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"
30 #include <linux/clocksource.h>
31 #include <linux/interrupt.h>
32 #include <linux/kvm.h>
34 #include <linux/vmalloc.h>
35 #include <linux/module.h>
36 #include <linux/mman.h>
37 #include <linux/highmem.h>
38 #include <linux/iommu.h>
39 #include <linux/intel-iommu.h>
40 #include <linux/cpufreq.h>
41 #include <linux/user-return-notifier.h>
42 #include <linux/srcu.h>
43 #include <linux/slab.h>
44 #include <linux/perf_event.h>
45 #include <linux/uaccess.h>
46 #include <linux/hash.h>
47 #include <linux/pci.h>
48 #include <trace/events/kvm.h>
50 #define CREATE_TRACE_POINTS
53 #include <asm/debugreg.h>
60 #include <asm/pvclock.h>
61 #include <asm/div64.h>
63 #define MAX_IO_MSRS 256
64 #define KVM_MAX_MCE_BANKS 32
65 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
67 #define emul_to_vcpu(ctxt) \
68 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
71 * - enable syscall per default because its emulated by KVM
72 * - enable LME and LMA per default on 64 bit KVM
76 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
78 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
81 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
82 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
84 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
85 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
86 struct kvm_cpuid_entry2 __user *entries);
87 static void process_nmi(struct kvm_vcpu *vcpu);
89 struct kvm_x86_ops *kvm_x86_ops;
90 EXPORT_SYMBOL_GPL(kvm_x86_ops);
93 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
95 bool kvm_has_tsc_control;
96 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
97 u32 kvm_max_guest_tsc_khz;
98 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
100 #define KVM_NR_SHARED_MSRS 16
102 struct kvm_shared_msrs_global {
104 u32 msrs[KVM_NR_SHARED_MSRS];
107 struct kvm_shared_msrs {
108 struct user_return_notifier urn;
110 struct kvm_shared_msr_values {
113 } values[KVM_NR_SHARED_MSRS];
116 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
117 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
119 struct kvm_stats_debugfs_item debugfs_entries[] = {
120 { "pf_fixed", VCPU_STAT(pf_fixed) },
121 { "pf_guest", VCPU_STAT(pf_guest) },
122 { "tlb_flush", VCPU_STAT(tlb_flush) },
123 { "invlpg", VCPU_STAT(invlpg) },
124 { "exits", VCPU_STAT(exits) },
125 { "io_exits", VCPU_STAT(io_exits) },
126 { "mmio_exits", VCPU_STAT(mmio_exits) },
127 { "signal_exits", VCPU_STAT(signal_exits) },
128 { "irq_window", VCPU_STAT(irq_window_exits) },
129 { "nmi_window", VCPU_STAT(nmi_window_exits) },
130 { "halt_exits", VCPU_STAT(halt_exits) },
131 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
132 { "hypercalls", VCPU_STAT(hypercalls) },
133 { "request_irq", VCPU_STAT(request_irq_exits) },
134 { "irq_exits", VCPU_STAT(irq_exits) },
135 { "host_state_reload", VCPU_STAT(host_state_reload) },
136 { "efer_reload", VCPU_STAT(efer_reload) },
137 { "fpu_reload", VCPU_STAT(fpu_reload) },
138 { "insn_emulation", VCPU_STAT(insn_emulation) },
139 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
140 { "irq_injections", VCPU_STAT(irq_injections) },
141 { "nmi_injections", VCPU_STAT(nmi_injections) },
142 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
143 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
144 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
145 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
146 { "mmu_flooded", VM_STAT(mmu_flooded) },
147 { "mmu_recycled", VM_STAT(mmu_recycled) },
148 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
149 { "mmu_unsync", VM_STAT(mmu_unsync) },
150 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
151 { "largepages", VM_STAT(lpages) },
155 u64 __read_mostly host_xcr0;
157 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
159 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
162 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
163 vcpu->arch.apf.gfns[i] = ~0;
166 static void kvm_on_user_return(struct user_return_notifier *urn)
169 struct kvm_shared_msrs *locals
170 = container_of(urn, struct kvm_shared_msrs, urn);
171 struct kvm_shared_msr_values *values;
173 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
174 values = &locals->values[slot];
175 if (values->host != values->curr) {
176 wrmsrl(shared_msrs_global.msrs[slot], values->host);
177 values->curr = values->host;
180 locals->registered = false;
181 user_return_notifier_unregister(urn);
184 static void shared_msr_update(unsigned slot, u32 msr)
186 struct kvm_shared_msrs *smsr;
189 smsr = &__get_cpu_var(shared_msrs);
190 /* only read, and nobody should modify it at this time,
191 * so don't need lock */
192 if (slot >= shared_msrs_global.nr) {
193 printk(KERN_ERR "kvm: invalid MSR slot!");
196 rdmsrl_safe(msr, &value);
197 smsr->values[slot].host = value;
198 smsr->values[slot].curr = value;
201 void kvm_define_shared_msr(unsigned slot, u32 msr)
203 if (slot >= shared_msrs_global.nr)
204 shared_msrs_global.nr = slot + 1;
205 shared_msrs_global.msrs[slot] = msr;
206 /* we need ensured the shared_msr_global have been updated */
209 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
211 static void kvm_shared_msr_cpu_online(void)
215 for (i = 0; i < shared_msrs_global.nr; ++i)
216 shared_msr_update(i, shared_msrs_global.msrs[i]);
219 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
221 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
223 if (((value ^ smsr->values[slot].curr) & mask) == 0)
225 smsr->values[slot].curr = value;
226 wrmsrl(shared_msrs_global.msrs[slot], value);
227 if (!smsr->registered) {
228 smsr->urn.on_user_return = kvm_on_user_return;
229 user_return_notifier_register(&smsr->urn);
230 smsr->registered = true;
233 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
235 static void drop_user_return_notifiers(void *ignore)
237 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
239 if (smsr->registered)
240 kvm_on_user_return(&smsr->urn);
243 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
245 if (irqchip_in_kernel(vcpu->kvm))
246 return vcpu->arch.apic_base;
248 return vcpu->arch.apic_base;
250 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
252 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
254 /* TODO: reserve bits check */
255 if (irqchip_in_kernel(vcpu->kvm))
256 kvm_lapic_set_base(vcpu, data);
258 vcpu->arch.apic_base = data;
260 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
262 #define EXCPT_BENIGN 0
263 #define EXCPT_CONTRIBUTORY 1
266 static int exception_class(int vector)
276 return EXCPT_CONTRIBUTORY;
283 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
284 unsigned nr, bool has_error, u32 error_code,
290 kvm_make_request(KVM_REQ_EVENT, vcpu);
292 if (!vcpu->arch.exception.pending) {
294 vcpu->arch.exception.pending = true;
295 vcpu->arch.exception.has_error_code = has_error;
296 vcpu->arch.exception.nr = nr;
297 vcpu->arch.exception.error_code = error_code;
298 vcpu->arch.exception.reinject = reinject;
302 /* to check exception */
303 prev_nr = vcpu->arch.exception.nr;
304 if (prev_nr == DF_VECTOR) {
305 /* triple fault -> shutdown */
306 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
309 class1 = exception_class(prev_nr);
310 class2 = exception_class(nr);
311 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
312 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
313 /* generate double fault per SDM Table 5-5 */
314 vcpu->arch.exception.pending = true;
315 vcpu->arch.exception.has_error_code = true;
316 vcpu->arch.exception.nr = DF_VECTOR;
317 vcpu->arch.exception.error_code = 0;
319 /* replace previous exception with a new one in a hope
320 that instruction re-execution will regenerate lost
325 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
327 kvm_multiple_exception(vcpu, nr, false, 0, false);
329 EXPORT_SYMBOL_GPL(kvm_queue_exception);
331 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
333 kvm_multiple_exception(vcpu, nr, false, 0, true);
335 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
337 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
340 kvm_inject_gp(vcpu, 0);
342 kvm_x86_ops->skip_emulated_instruction(vcpu);
344 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
346 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
348 ++vcpu->stat.pf_guest;
349 vcpu->arch.cr2 = fault->address;
350 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
352 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
354 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
356 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
357 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
359 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
362 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
364 atomic_inc(&vcpu->arch.nmi_queued);
365 kvm_make_request(KVM_REQ_NMI, vcpu);
367 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
369 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
371 kvm_multiple_exception(vcpu, nr, true, error_code, false);
373 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
375 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
377 kvm_multiple_exception(vcpu, nr, true, error_code, true);
379 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
382 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
383 * a #GP and return false.
385 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
387 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
389 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
392 EXPORT_SYMBOL_GPL(kvm_require_cpl);
395 * This function will be used to read from the physical memory of the currently
396 * running guest. The difference to kvm_read_guest_page is that this function
397 * can read from guest physical or from the guest's guest physical memory.
399 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
400 gfn_t ngfn, void *data, int offset, int len,
406 ngpa = gfn_to_gpa(ngfn);
407 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
408 if (real_gfn == UNMAPPED_GVA)
411 real_gfn = gpa_to_gfn(real_gfn);
413 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
415 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
417 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
418 void *data, int offset, int len, u32 access)
420 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
421 data, offset, len, access);
425 * Load the pae pdptrs. Return true is they are all valid.
427 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
429 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
430 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
433 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
435 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
436 offset * sizeof(u64), sizeof(pdpte),
437 PFERR_USER_MASK|PFERR_WRITE_MASK);
442 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
443 if (is_present_gpte(pdpte[i]) &&
444 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
451 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
452 __set_bit(VCPU_EXREG_PDPTR,
453 (unsigned long *)&vcpu->arch.regs_avail);
454 __set_bit(VCPU_EXREG_PDPTR,
455 (unsigned long *)&vcpu->arch.regs_dirty);
460 EXPORT_SYMBOL_GPL(load_pdptrs);
462 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
464 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
470 if (is_long_mode(vcpu) || !is_pae(vcpu))
473 if (!test_bit(VCPU_EXREG_PDPTR,
474 (unsigned long *)&vcpu->arch.regs_avail))
477 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
478 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
479 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
480 PFERR_USER_MASK | PFERR_WRITE_MASK);
483 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
489 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
491 unsigned long old_cr0 = kvm_read_cr0(vcpu);
492 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
493 X86_CR0_CD | X86_CR0_NW;
498 if (cr0 & 0xffffffff00000000UL)
502 cr0 &= ~CR0_RESERVED_BITS;
504 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
507 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
510 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
512 if ((vcpu->arch.efer & EFER_LME)) {
517 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
522 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
527 kvm_x86_ops->set_cr0(vcpu, cr0);
529 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
530 kvm_clear_async_pf_completion_queue(vcpu);
531 kvm_async_pf_hash_reset(vcpu);
534 if ((cr0 ^ old_cr0) & update_bits)
535 kvm_mmu_reset_context(vcpu);
538 EXPORT_SYMBOL_GPL(kvm_set_cr0);
540 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
542 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
544 EXPORT_SYMBOL_GPL(kvm_lmsw);
546 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
550 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
551 if (index != XCR_XFEATURE_ENABLED_MASK)
554 if (!(xcr0 & XSTATE_FP))
556 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
558 if (xcr0 & ~host_xcr0)
560 vcpu->arch.xcr0 = xcr0;
561 vcpu->guest_xcr0_loaded = 0;
565 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
567 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
568 __kvm_set_xcr(vcpu, index, xcr)) {
569 kvm_inject_gp(vcpu, 0);
574 EXPORT_SYMBOL_GPL(kvm_set_xcr);
576 static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
578 struct kvm_cpuid_entry2 *best;
580 if (!static_cpu_has(X86_FEATURE_XSAVE))
583 best = kvm_find_cpuid_entry(vcpu, 1, 0);
584 return best && (best->ecx & bit(X86_FEATURE_XSAVE));
587 static bool guest_cpuid_has_smep(struct kvm_vcpu *vcpu)
589 struct kvm_cpuid_entry2 *best;
591 best = kvm_find_cpuid_entry(vcpu, 7, 0);
592 return best && (best->ebx & bit(X86_FEATURE_SMEP));
595 static bool guest_cpuid_has_fsgsbase(struct kvm_vcpu *vcpu)
597 struct kvm_cpuid_entry2 *best;
599 best = kvm_find_cpuid_entry(vcpu, 7, 0);
600 return best && (best->ebx & bit(X86_FEATURE_FSGSBASE));
603 static void update_cpuid(struct kvm_vcpu *vcpu)
605 struct kvm_cpuid_entry2 *best;
606 struct kvm_lapic *apic = vcpu->arch.apic;
608 best = kvm_find_cpuid_entry(vcpu, 1, 0);
612 /* Update OSXSAVE bit */
613 if (cpu_has_xsave && best->function == 0x1) {
614 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
615 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
616 best->ecx |= bit(X86_FEATURE_OSXSAVE);
620 if (best->ecx & bit(X86_FEATURE_TSC_DEADLINE_TIMER))
621 apic->lapic_timer.timer_mode_mask = 3 << 17;
623 apic->lapic_timer.timer_mode_mask = 1 << 17;
627 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
629 unsigned long old_cr4 = kvm_read_cr4(vcpu);
630 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
631 X86_CR4_PAE | X86_CR4_SMEP;
632 if (cr4 & CR4_RESERVED_BITS)
635 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
638 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
641 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_RDWRGSFS))
644 if (is_long_mode(vcpu)) {
645 if (!(cr4 & X86_CR4_PAE))
647 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
648 && ((cr4 ^ old_cr4) & pdptr_bits)
649 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
653 if (kvm_x86_ops->set_cr4(vcpu, cr4))
656 if ((cr4 ^ old_cr4) & pdptr_bits)
657 kvm_mmu_reset_context(vcpu);
659 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
664 EXPORT_SYMBOL_GPL(kvm_set_cr4);
666 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
668 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
669 kvm_mmu_sync_roots(vcpu);
670 kvm_mmu_flush_tlb(vcpu);
674 if (is_long_mode(vcpu)) {
675 if (cr3 & CR3_L_MODE_RESERVED_BITS)
679 if (cr3 & CR3_PAE_RESERVED_BITS)
681 if (is_paging(vcpu) &&
682 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
686 * We don't check reserved bits in nonpae mode, because
687 * this isn't enforced, and VMware depends on this.
692 * Does the new cr3 value map to physical memory? (Note, we
693 * catch an invalid cr3 even in real-mode, because it would
694 * cause trouble later on when we turn on paging anyway.)
696 * A real CPU would silently accept an invalid cr3 and would
697 * attempt to use it - with largely undefined (and often hard
698 * to debug) behavior on the guest side.
700 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
702 vcpu->arch.cr3 = cr3;
703 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
704 vcpu->arch.mmu.new_cr3(vcpu);
707 EXPORT_SYMBOL_GPL(kvm_set_cr3);
709 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
711 if (cr8 & CR8_RESERVED_BITS)
713 if (irqchip_in_kernel(vcpu->kvm))
714 kvm_lapic_set_tpr(vcpu, cr8);
716 vcpu->arch.cr8 = cr8;
719 EXPORT_SYMBOL_GPL(kvm_set_cr8);
721 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
723 if (irqchip_in_kernel(vcpu->kvm))
724 return kvm_lapic_get_cr8(vcpu);
726 return vcpu->arch.cr8;
728 EXPORT_SYMBOL_GPL(kvm_get_cr8);
730 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
734 vcpu->arch.db[dr] = val;
735 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
736 vcpu->arch.eff_db[dr] = val;
739 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
743 if (val & 0xffffffff00000000ULL)
745 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
748 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
752 if (val & 0xffffffff00000000ULL)
754 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
755 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
756 kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
757 vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
765 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
769 res = __kvm_set_dr(vcpu, dr, val);
771 kvm_queue_exception(vcpu, UD_VECTOR);
773 kvm_inject_gp(vcpu, 0);
777 EXPORT_SYMBOL_GPL(kvm_set_dr);
779 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
783 *val = vcpu->arch.db[dr];
786 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
790 *val = vcpu->arch.dr6;
793 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
797 *val = vcpu->arch.dr7;
804 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
806 if (_kvm_get_dr(vcpu, dr, val)) {
807 kvm_queue_exception(vcpu, UD_VECTOR);
812 EXPORT_SYMBOL_GPL(kvm_get_dr);
815 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
816 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
818 * This list is modified at module load time to reflect the
819 * capabilities of the host cpu. This capabilities test skips MSRs that are
820 * kvm-specific. Those are put in the beginning of the list.
823 #define KVM_SAVE_MSRS_BEGIN 9
824 static u32 msrs_to_save[] = {
825 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
826 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
827 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
828 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
829 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
832 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
834 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
837 static unsigned num_msrs_to_save;
839 static u32 emulated_msrs[] = {
840 MSR_IA32_TSCDEADLINE,
841 MSR_IA32_MISC_ENABLE,
846 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
848 u64 old_efer = vcpu->arch.efer;
850 if (efer & efer_reserved_bits)
854 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
857 if (efer & EFER_FFXSR) {
858 struct kvm_cpuid_entry2 *feat;
860 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
861 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
865 if (efer & EFER_SVME) {
866 struct kvm_cpuid_entry2 *feat;
868 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
869 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
874 efer |= vcpu->arch.efer & EFER_LMA;
876 kvm_x86_ops->set_efer(vcpu, efer);
878 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
880 /* Update reserved bits */
881 if ((efer ^ old_efer) & EFER_NX)
882 kvm_mmu_reset_context(vcpu);
887 void kvm_enable_efer_bits(u64 mask)
889 efer_reserved_bits &= ~mask;
891 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
895 * Writes msr value into into the appropriate "register".
896 * Returns 0 on success, non-0 otherwise.
897 * Assumes vcpu_load() was already called.
899 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
901 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
905 * Adapt set_msr() to msr_io()'s calling convention
907 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
909 return kvm_set_msr(vcpu, index, *data);
912 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
916 struct pvclock_wall_clock wc;
917 struct timespec boot;
922 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
927 ++version; /* first time write, random junk */
931 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
934 * The guest calculates current wall clock time by adding
935 * system time (updated by kvm_guest_time_update below) to the
936 * wall clock specified here. guest system time equals host
937 * system time for us, thus we must fill in host boot time here.
941 wc.sec = boot.tv_sec;
942 wc.nsec = boot.tv_nsec;
943 wc.version = version;
945 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
948 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
951 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
953 uint32_t quotient, remainder;
955 /* Don't try to replace with do_div(), this one calculates
956 * "(dividend << 32) / divisor" */
958 : "=a" (quotient), "=d" (remainder)
959 : "0" (0), "1" (dividend), "r" (divisor) );
963 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
964 s8 *pshift, u32 *pmultiplier)
971 tps64 = base_khz * 1000LL;
972 scaled64 = scaled_khz * 1000LL;
973 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
978 tps32 = (uint32_t)tps64;
979 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
980 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
988 *pmultiplier = div_frac(scaled64, tps32);
990 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
991 __func__, base_khz, scaled_khz, shift, *pmultiplier);
994 static inline u64 get_kernel_ns(void)
998 WARN_ON(preemptible());
1000 monotonic_to_bootbased(&ts);
1001 return timespec_to_ns(&ts);
1004 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1005 unsigned long max_tsc_khz;
1007 static inline int kvm_tsc_changes_freq(void)
1009 int cpu = get_cpu();
1010 int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
1011 cpufreq_quick_get(cpu) != 0;
1016 u64 vcpu_tsc_khz(struct kvm_vcpu *vcpu)
1018 if (vcpu->arch.virtual_tsc_khz)
1019 return vcpu->arch.virtual_tsc_khz;
1021 return __this_cpu_read(cpu_tsc_khz);
1024 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1028 WARN_ON(preemptible());
1029 if (kvm_tsc_changes_freq())
1030 printk_once(KERN_WARNING
1031 "kvm: unreliable cycle conversion on adjustable rate TSC\n");
1032 ret = nsec * vcpu_tsc_khz(vcpu);
1033 do_div(ret, USEC_PER_SEC);
1037 static void kvm_init_tsc_catchup(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1039 /* Compute a scale to convert nanoseconds in TSC cycles */
1040 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1041 &vcpu->arch.tsc_catchup_shift,
1042 &vcpu->arch.tsc_catchup_mult);
1045 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1047 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1048 vcpu->arch.tsc_catchup_mult,
1049 vcpu->arch.tsc_catchup_shift);
1050 tsc += vcpu->arch.last_tsc_write;
1054 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1056 struct kvm *kvm = vcpu->kvm;
1057 u64 offset, ns, elapsed;
1058 unsigned long flags;
1061 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1062 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1063 ns = get_kernel_ns();
1064 elapsed = ns - kvm->arch.last_tsc_nsec;
1065 sdiff = data - kvm->arch.last_tsc_write;
1070 * Special case: close write to TSC within 5 seconds of
1071 * another CPU is interpreted as an attempt to synchronize
1072 * The 5 seconds is to accommodate host load / swapping as
1073 * well as any reset of TSC during the boot process.
1075 * In that case, for a reliable TSC, we can match TSC offsets,
1076 * or make a best guest using elapsed value.
1078 if (sdiff < nsec_to_cycles(vcpu, 5ULL * NSEC_PER_SEC) &&
1079 elapsed < 5ULL * NSEC_PER_SEC) {
1080 if (!check_tsc_unstable()) {
1081 offset = kvm->arch.last_tsc_offset;
1082 pr_debug("kvm: matched tsc offset for %llu\n", data);
1084 u64 delta = nsec_to_cycles(vcpu, elapsed);
1086 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1088 ns = kvm->arch.last_tsc_nsec;
1090 kvm->arch.last_tsc_nsec = ns;
1091 kvm->arch.last_tsc_write = data;
1092 kvm->arch.last_tsc_offset = offset;
1093 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1094 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1096 /* Reset of TSC must disable overshoot protection below */
1097 vcpu->arch.hv_clock.tsc_timestamp = 0;
1098 vcpu->arch.last_tsc_write = data;
1099 vcpu->arch.last_tsc_nsec = ns;
1101 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1103 static int kvm_guest_time_update(struct kvm_vcpu *v)
1105 unsigned long flags;
1106 struct kvm_vcpu_arch *vcpu = &v->arch;
1107 unsigned long this_tsc_khz;
1108 s64 kernel_ns, max_kernel_ns;
1111 /* Keep irq disabled to prevent changes to the clock */
1112 local_irq_save(flags);
1113 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v);
1114 kernel_ns = get_kernel_ns();
1115 this_tsc_khz = vcpu_tsc_khz(v);
1116 if (unlikely(this_tsc_khz == 0)) {
1117 local_irq_restore(flags);
1118 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1123 * We may have to catch up the TSC to match elapsed wall clock
1124 * time for two reasons, even if kvmclock is used.
1125 * 1) CPU could have been running below the maximum TSC rate
1126 * 2) Broken TSC compensation resets the base at each VCPU
1127 * entry to avoid unknown leaps of TSC even when running
1128 * again on the same CPU. This may cause apparent elapsed
1129 * time to disappear, and the guest to stand still or run
1132 if (vcpu->tsc_catchup) {
1133 u64 tsc = compute_guest_tsc(v, kernel_ns);
1134 if (tsc > tsc_timestamp) {
1135 kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1136 tsc_timestamp = tsc;
1140 local_irq_restore(flags);
1142 if (!vcpu->pv_time_enabled)
1146 * Time as measured by the TSC may go backwards when resetting the base
1147 * tsc_timestamp. The reason for this is that the TSC resolution is
1148 * higher than the resolution of the other clock scales. Thus, many
1149 * possible measurments of the TSC correspond to one measurement of any
1150 * other clock, and so a spread of values is possible. This is not a
1151 * problem for the computation of the nanosecond clock; with TSC rates
1152 * around 1GHZ, there can only be a few cycles which correspond to one
1153 * nanosecond value, and any path through this code will inevitably
1154 * take longer than that. However, with the kernel_ns value itself,
1155 * the precision may be much lower, down to HZ granularity. If the
1156 * first sampling of TSC against kernel_ns ends in the low part of the
1157 * range, and the second in the high end of the range, we can get:
1159 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1161 * As the sampling errors potentially range in the thousands of cycles,
1162 * it is possible such a time value has already been observed by the
1163 * guest. To protect against this, we must compute the system time as
1164 * observed by the guest and ensure the new system time is greater.
1167 if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1168 max_kernel_ns = vcpu->last_guest_tsc -
1169 vcpu->hv_clock.tsc_timestamp;
1170 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1171 vcpu->hv_clock.tsc_to_system_mul,
1172 vcpu->hv_clock.tsc_shift);
1173 max_kernel_ns += vcpu->last_kernel_ns;
1176 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1177 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1178 &vcpu->hv_clock.tsc_shift,
1179 &vcpu->hv_clock.tsc_to_system_mul);
1180 vcpu->hw_tsc_khz = this_tsc_khz;
1183 if (max_kernel_ns > kernel_ns)
1184 kernel_ns = max_kernel_ns;
1186 /* With all the info we got, fill in the values */
1187 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1188 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1189 vcpu->last_kernel_ns = kernel_ns;
1190 vcpu->last_guest_tsc = tsc_timestamp;
1191 vcpu->hv_clock.flags = 0;
1194 * The interface expects us to write an even number signaling that the
1195 * update is finished. Since the guest won't see the intermediate
1196 * state, we just increase by 2 at the end.
1198 vcpu->hv_clock.version += 2;
1200 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1202 sizeof(vcpu->hv_clock));
1206 static bool msr_mtrr_valid(unsigned msr)
1209 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1210 case MSR_MTRRfix64K_00000:
1211 case MSR_MTRRfix16K_80000:
1212 case MSR_MTRRfix16K_A0000:
1213 case MSR_MTRRfix4K_C0000:
1214 case MSR_MTRRfix4K_C8000:
1215 case MSR_MTRRfix4K_D0000:
1216 case MSR_MTRRfix4K_D8000:
1217 case MSR_MTRRfix4K_E0000:
1218 case MSR_MTRRfix4K_E8000:
1219 case MSR_MTRRfix4K_F0000:
1220 case MSR_MTRRfix4K_F8000:
1221 case MSR_MTRRdefType:
1222 case MSR_IA32_CR_PAT:
1230 static bool valid_pat_type(unsigned t)
1232 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1235 static bool valid_mtrr_type(unsigned t)
1237 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1240 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1244 if (!msr_mtrr_valid(msr))
1247 if (msr == MSR_IA32_CR_PAT) {
1248 for (i = 0; i < 8; i++)
1249 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1252 } else if (msr == MSR_MTRRdefType) {
1255 return valid_mtrr_type(data & 0xff);
1256 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1257 for (i = 0; i < 8 ; i++)
1258 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1263 /* variable MTRRs */
1264 return valid_mtrr_type(data & 0xff);
1267 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1269 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1271 if (!mtrr_valid(vcpu, msr, data))
1274 if (msr == MSR_MTRRdefType) {
1275 vcpu->arch.mtrr_state.def_type = data;
1276 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1277 } else if (msr == MSR_MTRRfix64K_00000)
1279 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1280 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1281 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1282 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1283 else if (msr == MSR_IA32_CR_PAT)
1284 vcpu->arch.pat = data;
1285 else { /* Variable MTRRs */
1286 int idx, is_mtrr_mask;
1289 idx = (msr - 0x200) / 2;
1290 is_mtrr_mask = msr - 0x200 - 2 * idx;
1293 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1296 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1300 kvm_mmu_reset_context(vcpu);
1304 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1306 u64 mcg_cap = vcpu->arch.mcg_cap;
1307 unsigned bank_num = mcg_cap & 0xff;
1310 case MSR_IA32_MCG_STATUS:
1311 vcpu->arch.mcg_status = data;
1313 case MSR_IA32_MCG_CTL:
1314 if (!(mcg_cap & MCG_CTL_P))
1316 if (data != 0 && data != ~(u64)0)
1318 vcpu->arch.mcg_ctl = data;
1321 if (msr >= MSR_IA32_MC0_CTL &&
1322 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1323 u32 offset = msr - MSR_IA32_MC0_CTL;
1324 /* only 0 or all 1s can be written to IA32_MCi_CTL
1325 * some Linux kernels though clear bit 10 in bank 4 to
1326 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1327 * this to avoid an uncatched #GP in the guest
1329 if ((offset & 0x3) == 0 &&
1330 data != 0 && (data | (1 << 10)) != ~(u64)0)
1332 vcpu->arch.mce_banks[offset] = data;
1340 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1342 struct kvm *kvm = vcpu->kvm;
1343 int lm = is_long_mode(vcpu);
1344 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1345 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1346 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1347 : kvm->arch.xen_hvm_config.blob_size_32;
1348 u32 page_num = data & ~PAGE_MASK;
1349 u64 page_addr = data & PAGE_MASK;
1354 if (page_num >= blob_size)
1357 page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1361 if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1363 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1372 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1374 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1377 static bool kvm_hv_msr_partition_wide(u32 msr)
1381 case HV_X64_MSR_GUEST_OS_ID:
1382 case HV_X64_MSR_HYPERCALL:
1390 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1392 struct kvm *kvm = vcpu->kvm;
1395 case HV_X64_MSR_GUEST_OS_ID:
1396 kvm->arch.hv_guest_os_id = data;
1397 /* setting guest os id to zero disables hypercall page */
1398 if (!kvm->arch.hv_guest_os_id)
1399 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1401 case HV_X64_MSR_HYPERCALL: {
1406 /* if guest os id is not set hypercall should remain disabled */
1407 if (!kvm->arch.hv_guest_os_id)
1409 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1410 kvm->arch.hv_hypercall = data;
1413 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1414 addr = gfn_to_hva(kvm, gfn);
1415 if (kvm_is_error_hva(addr))
1417 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1418 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1419 if (__copy_to_user((void __user *)addr, instructions, 4))
1421 kvm->arch.hv_hypercall = data;
1425 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1426 "data 0x%llx\n", msr, data);
1432 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1435 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1438 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1439 vcpu->arch.hv_vapic = data;
1442 addr = gfn_to_hva(vcpu->kvm, data >>
1443 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1444 if (kvm_is_error_hva(addr))
1446 if (__clear_user((void __user *)addr, PAGE_SIZE))
1448 vcpu->arch.hv_vapic = data;
1451 case HV_X64_MSR_EOI:
1452 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1453 case HV_X64_MSR_ICR:
1454 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1455 case HV_X64_MSR_TPR:
1456 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1458 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1459 "data 0x%llx\n", msr, data);
1466 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1468 gpa_t gpa = data & ~0x3f;
1470 /* Bits 2:5 are resrved, Should be zero */
1474 vcpu->arch.apf.msr_val = data;
1476 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1477 kvm_clear_async_pf_completion_queue(vcpu);
1478 kvm_async_pf_hash_reset(vcpu);
1482 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1486 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1487 kvm_async_pf_wakeup_all(vcpu);
1491 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1493 vcpu->arch.pv_time_enabled = false;
1496 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1500 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1503 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1504 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1505 vcpu->arch.st.accum_steal = delta;
1508 static void record_steal_time(struct kvm_vcpu *vcpu)
1510 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1513 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1514 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1517 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1518 vcpu->arch.st.steal.version += 2;
1519 vcpu->arch.st.accum_steal = 0;
1521 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1522 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1525 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1529 return set_efer(vcpu, data);
1531 data &= ~(u64)0x40; /* ignore flush filter disable */
1532 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1534 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1539 case MSR_FAM10H_MMIO_CONF_BASE:
1541 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1546 case MSR_AMD64_NB_CFG:
1548 case MSR_IA32_DEBUGCTLMSR:
1550 /* We support the non-activated case already */
1552 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1553 /* Values other than LBR and BTF are vendor-specific,
1554 thus reserved and should throw a #GP */
1557 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1560 case MSR_IA32_UCODE_REV:
1561 case MSR_IA32_UCODE_WRITE:
1562 case MSR_VM_HSAVE_PA:
1563 case MSR_AMD64_PATCH_LOADER:
1565 case 0x200 ... 0x2ff:
1566 return set_msr_mtrr(vcpu, msr, data);
1567 case MSR_IA32_APICBASE:
1568 kvm_set_apic_base(vcpu, data);
1570 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1571 return kvm_x2apic_msr_write(vcpu, msr, data);
1572 case MSR_IA32_TSCDEADLINE:
1573 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1575 case MSR_IA32_MISC_ENABLE:
1576 vcpu->arch.ia32_misc_enable_msr = data;
1578 case MSR_KVM_WALL_CLOCK_NEW:
1579 case MSR_KVM_WALL_CLOCK:
1580 vcpu->kvm->arch.wall_clock = data;
1581 kvm_write_wall_clock(vcpu->kvm, data);
1583 case MSR_KVM_SYSTEM_TIME_NEW:
1584 case MSR_KVM_SYSTEM_TIME: {
1586 kvmclock_reset(vcpu);
1588 vcpu->arch.time = data;
1589 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1591 /* we verify if the enable bit is set... */
1595 gpa_offset = data & ~(PAGE_MASK | 1);
1597 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
1598 &vcpu->arch.pv_time, data & ~1ULL,
1599 sizeof(struct pvclock_vcpu_time_info)))
1600 vcpu->arch.pv_time_enabled = false;
1602 vcpu->arch.pv_time_enabled = true;
1605 case MSR_KVM_ASYNC_PF_EN:
1606 if (kvm_pv_enable_async_pf(vcpu, data))
1609 case MSR_KVM_STEAL_TIME:
1611 if (unlikely(!sched_info_on()))
1614 if (data & KVM_STEAL_RESERVED_MASK)
1617 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
1618 data & KVM_STEAL_VALID_BITS,
1619 sizeof(struct kvm_steal_time)))
1622 vcpu->arch.st.msr_val = data;
1624 if (!(data & KVM_MSR_ENABLED))
1627 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1630 accumulate_steal_time(vcpu);
1633 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
1637 case MSR_IA32_MCG_CTL:
1638 case MSR_IA32_MCG_STATUS:
1639 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1640 return set_msr_mce(vcpu, msr, data);
1642 /* Performance counters are not protected by a CPUID bit,
1643 * so we should check all of them in the generic path for the sake of
1644 * cross vendor migration.
1645 * Writing a zero into the event select MSRs disables them,
1646 * which we perfectly emulate ;-). Any other value should be at least
1647 * reported, some guests depend on them.
1649 case MSR_P6_EVNTSEL0:
1650 case MSR_P6_EVNTSEL1:
1651 case MSR_K7_EVNTSEL0:
1652 case MSR_K7_EVNTSEL1:
1653 case MSR_K7_EVNTSEL2:
1654 case MSR_K7_EVNTSEL3:
1656 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1657 "0x%x data 0x%llx\n", msr, data);
1659 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1660 * so we ignore writes to make it happy.
1662 case MSR_P6_PERFCTR0:
1663 case MSR_P6_PERFCTR1:
1664 case MSR_K7_PERFCTR0:
1665 case MSR_K7_PERFCTR1:
1666 case MSR_K7_PERFCTR2:
1667 case MSR_K7_PERFCTR3:
1668 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1669 "0x%x data 0x%llx\n", msr, data);
1671 case MSR_K7_CLK_CTL:
1673 * Ignore all writes to this no longer documented MSR.
1674 * Writes are only relevant for old K7 processors,
1675 * all pre-dating SVM, but a recommended workaround from
1676 * AMD for these chips. It is possible to speicify the
1677 * affected processor models on the command line, hence
1678 * the need to ignore the workaround.
1681 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1682 if (kvm_hv_msr_partition_wide(msr)) {
1684 mutex_lock(&vcpu->kvm->lock);
1685 r = set_msr_hyperv_pw(vcpu, msr, data);
1686 mutex_unlock(&vcpu->kvm->lock);
1689 return set_msr_hyperv(vcpu, msr, data);
1691 case MSR_IA32_BBL_CR_CTL3:
1692 /* Drop writes to this legacy MSR -- see rdmsr
1693 * counterpart for further detail.
1695 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1698 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1699 return xen_hvm_config(vcpu, data);
1701 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1705 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1712 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1716 * Reads an msr value (of 'msr_index') into 'pdata'.
1717 * Returns 0 on success, non-0 otherwise.
1718 * Assumes vcpu_load() was already called.
1720 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1722 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1725 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1727 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1729 if (!msr_mtrr_valid(msr))
1732 if (msr == MSR_MTRRdefType)
1733 *pdata = vcpu->arch.mtrr_state.def_type +
1734 (vcpu->arch.mtrr_state.enabled << 10);
1735 else if (msr == MSR_MTRRfix64K_00000)
1737 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1738 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1739 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1740 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1741 else if (msr == MSR_IA32_CR_PAT)
1742 *pdata = vcpu->arch.pat;
1743 else { /* Variable MTRRs */
1744 int idx, is_mtrr_mask;
1747 idx = (msr - 0x200) / 2;
1748 is_mtrr_mask = msr - 0x200 - 2 * idx;
1751 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1754 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1761 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1764 u64 mcg_cap = vcpu->arch.mcg_cap;
1765 unsigned bank_num = mcg_cap & 0xff;
1768 case MSR_IA32_P5_MC_ADDR:
1769 case MSR_IA32_P5_MC_TYPE:
1772 case MSR_IA32_MCG_CAP:
1773 data = vcpu->arch.mcg_cap;
1775 case MSR_IA32_MCG_CTL:
1776 if (!(mcg_cap & MCG_CTL_P))
1778 data = vcpu->arch.mcg_ctl;
1780 case MSR_IA32_MCG_STATUS:
1781 data = vcpu->arch.mcg_status;
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 data = vcpu->arch.mce_banks[offset];
1796 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1799 struct kvm *kvm = vcpu->kvm;
1802 case HV_X64_MSR_GUEST_OS_ID:
1803 data = kvm->arch.hv_guest_os_id;
1805 case HV_X64_MSR_HYPERCALL:
1806 data = kvm->arch.hv_hypercall;
1809 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1817 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1822 case HV_X64_MSR_VP_INDEX: {
1825 kvm_for_each_vcpu(r, v, vcpu->kvm)
1830 case HV_X64_MSR_EOI:
1831 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1832 case HV_X64_MSR_ICR:
1833 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1834 case HV_X64_MSR_TPR:
1835 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1836 case HV_X64_MSR_APIC_ASSIST_PAGE:
1837 data = vcpu->arch.hv_vapic;
1840 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1847 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1852 case MSR_IA32_PLATFORM_ID:
1853 case MSR_IA32_EBL_CR_POWERON:
1854 case MSR_IA32_DEBUGCTLMSR:
1855 case MSR_IA32_LASTBRANCHFROMIP:
1856 case MSR_IA32_LASTBRANCHTOIP:
1857 case MSR_IA32_LASTINTFROMIP:
1858 case MSR_IA32_LASTINTTOIP:
1861 case MSR_VM_HSAVE_PA:
1862 case MSR_P6_PERFCTR0:
1863 case MSR_P6_PERFCTR1:
1864 case MSR_P6_EVNTSEL0:
1865 case MSR_P6_EVNTSEL1:
1866 case MSR_K7_EVNTSEL0:
1867 case MSR_K7_PERFCTR0:
1868 case MSR_K8_INT_PENDING_MSG:
1869 case MSR_AMD64_NB_CFG:
1870 case MSR_FAM10H_MMIO_CONF_BASE:
1873 case MSR_IA32_UCODE_REV:
1874 data = 0x100000000ULL;
1877 data = 0x500 | KVM_NR_VAR_MTRR;
1879 case 0x200 ... 0x2ff:
1880 return get_msr_mtrr(vcpu, msr, pdata);
1881 case 0xcd: /* fsb frequency */
1885 * MSR_EBC_FREQUENCY_ID
1886 * Conservative value valid for even the basic CPU models.
1887 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1888 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1889 * and 266MHz for model 3, or 4. Set Core Clock
1890 * Frequency to System Bus Frequency Ratio to 1 (bits
1891 * 31:24) even though these are only valid for CPU
1892 * models > 2, however guests may end up dividing or
1893 * multiplying by zero otherwise.
1895 case MSR_EBC_FREQUENCY_ID:
1898 case MSR_IA32_APICBASE:
1899 data = kvm_get_apic_base(vcpu);
1901 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1902 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1904 case MSR_IA32_TSCDEADLINE:
1905 data = kvm_get_lapic_tscdeadline_msr(vcpu);
1907 case MSR_IA32_MISC_ENABLE:
1908 data = vcpu->arch.ia32_misc_enable_msr;
1910 case MSR_IA32_PERF_STATUS:
1911 /* TSC increment by tick */
1913 /* CPU multiplier */
1914 data |= (((uint64_t)4ULL) << 40);
1917 data = vcpu->arch.efer;
1919 case MSR_KVM_WALL_CLOCK:
1920 case MSR_KVM_WALL_CLOCK_NEW:
1921 data = vcpu->kvm->arch.wall_clock;
1923 case MSR_KVM_SYSTEM_TIME:
1924 case MSR_KVM_SYSTEM_TIME_NEW:
1925 data = vcpu->arch.time;
1927 case MSR_KVM_ASYNC_PF_EN:
1928 data = vcpu->arch.apf.msr_val;
1930 case MSR_KVM_STEAL_TIME:
1931 data = vcpu->arch.st.msr_val;
1933 case MSR_IA32_P5_MC_ADDR:
1934 case MSR_IA32_P5_MC_TYPE:
1935 case MSR_IA32_MCG_CAP:
1936 case MSR_IA32_MCG_CTL:
1937 case MSR_IA32_MCG_STATUS:
1938 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1939 return get_msr_mce(vcpu, msr, pdata);
1940 case MSR_K7_CLK_CTL:
1942 * Provide expected ramp-up count for K7. All other
1943 * are set to zero, indicating minimum divisors for
1946 * This prevents guest kernels on AMD host with CPU
1947 * type 6, model 8 and higher from exploding due to
1948 * the rdmsr failing.
1952 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1953 if (kvm_hv_msr_partition_wide(msr)) {
1955 mutex_lock(&vcpu->kvm->lock);
1956 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1957 mutex_unlock(&vcpu->kvm->lock);
1960 return get_msr_hyperv(vcpu, msr, pdata);
1962 case MSR_IA32_BBL_CR_CTL3:
1963 /* This legacy MSR exists but isn't fully documented in current
1964 * silicon. It is however accessed by winxp in very narrow
1965 * scenarios where it sets bit #19, itself documented as
1966 * a "reserved" bit. Best effort attempt to source coherent
1967 * read data here should the balance of the register be
1968 * interpreted by the guest:
1970 * L2 cache control register 3: 64GB range, 256KB size,
1971 * enabled, latency 0x1, configured
1977 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1980 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1988 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1991 * Read or write a bunch of msrs. All parameters are kernel addresses.
1993 * @return number of msrs set successfully.
1995 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1996 struct kvm_msr_entry *entries,
1997 int (*do_msr)(struct kvm_vcpu *vcpu,
1998 unsigned index, u64 *data))
2002 idx = srcu_read_lock(&vcpu->kvm->srcu);
2003 for (i = 0; i < msrs->nmsrs; ++i)
2004 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2006 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2012 * Read or write a bunch of msrs. Parameters are user addresses.
2014 * @return number of msrs set successfully.
2016 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2017 int (*do_msr)(struct kvm_vcpu *vcpu,
2018 unsigned index, u64 *data),
2021 struct kvm_msrs msrs;
2022 struct kvm_msr_entry *entries;
2027 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2031 if (msrs.nmsrs >= MAX_IO_MSRS)
2035 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2036 entries = kmalloc(size, GFP_KERNEL);
2041 if (copy_from_user(entries, user_msrs->entries, size))
2044 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2049 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2060 int kvm_dev_ioctl_check_extension(long ext)
2065 case KVM_CAP_IRQCHIP:
2067 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2068 case KVM_CAP_SET_TSS_ADDR:
2069 case KVM_CAP_EXT_CPUID:
2070 case KVM_CAP_CLOCKSOURCE:
2072 case KVM_CAP_NOP_IO_DELAY:
2073 case KVM_CAP_MP_STATE:
2074 case KVM_CAP_SYNC_MMU:
2075 case KVM_CAP_USER_NMI:
2076 case KVM_CAP_REINJECT_CONTROL:
2077 case KVM_CAP_IRQ_INJECT_STATUS:
2078 case KVM_CAP_ASSIGN_DEV_IRQ:
2080 case KVM_CAP_IOEVENTFD:
2082 case KVM_CAP_PIT_STATE2:
2083 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2084 case KVM_CAP_XEN_HVM:
2085 case KVM_CAP_ADJUST_CLOCK:
2086 case KVM_CAP_VCPU_EVENTS:
2087 case KVM_CAP_HYPERV:
2088 case KVM_CAP_HYPERV_VAPIC:
2089 case KVM_CAP_HYPERV_SPIN:
2090 case KVM_CAP_PCI_SEGMENT:
2091 case KVM_CAP_DEBUGREGS:
2092 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2094 case KVM_CAP_ASYNC_PF:
2095 case KVM_CAP_GET_TSC_KHZ:
2098 case KVM_CAP_COALESCED_MMIO:
2099 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2102 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2104 case KVM_CAP_NR_VCPUS:
2105 r = KVM_SOFT_MAX_VCPUS;
2107 case KVM_CAP_MAX_VCPUS:
2110 case KVM_CAP_NR_MEMSLOTS:
2111 r = KVM_MEMORY_SLOTS;
2113 case KVM_CAP_PV_MMU: /* obsolete */
2117 r = iommu_present(&pci_bus_type);
2120 r = KVM_MAX_MCE_BANKS;
2125 case KVM_CAP_TSC_CONTROL:
2126 r = kvm_has_tsc_control;
2128 case KVM_CAP_TSC_DEADLINE_TIMER:
2129 r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2139 long kvm_arch_dev_ioctl(struct file *filp,
2140 unsigned int ioctl, unsigned long arg)
2142 void __user *argp = (void __user *)arg;
2146 case KVM_GET_MSR_INDEX_LIST: {
2147 struct kvm_msr_list __user *user_msr_list = argp;
2148 struct kvm_msr_list msr_list;
2152 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2155 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2156 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2159 if (n < msr_list.nmsrs)
2162 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2163 num_msrs_to_save * sizeof(u32)))
2165 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2167 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2172 case KVM_GET_SUPPORTED_CPUID: {
2173 struct kvm_cpuid2 __user *cpuid_arg = argp;
2174 struct kvm_cpuid2 cpuid;
2177 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2179 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2180 cpuid_arg->entries);
2185 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2190 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2193 mce_cap = KVM_MCE_CAP_SUPPORTED;
2195 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2207 static void wbinvd_ipi(void *garbage)
2212 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2214 return vcpu->kvm->arch.iommu_domain &&
2215 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2218 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2220 /* Address WBINVD may be executed by guest */
2221 if (need_emulate_wbinvd(vcpu)) {
2222 if (kvm_x86_ops->has_wbinvd_exit())
2223 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2224 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2225 smp_call_function_single(vcpu->cpu,
2226 wbinvd_ipi, NULL, 1);
2229 kvm_x86_ops->vcpu_load(vcpu, cpu);
2230 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2231 /* Make sure TSC doesn't go backwards */
2235 tsc = kvm_x86_ops->read_l1_tsc(vcpu);
2236 tsc_delta = !vcpu->arch.last_guest_tsc ? 0 :
2237 tsc - vcpu->arch.last_guest_tsc;
2240 mark_tsc_unstable("KVM discovered backwards TSC");
2241 if (check_tsc_unstable()) {
2242 kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2243 vcpu->arch.tsc_catchup = 1;
2245 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2246 if (vcpu->cpu != cpu)
2247 kvm_migrate_timers(vcpu);
2251 accumulate_steal_time(vcpu);
2252 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2255 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2257 kvm_x86_ops->vcpu_put(vcpu);
2258 kvm_put_guest_fpu(vcpu);
2259 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu);
2262 static int is_efer_nx(void)
2264 unsigned long long efer = 0;
2266 rdmsrl_safe(MSR_EFER, &efer);
2267 return efer & EFER_NX;
2270 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2273 struct kvm_cpuid_entry2 *e, *entry;
2276 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2277 e = &vcpu->arch.cpuid_entries[i];
2278 if (e->function == 0x80000001) {
2283 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2284 entry->edx &= ~(1 << 20);
2285 printk(KERN_INFO "kvm: guest NX capability removed\n");
2289 /* when an old userspace process fills a new kernel module */
2290 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2291 struct kvm_cpuid *cpuid,
2292 struct kvm_cpuid_entry __user *entries)
2295 struct kvm_cpuid_entry *cpuid_entries;
2298 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2301 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2305 if (copy_from_user(cpuid_entries, entries,
2306 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2308 for (i = 0; i < cpuid->nent; i++) {
2309 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2310 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2311 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2312 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2313 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2314 vcpu->arch.cpuid_entries[i].index = 0;
2315 vcpu->arch.cpuid_entries[i].flags = 0;
2316 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2317 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2318 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2320 vcpu->arch.cpuid_nent = cpuid->nent;
2321 cpuid_fix_nx_cap(vcpu);
2323 kvm_apic_set_version(vcpu);
2324 kvm_x86_ops->cpuid_update(vcpu);
2328 vfree(cpuid_entries);
2333 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2334 struct kvm_cpuid2 *cpuid,
2335 struct kvm_cpuid_entry2 __user *entries)
2340 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2343 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2344 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2346 vcpu->arch.cpuid_nent = cpuid->nent;
2347 kvm_apic_set_version(vcpu);
2348 kvm_x86_ops->cpuid_update(vcpu);
2356 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2357 struct kvm_cpuid2 *cpuid,
2358 struct kvm_cpuid_entry2 __user *entries)
2363 if (cpuid->nent < vcpu->arch.cpuid_nent)
2366 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2367 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2372 cpuid->nent = vcpu->arch.cpuid_nent;
2376 static void cpuid_mask(u32 *word, int wordnum)
2378 *word &= boot_cpu_data.x86_capability[wordnum];
2381 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2384 entry->function = function;
2385 entry->index = index;
2386 cpuid_count(entry->function, entry->index,
2387 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2391 static bool supported_xcr0_bit(unsigned bit)
2393 u64 mask = ((u64)1 << bit);
2395 return mask & (XSTATE_FP | XSTATE_SSE | XSTATE_YMM) & host_xcr0;
2398 #define F(x) bit(X86_FEATURE_##x)
2400 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2401 u32 index, int *nent, int maxnent)
2403 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2404 #ifdef CONFIG_X86_64
2405 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2407 unsigned f_lm = F(LM);
2409 unsigned f_gbpages = 0;
2412 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2415 const u32 kvm_supported_word0_x86_features =
2416 F(FPU) | F(VME) | F(DE) | F(PSE) |
2417 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2418 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2419 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2420 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2421 0 /* Reserved, DS, ACPI */ | F(MMX) |
2422 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2423 0 /* HTT, TM, Reserved, PBE */;
2424 /* cpuid 0x80000001.edx */
2425 const u32 kvm_supported_word1_x86_features =
2426 F(FPU) | F(VME) | F(DE) | F(PSE) |
2427 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2428 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2429 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2430 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2431 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2432 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2433 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2435 const u32 kvm_supported_word4_x86_features =
2436 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
2437 0 /* DS-CPL, VMX, SMX, EST */ |
2438 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2439 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2440 0 /* Reserved, DCA */ | F(XMM4_1) |
2441 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2442 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2443 F(F16C) | F(RDRAND);
2444 /* cpuid 0x80000001.ecx */
2445 const u32 kvm_supported_word6_x86_features =
2446 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2447 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2448 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2449 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2451 /* cpuid 0xC0000001.edx */
2452 const u32 kvm_supported_word5_x86_features =
2453 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
2454 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
2458 const u32 kvm_supported_word9_x86_features =
2459 F(SMEP) | F(FSGSBASE) | F(ERMS);
2461 /* all calls to cpuid_count() should be made on the same cpu */
2463 do_cpuid_1_ent(entry, function, index);
2468 entry->eax = min(entry->eax, (u32)0xd);
2471 entry->edx &= kvm_supported_word0_x86_features;
2472 cpuid_mask(&entry->edx, 0);
2473 entry->ecx &= kvm_supported_word4_x86_features;
2474 cpuid_mask(&entry->ecx, 4);
2475 /* we support x2apic emulation even if host does not support
2476 * it since we emulate x2apic in software */
2477 entry->ecx |= F(X2APIC);
2479 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2480 * may return different values. This forces us to get_cpu() before
2481 * issuing the first command, and also to emulate this annoying behavior
2482 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2484 int t, times = entry->eax & 0xff;
2486 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2487 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2488 for (t = 1; t < times && *nent < maxnent; ++t) {
2489 do_cpuid_1_ent(&entry[t], function, 0);
2490 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2495 /* function 4 has additional index. */
2499 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2500 /* read more entries until cache_type is zero */
2501 for (i = 1; *nent < maxnent; ++i) {
2502 cache_type = entry[i - 1].eax & 0x1f;
2505 do_cpuid_1_ent(&entry[i], function, i);
2507 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2513 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2514 /* Mask ebx against host capbability word 9 */
2516 entry->ebx &= kvm_supported_word9_x86_features;
2517 cpuid_mask(&entry->ebx, 9);
2527 /* function 0xb has additional index. */
2531 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2532 /* read more entries until level_type is zero */
2533 for (i = 1; *nent < maxnent; ++i) {
2534 level_type = entry[i - 1].ecx & 0xff00;
2537 do_cpuid_1_ent(&entry[i], function, i);
2539 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2547 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2548 for (idx = 1, i = 1; *nent < maxnent && idx < 64; ++idx) {
2549 do_cpuid_1_ent(&entry[i], function, idx);
2550 if (entry[i].eax == 0 || !supported_xcr0_bit(idx))
2553 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2559 case KVM_CPUID_SIGNATURE: {
2560 char signature[12] = "KVMKVMKVM\0\0";
2561 u32 *sigptr = (u32 *)signature;
2563 entry->ebx = sigptr[0];
2564 entry->ecx = sigptr[1];
2565 entry->edx = sigptr[2];
2568 case KVM_CPUID_FEATURES:
2569 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2570 (1 << KVM_FEATURE_NOP_IO_DELAY) |
2571 (1 << KVM_FEATURE_CLOCKSOURCE2) |
2572 (1 << KVM_FEATURE_ASYNC_PF) |
2573 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2575 if (sched_info_on())
2576 entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
2583 entry->eax = min(entry->eax, 0x8000001a);
2586 entry->edx &= kvm_supported_word1_x86_features;
2587 cpuid_mask(&entry->edx, 1);
2588 entry->ecx &= kvm_supported_word6_x86_features;
2589 cpuid_mask(&entry->ecx, 6);
2592 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
2593 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
2594 unsigned phys_as = entry->eax & 0xff;
2597 g_phys_as = phys_as;
2598 entry->eax = g_phys_as | (virt_as << 8);
2599 entry->ebx = entry->edx = 0;
2603 entry->ecx = entry->edx = 0;
2609 /*Add support for Centaur's CPUID instruction*/
2611 /*Just support up to 0xC0000004 now*/
2612 entry->eax = min(entry->eax, 0xC0000004);
2615 entry->edx &= kvm_supported_word5_x86_features;
2616 cpuid_mask(&entry->edx, 5);
2618 case 3: /* Processor serial number */
2619 case 5: /* MONITOR/MWAIT */
2620 case 6: /* Thermal management */
2621 case 0xA: /* Architectural Performance Monitoring */
2622 case 0x80000007: /* Advanced power management */
2627 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
2631 kvm_x86_ops->set_supported_cpuid(function, entry);
2638 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2639 struct kvm_cpuid_entry2 __user *entries)
2641 struct kvm_cpuid_entry2 *cpuid_entries;
2642 int limit, nent = 0, r = -E2BIG;
2645 if (cpuid->nent < 1)
2647 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2648 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2650 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2654 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2655 limit = cpuid_entries[0].eax;
2656 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2657 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2658 &nent, cpuid->nent);
2660 if (nent >= cpuid->nent)
2663 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2664 limit = cpuid_entries[nent - 1].eax;
2665 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2666 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2667 &nent, cpuid->nent);
2672 if (nent >= cpuid->nent)
2675 /* Add support for Centaur's CPUID instruction. */
2676 if (boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR) {
2677 do_cpuid_ent(&cpuid_entries[nent], 0xC0000000, 0,
2678 &nent, cpuid->nent);
2681 if (nent >= cpuid->nent)
2684 limit = cpuid_entries[nent - 1].eax;
2685 for (func = 0xC0000001;
2686 func <= limit && nent < cpuid->nent; ++func)
2687 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2688 &nent, cpuid->nent);
2691 if (nent >= cpuid->nent)
2695 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2699 if (nent >= cpuid->nent)
2702 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2706 if (nent >= cpuid->nent)
2710 if (copy_to_user(entries, cpuid_entries,
2711 nent * sizeof(struct kvm_cpuid_entry2)))
2717 vfree(cpuid_entries);
2722 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2723 struct kvm_lapic_state *s)
2725 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2730 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2731 struct kvm_lapic_state *s)
2733 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2734 kvm_apic_post_state_restore(vcpu);
2735 update_cr8_intercept(vcpu);
2740 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2741 struct kvm_interrupt *irq)
2743 if (irq->irq < 0 || irq->irq >= 256)
2745 if (irqchip_in_kernel(vcpu->kvm))
2748 kvm_queue_interrupt(vcpu, irq->irq, false);
2749 kvm_make_request(KVM_REQ_EVENT, vcpu);
2754 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2756 kvm_inject_nmi(vcpu);
2761 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2762 struct kvm_tpr_access_ctl *tac)
2766 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2770 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2774 unsigned bank_num = mcg_cap & 0xff, bank;
2777 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2779 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2782 vcpu->arch.mcg_cap = mcg_cap;
2783 /* Init IA32_MCG_CTL to all 1s */
2784 if (mcg_cap & MCG_CTL_P)
2785 vcpu->arch.mcg_ctl = ~(u64)0;
2786 /* Init IA32_MCi_CTL to all 1s */
2787 for (bank = 0; bank < bank_num; bank++)
2788 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2793 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2794 struct kvm_x86_mce *mce)
2796 u64 mcg_cap = vcpu->arch.mcg_cap;
2797 unsigned bank_num = mcg_cap & 0xff;
2798 u64 *banks = vcpu->arch.mce_banks;
2800 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2803 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2804 * reporting is disabled
2806 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2807 vcpu->arch.mcg_ctl != ~(u64)0)
2809 banks += 4 * mce->bank;
2811 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2812 * reporting is disabled for the bank
2814 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2816 if (mce->status & MCI_STATUS_UC) {
2817 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2818 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2819 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2822 if (banks[1] & MCI_STATUS_VAL)
2823 mce->status |= MCI_STATUS_OVER;
2824 banks[2] = mce->addr;
2825 banks[3] = mce->misc;
2826 vcpu->arch.mcg_status = mce->mcg_status;
2827 banks[1] = mce->status;
2828 kvm_queue_exception(vcpu, MC_VECTOR);
2829 } else if (!(banks[1] & MCI_STATUS_VAL)
2830 || !(banks[1] & MCI_STATUS_UC)) {
2831 if (banks[1] & MCI_STATUS_VAL)
2832 mce->status |= MCI_STATUS_OVER;
2833 banks[2] = mce->addr;
2834 banks[3] = mce->misc;
2835 banks[1] = mce->status;
2837 banks[1] |= MCI_STATUS_OVER;
2841 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2842 struct kvm_vcpu_events *events)
2845 events->exception.injected =
2846 vcpu->arch.exception.pending &&
2847 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2848 events->exception.nr = vcpu->arch.exception.nr;
2849 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2850 events->exception.pad = 0;
2851 events->exception.error_code = vcpu->arch.exception.error_code;
2853 events->interrupt.injected =
2854 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2855 events->interrupt.nr = vcpu->arch.interrupt.nr;
2856 events->interrupt.soft = 0;
2857 events->interrupt.shadow =
2858 kvm_x86_ops->get_interrupt_shadow(vcpu,
2859 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2861 events->nmi.injected = vcpu->arch.nmi_injected;
2862 events->nmi.pending = vcpu->arch.nmi_pending != 0;
2863 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2864 events->nmi.pad = 0;
2866 events->sipi_vector = vcpu->arch.sipi_vector;
2868 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2869 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2870 | KVM_VCPUEVENT_VALID_SHADOW);
2871 memset(&events->reserved, 0, sizeof(events->reserved));
2874 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2875 struct kvm_vcpu_events *events)
2877 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2878 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2879 | KVM_VCPUEVENT_VALID_SHADOW))
2883 vcpu->arch.exception.pending = events->exception.injected;
2884 vcpu->arch.exception.nr = events->exception.nr;
2885 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2886 vcpu->arch.exception.error_code = events->exception.error_code;
2888 vcpu->arch.interrupt.pending = events->interrupt.injected;
2889 vcpu->arch.interrupt.nr = events->interrupt.nr;
2890 vcpu->arch.interrupt.soft = events->interrupt.soft;
2891 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2892 kvm_x86_ops->set_interrupt_shadow(vcpu,
2893 events->interrupt.shadow);
2895 vcpu->arch.nmi_injected = events->nmi.injected;
2896 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2897 vcpu->arch.nmi_pending = events->nmi.pending;
2898 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2900 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2901 vcpu->arch.sipi_vector = events->sipi_vector;
2903 kvm_make_request(KVM_REQ_EVENT, vcpu);
2908 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2909 struct kvm_debugregs *dbgregs)
2911 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2912 dbgregs->dr6 = vcpu->arch.dr6;
2913 dbgregs->dr7 = vcpu->arch.dr7;
2915 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2918 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2919 struct kvm_debugregs *dbgregs)
2924 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2925 vcpu->arch.dr6 = dbgregs->dr6;
2926 vcpu->arch.dr7 = dbgregs->dr7;
2931 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2932 struct kvm_xsave *guest_xsave)
2935 memcpy(guest_xsave->region,
2936 &vcpu->arch.guest_fpu.state->xsave,
2939 memcpy(guest_xsave->region,
2940 &vcpu->arch.guest_fpu.state->fxsave,
2941 sizeof(struct i387_fxsave_struct));
2942 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2947 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2948 struct kvm_xsave *guest_xsave)
2951 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2954 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2955 guest_xsave->region, xstate_size);
2957 if (xstate_bv & ~XSTATE_FPSSE)
2959 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2960 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2965 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2966 struct kvm_xcrs *guest_xcrs)
2968 if (!cpu_has_xsave) {
2969 guest_xcrs->nr_xcrs = 0;
2973 guest_xcrs->nr_xcrs = 1;
2974 guest_xcrs->flags = 0;
2975 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2976 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2979 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2980 struct kvm_xcrs *guest_xcrs)
2987 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2990 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2991 /* Only support XCR0 currently */
2992 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2993 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2994 guest_xcrs->xcrs[0].value);
3002 long kvm_arch_vcpu_ioctl(struct file *filp,
3003 unsigned int ioctl, unsigned long arg)
3005 struct kvm_vcpu *vcpu = filp->private_data;
3006 void __user *argp = (void __user *)arg;
3009 struct kvm_lapic_state *lapic;
3010 struct kvm_xsave *xsave;
3011 struct kvm_xcrs *xcrs;
3017 case KVM_GET_LAPIC: {
3019 if (!vcpu->arch.apic)
3021 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3026 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3030 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3035 case KVM_SET_LAPIC: {
3037 if (!vcpu->arch.apic)
3039 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3044 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
3046 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3052 case KVM_INTERRUPT: {
3053 struct kvm_interrupt irq;
3056 if (copy_from_user(&irq, argp, sizeof irq))
3058 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3065 r = kvm_vcpu_ioctl_nmi(vcpu);
3071 case KVM_SET_CPUID: {
3072 struct kvm_cpuid __user *cpuid_arg = argp;
3073 struct kvm_cpuid cpuid;
3076 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3078 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3083 case KVM_SET_CPUID2: {
3084 struct kvm_cpuid2 __user *cpuid_arg = argp;
3085 struct kvm_cpuid2 cpuid;
3088 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3090 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3091 cpuid_arg->entries);
3096 case KVM_GET_CPUID2: {
3097 struct kvm_cpuid2 __user *cpuid_arg = argp;
3098 struct kvm_cpuid2 cpuid;
3101 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3103 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3104 cpuid_arg->entries);
3108 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3114 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3117 r = msr_io(vcpu, argp, do_set_msr, 0);
3119 case KVM_TPR_ACCESS_REPORTING: {
3120 struct kvm_tpr_access_ctl tac;
3123 if (copy_from_user(&tac, argp, sizeof tac))
3125 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3129 if (copy_to_user(argp, &tac, sizeof tac))
3134 case KVM_SET_VAPIC_ADDR: {
3135 struct kvm_vapic_addr va;
3138 if (!irqchip_in_kernel(vcpu->kvm))
3141 if (copy_from_user(&va, argp, sizeof va))
3144 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3147 case KVM_X86_SETUP_MCE: {
3151 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3153 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3156 case KVM_X86_SET_MCE: {
3157 struct kvm_x86_mce mce;
3160 if (copy_from_user(&mce, argp, sizeof mce))
3162 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3165 case KVM_GET_VCPU_EVENTS: {
3166 struct kvm_vcpu_events events;
3168 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3171 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3176 case KVM_SET_VCPU_EVENTS: {
3177 struct kvm_vcpu_events events;
3180 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3183 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3186 case KVM_GET_DEBUGREGS: {
3187 struct kvm_debugregs dbgregs;
3189 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3192 if (copy_to_user(argp, &dbgregs,
3193 sizeof(struct kvm_debugregs)))
3198 case KVM_SET_DEBUGREGS: {
3199 struct kvm_debugregs dbgregs;
3202 if (copy_from_user(&dbgregs, argp,
3203 sizeof(struct kvm_debugregs)))
3206 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3209 case KVM_GET_XSAVE: {
3210 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3215 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3218 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3223 case KVM_SET_XSAVE: {
3224 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3230 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
3233 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3236 case KVM_GET_XCRS: {
3237 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3242 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3245 if (copy_to_user(argp, u.xcrs,
3246 sizeof(struct kvm_xcrs)))
3251 case KVM_SET_XCRS: {
3252 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3258 if (copy_from_user(u.xcrs, argp,
3259 sizeof(struct kvm_xcrs)))
3262 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3265 case KVM_SET_TSC_KHZ: {
3269 if (!kvm_has_tsc_control)
3272 user_tsc_khz = (u32)arg;
3274 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3277 kvm_x86_ops->set_tsc_khz(vcpu, user_tsc_khz);
3282 case KVM_GET_TSC_KHZ: {
3284 if (check_tsc_unstable())
3287 r = vcpu_tsc_khz(vcpu);
3299 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3303 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3305 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3309 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3312 kvm->arch.ept_identity_map_addr = ident_addr;
3316 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3317 u32 kvm_nr_mmu_pages)
3319 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3322 mutex_lock(&kvm->slots_lock);
3323 spin_lock(&kvm->mmu_lock);
3325 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3326 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3328 spin_unlock(&kvm->mmu_lock);
3329 mutex_unlock(&kvm->slots_lock);
3333 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3335 return kvm->arch.n_max_mmu_pages;
3338 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3343 switch (chip->chip_id) {
3344 case KVM_IRQCHIP_PIC_MASTER:
3345 memcpy(&chip->chip.pic,
3346 &pic_irqchip(kvm)->pics[0],
3347 sizeof(struct kvm_pic_state));
3349 case KVM_IRQCHIP_PIC_SLAVE:
3350 memcpy(&chip->chip.pic,
3351 &pic_irqchip(kvm)->pics[1],
3352 sizeof(struct kvm_pic_state));
3354 case KVM_IRQCHIP_IOAPIC:
3355 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3364 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3369 switch (chip->chip_id) {
3370 case KVM_IRQCHIP_PIC_MASTER:
3371 spin_lock(&pic_irqchip(kvm)->lock);
3372 memcpy(&pic_irqchip(kvm)->pics[0],
3374 sizeof(struct kvm_pic_state));
3375 spin_unlock(&pic_irqchip(kvm)->lock);
3377 case KVM_IRQCHIP_PIC_SLAVE:
3378 spin_lock(&pic_irqchip(kvm)->lock);
3379 memcpy(&pic_irqchip(kvm)->pics[1],
3381 sizeof(struct kvm_pic_state));
3382 spin_unlock(&pic_irqchip(kvm)->lock);
3384 case KVM_IRQCHIP_IOAPIC:
3385 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3391 kvm_pic_update_irq(pic_irqchip(kvm));
3395 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3399 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3400 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3401 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3405 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3409 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3410 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3411 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3412 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3416 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3420 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3421 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3422 sizeof(ps->channels));
3423 ps->flags = kvm->arch.vpit->pit_state.flags;
3424 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3425 memset(&ps->reserved, 0, sizeof(ps->reserved));
3429 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3431 int r = 0, start = 0;
3432 u32 prev_legacy, cur_legacy;
3433 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3434 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3435 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3436 if (!prev_legacy && cur_legacy)
3438 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3439 sizeof(kvm->arch.vpit->pit_state.channels));
3440 kvm->arch.vpit->pit_state.flags = ps->flags;
3441 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3442 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3446 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3447 struct kvm_reinject_control *control)
3449 if (!kvm->arch.vpit)
3451 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3452 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3453 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3458 * Get (and clear) the dirty memory log for a memory slot.
3460 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3461 struct kvm_dirty_log *log)
3464 struct kvm_memory_slot *memslot;
3466 unsigned long is_dirty = 0;
3468 mutex_lock(&kvm->slots_lock);
3471 if (log->slot >= KVM_MEMORY_SLOTS)
3474 memslot = &kvm->memslots->memslots[log->slot];
3476 if (!memslot->dirty_bitmap)
3479 n = kvm_dirty_bitmap_bytes(memslot);
3481 for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3482 is_dirty = memslot->dirty_bitmap[i];
3484 /* If nothing is dirty, don't bother messing with page tables. */
3486 struct kvm_memslots *slots, *old_slots;
3487 unsigned long *dirty_bitmap;
3489 dirty_bitmap = memslot->dirty_bitmap_head;
3490 if (memslot->dirty_bitmap == dirty_bitmap)
3491 dirty_bitmap += n / sizeof(long);
3492 memset(dirty_bitmap, 0, n);
3495 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3498 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3499 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3500 slots->generation++;
3502 old_slots = kvm->memslots;
3503 rcu_assign_pointer(kvm->memslots, slots);
3504 synchronize_srcu_expedited(&kvm->srcu);
3505 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3508 spin_lock(&kvm->mmu_lock);
3509 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3510 spin_unlock(&kvm->mmu_lock);
3513 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
3517 if (clear_user(log->dirty_bitmap, n))
3523 mutex_unlock(&kvm->slots_lock);
3527 long kvm_arch_vm_ioctl(struct file *filp,
3528 unsigned int ioctl, unsigned long arg)
3530 struct kvm *kvm = filp->private_data;
3531 void __user *argp = (void __user *)arg;
3534 * This union makes it completely explicit to gcc-3.x
3535 * that these two variables' stack usage should be
3536 * combined, not added together.
3539 struct kvm_pit_state ps;
3540 struct kvm_pit_state2 ps2;
3541 struct kvm_pit_config pit_config;
3545 case KVM_SET_TSS_ADDR:
3546 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3550 case KVM_SET_IDENTITY_MAP_ADDR: {
3554 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3556 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3561 case KVM_SET_NR_MMU_PAGES:
3562 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3566 case KVM_GET_NR_MMU_PAGES:
3567 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3569 case KVM_CREATE_IRQCHIP: {
3570 struct kvm_pic *vpic;
3572 mutex_lock(&kvm->lock);
3575 goto create_irqchip_unlock;
3577 if (atomic_read(&kvm->online_vcpus))
3578 goto create_irqchip_unlock;
3580 vpic = kvm_create_pic(kvm);
3582 r = kvm_ioapic_init(kvm);
3584 mutex_lock(&kvm->slots_lock);
3585 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3587 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3589 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3591 mutex_unlock(&kvm->slots_lock);
3593 goto create_irqchip_unlock;
3596 goto create_irqchip_unlock;
3598 kvm->arch.vpic = vpic;
3600 r = kvm_setup_default_irq_routing(kvm);
3602 mutex_lock(&kvm->slots_lock);
3603 mutex_lock(&kvm->irq_lock);
3604 kvm_ioapic_destroy(kvm);
3605 kvm_destroy_pic(kvm);
3606 mutex_unlock(&kvm->irq_lock);
3607 mutex_unlock(&kvm->slots_lock);
3609 create_irqchip_unlock:
3610 mutex_unlock(&kvm->lock);
3613 case KVM_CREATE_PIT:
3614 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3616 case KVM_CREATE_PIT2:
3618 if (copy_from_user(&u.pit_config, argp,
3619 sizeof(struct kvm_pit_config)))
3622 mutex_lock(&kvm->slots_lock);
3625 goto create_pit_unlock;
3627 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3631 mutex_unlock(&kvm->slots_lock);
3633 case KVM_IRQ_LINE_STATUS:
3634 case KVM_IRQ_LINE: {
3635 struct kvm_irq_level irq_event;
3638 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3641 if (irqchip_in_kernel(kvm)) {
3643 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3644 irq_event.irq, irq_event.level);
3645 if (ioctl == KVM_IRQ_LINE_STATUS) {
3647 irq_event.status = status;
3648 if (copy_to_user(argp, &irq_event,
3656 case KVM_GET_IRQCHIP: {
3657 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3658 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3664 if (copy_from_user(chip, argp, sizeof *chip))
3665 goto get_irqchip_out;
3667 if (!irqchip_in_kernel(kvm))
3668 goto get_irqchip_out;
3669 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3671 goto get_irqchip_out;
3673 if (copy_to_user(argp, chip, sizeof *chip))
3674 goto get_irqchip_out;
3682 case KVM_SET_IRQCHIP: {
3683 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3684 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3690 if (copy_from_user(chip, argp, sizeof *chip))
3691 goto set_irqchip_out;
3693 if (!irqchip_in_kernel(kvm))
3694 goto set_irqchip_out;
3695 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3697 goto set_irqchip_out;
3707 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3710 if (!kvm->arch.vpit)
3712 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3716 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3723 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3726 if (!kvm->arch.vpit)
3728 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3734 case KVM_GET_PIT2: {
3736 if (!kvm->arch.vpit)
3738 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3742 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3747 case KVM_SET_PIT2: {
3749 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3752 if (!kvm->arch.vpit)
3754 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3760 case KVM_REINJECT_CONTROL: {
3761 struct kvm_reinject_control control;
3763 if (copy_from_user(&control, argp, sizeof(control)))
3765 r = kvm_vm_ioctl_reinject(kvm, &control);
3771 case KVM_XEN_HVM_CONFIG: {
3773 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3774 sizeof(struct kvm_xen_hvm_config)))
3777 if (kvm->arch.xen_hvm_config.flags)
3782 case KVM_SET_CLOCK: {
3783 struct kvm_clock_data user_ns;
3788 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3796 local_irq_disable();
3797 now_ns = get_kernel_ns();
3798 delta = user_ns.clock - now_ns;
3800 kvm->arch.kvmclock_offset = delta;
3803 case KVM_GET_CLOCK: {
3804 struct kvm_clock_data user_ns;
3807 local_irq_disable();
3808 now_ns = get_kernel_ns();
3809 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3812 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3815 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3828 static void kvm_init_msr_list(void)
3833 /* skip the first msrs in the list. KVM-specific */
3834 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3835 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3838 msrs_to_save[j] = msrs_to_save[i];
3841 num_msrs_to_save = j;
3844 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3852 if (!(vcpu->arch.apic &&
3853 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3854 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3865 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3872 if (!(vcpu->arch.apic &&
3873 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3874 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3876 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3886 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3887 struct kvm_segment *var, int seg)
3889 kvm_x86_ops->set_segment(vcpu, var, seg);
3892 void kvm_get_segment(struct kvm_vcpu *vcpu,
3893 struct kvm_segment *var, int seg)
3895 kvm_x86_ops->get_segment(vcpu, var, seg);
3898 static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3903 static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3906 struct x86_exception exception;
3908 BUG_ON(!mmu_is_nested(vcpu));
3910 /* NPT walks are always user-walks */
3911 access |= PFERR_USER_MASK;
3912 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3917 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3918 struct x86_exception *exception)
3920 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3921 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3924 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3925 struct x86_exception *exception)
3927 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3928 access |= PFERR_FETCH_MASK;
3929 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3932 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3933 struct x86_exception *exception)
3935 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3936 access |= PFERR_WRITE_MASK;
3937 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3940 /* uses this to access any guest's mapped memory without checking CPL */
3941 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3942 struct x86_exception *exception)
3944 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3947 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3948 struct kvm_vcpu *vcpu, u32 access,
3949 struct x86_exception *exception)
3952 int r = X86EMUL_CONTINUE;
3955 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3957 unsigned offset = addr & (PAGE_SIZE-1);
3958 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3961 if (gpa == UNMAPPED_GVA)
3962 return X86EMUL_PROPAGATE_FAULT;
3963 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3965 r = X86EMUL_IO_NEEDED;
3977 /* used for instruction fetching */
3978 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3979 gva_t addr, void *val, unsigned int bytes,
3980 struct x86_exception *exception)
3982 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3983 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3985 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3986 access | PFERR_FETCH_MASK,
3990 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3991 gva_t addr, void *val, unsigned int bytes,
3992 struct x86_exception *exception)
3994 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3995 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3997 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4000 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4002 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4003 gva_t addr, void *val, unsigned int bytes,
4004 struct x86_exception *exception)
4006 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4007 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4010 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4011 gva_t addr, void *val,
4013 struct x86_exception *exception)
4015 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4017 int r = X86EMUL_CONTINUE;
4020 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4023 unsigned offset = addr & (PAGE_SIZE-1);
4024 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4027 if (gpa == UNMAPPED_GVA)
4028 return X86EMUL_PROPAGATE_FAULT;
4029 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
4031 r = X86EMUL_IO_NEEDED;
4042 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4044 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4045 gpa_t *gpa, struct x86_exception *exception,
4048 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4050 if (vcpu_match_mmio_gva(vcpu, gva) &&
4051 check_write_user_access(vcpu, write, access,
4052 vcpu->arch.access)) {
4053 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4054 (gva & (PAGE_SIZE - 1));
4055 trace_vcpu_match_mmio(gva, *gpa, write, false);
4060 access |= PFERR_WRITE_MASK;
4062 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4064 if (*gpa == UNMAPPED_GVA)
4067 /* For APIC access vmexit */
4068 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4071 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4072 trace_vcpu_match_mmio(gva, *gpa, write, true);
4079 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4080 const void *val, int bytes)
4084 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4087 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
4091 struct read_write_emulator_ops {
4092 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4094 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4095 void *val, int bytes);
4096 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4097 int bytes, void *val);
4098 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4099 void *val, int bytes);
4103 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4105 if (vcpu->mmio_read_completed) {
4106 memcpy(val, vcpu->mmio_data, bytes);
4107 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4108 vcpu->mmio_phys_addr, *(u64 *)val);
4109 vcpu->mmio_read_completed = 0;
4116 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4117 void *val, int bytes)
4119 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4122 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4123 void *val, int bytes)
4125 return emulator_write_phys(vcpu, gpa, val, bytes);
4128 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4130 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4131 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4134 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4135 void *val, int bytes)
4137 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4138 return X86EMUL_IO_NEEDED;
4141 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4142 void *val, int bytes)
4144 memcpy(vcpu->mmio_data, val, bytes);
4145 memcpy(vcpu->run->mmio.data, vcpu->mmio_data, 8);
4146 return X86EMUL_CONTINUE;
4149 static struct read_write_emulator_ops read_emultor = {
4150 .read_write_prepare = read_prepare,
4151 .read_write_emulate = read_emulate,
4152 .read_write_mmio = vcpu_mmio_read,
4153 .read_write_exit_mmio = read_exit_mmio,
4156 static struct read_write_emulator_ops write_emultor = {
4157 .read_write_emulate = write_emulate,
4158 .read_write_mmio = write_mmio,
4159 .read_write_exit_mmio = write_exit_mmio,
4163 static int emulator_read_write_onepage(unsigned long addr, void *val,
4165 struct x86_exception *exception,
4166 struct kvm_vcpu *vcpu,
4167 struct read_write_emulator_ops *ops)
4171 bool write = ops->write;
4173 if (ops->read_write_prepare &&
4174 ops->read_write_prepare(vcpu, val, bytes))
4175 return X86EMUL_CONTINUE;
4177 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4180 return X86EMUL_PROPAGATE_FAULT;
4182 /* For APIC access vmexit */
4186 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4187 return X86EMUL_CONTINUE;
4191 * Is this MMIO handled locally?
4193 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4194 if (handled == bytes)
4195 return X86EMUL_CONTINUE;
4201 vcpu->mmio_needed = 1;
4202 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4203 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
4204 vcpu->mmio_size = bytes;
4205 vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
4206 vcpu->run->mmio.is_write = vcpu->mmio_is_write = write;
4207 vcpu->mmio_index = 0;
4209 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4212 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4213 void *val, unsigned int bytes,
4214 struct x86_exception *exception,
4215 struct read_write_emulator_ops *ops)
4217 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4219 /* Crossing a page boundary? */
4220 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4223 now = -addr & ~PAGE_MASK;
4224 rc = emulator_read_write_onepage(addr, val, now, exception,
4227 if (rc != X86EMUL_CONTINUE)
4234 return emulator_read_write_onepage(addr, val, bytes, exception,
4238 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4242 struct x86_exception *exception)
4244 return emulator_read_write(ctxt, addr, val, bytes,
4245 exception, &read_emultor);
4248 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4252 struct x86_exception *exception)
4254 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4255 exception, &write_emultor);
4258 #define CMPXCHG_TYPE(t, ptr, old, new) \
4259 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4261 #ifdef CONFIG_X86_64
4262 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4264 # define CMPXCHG64(ptr, old, new) \
4265 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4268 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4273 struct x86_exception *exception)
4275 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4281 /* guests cmpxchg8b have to be emulated atomically */
4282 if (bytes > 8 || (bytes & (bytes - 1)))
4285 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4287 if (gpa == UNMAPPED_GVA ||
4288 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4291 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4294 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4295 if (is_error_page(page)) {
4296 kvm_release_page_clean(page);
4300 kaddr = kmap_atomic(page, KM_USER0);
4301 kaddr += offset_in_page(gpa);
4304 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4307 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4310 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4313 exchanged = CMPXCHG64(kaddr, old, new);
4318 kunmap_atomic(kaddr, KM_USER0);
4319 kvm_release_page_dirty(page);
4322 return X86EMUL_CMPXCHG_FAILED;
4324 kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
4326 return X86EMUL_CONTINUE;
4329 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4331 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4334 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4336 /* TODO: String I/O for in kernel device */
4339 if (vcpu->arch.pio.in)
4340 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4341 vcpu->arch.pio.size, pd);
4343 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4344 vcpu->arch.pio.port, vcpu->arch.pio.size,
4350 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4351 int size, unsigned short port, void *val,
4354 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4356 if (vcpu->arch.pio.count)
4359 trace_kvm_pio(0, port, size, count);
4361 vcpu->arch.pio.port = port;
4362 vcpu->arch.pio.in = 1;
4363 vcpu->arch.pio.count = count;
4364 vcpu->arch.pio.size = size;
4366 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4368 memcpy(val, vcpu->arch.pio_data, size * count);
4369 vcpu->arch.pio.count = 0;
4373 vcpu->run->exit_reason = KVM_EXIT_IO;
4374 vcpu->run->io.direction = KVM_EXIT_IO_IN;
4375 vcpu->run->io.size = size;
4376 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4377 vcpu->run->io.count = count;
4378 vcpu->run->io.port = port;
4383 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4384 int size, unsigned short port,
4385 const void *val, unsigned int count)
4387 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4389 trace_kvm_pio(1, port, size, count);
4391 vcpu->arch.pio.port = port;
4392 vcpu->arch.pio.in = 0;
4393 vcpu->arch.pio.count = count;
4394 vcpu->arch.pio.size = size;
4396 memcpy(vcpu->arch.pio_data, val, size * count);
4398 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4399 vcpu->arch.pio.count = 0;
4403 vcpu->run->exit_reason = KVM_EXIT_IO;
4404 vcpu->run->io.direction = KVM_EXIT_IO_OUT;
4405 vcpu->run->io.size = size;
4406 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4407 vcpu->run->io.count = count;
4408 vcpu->run->io.port = port;
4413 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4415 return kvm_x86_ops->get_segment_base(vcpu, seg);
4418 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4420 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4423 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4425 if (!need_emulate_wbinvd(vcpu))
4426 return X86EMUL_CONTINUE;
4428 if (kvm_x86_ops->has_wbinvd_exit()) {
4429 int cpu = get_cpu();
4431 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4432 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4433 wbinvd_ipi, NULL, 1);
4435 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4438 return X86EMUL_CONTINUE;
4440 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4442 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4444 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4447 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4449 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4452 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4455 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4458 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4460 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4463 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4465 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4466 unsigned long value;
4470 value = kvm_read_cr0(vcpu);
4473 value = vcpu->arch.cr2;
4476 value = kvm_read_cr3(vcpu);
4479 value = kvm_read_cr4(vcpu);
4482 value = kvm_get_cr8(vcpu);
4485 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4492 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4494 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4499 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4502 vcpu->arch.cr2 = val;
4505 res = kvm_set_cr3(vcpu, val);
4508 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4511 res = kvm_set_cr8(vcpu, val);
4514 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4521 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4523 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4526 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4528 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4531 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4533 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4536 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4538 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4541 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4543 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4546 static unsigned long emulator_get_cached_segment_base(
4547 struct x86_emulate_ctxt *ctxt, int seg)
4549 return get_segment_base(emul_to_vcpu(ctxt), seg);
4552 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4553 struct desc_struct *desc, u32 *base3,
4556 struct kvm_segment var;
4558 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4559 *selector = var.selector;
4566 set_desc_limit(desc, var.limit);
4567 set_desc_base(desc, (unsigned long)var.base);
4568 #ifdef CONFIG_X86_64
4570 *base3 = var.base >> 32;
4572 desc->type = var.type;
4574 desc->dpl = var.dpl;
4575 desc->p = var.present;
4576 desc->avl = var.avl;
4584 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4585 struct desc_struct *desc, u32 base3,
4588 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4589 struct kvm_segment var;
4591 var.selector = selector;
4592 var.base = get_desc_base(desc);
4593 #ifdef CONFIG_X86_64
4594 var.base |= ((u64)base3) << 32;
4596 var.limit = get_desc_limit(desc);
4598 var.limit = (var.limit << 12) | 0xfff;
4599 var.type = desc->type;
4600 var.present = desc->p;
4601 var.dpl = desc->dpl;
4606 var.avl = desc->avl;
4607 var.present = desc->p;
4608 var.unusable = !var.present;
4611 kvm_set_segment(vcpu, &var, seg);
4615 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4616 u32 msr_index, u64 *pdata)
4618 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4621 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4622 u32 msr_index, u64 data)
4624 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
4627 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4629 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4632 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4635 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4637 * CR0.TS may reference the host fpu state, not the guest fpu state,
4638 * so it may be clear at this point.
4643 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4648 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4649 struct x86_instruction_info *info,
4650 enum x86_intercept_stage stage)
4652 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4655 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4656 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4658 struct kvm_cpuid_entry2 *cpuid = NULL;
4661 cpuid = kvm_find_cpuid_entry(emul_to_vcpu(ctxt),
4677 static struct x86_emulate_ops emulate_ops = {
4678 .read_std = kvm_read_guest_virt_system,
4679 .write_std = kvm_write_guest_virt_system,
4680 .fetch = kvm_fetch_guest_virt,
4681 .read_emulated = emulator_read_emulated,
4682 .write_emulated = emulator_write_emulated,
4683 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4684 .invlpg = emulator_invlpg,
4685 .pio_in_emulated = emulator_pio_in_emulated,
4686 .pio_out_emulated = emulator_pio_out_emulated,
4687 .get_segment = emulator_get_segment,
4688 .set_segment = emulator_set_segment,
4689 .get_cached_segment_base = emulator_get_cached_segment_base,
4690 .get_gdt = emulator_get_gdt,
4691 .get_idt = emulator_get_idt,
4692 .set_gdt = emulator_set_gdt,
4693 .set_idt = emulator_set_idt,
4694 .get_cr = emulator_get_cr,
4695 .set_cr = emulator_set_cr,
4696 .cpl = emulator_get_cpl,
4697 .get_dr = emulator_get_dr,
4698 .set_dr = emulator_set_dr,
4699 .set_msr = emulator_set_msr,
4700 .get_msr = emulator_get_msr,
4701 .halt = emulator_halt,
4702 .wbinvd = emulator_wbinvd,
4703 .fix_hypercall = emulator_fix_hypercall,
4704 .get_fpu = emulator_get_fpu,
4705 .put_fpu = emulator_put_fpu,
4706 .intercept = emulator_intercept,
4707 .get_cpuid = emulator_get_cpuid,
4710 static void cache_all_regs(struct kvm_vcpu *vcpu)
4712 kvm_register_read(vcpu, VCPU_REGS_RAX);
4713 kvm_register_read(vcpu, VCPU_REGS_RSP);
4714 kvm_register_read(vcpu, VCPU_REGS_RIP);
4715 vcpu->arch.regs_dirty = ~0;
4718 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4720 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4722 * an sti; sti; sequence only disable interrupts for the first
4723 * instruction. So, if the last instruction, be it emulated or
4724 * not, left the system with the INT_STI flag enabled, it
4725 * means that the last instruction is an sti. We should not
4726 * leave the flag on in this case. The same goes for mov ss
4728 if (!(int_shadow & mask))
4729 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4732 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4734 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4735 if (ctxt->exception.vector == PF_VECTOR)
4736 kvm_propagate_fault(vcpu, &ctxt->exception);
4737 else if (ctxt->exception.error_code_valid)
4738 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4739 ctxt->exception.error_code);
4741 kvm_queue_exception(vcpu, ctxt->exception.vector);
4744 static void init_decode_cache(struct x86_emulate_ctxt *ctxt,
4745 const unsigned long *regs)
4747 memset(&ctxt->twobyte, 0,
4748 (void *)&ctxt->regs - (void *)&ctxt->twobyte);
4749 memcpy(ctxt->regs, regs, sizeof(ctxt->regs));
4751 ctxt->fetch.start = 0;
4752 ctxt->fetch.end = 0;
4753 ctxt->io_read.pos = 0;
4754 ctxt->io_read.end = 0;
4755 ctxt->mem_read.pos = 0;
4756 ctxt->mem_read.end = 0;
4759 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4761 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4765 * TODO: fix emulate.c to use guest_read/write_register
4766 * instead of direct ->regs accesses, can save hundred cycles
4767 * on Intel for instructions that don't read/change RSP, for
4770 cache_all_regs(vcpu);
4772 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4774 ctxt->eflags = kvm_get_rflags(vcpu);
4775 ctxt->eip = kvm_rip_read(vcpu);
4776 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4777 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4778 cs_l ? X86EMUL_MODE_PROT64 :
4779 cs_db ? X86EMUL_MODE_PROT32 :
4780 X86EMUL_MODE_PROT16;
4781 ctxt->guest_mode = is_guest_mode(vcpu);
4783 init_decode_cache(ctxt, vcpu->arch.regs);
4784 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4787 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4789 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4792 init_emulate_ctxt(vcpu);
4796 ctxt->_eip = ctxt->eip + inc_eip;
4797 ret = emulate_int_real(ctxt, irq);
4799 if (ret != X86EMUL_CONTINUE)
4800 return EMULATE_FAIL;
4802 ctxt->eip = ctxt->_eip;
4803 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4804 kvm_rip_write(vcpu, ctxt->eip);
4805 kvm_set_rflags(vcpu, ctxt->eflags);
4807 if (irq == NMI_VECTOR)
4808 vcpu->arch.nmi_pending = 0;
4810 vcpu->arch.interrupt.pending = false;
4812 return EMULATE_DONE;
4814 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4816 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4818 int r = EMULATE_DONE;
4820 ++vcpu->stat.insn_emulation_fail;
4821 trace_kvm_emulate_insn_failed(vcpu);
4822 if (!is_guest_mode(vcpu)) {
4823 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4824 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4825 vcpu->run->internal.ndata = 0;
4828 kvm_queue_exception(vcpu, UD_VECTOR);
4833 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4841 * if emulation was due to access to shadowed page table
4842 * and it failed try to unshadow page and re-entetr the
4843 * guest to let CPU execute the instruction.
4845 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4848 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4850 if (gpa == UNMAPPED_GVA)
4851 return true; /* let cpu generate fault */
4853 if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4859 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4866 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4867 bool writeback = true;
4869 kvm_clear_exception_queue(vcpu);
4871 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4872 init_emulate_ctxt(vcpu);
4873 ctxt->interruptibility = 0;
4874 ctxt->have_exception = false;
4875 ctxt->perm_ok = false;
4877 ctxt->only_vendor_specific_insn
4878 = emulation_type & EMULTYPE_TRAP_UD;
4880 r = x86_decode_insn(ctxt, insn, insn_len);
4882 trace_kvm_emulate_insn_start(vcpu);
4883 ++vcpu->stat.insn_emulation;
4884 if (r != EMULATION_OK) {
4885 if (emulation_type & EMULTYPE_TRAP_UD)
4886 return EMULATE_FAIL;
4887 if (reexecute_instruction(vcpu, cr2))
4888 return EMULATE_DONE;
4889 if (emulation_type & EMULTYPE_SKIP)
4890 return EMULATE_FAIL;
4891 return handle_emulation_failure(vcpu);
4895 if (emulation_type & EMULTYPE_SKIP) {
4896 kvm_rip_write(vcpu, ctxt->_eip);
4897 return EMULATE_DONE;
4900 /* this is needed for vmware backdoor interface to work since it
4901 changes registers values during IO operation */
4902 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4903 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4904 memcpy(ctxt->regs, vcpu->arch.regs, sizeof ctxt->regs);
4908 r = x86_emulate_insn(ctxt);
4910 if (r == EMULATION_INTERCEPTED)
4911 return EMULATE_DONE;
4913 if (r == EMULATION_FAILED) {
4914 if (reexecute_instruction(vcpu, cr2))
4915 return EMULATE_DONE;
4917 return handle_emulation_failure(vcpu);
4920 if (ctxt->have_exception) {
4921 inject_emulated_exception(vcpu);
4923 } else if (vcpu->arch.pio.count) {
4924 if (!vcpu->arch.pio.in)
4925 vcpu->arch.pio.count = 0;
4928 r = EMULATE_DO_MMIO;
4929 } else if (vcpu->mmio_needed) {
4930 if (!vcpu->mmio_is_write)
4932 r = EMULATE_DO_MMIO;
4933 } else if (r == EMULATION_RESTART)
4939 toggle_interruptibility(vcpu, ctxt->interruptibility);
4940 kvm_set_rflags(vcpu, ctxt->eflags);
4941 kvm_make_request(KVM_REQ_EVENT, vcpu);
4942 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4943 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4944 kvm_rip_write(vcpu, ctxt->eip);
4946 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
4950 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4952 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4954 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4955 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4956 size, port, &val, 1);
4957 /* do not return to emulator after return from userspace */
4958 vcpu->arch.pio.count = 0;
4961 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4963 static void tsc_bad(void *info)
4965 __this_cpu_write(cpu_tsc_khz, 0);
4968 static void tsc_khz_changed(void *data)
4970 struct cpufreq_freqs *freq = data;
4971 unsigned long khz = 0;
4975 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4976 khz = cpufreq_quick_get(raw_smp_processor_id());
4979 __this_cpu_write(cpu_tsc_khz, khz);
4982 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4985 struct cpufreq_freqs *freq = data;
4987 struct kvm_vcpu *vcpu;
4988 int i, send_ipi = 0;
4991 * We allow guests to temporarily run on slowing clocks,
4992 * provided we notify them after, or to run on accelerating
4993 * clocks, provided we notify them before. Thus time never
4996 * However, we have a problem. We can't atomically update
4997 * the frequency of a given CPU from this function; it is
4998 * merely a notifier, which can be called from any CPU.
4999 * Changing the TSC frequency at arbitrary points in time
5000 * requires a recomputation of local variables related to
5001 * the TSC for each VCPU. We must flag these local variables
5002 * to be updated and be sure the update takes place with the
5003 * new frequency before any guests proceed.
5005 * Unfortunately, the combination of hotplug CPU and frequency
5006 * change creates an intractable locking scenario; the order
5007 * of when these callouts happen is undefined with respect to
5008 * CPU hotplug, and they can race with each other. As such,
5009 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5010 * undefined; you can actually have a CPU frequency change take
5011 * place in between the computation of X and the setting of the
5012 * variable. To protect against this problem, all updates of
5013 * the per_cpu tsc_khz variable are done in an interrupt
5014 * protected IPI, and all callers wishing to update the value
5015 * must wait for a synchronous IPI to complete (which is trivial
5016 * if the caller is on the CPU already). This establishes the
5017 * necessary total order on variable updates.
5019 * Note that because a guest time update may take place
5020 * anytime after the setting of the VCPU's request bit, the
5021 * correct TSC value must be set before the request. However,
5022 * to ensure the update actually makes it to any guest which
5023 * starts running in hardware virtualization between the set
5024 * and the acquisition of the spinlock, we must also ping the
5025 * CPU after setting the request bit.
5029 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5031 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5034 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5036 raw_spin_lock(&kvm_lock);
5037 list_for_each_entry(kvm, &vm_list, vm_list) {
5038 kvm_for_each_vcpu(i, vcpu, kvm) {
5039 if (vcpu->cpu != freq->cpu)
5041 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5042 if (vcpu->cpu != smp_processor_id())
5046 raw_spin_unlock(&kvm_lock);
5048 if (freq->old < freq->new && send_ipi) {
5050 * We upscale the frequency. Must make the guest
5051 * doesn't see old kvmclock values while running with
5052 * the new frequency, otherwise we risk the guest sees
5053 * time go backwards.
5055 * In case we update the frequency for another cpu
5056 * (which might be in guest context) send an interrupt
5057 * to kick the cpu out of guest context. Next time
5058 * guest context is entered kvmclock will be updated,
5059 * so the guest will not see stale values.
5061 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5066 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5067 .notifier_call = kvmclock_cpufreq_notifier
5070 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5071 unsigned long action, void *hcpu)
5073 unsigned int cpu = (unsigned long)hcpu;
5077 case CPU_DOWN_FAILED:
5078 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5080 case CPU_DOWN_PREPARE:
5081 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5087 static struct notifier_block kvmclock_cpu_notifier_block = {
5088 .notifier_call = kvmclock_cpu_notifier,
5089 .priority = -INT_MAX
5092 static void kvm_timer_init(void)
5096 max_tsc_khz = tsc_khz;
5097 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5098 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5099 #ifdef CONFIG_CPU_FREQ
5100 struct cpufreq_policy policy;
5101 memset(&policy, 0, sizeof(policy));
5103 cpufreq_get_policy(&policy, cpu);
5104 if (policy.cpuinfo.max_freq)
5105 max_tsc_khz = policy.cpuinfo.max_freq;
5108 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5109 CPUFREQ_TRANSITION_NOTIFIER);
5111 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5112 for_each_online_cpu(cpu)
5113 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5116 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5118 static int kvm_is_in_guest(void)
5120 return percpu_read(current_vcpu) != NULL;
5123 static int kvm_is_user_mode(void)
5127 if (percpu_read(current_vcpu))
5128 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
5130 return user_mode != 0;
5133 static unsigned long kvm_get_guest_ip(void)
5135 unsigned long ip = 0;
5137 if (percpu_read(current_vcpu))
5138 ip = kvm_rip_read(percpu_read(current_vcpu));
5143 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5144 .is_in_guest = kvm_is_in_guest,
5145 .is_user_mode = kvm_is_user_mode,
5146 .get_guest_ip = kvm_get_guest_ip,
5149 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5151 percpu_write(current_vcpu, vcpu);
5153 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5155 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5157 percpu_write(current_vcpu, NULL);
5159 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5161 static void kvm_set_mmio_spte_mask(void)
5164 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5167 * Set the reserved bits and the present bit of an paging-structure
5168 * entry to generate page fault with PFER.RSV = 1.
5170 mask = ((1ull << (62 - maxphyaddr + 1)) - 1) << maxphyaddr;
5173 #ifdef CONFIG_X86_64
5175 * If reserved bit is not supported, clear the present bit to disable
5178 if (maxphyaddr == 52)
5182 kvm_mmu_set_mmio_spte_mask(mask);
5185 int kvm_arch_init(void *opaque)
5188 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
5191 printk(KERN_ERR "kvm: already loaded the other module\n");
5196 if (!ops->cpu_has_kvm_support()) {
5197 printk(KERN_ERR "kvm: no hardware support\n");
5201 if (ops->disabled_by_bios()) {
5202 printk(KERN_ERR "kvm: disabled by bios\n");
5207 r = kvm_mmu_module_init();
5211 kvm_set_mmio_spte_mask();
5212 kvm_init_msr_list();
5215 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5216 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5220 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5223 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5231 void kvm_arch_exit(void)
5233 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5235 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5236 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5237 CPUFREQ_TRANSITION_NOTIFIER);
5238 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5240 kvm_mmu_module_exit();
5243 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5245 ++vcpu->stat.halt_exits;
5246 if (irqchip_in_kernel(vcpu->kvm)) {
5247 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5250 vcpu->run->exit_reason = KVM_EXIT_HLT;
5254 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5256 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
5259 if (is_long_mode(vcpu))
5262 return a0 | ((gpa_t)a1 << 32);
5265 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5267 u64 param, ingpa, outgpa, ret;
5268 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5269 bool fast, longmode;
5273 * hypercall generates UD from non zero cpl and real mode
5276 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5277 kvm_queue_exception(vcpu, UD_VECTOR);
5281 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5282 longmode = is_long_mode(vcpu) && cs_l == 1;
5285 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5286 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5287 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5288 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5289 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5290 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5292 #ifdef CONFIG_X86_64
5294 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5295 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5296 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5300 code = param & 0xffff;
5301 fast = (param >> 16) & 0x1;
5302 rep_cnt = (param >> 32) & 0xfff;
5303 rep_idx = (param >> 48) & 0xfff;
5305 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5308 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5309 kvm_vcpu_on_spin(vcpu);
5312 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5316 ret = res | (((u64)rep_done & 0xfff) << 32);
5318 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5320 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5321 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5327 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5329 unsigned long nr, a0, a1, a2, a3, ret;
5332 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5333 return kvm_hv_hypercall(vcpu);
5335 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5336 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5337 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5338 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5339 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5341 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5343 if (!is_long_mode(vcpu)) {
5351 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5357 case KVM_HC_VAPIC_POLL_IRQ:
5361 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
5368 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5369 ++vcpu->stat.hypercalls;
5372 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5374 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5376 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5377 char instruction[3];
5378 unsigned long rip = kvm_rip_read(vcpu);
5381 * Blow out the MMU to ensure that no other VCPU has an active mapping
5382 * to ensure that the updated hypercall appears atomically across all
5385 kvm_mmu_zap_all(vcpu->kvm);
5387 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5389 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5392 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
5394 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
5395 int j, nent = vcpu->arch.cpuid_nent;
5397 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
5398 /* when no next entry is found, the current entry[i] is reselected */
5399 for (j = i + 1; ; j = (j + 1) % nent) {
5400 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
5401 if (ej->function == e->function) {
5402 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
5406 return 0; /* silence gcc, even though control never reaches here */
5409 /* find an entry with matching function, matching index (if needed), and that
5410 * should be read next (if it's stateful) */
5411 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
5412 u32 function, u32 index)
5414 if (e->function != function)
5416 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
5418 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
5419 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
5424 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
5425 u32 function, u32 index)
5428 struct kvm_cpuid_entry2 *best = NULL;
5430 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
5431 struct kvm_cpuid_entry2 *e;
5433 e = &vcpu->arch.cpuid_entries[i];
5434 if (is_matching_cpuid_entry(e, function, index)) {
5435 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
5436 move_to_next_stateful_cpuid_entry(vcpu, i);
5443 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
5445 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
5447 struct kvm_cpuid_entry2 *best;
5449 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
5450 if (!best || best->eax < 0x80000008)
5452 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
5454 return best->eax & 0xff;
5460 * If no match is found, check whether we exceed the vCPU's limit
5461 * and return the content of the highest valid _standard_ leaf instead.
5462 * This is to satisfy the CPUID specification.
5464 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
5465 u32 function, u32 index)
5467 struct kvm_cpuid_entry2 *maxlevel;
5469 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
5470 if (!maxlevel || maxlevel->eax >= function)
5472 if (function & 0x80000000) {
5473 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
5477 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
5480 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
5482 u32 function, index;
5483 struct kvm_cpuid_entry2 *best;
5485 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
5486 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5487 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
5488 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
5489 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
5490 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
5491 best = kvm_find_cpuid_entry(vcpu, function, index);
5494 best = check_cpuid_limit(vcpu, function, index);
5497 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
5498 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
5499 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
5500 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
5502 kvm_x86_ops->skip_emulated_instruction(vcpu);
5503 trace_kvm_cpuid(function,
5504 kvm_register_read(vcpu, VCPU_REGS_RAX),
5505 kvm_register_read(vcpu, VCPU_REGS_RBX),
5506 kvm_register_read(vcpu, VCPU_REGS_RCX),
5507 kvm_register_read(vcpu, VCPU_REGS_RDX));
5509 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
5512 * Check if userspace requested an interrupt window, and that the
5513 * interrupt window is open.
5515 * No need to exit to userspace if we already have an interrupt queued.
5517 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5519 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5520 vcpu->run->request_interrupt_window &&
5521 kvm_arch_interrupt_allowed(vcpu));
5524 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5526 struct kvm_run *kvm_run = vcpu->run;
5528 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5529 kvm_run->cr8 = kvm_get_cr8(vcpu);
5530 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5531 if (irqchip_in_kernel(vcpu->kvm))
5532 kvm_run->ready_for_interrupt_injection = 1;
5534 kvm_run->ready_for_interrupt_injection =
5535 kvm_arch_interrupt_allowed(vcpu) &&
5536 !kvm_cpu_has_interrupt(vcpu) &&
5537 !kvm_event_needs_reinjection(vcpu);
5540 static void vapic_enter(struct kvm_vcpu *vcpu)
5542 struct kvm_lapic *apic = vcpu->arch.apic;
5545 if (!apic || !apic->vapic_addr)
5548 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5550 vcpu->arch.apic->vapic_page = page;
5553 static void vapic_exit(struct kvm_vcpu *vcpu)
5555 struct kvm_lapic *apic = vcpu->arch.apic;
5558 if (!apic || !apic->vapic_addr)
5561 idx = srcu_read_lock(&vcpu->kvm->srcu);
5562 kvm_release_page_dirty(apic->vapic_page);
5563 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5564 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5567 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5571 if (!kvm_x86_ops->update_cr8_intercept)
5574 if (!vcpu->arch.apic)
5577 if (!vcpu->arch.apic->vapic_addr)
5578 max_irr = kvm_lapic_find_highest_irr(vcpu);
5585 tpr = kvm_lapic_get_cr8(vcpu);
5587 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5590 static void inject_pending_event(struct kvm_vcpu *vcpu)
5592 /* try to reinject previous events if any */
5593 if (vcpu->arch.exception.pending) {
5594 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5595 vcpu->arch.exception.has_error_code,
5596 vcpu->arch.exception.error_code);
5597 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5598 vcpu->arch.exception.has_error_code,
5599 vcpu->arch.exception.error_code,
5600 vcpu->arch.exception.reinject);
5604 if (vcpu->arch.nmi_injected) {
5605 kvm_x86_ops->set_nmi(vcpu);
5609 if (vcpu->arch.interrupt.pending) {
5610 kvm_x86_ops->set_irq(vcpu);
5614 /* try to inject new event if pending */
5615 if (vcpu->arch.nmi_pending) {
5616 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5617 --vcpu->arch.nmi_pending;
5618 vcpu->arch.nmi_injected = true;
5619 kvm_x86_ops->set_nmi(vcpu);
5621 } else if (kvm_cpu_has_interrupt(vcpu)) {
5622 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5623 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5625 kvm_x86_ops->set_irq(vcpu);
5630 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5632 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5633 !vcpu->guest_xcr0_loaded) {
5634 /* kvm_set_xcr() also depends on this */
5635 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5636 vcpu->guest_xcr0_loaded = 1;
5640 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5642 if (vcpu->guest_xcr0_loaded) {
5643 if (vcpu->arch.xcr0 != host_xcr0)
5644 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5645 vcpu->guest_xcr0_loaded = 0;
5649 static void process_nmi(struct kvm_vcpu *vcpu)
5654 * x86 is limited to one NMI running, and one NMI pending after it.
5655 * If an NMI is already in progress, limit further NMIs to just one.
5656 * Otherwise, allow two (and we'll inject the first one immediately).
5658 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5661 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5662 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5663 kvm_make_request(KVM_REQ_EVENT, vcpu);
5666 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5669 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5670 vcpu->run->request_interrupt_window;
5672 if (vcpu->requests) {
5673 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5674 kvm_mmu_unload(vcpu);
5675 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5676 __kvm_migrate_timers(vcpu);
5677 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5678 r = kvm_guest_time_update(vcpu);
5682 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5683 kvm_mmu_sync_roots(vcpu);
5684 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5685 kvm_x86_ops->tlb_flush(vcpu);
5686 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5687 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5691 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5692 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5696 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5697 vcpu->fpu_active = 0;
5698 kvm_x86_ops->fpu_deactivate(vcpu);
5700 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5701 /* Page is swapped out. Do synthetic halt */
5702 vcpu->arch.apf.halted = true;
5706 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5707 record_steal_time(vcpu);
5708 if (kvm_check_request(KVM_REQ_NMI, vcpu))
5713 r = kvm_mmu_reload(vcpu);
5717 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5718 inject_pending_event(vcpu);
5720 /* enable NMI/IRQ window open exits if needed */
5721 if (vcpu->arch.nmi_pending)
5722 kvm_x86_ops->enable_nmi_window(vcpu);
5723 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5724 kvm_x86_ops->enable_irq_window(vcpu);
5726 if (kvm_lapic_enabled(vcpu)) {
5727 update_cr8_intercept(vcpu);
5728 kvm_lapic_sync_to_vapic(vcpu);
5734 kvm_x86_ops->prepare_guest_switch(vcpu);
5735 if (vcpu->fpu_active)
5736 kvm_load_guest_fpu(vcpu);
5737 kvm_load_guest_xcr0(vcpu);
5739 vcpu->mode = IN_GUEST_MODE;
5741 /* We should set ->mode before check ->requests,
5742 * see the comment in make_all_cpus_request.
5746 local_irq_disable();
5748 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5749 || need_resched() || signal_pending(current)) {
5750 vcpu->mode = OUTSIDE_GUEST_MODE;
5754 kvm_x86_ops->cancel_injection(vcpu);
5759 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5763 if (unlikely(vcpu->arch.switch_db_regs)) {
5765 set_debugreg(vcpu->arch.eff_db[0], 0);
5766 set_debugreg(vcpu->arch.eff_db[1], 1);
5767 set_debugreg(vcpu->arch.eff_db[2], 2);
5768 set_debugreg(vcpu->arch.eff_db[3], 3);
5771 trace_kvm_entry(vcpu->vcpu_id);
5772 kvm_x86_ops->run(vcpu);
5775 * If the guest has used debug registers, at least dr7
5776 * will be disabled while returning to the host.
5777 * If we don't have active breakpoints in the host, we don't
5778 * care about the messed up debug address registers. But if
5779 * we have some of them active, restore the old state.
5781 if (hw_breakpoint_active())
5782 hw_breakpoint_restore();
5784 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu);
5786 vcpu->mode = OUTSIDE_GUEST_MODE;
5793 * We must have an instruction between local_irq_enable() and
5794 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5795 * the interrupt shadow. The stat.exits increment will do nicely.
5796 * But we need to prevent reordering, hence this barrier():
5804 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5807 * Profile KVM exit RIPs:
5809 if (unlikely(prof_on == KVM_PROFILING)) {
5810 unsigned long rip = kvm_rip_read(vcpu);
5811 profile_hit(KVM_PROFILING, (void *)rip);
5815 kvm_lapic_sync_from_vapic(vcpu);
5817 r = kvm_x86_ops->handle_exit(vcpu);
5823 static int __vcpu_run(struct kvm_vcpu *vcpu)
5826 struct kvm *kvm = vcpu->kvm;
5828 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5829 pr_debug("vcpu %d received sipi with vector # %x\n",
5830 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5831 kvm_lapic_reset(vcpu);
5832 r = kvm_arch_vcpu_reset(vcpu);
5835 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5838 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5843 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5844 !vcpu->arch.apf.halted)
5845 r = vcpu_enter_guest(vcpu);
5847 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5848 kvm_vcpu_block(vcpu);
5849 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5850 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5852 switch(vcpu->arch.mp_state) {
5853 case KVM_MP_STATE_HALTED:
5854 vcpu->arch.mp_state =
5855 KVM_MP_STATE_RUNNABLE;
5856 case KVM_MP_STATE_RUNNABLE:
5857 vcpu->arch.apf.halted = false;
5859 case KVM_MP_STATE_SIPI_RECEIVED:
5870 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5871 if (kvm_cpu_has_pending_timer(vcpu))
5872 kvm_inject_pending_timer_irqs(vcpu);
5874 if (dm_request_for_irq_injection(vcpu)) {
5876 vcpu->run->exit_reason = KVM_EXIT_INTR;
5877 ++vcpu->stat.request_irq_exits;
5880 kvm_check_async_pf_completion(vcpu);
5882 if (signal_pending(current)) {
5884 vcpu->run->exit_reason = KVM_EXIT_INTR;
5885 ++vcpu->stat.signal_exits;
5887 if (need_resched()) {
5888 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5890 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5894 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5901 static int complete_mmio(struct kvm_vcpu *vcpu)
5903 struct kvm_run *run = vcpu->run;
5906 if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5909 if (vcpu->mmio_needed) {
5910 vcpu->mmio_needed = 0;
5911 if (!vcpu->mmio_is_write)
5912 memcpy(vcpu->mmio_data + vcpu->mmio_index,
5914 vcpu->mmio_index += 8;
5915 if (vcpu->mmio_index < vcpu->mmio_size) {
5916 run->exit_reason = KVM_EXIT_MMIO;
5917 run->mmio.phys_addr = vcpu->mmio_phys_addr + vcpu->mmio_index;
5918 memcpy(run->mmio.data, vcpu->mmio_data + vcpu->mmio_index, 8);
5919 run->mmio.len = min(vcpu->mmio_size - vcpu->mmio_index, 8);
5920 run->mmio.is_write = vcpu->mmio_is_write;
5921 vcpu->mmio_needed = 1;
5924 if (vcpu->mmio_is_write)
5926 vcpu->mmio_read_completed = 1;
5928 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5929 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5930 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5931 if (r != EMULATE_DONE)
5936 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5941 if (!tsk_used_math(current) && init_fpu(current))
5944 if (vcpu->sigset_active)
5945 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5947 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5948 kvm_vcpu_block(vcpu);
5949 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5954 /* re-sync apic's tpr */
5955 if (!irqchip_in_kernel(vcpu->kvm)) {
5956 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5962 r = complete_mmio(vcpu);
5966 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5967 kvm_register_write(vcpu, VCPU_REGS_RAX,
5968 kvm_run->hypercall.ret);
5970 r = __vcpu_run(vcpu);
5973 post_kvm_run_save(vcpu);
5974 if (vcpu->sigset_active)
5975 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5980 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5982 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5984 * We are here if userspace calls get_regs() in the middle of
5985 * instruction emulation. Registers state needs to be copied
5986 * back from emulation context to vcpu. Usrapace shouldn't do
5987 * that usually, but some bad designed PV devices (vmware
5988 * backdoor interface) need this to work
5990 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5991 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
5992 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5994 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5995 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5996 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5997 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5998 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5999 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6000 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6001 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6002 #ifdef CONFIG_X86_64
6003 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6004 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6005 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6006 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6007 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6008 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6009 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6010 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6013 regs->rip = kvm_rip_read(vcpu);
6014 regs->rflags = kvm_get_rflags(vcpu);
6019 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6021 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6022 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6024 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6025 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6026 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6027 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6028 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6029 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6030 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6031 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6032 #ifdef CONFIG_X86_64
6033 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6034 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6035 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6036 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6037 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6038 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6039 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6040 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6043 kvm_rip_write(vcpu, regs->rip);
6044 kvm_set_rflags(vcpu, regs->rflags);
6046 vcpu->arch.exception.pending = false;
6048 kvm_make_request(KVM_REQ_EVENT, vcpu);
6053 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6055 struct kvm_segment cs;
6057 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6061 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6063 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6064 struct kvm_sregs *sregs)
6068 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6069 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6070 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6071 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6072 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6073 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6075 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6076 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6078 kvm_x86_ops->get_idt(vcpu, &dt);
6079 sregs->idt.limit = dt.size;
6080 sregs->idt.base = dt.address;
6081 kvm_x86_ops->get_gdt(vcpu, &dt);
6082 sregs->gdt.limit = dt.size;
6083 sregs->gdt.base = dt.address;
6085 sregs->cr0 = kvm_read_cr0(vcpu);
6086 sregs->cr2 = vcpu->arch.cr2;
6087 sregs->cr3 = kvm_read_cr3(vcpu);
6088 sregs->cr4 = kvm_read_cr4(vcpu);
6089 sregs->cr8 = kvm_get_cr8(vcpu);
6090 sregs->efer = vcpu->arch.efer;
6091 sregs->apic_base = kvm_get_apic_base(vcpu);
6093 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6095 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6096 set_bit(vcpu->arch.interrupt.nr,
6097 (unsigned long *)sregs->interrupt_bitmap);
6102 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6103 struct kvm_mp_state *mp_state)
6105 mp_state->mp_state = vcpu->arch.mp_state;
6109 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6110 struct kvm_mp_state *mp_state)
6112 vcpu->arch.mp_state = mp_state->mp_state;
6113 kvm_make_request(KVM_REQ_EVENT, vcpu);
6117 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
6118 bool has_error_code, u32 error_code)
6120 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6123 init_emulate_ctxt(vcpu);
6125 ret = emulator_task_switch(ctxt, tss_selector, reason,
6126 has_error_code, error_code);
6129 return EMULATE_FAIL;
6131 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
6132 kvm_rip_write(vcpu, ctxt->eip);
6133 kvm_set_rflags(vcpu, ctxt->eflags);
6134 kvm_make_request(KVM_REQ_EVENT, vcpu);
6135 return EMULATE_DONE;
6137 EXPORT_SYMBOL_GPL(kvm_task_switch);
6139 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6140 struct kvm_sregs *sregs)
6142 int mmu_reset_needed = 0;
6143 int pending_vec, max_bits, idx;
6146 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6149 dt.size = sregs->idt.limit;
6150 dt.address = sregs->idt.base;
6151 kvm_x86_ops->set_idt(vcpu, &dt);
6152 dt.size = sregs->gdt.limit;
6153 dt.address = sregs->gdt.base;
6154 kvm_x86_ops->set_gdt(vcpu, &dt);
6156 vcpu->arch.cr2 = sregs->cr2;
6157 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6158 vcpu->arch.cr3 = sregs->cr3;
6159 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6161 kvm_set_cr8(vcpu, sregs->cr8);
6163 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6164 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6165 kvm_set_apic_base(vcpu, sregs->apic_base);
6167 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6168 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6169 vcpu->arch.cr0 = sregs->cr0;
6171 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6172 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6173 if (sregs->cr4 & X86_CR4_OSXSAVE)
6176 idx = srcu_read_lock(&vcpu->kvm->srcu);
6177 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6178 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6179 mmu_reset_needed = 1;
6181 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6183 if (mmu_reset_needed)
6184 kvm_mmu_reset_context(vcpu);
6186 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
6187 pending_vec = find_first_bit(
6188 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6189 if (pending_vec < max_bits) {
6190 kvm_queue_interrupt(vcpu, pending_vec, false);
6191 pr_debug("Set back pending irq %d\n", pending_vec);
6194 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6195 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6196 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6197 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6198 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6199 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6201 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6202 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6204 update_cr8_intercept(vcpu);
6206 /* Older userspace won't unhalt the vcpu on reset. */
6207 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6208 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6210 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6212 kvm_make_request(KVM_REQ_EVENT, vcpu);
6217 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6218 struct kvm_guest_debug *dbg)
6220 unsigned long rflags;
6223 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6225 if (vcpu->arch.exception.pending)
6227 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6228 kvm_queue_exception(vcpu, DB_VECTOR);
6230 kvm_queue_exception(vcpu, BP_VECTOR);
6234 * Read rflags as long as potentially injected trace flags are still
6237 rflags = kvm_get_rflags(vcpu);
6239 vcpu->guest_debug = dbg->control;
6240 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6241 vcpu->guest_debug = 0;
6243 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6244 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6245 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6246 vcpu->arch.switch_db_regs =
6247 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
6249 for (i = 0; i < KVM_NR_DB_REGS; i++)
6250 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6251 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
6254 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6255 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6256 get_segment_base(vcpu, VCPU_SREG_CS);
6259 * Trigger an rflags update that will inject or remove the trace
6262 kvm_set_rflags(vcpu, rflags);
6264 kvm_x86_ops->set_guest_debug(vcpu, dbg);
6274 * Translate a guest virtual address to a guest physical address.
6276 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6277 struct kvm_translation *tr)
6279 unsigned long vaddr = tr->linear_address;
6283 idx = srcu_read_lock(&vcpu->kvm->srcu);
6284 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6285 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6286 tr->physical_address = gpa;
6287 tr->valid = gpa != UNMAPPED_GVA;
6294 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6296 struct i387_fxsave_struct *fxsave =
6297 &vcpu->arch.guest_fpu.state->fxsave;
6299 memcpy(fpu->fpr, fxsave->st_space, 128);
6300 fpu->fcw = fxsave->cwd;
6301 fpu->fsw = fxsave->swd;
6302 fpu->ftwx = fxsave->twd;
6303 fpu->last_opcode = fxsave->fop;
6304 fpu->last_ip = fxsave->rip;
6305 fpu->last_dp = fxsave->rdp;
6306 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6311 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6313 struct i387_fxsave_struct *fxsave =
6314 &vcpu->arch.guest_fpu.state->fxsave;
6316 memcpy(fxsave->st_space, fpu->fpr, 128);
6317 fxsave->cwd = fpu->fcw;
6318 fxsave->swd = fpu->fsw;
6319 fxsave->twd = fpu->ftwx;
6320 fxsave->fop = fpu->last_opcode;
6321 fxsave->rip = fpu->last_ip;
6322 fxsave->rdp = fpu->last_dp;
6323 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6328 int fx_init(struct kvm_vcpu *vcpu)
6332 err = fpu_alloc(&vcpu->arch.guest_fpu);
6336 fpu_finit(&vcpu->arch.guest_fpu);
6339 * Ensure guest xcr0 is valid for loading
6341 vcpu->arch.xcr0 = XSTATE_FP;
6343 vcpu->arch.cr0 |= X86_CR0_ET;
6347 EXPORT_SYMBOL_GPL(fx_init);
6349 static void fx_free(struct kvm_vcpu *vcpu)
6351 fpu_free(&vcpu->arch.guest_fpu);
6354 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6356 if (vcpu->guest_fpu_loaded)
6360 * Restore all possible states in the guest,
6361 * and assume host would use all available bits.
6362 * Guest xcr0 would be loaded later.
6364 kvm_put_guest_xcr0(vcpu);
6365 vcpu->guest_fpu_loaded = 1;
6366 unlazy_fpu(current);
6367 fpu_restore_checking(&vcpu->arch.guest_fpu);
6371 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6373 kvm_put_guest_xcr0(vcpu);
6375 if (!vcpu->guest_fpu_loaded)
6378 vcpu->guest_fpu_loaded = 0;
6379 fpu_save_init(&vcpu->arch.guest_fpu);
6380 ++vcpu->stat.fpu_reload;
6381 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6385 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6387 kvmclock_reset(vcpu);
6389 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6391 kvm_x86_ops->vcpu_free(vcpu);
6394 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6397 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6398 printk_once(KERN_WARNING
6399 "kvm: SMP vm created on host with unstable TSC; "
6400 "guest TSC will not be reliable\n");
6401 return kvm_x86_ops->vcpu_create(kvm, id);
6404 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6408 vcpu->arch.mtrr_state.have_fixed = 1;
6410 r = kvm_arch_vcpu_reset(vcpu);
6412 r = kvm_mmu_setup(vcpu);
6418 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6420 vcpu->arch.apf.msr_val = 0;
6423 kvm_mmu_unload(vcpu);
6427 kvm_x86_ops->vcpu_free(vcpu);
6430 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
6432 atomic_set(&vcpu->arch.nmi_queued, 0);
6433 vcpu->arch.nmi_pending = 0;
6434 vcpu->arch.nmi_injected = false;
6436 vcpu->arch.switch_db_regs = 0;
6437 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6438 vcpu->arch.dr6 = DR6_FIXED_1;
6439 vcpu->arch.dr7 = DR7_FIXED_1;
6441 kvm_make_request(KVM_REQ_EVENT, vcpu);
6442 vcpu->arch.apf.msr_val = 0;
6443 vcpu->arch.st.msr_val = 0;
6445 kvmclock_reset(vcpu);
6447 kvm_clear_async_pf_completion_queue(vcpu);
6448 kvm_async_pf_hash_reset(vcpu);
6449 vcpu->arch.apf.halted = false;
6451 return kvm_x86_ops->vcpu_reset(vcpu);
6454 int kvm_arch_hardware_enable(void *garbage)
6457 struct kvm_vcpu *vcpu;
6460 kvm_shared_msr_cpu_online();
6461 list_for_each_entry(kvm, &vm_list, vm_list)
6462 kvm_for_each_vcpu(i, vcpu, kvm)
6463 if (vcpu->cpu == smp_processor_id())
6464 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6465 return kvm_x86_ops->hardware_enable(garbage);
6468 void kvm_arch_hardware_disable(void *garbage)
6470 kvm_x86_ops->hardware_disable(garbage);
6471 drop_user_return_notifiers(garbage);
6474 int kvm_arch_hardware_setup(void)
6476 return kvm_x86_ops->hardware_setup();
6479 void kvm_arch_hardware_unsetup(void)
6481 kvm_x86_ops->hardware_unsetup();
6484 void kvm_arch_check_processor_compat(void *rtn)
6486 kvm_x86_ops->check_processor_compatibility(rtn);
6489 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
6491 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
6494 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6500 BUG_ON(vcpu->kvm == NULL);
6503 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6504 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
6505 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
6506 vcpu->arch.mmu.translate_gpa = translate_gpa;
6507 vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
6508 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6509 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6511 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6513 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6518 vcpu->arch.pio_data = page_address(page);
6520 kvm_init_tsc_catchup(vcpu, max_tsc_khz);
6522 r = kvm_mmu_create(vcpu);
6524 goto fail_free_pio_data;
6526 if (irqchip_in_kernel(kvm)) {
6527 r = kvm_create_lapic(vcpu);
6529 goto fail_mmu_destroy;
6532 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6534 if (!vcpu->arch.mce_banks) {
6536 goto fail_free_lapic;
6538 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6540 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6541 goto fail_free_mce_banks;
6543 vcpu->arch.pv_time_enabled = false;
6544 kvm_async_pf_hash_reset(vcpu);
6547 fail_free_mce_banks:
6548 kfree(vcpu->arch.mce_banks);
6550 kvm_free_lapic(vcpu);
6552 kvm_mmu_destroy(vcpu);
6554 free_page((unsigned long)vcpu->arch.pio_data);
6559 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6563 kfree(vcpu->arch.mce_banks);
6564 kvm_free_lapic(vcpu);
6565 idx = srcu_read_lock(&vcpu->kvm->srcu);
6566 kvm_mmu_destroy(vcpu);
6567 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6568 free_page((unsigned long)vcpu->arch.pio_data);
6571 int kvm_arch_init_vm(struct kvm *kvm)
6573 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6574 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6576 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6577 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6579 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6584 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6587 kvm_mmu_unload(vcpu);
6591 static void kvm_free_vcpus(struct kvm *kvm)
6594 struct kvm_vcpu *vcpu;
6597 * Unpin any mmu pages first.
6599 kvm_for_each_vcpu(i, vcpu, kvm) {
6600 kvm_clear_async_pf_completion_queue(vcpu);
6601 kvm_unload_vcpu_mmu(vcpu);
6603 kvm_for_each_vcpu(i, vcpu, kvm)
6604 kvm_arch_vcpu_free(vcpu);
6606 mutex_lock(&kvm->lock);
6607 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6608 kvm->vcpus[i] = NULL;
6610 atomic_set(&kvm->online_vcpus, 0);
6611 mutex_unlock(&kvm->lock);
6614 void kvm_arch_sync_events(struct kvm *kvm)
6616 kvm_free_all_assigned_devices(kvm);
6620 void kvm_arch_destroy_vm(struct kvm *kvm)
6622 kvm_iommu_unmap_guest(kvm);
6623 kfree(kvm->arch.vpic);
6624 kfree(kvm->arch.vioapic);
6625 kvm_free_vcpus(kvm);
6626 if (kvm->arch.apic_access_page)
6627 put_page(kvm->arch.apic_access_page);
6628 if (kvm->arch.ept_identity_pagetable)
6629 put_page(kvm->arch.ept_identity_pagetable);
6632 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6633 struct kvm_memory_slot *memslot,
6634 struct kvm_memory_slot old,
6635 struct kvm_userspace_memory_region *mem,
6638 int npages = memslot->npages;
6639 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6641 /* Prevent internal slot pages from being moved by fork()/COW. */
6642 if (memslot->id >= KVM_MEMORY_SLOTS)
6643 map_flags = MAP_SHARED | MAP_ANONYMOUS;
6645 /*To keep backward compatibility with older userspace,
6646 *x86 needs to hanlde !user_alloc case.
6649 if (npages && !old.rmap) {
6650 unsigned long userspace_addr;
6652 down_write(¤t->mm->mmap_sem);
6653 userspace_addr = do_mmap(NULL, 0,
6655 PROT_READ | PROT_WRITE,
6658 up_write(¤t->mm->mmap_sem);
6660 if (IS_ERR((void *)userspace_addr))
6661 return PTR_ERR((void *)userspace_addr);
6663 memslot->userspace_addr = userspace_addr;
6671 void kvm_arch_commit_memory_region(struct kvm *kvm,
6672 struct kvm_userspace_memory_region *mem,
6673 struct kvm_memory_slot old,
6677 int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
6679 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6682 down_write(¤t->mm->mmap_sem);
6683 ret = do_munmap(current->mm, old.userspace_addr,
6684 old.npages * PAGE_SIZE);
6685 up_write(¤t->mm->mmap_sem);
6688 "kvm_vm_ioctl_set_memory_region: "
6689 "failed to munmap memory\n");
6692 if (!kvm->arch.n_requested_mmu_pages)
6693 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6695 spin_lock(&kvm->mmu_lock);
6697 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6698 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6699 spin_unlock(&kvm->mmu_lock);
6702 void kvm_arch_flush_shadow(struct kvm *kvm)
6704 kvm_mmu_zap_all(kvm);
6705 kvm_reload_remote_mmus(kvm);
6708 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6710 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6711 !vcpu->arch.apf.halted)
6712 || !list_empty_careful(&vcpu->async_pf.done)
6713 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6714 || atomic_read(&vcpu->arch.nmi_queued) ||
6715 (kvm_arch_interrupt_allowed(vcpu) &&
6716 kvm_cpu_has_interrupt(vcpu));
6719 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6722 int cpu = vcpu->cpu;
6724 if (waitqueue_active(&vcpu->wq)) {
6725 wake_up_interruptible(&vcpu->wq);
6726 ++vcpu->stat.halt_wakeup;
6730 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6731 if (kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE)
6732 smp_send_reschedule(cpu);
6736 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6738 return kvm_x86_ops->interrupt_allowed(vcpu);
6741 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6743 unsigned long current_rip = kvm_rip_read(vcpu) +
6744 get_segment_base(vcpu, VCPU_SREG_CS);
6746 return current_rip == linear_rip;
6748 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6750 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6752 unsigned long rflags;
6754 rflags = kvm_x86_ops->get_rflags(vcpu);
6755 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6756 rflags &= ~X86_EFLAGS_TF;
6759 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6761 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6763 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6764 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6765 rflags |= X86_EFLAGS_TF;
6766 kvm_x86_ops->set_rflags(vcpu, rflags);
6767 kvm_make_request(KVM_REQ_EVENT, vcpu);
6769 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6771 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6775 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6776 is_error_page(work->page))
6779 r = kvm_mmu_reload(vcpu);
6783 if (!vcpu->arch.mmu.direct_map &&
6784 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6787 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6790 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6792 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6795 static inline u32 kvm_async_pf_next_probe(u32 key)
6797 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6800 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6802 u32 key = kvm_async_pf_hash_fn(gfn);
6804 while (vcpu->arch.apf.gfns[key] != ~0)
6805 key = kvm_async_pf_next_probe(key);
6807 vcpu->arch.apf.gfns[key] = gfn;
6810 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6813 u32 key = kvm_async_pf_hash_fn(gfn);
6815 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6816 (vcpu->arch.apf.gfns[key] != gfn &&
6817 vcpu->arch.apf.gfns[key] != ~0); i++)
6818 key = kvm_async_pf_next_probe(key);
6823 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6825 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6828 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6832 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6834 vcpu->arch.apf.gfns[i] = ~0;
6836 j = kvm_async_pf_next_probe(j);
6837 if (vcpu->arch.apf.gfns[j] == ~0)
6839 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6841 * k lies cyclically in ]i,j]
6843 * |....j i.k.| or |.k..j i...|
6845 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6846 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6851 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6854 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6858 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6859 struct kvm_async_pf *work)
6861 struct x86_exception fault;
6863 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6864 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6866 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6867 (vcpu->arch.apf.send_user_only &&
6868 kvm_x86_ops->get_cpl(vcpu) == 0))
6869 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6870 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6871 fault.vector = PF_VECTOR;
6872 fault.error_code_valid = true;
6873 fault.error_code = 0;
6874 fault.nested_page_fault = false;
6875 fault.address = work->arch.token;
6876 kvm_inject_page_fault(vcpu, &fault);
6880 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6881 struct kvm_async_pf *work)
6883 struct x86_exception fault;
6885 trace_kvm_async_pf_ready(work->arch.token, work->gva);
6886 if (is_error_page(work->page))
6887 work->arch.token = ~0; /* broadcast wakeup */
6889 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6891 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6892 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6893 fault.vector = PF_VECTOR;
6894 fault.error_code_valid = true;
6895 fault.error_code = 0;
6896 fault.nested_page_fault = false;
6897 fault.address = work->arch.token;
6898 kvm_inject_page_fault(vcpu, &fault);
6900 vcpu->arch.apf.halted = false;
6903 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6905 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6908 return !kvm_event_needs_reinjection(vcpu) &&
6909 kvm_x86_ops->interrupt_allowed(vcpu);
6912 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6913 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6914 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6915 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6916 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6917 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6918 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6919 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6920 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6921 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6922 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6923 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);