projects / linux-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
KVM: x86: Don't attempt VMWare emulation on #GP with non-zero error code
[linux-block.git] / arch / x86 / kvm / x86.c
... / ...
CommitLineData
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Kernel-based Virtual Machine driver for Linux
4 *
5 * derived from drivers/kvm/kvm_main.c
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright (C) 2008 Qumranet, Inc.
9 * Copyright IBM Corporation, 2008
10 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 *
12 * Authors:
13 * Avi Kivity <avi@qumranet.com>
14 * Yaniv Kamay <yaniv@qumranet.com>
15 * Amit Shah <amit.shah@qumranet.com>
16 * Ben-Ami Yassour <benami@il.ibm.com>
17 */
18
19#include <linux/kvm_host.h>
20#include "irq.h"
21#include "mmu.h"
22#include "i8254.h"
23#include "tss.h"
24#include "kvm_cache_regs.h"
25#include "x86.h"
26#include "cpuid.h"
27#include "pmu.h"
28#include "hyperv.h"
29
30#include <linux/clocksource.h>
31#include <linux/interrupt.h>
32#include <linux/kvm.h>
33#include <linux/fs.h>
34#include <linux/vmalloc.h>
35#include <linux/export.h>
36#include <linux/moduleparam.h>
37#include <linux/mman.h>
38#include <linux/highmem.h>
39#include <linux/iommu.h>
40#include <linux/intel-iommu.h>
41#include <linux/cpufreq.h>
42#include <linux/user-return-notifier.h>
43#include <linux/srcu.h>
44#include <linux/slab.h>
45#include <linux/perf_event.h>
46#include <linux/uaccess.h>
47#include <linux/hash.h>
48#include <linux/pci.h>
49#include <linux/timekeeper_internal.h>
50#include <linux/pvclock_gtod.h>
51#include <linux/kvm_irqfd.h>
52#include <linux/irqbypass.h>
53#include <linux/sched/stat.h>
54#include <linux/sched/isolation.h>
55#include <linux/mem_encrypt.h>
56
57#include <trace/events/kvm.h>
58
59#include <asm/debugreg.h>
60#include <asm/msr.h>
61#include <asm/desc.h>
62#include <asm/mce.h>
63#include <linux/kernel_stat.h>
64#include <asm/fpu/internal.h> /* Ugh! */
65#include <asm/pvclock.h>
66#include <asm/div64.h>
67#include <asm/irq_remapping.h>
68#include <asm/mshyperv.h>
69#include <asm/hypervisor.h>
70#include <asm/intel_pt.h>
71#include <clocksource/hyperv_timer.h>
72
73#define CREATE_TRACE_POINTS
74#include "trace.h"
75
76#define MAX_IO_MSRS 256
77#define KVM_MAX_MCE_BANKS 32
78u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
79EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
80
81#define emul_to_vcpu(ctxt) \
82 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
83
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
89static
90u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
91#else
92static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
93#endif
94
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
97
98#define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
99 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
100
101static void update_cr8_intercept(struct kvm_vcpu *vcpu);
102static void process_nmi(struct kvm_vcpu *vcpu);
103static void enter_smm(struct kvm_vcpu *vcpu);
104static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
105static void store_regs(struct kvm_vcpu *vcpu);
106static int sync_regs(struct kvm_vcpu *vcpu);
107
108struct kvm_x86_ops *kvm_x86_ops __read_mostly;
109EXPORT_SYMBOL_GPL(kvm_x86_ops);
110
111static bool __read_mostly ignore_msrs = 0;
112module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
113
114static bool __read_mostly report_ignored_msrs = true;
115module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
116
117unsigned int min_timer_period_us = 200;
118module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
119
120static bool __read_mostly kvmclock_periodic_sync = true;
121module_param(kvmclock_periodic_sync, bool, S_IRUGO);
122
123bool __read_mostly kvm_has_tsc_control;
124EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
125u32 __read_mostly kvm_max_guest_tsc_khz;
126EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
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);
131u64 __read_mostly kvm_default_tsc_scaling_ratio;
132EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
133
134/* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
135static u32 __read_mostly tsc_tolerance_ppm = 250;
136module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
137
138/*
139 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
140 * adaptive tuning starting from default advancment of 1000ns. '0' disables
141 * advancement entirely. Any other value is used as-is and disables adaptive
142 * tuning, i.e. allows priveleged userspace to set an exact advancement time.
143 */
144static int __read_mostly lapic_timer_advance_ns = -1;
145module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
146
147static bool __read_mostly vector_hashing = true;
148module_param(vector_hashing, bool, S_IRUGO);
149
150bool __read_mostly enable_vmware_backdoor = false;
151module_param(enable_vmware_backdoor, bool, S_IRUGO);
152EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
153
154static bool __read_mostly force_emulation_prefix = false;
155module_param(force_emulation_prefix, bool, S_IRUGO);
156
157int __read_mostly pi_inject_timer = -1;
158module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
159
160#define KVM_NR_SHARED_MSRS 16
161
162struct kvm_shared_msrs_global {
163 int nr;
164 u32 msrs[KVM_NR_SHARED_MSRS];
165};
166
167struct kvm_shared_msrs {
168 struct user_return_notifier urn;
169 bool registered;
170 struct kvm_shared_msr_values {
171 u64 host;
172 u64 curr;
173 } values[KVM_NR_SHARED_MSRS];
174};
175
176static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
177static struct kvm_shared_msrs __percpu *shared_msrs;
178
179struct kvm_stats_debugfs_item debugfs_entries[] = {
180 { "pf_fixed", VCPU_STAT(pf_fixed) },
181 { "pf_guest", VCPU_STAT(pf_guest) },
182 { "tlb_flush", VCPU_STAT(tlb_flush) },
183 { "invlpg", VCPU_STAT(invlpg) },
184 { "exits", VCPU_STAT(exits) },
185 { "io_exits", VCPU_STAT(io_exits) },
186 { "mmio_exits", VCPU_STAT(mmio_exits) },
187 { "signal_exits", VCPU_STAT(signal_exits) },
188 { "irq_window", VCPU_STAT(irq_window_exits) },
189 { "nmi_window", VCPU_STAT(nmi_window_exits) },
190 { "halt_exits", VCPU_STAT(halt_exits) },
191 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
192 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
193 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
194 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
195 { "hypercalls", VCPU_STAT(hypercalls) },
196 { "request_irq", VCPU_STAT(request_irq_exits) },
197 { "irq_exits", VCPU_STAT(irq_exits) },
198 { "host_state_reload", VCPU_STAT(host_state_reload) },
199 { "fpu_reload", VCPU_STAT(fpu_reload) },
200 { "insn_emulation", VCPU_STAT(insn_emulation) },
201 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
202 { "irq_injections", VCPU_STAT(irq_injections) },
203 { "nmi_injections", VCPU_STAT(nmi_injections) },
204 { "req_event", VCPU_STAT(req_event) },
205 { "l1d_flush", VCPU_STAT(l1d_flush) },
206 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
207 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
208 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
209 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
210 { "mmu_flooded", VM_STAT(mmu_flooded) },
211 { "mmu_recycled", VM_STAT(mmu_recycled) },
212 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
213 { "mmu_unsync", VM_STAT(mmu_unsync) },
214 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
215 { "largepages", VM_STAT(lpages) },
216 { "max_mmu_page_hash_collisions",
217 VM_STAT(max_mmu_page_hash_collisions) },
218 { NULL }
219};
220
221u64 __read_mostly host_xcr0;
222
223struct kmem_cache *x86_fpu_cache;
224EXPORT_SYMBOL_GPL(x86_fpu_cache);
225
226static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
227
228static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
229{
230 int i;
231 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
232 vcpu->arch.apf.gfns[i] = ~0;
233}
234
235static void kvm_on_user_return(struct user_return_notifier *urn)
236{
237 unsigned slot;
238 struct kvm_shared_msrs *locals
239 = container_of(urn, struct kvm_shared_msrs, urn);
240 struct kvm_shared_msr_values *values;
241 unsigned long flags;
242
243 /*
244 * Disabling irqs at this point since the following code could be
245 * interrupted and executed through kvm_arch_hardware_disable()
246 */
247 local_irq_save(flags);
248 if (locals->registered) {
249 locals->registered = false;
250 user_return_notifier_unregister(urn);
251 }
252 local_irq_restore(flags);
253 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
254 values = &locals->values[slot];
255 if (values->host != values->curr) {
256 wrmsrl(shared_msrs_global.msrs[slot], values->host);
257 values->curr = values->host;
258 }
259 }
260}
261
262static void shared_msr_update(unsigned slot, u32 msr)
263{
264 u64 value;
265 unsigned int cpu = smp_processor_id();
266 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
267
268 /* only read, and nobody should modify it at this time,
269 * so don't need lock */
270 if (slot >= shared_msrs_global.nr) {
271 printk(KERN_ERR "kvm: invalid MSR slot!");
272 return;
273 }
274 rdmsrl_safe(msr, &value);
275 smsr->values[slot].host = value;
276 smsr->values[slot].curr = value;
277}
278
279void kvm_define_shared_msr(unsigned slot, u32 msr)
280{
281 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
282 shared_msrs_global.msrs[slot] = msr;
283 if (slot >= shared_msrs_global.nr)
284 shared_msrs_global.nr = slot + 1;
285}
286EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
287
288static void kvm_shared_msr_cpu_online(void)
289{
290 unsigned i;
291
292 for (i = 0; i < shared_msrs_global.nr; ++i)
293 shared_msr_update(i, shared_msrs_global.msrs[i]);
294}
295
296int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
297{
298 unsigned int cpu = smp_processor_id();
299 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
300 int err;
301
302 if (((value ^ smsr->values[slot].curr) & mask) == 0)
303 return 0;
304 smsr->values[slot].curr = value;
305 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
306 if (err)
307 return 1;
308
309 if (!smsr->registered) {
310 smsr->urn.on_user_return = kvm_on_user_return;
311 user_return_notifier_register(&smsr->urn);
312 smsr->registered = true;
313 }
314 return 0;
315}
316EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
317
318static void drop_user_return_notifiers(void)
319{
320 unsigned int cpu = smp_processor_id();
321 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
322
323 if (smsr->registered)
324 kvm_on_user_return(&smsr->urn);
325}
326
327u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
328{
329 return vcpu->arch.apic_base;
330}
331EXPORT_SYMBOL_GPL(kvm_get_apic_base);
332
333enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
334{
335 return kvm_apic_mode(kvm_get_apic_base(vcpu));
336}
337EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
338
339int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
340{
341 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
342 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
343 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
344 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
345
346 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
347 return 1;
348 if (!msr_info->host_initiated) {
349 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
350 return 1;
351 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
352 return 1;
353 }
354
355 kvm_lapic_set_base(vcpu, msr_info->data);
356 return 0;
357}
358EXPORT_SYMBOL_GPL(kvm_set_apic_base);
359
360asmlinkage __visible void kvm_spurious_fault(void)
361{
362 /* Fault while not rebooting. We want the trace. */
363 BUG();
364}
365EXPORT_SYMBOL_GPL(kvm_spurious_fault);
366
367#define EXCPT_BENIGN 0
368#define EXCPT_CONTRIBUTORY 1
369#define EXCPT_PF 2
370
371static int exception_class(int vector)
372{
373 switch (vector) {
374 case PF_VECTOR:
375 return EXCPT_PF;
376 case DE_VECTOR:
377 case TS_VECTOR:
378 case NP_VECTOR:
379 case SS_VECTOR:
380 case GP_VECTOR:
381 return EXCPT_CONTRIBUTORY;
382 default:
383 break;
384 }
385 return EXCPT_BENIGN;
386}
387
388#define EXCPT_FAULT 0
389#define EXCPT_TRAP 1
390#define EXCPT_ABORT 2
391#define EXCPT_INTERRUPT 3
392
393static int exception_type(int vector)
394{
395 unsigned int mask;
396
397 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
398 return EXCPT_INTERRUPT;
399
400 mask = 1 << vector;
401
402 /* #DB is trap, as instruction watchpoints are handled elsewhere */
403 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
404 return EXCPT_TRAP;
405
406 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
407 return EXCPT_ABORT;
408
409 /* Reserved exceptions will result in fault */
410 return EXCPT_FAULT;
411}
412
413void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
414{
415 unsigned nr = vcpu->arch.exception.nr;
416 bool has_payload = vcpu->arch.exception.has_payload;
417 unsigned long payload = vcpu->arch.exception.payload;
418
419 if (!has_payload)
420 return;
421
422 switch (nr) {
423 case DB_VECTOR:
424 /*
425 * "Certain debug exceptions may clear bit 0-3. The
426 * remaining contents of the DR6 register are never
427 * cleared by the processor".
428 */
429 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
430 /*
431 * DR6.RTM is set by all #DB exceptions that don't clear it.
432 */
433 vcpu->arch.dr6 |= DR6_RTM;
434 vcpu->arch.dr6 |= payload;
435 /*
436 * Bit 16 should be set in the payload whenever the #DB
437 * exception should clear DR6.RTM. This makes the payload
438 * compatible with the pending debug exceptions under VMX.
439 * Though not currently documented in the SDM, this also
440 * makes the payload compatible with the exit qualification
441 * for #DB exceptions under VMX.
442 */
443 vcpu->arch.dr6 ^= payload & DR6_RTM;
444 break;
445 case PF_VECTOR:
446 vcpu->arch.cr2 = payload;
447 break;
448 }
449
450 vcpu->arch.exception.has_payload = false;
451 vcpu->arch.exception.payload = 0;
452}
453EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
454
455static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
456 unsigned nr, bool has_error, u32 error_code,
457 bool has_payload, unsigned long payload, bool reinject)
458{
459 u32 prev_nr;
460 int class1, class2;
461
462 kvm_make_request(KVM_REQ_EVENT, vcpu);
463
464 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
465 queue:
466 if (has_error && !is_protmode(vcpu))
467 has_error = false;
468 if (reinject) {
469 /*
470 * On vmentry, vcpu->arch.exception.pending is only
471 * true if an event injection was blocked by
472 * nested_run_pending. In that case, however,
473 * vcpu_enter_guest requests an immediate exit,
474 * and the guest shouldn't proceed far enough to
475 * need reinjection.
476 */
477 WARN_ON_ONCE(vcpu->arch.exception.pending);
478 vcpu->arch.exception.injected = true;
479 if (WARN_ON_ONCE(has_payload)) {
480 /*
481 * A reinjected event has already
482 * delivered its payload.
483 */
484 has_payload = false;
485 payload = 0;
486 }
487 } else {
488 vcpu->arch.exception.pending = true;
489 vcpu->arch.exception.injected = false;
490 }
491 vcpu->arch.exception.has_error_code = has_error;
492 vcpu->arch.exception.nr = nr;
493 vcpu->arch.exception.error_code = error_code;
494 vcpu->arch.exception.has_payload = has_payload;
495 vcpu->arch.exception.payload = payload;
496 /*
497 * In guest mode, payload delivery should be deferred,
498 * so that the L1 hypervisor can intercept #PF before
499 * CR2 is modified (or intercept #DB before DR6 is
500 * modified under nVMX). However, for ABI
501 * compatibility with KVM_GET_VCPU_EVENTS and
502 * KVM_SET_VCPU_EVENTS, we can't delay payload
503 * delivery unless userspace has enabled this
504 * functionality via the per-VM capability,
505 * KVM_CAP_EXCEPTION_PAYLOAD.
506 */
507 if (!vcpu->kvm->arch.exception_payload_enabled ||
508 !is_guest_mode(vcpu))
509 kvm_deliver_exception_payload(vcpu);
510 return;
511 }
512
513 /* to check exception */
514 prev_nr = vcpu->arch.exception.nr;
515 if (prev_nr == DF_VECTOR) {
516 /* triple fault -> shutdown */
517 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
518 return;
519 }
520 class1 = exception_class(prev_nr);
521 class2 = exception_class(nr);
522 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
523 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
524 /*
525 * Generate double fault per SDM Table 5-5. Set
526 * exception.pending = true so that the double fault
527 * can trigger a nested vmexit.
528 */
529 vcpu->arch.exception.pending = true;
530 vcpu->arch.exception.injected = false;
531 vcpu->arch.exception.has_error_code = true;
532 vcpu->arch.exception.nr = DF_VECTOR;
533 vcpu->arch.exception.error_code = 0;
534 vcpu->arch.exception.has_payload = false;
535 vcpu->arch.exception.payload = 0;
536 } else
537 /* replace previous exception with a new one in a hope
538 that instruction re-execution will regenerate lost
539 exception */
540 goto queue;
541}
542
543void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
544{
545 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
546}
547EXPORT_SYMBOL_GPL(kvm_queue_exception);
548
549void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
550{
551 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
552}
553EXPORT_SYMBOL_GPL(kvm_requeue_exception);
554
555static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
556 unsigned long payload)
557{
558 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
559}
560
561static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
562 u32 error_code, unsigned long payload)
563{
564 kvm_multiple_exception(vcpu, nr, true, error_code,
565 true, payload, false);
566}
567
568int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
569{
570 if (err)
571 kvm_inject_gp(vcpu, 0);
572 else
573 return kvm_skip_emulated_instruction(vcpu);
574
575 return 1;
576}
577EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
578
579void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
580{
581 ++vcpu->stat.pf_guest;
582 vcpu->arch.exception.nested_apf =
583 is_guest_mode(vcpu) && fault->async_page_fault;
584 if (vcpu->arch.exception.nested_apf) {
585 vcpu->arch.apf.nested_apf_token = fault->address;
586 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
587 } else {
588 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
589 fault->address);
590 }
591}
592EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
593
594static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
595{
596 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
597 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
598 else
599 vcpu->arch.mmu->inject_page_fault(vcpu, fault);
600
601 return fault->nested_page_fault;
602}
603
604void kvm_inject_nmi(struct kvm_vcpu *vcpu)
605{
606 atomic_inc(&vcpu->arch.nmi_queued);
607 kvm_make_request(KVM_REQ_NMI, vcpu);
608}
609EXPORT_SYMBOL_GPL(kvm_inject_nmi);
610
611void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
612{
613 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
614}
615EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
616
617void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
618{
619 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
620}
621EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
622
623/*
624 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
625 * a #GP and return false.
626 */
627bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
628{
629 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
630 return true;
631 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
632 return false;
633}
634EXPORT_SYMBOL_GPL(kvm_require_cpl);
635
636bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
637{
638 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
639 return true;
640
641 kvm_queue_exception(vcpu, UD_VECTOR);
642 return false;
643}
644EXPORT_SYMBOL_GPL(kvm_require_dr);
645
646/*
647 * This function will be used to read from the physical memory of the currently
648 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
649 * can read from guest physical or from the guest's guest physical memory.
650 */
651int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
652 gfn_t ngfn, void *data, int offset, int len,
653 u32 access)
654{
655 struct x86_exception exception;
656 gfn_t real_gfn;
657 gpa_t ngpa;
658
659 ngpa = gfn_to_gpa(ngfn);
660 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
661 if (real_gfn == UNMAPPED_GVA)
662 return -EFAULT;
663
664 real_gfn = gpa_to_gfn(real_gfn);
665
666 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
667}
668EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
669
670static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
671 void *data, int offset, int len, u32 access)
672{
673 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
674 data, offset, len, access);
675}
676
677static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
678{
679 return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
680 rsvd_bits(1, 2);
681}
682
683/*
684 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
685 */
686int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
687{
688 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
689 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
690 int i;
691 int ret;
692 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
693
694 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
695 offset * sizeof(u64), sizeof(pdpte),
696 PFERR_USER_MASK|PFERR_WRITE_MASK);
697 if (ret < 0) {
698 ret = 0;
699 goto out;
700 }
701 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
702 if ((pdpte[i] & PT_PRESENT_MASK) &&
703 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
704 ret = 0;
705 goto out;
706 }
707 }
708 ret = 1;
709
710 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
711 __set_bit(VCPU_EXREG_PDPTR,
712 (unsigned long *)&vcpu->arch.regs_avail);
713 __set_bit(VCPU_EXREG_PDPTR,
714 (unsigned long *)&vcpu->arch.regs_dirty);
715out:
716
717 return ret;
718}
719EXPORT_SYMBOL_GPL(load_pdptrs);
720
721bool pdptrs_changed(struct kvm_vcpu *vcpu)
722{
723 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
724 bool changed = true;
725 int offset;
726 gfn_t gfn;
727 int r;
728
729 if (!is_pae_paging(vcpu))
730 return false;
731
732 if (!test_bit(VCPU_EXREG_PDPTR,
733 (unsigned long *)&vcpu->arch.regs_avail))
734 return true;
735
736 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
737 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
738 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
739 PFERR_USER_MASK | PFERR_WRITE_MASK);
740 if (r < 0)
741 goto out;
742 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
743out:
744
745 return changed;
746}
747EXPORT_SYMBOL_GPL(pdptrs_changed);
748
749int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
750{
751 unsigned long old_cr0 = kvm_read_cr0(vcpu);
752 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
753
754 cr0 |= X86_CR0_ET;
755
756#ifdef CONFIG_X86_64
757 if (cr0 & 0xffffffff00000000UL)
758 return 1;
759#endif
760
761 cr0 &= ~CR0_RESERVED_BITS;
762
763 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
764 return 1;
765
766 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
767 return 1;
768
769 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
770#ifdef CONFIG_X86_64
771 if ((vcpu->arch.efer & EFER_LME)) {
772 int cs_db, cs_l;
773
774 if (!is_pae(vcpu))
775 return 1;
776 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
777 if (cs_l)
778 return 1;
779 } else
780#endif
781 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
782 kvm_read_cr3(vcpu)))
783 return 1;
784 }
785
786 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
787 return 1;
788
789 kvm_x86_ops->set_cr0(vcpu, cr0);
790
791 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
792 kvm_clear_async_pf_completion_queue(vcpu);
793 kvm_async_pf_hash_reset(vcpu);
794 }
795
796 if ((cr0 ^ old_cr0) & update_bits)
797 kvm_mmu_reset_context(vcpu);
798
799 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
800 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
801 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
802 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
803
804 return 0;
805}
806EXPORT_SYMBOL_GPL(kvm_set_cr0);
807
808void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
809{
810 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
811}
812EXPORT_SYMBOL_GPL(kvm_lmsw);
813
814void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
815{
816 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
817 !vcpu->guest_xcr0_loaded) {
818 /* kvm_set_xcr() also depends on this */
819 if (vcpu->arch.xcr0 != host_xcr0)
820 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
821 vcpu->guest_xcr0_loaded = 1;
822 }
823}
824EXPORT_SYMBOL_GPL(kvm_load_guest_xcr0);
825
826void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
827{
828 if (vcpu->guest_xcr0_loaded) {
829 if (vcpu->arch.xcr0 != host_xcr0)
830 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
831 vcpu->guest_xcr0_loaded = 0;
832 }
833}
834EXPORT_SYMBOL_GPL(kvm_put_guest_xcr0);
835
836static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
837{
838 u64 xcr0 = xcr;
839 u64 old_xcr0 = vcpu->arch.xcr0;
840 u64 valid_bits;
841
842 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
843 if (index != XCR_XFEATURE_ENABLED_MASK)
844 return 1;
845 if (!(xcr0 & XFEATURE_MASK_FP))
846 return 1;
847 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
848 return 1;
849
850 /*
851 * Do not allow the guest to set bits that we do not support
852 * saving. However, xcr0 bit 0 is always set, even if the
853 * emulated CPU does not support XSAVE (see fx_init).
854 */
855 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
856 if (xcr0 & ~valid_bits)
857 return 1;
858
859 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
860 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
861 return 1;
862
863 if (xcr0 & XFEATURE_MASK_AVX512) {
864 if (!(xcr0 & XFEATURE_MASK_YMM))
865 return 1;
866 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
867 return 1;
868 }
869 vcpu->arch.xcr0 = xcr0;
870
871 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
872 kvm_update_cpuid(vcpu);
873 return 0;
874}
875
876int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
877{
878 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
879 __kvm_set_xcr(vcpu, index, xcr)) {
880 kvm_inject_gp(vcpu, 0);
881 return 1;
882 }
883 return 0;
884}
885EXPORT_SYMBOL_GPL(kvm_set_xcr);
886
887int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
888{
889 unsigned long old_cr4 = kvm_read_cr4(vcpu);
890 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
891 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
892
893 if (cr4 & CR4_RESERVED_BITS)
894 return 1;
895
896 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
897 return 1;
898
899 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
900 return 1;
901
902 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
903 return 1;
904
905 if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
906 return 1;
907
908 if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
909 return 1;
910
911 if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
912 return 1;
913
914 if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
915 return 1;
916
917 if (is_long_mode(vcpu)) {
918 if (!(cr4 & X86_CR4_PAE))
919 return 1;
920 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
921 && ((cr4 ^ old_cr4) & pdptr_bits)
922 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
923 kvm_read_cr3(vcpu)))
924 return 1;
925
926 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
927 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
928 return 1;
929
930 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
931 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
932 return 1;
933 }
934
935 if (kvm_x86_ops->set_cr4(vcpu, cr4))
936 return 1;
937
938 if (((cr4 ^ old_cr4) & pdptr_bits) ||
939 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
940 kvm_mmu_reset_context(vcpu);
941
942 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
943 kvm_update_cpuid(vcpu);
944
945 return 0;
946}
947EXPORT_SYMBOL_GPL(kvm_set_cr4);
948
949int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
950{
951 bool skip_tlb_flush = false;
952#ifdef CONFIG_X86_64
953 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
954
955 if (pcid_enabled) {
956 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
957 cr3 &= ~X86_CR3_PCID_NOFLUSH;
958 }
959#endif
960
961 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
962 if (!skip_tlb_flush) {
963 kvm_mmu_sync_roots(vcpu);
964 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
965 }
966 return 0;
967 }
968
969 if (is_long_mode(vcpu) &&
970 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
971 return 1;
972 else if (is_pae_paging(vcpu) &&
973 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
974 return 1;
975
976 kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
977 vcpu->arch.cr3 = cr3;
978 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
979
980 return 0;
981}
982EXPORT_SYMBOL_GPL(kvm_set_cr3);
983
984int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
985{
986 if (cr8 & CR8_RESERVED_BITS)
987 return 1;
988 if (lapic_in_kernel(vcpu))
989 kvm_lapic_set_tpr(vcpu, cr8);
990 else
991 vcpu->arch.cr8 = cr8;
992 return 0;
993}
994EXPORT_SYMBOL_GPL(kvm_set_cr8);
995
996unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
997{
998 if (lapic_in_kernel(vcpu))
999 return kvm_lapic_get_cr8(vcpu);
1000 else
1001 return vcpu->arch.cr8;
1002}
1003EXPORT_SYMBOL_GPL(kvm_get_cr8);
1004
1005static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1006{
1007 int i;
1008
1009 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1010 for (i = 0; i < KVM_NR_DB_REGS; i++)
1011 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1012 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1013 }
1014}
1015
1016static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1017{
1018 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1019 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
1020}
1021
1022static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1023{
1024 unsigned long dr7;
1025
1026 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1027 dr7 = vcpu->arch.guest_debug_dr7;
1028 else
1029 dr7 = vcpu->arch.dr7;
1030 kvm_x86_ops->set_dr7(vcpu, dr7);
1031 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1032 if (dr7 & DR7_BP_EN_MASK)
1033 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1034}
1035
1036static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1037{
1038 u64 fixed = DR6_FIXED_1;
1039
1040 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1041 fixed |= DR6_RTM;
1042 return fixed;
1043}
1044
1045static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1046{
1047 switch (dr) {
1048 case 0 ... 3:
1049 vcpu->arch.db[dr] = val;
1050 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1051 vcpu->arch.eff_db[dr] = val;
1052 break;
1053 case 4:
1054 /* fall through */
1055 case 6:
1056 if (val & 0xffffffff00000000ULL)
1057 return -1; /* #GP */
1058 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1059 kvm_update_dr6(vcpu);
1060 break;
1061 case 5:
1062 /* fall through */
1063 default: /* 7 */
1064 if (val & 0xffffffff00000000ULL)
1065 return -1; /* #GP */
1066 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1067 kvm_update_dr7(vcpu);
1068 break;
1069 }
1070
1071 return 0;
1072}
1073
1074int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1075{
1076 if (__kvm_set_dr(vcpu, dr, val)) {
1077 kvm_inject_gp(vcpu, 0);
1078 return 1;
1079 }
1080 return 0;
1081}
1082EXPORT_SYMBOL_GPL(kvm_set_dr);
1083
1084int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1085{
1086 switch (dr) {
1087 case 0 ... 3:
1088 *val = vcpu->arch.db[dr];
1089 break;
1090 case 4:
1091 /* fall through */
1092 case 6:
1093 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1094 *val = vcpu->arch.dr6;
1095 else
1096 *val = kvm_x86_ops->get_dr6(vcpu);
1097 break;
1098 case 5:
1099 /* fall through */
1100 default: /* 7 */
1101 *val = vcpu->arch.dr7;
1102 break;
1103 }
1104 return 0;
1105}
1106EXPORT_SYMBOL_GPL(kvm_get_dr);
1107
1108bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1109{
1110 u32 ecx = kvm_rcx_read(vcpu);
1111 u64 data;
1112 int err;
1113
1114 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1115 if (err)
1116 return err;
1117 kvm_rax_write(vcpu, (u32)data);
1118 kvm_rdx_write(vcpu, data >> 32);
1119 return err;
1120}
1121EXPORT_SYMBOL_GPL(kvm_rdpmc);
1122
1123/*
1124 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1125 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1126 *
1127 * This list is modified at module load time to reflect the
1128 * capabilities of the host cpu. This capabilities test skips MSRs that are
1129 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1130 * may depend on host virtualization features rather than host cpu features.
1131 */
1132
1133static u32 msrs_to_save[] = {
1134 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1135 MSR_STAR,
1136#ifdef CONFIG_X86_64
1137 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1138#endif
1139 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1140 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1141 MSR_IA32_SPEC_CTRL,
1142 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1143 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1144 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1145 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1146 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1147 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1148 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1149 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1150 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1151 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1152 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1153 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1154 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1155 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1156 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1157 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1158 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1159 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1160 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1161 MSR_ARCH_PERFMON_PERFCTR0 + 18, MSR_ARCH_PERFMON_PERFCTR0 + 19,
1162 MSR_ARCH_PERFMON_PERFCTR0 + 20, MSR_ARCH_PERFMON_PERFCTR0 + 21,
1163 MSR_ARCH_PERFMON_PERFCTR0 + 22, MSR_ARCH_PERFMON_PERFCTR0 + 23,
1164 MSR_ARCH_PERFMON_PERFCTR0 + 24, MSR_ARCH_PERFMON_PERFCTR0 + 25,
1165 MSR_ARCH_PERFMON_PERFCTR0 + 26, MSR_ARCH_PERFMON_PERFCTR0 + 27,
1166 MSR_ARCH_PERFMON_PERFCTR0 + 28, MSR_ARCH_PERFMON_PERFCTR0 + 29,
1167 MSR_ARCH_PERFMON_PERFCTR0 + 30, MSR_ARCH_PERFMON_PERFCTR0 + 31,
1168 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1169 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1170 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1171 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1172 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1173 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1174 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1175 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1176 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1177 MSR_ARCH_PERFMON_EVENTSEL0 + 18, MSR_ARCH_PERFMON_EVENTSEL0 + 19,
1178 MSR_ARCH_PERFMON_EVENTSEL0 + 20, MSR_ARCH_PERFMON_EVENTSEL0 + 21,
1179 MSR_ARCH_PERFMON_EVENTSEL0 + 22, MSR_ARCH_PERFMON_EVENTSEL0 + 23,
1180 MSR_ARCH_PERFMON_EVENTSEL0 + 24, MSR_ARCH_PERFMON_EVENTSEL0 + 25,
1181 MSR_ARCH_PERFMON_EVENTSEL0 + 26, MSR_ARCH_PERFMON_EVENTSEL0 + 27,
1182 MSR_ARCH_PERFMON_EVENTSEL0 + 28, MSR_ARCH_PERFMON_EVENTSEL0 + 29,
1183 MSR_ARCH_PERFMON_EVENTSEL0 + 30, MSR_ARCH_PERFMON_EVENTSEL0 + 31,
1184};
1185
1186static unsigned num_msrs_to_save;
1187
1188static u32 emulated_msrs[] = {
1189 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1190 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1191 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1192 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1193 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1194 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1195 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1196 HV_X64_MSR_RESET,
1197 HV_X64_MSR_VP_INDEX,
1198 HV_X64_MSR_VP_RUNTIME,
1199 HV_X64_MSR_SCONTROL,
1200 HV_X64_MSR_STIMER0_CONFIG,
1201 HV_X64_MSR_VP_ASSIST_PAGE,
1202 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1203 HV_X64_MSR_TSC_EMULATION_STATUS,
1204
1205 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1206 MSR_KVM_PV_EOI_EN,
1207
1208 MSR_IA32_TSC_ADJUST,
1209 MSR_IA32_TSCDEADLINE,
1210 MSR_IA32_ARCH_CAPABILITIES,
1211 MSR_IA32_MISC_ENABLE,
1212 MSR_IA32_MCG_STATUS,
1213 MSR_IA32_MCG_CTL,
1214 MSR_IA32_MCG_EXT_CTL,
1215 MSR_IA32_SMBASE,
1216 MSR_SMI_COUNT,
1217 MSR_PLATFORM_INFO,
1218 MSR_MISC_FEATURES_ENABLES,
1219 MSR_AMD64_VIRT_SPEC_CTRL,
1220 MSR_IA32_POWER_CTL,
1221
1222 /*
1223 * The following list leaves out MSRs whose values are determined
1224 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1225 * We always support the "true" VMX control MSRs, even if the host
1226 * processor does not, so I am putting these registers here rather
1227 * than in msrs_to_save.
1228 */
1229 MSR_IA32_VMX_BASIC,
1230 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1231 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1232 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1233 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1234 MSR_IA32_VMX_MISC,
1235 MSR_IA32_VMX_CR0_FIXED0,
1236 MSR_IA32_VMX_CR4_FIXED0,
1237 MSR_IA32_VMX_VMCS_ENUM,
1238 MSR_IA32_VMX_PROCBASED_CTLS2,
1239 MSR_IA32_VMX_EPT_VPID_CAP,
1240 MSR_IA32_VMX_VMFUNC,
1241
1242 MSR_K7_HWCR,
1243 MSR_KVM_POLL_CONTROL,
1244};
1245
1246static unsigned num_emulated_msrs;
1247
1248/*
1249 * List of msr numbers which are used to expose MSR-based features that
1250 * can be used by a hypervisor to validate requested CPU features.
1251 */
1252static u32 msr_based_features[] = {
1253 MSR_IA32_VMX_BASIC,
1254 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1255 MSR_IA32_VMX_PINBASED_CTLS,
1256 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1257 MSR_IA32_VMX_PROCBASED_CTLS,
1258 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1259 MSR_IA32_VMX_EXIT_CTLS,
1260 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1261 MSR_IA32_VMX_ENTRY_CTLS,
1262 MSR_IA32_VMX_MISC,
1263 MSR_IA32_VMX_CR0_FIXED0,
1264 MSR_IA32_VMX_CR0_FIXED1,
1265 MSR_IA32_VMX_CR4_FIXED0,
1266 MSR_IA32_VMX_CR4_FIXED1,
1267 MSR_IA32_VMX_VMCS_ENUM,
1268 MSR_IA32_VMX_PROCBASED_CTLS2,
1269 MSR_IA32_VMX_EPT_VPID_CAP,
1270 MSR_IA32_VMX_VMFUNC,
1271
1272 MSR_F10H_DECFG,
1273 MSR_IA32_UCODE_REV,
1274 MSR_IA32_ARCH_CAPABILITIES,
1275};
1276
1277static unsigned int num_msr_based_features;
1278
1279static u64 kvm_get_arch_capabilities(void)
1280{
1281 u64 data = 0;
1282
1283 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1284 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1285
1286 /*
1287 * If we're doing cache flushes (either "always" or "cond")
1288 * we will do one whenever the guest does a vmlaunch/vmresume.
1289 * If an outer hypervisor is doing the cache flush for us
1290 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1291 * capability to the guest too, and if EPT is disabled we're not
1292 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1293 * require a nested hypervisor to do a flush of its own.
1294 */
1295 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1296 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1297
1298 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1299 data |= ARCH_CAP_RDCL_NO;
1300 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1301 data |= ARCH_CAP_SSB_NO;
1302 if (!boot_cpu_has_bug(X86_BUG_MDS))
1303 data |= ARCH_CAP_MDS_NO;
1304
1305 return data;
1306}
1307
1308static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1309{
1310 switch (msr->index) {
1311 case MSR_IA32_ARCH_CAPABILITIES:
1312 msr->data = kvm_get_arch_capabilities();
1313 break;
1314 case MSR_IA32_UCODE_REV:
1315 rdmsrl_safe(msr->index, &msr->data);
1316 break;
1317 default:
1318 if (kvm_x86_ops->get_msr_feature(msr))
1319 return 1;
1320 }
1321 return 0;
1322}
1323
1324static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1325{
1326 struct kvm_msr_entry msr;
1327 int r;
1328
1329 msr.index = index;
1330 r = kvm_get_msr_feature(&msr);
1331 if (r)
1332 return r;
1333
1334 *data = msr.data;
1335
1336 return 0;
1337}
1338
1339static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1340{
1341 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1342 return false;
1343
1344 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1345 return false;
1346
1347 if (efer & (EFER_LME | EFER_LMA) &&
1348 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1349 return false;
1350
1351 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1352 return false;
1353
1354 return true;
1355
1356}
1357bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1358{
1359 if (efer & efer_reserved_bits)
1360 return false;
1361
1362 return __kvm_valid_efer(vcpu, efer);
1363}
1364EXPORT_SYMBOL_GPL(kvm_valid_efer);
1365
1366static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1367{
1368 u64 old_efer = vcpu->arch.efer;
1369 u64 efer = msr_info->data;
1370
1371 if (efer & efer_reserved_bits)
1372 return 1;
1373
1374 if (!msr_info->host_initiated) {
1375 if (!__kvm_valid_efer(vcpu, efer))
1376 return 1;
1377
1378 if (is_paging(vcpu) &&
1379 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1380 return 1;
1381 }
1382
1383 efer &= ~EFER_LMA;
1384 efer |= vcpu->arch.efer & EFER_LMA;
1385
1386 kvm_x86_ops->set_efer(vcpu, efer);
1387
1388 /* Update reserved bits */
1389 if ((efer ^ old_efer) & EFER_NX)
1390 kvm_mmu_reset_context(vcpu);
1391
1392 return 0;
1393}
1394
1395void kvm_enable_efer_bits(u64 mask)
1396{
1397 efer_reserved_bits &= ~mask;
1398}
1399EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1400
1401/*
1402 * Write @data into the MSR specified by @index. Select MSR specific fault
1403 * checks are bypassed if @host_initiated is %true.
1404 * Returns 0 on success, non-0 otherwise.
1405 * Assumes vcpu_load() was already called.
1406 */
1407static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1408 bool host_initiated)
1409{
1410 struct msr_data msr;
1411
1412 switch (index) {
1413 case MSR_FS_BASE:
1414 case MSR_GS_BASE:
1415 case MSR_KERNEL_GS_BASE:
1416 case MSR_CSTAR:
1417 case MSR_LSTAR:
1418 if (is_noncanonical_address(data, vcpu))
1419 return 1;
1420 break;
1421 case MSR_IA32_SYSENTER_EIP:
1422 case MSR_IA32_SYSENTER_ESP:
1423 /*
1424 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1425 * non-canonical address is written on Intel but not on
1426 * AMD (which ignores the top 32-bits, because it does
1427 * not implement 64-bit SYSENTER).
1428 *
1429 * 64-bit code should hence be able to write a non-canonical
1430 * value on AMD. Making the address canonical ensures that
1431 * vmentry does not fail on Intel after writing a non-canonical
1432 * value, and that something deterministic happens if the guest
1433 * invokes 64-bit SYSENTER.
1434 */
1435 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1436 }
1437
1438 msr.data = data;
1439 msr.index = index;
1440 msr.host_initiated = host_initiated;
1441
1442 return kvm_x86_ops->set_msr(vcpu, &msr);
1443}
1444
1445/*
1446 * Read the MSR specified by @index into @data. Select MSR specific fault
1447 * checks are bypassed if @host_initiated is %true.
1448 * Returns 0 on success, non-0 otherwise.
1449 * Assumes vcpu_load() was already called.
1450 */
1451static int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1452 bool host_initiated)
1453{
1454 struct msr_data msr;
1455 int ret;
1456
1457 msr.index = index;
1458 msr.host_initiated = host_initiated;
1459
1460 ret = kvm_x86_ops->get_msr(vcpu, &msr);
1461 if (!ret)
1462 *data = msr.data;
1463 return ret;
1464}
1465
1466int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1467{
1468 return __kvm_get_msr(vcpu, index, data, false);
1469}
1470EXPORT_SYMBOL_GPL(kvm_get_msr);
1471
1472int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1473{
1474 return __kvm_set_msr(vcpu, index, data, false);
1475}
1476EXPORT_SYMBOL_GPL(kvm_set_msr);
1477
1478int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1479{
1480 u32 ecx = kvm_rcx_read(vcpu);
1481 u64 data;
1482
1483 if (kvm_get_msr(vcpu, ecx, &data)) {
1484 trace_kvm_msr_read_ex(ecx);
1485 kvm_inject_gp(vcpu, 0);
1486 return 1;
1487 }
1488
1489 trace_kvm_msr_read(ecx, data);
1490
1491 kvm_rax_write(vcpu, data & -1u);
1492 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1493 return kvm_skip_emulated_instruction(vcpu);
1494}
1495EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1496
1497int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1498{
1499 u32 ecx = kvm_rcx_read(vcpu);
1500 u64 data = kvm_read_edx_eax(vcpu);
1501
1502 if (kvm_set_msr(vcpu, ecx, data)) {
1503 trace_kvm_msr_write_ex(ecx, data);
1504 kvm_inject_gp(vcpu, 0);
1505 return 1;
1506 }
1507
1508 trace_kvm_msr_write(ecx, data);
1509 return kvm_skip_emulated_instruction(vcpu);
1510}
1511EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1512
1513/*
1514 * Adapt set_msr() to msr_io()'s calling convention
1515 */
1516static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1517{
1518 return __kvm_get_msr(vcpu, index, data, true);
1519}
1520
1521static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1522{
1523 return __kvm_set_msr(vcpu, index, *data, true);
1524}
1525
1526#ifdef CONFIG_X86_64
1527struct pvclock_gtod_data {
1528 seqcount_t seq;
1529
1530 struct { /* extract of a clocksource struct */
1531 int vclock_mode;
1532 u64 cycle_last;
1533 u64 mask;
1534 u32 mult;
1535 u32 shift;
1536 } clock;
1537
1538 u64 boot_ns;
1539 u64 nsec_base;
1540 u64 wall_time_sec;
1541};
1542
1543static struct pvclock_gtod_data pvclock_gtod_data;
1544
1545static void update_pvclock_gtod(struct timekeeper *tk)
1546{
1547 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1548 u64 boot_ns;
1549
1550 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1551
1552 write_seqcount_begin(&vdata->seq);
1553
1554 /* copy pvclock gtod data */
1555 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1556 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1557 vdata->clock.mask = tk->tkr_mono.mask;
1558 vdata->clock.mult = tk->tkr_mono.mult;
1559 vdata->clock.shift = tk->tkr_mono.shift;
1560
1561 vdata->boot_ns = boot_ns;
1562 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1563
1564 vdata->wall_time_sec = tk->xtime_sec;
1565
1566 write_seqcount_end(&vdata->seq);
1567}
1568#endif
1569
1570void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1571{
1572 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1573 kvm_vcpu_kick(vcpu);
1574}
1575
1576static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1577{
1578 int version;
1579 int r;
1580 struct pvclock_wall_clock wc;
1581 struct timespec64 boot;
1582
1583 if (!wall_clock)
1584 return;
1585
1586 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1587 if (r)
1588 return;
1589
1590 if (version & 1)
1591 ++version; /* first time write, random junk */
1592
1593 ++version;
1594
1595 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1596 return;
1597
1598 /*
1599 * The guest calculates current wall clock time by adding
1600 * system time (updated by kvm_guest_time_update below) to the
1601 * wall clock specified here. guest system time equals host
1602 * system time for us, thus we must fill in host boot time here.
1603 */
1604 getboottime64(&boot);
1605
1606 if (kvm->arch.kvmclock_offset) {
1607 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1608 boot = timespec64_sub(boot, ts);
1609 }
1610 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1611 wc.nsec = boot.tv_nsec;
1612 wc.version = version;
1613
1614 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1615
1616 version++;
1617 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1618}
1619
1620static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1621{
1622 do_shl32_div32(dividend, divisor);
1623 return dividend;
1624}
1625
1626static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1627 s8 *pshift, u32 *pmultiplier)
1628{
1629 uint64_t scaled64;
1630 int32_t shift = 0;
1631 uint64_t tps64;
1632 uint32_t tps32;
1633
1634 tps64 = base_hz;
1635 scaled64 = scaled_hz;
1636 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1637 tps64 >>= 1;
1638 shift--;
1639 }
1640
1641 tps32 = (uint32_t)tps64;
1642 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1643 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1644 scaled64 >>= 1;
1645 else
1646 tps32 <<= 1;
1647 shift++;
1648 }
1649
1650 *pshift = shift;
1651 *pmultiplier = div_frac(scaled64, tps32);
1652}
1653
1654#ifdef CONFIG_X86_64
1655static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1656#endif
1657
1658static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1659static unsigned long max_tsc_khz;
1660
1661static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1662{
1663 u64 v = (u64)khz * (1000000 + ppm);
1664 do_div(v, 1000000);
1665 return v;
1666}
1667
1668static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1669{
1670 u64 ratio;
1671
1672 /* Guest TSC same frequency as host TSC? */
1673 if (!scale) {
1674 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1675 return 0;
1676 }
1677
1678 /* TSC scaling supported? */
1679 if (!kvm_has_tsc_control) {
1680 if (user_tsc_khz > tsc_khz) {
1681 vcpu->arch.tsc_catchup = 1;
1682 vcpu->arch.tsc_always_catchup = 1;
1683 return 0;
1684 } else {
1685 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
1686 return -1;
1687 }
1688 }
1689
1690 /* TSC scaling required - calculate ratio */
1691 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1692 user_tsc_khz, tsc_khz);
1693
1694 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1695 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1696 user_tsc_khz);
1697 return -1;
1698 }
1699
1700 vcpu->arch.tsc_scaling_ratio = ratio;
1701 return 0;
1702}
1703
1704static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1705{
1706 u32 thresh_lo, thresh_hi;
1707 int use_scaling = 0;
1708
1709 /* tsc_khz can be zero if TSC calibration fails */
1710 if (user_tsc_khz == 0) {
1711 /* set tsc_scaling_ratio to a safe value */
1712 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1713 return -1;
1714 }
1715
1716 /* Compute a scale to convert nanoseconds in TSC cycles */
1717 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1718 &vcpu->arch.virtual_tsc_shift,
1719 &vcpu->arch.virtual_tsc_mult);
1720 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1721
1722 /*
1723 * Compute the variation in TSC rate which is acceptable
1724 * within the range of tolerance and decide if the
1725 * rate being applied is within that bounds of the hardware
1726 * rate. If so, no scaling or compensation need be done.
1727 */
1728 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1729 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1730 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1731 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1732 use_scaling = 1;
1733 }
1734 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1735}
1736
1737static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1738{
1739 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1740 vcpu->arch.virtual_tsc_mult,
1741 vcpu->arch.virtual_tsc_shift);
1742 tsc += vcpu->arch.this_tsc_write;
1743 return tsc;
1744}
1745
1746static inline int gtod_is_based_on_tsc(int mode)
1747{
1748 return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1749}
1750
1751static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1752{
1753#ifdef CONFIG_X86_64
1754 bool vcpus_matched;
1755 struct kvm_arch *ka = &vcpu->kvm->arch;
1756 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1757
1758 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1759 atomic_read(&vcpu->kvm->online_vcpus));
1760
1761 /*
1762 * Once the masterclock is enabled, always perform request in
1763 * order to update it.
1764 *
1765 * In order to enable masterclock, the host clocksource must be TSC
1766 * and the vcpus need to have matched TSCs. When that happens,
1767 * perform request to enable masterclock.
1768 */
1769 if (ka->use_master_clock ||
1770 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1771 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1772
1773 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1774 atomic_read(&vcpu->kvm->online_vcpus),
1775 ka->use_master_clock, gtod->clock.vclock_mode);
1776#endif
1777}
1778
1779static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1780{
1781 u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1782 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1783}
1784
1785/*
1786 * Multiply tsc by a fixed point number represented by ratio.
1787 *
1788 * The most significant 64-N bits (mult) of ratio represent the
1789 * integral part of the fixed point number; the remaining N bits
1790 * (frac) represent the fractional part, ie. ratio represents a fixed
1791 * point number (mult + frac * 2^(-N)).
1792 *
1793 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1794 */
1795static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1796{
1797 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1798}
1799
1800u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1801{
1802 u64 _tsc = tsc;
1803 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1804
1805 if (ratio != kvm_default_tsc_scaling_ratio)
1806 _tsc = __scale_tsc(ratio, tsc);
1807
1808 return _tsc;
1809}
1810EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1811
1812static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1813{
1814 u64 tsc;
1815
1816 tsc = kvm_scale_tsc(vcpu, rdtsc());
1817
1818 return target_tsc - tsc;
1819}
1820
1821u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1822{
1823 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1824
1825 return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1826}
1827EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1828
1829static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1830{
1831 vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1832}
1833
1834static inline bool kvm_check_tsc_unstable(void)
1835{
1836#ifdef CONFIG_X86_64
1837 /*
1838 * TSC is marked unstable when we're running on Hyper-V,
1839 * 'TSC page' clocksource is good.
1840 */
1841 if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1842 return false;
1843#endif
1844 return check_tsc_unstable();
1845}
1846
1847void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1848{
1849 struct kvm *kvm = vcpu->kvm;
1850 u64 offset, ns, elapsed;
1851 unsigned long flags;
1852 bool matched;
1853 bool already_matched;
1854 u64 data = msr->data;
1855 bool synchronizing = false;
1856
1857 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1858 offset = kvm_compute_tsc_offset(vcpu, data);
1859 ns = ktime_get_boottime_ns();
1860 elapsed = ns - kvm->arch.last_tsc_nsec;
1861
1862 if (vcpu->arch.virtual_tsc_khz) {
1863 if (data == 0 && msr->host_initiated) {
1864 /*
1865 * detection of vcpu initialization -- need to sync
1866 * with other vCPUs. This particularly helps to keep
1867 * kvm_clock stable after CPU hotplug
1868 */
1869 synchronizing = true;
1870 } else {
1871 u64 tsc_exp = kvm->arch.last_tsc_write +
1872 nsec_to_cycles(vcpu, elapsed);
1873 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1874 /*
1875 * Special case: TSC write with a small delta (1 second)
1876 * of virtual cycle time against real time is
1877 * interpreted as an attempt to synchronize the CPU.
1878 */
1879 synchronizing = data < tsc_exp + tsc_hz &&
1880 data + tsc_hz > tsc_exp;
1881 }
1882 }
1883
1884 /*
1885 * For a reliable TSC, we can match TSC offsets, and for an unstable
1886 * TSC, we add elapsed time in this computation. We could let the
1887 * compensation code attempt to catch up if we fall behind, but
1888 * it's better to try to match offsets from the beginning.
1889 */
1890 if (synchronizing &&
1891 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1892 if (!kvm_check_tsc_unstable()) {
1893 offset = kvm->arch.cur_tsc_offset;
1894 } else {
1895 u64 delta = nsec_to_cycles(vcpu, elapsed);
1896 data += delta;
1897 offset = kvm_compute_tsc_offset(vcpu, data);
1898 }
1899 matched = true;
1900 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1901 } else {
1902 /*
1903 * We split periods of matched TSC writes into generations.
1904 * For each generation, we track the original measured
1905 * nanosecond time, offset, and write, so if TSCs are in
1906 * sync, we can match exact offset, and if not, we can match
1907 * exact software computation in compute_guest_tsc()
1908 *
1909 * These values are tracked in kvm->arch.cur_xxx variables.
1910 */
1911 kvm->arch.cur_tsc_generation++;
1912 kvm->arch.cur_tsc_nsec = ns;
1913 kvm->arch.cur_tsc_write = data;
1914 kvm->arch.cur_tsc_offset = offset;
1915 matched = false;
1916 }
1917
1918 /*
1919 * We also track th most recent recorded KHZ, write and time to
1920 * allow the matching interval to be extended at each write.
1921 */
1922 kvm->arch.last_tsc_nsec = ns;
1923 kvm->arch.last_tsc_write = data;
1924 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1925
1926 vcpu->arch.last_guest_tsc = data;
1927
1928 /* Keep track of which generation this VCPU has synchronized to */
1929 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1930 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1931 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1932
1933 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1934 update_ia32_tsc_adjust_msr(vcpu, offset);
1935
1936 kvm_vcpu_write_tsc_offset(vcpu, offset);
1937 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1938
1939 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1940 if (!matched) {
1941 kvm->arch.nr_vcpus_matched_tsc = 0;
1942 } else if (!already_matched) {
1943 kvm->arch.nr_vcpus_matched_tsc++;
1944 }
1945
1946 kvm_track_tsc_matching(vcpu);
1947 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1948}
1949
1950EXPORT_SYMBOL_GPL(kvm_write_tsc);
1951
1952static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1953 s64 adjustment)
1954{
1955 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1956 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
1957}
1958
1959static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1960{
1961 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1962 WARN_ON(adjustment < 0);
1963 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1964 adjust_tsc_offset_guest(vcpu, adjustment);
1965}
1966
1967#ifdef CONFIG_X86_64
1968
1969static u64 read_tsc(void)
1970{
1971 u64 ret = (u64)rdtsc_ordered();
1972 u64 last = pvclock_gtod_data.clock.cycle_last;
1973
1974 if (likely(ret >= last))
1975 return ret;
1976
1977 /*
1978 * GCC likes to generate cmov here, but this branch is extremely
1979 * predictable (it's just a function of time and the likely is
1980 * very likely) and there's a data dependence, so force GCC
1981 * to generate a branch instead. I don't barrier() because
1982 * we don't actually need a barrier, and if this function
1983 * ever gets inlined it will generate worse code.
1984 */
1985 asm volatile ("");
1986 return last;
1987}
1988
1989static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
1990{
1991 long v;
1992 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1993 u64 tsc_pg_val;
1994
1995 switch (gtod->clock.vclock_mode) {
1996 case VCLOCK_HVCLOCK:
1997 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
1998 tsc_timestamp);
1999 if (tsc_pg_val != U64_MAX) {
2000 /* TSC page valid */
2001 *mode = VCLOCK_HVCLOCK;
2002 v = (tsc_pg_val - gtod->clock.cycle_last) &
2003 gtod->clock.mask;
2004 } else {
2005 /* TSC page invalid */
2006 *mode = VCLOCK_NONE;
2007 }
2008 break;
2009 case VCLOCK_TSC:
2010 *mode = VCLOCK_TSC;
2011 *tsc_timestamp = read_tsc();
2012 v = (*tsc_timestamp - gtod->clock.cycle_last) &
2013 gtod->clock.mask;
2014 break;
2015 default:
2016 *mode = VCLOCK_NONE;
2017 }
2018
2019 if (*mode == VCLOCK_NONE)
2020 *tsc_timestamp = v = 0;
2021
2022 return v * gtod->clock.mult;
2023}
2024
2025static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
2026{
2027 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2028 unsigned long seq;
2029 int mode;
2030 u64 ns;
2031
2032 do {
2033 seq = read_seqcount_begin(&gtod->seq);
2034 ns = gtod->nsec_base;
2035 ns += vgettsc(tsc_timestamp, &mode);
2036 ns >>= gtod->clock.shift;
2037 ns += gtod->boot_ns;
2038 } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2039 *t = ns;
2040
2041 return mode;
2042}
2043
2044static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2045{
2046 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2047 unsigned long seq;
2048 int mode;
2049 u64 ns;
2050
2051 do {
2052 seq = read_seqcount_begin(&gtod->seq);
2053 ts->tv_sec = gtod->wall_time_sec;
2054 ns = gtod->nsec_base;
2055 ns += vgettsc(tsc_timestamp, &mode);
2056 ns >>= gtod->clock.shift;
2057 } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
2058
2059 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2060 ts->tv_nsec = ns;
2061
2062 return mode;
2063}
2064
2065/* returns true if host is using TSC based clocksource */
2066static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2067{
2068 /* checked again under seqlock below */
2069 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2070 return false;
2071
2072 return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
2073 tsc_timestamp));
2074}
2075
2076/* returns true if host is using TSC based clocksource */
2077static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2078 u64 *tsc_timestamp)
2079{
2080 /* checked again under seqlock below */
2081 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2082 return false;
2083
2084 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2085}
2086#endif
2087
2088/*
2089 *
2090 * Assuming a stable TSC across physical CPUS, and a stable TSC
2091 * across virtual CPUs, the following condition is possible.
2092 * Each numbered line represents an event visible to both
2093 * CPUs at the next numbered event.
2094 *
2095 * "timespecX" represents host monotonic time. "tscX" represents
2096 * RDTSC value.
2097 *
2098 * VCPU0 on CPU0 | VCPU1 on CPU1
2099 *
2100 * 1. read timespec0,tsc0
2101 * 2. | timespec1 = timespec0 + N
2102 * | tsc1 = tsc0 + M
2103 * 3. transition to guest | transition to guest
2104 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2105 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2106 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2107 *
2108 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2109 *
2110 * - ret0 < ret1
2111 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2112 * ...
2113 * - 0 < N - M => M < N
2114 *
2115 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2116 * always the case (the difference between two distinct xtime instances
2117 * might be smaller then the difference between corresponding TSC reads,
2118 * when updating guest vcpus pvclock areas).
2119 *
2120 * To avoid that problem, do not allow visibility of distinct
2121 * system_timestamp/tsc_timestamp values simultaneously: use a master
2122 * copy of host monotonic time values. Update that master copy
2123 * in lockstep.
2124 *
2125 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2126 *
2127 */
2128
2129static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2130{
2131#ifdef CONFIG_X86_64
2132 struct kvm_arch *ka = &kvm->arch;
2133 int vclock_mode;
2134 bool host_tsc_clocksource, vcpus_matched;
2135
2136 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2137 atomic_read(&kvm->online_vcpus));
2138
2139 /*
2140 * If the host uses TSC clock, then passthrough TSC as stable
2141 * to the guest.
2142 */
2143 host_tsc_clocksource = kvm_get_time_and_clockread(
2144 &ka->master_kernel_ns,
2145 &ka->master_cycle_now);
2146
2147 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2148 && !ka->backwards_tsc_observed
2149 && !ka->boot_vcpu_runs_old_kvmclock;
2150
2151 if (ka->use_master_clock)
2152 atomic_set(&kvm_guest_has_master_clock, 1);
2153
2154 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2155 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2156 vcpus_matched);
2157#endif
2158}
2159
2160void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2161{
2162 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2163}
2164
2165static void kvm_gen_update_masterclock(struct kvm *kvm)
2166{
2167#ifdef CONFIG_X86_64
2168 int i;
2169 struct kvm_vcpu *vcpu;
2170 struct kvm_arch *ka = &kvm->arch;
2171
2172 spin_lock(&ka->pvclock_gtod_sync_lock);
2173 kvm_make_mclock_inprogress_request(kvm);
2174 /* no guest entries from this point */
2175 pvclock_update_vm_gtod_copy(kvm);
2176
2177 kvm_for_each_vcpu(i, vcpu, kvm)
2178 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2179
2180 /* guest entries allowed */
2181 kvm_for_each_vcpu(i, vcpu, kvm)
2182 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2183
2184 spin_unlock(&ka->pvclock_gtod_sync_lock);
2185#endif
2186}
2187
2188u64 get_kvmclock_ns(struct kvm *kvm)
2189{
2190 struct kvm_arch *ka = &kvm->arch;
2191 struct pvclock_vcpu_time_info hv_clock;
2192 u64 ret;
2193
2194 spin_lock(&ka->pvclock_gtod_sync_lock);
2195 if (!ka->use_master_clock) {
2196 spin_unlock(&ka->pvclock_gtod_sync_lock);
2197 return ktime_get_boottime_ns() + ka->kvmclock_offset;
2198 }
2199
2200 hv_clock.tsc_timestamp = ka->master_cycle_now;
2201 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2202 spin_unlock(&ka->pvclock_gtod_sync_lock);
2203
2204 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2205 get_cpu();
2206
2207 if (__this_cpu_read(cpu_tsc_khz)) {
2208 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2209 &hv_clock.tsc_shift,
2210 &hv_clock.tsc_to_system_mul);
2211 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2212 } else
2213 ret = ktime_get_boottime_ns() + ka->kvmclock_offset;
2214
2215 put_cpu();
2216
2217 return ret;
2218}
2219
2220static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2221{
2222 struct kvm_vcpu_arch *vcpu = &v->arch;
2223 struct pvclock_vcpu_time_info guest_hv_clock;
2224
2225 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2226 &guest_hv_clock, sizeof(guest_hv_clock))))
2227 return;
2228
2229 /* This VCPU is paused, but it's legal for a guest to read another
2230 * VCPU's kvmclock, so we really have to follow the specification where
2231 * it says that version is odd if data is being modified, and even after
2232 * it is consistent.
2233 *
2234 * Version field updates must be kept separate. This is because
2235 * kvm_write_guest_cached might use a "rep movs" instruction, and
2236 * writes within a string instruction are weakly ordered. So there
2237 * are three writes overall.
2238 *
2239 * As a small optimization, only write the version field in the first
2240 * and third write. The vcpu->pv_time cache is still valid, because the
2241 * version field is the first in the struct.
2242 */
2243 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2244
2245 if (guest_hv_clock.version & 1)
2246 ++guest_hv_clock.version; /* first time write, random junk */
2247
2248 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2249 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2250 &vcpu->hv_clock,
2251 sizeof(vcpu->hv_clock.version));
2252
2253 smp_wmb();
2254
2255 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2256 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2257
2258 if (vcpu->pvclock_set_guest_stopped_request) {
2259 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2260 vcpu->pvclock_set_guest_stopped_request = false;
2261 }
2262
2263 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2264
2265 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2266 &vcpu->hv_clock,
2267 sizeof(vcpu->hv_clock));
2268
2269 smp_wmb();
2270
2271 vcpu->hv_clock.version++;
2272 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2273 &vcpu->hv_clock,
2274 sizeof(vcpu->hv_clock.version));
2275}
2276
2277static int kvm_guest_time_update(struct kvm_vcpu *v)
2278{
2279 unsigned long flags, tgt_tsc_khz;
2280 struct kvm_vcpu_arch *vcpu = &v->arch;
2281 struct kvm_arch *ka = &v->kvm->arch;
2282 s64 kernel_ns;
2283 u64 tsc_timestamp, host_tsc;
2284 u8 pvclock_flags;
2285 bool use_master_clock;
2286
2287 kernel_ns = 0;
2288 host_tsc = 0;
2289
2290 /*
2291 * If the host uses TSC clock, then passthrough TSC as stable
2292 * to the guest.
2293 */
2294 spin_lock(&ka->pvclock_gtod_sync_lock);
2295 use_master_clock = ka->use_master_clock;
2296 if (use_master_clock) {
2297 host_tsc = ka->master_cycle_now;
2298 kernel_ns = ka->master_kernel_ns;
2299 }
2300 spin_unlock(&ka->pvclock_gtod_sync_lock);
2301
2302 /* Keep irq disabled to prevent changes to the clock */
2303 local_irq_save(flags);
2304 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2305 if (unlikely(tgt_tsc_khz == 0)) {
2306 local_irq_restore(flags);
2307 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2308 return 1;
2309 }
2310 if (!use_master_clock) {
2311 host_tsc = rdtsc();
2312 kernel_ns = ktime_get_boottime_ns();
2313 }
2314
2315 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2316
2317 /*
2318 * We may have to catch up the TSC to match elapsed wall clock
2319 * time for two reasons, even if kvmclock is used.
2320 * 1) CPU could have been running below the maximum TSC rate
2321 * 2) Broken TSC compensation resets the base at each VCPU
2322 * entry to avoid unknown leaps of TSC even when running
2323 * again on the same CPU. This may cause apparent elapsed
2324 * time to disappear, and the guest to stand still or run
2325 * very slowly.
2326 */
2327 if (vcpu->tsc_catchup) {
2328 u64 tsc = compute_guest_tsc(v, kernel_ns);
2329 if (tsc > tsc_timestamp) {
2330 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2331 tsc_timestamp = tsc;
2332 }
2333 }
2334
2335 local_irq_restore(flags);
2336
2337 /* With all the info we got, fill in the values */
2338
2339 if (kvm_has_tsc_control)
2340 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2341
2342 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2343 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2344 &vcpu->hv_clock.tsc_shift,
2345 &vcpu->hv_clock.tsc_to_system_mul);
2346 vcpu->hw_tsc_khz = tgt_tsc_khz;
2347 }
2348
2349 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2350 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2351 vcpu->last_guest_tsc = tsc_timestamp;
2352
2353 /* If the host uses TSC clocksource, then it is stable */
2354 pvclock_flags = 0;
2355 if (use_master_clock)
2356 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2357
2358 vcpu->hv_clock.flags = pvclock_flags;
2359
2360 if (vcpu->pv_time_enabled)
2361 kvm_setup_pvclock_page(v);
2362 if (v == kvm_get_vcpu(v->kvm, 0))
2363 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2364 return 0;
2365}
2366
2367/*
2368 * kvmclock updates which are isolated to a given vcpu, such as
2369 * vcpu->cpu migration, should not allow system_timestamp from
2370 * the rest of the vcpus to remain static. Otherwise ntp frequency
2371 * correction applies to one vcpu's system_timestamp but not
2372 * the others.
2373 *
2374 * So in those cases, request a kvmclock update for all vcpus.
2375 * We need to rate-limit these requests though, as they can
2376 * considerably slow guests that have a large number of vcpus.
2377 * The time for a remote vcpu to update its kvmclock is bound
2378 * by the delay we use to rate-limit the updates.
2379 */
2380
2381#define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2382
2383static void kvmclock_update_fn(struct work_struct *work)
2384{
2385 int i;
2386 struct delayed_work *dwork = to_delayed_work(work);
2387 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2388 kvmclock_update_work);
2389 struct kvm *kvm = container_of(ka, struct kvm, arch);
2390 struct kvm_vcpu *vcpu;
2391
2392 kvm_for_each_vcpu(i, vcpu, kvm) {
2393 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2394 kvm_vcpu_kick(vcpu);
2395 }
2396}
2397
2398static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2399{
2400 struct kvm *kvm = v->kvm;
2401
2402 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2403 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2404 KVMCLOCK_UPDATE_DELAY);
2405}
2406
2407#define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2408
2409static void kvmclock_sync_fn(struct work_struct *work)
2410{
2411 struct delayed_work *dwork = to_delayed_work(work);
2412 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2413 kvmclock_sync_work);
2414 struct kvm *kvm = container_of(ka, struct kvm, arch);
2415
2416 if (!kvmclock_periodic_sync)
2417 return;
2418
2419 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2420 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2421 KVMCLOCK_SYNC_PERIOD);
2422}
2423
2424/*
2425 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2426 */
2427static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2428{
2429 /* McStatusWrEn enabled? */
2430 if (guest_cpuid_is_amd(vcpu))
2431 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2432
2433 return false;
2434}
2435
2436static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2437{
2438 u64 mcg_cap = vcpu->arch.mcg_cap;
2439 unsigned bank_num = mcg_cap & 0xff;
2440 u32 msr = msr_info->index;
2441 u64 data = msr_info->data;
2442
2443 switch (msr) {
2444 case MSR_IA32_MCG_STATUS:
2445 vcpu->arch.mcg_status = data;
2446 break;
2447 case MSR_IA32_MCG_CTL:
2448 if (!(mcg_cap & MCG_CTL_P) &&
2449 (data || !msr_info->host_initiated))
2450 return 1;
2451 if (data != 0 && data != ~(u64)0)
2452 return 1;
2453 vcpu->arch.mcg_ctl = data;
2454 break;
2455 default:
2456 if (msr >= MSR_IA32_MC0_CTL &&
2457 msr < MSR_IA32_MCx_CTL(bank_num)) {
2458 u32 offset = msr - MSR_IA32_MC0_CTL;
2459 /* only 0 or all 1s can be written to IA32_MCi_CTL
2460 * some Linux kernels though clear bit 10 in bank 4 to
2461 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2462 * this to avoid an uncatched #GP in the guest
2463 */
2464 if ((offset & 0x3) == 0 &&
2465 data != 0 && (data | (1 << 10)) != ~(u64)0)
2466 return -1;
2467
2468 /* MCi_STATUS */
2469 if (!msr_info->host_initiated &&
2470 (offset & 0x3) == 1 && data != 0) {
2471 if (!can_set_mci_status(vcpu))
2472 return -1;
2473 }
2474
2475 vcpu->arch.mce_banks[offset] = data;
2476 break;
2477 }
2478 return 1;
2479 }
2480 return 0;
2481}
2482
2483static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2484{
2485 struct kvm *kvm = vcpu->kvm;
2486 int lm = is_long_mode(vcpu);
2487 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2488 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2489 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2490 : kvm->arch.xen_hvm_config.blob_size_32;
2491 u32 page_num = data & ~PAGE_MASK;
2492 u64 page_addr = data & PAGE_MASK;
2493 u8 *page;
2494 int r;
2495
2496 r = -E2BIG;
2497 if (page_num >= blob_size)
2498 goto out;
2499 r = -ENOMEM;
2500 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2501 if (IS_ERR(page)) {
2502 r = PTR_ERR(page);
2503 goto out;
2504 }
2505 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2506 goto out_free;
2507 r = 0;
2508out_free:
2509 kfree(page);
2510out:
2511 return r;
2512}
2513
2514static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2515{
2516 gpa_t gpa = data & ~0x3f;
2517
2518 /* Bits 3:5 are reserved, Should be zero */
2519 if (data & 0x38)
2520 return 1;
2521
2522 vcpu->arch.apf.msr_val = data;
2523
2524 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2525 kvm_clear_async_pf_completion_queue(vcpu);
2526 kvm_async_pf_hash_reset(vcpu);
2527 return 0;
2528 }
2529
2530 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2531 sizeof(u32)))
2532 return 1;
2533
2534 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2535 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2536 kvm_async_pf_wakeup_all(vcpu);
2537 return 0;
2538}
2539
2540static void kvmclock_reset(struct kvm_vcpu *vcpu)
2541{
2542 vcpu->arch.pv_time_enabled = false;
2543}
2544
2545static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2546{
2547 ++vcpu->stat.tlb_flush;
2548 kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2549}
2550
2551static void record_steal_time(struct kvm_vcpu *vcpu)
2552{
2553 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2554 return;
2555
2556 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2557 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2558 return;
2559
2560 /*
2561 * Doing a TLB flush here, on the guest's behalf, can avoid
2562 * expensive IPIs.
2563 */
2564 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2565 vcpu->arch.st.steal.preempted & KVM_VCPU_FLUSH_TLB);
2566 if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2567 kvm_vcpu_flush_tlb(vcpu, false);
2568
2569 if (vcpu->arch.st.steal.version & 1)
2570 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2571
2572 vcpu->arch.st.steal.version += 1;
2573
2574 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2575 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2576
2577 smp_wmb();
2578
2579 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2580 vcpu->arch.st.last_steal;
2581 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2582
2583 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2584 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2585
2586 smp_wmb();
2587
2588 vcpu->arch.st.steal.version += 1;
2589
2590 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2591 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2592}
2593
2594int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2595{
2596 bool pr = false;
2597 u32 msr = msr_info->index;
2598 u64 data = msr_info->data;
2599
2600 switch (msr) {
2601 case MSR_AMD64_NB_CFG:
2602 case MSR_IA32_UCODE_WRITE:
2603 case MSR_VM_HSAVE_PA:
2604 case MSR_AMD64_PATCH_LOADER:
2605 case MSR_AMD64_BU_CFG2:
2606 case MSR_AMD64_DC_CFG:
2607 case MSR_F15H_EX_CFG:
2608 break;
2609
2610 case MSR_IA32_UCODE_REV:
2611 if (msr_info->host_initiated)
2612 vcpu->arch.microcode_version = data;
2613 break;
2614 case MSR_IA32_ARCH_CAPABILITIES:
2615 if (!msr_info->host_initiated)
2616 return 1;
2617 vcpu->arch.arch_capabilities = data;
2618 break;
2619 case MSR_EFER:
2620 return set_efer(vcpu, msr_info);
2621 case MSR_K7_HWCR:
2622 data &= ~(u64)0x40; /* ignore flush filter disable */
2623 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2624 data &= ~(u64)0x8; /* ignore TLB cache disable */
2625
2626 /* Handle McStatusWrEn */
2627 if (data == BIT_ULL(18)) {
2628 vcpu->arch.msr_hwcr = data;
2629 } else if (data != 0) {
2630 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2631 data);
2632 return 1;
2633 }
2634 break;
2635 case MSR_FAM10H_MMIO_CONF_BASE:
2636 if (data != 0) {
2637 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2638 "0x%llx\n", data);
2639 return 1;
2640 }
2641 break;
2642 case MSR_IA32_DEBUGCTLMSR:
2643 if (!data) {
2644 /* We support the non-activated case already */
2645 break;
2646 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2647 /* Values other than LBR and BTF are vendor-specific,
2648 thus reserved and should throw a #GP */
2649 return 1;
2650 }
2651 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2652 __func__, data);
2653 break;
2654 case 0x200 ... 0x2ff:
2655 return kvm_mtrr_set_msr(vcpu, msr, data);
2656 case MSR_IA32_APICBASE:
2657 return kvm_set_apic_base(vcpu, msr_info);
2658 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2659 return kvm_x2apic_msr_write(vcpu, msr, data);
2660 case MSR_IA32_TSCDEADLINE:
2661 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2662 break;
2663 case MSR_IA32_TSC_ADJUST:
2664 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2665 if (!msr_info->host_initiated) {
2666 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2667 adjust_tsc_offset_guest(vcpu, adj);
2668 }
2669 vcpu->arch.ia32_tsc_adjust_msr = data;
2670 }
2671 break;
2672 case MSR_IA32_MISC_ENABLE:
2673 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
2674 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
2675 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
2676 return 1;
2677 vcpu->arch.ia32_misc_enable_msr = data;
2678 kvm_update_cpuid(vcpu);
2679 } else {
2680 vcpu->arch.ia32_misc_enable_msr = data;
2681 }
2682 break;
2683 case MSR_IA32_SMBASE:
2684 if (!msr_info->host_initiated)
2685 return 1;
2686 vcpu->arch.smbase = data;
2687 break;
2688 case MSR_IA32_POWER_CTL:
2689 vcpu->arch.msr_ia32_power_ctl = data;
2690 break;
2691 case MSR_IA32_TSC:
2692 kvm_write_tsc(vcpu, msr_info);
2693 break;
2694 case MSR_SMI_COUNT:
2695 if (!msr_info->host_initiated)
2696 return 1;
2697 vcpu->arch.smi_count = data;
2698 break;
2699 case MSR_KVM_WALL_CLOCK_NEW:
2700 case MSR_KVM_WALL_CLOCK:
2701 vcpu->kvm->arch.wall_clock = data;
2702 kvm_write_wall_clock(vcpu->kvm, data);
2703 break;
2704 case MSR_KVM_SYSTEM_TIME_NEW:
2705 case MSR_KVM_SYSTEM_TIME: {
2706 struct kvm_arch *ka = &vcpu->kvm->arch;
2707
2708 kvmclock_reset(vcpu);
2709
2710 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2711 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2712
2713 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2714 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2715
2716 ka->boot_vcpu_runs_old_kvmclock = tmp;
2717 }
2718
2719 vcpu->arch.time = data;
2720 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2721
2722 /* we verify if the enable bit is set... */
2723 if (!(data & 1))
2724 break;
2725
2726 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2727 &vcpu->arch.pv_time, data & ~1ULL,
2728 sizeof(struct pvclock_vcpu_time_info)))
2729 vcpu->arch.pv_time_enabled = false;
2730 else
2731 vcpu->arch.pv_time_enabled = true;
2732
2733 break;
2734 }
2735 case MSR_KVM_ASYNC_PF_EN:
2736 if (kvm_pv_enable_async_pf(vcpu, data))
2737 return 1;
2738 break;
2739 case MSR_KVM_STEAL_TIME:
2740
2741 if (unlikely(!sched_info_on()))
2742 return 1;
2743
2744 if (data & KVM_STEAL_RESERVED_MASK)
2745 return 1;
2746
2747 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2748 data & KVM_STEAL_VALID_BITS,
2749 sizeof(struct kvm_steal_time)))
2750 return 1;
2751
2752 vcpu->arch.st.msr_val = data;
2753
2754 if (!(data & KVM_MSR_ENABLED))
2755 break;
2756
2757 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2758
2759 break;
2760 case MSR_KVM_PV_EOI_EN:
2761 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2762 return 1;
2763 break;
2764
2765 case MSR_KVM_POLL_CONTROL:
2766 /* only enable bit supported */
2767 if (data & (-1ULL << 1))
2768 return 1;
2769
2770 vcpu->arch.msr_kvm_poll_control = data;
2771 break;
2772
2773 case MSR_IA32_MCG_CTL:
2774 case MSR_IA32_MCG_STATUS:
2775 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2776 return set_msr_mce(vcpu, msr_info);
2777
2778 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2779 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2780 pr = true; /* fall through */
2781 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2782 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2783 if (kvm_pmu_is_valid_msr(vcpu, msr))
2784 return kvm_pmu_set_msr(vcpu, msr_info);
2785
2786 if (pr || data != 0)
2787 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2788 "0x%x data 0x%llx\n", msr, data);
2789 break;
2790 case MSR_K7_CLK_CTL:
2791 /*
2792 * Ignore all writes to this no longer documented MSR.
2793 * Writes are only relevant for old K7 processors,
2794 * all pre-dating SVM, but a recommended workaround from
2795 * AMD for these chips. It is possible to specify the
2796 * affected processor models on the command line, hence
2797 * the need to ignore the workaround.
2798 */
2799 break;
2800 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2801 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2802 case HV_X64_MSR_CRASH_CTL:
2803 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2804 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2805 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2806 case HV_X64_MSR_TSC_EMULATION_STATUS:
2807 return kvm_hv_set_msr_common(vcpu, msr, data,
2808 msr_info->host_initiated);
2809 case MSR_IA32_BBL_CR_CTL3:
2810 /* Drop writes to this legacy MSR -- see rdmsr
2811 * counterpart for further detail.
2812 */
2813 if (report_ignored_msrs)
2814 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2815 msr, data);
2816 break;
2817 case MSR_AMD64_OSVW_ID_LENGTH:
2818 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2819 return 1;
2820 vcpu->arch.osvw.length = data;
2821 break;
2822 case MSR_AMD64_OSVW_STATUS:
2823 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2824 return 1;
2825 vcpu->arch.osvw.status = data;
2826 break;
2827 case MSR_PLATFORM_INFO:
2828 if (!msr_info->host_initiated ||
2829 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2830 cpuid_fault_enabled(vcpu)))
2831 return 1;
2832 vcpu->arch.msr_platform_info = data;
2833 break;
2834 case MSR_MISC_FEATURES_ENABLES:
2835 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2836 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2837 !supports_cpuid_fault(vcpu)))
2838 return 1;
2839 vcpu->arch.msr_misc_features_enables = data;
2840 break;
2841 default:
2842 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2843 return xen_hvm_config(vcpu, data);
2844 if (kvm_pmu_is_valid_msr(vcpu, msr))
2845 return kvm_pmu_set_msr(vcpu, msr_info);
2846 if (!ignore_msrs) {
2847 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2848 msr, data);
2849 return 1;
2850 } else {
2851 if (report_ignored_msrs)
2852 vcpu_unimpl(vcpu,
2853 "ignored wrmsr: 0x%x data 0x%llx\n",
2854 msr, data);
2855 break;
2856 }
2857 }
2858 return 0;
2859}
2860EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2861
2862static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2863{
2864 u64 data;
2865 u64 mcg_cap = vcpu->arch.mcg_cap;
2866 unsigned bank_num = mcg_cap & 0xff;
2867
2868 switch (msr) {
2869 case MSR_IA32_P5_MC_ADDR:
2870 case MSR_IA32_P5_MC_TYPE:
2871 data = 0;
2872 break;
2873 case MSR_IA32_MCG_CAP:
2874 data = vcpu->arch.mcg_cap;
2875 break;
2876 case MSR_IA32_MCG_CTL:
2877 if (!(mcg_cap & MCG_CTL_P) && !host)
2878 return 1;
2879 data = vcpu->arch.mcg_ctl;
2880 break;
2881 case MSR_IA32_MCG_STATUS:
2882 data = vcpu->arch.mcg_status;
2883 break;
2884 default:
2885 if (msr >= MSR_IA32_MC0_CTL &&
2886 msr < MSR_IA32_MCx_CTL(bank_num)) {
2887 u32 offset = msr - MSR_IA32_MC0_CTL;
2888 data = vcpu->arch.mce_banks[offset];
2889 break;
2890 }
2891 return 1;
2892 }
2893 *pdata = data;
2894 return 0;
2895}
2896
2897int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2898{
2899 switch (msr_info->index) {
2900 case MSR_IA32_PLATFORM_ID:
2901 case MSR_IA32_EBL_CR_POWERON:
2902 case MSR_IA32_DEBUGCTLMSR:
2903 case MSR_IA32_LASTBRANCHFROMIP:
2904 case MSR_IA32_LASTBRANCHTOIP:
2905 case MSR_IA32_LASTINTFROMIP:
2906 case MSR_IA32_LASTINTTOIP:
2907 case MSR_K8_SYSCFG:
2908 case MSR_K8_TSEG_ADDR:
2909 case MSR_K8_TSEG_MASK:
2910 case MSR_VM_HSAVE_PA:
2911 case MSR_K8_INT_PENDING_MSG:
2912 case MSR_AMD64_NB_CFG:
2913 case MSR_FAM10H_MMIO_CONF_BASE:
2914 case MSR_AMD64_BU_CFG2:
2915 case MSR_IA32_PERF_CTL:
2916 case MSR_AMD64_DC_CFG:
2917 case MSR_F15H_EX_CFG:
2918 msr_info->data = 0;
2919 break;
2920 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2921 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2922 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2923 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2924 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2925 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2926 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2927 msr_info->data = 0;
2928 break;
2929 case MSR_IA32_UCODE_REV:
2930 msr_info->data = vcpu->arch.microcode_version;
2931 break;
2932 case MSR_IA32_ARCH_CAPABILITIES:
2933 if (!msr_info->host_initiated &&
2934 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
2935 return 1;
2936 msr_info->data = vcpu->arch.arch_capabilities;
2937 break;
2938 case MSR_IA32_POWER_CTL:
2939 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
2940 break;
2941 case MSR_IA32_TSC:
2942 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2943 break;
2944 case MSR_MTRRcap:
2945 case 0x200 ... 0x2ff:
2946 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2947 case 0xcd: /* fsb frequency */
2948 msr_info->data = 3;
2949 break;
2950 /*
2951 * MSR_EBC_FREQUENCY_ID
2952 * Conservative value valid for even the basic CPU models.
2953 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2954 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2955 * and 266MHz for model 3, or 4. Set Core Clock
2956 * Frequency to System Bus Frequency Ratio to 1 (bits
2957 * 31:24) even though these are only valid for CPU
2958 * models > 2, however guests may end up dividing or
2959 * multiplying by zero otherwise.
2960 */
2961 case MSR_EBC_FREQUENCY_ID:
2962 msr_info->data = 1 << 24;
2963 break;
2964 case MSR_IA32_APICBASE:
2965 msr_info->data = kvm_get_apic_base(vcpu);
2966 break;
2967 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2968 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2969 break;
2970 case MSR_IA32_TSCDEADLINE:
2971 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2972 break;
2973 case MSR_IA32_TSC_ADJUST:
2974 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2975 break;
2976 case MSR_IA32_MISC_ENABLE:
2977 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2978 break;
2979 case MSR_IA32_SMBASE:
2980 if (!msr_info->host_initiated)
2981 return 1;
2982 msr_info->data = vcpu->arch.smbase;
2983 break;
2984 case MSR_SMI_COUNT:
2985 msr_info->data = vcpu->arch.smi_count;
2986 break;
2987 case MSR_IA32_PERF_STATUS:
2988 /* TSC increment by tick */
2989 msr_info->data = 1000ULL;
2990 /* CPU multiplier */
2991 msr_info->data |= (((uint64_t)4ULL) << 40);
2992 break;
2993 case MSR_EFER:
2994 msr_info->data = vcpu->arch.efer;
2995 break;
2996 case MSR_KVM_WALL_CLOCK:
2997 case MSR_KVM_WALL_CLOCK_NEW:
2998 msr_info->data = vcpu->kvm->arch.wall_clock;
2999 break;
3000 case MSR_KVM_SYSTEM_TIME:
3001 case MSR_KVM_SYSTEM_TIME_NEW:
3002 msr_info->data = vcpu->arch.time;
3003 break;
3004 case MSR_KVM_ASYNC_PF_EN:
3005 msr_info->data = vcpu->arch.apf.msr_val;
3006 break;
3007 case MSR_KVM_STEAL_TIME:
3008 msr_info->data = vcpu->arch.st.msr_val;
3009 break;
3010 case MSR_KVM_PV_EOI_EN:
3011 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3012 break;
3013 case MSR_KVM_POLL_CONTROL:
3014 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3015 break;
3016 case MSR_IA32_P5_MC_ADDR:
3017 case MSR_IA32_P5_MC_TYPE:
3018 case MSR_IA32_MCG_CAP:
3019 case MSR_IA32_MCG_CTL:
3020 case MSR_IA32_MCG_STATUS:
3021 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3022 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3023 msr_info->host_initiated);
3024 case MSR_K7_CLK_CTL:
3025 /*
3026 * Provide expected ramp-up count for K7. All other
3027 * are set to zero, indicating minimum divisors for
3028 * every field.
3029 *
3030 * This prevents guest kernels on AMD host with CPU
3031 * type 6, model 8 and higher from exploding due to
3032 * the rdmsr failing.
3033 */
3034 msr_info->data = 0x20000000;
3035 break;
3036 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3037 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3038 case HV_X64_MSR_CRASH_CTL:
3039 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3040 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3041 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3042 case HV_X64_MSR_TSC_EMULATION_STATUS:
3043 return kvm_hv_get_msr_common(vcpu,
3044 msr_info->index, &msr_info->data,
3045 msr_info->host_initiated);
3046 break;
3047 case MSR_IA32_BBL_CR_CTL3:
3048 /* This legacy MSR exists but isn't fully documented in current
3049 * silicon. It is however accessed by winxp in very narrow
3050 * scenarios where it sets bit #19, itself documented as
3051 * a "reserved" bit. Best effort attempt to source coherent
3052 * read data here should the balance of the register be
3053 * interpreted by the guest:
3054 *
3055 * L2 cache control register 3: 64GB range, 256KB size,
3056 * enabled, latency 0x1, configured
3057 */
3058 msr_info->data = 0xbe702111;
3059 break;
3060 case MSR_AMD64_OSVW_ID_LENGTH:
3061 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3062 return 1;
3063 msr_info->data = vcpu->arch.osvw.length;
3064 break;
3065 case MSR_AMD64_OSVW_STATUS:
3066 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3067 return 1;
3068 msr_info->data = vcpu->arch.osvw.status;
3069 break;
3070 case MSR_PLATFORM_INFO:
3071 if (!msr_info->host_initiated &&
3072 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3073 return 1;
3074 msr_info->data = vcpu->arch.msr_platform_info;
3075 break;
3076 case MSR_MISC_FEATURES_ENABLES:
3077 msr_info->data = vcpu->arch.msr_misc_features_enables;
3078 break;
3079 case MSR_K7_HWCR:
3080 msr_info->data = vcpu->arch.msr_hwcr;
3081 break;
3082 default:
3083 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3084 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3085 if (!ignore_msrs) {
3086 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
3087 msr_info->index);
3088 return 1;
3089 } else {
3090 if (report_ignored_msrs)
3091 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
3092 msr_info->index);
3093 msr_info->data = 0;
3094 }
3095 break;
3096 }
3097 return 0;
3098}
3099EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3100
3101/*
3102 * Read or write a bunch of msrs. All parameters are kernel addresses.
3103 *
3104 * @return number of msrs set successfully.
3105 */
3106static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3107 struct kvm_msr_entry *entries,
3108 int (*do_msr)(struct kvm_vcpu *vcpu,
3109 unsigned index, u64 *data))
3110{
3111 int i;
3112
3113 for (i = 0; i < msrs->nmsrs; ++i)
3114 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3115 break;
3116
3117 return i;
3118}
3119
3120/*
3121 * Read or write a bunch of msrs. Parameters are user addresses.
3122 *
3123 * @return number of msrs set successfully.
3124 */
3125static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3126 int (*do_msr)(struct kvm_vcpu *vcpu,
3127 unsigned index, u64 *data),
3128 int writeback)
3129{
3130 struct kvm_msrs msrs;
3131 struct kvm_msr_entry *entries;
3132 int r, n;
3133 unsigned size;
3134
3135 r = -EFAULT;
3136 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3137 goto out;
3138
3139 r = -E2BIG;
3140 if (msrs.nmsrs >= MAX_IO_MSRS)
3141 goto out;
3142
3143 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3144 entries = memdup_user(user_msrs->entries, size);
3145 if (IS_ERR(entries)) {
3146 r = PTR_ERR(entries);
3147 goto out;
3148 }
3149
3150 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3151 if (r < 0)
3152 goto out_free;
3153
3154 r = -EFAULT;
3155 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3156 goto out_free;
3157
3158 r = n;
3159
3160out_free:
3161 kfree(entries);
3162out:
3163 return r;
3164}
3165
3166static inline bool kvm_can_mwait_in_guest(void)
3167{
3168 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3169 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3170 boot_cpu_has(X86_FEATURE_ARAT);
3171}
3172
3173int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3174{
3175 int r = 0;
3176
3177 switch (ext) {
3178 case KVM_CAP_IRQCHIP:
3179 case KVM_CAP_HLT:
3180 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3181 case KVM_CAP_SET_TSS_ADDR:
3182 case KVM_CAP_EXT_CPUID:
3183 case KVM_CAP_EXT_EMUL_CPUID:
3184 case KVM_CAP_CLOCKSOURCE:
3185 case KVM_CAP_PIT:
3186 case KVM_CAP_NOP_IO_DELAY:
3187 case KVM_CAP_MP_STATE:
3188 case KVM_CAP_SYNC_MMU:
3189 case KVM_CAP_USER_NMI:
3190 case KVM_CAP_REINJECT_CONTROL:
3191 case KVM_CAP_IRQ_INJECT_STATUS:
3192 case KVM_CAP_IOEVENTFD:
3193 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3194 case KVM_CAP_PIT2:
3195 case KVM_CAP_PIT_STATE2:
3196 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3197 case KVM_CAP_XEN_HVM:
3198 case KVM_CAP_VCPU_EVENTS:
3199 case KVM_CAP_HYPERV:
3200 case KVM_CAP_HYPERV_VAPIC:
3201 case KVM_CAP_HYPERV_SPIN:
3202 case KVM_CAP_HYPERV_SYNIC:
3203 case KVM_CAP_HYPERV_SYNIC2:
3204 case KVM_CAP_HYPERV_VP_INDEX:
3205 case KVM_CAP_HYPERV_EVENTFD:
3206 case KVM_CAP_HYPERV_TLBFLUSH:
3207 case KVM_CAP_HYPERV_SEND_IPI:
3208 case KVM_CAP_HYPERV_CPUID:
3209 case KVM_CAP_PCI_SEGMENT:
3210 case KVM_CAP_DEBUGREGS:
3211 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3212 case KVM_CAP_XSAVE:
3213 case KVM_CAP_ASYNC_PF:
3214 case KVM_CAP_GET_TSC_KHZ:
3215 case KVM_CAP_KVMCLOCK_CTRL:
3216 case KVM_CAP_READONLY_MEM:
3217 case KVM_CAP_HYPERV_TIME:
3218 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3219 case KVM_CAP_TSC_DEADLINE_TIMER:
3220 case KVM_CAP_DISABLE_QUIRKS:
3221 case KVM_CAP_SET_BOOT_CPU_ID:
3222 case KVM_CAP_SPLIT_IRQCHIP:
3223 case KVM_CAP_IMMEDIATE_EXIT:
3224 case KVM_CAP_PMU_EVENT_FILTER:
3225 case KVM_CAP_GET_MSR_FEATURES:
3226 case KVM_CAP_MSR_PLATFORM_INFO:
3227 case KVM_CAP_EXCEPTION_PAYLOAD:
3228 r = 1;
3229 break;
3230 case KVM_CAP_SYNC_REGS:
3231 r = KVM_SYNC_X86_VALID_FIELDS;
3232 break;
3233 case KVM_CAP_ADJUST_CLOCK:
3234 r = KVM_CLOCK_TSC_STABLE;
3235 break;
3236 case KVM_CAP_X86_DISABLE_EXITS:
3237 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3238 KVM_X86_DISABLE_EXITS_CSTATE;
3239 if(kvm_can_mwait_in_guest())
3240 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3241 break;
3242 case KVM_CAP_X86_SMM:
3243 /* SMBASE is usually relocated above 1M on modern chipsets,
3244 * and SMM handlers might indeed rely on 4G segment limits,
3245 * so do not report SMM to be available if real mode is
3246 * emulated via vm86 mode. Still, do not go to great lengths
3247 * to avoid userspace's usage of the feature, because it is a
3248 * fringe case that is not enabled except via specific settings
3249 * of the module parameters.
3250 */
3251 r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3252 break;
3253 case KVM_CAP_VAPIC:
3254 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3255 break;
3256 case KVM_CAP_NR_VCPUS:
3257 r = KVM_SOFT_MAX_VCPUS;
3258 break;
3259 case KVM_CAP_MAX_VCPUS:
3260 r = KVM_MAX_VCPUS;
3261 break;
3262 case KVM_CAP_MAX_VCPU_ID:
3263 r = KVM_MAX_VCPU_ID;
3264 break;
3265 case KVM_CAP_PV_MMU: /* obsolete */
3266 r = 0;
3267 break;
3268 case KVM_CAP_MCE:
3269 r = KVM_MAX_MCE_BANKS;
3270 break;
3271 case KVM_CAP_XCRS:
3272 r = boot_cpu_has(X86_FEATURE_XSAVE);
3273 break;
3274 case KVM_CAP_TSC_CONTROL:
3275 r = kvm_has_tsc_control;
3276 break;
3277 case KVM_CAP_X2APIC_API:
3278 r = KVM_X2APIC_API_VALID_FLAGS;
3279 break;
3280 case KVM_CAP_NESTED_STATE:
3281 r = kvm_x86_ops->get_nested_state ?
3282 kvm_x86_ops->get_nested_state(NULL, NULL, 0) : 0;
3283 break;
3284 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3285 r = kvm_x86_ops->enable_direct_tlbflush != NULL;
3286 break;
3287 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3288 r = kvm_x86_ops->nested_enable_evmcs != NULL;
3289 break;
3290 default:
3291 break;
3292 }
3293 return r;
3294
3295}
3296
3297long kvm_arch_dev_ioctl(struct file *filp,
3298 unsigned int ioctl, unsigned long arg)
3299{
3300 void __user *argp = (void __user *)arg;
3301 long r;
3302
3303 switch (ioctl) {
3304 case KVM_GET_MSR_INDEX_LIST: {
3305 struct kvm_msr_list __user *user_msr_list = argp;
3306 struct kvm_msr_list msr_list;
3307 unsigned n;
3308
3309 r = -EFAULT;
3310 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3311 goto out;
3312 n = msr_list.nmsrs;
3313 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3314 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3315 goto out;
3316 r = -E2BIG;
3317 if (n < msr_list.nmsrs)
3318 goto out;
3319 r = -EFAULT;
3320 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3321 num_msrs_to_save * sizeof(u32)))
3322 goto out;
3323 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3324 &emulated_msrs,
3325 num_emulated_msrs * sizeof(u32)))
3326 goto out;
3327 r = 0;
3328 break;
3329 }
3330 case KVM_GET_SUPPORTED_CPUID:
3331 case KVM_GET_EMULATED_CPUID: {
3332 struct kvm_cpuid2 __user *cpuid_arg = argp;
3333 struct kvm_cpuid2 cpuid;
3334
3335 r = -EFAULT;
3336 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3337 goto out;
3338
3339 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3340 ioctl);
3341 if (r)
3342 goto out;
3343
3344 r = -EFAULT;
3345 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3346 goto out;
3347 r = 0;
3348 break;
3349 }
3350 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3351 r = -EFAULT;
3352 if (copy_to_user(argp, &kvm_mce_cap_supported,
3353 sizeof(kvm_mce_cap_supported)))
3354 goto out;
3355 r = 0;
3356 break;
3357 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3358 struct kvm_msr_list __user *user_msr_list = argp;
3359 struct kvm_msr_list msr_list;
3360 unsigned int n;
3361
3362 r = -EFAULT;
3363 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3364 goto out;
3365 n = msr_list.nmsrs;
3366 msr_list.nmsrs = num_msr_based_features;
3367 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3368 goto out;
3369 r = -E2BIG;
3370 if (n < msr_list.nmsrs)
3371 goto out;
3372 r = -EFAULT;
3373 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3374 num_msr_based_features * sizeof(u32)))
3375 goto out;
3376 r = 0;
3377 break;
3378 }
3379 case KVM_GET_MSRS:
3380 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3381 break;
3382 }
3383 default:
3384 r = -EINVAL;
3385 }
3386out:
3387 return r;
3388}
3389
3390static void wbinvd_ipi(void *garbage)
3391{
3392 wbinvd();
3393}
3394
3395static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3396{
3397 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3398}
3399
3400void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3401{
3402 /* Address WBINVD may be executed by guest */
3403 if (need_emulate_wbinvd(vcpu)) {
3404 if (kvm_x86_ops->has_wbinvd_exit())
3405 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3406 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3407 smp_call_function_single(vcpu->cpu,
3408 wbinvd_ipi, NULL, 1);
3409 }
3410
3411 kvm_x86_ops->vcpu_load(vcpu, cpu);
3412
3413 fpregs_assert_state_consistent();
3414 if (test_thread_flag(TIF_NEED_FPU_LOAD))
3415 switch_fpu_return();
3416
3417 /* Apply any externally detected TSC adjustments (due to suspend) */
3418 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3419 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3420 vcpu->arch.tsc_offset_adjustment = 0;
3421 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3422 }
3423
3424 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3425 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3426 rdtsc() - vcpu->arch.last_host_tsc;
3427 if (tsc_delta < 0)
3428 mark_tsc_unstable("KVM discovered backwards TSC");
3429
3430 if (kvm_check_tsc_unstable()) {
3431 u64 offset = kvm_compute_tsc_offset(vcpu,
3432 vcpu->arch.last_guest_tsc);
3433 kvm_vcpu_write_tsc_offset(vcpu, offset);
3434 vcpu->arch.tsc_catchup = 1;
3435 }
3436
3437 if (kvm_lapic_hv_timer_in_use(vcpu))
3438 kvm_lapic_restart_hv_timer(vcpu);
3439
3440 /*
3441 * On a host with synchronized TSC, there is no need to update
3442 * kvmclock on vcpu->cpu migration
3443 */
3444 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3445 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3446 if (vcpu->cpu != cpu)
3447 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3448 vcpu->cpu = cpu;
3449 }
3450
3451 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3452}
3453
3454static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3455{
3456 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3457 return;
3458
3459 vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3460
3461 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3462 &vcpu->arch.st.steal.preempted,
3463 offsetof(struct kvm_steal_time, preempted),
3464 sizeof(vcpu->arch.st.steal.preempted));
3465}
3466
3467void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3468{
3469 int idx;
3470
3471 if (vcpu->preempted)
3472 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3473
3474 /*
3475 * Disable page faults because we're in atomic context here.
3476 * kvm_write_guest_offset_cached() would call might_fault()
3477 * that relies on pagefault_disable() to tell if there's a
3478 * bug. NOTE: the write to guest memory may not go through if
3479 * during postcopy live migration or if there's heavy guest
3480 * paging.
3481 */
3482 pagefault_disable();
3483 /*
3484 * kvm_memslots() will be called by
3485 * kvm_write_guest_offset_cached() so take the srcu lock.
3486 */
3487 idx = srcu_read_lock(&vcpu->kvm->srcu);
3488 kvm_steal_time_set_preempted(vcpu);
3489 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3490 pagefault_enable();
3491 kvm_x86_ops->vcpu_put(vcpu);
3492 vcpu->arch.last_host_tsc = rdtsc();
3493 /*
3494 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3495 * on every vmexit, but if not, we might have a stale dr6 from the
3496 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3497 */
3498 set_debugreg(0, 6);
3499}
3500
3501static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3502 struct kvm_lapic_state *s)
3503{
3504 if (vcpu->arch.apicv_active)
3505 kvm_x86_ops->sync_pir_to_irr(vcpu);
3506
3507 return kvm_apic_get_state(vcpu, s);
3508}
3509
3510static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3511 struct kvm_lapic_state *s)
3512{
3513 int r;
3514
3515 r = kvm_apic_set_state(vcpu, s);
3516 if (r)
3517 return r;
3518 update_cr8_intercept(vcpu);
3519
3520 return 0;
3521}
3522
3523static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3524{
3525 return (!lapic_in_kernel(vcpu) ||
3526 kvm_apic_accept_pic_intr(vcpu));
3527}
3528
3529/*
3530 * if userspace requested an interrupt window, check that the
3531 * interrupt window is open.
3532 *
3533 * No need to exit to userspace if we already have an interrupt queued.
3534 */
3535static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3536{
3537 return kvm_arch_interrupt_allowed(vcpu) &&
3538 !kvm_cpu_has_interrupt(vcpu) &&
3539 !kvm_event_needs_reinjection(vcpu) &&
3540 kvm_cpu_accept_dm_intr(vcpu);
3541}
3542
3543static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3544 struct kvm_interrupt *irq)
3545{
3546 if (irq->irq >= KVM_NR_INTERRUPTS)
3547 return -EINVAL;
3548
3549 if (!irqchip_in_kernel(vcpu->kvm)) {
3550 kvm_queue_interrupt(vcpu, irq->irq, false);
3551 kvm_make_request(KVM_REQ_EVENT, vcpu);
3552 return 0;
3553 }
3554
3555 /*
3556 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3557 * fail for in-kernel 8259.
3558 */
3559 if (pic_in_kernel(vcpu->kvm))
3560 return -ENXIO;
3561
3562 if (vcpu->arch.pending_external_vector != -1)
3563 return -EEXIST;
3564
3565 vcpu->arch.pending_external_vector = irq->irq;
3566 kvm_make_request(KVM_REQ_EVENT, vcpu);
3567 return 0;
3568}
3569
3570static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3571{
3572 kvm_inject_nmi(vcpu);
3573
3574 return 0;
3575}
3576
3577static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3578{
3579 kvm_make_request(KVM_REQ_SMI, vcpu);
3580
3581 return 0;
3582}
3583
3584static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3585 struct kvm_tpr_access_ctl *tac)
3586{
3587 if (tac->flags)
3588 return -EINVAL;
3589 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3590 return 0;
3591}
3592
3593static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3594 u64 mcg_cap)
3595{
3596 int r;
3597 unsigned bank_num = mcg_cap & 0xff, bank;
3598
3599 r = -EINVAL;
3600 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3601 goto out;
3602 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3603 goto out;
3604 r = 0;
3605 vcpu->arch.mcg_cap = mcg_cap;
3606 /* Init IA32_MCG_CTL to all 1s */
3607 if (mcg_cap & MCG_CTL_P)
3608 vcpu->arch.mcg_ctl = ~(u64)0;
3609 /* Init IA32_MCi_CTL to all 1s */
3610 for (bank = 0; bank < bank_num; bank++)
3611 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3612
3613 kvm_x86_ops->setup_mce(vcpu);
3614out:
3615 return r;
3616}
3617
3618static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3619 struct kvm_x86_mce *mce)
3620{
3621 u64 mcg_cap = vcpu->arch.mcg_cap;
3622 unsigned bank_num = mcg_cap & 0xff;
3623 u64 *banks = vcpu->arch.mce_banks;
3624
3625 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3626 return -EINVAL;
3627 /*
3628 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3629 * reporting is disabled
3630 */
3631 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3632 vcpu->arch.mcg_ctl != ~(u64)0)
3633 return 0;
3634 banks += 4 * mce->bank;
3635 /*
3636 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3637 * reporting is disabled for the bank
3638 */
3639 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3640 return 0;
3641 if (mce->status & MCI_STATUS_UC) {
3642 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3643 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3644 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3645 return 0;
3646 }
3647 if (banks[1] & MCI_STATUS_VAL)
3648 mce->status |= MCI_STATUS_OVER;
3649 banks[2] = mce->addr;
3650 banks[3] = mce->misc;
3651 vcpu->arch.mcg_status = mce->mcg_status;
3652 banks[1] = mce->status;
3653 kvm_queue_exception(vcpu, MC_VECTOR);
3654 } else if (!(banks[1] & MCI_STATUS_VAL)
3655 || !(banks[1] & MCI_STATUS_UC)) {
3656 if (banks[1] & MCI_STATUS_VAL)
3657 mce->status |= MCI_STATUS_OVER;
3658 banks[2] = mce->addr;
3659 banks[3] = mce->misc;
3660 banks[1] = mce->status;
3661 } else
3662 banks[1] |= MCI_STATUS_OVER;
3663 return 0;
3664}
3665
3666static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3667 struct kvm_vcpu_events *events)
3668{
3669 process_nmi(vcpu);
3670
3671 /*
3672 * The API doesn't provide the instruction length for software
3673 * exceptions, so don't report them. As long as the guest RIP
3674 * isn't advanced, we should expect to encounter the exception
3675 * again.
3676 */
3677 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3678 events->exception.injected = 0;
3679 events->exception.pending = 0;
3680 } else {
3681 events->exception.injected = vcpu->arch.exception.injected;
3682 events->exception.pending = vcpu->arch.exception.pending;
3683 /*
3684 * For ABI compatibility, deliberately conflate
3685 * pending and injected exceptions when
3686 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3687 */
3688 if (!vcpu->kvm->arch.exception_payload_enabled)
3689 events->exception.injected |=
3690 vcpu->arch.exception.pending;
3691 }
3692 events->exception.nr = vcpu->arch.exception.nr;
3693 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3694 events->exception.error_code = vcpu->arch.exception.error_code;
3695 events->exception_has_payload = vcpu->arch.exception.has_payload;
3696 events->exception_payload = vcpu->arch.exception.payload;
3697
3698 events->interrupt.injected =
3699 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3700 events->interrupt.nr = vcpu->arch.interrupt.nr;
3701 events->interrupt.soft = 0;
3702 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3703
3704 events->nmi.injected = vcpu->arch.nmi_injected;
3705 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3706 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3707 events->nmi.pad = 0;
3708
3709 events->sipi_vector = 0; /* never valid when reporting to user space */
3710
3711 events->smi.smm = is_smm(vcpu);
3712 events->smi.pending = vcpu->arch.smi_pending;
3713 events->smi.smm_inside_nmi =
3714 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3715 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3716
3717 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3718 | KVM_VCPUEVENT_VALID_SHADOW
3719 | KVM_VCPUEVENT_VALID_SMM);
3720 if (vcpu->kvm->arch.exception_payload_enabled)
3721 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3722
3723 memset(&events->reserved, 0, sizeof(events->reserved));
3724}
3725
3726static void kvm_smm_changed(struct kvm_vcpu *vcpu);
3727
3728static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3729 struct kvm_vcpu_events *events)
3730{
3731 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3732 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3733 | KVM_VCPUEVENT_VALID_SHADOW
3734 | KVM_VCPUEVENT_VALID_SMM
3735 | KVM_VCPUEVENT_VALID_PAYLOAD))
3736 return -EINVAL;
3737
3738 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3739 if (!vcpu->kvm->arch.exception_payload_enabled)
3740 return -EINVAL;
3741 if (events->exception.pending)
3742 events->exception.injected = 0;
3743 else
3744 events->exception_has_payload = 0;
3745 } else {
3746 events->exception.pending = 0;
3747 events->exception_has_payload = 0;
3748 }
3749
3750 if ((events->exception.injected || events->exception.pending) &&
3751 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3752 return -EINVAL;
3753
3754 /* INITs are latched while in SMM */
3755 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3756 (events->smi.smm || events->smi.pending) &&
3757 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3758 return -EINVAL;
3759
3760 process_nmi(vcpu);
3761 vcpu->arch.exception.injected = events->exception.injected;
3762 vcpu->arch.exception.pending = events->exception.pending;
3763 vcpu->arch.exception.nr = events->exception.nr;
3764 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3765 vcpu->arch.exception.error_code = events->exception.error_code;
3766 vcpu->arch.exception.has_payload = events->exception_has_payload;
3767 vcpu->arch.exception.payload = events->exception_payload;
3768
3769 vcpu->arch.interrupt.injected = events->interrupt.injected;
3770 vcpu->arch.interrupt.nr = events->interrupt.nr;
3771 vcpu->arch.interrupt.soft = events->interrupt.soft;
3772 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3773 kvm_x86_ops->set_interrupt_shadow(vcpu,
3774 events->interrupt.shadow);
3775
3776 vcpu->arch.nmi_injected = events->nmi.injected;
3777 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3778 vcpu->arch.nmi_pending = events->nmi.pending;
3779 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3780
3781 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3782 lapic_in_kernel(vcpu))
3783 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3784
3785 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3786 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
3787 if (events->smi.smm)
3788 vcpu->arch.hflags |= HF_SMM_MASK;
3789 else
3790 vcpu->arch.hflags &= ~HF_SMM_MASK;
3791 kvm_smm_changed(vcpu);
3792 }
3793
3794 vcpu->arch.smi_pending = events->smi.pending;
3795
3796 if (events->smi.smm) {
3797 if (events->smi.smm_inside_nmi)
3798 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3799 else
3800 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3801 if (lapic_in_kernel(vcpu)) {
3802 if (events->smi.latched_init)
3803 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3804 else
3805 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3806 }
3807 }
3808 }
3809
3810 kvm_make_request(KVM_REQ_EVENT, vcpu);
3811
3812 return 0;
3813}
3814
3815static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3816 struct kvm_debugregs *dbgregs)
3817{
3818 unsigned long val;
3819
3820 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3821 kvm_get_dr(vcpu, 6, &val);
3822 dbgregs->dr6 = val;
3823 dbgregs->dr7 = vcpu->arch.dr7;
3824 dbgregs->flags = 0;
3825 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3826}
3827
3828static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3829 struct kvm_debugregs *dbgregs)
3830{
3831 if (dbgregs->flags)
3832 return -EINVAL;
3833
3834 if (dbgregs->dr6 & ~0xffffffffull)
3835 return -EINVAL;
3836 if (dbgregs->dr7 & ~0xffffffffull)
3837 return -EINVAL;
3838
3839 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3840 kvm_update_dr0123(vcpu);
3841 vcpu->arch.dr6 = dbgregs->dr6;
3842 kvm_update_dr6(vcpu);
3843 vcpu->arch.dr7 = dbgregs->dr7;
3844 kvm_update_dr7(vcpu);
3845
3846 return 0;
3847}
3848
3849#define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3850
3851static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3852{
3853 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3854 u64 xstate_bv = xsave->header.xfeatures;
3855 u64 valid;
3856
3857 /*
3858 * Copy legacy XSAVE area, to avoid complications with CPUID
3859 * leaves 0 and 1 in the loop below.
3860 */
3861 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3862
3863 /* Set XSTATE_BV */
3864 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3865 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3866
3867 /*
3868 * Copy each region from the possibly compacted offset to the
3869 * non-compacted offset.
3870 */
3871 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3872 while (valid) {
3873 u64 xfeature_mask = valid & -valid;
3874 int xfeature_nr = fls64(xfeature_mask) - 1;
3875 void *src = get_xsave_addr(xsave, xfeature_nr);
3876
3877 if (src) {
3878 u32 size, offset, ecx, edx;
3879 cpuid_count(XSTATE_CPUID, xfeature_nr,
3880 &size, &offset, &ecx, &edx);
3881 if (xfeature_nr == XFEATURE_PKRU)
3882 memcpy(dest + offset, &vcpu->arch.pkru,
3883 sizeof(vcpu->arch.pkru));
3884 else
3885 memcpy(dest + offset, src, size);
3886
3887 }
3888
3889 valid -= xfeature_mask;
3890 }
3891}
3892
3893static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3894{
3895 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3896 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3897 u64 valid;
3898
3899 /*
3900 * Copy legacy XSAVE area, to avoid complications with CPUID
3901 * leaves 0 and 1 in the loop below.
3902 */
3903 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3904
3905 /* Set XSTATE_BV and possibly XCOMP_BV. */
3906 xsave->header.xfeatures = xstate_bv;
3907 if (boot_cpu_has(X86_FEATURE_XSAVES))
3908 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3909
3910 /*
3911 * Copy each region from the non-compacted offset to the
3912 * possibly compacted offset.
3913 */
3914 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3915 while (valid) {
3916 u64 xfeature_mask = valid & -valid;
3917 int xfeature_nr = fls64(xfeature_mask) - 1;
3918 void *dest = get_xsave_addr(xsave, xfeature_nr);
3919
3920 if (dest) {
3921 u32 size, offset, ecx, edx;
3922 cpuid_count(XSTATE_CPUID, xfeature_nr,
3923 &size, &offset, &ecx, &edx);
3924 if (xfeature_nr == XFEATURE_PKRU)
3925 memcpy(&vcpu->arch.pkru, src + offset,
3926 sizeof(vcpu->arch.pkru));
3927 else
3928 memcpy(dest, src + offset, size);
3929 }
3930
3931 valid -= xfeature_mask;
3932 }
3933}
3934
3935static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3936 struct kvm_xsave *guest_xsave)
3937{
3938 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3939 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3940 fill_xsave((u8 *) guest_xsave->region, vcpu);
3941 } else {
3942 memcpy(guest_xsave->region,
3943 &vcpu->arch.guest_fpu->state.fxsave,
3944 sizeof(struct fxregs_state));
3945 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3946 XFEATURE_MASK_FPSSE;
3947 }
3948}
3949
3950#define XSAVE_MXCSR_OFFSET 24
3951
3952static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3953 struct kvm_xsave *guest_xsave)
3954{
3955 u64 xstate_bv =
3956 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3957 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3958
3959 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3960 /*
3961 * Here we allow setting states that are not present in
3962 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3963 * with old userspace.
3964 */
3965 if (xstate_bv & ~kvm_supported_xcr0() ||
3966 mxcsr & ~mxcsr_feature_mask)
3967 return -EINVAL;
3968 load_xsave(vcpu, (u8 *)guest_xsave->region);
3969 } else {
3970 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3971 mxcsr & ~mxcsr_feature_mask)
3972 return -EINVAL;
3973 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
3974 guest_xsave->region, sizeof(struct fxregs_state));
3975 }
3976 return 0;
3977}
3978
3979static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3980 struct kvm_xcrs *guest_xcrs)
3981{
3982 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3983 guest_xcrs->nr_xcrs = 0;
3984 return;
3985 }
3986
3987 guest_xcrs->nr_xcrs = 1;
3988 guest_xcrs->flags = 0;
3989 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3990 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3991}
3992
3993static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3994 struct kvm_xcrs *guest_xcrs)
3995{
3996 int i, r = 0;
3997
3998 if (!boot_cpu_has(X86_FEATURE_XSAVE))
3999 return -EINVAL;
4000
4001 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4002 return -EINVAL;
4003
4004 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4005 /* Only support XCR0 currently */
4006 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4007 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4008 guest_xcrs->xcrs[i].value);
4009 break;
4010 }
4011 if (r)
4012 r = -EINVAL;
4013 return r;
4014}
4015
4016/*
4017 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4018 * stopped by the hypervisor. This function will be called from the host only.
4019 * EINVAL is returned when the host attempts to set the flag for a guest that
4020 * does not support pv clocks.
4021 */
4022static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4023{
4024 if (!vcpu->arch.pv_time_enabled)
4025 return -EINVAL;
4026 vcpu->arch.pvclock_set_guest_stopped_request = true;
4027 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4028 return 0;
4029}
4030
4031static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4032 struct kvm_enable_cap *cap)
4033{
4034 int r;
4035 uint16_t vmcs_version;
4036 void __user *user_ptr;
4037
4038 if (cap->flags)
4039 return -EINVAL;
4040
4041 switch (cap->cap) {
4042 case KVM_CAP_HYPERV_SYNIC2:
4043 if (cap->args[0])
4044 return -EINVAL;
4045 /* fall through */
4046
4047 case KVM_CAP_HYPERV_SYNIC:
4048 if (!irqchip_in_kernel(vcpu->kvm))
4049 return -EINVAL;
4050 return kvm_hv_activate_synic(vcpu, cap->cap ==
4051 KVM_CAP_HYPERV_SYNIC2);
4052 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4053 if (!kvm_x86_ops->nested_enable_evmcs)
4054 return -ENOTTY;
4055 r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
4056 if (!r) {
4057 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4058 if (copy_to_user(user_ptr, &vmcs_version,
4059 sizeof(vmcs_version)))
4060 r = -EFAULT;
4061 }
4062 return r;
4063 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4064 if (!kvm_x86_ops->enable_direct_tlbflush)
4065 return -ENOTTY;
4066
4067 return kvm_x86_ops->enable_direct_tlbflush(vcpu);
4068
4069 default:
4070 return -EINVAL;
4071 }
4072}
4073
4074long kvm_arch_vcpu_ioctl(struct file *filp,
4075 unsigned int ioctl, unsigned long arg)
4076{
4077 struct kvm_vcpu *vcpu = filp->private_data;
4078 void __user *argp = (void __user *)arg;
4079 int r;
4080 union {
4081 struct kvm_lapic_state *lapic;
4082 struct kvm_xsave *xsave;
4083 struct kvm_xcrs *xcrs;
4084 void *buffer;
4085 } u;
4086
4087 vcpu_load(vcpu);
4088
4089 u.buffer = NULL;
4090 switch (ioctl) {
4091 case KVM_GET_LAPIC: {
4092 r = -EINVAL;
4093 if (!lapic_in_kernel(vcpu))
4094 goto out;
4095 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4096 GFP_KERNEL_ACCOUNT);
4097
4098 r = -ENOMEM;
4099 if (!u.lapic)
4100 goto out;
4101 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4102 if (r)
4103 goto out;
4104 r = -EFAULT;
4105 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4106 goto out;
4107 r = 0;
4108 break;
4109 }
4110 case KVM_SET_LAPIC: {
4111 r = -EINVAL;
4112 if (!lapic_in_kernel(vcpu))
4113 goto out;
4114 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4115 if (IS_ERR(u.lapic)) {
4116 r = PTR_ERR(u.lapic);
4117 goto out_nofree;
4118 }
4119
4120 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4121 break;
4122 }
4123 case KVM_INTERRUPT: {
4124 struct kvm_interrupt irq;
4125
4126 r = -EFAULT;
4127 if (copy_from_user(&irq, argp, sizeof(irq)))
4128 goto out;
4129 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4130 break;
4131 }
4132 case KVM_NMI: {
4133 r = kvm_vcpu_ioctl_nmi(vcpu);
4134 break;
4135 }
4136 case KVM_SMI: {
4137 r = kvm_vcpu_ioctl_smi(vcpu);
4138 break;
4139 }
4140 case KVM_SET_CPUID: {
4141 struct kvm_cpuid __user *cpuid_arg = argp;
4142 struct kvm_cpuid cpuid;
4143
4144 r = -EFAULT;
4145 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4146 goto out;
4147 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4148 break;
4149 }
4150 case KVM_SET_CPUID2: {
4151 struct kvm_cpuid2 __user *cpuid_arg = argp;
4152 struct kvm_cpuid2 cpuid;
4153
4154 r = -EFAULT;
4155 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4156 goto out;
4157 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4158 cpuid_arg->entries);
4159 break;
4160 }
4161 case KVM_GET_CPUID2: {
4162 struct kvm_cpuid2 __user *cpuid_arg = argp;
4163 struct kvm_cpuid2 cpuid;
4164
4165 r = -EFAULT;
4166 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4167 goto out;
4168 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4169 cpuid_arg->entries);
4170 if (r)
4171 goto out;
4172 r = -EFAULT;
4173 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4174 goto out;
4175 r = 0;
4176 break;
4177 }
4178 case KVM_GET_MSRS: {
4179 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4180 r = msr_io(vcpu, argp, do_get_msr, 1);
4181 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4182 break;
4183 }
4184 case KVM_SET_MSRS: {
4185 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4186 r = msr_io(vcpu, argp, do_set_msr, 0);
4187 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4188 break;
4189 }
4190 case KVM_TPR_ACCESS_REPORTING: {
4191 struct kvm_tpr_access_ctl tac;
4192
4193 r = -EFAULT;
4194 if (copy_from_user(&tac, argp, sizeof(tac)))
4195 goto out;
4196 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4197 if (r)
4198 goto out;
4199 r = -EFAULT;
4200 if (copy_to_user(argp, &tac, sizeof(tac)))
4201 goto out;
4202 r = 0;
4203 break;
4204 };
4205 case KVM_SET_VAPIC_ADDR: {
4206 struct kvm_vapic_addr va;
4207 int idx;
4208
4209 r = -EINVAL;
4210 if (!lapic_in_kernel(vcpu))
4211 goto out;
4212 r = -EFAULT;
4213 if (copy_from_user(&va, argp, sizeof(va)))
4214 goto out;
4215 idx = srcu_read_lock(&vcpu->kvm->srcu);
4216 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4217 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4218 break;
4219 }
4220 case KVM_X86_SETUP_MCE: {
4221 u64 mcg_cap;
4222
4223 r = -EFAULT;
4224 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4225 goto out;
4226 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4227 break;
4228 }
4229 case KVM_X86_SET_MCE: {
4230 struct kvm_x86_mce mce;
4231
4232 r = -EFAULT;
4233 if (copy_from_user(&mce, argp, sizeof(mce)))
4234 goto out;
4235 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4236 break;
4237 }
4238 case KVM_GET_VCPU_EVENTS: {
4239 struct kvm_vcpu_events events;
4240
4241 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4242
4243 r = -EFAULT;
4244 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4245 break;
4246 r = 0;
4247 break;
4248 }
4249 case KVM_SET_VCPU_EVENTS: {
4250 struct kvm_vcpu_events events;
4251
4252 r = -EFAULT;
4253 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4254 break;
4255
4256 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4257 break;
4258 }
4259 case KVM_GET_DEBUGREGS: {
4260 struct kvm_debugregs dbgregs;
4261
4262 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4263
4264 r = -EFAULT;
4265 if (copy_to_user(argp, &dbgregs,
4266 sizeof(struct kvm_debugregs)))
4267 break;
4268 r = 0;
4269 break;
4270 }
4271 case KVM_SET_DEBUGREGS: {
4272 struct kvm_debugregs dbgregs;
4273
4274 r = -EFAULT;
4275 if (copy_from_user(&dbgregs, argp,
4276 sizeof(struct kvm_debugregs)))
4277 break;
4278
4279 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4280 break;
4281 }
4282 case KVM_GET_XSAVE: {
4283 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4284 r = -ENOMEM;
4285 if (!u.xsave)
4286 break;
4287
4288 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4289
4290 r = -EFAULT;
4291 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4292 break;
4293 r = 0;
4294 break;
4295 }
4296 case KVM_SET_XSAVE: {
4297 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4298 if (IS_ERR(u.xsave)) {
4299 r = PTR_ERR(u.xsave);
4300 goto out_nofree;
4301 }
4302
4303 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4304 break;
4305 }
4306 case KVM_GET_XCRS: {
4307 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4308 r = -ENOMEM;
4309 if (!u.xcrs)
4310 break;
4311
4312 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4313
4314 r = -EFAULT;
4315 if (copy_to_user(argp, u.xcrs,
4316 sizeof(struct kvm_xcrs)))
4317 break;
4318 r = 0;
4319 break;
4320 }
4321 case KVM_SET_XCRS: {
4322 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4323 if (IS_ERR(u.xcrs)) {
4324 r = PTR_ERR(u.xcrs);
4325 goto out_nofree;
4326 }
4327
4328 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4329 break;
4330 }
4331 case KVM_SET_TSC_KHZ: {
4332 u32 user_tsc_khz;
4333
4334 r = -EINVAL;
4335 user_tsc_khz = (u32)arg;
4336
4337 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4338 goto out;
4339
4340 if (user_tsc_khz == 0)
4341 user_tsc_khz = tsc_khz;
4342
4343 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4344 r = 0;
4345
4346 goto out;
4347 }
4348 case KVM_GET_TSC_KHZ: {
4349 r = vcpu->arch.virtual_tsc_khz;
4350 goto out;
4351 }
4352 case KVM_KVMCLOCK_CTRL: {
4353 r = kvm_set_guest_paused(vcpu);
4354 goto out;
4355 }
4356 case KVM_ENABLE_CAP: {
4357 struct kvm_enable_cap cap;
4358
4359 r = -EFAULT;
4360 if (copy_from_user(&cap, argp, sizeof(cap)))
4361 goto out;
4362 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4363 break;
4364 }
4365 case KVM_GET_NESTED_STATE: {
4366 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4367 u32 user_data_size;
4368
4369 r = -EINVAL;
4370 if (!kvm_x86_ops->get_nested_state)
4371 break;
4372
4373 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4374 r = -EFAULT;
4375 if (get_user(user_data_size, &user_kvm_nested_state->size))
4376 break;
4377
4378 r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4379 user_data_size);
4380 if (r < 0)
4381 break;
4382
4383 if (r > user_data_size) {
4384 if (put_user(r, &user_kvm_nested_state->size))
4385 r = -EFAULT;
4386 else
4387 r = -E2BIG;
4388 break;
4389 }
4390
4391 r = 0;
4392 break;
4393 }
4394 case KVM_SET_NESTED_STATE: {
4395 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4396 struct kvm_nested_state kvm_state;
4397
4398 r = -EINVAL;
4399 if (!kvm_x86_ops->set_nested_state)
4400 break;
4401
4402 r = -EFAULT;
4403 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4404 break;
4405
4406 r = -EINVAL;
4407 if (kvm_state.size < sizeof(kvm_state))
4408 break;
4409
4410 if (kvm_state.flags &
4411 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4412 | KVM_STATE_NESTED_EVMCS))
4413 break;
4414
4415 /* nested_run_pending implies guest_mode. */
4416 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4417 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4418 break;
4419
4420 r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4421 break;
4422 }
4423 case KVM_GET_SUPPORTED_HV_CPUID: {
4424 struct kvm_cpuid2 __user *cpuid_arg = argp;
4425 struct kvm_cpuid2 cpuid;
4426
4427 r = -EFAULT;
4428 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4429 goto out;
4430
4431 r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4432 cpuid_arg->entries);
4433 if (r)
4434 goto out;
4435
4436 r = -EFAULT;
4437 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4438 goto out;
4439 r = 0;
4440 break;
4441 }
4442 default:
4443 r = -EINVAL;
4444 }
4445out:
4446 kfree(u.buffer);
4447out_nofree:
4448 vcpu_put(vcpu);
4449 return r;
4450}
4451
4452vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4453{
4454 return VM_FAULT_SIGBUS;
4455}
4456
4457static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4458{
4459 int ret;
4460
4461 if (addr > (unsigned int)(-3 * PAGE_SIZE))
4462 return -EINVAL;
4463 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4464 return ret;
4465}
4466
4467static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4468 u64 ident_addr)
4469{
4470 return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4471}
4472
4473static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4474 unsigned long kvm_nr_mmu_pages)
4475{
4476 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4477 return -EINVAL;
4478
4479 mutex_lock(&kvm->slots_lock);
4480
4481 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4482 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4483
4484 mutex_unlock(&kvm->slots_lock);
4485 return 0;
4486}
4487
4488static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4489{
4490 return kvm->arch.n_max_mmu_pages;
4491}
4492
4493static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4494{
4495 struct kvm_pic *pic = kvm->arch.vpic;
4496 int r;
4497
4498 r = 0;
4499 switch (chip->chip_id) {
4500 case KVM_IRQCHIP_PIC_MASTER:
4501 memcpy(&chip->chip.pic, &pic->pics[0],
4502 sizeof(struct kvm_pic_state));
4503 break;
4504 case KVM_IRQCHIP_PIC_SLAVE:
4505 memcpy(&chip->chip.pic, &pic->pics[1],
4506 sizeof(struct kvm_pic_state));
4507 break;
4508 case KVM_IRQCHIP_IOAPIC:
4509 kvm_get_ioapic(kvm, &chip->chip.ioapic);
4510 break;
4511 default:
4512 r = -EINVAL;
4513 break;
4514 }
4515 return r;
4516}
4517
4518static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4519{
4520 struct kvm_pic *pic = kvm->arch.vpic;
4521 int r;
4522
4523 r = 0;
4524 switch (chip->chip_id) {
4525 case KVM_IRQCHIP_PIC_MASTER:
4526 spin_lock(&pic->lock);
4527 memcpy(&pic->pics[0], &chip->chip.pic,
4528 sizeof(struct kvm_pic_state));
4529 spin_unlock(&pic->lock);
4530 break;
4531 case KVM_IRQCHIP_PIC_SLAVE:
4532 spin_lock(&pic->lock);
4533 memcpy(&pic->pics[1], &chip->chip.pic,
4534 sizeof(struct kvm_pic_state));
4535 spin_unlock(&pic->lock);
4536 break;
4537 case KVM_IRQCHIP_IOAPIC:
4538 kvm_set_ioapic(kvm, &chip->chip.ioapic);
4539 break;
4540 default:
4541 r = -EINVAL;
4542 break;
4543 }
4544 kvm_pic_update_irq(pic);
4545 return r;
4546}
4547
4548static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4549{
4550 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4551
4552 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4553
4554 mutex_lock(&kps->lock);
4555 memcpy(ps, &kps->channels, sizeof(*ps));
4556 mutex_unlock(&kps->lock);
4557 return 0;
4558}
4559
4560static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4561{
4562 int i;
4563 struct kvm_pit *pit = kvm->arch.vpit;
4564
4565 mutex_lock(&pit->pit_state.lock);
4566 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4567 for (i = 0; i < 3; i++)
4568 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4569 mutex_unlock(&pit->pit_state.lock);
4570 return 0;
4571}
4572
4573static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4574{
4575 mutex_lock(&kvm->arch.vpit->pit_state.lock);
4576 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4577 sizeof(ps->channels));
4578 ps->flags = kvm->arch.vpit->pit_state.flags;
4579 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4580 memset(&ps->reserved, 0, sizeof(ps->reserved));
4581 return 0;
4582}
4583
4584static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4585{
4586 int start = 0;
4587 int i;
4588 u32 prev_legacy, cur_legacy;
4589 struct kvm_pit *pit = kvm->arch.vpit;
4590
4591 mutex_lock(&pit->pit_state.lock);
4592 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4593 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4594 if (!prev_legacy && cur_legacy)
4595 start = 1;
4596 memcpy(&pit->pit_state.channels, &ps->channels,
4597 sizeof(pit->pit_state.channels));
4598 pit->pit_state.flags = ps->flags;
4599 for (i = 0; i < 3; i++)
4600 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4601 start && i == 0);
4602 mutex_unlock(&pit->pit_state.lock);
4603 return 0;
4604}
4605
4606static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4607 struct kvm_reinject_control *control)
4608{
4609 struct kvm_pit *pit = kvm->arch.vpit;
4610
4611 if (!pit)
4612 return -ENXIO;
4613
4614 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4615 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4616 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4617 */
4618 mutex_lock(&pit->pit_state.lock);
4619 kvm_pit_set_reinject(pit, control->pit_reinject);
4620 mutex_unlock(&pit->pit_state.lock);
4621
4622 return 0;
4623}
4624
4625/**
4626 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4627 * @kvm: kvm instance
4628 * @log: slot id and address to which we copy the log
4629 *
4630 * Steps 1-4 below provide general overview of dirty page logging. See
4631 * kvm_get_dirty_log_protect() function description for additional details.
4632 *
4633 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4634 * always flush the TLB (step 4) even if previous step failed and the dirty
4635 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4636 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4637 * writes will be marked dirty for next log read.
4638 *
4639 * 1. Take a snapshot of the bit and clear it if needed.
4640 * 2. Write protect the corresponding page.
4641 * 3. Copy the snapshot to the userspace.
4642 * 4. Flush TLB's if needed.
4643 */
4644int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4645{
4646 bool flush = false;
4647 int r;
4648
4649 mutex_lock(&kvm->slots_lock);
4650
4651 /*
4652 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4653 */
4654 if (kvm_x86_ops->flush_log_dirty)
4655 kvm_x86_ops->flush_log_dirty(kvm);
4656
4657 r = kvm_get_dirty_log_protect(kvm, log, &flush);
4658
4659 /*
4660 * All the TLBs can be flushed out of mmu lock, see the comments in
4661 * kvm_mmu_slot_remove_write_access().
4662 */
4663 lockdep_assert_held(&kvm->slots_lock);
4664 if (flush)
4665 kvm_flush_remote_tlbs(kvm);
4666
4667 mutex_unlock(&kvm->slots_lock);
4668 return r;
4669}
4670
4671int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
4672{
4673 bool flush = false;
4674 int r;
4675
4676 mutex_lock(&kvm->slots_lock);
4677
4678 /*
4679 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4680 */
4681 if (kvm_x86_ops->flush_log_dirty)
4682 kvm_x86_ops->flush_log_dirty(kvm);
4683
4684 r = kvm_clear_dirty_log_protect(kvm, log, &flush);
4685
4686 /*
4687 * All the TLBs can be flushed out of mmu lock, see the comments in
4688 * kvm_mmu_slot_remove_write_access().
4689 */
4690 lockdep_assert_held(&kvm->slots_lock);
4691 if (flush)
4692 kvm_flush_remote_tlbs(kvm);
4693
4694 mutex_unlock(&kvm->slots_lock);
4695 return r;
4696}
4697
4698int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4699 bool line_status)
4700{
4701 if (!irqchip_in_kernel(kvm))
4702 return -ENXIO;
4703
4704 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4705 irq_event->irq, irq_event->level,
4706 line_status);
4707 return 0;
4708}
4709
4710int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4711 struct kvm_enable_cap *cap)
4712{
4713 int r;
4714
4715 if (cap->flags)
4716 return -EINVAL;
4717
4718 switch (cap->cap) {
4719 case KVM_CAP_DISABLE_QUIRKS:
4720 kvm->arch.disabled_quirks = cap->args[0];
4721 r = 0;
4722 break;
4723 case KVM_CAP_SPLIT_IRQCHIP: {
4724 mutex_lock(&kvm->lock);
4725 r = -EINVAL;
4726 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4727 goto split_irqchip_unlock;
4728 r = -EEXIST;
4729 if (irqchip_in_kernel(kvm))
4730 goto split_irqchip_unlock;
4731 if (kvm->created_vcpus)
4732 goto split_irqchip_unlock;
4733 r = kvm_setup_empty_irq_routing(kvm);
4734 if (r)
4735 goto split_irqchip_unlock;
4736 /* Pairs with irqchip_in_kernel. */
4737 smp_wmb();
4738 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4739 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4740 r = 0;
4741split_irqchip_unlock:
4742 mutex_unlock(&kvm->lock);
4743 break;
4744 }
4745 case KVM_CAP_X2APIC_API:
4746 r = -EINVAL;
4747 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4748 break;
4749
4750 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4751 kvm->arch.x2apic_format = true;
4752 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4753 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4754
4755 r = 0;
4756 break;
4757 case KVM_CAP_X86_DISABLE_EXITS:
4758 r = -EINVAL;
4759 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4760 break;
4761
4762 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4763 kvm_can_mwait_in_guest())
4764 kvm->arch.mwait_in_guest = true;
4765 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4766 kvm->arch.hlt_in_guest = true;
4767 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4768 kvm->arch.pause_in_guest = true;
4769 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
4770 kvm->arch.cstate_in_guest = true;
4771 r = 0;
4772 break;
4773 case KVM_CAP_MSR_PLATFORM_INFO:
4774 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4775 r = 0;
4776 break;
4777 case KVM_CAP_EXCEPTION_PAYLOAD:
4778 kvm->arch.exception_payload_enabled = cap->args[0];
4779 r = 0;
4780 break;
4781 default:
4782 r = -EINVAL;
4783 break;
4784 }
4785 return r;
4786}
4787
4788long kvm_arch_vm_ioctl(struct file *filp,
4789 unsigned int ioctl, unsigned long arg)
4790{
4791 struct kvm *kvm = filp->private_data;
4792 void __user *argp = (void __user *)arg;
4793 int r = -ENOTTY;
4794 /*
4795 * This union makes it completely explicit to gcc-3.x
4796 * that these two variables' stack usage should be
4797 * combined, not added together.
4798 */
4799 union {
4800 struct kvm_pit_state ps;
4801 struct kvm_pit_state2 ps2;
4802 struct kvm_pit_config pit_config;
4803 } u;
4804
4805 switch (ioctl) {
4806 case KVM_SET_TSS_ADDR:
4807 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4808 break;
4809 case KVM_SET_IDENTITY_MAP_ADDR: {
4810 u64 ident_addr;
4811
4812 mutex_lock(&kvm->lock);
4813 r = -EINVAL;
4814 if (kvm->created_vcpus)
4815 goto set_identity_unlock;
4816 r = -EFAULT;
4817 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4818 goto set_identity_unlock;
4819 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4820set_identity_unlock:
4821 mutex_unlock(&kvm->lock);
4822 break;
4823 }
4824 case KVM_SET_NR_MMU_PAGES:
4825 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4826 break;
4827 case KVM_GET_NR_MMU_PAGES:
4828 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4829 break;
4830 case KVM_CREATE_IRQCHIP: {
4831 mutex_lock(&kvm->lock);
4832
4833 r = -EEXIST;
4834 if (irqchip_in_kernel(kvm))
4835 goto create_irqchip_unlock;
4836
4837 r = -EINVAL;
4838 if (kvm->created_vcpus)
4839 goto create_irqchip_unlock;
4840
4841 r = kvm_pic_init(kvm);
4842 if (r)
4843 goto create_irqchip_unlock;
4844
4845 r = kvm_ioapic_init(kvm);
4846 if (r) {
4847 kvm_pic_destroy(kvm);
4848 goto create_irqchip_unlock;
4849 }
4850
4851 r = kvm_setup_default_irq_routing(kvm);
4852 if (r) {
4853 kvm_ioapic_destroy(kvm);
4854 kvm_pic_destroy(kvm);
4855 goto create_irqchip_unlock;
4856 }
4857 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4858 smp_wmb();
4859 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4860 create_irqchip_unlock:
4861 mutex_unlock(&kvm->lock);
4862 break;
4863 }
4864 case KVM_CREATE_PIT:
4865 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4866 goto create_pit;
4867 case KVM_CREATE_PIT2:
4868 r = -EFAULT;
4869 if (copy_from_user(&u.pit_config, argp,
4870 sizeof(struct kvm_pit_config)))
4871 goto out;
4872 create_pit:
4873 mutex_lock(&kvm->lock);
4874 r = -EEXIST;
4875 if (kvm->arch.vpit)
4876 goto create_pit_unlock;
4877 r = -ENOMEM;
4878 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4879 if (kvm->arch.vpit)
4880 r = 0;
4881 create_pit_unlock:
4882 mutex_unlock(&kvm->lock);
4883 break;
4884 case KVM_GET_IRQCHIP: {
4885 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4886 struct kvm_irqchip *chip;
4887
4888 chip = memdup_user(argp, sizeof(*chip));
4889 if (IS_ERR(chip)) {
4890 r = PTR_ERR(chip);
4891 goto out;
4892 }
4893
4894 r = -ENXIO;
4895 if (!irqchip_kernel(kvm))
4896 goto get_irqchip_out;
4897 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4898 if (r)
4899 goto get_irqchip_out;
4900 r = -EFAULT;
4901 if (copy_to_user(argp, chip, sizeof(*chip)))
4902 goto get_irqchip_out;
4903 r = 0;
4904 get_irqchip_out:
4905 kfree(chip);
4906 break;
4907 }
4908 case KVM_SET_IRQCHIP: {
4909 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4910 struct kvm_irqchip *chip;
4911
4912 chip = memdup_user(argp, sizeof(*chip));
4913 if (IS_ERR(chip)) {
4914 r = PTR_ERR(chip);
4915 goto out;
4916 }
4917
4918 r = -ENXIO;
4919 if (!irqchip_kernel(kvm))
4920 goto set_irqchip_out;
4921 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4922 if (r)
4923 goto set_irqchip_out;
4924 r = 0;
4925 set_irqchip_out:
4926 kfree(chip);
4927 break;
4928 }
4929 case KVM_GET_PIT: {
4930 r = -EFAULT;
4931 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4932 goto out;
4933 r = -ENXIO;
4934 if (!kvm->arch.vpit)
4935 goto out;
4936 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4937 if (r)
4938 goto out;
4939 r = -EFAULT;
4940 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4941 goto out;
4942 r = 0;
4943 break;
4944 }
4945 case KVM_SET_PIT: {
4946 r = -EFAULT;
4947 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
4948 goto out;
4949 r = -ENXIO;
4950 if (!kvm->arch.vpit)
4951 goto out;
4952 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4953 break;
4954 }
4955 case KVM_GET_PIT2: {
4956 r = -ENXIO;
4957 if (!kvm->arch.vpit)
4958 goto out;
4959 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4960 if (r)
4961 goto out;
4962 r = -EFAULT;
4963 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4964 goto out;
4965 r = 0;
4966 break;
4967 }
4968 case KVM_SET_PIT2: {
4969 r = -EFAULT;
4970 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4971 goto out;
4972 r = -ENXIO;
4973 if (!kvm->arch.vpit)
4974 goto out;
4975 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4976 break;
4977 }
4978 case KVM_REINJECT_CONTROL: {
4979 struct kvm_reinject_control control;
4980 r = -EFAULT;
4981 if (copy_from_user(&control, argp, sizeof(control)))
4982 goto out;
4983 r = kvm_vm_ioctl_reinject(kvm, &control);
4984 break;
4985 }
4986 case KVM_SET_BOOT_CPU_ID:
4987 r = 0;
4988 mutex_lock(&kvm->lock);
4989 if (kvm->created_vcpus)
4990 r = -EBUSY;
4991 else
4992 kvm->arch.bsp_vcpu_id = arg;
4993 mutex_unlock(&kvm->lock);
4994 break;
4995 case KVM_XEN_HVM_CONFIG: {
4996 struct kvm_xen_hvm_config xhc;
4997 r = -EFAULT;
4998 if (copy_from_user(&xhc, argp, sizeof(xhc)))
4999 goto out;
5000 r = -EINVAL;
5001 if (xhc.flags)
5002 goto out;
5003 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5004 r = 0;
5005 break;
5006 }
5007 case KVM_SET_CLOCK: {
5008 struct kvm_clock_data user_ns;
5009 u64 now_ns;
5010
5011 r = -EFAULT;
5012 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5013 goto out;
5014
5015 r = -EINVAL;
5016 if (user_ns.flags)
5017 goto out;
5018
5019 r = 0;
5020 /*
5021 * TODO: userspace has to take care of races with VCPU_RUN, so
5022 * kvm_gen_update_masterclock() can be cut down to locked
5023 * pvclock_update_vm_gtod_copy().
5024 */
5025 kvm_gen_update_masterclock(kvm);
5026 now_ns = get_kvmclock_ns(kvm);
5027 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5028 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5029 break;
5030 }
5031 case KVM_GET_CLOCK: {
5032 struct kvm_clock_data user_ns;
5033 u64 now_ns;
5034
5035 now_ns = get_kvmclock_ns(kvm);
5036 user_ns.clock = now_ns;
5037 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5038 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5039
5040 r = -EFAULT;
5041 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5042 goto out;
5043 r = 0;
5044 break;
5045 }
5046 case KVM_MEMORY_ENCRYPT_OP: {
5047 r = -ENOTTY;
5048 if (kvm_x86_ops->mem_enc_op)
5049 r = kvm_x86_ops->mem_enc_op(kvm, argp);
5050 break;
5051 }
5052 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5053 struct kvm_enc_region region;
5054
5055 r = -EFAULT;
5056 if (copy_from_user(&region, argp, sizeof(region)))
5057 goto out;
5058
5059 r = -ENOTTY;
5060 if (kvm_x86_ops->mem_enc_reg_region)
5061 r = kvm_x86_ops->mem_enc_reg_region(kvm, &region);
5062 break;
5063 }
5064 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5065 struct kvm_enc_region region;
5066
5067 r = -EFAULT;
5068 if (copy_from_user(&region, argp, sizeof(region)))
5069 goto out;
5070
5071 r = -ENOTTY;
5072 if (kvm_x86_ops->mem_enc_unreg_region)
5073 r = kvm_x86_ops->mem_enc_unreg_region(kvm, &region);
5074 break;
5075 }
5076 case KVM_HYPERV_EVENTFD: {
5077 struct kvm_hyperv_eventfd hvevfd;
5078
5079 r = -EFAULT;
5080 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5081 goto out;
5082 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5083 break;
5084 }
5085 case KVM_SET_PMU_EVENT_FILTER:
5086 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5087 break;
5088 default:
5089 r = -ENOTTY;
5090 }
5091out:
5092 return r;
5093}
5094
5095static void kvm_init_msr_list(void)
5096{
5097 u32 dummy[2];
5098 unsigned i, j;
5099
5100 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5101 "Please update the fixed PMCs in msrs_to_save[]");
5102 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_GENERIC != 32,
5103 "Please update the generic perfctr/eventsel MSRs in msrs_to_save[]");
5104
5105 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
5106 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
5107 continue;
5108
5109 /*
5110 * Even MSRs that are valid in the host may not be exposed
5111 * to the guests in some cases.
5112 */
5113 switch (msrs_to_save[i]) {
5114 case MSR_IA32_BNDCFGS:
5115 if (!kvm_mpx_supported())
5116 continue;
5117 break;
5118 case MSR_TSC_AUX:
5119 if (!kvm_x86_ops->rdtscp_supported())
5120 continue;
5121 break;
5122 case MSR_IA32_RTIT_CTL:
5123 case MSR_IA32_RTIT_STATUS:
5124 if (!kvm_x86_ops->pt_supported())
5125 continue;
5126 break;
5127 case MSR_IA32_RTIT_CR3_MATCH:
5128 if (!kvm_x86_ops->pt_supported() ||
5129 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5130 continue;
5131 break;
5132 case MSR_IA32_RTIT_OUTPUT_BASE:
5133 case MSR_IA32_RTIT_OUTPUT_MASK:
5134 if (!kvm_x86_ops->pt_supported() ||
5135 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5136 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5137 continue;
5138 break;
5139 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
5140 if (!kvm_x86_ops->pt_supported() ||
5141 msrs_to_save[i] - MSR_IA32_RTIT_ADDR0_A >=
5142 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5143 continue;
5144 break;
5145 }
5146 default:
5147 break;
5148 }
5149
5150 if (j < i)
5151 msrs_to_save[j] = msrs_to_save[i];
5152 j++;
5153 }
5154 num_msrs_to_save = j;
5155
5156 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
5157 if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
5158 continue;
5159
5160 if (j < i)
5161 emulated_msrs[j] = emulated_msrs[i];
5162 j++;
5163 }
5164 num_emulated_msrs = j;
5165
5166 for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
5167 struct kvm_msr_entry msr;
5168
5169 msr.index = msr_based_features[i];
5170 if (kvm_get_msr_feature(&msr))
5171 continue;
5172
5173 if (j < i)
5174 msr_based_features[j] = msr_based_features[i];
5175 j++;
5176 }
5177 num_msr_based_features = j;
5178}
5179
5180static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5181 const void *v)
5182{
5183 int handled = 0;
5184 int n;
5185
5186 do {
5187 n = min(len, 8);
5188 if (!(lapic_in_kernel(vcpu) &&
5189 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5190 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5191 break;
5192 handled += n;
5193 addr += n;
5194 len -= n;
5195 v += n;
5196 } while (len);
5197
5198 return handled;
5199}
5200
5201static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5202{
5203 int handled = 0;
5204 int n;
5205
5206 do {
5207 n = min(len, 8);
5208 if (!(lapic_in_kernel(vcpu) &&
5209 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5210 addr, n, v))
5211 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5212 break;
5213 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5214 handled += n;
5215 addr += n;
5216 len -= n;
5217 v += n;
5218 } while (len);
5219
5220 return handled;
5221}
5222
5223static void kvm_set_segment(struct kvm_vcpu *vcpu,
5224 struct kvm_segment *var, int seg)
5225{
5226 kvm_x86_ops->set_segment(vcpu, var, seg);
5227}
5228
5229void kvm_get_segment(struct kvm_vcpu *vcpu,
5230 struct kvm_segment *var, int seg)
5231{
5232 kvm_x86_ops->get_segment(vcpu, var, seg);
5233}
5234
5235gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5236 struct x86_exception *exception)
5237{
5238 gpa_t t_gpa;
5239
5240 BUG_ON(!mmu_is_nested(vcpu));
5241
5242 /* NPT walks are always user-walks */
5243 access |= PFERR_USER_MASK;
5244 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5245
5246 return t_gpa;
5247}
5248
5249gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5250 struct x86_exception *exception)
5251{
5252 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5253 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5254}
5255
5256 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5257 struct x86_exception *exception)
5258{
5259 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5260 access |= PFERR_FETCH_MASK;
5261 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5262}
5263
5264gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5265 struct x86_exception *exception)
5266{
5267 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5268 access |= PFERR_WRITE_MASK;
5269 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5270}
5271
5272/* uses this to access any guest's mapped memory without checking CPL */
5273gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5274 struct x86_exception *exception)
5275{
5276 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5277}
5278
5279static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5280 struct kvm_vcpu *vcpu, u32 access,
5281 struct x86_exception *exception)
5282{
5283 void *data = val;
5284 int r = X86EMUL_CONTINUE;
5285
5286 while (bytes) {
5287 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5288 exception);
5289 unsigned offset = addr & (PAGE_SIZE-1);
5290 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5291 int ret;
5292
5293 if (gpa == UNMAPPED_GVA)
5294 return X86EMUL_PROPAGATE_FAULT;
5295 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5296 offset, toread);
5297 if (ret < 0) {
5298 r = X86EMUL_IO_NEEDED;
5299 goto out;
5300 }
5301
5302 bytes -= toread;
5303 data += toread;
5304 addr += toread;
5305 }
5306out:
5307 return r;
5308}
5309
5310/* used for instruction fetching */
5311static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5312 gva_t addr, void *val, unsigned int bytes,
5313 struct x86_exception *exception)
5314{
5315 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5316 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5317 unsigned offset;
5318 int ret;
5319
5320 /* Inline kvm_read_guest_virt_helper for speed. */
5321 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5322 exception);
5323 if (unlikely(gpa == UNMAPPED_GVA))
5324 return X86EMUL_PROPAGATE_FAULT;
5325
5326 offset = addr & (PAGE_SIZE-1);
5327 if (WARN_ON(offset + bytes > PAGE_SIZE))
5328 bytes = (unsigned)PAGE_SIZE - offset;
5329 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5330 offset, bytes);
5331 if (unlikely(ret < 0))
5332 return X86EMUL_IO_NEEDED;
5333
5334 return X86EMUL_CONTINUE;
5335}
5336
5337int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5338 gva_t addr, void *val, unsigned int bytes,
5339 struct x86_exception *exception)
5340{
5341 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5342
5343 /*
5344 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5345 * is returned, but our callers are not ready for that and they blindly
5346 * call kvm_inject_page_fault. Ensure that they at least do not leak
5347 * uninitialized kernel stack memory into cr2 and error code.
5348 */
5349 memset(exception, 0, sizeof(*exception));
5350 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5351 exception);
5352}
5353EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5354
5355static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5356 gva_t addr, void *val, unsigned int bytes,
5357 struct x86_exception *exception, bool system)
5358{
5359 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5360 u32 access = 0;
5361
5362 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5363 access |= PFERR_USER_MASK;
5364
5365 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5366}
5367
5368static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5369 unsigned long addr, void *val, unsigned int bytes)
5370{
5371 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5372 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5373
5374 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5375}
5376
5377static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5378 struct kvm_vcpu *vcpu, u32 access,
5379 struct x86_exception *exception)
5380{
5381 void *data = val;
5382 int r = X86EMUL_CONTINUE;
5383
5384 while (bytes) {
5385 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5386 access,
5387 exception);
5388 unsigned offset = addr & (PAGE_SIZE-1);
5389 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5390 int ret;
5391
5392 if (gpa == UNMAPPED_GVA)
5393 return X86EMUL_PROPAGATE_FAULT;
5394 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5395 if (ret < 0) {
5396 r = X86EMUL_IO_NEEDED;
5397 goto out;
5398 }
5399
5400 bytes -= towrite;
5401 data += towrite;
5402 addr += towrite;
5403 }
5404out:
5405 return r;
5406}
5407
5408static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5409 unsigned int bytes, struct x86_exception *exception,
5410 bool system)
5411{
5412 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5413 u32 access = PFERR_WRITE_MASK;
5414
5415 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5416 access |= PFERR_USER_MASK;
5417
5418 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5419 access, exception);
5420}
5421
5422int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5423 unsigned int bytes, struct x86_exception *exception)
5424{
5425 /* kvm_write_guest_virt_system can pull in tons of pages. */
5426 vcpu->arch.l1tf_flush_l1d = true;
5427
5428 /*
5429 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5430 * is returned, but our callers are not ready for that and they blindly
5431 * call kvm_inject_page_fault. Ensure that they at least do not leak
5432 * uninitialized kernel stack memory into cr2 and error code.
5433 */
5434 memset(exception, 0, sizeof(*exception));
5435 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5436 PFERR_WRITE_MASK, exception);
5437}
5438EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5439
5440int handle_ud(struct kvm_vcpu *vcpu)
5441{
5442 int emul_type = EMULTYPE_TRAP_UD;
5443 enum emulation_result er;
5444 char sig[5]; /* ud2; .ascii "kvm" */
5445 struct x86_exception e;
5446
5447 if (force_emulation_prefix &&
5448 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5449 sig, sizeof(sig), &e) == 0 &&
5450 memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
5451 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5452 emul_type = 0;
5453 }
5454
5455 er = kvm_emulate_instruction(vcpu, emul_type);
5456 if (er == EMULATE_USER_EXIT)
5457 return 0;
5458 if (er != EMULATE_DONE)
5459 kvm_queue_exception(vcpu, UD_VECTOR);
5460 return 1;
5461}
5462EXPORT_SYMBOL_GPL(handle_ud);
5463
5464static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5465 gpa_t gpa, bool write)
5466{
5467 /* For APIC access vmexit */
5468 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5469 return 1;
5470
5471 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5472 trace_vcpu_match_mmio(gva, gpa, write, true);
5473 return 1;
5474 }
5475
5476 return 0;
5477}
5478
5479static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5480 gpa_t *gpa, struct x86_exception *exception,
5481 bool write)
5482{
5483 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5484 | (write ? PFERR_WRITE_MASK : 0);
5485
5486 /*
5487 * currently PKRU is only applied to ept enabled guest so
5488 * there is no pkey in EPT page table for L1 guest or EPT
5489 * shadow page table for L2 guest.
5490 */
5491 if (vcpu_match_mmio_gva(vcpu, gva)
5492 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5493 vcpu->arch.mmio_access, 0, access)) {
5494 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5495 (gva & (PAGE_SIZE - 1));
5496 trace_vcpu_match_mmio(gva, *gpa, write, false);
5497 return 1;
5498 }
5499
5500 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5501
5502 if (*gpa == UNMAPPED_GVA)
5503 return -1;
5504
5505 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5506}
5507
5508int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5509 const void *val, int bytes)
5510{
5511 int ret;
5512
5513 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5514 if (ret < 0)
5515 return 0;
5516 kvm_page_track_write(vcpu, gpa, val, bytes);
5517 return 1;
5518}
5519
5520struct read_write_emulator_ops {
5521 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5522 int bytes);
5523 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5524 void *val, int bytes);
5525 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5526 int bytes, void *val);
5527 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5528 void *val, int bytes);
5529 bool write;
5530};
5531
5532static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5533{
5534 if (vcpu->mmio_read_completed) {
5535 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5536 vcpu->mmio_fragments[0].gpa, val);
5537 vcpu->mmio_read_completed = 0;
5538 return 1;
5539 }
5540
5541 return 0;
5542}
5543
5544static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5545 void *val, int bytes)
5546{
5547 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5548}
5549
5550static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5551 void *val, int bytes)
5552{
5553 return emulator_write_phys(vcpu, gpa, val, bytes);
5554}
5555
5556static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5557{
5558 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5559 return vcpu_mmio_write(vcpu, gpa, bytes, val);
5560}
5561
5562static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5563 void *val, int bytes)
5564{
5565 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5566 return X86EMUL_IO_NEEDED;
5567}
5568
5569static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5570 void *val, int bytes)
5571{
5572 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5573
5574 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5575 return X86EMUL_CONTINUE;
5576}
5577
5578static const struct read_write_emulator_ops read_emultor = {
5579 .read_write_prepare = read_prepare,
5580 .read_write_emulate = read_emulate,
5581 .read_write_mmio = vcpu_mmio_read,
5582 .read_write_exit_mmio = read_exit_mmio,
5583};
5584
5585static const struct read_write_emulator_ops write_emultor = {
5586 .read_write_emulate = write_emulate,
5587 .read_write_mmio = write_mmio,
5588 .read_write_exit_mmio = write_exit_mmio,
5589 .write = true,
5590};
5591
5592static int emulator_read_write_onepage(unsigned long addr, void *val,
5593 unsigned int bytes,
5594 struct x86_exception *exception,
5595 struct kvm_vcpu *vcpu,
5596 const struct read_write_emulator_ops *ops)
5597{
5598 gpa_t gpa;
5599 int handled, ret;
5600 bool write = ops->write;
5601 struct kvm_mmio_fragment *frag;
5602 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5603
5604 /*
5605 * If the exit was due to a NPF we may already have a GPA.
5606 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5607 * Note, this cannot be used on string operations since string
5608 * operation using rep will only have the initial GPA from the NPF
5609 * occurred.
5610 */
5611 if (vcpu->arch.gpa_available &&
5612 emulator_can_use_gpa(ctxt) &&
5613 (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5614 gpa = vcpu->arch.gpa_val;
5615 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5616 } else {
5617 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5618 if (ret < 0)
5619 return X86EMUL_PROPAGATE_FAULT;
5620 }
5621
5622 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5623 return X86EMUL_CONTINUE;
5624
5625 /*
5626 * Is this MMIO handled locally?
5627 */
5628 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5629 if (handled == bytes)
5630 return X86EMUL_CONTINUE;
5631
5632 gpa += handled;
5633 bytes -= handled;
5634 val += handled;
5635
5636 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5637 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5638 frag->gpa = gpa;
5639 frag->data = val;
5640 frag->len = bytes;
5641 return X86EMUL_CONTINUE;
5642}
5643
5644static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5645 unsigned long addr,
5646 void *val, unsigned int bytes,
5647 struct x86_exception *exception,
5648 const struct read_write_emulator_ops *ops)
5649{
5650 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5651 gpa_t gpa;
5652 int rc;
5653
5654 if (ops->read_write_prepare &&
5655 ops->read_write_prepare(vcpu, val, bytes))
5656 return X86EMUL_CONTINUE;
5657
5658 vcpu->mmio_nr_fragments = 0;
5659
5660 /* Crossing a page boundary? */
5661 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5662 int now;
5663
5664 now = -addr & ~PAGE_MASK;
5665 rc = emulator_read_write_onepage(addr, val, now, exception,
5666 vcpu, ops);
5667
5668 if (rc != X86EMUL_CONTINUE)
5669 return rc;
5670 addr += now;
5671 if (ctxt->mode != X86EMUL_MODE_PROT64)
5672 addr = (u32)addr;
5673 val += now;
5674 bytes -= now;
5675 }
5676
5677 rc = emulator_read_write_onepage(addr, val, bytes, exception,
5678 vcpu, ops);
5679 if (rc != X86EMUL_CONTINUE)
5680 return rc;
5681
5682 if (!vcpu->mmio_nr_fragments)
5683 return rc;
5684
5685 gpa = vcpu->mmio_fragments[0].gpa;
5686
5687 vcpu->mmio_needed = 1;
5688 vcpu->mmio_cur_fragment = 0;
5689
5690 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5691 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5692 vcpu->run->exit_reason = KVM_EXIT_MMIO;
5693 vcpu->run->mmio.phys_addr = gpa;
5694
5695 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5696}
5697
5698static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5699 unsigned long addr,
5700 void *val,
5701 unsigned int bytes,
5702 struct x86_exception *exception)
5703{
5704 return emulator_read_write(ctxt, addr, val, bytes,
5705 exception, &read_emultor);
5706}
5707
5708static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5709 unsigned long addr,
5710 const void *val,
5711 unsigned int bytes,
5712 struct x86_exception *exception)
5713{
5714 return emulator_read_write(ctxt, addr, (void *)val, bytes,
5715 exception, &write_emultor);
5716}
5717
5718#define CMPXCHG_TYPE(t, ptr, old, new) \
5719 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5720
5721#ifdef CONFIG_X86_64
5722# define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5723#else
5724# define CMPXCHG64(ptr, old, new) \
5725 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5726#endif
5727
5728static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5729 unsigned long addr,
5730 const void *old,
5731 const void *new,
5732 unsigned int bytes,
5733 struct x86_exception *exception)
5734{
5735 struct kvm_host_map map;
5736 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5737 gpa_t gpa;
5738 char *kaddr;
5739 bool exchanged;
5740
5741 /* guests cmpxchg8b have to be emulated atomically */
5742 if (bytes > 8 || (bytes & (bytes - 1)))
5743 goto emul_write;
5744
5745 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5746
5747 if (gpa == UNMAPPED_GVA ||
5748 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5749 goto emul_write;
5750
5751 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5752 goto emul_write;
5753
5754 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
5755 goto emul_write;
5756
5757 kaddr = map.hva + offset_in_page(gpa);
5758
5759 switch (bytes) {
5760 case 1:
5761 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5762 break;
5763 case 2:
5764 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5765 break;
5766 case 4:
5767 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5768 break;
5769 case 8:
5770 exchanged = CMPXCHG64(kaddr, old, new);
5771 break;
5772 default:
5773 BUG();
5774 }
5775
5776 kvm_vcpu_unmap(vcpu, &map, true);
5777
5778 if (!exchanged)
5779 return X86EMUL_CMPXCHG_FAILED;
5780
5781 kvm_page_track_write(vcpu, gpa, new, bytes);
5782
5783 return X86EMUL_CONTINUE;
5784
5785emul_write:
5786 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5787
5788 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5789}
5790
5791static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5792{
5793 int r = 0, i;
5794
5795 for (i = 0; i < vcpu->arch.pio.count; i++) {
5796 if (vcpu->arch.pio.in)
5797 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5798 vcpu->arch.pio.size, pd);
5799 else
5800 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5801 vcpu->arch.pio.port, vcpu->arch.pio.size,
5802 pd);
5803 if (r)
5804 break;
5805 pd += vcpu->arch.pio.size;
5806 }
5807 return r;
5808}
5809
5810static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5811 unsigned short port, void *val,
5812 unsigned int count, bool in)
5813{
5814 vcpu->arch.pio.port = port;
5815 vcpu->arch.pio.in = in;
5816 vcpu->arch.pio.count = count;
5817 vcpu->arch.pio.size = size;
5818
5819 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5820 vcpu->arch.pio.count = 0;
5821 return 1;
5822 }
5823
5824 vcpu->run->exit_reason = KVM_EXIT_IO;
5825 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5826 vcpu->run->io.size = size;
5827 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5828 vcpu->run->io.count = count;
5829 vcpu->run->io.port = port;
5830
5831 return 0;
5832}
5833
5834static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5835 int size, unsigned short port, void *val,
5836 unsigned int count)
5837{
5838 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5839 int ret;
5840
5841 if (vcpu->arch.pio.count)
5842 goto data_avail;
5843
5844 memset(vcpu->arch.pio_data, 0, size * count);
5845
5846 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5847 if (ret) {
5848data_avail:
5849 memcpy(val, vcpu->arch.pio_data, size * count);
5850 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5851 vcpu->arch.pio.count = 0;
5852 return 1;
5853 }
5854
5855 return 0;
5856}
5857
5858static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5859 int size, unsigned short port,
5860 const void *val, unsigned int count)
5861{
5862 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5863
5864 memcpy(vcpu->arch.pio_data, val, size * count);
5865 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5866 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5867}
5868
5869static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5870{
5871 return kvm_x86_ops->get_segment_base(vcpu, seg);
5872}
5873
5874static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5875{
5876 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5877}
5878
5879static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5880{
5881 if (!need_emulate_wbinvd(vcpu))
5882 return X86EMUL_CONTINUE;
5883
5884 if (kvm_x86_ops->has_wbinvd_exit()) {
5885 int cpu = get_cpu();
5886
5887 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5888 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5889 wbinvd_ipi, NULL, 1);
5890 put_cpu();
5891 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5892 } else
5893 wbinvd();
5894 return X86EMUL_CONTINUE;
5895}
5896
5897int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5898{
5899 kvm_emulate_wbinvd_noskip(vcpu);
5900 return kvm_skip_emulated_instruction(vcpu);
5901}
5902EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5903
5904
5905
5906static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5907{
5908 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5909}
5910
5911static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5912 unsigned long *dest)
5913{
5914 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5915}
5916
5917static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5918 unsigned long value)
5919{
5920
5921 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5922}
5923
5924static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5925{
5926 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5927}
5928
5929static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5930{
5931 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5932 unsigned long value;
5933
5934 switch (cr) {
5935 case 0:
5936 value = kvm_read_cr0(vcpu);
5937 break;
5938 case 2:
5939 value = vcpu->arch.cr2;
5940 break;
5941 case 3:
5942 value = kvm_read_cr3(vcpu);
5943 break;
5944 case 4:
5945 value = kvm_read_cr4(vcpu);
5946 break;
5947 case 8:
5948 value = kvm_get_cr8(vcpu);
5949 break;
5950 default:
5951 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5952 return 0;
5953 }
5954
5955 return value;
5956}
5957
5958static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5959{
5960 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5961 int res = 0;
5962
5963 switch (cr) {
5964 case 0:
5965 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5966 break;
5967 case 2:
5968 vcpu->arch.cr2 = val;
5969 break;
5970 case 3:
5971 res = kvm_set_cr3(vcpu, val);
5972 break;
5973 case 4:
5974 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5975 break;
5976 case 8:
5977 res = kvm_set_cr8(vcpu, val);
5978 break;
5979 default:
5980 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5981 res = -1;
5982 }
5983
5984 return res;
5985}
5986
5987static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5988{
5989 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5990}
5991
5992static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5993{
5994 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5995}
5996
5997static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5998{
5999 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
6000}
6001
6002static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6003{
6004 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
6005}
6006
6007static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6008{
6009 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
6010}
6011
6012static unsigned long emulator_get_cached_segment_base(
6013 struct x86_emulate_ctxt *ctxt, int seg)
6014{
6015 return get_segment_base(emul_to_vcpu(ctxt), seg);
6016}
6017
6018static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6019 struct desc_struct *desc, u32 *base3,
6020 int seg)
6021{
6022 struct kvm_segment var;
6023
6024 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6025 *selector = var.selector;
6026
6027 if (var.unusable) {
6028 memset(desc, 0, sizeof(*desc));
6029 if (base3)
6030 *base3 = 0;
6031 return false;
6032 }
6033
6034 if (var.g)
6035 var.limit >>= 12;
6036 set_desc_limit(desc, var.limit);
6037 set_desc_base(desc, (unsigned long)var.base);
6038#ifdef CONFIG_X86_64
6039 if (base3)
6040 *base3 = var.base >> 32;
6041#endif
6042 desc->type = var.type;
6043 desc->s = var.s;
6044 desc->dpl = var.dpl;
6045 desc->p = var.present;
6046 desc->avl = var.avl;
6047 desc->l = var.l;
6048 desc->d = var.db;
6049 desc->g = var.g;
6050
6051 return true;
6052}
6053
6054static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6055 struct desc_struct *desc, u32 base3,
6056 int seg)
6057{
6058 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6059 struct kvm_segment var;
6060
6061 var.selector = selector;
6062 var.base = get_desc_base(desc);
6063#ifdef CONFIG_X86_64
6064 var.base |= ((u64)base3) << 32;
6065#endif
6066 var.limit = get_desc_limit(desc);
6067 if (desc->g)
6068 var.limit = (var.limit << 12) | 0xfff;
6069 var.type = desc->type;
6070 var.dpl = desc->dpl;
6071 var.db = desc->d;
6072 var.s = desc->s;
6073 var.l = desc->l;
6074 var.g = desc->g;
6075 var.avl = desc->avl;
6076 var.present = desc->p;
6077 var.unusable = !var.present;
6078 var.padding = 0;
6079
6080 kvm_set_segment(vcpu, &var, seg);
6081 return;
6082}
6083
6084static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6085 u32 msr_index, u64 *pdata)
6086{
6087 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
6088}
6089
6090static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6091 u32 msr_index, u64 data)
6092{
6093 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
6094}
6095
6096static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6097{
6098 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6099
6100 return vcpu->arch.smbase;
6101}
6102
6103static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6104{
6105 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6106
6107 vcpu->arch.smbase = smbase;
6108}
6109
6110static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6111 u32 pmc)
6112{
6113 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
6114}
6115
6116static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6117 u32 pmc, u64 *pdata)
6118{
6119 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6120}
6121
6122static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6123{
6124 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6125}
6126
6127static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6128 struct x86_instruction_info *info,
6129 enum x86_intercept_stage stage)
6130{
6131 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
6132}
6133
6134static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6135 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
6136{
6137 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
6138}
6139
6140static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6141{
6142 return kvm_register_read(emul_to_vcpu(ctxt), reg);
6143}
6144
6145static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6146{
6147 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6148}
6149
6150static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6151{
6152 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
6153}
6154
6155static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6156{
6157 return emul_to_vcpu(ctxt)->arch.hflags;
6158}
6159
6160static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6161{
6162 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6163}
6164
6165static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6166 const char *smstate)
6167{
6168 return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6169}
6170
6171static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6172{
6173 kvm_smm_changed(emul_to_vcpu(ctxt));
6174}
6175
6176static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6177{
6178 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6179}
6180
6181static const struct x86_emulate_ops emulate_ops = {
6182 .read_gpr = emulator_read_gpr,
6183 .write_gpr = emulator_write_gpr,
6184 .read_std = emulator_read_std,
6185 .write_std = emulator_write_std,
6186 .read_phys = kvm_read_guest_phys_system,
6187 .fetch = kvm_fetch_guest_virt,
6188 .read_emulated = emulator_read_emulated,
6189 .write_emulated = emulator_write_emulated,
6190 .cmpxchg_emulated = emulator_cmpxchg_emulated,
6191 .invlpg = emulator_invlpg,
6192 .pio_in_emulated = emulator_pio_in_emulated,
6193 .pio_out_emulated = emulator_pio_out_emulated,
6194 .get_segment = emulator_get_segment,
6195 .set_segment = emulator_set_segment,
6196 .get_cached_segment_base = emulator_get_cached_segment_base,
6197 .get_gdt = emulator_get_gdt,
6198 .get_idt = emulator_get_idt,
6199 .set_gdt = emulator_set_gdt,
6200 .set_idt = emulator_set_idt,
6201 .get_cr = emulator_get_cr,
6202 .set_cr = emulator_set_cr,
6203 .cpl = emulator_get_cpl,
6204 .get_dr = emulator_get_dr,
6205 .set_dr = emulator_set_dr,
6206 .get_smbase = emulator_get_smbase,
6207 .set_smbase = emulator_set_smbase,
6208 .set_msr = emulator_set_msr,
6209 .get_msr = emulator_get_msr,
6210 .check_pmc = emulator_check_pmc,
6211 .read_pmc = emulator_read_pmc,
6212 .halt = emulator_halt,
6213 .wbinvd = emulator_wbinvd,
6214 .fix_hypercall = emulator_fix_hypercall,
6215 .intercept = emulator_intercept,
6216 .get_cpuid = emulator_get_cpuid,
6217 .set_nmi_mask = emulator_set_nmi_mask,
6218 .get_hflags = emulator_get_hflags,
6219 .set_hflags = emulator_set_hflags,
6220 .pre_leave_smm = emulator_pre_leave_smm,
6221 .post_leave_smm = emulator_post_leave_smm,
6222 .set_xcr = emulator_set_xcr,
6223};
6224
6225static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6226{
6227 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
6228 /*
6229 * an sti; sti; sequence only disable interrupts for the first
6230 * instruction. So, if the last instruction, be it emulated or
6231 * not, left the system with the INT_STI flag enabled, it
6232 * means that the last instruction is an sti. We should not
6233 * leave the flag on in this case. The same goes for mov ss
6234 */
6235 if (int_shadow & mask)
6236 mask = 0;
6237 if (unlikely(int_shadow || mask)) {
6238 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6239 if (!mask)
6240 kvm_make_request(KVM_REQ_EVENT, vcpu);
6241 }
6242}
6243
6244static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6245{
6246 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6247 if (ctxt->exception.vector == PF_VECTOR)
6248 return kvm_propagate_fault(vcpu, &ctxt->exception);
6249
6250 if (ctxt->exception.error_code_valid)
6251 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6252 ctxt->exception.error_code);
6253 else
6254 kvm_queue_exception(vcpu, ctxt->exception.vector);
6255 return false;
6256}
6257
6258static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6259{
6260 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6261 int cs_db, cs_l;
6262
6263 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6264
6265 ctxt->eflags = kvm_get_rflags(vcpu);
6266 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6267
6268 ctxt->eip = kvm_rip_read(vcpu);
6269 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
6270 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
6271 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
6272 cs_db ? X86EMUL_MODE_PROT32 :
6273 X86EMUL_MODE_PROT16;
6274 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6275 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6276 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6277
6278 init_decode_cache(ctxt);
6279 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6280}
6281
6282int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6283{
6284 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6285 int ret;
6286
6287 init_emulate_ctxt(vcpu);
6288
6289 ctxt->op_bytes = 2;
6290 ctxt->ad_bytes = 2;
6291 ctxt->_eip = ctxt->eip + inc_eip;
6292 ret = emulate_int_real(ctxt, irq);
6293
6294 if (ret != X86EMUL_CONTINUE)
6295 return EMULATE_FAIL;
6296
6297 ctxt->eip = ctxt->_eip;
6298 kvm_rip_write(vcpu, ctxt->eip);
6299 kvm_set_rflags(vcpu, ctxt->eflags);
6300
6301 return EMULATE_DONE;
6302}
6303EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6304
6305static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6306{
6307 ++vcpu->stat.insn_emulation_fail;
6308 trace_kvm_emulate_insn_failed(vcpu);
6309
6310 if (emulation_type & EMULTYPE_NO_UD_ON_FAIL)
6311 return EMULATE_FAIL;
6312
6313 kvm_queue_exception(vcpu, UD_VECTOR);
6314
6315 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6316 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6317 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6318 vcpu->run->internal.ndata = 0;
6319 return EMULATE_USER_EXIT;
6320 }
6321
6322 return EMULATE_DONE;
6323}
6324
6325static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
6326 bool write_fault_to_shadow_pgtable,
6327 int emulation_type)
6328{
6329 gpa_t gpa = cr2;
6330 kvm_pfn_t pfn;
6331
6332 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6333 return false;
6334
6335 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6336 return false;
6337
6338 if (!vcpu->arch.mmu->direct_map) {
6339 /*
6340 * Write permission should be allowed since only
6341 * write access need to be emulated.
6342 */
6343 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6344
6345 /*
6346 * If the mapping is invalid in guest, let cpu retry
6347 * it to generate fault.
6348 */
6349 if (gpa == UNMAPPED_GVA)
6350 return true;
6351 }
6352
6353 /*
6354 * Do not retry the unhandleable instruction if it faults on the
6355 * readonly host memory, otherwise it will goto a infinite loop:
6356 * retry instruction -> write #PF -> emulation fail -> retry
6357 * instruction -> ...
6358 */
6359 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6360
6361 /*
6362 * If the instruction failed on the error pfn, it can not be fixed,
6363 * report the error to userspace.
6364 */
6365 if (is_error_noslot_pfn(pfn))
6366 return false;
6367
6368 kvm_release_pfn_clean(pfn);
6369
6370 /* The instructions are well-emulated on direct mmu. */
6371 if (vcpu->arch.mmu->direct_map) {
6372 unsigned int indirect_shadow_pages;
6373
6374 spin_lock(&vcpu->kvm->mmu_lock);
6375 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6376 spin_unlock(&vcpu->kvm->mmu_lock);
6377
6378 if (indirect_shadow_pages)
6379 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6380
6381 return true;
6382 }
6383
6384 /*
6385 * if emulation was due to access to shadowed page table
6386 * and it failed try to unshadow page and re-enter the
6387 * guest to let CPU execute the instruction.
6388 */
6389 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6390
6391 /*
6392 * If the access faults on its page table, it can not
6393 * be fixed by unprotecting shadow page and it should
6394 * be reported to userspace.
6395 */
6396 return !write_fault_to_shadow_pgtable;
6397}
6398
6399static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6400 unsigned long cr2, int emulation_type)
6401{
6402 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6403 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
6404
6405 last_retry_eip = vcpu->arch.last_retry_eip;
6406 last_retry_addr = vcpu->arch.last_retry_addr;
6407
6408 /*
6409 * If the emulation is caused by #PF and it is non-page_table
6410 * writing instruction, it means the VM-EXIT is caused by shadow
6411 * page protected, we can zap the shadow page and retry this
6412 * instruction directly.
6413 *
6414 * Note: if the guest uses a non-page-table modifying instruction
6415 * on the PDE that points to the instruction, then we will unmap
6416 * the instruction and go to an infinite loop. So, we cache the
6417 * last retried eip and the last fault address, if we meet the eip
6418 * and the address again, we can break out of the potential infinite
6419 * loop.
6420 */
6421 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6422
6423 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6424 return false;
6425
6426 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6427 return false;
6428
6429 if (x86_page_table_writing_insn(ctxt))
6430 return false;
6431
6432 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
6433 return false;
6434
6435 vcpu->arch.last_retry_eip = ctxt->eip;
6436 vcpu->arch.last_retry_addr = cr2;
6437
6438 if (!vcpu->arch.mmu->direct_map)
6439 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6440
6441 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6442
6443 return true;
6444}
6445
6446static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6447static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6448
6449static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6450{
6451 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6452 /* This is a good place to trace that we are exiting SMM. */
6453 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6454
6455 /* Process a latched INIT or SMI, if any. */
6456 kvm_make_request(KVM_REQ_EVENT, vcpu);
6457 }
6458
6459 kvm_mmu_reset_context(vcpu);
6460}
6461
6462static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6463 unsigned long *db)
6464{
6465 u32 dr6 = 0;
6466 int i;
6467 u32 enable, rwlen;
6468
6469 enable = dr7;
6470 rwlen = dr7 >> 16;
6471 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6472 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6473 dr6 |= (1 << i);
6474 return dr6;
6475}
6476
6477static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
6478{
6479 struct kvm_run *kvm_run = vcpu->run;
6480
6481 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6482 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6483 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6484 kvm_run->debug.arch.exception = DB_VECTOR;
6485 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6486 return EMULATE_USER_EXIT;
6487 }
6488 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6489 return EMULATE_DONE;
6490}
6491
6492int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6493{
6494 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6495 int r;
6496
6497 r = kvm_x86_ops->skip_emulated_instruction(vcpu);
6498 if (unlikely(r != EMULATE_DONE))
6499 return 0;
6500
6501 /*
6502 * rflags is the old, "raw" value of the flags. The new value has
6503 * not been saved yet.
6504 *
6505 * This is correct even for TF set by the guest, because "the
6506 * processor will not generate this exception after the instruction
6507 * that sets the TF flag".
6508 */
6509 if (unlikely(rflags & X86_EFLAGS_TF))
6510 r = kvm_vcpu_do_singlestep(vcpu);
6511 return r == EMULATE_DONE;
6512}
6513EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6514
6515static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6516{
6517 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6518 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6519 struct kvm_run *kvm_run = vcpu->run;
6520 unsigned long eip = kvm_get_linear_rip(vcpu);
6521 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6522 vcpu->arch.guest_debug_dr7,
6523 vcpu->arch.eff_db);
6524
6525 if (dr6 != 0) {
6526 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6527 kvm_run->debug.arch.pc = eip;
6528 kvm_run->debug.arch.exception = DB_VECTOR;
6529 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6530 *r = EMULATE_USER_EXIT;
6531 return true;
6532 }
6533 }
6534
6535 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6536 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6537 unsigned long eip = kvm_get_linear_rip(vcpu);
6538 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6539 vcpu->arch.dr7,
6540 vcpu->arch.db);
6541
6542 if (dr6 != 0) {
6543 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
6544 vcpu->arch.dr6 |= dr6 | DR6_RTM;
6545 kvm_queue_exception(vcpu, DB_VECTOR);
6546 *r = EMULATE_DONE;
6547 return true;
6548 }
6549 }
6550
6551 return false;
6552}
6553
6554static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6555{
6556 switch (ctxt->opcode_len) {
6557 case 1:
6558 switch (ctxt->b) {
6559 case 0xe4: /* IN */
6560 case 0xe5:
6561 case 0xec:
6562 case 0xed:
6563 case 0xe6: /* OUT */
6564 case 0xe7:
6565 case 0xee:
6566 case 0xef:
6567 case 0x6c: /* INS */
6568 case 0x6d:
6569 case 0x6e: /* OUTS */
6570 case 0x6f:
6571 return true;
6572 }
6573 break;
6574 case 2:
6575 switch (ctxt->b) {
6576 case 0x33: /* RDPMC */
6577 return true;
6578 }
6579 break;
6580 }
6581
6582 return false;
6583}
6584
6585int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6586 unsigned long cr2,
6587 int emulation_type,
6588 void *insn,
6589 int insn_len)
6590{
6591 int r;
6592 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6593 bool writeback = true;
6594 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6595
6596 vcpu->arch.l1tf_flush_l1d = true;
6597
6598 /*
6599 * Clear write_fault_to_shadow_pgtable here to ensure it is
6600 * never reused.
6601 */
6602 vcpu->arch.write_fault_to_shadow_pgtable = false;
6603 kvm_clear_exception_queue(vcpu);
6604
6605 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6606 init_emulate_ctxt(vcpu);
6607
6608 /*
6609 * We will reenter on the same instruction since
6610 * we do not set complete_userspace_io. This does not
6611 * handle watchpoints yet, those would be handled in
6612 * the emulate_ops.
6613 */
6614 if (!(emulation_type & EMULTYPE_SKIP) &&
6615 kvm_vcpu_check_breakpoint(vcpu, &r))
6616 return r;
6617
6618 ctxt->interruptibility = 0;
6619 ctxt->have_exception = false;
6620 ctxt->exception.vector = -1;
6621 ctxt->perm_ok = false;
6622
6623 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6624
6625 r = x86_decode_insn(ctxt, insn, insn_len);
6626
6627 trace_kvm_emulate_insn_start(vcpu);
6628 ++vcpu->stat.insn_emulation;
6629 if (r != EMULATION_OK) {
6630 if (emulation_type & EMULTYPE_TRAP_UD)
6631 return EMULATE_FAIL;
6632 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6633 emulation_type))
6634 return EMULATE_DONE;
6635 if (ctxt->have_exception) {
6636 /*
6637 * #UD should result in just EMULATION_FAILED, and trap-like
6638 * exception should not be encountered during decode.
6639 */
6640 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
6641 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
6642 inject_emulated_exception(vcpu);
6643 return EMULATE_DONE;
6644 }
6645 if (emulation_type & EMULTYPE_SKIP)
6646 return EMULATE_FAIL;
6647 return handle_emulation_failure(vcpu, emulation_type);
6648 }
6649 }
6650
6651 if ((emulation_type & EMULTYPE_VMWARE) &&
6652 !is_vmware_backdoor_opcode(ctxt))
6653 return EMULATE_FAIL;
6654
6655 if (emulation_type & EMULTYPE_SKIP) {
6656 kvm_rip_write(vcpu, ctxt->_eip);
6657 if (ctxt->eflags & X86_EFLAGS_RF)
6658 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6659 kvm_x86_ops->set_interrupt_shadow(vcpu, 0);
6660 return EMULATE_DONE;
6661 }
6662
6663 if (retry_instruction(ctxt, cr2, emulation_type))
6664 return EMULATE_DONE;
6665
6666 /* this is needed for vmware backdoor interface to work since it
6667 changes registers values during IO operation */
6668 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6669 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6670 emulator_invalidate_register_cache(ctxt);
6671 }
6672
6673restart:
6674 /* Save the faulting GPA (cr2) in the address field */
6675 ctxt->exception.address = cr2;
6676
6677 r = x86_emulate_insn(ctxt);
6678
6679 if (r == EMULATION_INTERCEPTED)
6680 return EMULATE_DONE;
6681
6682 if (r == EMULATION_FAILED) {
6683 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6684 emulation_type))
6685 return EMULATE_DONE;
6686
6687 return handle_emulation_failure(vcpu, emulation_type);
6688 }
6689
6690 if (ctxt->have_exception) {
6691 r = EMULATE_DONE;
6692 if (inject_emulated_exception(vcpu))
6693 return r;
6694 } else if (vcpu->arch.pio.count) {
6695 if (!vcpu->arch.pio.in) {
6696 /* FIXME: return into emulator if single-stepping. */
6697 vcpu->arch.pio.count = 0;
6698 } else {
6699 writeback = false;
6700 vcpu->arch.complete_userspace_io = complete_emulated_pio;
6701 }
6702 r = EMULATE_USER_EXIT;
6703 } else if (vcpu->mmio_needed) {
6704 ++vcpu->stat.mmio_exits;
6705
6706 if (!vcpu->mmio_is_write)
6707 writeback = false;
6708 r = EMULATE_USER_EXIT;
6709 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6710 } else if (r == EMULATION_RESTART)
6711 goto restart;
6712 else
6713 r = EMULATE_DONE;
6714
6715 if (writeback) {
6716 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6717 toggle_interruptibility(vcpu, ctxt->interruptibility);
6718 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6719 if (!ctxt->have_exception ||
6720 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
6721 kvm_rip_write(vcpu, ctxt->eip);
6722 if (r == EMULATE_DONE && ctxt->tf)
6723 r = kvm_vcpu_do_singlestep(vcpu);
6724 __kvm_set_rflags(vcpu, ctxt->eflags);
6725 }
6726
6727 /*
6728 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6729 * do nothing, and it will be requested again as soon as
6730 * the shadow expires. But we still need to check here,
6731 * because POPF has no interrupt shadow.
6732 */
6733 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6734 kvm_make_request(KVM_REQ_EVENT, vcpu);
6735 } else
6736 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6737
6738 return r;
6739}
6740
6741int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6742{
6743 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6744}
6745EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6746
6747int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6748 void *insn, int insn_len)
6749{
6750 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6751}
6752EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6753
6754static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
6755{
6756 vcpu->arch.pio.count = 0;
6757 return 1;
6758}
6759
6760static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
6761{
6762 vcpu->arch.pio.count = 0;
6763
6764 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
6765 return 1;
6766
6767 return kvm_skip_emulated_instruction(vcpu);
6768}
6769
6770static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6771 unsigned short port)
6772{
6773 unsigned long val = kvm_rax_read(vcpu);
6774 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6775 size, port, &val, 1);
6776 if (ret)
6777 return ret;
6778
6779 /*
6780 * Workaround userspace that relies on old KVM behavior of %rip being
6781 * incremented prior to exiting to userspace to handle "OUT 0x7e".
6782 */
6783 if (port == 0x7e &&
6784 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
6785 vcpu->arch.complete_userspace_io =
6786 complete_fast_pio_out_port_0x7e;
6787 kvm_skip_emulated_instruction(vcpu);
6788 } else {
6789 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6790 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
6791 }
6792 return 0;
6793}
6794
6795static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6796{
6797 unsigned long val;
6798
6799 /* We should only ever be called with arch.pio.count equal to 1 */
6800 BUG_ON(vcpu->arch.pio.count != 1);
6801
6802 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
6803 vcpu->arch.pio.count = 0;
6804 return 1;
6805 }
6806
6807 /* For size less than 4 we merge, else we zero extend */
6808 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
6809
6810 /*
6811 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6812 * the copy and tracing
6813 */
6814 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6815 vcpu->arch.pio.port, &val, 1);
6816 kvm_rax_write(vcpu, val);
6817
6818 return kvm_skip_emulated_instruction(vcpu);
6819}
6820
6821static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6822 unsigned short port)
6823{
6824 unsigned long val;
6825 int ret;
6826
6827 /* For size less than 4 we merge, else we zero extend */
6828 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
6829
6830 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6831 &val, 1);
6832 if (ret) {
6833 kvm_rax_write(vcpu, val);
6834 return ret;
6835 }
6836
6837 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6838 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6839
6840 return 0;
6841}
6842
6843int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6844{
6845 int ret;
6846
6847 if (in)
6848 ret = kvm_fast_pio_in(vcpu, size, port);
6849 else
6850 ret = kvm_fast_pio_out(vcpu, size, port);
6851 return ret && kvm_skip_emulated_instruction(vcpu);
6852}
6853EXPORT_SYMBOL_GPL(kvm_fast_pio);
6854
6855static int kvmclock_cpu_down_prep(unsigned int cpu)
6856{
6857 __this_cpu_write(cpu_tsc_khz, 0);
6858 return 0;
6859}
6860
6861static void tsc_khz_changed(void *data)
6862{
6863 struct cpufreq_freqs *freq = data;
6864 unsigned long khz = 0;
6865
6866 if (data)
6867 khz = freq->new;
6868 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6869 khz = cpufreq_quick_get(raw_smp_processor_id());
6870 if (!khz)
6871 khz = tsc_khz;
6872 __this_cpu_write(cpu_tsc_khz, khz);
6873}
6874
6875#ifdef CONFIG_X86_64
6876static void kvm_hyperv_tsc_notifier(void)
6877{
6878 struct kvm *kvm;
6879 struct kvm_vcpu *vcpu;
6880 int cpu;
6881
6882 mutex_lock(&kvm_lock);
6883 list_for_each_entry(kvm, &vm_list, vm_list)
6884 kvm_make_mclock_inprogress_request(kvm);
6885
6886 hyperv_stop_tsc_emulation();
6887
6888 /* TSC frequency always matches when on Hyper-V */
6889 for_each_present_cpu(cpu)
6890 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6891 kvm_max_guest_tsc_khz = tsc_khz;
6892
6893 list_for_each_entry(kvm, &vm_list, vm_list) {
6894 struct kvm_arch *ka = &kvm->arch;
6895
6896 spin_lock(&ka->pvclock_gtod_sync_lock);
6897
6898 pvclock_update_vm_gtod_copy(kvm);
6899
6900 kvm_for_each_vcpu(cpu, vcpu, kvm)
6901 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6902
6903 kvm_for_each_vcpu(cpu, vcpu, kvm)
6904 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6905
6906 spin_unlock(&ka->pvclock_gtod_sync_lock);
6907 }
6908 mutex_unlock(&kvm_lock);
6909}
6910#endif
6911
6912static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
6913{
6914 struct kvm *kvm;
6915 struct kvm_vcpu *vcpu;
6916 int i, send_ipi = 0;
6917
6918 /*
6919 * We allow guests to temporarily run on slowing clocks,
6920 * provided we notify them after, or to run on accelerating
6921 * clocks, provided we notify them before. Thus time never
6922 * goes backwards.
6923 *
6924 * However, we have a problem. We can't atomically update
6925 * the frequency of a given CPU from this function; it is
6926 * merely a notifier, which can be called from any CPU.
6927 * Changing the TSC frequency at arbitrary points in time
6928 * requires a recomputation of local variables related to
6929 * the TSC for each VCPU. We must flag these local variables
6930 * to be updated and be sure the update takes place with the
6931 * new frequency before any guests proceed.
6932 *
6933 * Unfortunately, the combination of hotplug CPU and frequency
6934 * change creates an intractable locking scenario; the order
6935 * of when these callouts happen is undefined with respect to
6936 * CPU hotplug, and they can race with each other. As such,
6937 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6938 * undefined; you can actually have a CPU frequency change take
6939 * place in between the computation of X and the setting of the
6940 * variable. To protect against this problem, all updates of
6941 * the per_cpu tsc_khz variable are done in an interrupt
6942 * protected IPI, and all callers wishing to update the value
6943 * must wait for a synchronous IPI to complete (which is trivial
6944 * if the caller is on the CPU already). This establishes the
6945 * necessary total order on variable updates.
6946 *
6947 * Note that because a guest time update may take place
6948 * anytime after the setting of the VCPU's request bit, the
6949 * correct TSC value must be set before the request. However,
6950 * to ensure the update actually makes it to any guest which
6951 * starts running in hardware virtualization between the set
6952 * and the acquisition of the spinlock, we must also ping the
6953 * CPU after setting the request bit.
6954 *
6955 */
6956
6957 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
6958
6959 mutex_lock(&kvm_lock);
6960 list_for_each_entry(kvm, &vm_list, vm_list) {
6961 kvm_for_each_vcpu(i, vcpu, kvm) {
6962 if (vcpu->cpu != cpu)
6963 continue;
6964 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6965 if (vcpu->cpu != raw_smp_processor_id())
6966 send_ipi = 1;
6967 }
6968 }
6969 mutex_unlock(&kvm_lock);
6970
6971 if (freq->old < freq->new && send_ipi) {
6972 /*
6973 * We upscale the frequency. Must make the guest
6974 * doesn't see old kvmclock values while running with
6975 * the new frequency, otherwise we risk the guest sees
6976 * time go backwards.
6977 *
6978 * In case we update the frequency for another cpu
6979 * (which might be in guest context) send an interrupt
6980 * to kick the cpu out of guest context. Next time
6981 * guest context is entered kvmclock will be updated,
6982 * so the guest will not see stale values.
6983 */
6984 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
6985 }
6986}
6987
6988static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
6989 void *data)
6990{
6991 struct cpufreq_freqs *freq = data;
6992 int cpu;
6993
6994 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
6995 return 0;
6996 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
6997 return 0;
6998
6999 for_each_cpu(cpu, freq->policy->cpus)
7000 __kvmclock_cpufreq_notifier(freq, cpu);
7001
7002 return 0;
7003}
7004
7005static struct notifier_block kvmclock_cpufreq_notifier_block = {
7006 .notifier_call = kvmclock_cpufreq_notifier
7007};
7008
7009static int kvmclock_cpu_online(unsigned int cpu)
7010{
7011 tsc_khz_changed(NULL);
7012 return 0;
7013}
7014
7015static void kvm_timer_init(void)
7016{
7017 max_tsc_khz = tsc_khz;
7018
7019 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7020#ifdef CONFIG_CPU_FREQ
7021 struct cpufreq_policy policy;
7022 int cpu;
7023
7024 memset(&policy, 0, sizeof(policy));
7025 cpu = get_cpu();
7026 cpufreq_get_policy(&policy, cpu);
7027 if (policy.cpuinfo.max_freq)
7028 max_tsc_khz = policy.cpuinfo.max_freq;
7029 put_cpu();
7030#endif
7031 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7032 CPUFREQ_TRANSITION_NOTIFIER);
7033 }
7034
7035 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7036 kvmclock_cpu_online, kvmclock_cpu_down_prep);
7037}
7038
7039DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7040EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7041
7042int kvm_is_in_guest(void)
7043{
7044 return __this_cpu_read(current_vcpu) != NULL;
7045}
7046
7047static int kvm_is_user_mode(void)
7048{
7049 int user_mode = 3;
7050
7051 if (__this_cpu_read(current_vcpu))
7052 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
7053
7054 return user_mode != 0;
7055}
7056
7057static unsigned long kvm_get_guest_ip(void)
7058{
7059 unsigned long ip = 0;
7060
7061 if (__this_cpu_read(current_vcpu))
7062 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7063
7064 return ip;
7065}
7066
7067static void kvm_handle_intel_pt_intr(void)
7068{
7069 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7070
7071 kvm_make_request(KVM_REQ_PMI, vcpu);
7072 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7073 (unsigned long *)&vcpu->arch.pmu.global_status);
7074}
7075
7076static struct perf_guest_info_callbacks kvm_guest_cbs = {
7077 .is_in_guest = kvm_is_in_guest,
7078 .is_user_mode = kvm_is_user_mode,
7079 .get_guest_ip = kvm_get_guest_ip,
7080 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
7081};
7082
7083#ifdef CONFIG_X86_64
7084static void pvclock_gtod_update_fn(struct work_struct *work)
7085{
7086 struct kvm *kvm;
7087
7088 struct kvm_vcpu *vcpu;
7089 int i;
7090
7091 mutex_lock(&kvm_lock);
7092 list_for_each_entry(kvm, &vm_list, vm_list)
7093 kvm_for_each_vcpu(i, vcpu, kvm)
7094 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7095 atomic_set(&kvm_guest_has_master_clock, 0);
7096 mutex_unlock(&kvm_lock);
7097}
7098
7099static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7100
7101/*
7102 * Notification about pvclock gtod data update.
7103 */
7104static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7105 void *priv)
7106{
7107 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7108 struct timekeeper *tk = priv;
7109
7110 update_pvclock_gtod(tk);
7111
7112 /* disable master clock if host does not trust, or does not
7113 * use, TSC based clocksource.
7114 */
7115 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7116 atomic_read(&kvm_guest_has_master_clock) != 0)
7117 queue_work(system_long_wq, &pvclock_gtod_work);
7118
7119 return 0;
7120}
7121
7122static struct notifier_block pvclock_gtod_notifier = {
7123 .notifier_call = pvclock_gtod_notify,
7124};
7125#endif
7126
7127int kvm_arch_init(void *opaque)
7128{
7129 int r;
7130 struct kvm_x86_ops *ops = opaque;
7131
7132 if (kvm_x86_ops) {
7133 printk(KERN_ERR "kvm: already loaded the other module\n");
7134 r = -EEXIST;
7135 goto out;
7136 }
7137
7138 if (!ops->cpu_has_kvm_support()) {
7139 printk(KERN_ERR "kvm: no hardware support\n");
7140 r = -EOPNOTSUPP;
7141 goto out;
7142 }
7143 if (ops->disabled_by_bios()) {
7144 printk(KERN_ERR "kvm: disabled by bios\n");
7145 r = -EOPNOTSUPP;
7146 goto out;
7147 }
7148
7149 /*
7150 * KVM explicitly assumes that the guest has an FPU and
7151 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7152 * vCPU's FPU state as a fxregs_state struct.
7153 */
7154 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7155 printk(KERN_ERR "kvm: inadequate fpu\n");
7156 r = -EOPNOTSUPP;
7157 goto out;
7158 }
7159
7160 r = -ENOMEM;
7161 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7162 __alignof__(struct fpu), SLAB_ACCOUNT,
7163 NULL);
7164 if (!x86_fpu_cache) {
7165 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7166 goto out;
7167 }
7168
7169 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
7170 if (!shared_msrs) {
7171 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
7172 goto out_free_x86_fpu_cache;
7173 }
7174
7175 r = kvm_mmu_module_init();
7176 if (r)
7177 goto out_free_percpu;
7178
7179 kvm_x86_ops = ops;
7180
7181 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7182 PT_DIRTY_MASK, PT64_NX_MASK, 0,
7183 PT_PRESENT_MASK, 0, sme_me_mask);
7184 kvm_timer_init();
7185
7186 perf_register_guest_info_callbacks(&kvm_guest_cbs);
7187
7188 if (boot_cpu_has(X86_FEATURE_XSAVE))
7189 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7190
7191 kvm_lapic_init();
7192 if (pi_inject_timer == -1)
7193 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7194#ifdef CONFIG_X86_64
7195 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7196
7197 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7198 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7199#endif
7200
7201 return 0;
7202
7203out_free_percpu:
7204 free_percpu(shared_msrs);
7205out_free_x86_fpu_cache:
7206 kmem_cache_destroy(x86_fpu_cache);
7207out:
7208 return r;
7209}
7210
7211void kvm_arch_exit(void)
7212{
7213#ifdef CONFIG_X86_64
7214 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7215 clear_hv_tscchange_cb();
7216#endif
7217 kvm_lapic_exit();
7218 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7219
7220 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7221 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7222 CPUFREQ_TRANSITION_NOTIFIER);
7223 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7224#ifdef CONFIG_X86_64
7225 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7226#endif
7227 kvm_x86_ops = NULL;
7228 kvm_mmu_module_exit();
7229 free_percpu(shared_msrs);
7230 kmem_cache_destroy(x86_fpu_cache);
7231}
7232
7233int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7234{
7235 ++vcpu->stat.halt_exits;
7236 if (lapic_in_kernel(vcpu)) {
7237 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7238 return 1;
7239 } else {
7240 vcpu->run->exit_reason = KVM_EXIT_HLT;
7241 return 0;
7242 }
7243}
7244EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7245
7246int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7247{
7248 int ret = kvm_skip_emulated_instruction(vcpu);
7249 /*
7250 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7251 * KVM_EXIT_DEBUG here.
7252 */
7253 return kvm_vcpu_halt(vcpu) && ret;
7254}
7255EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7256
7257#ifdef CONFIG_X86_64
7258static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7259 unsigned long clock_type)
7260{
7261 struct kvm_clock_pairing clock_pairing;
7262 struct timespec64 ts;
7263 u64 cycle;
7264 int ret;
7265
7266 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7267 return -KVM_EOPNOTSUPP;
7268
7269 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7270 return -KVM_EOPNOTSUPP;
7271
7272 clock_pairing.sec = ts.tv_sec;
7273 clock_pairing.nsec = ts.tv_nsec;
7274 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7275 clock_pairing.flags = 0;
7276 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7277
7278 ret = 0;
7279 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7280 sizeof(struct kvm_clock_pairing)))
7281 ret = -KVM_EFAULT;
7282
7283 return ret;
7284}
7285#endif
7286
7287/*
7288 * kvm_pv_kick_cpu_op: Kick a vcpu.
7289 *
7290 * @apicid - apicid of vcpu to be kicked.
7291 */
7292static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7293{
7294 struct kvm_lapic_irq lapic_irq;
7295
7296 lapic_irq.shorthand = 0;
7297 lapic_irq.dest_mode = 0;
7298 lapic_irq.level = 0;
7299 lapic_irq.dest_id = apicid;
7300 lapic_irq.msi_redir_hint = false;
7301
7302 lapic_irq.delivery_mode = APIC_DM_REMRD;
7303 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7304}
7305
7306void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
7307{
7308 if (!lapic_in_kernel(vcpu)) {
7309 WARN_ON_ONCE(vcpu->arch.apicv_active);
7310 return;
7311 }
7312 if (!vcpu->arch.apicv_active)
7313 return;
7314
7315 vcpu->arch.apicv_active = false;
7316 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
7317}
7318
7319static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
7320{
7321 struct kvm_vcpu *target = NULL;
7322 struct kvm_apic_map *map;
7323
7324 rcu_read_lock();
7325 map = rcu_dereference(kvm->arch.apic_map);
7326
7327 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
7328 target = map->phys_map[dest_id]->vcpu;
7329
7330 rcu_read_unlock();
7331
7332 if (target && READ_ONCE(target->ready))
7333 kvm_vcpu_yield_to(target);
7334}
7335
7336int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7337{
7338 unsigned long nr, a0, a1, a2, a3, ret;
7339 int op_64_bit;
7340
7341 if (kvm_hv_hypercall_enabled(vcpu->kvm))
7342 return kvm_hv_hypercall(vcpu);
7343
7344 nr = kvm_rax_read(vcpu);
7345 a0 = kvm_rbx_read(vcpu);
7346 a1 = kvm_rcx_read(vcpu);
7347 a2 = kvm_rdx_read(vcpu);
7348 a3 = kvm_rsi_read(vcpu);
7349
7350 trace_kvm_hypercall(nr, a0, a1, a2, a3);
7351
7352 op_64_bit = is_64_bit_mode(vcpu);
7353 if (!op_64_bit) {
7354 nr &= 0xFFFFFFFF;
7355 a0 &= 0xFFFFFFFF;
7356 a1 &= 0xFFFFFFFF;
7357 a2 &= 0xFFFFFFFF;
7358 a3 &= 0xFFFFFFFF;
7359 }
7360
7361 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
7362 ret = -KVM_EPERM;
7363 goto out;
7364 }
7365
7366 switch (nr) {
7367 case KVM_HC_VAPIC_POLL_IRQ:
7368 ret = 0;
7369 break;
7370 case KVM_HC_KICK_CPU:
7371 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7372 kvm_sched_yield(vcpu->kvm, a1);
7373 ret = 0;
7374 break;
7375#ifdef CONFIG_X86_64
7376 case KVM_HC_CLOCK_PAIRING:
7377 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7378 break;
7379#endif
7380 case KVM_HC_SEND_IPI:
7381 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7382 break;
7383 case KVM_HC_SCHED_YIELD:
7384 kvm_sched_yield(vcpu->kvm, a0);
7385 ret = 0;
7386 break;
7387 default:
7388 ret = -KVM_ENOSYS;
7389 break;
7390 }
7391out:
7392 if (!op_64_bit)
7393 ret = (u32)ret;
7394 kvm_rax_write(vcpu, ret);
7395
7396 ++vcpu->stat.hypercalls;
7397 return kvm_skip_emulated_instruction(vcpu);
7398}
7399EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7400
7401static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7402{
7403 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7404 char instruction[3];
7405 unsigned long rip = kvm_rip_read(vcpu);
7406
7407 kvm_x86_ops->patch_hypercall(vcpu, instruction);
7408
7409 return emulator_write_emulated(ctxt, rip, instruction, 3,
7410 &ctxt->exception);
7411}
7412
7413static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7414{
7415 return vcpu->run->request_interrupt_window &&
7416 likely(!pic_in_kernel(vcpu->kvm));
7417}
7418
7419static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7420{
7421 struct kvm_run *kvm_run = vcpu->run;
7422
7423 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7424 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7425 kvm_run->cr8 = kvm_get_cr8(vcpu);
7426 kvm_run->apic_base = kvm_get_apic_base(vcpu);
7427 kvm_run->ready_for_interrupt_injection =
7428 pic_in_kernel(vcpu->kvm) ||
7429 kvm_vcpu_ready_for_interrupt_injection(vcpu);
7430}
7431
7432static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7433{
7434 int max_irr, tpr;
7435
7436 if (!kvm_x86_ops->update_cr8_intercept)
7437 return;
7438
7439 if (!lapic_in_kernel(vcpu))
7440 return;
7441
7442 if (vcpu->arch.apicv_active)
7443 return;
7444
7445 if (!vcpu->arch.apic->vapic_addr)
7446 max_irr = kvm_lapic_find_highest_irr(vcpu);
7447 else
7448 max_irr = -1;
7449
7450 if (max_irr != -1)
7451 max_irr >>= 4;
7452
7453 tpr = kvm_lapic_get_cr8(vcpu);
7454
7455 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7456}
7457
7458static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7459{
7460 int r;
7461
7462 /* try to reinject previous events if any */
7463
7464 if (vcpu->arch.exception.injected)
7465 kvm_x86_ops->queue_exception(vcpu);
7466 /*
7467 * Do not inject an NMI or interrupt if there is a pending
7468 * exception. Exceptions and interrupts are recognized at
7469 * instruction boundaries, i.e. the start of an instruction.
7470 * Trap-like exceptions, e.g. #DB, have higher priority than
7471 * NMIs and interrupts, i.e. traps are recognized before an
7472 * NMI/interrupt that's pending on the same instruction.
7473 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7474 * priority, but are only generated (pended) during instruction
7475 * execution, i.e. a pending fault-like exception means the
7476 * fault occurred on the *previous* instruction and must be
7477 * serviced prior to recognizing any new events in order to
7478 * fully complete the previous instruction.
7479 */
7480 else if (!vcpu->arch.exception.pending) {
7481 if (vcpu->arch.nmi_injected)
7482 kvm_x86_ops->set_nmi(vcpu);
7483 else if (vcpu->arch.interrupt.injected)
7484 kvm_x86_ops->set_irq(vcpu);
7485 }
7486
7487 /*
7488 * Call check_nested_events() even if we reinjected a previous event
7489 * in order for caller to determine if it should require immediate-exit
7490 * from L2 to L1 due to pending L1 events which require exit
7491 * from L2 to L1.
7492 */
7493 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7494 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7495 if (r != 0)
7496 return r;
7497 }
7498
7499 /* try to inject new event if pending */
7500 if (vcpu->arch.exception.pending) {
7501 trace_kvm_inj_exception(vcpu->arch.exception.nr,
7502 vcpu->arch.exception.has_error_code,
7503 vcpu->arch.exception.error_code);
7504
7505 WARN_ON_ONCE(vcpu->arch.exception.injected);
7506 vcpu->arch.exception.pending = false;
7507 vcpu->arch.exception.injected = true;
7508
7509 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7510 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7511 X86_EFLAGS_RF);
7512
7513 if (vcpu->arch.exception.nr == DB_VECTOR) {
7514 /*
7515 * This code assumes that nSVM doesn't use
7516 * check_nested_events(). If it does, the
7517 * DR6/DR7 changes should happen before L1
7518 * gets a #VMEXIT for an intercepted #DB in
7519 * L2. (Under VMX, on the other hand, the
7520 * DR6/DR7 changes should not happen in the
7521 * event of a VM-exit to L1 for an intercepted
7522 * #DB in L2.)
7523 */
7524 kvm_deliver_exception_payload(vcpu);
7525 if (vcpu->arch.dr7 & DR7_GD) {
7526 vcpu->arch.dr7 &= ~DR7_GD;
7527 kvm_update_dr7(vcpu);
7528 }
7529 }
7530
7531 kvm_x86_ops->queue_exception(vcpu);
7532 }
7533
7534 /* Don't consider new event if we re-injected an event */
7535 if (kvm_event_needs_reinjection(vcpu))
7536 return 0;
7537
7538 if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7539 kvm_x86_ops->smi_allowed(vcpu)) {
7540 vcpu->arch.smi_pending = false;
7541 ++vcpu->arch.smi_count;
7542 enter_smm(vcpu);
7543 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7544 --vcpu->arch.nmi_pending;
7545 vcpu->arch.nmi_injected = true;
7546 kvm_x86_ops->set_nmi(vcpu);
7547 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
7548 /*
7549 * Because interrupts can be injected asynchronously, we are
7550 * calling check_nested_events again here to avoid a race condition.
7551 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7552 * proposal and current concerns. Perhaps we should be setting
7553 * KVM_REQ_EVENT only on certain events and not unconditionally?
7554 */
7555 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7556 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7557 if (r != 0)
7558 return r;
7559 }
7560 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7561 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7562 false);
7563 kvm_x86_ops->set_irq(vcpu);
7564 }
7565 }
7566
7567 return 0;
7568}
7569
7570static void process_nmi(struct kvm_vcpu *vcpu)
7571{
7572 unsigned limit = 2;
7573
7574 /*
7575 * x86 is limited to one NMI running, and one NMI pending after it.
7576 * If an NMI is already in progress, limit further NMIs to just one.
7577 * Otherwise, allow two (and we'll inject the first one immediately).
7578 */
7579 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7580 limit = 1;
7581
7582 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7583 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7584 kvm_make_request(KVM_REQ_EVENT, vcpu);
7585}
7586
7587static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7588{
7589 u32 flags = 0;
7590 flags |= seg->g << 23;
7591 flags |= seg->db << 22;
7592 flags |= seg->l << 21;
7593 flags |= seg->avl << 20;
7594 flags |= seg->present << 15;
7595 flags |= seg->dpl << 13;
7596 flags |= seg->s << 12;
7597 flags |= seg->type << 8;
7598 return flags;
7599}
7600
7601static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7602{
7603 struct kvm_segment seg;
7604 int offset;
7605
7606 kvm_get_segment(vcpu, &seg, n);
7607 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7608
7609 if (n < 3)
7610 offset = 0x7f84 + n * 12;
7611 else
7612 offset = 0x7f2c + (n - 3) * 12;
7613
7614 put_smstate(u32, buf, offset + 8, seg.base);
7615 put_smstate(u32, buf, offset + 4, seg.limit);
7616 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7617}
7618
7619#ifdef CONFIG_X86_64
7620static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7621{
7622 struct kvm_segment seg;
7623 int offset;
7624 u16 flags;
7625
7626 kvm_get_segment(vcpu, &seg, n);
7627 offset = 0x7e00 + n * 16;
7628
7629 flags = enter_smm_get_segment_flags(&seg) >> 8;
7630 put_smstate(u16, buf, offset, seg.selector);
7631 put_smstate(u16, buf, offset + 2, flags);
7632 put_smstate(u32, buf, offset + 4, seg.limit);
7633 put_smstate(u64, buf, offset + 8, seg.base);
7634}
7635#endif
7636
7637static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7638{
7639 struct desc_ptr dt;
7640 struct kvm_segment seg;
7641 unsigned long val;
7642 int i;
7643
7644 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7645 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7646 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7647 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7648
7649 for (i = 0; i < 8; i++)
7650 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7651
7652 kvm_get_dr(vcpu, 6, &val);
7653 put_smstate(u32, buf, 0x7fcc, (u32)val);
7654 kvm_get_dr(vcpu, 7, &val);
7655 put_smstate(u32, buf, 0x7fc8, (u32)val);
7656
7657 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7658 put_smstate(u32, buf, 0x7fc4, seg.selector);
7659 put_smstate(u32, buf, 0x7f64, seg.base);
7660 put_smstate(u32, buf, 0x7f60, seg.limit);
7661 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7662
7663 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7664 put_smstate(u32, buf, 0x7fc0, seg.selector);
7665 put_smstate(u32, buf, 0x7f80, seg.base);
7666 put_smstate(u32, buf, 0x7f7c, seg.limit);
7667 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7668
7669 kvm_x86_ops->get_gdt(vcpu, &dt);
7670 put_smstate(u32, buf, 0x7f74, dt.address);
7671 put_smstate(u32, buf, 0x7f70, dt.size);
7672
7673 kvm_x86_ops->get_idt(vcpu, &dt);
7674 put_smstate(u32, buf, 0x7f58, dt.address);
7675 put_smstate(u32, buf, 0x7f54, dt.size);
7676
7677 for (i = 0; i < 6; i++)
7678 enter_smm_save_seg_32(vcpu, buf, i);
7679
7680 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7681
7682 /* revision id */
7683 put_smstate(u32, buf, 0x7efc, 0x00020000);
7684 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7685}
7686
7687#ifdef CONFIG_X86_64
7688static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7689{
7690 struct desc_ptr dt;
7691 struct kvm_segment seg;
7692 unsigned long val;
7693 int i;
7694
7695 for (i = 0; i < 16; i++)
7696 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7697
7698 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7699 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7700
7701 kvm_get_dr(vcpu, 6, &val);
7702 put_smstate(u64, buf, 0x7f68, val);
7703 kvm_get_dr(vcpu, 7, &val);
7704 put_smstate(u64, buf, 0x7f60, val);
7705
7706 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7707 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7708 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7709
7710 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7711
7712 /* revision id */
7713 put_smstate(u32, buf, 0x7efc, 0x00020064);
7714
7715 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7716
7717 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7718 put_smstate(u16, buf, 0x7e90, seg.selector);
7719 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7720 put_smstate(u32, buf, 0x7e94, seg.limit);
7721 put_smstate(u64, buf, 0x7e98, seg.base);
7722
7723 kvm_x86_ops->get_idt(vcpu, &dt);
7724 put_smstate(u32, buf, 0x7e84, dt.size);
7725 put_smstate(u64, buf, 0x7e88, dt.address);
7726
7727 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7728 put_smstate(u16, buf, 0x7e70, seg.selector);
7729 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7730 put_smstate(u32, buf, 0x7e74, seg.limit);
7731 put_smstate(u64, buf, 0x7e78, seg.base);
7732
7733 kvm_x86_ops->get_gdt(vcpu, &dt);
7734 put_smstate(u32, buf, 0x7e64, dt.size);
7735 put_smstate(u64, buf, 0x7e68, dt.address);
7736
7737 for (i = 0; i < 6; i++)
7738 enter_smm_save_seg_64(vcpu, buf, i);
7739}
7740#endif
7741
7742static void enter_smm(struct kvm_vcpu *vcpu)
7743{
7744 struct kvm_segment cs, ds;
7745 struct desc_ptr dt;
7746 char buf[512];
7747 u32 cr0;
7748
7749 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7750 memset(buf, 0, 512);
7751#ifdef CONFIG_X86_64
7752 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7753 enter_smm_save_state_64(vcpu, buf);
7754 else
7755#endif
7756 enter_smm_save_state_32(vcpu, buf);
7757
7758 /*
7759 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7760 * vCPU state (e.g. leave guest mode) after we've saved the state into
7761 * the SMM state-save area.
7762 */
7763 kvm_x86_ops->pre_enter_smm(vcpu, buf);
7764
7765 vcpu->arch.hflags |= HF_SMM_MASK;
7766 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7767
7768 if (kvm_x86_ops->get_nmi_mask(vcpu))
7769 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7770 else
7771 kvm_x86_ops->set_nmi_mask(vcpu, true);
7772
7773 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7774 kvm_rip_write(vcpu, 0x8000);
7775
7776 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7777 kvm_x86_ops->set_cr0(vcpu, cr0);
7778 vcpu->arch.cr0 = cr0;
7779
7780 kvm_x86_ops->set_cr4(vcpu, 0);
7781
7782 /* Undocumented: IDT limit is set to zero on entry to SMM. */
7783 dt.address = dt.size = 0;
7784 kvm_x86_ops->set_idt(vcpu, &dt);
7785
7786 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7787
7788 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7789 cs.base = vcpu->arch.smbase;
7790
7791 ds.selector = 0;
7792 ds.base = 0;
7793
7794 cs.limit = ds.limit = 0xffffffff;
7795 cs.type = ds.type = 0x3;
7796 cs.dpl = ds.dpl = 0;
7797 cs.db = ds.db = 0;
7798 cs.s = ds.s = 1;
7799 cs.l = ds.l = 0;
7800 cs.g = ds.g = 1;
7801 cs.avl = ds.avl = 0;
7802 cs.present = ds.present = 1;
7803 cs.unusable = ds.unusable = 0;
7804 cs.padding = ds.padding = 0;
7805
7806 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7807 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7808 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7809 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7810 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7811 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7812
7813#ifdef CONFIG_X86_64
7814 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7815 kvm_x86_ops->set_efer(vcpu, 0);
7816#endif
7817
7818 kvm_update_cpuid(vcpu);
7819 kvm_mmu_reset_context(vcpu);
7820}
7821
7822static void process_smi(struct kvm_vcpu *vcpu)
7823{
7824 vcpu->arch.smi_pending = true;
7825 kvm_make_request(KVM_REQ_EVENT, vcpu);
7826}
7827
7828void kvm_make_scan_ioapic_request(struct kvm *kvm)
7829{
7830 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7831}
7832
7833static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7834{
7835 if (!kvm_apic_present(vcpu))
7836 return;
7837
7838 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7839
7840 if (irqchip_split(vcpu->kvm))
7841 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7842 else {
7843 if (vcpu->arch.apicv_active)
7844 kvm_x86_ops->sync_pir_to_irr(vcpu);
7845 if (ioapic_in_kernel(vcpu->kvm))
7846 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7847 }
7848
7849 if (is_guest_mode(vcpu))
7850 vcpu->arch.load_eoi_exitmap_pending = true;
7851 else
7852 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7853}
7854
7855static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7856{
7857 u64 eoi_exit_bitmap[4];
7858
7859 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7860 return;
7861
7862 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7863 vcpu_to_synic(vcpu)->vec_bitmap, 256);
7864 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7865}
7866
7867int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7868 unsigned long start, unsigned long end,
7869 bool blockable)
7870{
7871 unsigned long apic_address;
7872
7873 /*
7874 * The physical address of apic access page is stored in the VMCS.
7875 * Update it when it becomes invalid.
7876 */
7877 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7878 if (start <= apic_address && apic_address < end)
7879 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7880
7881 return 0;
7882}
7883
7884void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7885{
7886 struct page *page = NULL;
7887
7888 if (!lapic_in_kernel(vcpu))
7889 return;
7890
7891 if (!kvm_x86_ops->set_apic_access_page_addr)
7892 return;
7893
7894 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7895 if (is_error_page(page))
7896 return;
7897 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7898
7899 /*
7900 * Do not pin apic access page in memory, the MMU notifier
7901 * will call us again if it is migrated or swapped out.
7902 */
7903 put_page(page);
7904}
7905EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7906
7907void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
7908{
7909 smp_send_reschedule(vcpu->cpu);
7910}
7911EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
7912
7913/*
7914 * Returns 1 to let vcpu_run() continue the guest execution loop without
7915 * exiting to the userspace. Otherwise, the value will be returned to the
7916 * userspace.
7917 */
7918static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7919{
7920 int r;
7921 bool req_int_win =
7922 dm_request_for_irq_injection(vcpu) &&
7923 kvm_cpu_accept_dm_intr(vcpu);
7924
7925 bool req_immediate_exit = false;
7926
7927 if (kvm_request_pending(vcpu)) {
7928 if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu))
7929 kvm_x86_ops->get_vmcs12_pages(vcpu);
7930 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7931 kvm_mmu_unload(vcpu);
7932 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7933 __kvm_migrate_timers(vcpu);
7934 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7935 kvm_gen_update_masterclock(vcpu->kvm);
7936 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7937 kvm_gen_kvmclock_update(vcpu);
7938 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7939 r = kvm_guest_time_update(vcpu);
7940 if (unlikely(r))
7941 goto out;
7942 }
7943 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7944 kvm_mmu_sync_roots(vcpu);
7945 if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
7946 kvm_mmu_load_cr3(vcpu);
7947 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7948 kvm_vcpu_flush_tlb(vcpu, true);
7949 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7950 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7951 r = 0;
7952 goto out;
7953 }
7954 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
7955 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
7956 vcpu->mmio_needed = 0;
7957 r = 0;
7958 goto out;
7959 }
7960 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7961 /* Page is swapped out. Do synthetic halt */
7962 vcpu->arch.apf.halted = true;
7963 r = 1;
7964 goto out;
7965 }
7966 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
7967 record_steal_time(vcpu);
7968 if (kvm_check_request(KVM_REQ_SMI, vcpu))
7969 process_smi(vcpu);
7970 if (kvm_check_request(KVM_REQ_NMI, vcpu))
7971 process_nmi(vcpu);
7972 if (kvm_check_request(KVM_REQ_PMU, vcpu))
7973 kvm_pmu_handle_event(vcpu);
7974 if (kvm_check_request(KVM_REQ_PMI, vcpu))
7975 kvm_pmu_deliver_pmi(vcpu);
7976 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7977 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7978 if (test_bit(vcpu->arch.pending_ioapic_eoi,
7979 vcpu->arch.ioapic_handled_vectors)) {
7980 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7981 vcpu->run->eoi.vector =
7982 vcpu->arch.pending_ioapic_eoi;
7983 r = 0;
7984 goto out;
7985 }
7986 }
7987 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
7988 vcpu_scan_ioapic(vcpu);
7989 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
7990 vcpu_load_eoi_exitmap(vcpu);
7991 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
7992 kvm_vcpu_reload_apic_access_page(vcpu);
7993 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
7994 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7995 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
7996 r = 0;
7997 goto out;
7998 }
7999 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8000 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8001 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8002 r = 0;
8003 goto out;
8004 }
8005 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8006 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8007 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8008 r = 0;
8009 goto out;
8010 }
8011
8012 /*
8013 * KVM_REQ_HV_STIMER has to be processed after
8014 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8015 * depend on the guest clock being up-to-date
8016 */
8017 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8018 kvm_hv_process_stimers(vcpu);
8019 }
8020
8021 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8022 ++vcpu->stat.req_event;
8023 kvm_apic_accept_events(vcpu);
8024 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8025 r = 1;
8026 goto out;
8027 }
8028
8029 if (inject_pending_event(vcpu, req_int_win) != 0)
8030 req_immediate_exit = true;
8031 else {
8032 /* Enable SMI/NMI/IRQ window open exits if needed.
8033 *
8034 * SMIs have three cases:
8035 * 1) They can be nested, and then there is nothing to
8036 * do here because RSM will cause a vmexit anyway.
8037 * 2) There is an ISA-specific reason why SMI cannot be
8038 * injected, and the moment when this changes can be
8039 * intercepted.
8040 * 3) Or the SMI can be pending because
8041 * inject_pending_event has completed the injection
8042 * of an IRQ or NMI from the previous vmexit, and
8043 * then we request an immediate exit to inject the
8044 * SMI.
8045 */
8046 if (vcpu->arch.smi_pending && !is_smm(vcpu))
8047 if (!kvm_x86_ops->enable_smi_window(vcpu))
8048 req_immediate_exit = true;
8049 if (vcpu->arch.nmi_pending)
8050 kvm_x86_ops->enable_nmi_window(vcpu);
8051 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
8052 kvm_x86_ops->enable_irq_window(vcpu);
8053 WARN_ON(vcpu->arch.exception.pending);
8054 }
8055
8056 if (kvm_lapic_enabled(vcpu)) {
8057 update_cr8_intercept(vcpu);
8058 kvm_lapic_sync_to_vapic(vcpu);
8059 }
8060 }
8061
8062 r = kvm_mmu_reload(vcpu);
8063 if (unlikely(r)) {
8064 goto cancel_injection;
8065 }
8066
8067 preempt_disable();
8068
8069 kvm_x86_ops->prepare_guest_switch(vcpu);
8070
8071 /*
8072 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
8073 * IPI are then delayed after guest entry, which ensures that they
8074 * result in virtual interrupt delivery.
8075 */
8076 local_irq_disable();
8077 vcpu->mode = IN_GUEST_MODE;
8078
8079 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8080
8081 /*
8082 * 1) We should set ->mode before checking ->requests. Please see
8083 * the comment in kvm_vcpu_exiting_guest_mode().
8084 *
8085 * 2) For APICv, we should set ->mode before checking PID.ON. This
8086 * pairs with the memory barrier implicit in pi_test_and_set_on
8087 * (see vmx_deliver_posted_interrupt).
8088 *
8089 * 3) This also orders the write to mode from any reads to the page
8090 * tables done while the VCPU is running. Please see the comment
8091 * in kvm_flush_remote_tlbs.
8092 */
8093 smp_mb__after_srcu_read_unlock();
8094
8095 /*
8096 * This handles the case where a posted interrupt was
8097 * notified with kvm_vcpu_kick.
8098 */
8099 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8100 kvm_x86_ops->sync_pir_to_irr(vcpu);
8101
8102 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
8103 || need_resched() || signal_pending(current)) {
8104 vcpu->mode = OUTSIDE_GUEST_MODE;
8105 smp_wmb();
8106 local_irq_enable();
8107 preempt_enable();
8108 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8109 r = 1;
8110 goto cancel_injection;
8111 }
8112
8113 if (req_immediate_exit) {
8114 kvm_make_request(KVM_REQ_EVENT, vcpu);
8115 kvm_x86_ops->request_immediate_exit(vcpu);
8116 }
8117
8118 trace_kvm_entry(vcpu->vcpu_id);
8119 guest_enter_irqoff();
8120
8121 /* The preempt notifier should have taken care of the FPU already. */
8122 WARN_ON_ONCE(test_thread_flag(TIF_NEED_FPU_LOAD));
8123
8124 if (unlikely(vcpu->arch.switch_db_regs)) {
8125 set_debugreg(0, 7);
8126 set_debugreg(vcpu->arch.eff_db[0], 0);
8127 set_debugreg(vcpu->arch.eff_db[1], 1);
8128 set_debugreg(vcpu->arch.eff_db[2], 2);
8129 set_debugreg(vcpu->arch.eff_db[3], 3);
8130 set_debugreg(vcpu->arch.dr6, 6);
8131 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8132 }
8133
8134 kvm_x86_ops->run(vcpu);
8135
8136 /*
8137 * Do this here before restoring debug registers on the host. And
8138 * since we do this before handling the vmexit, a DR access vmexit
8139 * can (a) read the correct value of the debug registers, (b) set
8140 * KVM_DEBUGREG_WONT_EXIT again.
8141 */
8142 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8143 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8144 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
8145 kvm_update_dr0123(vcpu);
8146 kvm_update_dr6(vcpu);
8147 kvm_update_dr7(vcpu);
8148 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8149 }
8150
8151 /*
8152 * If the guest has used debug registers, at least dr7
8153 * will be disabled while returning to the host.
8154 * If we don't have active breakpoints in the host, we don't
8155 * care about the messed up debug address registers. But if
8156 * we have some of them active, restore the old state.
8157 */
8158 if (hw_breakpoint_active())
8159 hw_breakpoint_restore();
8160
8161 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8162
8163 vcpu->mode = OUTSIDE_GUEST_MODE;
8164 smp_wmb();
8165
8166 kvm_x86_ops->handle_exit_irqoff(vcpu);
8167
8168 /*
8169 * Consume any pending interrupts, including the possible source of
8170 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8171 * An instruction is required after local_irq_enable() to fully unblock
8172 * interrupts on processors that implement an interrupt shadow, the
8173 * stat.exits increment will do nicely.
8174 */
8175 kvm_before_interrupt(vcpu);
8176 local_irq_enable();
8177 ++vcpu->stat.exits;
8178 local_irq_disable();
8179 kvm_after_interrupt(vcpu);
8180
8181 guest_exit_irqoff();
8182 if (lapic_in_kernel(vcpu)) {
8183 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8184 if (delta != S64_MIN) {
8185 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8186 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8187 }
8188 }
8189
8190 local_irq_enable();
8191 preempt_enable();
8192
8193 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8194
8195 /*
8196 * Profile KVM exit RIPs:
8197 */
8198 if (unlikely(prof_on == KVM_PROFILING)) {
8199 unsigned long rip = kvm_rip_read(vcpu);
8200 profile_hit(KVM_PROFILING, (void *)rip);
8201 }
8202
8203 if (unlikely(vcpu->arch.tsc_always_catchup))
8204 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8205
8206 if (vcpu->arch.apic_attention)
8207 kvm_lapic_sync_from_vapic(vcpu);
8208
8209 vcpu->arch.gpa_available = false;
8210 r = kvm_x86_ops->handle_exit(vcpu);
8211 return r;
8212
8213cancel_injection:
8214 kvm_x86_ops->cancel_injection(vcpu);
8215 if (unlikely(vcpu->arch.apic_attention))
8216 kvm_lapic_sync_from_vapic(vcpu);
8217out:
8218 return r;
8219}
8220
8221static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8222{
8223 if (!kvm_arch_vcpu_runnable(vcpu) &&
8224 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
8225 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8226 kvm_vcpu_block(vcpu);
8227 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8228
8229 if (kvm_x86_ops->post_block)
8230 kvm_x86_ops->post_block(vcpu);
8231
8232 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8233 return 1;
8234 }
8235
8236 kvm_apic_accept_events(vcpu);
8237 switch(vcpu->arch.mp_state) {
8238 case KVM_MP_STATE_HALTED:
8239 vcpu->arch.pv.pv_unhalted = false;
8240 vcpu->arch.mp_state =
8241 KVM_MP_STATE_RUNNABLE;
8242 /* fall through */
8243 case KVM_MP_STATE_RUNNABLE:
8244 vcpu->arch.apf.halted = false;
8245 break;
8246 case KVM_MP_STATE_INIT_RECEIVED:
8247 break;
8248 default:
8249 return -EINTR;
8250 break;
8251 }
8252 return 1;
8253}
8254
8255static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8256{
8257 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8258 kvm_x86_ops->check_nested_events(vcpu, false);
8259
8260 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8261 !vcpu->arch.apf.halted);
8262}
8263
8264static int vcpu_run(struct kvm_vcpu *vcpu)
8265{
8266 int r;
8267 struct kvm *kvm = vcpu->kvm;
8268
8269 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8270 vcpu->arch.l1tf_flush_l1d = true;
8271
8272 for (;;) {
8273 if (kvm_vcpu_running(vcpu)) {
8274 r = vcpu_enter_guest(vcpu);
8275 } else {
8276 r = vcpu_block(kvm, vcpu);
8277 }
8278
8279 if (r <= 0)
8280 break;
8281
8282 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8283 if (kvm_cpu_has_pending_timer(vcpu))
8284 kvm_inject_pending_timer_irqs(vcpu);
8285
8286 if (dm_request_for_irq_injection(vcpu) &&
8287 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8288 r = 0;
8289 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8290 ++vcpu->stat.request_irq_exits;
8291 break;
8292 }
8293
8294 kvm_check_async_pf_completion(vcpu);
8295
8296 if (signal_pending(current)) {
8297 r = -EINTR;
8298 vcpu->run->exit_reason = KVM_EXIT_INTR;
8299 ++vcpu->stat.signal_exits;
8300 break;
8301 }
8302 if (need_resched()) {
8303 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8304 cond_resched();
8305 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8306 }
8307 }
8308
8309 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8310
8311 return r;
8312}
8313
8314static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8315{
8316 int r;
8317 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8318 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8319 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8320 if (r != EMULATE_DONE)
8321 return 0;
8322 return 1;
8323}
8324
8325static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8326{
8327 BUG_ON(!vcpu->arch.pio.count);
8328
8329 return complete_emulated_io(vcpu);
8330}
8331
8332/*
8333 * Implements the following, as a state machine:
8334 *
8335 * read:
8336 * for each fragment
8337 * for each mmio piece in the fragment
8338 * write gpa, len
8339 * exit
8340 * copy data
8341 * execute insn
8342 *
8343 * write:
8344 * for each fragment
8345 * for each mmio piece in the fragment
8346 * write gpa, len
8347 * copy data
8348 * exit
8349 */
8350static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8351{
8352 struct kvm_run *run = vcpu->run;
8353 struct kvm_mmio_fragment *frag;
8354 unsigned len;
8355
8356 BUG_ON(!vcpu->mmio_needed);
8357
8358 /* Complete previous fragment */
8359 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8360 len = min(8u, frag->len);
8361 if (!vcpu->mmio_is_write)
8362 memcpy(frag->data, run->mmio.data, len);
8363
8364 if (frag->len <= 8) {
8365 /* Switch to the next fragment. */
8366 frag++;
8367 vcpu->mmio_cur_fragment++;
8368 } else {
8369 /* Go forward to the next mmio piece. */
8370 frag->data += len;
8371 frag->gpa += len;
8372 frag->len -= len;
8373 }
8374
8375 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8376 vcpu->mmio_needed = 0;
8377
8378 /* FIXME: return into emulator if single-stepping. */
8379 if (vcpu->mmio_is_write)
8380 return 1;
8381 vcpu->mmio_read_completed = 1;
8382 return complete_emulated_io(vcpu);
8383 }
8384
8385 run->exit_reason = KVM_EXIT_MMIO;
8386 run->mmio.phys_addr = frag->gpa;
8387 if (vcpu->mmio_is_write)
8388 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8389 run->mmio.len = min(8u, frag->len);
8390 run->mmio.is_write = vcpu->mmio_is_write;
8391 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8392 return 0;
8393}
8394
8395/* Swap (qemu) user FPU context for the guest FPU context. */
8396static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8397{
8398 fpregs_lock();
8399
8400 copy_fpregs_to_fpstate(vcpu->arch.user_fpu);
8401 /* PKRU is separately restored in kvm_x86_ops->run. */
8402 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8403 ~XFEATURE_MASK_PKRU);
8404
8405 fpregs_mark_activate();
8406 fpregs_unlock();
8407
8408 trace_kvm_fpu(1);
8409}
8410
8411/* When vcpu_run ends, restore user space FPU context. */
8412static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8413{
8414 fpregs_lock();
8415
8416 copy_fpregs_to_fpstate(vcpu->arch.guest_fpu);
8417 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
8418
8419 fpregs_mark_activate();
8420 fpregs_unlock();
8421
8422 ++vcpu->stat.fpu_reload;
8423 trace_kvm_fpu(0);
8424}
8425
8426int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8427{
8428 int r;
8429
8430 vcpu_load(vcpu);
8431 kvm_sigset_activate(vcpu);
8432 kvm_load_guest_fpu(vcpu);
8433
8434 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8435 if (kvm_run->immediate_exit) {
8436 r = -EINTR;
8437 goto out;
8438 }
8439 kvm_vcpu_block(vcpu);
8440 kvm_apic_accept_events(vcpu);
8441 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8442 r = -EAGAIN;
8443 if (signal_pending(current)) {
8444 r = -EINTR;
8445 vcpu->run->exit_reason = KVM_EXIT_INTR;
8446 ++vcpu->stat.signal_exits;
8447 }
8448 goto out;
8449 }
8450
8451 if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8452 r = -EINVAL;
8453 goto out;
8454 }
8455
8456 if (vcpu->run->kvm_dirty_regs) {
8457 r = sync_regs(vcpu);
8458 if (r != 0)
8459 goto out;
8460 }
8461
8462 /* re-sync apic's tpr */
8463 if (!lapic_in_kernel(vcpu)) {
8464 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8465 r = -EINVAL;
8466 goto out;
8467 }
8468 }
8469
8470 if (unlikely(vcpu->arch.complete_userspace_io)) {
8471 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8472 vcpu->arch.complete_userspace_io = NULL;
8473 r = cui(vcpu);
8474 if (r <= 0)
8475 goto out;
8476 } else
8477 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8478
8479 if (kvm_run->immediate_exit)
8480 r = -EINTR;
8481 else
8482 r = vcpu_run(vcpu);
8483
8484out:
8485 kvm_put_guest_fpu(vcpu);
8486 if (vcpu->run->kvm_valid_regs)
8487 store_regs(vcpu);
8488 post_kvm_run_save(vcpu);
8489 kvm_sigset_deactivate(vcpu);
8490
8491 vcpu_put(vcpu);
8492 return r;
8493}
8494
8495static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8496{
8497 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8498 /*
8499 * We are here if userspace calls get_regs() in the middle of
8500 * instruction emulation. Registers state needs to be copied
8501 * back from emulation context to vcpu. Userspace shouldn't do
8502 * that usually, but some bad designed PV devices (vmware
8503 * backdoor interface) need this to work
8504 */
8505 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8506 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8507 }
8508 regs->rax = kvm_rax_read(vcpu);
8509 regs->rbx = kvm_rbx_read(vcpu);
8510 regs->rcx = kvm_rcx_read(vcpu);
8511 regs->rdx = kvm_rdx_read(vcpu);
8512 regs->rsi = kvm_rsi_read(vcpu);
8513 regs->rdi = kvm_rdi_read(vcpu);
8514 regs->rsp = kvm_rsp_read(vcpu);
8515 regs->rbp = kvm_rbp_read(vcpu);
8516#ifdef CONFIG_X86_64
8517 regs->r8 = kvm_r8_read(vcpu);
8518 regs->r9 = kvm_r9_read(vcpu);
8519 regs->r10 = kvm_r10_read(vcpu);
8520 regs->r11 = kvm_r11_read(vcpu);
8521 regs->r12 = kvm_r12_read(vcpu);
8522 regs->r13 = kvm_r13_read(vcpu);
8523 regs->r14 = kvm_r14_read(vcpu);
8524 regs->r15 = kvm_r15_read(vcpu);
8525#endif
8526
8527 regs->rip = kvm_rip_read(vcpu);
8528 regs->rflags = kvm_get_rflags(vcpu);
8529}
8530
8531int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8532{
8533 vcpu_load(vcpu);
8534 __get_regs(vcpu, regs);
8535 vcpu_put(vcpu);
8536 return 0;
8537}
8538
8539static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8540{
8541 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8542 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8543
8544 kvm_rax_write(vcpu, regs->rax);
8545 kvm_rbx_write(vcpu, regs->rbx);
8546 kvm_rcx_write(vcpu, regs->rcx);
8547 kvm_rdx_write(vcpu, regs->rdx);
8548 kvm_rsi_write(vcpu, regs->rsi);
8549 kvm_rdi_write(vcpu, regs->rdi);
8550 kvm_rsp_write(vcpu, regs->rsp);
8551 kvm_rbp_write(vcpu, regs->rbp);
8552#ifdef CONFIG_X86_64
8553 kvm_r8_write(vcpu, regs->r8);
8554 kvm_r9_write(vcpu, regs->r9);
8555 kvm_r10_write(vcpu, regs->r10);
8556 kvm_r11_write(vcpu, regs->r11);
8557 kvm_r12_write(vcpu, regs->r12);
8558 kvm_r13_write(vcpu, regs->r13);
8559 kvm_r14_write(vcpu, regs->r14);
8560 kvm_r15_write(vcpu, regs->r15);
8561#endif
8562
8563 kvm_rip_write(vcpu, regs->rip);
8564 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8565
8566 vcpu->arch.exception.pending = false;
8567
8568 kvm_make_request(KVM_REQ_EVENT, vcpu);
8569}
8570
8571int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8572{
8573 vcpu_load(vcpu);
8574 __set_regs(vcpu, regs);
8575 vcpu_put(vcpu);
8576 return 0;
8577}
8578
8579void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8580{
8581 struct kvm_segment cs;
8582
8583 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8584 *db = cs.db;
8585 *l = cs.l;
8586}
8587EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8588
8589static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8590{
8591 struct desc_ptr dt;
8592
8593 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8594 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8595 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8596 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8597 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8598 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8599
8600 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8601 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8602
8603 kvm_x86_ops->get_idt(vcpu, &dt);
8604 sregs->idt.limit = dt.size;
8605 sregs->idt.base = dt.address;
8606 kvm_x86_ops->get_gdt(vcpu, &dt);
8607 sregs->gdt.limit = dt.size;
8608 sregs->gdt.base = dt.address;
8609
8610 sregs->cr0 = kvm_read_cr0(vcpu);
8611 sregs->cr2 = vcpu->arch.cr2;
8612 sregs->cr3 = kvm_read_cr3(vcpu);
8613 sregs->cr4 = kvm_read_cr4(vcpu);
8614 sregs->cr8 = kvm_get_cr8(vcpu);
8615 sregs->efer = vcpu->arch.efer;
8616 sregs->apic_base = kvm_get_apic_base(vcpu);
8617
8618 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8619
8620 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8621 set_bit(vcpu->arch.interrupt.nr,
8622 (unsigned long *)sregs->interrupt_bitmap);
8623}
8624
8625int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8626 struct kvm_sregs *sregs)
8627{
8628 vcpu_load(vcpu);
8629 __get_sregs(vcpu, sregs);
8630 vcpu_put(vcpu);
8631 return 0;
8632}
8633
8634int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8635 struct kvm_mp_state *mp_state)
8636{
8637 vcpu_load(vcpu);
8638
8639 kvm_apic_accept_events(vcpu);
8640 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8641 vcpu->arch.pv.pv_unhalted)
8642 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8643 else
8644 mp_state->mp_state = vcpu->arch.mp_state;
8645
8646 vcpu_put(vcpu);
8647 return 0;
8648}
8649
8650int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8651 struct kvm_mp_state *mp_state)
8652{
8653 int ret = -EINVAL;
8654
8655 vcpu_load(vcpu);
8656
8657 if (!lapic_in_kernel(vcpu) &&
8658 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8659 goto out;
8660
8661 /* INITs are latched while in SMM */
8662 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
8663 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8664 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8665 goto out;
8666
8667 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8668 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8669 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8670 } else
8671 vcpu->arch.mp_state = mp_state->mp_state;
8672 kvm_make_request(KVM_REQ_EVENT, vcpu);
8673
8674 ret = 0;
8675out:
8676 vcpu_put(vcpu);
8677 return ret;
8678}
8679
8680int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8681 int reason, bool has_error_code, u32 error_code)
8682{
8683 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8684 int ret;
8685
8686 init_emulate_ctxt(vcpu);
8687
8688 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8689 has_error_code, error_code);
8690
8691 if (ret)
8692 return EMULATE_FAIL;
8693
8694 kvm_rip_write(vcpu, ctxt->eip);
8695 kvm_set_rflags(vcpu, ctxt->eflags);
8696 kvm_make_request(KVM_REQ_EVENT, vcpu);
8697 return EMULATE_DONE;
8698}
8699EXPORT_SYMBOL_GPL(kvm_task_switch);
8700
8701static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8702{
8703 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
8704 (sregs->cr4 & X86_CR4_OSXSAVE))
8705 return -EINVAL;
8706
8707 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8708 /*
8709 * When EFER.LME and CR0.PG are set, the processor is in
8710 * 64-bit mode (though maybe in a 32-bit code segment).
8711 * CR4.PAE and EFER.LMA must be set.
8712 */
8713 if (!(sregs->cr4 & X86_CR4_PAE)
8714 || !(sregs->efer & EFER_LMA))
8715 return -EINVAL;
8716 } else {
8717 /*
8718 * Not in 64-bit mode: EFER.LMA is clear and the code
8719 * segment cannot be 64-bit.
8720 */
8721 if (sregs->efer & EFER_LMA || sregs->cs.l)
8722 return -EINVAL;
8723 }
8724
8725 return 0;
8726}
8727
8728static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8729{
8730 struct msr_data apic_base_msr;
8731 int mmu_reset_needed = 0;
8732 int cpuid_update_needed = 0;
8733 int pending_vec, max_bits, idx;
8734 struct desc_ptr dt;
8735 int ret = -EINVAL;
8736
8737 if (kvm_valid_sregs(vcpu, sregs))
8738 goto out;
8739
8740 apic_base_msr.data = sregs->apic_base;
8741 apic_base_msr.host_initiated = true;
8742 if (kvm_set_apic_base(vcpu, &apic_base_msr))
8743 goto out;
8744
8745 dt.size = sregs->idt.limit;
8746 dt.address = sregs->idt.base;
8747 kvm_x86_ops->set_idt(vcpu, &dt);
8748 dt.size = sregs->gdt.limit;
8749 dt.address = sregs->gdt.base;
8750 kvm_x86_ops->set_gdt(vcpu, &dt);
8751
8752 vcpu->arch.cr2 = sregs->cr2;
8753 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8754 vcpu->arch.cr3 = sregs->cr3;
8755 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8756
8757 kvm_set_cr8(vcpu, sregs->cr8);
8758
8759 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8760 kvm_x86_ops->set_efer(vcpu, sregs->efer);
8761
8762 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8763 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8764 vcpu->arch.cr0 = sregs->cr0;
8765
8766 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8767 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8768 (X86_CR4_OSXSAVE | X86_CR4_PKE));
8769 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8770 if (cpuid_update_needed)
8771 kvm_update_cpuid(vcpu);
8772
8773 idx = srcu_read_lock(&vcpu->kvm->srcu);
8774 if (is_pae_paging(vcpu)) {
8775 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8776 mmu_reset_needed = 1;
8777 }
8778 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8779
8780 if (mmu_reset_needed)
8781 kvm_mmu_reset_context(vcpu);
8782
8783 max_bits = KVM_NR_INTERRUPTS;
8784 pending_vec = find_first_bit(
8785 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
8786 if (pending_vec < max_bits) {
8787 kvm_queue_interrupt(vcpu, pending_vec, false);
8788 pr_debug("Set back pending irq %d\n", pending_vec);
8789 }
8790
8791 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8792 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8793 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8794 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8795 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8796 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8797
8798 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8799 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8800
8801 update_cr8_intercept(vcpu);
8802
8803 /* Older userspace won't unhalt the vcpu on reset. */
8804 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8805 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8806 !is_protmode(vcpu))
8807 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8808
8809 kvm_make_request(KVM_REQ_EVENT, vcpu);
8810
8811 ret = 0;
8812out:
8813 return ret;
8814}
8815
8816int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8817 struct kvm_sregs *sregs)
8818{
8819 int ret;
8820
8821 vcpu_load(vcpu);
8822 ret = __set_sregs(vcpu, sregs);
8823 vcpu_put(vcpu);
8824 return ret;
8825}
8826
8827int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8828 struct kvm_guest_debug *dbg)
8829{
8830 unsigned long rflags;
8831 int i, r;
8832
8833 vcpu_load(vcpu);
8834
8835 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8836 r = -EBUSY;
8837 if (vcpu->arch.exception.pending)
8838 goto out;
8839 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8840 kvm_queue_exception(vcpu, DB_VECTOR);
8841 else
8842 kvm_queue_exception(vcpu, BP_VECTOR);
8843 }
8844
8845 /*
8846 * Read rflags as long as potentially injected trace flags are still
8847 * filtered out.
8848 */
8849 rflags = kvm_get_rflags(vcpu);
8850
8851 vcpu->guest_debug = dbg->control;
8852 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8853 vcpu->guest_debug = 0;
8854
8855 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8856 for (i = 0; i < KVM_NR_DB_REGS; ++i)
8857 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8858 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8859 } else {
8860 for (i = 0; i < KVM_NR_DB_REGS; i++)
8861 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8862 }
8863 kvm_update_dr7(vcpu);
8864
8865 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8866 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8867 get_segment_base(vcpu, VCPU_SREG_CS);
8868
8869 /*
8870 * Trigger an rflags update that will inject or remove the trace
8871 * flags.
8872 */
8873 kvm_set_rflags(vcpu, rflags);
8874
8875 kvm_x86_ops->update_bp_intercept(vcpu);
8876
8877 r = 0;
8878
8879out:
8880 vcpu_put(vcpu);
8881 return r;
8882}
8883
8884/*
8885 * Translate a guest virtual address to a guest physical address.
8886 */
8887int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8888 struct kvm_translation *tr)
8889{
8890 unsigned long vaddr = tr->linear_address;
8891 gpa_t gpa;
8892 int idx;
8893
8894 vcpu_load(vcpu);
8895
8896 idx = srcu_read_lock(&vcpu->kvm->srcu);
8897 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8898 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8899 tr->physical_address = gpa;
8900 tr->valid = gpa != UNMAPPED_GVA;
8901 tr->writeable = 1;
8902 tr->usermode = 0;
8903
8904 vcpu_put(vcpu);
8905 return 0;
8906}
8907
8908int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8909{
8910 struct fxregs_state *fxsave;
8911
8912 vcpu_load(vcpu);
8913
8914 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8915 memcpy(fpu->fpr, fxsave->st_space, 128);
8916 fpu->fcw = fxsave->cwd;
8917 fpu->fsw = fxsave->swd;
8918 fpu->ftwx = fxsave->twd;
8919 fpu->last_opcode = fxsave->fop;
8920 fpu->last_ip = fxsave->rip;
8921 fpu->last_dp = fxsave->rdp;
8922 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
8923
8924 vcpu_put(vcpu);
8925 return 0;
8926}
8927
8928int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8929{
8930 struct fxregs_state *fxsave;
8931
8932 vcpu_load(vcpu);
8933
8934 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8935
8936 memcpy(fxsave->st_space, fpu->fpr, 128);
8937 fxsave->cwd = fpu->fcw;
8938 fxsave->swd = fpu->fsw;
8939 fxsave->twd = fpu->ftwx;
8940 fxsave->fop = fpu->last_opcode;
8941 fxsave->rip = fpu->last_ip;
8942 fxsave->rdp = fpu->last_dp;
8943 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
8944
8945 vcpu_put(vcpu);
8946 return 0;
8947}
8948
8949static void store_regs(struct kvm_vcpu *vcpu)
8950{
8951 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8952
8953 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8954 __get_regs(vcpu, &vcpu->run->s.regs.regs);
8955
8956 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8957 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8958
8959 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8960 kvm_vcpu_ioctl_x86_get_vcpu_events(
8961 vcpu, &vcpu->run->s.regs.events);
8962}
8963
8964static int sync_regs(struct kvm_vcpu *vcpu)
8965{
8966 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
8967 return -EINVAL;
8968
8969 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
8970 __set_regs(vcpu, &vcpu->run->s.regs.regs);
8971 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
8972 }
8973 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
8974 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
8975 return -EINVAL;
8976 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
8977 }
8978 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
8979 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
8980 vcpu, &vcpu->run->s.regs.events))
8981 return -EINVAL;
8982 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
8983 }
8984
8985 return 0;
8986}
8987
8988static void fx_init(struct kvm_vcpu *vcpu)
8989{
8990 fpstate_init(&vcpu->arch.guest_fpu->state);
8991 if (boot_cpu_has(X86_FEATURE_XSAVES))
8992 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
8993 host_xcr0 | XSTATE_COMPACTION_ENABLED;
8994
8995 /*
8996 * Ensure guest xcr0 is valid for loading
8997 */
8998 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8999
9000 vcpu->arch.cr0 |= X86_CR0_ET;
9001}
9002
9003void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
9004{
9005 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
9006
9007 kvmclock_reset(vcpu);
9008
9009 kvm_x86_ops->vcpu_free(vcpu);
9010 free_cpumask_var(wbinvd_dirty_mask);
9011}
9012
9013struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
9014 unsigned int id)
9015{
9016 struct kvm_vcpu *vcpu;
9017
9018 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9019 printk_once(KERN_WARNING
9020 "kvm: SMP vm created on host with unstable TSC; "
9021 "guest TSC will not be reliable\n");
9022
9023 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
9024
9025 return vcpu;
9026}
9027
9028int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
9029{
9030 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
9031 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
9032 kvm_vcpu_mtrr_init(vcpu);
9033 vcpu_load(vcpu);
9034 kvm_vcpu_reset(vcpu, false);
9035 kvm_init_mmu(vcpu, false);
9036 vcpu_put(vcpu);
9037 return 0;
9038}
9039
9040void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
9041{
9042 struct msr_data msr;
9043 struct kvm *kvm = vcpu->kvm;
9044
9045 kvm_hv_vcpu_postcreate(vcpu);
9046
9047 if (mutex_lock_killable(&vcpu->mutex))
9048 return;
9049 vcpu_load(vcpu);
9050 msr.data = 0x0;
9051 msr.index = MSR_IA32_TSC;
9052 msr.host_initiated = true;
9053 kvm_write_tsc(vcpu, &msr);
9054 vcpu_put(vcpu);
9055
9056 /* poll control enabled by default */
9057 vcpu->arch.msr_kvm_poll_control = 1;
9058
9059 mutex_unlock(&vcpu->mutex);
9060
9061 if (!kvmclock_periodic_sync)
9062 return;
9063
9064 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9065 KVMCLOCK_SYNC_PERIOD);
9066}
9067
9068void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9069{
9070 vcpu->arch.apf.msr_val = 0;
9071
9072 vcpu_load(vcpu);
9073 kvm_mmu_unload(vcpu);
9074 vcpu_put(vcpu);
9075
9076 kvm_x86_ops->vcpu_free(vcpu);
9077}
9078
9079void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
9080{
9081 kvm_lapic_reset(vcpu, init_event);
9082
9083 vcpu->arch.hflags = 0;
9084
9085 vcpu->arch.smi_pending = 0;
9086 vcpu->arch.smi_count = 0;
9087 atomic_set(&vcpu->arch.nmi_queued, 0);
9088 vcpu->arch.nmi_pending = 0;
9089 vcpu->arch.nmi_injected = false;
9090 kvm_clear_interrupt_queue(vcpu);
9091 kvm_clear_exception_queue(vcpu);
9092 vcpu->arch.exception.pending = false;
9093
9094 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
9095 kvm_update_dr0123(vcpu);
9096 vcpu->arch.dr6 = DR6_INIT;
9097 kvm_update_dr6(vcpu);
9098 vcpu->arch.dr7 = DR7_FIXED_1;
9099 kvm_update_dr7(vcpu);
9100
9101 vcpu->arch.cr2 = 0;
9102
9103 kvm_make_request(KVM_REQ_EVENT, vcpu);
9104 vcpu->arch.apf.msr_val = 0;
9105 vcpu->arch.st.msr_val = 0;
9106
9107 kvmclock_reset(vcpu);
9108
9109 kvm_clear_async_pf_completion_queue(vcpu);
9110 kvm_async_pf_hash_reset(vcpu);
9111 vcpu->arch.apf.halted = false;
9112
9113 if (kvm_mpx_supported()) {
9114 void *mpx_state_buffer;
9115
9116 /*
9117 * To avoid have the INIT path from kvm_apic_has_events() that be
9118 * called with loaded FPU and does not let userspace fix the state.
9119 */
9120 if (init_event)
9121 kvm_put_guest_fpu(vcpu);
9122 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9123 XFEATURE_BNDREGS);
9124 if (mpx_state_buffer)
9125 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
9126 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9127 XFEATURE_BNDCSR);
9128 if (mpx_state_buffer)
9129 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
9130 if (init_event)
9131 kvm_load_guest_fpu(vcpu);
9132 }
9133
9134 if (!init_event) {
9135 kvm_pmu_reset(vcpu);
9136 vcpu->arch.smbase = 0x30000;
9137
9138 vcpu->arch.msr_misc_features_enables = 0;
9139
9140 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9141 }
9142
9143 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
9144 vcpu->arch.regs_avail = ~0;
9145 vcpu->arch.regs_dirty = ~0;
9146
9147 vcpu->arch.ia32_xss = 0;
9148
9149 kvm_x86_ops->vcpu_reset(vcpu, init_event);
9150}
9151
9152void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
9153{
9154 struct kvm_segment cs;
9155
9156 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9157 cs.selector = vector << 8;
9158 cs.base = vector << 12;
9159 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9160 kvm_rip_write(vcpu, 0);
9161}
9162
9163int kvm_arch_hardware_enable(void)
9164{
9165 struct kvm *kvm;
9166 struct kvm_vcpu *vcpu;
9167 int i;
9168 int ret;
9169 u64 local_tsc;
9170 u64 max_tsc = 0;
9171 bool stable, backwards_tsc = false;
9172
9173 kvm_shared_msr_cpu_online();
9174 ret = kvm_x86_ops->hardware_enable();
9175 if (ret != 0)
9176 return ret;
9177
9178 local_tsc = rdtsc();
9179 stable = !kvm_check_tsc_unstable();
9180 list_for_each_entry(kvm, &vm_list, vm_list) {
9181 kvm_for_each_vcpu(i, vcpu, kvm) {
9182 if (!stable && vcpu->cpu == smp_processor_id())
9183 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9184 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
9185 backwards_tsc = true;
9186 if (vcpu->arch.last_host_tsc > max_tsc)
9187 max_tsc = vcpu->arch.last_host_tsc;
9188 }
9189 }
9190 }
9191
9192 /*
9193 * Sometimes, even reliable TSCs go backwards. This happens on
9194 * platforms that reset TSC during suspend or hibernate actions, but
9195 * maintain synchronization. We must compensate. Fortunately, we can
9196 * detect that condition here, which happens early in CPU bringup,
9197 * before any KVM threads can be running. Unfortunately, we can't
9198 * bring the TSCs fully up to date with real time, as we aren't yet far
9199 * enough into CPU bringup that we know how much real time has actually
9200 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
9201 * variables that haven't been updated yet.
9202 *
9203 * So we simply find the maximum observed TSC above, then record the
9204 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
9205 * the adjustment will be applied. Note that we accumulate
9206 * adjustments, in case multiple suspend cycles happen before some VCPU
9207 * gets a chance to run again. In the event that no KVM threads get a
9208 * chance to run, we will miss the entire elapsed period, as we'll have
9209 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
9210 * loose cycle time. This isn't too big a deal, since the loss will be
9211 * uniform across all VCPUs (not to mention the scenario is extremely
9212 * unlikely). It is possible that a second hibernate recovery happens
9213 * much faster than a first, causing the observed TSC here to be
9214 * smaller; this would require additional padding adjustment, which is
9215 * why we set last_host_tsc to the local tsc observed here.
9216 *
9217 * N.B. - this code below runs only on platforms with reliable TSC,
9218 * as that is the only way backwards_tsc is set above. Also note
9219 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
9220 * have the same delta_cyc adjustment applied if backwards_tsc
9221 * is detected. Note further, this adjustment is only done once,
9222 * as we reset last_host_tsc on all VCPUs to stop this from being
9223 * called multiple times (one for each physical CPU bringup).
9224 *
9225 * Platforms with unreliable TSCs don't have to deal with this, they
9226 * will be compensated by the logic in vcpu_load, which sets the TSC to
9227 * catchup mode. This will catchup all VCPUs to real time, but cannot
9228 * guarantee that they stay in perfect synchronization.
9229 */
9230 if (backwards_tsc) {
9231 u64 delta_cyc = max_tsc - local_tsc;
9232 list_for_each_entry(kvm, &vm_list, vm_list) {
9233 kvm->arch.backwards_tsc_observed = true;
9234 kvm_for_each_vcpu(i, vcpu, kvm) {
9235 vcpu->arch.tsc_offset_adjustment += delta_cyc;
9236 vcpu->arch.last_host_tsc = local_tsc;
9237 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9238 }
9239
9240 /*
9241 * We have to disable TSC offset matching.. if you were
9242 * booting a VM while issuing an S4 host suspend....
9243 * you may have some problem. Solving this issue is
9244 * left as an exercise to the reader.
9245 */
9246 kvm->arch.last_tsc_nsec = 0;
9247 kvm->arch.last_tsc_write = 0;
9248 }
9249
9250 }
9251 return 0;
9252}
9253
9254void kvm_arch_hardware_disable(void)
9255{
9256 kvm_x86_ops->hardware_disable();
9257 drop_user_return_notifiers();
9258}
9259
9260int kvm_arch_hardware_setup(void)
9261{
9262 int r;
9263
9264 r = kvm_x86_ops->hardware_setup();
9265 if (r != 0)
9266 return r;
9267
9268 if (kvm_has_tsc_control) {
9269 /*
9270 * Make sure the user can only configure tsc_khz values that
9271 * fit into a signed integer.
9272 * A min value is not calculated because it will always
9273 * be 1 on all machines.
9274 */
9275 u64 max = min(0x7fffffffULL,
9276 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9277 kvm_max_guest_tsc_khz = max;
9278
9279 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9280 }
9281
9282 kvm_init_msr_list();
9283 return 0;
9284}
9285
9286void kvm_arch_hardware_unsetup(void)
9287{
9288 kvm_x86_ops->hardware_unsetup();
9289}
9290
9291int kvm_arch_check_processor_compat(void)
9292{
9293 return kvm_x86_ops->check_processor_compatibility();
9294}
9295
9296bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9297{
9298 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9299}
9300EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9301
9302bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9303{
9304 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9305}
9306
9307struct static_key kvm_no_apic_vcpu __read_mostly;
9308EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9309
9310int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
9311{
9312 struct page *page;
9313 int r;
9314
9315 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
9316 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9317 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9318 else
9319 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9320
9321 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9322 if (!page) {
9323 r = -ENOMEM;
9324 goto fail;
9325 }
9326 vcpu->arch.pio_data = page_address(page);
9327
9328 kvm_set_tsc_khz(vcpu, max_tsc_khz);
9329
9330 r = kvm_mmu_create(vcpu);
9331 if (r < 0)
9332 goto fail_free_pio_data;
9333
9334 if (irqchip_in_kernel(vcpu->kvm)) {
9335 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
9336 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9337 if (r < 0)
9338 goto fail_mmu_destroy;
9339 } else
9340 static_key_slow_inc(&kvm_no_apic_vcpu);
9341
9342 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9343 GFP_KERNEL_ACCOUNT);
9344 if (!vcpu->arch.mce_banks) {
9345 r = -ENOMEM;
9346 goto fail_free_lapic;
9347 }
9348 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9349
9350 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9351 GFP_KERNEL_ACCOUNT)) {
9352 r = -ENOMEM;
9353 goto fail_free_mce_banks;
9354 }
9355
9356 fx_init(vcpu);
9357
9358 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9359
9360 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9361
9362 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9363
9364 kvm_async_pf_hash_reset(vcpu);
9365 kvm_pmu_init(vcpu);
9366
9367 vcpu->arch.pending_external_vector = -1;
9368 vcpu->arch.preempted_in_kernel = false;
9369
9370 kvm_hv_vcpu_init(vcpu);
9371
9372 return 0;
9373
9374fail_free_mce_banks:
9375 kfree(vcpu->arch.mce_banks);
9376fail_free_lapic:
9377 kvm_free_lapic(vcpu);
9378fail_mmu_destroy:
9379 kvm_mmu_destroy(vcpu);
9380fail_free_pio_data:
9381 free_page((unsigned long)vcpu->arch.pio_data);
9382fail:
9383 return r;
9384}
9385
9386void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
9387{
9388 int idx;
9389
9390 kvm_hv_vcpu_uninit(vcpu);
9391 kvm_pmu_destroy(vcpu);
9392 kfree(vcpu->arch.mce_banks);
9393 kvm_free_lapic(vcpu);
9394 idx = srcu_read_lock(&vcpu->kvm->srcu);
9395 kvm_mmu_destroy(vcpu);
9396 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9397 free_page((unsigned long)vcpu->arch.pio_data);
9398 if (!lapic_in_kernel(vcpu))
9399 static_key_slow_dec(&kvm_no_apic_vcpu);
9400}
9401
9402void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9403{
9404 vcpu->arch.l1tf_flush_l1d = true;
9405 kvm_x86_ops->sched_in(vcpu, cpu);
9406}
9407
9408int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9409{
9410 if (type)
9411 return -EINVAL;
9412
9413 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9414 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9415 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9416 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9417
9418 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9419 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9420 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9421 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9422 &kvm->arch.irq_sources_bitmap);
9423
9424 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9425 mutex_init(&kvm->arch.apic_map_lock);
9426 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9427
9428 kvm->arch.kvmclock_offset = -ktime_get_boottime_ns();
9429 pvclock_update_vm_gtod_copy(kvm);
9430
9431 kvm->arch.guest_can_read_msr_platform_info = true;
9432
9433 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9434 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9435
9436 kvm_hv_init_vm(kvm);
9437 kvm_page_track_init(kvm);
9438 kvm_mmu_init_vm(kvm);
9439
9440 return kvm_x86_ops->vm_init(kvm);
9441}
9442
9443static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9444{
9445 vcpu_load(vcpu);
9446 kvm_mmu_unload(vcpu);
9447 vcpu_put(vcpu);
9448}
9449
9450static void kvm_free_vcpus(struct kvm *kvm)
9451{
9452 unsigned int i;
9453 struct kvm_vcpu *vcpu;
9454
9455 /*
9456 * Unpin any mmu pages first.
9457 */
9458 kvm_for_each_vcpu(i, vcpu, kvm) {
9459 kvm_clear_async_pf_completion_queue(vcpu);
9460 kvm_unload_vcpu_mmu(vcpu);
9461 }
9462 kvm_for_each_vcpu(i, vcpu, kvm)
9463 kvm_arch_vcpu_free(vcpu);
9464
9465 mutex_lock(&kvm->lock);
9466 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9467 kvm->vcpus[i] = NULL;
9468
9469 atomic_set(&kvm->online_vcpus, 0);
9470 mutex_unlock(&kvm->lock);
9471}
9472
9473void kvm_arch_sync_events(struct kvm *kvm)
9474{
9475 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9476 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9477 kvm_free_pit(kvm);
9478}
9479
9480int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9481{
9482 int i, r;
9483 unsigned long hva;
9484 struct kvm_memslots *slots = kvm_memslots(kvm);
9485 struct kvm_memory_slot *slot, old;
9486
9487 /* Called with kvm->slots_lock held. */
9488 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9489 return -EINVAL;
9490
9491 slot = id_to_memslot(slots, id);
9492 if (size) {
9493 if (slot->npages)
9494 return -EEXIST;
9495
9496 /*
9497 * MAP_SHARED to prevent internal slot pages from being moved
9498 * by fork()/COW.
9499 */
9500 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9501 MAP_SHARED | MAP_ANONYMOUS, 0);
9502 if (IS_ERR((void *)hva))
9503 return PTR_ERR((void *)hva);
9504 } else {
9505 if (!slot->npages)
9506 return 0;
9507
9508 hva = 0;
9509 }
9510
9511 old = *slot;
9512 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9513 struct kvm_userspace_memory_region m;
9514
9515 m.slot = id | (i << 16);
9516 m.flags = 0;
9517 m.guest_phys_addr = gpa;
9518 m.userspace_addr = hva;
9519 m.memory_size = size;
9520 r = __kvm_set_memory_region(kvm, &m);
9521 if (r < 0)
9522 return r;
9523 }
9524
9525 if (!size)
9526 vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9527
9528 return 0;
9529}
9530EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9531
9532int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9533{
9534 int r;
9535
9536 mutex_lock(&kvm->slots_lock);
9537 r = __x86_set_memory_region(kvm, id, gpa, size);
9538 mutex_unlock(&kvm->slots_lock);
9539
9540 return r;
9541}
9542EXPORT_SYMBOL_GPL(x86_set_memory_region);
9543
9544void kvm_arch_destroy_vm(struct kvm *kvm)
9545{
9546 if (current->mm == kvm->mm) {
9547 /*
9548 * Free memory regions allocated on behalf of userspace,
9549 * unless the the memory map has changed due to process exit
9550 * or fd copying.
9551 */
9552 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9553 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9554 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9555 }
9556 if (kvm_x86_ops->vm_destroy)
9557 kvm_x86_ops->vm_destroy(kvm);
9558 kvm_pic_destroy(kvm);
9559 kvm_ioapic_destroy(kvm);
9560 kvm_free_vcpus(kvm);
9561 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9562 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
9563 kvm_mmu_uninit_vm(kvm);
9564 kvm_page_track_cleanup(kvm);
9565 kvm_hv_destroy_vm(kvm);
9566}
9567
9568void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9569 struct kvm_memory_slot *dont)
9570{
9571 int i;
9572
9573 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9574 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9575 kvfree(free->arch.rmap[i]);
9576 free->arch.rmap[i] = NULL;
9577 }
9578 if (i == 0)
9579 continue;
9580
9581 if (!dont || free->arch.lpage_info[i - 1] !=
9582 dont->arch.lpage_info[i - 1]) {
9583 kvfree(free->arch.lpage_info[i - 1]);
9584 free->arch.lpage_info[i - 1] = NULL;
9585 }
9586 }
9587
9588 kvm_page_track_free_memslot(free, dont);
9589}
9590
9591int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9592 unsigned long npages)
9593{
9594 int i;
9595
9596 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9597 struct kvm_lpage_info *linfo;
9598 unsigned long ugfn;
9599 int lpages;
9600 int level = i + 1;
9601
9602 lpages = gfn_to_index(slot->base_gfn + npages - 1,
9603 slot->base_gfn, level) + 1;
9604
9605 slot->arch.rmap[i] =
9606 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9607 GFP_KERNEL_ACCOUNT);
9608 if (!slot->arch.rmap[i])
9609 goto out_free;
9610 if (i == 0)
9611 continue;
9612
9613 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
9614 if (!linfo)
9615 goto out_free;
9616
9617 slot->arch.lpage_info[i - 1] = linfo;
9618
9619 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9620 linfo[0].disallow_lpage = 1;
9621 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9622 linfo[lpages - 1].disallow_lpage = 1;
9623 ugfn = slot->userspace_addr >> PAGE_SHIFT;
9624 /*
9625 * If the gfn and userspace address are not aligned wrt each
9626 * other, or if explicitly asked to, disable large page
9627 * support for this slot
9628 */
9629 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9630 !kvm_largepages_enabled()) {
9631 unsigned long j;
9632
9633 for (j = 0; j < lpages; ++j)
9634 linfo[j].disallow_lpage = 1;
9635 }
9636 }
9637
9638 if (kvm_page_track_create_memslot(slot, npages))
9639 goto out_free;
9640
9641 return 0;
9642
9643out_free:
9644 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9645 kvfree(slot->arch.rmap[i]);
9646 slot->arch.rmap[i] = NULL;
9647 if (i == 0)
9648 continue;
9649
9650 kvfree(slot->arch.lpage_info[i - 1]);
9651 slot->arch.lpage_info[i - 1] = NULL;
9652 }
9653 return -ENOMEM;
9654}
9655
9656void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
9657{
9658 /*
9659 * memslots->generation has been incremented.
9660 * mmio generation may have reached its maximum value.
9661 */
9662 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
9663}
9664
9665int kvm_arch_prepare_memory_region(struct kvm *kvm,
9666 struct kvm_memory_slot *memslot,
9667 const struct kvm_userspace_memory_region *mem,
9668 enum kvm_mr_change change)
9669{
9670 return 0;
9671}
9672
9673static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9674 struct kvm_memory_slot *new)
9675{
9676 /* Still write protect RO slot */
9677 if (new->flags & KVM_MEM_READONLY) {
9678 kvm_mmu_slot_remove_write_access(kvm, new);
9679 return;
9680 }
9681
9682 /*
9683 * Call kvm_x86_ops dirty logging hooks when they are valid.
9684 *
9685 * kvm_x86_ops->slot_disable_log_dirty is called when:
9686 *
9687 * - KVM_MR_CREATE with dirty logging is disabled
9688 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9689 *
9690 * The reason is, in case of PML, we need to set D-bit for any slots
9691 * with dirty logging disabled in order to eliminate unnecessary GPA
9692 * logging in PML buffer (and potential PML buffer full VMEXT). This
9693 * guarantees leaving PML enabled during guest's lifetime won't have
9694 * any additional overhead from PML when guest is running with dirty
9695 * logging disabled for memory slots.
9696 *
9697 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9698 * to dirty logging mode.
9699 *
9700 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9701 *
9702 * In case of write protect:
9703 *
9704 * Write protect all pages for dirty logging.
9705 *
9706 * All the sptes including the large sptes which point to this
9707 * slot are set to readonly. We can not create any new large
9708 * spte on this slot until the end of the logging.
9709 *
9710 * See the comments in fast_page_fault().
9711 */
9712 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9713 if (kvm_x86_ops->slot_enable_log_dirty)
9714 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9715 else
9716 kvm_mmu_slot_remove_write_access(kvm, new);
9717 } else {
9718 if (kvm_x86_ops->slot_disable_log_dirty)
9719 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9720 }
9721}
9722
9723void kvm_arch_commit_memory_region(struct kvm *kvm,
9724 const struct kvm_userspace_memory_region *mem,
9725 const struct kvm_memory_slot *old,
9726 const struct kvm_memory_slot *new,
9727 enum kvm_mr_change change)
9728{
9729 if (!kvm->arch.n_requested_mmu_pages)
9730 kvm_mmu_change_mmu_pages(kvm,
9731 kvm_mmu_calculate_default_mmu_pages(kvm));
9732
9733 /*
9734 * Dirty logging tracks sptes in 4k granularity, meaning that large
9735 * sptes have to be split. If live migration is successful, the guest
9736 * in the source machine will be destroyed and large sptes will be
9737 * created in the destination. However, if the guest continues to run
9738 * in the source machine (for example if live migration fails), small
9739 * sptes will remain around and cause bad performance.
9740 *
9741 * Scan sptes if dirty logging has been stopped, dropping those
9742 * which can be collapsed into a single large-page spte. Later
9743 * page faults will create the large-page sptes.
9744 *
9745 * There is no need to do this in any of the following cases:
9746 * CREATE: No dirty mappings will already exist.
9747 * MOVE/DELETE: The old mappings will already have been cleaned up by
9748 * kvm_arch_flush_shadow_memslot()
9749 */
9750 if (change == KVM_MR_FLAGS_ONLY &&
9751 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9752 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9753 kvm_mmu_zap_collapsible_sptes(kvm, new);
9754
9755 /*
9756 * Set up write protection and/or dirty logging for the new slot.
9757 *
9758 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9759 * been zapped so no dirty logging staff is needed for old slot. For
9760 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9761 * new and it's also covered when dealing with the new slot.
9762 *
9763 * FIXME: const-ify all uses of struct kvm_memory_slot.
9764 */
9765 if (change != KVM_MR_DELETE)
9766 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9767}
9768
9769void kvm_arch_flush_shadow_all(struct kvm *kvm)
9770{
9771 kvm_mmu_zap_all(kvm);
9772}
9773
9774void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9775 struct kvm_memory_slot *slot)
9776{
9777 kvm_page_track_flush_slot(kvm, slot);
9778}
9779
9780static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9781{
9782 return (is_guest_mode(vcpu) &&
9783 kvm_x86_ops->guest_apic_has_interrupt &&
9784 kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9785}
9786
9787static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9788{
9789 if (!list_empty_careful(&vcpu->async_pf.done))
9790 return true;
9791
9792 if (kvm_apic_has_events(vcpu))
9793 return true;
9794
9795 if (vcpu->arch.pv.pv_unhalted)
9796 return true;
9797
9798 if (vcpu->arch.exception.pending)
9799 return true;
9800
9801 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9802 (vcpu->arch.nmi_pending &&
9803 kvm_x86_ops->nmi_allowed(vcpu)))
9804 return true;
9805
9806 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9807 (vcpu->arch.smi_pending && !is_smm(vcpu)))
9808 return true;
9809
9810 if (kvm_arch_interrupt_allowed(vcpu) &&
9811 (kvm_cpu_has_interrupt(vcpu) ||
9812 kvm_guest_apic_has_interrupt(vcpu)))
9813 return true;
9814
9815 if (kvm_hv_has_stimer_pending(vcpu))
9816 return true;
9817
9818 return false;
9819}
9820
9821int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9822{
9823 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9824}
9825
9826bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
9827{
9828 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
9829 return true;
9830
9831 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9832 kvm_test_request(KVM_REQ_SMI, vcpu) ||
9833 kvm_test_request(KVM_REQ_EVENT, vcpu))
9834 return true;
9835
9836 if (vcpu->arch.apicv_active && kvm_x86_ops->dy_apicv_has_pending_interrupt(vcpu))
9837 return true;
9838
9839 return false;
9840}
9841
9842bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9843{
9844 return vcpu->arch.preempted_in_kernel;
9845}
9846
9847int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9848{
9849 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9850}
9851
9852int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9853{
9854 return kvm_x86_ops->interrupt_allowed(vcpu);
9855}
9856
9857unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9858{
9859 if (is_64_bit_mode(vcpu))
9860 return kvm_rip_read(vcpu);
9861 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9862 kvm_rip_read(vcpu));
9863}
9864EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9865
9866bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9867{
9868 return kvm_get_linear_rip(vcpu) == linear_rip;
9869}
9870EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9871
9872unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9873{
9874 unsigned long rflags;
9875
9876 rflags = kvm_x86_ops->get_rflags(vcpu);
9877 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9878 rflags &= ~X86_EFLAGS_TF;
9879 return rflags;
9880}
9881EXPORT_SYMBOL_GPL(kvm_get_rflags);
9882
9883static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9884{
9885 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9886 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9887 rflags |= X86_EFLAGS_TF;
9888 kvm_x86_ops->set_rflags(vcpu, rflags);
9889}
9890
9891void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9892{
9893 __kvm_set_rflags(vcpu, rflags);
9894 kvm_make_request(KVM_REQ_EVENT, vcpu);
9895}
9896EXPORT_SYMBOL_GPL(kvm_set_rflags);
9897
9898void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9899{
9900 int r;
9901
9902 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
9903 work->wakeup_all)
9904 return;
9905
9906 r = kvm_mmu_reload(vcpu);
9907 if (unlikely(r))
9908 return;
9909
9910 if (!vcpu->arch.mmu->direct_map &&
9911 work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
9912 return;
9913
9914 vcpu->arch.mmu->page_fault(vcpu, work->gva, 0, true);
9915}
9916
9917static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9918{
9919 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9920}
9921
9922static inline u32 kvm_async_pf_next_probe(u32 key)
9923{
9924 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9925}
9926
9927static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9928{
9929 u32 key = kvm_async_pf_hash_fn(gfn);
9930
9931 while (vcpu->arch.apf.gfns[key] != ~0)
9932 key = kvm_async_pf_next_probe(key);
9933
9934 vcpu->arch.apf.gfns[key] = gfn;
9935}
9936
9937static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9938{
9939 int i;
9940 u32 key = kvm_async_pf_hash_fn(gfn);
9941
9942 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9943 (vcpu->arch.apf.gfns[key] != gfn &&
9944 vcpu->arch.apf.gfns[key] != ~0); i++)
9945 key = kvm_async_pf_next_probe(key);
9946
9947 return key;
9948}
9949
9950bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9951{
9952 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9953}
9954
9955static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9956{
9957 u32 i, j, k;
9958
9959 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9960 while (true) {
9961 vcpu->arch.apf.gfns[i] = ~0;
9962 do {
9963 j = kvm_async_pf_next_probe(j);
9964 if (vcpu->arch.apf.gfns[j] == ~0)
9965 return;
9966 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
9967 /*
9968 * k lies cyclically in ]i,j]
9969 * | i.k.j |
9970 * |....j i.k.| or |.k..j i...|
9971 */
9972 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
9973 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
9974 i = j;
9975 }
9976}
9977
9978static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
9979{
9980
9981 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
9982 sizeof(val));
9983}
9984
9985static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
9986{
9987
9988 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
9989 sizeof(u32));
9990}
9991
9992static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
9993{
9994 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
9995 return false;
9996
9997 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
9998 (vcpu->arch.apf.send_user_only &&
9999 kvm_x86_ops->get_cpl(vcpu) == 0))
10000 return false;
10001
10002 return true;
10003}
10004
10005bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
10006{
10007 if (unlikely(!lapic_in_kernel(vcpu) ||
10008 kvm_event_needs_reinjection(vcpu) ||
10009 vcpu->arch.exception.pending))
10010 return false;
10011
10012 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
10013 return false;
10014
10015 /*
10016 * If interrupts are off we cannot even use an artificial
10017 * halt state.
10018 */
10019 return kvm_x86_ops->interrupt_allowed(vcpu);
10020}
10021
10022void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
10023 struct kvm_async_pf *work)
10024{
10025 struct x86_exception fault;
10026
10027 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
10028 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
10029
10030 if (kvm_can_deliver_async_pf(vcpu) &&
10031 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
10032 fault.vector = PF_VECTOR;
10033 fault.error_code_valid = true;
10034 fault.error_code = 0;
10035 fault.nested_page_fault = false;
10036 fault.address = work->arch.token;
10037 fault.async_page_fault = true;
10038 kvm_inject_page_fault(vcpu, &fault);
10039 } else {
10040 /*
10041 * It is not possible to deliver a paravirtualized asynchronous
10042 * page fault, but putting the guest in an artificial halt state
10043 * can be beneficial nevertheless: if an interrupt arrives, we
10044 * can deliver it timely and perhaps the guest will schedule
10045 * another process. When the instruction that triggered a page
10046 * fault is retried, hopefully the page will be ready in the host.
10047 */
10048 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
10049 }
10050}
10051
10052void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
10053 struct kvm_async_pf *work)
10054{
10055 struct x86_exception fault;
10056 u32 val;
10057
10058 if (work->wakeup_all)
10059 work->arch.token = ~0; /* broadcast wakeup */
10060 else
10061 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10062 trace_kvm_async_pf_ready(work->arch.token, work->gva);
10063
10064 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
10065 !apf_get_user(vcpu, &val)) {
10066 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
10067 vcpu->arch.exception.pending &&
10068 vcpu->arch.exception.nr == PF_VECTOR &&
10069 !apf_put_user(vcpu, 0)) {
10070 vcpu->arch.exception.injected = false;
10071 vcpu->arch.exception.pending = false;
10072 vcpu->arch.exception.nr = 0;
10073 vcpu->arch.exception.has_error_code = false;
10074 vcpu->arch.exception.error_code = 0;
10075 vcpu->arch.exception.has_payload = false;
10076 vcpu->arch.exception.payload = 0;
10077 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
10078 fault.vector = PF_VECTOR;
10079 fault.error_code_valid = true;
10080 fault.error_code = 0;
10081 fault.nested_page_fault = false;
10082 fault.address = work->arch.token;
10083 fault.async_page_fault = true;
10084 kvm_inject_page_fault(vcpu, &fault);
10085 }
10086 }
10087 vcpu->arch.apf.halted = false;
10088 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10089}
10090
10091bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
10092{
10093 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
10094 return true;
10095 else
10096 return kvm_can_do_async_pf(vcpu);
10097}
10098
10099void kvm_arch_start_assignment(struct kvm *kvm)
10100{
10101 atomic_inc(&kvm->arch.assigned_device_count);
10102}
10103EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
10104
10105void kvm_arch_end_assignment(struct kvm *kvm)
10106{
10107 atomic_dec(&kvm->arch.assigned_device_count);
10108}
10109EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
10110
10111bool kvm_arch_has_assigned_device(struct kvm *kvm)
10112{
10113 return atomic_read(&kvm->arch.assigned_device_count);
10114}
10115EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
10116
10117void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
10118{
10119 atomic_inc(&kvm->arch.noncoherent_dma_count);
10120}
10121EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
10122
10123void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
10124{
10125 atomic_dec(&kvm->arch.noncoherent_dma_count);
10126}
10127EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
10128
10129bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
10130{
10131 return atomic_read(&kvm->arch.noncoherent_dma_count);
10132}
10133EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
10134
10135bool kvm_arch_has_irq_bypass(void)
10136{
10137 return true;
10138}
10139
10140int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
10141 struct irq_bypass_producer *prod)
10142{
10143 struct kvm_kernel_irqfd *irqfd =
10144 container_of(cons, struct kvm_kernel_irqfd, consumer);
10145
10146 irqfd->producer = prod;
10147
10148 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
10149 prod->irq, irqfd->gsi, 1);
10150}
10151
10152void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
10153 struct irq_bypass_producer *prod)
10154{
10155 int ret;
10156 struct kvm_kernel_irqfd *irqfd =
10157 container_of(cons, struct kvm_kernel_irqfd, consumer);
10158
10159 WARN_ON(irqfd->producer != prod);
10160 irqfd->producer = NULL;
10161
10162 /*
10163 * When producer of consumer is unregistered, we change back to
10164 * remapped mode, so we can re-use the current implementation
10165 * when the irq is masked/disabled or the consumer side (KVM
10166 * int this case doesn't want to receive the interrupts.
10167 */
10168 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
10169 if (ret)
10170 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
10171 " fails: %d\n", irqfd->consumer.token, ret);
10172}
10173
10174int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
10175 uint32_t guest_irq, bool set)
10176{
10177 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
10178}
10179
10180bool kvm_vector_hashing_enabled(void)
10181{
10182 return vector_hashing;
10183}
10184EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
10185
10186bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
10187{
10188 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
10189}
10190EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
10191
10192
10193EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
10194EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
10195EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
10196EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
10197EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
10198EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
10199EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
10200EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
10201EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
10202EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
10203EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
10204EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
10205EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
10206EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
10207EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
10208EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
10209EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
10210EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
10211EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
10212EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
Linux 6.x block layer and io_uring tree(s)