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)
66 #define emul_to_vcpu(ctxt) \
67 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
70 * - enable syscall per default because its emulated by KVM
71 * - enable LME and LMA per default on 64 bit KVM
75 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
77 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
80 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
81 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
83 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
84 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
85 struct kvm_cpuid_entry2 __user *entries);
87 struct kvm_x86_ops *kvm_x86_ops;
88 EXPORT_SYMBOL_GPL(kvm_x86_ops);
91 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
93 bool kvm_has_tsc_control;
94 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
95 u32 kvm_max_guest_tsc_khz;
96 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
98 #define KVM_NR_SHARED_MSRS 16
100 struct kvm_shared_msrs_global {
102 u32 msrs[KVM_NR_SHARED_MSRS];
105 struct kvm_shared_msrs {
106 struct user_return_notifier urn;
108 struct kvm_shared_msr_values {
111 } values[KVM_NR_SHARED_MSRS];
114 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
115 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
117 struct kvm_stats_debugfs_item debugfs_entries[] = {
118 { "pf_fixed", VCPU_STAT(pf_fixed) },
119 { "pf_guest", VCPU_STAT(pf_guest) },
120 { "tlb_flush", VCPU_STAT(tlb_flush) },
121 { "invlpg", VCPU_STAT(invlpg) },
122 { "exits", VCPU_STAT(exits) },
123 { "io_exits", VCPU_STAT(io_exits) },
124 { "mmio_exits", VCPU_STAT(mmio_exits) },
125 { "signal_exits", VCPU_STAT(signal_exits) },
126 { "irq_window", VCPU_STAT(irq_window_exits) },
127 { "nmi_window", VCPU_STAT(nmi_window_exits) },
128 { "halt_exits", VCPU_STAT(halt_exits) },
129 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
130 { "hypercalls", VCPU_STAT(hypercalls) },
131 { "request_irq", VCPU_STAT(request_irq_exits) },
132 { "irq_exits", VCPU_STAT(irq_exits) },
133 { "host_state_reload", VCPU_STAT(host_state_reload) },
134 { "efer_reload", VCPU_STAT(efer_reload) },
135 { "fpu_reload", VCPU_STAT(fpu_reload) },
136 { "insn_emulation", VCPU_STAT(insn_emulation) },
137 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
138 { "irq_injections", VCPU_STAT(irq_injections) },
139 { "nmi_injections", VCPU_STAT(nmi_injections) },
140 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
141 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
142 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
143 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
144 { "mmu_flooded", VM_STAT(mmu_flooded) },
145 { "mmu_recycled", VM_STAT(mmu_recycled) },
146 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
147 { "mmu_unsync", VM_STAT(mmu_unsync) },
148 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
149 { "largepages", VM_STAT(lpages) },
153 u64 __read_mostly host_xcr0;
155 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
157 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
160 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
161 vcpu->arch.apf.gfns[i] = ~0;
164 static void kvm_on_user_return(struct user_return_notifier *urn)
167 struct kvm_shared_msrs *locals
168 = container_of(urn, struct kvm_shared_msrs, urn);
169 struct kvm_shared_msr_values *values;
171 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
172 values = &locals->values[slot];
173 if (values->host != values->curr) {
174 wrmsrl(shared_msrs_global.msrs[slot], values->host);
175 values->curr = values->host;
178 locals->registered = false;
179 user_return_notifier_unregister(urn);
182 static void shared_msr_update(unsigned slot, u32 msr)
184 struct kvm_shared_msrs *smsr;
187 smsr = &__get_cpu_var(shared_msrs);
188 /* only read, and nobody should modify it at this time,
189 * so don't need lock */
190 if (slot >= shared_msrs_global.nr) {
191 printk(KERN_ERR "kvm: invalid MSR slot!");
194 rdmsrl_safe(msr, &value);
195 smsr->values[slot].host = value;
196 smsr->values[slot].curr = value;
199 void kvm_define_shared_msr(unsigned slot, u32 msr)
201 if (slot >= shared_msrs_global.nr)
202 shared_msrs_global.nr = slot + 1;
203 shared_msrs_global.msrs[slot] = msr;
204 /* we need ensured the shared_msr_global have been updated */
207 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
209 static void kvm_shared_msr_cpu_online(void)
213 for (i = 0; i < shared_msrs_global.nr; ++i)
214 shared_msr_update(i, shared_msrs_global.msrs[i]);
217 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
219 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
221 if (((value ^ smsr->values[slot].curr) & mask) == 0)
223 smsr->values[slot].curr = value;
224 wrmsrl(shared_msrs_global.msrs[slot], value);
225 if (!smsr->registered) {
226 smsr->urn.on_user_return = kvm_on_user_return;
227 user_return_notifier_register(&smsr->urn);
228 smsr->registered = true;
231 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
233 static void drop_user_return_notifiers(void *ignore)
235 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
237 if (smsr->registered)
238 kvm_on_user_return(&smsr->urn);
241 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
243 if (irqchip_in_kernel(vcpu->kvm))
244 return vcpu->arch.apic_base;
246 return vcpu->arch.apic_base;
248 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
250 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
252 /* TODO: reserve bits check */
253 if (irqchip_in_kernel(vcpu->kvm))
254 kvm_lapic_set_base(vcpu, data);
256 vcpu->arch.apic_base = data;
258 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
260 #define EXCPT_BENIGN 0
261 #define EXCPT_CONTRIBUTORY 1
264 static int exception_class(int vector)
274 return EXCPT_CONTRIBUTORY;
281 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
282 unsigned nr, bool has_error, u32 error_code,
288 kvm_make_request(KVM_REQ_EVENT, vcpu);
290 if (!vcpu->arch.exception.pending) {
292 vcpu->arch.exception.pending = true;
293 vcpu->arch.exception.has_error_code = has_error;
294 vcpu->arch.exception.nr = nr;
295 vcpu->arch.exception.error_code = error_code;
296 vcpu->arch.exception.reinject = reinject;
300 /* to check exception */
301 prev_nr = vcpu->arch.exception.nr;
302 if (prev_nr == DF_VECTOR) {
303 /* triple fault -> shutdown */
304 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
307 class1 = exception_class(prev_nr);
308 class2 = exception_class(nr);
309 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
310 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
311 /* generate double fault per SDM Table 5-5 */
312 vcpu->arch.exception.pending = true;
313 vcpu->arch.exception.has_error_code = true;
314 vcpu->arch.exception.nr = DF_VECTOR;
315 vcpu->arch.exception.error_code = 0;
317 /* replace previous exception with a new one in a hope
318 that instruction re-execution will regenerate lost
323 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
325 kvm_multiple_exception(vcpu, nr, false, 0, false);
327 EXPORT_SYMBOL_GPL(kvm_queue_exception);
329 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
331 kvm_multiple_exception(vcpu, nr, false, 0, true);
333 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
335 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
338 kvm_inject_gp(vcpu, 0);
340 kvm_x86_ops->skip_emulated_instruction(vcpu);
342 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
344 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
346 ++vcpu->stat.pf_guest;
347 vcpu->arch.cr2 = fault->address;
348 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
351 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
353 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
354 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
356 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
359 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
361 kvm_make_request(KVM_REQ_EVENT, vcpu);
362 vcpu->arch.nmi_pending = 1;
364 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
366 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
368 kvm_multiple_exception(vcpu, nr, true, error_code, false);
370 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
372 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
374 kvm_multiple_exception(vcpu, nr, true, error_code, true);
376 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
379 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
380 * a #GP and return false.
382 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
384 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
386 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
389 EXPORT_SYMBOL_GPL(kvm_require_cpl);
392 * This function will be used to read from the physical memory of the currently
393 * running guest. The difference to kvm_read_guest_page is that this function
394 * can read from guest physical or from the guest's guest physical memory.
396 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
397 gfn_t ngfn, void *data, int offset, int len,
403 ngpa = gfn_to_gpa(ngfn);
404 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
405 if (real_gfn == UNMAPPED_GVA)
408 real_gfn = gpa_to_gfn(real_gfn);
410 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
412 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
414 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
415 void *data, int offset, int len, u32 access)
417 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
418 data, offset, len, access);
422 * Load the pae pdptrs. Return true is they are all valid.
424 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
426 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
427 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
430 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
432 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
433 offset * sizeof(u64), sizeof(pdpte),
434 PFERR_USER_MASK|PFERR_WRITE_MASK);
439 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
440 if (is_present_gpte(pdpte[i]) &&
441 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
448 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
449 __set_bit(VCPU_EXREG_PDPTR,
450 (unsigned long *)&vcpu->arch.regs_avail);
451 __set_bit(VCPU_EXREG_PDPTR,
452 (unsigned long *)&vcpu->arch.regs_dirty);
457 EXPORT_SYMBOL_GPL(load_pdptrs);
459 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
461 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
467 if (is_long_mode(vcpu) || !is_pae(vcpu))
470 if (!test_bit(VCPU_EXREG_PDPTR,
471 (unsigned long *)&vcpu->arch.regs_avail))
474 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
475 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
476 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
477 PFERR_USER_MASK | PFERR_WRITE_MASK);
480 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
486 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
488 unsigned long old_cr0 = kvm_read_cr0(vcpu);
489 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
490 X86_CR0_CD | X86_CR0_NW;
495 if (cr0 & 0xffffffff00000000UL)
499 cr0 &= ~CR0_RESERVED_BITS;
501 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
504 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
507 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
509 if ((vcpu->arch.efer & EFER_LME)) {
514 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
519 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
524 kvm_x86_ops->set_cr0(vcpu, cr0);
526 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
527 kvm_clear_async_pf_completion_queue(vcpu);
528 kvm_async_pf_hash_reset(vcpu);
531 if ((cr0 ^ old_cr0) & update_bits)
532 kvm_mmu_reset_context(vcpu);
535 EXPORT_SYMBOL_GPL(kvm_set_cr0);
537 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
539 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
541 EXPORT_SYMBOL_GPL(kvm_lmsw);
543 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
547 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
548 if (index != XCR_XFEATURE_ENABLED_MASK)
551 if (kvm_x86_ops->get_cpl(vcpu) != 0)
553 if (!(xcr0 & XSTATE_FP))
555 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
557 if (xcr0 & ~host_xcr0)
559 vcpu->arch.xcr0 = xcr0;
560 vcpu->guest_xcr0_loaded = 0;
564 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
566 if (__kvm_set_xcr(vcpu, index, xcr)) {
567 kvm_inject_gp(vcpu, 0);
572 EXPORT_SYMBOL_GPL(kvm_set_xcr);
574 static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
576 struct kvm_cpuid_entry2 *best;
578 best = kvm_find_cpuid_entry(vcpu, 1, 0);
579 return best && (best->ecx & bit(X86_FEATURE_XSAVE));
582 static void update_cpuid(struct kvm_vcpu *vcpu)
584 struct kvm_cpuid_entry2 *best;
586 best = kvm_find_cpuid_entry(vcpu, 1, 0);
590 /* Update OSXSAVE bit */
591 if (cpu_has_xsave && best->function == 0x1) {
592 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
593 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
594 best->ecx |= bit(X86_FEATURE_OSXSAVE);
598 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
600 unsigned long old_cr4 = kvm_read_cr4(vcpu);
601 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
603 if (cr4 & CR4_RESERVED_BITS)
606 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
609 if (is_long_mode(vcpu)) {
610 if (!(cr4 & X86_CR4_PAE))
612 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
613 && ((cr4 ^ old_cr4) & pdptr_bits)
614 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
618 if (cr4 & X86_CR4_VMXE)
621 kvm_x86_ops->set_cr4(vcpu, cr4);
623 if ((cr4 ^ old_cr4) & pdptr_bits)
624 kvm_mmu_reset_context(vcpu);
626 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
631 EXPORT_SYMBOL_GPL(kvm_set_cr4);
633 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
635 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
636 kvm_mmu_sync_roots(vcpu);
637 kvm_mmu_flush_tlb(vcpu);
641 if (is_long_mode(vcpu)) {
642 if (cr3 & CR3_L_MODE_RESERVED_BITS)
646 if (cr3 & CR3_PAE_RESERVED_BITS)
648 if (is_paging(vcpu) &&
649 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
653 * We don't check reserved bits in nonpae mode, because
654 * this isn't enforced, and VMware depends on this.
659 * Does the new cr3 value map to physical memory? (Note, we
660 * catch an invalid cr3 even in real-mode, because it would
661 * cause trouble later on when we turn on paging anyway.)
663 * A real CPU would silently accept an invalid cr3 and would
664 * attempt to use it - with largely undefined (and often hard
665 * to debug) behavior on the guest side.
667 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
669 vcpu->arch.cr3 = cr3;
670 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
671 vcpu->arch.mmu.new_cr3(vcpu);
674 EXPORT_SYMBOL_GPL(kvm_set_cr3);
676 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
678 if (cr8 & CR8_RESERVED_BITS)
680 if (irqchip_in_kernel(vcpu->kvm))
681 kvm_lapic_set_tpr(vcpu, cr8);
683 vcpu->arch.cr8 = cr8;
686 EXPORT_SYMBOL_GPL(kvm_set_cr8);
688 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
690 if (irqchip_in_kernel(vcpu->kvm))
691 return kvm_lapic_get_cr8(vcpu);
693 return vcpu->arch.cr8;
695 EXPORT_SYMBOL_GPL(kvm_get_cr8);
697 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
701 vcpu->arch.db[dr] = val;
702 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
703 vcpu->arch.eff_db[dr] = val;
706 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
710 if (val & 0xffffffff00000000ULL)
712 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
715 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
719 if (val & 0xffffffff00000000ULL)
721 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
722 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
723 kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
724 vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
732 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
736 res = __kvm_set_dr(vcpu, dr, val);
738 kvm_queue_exception(vcpu, UD_VECTOR);
740 kvm_inject_gp(vcpu, 0);
744 EXPORT_SYMBOL_GPL(kvm_set_dr);
746 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
750 *val = vcpu->arch.db[dr];
753 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
757 *val = vcpu->arch.dr6;
760 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
764 *val = vcpu->arch.dr7;
771 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
773 if (_kvm_get_dr(vcpu, dr, val)) {
774 kvm_queue_exception(vcpu, UD_VECTOR);
779 EXPORT_SYMBOL_GPL(kvm_get_dr);
782 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
783 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
785 * This list is modified at module load time to reflect the
786 * capabilities of the host cpu. This capabilities test skips MSRs that are
787 * kvm-specific. Those are put in the beginning of the list.
790 #define KVM_SAVE_MSRS_BEGIN 8
791 static u32 msrs_to_save[] = {
792 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
793 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
794 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
795 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN,
796 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
799 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
801 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
804 static unsigned num_msrs_to_save;
806 static u32 emulated_msrs[] = {
807 MSR_IA32_MISC_ENABLE,
812 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
814 u64 old_efer = vcpu->arch.efer;
816 if (efer & efer_reserved_bits)
820 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
823 if (efer & EFER_FFXSR) {
824 struct kvm_cpuid_entry2 *feat;
826 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
827 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
831 if (efer & EFER_SVME) {
832 struct kvm_cpuid_entry2 *feat;
834 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
835 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
840 efer |= vcpu->arch.efer & EFER_LMA;
842 kvm_x86_ops->set_efer(vcpu, efer);
844 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
846 /* Update reserved bits */
847 if ((efer ^ old_efer) & EFER_NX)
848 kvm_mmu_reset_context(vcpu);
853 void kvm_enable_efer_bits(u64 mask)
855 efer_reserved_bits &= ~mask;
857 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
861 * Writes msr value into into the appropriate "register".
862 * Returns 0 on success, non-0 otherwise.
863 * Assumes vcpu_load() was already called.
865 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
867 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
871 * Adapt set_msr() to msr_io()'s calling convention
873 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
875 return kvm_set_msr(vcpu, index, *data);
878 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
882 struct pvclock_wall_clock wc;
883 struct timespec boot;
888 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
893 ++version; /* first time write, random junk */
897 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
900 * The guest calculates current wall clock time by adding
901 * system time (updated by kvm_guest_time_update below) to the
902 * wall clock specified here. guest system time equals host
903 * system time for us, thus we must fill in host boot time here.
907 wc.sec = boot.tv_sec;
908 wc.nsec = boot.tv_nsec;
909 wc.version = version;
911 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
914 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
917 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
919 uint32_t quotient, remainder;
921 /* Don't try to replace with do_div(), this one calculates
922 * "(dividend << 32) / divisor" */
924 : "=a" (quotient), "=d" (remainder)
925 : "0" (0), "1" (dividend), "r" (divisor) );
929 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
930 s8 *pshift, u32 *pmultiplier)
937 tps64 = base_khz * 1000LL;
938 scaled64 = scaled_khz * 1000LL;
939 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
944 tps32 = (uint32_t)tps64;
945 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
946 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
954 *pmultiplier = div_frac(scaled64, tps32);
956 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
957 __func__, base_khz, scaled_khz, shift, *pmultiplier);
960 static inline u64 get_kernel_ns(void)
964 WARN_ON(preemptible());
966 monotonic_to_bootbased(&ts);
967 return timespec_to_ns(&ts);
970 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
971 unsigned long max_tsc_khz;
973 static inline int kvm_tsc_changes_freq(void)
976 int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
977 cpufreq_quick_get(cpu) != 0;
982 static u64 vcpu_tsc_khz(struct kvm_vcpu *vcpu)
984 if (vcpu->arch.virtual_tsc_khz)
985 return vcpu->arch.virtual_tsc_khz;
987 return __this_cpu_read(cpu_tsc_khz);
990 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
994 WARN_ON(preemptible());
995 if (kvm_tsc_changes_freq())
996 printk_once(KERN_WARNING
997 "kvm: unreliable cycle conversion on adjustable rate TSC\n");
998 ret = nsec * vcpu_tsc_khz(vcpu);
999 do_div(ret, USEC_PER_SEC);
1003 static void kvm_init_tsc_catchup(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1005 /* Compute a scale to convert nanoseconds in TSC cycles */
1006 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1007 &vcpu->arch.tsc_catchup_shift,
1008 &vcpu->arch.tsc_catchup_mult);
1011 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1013 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1014 vcpu->arch.tsc_catchup_mult,
1015 vcpu->arch.tsc_catchup_shift);
1016 tsc += vcpu->arch.last_tsc_write;
1020 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1022 struct kvm *kvm = vcpu->kvm;
1023 u64 offset, ns, elapsed;
1024 unsigned long flags;
1027 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1028 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1029 ns = get_kernel_ns();
1030 elapsed = ns - kvm->arch.last_tsc_nsec;
1031 sdiff = data - kvm->arch.last_tsc_write;
1036 * Special case: close write to TSC within 5 seconds of
1037 * another CPU is interpreted as an attempt to synchronize
1038 * The 5 seconds is to accommodate host load / swapping as
1039 * well as any reset of TSC during the boot process.
1041 * In that case, for a reliable TSC, we can match TSC offsets,
1042 * or make a best guest using elapsed value.
1044 if (sdiff < nsec_to_cycles(vcpu, 5ULL * NSEC_PER_SEC) &&
1045 elapsed < 5ULL * NSEC_PER_SEC) {
1046 if (!check_tsc_unstable()) {
1047 offset = kvm->arch.last_tsc_offset;
1048 pr_debug("kvm: matched tsc offset for %llu\n", data);
1050 u64 delta = nsec_to_cycles(vcpu, elapsed);
1052 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1054 ns = kvm->arch.last_tsc_nsec;
1056 kvm->arch.last_tsc_nsec = ns;
1057 kvm->arch.last_tsc_write = data;
1058 kvm->arch.last_tsc_offset = offset;
1059 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1060 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1062 /* Reset of TSC must disable overshoot protection below */
1063 vcpu->arch.hv_clock.tsc_timestamp = 0;
1064 vcpu->arch.last_tsc_write = data;
1065 vcpu->arch.last_tsc_nsec = ns;
1067 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1069 static int kvm_guest_time_update(struct kvm_vcpu *v)
1071 unsigned long flags;
1072 struct kvm_vcpu_arch *vcpu = &v->arch;
1074 unsigned long this_tsc_khz;
1075 s64 kernel_ns, max_kernel_ns;
1078 /* Keep irq disabled to prevent changes to the clock */
1079 local_irq_save(flags);
1080 kvm_get_msr(v, MSR_IA32_TSC, &tsc_timestamp);
1081 kernel_ns = get_kernel_ns();
1082 this_tsc_khz = vcpu_tsc_khz(v);
1083 if (unlikely(this_tsc_khz == 0)) {
1084 local_irq_restore(flags);
1085 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1090 * We may have to catch up the TSC to match elapsed wall clock
1091 * time for two reasons, even if kvmclock is used.
1092 * 1) CPU could have been running below the maximum TSC rate
1093 * 2) Broken TSC compensation resets the base at each VCPU
1094 * entry to avoid unknown leaps of TSC even when running
1095 * again on the same CPU. This may cause apparent elapsed
1096 * time to disappear, and the guest to stand still or run
1099 if (vcpu->tsc_catchup) {
1100 u64 tsc = compute_guest_tsc(v, kernel_ns);
1101 if (tsc > tsc_timestamp) {
1102 kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1103 tsc_timestamp = tsc;
1107 local_irq_restore(flags);
1109 if (!vcpu->time_page)
1113 * Time as measured by the TSC may go backwards when resetting the base
1114 * tsc_timestamp. The reason for this is that the TSC resolution is
1115 * higher than the resolution of the other clock scales. Thus, many
1116 * possible measurments of the TSC correspond to one measurement of any
1117 * other clock, and so a spread of values is possible. This is not a
1118 * problem for the computation of the nanosecond clock; with TSC rates
1119 * around 1GHZ, there can only be a few cycles which correspond to one
1120 * nanosecond value, and any path through this code will inevitably
1121 * take longer than that. However, with the kernel_ns value itself,
1122 * the precision may be much lower, down to HZ granularity. If the
1123 * first sampling of TSC against kernel_ns ends in the low part of the
1124 * range, and the second in the high end of the range, we can get:
1126 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1128 * As the sampling errors potentially range in the thousands of cycles,
1129 * it is possible such a time value has already been observed by the
1130 * guest. To protect against this, we must compute the system time as
1131 * observed by the guest and ensure the new system time is greater.
1134 if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1135 max_kernel_ns = vcpu->last_guest_tsc -
1136 vcpu->hv_clock.tsc_timestamp;
1137 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1138 vcpu->hv_clock.tsc_to_system_mul,
1139 vcpu->hv_clock.tsc_shift);
1140 max_kernel_ns += vcpu->last_kernel_ns;
1143 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1144 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1145 &vcpu->hv_clock.tsc_shift,
1146 &vcpu->hv_clock.tsc_to_system_mul);
1147 vcpu->hw_tsc_khz = this_tsc_khz;
1150 if (max_kernel_ns > kernel_ns)
1151 kernel_ns = max_kernel_ns;
1153 /* With all the info we got, fill in the values */
1154 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1155 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1156 vcpu->last_kernel_ns = kernel_ns;
1157 vcpu->last_guest_tsc = tsc_timestamp;
1158 vcpu->hv_clock.flags = 0;
1161 * The interface expects us to write an even number signaling that the
1162 * update is finished. Since the guest won't see the intermediate
1163 * state, we just increase by 2 at the end.
1165 vcpu->hv_clock.version += 2;
1167 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1169 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1170 sizeof(vcpu->hv_clock));
1172 kunmap_atomic(shared_kaddr, KM_USER0);
1174 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1178 static bool msr_mtrr_valid(unsigned msr)
1181 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1182 case MSR_MTRRfix64K_00000:
1183 case MSR_MTRRfix16K_80000:
1184 case MSR_MTRRfix16K_A0000:
1185 case MSR_MTRRfix4K_C0000:
1186 case MSR_MTRRfix4K_C8000:
1187 case MSR_MTRRfix4K_D0000:
1188 case MSR_MTRRfix4K_D8000:
1189 case MSR_MTRRfix4K_E0000:
1190 case MSR_MTRRfix4K_E8000:
1191 case MSR_MTRRfix4K_F0000:
1192 case MSR_MTRRfix4K_F8000:
1193 case MSR_MTRRdefType:
1194 case MSR_IA32_CR_PAT:
1202 static bool valid_pat_type(unsigned t)
1204 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1207 static bool valid_mtrr_type(unsigned t)
1209 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1212 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1216 if (!msr_mtrr_valid(msr))
1219 if (msr == MSR_IA32_CR_PAT) {
1220 for (i = 0; i < 8; i++)
1221 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1224 } else if (msr == MSR_MTRRdefType) {
1227 return valid_mtrr_type(data & 0xff);
1228 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1229 for (i = 0; i < 8 ; i++)
1230 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1235 /* variable MTRRs */
1236 return valid_mtrr_type(data & 0xff);
1239 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1241 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1243 if (!mtrr_valid(vcpu, msr, data))
1246 if (msr == MSR_MTRRdefType) {
1247 vcpu->arch.mtrr_state.def_type = data;
1248 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1249 } else if (msr == MSR_MTRRfix64K_00000)
1251 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1252 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1253 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1254 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1255 else if (msr == MSR_IA32_CR_PAT)
1256 vcpu->arch.pat = data;
1257 else { /* Variable MTRRs */
1258 int idx, is_mtrr_mask;
1261 idx = (msr - 0x200) / 2;
1262 is_mtrr_mask = msr - 0x200 - 2 * idx;
1265 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1268 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1272 kvm_mmu_reset_context(vcpu);
1276 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1278 u64 mcg_cap = vcpu->arch.mcg_cap;
1279 unsigned bank_num = mcg_cap & 0xff;
1282 case MSR_IA32_MCG_STATUS:
1283 vcpu->arch.mcg_status = data;
1285 case MSR_IA32_MCG_CTL:
1286 if (!(mcg_cap & MCG_CTL_P))
1288 if (data != 0 && data != ~(u64)0)
1290 vcpu->arch.mcg_ctl = data;
1293 if (msr >= MSR_IA32_MC0_CTL &&
1294 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1295 u32 offset = msr - MSR_IA32_MC0_CTL;
1296 /* only 0 or all 1s can be written to IA32_MCi_CTL
1297 * some Linux kernels though clear bit 10 in bank 4 to
1298 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1299 * this to avoid an uncatched #GP in the guest
1301 if ((offset & 0x3) == 0 &&
1302 data != 0 && (data | (1 << 10)) != ~(u64)0)
1304 vcpu->arch.mce_banks[offset] = data;
1312 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1314 struct kvm *kvm = vcpu->kvm;
1315 int lm = is_long_mode(vcpu);
1316 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1317 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1318 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1319 : kvm->arch.xen_hvm_config.blob_size_32;
1320 u32 page_num = data & ~PAGE_MASK;
1321 u64 page_addr = data & PAGE_MASK;
1326 if (page_num >= blob_size)
1329 page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1333 if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1335 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1344 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1346 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1349 static bool kvm_hv_msr_partition_wide(u32 msr)
1353 case HV_X64_MSR_GUEST_OS_ID:
1354 case HV_X64_MSR_HYPERCALL:
1362 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1364 struct kvm *kvm = vcpu->kvm;
1367 case HV_X64_MSR_GUEST_OS_ID:
1368 kvm->arch.hv_guest_os_id = data;
1369 /* setting guest os id to zero disables hypercall page */
1370 if (!kvm->arch.hv_guest_os_id)
1371 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1373 case HV_X64_MSR_HYPERCALL: {
1378 /* if guest os id is not set hypercall should remain disabled */
1379 if (!kvm->arch.hv_guest_os_id)
1381 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1382 kvm->arch.hv_hypercall = data;
1385 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1386 addr = gfn_to_hva(kvm, gfn);
1387 if (kvm_is_error_hva(addr))
1389 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1390 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1391 if (copy_to_user((void __user *)addr, instructions, 4))
1393 kvm->arch.hv_hypercall = data;
1397 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1398 "data 0x%llx\n", msr, data);
1404 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1407 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1410 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1411 vcpu->arch.hv_vapic = data;
1414 addr = gfn_to_hva(vcpu->kvm, data >>
1415 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1416 if (kvm_is_error_hva(addr))
1418 if (clear_user((void __user *)addr, PAGE_SIZE))
1420 vcpu->arch.hv_vapic = data;
1423 case HV_X64_MSR_EOI:
1424 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1425 case HV_X64_MSR_ICR:
1426 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1427 case HV_X64_MSR_TPR:
1428 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1430 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1431 "data 0x%llx\n", msr, data);
1438 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1440 gpa_t gpa = data & ~0x3f;
1442 /* Bits 2:5 are resrved, Should be zero */
1446 vcpu->arch.apf.msr_val = data;
1448 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1449 kvm_clear_async_pf_completion_queue(vcpu);
1450 kvm_async_pf_hash_reset(vcpu);
1454 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1457 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1458 kvm_async_pf_wakeup_all(vcpu);
1462 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1464 if (vcpu->arch.time_page) {
1465 kvm_release_page_dirty(vcpu->arch.time_page);
1466 vcpu->arch.time_page = NULL;
1470 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1474 return set_efer(vcpu, data);
1476 data &= ~(u64)0x40; /* ignore flush filter disable */
1477 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1479 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1484 case MSR_FAM10H_MMIO_CONF_BASE:
1486 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1491 case MSR_AMD64_NB_CFG:
1493 case MSR_IA32_DEBUGCTLMSR:
1495 /* We support the non-activated case already */
1497 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1498 /* Values other than LBR and BTF are vendor-specific,
1499 thus reserved and should throw a #GP */
1502 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1505 case MSR_IA32_UCODE_REV:
1506 case MSR_IA32_UCODE_WRITE:
1507 case MSR_VM_HSAVE_PA:
1508 case MSR_AMD64_PATCH_LOADER:
1510 case 0x200 ... 0x2ff:
1511 return set_msr_mtrr(vcpu, msr, data);
1512 case MSR_IA32_APICBASE:
1513 kvm_set_apic_base(vcpu, data);
1515 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1516 return kvm_x2apic_msr_write(vcpu, msr, data);
1517 case MSR_IA32_MISC_ENABLE:
1518 vcpu->arch.ia32_misc_enable_msr = data;
1520 case MSR_KVM_WALL_CLOCK_NEW:
1521 case MSR_KVM_WALL_CLOCK:
1522 vcpu->kvm->arch.wall_clock = data;
1523 kvm_write_wall_clock(vcpu->kvm, data);
1525 case MSR_KVM_SYSTEM_TIME_NEW:
1526 case MSR_KVM_SYSTEM_TIME: {
1527 kvmclock_reset(vcpu);
1529 vcpu->arch.time = data;
1530 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1532 /* we verify if the enable bit is set... */
1536 /* ...but clean it before doing the actual write */
1537 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1539 vcpu->arch.time_page =
1540 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1542 if (is_error_page(vcpu->arch.time_page)) {
1543 kvm_release_page_clean(vcpu->arch.time_page);
1544 vcpu->arch.time_page = NULL;
1548 case MSR_KVM_ASYNC_PF_EN:
1549 if (kvm_pv_enable_async_pf(vcpu, data))
1552 case MSR_IA32_MCG_CTL:
1553 case MSR_IA32_MCG_STATUS:
1554 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1555 return set_msr_mce(vcpu, msr, data);
1557 /* Performance counters are not protected by a CPUID bit,
1558 * so we should check all of them in the generic path for the sake of
1559 * cross vendor migration.
1560 * Writing a zero into the event select MSRs disables them,
1561 * which we perfectly emulate ;-). Any other value should be at least
1562 * reported, some guests depend on them.
1564 case MSR_P6_EVNTSEL0:
1565 case MSR_P6_EVNTSEL1:
1566 case MSR_K7_EVNTSEL0:
1567 case MSR_K7_EVNTSEL1:
1568 case MSR_K7_EVNTSEL2:
1569 case MSR_K7_EVNTSEL3:
1571 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1572 "0x%x data 0x%llx\n", msr, data);
1574 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1575 * so we ignore writes to make it happy.
1577 case MSR_P6_PERFCTR0:
1578 case MSR_P6_PERFCTR1:
1579 case MSR_K7_PERFCTR0:
1580 case MSR_K7_PERFCTR1:
1581 case MSR_K7_PERFCTR2:
1582 case MSR_K7_PERFCTR3:
1583 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1584 "0x%x data 0x%llx\n", msr, data);
1586 case MSR_K7_CLK_CTL:
1588 * Ignore all writes to this no longer documented MSR.
1589 * Writes are only relevant for old K7 processors,
1590 * all pre-dating SVM, but a recommended workaround from
1591 * AMD for these chips. It is possible to speicify the
1592 * affected processor models on the command line, hence
1593 * the need to ignore the workaround.
1596 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1597 if (kvm_hv_msr_partition_wide(msr)) {
1599 mutex_lock(&vcpu->kvm->lock);
1600 r = set_msr_hyperv_pw(vcpu, msr, data);
1601 mutex_unlock(&vcpu->kvm->lock);
1604 return set_msr_hyperv(vcpu, msr, data);
1606 case MSR_IA32_BBL_CR_CTL3:
1607 /* Drop writes to this legacy MSR -- see rdmsr
1608 * counterpart for further detail.
1610 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1613 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1614 return xen_hvm_config(vcpu, data);
1616 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1620 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1627 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1631 * Reads an msr value (of 'msr_index') into 'pdata'.
1632 * Returns 0 on success, non-0 otherwise.
1633 * Assumes vcpu_load() was already called.
1635 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1637 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1640 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1642 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1644 if (!msr_mtrr_valid(msr))
1647 if (msr == MSR_MTRRdefType)
1648 *pdata = vcpu->arch.mtrr_state.def_type +
1649 (vcpu->arch.mtrr_state.enabled << 10);
1650 else if (msr == MSR_MTRRfix64K_00000)
1652 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1653 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1654 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1655 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1656 else if (msr == MSR_IA32_CR_PAT)
1657 *pdata = vcpu->arch.pat;
1658 else { /* Variable MTRRs */
1659 int idx, is_mtrr_mask;
1662 idx = (msr - 0x200) / 2;
1663 is_mtrr_mask = msr - 0x200 - 2 * idx;
1666 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1669 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1676 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1679 u64 mcg_cap = vcpu->arch.mcg_cap;
1680 unsigned bank_num = mcg_cap & 0xff;
1683 case MSR_IA32_P5_MC_ADDR:
1684 case MSR_IA32_P5_MC_TYPE:
1687 case MSR_IA32_MCG_CAP:
1688 data = vcpu->arch.mcg_cap;
1690 case MSR_IA32_MCG_CTL:
1691 if (!(mcg_cap & MCG_CTL_P))
1693 data = vcpu->arch.mcg_ctl;
1695 case MSR_IA32_MCG_STATUS:
1696 data = vcpu->arch.mcg_status;
1699 if (msr >= MSR_IA32_MC0_CTL &&
1700 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1701 u32 offset = msr - MSR_IA32_MC0_CTL;
1702 data = vcpu->arch.mce_banks[offset];
1711 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1714 struct kvm *kvm = vcpu->kvm;
1717 case HV_X64_MSR_GUEST_OS_ID:
1718 data = kvm->arch.hv_guest_os_id;
1720 case HV_X64_MSR_HYPERCALL:
1721 data = kvm->arch.hv_hypercall;
1724 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1732 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1737 case HV_X64_MSR_VP_INDEX: {
1740 kvm_for_each_vcpu(r, v, vcpu->kvm)
1745 case HV_X64_MSR_EOI:
1746 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1747 case HV_X64_MSR_ICR:
1748 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1749 case HV_X64_MSR_TPR:
1750 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1752 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1759 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1764 case MSR_IA32_PLATFORM_ID:
1765 case MSR_IA32_UCODE_REV:
1766 case MSR_IA32_EBL_CR_POWERON:
1767 case MSR_IA32_DEBUGCTLMSR:
1768 case MSR_IA32_LASTBRANCHFROMIP:
1769 case MSR_IA32_LASTBRANCHTOIP:
1770 case MSR_IA32_LASTINTFROMIP:
1771 case MSR_IA32_LASTINTTOIP:
1774 case MSR_VM_HSAVE_PA:
1775 case MSR_P6_PERFCTR0:
1776 case MSR_P6_PERFCTR1:
1777 case MSR_P6_EVNTSEL0:
1778 case MSR_P6_EVNTSEL1:
1779 case MSR_K7_EVNTSEL0:
1780 case MSR_K7_PERFCTR0:
1781 case MSR_K8_INT_PENDING_MSG:
1782 case MSR_AMD64_NB_CFG:
1783 case MSR_FAM10H_MMIO_CONF_BASE:
1787 data = 0x500 | KVM_NR_VAR_MTRR;
1789 case 0x200 ... 0x2ff:
1790 return get_msr_mtrr(vcpu, msr, pdata);
1791 case 0xcd: /* fsb frequency */
1795 * MSR_EBC_FREQUENCY_ID
1796 * Conservative value valid for even the basic CPU models.
1797 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1798 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1799 * and 266MHz for model 3, or 4. Set Core Clock
1800 * Frequency to System Bus Frequency Ratio to 1 (bits
1801 * 31:24) even though these are only valid for CPU
1802 * models > 2, however guests may end up dividing or
1803 * multiplying by zero otherwise.
1805 case MSR_EBC_FREQUENCY_ID:
1808 case MSR_IA32_APICBASE:
1809 data = kvm_get_apic_base(vcpu);
1811 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1812 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1814 case MSR_IA32_MISC_ENABLE:
1815 data = vcpu->arch.ia32_misc_enable_msr;
1817 case MSR_IA32_PERF_STATUS:
1818 /* TSC increment by tick */
1820 /* CPU multiplier */
1821 data |= (((uint64_t)4ULL) << 40);
1824 data = vcpu->arch.efer;
1826 case MSR_KVM_WALL_CLOCK:
1827 case MSR_KVM_WALL_CLOCK_NEW:
1828 data = vcpu->kvm->arch.wall_clock;
1830 case MSR_KVM_SYSTEM_TIME:
1831 case MSR_KVM_SYSTEM_TIME_NEW:
1832 data = vcpu->arch.time;
1834 case MSR_KVM_ASYNC_PF_EN:
1835 data = vcpu->arch.apf.msr_val;
1837 case MSR_IA32_P5_MC_ADDR:
1838 case MSR_IA32_P5_MC_TYPE:
1839 case MSR_IA32_MCG_CAP:
1840 case MSR_IA32_MCG_CTL:
1841 case MSR_IA32_MCG_STATUS:
1842 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1843 return get_msr_mce(vcpu, msr, pdata);
1844 case MSR_K7_CLK_CTL:
1846 * Provide expected ramp-up count for K7. All other
1847 * are set to zero, indicating minimum divisors for
1850 * This prevents guest kernels on AMD host with CPU
1851 * type 6, model 8 and higher from exploding due to
1852 * the rdmsr failing.
1856 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1857 if (kvm_hv_msr_partition_wide(msr)) {
1859 mutex_lock(&vcpu->kvm->lock);
1860 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1861 mutex_unlock(&vcpu->kvm->lock);
1864 return get_msr_hyperv(vcpu, msr, pdata);
1866 case MSR_IA32_BBL_CR_CTL3:
1867 /* This legacy MSR exists but isn't fully documented in current
1868 * silicon. It is however accessed by winxp in very narrow
1869 * scenarios where it sets bit #19, itself documented as
1870 * a "reserved" bit. Best effort attempt to source coherent
1871 * read data here should the balance of the register be
1872 * interpreted by the guest:
1874 * L2 cache control register 3: 64GB range, 256KB size,
1875 * enabled, latency 0x1, configured
1881 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1884 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1892 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1895 * Read or write a bunch of msrs. All parameters are kernel addresses.
1897 * @return number of msrs set successfully.
1899 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1900 struct kvm_msr_entry *entries,
1901 int (*do_msr)(struct kvm_vcpu *vcpu,
1902 unsigned index, u64 *data))
1906 idx = srcu_read_lock(&vcpu->kvm->srcu);
1907 for (i = 0; i < msrs->nmsrs; ++i)
1908 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1910 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1916 * Read or write a bunch of msrs. Parameters are user addresses.
1918 * @return number of msrs set successfully.
1920 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1921 int (*do_msr)(struct kvm_vcpu *vcpu,
1922 unsigned index, u64 *data),
1925 struct kvm_msrs msrs;
1926 struct kvm_msr_entry *entries;
1931 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1935 if (msrs.nmsrs >= MAX_IO_MSRS)
1939 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1940 entries = kmalloc(size, GFP_KERNEL);
1945 if (copy_from_user(entries, user_msrs->entries, size))
1948 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1953 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1964 int kvm_dev_ioctl_check_extension(long ext)
1969 case KVM_CAP_IRQCHIP:
1971 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1972 case KVM_CAP_SET_TSS_ADDR:
1973 case KVM_CAP_EXT_CPUID:
1974 case KVM_CAP_CLOCKSOURCE:
1976 case KVM_CAP_NOP_IO_DELAY:
1977 case KVM_CAP_MP_STATE:
1978 case KVM_CAP_SYNC_MMU:
1979 case KVM_CAP_USER_NMI:
1980 case KVM_CAP_REINJECT_CONTROL:
1981 case KVM_CAP_IRQ_INJECT_STATUS:
1982 case KVM_CAP_ASSIGN_DEV_IRQ:
1984 case KVM_CAP_IOEVENTFD:
1986 case KVM_CAP_PIT_STATE2:
1987 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1988 case KVM_CAP_XEN_HVM:
1989 case KVM_CAP_ADJUST_CLOCK:
1990 case KVM_CAP_VCPU_EVENTS:
1991 case KVM_CAP_HYPERV:
1992 case KVM_CAP_HYPERV_VAPIC:
1993 case KVM_CAP_HYPERV_SPIN:
1994 case KVM_CAP_PCI_SEGMENT:
1995 case KVM_CAP_DEBUGREGS:
1996 case KVM_CAP_X86_ROBUST_SINGLESTEP:
1998 case KVM_CAP_ASYNC_PF:
1999 case KVM_CAP_GET_TSC_KHZ:
2002 case KVM_CAP_COALESCED_MMIO:
2003 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2006 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2008 case KVM_CAP_NR_VCPUS:
2011 case KVM_CAP_NR_MEMSLOTS:
2012 r = KVM_MEMORY_SLOTS;
2014 case KVM_CAP_PV_MMU: /* obsolete */
2021 r = KVM_MAX_MCE_BANKS;
2026 case KVM_CAP_TSC_CONTROL:
2027 r = kvm_has_tsc_control;
2037 long kvm_arch_dev_ioctl(struct file *filp,
2038 unsigned int ioctl, unsigned long arg)
2040 void __user *argp = (void __user *)arg;
2044 case KVM_GET_MSR_INDEX_LIST: {
2045 struct kvm_msr_list __user *user_msr_list = argp;
2046 struct kvm_msr_list msr_list;
2050 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2053 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2054 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2057 if (n < msr_list.nmsrs)
2060 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2061 num_msrs_to_save * sizeof(u32)))
2063 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2065 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2070 case KVM_GET_SUPPORTED_CPUID: {
2071 struct kvm_cpuid2 __user *cpuid_arg = argp;
2072 struct kvm_cpuid2 cpuid;
2075 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2077 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2078 cpuid_arg->entries);
2083 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2088 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2091 mce_cap = KVM_MCE_CAP_SUPPORTED;
2093 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2105 static void wbinvd_ipi(void *garbage)
2110 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2112 return vcpu->kvm->arch.iommu_domain &&
2113 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2116 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2118 /* Address WBINVD may be executed by guest */
2119 if (need_emulate_wbinvd(vcpu)) {
2120 if (kvm_x86_ops->has_wbinvd_exit())
2121 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2122 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2123 smp_call_function_single(vcpu->cpu,
2124 wbinvd_ipi, NULL, 1);
2127 kvm_x86_ops->vcpu_load(vcpu, cpu);
2128 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2129 /* Make sure TSC doesn't go backwards */
2133 kvm_get_msr(vcpu, MSR_IA32_TSC, &tsc);
2134 tsc_delta = !vcpu->arch.last_guest_tsc ? 0 :
2135 tsc - vcpu->arch.last_guest_tsc;
2138 mark_tsc_unstable("KVM discovered backwards TSC");
2139 if (check_tsc_unstable()) {
2140 kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2141 vcpu->arch.tsc_catchup = 1;
2143 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2144 if (vcpu->cpu != cpu)
2145 kvm_migrate_timers(vcpu);
2150 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2152 kvm_x86_ops->vcpu_put(vcpu);
2153 kvm_put_guest_fpu(vcpu);
2154 kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
2157 static int is_efer_nx(void)
2159 unsigned long long efer = 0;
2161 rdmsrl_safe(MSR_EFER, &efer);
2162 return efer & EFER_NX;
2165 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2168 struct kvm_cpuid_entry2 *e, *entry;
2171 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2172 e = &vcpu->arch.cpuid_entries[i];
2173 if (e->function == 0x80000001) {
2178 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2179 entry->edx &= ~(1 << 20);
2180 printk(KERN_INFO "kvm: guest NX capability removed\n");
2184 /* when an old userspace process fills a new kernel module */
2185 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2186 struct kvm_cpuid *cpuid,
2187 struct kvm_cpuid_entry __user *entries)
2190 struct kvm_cpuid_entry *cpuid_entries;
2193 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2196 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2200 if (copy_from_user(cpuid_entries, entries,
2201 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2203 for (i = 0; i < cpuid->nent; i++) {
2204 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2205 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2206 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2207 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2208 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2209 vcpu->arch.cpuid_entries[i].index = 0;
2210 vcpu->arch.cpuid_entries[i].flags = 0;
2211 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2212 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2213 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2215 vcpu->arch.cpuid_nent = cpuid->nent;
2216 cpuid_fix_nx_cap(vcpu);
2218 kvm_apic_set_version(vcpu);
2219 kvm_x86_ops->cpuid_update(vcpu);
2223 vfree(cpuid_entries);
2228 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2229 struct kvm_cpuid2 *cpuid,
2230 struct kvm_cpuid_entry2 __user *entries)
2235 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2238 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2239 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2241 vcpu->arch.cpuid_nent = cpuid->nent;
2242 kvm_apic_set_version(vcpu);
2243 kvm_x86_ops->cpuid_update(vcpu);
2251 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2252 struct kvm_cpuid2 *cpuid,
2253 struct kvm_cpuid_entry2 __user *entries)
2258 if (cpuid->nent < vcpu->arch.cpuid_nent)
2261 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2262 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2267 cpuid->nent = vcpu->arch.cpuid_nent;
2271 static void cpuid_mask(u32 *word, int wordnum)
2273 *word &= boot_cpu_data.x86_capability[wordnum];
2276 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2279 entry->function = function;
2280 entry->index = index;
2281 cpuid_count(entry->function, entry->index,
2282 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2286 #define F(x) bit(X86_FEATURE_##x)
2288 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2289 u32 index, int *nent, int maxnent)
2291 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2292 #ifdef CONFIG_X86_64
2293 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2295 unsigned f_lm = F(LM);
2297 unsigned f_gbpages = 0;
2300 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2303 const u32 kvm_supported_word0_x86_features =
2304 F(FPU) | F(VME) | F(DE) | F(PSE) |
2305 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2306 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2307 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2308 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2309 0 /* Reserved, DS, ACPI */ | F(MMX) |
2310 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2311 0 /* HTT, TM, Reserved, PBE */;
2312 /* cpuid 0x80000001.edx */
2313 const u32 kvm_supported_word1_x86_features =
2314 F(FPU) | F(VME) | F(DE) | F(PSE) |
2315 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2316 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2317 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2318 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2319 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2320 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2321 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2323 const u32 kvm_supported_word4_x86_features =
2324 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
2325 0 /* DS-CPL, VMX, SMX, EST */ |
2326 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2327 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2328 0 /* Reserved, DCA */ | F(XMM4_1) |
2329 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2330 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2332 /* cpuid 0x80000001.ecx */
2333 const u32 kvm_supported_word6_x86_features =
2334 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2335 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2336 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2337 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2339 /* cpuid 0xC0000001.edx */
2340 const u32 kvm_supported_word5_x86_features =
2341 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
2342 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
2345 /* all calls to cpuid_count() should be made on the same cpu */
2347 do_cpuid_1_ent(entry, function, index);
2352 entry->eax = min(entry->eax, (u32)0xd);
2355 entry->edx &= kvm_supported_word0_x86_features;
2356 cpuid_mask(&entry->edx, 0);
2357 entry->ecx &= kvm_supported_word4_x86_features;
2358 cpuid_mask(&entry->ecx, 4);
2359 /* we support x2apic emulation even if host does not support
2360 * it since we emulate x2apic in software */
2361 entry->ecx |= F(X2APIC);
2363 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2364 * may return different values. This forces us to get_cpu() before
2365 * issuing the first command, and also to emulate this annoying behavior
2366 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2368 int t, times = entry->eax & 0xff;
2370 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2371 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2372 for (t = 1; t < times && *nent < maxnent; ++t) {
2373 do_cpuid_1_ent(&entry[t], function, 0);
2374 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2379 /* function 4 and 0xb have additional index. */
2383 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2384 /* read more entries until cache_type is zero */
2385 for (i = 1; *nent < maxnent; ++i) {
2386 cache_type = entry[i - 1].eax & 0x1f;
2389 do_cpuid_1_ent(&entry[i], function, i);
2391 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2399 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2400 /* read more entries until level_type is zero */
2401 for (i = 1; *nent < maxnent; ++i) {
2402 level_type = entry[i - 1].ecx & 0xff00;
2405 do_cpuid_1_ent(&entry[i], function, i);
2407 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2415 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2416 for (i = 1; *nent < maxnent && i < 64; ++i) {
2417 if (entry[i].eax == 0)
2419 do_cpuid_1_ent(&entry[i], function, i);
2421 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2426 case KVM_CPUID_SIGNATURE: {
2427 char signature[12] = "KVMKVMKVM\0\0";
2428 u32 *sigptr = (u32 *)signature;
2430 entry->ebx = sigptr[0];
2431 entry->ecx = sigptr[1];
2432 entry->edx = sigptr[2];
2435 case KVM_CPUID_FEATURES:
2436 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2437 (1 << KVM_FEATURE_NOP_IO_DELAY) |
2438 (1 << KVM_FEATURE_CLOCKSOURCE2) |
2439 (1 << KVM_FEATURE_ASYNC_PF) |
2440 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2446 entry->eax = min(entry->eax, 0x8000001a);
2449 entry->edx &= kvm_supported_word1_x86_features;
2450 cpuid_mask(&entry->edx, 1);
2451 entry->ecx &= kvm_supported_word6_x86_features;
2452 cpuid_mask(&entry->ecx, 6);
2454 /*Add support for Centaur's CPUID instruction*/
2456 /*Just support up to 0xC0000004 now*/
2457 entry->eax = min(entry->eax, 0xC0000004);
2460 entry->edx &= kvm_supported_word5_x86_features;
2461 cpuid_mask(&entry->edx, 5);
2466 /*Now nothing to do, reserved for the future*/
2470 kvm_x86_ops->set_supported_cpuid(function, entry);
2477 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2478 struct kvm_cpuid_entry2 __user *entries)
2480 struct kvm_cpuid_entry2 *cpuid_entries;
2481 int limit, nent = 0, r = -E2BIG;
2484 if (cpuid->nent < 1)
2486 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2487 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2489 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2493 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2494 limit = cpuid_entries[0].eax;
2495 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2496 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2497 &nent, cpuid->nent);
2499 if (nent >= cpuid->nent)
2502 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2503 limit = cpuid_entries[nent - 1].eax;
2504 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2505 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2506 &nent, cpuid->nent);
2511 if (nent >= cpuid->nent)
2514 /* Add support for Centaur's CPUID instruction. */
2515 if (boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR) {
2516 do_cpuid_ent(&cpuid_entries[nent], 0xC0000000, 0,
2517 &nent, cpuid->nent);
2520 if (nent >= cpuid->nent)
2523 limit = cpuid_entries[nent - 1].eax;
2524 for (func = 0xC0000001;
2525 func <= limit && nent < cpuid->nent; ++func)
2526 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2527 &nent, cpuid->nent);
2530 if (nent >= cpuid->nent)
2534 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2538 if (nent >= cpuid->nent)
2541 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2545 if (nent >= cpuid->nent)
2549 if (copy_to_user(entries, cpuid_entries,
2550 nent * sizeof(struct kvm_cpuid_entry2)))
2556 vfree(cpuid_entries);
2561 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2562 struct kvm_lapic_state *s)
2564 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2569 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2570 struct kvm_lapic_state *s)
2572 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2573 kvm_apic_post_state_restore(vcpu);
2574 update_cr8_intercept(vcpu);
2579 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2580 struct kvm_interrupt *irq)
2582 if (irq->irq < 0 || irq->irq >= 256)
2584 if (irqchip_in_kernel(vcpu->kvm))
2587 kvm_queue_interrupt(vcpu, irq->irq, false);
2588 kvm_make_request(KVM_REQ_EVENT, vcpu);
2593 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2595 kvm_inject_nmi(vcpu);
2600 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2601 struct kvm_tpr_access_ctl *tac)
2605 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2609 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2613 unsigned bank_num = mcg_cap & 0xff, bank;
2616 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2618 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2621 vcpu->arch.mcg_cap = mcg_cap;
2622 /* Init IA32_MCG_CTL to all 1s */
2623 if (mcg_cap & MCG_CTL_P)
2624 vcpu->arch.mcg_ctl = ~(u64)0;
2625 /* Init IA32_MCi_CTL to all 1s */
2626 for (bank = 0; bank < bank_num; bank++)
2627 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2632 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2633 struct kvm_x86_mce *mce)
2635 u64 mcg_cap = vcpu->arch.mcg_cap;
2636 unsigned bank_num = mcg_cap & 0xff;
2637 u64 *banks = vcpu->arch.mce_banks;
2639 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2642 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2643 * reporting is disabled
2645 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2646 vcpu->arch.mcg_ctl != ~(u64)0)
2648 banks += 4 * mce->bank;
2650 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2651 * reporting is disabled for the bank
2653 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2655 if (mce->status & MCI_STATUS_UC) {
2656 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2657 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2658 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2661 if (banks[1] & MCI_STATUS_VAL)
2662 mce->status |= MCI_STATUS_OVER;
2663 banks[2] = mce->addr;
2664 banks[3] = mce->misc;
2665 vcpu->arch.mcg_status = mce->mcg_status;
2666 banks[1] = mce->status;
2667 kvm_queue_exception(vcpu, MC_VECTOR);
2668 } else if (!(banks[1] & MCI_STATUS_VAL)
2669 || !(banks[1] & MCI_STATUS_UC)) {
2670 if (banks[1] & MCI_STATUS_VAL)
2671 mce->status |= MCI_STATUS_OVER;
2672 banks[2] = mce->addr;
2673 banks[3] = mce->misc;
2674 banks[1] = mce->status;
2676 banks[1] |= MCI_STATUS_OVER;
2680 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2681 struct kvm_vcpu_events *events)
2683 events->exception.injected =
2684 vcpu->arch.exception.pending &&
2685 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2686 events->exception.nr = vcpu->arch.exception.nr;
2687 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2688 events->exception.pad = 0;
2689 events->exception.error_code = vcpu->arch.exception.error_code;
2691 events->interrupt.injected =
2692 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2693 events->interrupt.nr = vcpu->arch.interrupt.nr;
2694 events->interrupt.soft = 0;
2695 events->interrupt.shadow =
2696 kvm_x86_ops->get_interrupt_shadow(vcpu,
2697 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2699 events->nmi.injected = vcpu->arch.nmi_injected;
2700 events->nmi.pending = vcpu->arch.nmi_pending;
2701 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2702 events->nmi.pad = 0;
2704 events->sipi_vector = vcpu->arch.sipi_vector;
2706 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2707 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2708 | KVM_VCPUEVENT_VALID_SHADOW);
2709 memset(&events->reserved, 0, sizeof(events->reserved));
2712 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2713 struct kvm_vcpu_events *events)
2715 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2716 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2717 | KVM_VCPUEVENT_VALID_SHADOW))
2720 vcpu->arch.exception.pending = events->exception.injected;
2721 vcpu->arch.exception.nr = events->exception.nr;
2722 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2723 vcpu->arch.exception.error_code = events->exception.error_code;
2725 vcpu->arch.interrupt.pending = events->interrupt.injected;
2726 vcpu->arch.interrupt.nr = events->interrupt.nr;
2727 vcpu->arch.interrupt.soft = events->interrupt.soft;
2728 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2729 kvm_x86_ops->set_interrupt_shadow(vcpu,
2730 events->interrupt.shadow);
2732 vcpu->arch.nmi_injected = events->nmi.injected;
2733 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2734 vcpu->arch.nmi_pending = events->nmi.pending;
2735 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2737 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2738 vcpu->arch.sipi_vector = events->sipi_vector;
2740 kvm_make_request(KVM_REQ_EVENT, vcpu);
2745 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2746 struct kvm_debugregs *dbgregs)
2748 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2749 dbgregs->dr6 = vcpu->arch.dr6;
2750 dbgregs->dr7 = vcpu->arch.dr7;
2752 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2755 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2756 struct kvm_debugregs *dbgregs)
2761 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2762 vcpu->arch.dr6 = dbgregs->dr6;
2763 vcpu->arch.dr7 = dbgregs->dr7;
2768 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2769 struct kvm_xsave *guest_xsave)
2772 memcpy(guest_xsave->region,
2773 &vcpu->arch.guest_fpu.state->xsave,
2776 memcpy(guest_xsave->region,
2777 &vcpu->arch.guest_fpu.state->fxsave,
2778 sizeof(struct i387_fxsave_struct));
2779 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2784 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2785 struct kvm_xsave *guest_xsave)
2788 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2791 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2792 guest_xsave->region, xstate_size);
2794 if (xstate_bv & ~XSTATE_FPSSE)
2796 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2797 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2802 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2803 struct kvm_xcrs *guest_xcrs)
2805 if (!cpu_has_xsave) {
2806 guest_xcrs->nr_xcrs = 0;
2810 guest_xcrs->nr_xcrs = 1;
2811 guest_xcrs->flags = 0;
2812 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2813 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2816 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2817 struct kvm_xcrs *guest_xcrs)
2824 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2827 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2828 /* Only support XCR0 currently */
2829 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2830 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2831 guest_xcrs->xcrs[0].value);
2839 long kvm_arch_vcpu_ioctl(struct file *filp,
2840 unsigned int ioctl, unsigned long arg)
2842 struct kvm_vcpu *vcpu = filp->private_data;
2843 void __user *argp = (void __user *)arg;
2846 struct kvm_lapic_state *lapic;
2847 struct kvm_xsave *xsave;
2848 struct kvm_xcrs *xcrs;
2854 case KVM_GET_LAPIC: {
2856 if (!vcpu->arch.apic)
2858 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2863 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
2867 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
2872 case KVM_SET_LAPIC: {
2874 if (!vcpu->arch.apic)
2876 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2881 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
2883 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
2889 case KVM_INTERRUPT: {
2890 struct kvm_interrupt irq;
2893 if (copy_from_user(&irq, argp, sizeof irq))
2895 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2902 r = kvm_vcpu_ioctl_nmi(vcpu);
2908 case KVM_SET_CPUID: {
2909 struct kvm_cpuid __user *cpuid_arg = argp;
2910 struct kvm_cpuid cpuid;
2913 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2915 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2920 case KVM_SET_CPUID2: {
2921 struct kvm_cpuid2 __user *cpuid_arg = argp;
2922 struct kvm_cpuid2 cpuid;
2925 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2927 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2928 cpuid_arg->entries);
2933 case KVM_GET_CPUID2: {
2934 struct kvm_cpuid2 __user *cpuid_arg = argp;
2935 struct kvm_cpuid2 cpuid;
2938 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2940 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2941 cpuid_arg->entries);
2945 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2951 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2954 r = msr_io(vcpu, argp, do_set_msr, 0);
2956 case KVM_TPR_ACCESS_REPORTING: {
2957 struct kvm_tpr_access_ctl tac;
2960 if (copy_from_user(&tac, argp, sizeof tac))
2962 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2966 if (copy_to_user(argp, &tac, sizeof tac))
2971 case KVM_SET_VAPIC_ADDR: {
2972 struct kvm_vapic_addr va;
2975 if (!irqchip_in_kernel(vcpu->kvm))
2978 if (copy_from_user(&va, argp, sizeof va))
2981 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2984 case KVM_X86_SETUP_MCE: {
2988 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2990 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2993 case KVM_X86_SET_MCE: {
2994 struct kvm_x86_mce mce;
2997 if (copy_from_user(&mce, argp, sizeof mce))
2999 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3002 case KVM_GET_VCPU_EVENTS: {
3003 struct kvm_vcpu_events events;
3005 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3008 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3013 case KVM_SET_VCPU_EVENTS: {
3014 struct kvm_vcpu_events events;
3017 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3020 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3023 case KVM_GET_DEBUGREGS: {
3024 struct kvm_debugregs dbgregs;
3026 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3029 if (copy_to_user(argp, &dbgregs,
3030 sizeof(struct kvm_debugregs)))
3035 case KVM_SET_DEBUGREGS: {
3036 struct kvm_debugregs dbgregs;
3039 if (copy_from_user(&dbgregs, argp,
3040 sizeof(struct kvm_debugregs)))
3043 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3046 case KVM_GET_XSAVE: {
3047 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3052 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3055 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3060 case KVM_SET_XSAVE: {
3061 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3067 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
3070 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3073 case KVM_GET_XCRS: {
3074 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3079 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3082 if (copy_to_user(argp, u.xcrs,
3083 sizeof(struct kvm_xcrs)))
3088 case KVM_SET_XCRS: {
3089 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3095 if (copy_from_user(u.xcrs, argp,
3096 sizeof(struct kvm_xcrs)))
3099 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3102 case KVM_SET_TSC_KHZ: {
3106 if (!kvm_has_tsc_control)
3109 user_tsc_khz = (u32)arg;
3111 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3114 kvm_x86_ops->set_tsc_khz(vcpu, user_tsc_khz);
3119 case KVM_GET_TSC_KHZ: {
3121 if (check_tsc_unstable())
3124 r = vcpu_tsc_khz(vcpu);
3136 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3140 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3142 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3146 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3149 kvm->arch.ept_identity_map_addr = ident_addr;
3153 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3154 u32 kvm_nr_mmu_pages)
3156 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3159 mutex_lock(&kvm->slots_lock);
3160 spin_lock(&kvm->mmu_lock);
3162 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3163 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3165 spin_unlock(&kvm->mmu_lock);
3166 mutex_unlock(&kvm->slots_lock);
3170 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3172 return kvm->arch.n_max_mmu_pages;
3175 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3180 switch (chip->chip_id) {
3181 case KVM_IRQCHIP_PIC_MASTER:
3182 memcpy(&chip->chip.pic,
3183 &pic_irqchip(kvm)->pics[0],
3184 sizeof(struct kvm_pic_state));
3186 case KVM_IRQCHIP_PIC_SLAVE:
3187 memcpy(&chip->chip.pic,
3188 &pic_irqchip(kvm)->pics[1],
3189 sizeof(struct kvm_pic_state));
3191 case KVM_IRQCHIP_IOAPIC:
3192 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3201 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3206 switch (chip->chip_id) {
3207 case KVM_IRQCHIP_PIC_MASTER:
3208 spin_lock(&pic_irqchip(kvm)->lock);
3209 memcpy(&pic_irqchip(kvm)->pics[0],
3211 sizeof(struct kvm_pic_state));
3212 spin_unlock(&pic_irqchip(kvm)->lock);
3214 case KVM_IRQCHIP_PIC_SLAVE:
3215 spin_lock(&pic_irqchip(kvm)->lock);
3216 memcpy(&pic_irqchip(kvm)->pics[1],
3218 sizeof(struct kvm_pic_state));
3219 spin_unlock(&pic_irqchip(kvm)->lock);
3221 case KVM_IRQCHIP_IOAPIC:
3222 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3228 kvm_pic_update_irq(pic_irqchip(kvm));
3232 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3236 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3237 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3238 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3242 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3246 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3247 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3248 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3249 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3253 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3257 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3258 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3259 sizeof(ps->channels));
3260 ps->flags = kvm->arch.vpit->pit_state.flags;
3261 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3262 memset(&ps->reserved, 0, sizeof(ps->reserved));
3266 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3268 int r = 0, start = 0;
3269 u32 prev_legacy, cur_legacy;
3270 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3271 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3272 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3273 if (!prev_legacy && cur_legacy)
3275 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3276 sizeof(kvm->arch.vpit->pit_state.channels));
3277 kvm->arch.vpit->pit_state.flags = ps->flags;
3278 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3279 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3283 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3284 struct kvm_reinject_control *control)
3286 if (!kvm->arch.vpit)
3288 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3289 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3290 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3295 * Get (and clear) the dirty memory log for a memory slot.
3297 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3298 struct kvm_dirty_log *log)
3301 struct kvm_memory_slot *memslot;
3303 unsigned long is_dirty = 0;
3305 mutex_lock(&kvm->slots_lock);
3308 if (log->slot >= KVM_MEMORY_SLOTS)
3311 memslot = &kvm->memslots->memslots[log->slot];
3313 if (!memslot->dirty_bitmap)
3316 n = kvm_dirty_bitmap_bytes(memslot);
3318 for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3319 is_dirty = memslot->dirty_bitmap[i];
3321 /* If nothing is dirty, don't bother messing with page tables. */
3323 struct kvm_memslots *slots, *old_slots;
3324 unsigned long *dirty_bitmap;
3326 dirty_bitmap = memslot->dirty_bitmap_head;
3327 if (memslot->dirty_bitmap == dirty_bitmap)
3328 dirty_bitmap += n / sizeof(long);
3329 memset(dirty_bitmap, 0, n);
3332 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3335 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3336 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3337 slots->generation++;
3339 old_slots = kvm->memslots;
3340 rcu_assign_pointer(kvm->memslots, slots);
3341 synchronize_srcu_expedited(&kvm->srcu);
3342 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3345 spin_lock(&kvm->mmu_lock);
3346 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3347 spin_unlock(&kvm->mmu_lock);
3350 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
3354 if (clear_user(log->dirty_bitmap, n))
3360 mutex_unlock(&kvm->slots_lock);
3364 long kvm_arch_vm_ioctl(struct file *filp,
3365 unsigned int ioctl, unsigned long arg)
3367 struct kvm *kvm = filp->private_data;
3368 void __user *argp = (void __user *)arg;
3371 * This union makes it completely explicit to gcc-3.x
3372 * that these two variables' stack usage should be
3373 * combined, not added together.
3376 struct kvm_pit_state ps;
3377 struct kvm_pit_state2 ps2;
3378 struct kvm_pit_config pit_config;
3382 case KVM_SET_TSS_ADDR:
3383 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3387 case KVM_SET_IDENTITY_MAP_ADDR: {
3391 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3393 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3398 case KVM_SET_NR_MMU_PAGES:
3399 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3403 case KVM_GET_NR_MMU_PAGES:
3404 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3406 case KVM_CREATE_IRQCHIP: {
3407 struct kvm_pic *vpic;
3409 mutex_lock(&kvm->lock);
3412 goto create_irqchip_unlock;
3414 vpic = kvm_create_pic(kvm);
3416 r = kvm_ioapic_init(kvm);
3418 mutex_lock(&kvm->slots_lock);
3419 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3421 mutex_unlock(&kvm->slots_lock);
3423 goto create_irqchip_unlock;
3426 goto create_irqchip_unlock;
3428 kvm->arch.vpic = vpic;
3430 r = kvm_setup_default_irq_routing(kvm);
3432 mutex_lock(&kvm->slots_lock);
3433 mutex_lock(&kvm->irq_lock);
3434 kvm_ioapic_destroy(kvm);
3435 kvm_destroy_pic(kvm);
3436 mutex_unlock(&kvm->irq_lock);
3437 mutex_unlock(&kvm->slots_lock);
3439 create_irqchip_unlock:
3440 mutex_unlock(&kvm->lock);
3443 case KVM_CREATE_PIT:
3444 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3446 case KVM_CREATE_PIT2:
3448 if (copy_from_user(&u.pit_config, argp,
3449 sizeof(struct kvm_pit_config)))
3452 mutex_lock(&kvm->slots_lock);
3455 goto create_pit_unlock;
3457 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3461 mutex_unlock(&kvm->slots_lock);
3463 case KVM_IRQ_LINE_STATUS:
3464 case KVM_IRQ_LINE: {
3465 struct kvm_irq_level irq_event;
3468 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3471 if (irqchip_in_kernel(kvm)) {
3473 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3474 irq_event.irq, irq_event.level);
3475 if (ioctl == KVM_IRQ_LINE_STATUS) {
3477 irq_event.status = status;
3478 if (copy_to_user(argp, &irq_event,
3486 case KVM_GET_IRQCHIP: {
3487 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3488 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3494 if (copy_from_user(chip, argp, sizeof *chip))
3495 goto get_irqchip_out;
3497 if (!irqchip_in_kernel(kvm))
3498 goto get_irqchip_out;
3499 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3501 goto get_irqchip_out;
3503 if (copy_to_user(argp, chip, sizeof *chip))
3504 goto get_irqchip_out;
3512 case KVM_SET_IRQCHIP: {
3513 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3514 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3520 if (copy_from_user(chip, argp, sizeof *chip))
3521 goto set_irqchip_out;
3523 if (!irqchip_in_kernel(kvm))
3524 goto set_irqchip_out;
3525 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3527 goto set_irqchip_out;
3537 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3540 if (!kvm->arch.vpit)
3542 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3546 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3553 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3556 if (!kvm->arch.vpit)
3558 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3564 case KVM_GET_PIT2: {
3566 if (!kvm->arch.vpit)
3568 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3572 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3577 case KVM_SET_PIT2: {
3579 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3582 if (!kvm->arch.vpit)
3584 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3590 case KVM_REINJECT_CONTROL: {
3591 struct kvm_reinject_control control;
3593 if (copy_from_user(&control, argp, sizeof(control)))
3595 r = kvm_vm_ioctl_reinject(kvm, &control);
3601 case KVM_XEN_HVM_CONFIG: {
3603 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3604 sizeof(struct kvm_xen_hvm_config)))
3607 if (kvm->arch.xen_hvm_config.flags)
3612 case KVM_SET_CLOCK: {
3613 struct kvm_clock_data user_ns;
3618 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3626 local_irq_disable();
3627 now_ns = get_kernel_ns();
3628 delta = user_ns.clock - now_ns;
3630 kvm->arch.kvmclock_offset = delta;
3633 case KVM_GET_CLOCK: {
3634 struct kvm_clock_data user_ns;
3637 local_irq_disable();
3638 now_ns = get_kernel_ns();
3639 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3642 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3645 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3658 static void kvm_init_msr_list(void)
3663 /* skip the first msrs in the list. KVM-specific */
3664 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3665 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3668 msrs_to_save[j] = msrs_to_save[i];
3671 num_msrs_to_save = j;
3674 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3682 if (!(vcpu->arch.apic &&
3683 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3684 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3695 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3702 if (!(vcpu->arch.apic &&
3703 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3704 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3706 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3716 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3717 struct kvm_segment *var, int seg)
3719 kvm_x86_ops->set_segment(vcpu, var, seg);
3722 void kvm_get_segment(struct kvm_vcpu *vcpu,
3723 struct kvm_segment *var, int seg)
3725 kvm_x86_ops->get_segment(vcpu, var, seg);
3728 static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3733 static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3736 struct x86_exception exception;
3738 BUG_ON(!mmu_is_nested(vcpu));
3740 /* NPT walks are always user-walks */
3741 access |= PFERR_USER_MASK;
3742 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3747 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3748 struct x86_exception *exception)
3750 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3751 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3754 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3755 struct x86_exception *exception)
3757 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3758 access |= PFERR_FETCH_MASK;
3759 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3762 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3763 struct x86_exception *exception)
3765 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3766 access |= PFERR_WRITE_MASK;
3767 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3770 /* uses this to access any guest's mapped memory without checking CPL */
3771 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3772 struct x86_exception *exception)
3774 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3777 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3778 struct kvm_vcpu *vcpu, u32 access,
3779 struct x86_exception *exception)
3782 int r = X86EMUL_CONTINUE;
3785 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3787 unsigned offset = addr & (PAGE_SIZE-1);
3788 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3791 if (gpa == UNMAPPED_GVA)
3792 return X86EMUL_PROPAGATE_FAULT;
3793 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3795 r = X86EMUL_IO_NEEDED;
3807 /* used for instruction fetching */
3808 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3809 gva_t addr, void *val, unsigned int bytes,
3810 struct x86_exception *exception)
3812 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3813 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3815 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3816 access | PFERR_FETCH_MASK,
3820 static int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3821 gva_t addr, void *val, unsigned int bytes,
3822 struct x86_exception *exception)
3824 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3825 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3827 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3831 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3832 gva_t addr, void *val, unsigned int bytes,
3833 struct x86_exception *exception)
3835 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3836 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3839 static int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3840 gva_t addr, void *val,
3842 struct x86_exception *exception)
3844 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3846 int r = X86EMUL_CONTINUE;
3849 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3852 unsigned offset = addr & (PAGE_SIZE-1);
3853 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3856 if (gpa == UNMAPPED_GVA)
3857 return X86EMUL_PROPAGATE_FAULT;
3858 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3860 r = X86EMUL_IO_NEEDED;
3872 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
3876 struct x86_exception *exception)
3878 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3882 if (vcpu->mmio_read_completed) {
3883 memcpy(val, vcpu->mmio_data, bytes);
3884 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3885 vcpu->mmio_phys_addr, *(u64 *)val);
3886 vcpu->mmio_read_completed = 0;
3887 return X86EMUL_CONTINUE;
3890 gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, exception);
3892 if (gpa == UNMAPPED_GVA)
3893 return X86EMUL_PROPAGATE_FAULT;
3895 /* For APIC access vmexit */
3896 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3899 if (kvm_read_guest_virt(ctxt, addr, val, bytes, exception)
3900 == X86EMUL_CONTINUE)
3901 return X86EMUL_CONTINUE;
3905 * Is this MMIO handled locally?
3907 handled = vcpu_mmio_read(vcpu, gpa, bytes, val);
3909 if (handled == bytes)
3910 return X86EMUL_CONTINUE;
3916 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3918 vcpu->mmio_needed = 1;
3919 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3920 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3921 vcpu->mmio_size = bytes;
3922 vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
3923 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 0;
3924 vcpu->mmio_index = 0;
3926 return X86EMUL_IO_NEEDED;
3929 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3930 const void *val, int bytes)
3934 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3937 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3941 static int emulator_write_emulated_onepage(unsigned long addr,
3944 struct x86_exception *exception,
3945 struct kvm_vcpu *vcpu)
3950 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, exception);
3952 if (gpa == UNMAPPED_GVA)
3953 return X86EMUL_PROPAGATE_FAULT;
3955 /* For APIC access vmexit */
3956 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3959 if (emulator_write_phys(vcpu, gpa, val, bytes))
3960 return X86EMUL_CONTINUE;
3963 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3965 * Is this MMIO handled locally?
3967 handled = vcpu_mmio_write(vcpu, gpa, bytes, val);
3968 if (handled == bytes)
3969 return X86EMUL_CONTINUE;
3975 vcpu->mmio_needed = 1;
3976 memcpy(vcpu->mmio_data, val, bytes);
3977 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3978 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3979 vcpu->mmio_size = bytes;
3980 vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
3981 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 1;
3982 memcpy(vcpu->run->mmio.data, vcpu->mmio_data, 8);
3983 vcpu->mmio_index = 0;
3985 return X86EMUL_CONTINUE;
3988 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
3992 struct x86_exception *exception)
3994 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3996 /* Crossing a page boundary? */
3997 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4000 now = -addr & ~PAGE_MASK;
4001 rc = emulator_write_emulated_onepage(addr, val, now, exception,
4003 if (rc != X86EMUL_CONTINUE)
4009 return emulator_write_emulated_onepage(addr, val, bytes, exception,
4013 #define CMPXCHG_TYPE(t, ptr, old, new) \
4014 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4016 #ifdef CONFIG_X86_64
4017 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4019 # define CMPXCHG64(ptr, old, new) \
4020 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4023 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4028 struct x86_exception *exception)
4030 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4036 /* guests cmpxchg8b have to be emulated atomically */
4037 if (bytes > 8 || (bytes & (bytes - 1)))
4040 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4042 if (gpa == UNMAPPED_GVA ||
4043 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4046 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4049 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4050 if (is_error_page(page)) {
4051 kvm_release_page_clean(page);
4055 kaddr = kmap_atomic(page, KM_USER0);
4056 kaddr += offset_in_page(gpa);
4059 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4062 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4065 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4068 exchanged = CMPXCHG64(kaddr, old, new);
4073 kunmap_atomic(kaddr, KM_USER0);
4074 kvm_release_page_dirty(page);
4077 return X86EMUL_CMPXCHG_FAILED;
4079 kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
4081 return X86EMUL_CONTINUE;
4084 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4086 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4089 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4091 /* TODO: String I/O for in kernel device */
4094 if (vcpu->arch.pio.in)
4095 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4096 vcpu->arch.pio.size, pd);
4098 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4099 vcpu->arch.pio.port, vcpu->arch.pio.size,
4105 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4106 int size, unsigned short port, void *val,
4109 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4111 if (vcpu->arch.pio.count)
4114 trace_kvm_pio(0, port, size, count);
4116 vcpu->arch.pio.port = port;
4117 vcpu->arch.pio.in = 1;
4118 vcpu->arch.pio.count = count;
4119 vcpu->arch.pio.size = size;
4121 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4123 memcpy(val, vcpu->arch.pio_data, size * count);
4124 vcpu->arch.pio.count = 0;
4128 vcpu->run->exit_reason = KVM_EXIT_IO;
4129 vcpu->run->io.direction = KVM_EXIT_IO_IN;
4130 vcpu->run->io.size = size;
4131 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4132 vcpu->run->io.count = count;
4133 vcpu->run->io.port = port;
4138 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4139 int size, unsigned short port,
4140 const void *val, unsigned int count)
4142 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4144 trace_kvm_pio(1, port, size, count);
4146 vcpu->arch.pio.port = port;
4147 vcpu->arch.pio.in = 0;
4148 vcpu->arch.pio.count = count;
4149 vcpu->arch.pio.size = size;
4151 memcpy(vcpu->arch.pio_data, val, size * count);
4153 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4154 vcpu->arch.pio.count = 0;
4158 vcpu->run->exit_reason = KVM_EXIT_IO;
4159 vcpu->run->io.direction = KVM_EXIT_IO_OUT;
4160 vcpu->run->io.size = size;
4161 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4162 vcpu->run->io.count = count;
4163 vcpu->run->io.port = port;
4168 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4170 return kvm_x86_ops->get_segment_base(vcpu, seg);
4173 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4175 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4178 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4180 if (!need_emulate_wbinvd(vcpu))
4181 return X86EMUL_CONTINUE;
4183 if (kvm_x86_ops->has_wbinvd_exit()) {
4184 int cpu = get_cpu();
4186 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4187 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4188 wbinvd_ipi, NULL, 1);
4190 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4193 return X86EMUL_CONTINUE;
4195 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4197 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4199 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4202 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4204 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4207 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4210 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4213 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4215 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4218 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4220 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4221 unsigned long value;
4225 value = kvm_read_cr0(vcpu);
4228 value = vcpu->arch.cr2;
4231 value = kvm_read_cr3(vcpu);
4234 value = kvm_read_cr4(vcpu);
4237 value = kvm_get_cr8(vcpu);
4240 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4247 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4249 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4254 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4257 vcpu->arch.cr2 = val;
4260 res = kvm_set_cr3(vcpu, val);
4263 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4266 res = kvm_set_cr8(vcpu, val);
4269 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4276 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4278 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4281 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4283 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4286 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4288 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4291 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4293 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4296 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4298 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4301 static unsigned long emulator_get_cached_segment_base(
4302 struct x86_emulate_ctxt *ctxt, int seg)
4304 return get_segment_base(emul_to_vcpu(ctxt), seg);
4307 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4308 struct desc_struct *desc, u32 *base3,
4311 struct kvm_segment var;
4313 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4314 *selector = var.selector;
4321 set_desc_limit(desc, var.limit);
4322 set_desc_base(desc, (unsigned long)var.base);
4323 #ifdef CONFIG_X86_64
4325 *base3 = var.base >> 32;
4327 desc->type = var.type;
4329 desc->dpl = var.dpl;
4330 desc->p = var.present;
4331 desc->avl = var.avl;
4339 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4340 struct desc_struct *desc, u32 base3,
4343 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4344 struct kvm_segment var;
4346 var.selector = selector;
4347 var.base = get_desc_base(desc);
4348 #ifdef CONFIG_X86_64
4349 var.base |= ((u64)base3) << 32;
4351 var.limit = get_desc_limit(desc);
4353 var.limit = (var.limit << 12) | 0xfff;
4354 var.type = desc->type;
4355 var.present = desc->p;
4356 var.dpl = desc->dpl;
4361 var.avl = desc->avl;
4362 var.present = desc->p;
4363 var.unusable = !var.present;
4366 kvm_set_segment(vcpu, &var, seg);
4370 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4371 u32 msr_index, u64 *pdata)
4373 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4376 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4377 u32 msr_index, u64 data)
4379 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
4382 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4384 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4387 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4390 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4392 * CR0.TS may reference the host fpu state, not the guest fpu state,
4393 * so it may be clear at this point.
4398 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4403 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4404 struct x86_instruction_info *info,
4405 enum x86_intercept_stage stage)
4407 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4410 static struct x86_emulate_ops emulate_ops = {
4411 .read_std = kvm_read_guest_virt_system,
4412 .write_std = kvm_write_guest_virt_system,
4413 .fetch = kvm_fetch_guest_virt,
4414 .read_emulated = emulator_read_emulated,
4415 .write_emulated = emulator_write_emulated,
4416 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4417 .invlpg = emulator_invlpg,
4418 .pio_in_emulated = emulator_pio_in_emulated,
4419 .pio_out_emulated = emulator_pio_out_emulated,
4420 .get_segment = emulator_get_segment,
4421 .set_segment = emulator_set_segment,
4422 .get_cached_segment_base = emulator_get_cached_segment_base,
4423 .get_gdt = emulator_get_gdt,
4424 .get_idt = emulator_get_idt,
4425 .set_gdt = emulator_set_gdt,
4426 .set_idt = emulator_set_idt,
4427 .get_cr = emulator_get_cr,
4428 .set_cr = emulator_set_cr,
4429 .cpl = emulator_get_cpl,
4430 .get_dr = emulator_get_dr,
4431 .set_dr = emulator_set_dr,
4432 .set_msr = emulator_set_msr,
4433 .get_msr = emulator_get_msr,
4434 .halt = emulator_halt,
4435 .wbinvd = emulator_wbinvd,
4436 .fix_hypercall = emulator_fix_hypercall,
4437 .get_fpu = emulator_get_fpu,
4438 .put_fpu = emulator_put_fpu,
4439 .intercept = emulator_intercept,
4442 static void cache_all_regs(struct kvm_vcpu *vcpu)
4444 kvm_register_read(vcpu, VCPU_REGS_RAX);
4445 kvm_register_read(vcpu, VCPU_REGS_RSP);
4446 kvm_register_read(vcpu, VCPU_REGS_RIP);
4447 vcpu->arch.regs_dirty = ~0;
4450 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4452 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4454 * an sti; sti; sequence only disable interrupts for the first
4455 * instruction. So, if the last instruction, be it emulated or
4456 * not, left the system with the INT_STI flag enabled, it
4457 * means that the last instruction is an sti. We should not
4458 * leave the flag on in this case. The same goes for mov ss
4460 if (!(int_shadow & mask))
4461 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4464 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4466 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4467 if (ctxt->exception.vector == PF_VECTOR)
4468 kvm_propagate_fault(vcpu, &ctxt->exception);
4469 else if (ctxt->exception.error_code_valid)
4470 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4471 ctxt->exception.error_code);
4473 kvm_queue_exception(vcpu, ctxt->exception.vector);
4476 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4478 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4482 * TODO: fix emulate.c to use guest_read/write_register
4483 * instead of direct ->regs accesses, can save hundred cycles
4484 * on Intel for instructions that don't read/change RSP, for
4487 cache_all_regs(vcpu);
4489 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4491 vcpu->arch.emulate_ctxt.eflags = kvm_get_rflags(vcpu);
4492 vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
4493 vcpu->arch.emulate_ctxt.mode =
4494 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4495 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
4496 ? X86EMUL_MODE_VM86 : cs_l
4497 ? X86EMUL_MODE_PROT64 : cs_db
4498 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
4499 vcpu->arch.emulate_ctxt.guest_mode = is_guest_mode(vcpu);
4500 memset(c, 0, sizeof(struct decode_cache));
4501 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4502 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4505 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4507 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4510 init_emulate_ctxt(vcpu);
4512 vcpu->arch.emulate_ctxt.decode.op_bytes = 2;
4513 vcpu->arch.emulate_ctxt.decode.ad_bytes = 2;
4514 vcpu->arch.emulate_ctxt.decode.eip = vcpu->arch.emulate_ctxt.eip +
4516 ret = emulate_int_real(&vcpu->arch.emulate_ctxt, &emulate_ops, irq);
4518 if (ret != X86EMUL_CONTINUE)
4519 return EMULATE_FAIL;
4521 vcpu->arch.emulate_ctxt.eip = c->eip;
4522 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4523 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4524 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4526 if (irq == NMI_VECTOR)
4527 vcpu->arch.nmi_pending = false;
4529 vcpu->arch.interrupt.pending = false;
4531 return EMULATE_DONE;
4533 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4535 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4537 int r = EMULATE_DONE;
4539 ++vcpu->stat.insn_emulation_fail;
4540 trace_kvm_emulate_insn_failed(vcpu);
4541 if (!is_guest_mode(vcpu)) {
4542 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4543 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4544 vcpu->run->internal.ndata = 0;
4547 kvm_queue_exception(vcpu, UD_VECTOR);
4552 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4560 * if emulation was due to access to shadowed page table
4561 * and it failed try to unshadow page and re-entetr the
4562 * guest to let CPU execute the instruction.
4564 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4567 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4569 if (gpa == UNMAPPED_GVA)
4570 return true; /* let cpu generate fault */
4572 if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4578 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4585 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4586 bool writeback = true;
4588 kvm_clear_exception_queue(vcpu);
4590 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4591 init_emulate_ctxt(vcpu);
4592 vcpu->arch.emulate_ctxt.interruptibility = 0;
4593 vcpu->arch.emulate_ctxt.have_exception = false;
4594 vcpu->arch.emulate_ctxt.perm_ok = false;
4596 vcpu->arch.emulate_ctxt.only_vendor_specific_insn
4597 = emulation_type & EMULTYPE_TRAP_UD;
4599 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, insn, insn_len);
4601 trace_kvm_emulate_insn_start(vcpu);
4602 ++vcpu->stat.insn_emulation;
4604 if (emulation_type & EMULTYPE_TRAP_UD)
4605 return EMULATE_FAIL;
4606 if (reexecute_instruction(vcpu, cr2))
4607 return EMULATE_DONE;
4608 if (emulation_type & EMULTYPE_SKIP)
4609 return EMULATE_FAIL;
4610 return handle_emulation_failure(vcpu);
4614 if (emulation_type & EMULTYPE_SKIP) {
4615 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
4616 return EMULATE_DONE;
4619 /* this is needed for vmware backdoor interface to work since it
4620 changes registers values during IO operation */
4621 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4622 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4623 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4627 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt);
4629 if (r == EMULATION_INTERCEPTED)
4630 return EMULATE_DONE;
4632 if (r == EMULATION_FAILED) {
4633 if (reexecute_instruction(vcpu, cr2))
4634 return EMULATE_DONE;
4636 return handle_emulation_failure(vcpu);
4639 if (vcpu->arch.emulate_ctxt.have_exception) {
4640 inject_emulated_exception(vcpu);
4642 } else if (vcpu->arch.pio.count) {
4643 if (!vcpu->arch.pio.in)
4644 vcpu->arch.pio.count = 0;
4647 r = EMULATE_DO_MMIO;
4648 } else if (vcpu->mmio_needed) {
4649 if (!vcpu->mmio_is_write)
4651 r = EMULATE_DO_MMIO;
4652 } else if (r == EMULATION_RESTART)
4658 toggle_interruptibility(vcpu,
4659 vcpu->arch.emulate_ctxt.interruptibility);
4660 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4661 kvm_make_request(KVM_REQ_EVENT, vcpu);
4662 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4663 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4664 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4666 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
4670 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4672 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4674 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4675 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4676 size, port, &val, 1);
4677 /* do not return to emulator after return from userspace */
4678 vcpu->arch.pio.count = 0;
4681 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4683 static void tsc_bad(void *info)
4685 __this_cpu_write(cpu_tsc_khz, 0);
4688 static void tsc_khz_changed(void *data)
4690 struct cpufreq_freqs *freq = data;
4691 unsigned long khz = 0;
4695 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4696 khz = cpufreq_quick_get(raw_smp_processor_id());
4699 __this_cpu_write(cpu_tsc_khz, khz);
4702 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4705 struct cpufreq_freqs *freq = data;
4707 struct kvm_vcpu *vcpu;
4708 int i, send_ipi = 0;
4711 * We allow guests to temporarily run on slowing clocks,
4712 * provided we notify them after, or to run on accelerating
4713 * clocks, provided we notify them before. Thus time never
4716 * However, we have a problem. We can't atomically update
4717 * the frequency of a given CPU from this function; it is
4718 * merely a notifier, which can be called from any CPU.
4719 * Changing the TSC frequency at arbitrary points in time
4720 * requires a recomputation of local variables related to
4721 * the TSC for each VCPU. We must flag these local variables
4722 * to be updated and be sure the update takes place with the
4723 * new frequency before any guests proceed.
4725 * Unfortunately, the combination of hotplug CPU and frequency
4726 * change creates an intractable locking scenario; the order
4727 * of when these callouts happen is undefined with respect to
4728 * CPU hotplug, and they can race with each other. As such,
4729 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4730 * undefined; you can actually have a CPU frequency change take
4731 * place in between the computation of X and the setting of the
4732 * variable. To protect against this problem, all updates of
4733 * the per_cpu tsc_khz variable are done in an interrupt
4734 * protected IPI, and all callers wishing to update the value
4735 * must wait for a synchronous IPI to complete (which is trivial
4736 * if the caller is on the CPU already). This establishes the
4737 * necessary total order on variable updates.
4739 * Note that because a guest time update may take place
4740 * anytime after the setting of the VCPU's request bit, the
4741 * correct TSC value must be set before the request. However,
4742 * to ensure the update actually makes it to any guest which
4743 * starts running in hardware virtualization between the set
4744 * and the acquisition of the spinlock, we must also ping the
4745 * CPU after setting the request bit.
4749 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4751 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4754 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4756 raw_spin_lock(&kvm_lock);
4757 list_for_each_entry(kvm, &vm_list, vm_list) {
4758 kvm_for_each_vcpu(i, vcpu, kvm) {
4759 if (vcpu->cpu != freq->cpu)
4761 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4762 if (vcpu->cpu != smp_processor_id())
4766 raw_spin_unlock(&kvm_lock);
4768 if (freq->old < freq->new && send_ipi) {
4770 * We upscale the frequency. Must make the guest
4771 * doesn't see old kvmclock values while running with
4772 * the new frequency, otherwise we risk the guest sees
4773 * time go backwards.
4775 * In case we update the frequency for another cpu
4776 * (which might be in guest context) send an interrupt
4777 * to kick the cpu out of guest context. Next time
4778 * guest context is entered kvmclock will be updated,
4779 * so the guest will not see stale values.
4781 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4786 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4787 .notifier_call = kvmclock_cpufreq_notifier
4790 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4791 unsigned long action, void *hcpu)
4793 unsigned int cpu = (unsigned long)hcpu;
4797 case CPU_DOWN_FAILED:
4798 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4800 case CPU_DOWN_PREPARE:
4801 smp_call_function_single(cpu, tsc_bad, NULL, 1);
4807 static struct notifier_block kvmclock_cpu_notifier_block = {
4808 .notifier_call = kvmclock_cpu_notifier,
4809 .priority = -INT_MAX
4812 static void kvm_timer_init(void)
4816 max_tsc_khz = tsc_khz;
4817 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4818 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4819 #ifdef CONFIG_CPU_FREQ
4820 struct cpufreq_policy policy;
4821 memset(&policy, 0, sizeof(policy));
4823 cpufreq_get_policy(&policy, cpu);
4824 if (policy.cpuinfo.max_freq)
4825 max_tsc_khz = policy.cpuinfo.max_freq;
4828 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4829 CPUFREQ_TRANSITION_NOTIFIER);
4831 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
4832 for_each_online_cpu(cpu)
4833 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4836 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4838 static int kvm_is_in_guest(void)
4840 return percpu_read(current_vcpu) != NULL;
4843 static int kvm_is_user_mode(void)
4847 if (percpu_read(current_vcpu))
4848 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
4850 return user_mode != 0;
4853 static unsigned long kvm_get_guest_ip(void)
4855 unsigned long ip = 0;
4857 if (percpu_read(current_vcpu))
4858 ip = kvm_rip_read(percpu_read(current_vcpu));
4863 static struct perf_guest_info_callbacks kvm_guest_cbs = {
4864 .is_in_guest = kvm_is_in_guest,
4865 .is_user_mode = kvm_is_user_mode,
4866 .get_guest_ip = kvm_get_guest_ip,
4869 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4871 percpu_write(current_vcpu, vcpu);
4873 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4875 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4877 percpu_write(current_vcpu, NULL);
4879 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4881 int kvm_arch_init(void *opaque)
4884 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4887 printk(KERN_ERR "kvm: already loaded the other module\n");
4892 if (!ops->cpu_has_kvm_support()) {
4893 printk(KERN_ERR "kvm: no hardware support\n");
4897 if (ops->disabled_by_bios()) {
4898 printk(KERN_ERR "kvm: disabled by bios\n");
4903 r = kvm_mmu_module_init();
4907 kvm_init_msr_list();
4910 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
4911 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4912 PT_DIRTY_MASK, PT64_NX_MASK, 0);
4916 perf_register_guest_info_callbacks(&kvm_guest_cbs);
4919 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
4927 void kvm_arch_exit(void)
4929 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4931 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4932 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4933 CPUFREQ_TRANSITION_NOTIFIER);
4934 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4936 kvm_mmu_module_exit();
4939 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4941 ++vcpu->stat.halt_exits;
4942 if (irqchip_in_kernel(vcpu->kvm)) {
4943 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
4946 vcpu->run->exit_reason = KVM_EXIT_HLT;
4950 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4952 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
4955 if (is_long_mode(vcpu))
4958 return a0 | ((gpa_t)a1 << 32);
4961 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4963 u64 param, ingpa, outgpa, ret;
4964 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4965 bool fast, longmode;
4969 * hypercall generates UD from non zero cpl and real mode
4972 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
4973 kvm_queue_exception(vcpu, UD_VECTOR);
4977 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4978 longmode = is_long_mode(vcpu) && cs_l == 1;
4981 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4982 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4983 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4984 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4985 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4986 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
4988 #ifdef CONFIG_X86_64
4990 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4991 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4992 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4996 code = param & 0xffff;
4997 fast = (param >> 16) & 0x1;
4998 rep_cnt = (param >> 32) & 0xfff;
4999 rep_idx = (param >> 48) & 0xfff;
5001 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5004 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5005 kvm_vcpu_on_spin(vcpu);
5008 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5012 ret = res | (((u64)rep_done & 0xfff) << 32);
5014 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5016 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5017 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5023 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5025 unsigned long nr, a0, a1, a2, a3, ret;
5028 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5029 return kvm_hv_hypercall(vcpu);
5031 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5032 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5033 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5034 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5035 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5037 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5039 if (!is_long_mode(vcpu)) {
5047 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5053 case KVM_HC_VAPIC_POLL_IRQ:
5057 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
5064 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5065 ++vcpu->stat.hypercalls;
5068 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5070 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5072 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5073 char instruction[3];
5074 unsigned long rip = kvm_rip_read(vcpu);
5077 * Blow out the MMU to ensure that no other VCPU has an active mapping
5078 * to ensure that the updated hypercall appears atomically across all
5081 kvm_mmu_zap_all(vcpu->kvm);
5083 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5085 return emulator_write_emulated(&vcpu->arch.emulate_ctxt,
5086 rip, instruction, 3, NULL);
5089 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
5091 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
5092 int j, nent = vcpu->arch.cpuid_nent;
5094 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
5095 /* when no next entry is found, the current entry[i] is reselected */
5096 for (j = i + 1; ; j = (j + 1) % nent) {
5097 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
5098 if (ej->function == e->function) {
5099 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
5103 return 0; /* silence gcc, even though control never reaches here */
5106 /* find an entry with matching function, matching index (if needed), and that
5107 * should be read next (if it's stateful) */
5108 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
5109 u32 function, u32 index)
5111 if (e->function != function)
5113 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
5115 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
5116 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
5121 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
5122 u32 function, u32 index)
5125 struct kvm_cpuid_entry2 *best = NULL;
5127 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
5128 struct kvm_cpuid_entry2 *e;
5130 e = &vcpu->arch.cpuid_entries[i];
5131 if (is_matching_cpuid_entry(e, function, index)) {
5132 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
5133 move_to_next_stateful_cpuid_entry(vcpu, i);
5140 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
5142 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
5144 struct kvm_cpuid_entry2 *best;
5146 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
5147 if (!best || best->eax < 0x80000008)
5149 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
5151 return best->eax & 0xff;
5157 * If no match is found, check whether we exceed the vCPU's limit
5158 * and return the content of the highest valid _standard_ leaf instead.
5159 * This is to satisfy the CPUID specification.
5161 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
5162 u32 function, u32 index)
5164 struct kvm_cpuid_entry2 *maxlevel;
5166 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
5167 if (!maxlevel || maxlevel->eax >= function)
5169 if (function & 0x80000000) {
5170 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
5174 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
5177 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
5179 u32 function, index;
5180 struct kvm_cpuid_entry2 *best;
5182 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
5183 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5184 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
5185 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
5186 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
5187 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
5188 best = kvm_find_cpuid_entry(vcpu, function, index);
5191 best = check_cpuid_limit(vcpu, function, index);
5194 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
5195 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
5196 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
5197 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
5199 kvm_x86_ops->skip_emulated_instruction(vcpu);
5200 trace_kvm_cpuid(function,
5201 kvm_register_read(vcpu, VCPU_REGS_RAX),
5202 kvm_register_read(vcpu, VCPU_REGS_RBX),
5203 kvm_register_read(vcpu, VCPU_REGS_RCX),
5204 kvm_register_read(vcpu, VCPU_REGS_RDX));
5206 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
5209 * Check if userspace requested an interrupt window, and that the
5210 * interrupt window is open.
5212 * No need to exit to userspace if we already have an interrupt queued.
5214 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5216 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5217 vcpu->run->request_interrupt_window &&
5218 kvm_arch_interrupt_allowed(vcpu));
5221 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5223 struct kvm_run *kvm_run = vcpu->run;
5225 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5226 kvm_run->cr8 = kvm_get_cr8(vcpu);
5227 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5228 if (irqchip_in_kernel(vcpu->kvm))
5229 kvm_run->ready_for_interrupt_injection = 1;
5231 kvm_run->ready_for_interrupt_injection =
5232 kvm_arch_interrupt_allowed(vcpu) &&
5233 !kvm_cpu_has_interrupt(vcpu) &&
5234 !kvm_event_needs_reinjection(vcpu);
5237 static void vapic_enter(struct kvm_vcpu *vcpu)
5239 struct kvm_lapic *apic = vcpu->arch.apic;
5242 if (!apic || !apic->vapic_addr)
5245 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5247 vcpu->arch.apic->vapic_page = page;
5250 static void vapic_exit(struct kvm_vcpu *vcpu)
5252 struct kvm_lapic *apic = vcpu->arch.apic;
5255 if (!apic || !apic->vapic_addr)
5258 idx = srcu_read_lock(&vcpu->kvm->srcu);
5259 kvm_release_page_dirty(apic->vapic_page);
5260 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5261 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5264 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5268 if (!kvm_x86_ops->update_cr8_intercept)
5271 if (!vcpu->arch.apic)
5274 if (!vcpu->arch.apic->vapic_addr)
5275 max_irr = kvm_lapic_find_highest_irr(vcpu);
5282 tpr = kvm_lapic_get_cr8(vcpu);
5284 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5287 static void inject_pending_event(struct kvm_vcpu *vcpu)
5289 /* try to reinject previous events if any */
5290 if (vcpu->arch.exception.pending) {
5291 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5292 vcpu->arch.exception.has_error_code,
5293 vcpu->arch.exception.error_code);
5294 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5295 vcpu->arch.exception.has_error_code,
5296 vcpu->arch.exception.error_code,
5297 vcpu->arch.exception.reinject);
5301 if (vcpu->arch.nmi_injected) {
5302 kvm_x86_ops->set_nmi(vcpu);
5306 if (vcpu->arch.interrupt.pending) {
5307 kvm_x86_ops->set_irq(vcpu);
5311 /* try to inject new event if pending */
5312 if (vcpu->arch.nmi_pending) {
5313 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5314 vcpu->arch.nmi_pending = false;
5315 vcpu->arch.nmi_injected = true;
5316 kvm_x86_ops->set_nmi(vcpu);
5318 } else if (kvm_cpu_has_interrupt(vcpu)) {
5319 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5320 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5322 kvm_x86_ops->set_irq(vcpu);
5327 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5329 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5330 !vcpu->guest_xcr0_loaded) {
5331 /* kvm_set_xcr() also depends on this */
5332 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5333 vcpu->guest_xcr0_loaded = 1;
5337 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5339 if (vcpu->guest_xcr0_loaded) {
5340 if (vcpu->arch.xcr0 != host_xcr0)
5341 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5342 vcpu->guest_xcr0_loaded = 0;
5346 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5350 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5351 vcpu->run->request_interrupt_window;
5353 if (vcpu->requests) {
5354 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5355 kvm_mmu_unload(vcpu);
5356 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5357 __kvm_migrate_timers(vcpu);
5358 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5359 r = kvm_guest_time_update(vcpu);
5363 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5364 kvm_mmu_sync_roots(vcpu);
5365 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5366 kvm_x86_ops->tlb_flush(vcpu);
5367 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5368 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5372 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5373 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5377 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5378 vcpu->fpu_active = 0;
5379 kvm_x86_ops->fpu_deactivate(vcpu);
5381 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5382 /* Page is swapped out. Do synthetic halt */
5383 vcpu->arch.apf.halted = true;
5389 r = kvm_mmu_reload(vcpu);
5394 * An NMI can be injected between local nmi_pending read and
5395 * vcpu->arch.nmi_pending read inside inject_pending_event().
5396 * But in that case, KVM_REQ_EVENT will be set, which makes
5397 * the race described above benign.
5399 nmi_pending = ACCESS_ONCE(vcpu->arch.nmi_pending);
5401 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5402 inject_pending_event(vcpu);
5404 /* enable NMI/IRQ window open exits if needed */
5406 kvm_x86_ops->enable_nmi_window(vcpu);
5407 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5408 kvm_x86_ops->enable_irq_window(vcpu);
5410 if (kvm_lapic_enabled(vcpu)) {
5411 update_cr8_intercept(vcpu);
5412 kvm_lapic_sync_to_vapic(vcpu);
5418 kvm_x86_ops->prepare_guest_switch(vcpu);
5419 if (vcpu->fpu_active)
5420 kvm_load_guest_fpu(vcpu);
5421 kvm_load_guest_xcr0(vcpu);
5423 vcpu->mode = IN_GUEST_MODE;
5425 /* We should set ->mode before check ->requests,
5426 * see the comment in make_all_cpus_request.
5430 local_irq_disable();
5432 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5433 || need_resched() || signal_pending(current)) {
5434 vcpu->mode = OUTSIDE_GUEST_MODE;
5438 kvm_x86_ops->cancel_injection(vcpu);
5443 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5447 if (unlikely(vcpu->arch.switch_db_regs)) {
5449 set_debugreg(vcpu->arch.eff_db[0], 0);
5450 set_debugreg(vcpu->arch.eff_db[1], 1);
5451 set_debugreg(vcpu->arch.eff_db[2], 2);
5452 set_debugreg(vcpu->arch.eff_db[3], 3);
5455 trace_kvm_entry(vcpu->vcpu_id);
5456 kvm_x86_ops->run(vcpu);
5459 * If the guest has used debug registers, at least dr7
5460 * will be disabled while returning to the host.
5461 * If we don't have active breakpoints in the host, we don't
5462 * care about the messed up debug address registers. But if
5463 * we have some of them active, restore the old state.
5465 if (hw_breakpoint_active())
5466 hw_breakpoint_restore();
5468 kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
5470 vcpu->mode = OUTSIDE_GUEST_MODE;
5477 * We must have an instruction between local_irq_enable() and
5478 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5479 * the interrupt shadow. The stat.exits increment will do nicely.
5480 * But we need to prevent reordering, hence this barrier():
5488 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5491 * Profile KVM exit RIPs:
5493 if (unlikely(prof_on == KVM_PROFILING)) {
5494 unsigned long rip = kvm_rip_read(vcpu);
5495 profile_hit(KVM_PROFILING, (void *)rip);
5499 kvm_lapic_sync_from_vapic(vcpu);
5501 r = kvm_x86_ops->handle_exit(vcpu);
5507 static int __vcpu_run(struct kvm_vcpu *vcpu)
5510 struct kvm *kvm = vcpu->kvm;
5512 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5513 pr_debug("vcpu %d received sipi with vector # %x\n",
5514 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5515 kvm_lapic_reset(vcpu);
5516 r = kvm_arch_vcpu_reset(vcpu);
5519 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5522 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5527 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5528 !vcpu->arch.apf.halted)
5529 r = vcpu_enter_guest(vcpu);
5531 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5532 kvm_vcpu_block(vcpu);
5533 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5534 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5536 switch(vcpu->arch.mp_state) {
5537 case KVM_MP_STATE_HALTED:
5538 vcpu->arch.mp_state =
5539 KVM_MP_STATE_RUNNABLE;
5540 case KVM_MP_STATE_RUNNABLE:
5541 vcpu->arch.apf.halted = false;
5543 case KVM_MP_STATE_SIPI_RECEIVED:
5554 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5555 if (kvm_cpu_has_pending_timer(vcpu))
5556 kvm_inject_pending_timer_irqs(vcpu);
5558 if (dm_request_for_irq_injection(vcpu)) {
5560 vcpu->run->exit_reason = KVM_EXIT_INTR;
5561 ++vcpu->stat.request_irq_exits;
5564 kvm_check_async_pf_completion(vcpu);
5566 if (signal_pending(current)) {
5568 vcpu->run->exit_reason = KVM_EXIT_INTR;
5569 ++vcpu->stat.signal_exits;
5571 if (need_resched()) {
5572 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5574 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5578 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5585 static int complete_mmio(struct kvm_vcpu *vcpu)
5587 struct kvm_run *run = vcpu->run;
5590 if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5593 if (vcpu->mmio_needed) {
5594 vcpu->mmio_needed = 0;
5595 if (!vcpu->mmio_is_write)
5596 memcpy(vcpu->mmio_data + vcpu->mmio_index,
5598 vcpu->mmio_index += 8;
5599 if (vcpu->mmio_index < vcpu->mmio_size) {
5600 run->exit_reason = KVM_EXIT_MMIO;
5601 run->mmio.phys_addr = vcpu->mmio_phys_addr + vcpu->mmio_index;
5602 memcpy(run->mmio.data, vcpu->mmio_data + vcpu->mmio_index, 8);
5603 run->mmio.len = min(vcpu->mmio_size - vcpu->mmio_index, 8);
5604 run->mmio.is_write = vcpu->mmio_is_write;
5605 vcpu->mmio_needed = 1;
5608 if (vcpu->mmio_is_write)
5610 vcpu->mmio_read_completed = 1;
5612 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5613 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5614 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5615 if (r != EMULATE_DONE)
5620 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5625 if (!tsk_used_math(current) && init_fpu(current))
5628 if (vcpu->sigset_active)
5629 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5631 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5632 kvm_vcpu_block(vcpu);
5633 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5638 /* re-sync apic's tpr */
5639 if (!irqchip_in_kernel(vcpu->kvm)) {
5640 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5646 r = complete_mmio(vcpu);
5650 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5651 kvm_register_write(vcpu, VCPU_REGS_RAX,
5652 kvm_run->hypercall.ret);
5654 r = __vcpu_run(vcpu);
5657 post_kvm_run_save(vcpu);
5658 if (vcpu->sigset_active)
5659 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5664 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5666 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5668 * We are here if userspace calls get_regs() in the middle of
5669 * instruction emulation. Registers state needs to be copied
5670 * back from emulation context to vcpu. Usrapace shouldn't do
5671 * that usually, but some bad designed PV devices (vmware
5672 * backdoor interface) need this to work
5674 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5675 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5676 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5678 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5679 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5680 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5681 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5682 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5683 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5684 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5685 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5686 #ifdef CONFIG_X86_64
5687 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5688 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5689 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5690 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5691 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5692 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5693 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5694 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5697 regs->rip = kvm_rip_read(vcpu);
5698 regs->rflags = kvm_get_rflags(vcpu);
5703 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5705 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
5706 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5708 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5709 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5710 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5711 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5712 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5713 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5714 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5715 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5716 #ifdef CONFIG_X86_64
5717 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5718 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5719 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5720 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5721 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5722 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5723 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5724 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5727 kvm_rip_write(vcpu, regs->rip);
5728 kvm_set_rflags(vcpu, regs->rflags);
5730 vcpu->arch.exception.pending = false;
5732 kvm_make_request(KVM_REQ_EVENT, vcpu);
5737 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5739 struct kvm_segment cs;
5741 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
5745 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5747 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5748 struct kvm_sregs *sregs)
5752 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5753 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5754 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5755 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5756 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5757 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5759 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5760 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5762 kvm_x86_ops->get_idt(vcpu, &dt);
5763 sregs->idt.limit = dt.size;
5764 sregs->idt.base = dt.address;
5765 kvm_x86_ops->get_gdt(vcpu, &dt);
5766 sregs->gdt.limit = dt.size;
5767 sregs->gdt.base = dt.address;
5769 sregs->cr0 = kvm_read_cr0(vcpu);
5770 sregs->cr2 = vcpu->arch.cr2;
5771 sregs->cr3 = kvm_read_cr3(vcpu);
5772 sregs->cr4 = kvm_read_cr4(vcpu);
5773 sregs->cr8 = kvm_get_cr8(vcpu);
5774 sregs->efer = vcpu->arch.efer;
5775 sregs->apic_base = kvm_get_apic_base(vcpu);
5777 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
5779 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
5780 set_bit(vcpu->arch.interrupt.nr,
5781 (unsigned long *)sregs->interrupt_bitmap);
5786 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
5787 struct kvm_mp_state *mp_state)
5789 mp_state->mp_state = vcpu->arch.mp_state;
5793 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
5794 struct kvm_mp_state *mp_state)
5796 vcpu->arch.mp_state = mp_state->mp_state;
5797 kvm_make_request(KVM_REQ_EVENT, vcpu);
5801 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
5802 bool has_error_code, u32 error_code)
5804 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5807 init_emulate_ctxt(vcpu);
5809 ret = emulator_task_switch(&vcpu->arch.emulate_ctxt,
5810 tss_selector, reason, has_error_code,
5814 return EMULATE_FAIL;
5816 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5817 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
5818 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
5819 kvm_make_request(KVM_REQ_EVENT, vcpu);
5820 return EMULATE_DONE;
5822 EXPORT_SYMBOL_GPL(kvm_task_switch);
5824 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5825 struct kvm_sregs *sregs)
5827 int mmu_reset_needed = 0;
5828 int pending_vec, max_bits, idx;
5831 dt.size = sregs->idt.limit;
5832 dt.address = sregs->idt.base;
5833 kvm_x86_ops->set_idt(vcpu, &dt);
5834 dt.size = sregs->gdt.limit;
5835 dt.address = sregs->gdt.base;
5836 kvm_x86_ops->set_gdt(vcpu, &dt);
5838 vcpu->arch.cr2 = sregs->cr2;
5839 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
5840 vcpu->arch.cr3 = sregs->cr3;
5841 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
5843 kvm_set_cr8(vcpu, sregs->cr8);
5845 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5846 kvm_x86_ops->set_efer(vcpu, sregs->efer);
5847 kvm_set_apic_base(vcpu, sregs->apic_base);
5849 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5850 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5851 vcpu->arch.cr0 = sregs->cr0;
5853 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5854 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5855 if (sregs->cr4 & X86_CR4_OSXSAVE)
5858 idx = srcu_read_lock(&vcpu->kvm->srcu);
5859 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5860 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
5861 mmu_reset_needed = 1;
5863 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5865 if (mmu_reset_needed)
5866 kvm_mmu_reset_context(vcpu);
5868 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5869 pending_vec = find_first_bit(
5870 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5871 if (pending_vec < max_bits) {
5872 kvm_queue_interrupt(vcpu, pending_vec, false);
5873 pr_debug("Set back pending irq %d\n", pending_vec);
5876 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5877 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5878 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5879 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5880 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5881 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5883 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5884 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5886 update_cr8_intercept(vcpu);
5888 /* Older userspace won't unhalt the vcpu on reset. */
5889 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5890 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5892 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5894 kvm_make_request(KVM_REQ_EVENT, vcpu);
5899 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5900 struct kvm_guest_debug *dbg)
5902 unsigned long rflags;
5905 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5907 if (vcpu->arch.exception.pending)
5909 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5910 kvm_queue_exception(vcpu, DB_VECTOR);
5912 kvm_queue_exception(vcpu, BP_VECTOR);
5916 * Read rflags as long as potentially injected trace flags are still
5919 rflags = kvm_get_rflags(vcpu);
5921 vcpu->guest_debug = dbg->control;
5922 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5923 vcpu->guest_debug = 0;
5925 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5926 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5927 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5928 vcpu->arch.switch_db_regs =
5929 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5931 for (i = 0; i < KVM_NR_DB_REGS; i++)
5932 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5933 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5936 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5937 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5938 get_segment_base(vcpu, VCPU_SREG_CS);
5941 * Trigger an rflags update that will inject or remove the trace
5944 kvm_set_rflags(vcpu, rflags);
5946 kvm_x86_ops->set_guest_debug(vcpu, dbg);
5956 * Translate a guest virtual address to a guest physical address.
5958 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5959 struct kvm_translation *tr)
5961 unsigned long vaddr = tr->linear_address;
5965 idx = srcu_read_lock(&vcpu->kvm->srcu);
5966 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5967 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5968 tr->physical_address = gpa;
5969 tr->valid = gpa != UNMAPPED_GVA;
5976 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5978 struct i387_fxsave_struct *fxsave =
5979 &vcpu->arch.guest_fpu.state->fxsave;
5981 memcpy(fpu->fpr, fxsave->st_space, 128);
5982 fpu->fcw = fxsave->cwd;
5983 fpu->fsw = fxsave->swd;
5984 fpu->ftwx = fxsave->twd;
5985 fpu->last_opcode = fxsave->fop;
5986 fpu->last_ip = fxsave->rip;
5987 fpu->last_dp = fxsave->rdp;
5988 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5993 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5995 struct i387_fxsave_struct *fxsave =
5996 &vcpu->arch.guest_fpu.state->fxsave;
5998 memcpy(fxsave->st_space, fpu->fpr, 128);
5999 fxsave->cwd = fpu->fcw;
6000 fxsave->swd = fpu->fsw;
6001 fxsave->twd = fpu->ftwx;
6002 fxsave->fop = fpu->last_opcode;
6003 fxsave->rip = fpu->last_ip;
6004 fxsave->rdp = fpu->last_dp;
6005 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6010 int fx_init(struct kvm_vcpu *vcpu)
6014 err = fpu_alloc(&vcpu->arch.guest_fpu);
6018 fpu_finit(&vcpu->arch.guest_fpu);
6021 * Ensure guest xcr0 is valid for loading
6023 vcpu->arch.xcr0 = XSTATE_FP;
6025 vcpu->arch.cr0 |= X86_CR0_ET;
6029 EXPORT_SYMBOL_GPL(fx_init);
6031 static void fx_free(struct kvm_vcpu *vcpu)
6033 fpu_free(&vcpu->arch.guest_fpu);
6036 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6038 if (vcpu->guest_fpu_loaded)
6042 * Restore all possible states in the guest,
6043 * and assume host would use all available bits.
6044 * Guest xcr0 would be loaded later.
6046 kvm_put_guest_xcr0(vcpu);
6047 vcpu->guest_fpu_loaded = 1;
6048 unlazy_fpu(current);
6049 fpu_restore_checking(&vcpu->arch.guest_fpu);
6053 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6055 kvm_put_guest_xcr0(vcpu);
6057 if (!vcpu->guest_fpu_loaded)
6060 vcpu->guest_fpu_loaded = 0;
6061 fpu_save_init(&vcpu->arch.guest_fpu);
6062 ++vcpu->stat.fpu_reload;
6063 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6067 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6069 kvmclock_reset(vcpu);
6071 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6073 kvm_x86_ops->vcpu_free(vcpu);
6076 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6079 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6080 printk_once(KERN_WARNING
6081 "kvm: SMP vm created on host with unstable TSC; "
6082 "guest TSC will not be reliable\n");
6083 return kvm_x86_ops->vcpu_create(kvm, id);
6086 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6090 vcpu->arch.mtrr_state.have_fixed = 1;
6092 r = kvm_arch_vcpu_reset(vcpu);
6094 r = kvm_mmu_setup(vcpu);
6101 kvm_x86_ops->vcpu_free(vcpu);
6105 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6107 vcpu->arch.apf.msr_val = 0;
6110 kvm_mmu_unload(vcpu);
6114 kvm_x86_ops->vcpu_free(vcpu);
6117 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
6119 vcpu->arch.nmi_pending = false;
6120 vcpu->arch.nmi_injected = false;
6122 vcpu->arch.switch_db_regs = 0;
6123 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6124 vcpu->arch.dr6 = DR6_FIXED_1;
6125 vcpu->arch.dr7 = DR7_FIXED_1;
6127 kvm_make_request(KVM_REQ_EVENT, vcpu);
6128 vcpu->arch.apf.msr_val = 0;
6130 kvmclock_reset(vcpu);
6132 kvm_clear_async_pf_completion_queue(vcpu);
6133 kvm_async_pf_hash_reset(vcpu);
6134 vcpu->arch.apf.halted = false;
6136 return kvm_x86_ops->vcpu_reset(vcpu);
6139 int kvm_arch_hardware_enable(void *garbage)
6142 struct kvm_vcpu *vcpu;
6145 kvm_shared_msr_cpu_online();
6146 list_for_each_entry(kvm, &vm_list, vm_list)
6147 kvm_for_each_vcpu(i, vcpu, kvm)
6148 if (vcpu->cpu == smp_processor_id())
6149 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6150 return kvm_x86_ops->hardware_enable(garbage);
6153 void kvm_arch_hardware_disable(void *garbage)
6155 kvm_x86_ops->hardware_disable(garbage);
6156 drop_user_return_notifiers(garbage);
6159 int kvm_arch_hardware_setup(void)
6161 return kvm_x86_ops->hardware_setup();
6164 void kvm_arch_hardware_unsetup(void)
6166 kvm_x86_ops->hardware_unsetup();
6169 void kvm_arch_check_processor_compat(void *rtn)
6171 kvm_x86_ops->check_processor_compatibility(rtn);
6174 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6180 BUG_ON(vcpu->kvm == NULL);
6183 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6184 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
6185 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
6186 vcpu->arch.mmu.translate_gpa = translate_gpa;
6187 vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
6188 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6189 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6191 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6193 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6198 vcpu->arch.pio_data = page_address(page);
6200 kvm_init_tsc_catchup(vcpu, max_tsc_khz);
6202 r = kvm_mmu_create(vcpu);
6204 goto fail_free_pio_data;
6206 if (irqchip_in_kernel(kvm)) {
6207 r = kvm_create_lapic(vcpu);
6209 goto fail_mmu_destroy;
6212 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6214 if (!vcpu->arch.mce_banks) {
6216 goto fail_free_lapic;
6218 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6220 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6221 goto fail_free_mce_banks;
6223 kvm_async_pf_hash_reset(vcpu);
6226 fail_free_mce_banks:
6227 kfree(vcpu->arch.mce_banks);
6229 kvm_free_lapic(vcpu);
6231 kvm_mmu_destroy(vcpu);
6233 free_page((unsigned long)vcpu->arch.pio_data);
6238 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6242 kfree(vcpu->arch.mce_banks);
6243 kvm_free_lapic(vcpu);
6244 idx = srcu_read_lock(&vcpu->kvm->srcu);
6245 kvm_mmu_destroy(vcpu);
6246 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6247 free_page((unsigned long)vcpu->arch.pio_data);
6250 int kvm_arch_init_vm(struct kvm *kvm)
6252 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6253 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6255 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6256 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6258 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6263 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6266 kvm_mmu_unload(vcpu);
6270 static void kvm_free_vcpus(struct kvm *kvm)
6273 struct kvm_vcpu *vcpu;
6276 * Unpin any mmu pages first.
6278 kvm_for_each_vcpu(i, vcpu, kvm) {
6279 kvm_clear_async_pf_completion_queue(vcpu);
6280 kvm_unload_vcpu_mmu(vcpu);
6282 kvm_for_each_vcpu(i, vcpu, kvm)
6283 kvm_arch_vcpu_free(vcpu);
6285 mutex_lock(&kvm->lock);
6286 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6287 kvm->vcpus[i] = NULL;
6289 atomic_set(&kvm->online_vcpus, 0);
6290 mutex_unlock(&kvm->lock);
6293 void kvm_arch_sync_events(struct kvm *kvm)
6295 kvm_free_all_assigned_devices(kvm);
6299 void kvm_arch_destroy_vm(struct kvm *kvm)
6301 kvm_iommu_unmap_guest(kvm);
6302 kfree(kvm->arch.vpic);
6303 kfree(kvm->arch.vioapic);
6304 kvm_free_vcpus(kvm);
6305 if (kvm->arch.apic_access_page)
6306 put_page(kvm->arch.apic_access_page);
6307 if (kvm->arch.ept_identity_pagetable)
6308 put_page(kvm->arch.ept_identity_pagetable);
6311 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6312 struct kvm_memory_slot *memslot,
6313 struct kvm_memory_slot old,
6314 struct kvm_userspace_memory_region *mem,
6317 int npages = memslot->npages;
6318 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6320 /* Prevent internal slot pages from being moved by fork()/COW. */
6321 if (memslot->id >= KVM_MEMORY_SLOTS)
6322 map_flags = MAP_SHARED | MAP_ANONYMOUS;
6324 /*To keep backward compatibility with older userspace,
6325 *x86 needs to hanlde !user_alloc case.
6328 if (npages && !old.rmap) {
6329 unsigned long userspace_addr;
6331 down_write(¤t->mm->mmap_sem);
6332 userspace_addr = do_mmap(NULL, 0,
6334 PROT_READ | PROT_WRITE,
6337 up_write(¤t->mm->mmap_sem);
6339 if (IS_ERR((void *)userspace_addr))
6340 return PTR_ERR((void *)userspace_addr);
6342 memslot->userspace_addr = userspace_addr;
6350 void kvm_arch_commit_memory_region(struct kvm *kvm,
6351 struct kvm_userspace_memory_region *mem,
6352 struct kvm_memory_slot old,
6356 int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
6358 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6361 down_write(¤t->mm->mmap_sem);
6362 ret = do_munmap(current->mm, old.userspace_addr,
6363 old.npages * PAGE_SIZE);
6364 up_write(¤t->mm->mmap_sem);
6367 "kvm_vm_ioctl_set_memory_region: "
6368 "failed to munmap memory\n");
6371 if (!kvm->arch.n_requested_mmu_pages)
6372 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6374 spin_lock(&kvm->mmu_lock);
6376 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6377 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6378 spin_unlock(&kvm->mmu_lock);
6381 void kvm_arch_flush_shadow(struct kvm *kvm)
6383 kvm_mmu_zap_all(kvm);
6384 kvm_reload_remote_mmus(kvm);
6387 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6389 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6390 !vcpu->arch.apf.halted)
6391 || !list_empty_careful(&vcpu->async_pf.done)
6392 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6393 || vcpu->arch.nmi_pending ||
6394 (kvm_arch_interrupt_allowed(vcpu) &&
6395 kvm_cpu_has_interrupt(vcpu));
6398 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6401 int cpu = vcpu->cpu;
6403 if (waitqueue_active(&vcpu->wq)) {
6404 wake_up_interruptible(&vcpu->wq);
6405 ++vcpu->stat.halt_wakeup;
6409 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6410 if (kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE)
6411 smp_send_reschedule(cpu);
6415 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6417 return kvm_x86_ops->interrupt_allowed(vcpu);
6420 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6422 unsigned long current_rip = kvm_rip_read(vcpu) +
6423 get_segment_base(vcpu, VCPU_SREG_CS);
6425 return current_rip == linear_rip;
6427 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6429 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6431 unsigned long rflags;
6433 rflags = kvm_x86_ops->get_rflags(vcpu);
6434 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6435 rflags &= ~X86_EFLAGS_TF;
6438 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6440 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6442 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6443 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6444 rflags |= X86_EFLAGS_TF;
6445 kvm_x86_ops->set_rflags(vcpu, rflags);
6446 kvm_make_request(KVM_REQ_EVENT, vcpu);
6448 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6450 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6454 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6455 is_error_page(work->page))
6458 r = kvm_mmu_reload(vcpu);
6462 if (!vcpu->arch.mmu.direct_map &&
6463 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6466 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6469 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6471 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6474 static inline u32 kvm_async_pf_next_probe(u32 key)
6476 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6479 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6481 u32 key = kvm_async_pf_hash_fn(gfn);
6483 while (vcpu->arch.apf.gfns[key] != ~0)
6484 key = kvm_async_pf_next_probe(key);
6486 vcpu->arch.apf.gfns[key] = gfn;
6489 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6492 u32 key = kvm_async_pf_hash_fn(gfn);
6494 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6495 (vcpu->arch.apf.gfns[key] != gfn &&
6496 vcpu->arch.apf.gfns[key] != ~0); i++)
6497 key = kvm_async_pf_next_probe(key);
6502 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6504 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6507 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6511 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6513 vcpu->arch.apf.gfns[i] = ~0;
6515 j = kvm_async_pf_next_probe(j);
6516 if (vcpu->arch.apf.gfns[j] == ~0)
6518 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6520 * k lies cyclically in ]i,j]
6522 * |....j i.k.| or |.k..j i...|
6524 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6525 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6530 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6533 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6537 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6538 struct kvm_async_pf *work)
6540 struct x86_exception fault;
6542 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6543 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6545 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6546 (vcpu->arch.apf.send_user_only &&
6547 kvm_x86_ops->get_cpl(vcpu) == 0))
6548 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6549 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6550 fault.vector = PF_VECTOR;
6551 fault.error_code_valid = true;
6552 fault.error_code = 0;
6553 fault.nested_page_fault = false;
6554 fault.address = work->arch.token;
6555 kvm_inject_page_fault(vcpu, &fault);
6559 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6560 struct kvm_async_pf *work)
6562 struct x86_exception fault;
6564 trace_kvm_async_pf_ready(work->arch.token, work->gva);
6565 if (is_error_page(work->page))
6566 work->arch.token = ~0; /* broadcast wakeup */
6568 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6570 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6571 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6572 fault.vector = PF_VECTOR;
6573 fault.error_code_valid = true;
6574 fault.error_code = 0;
6575 fault.nested_page_fault = false;
6576 fault.address = work->arch.token;
6577 kvm_inject_page_fault(vcpu, &fault);
6579 vcpu->arch.apf.halted = false;
6582 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6584 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6587 return !kvm_event_needs_reinjection(vcpu) &&
6588 kvm_x86_ops->interrupt_allowed(vcpu);
6591 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6592 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6593 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6594 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6595 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6596 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6597 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6598 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6599 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6600 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6601 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6602 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);