {
6469 .notifier_call = pvclock_gtod_notify,
6470};
6471#endif
6472
f8c16bba 6473int kvm_arch_init(void *opaque)
043405e1 6474{
b820cc0c 6475 int r;
6b61edf7 6476 struct kvm_x86_ops *ops = opaque;
f8c16bba 6477
f8c16bba
ZX		 6478 if (kvm_x86_ops) {
6479 printk(KERN_ERR "kvm: already loaded the other module\n");
56c6d28a
ZX		 6480 r = -EEXIST;
6481 goto out;
f8c16bba
ZX		 6482 }
6483
6484 if (!ops->cpu_has_kvm_support()) {
6485 printk(KERN_ERR "kvm: no hardware support\n");
56c6d28a
ZX		 6486 r = -EOPNOTSUPP;
6487 goto out;
f8c16bba
ZX		 6488 }
6489 if (ops->disabled_by_bios()) {
6490 printk(KERN_ERR "kvm: disabled by bios\n");
56c6d28a
ZX		 6491 r = -EOPNOTSUPP;
6492 goto out;
f8c16bba
ZX		 6493 }
6494
013f6a5d
MT		 6495 r = -ENOMEM;
6496 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6497 if (!shared_msrs) {
6498 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6499 goto out;
6500 }
6501
97db56ce
AK		 6502 r = kvm_mmu_module_init();
6503 if (r)
013f6a5d 6504 goto out_free_percpu;
97db56ce 6505
ce88decf 6506 kvm_set_mmio_spte_mask();
97db56ce 6507
f8c16bba 6508 kvm_x86_ops = ops;
920c8377 6509
7b52345e 6510 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
ffb128c8 6511 PT_DIRTY_MASK, PT64_NX_MASK, 0,
d0ec49d4 6512 PT_PRESENT_MASK, 0, sme_me_mask);
b820cc0c 6513 kvm_timer_init();
c8076604 6514
ff9d07a0
ZY		 6515 perf_register_guest_info_callbacks(&kvm_guest_cbs);
6516
d366bf7e 6517 if (boot_cpu_has(X86_FEATURE_XSAVE))
2acf923e
DC		 6518 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6519
c5cc421b 6520 kvm_lapic_init();
16e8d74d
MT		 6521#ifdef CONFIG_X86_64
6522 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
0092e434 6523
5fa4ec9c 6524 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
0092e434 6525 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
16e8d74d
MT		 6526#endif
6527
f8c16bba 6528 return 0;
56c6d28a 6529
013f6a5d
MT		 6530out_free_percpu:
6531 free_percpu(shared_msrs);
56c6d28a 6532out:
56c6d28a 6533 return r;
043405e1 6534}
8776e519 6535
f8c16bba
ZX		 6536void kvm_arch_exit(void)
6537{
0092e434 6538#ifdef CONFIG_X86_64
5fa4ec9c 6539 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
0092e434
VK		 6540 clear_hv_tscchange_cb();
6541#endif
cef84c30 6542 kvm_lapic_exit();
ff9d07a0
ZY		 6543 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6544
888d256e
JK		 6545 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6546 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6547 CPUFREQ_TRANSITION_NOTIFIER);
251a5fd6 6548 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
16e8d74d
MT		 6549#ifdef CONFIG_X86_64
6550 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6551#endif
f8c16bba 6552 kvm_x86_ops = NULL;
56c6d28a 6553 kvm_mmu_module_exit();
013f6a5d 6554 free_percpu(shared_msrs);
56c6d28a 6555}
f8c16bba 6556
5cb56059 6557int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8776e519
HB		 6558{
6559 ++vcpu->stat.halt_exits;
35754c98 6560 if (lapic_in_kernel(vcpu)) {
a4535290 6561 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
8776e519
HB		 6562 return 1;
6563 } else {
6564 vcpu->run->exit_reason = KVM_EXIT_HLT;
6565 return 0;
6566 }
6567}
5cb56059
JS		 6568EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6569
6570int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6571{
6affcbed
KH		 6572 int ret = kvm_skip_emulated_instruction(vcpu);
6573 /*
6574 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6575 * KVM_EXIT_DEBUG here.
6576 */
6577 return kvm_vcpu_halt(vcpu) && ret;
5cb56059 6578}
8776e519
HB		 6579EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6580
8ef81a9a 6581#ifdef CONFIG_X86_64
55dd00a7
MT		 6582static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6583 unsigned long clock_type)
6584{
6585 struct kvm_clock_pairing clock_pairing;
6586 struct timespec ts;
80fbd89c 6587 u64 cycle;
55dd00a7
MT		 6588 int ret;
6589
6590 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6591 return -KVM_EOPNOTSUPP;
6592
6593 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6594 return -KVM_EOPNOTSUPP;
6595
6596 clock_pairing.sec = ts.tv_sec;
6597 clock_pairing.nsec = ts.tv_nsec;
6598 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6599 clock_pairing.flags = 0;
6600
6601 ret = 0;
6602 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6603 sizeof(struct kvm_clock_pairing)))
6604 ret = -KVM_EFAULT;
6605
6606 return ret;
6607}
8ef81a9a 6608#endif
55dd00a7 6609
6aef266c
SV		 6610/*
6611 * kvm_pv_kick_cpu_op: Kick a vcpu.
6612 *
6613 * @apicid - apicid of vcpu to be kicked.
6614 */
6615static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6616{
24d2166b 6617 struct kvm_lapic_irq lapic_irq;
6aef266c 6618
24d2166b
R		 6619 lapic_irq.shorthand = 0;
6620 lapic_irq.dest_mode = 0;
ebd28fcb 6621 lapic_irq.level = 0;
24d2166b 6622 lapic_irq.dest_id = apicid;
93bbf0b8 6623 lapic_irq.msi_redir_hint = false;
6aef266c 6624
24d2166b 6625 lapic_irq.delivery_mode = APIC_DM_REMRD;
795a149e 6626 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6aef266c
SV		 6627}
6628
d62caabb
AS		 6629void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6630{
6631 vcpu->arch.apicv_active = false;
6632 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6633}
6634
8776e519
HB		 6635int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6636{
6637 unsigned long nr, a0, a1, a2, a3, ret;
6affcbed 6638 int op_64_bit, r;
8776e519 6639
6affcbed 6640 r = kvm_skip_emulated_instruction(vcpu);
5cb56059 6641
55cd8e5a
GN		 6642 if (kvm_hv_hypercall_enabled(vcpu->kvm))
6643 return kvm_hv_hypercall(vcpu);
6644
5fdbf976
MT		 6645 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6646 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6647 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6648 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6649 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
8776e519 6650
229456fc 6651 trace_kvm_hypercall(nr, a0, a1, a2, a3);
2714d1d3 6652
a449c7aa
NA		 6653 op_64_bit = is_64_bit_mode(vcpu);
6654 if (!op_64_bit) {
8776e519
HB		 6655 nr &= 0xFFFFFFFF;
6656 a0 &= 0xFFFFFFFF;
6657 a1 &= 0xFFFFFFFF;
6658 a2 &= 0xFFFFFFFF;
6659 a3 &= 0xFFFFFFFF;
6660 }
6661
07708c4a
JK		 6662 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6663 ret = -KVM_EPERM;
6664 goto out;
6665 }
6666
8776e519 6667 switch (nr) {
b93463aa
AK		 6668 case KVM_HC_VAPIC_POLL_IRQ:
6669 ret = 0;
6670 break;
6aef266c
SV		 6671 case KVM_HC_KICK_CPU:
6672 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6673 ret = 0;
6674 break;
8ef81a9a 6675#ifdef CONFIG_X86_64
55dd00a7
MT		 6676 case KVM_HC_CLOCK_PAIRING:
6677 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
6678 break;
8ef81a9a 6679#endif
8776e519
HB		 6680 default:
6681 ret = -KVM_ENOSYS;
6682 break;
6683 }
07708c4a 6684out:
a449c7aa
NA		 6685 if (!op_64_bit)
6686 ret = (u32)ret;
5fdbf976 6687 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
f11c3a8d 6688 ++vcpu->stat.hypercalls;
2f333bcb 6689 return r;
8776e519
HB		 6690}
6691EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6692
b6785def 6693static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8776e519 6694{
d6aa1000 6695 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8776e519 6696 char instruction[3];
5fdbf976 6697 unsigned long rip = kvm_rip_read(vcpu);
8776e519 6698
8776e519 6699 kvm_x86_ops->patch_hypercall(vcpu, instruction);
8776e519 6700
ce2e852e
DV		 6701 return emulator_write_emulated(ctxt, rip, instruction, 3,
6702 &ctxt->exception);
8776e519
HB		 6703}
6704
851ba692 6705static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
b6c7a5dc 6706{
782d422b
MG		 6707 return vcpu->run->request_interrupt_window &&
6708 likely(!pic_in_kernel(vcpu->kvm));
b6c7a5dc
HB		 6709}
6710
851ba692 6711static void post_kvm_run_save(struct kvm_vcpu *vcpu)
b6c7a5dc 6712{
851ba692
AK		 6713 struct kvm_run *kvm_run = vcpu->run;
6714
91586a3b 6715 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
f077825a 6716 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
2d3ad1f4 6717 kvm_run->cr8 = kvm_get_cr8(vcpu);
b6c7a5dc 6718 kvm_run->apic_base = kvm_get_apic_base(vcpu);
127a457a
MG		 6719 kvm_run->ready_for_interrupt_injection =
6720 pic_in_kernel(vcpu->kvm) ||
782d422b 6721 kvm_vcpu_ready_for_interrupt_injection(vcpu);
b6c7a5dc
HB		 6722}
6723
95ba8273
GN		 6724static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6725{
6726 int max_irr, tpr;
6727
6728 if (!kvm_x86_ops->update_cr8_intercept)
6729 return;
6730
bce87cce 6731 if (!lapic_in_kernel(vcpu))
88c808fd
AK		 6732 return;
6733
d62caabb
AS		 6734 if (vcpu->arch.apicv_active)
6735 return;
6736
8db3baa2
GN		 6737 if (!vcpu->arch.apic->vapic_addr)
6738 max_irr = kvm_lapic_find_highest_irr(vcpu);
6739 else
6740 max_irr = -1;
95ba8273
GN		 6741
6742 if (max_irr != -1)
6743 max_irr >>= 4;
6744
6745 tpr = kvm_lapic_get_cr8(vcpu);
6746
6747 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6748}
6749
b6b8a145 6750static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
95ba8273 6751{
b6b8a145
JK		 6752 int r;
6753
95ba8273 6754 /* try to reinject previous events if any */
664f8e26
WL		 6755 if (vcpu->arch.exception.injected) {
6756 kvm_x86_ops->queue_exception(vcpu);
6757 return 0;
6758 }
6759
6760 /*
6761 * Exceptions must be injected immediately, or the exception
6762 * frame will have the address of the NMI or interrupt handler.
6763 */
6764 if (!vcpu->arch.exception.pending) {
6765 if (vcpu->arch.nmi_injected) {
6766 kvm_x86_ops->set_nmi(vcpu);
6767 return 0;
6768 }
6769
04140b41 6770 if (vcpu->arch.interrupt.injected) {
664f8e26
WL		 6771 kvm_x86_ops->set_irq(vcpu);
6772 return 0;
6773 }
6774 }
6775
6776 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6777 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6778 if (r != 0)
6779 return r;
6780 }
6781
6782 /* try to inject new event if pending */
b59bb7bd 6783 if (vcpu->arch.exception.pending) {
5c1c85d0
AK		 6784 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6785 vcpu->arch.exception.has_error_code,
6786 vcpu->arch.exception.error_code);
d6e8c854 6787
664f8e26
WL		 6788 vcpu->arch.exception.pending = false;
6789 vcpu->arch.exception.injected = true;
6790
d6e8c854
NA		 6791 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6792 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6793 X86_EFLAGS_RF);
6794
6bdf0662
NA		 6795 if (vcpu->arch.exception.nr == DB_VECTOR &&
6796 (vcpu->arch.dr7 & DR7_GD)) {
6797 vcpu->arch.dr7 &= ~DR7_GD;
6798 kvm_update_dr7(vcpu);
6799 }
6800
cfcd20e5 6801 kvm_x86_ops->queue_exception(vcpu);
72d7b374 6802 } else if (vcpu->arch.smi_pending && !is_smm(vcpu) && kvm_x86_ops->smi_allowed(vcpu)) {
c43203ca 6803 vcpu->arch.smi_pending = false;
52797bf9 6804 ++vcpu->arch.smi_count;
ee2cd4b7 6805 enter_smm(vcpu);
c43203ca 6806 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
321c5658
YS		 6807 --vcpu->arch.nmi_pending;
6808 vcpu->arch.nmi_injected = true;
6809 kvm_x86_ops->set_nmi(vcpu);
c7c9c56c 6810 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
9242b5b6
BD		 6811 /*
6812 * Because interrupts can be injected asynchronously, we are
6813 * calling check_nested_events again here to avoid a race condition.
6814 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6815 * proposal and current concerns. Perhaps we should be setting
6816 * KVM_REQ_EVENT only on certain events and not unconditionally?
6817 */
6818 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6819 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6820 if (r != 0)
6821 return r;
6822 }
95ba8273 6823 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
66fd3f7f
GN		 6824 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6825 false);
6826 kvm_x86_ops->set_irq(vcpu);
95ba8273
GN		 6827 }
6828 }
ee2cd4b7 6829
b6b8a145 6830 return 0;
95ba8273
GN		 6831}
6832
7460fb4a
AK		 6833static void process_nmi(struct kvm_vcpu *vcpu)
6834{
6835 unsigned limit = 2;
6836
6837 /*
6838 * x86 is limited to one NMI running, and one NMI pending after it.
6839 * If an NMI is already in progress, limit further NMIs to just one.
6840 * Otherwise, allow two (and we'll inject the first one immediately).
6841 */
6842 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6843 limit = 1;
6844
6845 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6846 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6847 kvm_make_request(KVM_REQ_EVENT, vcpu);
6848}
6849
ee2cd4b7 6850static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
660a5d51
PB		 6851{
6852 u32 flags = 0;
6853 flags |= seg->g << 23;
6854 flags |= seg->db << 22;
6855 flags |= seg->l << 21;
6856 flags |= seg->avl << 20;
6857 flags |= seg->present << 15;
6858 flags |= seg->dpl << 13;
6859 flags |= seg->s << 12;
6860 flags |= seg->type << 8;
6861 return flags;
6862}
6863
ee2cd4b7 6864static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
660a5d51
PB		 6865{
6866 struct kvm_segment seg;
6867 int offset;
6868
6869 kvm_get_segment(vcpu, &seg, n);
6870 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6871
6872 if (n < 3)
6873 offset = 0x7f84 + n * 12;
6874 else
6875 offset = 0x7f2c + (n - 3) * 12;
6876
6877 put_smstate(u32, buf, offset + 8, seg.base);
6878 put_smstate(u32, buf, offset + 4, seg.limit);
ee2cd4b7 6879 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
660a5d51
PB		 6880}
6881
efbb288a 6882#ifdef CONFIG_X86_64
ee2cd4b7 6883static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
660a5d51
PB		 6884{
6885 struct kvm_segment seg;
6886 int offset;
6887 u16 flags;
6888
6889 kvm_get_segment(vcpu, &seg, n);
6890 offset = 0x7e00 + n * 16;
6891
ee2cd4b7 6892 flags = enter_smm_get_segment_flags(&seg) >> 8;
660a5d51
PB		 6893 put_smstate(u16, buf, offset, seg.selector);
6894 put_smstate(u16, buf, offset + 2, flags);
6895 put_smstate(u32, buf, offset + 4, seg.limit);
6896 put_smstate(u64, buf, offset + 8, seg.base);
6897}
efbb288a 6898#endif
660a5d51 6899
ee2cd4b7 6900static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
660a5d51
PB		 6901{
6902 struct desc_ptr dt;
6903 struct kvm_segment seg;
6904 unsigned long val;
6905 int i;
6906
6907 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6908 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6909 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6910 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6911
6912 for (i = 0; i < 8; i++)
6913 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6914
6915 kvm_get_dr(vcpu, 6, &val);
6916 put_smstate(u32, buf, 0x7fcc, (u32)val);
6917 kvm_get_dr(vcpu, 7, &val);
6918 put_smstate(u32, buf, 0x7fc8, (u32)val);
6919
6920 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6921 put_smstate(u32, buf, 0x7fc4, seg.selector);
6922 put_smstate(u32, buf, 0x7f64, seg.base);
6923 put_smstate(u32, buf, 0x7f60, seg.limit);
ee2cd4b7 6924 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
660a5d51
PB		 6925
6926 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6927 put_smstate(u32, buf, 0x7fc0, seg.selector);
6928 put_smstate(u32, buf, 0x7f80, seg.base);
6929 put_smstate(u32, buf, 0x7f7c, seg.limit);
ee2cd4b7 6930 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
660a5d51
PB		 6931
6932 kvm_x86_ops->get_gdt(vcpu, &dt);
6933 put_smstate(u32, buf, 0x7f74, dt.address);
6934 put_smstate(u32, buf, 0x7f70, dt.size);
6935
6936 kvm_x86_ops->get_idt(vcpu, &dt);
6937 put_smstate(u32, buf, 0x7f58, dt.address);
6938 put_smstate(u32, buf, 0x7f54, dt.size);
6939
6940 for (i = 0; i < 6; i++)
ee2cd4b7 6941 enter_smm_save_seg_32(vcpu, buf, i);
660a5d51
PB		 6942
6943 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6944
6945 /* revision id */
6946 put_smstate(u32, buf, 0x7efc, 0x00020000);
6947 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6948}
6949
ee2cd4b7 6950static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
660a5d51
PB		 6951{
6952#ifdef CONFIG_X86_64
6953 struct desc_ptr dt;
6954 struct kvm_segment seg;
6955 unsigned long val;
6956 int i;
6957
6958 for (i = 0; i < 16; i++)
6959 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6960
6961 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6962 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6963
6964 kvm_get_dr(vcpu, 6, &val);
6965 put_smstate(u64, buf, 0x7f68, val);
6966 kvm_get_dr(vcpu, 7, &val);
6967 put_smstate(u64, buf, 0x7f60, val);
6968
6969 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6970 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6971 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6972
6973 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6974
6975 /* revision id */
6976 put_smstate(u32, buf, 0x7efc, 0x00020064);
6977
6978 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6979
6980 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6981 put_smstate(u16, buf, 0x7e90, seg.selector);
ee2cd4b7 6982 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
660a5d51
PB		 6983 put_smstate(u32, buf, 0x7e94, seg.limit);
6984 put_smstate(u64, buf, 0x7e98, seg.base);
6985
6986 kvm_x86_ops->get_idt(vcpu, &dt);
6987 put_smstate(u32, buf, 0x7e84, dt.size);
6988 put_smstate(u64, buf, 0x7e88, dt.address);
6989
6990 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6991 put_smstate(u16, buf, 0x7e70, seg.selector);
ee2cd4b7 6992 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
660a5d51
PB		 6993 put_smstate(u32, buf, 0x7e74, seg.limit);
6994 put_smstate(u64, buf, 0x7e78, seg.base);
6995
6996 kvm_x86_ops->get_gdt(vcpu, &dt);
6997 put_smstate(u32, buf, 0x7e64, dt.size);
6998 put_smstate(u64, buf, 0x7e68, dt.address);
6999
7000 for (i = 0; i < 6; i++)
ee2cd4b7 7001 enter_smm_save_seg_64(vcpu, buf, i);
660a5d51
PB		 7002#else
7003 WARN_ON_ONCE(1);
7004#endif
7005}
7006
ee2cd4b7 7007static void enter_smm(struct kvm_vcpu *vcpu)
64d60670 7008{
660a5d51 7009 struct kvm_segment cs, ds;
18c3626e 7010 struct desc_ptr dt;
660a5d51
PB		 7011 char buf[512];
7012 u32 cr0;
7013
660a5d51 7014 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
660a5d51 7015 memset(buf, 0, 512);
d6321d49 7016 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
ee2cd4b7 7017 enter_smm_save_state_64(vcpu, buf);
660a5d51 7018 else
ee2cd4b7 7019 enter_smm_save_state_32(vcpu, buf);
660a5d51 7020
0234bf88
LP		 7021 /*
7022 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7023 * vCPU state (e.g. leave guest mode) after we've saved the state into
7024 * the SMM state-save area.
7025 */
7026 kvm_x86_ops->pre_enter_smm(vcpu, buf);
7027
7028 vcpu->arch.hflags |= HF_SMM_MASK;
54bf36aa 7029 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
660a5d51
PB		 7030
7031 if (kvm_x86_ops->get_nmi_mask(vcpu))
7032 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7033 else
7034 kvm_x86_ops->set_nmi_mask(vcpu, true);
7035
7036 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7037 kvm_rip_write(vcpu, 0x8000);
7038
7039 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7040 kvm_x86_ops->set_cr0(vcpu, cr0);
7041 vcpu->arch.cr0 = cr0;
7042
7043 kvm_x86_ops->set_cr4(vcpu, 0);
7044
18c3626e
PB		 7045 /* Undocumented: IDT limit is set to zero on entry to SMM. */
7046 dt.address = dt.size = 0;
7047 kvm_x86_ops->set_idt(vcpu, &dt);
7048
660a5d51
PB		 7049 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7050
7051 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7052 cs.base = vcpu->arch.smbase;
7053
7054 ds.selector = 0;
7055 ds.base = 0;
7056
7057 cs.limit = ds.limit = 0xffffffff;
7058 cs.type = ds.type = 0x3;
7059 cs.dpl = ds.dpl = 0;
7060 cs.db = ds.db = 0;
7061 cs.s = ds.s = 1;
7062 cs.l = ds.l = 0;
7063 cs.g = ds.g = 1;
7064 cs.avl = ds.avl = 0;
7065 cs.present = ds.present = 1;
7066 cs.unusable = ds.unusable = 0;
7067 cs.padding = ds.padding = 0;
7068
7069 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7070 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7071 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7072 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7073 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7074 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7075
d6321d49 7076 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
660a5d51
PB		 7077 kvm_x86_ops->set_efer(vcpu, 0);
7078
7079 kvm_update_cpuid(vcpu);
7080 kvm_mmu_reset_context(vcpu);
64d60670
PB		 7081}
7082
ee2cd4b7 7083static void process_smi(struct kvm_vcpu *vcpu)
c43203ca
PB		 7084{
7085 vcpu->arch.smi_pending = true;
7086 kvm_make_request(KVM_REQ_EVENT, vcpu);
7087}
7088
2860c4b1
PB		 7089void kvm_make_scan_ioapic_request(struct kvm *kvm)
7090{
7091 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7092}
7093
3d81bc7e 7094static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
c7c9c56c 7095{
3d81bc7e
YZ		 7096 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7097 return;
c7c9c56c 7098
6308630b 7099 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
c7c9c56c 7100
b053b2ae 7101 if (irqchip_split(vcpu->kvm))
6308630b 7102 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
db2bdcbb 7103 else {
fa59cc00 7104 if (vcpu->arch.apicv_active)
d62caabb 7105 kvm_x86_ops->sync_pir_to_irr(vcpu);
6308630b 7106 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
db2bdcbb 7107 }
e40ff1d6
LA		 7108
7109 if (is_guest_mode(vcpu))
7110 vcpu->arch.load_eoi_exitmap_pending = true;
7111 else
7112 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7113}
7114
7115static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7116{
7117 u64 eoi_exit_bitmap[4];
7118
7119 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7120 return;
7121
5c919412
AS		 7122 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7123 vcpu_to_synic(vcpu)->vec_bitmap, 256);
7124 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
c7c9c56c
YZ		 7125}
7126
b1394e74
RK		 7127void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7128 unsigned long start, unsigned long end)
7129{
7130 unsigned long apic_address;
7131
7132 /*
7133 * The physical address of apic access page is stored in the VMCS.
7134 * Update it when it becomes invalid.
7135 */
7136 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7137 if (start <= apic_address && apic_address < end)
7138 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7139}
7140
4256f43f
TC		 7141void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7142{
c24ae0dc
TC		 7143 struct page *page = NULL;
7144
35754c98 7145 if (!lapic_in_kernel(vcpu))
f439ed27
PB		 7146 return;
7147
4256f43f
TC		 7148 if (!kvm_x86_ops->set_apic_access_page_addr)
7149 return;
7150
c24ae0dc 7151 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
e8fd5e9e
AA		 7152 if (is_error_page(page))
7153 return;
c24ae0dc
TC		 7154 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7155
7156 /*
7157 * Do not pin apic access page in memory, the MMU notifier
7158 * will call us again if it is migrated or swapped out.
7159 */
7160 put_page(page);
4256f43f
TC		 7161}
7162EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7163
9357d939 7164/*
362c698f 7165 * Returns 1 to let vcpu_run() continue the guest execution loop without
9357d939
TY		 7166 * exiting to the userspace. Otherwise, the value will be returned to the
7167 * userspace.
7168 */
851ba692 7169static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
b6c7a5dc
HB		 7170{
7171 int r;
62a193ed
MG		 7172 bool req_int_win =
7173 dm_request_for_irq_injection(vcpu) &&
7174 kvm_cpu_accept_dm_intr(vcpu);
7175
730dca42 7176 bool req_immediate_exit = false;
b6c7a5dc 7177
2fa6e1e1 7178 if (kvm_request_pending(vcpu)) {
a8eeb04a 7179 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
2e53d63a 7180 kvm_mmu_unload(vcpu);
a8eeb04a 7181 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
2f599714 7182 __kvm_migrate_timers(vcpu);
d828199e
MT		 7183 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7184 kvm_gen_update_masterclock(vcpu->kvm);
0061d53d
MT		 7185 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7186 kvm_gen_kvmclock_update(vcpu);
34c238a1
ZA		 7187 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7188 r = kvm_guest_time_update(vcpu);
8cfdc000
ZA		 7189 if (unlikely(r))
7190 goto out;
7191 }
a8eeb04a 7192 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
4731d4c7 7193 kvm_mmu_sync_roots(vcpu);
a8eeb04a 7194 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
c2ba05cc 7195 kvm_vcpu_flush_tlb(vcpu, true);
a8eeb04a 7196 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
851ba692 7197 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
b93463aa
AK		 7198 r = 0;
7199 goto out;
7200 }
a8eeb04a 7201 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
851ba692 7202 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
bbeac283 7203 vcpu->mmio_needed = 0;
71c4dfaf
JR		 7204 r = 0;
7205 goto out;
7206 }
af585b92
GN		 7207 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7208 /* Page is swapped out. Do synthetic halt */
7209 vcpu->arch.apf.halted = true;
7210 r = 1;
7211 goto out;
7212 }
c9aaa895
GC		 7213 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
7214 record_steal_time(vcpu);
64d60670
PB		 7215 if (kvm_check_request(KVM_REQ_SMI, vcpu))
7216 process_smi(vcpu);
7460fb4a
AK		 7217 if (kvm_check_request(KVM_REQ_NMI, vcpu))
7218 process_nmi(vcpu);
f5132b01 7219 if (kvm_check_request(KVM_REQ_PMU, vcpu))
c6702c9d 7220 kvm_pmu_handle_event(vcpu);
f5132b01 7221 if (kvm_check_request(KVM_REQ_PMI, vcpu))
c6702c9d 7222 kvm_pmu_deliver_pmi(vcpu);
7543a635
SR		 7223 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7224 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7225 if (test_bit(vcpu->arch.pending_ioapic_eoi,
6308630b 7226 vcpu->arch.ioapic_handled_vectors)) {
7543a635
SR		 7227 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7228 vcpu->run->eoi.vector =
7229 vcpu->arch.pending_ioapic_eoi;
7230 r = 0;
7231 goto out;
7232 }
7233 }
3d81bc7e
YZ		 7234 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
7235 vcpu_scan_ioapic(vcpu);
e40ff1d6
LA		 7236 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
7237 vcpu_load_eoi_exitmap(vcpu);
4256f43f
TC		 7238 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
7239 kvm_vcpu_reload_apic_access_page(vcpu);
2ce79189
AS		 7240 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
7241 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7242 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
7243 r = 0;
7244 goto out;
7245 }
e516cebb
AS		 7246 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
7247 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7248 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
7249 r = 0;
7250 goto out;
7251 }
db397571
AS		 7252 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
7253 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
7254 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
7255 r = 0;
7256 goto out;
7257 }
f3b138c5
AS		 7258
7259 /*
7260 * KVM_REQ_HV_STIMER has to be processed after
7261 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
7262 * depend on the guest clock being up-to-date
7263 */
1f4b34f8
AS		 7264 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
7265 kvm_hv_process_stimers(vcpu);
2f52d58c 7266 }
b93463aa 7267
b463a6f7 7268 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
0f1e261e 7269 ++vcpu->stat.req_event;
66450a21
JK		 7270 kvm_apic_accept_events(vcpu);
7271 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
7272 r = 1;
7273 goto out;
7274 }
7275
b6b8a145
JK		 7276 if (inject_pending_event(vcpu, req_int_win) != 0)
7277 req_immediate_exit = true;
321c5658 7278 else {
cc3d967f 7279 /* Enable SMI/NMI/IRQ window open exits if needed.
c43203ca 7280 *
cc3d967f
LP		 7281 * SMIs have three cases:
7282 * 1) They can be nested, and then there is nothing to
7283 * do here because RSM will cause a vmexit anyway.
7284 * 2) There is an ISA-specific reason why SMI cannot be
7285 * injected, and the moment when this changes can be
7286 * intercepted.
7287 * 3) Or the SMI can be pending because
7288 * inject_pending_event has completed the injection
7289 * of an IRQ or NMI from the previous vmexit, and
7290 * then we request an immediate exit to inject the
7291 * SMI.
c43203ca
PB		 7292 */
7293 if (vcpu->arch.smi_pending && !is_smm(vcpu))
cc3d967f
LP		 7294 if (!kvm_x86_ops->enable_smi_window(vcpu))
7295 req_immediate_exit = true;
321c5658
YS		 7296 if (vcpu->arch.nmi_pending)
7297 kvm_x86_ops->enable_nmi_window(vcpu);
7298 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
7299 kvm_x86_ops->enable_irq_window(vcpu);
664f8e26 7300 WARN_ON(vcpu->arch.exception.pending);
321c5658 7301 }
b463a6f7
AK		 7302
7303 if (kvm_lapic_enabled(vcpu)) {
7304 update_cr8_intercept(vcpu);
7305 kvm_lapic_sync_to_vapic(vcpu);
7306 }
7307 }
7308
d8368af8
AK		 7309 r = kvm_mmu_reload(vcpu);
7310 if (unlikely(r)) {
d905c069 7311 goto cancel_injection;
d8368af8
AK		 7312 }
7313
b6c7a5dc
HB		 7314 preempt_disable();
7315
7316 kvm_x86_ops->prepare_guest_switch(vcpu);
b95234c8
PB		 7317
7318 /*
7319 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
7320 * IPI are then delayed after guest entry, which ensures that they
7321 * result in virtual interrupt delivery.
7322 */
7323 local_irq_disable();
6b7e2d09
XG		 7324 vcpu->mode = IN_GUEST_MODE;
7325
01b71917
MT		 7326 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7327
0f127d12 7328 /*
b95234c8 7329 * 1) We should set ->mode before checking ->requests. Please see
cde9af6e 7330 * the comment in kvm_vcpu_exiting_guest_mode().
b95234c8
PB		 7331 *
7332 * 2) For APICv, we should set ->mode before checking PIR.ON. This
7333 * pairs with the memory barrier implicit in pi_test_and_set_on
7334 * (see vmx_deliver_posted_interrupt).
7335 *
7336 * 3) This also orders the write to mode from any reads to the page
7337 * tables done while the VCPU is running. Please see the comment
7338 * in kvm_flush_remote_tlbs.
6b7e2d09 7339 */
01b71917 7340 smp_mb__after_srcu_read_unlock();
b6c7a5dc 7341
b95234c8
PB		 7342 /*
7343 * This handles the case where a posted interrupt was
7344 * notified with kvm_vcpu_kick.
7345 */
fa59cc00
LA		 7346 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
7347 kvm_x86_ops->sync_pir_to_irr(vcpu);
32f88400 7348
2fa6e1e1 7349 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
d94e1dc9 7350 || need_resched() || signal_pending(current)) {
6b7e2d09 7351 vcpu->mode = OUTSIDE_GUEST_MODE;
d94e1dc9 7352 smp_wmb();
6c142801
AK		 7353 local_irq_enable();
7354 preempt_enable();
01b71917 7355 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6c142801 7356 r = 1;
d905c069 7357 goto cancel_injection;
6c142801
AK		 7358 }
7359
fc5b7f3b
DM		 7360 kvm_load_guest_xcr0(vcpu);
7361
c43203ca
PB		 7362 if (req_immediate_exit) {
7363 kvm_make_request(KVM_REQ_EVENT, vcpu);
d6185f20 7364 smp_send_reschedule(vcpu->cpu);
c43203ca 7365 }
d6185f20 7366
8b89fe1f 7367 trace_kvm_entry(vcpu->vcpu_id);
9c48d517
WL		 7368 if (lapic_timer_advance_ns)
7369 wait_lapic_expire(vcpu);
6edaa530 7370 guest_enter_irqoff();
b6c7a5dc 7371
42dbaa5a 7372 if (unlikely(vcpu->arch.switch_db_regs)) {
42dbaa5a
JK		 7373 set_debugreg(0, 7);
7374 set_debugreg(vcpu->arch.eff_db[0], 0);
7375 set_debugreg(vcpu->arch.eff_db[1], 1);
7376 set_debugreg(vcpu->arch.eff_db[2], 2);
7377 set_debugreg(vcpu->arch.eff_db[3], 3);
c77fb5fe 7378 set_debugreg(vcpu->arch.dr6, 6);
ae561ede 7379 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
42dbaa5a 7380 }
b6c7a5dc 7381
851ba692 7382 kvm_x86_ops->run(vcpu);
b6c7a5dc 7383
c77fb5fe
PB		 7384 /*
7385 * Do this here before restoring debug registers on the host. And
7386 * since we do this before handling the vmexit, a DR access vmexit
7387 * can (a) read the correct value of the debug registers, (b) set
7388 * KVM_DEBUGREG_WONT_EXIT again.
7389 */
7390 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
c77fb5fe
PB		 7391 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
7392 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
70e4da7a
PB		 7393 kvm_update_dr0123(vcpu);
7394 kvm_update_dr6(vcpu);
7395 kvm_update_dr7(vcpu);
7396 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
c77fb5fe
PB		 7397 }
7398
24f1e32c
FW		 7399 /*
7400 * If the guest has used debug registers, at least dr7
7401 * will be disabled while returning to the host.
7402 * If we don't have active breakpoints in the host, we don't
7403 * care about the messed up debug address registers. But if
7404 * we have some of them active, restore the old state.
7405 */
59d8eb53 7406 if (hw_breakpoint_active())
24f1e32c 7407 hw_breakpoint_restore();
42dbaa5a 7408
4ba76538 7409 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
1d5f066e 7410
6b7e2d09 7411 vcpu->mode = OUTSIDE_GUEST_MODE;
d94e1dc9 7412 smp_wmb();
a547c6db 7413
fc5b7f3b
DM		 7414 kvm_put_guest_xcr0(vcpu);
7415
dd60d217 7416 kvm_before_interrupt(vcpu);
a547c6db 7417 kvm_x86_ops->handle_external_intr(vcpu);
dd60d217 7418 kvm_after_interrupt(vcpu);
b6c7a5dc
HB		 7419
7420 ++vcpu->stat.exits;
7421
f2485b3e 7422 guest_exit_irqoff();
b6c7a5dc 7423
f2485b3e 7424 local_irq_enable();
b6c7a5dc
HB		 7425 preempt_enable();
7426
f656ce01 7427 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3200f405 7428
b6c7a5dc
HB		 7429 /*
7430 * Profile KVM exit RIPs:
7431 */
7432 if (unlikely(prof_on == KVM_PROFILING)) {
5fdbf976
MT		 7433 unsigned long rip = kvm_rip_read(vcpu);
7434 profile_hit(KVM_PROFILING, (void *)rip);
b6c7a5dc
HB		 7435 }
7436
cc578287
ZA		 7437 if (unlikely(vcpu->arch.tsc_always_catchup))
7438 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
298101da 7439
5cfb1d5a
MT		 7440 if (vcpu->arch.apic_attention)
7441 kvm_lapic_sync_from_vapic(vcpu);
b93463aa 7442
618232e2 7443 vcpu->arch.gpa_available = false;
851ba692 7444 r = kvm_x86_ops->handle_exit(vcpu);
d905c069
MT		 7445 return r;
7446
7447cancel_injection:
7448 kvm_x86_ops->cancel_injection(vcpu);
ae7a2a3f
MT		 7449 if (unlikely(vcpu->arch.apic_attention))
7450 kvm_lapic_sync_from_vapic(vcpu);
d7690175
MT		 7451out:
7452 return r;
7453}
b6c7a5dc 7454
362c698f
PB		 7455static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7456{
bf9f6ac8
FW		 7457 if (!kvm_arch_vcpu_runnable(vcpu) &&
7458 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
9c8fd1ba
PB		 7459 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7460 kvm_vcpu_block(vcpu);
7461 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
bf9f6ac8
FW		 7462
7463 if (kvm_x86_ops->post_block)
7464 kvm_x86_ops->post_block(vcpu);
7465
9c8fd1ba
PB		 7466 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7467 return 1;
7468 }
362c698f
PB		 7469
7470 kvm_apic_accept_events(vcpu);
7471 switch(vcpu->arch.mp_state) {
7472 case KVM_MP_STATE_HALTED:
7473 vcpu->arch.pv.pv_unhalted = false;
7474 vcpu->arch.mp_state =
7475 KVM_MP_STATE_RUNNABLE;
7476 case KVM_MP_STATE_RUNNABLE:
7477 vcpu->arch.apf.halted = false;
7478 break;
7479 case KVM_MP_STATE_INIT_RECEIVED:
7480 break;
7481 default:
7482 return -EINTR;
7483 break;
7484 }
7485 return 1;
7486}
09cec754 7487
5d9bc648
PB		 7488static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7489{
0ad3bed6
PB		 7490 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7491 kvm_x86_ops->check_nested_events(vcpu, false);
7492
5d9bc648
PB		 7493 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7494 !vcpu->arch.apf.halted);
7495}
7496
362c698f 7497static int vcpu_run(struct kvm_vcpu *vcpu)
d7690175
MT		 7498{
7499 int r;
f656ce01 7500 struct kvm *kvm = vcpu->kvm;
d7690175 7501
f656ce01 7502 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
d7690175 7503
362c698f 7504 for (;;) {
58f800d5 7505 if (kvm_vcpu_running(vcpu)) {
851ba692 7506 r = vcpu_enter_guest(vcpu);
bf9f6ac8 7507 } else {
362c698f 7508 r = vcpu_block(kvm, vcpu);
bf9f6ac8
FW		 7509 }
7510
09cec754
GN		 7511 if (r <= 0)
7512 break;
7513
72875d8a 7514 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
09cec754
GN		 7515 if (kvm_cpu_has_pending_timer(vcpu))
7516 kvm_inject_pending_timer_irqs(vcpu);
7517
782d422b
MG		 7518 if (dm_request_for_irq_injection(vcpu) &&
7519 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
4ca7dd8c
PB		 7520 r = 0;
7521 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
09cec754 7522 ++vcpu->stat.request_irq_exits;
362c698f 7523 break;
09cec754 7524 }
af585b92
GN		 7525
7526 kvm_check_async_pf_completion(vcpu);
7527
09cec754
GN		 7528 if (signal_pending(current)) {
7529 r = -EINTR;
851ba692 7530 vcpu->run->exit_reason = KVM_EXIT_INTR;
09cec754 7531 ++vcpu->stat.signal_exits;
362c698f 7532 break;
09cec754
GN		 7533 }
7534 if (need_resched()) {
f656ce01 7535 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
c08ac06a 7536 cond_resched();
f656ce01 7537 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
d7690175 7538 }
b6c7a5dc
HB		 7539 }
7540
f656ce01 7541 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
b6c7a5dc
HB		 7542
7543 return r;
7544}
7545
716d51ab
GN		 7546static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7547{
7548 int r;
7549 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7550 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7551 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7552 if (r != EMULATE_DONE)
7553 return 0;
7554 return 1;
7555}
7556
7557static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7558{
7559 BUG_ON(!vcpu->arch.pio.count);
7560
7561 return complete_emulated_io(vcpu);
7562}
7563
f78146b0
AK		 7564/*
7565 * Implements the following, as a state machine:
7566 *
7567 * read:
7568 * for each fragment
87da7e66
XG		 7569 * for each mmio piece in the fragment
7570 * write gpa, len
7571 * exit
7572 * copy data
f78146b0
AK		 7573 * execute insn
7574 *
7575 * write:
7576 * for each fragment
87da7e66
XG		 7577 * for each mmio piece in the fragment
7578 * write gpa, len
7579 * copy data
7580 * exit
f78146b0 7581 */
716d51ab 7582static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
5287f194
AK		 7583{
7584 struct kvm_run *run = vcpu->run;
f78146b0 7585 struct kvm_mmio_fragment *frag;
87da7e66 7586 unsigned len;
5287f194 7587
716d51ab 7588 BUG_ON(!vcpu->mmio_needed);
5287f194 7589
716d51ab 7590 /* Complete previous fragment */
87da7e66
XG		 7591 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7592 len = min(8u, frag->len);
716d51ab 7593 if (!vcpu->mmio_is_write)
87da7e66
XG		 7594 memcpy(frag->data, run->mmio.data, len);
7595
7596 if (frag->len <= 8) {
7597 /* Switch to the next fragment. */
7598 frag++;
7599 vcpu->mmio_cur_fragment++;
7600 } else {
7601 /* Go forward to the next mmio piece. */
7602 frag->data += len;
7603 frag->gpa += len;
7604 frag->len -= len;
7605 }
7606
a08d3b3b 7607 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
716d51ab 7608 vcpu->mmio_needed = 0;
0912c977
PB		 7609
7610 /* FIXME: return into emulator if single-stepping. */
cef4dea0 7611 if (vcpu->mmio_is_write)
716d51ab
GN		 7612 return 1;
7613 vcpu->mmio_read_completed = 1;
7614 return complete_emulated_io(vcpu);
7615 }
87da7e66 7616
716d51ab
GN		 7617 run->exit_reason = KVM_EXIT_MMIO;
7618 run->mmio.phys_addr = frag->gpa;
7619 if (vcpu->mmio_is_write)
87da7e66
XG		 7620 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7621 run->mmio.len = min(8u, frag->len);
716d51ab
GN		 7622 run->mmio.is_write = vcpu->mmio_is_write;
7623 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7624 return 0;
5287f194
AK		 7625}
7626
b6c7a5dc
HB		 7627int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7628{
7629 int r;
b6c7a5dc 7630
accb757d 7631 vcpu_load(vcpu);
20b7035c 7632 kvm_sigset_activate(vcpu);
5663d8f9
PX		 7633 kvm_load_guest_fpu(vcpu);
7634
a4535290 7635 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
2f173d26
JS		 7636 if (kvm_run->immediate_exit) {
7637 r = -EINTR;
7638 goto out;
7639 }
b6c7a5dc 7640 kvm_vcpu_block(vcpu);
66450a21 7641 kvm_apic_accept_events(vcpu);
72875d8a 7642 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
ac9f6dc0 7643 r = -EAGAIN;
a0595000
JS		 7644 if (signal_pending(current)) {
7645 r = -EINTR;
7646 vcpu->run->exit_reason = KVM_EXIT_INTR;
7647 ++vcpu->stat.signal_exits;
7648 }
ac9f6dc0 7649 goto out;
b6c7a5dc
HB		 7650 }
7651
01643c51
KH		 7652 if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
7653 r = -EINVAL;
7654 goto out;
7655 }
7656
7657 if (vcpu->run->kvm_dirty_regs) {
7658 r = sync_regs(vcpu);
7659 if (r != 0)
7660 goto out;
7661 }
7662
b6c7a5dc 7663 /* re-sync apic's tpr */
35754c98 7664 if (!lapic_in_kernel(vcpu)) {
eea1cff9
AP		 7665 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7666 r = -EINVAL;
7667 goto out;
7668 }
7669 }
b6c7a5dc 7670
716d51ab
GN		 7671 if (unlikely(vcpu->arch.complete_userspace_io)) {
7672 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7673 vcpu->arch.complete_userspace_io = NULL;
7674 r = cui(vcpu);
7675 if (r <= 0)
5663d8f9 7676 goto out;
716d51ab
GN		 7677 } else
7678 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
5287f194 7679
460df4c1
PB		 7680 if (kvm_run->immediate_exit)
7681 r = -EINTR;
7682 else
7683 r = vcpu_run(vcpu);
b6c7a5dc
HB		 7684
7685out:
5663d8f9 7686 kvm_put_guest_fpu(vcpu);
01643c51
KH		 7687 if (vcpu->run->kvm_valid_regs)
7688 store_regs(vcpu);
f1d86e46 7689 post_kvm_run_save(vcpu);
20b7035c 7690 kvm_sigset_deactivate(vcpu);
b6c7a5dc 7691
accb757d 7692 vcpu_put(vcpu);
b6c7a5dc
HB		 7693 return r;
7694}
7695
01643c51 7696static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
b6c7a5dc 7697{
7ae441ea
GN		 7698 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
7699 /*
7700 * We are here if userspace calls get_regs() in the middle of
7701 * instruction emulation. Registers state needs to be copied
4a969980 7702 * back from emulation context to vcpu. Userspace shouldn't do
7ae441ea
GN		 7703 * that usually, but some bad designed PV devices (vmware
7704 * backdoor interface) need this to work
7705 */
dd856efa 7706 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7ae441ea
GN		 7707 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7708 }
5fdbf976
MT		 7709 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
7710 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
7711 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
7712 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
7713 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
7714 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
7715 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7716 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
b6c7a5dc 7717#ifdef CONFIG_X86_64
5fdbf976
MT		 7718 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7719 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7720 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7721 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7722 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7723 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7724 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7725 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
b6c7a5dc
HB		 7726#endif
7727
5fdbf976 7728 regs->rip = kvm_rip_read(vcpu);
91586a3b 7729 regs->rflags = kvm_get_rflags(vcpu);
01643c51 7730}
b6c7a5dc 7731
01643c51
KH		 7732int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7733{
7734 vcpu_load(vcpu);
7735 __get_regs(vcpu, regs);
1fc9b76b 7736 vcpu_put(vcpu);
b6c7a5dc
HB		 7737 return 0;
7738}
7739
01643c51 7740static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
b6c7a5dc 7741{
7ae441ea
GN		 7742 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7743 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7744
5fdbf976
MT		 7745 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7746 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7747 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7748 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7749 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7750 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7751 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7752 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
b6c7a5dc 7753#ifdef CONFIG_X86_64
5fdbf976
MT		 7754 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7755 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7756 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7757 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7758 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7759 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7760 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7761 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
b6c7a5dc
HB		 7762#endif
7763
5fdbf976 7764 kvm_rip_write(vcpu, regs->rip);
d73235d1 7765 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
b6c7a5dc 7766
b4f14abd
JK		 7767 vcpu->arch.exception.pending = false;
7768
3842d135 7769 kvm_make_request(KVM_REQ_EVENT, vcpu);
01643c51 7770}
3842d135 7771
01643c51
KH		 7772int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7773{
7774 vcpu_load(vcpu);
7775 __set_regs(vcpu, regs);
875656fe 7776 vcpu_put(vcpu);
b6c7a5dc
HB		 7777 return 0;
7778}
7779
b6c7a5dc
HB		 7780void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7781{
7782 struct kvm_segment cs;
7783
3e6e0aab 7784 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
b6c7a5dc
HB		 7785 *db = cs.db;
7786 *l = cs.l;
7787}
7788EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7789
01643c51 7790static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
b6c7a5dc 7791{
89a27f4d 7792 struct desc_ptr dt;
b6c7a5dc 7793
3e6e0aab
GT		 7794 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7795 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7796 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7797 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7798 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7799 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
b6c7a5dc 7800
3e6e0aab
GT		 7801 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7802 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
b6c7a5dc
HB		 7803
7804 kvm_x86_ops->get_idt(vcpu, &dt);
89a27f4d
GN		 7805 sregs->idt.limit = dt.size;
7806 sregs->idt.base = dt.address;
b6c7a5dc 7807 kvm_x86_ops->get_gdt(vcpu, &dt);
89a27f4d
GN		 7808 sregs->gdt.limit = dt.size;
7809 sregs->gdt.base = dt.address;
b6c7a5dc 7810
4d4ec087 7811 sregs->cr0 = kvm_read_cr0(vcpu);
ad312c7c 7812 sregs->cr2 = vcpu->arch.cr2;
9f8fe504 7813 sregs->cr3 = kvm_read_cr3(vcpu);
fc78f519 7814 sregs->cr4 = kvm_read_cr4(vcpu);
2d3ad1f4 7815 sregs->cr8 = kvm_get_cr8(vcpu);
f6801dff 7816 sregs->efer = vcpu->arch.efer;
b6c7a5dc
HB		 7817 sregs->apic_base = kvm_get_apic_base(vcpu);
7818
923c61bb 7819 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
b6c7a5dc 7820
04140b41 7821 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
14d0bc1f
GN		 7822 set_bit(vcpu->arch.interrupt.nr,
7823 (unsigned long *)sregs->interrupt_bitmap);
01643c51 7824}
16d7a191 7825
01643c51
KH		 7826int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7827 struct kvm_sregs *sregs)
7828{
7829 vcpu_load(vcpu);
7830 __get_sregs(vcpu, sregs);
bcdec41c 7831 vcpu_put(vcpu);
b6c7a5dc
HB		 7832 return 0;
7833}
7834
62d9f0db
MT		 7835int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7836 struct kvm_mp_state *mp_state)
7837{
fd232561
CD		 7838 vcpu_load(vcpu);
7839
66450a21 7840 kvm_apic_accept_events(vcpu);
6aef266c
SV		 7841 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7842 vcpu->arch.pv.pv_unhalted)
7843 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7844 else
7845 mp_state->mp_state = vcpu->arch.mp_state;
7846
fd232561 7847 vcpu_put(vcpu);
62d9f0db
MT		 7848 return 0;
7849}
7850
7851int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7852 struct kvm_mp_state *mp_state)
7853{
e83dff5e
CD		 7854 int ret = -EINVAL;
7855
7856 vcpu_load(vcpu);
7857
bce87cce 7858 if (!lapic_in_kernel(vcpu) &&
66450a21 7859 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
e83dff5e 7860 goto out;
66450a21 7861
28bf2888
DH		 7862 /* INITs are latched while in SMM */
7863 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
7864 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
7865 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
e83dff5e 7866 goto out;
28bf2888 7867
66450a21
JK		 7868 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7869 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7870 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7871 } else
7872 vcpu->arch.mp_state = mp_state->mp_state;
3842d135 7873 kvm_make_request(KVM_REQ_EVENT, vcpu);
e83dff5e
CD		 7874
7875 ret = 0;
7876out:
7877 vcpu_put(vcpu);
7878 return ret;
62d9f0db
MT		 7879}
7880
7f3d35fd
KW		 7881int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7882 int reason, bool has_error_code, u32 error_code)
b6c7a5dc 7883{
9d74191a 7884 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8ec4722d 7885 int ret;
e01c2426 7886
8ec4722d 7887 init_emulate_ctxt(vcpu);
c697518a 7888
7f3d35fd 7889 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9d74191a 7890 has_error_code, error_code);
c697518a 7891
c697518a 7892 if (ret)
19d04437 7893 return EMULATE_FAIL;
37817f29 7894
9d74191a
TY		 7895 kvm_rip_write(vcpu, ctxt->eip);
7896 kvm_set_rflags(vcpu, ctxt->eflags);
3842d135 7897 kvm_make_request(KVM_REQ_EVENT, vcpu);
19d04437 7898 return EMULATE_DONE;
37817f29
IE		 7899}
7900EXPORT_SYMBOL_GPL(kvm_task_switch);
7901
f2981033
LT		 7902int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
7903{
37b95951 7904 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
f2981033
LT		 7905 /*
7906 * When EFER.LME and CR0.PG are set, the processor is in
7907 * 64-bit mode (though maybe in a 32-bit code segment).
7908 * CR4.PAE and EFER.LMA must be set.
7909 */
37b95951 7910 if (!(sregs->cr4 & X86_CR4_PAE)
f2981033
LT		 7911 || !(sregs->efer & EFER_LMA))
7912 return -EINVAL;
7913 } else {
7914 /*
7915 * Not in 64-bit mode: EFER.LMA is clear and the code
7916 * segment cannot be 64-bit.
7917 */
7918 if (sregs->efer & EFER_LMA || sregs->cs.l)
7919 return -EINVAL;
7920 }
7921
7922 return 0;
7923}
7924
01643c51 7925static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
b6c7a5dc 7926{
58cb628d 7927 struct msr_data apic_base_msr;
b6c7a5dc 7928 int mmu_reset_needed = 0;
63f42e02 7929 int pending_vec, max_bits, idx;
89a27f4d 7930 struct desc_ptr dt;
b4ef9d4e
CD		 7931 int ret = -EINVAL;
7932
d6321d49
RK		 7933 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
7934 (sregs->cr4 & X86_CR4_OSXSAVE))
b4ef9d4e 7935 goto out;
6d1068b3 7936
f2981033 7937 if (kvm_valid_sregs(vcpu, sregs))
8dbfb2bf 7938 goto out;
f2981033 7939
d3802286
JM		 7940 apic_base_msr.data = sregs->apic_base;
7941 apic_base_msr.host_initiated = true;
7942 if (kvm_set_apic_base(vcpu, &apic_base_msr))
b4ef9d4e 7943 goto out;
6d1068b3 7944
89a27f4d
GN		 7945 dt.size = sregs->idt.limit;
7946 dt.address = sregs->idt.base;
b6c7a5dc 7947 kvm_x86_ops->set_idt(vcpu, &dt);
89a27f4d
GN		 7948 dt.size = sregs->gdt.limit;
7949 dt.address = sregs->gdt.base;
b6c7a5dc
HB		 7950 kvm_x86_ops->set_gdt(vcpu, &dt);
7951
ad312c7c 7952 vcpu->arch.cr2 = sregs->cr2;
9f8fe504 7953 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
dc7e795e 7954 vcpu->arch.cr3 = sregs->cr3;
aff48baa 7955 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
b6c7a5dc 7956
2d3ad1f4 7957 kvm_set_cr8(vcpu, sregs->cr8);
b6c7a5dc 7958
f6801dff 7959 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
b6c7a5dc 7960 kvm_x86_ops->set_efer(vcpu, sregs->efer);
b6c7a5dc 7961
4d4ec087 7962 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
b6c7a5dc 7963 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
d7306163 7964 vcpu->arch.cr0 = sregs->cr0;
b6c7a5dc 7965
fc78f519 7966 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
b6c7a5dc 7967 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
b9baba86 7968 if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
00b27a3e 7969 kvm_update_cpuid(vcpu);
63f42e02
XG		 7970
7971 idx = srcu_read_lock(&vcpu->kvm->srcu);
7c93be44 7972 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
9f8fe504 7973 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7c93be44
MT		 7974 mmu_reset_needed = 1;
7975 }
63f42e02 7976 srcu_read_unlock(&vcpu->kvm->srcu, idx);
b6c7a5dc
HB		 7977
7978 if (mmu_reset_needed)
7979 kvm_mmu_reset_context(vcpu);
7980
a50abc3b 7981 max_bits = KVM_NR_INTERRUPTS;
923c61bb
GN		 7982 pending_vec = find_first_bit(
7983 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7984 if (pending_vec < max_bits) {
66fd3f7f 7985 kvm_queue_interrupt(vcpu, pending_vec, false);
923c61bb 7986 pr_debug("Set back pending irq %d\n", pending_vec);
b6c7a5dc
HB		 7987 }
7988
3e6e0aab
GT		 7989 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7990 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7991 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7992 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7993 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7994 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
b6c7a5dc 7995
3e6e0aab
GT		 7996 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7997 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
b6c7a5dc 7998
5f0269f5
ME		 7999 update_cr8_intercept(vcpu);
8000
9c3e4aab 8001 /* Older userspace won't unhalt the vcpu on reset. */
c5af89b6 8002 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9c3e4aab 8003 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
3eeb3288 8004 !is_protmode(vcpu))
9c3e4aab
MT		 8005 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8006
3842d135
AK		 8007 kvm_make_request(KVM_REQ_EVENT, vcpu);
8008
b4ef9d4e
CD		 8009 ret = 0;
8010out:
01643c51
KH		 8011 return ret;
8012}
8013
8014int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8015 struct kvm_sregs *sregs)
8016{
8017 int ret;
8018
8019 vcpu_load(vcpu);
8020 ret = __set_sregs(vcpu, sregs);
b4ef9d4e
CD		 8021 vcpu_put(vcpu);
8022 return ret;
b6c7a5dc
HB		 8023}
8024
d0bfb940
JK		 8025int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8026 struct kvm_guest_debug *dbg)
b6c7a5dc 8027{
355be0b9 8028 unsigned long rflags;
ae675ef0 8029 int i, r;
b6c7a5dc 8030
66b56562
CD		 8031 vcpu_load(vcpu);
8032
4f926bf2
JK		 8033 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8034 r = -EBUSY;
8035 if (vcpu->arch.exception.pending)
2122ff5e 8036 goto out;
4f926bf2
JK		 8037 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8038 kvm_queue_exception(vcpu, DB_VECTOR);
8039 else
8040 kvm_queue_exception(vcpu, BP_VECTOR);
8041 }
8042
91586a3b
JK		 8043 /*
8044 * Read rflags as long as potentially injected trace flags are still
8045 * filtered out.
8046 */
8047 rflags = kvm_get_rflags(vcpu);
355be0b9
JK		 8048
8049 vcpu->guest_debug = dbg->control;
8050 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8051 vcpu->guest_debug = 0;
8052
8053 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
ae675ef0
JK		 8054 for (i = 0; i < KVM_NR_DB_REGS; ++i)
8055 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
c8639010 8056 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
ae675ef0
JK		 8057 } else {
8058 for (i = 0; i < KVM_NR_DB_REGS; i++)
8059 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
ae675ef0 8060 }
c8639010 8061 kvm_update_dr7(vcpu);
ae675ef0 8062
f92653ee
JK		 8063 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8064 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8065 get_segment_base(vcpu, VCPU_SREG_CS);
94fe45da 8066
91586a3b
JK		 8067 /*
8068 * Trigger an rflags update that will inject or remove the trace
8069 * flags.
8070 */
8071 kvm_set_rflags(vcpu, rflags);
b6c7a5dc 8072
a96036b8 8073 kvm_x86_ops->update_bp_intercept(vcpu);
b6c7a5dc 8074
4f926bf2 8075 r = 0;
d0bfb940 8076
2122ff5e 8077out:
66b56562 8078 vcpu_put(vcpu);
b6c7a5dc
HB		 8079 return r;
8080}
8081
8b006791
ZX		 8082/*
8083 * Translate a guest virtual address to a guest physical address.
8084 */
8085int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8086 struct kvm_translation *tr)
8087{
8088 unsigned long vaddr = tr->linear_address;
8089 gpa_t gpa;
f656ce01 8090 int idx;
8b006791 8091
1da5b61d
CD		 8092 vcpu_load(vcpu);
8093
f656ce01 8094 idx = srcu_read_lock(&vcpu->kvm->srcu);
1871c602 8095 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
f656ce01 8096 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8b006791
ZX		 8097 tr->physical_address = gpa;
8098 tr->valid = gpa != UNMAPPED_GVA;
8099 tr->writeable = 1;
8100 tr->usermode = 0;
8b006791 8101
1da5b61d 8102 vcpu_put(vcpu);
8b006791
ZX		 8103 return 0;
8104}
8105
d0752060
HB		 8106int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8107{
1393123e 8108 struct fxregs_state *fxsave;
d0752060 8109
1393123e 8110 vcpu_load(vcpu);
d0752060 8111
1393123e 8112 fxsave = &vcpu->arch.guest_fpu.state.fxsave;
d0752060
HB		 8113 memcpy(fpu->fpr, fxsave->st_space, 128);
8114 fpu->fcw = fxsave->cwd;
8115 fpu->fsw = fxsave->swd;
8116 fpu->ftwx = fxsave->twd;
8117 fpu->last_opcode = fxsave->fop;
8118 fpu->last_ip = fxsave->rip;
8119 fpu->last_dp = fxsave->rdp;
8120 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
8121
1393123e 8122 vcpu_put(vcpu);
d0752060
HB		 8123 return 0;
8124}
8125
8126int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8127{
6a96bc7f
CD		 8128 struct fxregs_state *fxsave;
8129
8130 vcpu_load(vcpu);
8131
8132 fxsave = &vcpu->arch.guest_fpu.state.fxsave;
d0752060 8133
d0752060
HB		 8134 memcpy(fxsave->st_space, fpu->fpr, 128);
8135 fxsave->cwd = fpu->fcw;
8136 fxsave->swd = fpu->fsw;
8137 fxsave->twd = fpu->ftwx;
8138 fxsave->fop = fpu->last_opcode;
8139 fxsave->rip = fpu->last_ip;
8140 fxsave->rdp = fpu->last_dp;
8141 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
8142
6a96bc7f 8143 vcpu_put(vcpu);
d0752060
HB		 8144 return 0;
8145}
8146
01643c51
KH		 8147static void store_regs(struct kvm_vcpu *vcpu)
8148{
8149 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8150
8151 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8152 __get_regs(vcpu, &vcpu->run->s.regs.regs);
8153
8154 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8155 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8156
8157 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8158 kvm_vcpu_ioctl_x86_get_vcpu_events(
8159 vcpu, &vcpu->run->s.regs.events);
8160}
8161
8162static int sync_regs(struct kvm_vcpu *vcpu)
8163{
8164 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
8165 return -EINVAL;
8166
8167 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
8168 __set_regs(vcpu, &vcpu->run->s.regs.regs);
8169 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
8170 }
8171 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
8172 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
8173 return -EINVAL;
8174 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
8175 }
8176 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
8177 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
8178 vcpu, &vcpu->run->s.regs.events))
8179 return -EINVAL;
8180 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
8181 }
8182
8183 return 0;
8184}
8185
0ee6a517 8186static void fx_init(struct kvm_vcpu *vcpu)
d0752060 8187{
bf935b0b 8188 fpstate_init(&vcpu->arch.guest_fpu.state);
782511b0 8189 if (boot_cpu_has(X86_FEATURE_XSAVES))
7366ed77 8190 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
df1daba7 8191 host_xcr0 | XSTATE_COMPACTION_ENABLED;
d0752060 8192
2acf923e
DC		 8193 /*
8194 * Ensure guest xcr0 is valid for loading
8195 */
d91cab78 8196 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
2acf923e 8197
ad312c7c 8198 vcpu->arch.cr0 |= X86_CR0_ET;
d0752060 8199}
d0752060 8200
f775b13e 8201/* Swap (qemu) user FPU context for the guest FPU context. */
d0752060
HB		 8202void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8203{
f775b13e
RR		 8204 preempt_disable();
8205 copy_fpregs_to_fpstate(&vcpu->arch.user_fpu);
38cfd5e3
PB		 8206 /* PKRU is separately restored in kvm_x86_ops->run. */
8207 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state,
8208 ~XFEATURE_MASK_PKRU);
f775b13e 8209 preempt_enable();
0c04851c 8210 trace_kvm_fpu(1);
d0752060 8211}
d0752060 8212
f775b13e 8213/* When vcpu_run ends, restore user space FPU context. */
d0752060
HB		 8214void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8215{
f775b13e 8216 preempt_disable();
4f836347 8217 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
f775b13e
RR		 8218 copy_kernel_to_fpregs(&vcpu->arch.user_fpu.state);
8219 preempt_enable();
f096ed85 8220 ++vcpu->stat.fpu_reload;
0c04851c 8221 trace_kvm_fpu(0);
d0752060 8222}
e9b11c17
ZX		 8223
8224void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
8225{
bd768e14
IY		 8226 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
8227
12f9a48f 8228 kvmclock_reset(vcpu);
7f1ea208 8229
e9b11c17 8230 kvm_x86_ops->vcpu_free(vcpu);
bd768e14 8231 free_cpumask_var(wbinvd_dirty_mask);
e9b11c17
ZX		 8232}
8233
8234struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
8235 unsigned int id)
8236{
c447e76b
LL		 8237 struct kvm_vcpu *vcpu;
8238
b0c39dc6 8239 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6755bae8
ZA		 8240 printk_once(KERN_WARNING
8241 "kvm: SMP vm created on host with unstable TSC; "
8242 "guest TSC will not be reliable\n");
c447e76b
LL		 8243
8244 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
8245
c447e76b 8246 return vcpu;
26e5215f 8247}
e9b11c17 8248
26e5215f
AK		 8249int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
8250{
19efffa2 8251 kvm_vcpu_mtrr_init(vcpu);
ec7660cc 8252 vcpu_load(vcpu);
d28bc9dd 8253 kvm_vcpu_reset(vcpu, false);
8a3c1a33 8254 kvm_mmu_setup(vcpu);
e9b11c17 8255 vcpu_put(vcpu);
ec7660cc 8256 return 0;
e9b11c17
ZX		 8257}
8258
31928aa5 8259void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
42897d86 8260{
8fe8ab46 8261 struct msr_data msr;
332967a3 8262 struct kvm *kvm = vcpu->kvm;
42897d86 8263
d3457c87
RK		 8264 kvm_hv_vcpu_postcreate(vcpu);
8265
ec7660cc 8266 if (mutex_lock_killable(&vcpu->mutex))
31928aa5 8267 return;
ec7660cc 8268 vcpu_load(vcpu);
8fe8ab46
WA		 8269 msr.data = 0x0;
8270 msr.index = MSR_IA32_TSC;
8271 msr.host_initiated = true;
8272 kvm_write_tsc(vcpu, &msr);
42897d86 8273 vcpu_put(vcpu);
ec7660cc 8274 mutex_unlock(&vcpu->mutex);
42897d86 8275
630994b3
MT		 8276 if (!kvmclock_periodic_sync)
8277 return;
8278
332967a3
AJ		 8279 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
8280 KVMCLOCK_SYNC_PERIOD);
42897d86
MT		 8281}
8282
d40ccc62 8283void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
e9b11c17 8284{
344d9588
GN		 8285 vcpu->arch.apf.msr_val = 0;
8286
ec7660cc 8287 vcpu_load(vcpu);
e9b11c17
ZX		 8288 kvm_mmu_unload(vcpu);
8289 vcpu_put(vcpu);
8290
8291 kvm_x86_ops->vcpu_free(vcpu);
8292}
8293
d28bc9dd 8294void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
e9b11c17 8295{
b7e31be3
RK		 8296 kvm_lapic_reset(vcpu, init_event);
8297
e69fab5d
PB		 8298 vcpu->arch.hflags = 0;
8299
c43203ca 8300 vcpu->arch.smi_pending = 0;
52797bf9 8301 vcpu->arch.smi_count = 0;
7460fb4a
AK		 8302 atomic_set(&vcpu->arch.nmi_queued, 0);
8303 vcpu->arch.nmi_pending = 0;
448fa4a9 8304 vcpu->arch.nmi_injected = false;
5f7552d4
NA		 8305 kvm_clear_interrupt_queue(vcpu);
8306 kvm_clear_exception_queue(vcpu);
664f8e26 8307 vcpu->arch.exception.pending = false;
448fa4a9 8308
42dbaa5a 8309 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
ae561ede 8310 kvm_update_dr0123(vcpu);
6f43ed01 8311 vcpu->arch.dr6 = DR6_INIT;
73aaf249 8312 kvm_update_dr6(vcpu);
42dbaa5a 8313 vcpu->arch.dr7 = DR7_FIXED_1;
c8639010 8314 kvm_update_dr7(vcpu);
42dbaa5a 8315
1119022c
NA		 8316 vcpu->arch.cr2 = 0;
8317
3842d135 8318 kvm_make_request(KVM_REQ_EVENT, vcpu);
344d9588 8319 vcpu->arch.apf.msr_val = 0;
c9aaa895 8320 vcpu->arch.st.msr_val = 0;
3842d135 8321
12f9a48f
GC		 8322 kvmclock_reset(vcpu);
8323
af585b92
GN		 8324 kvm_clear_async_pf_completion_queue(vcpu);
8325 kvm_async_pf_hash_reset(vcpu);
8326 vcpu->arch.apf.halted = false;
3842d135 8327
a554d207
WL		 8328 if (kvm_mpx_supported()) {
8329 void *mpx_state_buffer;
8330
8331 /*
8332 * To avoid have the INIT path from kvm_apic_has_events() that be
8333 * called with loaded FPU and does not let userspace fix the state.
8334 */
f775b13e
RR		 8335 if (init_event)
8336 kvm_put_guest_fpu(vcpu);
a554d207
WL		 8337 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
8338 XFEATURE_MASK_BNDREGS);
8339 if (mpx_state_buffer)
8340 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
8341 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
8342 XFEATURE_MASK_BNDCSR);
8343 if (mpx_state_buffer)
8344 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
f775b13e
RR		 8345 if (init_event)
8346 kvm_load_guest_fpu(vcpu);
a554d207
WL		 8347 }
8348
64d60670 8349 if (!init_event) {
d28bc9dd 8350 kvm_pmu_reset(vcpu);
64d60670 8351 vcpu->arch.smbase = 0x30000;
db2336a8
KH		 8352
8353 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
8354 vcpu->arch.msr_misc_features_enables = 0;
a554d207
WL		 8355
8356 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
64d60670 8357 }
f5132b01 8358
66f7b72e
JS		 8359 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
8360 vcpu->arch.regs_avail = ~0;
8361 vcpu->arch.regs_dirty = ~0;
8362
a554d207
WL		 8363 vcpu->arch.ia32_xss = 0;
8364
d28bc9dd 8365 kvm_x86_ops->vcpu_reset(vcpu, init_event);
e9b11c17
ZX		 8366}
8367
2b4a273b 8368void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
66450a21
JK		 8369{
8370 struct kvm_segment cs;
8371
8372 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8373 cs.selector = vector << 8;
8374 cs.base = vector << 12;
8375 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8376 kvm_rip_write(vcpu, 0);
e9b11c17
ZX		 8377}
8378
13a34e06 8379int kvm_arch_hardware_enable(void)
e9b11c17 8380{
ca84d1a2
ZA		 8381 struct kvm *kvm;
8382 struct kvm_vcpu *vcpu;
8383 int i;
0dd6a6ed
ZA		 8384 int ret;
8385 u64 local_tsc;
8386 u64 max_tsc = 0;
8387 bool stable, backwards_tsc = false;
18863bdd
AK		 8388
8389 kvm_shared_msr_cpu_online();
13a34e06 8390 ret = kvm_x86_ops->hardware_enable();
0dd6a6ed
ZA		 8391 if (ret != 0)
8392 return ret;
8393
4ea1636b 8394 local_tsc = rdtsc();
b0c39dc6 8395 stable = !kvm_check_tsc_unstable();
0dd6a6ed
ZA		 8396 list_for_each_entry(kvm, &vm_list, vm_list) {
8397 kvm_for_each_vcpu(i, vcpu, kvm) {
8398 if (!stable && vcpu->cpu == smp_processor_id())
105b21bb 8399 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
0dd6a6ed
ZA		 8400 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
8401 backwards_tsc = true;
8402 if (vcpu->arch.last_host_tsc > max_tsc)
8403 max_tsc = vcpu->arch.last_host_tsc;
8404 }
8405 }
8406 }
8407
8408 /*
8409 * Sometimes, even reliable TSCs go backwards. This happens on
8410 * platforms that reset TSC during suspend or hibernate actions, but
8411 * maintain synchronization. We must compensate. Fortunately, we can
8412 * detect that condition here, which happens early in CPU bringup,
8413 * before any KVM threads can be running. Unfortunately, we can't
8414 * bring the TSCs fully up to date with real time, as we aren't yet far
8415 * enough into CPU bringup that we know how much real time has actually
108b249c 8416 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
0dd6a6ed
ZA		 8417 * variables that haven't been updated yet.
8418 *
8419 * So we simply find the maximum observed TSC above, then record the
8420 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
8421 * the adjustment will be applied. Note that we accumulate
8422 * adjustments, in case multiple suspend cycles happen before some VCPU
8423 * gets a chance to run again. In the event that no KVM threads get a
8424 * chance to run, we will miss the entire elapsed period, as we'll have
8425 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
8426 * loose cycle time. This isn't too big a deal, since the loss will be
8427 * uniform across all VCPUs (not to mention the scenario is extremely
8428 * unlikely). It is possible that a second hibernate recovery happens
8429 * much faster than a first, causing the observed TSC here to be
8430 * smaller; this would require additional padding adjustment, which is
8431 * why we set last_host_tsc to the local tsc observed here.
8432 *
8433 * N.B. - this code below runs only on platforms with reliable TSC,
8434 * as that is the only way backwards_tsc is set above. Also note
8435 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
8436 * have the same delta_cyc adjustment applied if backwards_tsc
8437 * is detected. Note further, this adjustment is only done once,
8438 * as we reset last_host_tsc on all VCPUs to stop this from being
8439 * called multiple times (one for each physical CPU bringup).
8440 *
4a969980 8441 * Platforms with unreliable TSCs don't have to deal with this, they
0dd6a6ed
ZA		 8442 * will be compensated by the logic in vcpu_load, which sets the TSC to
8443 * catchup mode. This will catchup all VCPUs to real time, but cannot
8444 * guarantee that they stay in perfect synchronization.
8445 */
8446 if (backwards_tsc) {
8447 u64 delta_cyc = max_tsc - local_tsc;
8448 list_for_each_entry(kvm, &vm_list, vm_list) {
a826faf1 8449 kvm->arch.backwards_tsc_observed = true;
0dd6a6ed
ZA		 8450 kvm_for_each_vcpu(i, vcpu, kvm) {
8451 vcpu->arch.tsc_offset_adjustment += delta_cyc;
8452 vcpu->arch.last_host_tsc = local_tsc;
105b21bb 8453 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
0dd6a6ed
ZA		 8454 }
8455
8456 /*
8457 * We have to disable TSC offset matching.. if you were
8458 * booting a VM while issuing an S4 host suspend....
8459 * you may have some problem. Solving this issue is
8460 * left as an exercise to the reader.
8461 */
8462 kvm->arch.last_tsc_nsec = 0;
8463 kvm->arch.last_tsc_write = 0;
8464 }
8465
8466 }
8467 return 0;
e9b11c17
ZX		 8468}
8469
13a34e06 8470void kvm_arch_hardware_disable(void)
e9b11c17 8471{
13a34e06
RK		 8472 kvm_x86_ops->hardware_disable();
8473 drop_user_return_notifiers();
e9b11c17
ZX		 8474}
8475
8476int kvm_arch_hardware_setup(void)
8477{
9e9c3fe4
NA		 8478 int r;
8479
8480 r = kvm_x86_ops->hardware_setup();
8481 if (r != 0)
8482 return r;
8483
35181e86
HZ		 8484 if (kvm_has_tsc_control) {
8485 /*
8486 * Make sure the user can only configure tsc_khz values that
8487 * fit into a signed integer.
8488 * A min value is not calculated needed because it will always
8489 * be 1 on all machines.
8490 */
8491 u64 max = min(0x7fffffffULL,
8492 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
8493 kvm_max_guest_tsc_khz = max;
8494
ad721883 8495 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
35181e86 8496 }
ad721883 8497
9e9c3fe4
NA		 8498 kvm_init_msr_list();
8499 return 0;
e9b11c17
ZX		 8500}
8501
8502void kvm_arch_hardware_unsetup(void)
8503{
8504 kvm_x86_ops->hardware_unsetup();
8505}
8506
8507void kvm_arch_check_processor_compat(void *rtn)
8508{
8509 kvm_x86_ops->check_processor_compatibility(rtn);
d71ba788
PB		 8510}
8511
8512bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
8513{
8514 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
8515}
8516EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
8517
8518bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
8519{
8520 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
e9b11c17
ZX		 8521}
8522
54e9818f 8523struct static_key kvm_no_apic_vcpu __read_mostly;
bce87cce 8524EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
54e9818f 8525
e9b11c17
ZX		 8526int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
8527{
8528 struct page *page;
e9b11c17
ZX		 8529 int r;
8530
b2a05fef 8531 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
9aabc88f 8532 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
26de7988 8533 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
a4535290 8534 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
e9b11c17 8535 else
a4535290 8536 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
e9b11c17
ZX		 8537
8538 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
8539 if (!page) {
8540 r = -ENOMEM;
8541 goto fail;
8542 }
ad312c7c 8543 vcpu->arch.pio_data = page_address(page);
e9b11c17 8544
cc578287 8545 kvm_set_tsc_khz(vcpu, max_tsc_khz);
c285545f 8546
e9b11c17
ZX		 8547 r = kvm_mmu_create(vcpu);
8548 if (r < 0)
8549 goto fail_free_pio_data;
8550
26de7988 8551 if (irqchip_in_kernel(vcpu->kvm)) {
e9b11c17
ZX		 8552 r = kvm_create_lapic(vcpu);
8553 if (r < 0)
8554 goto fail_mmu_destroy;
54e9818f
GN		 8555 } else
8556 static_key_slow_inc(&kvm_no_apic_vcpu);
e9b11c17 8557
890ca9ae
HY		 8558 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
8559 GFP_KERNEL);
8560 if (!vcpu->arch.mce_banks) {
8561 r = -ENOMEM;
443c39bc 8562 goto fail_free_lapic;
890ca9ae
HY		 8563 }
8564 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
8565
f1797359
WY		 8566 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
8567 r = -ENOMEM;
f5f48ee1 8568 goto fail_free_mce_banks;
f1797359 8569 }
f5f48ee1 8570
0ee6a517 8571 fx_init(vcpu);
66f7b72e 8572
4344ee98 8573 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
d7876f1b 8574
5a4f55cd
EK		 8575 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
8576
74545705
RK		 8577 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
8578
af585b92 8579 kvm_async_pf_hash_reset(vcpu);
f5132b01 8580 kvm_pmu_init(vcpu);
af585b92 8581
1c1a9ce9 8582 vcpu->arch.pending_external_vector = -1;
de63ad4c 8583 vcpu->arch.preempted_in_kernel = false;
1c1a9ce9 8584
5c919412
AS		 8585 kvm_hv_vcpu_init(vcpu);
8586
e9b11c17 8587 return 0;
0ee6a517 8588
f5f48ee1
SY		 8589fail_free_mce_banks:
8590 kfree(vcpu->arch.mce_banks);
443c39bc
WY		 8591fail_free_lapic:
8592 kvm_free_lapic(vcpu);
e9b11c17
ZX		 8593fail_mmu_destroy:
8594 kvm_mmu_destroy(vcpu);
8595fail_free_pio_data:
ad312c7c 8596 free_page((unsigned long)vcpu->arch.pio_data);
e9b11c17
ZX		 8597fail:
8598 return r;
8599}
8600
8601void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
8602{
f656ce01
MT		 8603 int idx;
8604
1f4b34f8 8605 kvm_hv_vcpu_uninit(vcpu);
f5132b01 8606 kvm_pmu_destroy(vcpu);
36cb93fd 8607 kfree(vcpu->arch.mce_banks);
e9b11c17 8608 kvm_free_lapic(vcpu);
f656ce01 8609 idx = srcu_read_lock(&vcpu->kvm->srcu);
e9b11c17 8610 kvm_mmu_destroy(vcpu);
f656ce01 8611 srcu_read_unlock(&vcpu->kvm->srcu, idx);
ad312c7c 8612 free_page((unsigned long)vcpu->arch.pio_data);
35754c98 8613 if (!lapic_in_kernel(vcpu))
54e9818f 8614 static_key_slow_dec(&kvm_no_apic_vcpu);
e9b11c17 8615}
d19a9cd2 8616
e790d9ef
RK		 8617void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
8618{
ae97a3b8 8619 kvm_x86_ops->sched_in(vcpu, cpu);
e790d9ef
RK		 8620}
8621
e08b9637 8622int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
d19a9cd2 8623{
e08b9637
CO		 8624 if (type)
8625 return -EINVAL;
8626
6ef768fa 8627 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
f05e70ac 8628 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
365c8868 8629 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
4d5c5d0f 8630 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
e0f0bbc5 8631 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
d19a9cd2 8632
5550af4d
SY		 8633 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
8634 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7a84428a
AW		 8635 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
8636 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
8637 &kvm->arch.irq_sources_bitmap);
5550af4d 8638
038f8c11 8639 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
1e08ec4a 8640 mutex_init(&kvm->arch.apic_map_lock);
d828199e
MT		 8641 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
8642
108b249c 8643 kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
d828199e 8644 pvclock_update_vm_gtod_copy(kvm);
53f658b3 8645
7e44e449 8646 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
332967a3 8647 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7e44e449 8648
cbc0236a 8649 kvm_hv_init_vm(kvm);
0eb05bf2 8650 kvm_page_track_init(kvm);
13d268ca 8651 kvm_mmu_init_vm(kvm);
0eb05bf2 8652
03543133
SS		 8653 if (kvm_x86_ops->vm_init)
8654 return kvm_x86_ops->vm_init(kvm);
8655
d89f5eff 8656 return 0;
d19a9cd2
ZX		 8657}
8658
8659static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
8660{
ec7660cc 8661 vcpu_load(vcpu);
d19a9cd2
ZX		 8662 kvm_mmu_unload(vcpu);
8663 vcpu_put(vcpu);
8664}
8665
8666static void kvm_free_vcpus(struct kvm *kvm)
8667{
8668 unsigned int i;
988a2cae 8669 struct kvm_vcpu *vcpu;
d19a9cd2
ZX		 8670
8671 /*
8672 * Unpin any mmu pages first.
8673 */
af585b92
GN		 8674 kvm_for_each_vcpu(i, vcpu, kvm) {
8675 kvm_clear_async_pf_completion_queue(vcpu);
988a2cae 8676 kvm_unload_vcpu_mmu(vcpu);
af585b92 8677 }
988a2cae
GN		 8678 kvm_for_each_vcpu(i, vcpu, kvm)
8679 kvm_arch_vcpu_free(vcpu);
8680
8681 mutex_lock(&kvm->lock);
8682 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
8683 kvm->vcpus[i] = NULL;
d19a9cd2 8684
988a2cae
GN		 8685 atomic_set(&kvm->online_vcpus, 0);
8686 mutex_unlock(&kvm->lock);
d19a9cd2
ZX		 8687}
8688
ad8ba2cd
SY		 8689void kvm_arch_sync_events(struct kvm *kvm)
8690{
332967a3 8691 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7e44e449 8692 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
aea924f6 8693 kvm_free_pit(kvm);
ad8ba2cd
SY		 8694}
8695
1d8007bd 8696int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9da0e4d5
PB		 8697{
8698 int i, r;
25188b99 8699 unsigned long hva;
f0d648bd
PB		 8700 struct kvm_memslots *slots = kvm_memslots(kvm);
8701 struct kvm_memory_slot *slot, old;
9da0e4d5
PB		 8702
8703 /* Called with kvm->slots_lock held. */
1d8007bd
PB		 8704 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
8705 return -EINVAL;
9da0e4d5 8706
f0d648bd
PB		 8707 slot = id_to_memslot(slots, id);
8708 if (size) {
b21629da 8709 if (slot->npages)
f0d648bd
PB		 8710 return -EEXIST;
8711
8712 /*
8713 * MAP_SHARED to prevent internal slot pages from being moved
8714 * by fork()/COW.
8715 */
8716 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
8717 MAP_SHARED | MAP_ANONYMOUS, 0);
8718 if (IS_ERR((void *)hva))
8719 return PTR_ERR((void *)hva);
8720 } else {
8721 if (!slot->npages)
8722 return 0;
8723
8724 hva = 0;
8725 }
8726
8727 old = *slot;
9da0e4d5 8728 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1d8007bd 8729 struct kvm_userspace_memory_region m;
9da0e4d5 8730
1d8007bd
PB		 8731 m.slot = id | (i << 16);
8732 m.flags = 0;
8733 m.guest_phys_addr = gpa;
f0d648bd 8734 m.userspace_addr = hva;
1d8007bd 8735 m.memory_size = size;
9da0e4d5
PB		 8736 r = __kvm_set_memory_region(kvm, &m);
8737 if (r < 0)
8738 return r;
8739 }
8740
103c763c
EB		 8741 if (!size)
8742 vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
f0d648bd 8743
9da0e4d5
PB		 8744 return 0;
8745}
8746EXPORT_SYMBOL_GPL(__x86_set_memory_region);
8747
1d8007bd 8748int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9da0e4d5
PB		 8749{
8750 int r;
8751
8752 mutex_lock(&kvm->slots_lock);
1d8007bd 8753 r = __x86_set_memory_region(kvm, id, gpa, size);
9da0e4d5
PB		 8754 mutex_unlock(&kvm->slots_lock);
8755
8756 return r;
8757}
8758EXPORT_SYMBOL_GPL(x86_set_memory_region);
8759
d19a9cd2
ZX		 8760void kvm_arch_destroy_vm(struct kvm *kvm)
8761{
27469d29
AH		 8762 if (current->mm == kvm->mm) {
8763 /*
8764 * Free memory regions allocated on behalf of userspace,
8765 * unless the the memory map has changed due to process exit
8766 * or fd copying.
8767 */
1d8007bd
PB		 8768 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
8769 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
8770 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
27469d29 8771 }
03543133
SS		 8772 if (kvm_x86_ops->vm_destroy)
8773 kvm_x86_ops->vm_destroy(kvm);
c761159c
PX		 8774 kvm_pic_destroy(kvm);
8775 kvm_ioapic_destroy(kvm);
d19a9cd2 8776 kvm_free_vcpus(kvm);
af1bae54 8777 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
13d268ca 8778 kvm_mmu_uninit_vm(kvm);
2beb6dad 8779 kvm_page_track_cleanup(kvm);
cbc0236a 8780 kvm_hv_destroy_vm(kvm);
d19a9cd2 8781}
0de10343 8782
5587027c 8783void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
db3fe4eb
TY		 8784 struct kvm_memory_slot *dont)
8785{
8786 int i;
8787
d89cc617
TY		 8788 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8789 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
548ef284 8790 kvfree(free->arch.rmap[i]);
d89cc617 8791 free->arch.rmap[i] = NULL;
77d11309 8792 }
d89cc617
TY		 8793 if (i == 0)
8794 continue;
8795
8796 if (!dont || free->arch.lpage_info[i - 1] !=
8797 dont->arch.lpage_info[i - 1]) {
548ef284 8798 kvfree(free->arch.lpage_info[i - 1]);
d89cc617 8799 free->arch.lpage_info[i - 1] = NULL;
db3fe4eb
TY		 8800 }
8801 }
21ebbeda
XG		 8802
8803 kvm_page_track_free_memslot(free, dont);
db3fe4eb
TY		 8804}
8805
5587027c
AK		 8806int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
8807 unsigned long npages)
db3fe4eb
TY		 8808{
8809 int i;
8810
d89cc617 8811 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
92f94f1e 8812 struct kvm_lpage_info *linfo;
db3fe4eb
TY		 8813 unsigned long ugfn;
8814 int lpages;
d89cc617 8815 int level = i + 1;
db3fe4eb
TY		 8816
8817 lpages = gfn_to_index(slot->base_gfn + npages - 1,
8818 slot->base_gfn, level) + 1;
8819
d89cc617 8820 slot->arch.rmap[i] =
a7c3e901 8821 kvzalloc(lpages * sizeof(*slot->arch.rmap[i]), GFP_KERNEL);
d89cc617 8822 if (!slot->arch.rmap[i])
77d11309 8823 goto out_free;
d89cc617
TY		 8824 if (i == 0)
8825 continue;
77d11309 8826
a7c3e901 8827 linfo = kvzalloc(lpages * sizeof(*linfo), GFP_KERNEL);
92f94f1e 8828 if (!linfo)
db3fe4eb
TY		 8829 goto out_free;
8830
92f94f1e
XG		 8831 slot->arch.lpage_info[i - 1] = linfo;
8832
db3fe4eb 8833 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
92f94f1e 8834 linfo[0].disallow_lpage = 1;
db3fe4eb 8835 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
92f94f1e 8836 linfo[lpages - 1].disallow_lpage = 1;
db3fe4eb
TY		 8837 ugfn = slot->userspace_addr >> PAGE_SHIFT;
8838 /*
8839 * If the gfn and userspace address are not aligned wrt each
8840 * other, or if explicitly asked to, disable large page
8841 * support for this slot
8842 */
8843 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
8844 !kvm_largepages_enabled()) {
8845 unsigned long j;
8846
8847 for (j = 0; j < lpages; ++j)
92f94f1e 8848 linfo[j].disallow_lpage = 1;
db3fe4eb
TY		 8849 }
8850 }
8851
21ebbeda
XG		 8852 if (kvm_page_track_create_memslot(slot, npages))
8853 goto out_free;
8854
db3fe4eb
TY		 8855 return 0;
8856
8857out_free:
d89cc617 8858 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
548ef284 8859 kvfree(slot->arch.rmap[i]);
d89cc617
TY		 8860 slot->arch.rmap[i] = NULL;
8861 if (i == 0)
8862 continue;
8863
548ef284 8864 kvfree(slot->arch.lpage_info[i - 1]);
d89cc617 8865 slot->arch.lpage_info[i - 1] = NULL;
db3fe4eb
TY		 8866 }
8867 return -ENOMEM;
8868}
8869
15f46015 8870void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
e59dbe09 8871{
e6dff7d1
TY		 8872 /*
8873 * memslots->generation has been incremented.
8874 * mmio generation may have reached its maximum value.
8875 */
54bf36aa 8876 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
e59dbe09
TY		 8877}
8878
f7784b8e
MT		 8879int kvm_arch_prepare_memory_region(struct kvm *kvm,
8880 struct kvm_memory_slot *memslot,
09170a49 8881 const struct kvm_userspace_memory_region *mem,
7b6195a9 8882 enum kvm_mr_change change)
0de10343 8883{
f7784b8e
MT		 8884 return 0;
8885}
8886
88178fd4
KH		 8887static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8888 struct kvm_memory_slot *new)
8889{
8890 /* Still write protect RO slot */
8891 if (new->flags & KVM_MEM_READONLY) {
8892 kvm_mmu_slot_remove_write_access(kvm, new);
8893 return;
8894 }
8895
8896 /*
8897 * Call kvm_x86_ops dirty logging hooks when they are valid.
8898 *
8899 * kvm_x86_ops->slot_disable_log_dirty is called when:
8900 *
8901 * - KVM_MR_CREATE with dirty logging is disabled
8902 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8903 *
8904 * The reason is, in case of PML, we need to set D-bit for any slots
8905 * with dirty logging disabled in order to eliminate unnecessary GPA
8906 * logging in PML buffer (and potential PML buffer full VMEXT). This
8907 * guarantees leaving PML enabled during guest's lifetime won't have
8908 * any additonal overhead from PML when guest is running with dirty
8909 * logging disabled for memory slots.
8910 *
8911 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8912 * to dirty logging mode.
8913 *
8914 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8915 *
8916 * In case of write protect:
8917 *
8918 * Write protect all pages for dirty logging.
8919 *
8920 * All the sptes including the large sptes which point to this
8921 * slot are set to readonly. We can not create any new large
8922 * spte on this slot until the end of the logging.
8923 *
8924 * See the comments in fast_page_fault().
8925 */
8926 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8927 if (kvm_x86_ops->slot_enable_log_dirty)
8928 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8929 else
8930 kvm_mmu_slot_remove_write_access(kvm, new);
8931 } else {
8932 if (kvm_x86_ops->slot_disable_log_dirty)
8933 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8934 }
8935}
8936
f7784b8e 8937void kvm_arch_commit_memory_region(struct kvm *kvm,
09170a49 8938 const struct kvm_userspace_memory_region *mem,
8482644a 8939 const struct kvm_memory_slot *old,
f36f3f28 8940 const struct kvm_memory_slot *new,
8482644a 8941 enum kvm_mr_change change)
f7784b8e 8942{
8482644a 8943 int nr_mmu_pages = 0;
f7784b8e 8944
48c0e4e9
XG		 8945 if (!kvm->arch.n_requested_mmu_pages)
8946 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8947
48c0e4e9 8948 if (nr_mmu_pages)
0de10343 8949 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
1c91cad4 8950
3ea3b7fa
WL		 8951 /*
8952 * Dirty logging tracks sptes in 4k granularity, meaning that large
8953 * sptes have to be split. If live migration is successful, the guest
8954 * in the source machine will be destroyed and large sptes will be
8955 * created in the destination. However, if the guest continues to run
8956 * in the source machine (for example if live migration fails), small
8957 * sptes will remain around and cause bad performance.
8958 *
8959 * Scan sptes if dirty logging has been stopped, dropping those
8960 * which can be collapsed into a single large-page spte. Later
8961 * page faults will create the large-page sptes.
8962 */
8963 if ((change != KVM_MR_DELETE) &&
8964 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8965 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8966 kvm_mmu_zap_collapsible_sptes(kvm, new);
8967
c972f3b1 8968 /*
88178fd4 8969 * Set up write protection and/or dirty logging for the new slot.
c126d94f 8970 *
88178fd4
KH		 8971 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8972 * been zapped so no dirty logging staff is needed for old slot. For
8973 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8974 * new and it's also covered when dealing with the new slot.
f36f3f28
PB		 8975 *
8976 * FIXME: const-ify all uses of struct kvm_memory_slot.
c972f3b1 8977 */
88178fd4 8978 if (change != KVM_MR_DELETE)
f36f3f28 8979 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
0de10343 8980}
1d737c8a 8981
2df72e9b 8982void kvm_arch_flush_shadow_all(struct kvm *kvm)
34d4cb8f 8983{
6ca18b69 8984 kvm_mmu_invalidate_zap_all_pages(kvm);
34d4cb8f
MT		 8985}
8986
2df72e9b
MT		 8987void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8988 struct kvm_memory_slot *slot)
8989{
ae7cd873 8990 kvm_page_track_flush_slot(kvm, slot);
2df72e9b
MT		 8991}
8992
5d9bc648
PB		 8993static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8994{
8995 if (!list_empty_careful(&vcpu->async_pf.done))
8996 return true;
8997
8998 if (kvm_apic_has_events(vcpu))
8999 return true;
9000
9001 if (vcpu->arch.pv.pv_unhalted)
9002 return true;
9003
a5f01f8e
WL		 9004 if (vcpu->arch.exception.pending)
9005 return true;
9006
47a66eed
Z		 9007 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9008 (vcpu->arch.nmi_pending &&
9009 kvm_x86_ops->nmi_allowed(vcpu)))
5d9bc648
PB		 9010 return true;
9011
47a66eed
Z		 9012 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9013 (vcpu->arch.smi_pending && !is_smm(vcpu)))
73917739
PB		 9014 return true;
9015
5d9bc648
PB		 9016 if (kvm_arch_interrupt_allowed(vcpu) &&
9017 kvm_cpu_has_interrupt(vcpu))
9018 return true;
9019
1f4b34f8
AS		 9020 if (kvm_hv_has_stimer_pending(vcpu))
9021 return true;
9022
5d9bc648
PB		 9023 return false;
9024}
9025
1d737c8a
ZX		 9026int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9027{
5d9bc648 9028 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
1d737c8a 9029}
5736199a 9030
199b5763
LM		 9031bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9032{
de63ad4c 9033 return vcpu->arch.preempted_in_kernel;
199b5763
LM		 9034}
9035
b6d33834 9036int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
5736199a 9037{
b6d33834 9038 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
5736199a 9039}
78646121
GN		 9040
9041int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9042{
9043 return kvm_x86_ops->interrupt_allowed(vcpu);
9044}
229456fc 9045
82b32774 9046unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
f92653ee 9047{
82b32774
NA		 9048 if (is_64_bit_mode(vcpu))
9049 return kvm_rip_read(vcpu);
9050 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9051 kvm_rip_read(vcpu));
9052}
9053EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
f92653ee 9054
82b32774
NA		 9055bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9056{
9057 return kvm_get_linear_rip(vcpu) == linear_rip;
f92653ee
JK		 9058}
9059EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9060
94fe45da
JK		 9061unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9062{
9063 unsigned long rflags;
9064
9065 rflags = kvm_x86_ops->get_rflags(vcpu);
9066 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
c310bac5 9067 rflags &= ~X86_EFLAGS_TF;
94fe45da
JK		 9068 return rflags;
9069}
9070EXPORT_SYMBOL_GPL(kvm_get_rflags);
9071
6addfc42 9072static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
94fe45da
JK		 9073{
9074 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
f92653ee 9075 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
c310bac5 9076 rflags |= X86_EFLAGS_TF;
94fe45da 9077 kvm_x86_ops->set_rflags(vcpu, rflags);
6addfc42
PB		 9078}
9079
9080void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9081{
9082 __kvm_set_rflags(vcpu, rflags);
3842d135 9083 kvm_make_request(KVM_REQ_EVENT, vcpu);
94fe45da
JK		 9084}
9085EXPORT_SYMBOL_GPL(kvm_set_rflags);
9086
56028d08
GN		 9087void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9088{
9089 int r;
9090
fb67e14f 9091 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
f2e10669 9092 work->wakeup_all)
56028d08
GN		 9093 return;
9094
9095 r = kvm_mmu_reload(vcpu);
9096 if (unlikely(r))
9097 return;
9098
fb67e14f
XG		 9099 if (!vcpu->arch.mmu.direct_map &&
9100 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
9101 return;
9102
56028d08
GN		 9103 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
9104}
9105
af585b92
GN		 9106static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9107{
9108 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9109}
9110
9111static inline u32 kvm_async_pf_next_probe(u32 key)
9112{
9113 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9114}
9115
9116static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9117{
9118 u32 key = kvm_async_pf_hash_fn(gfn);
9119
9120 while (vcpu->arch.apf.gfns[key] != ~0)
9121 key = kvm_async_pf_next_probe(key);
9122
9123 vcpu->arch.apf.gfns[key] = gfn;
9124}
9125
9126static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9127{
9128 int i;
9129 u32 key = kvm_async_pf_hash_fn(gfn);
9130
9131 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
c7d28c24
XG		 9132 (vcpu->arch.apf.gfns[key] != gfn &&
9133 vcpu->arch.apf.gfns[key] != ~0); i++)
af585b92
GN		 9134 key = kvm_async_pf_next_probe(key);
9135
9136 return key;
9137}
9138
9139bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9140{
9141 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9142}
9143
9144static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9145{
9146 u32 i, j, k;
9147
9148 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9149 while (true) {
9150 vcpu->arch.apf.gfns[i] = ~0;
9151 do {
9152 j = kvm_async_pf_next_probe(j);
9153 if (vcpu->arch.apf.gfns[j] == ~0)
9154 return;
9155 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
9156 /*
9157 * k lies cyclically in ]i,j]
9158 * | i.k.j |
9159 * |....j i.k.| or |.k..j i...|
9160 */
9161 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
9162 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
9163 i = j;
9164 }
9165}
9166
7c90705b
GN		 9167static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
9168{
4e335d9e
PB		 9169
9170 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
9171 sizeof(val));
7c90705b
GN		 9172}
9173
9a6e7c39
WL		 9174static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
9175{
9176
9177 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
9178 sizeof(u32));
9179}
9180
af585b92
GN		 9181void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
9182 struct kvm_async_pf *work)
9183{
6389ee94
AK		 9184 struct x86_exception fault;
9185
7c90705b 9186 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
af585b92 9187 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7c90705b
GN		 9188
9189 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
fc5f06fa
GN		 9190 (vcpu->arch.apf.send_user_only &&
9191 kvm_x86_ops->get_cpl(vcpu) == 0))
7c90705b
GN		 9192 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
9193 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6389ee94
AK		 9194 fault.vector = PF_VECTOR;
9195 fault.error_code_valid = true;
9196 fault.error_code = 0;
9197 fault.nested_page_fault = false;
9198 fault.address = work->arch.token;
adfe20fb 9199 fault.async_page_fault = true;
6389ee94 9200 kvm_inject_page_fault(vcpu, &fault);
7c90705b 9201 }
af585b92
GN		 9202}
9203
9204void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
9205 struct kvm_async_pf *work)
9206{
6389ee94 9207 struct x86_exception fault;
9a6e7c39 9208 u32 val;
6389ee94 9209
f2e10669 9210 if (work->wakeup_all)
7c90705b
GN		 9211 work->arch.token = ~0; /* broadcast wakeup */
9212 else
9213 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
24dccf83 9214 trace_kvm_async_pf_ready(work->arch.token, work->gva);
7c90705b 9215
9a6e7c39
WL		 9216 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
9217 !apf_get_user(vcpu, &val)) {
9218 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
9219 vcpu->arch.exception.pending &&
9220 vcpu->arch.exception.nr == PF_VECTOR &&
9221 !apf_put_user(vcpu, 0)) {
9222 vcpu->arch.exception.injected = false;
9223 vcpu->arch.exception.pending = false;
9224 vcpu->arch.exception.nr = 0;
9225 vcpu->arch.exception.has_error_code = false;
9226 vcpu->arch.exception.error_code = 0;
9227 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
9228 fault.vector = PF_VECTOR;
9229 fault.error_code_valid = true;
9230 fault.error_code = 0;
9231 fault.nested_page_fault = false;
9232 fault.address = work->arch.token;
9233 fault.async_page_fault = true;
9234 kvm_inject_page_fault(vcpu, &fault);
9235 }
7c90705b 9236 }
e6d53e3b 9237 vcpu->arch.apf.halted = false;
a4fa1635 9238 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7c90705b
GN		 9239}
9240
9241bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
9242{
9243 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
9244 return true;
9245 else
9bc1f09f 9246 return kvm_can_do_async_pf(vcpu);
af585b92
GN		 9247}
9248
5544eb9b
PB		 9249void kvm_arch_start_assignment(struct kvm *kvm)
9250{
9251 atomic_inc(&kvm->arch.assigned_device_count);
9252}
9253EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
9254
9255void kvm_arch_end_assignment(struct kvm *kvm)
9256{
9257 atomic_dec(&kvm->arch.assigned_device_count);
9258}
9259EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
9260
9261bool kvm_arch_has_assigned_device(struct kvm *kvm)
9262{
9263 return atomic_read(&kvm->arch.assigned_device_count);
9264}
9265EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
9266
e0f0bbc5
AW		 9267void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
9268{
9269 atomic_inc(&kvm->arch.noncoherent_dma_count);
9270}
9271EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
9272
9273void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
9274{
9275 atomic_dec(&kvm->arch.noncoherent_dma_count);
9276}
9277EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
9278
9279bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
9280{
9281 return atomic_read(&kvm->arch.noncoherent_dma_count);
9282}
9283EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
9284
14717e20
AW		 9285bool kvm_arch_has_irq_bypass(void)
9286{
9287 return kvm_x86_ops->update_pi_irte != NULL;
9288}
9289
87276880
FW		 9290int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
9291 struct irq_bypass_producer *prod)
9292{
9293 struct kvm_kernel_irqfd *irqfd =
9294 container_of(cons, struct kvm_kernel_irqfd, consumer);
9295
14717e20 9296 irqfd->producer = prod;
87276880 9297
14717e20
AW		 9298 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
9299 prod->irq, irqfd->gsi, 1);
87276880
FW		 9300}
9301
9302void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
9303 struct irq_bypass_producer *prod)
9304{
9305 int ret;
9306 struct kvm_kernel_irqfd *irqfd =
9307 container_of(cons, struct kvm_kernel_irqfd, consumer);
9308
87276880
FW		 9309 WARN_ON(irqfd->producer != prod);
9310 irqfd->producer = NULL;
9311
9312 /*
9313 * When producer of consumer is unregistered, we change back to
9314 * remapped mode, so we can re-use the current implementation
bb3541f1 9315 * when the irq is masked/disabled or the consumer side (KVM
87276880
FW		 9316 * int this case doesn't want to receive the interrupts.
9317 */
9318 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
9319 if (ret)
9320 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
9321 " fails: %d\n", irqfd->consumer.token, ret);
9322}
9323
9324int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
9325 uint32_t guest_irq, bool set)
9326{
9327 if (!kvm_x86_ops->update_pi_irte)
9328 return -EINVAL;
9329
9330 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
9331}
9332
52004014
FW		 9333bool kvm_vector_hashing_enabled(void)
9334{
9335 return vector_hashing;
9336}
9337EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
9338
229456fc 9339EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
931c33b1 9340EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
229456fc
MT		 9341EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
9342EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
9343EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
9344EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
0ac406de 9345EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
d8cabddf 9346EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
17897f36 9347EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
236649de 9348EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
ec1ff790 9349EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
532a46b9 9350EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
2e554e8d 9351EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
489223ed 9352EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
7b46268d 9353EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
843e4330 9354EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
efc64404 9355EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
18f40c53
SS		 9356EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
9357EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
Linux 6.x block layer and io_uring tree(s)