2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
14 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
22 #include <linux/kvm_host.h>
27 #include "kvm_cache_regs.h"
30 #include <linux/clocksource.h>
31 #include <linux/interrupt.h>
32 #include <linux/kvm.h>
34 #include <linux/vmalloc.h>
35 #include <linux/module.h>
36 #include <linux/mman.h>
37 #include <linux/highmem.h>
38 #include <linux/iommu.h>
39 #include <linux/intel-iommu.h>
40 #include <linux/cpufreq.h>
41 #include <linux/user-return-notifier.h>
42 #include <linux/srcu.h>
43 #include <linux/slab.h>
44 #include <linux/perf_event.h>
45 #include <linux/uaccess.h>
46 #include <linux/hash.h>
47 #include <linux/pci.h>
48 #include <trace/events/kvm.h>
50 #define CREATE_TRACE_POINTS
53 #include <asm/debugreg.h>
60 #include <asm/pvclock.h>
61 #include <asm/div64.h>
63 #define MAX_IO_MSRS 256
64 #define KVM_MAX_MCE_BANKS 32
65 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
67 #define emul_to_vcpu(ctxt) \
68 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
71 * - enable syscall per default because its emulated by KVM
72 * - enable LME and LMA per default on 64 bit KVM
76 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
78 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
81 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
82 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
84 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
85 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
86 struct kvm_cpuid_entry2 __user *entries);
87 static void process_nmi(struct kvm_vcpu *vcpu);
89 struct kvm_x86_ops *kvm_x86_ops;
90 EXPORT_SYMBOL_GPL(kvm_x86_ops);
93 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
95 bool kvm_has_tsc_control;
96 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
97 u32 kvm_max_guest_tsc_khz;
98 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
100 #define KVM_NR_SHARED_MSRS 16
102 struct kvm_shared_msrs_global {
104 u32 msrs[KVM_NR_SHARED_MSRS];
107 struct kvm_shared_msrs {
108 struct user_return_notifier urn;
110 struct kvm_shared_msr_values {
113 } values[KVM_NR_SHARED_MSRS];
116 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
117 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
119 struct kvm_stats_debugfs_item debugfs_entries[] = {
120 { "pf_fixed", VCPU_STAT(pf_fixed) },
121 { "pf_guest", VCPU_STAT(pf_guest) },
122 { "tlb_flush", VCPU_STAT(tlb_flush) },
123 { "invlpg", VCPU_STAT(invlpg) },
124 { "exits", VCPU_STAT(exits) },
125 { "io_exits", VCPU_STAT(io_exits) },
126 { "mmio_exits", VCPU_STAT(mmio_exits) },
127 { "signal_exits", VCPU_STAT(signal_exits) },
128 { "irq_window", VCPU_STAT(irq_window_exits) },
129 { "nmi_window", VCPU_STAT(nmi_window_exits) },
130 { "halt_exits", VCPU_STAT(halt_exits) },
131 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
132 { "hypercalls", VCPU_STAT(hypercalls) },
133 { "request_irq", VCPU_STAT(request_irq_exits) },
134 { "irq_exits", VCPU_STAT(irq_exits) },
135 { "host_state_reload", VCPU_STAT(host_state_reload) },
136 { "efer_reload", VCPU_STAT(efer_reload) },
137 { "fpu_reload", VCPU_STAT(fpu_reload) },
138 { "insn_emulation", VCPU_STAT(insn_emulation) },
139 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
140 { "irq_injections", VCPU_STAT(irq_injections) },
141 { "nmi_injections", VCPU_STAT(nmi_injections) },
142 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
143 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
144 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
145 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
146 { "mmu_flooded", VM_STAT(mmu_flooded) },
147 { "mmu_recycled", VM_STAT(mmu_recycled) },
148 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
149 { "mmu_unsync", VM_STAT(mmu_unsync) },
150 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
151 { "largepages", VM_STAT(lpages) },
155 u64 __read_mostly host_xcr0;
157 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
159 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
162 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
163 vcpu->arch.apf.gfns[i] = ~0;
166 static void kvm_on_user_return(struct user_return_notifier *urn)
169 struct kvm_shared_msrs *locals
170 = container_of(urn, struct kvm_shared_msrs, urn);
171 struct kvm_shared_msr_values *values;
173 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
174 values = &locals->values[slot];
175 if (values->host != values->curr) {
176 wrmsrl(shared_msrs_global.msrs[slot], values->host);
177 values->curr = values->host;
180 locals->registered = false;
181 user_return_notifier_unregister(urn);
184 static void shared_msr_update(unsigned slot, u32 msr)
186 struct kvm_shared_msrs *smsr;
189 smsr = &__get_cpu_var(shared_msrs);
190 /* only read, and nobody should modify it at this time,
191 * so don't need lock */
192 if (slot >= shared_msrs_global.nr) {
193 printk(KERN_ERR "kvm: invalid MSR slot!");
196 rdmsrl_safe(msr, &value);
197 smsr->values[slot].host = value;
198 smsr->values[slot].curr = value;
201 void kvm_define_shared_msr(unsigned slot, u32 msr)
203 if (slot >= shared_msrs_global.nr)
204 shared_msrs_global.nr = slot + 1;
205 shared_msrs_global.msrs[slot] = msr;
206 /* we need ensured the shared_msr_global have been updated */
209 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
211 static void kvm_shared_msr_cpu_online(void)
215 for (i = 0; i < shared_msrs_global.nr; ++i)
216 shared_msr_update(i, shared_msrs_global.msrs[i]);
219 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
221 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
223 if (((value ^ smsr->values[slot].curr) & mask) == 0)
225 smsr->values[slot].curr = value;
226 wrmsrl(shared_msrs_global.msrs[slot], value);
227 if (!smsr->registered) {
228 smsr->urn.on_user_return = kvm_on_user_return;
229 user_return_notifier_register(&smsr->urn);
230 smsr->registered = true;
233 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
235 static void drop_user_return_notifiers(void *ignore)
237 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
239 if (smsr->registered)
240 kvm_on_user_return(&smsr->urn);
243 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
245 if (irqchip_in_kernel(vcpu->kvm))
246 return vcpu->arch.apic_base;
248 return vcpu->arch.apic_base;
250 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
252 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
254 /* TODO: reserve bits check */
255 if (irqchip_in_kernel(vcpu->kvm))
256 kvm_lapic_set_base(vcpu, data);
258 vcpu->arch.apic_base = data;
260 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
262 #define EXCPT_BENIGN 0
263 #define EXCPT_CONTRIBUTORY 1
266 static int exception_class(int vector)
276 return EXCPT_CONTRIBUTORY;
283 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
284 unsigned nr, bool has_error, u32 error_code,
290 kvm_make_request(KVM_REQ_EVENT, vcpu);
292 if (!vcpu->arch.exception.pending) {
294 vcpu->arch.exception.pending = true;
295 vcpu->arch.exception.has_error_code = has_error;
296 vcpu->arch.exception.nr = nr;
297 vcpu->arch.exception.error_code = error_code;
298 vcpu->arch.exception.reinject = reinject;
302 /* to check exception */
303 prev_nr = vcpu->arch.exception.nr;
304 if (prev_nr == DF_VECTOR) {
305 /* triple fault -> shutdown */
306 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
309 class1 = exception_class(prev_nr);
310 class2 = exception_class(nr);
311 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
312 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
313 /* generate double fault per SDM Table 5-5 */
314 vcpu->arch.exception.pending = true;
315 vcpu->arch.exception.has_error_code = true;
316 vcpu->arch.exception.nr = DF_VECTOR;
317 vcpu->arch.exception.error_code = 0;
319 /* replace previous exception with a new one in a hope
320 that instruction re-execution will regenerate lost
325 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
327 kvm_multiple_exception(vcpu, nr, false, 0, false);
329 EXPORT_SYMBOL_GPL(kvm_queue_exception);
331 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
333 kvm_multiple_exception(vcpu, nr, false, 0, true);
335 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
337 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
340 kvm_inject_gp(vcpu, 0);
342 kvm_x86_ops->skip_emulated_instruction(vcpu);
344 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
346 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
348 ++vcpu->stat.pf_guest;
349 vcpu->arch.cr2 = fault->address;
350 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
352 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
354 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
356 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
357 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
359 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
362 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
364 atomic_inc(&vcpu->arch.nmi_queued);
365 kvm_make_request(KVM_REQ_NMI, vcpu);
367 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
369 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
371 kvm_multiple_exception(vcpu, nr, true, error_code, false);
373 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
375 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
377 kvm_multiple_exception(vcpu, nr, true, error_code, true);
379 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
382 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
383 * a #GP and return false.
385 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
387 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
389 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
392 EXPORT_SYMBOL_GPL(kvm_require_cpl);
395 * This function will be used to read from the physical memory of the currently
396 * running guest. The difference to kvm_read_guest_page is that this function
397 * can read from guest physical or from the guest's guest physical memory.
399 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
400 gfn_t ngfn, void *data, int offset, int len,
406 ngpa = gfn_to_gpa(ngfn);
407 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
408 if (real_gfn == UNMAPPED_GVA)
411 real_gfn = gpa_to_gfn(real_gfn);
413 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
415 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
417 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
418 void *data, int offset, int len, u32 access)
420 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
421 data, offset, len, access);
425 * Load the pae pdptrs. Return true is they are all valid.
427 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
429 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
430 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
433 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
435 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
436 offset * sizeof(u64), sizeof(pdpte),
437 PFERR_USER_MASK|PFERR_WRITE_MASK);
442 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
443 if (is_present_gpte(pdpte[i]) &&
444 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
451 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
452 __set_bit(VCPU_EXREG_PDPTR,
453 (unsigned long *)&vcpu->arch.regs_avail);
454 __set_bit(VCPU_EXREG_PDPTR,
455 (unsigned long *)&vcpu->arch.regs_dirty);
460 EXPORT_SYMBOL_GPL(load_pdptrs);
462 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
464 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
470 if (is_long_mode(vcpu) || !is_pae(vcpu))
473 if (!test_bit(VCPU_EXREG_PDPTR,
474 (unsigned long *)&vcpu->arch.regs_avail))
477 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
478 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
479 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
480 PFERR_USER_MASK | PFERR_WRITE_MASK);
483 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
489 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
491 unsigned long old_cr0 = kvm_read_cr0(vcpu);
492 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
493 X86_CR0_CD | X86_CR0_NW;
498 if (cr0 & 0xffffffff00000000UL)
502 cr0 &= ~CR0_RESERVED_BITS;
504 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
507 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
510 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
512 if ((vcpu->arch.efer & EFER_LME)) {
517 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
522 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
527 kvm_x86_ops->set_cr0(vcpu, cr0);
529 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
530 kvm_clear_async_pf_completion_queue(vcpu);
531 kvm_async_pf_hash_reset(vcpu);
534 if ((cr0 ^ old_cr0) & update_bits)
535 kvm_mmu_reset_context(vcpu);
538 EXPORT_SYMBOL_GPL(kvm_set_cr0);
540 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
542 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
544 EXPORT_SYMBOL_GPL(kvm_lmsw);
546 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
550 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
551 if (index != XCR_XFEATURE_ENABLED_MASK)
554 if (kvm_x86_ops->get_cpl(vcpu) != 0)
556 if (!(xcr0 & XSTATE_FP))
558 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
560 if (xcr0 & ~host_xcr0)
562 vcpu->arch.xcr0 = xcr0;
563 vcpu->guest_xcr0_loaded = 0;
567 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
569 if (__kvm_set_xcr(vcpu, index, xcr)) {
570 kvm_inject_gp(vcpu, 0);
575 EXPORT_SYMBOL_GPL(kvm_set_xcr);
577 static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
579 struct kvm_cpuid_entry2 *best;
581 if (!static_cpu_has(X86_FEATURE_XSAVE))
584 best = kvm_find_cpuid_entry(vcpu, 1, 0);
585 return best && (best->ecx & bit(X86_FEATURE_XSAVE));
588 static bool guest_cpuid_has_smep(struct kvm_vcpu *vcpu)
590 struct kvm_cpuid_entry2 *best;
592 best = kvm_find_cpuid_entry(vcpu, 7, 0);
593 return best && (best->ebx & bit(X86_FEATURE_SMEP));
596 static bool guest_cpuid_has_fsgsbase(struct kvm_vcpu *vcpu)
598 struct kvm_cpuid_entry2 *best;
600 best = kvm_find_cpuid_entry(vcpu, 7, 0);
601 return best && (best->ebx & bit(X86_FEATURE_FSGSBASE));
604 static void update_cpuid(struct kvm_vcpu *vcpu)
606 struct kvm_cpuid_entry2 *best;
607 struct kvm_lapic *apic = vcpu->arch.apic;
609 best = kvm_find_cpuid_entry(vcpu, 1, 0);
613 /* Update OSXSAVE bit */
614 if (cpu_has_xsave && best->function == 0x1) {
615 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
616 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
617 best->ecx |= bit(X86_FEATURE_OSXSAVE);
621 if (best->ecx & bit(X86_FEATURE_TSC_DEADLINE_TIMER))
622 apic->lapic_timer.timer_mode_mask = 3 << 17;
624 apic->lapic_timer.timer_mode_mask = 1 << 17;
628 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
630 unsigned long old_cr4 = kvm_read_cr4(vcpu);
631 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
632 X86_CR4_PAE | X86_CR4_SMEP;
633 if (cr4 & CR4_RESERVED_BITS)
636 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
639 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
642 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_RDWRGSFS))
645 if (is_long_mode(vcpu)) {
646 if (!(cr4 & X86_CR4_PAE))
648 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
649 && ((cr4 ^ old_cr4) & pdptr_bits)
650 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
654 if (kvm_x86_ops->set_cr4(vcpu, cr4))
657 if ((cr4 ^ old_cr4) & pdptr_bits)
658 kvm_mmu_reset_context(vcpu);
660 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
665 EXPORT_SYMBOL_GPL(kvm_set_cr4);
667 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
669 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
670 kvm_mmu_sync_roots(vcpu);
671 kvm_mmu_flush_tlb(vcpu);
675 if (is_long_mode(vcpu)) {
676 if (cr3 & CR3_L_MODE_RESERVED_BITS)
680 if (cr3 & CR3_PAE_RESERVED_BITS)
682 if (is_paging(vcpu) &&
683 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
687 * We don't check reserved bits in nonpae mode, because
688 * this isn't enforced, and VMware depends on this.
693 * Does the new cr3 value map to physical memory? (Note, we
694 * catch an invalid cr3 even in real-mode, because it would
695 * cause trouble later on when we turn on paging anyway.)
697 * A real CPU would silently accept an invalid cr3 and would
698 * attempt to use it - with largely undefined (and often hard
699 * to debug) behavior on the guest side.
701 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
703 vcpu->arch.cr3 = cr3;
704 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
705 vcpu->arch.mmu.new_cr3(vcpu);
708 EXPORT_SYMBOL_GPL(kvm_set_cr3);
710 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
712 if (cr8 & CR8_RESERVED_BITS)
714 if (irqchip_in_kernel(vcpu->kvm))
715 kvm_lapic_set_tpr(vcpu, cr8);
717 vcpu->arch.cr8 = cr8;
720 EXPORT_SYMBOL_GPL(kvm_set_cr8);
722 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
724 if (irqchip_in_kernel(vcpu->kvm))
725 return kvm_lapic_get_cr8(vcpu);
727 return vcpu->arch.cr8;
729 EXPORT_SYMBOL_GPL(kvm_get_cr8);
731 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
735 vcpu->arch.db[dr] = val;
736 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
737 vcpu->arch.eff_db[dr] = val;
740 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
744 if (val & 0xffffffff00000000ULL)
746 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
749 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
753 if (val & 0xffffffff00000000ULL)
755 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
756 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
757 kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
758 vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
766 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
770 res = __kvm_set_dr(vcpu, dr, val);
772 kvm_queue_exception(vcpu, UD_VECTOR);
774 kvm_inject_gp(vcpu, 0);
778 EXPORT_SYMBOL_GPL(kvm_set_dr);
780 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
784 *val = vcpu->arch.db[dr];
787 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
791 *val = vcpu->arch.dr6;
794 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
798 *val = vcpu->arch.dr7;
805 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
807 if (_kvm_get_dr(vcpu, dr, val)) {
808 kvm_queue_exception(vcpu, UD_VECTOR);
813 EXPORT_SYMBOL_GPL(kvm_get_dr);
816 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
817 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
819 * This list is modified at module load time to reflect the
820 * capabilities of the host cpu. This capabilities test skips MSRs that are
821 * kvm-specific. Those are put in the beginning of the list.
824 #define KVM_SAVE_MSRS_BEGIN 9
825 static u32 msrs_to_save[] = {
826 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
827 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
828 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
829 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
830 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
833 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
835 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
838 static unsigned num_msrs_to_save;
840 static u32 emulated_msrs[] = {
841 MSR_IA32_TSCDEADLINE,
842 MSR_IA32_MISC_ENABLE,
847 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
849 u64 old_efer = vcpu->arch.efer;
851 if (efer & efer_reserved_bits)
855 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
858 if (efer & EFER_FFXSR) {
859 struct kvm_cpuid_entry2 *feat;
861 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
862 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
866 if (efer & EFER_SVME) {
867 struct kvm_cpuid_entry2 *feat;
869 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
870 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
875 efer |= vcpu->arch.efer & EFER_LMA;
877 kvm_x86_ops->set_efer(vcpu, efer);
879 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
881 /* Update reserved bits */
882 if ((efer ^ old_efer) & EFER_NX)
883 kvm_mmu_reset_context(vcpu);
888 void kvm_enable_efer_bits(u64 mask)
890 efer_reserved_bits &= ~mask;
892 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
896 * Writes msr value into into the appropriate "register".
897 * Returns 0 on success, non-0 otherwise.
898 * Assumes vcpu_load() was already called.
900 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
902 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
906 * Adapt set_msr() to msr_io()'s calling convention
908 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
910 return kvm_set_msr(vcpu, index, *data);
913 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
917 struct pvclock_wall_clock wc;
918 struct timespec boot;
923 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
928 ++version; /* first time write, random junk */
932 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
935 * The guest calculates current wall clock time by adding
936 * system time (updated by kvm_guest_time_update below) to the
937 * wall clock specified here. guest system time equals host
938 * system time for us, thus we must fill in host boot time here.
942 wc.sec = boot.tv_sec;
943 wc.nsec = boot.tv_nsec;
944 wc.version = version;
946 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
949 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
952 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
954 uint32_t quotient, remainder;
956 /* Don't try to replace with do_div(), this one calculates
957 * "(dividend << 32) / divisor" */
959 : "=a" (quotient), "=d" (remainder)
960 : "0" (0), "1" (dividend), "r" (divisor) );
964 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
965 s8 *pshift, u32 *pmultiplier)
972 tps64 = base_khz * 1000LL;
973 scaled64 = scaled_khz * 1000LL;
974 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
979 tps32 = (uint32_t)tps64;
980 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
981 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
989 *pmultiplier = div_frac(scaled64, tps32);
991 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
992 __func__, base_khz, scaled_khz, shift, *pmultiplier);
995 static inline u64 get_kernel_ns(void)
999 WARN_ON(preemptible());
1001 monotonic_to_bootbased(&ts);
1002 return timespec_to_ns(&ts);
1005 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1006 unsigned long max_tsc_khz;
1008 static inline int kvm_tsc_changes_freq(void)
1010 int cpu = get_cpu();
1011 int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
1012 cpufreq_quick_get(cpu) != 0;
1017 u64 vcpu_tsc_khz(struct kvm_vcpu *vcpu)
1019 if (vcpu->arch.virtual_tsc_khz)
1020 return vcpu->arch.virtual_tsc_khz;
1022 return __this_cpu_read(cpu_tsc_khz);
1025 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1029 WARN_ON(preemptible());
1030 if (kvm_tsc_changes_freq())
1031 printk_once(KERN_WARNING
1032 "kvm: unreliable cycle conversion on adjustable rate TSC\n");
1033 ret = nsec * vcpu_tsc_khz(vcpu);
1034 do_div(ret, USEC_PER_SEC);
1038 static void kvm_init_tsc_catchup(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1040 /* Compute a scale to convert nanoseconds in TSC cycles */
1041 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1042 &vcpu->arch.tsc_catchup_shift,
1043 &vcpu->arch.tsc_catchup_mult);
1046 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1048 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1049 vcpu->arch.tsc_catchup_mult,
1050 vcpu->arch.tsc_catchup_shift);
1051 tsc += vcpu->arch.last_tsc_write;
1055 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1057 struct kvm *kvm = vcpu->kvm;
1058 u64 offset, ns, elapsed;
1059 unsigned long flags;
1062 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1063 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1064 ns = get_kernel_ns();
1065 elapsed = ns - kvm->arch.last_tsc_nsec;
1066 sdiff = data - kvm->arch.last_tsc_write;
1071 * Special case: close write to TSC within 5 seconds of
1072 * another CPU is interpreted as an attempt to synchronize
1073 * The 5 seconds is to accommodate host load / swapping as
1074 * well as any reset of TSC during the boot process.
1076 * In that case, for a reliable TSC, we can match TSC offsets,
1077 * or make a best guest using elapsed value.
1079 if (sdiff < nsec_to_cycles(vcpu, 5ULL * NSEC_PER_SEC) &&
1080 elapsed < 5ULL * NSEC_PER_SEC) {
1081 if (!check_tsc_unstable()) {
1082 offset = kvm->arch.last_tsc_offset;
1083 pr_debug("kvm: matched tsc offset for %llu\n", data);
1085 u64 delta = nsec_to_cycles(vcpu, elapsed);
1087 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1089 ns = kvm->arch.last_tsc_nsec;
1091 kvm->arch.last_tsc_nsec = ns;
1092 kvm->arch.last_tsc_write = data;
1093 kvm->arch.last_tsc_offset = offset;
1094 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1095 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1097 /* Reset of TSC must disable overshoot protection below */
1098 vcpu->arch.hv_clock.tsc_timestamp = 0;
1099 vcpu->arch.last_tsc_write = data;
1100 vcpu->arch.last_tsc_nsec = ns;
1102 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1104 static int kvm_guest_time_update(struct kvm_vcpu *v)
1106 unsigned long flags;
1107 struct kvm_vcpu_arch *vcpu = &v->arch;
1108 unsigned long this_tsc_khz;
1109 s64 kernel_ns, max_kernel_ns;
1112 /* Keep irq disabled to prevent changes to the clock */
1113 local_irq_save(flags);
1114 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v);
1115 kernel_ns = get_kernel_ns();
1116 this_tsc_khz = vcpu_tsc_khz(v);
1117 if (unlikely(this_tsc_khz == 0)) {
1118 local_irq_restore(flags);
1119 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1124 * We may have to catch up the TSC to match elapsed wall clock
1125 * time for two reasons, even if kvmclock is used.
1126 * 1) CPU could have been running below the maximum TSC rate
1127 * 2) Broken TSC compensation resets the base at each VCPU
1128 * entry to avoid unknown leaps of TSC even when running
1129 * again on the same CPU. This may cause apparent elapsed
1130 * time to disappear, and the guest to stand still or run
1133 if (vcpu->tsc_catchup) {
1134 u64 tsc = compute_guest_tsc(v, kernel_ns);
1135 if (tsc > tsc_timestamp) {
1136 kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1137 tsc_timestamp = tsc;
1141 local_irq_restore(flags);
1143 if (!vcpu->pv_time_enabled)
1147 * Time as measured by the TSC may go backwards when resetting the base
1148 * tsc_timestamp. The reason for this is that the TSC resolution is
1149 * higher than the resolution of the other clock scales. Thus, many
1150 * possible measurments of the TSC correspond to one measurement of any
1151 * other clock, and so a spread of values is possible. This is not a
1152 * problem for the computation of the nanosecond clock; with TSC rates
1153 * around 1GHZ, there can only be a few cycles which correspond to one
1154 * nanosecond value, and any path through this code will inevitably
1155 * take longer than that. However, with the kernel_ns value itself,
1156 * the precision may be much lower, down to HZ granularity. If the
1157 * first sampling of TSC against kernel_ns ends in the low part of the
1158 * range, and the second in the high end of the range, we can get:
1160 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1162 * As the sampling errors potentially range in the thousands of cycles,
1163 * it is possible such a time value has already been observed by the
1164 * guest. To protect against this, we must compute the system time as
1165 * observed by the guest and ensure the new system time is greater.
1168 if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1169 max_kernel_ns = vcpu->last_guest_tsc -
1170 vcpu->hv_clock.tsc_timestamp;
1171 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1172 vcpu->hv_clock.tsc_to_system_mul,
1173 vcpu->hv_clock.tsc_shift);
1174 max_kernel_ns += vcpu->last_kernel_ns;
1177 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1178 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1179 &vcpu->hv_clock.tsc_shift,
1180 &vcpu->hv_clock.tsc_to_system_mul);
1181 vcpu->hw_tsc_khz = this_tsc_khz;
1184 if (max_kernel_ns > kernel_ns)
1185 kernel_ns = max_kernel_ns;
1187 /* With all the info we got, fill in the values */
1188 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1189 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1190 vcpu->last_kernel_ns = kernel_ns;
1191 vcpu->last_guest_tsc = tsc_timestamp;
1192 vcpu->hv_clock.flags = 0;
1195 * The interface expects us to write an even number signaling that the
1196 * update is finished. Since the guest won't see the intermediate
1197 * state, we just increase by 2 at the end.
1199 vcpu->hv_clock.version += 2;
1201 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1203 sizeof(vcpu->hv_clock));
1207 static bool msr_mtrr_valid(unsigned msr)
1210 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1211 case MSR_MTRRfix64K_00000:
1212 case MSR_MTRRfix16K_80000:
1213 case MSR_MTRRfix16K_A0000:
1214 case MSR_MTRRfix4K_C0000:
1215 case MSR_MTRRfix4K_C8000:
1216 case MSR_MTRRfix4K_D0000:
1217 case MSR_MTRRfix4K_D8000:
1218 case MSR_MTRRfix4K_E0000:
1219 case MSR_MTRRfix4K_E8000:
1220 case MSR_MTRRfix4K_F0000:
1221 case MSR_MTRRfix4K_F8000:
1222 case MSR_MTRRdefType:
1223 case MSR_IA32_CR_PAT:
1231 static bool valid_pat_type(unsigned t)
1233 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1236 static bool valid_mtrr_type(unsigned t)
1238 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1241 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1245 if (!msr_mtrr_valid(msr))
1248 if (msr == MSR_IA32_CR_PAT) {
1249 for (i = 0; i < 8; i++)
1250 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1253 } else if (msr == MSR_MTRRdefType) {
1256 return valid_mtrr_type(data & 0xff);
1257 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1258 for (i = 0; i < 8 ; i++)
1259 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1264 /* variable MTRRs */
1265 return valid_mtrr_type(data & 0xff);
1268 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1270 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1272 if (!mtrr_valid(vcpu, msr, data))
1275 if (msr == MSR_MTRRdefType) {
1276 vcpu->arch.mtrr_state.def_type = data;
1277 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1278 } else if (msr == MSR_MTRRfix64K_00000)
1280 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1281 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1282 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1283 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1284 else if (msr == MSR_IA32_CR_PAT)
1285 vcpu->arch.pat = data;
1286 else { /* Variable MTRRs */
1287 int idx, is_mtrr_mask;
1290 idx = (msr - 0x200) / 2;
1291 is_mtrr_mask = msr - 0x200 - 2 * idx;
1294 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1297 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1301 kvm_mmu_reset_context(vcpu);
1305 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1307 u64 mcg_cap = vcpu->arch.mcg_cap;
1308 unsigned bank_num = mcg_cap & 0xff;
1311 case MSR_IA32_MCG_STATUS:
1312 vcpu->arch.mcg_status = data;
1314 case MSR_IA32_MCG_CTL:
1315 if (!(mcg_cap & MCG_CTL_P))
1317 if (data != 0 && data != ~(u64)0)
1319 vcpu->arch.mcg_ctl = data;
1322 if (msr >= MSR_IA32_MC0_CTL &&
1323 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1324 u32 offset = msr - MSR_IA32_MC0_CTL;
1325 /* only 0 or all 1s can be written to IA32_MCi_CTL
1326 * some Linux kernels though clear bit 10 in bank 4 to
1327 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1328 * this to avoid an uncatched #GP in the guest
1330 if ((offset & 0x3) == 0 &&
1331 data != 0 && (data | (1 << 10)) != ~(u64)0)
1333 vcpu->arch.mce_banks[offset] = data;
1341 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1343 struct kvm *kvm = vcpu->kvm;
1344 int lm = is_long_mode(vcpu);
1345 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1346 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1347 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1348 : kvm->arch.xen_hvm_config.blob_size_32;
1349 u32 page_num = data & ~PAGE_MASK;
1350 u64 page_addr = data & PAGE_MASK;
1355 if (page_num >= blob_size)
1358 page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1362 if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1364 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1373 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1375 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1378 static bool kvm_hv_msr_partition_wide(u32 msr)
1382 case HV_X64_MSR_GUEST_OS_ID:
1383 case HV_X64_MSR_HYPERCALL:
1391 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1393 struct kvm *kvm = vcpu->kvm;
1396 case HV_X64_MSR_GUEST_OS_ID:
1397 kvm->arch.hv_guest_os_id = data;
1398 /* setting guest os id to zero disables hypercall page */
1399 if (!kvm->arch.hv_guest_os_id)
1400 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1402 case HV_X64_MSR_HYPERCALL: {
1407 /* if guest os id is not set hypercall should remain disabled */
1408 if (!kvm->arch.hv_guest_os_id)
1410 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1411 kvm->arch.hv_hypercall = data;
1414 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1415 addr = gfn_to_hva(kvm, gfn);
1416 if (kvm_is_error_hva(addr))
1418 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1419 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1420 if (__copy_to_user((void __user *)addr, instructions, 4))
1422 kvm->arch.hv_hypercall = data;
1426 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1427 "data 0x%llx\n", msr, data);
1433 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1436 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1439 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1440 vcpu->arch.hv_vapic = data;
1443 addr = gfn_to_hva(vcpu->kvm, data >>
1444 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1445 if (kvm_is_error_hva(addr))
1447 if (__clear_user((void __user *)addr, PAGE_SIZE))
1449 vcpu->arch.hv_vapic = data;
1452 case HV_X64_MSR_EOI:
1453 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1454 case HV_X64_MSR_ICR:
1455 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1456 case HV_X64_MSR_TPR:
1457 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1459 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1460 "data 0x%llx\n", msr, data);
1467 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1469 gpa_t gpa = data & ~0x3f;
1471 /* Bits 2:5 are resrved, Should be zero */
1475 vcpu->arch.apf.msr_val = data;
1477 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1478 kvm_clear_async_pf_completion_queue(vcpu);
1479 kvm_async_pf_hash_reset(vcpu);
1483 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1486 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1487 kvm_async_pf_wakeup_all(vcpu);
1491 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1493 vcpu->arch.pv_time_enabled = false;
1496 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1500 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1503 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1504 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1505 vcpu->arch.st.accum_steal = delta;
1508 static void record_steal_time(struct kvm_vcpu *vcpu)
1510 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1513 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1514 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1517 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1518 vcpu->arch.st.steal.version += 2;
1519 vcpu->arch.st.accum_steal = 0;
1521 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1522 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1525 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1529 return set_efer(vcpu, data);
1531 data &= ~(u64)0x40; /* ignore flush filter disable */
1532 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1534 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1539 case MSR_FAM10H_MMIO_CONF_BASE:
1541 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1546 case MSR_AMD64_NB_CFG:
1548 case MSR_IA32_DEBUGCTLMSR:
1550 /* We support the non-activated case already */
1552 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1553 /* Values other than LBR and BTF are vendor-specific,
1554 thus reserved and should throw a #GP */
1557 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1560 case MSR_IA32_UCODE_REV:
1561 case MSR_IA32_UCODE_WRITE:
1562 case MSR_VM_HSAVE_PA:
1563 case MSR_AMD64_PATCH_LOADER:
1565 case 0x200 ... 0x2ff:
1566 return set_msr_mtrr(vcpu, msr, data);
1567 case MSR_IA32_APICBASE:
1568 kvm_set_apic_base(vcpu, data);
1570 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1571 return kvm_x2apic_msr_write(vcpu, msr, data);
1572 case MSR_IA32_TSCDEADLINE:
1573 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1575 case MSR_IA32_MISC_ENABLE:
1576 vcpu->arch.ia32_misc_enable_msr = data;
1578 case MSR_KVM_WALL_CLOCK_NEW:
1579 case MSR_KVM_WALL_CLOCK:
1580 vcpu->kvm->arch.wall_clock = data;
1581 kvm_write_wall_clock(vcpu->kvm, data);
1583 case MSR_KVM_SYSTEM_TIME_NEW:
1584 case MSR_KVM_SYSTEM_TIME: {
1586 kvmclock_reset(vcpu);
1588 vcpu->arch.time = data;
1589 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1591 /* we verify if the enable bit is set... */
1595 gpa_offset = data & ~(PAGE_MASK | 1);
1597 /* Check that the address is 32-byte aligned. */
1598 if (gpa_offset & (sizeof(struct pvclock_vcpu_time_info) - 1))
1601 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
1602 &vcpu->arch.pv_time, data & ~1ULL))
1603 vcpu->arch.pv_time_enabled = false;
1605 vcpu->arch.pv_time_enabled = true;
1608 case MSR_KVM_ASYNC_PF_EN:
1609 if (kvm_pv_enable_async_pf(vcpu, data))
1612 case MSR_KVM_STEAL_TIME:
1614 if (unlikely(!sched_info_on()))
1617 if (data & KVM_STEAL_RESERVED_MASK)
1620 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
1621 data & KVM_STEAL_VALID_BITS))
1624 vcpu->arch.st.msr_val = data;
1626 if (!(data & KVM_MSR_ENABLED))
1629 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1632 accumulate_steal_time(vcpu);
1635 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
1639 case MSR_IA32_MCG_CTL:
1640 case MSR_IA32_MCG_STATUS:
1641 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1642 return set_msr_mce(vcpu, msr, data);
1644 /* Performance counters are not protected by a CPUID bit,
1645 * so we should check all of them in the generic path for the sake of
1646 * cross vendor migration.
1647 * Writing a zero into the event select MSRs disables them,
1648 * which we perfectly emulate ;-). Any other value should be at least
1649 * reported, some guests depend on them.
1651 case MSR_P6_EVNTSEL0:
1652 case MSR_P6_EVNTSEL1:
1653 case MSR_K7_EVNTSEL0:
1654 case MSR_K7_EVNTSEL1:
1655 case MSR_K7_EVNTSEL2:
1656 case MSR_K7_EVNTSEL3:
1658 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1659 "0x%x data 0x%llx\n", msr, data);
1661 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1662 * so we ignore writes to make it happy.
1664 case MSR_P6_PERFCTR0:
1665 case MSR_P6_PERFCTR1:
1666 case MSR_K7_PERFCTR0:
1667 case MSR_K7_PERFCTR1:
1668 case MSR_K7_PERFCTR2:
1669 case MSR_K7_PERFCTR3:
1670 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1671 "0x%x data 0x%llx\n", msr, data);
1673 case MSR_K7_CLK_CTL:
1675 * Ignore all writes to this no longer documented MSR.
1676 * Writes are only relevant for old K7 processors,
1677 * all pre-dating SVM, but a recommended workaround from
1678 * AMD for these chips. It is possible to speicify the
1679 * affected processor models on the command line, hence
1680 * the need to ignore the workaround.
1683 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1684 if (kvm_hv_msr_partition_wide(msr)) {
1686 mutex_lock(&vcpu->kvm->lock);
1687 r = set_msr_hyperv_pw(vcpu, msr, data);
1688 mutex_unlock(&vcpu->kvm->lock);
1691 return set_msr_hyperv(vcpu, msr, data);
1693 case MSR_IA32_BBL_CR_CTL3:
1694 /* Drop writes to this legacy MSR -- see rdmsr
1695 * counterpart for further detail.
1697 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1700 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1701 return xen_hvm_config(vcpu, data);
1703 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1707 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1714 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1718 * Reads an msr value (of 'msr_index') into 'pdata'.
1719 * Returns 0 on success, non-0 otherwise.
1720 * Assumes vcpu_load() was already called.
1722 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1724 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1727 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1729 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1731 if (!msr_mtrr_valid(msr))
1734 if (msr == MSR_MTRRdefType)
1735 *pdata = vcpu->arch.mtrr_state.def_type +
1736 (vcpu->arch.mtrr_state.enabled << 10);
1737 else if (msr == MSR_MTRRfix64K_00000)
1739 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1740 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1741 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1742 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1743 else if (msr == MSR_IA32_CR_PAT)
1744 *pdata = vcpu->arch.pat;
1745 else { /* Variable MTRRs */
1746 int idx, is_mtrr_mask;
1749 idx = (msr - 0x200) / 2;
1750 is_mtrr_mask = msr - 0x200 - 2 * idx;
1753 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1756 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1763 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1766 u64 mcg_cap = vcpu->arch.mcg_cap;
1767 unsigned bank_num = mcg_cap & 0xff;
1770 case MSR_IA32_P5_MC_ADDR:
1771 case MSR_IA32_P5_MC_TYPE:
1774 case MSR_IA32_MCG_CAP:
1775 data = vcpu->arch.mcg_cap;
1777 case MSR_IA32_MCG_CTL:
1778 if (!(mcg_cap & MCG_CTL_P))
1780 data = vcpu->arch.mcg_ctl;
1782 case MSR_IA32_MCG_STATUS:
1783 data = vcpu->arch.mcg_status;
1786 if (msr >= MSR_IA32_MC0_CTL &&
1787 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1788 u32 offset = msr - MSR_IA32_MC0_CTL;
1789 data = vcpu->arch.mce_banks[offset];
1798 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1801 struct kvm *kvm = vcpu->kvm;
1804 case HV_X64_MSR_GUEST_OS_ID:
1805 data = kvm->arch.hv_guest_os_id;
1807 case HV_X64_MSR_HYPERCALL:
1808 data = kvm->arch.hv_hypercall;
1811 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1819 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1824 case HV_X64_MSR_VP_INDEX: {
1827 kvm_for_each_vcpu(r, v, vcpu->kvm)
1832 case HV_X64_MSR_EOI:
1833 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1834 case HV_X64_MSR_ICR:
1835 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1836 case HV_X64_MSR_TPR:
1837 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1838 case HV_X64_MSR_APIC_ASSIST_PAGE:
1839 data = vcpu->arch.hv_vapic;
1842 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1849 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1854 case MSR_IA32_PLATFORM_ID:
1855 case MSR_IA32_EBL_CR_POWERON:
1856 case MSR_IA32_DEBUGCTLMSR:
1857 case MSR_IA32_LASTBRANCHFROMIP:
1858 case MSR_IA32_LASTBRANCHTOIP:
1859 case MSR_IA32_LASTINTFROMIP:
1860 case MSR_IA32_LASTINTTOIP:
1863 case MSR_VM_HSAVE_PA:
1864 case MSR_P6_PERFCTR0:
1865 case MSR_P6_PERFCTR1:
1866 case MSR_P6_EVNTSEL0:
1867 case MSR_P6_EVNTSEL1:
1868 case MSR_K7_EVNTSEL0:
1869 case MSR_K7_PERFCTR0:
1870 case MSR_K8_INT_PENDING_MSG:
1871 case MSR_AMD64_NB_CFG:
1872 case MSR_FAM10H_MMIO_CONF_BASE:
1875 case MSR_IA32_UCODE_REV:
1876 data = 0x100000000ULL;
1879 data = 0x500 | KVM_NR_VAR_MTRR;
1881 case 0x200 ... 0x2ff:
1882 return get_msr_mtrr(vcpu, msr, pdata);
1883 case 0xcd: /* fsb frequency */
1887 * MSR_EBC_FREQUENCY_ID
1888 * Conservative value valid for even the basic CPU models.
1889 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1890 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1891 * and 266MHz for model 3, or 4. Set Core Clock
1892 * Frequency to System Bus Frequency Ratio to 1 (bits
1893 * 31:24) even though these are only valid for CPU
1894 * models > 2, however guests may end up dividing or
1895 * multiplying by zero otherwise.
1897 case MSR_EBC_FREQUENCY_ID:
1900 case MSR_IA32_APICBASE:
1901 data = kvm_get_apic_base(vcpu);
1903 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1904 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1906 case MSR_IA32_TSCDEADLINE:
1907 data = kvm_get_lapic_tscdeadline_msr(vcpu);
1909 case MSR_IA32_MISC_ENABLE:
1910 data = vcpu->arch.ia32_misc_enable_msr;
1912 case MSR_IA32_PERF_STATUS:
1913 /* TSC increment by tick */
1915 /* CPU multiplier */
1916 data |= (((uint64_t)4ULL) << 40);
1919 data = vcpu->arch.efer;
1921 case MSR_KVM_WALL_CLOCK:
1922 case MSR_KVM_WALL_CLOCK_NEW:
1923 data = vcpu->kvm->arch.wall_clock;
1925 case MSR_KVM_SYSTEM_TIME:
1926 case MSR_KVM_SYSTEM_TIME_NEW:
1927 data = vcpu->arch.time;
1929 case MSR_KVM_ASYNC_PF_EN:
1930 data = vcpu->arch.apf.msr_val;
1932 case MSR_KVM_STEAL_TIME:
1933 data = vcpu->arch.st.msr_val;
1935 case MSR_IA32_P5_MC_ADDR:
1936 case MSR_IA32_P5_MC_TYPE:
1937 case MSR_IA32_MCG_CAP:
1938 case MSR_IA32_MCG_CTL:
1939 case MSR_IA32_MCG_STATUS:
1940 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1941 return get_msr_mce(vcpu, msr, pdata);
1942 case MSR_K7_CLK_CTL:
1944 * Provide expected ramp-up count for K7. All other
1945 * are set to zero, indicating minimum divisors for
1948 * This prevents guest kernels on AMD host with CPU
1949 * type 6, model 8 and higher from exploding due to
1950 * the rdmsr failing.
1954 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1955 if (kvm_hv_msr_partition_wide(msr)) {
1957 mutex_lock(&vcpu->kvm->lock);
1958 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1959 mutex_unlock(&vcpu->kvm->lock);
1962 return get_msr_hyperv(vcpu, msr, pdata);
1964 case MSR_IA32_BBL_CR_CTL3:
1965 /* This legacy MSR exists but isn't fully documented in current
1966 * silicon. It is however accessed by winxp in very narrow
1967 * scenarios where it sets bit #19, itself documented as
1968 * a "reserved" bit. Best effort attempt to source coherent
1969 * read data here should the balance of the register be
1970 * interpreted by the guest:
1972 * L2 cache control register 3: 64GB range, 256KB size,
1973 * enabled, latency 0x1, configured
1979 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1982 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1990 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1993 * Read or write a bunch of msrs. All parameters are kernel addresses.
1995 * @return number of msrs set successfully.
1997 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1998 struct kvm_msr_entry *entries,
1999 int (*do_msr)(struct kvm_vcpu *vcpu,
2000 unsigned index, u64 *data))
2004 idx = srcu_read_lock(&vcpu->kvm->srcu);
2005 for (i = 0; i < msrs->nmsrs; ++i)
2006 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2008 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2014 * Read or write a bunch of msrs. Parameters are user addresses.
2016 * @return number of msrs set successfully.
2018 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2019 int (*do_msr)(struct kvm_vcpu *vcpu,
2020 unsigned index, u64 *data),
2023 struct kvm_msrs msrs;
2024 struct kvm_msr_entry *entries;
2029 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2033 if (msrs.nmsrs >= MAX_IO_MSRS)
2037 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2038 entries = kmalloc(size, GFP_KERNEL);
2043 if (copy_from_user(entries, user_msrs->entries, size))
2046 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2051 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2062 int kvm_dev_ioctl_check_extension(long ext)
2067 case KVM_CAP_IRQCHIP:
2069 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2070 case KVM_CAP_SET_TSS_ADDR:
2071 case KVM_CAP_EXT_CPUID:
2072 case KVM_CAP_CLOCKSOURCE:
2074 case KVM_CAP_NOP_IO_DELAY:
2075 case KVM_CAP_MP_STATE:
2076 case KVM_CAP_SYNC_MMU:
2077 case KVM_CAP_USER_NMI:
2078 case KVM_CAP_REINJECT_CONTROL:
2079 case KVM_CAP_IRQ_INJECT_STATUS:
2080 case KVM_CAP_ASSIGN_DEV_IRQ:
2082 case KVM_CAP_IOEVENTFD:
2084 case KVM_CAP_PIT_STATE2:
2085 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2086 case KVM_CAP_XEN_HVM:
2087 case KVM_CAP_ADJUST_CLOCK:
2088 case KVM_CAP_VCPU_EVENTS:
2089 case KVM_CAP_HYPERV:
2090 case KVM_CAP_HYPERV_VAPIC:
2091 case KVM_CAP_HYPERV_SPIN:
2092 case KVM_CAP_PCI_SEGMENT:
2093 case KVM_CAP_DEBUGREGS:
2094 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2096 case KVM_CAP_ASYNC_PF:
2097 case KVM_CAP_GET_TSC_KHZ:
2100 case KVM_CAP_COALESCED_MMIO:
2101 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2104 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2106 case KVM_CAP_NR_VCPUS:
2107 r = KVM_SOFT_MAX_VCPUS;
2109 case KVM_CAP_MAX_VCPUS:
2112 case KVM_CAP_NR_MEMSLOTS:
2113 r = KVM_MEMORY_SLOTS;
2115 case KVM_CAP_PV_MMU: /* obsolete */
2119 r = iommu_present(&pci_bus_type);
2122 r = KVM_MAX_MCE_BANKS;
2127 case KVM_CAP_TSC_CONTROL:
2128 r = kvm_has_tsc_control;
2130 case KVM_CAP_TSC_DEADLINE_TIMER:
2131 r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2141 long kvm_arch_dev_ioctl(struct file *filp,
2142 unsigned int ioctl, unsigned long arg)
2144 void __user *argp = (void __user *)arg;
2148 case KVM_GET_MSR_INDEX_LIST: {
2149 struct kvm_msr_list __user *user_msr_list = argp;
2150 struct kvm_msr_list msr_list;
2154 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2157 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2158 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2161 if (n < msr_list.nmsrs)
2164 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2165 num_msrs_to_save * sizeof(u32)))
2167 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2169 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2174 case KVM_GET_SUPPORTED_CPUID: {
2175 struct kvm_cpuid2 __user *cpuid_arg = argp;
2176 struct kvm_cpuid2 cpuid;
2179 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2181 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2182 cpuid_arg->entries);
2187 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2192 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2195 mce_cap = KVM_MCE_CAP_SUPPORTED;
2197 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2209 static void wbinvd_ipi(void *garbage)
2214 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2216 return vcpu->kvm->arch.iommu_domain &&
2217 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2220 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2222 /* Address WBINVD may be executed by guest */
2223 if (need_emulate_wbinvd(vcpu)) {
2224 if (kvm_x86_ops->has_wbinvd_exit())
2225 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2226 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2227 smp_call_function_single(vcpu->cpu,
2228 wbinvd_ipi, NULL, 1);
2231 kvm_x86_ops->vcpu_load(vcpu, cpu);
2232 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2233 /* Make sure TSC doesn't go backwards */
2237 tsc = kvm_x86_ops->read_l1_tsc(vcpu);
2238 tsc_delta = !vcpu->arch.last_guest_tsc ? 0 :
2239 tsc - vcpu->arch.last_guest_tsc;
2242 mark_tsc_unstable("KVM discovered backwards TSC");
2243 if (check_tsc_unstable()) {
2244 kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2245 vcpu->arch.tsc_catchup = 1;
2247 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2248 if (vcpu->cpu != cpu)
2249 kvm_migrate_timers(vcpu);
2253 accumulate_steal_time(vcpu);
2254 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2257 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2259 kvm_x86_ops->vcpu_put(vcpu);
2260 kvm_put_guest_fpu(vcpu);
2261 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu);
2264 static int is_efer_nx(void)
2266 unsigned long long efer = 0;
2268 rdmsrl_safe(MSR_EFER, &efer);
2269 return efer & EFER_NX;
2272 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2275 struct kvm_cpuid_entry2 *e, *entry;
2278 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2279 e = &vcpu->arch.cpuid_entries[i];
2280 if (e->function == 0x80000001) {
2285 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2286 entry->edx &= ~(1 << 20);
2287 printk(KERN_INFO "kvm: guest NX capability removed\n");
2291 /* when an old userspace process fills a new kernel module */
2292 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2293 struct kvm_cpuid *cpuid,
2294 struct kvm_cpuid_entry __user *entries)
2297 struct kvm_cpuid_entry *cpuid_entries;
2300 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2303 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2307 if (copy_from_user(cpuid_entries, entries,
2308 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2310 for (i = 0; i < cpuid->nent; i++) {
2311 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2312 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2313 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2314 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2315 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2316 vcpu->arch.cpuid_entries[i].index = 0;
2317 vcpu->arch.cpuid_entries[i].flags = 0;
2318 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2319 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2320 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2322 vcpu->arch.cpuid_nent = cpuid->nent;
2323 cpuid_fix_nx_cap(vcpu);
2325 kvm_apic_set_version(vcpu);
2326 kvm_x86_ops->cpuid_update(vcpu);
2330 vfree(cpuid_entries);
2335 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2336 struct kvm_cpuid2 *cpuid,
2337 struct kvm_cpuid_entry2 __user *entries)
2342 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2345 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2346 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2348 vcpu->arch.cpuid_nent = cpuid->nent;
2349 kvm_apic_set_version(vcpu);
2350 kvm_x86_ops->cpuid_update(vcpu);
2358 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2359 struct kvm_cpuid2 *cpuid,
2360 struct kvm_cpuid_entry2 __user *entries)
2365 if (cpuid->nent < vcpu->arch.cpuid_nent)
2368 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2369 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2374 cpuid->nent = vcpu->arch.cpuid_nent;
2378 static void cpuid_mask(u32 *word, int wordnum)
2380 *word &= boot_cpu_data.x86_capability[wordnum];
2383 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2386 entry->function = function;
2387 entry->index = index;
2388 cpuid_count(entry->function, entry->index,
2389 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2393 static bool supported_xcr0_bit(unsigned bit)
2395 u64 mask = ((u64)1 << bit);
2397 return mask & (XSTATE_FP | XSTATE_SSE | XSTATE_YMM) & host_xcr0;
2400 #define F(x) bit(X86_FEATURE_##x)
2402 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2403 u32 index, int *nent, int maxnent)
2405 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2406 #ifdef CONFIG_X86_64
2407 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2409 unsigned f_lm = F(LM);
2411 unsigned f_gbpages = 0;
2414 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2417 const u32 kvm_supported_word0_x86_features =
2418 F(FPU) | F(VME) | F(DE) | F(PSE) |
2419 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2420 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2421 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2422 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2423 0 /* Reserved, DS, ACPI */ | F(MMX) |
2424 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2425 0 /* HTT, TM, Reserved, PBE */;
2426 /* cpuid 0x80000001.edx */
2427 const u32 kvm_supported_word1_x86_features =
2428 F(FPU) | F(VME) | F(DE) | F(PSE) |
2429 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2430 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2431 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2432 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2433 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2434 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2435 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2437 const u32 kvm_supported_word4_x86_features =
2438 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
2439 0 /* DS-CPL, VMX, SMX, EST */ |
2440 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2441 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2442 0 /* Reserved, DCA */ | F(XMM4_1) |
2443 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2444 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2445 F(F16C) | F(RDRAND);
2446 /* cpuid 0x80000001.ecx */
2447 const u32 kvm_supported_word6_x86_features =
2448 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2449 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2450 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2451 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2453 /* cpuid 0xC0000001.edx */
2454 const u32 kvm_supported_word5_x86_features =
2455 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
2456 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
2460 const u32 kvm_supported_word9_x86_features =
2461 F(SMEP) | F(FSGSBASE) | F(ERMS);
2463 /* all calls to cpuid_count() should be made on the same cpu */
2465 do_cpuid_1_ent(entry, function, index);
2470 entry->eax = min(entry->eax, (u32)0xd);
2473 entry->edx &= kvm_supported_word0_x86_features;
2474 cpuid_mask(&entry->edx, 0);
2475 entry->ecx &= kvm_supported_word4_x86_features;
2476 cpuid_mask(&entry->ecx, 4);
2477 /* we support x2apic emulation even if host does not support
2478 * it since we emulate x2apic in software */
2479 entry->ecx |= F(X2APIC);
2481 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2482 * may return different values. This forces us to get_cpu() before
2483 * issuing the first command, and also to emulate this annoying behavior
2484 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2486 int t, times = entry->eax & 0xff;
2488 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2489 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2490 for (t = 1; t < times && *nent < maxnent; ++t) {
2491 do_cpuid_1_ent(&entry[t], function, 0);
2492 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2497 /* function 4 has additional index. */
2501 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2502 /* read more entries until cache_type is zero */
2503 for (i = 1; *nent < maxnent; ++i) {
2504 cache_type = entry[i - 1].eax & 0x1f;
2507 do_cpuid_1_ent(&entry[i], function, i);
2509 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2515 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2516 /* Mask ebx against host capbability word 9 */
2518 entry->ebx &= kvm_supported_word9_x86_features;
2519 cpuid_mask(&entry->ebx, 9);
2529 /* function 0xb has additional index. */
2533 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2534 /* read more entries until level_type is zero */
2535 for (i = 1; *nent < maxnent; ++i) {
2536 level_type = entry[i - 1].ecx & 0xff00;
2539 do_cpuid_1_ent(&entry[i], function, i);
2541 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2549 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2550 for (idx = 1, i = 1; *nent < maxnent && idx < 64; ++idx) {
2551 do_cpuid_1_ent(&entry[i], function, idx);
2552 if (entry[i].eax == 0 || !supported_xcr0_bit(idx))
2555 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2561 case KVM_CPUID_SIGNATURE: {
2562 char signature[12] = "KVMKVMKVM\0\0";
2563 u32 *sigptr = (u32 *)signature;
2565 entry->ebx = sigptr[0];
2566 entry->ecx = sigptr[1];
2567 entry->edx = sigptr[2];
2570 case KVM_CPUID_FEATURES:
2571 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2572 (1 << KVM_FEATURE_NOP_IO_DELAY) |
2573 (1 << KVM_FEATURE_CLOCKSOURCE2) |
2574 (1 << KVM_FEATURE_ASYNC_PF) |
2575 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2577 if (sched_info_on())
2578 entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
2585 entry->eax = min(entry->eax, 0x8000001a);
2588 entry->edx &= kvm_supported_word1_x86_features;
2589 cpuid_mask(&entry->edx, 1);
2590 entry->ecx &= kvm_supported_word6_x86_features;
2591 cpuid_mask(&entry->ecx, 6);
2594 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
2595 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
2596 unsigned phys_as = entry->eax & 0xff;
2599 g_phys_as = phys_as;
2600 entry->eax = g_phys_as | (virt_as << 8);
2601 entry->ebx = entry->edx = 0;
2605 entry->ecx = entry->edx = 0;
2611 /*Add support for Centaur's CPUID instruction*/
2613 /*Just support up to 0xC0000004 now*/
2614 entry->eax = min(entry->eax, 0xC0000004);
2617 entry->edx &= kvm_supported_word5_x86_features;
2618 cpuid_mask(&entry->edx, 5);
2620 case 3: /* Processor serial number */
2621 case 5: /* MONITOR/MWAIT */
2622 case 6: /* Thermal management */
2623 case 0xA: /* Architectural Performance Monitoring */
2624 case 0x80000007: /* Advanced power management */
2629 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
2633 kvm_x86_ops->set_supported_cpuid(function, entry);
2640 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2641 struct kvm_cpuid_entry2 __user *entries)
2643 struct kvm_cpuid_entry2 *cpuid_entries;
2644 int limit, nent = 0, r = -E2BIG;
2647 if (cpuid->nent < 1)
2649 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2650 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2652 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2656 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2657 limit = cpuid_entries[0].eax;
2658 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2659 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2660 &nent, cpuid->nent);
2662 if (nent >= cpuid->nent)
2665 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2666 limit = cpuid_entries[nent - 1].eax;
2667 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2668 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2669 &nent, cpuid->nent);
2674 if (nent >= cpuid->nent)
2677 /* Add support for Centaur's CPUID instruction. */
2678 if (boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR) {
2679 do_cpuid_ent(&cpuid_entries[nent], 0xC0000000, 0,
2680 &nent, cpuid->nent);
2683 if (nent >= cpuid->nent)
2686 limit = cpuid_entries[nent - 1].eax;
2687 for (func = 0xC0000001;
2688 func <= limit && nent < cpuid->nent; ++func)
2689 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2690 &nent, cpuid->nent);
2693 if (nent >= cpuid->nent)
2697 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2701 if (nent >= cpuid->nent)
2704 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2708 if (nent >= cpuid->nent)
2712 if (copy_to_user(entries, cpuid_entries,
2713 nent * sizeof(struct kvm_cpuid_entry2)))
2719 vfree(cpuid_entries);
2724 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2725 struct kvm_lapic_state *s)
2727 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2732 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2733 struct kvm_lapic_state *s)
2735 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2736 kvm_apic_post_state_restore(vcpu);
2737 update_cr8_intercept(vcpu);
2742 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2743 struct kvm_interrupt *irq)
2745 if (irq->irq < 0 || irq->irq >= 256)
2747 if (irqchip_in_kernel(vcpu->kvm))
2750 kvm_queue_interrupt(vcpu, irq->irq, false);
2751 kvm_make_request(KVM_REQ_EVENT, vcpu);
2756 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2758 kvm_inject_nmi(vcpu);
2763 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2764 struct kvm_tpr_access_ctl *tac)
2768 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2772 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2776 unsigned bank_num = mcg_cap & 0xff, bank;
2779 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2781 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2784 vcpu->arch.mcg_cap = mcg_cap;
2785 /* Init IA32_MCG_CTL to all 1s */
2786 if (mcg_cap & MCG_CTL_P)
2787 vcpu->arch.mcg_ctl = ~(u64)0;
2788 /* Init IA32_MCi_CTL to all 1s */
2789 for (bank = 0; bank < bank_num; bank++)
2790 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2795 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2796 struct kvm_x86_mce *mce)
2798 u64 mcg_cap = vcpu->arch.mcg_cap;
2799 unsigned bank_num = mcg_cap & 0xff;
2800 u64 *banks = vcpu->arch.mce_banks;
2802 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2805 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2806 * reporting is disabled
2808 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2809 vcpu->arch.mcg_ctl != ~(u64)0)
2811 banks += 4 * mce->bank;
2813 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2814 * reporting is disabled for the bank
2816 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2818 if (mce->status & MCI_STATUS_UC) {
2819 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2820 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2821 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2824 if (banks[1] & MCI_STATUS_VAL)
2825 mce->status |= MCI_STATUS_OVER;
2826 banks[2] = mce->addr;
2827 banks[3] = mce->misc;
2828 vcpu->arch.mcg_status = mce->mcg_status;
2829 banks[1] = mce->status;
2830 kvm_queue_exception(vcpu, MC_VECTOR);
2831 } else if (!(banks[1] & MCI_STATUS_VAL)
2832 || !(banks[1] & MCI_STATUS_UC)) {
2833 if (banks[1] & MCI_STATUS_VAL)
2834 mce->status |= MCI_STATUS_OVER;
2835 banks[2] = mce->addr;
2836 banks[3] = mce->misc;
2837 banks[1] = mce->status;
2839 banks[1] |= MCI_STATUS_OVER;
2843 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2844 struct kvm_vcpu_events *events)
2847 events->exception.injected =
2848 vcpu->arch.exception.pending &&
2849 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2850 events->exception.nr = vcpu->arch.exception.nr;
2851 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2852 events->exception.pad = 0;
2853 events->exception.error_code = vcpu->arch.exception.error_code;
2855 events->interrupt.injected =
2856 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2857 events->interrupt.nr = vcpu->arch.interrupt.nr;
2858 events->interrupt.soft = 0;
2859 events->interrupt.shadow =
2860 kvm_x86_ops->get_interrupt_shadow(vcpu,
2861 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2863 events->nmi.injected = vcpu->arch.nmi_injected;
2864 events->nmi.pending = vcpu->arch.nmi_pending != 0;
2865 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2866 events->nmi.pad = 0;
2868 events->sipi_vector = vcpu->arch.sipi_vector;
2870 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2871 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2872 | KVM_VCPUEVENT_VALID_SHADOW);
2873 memset(&events->reserved, 0, sizeof(events->reserved));
2876 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2877 struct kvm_vcpu_events *events)
2879 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2880 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2881 | KVM_VCPUEVENT_VALID_SHADOW))
2885 vcpu->arch.exception.pending = events->exception.injected;
2886 vcpu->arch.exception.nr = events->exception.nr;
2887 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2888 vcpu->arch.exception.error_code = events->exception.error_code;
2890 vcpu->arch.interrupt.pending = events->interrupt.injected;
2891 vcpu->arch.interrupt.nr = events->interrupt.nr;
2892 vcpu->arch.interrupt.soft = events->interrupt.soft;
2893 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2894 kvm_x86_ops->set_interrupt_shadow(vcpu,
2895 events->interrupt.shadow);
2897 vcpu->arch.nmi_injected = events->nmi.injected;
2898 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2899 vcpu->arch.nmi_pending = events->nmi.pending;
2900 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2902 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2903 vcpu->arch.sipi_vector = events->sipi_vector;
2905 kvm_make_request(KVM_REQ_EVENT, vcpu);
2910 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2911 struct kvm_debugregs *dbgregs)
2913 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2914 dbgregs->dr6 = vcpu->arch.dr6;
2915 dbgregs->dr7 = vcpu->arch.dr7;
2917 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2920 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2921 struct kvm_debugregs *dbgregs)
2926 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2927 vcpu->arch.dr6 = dbgregs->dr6;
2928 vcpu->arch.dr7 = dbgregs->dr7;
2933 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2934 struct kvm_xsave *guest_xsave)
2937 memcpy(guest_xsave->region,
2938 &vcpu->arch.guest_fpu.state->xsave,
2941 memcpy(guest_xsave->region,
2942 &vcpu->arch.guest_fpu.state->fxsave,
2943 sizeof(struct i387_fxsave_struct));
2944 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2949 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2950 struct kvm_xsave *guest_xsave)
2953 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2956 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2957 guest_xsave->region, xstate_size);
2959 if (xstate_bv & ~XSTATE_FPSSE)
2961 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2962 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2967 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2968 struct kvm_xcrs *guest_xcrs)
2970 if (!cpu_has_xsave) {
2971 guest_xcrs->nr_xcrs = 0;
2975 guest_xcrs->nr_xcrs = 1;
2976 guest_xcrs->flags = 0;
2977 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2978 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2981 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2982 struct kvm_xcrs *guest_xcrs)
2989 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2992 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2993 /* Only support XCR0 currently */
2994 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2995 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2996 guest_xcrs->xcrs[0].value);
3004 long kvm_arch_vcpu_ioctl(struct file *filp,
3005 unsigned int ioctl, unsigned long arg)
3007 struct kvm_vcpu *vcpu = filp->private_data;
3008 void __user *argp = (void __user *)arg;
3011 struct kvm_lapic_state *lapic;
3012 struct kvm_xsave *xsave;
3013 struct kvm_xcrs *xcrs;
3019 case KVM_GET_LAPIC: {
3021 if (!vcpu->arch.apic)
3023 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3028 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3032 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3037 case KVM_SET_LAPIC: {
3039 if (!vcpu->arch.apic)
3041 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3046 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
3048 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3054 case KVM_INTERRUPT: {
3055 struct kvm_interrupt irq;
3058 if (copy_from_user(&irq, argp, sizeof irq))
3060 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3067 r = kvm_vcpu_ioctl_nmi(vcpu);
3073 case KVM_SET_CPUID: {
3074 struct kvm_cpuid __user *cpuid_arg = argp;
3075 struct kvm_cpuid cpuid;
3078 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3080 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3085 case KVM_SET_CPUID2: {
3086 struct kvm_cpuid2 __user *cpuid_arg = argp;
3087 struct kvm_cpuid2 cpuid;
3090 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3092 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3093 cpuid_arg->entries);
3098 case KVM_GET_CPUID2: {
3099 struct kvm_cpuid2 __user *cpuid_arg = argp;
3100 struct kvm_cpuid2 cpuid;
3103 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3105 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3106 cpuid_arg->entries);
3110 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3116 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3119 r = msr_io(vcpu, argp, do_set_msr, 0);
3121 case KVM_TPR_ACCESS_REPORTING: {
3122 struct kvm_tpr_access_ctl tac;
3125 if (copy_from_user(&tac, argp, sizeof tac))
3127 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3131 if (copy_to_user(argp, &tac, sizeof tac))
3136 case KVM_SET_VAPIC_ADDR: {
3137 struct kvm_vapic_addr va;
3140 if (!irqchip_in_kernel(vcpu->kvm))
3143 if (copy_from_user(&va, argp, sizeof va))
3146 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3149 case KVM_X86_SETUP_MCE: {
3153 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3155 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3158 case KVM_X86_SET_MCE: {
3159 struct kvm_x86_mce mce;
3162 if (copy_from_user(&mce, argp, sizeof mce))
3164 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3167 case KVM_GET_VCPU_EVENTS: {
3168 struct kvm_vcpu_events events;
3170 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3173 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3178 case KVM_SET_VCPU_EVENTS: {
3179 struct kvm_vcpu_events events;
3182 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3185 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3188 case KVM_GET_DEBUGREGS: {
3189 struct kvm_debugregs dbgregs;
3191 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3194 if (copy_to_user(argp, &dbgregs,
3195 sizeof(struct kvm_debugregs)))
3200 case KVM_SET_DEBUGREGS: {
3201 struct kvm_debugregs dbgregs;
3204 if (copy_from_user(&dbgregs, argp,
3205 sizeof(struct kvm_debugregs)))
3208 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3211 case KVM_GET_XSAVE: {
3212 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3217 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3220 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3225 case KVM_SET_XSAVE: {
3226 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3232 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
3235 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3238 case KVM_GET_XCRS: {
3239 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3244 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3247 if (copy_to_user(argp, u.xcrs,
3248 sizeof(struct kvm_xcrs)))
3253 case KVM_SET_XCRS: {
3254 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3260 if (copy_from_user(u.xcrs, argp,
3261 sizeof(struct kvm_xcrs)))
3264 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3267 case KVM_SET_TSC_KHZ: {
3271 if (!kvm_has_tsc_control)
3274 user_tsc_khz = (u32)arg;
3276 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3279 kvm_x86_ops->set_tsc_khz(vcpu, user_tsc_khz);
3284 case KVM_GET_TSC_KHZ: {
3286 if (check_tsc_unstable())
3289 r = vcpu_tsc_khz(vcpu);
3301 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3305 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3307 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3311 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3314 kvm->arch.ept_identity_map_addr = ident_addr;
3318 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3319 u32 kvm_nr_mmu_pages)
3321 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3324 mutex_lock(&kvm->slots_lock);
3325 spin_lock(&kvm->mmu_lock);
3327 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3328 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3330 spin_unlock(&kvm->mmu_lock);
3331 mutex_unlock(&kvm->slots_lock);
3335 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3337 return kvm->arch.n_max_mmu_pages;
3340 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3345 switch (chip->chip_id) {
3346 case KVM_IRQCHIP_PIC_MASTER:
3347 memcpy(&chip->chip.pic,
3348 &pic_irqchip(kvm)->pics[0],
3349 sizeof(struct kvm_pic_state));
3351 case KVM_IRQCHIP_PIC_SLAVE:
3352 memcpy(&chip->chip.pic,
3353 &pic_irqchip(kvm)->pics[1],
3354 sizeof(struct kvm_pic_state));
3356 case KVM_IRQCHIP_IOAPIC:
3357 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3366 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3371 switch (chip->chip_id) {
3372 case KVM_IRQCHIP_PIC_MASTER:
3373 spin_lock(&pic_irqchip(kvm)->lock);
3374 memcpy(&pic_irqchip(kvm)->pics[0],
3376 sizeof(struct kvm_pic_state));
3377 spin_unlock(&pic_irqchip(kvm)->lock);
3379 case KVM_IRQCHIP_PIC_SLAVE:
3380 spin_lock(&pic_irqchip(kvm)->lock);
3381 memcpy(&pic_irqchip(kvm)->pics[1],
3383 sizeof(struct kvm_pic_state));
3384 spin_unlock(&pic_irqchip(kvm)->lock);
3386 case KVM_IRQCHIP_IOAPIC:
3387 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3393 kvm_pic_update_irq(pic_irqchip(kvm));
3397 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3401 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3402 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3403 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3407 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3411 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3412 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3413 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3414 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3418 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3422 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3423 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3424 sizeof(ps->channels));
3425 ps->flags = kvm->arch.vpit->pit_state.flags;
3426 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3427 memset(&ps->reserved, 0, sizeof(ps->reserved));
3431 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3433 int r = 0, start = 0;
3434 u32 prev_legacy, cur_legacy;
3435 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3436 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3437 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3438 if (!prev_legacy && cur_legacy)
3440 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3441 sizeof(kvm->arch.vpit->pit_state.channels));
3442 kvm->arch.vpit->pit_state.flags = ps->flags;
3443 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3444 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3448 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3449 struct kvm_reinject_control *control)
3451 if (!kvm->arch.vpit)
3453 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3454 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3455 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3460 * Get (and clear) the dirty memory log for a memory slot.
3462 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3463 struct kvm_dirty_log *log)
3466 struct kvm_memory_slot *memslot;
3468 unsigned long is_dirty = 0;
3470 mutex_lock(&kvm->slots_lock);
3473 if (log->slot >= KVM_MEMORY_SLOTS)
3476 memslot = &kvm->memslots->memslots[log->slot];
3478 if (!memslot->dirty_bitmap)
3481 n = kvm_dirty_bitmap_bytes(memslot);
3483 for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3484 is_dirty = memslot->dirty_bitmap[i];
3486 /* If nothing is dirty, don't bother messing with page tables. */
3488 struct kvm_memslots *slots, *old_slots;
3489 unsigned long *dirty_bitmap;
3491 dirty_bitmap = memslot->dirty_bitmap_head;
3492 if (memslot->dirty_bitmap == dirty_bitmap)
3493 dirty_bitmap += n / sizeof(long);
3494 memset(dirty_bitmap, 0, n);
3497 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3500 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3501 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3502 slots->generation++;
3504 old_slots = kvm->memslots;
3505 rcu_assign_pointer(kvm->memslots, slots);
3506 synchronize_srcu_expedited(&kvm->srcu);
3507 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3510 spin_lock(&kvm->mmu_lock);
3511 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3512 spin_unlock(&kvm->mmu_lock);
3515 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
3519 if (clear_user(log->dirty_bitmap, n))
3525 mutex_unlock(&kvm->slots_lock);
3529 long kvm_arch_vm_ioctl(struct file *filp,
3530 unsigned int ioctl, unsigned long arg)
3532 struct kvm *kvm = filp->private_data;
3533 void __user *argp = (void __user *)arg;
3536 * This union makes it completely explicit to gcc-3.x
3537 * that these two variables' stack usage should be
3538 * combined, not added together.
3541 struct kvm_pit_state ps;
3542 struct kvm_pit_state2 ps2;
3543 struct kvm_pit_config pit_config;
3547 case KVM_SET_TSS_ADDR:
3548 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3552 case KVM_SET_IDENTITY_MAP_ADDR: {
3556 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3558 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3563 case KVM_SET_NR_MMU_PAGES:
3564 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3568 case KVM_GET_NR_MMU_PAGES:
3569 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3571 case KVM_CREATE_IRQCHIP: {
3572 struct kvm_pic *vpic;
3574 mutex_lock(&kvm->lock);
3577 goto create_irqchip_unlock;
3579 if (atomic_read(&kvm->online_vcpus))
3580 goto create_irqchip_unlock;
3582 vpic = kvm_create_pic(kvm);
3584 r = kvm_ioapic_init(kvm);
3586 mutex_lock(&kvm->slots_lock);
3587 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3589 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3591 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3593 mutex_unlock(&kvm->slots_lock);
3595 goto create_irqchip_unlock;
3598 goto create_irqchip_unlock;
3600 kvm->arch.vpic = vpic;
3602 r = kvm_setup_default_irq_routing(kvm);
3604 mutex_lock(&kvm->slots_lock);
3605 mutex_lock(&kvm->irq_lock);
3606 kvm_ioapic_destroy(kvm);
3607 kvm_destroy_pic(kvm);
3608 mutex_unlock(&kvm->irq_lock);
3609 mutex_unlock(&kvm->slots_lock);
3611 create_irqchip_unlock:
3612 mutex_unlock(&kvm->lock);
3615 case KVM_CREATE_PIT:
3616 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3618 case KVM_CREATE_PIT2:
3620 if (copy_from_user(&u.pit_config, argp,
3621 sizeof(struct kvm_pit_config)))
3624 mutex_lock(&kvm->slots_lock);
3627 goto create_pit_unlock;
3629 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3633 mutex_unlock(&kvm->slots_lock);
3635 case KVM_IRQ_LINE_STATUS:
3636 case KVM_IRQ_LINE: {
3637 struct kvm_irq_level irq_event;
3640 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3643 if (irqchip_in_kernel(kvm)) {
3645 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3646 irq_event.irq, irq_event.level);
3647 if (ioctl == KVM_IRQ_LINE_STATUS) {
3649 irq_event.status = status;
3650 if (copy_to_user(argp, &irq_event,
3658 case KVM_GET_IRQCHIP: {
3659 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3660 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3666 if (copy_from_user(chip, argp, sizeof *chip))
3667 goto get_irqchip_out;
3669 if (!irqchip_in_kernel(kvm))
3670 goto get_irqchip_out;
3671 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3673 goto get_irqchip_out;
3675 if (copy_to_user(argp, chip, sizeof *chip))
3676 goto get_irqchip_out;
3684 case KVM_SET_IRQCHIP: {
3685 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3686 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3692 if (copy_from_user(chip, argp, sizeof *chip))
3693 goto set_irqchip_out;
3695 if (!irqchip_in_kernel(kvm))
3696 goto set_irqchip_out;
3697 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3699 goto set_irqchip_out;
3709 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3712 if (!kvm->arch.vpit)
3714 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3718 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3725 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3728 if (!kvm->arch.vpit)
3730 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3736 case KVM_GET_PIT2: {
3738 if (!kvm->arch.vpit)
3740 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3744 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3749 case KVM_SET_PIT2: {
3751 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3754 if (!kvm->arch.vpit)
3756 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3762 case KVM_REINJECT_CONTROL: {
3763 struct kvm_reinject_control control;
3765 if (copy_from_user(&control, argp, sizeof(control)))
3767 r = kvm_vm_ioctl_reinject(kvm, &control);
3773 case KVM_XEN_HVM_CONFIG: {
3775 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3776 sizeof(struct kvm_xen_hvm_config)))
3779 if (kvm->arch.xen_hvm_config.flags)
3784 case KVM_SET_CLOCK: {
3785 struct kvm_clock_data user_ns;
3790 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3798 local_irq_disable();
3799 now_ns = get_kernel_ns();
3800 delta = user_ns.clock - now_ns;
3802 kvm->arch.kvmclock_offset = delta;
3805 case KVM_GET_CLOCK: {
3806 struct kvm_clock_data user_ns;
3809 local_irq_disable();
3810 now_ns = get_kernel_ns();
3811 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3814 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3817 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3830 static void kvm_init_msr_list(void)
3835 /* skip the first msrs in the list. KVM-specific */
3836 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3837 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3840 msrs_to_save[j] = msrs_to_save[i];
3843 num_msrs_to_save = j;
3846 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3854 if (!(vcpu->arch.apic &&
3855 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3856 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3867 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3874 if (!(vcpu->arch.apic &&
3875 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3876 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3878 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3888 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3889 struct kvm_segment *var, int seg)
3891 kvm_x86_ops->set_segment(vcpu, var, seg);
3894 void kvm_get_segment(struct kvm_vcpu *vcpu,
3895 struct kvm_segment *var, int seg)
3897 kvm_x86_ops->get_segment(vcpu, var, seg);
3900 static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3905 static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3908 struct x86_exception exception;
3910 BUG_ON(!mmu_is_nested(vcpu));
3912 /* NPT walks are always user-walks */
3913 access |= PFERR_USER_MASK;
3914 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3919 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3920 struct x86_exception *exception)
3922 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3923 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3926 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3927 struct x86_exception *exception)
3929 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3930 access |= PFERR_FETCH_MASK;
3931 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3934 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3935 struct x86_exception *exception)
3937 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3938 access |= PFERR_WRITE_MASK;
3939 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3942 /* uses this to access any guest's mapped memory without checking CPL */
3943 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3944 struct x86_exception *exception)
3946 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3949 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3950 struct kvm_vcpu *vcpu, u32 access,
3951 struct x86_exception *exception)
3954 int r = X86EMUL_CONTINUE;
3957 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3959 unsigned offset = addr & (PAGE_SIZE-1);
3960 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3963 if (gpa == UNMAPPED_GVA)
3964 return X86EMUL_PROPAGATE_FAULT;
3965 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3967 r = X86EMUL_IO_NEEDED;
3979 /* used for instruction fetching */
3980 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3981 gva_t addr, void *val, unsigned int bytes,
3982 struct x86_exception *exception)
3984 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3985 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3987 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3988 access | PFERR_FETCH_MASK,
3992 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3993 gva_t addr, void *val, unsigned int bytes,
3994 struct x86_exception *exception)
3996 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3997 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3999 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4002 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4004 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4005 gva_t addr, void *val, unsigned int bytes,
4006 struct x86_exception *exception)
4008 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4009 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4012 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4013 gva_t addr, void *val,
4015 struct x86_exception *exception)
4017 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4019 int r = X86EMUL_CONTINUE;
4022 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4025 unsigned offset = addr & (PAGE_SIZE-1);
4026 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4029 if (gpa == UNMAPPED_GVA)
4030 return X86EMUL_PROPAGATE_FAULT;
4031 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
4033 r = X86EMUL_IO_NEEDED;
4044 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4046 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4047 gpa_t *gpa, struct x86_exception *exception,
4050 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4052 if (vcpu_match_mmio_gva(vcpu, gva) &&
4053 check_write_user_access(vcpu, write, access,
4054 vcpu->arch.access)) {
4055 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4056 (gva & (PAGE_SIZE - 1));
4057 trace_vcpu_match_mmio(gva, *gpa, write, false);
4062 access |= PFERR_WRITE_MASK;
4064 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4066 if (*gpa == UNMAPPED_GVA)
4069 /* For APIC access vmexit */
4070 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4073 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4074 trace_vcpu_match_mmio(gva, *gpa, write, true);
4081 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4082 const void *val, int bytes)
4086 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4089 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
4093 struct read_write_emulator_ops {
4094 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4096 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4097 void *val, int bytes);
4098 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4099 int bytes, void *val);
4100 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4101 void *val, int bytes);
4105 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4107 if (vcpu->mmio_read_completed) {
4108 memcpy(val, vcpu->mmio_data, bytes);
4109 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4110 vcpu->mmio_phys_addr, *(u64 *)val);
4111 vcpu->mmio_read_completed = 0;
4118 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4119 void *val, int bytes)
4121 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4124 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4125 void *val, int bytes)
4127 return emulator_write_phys(vcpu, gpa, val, bytes);
4130 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4132 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4133 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4136 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4137 void *val, int bytes)
4139 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4140 return X86EMUL_IO_NEEDED;
4143 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4144 void *val, int bytes)
4146 memcpy(vcpu->mmio_data, val, bytes);
4147 memcpy(vcpu->run->mmio.data, vcpu->mmio_data, 8);
4148 return X86EMUL_CONTINUE;
4151 static struct read_write_emulator_ops read_emultor = {
4152 .read_write_prepare = read_prepare,
4153 .read_write_emulate = read_emulate,
4154 .read_write_mmio = vcpu_mmio_read,
4155 .read_write_exit_mmio = read_exit_mmio,
4158 static struct read_write_emulator_ops write_emultor = {
4159 .read_write_emulate = write_emulate,
4160 .read_write_mmio = write_mmio,
4161 .read_write_exit_mmio = write_exit_mmio,
4165 static int emulator_read_write_onepage(unsigned long addr, void *val,
4167 struct x86_exception *exception,
4168 struct kvm_vcpu *vcpu,
4169 struct read_write_emulator_ops *ops)
4173 bool write = ops->write;
4175 if (ops->read_write_prepare &&
4176 ops->read_write_prepare(vcpu, val, bytes))
4177 return X86EMUL_CONTINUE;
4179 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4182 return X86EMUL_PROPAGATE_FAULT;
4184 /* For APIC access vmexit */
4188 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4189 return X86EMUL_CONTINUE;
4193 * Is this MMIO handled locally?
4195 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4196 if (handled == bytes)
4197 return X86EMUL_CONTINUE;
4203 vcpu->mmio_needed = 1;
4204 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4205 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
4206 vcpu->mmio_size = bytes;
4207 vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
4208 vcpu->run->mmio.is_write = vcpu->mmio_is_write = write;
4209 vcpu->mmio_index = 0;
4211 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4214 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4215 void *val, unsigned int bytes,
4216 struct x86_exception *exception,
4217 struct read_write_emulator_ops *ops)
4219 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4221 /* Crossing a page boundary? */
4222 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4225 now = -addr & ~PAGE_MASK;
4226 rc = emulator_read_write_onepage(addr, val, now, exception,
4229 if (rc != X86EMUL_CONTINUE)
4236 return emulator_read_write_onepage(addr, val, bytes, exception,
4240 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4244 struct x86_exception *exception)
4246 return emulator_read_write(ctxt, addr, val, bytes,
4247 exception, &read_emultor);
4250 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4254 struct x86_exception *exception)
4256 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4257 exception, &write_emultor);
4260 #define CMPXCHG_TYPE(t, ptr, old, new) \
4261 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4263 #ifdef CONFIG_X86_64
4264 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4266 # define CMPXCHG64(ptr, old, new) \
4267 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4270 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4275 struct x86_exception *exception)
4277 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4283 /* guests cmpxchg8b have to be emulated atomically */
4284 if (bytes > 8 || (bytes & (bytes - 1)))
4287 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4289 if (gpa == UNMAPPED_GVA ||
4290 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4293 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4296 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4297 if (is_error_page(page)) {
4298 kvm_release_page_clean(page);
4302 kaddr = kmap_atomic(page, KM_USER0);
4303 kaddr += offset_in_page(gpa);
4306 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4309 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4312 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4315 exchanged = CMPXCHG64(kaddr, old, new);
4320 kunmap_atomic(kaddr, KM_USER0);
4321 kvm_release_page_dirty(page);
4324 return X86EMUL_CMPXCHG_FAILED;
4326 kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
4328 return X86EMUL_CONTINUE;
4331 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4333 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4336 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4338 /* TODO: String I/O for in kernel device */
4341 if (vcpu->arch.pio.in)
4342 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4343 vcpu->arch.pio.size, pd);
4345 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4346 vcpu->arch.pio.port, vcpu->arch.pio.size,
4352 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4353 int size, unsigned short port, void *val,
4356 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4358 if (vcpu->arch.pio.count)
4361 trace_kvm_pio(0, port, size, count);
4363 vcpu->arch.pio.port = port;
4364 vcpu->arch.pio.in = 1;
4365 vcpu->arch.pio.count = count;
4366 vcpu->arch.pio.size = size;
4368 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4370 memcpy(val, vcpu->arch.pio_data, size * count);
4371 vcpu->arch.pio.count = 0;
4375 vcpu->run->exit_reason = KVM_EXIT_IO;
4376 vcpu->run->io.direction = KVM_EXIT_IO_IN;
4377 vcpu->run->io.size = size;
4378 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4379 vcpu->run->io.count = count;
4380 vcpu->run->io.port = port;
4385 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4386 int size, unsigned short port,
4387 const void *val, unsigned int count)
4389 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4391 trace_kvm_pio(1, port, size, count);
4393 vcpu->arch.pio.port = port;
4394 vcpu->arch.pio.in = 0;
4395 vcpu->arch.pio.count = count;
4396 vcpu->arch.pio.size = size;
4398 memcpy(vcpu->arch.pio_data, val, size * count);
4400 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4401 vcpu->arch.pio.count = 0;
4405 vcpu->run->exit_reason = KVM_EXIT_IO;
4406 vcpu->run->io.direction = KVM_EXIT_IO_OUT;
4407 vcpu->run->io.size = size;
4408 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4409 vcpu->run->io.count = count;
4410 vcpu->run->io.port = port;
4415 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4417 return kvm_x86_ops->get_segment_base(vcpu, seg);
4420 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4422 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4425 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4427 if (!need_emulate_wbinvd(vcpu))
4428 return X86EMUL_CONTINUE;
4430 if (kvm_x86_ops->has_wbinvd_exit()) {
4431 int cpu = get_cpu();
4433 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4434 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4435 wbinvd_ipi, NULL, 1);
4437 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4440 return X86EMUL_CONTINUE;
4442 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4444 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4446 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4449 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4451 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4454 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4457 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4460 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4462 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4465 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4467 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4468 unsigned long value;
4472 value = kvm_read_cr0(vcpu);
4475 value = vcpu->arch.cr2;
4478 value = kvm_read_cr3(vcpu);
4481 value = kvm_read_cr4(vcpu);
4484 value = kvm_get_cr8(vcpu);
4487 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4494 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4496 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4501 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4504 vcpu->arch.cr2 = val;
4507 res = kvm_set_cr3(vcpu, val);
4510 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4513 res = kvm_set_cr8(vcpu, val);
4516 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4523 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4525 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4528 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4530 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4533 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4535 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4538 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4540 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4543 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4545 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4548 static unsigned long emulator_get_cached_segment_base(
4549 struct x86_emulate_ctxt *ctxt, int seg)
4551 return get_segment_base(emul_to_vcpu(ctxt), seg);
4554 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4555 struct desc_struct *desc, u32 *base3,
4558 struct kvm_segment var;
4560 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4561 *selector = var.selector;
4568 set_desc_limit(desc, var.limit);
4569 set_desc_base(desc, (unsigned long)var.base);
4570 #ifdef CONFIG_X86_64
4572 *base3 = var.base >> 32;
4574 desc->type = var.type;
4576 desc->dpl = var.dpl;
4577 desc->p = var.present;
4578 desc->avl = var.avl;
4586 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4587 struct desc_struct *desc, u32 base3,
4590 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4591 struct kvm_segment var;
4593 var.selector = selector;
4594 var.base = get_desc_base(desc);
4595 #ifdef CONFIG_X86_64
4596 var.base |= ((u64)base3) << 32;
4598 var.limit = get_desc_limit(desc);
4600 var.limit = (var.limit << 12) | 0xfff;
4601 var.type = desc->type;
4602 var.present = desc->p;
4603 var.dpl = desc->dpl;
4608 var.avl = desc->avl;
4609 var.present = desc->p;
4610 var.unusable = !var.present;
4613 kvm_set_segment(vcpu, &var, seg);
4617 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4618 u32 msr_index, u64 *pdata)
4620 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4623 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4624 u32 msr_index, u64 data)
4626 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
4629 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4631 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4634 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4637 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4639 * CR0.TS may reference the host fpu state, not the guest fpu state,
4640 * so it may be clear at this point.
4645 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4650 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4651 struct x86_instruction_info *info,
4652 enum x86_intercept_stage stage)
4654 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4657 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4658 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4660 struct kvm_cpuid_entry2 *cpuid = NULL;
4663 cpuid = kvm_find_cpuid_entry(emul_to_vcpu(ctxt),
4679 static struct x86_emulate_ops emulate_ops = {
4680 .read_std = kvm_read_guest_virt_system,
4681 .write_std = kvm_write_guest_virt_system,
4682 .fetch = kvm_fetch_guest_virt,
4683 .read_emulated = emulator_read_emulated,
4684 .write_emulated = emulator_write_emulated,
4685 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4686 .invlpg = emulator_invlpg,
4687 .pio_in_emulated = emulator_pio_in_emulated,
4688 .pio_out_emulated = emulator_pio_out_emulated,
4689 .get_segment = emulator_get_segment,
4690 .set_segment = emulator_set_segment,
4691 .get_cached_segment_base = emulator_get_cached_segment_base,
4692 .get_gdt = emulator_get_gdt,
4693 .get_idt = emulator_get_idt,
4694 .set_gdt = emulator_set_gdt,
4695 .set_idt = emulator_set_idt,
4696 .get_cr = emulator_get_cr,
4697 .set_cr = emulator_set_cr,
4698 .cpl = emulator_get_cpl,
4699 .get_dr = emulator_get_dr,
4700 .set_dr = emulator_set_dr,
4701 .set_msr = emulator_set_msr,
4702 .get_msr = emulator_get_msr,
4703 .halt = emulator_halt,
4704 .wbinvd = emulator_wbinvd,
4705 .fix_hypercall = emulator_fix_hypercall,
4706 .get_fpu = emulator_get_fpu,
4707 .put_fpu = emulator_put_fpu,
4708 .intercept = emulator_intercept,
4709 .get_cpuid = emulator_get_cpuid,
4712 static void cache_all_regs(struct kvm_vcpu *vcpu)
4714 kvm_register_read(vcpu, VCPU_REGS_RAX);
4715 kvm_register_read(vcpu, VCPU_REGS_RSP);
4716 kvm_register_read(vcpu, VCPU_REGS_RIP);
4717 vcpu->arch.regs_dirty = ~0;
4720 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4722 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4724 * an sti; sti; sequence only disable interrupts for the first
4725 * instruction. So, if the last instruction, be it emulated or
4726 * not, left the system with the INT_STI flag enabled, it
4727 * means that the last instruction is an sti. We should not
4728 * leave the flag on in this case. The same goes for mov ss
4730 if (!(int_shadow & mask))
4731 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4734 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4736 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4737 if (ctxt->exception.vector == PF_VECTOR)
4738 kvm_propagate_fault(vcpu, &ctxt->exception);
4739 else if (ctxt->exception.error_code_valid)
4740 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4741 ctxt->exception.error_code);
4743 kvm_queue_exception(vcpu, ctxt->exception.vector);
4746 static void init_decode_cache(struct x86_emulate_ctxt *ctxt,
4747 const unsigned long *regs)
4749 memset(&ctxt->twobyte, 0,
4750 (void *)&ctxt->regs - (void *)&ctxt->twobyte);
4751 memcpy(ctxt->regs, regs, sizeof(ctxt->regs));
4753 ctxt->fetch.start = 0;
4754 ctxt->fetch.end = 0;
4755 ctxt->io_read.pos = 0;
4756 ctxt->io_read.end = 0;
4757 ctxt->mem_read.pos = 0;
4758 ctxt->mem_read.end = 0;
4761 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4763 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4767 * TODO: fix emulate.c to use guest_read/write_register
4768 * instead of direct ->regs accesses, can save hundred cycles
4769 * on Intel for instructions that don't read/change RSP, for
4772 cache_all_regs(vcpu);
4774 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4776 ctxt->eflags = kvm_get_rflags(vcpu);
4777 ctxt->eip = kvm_rip_read(vcpu);
4778 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4779 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4780 cs_l ? X86EMUL_MODE_PROT64 :
4781 cs_db ? X86EMUL_MODE_PROT32 :
4782 X86EMUL_MODE_PROT16;
4783 ctxt->guest_mode = is_guest_mode(vcpu);
4785 init_decode_cache(ctxt, vcpu->arch.regs);
4786 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4789 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4791 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4794 init_emulate_ctxt(vcpu);
4798 ctxt->_eip = ctxt->eip + inc_eip;
4799 ret = emulate_int_real(ctxt, irq);
4801 if (ret != X86EMUL_CONTINUE)
4802 return EMULATE_FAIL;
4804 ctxt->eip = ctxt->_eip;
4805 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4806 kvm_rip_write(vcpu, ctxt->eip);
4807 kvm_set_rflags(vcpu, ctxt->eflags);
4809 if (irq == NMI_VECTOR)
4810 vcpu->arch.nmi_pending = 0;
4812 vcpu->arch.interrupt.pending = false;
4814 return EMULATE_DONE;
4816 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4818 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4820 int r = EMULATE_DONE;
4822 ++vcpu->stat.insn_emulation_fail;
4823 trace_kvm_emulate_insn_failed(vcpu);
4824 if (!is_guest_mode(vcpu)) {
4825 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4826 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4827 vcpu->run->internal.ndata = 0;
4830 kvm_queue_exception(vcpu, UD_VECTOR);
4835 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4843 * if emulation was due to access to shadowed page table
4844 * and it failed try to unshadow page and re-entetr the
4845 * guest to let CPU execute the instruction.
4847 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4850 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4852 if (gpa == UNMAPPED_GVA)
4853 return true; /* let cpu generate fault */
4855 if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4861 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4868 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4869 bool writeback = true;
4871 kvm_clear_exception_queue(vcpu);
4873 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4874 init_emulate_ctxt(vcpu);
4875 ctxt->interruptibility = 0;
4876 ctxt->have_exception = false;
4877 ctxt->perm_ok = false;
4879 ctxt->only_vendor_specific_insn
4880 = emulation_type & EMULTYPE_TRAP_UD;
4882 r = x86_decode_insn(ctxt, insn, insn_len);
4884 trace_kvm_emulate_insn_start(vcpu);
4885 ++vcpu->stat.insn_emulation;
4886 if (r != EMULATION_OK) {
4887 if (emulation_type & EMULTYPE_TRAP_UD)
4888 return EMULATE_FAIL;
4889 if (reexecute_instruction(vcpu, cr2))
4890 return EMULATE_DONE;
4891 if (emulation_type & EMULTYPE_SKIP)
4892 return EMULATE_FAIL;
4893 return handle_emulation_failure(vcpu);
4897 if (emulation_type & EMULTYPE_SKIP) {
4898 kvm_rip_write(vcpu, ctxt->_eip);
4899 return EMULATE_DONE;
4902 /* this is needed for vmware backdoor interface to work since it
4903 changes registers values during IO operation */
4904 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4905 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4906 memcpy(ctxt->regs, vcpu->arch.regs, sizeof ctxt->regs);
4910 r = x86_emulate_insn(ctxt);
4912 if (r == EMULATION_INTERCEPTED)
4913 return EMULATE_DONE;
4915 if (r == EMULATION_FAILED) {
4916 if (reexecute_instruction(vcpu, cr2))
4917 return EMULATE_DONE;
4919 return handle_emulation_failure(vcpu);
4922 if (ctxt->have_exception) {
4923 inject_emulated_exception(vcpu);
4925 } else if (vcpu->arch.pio.count) {
4926 if (!vcpu->arch.pio.in)
4927 vcpu->arch.pio.count = 0;
4930 r = EMULATE_DO_MMIO;
4931 } else if (vcpu->mmio_needed) {
4932 if (!vcpu->mmio_is_write)
4934 r = EMULATE_DO_MMIO;
4935 } else if (r == EMULATION_RESTART)
4941 toggle_interruptibility(vcpu, ctxt->interruptibility);
4942 kvm_set_rflags(vcpu, ctxt->eflags);
4943 kvm_make_request(KVM_REQ_EVENT, vcpu);
4944 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4945 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4946 kvm_rip_write(vcpu, ctxt->eip);
4948 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
4952 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4954 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4956 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4957 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4958 size, port, &val, 1);
4959 /* do not return to emulator after return from userspace */
4960 vcpu->arch.pio.count = 0;
4963 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4965 static void tsc_bad(void *info)
4967 __this_cpu_write(cpu_tsc_khz, 0);
4970 static void tsc_khz_changed(void *data)
4972 struct cpufreq_freqs *freq = data;
4973 unsigned long khz = 0;
4977 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4978 khz = cpufreq_quick_get(raw_smp_processor_id());
4981 __this_cpu_write(cpu_tsc_khz, khz);
4984 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4987 struct cpufreq_freqs *freq = data;
4989 struct kvm_vcpu *vcpu;
4990 int i, send_ipi = 0;
4993 * We allow guests to temporarily run on slowing clocks,
4994 * provided we notify them after, or to run on accelerating
4995 * clocks, provided we notify them before. Thus time never
4998 * However, we have a problem. We can't atomically update
4999 * the frequency of a given CPU from this function; it is
5000 * merely a notifier, which can be called from any CPU.
5001 * Changing the TSC frequency at arbitrary points in time
5002 * requires a recomputation of local variables related to
5003 * the TSC for each VCPU. We must flag these local variables
5004 * to be updated and be sure the update takes place with the
5005 * new frequency before any guests proceed.
5007 * Unfortunately, the combination of hotplug CPU and frequency
5008 * change creates an intractable locking scenario; the order
5009 * of when these callouts happen is undefined with respect to
5010 * CPU hotplug, and they can race with each other. As such,
5011 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5012 * undefined; you can actually have a CPU frequency change take
5013 * place in between the computation of X and the setting of the
5014 * variable. To protect against this problem, all updates of
5015 * the per_cpu tsc_khz variable are done in an interrupt
5016 * protected IPI, and all callers wishing to update the value
5017 * must wait for a synchronous IPI to complete (which is trivial
5018 * if the caller is on the CPU already). This establishes the
5019 * necessary total order on variable updates.
5021 * Note that because a guest time update may take place
5022 * anytime after the setting of the VCPU's request bit, the
5023 * correct TSC value must be set before the request. However,
5024 * to ensure the update actually makes it to any guest which
5025 * starts running in hardware virtualization between the set
5026 * and the acquisition of the spinlock, we must also ping the
5027 * CPU after setting the request bit.
5031 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5033 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5036 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5038 raw_spin_lock(&kvm_lock);
5039 list_for_each_entry(kvm, &vm_list, vm_list) {
5040 kvm_for_each_vcpu(i, vcpu, kvm) {
5041 if (vcpu->cpu != freq->cpu)
5043 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5044 if (vcpu->cpu != smp_processor_id())
5048 raw_spin_unlock(&kvm_lock);
5050 if (freq->old < freq->new && send_ipi) {
5052 * We upscale the frequency. Must make the guest
5053 * doesn't see old kvmclock values while running with
5054 * the new frequency, otherwise we risk the guest sees
5055 * time go backwards.
5057 * In case we update the frequency for another cpu
5058 * (which might be in guest context) send an interrupt
5059 * to kick the cpu out of guest context. Next time
5060 * guest context is entered kvmclock will be updated,
5061 * so the guest will not see stale values.
5063 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5068 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5069 .notifier_call = kvmclock_cpufreq_notifier
5072 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5073 unsigned long action, void *hcpu)
5075 unsigned int cpu = (unsigned long)hcpu;
5079 case CPU_DOWN_FAILED:
5080 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5082 case CPU_DOWN_PREPARE:
5083 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5089 static struct notifier_block kvmclock_cpu_notifier_block = {
5090 .notifier_call = kvmclock_cpu_notifier,
5091 .priority = -INT_MAX
5094 static void kvm_timer_init(void)
5098 max_tsc_khz = tsc_khz;
5099 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5100 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5101 #ifdef CONFIG_CPU_FREQ
5102 struct cpufreq_policy policy;
5103 memset(&policy, 0, sizeof(policy));
5105 cpufreq_get_policy(&policy, cpu);
5106 if (policy.cpuinfo.max_freq)
5107 max_tsc_khz = policy.cpuinfo.max_freq;
5110 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5111 CPUFREQ_TRANSITION_NOTIFIER);
5113 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5114 for_each_online_cpu(cpu)
5115 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5118 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5120 static int kvm_is_in_guest(void)
5122 return percpu_read(current_vcpu) != NULL;
5125 static int kvm_is_user_mode(void)
5129 if (percpu_read(current_vcpu))
5130 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
5132 return user_mode != 0;
5135 static unsigned long kvm_get_guest_ip(void)
5137 unsigned long ip = 0;
5139 if (percpu_read(current_vcpu))
5140 ip = kvm_rip_read(percpu_read(current_vcpu));
5145 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5146 .is_in_guest = kvm_is_in_guest,
5147 .is_user_mode = kvm_is_user_mode,
5148 .get_guest_ip = kvm_get_guest_ip,
5151 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5153 percpu_write(current_vcpu, vcpu);
5155 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5157 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5159 percpu_write(current_vcpu, NULL);
5161 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5163 static void kvm_set_mmio_spte_mask(void)
5166 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5169 * Set the reserved bits and the present bit of an paging-structure
5170 * entry to generate page fault with PFER.RSV = 1.
5172 mask = ((1ull << (62 - maxphyaddr + 1)) - 1) << maxphyaddr;
5175 #ifdef CONFIG_X86_64
5177 * If reserved bit is not supported, clear the present bit to disable
5180 if (maxphyaddr == 52)
5184 kvm_mmu_set_mmio_spte_mask(mask);
5187 int kvm_arch_init(void *opaque)
5190 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
5193 printk(KERN_ERR "kvm: already loaded the other module\n");
5198 if (!ops->cpu_has_kvm_support()) {
5199 printk(KERN_ERR "kvm: no hardware support\n");
5203 if (ops->disabled_by_bios()) {
5204 printk(KERN_ERR "kvm: disabled by bios\n");
5209 r = kvm_mmu_module_init();
5213 kvm_set_mmio_spte_mask();
5214 kvm_init_msr_list();
5217 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5218 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5222 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5225 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5233 void kvm_arch_exit(void)
5235 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5237 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5238 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5239 CPUFREQ_TRANSITION_NOTIFIER);
5240 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5242 kvm_mmu_module_exit();
5245 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5247 ++vcpu->stat.halt_exits;
5248 if (irqchip_in_kernel(vcpu->kvm)) {
5249 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5252 vcpu->run->exit_reason = KVM_EXIT_HLT;
5256 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5258 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
5261 if (is_long_mode(vcpu))
5264 return a0 | ((gpa_t)a1 << 32);
5267 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5269 u64 param, ingpa, outgpa, ret;
5270 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5271 bool fast, longmode;
5275 * hypercall generates UD from non zero cpl and real mode
5278 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5279 kvm_queue_exception(vcpu, UD_VECTOR);
5283 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5284 longmode = is_long_mode(vcpu) && cs_l == 1;
5287 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5288 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5289 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5290 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5291 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5292 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5294 #ifdef CONFIG_X86_64
5296 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5297 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5298 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5302 code = param & 0xffff;
5303 fast = (param >> 16) & 0x1;
5304 rep_cnt = (param >> 32) & 0xfff;
5305 rep_idx = (param >> 48) & 0xfff;
5307 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5310 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5311 kvm_vcpu_on_spin(vcpu);
5314 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5318 ret = res | (((u64)rep_done & 0xfff) << 32);
5320 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5322 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5323 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5329 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5331 unsigned long nr, a0, a1, a2, a3, ret;
5334 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5335 return kvm_hv_hypercall(vcpu);
5337 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5338 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5339 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5340 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5341 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5343 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5345 if (!is_long_mode(vcpu)) {
5353 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5359 case KVM_HC_VAPIC_POLL_IRQ:
5363 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
5370 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5371 ++vcpu->stat.hypercalls;
5374 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5376 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5378 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5379 char instruction[3];
5380 unsigned long rip = kvm_rip_read(vcpu);
5383 * Blow out the MMU to ensure that no other VCPU has an active mapping
5384 * to ensure that the updated hypercall appears atomically across all
5387 kvm_mmu_zap_all(vcpu->kvm);
5389 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5391 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5394 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
5396 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
5397 int j, nent = vcpu->arch.cpuid_nent;
5399 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
5400 /* when no next entry is found, the current entry[i] is reselected */
5401 for (j = i + 1; ; j = (j + 1) % nent) {
5402 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
5403 if (ej->function == e->function) {
5404 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
5408 return 0; /* silence gcc, even though control never reaches here */
5411 /* find an entry with matching function, matching index (if needed), and that
5412 * should be read next (if it's stateful) */
5413 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
5414 u32 function, u32 index)
5416 if (e->function != function)
5418 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
5420 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
5421 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
5426 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
5427 u32 function, u32 index)
5430 struct kvm_cpuid_entry2 *best = NULL;
5432 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
5433 struct kvm_cpuid_entry2 *e;
5435 e = &vcpu->arch.cpuid_entries[i];
5436 if (is_matching_cpuid_entry(e, function, index)) {
5437 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
5438 move_to_next_stateful_cpuid_entry(vcpu, i);
5445 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
5447 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
5449 struct kvm_cpuid_entry2 *best;
5451 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
5452 if (!best || best->eax < 0x80000008)
5454 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
5456 return best->eax & 0xff;
5462 * If no match is found, check whether we exceed the vCPU's limit
5463 * and return the content of the highest valid _standard_ leaf instead.
5464 * This is to satisfy the CPUID specification.
5466 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
5467 u32 function, u32 index)
5469 struct kvm_cpuid_entry2 *maxlevel;
5471 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
5472 if (!maxlevel || maxlevel->eax >= function)
5474 if (function & 0x80000000) {
5475 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
5479 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
5482 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
5484 u32 function, index;
5485 struct kvm_cpuid_entry2 *best;
5487 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
5488 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5489 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
5490 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
5491 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
5492 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
5493 best = kvm_find_cpuid_entry(vcpu, function, index);
5496 best = check_cpuid_limit(vcpu, function, index);
5499 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
5500 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
5501 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
5502 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
5504 kvm_x86_ops->skip_emulated_instruction(vcpu);
5505 trace_kvm_cpuid(function,
5506 kvm_register_read(vcpu, VCPU_REGS_RAX),
5507 kvm_register_read(vcpu, VCPU_REGS_RBX),
5508 kvm_register_read(vcpu, VCPU_REGS_RCX),
5509 kvm_register_read(vcpu, VCPU_REGS_RDX));
5511 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
5514 * Check if userspace requested an interrupt window, and that the
5515 * interrupt window is open.
5517 * No need to exit to userspace if we already have an interrupt queued.
5519 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5521 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5522 vcpu->run->request_interrupt_window &&
5523 kvm_arch_interrupt_allowed(vcpu));
5526 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5528 struct kvm_run *kvm_run = vcpu->run;
5530 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5531 kvm_run->cr8 = kvm_get_cr8(vcpu);
5532 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5533 if (irqchip_in_kernel(vcpu->kvm))
5534 kvm_run->ready_for_interrupt_injection = 1;
5536 kvm_run->ready_for_interrupt_injection =
5537 kvm_arch_interrupt_allowed(vcpu) &&
5538 !kvm_cpu_has_interrupt(vcpu) &&
5539 !kvm_event_needs_reinjection(vcpu);
5542 static void vapic_enter(struct kvm_vcpu *vcpu)
5544 struct kvm_lapic *apic = vcpu->arch.apic;
5547 if (!apic || !apic->vapic_addr)
5550 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5552 vcpu->arch.apic->vapic_page = page;
5555 static void vapic_exit(struct kvm_vcpu *vcpu)
5557 struct kvm_lapic *apic = vcpu->arch.apic;
5560 if (!apic || !apic->vapic_addr)
5563 idx = srcu_read_lock(&vcpu->kvm->srcu);
5564 kvm_release_page_dirty(apic->vapic_page);
5565 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5566 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5569 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5573 if (!kvm_x86_ops->update_cr8_intercept)
5576 if (!vcpu->arch.apic)
5579 if (!vcpu->arch.apic->vapic_addr)
5580 max_irr = kvm_lapic_find_highest_irr(vcpu);
5587 tpr = kvm_lapic_get_cr8(vcpu);
5589 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5592 static void inject_pending_event(struct kvm_vcpu *vcpu)
5594 /* try to reinject previous events if any */
5595 if (vcpu->arch.exception.pending) {
5596 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5597 vcpu->arch.exception.has_error_code,
5598 vcpu->arch.exception.error_code);
5599 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5600 vcpu->arch.exception.has_error_code,
5601 vcpu->arch.exception.error_code,
5602 vcpu->arch.exception.reinject);
5606 if (vcpu->arch.nmi_injected) {
5607 kvm_x86_ops->set_nmi(vcpu);
5611 if (vcpu->arch.interrupt.pending) {
5612 kvm_x86_ops->set_irq(vcpu);
5616 /* try to inject new event if pending */
5617 if (vcpu->arch.nmi_pending) {
5618 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5619 --vcpu->arch.nmi_pending;
5620 vcpu->arch.nmi_injected = true;
5621 kvm_x86_ops->set_nmi(vcpu);
5623 } else if (kvm_cpu_has_interrupt(vcpu)) {
5624 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5625 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5627 kvm_x86_ops->set_irq(vcpu);
5632 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5634 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5635 !vcpu->guest_xcr0_loaded) {
5636 /* kvm_set_xcr() also depends on this */
5637 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5638 vcpu->guest_xcr0_loaded = 1;
5642 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5644 if (vcpu->guest_xcr0_loaded) {
5645 if (vcpu->arch.xcr0 != host_xcr0)
5646 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5647 vcpu->guest_xcr0_loaded = 0;
5651 static void process_nmi(struct kvm_vcpu *vcpu)
5656 * x86 is limited to one NMI running, and one NMI pending after it.
5657 * If an NMI is already in progress, limit further NMIs to just one.
5658 * Otherwise, allow two (and we'll inject the first one immediately).
5660 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5663 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5664 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5665 kvm_make_request(KVM_REQ_EVENT, vcpu);
5668 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5671 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5672 vcpu->run->request_interrupt_window;
5674 if (vcpu->requests) {
5675 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5676 kvm_mmu_unload(vcpu);
5677 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5678 __kvm_migrate_timers(vcpu);
5679 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5680 r = kvm_guest_time_update(vcpu);
5684 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5685 kvm_mmu_sync_roots(vcpu);
5686 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5687 kvm_x86_ops->tlb_flush(vcpu);
5688 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5689 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5693 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5694 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5698 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5699 vcpu->fpu_active = 0;
5700 kvm_x86_ops->fpu_deactivate(vcpu);
5702 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5703 /* Page is swapped out. Do synthetic halt */
5704 vcpu->arch.apf.halted = true;
5708 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5709 record_steal_time(vcpu);
5710 if (kvm_check_request(KVM_REQ_NMI, vcpu))
5715 r = kvm_mmu_reload(vcpu);
5719 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5720 inject_pending_event(vcpu);
5722 /* enable NMI/IRQ window open exits if needed */
5723 if (vcpu->arch.nmi_pending)
5724 kvm_x86_ops->enable_nmi_window(vcpu);
5725 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5726 kvm_x86_ops->enable_irq_window(vcpu);
5728 if (kvm_lapic_enabled(vcpu)) {
5729 update_cr8_intercept(vcpu);
5730 kvm_lapic_sync_to_vapic(vcpu);
5736 kvm_x86_ops->prepare_guest_switch(vcpu);
5737 if (vcpu->fpu_active)
5738 kvm_load_guest_fpu(vcpu);
5739 kvm_load_guest_xcr0(vcpu);
5741 vcpu->mode = IN_GUEST_MODE;
5743 /* We should set ->mode before check ->requests,
5744 * see the comment in make_all_cpus_request.
5748 local_irq_disable();
5750 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5751 || need_resched() || signal_pending(current)) {
5752 vcpu->mode = OUTSIDE_GUEST_MODE;
5756 kvm_x86_ops->cancel_injection(vcpu);
5761 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5765 if (unlikely(vcpu->arch.switch_db_regs)) {
5767 set_debugreg(vcpu->arch.eff_db[0], 0);
5768 set_debugreg(vcpu->arch.eff_db[1], 1);
5769 set_debugreg(vcpu->arch.eff_db[2], 2);
5770 set_debugreg(vcpu->arch.eff_db[3], 3);
5773 trace_kvm_entry(vcpu->vcpu_id);
5774 kvm_x86_ops->run(vcpu);
5777 * If the guest has used debug registers, at least dr7
5778 * will be disabled while returning to the host.
5779 * If we don't have active breakpoints in the host, we don't
5780 * care about the messed up debug address registers. But if
5781 * we have some of them active, restore the old state.
5783 if (hw_breakpoint_active())
5784 hw_breakpoint_restore();
5786 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu);
5788 vcpu->mode = OUTSIDE_GUEST_MODE;
5795 * We must have an instruction between local_irq_enable() and
5796 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5797 * the interrupt shadow. The stat.exits increment will do nicely.
5798 * But we need to prevent reordering, hence this barrier():
5806 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5809 * Profile KVM exit RIPs:
5811 if (unlikely(prof_on == KVM_PROFILING)) {
5812 unsigned long rip = kvm_rip_read(vcpu);
5813 profile_hit(KVM_PROFILING, (void *)rip);
5817 kvm_lapic_sync_from_vapic(vcpu);
5819 r = kvm_x86_ops->handle_exit(vcpu);
5825 static int __vcpu_run(struct kvm_vcpu *vcpu)
5828 struct kvm *kvm = vcpu->kvm;
5830 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5831 pr_debug("vcpu %d received sipi with vector # %x\n",
5832 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5833 kvm_lapic_reset(vcpu);
5834 r = kvm_arch_vcpu_reset(vcpu);
5837 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5840 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5845 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5846 !vcpu->arch.apf.halted)
5847 r = vcpu_enter_guest(vcpu);
5849 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5850 kvm_vcpu_block(vcpu);
5851 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5852 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5854 switch(vcpu->arch.mp_state) {
5855 case KVM_MP_STATE_HALTED:
5856 vcpu->arch.mp_state =
5857 KVM_MP_STATE_RUNNABLE;
5858 case KVM_MP_STATE_RUNNABLE:
5859 vcpu->arch.apf.halted = false;
5861 case KVM_MP_STATE_SIPI_RECEIVED:
5872 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5873 if (kvm_cpu_has_pending_timer(vcpu))
5874 kvm_inject_pending_timer_irqs(vcpu);
5876 if (dm_request_for_irq_injection(vcpu)) {
5878 vcpu->run->exit_reason = KVM_EXIT_INTR;
5879 ++vcpu->stat.request_irq_exits;
5882 kvm_check_async_pf_completion(vcpu);
5884 if (signal_pending(current)) {
5886 vcpu->run->exit_reason = KVM_EXIT_INTR;
5887 ++vcpu->stat.signal_exits;
5889 if (need_resched()) {
5890 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5892 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5896 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5903 static int complete_mmio(struct kvm_vcpu *vcpu)
5905 struct kvm_run *run = vcpu->run;
5908 if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5911 if (vcpu->mmio_needed) {
5912 vcpu->mmio_needed = 0;
5913 if (!vcpu->mmio_is_write)
5914 memcpy(vcpu->mmio_data + vcpu->mmio_index,
5916 vcpu->mmio_index += 8;
5917 if (vcpu->mmio_index < vcpu->mmio_size) {
5918 run->exit_reason = KVM_EXIT_MMIO;
5919 run->mmio.phys_addr = vcpu->mmio_phys_addr + vcpu->mmio_index;
5920 memcpy(run->mmio.data, vcpu->mmio_data + vcpu->mmio_index, 8);
5921 run->mmio.len = min(vcpu->mmio_size - vcpu->mmio_index, 8);
5922 run->mmio.is_write = vcpu->mmio_is_write;
5923 vcpu->mmio_needed = 1;
5926 if (vcpu->mmio_is_write)
5928 vcpu->mmio_read_completed = 1;
5930 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5931 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5932 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5933 if (r != EMULATE_DONE)
5938 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5943 if (!tsk_used_math(current) && init_fpu(current))
5946 if (vcpu->sigset_active)
5947 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5949 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5950 kvm_vcpu_block(vcpu);
5951 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5956 /* re-sync apic's tpr */
5957 if (!irqchip_in_kernel(vcpu->kvm)) {
5958 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5964 r = complete_mmio(vcpu);
5968 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5969 kvm_register_write(vcpu, VCPU_REGS_RAX,
5970 kvm_run->hypercall.ret);
5972 r = __vcpu_run(vcpu);
5975 post_kvm_run_save(vcpu);
5976 if (vcpu->sigset_active)
5977 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5982 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5984 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5986 * We are here if userspace calls get_regs() in the middle of
5987 * instruction emulation. Registers state needs to be copied
5988 * back from emulation context to vcpu. Usrapace shouldn't do
5989 * that usually, but some bad designed PV devices (vmware
5990 * backdoor interface) need this to work
5992 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5993 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
5994 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5996 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5997 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5998 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5999 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6000 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6001 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6002 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6003 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6004 #ifdef CONFIG_X86_64
6005 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6006 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6007 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6008 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6009 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6010 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6011 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6012 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6015 regs->rip = kvm_rip_read(vcpu);
6016 regs->rflags = kvm_get_rflags(vcpu);
6021 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6023 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6024 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6026 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6027 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6028 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6029 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6030 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6031 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6032 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6033 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6034 #ifdef CONFIG_X86_64
6035 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6036 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6037 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6038 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6039 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6040 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6041 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6042 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6045 kvm_rip_write(vcpu, regs->rip);
6046 kvm_set_rflags(vcpu, regs->rflags);
6048 vcpu->arch.exception.pending = false;
6050 kvm_make_request(KVM_REQ_EVENT, vcpu);
6055 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6057 struct kvm_segment cs;
6059 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6063 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6065 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6066 struct kvm_sregs *sregs)
6070 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6071 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6072 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6073 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6074 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6075 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6077 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6078 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6080 kvm_x86_ops->get_idt(vcpu, &dt);
6081 sregs->idt.limit = dt.size;
6082 sregs->idt.base = dt.address;
6083 kvm_x86_ops->get_gdt(vcpu, &dt);
6084 sregs->gdt.limit = dt.size;
6085 sregs->gdt.base = dt.address;
6087 sregs->cr0 = kvm_read_cr0(vcpu);
6088 sregs->cr2 = vcpu->arch.cr2;
6089 sregs->cr3 = kvm_read_cr3(vcpu);
6090 sregs->cr4 = kvm_read_cr4(vcpu);
6091 sregs->cr8 = kvm_get_cr8(vcpu);
6092 sregs->efer = vcpu->arch.efer;
6093 sregs->apic_base = kvm_get_apic_base(vcpu);
6095 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6097 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6098 set_bit(vcpu->arch.interrupt.nr,
6099 (unsigned long *)sregs->interrupt_bitmap);
6104 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6105 struct kvm_mp_state *mp_state)
6107 mp_state->mp_state = vcpu->arch.mp_state;
6111 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6112 struct kvm_mp_state *mp_state)
6114 vcpu->arch.mp_state = mp_state->mp_state;
6115 kvm_make_request(KVM_REQ_EVENT, vcpu);
6119 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
6120 bool has_error_code, u32 error_code)
6122 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6125 init_emulate_ctxt(vcpu);
6127 ret = emulator_task_switch(ctxt, tss_selector, reason,
6128 has_error_code, error_code);
6131 return EMULATE_FAIL;
6133 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
6134 kvm_rip_write(vcpu, ctxt->eip);
6135 kvm_set_rflags(vcpu, ctxt->eflags);
6136 kvm_make_request(KVM_REQ_EVENT, vcpu);
6137 return EMULATE_DONE;
6139 EXPORT_SYMBOL_GPL(kvm_task_switch);
6141 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6142 struct kvm_sregs *sregs)
6144 int mmu_reset_needed = 0;
6145 int pending_vec, max_bits, idx;
6148 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6151 dt.size = sregs->idt.limit;
6152 dt.address = sregs->idt.base;
6153 kvm_x86_ops->set_idt(vcpu, &dt);
6154 dt.size = sregs->gdt.limit;
6155 dt.address = sregs->gdt.base;
6156 kvm_x86_ops->set_gdt(vcpu, &dt);
6158 vcpu->arch.cr2 = sregs->cr2;
6159 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6160 vcpu->arch.cr3 = sregs->cr3;
6161 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6163 kvm_set_cr8(vcpu, sregs->cr8);
6165 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6166 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6167 kvm_set_apic_base(vcpu, sregs->apic_base);
6169 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6170 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6171 vcpu->arch.cr0 = sregs->cr0;
6173 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6174 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6175 if (sregs->cr4 & X86_CR4_OSXSAVE)
6178 idx = srcu_read_lock(&vcpu->kvm->srcu);
6179 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6180 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6181 mmu_reset_needed = 1;
6183 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6185 if (mmu_reset_needed)
6186 kvm_mmu_reset_context(vcpu);
6188 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
6189 pending_vec = find_first_bit(
6190 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6191 if (pending_vec < max_bits) {
6192 kvm_queue_interrupt(vcpu, pending_vec, false);
6193 pr_debug("Set back pending irq %d\n", pending_vec);
6196 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6197 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6198 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6199 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6200 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6201 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6203 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6204 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6206 update_cr8_intercept(vcpu);
6208 /* Older userspace won't unhalt the vcpu on reset. */
6209 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6210 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6212 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6214 kvm_make_request(KVM_REQ_EVENT, vcpu);
6219 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6220 struct kvm_guest_debug *dbg)
6222 unsigned long rflags;
6225 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6227 if (vcpu->arch.exception.pending)
6229 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6230 kvm_queue_exception(vcpu, DB_VECTOR);
6232 kvm_queue_exception(vcpu, BP_VECTOR);
6236 * Read rflags as long as potentially injected trace flags are still
6239 rflags = kvm_get_rflags(vcpu);
6241 vcpu->guest_debug = dbg->control;
6242 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6243 vcpu->guest_debug = 0;
6245 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6246 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6247 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6248 vcpu->arch.switch_db_regs =
6249 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
6251 for (i = 0; i < KVM_NR_DB_REGS; i++)
6252 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6253 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
6256 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6257 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6258 get_segment_base(vcpu, VCPU_SREG_CS);
6261 * Trigger an rflags update that will inject or remove the trace
6264 kvm_set_rflags(vcpu, rflags);
6266 kvm_x86_ops->set_guest_debug(vcpu, dbg);
6276 * Translate a guest virtual address to a guest physical address.
6278 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6279 struct kvm_translation *tr)
6281 unsigned long vaddr = tr->linear_address;
6285 idx = srcu_read_lock(&vcpu->kvm->srcu);
6286 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6287 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6288 tr->physical_address = gpa;
6289 tr->valid = gpa != UNMAPPED_GVA;
6296 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6298 struct i387_fxsave_struct *fxsave =
6299 &vcpu->arch.guest_fpu.state->fxsave;
6301 memcpy(fpu->fpr, fxsave->st_space, 128);
6302 fpu->fcw = fxsave->cwd;
6303 fpu->fsw = fxsave->swd;
6304 fpu->ftwx = fxsave->twd;
6305 fpu->last_opcode = fxsave->fop;
6306 fpu->last_ip = fxsave->rip;
6307 fpu->last_dp = fxsave->rdp;
6308 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6313 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6315 struct i387_fxsave_struct *fxsave =
6316 &vcpu->arch.guest_fpu.state->fxsave;
6318 memcpy(fxsave->st_space, fpu->fpr, 128);
6319 fxsave->cwd = fpu->fcw;
6320 fxsave->swd = fpu->fsw;
6321 fxsave->twd = fpu->ftwx;
6322 fxsave->fop = fpu->last_opcode;
6323 fxsave->rip = fpu->last_ip;
6324 fxsave->rdp = fpu->last_dp;
6325 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6330 int fx_init(struct kvm_vcpu *vcpu)
6334 err = fpu_alloc(&vcpu->arch.guest_fpu);
6338 fpu_finit(&vcpu->arch.guest_fpu);
6341 * Ensure guest xcr0 is valid for loading
6343 vcpu->arch.xcr0 = XSTATE_FP;
6345 vcpu->arch.cr0 |= X86_CR0_ET;
6349 EXPORT_SYMBOL_GPL(fx_init);
6351 static void fx_free(struct kvm_vcpu *vcpu)
6353 fpu_free(&vcpu->arch.guest_fpu);
6356 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6358 if (vcpu->guest_fpu_loaded)
6362 * Restore all possible states in the guest,
6363 * and assume host would use all available bits.
6364 * Guest xcr0 would be loaded later.
6366 kvm_put_guest_xcr0(vcpu);
6367 vcpu->guest_fpu_loaded = 1;
6368 unlazy_fpu(current);
6369 fpu_restore_checking(&vcpu->arch.guest_fpu);
6373 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6375 kvm_put_guest_xcr0(vcpu);
6377 if (!vcpu->guest_fpu_loaded)
6380 vcpu->guest_fpu_loaded = 0;
6381 fpu_save_init(&vcpu->arch.guest_fpu);
6382 ++vcpu->stat.fpu_reload;
6383 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6387 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6389 kvmclock_reset(vcpu);
6391 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6393 kvm_x86_ops->vcpu_free(vcpu);
6396 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6399 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6400 printk_once(KERN_WARNING
6401 "kvm: SMP vm created on host with unstable TSC; "
6402 "guest TSC will not be reliable\n");
6403 return kvm_x86_ops->vcpu_create(kvm, id);
6406 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6410 vcpu->arch.mtrr_state.have_fixed = 1;
6412 r = kvm_arch_vcpu_reset(vcpu);
6414 r = kvm_mmu_setup(vcpu);
6420 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6422 vcpu->arch.apf.msr_val = 0;
6425 kvm_mmu_unload(vcpu);
6429 kvm_x86_ops->vcpu_free(vcpu);
6432 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
6434 atomic_set(&vcpu->arch.nmi_queued, 0);
6435 vcpu->arch.nmi_pending = 0;
6436 vcpu->arch.nmi_injected = false;
6438 vcpu->arch.switch_db_regs = 0;
6439 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6440 vcpu->arch.dr6 = DR6_FIXED_1;
6441 vcpu->arch.dr7 = DR7_FIXED_1;
6443 kvm_make_request(KVM_REQ_EVENT, vcpu);
6444 vcpu->arch.apf.msr_val = 0;
6445 vcpu->arch.st.msr_val = 0;
6447 kvmclock_reset(vcpu);
6449 kvm_clear_async_pf_completion_queue(vcpu);
6450 kvm_async_pf_hash_reset(vcpu);
6451 vcpu->arch.apf.halted = false;
6453 return kvm_x86_ops->vcpu_reset(vcpu);
6456 int kvm_arch_hardware_enable(void *garbage)
6459 struct kvm_vcpu *vcpu;
6462 kvm_shared_msr_cpu_online();
6463 list_for_each_entry(kvm, &vm_list, vm_list)
6464 kvm_for_each_vcpu(i, vcpu, kvm)
6465 if (vcpu->cpu == smp_processor_id())
6466 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6467 return kvm_x86_ops->hardware_enable(garbage);
6470 void kvm_arch_hardware_disable(void *garbage)
6472 kvm_x86_ops->hardware_disable(garbage);
6473 drop_user_return_notifiers(garbage);
6476 int kvm_arch_hardware_setup(void)
6478 return kvm_x86_ops->hardware_setup();
6481 void kvm_arch_hardware_unsetup(void)
6483 kvm_x86_ops->hardware_unsetup();
6486 void kvm_arch_check_processor_compat(void *rtn)
6488 kvm_x86_ops->check_processor_compatibility(rtn);
6491 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
6493 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
6496 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6502 BUG_ON(vcpu->kvm == NULL);
6505 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6506 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
6507 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
6508 vcpu->arch.mmu.translate_gpa = translate_gpa;
6509 vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
6510 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6511 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6513 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6515 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6520 vcpu->arch.pio_data = page_address(page);
6522 kvm_init_tsc_catchup(vcpu, max_tsc_khz);
6524 r = kvm_mmu_create(vcpu);
6526 goto fail_free_pio_data;
6528 if (irqchip_in_kernel(kvm)) {
6529 r = kvm_create_lapic(vcpu);
6531 goto fail_mmu_destroy;
6534 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6536 if (!vcpu->arch.mce_banks) {
6538 goto fail_free_lapic;
6540 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6542 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6543 goto fail_free_mce_banks;
6545 vcpu->arch.pv_time_enabled = false;
6546 kvm_async_pf_hash_reset(vcpu);
6549 fail_free_mce_banks:
6550 kfree(vcpu->arch.mce_banks);
6552 kvm_free_lapic(vcpu);
6554 kvm_mmu_destroy(vcpu);
6556 free_page((unsigned long)vcpu->arch.pio_data);
6561 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6565 kfree(vcpu->arch.mce_banks);
6566 kvm_free_lapic(vcpu);
6567 idx = srcu_read_lock(&vcpu->kvm->srcu);
6568 kvm_mmu_destroy(vcpu);
6569 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6570 free_page((unsigned long)vcpu->arch.pio_data);
6573 int kvm_arch_init_vm(struct kvm *kvm)
6575 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6576 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6578 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6579 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6581 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6586 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6589 kvm_mmu_unload(vcpu);
6593 static void kvm_free_vcpus(struct kvm *kvm)
6596 struct kvm_vcpu *vcpu;
6599 * Unpin any mmu pages first.
6601 kvm_for_each_vcpu(i, vcpu, kvm) {
6602 kvm_clear_async_pf_completion_queue(vcpu);
6603 kvm_unload_vcpu_mmu(vcpu);
6605 kvm_for_each_vcpu(i, vcpu, kvm)
6606 kvm_arch_vcpu_free(vcpu);
6608 mutex_lock(&kvm->lock);
6609 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6610 kvm->vcpus[i] = NULL;
6612 atomic_set(&kvm->online_vcpus, 0);
6613 mutex_unlock(&kvm->lock);
6616 void kvm_arch_sync_events(struct kvm *kvm)
6618 kvm_free_all_assigned_devices(kvm);
6622 void kvm_arch_destroy_vm(struct kvm *kvm)
6624 kvm_iommu_unmap_guest(kvm);
6625 kfree(kvm->arch.vpic);
6626 kfree(kvm->arch.vioapic);
6627 kvm_free_vcpus(kvm);
6628 if (kvm->arch.apic_access_page)
6629 put_page(kvm->arch.apic_access_page);
6630 if (kvm->arch.ept_identity_pagetable)
6631 put_page(kvm->arch.ept_identity_pagetable);
6634 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6635 struct kvm_memory_slot *memslot,
6636 struct kvm_memory_slot old,
6637 struct kvm_userspace_memory_region *mem,
6640 int npages = memslot->npages;
6641 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6643 /* Prevent internal slot pages from being moved by fork()/COW. */
6644 if (memslot->id >= KVM_MEMORY_SLOTS)
6645 map_flags = MAP_SHARED | MAP_ANONYMOUS;
6647 /*To keep backward compatibility with older userspace,
6648 *x86 needs to hanlde !user_alloc case.
6651 if (npages && !old.rmap) {
6652 unsigned long userspace_addr;
6654 down_write(¤t->mm->mmap_sem);
6655 userspace_addr = do_mmap(NULL, 0,
6657 PROT_READ | PROT_WRITE,
6660 up_write(¤t->mm->mmap_sem);
6662 if (IS_ERR((void *)userspace_addr))
6663 return PTR_ERR((void *)userspace_addr);
6665 memslot->userspace_addr = userspace_addr;
6673 void kvm_arch_commit_memory_region(struct kvm *kvm,
6674 struct kvm_userspace_memory_region *mem,
6675 struct kvm_memory_slot old,
6679 int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
6681 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6684 down_write(¤t->mm->mmap_sem);
6685 ret = do_munmap(current->mm, old.userspace_addr,
6686 old.npages * PAGE_SIZE);
6687 up_write(¤t->mm->mmap_sem);
6690 "kvm_vm_ioctl_set_memory_region: "
6691 "failed to munmap memory\n");
6694 if (!kvm->arch.n_requested_mmu_pages)
6695 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6697 spin_lock(&kvm->mmu_lock);
6699 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6700 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6701 spin_unlock(&kvm->mmu_lock);
6704 void kvm_arch_flush_shadow(struct kvm *kvm)
6706 kvm_mmu_zap_all(kvm);
6707 kvm_reload_remote_mmus(kvm);
6710 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6712 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6713 !vcpu->arch.apf.halted)
6714 || !list_empty_careful(&vcpu->async_pf.done)
6715 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6716 || atomic_read(&vcpu->arch.nmi_queued) ||
6717 (kvm_arch_interrupt_allowed(vcpu) &&
6718 kvm_cpu_has_interrupt(vcpu));
6721 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6724 int cpu = vcpu->cpu;
6726 if (waitqueue_active(&vcpu->wq)) {
6727 wake_up_interruptible(&vcpu->wq);
6728 ++vcpu->stat.halt_wakeup;
6732 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6733 if (kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE)
6734 smp_send_reschedule(cpu);
6738 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6740 return kvm_x86_ops->interrupt_allowed(vcpu);
6743 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6745 unsigned long current_rip = kvm_rip_read(vcpu) +
6746 get_segment_base(vcpu, VCPU_SREG_CS);
6748 return current_rip == linear_rip;
6750 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6752 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6754 unsigned long rflags;
6756 rflags = kvm_x86_ops->get_rflags(vcpu);
6757 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6758 rflags &= ~X86_EFLAGS_TF;
6761 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6763 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6765 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6766 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6767 rflags |= X86_EFLAGS_TF;
6768 kvm_x86_ops->set_rflags(vcpu, rflags);
6769 kvm_make_request(KVM_REQ_EVENT, vcpu);
6771 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6773 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6777 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6778 is_error_page(work->page))
6781 r = kvm_mmu_reload(vcpu);
6785 if (!vcpu->arch.mmu.direct_map &&
6786 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6789 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6792 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6794 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6797 static inline u32 kvm_async_pf_next_probe(u32 key)
6799 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6802 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6804 u32 key = kvm_async_pf_hash_fn(gfn);
6806 while (vcpu->arch.apf.gfns[key] != ~0)
6807 key = kvm_async_pf_next_probe(key);
6809 vcpu->arch.apf.gfns[key] = gfn;
6812 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6815 u32 key = kvm_async_pf_hash_fn(gfn);
6817 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6818 (vcpu->arch.apf.gfns[key] != gfn &&
6819 vcpu->arch.apf.gfns[key] != ~0); i++)
6820 key = kvm_async_pf_next_probe(key);
6825 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6827 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6830 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6834 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6836 vcpu->arch.apf.gfns[i] = ~0;
6838 j = kvm_async_pf_next_probe(j);
6839 if (vcpu->arch.apf.gfns[j] == ~0)
6841 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6843 * k lies cyclically in ]i,j]
6845 * |....j i.k.| or |.k..j i...|
6847 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6848 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6853 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6856 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6860 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6861 struct kvm_async_pf *work)
6863 struct x86_exception fault;
6865 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6866 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6868 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6869 (vcpu->arch.apf.send_user_only &&
6870 kvm_x86_ops->get_cpl(vcpu) == 0))
6871 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6872 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6873 fault.vector = PF_VECTOR;
6874 fault.error_code_valid = true;
6875 fault.error_code = 0;
6876 fault.nested_page_fault = false;
6877 fault.address = work->arch.token;
6878 kvm_inject_page_fault(vcpu, &fault);
6882 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6883 struct kvm_async_pf *work)
6885 struct x86_exception fault;
6887 trace_kvm_async_pf_ready(work->arch.token, work->gva);
6888 if (is_error_page(work->page))
6889 work->arch.token = ~0; /* broadcast wakeup */
6891 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6893 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6894 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6895 fault.vector = PF_VECTOR;
6896 fault.error_code_valid = true;
6897 fault.error_code = 0;
6898 fault.nested_page_fault = false;
6899 fault.address = work->arch.token;
6900 kvm_inject_page_fault(vcpu, &fault);
6902 vcpu->arch.apf.halted = false;
6905 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6907 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6910 return !kvm_event_needs_reinjection(vcpu) &&
6911 kvm_x86_ops->interrupt_allowed(vcpu);
6914 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6915 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6916 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6917 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6918 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6919 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6920 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6921 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6922 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6923 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6924 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6925 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);