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) && (cr3 & CR3_L_MODE_RESERVED_BITS))
706 if (is_pae(vcpu) && is_paging(vcpu) &&
707 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
710 vcpu->arch.cr3 = cr3;
711 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
712 kvm_mmu_new_cr3(vcpu);
715 EXPORT_SYMBOL_GPL(kvm_set_cr3);
717 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
719 if (cr8 & CR8_RESERVED_BITS)
721 if (irqchip_in_kernel(vcpu->kvm))
722 kvm_lapic_set_tpr(vcpu, cr8);
724 vcpu->arch.cr8 = cr8;
727 EXPORT_SYMBOL_GPL(kvm_set_cr8);
729 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
731 if (irqchip_in_kernel(vcpu->kvm))
732 return kvm_lapic_get_cr8(vcpu);
734 return vcpu->arch.cr8;
736 EXPORT_SYMBOL_GPL(kvm_get_cr8);
738 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
740 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
741 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
744 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
748 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
749 dr7 = vcpu->arch.guest_debug_dr7;
751 dr7 = vcpu->arch.dr7;
752 kvm_x86_ops->set_dr7(vcpu, dr7);
753 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
754 if (dr7 & DR7_BP_EN_MASK)
755 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
758 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
762 vcpu->arch.db[dr] = val;
763 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
764 vcpu->arch.eff_db[dr] = val;
767 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
771 if (val & 0xffffffff00000000ULL)
773 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
774 kvm_update_dr6(vcpu);
777 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
781 if (val & 0xffffffff00000000ULL)
783 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
784 kvm_update_dr7(vcpu);
791 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
795 res = __kvm_set_dr(vcpu, dr, val);
797 kvm_queue_exception(vcpu, UD_VECTOR);
799 kvm_inject_gp(vcpu, 0);
803 EXPORT_SYMBOL_GPL(kvm_set_dr);
805 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
809 *val = vcpu->arch.db[dr];
812 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
816 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
817 *val = vcpu->arch.dr6;
819 *val = kvm_x86_ops->get_dr6(vcpu);
822 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
826 *val = vcpu->arch.dr7;
833 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
835 if (_kvm_get_dr(vcpu, dr, val)) {
836 kvm_queue_exception(vcpu, UD_VECTOR);
841 EXPORT_SYMBOL_GPL(kvm_get_dr);
843 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
845 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
849 err = kvm_pmu_read_pmc(vcpu, ecx, &data);
852 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
853 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
856 EXPORT_SYMBOL_GPL(kvm_rdpmc);
859 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
860 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
862 * This list is modified at module load time to reflect the
863 * capabilities of the host cpu. This capabilities test skips MSRs that are
864 * kvm-specific. Those are put in the beginning of the list.
867 #define KVM_SAVE_MSRS_BEGIN 12
868 static u32 msrs_to_save[] = {
869 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
870 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
871 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
872 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
873 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
875 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
878 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
880 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
881 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS
884 static unsigned num_msrs_to_save;
886 static const u32 emulated_msrs[] = {
888 MSR_IA32_TSCDEADLINE,
889 MSR_IA32_MISC_ENABLE,
894 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
896 if (efer & efer_reserved_bits)
899 if (efer & EFER_FFXSR) {
900 struct kvm_cpuid_entry2 *feat;
902 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
903 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
907 if (efer & EFER_SVME) {
908 struct kvm_cpuid_entry2 *feat;
910 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
911 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
917 EXPORT_SYMBOL_GPL(kvm_valid_efer);
919 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
921 u64 old_efer = vcpu->arch.efer;
923 if (!kvm_valid_efer(vcpu, efer))
927 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
931 efer |= vcpu->arch.efer & EFER_LMA;
933 kvm_x86_ops->set_efer(vcpu, efer);
935 /* Update reserved bits */
936 if ((efer ^ old_efer) & EFER_NX)
937 kvm_mmu_reset_context(vcpu);
942 void kvm_enable_efer_bits(u64 mask)
944 efer_reserved_bits &= ~mask;
946 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
950 * Writes msr value into into the appropriate "register".
951 * Returns 0 on success, non-0 otherwise.
952 * Assumes vcpu_load() was already called.
954 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
956 return kvm_x86_ops->set_msr(vcpu, msr);
960 * Adapt set_msr() to msr_io()'s calling convention
962 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
968 msr.host_initiated = true;
969 return kvm_set_msr(vcpu, &msr);
973 struct pvclock_gtod_data {
976 struct { /* extract of a clocksource struct */
984 /* open coded 'struct timespec' */
985 u64 monotonic_time_snsec;
986 time_t monotonic_time_sec;
989 static struct pvclock_gtod_data pvclock_gtod_data;
991 static void update_pvclock_gtod(struct timekeeper *tk)
993 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
995 write_seqcount_begin(&vdata->seq);
997 /* copy pvclock gtod data */
998 vdata->clock.vclock_mode = tk->clock->archdata.vclock_mode;
999 vdata->clock.cycle_last = tk->clock->cycle_last;
1000 vdata->clock.mask = tk->clock->mask;
1001 vdata->clock.mult = tk->mult;
1002 vdata->clock.shift = tk->shift;
1004 vdata->monotonic_time_sec = tk->xtime_sec
1005 + tk->wall_to_monotonic.tv_sec;
1006 vdata->monotonic_time_snsec = tk->xtime_nsec
1007 + (tk->wall_to_monotonic.tv_nsec
1009 while (vdata->monotonic_time_snsec >=
1010 (((u64)NSEC_PER_SEC) << tk->shift)) {
1011 vdata->monotonic_time_snsec -=
1012 ((u64)NSEC_PER_SEC) << tk->shift;
1013 vdata->monotonic_time_sec++;
1016 write_seqcount_end(&vdata->seq);
1021 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1025 struct pvclock_wall_clock wc;
1026 struct timespec boot;
1031 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1036 ++version; /* first time write, random junk */
1040 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1043 * The guest calculates current wall clock time by adding
1044 * system time (updated by kvm_guest_time_update below) to the
1045 * wall clock specified here. guest system time equals host
1046 * system time for us, thus we must fill in host boot time here.
1050 if (kvm->arch.kvmclock_offset) {
1051 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1052 boot = timespec_sub(boot, ts);
1054 wc.sec = boot.tv_sec;
1055 wc.nsec = boot.tv_nsec;
1056 wc.version = version;
1058 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1061 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1064 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1066 uint32_t quotient, remainder;
1068 /* Don't try to replace with do_div(), this one calculates
1069 * "(dividend << 32) / divisor" */
1071 : "=a" (quotient), "=d" (remainder)
1072 : "0" (0), "1" (dividend), "r" (divisor) );
1076 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1077 s8 *pshift, u32 *pmultiplier)
1084 tps64 = base_khz * 1000LL;
1085 scaled64 = scaled_khz * 1000LL;
1086 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1091 tps32 = (uint32_t)tps64;
1092 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1093 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1101 *pmultiplier = div_frac(scaled64, tps32);
1103 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1104 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1107 static inline u64 get_kernel_ns(void)
1112 monotonic_to_bootbased(&ts);
1113 return timespec_to_ns(&ts);
1116 #ifdef CONFIG_X86_64
1117 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1120 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1121 unsigned long max_tsc_khz;
1123 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1125 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1126 vcpu->arch.virtual_tsc_shift);
1129 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1131 u64 v = (u64)khz * (1000000 + ppm);
1136 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1138 u32 thresh_lo, thresh_hi;
1139 int use_scaling = 0;
1141 /* tsc_khz can be zero if TSC calibration fails */
1142 if (this_tsc_khz == 0)
1145 /* Compute a scale to convert nanoseconds in TSC cycles */
1146 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1147 &vcpu->arch.virtual_tsc_shift,
1148 &vcpu->arch.virtual_tsc_mult);
1149 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1152 * Compute the variation in TSC rate which is acceptable
1153 * within the range of tolerance and decide if the
1154 * rate being applied is within that bounds of the hardware
1155 * rate. If so, no scaling or compensation need be done.
1157 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1158 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1159 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1160 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1163 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1166 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1168 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1169 vcpu->arch.virtual_tsc_mult,
1170 vcpu->arch.virtual_tsc_shift);
1171 tsc += vcpu->arch.this_tsc_write;
1175 void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1177 #ifdef CONFIG_X86_64
1179 bool do_request = false;
1180 struct kvm_arch *ka = &vcpu->kvm->arch;
1181 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1183 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1184 atomic_read(&vcpu->kvm->online_vcpus));
1186 if (vcpus_matched && gtod->clock.vclock_mode == VCLOCK_TSC)
1187 if (!ka->use_master_clock)
1190 if (!vcpus_matched && ka->use_master_clock)
1194 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1196 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1197 atomic_read(&vcpu->kvm->online_vcpus),
1198 ka->use_master_clock, gtod->clock.vclock_mode);
1202 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1204 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1205 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1208 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1210 struct kvm *kvm = vcpu->kvm;
1211 u64 offset, ns, elapsed;
1212 unsigned long flags;
1215 u64 data = msr->data;
1217 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1218 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1219 ns = get_kernel_ns();
1220 elapsed = ns - kvm->arch.last_tsc_nsec;
1222 if (vcpu->arch.virtual_tsc_khz) {
1225 /* n.b - signed multiplication and division required */
1226 usdiff = data - kvm->arch.last_tsc_write;
1227 #ifdef CONFIG_X86_64
1228 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1230 /* do_div() only does unsigned */
1231 asm("1: idivl %[divisor]\n"
1232 "2: xor %%edx, %%edx\n"
1233 " movl $0, %[faulted]\n"
1235 ".section .fixup,\"ax\"\n"
1236 "4: movl $1, %[faulted]\n"
1240 _ASM_EXTABLE(1b, 4b)
1242 : "=A"(usdiff), [faulted] "=r" (faulted)
1243 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1246 do_div(elapsed, 1000);
1251 /* idivl overflow => difference is larger than USEC_PER_SEC */
1253 usdiff = USEC_PER_SEC;
1255 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1258 * Special case: TSC write with a small delta (1 second) of virtual
1259 * cycle time against real time is interpreted as an attempt to
1260 * synchronize the CPU.
1262 * For a reliable TSC, we can match TSC offsets, and for an unstable
1263 * TSC, we add elapsed time in this computation. We could let the
1264 * compensation code attempt to catch up if we fall behind, but
1265 * it's better to try to match offsets from the beginning.
1267 if (usdiff < USEC_PER_SEC &&
1268 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1269 if (!check_tsc_unstable()) {
1270 offset = kvm->arch.cur_tsc_offset;
1271 pr_debug("kvm: matched tsc offset for %llu\n", data);
1273 u64 delta = nsec_to_cycles(vcpu, elapsed);
1275 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1276 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1281 * We split periods of matched TSC writes into generations.
1282 * For each generation, we track the original measured
1283 * nanosecond time, offset, and write, so if TSCs are in
1284 * sync, we can match exact offset, and if not, we can match
1285 * exact software computation in compute_guest_tsc()
1287 * These values are tracked in kvm->arch.cur_xxx variables.
1289 kvm->arch.cur_tsc_generation++;
1290 kvm->arch.cur_tsc_nsec = ns;
1291 kvm->arch.cur_tsc_write = data;
1292 kvm->arch.cur_tsc_offset = offset;
1294 pr_debug("kvm: new tsc generation %u, clock %llu\n",
1295 kvm->arch.cur_tsc_generation, data);
1299 * We also track th most recent recorded KHZ, write and time to
1300 * allow the matching interval to be extended at each write.
1302 kvm->arch.last_tsc_nsec = ns;
1303 kvm->arch.last_tsc_write = data;
1304 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1306 vcpu->arch.last_guest_tsc = data;
1308 /* Keep track of which generation this VCPU has synchronized to */
1309 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1310 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1311 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1313 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1314 update_ia32_tsc_adjust_msr(vcpu, offset);
1315 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1316 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1318 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1320 kvm->arch.nr_vcpus_matched_tsc++;
1322 kvm->arch.nr_vcpus_matched_tsc = 0;
1324 kvm_track_tsc_matching(vcpu);
1325 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1328 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1330 #ifdef CONFIG_X86_64
1332 static cycle_t read_tsc(void)
1338 * Empirically, a fence (of type that depends on the CPU)
1339 * before rdtsc is enough to ensure that rdtsc is ordered
1340 * with respect to loads. The various CPU manuals are unclear
1341 * as to whether rdtsc can be reordered with later loads,
1342 * but no one has ever seen it happen.
1345 ret = (cycle_t)vget_cycles();
1347 last = pvclock_gtod_data.clock.cycle_last;
1349 if (likely(ret >= last))
1353 * GCC likes to generate cmov here, but this branch is extremely
1354 * predictable (it's just a funciton of time and the likely is
1355 * very likely) and there's a data dependence, so force GCC
1356 * to generate a branch instead. I don't barrier() because
1357 * we don't actually need a barrier, and if this function
1358 * ever gets inlined it will generate worse code.
1364 static inline u64 vgettsc(cycle_t *cycle_now)
1367 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1369 *cycle_now = read_tsc();
1371 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1372 return v * gtod->clock.mult;
1375 static int do_monotonic(struct timespec *ts, cycle_t *cycle_now)
1380 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1384 seq = read_seqcount_begin(>od->seq);
1385 mode = gtod->clock.vclock_mode;
1386 ts->tv_sec = gtod->monotonic_time_sec;
1387 ns = gtod->monotonic_time_snsec;
1388 ns += vgettsc(cycle_now);
1389 ns >>= gtod->clock.shift;
1390 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1391 timespec_add_ns(ts, ns);
1396 /* returns true if host is using tsc clocksource */
1397 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1401 /* checked again under seqlock below */
1402 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1405 if (do_monotonic(&ts, cycle_now) != VCLOCK_TSC)
1408 monotonic_to_bootbased(&ts);
1409 *kernel_ns = timespec_to_ns(&ts);
1417 * Assuming a stable TSC across physical CPUS, and a stable TSC
1418 * across virtual CPUs, the following condition is possible.
1419 * Each numbered line represents an event visible to both
1420 * CPUs at the next numbered event.
1422 * "timespecX" represents host monotonic time. "tscX" represents
1425 * VCPU0 on CPU0 | VCPU1 on CPU1
1427 * 1. read timespec0,tsc0
1428 * 2. | timespec1 = timespec0 + N
1430 * 3. transition to guest | transition to guest
1431 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1432 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1433 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1435 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1438 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1440 * - 0 < N - M => M < N
1442 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1443 * always the case (the difference between two distinct xtime instances
1444 * might be smaller then the difference between corresponding TSC reads,
1445 * when updating guest vcpus pvclock areas).
1447 * To avoid that problem, do not allow visibility of distinct
1448 * system_timestamp/tsc_timestamp values simultaneously: use a master
1449 * copy of host monotonic time values. Update that master copy
1452 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1456 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1458 #ifdef CONFIG_X86_64
1459 struct kvm_arch *ka = &kvm->arch;
1461 bool host_tsc_clocksource, vcpus_matched;
1463 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1464 atomic_read(&kvm->online_vcpus));
1467 * If the host uses TSC clock, then passthrough TSC as stable
1470 host_tsc_clocksource = kvm_get_time_and_clockread(
1471 &ka->master_kernel_ns,
1472 &ka->master_cycle_now);
1474 ka->use_master_clock = host_tsc_clocksource & vcpus_matched;
1476 if (ka->use_master_clock)
1477 atomic_set(&kvm_guest_has_master_clock, 1);
1479 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1480 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1485 static void kvm_gen_update_masterclock(struct kvm *kvm)
1487 #ifdef CONFIG_X86_64
1489 struct kvm_vcpu *vcpu;
1490 struct kvm_arch *ka = &kvm->arch;
1492 spin_lock(&ka->pvclock_gtod_sync_lock);
1493 kvm_make_mclock_inprogress_request(kvm);
1494 /* no guest entries from this point */
1495 pvclock_update_vm_gtod_copy(kvm);
1497 kvm_for_each_vcpu(i, vcpu, kvm)
1498 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
1500 /* guest entries allowed */
1501 kvm_for_each_vcpu(i, vcpu, kvm)
1502 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1504 spin_unlock(&ka->pvclock_gtod_sync_lock);
1508 static int kvm_guest_time_update(struct kvm_vcpu *v)
1510 unsigned long flags, this_tsc_khz;
1511 struct kvm_vcpu_arch *vcpu = &v->arch;
1512 struct kvm_arch *ka = &v->kvm->arch;
1514 u64 tsc_timestamp, host_tsc;
1515 struct pvclock_vcpu_time_info guest_hv_clock;
1517 bool use_master_clock;
1523 * If the host uses TSC clock, then passthrough TSC as stable
1526 spin_lock(&ka->pvclock_gtod_sync_lock);
1527 use_master_clock = ka->use_master_clock;
1528 if (use_master_clock) {
1529 host_tsc = ka->master_cycle_now;
1530 kernel_ns = ka->master_kernel_ns;
1532 spin_unlock(&ka->pvclock_gtod_sync_lock);
1534 /* Keep irq disabled to prevent changes to the clock */
1535 local_irq_save(flags);
1536 this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
1537 if (unlikely(this_tsc_khz == 0)) {
1538 local_irq_restore(flags);
1539 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1542 if (!use_master_clock) {
1543 host_tsc = native_read_tsc();
1544 kernel_ns = get_kernel_ns();
1547 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1550 * We may have to catch up the TSC to match elapsed wall clock
1551 * time for two reasons, even if kvmclock is used.
1552 * 1) CPU could have been running below the maximum TSC rate
1553 * 2) Broken TSC compensation resets the base at each VCPU
1554 * entry to avoid unknown leaps of TSC even when running
1555 * again on the same CPU. This may cause apparent elapsed
1556 * time to disappear, and the guest to stand still or run
1559 if (vcpu->tsc_catchup) {
1560 u64 tsc = compute_guest_tsc(v, kernel_ns);
1561 if (tsc > tsc_timestamp) {
1562 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1563 tsc_timestamp = tsc;
1567 local_irq_restore(flags);
1569 if (!vcpu->pv_time_enabled)
1572 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1573 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1574 &vcpu->hv_clock.tsc_shift,
1575 &vcpu->hv_clock.tsc_to_system_mul);
1576 vcpu->hw_tsc_khz = this_tsc_khz;
1579 /* With all the info we got, fill in the values */
1580 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1581 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1582 vcpu->last_guest_tsc = tsc_timestamp;
1585 * The interface expects us to write an even number signaling that the
1586 * update is finished. Since the guest won't see the intermediate
1587 * state, we just increase by 2 at the end.
1589 vcpu->hv_clock.version += 2;
1591 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1592 &guest_hv_clock, sizeof(guest_hv_clock))))
1595 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1596 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1598 if (vcpu->pvclock_set_guest_stopped_request) {
1599 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1600 vcpu->pvclock_set_guest_stopped_request = false;
1603 /* If the host uses TSC clocksource, then it is stable */
1604 if (use_master_clock)
1605 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1607 vcpu->hv_clock.flags = pvclock_flags;
1609 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1611 sizeof(vcpu->hv_clock));
1616 * kvmclock updates which are isolated to a given vcpu, such as
1617 * vcpu->cpu migration, should not allow system_timestamp from
1618 * the rest of the vcpus to remain static. Otherwise ntp frequency
1619 * correction applies to one vcpu's system_timestamp but not
1622 * So in those cases, request a kvmclock update for all vcpus.
1623 * We need to rate-limit these requests though, as they can
1624 * considerably slow guests that have a large number of vcpus.
1625 * The time for a remote vcpu to update its kvmclock is bound
1626 * by the delay we use to rate-limit the updates.
1629 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1631 static void kvmclock_update_fn(struct work_struct *work)
1634 struct delayed_work *dwork = to_delayed_work(work);
1635 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1636 kvmclock_update_work);
1637 struct kvm *kvm = container_of(ka, struct kvm, arch);
1638 struct kvm_vcpu *vcpu;
1640 kvm_for_each_vcpu(i, vcpu, kvm) {
1641 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
1642 kvm_vcpu_kick(vcpu);
1646 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1648 struct kvm *kvm = v->kvm;
1650 set_bit(KVM_REQ_CLOCK_UPDATE, &v->requests);
1651 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1652 KVMCLOCK_UPDATE_DELAY);
1655 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1657 static void kvmclock_sync_fn(struct work_struct *work)
1659 struct delayed_work *dwork = to_delayed_work(work);
1660 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1661 kvmclock_sync_work);
1662 struct kvm *kvm = container_of(ka, struct kvm, arch);
1664 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1665 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1666 KVMCLOCK_SYNC_PERIOD);
1669 static bool msr_mtrr_valid(unsigned msr)
1672 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1673 case MSR_MTRRfix64K_00000:
1674 case MSR_MTRRfix16K_80000:
1675 case MSR_MTRRfix16K_A0000:
1676 case MSR_MTRRfix4K_C0000:
1677 case MSR_MTRRfix4K_C8000:
1678 case MSR_MTRRfix4K_D0000:
1679 case MSR_MTRRfix4K_D8000:
1680 case MSR_MTRRfix4K_E0000:
1681 case MSR_MTRRfix4K_E8000:
1682 case MSR_MTRRfix4K_F0000:
1683 case MSR_MTRRfix4K_F8000:
1684 case MSR_MTRRdefType:
1685 case MSR_IA32_CR_PAT:
1693 static bool valid_pat_type(unsigned t)
1695 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1698 static bool valid_mtrr_type(unsigned t)
1700 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1703 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1707 if (!msr_mtrr_valid(msr))
1710 if (msr == MSR_IA32_CR_PAT) {
1711 for (i = 0; i < 8; i++)
1712 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1715 } else if (msr == MSR_MTRRdefType) {
1718 return valid_mtrr_type(data & 0xff);
1719 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1720 for (i = 0; i < 8 ; i++)
1721 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1726 /* variable MTRRs */
1727 return valid_mtrr_type(data & 0xff);
1730 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1732 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1734 if (!mtrr_valid(vcpu, msr, data))
1737 if (msr == MSR_MTRRdefType) {
1738 vcpu->arch.mtrr_state.def_type = data;
1739 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1740 } else if (msr == MSR_MTRRfix64K_00000)
1742 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1743 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1744 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1745 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1746 else if (msr == MSR_IA32_CR_PAT)
1747 vcpu->arch.pat = data;
1748 else { /* Variable MTRRs */
1749 int idx, is_mtrr_mask;
1752 idx = (msr - 0x200) / 2;
1753 is_mtrr_mask = msr - 0x200 - 2 * idx;
1756 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1759 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1763 kvm_mmu_reset_context(vcpu);
1767 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1769 u64 mcg_cap = vcpu->arch.mcg_cap;
1770 unsigned bank_num = mcg_cap & 0xff;
1773 case MSR_IA32_MCG_STATUS:
1774 vcpu->arch.mcg_status = data;
1776 case MSR_IA32_MCG_CTL:
1777 if (!(mcg_cap & MCG_CTL_P))
1779 if (data != 0 && data != ~(u64)0)
1781 vcpu->arch.mcg_ctl = data;
1784 if (msr >= MSR_IA32_MC0_CTL &&
1785 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1786 u32 offset = msr - MSR_IA32_MC0_CTL;
1787 /* only 0 or all 1s can be written to IA32_MCi_CTL
1788 * some Linux kernels though clear bit 10 in bank 4 to
1789 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1790 * this to avoid an uncatched #GP in the guest
1792 if ((offset & 0x3) == 0 &&
1793 data != 0 && (data | (1 << 10)) != ~(u64)0)
1795 vcpu->arch.mce_banks[offset] = data;
1803 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1805 struct kvm *kvm = vcpu->kvm;
1806 int lm = is_long_mode(vcpu);
1807 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1808 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1809 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1810 : kvm->arch.xen_hvm_config.blob_size_32;
1811 u32 page_num = data & ~PAGE_MASK;
1812 u64 page_addr = data & PAGE_MASK;
1817 if (page_num >= blob_size)
1820 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1825 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1834 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1836 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1839 static bool kvm_hv_msr_partition_wide(u32 msr)
1843 case HV_X64_MSR_GUEST_OS_ID:
1844 case HV_X64_MSR_HYPERCALL:
1845 case HV_X64_MSR_REFERENCE_TSC:
1846 case HV_X64_MSR_TIME_REF_COUNT:
1854 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1856 struct kvm *kvm = vcpu->kvm;
1859 case HV_X64_MSR_GUEST_OS_ID:
1860 kvm->arch.hv_guest_os_id = data;
1861 /* setting guest os id to zero disables hypercall page */
1862 if (!kvm->arch.hv_guest_os_id)
1863 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1865 case HV_X64_MSR_HYPERCALL: {
1870 /* if guest os id is not set hypercall should remain disabled */
1871 if (!kvm->arch.hv_guest_os_id)
1873 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1874 kvm->arch.hv_hypercall = data;
1877 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1878 addr = gfn_to_hva(kvm, gfn);
1879 if (kvm_is_error_hva(addr))
1881 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1882 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1883 if (__copy_to_user((void __user *)addr, instructions, 4))
1885 kvm->arch.hv_hypercall = data;
1886 mark_page_dirty(kvm, gfn);
1889 case HV_X64_MSR_REFERENCE_TSC: {
1891 HV_REFERENCE_TSC_PAGE tsc_ref;
1892 memset(&tsc_ref, 0, sizeof(tsc_ref));
1893 kvm->arch.hv_tsc_page = data;
1894 if (!(data & HV_X64_MSR_TSC_REFERENCE_ENABLE))
1896 gfn = data >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
1897 if (kvm_write_guest(kvm, data,
1898 &tsc_ref, sizeof(tsc_ref)))
1900 mark_page_dirty(kvm, gfn);
1904 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1905 "data 0x%llx\n", msr, data);
1911 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1914 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1918 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1919 vcpu->arch.hv_vapic = data;
1920 if (kvm_lapic_enable_pv_eoi(vcpu, 0))
1924 gfn = data >> HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT;
1925 addr = gfn_to_hva(vcpu->kvm, gfn);
1926 if (kvm_is_error_hva(addr))
1928 if (__clear_user((void __user *)addr, PAGE_SIZE))
1930 vcpu->arch.hv_vapic = data;
1931 mark_page_dirty(vcpu->kvm, gfn);
1932 if (kvm_lapic_enable_pv_eoi(vcpu, gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
1936 case HV_X64_MSR_EOI:
1937 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1938 case HV_X64_MSR_ICR:
1939 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1940 case HV_X64_MSR_TPR:
1941 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1943 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1944 "data 0x%llx\n", msr, data);
1951 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1953 gpa_t gpa = data & ~0x3f;
1955 /* Bits 2:5 are reserved, Should be zero */
1959 vcpu->arch.apf.msr_val = data;
1961 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1962 kvm_clear_async_pf_completion_queue(vcpu);
1963 kvm_async_pf_hash_reset(vcpu);
1967 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1971 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1972 kvm_async_pf_wakeup_all(vcpu);
1976 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1978 vcpu->arch.pv_time_enabled = false;
1981 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1985 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1988 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1989 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1990 vcpu->arch.st.accum_steal = delta;
1993 static void record_steal_time(struct kvm_vcpu *vcpu)
1995 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1998 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1999 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2002 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
2003 vcpu->arch.st.steal.version += 2;
2004 vcpu->arch.st.accum_steal = 0;
2006 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2007 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2010 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2013 u32 msr = msr_info->index;
2014 u64 data = msr_info->data;
2017 case MSR_AMD64_NB_CFG:
2018 case MSR_IA32_UCODE_REV:
2019 case MSR_IA32_UCODE_WRITE:
2020 case MSR_VM_HSAVE_PA:
2021 case MSR_AMD64_PATCH_LOADER:
2022 case MSR_AMD64_BU_CFG2:
2026 return set_efer(vcpu, data);
2028 data &= ~(u64)0x40; /* ignore flush filter disable */
2029 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2030 data &= ~(u64)0x8; /* ignore TLB cache disable */
2032 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2037 case MSR_FAM10H_MMIO_CONF_BASE:
2039 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2044 case MSR_IA32_DEBUGCTLMSR:
2046 /* We support the non-activated case already */
2048 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2049 /* Values other than LBR and BTF are vendor-specific,
2050 thus reserved and should throw a #GP */
2053 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2056 case 0x200 ... 0x2ff:
2057 return set_msr_mtrr(vcpu, msr, data);
2058 case MSR_IA32_APICBASE:
2059 return kvm_set_apic_base(vcpu, msr_info);
2060 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2061 return kvm_x2apic_msr_write(vcpu, msr, data);
2062 case MSR_IA32_TSCDEADLINE:
2063 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2065 case MSR_IA32_TSC_ADJUST:
2066 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2067 if (!msr_info->host_initiated) {
2068 u64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2069 kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
2071 vcpu->arch.ia32_tsc_adjust_msr = data;
2074 case MSR_IA32_MISC_ENABLE:
2075 vcpu->arch.ia32_misc_enable_msr = data;
2077 case MSR_KVM_WALL_CLOCK_NEW:
2078 case MSR_KVM_WALL_CLOCK:
2079 vcpu->kvm->arch.wall_clock = data;
2080 kvm_write_wall_clock(vcpu->kvm, data);
2082 case MSR_KVM_SYSTEM_TIME_NEW:
2083 case MSR_KVM_SYSTEM_TIME: {
2085 kvmclock_reset(vcpu);
2087 vcpu->arch.time = data;
2088 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2090 /* we verify if the enable bit is set... */
2094 gpa_offset = data & ~(PAGE_MASK | 1);
2096 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2097 &vcpu->arch.pv_time, data & ~1ULL,
2098 sizeof(struct pvclock_vcpu_time_info)))
2099 vcpu->arch.pv_time_enabled = false;
2101 vcpu->arch.pv_time_enabled = true;
2105 case MSR_KVM_ASYNC_PF_EN:
2106 if (kvm_pv_enable_async_pf(vcpu, data))
2109 case MSR_KVM_STEAL_TIME:
2111 if (unlikely(!sched_info_on()))
2114 if (data & KVM_STEAL_RESERVED_MASK)
2117 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2118 data & KVM_STEAL_VALID_BITS,
2119 sizeof(struct kvm_steal_time)))
2122 vcpu->arch.st.msr_val = data;
2124 if (!(data & KVM_MSR_ENABLED))
2127 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2130 accumulate_steal_time(vcpu);
2133 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2136 case MSR_KVM_PV_EOI_EN:
2137 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2141 case MSR_IA32_MCG_CTL:
2142 case MSR_IA32_MCG_STATUS:
2143 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2144 return set_msr_mce(vcpu, msr, data);
2146 /* Performance counters are not protected by a CPUID bit,
2147 * so we should check all of them in the generic path for the sake of
2148 * cross vendor migration.
2149 * Writing a zero into the event select MSRs disables them,
2150 * which we perfectly emulate ;-). Any other value should be at least
2151 * reported, some guests depend on them.
2153 case MSR_K7_EVNTSEL0:
2154 case MSR_K7_EVNTSEL1:
2155 case MSR_K7_EVNTSEL2:
2156 case MSR_K7_EVNTSEL3:
2158 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2159 "0x%x data 0x%llx\n", msr, data);
2161 /* at least RHEL 4 unconditionally writes to the perfctr registers,
2162 * so we ignore writes to make it happy.
2164 case MSR_K7_PERFCTR0:
2165 case MSR_K7_PERFCTR1:
2166 case MSR_K7_PERFCTR2:
2167 case MSR_K7_PERFCTR3:
2168 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2169 "0x%x data 0x%llx\n", msr, data);
2171 case MSR_P6_PERFCTR0:
2172 case MSR_P6_PERFCTR1:
2174 case MSR_P6_EVNTSEL0:
2175 case MSR_P6_EVNTSEL1:
2176 if (kvm_pmu_msr(vcpu, msr))
2177 return kvm_pmu_set_msr(vcpu, msr_info);
2179 if (pr || data != 0)
2180 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2181 "0x%x data 0x%llx\n", msr, data);
2183 case MSR_K7_CLK_CTL:
2185 * Ignore all writes to this no longer documented MSR.
2186 * Writes are only relevant for old K7 processors,
2187 * all pre-dating SVM, but a recommended workaround from
2188 * AMD for these chips. It is possible to specify the
2189 * affected processor models on the command line, hence
2190 * the need to ignore the workaround.
2193 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2194 if (kvm_hv_msr_partition_wide(msr)) {
2196 mutex_lock(&vcpu->kvm->lock);
2197 r = set_msr_hyperv_pw(vcpu, msr, data);
2198 mutex_unlock(&vcpu->kvm->lock);
2201 return set_msr_hyperv(vcpu, msr, data);
2203 case MSR_IA32_BBL_CR_CTL3:
2204 /* Drop writes to this legacy MSR -- see rdmsr
2205 * counterpart for further detail.
2207 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2209 case MSR_AMD64_OSVW_ID_LENGTH:
2210 if (!guest_cpuid_has_osvw(vcpu))
2212 vcpu->arch.osvw.length = data;
2214 case MSR_AMD64_OSVW_STATUS:
2215 if (!guest_cpuid_has_osvw(vcpu))
2217 vcpu->arch.osvw.status = data;
2220 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2221 return xen_hvm_config(vcpu, data);
2222 if (kvm_pmu_msr(vcpu, msr))
2223 return kvm_pmu_set_msr(vcpu, msr_info);
2225 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2229 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2236 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2240 * Reads an msr value (of 'msr_index') into 'pdata'.
2241 * Returns 0 on success, non-0 otherwise.
2242 * Assumes vcpu_load() was already called.
2244 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2246 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
2249 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2251 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2253 if (!msr_mtrr_valid(msr))
2256 if (msr == MSR_MTRRdefType)
2257 *pdata = vcpu->arch.mtrr_state.def_type +
2258 (vcpu->arch.mtrr_state.enabled << 10);
2259 else if (msr == MSR_MTRRfix64K_00000)
2261 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2262 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2263 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2264 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2265 else if (msr == MSR_IA32_CR_PAT)
2266 *pdata = vcpu->arch.pat;
2267 else { /* Variable MTRRs */
2268 int idx, is_mtrr_mask;
2271 idx = (msr - 0x200) / 2;
2272 is_mtrr_mask = msr - 0x200 - 2 * idx;
2275 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2278 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2285 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2288 u64 mcg_cap = vcpu->arch.mcg_cap;
2289 unsigned bank_num = mcg_cap & 0xff;
2292 case MSR_IA32_P5_MC_ADDR:
2293 case MSR_IA32_P5_MC_TYPE:
2296 case MSR_IA32_MCG_CAP:
2297 data = vcpu->arch.mcg_cap;
2299 case MSR_IA32_MCG_CTL:
2300 if (!(mcg_cap & MCG_CTL_P))
2302 data = vcpu->arch.mcg_ctl;
2304 case MSR_IA32_MCG_STATUS:
2305 data = vcpu->arch.mcg_status;
2308 if (msr >= MSR_IA32_MC0_CTL &&
2309 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
2310 u32 offset = msr - MSR_IA32_MC0_CTL;
2311 data = vcpu->arch.mce_banks[offset];
2320 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2323 struct kvm *kvm = vcpu->kvm;
2326 case HV_X64_MSR_GUEST_OS_ID:
2327 data = kvm->arch.hv_guest_os_id;
2329 case HV_X64_MSR_HYPERCALL:
2330 data = kvm->arch.hv_hypercall;
2332 case HV_X64_MSR_TIME_REF_COUNT: {
2334 div_u64(get_kernel_ns() + kvm->arch.kvmclock_offset, 100);
2337 case HV_X64_MSR_REFERENCE_TSC:
2338 data = kvm->arch.hv_tsc_page;
2341 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2349 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2354 case HV_X64_MSR_VP_INDEX: {
2357 kvm_for_each_vcpu(r, v, vcpu->kvm) {
2365 case HV_X64_MSR_EOI:
2366 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2367 case HV_X64_MSR_ICR:
2368 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2369 case HV_X64_MSR_TPR:
2370 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
2371 case HV_X64_MSR_APIC_ASSIST_PAGE:
2372 data = vcpu->arch.hv_vapic;
2375 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2382 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2387 case MSR_IA32_PLATFORM_ID:
2388 case MSR_IA32_EBL_CR_POWERON:
2389 case MSR_IA32_DEBUGCTLMSR:
2390 case MSR_IA32_LASTBRANCHFROMIP:
2391 case MSR_IA32_LASTBRANCHTOIP:
2392 case MSR_IA32_LASTINTFROMIP:
2393 case MSR_IA32_LASTINTTOIP:
2396 case MSR_VM_HSAVE_PA:
2397 case MSR_K7_EVNTSEL0:
2398 case MSR_K7_PERFCTR0:
2399 case MSR_K8_INT_PENDING_MSG:
2400 case MSR_AMD64_NB_CFG:
2401 case MSR_FAM10H_MMIO_CONF_BASE:
2402 case MSR_AMD64_BU_CFG2:
2405 case MSR_P6_PERFCTR0:
2406 case MSR_P6_PERFCTR1:
2407 case MSR_P6_EVNTSEL0:
2408 case MSR_P6_EVNTSEL1:
2409 if (kvm_pmu_msr(vcpu, msr))
2410 return kvm_pmu_get_msr(vcpu, msr, pdata);
2413 case MSR_IA32_UCODE_REV:
2414 data = 0x100000000ULL;
2417 data = 0x500 | KVM_NR_VAR_MTRR;
2419 case 0x200 ... 0x2ff:
2420 return get_msr_mtrr(vcpu, msr, pdata);
2421 case 0xcd: /* fsb frequency */
2425 * MSR_EBC_FREQUENCY_ID
2426 * Conservative value valid for even the basic CPU models.
2427 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2428 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2429 * and 266MHz for model 3, or 4. Set Core Clock
2430 * Frequency to System Bus Frequency Ratio to 1 (bits
2431 * 31:24) even though these are only valid for CPU
2432 * models > 2, however guests may end up dividing or
2433 * multiplying by zero otherwise.
2435 case MSR_EBC_FREQUENCY_ID:
2438 case MSR_IA32_APICBASE:
2439 data = kvm_get_apic_base(vcpu);
2441 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2442 return kvm_x2apic_msr_read(vcpu, msr, pdata);
2444 case MSR_IA32_TSCDEADLINE:
2445 data = kvm_get_lapic_tscdeadline_msr(vcpu);
2447 case MSR_IA32_TSC_ADJUST:
2448 data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2450 case MSR_IA32_MISC_ENABLE:
2451 data = vcpu->arch.ia32_misc_enable_msr;
2453 case MSR_IA32_PERF_STATUS:
2454 /* TSC increment by tick */
2456 /* CPU multiplier */
2457 data |= (((uint64_t)4ULL) << 40);
2460 data = vcpu->arch.efer;
2462 case MSR_KVM_WALL_CLOCK:
2463 case MSR_KVM_WALL_CLOCK_NEW:
2464 data = vcpu->kvm->arch.wall_clock;
2466 case MSR_KVM_SYSTEM_TIME:
2467 case MSR_KVM_SYSTEM_TIME_NEW:
2468 data = vcpu->arch.time;
2470 case MSR_KVM_ASYNC_PF_EN:
2471 data = vcpu->arch.apf.msr_val;
2473 case MSR_KVM_STEAL_TIME:
2474 data = vcpu->arch.st.msr_val;
2476 case MSR_KVM_PV_EOI_EN:
2477 data = vcpu->arch.pv_eoi.msr_val;
2479 case MSR_IA32_P5_MC_ADDR:
2480 case MSR_IA32_P5_MC_TYPE:
2481 case MSR_IA32_MCG_CAP:
2482 case MSR_IA32_MCG_CTL:
2483 case MSR_IA32_MCG_STATUS:
2484 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2485 return get_msr_mce(vcpu, msr, pdata);
2486 case MSR_K7_CLK_CTL:
2488 * Provide expected ramp-up count for K7. All other
2489 * are set to zero, indicating minimum divisors for
2492 * This prevents guest kernels on AMD host with CPU
2493 * type 6, model 8 and higher from exploding due to
2494 * the rdmsr failing.
2498 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2499 if (kvm_hv_msr_partition_wide(msr)) {
2501 mutex_lock(&vcpu->kvm->lock);
2502 r = get_msr_hyperv_pw(vcpu, msr, pdata);
2503 mutex_unlock(&vcpu->kvm->lock);
2506 return get_msr_hyperv(vcpu, msr, pdata);
2508 case MSR_IA32_BBL_CR_CTL3:
2509 /* This legacy MSR exists but isn't fully documented in current
2510 * silicon. It is however accessed by winxp in very narrow
2511 * scenarios where it sets bit #19, itself documented as
2512 * a "reserved" bit. Best effort attempt to source coherent
2513 * read data here should the balance of the register be
2514 * interpreted by the guest:
2516 * L2 cache control register 3: 64GB range, 256KB size,
2517 * enabled, latency 0x1, configured
2521 case MSR_AMD64_OSVW_ID_LENGTH:
2522 if (!guest_cpuid_has_osvw(vcpu))
2524 data = vcpu->arch.osvw.length;
2526 case MSR_AMD64_OSVW_STATUS:
2527 if (!guest_cpuid_has_osvw(vcpu))
2529 data = vcpu->arch.osvw.status;
2532 if (kvm_pmu_msr(vcpu, msr))
2533 return kvm_pmu_get_msr(vcpu, msr, pdata);
2535 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2538 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2546 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2549 * Read or write a bunch of msrs. All parameters are kernel addresses.
2551 * @return number of msrs set successfully.
2553 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2554 struct kvm_msr_entry *entries,
2555 int (*do_msr)(struct kvm_vcpu *vcpu,
2556 unsigned index, u64 *data))
2560 idx = srcu_read_lock(&vcpu->kvm->srcu);
2561 for (i = 0; i < msrs->nmsrs; ++i)
2562 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2564 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2570 * Read or write a bunch of msrs. Parameters are user addresses.
2572 * @return number of msrs set successfully.
2574 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2575 int (*do_msr)(struct kvm_vcpu *vcpu,
2576 unsigned index, u64 *data),
2579 struct kvm_msrs msrs;
2580 struct kvm_msr_entry *entries;
2585 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2589 if (msrs.nmsrs >= MAX_IO_MSRS)
2592 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2593 entries = memdup_user(user_msrs->entries, size);
2594 if (IS_ERR(entries)) {
2595 r = PTR_ERR(entries);
2599 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2604 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2615 int kvm_dev_ioctl_check_extension(long ext)
2620 case KVM_CAP_IRQCHIP:
2622 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2623 case KVM_CAP_SET_TSS_ADDR:
2624 case KVM_CAP_EXT_CPUID:
2625 case KVM_CAP_EXT_EMUL_CPUID:
2626 case KVM_CAP_CLOCKSOURCE:
2628 case KVM_CAP_NOP_IO_DELAY:
2629 case KVM_CAP_MP_STATE:
2630 case KVM_CAP_SYNC_MMU:
2631 case KVM_CAP_USER_NMI:
2632 case KVM_CAP_REINJECT_CONTROL:
2633 case KVM_CAP_IRQ_INJECT_STATUS:
2635 case KVM_CAP_IOEVENTFD:
2636 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2638 case KVM_CAP_PIT_STATE2:
2639 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2640 case KVM_CAP_XEN_HVM:
2641 case KVM_CAP_ADJUST_CLOCK:
2642 case KVM_CAP_VCPU_EVENTS:
2643 case KVM_CAP_HYPERV:
2644 case KVM_CAP_HYPERV_VAPIC:
2645 case KVM_CAP_HYPERV_SPIN:
2646 case KVM_CAP_PCI_SEGMENT:
2647 case KVM_CAP_DEBUGREGS:
2648 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2650 case KVM_CAP_ASYNC_PF:
2651 case KVM_CAP_GET_TSC_KHZ:
2652 case KVM_CAP_KVMCLOCK_CTRL:
2653 case KVM_CAP_READONLY_MEM:
2654 case KVM_CAP_HYPERV_TIME:
2655 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2656 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2657 case KVM_CAP_ASSIGN_DEV_IRQ:
2658 case KVM_CAP_PCI_2_3:
2662 case KVM_CAP_COALESCED_MMIO:
2663 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2666 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2668 case KVM_CAP_NR_VCPUS:
2669 r = KVM_SOFT_MAX_VCPUS;
2671 case KVM_CAP_MAX_VCPUS:
2674 case KVM_CAP_NR_MEMSLOTS:
2675 r = KVM_USER_MEM_SLOTS;
2677 case KVM_CAP_PV_MMU: /* obsolete */
2680 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2682 r = iommu_present(&pci_bus_type);
2686 r = KVM_MAX_MCE_BANKS;
2691 case KVM_CAP_TSC_CONTROL:
2692 r = kvm_has_tsc_control;
2694 case KVM_CAP_TSC_DEADLINE_TIMER:
2695 r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2705 long kvm_arch_dev_ioctl(struct file *filp,
2706 unsigned int ioctl, unsigned long arg)
2708 void __user *argp = (void __user *)arg;
2712 case KVM_GET_MSR_INDEX_LIST: {
2713 struct kvm_msr_list __user *user_msr_list = argp;
2714 struct kvm_msr_list msr_list;
2718 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2721 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2722 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2725 if (n < msr_list.nmsrs)
2728 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2729 num_msrs_to_save * sizeof(u32)))
2731 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2733 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2738 case KVM_GET_SUPPORTED_CPUID:
2739 case KVM_GET_EMULATED_CPUID: {
2740 struct kvm_cpuid2 __user *cpuid_arg = argp;
2741 struct kvm_cpuid2 cpuid;
2744 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2747 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2753 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2758 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2761 mce_cap = KVM_MCE_CAP_SUPPORTED;
2763 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2775 static void wbinvd_ipi(void *garbage)
2780 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2782 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2785 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2787 /* Address WBINVD may be executed by guest */
2788 if (need_emulate_wbinvd(vcpu)) {
2789 if (kvm_x86_ops->has_wbinvd_exit())
2790 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2791 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2792 smp_call_function_single(vcpu->cpu,
2793 wbinvd_ipi, NULL, 1);
2796 kvm_x86_ops->vcpu_load(vcpu, cpu);
2798 /* Apply any externally detected TSC adjustments (due to suspend) */
2799 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2800 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2801 vcpu->arch.tsc_offset_adjustment = 0;
2802 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
2805 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2806 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2807 native_read_tsc() - vcpu->arch.last_host_tsc;
2809 mark_tsc_unstable("KVM discovered backwards TSC");
2810 if (check_tsc_unstable()) {
2811 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2812 vcpu->arch.last_guest_tsc);
2813 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2814 vcpu->arch.tsc_catchup = 1;
2817 * On a host with synchronized TSC, there is no need to update
2818 * kvmclock on vcpu->cpu migration
2820 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2821 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2822 if (vcpu->cpu != cpu)
2823 kvm_migrate_timers(vcpu);
2827 accumulate_steal_time(vcpu);
2828 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2831 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2833 kvm_x86_ops->vcpu_put(vcpu);
2834 kvm_put_guest_fpu(vcpu);
2835 vcpu->arch.last_host_tsc = native_read_tsc();
2838 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2839 struct kvm_lapic_state *s)
2841 kvm_x86_ops->sync_pir_to_irr(vcpu);
2842 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2847 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2848 struct kvm_lapic_state *s)
2850 kvm_apic_post_state_restore(vcpu, s);
2851 update_cr8_intercept(vcpu);
2856 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2857 struct kvm_interrupt *irq)
2859 if (irq->irq >= KVM_NR_INTERRUPTS)
2861 if (irqchip_in_kernel(vcpu->kvm))
2864 kvm_queue_interrupt(vcpu, irq->irq, false);
2865 kvm_make_request(KVM_REQ_EVENT, vcpu);
2870 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2872 kvm_inject_nmi(vcpu);
2877 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2878 struct kvm_tpr_access_ctl *tac)
2882 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2886 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2890 unsigned bank_num = mcg_cap & 0xff, bank;
2893 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2895 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2898 vcpu->arch.mcg_cap = mcg_cap;
2899 /* Init IA32_MCG_CTL to all 1s */
2900 if (mcg_cap & MCG_CTL_P)
2901 vcpu->arch.mcg_ctl = ~(u64)0;
2902 /* Init IA32_MCi_CTL to all 1s */
2903 for (bank = 0; bank < bank_num; bank++)
2904 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2909 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2910 struct kvm_x86_mce *mce)
2912 u64 mcg_cap = vcpu->arch.mcg_cap;
2913 unsigned bank_num = mcg_cap & 0xff;
2914 u64 *banks = vcpu->arch.mce_banks;
2916 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2919 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2920 * reporting is disabled
2922 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2923 vcpu->arch.mcg_ctl != ~(u64)0)
2925 banks += 4 * mce->bank;
2927 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2928 * reporting is disabled for the bank
2930 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2932 if (mce->status & MCI_STATUS_UC) {
2933 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2934 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2935 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2938 if (banks[1] & MCI_STATUS_VAL)
2939 mce->status |= MCI_STATUS_OVER;
2940 banks[2] = mce->addr;
2941 banks[3] = mce->misc;
2942 vcpu->arch.mcg_status = mce->mcg_status;
2943 banks[1] = mce->status;
2944 kvm_queue_exception(vcpu, MC_VECTOR);
2945 } else if (!(banks[1] & MCI_STATUS_VAL)
2946 || !(banks[1] & MCI_STATUS_UC)) {
2947 if (banks[1] & MCI_STATUS_VAL)
2948 mce->status |= MCI_STATUS_OVER;
2949 banks[2] = mce->addr;
2950 banks[3] = mce->misc;
2951 banks[1] = mce->status;
2953 banks[1] |= MCI_STATUS_OVER;
2957 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2958 struct kvm_vcpu_events *events)
2961 events->exception.injected =
2962 vcpu->arch.exception.pending &&
2963 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2964 events->exception.nr = vcpu->arch.exception.nr;
2965 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2966 events->exception.pad = 0;
2967 events->exception.error_code = vcpu->arch.exception.error_code;
2969 events->interrupt.injected =
2970 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2971 events->interrupt.nr = vcpu->arch.interrupt.nr;
2972 events->interrupt.soft = 0;
2973 events->interrupt.shadow =
2974 kvm_x86_ops->get_interrupt_shadow(vcpu,
2975 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2977 events->nmi.injected = vcpu->arch.nmi_injected;
2978 events->nmi.pending = vcpu->arch.nmi_pending != 0;
2979 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2980 events->nmi.pad = 0;
2982 events->sipi_vector = 0; /* never valid when reporting to user space */
2984 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2985 | KVM_VCPUEVENT_VALID_SHADOW);
2986 memset(&events->reserved, 0, sizeof(events->reserved));
2989 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2990 struct kvm_vcpu_events *events)
2992 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2993 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2994 | KVM_VCPUEVENT_VALID_SHADOW))
2998 vcpu->arch.exception.pending = events->exception.injected;
2999 vcpu->arch.exception.nr = events->exception.nr;
3000 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3001 vcpu->arch.exception.error_code = events->exception.error_code;
3003 vcpu->arch.interrupt.pending = events->interrupt.injected;
3004 vcpu->arch.interrupt.nr = events->interrupt.nr;
3005 vcpu->arch.interrupt.soft = events->interrupt.soft;
3006 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3007 kvm_x86_ops->set_interrupt_shadow(vcpu,
3008 events->interrupt.shadow);
3010 vcpu->arch.nmi_injected = events->nmi.injected;
3011 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3012 vcpu->arch.nmi_pending = events->nmi.pending;
3013 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3015 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3016 kvm_vcpu_has_lapic(vcpu))
3017 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3019 kvm_make_request(KVM_REQ_EVENT, vcpu);
3024 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3025 struct kvm_debugregs *dbgregs)
3029 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3030 _kvm_get_dr(vcpu, 6, &val);
3032 dbgregs->dr7 = vcpu->arch.dr7;
3034 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3037 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3038 struct kvm_debugregs *dbgregs)
3043 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3044 vcpu->arch.dr6 = dbgregs->dr6;
3045 kvm_update_dr6(vcpu);
3046 vcpu->arch.dr7 = dbgregs->dr7;
3047 kvm_update_dr7(vcpu);
3052 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3053 struct kvm_xsave *guest_xsave)
3055 if (cpu_has_xsave) {
3056 memcpy(guest_xsave->region,
3057 &vcpu->arch.guest_fpu.state->xsave,
3058 vcpu->arch.guest_xstate_size);
3059 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] &=
3060 vcpu->arch.guest_supported_xcr0 | XSTATE_FPSSE;
3062 memcpy(guest_xsave->region,
3063 &vcpu->arch.guest_fpu.state->fxsave,
3064 sizeof(struct i387_fxsave_struct));
3065 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3070 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3071 struct kvm_xsave *guest_xsave)
3074 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3076 if (cpu_has_xsave) {
3078 * Here we allow setting states that are not present in
3079 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3080 * with old userspace.
3082 if (xstate_bv & ~kvm_supported_xcr0())
3084 memcpy(&vcpu->arch.guest_fpu.state->xsave,
3085 guest_xsave->region, vcpu->arch.guest_xstate_size);
3087 if (xstate_bv & ~XSTATE_FPSSE)
3089 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
3090 guest_xsave->region, sizeof(struct i387_fxsave_struct));
3095 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3096 struct kvm_xcrs *guest_xcrs)
3098 if (!cpu_has_xsave) {
3099 guest_xcrs->nr_xcrs = 0;
3103 guest_xcrs->nr_xcrs = 1;
3104 guest_xcrs->flags = 0;
3105 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3106 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3109 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3110 struct kvm_xcrs *guest_xcrs)
3117 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3120 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3121 /* Only support XCR0 currently */
3122 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3123 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3124 guest_xcrs->xcrs[i].value);
3133 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3134 * stopped by the hypervisor. This function will be called from the host only.
3135 * EINVAL is returned when the host attempts to set the flag for a guest that
3136 * does not support pv clocks.
3138 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3140 if (!vcpu->arch.pv_time_enabled)
3142 vcpu->arch.pvclock_set_guest_stopped_request = true;
3143 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3147 long kvm_arch_vcpu_ioctl(struct file *filp,
3148 unsigned int ioctl, unsigned long arg)
3150 struct kvm_vcpu *vcpu = filp->private_data;
3151 void __user *argp = (void __user *)arg;
3154 struct kvm_lapic_state *lapic;
3155 struct kvm_xsave *xsave;
3156 struct kvm_xcrs *xcrs;
3162 case KVM_GET_LAPIC: {
3164 if (!vcpu->arch.apic)
3166 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3171 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3175 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3180 case KVM_SET_LAPIC: {
3182 if (!vcpu->arch.apic)
3184 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3185 if (IS_ERR(u.lapic))
3186 return PTR_ERR(u.lapic);
3188 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3191 case KVM_INTERRUPT: {
3192 struct kvm_interrupt irq;
3195 if (copy_from_user(&irq, argp, sizeof irq))
3197 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3201 r = kvm_vcpu_ioctl_nmi(vcpu);
3204 case KVM_SET_CPUID: {
3205 struct kvm_cpuid __user *cpuid_arg = argp;
3206 struct kvm_cpuid cpuid;
3209 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3211 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3214 case KVM_SET_CPUID2: {
3215 struct kvm_cpuid2 __user *cpuid_arg = argp;
3216 struct kvm_cpuid2 cpuid;
3219 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3221 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3222 cpuid_arg->entries);
3225 case KVM_GET_CPUID2: {
3226 struct kvm_cpuid2 __user *cpuid_arg = argp;
3227 struct kvm_cpuid2 cpuid;
3230 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3232 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3233 cpuid_arg->entries);
3237 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3243 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3246 r = msr_io(vcpu, argp, do_set_msr, 0);
3248 case KVM_TPR_ACCESS_REPORTING: {
3249 struct kvm_tpr_access_ctl tac;
3252 if (copy_from_user(&tac, argp, sizeof tac))
3254 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3258 if (copy_to_user(argp, &tac, sizeof tac))
3263 case KVM_SET_VAPIC_ADDR: {
3264 struct kvm_vapic_addr va;
3267 if (!irqchip_in_kernel(vcpu->kvm))
3270 if (copy_from_user(&va, argp, sizeof va))
3272 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3275 case KVM_X86_SETUP_MCE: {
3279 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3281 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3284 case KVM_X86_SET_MCE: {
3285 struct kvm_x86_mce mce;
3288 if (copy_from_user(&mce, argp, sizeof mce))
3290 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3293 case KVM_GET_VCPU_EVENTS: {
3294 struct kvm_vcpu_events events;
3296 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3299 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3304 case KVM_SET_VCPU_EVENTS: {
3305 struct kvm_vcpu_events events;
3308 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3311 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3314 case KVM_GET_DEBUGREGS: {
3315 struct kvm_debugregs dbgregs;
3317 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3320 if (copy_to_user(argp, &dbgregs,
3321 sizeof(struct kvm_debugregs)))
3326 case KVM_SET_DEBUGREGS: {
3327 struct kvm_debugregs dbgregs;
3330 if (copy_from_user(&dbgregs, argp,
3331 sizeof(struct kvm_debugregs)))
3334 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3337 case KVM_GET_XSAVE: {
3338 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3343 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3346 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3351 case KVM_SET_XSAVE: {
3352 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3353 if (IS_ERR(u.xsave))
3354 return PTR_ERR(u.xsave);
3356 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3359 case KVM_GET_XCRS: {
3360 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3365 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3368 if (copy_to_user(argp, u.xcrs,
3369 sizeof(struct kvm_xcrs)))
3374 case KVM_SET_XCRS: {
3375 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3377 return PTR_ERR(u.xcrs);
3379 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3382 case KVM_SET_TSC_KHZ: {
3386 user_tsc_khz = (u32)arg;
3388 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3391 if (user_tsc_khz == 0)
3392 user_tsc_khz = tsc_khz;
3394 kvm_set_tsc_khz(vcpu, user_tsc_khz);
3399 case KVM_GET_TSC_KHZ: {
3400 r = vcpu->arch.virtual_tsc_khz;
3403 case KVM_KVMCLOCK_CTRL: {
3404 r = kvm_set_guest_paused(vcpu);
3415 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3417 return VM_FAULT_SIGBUS;
3420 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3424 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3426 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3430 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3433 kvm->arch.ept_identity_map_addr = ident_addr;
3437 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3438 u32 kvm_nr_mmu_pages)
3440 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3443 mutex_lock(&kvm->slots_lock);
3445 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3446 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3448 mutex_unlock(&kvm->slots_lock);
3452 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3454 return kvm->arch.n_max_mmu_pages;
3457 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3462 switch (chip->chip_id) {
3463 case KVM_IRQCHIP_PIC_MASTER:
3464 memcpy(&chip->chip.pic,
3465 &pic_irqchip(kvm)->pics[0],
3466 sizeof(struct kvm_pic_state));
3468 case KVM_IRQCHIP_PIC_SLAVE:
3469 memcpy(&chip->chip.pic,
3470 &pic_irqchip(kvm)->pics[1],
3471 sizeof(struct kvm_pic_state));
3473 case KVM_IRQCHIP_IOAPIC:
3474 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3483 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3488 switch (chip->chip_id) {
3489 case KVM_IRQCHIP_PIC_MASTER:
3490 spin_lock(&pic_irqchip(kvm)->lock);
3491 memcpy(&pic_irqchip(kvm)->pics[0],
3493 sizeof(struct kvm_pic_state));
3494 spin_unlock(&pic_irqchip(kvm)->lock);
3496 case KVM_IRQCHIP_PIC_SLAVE:
3497 spin_lock(&pic_irqchip(kvm)->lock);
3498 memcpy(&pic_irqchip(kvm)->pics[1],
3500 sizeof(struct kvm_pic_state));
3501 spin_unlock(&pic_irqchip(kvm)->lock);
3503 case KVM_IRQCHIP_IOAPIC:
3504 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3510 kvm_pic_update_irq(pic_irqchip(kvm));
3514 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3518 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3519 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3520 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3524 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3528 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3529 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3530 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3531 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3535 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3539 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3540 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3541 sizeof(ps->channels));
3542 ps->flags = kvm->arch.vpit->pit_state.flags;
3543 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3544 memset(&ps->reserved, 0, sizeof(ps->reserved));
3548 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3550 int r = 0, start = 0;
3551 u32 prev_legacy, cur_legacy;
3552 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3553 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3554 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3555 if (!prev_legacy && cur_legacy)
3557 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3558 sizeof(kvm->arch.vpit->pit_state.channels));
3559 kvm->arch.vpit->pit_state.flags = ps->flags;
3560 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3561 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3565 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3566 struct kvm_reinject_control *control)
3568 if (!kvm->arch.vpit)
3570 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3571 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3572 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3577 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3578 * @kvm: kvm instance
3579 * @log: slot id and address to which we copy the log
3581 * We need to keep it in mind that VCPU threads can write to the bitmap
3582 * concurrently. So, to avoid losing data, we keep the following order for
3585 * 1. Take a snapshot of the bit and clear it if needed.
3586 * 2. Write protect the corresponding page.
3587 * 3. Flush TLB's if needed.
3588 * 4. Copy the snapshot to the userspace.
3590 * Between 2 and 3, the guest may write to the page using the remaining TLB
3591 * entry. This is not a problem because the page will be reported dirty at
3592 * step 4 using the snapshot taken before and step 3 ensures that successive
3593 * writes will be logged for the next call.
3595 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3598 struct kvm_memory_slot *memslot;
3600 unsigned long *dirty_bitmap;
3601 unsigned long *dirty_bitmap_buffer;
3602 bool is_dirty = false;
3604 mutex_lock(&kvm->slots_lock);
3607 if (log->slot >= KVM_USER_MEM_SLOTS)
3610 memslot = id_to_memslot(kvm->memslots, log->slot);
3612 dirty_bitmap = memslot->dirty_bitmap;
3617 n = kvm_dirty_bitmap_bytes(memslot);
3619 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
3620 memset(dirty_bitmap_buffer, 0, n);
3622 spin_lock(&kvm->mmu_lock);
3624 for (i = 0; i < n / sizeof(long); i++) {
3628 if (!dirty_bitmap[i])
3633 mask = xchg(&dirty_bitmap[i], 0);
3634 dirty_bitmap_buffer[i] = mask;
3636 offset = i * BITS_PER_LONG;
3637 kvm_mmu_write_protect_pt_masked(kvm, memslot, offset, mask);
3640 spin_unlock(&kvm->mmu_lock);
3642 /* See the comments in kvm_mmu_slot_remove_write_access(). */
3643 lockdep_assert_held(&kvm->slots_lock);
3646 * All the TLBs can be flushed out of mmu lock, see the comments in
3647 * kvm_mmu_slot_remove_write_access().
3650 kvm_flush_remote_tlbs(kvm);
3653 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
3658 mutex_unlock(&kvm->slots_lock);
3662 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3665 if (!irqchip_in_kernel(kvm))
3668 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3669 irq_event->irq, irq_event->level,
3674 long kvm_arch_vm_ioctl(struct file *filp,
3675 unsigned int ioctl, unsigned long arg)
3677 struct kvm *kvm = filp->private_data;
3678 void __user *argp = (void __user *)arg;
3681 * This union makes it completely explicit to gcc-3.x
3682 * that these two variables' stack usage should be
3683 * combined, not added together.
3686 struct kvm_pit_state ps;
3687 struct kvm_pit_state2 ps2;
3688 struct kvm_pit_config pit_config;
3692 case KVM_SET_TSS_ADDR:
3693 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3695 case KVM_SET_IDENTITY_MAP_ADDR: {
3699 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3701 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3704 case KVM_SET_NR_MMU_PAGES:
3705 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3707 case KVM_GET_NR_MMU_PAGES:
3708 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3710 case KVM_CREATE_IRQCHIP: {
3711 struct kvm_pic *vpic;
3713 mutex_lock(&kvm->lock);
3716 goto create_irqchip_unlock;
3718 if (atomic_read(&kvm->online_vcpus))
3719 goto create_irqchip_unlock;
3721 vpic = kvm_create_pic(kvm);
3723 r = kvm_ioapic_init(kvm);
3725 mutex_lock(&kvm->slots_lock);
3726 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3728 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3730 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3732 mutex_unlock(&kvm->slots_lock);
3734 goto create_irqchip_unlock;
3737 goto create_irqchip_unlock;
3739 kvm->arch.vpic = vpic;
3741 r = kvm_setup_default_irq_routing(kvm);
3743 mutex_lock(&kvm->slots_lock);
3744 mutex_lock(&kvm->irq_lock);
3745 kvm_ioapic_destroy(kvm);
3746 kvm_destroy_pic(kvm);
3747 mutex_unlock(&kvm->irq_lock);
3748 mutex_unlock(&kvm->slots_lock);
3750 create_irqchip_unlock:
3751 mutex_unlock(&kvm->lock);
3754 case KVM_CREATE_PIT:
3755 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3757 case KVM_CREATE_PIT2:
3759 if (copy_from_user(&u.pit_config, argp,
3760 sizeof(struct kvm_pit_config)))
3763 mutex_lock(&kvm->slots_lock);
3766 goto create_pit_unlock;
3768 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3772 mutex_unlock(&kvm->slots_lock);
3774 case KVM_GET_IRQCHIP: {
3775 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3776 struct kvm_irqchip *chip;
3778 chip = memdup_user(argp, sizeof(*chip));
3785 if (!irqchip_in_kernel(kvm))
3786 goto get_irqchip_out;
3787 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3789 goto get_irqchip_out;
3791 if (copy_to_user(argp, chip, sizeof *chip))
3792 goto get_irqchip_out;
3798 case KVM_SET_IRQCHIP: {
3799 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3800 struct kvm_irqchip *chip;
3802 chip = memdup_user(argp, sizeof(*chip));
3809 if (!irqchip_in_kernel(kvm))
3810 goto set_irqchip_out;
3811 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3813 goto set_irqchip_out;
3821 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3824 if (!kvm->arch.vpit)
3826 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3830 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3837 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3840 if (!kvm->arch.vpit)
3842 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3845 case KVM_GET_PIT2: {
3847 if (!kvm->arch.vpit)
3849 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3853 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3858 case KVM_SET_PIT2: {
3860 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3863 if (!kvm->arch.vpit)
3865 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3868 case KVM_REINJECT_CONTROL: {
3869 struct kvm_reinject_control control;
3871 if (copy_from_user(&control, argp, sizeof(control)))
3873 r = kvm_vm_ioctl_reinject(kvm, &control);
3876 case KVM_XEN_HVM_CONFIG: {
3878 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3879 sizeof(struct kvm_xen_hvm_config)))
3882 if (kvm->arch.xen_hvm_config.flags)
3887 case KVM_SET_CLOCK: {
3888 struct kvm_clock_data user_ns;
3893 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3901 local_irq_disable();
3902 now_ns = get_kernel_ns();
3903 delta = user_ns.clock - now_ns;
3905 kvm->arch.kvmclock_offset = delta;
3906 kvm_gen_update_masterclock(kvm);
3909 case KVM_GET_CLOCK: {
3910 struct kvm_clock_data user_ns;
3913 local_irq_disable();
3914 now_ns = get_kernel_ns();
3915 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3918 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3921 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3934 static void kvm_init_msr_list(void)
3939 /* skip the first msrs in the list. KVM-specific */
3940 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3941 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3945 * Even MSRs that are valid in the host may not be exposed
3946 * to the guests in some cases. We could work around this
3947 * in VMX with the generic MSR save/load machinery, but it
3948 * is not really worthwhile since it will really only
3949 * happen with nested virtualization.
3951 switch (msrs_to_save[i]) {
3952 case MSR_IA32_BNDCFGS:
3953 if (!kvm_x86_ops->mpx_supported())
3961 msrs_to_save[j] = msrs_to_save[i];
3964 num_msrs_to_save = j;
3967 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3975 if (!(vcpu->arch.apic &&
3976 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3977 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3988 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3995 if (!(vcpu->arch.apic &&
3996 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3997 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3999 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4009 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4010 struct kvm_segment *var, int seg)
4012 kvm_x86_ops->set_segment(vcpu, var, seg);
4015 void kvm_get_segment(struct kvm_vcpu *vcpu,
4016 struct kvm_segment *var, int seg)
4018 kvm_x86_ops->get_segment(vcpu, var, seg);
4021 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
4024 struct x86_exception exception;
4026 BUG_ON(!mmu_is_nested(vcpu));
4028 /* NPT walks are always user-walks */
4029 access |= PFERR_USER_MASK;
4030 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
4035 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4036 struct x86_exception *exception)
4038 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4039 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4042 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4043 struct x86_exception *exception)
4045 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4046 access |= PFERR_FETCH_MASK;
4047 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4050 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4051 struct x86_exception *exception)
4053 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4054 access |= PFERR_WRITE_MASK;
4055 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4058 /* uses this to access any guest's mapped memory without checking CPL */
4059 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4060 struct x86_exception *exception)
4062 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4065 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4066 struct kvm_vcpu *vcpu, u32 access,
4067 struct x86_exception *exception)
4070 int r = X86EMUL_CONTINUE;
4073 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4075 unsigned offset = addr & (PAGE_SIZE-1);
4076 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4079 if (gpa == UNMAPPED_GVA)
4080 return X86EMUL_PROPAGATE_FAULT;
4081 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
4083 r = X86EMUL_IO_NEEDED;
4095 /* used for instruction fetching */
4096 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4097 gva_t addr, void *val, unsigned int bytes,
4098 struct x86_exception *exception)
4100 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4101 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4103 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
4104 access | PFERR_FETCH_MASK,
4108 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4109 gva_t addr, void *val, unsigned int bytes,
4110 struct x86_exception *exception)
4112 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4113 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4115 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4118 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4120 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4121 gva_t addr, void *val, unsigned int bytes,
4122 struct x86_exception *exception)
4124 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4125 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4128 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4129 gva_t addr, void *val,
4131 struct x86_exception *exception)
4133 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4135 int r = X86EMUL_CONTINUE;
4138 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4141 unsigned offset = addr & (PAGE_SIZE-1);
4142 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4145 if (gpa == UNMAPPED_GVA)
4146 return X86EMUL_PROPAGATE_FAULT;
4147 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
4149 r = X86EMUL_IO_NEEDED;
4160 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4162 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4163 gpa_t *gpa, struct x86_exception *exception,
4166 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4167 | (write ? PFERR_WRITE_MASK : 0);
4169 if (vcpu_match_mmio_gva(vcpu, gva)
4170 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4171 vcpu->arch.access, access)) {
4172 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4173 (gva & (PAGE_SIZE - 1));
4174 trace_vcpu_match_mmio(gva, *gpa, write, false);
4178 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4180 if (*gpa == UNMAPPED_GVA)
4183 /* For APIC access vmexit */
4184 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4187 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4188 trace_vcpu_match_mmio(gva, *gpa, write, true);
4195 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4196 const void *val, int bytes)
4200 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4203 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4207 struct read_write_emulator_ops {
4208 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4210 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4211 void *val, int bytes);
4212 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4213 int bytes, void *val);
4214 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4215 void *val, int bytes);
4219 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4221 if (vcpu->mmio_read_completed) {
4222 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4223 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4224 vcpu->mmio_read_completed = 0;
4231 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4232 void *val, int bytes)
4234 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4237 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4238 void *val, int bytes)
4240 return emulator_write_phys(vcpu, gpa, val, bytes);
4243 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4245 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4246 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4249 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4250 void *val, int bytes)
4252 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4253 return X86EMUL_IO_NEEDED;
4256 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4257 void *val, int bytes)
4259 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4261 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4262 return X86EMUL_CONTINUE;
4265 static const struct read_write_emulator_ops read_emultor = {
4266 .read_write_prepare = read_prepare,
4267 .read_write_emulate = read_emulate,
4268 .read_write_mmio = vcpu_mmio_read,
4269 .read_write_exit_mmio = read_exit_mmio,
4272 static const struct read_write_emulator_ops write_emultor = {
4273 .read_write_emulate = write_emulate,
4274 .read_write_mmio = write_mmio,
4275 .read_write_exit_mmio = write_exit_mmio,
4279 static int emulator_read_write_onepage(unsigned long addr, void *val,
4281 struct x86_exception *exception,
4282 struct kvm_vcpu *vcpu,
4283 const struct read_write_emulator_ops *ops)
4287 bool write = ops->write;
4288 struct kvm_mmio_fragment *frag;
4290 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4293 return X86EMUL_PROPAGATE_FAULT;
4295 /* For APIC access vmexit */
4299 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4300 return X86EMUL_CONTINUE;
4304 * Is this MMIO handled locally?
4306 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4307 if (handled == bytes)
4308 return X86EMUL_CONTINUE;
4314 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4315 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4319 return X86EMUL_CONTINUE;
4322 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4323 void *val, unsigned int bytes,
4324 struct x86_exception *exception,
4325 const struct read_write_emulator_ops *ops)
4327 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4331 if (ops->read_write_prepare &&
4332 ops->read_write_prepare(vcpu, val, bytes))
4333 return X86EMUL_CONTINUE;
4335 vcpu->mmio_nr_fragments = 0;
4337 /* Crossing a page boundary? */
4338 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4341 now = -addr & ~PAGE_MASK;
4342 rc = emulator_read_write_onepage(addr, val, now, exception,
4345 if (rc != X86EMUL_CONTINUE)
4352 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4354 if (rc != X86EMUL_CONTINUE)
4357 if (!vcpu->mmio_nr_fragments)
4360 gpa = vcpu->mmio_fragments[0].gpa;
4362 vcpu->mmio_needed = 1;
4363 vcpu->mmio_cur_fragment = 0;
4365 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4366 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4367 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4368 vcpu->run->mmio.phys_addr = gpa;
4370 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4373 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4377 struct x86_exception *exception)
4379 return emulator_read_write(ctxt, addr, val, bytes,
4380 exception, &read_emultor);
4383 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4387 struct x86_exception *exception)
4389 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4390 exception, &write_emultor);
4393 #define CMPXCHG_TYPE(t, ptr, old, new) \
4394 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4396 #ifdef CONFIG_X86_64
4397 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4399 # define CMPXCHG64(ptr, old, new) \
4400 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4403 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4408 struct x86_exception *exception)
4410 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4416 /* guests cmpxchg8b have to be emulated atomically */
4417 if (bytes > 8 || (bytes & (bytes - 1)))
4420 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4422 if (gpa == UNMAPPED_GVA ||
4423 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4426 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4429 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4430 if (is_error_page(page))
4433 kaddr = kmap_atomic(page);
4434 kaddr += offset_in_page(gpa);
4437 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4440 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4443 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4446 exchanged = CMPXCHG64(kaddr, old, new);
4451 kunmap_atomic(kaddr);
4452 kvm_release_page_dirty(page);
4455 return X86EMUL_CMPXCHG_FAILED;
4457 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
4458 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4460 return X86EMUL_CONTINUE;
4463 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4465 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4468 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4470 /* TODO: String I/O for in kernel device */
4473 if (vcpu->arch.pio.in)
4474 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4475 vcpu->arch.pio.size, pd);
4477 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4478 vcpu->arch.pio.port, vcpu->arch.pio.size,
4483 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4484 unsigned short port, void *val,
4485 unsigned int count, bool in)
4487 vcpu->arch.pio.port = port;
4488 vcpu->arch.pio.in = in;
4489 vcpu->arch.pio.count = count;
4490 vcpu->arch.pio.size = size;
4492 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4493 vcpu->arch.pio.count = 0;
4497 vcpu->run->exit_reason = KVM_EXIT_IO;
4498 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4499 vcpu->run->io.size = size;
4500 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4501 vcpu->run->io.count = count;
4502 vcpu->run->io.port = port;
4507 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4508 int size, unsigned short port, void *val,
4511 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4514 if (vcpu->arch.pio.count)
4517 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4520 memcpy(val, vcpu->arch.pio_data, size * count);
4521 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4522 vcpu->arch.pio.count = 0;
4529 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4530 int size, unsigned short port,
4531 const void *val, unsigned int count)
4533 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4535 memcpy(vcpu->arch.pio_data, val, size * count);
4536 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4537 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4540 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4542 return kvm_x86_ops->get_segment_base(vcpu, seg);
4545 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4547 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4550 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4552 if (!need_emulate_wbinvd(vcpu))
4553 return X86EMUL_CONTINUE;
4555 if (kvm_x86_ops->has_wbinvd_exit()) {
4556 int cpu = get_cpu();
4558 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4559 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4560 wbinvd_ipi, NULL, 1);
4562 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4565 return X86EMUL_CONTINUE;
4567 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4569 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4571 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4574 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4576 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4579 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4582 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4585 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4587 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4590 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4592 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4593 unsigned long value;
4597 value = kvm_read_cr0(vcpu);
4600 value = vcpu->arch.cr2;
4603 value = kvm_read_cr3(vcpu);
4606 value = kvm_read_cr4(vcpu);
4609 value = kvm_get_cr8(vcpu);
4612 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4619 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4621 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4626 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4629 vcpu->arch.cr2 = val;
4632 res = kvm_set_cr3(vcpu, val);
4635 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4638 res = kvm_set_cr8(vcpu, val);
4641 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4648 static void emulator_set_rflags(struct x86_emulate_ctxt *ctxt, ulong val)
4650 kvm_set_rflags(emul_to_vcpu(ctxt), val);
4653 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4655 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4658 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4660 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4663 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4665 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4668 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4670 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4673 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4675 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4678 static unsigned long emulator_get_cached_segment_base(
4679 struct x86_emulate_ctxt *ctxt, int seg)
4681 return get_segment_base(emul_to_vcpu(ctxt), seg);
4684 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4685 struct desc_struct *desc, u32 *base3,
4688 struct kvm_segment var;
4690 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4691 *selector = var.selector;
4694 memset(desc, 0, sizeof(*desc));
4700 set_desc_limit(desc, var.limit);
4701 set_desc_base(desc, (unsigned long)var.base);
4702 #ifdef CONFIG_X86_64
4704 *base3 = var.base >> 32;
4706 desc->type = var.type;
4708 desc->dpl = var.dpl;
4709 desc->p = var.present;
4710 desc->avl = var.avl;
4718 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4719 struct desc_struct *desc, u32 base3,
4722 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4723 struct kvm_segment var;
4725 var.selector = selector;
4726 var.base = get_desc_base(desc);
4727 #ifdef CONFIG_X86_64
4728 var.base |= ((u64)base3) << 32;
4730 var.limit = get_desc_limit(desc);
4732 var.limit = (var.limit << 12) | 0xfff;
4733 var.type = desc->type;
4734 var.present = desc->p;
4735 var.dpl = desc->dpl;
4740 var.avl = desc->avl;
4741 var.present = desc->p;
4742 var.unusable = !var.present;
4745 kvm_set_segment(vcpu, &var, seg);
4749 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4750 u32 msr_index, u64 *pdata)
4752 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4755 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4756 u32 msr_index, u64 data)
4758 struct msr_data msr;
4761 msr.index = msr_index;
4762 msr.host_initiated = false;
4763 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4766 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4767 u32 pmc, u64 *pdata)
4769 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4772 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4774 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4777 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4780 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4782 * CR0.TS may reference the host fpu state, not the guest fpu state,
4783 * so it may be clear at this point.
4788 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4793 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4794 struct x86_instruction_info *info,
4795 enum x86_intercept_stage stage)
4797 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4800 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4801 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4803 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4806 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4808 return kvm_register_read(emul_to_vcpu(ctxt), reg);
4811 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4813 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4816 static const struct x86_emulate_ops emulate_ops = {
4817 .read_gpr = emulator_read_gpr,
4818 .write_gpr = emulator_write_gpr,
4819 .read_std = kvm_read_guest_virt_system,
4820 .write_std = kvm_write_guest_virt_system,
4821 .fetch = kvm_fetch_guest_virt,
4822 .read_emulated = emulator_read_emulated,
4823 .write_emulated = emulator_write_emulated,
4824 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4825 .invlpg = emulator_invlpg,
4826 .pio_in_emulated = emulator_pio_in_emulated,
4827 .pio_out_emulated = emulator_pio_out_emulated,
4828 .get_segment = emulator_get_segment,
4829 .set_segment = emulator_set_segment,
4830 .get_cached_segment_base = emulator_get_cached_segment_base,
4831 .get_gdt = emulator_get_gdt,
4832 .get_idt = emulator_get_idt,
4833 .set_gdt = emulator_set_gdt,
4834 .set_idt = emulator_set_idt,
4835 .get_cr = emulator_get_cr,
4836 .set_cr = emulator_set_cr,
4837 .set_rflags = emulator_set_rflags,
4838 .cpl = emulator_get_cpl,
4839 .get_dr = emulator_get_dr,
4840 .set_dr = emulator_set_dr,
4841 .set_msr = emulator_set_msr,
4842 .get_msr = emulator_get_msr,
4843 .read_pmc = emulator_read_pmc,
4844 .halt = emulator_halt,
4845 .wbinvd = emulator_wbinvd,
4846 .fix_hypercall = emulator_fix_hypercall,
4847 .get_fpu = emulator_get_fpu,
4848 .put_fpu = emulator_put_fpu,
4849 .intercept = emulator_intercept,
4850 .get_cpuid = emulator_get_cpuid,
4853 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4855 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4857 * an sti; sti; sequence only disable interrupts for the first
4858 * instruction. So, if the last instruction, be it emulated or
4859 * not, left the system with the INT_STI flag enabled, it
4860 * means that the last instruction is an sti. We should not
4861 * leave the flag on in this case. The same goes for mov ss
4863 if (!(int_shadow & mask))
4864 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4867 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4869 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4870 if (ctxt->exception.vector == PF_VECTOR)
4871 kvm_propagate_fault(vcpu, &ctxt->exception);
4872 else if (ctxt->exception.error_code_valid)
4873 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4874 ctxt->exception.error_code);
4876 kvm_queue_exception(vcpu, ctxt->exception.vector);
4879 static void init_decode_cache(struct x86_emulate_ctxt *ctxt)
4881 memset(&ctxt->opcode_len, 0,
4882 (void *)&ctxt->_regs - (void *)&ctxt->opcode_len);
4884 ctxt->fetch.start = 0;
4885 ctxt->fetch.end = 0;
4886 ctxt->io_read.pos = 0;
4887 ctxt->io_read.end = 0;
4888 ctxt->mem_read.pos = 0;
4889 ctxt->mem_read.end = 0;
4892 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4894 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4897 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4899 ctxt->eflags = kvm_get_rflags(vcpu);
4900 ctxt->eip = kvm_rip_read(vcpu);
4901 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4902 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4903 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
4904 cs_db ? X86EMUL_MODE_PROT32 :
4905 X86EMUL_MODE_PROT16;
4906 ctxt->guest_mode = is_guest_mode(vcpu);
4908 init_decode_cache(ctxt);
4909 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4912 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4914 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4917 init_emulate_ctxt(vcpu);
4921 ctxt->_eip = ctxt->eip + inc_eip;
4922 ret = emulate_int_real(ctxt, irq);
4924 if (ret != X86EMUL_CONTINUE)
4925 return EMULATE_FAIL;
4927 ctxt->eip = ctxt->_eip;
4928 kvm_rip_write(vcpu, ctxt->eip);
4929 kvm_set_rflags(vcpu, ctxt->eflags);
4931 if (irq == NMI_VECTOR)
4932 vcpu->arch.nmi_pending = 0;
4934 vcpu->arch.interrupt.pending = false;
4936 return EMULATE_DONE;
4938 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4940 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4942 int r = EMULATE_DONE;
4944 ++vcpu->stat.insn_emulation_fail;
4945 trace_kvm_emulate_insn_failed(vcpu);
4946 if (!is_guest_mode(vcpu)) {
4947 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4948 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4949 vcpu->run->internal.ndata = 0;
4952 kvm_queue_exception(vcpu, UD_VECTOR);
4957 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
4958 bool write_fault_to_shadow_pgtable,
4964 if (emulation_type & EMULTYPE_NO_REEXECUTE)
4967 if (!vcpu->arch.mmu.direct_map) {
4969 * Write permission should be allowed since only
4970 * write access need to be emulated.
4972 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4975 * If the mapping is invalid in guest, let cpu retry
4976 * it to generate fault.
4978 if (gpa == UNMAPPED_GVA)
4983 * Do not retry the unhandleable instruction if it faults on the
4984 * readonly host memory, otherwise it will goto a infinite loop:
4985 * retry instruction -> write #PF -> emulation fail -> retry
4986 * instruction -> ...
4988 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
4991 * If the instruction failed on the error pfn, it can not be fixed,
4992 * report the error to userspace.
4994 if (is_error_noslot_pfn(pfn))
4997 kvm_release_pfn_clean(pfn);
4999 /* The instructions are well-emulated on direct mmu. */
5000 if (vcpu->arch.mmu.direct_map) {
5001 unsigned int indirect_shadow_pages;
5003 spin_lock(&vcpu->kvm->mmu_lock);
5004 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5005 spin_unlock(&vcpu->kvm->mmu_lock);
5007 if (indirect_shadow_pages)
5008 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5014 * if emulation was due to access to shadowed page table
5015 * and it failed try to unshadow page and re-enter the
5016 * guest to let CPU execute the instruction.
5018 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5021 * If the access faults on its page table, it can not
5022 * be fixed by unprotecting shadow page and it should
5023 * be reported to userspace.
5025 return !write_fault_to_shadow_pgtable;
5028 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5029 unsigned long cr2, int emulation_type)
5031 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5032 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5034 last_retry_eip = vcpu->arch.last_retry_eip;
5035 last_retry_addr = vcpu->arch.last_retry_addr;
5038 * If the emulation is caused by #PF and it is non-page_table
5039 * writing instruction, it means the VM-EXIT is caused by shadow
5040 * page protected, we can zap the shadow page and retry this
5041 * instruction directly.
5043 * Note: if the guest uses a non-page-table modifying instruction
5044 * on the PDE that points to the instruction, then we will unmap
5045 * the instruction and go to an infinite loop. So, we cache the
5046 * last retried eip and the last fault address, if we meet the eip
5047 * and the address again, we can break out of the potential infinite
5050 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5052 if (!(emulation_type & EMULTYPE_RETRY))
5055 if (x86_page_table_writing_insn(ctxt))
5058 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5061 vcpu->arch.last_retry_eip = ctxt->eip;
5062 vcpu->arch.last_retry_addr = cr2;
5064 if (!vcpu->arch.mmu.direct_map)
5065 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5067 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5072 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5073 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5075 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5084 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5085 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5090 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, int *r)
5092 struct kvm_run *kvm_run = vcpu->run;
5095 * Use the "raw" value to see if TF was passed to the processor.
5096 * Note that the new value of the flags has not been saved yet.
5098 * This is correct even for TF set by the guest, because "the
5099 * processor will not generate this exception after the instruction
5100 * that sets the TF flag".
5102 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5104 if (unlikely(rflags & X86_EFLAGS_TF)) {
5105 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5106 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1;
5107 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5108 kvm_run->debug.arch.exception = DB_VECTOR;
5109 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5110 *r = EMULATE_USER_EXIT;
5112 vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5114 * "Certain debug exceptions may clear bit 0-3. The
5115 * remaining contents of the DR6 register are never
5116 * cleared by the processor".
5118 vcpu->arch.dr6 &= ~15;
5119 vcpu->arch.dr6 |= DR6_BS;
5120 kvm_queue_exception(vcpu, DB_VECTOR);
5125 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5127 struct kvm_run *kvm_run = vcpu->run;
5128 unsigned long eip = vcpu->arch.emulate_ctxt.eip;
5131 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5132 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5133 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5134 vcpu->arch.guest_debug_dr7,
5138 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
5139 kvm_run->debug.arch.pc = kvm_rip_read(vcpu) +
5140 get_segment_base(vcpu, VCPU_SREG_CS);
5142 kvm_run->debug.arch.exception = DB_VECTOR;
5143 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5144 *r = EMULATE_USER_EXIT;
5149 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK)) {
5150 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5155 vcpu->arch.dr6 &= ~15;
5156 vcpu->arch.dr6 |= dr6;
5157 kvm_queue_exception(vcpu, DB_VECTOR);
5166 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5173 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5174 bool writeback = true;
5175 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5178 * Clear write_fault_to_shadow_pgtable here to ensure it is
5181 vcpu->arch.write_fault_to_shadow_pgtable = false;
5182 kvm_clear_exception_queue(vcpu);
5184 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5185 init_emulate_ctxt(vcpu);
5188 * We will reenter on the same instruction since
5189 * we do not set complete_userspace_io. This does not
5190 * handle watchpoints yet, those would be handled in
5193 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5196 ctxt->interruptibility = 0;
5197 ctxt->have_exception = false;
5198 ctxt->perm_ok = false;
5200 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5202 r = x86_decode_insn(ctxt, insn, insn_len);
5204 trace_kvm_emulate_insn_start(vcpu);
5205 ++vcpu->stat.insn_emulation;
5206 if (r != EMULATION_OK) {
5207 if (emulation_type & EMULTYPE_TRAP_UD)
5208 return EMULATE_FAIL;
5209 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5211 return EMULATE_DONE;
5212 if (emulation_type & EMULTYPE_SKIP)
5213 return EMULATE_FAIL;
5214 return handle_emulation_failure(vcpu);
5218 if (emulation_type & EMULTYPE_SKIP) {
5219 kvm_rip_write(vcpu, ctxt->_eip);
5220 return EMULATE_DONE;
5223 if (retry_instruction(ctxt, cr2, emulation_type))
5224 return EMULATE_DONE;
5226 /* this is needed for vmware backdoor interface to work since it
5227 changes registers values during IO operation */
5228 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5229 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5230 emulator_invalidate_register_cache(ctxt);
5234 r = x86_emulate_insn(ctxt);
5236 if (r == EMULATION_INTERCEPTED)
5237 return EMULATE_DONE;
5239 if (r == EMULATION_FAILED) {
5240 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5242 return EMULATE_DONE;
5244 return handle_emulation_failure(vcpu);
5247 if (ctxt->have_exception) {
5248 inject_emulated_exception(vcpu);
5250 } else if (vcpu->arch.pio.count) {
5251 if (!vcpu->arch.pio.in) {
5252 /* FIXME: return into emulator if single-stepping. */
5253 vcpu->arch.pio.count = 0;
5256 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5258 r = EMULATE_USER_EXIT;
5259 } else if (vcpu->mmio_needed) {
5260 if (!vcpu->mmio_is_write)
5262 r = EMULATE_USER_EXIT;
5263 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5264 } else if (r == EMULATION_RESTART)
5270 toggle_interruptibility(vcpu, ctxt->interruptibility);
5271 kvm_make_request(KVM_REQ_EVENT, vcpu);
5272 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5273 kvm_rip_write(vcpu, ctxt->eip);
5274 if (r == EMULATE_DONE)
5275 kvm_vcpu_check_singlestep(vcpu, &r);
5276 kvm_set_rflags(vcpu, ctxt->eflags);
5278 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5282 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5284 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5286 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5287 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5288 size, port, &val, 1);
5289 /* do not return to emulator after return from userspace */
5290 vcpu->arch.pio.count = 0;
5293 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5295 static void tsc_bad(void *info)
5297 __this_cpu_write(cpu_tsc_khz, 0);
5300 static void tsc_khz_changed(void *data)
5302 struct cpufreq_freqs *freq = data;
5303 unsigned long khz = 0;
5307 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5308 khz = cpufreq_quick_get(raw_smp_processor_id());
5311 __this_cpu_write(cpu_tsc_khz, khz);
5314 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5317 struct cpufreq_freqs *freq = data;
5319 struct kvm_vcpu *vcpu;
5320 int i, send_ipi = 0;
5323 * We allow guests to temporarily run on slowing clocks,
5324 * provided we notify them after, or to run on accelerating
5325 * clocks, provided we notify them before. Thus time never
5328 * However, we have a problem. We can't atomically update
5329 * the frequency of a given CPU from this function; it is
5330 * merely a notifier, which can be called from any CPU.
5331 * Changing the TSC frequency at arbitrary points in time
5332 * requires a recomputation of local variables related to
5333 * the TSC for each VCPU. We must flag these local variables
5334 * to be updated and be sure the update takes place with the
5335 * new frequency before any guests proceed.
5337 * Unfortunately, the combination of hotplug CPU and frequency
5338 * change creates an intractable locking scenario; the order
5339 * of when these callouts happen is undefined with respect to
5340 * CPU hotplug, and they can race with each other. As such,
5341 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5342 * undefined; you can actually have a CPU frequency change take
5343 * place in between the computation of X and the setting of the
5344 * variable. To protect against this problem, all updates of
5345 * the per_cpu tsc_khz variable are done in an interrupt
5346 * protected IPI, and all callers wishing to update the value
5347 * must wait for a synchronous IPI to complete (which is trivial
5348 * if the caller is on the CPU already). This establishes the
5349 * necessary total order on variable updates.
5351 * Note that because a guest time update may take place
5352 * anytime after the setting of the VCPU's request bit, the
5353 * correct TSC value must be set before the request. However,
5354 * to ensure the update actually makes it to any guest which
5355 * starts running in hardware virtualization between the set
5356 * and the acquisition of the spinlock, we must also ping the
5357 * CPU after setting the request bit.
5361 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5363 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5366 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5368 spin_lock(&kvm_lock);
5369 list_for_each_entry(kvm, &vm_list, vm_list) {
5370 kvm_for_each_vcpu(i, vcpu, kvm) {
5371 if (vcpu->cpu != freq->cpu)
5373 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5374 if (vcpu->cpu != smp_processor_id())
5378 spin_unlock(&kvm_lock);
5380 if (freq->old < freq->new && send_ipi) {
5382 * We upscale the frequency. Must make the guest
5383 * doesn't see old kvmclock values while running with
5384 * the new frequency, otherwise we risk the guest sees
5385 * time go backwards.
5387 * In case we update the frequency for another cpu
5388 * (which might be in guest context) send an interrupt
5389 * to kick the cpu out of guest context. Next time
5390 * guest context is entered kvmclock will be updated,
5391 * so the guest will not see stale values.
5393 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5398 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5399 .notifier_call = kvmclock_cpufreq_notifier
5402 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5403 unsigned long action, void *hcpu)
5405 unsigned int cpu = (unsigned long)hcpu;
5409 case CPU_DOWN_FAILED:
5410 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5412 case CPU_DOWN_PREPARE:
5413 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5419 static struct notifier_block kvmclock_cpu_notifier_block = {
5420 .notifier_call = kvmclock_cpu_notifier,
5421 .priority = -INT_MAX
5424 static void kvm_timer_init(void)
5428 max_tsc_khz = tsc_khz;
5430 cpu_notifier_register_begin();
5431 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5432 #ifdef CONFIG_CPU_FREQ
5433 struct cpufreq_policy policy;
5434 memset(&policy, 0, sizeof(policy));
5436 cpufreq_get_policy(&policy, cpu);
5437 if (policy.cpuinfo.max_freq)
5438 max_tsc_khz = policy.cpuinfo.max_freq;
5441 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5442 CPUFREQ_TRANSITION_NOTIFIER);
5444 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5445 for_each_online_cpu(cpu)
5446 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5448 __register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5449 cpu_notifier_register_done();
5453 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5455 int kvm_is_in_guest(void)
5457 return __this_cpu_read(current_vcpu) != NULL;
5460 static int kvm_is_user_mode(void)
5464 if (__this_cpu_read(current_vcpu))
5465 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5467 return user_mode != 0;
5470 static unsigned long kvm_get_guest_ip(void)
5472 unsigned long ip = 0;
5474 if (__this_cpu_read(current_vcpu))
5475 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5480 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5481 .is_in_guest = kvm_is_in_guest,
5482 .is_user_mode = kvm_is_user_mode,
5483 .get_guest_ip = kvm_get_guest_ip,
5486 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5488 __this_cpu_write(current_vcpu, vcpu);
5490 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5492 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5494 __this_cpu_write(current_vcpu, NULL);
5496 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5498 static void kvm_set_mmio_spte_mask(void)
5501 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5504 * Set the reserved bits and the present bit of an paging-structure
5505 * entry to generate page fault with PFER.RSV = 1.
5507 /* Mask the reserved physical address bits. */
5508 mask = ((1ull << (51 - maxphyaddr + 1)) - 1) << maxphyaddr;
5510 /* Bit 62 is always reserved for 32bit host. */
5511 mask |= 0x3ull << 62;
5513 /* Set the present bit. */
5516 #ifdef CONFIG_X86_64
5518 * If reserved bit is not supported, clear the present bit to disable
5521 if (maxphyaddr == 52)
5525 kvm_mmu_set_mmio_spte_mask(mask);
5528 #ifdef CONFIG_X86_64
5529 static void pvclock_gtod_update_fn(struct work_struct *work)
5533 struct kvm_vcpu *vcpu;
5536 spin_lock(&kvm_lock);
5537 list_for_each_entry(kvm, &vm_list, vm_list)
5538 kvm_for_each_vcpu(i, vcpu, kvm)
5539 set_bit(KVM_REQ_MASTERCLOCK_UPDATE, &vcpu->requests);
5540 atomic_set(&kvm_guest_has_master_clock, 0);
5541 spin_unlock(&kvm_lock);
5544 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5547 * Notification about pvclock gtod data update.
5549 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5552 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5553 struct timekeeper *tk = priv;
5555 update_pvclock_gtod(tk);
5557 /* disable master clock if host does not trust, or does not
5558 * use, TSC clocksource
5560 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5561 atomic_read(&kvm_guest_has_master_clock) != 0)
5562 queue_work(system_long_wq, &pvclock_gtod_work);
5567 static struct notifier_block pvclock_gtod_notifier = {
5568 .notifier_call = pvclock_gtod_notify,
5572 int kvm_arch_init(void *opaque)
5575 struct kvm_x86_ops *ops = opaque;
5578 printk(KERN_ERR "kvm: already loaded the other module\n");
5583 if (!ops->cpu_has_kvm_support()) {
5584 printk(KERN_ERR "kvm: no hardware support\n");
5588 if (ops->disabled_by_bios()) {
5589 printk(KERN_ERR "kvm: disabled by bios\n");
5595 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5597 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5601 r = kvm_mmu_module_init();
5603 goto out_free_percpu;
5605 kvm_set_mmio_spte_mask();
5608 kvm_init_msr_list();
5610 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5611 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5615 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5618 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5621 #ifdef CONFIG_X86_64
5622 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5628 free_percpu(shared_msrs);
5633 void kvm_arch_exit(void)
5635 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5637 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5638 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5639 CPUFREQ_TRANSITION_NOTIFIER);
5640 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5641 #ifdef CONFIG_X86_64
5642 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5645 kvm_mmu_module_exit();
5646 free_percpu(shared_msrs);
5649 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5651 ++vcpu->stat.halt_exits;
5652 if (irqchip_in_kernel(vcpu->kvm)) {
5653 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5656 vcpu->run->exit_reason = KVM_EXIT_HLT;
5660 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5662 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5664 u64 param, ingpa, outgpa, ret;
5665 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5666 bool fast, longmode;
5670 * hypercall generates UD from non zero cpl and real mode
5673 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5674 kvm_queue_exception(vcpu, UD_VECTOR);
5678 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5679 longmode = is_long_mode(vcpu) && cs_l == 1;
5682 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5683 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5684 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5685 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5686 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5687 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5689 #ifdef CONFIG_X86_64
5691 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5692 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5693 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5697 code = param & 0xffff;
5698 fast = (param >> 16) & 0x1;
5699 rep_cnt = (param >> 32) & 0xfff;
5700 rep_idx = (param >> 48) & 0xfff;
5702 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5705 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5706 kvm_vcpu_on_spin(vcpu);
5709 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5713 ret = res | (((u64)rep_done & 0xfff) << 32);
5715 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5717 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5718 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5725 * kvm_pv_kick_cpu_op: Kick a vcpu.
5727 * @apicid - apicid of vcpu to be kicked.
5729 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5731 struct kvm_lapic_irq lapic_irq;
5733 lapic_irq.shorthand = 0;
5734 lapic_irq.dest_mode = 0;
5735 lapic_irq.dest_id = apicid;
5737 lapic_irq.delivery_mode = APIC_DM_REMRD;
5738 kvm_irq_delivery_to_apic(kvm, 0, &lapic_irq, NULL);
5741 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5743 unsigned long nr, a0, a1, a2, a3, ret;
5746 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5747 return kvm_hv_hypercall(vcpu);
5749 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5750 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5751 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5752 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5753 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5755 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5757 if (!is_long_mode(vcpu)) {
5765 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5771 case KVM_HC_VAPIC_POLL_IRQ:
5774 case KVM_HC_KICK_CPU:
5775 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5783 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5784 ++vcpu->stat.hypercalls;
5787 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5789 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5791 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5792 char instruction[3];
5793 unsigned long rip = kvm_rip_read(vcpu);
5795 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5797 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5801 * Check if userspace requested an interrupt window, and that the
5802 * interrupt window is open.
5804 * No need to exit to userspace if we already have an interrupt queued.
5806 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5808 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5809 vcpu->run->request_interrupt_window &&
5810 kvm_arch_interrupt_allowed(vcpu));
5813 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5815 struct kvm_run *kvm_run = vcpu->run;
5817 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5818 kvm_run->cr8 = kvm_get_cr8(vcpu);
5819 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5820 if (irqchip_in_kernel(vcpu->kvm))
5821 kvm_run->ready_for_interrupt_injection = 1;
5823 kvm_run->ready_for_interrupt_injection =
5824 kvm_arch_interrupt_allowed(vcpu) &&
5825 !kvm_cpu_has_interrupt(vcpu) &&
5826 !kvm_event_needs_reinjection(vcpu);
5829 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5833 if (!kvm_x86_ops->update_cr8_intercept)
5836 if (!vcpu->arch.apic)
5839 if (!vcpu->arch.apic->vapic_addr)
5840 max_irr = kvm_lapic_find_highest_irr(vcpu);
5847 tpr = kvm_lapic_get_cr8(vcpu);
5849 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5852 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
5856 /* try to reinject previous events if any */
5857 if (vcpu->arch.exception.pending) {
5858 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5859 vcpu->arch.exception.has_error_code,
5860 vcpu->arch.exception.error_code);
5861 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5862 vcpu->arch.exception.has_error_code,
5863 vcpu->arch.exception.error_code,
5864 vcpu->arch.exception.reinject);
5868 if (vcpu->arch.nmi_injected) {
5869 kvm_x86_ops->set_nmi(vcpu);
5873 if (vcpu->arch.interrupt.pending) {
5874 kvm_x86_ops->set_irq(vcpu);
5878 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
5879 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
5884 /* try to inject new event if pending */
5885 if (vcpu->arch.nmi_pending) {
5886 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5887 --vcpu->arch.nmi_pending;
5888 vcpu->arch.nmi_injected = true;
5889 kvm_x86_ops->set_nmi(vcpu);
5891 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
5892 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5893 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5895 kvm_x86_ops->set_irq(vcpu);
5901 static void process_nmi(struct kvm_vcpu *vcpu)
5906 * x86 is limited to one NMI running, and one NMI pending after it.
5907 * If an NMI is already in progress, limit further NMIs to just one.
5908 * Otherwise, allow two (and we'll inject the first one immediately).
5910 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5913 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5914 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5915 kvm_make_request(KVM_REQ_EVENT, vcpu);
5918 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
5920 u64 eoi_exit_bitmap[4];
5923 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
5926 memset(eoi_exit_bitmap, 0, 32);
5929 kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
5930 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
5931 kvm_apic_update_tmr(vcpu, tmr);
5935 * Returns 1 to let __vcpu_run() continue the guest execution loop without
5936 * exiting to the userspace. Otherwise, the value will be returned to the
5939 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5942 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5943 vcpu->run->request_interrupt_window;
5944 bool req_immediate_exit = false;
5946 if (vcpu->requests) {
5947 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5948 kvm_mmu_unload(vcpu);
5949 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5950 __kvm_migrate_timers(vcpu);
5951 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
5952 kvm_gen_update_masterclock(vcpu->kvm);
5953 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
5954 kvm_gen_kvmclock_update(vcpu);
5955 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5956 r = kvm_guest_time_update(vcpu);
5960 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5961 kvm_mmu_sync_roots(vcpu);
5962 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5963 kvm_x86_ops->tlb_flush(vcpu);
5964 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5965 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5969 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5970 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5974 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5975 vcpu->fpu_active = 0;
5976 kvm_x86_ops->fpu_deactivate(vcpu);
5978 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5979 /* Page is swapped out. Do synthetic halt */
5980 vcpu->arch.apf.halted = true;
5984 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5985 record_steal_time(vcpu);
5986 if (kvm_check_request(KVM_REQ_NMI, vcpu))
5988 if (kvm_check_request(KVM_REQ_PMU, vcpu))
5989 kvm_handle_pmu_event(vcpu);
5990 if (kvm_check_request(KVM_REQ_PMI, vcpu))
5991 kvm_deliver_pmi(vcpu);
5992 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
5993 vcpu_scan_ioapic(vcpu);
5996 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5997 kvm_apic_accept_events(vcpu);
5998 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6003 if (inject_pending_event(vcpu, req_int_win) != 0)
6004 req_immediate_exit = true;
6005 /* enable NMI/IRQ window open exits if needed */
6006 else if (vcpu->arch.nmi_pending)
6007 kvm_x86_ops->enable_nmi_window(vcpu);
6008 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6009 kvm_x86_ops->enable_irq_window(vcpu);
6011 if (kvm_lapic_enabled(vcpu)) {
6013 * Update architecture specific hints for APIC
6014 * virtual interrupt delivery.
6016 if (kvm_x86_ops->hwapic_irr_update)
6017 kvm_x86_ops->hwapic_irr_update(vcpu,
6018 kvm_lapic_find_highest_irr(vcpu));
6019 update_cr8_intercept(vcpu);
6020 kvm_lapic_sync_to_vapic(vcpu);
6024 r = kvm_mmu_reload(vcpu);
6026 goto cancel_injection;
6031 kvm_x86_ops->prepare_guest_switch(vcpu);
6032 if (vcpu->fpu_active)
6033 kvm_load_guest_fpu(vcpu);
6034 kvm_load_guest_xcr0(vcpu);
6036 vcpu->mode = IN_GUEST_MODE;
6038 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6040 /* We should set ->mode before check ->requests,
6041 * see the comment in make_all_cpus_request.
6043 smp_mb__after_srcu_read_unlock();
6045 local_irq_disable();
6047 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6048 || need_resched() || signal_pending(current)) {
6049 vcpu->mode = OUTSIDE_GUEST_MODE;
6053 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6055 goto cancel_injection;
6058 if (req_immediate_exit)
6059 smp_send_reschedule(vcpu->cpu);
6063 if (unlikely(vcpu->arch.switch_db_regs)) {
6065 set_debugreg(vcpu->arch.eff_db[0], 0);
6066 set_debugreg(vcpu->arch.eff_db[1], 1);
6067 set_debugreg(vcpu->arch.eff_db[2], 2);
6068 set_debugreg(vcpu->arch.eff_db[3], 3);
6069 set_debugreg(vcpu->arch.dr6, 6);
6072 trace_kvm_entry(vcpu->vcpu_id);
6073 kvm_x86_ops->run(vcpu);
6076 * Do this here before restoring debug registers on the host. And
6077 * since we do this before handling the vmexit, a DR access vmexit
6078 * can (a) read the correct value of the debug registers, (b) set
6079 * KVM_DEBUGREG_WONT_EXIT again.
6081 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6084 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6085 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6086 for (i = 0; i < KVM_NR_DB_REGS; i++)
6087 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6091 * If the guest has used debug registers, at least dr7
6092 * will be disabled while returning to the host.
6093 * If we don't have active breakpoints in the host, we don't
6094 * care about the messed up debug address registers. But if
6095 * we have some of them active, restore the old state.
6097 if (hw_breakpoint_active())
6098 hw_breakpoint_restore();
6100 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
6103 vcpu->mode = OUTSIDE_GUEST_MODE;
6106 /* Interrupt is enabled by handle_external_intr() */
6107 kvm_x86_ops->handle_external_intr(vcpu);
6112 * We must have an instruction between local_irq_enable() and
6113 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6114 * the interrupt shadow. The stat.exits increment will do nicely.
6115 * But we need to prevent reordering, hence this barrier():
6123 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6126 * Profile KVM exit RIPs:
6128 if (unlikely(prof_on == KVM_PROFILING)) {
6129 unsigned long rip = kvm_rip_read(vcpu);
6130 profile_hit(KVM_PROFILING, (void *)rip);
6133 if (unlikely(vcpu->arch.tsc_always_catchup))
6134 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6136 if (vcpu->arch.apic_attention)
6137 kvm_lapic_sync_from_vapic(vcpu);
6139 r = kvm_x86_ops->handle_exit(vcpu);
6143 kvm_x86_ops->cancel_injection(vcpu);
6144 if (unlikely(vcpu->arch.apic_attention))
6145 kvm_lapic_sync_from_vapic(vcpu);
6151 static int __vcpu_run(struct kvm_vcpu *vcpu)
6154 struct kvm *kvm = vcpu->kvm;
6156 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6160 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6161 !vcpu->arch.apf.halted)
6162 r = vcpu_enter_guest(vcpu);
6164 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6165 kvm_vcpu_block(vcpu);
6166 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6167 if (kvm_check_request(KVM_REQ_UNHALT, vcpu)) {
6168 kvm_apic_accept_events(vcpu);
6169 switch(vcpu->arch.mp_state) {
6170 case KVM_MP_STATE_HALTED:
6171 vcpu->arch.pv.pv_unhalted = false;
6172 vcpu->arch.mp_state =
6173 KVM_MP_STATE_RUNNABLE;
6174 case KVM_MP_STATE_RUNNABLE:
6175 vcpu->arch.apf.halted = false;
6177 case KVM_MP_STATE_INIT_RECEIVED:
6189 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6190 if (kvm_cpu_has_pending_timer(vcpu))
6191 kvm_inject_pending_timer_irqs(vcpu);
6193 if (dm_request_for_irq_injection(vcpu)) {
6195 vcpu->run->exit_reason = KVM_EXIT_INTR;
6196 ++vcpu->stat.request_irq_exits;
6199 kvm_check_async_pf_completion(vcpu);
6201 if (signal_pending(current)) {
6203 vcpu->run->exit_reason = KVM_EXIT_INTR;
6204 ++vcpu->stat.signal_exits;
6206 if (need_resched()) {
6207 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6209 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6213 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6218 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6221 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6222 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6223 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6224 if (r != EMULATE_DONE)
6229 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6231 BUG_ON(!vcpu->arch.pio.count);
6233 return complete_emulated_io(vcpu);
6237 * Implements the following, as a state machine:
6241 * for each mmio piece in the fragment
6249 * for each mmio piece in the fragment
6254 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6256 struct kvm_run *run = vcpu->run;
6257 struct kvm_mmio_fragment *frag;
6260 BUG_ON(!vcpu->mmio_needed);
6262 /* Complete previous fragment */
6263 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6264 len = min(8u, frag->len);
6265 if (!vcpu->mmio_is_write)
6266 memcpy(frag->data, run->mmio.data, len);
6268 if (frag->len <= 8) {
6269 /* Switch to the next fragment. */
6271 vcpu->mmio_cur_fragment++;
6273 /* Go forward to the next mmio piece. */
6279 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6280 vcpu->mmio_needed = 0;
6282 /* FIXME: return into emulator if single-stepping. */
6283 if (vcpu->mmio_is_write)
6285 vcpu->mmio_read_completed = 1;
6286 return complete_emulated_io(vcpu);
6289 run->exit_reason = KVM_EXIT_MMIO;
6290 run->mmio.phys_addr = frag->gpa;
6291 if (vcpu->mmio_is_write)
6292 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6293 run->mmio.len = min(8u, frag->len);
6294 run->mmio.is_write = vcpu->mmio_is_write;
6295 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6300 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6305 if (!tsk_used_math(current) && init_fpu(current))
6308 if (vcpu->sigset_active)
6309 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6311 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6312 kvm_vcpu_block(vcpu);
6313 kvm_apic_accept_events(vcpu);
6314 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6319 /* re-sync apic's tpr */
6320 if (!irqchip_in_kernel(vcpu->kvm)) {
6321 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6327 if (unlikely(vcpu->arch.complete_userspace_io)) {
6328 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6329 vcpu->arch.complete_userspace_io = NULL;
6334 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6336 r = __vcpu_run(vcpu);
6339 post_kvm_run_save(vcpu);
6340 if (vcpu->sigset_active)
6341 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6346 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6348 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6350 * We are here if userspace calls get_regs() in the middle of
6351 * instruction emulation. Registers state needs to be copied
6352 * back from emulation context to vcpu. Userspace shouldn't do
6353 * that usually, but some bad designed PV devices (vmware
6354 * backdoor interface) need this to work
6356 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6357 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6359 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6360 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6361 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6362 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6363 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6364 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6365 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6366 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6367 #ifdef CONFIG_X86_64
6368 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6369 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6370 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6371 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6372 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6373 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6374 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6375 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6378 regs->rip = kvm_rip_read(vcpu);
6379 regs->rflags = kvm_get_rflags(vcpu);
6384 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6386 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6387 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6389 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6390 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6391 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6392 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6393 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6394 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6395 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6396 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6397 #ifdef CONFIG_X86_64
6398 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6399 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6400 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6401 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6402 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6403 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6404 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6405 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6408 kvm_rip_write(vcpu, regs->rip);
6409 kvm_set_rflags(vcpu, regs->rflags);
6411 vcpu->arch.exception.pending = false;
6413 kvm_make_request(KVM_REQ_EVENT, vcpu);
6418 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6420 struct kvm_segment cs;
6422 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6426 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6428 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6429 struct kvm_sregs *sregs)
6433 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6434 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6435 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6436 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6437 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6438 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6440 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6441 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6443 kvm_x86_ops->get_idt(vcpu, &dt);
6444 sregs->idt.limit = dt.size;
6445 sregs->idt.base = dt.address;
6446 kvm_x86_ops->get_gdt(vcpu, &dt);
6447 sregs->gdt.limit = dt.size;
6448 sregs->gdt.base = dt.address;
6450 sregs->cr0 = kvm_read_cr0(vcpu);
6451 sregs->cr2 = vcpu->arch.cr2;
6452 sregs->cr3 = kvm_read_cr3(vcpu);
6453 sregs->cr4 = kvm_read_cr4(vcpu);
6454 sregs->cr8 = kvm_get_cr8(vcpu);
6455 sregs->efer = vcpu->arch.efer;
6456 sregs->apic_base = kvm_get_apic_base(vcpu);
6458 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6460 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6461 set_bit(vcpu->arch.interrupt.nr,
6462 (unsigned long *)sregs->interrupt_bitmap);
6467 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6468 struct kvm_mp_state *mp_state)
6470 kvm_apic_accept_events(vcpu);
6471 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
6472 vcpu->arch.pv.pv_unhalted)
6473 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
6475 mp_state->mp_state = vcpu->arch.mp_state;
6480 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6481 struct kvm_mp_state *mp_state)
6483 if (!kvm_vcpu_has_lapic(vcpu) &&
6484 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6487 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6488 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6489 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6491 vcpu->arch.mp_state = mp_state->mp_state;
6492 kvm_make_request(KVM_REQ_EVENT, vcpu);
6496 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6497 int reason, bool has_error_code, u32 error_code)
6499 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6502 init_emulate_ctxt(vcpu);
6504 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6505 has_error_code, error_code);
6508 return EMULATE_FAIL;
6510 kvm_rip_write(vcpu, ctxt->eip);
6511 kvm_set_rflags(vcpu, ctxt->eflags);
6512 kvm_make_request(KVM_REQ_EVENT, vcpu);
6513 return EMULATE_DONE;
6515 EXPORT_SYMBOL_GPL(kvm_task_switch);
6517 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6518 struct kvm_sregs *sregs)
6520 struct msr_data apic_base_msr;
6521 int mmu_reset_needed = 0;
6522 int pending_vec, max_bits, idx;
6525 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6528 dt.size = sregs->idt.limit;
6529 dt.address = sregs->idt.base;
6530 kvm_x86_ops->set_idt(vcpu, &dt);
6531 dt.size = sregs->gdt.limit;
6532 dt.address = sregs->gdt.base;
6533 kvm_x86_ops->set_gdt(vcpu, &dt);
6535 vcpu->arch.cr2 = sregs->cr2;
6536 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6537 vcpu->arch.cr3 = sregs->cr3;
6538 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6540 kvm_set_cr8(vcpu, sregs->cr8);
6542 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6543 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6544 apic_base_msr.data = sregs->apic_base;
6545 apic_base_msr.host_initiated = true;
6546 kvm_set_apic_base(vcpu, &apic_base_msr);
6548 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6549 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6550 vcpu->arch.cr0 = sregs->cr0;
6552 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6553 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6554 if (sregs->cr4 & X86_CR4_OSXSAVE)
6555 kvm_update_cpuid(vcpu);
6557 idx = srcu_read_lock(&vcpu->kvm->srcu);
6558 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6559 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6560 mmu_reset_needed = 1;
6562 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6564 if (mmu_reset_needed)
6565 kvm_mmu_reset_context(vcpu);
6567 max_bits = KVM_NR_INTERRUPTS;
6568 pending_vec = find_first_bit(
6569 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6570 if (pending_vec < max_bits) {
6571 kvm_queue_interrupt(vcpu, pending_vec, false);
6572 pr_debug("Set back pending irq %d\n", pending_vec);
6575 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6576 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6577 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6578 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6579 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6580 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6582 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6583 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6585 update_cr8_intercept(vcpu);
6587 /* Older userspace won't unhalt the vcpu on reset. */
6588 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6589 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6591 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6593 kvm_make_request(KVM_REQ_EVENT, vcpu);
6598 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6599 struct kvm_guest_debug *dbg)
6601 unsigned long rflags;
6604 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6606 if (vcpu->arch.exception.pending)
6608 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6609 kvm_queue_exception(vcpu, DB_VECTOR);
6611 kvm_queue_exception(vcpu, BP_VECTOR);
6615 * Read rflags as long as potentially injected trace flags are still
6618 rflags = kvm_get_rflags(vcpu);
6620 vcpu->guest_debug = dbg->control;
6621 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6622 vcpu->guest_debug = 0;
6624 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6625 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6626 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6627 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6629 for (i = 0; i < KVM_NR_DB_REGS; i++)
6630 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6632 kvm_update_dr7(vcpu);
6634 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6635 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6636 get_segment_base(vcpu, VCPU_SREG_CS);
6639 * Trigger an rflags update that will inject or remove the trace
6642 kvm_set_rflags(vcpu, rflags);
6644 kvm_x86_ops->update_db_bp_intercept(vcpu);
6654 * Translate a guest virtual address to a guest physical address.
6656 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6657 struct kvm_translation *tr)
6659 unsigned long vaddr = tr->linear_address;
6663 idx = srcu_read_lock(&vcpu->kvm->srcu);
6664 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6665 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6666 tr->physical_address = gpa;
6667 tr->valid = gpa != UNMAPPED_GVA;
6674 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6676 struct i387_fxsave_struct *fxsave =
6677 &vcpu->arch.guest_fpu.state->fxsave;
6679 memcpy(fpu->fpr, fxsave->st_space, 128);
6680 fpu->fcw = fxsave->cwd;
6681 fpu->fsw = fxsave->swd;
6682 fpu->ftwx = fxsave->twd;
6683 fpu->last_opcode = fxsave->fop;
6684 fpu->last_ip = fxsave->rip;
6685 fpu->last_dp = fxsave->rdp;
6686 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6691 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6693 struct i387_fxsave_struct *fxsave =
6694 &vcpu->arch.guest_fpu.state->fxsave;
6696 memcpy(fxsave->st_space, fpu->fpr, 128);
6697 fxsave->cwd = fpu->fcw;
6698 fxsave->swd = fpu->fsw;
6699 fxsave->twd = fpu->ftwx;
6700 fxsave->fop = fpu->last_opcode;
6701 fxsave->rip = fpu->last_ip;
6702 fxsave->rdp = fpu->last_dp;
6703 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6708 int fx_init(struct kvm_vcpu *vcpu)
6712 err = fpu_alloc(&vcpu->arch.guest_fpu);
6716 fpu_finit(&vcpu->arch.guest_fpu);
6719 * Ensure guest xcr0 is valid for loading
6721 vcpu->arch.xcr0 = XSTATE_FP;
6723 vcpu->arch.cr0 |= X86_CR0_ET;
6727 EXPORT_SYMBOL_GPL(fx_init);
6729 static void fx_free(struct kvm_vcpu *vcpu)
6731 fpu_free(&vcpu->arch.guest_fpu);
6734 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6736 if (vcpu->guest_fpu_loaded)
6740 * Restore all possible states in the guest,
6741 * and assume host would use all available bits.
6742 * Guest xcr0 would be loaded later.
6744 kvm_put_guest_xcr0(vcpu);
6745 vcpu->guest_fpu_loaded = 1;
6746 __kernel_fpu_begin();
6747 fpu_restore_checking(&vcpu->arch.guest_fpu);
6751 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6753 kvm_put_guest_xcr0(vcpu);
6755 if (!vcpu->guest_fpu_loaded)
6758 vcpu->guest_fpu_loaded = 0;
6759 fpu_save_init(&vcpu->arch.guest_fpu);
6761 ++vcpu->stat.fpu_reload;
6762 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6766 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6768 kvmclock_reset(vcpu);
6770 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6772 kvm_x86_ops->vcpu_free(vcpu);
6775 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6778 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6779 printk_once(KERN_WARNING
6780 "kvm: SMP vm created on host with unstable TSC; "
6781 "guest TSC will not be reliable\n");
6782 return kvm_x86_ops->vcpu_create(kvm, id);
6785 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6789 vcpu->arch.mtrr_state.have_fixed = 1;
6790 r = vcpu_load(vcpu);
6793 kvm_vcpu_reset(vcpu);
6794 kvm_mmu_setup(vcpu);
6800 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
6803 struct msr_data msr;
6804 struct kvm *kvm = vcpu->kvm;
6806 r = vcpu_load(vcpu);
6810 msr.index = MSR_IA32_TSC;
6811 msr.host_initiated = true;
6812 kvm_write_tsc(vcpu, &msr);
6815 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
6816 KVMCLOCK_SYNC_PERIOD);
6821 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6824 vcpu->arch.apf.msr_val = 0;
6826 r = vcpu_load(vcpu);
6828 kvm_mmu_unload(vcpu);
6832 kvm_x86_ops->vcpu_free(vcpu);
6835 void kvm_vcpu_reset(struct kvm_vcpu *vcpu)
6837 atomic_set(&vcpu->arch.nmi_queued, 0);
6838 vcpu->arch.nmi_pending = 0;
6839 vcpu->arch.nmi_injected = false;
6841 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6842 vcpu->arch.dr6 = DR6_FIXED_1;
6843 kvm_update_dr6(vcpu);
6844 vcpu->arch.dr7 = DR7_FIXED_1;
6845 kvm_update_dr7(vcpu);
6847 kvm_make_request(KVM_REQ_EVENT, vcpu);
6848 vcpu->arch.apf.msr_val = 0;
6849 vcpu->arch.st.msr_val = 0;
6851 kvmclock_reset(vcpu);
6853 kvm_clear_async_pf_completion_queue(vcpu);
6854 kvm_async_pf_hash_reset(vcpu);
6855 vcpu->arch.apf.halted = false;
6857 kvm_pmu_reset(vcpu);
6859 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
6860 vcpu->arch.regs_avail = ~0;
6861 vcpu->arch.regs_dirty = ~0;
6863 kvm_x86_ops->vcpu_reset(vcpu);
6866 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, unsigned int vector)
6868 struct kvm_segment cs;
6870 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6871 cs.selector = vector << 8;
6872 cs.base = vector << 12;
6873 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6874 kvm_rip_write(vcpu, 0);
6877 int kvm_arch_hardware_enable(void *garbage)
6880 struct kvm_vcpu *vcpu;
6885 bool stable, backwards_tsc = false;
6887 kvm_shared_msr_cpu_online();
6888 ret = kvm_x86_ops->hardware_enable(garbage);
6892 local_tsc = native_read_tsc();
6893 stable = !check_tsc_unstable();
6894 list_for_each_entry(kvm, &vm_list, vm_list) {
6895 kvm_for_each_vcpu(i, vcpu, kvm) {
6896 if (!stable && vcpu->cpu == smp_processor_id())
6897 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
6898 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
6899 backwards_tsc = true;
6900 if (vcpu->arch.last_host_tsc > max_tsc)
6901 max_tsc = vcpu->arch.last_host_tsc;
6907 * Sometimes, even reliable TSCs go backwards. This happens on
6908 * platforms that reset TSC during suspend or hibernate actions, but
6909 * maintain synchronization. We must compensate. Fortunately, we can
6910 * detect that condition here, which happens early in CPU bringup,
6911 * before any KVM threads can be running. Unfortunately, we can't
6912 * bring the TSCs fully up to date with real time, as we aren't yet far
6913 * enough into CPU bringup that we know how much real time has actually
6914 * elapsed; our helper function, get_kernel_ns() will be using boot
6915 * variables that haven't been updated yet.
6917 * So we simply find the maximum observed TSC above, then record the
6918 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
6919 * the adjustment will be applied. Note that we accumulate
6920 * adjustments, in case multiple suspend cycles happen before some VCPU
6921 * gets a chance to run again. In the event that no KVM threads get a
6922 * chance to run, we will miss the entire elapsed period, as we'll have
6923 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
6924 * loose cycle time. This isn't too big a deal, since the loss will be
6925 * uniform across all VCPUs (not to mention the scenario is extremely
6926 * unlikely). It is possible that a second hibernate recovery happens
6927 * much faster than a first, causing the observed TSC here to be
6928 * smaller; this would require additional padding adjustment, which is
6929 * why we set last_host_tsc to the local tsc observed here.
6931 * N.B. - this code below runs only on platforms with reliable TSC,
6932 * as that is the only way backwards_tsc is set above. Also note
6933 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
6934 * have the same delta_cyc adjustment applied if backwards_tsc
6935 * is detected. Note further, this adjustment is only done once,
6936 * as we reset last_host_tsc on all VCPUs to stop this from being
6937 * called multiple times (one for each physical CPU bringup).
6939 * Platforms with unreliable TSCs don't have to deal with this, they
6940 * will be compensated by the logic in vcpu_load, which sets the TSC to
6941 * catchup mode. This will catchup all VCPUs to real time, but cannot
6942 * guarantee that they stay in perfect synchronization.
6944 if (backwards_tsc) {
6945 u64 delta_cyc = max_tsc - local_tsc;
6946 list_for_each_entry(kvm, &vm_list, vm_list) {
6947 kvm_for_each_vcpu(i, vcpu, kvm) {
6948 vcpu->arch.tsc_offset_adjustment += delta_cyc;
6949 vcpu->arch.last_host_tsc = local_tsc;
6950 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
6955 * We have to disable TSC offset matching.. if you were
6956 * booting a VM while issuing an S4 host suspend....
6957 * you may have some problem. Solving this issue is
6958 * left as an exercise to the reader.
6960 kvm->arch.last_tsc_nsec = 0;
6961 kvm->arch.last_tsc_write = 0;
6968 void kvm_arch_hardware_disable(void *garbage)
6970 kvm_x86_ops->hardware_disable(garbage);
6971 drop_user_return_notifiers(garbage);
6974 int kvm_arch_hardware_setup(void)
6976 return kvm_x86_ops->hardware_setup();
6979 void kvm_arch_hardware_unsetup(void)
6981 kvm_x86_ops->hardware_unsetup();
6984 void kvm_arch_check_processor_compat(void *rtn)
6986 kvm_x86_ops->check_processor_compatibility(rtn);
6989 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
6991 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
6994 struct static_key kvm_no_apic_vcpu __read_mostly;
6996 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7002 BUG_ON(vcpu->kvm == NULL);
7005 vcpu->arch.pv.pv_unhalted = false;
7006 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7007 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
7008 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7010 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7012 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7017 vcpu->arch.pio_data = page_address(page);
7019 kvm_set_tsc_khz(vcpu, max_tsc_khz);
7021 r = kvm_mmu_create(vcpu);
7023 goto fail_free_pio_data;
7025 if (irqchip_in_kernel(kvm)) {
7026 r = kvm_create_lapic(vcpu);
7028 goto fail_mmu_destroy;
7030 static_key_slow_inc(&kvm_no_apic_vcpu);
7032 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7034 if (!vcpu->arch.mce_banks) {
7036 goto fail_free_lapic;
7038 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7040 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7042 goto fail_free_mce_banks;
7047 goto fail_free_wbinvd_dirty_mask;
7049 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7050 vcpu->arch.pv_time_enabled = false;
7052 vcpu->arch.guest_supported_xcr0 = 0;
7053 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7055 kvm_async_pf_hash_reset(vcpu);
7059 fail_free_wbinvd_dirty_mask:
7060 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7061 fail_free_mce_banks:
7062 kfree(vcpu->arch.mce_banks);
7064 kvm_free_lapic(vcpu);
7066 kvm_mmu_destroy(vcpu);
7068 free_page((unsigned long)vcpu->arch.pio_data);
7073 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7077 kvm_pmu_destroy(vcpu);
7078 kfree(vcpu->arch.mce_banks);
7079 kvm_free_lapic(vcpu);
7080 idx = srcu_read_lock(&vcpu->kvm->srcu);
7081 kvm_mmu_destroy(vcpu);
7082 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7083 free_page((unsigned long)vcpu->arch.pio_data);
7084 if (!irqchip_in_kernel(vcpu->kvm))
7085 static_key_slow_dec(&kvm_no_apic_vcpu);
7088 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7093 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7094 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7095 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7096 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7098 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7099 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7100 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7101 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7102 &kvm->arch.irq_sources_bitmap);
7104 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7105 mutex_init(&kvm->arch.apic_map_lock);
7106 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7108 pvclock_update_vm_gtod_copy(kvm);
7110 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7111 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7116 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7119 r = vcpu_load(vcpu);
7121 kvm_mmu_unload(vcpu);
7125 static void kvm_free_vcpus(struct kvm *kvm)
7128 struct kvm_vcpu *vcpu;
7131 * Unpin any mmu pages first.
7133 kvm_for_each_vcpu(i, vcpu, kvm) {
7134 kvm_clear_async_pf_completion_queue(vcpu);
7135 kvm_unload_vcpu_mmu(vcpu);
7137 kvm_for_each_vcpu(i, vcpu, kvm)
7138 kvm_arch_vcpu_free(vcpu);
7140 mutex_lock(&kvm->lock);
7141 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7142 kvm->vcpus[i] = NULL;
7144 atomic_set(&kvm->online_vcpus, 0);
7145 mutex_unlock(&kvm->lock);
7148 void kvm_arch_sync_events(struct kvm *kvm)
7150 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7151 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7152 kvm_free_all_assigned_devices(kvm);
7156 void kvm_arch_destroy_vm(struct kvm *kvm)
7158 if (current->mm == kvm->mm) {
7160 * Free memory regions allocated on behalf of userspace,
7161 * unless the the memory map has changed due to process exit
7164 struct kvm_userspace_memory_region mem;
7165 memset(&mem, 0, sizeof(mem));
7166 mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
7167 kvm_set_memory_region(kvm, &mem);
7169 mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
7170 kvm_set_memory_region(kvm, &mem);
7172 mem.slot = TSS_PRIVATE_MEMSLOT;
7173 kvm_set_memory_region(kvm, &mem);
7175 kvm_iommu_unmap_guest(kvm);
7176 kfree(kvm->arch.vpic);
7177 kfree(kvm->arch.vioapic);
7178 kvm_free_vcpus(kvm);
7179 if (kvm->arch.apic_access_page)
7180 put_page(kvm->arch.apic_access_page);
7181 if (kvm->arch.ept_identity_pagetable)
7182 put_page(kvm->arch.ept_identity_pagetable);
7183 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7186 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7187 struct kvm_memory_slot *dont)
7191 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7192 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7193 kvm_kvfree(free->arch.rmap[i]);
7194 free->arch.rmap[i] = NULL;
7199 if (!dont || free->arch.lpage_info[i - 1] !=
7200 dont->arch.lpage_info[i - 1]) {
7201 kvm_kvfree(free->arch.lpage_info[i - 1]);
7202 free->arch.lpage_info[i - 1] = NULL;
7207 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7208 unsigned long npages)
7212 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7217 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7218 slot->base_gfn, level) + 1;
7220 slot->arch.rmap[i] =
7221 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7222 if (!slot->arch.rmap[i])
7227 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7228 sizeof(*slot->arch.lpage_info[i - 1]));
7229 if (!slot->arch.lpage_info[i - 1])
7232 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7233 slot->arch.lpage_info[i - 1][0].write_count = 1;
7234 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7235 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7236 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7238 * If the gfn and userspace address are not aligned wrt each
7239 * other, or if explicitly asked to, disable large page
7240 * support for this slot
7242 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7243 !kvm_largepages_enabled()) {
7246 for (j = 0; j < lpages; ++j)
7247 slot->arch.lpage_info[i - 1][j].write_count = 1;
7254 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7255 kvm_kvfree(slot->arch.rmap[i]);
7256 slot->arch.rmap[i] = NULL;
7260 kvm_kvfree(slot->arch.lpage_info[i - 1]);
7261 slot->arch.lpage_info[i - 1] = NULL;
7266 void kvm_arch_memslots_updated(struct kvm *kvm)
7269 * memslots->generation has been incremented.
7270 * mmio generation may have reached its maximum value.
7272 kvm_mmu_invalidate_mmio_sptes(kvm);
7275 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7276 struct kvm_memory_slot *memslot,
7277 struct kvm_userspace_memory_region *mem,
7278 enum kvm_mr_change change)
7281 * Only private memory slots need to be mapped here since
7282 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
7284 if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
7285 unsigned long userspace_addr;
7288 * MAP_SHARED to prevent internal slot pages from being moved
7291 userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
7292 PROT_READ | PROT_WRITE,
7293 MAP_SHARED | MAP_ANONYMOUS, 0);
7295 if (IS_ERR((void *)userspace_addr))
7296 return PTR_ERR((void *)userspace_addr);
7298 memslot->userspace_addr = userspace_addr;
7304 void kvm_arch_commit_memory_region(struct kvm *kvm,
7305 struct kvm_userspace_memory_region *mem,
7306 const struct kvm_memory_slot *old,
7307 enum kvm_mr_change change)
7310 int nr_mmu_pages = 0;
7312 if ((mem->slot >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_DELETE)) {
7315 ret = vm_munmap(old->userspace_addr,
7316 old->npages * PAGE_SIZE);
7319 "kvm_vm_ioctl_set_memory_region: "
7320 "failed to munmap memory\n");
7323 if (!kvm->arch.n_requested_mmu_pages)
7324 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7327 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7329 * Write protect all pages for dirty logging.
7331 * All the sptes including the large sptes which point to this
7332 * slot are set to readonly. We can not create any new large
7333 * spte on this slot until the end of the logging.
7335 * See the comments in fast_page_fault().
7337 if ((change != KVM_MR_DELETE) && (mem->flags & KVM_MEM_LOG_DIRTY_PAGES))
7338 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
7341 void kvm_arch_flush_shadow_all(struct kvm *kvm)
7343 kvm_mmu_invalidate_zap_all_pages(kvm);
7346 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7347 struct kvm_memory_slot *slot)
7349 kvm_mmu_invalidate_zap_all_pages(kvm);
7352 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
7354 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7355 kvm_x86_ops->check_nested_events(vcpu, false);
7357 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7358 !vcpu->arch.apf.halted)
7359 || !list_empty_careful(&vcpu->async_pf.done)
7360 || kvm_apic_has_events(vcpu)
7361 || vcpu->arch.pv.pv_unhalted
7362 || atomic_read(&vcpu->arch.nmi_queued) ||
7363 (kvm_arch_interrupt_allowed(vcpu) &&
7364 kvm_cpu_has_interrupt(vcpu));
7367 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
7369 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
7372 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7374 return kvm_x86_ops->interrupt_allowed(vcpu);
7377 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7379 unsigned long current_rip = kvm_rip_read(vcpu) +
7380 get_segment_base(vcpu, VCPU_SREG_CS);
7382 return current_rip == linear_rip;
7384 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7386 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7388 unsigned long rflags;
7390 rflags = kvm_x86_ops->get_rflags(vcpu);
7391 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7392 rflags &= ~X86_EFLAGS_TF;
7395 EXPORT_SYMBOL_GPL(kvm_get_rflags);
7397 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7399 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
7400 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
7401 rflags |= X86_EFLAGS_TF;
7402 kvm_x86_ops->set_rflags(vcpu, rflags);
7403 kvm_make_request(KVM_REQ_EVENT, vcpu);
7405 EXPORT_SYMBOL_GPL(kvm_set_rflags);
7407 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7411 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
7415 r = kvm_mmu_reload(vcpu);
7419 if (!vcpu->arch.mmu.direct_map &&
7420 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7423 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7426 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7428 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7431 static inline u32 kvm_async_pf_next_probe(u32 key)
7433 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7436 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7438 u32 key = kvm_async_pf_hash_fn(gfn);
7440 while (vcpu->arch.apf.gfns[key] != ~0)
7441 key = kvm_async_pf_next_probe(key);
7443 vcpu->arch.apf.gfns[key] = gfn;
7446 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7449 u32 key = kvm_async_pf_hash_fn(gfn);
7451 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
7452 (vcpu->arch.apf.gfns[key] != gfn &&
7453 vcpu->arch.apf.gfns[key] != ~0); i++)
7454 key = kvm_async_pf_next_probe(key);
7459 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7461 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7464 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7468 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7470 vcpu->arch.apf.gfns[i] = ~0;
7472 j = kvm_async_pf_next_probe(j);
7473 if (vcpu->arch.apf.gfns[j] == ~0)
7475 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7477 * k lies cyclically in ]i,j]
7479 * |....j i.k.| or |.k..j i...|
7481 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7482 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7487 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7490 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7494 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7495 struct kvm_async_pf *work)
7497 struct x86_exception fault;
7499 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
7500 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7502 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
7503 (vcpu->arch.apf.send_user_only &&
7504 kvm_x86_ops->get_cpl(vcpu) == 0))
7505 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7506 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
7507 fault.vector = PF_VECTOR;
7508 fault.error_code_valid = true;
7509 fault.error_code = 0;
7510 fault.nested_page_fault = false;
7511 fault.address = work->arch.token;
7512 kvm_inject_page_fault(vcpu, &fault);
7516 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7517 struct kvm_async_pf *work)
7519 struct x86_exception fault;
7521 trace_kvm_async_pf_ready(work->arch.token, work->gva);
7522 if (work->wakeup_all)
7523 work->arch.token = ~0; /* broadcast wakeup */
7525 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7527 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7528 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
7529 fault.vector = PF_VECTOR;
7530 fault.error_code_valid = true;
7531 fault.error_code = 0;
7532 fault.nested_page_fault = false;
7533 fault.address = work->arch.token;
7534 kvm_inject_page_fault(vcpu, &fault);
7536 vcpu->arch.apf.halted = false;
7537 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7540 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7542 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7545 return !kvm_event_needs_reinjection(vcpu) &&
7546 kvm_x86_ops->interrupt_allowed(vcpu);
7549 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
7551 atomic_inc(&kvm->arch.noncoherent_dma_count);
7553 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
7555 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
7557 atomic_dec(&kvm->arch.noncoherent_dma_count);
7559 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
7561 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
7563 return atomic_read(&kvm->arch.noncoherent_dma_count);
7565 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
7567 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
7568 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
7569 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
7570 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
7571 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
7572 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
7573 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
7574 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
7575 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
7576 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
7577 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
7578 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
7579 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);