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 <trace/events/kvm.h>
49 #define CREATE_TRACE_POINTS
52 #include <asm/debugreg.h>
59 #include <asm/pvclock.h>
60 #include <asm/div64.h>
62 #define MAX_IO_MSRS 256
63 #define KVM_MAX_MCE_BANKS 32
64 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
67 * - enable syscall per default because its emulated by KVM
68 * - enable LME and LMA per default on 64 bit KVM
72 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
74 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
77 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
78 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
80 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
81 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
82 struct kvm_cpuid_entry2 __user *entries);
84 struct kvm_x86_ops *kvm_x86_ops;
85 EXPORT_SYMBOL_GPL(kvm_x86_ops);
88 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
90 #define KVM_NR_SHARED_MSRS 16
92 struct kvm_shared_msrs_global {
94 u32 msrs[KVM_NR_SHARED_MSRS];
97 struct kvm_shared_msrs {
98 struct user_return_notifier urn;
100 struct kvm_shared_msr_values {
103 } values[KVM_NR_SHARED_MSRS];
106 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
107 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
109 struct kvm_stats_debugfs_item debugfs_entries[] = {
110 { "pf_fixed", VCPU_STAT(pf_fixed) },
111 { "pf_guest", VCPU_STAT(pf_guest) },
112 { "tlb_flush", VCPU_STAT(tlb_flush) },
113 { "invlpg", VCPU_STAT(invlpg) },
114 { "exits", VCPU_STAT(exits) },
115 { "io_exits", VCPU_STAT(io_exits) },
116 { "mmio_exits", VCPU_STAT(mmio_exits) },
117 { "signal_exits", VCPU_STAT(signal_exits) },
118 { "irq_window", VCPU_STAT(irq_window_exits) },
119 { "nmi_window", VCPU_STAT(nmi_window_exits) },
120 { "halt_exits", VCPU_STAT(halt_exits) },
121 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
122 { "hypercalls", VCPU_STAT(hypercalls) },
123 { "request_irq", VCPU_STAT(request_irq_exits) },
124 { "irq_exits", VCPU_STAT(irq_exits) },
125 { "host_state_reload", VCPU_STAT(host_state_reload) },
126 { "efer_reload", VCPU_STAT(efer_reload) },
127 { "fpu_reload", VCPU_STAT(fpu_reload) },
128 { "insn_emulation", VCPU_STAT(insn_emulation) },
129 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
130 { "irq_injections", VCPU_STAT(irq_injections) },
131 { "nmi_injections", VCPU_STAT(nmi_injections) },
132 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
133 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
134 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
135 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
136 { "mmu_flooded", VM_STAT(mmu_flooded) },
137 { "mmu_recycled", VM_STAT(mmu_recycled) },
138 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
139 { "mmu_unsync", VM_STAT(mmu_unsync) },
140 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
141 { "largepages", VM_STAT(lpages) },
145 u64 __read_mostly host_xcr0;
147 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
150 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
151 vcpu->arch.apf.gfns[i] = ~0;
154 static void kvm_on_user_return(struct user_return_notifier *urn)
157 struct kvm_shared_msrs *locals
158 = container_of(urn, struct kvm_shared_msrs, urn);
159 struct kvm_shared_msr_values *values;
161 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
162 values = &locals->values[slot];
163 if (values->host != values->curr) {
164 wrmsrl(shared_msrs_global.msrs[slot], values->host);
165 values->curr = values->host;
168 locals->registered = false;
169 user_return_notifier_unregister(urn);
172 static void shared_msr_update(unsigned slot, u32 msr)
174 struct kvm_shared_msrs *smsr;
177 smsr = &__get_cpu_var(shared_msrs);
178 /* only read, and nobody should modify it at this time,
179 * so don't need lock */
180 if (slot >= shared_msrs_global.nr) {
181 printk(KERN_ERR "kvm: invalid MSR slot!");
184 rdmsrl_safe(msr, &value);
185 smsr->values[slot].host = value;
186 smsr->values[slot].curr = value;
189 void kvm_define_shared_msr(unsigned slot, u32 msr)
191 if (slot >= shared_msrs_global.nr)
192 shared_msrs_global.nr = slot + 1;
193 shared_msrs_global.msrs[slot] = msr;
194 /* we need ensured the shared_msr_global have been updated */
197 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
199 static void kvm_shared_msr_cpu_online(void)
203 for (i = 0; i < shared_msrs_global.nr; ++i)
204 shared_msr_update(i, shared_msrs_global.msrs[i]);
207 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
209 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
211 if (((value ^ smsr->values[slot].curr) & mask) == 0)
213 smsr->values[slot].curr = value;
214 wrmsrl(shared_msrs_global.msrs[slot], value);
215 if (!smsr->registered) {
216 smsr->urn.on_user_return = kvm_on_user_return;
217 user_return_notifier_register(&smsr->urn);
218 smsr->registered = true;
221 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
223 static void drop_user_return_notifiers(void *ignore)
225 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
227 if (smsr->registered)
228 kvm_on_user_return(&smsr->urn);
231 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
233 if (irqchip_in_kernel(vcpu->kvm))
234 return vcpu->arch.apic_base;
236 return vcpu->arch.apic_base;
238 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
240 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
242 /* TODO: reserve bits check */
243 if (irqchip_in_kernel(vcpu->kvm))
244 kvm_lapic_set_base(vcpu, data);
246 vcpu->arch.apic_base = data;
248 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
250 #define EXCPT_BENIGN 0
251 #define EXCPT_CONTRIBUTORY 1
254 static int exception_class(int vector)
264 return EXCPT_CONTRIBUTORY;
271 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
272 unsigned nr, bool has_error, u32 error_code,
278 kvm_make_request(KVM_REQ_EVENT, vcpu);
280 if (!vcpu->arch.exception.pending) {
282 vcpu->arch.exception.pending = true;
283 vcpu->arch.exception.has_error_code = has_error;
284 vcpu->arch.exception.nr = nr;
285 vcpu->arch.exception.error_code = error_code;
286 vcpu->arch.exception.reinject = reinject;
290 /* to check exception */
291 prev_nr = vcpu->arch.exception.nr;
292 if (prev_nr == DF_VECTOR) {
293 /* triple fault -> shutdown */
294 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
297 class1 = exception_class(prev_nr);
298 class2 = exception_class(nr);
299 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
300 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
301 /* generate double fault per SDM Table 5-5 */
302 vcpu->arch.exception.pending = true;
303 vcpu->arch.exception.has_error_code = true;
304 vcpu->arch.exception.nr = DF_VECTOR;
305 vcpu->arch.exception.error_code = 0;
307 /* replace previous exception with a new one in a hope
308 that instruction re-execution will regenerate lost
313 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
315 kvm_multiple_exception(vcpu, nr, false, 0, false);
317 EXPORT_SYMBOL_GPL(kvm_queue_exception);
319 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
321 kvm_multiple_exception(vcpu, nr, false, 0, true);
323 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
325 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
328 kvm_inject_gp(vcpu, 0);
330 kvm_x86_ops->skip_emulated_instruction(vcpu);
332 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
334 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
336 ++vcpu->stat.pf_guest;
337 vcpu->arch.cr2 = fault->address;
338 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
341 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
343 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
344 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
346 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
349 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
351 kvm_make_request(KVM_REQ_EVENT, vcpu);
352 vcpu->arch.nmi_pending = 1;
354 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
356 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
358 kvm_multiple_exception(vcpu, nr, true, error_code, false);
360 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
362 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
364 kvm_multiple_exception(vcpu, nr, true, error_code, true);
366 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
369 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
370 * a #GP and return false.
372 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
374 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
376 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
379 EXPORT_SYMBOL_GPL(kvm_require_cpl);
382 * This function will be used to read from the physical memory of the currently
383 * running guest. The difference to kvm_read_guest_page is that this function
384 * can read from guest physical or from the guest's guest physical memory.
386 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
387 gfn_t ngfn, void *data, int offset, int len,
393 ngpa = gfn_to_gpa(ngfn);
394 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
395 if (real_gfn == UNMAPPED_GVA)
398 real_gfn = gpa_to_gfn(real_gfn);
400 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
402 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
404 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
405 void *data, int offset, int len, u32 access)
407 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
408 data, offset, len, access);
412 * Load the pae pdptrs. Return true is they are all valid.
414 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
416 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
417 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
420 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
422 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
423 offset * sizeof(u64), sizeof(pdpte),
424 PFERR_USER_MASK|PFERR_WRITE_MASK);
429 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
430 if (is_present_gpte(pdpte[i]) &&
431 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
438 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
439 __set_bit(VCPU_EXREG_PDPTR,
440 (unsigned long *)&vcpu->arch.regs_avail);
441 __set_bit(VCPU_EXREG_PDPTR,
442 (unsigned long *)&vcpu->arch.regs_dirty);
447 EXPORT_SYMBOL_GPL(load_pdptrs);
449 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
451 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
457 if (is_long_mode(vcpu) || !is_pae(vcpu))
460 if (!test_bit(VCPU_EXREG_PDPTR,
461 (unsigned long *)&vcpu->arch.regs_avail))
464 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
465 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
466 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
467 PFERR_USER_MASK | PFERR_WRITE_MASK);
470 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
476 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
478 unsigned long old_cr0 = kvm_read_cr0(vcpu);
479 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
480 X86_CR0_CD | X86_CR0_NW;
485 if (cr0 & 0xffffffff00000000UL)
489 cr0 &= ~CR0_RESERVED_BITS;
491 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
494 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
497 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
499 if ((vcpu->arch.efer & EFER_LME)) {
504 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
509 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
514 kvm_x86_ops->set_cr0(vcpu, cr0);
516 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
517 kvm_clear_async_pf_completion_queue(vcpu);
518 kvm_async_pf_hash_reset(vcpu);
521 if ((cr0 ^ old_cr0) & update_bits)
522 kvm_mmu_reset_context(vcpu);
525 EXPORT_SYMBOL_GPL(kvm_set_cr0);
527 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
529 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
531 EXPORT_SYMBOL_GPL(kvm_lmsw);
533 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
537 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
538 if (index != XCR_XFEATURE_ENABLED_MASK)
541 if (kvm_x86_ops->get_cpl(vcpu) != 0)
543 if (!(xcr0 & XSTATE_FP))
545 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
547 if (xcr0 & ~host_xcr0)
549 vcpu->arch.xcr0 = xcr0;
550 vcpu->guest_xcr0_loaded = 0;
554 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
556 if (__kvm_set_xcr(vcpu, index, xcr)) {
557 kvm_inject_gp(vcpu, 0);
562 EXPORT_SYMBOL_GPL(kvm_set_xcr);
564 static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
566 struct kvm_cpuid_entry2 *best;
568 best = kvm_find_cpuid_entry(vcpu, 1, 0);
569 return best && (best->ecx & bit(X86_FEATURE_XSAVE));
572 static void update_cpuid(struct kvm_vcpu *vcpu)
574 struct kvm_cpuid_entry2 *best;
576 best = kvm_find_cpuid_entry(vcpu, 1, 0);
580 /* Update OSXSAVE bit */
581 if (cpu_has_xsave && best->function == 0x1) {
582 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
583 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
584 best->ecx |= bit(X86_FEATURE_OSXSAVE);
588 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
590 unsigned long old_cr4 = kvm_read_cr4(vcpu);
591 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
593 if (cr4 & CR4_RESERVED_BITS)
596 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
599 if (is_long_mode(vcpu)) {
600 if (!(cr4 & X86_CR4_PAE))
602 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
603 && ((cr4 ^ old_cr4) & pdptr_bits)
604 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
608 if (cr4 & X86_CR4_VMXE)
611 kvm_x86_ops->set_cr4(vcpu, cr4);
613 if ((cr4 ^ old_cr4) & pdptr_bits)
614 kvm_mmu_reset_context(vcpu);
616 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
621 EXPORT_SYMBOL_GPL(kvm_set_cr4);
623 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
625 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
626 kvm_mmu_sync_roots(vcpu);
627 kvm_mmu_flush_tlb(vcpu);
631 if (is_long_mode(vcpu)) {
632 if (cr3 & CR3_L_MODE_RESERVED_BITS)
636 if (cr3 & CR3_PAE_RESERVED_BITS)
638 if (is_paging(vcpu) &&
639 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
643 * We don't check reserved bits in nonpae mode, because
644 * this isn't enforced, and VMware depends on this.
649 * Does the new cr3 value map to physical memory? (Note, we
650 * catch an invalid cr3 even in real-mode, because it would
651 * cause trouble later on when we turn on paging anyway.)
653 * A real CPU would silently accept an invalid cr3 and would
654 * attempt to use it - with largely undefined (and often hard
655 * to debug) behavior on the guest side.
657 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
659 vcpu->arch.cr3 = cr3;
660 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
661 vcpu->arch.mmu.new_cr3(vcpu);
664 EXPORT_SYMBOL_GPL(kvm_set_cr3);
666 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
668 if (cr8 & CR8_RESERVED_BITS)
670 if (irqchip_in_kernel(vcpu->kvm))
671 kvm_lapic_set_tpr(vcpu, cr8);
673 vcpu->arch.cr8 = cr8;
676 EXPORT_SYMBOL_GPL(kvm_set_cr8);
678 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
680 if (irqchip_in_kernel(vcpu->kvm))
681 return kvm_lapic_get_cr8(vcpu);
683 return vcpu->arch.cr8;
685 EXPORT_SYMBOL_GPL(kvm_get_cr8);
687 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
691 vcpu->arch.db[dr] = val;
692 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
693 vcpu->arch.eff_db[dr] = val;
696 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
700 if (val & 0xffffffff00000000ULL)
702 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
705 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
709 if (val & 0xffffffff00000000ULL)
711 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
712 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
713 kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
714 vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
722 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
726 res = __kvm_set_dr(vcpu, dr, val);
728 kvm_queue_exception(vcpu, UD_VECTOR);
730 kvm_inject_gp(vcpu, 0);
734 EXPORT_SYMBOL_GPL(kvm_set_dr);
736 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
740 *val = vcpu->arch.db[dr];
743 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
747 *val = vcpu->arch.dr6;
750 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
754 *val = vcpu->arch.dr7;
761 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
763 if (_kvm_get_dr(vcpu, dr, val)) {
764 kvm_queue_exception(vcpu, UD_VECTOR);
769 EXPORT_SYMBOL_GPL(kvm_get_dr);
772 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
773 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
775 * This list is modified at module load time to reflect the
776 * capabilities of the host cpu. This capabilities test skips MSRs that are
777 * kvm-specific. Those are put in the beginning of the list.
780 #define KVM_SAVE_MSRS_BEGIN 8
781 static u32 msrs_to_save[] = {
782 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
783 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
784 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
785 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN,
786 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
789 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
791 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
794 static unsigned num_msrs_to_save;
796 static u32 emulated_msrs[] = {
797 MSR_IA32_MISC_ENABLE,
802 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
804 u64 old_efer = vcpu->arch.efer;
806 if (efer & efer_reserved_bits)
810 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
813 if (efer & EFER_FFXSR) {
814 struct kvm_cpuid_entry2 *feat;
816 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
817 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
821 if (efer & EFER_SVME) {
822 struct kvm_cpuid_entry2 *feat;
824 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
825 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
830 efer |= vcpu->arch.efer & EFER_LMA;
832 kvm_x86_ops->set_efer(vcpu, efer);
834 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
836 /* Update reserved bits */
837 if ((efer ^ old_efer) & EFER_NX)
838 kvm_mmu_reset_context(vcpu);
843 void kvm_enable_efer_bits(u64 mask)
845 efer_reserved_bits &= ~mask;
847 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
851 * Writes msr value into into the appropriate "register".
852 * Returns 0 on success, non-0 otherwise.
853 * Assumes vcpu_load() was already called.
855 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
857 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
861 * Adapt set_msr() to msr_io()'s calling convention
863 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
865 return kvm_set_msr(vcpu, index, *data);
868 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
872 struct pvclock_wall_clock wc;
873 struct timespec boot;
878 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
883 ++version; /* first time write, random junk */
887 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
890 * The guest calculates current wall clock time by adding
891 * system time (updated by kvm_guest_time_update below) to the
892 * wall clock specified here. guest system time equals host
893 * system time for us, thus we must fill in host boot time here.
897 wc.sec = boot.tv_sec;
898 wc.nsec = boot.tv_nsec;
899 wc.version = version;
901 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
904 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
907 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
909 uint32_t quotient, remainder;
911 /* Don't try to replace with do_div(), this one calculates
912 * "(dividend << 32) / divisor" */
914 : "=a" (quotient), "=d" (remainder)
915 : "0" (0), "1" (dividend), "r" (divisor) );
919 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
920 s8 *pshift, u32 *pmultiplier)
927 tps64 = base_khz * 1000LL;
928 scaled64 = scaled_khz * 1000LL;
929 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
934 tps32 = (uint32_t)tps64;
935 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
936 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
944 *pmultiplier = div_frac(scaled64, tps32);
946 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
947 __func__, base_khz, scaled_khz, shift, *pmultiplier);
950 static inline u64 get_kernel_ns(void)
954 WARN_ON(preemptible());
956 monotonic_to_bootbased(&ts);
957 return timespec_to_ns(&ts);
960 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
961 unsigned long max_tsc_khz;
963 static inline int kvm_tsc_changes_freq(void)
966 int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
967 cpufreq_quick_get(cpu) != 0;
972 static u64 vcpu_tsc_khz(struct kvm_vcpu *vcpu)
974 if (vcpu->arch.virtual_tsc_khz)
975 return vcpu->arch.virtual_tsc_khz;
977 return __this_cpu_read(cpu_tsc_khz);
980 static inline u64 nsec_to_cycles(u64 nsec)
984 WARN_ON(preemptible());
985 if (kvm_tsc_changes_freq())
986 printk_once(KERN_WARNING
987 "kvm: unreliable cycle conversion on adjustable rate TSC\n");
988 ret = nsec * __this_cpu_read(cpu_tsc_khz);
989 do_div(ret, USEC_PER_SEC);
993 static void kvm_init_tsc_catchup(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
995 /* Compute a scale to convert nanoseconds in TSC cycles */
996 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
997 &vcpu->arch.tsc_catchup_shift,
998 &vcpu->arch.tsc_catchup_mult);
1001 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1003 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1004 vcpu->arch.tsc_catchup_mult,
1005 vcpu->arch.tsc_catchup_shift);
1006 tsc += vcpu->arch.last_tsc_write;
1010 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1012 struct kvm *kvm = vcpu->kvm;
1013 u64 offset, ns, elapsed;
1014 unsigned long flags;
1017 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1018 offset = data - native_read_tsc();
1019 ns = get_kernel_ns();
1020 elapsed = ns - kvm->arch.last_tsc_nsec;
1021 sdiff = data - kvm->arch.last_tsc_write;
1026 * Special case: close write to TSC within 5 seconds of
1027 * another CPU is interpreted as an attempt to synchronize
1028 * The 5 seconds is to accommodate host load / swapping as
1029 * well as any reset of TSC during the boot process.
1031 * In that case, for a reliable TSC, we can match TSC offsets,
1032 * or make a best guest using elapsed value.
1034 if (sdiff < nsec_to_cycles(5ULL * NSEC_PER_SEC) &&
1035 elapsed < 5ULL * NSEC_PER_SEC) {
1036 if (!check_tsc_unstable()) {
1037 offset = kvm->arch.last_tsc_offset;
1038 pr_debug("kvm: matched tsc offset for %llu\n", data);
1040 u64 delta = nsec_to_cycles(elapsed);
1042 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1044 ns = kvm->arch.last_tsc_nsec;
1046 kvm->arch.last_tsc_nsec = ns;
1047 kvm->arch.last_tsc_write = data;
1048 kvm->arch.last_tsc_offset = offset;
1049 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1050 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1052 /* Reset of TSC must disable overshoot protection below */
1053 vcpu->arch.hv_clock.tsc_timestamp = 0;
1054 vcpu->arch.last_tsc_write = data;
1055 vcpu->arch.last_tsc_nsec = ns;
1057 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1059 static int kvm_guest_time_update(struct kvm_vcpu *v)
1061 unsigned long flags;
1062 struct kvm_vcpu_arch *vcpu = &v->arch;
1064 unsigned long this_tsc_khz;
1065 s64 kernel_ns, max_kernel_ns;
1068 /* Keep irq disabled to prevent changes to the clock */
1069 local_irq_save(flags);
1070 kvm_get_msr(v, MSR_IA32_TSC, &tsc_timestamp);
1071 kernel_ns = get_kernel_ns();
1072 this_tsc_khz = vcpu_tsc_khz(v);
1073 if (unlikely(this_tsc_khz == 0)) {
1074 local_irq_restore(flags);
1075 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1080 * We may have to catch up the TSC to match elapsed wall clock
1081 * time for two reasons, even if kvmclock is used.
1082 * 1) CPU could have been running below the maximum TSC rate
1083 * 2) Broken TSC compensation resets the base at each VCPU
1084 * entry to avoid unknown leaps of TSC even when running
1085 * again on the same CPU. This may cause apparent elapsed
1086 * time to disappear, and the guest to stand still or run
1089 if (vcpu->tsc_catchup) {
1090 u64 tsc = compute_guest_tsc(v, kernel_ns);
1091 if (tsc > tsc_timestamp) {
1092 kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1093 tsc_timestamp = tsc;
1097 local_irq_restore(flags);
1099 if (!vcpu->time_page)
1103 * Time as measured by the TSC may go backwards when resetting the base
1104 * tsc_timestamp. The reason for this is that the TSC resolution is
1105 * higher than the resolution of the other clock scales. Thus, many
1106 * possible measurments of the TSC correspond to one measurement of any
1107 * other clock, and so a spread of values is possible. This is not a
1108 * problem for the computation of the nanosecond clock; with TSC rates
1109 * around 1GHZ, there can only be a few cycles which correspond to one
1110 * nanosecond value, and any path through this code will inevitably
1111 * take longer than that. However, with the kernel_ns value itself,
1112 * the precision may be much lower, down to HZ granularity. If the
1113 * first sampling of TSC against kernel_ns ends in the low part of the
1114 * range, and the second in the high end of the range, we can get:
1116 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1118 * As the sampling errors potentially range in the thousands of cycles,
1119 * it is possible such a time value has already been observed by the
1120 * guest. To protect against this, we must compute the system time as
1121 * observed by the guest and ensure the new system time is greater.
1124 if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1125 max_kernel_ns = vcpu->last_guest_tsc -
1126 vcpu->hv_clock.tsc_timestamp;
1127 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1128 vcpu->hv_clock.tsc_to_system_mul,
1129 vcpu->hv_clock.tsc_shift);
1130 max_kernel_ns += vcpu->last_kernel_ns;
1133 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1134 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1135 &vcpu->hv_clock.tsc_shift,
1136 &vcpu->hv_clock.tsc_to_system_mul);
1137 vcpu->hw_tsc_khz = this_tsc_khz;
1140 if (max_kernel_ns > kernel_ns)
1141 kernel_ns = max_kernel_ns;
1143 /* With all the info we got, fill in the values */
1144 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1145 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1146 vcpu->last_kernel_ns = kernel_ns;
1147 vcpu->last_guest_tsc = tsc_timestamp;
1148 vcpu->hv_clock.flags = 0;
1151 * The interface expects us to write an even number signaling that the
1152 * update is finished. Since the guest won't see the intermediate
1153 * state, we just increase by 2 at the end.
1155 vcpu->hv_clock.version += 2;
1157 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1159 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1160 sizeof(vcpu->hv_clock));
1162 kunmap_atomic(shared_kaddr, KM_USER0);
1164 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1168 static bool msr_mtrr_valid(unsigned msr)
1171 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1172 case MSR_MTRRfix64K_00000:
1173 case MSR_MTRRfix16K_80000:
1174 case MSR_MTRRfix16K_A0000:
1175 case MSR_MTRRfix4K_C0000:
1176 case MSR_MTRRfix4K_C8000:
1177 case MSR_MTRRfix4K_D0000:
1178 case MSR_MTRRfix4K_D8000:
1179 case MSR_MTRRfix4K_E0000:
1180 case MSR_MTRRfix4K_E8000:
1181 case MSR_MTRRfix4K_F0000:
1182 case MSR_MTRRfix4K_F8000:
1183 case MSR_MTRRdefType:
1184 case MSR_IA32_CR_PAT:
1192 static bool valid_pat_type(unsigned t)
1194 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1197 static bool valid_mtrr_type(unsigned t)
1199 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1202 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1206 if (!msr_mtrr_valid(msr))
1209 if (msr == MSR_IA32_CR_PAT) {
1210 for (i = 0; i < 8; i++)
1211 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1214 } else if (msr == MSR_MTRRdefType) {
1217 return valid_mtrr_type(data & 0xff);
1218 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1219 for (i = 0; i < 8 ; i++)
1220 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1225 /* variable MTRRs */
1226 return valid_mtrr_type(data & 0xff);
1229 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1231 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1233 if (!mtrr_valid(vcpu, msr, data))
1236 if (msr == MSR_MTRRdefType) {
1237 vcpu->arch.mtrr_state.def_type = data;
1238 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1239 } else if (msr == MSR_MTRRfix64K_00000)
1241 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1242 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1243 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1244 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1245 else if (msr == MSR_IA32_CR_PAT)
1246 vcpu->arch.pat = data;
1247 else { /* Variable MTRRs */
1248 int idx, is_mtrr_mask;
1251 idx = (msr - 0x200) / 2;
1252 is_mtrr_mask = msr - 0x200 - 2 * idx;
1255 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1258 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1262 kvm_mmu_reset_context(vcpu);
1266 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1268 u64 mcg_cap = vcpu->arch.mcg_cap;
1269 unsigned bank_num = mcg_cap & 0xff;
1272 case MSR_IA32_MCG_STATUS:
1273 vcpu->arch.mcg_status = data;
1275 case MSR_IA32_MCG_CTL:
1276 if (!(mcg_cap & MCG_CTL_P))
1278 if (data != 0 && data != ~(u64)0)
1280 vcpu->arch.mcg_ctl = data;
1283 if (msr >= MSR_IA32_MC0_CTL &&
1284 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1285 u32 offset = msr - MSR_IA32_MC0_CTL;
1286 /* only 0 or all 1s can be written to IA32_MCi_CTL
1287 * some Linux kernels though clear bit 10 in bank 4 to
1288 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1289 * this to avoid an uncatched #GP in the guest
1291 if ((offset & 0x3) == 0 &&
1292 data != 0 && (data | (1 << 10)) != ~(u64)0)
1294 vcpu->arch.mce_banks[offset] = data;
1302 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1304 struct kvm *kvm = vcpu->kvm;
1305 int lm = is_long_mode(vcpu);
1306 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1307 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1308 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1309 : kvm->arch.xen_hvm_config.blob_size_32;
1310 u32 page_num = data & ~PAGE_MASK;
1311 u64 page_addr = data & PAGE_MASK;
1316 if (page_num >= blob_size)
1319 page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1323 if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1325 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1334 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1336 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1339 static bool kvm_hv_msr_partition_wide(u32 msr)
1343 case HV_X64_MSR_GUEST_OS_ID:
1344 case HV_X64_MSR_HYPERCALL:
1352 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1354 struct kvm *kvm = vcpu->kvm;
1357 case HV_X64_MSR_GUEST_OS_ID:
1358 kvm->arch.hv_guest_os_id = data;
1359 /* setting guest os id to zero disables hypercall page */
1360 if (!kvm->arch.hv_guest_os_id)
1361 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1363 case HV_X64_MSR_HYPERCALL: {
1368 /* if guest os id is not set hypercall should remain disabled */
1369 if (!kvm->arch.hv_guest_os_id)
1371 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1372 kvm->arch.hv_hypercall = data;
1375 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1376 addr = gfn_to_hva(kvm, gfn);
1377 if (kvm_is_error_hva(addr))
1379 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1380 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1381 if (copy_to_user((void __user *)addr, instructions, 4))
1383 kvm->arch.hv_hypercall = data;
1387 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1388 "data 0x%llx\n", msr, data);
1394 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1397 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1400 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1401 vcpu->arch.hv_vapic = data;
1404 addr = gfn_to_hva(vcpu->kvm, data >>
1405 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1406 if (kvm_is_error_hva(addr))
1408 if (clear_user((void __user *)addr, PAGE_SIZE))
1410 vcpu->arch.hv_vapic = data;
1413 case HV_X64_MSR_EOI:
1414 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1415 case HV_X64_MSR_ICR:
1416 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1417 case HV_X64_MSR_TPR:
1418 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1420 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1421 "data 0x%llx\n", msr, data);
1428 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1430 gpa_t gpa = data & ~0x3f;
1432 /* Bits 2:5 are resrved, Should be zero */
1436 vcpu->arch.apf.msr_val = data;
1438 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1439 kvm_clear_async_pf_completion_queue(vcpu);
1440 kvm_async_pf_hash_reset(vcpu);
1444 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1447 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1448 kvm_async_pf_wakeup_all(vcpu);
1452 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1454 if (vcpu->arch.time_page) {
1455 kvm_release_page_dirty(vcpu->arch.time_page);
1456 vcpu->arch.time_page = NULL;
1460 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1464 return set_efer(vcpu, data);
1466 data &= ~(u64)0x40; /* ignore flush filter disable */
1467 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1469 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1474 case MSR_FAM10H_MMIO_CONF_BASE:
1476 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1481 case MSR_AMD64_NB_CFG:
1483 case MSR_IA32_DEBUGCTLMSR:
1485 /* We support the non-activated case already */
1487 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1488 /* Values other than LBR and BTF are vendor-specific,
1489 thus reserved and should throw a #GP */
1492 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1495 case MSR_IA32_UCODE_REV:
1496 case MSR_IA32_UCODE_WRITE:
1497 case MSR_VM_HSAVE_PA:
1498 case MSR_AMD64_PATCH_LOADER:
1500 case 0x200 ... 0x2ff:
1501 return set_msr_mtrr(vcpu, msr, data);
1502 case MSR_IA32_APICBASE:
1503 kvm_set_apic_base(vcpu, data);
1505 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1506 return kvm_x2apic_msr_write(vcpu, msr, data);
1507 case MSR_IA32_MISC_ENABLE:
1508 vcpu->arch.ia32_misc_enable_msr = data;
1510 case MSR_KVM_WALL_CLOCK_NEW:
1511 case MSR_KVM_WALL_CLOCK:
1512 vcpu->kvm->arch.wall_clock = data;
1513 kvm_write_wall_clock(vcpu->kvm, data);
1515 case MSR_KVM_SYSTEM_TIME_NEW:
1516 case MSR_KVM_SYSTEM_TIME: {
1517 kvmclock_reset(vcpu);
1519 vcpu->arch.time = data;
1520 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1522 /* we verify if the enable bit is set... */
1526 /* ...but clean it before doing the actual write */
1527 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1529 vcpu->arch.time_page =
1530 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1532 if (is_error_page(vcpu->arch.time_page)) {
1533 kvm_release_page_clean(vcpu->arch.time_page);
1534 vcpu->arch.time_page = NULL;
1538 case MSR_KVM_ASYNC_PF_EN:
1539 if (kvm_pv_enable_async_pf(vcpu, data))
1542 case MSR_IA32_MCG_CTL:
1543 case MSR_IA32_MCG_STATUS:
1544 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1545 return set_msr_mce(vcpu, msr, data);
1547 /* Performance counters are not protected by a CPUID bit,
1548 * so we should check all of them in the generic path for the sake of
1549 * cross vendor migration.
1550 * Writing a zero into the event select MSRs disables them,
1551 * which we perfectly emulate ;-). Any other value should be at least
1552 * reported, some guests depend on them.
1554 case MSR_P6_EVNTSEL0:
1555 case MSR_P6_EVNTSEL1:
1556 case MSR_K7_EVNTSEL0:
1557 case MSR_K7_EVNTSEL1:
1558 case MSR_K7_EVNTSEL2:
1559 case MSR_K7_EVNTSEL3:
1561 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1562 "0x%x data 0x%llx\n", msr, data);
1564 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1565 * so we ignore writes to make it happy.
1567 case MSR_P6_PERFCTR0:
1568 case MSR_P6_PERFCTR1:
1569 case MSR_K7_PERFCTR0:
1570 case MSR_K7_PERFCTR1:
1571 case MSR_K7_PERFCTR2:
1572 case MSR_K7_PERFCTR3:
1573 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1574 "0x%x data 0x%llx\n", msr, data);
1576 case MSR_K7_CLK_CTL:
1578 * Ignore all writes to this no longer documented MSR.
1579 * Writes are only relevant for old K7 processors,
1580 * all pre-dating SVM, but a recommended workaround from
1581 * AMD for these chips. It is possible to speicify the
1582 * affected processor models on the command line, hence
1583 * the need to ignore the workaround.
1586 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1587 if (kvm_hv_msr_partition_wide(msr)) {
1589 mutex_lock(&vcpu->kvm->lock);
1590 r = set_msr_hyperv_pw(vcpu, msr, data);
1591 mutex_unlock(&vcpu->kvm->lock);
1594 return set_msr_hyperv(vcpu, msr, data);
1596 case MSR_IA32_BBL_CR_CTL3:
1597 /* Drop writes to this legacy MSR -- see rdmsr
1598 * counterpart for further detail.
1600 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1603 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1604 return xen_hvm_config(vcpu, data);
1606 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1610 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1617 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1621 * Reads an msr value (of 'msr_index') into 'pdata'.
1622 * Returns 0 on success, non-0 otherwise.
1623 * Assumes vcpu_load() was already called.
1625 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1627 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1630 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1632 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1634 if (!msr_mtrr_valid(msr))
1637 if (msr == MSR_MTRRdefType)
1638 *pdata = vcpu->arch.mtrr_state.def_type +
1639 (vcpu->arch.mtrr_state.enabled << 10);
1640 else if (msr == MSR_MTRRfix64K_00000)
1642 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1643 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1644 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1645 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1646 else if (msr == MSR_IA32_CR_PAT)
1647 *pdata = vcpu->arch.pat;
1648 else { /* Variable MTRRs */
1649 int idx, is_mtrr_mask;
1652 idx = (msr - 0x200) / 2;
1653 is_mtrr_mask = msr - 0x200 - 2 * idx;
1656 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1659 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1666 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1669 u64 mcg_cap = vcpu->arch.mcg_cap;
1670 unsigned bank_num = mcg_cap & 0xff;
1673 case MSR_IA32_P5_MC_ADDR:
1674 case MSR_IA32_P5_MC_TYPE:
1677 case MSR_IA32_MCG_CAP:
1678 data = vcpu->arch.mcg_cap;
1680 case MSR_IA32_MCG_CTL:
1681 if (!(mcg_cap & MCG_CTL_P))
1683 data = vcpu->arch.mcg_ctl;
1685 case MSR_IA32_MCG_STATUS:
1686 data = vcpu->arch.mcg_status;
1689 if (msr >= MSR_IA32_MC0_CTL &&
1690 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1691 u32 offset = msr - MSR_IA32_MC0_CTL;
1692 data = vcpu->arch.mce_banks[offset];
1701 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1704 struct kvm *kvm = vcpu->kvm;
1707 case HV_X64_MSR_GUEST_OS_ID:
1708 data = kvm->arch.hv_guest_os_id;
1710 case HV_X64_MSR_HYPERCALL:
1711 data = kvm->arch.hv_hypercall;
1714 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1722 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1727 case HV_X64_MSR_VP_INDEX: {
1730 kvm_for_each_vcpu(r, v, vcpu->kvm)
1735 case HV_X64_MSR_EOI:
1736 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1737 case HV_X64_MSR_ICR:
1738 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1739 case HV_X64_MSR_TPR:
1740 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1742 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1749 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1754 case MSR_IA32_PLATFORM_ID:
1755 case MSR_IA32_UCODE_REV:
1756 case MSR_IA32_EBL_CR_POWERON:
1757 case MSR_IA32_DEBUGCTLMSR:
1758 case MSR_IA32_LASTBRANCHFROMIP:
1759 case MSR_IA32_LASTBRANCHTOIP:
1760 case MSR_IA32_LASTINTFROMIP:
1761 case MSR_IA32_LASTINTTOIP:
1764 case MSR_VM_HSAVE_PA:
1765 case MSR_P6_PERFCTR0:
1766 case MSR_P6_PERFCTR1:
1767 case MSR_P6_EVNTSEL0:
1768 case MSR_P6_EVNTSEL1:
1769 case MSR_K7_EVNTSEL0:
1770 case MSR_K7_PERFCTR0:
1771 case MSR_K8_INT_PENDING_MSG:
1772 case MSR_AMD64_NB_CFG:
1773 case MSR_FAM10H_MMIO_CONF_BASE:
1777 data = 0x500 | KVM_NR_VAR_MTRR;
1779 case 0x200 ... 0x2ff:
1780 return get_msr_mtrr(vcpu, msr, pdata);
1781 case 0xcd: /* fsb frequency */
1785 * MSR_EBC_FREQUENCY_ID
1786 * Conservative value valid for even the basic CPU models.
1787 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1788 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1789 * and 266MHz for model 3, or 4. Set Core Clock
1790 * Frequency to System Bus Frequency Ratio to 1 (bits
1791 * 31:24) even though these are only valid for CPU
1792 * models > 2, however guests may end up dividing or
1793 * multiplying by zero otherwise.
1795 case MSR_EBC_FREQUENCY_ID:
1798 case MSR_IA32_APICBASE:
1799 data = kvm_get_apic_base(vcpu);
1801 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1802 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1804 case MSR_IA32_MISC_ENABLE:
1805 data = vcpu->arch.ia32_misc_enable_msr;
1807 case MSR_IA32_PERF_STATUS:
1808 /* TSC increment by tick */
1810 /* CPU multiplier */
1811 data |= (((uint64_t)4ULL) << 40);
1814 data = vcpu->arch.efer;
1816 case MSR_KVM_WALL_CLOCK:
1817 case MSR_KVM_WALL_CLOCK_NEW:
1818 data = vcpu->kvm->arch.wall_clock;
1820 case MSR_KVM_SYSTEM_TIME:
1821 case MSR_KVM_SYSTEM_TIME_NEW:
1822 data = vcpu->arch.time;
1824 case MSR_KVM_ASYNC_PF_EN:
1825 data = vcpu->arch.apf.msr_val;
1827 case MSR_IA32_P5_MC_ADDR:
1828 case MSR_IA32_P5_MC_TYPE:
1829 case MSR_IA32_MCG_CAP:
1830 case MSR_IA32_MCG_CTL:
1831 case MSR_IA32_MCG_STATUS:
1832 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1833 return get_msr_mce(vcpu, msr, pdata);
1834 case MSR_K7_CLK_CTL:
1836 * Provide expected ramp-up count for K7. All other
1837 * are set to zero, indicating minimum divisors for
1840 * This prevents guest kernels on AMD host with CPU
1841 * type 6, model 8 and higher from exploding due to
1842 * the rdmsr failing.
1846 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1847 if (kvm_hv_msr_partition_wide(msr)) {
1849 mutex_lock(&vcpu->kvm->lock);
1850 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1851 mutex_unlock(&vcpu->kvm->lock);
1854 return get_msr_hyperv(vcpu, msr, pdata);
1856 case MSR_IA32_BBL_CR_CTL3:
1857 /* This legacy MSR exists but isn't fully documented in current
1858 * silicon. It is however accessed by winxp in very narrow
1859 * scenarios where it sets bit #19, itself documented as
1860 * a "reserved" bit. Best effort attempt to source coherent
1861 * read data here should the balance of the register be
1862 * interpreted by the guest:
1864 * L2 cache control register 3: 64GB range, 256KB size,
1865 * enabled, latency 0x1, configured
1871 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1874 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1882 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1885 * Read or write a bunch of msrs. All parameters are kernel addresses.
1887 * @return number of msrs set successfully.
1889 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1890 struct kvm_msr_entry *entries,
1891 int (*do_msr)(struct kvm_vcpu *vcpu,
1892 unsigned index, u64 *data))
1896 idx = srcu_read_lock(&vcpu->kvm->srcu);
1897 for (i = 0; i < msrs->nmsrs; ++i)
1898 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1900 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1906 * Read or write a bunch of msrs. Parameters are user addresses.
1908 * @return number of msrs set successfully.
1910 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1911 int (*do_msr)(struct kvm_vcpu *vcpu,
1912 unsigned index, u64 *data),
1915 struct kvm_msrs msrs;
1916 struct kvm_msr_entry *entries;
1921 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1925 if (msrs.nmsrs >= MAX_IO_MSRS)
1929 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1930 entries = kmalloc(size, GFP_KERNEL);
1935 if (copy_from_user(entries, user_msrs->entries, size))
1938 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1943 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1954 int kvm_dev_ioctl_check_extension(long ext)
1959 case KVM_CAP_IRQCHIP:
1961 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1962 case KVM_CAP_SET_TSS_ADDR:
1963 case KVM_CAP_EXT_CPUID:
1964 case KVM_CAP_CLOCKSOURCE:
1966 case KVM_CAP_NOP_IO_DELAY:
1967 case KVM_CAP_MP_STATE:
1968 case KVM_CAP_SYNC_MMU:
1969 case KVM_CAP_USER_NMI:
1970 case KVM_CAP_REINJECT_CONTROL:
1971 case KVM_CAP_IRQ_INJECT_STATUS:
1972 case KVM_CAP_ASSIGN_DEV_IRQ:
1974 case KVM_CAP_IOEVENTFD:
1976 case KVM_CAP_PIT_STATE2:
1977 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1978 case KVM_CAP_XEN_HVM:
1979 case KVM_CAP_ADJUST_CLOCK:
1980 case KVM_CAP_VCPU_EVENTS:
1981 case KVM_CAP_HYPERV:
1982 case KVM_CAP_HYPERV_VAPIC:
1983 case KVM_CAP_HYPERV_SPIN:
1984 case KVM_CAP_PCI_SEGMENT:
1985 case KVM_CAP_DEBUGREGS:
1986 case KVM_CAP_X86_ROBUST_SINGLESTEP:
1988 case KVM_CAP_ASYNC_PF:
1991 case KVM_CAP_COALESCED_MMIO:
1992 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1995 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1997 case KVM_CAP_NR_VCPUS:
2000 case KVM_CAP_NR_MEMSLOTS:
2001 r = KVM_MEMORY_SLOTS;
2003 case KVM_CAP_PV_MMU: /* obsolete */
2010 r = KVM_MAX_MCE_BANKS;
2023 long kvm_arch_dev_ioctl(struct file *filp,
2024 unsigned int ioctl, unsigned long arg)
2026 void __user *argp = (void __user *)arg;
2030 case KVM_GET_MSR_INDEX_LIST: {
2031 struct kvm_msr_list __user *user_msr_list = argp;
2032 struct kvm_msr_list msr_list;
2036 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2039 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2040 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2043 if (n < msr_list.nmsrs)
2046 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2047 num_msrs_to_save * sizeof(u32)))
2049 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2051 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2056 case KVM_GET_SUPPORTED_CPUID: {
2057 struct kvm_cpuid2 __user *cpuid_arg = argp;
2058 struct kvm_cpuid2 cpuid;
2061 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2063 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2064 cpuid_arg->entries);
2069 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2074 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2077 mce_cap = KVM_MCE_CAP_SUPPORTED;
2079 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2091 static void wbinvd_ipi(void *garbage)
2096 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2098 return vcpu->kvm->arch.iommu_domain &&
2099 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2102 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2104 /* Address WBINVD may be executed by guest */
2105 if (need_emulate_wbinvd(vcpu)) {
2106 if (kvm_x86_ops->has_wbinvd_exit())
2107 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2108 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2109 smp_call_function_single(vcpu->cpu,
2110 wbinvd_ipi, NULL, 1);
2113 kvm_x86_ops->vcpu_load(vcpu, cpu);
2114 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2115 /* Make sure TSC doesn't go backwards */
2116 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2117 native_read_tsc() - vcpu->arch.last_host_tsc;
2119 mark_tsc_unstable("KVM discovered backwards TSC");
2120 if (check_tsc_unstable()) {
2121 kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2122 vcpu->arch.tsc_catchup = 1;
2124 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2125 if (vcpu->cpu != cpu)
2126 kvm_migrate_timers(vcpu);
2131 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2133 kvm_x86_ops->vcpu_put(vcpu);
2134 kvm_put_guest_fpu(vcpu);
2135 vcpu->arch.last_host_tsc = native_read_tsc();
2138 static int is_efer_nx(void)
2140 unsigned long long efer = 0;
2142 rdmsrl_safe(MSR_EFER, &efer);
2143 return efer & EFER_NX;
2146 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2149 struct kvm_cpuid_entry2 *e, *entry;
2152 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2153 e = &vcpu->arch.cpuid_entries[i];
2154 if (e->function == 0x80000001) {
2159 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2160 entry->edx &= ~(1 << 20);
2161 printk(KERN_INFO "kvm: guest NX capability removed\n");
2165 /* when an old userspace process fills a new kernel module */
2166 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2167 struct kvm_cpuid *cpuid,
2168 struct kvm_cpuid_entry __user *entries)
2171 struct kvm_cpuid_entry *cpuid_entries;
2174 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2177 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2181 if (copy_from_user(cpuid_entries, entries,
2182 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2184 for (i = 0; i < cpuid->nent; i++) {
2185 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2186 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2187 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2188 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2189 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2190 vcpu->arch.cpuid_entries[i].index = 0;
2191 vcpu->arch.cpuid_entries[i].flags = 0;
2192 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2193 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2194 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2196 vcpu->arch.cpuid_nent = cpuid->nent;
2197 cpuid_fix_nx_cap(vcpu);
2199 kvm_apic_set_version(vcpu);
2200 kvm_x86_ops->cpuid_update(vcpu);
2204 vfree(cpuid_entries);
2209 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2210 struct kvm_cpuid2 *cpuid,
2211 struct kvm_cpuid_entry2 __user *entries)
2216 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2219 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2220 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2222 vcpu->arch.cpuid_nent = cpuid->nent;
2223 kvm_apic_set_version(vcpu);
2224 kvm_x86_ops->cpuid_update(vcpu);
2232 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2233 struct kvm_cpuid2 *cpuid,
2234 struct kvm_cpuid_entry2 __user *entries)
2239 if (cpuid->nent < vcpu->arch.cpuid_nent)
2242 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2243 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2248 cpuid->nent = vcpu->arch.cpuid_nent;
2252 static void cpuid_mask(u32 *word, int wordnum)
2254 *word &= boot_cpu_data.x86_capability[wordnum];
2257 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2260 entry->function = function;
2261 entry->index = index;
2262 cpuid_count(entry->function, entry->index,
2263 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2267 #define F(x) bit(X86_FEATURE_##x)
2269 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2270 u32 index, int *nent, int maxnent)
2272 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2273 #ifdef CONFIG_X86_64
2274 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2276 unsigned f_lm = F(LM);
2278 unsigned f_gbpages = 0;
2281 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2284 const u32 kvm_supported_word0_x86_features =
2285 F(FPU) | F(VME) | F(DE) | F(PSE) |
2286 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2287 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2288 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2289 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2290 0 /* Reserved, DS, ACPI */ | F(MMX) |
2291 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2292 0 /* HTT, TM, Reserved, PBE */;
2293 /* cpuid 0x80000001.edx */
2294 const u32 kvm_supported_word1_x86_features =
2295 F(FPU) | F(VME) | F(DE) | F(PSE) |
2296 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2297 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2298 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2299 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2300 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2301 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2302 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2304 const u32 kvm_supported_word4_x86_features =
2305 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
2306 0 /* DS-CPL, VMX, SMX, EST */ |
2307 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2308 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2309 0 /* Reserved, DCA */ | F(XMM4_1) |
2310 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2311 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2313 /* cpuid 0x80000001.ecx */
2314 const u32 kvm_supported_word6_x86_features =
2315 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2316 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2317 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2318 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2320 /* all calls to cpuid_count() should be made on the same cpu */
2322 do_cpuid_1_ent(entry, function, index);
2327 entry->eax = min(entry->eax, (u32)0xd);
2330 entry->edx &= kvm_supported_word0_x86_features;
2331 cpuid_mask(&entry->edx, 0);
2332 entry->ecx &= kvm_supported_word4_x86_features;
2333 cpuid_mask(&entry->ecx, 4);
2334 /* we support x2apic emulation even if host does not support
2335 * it since we emulate x2apic in software */
2336 entry->ecx |= F(X2APIC);
2338 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2339 * may return different values. This forces us to get_cpu() before
2340 * issuing the first command, and also to emulate this annoying behavior
2341 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2343 int t, times = entry->eax & 0xff;
2345 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2346 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2347 for (t = 1; t < times && *nent < maxnent; ++t) {
2348 do_cpuid_1_ent(&entry[t], function, 0);
2349 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2354 /* function 4 and 0xb have additional index. */
2358 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2359 /* read more entries until cache_type is zero */
2360 for (i = 1; *nent < maxnent; ++i) {
2361 cache_type = entry[i - 1].eax & 0x1f;
2364 do_cpuid_1_ent(&entry[i], function, i);
2366 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2374 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2375 /* read more entries until level_type is zero */
2376 for (i = 1; *nent < maxnent; ++i) {
2377 level_type = entry[i - 1].ecx & 0xff00;
2380 do_cpuid_1_ent(&entry[i], function, i);
2382 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2390 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2391 for (i = 1; *nent < maxnent && i < 64; ++i) {
2392 if (entry[i].eax == 0)
2394 do_cpuid_1_ent(&entry[i], function, i);
2396 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2401 case KVM_CPUID_SIGNATURE: {
2402 char signature[12] = "KVMKVMKVM\0\0";
2403 u32 *sigptr = (u32 *)signature;
2405 entry->ebx = sigptr[0];
2406 entry->ecx = sigptr[1];
2407 entry->edx = sigptr[2];
2410 case KVM_CPUID_FEATURES:
2411 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2412 (1 << KVM_FEATURE_NOP_IO_DELAY) |
2413 (1 << KVM_FEATURE_CLOCKSOURCE2) |
2414 (1 << KVM_FEATURE_ASYNC_PF) |
2415 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2421 entry->eax = min(entry->eax, 0x8000001a);
2424 entry->edx &= kvm_supported_word1_x86_features;
2425 cpuid_mask(&entry->edx, 1);
2426 entry->ecx &= kvm_supported_word6_x86_features;
2427 cpuid_mask(&entry->ecx, 6);
2431 kvm_x86_ops->set_supported_cpuid(function, entry);
2438 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2439 struct kvm_cpuid_entry2 __user *entries)
2441 struct kvm_cpuid_entry2 *cpuid_entries;
2442 int limit, nent = 0, r = -E2BIG;
2445 if (cpuid->nent < 1)
2447 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2448 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2450 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2454 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2455 limit = cpuid_entries[0].eax;
2456 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2457 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2458 &nent, cpuid->nent);
2460 if (nent >= cpuid->nent)
2463 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2464 limit = cpuid_entries[nent - 1].eax;
2465 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2466 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2467 &nent, cpuid->nent);
2472 if (nent >= cpuid->nent)
2475 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2479 if (nent >= cpuid->nent)
2482 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2486 if (nent >= cpuid->nent)
2490 if (copy_to_user(entries, cpuid_entries,
2491 nent * sizeof(struct kvm_cpuid_entry2)))
2497 vfree(cpuid_entries);
2502 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2503 struct kvm_lapic_state *s)
2505 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2510 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2511 struct kvm_lapic_state *s)
2513 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2514 kvm_apic_post_state_restore(vcpu);
2515 update_cr8_intercept(vcpu);
2520 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2521 struct kvm_interrupt *irq)
2523 if (irq->irq < 0 || irq->irq >= 256)
2525 if (irqchip_in_kernel(vcpu->kvm))
2528 kvm_queue_interrupt(vcpu, irq->irq, false);
2529 kvm_make_request(KVM_REQ_EVENT, vcpu);
2534 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2536 kvm_inject_nmi(vcpu);
2541 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2542 struct kvm_tpr_access_ctl *tac)
2546 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2550 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2554 unsigned bank_num = mcg_cap & 0xff, bank;
2557 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2559 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2562 vcpu->arch.mcg_cap = mcg_cap;
2563 /* Init IA32_MCG_CTL to all 1s */
2564 if (mcg_cap & MCG_CTL_P)
2565 vcpu->arch.mcg_ctl = ~(u64)0;
2566 /* Init IA32_MCi_CTL to all 1s */
2567 for (bank = 0; bank < bank_num; bank++)
2568 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2573 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2574 struct kvm_x86_mce *mce)
2576 u64 mcg_cap = vcpu->arch.mcg_cap;
2577 unsigned bank_num = mcg_cap & 0xff;
2578 u64 *banks = vcpu->arch.mce_banks;
2580 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2583 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2584 * reporting is disabled
2586 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2587 vcpu->arch.mcg_ctl != ~(u64)0)
2589 banks += 4 * mce->bank;
2591 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2592 * reporting is disabled for the bank
2594 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2596 if (mce->status & MCI_STATUS_UC) {
2597 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2598 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2599 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2602 if (banks[1] & MCI_STATUS_VAL)
2603 mce->status |= MCI_STATUS_OVER;
2604 banks[2] = mce->addr;
2605 banks[3] = mce->misc;
2606 vcpu->arch.mcg_status = mce->mcg_status;
2607 banks[1] = mce->status;
2608 kvm_queue_exception(vcpu, MC_VECTOR);
2609 } else if (!(banks[1] & MCI_STATUS_VAL)
2610 || !(banks[1] & MCI_STATUS_UC)) {
2611 if (banks[1] & MCI_STATUS_VAL)
2612 mce->status |= MCI_STATUS_OVER;
2613 banks[2] = mce->addr;
2614 banks[3] = mce->misc;
2615 banks[1] = mce->status;
2617 banks[1] |= MCI_STATUS_OVER;
2621 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2622 struct kvm_vcpu_events *events)
2624 events->exception.injected =
2625 vcpu->arch.exception.pending &&
2626 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2627 events->exception.nr = vcpu->arch.exception.nr;
2628 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2629 events->exception.pad = 0;
2630 events->exception.error_code = vcpu->arch.exception.error_code;
2632 events->interrupt.injected =
2633 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2634 events->interrupt.nr = vcpu->arch.interrupt.nr;
2635 events->interrupt.soft = 0;
2636 events->interrupt.shadow =
2637 kvm_x86_ops->get_interrupt_shadow(vcpu,
2638 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2640 events->nmi.injected = vcpu->arch.nmi_injected;
2641 events->nmi.pending = vcpu->arch.nmi_pending;
2642 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2643 events->nmi.pad = 0;
2645 events->sipi_vector = vcpu->arch.sipi_vector;
2647 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2648 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2649 | KVM_VCPUEVENT_VALID_SHADOW);
2650 memset(&events->reserved, 0, sizeof(events->reserved));
2653 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2654 struct kvm_vcpu_events *events)
2656 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2657 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2658 | KVM_VCPUEVENT_VALID_SHADOW))
2661 vcpu->arch.exception.pending = events->exception.injected;
2662 vcpu->arch.exception.nr = events->exception.nr;
2663 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2664 vcpu->arch.exception.error_code = events->exception.error_code;
2666 vcpu->arch.interrupt.pending = events->interrupt.injected;
2667 vcpu->arch.interrupt.nr = events->interrupt.nr;
2668 vcpu->arch.interrupt.soft = events->interrupt.soft;
2669 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2670 kvm_x86_ops->set_interrupt_shadow(vcpu,
2671 events->interrupt.shadow);
2673 vcpu->arch.nmi_injected = events->nmi.injected;
2674 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2675 vcpu->arch.nmi_pending = events->nmi.pending;
2676 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2678 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2679 vcpu->arch.sipi_vector = events->sipi_vector;
2681 kvm_make_request(KVM_REQ_EVENT, vcpu);
2686 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2687 struct kvm_debugregs *dbgregs)
2689 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2690 dbgregs->dr6 = vcpu->arch.dr6;
2691 dbgregs->dr7 = vcpu->arch.dr7;
2693 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2696 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2697 struct kvm_debugregs *dbgregs)
2702 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2703 vcpu->arch.dr6 = dbgregs->dr6;
2704 vcpu->arch.dr7 = dbgregs->dr7;
2709 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2710 struct kvm_xsave *guest_xsave)
2713 memcpy(guest_xsave->region,
2714 &vcpu->arch.guest_fpu.state->xsave,
2717 memcpy(guest_xsave->region,
2718 &vcpu->arch.guest_fpu.state->fxsave,
2719 sizeof(struct i387_fxsave_struct));
2720 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2725 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2726 struct kvm_xsave *guest_xsave)
2729 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2732 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2733 guest_xsave->region, xstate_size);
2735 if (xstate_bv & ~XSTATE_FPSSE)
2737 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2738 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2743 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2744 struct kvm_xcrs *guest_xcrs)
2746 if (!cpu_has_xsave) {
2747 guest_xcrs->nr_xcrs = 0;
2751 guest_xcrs->nr_xcrs = 1;
2752 guest_xcrs->flags = 0;
2753 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2754 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2757 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2758 struct kvm_xcrs *guest_xcrs)
2765 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2768 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2769 /* Only support XCR0 currently */
2770 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2771 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2772 guest_xcrs->xcrs[0].value);
2780 long kvm_arch_vcpu_ioctl(struct file *filp,
2781 unsigned int ioctl, unsigned long arg)
2783 struct kvm_vcpu *vcpu = filp->private_data;
2784 void __user *argp = (void __user *)arg;
2787 struct kvm_lapic_state *lapic;
2788 struct kvm_xsave *xsave;
2789 struct kvm_xcrs *xcrs;
2795 case KVM_GET_LAPIC: {
2797 if (!vcpu->arch.apic)
2799 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2804 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
2808 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
2813 case KVM_SET_LAPIC: {
2815 if (!vcpu->arch.apic)
2817 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2822 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
2824 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
2830 case KVM_INTERRUPT: {
2831 struct kvm_interrupt irq;
2834 if (copy_from_user(&irq, argp, sizeof irq))
2836 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2843 r = kvm_vcpu_ioctl_nmi(vcpu);
2849 case KVM_SET_CPUID: {
2850 struct kvm_cpuid __user *cpuid_arg = argp;
2851 struct kvm_cpuid cpuid;
2854 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2856 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2861 case KVM_SET_CPUID2: {
2862 struct kvm_cpuid2 __user *cpuid_arg = argp;
2863 struct kvm_cpuid2 cpuid;
2866 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2868 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2869 cpuid_arg->entries);
2874 case KVM_GET_CPUID2: {
2875 struct kvm_cpuid2 __user *cpuid_arg = argp;
2876 struct kvm_cpuid2 cpuid;
2879 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2881 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2882 cpuid_arg->entries);
2886 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2892 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2895 r = msr_io(vcpu, argp, do_set_msr, 0);
2897 case KVM_TPR_ACCESS_REPORTING: {
2898 struct kvm_tpr_access_ctl tac;
2901 if (copy_from_user(&tac, argp, sizeof tac))
2903 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2907 if (copy_to_user(argp, &tac, sizeof tac))
2912 case KVM_SET_VAPIC_ADDR: {
2913 struct kvm_vapic_addr va;
2916 if (!irqchip_in_kernel(vcpu->kvm))
2919 if (copy_from_user(&va, argp, sizeof va))
2922 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2925 case KVM_X86_SETUP_MCE: {
2929 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2931 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2934 case KVM_X86_SET_MCE: {
2935 struct kvm_x86_mce mce;
2938 if (copy_from_user(&mce, argp, sizeof mce))
2940 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2943 case KVM_GET_VCPU_EVENTS: {
2944 struct kvm_vcpu_events events;
2946 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2949 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2954 case KVM_SET_VCPU_EVENTS: {
2955 struct kvm_vcpu_events events;
2958 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2961 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2964 case KVM_GET_DEBUGREGS: {
2965 struct kvm_debugregs dbgregs;
2967 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2970 if (copy_to_user(argp, &dbgregs,
2971 sizeof(struct kvm_debugregs)))
2976 case KVM_SET_DEBUGREGS: {
2977 struct kvm_debugregs dbgregs;
2980 if (copy_from_user(&dbgregs, argp,
2981 sizeof(struct kvm_debugregs)))
2984 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2987 case KVM_GET_XSAVE: {
2988 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2993 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
2996 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3001 case KVM_SET_XSAVE: {
3002 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3008 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
3011 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3014 case KVM_GET_XCRS: {
3015 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3020 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3023 if (copy_to_user(argp, u.xcrs,
3024 sizeof(struct kvm_xcrs)))
3029 case KVM_SET_XCRS: {
3030 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3036 if (copy_from_user(u.xcrs, argp,
3037 sizeof(struct kvm_xcrs)))
3040 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3051 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3055 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3057 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3061 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3064 kvm->arch.ept_identity_map_addr = ident_addr;
3068 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3069 u32 kvm_nr_mmu_pages)
3071 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3074 mutex_lock(&kvm->slots_lock);
3075 spin_lock(&kvm->mmu_lock);
3077 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3078 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3080 spin_unlock(&kvm->mmu_lock);
3081 mutex_unlock(&kvm->slots_lock);
3085 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3087 return kvm->arch.n_max_mmu_pages;
3090 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3095 switch (chip->chip_id) {
3096 case KVM_IRQCHIP_PIC_MASTER:
3097 memcpy(&chip->chip.pic,
3098 &pic_irqchip(kvm)->pics[0],
3099 sizeof(struct kvm_pic_state));
3101 case KVM_IRQCHIP_PIC_SLAVE:
3102 memcpy(&chip->chip.pic,
3103 &pic_irqchip(kvm)->pics[1],
3104 sizeof(struct kvm_pic_state));
3106 case KVM_IRQCHIP_IOAPIC:
3107 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3116 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3121 switch (chip->chip_id) {
3122 case KVM_IRQCHIP_PIC_MASTER:
3123 spin_lock(&pic_irqchip(kvm)->lock);
3124 memcpy(&pic_irqchip(kvm)->pics[0],
3126 sizeof(struct kvm_pic_state));
3127 spin_unlock(&pic_irqchip(kvm)->lock);
3129 case KVM_IRQCHIP_PIC_SLAVE:
3130 spin_lock(&pic_irqchip(kvm)->lock);
3131 memcpy(&pic_irqchip(kvm)->pics[1],
3133 sizeof(struct kvm_pic_state));
3134 spin_unlock(&pic_irqchip(kvm)->lock);
3136 case KVM_IRQCHIP_IOAPIC:
3137 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3143 kvm_pic_update_irq(pic_irqchip(kvm));
3147 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3151 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3152 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3153 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3157 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3161 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3162 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3163 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3164 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3168 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3172 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3173 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3174 sizeof(ps->channels));
3175 ps->flags = kvm->arch.vpit->pit_state.flags;
3176 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3177 memset(&ps->reserved, 0, sizeof(ps->reserved));
3181 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3183 int r = 0, start = 0;
3184 u32 prev_legacy, cur_legacy;
3185 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3186 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3187 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3188 if (!prev_legacy && cur_legacy)
3190 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3191 sizeof(kvm->arch.vpit->pit_state.channels));
3192 kvm->arch.vpit->pit_state.flags = ps->flags;
3193 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3194 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3198 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3199 struct kvm_reinject_control *control)
3201 if (!kvm->arch.vpit)
3203 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3204 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3205 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3210 * Get (and clear) the dirty memory log for a memory slot.
3212 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3213 struct kvm_dirty_log *log)
3216 struct kvm_memory_slot *memslot;
3218 unsigned long is_dirty = 0;
3220 mutex_lock(&kvm->slots_lock);
3223 if (log->slot >= KVM_MEMORY_SLOTS)
3226 memslot = &kvm->memslots->memslots[log->slot];
3228 if (!memslot->dirty_bitmap)
3231 n = kvm_dirty_bitmap_bytes(memslot);
3233 for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3234 is_dirty = memslot->dirty_bitmap[i];
3236 /* If nothing is dirty, don't bother messing with page tables. */
3238 struct kvm_memslots *slots, *old_slots;
3239 unsigned long *dirty_bitmap;
3241 dirty_bitmap = memslot->dirty_bitmap_head;
3242 if (memslot->dirty_bitmap == dirty_bitmap)
3243 dirty_bitmap += n / sizeof(long);
3244 memset(dirty_bitmap, 0, n);
3247 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3250 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3251 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3252 slots->generation++;
3254 old_slots = kvm->memslots;
3255 rcu_assign_pointer(kvm->memslots, slots);
3256 synchronize_srcu_expedited(&kvm->srcu);
3257 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3260 spin_lock(&kvm->mmu_lock);
3261 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3262 spin_unlock(&kvm->mmu_lock);
3265 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
3269 if (clear_user(log->dirty_bitmap, n))
3275 mutex_unlock(&kvm->slots_lock);
3279 long kvm_arch_vm_ioctl(struct file *filp,
3280 unsigned int ioctl, unsigned long arg)
3282 struct kvm *kvm = filp->private_data;
3283 void __user *argp = (void __user *)arg;
3286 * This union makes it completely explicit to gcc-3.x
3287 * that these two variables' stack usage should be
3288 * combined, not added together.
3291 struct kvm_pit_state ps;
3292 struct kvm_pit_state2 ps2;
3293 struct kvm_pit_config pit_config;
3297 case KVM_SET_TSS_ADDR:
3298 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3302 case KVM_SET_IDENTITY_MAP_ADDR: {
3306 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3308 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3313 case KVM_SET_NR_MMU_PAGES:
3314 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3318 case KVM_GET_NR_MMU_PAGES:
3319 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3321 case KVM_CREATE_IRQCHIP: {
3322 struct kvm_pic *vpic;
3324 mutex_lock(&kvm->lock);
3327 goto create_irqchip_unlock;
3329 vpic = kvm_create_pic(kvm);
3331 r = kvm_ioapic_init(kvm);
3333 mutex_lock(&kvm->slots_lock);
3334 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3336 mutex_unlock(&kvm->slots_lock);
3338 goto create_irqchip_unlock;
3341 goto create_irqchip_unlock;
3343 kvm->arch.vpic = vpic;
3345 r = kvm_setup_default_irq_routing(kvm);
3347 mutex_lock(&kvm->slots_lock);
3348 mutex_lock(&kvm->irq_lock);
3349 kvm_ioapic_destroy(kvm);
3350 kvm_destroy_pic(kvm);
3351 mutex_unlock(&kvm->irq_lock);
3352 mutex_unlock(&kvm->slots_lock);
3354 create_irqchip_unlock:
3355 mutex_unlock(&kvm->lock);
3358 case KVM_CREATE_PIT:
3359 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3361 case KVM_CREATE_PIT2:
3363 if (copy_from_user(&u.pit_config, argp,
3364 sizeof(struct kvm_pit_config)))
3367 mutex_lock(&kvm->slots_lock);
3370 goto create_pit_unlock;
3372 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3376 mutex_unlock(&kvm->slots_lock);
3378 case KVM_IRQ_LINE_STATUS:
3379 case KVM_IRQ_LINE: {
3380 struct kvm_irq_level irq_event;
3383 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3386 if (irqchip_in_kernel(kvm)) {
3388 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3389 irq_event.irq, irq_event.level);
3390 if (ioctl == KVM_IRQ_LINE_STATUS) {
3392 irq_event.status = status;
3393 if (copy_to_user(argp, &irq_event,
3401 case KVM_GET_IRQCHIP: {
3402 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3403 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3409 if (copy_from_user(chip, argp, sizeof *chip))
3410 goto get_irqchip_out;
3412 if (!irqchip_in_kernel(kvm))
3413 goto get_irqchip_out;
3414 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3416 goto get_irqchip_out;
3418 if (copy_to_user(argp, chip, sizeof *chip))
3419 goto get_irqchip_out;
3427 case KVM_SET_IRQCHIP: {
3428 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3429 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3435 if (copy_from_user(chip, argp, sizeof *chip))
3436 goto set_irqchip_out;
3438 if (!irqchip_in_kernel(kvm))
3439 goto set_irqchip_out;
3440 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3442 goto set_irqchip_out;
3452 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3455 if (!kvm->arch.vpit)
3457 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3461 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3468 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3471 if (!kvm->arch.vpit)
3473 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3479 case KVM_GET_PIT2: {
3481 if (!kvm->arch.vpit)
3483 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3487 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3492 case KVM_SET_PIT2: {
3494 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3497 if (!kvm->arch.vpit)
3499 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3505 case KVM_REINJECT_CONTROL: {
3506 struct kvm_reinject_control control;
3508 if (copy_from_user(&control, argp, sizeof(control)))
3510 r = kvm_vm_ioctl_reinject(kvm, &control);
3516 case KVM_XEN_HVM_CONFIG: {
3518 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3519 sizeof(struct kvm_xen_hvm_config)))
3522 if (kvm->arch.xen_hvm_config.flags)
3527 case KVM_SET_CLOCK: {
3528 struct kvm_clock_data user_ns;
3533 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3541 local_irq_disable();
3542 now_ns = get_kernel_ns();
3543 delta = user_ns.clock - now_ns;
3545 kvm->arch.kvmclock_offset = delta;
3548 case KVM_GET_CLOCK: {
3549 struct kvm_clock_data user_ns;
3552 local_irq_disable();
3553 now_ns = get_kernel_ns();
3554 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3557 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3560 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3573 static void kvm_init_msr_list(void)
3578 /* skip the first msrs in the list. KVM-specific */
3579 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3580 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3583 msrs_to_save[j] = msrs_to_save[i];
3586 num_msrs_to_save = j;
3589 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3597 if (!(vcpu->arch.apic &&
3598 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3599 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3610 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3617 if (!(vcpu->arch.apic &&
3618 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3619 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3621 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3631 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3632 struct kvm_segment *var, int seg)
3634 kvm_x86_ops->set_segment(vcpu, var, seg);
3637 void kvm_get_segment(struct kvm_vcpu *vcpu,
3638 struct kvm_segment *var, int seg)
3640 kvm_x86_ops->get_segment(vcpu, var, seg);
3643 static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3648 static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3651 struct x86_exception exception;
3653 BUG_ON(!mmu_is_nested(vcpu));
3655 /* NPT walks are always user-walks */
3656 access |= PFERR_USER_MASK;
3657 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3662 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3663 struct x86_exception *exception)
3665 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3666 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3669 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3670 struct x86_exception *exception)
3672 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3673 access |= PFERR_FETCH_MASK;
3674 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3677 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3678 struct x86_exception *exception)
3680 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3681 access |= PFERR_WRITE_MASK;
3682 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3685 /* uses this to access any guest's mapped memory without checking CPL */
3686 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3687 struct x86_exception *exception)
3689 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3692 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3693 struct kvm_vcpu *vcpu, u32 access,
3694 struct x86_exception *exception)
3697 int r = X86EMUL_CONTINUE;
3700 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3702 unsigned offset = addr & (PAGE_SIZE-1);
3703 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3706 if (gpa == UNMAPPED_GVA)
3707 return X86EMUL_PROPAGATE_FAULT;
3708 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3710 r = X86EMUL_IO_NEEDED;
3722 /* used for instruction fetching */
3723 static int kvm_fetch_guest_virt(gva_t addr, void *val, unsigned int bytes,
3724 struct kvm_vcpu *vcpu,
3725 struct x86_exception *exception)
3727 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3728 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3729 access | PFERR_FETCH_MASK,
3733 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3734 struct kvm_vcpu *vcpu,
3735 struct x86_exception *exception)
3737 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3738 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3742 static int kvm_read_guest_virt_system(gva_t addr, void *val, unsigned int bytes,
3743 struct kvm_vcpu *vcpu,
3744 struct x86_exception *exception)
3746 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3749 static int kvm_write_guest_virt_system(gva_t addr, void *val,
3751 struct kvm_vcpu *vcpu,
3752 struct x86_exception *exception)
3755 int r = X86EMUL_CONTINUE;
3758 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3761 unsigned offset = addr & (PAGE_SIZE-1);
3762 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3765 if (gpa == UNMAPPED_GVA)
3766 return X86EMUL_PROPAGATE_FAULT;
3767 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3769 r = X86EMUL_IO_NEEDED;
3781 static int emulator_read_emulated(unsigned long addr,
3784 struct x86_exception *exception,
3785 struct kvm_vcpu *vcpu)
3790 if (vcpu->mmio_read_completed) {
3791 memcpy(val, vcpu->mmio_data, bytes);
3792 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3793 vcpu->mmio_phys_addr, *(u64 *)val);
3794 vcpu->mmio_read_completed = 0;
3795 return X86EMUL_CONTINUE;
3798 gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, exception);
3800 if (gpa == UNMAPPED_GVA)
3801 return X86EMUL_PROPAGATE_FAULT;
3803 /* For APIC access vmexit */
3804 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3807 if (kvm_read_guest_virt(addr, val, bytes, vcpu, exception)
3808 == X86EMUL_CONTINUE)
3809 return X86EMUL_CONTINUE;
3813 * Is this MMIO handled locally?
3815 handled = vcpu_mmio_read(vcpu, gpa, bytes, val);
3817 if (handled == bytes)
3818 return X86EMUL_CONTINUE;
3824 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3826 vcpu->mmio_needed = 1;
3827 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3828 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3829 vcpu->mmio_size = bytes;
3830 vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
3831 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 0;
3832 vcpu->mmio_index = 0;
3834 return X86EMUL_IO_NEEDED;
3837 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3838 const void *val, int bytes)
3842 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3845 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3849 static int emulator_write_emulated_onepage(unsigned long addr,
3852 struct x86_exception *exception,
3853 struct kvm_vcpu *vcpu)
3858 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, exception);
3860 if (gpa == UNMAPPED_GVA)
3861 return X86EMUL_PROPAGATE_FAULT;
3863 /* For APIC access vmexit */
3864 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3867 if (emulator_write_phys(vcpu, gpa, val, bytes))
3868 return X86EMUL_CONTINUE;
3871 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3873 * Is this MMIO handled locally?
3875 handled = vcpu_mmio_write(vcpu, gpa, bytes, val);
3876 if (handled == bytes)
3877 return X86EMUL_CONTINUE;
3883 vcpu->mmio_needed = 1;
3884 memcpy(vcpu->mmio_data, val, bytes);
3885 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3886 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3887 vcpu->mmio_size = bytes;
3888 vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
3889 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 1;
3890 memcpy(vcpu->run->mmio.data, vcpu->mmio_data, 8);
3891 vcpu->mmio_index = 0;
3893 return X86EMUL_CONTINUE;
3896 int emulator_write_emulated(unsigned long addr,
3899 struct x86_exception *exception,
3900 struct kvm_vcpu *vcpu)
3902 /* Crossing a page boundary? */
3903 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3906 now = -addr & ~PAGE_MASK;
3907 rc = emulator_write_emulated_onepage(addr, val, now, exception,
3909 if (rc != X86EMUL_CONTINUE)
3915 return emulator_write_emulated_onepage(addr, val, bytes, exception,
3919 #define CMPXCHG_TYPE(t, ptr, old, new) \
3920 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3922 #ifdef CONFIG_X86_64
3923 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3925 # define CMPXCHG64(ptr, old, new) \
3926 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
3929 static int emulator_cmpxchg_emulated(unsigned long addr,
3933 struct x86_exception *exception,
3934 struct kvm_vcpu *vcpu)
3941 /* guests cmpxchg8b have to be emulated atomically */
3942 if (bytes > 8 || (bytes & (bytes - 1)))
3945 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3947 if (gpa == UNMAPPED_GVA ||
3948 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3951 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3954 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3955 if (is_error_page(page)) {
3956 kvm_release_page_clean(page);
3960 kaddr = kmap_atomic(page, KM_USER0);
3961 kaddr += offset_in_page(gpa);
3964 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
3967 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
3970 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
3973 exchanged = CMPXCHG64(kaddr, old, new);
3978 kunmap_atomic(kaddr, KM_USER0);
3979 kvm_release_page_dirty(page);
3982 return X86EMUL_CMPXCHG_FAILED;
3984 kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
3986 return X86EMUL_CONTINUE;
3989 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3991 return emulator_write_emulated(addr, new, bytes, exception, vcpu);
3994 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3996 /* TODO: String I/O for in kernel device */
3999 if (vcpu->arch.pio.in)
4000 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4001 vcpu->arch.pio.size, pd);
4003 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4004 vcpu->arch.pio.port, vcpu->arch.pio.size,
4010 static int emulator_pio_in_emulated(int size, unsigned short port, void *val,
4011 unsigned int count, struct kvm_vcpu *vcpu)
4013 if (vcpu->arch.pio.count)
4016 trace_kvm_pio(0, port, size, count);
4018 vcpu->arch.pio.port = port;
4019 vcpu->arch.pio.in = 1;
4020 vcpu->arch.pio.count = count;
4021 vcpu->arch.pio.size = size;
4023 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4025 memcpy(val, vcpu->arch.pio_data, size * count);
4026 vcpu->arch.pio.count = 0;
4030 vcpu->run->exit_reason = KVM_EXIT_IO;
4031 vcpu->run->io.direction = KVM_EXIT_IO_IN;
4032 vcpu->run->io.size = size;
4033 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4034 vcpu->run->io.count = count;
4035 vcpu->run->io.port = port;
4040 static int emulator_pio_out_emulated(int size, unsigned short port,
4041 const void *val, unsigned int count,
4042 struct kvm_vcpu *vcpu)
4044 trace_kvm_pio(1, port, size, count);
4046 vcpu->arch.pio.port = port;
4047 vcpu->arch.pio.in = 0;
4048 vcpu->arch.pio.count = count;
4049 vcpu->arch.pio.size = size;
4051 memcpy(vcpu->arch.pio_data, val, size * count);
4053 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4054 vcpu->arch.pio.count = 0;
4058 vcpu->run->exit_reason = KVM_EXIT_IO;
4059 vcpu->run->io.direction = KVM_EXIT_IO_OUT;
4060 vcpu->run->io.size = size;
4061 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4062 vcpu->run->io.count = count;
4063 vcpu->run->io.port = port;
4068 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4070 return kvm_x86_ops->get_segment_base(vcpu, seg);
4073 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
4075 kvm_mmu_invlpg(vcpu, address);
4076 return X86EMUL_CONTINUE;
4079 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4081 if (!need_emulate_wbinvd(vcpu))
4082 return X86EMUL_CONTINUE;
4084 if (kvm_x86_ops->has_wbinvd_exit()) {
4085 int cpu = get_cpu();
4087 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4088 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4089 wbinvd_ipi, NULL, 1);
4091 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4094 return X86EMUL_CONTINUE;
4096 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4098 int emulate_clts(struct kvm_vcpu *vcpu)
4100 kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4101 kvm_x86_ops->fpu_activate(vcpu);
4102 return X86EMUL_CONTINUE;
4105 int emulator_get_dr(int dr, unsigned long *dest, struct kvm_vcpu *vcpu)
4107 return _kvm_get_dr(vcpu, dr, dest);
4110 int emulator_set_dr(int dr, unsigned long value, struct kvm_vcpu *vcpu)
4113 return __kvm_set_dr(vcpu, dr, value);
4116 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4118 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4121 static unsigned long emulator_get_cr(int cr, struct kvm_vcpu *vcpu)
4123 unsigned long value;
4127 value = kvm_read_cr0(vcpu);
4130 value = vcpu->arch.cr2;
4133 value = kvm_read_cr3(vcpu);
4136 value = kvm_read_cr4(vcpu);
4139 value = kvm_get_cr8(vcpu);
4142 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4149 static int emulator_set_cr(int cr, unsigned long val, struct kvm_vcpu *vcpu)
4155 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4158 vcpu->arch.cr2 = val;
4161 res = kvm_set_cr3(vcpu, val);
4164 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4167 res = kvm_set_cr8(vcpu, val);
4170 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4177 static int emulator_get_cpl(struct kvm_vcpu *vcpu)
4179 return kvm_x86_ops->get_cpl(vcpu);
4182 static void emulator_get_gdt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
4184 kvm_x86_ops->get_gdt(vcpu, dt);
4187 static void emulator_get_idt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
4189 kvm_x86_ops->get_idt(vcpu, dt);
4192 static unsigned long emulator_get_cached_segment_base(int seg,
4193 struct kvm_vcpu *vcpu)
4195 return get_segment_base(vcpu, seg);
4198 static bool emulator_get_cached_descriptor(struct desc_struct *desc, u32 *base3,
4199 int seg, struct kvm_vcpu *vcpu)
4201 struct kvm_segment var;
4203 kvm_get_segment(vcpu, &var, seg);
4210 set_desc_limit(desc, var.limit);
4211 set_desc_base(desc, (unsigned long)var.base);
4212 #ifdef CONFIG_X86_64
4214 *base3 = var.base >> 32;
4216 desc->type = var.type;
4218 desc->dpl = var.dpl;
4219 desc->p = var.present;
4220 desc->avl = var.avl;
4228 static void emulator_set_cached_descriptor(struct desc_struct *desc, u32 base3,
4229 int seg, struct kvm_vcpu *vcpu)
4231 struct kvm_segment var;
4233 /* needed to preserve selector */
4234 kvm_get_segment(vcpu, &var, seg);
4236 var.base = get_desc_base(desc);
4237 #ifdef CONFIG_X86_64
4238 var.base |= ((u64)base3) << 32;
4240 var.limit = get_desc_limit(desc);
4242 var.limit = (var.limit << 12) | 0xfff;
4243 var.type = desc->type;
4244 var.present = desc->p;
4245 var.dpl = desc->dpl;
4250 var.avl = desc->avl;
4251 var.present = desc->p;
4252 var.unusable = !var.present;
4255 kvm_set_segment(vcpu, &var, seg);
4259 static u16 emulator_get_segment_selector(int seg, struct kvm_vcpu *vcpu)
4261 struct kvm_segment kvm_seg;
4263 kvm_get_segment(vcpu, &kvm_seg, seg);
4264 return kvm_seg.selector;
4267 static void emulator_set_segment_selector(u16 sel, int seg,
4268 struct kvm_vcpu *vcpu)
4270 struct kvm_segment kvm_seg;
4272 kvm_get_segment(vcpu, &kvm_seg, seg);
4273 kvm_seg.selector = sel;
4274 kvm_set_segment(vcpu, &kvm_seg, seg);
4277 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4280 kvm_load_guest_fpu(ctxt->vcpu);
4282 * CR0.TS may reference the host fpu state, not the guest fpu state,
4283 * so it may be clear at this point.
4288 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4293 static int emulator_intercept(struct kvm_vcpu *vcpu,
4294 struct x86_instruction_info *info,
4295 enum x86_intercept_stage stage)
4297 return kvm_x86_ops->check_intercept(vcpu, info, stage);
4300 static struct x86_emulate_ops emulate_ops = {
4301 .read_std = kvm_read_guest_virt_system,
4302 .write_std = kvm_write_guest_virt_system,
4303 .fetch = kvm_fetch_guest_virt,
4304 .read_emulated = emulator_read_emulated,
4305 .write_emulated = emulator_write_emulated,
4306 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4307 .pio_in_emulated = emulator_pio_in_emulated,
4308 .pio_out_emulated = emulator_pio_out_emulated,
4309 .get_cached_descriptor = emulator_get_cached_descriptor,
4310 .set_cached_descriptor = emulator_set_cached_descriptor,
4311 .get_segment_selector = emulator_get_segment_selector,
4312 .set_segment_selector = emulator_set_segment_selector,
4313 .get_cached_segment_base = emulator_get_cached_segment_base,
4314 .get_gdt = emulator_get_gdt,
4315 .get_idt = emulator_get_idt,
4316 .get_cr = emulator_get_cr,
4317 .set_cr = emulator_set_cr,
4318 .cpl = emulator_get_cpl,
4319 .get_dr = emulator_get_dr,
4320 .set_dr = emulator_set_dr,
4321 .set_msr = kvm_set_msr,
4322 .get_msr = kvm_get_msr,
4323 .get_fpu = emulator_get_fpu,
4324 .put_fpu = emulator_put_fpu,
4325 .intercept = emulator_intercept,
4328 static void cache_all_regs(struct kvm_vcpu *vcpu)
4330 kvm_register_read(vcpu, VCPU_REGS_RAX);
4331 kvm_register_read(vcpu, VCPU_REGS_RSP);
4332 kvm_register_read(vcpu, VCPU_REGS_RIP);
4333 vcpu->arch.regs_dirty = ~0;
4336 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4338 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4340 * an sti; sti; sequence only disable interrupts for the first
4341 * instruction. So, if the last instruction, be it emulated or
4342 * not, left the system with the INT_STI flag enabled, it
4343 * means that the last instruction is an sti. We should not
4344 * leave the flag on in this case. The same goes for mov ss
4346 if (!(int_shadow & mask))
4347 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4350 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4352 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4353 if (ctxt->exception.vector == PF_VECTOR)
4354 kvm_propagate_fault(vcpu, &ctxt->exception);
4355 else if (ctxt->exception.error_code_valid)
4356 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4357 ctxt->exception.error_code);
4359 kvm_queue_exception(vcpu, ctxt->exception.vector);
4362 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4364 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4367 cache_all_regs(vcpu);
4369 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4371 vcpu->arch.emulate_ctxt.vcpu = vcpu;
4372 vcpu->arch.emulate_ctxt.eflags = kvm_get_rflags(vcpu);
4373 vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
4374 vcpu->arch.emulate_ctxt.mode =
4375 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4376 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
4377 ? X86EMUL_MODE_VM86 : cs_l
4378 ? X86EMUL_MODE_PROT64 : cs_db
4379 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
4380 vcpu->arch.emulate_ctxt.guest_mode = is_guest_mode(vcpu);
4381 memset(c, 0, sizeof(struct decode_cache));
4382 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4385 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq)
4387 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4390 init_emulate_ctxt(vcpu);
4392 vcpu->arch.emulate_ctxt.decode.op_bytes = 2;
4393 vcpu->arch.emulate_ctxt.decode.ad_bytes = 2;
4394 vcpu->arch.emulate_ctxt.decode.eip = vcpu->arch.emulate_ctxt.eip;
4395 ret = emulate_int_real(&vcpu->arch.emulate_ctxt, &emulate_ops, irq);
4397 if (ret != X86EMUL_CONTINUE)
4398 return EMULATE_FAIL;
4400 vcpu->arch.emulate_ctxt.eip = c->eip;
4401 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4402 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4403 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4405 if (irq == NMI_VECTOR)
4406 vcpu->arch.nmi_pending = false;
4408 vcpu->arch.interrupt.pending = false;
4410 return EMULATE_DONE;
4412 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4414 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4416 int r = EMULATE_DONE;
4418 ++vcpu->stat.insn_emulation_fail;
4419 trace_kvm_emulate_insn_failed(vcpu);
4420 if (!is_guest_mode(vcpu)) {
4421 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4422 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4423 vcpu->run->internal.ndata = 0;
4426 kvm_queue_exception(vcpu, UD_VECTOR);
4431 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4439 * if emulation was due to access to shadowed page table
4440 * and it failed try to unshadow page and re-entetr the
4441 * guest to let CPU execute the instruction.
4443 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4446 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4448 if (gpa == UNMAPPED_GVA)
4449 return true; /* let cpu generate fault */
4451 if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4457 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4464 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4466 kvm_clear_exception_queue(vcpu);
4467 vcpu->arch.mmio_fault_cr2 = cr2;
4469 * TODO: fix emulate.c to use guest_read/write_register
4470 * instead of direct ->regs accesses, can save hundred cycles
4471 * on Intel for instructions that don't read/change RSP, for
4474 cache_all_regs(vcpu);
4476 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4477 init_emulate_ctxt(vcpu);
4478 vcpu->arch.emulate_ctxt.interruptibility = 0;
4479 vcpu->arch.emulate_ctxt.have_exception = false;
4480 vcpu->arch.emulate_ctxt.perm_ok = false;
4482 vcpu->arch.emulate_ctxt.only_vendor_specific_insn
4483 = emulation_type & EMULTYPE_TRAP_UD;
4485 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, insn, insn_len);
4487 trace_kvm_emulate_insn_start(vcpu);
4488 ++vcpu->stat.insn_emulation;
4490 if (emulation_type & EMULTYPE_TRAP_UD)
4491 return EMULATE_FAIL;
4492 if (reexecute_instruction(vcpu, cr2))
4493 return EMULATE_DONE;
4494 if (emulation_type & EMULTYPE_SKIP)
4495 return EMULATE_FAIL;
4496 return handle_emulation_failure(vcpu);
4500 if (emulation_type & EMULTYPE_SKIP) {
4501 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
4502 return EMULATE_DONE;
4505 /* this is needed for vmware backdor interface to work since it
4506 changes registers values during IO operation */
4507 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4510 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt);
4512 if (r == EMULATION_INTERCEPTED)
4513 return EMULATE_DONE;
4515 if (r == EMULATION_FAILED) {
4516 if (reexecute_instruction(vcpu, cr2))
4517 return EMULATE_DONE;
4519 return handle_emulation_failure(vcpu);
4522 if (vcpu->arch.emulate_ctxt.have_exception) {
4523 inject_emulated_exception(vcpu);
4525 } else if (vcpu->arch.pio.count) {
4526 if (!vcpu->arch.pio.in)
4527 vcpu->arch.pio.count = 0;
4528 r = EMULATE_DO_MMIO;
4529 } else if (vcpu->mmio_needed)
4530 r = EMULATE_DO_MMIO;
4531 else if (r == EMULATION_RESTART)
4536 toggle_interruptibility(vcpu, vcpu->arch.emulate_ctxt.interruptibility);
4537 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4538 kvm_make_request(KVM_REQ_EVENT, vcpu);
4539 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4540 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4544 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4546 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4548 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4549 int ret = emulator_pio_out_emulated(size, port, &val, 1, vcpu);
4550 /* do not return to emulator after return from userspace */
4551 vcpu->arch.pio.count = 0;
4554 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4556 static void tsc_bad(void *info)
4558 __this_cpu_write(cpu_tsc_khz, 0);
4561 static void tsc_khz_changed(void *data)
4563 struct cpufreq_freqs *freq = data;
4564 unsigned long khz = 0;
4568 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4569 khz = cpufreq_quick_get(raw_smp_processor_id());
4572 __this_cpu_write(cpu_tsc_khz, khz);
4575 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4578 struct cpufreq_freqs *freq = data;
4580 struct kvm_vcpu *vcpu;
4581 int i, send_ipi = 0;
4584 * We allow guests to temporarily run on slowing clocks,
4585 * provided we notify them after, or to run on accelerating
4586 * clocks, provided we notify them before. Thus time never
4589 * However, we have a problem. We can't atomically update
4590 * the frequency of a given CPU from this function; it is
4591 * merely a notifier, which can be called from any CPU.
4592 * Changing the TSC frequency at arbitrary points in time
4593 * requires a recomputation of local variables related to
4594 * the TSC for each VCPU. We must flag these local variables
4595 * to be updated and be sure the update takes place with the
4596 * new frequency before any guests proceed.
4598 * Unfortunately, the combination of hotplug CPU and frequency
4599 * change creates an intractable locking scenario; the order
4600 * of when these callouts happen is undefined with respect to
4601 * CPU hotplug, and they can race with each other. As such,
4602 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4603 * undefined; you can actually have a CPU frequency change take
4604 * place in between the computation of X and the setting of the
4605 * variable. To protect against this problem, all updates of
4606 * the per_cpu tsc_khz variable are done in an interrupt
4607 * protected IPI, and all callers wishing to update the value
4608 * must wait for a synchronous IPI to complete (which is trivial
4609 * if the caller is on the CPU already). This establishes the
4610 * necessary total order on variable updates.
4612 * Note that because a guest time update may take place
4613 * anytime after the setting of the VCPU's request bit, the
4614 * correct TSC value must be set before the request. However,
4615 * to ensure the update actually makes it to any guest which
4616 * starts running in hardware virtualization between the set
4617 * and the acquisition of the spinlock, we must also ping the
4618 * CPU after setting the request bit.
4622 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4624 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4627 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4629 raw_spin_lock(&kvm_lock);
4630 list_for_each_entry(kvm, &vm_list, vm_list) {
4631 kvm_for_each_vcpu(i, vcpu, kvm) {
4632 if (vcpu->cpu != freq->cpu)
4634 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4635 if (vcpu->cpu != smp_processor_id())
4639 raw_spin_unlock(&kvm_lock);
4641 if (freq->old < freq->new && send_ipi) {
4643 * We upscale the frequency. Must make the guest
4644 * doesn't see old kvmclock values while running with
4645 * the new frequency, otherwise we risk the guest sees
4646 * time go backwards.
4648 * In case we update the frequency for another cpu
4649 * (which might be in guest context) send an interrupt
4650 * to kick the cpu out of guest context. Next time
4651 * guest context is entered kvmclock will be updated,
4652 * so the guest will not see stale values.
4654 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4659 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4660 .notifier_call = kvmclock_cpufreq_notifier
4663 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4664 unsigned long action, void *hcpu)
4666 unsigned int cpu = (unsigned long)hcpu;
4670 case CPU_DOWN_FAILED:
4671 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4673 case CPU_DOWN_PREPARE:
4674 smp_call_function_single(cpu, tsc_bad, NULL, 1);
4680 static struct notifier_block kvmclock_cpu_notifier_block = {
4681 .notifier_call = kvmclock_cpu_notifier,
4682 .priority = -INT_MAX
4685 static void kvm_timer_init(void)
4689 max_tsc_khz = tsc_khz;
4690 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4691 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4692 #ifdef CONFIG_CPU_FREQ
4693 struct cpufreq_policy policy;
4694 memset(&policy, 0, sizeof(policy));
4696 cpufreq_get_policy(&policy, cpu);
4697 if (policy.cpuinfo.max_freq)
4698 max_tsc_khz = policy.cpuinfo.max_freq;
4701 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4702 CPUFREQ_TRANSITION_NOTIFIER);
4704 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
4705 for_each_online_cpu(cpu)
4706 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4709 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4711 static int kvm_is_in_guest(void)
4713 return percpu_read(current_vcpu) != NULL;
4716 static int kvm_is_user_mode(void)
4720 if (percpu_read(current_vcpu))
4721 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
4723 return user_mode != 0;
4726 static unsigned long kvm_get_guest_ip(void)
4728 unsigned long ip = 0;
4730 if (percpu_read(current_vcpu))
4731 ip = kvm_rip_read(percpu_read(current_vcpu));
4736 static struct perf_guest_info_callbacks kvm_guest_cbs = {
4737 .is_in_guest = kvm_is_in_guest,
4738 .is_user_mode = kvm_is_user_mode,
4739 .get_guest_ip = kvm_get_guest_ip,
4742 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4744 percpu_write(current_vcpu, vcpu);
4746 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4748 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4750 percpu_write(current_vcpu, NULL);
4752 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4754 int kvm_arch_init(void *opaque)
4757 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4760 printk(KERN_ERR "kvm: already loaded the other module\n");
4765 if (!ops->cpu_has_kvm_support()) {
4766 printk(KERN_ERR "kvm: no hardware support\n");
4770 if (ops->disabled_by_bios()) {
4771 printk(KERN_ERR "kvm: disabled by bios\n");
4776 r = kvm_mmu_module_init();
4780 kvm_init_msr_list();
4783 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
4784 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4785 PT_DIRTY_MASK, PT64_NX_MASK, 0);
4789 perf_register_guest_info_callbacks(&kvm_guest_cbs);
4792 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
4800 void kvm_arch_exit(void)
4802 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4804 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4805 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4806 CPUFREQ_TRANSITION_NOTIFIER);
4807 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4809 kvm_mmu_module_exit();
4812 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4814 ++vcpu->stat.halt_exits;
4815 if (irqchip_in_kernel(vcpu->kvm)) {
4816 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
4819 vcpu->run->exit_reason = KVM_EXIT_HLT;
4823 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4825 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
4828 if (is_long_mode(vcpu))
4831 return a0 | ((gpa_t)a1 << 32);
4834 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4836 u64 param, ingpa, outgpa, ret;
4837 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4838 bool fast, longmode;
4842 * hypercall generates UD from non zero cpl and real mode
4845 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
4846 kvm_queue_exception(vcpu, UD_VECTOR);
4850 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4851 longmode = is_long_mode(vcpu) && cs_l == 1;
4854 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4855 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4856 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4857 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4858 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4859 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
4861 #ifdef CONFIG_X86_64
4863 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4864 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4865 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4869 code = param & 0xffff;
4870 fast = (param >> 16) & 0x1;
4871 rep_cnt = (param >> 32) & 0xfff;
4872 rep_idx = (param >> 48) & 0xfff;
4874 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
4877 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
4878 kvm_vcpu_on_spin(vcpu);
4881 res = HV_STATUS_INVALID_HYPERCALL_CODE;
4885 ret = res | (((u64)rep_done & 0xfff) << 32);
4887 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4889 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
4890 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
4896 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
4898 unsigned long nr, a0, a1, a2, a3, ret;
4901 if (kvm_hv_hypercall_enabled(vcpu->kvm))
4902 return kvm_hv_hypercall(vcpu);
4904 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
4905 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
4906 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
4907 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
4908 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
4910 trace_kvm_hypercall(nr, a0, a1, a2, a3);
4912 if (!is_long_mode(vcpu)) {
4920 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
4926 case KVM_HC_VAPIC_POLL_IRQ:
4930 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
4937 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4938 ++vcpu->stat.hypercalls;
4941 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
4943 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
4945 char instruction[3];
4946 unsigned long rip = kvm_rip_read(vcpu);
4949 * Blow out the MMU to ensure that no other VCPU has an active mapping
4950 * to ensure that the updated hypercall appears atomically across all
4953 kvm_mmu_zap_all(vcpu->kvm);
4955 kvm_x86_ops->patch_hypercall(vcpu, instruction);
4957 return emulator_write_emulated(rip, instruction, 3, NULL, vcpu);
4960 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4962 struct desc_ptr dt = { limit, base };
4964 kvm_x86_ops->set_gdt(vcpu, &dt);
4967 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4969 struct desc_ptr dt = { limit, base };
4971 kvm_x86_ops->set_idt(vcpu, &dt);
4974 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
4976 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
4977 int j, nent = vcpu->arch.cpuid_nent;
4979 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
4980 /* when no next entry is found, the current entry[i] is reselected */
4981 for (j = i + 1; ; j = (j + 1) % nent) {
4982 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
4983 if (ej->function == e->function) {
4984 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
4988 return 0; /* silence gcc, even though control never reaches here */
4991 /* find an entry with matching function, matching index (if needed), and that
4992 * should be read next (if it's stateful) */
4993 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
4994 u32 function, u32 index)
4996 if (e->function != function)
4998 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
5000 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
5001 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
5006 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
5007 u32 function, u32 index)
5010 struct kvm_cpuid_entry2 *best = NULL;
5012 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
5013 struct kvm_cpuid_entry2 *e;
5015 e = &vcpu->arch.cpuid_entries[i];
5016 if (is_matching_cpuid_entry(e, function, index)) {
5017 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
5018 move_to_next_stateful_cpuid_entry(vcpu, i);
5025 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
5027 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
5029 struct kvm_cpuid_entry2 *best;
5031 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
5032 if (!best || best->eax < 0x80000008)
5034 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
5036 return best->eax & 0xff;
5042 * If no match is found, check whether we exceed the vCPU's limit
5043 * and return the content of the highest valid _standard_ leaf instead.
5044 * This is to satisfy the CPUID specification.
5046 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
5047 u32 function, u32 index)
5049 struct kvm_cpuid_entry2 *maxlevel;
5051 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
5052 if (!maxlevel || maxlevel->eax >= function)
5054 if (function & 0x80000000) {
5055 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
5059 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
5062 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
5064 u32 function, index;
5065 struct kvm_cpuid_entry2 *best;
5067 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
5068 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5069 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
5070 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
5071 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
5072 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
5073 best = kvm_find_cpuid_entry(vcpu, function, index);
5076 best = check_cpuid_limit(vcpu, function, index);
5079 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
5080 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
5081 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
5082 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
5084 kvm_x86_ops->skip_emulated_instruction(vcpu);
5085 trace_kvm_cpuid(function,
5086 kvm_register_read(vcpu, VCPU_REGS_RAX),
5087 kvm_register_read(vcpu, VCPU_REGS_RBX),
5088 kvm_register_read(vcpu, VCPU_REGS_RCX),
5089 kvm_register_read(vcpu, VCPU_REGS_RDX));
5091 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
5094 * Check if userspace requested an interrupt window, and that the
5095 * interrupt window is open.
5097 * No need to exit to userspace if we already have an interrupt queued.
5099 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5101 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5102 vcpu->run->request_interrupt_window &&
5103 kvm_arch_interrupt_allowed(vcpu));
5106 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5108 struct kvm_run *kvm_run = vcpu->run;
5110 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5111 kvm_run->cr8 = kvm_get_cr8(vcpu);
5112 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5113 if (irqchip_in_kernel(vcpu->kvm))
5114 kvm_run->ready_for_interrupt_injection = 1;
5116 kvm_run->ready_for_interrupt_injection =
5117 kvm_arch_interrupt_allowed(vcpu) &&
5118 !kvm_cpu_has_interrupt(vcpu) &&
5119 !kvm_event_needs_reinjection(vcpu);
5122 static void vapic_enter(struct kvm_vcpu *vcpu)
5124 struct kvm_lapic *apic = vcpu->arch.apic;
5127 if (!apic || !apic->vapic_addr)
5130 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5132 vcpu->arch.apic->vapic_page = page;
5135 static void vapic_exit(struct kvm_vcpu *vcpu)
5137 struct kvm_lapic *apic = vcpu->arch.apic;
5140 if (!apic || !apic->vapic_addr)
5143 idx = srcu_read_lock(&vcpu->kvm->srcu);
5144 kvm_release_page_dirty(apic->vapic_page);
5145 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5146 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5149 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5153 if (!kvm_x86_ops->update_cr8_intercept)
5156 if (!vcpu->arch.apic)
5159 if (!vcpu->arch.apic->vapic_addr)
5160 max_irr = kvm_lapic_find_highest_irr(vcpu);
5167 tpr = kvm_lapic_get_cr8(vcpu);
5169 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5172 static void inject_pending_event(struct kvm_vcpu *vcpu)
5174 /* try to reinject previous events if any */
5175 if (vcpu->arch.exception.pending) {
5176 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5177 vcpu->arch.exception.has_error_code,
5178 vcpu->arch.exception.error_code);
5179 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5180 vcpu->arch.exception.has_error_code,
5181 vcpu->arch.exception.error_code,
5182 vcpu->arch.exception.reinject);
5186 if (vcpu->arch.nmi_injected) {
5187 kvm_x86_ops->set_nmi(vcpu);
5191 if (vcpu->arch.interrupt.pending) {
5192 kvm_x86_ops->set_irq(vcpu);
5196 /* try to inject new event if pending */
5197 if (vcpu->arch.nmi_pending) {
5198 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5199 vcpu->arch.nmi_pending = false;
5200 vcpu->arch.nmi_injected = true;
5201 kvm_x86_ops->set_nmi(vcpu);
5203 } else if (kvm_cpu_has_interrupt(vcpu)) {
5204 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5205 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5207 kvm_x86_ops->set_irq(vcpu);
5212 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5214 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5215 !vcpu->guest_xcr0_loaded) {
5216 /* kvm_set_xcr() also depends on this */
5217 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5218 vcpu->guest_xcr0_loaded = 1;
5222 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5224 if (vcpu->guest_xcr0_loaded) {
5225 if (vcpu->arch.xcr0 != host_xcr0)
5226 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5227 vcpu->guest_xcr0_loaded = 0;
5231 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5235 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5236 vcpu->run->request_interrupt_window;
5238 if (vcpu->requests) {
5239 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5240 kvm_mmu_unload(vcpu);
5241 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5242 __kvm_migrate_timers(vcpu);
5243 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5244 r = kvm_guest_time_update(vcpu);
5248 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5249 kvm_mmu_sync_roots(vcpu);
5250 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5251 kvm_x86_ops->tlb_flush(vcpu);
5252 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5253 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5257 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5258 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5262 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5263 vcpu->fpu_active = 0;
5264 kvm_x86_ops->fpu_deactivate(vcpu);
5266 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5267 /* Page is swapped out. Do synthetic halt */
5268 vcpu->arch.apf.halted = true;
5274 r = kvm_mmu_reload(vcpu);
5279 * An NMI can be injected between local nmi_pending read and
5280 * vcpu->arch.nmi_pending read inside inject_pending_event().
5281 * But in that case, KVM_REQ_EVENT will be set, which makes
5282 * the race described above benign.
5284 nmi_pending = ACCESS_ONCE(vcpu->arch.nmi_pending);
5286 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5287 inject_pending_event(vcpu);
5289 /* enable NMI/IRQ window open exits if needed */
5291 kvm_x86_ops->enable_nmi_window(vcpu);
5292 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5293 kvm_x86_ops->enable_irq_window(vcpu);
5295 if (kvm_lapic_enabled(vcpu)) {
5296 update_cr8_intercept(vcpu);
5297 kvm_lapic_sync_to_vapic(vcpu);
5303 kvm_x86_ops->prepare_guest_switch(vcpu);
5304 if (vcpu->fpu_active)
5305 kvm_load_guest_fpu(vcpu);
5306 kvm_load_guest_xcr0(vcpu);
5308 vcpu->mode = IN_GUEST_MODE;
5310 /* We should set ->mode before check ->requests,
5311 * see the comment in make_all_cpus_request.
5315 local_irq_disable();
5317 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5318 || need_resched() || signal_pending(current)) {
5319 vcpu->mode = OUTSIDE_GUEST_MODE;
5323 kvm_x86_ops->cancel_injection(vcpu);
5328 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5332 if (unlikely(vcpu->arch.switch_db_regs)) {
5334 set_debugreg(vcpu->arch.eff_db[0], 0);
5335 set_debugreg(vcpu->arch.eff_db[1], 1);
5336 set_debugreg(vcpu->arch.eff_db[2], 2);
5337 set_debugreg(vcpu->arch.eff_db[3], 3);
5340 trace_kvm_entry(vcpu->vcpu_id);
5341 kvm_x86_ops->run(vcpu);
5344 * If the guest has used debug registers, at least dr7
5345 * will be disabled while returning to the host.
5346 * If we don't have active breakpoints in the host, we don't
5347 * care about the messed up debug address registers. But if
5348 * we have some of them active, restore the old state.
5350 if (hw_breakpoint_active())
5351 hw_breakpoint_restore();
5353 kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
5355 vcpu->mode = OUTSIDE_GUEST_MODE;
5362 * We must have an instruction between local_irq_enable() and
5363 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5364 * the interrupt shadow. The stat.exits increment will do nicely.
5365 * But we need to prevent reordering, hence this barrier():
5373 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5376 * Profile KVM exit RIPs:
5378 if (unlikely(prof_on == KVM_PROFILING)) {
5379 unsigned long rip = kvm_rip_read(vcpu);
5380 profile_hit(KVM_PROFILING, (void *)rip);
5384 kvm_lapic_sync_from_vapic(vcpu);
5386 r = kvm_x86_ops->handle_exit(vcpu);
5392 static int __vcpu_run(struct kvm_vcpu *vcpu)
5395 struct kvm *kvm = vcpu->kvm;
5397 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5398 pr_debug("vcpu %d received sipi with vector # %x\n",
5399 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5400 kvm_lapic_reset(vcpu);
5401 r = kvm_arch_vcpu_reset(vcpu);
5404 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5407 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5412 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5413 !vcpu->arch.apf.halted)
5414 r = vcpu_enter_guest(vcpu);
5416 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5417 kvm_vcpu_block(vcpu);
5418 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5419 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5421 switch(vcpu->arch.mp_state) {
5422 case KVM_MP_STATE_HALTED:
5423 vcpu->arch.mp_state =
5424 KVM_MP_STATE_RUNNABLE;
5425 case KVM_MP_STATE_RUNNABLE:
5426 vcpu->arch.apf.halted = false;
5428 case KVM_MP_STATE_SIPI_RECEIVED:
5439 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5440 if (kvm_cpu_has_pending_timer(vcpu))
5441 kvm_inject_pending_timer_irqs(vcpu);
5443 if (dm_request_for_irq_injection(vcpu)) {
5445 vcpu->run->exit_reason = KVM_EXIT_INTR;
5446 ++vcpu->stat.request_irq_exits;
5449 kvm_check_async_pf_completion(vcpu);
5451 if (signal_pending(current)) {
5453 vcpu->run->exit_reason = KVM_EXIT_INTR;
5454 ++vcpu->stat.signal_exits;
5456 if (need_resched()) {
5457 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5459 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5463 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5470 static int complete_mmio(struct kvm_vcpu *vcpu)
5472 struct kvm_run *run = vcpu->run;
5475 if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5478 if (vcpu->mmio_needed) {
5479 vcpu->mmio_needed = 0;
5480 if (!vcpu->mmio_is_write)
5481 memcpy(vcpu->mmio_data, run->mmio.data, 8);
5482 vcpu->mmio_index += 8;
5483 if (vcpu->mmio_index < vcpu->mmio_size) {
5484 run->exit_reason = KVM_EXIT_MMIO;
5485 run->mmio.phys_addr = vcpu->mmio_phys_addr + vcpu->mmio_index;
5486 memcpy(run->mmio.data, vcpu->mmio_data + vcpu->mmio_index, 8);
5487 run->mmio.len = min(vcpu->mmio_size - vcpu->mmio_index, 8);
5488 run->mmio.is_write = vcpu->mmio_is_write;
5489 vcpu->mmio_needed = 1;
5492 if (vcpu->mmio_is_write)
5494 vcpu->mmio_read_completed = 1;
5496 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5497 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5498 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5499 if (r != EMULATE_DONE)
5504 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5509 if (!tsk_used_math(current) && init_fpu(current))
5512 if (vcpu->sigset_active)
5513 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5515 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5516 kvm_vcpu_block(vcpu);
5517 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5522 /* re-sync apic's tpr */
5523 if (!irqchip_in_kernel(vcpu->kvm)) {
5524 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5530 r = complete_mmio(vcpu);
5534 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5535 kvm_register_write(vcpu, VCPU_REGS_RAX,
5536 kvm_run->hypercall.ret);
5538 r = __vcpu_run(vcpu);
5541 post_kvm_run_save(vcpu);
5542 if (vcpu->sigset_active)
5543 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5548 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5550 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5551 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5552 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5553 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5554 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5555 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5556 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5557 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5558 #ifdef CONFIG_X86_64
5559 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5560 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5561 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5562 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5563 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5564 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5565 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5566 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5569 regs->rip = kvm_rip_read(vcpu);
5570 regs->rflags = kvm_get_rflags(vcpu);
5575 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5577 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5578 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5579 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5580 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5581 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5582 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5583 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5584 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5585 #ifdef CONFIG_X86_64
5586 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5587 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5588 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5589 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5590 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5591 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5592 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5593 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5596 kvm_rip_write(vcpu, regs->rip);
5597 kvm_set_rflags(vcpu, regs->rflags);
5599 vcpu->arch.exception.pending = false;
5601 kvm_make_request(KVM_REQ_EVENT, vcpu);
5606 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5608 struct kvm_segment cs;
5610 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
5614 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5616 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5617 struct kvm_sregs *sregs)
5621 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5622 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5623 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5624 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5625 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5626 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5628 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5629 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5631 kvm_x86_ops->get_idt(vcpu, &dt);
5632 sregs->idt.limit = dt.size;
5633 sregs->idt.base = dt.address;
5634 kvm_x86_ops->get_gdt(vcpu, &dt);
5635 sregs->gdt.limit = dt.size;
5636 sregs->gdt.base = dt.address;
5638 sregs->cr0 = kvm_read_cr0(vcpu);
5639 sregs->cr2 = vcpu->arch.cr2;
5640 sregs->cr3 = kvm_read_cr3(vcpu);
5641 sregs->cr4 = kvm_read_cr4(vcpu);
5642 sregs->cr8 = kvm_get_cr8(vcpu);
5643 sregs->efer = vcpu->arch.efer;
5644 sregs->apic_base = kvm_get_apic_base(vcpu);
5646 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
5648 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
5649 set_bit(vcpu->arch.interrupt.nr,
5650 (unsigned long *)sregs->interrupt_bitmap);
5655 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
5656 struct kvm_mp_state *mp_state)
5658 mp_state->mp_state = vcpu->arch.mp_state;
5662 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
5663 struct kvm_mp_state *mp_state)
5665 vcpu->arch.mp_state = mp_state->mp_state;
5666 kvm_make_request(KVM_REQ_EVENT, vcpu);
5670 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
5671 bool has_error_code, u32 error_code)
5673 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5676 init_emulate_ctxt(vcpu);
5678 ret = emulator_task_switch(&vcpu->arch.emulate_ctxt,
5679 tss_selector, reason, has_error_code,
5683 return EMULATE_FAIL;
5685 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5686 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
5687 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
5688 kvm_make_request(KVM_REQ_EVENT, vcpu);
5689 return EMULATE_DONE;
5691 EXPORT_SYMBOL_GPL(kvm_task_switch);
5693 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5694 struct kvm_sregs *sregs)
5696 int mmu_reset_needed = 0;
5697 int pending_vec, max_bits, idx;
5700 dt.size = sregs->idt.limit;
5701 dt.address = sregs->idt.base;
5702 kvm_x86_ops->set_idt(vcpu, &dt);
5703 dt.size = sregs->gdt.limit;
5704 dt.address = sregs->gdt.base;
5705 kvm_x86_ops->set_gdt(vcpu, &dt);
5707 vcpu->arch.cr2 = sregs->cr2;
5708 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
5709 vcpu->arch.cr3 = sregs->cr3;
5710 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
5712 kvm_set_cr8(vcpu, sregs->cr8);
5714 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5715 kvm_x86_ops->set_efer(vcpu, sregs->efer);
5716 kvm_set_apic_base(vcpu, sregs->apic_base);
5718 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5719 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5720 vcpu->arch.cr0 = sregs->cr0;
5722 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5723 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5724 if (sregs->cr4 & X86_CR4_OSXSAVE)
5727 idx = srcu_read_lock(&vcpu->kvm->srcu);
5728 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5729 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
5730 mmu_reset_needed = 1;
5732 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5734 if (mmu_reset_needed)
5735 kvm_mmu_reset_context(vcpu);
5737 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5738 pending_vec = find_first_bit(
5739 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5740 if (pending_vec < max_bits) {
5741 kvm_queue_interrupt(vcpu, pending_vec, false);
5742 pr_debug("Set back pending irq %d\n", pending_vec);
5745 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5746 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5747 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5748 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5749 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5750 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5752 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5753 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5755 update_cr8_intercept(vcpu);
5757 /* Older userspace won't unhalt the vcpu on reset. */
5758 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5759 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5761 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5763 kvm_make_request(KVM_REQ_EVENT, vcpu);
5768 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5769 struct kvm_guest_debug *dbg)
5771 unsigned long rflags;
5774 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5776 if (vcpu->arch.exception.pending)
5778 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5779 kvm_queue_exception(vcpu, DB_VECTOR);
5781 kvm_queue_exception(vcpu, BP_VECTOR);
5785 * Read rflags as long as potentially injected trace flags are still
5788 rflags = kvm_get_rflags(vcpu);
5790 vcpu->guest_debug = dbg->control;
5791 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5792 vcpu->guest_debug = 0;
5794 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5795 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5796 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5797 vcpu->arch.switch_db_regs =
5798 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5800 for (i = 0; i < KVM_NR_DB_REGS; i++)
5801 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5802 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5805 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5806 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5807 get_segment_base(vcpu, VCPU_SREG_CS);
5810 * Trigger an rflags update that will inject or remove the trace
5813 kvm_set_rflags(vcpu, rflags);
5815 kvm_x86_ops->set_guest_debug(vcpu, dbg);
5825 * Translate a guest virtual address to a guest physical address.
5827 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5828 struct kvm_translation *tr)
5830 unsigned long vaddr = tr->linear_address;
5834 idx = srcu_read_lock(&vcpu->kvm->srcu);
5835 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5836 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5837 tr->physical_address = gpa;
5838 tr->valid = gpa != UNMAPPED_GVA;
5845 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5847 struct i387_fxsave_struct *fxsave =
5848 &vcpu->arch.guest_fpu.state->fxsave;
5850 memcpy(fpu->fpr, fxsave->st_space, 128);
5851 fpu->fcw = fxsave->cwd;
5852 fpu->fsw = fxsave->swd;
5853 fpu->ftwx = fxsave->twd;
5854 fpu->last_opcode = fxsave->fop;
5855 fpu->last_ip = fxsave->rip;
5856 fpu->last_dp = fxsave->rdp;
5857 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5862 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5864 struct i387_fxsave_struct *fxsave =
5865 &vcpu->arch.guest_fpu.state->fxsave;
5867 memcpy(fxsave->st_space, fpu->fpr, 128);
5868 fxsave->cwd = fpu->fcw;
5869 fxsave->swd = fpu->fsw;
5870 fxsave->twd = fpu->ftwx;
5871 fxsave->fop = fpu->last_opcode;
5872 fxsave->rip = fpu->last_ip;
5873 fxsave->rdp = fpu->last_dp;
5874 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5879 int fx_init(struct kvm_vcpu *vcpu)
5883 err = fpu_alloc(&vcpu->arch.guest_fpu);
5887 fpu_finit(&vcpu->arch.guest_fpu);
5890 * Ensure guest xcr0 is valid for loading
5892 vcpu->arch.xcr0 = XSTATE_FP;
5894 vcpu->arch.cr0 |= X86_CR0_ET;
5898 EXPORT_SYMBOL_GPL(fx_init);
5900 static void fx_free(struct kvm_vcpu *vcpu)
5902 fpu_free(&vcpu->arch.guest_fpu);
5905 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5907 if (vcpu->guest_fpu_loaded)
5911 * Restore all possible states in the guest,
5912 * and assume host would use all available bits.
5913 * Guest xcr0 would be loaded later.
5915 kvm_put_guest_xcr0(vcpu);
5916 vcpu->guest_fpu_loaded = 1;
5917 unlazy_fpu(current);
5918 fpu_restore_checking(&vcpu->arch.guest_fpu);
5922 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5924 kvm_put_guest_xcr0(vcpu);
5926 if (!vcpu->guest_fpu_loaded)
5929 vcpu->guest_fpu_loaded = 0;
5930 fpu_save_init(&vcpu->arch.guest_fpu);
5931 ++vcpu->stat.fpu_reload;
5932 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
5936 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5938 kvmclock_reset(vcpu);
5940 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
5942 kvm_x86_ops->vcpu_free(vcpu);
5945 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5948 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
5949 printk_once(KERN_WARNING
5950 "kvm: SMP vm created on host with unstable TSC; "
5951 "guest TSC will not be reliable\n");
5952 return kvm_x86_ops->vcpu_create(kvm, id);
5955 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5959 vcpu->arch.mtrr_state.have_fixed = 1;
5961 r = kvm_arch_vcpu_reset(vcpu);
5963 r = kvm_mmu_setup(vcpu);
5970 kvm_x86_ops->vcpu_free(vcpu);
5974 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5976 vcpu->arch.apf.msr_val = 0;
5979 kvm_mmu_unload(vcpu);
5983 kvm_x86_ops->vcpu_free(vcpu);
5986 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5988 vcpu->arch.nmi_pending = false;
5989 vcpu->arch.nmi_injected = false;
5991 vcpu->arch.switch_db_regs = 0;
5992 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5993 vcpu->arch.dr6 = DR6_FIXED_1;
5994 vcpu->arch.dr7 = DR7_FIXED_1;
5996 kvm_make_request(KVM_REQ_EVENT, vcpu);
5997 vcpu->arch.apf.msr_val = 0;
5999 kvmclock_reset(vcpu);
6001 kvm_clear_async_pf_completion_queue(vcpu);
6002 kvm_async_pf_hash_reset(vcpu);
6003 vcpu->arch.apf.halted = false;
6005 return kvm_x86_ops->vcpu_reset(vcpu);
6008 int kvm_arch_hardware_enable(void *garbage)
6011 struct kvm_vcpu *vcpu;
6014 kvm_shared_msr_cpu_online();
6015 list_for_each_entry(kvm, &vm_list, vm_list)
6016 kvm_for_each_vcpu(i, vcpu, kvm)
6017 if (vcpu->cpu == smp_processor_id())
6018 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6019 return kvm_x86_ops->hardware_enable(garbage);
6022 void kvm_arch_hardware_disable(void *garbage)
6024 kvm_x86_ops->hardware_disable(garbage);
6025 drop_user_return_notifiers(garbage);
6028 int kvm_arch_hardware_setup(void)
6030 return kvm_x86_ops->hardware_setup();
6033 void kvm_arch_hardware_unsetup(void)
6035 kvm_x86_ops->hardware_unsetup();
6038 void kvm_arch_check_processor_compat(void *rtn)
6040 kvm_x86_ops->check_processor_compatibility(rtn);
6043 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6049 BUG_ON(vcpu->kvm == NULL);
6052 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6053 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
6054 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
6055 vcpu->arch.mmu.translate_gpa = translate_gpa;
6056 vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
6057 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6058 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6060 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6062 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6067 vcpu->arch.pio_data = page_address(page);
6069 kvm_init_tsc_catchup(vcpu, max_tsc_khz);
6071 r = kvm_mmu_create(vcpu);
6073 goto fail_free_pio_data;
6075 if (irqchip_in_kernel(kvm)) {
6076 r = kvm_create_lapic(vcpu);
6078 goto fail_mmu_destroy;
6081 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6083 if (!vcpu->arch.mce_banks) {
6085 goto fail_free_lapic;
6087 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6089 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6090 goto fail_free_mce_banks;
6092 kvm_async_pf_hash_reset(vcpu);
6095 fail_free_mce_banks:
6096 kfree(vcpu->arch.mce_banks);
6098 kvm_free_lapic(vcpu);
6100 kvm_mmu_destroy(vcpu);
6102 free_page((unsigned long)vcpu->arch.pio_data);
6107 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6111 kfree(vcpu->arch.mce_banks);
6112 kvm_free_lapic(vcpu);
6113 idx = srcu_read_lock(&vcpu->kvm->srcu);
6114 kvm_mmu_destroy(vcpu);
6115 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6116 free_page((unsigned long)vcpu->arch.pio_data);
6119 int kvm_arch_init_vm(struct kvm *kvm)
6121 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6122 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6124 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6125 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6127 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6132 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6135 kvm_mmu_unload(vcpu);
6139 static void kvm_free_vcpus(struct kvm *kvm)
6142 struct kvm_vcpu *vcpu;
6145 * Unpin any mmu pages first.
6147 kvm_for_each_vcpu(i, vcpu, kvm) {
6148 kvm_clear_async_pf_completion_queue(vcpu);
6149 kvm_unload_vcpu_mmu(vcpu);
6151 kvm_for_each_vcpu(i, vcpu, kvm)
6152 kvm_arch_vcpu_free(vcpu);
6154 mutex_lock(&kvm->lock);
6155 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6156 kvm->vcpus[i] = NULL;
6158 atomic_set(&kvm->online_vcpus, 0);
6159 mutex_unlock(&kvm->lock);
6162 void kvm_arch_sync_events(struct kvm *kvm)
6164 kvm_free_all_assigned_devices(kvm);
6168 void kvm_arch_destroy_vm(struct kvm *kvm)
6170 kvm_iommu_unmap_guest(kvm);
6171 kfree(kvm->arch.vpic);
6172 kfree(kvm->arch.vioapic);
6173 kvm_free_vcpus(kvm);
6174 if (kvm->arch.apic_access_page)
6175 put_page(kvm->arch.apic_access_page);
6176 if (kvm->arch.ept_identity_pagetable)
6177 put_page(kvm->arch.ept_identity_pagetable);
6180 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6181 struct kvm_memory_slot *memslot,
6182 struct kvm_memory_slot old,
6183 struct kvm_userspace_memory_region *mem,
6186 int npages = memslot->npages;
6187 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6189 /* Prevent internal slot pages from being moved by fork()/COW. */
6190 if (memslot->id >= KVM_MEMORY_SLOTS)
6191 map_flags = MAP_SHARED | MAP_ANONYMOUS;
6193 /*To keep backward compatibility with older userspace,
6194 *x86 needs to hanlde !user_alloc case.
6197 if (npages && !old.rmap) {
6198 unsigned long userspace_addr;
6200 down_write(¤t->mm->mmap_sem);
6201 userspace_addr = do_mmap(NULL, 0,
6203 PROT_READ | PROT_WRITE,
6206 up_write(¤t->mm->mmap_sem);
6208 if (IS_ERR((void *)userspace_addr))
6209 return PTR_ERR((void *)userspace_addr);
6211 memslot->userspace_addr = userspace_addr;
6219 void kvm_arch_commit_memory_region(struct kvm *kvm,
6220 struct kvm_userspace_memory_region *mem,
6221 struct kvm_memory_slot old,
6225 int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
6227 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6230 down_write(¤t->mm->mmap_sem);
6231 ret = do_munmap(current->mm, old.userspace_addr,
6232 old.npages * PAGE_SIZE);
6233 up_write(¤t->mm->mmap_sem);
6236 "kvm_vm_ioctl_set_memory_region: "
6237 "failed to munmap memory\n");
6240 if (!kvm->arch.n_requested_mmu_pages)
6241 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6243 spin_lock(&kvm->mmu_lock);
6245 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6246 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6247 spin_unlock(&kvm->mmu_lock);
6250 void kvm_arch_flush_shadow(struct kvm *kvm)
6252 kvm_mmu_zap_all(kvm);
6253 kvm_reload_remote_mmus(kvm);
6256 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6258 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6259 !vcpu->arch.apf.halted)
6260 || !list_empty_careful(&vcpu->async_pf.done)
6261 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6262 || vcpu->arch.nmi_pending ||
6263 (kvm_arch_interrupt_allowed(vcpu) &&
6264 kvm_cpu_has_interrupt(vcpu));
6267 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6270 int cpu = vcpu->cpu;
6272 if (waitqueue_active(&vcpu->wq)) {
6273 wake_up_interruptible(&vcpu->wq);
6274 ++vcpu->stat.halt_wakeup;
6278 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6279 if (kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE)
6280 smp_send_reschedule(cpu);
6284 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6286 return kvm_x86_ops->interrupt_allowed(vcpu);
6289 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6291 unsigned long current_rip = kvm_rip_read(vcpu) +
6292 get_segment_base(vcpu, VCPU_SREG_CS);
6294 return current_rip == linear_rip;
6296 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6298 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6300 unsigned long rflags;
6302 rflags = kvm_x86_ops->get_rflags(vcpu);
6303 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6304 rflags &= ~X86_EFLAGS_TF;
6307 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6309 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6311 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6312 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6313 rflags |= X86_EFLAGS_TF;
6314 kvm_x86_ops->set_rflags(vcpu, rflags);
6315 kvm_make_request(KVM_REQ_EVENT, vcpu);
6317 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6319 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6323 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6324 is_error_page(work->page))
6327 r = kvm_mmu_reload(vcpu);
6331 if (!vcpu->arch.mmu.direct_map &&
6332 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6335 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6338 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6340 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6343 static inline u32 kvm_async_pf_next_probe(u32 key)
6345 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6348 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6350 u32 key = kvm_async_pf_hash_fn(gfn);
6352 while (vcpu->arch.apf.gfns[key] != ~0)
6353 key = kvm_async_pf_next_probe(key);
6355 vcpu->arch.apf.gfns[key] = gfn;
6358 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6361 u32 key = kvm_async_pf_hash_fn(gfn);
6363 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6364 (vcpu->arch.apf.gfns[key] != gfn &&
6365 vcpu->arch.apf.gfns[key] != ~0); i++)
6366 key = kvm_async_pf_next_probe(key);
6371 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6373 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6376 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6380 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6382 vcpu->arch.apf.gfns[i] = ~0;
6384 j = kvm_async_pf_next_probe(j);
6385 if (vcpu->arch.apf.gfns[j] == ~0)
6387 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6389 * k lies cyclically in ]i,j]
6391 * |....j i.k.| or |.k..j i...|
6393 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6394 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6399 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6402 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6406 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6407 struct kvm_async_pf *work)
6409 struct x86_exception fault;
6411 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6412 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6414 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6415 (vcpu->arch.apf.send_user_only &&
6416 kvm_x86_ops->get_cpl(vcpu) == 0))
6417 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6418 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6419 fault.vector = PF_VECTOR;
6420 fault.error_code_valid = true;
6421 fault.error_code = 0;
6422 fault.nested_page_fault = false;
6423 fault.address = work->arch.token;
6424 kvm_inject_page_fault(vcpu, &fault);
6428 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6429 struct kvm_async_pf *work)
6431 struct x86_exception fault;
6433 trace_kvm_async_pf_ready(work->arch.token, work->gva);
6434 if (is_error_page(work->page))
6435 work->arch.token = ~0; /* broadcast wakeup */
6437 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6439 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6440 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6441 fault.vector = PF_VECTOR;
6442 fault.error_code_valid = true;
6443 fault.error_code = 0;
6444 fault.nested_page_fault = false;
6445 fault.address = work->arch.token;
6446 kvm_inject_page_fault(vcpu, &fault);
6448 vcpu->arch.apf.halted = false;
6451 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6453 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6456 return !kvm_event_needs_reinjection(vcpu) &&
6457 kvm_x86_ops->interrupt_allowed(vcpu);
6460 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6461 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6462 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6463 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6464 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6465 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6466 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6467 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6468 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6469 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6470 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6471 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);