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 (kvm_x86_ops->set_cr4(vcpu, cr4))
621 if ((cr4 ^ old_cr4) & pdptr_bits)
622 kvm_mmu_reset_context(vcpu);
624 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
629 EXPORT_SYMBOL_GPL(kvm_set_cr4);
631 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
633 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
634 kvm_mmu_sync_roots(vcpu);
635 kvm_mmu_flush_tlb(vcpu);
639 if (is_long_mode(vcpu)) {
640 if (cr3 & CR3_L_MODE_RESERVED_BITS)
644 if (cr3 & CR3_PAE_RESERVED_BITS)
646 if (is_paging(vcpu) &&
647 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
651 * We don't check reserved bits in nonpae mode, because
652 * this isn't enforced, and VMware depends on this.
657 * Does the new cr3 value map to physical memory? (Note, we
658 * catch an invalid cr3 even in real-mode, because it would
659 * cause trouble later on when we turn on paging anyway.)
661 * A real CPU would silently accept an invalid cr3 and would
662 * attempt to use it - with largely undefined (and often hard
663 * to debug) behavior on the guest side.
665 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
667 vcpu->arch.cr3 = cr3;
668 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
669 vcpu->arch.mmu.new_cr3(vcpu);
672 EXPORT_SYMBOL_GPL(kvm_set_cr3);
674 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
676 if (cr8 & CR8_RESERVED_BITS)
678 if (irqchip_in_kernel(vcpu->kvm))
679 kvm_lapic_set_tpr(vcpu, cr8);
681 vcpu->arch.cr8 = cr8;
684 EXPORT_SYMBOL_GPL(kvm_set_cr8);
686 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
688 if (irqchip_in_kernel(vcpu->kvm))
689 return kvm_lapic_get_cr8(vcpu);
691 return vcpu->arch.cr8;
693 EXPORT_SYMBOL_GPL(kvm_get_cr8);
695 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
699 vcpu->arch.db[dr] = val;
700 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
701 vcpu->arch.eff_db[dr] = val;
704 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
708 if (val & 0xffffffff00000000ULL)
710 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
713 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
717 if (val & 0xffffffff00000000ULL)
719 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
720 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
721 kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
722 vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
730 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
734 res = __kvm_set_dr(vcpu, dr, val);
736 kvm_queue_exception(vcpu, UD_VECTOR);
738 kvm_inject_gp(vcpu, 0);
742 EXPORT_SYMBOL_GPL(kvm_set_dr);
744 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
748 *val = vcpu->arch.db[dr];
751 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
755 *val = vcpu->arch.dr6;
758 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
762 *val = vcpu->arch.dr7;
769 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
771 if (_kvm_get_dr(vcpu, dr, val)) {
772 kvm_queue_exception(vcpu, UD_VECTOR);
777 EXPORT_SYMBOL_GPL(kvm_get_dr);
780 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
781 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
783 * This list is modified at module load time to reflect the
784 * capabilities of the host cpu. This capabilities test skips MSRs that are
785 * kvm-specific. Those are put in the beginning of the list.
788 #define KVM_SAVE_MSRS_BEGIN 8
789 static u32 msrs_to_save[] = {
790 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
791 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
792 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
793 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN,
794 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
797 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
799 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
802 static unsigned num_msrs_to_save;
804 static u32 emulated_msrs[] = {
805 MSR_IA32_MISC_ENABLE,
810 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
812 u64 old_efer = vcpu->arch.efer;
814 if (efer & efer_reserved_bits)
818 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
821 if (efer & EFER_FFXSR) {
822 struct kvm_cpuid_entry2 *feat;
824 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
825 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
829 if (efer & EFER_SVME) {
830 struct kvm_cpuid_entry2 *feat;
832 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
833 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
838 efer |= vcpu->arch.efer & EFER_LMA;
840 kvm_x86_ops->set_efer(vcpu, efer);
842 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
844 /* Update reserved bits */
845 if ((efer ^ old_efer) & EFER_NX)
846 kvm_mmu_reset_context(vcpu);
851 void kvm_enable_efer_bits(u64 mask)
853 efer_reserved_bits &= ~mask;
855 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
859 * Writes msr value into into the appropriate "register".
860 * Returns 0 on success, non-0 otherwise.
861 * Assumes vcpu_load() was already called.
863 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
865 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
869 * Adapt set_msr() to msr_io()'s calling convention
871 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
873 return kvm_set_msr(vcpu, index, *data);
876 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
880 struct pvclock_wall_clock wc;
881 struct timespec boot;
886 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
891 ++version; /* first time write, random junk */
895 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
898 * The guest calculates current wall clock time by adding
899 * system time (updated by kvm_guest_time_update below) to the
900 * wall clock specified here. guest system time equals host
901 * system time for us, thus we must fill in host boot time here.
905 wc.sec = boot.tv_sec;
906 wc.nsec = boot.tv_nsec;
907 wc.version = version;
909 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
912 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
915 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
917 uint32_t quotient, remainder;
919 /* Don't try to replace with do_div(), this one calculates
920 * "(dividend << 32) / divisor" */
922 : "=a" (quotient), "=d" (remainder)
923 : "0" (0), "1" (dividend), "r" (divisor) );
927 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
928 s8 *pshift, u32 *pmultiplier)
935 tps64 = base_khz * 1000LL;
936 scaled64 = scaled_khz * 1000LL;
937 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
942 tps32 = (uint32_t)tps64;
943 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
944 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
952 *pmultiplier = div_frac(scaled64, tps32);
954 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
955 __func__, base_khz, scaled_khz, shift, *pmultiplier);
958 static inline u64 get_kernel_ns(void)
962 WARN_ON(preemptible());
964 monotonic_to_bootbased(&ts);
965 return timespec_to_ns(&ts);
968 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
969 unsigned long max_tsc_khz;
971 static inline int kvm_tsc_changes_freq(void)
974 int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
975 cpufreq_quick_get(cpu) != 0;
980 static u64 vcpu_tsc_khz(struct kvm_vcpu *vcpu)
982 if (vcpu->arch.virtual_tsc_khz)
983 return vcpu->arch.virtual_tsc_khz;
985 return __this_cpu_read(cpu_tsc_khz);
988 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
992 WARN_ON(preemptible());
993 if (kvm_tsc_changes_freq())
994 printk_once(KERN_WARNING
995 "kvm: unreliable cycle conversion on adjustable rate TSC\n");
996 ret = nsec * vcpu_tsc_khz(vcpu);
997 do_div(ret, USEC_PER_SEC);
1001 static void kvm_init_tsc_catchup(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1003 /* Compute a scale to convert nanoseconds in TSC cycles */
1004 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1005 &vcpu->arch.tsc_catchup_shift,
1006 &vcpu->arch.tsc_catchup_mult);
1009 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1011 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1012 vcpu->arch.tsc_catchup_mult,
1013 vcpu->arch.tsc_catchup_shift);
1014 tsc += vcpu->arch.last_tsc_write;
1018 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1020 struct kvm *kvm = vcpu->kvm;
1021 u64 offset, ns, elapsed;
1022 unsigned long flags;
1025 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1026 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1027 ns = get_kernel_ns();
1028 elapsed = ns - kvm->arch.last_tsc_nsec;
1029 sdiff = data - kvm->arch.last_tsc_write;
1034 * Special case: close write to TSC within 5 seconds of
1035 * another CPU is interpreted as an attempt to synchronize
1036 * The 5 seconds is to accommodate host load / swapping as
1037 * well as any reset of TSC during the boot process.
1039 * In that case, for a reliable TSC, we can match TSC offsets,
1040 * or make a best guest using elapsed value.
1042 if (sdiff < nsec_to_cycles(vcpu, 5ULL * NSEC_PER_SEC) &&
1043 elapsed < 5ULL * NSEC_PER_SEC) {
1044 if (!check_tsc_unstable()) {
1045 offset = kvm->arch.last_tsc_offset;
1046 pr_debug("kvm: matched tsc offset for %llu\n", data);
1048 u64 delta = nsec_to_cycles(vcpu, elapsed);
1050 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1052 ns = kvm->arch.last_tsc_nsec;
1054 kvm->arch.last_tsc_nsec = ns;
1055 kvm->arch.last_tsc_write = data;
1056 kvm->arch.last_tsc_offset = offset;
1057 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1058 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1060 /* Reset of TSC must disable overshoot protection below */
1061 vcpu->arch.hv_clock.tsc_timestamp = 0;
1062 vcpu->arch.last_tsc_write = data;
1063 vcpu->arch.last_tsc_nsec = ns;
1065 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1067 static int kvm_guest_time_update(struct kvm_vcpu *v)
1069 unsigned long flags;
1070 struct kvm_vcpu_arch *vcpu = &v->arch;
1072 unsigned long this_tsc_khz;
1073 s64 kernel_ns, max_kernel_ns;
1076 /* Keep irq disabled to prevent changes to the clock */
1077 local_irq_save(flags);
1078 kvm_get_msr(v, MSR_IA32_TSC, &tsc_timestamp);
1079 kernel_ns = get_kernel_ns();
1080 this_tsc_khz = vcpu_tsc_khz(v);
1081 if (unlikely(this_tsc_khz == 0)) {
1082 local_irq_restore(flags);
1083 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1088 * We may have to catch up the TSC to match elapsed wall clock
1089 * time for two reasons, even if kvmclock is used.
1090 * 1) CPU could have been running below the maximum TSC rate
1091 * 2) Broken TSC compensation resets the base at each VCPU
1092 * entry to avoid unknown leaps of TSC even when running
1093 * again on the same CPU. This may cause apparent elapsed
1094 * time to disappear, and the guest to stand still or run
1097 if (vcpu->tsc_catchup) {
1098 u64 tsc = compute_guest_tsc(v, kernel_ns);
1099 if (tsc > tsc_timestamp) {
1100 kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1101 tsc_timestamp = tsc;
1105 local_irq_restore(flags);
1107 if (!vcpu->time_page)
1111 * Time as measured by the TSC may go backwards when resetting the base
1112 * tsc_timestamp. The reason for this is that the TSC resolution is
1113 * higher than the resolution of the other clock scales. Thus, many
1114 * possible measurments of the TSC correspond to one measurement of any
1115 * other clock, and so a spread of values is possible. This is not a
1116 * problem for the computation of the nanosecond clock; with TSC rates
1117 * around 1GHZ, there can only be a few cycles which correspond to one
1118 * nanosecond value, and any path through this code will inevitably
1119 * take longer than that. However, with the kernel_ns value itself,
1120 * the precision may be much lower, down to HZ granularity. If the
1121 * first sampling of TSC against kernel_ns ends in the low part of the
1122 * range, and the second in the high end of the range, we can get:
1124 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1126 * As the sampling errors potentially range in the thousands of cycles,
1127 * it is possible such a time value has already been observed by the
1128 * guest. To protect against this, we must compute the system time as
1129 * observed by the guest and ensure the new system time is greater.
1132 if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1133 max_kernel_ns = vcpu->last_guest_tsc -
1134 vcpu->hv_clock.tsc_timestamp;
1135 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1136 vcpu->hv_clock.tsc_to_system_mul,
1137 vcpu->hv_clock.tsc_shift);
1138 max_kernel_ns += vcpu->last_kernel_ns;
1141 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1142 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1143 &vcpu->hv_clock.tsc_shift,
1144 &vcpu->hv_clock.tsc_to_system_mul);
1145 vcpu->hw_tsc_khz = this_tsc_khz;
1148 if (max_kernel_ns > kernel_ns)
1149 kernel_ns = max_kernel_ns;
1151 /* With all the info we got, fill in the values */
1152 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1153 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1154 vcpu->last_kernel_ns = kernel_ns;
1155 vcpu->last_guest_tsc = tsc_timestamp;
1156 vcpu->hv_clock.flags = 0;
1159 * The interface expects us to write an even number signaling that the
1160 * update is finished. Since the guest won't see the intermediate
1161 * state, we just increase by 2 at the end.
1163 vcpu->hv_clock.version += 2;
1165 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1167 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1168 sizeof(vcpu->hv_clock));
1170 kunmap_atomic(shared_kaddr, KM_USER0);
1172 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1176 static bool msr_mtrr_valid(unsigned msr)
1179 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1180 case MSR_MTRRfix64K_00000:
1181 case MSR_MTRRfix16K_80000:
1182 case MSR_MTRRfix16K_A0000:
1183 case MSR_MTRRfix4K_C0000:
1184 case MSR_MTRRfix4K_C8000:
1185 case MSR_MTRRfix4K_D0000:
1186 case MSR_MTRRfix4K_D8000:
1187 case MSR_MTRRfix4K_E0000:
1188 case MSR_MTRRfix4K_E8000:
1189 case MSR_MTRRfix4K_F0000:
1190 case MSR_MTRRfix4K_F8000:
1191 case MSR_MTRRdefType:
1192 case MSR_IA32_CR_PAT:
1200 static bool valid_pat_type(unsigned t)
1202 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1205 static bool valid_mtrr_type(unsigned t)
1207 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1210 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1214 if (!msr_mtrr_valid(msr))
1217 if (msr == MSR_IA32_CR_PAT) {
1218 for (i = 0; i < 8; i++)
1219 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1222 } else if (msr == MSR_MTRRdefType) {
1225 return valid_mtrr_type(data & 0xff);
1226 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1227 for (i = 0; i < 8 ; i++)
1228 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1233 /* variable MTRRs */
1234 return valid_mtrr_type(data & 0xff);
1237 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1239 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1241 if (!mtrr_valid(vcpu, msr, data))
1244 if (msr == MSR_MTRRdefType) {
1245 vcpu->arch.mtrr_state.def_type = data;
1246 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1247 } else if (msr == MSR_MTRRfix64K_00000)
1249 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1250 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1251 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1252 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1253 else if (msr == MSR_IA32_CR_PAT)
1254 vcpu->arch.pat = data;
1255 else { /* Variable MTRRs */
1256 int idx, is_mtrr_mask;
1259 idx = (msr - 0x200) / 2;
1260 is_mtrr_mask = msr - 0x200 - 2 * idx;
1263 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1266 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1270 kvm_mmu_reset_context(vcpu);
1274 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1276 u64 mcg_cap = vcpu->arch.mcg_cap;
1277 unsigned bank_num = mcg_cap & 0xff;
1280 case MSR_IA32_MCG_STATUS:
1281 vcpu->arch.mcg_status = data;
1283 case MSR_IA32_MCG_CTL:
1284 if (!(mcg_cap & MCG_CTL_P))
1286 if (data != 0 && data != ~(u64)0)
1288 vcpu->arch.mcg_ctl = data;
1291 if (msr >= MSR_IA32_MC0_CTL &&
1292 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1293 u32 offset = msr - MSR_IA32_MC0_CTL;
1294 /* only 0 or all 1s can be written to IA32_MCi_CTL
1295 * some Linux kernels though clear bit 10 in bank 4 to
1296 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1297 * this to avoid an uncatched #GP in the guest
1299 if ((offset & 0x3) == 0 &&
1300 data != 0 && (data | (1 << 10)) != ~(u64)0)
1302 vcpu->arch.mce_banks[offset] = data;
1310 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1312 struct kvm *kvm = vcpu->kvm;
1313 int lm = is_long_mode(vcpu);
1314 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1315 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1316 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1317 : kvm->arch.xen_hvm_config.blob_size_32;
1318 u32 page_num = data & ~PAGE_MASK;
1319 u64 page_addr = data & PAGE_MASK;
1324 if (page_num >= blob_size)
1327 page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1331 if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1333 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1342 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1344 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1347 static bool kvm_hv_msr_partition_wide(u32 msr)
1351 case HV_X64_MSR_GUEST_OS_ID:
1352 case HV_X64_MSR_HYPERCALL:
1360 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1362 struct kvm *kvm = vcpu->kvm;
1365 case HV_X64_MSR_GUEST_OS_ID:
1366 kvm->arch.hv_guest_os_id = data;
1367 /* setting guest os id to zero disables hypercall page */
1368 if (!kvm->arch.hv_guest_os_id)
1369 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1371 case HV_X64_MSR_HYPERCALL: {
1376 /* if guest os id is not set hypercall should remain disabled */
1377 if (!kvm->arch.hv_guest_os_id)
1379 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1380 kvm->arch.hv_hypercall = data;
1383 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1384 addr = gfn_to_hva(kvm, gfn);
1385 if (kvm_is_error_hva(addr))
1387 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1388 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1389 if (__copy_to_user((void __user *)addr, instructions, 4))
1391 kvm->arch.hv_hypercall = data;
1395 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1396 "data 0x%llx\n", msr, data);
1402 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1405 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1408 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1409 vcpu->arch.hv_vapic = data;
1412 addr = gfn_to_hva(vcpu->kvm, data >>
1413 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1414 if (kvm_is_error_hva(addr))
1416 if (__clear_user((void __user *)addr, PAGE_SIZE))
1418 vcpu->arch.hv_vapic = data;
1421 case HV_X64_MSR_EOI:
1422 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1423 case HV_X64_MSR_ICR:
1424 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1425 case HV_X64_MSR_TPR:
1426 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1428 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1429 "data 0x%llx\n", msr, data);
1436 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1438 gpa_t gpa = data & ~0x3f;
1440 /* Bits 2:5 are resrved, Should be zero */
1444 vcpu->arch.apf.msr_val = data;
1446 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1447 kvm_clear_async_pf_completion_queue(vcpu);
1448 kvm_async_pf_hash_reset(vcpu);
1452 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1455 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1456 kvm_async_pf_wakeup_all(vcpu);
1460 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1462 if (vcpu->arch.time_page) {
1463 kvm_release_page_dirty(vcpu->arch.time_page);
1464 vcpu->arch.time_page = NULL;
1468 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1472 return set_efer(vcpu, data);
1474 data &= ~(u64)0x40; /* ignore flush filter disable */
1475 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1477 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1482 case MSR_FAM10H_MMIO_CONF_BASE:
1484 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1489 case MSR_AMD64_NB_CFG:
1491 case MSR_IA32_DEBUGCTLMSR:
1493 /* We support the non-activated case already */
1495 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1496 /* Values other than LBR and BTF are vendor-specific,
1497 thus reserved and should throw a #GP */
1500 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1503 case MSR_IA32_UCODE_REV:
1504 case MSR_IA32_UCODE_WRITE:
1505 case MSR_VM_HSAVE_PA:
1506 case MSR_AMD64_PATCH_LOADER:
1508 case 0x200 ... 0x2ff:
1509 return set_msr_mtrr(vcpu, msr, data);
1510 case MSR_IA32_APICBASE:
1511 kvm_set_apic_base(vcpu, data);
1513 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1514 return kvm_x2apic_msr_write(vcpu, msr, data);
1515 case MSR_IA32_MISC_ENABLE:
1516 vcpu->arch.ia32_misc_enable_msr = data;
1518 case MSR_KVM_WALL_CLOCK_NEW:
1519 case MSR_KVM_WALL_CLOCK:
1520 vcpu->kvm->arch.wall_clock = data;
1521 kvm_write_wall_clock(vcpu->kvm, data);
1523 case MSR_KVM_SYSTEM_TIME_NEW:
1524 case MSR_KVM_SYSTEM_TIME: {
1525 kvmclock_reset(vcpu);
1527 vcpu->arch.time = data;
1528 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1530 /* we verify if the enable bit is set... */
1534 /* ...but clean it before doing the actual write */
1535 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1537 vcpu->arch.time_page =
1538 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1540 if (is_error_page(vcpu->arch.time_page)) {
1541 kvm_release_page_clean(vcpu->arch.time_page);
1542 vcpu->arch.time_page = NULL;
1546 case MSR_KVM_ASYNC_PF_EN:
1547 if (kvm_pv_enable_async_pf(vcpu, data))
1550 case MSR_IA32_MCG_CTL:
1551 case MSR_IA32_MCG_STATUS:
1552 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1553 return set_msr_mce(vcpu, msr, data);
1555 /* Performance counters are not protected by a CPUID bit,
1556 * so we should check all of them in the generic path for the sake of
1557 * cross vendor migration.
1558 * Writing a zero into the event select MSRs disables them,
1559 * which we perfectly emulate ;-). Any other value should be at least
1560 * reported, some guests depend on them.
1562 case MSR_P6_EVNTSEL0:
1563 case MSR_P6_EVNTSEL1:
1564 case MSR_K7_EVNTSEL0:
1565 case MSR_K7_EVNTSEL1:
1566 case MSR_K7_EVNTSEL2:
1567 case MSR_K7_EVNTSEL3:
1569 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1570 "0x%x data 0x%llx\n", msr, data);
1572 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1573 * so we ignore writes to make it happy.
1575 case MSR_P6_PERFCTR0:
1576 case MSR_P6_PERFCTR1:
1577 case MSR_K7_PERFCTR0:
1578 case MSR_K7_PERFCTR1:
1579 case MSR_K7_PERFCTR2:
1580 case MSR_K7_PERFCTR3:
1581 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1582 "0x%x data 0x%llx\n", msr, data);
1584 case MSR_K7_CLK_CTL:
1586 * Ignore all writes to this no longer documented MSR.
1587 * Writes are only relevant for old K7 processors,
1588 * all pre-dating SVM, but a recommended workaround from
1589 * AMD for these chips. It is possible to speicify the
1590 * affected processor models on the command line, hence
1591 * the need to ignore the workaround.
1594 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1595 if (kvm_hv_msr_partition_wide(msr)) {
1597 mutex_lock(&vcpu->kvm->lock);
1598 r = set_msr_hyperv_pw(vcpu, msr, data);
1599 mutex_unlock(&vcpu->kvm->lock);
1602 return set_msr_hyperv(vcpu, msr, data);
1604 case MSR_IA32_BBL_CR_CTL3:
1605 /* Drop writes to this legacy MSR -- see rdmsr
1606 * counterpart for further detail.
1608 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1611 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1612 return xen_hvm_config(vcpu, data);
1614 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1618 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1625 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1629 * Reads an msr value (of 'msr_index') into 'pdata'.
1630 * Returns 0 on success, non-0 otherwise.
1631 * Assumes vcpu_load() was already called.
1633 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1635 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1638 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1640 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1642 if (!msr_mtrr_valid(msr))
1645 if (msr == MSR_MTRRdefType)
1646 *pdata = vcpu->arch.mtrr_state.def_type +
1647 (vcpu->arch.mtrr_state.enabled << 10);
1648 else if (msr == MSR_MTRRfix64K_00000)
1650 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1651 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1652 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1653 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1654 else if (msr == MSR_IA32_CR_PAT)
1655 *pdata = vcpu->arch.pat;
1656 else { /* Variable MTRRs */
1657 int idx, is_mtrr_mask;
1660 idx = (msr - 0x200) / 2;
1661 is_mtrr_mask = msr - 0x200 - 2 * idx;
1664 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1667 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1674 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1677 u64 mcg_cap = vcpu->arch.mcg_cap;
1678 unsigned bank_num = mcg_cap & 0xff;
1681 case MSR_IA32_P5_MC_ADDR:
1682 case MSR_IA32_P5_MC_TYPE:
1685 case MSR_IA32_MCG_CAP:
1686 data = vcpu->arch.mcg_cap;
1688 case MSR_IA32_MCG_CTL:
1689 if (!(mcg_cap & MCG_CTL_P))
1691 data = vcpu->arch.mcg_ctl;
1693 case MSR_IA32_MCG_STATUS:
1694 data = vcpu->arch.mcg_status;
1697 if (msr >= MSR_IA32_MC0_CTL &&
1698 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1699 u32 offset = msr - MSR_IA32_MC0_CTL;
1700 data = vcpu->arch.mce_banks[offset];
1709 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1712 struct kvm *kvm = vcpu->kvm;
1715 case HV_X64_MSR_GUEST_OS_ID:
1716 data = kvm->arch.hv_guest_os_id;
1718 case HV_X64_MSR_HYPERCALL:
1719 data = kvm->arch.hv_hypercall;
1722 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1730 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1735 case HV_X64_MSR_VP_INDEX: {
1738 kvm_for_each_vcpu(r, v, vcpu->kvm)
1743 case HV_X64_MSR_EOI:
1744 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1745 case HV_X64_MSR_ICR:
1746 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1747 case HV_X64_MSR_TPR:
1748 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1750 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1757 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1762 case MSR_IA32_PLATFORM_ID:
1763 case MSR_IA32_UCODE_REV:
1764 case MSR_IA32_EBL_CR_POWERON:
1765 case MSR_IA32_DEBUGCTLMSR:
1766 case MSR_IA32_LASTBRANCHFROMIP:
1767 case MSR_IA32_LASTBRANCHTOIP:
1768 case MSR_IA32_LASTINTFROMIP:
1769 case MSR_IA32_LASTINTTOIP:
1772 case MSR_VM_HSAVE_PA:
1773 case MSR_P6_PERFCTR0:
1774 case MSR_P6_PERFCTR1:
1775 case MSR_P6_EVNTSEL0:
1776 case MSR_P6_EVNTSEL1:
1777 case MSR_K7_EVNTSEL0:
1778 case MSR_K7_PERFCTR0:
1779 case MSR_K8_INT_PENDING_MSG:
1780 case MSR_AMD64_NB_CFG:
1781 case MSR_FAM10H_MMIO_CONF_BASE:
1785 data = 0x500 | KVM_NR_VAR_MTRR;
1787 case 0x200 ... 0x2ff:
1788 return get_msr_mtrr(vcpu, msr, pdata);
1789 case 0xcd: /* fsb frequency */
1793 * MSR_EBC_FREQUENCY_ID
1794 * Conservative value valid for even the basic CPU models.
1795 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1796 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1797 * and 266MHz for model 3, or 4. Set Core Clock
1798 * Frequency to System Bus Frequency Ratio to 1 (bits
1799 * 31:24) even though these are only valid for CPU
1800 * models > 2, however guests may end up dividing or
1801 * multiplying by zero otherwise.
1803 case MSR_EBC_FREQUENCY_ID:
1806 case MSR_IA32_APICBASE:
1807 data = kvm_get_apic_base(vcpu);
1809 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1810 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1812 case MSR_IA32_MISC_ENABLE:
1813 data = vcpu->arch.ia32_misc_enable_msr;
1815 case MSR_IA32_PERF_STATUS:
1816 /* TSC increment by tick */
1818 /* CPU multiplier */
1819 data |= (((uint64_t)4ULL) << 40);
1822 data = vcpu->arch.efer;
1824 case MSR_KVM_WALL_CLOCK:
1825 case MSR_KVM_WALL_CLOCK_NEW:
1826 data = vcpu->kvm->arch.wall_clock;
1828 case MSR_KVM_SYSTEM_TIME:
1829 case MSR_KVM_SYSTEM_TIME_NEW:
1830 data = vcpu->arch.time;
1832 case MSR_KVM_ASYNC_PF_EN:
1833 data = vcpu->arch.apf.msr_val;
1835 case MSR_IA32_P5_MC_ADDR:
1836 case MSR_IA32_P5_MC_TYPE:
1837 case MSR_IA32_MCG_CAP:
1838 case MSR_IA32_MCG_CTL:
1839 case MSR_IA32_MCG_STATUS:
1840 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1841 return get_msr_mce(vcpu, msr, pdata);
1842 case MSR_K7_CLK_CTL:
1844 * Provide expected ramp-up count for K7. All other
1845 * are set to zero, indicating minimum divisors for
1848 * This prevents guest kernels on AMD host with CPU
1849 * type 6, model 8 and higher from exploding due to
1850 * the rdmsr failing.
1854 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1855 if (kvm_hv_msr_partition_wide(msr)) {
1857 mutex_lock(&vcpu->kvm->lock);
1858 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1859 mutex_unlock(&vcpu->kvm->lock);
1862 return get_msr_hyperv(vcpu, msr, pdata);
1864 case MSR_IA32_BBL_CR_CTL3:
1865 /* This legacy MSR exists but isn't fully documented in current
1866 * silicon. It is however accessed by winxp in very narrow
1867 * scenarios where it sets bit #19, itself documented as
1868 * a "reserved" bit. Best effort attempt to source coherent
1869 * read data here should the balance of the register be
1870 * interpreted by the guest:
1872 * L2 cache control register 3: 64GB range, 256KB size,
1873 * enabled, latency 0x1, configured
1879 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1882 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1890 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1893 * Read or write a bunch of msrs. All parameters are kernel addresses.
1895 * @return number of msrs set successfully.
1897 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1898 struct kvm_msr_entry *entries,
1899 int (*do_msr)(struct kvm_vcpu *vcpu,
1900 unsigned index, u64 *data))
1904 idx = srcu_read_lock(&vcpu->kvm->srcu);
1905 for (i = 0; i < msrs->nmsrs; ++i)
1906 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1908 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1914 * Read or write a bunch of msrs. Parameters are user addresses.
1916 * @return number of msrs set successfully.
1918 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1919 int (*do_msr)(struct kvm_vcpu *vcpu,
1920 unsigned index, u64 *data),
1923 struct kvm_msrs msrs;
1924 struct kvm_msr_entry *entries;
1929 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1933 if (msrs.nmsrs >= MAX_IO_MSRS)
1937 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1938 entries = kmalloc(size, GFP_KERNEL);
1943 if (copy_from_user(entries, user_msrs->entries, size))
1946 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1951 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1962 int kvm_dev_ioctl_check_extension(long ext)
1967 case KVM_CAP_IRQCHIP:
1969 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1970 case KVM_CAP_SET_TSS_ADDR:
1971 case KVM_CAP_EXT_CPUID:
1972 case KVM_CAP_CLOCKSOURCE:
1974 case KVM_CAP_NOP_IO_DELAY:
1975 case KVM_CAP_MP_STATE:
1976 case KVM_CAP_SYNC_MMU:
1977 case KVM_CAP_USER_NMI:
1978 case KVM_CAP_REINJECT_CONTROL:
1979 case KVM_CAP_IRQ_INJECT_STATUS:
1980 case KVM_CAP_ASSIGN_DEV_IRQ:
1982 case KVM_CAP_IOEVENTFD:
1984 case KVM_CAP_PIT_STATE2:
1985 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1986 case KVM_CAP_XEN_HVM:
1987 case KVM_CAP_ADJUST_CLOCK:
1988 case KVM_CAP_VCPU_EVENTS:
1989 case KVM_CAP_HYPERV:
1990 case KVM_CAP_HYPERV_VAPIC:
1991 case KVM_CAP_HYPERV_SPIN:
1992 case KVM_CAP_PCI_SEGMENT:
1993 case KVM_CAP_DEBUGREGS:
1994 case KVM_CAP_X86_ROBUST_SINGLESTEP:
1996 case KVM_CAP_ASYNC_PF:
1997 case KVM_CAP_GET_TSC_KHZ:
2000 case KVM_CAP_COALESCED_MMIO:
2001 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2004 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2006 case KVM_CAP_NR_VCPUS:
2009 case KVM_CAP_NR_MEMSLOTS:
2010 r = KVM_MEMORY_SLOTS;
2012 case KVM_CAP_PV_MMU: /* obsolete */
2019 r = KVM_MAX_MCE_BANKS;
2024 case KVM_CAP_TSC_CONTROL:
2025 r = kvm_has_tsc_control;
2035 long kvm_arch_dev_ioctl(struct file *filp,
2036 unsigned int ioctl, unsigned long arg)
2038 void __user *argp = (void __user *)arg;
2042 case KVM_GET_MSR_INDEX_LIST: {
2043 struct kvm_msr_list __user *user_msr_list = argp;
2044 struct kvm_msr_list msr_list;
2048 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2051 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2052 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2055 if (n < msr_list.nmsrs)
2058 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2059 num_msrs_to_save * sizeof(u32)))
2061 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2063 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2068 case KVM_GET_SUPPORTED_CPUID: {
2069 struct kvm_cpuid2 __user *cpuid_arg = argp;
2070 struct kvm_cpuid2 cpuid;
2073 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2075 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2076 cpuid_arg->entries);
2081 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2086 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2089 mce_cap = KVM_MCE_CAP_SUPPORTED;
2091 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2103 static void wbinvd_ipi(void *garbage)
2108 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2110 return vcpu->kvm->arch.iommu_domain &&
2111 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2114 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2116 /* Address WBINVD may be executed by guest */
2117 if (need_emulate_wbinvd(vcpu)) {
2118 if (kvm_x86_ops->has_wbinvd_exit())
2119 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2120 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2121 smp_call_function_single(vcpu->cpu,
2122 wbinvd_ipi, NULL, 1);
2125 kvm_x86_ops->vcpu_load(vcpu, cpu);
2126 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2127 /* Make sure TSC doesn't go backwards */
2131 kvm_get_msr(vcpu, MSR_IA32_TSC, &tsc);
2132 tsc_delta = !vcpu->arch.last_guest_tsc ? 0 :
2133 tsc - vcpu->arch.last_guest_tsc;
2136 mark_tsc_unstable("KVM discovered backwards TSC");
2137 if (check_tsc_unstable()) {
2138 kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2139 vcpu->arch.tsc_catchup = 1;
2141 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2142 if (vcpu->cpu != cpu)
2143 kvm_migrate_timers(vcpu);
2148 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2150 kvm_x86_ops->vcpu_put(vcpu);
2151 kvm_put_guest_fpu(vcpu);
2152 kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
2155 static int is_efer_nx(void)
2157 unsigned long long efer = 0;
2159 rdmsrl_safe(MSR_EFER, &efer);
2160 return efer & EFER_NX;
2163 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2166 struct kvm_cpuid_entry2 *e, *entry;
2169 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2170 e = &vcpu->arch.cpuid_entries[i];
2171 if (e->function == 0x80000001) {
2176 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2177 entry->edx &= ~(1 << 20);
2178 printk(KERN_INFO "kvm: guest NX capability removed\n");
2182 /* when an old userspace process fills a new kernel module */
2183 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2184 struct kvm_cpuid *cpuid,
2185 struct kvm_cpuid_entry __user *entries)
2188 struct kvm_cpuid_entry *cpuid_entries;
2191 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2194 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2198 if (copy_from_user(cpuid_entries, entries,
2199 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2201 for (i = 0; i < cpuid->nent; i++) {
2202 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2203 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2204 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2205 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2206 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2207 vcpu->arch.cpuid_entries[i].index = 0;
2208 vcpu->arch.cpuid_entries[i].flags = 0;
2209 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2210 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2211 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2213 vcpu->arch.cpuid_nent = cpuid->nent;
2214 cpuid_fix_nx_cap(vcpu);
2216 kvm_apic_set_version(vcpu);
2217 kvm_x86_ops->cpuid_update(vcpu);
2221 vfree(cpuid_entries);
2226 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2227 struct kvm_cpuid2 *cpuid,
2228 struct kvm_cpuid_entry2 __user *entries)
2233 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2236 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2237 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2239 vcpu->arch.cpuid_nent = cpuid->nent;
2240 kvm_apic_set_version(vcpu);
2241 kvm_x86_ops->cpuid_update(vcpu);
2249 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2250 struct kvm_cpuid2 *cpuid,
2251 struct kvm_cpuid_entry2 __user *entries)
2256 if (cpuid->nent < vcpu->arch.cpuid_nent)
2259 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2260 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2265 cpuid->nent = vcpu->arch.cpuid_nent;
2269 static void cpuid_mask(u32 *word, int wordnum)
2271 *word &= boot_cpu_data.x86_capability[wordnum];
2274 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2277 entry->function = function;
2278 entry->index = index;
2279 cpuid_count(entry->function, entry->index,
2280 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2284 static bool supported_xcr0_bit(unsigned bit)
2286 u64 mask = ((u64)1 << bit);
2288 return mask & (XSTATE_FP | XSTATE_SSE | XSTATE_YMM) & host_xcr0;
2291 #define F(x) bit(X86_FEATURE_##x)
2293 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2294 u32 index, int *nent, int maxnent)
2296 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2297 #ifdef CONFIG_X86_64
2298 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2300 unsigned f_lm = F(LM);
2302 unsigned f_gbpages = 0;
2305 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2308 const u32 kvm_supported_word0_x86_features =
2309 F(FPU) | F(VME) | F(DE) | F(PSE) |
2310 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2311 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2312 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2313 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2314 0 /* Reserved, DS, ACPI */ | F(MMX) |
2315 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2316 0 /* HTT, TM, Reserved, PBE */;
2317 /* cpuid 0x80000001.edx */
2318 const u32 kvm_supported_word1_x86_features =
2319 F(FPU) | F(VME) | F(DE) | F(PSE) |
2320 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2321 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2322 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2323 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2324 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2325 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2326 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2328 const u32 kvm_supported_word4_x86_features =
2329 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
2330 0 /* DS-CPL, VMX, SMX, EST */ |
2331 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2332 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2333 0 /* Reserved, DCA */ | F(XMM4_1) |
2334 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2335 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2337 /* cpuid 0x80000001.ecx */
2338 const u32 kvm_supported_word6_x86_features =
2339 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2340 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2341 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2342 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2344 /* cpuid 0xC0000001.edx */
2345 const u32 kvm_supported_word5_x86_features =
2346 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
2347 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
2350 /* all calls to cpuid_count() should be made on the same cpu */
2352 do_cpuid_1_ent(entry, function, index);
2357 entry->eax = min(entry->eax, (u32)0xd);
2360 entry->edx &= kvm_supported_word0_x86_features;
2361 cpuid_mask(&entry->edx, 0);
2362 entry->ecx &= kvm_supported_word4_x86_features;
2363 cpuid_mask(&entry->ecx, 4);
2364 /* we support x2apic emulation even if host does not support
2365 * it since we emulate x2apic in software */
2366 entry->ecx |= F(X2APIC);
2368 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2369 * may return different values. This forces us to get_cpu() before
2370 * issuing the first command, and also to emulate this annoying behavior
2371 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2373 int t, times = entry->eax & 0xff;
2375 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2376 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2377 for (t = 1; t < times && *nent < maxnent; ++t) {
2378 do_cpuid_1_ent(&entry[t], function, 0);
2379 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2384 /* function 4 and 0xb have additional index. */
2388 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2389 /* read more entries until cache_type is zero */
2390 for (i = 1; *nent < maxnent; ++i) {
2391 cache_type = entry[i - 1].eax & 0x1f;
2394 do_cpuid_1_ent(&entry[i], function, i);
2396 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2406 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2407 /* read more entries until level_type is zero */
2408 for (i = 1; *nent < maxnent; ++i) {
2409 level_type = entry[i - 1].ecx & 0xff00;
2412 do_cpuid_1_ent(&entry[i], function, i);
2414 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2422 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2423 for (i = 1; *nent < maxnent && i < 64; ++i) {
2424 if (entry[i].eax == 0 || !supported_xcr0_bit(i))
2426 do_cpuid_1_ent(&entry[i], function, i);
2428 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2433 case KVM_CPUID_SIGNATURE: {
2434 char signature[12] = "KVMKVMKVM\0\0";
2435 u32 *sigptr = (u32 *)signature;
2437 entry->ebx = sigptr[0];
2438 entry->ecx = sigptr[1];
2439 entry->edx = sigptr[2];
2442 case KVM_CPUID_FEATURES:
2443 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2444 (1 << KVM_FEATURE_NOP_IO_DELAY) |
2445 (1 << KVM_FEATURE_CLOCKSOURCE2) |
2446 (1 << KVM_FEATURE_ASYNC_PF) |
2447 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2453 entry->eax = min(entry->eax, 0x8000001a);
2456 entry->edx &= kvm_supported_word1_x86_features;
2457 cpuid_mask(&entry->edx, 1);
2458 entry->ecx &= kvm_supported_word6_x86_features;
2459 cpuid_mask(&entry->ecx, 6);
2462 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
2463 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
2464 unsigned phys_as = entry->eax & 0xff;
2467 g_phys_as = phys_as;
2468 entry->eax = g_phys_as | (virt_as << 8);
2469 entry->ebx = entry->edx = 0;
2473 entry->ecx = entry->edx = 0;
2479 /*Add support for Centaur's CPUID instruction*/
2481 /*Just support up to 0xC0000004 now*/
2482 entry->eax = min(entry->eax, 0xC0000004);
2485 entry->edx &= kvm_supported_word5_x86_features;
2486 cpuid_mask(&entry->edx, 5);
2488 case 3: /* Processor serial number */
2489 case 5: /* MONITOR/MWAIT */
2490 case 6: /* Thermal management */
2491 case 0xA: /* Architectural Performance Monitoring */
2492 case 0x80000007: /* Advanced power management */
2497 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
2501 kvm_x86_ops->set_supported_cpuid(function, entry);
2508 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2509 struct kvm_cpuid_entry2 __user *entries)
2511 struct kvm_cpuid_entry2 *cpuid_entries;
2512 int limit, nent = 0, r = -E2BIG;
2515 if (cpuid->nent < 1)
2517 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2518 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2520 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2524 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2525 limit = cpuid_entries[0].eax;
2526 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2527 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2528 &nent, cpuid->nent);
2530 if (nent >= cpuid->nent)
2533 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2534 limit = cpuid_entries[nent - 1].eax;
2535 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2536 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2537 &nent, cpuid->nent);
2542 if (nent >= cpuid->nent)
2545 /* Add support for Centaur's CPUID instruction. */
2546 if (boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR) {
2547 do_cpuid_ent(&cpuid_entries[nent], 0xC0000000, 0,
2548 &nent, cpuid->nent);
2551 if (nent >= cpuid->nent)
2554 limit = cpuid_entries[nent - 1].eax;
2555 for (func = 0xC0000001;
2556 func <= limit && nent < cpuid->nent; ++func)
2557 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2558 &nent, cpuid->nent);
2561 if (nent >= cpuid->nent)
2565 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2569 if (nent >= cpuid->nent)
2572 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2576 if (nent >= cpuid->nent)
2580 if (copy_to_user(entries, cpuid_entries,
2581 nent * sizeof(struct kvm_cpuid_entry2)))
2587 vfree(cpuid_entries);
2592 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2593 struct kvm_lapic_state *s)
2595 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2600 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2601 struct kvm_lapic_state *s)
2603 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2604 kvm_apic_post_state_restore(vcpu);
2605 update_cr8_intercept(vcpu);
2610 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2611 struct kvm_interrupt *irq)
2613 if (irq->irq < 0 || irq->irq >= 256)
2615 if (irqchip_in_kernel(vcpu->kvm))
2618 kvm_queue_interrupt(vcpu, irq->irq, false);
2619 kvm_make_request(KVM_REQ_EVENT, vcpu);
2624 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2626 kvm_inject_nmi(vcpu);
2631 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2632 struct kvm_tpr_access_ctl *tac)
2636 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2640 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2644 unsigned bank_num = mcg_cap & 0xff, bank;
2647 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2649 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2652 vcpu->arch.mcg_cap = mcg_cap;
2653 /* Init IA32_MCG_CTL to all 1s */
2654 if (mcg_cap & MCG_CTL_P)
2655 vcpu->arch.mcg_ctl = ~(u64)0;
2656 /* Init IA32_MCi_CTL to all 1s */
2657 for (bank = 0; bank < bank_num; bank++)
2658 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2663 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2664 struct kvm_x86_mce *mce)
2666 u64 mcg_cap = vcpu->arch.mcg_cap;
2667 unsigned bank_num = mcg_cap & 0xff;
2668 u64 *banks = vcpu->arch.mce_banks;
2670 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2673 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2674 * reporting is disabled
2676 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2677 vcpu->arch.mcg_ctl != ~(u64)0)
2679 banks += 4 * mce->bank;
2681 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2682 * reporting is disabled for the bank
2684 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2686 if (mce->status & MCI_STATUS_UC) {
2687 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2688 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2689 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2692 if (banks[1] & MCI_STATUS_VAL)
2693 mce->status |= MCI_STATUS_OVER;
2694 banks[2] = mce->addr;
2695 banks[3] = mce->misc;
2696 vcpu->arch.mcg_status = mce->mcg_status;
2697 banks[1] = mce->status;
2698 kvm_queue_exception(vcpu, MC_VECTOR);
2699 } else if (!(banks[1] & MCI_STATUS_VAL)
2700 || !(banks[1] & MCI_STATUS_UC)) {
2701 if (banks[1] & MCI_STATUS_VAL)
2702 mce->status |= MCI_STATUS_OVER;
2703 banks[2] = mce->addr;
2704 banks[3] = mce->misc;
2705 banks[1] = mce->status;
2707 banks[1] |= MCI_STATUS_OVER;
2711 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2712 struct kvm_vcpu_events *events)
2714 events->exception.injected =
2715 vcpu->arch.exception.pending &&
2716 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2717 events->exception.nr = vcpu->arch.exception.nr;
2718 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2719 events->exception.pad = 0;
2720 events->exception.error_code = vcpu->arch.exception.error_code;
2722 events->interrupt.injected =
2723 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2724 events->interrupt.nr = vcpu->arch.interrupt.nr;
2725 events->interrupt.soft = 0;
2726 events->interrupt.shadow =
2727 kvm_x86_ops->get_interrupt_shadow(vcpu,
2728 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2730 events->nmi.injected = vcpu->arch.nmi_injected;
2731 events->nmi.pending = vcpu->arch.nmi_pending;
2732 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2733 events->nmi.pad = 0;
2735 events->sipi_vector = vcpu->arch.sipi_vector;
2737 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2738 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2739 | KVM_VCPUEVENT_VALID_SHADOW);
2740 memset(&events->reserved, 0, sizeof(events->reserved));
2743 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2744 struct kvm_vcpu_events *events)
2746 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2747 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2748 | KVM_VCPUEVENT_VALID_SHADOW))
2751 vcpu->arch.exception.pending = events->exception.injected;
2752 vcpu->arch.exception.nr = events->exception.nr;
2753 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2754 vcpu->arch.exception.error_code = events->exception.error_code;
2756 vcpu->arch.interrupt.pending = events->interrupt.injected;
2757 vcpu->arch.interrupt.nr = events->interrupt.nr;
2758 vcpu->arch.interrupt.soft = events->interrupt.soft;
2759 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2760 kvm_x86_ops->set_interrupt_shadow(vcpu,
2761 events->interrupt.shadow);
2763 vcpu->arch.nmi_injected = events->nmi.injected;
2764 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2765 vcpu->arch.nmi_pending = events->nmi.pending;
2766 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2768 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2769 vcpu->arch.sipi_vector = events->sipi_vector;
2771 kvm_make_request(KVM_REQ_EVENT, vcpu);
2776 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2777 struct kvm_debugregs *dbgregs)
2779 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2780 dbgregs->dr6 = vcpu->arch.dr6;
2781 dbgregs->dr7 = vcpu->arch.dr7;
2783 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2786 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2787 struct kvm_debugregs *dbgregs)
2792 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2793 vcpu->arch.dr6 = dbgregs->dr6;
2794 vcpu->arch.dr7 = dbgregs->dr7;
2799 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2800 struct kvm_xsave *guest_xsave)
2803 memcpy(guest_xsave->region,
2804 &vcpu->arch.guest_fpu.state->xsave,
2807 memcpy(guest_xsave->region,
2808 &vcpu->arch.guest_fpu.state->fxsave,
2809 sizeof(struct i387_fxsave_struct));
2810 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2815 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2816 struct kvm_xsave *guest_xsave)
2819 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2822 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2823 guest_xsave->region, xstate_size);
2825 if (xstate_bv & ~XSTATE_FPSSE)
2827 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2828 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2833 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2834 struct kvm_xcrs *guest_xcrs)
2836 if (!cpu_has_xsave) {
2837 guest_xcrs->nr_xcrs = 0;
2841 guest_xcrs->nr_xcrs = 1;
2842 guest_xcrs->flags = 0;
2843 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2844 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2847 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2848 struct kvm_xcrs *guest_xcrs)
2855 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2858 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2859 /* Only support XCR0 currently */
2860 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2861 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2862 guest_xcrs->xcrs[0].value);
2870 long kvm_arch_vcpu_ioctl(struct file *filp,
2871 unsigned int ioctl, unsigned long arg)
2873 struct kvm_vcpu *vcpu = filp->private_data;
2874 void __user *argp = (void __user *)arg;
2877 struct kvm_lapic_state *lapic;
2878 struct kvm_xsave *xsave;
2879 struct kvm_xcrs *xcrs;
2885 case KVM_GET_LAPIC: {
2887 if (!vcpu->arch.apic)
2889 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2894 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
2898 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
2903 case KVM_SET_LAPIC: {
2905 if (!vcpu->arch.apic)
2907 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2912 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
2914 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
2920 case KVM_INTERRUPT: {
2921 struct kvm_interrupt irq;
2924 if (copy_from_user(&irq, argp, sizeof irq))
2926 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2933 r = kvm_vcpu_ioctl_nmi(vcpu);
2939 case KVM_SET_CPUID: {
2940 struct kvm_cpuid __user *cpuid_arg = argp;
2941 struct kvm_cpuid cpuid;
2944 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2946 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2951 case KVM_SET_CPUID2: {
2952 struct kvm_cpuid2 __user *cpuid_arg = argp;
2953 struct kvm_cpuid2 cpuid;
2956 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2958 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2959 cpuid_arg->entries);
2964 case KVM_GET_CPUID2: {
2965 struct kvm_cpuid2 __user *cpuid_arg = argp;
2966 struct kvm_cpuid2 cpuid;
2969 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2971 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2972 cpuid_arg->entries);
2976 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2982 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2985 r = msr_io(vcpu, argp, do_set_msr, 0);
2987 case KVM_TPR_ACCESS_REPORTING: {
2988 struct kvm_tpr_access_ctl tac;
2991 if (copy_from_user(&tac, argp, sizeof tac))
2993 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2997 if (copy_to_user(argp, &tac, sizeof tac))
3002 case KVM_SET_VAPIC_ADDR: {
3003 struct kvm_vapic_addr va;
3006 if (!irqchip_in_kernel(vcpu->kvm))
3009 if (copy_from_user(&va, argp, sizeof va))
3012 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3015 case KVM_X86_SETUP_MCE: {
3019 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3021 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3024 case KVM_X86_SET_MCE: {
3025 struct kvm_x86_mce mce;
3028 if (copy_from_user(&mce, argp, sizeof mce))
3030 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3033 case KVM_GET_VCPU_EVENTS: {
3034 struct kvm_vcpu_events events;
3036 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3039 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3044 case KVM_SET_VCPU_EVENTS: {
3045 struct kvm_vcpu_events events;
3048 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3051 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3054 case KVM_GET_DEBUGREGS: {
3055 struct kvm_debugregs dbgregs;
3057 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3060 if (copy_to_user(argp, &dbgregs,
3061 sizeof(struct kvm_debugregs)))
3066 case KVM_SET_DEBUGREGS: {
3067 struct kvm_debugregs dbgregs;
3070 if (copy_from_user(&dbgregs, argp,
3071 sizeof(struct kvm_debugregs)))
3074 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3077 case KVM_GET_XSAVE: {
3078 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3083 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3086 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3091 case KVM_SET_XSAVE: {
3092 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3098 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
3101 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3104 case KVM_GET_XCRS: {
3105 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3110 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3113 if (copy_to_user(argp, u.xcrs,
3114 sizeof(struct kvm_xcrs)))
3119 case KVM_SET_XCRS: {
3120 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3126 if (copy_from_user(u.xcrs, argp,
3127 sizeof(struct kvm_xcrs)))
3130 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3133 case KVM_SET_TSC_KHZ: {
3137 if (!kvm_has_tsc_control)
3140 user_tsc_khz = (u32)arg;
3142 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3145 kvm_x86_ops->set_tsc_khz(vcpu, user_tsc_khz);
3150 case KVM_GET_TSC_KHZ: {
3152 if (check_tsc_unstable())
3155 r = vcpu_tsc_khz(vcpu);
3167 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3171 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3173 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3177 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3180 kvm->arch.ept_identity_map_addr = ident_addr;
3184 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3185 u32 kvm_nr_mmu_pages)
3187 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3190 mutex_lock(&kvm->slots_lock);
3191 spin_lock(&kvm->mmu_lock);
3193 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3194 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3196 spin_unlock(&kvm->mmu_lock);
3197 mutex_unlock(&kvm->slots_lock);
3201 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3203 return kvm->arch.n_max_mmu_pages;
3206 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3211 switch (chip->chip_id) {
3212 case KVM_IRQCHIP_PIC_MASTER:
3213 memcpy(&chip->chip.pic,
3214 &pic_irqchip(kvm)->pics[0],
3215 sizeof(struct kvm_pic_state));
3217 case KVM_IRQCHIP_PIC_SLAVE:
3218 memcpy(&chip->chip.pic,
3219 &pic_irqchip(kvm)->pics[1],
3220 sizeof(struct kvm_pic_state));
3222 case KVM_IRQCHIP_IOAPIC:
3223 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3232 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3237 switch (chip->chip_id) {
3238 case KVM_IRQCHIP_PIC_MASTER:
3239 spin_lock(&pic_irqchip(kvm)->lock);
3240 memcpy(&pic_irqchip(kvm)->pics[0],
3242 sizeof(struct kvm_pic_state));
3243 spin_unlock(&pic_irqchip(kvm)->lock);
3245 case KVM_IRQCHIP_PIC_SLAVE:
3246 spin_lock(&pic_irqchip(kvm)->lock);
3247 memcpy(&pic_irqchip(kvm)->pics[1],
3249 sizeof(struct kvm_pic_state));
3250 spin_unlock(&pic_irqchip(kvm)->lock);
3252 case KVM_IRQCHIP_IOAPIC:
3253 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3259 kvm_pic_update_irq(pic_irqchip(kvm));
3263 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3267 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3268 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3269 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3273 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3277 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3278 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3279 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3280 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3284 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3288 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3289 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3290 sizeof(ps->channels));
3291 ps->flags = kvm->arch.vpit->pit_state.flags;
3292 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3293 memset(&ps->reserved, 0, sizeof(ps->reserved));
3297 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3299 int r = 0, start = 0;
3300 u32 prev_legacy, cur_legacy;
3301 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3302 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3303 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3304 if (!prev_legacy && cur_legacy)
3306 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3307 sizeof(kvm->arch.vpit->pit_state.channels));
3308 kvm->arch.vpit->pit_state.flags = ps->flags;
3309 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3310 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3314 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3315 struct kvm_reinject_control *control)
3317 if (!kvm->arch.vpit)
3319 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3320 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3321 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3326 * Get (and clear) the dirty memory log for a memory slot.
3328 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3329 struct kvm_dirty_log *log)
3332 struct kvm_memory_slot *memslot;
3334 unsigned long is_dirty = 0;
3336 mutex_lock(&kvm->slots_lock);
3339 if (log->slot >= KVM_MEMORY_SLOTS)
3342 memslot = &kvm->memslots->memslots[log->slot];
3344 if (!memslot->dirty_bitmap)
3347 n = kvm_dirty_bitmap_bytes(memslot);
3349 for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3350 is_dirty = memslot->dirty_bitmap[i];
3352 /* If nothing is dirty, don't bother messing with page tables. */
3354 struct kvm_memslots *slots, *old_slots;
3355 unsigned long *dirty_bitmap;
3357 dirty_bitmap = memslot->dirty_bitmap_head;
3358 if (memslot->dirty_bitmap == dirty_bitmap)
3359 dirty_bitmap += n / sizeof(long);
3360 memset(dirty_bitmap, 0, n);
3363 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3366 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3367 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3368 slots->generation++;
3370 old_slots = kvm->memslots;
3371 rcu_assign_pointer(kvm->memslots, slots);
3372 synchronize_srcu_expedited(&kvm->srcu);
3373 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3376 spin_lock(&kvm->mmu_lock);
3377 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3378 spin_unlock(&kvm->mmu_lock);
3381 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
3385 if (clear_user(log->dirty_bitmap, n))
3391 mutex_unlock(&kvm->slots_lock);
3395 long kvm_arch_vm_ioctl(struct file *filp,
3396 unsigned int ioctl, unsigned long arg)
3398 struct kvm *kvm = filp->private_data;
3399 void __user *argp = (void __user *)arg;
3402 * This union makes it completely explicit to gcc-3.x
3403 * that these two variables' stack usage should be
3404 * combined, not added together.
3407 struct kvm_pit_state ps;
3408 struct kvm_pit_state2 ps2;
3409 struct kvm_pit_config pit_config;
3413 case KVM_SET_TSS_ADDR:
3414 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3418 case KVM_SET_IDENTITY_MAP_ADDR: {
3422 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3424 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3429 case KVM_SET_NR_MMU_PAGES:
3430 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3434 case KVM_GET_NR_MMU_PAGES:
3435 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3437 case KVM_CREATE_IRQCHIP: {
3438 struct kvm_pic *vpic;
3440 mutex_lock(&kvm->lock);
3443 goto create_irqchip_unlock;
3445 vpic = kvm_create_pic(kvm);
3447 r = kvm_ioapic_init(kvm);
3449 mutex_lock(&kvm->slots_lock);
3450 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3452 mutex_unlock(&kvm->slots_lock);
3454 goto create_irqchip_unlock;
3457 goto create_irqchip_unlock;
3459 kvm->arch.vpic = vpic;
3461 r = kvm_setup_default_irq_routing(kvm);
3463 mutex_lock(&kvm->slots_lock);
3464 mutex_lock(&kvm->irq_lock);
3465 kvm_ioapic_destroy(kvm);
3466 kvm_destroy_pic(kvm);
3467 mutex_unlock(&kvm->irq_lock);
3468 mutex_unlock(&kvm->slots_lock);
3470 create_irqchip_unlock:
3471 mutex_unlock(&kvm->lock);
3474 case KVM_CREATE_PIT:
3475 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3477 case KVM_CREATE_PIT2:
3479 if (copy_from_user(&u.pit_config, argp,
3480 sizeof(struct kvm_pit_config)))
3483 mutex_lock(&kvm->slots_lock);
3486 goto create_pit_unlock;
3488 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3492 mutex_unlock(&kvm->slots_lock);
3494 case KVM_IRQ_LINE_STATUS:
3495 case KVM_IRQ_LINE: {
3496 struct kvm_irq_level irq_event;
3499 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3502 if (irqchip_in_kernel(kvm)) {
3504 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3505 irq_event.irq, irq_event.level);
3506 if (ioctl == KVM_IRQ_LINE_STATUS) {
3508 irq_event.status = status;
3509 if (copy_to_user(argp, &irq_event,
3517 case KVM_GET_IRQCHIP: {
3518 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3519 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3525 if (copy_from_user(chip, argp, sizeof *chip))
3526 goto get_irqchip_out;
3528 if (!irqchip_in_kernel(kvm))
3529 goto get_irqchip_out;
3530 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3532 goto get_irqchip_out;
3534 if (copy_to_user(argp, chip, sizeof *chip))
3535 goto get_irqchip_out;
3543 case KVM_SET_IRQCHIP: {
3544 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3545 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3551 if (copy_from_user(chip, argp, sizeof *chip))
3552 goto set_irqchip_out;
3554 if (!irqchip_in_kernel(kvm))
3555 goto set_irqchip_out;
3556 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3558 goto set_irqchip_out;
3568 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3571 if (!kvm->arch.vpit)
3573 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3577 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3584 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3587 if (!kvm->arch.vpit)
3589 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3595 case KVM_GET_PIT2: {
3597 if (!kvm->arch.vpit)
3599 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3603 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3608 case KVM_SET_PIT2: {
3610 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3613 if (!kvm->arch.vpit)
3615 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3621 case KVM_REINJECT_CONTROL: {
3622 struct kvm_reinject_control control;
3624 if (copy_from_user(&control, argp, sizeof(control)))
3626 r = kvm_vm_ioctl_reinject(kvm, &control);
3632 case KVM_XEN_HVM_CONFIG: {
3634 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3635 sizeof(struct kvm_xen_hvm_config)))
3638 if (kvm->arch.xen_hvm_config.flags)
3643 case KVM_SET_CLOCK: {
3644 struct kvm_clock_data user_ns;
3649 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3657 local_irq_disable();
3658 now_ns = get_kernel_ns();
3659 delta = user_ns.clock - now_ns;
3661 kvm->arch.kvmclock_offset = delta;
3664 case KVM_GET_CLOCK: {
3665 struct kvm_clock_data user_ns;
3668 local_irq_disable();
3669 now_ns = get_kernel_ns();
3670 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3673 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3676 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3689 static void kvm_init_msr_list(void)
3694 /* skip the first msrs in the list. KVM-specific */
3695 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3696 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3699 msrs_to_save[j] = msrs_to_save[i];
3702 num_msrs_to_save = j;
3705 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3713 if (!(vcpu->arch.apic &&
3714 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3715 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3726 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3733 if (!(vcpu->arch.apic &&
3734 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3735 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3737 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3747 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3748 struct kvm_segment *var, int seg)
3750 kvm_x86_ops->set_segment(vcpu, var, seg);
3753 void kvm_get_segment(struct kvm_vcpu *vcpu,
3754 struct kvm_segment *var, int seg)
3756 kvm_x86_ops->get_segment(vcpu, var, seg);
3759 static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3764 static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3767 struct x86_exception exception;
3769 BUG_ON(!mmu_is_nested(vcpu));
3771 /* NPT walks are always user-walks */
3772 access |= PFERR_USER_MASK;
3773 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3778 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3779 struct x86_exception *exception)
3781 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3782 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3785 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3786 struct x86_exception *exception)
3788 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3789 access |= PFERR_FETCH_MASK;
3790 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3793 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3794 struct x86_exception *exception)
3796 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3797 access |= PFERR_WRITE_MASK;
3798 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3801 /* uses this to access any guest's mapped memory without checking CPL */
3802 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3803 struct x86_exception *exception)
3805 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3808 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3809 struct kvm_vcpu *vcpu, u32 access,
3810 struct x86_exception *exception)
3813 int r = X86EMUL_CONTINUE;
3816 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3818 unsigned offset = addr & (PAGE_SIZE-1);
3819 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3822 if (gpa == UNMAPPED_GVA)
3823 return X86EMUL_PROPAGATE_FAULT;
3824 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3826 r = X86EMUL_IO_NEEDED;
3838 /* used for instruction fetching */
3839 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3840 gva_t addr, void *val, unsigned int bytes,
3841 struct x86_exception *exception)
3843 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3844 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3846 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3847 access | PFERR_FETCH_MASK,
3851 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3852 gva_t addr, void *val, unsigned int bytes,
3853 struct x86_exception *exception)
3855 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3856 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3858 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3861 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
3863 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3864 gva_t addr, void *val, unsigned int bytes,
3865 struct x86_exception *exception)
3867 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3868 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3871 static int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3872 gva_t addr, void *val,
3874 struct x86_exception *exception)
3876 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3878 int r = X86EMUL_CONTINUE;
3881 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3884 unsigned offset = addr & (PAGE_SIZE-1);
3885 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3888 if (gpa == UNMAPPED_GVA)
3889 return X86EMUL_PROPAGATE_FAULT;
3890 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3892 r = X86EMUL_IO_NEEDED;
3904 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
3908 struct x86_exception *exception)
3910 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3914 if (vcpu->mmio_read_completed) {
3915 memcpy(val, vcpu->mmio_data, bytes);
3916 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3917 vcpu->mmio_phys_addr, *(u64 *)val);
3918 vcpu->mmio_read_completed = 0;
3919 return X86EMUL_CONTINUE;
3922 gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, exception);
3924 if (gpa == UNMAPPED_GVA)
3925 return X86EMUL_PROPAGATE_FAULT;
3927 /* For APIC access vmexit */
3928 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3931 if (kvm_read_guest_virt(ctxt, addr, val, bytes, exception)
3932 == X86EMUL_CONTINUE)
3933 return X86EMUL_CONTINUE;
3937 * Is this MMIO handled locally?
3939 handled = vcpu_mmio_read(vcpu, gpa, bytes, val);
3941 if (handled == bytes)
3942 return X86EMUL_CONTINUE;
3948 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3950 vcpu->mmio_needed = 1;
3951 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3952 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3953 vcpu->mmio_size = bytes;
3954 vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
3955 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 0;
3956 vcpu->mmio_index = 0;
3958 return X86EMUL_IO_NEEDED;
3961 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3962 const void *val, int bytes)
3966 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3969 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3973 static int emulator_write_emulated_onepage(unsigned long addr,
3976 struct x86_exception *exception,
3977 struct kvm_vcpu *vcpu)
3982 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, exception);
3984 if (gpa == UNMAPPED_GVA)
3985 return X86EMUL_PROPAGATE_FAULT;
3987 /* For APIC access vmexit */
3988 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3991 if (emulator_write_phys(vcpu, gpa, val, bytes))
3992 return X86EMUL_CONTINUE;
3995 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3997 * Is this MMIO handled locally?
3999 handled = vcpu_mmio_write(vcpu, gpa, bytes, val);
4000 if (handled == bytes)
4001 return X86EMUL_CONTINUE;
4007 vcpu->mmio_needed = 1;
4008 memcpy(vcpu->mmio_data, val, bytes);
4009 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4010 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
4011 vcpu->mmio_size = bytes;
4012 vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
4013 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 1;
4014 memcpy(vcpu->run->mmio.data, vcpu->mmio_data, 8);
4015 vcpu->mmio_index = 0;
4017 return X86EMUL_CONTINUE;
4020 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4024 struct x86_exception *exception)
4026 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4028 /* Crossing a page boundary? */
4029 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4032 now = -addr & ~PAGE_MASK;
4033 rc = emulator_write_emulated_onepage(addr, val, now, exception,
4035 if (rc != X86EMUL_CONTINUE)
4041 return emulator_write_emulated_onepage(addr, val, bytes, exception,
4045 #define CMPXCHG_TYPE(t, ptr, old, new) \
4046 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4048 #ifdef CONFIG_X86_64
4049 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4051 # define CMPXCHG64(ptr, old, new) \
4052 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4055 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4060 struct x86_exception *exception)
4062 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4068 /* guests cmpxchg8b have to be emulated atomically */
4069 if (bytes > 8 || (bytes & (bytes - 1)))
4072 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4074 if (gpa == UNMAPPED_GVA ||
4075 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4078 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4081 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4082 if (is_error_page(page)) {
4083 kvm_release_page_clean(page);
4087 kaddr = kmap_atomic(page, KM_USER0);
4088 kaddr += offset_in_page(gpa);
4091 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4094 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4097 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4100 exchanged = CMPXCHG64(kaddr, old, new);
4105 kunmap_atomic(kaddr, KM_USER0);
4106 kvm_release_page_dirty(page);
4109 return X86EMUL_CMPXCHG_FAILED;
4111 kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
4113 return X86EMUL_CONTINUE;
4116 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4118 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4121 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4123 /* TODO: String I/O for in kernel device */
4126 if (vcpu->arch.pio.in)
4127 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4128 vcpu->arch.pio.size, pd);
4130 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4131 vcpu->arch.pio.port, vcpu->arch.pio.size,
4137 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4138 int size, unsigned short port, void *val,
4141 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4143 if (vcpu->arch.pio.count)
4146 trace_kvm_pio(0, port, size, count);
4148 vcpu->arch.pio.port = port;
4149 vcpu->arch.pio.in = 1;
4150 vcpu->arch.pio.count = count;
4151 vcpu->arch.pio.size = size;
4153 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4155 memcpy(val, vcpu->arch.pio_data, size * count);
4156 vcpu->arch.pio.count = 0;
4160 vcpu->run->exit_reason = KVM_EXIT_IO;
4161 vcpu->run->io.direction = KVM_EXIT_IO_IN;
4162 vcpu->run->io.size = size;
4163 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4164 vcpu->run->io.count = count;
4165 vcpu->run->io.port = port;
4170 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4171 int size, unsigned short port,
4172 const void *val, unsigned int count)
4174 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4176 trace_kvm_pio(1, port, size, count);
4178 vcpu->arch.pio.port = port;
4179 vcpu->arch.pio.in = 0;
4180 vcpu->arch.pio.count = count;
4181 vcpu->arch.pio.size = size;
4183 memcpy(vcpu->arch.pio_data, val, size * count);
4185 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4186 vcpu->arch.pio.count = 0;
4190 vcpu->run->exit_reason = KVM_EXIT_IO;
4191 vcpu->run->io.direction = KVM_EXIT_IO_OUT;
4192 vcpu->run->io.size = size;
4193 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4194 vcpu->run->io.count = count;
4195 vcpu->run->io.port = port;
4200 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4202 return kvm_x86_ops->get_segment_base(vcpu, seg);
4205 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4207 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4210 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4212 if (!need_emulate_wbinvd(vcpu))
4213 return X86EMUL_CONTINUE;
4215 if (kvm_x86_ops->has_wbinvd_exit()) {
4216 int cpu = get_cpu();
4218 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4219 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4220 wbinvd_ipi, NULL, 1);
4222 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4225 return X86EMUL_CONTINUE;
4227 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4229 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4231 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4234 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4236 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4239 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4242 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4245 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4247 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4250 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4252 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4253 unsigned long value;
4257 value = kvm_read_cr0(vcpu);
4260 value = vcpu->arch.cr2;
4263 value = kvm_read_cr3(vcpu);
4266 value = kvm_read_cr4(vcpu);
4269 value = kvm_get_cr8(vcpu);
4272 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4279 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4281 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4286 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4289 vcpu->arch.cr2 = val;
4292 res = kvm_set_cr3(vcpu, val);
4295 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4298 res = kvm_set_cr8(vcpu, val);
4301 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4308 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4310 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4313 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4315 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4318 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4320 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4323 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4325 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4328 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4330 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4333 static unsigned long emulator_get_cached_segment_base(
4334 struct x86_emulate_ctxt *ctxt, int seg)
4336 return get_segment_base(emul_to_vcpu(ctxt), seg);
4339 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4340 struct desc_struct *desc, u32 *base3,
4343 struct kvm_segment var;
4345 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4346 *selector = var.selector;
4353 set_desc_limit(desc, var.limit);
4354 set_desc_base(desc, (unsigned long)var.base);
4355 #ifdef CONFIG_X86_64
4357 *base3 = var.base >> 32;
4359 desc->type = var.type;
4361 desc->dpl = var.dpl;
4362 desc->p = var.present;
4363 desc->avl = var.avl;
4371 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4372 struct desc_struct *desc, u32 base3,
4375 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4376 struct kvm_segment var;
4378 var.selector = selector;
4379 var.base = get_desc_base(desc);
4380 #ifdef CONFIG_X86_64
4381 var.base |= ((u64)base3) << 32;
4383 var.limit = get_desc_limit(desc);
4385 var.limit = (var.limit << 12) | 0xfff;
4386 var.type = desc->type;
4387 var.present = desc->p;
4388 var.dpl = desc->dpl;
4393 var.avl = desc->avl;
4394 var.present = desc->p;
4395 var.unusable = !var.present;
4398 kvm_set_segment(vcpu, &var, seg);
4402 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4403 u32 msr_index, u64 *pdata)
4405 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4408 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4409 u32 msr_index, u64 data)
4411 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
4414 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4416 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4419 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4422 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4424 * CR0.TS may reference the host fpu state, not the guest fpu state,
4425 * so it may be clear at this point.
4430 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4435 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4436 struct x86_instruction_info *info,
4437 enum x86_intercept_stage stage)
4439 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4442 static struct x86_emulate_ops emulate_ops = {
4443 .read_std = kvm_read_guest_virt_system,
4444 .write_std = kvm_write_guest_virt_system,
4445 .fetch = kvm_fetch_guest_virt,
4446 .read_emulated = emulator_read_emulated,
4447 .write_emulated = emulator_write_emulated,
4448 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4449 .invlpg = emulator_invlpg,
4450 .pio_in_emulated = emulator_pio_in_emulated,
4451 .pio_out_emulated = emulator_pio_out_emulated,
4452 .get_segment = emulator_get_segment,
4453 .set_segment = emulator_set_segment,
4454 .get_cached_segment_base = emulator_get_cached_segment_base,
4455 .get_gdt = emulator_get_gdt,
4456 .get_idt = emulator_get_idt,
4457 .set_gdt = emulator_set_gdt,
4458 .set_idt = emulator_set_idt,
4459 .get_cr = emulator_get_cr,
4460 .set_cr = emulator_set_cr,
4461 .cpl = emulator_get_cpl,
4462 .get_dr = emulator_get_dr,
4463 .set_dr = emulator_set_dr,
4464 .set_msr = emulator_set_msr,
4465 .get_msr = emulator_get_msr,
4466 .halt = emulator_halt,
4467 .wbinvd = emulator_wbinvd,
4468 .fix_hypercall = emulator_fix_hypercall,
4469 .get_fpu = emulator_get_fpu,
4470 .put_fpu = emulator_put_fpu,
4471 .intercept = emulator_intercept,
4474 static void cache_all_regs(struct kvm_vcpu *vcpu)
4476 kvm_register_read(vcpu, VCPU_REGS_RAX);
4477 kvm_register_read(vcpu, VCPU_REGS_RSP);
4478 kvm_register_read(vcpu, VCPU_REGS_RIP);
4479 vcpu->arch.regs_dirty = ~0;
4482 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4484 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4486 * an sti; sti; sequence only disable interrupts for the first
4487 * instruction. So, if the last instruction, be it emulated or
4488 * not, left the system with the INT_STI flag enabled, it
4489 * means that the last instruction is an sti. We should not
4490 * leave the flag on in this case. The same goes for mov ss
4492 if (!(int_shadow & mask))
4493 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4496 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4498 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4499 if (ctxt->exception.vector == PF_VECTOR)
4500 kvm_propagate_fault(vcpu, &ctxt->exception);
4501 else if (ctxt->exception.error_code_valid)
4502 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4503 ctxt->exception.error_code);
4505 kvm_queue_exception(vcpu, ctxt->exception.vector);
4508 static void init_decode_cache(struct decode_cache *c,
4509 const unsigned long *regs)
4511 memset(c, 0, offsetof(struct decode_cache, regs));
4512 memcpy(c->regs, regs, sizeof(c->regs));
4518 c->mem_read.pos = 0;
4519 c->mem_read.end = 0;
4522 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4524 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4525 struct decode_cache *c = &ctxt->decode;
4529 * TODO: fix emulate.c to use guest_read/write_register
4530 * instead of direct ->regs accesses, can save hundred cycles
4531 * on Intel for instructions that don't read/change RSP, for
4534 cache_all_regs(vcpu);
4536 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4538 ctxt->eflags = kvm_get_rflags(vcpu);
4539 ctxt->eip = kvm_rip_read(vcpu);
4540 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4541 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4542 cs_l ? X86EMUL_MODE_PROT64 :
4543 cs_db ? X86EMUL_MODE_PROT32 :
4544 X86EMUL_MODE_PROT16;
4545 ctxt->guest_mode = is_guest_mode(vcpu);
4547 init_decode_cache(c, vcpu->arch.regs);
4548 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4551 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4553 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4556 init_emulate_ctxt(vcpu);
4558 vcpu->arch.emulate_ctxt.decode.op_bytes = 2;
4559 vcpu->arch.emulate_ctxt.decode.ad_bytes = 2;
4560 vcpu->arch.emulate_ctxt.decode.eip = vcpu->arch.emulate_ctxt.eip +
4562 ret = emulate_int_real(&vcpu->arch.emulate_ctxt, irq);
4564 if (ret != X86EMUL_CONTINUE)
4565 return EMULATE_FAIL;
4567 vcpu->arch.emulate_ctxt.eip = c->eip;
4568 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4569 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4570 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4572 if (irq == NMI_VECTOR)
4573 vcpu->arch.nmi_pending = false;
4575 vcpu->arch.interrupt.pending = false;
4577 return EMULATE_DONE;
4579 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4581 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4583 int r = EMULATE_DONE;
4585 ++vcpu->stat.insn_emulation_fail;
4586 trace_kvm_emulate_insn_failed(vcpu);
4587 if (!is_guest_mode(vcpu)) {
4588 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4589 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4590 vcpu->run->internal.ndata = 0;
4593 kvm_queue_exception(vcpu, UD_VECTOR);
4598 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4606 * if emulation was due to access to shadowed page table
4607 * and it failed try to unshadow page and re-entetr the
4608 * guest to let CPU execute the instruction.
4610 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4613 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4615 if (gpa == UNMAPPED_GVA)
4616 return true; /* let cpu generate fault */
4618 if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4624 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4631 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4632 bool writeback = true;
4634 kvm_clear_exception_queue(vcpu);
4636 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4637 init_emulate_ctxt(vcpu);
4638 vcpu->arch.emulate_ctxt.interruptibility = 0;
4639 vcpu->arch.emulate_ctxt.have_exception = false;
4640 vcpu->arch.emulate_ctxt.perm_ok = false;
4642 vcpu->arch.emulate_ctxt.only_vendor_specific_insn
4643 = emulation_type & EMULTYPE_TRAP_UD;
4645 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, insn, insn_len);
4647 trace_kvm_emulate_insn_start(vcpu);
4648 ++vcpu->stat.insn_emulation;
4650 if (emulation_type & EMULTYPE_TRAP_UD)
4651 return EMULATE_FAIL;
4652 if (reexecute_instruction(vcpu, cr2))
4653 return EMULATE_DONE;
4654 if (emulation_type & EMULTYPE_SKIP)
4655 return EMULATE_FAIL;
4656 return handle_emulation_failure(vcpu);
4660 if (emulation_type & EMULTYPE_SKIP) {
4661 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
4662 return EMULATE_DONE;
4665 /* this is needed for vmware backdoor interface to work since it
4666 changes registers values during IO operation */
4667 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4668 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4669 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4673 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt);
4675 if (r == EMULATION_INTERCEPTED)
4676 return EMULATE_DONE;
4678 if (r == EMULATION_FAILED) {
4679 if (reexecute_instruction(vcpu, cr2))
4680 return EMULATE_DONE;
4682 return handle_emulation_failure(vcpu);
4685 if (vcpu->arch.emulate_ctxt.have_exception) {
4686 inject_emulated_exception(vcpu);
4688 } else if (vcpu->arch.pio.count) {
4689 if (!vcpu->arch.pio.in)
4690 vcpu->arch.pio.count = 0;
4693 r = EMULATE_DO_MMIO;
4694 } else if (vcpu->mmio_needed) {
4695 if (!vcpu->mmio_is_write)
4697 r = EMULATE_DO_MMIO;
4698 } else if (r == EMULATION_RESTART)
4704 toggle_interruptibility(vcpu,
4705 vcpu->arch.emulate_ctxt.interruptibility);
4706 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4707 kvm_make_request(KVM_REQ_EVENT, vcpu);
4708 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4709 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4710 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4712 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
4716 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4718 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4720 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4721 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4722 size, port, &val, 1);
4723 /* do not return to emulator after return from userspace */
4724 vcpu->arch.pio.count = 0;
4727 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4729 static void tsc_bad(void *info)
4731 __this_cpu_write(cpu_tsc_khz, 0);
4734 static void tsc_khz_changed(void *data)
4736 struct cpufreq_freqs *freq = data;
4737 unsigned long khz = 0;
4741 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4742 khz = cpufreq_quick_get(raw_smp_processor_id());
4745 __this_cpu_write(cpu_tsc_khz, khz);
4748 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4751 struct cpufreq_freqs *freq = data;
4753 struct kvm_vcpu *vcpu;
4754 int i, send_ipi = 0;
4757 * We allow guests to temporarily run on slowing clocks,
4758 * provided we notify them after, or to run on accelerating
4759 * clocks, provided we notify them before. Thus time never
4762 * However, we have a problem. We can't atomically update
4763 * the frequency of a given CPU from this function; it is
4764 * merely a notifier, which can be called from any CPU.
4765 * Changing the TSC frequency at arbitrary points in time
4766 * requires a recomputation of local variables related to
4767 * the TSC for each VCPU. We must flag these local variables
4768 * to be updated and be sure the update takes place with the
4769 * new frequency before any guests proceed.
4771 * Unfortunately, the combination of hotplug CPU and frequency
4772 * change creates an intractable locking scenario; the order
4773 * of when these callouts happen is undefined with respect to
4774 * CPU hotplug, and they can race with each other. As such,
4775 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4776 * undefined; you can actually have a CPU frequency change take
4777 * place in between the computation of X and the setting of the
4778 * variable. To protect against this problem, all updates of
4779 * the per_cpu tsc_khz variable are done in an interrupt
4780 * protected IPI, and all callers wishing to update the value
4781 * must wait for a synchronous IPI to complete (which is trivial
4782 * if the caller is on the CPU already). This establishes the
4783 * necessary total order on variable updates.
4785 * Note that because a guest time update may take place
4786 * anytime after the setting of the VCPU's request bit, the
4787 * correct TSC value must be set before the request. However,
4788 * to ensure the update actually makes it to any guest which
4789 * starts running in hardware virtualization between the set
4790 * and the acquisition of the spinlock, we must also ping the
4791 * CPU after setting the request bit.
4795 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4797 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4800 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4802 raw_spin_lock(&kvm_lock);
4803 list_for_each_entry(kvm, &vm_list, vm_list) {
4804 kvm_for_each_vcpu(i, vcpu, kvm) {
4805 if (vcpu->cpu != freq->cpu)
4807 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4808 if (vcpu->cpu != smp_processor_id())
4812 raw_spin_unlock(&kvm_lock);
4814 if (freq->old < freq->new && send_ipi) {
4816 * We upscale the frequency. Must make the guest
4817 * doesn't see old kvmclock values while running with
4818 * the new frequency, otherwise we risk the guest sees
4819 * time go backwards.
4821 * In case we update the frequency for another cpu
4822 * (which might be in guest context) send an interrupt
4823 * to kick the cpu out of guest context. Next time
4824 * guest context is entered kvmclock will be updated,
4825 * so the guest will not see stale values.
4827 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4832 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4833 .notifier_call = kvmclock_cpufreq_notifier
4836 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4837 unsigned long action, void *hcpu)
4839 unsigned int cpu = (unsigned long)hcpu;
4843 case CPU_DOWN_FAILED:
4844 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4846 case CPU_DOWN_PREPARE:
4847 smp_call_function_single(cpu, tsc_bad, NULL, 1);
4853 static struct notifier_block kvmclock_cpu_notifier_block = {
4854 .notifier_call = kvmclock_cpu_notifier,
4855 .priority = -INT_MAX
4858 static void kvm_timer_init(void)
4862 max_tsc_khz = tsc_khz;
4863 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4864 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4865 #ifdef CONFIG_CPU_FREQ
4866 struct cpufreq_policy policy;
4867 memset(&policy, 0, sizeof(policy));
4869 cpufreq_get_policy(&policy, cpu);
4870 if (policy.cpuinfo.max_freq)
4871 max_tsc_khz = policy.cpuinfo.max_freq;
4874 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4875 CPUFREQ_TRANSITION_NOTIFIER);
4877 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
4878 for_each_online_cpu(cpu)
4879 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4882 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4884 static int kvm_is_in_guest(void)
4886 return percpu_read(current_vcpu) != NULL;
4889 static int kvm_is_user_mode(void)
4893 if (percpu_read(current_vcpu))
4894 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
4896 return user_mode != 0;
4899 static unsigned long kvm_get_guest_ip(void)
4901 unsigned long ip = 0;
4903 if (percpu_read(current_vcpu))
4904 ip = kvm_rip_read(percpu_read(current_vcpu));
4909 static struct perf_guest_info_callbacks kvm_guest_cbs = {
4910 .is_in_guest = kvm_is_in_guest,
4911 .is_user_mode = kvm_is_user_mode,
4912 .get_guest_ip = kvm_get_guest_ip,
4915 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4917 percpu_write(current_vcpu, vcpu);
4919 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4921 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4923 percpu_write(current_vcpu, NULL);
4925 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4927 int kvm_arch_init(void *opaque)
4930 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4933 printk(KERN_ERR "kvm: already loaded the other module\n");
4938 if (!ops->cpu_has_kvm_support()) {
4939 printk(KERN_ERR "kvm: no hardware support\n");
4943 if (ops->disabled_by_bios()) {
4944 printk(KERN_ERR "kvm: disabled by bios\n");
4949 r = kvm_mmu_module_init();
4953 kvm_init_msr_list();
4956 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
4957 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4958 PT_DIRTY_MASK, PT64_NX_MASK, 0);
4962 perf_register_guest_info_callbacks(&kvm_guest_cbs);
4965 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
4973 void kvm_arch_exit(void)
4975 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4977 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4978 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4979 CPUFREQ_TRANSITION_NOTIFIER);
4980 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4982 kvm_mmu_module_exit();
4985 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4987 ++vcpu->stat.halt_exits;
4988 if (irqchip_in_kernel(vcpu->kvm)) {
4989 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
4992 vcpu->run->exit_reason = KVM_EXIT_HLT;
4996 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4998 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
5001 if (is_long_mode(vcpu))
5004 return a0 | ((gpa_t)a1 << 32);
5007 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5009 u64 param, ingpa, outgpa, ret;
5010 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5011 bool fast, longmode;
5015 * hypercall generates UD from non zero cpl and real mode
5018 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5019 kvm_queue_exception(vcpu, UD_VECTOR);
5023 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5024 longmode = is_long_mode(vcpu) && cs_l == 1;
5027 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5028 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5029 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5030 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5031 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5032 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5034 #ifdef CONFIG_X86_64
5036 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5037 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5038 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5042 code = param & 0xffff;
5043 fast = (param >> 16) & 0x1;
5044 rep_cnt = (param >> 32) & 0xfff;
5045 rep_idx = (param >> 48) & 0xfff;
5047 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5050 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5051 kvm_vcpu_on_spin(vcpu);
5054 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5058 ret = res | (((u64)rep_done & 0xfff) << 32);
5060 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5062 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5063 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5069 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5071 unsigned long nr, a0, a1, a2, a3, ret;
5074 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5075 return kvm_hv_hypercall(vcpu);
5077 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5078 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5079 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5080 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5081 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5083 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5085 if (!is_long_mode(vcpu)) {
5093 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5099 case KVM_HC_VAPIC_POLL_IRQ:
5103 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
5110 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5111 ++vcpu->stat.hypercalls;
5114 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5116 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5118 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5119 char instruction[3];
5120 unsigned long rip = kvm_rip_read(vcpu);
5123 * Blow out the MMU to ensure that no other VCPU has an active mapping
5124 * to ensure that the updated hypercall appears atomically across all
5127 kvm_mmu_zap_all(vcpu->kvm);
5129 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5131 return emulator_write_emulated(&vcpu->arch.emulate_ctxt,
5132 rip, instruction, 3, NULL);
5135 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
5137 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
5138 int j, nent = vcpu->arch.cpuid_nent;
5140 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
5141 /* when no next entry is found, the current entry[i] is reselected */
5142 for (j = i + 1; ; j = (j + 1) % nent) {
5143 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
5144 if (ej->function == e->function) {
5145 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
5149 return 0; /* silence gcc, even though control never reaches here */
5152 /* find an entry with matching function, matching index (if needed), and that
5153 * should be read next (if it's stateful) */
5154 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
5155 u32 function, u32 index)
5157 if (e->function != function)
5159 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
5161 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
5162 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
5167 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
5168 u32 function, u32 index)
5171 struct kvm_cpuid_entry2 *best = NULL;
5173 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
5174 struct kvm_cpuid_entry2 *e;
5176 e = &vcpu->arch.cpuid_entries[i];
5177 if (is_matching_cpuid_entry(e, function, index)) {
5178 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
5179 move_to_next_stateful_cpuid_entry(vcpu, i);
5186 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
5188 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
5190 struct kvm_cpuid_entry2 *best;
5192 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
5193 if (!best || best->eax < 0x80000008)
5195 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
5197 return best->eax & 0xff;
5203 * If no match is found, check whether we exceed the vCPU's limit
5204 * and return the content of the highest valid _standard_ leaf instead.
5205 * This is to satisfy the CPUID specification.
5207 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
5208 u32 function, u32 index)
5210 struct kvm_cpuid_entry2 *maxlevel;
5212 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
5213 if (!maxlevel || maxlevel->eax >= function)
5215 if (function & 0x80000000) {
5216 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
5220 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
5223 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
5225 u32 function, index;
5226 struct kvm_cpuid_entry2 *best;
5228 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
5229 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5230 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
5231 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
5232 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
5233 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
5234 best = kvm_find_cpuid_entry(vcpu, function, index);
5237 best = check_cpuid_limit(vcpu, function, index);
5240 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
5241 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
5242 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
5243 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
5245 kvm_x86_ops->skip_emulated_instruction(vcpu);
5246 trace_kvm_cpuid(function,
5247 kvm_register_read(vcpu, VCPU_REGS_RAX),
5248 kvm_register_read(vcpu, VCPU_REGS_RBX),
5249 kvm_register_read(vcpu, VCPU_REGS_RCX),
5250 kvm_register_read(vcpu, VCPU_REGS_RDX));
5252 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
5255 * Check if userspace requested an interrupt window, and that the
5256 * interrupt window is open.
5258 * No need to exit to userspace if we already have an interrupt queued.
5260 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5262 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5263 vcpu->run->request_interrupt_window &&
5264 kvm_arch_interrupt_allowed(vcpu));
5267 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5269 struct kvm_run *kvm_run = vcpu->run;
5271 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5272 kvm_run->cr8 = kvm_get_cr8(vcpu);
5273 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5274 if (irqchip_in_kernel(vcpu->kvm))
5275 kvm_run->ready_for_interrupt_injection = 1;
5277 kvm_run->ready_for_interrupt_injection =
5278 kvm_arch_interrupt_allowed(vcpu) &&
5279 !kvm_cpu_has_interrupt(vcpu) &&
5280 !kvm_event_needs_reinjection(vcpu);
5283 static void vapic_enter(struct kvm_vcpu *vcpu)
5285 struct kvm_lapic *apic = vcpu->arch.apic;
5288 if (!apic || !apic->vapic_addr)
5291 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5293 vcpu->arch.apic->vapic_page = page;
5296 static void vapic_exit(struct kvm_vcpu *vcpu)
5298 struct kvm_lapic *apic = vcpu->arch.apic;
5301 if (!apic || !apic->vapic_addr)
5304 idx = srcu_read_lock(&vcpu->kvm->srcu);
5305 kvm_release_page_dirty(apic->vapic_page);
5306 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5307 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5310 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5314 if (!kvm_x86_ops->update_cr8_intercept)
5317 if (!vcpu->arch.apic)
5320 if (!vcpu->arch.apic->vapic_addr)
5321 max_irr = kvm_lapic_find_highest_irr(vcpu);
5328 tpr = kvm_lapic_get_cr8(vcpu);
5330 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5333 static void inject_pending_event(struct kvm_vcpu *vcpu)
5335 /* try to reinject previous events if any */
5336 if (vcpu->arch.exception.pending) {
5337 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5338 vcpu->arch.exception.has_error_code,
5339 vcpu->arch.exception.error_code);
5340 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5341 vcpu->arch.exception.has_error_code,
5342 vcpu->arch.exception.error_code,
5343 vcpu->arch.exception.reinject);
5347 if (vcpu->arch.nmi_injected) {
5348 kvm_x86_ops->set_nmi(vcpu);
5352 if (vcpu->arch.interrupt.pending) {
5353 kvm_x86_ops->set_irq(vcpu);
5357 /* try to inject new event if pending */
5358 if (vcpu->arch.nmi_pending) {
5359 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5360 vcpu->arch.nmi_pending = false;
5361 vcpu->arch.nmi_injected = true;
5362 kvm_x86_ops->set_nmi(vcpu);
5364 } else if (kvm_cpu_has_interrupt(vcpu)) {
5365 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5366 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5368 kvm_x86_ops->set_irq(vcpu);
5373 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5375 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5376 !vcpu->guest_xcr0_loaded) {
5377 /* kvm_set_xcr() also depends on this */
5378 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5379 vcpu->guest_xcr0_loaded = 1;
5383 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5385 if (vcpu->guest_xcr0_loaded) {
5386 if (vcpu->arch.xcr0 != host_xcr0)
5387 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5388 vcpu->guest_xcr0_loaded = 0;
5392 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5396 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5397 vcpu->run->request_interrupt_window;
5399 if (vcpu->requests) {
5400 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5401 kvm_mmu_unload(vcpu);
5402 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5403 __kvm_migrate_timers(vcpu);
5404 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5405 r = kvm_guest_time_update(vcpu);
5409 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5410 kvm_mmu_sync_roots(vcpu);
5411 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5412 kvm_x86_ops->tlb_flush(vcpu);
5413 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5414 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5418 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5419 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5423 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5424 vcpu->fpu_active = 0;
5425 kvm_x86_ops->fpu_deactivate(vcpu);
5427 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5428 /* Page is swapped out. Do synthetic halt */
5429 vcpu->arch.apf.halted = true;
5435 r = kvm_mmu_reload(vcpu);
5440 * An NMI can be injected between local nmi_pending read and
5441 * vcpu->arch.nmi_pending read inside inject_pending_event().
5442 * But in that case, KVM_REQ_EVENT will be set, which makes
5443 * the race described above benign.
5445 nmi_pending = ACCESS_ONCE(vcpu->arch.nmi_pending);
5447 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5448 inject_pending_event(vcpu);
5450 /* enable NMI/IRQ window open exits if needed */
5452 kvm_x86_ops->enable_nmi_window(vcpu);
5453 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5454 kvm_x86_ops->enable_irq_window(vcpu);
5456 if (kvm_lapic_enabled(vcpu)) {
5457 update_cr8_intercept(vcpu);
5458 kvm_lapic_sync_to_vapic(vcpu);
5464 kvm_x86_ops->prepare_guest_switch(vcpu);
5465 if (vcpu->fpu_active)
5466 kvm_load_guest_fpu(vcpu);
5467 kvm_load_guest_xcr0(vcpu);
5469 vcpu->mode = IN_GUEST_MODE;
5471 /* We should set ->mode before check ->requests,
5472 * see the comment in make_all_cpus_request.
5476 local_irq_disable();
5478 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5479 || need_resched() || signal_pending(current)) {
5480 vcpu->mode = OUTSIDE_GUEST_MODE;
5484 kvm_x86_ops->cancel_injection(vcpu);
5489 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5493 if (unlikely(vcpu->arch.switch_db_regs)) {
5495 set_debugreg(vcpu->arch.eff_db[0], 0);
5496 set_debugreg(vcpu->arch.eff_db[1], 1);
5497 set_debugreg(vcpu->arch.eff_db[2], 2);
5498 set_debugreg(vcpu->arch.eff_db[3], 3);
5501 trace_kvm_entry(vcpu->vcpu_id);
5502 kvm_x86_ops->run(vcpu);
5505 * If the guest has used debug registers, at least dr7
5506 * will be disabled while returning to the host.
5507 * If we don't have active breakpoints in the host, we don't
5508 * care about the messed up debug address registers. But if
5509 * we have some of them active, restore the old state.
5511 if (hw_breakpoint_active())
5512 hw_breakpoint_restore();
5514 kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
5516 vcpu->mode = OUTSIDE_GUEST_MODE;
5523 * We must have an instruction between local_irq_enable() and
5524 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5525 * the interrupt shadow. The stat.exits increment will do nicely.
5526 * But we need to prevent reordering, hence this barrier():
5534 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5537 * Profile KVM exit RIPs:
5539 if (unlikely(prof_on == KVM_PROFILING)) {
5540 unsigned long rip = kvm_rip_read(vcpu);
5541 profile_hit(KVM_PROFILING, (void *)rip);
5545 kvm_lapic_sync_from_vapic(vcpu);
5547 r = kvm_x86_ops->handle_exit(vcpu);
5553 static int __vcpu_run(struct kvm_vcpu *vcpu)
5556 struct kvm *kvm = vcpu->kvm;
5558 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5559 pr_debug("vcpu %d received sipi with vector # %x\n",
5560 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5561 kvm_lapic_reset(vcpu);
5562 r = kvm_arch_vcpu_reset(vcpu);
5565 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5568 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5573 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5574 !vcpu->arch.apf.halted)
5575 r = vcpu_enter_guest(vcpu);
5577 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5578 kvm_vcpu_block(vcpu);
5579 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5580 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5582 switch(vcpu->arch.mp_state) {
5583 case KVM_MP_STATE_HALTED:
5584 vcpu->arch.mp_state =
5585 KVM_MP_STATE_RUNNABLE;
5586 case KVM_MP_STATE_RUNNABLE:
5587 vcpu->arch.apf.halted = false;
5589 case KVM_MP_STATE_SIPI_RECEIVED:
5600 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5601 if (kvm_cpu_has_pending_timer(vcpu))
5602 kvm_inject_pending_timer_irqs(vcpu);
5604 if (dm_request_for_irq_injection(vcpu)) {
5606 vcpu->run->exit_reason = KVM_EXIT_INTR;
5607 ++vcpu->stat.request_irq_exits;
5610 kvm_check_async_pf_completion(vcpu);
5612 if (signal_pending(current)) {
5614 vcpu->run->exit_reason = KVM_EXIT_INTR;
5615 ++vcpu->stat.signal_exits;
5617 if (need_resched()) {
5618 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5620 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5624 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5631 static int complete_mmio(struct kvm_vcpu *vcpu)
5633 struct kvm_run *run = vcpu->run;
5636 if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5639 if (vcpu->mmio_needed) {
5640 vcpu->mmio_needed = 0;
5641 if (!vcpu->mmio_is_write)
5642 memcpy(vcpu->mmio_data + vcpu->mmio_index,
5644 vcpu->mmio_index += 8;
5645 if (vcpu->mmio_index < vcpu->mmio_size) {
5646 run->exit_reason = KVM_EXIT_MMIO;
5647 run->mmio.phys_addr = vcpu->mmio_phys_addr + vcpu->mmio_index;
5648 memcpy(run->mmio.data, vcpu->mmio_data + vcpu->mmio_index, 8);
5649 run->mmio.len = min(vcpu->mmio_size - vcpu->mmio_index, 8);
5650 run->mmio.is_write = vcpu->mmio_is_write;
5651 vcpu->mmio_needed = 1;
5654 if (vcpu->mmio_is_write)
5656 vcpu->mmio_read_completed = 1;
5658 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5659 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5660 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5661 if (r != EMULATE_DONE)
5666 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5671 if (!tsk_used_math(current) && init_fpu(current))
5674 if (vcpu->sigset_active)
5675 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5677 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5678 kvm_vcpu_block(vcpu);
5679 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5684 /* re-sync apic's tpr */
5685 if (!irqchip_in_kernel(vcpu->kvm)) {
5686 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5692 r = complete_mmio(vcpu);
5696 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5697 kvm_register_write(vcpu, VCPU_REGS_RAX,
5698 kvm_run->hypercall.ret);
5700 r = __vcpu_run(vcpu);
5703 post_kvm_run_save(vcpu);
5704 if (vcpu->sigset_active)
5705 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5710 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5712 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5714 * We are here if userspace calls get_regs() in the middle of
5715 * instruction emulation. Registers state needs to be copied
5716 * back from emulation context to vcpu. Usrapace shouldn't do
5717 * that usually, but some bad designed PV devices (vmware
5718 * backdoor interface) need this to work
5720 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5721 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5722 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5724 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5725 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5726 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5727 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5728 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5729 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5730 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5731 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5732 #ifdef CONFIG_X86_64
5733 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5734 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5735 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5736 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5737 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5738 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5739 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5740 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5743 regs->rip = kvm_rip_read(vcpu);
5744 regs->rflags = kvm_get_rflags(vcpu);
5749 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5751 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
5752 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5754 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5755 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5756 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5757 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5758 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5759 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5760 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5761 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5762 #ifdef CONFIG_X86_64
5763 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5764 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5765 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5766 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5767 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5768 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5769 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5770 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5773 kvm_rip_write(vcpu, regs->rip);
5774 kvm_set_rflags(vcpu, regs->rflags);
5776 vcpu->arch.exception.pending = false;
5778 kvm_make_request(KVM_REQ_EVENT, vcpu);
5783 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5785 struct kvm_segment cs;
5787 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
5791 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5793 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5794 struct kvm_sregs *sregs)
5798 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5799 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5800 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5801 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5802 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5803 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5805 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5806 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5808 kvm_x86_ops->get_idt(vcpu, &dt);
5809 sregs->idt.limit = dt.size;
5810 sregs->idt.base = dt.address;
5811 kvm_x86_ops->get_gdt(vcpu, &dt);
5812 sregs->gdt.limit = dt.size;
5813 sregs->gdt.base = dt.address;
5815 sregs->cr0 = kvm_read_cr0(vcpu);
5816 sregs->cr2 = vcpu->arch.cr2;
5817 sregs->cr3 = kvm_read_cr3(vcpu);
5818 sregs->cr4 = kvm_read_cr4(vcpu);
5819 sregs->cr8 = kvm_get_cr8(vcpu);
5820 sregs->efer = vcpu->arch.efer;
5821 sregs->apic_base = kvm_get_apic_base(vcpu);
5823 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
5825 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
5826 set_bit(vcpu->arch.interrupt.nr,
5827 (unsigned long *)sregs->interrupt_bitmap);
5832 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
5833 struct kvm_mp_state *mp_state)
5835 mp_state->mp_state = vcpu->arch.mp_state;
5839 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
5840 struct kvm_mp_state *mp_state)
5842 vcpu->arch.mp_state = mp_state->mp_state;
5843 kvm_make_request(KVM_REQ_EVENT, vcpu);
5847 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
5848 bool has_error_code, u32 error_code)
5850 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5853 init_emulate_ctxt(vcpu);
5855 ret = emulator_task_switch(&vcpu->arch.emulate_ctxt,
5856 tss_selector, reason, has_error_code,
5860 return EMULATE_FAIL;
5862 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5863 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
5864 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
5865 kvm_make_request(KVM_REQ_EVENT, vcpu);
5866 return EMULATE_DONE;
5868 EXPORT_SYMBOL_GPL(kvm_task_switch);
5870 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5871 struct kvm_sregs *sregs)
5873 int mmu_reset_needed = 0;
5874 int pending_vec, max_bits, idx;
5877 dt.size = sregs->idt.limit;
5878 dt.address = sregs->idt.base;
5879 kvm_x86_ops->set_idt(vcpu, &dt);
5880 dt.size = sregs->gdt.limit;
5881 dt.address = sregs->gdt.base;
5882 kvm_x86_ops->set_gdt(vcpu, &dt);
5884 vcpu->arch.cr2 = sregs->cr2;
5885 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
5886 vcpu->arch.cr3 = sregs->cr3;
5887 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
5889 kvm_set_cr8(vcpu, sregs->cr8);
5891 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5892 kvm_x86_ops->set_efer(vcpu, sregs->efer);
5893 kvm_set_apic_base(vcpu, sregs->apic_base);
5895 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5896 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5897 vcpu->arch.cr0 = sregs->cr0;
5899 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5900 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5901 if (sregs->cr4 & X86_CR4_OSXSAVE)
5904 idx = srcu_read_lock(&vcpu->kvm->srcu);
5905 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5906 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
5907 mmu_reset_needed = 1;
5909 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5911 if (mmu_reset_needed)
5912 kvm_mmu_reset_context(vcpu);
5914 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5915 pending_vec = find_first_bit(
5916 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5917 if (pending_vec < max_bits) {
5918 kvm_queue_interrupt(vcpu, pending_vec, false);
5919 pr_debug("Set back pending irq %d\n", pending_vec);
5922 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5923 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5924 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5925 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5926 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5927 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5929 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5930 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5932 update_cr8_intercept(vcpu);
5934 /* Older userspace won't unhalt the vcpu on reset. */
5935 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5936 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5938 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5940 kvm_make_request(KVM_REQ_EVENT, vcpu);
5945 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5946 struct kvm_guest_debug *dbg)
5948 unsigned long rflags;
5951 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5953 if (vcpu->arch.exception.pending)
5955 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5956 kvm_queue_exception(vcpu, DB_VECTOR);
5958 kvm_queue_exception(vcpu, BP_VECTOR);
5962 * Read rflags as long as potentially injected trace flags are still
5965 rflags = kvm_get_rflags(vcpu);
5967 vcpu->guest_debug = dbg->control;
5968 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5969 vcpu->guest_debug = 0;
5971 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5972 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5973 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5974 vcpu->arch.switch_db_regs =
5975 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5977 for (i = 0; i < KVM_NR_DB_REGS; i++)
5978 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5979 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5982 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5983 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5984 get_segment_base(vcpu, VCPU_SREG_CS);
5987 * Trigger an rflags update that will inject or remove the trace
5990 kvm_set_rflags(vcpu, rflags);
5992 kvm_x86_ops->set_guest_debug(vcpu, dbg);
6002 * Translate a guest virtual address to a guest physical address.
6004 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6005 struct kvm_translation *tr)
6007 unsigned long vaddr = tr->linear_address;
6011 idx = srcu_read_lock(&vcpu->kvm->srcu);
6012 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6013 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6014 tr->physical_address = gpa;
6015 tr->valid = gpa != UNMAPPED_GVA;
6022 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6024 struct i387_fxsave_struct *fxsave =
6025 &vcpu->arch.guest_fpu.state->fxsave;
6027 memcpy(fpu->fpr, fxsave->st_space, 128);
6028 fpu->fcw = fxsave->cwd;
6029 fpu->fsw = fxsave->swd;
6030 fpu->ftwx = fxsave->twd;
6031 fpu->last_opcode = fxsave->fop;
6032 fpu->last_ip = fxsave->rip;
6033 fpu->last_dp = fxsave->rdp;
6034 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6039 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6041 struct i387_fxsave_struct *fxsave =
6042 &vcpu->arch.guest_fpu.state->fxsave;
6044 memcpy(fxsave->st_space, fpu->fpr, 128);
6045 fxsave->cwd = fpu->fcw;
6046 fxsave->swd = fpu->fsw;
6047 fxsave->twd = fpu->ftwx;
6048 fxsave->fop = fpu->last_opcode;
6049 fxsave->rip = fpu->last_ip;
6050 fxsave->rdp = fpu->last_dp;
6051 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6056 int fx_init(struct kvm_vcpu *vcpu)
6060 err = fpu_alloc(&vcpu->arch.guest_fpu);
6064 fpu_finit(&vcpu->arch.guest_fpu);
6067 * Ensure guest xcr0 is valid for loading
6069 vcpu->arch.xcr0 = XSTATE_FP;
6071 vcpu->arch.cr0 |= X86_CR0_ET;
6075 EXPORT_SYMBOL_GPL(fx_init);
6077 static void fx_free(struct kvm_vcpu *vcpu)
6079 fpu_free(&vcpu->arch.guest_fpu);
6082 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6084 if (vcpu->guest_fpu_loaded)
6088 * Restore all possible states in the guest,
6089 * and assume host would use all available bits.
6090 * Guest xcr0 would be loaded later.
6092 kvm_put_guest_xcr0(vcpu);
6093 vcpu->guest_fpu_loaded = 1;
6094 unlazy_fpu(current);
6095 fpu_restore_checking(&vcpu->arch.guest_fpu);
6099 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6101 kvm_put_guest_xcr0(vcpu);
6103 if (!vcpu->guest_fpu_loaded)
6106 vcpu->guest_fpu_loaded = 0;
6107 fpu_save_init(&vcpu->arch.guest_fpu);
6108 ++vcpu->stat.fpu_reload;
6109 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6113 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6115 kvmclock_reset(vcpu);
6117 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6119 kvm_x86_ops->vcpu_free(vcpu);
6122 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6125 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6126 printk_once(KERN_WARNING
6127 "kvm: SMP vm created on host with unstable TSC; "
6128 "guest TSC will not be reliable\n");
6129 return kvm_x86_ops->vcpu_create(kvm, id);
6132 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6136 vcpu->arch.mtrr_state.have_fixed = 1;
6138 r = kvm_arch_vcpu_reset(vcpu);
6140 r = kvm_mmu_setup(vcpu);
6146 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6148 vcpu->arch.apf.msr_val = 0;
6151 kvm_mmu_unload(vcpu);
6155 kvm_x86_ops->vcpu_free(vcpu);
6158 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
6160 vcpu->arch.nmi_pending = false;
6161 vcpu->arch.nmi_injected = false;
6163 vcpu->arch.switch_db_regs = 0;
6164 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6165 vcpu->arch.dr6 = DR6_FIXED_1;
6166 vcpu->arch.dr7 = DR7_FIXED_1;
6168 kvm_make_request(KVM_REQ_EVENT, vcpu);
6169 vcpu->arch.apf.msr_val = 0;
6171 kvmclock_reset(vcpu);
6173 kvm_clear_async_pf_completion_queue(vcpu);
6174 kvm_async_pf_hash_reset(vcpu);
6175 vcpu->arch.apf.halted = false;
6177 return kvm_x86_ops->vcpu_reset(vcpu);
6180 int kvm_arch_hardware_enable(void *garbage)
6183 struct kvm_vcpu *vcpu;
6186 kvm_shared_msr_cpu_online();
6187 list_for_each_entry(kvm, &vm_list, vm_list)
6188 kvm_for_each_vcpu(i, vcpu, kvm)
6189 if (vcpu->cpu == smp_processor_id())
6190 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6191 return kvm_x86_ops->hardware_enable(garbage);
6194 void kvm_arch_hardware_disable(void *garbage)
6196 kvm_x86_ops->hardware_disable(garbage);
6197 drop_user_return_notifiers(garbage);
6200 int kvm_arch_hardware_setup(void)
6202 return kvm_x86_ops->hardware_setup();
6205 void kvm_arch_hardware_unsetup(void)
6207 kvm_x86_ops->hardware_unsetup();
6210 void kvm_arch_check_processor_compat(void *rtn)
6212 kvm_x86_ops->check_processor_compatibility(rtn);
6215 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6221 BUG_ON(vcpu->kvm == NULL);
6224 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6225 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
6226 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
6227 vcpu->arch.mmu.translate_gpa = translate_gpa;
6228 vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
6229 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6230 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6232 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6234 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6239 vcpu->arch.pio_data = page_address(page);
6241 kvm_init_tsc_catchup(vcpu, max_tsc_khz);
6243 r = kvm_mmu_create(vcpu);
6245 goto fail_free_pio_data;
6247 if (irqchip_in_kernel(kvm)) {
6248 r = kvm_create_lapic(vcpu);
6250 goto fail_mmu_destroy;
6253 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6255 if (!vcpu->arch.mce_banks) {
6257 goto fail_free_lapic;
6259 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6261 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6262 goto fail_free_mce_banks;
6264 kvm_async_pf_hash_reset(vcpu);
6267 fail_free_mce_banks:
6268 kfree(vcpu->arch.mce_banks);
6270 kvm_free_lapic(vcpu);
6272 kvm_mmu_destroy(vcpu);
6274 free_page((unsigned long)vcpu->arch.pio_data);
6279 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6283 kfree(vcpu->arch.mce_banks);
6284 kvm_free_lapic(vcpu);
6285 idx = srcu_read_lock(&vcpu->kvm->srcu);
6286 kvm_mmu_destroy(vcpu);
6287 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6288 free_page((unsigned long)vcpu->arch.pio_data);
6291 int kvm_arch_init_vm(struct kvm *kvm)
6293 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6294 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6296 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6297 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6299 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6304 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6307 kvm_mmu_unload(vcpu);
6311 static void kvm_free_vcpus(struct kvm *kvm)
6314 struct kvm_vcpu *vcpu;
6317 * Unpin any mmu pages first.
6319 kvm_for_each_vcpu(i, vcpu, kvm) {
6320 kvm_clear_async_pf_completion_queue(vcpu);
6321 kvm_unload_vcpu_mmu(vcpu);
6323 kvm_for_each_vcpu(i, vcpu, kvm)
6324 kvm_arch_vcpu_free(vcpu);
6326 mutex_lock(&kvm->lock);
6327 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6328 kvm->vcpus[i] = NULL;
6330 atomic_set(&kvm->online_vcpus, 0);
6331 mutex_unlock(&kvm->lock);
6334 void kvm_arch_sync_events(struct kvm *kvm)
6336 kvm_free_all_assigned_devices(kvm);
6340 void kvm_arch_destroy_vm(struct kvm *kvm)
6342 kvm_iommu_unmap_guest(kvm);
6343 kfree(kvm->arch.vpic);
6344 kfree(kvm->arch.vioapic);
6345 kvm_free_vcpus(kvm);
6346 if (kvm->arch.apic_access_page)
6347 put_page(kvm->arch.apic_access_page);
6348 if (kvm->arch.ept_identity_pagetable)
6349 put_page(kvm->arch.ept_identity_pagetable);
6352 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6353 struct kvm_memory_slot *memslot,
6354 struct kvm_memory_slot old,
6355 struct kvm_userspace_memory_region *mem,
6358 int npages = memslot->npages;
6359 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6361 /* Prevent internal slot pages from being moved by fork()/COW. */
6362 if (memslot->id >= KVM_MEMORY_SLOTS)
6363 map_flags = MAP_SHARED | MAP_ANONYMOUS;
6365 /*To keep backward compatibility with older userspace,
6366 *x86 needs to hanlde !user_alloc case.
6369 if (npages && !old.rmap) {
6370 unsigned long userspace_addr;
6372 down_write(¤t->mm->mmap_sem);
6373 userspace_addr = do_mmap(NULL, 0,
6375 PROT_READ | PROT_WRITE,
6378 up_write(¤t->mm->mmap_sem);
6380 if (IS_ERR((void *)userspace_addr))
6381 return PTR_ERR((void *)userspace_addr);
6383 memslot->userspace_addr = userspace_addr;
6391 void kvm_arch_commit_memory_region(struct kvm *kvm,
6392 struct kvm_userspace_memory_region *mem,
6393 struct kvm_memory_slot old,
6397 int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
6399 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6402 down_write(¤t->mm->mmap_sem);
6403 ret = do_munmap(current->mm, old.userspace_addr,
6404 old.npages * PAGE_SIZE);
6405 up_write(¤t->mm->mmap_sem);
6408 "kvm_vm_ioctl_set_memory_region: "
6409 "failed to munmap memory\n");
6412 if (!kvm->arch.n_requested_mmu_pages)
6413 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6415 spin_lock(&kvm->mmu_lock);
6417 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6418 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6419 spin_unlock(&kvm->mmu_lock);
6422 void kvm_arch_flush_shadow(struct kvm *kvm)
6424 kvm_mmu_zap_all(kvm);
6425 kvm_reload_remote_mmus(kvm);
6428 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6430 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6431 !vcpu->arch.apf.halted)
6432 || !list_empty_careful(&vcpu->async_pf.done)
6433 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6434 || vcpu->arch.nmi_pending ||
6435 (kvm_arch_interrupt_allowed(vcpu) &&
6436 kvm_cpu_has_interrupt(vcpu));
6439 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6442 int cpu = vcpu->cpu;
6444 if (waitqueue_active(&vcpu->wq)) {
6445 wake_up_interruptible(&vcpu->wq);
6446 ++vcpu->stat.halt_wakeup;
6450 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6451 if (kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE)
6452 smp_send_reschedule(cpu);
6456 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6458 return kvm_x86_ops->interrupt_allowed(vcpu);
6461 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6463 unsigned long current_rip = kvm_rip_read(vcpu) +
6464 get_segment_base(vcpu, VCPU_SREG_CS);
6466 return current_rip == linear_rip;
6468 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6470 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6472 unsigned long rflags;
6474 rflags = kvm_x86_ops->get_rflags(vcpu);
6475 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6476 rflags &= ~X86_EFLAGS_TF;
6479 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6481 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6483 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6484 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6485 rflags |= X86_EFLAGS_TF;
6486 kvm_x86_ops->set_rflags(vcpu, rflags);
6487 kvm_make_request(KVM_REQ_EVENT, vcpu);
6489 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6491 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6495 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6496 is_error_page(work->page))
6499 r = kvm_mmu_reload(vcpu);
6503 if (!vcpu->arch.mmu.direct_map &&
6504 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6507 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6510 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6512 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6515 static inline u32 kvm_async_pf_next_probe(u32 key)
6517 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6520 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6522 u32 key = kvm_async_pf_hash_fn(gfn);
6524 while (vcpu->arch.apf.gfns[key] != ~0)
6525 key = kvm_async_pf_next_probe(key);
6527 vcpu->arch.apf.gfns[key] = gfn;
6530 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6533 u32 key = kvm_async_pf_hash_fn(gfn);
6535 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6536 (vcpu->arch.apf.gfns[key] != gfn &&
6537 vcpu->arch.apf.gfns[key] != ~0); i++)
6538 key = kvm_async_pf_next_probe(key);
6543 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6545 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6548 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6552 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6554 vcpu->arch.apf.gfns[i] = ~0;
6556 j = kvm_async_pf_next_probe(j);
6557 if (vcpu->arch.apf.gfns[j] == ~0)
6559 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6561 * k lies cyclically in ]i,j]
6563 * |....j i.k.| or |.k..j i...|
6565 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6566 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6571 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6574 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6578 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6579 struct kvm_async_pf *work)
6581 struct x86_exception fault;
6583 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6584 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6586 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6587 (vcpu->arch.apf.send_user_only &&
6588 kvm_x86_ops->get_cpl(vcpu) == 0))
6589 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6590 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6591 fault.vector = PF_VECTOR;
6592 fault.error_code_valid = true;
6593 fault.error_code = 0;
6594 fault.nested_page_fault = false;
6595 fault.address = work->arch.token;
6596 kvm_inject_page_fault(vcpu, &fault);
6600 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6601 struct kvm_async_pf *work)
6603 struct x86_exception fault;
6605 trace_kvm_async_pf_ready(work->arch.token, work->gva);
6606 if (is_error_page(work->page))
6607 work->arch.token = ~0; /* broadcast wakeup */
6609 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6611 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6612 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6613 fault.vector = PF_VECTOR;
6614 fault.error_code_valid = true;
6615 fault.error_code = 0;
6616 fault.nested_page_fault = false;
6617 fault.address = work->arch.token;
6618 kvm_inject_page_fault(vcpu, &fault);
6620 vcpu->arch.apf.halted = false;
6623 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6625 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6628 return !kvm_event_needs_reinjection(vcpu) &&
6629 kvm_x86_ops->interrupt_allowed(vcpu);
6632 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6633 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6634 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6635 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6636 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6637 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6638 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6639 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6640 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6641 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6642 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6643 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);