2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
14 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
22 #include <linux/kvm_host.h>
27 #include "kvm_cache_regs.h"
31 #include <linux/clocksource.h>
32 #include <linux/interrupt.h>
33 #include <linux/kvm.h>
35 #include <linux/vmalloc.h>
36 #include <linux/module.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 <trace/events/kvm.h>
53 #define CREATE_TRACE_POINTS
56 #include <asm/debugreg.h>
62 #include <asm/fpu-internal.h> /* Ugh! */
64 #include <asm/pvclock.h>
65 #include <asm/div64.h>
67 #define MAX_IO_MSRS 256
68 #define KVM_MAX_MCE_BANKS 32
69 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
71 #define emul_to_vcpu(ctxt) \
72 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
75 * - enable syscall per default because its emulated by KVM
76 * - enable LME and LMA per default on 64 bit KVM
80 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
82 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
85 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
86 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
88 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
89 static void process_nmi(struct kvm_vcpu *vcpu);
91 struct kvm_x86_ops *kvm_x86_ops;
92 EXPORT_SYMBOL_GPL(kvm_x86_ops);
94 static bool ignore_msrs = 0;
95 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
97 unsigned int min_timer_period_us = 500;
98 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
100 bool kvm_has_tsc_control;
101 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
102 u32 kvm_max_guest_tsc_khz;
103 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
105 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
106 static u32 tsc_tolerance_ppm = 250;
107 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
109 #define KVM_NR_SHARED_MSRS 16
111 struct kvm_shared_msrs_global {
113 u32 msrs[KVM_NR_SHARED_MSRS];
116 struct kvm_shared_msrs {
117 struct user_return_notifier urn;
119 struct kvm_shared_msr_values {
122 } values[KVM_NR_SHARED_MSRS];
125 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
126 static struct kvm_shared_msrs __percpu *shared_msrs;
128 struct kvm_stats_debugfs_item debugfs_entries[] = {
129 { "pf_fixed", VCPU_STAT(pf_fixed) },
130 { "pf_guest", VCPU_STAT(pf_guest) },
131 { "tlb_flush", VCPU_STAT(tlb_flush) },
132 { "invlpg", VCPU_STAT(invlpg) },
133 { "exits", VCPU_STAT(exits) },
134 { "io_exits", VCPU_STAT(io_exits) },
135 { "mmio_exits", VCPU_STAT(mmio_exits) },
136 { "signal_exits", VCPU_STAT(signal_exits) },
137 { "irq_window", VCPU_STAT(irq_window_exits) },
138 { "nmi_window", VCPU_STAT(nmi_window_exits) },
139 { "halt_exits", VCPU_STAT(halt_exits) },
140 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
141 { "hypercalls", VCPU_STAT(hypercalls) },
142 { "request_irq", VCPU_STAT(request_irq_exits) },
143 { "irq_exits", VCPU_STAT(irq_exits) },
144 { "host_state_reload", VCPU_STAT(host_state_reload) },
145 { "efer_reload", VCPU_STAT(efer_reload) },
146 { "fpu_reload", VCPU_STAT(fpu_reload) },
147 { "insn_emulation", VCPU_STAT(insn_emulation) },
148 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
149 { "irq_injections", VCPU_STAT(irq_injections) },
150 { "nmi_injections", VCPU_STAT(nmi_injections) },
151 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
152 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
153 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
154 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
155 { "mmu_flooded", VM_STAT(mmu_flooded) },
156 { "mmu_recycled", VM_STAT(mmu_recycled) },
157 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
158 { "mmu_unsync", VM_STAT(mmu_unsync) },
159 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
160 { "largepages", VM_STAT(lpages) },
164 u64 __read_mostly host_xcr0;
166 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
168 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
171 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
172 vcpu->arch.apf.gfns[i] = ~0;
175 static void kvm_on_user_return(struct user_return_notifier *urn)
178 struct kvm_shared_msrs *locals
179 = container_of(urn, struct kvm_shared_msrs, urn);
180 struct kvm_shared_msr_values *values;
182 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
183 values = &locals->values[slot];
184 if (values->host != values->curr) {
185 wrmsrl(shared_msrs_global.msrs[slot], values->host);
186 values->curr = values->host;
189 locals->registered = false;
190 user_return_notifier_unregister(urn);
193 static void shared_msr_update(unsigned slot, u32 msr)
196 unsigned int cpu = smp_processor_id();
197 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
199 /* only read, and nobody should modify it at this time,
200 * so don't need lock */
201 if (slot >= shared_msrs_global.nr) {
202 printk(KERN_ERR "kvm: invalid MSR slot!");
205 rdmsrl_safe(msr, &value);
206 smsr->values[slot].host = value;
207 smsr->values[slot].curr = value;
210 void kvm_define_shared_msr(unsigned slot, u32 msr)
212 if (slot >= shared_msrs_global.nr)
213 shared_msrs_global.nr = slot + 1;
214 shared_msrs_global.msrs[slot] = msr;
215 /* we need ensured the shared_msr_global have been updated */
218 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
220 static void kvm_shared_msr_cpu_online(void)
224 for (i = 0; i < shared_msrs_global.nr; ++i)
225 shared_msr_update(i, shared_msrs_global.msrs[i]);
228 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
230 unsigned int cpu = smp_processor_id();
231 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
233 if (((value ^ smsr->values[slot].curr) & mask) == 0)
235 smsr->values[slot].curr = value;
236 wrmsrl(shared_msrs_global.msrs[slot], value);
237 if (!smsr->registered) {
238 smsr->urn.on_user_return = kvm_on_user_return;
239 user_return_notifier_register(&smsr->urn);
240 smsr->registered = true;
243 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
245 static void drop_user_return_notifiers(void *ignore)
247 unsigned int cpu = smp_processor_id();
248 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
250 if (smsr->registered)
251 kvm_on_user_return(&smsr->urn);
254 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
256 return vcpu->arch.apic_base;
258 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
260 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
262 u64 old_state = vcpu->arch.apic_base &
263 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
264 u64 new_state = msr_info->data &
265 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
266 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
267 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
269 if (!msr_info->host_initiated &&
270 ((msr_info->data & reserved_bits) != 0 ||
271 new_state == X2APIC_ENABLE ||
272 (new_state == MSR_IA32_APICBASE_ENABLE &&
273 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
274 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
278 kvm_lapic_set_base(vcpu, msr_info->data);
281 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
283 asmlinkage void kvm_spurious_fault(void)
285 /* Fault while not rebooting. We want the trace. */
288 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
290 #define EXCPT_BENIGN 0
291 #define EXCPT_CONTRIBUTORY 1
294 static int exception_class(int vector)
304 return EXCPT_CONTRIBUTORY;
311 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
312 unsigned nr, bool has_error, u32 error_code,
318 kvm_make_request(KVM_REQ_EVENT, vcpu);
320 if (!vcpu->arch.exception.pending) {
322 vcpu->arch.exception.pending = true;
323 vcpu->arch.exception.has_error_code = has_error;
324 vcpu->arch.exception.nr = nr;
325 vcpu->arch.exception.error_code = error_code;
326 vcpu->arch.exception.reinject = reinject;
330 /* to check exception */
331 prev_nr = vcpu->arch.exception.nr;
332 if (prev_nr == DF_VECTOR) {
333 /* triple fault -> shutdown */
334 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
337 class1 = exception_class(prev_nr);
338 class2 = exception_class(nr);
339 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
340 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
341 /* generate double fault per SDM Table 5-5 */
342 vcpu->arch.exception.pending = true;
343 vcpu->arch.exception.has_error_code = true;
344 vcpu->arch.exception.nr = DF_VECTOR;
345 vcpu->arch.exception.error_code = 0;
347 /* replace previous exception with a new one in a hope
348 that instruction re-execution will regenerate lost
353 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
355 kvm_multiple_exception(vcpu, nr, false, 0, false);
357 EXPORT_SYMBOL_GPL(kvm_queue_exception);
359 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
361 kvm_multiple_exception(vcpu, nr, false, 0, true);
363 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
365 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
368 kvm_inject_gp(vcpu, 0);
370 kvm_x86_ops->skip_emulated_instruction(vcpu);
372 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
374 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
376 ++vcpu->stat.pf_guest;
377 vcpu->arch.cr2 = fault->address;
378 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
380 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
382 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
384 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
385 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
387 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
390 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
392 atomic_inc(&vcpu->arch.nmi_queued);
393 kvm_make_request(KVM_REQ_NMI, vcpu);
395 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
397 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
399 kvm_multiple_exception(vcpu, nr, true, error_code, false);
401 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
403 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
405 kvm_multiple_exception(vcpu, nr, true, error_code, true);
407 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
410 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
411 * a #GP and return false.
413 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
415 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
417 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
420 EXPORT_SYMBOL_GPL(kvm_require_cpl);
423 * This function will be used to read from the physical memory of the currently
424 * running guest. The difference to kvm_read_guest_page is that this function
425 * can read from guest physical or from the guest's guest physical memory.
427 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
428 gfn_t ngfn, void *data, int offset, int len,
434 ngpa = gfn_to_gpa(ngfn);
435 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
436 if (real_gfn == UNMAPPED_GVA)
439 real_gfn = gpa_to_gfn(real_gfn);
441 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
443 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
445 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
446 void *data, int offset, int len, u32 access)
448 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
449 data, offset, len, access);
453 * Load the pae pdptrs. Return true is they are all valid.
455 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
457 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
458 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
461 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
463 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
464 offset * sizeof(u64), sizeof(pdpte),
465 PFERR_USER_MASK|PFERR_WRITE_MASK);
470 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
471 if (is_present_gpte(pdpte[i]) &&
472 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
479 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
480 __set_bit(VCPU_EXREG_PDPTR,
481 (unsigned long *)&vcpu->arch.regs_avail);
482 __set_bit(VCPU_EXREG_PDPTR,
483 (unsigned long *)&vcpu->arch.regs_dirty);
488 EXPORT_SYMBOL_GPL(load_pdptrs);
490 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
492 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
498 if (is_long_mode(vcpu) || !is_pae(vcpu))
501 if (!test_bit(VCPU_EXREG_PDPTR,
502 (unsigned long *)&vcpu->arch.regs_avail))
505 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
506 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
507 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
508 PFERR_USER_MASK | PFERR_WRITE_MASK);
511 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
517 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
519 unsigned long old_cr0 = kvm_read_cr0(vcpu);
520 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
521 X86_CR0_CD | X86_CR0_NW;
526 if (cr0 & 0xffffffff00000000UL)
530 cr0 &= ~CR0_RESERVED_BITS;
532 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
535 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
538 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
540 if ((vcpu->arch.efer & EFER_LME)) {
545 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
550 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
555 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
558 kvm_x86_ops->set_cr0(vcpu, cr0);
560 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
561 kvm_clear_async_pf_completion_queue(vcpu);
562 kvm_async_pf_hash_reset(vcpu);
565 if ((cr0 ^ old_cr0) & update_bits)
566 kvm_mmu_reset_context(vcpu);
569 EXPORT_SYMBOL_GPL(kvm_set_cr0);
571 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
573 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
575 EXPORT_SYMBOL_GPL(kvm_lmsw);
577 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
579 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
580 !vcpu->guest_xcr0_loaded) {
581 /* kvm_set_xcr() also depends on this */
582 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
583 vcpu->guest_xcr0_loaded = 1;
587 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
589 if (vcpu->guest_xcr0_loaded) {
590 if (vcpu->arch.xcr0 != host_xcr0)
591 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
592 vcpu->guest_xcr0_loaded = 0;
596 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
599 u64 old_xcr0 = vcpu->arch.xcr0;
602 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
603 if (index != XCR_XFEATURE_ENABLED_MASK)
605 if (!(xcr0 & XSTATE_FP))
607 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
611 * Do not allow the guest to set bits that we do not support
612 * saving. However, xcr0 bit 0 is always set, even if the
613 * emulated CPU does not support XSAVE (see fx_init).
615 valid_bits = vcpu->arch.guest_supported_xcr0 | XSTATE_FP;
616 if (xcr0 & ~valid_bits)
619 if ((!(xcr0 & XSTATE_BNDREGS)) != (!(xcr0 & XSTATE_BNDCSR)))
622 kvm_put_guest_xcr0(vcpu);
623 vcpu->arch.xcr0 = xcr0;
625 if ((xcr0 ^ old_xcr0) & XSTATE_EXTEND_MASK)
626 kvm_update_cpuid(vcpu);
630 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
632 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
633 __kvm_set_xcr(vcpu, index, xcr)) {
634 kvm_inject_gp(vcpu, 0);
639 EXPORT_SYMBOL_GPL(kvm_set_xcr);
641 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
643 unsigned long old_cr4 = kvm_read_cr4(vcpu);
644 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
645 X86_CR4_PAE | X86_CR4_SMEP;
646 if (cr4 & CR4_RESERVED_BITS)
649 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
652 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
655 if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
658 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
661 if (is_long_mode(vcpu)) {
662 if (!(cr4 & X86_CR4_PAE))
664 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
665 && ((cr4 ^ old_cr4) & pdptr_bits)
666 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
670 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
671 if (!guest_cpuid_has_pcid(vcpu))
674 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
675 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
679 if (kvm_x86_ops->set_cr4(vcpu, cr4))
682 if (((cr4 ^ old_cr4) & pdptr_bits) ||
683 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
684 kvm_mmu_reset_context(vcpu);
686 if ((cr4 ^ old_cr4) & X86_CR4_SMAP)
687 update_permission_bitmask(vcpu, vcpu->arch.walk_mmu, false);
689 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
690 kvm_update_cpuid(vcpu);
694 EXPORT_SYMBOL_GPL(kvm_set_cr4);
696 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
698 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
699 kvm_mmu_sync_roots(vcpu);
700 kvm_mmu_flush_tlb(vcpu);
704 if (is_long_mode(vcpu)) {
705 if (kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)) {
706 if (cr3 & CR3_PCID_ENABLED_RESERVED_BITS)
709 if (cr3 & CR3_L_MODE_RESERVED_BITS)
713 if (cr3 & CR3_PAE_RESERVED_BITS)
715 if (is_paging(vcpu) &&
716 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
720 * We don't check reserved bits in nonpae mode, because
721 * this isn't enforced, and VMware depends on this.
725 vcpu->arch.cr3 = cr3;
726 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
727 kvm_mmu_new_cr3(vcpu);
730 EXPORT_SYMBOL_GPL(kvm_set_cr3);
732 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
734 if (cr8 & CR8_RESERVED_BITS)
736 if (irqchip_in_kernel(vcpu->kvm))
737 kvm_lapic_set_tpr(vcpu, cr8);
739 vcpu->arch.cr8 = cr8;
742 EXPORT_SYMBOL_GPL(kvm_set_cr8);
744 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
746 if (irqchip_in_kernel(vcpu->kvm))
747 return kvm_lapic_get_cr8(vcpu);
749 return vcpu->arch.cr8;
751 EXPORT_SYMBOL_GPL(kvm_get_cr8);
753 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
755 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
756 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
759 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
763 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
764 dr7 = vcpu->arch.guest_debug_dr7;
766 dr7 = vcpu->arch.dr7;
767 kvm_x86_ops->set_dr7(vcpu, dr7);
768 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
769 if (dr7 & DR7_BP_EN_MASK)
770 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
773 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
777 vcpu->arch.db[dr] = val;
778 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
779 vcpu->arch.eff_db[dr] = val;
782 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
786 if (val & 0xffffffff00000000ULL)
788 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
789 kvm_update_dr6(vcpu);
792 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
796 if (val & 0xffffffff00000000ULL)
798 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
799 kvm_update_dr7(vcpu);
806 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
810 res = __kvm_set_dr(vcpu, dr, val);
812 kvm_queue_exception(vcpu, UD_VECTOR);
814 kvm_inject_gp(vcpu, 0);
818 EXPORT_SYMBOL_GPL(kvm_set_dr);
820 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
824 *val = vcpu->arch.db[dr];
827 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
831 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
832 *val = vcpu->arch.dr6;
834 *val = kvm_x86_ops->get_dr6(vcpu);
837 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
841 *val = vcpu->arch.dr7;
848 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
850 if (_kvm_get_dr(vcpu, dr, val)) {
851 kvm_queue_exception(vcpu, UD_VECTOR);
856 EXPORT_SYMBOL_GPL(kvm_get_dr);
858 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
860 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
864 err = kvm_pmu_read_pmc(vcpu, ecx, &data);
867 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
868 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
871 EXPORT_SYMBOL_GPL(kvm_rdpmc);
874 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
875 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
877 * This list is modified at module load time to reflect the
878 * capabilities of the host cpu. This capabilities test skips MSRs that are
879 * kvm-specific. Those are put in the beginning of the list.
882 #define KVM_SAVE_MSRS_BEGIN 12
883 static u32 msrs_to_save[] = {
884 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
885 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
886 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
887 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
888 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
890 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
893 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
895 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
896 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS
899 static unsigned num_msrs_to_save;
901 static const u32 emulated_msrs[] = {
903 MSR_IA32_TSCDEADLINE,
904 MSR_IA32_MISC_ENABLE,
909 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
911 if (efer & efer_reserved_bits)
914 if (efer & EFER_FFXSR) {
915 struct kvm_cpuid_entry2 *feat;
917 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
918 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
922 if (efer & EFER_SVME) {
923 struct kvm_cpuid_entry2 *feat;
925 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
926 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
932 EXPORT_SYMBOL_GPL(kvm_valid_efer);
934 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
936 u64 old_efer = vcpu->arch.efer;
938 if (!kvm_valid_efer(vcpu, efer))
942 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
946 efer |= vcpu->arch.efer & EFER_LMA;
948 kvm_x86_ops->set_efer(vcpu, efer);
950 /* Update reserved bits */
951 if ((efer ^ old_efer) & EFER_NX)
952 kvm_mmu_reset_context(vcpu);
957 void kvm_enable_efer_bits(u64 mask)
959 efer_reserved_bits &= ~mask;
961 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
965 * Writes msr value into into the appropriate "register".
966 * Returns 0 on success, non-0 otherwise.
967 * Assumes vcpu_load() was already called.
969 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
971 return kvm_x86_ops->set_msr(vcpu, msr);
975 * Adapt set_msr() to msr_io()'s calling convention
977 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
983 msr.host_initiated = true;
984 return kvm_set_msr(vcpu, &msr);
988 struct pvclock_gtod_data {
991 struct { /* extract of a clocksource struct */
999 /* open coded 'struct timespec' */
1000 u64 monotonic_time_snsec;
1001 time_t monotonic_time_sec;
1004 static struct pvclock_gtod_data pvclock_gtod_data;
1006 static void update_pvclock_gtod(struct timekeeper *tk)
1008 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1010 write_seqcount_begin(&vdata->seq);
1012 /* copy pvclock gtod data */
1013 vdata->clock.vclock_mode = tk->clock->archdata.vclock_mode;
1014 vdata->clock.cycle_last = tk->clock->cycle_last;
1015 vdata->clock.mask = tk->clock->mask;
1016 vdata->clock.mult = tk->mult;
1017 vdata->clock.shift = tk->shift;
1019 vdata->monotonic_time_sec = tk->xtime_sec
1020 + tk->wall_to_monotonic.tv_sec;
1021 vdata->monotonic_time_snsec = tk->xtime_nsec
1022 + (tk->wall_to_monotonic.tv_nsec
1024 while (vdata->monotonic_time_snsec >=
1025 (((u64)NSEC_PER_SEC) << tk->shift)) {
1026 vdata->monotonic_time_snsec -=
1027 ((u64)NSEC_PER_SEC) << tk->shift;
1028 vdata->monotonic_time_sec++;
1031 write_seqcount_end(&vdata->seq);
1036 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1040 struct pvclock_wall_clock wc;
1041 struct timespec boot;
1046 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1051 ++version; /* first time write, random junk */
1055 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1058 * The guest calculates current wall clock time by adding
1059 * system time (updated by kvm_guest_time_update below) to the
1060 * wall clock specified here. guest system time equals host
1061 * system time for us, thus we must fill in host boot time here.
1065 if (kvm->arch.kvmclock_offset) {
1066 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1067 boot = timespec_sub(boot, ts);
1069 wc.sec = boot.tv_sec;
1070 wc.nsec = boot.tv_nsec;
1071 wc.version = version;
1073 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1076 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1079 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1081 uint32_t quotient, remainder;
1083 /* Don't try to replace with do_div(), this one calculates
1084 * "(dividend << 32) / divisor" */
1086 : "=a" (quotient), "=d" (remainder)
1087 : "0" (0), "1" (dividend), "r" (divisor) );
1091 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1092 s8 *pshift, u32 *pmultiplier)
1099 tps64 = base_khz * 1000LL;
1100 scaled64 = scaled_khz * 1000LL;
1101 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1106 tps32 = (uint32_t)tps64;
1107 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1108 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1116 *pmultiplier = div_frac(scaled64, tps32);
1118 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1119 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1122 static inline u64 get_kernel_ns(void)
1127 monotonic_to_bootbased(&ts);
1128 return timespec_to_ns(&ts);
1131 #ifdef CONFIG_X86_64
1132 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1135 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1136 unsigned long max_tsc_khz;
1138 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1140 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1141 vcpu->arch.virtual_tsc_shift);
1144 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1146 u64 v = (u64)khz * (1000000 + ppm);
1151 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1153 u32 thresh_lo, thresh_hi;
1154 int use_scaling = 0;
1156 /* tsc_khz can be zero if TSC calibration fails */
1157 if (this_tsc_khz == 0)
1160 /* Compute a scale to convert nanoseconds in TSC cycles */
1161 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1162 &vcpu->arch.virtual_tsc_shift,
1163 &vcpu->arch.virtual_tsc_mult);
1164 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1167 * Compute the variation in TSC rate which is acceptable
1168 * within the range of tolerance and decide if the
1169 * rate being applied is within that bounds of the hardware
1170 * rate. If so, no scaling or compensation need be done.
1172 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1173 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1174 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1175 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1178 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1181 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1183 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1184 vcpu->arch.virtual_tsc_mult,
1185 vcpu->arch.virtual_tsc_shift);
1186 tsc += vcpu->arch.this_tsc_write;
1190 void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1192 #ifdef CONFIG_X86_64
1194 bool do_request = false;
1195 struct kvm_arch *ka = &vcpu->kvm->arch;
1196 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1198 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1199 atomic_read(&vcpu->kvm->online_vcpus));
1201 if (vcpus_matched && gtod->clock.vclock_mode == VCLOCK_TSC)
1202 if (!ka->use_master_clock)
1205 if (!vcpus_matched && ka->use_master_clock)
1209 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1211 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1212 atomic_read(&vcpu->kvm->online_vcpus),
1213 ka->use_master_clock, gtod->clock.vclock_mode);
1217 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1219 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1220 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1223 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1225 struct kvm *kvm = vcpu->kvm;
1226 u64 offset, ns, elapsed;
1227 unsigned long flags;
1230 u64 data = msr->data;
1232 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1233 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1234 ns = get_kernel_ns();
1235 elapsed = ns - kvm->arch.last_tsc_nsec;
1237 if (vcpu->arch.virtual_tsc_khz) {
1240 /* n.b - signed multiplication and division required */
1241 usdiff = data - kvm->arch.last_tsc_write;
1242 #ifdef CONFIG_X86_64
1243 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1245 /* do_div() only does unsigned */
1246 asm("1: idivl %[divisor]\n"
1247 "2: xor %%edx, %%edx\n"
1248 " movl $0, %[faulted]\n"
1250 ".section .fixup,\"ax\"\n"
1251 "4: movl $1, %[faulted]\n"
1255 _ASM_EXTABLE(1b, 4b)
1257 : "=A"(usdiff), [faulted] "=r" (faulted)
1258 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1261 do_div(elapsed, 1000);
1266 /* idivl overflow => difference is larger than USEC_PER_SEC */
1268 usdiff = USEC_PER_SEC;
1270 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1273 * Special case: TSC write with a small delta (1 second) of virtual
1274 * cycle time against real time is interpreted as an attempt to
1275 * synchronize the CPU.
1277 * For a reliable TSC, we can match TSC offsets, and for an unstable
1278 * TSC, we add elapsed time in this computation. We could let the
1279 * compensation code attempt to catch up if we fall behind, but
1280 * it's better to try to match offsets from the beginning.
1282 if (usdiff < USEC_PER_SEC &&
1283 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1284 if (!check_tsc_unstable()) {
1285 offset = kvm->arch.cur_tsc_offset;
1286 pr_debug("kvm: matched tsc offset for %llu\n", data);
1288 u64 delta = nsec_to_cycles(vcpu, elapsed);
1290 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1291 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1296 * We split periods of matched TSC writes into generations.
1297 * For each generation, we track the original measured
1298 * nanosecond time, offset, and write, so if TSCs are in
1299 * sync, we can match exact offset, and if not, we can match
1300 * exact software computation in compute_guest_tsc()
1302 * These values are tracked in kvm->arch.cur_xxx variables.
1304 kvm->arch.cur_tsc_generation++;
1305 kvm->arch.cur_tsc_nsec = ns;
1306 kvm->arch.cur_tsc_write = data;
1307 kvm->arch.cur_tsc_offset = offset;
1309 pr_debug("kvm: new tsc generation %u, clock %llu\n",
1310 kvm->arch.cur_tsc_generation, data);
1314 * We also track th most recent recorded KHZ, write and time to
1315 * allow the matching interval to be extended at each write.
1317 kvm->arch.last_tsc_nsec = ns;
1318 kvm->arch.last_tsc_write = data;
1319 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1321 vcpu->arch.last_guest_tsc = data;
1323 /* Keep track of which generation this VCPU has synchronized to */
1324 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1325 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1326 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1328 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1329 update_ia32_tsc_adjust_msr(vcpu, offset);
1330 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1331 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1333 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1335 kvm->arch.nr_vcpus_matched_tsc++;
1337 kvm->arch.nr_vcpus_matched_tsc = 0;
1339 kvm_track_tsc_matching(vcpu);
1340 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1343 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1345 #ifdef CONFIG_X86_64
1347 static cycle_t read_tsc(void)
1353 * Empirically, a fence (of type that depends on the CPU)
1354 * before rdtsc is enough to ensure that rdtsc is ordered
1355 * with respect to loads. The various CPU manuals are unclear
1356 * as to whether rdtsc can be reordered with later loads,
1357 * but no one has ever seen it happen.
1360 ret = (cycle_t)vget_cycles();
1362 last = pvclock_gtod_data.clock.cycle_last;
1364 if (likely(ret >= last))
1368 * GCC likes to generate cmov here, but this branch is extremely
1369 * predictable (it's just a funciton of time and the likely is
1370 * very likely) and there's a data dependence, so force GCC
1371 * to generate a branch instead. I don't barrier() because
1372 * we don't actually need a barrier, and if this function
1373 * ever gets inlined it will generate worse code.
1379 static inline u64 vgettsc(cycle_t *cycle_now)
1382 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1384 *cycle_now = read_tsc();
1386 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1387 return v * gtod->clock.mult;
1390 static int do_monotonic(struct timespec *ts, cycle_t *cycle_now)
1395 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1399 seq = read_seqcount_begin(>od->seq);
1400 mode = gtod->clock.vclock_mode;
1401 ts->tv_sec = gtod->monotonic_time_sec;
1402 ns = gtod->monotonic_time_snsec;
1403 ns += vgettsc(cycle_now);
1404 ns >>= gtod->clock.shift;
1405 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1406 timespec_add_ns(ts, ns);
1411 /* returns true if host is using tsc clocksource */
1412 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1416 /* checked again under seqlock below */
1417 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1420 if (do_monotonic(&ts, cycle_now) != VCLOCK_TSC)
1423 monotonic_to_bootbased(&ts);
1424 *kernel_ns = timespec_to_ns(&ts);
1432 * Assuming a stable TSC across physical CPUS, and a stable TSC
1433 * across virtual CPUs, the following condition is possible.
1434 * Each numbered line represents an event visible to both
1435 * CPUs at the next numbered event.
1437 * "timespecX" represents host monotonic time. "tscX" represents
1440 * VCPU0 on CPU0 | VCPU1 on CPU1
1442 * 1. read timespec0,tsc0
1443 * 2. | timespec1 = timespec0 + N
1445 * 3. transition to guest | transition to guest
1446 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1447 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1448 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1450 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1453 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1455 * - 0 < N - M => M < N
1457 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1458 * always the case (the difference between two distinct xtime instances
1459 * might be smaller then the difference between corresponding TSC reads,
1460 * when updating guest vcpus pvclock areas).
1462 * To avoid that problem, do not allow visibility of distinct
1463 * system_timestamp/tsc_timestamp values simultaneously: use a master
1464 * copy of host monotonic time values. Update that master copy
1467 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1471 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1473 #ifdef CONFIG_X86_64
1474 struct kvm_arch *ka = &kvm->arch;
1476 bool host_tsc_clocksource, vcpus_matched;
1478 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1479 atomic_read(&kvm->online_vcpus));
1482 * If the host uses TSC clock, then passthrough TSC as stable
1485 host_tsc_clocksource = kvm_get_time_and_clockread(
1486 &ka->master_kernel_ns,
1487 &ka->master_cycle_now);
1489 ka->use_master_clock = host_tsc_clocksource & vcpus_matched;
1491 if (ka->use_master_clock)
1492 atomic_set(&kvm_guest_has_master_clock, 1);
1494 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1495 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1500 static void kvm_gen_update_masterclock(struct kvm *kvm)
1502 #ifdef CONFIG_X86_64
1504 struct kvm_vcpu *vcpu;
1505 struct kvm_arch *ka = &kvm->arch;
1507 spin_lock(&ka->pvclock_gtod_sync_lock);
1508 kvm_make_mclock_inprogress_request(kvm);
1509 /* no guest entries from this point */
1510 pvclock_update_vm_gtod_copy(kvm);
1512 kvm_for_each_vcpu(i, vcpu, kvm)
1513 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
1515 /* guest entries allowed */
1516 kvm_for_each_vcpu(i, vcpu, kvm)
1517 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1519 spin_unlock(&ka->pvclock_gtod_sync_lock);
1523 static int kvm_guest_time_update(struct kvm_vcpu *v)
1525 unsigned long flags, this_tsc_khz;
1526 struct kvm_vcpu_arch *vcpu = &v->arch;
1527 struct kvm_arch *ka = &v->kvm->arch;
1529 u64 tsc_timestamp, host_tsc;
1530 struct pvclock_vcpu_time_info guest_hv_clock;
1532 bool use_master_clock;
1538 * If the host uses TSC clock, then passthrough TSC as stable
1541 spin_lock(&ka->pvclock_gtod_sync_lock);
1542 use_master_clock = ka->use_master_clock;
1543 if (use_master_clock) {
1544 host_tsc = ka->master_cycle_now;
1545 kernel_ns = ka->master_kernel_ns;
1547 spin_unlock(&ka->pvclock_gtod_sync_lock);
1549 /* Keep irq disabled to prevent changes to the clock */
1550 local_irq_save(flags);
1551 this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
1552 if (unlikely(this_tsc_khz == 0)) {
1553 local_irq_restore(flags);
1554 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1557 if (!use_master_clock) {
1558 host_tsc = native_read_tsc();
1559 kernel_ns = get_kernel_ns();
1562 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1565 * We may have to catch up the TSC to match elapsed wall clock
1566 * time for two reasons, even if kvmclock is used.
1567 * 1) CPU could have been running below the maximum TSC rate
1568 * 2) Broken TSC compensation resets the base at each VCPU
1569 * entry to avoid unknown leaps of TSC even when running
1570 * again on the same CPU. This may cause apparent elapsed
1571 * time to disappear, and the guest to stand still or run
1574 if (vcpu->tsc_catchup) {
1575 u64 tsc = compute_guest_tsc(v, kernel_ns);
1576 if (tsc > tsc_timestamp) {
1577 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1578 tsc_timestamp = tsc;
1582 local_irq_restore(flags);
1584 if (!vcpu->pv_time_enabled)
1587 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1588 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1589 &vcpu->hv_clock.tsc_shift,
1590 &vcpu->hv_clock.tsc_to_system_mul);
1591 vcpu->hw_tsc_khz = this_tsc_khz;
1594 /* With all the info we got, fill in the values */
1595 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1596 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1597 vcpu->last_guest_tsc = tsc_timestamp;
1600 * The interface expects us to write an even number signaling that the
1601 * update is finished. Since the guest won't see the intermediate
1602 * state, we just increase by 2 at the end.
1604 vcpu->hv_clock.version += 2;
1606 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1607 &guest_hv_clock, sizeof(guest_hv_clock))))
1610 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1611 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1613 if (vcpu->pvclock_set_guest_stopped_request) {
1614 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1615 vcpu->pvclock_set_guest_stopped_request = false;
1618 /* If the host uses TSC clocksource, then it is stable */
1619 if (use_master_clock)
1620 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1622 vcpu->hv_clock.flags = pvclock_flags;
1624 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1626 sizeof(vcpu->hv_clock));
1631 * kvmclock updates which are isolated to a given vcpu, such as
1632 * vcpu->cpu migration, should not allow system_timestamp from
1633 * the rest of the vcpus to remain static. Otherwise ntp frequency
1634 * correction applies to one vcpu's system_timestamp but not
1637 * So in those cases, request a kvmclock update for all vcpus.
1638 * We need to rate-limit these requests though, as they can
1639 * considerably slow guests that have a large number of vcpus.
1640 * The time for a remote vcpu to update its kvmclock is bound
1641 * by the delay we use to rate-limit the updates.
1644 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1646 static void kvmclock_update_fn(struct work_struct *work)
1649 struct delayed_work *dwork = to_delayed_work(work);
1650 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1651 kvmclock_update_work);
1652 struct kvm *kvm = container_of(ka, struct kvm, arch);
1653 struct kvm_vcpu *vcpu;
1655 kvm_for_each_vcpu(i, vcpu, kvm) {
1656 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
1657 kvm_vcpu_kick(vcpu);
1661 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1663 struct kvm *kvm = v->kvm;
1665 set_bit(KVM_REQ_CLOCK_UPDATE, &v->requests);
1666 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1667 KVMCLOCK_UPDATE_DELAY);
1670 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1672 static void kvmclock_sync_fn(struct work_struct *work)
1674 struct delayed_work *dwork = to_delayed_work(work);
1675 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1676 kvmclock_sync_work);
1677 struct kvm *kvm = container_of(ka, struct kvm, arch);
1679 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1680 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1681 KVMCLOCK_SYNC_PERIOD);
1684 static bool msr_mtrr_valid(unsigned msr)
1687 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1688 case MSR_MTRRfix64K_00000:
1689 case MSR_MTRRfix16K_80000:
1690 case MSR_MTRRfix16K_A0000:
1691 case MSR_MTRRfix4K_C0000:
1692 case MSR_MTRRfix4K_C8000:
1693 case MSR_MTRRfix4K_D0000:
1694 case MSR_MTRRfix4K_D8000:
1695 case MSR_MTRRfix4K_E0000:
1696 case MSR_MTRRfix4K_E8000:
1697 case MSR_MTRRfix4K_F0000:
1698 case MSR_MTRRfix4K_F8000:
1699 case MSR_MTRRdefType:
1700 case MSR_IA32_CR_PAT:
1708 static bool valid_pat_type(unsigned t)
1710 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1713 static bool valid_mtrr_type(unsigned t)
1715 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1718 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1722 if (!msr_mtrr_valid(msr))
1725 if (msr == MSR_IA32_CR_PAT) {
1726 for (i = 0; i < 8; i++)
1727 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1730 } else if (msr == MSR_MTRRdefType) {
1733 return valid_mtrr_type(data & 0xff);
1734 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1735 for (i = 0; i < 8 ; i++)
1736 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1741 /* variable MTRRs */
1742 return valid_mtrr_type(data & 0xff);
1745 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1747 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1749 if (!mtrr_valid(vcpu, msr, data))
1752 if (msr == MSR_MTRRdefType) {
1753 vcpu->arch.mtrr_state.def_type = data;
1754 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1755 } else if (msr == MSR_MTRRfix64K_00000)
1757 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1758 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1759 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1760 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1761 else if (msr == MSR_IA32_CR_PAT)
1762 vcpu->arch.pat = data;
1763 else { /* Variable MTRRs */
1764 int idx, is_mtrr_mask;
1767 idx = (msr - 0x200) / 2;
1768 is_mtrr_mask = msr - 0x200 - 2 * idx;
1771 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1774 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1778 kvm_mmu_reset_context(vcpu);
1782 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1784 u64 mcg_cap = vcpu->arch.mcg_cap;
1785 unsigned bank_num = mcg_cap & 0xff;
1788 case MSR_IA32_MCG_STATUS:
1789 vcpu->arch.mcg_status = data;
1791 case MSR_IA32_MCG_CTL:
1792 if (!(mcg_cap & MCG_CTL_P))
1794 if (data != 0 && data != ~(u64)0)
1796 vcpu->arch.mcg_ctl = data;
1799 if (msr >= MSR_IA32_MC0_CTL &&
1800 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1801 u32 offset = msr - MSR_IA32_MC0_CTL;
1802 /* only 0 or all 1s can be written to IA32_MCi_CTL
1803 * some Linux kernels though clear bit 10 in bank 4 to
1804 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1805 * this to avoid an uncatched #GP in the guest
1807 if ((offset & 0x3) == 0 &&
1808 data != 0 && (data | (1 << 10)) != ~(u64)0)
1810 vcpu->arch.mce_banks[offset] = data;
1818 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1820 struct kvm *kvm = vcpu->kvm;
1821 int lm = is_long_mode(vcpu);
1822 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1823 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1824 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1825 : kvm->arch.xen_hvm_config.blob_size_32;
1826 u32 page_num = data & ~PAGE_MASK;
1827 u64 page_addr = data & PAGE_MASK;
1832 if (page_num >= blob_size)
1835 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1840 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1849 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1851 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1854 static bool kvm_hv_msr_partition_wide(u32 msr)
1858 case HV_X64_MSR_GUEST_OS_ID:
1859 case HV_X64_MSR_HYPERCALL:
1860 case HV_X64_MSR_REFERENCE_TSC:
1861 case HV_X64_MSR_TIME_REF_COUNT:
1869 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1871 struct kvm *kvm = vcpu->kvm;
1874 case HV_X64_MSR_GUEST_OS_ID:
1875 kvm->arch.hv_guest_os_id = data;
1876 /* setting guest os id to zero disables hypercall page */
1877 if (!kvm->arch.hv_guest_os_id)
1878 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1880 case HV_X64_MSR_HYPERCALL: {
1885 /* if guest os id is not set hypercall should remain disabled */
1886 if (!kvm->arch.hv_guest_os_id)
1888 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1889 kvm->arch.hv_hypercall = data;
1892 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1893 addr = gfn_to_hva(kvm, gfn);
1894 if (kvm_is_error_hva(addr))
1896 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1897 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1898 if (__copy_to_user((void __user *)addr, instructions, 4))
1900 kvm->arch.hv_hypercall = data;
1901 mark_page_dirty(kvm, gfn);
1904 case HV_X64_MSR_REFERENCE_TSC: {
1906 HV_REFERENCE_TSC_PAGE tsc_ref;
1907 memset(&tsc_ref, 0, sizeof(tsc_ref));
1908 kvm->arch.hv_tsc_page = data;
1909 if (!(data & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1911 gfn = data >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
1912 if (kvm_write_guest(kvm, data,
1913 &tsc_ref, sizeof(tsc_ref)))
1915 mark_page_dirty(kvm, gfn);
1919 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1920 "data 0x%llx\n", msr, data);
1926 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1929 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1933 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1934 vcpu->arch.hv_vapic = data;
1937 gfn = data >> HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT;
1938 addr = gfn_to_hva(vcpu->kvm, gfn);
1939 if (kvm_is_error_hva(addr))
1941 if (__clear_user((void __user *)addr, PAGE_SIZE))
1943 vcpu->arch.hv_vapic = data;
1944 mark_page_dirty(vcpu->kvm, gfn);
1947 case HV_X64_MSR_EOI:
1948 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1949 case HV_X64_MSR_ICR:
1950 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1951 case HV_X64_MSR_TPR:
1952 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1954 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1955 "data 0x%llx\n", msr, data);
1962 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1964 gpa_t gpa = data & ~0x3f;
1966 /* Bits 2:5 are reserved, Should be zero */
1970 vcpu->arch.apf.msr_val = data;
1972 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1973 kvm_clear_async_pf_completion_queue(vcpu);
1974 kvm_async_pf_hash_reset(vcpu);
1978 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1982 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1983 kvm_async_pf_wakeup_all(vcpu);
1987 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1989 vcpu->arch.pv_time_enabled = false;
1992 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1996 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1999 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
2000 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2001 vcpu->arch.st.accum_steal = delta;
2004 static void record_steal_time(struct kvm_vcpu *vcpu)
2006 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2009 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2010 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2013 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
2014 vcpu->arch.st.steal.version += 2;
2015 vcpu->arch.st.accum_steal = 0;
2017 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2018 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2021 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2024 u32 msr = msr_info->index;
2025 u64 data = msr_info->data;
2028 case MSR_AMD64_NB_CFG:
2029 case MSR_IA32_UCODE_REV:
2030 case MSR_IA32_UCODE_WRITE:
2031 case MSR_VM_HSAVE_PA:
2032 case MSR_AMD64_PATCH_LOADER:
2033 case MSR_AMD64_BU_CFG2:
2037 return set_efer(vcpu, data);
2039 data &= ~(u64)0x40; /* ignore flush filter disable */
2040 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2041 data &= ~(u64)0x8; /* ignore TLB cache disable */
2043 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2048 case MSR_FAM10H_MMIO_CONF_BASE:
2050 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2055 case MSR_IA32_DEBUGCTLMSR:
2057 /* We support the non-activated case already */
2059 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2060 /* Values other than LBR and BTF are vendor-specific,
2061 thus reserved and should throw a #GP */
2064 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2067 case 0x200 ... 0x2ff:
2068 return set_msr_mtrr(vcpu, msr, data);
2069 case MSR_IA32_APICBASE:
2070 return kvm_set_apic_base(vcpu, msr_info);
2071 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2072 return kvm_x2apic_msr_write(vcpu, msr, data);
2073 case MSR_IA32_TSCDEADLINE:
2074 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2076 case MSR_IA32_TSC_ADJUST:
2077 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2078 if (!msr_info->host_initiated) {
2079 u64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2080 kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
2082 vcpu->arch.ia32_tsc_adjust_msr = data;
2085 case MSR_IA32_MISC_ENABLE:
2086 vcpu->arch.ia32_misc_enable_msr = data;
2088 case MSR_KVM_WALL_CLOCK_NEW:
2089 case MSR_KVM_WALL_CLOCK:
2090 vcpu->kvm->arch.wall_clock = data;
2091 kvm_write_wall_clock(vcpu->kvm, data);
2093 case MSR_KVM_SYSTEM_TIME_NEW:
2094 case MSR_KVM_SYSTEM_TIME: {
2096 kvmclock_reset(vcpu);
2098 vcpu->arch.time = data;
2099 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2101 /* we verify if the enable bit is set... */
2105 gpa_offset = data & ~(PAGE_MASK | 1);
2107 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2108 &vcpu->arch.pv_time, data & ~1ULL,
2109 sizeof(struct pvclock_vcpu_time_info)))
2110 vcpu->arch.pv_time_enabled = false;
2112 vcpu->arch.pv_time_enabled = true;
2116 case MSR_KVM_ASYNC_PF_EN:
2117 if (kvm_pv_enable_async_pf(vcpu, data))
2120 case MSR_KVM_STEAL_TIME:
2122 if (unlikely(!sched_info_on()))
2125 if (data & KVM_STEAL_RESERVED_MASK)
2128 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2129 data & KVM_STEAL_VALID_BITS,
2130 sizeof(struct kvm_steal_time)))
2133 vcpu->arch.st.msr_val = data;
2135 if (!(data & KVM_MSR_ENABLED))
2138 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2141 accumulate_steal_time(vcpu);
2144 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2147 case MSR_KVM_PV_EOI_EN:
2148 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2152 case MSR_IA32_MCG_CTL:
2153 case MSR_IA32_MCG_STATUS:
2154 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2155 return set_msr_mce(vcpu, msr, data);
2157 /* Performance counters are not protected by a CPUID bit,
2158 * so we should check all of them in the generic path for the sake of
2159 * cross vendor migration.
2160 * Writing a zero into the event select MSRs disables them,
2161 * which we perfectly emulate ;-). Any other value should be at least
2162 * reported, some guests depend on them.
2164 case MSR_K7_EVNTSEL0:
2165 case MSR_K7_EVNTSEL1:
2166 case MSR_K7_EVNTSEL2:
2167 case MSR_K7_EVNTSEL3:
2169 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2170 "0x%x data 0x%llx\n", msr, data);
2172 /* at least RHEL 4 unconditionally writes to the perfctr registers,
2173 * so we ignore writes to make it happy.
2175 case MSR_K7_PERFCTR0:
2176 case MSR_K7_PERFCTR1:
2177 case MSR_K7_PERFCTR2:
2178 case MSR_K7_PERFCTR3:
2179 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2180 "0x%x data 0x%llx\n", msr, data);
2182 case MSR_P6_PERFCTR0:
2183 case MSR_P6_PERFCTR1:
2185 case MSR_P6_EVNTSEL0:
2186 case MSR_P6_EVNTSEL1:
2187 if (kvm_pmu_msr(vcpu, msr))
2188 return kvm_pmu_set_msr(vcpu, msr_info);
2190 if (pr || data != 0)
2191 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2192 "0x%x data 0x%llx\n", msr, data);
2194 case MSR_K7_CLK_CTL:
2196 * Ignore all writes to this no longer documented MSR.
2197 * Writes are only relevant for old K7 processors,
2198 * all pre-dating SVM, but a recommended workaround from
2199 * AMD for these chips. It is possible to specify the
2200 * affected processor models on the command line, hence
2201 * the need to ignore the workaround.
2204 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2205 if (kvm_hv_msr_partition_wide(msr)) {
2207 mutex_lock(&vcpu->kvm->lock);
2208 r = set_msr_hyperv_pw(vcpu, msr, data);
2209 mutex_unlock(&vcpu->kvm->lock);
2212 return set_msr_hyperv(vcpu, msr, data);
2214 case MSR_IA32_BBL_CR_CTL3:
2215 /* Drop writes to this legacy MSR -- see rdmsr
2216 * counterpart for further detail.
2218 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2220 case MSR_AMD64_OSVW_ID_LENGTH:
2221 if (!guest_cpuid_has_osvw(vcpu))
2223 vcpu->arch.osvw.length = data;
2225 case MSR_AMD64_OSVW_STATUS:
2226 if (!guest_cpuid_has_osvw(vcpu))
2228 vcpu->arch.osvw.status = data;
2231 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2232 return xen_hvm_config(vcpu, data);
2233 if (kvm_pmu_msr(vcpu, msr))
2234 return kvm_pmu_set_msr(vcpu, msr_info);
2236 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2240 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2247 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2251 * Reads an msr value (of 'msr_index') into 'pdata'.
2252 * Returns 0 on success, non-0 otherwise.
2253 * Assumes vcpu_load() was already called.
2255 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2257 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
2260 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2262 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2264 if (!msr_mtrr_valid(msr))
2267 if (msr == MSR_MTRRdefType)
2268 *pdata = vcpu->arch.mtrr_state.def_type +
2269 (vcpu->arch.mtrr_state.enabled << 10);
2270 else if (msr == MSR_MTRRfix64K_00000)
2272 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2273 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2274 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2275 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2276 else if (msr == MSR_IA32_CR_PAT)
2277 *pdata = vcpu->arch.pat;
2278 else { /* Variable MTRRs */
2279 int idx, is_mtrr_mask;
2282 idx = (msr - 0x200) / 2;
2283 is_mtrr_mask = msr - 0x200 - 2 * idx;
2286 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2289 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2296 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2299 u64 mcg_cap = vcpu->arch.mcg_cap;
2300 unsigned bank_num = mcg_cap & 0xff;
2303 case MSR_IA32_P5_MC_ADDR:
2304 case MSR_IA32_P5_MC_TYPE:
2307 case MSR_IA32_MCG_CAP:
2308 data = vcpu->arch.mcg_cap;
2310 case MSR_IA32_MCG_CTL:
2311 if (!(mcg_cap & MCG_CTL_P))
2313 data = vcpu->arch.mcg_ctl;
2315 case MSR_IA32_MCG_STATUS:
2316 data = vcpu->arch.mcg_status;
2319 if (msr >= MSR_IA32_MC0_CTL &&
2320 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
2321 u32 offset = msr - MSR_IA32_MC0_CTL;
2322 data = vcpu->arch.mce_banks[offset];
2331 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2334 struct kvm *kvm = vcpu->kvm;
2337 case HV_X64_MSR_GUEST_OS_ID:
2338 data = kvm->arch.hv_guest_os_id;
2340 case HV_X64_MSR_HYPERCALL:
2341 data = kvm->arch.hv_hypercall;
2343 case HV_X64_MSR_TIME_REF_COUNT: {
2345 div_u64(get_kernel_ns() + kvm->arch.kvmclock_offset, 100);
2348 case HV_X64_MSR_REFERENCE_TSC:
2349 data = kvm->arch.hv_tsc_page;
2352 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2360 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2365 case HV_X64_MSR_VP_INDEX: {
2368 kvm_for_each_vcpu(r, v, vcpu->kvm) {
2376 case HV_X64_MSR_EOI:
2377 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2378 case HV_X64_MSR_ICR:
2379 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2380 case HV_X64_MSR_TPR:
2381 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
2382 case HV_X64_MSR_APIC_ASSIST_PAGE:
2383 data = vcpu->arch.hv_vapic;
2386 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2393 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2398 case MSR_IA32_PLATFORM_ID:
2399 case MSR_IA32_EBL_CR_POWERON:
2400 case MSR_IA32_DEBUGCTLMSR:
2401 case MSR_IA32_LASTBRANCHFROMIP:
2402 case MSR_IA32_LASTBRANCHTOIP:
2403 case MSR_IA32_LASTINTFROMIP:
2404 case MSR_IA32_LASTINTTOIP:
2407 case MSR_VM_HSAVE_PA:
2408 case MSR_K7_EVNTSEL0:
2409 case MSR_K7_PERFCTR0:
2410 case MSR_K8_INT_PENDING_MSG:
2411 case MSR_AMD64_NB_CFG:
2412 case MSR_FAM10H_MMIO_CONF_BASE:
2413 case MSR_AMD64_BU_CFG2:
2416 case MSR_P6_PERFCTR0:
2417 case MSR_P6_PERFCTR1:
2418 case MSR_P6_EVNTSEL0:
2419 case MSR_P6_EVNTSEL1:
2420 if (kvm_pmu_msr(vcpu, msr))
2421 return kvm_pmu_get_msr(vcpu, msr, pdata);
2424 case MSR_IA32_UCODE_REV:
2425 data = 0x100000000ULL;
2428 data = 0x500 | KVM_NR_VAR_MTRR;
2430 case 0x200 ... 0x2ff:
2431 return get_msr_mtrr(vcpu, msr, pdata);
2432 case 0xcd: /* fsb frequency */
2436 * MSR_EBC_FREQUENCY_ID
2437 * Conservative value valid for even the basic CPU models.
2438 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2439 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2440 * and 266MHz for model 3, or 4. Set Core Clock
2441 * Frequency to System Bus Frequency Ratio to 1 (bits
2442 * 31:24) even though these are only valid for CPU
2443 * models > 2, however guests may end up dividing or
2444 * multiplying by zero otherwise.
2446 case MSR_EBC_FREQUENCY_ID:
2449 case MSR_IA32_APICBASE:
2450 data = kvm_get_apic_base(vcpu);
2452 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2453 return kvm_x2apic_msr_read(vcpu, msr, pdata);
2455 case MSR_IA32_TSCDEADLINE:
2456 data = kvm_get_lapic_tscdeadline_msr(vcpu);
2458 case MSR_IA32_TSC_ADJUST:
2459 data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2461 case MSR_IA32_MISC_ENABLE:
2462 data = vcpu->arch.ia32_misc_enable_msr;
2464 case MSR_IA32_PERF_STATUS:
2465 /* TSC increment by tick */
2467 /* CPU multiplier */
2468 data |= (((uint64_t)4ULL) << 40);
2471 data = vcpu->arch.efer;
2473 case MSR_KVM_WALL_CLOCK:
2474 case MSR_KVM_WALL_CLOCK_NEW:
2475 data = vcpu->kvm->arch.wall_clock;
2477 case MSR_KVM_SYSTEM_TIME:
2478 case MSR_KVM_SYSTEM_TIME_NEW:
2479 data = vcpu->arch.time;
2481 case MSR_KVM_ASYNC_PF_EN:
2482 data = vcpu->arch.apf.msr_val;
2484 case MSR_KVM_STEAL_TIME:
2485 data = vcpu->arch.st.msr_val;
2487 case MSR_KVM_PV_EOI_EN:
2488 data = vcpu->arch.pv_eoi.msr_val;
2490 case MSR_IA32_P5_MC_ADDR:
2491 case MSR_IA32_P5_MC_TYPE:
2492 case MSR_IA32_MCG_CAP:
2493 case MSR_IA32_MCG_CTL:
2494 case MSR_IA32_MCG_STATUS:
2495 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2496 return get_msr_mce(vcpu, msr, pdata);
2497 case MSR_K7_CLK_CTL:
2499 * Provide expected ramp-up count for K7. All other
2500 * are set to zero, indicating minimum divisors for
2503 * This prevents guest kernels on AMD host with CPU
2504 * type 6, model 8 and higher from exploding due to
2505 * the rdmsr failing.
2509 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2510 if (kvm_hv_msr_partition_wide(msr)) {
2512 mutex_lock(&vcpu->kvm->lock);
2513 r = get_msr_hyperv_pw(vcpu, msr, pdata);
2514 mutex_unlock(&vcpu->kvm->lock);
2517 return get_msr_hyperv(vcpu, msr, pdata);
2519 case MSR_IA32_BBL_CR_CTL3:
2520 /* This legacy MSR exists but isn't fully documented in current
2521 * silicon. It is however accessed by winxp in very narrow
2522 * scenarios where it sets bit #19, itself documented as
2523 * a "reserved" bit. Best effort attempt to source coherent
2524 * read data here should the balance of the register be
2525 * interpreted by the guest:
2527 * L2 cache control register 3: 64GB range, 256KB size,
2528 * enabled, latency 0x1, configured
2532 case MSR_AMD64_OSVW_ID_LENGTH:
2533 if (!guest_cpuid_has_osvw(vcpu))
2535 data = vcpu->arch.osvw.length;
2537 case MSR_AMD64_OSVW_STATUS:
2538 if (!guest_cpuid_has_osvw(vcpu))
2540 data = vcpu->arch.osvw.status;
2543 if (kvm_pmu_msr(vcpu, msr))
2544 return kvm_pmu_get_msr(vcpu, msr, pdata);
2546 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2549 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2557 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2560 * Read or write a bunch of msrs. All parameters are kernel addresses.
2562 * @return number of msrs set successfully.
2564 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2565 struct kvm_msr_entry *entries,
2566 int (*do_msr)(struct kvm_vcpu *vcpu,
2567 unsigned index, u64 *data))
2571 idx = srcu_read_lock(&vcpu->kvm->srcu);
2572 for (i = 0; i < msrs->nmsrs; ++i)
2573 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2575 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2581 * Read or write a bunch of msrs. Parameters are user addresses.
2583 * @return number of msrs set successfully.
2585 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2586 int (*do_msr)(struct kvm_vcpu *vcpu,
2587 unsigned index, u64 *data),
2590 struct kvm_msrs msrs;
2591 struct kvm_msr_entry *entries;
2596 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2600 if (msrs.nmsrs >= MAX_IO_MSRS)
2603 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2604 entries = memdup_user(user_msrs->entries, size);
2605 if (IS_ERR(entries)) {
2606 r = PTR_ERR(entries);
2610 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2615 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2626 int kvm_dev_ioctl_check_extension(long ext)
2631 case KVM_CAP_IRQCHIP:
2633 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2634 case KVM_CAP_SET_TSS_ADDR:
2635 case KVM_CAP_EXT_CPUID:
2636 case KVM_CAP_EXT_EMUL_CPUID:
2637 case KVM_CAP_CLOCKSOURCE:
2639 case KVM_CAP_NOP_IO_DELAY:
2640 case KVM_CAP_MP_STATE:
2641 case KVM_CAP_SYNC_MMU:
2642 case KVM_CAP_USER_NMI:
2643 case KVM_CAP_REINJECT_CONTROL:
2644 case KVM_CAP_IRQ_INJECT_STATUS:
2646 case KVM_CAP_IOEVENTFD:
2648 case KVM_CAP_PIT_STATE2:
2649 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2650 case KVM_CAP_XEN_HVM:
2651 case KVM_CAP_ADJUST_CLOCK:
2652 case KVM_CAP_VCPU_EVENTS:
2653 case KVM_CAP_HYPERV:
2654 case KVM_CAP_HYPERV_VAPIC:
2655 case KVM_CAP_HYPERV_SPIN:
2656 case KVM_CAP_PCI_SEGMENT:
2657 case KVM_CAP_DEBUGREGS:
2658 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2660 case KVM_CAP_ASYNC_PF:
2661 case KVM_CAP_GET_TSC_KHZ:
2662 case KVM_CAP_KVMCLOCK_CTRL:
2663 case KVM_CAP_READONLY_MEM:
2664 case KVM_CAP_HYPERV_TIME:
2665 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2666 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2667 case KVM_CAP_ASSIGN_DEV_IRQ:
2668 case KVM_CAP_PCI_2_3:
2672 case KVM_CAP_COALESCED_MMIO:
2673 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2676 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2678 case KVM_CAP_NR_VCPUS:
2679 r = KVM_SOFT_MAX_VCPUS;
2681 case KVM_CAP_MAX_VCPUS:
2684 case KVM_CAP_NR_MEMSLOTS:
2685 r = KVM_USER_MEM_SLOTS;
2687 case KVM_CAP_PV_MMU: /* obsolete */
2690 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2692 r = iommu_present(&pci_bus_type);
2696 r = KVM_MAX_MCE_BANKS;
2701 case KVM_CAP_TSC_CONTROL:
2702 r = kvm_has_tsc_control;
2704 case KVM_CAP_TSC_DEADLINE_TIMER:
2705 r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2715 long kvm_arch_dev_ioctl(struct file *filp,
2716 unsigned int ioctl, unsigned long arg)
2718 void __user *argp = (void __user *)arg;
2722 case KVM_GET_MSR_INDEX_LIST: {
2723 struct kvm_msr_list __user *user_msr_list = argp;
2724 struct kvm_msr_list msr_list;
2728 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2731 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2732 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2735 if (n < msr_list.nmsrs)
2738 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2739 num_msrs_to_save * sizeof(u32)))
2741 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2743 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2748 case KVM_GET_SUPPORTED_CPUID:
2749 case KVM_GET_EMULATED_CPUID: {
2750 struct kvm_cpuid2 __user *cpuid_arg = argp;
2751 struct kvm_cpuid2 cpuid;
2754 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2757 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2763 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2768 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2771 mce_cap = KVM_MCE_CAP_SUPPORTED;
2773 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2785 static void wbinvd_ipi(void *garbage)
2790 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2792 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2795 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2797 /* Address WBINVD may be executed by guest */
2798 if (need_emulate_wbinvd(vcpu)) {
2799 if (kvm_x86_ops->has_wbinvd_exit())
2800 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2801 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2802 smp_call_function_single(vcpu->cpu,
2803 wbinvd_ipi, NULL, 1);
2806 kvm_x86_ops->vcpu_load(vcpu, cpu);
2808 /* Apply any externally detected TSC adjustments (due to suspend) */
2809 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2810 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2811 vcpu->arch.tsc_offset_adjustment = 0;
2812 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
2815 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2816 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2817 native_read_tsc() - vcpu->arch.last_host_tsc;
2819 mark_tsc_unstable("KVM discovered backwards TSC");
2820 if (check_tsc_unstable()) {
2821 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2822 vcpu->arch.last_guest_tsc);
2823 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2824 vcpu->arch.tsc_catchup = 1;
2827 * On a host with synchronized TSC, there is no need to update
2828 * kvmclock on vcpu->cpu migration
2830 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2831 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2832 if (vcpu->cpu != cpu)
2833 kvm_migrate_timers(vcpu);
2837 accumulate_steal_time(vcpu);
2838 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2841 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2843 kvm_x86_ops->vcpu_put(vcpu);
2844 kvm_put_guest_fpu(vcpu);
2845 vcpu->arch.last_host_tsc = native_read_tsc();
2848 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2849 struct kvm_lapic_state *s)
2851 kvm_x86_ops->sync_pir_to_irr(vcpu);
2852 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2857 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2858 struct kvm_lapic_state *s)
2860 kvm_apic_post_state_restore(vcpu, s);
2861 update_cr8_intercept(vcpu);
2866 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2867 struct kvm_interrupt *irq)
2869 if (irq->irq >= KVM_NR_INTERRUPTS)
2871 if (irqchip_in_kernel(vcpu->kvm))
2874 kvm_queue_interrupt(vcpu, irq->irq, false);
2875 kvm_make_request(KVM_REQ_EVENT, vcpu);
2880 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2882 kvm_inject_nmi(vcpu);
2887 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2888 struct kvm_tpr_access_ctl *tac)
2892 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2896 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2900 unsigned bank_num = mcg_cap & 0xff, bank;
2903 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2905 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2908 vcpu->arch.mcg_cap = mcg_cap;
2909 /* Init IA32_MCG_CTL to all 1s */
2910 if (mcg_cap & MCG_CTL_P)
2911 vcpu->arch.mcg_ctl = ~(u64)0;
2912 /* Init IA32_MCi_CTL to all 1s */
2913 for (bank = 0; bank < bank_num; bank++)
2914 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2919 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2920 struct kvm_x86_mce *mce)
2922 u64 mcg_cap = vcpu->arch.mcg_cap;
2923 unsigned bank_num = mcg_cap & 0xff;
2924 u64 *banks = vcpu->arch.mce_banks;
2926 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2929 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2930 * reporting is disabled
2932 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2933 vcpu->arch.mcg_ctl != ~(u64)0)
2935 banks += 4 * mce->bank;
2937 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2938 * reporting is disabled for the bank
2940 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2942 if (mce->status & MCI_STATUS_UC) {
2943 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2944 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2945 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2948 if (banks[1] & MCI_STATUS_VAL)
2949 mce->status |= MCI_STATUS_OVER;
2950 banks[2] = mce->addr;
2951 banks[3] = mce->misc;
2952 vcpu->arch.mcg_status = mce->mcg_status;
2953 banks[1] = mce->status;
2954 kvm_queue_exception(vcpu, MC_VECTOR);
2955 } else if (!(banks[1] & MCI_STATUS_VAL)
2956 || !(banks[1] & MCI_STATUS_UC)) {
2957 if (banks[1] & MCI_STATUS_VAL)
2958 mce->status |= MCI_STATUS_OVER;
2959 banks[2] = mce->addr;
2960 banks[3] = mce->misc;
2961 banks[1] = mce->status;
2963 banks[1] |= MCI_STATUS_OVER;
2967 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2968 struct kvm_vcpu_events *events)
2971 events->exception.injected =
2972 vcpu->arch.exception.pending &&
2973 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2974 events->exception.nr = vcpu->arch.exception.nr;
2975 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2976 events->exception.pad = 0;
2977 events->exception.error_code = vcpu->arch.exception.error_code;
2979 events->interrupt.injected =
2980 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2981 events->interrupt.nr = vcpu->arch.interrupt.nr;
2982 events->interrupt.soft = 0;
2983 events->interrupt.shadow =
2984 kvm_x86_ops->get_interrupt_shadow(vcpu,
2985 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2987 events->nmi.injected = vcpu->arch.nmi_injected;
2988 events->nmi.pending = vcpu->arch.nmi_pending != 0;
2989 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2990 events->nmi.pad = 0;
2992 events->sipi_vector = 0; /* never valid when reporting to user space */
2994 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2995 | KVM_VCPUEVENT_VALID_SHADOW);
2996 memset(&events->reserved, 0, sizeof(events->reserved));
2999 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3000 struct kvm_vcpu_events *events)
3002 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3003 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3004 | KVM_VCPUEVENT_VALID_SHADOW))
3008 vcpu->arch.exception.pending = events->exception.injected;
3009 vcpu->arch.exception.nr = events->exception.nr;
3010 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3011 vcpu->arch.exception.error_code = events->exception.error_code;
3013 vcpu->arch.interrupt.pending = events->interrupt.injected;
3014 vcpu->arch.interrupt.nr = events->interrupt.nr;
3015 vcpu->arch.interrupt.soft = events->interrupt.soft;
3016 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3017 kvm_x86_ops->set_interrupt_shadow(vcpu,
3018 events->interrupt.shadow);
3020 vcpu->arch.nmi_injected = events->nmi.injected;
3021 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3022 vcpu->arch.nmi_pending = events->nmi.pending;
3023 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3025 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3026 kvm_vcpu_has_lapic(vcpu))
3027 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3029 kvm_make_request(KVM_REQ_EVENT, vcpu);
3034 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3035 struct kvm_debugregs *dbgregs)
3039 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3040 _kvm_get_dr(vcpu, 6, &val);
3042 dbgregs->dr7 = vcpu->arch.dr7;
3044 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3047 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3048 struct kvm_debugregs *dbgregs)
3053 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3054 vcpu->arch.dr6 = dbgregs->dr6;
3055 kvm_update_dr6(vcpu);
3056 vcpu->arch.dr7 = dbgregs->dr7;
3057 kvm_update_dr7(vcpu);
3062 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3063 struct kvm_xsave *guest_xsave)
3065 if (cpu_has_xsave) {
3066 memcpy(guest_xsave->region,
3067 &vcpu->arch.guest_fpu.state->xsave,
3068 vcpu->arch.guest_xstate_size);
3069 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] &=
3070 vcpu->arch.guest_supported_xcr0 | XSTATE_FPSSE;
3072 memcpy(guest_xsave->region,
3073 &vcpu->arch.guest_fpu.state->fxsave,
3074 sizeof(struct i387_fxsave_struct));
3075 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3080 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3081 struct kvm_xsave *guest_xsave)
3084 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3086 if (cpu_has_xsave) {
3088 * Here we allow setting states that are not present in
3089 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3090 * with old userspace.
3092 if (xstate_bv & ~kvm_supported_xcr0())
3094 memcpy(&vcpu->arch.guest_fpu.state->xsave,
3095 guest_xsave->region, vcpu->arch.guest_xstate_size);
3097 if (xstate_bv & ~XSTATE_FPSSE)
3099 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
3100 guest_xsave->region, sizeof(struct i387_fxsave_struct));
3105 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3106 struct kvm_xcrs *guest_xcrs)
3108 if (!cpu_has_xsave) {
3109 guest_xcrs->nr_xcrs = 0;
3113 guest_xcrs->nr_xcrs = 1;
3114 guest_xcrs->flags = 0;
3115 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3116 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3119 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3120 struct kvm_xcrs *guest_xcrs)
3127 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3130 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3131 /* Only support XCR0 currently */
3132 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3133 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3134 guest_xcrs->xcrs[i].value);
3143 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3144 * stopped by the hypervisor. This function will be called from the host only.
3145 * EINVAL is returned when the host attempts to set the flag for a guest that
3146 * does not support pv clocks.
3148 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3150 if (!vcpu->arch.pv_time_enabled)
3152 vcpu->arch.pvclock_set_guest_stopped_request = true;
3153 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3157 long kvm_arch_vcpu_ioctl(struct file *filp,
3158 unsigned int ioctl, unsigned long arg)
3160 struct kvm_vcpu *vcpu = filp->private_data;
3161 void __user *argp = (void __user *)arg;
3164 struct kvm_lapic_state *lapic;
3165 struct kvm_xsave *xsave;
3166 struct kvm_xcrs *xcrs;
3172 case KVM_GET_LAPIC: {
3174 if (!vcpu->arch.apic)
3176 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3181 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3185 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3190 case KVM_SET_LAPIC: {
3192 if (!vcpu->arch.apic)
3194 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3195 if (IS_ERR(u.lapic))
3196 return PTR_ERR(u.lapic);
3198 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3201 case KVM_INTERRUPT: {
3202 struct kvm_interrupt irq;
3205 if (copy_from_user(&irq, argp, sizeof irq))
3207 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3211 r = kvm_vcpu_ioctl_nmi(vcpu);
3214 case KVM_SET_CPUID: {
3215 struct kvm_cpuid __user *cpuid_arg = argp;
3216 struct kvm_cpuid cpuid;
3219 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3221 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3224 case KVM_SET_CPUID2: {
3225 struct kvm_cpuid2 __user *cpuid_arg = argp;
3226 struct kvm_cpuid2 cpuid;
3229 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3231 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3232 cpuid_arg->entries);
3235 case KVM_GET_CPUID2: {
3236 struct kvm_cpuid2 __user *cpuid_arg = argp;
3237 struct kvm_cpuid2 cpuid;
3240 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3242 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3243 cpuid_arg->entries);
3247 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3253 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3256 r = msr_io(vcpu, argp, do_set_msr, 0);
3258 case KVM_TPR_ACCESS_REPORTING: {
3259 struct kvm_tpr_access_ctl tac;
3262 if (copy_from_user(&tac, argp, sizeof tac))
3264 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3268 if (copy_to_user(argp, &tac, sizeof tac))
3273 case KVM_SET_VAPIC_ADDR: {
3274 struct kvm_vapic_addr va;
3277 if (!irqchip_in_kernel(vcpu->kvm))
3280 if (copy_from_user(&va, argp, sizeof va))
3282 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3285 case KVM_X86_SETUP_MCE: {
3289 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3291 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3294 case KVM_X86_SET_MCE: {
3295 struct kvm_x86_mce mce;
3298 if (copy_from_user(&mce, argp, sizeof mce))
3300 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3303 case KVM_GET_VCPU_EVENTS: {
3304 struct kvm_vcpu_events events;
3306 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3309 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3314 case KVM_SET_VCPU_EVENTS: {
3315 struct kvm_vcpu_events events;
3318 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3321 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3324 case KVM_GET_DEBUGREGS: {
3325 struct kvm_debugregs dbgregs;
3327 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3330 if (copy_to_user(argp, &dbgregs,
3331 sizeof(struct kvm_debugregs)))
3336 case KVM_SET_DEBUGREGS: {
3337 struct kvm_debugregs dbgregs;
3340 if (copy_from_user(&dbgregs, argp,
3341 sizeof(struct kvm_debugregs)))
3344 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3347 case KVM_GET_XSAVE: {
3348 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3353 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3356 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3361 case KVM_SET_XSAVE: {
3362 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3363 if (IS_ERR(u.xsave))
3364 return PTR_ERR(u.xsave);
3366 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3369 case KVM_GET_XCRS: {
3370 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3375 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3378 if (copy_to_user(argp, u.xcrs,
3379 sizeof(struct kvm_xcrs)))
3384 case KVM_SET_XCRS: {
3385 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3387 return PTR_ERR(u.xcrs);
3389 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3392 case KVM_SET_TSC_KHZ: {
3396 user_tsc_khz = (u32)arg;
3398 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3401 if (user_tsc_khz == 0)
3402 user_tsc_khz = tsc_khz;
3404 kvm_set_tsc_khz(vcpu, user_tsc_khz);
3409 case KVM_GET_TSC_KHZ: {
3410 r = vcpu->arch.virtual_tsc_khz;
3413 case KVM_KVMCLOCK_CTRL: {
3414 r = kvm_set_guest_paused(vcpu);
3425 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3427 return VM_FAULT_SIGBUS;
3430 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3434 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3436 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3440 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3443 kvm->arch.ept_identity_map_addr = ident_addr;
3447 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3448 u32 kvm_nr_mmu_pages)
3450 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3453 mutex_lock(&kvm->slots_lock);
3455 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3456 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3458 mutex_unlock(&kvm->slots_lock);
3462 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3464 return kvm->arch.n_max_mmu_pages;
3467 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3472 switch (chip->chip_id) {
3473 case KVM_IRQCHIP_PIC_MASTER:
3474 memcpy(&chip->chip.pic,
3475 &pic_irqchip(kvm)->pics[0],
3476 sizeof(struct kvm_pic_state));
3478 case KVM_IRQCHIP_PIC_SLAVE:
3479 memcpy(&chip->chip.pic,
3480 &pic_irqchip(kvm)->pics[1],
3481 sizeof(struct kvm_pic_state));
3483 case KVM_IRQCHIP_IOAPIC:
3484 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3493 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3498 switch (chip->chip_id) {
3499 case KVM_IRQCHIP_PIC_MASTER:
3500 spin_lock(&pic_irqchip(kvm)->lock);
3501 memcpy(&pic_irqchip(kvm)->pics[0],
3503 sizeof(struct kvm_pic_state));
3504 spin_unlock(&pic_irqchip(kvm)->lock);
3506 case KVM_IRQCHIP_PIC_SLAVE:
3507 spin_lock(&pic_irqchip(kvm)->lock);
3508 memcpy(&pic_irqchip(kvm)->pics[1],
3510 sizeof(struct kvm_pic_state));
3511 spin_unlock(&pic_irqchip(kvm)->lock);
3513 case KVM_IRQCHIP_IOAPIC:
3514 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3520 kvm_pic_update_irq(pic_irqchip(kvm));
3524 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3528 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3529 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3530 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3534 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3538 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3539 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3540 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3541 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3545 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3549 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3550 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3551 sizeof(ps->channels));
3552 ps->flags = kvm->arch.vpit->pit_state.flags;
3553 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3554 memset(&ps->reserved, 0, sizeof(ps->reserved));
3558 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3560 int r = 0, start = 0;
3561 u32 prev_legacy, cur_legacy;
3562 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3563 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3564 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3565 if (!prev_legacy && cur_legacy)
3567 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3568 sizeof(kvm->arch.vpit->pit_state.channels));
3569 kvm->arch.vpit->pit_state.flags = ps->flags;
3570 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3571 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3575 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3576 struct kvm_reinject_control *control)
3578 if (!kvm->arch.vpit)
3580 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3581 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3582 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3587 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3588 * @kvm: kvm instance
3589 * @log: slot id and address to which we copy the log
3591 * We need to keep it in mind that VCPU threads can write to the bitmap
3592 * concurrently. So, to avoid losing data, we keep the following order for
3595 * 1. Take a snapshot of the bit and clear it if needed.
3596 * 2. Write protect the corresponding page.
3597 * 3. Flush TLB's if needed.
3598 * 4. Copy the snapshot to the userspace.
3600 * Between 2 and 3, the guest may write to the page using the remaining TLB
3601 * entry. This is not a problem because the page will be reported dirty at
3602 * step 4 using the snapshot taken before and step 3 ensures that successive
3603 * writes will be logged for the next call.
3605 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3608 struct kvm_memory_slot *memslot;
3610 unsigned long *dirty_bitmap;
3611 unsigned long *dirty_bitmap_buffer;
3612 bool is_dirty = false;
3614 mutex_lock(&kvm->slots_lock);
3617 if (log->slot >= KVM_USER_MEM_SLOTS)
3620 memslot = id_to_memslot(kvm->memslots, log->slot);
3622 dirty_bitmap = memslot->dirty_bitmap;
3627 n = kvm_dirty_bitmap_bytes(memslot);
3629 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
3630 memset(dirty_bitmap_buffer, 0, n);
3632 spin_lock(&kvm->mmu_lock);
3634 for (i = 0; i < n / sizeof(long); i++) {
3638 if (!dirty_bitmap[i])
3643 mask = xchg(&dirty_bitmap[i], 0);
3644 dirty_bitmap_buffer[i] = mask;
3646 offset = i * BITS_PER_LONG;
3647 kvm_mmu_write_protect_pt_masked(kvm, memslot, offset, mask);
3650 kvm_flush_remote_tlbs(kvm);
3652 spin_unlock(&kvm->mmu_lock);
3655 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
3660 mutex_unlock(&kvm->slots_lock);
3664 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3667 if (!irqchip_in_kernel(kvm))
3670 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3671 irq_event->irq, irq_event->level,
3676 long kvm_arch_vm_ioctl(struct file *filp,
3677 unsigned int ioctl, unsigned long arg)
3679 struct kvm *kvm = filp->private_data;
3680 void __user *argp = (void __user *)arg;
3683 * This union makes it completely explicit to gcc-3.x
3684 * that these two variables' stack usage should be
3685 * combined, not added together.
3688 struct kvm_pit_state ps;
3689 struct kvm_pit_state2 ps2;
3690 struct kvm_pit_config pit_config;
3694 case KVM_SET_TSS_ADDR:
3695 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3697 case KVM_SET_IDENTITY_MAP_ADDR: {
3701 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3703 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3706 case KVM_SET_NR_MMU_PAGES:
3707 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3709 case KVM_GET_NR_MMU_PAGES:
3710 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3712 case KVM_CREATE_IRQCHIP: {
3713 struct kvm_pic *vpic;
3715 mutex_lock(&kvm->lock);
3718 goto create_irqchip_unlock;
3720 if (atomic_read(&kvm->online_vcpus))
3721 goto create_irqchip_unlock;
3723 vpic = kvm_create_pic(kvm);
3725 r = kvm_ioapic_init(kvm);
3727 mutex_lock(&kvm->slots_lock);
3728 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3730 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3732 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3734 mutex_unlock(&kvm->slots_lock);
3736 goto create_irqchip_unlock;
3739 goto create_irqchip_unlock;
3741 kvm->arch.vpic = vpic;
3743 r = kvm_setup_default_irq_routing(kvm);
3745 mutex_lock(&kvm->slots_lock);
3746 mutex_lock(&kvm->irq_lock);
3747 kvm_ioapic_destroy(kvm);
3748 kvm_destroy_pic(kvm);
3749 mutex_unlock(&kvm->irq_lock);
3750 mutex_unlock(&kvm->slots_lock);
3752 create_irqchip_unlock:
3753 mutex_unlock(&kvm->lock);
3756 case KVM_CREATE_PIT:
3757 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3759 case KVM_CREATE_PIT2:
3761 if (copy_from_user(&u.pit_config, argp,
3762 sizeof(struct kvm_pit_config)))
3765 mutex_lock(&kvm->slots_lock);
3768 goto create_pit_unlock;
3770 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3774 mutex_unlock(&kvm->slots_lock);
3776 case KVM_GET_IRQCHIP: {
3777 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3778 struct kvm_irqchip *chip;
3780 chip = memdup_user(argp, sizeof(*chip));
3787 if (!irqchip_in_kernel(kvm))
3788 goto get_irqchip_out;
3789 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3791 goto get_irqchip_out;
3793 if (copy_to_user(argp, chip, sizeof *chip))
3794 goto get_irqchip_out;
3800 case KVM_SET_IRQCHIP: {
3801 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3802 struct kvm_irqchip *chip;
3804 chip = memdup_user(argp, sizeof(*chip));
3811 if (!irqchip_in_kernel(kvm))
3812 goto set_irqchip_out;
3813 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3815 goto set_irqchip_out;
3823 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3826 if (!kvm->arch.vpit)
3828 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3832 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3839 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3842 if (!kvm->arch.vpit)
3844 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3847 case KVM_GET_PIT2: {
3849 if (!kvm->arch.vpit)
3851 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3855 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3860 case KVM_SET_PIT2: {
3862 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3865 if (!kvm->arch.vpit)
3867 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3870 case KVM_REINJECT_CONTROL: {
3871 struct kvm_reinject_control control;
3873 if (copy_from_user(&control, argp, sizeof(control)))
3875 r = kvm_vm_ioctl_reinject(kvm, &control);
3878 case KVM_XEN_HVM_CONFIG: {
3880 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3881 sizeof(struct kvm_xen_hvm_config)))
3884 if (kvm->arch.xen_hvm_config.flags)
3889 case KVM_SET_CLOCK: {
3890 struct kvm_clock_data user_ns;
3895 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3903 local_irq_disable();
3904 now_ns = get_kernel_ns();
3905 delta = user_ns.clock - now_ns;
3907 kvm->arch.kvmclock_offset = delta;
3908 kvm_gen_update_masterclock(kvm);
3911 case KVM_GET_CLOCK: {
3912 struct kvm_clock_data user_ns;
3915 local_irq_disable();
3916 now_ns = get_kernel_ns();
3917 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3920 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3923 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3936 static void kvm_init_msr_list(void)
3941 /* skip the first msrs in the list. KVM-specific */
3942 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3943 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3947 * Even MSRs that are valid in the host may not be exposed
3948 * to the guests in some cases. We could work around this
3949 * in VMX with the generic MSR save/load machinery, but it
3950 * is not really worthwhile since it will really only
3951 * happen with nested virtualization.
3953 switch (msrs_to_save[i]) {
3954 case MSR_IA32_BNDCFGS:
3955 if (!kvm_x86_ops->mpx_supported())
3963 msrs_to_save[j] = msrs_to_save[i];
3966 num_msrs_to_save = j;
3969 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3977 if (!(vcpu->arch.apic &&
3978 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3979 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3990 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3997 if (!(vcpu->arch.apic &&
3998 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3999 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
4001 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4011 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4012 struct kvm_segment *var, int seg)
4014 kvm_x86_ops->set_segment(vcpu, var, seg);
4017 void kvm_get_segment(struct kvm_vcpu *vcpu,
4018 struct kvm_segment *var, int seg)
4020 kvm_x86_ops->get_segment(vcpu, var, seg);
4023 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
4026 struct x86_exception exception;
4028 BUG_ON(!mmu_is_nested(vcpu));
4030 /* NPT walks are always user-walks */
4031 access |= PFERR_USER_MASK;
4032 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
4037 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4038 struct x86_exception *exception)
4040 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4041 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4044 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4045 struct x86_exception *exception)
4047 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4048 access |= PFERR_FETCH_MASK;
4049 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4052 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4053 struct x86_exception *exception)
4055 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4056 access |= PFERR_WRITE_MASK;
4057 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4060 /* uses this to access any guest's mapped memory without checking CPL */
4061 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4062 struct x86_exception *exception)
4064 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4067 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4068 struct kvm_vcpu *vcpu, u32 access,
4069 struct x86_exception *exception)
4072 int r = X86EMUL_CONTINUE;
4075 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4077 unsigned offset = addr & (PAGE_SIZE-1);
4078 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4081 if (gpa == UNMAPPED_GVA)
4082 return X86EMUL_PROPAGATE_FAULT;
4083 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
4085 r = X86EMUL_IO_NEEDED;
4097 /* used for instruction fetching */
4098 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4099 gva_t addr, void *val, unsigned int bytes,
4100 struct x86_exception *exception)
4102 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4103 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4105 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
4106 access | PFERR_FETCH_MASK,
4110 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4111 gva_t addr, void *val, unsigned int bytes,
4112 struct x86_exception *exception)
4114 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4115 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4117 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4120 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4122 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4123 gva_t addr, void *val, unsigned int bytes,
4124 struct x86_exception *exception)
4126 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4127 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4130 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4131 gva_t addr, void *val,
4133 struct x86_exception *exception)
4135 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4137 int r = X86EMUL_CONTINUE;
4140 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4143 unsigned offset = addr & (PAGE_SIZE-1);
4144 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4147 if (gpa == UNMAPPED_GVA)
4148 return X86EMUL_PROPAGATE_FAULT;
4149 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
4151 r = X86EMUL_IO_NEEDED;
4162 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4164 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4165 gpa_t *gpa, struct x86_exception *exception,
4168 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4169 | (write ? PFERR_WRITE_MASK : 0);
4171 if (vcpu_match_mmio_gva(vcpu, gva)
4172 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4173 vcpu->arch.access, access)) {
4174 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4175 (gva & (PAGE_SIZE - 1));
4176 trace_vcpu_match_mmio(gva, *gpa, write, false);
4180 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4182 if (*gpa == UNMAPPED_GVA)
4185 /* For APIC access vmexit */
4186 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4189 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4190 trace_vcpu_match_mmio(gva, *gpa, write, true);
4197 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4198 const void *val, int bytes)
4202 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4205 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4209 struct read_write_emulator_ops {
4210 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4212 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4213 void *val, int bytes);
4214 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4215 int bytes, void *val);
4216 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4217 void *val, int bytes);
4221 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4223 if (vcpu->mmio_read_completed) {
4224 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4225 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4226 vcpu->mmio_read_completed = 0;
4233 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4234 void *val, int bytes)
4236 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4239 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4240 void *val, int bytes)
4242 return emulator_write_phys(vcpu, gpa, val, bytes);
4245 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4247 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4248 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4251 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4252 void *val, int bytes)
4254 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4255 return X86EMUL_IO_NEEDED;
4258 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4259 void *val, int bytes)
4261 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4263 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4264 return X86EMUL_CONTINUE;
4267 static const struct read_write_emulator_ops read_emultor = {
4268 .read_write_prepare = read_prepare,
4269 .read_write_emulate = read_emulate,
4270 .read_write_mmio = vcpu_mmio_read,
4271 .read_write_exit_mmio = read_exit_mmio,
4274 static const struct read_write_emulator_ops write_emultor = {
4275 .read_write_emulate = write_emulate,
4276 .read_write_mmio = write_mmio,
4277 .read_write_exit_mmio = write_exit_mmio,
4281 static int emulator_read_write_onepage(unsigned long addr, void *val,
4283 struct x86_exception *exception,
4284 struct kvm_vcpu *vcpu,
4285 const struct read_write_emulator_ops *ops)
4289 bool write = ops->write;
4290 struct kvm_mmio_fragment *frag;
4292 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4295 return X86EMUL_PROPAGATE_FAULT;
4297 /* For APIC access vmexit */
4301 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4302 return X86EMUL_CONTINUE;
4306 * Is this MMIO handled locally?
4308 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4309 if (handled == bytes)
4310 return X86EMUL_CONTINUE;
4316 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4317 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4321 return X86EMUL_CONTINUE;
4324 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4325 void *val, unsigned int bytes,
4326 struct x86_exception *exception,
4327 const struct read_write_emulator_ops *ops)
4329 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4333 if (ops->read_write_prepare &&
4334 ops->read_write_prepare(vcpu, val, bytes))
4335 return X86EMUL_CONTINUE;
4337 vcpu->mmio_nr_fragments = 0;
4339 /* Crossing a page boundary? */
4340 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4343 now = -addr & ~PAGE_MASK;
4344 rc = emulator_read_write_onepage(addr, val, now, exception,
4347 if (rc != X86EMUL_CONTINUE)
4354 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4356 if (rc != X86EMUL_CONTINUE)
4359 if (!vcpu->mmio_nr_fragments)
4362 gpa = vcpu->mmio_fragments[0].gpa;
4364 vcpu->mmio_needed = 1;
4365 vcpu->mmio_cur_fragment = 0;
4367 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4368 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4369 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4370 vcpu->run->mmio.phys_addr = gpa;
4372 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4375 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4379 struct x86_exception *exception)
4381 return emulator_read_write(ctxt, addr, val, bytes,
4382 exception, &read_emultor);
4385 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4389 struct x86_exception *exception)
4391 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4392 exception, &write_emultor);
4395 #define CMPXCHG_TYPE(t, ptr, old, new) \
4396 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4398 #ifdef CONFIG_X86_64
4399 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4401 # define CMPXCHG64(ptr, old, new) \
4402 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4405 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4410 struct x86_exception *exception)
4412 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4418 /* guests cmpxchg8b have to be emulated atomically */
4419 if (bytes > 8 || (bytes & (bytes - 1)))
4422 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4424 if (gpa == UNMAPPED_GVA ||
4425 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4428 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4431 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4432 if (is_error_page(page))
4435 kaddr = kmap_atomic(page);
4436 kaddr += offset_in_page(gpa);
4439 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4442 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4445 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4448 exchanged = CMPXCHG64(kaddr, old, new);
4453 kunmap_atomic(kaddr);
4454 kvm_release_page_dirty(page);
4457 return X86EMUL_CMPXCHG_FAILED;
4459 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
4460 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4462 return X86EMUL_CONTINUE;
4465 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4467 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4470 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4472 /* TODO: String I/O for in kernel device */
4475 if (vcpu->arch.pio.in)
4476 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4477 vcpu->arch.pio.size, pd);
4479 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4480 vcpu->arch.pio.port, vcpu->arch.pio.size,
4485 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4486 unsigned short port, void *val,
4487 unsigned int count, bool in)
4489 trace_kvm_pio(!in, port, size, count);
4491 vcpu->arch.pio.port = port;
4492 vcpu->arch.pio.in = in;
4493 vcpu->arch.pio.count = count;
4494 vcpu->arch.pio.size = size;
4496 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4497 vcpu->arch.pio.count = 0;
4501 vcpu->run->exit_reason = KVM_EXIT_IO;
4502 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4503 vcpu->run->io.size = size;
4504 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4505 vcpu->run->io.count = count;
4506 vcpu->run->io.port = port;
4511 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4512 int size, unsigned short port, void *val,
4515 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4518 if (vcpu->arch.pio.count)
4521 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4524 memcpy(val, vcpu->arch.pio_data, size * count);
4525 vcpu->arch.pio.count = 0;
4532 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4533 int size, unsigned short port,
4534 const void *val, unsigned int count)
4536 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4538 memcpy(vcpu->arch.pio_data, val, size * count);
4539 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4542 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4544 return kvm_x86_ops->get_segment_base(vcpu, seg);
4547 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4549 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4552 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4554 if (!need_emulate_wbinvd(vcpu))
4555 return X86EMUL_CONTINUE;
4557 if (kvm_x86_ops->has_wbinvd_exit()) {
4558 int cpu = get_cpu();
4560 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4561 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4562 wbinvd_ipi, NULL, 1);
4564 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4567 return X86EMUL_CONTINUE;
4569 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4571 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4573 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4576 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4578 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4581 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4584 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4587 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4589 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4592 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4594 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4595 unsigned long value;
4599 value = kvm_read_cr0(vcpu);
4602 value = vcpu->arch.cr2;
4605 value = kvm_read_cr3(vcpu);
4608 value = kvm_read_cr4(vcpu);
4611 value = kvm_get_cr8(vcpu);
4614 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4621 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4623 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4628 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4631 vcpu->arch.cr2 = val;
4634 res = kvm_set_cr3(vcpu, val);
4637 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4640 res = kvm_set_cr8(vcpu, val);
4643 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4650 static void emulator_set_rflags(struct x86_emulate_ctxt *ctxt, ulong val)
4652 kvm_set_rflags(emul_to_vcpu(ctxt), val);
4655 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4657 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4660 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4662 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4665 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4667 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4670 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4672 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4675 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4677 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4680 static unsigned long emulator_get_cached_segment_base(
4681 struct x86_emulate_ctxt *ctxt, int seg)
4683 return get_segment_base(emul_to_vcpu(ctxt), seg);
4686 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4687 struct desc_struct *desc, u32 *base3,
4690 struct kvm_segment var;
4692 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4693 *selector = var.selector;
4696 memset(desc, 0, sizeof(*desc));
4702 set_desc_limit(desc, var.limit);
4703 set_desc_base(desc, (unsigned long)var.base);
4704 #ifdef CONFIG_X86_64
4706 *base3 = var.base >> 32;
4708 desc->type = var.type;
4710 desc->dpl = var.dpl;
4711 desc->p = var.present;
4712 desc->avl = var.avl;
4720 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4721 struct desc_struct *desc, u32 base3,
4724 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4725 struct kvm_segment var;
4727 var.selector = selector;
4728 var.base = get_desc_base(desc);
4729 #ifdef CONFIG_X86_64
4730 var.base |= ((u64)base3) << 32;
4732 var.limit = get_desc_limit(desc);
4734 var.limit = (var.limit << 12) | 0xfff;
4735 var.type = desc->type;
4736 var.present = desc->p;
4737 var.dpl = desc->dpl;
4742 var.avl = desc->avl;
4743 var.present = desc->p;
4744 var.unusable = !var.present;
4747 kvm_set_segment(vcpu, &var, seg);
4751 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4752 u32 msr_index, u64 *pdata)
4754 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4757 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4758 u32 msr_index, u64 data)
4760 struct msr_data msr;
4763 msr.index = msr_index;
4764 msr.host_initiated = false;
4765 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4768 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4769 u32 pmc, u64 *pdata)
4771 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4774 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4776 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4779 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4782 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4784 * CR0.TS may reference the host fpu state, not the guest fpu state,
4785 * so it may be clear at this point.
4790 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4795 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4796 struct x86_instruction_info *info,
4797 enum x86_intercept_stage stage)
4799 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4802 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4803 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4805 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4808 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4810 return kvm_register_read(emul_to_vcpu(ctxt), reg);
4813 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4815 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4818 static const struct x86_emulate_ops emulate_ops = {
4819 .read_gpr = emulator_read_gpr,
4820 .write_gpr = emulator_write_gpr,
4821 .read_std = kvm_read_guest_virt_system,
4822 .write_std = kvm_write_guest_virt_system,
4823 .fetch = kvm_fetch_guest_virt,
4824 .read_emulated = emulator_read_emulated,
4825 .write_emulated = emulator_write_emulated,
4826 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4827 .invlpg = emulator_invlpg,
4828 .pio_in_emulated = emulator_pio_in_emulated,
4829 .pio_out_emulated = emulator_pio_out_emulated,
4830 .get_segment = emulator_get_segment,
4831 .set_segment = emulator_set_segment,
4832 .get_cached_segment_base = emulator_get_cached_segment_base,
4833 .get_gdt = emulator_get_gdt,
4834 .get_idt = emulator_get_idt,
4835 .set_gdt = emulator_set_gdt,
4836 .set_idt = emulator_set_idt,
4837 .get_cr = emulator_get_cr,
4838 .set_cr = emulator_set_cr,
4839 .set_rflags = emulator_set_rflags,
4840 .cpl = emulator_get_cpl,
4841 .get_dr = emulator_get_dr,
4842 .set_dr = emulator_set_dr,
4843 .set_msr = emulator_set_msr,
4844 .get_msr = emulator_get_msr,
4845 .read_pmc = emulator_read_pmc,
4846 .halt = emulator_halt,
4847 .wbinvd = emulator_wbinvd,
4848 .fix_hypercall = emulator_fix_hypercall,
4849 .get_fpu = emulator_get_fpu,
4850 .put_fpu = emulator_put_fpu,
4851 .intercept = emulator_intercept,
4852 .get_cpuid = emulator_get_cpuid,
4855 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4857 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4859 * an sti; sti; sequence only disable interrupts for the first
4860 * instruction. So, if the last instruction, be it emulated or
4861 * not, left the system with the INT_STI flag enabled, it
4862 * means that the last instruction is an sti. We should not
4863 * leave the flag on in this case. The same goes for mov ss
4865 if (!(int_shadow & mask))
4866 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4869 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4871 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4872 if (ctxt->exception.vector == PF_VECTOR)
4873 kvm_propagate_fault(vcpu, &ctxt->exception);
4874 else if (ctxt->exception.error_code_valid)
4875 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4876 ctxt->exception.error_code);
4878 kvm_queue_exception(vcpu, ctxt->exception.vector);
4881 static void init_decode_cache(struct x86_emulate_ctxt *ctxt)
4883 memset(&ctxt->opcode_len, 0,
4884 (void *)&ctxt->_regs - (void *)&ctxt->opcode_len);
4886 ctxt->fetch.start = 0;
4887 ctxt->fetch.end = 0;
4888 ctxt->io_read.pos = 0;
4889 ctxt->io_read.end = 0;
4890 ctxt->mem_read.pos = 0;
4891 ctxt->mem_read.end = 0;
4894 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4896 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4899 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4901 ctxt->eflags = kvm_get_rflags(vcpu);
4902 ctxt->eip = kvm_rip_read(vcpu);
4903 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4904 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4905 cs_l ? X86EMUL_MODE_PROT64 :
4906 cs_db ? X86EMUL_MODE_PROT32 :
4907 X86EMUL_MODE_PROT16;
4908 ctxt->guest_mode = is_guest_mode(vcpu);
4910 init_decode_cache(ctxt);
4911 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4914 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4916 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4919 init_emulate_ctxt(vcpu);
4923 ctxt->_eip = ctxt->eip + inc_eip;
4924 ret = emulate_int_real(ctxt, irq);
4926 if (ret != X86EMUL_CONTINUE)
4927 return EMULATE_FAIL;
4929 ctxt->eip = ctxt->_eip;
4930 kvm_rip_write(vcpu, ctxt->eip);
4931 kvm_set_rflags(vcpu, ctxt->eflags);
4933 if (irq == NMI_VECTOR)
4934 vcpu->arch.nmi_pending = 0;
4936 vcpu->arch.interrupt.pending = false;
4938 return EMULATE_DONE;
4940 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4942 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4944 int r = EMULATE_DONE;
4946 ++vcpu->stat.insn_emulation_fail;
4947 trace_kvm_emulate_insn_failed(vcpu);
4948 if (!is_guest_mode(vcpu)) {
4949 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4950 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4951 vcpu->run->internal.ndata = 0;
4954 kvm_queue_exception(vcpu, UD_VECTOR);
4959 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
4960 bool write_fault_to_shadow_pgtable,
4966 if (emulation_type & EMULTYPE_NO_REEXECUTE)
4969 if (!vcpu->arch.mmu.direct_map) {
4971 * Write permission should be allowed since only
4972 * write access need to be emulated.
4974 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4977 * If the mapping is invalid in guest, let cpu retry
4978 * it to generate fault.
4980 if (gpa == UNMAPPED_GVA)
4985 * Do not retry the unhandleable instruction if it faults on the
4986 * readonly host memory, otherwise it will goto a infinite loop:
4987 * retry instruction -> write #PF -> emulation fail -> retry
4988 * instruction -> ...
4990 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
4993 * If the instruction failed on the error pfn, it can not be fixed,
4994 * report the error to userspace.
4996 if (is_error_noslot_pfn(pfn))
4999 kvm_release_pfn_clean(pfn);
5001 /* The instructions are well-emulated on direct mmu. */
5002 if (vcpu->arch.mmu.direct_map) {
5003 unsigned int indirect_shadow_pages;
5005 spin_lock(&vcpu->kvm->mmu_lock);
5006 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5007 spin_unlock(&vcpu->kvm->mmu_lock);
5009 if (indirect_shadow_pages)
5010 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5016 * if emulation was due to access to shadowed page table
5017 * and it failed try to unshadow page and re-enter the
5018 * guest to let CPU execute the instruction.
5020 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5023 * If the access faults on its page table, it can not
5024 * be fixed by unprotecting shadow page and it should
5025 * be reported to userspace.
5027 return !write_fault_to_shadow_pgtable;
5030 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5031 unsigned long cr2, int emulation_type)
5033 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5034 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5036 last_retry_eip = vcpu->arch.last_retry_eip;
5037 last_retry_addr = vcpu->arch.last_retry_addr;
5040 * If the emulation is caused by #PF and it is non-page_table
5041 * writing instruction, it means the VM-EXIT is caused by shadow
5042 * page protected, we can zap the shadow page and retry this
5043 * instruction directly.
5045 * Note: if the guest uses a non-page-table modifying instruction
5046 * on the PDE that points to the instruction, then we will unmap
5047 * the instruction and go to an infinite loop. So, we cache the
5048 * last retried eip and the last fault address, if we meet the eip
5049 * and the address again, we can break out of the potential infinite
5052 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5054 if (!(emulation_type & EMULTYPE_RETRY))
5057 if (x86_page_table_writing_insn(ctxt))
5060 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5063 vcpu->arch.last_retry_eip = ctxt->eip;
5064 vcpu->arch.last_retry_addr = cr2;
5066 if (!vcpu->arch.mmu.direct_map)
5067 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5069 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5074 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5075 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5077 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5086 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5087 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5092 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, int *r)
5094 struct kvm_run *kvm_run = vcpu->run;
5097 * Use the "raw" value to see if TF was passed to the processor.
5098 * Note that the new value of the flags has not been saved yet.
5100 * This is correct even for TF set by the guest, because "the
5101 * processor will not generate this exception after the instruction
5102 * that sets the TF flag".
5104 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5106 if (unlikely(rflags & X86_EFLAGS_TF)) {
5107 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5108 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1;
5109 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5110 kvm_run->debug.arch.exception = DB_VECTOR;
5111 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5112 *r = EMULATE_USER_EXIT;
5114 vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5116 * "Certain debug exceptions may clear bit 0-3. The
5117 * remaining contents of the DR6 register are never
5118 * cleared by the processor".
5120 vcpu->arch.dr6 &= ~15;
5121 vcpu->arch.dr6 |= DR6_BS;
5122 kvm_queue_exception(vcpu, DB_VECTOR);
5127 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5129 struct kvm_run *kvm_run = vcpu->run;
5130 unsigned long eip = vcpu->arch.emulate_ctxt.eip;
5133 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5134 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5135 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5136 vcpu->arch.guest_debug_dr7,
5140 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
5141 kvm_run->debug.arch.pc = kvm_rip_read(vcpu) +
5142 get_segment_base(vcpu, VCPU_SREG_CS);
5144 kvm_run->debug.arch.exception = DB_VECTOR;
5145 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5146 *r = EMULATE_USER_EXIT;
5151 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK)) {
5152 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5157 vcpu->arch.dr6 &= ~15;
5158 vcpu->arch.dr6 |= dr6;
5159 kvm_queue_exception(vcpu, DB_VECTOR);
5168 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5175 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5176 bool writeback = true;
5177 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5180 * Clear write_fault_to_shadow_pgtable here to ensure it is
5183 vcpu->arch.write_fault_to_shadow_pgtable = false;
5184 kvm_clear_exception_queue(vcpu);
5186 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5187 init_emulate_ctxt(vcpu);
5190 * We will reenter on the same instruction since
5191 * we do not set complete_userspace_io. This does not
5192 * handle watchpoints yet, those would be handled in
5195 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5198 ctxt->interruptibility = 0;
5199 ctxt->have_exception = false;
5200 ctxt->perm_ok = false;
5202 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5204 r = x86_decode_insn(ctxt, insn, insn_len);
5206 trace_kvm_emulate_insn_start(vcpu);
5207 ++vcpu->stat.insn_emulation;
5208 if (r != EMULATION_OK) {
5209 if (emulation_type & EMULTYPE_TRAP_UD)
5210 return EMULATE_FAIL;
5211 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5213 return EMULATE_DONE;
5214 if (emulation_type & EMULTYPE_SKIP)
5215 return EMULATE_FAIL;
5216 return handle_emulation_failure(vcpu);
5220 if (emulation_type & EMULTYPE_SKIP) {
5221 kvm_rip_write(vcpu, ctxt->_eip);
5222 return EMULATE_DONE;
5225 if (retry_instruction(ctxt, cr2, emulation_type))
5226 return EMULATE_DONE;
5228 /* this is needed for vmware backdoor interface to work since it
5229 changes registers values during IO operation */
5230 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5231 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5232 emulator_invalidate_register_cache(ctxt);
5236 r = x86_emulate_insn(ctxt);
5238 if (r == EMULATION_INTERCEPTED)
5239 return EMULATE_DONE;
5241 if (r == EMULATION_FAILED) {
5242 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5244 return EMULATE_DONE;
5246 return handle_emulation_failure(vcpu);
5249 if (ctxt->have_exception) {
5250 inject_emulated_exception(vcpu);
5252 } else if (vcpu->arch.pio.count) {
5253 if (!vcpu->arch.pio.in) {
5254 /* FIXME: return into emulator if single-stepping. */
5255 vcpu->arch.pio.count = 0;
5258 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5260 r = EMULATE_USER_EXIT;
5261 } else if (vcpu->mmio_needed) {
5262 if (!vcpu->mmio_is_write)
5264 r = EMULATE_USER_EXIT;
5265 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5266 } else if (r == EMULATION_RESTART)
5272 toggle_interruptibility(vcpu, ctxt->interruptibility);
5273 kvm_make_request(KVM_REQ_EVENT, vcpu);
5274 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5275 kvm_rip_write(vcpu, ctxt->eip);
5276 if (r == EMULATE_DONE)
5277 kvm_vcpu_check_singlestep(vcpu, &r);
5278 kvm_set_rflags(vcpu, ctxt->eflags);
5280 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5284 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5286 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5288 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5289 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5290 size, port, &val, 1);
5291 /* do not return to emulator after return from userspace */
5292 vcpu->arch.pio.count = 0;
5295 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5297 static void tsc_bad(void *info)
5299 __this_cpu_write(cpu_tsc_khz, 0);
5302 static void tsc_khz_changed(void *data)
5304 struct cpufreq_freqs *freq = data;
5305 unsigned long khz = 0;
5309 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5310 khz = cpufreq_quick_get(raw_smp_processor_id());
5313 __this_cpu_write(cpu_tsc_khz, khz);
5316 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5319 struct cpufreq_freqs *freq = data;
5321 struct kvm_vcpu *vcpu;
5322 int i, send_ipi = 0;
5325 * We allow guests to temporarily run on slowing clocks,
5326 * provided we notify them after, or to run on accelerating
5327 * clocks, provided we notify them before. Thus time never
5330 * However, we have a problem. We can't atomically update
5331 * the frequency of a given CPU from this function; it is
5332 * merely a notifier, which can be called from any CPU.
5333 * Changing the TSC frequency at arbitrary points in time
5334 * requires a recomputation of local variables related to
5335 * the TSC for each VCPU. We must flag these local variables
5336 * to be updated and be sure the update takes place with the
5337 * new frequency before any guests proceed.
5339 * Unfortunately, the combination of hotplug CPU and frequency
5340 * change creates an intractable locking scenario; the order
5341 * of when these callouts happen is undefined with respect to
5342 * CPU hotplug, and they can race with each other. As such,
5343 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5344 * undefined; you can actually have a CPU frequency change take
5345 * place in between the computation of X and the setting of the
5346 * variable. To protect against this problem, all updates of
5347 * the per_cpu tsc_khz variable are done in an interrupt
5348 * protected IPI, and all callers wishing to update the value
5349 * must wait for a synchronous IPI to complete (which is trivial
5350 * if the caller is on the CPU already). This establishes the
5351 * necessary total order on variable updates.
5353 * Note that because a guest time update may take place
5354 * anytime after the setting of the VCPU's request bit, the
5355 * correct TSC value must be set before the request. However,
5356 * to ensure the update actually makes it to any guest which
5357 * starts running in hardware virtualization between the set
5358 * and the acquisition of the spinlock, we must also ping the
5359 * CPU after setting the request bit.
5363 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5365 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5368 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5370 spin_lock(&kvm_lock);
5371 list_for_each_entry(kvm, &vm_list, vm_list) {
5372 kvm_for_each_vcpu(i, vcpu, kvm) {
5373 if (vcpu->cpu != freq->cpu)
5375 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5376 if (vcpu->cpu != smp_processor_id())
5380 spin_unlock(&kvm_lock);
5382 if (freq->old < freq->new && send_ipi) {
5384 * We upscale the frequency. Must make the guest
5385 * doesn't see old kvmclock values while running with
5386 * the new frequency, otherwise we risk the guest sees
5387 * time go backwards.
5389 * In case we update the frequency for another cpu
5390 * (which might be in guest context) send an interrupt
5391 * to kick the cpu out of guest context. Next time
5392 * guest context is entered kvmclock will be updated,
5393 * so the guest will not see stale values.
5395 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5400 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5401 .notifier_call = kvmclock_cpufreq_notifier
5404 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5405 unsigned long action, void *hcpu)
5407 unsigned int cpu = (unsigned long)hcpu;
5411 case CPU_DOWN_FAILED:
5412 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5414 case CPU_DOWN_PREPARE:
5415 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5421 static struct notifier_block kvmclock_cpu_notifier_block = {
5422 .notifier_call = kvmclock_cpu_notifier,
5423 .priority = -INT_MAX
5426 static void kvm_timer_init(void)
5430 max_tsc_khz = tsc_khz;
5431 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5432 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5433 #ifdef CONFIG_CPU_FREQ
5434 struct cpufreq_policy policy;
5435 memset(&policy, 0, sizeof(policy));
5437 cpufreq_get_policy(&policy, cpu);
5438 if (policy.cpuinfo.max_freq)
5439 max_tsc_khz = policy.cpuinfo.max_freq;
5442 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5443 CPUFREQ_TRANSITION_NOTIFIER);
5445 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5446 for_each_online_cpu(cpu)
5447 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5450 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5452 int kvm_is_in_guest(void)
5454 return __this_cpu_read(current_vcpu) != NULL;
5457 static int kvm_is_user_mode(void)
5461 if (__this_cpu_read(current_vcpu))
5462 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5464 return user_mode != 0;
5467 static unsigned long kvm_get_guest_ip(void)
5469 unsigned long ip = 0;
5471 if (__this_cpu_read(current_vcpu))
5472 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5477 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5478 .is_in_guest = kvm_is_in_guest,
5479 .is_user_mode = kvm_is_user_mode,
5480 .get_guest_ip = kvm_get_guest_ip,
5483 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5485 __this_cpu_write(current_vcpu, vcpu);
5487 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5489 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5491 __this_cpu_write(current_vcpu, NULL);
5493 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5495 static void kvm_set_mmio_spte_mask(void)
5498 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5501 * Set the reserved bits and the present bit of an paging-structure
5502 * entry to generate page fault with PFER.RSV = 1.
5504 /* Mask the reserved physical address bits. */
5505 mask = ((1ull << (51 - maxphyaddr + 1)) - 1) << maxphyaddr;
5507 /* Bit 62 is always reserved for 32bit host. */
5508 mask |= 0x3ull << 62;
5510 /* Set the present bit. */
5513 #ifdef CONFIG_X86_64
5515 * If reserved bit is not supported, clear the present bit to disable
5518 if (maxphyaddr == 52)
5522 kvm_mmu_set_mmio_spte_mask(mask);
5525 #ifdef CONFIG_X86_64
5526 static void pvclock_gtod_update_fn(struct work_struct *work)
5530 struct kvm_vcpu *vcpu;
5533 spin_lock(&kvm_lock);
5534 list_for_each_entry(kvm, &vm_list, vm_list)
5535 kvm_for_each_vcpu(i, vcpu, kvm)
5536 set_bit(KVM_REQ_MASTERCLOCK_UPDATE, &vcpu->requests);
5537 atomic_set(&kvm_guest_has_master_clock, 0);
5538 spin_unlock(&kvm_lock);
5541 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5544 * Notification about pvclock gtod data update.
5546 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5549 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5550 struct timekeeper *tk = priv;
5552 update_pvclock_gtod(tk);
5554 /* disable master clock if host does not trust, or does not
5555 * use, TSC clocksource
5557 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5558 atomic_read(&kvm_guest_has_master_clock) != 0)
5559 queue_work(system_long_wq, &pvclock_gtod_work);
5564 static struct notifier_block pvclock_gtod_notifier = {
5565 .notifier_call = pvclock_gtod_notify,
5569 int kvm_arch_init(void *opaque)
5572 struct kvm_x86_ops *ops = opaque;
5575 printk(KERN_ERR "kvm: already loaded the other module\n");
5580 if (!ops->cpu_has_kvm_support()) {
5581 printk(KERN_ERR "kvm: no hardware support\n");
5585 if (ops->disabled_by_bios()) {
5586 printk(KERN_ERR "kvm: disabled by bios\n");
5592 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5594 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5598 r = kvm_mmu_module_init();
5600 goto out_free_percpu;
5602 kvm_set_mmio_spte_mask();
5605 kvm_init_msr_list();
5607 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5608 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5612 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5615 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5618 #ifdef CONFIG_X86_64
5619 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5625 free_percpu(shared_msrs);
5630 void kvm_arch_exit(void)
5632 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5634 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5635 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5636 CPUFREQ_TRANSITION_NOTIFIER);
5637 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5638 #ifdef CONFIG_X86_64
5639 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5642 kvm_mmu_module_exit();
5643 free_percpu(shared_msrs);
5646 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5648 ++vcpu->stat.halt_exits;
5649 if (irqchip_in_kernel(vcpu->kvm)) {
5650 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5653 vcpu->run->exit_reason = KVM_EXIT_HLT;
5657 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5659 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5661 u64 param, ingpa, outgpa, ret;
5662 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5663 bool fast, longmode;
5667 * hypercall generates UD from non zero cpl and real mode
5670 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5671 kvm_queue_exception(vcpu, UD_VECTOR);
5675 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5676 longmode = is_long_mode(vcpu) && cs_l == 1;
5679 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5680 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5681 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5682 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5683 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5684 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5686 #ifdef CONFIG_X86_64
5688 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5689 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5690 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5694 code = param & 0xffff;
5695 fast = (param >> 16) & 0x1;
5696 rep_cnt = (param >> 32) & 0xfff;
5697 rep_idx = (param >> 48) & 0xfff;
5699 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5702 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5703 kvm_vcpu_on_spin(vcpu);
5706 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5710 ret = res | (((u64)rep_done & 0xfff) << 32);
5712 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5714 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5715 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5722 * kvm_pv_kick_cpu_op: Kick a vcpu.
5724 * @apicid - apicid of vcpu to be kicked.
5726 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5728 struct kvm_lapic_irq lapic_irq;
5730 lapic_irq.shorthand = 0;
5731 lapic_irq.dest_mode = 0;
5732 lapic_irq.dest_id = apicid;
5734 lapic_irq.delivery_mode = APIC_DM_REMRD;
5735 kvm_irq_delivery_to_apic(kvm, 0, &lapic_irq, NULL);
5738 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5740 unsigned long nr, a0, a1, a2, a3, ret;
5743 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5744 return kvm_hv_hypercall(vcpu);
5746 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5747 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5748 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5749 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5750 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5752 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5754 if (!is_long_mode(vcpu)) {
5762 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5768 case KVM_HC_VAPIC_POLL_IRQ:
5771 case KVM_HC_KICK_CPU:
5772 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5780 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5781 ++vcpu->stat.hypercalls;
5784 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5786 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5788 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5789 char instruction[3];
5790 unsigned long rip = kvm_rip_read(vcpu);
5792 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5794 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5798 * Check if userspace requested an interrupt window, and that the
5799 * interrupt window is open.
5801 * No need to exit to userspace if we already have an interrupt queued.
5803 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5805 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5806 vcpu->run->request_interrupt_window &&
5807 kvm_arch_interrupt_allowed(vcpu));
5810 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5812 struct kvm_run *kvm_run = vcpu->run;
5814 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5815 kvm_run->cr8 = kvm_get_cr8(vcpu);
5816 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5817 if (irqchip_in_kernel(vcpu->kvm))
5818 kvm_run->ready_for_interrupt_injection = 1;
5820 kvm_run->ready_for_interrupt_injection =
5821 kvm_arch_interrupt_allowed(vcpu) &&
5822 !kvm_cpu_has_interrupt(vcpu) &&
5823 !kvm_event_needs_reinjection(vcpu);
5826 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5830 if (!kvm_x86_ops->update_cr8_intercept)
5833 if (!vcpu->arch.apic)
5836 if (!vcpu->arch.apic->vapic_addr)
5837 max_irr = kvm_lapic_find_highest_irr(vcpu);
5844 tpr = kvm_lapic_get_cr8(vcpu);
5846 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5849 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
5853 /* try to reinject previous events if any */
5854 if (vcpu->arch.exception.pending) {
5855 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5856 vcpu->arch.exception.has_error_code,
5857 vcpu->arch.exception.error_code);
5858 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5859 vcpu->arch.exception.has_error_code,
5860 vcpu->arch.exception.error_code,
5861 vcpu->arch.exception.reinject);
5865 if (vcpu->arch.nmi_injected) {
5866 kvm_x86_ops->set_nmi(vcpu);
5870 if (vcpu->arch.interrupt.pending) {
5871 kvm_x86_ops->set_irq(vcpu);
5875 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
5876 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
5881 /* try to inject new event if pending */
5882 if (vcpu->arch.nmi_pending) {
5883 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5884 --vcpu->arch.nmi_pending;
5885 vcpu->arch.nmi_injected = true;
5886 kvm_x86_ops->set_nmi(vcpu);
5888 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
5889 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5890 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5892 kvm_x86_ops->set_irq(vcpu);
5898 static void process_nmi(struct kvm_vcpu *vcpu)
5903 * x86 is limited to one NMI running, and one NMI pending after it.
5904 * If an NMI is already in progress, limit further NMIs to just one.
5905 * Otherwise, allow two (and we'll inject the first one immediately).
5907 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5910 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5911 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5912 kvm_make_request(KVM_REQ_EVENT, vcpu);
5915 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
5917 u64 eoi_exit_bitmap[4];
5920 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
5923 memset(eoi_exit_bitmap, 0, 32);
5926 kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
5927 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
5928 kvm_apic_update_tmr(vcpu, tmr);
5932 * Returns 1 to let __vcpu_run() continue the guest execution loop without
5933 * exiting to the userspace. Otherwise, the value will be returned to the
5936 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5939 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5940 vcpu->run->request_interrupt_window;
5941 bool req_immediate_exit = false;
5943 if (vcpu->requests) {
5944 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5945 kvm_mmu_unload(vcpu);
5946 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5947 __kvm_migrate_timers(vcpu);
5948 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
5949 kvm_gen_update_masterclock(vcpu->kvm);
5950 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
5951 kvm_gen_kvmclock_update(vcpu);
5952 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5953 r = kvm_guest_time_update(vcpu);
5957 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5958 kvm_mmu_sync_roots(vcpu);
5959 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5960 kvm_x86_ops->tlb_flush(vcpu);
5961 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5962 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5966 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5967 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5971 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5972 vcpu->fpu_active = 0;
5973 kvm_x86_ops->fpu_deactivate(vcpu);
5975 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5976 /* Page is swapped out. Do synthetic halt */
5977 vcpu->arch.apf.halted = true;
5981 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5982 record_steal_time(vcpu);
5983 if (kvm_check_request(KVM_REQ_NMI, vcpu))
5985 if (kvm_check_request(KVM_REQ_PMU, vcpu))
5986 kvm_handle_pmu_event(vcpu);
5987 if (kvm_check_request(KVM_REQ_PMI, vcpu))
5988 kvm_deliver_pmi(vcpu);
5989 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
5990 vcpu_scan_ioapic(vcpu);
5993 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5994 kvm_apic_accept_events(vcpu);
5995 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6000 if (inject_pending_event(vcpu, req_int_win) != 0)
6001 req_immediate_exit = true;
6002 /* enable NMI/IRQ window open exits if needed */
6003 else if (vcpu->arch.nmi_pending)
6004 kvm_x86_ops->enable_nmi_window(vcpu);
6005 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6006 kvm_x86_ops->enable_irq_window(vcpu);
6008 if (kvm_lapic_enabled(vcpu)) {
6010 * Update architecture specific hints for APIC
6011 * virtual interrupt delivery.
6013 if (kvm_x86_ops->hwapic_irr_update)
6014 kvm_x86_ops->hwapic_irr_update(vcpu,
6015 kvm_lapic_find_highest_irr(vcpu));
6016 update_cr8_intercept(vcpu);
6017 kvm_lapic_sync_to_vapic(vcpu);
6021 r = kvm_mmu_reload(vcpu);
6023 goto cancel_injection;
6028 kvm_x86_ops->prepare_guest_switch(vcpu);
6029 if (vcpu->fpu_active)
6030 kvm_load_guest_fpu(vcpu);
6031 kvm_load_guest_xcr0(vcpu);
6033 vcpu->mode = IN_GUEST_MODE;
6035 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6037 /* We should set ->mode before check ->requests,
6038 * see the comment in make_all_cpus_request.
6040 smp_mb__after_srcu_read_unlock();
6042 local_irq_disable();
6044 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6045 || need_resched() || signal_pending(current)) {
6046 vcpu->mode = OUTSIDE_GUEST_MODE;
6050 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6052 goto cancel_injection;
6055 if (req_immediate_exit)
6056 smp_send_reschedule(vcpu->cpu);
6060 if (unlikely(vcpu->arch.switch_db_regs)) {
6062 set_debugreg(vcpu->arch.eff_db[0], 0);
6063 set_debugreg(vcpu->arch.eff_db[1], 1);
6064 set_debugreg(vcpu->arch.eff_db[2], 2);
6065 set_debugreg(vcpu->arch.eff_db[3], 3);
6066 set_debugreg(vcpu->arch.dr6, 6);
6069 trace_kvm_entry(vcpu->vcpu_id);
6070 kvm_x86_ops->run(vcpu);
6073 * Do this here before restoring debug registers on the host. And
6074 * since we do this before handling the vmexit, a DR access vmexit
6075 * can (a) read the correct value of the debug registers, (b) set
6076 * KVM_DEBUGREG_WONT_EXIT again.
6078 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6081 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6082 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6083 for (i = 0; i < KVM_NR_DB_REGS; i++)
6084 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6088 * If the guest has used debug registers, at least dr7
6089 * will be disabled while returning to the host.
6090 * If we don't have active breakpoints in the host, we don't
6091 * care about the messed up debug address registers. But if
6092 * we have some of them active, restore the old state.
6094 if (hw_breakpoint_active())
6095 hw_breakpoint_restore();
6097 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
6100 vcpu->mode = OUTSIDE_GUEST_MODE;
6103 /* Interrupt is enabled by handle_external_intr() */
6104 kvm_x86_ops->handle_external_intr(vcpu);
6109 * We must have an instruction between local_irq_enable() and
6110 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6111 * the interrupt shadow. The stat.exits increment will do nicely.
6112 * But we need to prevent reordering, hence this barrier():
6120 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6123 * Profile KVM exit RIPs:
6125 if (unlikely(prof_on == KVM_PROFILING)) {
6126 unsigned long rip = kvm_rip_read(vcpu);
6127 profile_hit(KVM_PROFILING, (void *)rip);
6130 if (unlikely(vcpu->arch.tsc_always_catchup))
6131 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6133 if (vcpu->arch.apic_attention)
6134 kvm_lapic_sync_from_vapic(vcpu);
6136 r = kvm_x86_ops->handle_exit(vcpu);
6140 kvm_x86_ops->cancel_injection(vcpu);
6141 if (unlikely(vcpu->arch.apic_attention))
6142 kvm_lapic_sync_from_vapic(vcpu);
6148 static int __vcpu_run(struct kvm_vcpu *vcpu)
6151 struct kvm *kvm = vcpu->kvm;
6153 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6157 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6158 !vcpu->arch.apf.halted)
6159 r = vcpu_enter_guest(vcpu);
6161 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6162 kvm_vcpu_block(vcpu);
6163 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6164 if (kvm_check_request(KVM_REQ_UNHALT, vcpu)) {
6165 kvm_apic_accept_events(vcpu);
6166 switch(vcpu->arch.mp_state) {
6167 case KVM_MP_STATE_HALTED:
6168 vcpu->arch.pv.pv_unhalted = false;
6169 vcpu->arch.mp_state =
6170 KVM_MP_STATE_RUNNABLE;
6171 case KVM_MP_STATE_RUNNABLE:
6172 vcpu->arch.apf.halted = false;
6174 case KVM_MP_STATE_INIT_RECEIVED:
6186 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6187 if (kvm_cpu_has_pending_timer(vcpu))
6188 kvm_inject_pending_timer_irqs(vcpu);
6190 if (dm_request_for_irq_injection(vcpu)) {
6192 vcpu->run->exit_reason = KVM_EXIT_INTR;
6193 ++vcpu->stat.request_irq_exits;
6196 kvm_check_async_pf_completion(vcpu);
6198 if (signal_pending(current)) {
6200 vcpu->run->exit_reason = KVM_EXIT_INTR;
6201 ++vcpu->stat.signal_exits;
6203 if (need_resched()) {
6204 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6206 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6210 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6215 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6218 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6219 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6220 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6221 if (r != EMULATE_DONE)
6226 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6228 BUG_ON(!vcpu->arch.pio.count);
6230 return complete_emulated_io(vcpu);
6234 * Implements the following, as a state machine:
6238 * for each mmio piece in the fragment
6246 * for each mmio piece in the fragment
6251 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6253 struct kvm_run *run = vcpu->run;
6254 struct kvm_mmio_fragment *frag;
6257 BUG_ON(!vcpu->mmio_needed);
6259 /* Complete previous fragment */
6260 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6261 len = min(8u, frag->len);
6262 if (!vcpu->mmio_is_write)
6263 memcpy(frag->data, run->mmio.data, len);
6265 if (frag->len <= 8) {
6266 /* Switch to the next fragment. */
6268 vcpu->mmio_cur_fragment++;
6270 /* Go forward to the next mmio piece. */
6276 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6277 vcpu->mmio_needed = 0;
6279 /* FIXME: return into emulator if single-stepping. */
6280 if (vcpu->mmio_is_write)
6282 vcpu->mmio_read_completed = 1;
6283 return complete_emulated_io(vcpu);
6286 run->exit_reason = KVM_EXIT_MMIO;
6287 run->mmio.phys_addr = frag->gpa;
6288 if (vcpu->mmio_is_write)
6289 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6290 run->mmio.len = min(8u, frag->len);
6291 run->mmio.is_write = vcpu->mmio_is_write;
6292 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6297 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6302 if (!tsk_used_math(current) && init_fpu(current))
6305 if (vcpu->sigset_active)
6306 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6308 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6309 kvm_vcpu_block(vcpu);
6310 kvm_apic_accept_events(vcpu);
6311 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6316 /* re-sync apic's tpr */
6317 if (!irqchip_in_kernel(vcpu->kvm)) {
6318 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6324 if (unlikely(vcpu->arch.complete_userspace_io)) {
6325 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6326 vcpu->arch.complete_userspace_io = NULL;
6331 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6333 r = __vcpu_run(vcpu);
6336 post_kvm_run_save(vcpu);
6337 if (vcpu->sigset_active)
6338 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6343 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6345 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6347 * We are here if userspace calls get_regs() in the middle of
6348 * instruction emulation. Registers state needs to be copied
6349 * back from emulation context to vcpu. Userspace shouldn't do
6350 * that usually, but some bad designed PV devices (vmware
6351 * backdoor interface) need this to work
6353 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6354 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6356 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6357 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6358 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6359 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6360 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6361 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6362 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6363 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6364 #ifdef CONFIG_X86_64
6365 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6366 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6367 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6368 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6369 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6370 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6371 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6372 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6375 regs->rip = kvm_rip_read(vcpu);
6376 regs->rflags = kvm_get_rflags(vcpu);
6381 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6383 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6384 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6386 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6387 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6388 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6389 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6390 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6391 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6392 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6393 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6394 #ifdef CONFIG_X86_64
6395 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6396 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6397 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6398 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6399 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6400 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6401 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6402 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6405 kvm_rip_write(vcpu, regs->rip);
6406 kvm_set_rflags(vcpu, regs->rflags);
6408 vcpu->arch.exception.pending = false;
6410 kvm_make_request(KVM_REQ_EVENT, vcpu);
6415 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6417 struct kvm_segment cs;
6419 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6423 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6425 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6426 struct kvm_sregs *sregs)
6430 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6431 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6432 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6433 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6434 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6435 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6437 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6438 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6440 kvm_x86_ops->get_idt(vcpu, &dt);
6441 sregs->idt.limit = dt.size;
6442 sregs->idt.base = dt.address;
6443 kvm_x86_ops->get_gdt(vcpu, &dt);
6444 sregs->gdt.limit = dt.size;
6445 sregs->gdt.base = dt.address;
6447 sregs->cr0 = kvm_read_cr0(vcpu);
6448 sregs->cr2 = vcpu->arch.cr2;
6449 sregs->cr3 = kvm_read_cr3(vcpu);
6450 sregs->cr4 = kvm_read_cr4(vcpu);
6451 sregs->cr8 = kvm_get_cr8(vcpu);
6452 sregs->efer = vcpu->arch.efer;
6453 sregs->apic_base = kvm_get_apic_base(vcpu);
6455 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6457 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6458 set_bit(vcpu->arch.interrupt.nr,
6459 (unsigned long *)sregs->interrupt_bitmap);
6464 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6465 struct kvm_mp_state *mp_state)
6467 kvm_apic_accept_events(vcpu);
6468 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
6469 vcpu->arch.pv.pv_unhalted)
6470 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
6472 mp_state->mp_state = vcpu->arch.mp_state;
6477 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6478 struct kvm_mp_state *mp_state)
6480 if (!kvm_vcpu_has_lapic(vcpu) &&
6481 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6484 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6485 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6486 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6488 vcpu->arch.mp_state = mp_state->mp_state;
6489 kvm_make_request(KVM_REQ_EVENT, vcpu);
6493 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6494 int reason, bool has_error_code, u32 error_code)
6496 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6499 init_emulate_ctxt(vcpu);
6501 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6502 has_error_code, error_code);
6505 return EMULATE_FAIL;
6507 kvm_rip_write(vcpu, ctxt->eip);
6508 kvm_set_rflags(vcpu, ctxt->eflags);
6509 kvm_make_request(KVM_REQ_EVENT, vcpu);
6510 return EMULATE_DONE;
6512 EXPORT_SYMBOL_GPL(kvm_task_switch);
6514 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6515 struct kvm_sregs *sregs)
6517 struct msr_data apic_base_msr;
6518 int mmu_reset_needed = 0;
6519 int pending_vec, max_bits, idx;
6522 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6525 dt.size = sregs->idt.limit;
6526 dt.address = sregs->idt.base;
6527 kvm_x86_ops->set_idt(vcpu, &dt);
6528 dt.size = sregs->gdt.limit;
6529 dt.address = sregs->gdt.base;
6530 kvm_x86_ops->set_gdt(vcpu, &dt);
6532 vcpu->arch.cr2 = sregs->cr2;
6533 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6534 vcpu->arch.cr3 = sregs->cr3;
6535 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6537 kvm_set_cr8(vcpu, sregs->cr8);
6539 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6540 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6541 apic_base_msr.data = sregs->apic_base;
6542 apic_base_msr.host_initiated = true;
6543 kvm_set_apic_base(vcpu, &apic_base_msr);
6545 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6546 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6547 vcpu->arch.cr0 = sregs->cr0;
6549 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6550 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6551 if (sregs->cr4 & X86_CR4_OSXSAVE)
6552 kvm_update_cpuid(vcpu);
6554 idx = srcu_read_lock(&vcpu->kvm->srcu);
6555 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6556 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6557 mmu_reset_needed = 1;
6559 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6561 if (mmu_reset_needed)
6562 kvm_mmu_reset_context(vcpu);
6564 max_bits = KVM_NR_INTERRUPTS;
6565 pending_vec = find_first_bit(
6566 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6567 if (pending_vec < max_bits) {
6568 kvm_queue_interrupt(vcpu, pending_vec, false);
6569 pr_debug("Set back pending irq %d\n", pending_vec);
6572 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6573 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6574 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6575 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6576 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6577 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6579 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6580 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6582 update_cr8_intercept(vcpu);
6584 /* Older userspace won't unhalt the vcpu on reset. */
6585 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6586 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6588 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6590 kvm_make_request(KVM_REQ_EVENT, vcpu);
6595 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6596 struct kvm_guest_debug *dbg)
6598 unsigned long rflags;
6601 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6603 if (vcpu->arch.exception.pending)
6605 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6606 kvm_queue_exception(vcpu, DB_VECTOR);
6608 kvm_queue_exception(vcpu, BP_VECTOR);
6612 * Read rflags as long as potentially injected trace flags are still
6615 rflags = kvm_get_rflags(vcpu);
6617 vcpu->guest_debug = dbg->control;
6618 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6619 vcpu->guest_debug = 0;
6621 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6622 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6623 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6624 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6626 for (i = 0; i < KVM_NR_DB_REGS; i++)
6627 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6629 kvm_update_dr7(vcpu);
6631 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6632 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6633 get_segment_base(vcpu, VCPU_SREG_CS);
6636 * Trigger an rflags update that will inject or remove the trace
6639 kvm_set_rflags(vcpu, rflags);
6641 kvm_x86_ops->update_db_bp_intercept(vcpu);
6651 * Translate a guest virtual address to a guest physical address.
6653 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6654 struct kvm_translation *tr)
6656 unsigned long vaddr = tr->linear_address;
6660 idx = srcu_read_lock(&vcpu->kvm->srcu);
6661 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6662 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6663 tr->physical_address = gpa;
6664 tr->valid = gpa != UNMAPPED_GVA;
6671 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6673 struct i387_fxsave_struct *fxsave =
6674 &vcpu->arch.guest_fpu.state->fxsave;
6676 memcpy(fpu->fpr, fxsave->st_space, 128);
6677 fpu->fcw = fxsave->cwd;
6678 fpu->fsw = fxsave->swd;
6679 fpu->ftwx = fxsave->twd;
6680 fpu->last_opcode = fxsave->fop;
6681 fpu->last_ip = fxsave->rip;
6682 fpu->last_dp = fxsave->rdp;
6683 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6688 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6690 struct i387_fxsave_struct *fxsave =
6691 &vcpu->arch.guest_fpu.state->fxsave;
6693 memcpy(fxsave->st_space, fpu->fpr, 128);
6694 fxsave->cwd = fpu->fcw;
6695 fxsave->swd = fpu->fsw;
6696 fxsave->twd = fpu->ftwx;
6697 fxsave->fop = fpu->last_opcode;
6698 fxsave->rip = fpu->last_ip;
6699 fxsave->rdp = fpu->last_dp;
6700 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6705 int fx_init(struct kvm_vcpu *vcpu)
6709 err = fpu_alloc(&vcpu->arch.guest_fpu);
6713 fpu_finit(&vcpu->arch.guest_fpu);
6716 * Ensure guest xcr0 is valid for loading
6718 vcpu->arch.xcr0 = XSTATE_FP;
6720 vcpu->arch.cr0 |= X86_CR0_ET;
6724 EXPORT_SYMBOL_GPL(fx_init);
6726 static void fx_free(struct kvm_vcpu *vcpu)
6728 fpu_free(&vcpu->arch.guest_fpu);
6731 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6733 if (vcpu->guest_fpu_loaded)
6737 * Restore all possible states in the guest,
6738 * and assume host would use all available bits.
6739 * Guest xcr0 would be loaded later.
6741 kvm_put_guest_xcr0(vcpu);
6742 vcpu->guest_fpu_loaded = 1;
6743 __kernel_fpu_begin();
6744 fpu_restore_checking(&vcpu->arch.guest_fpu);
6748 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6750 kvm_put_guest_xcr0(vcpu);
6752 if (!vcpu->guest_fpu_loaded)
6755 vcpu->guest_fpu_loaded = 0;
6756 fpu_save_init(&vcpu->arch.guest_fpu);
6758 ++vcpu->stat.fpu_reload;
6759 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6763 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6765 kvmclock_reset(vcpu);
6767 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6769 kvm_x86_ops->vcpu_free(vcpu);
6772 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6775 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6776 printk_once(KERN_WARNING
6777 "kvm: SMP vm created on host with unstable TSC; "
6778 "guest TSC will not be reliable\n");
6779 return kvm_x86_ops->vcpu_create(kvm, id);
6782 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6786 vcpu->arch.mtrr_state.have_fixed = 1;
6787 r = vcpu_load(vcpu);
6790 kvm_vcpu_reset(vcpu);
6791 kvm_mmu_setup(vcpu);
6797 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
6800 struct msr_data msr;
6801 struct kvm *kvm = vcpu->kvm;
6803 r = vcpu_load(vcpu);
6807 msr.index = MSR_IA32_TSC;
6808 msr.host_initiated = true;
6809 kvm_write_tsc(vcpu, &msr);
6812 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
6813 KVMCLOCK_SYNC_PERIOD);
6818 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6821 vcpu->arch.apf.msr_val = 0;
6823 r = vcpu_load(vcpu);
6825 kvm_mmu_unload(vcpu);
6829 kvm_x86_ops->vcpu_free(vcpu);
6832 void kvm_vcpu_reset(struct kvm_vcpu *vcpu)
6834 atomic_set(&vcpu->arch.nmi_queued, 0);
6835 vcpu->arch.nmi_pending = 0;
6836 vcpu->arch.nmi_injected = false;
6838 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6839 vcpu->arch.dr6 = DR6_FIXED_1;
6840 kvm_update_dr6(vcpu);
6841 vcpu->arch.dr7 = DR7_FIXED_1;
6842 kvm_update_dr7(vcpu);
6844 kvm_make_request(KVM_REQ_EVENT, vcpu);
6845 vcpu->arch.apf.msr_val = 0;
6846 vcpu->arch.st.msr_val = 0;
6848 kvmclock_reset(vcpu);
6850 kvm_clear_async_pf_completion_queue(vcpu);
6851 kvm_async_pf_hash_reset(vcpu);
6852 vcpu->arch.apf.halted = false;
6854 kvm_pmu_reset(vcpu);
6856 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
6857 vcpu->arch.regs_avail = ~0;
6858 vcpu->arch.regs_dirty = ~0;
6860 kvm_x86_ops->vcpu_reset(vcpu);
6863 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, unsigned int vector)
6865 struct kvm_segment cs;
6867 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6868 cs.selector = vector << 8;
6869 cs.base = vector << 12;
6870 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6871 kvm_rip_write(vcpu, 0);
6874 int kvm_arch_hardware_enable(void *garbage)
6877 struct kvm_vcpu *vcpu;
6882 bool stable, backwards_tsc = false;
6884 kvm_shared_msr_cpu_online();
6885 ret = kvm_x86_ops->hardware_enable(garbage);
6889 local_tsc = native_read_tsc();
6890 stable = !check_tsc_unstable();
6891 list_for_each_entry(kvm, &vm_list, vm_list) {
6892 kvm_for_each_vcpu(i, vcpu, kvm) {
6893 if (!stable && vcpu->cpu == smp_processor_id())
6894 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
6895 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
6896 backwards_tsc = true;
6897 if (vcpu->arch.last_host_tsc > max_tsc)
6898 max_tsc = vcpu->arch.last_host_tsc;
6904 * Sometimes, even reliable TSCs go backwards. This happens on
6905 * platforms that reset TSC during suspend or hibernate actions, but
6906 * maintain synchronization. We must compensate. Fortunately, we can
6907 * detect that condition here, which happens early in CPU bringup,
6908 * before any KVM threads can be running. Unfortunately, we can't
6909 * bring the TSCs fully up to date with real time, as we aren't yet far
6910 * enough into CPU bringup that we know how much real time has actually
6911 * elapsed; our helper function, get_kernel_ns() will be using boot
6912 * variables that haven't been updated yet.
6914 * So we simply find the maximum observed TSC above, then record the
6915 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
6916 * the adjustment will be applied. Note that we accumulate
6917 * adjustments, in case multiple suspend cycles happen before some VCPU
6918 * gets a chance to run again. In the event that no KVM threads get a
6919 * chance to run, we will miss the entire elapsed period, as we'll have
6920 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
6921 * loose cycle time. This isn't too big a deal, since the loss will be
6922 * uniform across all VCPUs (not to mention the scenario is extremely
6923 * unlikely). It is possible that a second hibernate recovery happens
6924 * much faster than a first, causing the observed TSC here to be
6925 * smaller; this would require additional padding adjustment, which is
6926 * why we set last_host_tsc to the local tsc observed here.
6928 * N.B. - this code below runs only on platforms with reliable TSC,
6929 * as that is the only way backwards_tsc is set above. Also note
6930 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
6931 * have the same delta_cyc adjustment applied if backwards_tsc
6932 * is detected. Note further, this adjustment is only done once,
6933 * as we reset last_host_tsc on all VCPUs to stop this from being
6934 * called multiple times (one for each physical CPU bringup).
6936 * Platforms with unreliable TSCs don't have to deal with this, they
6937 * will be compensated by the logic in vcpu_load, which sets the TSC to
6938 * catchup mode. This will catchup all VCPUs to real time, but cannot
6939 * guarantee that they stay in perfect synchronization.
6941 if (backwards_tsc) {
6942 u64 delta_cyc = max_tsc - local_tsc;
6943 list_for_each_entry(kvm, &vm_list, vm_list) {
6944 kvm_for_each_vcpu(i, vcpu, kvm) {
6945 vcpu->arch.tsc_offset_adjustment += delta_cyc;
6946 vcpu->arch.last_host_tsc = local_tsc;
6947 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
6952 * We have to disable TSC offset matching.. if you were
6953 * booting a VM while issuing an S4 host suspend....
6954 * you may have some problem. Solving this issue is
6955 * left as an exercise to the reader.
6957 kvm->arch.last_tsc_nsec = 0;
6958 kvm->arch.last_tsc_write = 0;
6965 void kvm_arch_hardware_disable(void *garbage)
6967 kvm_x86_ops->hardware_disable(garbage);
6968 drop_user_return_notifiers(garbage);
6971 int kvm_arch_hardware_setup(void)
6973 return kvm_x86_ops->hardware_setup();
6976 void kvm_arch_hardware_unsetup(void)
6978 kvm_x86_ops->hardware_unsetup();
6981 void kvm_arch_check_processor_compat(void *rtn)
6983 kvm_x86_ops->check_processor_compatibility(rtn);
6986 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
6988 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
6991 struct static_key kvm_no_apic_vcpu __read_mostly;
6993 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6999 BUG_ON(vcpu->kvm == NULL);
7002 vcpu->arch.pv.pv_unhalted = false;
7003 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7004 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
7005 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7007 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7009 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7014 vcpu->arch.pio_data = page_address(page);
7016 kvm_set_tsc_khz(vcpu, max_tsc_khz);
7018 r = kvm_mmu_create(vcpu);
7020 goto fail_free_pio_data;
7022 if (irqchip_in_kernel(kvm)) {
7023 r = kvm_create_lapic(vcpu);
7025 goto fail_mmu_destroy;
7027 static_key_slow_inc(&kvm_no_apic_vcpu);
7029 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7031 if (!vcpu->arch.mce_banks) {
7033 goto fail_free_lapic;
7035 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7037 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7039 goto fail_free_mce_banks;
7044 goto fail_free_wbinvd_dirty_mask;
7046 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7047 vcpu->arch.pv_time_enabled = false;
7049 vcpu->arch.guest_supported_xcr0 = 0;
7050 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7052 kvm_async_pf_hash_reset(vcpu);
7056 fail_free_wbinvd_dirty_mask:
7057 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7058 fail_free_mce_banks:
7059 kfree(vcpu->arch.mce_banks);
7061 kvm_free_lapic(vcpu);
7063 kvm_mmu_destroy(vcpu);
7065 free_page((unsigned long)vcpu->arch.pio_data);
7070 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7074 kvm_pmu_destroy(vcpu);
7075 kfree(vcpu->arch.mce_banks);
7076 kvm_free_lapic(vcpu);
7077 idx = srcu_read_lock(&vcpu->kvm->srcu);
7078 kvm_mmu_destroy(vcpu);
7079 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7080 free_page((unsigned long)vcpu->arch.pio_data);
7081 if (!irqchip_in_kernel(vcpu->kvm))
7082 static_key_slow_dec(&kvm_no_apic_vcpu);
7085 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7090 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7091 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7092 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7093 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7095 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7096 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7097 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7098 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7099 &kvm->arch.irq_sources_bitmap);
7101 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7102 mutex_init(&kvm->arch.apic_map_lock);
7103 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7105 pvclock_update_vm_gtod_copy(kvm);
7107 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7108 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7113 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7116 r = vcpu_load(vcpu);
7118 kvm_mmu_unload(vcpu);
7122 static void kvm_free_vcpus(struct kvm *kvm)
7125 struct kvm_vcpu *vcpu;
7128 * Unpin any mmu pages first.
7130 kvm_for_each_vcpu(i, vcpu, kvm) {
7131 kvm_clear_async_pf_completion_queue(vcpu);
7132 kvm_unload_vcpu_mmu(vcpu);
7134 kvm_for_each_vcpu(i, vcpu, kvm)
7135 kvm_arch_vcpu_free(vcpu);
7137 mutex_lock(&kvm->lock);
7138 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7139 kvm->vcpus[i] = NULL;
7141 atomic_set(&kvm->online_vcpus, 0);
7142 mutex_unlock(&kvm->lock);
7145 void kvm_arch_sync_events(struct kvm *kvm)
7147 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7148 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7149 kvm_free_all_assigned_devices(kvm);
7153 void kvm_arch_destroy_vm(struct kvm *kvm)
7155 if (current->mm == kvm->mm) {
7157 * Free memory regions allocated on behalf of userspace,
7158 * unless the the memory map has changed due to process exit
7161 struct kvm_userspace_memory_region mem;
7162 memset(&mem, 0, sizeof(mem));
7163 mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
7164 kvm_set_memory_region(kvm, &mem);
7166 mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
7167 kvm_set_memory_region(kvm, &mem);
7169 mem.slot = TSS_PRIVATE_MEMSLOT;
7170 kvm_set_memory_region(kvm, &mem);
7172 kvm_iommu_unmap_guest(kvm);
7173 kfree(kvm->arch.vpic);
7174 kfree(kvm->arch.vioapic);
7175 kvm_free_vcpus(kvm);
7176 if (kvm->arch.apic_access_page)
7177 put_page(kvm->arch.apic_access_page);
7178 if (kvm->arch.ept_identity_pagetable)
7179 put_page(kvm->arch.ept_identity_pagetable);
7180 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7183 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7184 struct kvm_memory_slot *dont)
7188 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7189 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7190 kvm_kvfree(free->arch.rmap[i]);
7191 free->arch.rmap[i] = NULL;
7196 if (!dont || free->arch.lpage_info[i - 1] !=
7197 dont->arch.lpage_info[i - 1]) {
7198 kvm_kvfree(free->arch.lpage_info[i - 1]);
7199 free->arch.lpage_info[i - 1] = NULL;
7204 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7205 unsigned long npages)
7209 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7214 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7215 slot->base_gfn, level) + 1;
7217 slot->arch.rmap[i] =
7218 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7219 if (!slot->arch.rmap[i])
7224 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7225 sizeof(*slot->arch.lpage_info[i - 1]));
7226 if (!slot->arch.lpage_info[i - 1])
7229 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7230 slot->arch.lpage_info[i - 1][0].write_count = 1;
7231 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7232 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7233 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7235 * If the gfn and userspace address are not aligned wrt each
7236 * other, or if explicitly asked to, disable large page
7237 * support for this slot
7239 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7240 !kvm_largepages_enabled()) {
7243 for (j = 0; j < lpages; ++j)
7244 slot->arch.lpage_info[i - 1][j].write_count = 1;
7251 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7252 kvm_kvfree(slot->arch.rmap[i]);
7253 slot->arch.rmap[i] = NULL;
7257 kvm_kvfree(slot->arch.lpage_info[i - 1]);
7258 slot->arch.lpage_info[i - 1] = NULL;
7263 void kvm_arch_memslots_updated(struct kvm *kvm)
7266 * memslots->generation has been incremented.
7267 * mmio generation may have reached its maximum value.
7269 kvm_mmu_invalidate_mmio_sptes(kvm);
7272 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7273 struct kvm_memory_slot *memslot,
7274 struct kvm_userspace_memory_region *mem,
7275 enum kvm_mr_change change)
7278 * Only private memory slots need to be mapped here since
7279 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
7281 if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
7282 unsigned long userspace_addr;
7285 * MAP_SHARED to prevent internal slot pages from being moved
7288 userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
7289 PROT_READ | PROT_WRITE,
7290 MAP_SHARED | MAP_ANONYMOUS, 0);
7292 if (IS_ERR((void *)userspace_addr))
7293 return PTR_ERR((void *)userspace_addr);
7295 memslot->userspace_addr = userspace_addr;
7301 void kvm_arch_commit_memory_region(struct kvm *kvm,
7302 struct kvm_userspace_memory_region *mem,
7303 const struct kvm_memory_slot *old,
7304 enum kvm_mr_change change)
7307 int nr_mmu_pages = 0;
7309 if ((mem->slot >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_DELETE)) {
7312 ret = vm_munmap(old->userspace_addr,
7313 old->npages * PAGE_SIZE);
7316 "kvm_vm_ioctl_set_memory_region: "
7317 "failed to munmap memory\n");
7320 if (!kvm->arch.n_requested_mmu_pages)
7321 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7324 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7326 * Write protect all pages for dirty logging.
7327 * Existing largepage mappings are destroyed here and new ones will
7328 * not be created until the end of the logging.
7330 if ((change != KVM_MR_DELETE) && (mem->flags & KVM_MEM_LOG_DIRTY_PAGES))
7331 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
7334 void kvm_arch_flush_shadow_all(struct kvm *kvm)
7336 kvm_mmu_invalidate_zap_all_pages(kvm);
7339 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7340 struct kvm_memory_slot *slot)
7342 kvm_mmu_invalidate_zap_all_pages(kvm);
7345 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
7347 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7348 kvm_x86_ops->check_nested_events(vcpu, false);
7350 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7351 !vcpu->arch.apf.halted)
7352 || !list_empty_careful(&vcpu->async_pf.done)
7353 || kvm_apic_has_events(vcpu)
7354 || vcpu->arch.pv.pv_unhalted
7355 || atomic_read(&vcpu->arch.nmi_queued) ||
7356 (kvm_arch_interrupt_allowed(vcpu) &&
7357 kvm_cpu_has_interrupt(vcpu));
7360 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
7362 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
7365 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7367 return kvm_x86_ops->interrupt_allowed(vcpu);
7370 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7372 unsigned long current_rip = kvm_rip_read(vcpu) +
7373 get_segment_base(vcpu, VCPU_SREG_CS);
7375 return current_rip == linear_rip;
7377 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7379 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7381 unsigned long rflags;
7383 rflags = kvm_x86_ops->get_rflags(vcpu);
7384 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7385 rflags &= ~X86_EFLAGS_TF;
7388 EXPORT_SYMBOL_GPL(kvm_get_rflags);
7390 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7392 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
7393 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
7394 rflags |= X86_EFLAGS_TF;
7395 kvm_x86_ops->set_rflags(vcpu, rflags);
7396 kvm_make_request(KVM_REQ_EVENT, vcpu);
7398 EXPORT_SYMBOL_GPL(kvm_set_rflags);
7400 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7404 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
7408 r = kvm_mmu_reload(vcpu);
7412 if (!vcpu->arch.mmu.direct_map &&
7413 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7416 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7419 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7421 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7424 static inline u32 kvm_async_pf_next_probe(u32 key)
7426 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7429 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7431 u32 key = kvm_async_pf_hash_fn(gfn);
7433 while (vcpu->arch.apf.gfns[key] != ~0)
7434 key = kvm_async_pf_next_probe(key);
7436 vcpu->arch.apf.gfns[key] = gfn;
7439 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7442 u32 key = kvm_async_pf_hash_fn(gfn);
7444 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
7445 (vcpu->arch.apf.gfns[key] != gfn &&
7446 vcpu->arch.apf.gfns[key] != ~0); i++)
7447 key = kvm_async_pf_next_probe(key);
7452 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7454 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7457 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7461 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7463 vcpu->arch.apf.gfns[i] = ~0;
7465 j = kvm_async_pf_next_probe(j);
7466 if (vcpu->arch.apf.gfns[j] == ~0)
7468 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7470 * k lies cyclically in ]i,j]
7472 * |....j i.k.| or |.k..j i...|
7474 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7475 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7480 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7483 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7487 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7488 struct kvm_async_pf *work)
7490 struct x86_exception fault;
7492 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
7493 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7495 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
7496 (vcpu->arch.apf.send_user_only &&
7497 kvm_x86_ops->get_cpl(vcpu) == 0))
7498 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7499 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
7500 fault.vector = PF_VECTOR;
7501 fault.error_code_valid = true;
7502 fault.error_code = 0;
7503 fault.nested_page_fault = false;
7504 fault.address = work->arch.token;
7505 kvm_inject_page_fault(vcpu, &fault);
7509 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7510 struct kvm_async_pf *work)
7512 struct x86_exception fault;
7514 trace_kvm_async_pf_ready(work->arch.token, work->gva);
7515 if (work->wakeup_all)
7516 work->arch.token = ~0; /* broadcast wakeup */
7518 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7520 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7521 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
7522 fault.vector = PF_VECTOR;
7523 fault.error_code_valid = true;
7524 fault.error_code = 0;
7525 fault.nested_page_fault = false;
7526 fault.address = work->arch.token;
7527 kvm_inject_page_fault(vcpu, &fault);
7529 vcpu->arch.apf.halted = false;
7530 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7533 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7535 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7538 return !kvm_event_needs_reinjection(vcpu) &&
7539 kvm_x86_ops->interrupt_allowed(vcpu);
7542 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
7544 atomic_inc(&kvm->arch.noncoherent_dma_count);
7546 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
7548 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
7550 atomic_dec(&kvm->arch.noncoherent_dma_count);
7552 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
7554 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
7556 return atomic_read(&kvm->arch.noncoherent_dma_count);
7558 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
7560 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
7561 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
7562 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
7563 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
7564 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
7565 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
7566 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
7567 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
7568 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
7569 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
7570 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
7571 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
7572 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);