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);
350 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
352 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
354 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
355 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
357 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
360 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
362 kvm_make_request(KVM_REQ_EVENT, vcpu);
363 vcpu->arch.nmi_pending = 1;
365 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
367 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
369 kvm_multiple_exception(vcpu, nr, true, error_code, false);
371 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
373 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
375 kvm_multiple_exception(vcpu, nr, true, error_code, true);
377 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
380 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
381 * a #GP and return false.
383 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
385 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
387 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
390 EXPORT_SYMBOL_GPL(kvm_require_cpl);
393 * This function will be used to read from the physical memory of the currently
394 * running guest. The difference to kvm_read_guest_page is that this function
395 * can read from guest physical or from the guest's guest physical memory.
397 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
398 gfn_t ngfn, void *data, int offset, int len,
404 ngpa = gfn_to_gpa(ngfn);
405 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
406 if (real_gfn == UNMAPPED_GVA)
409 real_gfn = gpa_to_gfn(real_gfn);
411 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
413 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
415 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
416 void *data, int offset, int len, u32 access)
418 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
419 data, offset, len, access);
423 * Load the pae pdptrs. Return true is they are all valid.
425 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
427 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
428 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
431 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
433 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
434 offset * sizeof(u64), sizeof(pdpte),
435 PFERR_USER_MASK|PFERR_WRITE_MASK);
440 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
441 if (is_present_gpte(pdpte[i]) &&
442 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
449 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
450 __set_bit(VCPU_EXREG_PDPTR,
451 (unsigned long *)&vcpu->arch.regs_avail);
452 __set_bit(VCPU_EXREG_PDPTR,
453 (unsigned long *)&vcpu->arch.regs_dirty);
458 EXPORT_SYMBOL_GPL(load_pdptrs);
460 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
462 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
468 if (is_long_mode(vcpu) || !is_pae(vcpu))
471 if (!test_bit(VCPU_EXREG_PDPTR,
472 (unsigned long *)&vcpu->arch.regs_avail))
475 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
476 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
477 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
478 PFERR_USER_MASK | PFERR_WRITE_MASK);
481 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
487 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
489 unsigned long old_cr0 = kvm_read_cr0(vcpu);
490 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
491 X86_CR0_CD | X86_CR0_NW;
496 if (cr0 & 0xffffffff00000000UL)
500 cr0 &= ~CR0_RESERVED_BITS;
502 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
505 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
508 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
510 if ((vcpu->arch.efer & EFER_LME)) {
515 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
520 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
525 kvm_x86_ops->set_cr0(vcpu, cr0);
527 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
528 kvm_clear_async_pf_completion_queue(vcpu);
529 kvm_async_pf_hash_reset(vcpu);
532 if ((cr0 ^ old_cr0) & update_bits)
533 kvm_mmu_reset_context(vcpu);
536 EXPORT_SYMBOL_GPL(kvm_set_cr0);
538 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
540 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
542 EXPORT_SYMBOL_GPL(kvm_lmsw);
544 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
548 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
549 if (index != XCR_XFEATURE_ENABLED_MASK)
552 if (kvm_x86_ops->get_cpl(vcpu) != 0)
554 if (!(xcr0 & XSTATE_FP))
556 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
558 if (xcr0 & ~host_xcr0)
560 vcpu->arch.xcr0 = xcr0;
561 vcpu->guest_xcr0_loaded = 0;
565 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
567 if (__kvm_set_xcr(vcpu, index, xcr)) {
568 kvm_inject_gp(vcpu, 0);
573 EXPORT_SYMBOL_GPL(kvm_set_xcr);
575 static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
577 struct kvm_cpuid_entry2 *best;
579 best = kvm_find_cpuid_entry(vcpu, 1, 0);
580 return best && (best->ecx & bit(X86_FEATURE_XSAVE));
583 static bool guest_cpuid_has_smep(struct kvm_vcpu *vcpu)
585 struct kvm_cpuid_entry2 *best;
587 best = kvm_find_cpuid_entry(vcpu, 7, 0);
588 return best && (best->ebx & bit(X86_FEATURE_SMEP));
591 static bool guest_cpuid_has_fsgsbase(struct kvm_vcpu *vcpu)
593 struct kvm_cpuid_entry2 *best;
595 best = kvm_find_cpuid_entry(vcpu, 7, 0);
596 return best && (best->ebx & bit(X86_FEATURE_FSGSBASE));
599 static void update_cpuid(struct kvm_vcpu *vcpu)
601 struct kvm_cpuid_entry2 *best;
603 best = kvm_find_cpuid_entry(vcpu, 1, 0);
607 /* Update OSXSAVE bit */
608 if (cpu_has_xsave && best->function == 0x1) {
609 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
610 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
611 best->ecx |= bit(X86_FEATURE_OSXSAVE);
615 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
617 unsigned long old_cr4 = kvm_read_cr4(vcpu);
618 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
619 X86_CR4_PAE | X86_CR4_SMEP;
620 if (cr4 & CR4_RESERVED_BITS)
623 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
626 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
629 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_RDWRGSFS))
632 if (is_long_mode(vcpu)) {
633 if (!(cr4 & X86_CR4_PAE))
635 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
636 && ((cr4 ^ old_cr4) & pdptr_bits)
637 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
641 if (kvm_x86_ops->set_cr4(vcpu, cr4))
644 if ((cr4 ^ old_cr4) & pdptr_bits)
645 kvm_mmu_reset_context(vcpu);
647 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
652 EXPORT_SYMBOL_GPL(kvm_set_cr4);
654 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
656 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
657 kvm_mmu_sync_roots(vcpu);
658 kvm_mmu_flush_tlb(vcpu);
662 if (is_long_mode(vcpu)) {
663 if (cr3 & CR3_L_MODE_RESERVED_BITS)
667 if (cr3 & CR3_PAE_RESERVED_BITS)
669 if (is_paging(vcpu) &&
670 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
674 * We don't check reserved bits in nonpae mode, because
675 * this isn't enforced, and VMware depends on this.
680 * Does the new cr3 value map to physical memory? (Note, we
681 * catch an invalid cr3 even in real-mode, because it would
682 * cause trouble later on when we turn on paging anyway.)
684 * A real CPU would silently accept an invalid cr3 and would
685 * attempt to use it - with largely undefined (and often hard
686 * to debug) behavior on the guest side.
688 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
690 vcpu->arch.cr3 = cr3;
691 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
692 vcpu->arch.mmu.new_cr3(vcpu);
695 EXPORT_SYMBOL_GPL(kvm_set_cr3);
697 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
699 if (cr8 & CR8_RESERVED_BITS)
701 if (irqchip_in_kernel(vcpu->kvm))
702 kvm_lapic_set_tpr(vcpu, cr8);
704 vcpu->arch.cr8 = cr8;
707 EXPORT_SYMBOL_GPL(kvm_set_cr8);
709 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
711 if (irqchip_in_kernel(vcpu->kvm))
712 return kvm_lapic_get_cr8(vcpu);
714 return vcpu->arch.cr8;
716 EXPORT_SYMBOL_GPL(kvm_get_cr8);
718 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
722 vcpu->arch.db[dr] = val;
723 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
724 vcpu->arch.eff_db[dr] = val;
727 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
731 if (val & 0xffffffff00000000ULL)
733 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
736 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
740 if (val & 0xffffffff00000000ULL)
742 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
743 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
744 kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
745 vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
753 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
757 res = __kvm_set_dr(vcpu, dr, val);
759 kvm_queue_exception(vcpu, UD_VECTOR);
761 kvm_inject_gp(vcpu, 0);
765 EXPORT_SYMBOL_GPL(kvm_set_dr);
767 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
771 *val = vcpu->arch.db[dr];
774 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
778 *val = vcpu->arch.dr6;
781 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
785 *val = vcpu->arch.dr7;
792 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
794 if (_kvm_get_dr(vcpu, dr, val)) {
795 kvm_queue_exception(vcpu, UD_VECTOR);
800 EXPORT_SYMBOL_GPL(kvm_get_dr);
803 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
804 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
806 * This list is modified at module load time to reflect the
807 * capabilities of the host cpu. This capabilities test skips MSRs that are
808 * kvm-specific. Those are put in the beginning of the list.
811 #define KVM_SAVE_MSRS_BEGIN 9
812 static u32 msrs_to_save[] = {
813 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
814 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
815 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
816 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
817 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
820 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
822 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
825 static unsigned num_msrs_to_save;
827 static u32 emulated_msrs[] = {
828 MSR_IA32_MISC_ENABLE,
833 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
835 u64 old_efer = vcpu->arch.efer;
837 if (efer & efer_reserved_bits)
841 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
844 if (efer & EFER_FFXSR) {
845 struct kvm_cpuid_entry2 *feat;
847 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
848 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
852 if (efer & EFER_SVME) {
853 struct kvm_cpuid_entry2 *feat;
855 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
856 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
861 efer |= vcpu->arch.efer & EFER_LMA;
863 kvm_x86_ops->set_efer(vcpu, efer);
865 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
867 /* Update reserved bits */
868 if ((efer ^ old_efer) & EFER_NX)
869 kvm_mmu_reset_context(vcpu);
874 void kvm_enable_efer_bits(u64 mask)
876 efer_reserved_bits &= ~mask;
878 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
882 * Writes msr value into into the appropriate "register".
883 * Returns 0 on success, non-0 otherwise.
884 * Assumes vcpu_load() was already called.
886 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
888 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
892 * Adapt set_msr() to msr_io()'s calling convention
894 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
896 return kvm_set_msr(vcpu, index, *data);
899 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
903 struct pvclock_wall_clock wc;
904 struct timespec boot;
909 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
914 ++version; /* first time write, random junk */
918 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
921 * The guest calculates current wall clock time by adding
922 * system time (updated by kvm_guest_time_update below) to the
923 * wall clock specified here. guest system time equals host
924 * system time for us, thus we must fill in host boot time here.
928 wc.sec = boot.tv_sec;
929 wc.nsec = boot.tv_nsec;
930 wc.version = version;
932 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
935 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
938 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
940 uint32_t quotient, remainder;
942 /* Don't try to replace with do_div(), this one calculates
943 * "(dividend << 32) / divisor" */
945 : "=a" (quotient), "=d" (remainder)
946 : "0" (0), "1" (dividend), "r" (divisor) );
950 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
951 s8 *pshift, u32 *pmultiplier)
958 tps64 = base_khz * 1000LL;
959 scaled64 = scaled_khz * 1000LL;
960 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
965 tps32 = (uint32_t)tps64;
966 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
967 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
975 *pmultiplier = div_frac(scaled64, tps32);
977 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
978 __func__, base_khz, scaled_khz, shift, *pmultiplier);
981 static inline u64 get_kernel_ns(void)
985 WARN_ON(preemptible());
987 monotonic_to_bootbased(&ts);
988 return timespec_to_ns(&ts);
991 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
992 unsigned long max_tsc_khz;
994 static inline int kvm_tsc_changes_freq(void)
997 int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
998 cpufreq_quick_get(cpu) != 0;
1003 static u64 vcpu_tsc_khz(struct kvm_vcpu *vcpu)
1005 if (vcpu->arch.virtual_tsc_khz)
1006 return vcpu->arch.virtual_tsc_khz;
1008 return __this_cpu_read(cpu_tsc_khz);
1011 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1015 WARN_ON(preemptible());
1016 if (kvm_tsc_changes_freq())
1017 printk_once(KERN_WARNING
1018 "kvm: unreliable cycle conversion on adjustable rate TSC\n");
1019 ret = nsec * vcpu_tsc_khz(vcpu);
1020 do_div(ret, USEC_PER_SEC);
1024 static void kvm_init_tsc_catchup(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1026 /* Compute a scale to convert nanoseconds in TSC cycles */
1027 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1028 &vcpu->arch.tsc_catchup_shift,
1029 &vcpu->arch.tsc_catchup_mult);
1032 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1034 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1035 vcpu->arch.tsc_catchup_mult,
1036 vcpu->arch.tsc_catchup_shift);
1037 tsc += vcpu->arch.last_tsc_write;
1041 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1043 struct kvm *kvm = vcpu->kvm;
1044 u64 offset, ns, elapsed;
1045 unsigned long flags;
1048 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1049 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1050 ns = get_kernel_ns();
1051 elapsed = ns - kvm->arch.last_tsc_nsec;
1052 sdiff = data - kvm->arch.last_tsc_write;
1057 * Special case: close write to TSC within 5 seconds of
1058 * another CPU is interpreted as an attempt to synchronize
1059 * The 5 seconds is to accommodate host load / swapping as
1060 * well as any reset of TSC during the boot process.
1062 * In that case, for a reliable TSC, we can match TSC offsets,
1063 * or make a best guest using elapsed value.
1065 if (sdiff < nsec_to_cycles(vcpu, 5ULL * NSEC_PER_SEC) &&
1066 elapsed < 5ULL * NSEC_PER_SEC) {
1067 if (!check_tsc_unstable()) {
1068 offset = kvm->arch.last_tsc_offset;
1069 pr_debug("kvm: matched tsc offset for %llu\n", data);
1071 u64 delta = nsec_to_cycles(vcpu, elapsed);
1073 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1075 ns = kvm->arch.last_tsc_nsec;
1077 kvm->arch.last_tsc_nsec = ns;
1078 kvm->arch.last_tsc_write = data;
1079 kvm->arch.last_tsc_offset = offset;
1080 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1081 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1083 /* Reset of TSC must disable overshoot protection below */
1084 vcpu->arch.hv_clock.tsc_timestamp = 0;
1085 vcpu->arch.last_tsc_write = data;
1086 vcpu->arch.last_tsc_nsec = ns;
1088 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1090 static int kvm_guest_time_update(struct kvm_vcpu *v)
1092 unsigned long flags;
1093 struct kvm_vcpu_arch *vcpu = &v->arch;
1095 unsigned long this_tsc_khz;
1096 s64 kernel_ns, max_kernel_ns;
1099 /* Keep irq disabled to prevent changes to the clock */
1100 local_irq_save(flags);
1101 kvm_get_msr(v, MSR_IA32_TSC, &tsc_timestamp);
1102 kernel_ns = get_kernel_ns();
1103 this_tsc_khz = vcpu_tsc_khz(v);
1104 if (unlikely(this_tsc_khz == 0)) {
1105 local_irq_restore(flags);
1106 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1111 * We may have to catch up the TSC to match elapsed wall clock
1112 * time for two reasons, even if kvmclock is used.
1113 * 1) CPU could have been running below the maximum TSC rate
1114 * 2) Broken TSC compensation resets the base at each VCPU
1115 * entry to avoid unknown leaps of TSC even when running
1116 * again on the same CPU. This may cause apparent elapsed
1117 * time to disappear, and the guest to stand still or run
1120 if (vcpu->tsc_catchup) {
1121 u64 tsc = compute_guest_tsc(v, kernel_ns);
1122 if (tsc > tsc_timestamp) {
1123 kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1124 tsc_timestamp = tsc;
1128 local_irq_restore(flags);
1130 if (!vcpu->time_page)
1134 * Time as measured by the TSC may go backwards when resetting the base
1135 * tsc_timestamp. The reason for this is that the TSC resolution is
1136 * higher than the resolution of the other clock scales. Thus, many
1137 * possible measurments of the TSC correspond to one measurement of any
1138 * other clock, and so a spread of values is possible. This is not a
1139 * problem for the computation of the nanosecond clock; with TSC rates
1140 * around 1GHZ, there can only be a few cycles which correspond to one
1141 * nanosecond value, and any path through this code will inevitably
1142 * take longer than that. However, with the kernel_ns value itself,
1143 * the precision may be much lower, down to HZ granularity. If the
1144 * first sampling of TSC against kernel_ns ends in the low part of the
1145 * range, and the second in the high end of the range, we can get:
1147 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1149 * As the sampling errors potentially range in the thousands of cycles,
1150 * it is possible such a time value has already been observed by the
1151 * guest. To protect against this, we must compute the system time as
1152 * observed by the guest and ensure the new system time is greater.
1155 if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1156 max_kernel_ns = vcpu->last_guest_tsc -
1157 vcpu->hv_clock.tsc_timestamp;
1158 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1159 vcpu->hv_clock.tsc_to_system_mul,
1160 vcpu->hv_clock.tsc_shift);
1161 max_kernel_ns += vcpu->last_kernel_ns;
1164 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1165 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1166 &vcpu->hv_clock.tsc_shift,
1167 &vcpu->hv_clock.tsc_to_system_mul);
1168 vcpu->hw_tsc_khz = this_tsc_khz;
1171 if (max_kernel_ns > kernel_ns)
1172 kernel_ns = max_kernel_ns;
1174 /* With all the info we got, fill in the values */
1175 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1176 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1177 vcpu->last_kernel_ns = kernel_ns;
1178 vcpu->last_guest_tsc = tsc_timestamp;
1179 vcpu->hv_clock.flags = 0;
1182 * The interface expects us to write an even number signaling that the
1183 * update is finished. Since the guest won't see the intermediate
1184 * state, we just increase by 2 at the end.
1186 vcpu->hv_clock.version += 2;
1188 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1190 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1191 sizeof(vcpu->hv_clock));
1193 kunmap_atomic(shared_kaddr, KM_USER0);
1195 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1199 static bool msr_mtrr_valid(unsigned msr)
1202 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1203 case MSR_MTRRfix64K_00000:
1204 case MSR_MTRRfix16K_80000:
1205 case MSR_MTRRfix16K_A0000:
1206 case MSR_MTRRfix4K_C0000:
1207 case MSR_MTRRfix4K_C8000:
1208 case MSR_MTRRfix4K_D0000:
1209 case MSR_MTRRfix4K_D8000:
1210 case MSR_MTRRfix4K_E0000:
1211 case MSR_MTRRfix4K_E8000:
1212 case MSR_MTRRfix4K_F0000:
1213 case MSR_MTRRfix4K_F8000:
1214 case MSR_MTRRdefType:
1215 case MSR_IA32_CR_PAT:
1223 static bool valid_pat_type(unsigned t)
1225 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1228 static bool valid_mtrr_type(unsigned t)
1230 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1233 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1237 if (!msr_mtrr_valid(msr))
1240 if (msr == MSR_IA32_CR_PAT) {
1241 for (i = 0; i < 8; i++)
1242 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1245 } else if (msr == MSR_MTRRdefType) {
1248 return valid_mtrr_type(data & 0xff);
1249 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1250 for (i = 0; i < 8 ; i++)
1251 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1256 /* variable MTRRs */
1257 return valid_mtrr_type(data & 0xff);
1260 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1262 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1264 if (!mtrr_valid(vcpu, msr, data))
1267 if (msr == MSR_MTRRdefType) {
1268 vcpu->arch.mtrr_state.def_type = data;
1269 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1270 } else if (msr == MSR_MTRRfix64K_00000)
1272 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1273 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1274 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1275 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1276 else if (msr == MSR_IA32_CR_PAT)
1277 vcpu->arch.pat = data;
1278 else { /* Variable MTRRs */
1279 int idx, is_mtrr_mask;
1282 idx = (msr - 0x200) / 2;
1283 is_mtrr_mask = msr - 0x200 - 2 * idx;
1286 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1289 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1293 kvm_mmu_reset_context(vcpu);
1297 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1299 u64 mcg_cap = vcpu->arch.mcg_cap;
1300 unsigned bank_num = mcg_cap & 0xff;
1303 case MSR_IA32_MCG_STATUS:
1304 vcpu->arch.mcg_status = data;
1306 case MSR_IA32_MCG_CTL:
1307 if (!(mcg_cap & MCG_CTL_P))
1309 if (data != 0 && data != ~(u64)0)
1311 vcpu->arch.mcg_ctl = data;
1314 if (msr >= MSR_IA32_MC0_CTL &&
1315 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1316 u32 offset = msr - MSR_IA32_MC0_CTL;
1317 /* only 0 or all 1s can be written to IA32_MCi_CTL
1318 * some Linux kernels though clear bit 10 in bank 4 to
1319 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1320 * this to avoid an uncatched #GP in the guest
1322 if ((offset & 0x3) == 0 &&
1323 data != 0 && (data | (1 << 10)) != ~(u64)0)
1325 vcpu->arch.mce_banks[offset] = data;
1333 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1335 struct kvm *kvm = vcpu->kvm;
1336 int lm = is_long_mode(vcpu);
1337 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1338 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1339 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1340 : kvm->arch.xen_hvm_config.blob_size_32;
1341 u32 page_num = data & ~PAGE_MASK;
1342 u64 page_addr = data & PAGE_MASK;
1347 if (page_num >= blob_size)
1350 page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1354 if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1356 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1365 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1367 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1370 static bool kvm_hv_msr_partition_wide(u32 msr)
1374 case HV_X64_MSR_GUEST_OS_ID:
1375 case HV_X64_MSR_HYPERCALL:
1383 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1385 struct kvm *kvm = vcpu->kvm;
1388 case HV_X64_MSR_GUEST_OS_ID:
1389 kvm->arch.hv_guest_os_id = data;
1390 /* setting guest os id to zero disables hypercall page */
1391 if (!kvm->arch.hv_guest_os_id)
1392 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1394 case HV_X64_MSR_HYPERCALL: {
1399 /* if guest os id is not set hypercall should remain disabled */
1400 if (!kvm->arch.hv_guest_os_id)
1402 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1403 kvm->arch.hv_hypercall = data;
1406 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1407 addr = gfn_to_hva(kvm, gfn);
1408 if (kvm_is_error_hva(addr))
1410 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1411 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1412 if (__copy_to_user((void __user *)addr, instructions, 4))
1414 kvm->arch.hv_hypercall = data;
1418 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1419 "data 0x%llx\n", msr, data);
1425 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1428 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1431 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1432 vcpu->arch.hv_vapic = data;
1435 addr = gfn_to_hva(vcpu->kvm, data >>
1436 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1437 if (kvm_is_error_hva(addr))
1439 if (__clear_user((void __user *)addr, PAGE_SIZE))
1441 vcpu->arch.hv_vapic = data;
1444 case HV_X64_MSR_EOI:
1445 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1446 case HV_X64_MSR_ICR:
1447 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1448 case HV_X64_MSR_TPR:
1449 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1451 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1452 "data 0x%llx\n", msr, data);
1459 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1461 gpa_t gpa = data & ~0x3f;
1463 /* Bits 2:5 are resrved, Should be zero */
1467 vcpu->arch.apf.msr_val = data;
1469 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1470 kvm_clear_async_pf_completion_queue(vcpu);
1471 kvm_async_pf_hash_reset(vcpu);
1475 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1478 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1479 kvm_async_pf_wakeup_all(vcpu);
1483 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1485 if (vcpu->arch.time_page) {
1486 kvm_release_page_dirty(vcpu->arch.time_page);
1487 vcpu->arch.time_page = NULL;
1491 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1495 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1498 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1499 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1500 vcpu->arch.st.accum_steal = delta;
1503 static void record_steal_time(struct kvm_vcpu *vcpu)
1505 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1508 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1509 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1512 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1513 vcpu->arch.st.steal.version += 2;
1514 vcpu->arch.st.accum_steal = 0;
1516 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1517 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1520 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1524 return set_efer(vcpu, data);
1526 data &= ~(u64)0x40; /* ignore flush filter disable */
1527 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1529 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1534 case MSR_FAM10H_MMIO_CONF_BASE:
1536 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1541 case MSR_AMD64_NB_CFG:
1543 case MSR_IA32_DEBUGCTLMSR:
1545 /* We support the non-activated case already */
1547 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1548 /* Values other than LBR and BTF are vendor-specific,
1549 thus reserved and should throw a #GP */
1552 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1555 case MSR_IA32_UCODE_REV:
1556 case MSR_IA32_UCODE_WRITE:
1557 case MSR_VM_HSAVE_PA:
1558 case MSR_AMD64_PATCH_LOADER:
1560 case 0x200 ... 0x2ff:
1561 return set_msr_mtrr(vcpu, msr, data);
1562 case MSR_IA32_APICBASE:
1563 kvm_set_apic_base(vcpu, data);
1565 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1566 return kvm_x2apic_msr_write(vcpu, msr, data);
1567 case MSR_IA32_MISC_ENABLE:
1568 vcpu->arch.ia32_misc_enable_msr = data;
1570 case MSR_KVM_WALL_CLOCK_NEW:
1571 case MSR_KVM_WALL_CLOCK:
1572 vcpu->kvm->arch.wall_clock = data;
1573 kvm_write_wall_clock(vcpu->kvm, data);
1575 case MSR_KVM_SYSTEM_TIME_NEW:
1576 case MSR_KVM_SYSTEM_TIME: {
1577 kvmclock_reset(vcpu);
1579 vcpu->arch.time = data;
1580 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1582 /* we verify if the enable bit is set... */
1586 /* ...but clean it before doing the actual write */
1587 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1589 vcpu->arch.time_page =
1590 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1592 if (is_error_page(vcpu->arch.time_page)) {
1593 kvm_release_page_clean(vcpu->arch.time_page);
1594 vcpu->arch.time_page = NULL;
1598 case MSR_KVM_ASYNC_PF_EN:
1599 if (kvm_pv_enable_async_pf(vcpu, data))
1602 case MSR_KVM_STEAL_TIME:
1604 if (unlikely(!sched_info_on()))
1607 if (data & KVM_STEAL_RESERVED_MASK)
1610 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
1611 data & KVM_STEAL_VALID_BITS))
1614 vcpu->arch.st.msr_val = data;
1616 if (!(data & KVM_MSR_ENABLED))
1619 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1622 accumulate_steal_time(vcpu);
1625 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
1629 case MSR_IA32_MCG_CTL:
1630 case MSR_IA32_MCG_STATUS:
1631 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1632 return set_msr_mce(vcpu, msr, data);
1634 /* Performance counters are not protected by a CPUID bit,
1635 * so we should check all of them in the generic path for the sake of
1636 * cross vendor migration.
1637 * Writing a zero into the event select MSRs disables them,
1638 * which we perfectly emulate ;-). Any other value should be at least
1639 * reported, some guests depend on them.
1641 case MSR_P6_EVNTSEL0:
1642 case MSR_P6_EVNTSEL1:
1643 case MSR_K7_EVNTSEL0:
1644 case MSR_K7_EVNTSEL1:
1645 case MSR_K7_EVNTSEL2:
1646 case MSR_K7_EVNTSEL3:
1648 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1649 "0x%x data 0x%llx\n", msr, data);
1651 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1652 * so we ignore writes to make it happy.
1654 case MSR_P6_PERFCTR0:
1655 case MSR_P6_PERFCTR1:
1656 case MSR_K7_PERFCTR0:
1657 case MSR_K7_PERFCTR1:
1658 case MSR_K7_PERFCTR2:
1659 case MSR_K7_PERFCTR3:
1660 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1661 "0x%x data 0x%llx\n", msr, data);
1663 case MSR_K7_CLK_CTL:
1665 * Ignore all writes to this no longer documented MSR.
1666 * Writes are only relevant for old K7 processors,
1667 * all pre-dating SVM, but a recommended workaround from
1668 * AMD for these chips. It is possible to speicify the
1669 * affected processor models on the command line, hence
1670 * the need to ignore the workaround.
1673 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1674 if (kvm_hv_msr_partition_wide(msr)) {
1676 mutex_lock(&vcpu->kvm->lock);
1677 r = set_msr_hyperv_pw(vcpu, msr, data);
1678 mutex_unlock(&vcpu->kvm->lock);
1681 return set_msr_hyperv(vcpu, msr, data);
1683 case MSR_IA32_BBL_CR_CTL3:
1684 /* Drop writes to this legacy MSR -- see rdmsr
1685 * counterpart for further detail.
1687 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1690 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1691 return xen_hvm_config(vcpu, data);
1693 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1697 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1704 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1708 * Reads an msr value (of 'msr_index') into 'pdata'.
1709 * Returns 0 on success, non-0 otherwise.
1710 * Assumes vcpu_load() was already called.
1712 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1714 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1717 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1719 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1721 if (!msr_mtrr_valid(msr))
1724 if (msr == MSR_MTRRdefType)
1725 *pdata = vcpu->arch.mtrr_state.def_type +
1726 (vcpu->arch.mtrr_state.enabled << 10);
1727 else if (msr == MSR_MTRRfix64K_00000)
1729 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1730 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1731 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1732 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1733 else if (msr == MSR_IA32_CR_PAT)
1734 *pdata = vcpu->arch.pat;
1735 else { /* Variable MTRRs */
1736 int idx, is_mtrr_mask;
1739 idx = (msr - 0x200) / 2;
1740 is_mtrr_mask = msr - 0x200 - 2 * idx;
1743 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1746 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1753 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1756 u64 mcg_cap = vcpu->arch.mcg_cap;
1757 unsigned bank_num = mcg_cap & 0xff;
1760 case MSR_IA32_P5_MC_ADDR:
1761 case MSR_IA32_P5_MC_TYPE:
1764 case MSR_IA32_MCG_CAP:
1765 data = vcpu->arch.mcg_cap;
1767 case MSR_IA32_MCG_CTL:
1768 if (!(mcg_cap & MCG_CTL_P))
1770 data = vcpu->arch.mcg_ctl;
1772 case MSR_IA32_MCG_STATUS:
1773 data = vcpu->arch.mcg_status;
1776 if (msr >= MSR_IA32_MC0_CTL &&
1777 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1778 u32 offset = msr - MSR_IA32_MC0_CTL;
1779 data = vcpu->arch.mce_banks[offset];
1788 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1791 struct kvm *kvm = vcpu->kvm;
1794 case HV_X64_MSR_GUEST_OS_ID:
1795 data = kvm->arch.hv_guest_os_id;
1797 case HV_X64_MSR_HYPERCALL:
1798 data = kvm->arch.hv_hypercall;
1801 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1809 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1814 case HV_X64_MSR_VP_INDEX: {
1817 kvm_for_each_vcpu(r, v, vcpu->kvm)
1822 case HV_X64_MSR_EOI:
1823 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1824 case HV_X64_MSR_ICR:
1825 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1826 case HV_X64_MSR_TPR:
1827 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1828 case HV_X64_MSR_APIC_ASSIST_PAGE:
1829 data = vcpu->arch.hv_vapic;
1832 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1839 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1844 case MSR_IA32_PLATFORM_ID:
1845 case MSR_IA32_EBL_CR_POWERON:
1846 case MSR_IA32_DEBUGCTLMSR:
1847 case MSR_IA32_LASTBRANCHFROMIP:
1848 case MSR_IA32_LASTBRANCHTOIP:
1849 case MSR_IA32_LASTINTFROMIP:
1850 case MSR_IA32_LASTINTTOIP:
1853 case MSR_VM_HSAVE_PA:
1854 case MSR_P6_PERFCTR0:
1855 case MSR_P6_PERFCTR1:
1856 case MSR_P6_EVNTSEL0:
1857 case MSR_P6_EVNTSEL1:
1858 case MSR_K7_EVNTSEL0:
1859 case MSR_K7_PERFCTR0:
1860 case MSR_K8_INT_PENDING_MSG:
1861 case MSR_AMD64_NB_CFG:
1862 case MSR_FAM10H_MMIO_CONF_BASE:
1865 case MSR_IA32_UCODE_REV:
1866 data = 0x100000000ULL;
1869 data = 0x500 | KVM_NR_VAR_MTRR;
1871 case 0x200 ... 0x2ff:
1872 return get_msr_mtrr(vcpu, msr, pdata);
1873 case 0xcd: /* fsb frequency */
1877 * MSR_EBC_FREQUENCY_ID
1878 * Conservative value valid for even the basic CPU models.
1879 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1880 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1881 * and 266MHz for model 3, or 4. Set Core Clock
1882 * Frequency to System Bus Frequency Ratio to 1 (bits
1883 * 31:24) even though these are only valid for CPU
1884 * models > 2, however guests may end up dividing or
1885 * multiplying by zero otherwise.
1887 case MSR_EBC_FREQUENCY_ID:
1890 case MSR_IA32_APICBASE:
1891 data = kvm_get_apic_base(vcpu);
1893 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1894 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1896 case MSR_IA32_MISC_ENABLE:
1897 data = vcpu->arch.ia32_misc_enable_msr;
1899 case MSR_IA32_PERF_STATUS:
1900 /* TSC increment by tick */
1902 /* CPU multiplier */
1903 data |= (((uint64_t)4ULL) << 40);
1906 data = vcpu->arch.efer;
1908 case MSR_KVM_WALL_CLOCK:
1909 case MSR_KVM_WALL_CLOCK_NEW:
1910 data = vcpu->kvm->arch.wall_clock;
1912 case MSR_KVM_SYSTEM_TIME:
1913 case MSR_KVM_SYSTEM_TIME_NEW:
1914 data = vcpu->arch.time;
1916 case MSR_KVM_ASYNC_PF_EN:
1917 data = vcpu->arch.apf.msr_val;
1919 case MSR_KVM_STEAL_TIME:
1920 data = vcpu->arch.st.msr_val;
1922 case MSR_IA32_P5_MC_ADDR:
1923 case MSR_IA32_P5_MC_TYPE:
1924 case MSR_IA32_MCG_CAP:
1925 case MSR_IA32_MCG_CTL:
1926 case MSR_IA32_MCG_STATUS:
1927 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1928 return get_msr_mce(vcpu, msr, pdata);
1929 case MSR_K7_CLK_CTL:
1931 * Provide expected ramp-up count for K7. All other
1932 * are set to zero, indicating minimum divisors for
1935 * This prevents guest kernels on AMD host with CPU
1936 * type 6, model 8 and higher from exploding due to
1937 * the rdmsr failing.
1941 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1942 if (kvm_hv_msr_partition_wide(msr)) {
1944 mutex_lock(&vcpu->kvm->lock);
1945 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1946 mutex_unlock(&vcpu->kvm->lock);
1949 return get_msr_hyperv(vcpu, msr, pdata);
1951 case MSR_IA32_BBL_CR_CTL3:
1952 /* This legacy MSR exists but isn't fully documented in current
1953 * silicon. It is however accessed by winxp in very narrow
1954 * scenarios where it sets bit #19, itself documented as
1955 * a "reserved" bit. Best effort attempt to source coherent
1956 * read data here should the balance of the register be
1957 * interpreted by the guest:
1959 * L2 cache control register 3: 64GB range, 256KB size,
1960 * enabled, latency 0x1, configured
1966 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1969 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1977 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1980 * Read or write a bunch of msrs. All parameters are kernel addresses.
1982 * @return number of msrs set successfully.
1984 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1985 struct kvm_msr_entry *entries,
1986 int (*do_msr)(struct kvm_vcpu *vcpu,
1987 unsigned index, u64 *data))
1991 idx = srcu_read_lock(&vcpu->kvm->srcu);
1992 for (i = 0; i < msrs->nmsrs; ++i)
1993 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1995 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2001 * Read or write a bunch of msrs. Parameters are user addresses.
2003 * @return number of msrs set successfully.
2005 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2006 int (*do_msr)(struct kvm_vcpu *vcpu,
2007 unsigned index, u64 *data),
2010 struct kvm_msrs msrs;
2011 struct kvm_msr_entry *entries;
2016 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2020 if (msrs.nmsrs >= MAX_IO_MSRS)
2024 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2025 entries = kmalloc(size, GFP_KERNEL);
2030 if (copy_from_user(entries, user_msrs->entries, size))
2033 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2038 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2049 int kvm_dev_ioctl_check_extension(long ext)
2054 case KVM_CAP_IRQCHIP:
2056 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2057 case KVM_CAP_SET_TSS_ADDR:
2058 case KVM_CAP_EXT_CPUID:
2059 case KVM_CAP_CLOCKSOURCE:
2061 case KVM_CAP_NOP_IO_DELAY:
2062 case KVM_CAP_MP_STATE:
2063 case KVM_CAP_SYNC_MMU:
2064 case KVM_CAP_USER_NMI:
2065 case KVM_CAP_REINJECT_CONTROL:
2066 case KVM_CAP_IRQ_INJECT_STATUS:
2067 case KVM_CAP_ASSIGN_DEV_IRQ:
2069 case KVM_CAP_IOEVENTFD:
2071 case KVM_CAP_PIT_STATE2:
2072 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2073 case KVM_CAP_XEN_HVM:
2074 case KVM_CAP_ADJUST_CLOCK:
2075 case KVM_CAP_VCPU_EVENTS:
2076 case KVM_CAP_HYPERV:
2077 case KVM_CAP_HYPERV_VAPIC:
2078 case KVM_CAP_HYPERV_SPIN:
2079 case KVM_CAP_PCI_SEGMENT:
2080 case KVM_CAP_DEBUGREGS:
2081 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2083 case KVM_CAP_ASYNC_PF:
2084 case KVM_CAP_GET_TSC_KHZ:
2087 case KVM_CAP_COALESCED_MMIO:
2088 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2091 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2093 case KVM_CAP_NR_VCPUS:
2094 r = KVM_SOFT_MAX_VCPUS;
2096 case KVM_CAP_MAX_VCPUS:
2099 case KVM_CAP_NR_MEMSLOTS:
2100 r = KVM_MEMORY_SLOTS;
2102 case KVM_CAP_PV_MMU: /* obsolete */
2109 r = KVM_MAX_MCE_BANKS;
2114 case KVM_CAP_TSC_CONTROL:
2115 r = kvm_has_tsc_control;
2125 long kvm_arch_dev_ioctl(struct file *filp,
2126 unsigned int ioctl, unsigned long arg)
2128 void __user *argp = (void __user *)arg;
2132 case KVM_GET_MSR_INDEX_LIST: {
2133 struct kvm_msr_list __user *user_msr_list = argp;
2134 struct kvm_msr_list msr_list;
2138 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2141 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2142 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2145 if (n < msr_list.nmsrs)
2148 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2149 num_msrs_to_save * sizeof(u32)))
2151 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2153 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2158 case KVM_GET_SUPPORTED_CPUID: {
2159 struct kvm_cpuid2 __user *cpuid_arg = argp;
2160 struct kvm_cpuid2 cpuid;
2163 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2165 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2166 cpuid_arg->entries);
2171 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2176 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2179 mce_cap = KVM_MCE_CAP_SUPPORTED;
2181 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2193 static void wbinvd_ipi(void *garbage)
2198 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2200 return vcpu->kvm->arch.iommu_domain &&
2201 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2204 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2206 /* Address WBINVD may be executed by guest */
2207 if (need_emulate_wbinvd(vcpu)) {
2208 if (kvm_x86_ops->has_wbinvd_exit())
2209 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2210 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2211 smp_call_function_single(vcpu->cpu,
2212 wbinvd_ipi, NULL, 1);
2215 kvm_x86_ops->vcpu_load(vcpu, cpu);
2216 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2217 /* Make sure TSC doesn't go backwards */
2221 kvm_get_msr(vcpu, MSR_IA32_TSC, &tsc);
2222 tsc_delta = !vcpu->arch.last_guest_tsc ? 0 :
2223 tsc - vcpu->arch.last_guest_tsc;
2226 mark_tsc_unstable("KVM discovered backwards TSC");
2227 if (check_tsc_unstable()) {
2228 kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2229 vcpu->arch.tsc_catchup = 1;
2231 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2232 if (vcpu->cpu != cpu)
2233 kvm_migrate_timers(vcpu);
2237 accumulate_steal_time(vcpu);
2238 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2241 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2243 kvm_x86_ops->vcpu_put(vcpu);
2244 kvm_put_guest_fpu(vcpu);
2245 kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
2248 static int is_efer_nx(void)
2250 unsigned long long efer = 0;
2252 rdmsrl_safe(MSR_EFER, &efer);
2253 return efer & EFER_NX;
2256 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2259 struct kvm_cpuid_entry2 *e, *entry;
2262 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2263 e = &vcpu->arch.cpuid_entries[i];
2264 if (e->function == 0x80000001) {
2269 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2270 entry->edx &= ~(1 << 20);
2271 printk(KERN_INFO "kvm: guest NX capability removed\n");
2275 /* when an old userspace process fills a new kernel module */
2276 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2277 struct kvm_cpuid *cpuid,
2278 struct kvm_cpuid_entry __user *entries)
2281 struct kvm_cpuid_entry *cpuid_entries;
2284 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2287 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2291 if (copy_from_user(cpuid_entries, entries,
2292 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2294 for (i = 0; i < cpuid->nent; i++) {
2295 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2296 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2297 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2298 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2299 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2300 vcpu->arch.cpuid_entries[i].index = 0;
2301 vcpu->arch.cpuid_entries[i].flags = 0;
2302 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2303 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2304 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2306 vcpu->arch.cpuid_nent = cpuid->nent;
2307 cpuid_fix_nx_cap(vcpu);
2309 kvm_apic_set_version(vcpu);
2310 kvm_x86_ops->cpuid_update(vcpu);
2314 vfree(cpuid_entries);
2319 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2320 struct kvm_cpuid2 *cpuid,
2321 struct kvm_cpuid_entry2 __user *entries)
2326 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2329 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2330 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2332 vcpu->arch.cpuid_nent = cpuid->nent;
2333 kvm_apic_set_version(vcpu);
2334 kvm_x86_ops->cpuid_update(vcpu);
2342 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2343 struct kvm_cpuid2 *cpuid,
2344 struct kvm_cpuid_entry2 __user *entries)
2349 if (cpuid->nent < vcpu->arch.cpuid_nent)
2352 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2353 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2358 cpuid->nent = vcpu->arch.cpuid_nent;
2362 static void cpuid_mask(u32 *word, int wordnum)
2364 *word &= boot_cpu_data.x86_capability[wordnum];
2367 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2370 entry->function = function;
2371 entry->index = index;
2372 cpuid_count(entry->function, entry->index,
2373 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2377 static bool supported_xcr0_bit(unsigned bit)
2379 u64 mask = ((u64)1 << bit);
2381 return mask & (XSTATE_FP | XSTATE_SSE | XSTATE_YMM) & host_xcr0;
2384 #define F(x) bit(X86_FEATURE_##x)
2386 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2387 u32 index, int *nent, int maxnent)
2389 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2390 #ifdef CONFIG_X86_64
2391 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2393 unsigned f_lm = F(LM);
2395 unsigned f_gbpages = 0;
2398 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2401 const u32 kvm_supported_word0_x86_features =
2402 F(FPU) | F(VME) | F(DE) | F(PSE) |
2403 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2404 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2405 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2406 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2407 0 /* Reserved, DS, ACPI */ | F(MMX) |
2408 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2409 0 /* HTT, TM, Reserved, PBE */;
2410 /* cpuid 0x80000001.edx */
2411 const u32 kvm_supported_word1_x86_features =
2412 F(FPU) | F(VME) | F(DE) | F(PSE) |
2413 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2414 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2415 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2416 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2417 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2418 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2419 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2421 const u32 kvm_supported_word4_x86_features =
2422 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
2423 0 /* DS-CPL, VMX, SMX, EST */ |
2424 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2425 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2426 0 /* Reserved, DCA */ | F(XMM4_1) |
2427 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2428 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2429 F(F16C) | F(RDRAND);
2430 /* cpuid 0x80000001.ecx */
2431 const u32 kvm_supported_word6_x86_features =
2432 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2433 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2434 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2435 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2437 /* cpuid 0xC0000001.edx */
2438 const u32 kvm_supported_word5_x86_features =
2439 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
2440 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
2444 const u32 kvm_supported_word9_x86_features =
2445 F(SMEP) | F(FSGSBASE) | F(ERMS);
2447 /* all calls to cpuid_count() should be made on the same cpu */
2449 do_cpuid_1_ent(entry, function, index);
2454 entry->eax = min(entry->eax, (u32)0xd);
2457 entry->edx &= kvm_supported_word0_x86_features;
2458 cpuid_mask(&entry->edx, 0);
2459 entry->ecx &= kvm_supported_word4_x86_features;
2460 cpuid_mask(&entry->ecx, 4);
2461 /* we support x2apic emulation even if host does not support
2462 * it since we emulate x2apic in software */
2463 entry->ecx |= F(X2APIC);
2465 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2466 * may return different values. This forces us to get_cpu() before
2467 * issuing the first command, and also to emulate this annoying behavior
2468 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2470 int t, times = entry->eax & 0xff;
2472 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2473 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2474 for (t = 1; t < times && *nent < maxnent; ++t) {
2475 do_cpuid_1_ent(&entry[t], function, 0);
2476 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2481 /* function 4 has additional index. */
2485 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2486 /* read more entries until cache_type is zero */
2487 for (i = 1; *nent < maxnent; ++i) {
2488 cache_type = entry[i - 1].eax & 0x1f;
2491 do_cpuid_1_ent(&entry[i], function, i);
2493 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2499 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2500 /* Mask ebx against host capbability word 9 */
2502 entry->ebx &= kvm_supported_word9_x86_features;
2503 cpuid_mask(&entry->ebx, 9);
2513 /* function 0xb has additional index. */
2517 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2518 /* read more entries until level_type is zero */
2519 for (i = 1; *nent < maxnent; ++i) {
2520 level_type = entry[i - 1].ecx & 0xff00;
2523 do_cpuid_1_ent(&entry[i], function, i);
2525 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2533 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2534 for (idx = 1, i = 1; *nent < maxnent && idx < 64; ++idx) {
2535 do_cpuid_1_ent(&entry[i], function, idx);
2536 if (entry[i].eax == 0 || !supported_xcr0_bit(idx))
2539 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2545 case KVM_CPUID_SIGNATURE: {
2546 char signature[12] = "KVMKVMKVM\0\0";
2547 u32 *sigptr = (u32 *)signature;
2549 entry->ebx = sigptr[0];
2550 entry->ecx = sigptr[1];
2551 entry->edx = sigptr[2];
2554 case KVM_CPUID_FEATURES:
2555 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2556 (1 << KVM_FEATURE_NOP_IO_DELAY) |
2557 (1 << KVM_FEATURE_CLOCKSOURCE2) |
2558 (1 << KVM_FEATURE_ASYNC_PF) |
2559 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2561 if (sched_info_on())
2562 entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
2569 entry->eax = min(entry->eax, 0x8000001a);
2572 entry->edx &= kvm_supported_word1_x86_features;
2573 cpuid_mask(&entry->edx, 1);
2574 entry->ecx &= kvm_supported_word6_x86_features;
2575 cpuid_mask(&entry->ecx, 6);
2578 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
2579 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
2580 unsigned phys_as = entry->eax & 0xff;
2583 g_phys_as = phys_as;
2584 entry->eax = g_phys_as | (virt_as << 8);
2585 entry->ebx = entry->edx = 0;
2589 entry->ecx = entry->edx = 0;
2595 /*Add support for Centaur's CPUID instruction*/
2597 /*Just support up to 0xC0000004 now*/
2598 entry->eax = min(entry->eax, 0xC0000004);
2601 entry->edx &= kvm_supported_word5_x86_features;
2602 cpuid_mask(&entry->edx, 5);
2604 case 3: /* Processor serial number */
2605 case 5: /* MONITOR/MWAIT */
2606 case 6: /* Thermal management */
2607 case 0xA: /* Architectural Performance Monitoring */
2608 case 0x80000007: /* Advanced power management */
2613 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
2617 kvm_x86_ops->set_supported_cpuid(function, entry);
2624 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2625 struct kvm_cpuid_entry2 __user *entries)
2627 struct kvm_cpuid_entry2 *cpuid_entries;
2628 int limit, nent = 0, r = -E2BIG;
2631 if (cpuid->nent < 1)
2633 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2634 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2636 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2640 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2641 limit = cpuid_entries[0].eax;
2642 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2643 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2644 &nent, cpuid->nent);
2646 if (nent >= cpuid->nent)
2649 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2650 limit = cpuid_entries[nent - 1].eax;
2651 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2652 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2653 &nent, cpuid->nent);
2658 if (nent >= cpuid->nent)
2661 /* Add support for Centaur's CPUID instruction. */
2662 if (boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR) {
2663 do_cpuid_ent(&cpuid_entries[nent], 0xC0000000, 0,
2664 &nent, cpuid->nent);
2667 if (nent >= cpuid->nent)
2670 limit = cpuid_entries[nent - 1].eax;
2671 for (func = 0xC0000001;
2672 func <= limit && nent < cpuid->nent; ++func)
2673 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2674 &nent, cpuid->nent);
2677 if (nent >= cpuid->nent)
2681 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2685 if (nent >= cpuid->nent)
2688 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2692 if (nent >= cpuid->nent)
2696 if (copy_to_user(entries, cpuid_entries,
2697 nent * sizeof(struct kvm_cpuid_entry2)))
2703 vfree(cpuid_entries);
2708 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2709 struct kvm_lapic_state *s)
2711 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2716 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2717 struct kvm_lapic_state *s)
2719 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2720 kvm_apic_post_state_restore(vcpu);
2721 update_cr8_intercept(vcpu);
2726 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2727 struct kvm_interrupt *irq)
2729 if (irq->irq < 0 || irq->irq >= 256)
2731 if (irqchip_in_kernel(vcpu->kvm))
2734 kvm_queue_interrupt(vcpu, irq->irq, false);
2735 kvm_make_request(KVM_REQ_EVENT, vcpu);
2740 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2742 kvm_inject_nmi(vcpu);
2747 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2748 struct kvm_tpr_access_ctl *tac)
2752 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2756 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2760 unsigned bank_num = mcg_cap & 0xff, bank;
2763 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2765 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2768 vcpu->arch.mcg_cap = mcg_cap;
2769 /* Init IA32_MCG_CTL to all 1s */
2770 if (mcg_cap & MCG_CTL_P)
2771 vcpu->arch.mcg_ctl = ~(u64)0;
2772 /* Init IA32_MCi_CTL to all 1s */
2773 for (bank = 0; bank < bank_num; bank++)
2774 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2779 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2780 struct kvm_x86_mce *mce)
2782 u64 mcg_cap = vcpu->arch.mcg_cap;
2783 unsigned bank_num = mcg_cap & 0xff;
2784 u64 *banks = vcpu->arch.mce_banks;
2786 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2789 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2790 * reporting is disabled
2792 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2793 vcpu->arch.mcg_ctl != ~(u64)0)
2795 banks += 4 * mce->bank;
2797 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2798 * reporting is disabled for the bank
2800 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2802 if (mce->status & MCI_STATUS_UC) {
2803 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2804 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2805 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2808 if (banks[1] & MCI_STATUS_VAL)
2809 mce->status |= MCI_STATUS_OVER;
2810 banks[2] = mce->addr;
2811 banks[3] = mce->misc;
2812 vcpu->arch.mcg_status = mce->mcg_status;
2813 banks[1] = mce->status;
2814 kvm_queue_exception(vcpu, MC_VECTOR);
2815 } else if (!(banks[1] & MCI_STATUS_VAL)
2816 || !(banks[1] & MCI_STATUS_UC)) {
2817 if (banks[1] & MCI_STATUS_VAL)
2818 mce->status |= MCI_STATUS_OVER;
2819 banks[2] = mce->addr;
2820 banks[3] = mce->misc;
2821 banks[1] = mce->status;
2823 banks[1] |= MCI_STATUS_OVER;
2827 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2828 struct kvm_vcpu_events *events)
2830 events->exception.injected =
2831 vcpu->arch.exception.pending &&
2832 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2833 events->exception.nr = vcpu->arch.exception.nr;
2834 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2835 events->exception.pad = 0;
2836 events->exception.error_code = vcpu->arch.exception.error_code;
2838 events->interrupt.injected =
2839 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2840 events->interrupt.nr = vcpu->arch.interrupt.nr;
2841 events->interrupt.soft = 0;
2842 events->interrupt.shadow =
2843 kvm_x86_ops->get_interrupt_shadow(vcpu,
2844 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2846 events->nmi.injected = vcpu->arch.nmi_injected;
2847 events->nmi.pending = vcpu->arch.nmi_pending;
2848 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2849 events->nmi.pad = 0;
2851 events->sipi_vector = vcpu->arch.sipi_vector;
2853 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2854 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2855 | KVM_VCPUEVENT_VALID_SHADOW);
2856 memset(&events->reserved, 0, sizeof(events->reserved));
2859 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2860 struct kvm_vcpu_events *events)
2862 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2863 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2864 | KVM_VCPUEVENT_VALID_SHADOW))
2867 vcpu->arch.exception.pending = events->exception.injected;
2868 vcpu->arch.exception.nr = events->exception.nr;
2869 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2870 vcpu->arch.exception.error_code = events->exception.error_code;
2872 vcpu->arch.interrupt.pending = events->interrupt.injected;
2873 vcpu->arch.interrupt.nr = events->interrupt.nr;
2874 vcpu->arch.interrupt.soft = events->interrupt.soft;
2875 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2876 kvm_x86_ops->set_interrupt_shadow(vcpu,
2877 events->interrupt.shadow);
2879 vcpu->arch.nmi_injected = events->nmi.injected;
2880 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2881 vcpu->arch.nmi_pending = events->nmi.pending;
2882 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2884 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2885 vcpu->arch.sipi_vector = events->sipi_vector;
2887 kvm_make_request(KVM_REQ_EVENT, vcpu);
2892 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2893 struct kvm_debugregs *dbgregs)
2895 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2896 dbgregs->dr6 = vcpu->arch.dr6;
2897 dbgregs->dr7 = vcpu->arch.dr7;
2899 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2902 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2903 struct kvm_debugregs *dbgregs)
2908 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2909 vcpu->arch.dr6 = dbgregs->dr6;
2910 vcpu->arch.dr7 = dbgregs->dr7;
2915 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2916 struct kvm_xsave *guest_xsave)
2919 memcpy(guest_xsave->region,
2920 &vcpu->arch.guest_fpu.state->xsave,
2923 memcpy(guest_xsave->region,
2924 &vcpu->arch.guest_fpu.state->fxsave,
2925 sizeof(struct i387_fxsave_struct));
2926 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2931 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2932 struct kvm_xsave *guest_xsave)
2935 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2938 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2939 guest_xsave->region, xstate_size);
2941 if (xstate_bv & ~XSTATE_FPSSE)
2943 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2944 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2949 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2950 struct kvm_xcrs *guest_xcrs)
2952 if (!cpu_has_xsave) {
2953 guest_xcrs->nr_xcrs = 0;
2957 guest_xcrs->nr_xcrs = 1;
2958 guest_xcrs->flags = 0;
2959 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2960 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2963 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2964 struct kvm_xcrs *guest_xcrs)
2971 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2974 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2975 /* Only support XCR0 currently */
2976 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2977 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2978 guest_xcrs->xcrs[0].value);
2986 long kvm_arch_vcpu_ioctl(struct file *filp,
2987 unsigned int ioctl, unsigned long arg)
2989 struct kvm_vcpu *vcpu = filp->private_data;
2990 void __user *argp = (void __user *)arg;
2993 struct kvm_lapic_state *lapic;
2994 struct kvm_xsave *xsave;
2995 struct kvm_xcrs *xcrs;
3001 case KVM_GET_LAPIC: {
3003 if (!vcpu->arch.apic)
3005 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3010 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3014 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3019 case KVM_SET_LAPIC: {
3021 if (!vcpu->arch.apic)
3023 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3028 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
3030 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3036 case KVM_INTERRUPT: {
3037 struct kvm_interrupt irq;
3040 if (copy_from_user(&irq, argp, sizeof irq))
3042 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3049 r = kvm_vcpu_ioctl_nmi(vcpu);
3055 case KVM_SET_CPUID: {
3056 struct kvm_cpuid __user *cpuid_arg = argp;
3057 struct kvm_cpuid cpuid;
3060 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3062 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3067 case KVM_SET_CPUID2: {
3068 struct kvm_cpuid2 __user *cpuid_arg = argp;
3069 struct kvm_cpuid2 cpuid;
3072 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3074 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3075 cpuid_arg->entries);
3080 case KVM_GET_CPUID2: {
3081 struct kvm_cpuid2 __user *cpuid_arg = argp;
3082 struct kvm_cpuid2 cpuid;
3085 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3087 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3088 cpuid_arg->entries);
3092 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3098 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3101 r = msr_io(vcpu, argp, do_set_msr, 0);
3103 case KVM_TPR_ACCESS_REPORTING: {
3104 struct kvm_tpr_access_ctl tac;
3107 if (copy_from_user(&tac, argp, sizeof tac))
3109 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3113 if (copy_to_user(argp, &tac, sizeof tac))
3118 case KVM_SET_VAPIC_ADDR: {
3119 struct kvm_vapic_addr va;
3122 if (!irqchip_in_kernel(vcpu->kvm))
3125 if (copy_from_user(&va, argp, sizeof va))
3128 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3131 case KVM_X86_SETUP_MCE: {
3135 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3137 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3140 case KVM_X86_SET_MCE: {
3141 struct kvm_x86_mce mce;
3144 if (copy_from_user(&mce, argp, sizeof mce))
3146 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3149 case KVM_GET_VCPU_EVENTS: {
3150 struct kvm_vcpu_events events;
3152 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3155 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3160 case KVM_SET_VCPU_EVENTS: {
3161 struct kvm_vcpu_events events;
3164 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3167 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3170 case KVM_GET_DEBUGREGS: {
3171 struct kvm_debugregs dbgregs;
3173 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3176 if (copy_to_user(argp, &dbgregs,
3177 sizeof(struct kvm_debugregs)))
3182 case KVM_SET_DEBUGREGS: {
3183 struct kvm_debugregs dbgregs;
3186 if (copy_from_user(&dbgregs, argp,
3187 sizeof(struct kvm_debugregs)))
3190 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3193 case KVM_GET_XSAVE: {
3194 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3199 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3202 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3207 case KVM_SET_XSAVE: {
3208 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3214 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
3217 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3220 case KVM_GET_XCRS: {
3221 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3226 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3229 if (copy_to_user(argp, u.xcrs,
3230 sizeof(struct kvm_xcrs)))
3235 case KVM_SET_XCRS: {
3236 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3242 if (copy_from_user(u.xcrs, argp,
3243 sizeof(struct kvm_xcrs)))
3246 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3249 case KVM_SET_TSC_KHZ: {
3253 if (!kvm_has_tsc_control)
3256 user_tsc_khz = (u32)arg;
3258 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3261 kvm_x86_ops->set_tsc_khz(vcpu, user_tsc_khz);
3266 case KVM_GET_TSC_KHZ: {
3268 if (check_tsc_unstable())
3271 r = vcpu_tsc_khz(vcpu);
3283 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3287 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3289 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3293 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3296 kvm->arch.ept_identity_map_addr = ident_addr;
3300 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3301 u32 kvm_nr_mmu_pages)
3303 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3306 mutex_lock(&kvm->slots_lock);
3307 spin_lock(&kvm->mmu_lock);
3309 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3310 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3312 spin_unlock(&kvm->mmu_lock);
3313 mutex_unlock(&kvm->slots_lock);
3317 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3319 return kvm->arch.n_max_mmu_pages;
3322 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3327 switch (chip->chip_id) {
3328 case KVM_IRQCHIP_PIC_MASTER:
3329 memcpy(&chip->chip.pic,
3330 &pic_irqchip(kvm)->pics[0],
3331 sizeof(struct kvm_pic_state));
3333 case KVM_IRQCHIP_PIC_SLAVE:
3334 memcpy(&chip->chip.pic,
3335 &pic_irqchip(kvm)->pics[1],
3336 sizeof(struct kvm_pic_state));
3338 case KVM_IRQCHIP_IOAPIC:
3339 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3348 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3353 switch (chip->chip_id) {
3354 case KVM_IRQCHIP_PIC_MASTER:
3355 spin_lock(&pic_irqchip(kvm)->lock);
3356 memcpy(&pic_irqchip(kvm)->pics[0],
3358 sizeof(struct kvm_pic_state));
3359 spin_unlock(&pic_irqchip(kvm)->lock);
3361 case KVM_IRQCHIP_PIC_SLAVE:
3362 spin_lock(&pic_irqchip(kvm)->lock);
3363 memcpy(&pic_irqchip(kvm)->pics[1],
3365 sizeof(struct kvm_pic_state));
3366 spin_unlock(&pic_irqchip(kvm)->lock);
3368 case KVM_IRQCHIP_IOAPIC:
3369 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3375 kvm_pic_update_irq(pic_irqchip(kvm));
3379 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3383 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3384 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3385 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3389 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3393 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3394 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3395 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3396 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3400 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3404 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3405 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3406 sizeof(ps->channels));
3407 ps->flags = kvm->arch.vpit->pit_state.flags;
3408 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3409 memset(&ps->reserved, 0, sizeof(ps->reserved));
3413 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3415 int r = 0, start = 0;
3416 u32 prev_legacy, cur_legacy;
3417 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3418 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3419 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3420 if (!prev_legacy && cur_legacy)
3422 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3423 sizeof(kvm->arch.vpit->pit_state.channels));
3424 kvm->arch.vpit->pit_state.flags = ps->flags;
3425 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3426 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3430 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3431 struct kvm_reinject_control *control)
3433 if (!kvm->arch.vpit)
3435 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3436 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3437 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3442 * Get (and clear) the dirty memory log for a memory slot.
3444 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3445 struct kvm_dirty_log *log)
3448 struct kvm_memory_slot *memslot;
3450 unsigned long is_dirty = 0;
3452 mutex_lock(&kvm->slots_lock);
3455 if (log->slot >= KVM_MEMORY_SLOTS)
3458 memslot = &kvm->memslots->memslots[log->slot];
3460 if (!memslot->dirty_bitmap)
3463 n = kvm_dirty_bitmap_bytes(memslot);
3465 for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3466 is_dirty = memslot->dirty_bitmap[i];
3468 /* If nothing is dirty, don't bother messing with page tables. */
3470 struct kvm_memslots *slots, *old_slots;
3471 unsigned long *dirty_bitmap;
3473 dirty_bitmap = memslot->dirty_bitmap_head;
3474 if (memslot->dirty_bitmap == dirty_bitmap)
3475 dirty_bitmap += n / sizeof(long);
3476 memset(dirty_bitmap, 0, n);
3479 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3482 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3483 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3484 slots->generation++;
3486 old_slots = kvm->memslots;
3487 rcu_assign_pointer(kvm->memslots, slots);
3488 synchronize_srcu_expedited(&kvm->srcu);
3489 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3492 spin_lock(&kvm->mmu_lock);
3493 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3494 spin_unlock(&kvm->mmu_lock);
3497 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
3501 if (clear_user(log->dirty_bitmap, n))
3507 mutex_unlock(&kvm->slots_lock);
3511 long kvm_arch_vm_ioctl(struct file *filp,
3512 unsigned int ioctl, unsigned long arg)
3514 struct kvm *kvm = filp->private_data;
3515 void __user *argp = (void __user *)arg;
3518 * This union makes it completely explicit to gcc-3.x
3519 * that these two variables' stack usage should be
3520 * combined, not added together.
3523 struct kvm_pit_state ps;
3524 struct kvm_pit_state2 ps2;
3525 struct kvm_pit_config pit_config;
3529 case KVM_SET_TSS_ADDR:
3530 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3534 case KVM_SET_IDENTITY_MAP_ADDR: {
3538 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3540 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3545 case KVM_SET_NR_MMU_PAGES:
3546 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3550 case KVM_GET_NR_MMU_PAGES:
3551 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3553 case KVM_CREATE_IRQCHIP: {
3554 struct kvm_pic *vpic;
3556 mutex_lock(&kvm->lock);
3559 goto create_irqchip_unlock;
3561 vpic = kvm_create_pic(kvm);
3563 r = kvm_ioapic_init(kvm);
3565 mutex_lock(&kvm->slots_lock);
3566 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3568 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3570 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3572 mutex_unlock(&kvm->slots_lock);
3574 goto create_irqchip_unlock;
3577 goto create_irqchip_unlock;
3579 kvm->arch.vpic = vpic;
3581 r = kvm_setup_default_irq_routing(kvm);
3583 mutex_lock(&kvm->slots_lock);
3584 mutex_lock(&kvm->irq_lock);
3585 kvm_ioapic_destroy(kvm);
3586 kvm_destroy_pic(kvm);
3587 mutex_unlock(&kvm->irq_lock);
3588 mutex_unlock(&kvm->slots_lock);
3590 create_irqchip_unlock:
3591 mutex_unlock(&kvm->lock);
3594 case KVM_CREATE_PIT:
3595 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3597 case KVM_CREATE_PIT2:
3599 if (copy_from_user(&u.pit_config, argp,
3600 sizeof(struct kvm_pit_config)))
3603 mutex_lock(&kvm->slots_lock);
3606 goto create_pit_unlock;
3608 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3612 mutex_unlock(&kvm->slots_lock);
3614 case KVM_IRQ_LINE_STATUS:
3615 case KVM_IRQ_LINE: {
3616 struct kvm_irq_level irq_event;
3619 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3622 if (irqchip_in_kernel(kvm)) {
3624 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3625 irq_event.irq, irq_event.level);
3626 if (ioctl == KVM_IRQ_LINE_STATUS) {
3628 irq_event.status = status;
3629 if (copy_to_user(argp, &irq_event,
3637 case KVM_GET_IRQCHIP: {
3638 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3639 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3645 if (copy_from_user(chip, argp, sizeof *chip))
3646 goto get_irqchip_out;
3648 if (!irqchip_in_kernel(kvm))
3649 goto get_irqchip_out;
3650 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3652 goto get_irqchip_out;
3654 if (copy_to_user(argp, chip, sizeof *chip))
3655 goto get_irqchip_out;
3663 case KVM_SET_IRQCHIP: {
3664 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3665 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3671 if (copy_from_user(chip, argp, sizeof *chip))
3672 goto set_irqchip_out;
3674 if (!irqchip_in_kernel(kvm))
3675 goto set_irqchip_out;
3676 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3678 goto set_irqchip_out;
3688 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3691 if (!kvm->arch.vpit)
3693 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3697 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3704 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3707 if (!kvm->arch.vpit)
3709 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3715 case KVM_GET_PIT2: {
3717 if (!kvm->arch.vpit)
3719 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3723 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3728 case KVM_SET_PIT2: {
3730 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3733 if (!kvm->arch.vpit)
3735 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3741 case KVM_REINJECT_CONTROL: {
3742 struct kvm_reinject_control control;
3744 if (copy_from_user(&control, argp, sizeof(control)))
3746 r = kvm_vm_ioctl_reinject(kvm, &control);
3752 case KVM_XEN_HVM_CONFIG: {
3754 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3755 sizeof(struct kvm_xen_hvm_config)))
3758 if (kvm->arch.xen_hvm_config.flags)
3763 case KVM_SET_CLOCK: {
3764 struct kvm_clock_data user_ns;
3769 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3777 local_irq_disable();
3778 now_ns = get_kernel_ns();
3779 delta = user_ns.clock - now_ns;
3781 kvm->arch.kvmclock_offset = delta;
3784 case KVM_GET_CLOCK: {
3785 struct kvm_clock_data user_ns;
3788 local_irq_disable();
3789 now_ns = get_kernel_ns();
3790 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3793 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3796 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3809 static void kvm_init_msr_list(void)
3814 /* skip the first msrs in the list. KVM-specific */
3815 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3816 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3819 msrs_to_save[j] = msrs_to_save[i];
3822 num_msrs_to_save = j;
3825 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3833 if (!(vcpu->arch.apic &&
3834 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3835 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3846 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3853 if (!(vcpu->arch.apic &&
3854 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3855 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3857 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3867 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3868 struct kvm_segment *var, int seg)
3870 kvm_x86_ops->set_segment(vcpu, var, seg);
3873 void kvm_get_segment(struct kvm_vcpu *vcpu,
3874 struct kvm_segment *var, int seg)
3876 kvm_x86_ops->get_segment(vcpu, var, seg);
3879 static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3884 static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3887 struct x86_exception exception;
3889 BUG_ON(!mmu_is_nested(vcpu));
3891 /* NPT walks are always user-walks */
3892 access |= PFERR_USER_MASK;
3893 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3898 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3899 struct x86_exception *exception)
3901 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3902 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3905 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3906 struct x86_exception *exception)
3908 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3909 access |= PFERR_FETCH_MASK;
3910 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3913 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3914 struct x86_exception *exception)
3916 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3917 access |= PFERR_WRITE_MASK;
3918 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3921 /* uses this to access any guest's mapped memory without checking CPL */
3922 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3923 struct x86_exception *exception)
3925 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3928 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3929 struct kvm_vcpu *vcpu, u32 access,
3930 struct x86_exception *exception)
3933 int r = X86EMUL_CONTINUE;
3936 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3938 unsigned offset = addr & (PAGE_SIZE-1);
3939 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3942 if (gpa == UNMAPPED_GVA)
3943 return X86EMUL_PROPAGATE_FAULT;
3944 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3946 r = X86EMUL_IO_NEEDED;
3958 /* used for instruction fetching */
3959 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3960 gva_t addr, void *val, unsigned int bytes,
3961 struct x86_exception *exception)
3963 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3964 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3966 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3967 access | PFERR_FETCH_MASK,
3971 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3972 gva_t addr, void *val, unsigned int bytes,
3973 struct x86_exception *exception)
3975 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3976 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3978 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3981 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
3983 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3984 gva_t addr, void *val, unsigned int bytes,
3985 struct x86_exception *exception)
3987 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3988 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3991 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3992 gva_t addr, void *val,
3994 struct x86_exception *exception)
3996 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3998 int r = X86EMUL_CONTINUE;
4001 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4004 unsigned offset = addr & (PAGE_SIZE-1);
4005 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4008 if (gpa == UNMAPPED_GVA)
4009 return X86EMUL_PROPAGATE_FAULT;
4010 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
4012 r = X86EMUL_IO_NEEDED;
4023 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4025 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4026 gpa_t *gpa, struct x86_exception *exception,
4029 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4031 if (vcpu_match_mmio_gva(vcpu, gva) &&
4032 check_write_user_access(vcpu, write, access,
4033 vcpu->arch.access)) {
4034 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4035 (gva & (PAGE_SIZE - 1));
4036 trace_vcpu_match_mmio(gva, *gpa, write, false);
4041 access |= PFERR_WRITE_MASK;
4043 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4045 if (*gpa == UNMAPPED_GVA)
4048 /* For APIC access vmexit */
4049 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4052 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4053 trace_vcpu_match_mmio(gva, *gpa, write, true);
4060 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4061 const void *val, int bytes)
4065 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4068 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
4072 struct read_write_emulator_ops {
4073 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4075 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4076 void *val, int bytes);
4077 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4078 int bytes, void *val);
4079 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4080 void *val, int bytes);
4084 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4086 if (vcpu->mmio_read_completed) {
4087 memcpy(val, vcpu->mmio_data, bytes);
4088 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4089 vcpu->mmio_phys_addr, *(u64 *)val);
4090 vcpu->mmio_read_completed = 0;
4097 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4098 void *val, int bytes)
4100 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4103 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4104 void *val, int bytes)
4106 return emulator_write_phys(vcpu, gpa, val, bytes);
4109 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4111 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4112 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4115 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4116 void *val, int bytes)
4118 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4119 return X86EMUL_IO_NEEDED;
4122 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4123 void *val, int bytes)
4125 memcpy(vcpu->mmio_data, val, bytes);
4126 memcpy(vcpu->run->mmio.data, vcpu->mmio_data, 8);
4127 return X86EMUL_CONTINUE;
4130 static struct read_write_emulator_ops read_emultor = {
4131 .read_write_prepare = read_prepare,
4132 .read_write_emulate = read_emulate,
4133 .read_write_mmio = vcpu_mmio_read,
4134 .read_write_exit_mmio = read_exit_mmio,
4137 static struct read_write_emulator_ops write_emultor = {
4138 .read_write_emulate = write_emulate,
4139 .read_write_mmio = write_mmio,
4140 .read_write_exit_mmio = write_exit_mmio,
4144 static int emulator_read_write_onepage(unsigned long addr, void *val,
4146 struct x86_exception *exception,
4147 struct kvm_vcpu *vcpu,
4148 struct read_write_emulator_ops *ops)
4152 bool write = ops->write;
4154 if (ops->read_write_prepare &&
4155 ops->read_write_prepare(vcpu, val, bytes))
4156 return X86EMUL_CONTINUE;
4158 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4161 return X86EMUL_PROPAGATE_FAULT;
4163 /* For APIC access vmexit */
4167 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4168 return X86EMUL_CONTINUE;
4172 * Is this MMIO handled locally?
4174 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4175 if (handled == bytes)
4176 return X86EMUL_CONTINUE;
4182 vcpu->mmio_needed = 1;
4183 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4184 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
4185 vcpu->mmio_size = bytes;
4186 vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
4187 vcpu->run->mmio.is_write = vcpu->mmio_is_write = write;
4188 vcpu->mmio_index = 0;
4190 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4193 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4194 void *val, unsigned int bytes,
4195 struct x86_exception *exception,
4196 struct read_write_emulator_ops *ops)
4198 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4200 /* Crossing a page boundary? */
4201 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4204 now = -addr & ~PAGE_MASK;
4205 rc = emulator_read_write_onepage(addr, val, now, exception,
4208 if (rc != X86EMUL_CONTINUE)
4215 return emulator_read_write_onepage(addr, val, bytes, exception,
4219 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4223 struct x86_exception *exception)
4225 return emulator_read_write(ctxt, addr, val, bytes,
4226 exception, &read_emultor);
4229 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4233 struct x86_exception *exception)
4235 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4236 exception, &write_emultor);
4239 #define CMPXCHG_TYPE(t, ptr, old, new) \
4240 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4242 #ifdef CONFIG_X86_64
4243 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4245 # define CMPXCHG64(ptr, old, new) \
4246 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4249 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4254 struct x86_exception *exception)
4256 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4262 /* guests cmpxchg8b have to be emulated atomically */
4263 if (bytes > 8 || (bytes & (bytes - 1)))
4266 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4268 if (gpa == UNMAPPED_GVA ||
4269 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4272 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4275 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4276 if (is_error_page(page)) {
4277 kvm_release_page_clean(page);
4281 kaddr = kmap_atomic(page, KM_USER0);
4282 kaddr += offset_in_page(gpa);
4285 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4288 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4291 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4294 exchanged = CMPXCHG64(kaddr, old, new);
4299 kunmap_atomic(kaddr, KM_USER0);
4300 kvm_release_page_dirty(page);
4303 return X86EMUL_CMPXCHG_FAILED;
4305 kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
4307 return X86EMUL_CONTINUE;
4310 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4312 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4315 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4317 /* TODO: String I/O for in kernel device */
4320 if (vcpu->arch.pio.in)
4321 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4322 vcpu->arch.pio.size, pd);
4324 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4325 vcpu->arch.pio.port, vcpu->arch.pio.size,
4331 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4332 int size, unsigned short port, void *val,
4335 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4337 if (vcpu->arch.pio.count)
4340 trace_kvm_pio(0, port, size, count);
4342 vcpu->arch.pio.port = port;
4343 vcpu->arch.pio.in = 1;
4344 vcpu->arch.pio.count = count;
4345 vcpu->arch.pio.size = size;
4347 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4349 memcpy(val, vcpu->arch.pio_data, size * count);
4350 vcpu->arch.pio.count = 0;
4354 vcpu->run->exit_reason = KVM_EXIT_IO;
4355 vcpu->run->io.direction = KVM_EXIT_IO_IN;
4356 vcpu->run->io.size = size;
4357 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4358 vcpu->run->io.count = count;
4359 vcpu->run->io.port = port;
4364 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4365 int size, unsigned short port,
4366 const void *val, unsigned int count)
4368 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4370 trace_kvm_pio(1, port, size, count);
4372 vcpu->arch.pio.port = port;
4373 vcpu->arch.pio.in = 0;
4374 vcpu->arch.pio.count = count;
4375 vcpu->arch.pio.size = size;
4377 memcpy(vcpu->arch.pio_data, val, size * count);
4379 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4380 vcpu->arch.pio.count = 0;
4384 vcpu->run->exit_reason = KVM_EXIT_IO;
4385 vcpu->run->io.direction = KVM_EXIT_IO_OUT;
4386 vcpu->run->io.size = size;
4387 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4388 vcpu->run->io.count = count;
4389 vcpu->run->io.port = port;
4394 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4396 return kvm_x86_ops->get_segment_base(vcpu, seg);
4399 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4401 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4404 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4406 if (!need_emulate_wbinvd(vcpu))
4407 return X86EMUL_CONTINUE;
4409 if (kvm_x86_ops->has_wbinvd_exit()) {
4410 int cpu = get_cpu();
4412 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4413 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4414 wbinvd_ipi, NULL, 1);
4416 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4419 return X86EMUL_CONTINUE;
4421 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4423 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4425 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4428 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4430 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4433 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4436 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4439 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4441 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4444 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4446 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4447 unsigned long value;
4451 value = kvm_read_cr0(vcpu);
4454 value = vcpu->arch.cr2;
4457 value = kvm_read_cr3(vcpu);
4460 value = kvm_read_cr4(vcpu);
4463 value = kvm_get_cr8(vcpu);
4466 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4473 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4475 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4480 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4483 vcpu->arch.cr2 = val;
4486 res = kvm_set_cr3(vcpu, val);
4489 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4492 res = kvm_set_cr8(vcpu, val);
4495 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4502 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4504 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4507 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4509 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4512 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4514 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4517 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4519 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4522 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4524 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4527 static unsigned long emulator_get_cached_segment_base(
4528 struct x86_emulate_ctxt *ctxt, int seg)
4530 return get_segment_base(emul_to_vcpu(ctxt), seg);
4533 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4534 struct desc_struct *desc, u32 *base3,
4537 struct kvm_segment var;
4539 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4540 *selector = var.selector;
4547 set_desc_limit(desc, var.limit);
4548 set_desc_base(desc, (unsigned long)var.base);
4549 #ifdef CONFIG_X86_64
4551 *base3 = var.base >> 32;
4553 desc->type = var.type;
4555 desc->dpl = var.dpl;
4556 desc->p = var.present;
4557 desc->avl = var.avl;
4565 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4566 struct desc_struct *desc, u32 base3,
4569 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4570 struct kvm_segment var;
4572 var.selector = selector;
4573 var.base = get_desc_base(desc);
4574 #ifdef CONFIG_X86_64
4575 var.base |= ((u64)base3) << 32;
4577 var.limit = get_desc_limit(desc);
4579 var.limit = (var.limit << 12) | 0xfff;
4580 var.type = desc->type;
4581 var.present = desc->p;
4582 var.dpl = desc->dpl;
4587 var.avl = desc->avl;
4588 var.present = desc->p;
4589 var.unusable = !var.present;
4592 kvm_set_segment(vcpu, &var, seg);
4596 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4597 u32 msr_index, u64 *pdata)
4599 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4602 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4603 u32 msr_index, u64 data)
4605 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
4608 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4610 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4613 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4616 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4618 * CR0.TS may reference the host fpu state, not the guest fpu state,
4619 * so it may be clear at this point.
4624 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4629 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4630 struct x86_instruction_info *info,
4631 enum x86_intercept_stage stage)
4633 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4636 static struct x86_emulate_ops emulate_ops = {
4637 .read_std = kvm_read_guest_virt_system,
4638 .write_std = kvm_write_guest_virt_system,
4639 .fetch = kvm_fetch_guest_virt,
4640 .read_emulated = emulator_read_emulated,
4641 .write_emulated = emulator_write_emulated,
4642 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4643 .invlpg = emulator_invlpg,
4644 .pio_in_emulated = emulator_pio_in_emulated,
4645 .pio_out_emulated = emulator_pio_out_emulated,
4646 .get_segment = emulator_get_segment,
4647 .set_segment = emulator_set_segment,
4648 .get_cached_segment_base = emulator_get_cached_segment_base,
4649 .get_gdt = emulator_get_gdt,
4650 .get_idt = emulator_get_idt,
4651 .set_gdt = emulator_set_gdt,
4652 .set_idt = emulator_set_idt,
4653 .get_cr = emulator_get_cr,
4654 .set_cr = emulator_set_cr,
4655 .cpl = emulator_get_cpl,
4656 .get_dr = emulator_get_dr,
4657 .set_dr = emulator_set_dr,
4658 .set_msr = emulator_set_msr,
4659 .get_msr = emulator_get_msr,
4660 .halt = emulator_halt,
4661 .wbinvd = emulator_wbinvd,
4662 .fix_hypercall = emulator_fix_hypercall,
4663 .get_fpu = emulator_get_fpu,
4664 .put_fpu = emulator_put_fpu,
4665 .intercept = emulator_intercept,
4668 static void cache_all_regs(struct kvm_vcpu *vcpu)
4670 kvm_register_read(vcpu, VCPU_REGS_RAX);
4671 kvm_register_read(vcpu, VCPU_REGS_RSP);
4672 kvm_register_read(vcpu, VCPU_REGS_RIP);
4673 vcpu->arch.regs_dirty = ~0;
4676 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4678 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4680 * an sti; sti; sequence only disable interrupts for the first
4681 * instruction. So, if the last instruction, be it emulated or
4682 * not, left the system with the INT_STI flag enabled, it
4683 * means that the last instruction is an sti. We should not
4684 * leave the flag on in this case. The same goes for mov ss
4686 if (!(int_shadow & mask))
4687 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4690 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4692 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4693 if (ctxt->exception.vector == PF_VECTOR)
4694 kvm_propagate_fault(vcpu, &ctxt->exception);
4695 else if (ctxt->exception.error_code_valid)
4696 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4697 ctxt->exception.error_code);
4699 kvm_queue_exception(vcpu, ctxt->exception.vector);
4702 static void init_decode_cache(struct x86_emulate_ctxt *ctxt,
4703 const unsigned long *regs)
4705 memset(&ctxt->twobyte, 0,
4706 (void *)&ctxt->regs - (void *)&ctxt->twobyte);
4707 memcpy(ctxt->regs, regs, sizeof(ctxt->regs));
4709 ctxt->fetch.start = 0;
4710 ctxt->fetch.end = 0;
4711 ctxt->io_read.pos = 0;
4712 ctxt->io_read.end = 0;
4713 ctxt->mem_read.pos = 0;
4714 ctxt->mem_read.end = 0;
4717 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4719 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4723 * TODO: fix emulate.c to use guest_read/write_register
4724 * instead of direct ->regs accesses, can save hundred cycles
4725 * on Intel for instructions that don't read/change RSP, for
4728 cache_all_regs(vcpu);
4730 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4732 ctxt->eflags = kvm_get_rflags(vcpu);
4733 ctxt->eip = kvm_rip_read(vcpu);
4734 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4735 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4736 cs_l ? X86EMUL_MODE_PROT64 :
4737 cs_db ? X86EMUL_MODE_PROT32 :
4738 X86EMUL_MODE_PROT16;
4739 ctxt->guest_mode = is_guest_mode(vcpu);
4741 init_decode_cache(ctxt, vcpu->arch.regs);
4742 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4745 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4747 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4750 init_emulate_ctxt(vcpu);
4754 ctxt->_eip = ctxt->eip + inc_eip;
4755 ret = emulate_int_real(ctxt, irq);
4757 if (ret != X86EMUL_CONTINUE)
4758 return EMULATE_FAIL;
4760 ctxt->eip = ctxt->_eip;
4761 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4762 kvm_rip_write(vcpu, ctxt->eip);
4763 kvm_set_rflags(vcpu, ctxt->eflags);
4765 if (irq == NMI_VECTOR)
4766 vcpu->arch.nmi_pending = false;
4768 vcpu->arch.interrupt.pending = false;
4770 return EMULATE_DONE;
4772 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4774 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4776 int r = EMULATE_DONE;
4778 ++vcpu->stat.insn_emulation_fail;
4779 trace_kvm_emulate_insn_failed(vcpu);
4780 if (!is_guest_mode(vcpu)) {
4781 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4782 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4783 vcpu->run->internal.ndata = 0;
4786 kvm_queue_exception(vcpu, UD_VECTOR);
4791 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4799 * if emulation was due to access to shadowed page table
4800 * and it failed try to unshadow page and re-entetr the
4801 * guest to let CPU execute the instruction.
4803 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4806 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4808 if (gpa == UNMAPPED_GVA)
4809 return true; /* let cpu generate fault */
4811 if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4817 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4824 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4825 bool writeback = true;
4827 kvm_clear_exception_queue(vcpu);
4829 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4830 init_emulate_ctxt(vcpu);
4831 ctxt->interruptibility = 0;
4832 ctxt->have_exception = false;
4833 ctxt->perm_ok = false;
4835 ctxt->only_vendor_specific_insn
4836 = emulation_type & EMULTYPE_TRAP_UD;
4838 r = x86_decode_insn(ctxt, insn, insn_len);
4840 trace_kvm_emulate_insn_start(vcpu);
4841 ++vcpu->stat.insn_emulation;
4842 if (r != EMULATION_OK) {
4843 if (emulation_type & EMULTYPE_TRAP_UD)
4844 return EMULATE_FAIL;
4845 if (reexecute_instruction(vcpu, cr2))
4846 return EMULATE_DONE;
4847 if (emulation_type & EMULTYPE_SKIP)
4848 return EMULATE_FAIL;
4849 return handle_emulation_failure(vcpu);
4853 if (emulation_type & EMULTYPE_SKIP) {
4854 kvm_rip_write(vcpu, ctxt->_eip);
4855 return EMULATE_DONE;
4858 /* this is needed for vmware backdoor interface to work since it
4859 changes registers values during IO operation */
4860 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4861 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4862 memcpy(ctxt->regs, vcpu->arch.regs, sizeof ctxt->regs);
4866 r = x86_emulate_insn(ctxt);
4868 if (r == EMULATION_INTERCEPTED)
4869 return EMULATE_DONE;
4871 if (r == EMULATION_FAILED) {
4872 if (reexecute_instruction(vcpu, cr2))
4873 return EMULATE_DONE;
4875 return handle_emulation_failure(vcpu);
4878 if (ctxt->have_exception) {
4879 inject_emulated_exception(vcpu);
4881 } else if (vcpu->arch.pio.count) {
4882 if (!vcpu->arch.pio.in)
4883 vcpu->arch.pio.count = 0;
4886 r = EMULATE_DO_MMIO;
4887 } else if (vcpu->mmio_needed) {
4888 if (!vcpu->mmio_is_write)
4890 r = EMULATE_DO_MMIO;
4891 } else if (r == EMULATION_RESTART)
4897 toggle_interruptibility(vcpu, ctxt->interruptibility);
4898 kvm_set_rflags(vcpu, ctxt->eflags);
4899 kvm_make_request(KVM_REQ_EVENT, vcpu);
4900 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4901 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4902 kvm_rip_write(vcpu, ctxt->eip);
4904 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
4908 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4910 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4912 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4913 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4914 size, port, &val, 1);
4915 /* do not return to emulator after return from userspace */
4916 vcpu->arch.pio.count = 0;
4919 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4921 static void tsc_bad(void *info)
4923 __this_cpu_write(cpu_tsc_khz, 0);
4926 static void tsc_khz_changed(void *data)
4928 struct cpufreq_freqs *freq = data;
4929 unsigned long khz = 0;
4933 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4934 khz = cpufreq_quick_get(raw_smp_processor_id());
4937 __this_cpu_write(cpu_tsc_khz, khz);
4940 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4943 struct cpufreq_freqs *freq = data;
4945 struct kvm_vcpu *vcpu;
4946 int i, send_ipi = 0;
4949 * We allow guests to temporarily run on slowing clocks,
4950 * provided we notify them after, or to run on accelerating
4951 * clocks, provided we notify them before. Thus time never
4954 * However, we have a problem. We can't atomically update
4955 * the frequency of a given CPU from this function; it is
4956 * merely a notifier, which can be called from any CPU.
4957 * Changing the TSC frequency at arbitrary points in time
4958 * requires a recomputation of local variables related to
4959 * the TSC for each VCPU. We must flag these local variables
4960 * to be updated and be sure the update takes place with the
4961 * new frequency before any guests proceed.
4963 * Unfortunately, the combination of hotplug CPU and frequency
4964 * change creates an intractable locking scenario; the order
4965 * of when these callouts happen is undefined with respect to
4966 * CPU hotplug, and they can race with each other. As such,
4967 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4968 * undefined; you can actually have a CPU frequency change take
4969 * place in between the computation of X and the setting of the
4970 * variable. To protect against this problem, all updates of
4971 * the per_cpu tsc_khz variable are done in an interrupt
4972 * protected IPI, and all callers wishing to update the value
4973 * must wait for a synchronous IPI to complete (which is trivial
4974 * if the caller is on the CPU already). This establishes the
4975 * necessary total order on variable updates.
4977 * Note that because a guest time update may take place
4978 * anytime after the setting of the VCPU's request bit, the
4979 * correct TSC value must be set before the request. However,
4980 * to ensure the update actually makes it to any guest which
4981 * starts running in hardware virtualization between the set
4982 * and the acquisition of the spinlock, we must also ping the
4983 * CPU after setting the request bit.
4987 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4989 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4992 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4994 raw_spin_lock(&kvm_lock);
4995 list_for_each_entry(kvm, &vm_list, vm_list) {
4996 kvm_for_each_vcpu(i, vcpu, kvm) {
4997 if (vcpu->cpu != freq->cpu)
4999 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5000 if (vcpu->cpu != smp_processor_id())
5004 raw_spin_unlock(&kvm_lock);
5006 if (freq->old < freq->new && send_ipi) {
5008 * We upscale the frequency. Must make the guest
5009 * doesn't see old kvmclock values while running with
5010 * the new frequency, otherwise we risk the guest sees
5011 * time go backwards.
5013 * In case we update the frequency for another cpu
5014 * (which might be in guest context) send an interrupt
5015 * to kick the cpu out of guest context. Next time
5016 * guest context is entered kvmclock will be updated,
5017 * so the guest will not see stale values.
5019 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5024 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5025 .notifier_call = kvmclock_cpufreq_notifier
5028 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5029 unsigned long action, void *hcpu)
5031 unsigned int cpu = (unsigned long)hcpu;
5035 case CPU_DOWN_FAILED:
5036 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5038 case CPU_DOWN_PREPARE:
5039 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5045 static struct notifier_block kvmclock_cpu_notifier_block = {
5046 .notifier_call = kvmclock_cpu_notifier,
5047 .priority = -INT_MAX
5050 static void kvm_timer_init(void)
5054 max_tsc_khz = tsc_khz;
5055 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5056 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5057 #ifdef CONFIG_CPU_FREQ
5058 struct cpufreq_policy policy;
5059 memset(&policy, 0, sizeof(policy));
5061 cpufreq_get_policy(&policy, cpu);
5062 if (policy.cpuinfo.max_freq)
5063 max_tsc_khz = policy.cpuinfo.max_freq;
5066 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5067 CPUFREQ_TRANSITION_NOTIFIER);
5069 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5070 for_each_online_cpu(cpu)
5071 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5074 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5076 static int kvm_is_in_guest(void)
5078 return percpu_read(current_vcpu) != NULL;
5081 static int kvm_is_user_mode(void)
5085 if (percpu_read(current_vcpu))
5086 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
5088 return user_mode != 0;
5091 static unsigned long kvm_get_guest_ip(void)
5093 unsigned long ip = 0;
5095 if (percpu_read(current_vcpu))
5096 ip = kvm_rip_read(percpu_read(current_vcpu));
5101 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5102 .is_in_guest = kvm_is_in_guest,
5103 .is_user_mode = kvm_is_user_mode,
5104 .get_guest_ip = kvm_get_guest_ip,
5107 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5109 percpu_write(current_vcpu, vcpu);
5111 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5113 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5115 percpu_write(current_vcpu, NULL);
5117 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5119 static void kvm_set_mmio_spte_mask(void)
5122 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5125 * Set the reserved bits and the present bit of an paging-structure
5126 * entry to generate page fault with PFER.RSV = 1.
5128 mask = ((1ull << (62 - maxphyaddr + 1)) - 1) << maxphyaddr;
5131 #ifdef CONFIG_X86_64
5133 * If reserved bit is not supported, clear the present bit to disable
5136 if (maxphyaddr == 52)
5140 kvm_mmu_set_mmio_spte_mask(mask);
5143 int kvm_arch_init(void *opaque)
5146 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
5149 printk(KERN_ERR "kvm: already loaded the other module\n");
5154 if (!ops->cpu_has_kvm_support()) {
5155 printk(KERN_ERR "kvm: no hardware support\n");
5159 if (ops->disabled_by_bios()) {
5160 printk(KERN_ERR "kvm: disabled by bios\n");
5165 r = kvm_mmu_module_init();
5169 kvm_set_mmio_spte_mask();
5170 kvm_init_msr_list();
5173 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5174 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5178 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5181 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5189 void kvm_arch_exit(void)
5191 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5193 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5194 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5195 CPUFREQ_TRANSITION_NOTIFIER);
5196 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5198 kvm_mmu_module_exit();
5201 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5203 ++vcpu->stat.halt_exits;
5204 if (irqchip_in_kernel(vcpu->kvm)) {
5205 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5208 vcpu->run->exit_reason = KVM_EXIT_HLT;
5212 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5214 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
5217 if (is_long_mode(vcpu))
5220 return a0 | ((gpa_t)a1 << 32);
5223 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5225 u64 param, ingpa, outgpa, ret;
5226 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5227 bool fast, longmode;
5231 * hypercall generates UD from non zero cpl and real mode
5234 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5235 kvm_queue_exception(vcpu, UD_VECTOR);
5239 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5240 longmode = is_long_mode(vcpu) && cs_l == 1;
5243 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5244 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5245 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5246 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5247 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5248 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5250 #ifdef CONFIG_X86_64
5252 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5253 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5254 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5258 code = param & 0xffff;
5259 fast = (param >> 16) & 0x1;
5260 rep_cnt = (param >> 32) & 0xfff;
5261 rep_idx = (param >> 48) & 0xfff;
5263 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5266 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5267 kvm_vcpu_on_spin(vcpu);
5270 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5274 ret = res | (((u64)rep_done & 0xfff) << 32);
5276 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5278 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5279 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5285 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5287 unsigned long nr, a0, a1, a2, a3, ret;
5290 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5291 return kvm_hv_hypercall(vcpu);
5293 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5294 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5295 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5296 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5297 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5299 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5301 if (!is_long_mode(vcpu)) {
5309 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5315 case KVM_HC_VAPIC_POLL_IRQ:
5319 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
5326 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5327 ++vcpu->stat.hypercalls;
5330 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5332 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5334 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5335 char instruction[3];
5336 unsigned long rip = kvm_rip_read(vcpu);
5339 * Blow out the MMU to ensure that no other VCPU has an active mapping
5340 * to ensure that the updated hypercall appears atomically across all
5343 kvm_mmu_zap_all(vcpu->kvm);
5345 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5347 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5350 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
5352 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
5353 int j, nent = vcpu->arch.cpuid_nent;
5355 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
5356 /* when no next entry is found, the current entry[i] is reselected */
5357 for (j = i + 1; ; j = (j + 1) % nent) {
5358 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
5359 if (ej->function == e->function) {
5360 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
5364 return 0; /* silence gcc, even though control never reaches here */
5367 /* find an entry with matching function, matching index (if needed), and that
5368 * should be read next (if it's stateful) */
5369 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
5370 u32 function, u32 index)
5372 if (e->function != function)
5374 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
5376 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
5377 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
5382 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
5383 u32 function, u32 index)
5386 struct kvm_cpuid_entry2 *best = NULL;
5388 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
5389 struct kvm_cpuid_entry2 *e;
5391 e = &vcpu->arch.cpuid_entries[i];
5392 if (is_matching_cpuid_entry(e, function, index)) {
5393 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
5394 move_to_next_stateful_cpuid_entry(vcpu, i);
5401 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
5403 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
5405 struct kvm_cpuid_entry2 *best;
5407 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
5408 if (!best || best->eax < 0x80000008)
5410 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
5412 return best->eax & 0xff;
5418 * If no match is found, check whether we exceed the vCPU's limit
5419 * and return the content of the highest valid _standard_ leaf instead.
5420 * This is to satisfy the CPUID specification.
5422 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
5423 u32 function, u32 index)
5425 struct kvm_cpuid_entry2 *maxlevel;
5427 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
5428 if (!maxlevel || maxlevel->eax >= function)
5430 if (function & 0x80000000) {
5431 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
5435 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
5438 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
5440 u32 function, index;
5441 struct kvm_cpuid_entry2 *best;
5443 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
5444 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5445 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
5446 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
5447 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
5448 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
5449 best = kvm_find_cpuid_entry(vcpu, function, index);
5452 best = check_cpuid_limit(vcpu, function, index);
5455 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
5456 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
5457 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
5458 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
5460 kvm_x86_ops->skip_emulated_instruction(vcpu);
5461 trace_kvm_cpuid(function,
5462 kvm_register_read(vcpu, VCPU_REGS_RAX),
5463 kvm_register_read(vcpu, VCPU_REGS_RBX),
5464 kvm_register_read(vcpu, VCPU_REGS_RCX),
5465 kvm_register_read(vcpu, VCPU_REGS_RDX));
5467 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
5470 * Check if userspace requested an interrupt window, and that the
5471 * interrupt window is open.
5473 * No need to exit to userspace if we already have an interrupt queued.
5475 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5477 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5478 vcpu->run->request_interrupt_window &&
5479 kvm_arch_interrupt_allowed(vcpu));
5482 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5484 struct kvm_run *kvm_run = vcpu->run;
5486 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5487 kvm_run->cr8 = kvm_get_cr8(vcpu);
5488 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5489 if (irqchip_in_kernel(vcpu->kvm))
5490 kvm_run->ready_for_interrupt_injection = 1;
5492 kvm_run->ready_for_interrupt_injection =
5493 kvm_arch_interrupt_allowed(vcpu) &&
5494 !kvm_cpu_has_interrupt(vcpu) &&
5495 !kvm_event_needs_reinjection(vcpu);
5498 static void vapic_enter(struct kvm_vcpu *vcpu)
5500 struct kvm_lapic *apic = vcpu->arch.apic;
5503 if (!apic || !apic->vapic_addr)
5506 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5508 vcpu->arch.apic->vapic_page = page;
5511 static void vapic_exit(struct kvm_vcpu *vcpu)
5513 struct kvm_lapic *apic = vcpu->arch.apic;
5516 if (!apic || !apic->vapic_addr)
5519 idx = srcu_read_lock(&vcpu->kvm->srcu);
5520 kvm_release_page_dirty(apic->vapic_page);
5521 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5522 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5525 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5529 if (!kvm_x86_ops->update_cr8_intercept)
5532 if (!vcpu->arch.apic)
5535 if (!vcpu->arch.apic->vapic_addr)
5536 max_irr = kvm_lapic_find_highest_irr(vcpu);
5543 tpr = kvm_lapic_get_cr8(vcpu);
5545 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5548 static void inject_pending_event(struct kvm_vcpu *vcpu)
5550 /* try to reinject previous events if any */
5551 if (vcpu->arch.exception.pending) {
5552 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5553 vcpu->arch.exception.has_error_code,
5554 vcpu->arch.exception.error_code);
5555 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5556 vcpu->arch.exception.has_error_code,
5557 vcpu->arch.exception.error_code,
5558 vcpu->arch.exception.reinject);
5562 if (vcpu->arch.nmi_injected) {
5563 kvm_x86_ops->set_nmi(vcpu);
5567 if (vcpu->arch.interrupt.pending) {
5568 kvm_x86_ops->set_irq(vcpu);
5572 /* try to inject new event if pending */
5573 if (vcpu->arch.nmi_pending) {
5574 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5575 vcpu->arch.nmi_pending = false;
5576 vcpu->arch.nmi_injected = true;
5577 kvm_x86_ops->set_nmi(vcpu);
5579 } else if (kvm_cpu_has_interrupt(vcpu)) {
5580 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5581 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5583 kvm_x86_ops->set_irq(vcpu);
5588 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5590 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5591 !vcpu->guest_xcr0_loaded) {
5592 /* kvm_set_xcr() also depends on this */
5593 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5594 vcpu->guest_xcr0_loaded = 1;
5598 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5600 if (vcpu->guest_xcr0_loaded) {
5601 if (vcpu->arch.xcr0 != host_xcr0)
5602 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5603 vcpu->guest_xcr0_loaded = 0;
5607 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5611 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5612 vcpu->run->request_interrupt_window;
5614 if (vcpu->requests) {
5615 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5616 kvm_mmu_unload(vcpu);
5617 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5618 __kvm_migrate_timers(vcpu);
5619 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5620 r = kvm_guest_time_update(vcpu);
5624 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5625 kvm_mmu_sync_roots(vcpu);
5626 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5627 kvm_x86_ops->tlb_flush(vcpu);
5628 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5629 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5633 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5634 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5638 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5639 vcpu->fpu_active = 0;
5640 kvm_x86_ops->fpu_deactivate(vcpu);
5642 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5643 /* Page is swapped out. Do synthetic halt */
5644 vcpu->arch.apf.halted = true;
5648 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5649 record_steal_time(vcpu);
5653 r = kvm_mmu_reload(vcpu);
5658 * An NMI can be injected between local nmi_pending read and
5659 * vcpu->arch.nmi_pending read inside inject_pending_event().
5660 * But in that case, KVM_REQ_EVENT will be set, which makes
5661 * the race described above benign.
5663 nmi_pending = ACCESS_ONCE(vcpu->arch.nmi_pending);
5665 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5666 inject_pending_event(vcpu);
5668 /* enable NMI/IRQ window open exits if needed */
5670 kvm_x86_ops->enable_nmi_window(vcpu);
5671 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5672 kvm_x86_ops->enable_irq_window(vcpu);
5674 if (kvm_lapic_enabled(vcpu)) {
5675 update_cr8_intercept(vcpu);
5676 kvm_lapic_sync_to_vapic(vcpu);
5682 kvm_x86_ops->prepare_guest_switch(vcpu);
5683 if (vcpu->fpu_active)
5684 kvm_load_guest_fpu(vcpu);
5685 kvm_load_guest_xcr0(vcpu);
5687 vcpu->mode = IN_GUEST_MODE;
5689 /* We should set ->mode before check ->requests,
5690 * see the comment in make_all_cpus_request.
5694 local_irq_disable();
5696 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5697 || need_resched() || signal_pending(current)) {
5698 vcpu->mode = OUTSIDE_GUEST_MODE;
5702 kvm_x86_ops->cancel_injection(vcpu);
5707 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5711 if (unlikely(vcpu->arch.switch_db_regs)) {
5713 set_debugreg(vcpu->arch.eff_db[0], 0);
5714 set_debugreg(vcpu->arch.eff_db[1], 1);
5715 set_debugreg(vcpu->arch.eff_db[2], 2);
5716 set_debugreg(vcpu->arch.eff_db[3], 3);
5719 trace_kvm_entry(vcpu->vcpu_id);
5720 kvm_x86_ops->run(vcpu);
5723 * If the guest has used debug registers, at least dr7
5724 * will be disabled while returning to the host.
5725 * If we don't have active breakpoints in the host, we don't
5726 * care about the messed up debug address registers. But if
5727 * we have some of them active, restore the old state.
5729 if (hw_breakpoint_active())
5730 hw_breakpoint_restore();
5732 kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
5734 vcpu->mode = OUTSIDE_GUEST_MODE;
5741 * We must have an instruction between local_irq_enable() and
5742 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5743 * the interrupt shadow. The stat.exits increment will do nicely.
5744 * But we need to prevent reordering, hence this barrier():
5752 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5755 * Profile KVM exit RIPs:
5757 if (unlikely(prof_on == KVM_PROFILING)) {
5758 unsigned long rip = kvm_rip_read(vcpu);
5759 profile_hit(KVM_PROFILING, (void *)rip);
5763 kvm_lapic_sync_from_vapic(vcpu);
5765 r = kvm_x86_ops->handle_exit(vcpu);
5771 static int __vcpu_run(struct kvm_vcpu *vcpu)
5774 struct kvm *kvm = vcpu->kvm;
5776 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5777 pr_debug("vcpu %d received sipi with vector # %x\n",
5778 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5779 kvm_lapic_reset(vcpu);
5780 r = kvm_arch_vcpu_reset(vcpu);
5783 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5786 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5791 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5792 !vcpu->arch.apf.halted)
5793 r = vcpu_enter_guest(vcpu);
5795 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5796 kvm_vcpu_block(vcpu);
5797 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5798 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5800 switch(vcpu->arch.mp_state) {
5801 case KVM_MP_STATE_HALTED:
5802 vcpu->arch.mp_state =
5803 KVM_MP_STATE_RUNNABLE;
5804 case KVM_MP_STATE_RUNNABLE:
5805 vcpu->arch.apf.halted = false;
5807 case KVM_MP_STATE_SIPI_RECEIVED:
5818 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5819 if (kvm_cpu_has_pending_timer(vcpu))
5820 kvm_inject_pending_timer_irqs(vcpu);
5822 if (dm_request_for_irq_injection(vcpu)) {
5824 vcpu->run->exit_reason = KVM_EXIT_INTR;
5825 ++vcpu->stat.request_irq_exits;
5828 kvm_check_async_pf_completion(vcpu);
5830 if (signal_pending(current)) {
5832 vcpu->run->exit_reason = KVM_EXIT_INTR;
5833 ++vcpu->stat.signal_exits;
5835 if (need_resched()) {
5836 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5838 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5842 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5849 static int complete_mmio(struct kvm_vcpu *vcpu)
5851 struct kvm_run *run = vcpu->run;
5854 if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5857 if (vcpu->mmio_needed) {
5858 vcpu->mmio_needed = 0;
5859 if (!vcpu->mmio_is_write)
5860 memcpy(vcpu->mmio_data + vcpu->mmio_index,
5862 vcpu->mmio_index += 8;
5863 if (vcpu->mmio_index < vcpu->mmio_size) {
5864 run->exit_reason = KVM_EXIT_MMIO;
5865 run->mmio.phys_addr = vcpu->mmio_phys_addr + vcpu->mmio_index;
5866 memcpy(run->mmio.data, vcpu->mmio_data + vcpu->mmio_index, 8);
5867 run->mmio.len = min(vcpu->mmio_size - vcpu->mmio_index, 8);
5868 run->mmio.is_write = vcpu->mmio_is_write;
5869 vcpu->mmio_needed = 1;
5872 if (vcpu->mmio_is_write)
5874 vcpu->mmio_read_completed = 1;
5876 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5877 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5878 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5879 if (r != EMULATE_DONE)
5884 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5889 if (!tsk_used_math(current) && init_fpu(current))
5892 if (vcpu->sigset_active)
5893 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5895 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5896 kvm_vcpu_block(vcpu);
5897 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5902 /* re-sync apic's tpr */
5903 if (!irqchip_in_kernel(vcpu->kvm)) {
5904 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5910 r = complete_mmio(vcpu);
5914 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5915 kvm_register_write(vcpu, VCPU_REGS_RAX,
5916 kvm_run->hypercall.ret);
5918 r = __vcpu_run(vcpu);
5921 post_kvm_run_save(vcpu);
5922 if (vcpu->sigset_active)
5923 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5928 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5930 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5932 * We are here if userspace calls get_regs() in the middle of
5933 * instruction emulation. Registers state needs to be copied
5934 * back from emulation context to vcpu. Usrapace shouldn't do
5935 * that usually, but some bad designed PV devices (vmware
5936 * backdoor interface) need this to work
5938 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5939 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
5940 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5942 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5943 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5944 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5945 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5946 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5947 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5948 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5949 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5950 #ifdef CONFIG_X86_64
5951 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5952 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5953 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5954 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5955 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5956 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5957 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5958 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5961 regs->rip = kvm_rip_read(vcpu);
5962 regs->rflags = kvm_get_rflags(vcpu);
5967 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5969 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
5970 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5972 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5973 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5974 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5975 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5976 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5977 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5978 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5979 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5980 #ifdef CONFIG_X86_64
5981 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5982 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5983 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5984 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5985 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5986 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5987 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5988 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5991 kvm_rip_write(vcpu, regs->rip);
5992 kvm_set_rflags(vcpu, regs->rflags);
5994 vcpu->arch.exception.pending = false;
5996 kvm_make_request(KVM_REQ_EVENT, vcpu);
6001 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6003 struct kvm_segment cs;
6005 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6009 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6011 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6012 struct kvm_sregs *sregs)
6016 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6017 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6018 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6019 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6020 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6021 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6023 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6024 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6026 kvm_x86_ops->get_idt(vcpu, &dt);
6027 sregs->idt.limit = dt.size;
6028 sregs->idt.base = dt.address;
6029 kvm_x86_ops->get_gdt(vcpu, &dt);
6030 sregs->gdt.limit = dt.size;
6031 sregs->gdt.base = dt.address;
6033 sregs->cr0 = kvm_read_cr0(vcpu);
6034 sregs->cr2 = vcpu->arch.cr2;
6035 sregs->cr3 = kvm_read_cr3(vcpu);
6036 sregs->cr4 = kvm_read_cr4(vcpu);
6037 sregs->cr8 = kvm_get_cr8(vcpu);
6038 sregs->efer = vcpu->arch.efer;
6039 sregs->apic_base = kvm_get_apic_base(vcpu);
6041 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6043 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6044 set_bit(vcpu->arch.interrupt.nr,
6045 (unsigned long *)sregs->interrupt_bitmap);
6050 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6051 struct kvm_mp_state *mp_state)
6053 mp_state->mp_state = vcpu->arch.mp_state;
6057 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6058 struct kvm_mp_state *mp_state)
6060 vcpu->arch.mp_state = mp_state->mp_state;
6061 kvm_make_request(KVM_REQ_EVENT, vcpu);
6065 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
6066 bool has_error_code, u32 error_code)
6068 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6071 init_emulate_ctxt(vcpu);
6073 ret = emulator_task_switch(ctxt, tss_selector, reason,
6074 has_error_code, error_code);
6077 return EMULATE_FAIL;
6079 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
6080 kvm_rip_write(vcpu, ctxt->eip);
6081 kvm_set_rflags(vcpu, ctxt->eflags);
6082 kvm_make_request(KVM_REQ_EVENT, vcpu);
6083 return EMULATE_DONE;
6085 EXPORT_SYMBOL_GPL(kvm_task_switch);
6087 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6088 struct kvm_sregs *sregs)
6090 int mmu_reset_needed = 0;
6091 int pending_vec, max_bits, idx;
6094 dt.size = sregs->idt.limit;
6095 dt.address = sregs->idt.base;
6096 kvm_x86_ops->set_idt(vcpu, &dt);
6097 dt.size = sregs->gdt.limit;
6098 dt.address = sregs->gdt.base;
6099 kvm_x86_ops->set_gdt(vcpu, &dt);
6101 vcpu->arch.cr2 = sregs->cr2;
6102 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6103 vcpu->arch.cr3 = sregs->cr3;
6104 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6106 kvm_set_cr8(vcpu, sregs->cr8);
6108 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6109 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6110 kvm_set_apic_base(vcpu, sregs->apic_base);
6112 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6113 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6114 vcpu->arch.cr0 = sregs->cr0;
6116 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6117 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6118 if (sregs->cr4 & X86_CR4_OSXSAVE)
6121 idx = srcu_read_lock(&vcpu->kvm->srcu);
6122 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6123 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6124 mmu_reset_needed = 1;
6126 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6128 if (mmu_reset_needed)
6129 kvm_mmu_reset_context(vcpu);
6131 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
6132 pending_vec = find_first_bit(
6133 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6134 if (pending_vec < max_bits) {
6135 kvm_queue_interrupt(vcpu, pending_vec, false);
6136 pr_debug("Set back pending irq %d\n", pending_vec);
6139 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6140 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6141 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6142 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6143 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6144 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6146 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6147 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6149 update_cr8_intercept(vcpu);
6151 /* Older userspace won't unhalt the vcpu on reset. */
6152 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6153 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6155 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6157 kvm_make_request(KVM_REQ_EVENT, vcpu);
6162 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6163 struct kvm_guest_debug *dbg)
6165 unsigned long rflags;
6168 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6170 if (vcpu->arch.exception.pending)
6172 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6173 kvm_queue_exception(vcpu, DB_VECTOR);
6175 kvm_queue_exception(vcpu, BP_VECTOR);
6179 * Read rflags as long as potentially injected trace flags are still
6182 rflags = kvm_get_rflags(vcpu);
6184 vcpu->guest_debug = dbg->control;
6185 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6186 vcpu->guest_debug = 0;
6188 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6189 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6190 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6191 vcpu->arch.switch_db_regs =
6192 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
6194 for (i = 0; i < KVM_NR_DB_REGS; i++)
6195 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6196 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
6199 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6200 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6201 get_segment_base(vcpu, VCPU_SREG_CS);
6204 * Trigger an rflags update that will inject or remove the trace
6207 kvm_set_rflags(vcpu, rflags);
6209 kvm_x86_ops->set_guest_debug(vcpu, dbg);
6219 * Translate a guest virtual address to a guest physical address.
6221 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6222 struct kvm_translation *tr)
6224 unsigned long vaddr = tr->linear_address;
6228 idx = srcu_read_lock(&vcpu->kvm->srcu);
6229 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6230 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6231 tr->physical_address = gpa;
6232 tr->valid = gpa != UNMAPPED_GVA;
6239 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6241 struct i387_fxsave_struct *fxsave =
6242 &vcpu->arch.guest_fpu.state->fxsave;
6244 memcpy(fpu->fpr, fxsave->st_space, 128);
6245 fpu->fcw = fxsave->cwd;
6246 fpu->fsw = fxsave->swd;
6247 fpu->ftwx = fxsave->twd;
6248 fpu->last_opcode = fxsave->fop;
6249 fpu->last_ip = fxsave->rip;
6250 fpu->last_dp = fxsave->rdp;
6251 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6256 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6258 struct i387_fxsave_struct *fxsave =
6259 &vcpu->arch.guest_fpu.state->fxsave;
6261 memcpy(fxsave->st_space, fpu->fpr, 128);
6262 fxsave->cwd = fpu->fcw;
6263 fxsave->swd = fpu->fsw;
6264 fxsave->twd = fpu->ftwx;
6265 fxsave->fop = fpu->last_opcode;
6266 fxsave->rip = fpu->last_ip;
6267 fxsave->rdp = fpu->last_dp;
6268 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6273 int fx_init(struct kvm_vcpu *vcpu)
6277 err = fpu_alloc(&vcpu->arch.guest_fpu);
6281 fpu_finit(&vcpu->arch.guest_fpu);
6284 * Ensure guest xcr0 is valid for loading
6286 vcpu->arch.xcr0 = XSTATE_FP;
6288 vcpu->arch.cr0 |= X86_CR0_ET;
6292 EXPORT_SYMBOL_GPL(fx_init);
6294 static void fx_free(struct kvm_vcpu *vcpu)
6296 fpu_free(&vcpu->arch.guest_fpu);
6299 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6301 if (vcpu->guest_fpu_loaded)
6305 * Restore all possible states in the guest,
6306 * and assume host would use all available bits.
6307 * Guest xcr0 would be loaded later.
6309 kvm_put_guest_xcr0(vcpu);
6310 vcpu->guest_fpu_loaded = 1;
6311 unlazy_fpu(current);
6312 fpu_restore_checking(&vcpu->arch.guest_fpu);
6316 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6318 kvm_put_guest_xcr0(vcpu);
6320 if (!vcpu->guest_fpu_loaded)
6323 vcpu->guest_fpu_loaded = 0;
6324 fpu_save_init(&vcpu->arch.guest_fpu);
6325 ++vcpu->stat.fpu_reload;
6326 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6330 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6332 kvmclock_reset(vcpu);
6334 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6336 kvm_x86_ops->vcpu_free(vcpu);
6339 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6342 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6343 printk_once(KERN_WARNING
6344 "kvm: SMP vm created on host with unstable TSC; "
6345 "guest TSC will not be reliable\n");
6346 return kvm_x86_ops->vcpu_create(kvm, id);
6349 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6353 vcpu->arch.mtrr_state.have_fixed = 1;
6355 r = kvm_arch_vcpu_reset(vcpu);
6357 r = kvm_mmu_setup(vcpu);
6363 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6365 vcpu->arch.apf.msr_val = 0;
6368 kvm_mmu_unload(vcpu);
6372 kvm_x86_ops->vcpu_free(vcpu);
6375 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
6377 vcpu->arch.nmi_pending = false;
6378 vcpu->arch.nmi_injected = false;
6380 vcpu->arch.switch_db_regs = 0;
6381 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6382 vcpu->arch.dr6 = DR6_FIXED_1;
6383 vcpu->arch.dr7 = DR7_FIXED_1;
6385 kvm_make_request(KVM_REQ_EVENT, vcpu);
6386 vcpu->arch.apf.msr_val = 0;
6387 vcpu->arch.st.msr_val = 0;
6389 kvmclock_reset(vcpu);
6391 kvm_clear_async_pf_completion_queue(vcpu);
6392 kvm_async_pf_hash_reset(vcpu);
6393 vcpu->arch.apf.halted = false;
6395 return kvm_x86_ops->vcpu_reset(vcpu);
6398 int kvm_arch_hardware_enable(void *garbage)
6401 struct kvm_vcpu *vcpu;
6404 kvm_shared_msr_cpu_online();
6405 list_for_each_entry(kvm, &vm_list, vm_list)
6406 kvm_for_each_vcpu(i, vcpu, kvm)
6407 if (vcpu->cpu == smp_processor_id())
6408 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6409 return kvm_x86_ops->hardware_enable(garbage);
6412 void kvm_arch_hardware_disable(void *garbage)
6414 kvm_x86_ops->hardware_disable(garbage);
6415 drop_user_return_notifiers(garbage);
6418 int kvm_arch_hardware_setup(void)
6420 return kvm_x86_ops->hardware_setup();
6423 void kvm_arch_hardware_unsetup(void)
6425 kvm_x86_ops->hardware_unsetup();
6428 void kvm_arch_check_processor_compat(void *rtn)
6430 kvm_x86_ops->check_processor_compatibility(rtn);
6433 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6439 BUG_ON(vcpu->kvm == NULL);
6442 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6443 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
6444 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
6445 vcpu->arch.mmu.translate_gpa = translate_gpa;
6446 vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
6447 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6448 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6450 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6452 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6457 vcpu->arch.pio_data = page_address(page);
6459 kvm_init_tsc_catchup(vcpu, max_tsc_khz);
6461 r = kvm_mmu_create(vcpu);
6463 goto fail_free_pio_data;
6465 if (irqchip_in_kernel(kvm)) {
6466 r = kvm_create_lapic(vcpu);
6468 goto fail_mmu_destroy;
6471 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6473 if (!vcpu->arch.mce_banks) {
6475 goto fail_free_lapic;
6477 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6479 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6480 goto fail_free_mce_banks;
6482 kvm_async_pf_hash_reset(vcpu);
6485 fail_free_mce_banks:
6486 kfree(vcpu->arch.mce_banks);
6488 kvm_free_lapic(vcpu);
6490 kvm_mmu_destroy(vcpu);
6492 free_page((unsigned long)vcpu->arch.pio_data);
6497 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6501 kfree(vcpu->arch.mce_banks);
6502 kvm_free_lapic(vcpu);
6503 idx = srcu_read_lock(&vcpu->kvm->srcu);
6504 kvm_mmu_destroy(vcpu);
6505 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6506 free_page((unsigned long)vcpu->arch.pio_data);
6509 int kvm_arch_init_vm(struct kvm *kvm)
6511 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6512 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6514 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6515 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6517 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6522 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6525 kvm_mmu_unload(vcpu);
6529 static void kvm_free_vcpus(struct kvm *kvm)
6532 struct kvm_vcpu *vcpu;
6535 * Unpin any mmu pages first.
6537 kvm_for_each_vcpu(i, vcpu, kvm) {
6538 kvm_clear_async_pf_completion_queue(vcpu);
6539 kvm_unload_vcpu_mmu(vcpu);
6541 kvm_for_each_vcpu(i, vcpu, kvm)
6542 kvm_arch_vcpu_free(vcpu);
6544 mutex_lock(&kvm->lock);
6545 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6546 kvm->vcpus[i] = NULL;
6548 atomic_set(&kvm->online_vcpus, 0);
6549 mutex_unlock(&kvm->lock);
6552 void kvm_arch_sync_events(struct kvm *kvm)
6554 kvm_free_all_assigned_devices(kvm);
6558 void kvm_arch_destroy_vm(struct kvm *kvm)
6560 kvm_iommu_unmap_guest(kvm);
6561 kfree(kvm->arch.vpic);
6562 kfree(kvm->arch.vioapic);
6563 kvm_free_vcpus(kvm);
6564 if (kvm->arch.apic_access_page)
6565 put_page(kvm->arch.apic_access_page);
6566 if (kvm->arch.ept_identity_pagetable)
6567 put_page(kvm->arch.ept_identity_pagetable);
6570 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6571 struct kvm_memory_slot *memslot,
6572 struct kvm_memory_slot old,
6573 struct kvm_userspace_memory_region *mem,
6576 int npages = memslot->npages;
6577 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6579 /* Prevent internal slot pages from being moved by fork()/COW. */
6580 if (memslot->id >= KVM_MEMORY_SLOTS)
6581 map_flags = MAP_SHARED | MAP_ANONYMOUS;
6583 /*To keep backward compatibility with older userspace,
6584 *x86 needs to hanlde !user_alloc case.
6587 if (npages && !old.rmap) {
6588 unsigned long userspace_addr;
6590 down_write(¤t->mm->mmap_sem);
6591 userspace_addr = do_mmap(NULL, 0,
6593 PROT_READ | PROT_WRITE,
6596 up_write(¤t->mm->mmap_sem);
6598 if (IS_ERR((void *)userspace_addr))
6599 return PTR_ERR((void *)userspace_addr);
6601 memslot->userspace_addr = userspace_addr;
6609 void kvm_arch_commit_memory_region(struct kvm *kvm,
6610 struct kvm_userspace_memory_region *mem,
6611 struct kvm_memory_slot old,
6615 int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
6617 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6620 down_write(¤t->mm->mmap_sem);
6621 ret = do_munmap(current->mm, old.userspace_addr,
6622 old.npages * PAGE_SIZE);
6623 up_write(¤t->mm->mmap_sem);
6626 "kvm_vm_ioctl_set_memory_region: "
6627 "failed to munmap memory\n");
6630 if (!kvm->arch.n_requested_mmu_pages)
6631 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6633 spin_lock(&kvm->mmu_lock);
6635 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6636 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6637 spin_unlock(&kvm->mmu_lock);
6640 void kvm_arch_flush_shadow(struct kvm *kvm)
6642 kvm_mmu_zap_all(kvm);
6643 kvm_reload_remote_mmus(kvm);
6646 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6648 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6649 !vcpu->arch.apf.halted)
6650 || !list_empty_careful(&vcpu->async_pf.done)
6651 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6652 || vcpu->arch.nmi_pending ||
6653 (kvm_arch_interrupt_allowed(vcpu) &&
6654 kvm_cpu_has_interrupt(vcpu));
6657 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6660 int cpu = vcpu->cpu;
6662 if (waitqueue_active(&vcpu->wq)) {
6663 wake_up_interruptible(&vcpu->wq);
6664 ++vcpu->stat.halt_wakeup;
6668 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6669 if (kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE)
6670 smp_send_reschedule(cpu);
6674 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6676 return kvm_x86_ops->interrupt_allowed(vcpu);
6679 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6681 unsigned long current_rip = kvm_rip_read(vcpu) +
6682 get_segment_base(vcpu, VCPU_SREG_CS);
6684 return current_rip == linear_rip;
6686 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6688 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6690 unsigned long rflags;
6692 rflags = kvm_x86_ops->get_rflags(vcpu);
6693 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6694 rflags &= ~X86_EFLAGS_TF;
6697 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6699 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6701 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6702 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6703 rflags |= X86_EFLAGS_TF;
6704 kvm_x86_ops->set_rflags(vcpu, rflags);
6705 kvm_make_request(KVM_REQ_EVENT, vcpu);
6707 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6709 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6713 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6714 is_error_page(work->page))
6717 r = kvm_mmu_reload(vcpu);
6721 if (!vcpu->arch.mmu.direct_map &&
6722 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6725 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6728 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6730 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6733 static inline u32 kvm_async_pf_next_probe(u32 key)
6735 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6738 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6740 u32 key = kvm_async_pf_hash_fn(gfn);
6742 while (vcpu->arch.apf.gfns[key] != ~0)
6743 key = kvm_async_pf_next_probe(key);
6745 vcpu->arch.apf.gfns[key] = gfn;
6748 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6751 u32 key = kvm_async_pf_hash_fn(gfn);
6753 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6754 (vcpu->arch.apf.gfns[key] != gfn &&
6755 vcpu->arch.apf.gfns[key] != ~0); i++)
6756 key = kvm_async_pf_next_probe(key);
6761 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6763 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6766 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6770 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6772 vcpu->arch.apf.gfns[i] = ~0;
6774 j = kvm_async_pf_next_probe(j);
6775 if (vcpu->arch.apf.gfns[j] == ~0)
6777 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6779 * k lies cyclically in ]i,j]
6781 * |....j i.k.| or |.k..j i...|
6783 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6784 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6789 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6792 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6796 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6797 struct kvm_async_pf *work)
6799 struct x86_exception fault;
6801 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6802 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6804 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6805 (vcpu->arch.apf.send_user_only &&
6806 kvm_x86_ops->get_cpl(vcpu) == 0))
6807 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6808 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6809 fault.vector = PF_VECTOR;
6810 fault.error_code_valid = true;
6811 fault.error_code = 0;
6812 fault.nested_page_fault = false;
6813 fault.address = work->arch.token;
6814 kvm_inject_page_fault(vcpu, &fault);
6818 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6819 struct kvm_async_pf *work)
6821 struct x86_exception fault;
6823 trace_kvm_async_pf_ready(work->arch.token, work->gva);
6824 if (is_error_page(work->page))
6825 work->arch.token = ~0; /* broadcast wakeup */
6827 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6829 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6830 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6831 fault.vector = PF_VECTOR;
6832 fault.error_code_valid = true;
6833 fault.error_code = 0;
6834 fault.nested_page_fault = false;
6835 fault.address = work->arch.token;
6836 kvm_inject_page_fault(vcpu, &fault);
6838 vcpu->arch.apf.halted = false;
6841 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6843 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6846 return !kvm_event_needs_reinjection(vcpu) &&
6847 kvm_x86_ops->interrupt_allowed(vcpu);
6850 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6851 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6852 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6853 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6854 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6855 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6856 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6857 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6858 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6859 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6860 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6861 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);