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"
31 #include <linux/clocksource.h>
32 #include <linux/interrupt.h>
33 #include <linux/kvm.h>
35 #include <linux/vmalloc.h>
36 #include <linux/module.h>
37 #include <linux/mman.h>
38 #include <linux/highmem.h>
39 #include <linux/iommu.h>
40 #include <linux/intel-iommu.h>
41 #include <linux/cpufreq.h>
42 #include <linux/user-return-notifier.h>
43 #include <linux/srcu.h>
44 #include <linux/slab.h>
45 #include <linux/perf_event.h>
46 #include <linux/uaccess.h>
47 #include <linux/hash.h>
48 #include <linux/pci.h>
49 #include <linux/timekeeper_internal.h>
50 #include <linux/pvclock_gtod.h>
51 #include <trace/events/kvm.h>
53 #define CREATE_TRACE_POINTS
56 #include <asm/debugreg.h>
62 #include <asm/fpu-internal.h> /* Ugh! */
64 #include <asm/pvclock.h>
65 #include <asm/div64.h>
67 #define MAX_IO_MSRS 256
68 #define KVM_MAX_MCE_BANKS 32
69 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
71 #define emul_to_vcpu(ctxt) \
72 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
75 * - enable syscall per default because its emulated by KVM
76 * - enable LME and LMA per default on 64 bit KVM
80 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
82 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
85 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
86 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
88 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
89 static void process_nmi(struct kvm_vcpu *vcpu);
91 struct kvm_x86_ops *kvm_x86_ops;
92 EXPORT_SYMBOL_GPL(kvm_x86_ops);
94 static bool ignore_msrs = 0;
95 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
97 bool kvm_has_tsc_control;
98 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
99 u32 kvm_max_guest_tsc_khz;
100 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
102 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
103 static u32 tsc_tolerance_ppm = 250;
104 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
106 #define KVM_NR_SHARED_MSRS 16
108 struct kvm_shared_msrs_global {
110 u32 msrs[KVM_NR_SHARED_MSRS];
113 struct kvm_shared_msrs {
114 struct user_return_notifier urn;
116 struct kvm_shared_msr_values {
119 } values[KVM_NR_SHARED_MSRS];
122 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
123 static struct kvm_shared_msrs __percpu *shared_msrs;
125 struct kvm_stats_debugfs_item debugfs_entries[] = {
126 { "pf_fixed", VCPU_STAT(pf_fixed) },
127 { "pf_guest", VCPU_STAT(pf_guest) },
128 { "tlb_flush", VCPU_STAT(tlb_flush) },
129 { "invlpg", VCPU_STAT(invlpg) },
130 { "exits", VCPU_STAT(exits) },
131 { "io_exits", VCPU_STAT(io_exits) },
132 { "mmio_exits", VCPU_STAT(mmio_exits) },
133 { "signal_exits", VCPU_STAT(signal_exits) },
134 { "irq_window", VCPU_STAT(irq_window_exits) },
135 { "nmi_window", VCPU_STAT(nmi_window_exits) },
136 { "halt_exits", VCPU_STAT(halt_exits) },
137 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
138 { "hypercalls", VCPU_STAT(hypercalls) },
139 { "request_irq", VCPU_STAT(request_irq_exits) },
140 { "irq_exits", VCPU_STAT(irq_exits) },
141 { "host_state_reload", VCPU_STAT(host_state_reload) },
142 { "efer_reload", VCPU_STAT(efer_reload) },
143 { "fpu_reload", VCPU_STAT(fpu_reload) },
144 { "insn_emulation", VCPU_STAT(insn_emulation) },
145 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
146 { "irq_injections", VCPU_STAT(irq_injections) },
147 { "nmi_injections", VCPU_STAT(nmi_injections) },
148 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
149 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
150 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
151 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
152 { "mmu_flooded", VM_STAT(mmu_flooded) },
153 { "mmu_recycled", VM_STAT(mmu_recycled) },
154 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
155 { "mmu_unsync", VM_STAT(mmu_unsync) },
156 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
157 { "largepages", VM_STAT(lpages) },
161 u64 __read_mostly host_xcr0;
163 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
165 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
168 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
169 vcpu->arch.apf.gfns[i] = ~0;
172 static void kvm_on_user_return(struct user_return_notifier *urn)
175 struct kvm_shared_msrs *locals
176 = container_of(urn, struct kvm_shared_msrs, urn);
177 struct kvm_shared_msr_values *values;
179 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
180 values = &locals->values[slot];
181 if (values->host != values->curr) {
182 wrmsrl(shared_msrs_global.msrs[slot], values->host);
183 values->curr = values->host;
186 locals->registered = false;
187 user_return_notifier_unregister(urn);
190 static void shared_msr_update(unsigned slot, u32 msr)
193 unsigned int cpu = smp_processor_id();
194 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
196 /* only read, and nobody should modify it at this time,
197 * so don't need lock */
198 if (slot >= shared_msrs_global.nr) {
199 printk(KERN_ERR "kvm: invalid MSR slot!");
202 rdmsrl_safe(msr, &value);
203 smsr->values[slot].host = value;
204 smsr->values[slot].curr = value;
207 void kvm_define_shared_msr(unsigned slot, u32 msr)
209 if (slot >= shared_msrs_global.nr)
210 shared_msrs_global.nr = slot + 1;
211 shared_msrs_global.msrs[slot] = msr;
212 /* we need ensured the shared_msr_global have been updated */
215 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
217 static void kvm_shared_msr_cpu_online(void)
221 for (i = 0; i < shared_msrs_global.nr; ++i)
222 shared_msr_update(i, shared_msrs_global.msrs[i]);
225 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
227 unsigned int cpu = smp_processor_id();
228 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
230 if (((value ^ smsr->values[slot].curr) & mask) == 0)
232 smsr->values[slot].curr = value;
233 wrmsrl(shared_msrs_global.msrs[slot], value);
234 if (!smsr->registered) {
235 smsr->urn.on_user_return = kvm_on_user_return;
236 user_return_notifier_register(&smsr->urn);
237 smsr->registered = true;
240 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
242 static void drop_user_return_notifiers(void *ignore)
244 unsigned int cpu = smp_processor_id();
245 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
247 if (smsr->registered)
248 kvm_on_user_return(&smsr->urn);
251 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
253 return vcpu->arch.apic_base;
255 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
257 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
259 /* TODO: reserve bits check */
260 kvm_lapic_set_base(vcpu, data);
262 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
264 asmlinkage void kvm_spurious_fault(void)
266 /* Fault while not rebooting. We want the trace. */
269 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
271 #define EXCPT_BENIGN 0
272 #define EXCPT_CONTRIBUTORY 1
275 static int exception_class(int vector)
285 return EXCPT_CONTRIBUTORY;
292 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
293 unsigned nr, bool has_error, u32 error_code,
299 kvm_make_request(KVM_REQ_EVENT, vcpu);
301 if (!vcpu->arch.exception.pending) {
303 vcpu->arch.exception.pending = true;
304 vcpu->arch.exception.has_error_code = has_error;
305 vcpu->arch.exception.nr = nr;
306 vcpu->arch.exception.error_code = error_code;
307 vcpu->arch.exception.reinject = reinject;
311 /* to check exception */
312 prev_nr = vcpu->arch.exception.nr;
313 if (prev_nr == DF_VECTOR) {
314 /* triple fault -> shutdown */
315 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
318 class1 = exception_class(prev_nr);
319 class2 = exception_class(nr);
320 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
321 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
322 /* generate double fault per SDM Table 5-5 */
323 vcpu->arch.exception.pending = true;
324 vcpu->arch.exception.has_error_code = true;
325 vcpu->arch.exception.nr = DF_VECTOR;
326 vcpu->arch.exception.error_code = 0;
328 /* replace previous exception with a new one in a hope
329 that instruction re-execution will regenerate lost
334 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
336 kvm_multiple_exception(vcpu, nr, false, 0, false);
338 EXPORT_SYMBOL_GPL(kvm_queue_exception);
340 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
342 kvm_multiple_exception(vcpu, nr, false, 0, true);
344 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
346 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
349 kvm_inject_gp(vcpu, 0);
351 kvm_x86_ops->skip_emulated_instruction(vcpu);
353 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
355 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
357 ++vcpu->stat.pf_guest;
358 vcpu->arch.cr2 = fault->address;
359 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
361 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
363 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
365 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
366 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
368 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
371 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
373 atomic_inc(&vcpu->arch.nmi_queued);
374 kvm_make_request(KVM_REQ_NMI, vcpu);
376 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
378 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
380 kvm_multiple_exception(vcpu, nr, true, error_code, false);
382 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
384 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
386 kvm_multiple_exception(vcpu, nr, true, error_code, true);
388 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
391 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
392 * a #GP and return false.
394 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
396 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
398 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
401 EXPORT_SYMBOL_GPL(kvm_require_cpl);
404 * This function will be used to read from the physical memory of the currently
405 * running guest. The difference to kvm_read_guest_page is that this function
406 * can read from guest physical or from the guest's guest physical memory.
408 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
409 gfn_t ngfn, void *data, int offset, int len,
415 ngpa = gfn_to_gpa(ngfn);
416 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
417 if (real_gfn == UNMAPPED_GVA)
420 real_gfn = gpa_to_gfn(real_gfn);
422 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
424 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
426 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
427 void *data, int offset, int len, u32 access)
429 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
430 data, offset, len, access);
434 * Load the pae pdptrs. Return true is they are all valid.
436 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
438 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
439 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
442 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
444 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
445 offset * sizeof(u64), sizeof(pdpte),
446 PFERR_USER_MASK|PFERR_WRITE_MASK);
451 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
452 if (is_present_gpte(pdpte[i]) &&
453 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
460 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
461 __set_bit(VCPU_EXREG_PDPTR,
462 (unsigned long *)&vcpu->arch.regs_avail);
463 __set_bit(VCPU_EXREG_PDPTR,
464 (unsigned long *)&vcpu->arch.regs_dirty);
469 EXPORT_SYMBOL_GPL(load_pdptrs);
471 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
473 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
479 if (is_long_mode(vcpu) || !is_pae(vcpu))
482 if (!test_bit(VCPU_EXREG_PDPTR,
483 (unsigned long *)&vcpu->arch.regs_avail))
486 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
487 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
488 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
489 PFERR_USER_MASK | PFERR_WRITE_MASK);
492 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
498 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
500 unsigned long old_cr0 = kvm_read_cr0(vcpu);
501 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
502 X86_CR0_CD | X86_CR0_NW;
507 if (cr0 & 0xffffffff00000000UL)
511 cr0 &= ~CR0_RESERVED_BITS;
513 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
516 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
519 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
521 if ((vcpu->arch.efer & EFER_LME)) {
526 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
531 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
536 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
539 kvm_x86_ops->set_cr0(vcpu, cr0);
541 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
542 kvm_clear_async_pf_completion_queue(vcpu);
543 kvm_async_pf_hash_reset(vcpu);
546 if ((cr0 ^ old_cr0) & update_bits)
547 kvm_mmu_reset_context(vcpu);
550 EXPORT_SYMBOL_GPL(kvm_set_cr0);
552 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
554 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
556 EXPORT_SYMBOL_GPL(kvm_lmsw);
558 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
560 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
561 !vcpu->guest_xcr0_loaded) {
562 /* kvm_set_xcr() also depends on this */
563 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
564 vcpu->guest_xcr0_loaded = 1;
568 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
570 if (vcpu->guest_xcr0_loaded) {
571 if (vcpu->arch.xcr0 != host_xcr0)
572 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
573 vcpu->guest_xcr0_loaded = 0;
577 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
582 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
583 if (index != XCR_XFEATURE_ENABLED_MASK)
586 if (!(xcr0 & XSTATE_FP))
588 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
592 * Do not allow the guest to set bits that we do not support
593 * saving. However, xcr0 bit 0 is always set, even if the
594 * emulated CPU does not support XSAVE (see fx_init).
596 valid_bits = vcpu->arch.guest_supported_xcr0 | XSTATE_FP;
597 if (xcr0 & ~valid_bits)
600 kvm_put_guest_xcr0(vcpu);
601 vcpu->arch.xcr0 = xcr0;
605 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
607 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
608 __kvm_set_xcr(vcpu, index, xcr)) {
609 kvm_inject_gp(vcpu, 0);
614 EXPORT_SYMBOL_GPL(kvm_set_xcr);
616 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
618 unsigned long old_cr4 = kvm_read_cr4(vcpu);
619 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
620 X86_CR4_PAE | X86_CR4_SMEP;
621 if (cr4 & CR4_RESERVED_BITS)
624 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
627 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
630 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
633 if (is_long_mode(vcpu)) {
634 if (!(cr4 & X86_CR4_PAE))
636 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
637 && ((cr4 ^ old_cr4) & pdptr_bits)
638 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
642 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
643 if (!guest_cpuid_has_pcid(vcpu))
646 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
647 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
651 if (kvm_x86_ops->set_cr4(vcpu, cr4))
654 if (((cr4 ^ old_cr4) & pdptr_bits) ||
655 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
656 kvm_mmu_reset_context(vcpu);
658 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
659 kvm_update_cpuid(vcpu);
663 EXPORT_SYMBOL_GPL(kvm_set_cr4);
665 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
667 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
668 kvm_mmu_sync_roots(vcpu);
669 kvm_mmu_flush_tlb(vcpu);
673 if (is_long_mode(vcpu)) {
674 if (kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)) {
675 if (cr3 & CR3_PCID_ENABLED_RESERVED_BITS)
678 if (cr3 & CR3_L_MODE_RESERVED_BITS)
682 if (cr3 & CR3_PAE_RESERVED_BITS)
684 if (is_paging(vcpu) &&
685 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
689 * We don't check reserved bits in nonpae mode, because
690 * this isn't enforced, and VMware depends on this.
694 vcpu->arch.cr3 = cr3;
695 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
696 kvm_mmu_new_cr3(vcpu);
699 EXPORT_SYMBOL_GPL(kvm_set_cr3);
701 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
703 if (cr8 & CR8_RESERVED_BITS)
705 if (irqchip_in_kernel(vcpu->kvm))
706 kvm_lapic_set_tpr(vcpu, cr8);
708 vcpu->arch.cr8 = cr8;
711 EXPORT_SYMBOL_GPL(kvm_set_cr8);
713 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
715 if (irqchip_in_kernel(vcpu->kvm))
716 return kvm_lapic_get_cr8(vcpu);
718 return vcpu->arch.cr8;
720 EXPORT_SYMBOL_GPL(kvm_get_cr8);
722 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
726 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
727 dr7 = vcpu->arch.guest_debug_dr7;
729 dr7 = vcpu->arch.dr7;
730 kvm_x86_ops->set_dr7(vcpu, dr7);
731 vcpu->arch.switch_db_regs = (dr7 & DR7_BP_EN_MASK);
734 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
738 vcpu->arch.db[dr] = val;
739 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
740 vcpu->arch.eff_db[dr] = val;
743 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
747 if (val & 0xffffffff00000000ULL)
749 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
752 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
756 if (val & 0xffffffff00000000ULL)
758 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
759 kvm_update_dr7(vcpu);
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);
815 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
817 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
821 err = kvm_pmu_read_pmc(vcpu, ecx, &data);
824 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
825 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
828 EXPORT_SYMBOL_GPL(kvm_rdpmc);
831 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
832 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
834 * This list is modified at module load time to reflect the
835 * capabilities of the host cpu. This capabilities test skips MSRs that are
836 * kvm-specific. Those are put in the beginning of the list.
839 #define KVM_SAVE_MSRS_BEGIN 10
840 static u32 msrs_to_save[] = {
841 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
842 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
843 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
844 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
846 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
849 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
851 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
852 MSR_IA32_FEATURE_CONTROL
855 static unsigned num_msrs_to_save;
857 static const u32 emulated_msrs[] = {
859 MSR_IA32_TSCDEADLINE,
860 MSR_IA32_MISC_ENABLE,
865 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
867 if (efer & efer_reserved_bits)
870 if (efer & EFER_FFXSR) {
871 struct kvm_cpuid_entry2 *feat;
873 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
874 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
878 if (efer & EFER_SVME) {
879 struct kvm_cpuid_entry2 *feat;
881 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
882 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
888 EXPORT_SYMBOL_GPL(kvm_valid_efer);
890 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
892 u64 old_efer = vcpu->arch.efer;
894 if (!kvm_valid_efer(vcpu, efer))
898 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
902 efer |= vcpu->arch.efer & EFER_LMA;
904 kvm_x86_ops->set_efer(vcpu, efer);
906 /* Update reserved bits */
907 if ((efer ^ old_efer) & EFER_NX)
908 kvm_mmu_reset_context(vcpu);
913 void kvm_enable_efer_bits(u64 mask)
915 efer_reserved_bits &= ~mask;
917 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
921 * Writes msr value into into the appropriate "register".
922 * Returns 0 on success, non-0 otherwise.
923 * Assumes vcpu_load() was already called.
925 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
927 return kvm_x86_ops->set_msr(vcpu, msr);
931 * Adapt set_msr() to msr_io()'s calling convention
933 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
939 msr.host_initiated = true;
940 return kvm_set_msr(vcpu, &msr);
944 struct pvclock_gtod_data {
947 struct { /* extract of a clocksource struct */
955 /* open coded 'struct timespec' */
956 u64 monotonic_time_snsec;
957 time_t monotonic_time_sec;
960 static struct pvclock_gtod_data pvclock_gtod_data;
962 static void update_pvclock_gtod(struct timekeeper *tk)
964 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
966 write_seqcount_begin(&vdata->seq);
968 /* copy pvclock gtod data */
969 vdata->clock.vclock_mode = tk->clock->archdata.vclock_mode;
970 vdata->clock.cycle_last = tk->clock->cycle_last;
971 vdata->clock.mask = tk->clock->mask;
972 vdata->clock.mult = tk->mult;
973 vdata->clock.shift = tk->shift;
975 vdata->monotonic_time_sec = tk->xtime_sec
976 + tk->wall_to_monotonic.tv_sec;
977 vdata->monotonic_time_snsec = tk->xtime_nsec
978 + (tk->wall_to_monotonic.tv_nsec
980 while (vdata->monotonic_time_snsec >=
981 (((u64)NSEC_PER_SEC) << tk->shift)) {
982 vdata->monotonic_time_snsec -=
983 ((u64)NSEC_PER_SEC) << tk->shift;
984 vdata->monotonic_time_sec++;
987 write_seqcount_end(&vdata->seq);
992 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
996 struct pvclock_wall_clock wc;
997 struct timespec boot;
1002 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1007 ++version; /* first time write, random junk */
1011 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1014 * The guest calculates current wall clock time by adding
1015 * system time (updated by kvm_guest_time_update below) to the
1016 * wall clock specified here. guest system time equals host
1017 * system time for us, thus we must fill in host boot time here.
1021 if (kvm->arch.kvmclock_offset) {
1022 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1023 boot = timespec_sub(boot, ts);
1025 wc.sec = boot.tv_sec;
1026 wc.nsec = boot.tv_nsec;
1027 wc.version = version;
1029 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1032 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1035 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1037 uint32_t quotient, remainder;
1039 /* Don't try to replace with do_div(), this one calculates
1040 * "(dividend << 32) / divisor" */
1042 : "=a" (quotient), "=d" (remainder)
1043 : "0" (0), "1" (dividend), "r" (divisor) );
1047 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1048 s8 *pshift, u32 *pmultiplier)
1055 tps64 = base_khz * 1000LL;
1056 scaled64 = scaled_khz * 1000LL;
1057 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1062 tps32 = (uint32_t)tps64;
1063 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1064 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1072 *pmultiplier = div_frac(scaled64, tps32);
1074 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1075 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1078 static inline u64 get_kernel_ns(void)
1082 WARN_ON(preemptible());
1084 monotonic_to_bootbased(&ts);
1085 return timespec_to_ns(&ts);
1088 #ifdef CONFIG_X86_64
1089 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1092 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1093 unsigned long max_tsc_khz;
1095 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1097 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1098 vcpu->arch.virtual_tsc_shift);
1101 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1103 u64 v = (u64)khz * (1000000 + ppm);
1108 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1110 u32 thresh_lo, thresh_hi;
1111 int use_scaling = 0;
1113 /* tsc_khz can be zero if TSC calibration fails */
1114 if (this_tsc_khz == 0)
1117 /* Compute a scale to convert nanoseconds in TSC cycles */
1118 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1119 &vcpu->arch.virtual_tsc_shift,
1120 &vcpu->arch.virtual_tsc_mult);
1121 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1124 * Compute the variation in TSC rate which is acceptable
1125 * within the range of tolerance and decide if the
1126 * rate being applied is within that bounds of the hardware
1127 * rate. If so, no scaling or compensation need be done.
1129 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1130 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1131 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1132 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1135 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1138 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1140 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1141 vcpu->arch.virtual_tsc_mult,
1142 vcpu->arch.virtual_tsc_shift);
1143 tsc += vcpu->arch.this_tsc_write;
1147 void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1149 #ifdef CONFIG_X86_64
1151 bool do_request = false;
1152 struct kvm_arch *ka = &vcpu->kvm->arch;
1153 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1155 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1156 atomic_read(&vcpu->kvm->online_vcpus));
1158 if (vcpus_matched && gtod->clock.vclock_mode == VCLOCK_TSC)
1159 if (!ka->use_master_clock)
1162 if (!vcpus_matched && ka->use_master_clock)
1166 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1168 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1169 atomic_read(&vcpu->kvm->online_vcpus),
1170 ka->use_master_clock, gtod->clock.vclock_mode);
1174 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1176 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1177 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1180 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1182 struct kvm *kvm = vcpu->kvm;
1183 u64 offset, ns, elapsed;
1184 unsigned long flags;
1187 u64 data = msr->data;
1189 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1190 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1191 ns = get_kernel_ns();
1192 elapsed = ns - kvm->arch.last_tsc_nsec;
1194 if (vcpu->arch.virtual_tsc_khz) {
1197 /* n.b - signed multiplication and division required */
1198 usdiff = data - kvm->arch.last_tsc_write;
1199 #ifdef CONFIG_X86_64
1200 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1202 /* do_div() only does unsigned */
1203 asm("1: idivl %[divisor]\n"
1204 "2: xor %%edx, %%edx\n"
1205 " movl $0, %[faulted]\n"
1207 ".section .fixup,\"ax\"\n"
1208 "4: movl $1, %[faulted]\n"
1212 _ASM_EXTABLE(1b, 4b)
1214 : "=A"(usdiff), [faulted] "=r" (faulted)
1215 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1218 do_div(elapsed, 1000);
1223 /* idivl overflow => difference is larger than USEC_PER_SEC */
1225 usdiff = USEC_PER_SEC;
1227 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1230 * Special case: TSC write with a small delta (1 second) of virtual
1231 * cycle time against real time is interpreted as an attempt to
1232 * synchronize the CPU.
1234 * For a reliable TSC, we can match TSC offsets, and for an unstable
1235 * TSC, we add elapsed time in this computation. We could let the
1236 * compensation code attempt to catch up if we fall behind, but
1237 * it's better to try to match offsets from the beginning.
1239 if (usdiff < USEC_PER_SEC &&
1240 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1241 if (!check_tsc_unstable()) {
1242 offset = kvm->arch.cur_tsc_offset;
1243 pr_debug("kvm: matched tsc offset for %llu\n", data);
1245 u64 delta = nsec_to_cycles(vcpu, elapsed);
1247 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1248 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1253 * We split periods of matched TSC writes into generations.
1254 * For each generation, we track the original measured
1255 * nanosecond time, offset, and write, so if TSCs are in
1256 * sync, we can match exact offset, and if not, we can match
1257 * exact software computation in compute_guest_tsc()
1259 * These values are tracked in kvm->arch.cur_xxx variables.
1261 kvm->arch.cur_tsc_generation++;
1262 kvm->arch.cur_tsc_nsec = ns;
1263 kvm->arch.cur_tsc_write = data;
1264 kvm->arch.cur_tsc_offset = offset;
1266 pr_debug("kvm: new tsc generation %u, clock %llu\n",
1267 kvm->arch.cur_tsc_generation, data);
1271 * We also track th most recent recorded KHZ, write and time to
1272 * allow the matching interval to be extended at each write.
1274 kvm->arch.last_tsc_nsec = ns;
1275 kvm->arch.last_tsc_write = data;
1276 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1278 /* Reset of TSC must disable overshoot protection below */
1279 vcpu->arch.hv_clock.tsc_timestamp = 0;
1280 vcpu->arch.last_guest_tsc = data;
1282 /* Keep track of which generation this VCPU has synchronized to */
1283 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1284 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1285 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1287 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1288 update_ia32_tsc_adjust_msr(vcpu, offset);
1289 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1290 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1292 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1294 kvm->arch.nr_vcpus_matched_tsc++;
1296 kvm->arch.nr_vcpus_matched_tsc = 0;
1298 kvm_track_tsc_matching(vcpu);
1299 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1302 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1304 #ifdef CONFIG_X86_64
1306 static cycle_t read_tsc(void)
1312 * Empirically, a fence (of type that depends on the CPU)
1313 * before rdtsc is enough to ensure that rdtsc is ordered
1314 * with respect to loads. The various CPU manuals are unclear
1315 * as to whether rdtsc can be reordered with later loads,
1316 * but no one has ever seen it happen.
1319 ret = (cycle_t)vget_cycles();
1321 last = pvclock_gtod_data.clock.cycle_last;
1323 if (likely(ret >= last))
1327 * GCC likes to generate cmov here, but this branch is extremely
1328 * predictable (it's just a funciton of time and the likely is
1329 * very likely) and there's a data dependence, so force GCC
1330 * to generate a branch instead. I don't barrier() because
1331 * we don't actually need a barrier, and if this function
1332 * ever gets inlined it will generate worse code.
1338 static inline u64 vgettsc(cycle_t *cycle_now)
1341 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1343 *cycle_now = read_tsc();
1345 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1346 return v * gtod->clock.mult;
1349 static int do_monotonic(struct timespec *ts, cycle_t *cycle_now)
1354 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1358 seq = read_seqcount_begin(>od->seq);
1359 mode = gtod->clock.vclock_mode;
1360 ts->tv_sec = gtod->monotonic_time_sec;
1361 ns = gtod->monotonic_time_snsec;
1362 ns += vgettsc(cycle_now);
1363 ns >>= gtod->clock.shift;
1364 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1365 timespec_add_ns(ts, ns);
1370 /* returns true if host is using tsc clocksource */
1371 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1375 /* checked again under seqlock below */
1376 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1379 if (do_monotonic(&ts, cycle_now) != VCLOCK_TSC)
1382 monotonic_to_bootbased(&ts);
1383 *kernel_ns = timespec_to_ns(&ts);
1391 * Assuming a stable TSC across physical CPUS, and a stable TSC
1392 * across virtual CPUs, the following condition is possible.
1393 * Each numbered line represents an event visible to both
1394 * CPUs at the next numbered event.
1396 * "timespecX" represents host monotonic time. "tscX" represents
1399 * VCPU0 on CPU0 | VCPU1 on CPU1
1401 * 1. read timespec0,tsc0
1402 * 2. | timespec1 = timespec0 + N
1404 * 3. transition to guest | transition to guest
1405 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1406 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1407 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1409 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1412 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1414 * - 0 < N - M => M < N
1416 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1417 * always the case (the difference between two distinct xtime instances
1418 * might be smaller then the difference between corresponding TSC reads,
1419 * when updating guest vcpus pvclock areas).
1421 * To avoid that problem, do not allow visibility of distinct
1422 * system_timestamp/tsc_timestamp values simultaneously: use a master
1423 * copy of host monotonic time values. Update that master copy
1426 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1430 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1432 #ifdef CONFIG_X86_64
1433 struct kvm_arch *ka = &kvm->arch;
1435 bool host_tsc_clocksource, vcpus_matched;
1437 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1438 atomic_read(&kvm->online_vcpus));
1441 * If the host uses TSC clock, then passthrough TSC as stable
1444 host_tsc_clocksource = kvm_get_time_and_clockread(
1445 &ka->master_kernel_ns,
1446 &ka->master_cycle_now);
1448 ka->use_master_clock = host_tsc_clocksource & vcpus_matched;
1450 if (ka->use_master_clock)
1451 atomic_set(&kvm_guest_has_master_clock, 1);
1453 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1454 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1459 static void kvm_gen_update_masterclock(struct kvm *kvm)
1461 #ifdef CONFIG_X86_64
1463 struct kvm_vcpu *vcpu;
1464 struct kvm_arch *ka = &kvm->arch;
1466 spin_lock(&ka->pvclock_gtod_sync_lock);
1467 kvm_make_mclock_inprogress_request(kvm);
1468 /* no guest entries from this point */
1469 pvclock_update_vm_gtod_copy(kvm);
1471 kvm_for_each_vcpu(i, vcpu, kvm)
1472 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
1474 /* guest entries allowed */
1475 kvm_for_each_vcpu(i, vcpu, kvm)
1476 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1478 spin_unlock(&ka->pvclock_gtod_sync_lock);
1482 static int kvm_guest_time_update(struct kvm_vcpu *v)
1484 unsigned long flags, this_tsc_khz;
1485 struct kvm_vcpu_arch *vcpu = &v->arch;
1486 struct kvm_arch *ka = &v->kvm->arch;
1487 s64 kernel_ns, max_kernel_ns;
1488 u64 tsc_timestamp, host_tsc;
1489 struct pvclock_vcpu_time_info guest_hv_clock;
1491 bool use_master_clock;
1497 * If the host uses TSC clock, then passthrough TSC as stable
1500 spin_lock(&ka->pvclock_gtod_sync_lock);
1501 use_master_clock = ka->use_master_clock;
1502 if (use_master_clock) {
1503 host_tsc = ka->master_cycle_now;
1504 kernel_ns = ka->master_kernel_ns;
1506 spin_unlock(&ka->pvclock_gtod_sync_lock);
1508 /* Keep irq disabled to prevent changes to the clock */
1509 local_irq_save(flags);
1510 this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
1511 if (unlikely(this_tsc_khz == 0)) {
1512 local_irq_restore(flags);
1513 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1516 if (!use_master_clock) {
1517 host_tsc = native_read_tsc();
1518 kernel_ns = get_kernel_ns();
1521 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1524 * We may have to catch up the TSC to match elapsed wall clock
1525 * time for two reasons, even if kvmclock is used.
1526 * 1) CPU could have been running below the maximum TSC rate
1527 * 2) Broken TSC compensation resets the base at each VCPU
1528 * entry to avoid unknown leaps of TSC even when running
1529 * again on the same CPU. This may cause apparent elapsed
1530 * time to disappear, and the guest to stand still or run
1533 if (vcpu->tsc_catchup) {
1534 u64 tsc = compute_guest_tsc(v, kernel_ns);
1535 if (tsc > tsc_timestamp) {
1536 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1537 tsc_timestamp = tsc;
1541 local_irq_restore(flags);
1543 if (!vcpu->pv_time_enabled)
1547 * Time as measured by the TSC may go backwards when resetting the base
1548 * tsc_timestamp. The reason for this is that the TSC resolution is
1549 * higher than the resolution of the other clock scales. Thus, many
1550 * possible measurments of the TSC correspond to one measurement of any
1551 * other clock, and so a spread of values is possible. This is not a
1552 * problem for the computation of the nanosecond clock; with TSC rates
1553 * around 1GHZ, there can only be a few cycles which correspond to one
1554 * nanosecond value, and any path through this code will inevitably
1555 * take longer than that. However, with the kernel_ns value itself,
1556 * the precision may be much lower, down to HZ granularity. If the
1557 * first sampling of TSC against kernel_ns ends in the low part of the
1558 * range, and the second in the high end of the range, we can get:
1560 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1562 * As the sampling errors potentially range in the thousands of cycles,
1563 * it is possible such a time value has already been observed by the
1564 * guest. To protect against this, we must compute the system time as
1565 * observed by the guest and ensure the new system time is greater.
1568 if (vcpu->hv_clock.tsc_timestamp) {
1569 max_kernel_ns = vcpu->last_guest_tsc -
1570 vcpu->hv_clock.tsc_timestamp;
1571 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1572 vcpu->hv_clock.tsc_to_system_mul,
1573 vcpu->hv_clock.tsc_shift);
1574 max_kernel_ns += vcpu->last_kernel_ns;
1577 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1578 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1579 &vcpu->hv_clock.tsc_shift,
1580 &vcpu->hv_clock.tsc_to_system_mul);
1581 vcpu->hw_tsc_khz = this_tsc_khz;
1584 /* with a master <monotonic time, tsc value> tuple,
1585 * pvclock clock reads always increase at the (scaled) rate
1586 * of guest TSC - no need to deal with sampling errors.
1588 if (!use_master_clock) {
1589 if (max_kernel_ns > kernel_ns)
1590 kernel_ns = max_kernel_ns;
1592 /* With all the info we got, fill in the values */
1593 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1594 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1595 vcpu->last_kernel_ns = kernel_ns;
1596 vcpu->last_guest_tsc = tsc_timestamp;
1599 * The interface expects us to write an even number signaling that the
1600 * update is finished. Since the guest won't see the intermediate
1601 * state, we just increase by 2 at the end.
1603 vcpu->hv_clock.version += 2;
1605 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1606 &guest_hv_clock, sizeof(guest_hv_clock))))
1609 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1610 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1612 if (vcpu->pvclock_set_guest_stopped_request) {
1613 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1614 vcpu->pvclock_set_guest_stopped_request = false;
1617 /* If the host uses TSC clocksource, then it is stable */
1618 if (use_master_clock)
1619 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1621 vcpu->hv_clock.flags = pvclock_flags;
1623 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1625 sizeof(vcpu->hv_clock));
1630 * kvmclock updates which are isolated to a given vcpu, such as
1631 * vcpu->cpu migration, should not allow system_timestamp from
1632 * the rest of the vcpus to remain static. Otherwise ntp frequency
1633 * correction applies to one vcpu's system_timestamp but not
1636 * So in those cases, request a kvmclock update for all vcpus.
1637 * The worst case for a remote vcpu to update its kvmclock
1638 * is then bounded by maximum nohz sleep latency.
1641 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1644 struct kvm *kvm = v->kvm;
1645 struct kvm_vcpu *vcpu;
1647 kvm_for_each_vcpu(i, vcpu, kvm) {
1648 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
1649 kvm_vcpu_kick(vcpu);
1653 static bool msr_mtrr_valid(unsigned msr)
1656 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1657 case MSR_MTRRfix64K_00000:
1658 case MSR_MTRRfix16K_80000:
1659 case MSR_MTRRfix16K_A0000:
1660 case MSR_MTRRfix4K_C0000:
1661 case MSR_MTRRfix4K_C8000:
1662 case MSR_MTRRfix4K_D0000:
1663 case MSR_MTRRfix4K_D8000:
1664 case MSR_MTRRfix4K_E0000:
1665 case MSR_MTRRfix4K_E8000:
1666 case MSR_MTRRfix4K_F0000:
1667 case MSR_MTRRfix4K_F8000:
1668 case MSR_MTRRdefType:
1669 case MSR_IA32_CR_PAT:
1677 static bool valid_pat_type(unsigned t)
1679 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1682 static bool valid_mtrr_type(unsigned t)
1684 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1687 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1691 if (!msr_mtrr_valid(msr))
1694 if (msr == MSR_IA32_CR_PAT) {
1695 for (i = 0; i < 8; i++)
1696 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1699 } else if (msr == MSR_MTRRdefType) {
1702 return valid_mtrr_type(data & 0xff);
1703 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1704 for (i = 0; i < 8 ; i++)
1705 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1710 /* variable MTRRs */
1711 return valid_mtrr_type(data & 0xff);
1714 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1716 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1718 if (!mtrr_valid(vcpu, msr, data))
1721 if (msr == MSR_MTRRdefType) {
1722 vcpu->arch.mtrr_state.def_type = data;
1723 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1724 } else if (msr == MSR_MTRRfix64K_00000)
1726 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1727 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1728 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1729 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1730 else if (msr == MSR_IA32_CR_PAT)
1731 vcpu->arch.pat = data;
1732 else { /* Variable MTRRs */
1733 int idx, is_mtrr_mask;
1736 idx = (msr - 0x200) / 2;
1737 is_mtrr_mask = msr - 0x200 - 2 * idx;
1740 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1743 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1747 kvm_mmu_reset_context(vcpu);
1751 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1753 u64 mcg_cap = vcpu->arch.mcg_cap;
1754 unsigned bank_num = mcg_cap & 0xff;
1757 case MSR_IA32_MCG_STATUS:
1758 vcpu->arch.mcg_status = data;
1760 case MSR_IA32_MCG_CTL:
1761 if (!(mcg_cap & MCG_CTL_P))
1763 if (data != 0 && data != ~(u64)0)
1765 vcpu->arch.mcg_ctl = data;
1768 if (msr >= MSR_IA32_MC0_CTL &&
1769 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1770 u32 offset = msr - MSR_IA32_MC0_CTL;
1771 /* only 0 or all 1s can be written to IA32_MCi_CTL
1772 * some Linux kernels though clear bit 10 in bank 4 to
1773 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1774 * this to avoid an uncatched #GP in the guest
1776 if ((offset & 0x3) == 0 &&
1777 data != 0 && (data | (1 << 10)) != ~(u64)0)
1779 vcpu->arch.mce_banks[offset] = data;
1787 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1789 struct kvm *kvm = vcpu->kvm;
1790 int lm = is_long_mode(vcpu);
1791 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1792 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1793 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1794 : kvm->arch.xen_hvm_config.blob_size_32;
1795 u32 page_num = data & ~PAGE_MASK;
1796 u64 page_addr = data & PAGE_MASK;
1801 if (page_num >= blob_size)
1804 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1809 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1818 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1820 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1823 static bool kvm_hv_msr_partition_wide(u32 msr)
1827 case HV_X64_MSR_GUEST_OS_ID:
1828 case HV_X64_MSR_HYPERCALL:
1836 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1838 struct kvm *kvm = vcpu->kvm;
1841 case HV_X64_MSR_GUEST_OS_ID:
1842 kvm->arch.hv_guest_os_id = data;
1843 /* setting guest os id to zero disables hypercall page */
1844 if (!kvm->arch.hv_guest_os_id)
1845 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1847 case HV_X64_MSR_HYPERCALL: {
1852 /* if guest os id is not set hypercall should remain disabled */
1853 if (!kvm->arch.hv_guest_os_id)
1855 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1856 kvm->arch.hv_hypercall = data;
1859 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1860 addr = gfn_to_hva(kvm, gfn);
1861 if (kvm_is_error_hva(addr))
1863 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1864 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1865 if (__copy_to_user((void __user *)addr, instructions, 4))
1867 kvm->arch.hv_hypercall = data;
1871 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1872 "data 0x%llx\n", msr, data);
1878 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1881 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1884 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1885 vcpu->arch.hv_vapic = data;
1888 addr = gfn_to_hva(vcpu->kvm, data >>
1889 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1890 if (kvm_is_error_hva(addr))
1892 if (__clear_user((void __user *)addr, PAGE_SIZE))
1894 vcpu->arch.hv_vapic = data;
1897 case HV_X64_MSR_EOI:
1898 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1899 case HV_X64_MSR_ICR:
1900 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1901 case HV_X64_MSR_TPR:
1902 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1904 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1905 "data 0x%llx\n", msr, data);
1912 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1914 gpa_t gpa = data & ~0x3f;
1916 /* Bits 2:5 are reserved, Should be zero */
1920 vcpu->arch.apf.msr_val = data;
1922 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1923 kvm_clear_async_pf_completion_queue(vcpu);
1924 kvm_async_pf_hash_reset(vcpu);
1928 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1932 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1933 kvm_async_pf_wakeup_all(vcpu);
1937 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1939 vcpu->arch.pv_time_enabled = false;
1942 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1946 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1949 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1950 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1951 vcpu->arch.st.accum_steal = delta;
1954 static void record_steal_time(struct kvm_vcpu *vcpu)
1956 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1959 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1960 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1963 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1964 vcpu->arch.st.steal.version += 2;
1965 vcpu->arch.st.accum_steal = 0;
1967 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1968 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1971 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1974 u32 msr = msr_info->index;
1975 u64 data = msr_info->data;
1978 case MSR_AMD64_NB_CFG:
1979 case MSR_IA32_UCODE_REV:
1980 case MSR_IA32_UCODE_WRITE:
1981 case MSR_VM_HSAVE_PA:
1982 case MSR_AMD64_PATCH_LOADER:
1983 case MSR_AMD64_BU_CFG2:
1987 return set_efer(vcpu, data);
1989 data &= ~(u64)0x40; /* ignore flush filter disable */
1990 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1991 data &= ~(u64)0x8; /* ignore TLB cache disable */
1993 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1998 case MSR_FAM10H_MMIO_CONF_BASE:
2000 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2005 case MSR_IA32_DEBUGCTLMSR:
2007 /* We support the non-activated case already */
2009 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2010 /* Values other than LBR and BTF are vendor-specific,
2011 thus reserved and should throw a #GP */
2014 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2017 case 0x200 ... 0x2ff:
2018 return set_msr_mtrr(vcpu, msr, data);
2019 case MSR_IA32_APICBASE:
2020 kvm_set_apic_base(vcpu, data);
2022 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2023 return kvm_x2apic_msr_write(vcpu, msr, data);
2024 case MSR_IA32_TSCDEADLINE:
2025 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2027 case MSR_IA32_TSC_ADJUST:
2028 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2029 if (!msr_info->host_initiated) {
2030 u64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2031 kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
2033 vcpu->arch.ia32_tsc_adjust_msr = data;
2036 case MSR_IA32_MISC_ENABLE:
2037 vcpu->arch.ia32_misc_enable_msr = data;
2039 case MSR_KVM_WALL_CLOCK_NEW:
2040 case MSR_KVM_WALL_CLOCK:
2041 vcpu->kvm->arch.wall_clock = data;
2042 kvm_write_wall_clock(vcpu->kvm, data);
2044 case MSR_KVM_SYSTEM_TIME_NEW:
2045 case MSR_KVM_SYSTEM_TIME: {
2047 kvmclock_reset(vcpu);
2049 vcpu->arch.time = data;
2050 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2052 /* we verify if the enable bit is set... */
2056 gpa_offset = data & ~(PAGE_MASK | 1);
2058 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2059 &vcpu->arch.pv_time, data & ~1ULL,
2060 sizeof(struct pvclock_vcpu_time_info)))
2061 vcpu->arch.pv_time_enabled = false;
2063 vcpu->arch.pv_time_enabled = true;
2067 case MSR_KVM_ASYNC_PF_EN:
2068 if (kvm_pv_enable_async_pf(vcpu, data))
2071 case MSR_KVM_STEAL_TIME:
2073 if (unlikely(!sched_info_on()))
2076 if (data & KVM_STEAL_RESERVED_MASK)
2079 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2080 data & KVM_STEAL_VALID_BITS,
2081 sizeof(struct kvm_steal_time)))
2084 vcpu->arch.st.msr_val = data;
2086 if (!(data & KVM_MSR_ENABLED))
2089 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2092 accumulate_steal_time(vcpu);
2095 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2098 case MSR_KVM_PV_EOI_EN:
2099 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2103 case MSR_IA32_MCG_CTL:
2104 case MSR_IA32_MCG_STATUS:
2105 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2106 return set_msr_mce(vcpu, msr, data);
2108 /* Performance counters are not protected by a CPUID bit,
2109 * so we should check all of them in the generic path for the sake of
2110 * cross vendor migration.
2111 * Writing a zero into the event select MSRs disables them,
2112 * which we perfectly emulate ;-). Any other value should be at least
2113 * reported, some guests depend on them.
2115 case MSR_K7_EVNTSEL0:
2116 case MSR_K7_EVNTSEL1:
2117 case MSR_K7_EVNTSEL2:
2118 case MSR_K7_EVNTSEL3:
2120 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2121 "0x%x data 0x%llx\n", msr, data);
2123 /* at least RHEL 4 unconditionally writes to the perfctr registers,
2124 * so we ignore writes to make it happy.
2126 case MSR_K7_PERFCTR0:
2127 case MSR_K7_PERFCTR1:
2128 case MSR_K7_PERFCTR2:
2129 case MSR_K7_PERFCTR3:
2130 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2131 "0x%x data 0x%llx\n", msr, data);
2133 case MSR_P6_PERFCTR0:
2134 case MSR_P6_PERFCTR1:
2136 case MSR_P6_EVNTSEL0:
2137 case MSR_P6_EVNTSEL1:
2138 if (kvm_pmu_msr(vcpu, msr))
2139 return kvm_pmu_set_msr(vcpu, msr_info);
2141 if (pr || data != 0)
2142 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2143 "0x%x data 0x%llx\n", msr, data);
2145 case MSR_K7_CLK_CTL:
2147 * Ignore all writes to this no longer documented MSR.
2148 * Writes are only relevant for old K7 processors,
2149 * all pre-dating SVM, but a recommended workaround from
2150 * AMD for these chips. It is possible to specify the
2151 * affected processor models on the command line, hence
2152 * the need to ignore the workaround.
2155 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2156 if (kvm_hv_msr_partition_wide(msr)) {
2158 mutex_lock(&vcpu->kvm->lock);
2159 r = set_msr_hyperv_pw(vcpu, msr, data);
2160 mutex_unlock(&vcpu->kvm->lock);
2163 return set_msr_hyperv(vcpu, msr, data);
2165 case MSR_IA32_BBL_CR_CTL3:
2166 /* Drop writes to this legacy MSR -- see rdmsr
2167 * counterpart for further detail.
2169 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2171 case MSR_AMD64_OSVW_ID_LENGTH:
2172 if (!guest_cpuid_has_osvw(vcpu))
2174 vcpu->arch.osvw.length = data;
2176 case MSR_AMD64_OSVW_STATUS:
2177 if (!guest_cpuid_has_osvw(vcpu))
2179 vcpu->arch.osvw.status = data;
2182 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2183 return xen_hvm_config(vcpu, data);
2184 if (kvm_pmu_msr(vcpu, msr))
2185 return kvm_pmu_set_msr(vcpu, msr_info);
2187 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2191 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2198 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2202 * Reads an msr value (of 'msr_index') into 'pdata'.
2203 * Returns 0 on success, non-0 otherwise.
2204 * Assumes vcpu_load() was already called.
2206 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2208 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
2211 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2213 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2215 if (!msr_mtrr_valid(msr))
2218 if (msr == MSR_MTRRdefType)
2219 *pdata = vcpu->arch.mtrr_state.def_type +
2220 (vcpu->arch.mtrr_state.enabled << 10);
2221 else if (msr == MSR_MTRRfix64K_00000)
2223 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2224 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2225 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2226 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2227 else if (msr == MSR_IA32_CR_PAT)
2228 *pdata = vcpu->arch.pat;
2229 else { /* Variable MTRRs */
2230 int idx, is_mtrr_mask;
2233 idx = (msr - 0x200) / 2;
2234 is_mtrr_mask = msr - 0x200 - 2 * idx;
2237 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2240 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2247 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2250 u64 mcg_cap = vcpu->arch.mcg_cap;
2251 unsigned bank_num = mcg_cap & 0xff;
2254 case MSR_IA32_P5_MC_ADDR:
2255 case MSR_IA32_P5_MC_TYPE:
2258 case MSR_IA32_MCG_CAP:
2259 data = vcpu->arch.mcg_cap;
2261 case MSR_IA32_MCG_CTL:
2262 if (!(mcg_cap & MCG_CTL_P))
2264 data = vcpu->arch.mcg_ctl;
2266 case MSR_IA32_MCG_STATUS:
2267 data = vcpu->arch.mcg_status;
2270 if (msr >= MSR_IA32_MC0_CTL &&
2271 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
2272 u32 offset = msr - MSR_IA32_MC0_CTL;
2273 data = vcpu->arch.mce_banks[offset];
2282 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2285 struct kvm *kvm = vcpu->kvm;
2288 case HV_X64_MSR_GUEST_OS_ID:
2289 data = kvm->arch.hv_guest_os_id;
2291 case HV_X64_MSR_HYPERCALL:
2292 data = kvm->arch.hv_hypercall;
2295 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2303 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2308 case HV_X64_MSR_VP_INDEX: {
2311 kvm_for_each_vcpu(r, v, vcpu->kvm)
2316 case HV_X64_MSR_EOI:
2317 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2318 case HV_X64_MSR_ICR:
2319 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2320 case HV_X64_MSR_TPR:
2321 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
2322 case HV_X64_MSR_APIC_ASSIST_PAGE:
2323 data = vcpu->arch.hv_vapic;
2326 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2333 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2338 case MSR_IA32_PLATFORM_ID:
2339 case MSR_IA32_EBL_CR_POWERON:
2340 case MSR_IA32_DEBUGCTLMSR:
2341 case MSR_IA32_LASTBRANCHFROMIP:
2342 case MSR_IA32_LASTBRANCHTOIP:
2343 case MSR_IA32_LASTINTFROMIP:
2344 case MSR_IA32_LASTINTTOIP:
2347 case MSR_VM_HSAVE_PA:
2348 case MSR_K7_EVNTSEL0:
2349 case MSR_K7_PERFCTR0:
2350 case MSR_K8_INT_PENDING_MSG:
2351 case MSR_AMD64_NB_CFG:
2352 case MSR_FAM10H_MMIO_CONF_BASE:
2353 case MSR_AMD64_BU_CFG2:
2356 case MSR_P6_PERFCTR0:
2357 case MSR_P6_PERFCTR1:
2358 case MSR_P6_EVNTSEL0:
2359 case MSR_P6_EVNTSEL1:
2360 if (kvm_pmu_msr(vcpu, msr))
2361 return kvm_pmu_get_msr(vcpu, msr, pdata);
2364 case MSR_IA32_UCODE_REV:
2365 data = 0x100000000ULL;
2368 data = 0x500 | KVM_NR_VAR_MTRR;
2370 case 0x200 ... 0x2ff:
2371 return get_msr_mtrr(vcpu, msr, pdata);
2372 case 0xcd: /* fsb frequency */
2376 * MSR_EBC_FREQUENCY_ID
2377 * Conservative value valid for even the basic CPU models.
2378 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2379 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2380 * and 266MHz for model 3, or 4. Set Core Clock
2381 * Frequency to System Bus Frequency Ratio to 1 (bits
2382 * 31:24) even though these are only valid for CPU
2383 * models > 2, however guests may end up dividing or
2384 * multiplying by zero otherwise.
2386 case MSR_EBC_FREQUENCY_ID:
2389 case MSR_IA32_APICBASE:
2390 data = kvm_get_apic_base(vcpu);
2392 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2393 return kvm_x2apic_msr_read(vcpu, msr, pdata);
2395 case MSR_IA32_TSCDEADLINE:
2396 data = kvm_get_lapic_tscdeadline_msr(vcpu);
2398 case MSR_IA32_TSC_ADJUST:
2399 data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2401 case MSR_IA32_MISC_ENABLE:
2402 data = vcpu->arch.ia32_misc_enable_msr;
2404 case MSR_IA32_PERF_STATUS:
2405 /* TSC increment by tick */
2407 /* CPU multiplier */
2408 data |= (((uint64_t)4ULL) << 40);
2411 data = vcpu->arch.efer;
2413 case MSR_KVM_WALL_CLOCK:
2414 case MSR_KVM_WALL_CLOCK_NEW:
2415 data = vcpu->kvm->arch.wall_clock;
2417 case MSR_KVM_SYSTEM_TIME:
2418 case MSR_KVM_SYSTEM_TIME_NEW:
2419 data = vcpu->arch.time;
2421 case MSR_KVM_ASYNC_PF_EN:
2422 data = vcpu->arch.apf.msr_val;
2424 case MSR_KVM_STEAL_TIME:
2425 data = vcpu->arch.st.msr_val;
2427 case MSR_KVM_PV_EOI_EN:
2428 data = vcpu->arch.pv_eoi.msr_val;
2430 case MSR_IA32_P5_MC_ADDR:
2431 case MSR_IA32_P5_MC_TYPE:
2432 case MSR_IA32_MCG_CAP:
2433 case MSR_IA32_MCG_CTL:
2434 case MSR_IA32_MCG_STATUS:
2435 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2436 return get_msr_mce(vcpu, msr, pdata);
2437 case MSR_K7_CLK_CTL:
2439 * Provide expected ramp-up count for K7. All other
2440 * are set to zero, indicating minimum divisors for
2443 * This prevents guest kernels on AMD host with CPU
2444 * type 6, model 8 and higher from exploding due to
2445 * the rdmsr failing.
2449 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2450 if (kvm_hv_msr_partition_wide(msr)) {
2452 mutex_lock(&vcpu->kvm->lock);
2453 r = get_msr_hyperv_pw(vcpu, msr, pdata);
2454 mutex_unlock(&vcpu->kvm->lock);
2457 return get_msr_hyperv(vcpu, msr, pdata);
2459 case MSR_IA32_BBL_CR_CTL3:
2460 /* This legacy MSR exists but isn't fully documented in current
2461 * silicon. It is however accessed by winxp in very narrow
2462 * scenarios where it sets bit #19, itself documented as
2463 * a "reserved" bit. Best effort attempt to source coherent
2464 * read data here should the balance of the register be
2465 * interpreted by the guest:
2467 * L2 cache control register 3: 64GB range, 256KB size,
2468 * enabled, latency 0x1, configured
2472 case MSR_AMD64_OSVW_ID_LENGTH:
2473 if (!guest_cpuid_has_osvw(vcpu))
2475 data = vcpu->arch.osvw.length;
2477 case MSR_AMD64_OSVW_STATUS:
2478 if (!guest_cpuid_has_osvw(vcpu))
2480 data = vcpu->arch.osvw.status;
2483 if (kvm_pmu_msr(vcpu, msr))
2484 return kvm_pmu_get_msr(vcpu, msr, pdata);
2486 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2489 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2497 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2500 * Read or write a bunch of msrs. All parameters are kernel addresses.
2502 * @return number of msrs set successfully.
2504 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2505 struct kvm_msr_entry *entries,
2506 int (*do_msr)(struct kvm_vcpu *vcpu,
2507 unsigned index, u64 *data))
2511 idx = srcu_read_lock(&vcpu->kvm->srcu);
2512 for (i = 0; i < msrs->nmsrs; ++i)
2513 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2515 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2521 * Read or write a bunch of msrs. Parameters are user addresses.
2523 * @return number of msrs set successfully.
2525 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2526 int (*do_msr)(struct kvm_vcpu *vcpu,
2527 unsigned index, u64 *data),
2530 struct kvm_msrs msrs;
2531 struct kvm_msr_entry *entries;
2536 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2540 if (msrs.nmsrs >= MAX_IO_MSRS)
2543 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2544 entries = memdup_user(user_msrs->entries, size);
2545 if (IS_ERR(entries)) {
2546 r = PTR_ERR(entries);
2550 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2555 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2566 int kvm_dev_ioctl_check_extension(long ext)
2571 case KVM_CAP_IRQCHIP:
2573 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2574 case KVM_CAP_SET_TSS_ADDR:
2575 case KVM_CAP_EXT_CPUID:
2576 case KVM_CAP_EXT_EMUL_CPUID:
2577 case KVM_CAP_CLOCKSOURCE:
2579 case KVM_CAP_NOP_IO_DELAY:
2580 case KVM_CAP_MP_STATE:
2581 case KVM_CAP_SYNC_MMU:
2582 case KVM_CAP_USER_NMI:
2583 case KVM_CAP_REINJECT_CONTROL:
2584 case KVM_CAP_IRQ_INJECT_STATUS:
2586 case KVM_CAP_IOEVENTFD:
2588 case KVM_CAP_PIT_STATE2:
2589 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2590 case KVM_CAP_XEN_HVM:
2591 case KVM_CAP_ADJUST_CLOCK:
2592 case KVM_CAP_VCPU_EVENTS:
2593 case KVM_CAP_HYPERV:
2594 case KVM_CAP_HYPERV_VAPIC:
2595 case KVM_CAP_HYPERV_SPIN:
2596 case KVM_CAP_PCI_SEGMENT:
2597 case KVM_CAP_DEBUGREGS:
2598 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2600 case KVM_CAP_ASYNC_PF:
2601 case KVM_CAP_GET_TSC_KHZ:
2602 case KVM_CAP_KVMCLOCK_CTRL:
2603 case KVM_CAP_READONLY_MEM:
2604 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2605 case KVM_CAP_ASSIGN_DEV_IRQ:
2606 case KVM_CAP_PCI_2_3:
2610 case KVM_CAP_COALESCED_MMIO:
2611 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2614 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2616 case KVM_CAP_NR_VCPUS:
2617 r = KVM_SOFT_MAX_VCPUS;
2619 case KVM_CAP_MAX_VCPUS:
2622 case KVM_CAP_NR_MEMSLOTS:
2623 r = KVM_USER_MEM_SLOTS;
2625 case KVM_CAP_PV_MMU: /* obsolete */
2628 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2630 r = iommu_present(&pci_bus_type);
2634 r = KVM_MAX_MCE_BANKS;
2639 case KVM_CAP_TSC_CONTROL:
2640 r = kvm_has_tsc_control;
2642 case KVM_CAP_TSC_DEADLINE_TIMER:
2643 r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2653 long kvm_arch_dev_ioctl(struct file *filp,
2654 unsigned int ioctl, unsigned long arg)
2656 void __user *argp = (void __user *)arg;
2660 case KVM_GET_MSR_INDEX_LIST: {
2661 struct kvm_msr_list __user *user_msr_list = argp;
2662 struct kvm_msr_list msr_list;
2666 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2669 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2670 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2673 if (n < msr_list.nmsrs)
2676 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2677 num_msrs_to_save * sizeof(u32)))
2679 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2681 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2686 case KVM_GET_SUPPORTED_CPUID:
2687 case KVM_GET_EMULATED_CPUID: {
2688 struct kvm_cpuid2 __user *cpuid_arg = argp;
2689 struct kvm_cpuid2 cpuid;
2692 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2695 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2701 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2706 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2709 mce_cap = KVM_MCE_CAP_SUPPORTED;
2711 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2723 static void wbinvd_ipi(void *garbage)
2728 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2730 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2733 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2735 /* Address WBINVD may be executed by guest */
2736 if (need_emulate_wbinvd(vcpu)) {
2737 if (kvm_x86_ops->has_wbinvd_exit())
2738 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2739 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2740 smp_call_function_single(vcpu->cpu,
2741 wbinvd_ipi, NULL, 1);
2744 kvm_x86_ops->vcpu_load(vcpu, cpu);
2746 /* Apply any externally detected TSC adjustments (due to suspend) */
2747 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2748 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2749 vcpu->arch.tsc_offset_adjustment = 0;
2750 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
2753 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2754 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2755 native_read_tsc() - vcpu->arch.last_host_tsc;
2757 mark_tsc_unstable("KVM discovered backwards TSC");
2758 if (check_tsc_unstable()) {
2759 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2760 vcpu->arch.last_guest_tsc);
2761 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2762 vcpu->arch.tsc_catchup = 1;
2765 * On a host with synchronized TSC, there is no need to update
2766 * kvmclock on vcpu->cpu migration
2768 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2769 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2770 if (vcpu->cpu != cpu)
2771 kvm_migrate_timers(vcpu);
2775 accumulate_steal_time(vcpu);
2776 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2779 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2781 kvm_x86_ops->vcpu_put(vcpu);
2782 kvm_put_guest_fpu(vcpu);
2783 vcpu->arch.last_host_tsc = native_read_tsc();
2786 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2787 struct kvm_lapic_state *s)
2789 kvm_x86_ops->sync_pir_to_irr(vcpu);
2790 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2795 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2796 struct kvm_lapic_state *s)
2798 kvm_apic_post_state_restore(vcpu, s);
2799 update_cr8_intercept(vcpu);
2804 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2805 struct kvm_interrupt *irq)
2807 if (irq->irq >= KVM_NR_INTERRUPTS)
2809 if (irqchip_in_kernel(vcpu->kvm))
2812 kvm_queue_interrupt(vcpu, irq->irq, false);
2813 kvm_make_request(KVM_REQ_EVENT, vcpu);
2818 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2820 kvm_inject_nmi(vcpu);
2825 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2826 struct kvm_tpr_access_ctl *tac)
2830 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2834 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2838 unsigned bank_num = mcg_cap & 0xff, bank;
2841 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2843 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2846 vcpu->arch.mcg_cap = mcg_cap;
2847 /* Init IA32_MCG_CTL to all 1s */
2848 if (mcg_cap & MCG_CTL_P)
2849 vcpu->arch.mcg_ctl = ~(u64)0;
2850 /* Init IA32_MCi_CTL to all 1s */
2851 for (bank = 0; bank < bank_num; bank++)
2852 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2857 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2858 struct kvm_x86_mce *mce)
2860 u64 mcg_cap = vcpu->arch.mcg_cap;
2861 unsigned bank_num = mcg_cap & 0xff;
2862 u64 *banks = vcpu->arch.mce_banks;
2864 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2867 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2868 * reporting is disabled
2870 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2871 vcpu->arch.mcg_ctl != ~(u64)0)
2873 banks += 4 * mce->bank;
2875 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2876 * reporting is disabled for the bank
2878 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2880 if (mce->status & MCI_STATUS_UC) {
2881 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2882 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2883 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2886 if (banks[1] & MCI_STATUS_VAL)
2887 mce->status |= MCI_STATUS_OVER;
2888 banks[2] = mce->addr;
2889 banks[3] = mce->misc;
2890 vcpu->arch.mcg_status = mce->mcg_status;
2891 banks[1] = mce->status;
2892 kvm_queue_exception(vcpu, MC_VECTOR);
2893 } else if (!(banks[1] & MCI_STATUS_VAL)
2894 || !(banks[1] & MCI_STATUS_UC)) {
2895 if (banks[1] & MCI_STATUS_VAL)
2896 mce->status |= MCI_STATUS_OVER;
2897 banks[2] = mce->addr;
2898 banks[3] = mce->misc;
2899 banks[1] = mce->status;
2901 banks[1] |= MCI_STATUS_OVER;
2905 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2906 struct kvm_vcpu_events *events)
2909 events->exception.injected =
2910 vcpu->arch.exception.pending &&
2911 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2912 events->exception.nr = vcpu->arch.exception.nr;
2913 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2914 events->exception.pad = 0;
2915 events->exception.error_code = vcpu->arch.exception.error_code;
2917 events->interrupt.injected =
2918 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2919 events->interrupt.nr = vcpu->arch.interrupt.nr;
2920 events->interrupt.soft = 0;
2921 events->interrupt.shadow =
2922 kvm_x86_ops->get_interrupt_shadow(vcpu,
2923 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2925 events->nmi.injected = vcpu->arch.nmi_injected;
2926 events->nmi.pending = vcpu->arch.nmi_pending != 0;
2927 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2928 events->nmi.pad = 0;
2930 events->sipi_vector = 0; /* never valid when reporting to user space */
2932 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2933 | KVM_VCPUEVENT_VALID_SHADOW);
2934 memset(&events->reserved, 0, sizeof(events->reserved));
2937 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2938 struct kvm_vcpu_events *events)
2940 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2941 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2942 | KVM_VCPUEVENT_VALID_SHADOW))
2946 vcpu->arch.exception.pending = events->exception.injected;
2947 vcpu->arch.exception.nr = events->exception.nr;
2948 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2949 vcpu->arch.exception.error_code = events->exception.error_code;
2951 vcpu->arch.interrupt.pending = events->interrupt.injected;
2952 vcpu->arch.interrupt.nr = events->interrupt.nr;
2953 vcpu->arch.interrupt.soft = events->interrupt.soft;
2954 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2955 kvm_x86_ops->set_interrupt_shadow(vcpu,
2956 events->interrupt.shadow);
2958 vcpu->arch.nmi_injected = events->nmi.injected;
2959 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2960 vcpu->arch.nmi_pending = events->nmi.pending;
2961 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2963 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
2964 kvm_vcpu_has_lapic(vcpu))
2965 vcpu->arch.apic->sipi_vector = events->sipi_vector;
2967 kvm_make_request(KVM_REQ_EVENT, vcpu);
2972 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2973 struct kvm_debugregs *dbgregs)
2975 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2976 dbgregs->dr6 = vcpu->arch.dr6;
2977 dbgregs->dr7 = vcpu->arch.dr7;
2979 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2982 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2983 struct kvm_debugregs *dbgregs)
2988 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2989 vcpu->arch.dr6 = dbgregs->dr6;
2990 vcpu->arch.dr7 = dbgregs->dr7;
2995 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2996 struct kvm_xsave *guest_xsave)
2998 if (cpu_has_xsave) {
2999 memcpy(guest_xsave->region,
3000 &vcpu->arch.guest_fpu.state->xsave,
3001 vcpu->arch.guest_xstate_size);
3002 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] &=
3003 vcpu->arch.guest_supported_xcr0 | XSTATE_FPSSE;
3005 memcpy(guest_xsave->region,
3006 &vcpu->arch.guest_fpu.state->fxsave,
3007 sizeof(struct i387_fxsave_struct));
3008 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3013 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3014 struct kvm_xsave *guest_xsave)
3017 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3019 if (cpu_has_xsave) {
3021 * Here we allow setting states that are not present in
3022 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3023 * with old userspace.
3025 if (xstate_bv & ~KVM_SUPPORTED_XCR0)
3027 if (xstate_bv & ~host_xcr0)
3029 memcpy(&vcpu->arch.guest_fpu.state->xsave,
3030 guest_xsave->region, vcpu->arch.guest_xstate_size);
3032 if (xstate_bv & ~XSTATE_FPSSE)
3034 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
3035 guest_xsave->region, sizeof(struct i387_fxsave_struct));
3040 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3041 struct kvm_xcrs *guest_xcrs)
3043 if (!cpu_has_xsave) {
3044 guest_xcrs->nr_xcrs = 0;
3048 guest_xcrs->nr_xcrs = 1;
3049 guest_xcrs->flags = 0;
3050 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3051 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3054 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3055 struct kvm_xcrs *guest_xcrs)
3062 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3065 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3066 /* Only support XCR0 currently */
3067 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3068 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3069 guest_xcrs->xcrs[i].value);
3078 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3079 * stopped by the hypervisor. This function will be called from the host only.
3080 * EINVAL is returned when the host attempts to set the flag for a guest that
3081 * does not support pv clocks.
3083 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3085 if (!vcpu->arch.pv_time_enabled)
3087 vcpu->arch.pvclock_set_guest_stopped_request = true;
3088 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3092 long kvm_arch_vcpu_ioctl(struct file *filp,
3093 unsigned int ioctl, unsigned long arg)
3095 struct kvm_vcpu *vcpu = filp->private_data;
3096 void __user *argp = (void __user *)arg;
3099 struct kvm_lapic_state *lapic;
3100 struct kvm_xsave *xsave;
3101 struct kvm_xcrs *xcrs;
3107 case KVM_GET_LAPIC: {
3109 if (!vcpu->arch.apic)
3111 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3116 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3120 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3125 case KVM_SET_LAPIC: {
3127 if (!vcpu->arch.apic)
3129 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3130 if (IS_ERR(u.lapic))
3131 return PTR_ERR(u.lapic);
3133 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3136 case KVM_INTERRUPT: {
3137 struct kvm_interrupt irq;
3140 if (copy_from_user(&irq, argp, sizeof irq))
3142 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3146 r = kvm_vcpu_ioctl_nmi(vcpu);
3149 case KVM_SET_CPUID: {
3150 struct kvm_cpuid __user *cpuid_arg = argp;
3151 struct kvm_cpuid cpuid;
3154 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3156 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3159 case KVM_SET_CPUID2: {
3160 struct kvm_cpuid2 __user *cpuid_arg = argp;
3161 struct kvm_cpuid2 cpuid;
3164 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3166 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3167 cpuid_arg->entries);
3170 case KVM_GET_CPUID2: {
3171 struct kvm_cpuid2 __user *cpuid_arg = argp;
3172 struct kvm_cpuid2 cpuid;
3175 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3177 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3178 cpuid_arg->entries);
3182 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3188 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3191 r = msr_io(vcpu, argp, do_set_msr, 0);
3193 case KVM_TPR_ACCESS_REPORTING: {
3194 struct kvm_tpr_access_ctl tac;
3197 if (copy_from_user(&tac, argp, sizeof tac))
3199 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3203 if (copy_to_user(argp, &tac, sizeof tac))
3208 case KVM_SET_VAPIC_ADDR: {
3209 struct kvm_vapic_addr va;
3212 if (!irqchip_in_kernel(vcpu->kvm))
3215 if (copy_from_user(&va, argp, sizeof va))
3218 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3221 case KVM_X86_SETUP_MCE: {
3225 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3227 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3230 case KVM_X86_SET_MCE: {
3231 struct kvm_x86_mce mce;
3234 if (copy_from_user(&mce, argp, sizeof mce))
3236 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3239 case KVM_GET_VCPU_EVENTS: {
3240 struct kvm_vcpu_events events;
3242 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3245 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3250 case KVM_SET_VCPU_EVENTS: {
3251 struct kvm_vcpu_events events;
3254 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3257 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3260 case KVM_GET_DEBUGREGS: {
3261 struct kvm_debugregs dbgregs;
3263 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3266 if (copy_to_user(argp, &dbgregs,
3267 sizeof(struct kvm_debugregs)))
3272 case KVM_SET_DEBUGREGS: {
3273 struct kvm_debugregs dbgregs;
3276 if (copy_from_user(&dbgregs, argp,
3277 sizeof(struct kvm_debugregs)))
3280 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3283 case KVM_GET_XSAVE: {
3284 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3289 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3292 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3297 case KVM_SET_XSAVE: {
3298 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3299 if (IS_ERR(u.xsave))
3300 return PTR_ERR(u.xsave);
3302 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3305 case KVM_GET_XCRS: {
3306 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3311 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3314 if (copy_to_user(argp, u.xcrs,
3315 sizeof(struct kvm_xcrs)))
3320 case KVM_SET_XCRS: {
3321 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3323 return PTR_ERR(u.xcrs);
3325 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3328 case KVM_SET_TSC_KHZ: {
3332 user_tsc_khz = (u32)arg;
3334 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3337 if (user_tsc_khz == 0)
3338 user_tsc_khz = tsc_khz;
3340 kvm_set_tsc_khz(vcpu, user_tsc_khz);
3345 case KVM_GET_TSC_KHZ: {
3346 r = vcpu->arch.virtual_tsc_khz;
3349 case KVM_KVMCLOCK_CTRL: {
3350 r = kvm_set_guest_paused(vcpu);
3361 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3363 return VM_FAULT_SIGBUS;
3366 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3370 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3372 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3376 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3379 kvm->arch.ept_identity_map_addr = ident_addr;
3383 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3384 u32 kvm_nr_mmu_pages)
3386 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3389 mutex_lock(&kvm->slots_lock);
3391 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3392 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3394 mutex_unlock(&kvm->slots_lock);
3398 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3400 return kvm->arch.n_max_mmu_pages;
3403 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3408 switch (chip->chip_id) {
3409 case KVM_IRQCHIP_PIC_MASTER:
3410 memcpy(&chip->chip.pic,
3411 &pic_irqchip(kvm)->pics[0],
3412 sizeof(struct kvm_pic_state));
3414 case KVM_IRQCHIP_PIC_SLAVE:
3415 memcpy(&chip->chip.pic,
3416 &pic_irqchip(kvm)->pics[1],
3417 sizeof(struct kvm_pic_state));
3419 case KVM_IRQCHIP_IOAPIC:
3420 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3429 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3434 switch (chip->chip_id) {
3435 case KVM_IRQCHIP_PIC_MASTER:
3436 spin_lock(&pic_irqchip(kvm)->lock);
3437 memcpy(&pic_irqchip(kvm)->pics[0],
3439 sizeof(struct kvm_pic_state));
3440 spin_unlock(&pic_irqchip(kvm)->lock);
3442 case KVM_IRQCHIP_PIC_SLAVE:
3443 spin_lock(&pic_irqchip(kvm)->lock);
3444 memcpy(&pic_irqchip(kvm)->pics[1],
3446 sizeof(struct kvm_pic_state));
3447 spin_unlock(&pic_irqchip(kvm)->lock);
3449 case KVM_IRQCHIP_IOAPIC:
3450 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3456 kvm_pic_update_irq(pic_irqchip(kvm));
3460 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3464 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3465 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3466 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3470 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3474 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3475 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3476 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3477 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3481 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3485 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3486 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3487 sizeof(ps->channels));
3488 ps->flags = kvm->arch.vpit->pit_state.flags;
3489 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3490 memset(&ps->reserved, 0, sizeof(ps->reserved));
3494 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3496 int r = 0, start = 0;
3497 u32 prev_legacy, cur_legacy;
3498 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3499 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3500 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3501 if (!prev_legacy && cur_legacy)
3503 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3504 sizeof(kvm->arch.vpit->pit_state.channels));
3505 kvm->arch.vpit->pit_state.flags = ps->flags;
3506 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3507 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3511 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3512 struct kvm_reinject_control *control)
3514 if (!kvm->arch.vpit)
3516 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3517 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3518 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3523 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3524 * @kvm: kvm instance
3525 * @log: slot id and address to which we copy the log
3527 * We need to keep it in mind that VCPU threads can write to the bitmap
3528 * concurrently. So, to avoid losing data, we keep the following order for
3531 * 1. Take a snapshot of the bit and clear it if needed.
3532 * 2. Write protect the corresponding page.
3533 * 3. Flush TLB's if needed.
3534 * 4. Copy the snapshot to the userspace.
3536 * Between 2 and 3, the guest may write to the page using the remaining TLB
3537 * entry. This is not a problem because the page will be reported dirty at
3538 * step 4 using the snapshot taken before and step 3 ensures that successive
3539 * writes will be logged for the next call.
3541 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3544 struct kvm_memory_slot *memslot;
3546 unsigned long *dirty_bitmap;
3547 unsigned long *dirty_bitmap_buffer;
3548 bool is_dirty = false;
3550 mutex_lock(&kvm->slots_lock);
3553 if (log->slot >= KVM_USER_MEM_SLOTS)
3556 memslot = id_to_memslot(kvm->memslots, log->slot);
3558 dirty_bitmap = memslot->dirty_bitmap;
3563 n = kvm_dirty_bitmap_bytes(memslot);
3565 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
3566 memset(dirty_bitmap_buffer, 0, n);
3568 spin_lock(&kvm->mmu_lock);
3570 for (i = 0; i < n / sizeof(long); i++) {
3574 if (!dirty_bitmap[i])
3579 mask = xchg(&dirty_bitmap[i], 0);
3580 dirty_bitmap_buffer[i] = mask;
3582 offset = i * BITS_PER_LONG;
3583 kvm_mmu_write_protect_pt_masked(kvm, memslot, offset, mask);
3586 kvm_flush_remote_tlbs(kvm);
3588 spin_unlock(&kvm->mmu_lock);
3591 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
3596 mutex_unlock(&kvm->slots_lock);
3600 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3603 if (!irqchip_in_kernel(kvm))
3606 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3607 irq_event->irq, irq_event->level,
3612 long kvm_arch_vm_ioctl(struct file *filp,
3613 unsigned int ioctl, unsigned long arg)
3615 struct kvm *kvm = filp->private_data;
3616 void __user *argp = (void __user *)arg;
3619 * This union makes it completely explicit to gcc-3.x
3620 * that these two variables' stack usage should be
3621 * combined, not added together.
3624 struct kvm_pit_state ps;
3625 struct kvm_pit_state2 ps2;
3626 struct kvm_pit_config pit_config;
3630 case KVM_SET_TSS_ADDR:
3631 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3633 case KVM_SET_IDENTITY_MAP_ADDR: {
3637 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3639 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3642 case KVM_SET_NR_MMU_PAGES:
3643 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3645 case KVM_GET_NR_MMU_PAGES:
3646 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3648 case KVM_CREATE_IRQCHIP: {
3649 struct kvm_pic *vpic;
3651 mutex_lock(&kvm->lock);
3654 goto create_irqchip_unlock;
3656 if (atomic_read(&kvm->online_vcpus))
3657 goto create_irqchip_unlock;
3659 vpic = kvm_create_pic(kvm);
3661 r = kvm_ioapic_init(kvm);
3663 mutex_lock(&kvm->slots_lock);
3664 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3666 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3668 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3670 mutex_unlock(&kvm->slots_lock);
3672 goto create_irqchip_unlock;
3675 goto create_irqchip_unlock;
3677 kvm->arch.vpic = vpic;
3679 r = kvm_setup_default_irq_routing(kvm);
3681 mutex_lock(&kvm->slots_lock);
3682 mutex_lock(&kvm->irq_lock);
3683 kvm_ioapic_destroy(kvm);
3684 kvm_destroy_pic(kvm);
3685 mutex_unlock(&kvm->irq_lock);
3686 mutex_unlock(&kvm->slots_lock);
3688 create_irqchip_unlock:
3689 mutex_unlock(&kvm->lock);
3692 case KVM_CREATE_PIT:
3693 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3695 case KVM_CREATE_PIT2:
3697 if (copy_from_user(&u.pit_config, argp,
3698 sizeof(struct kvm_pit_config)))
3701 mutex_lock(&kvm->slots_lock);
3704 goto create_pit_unlock;
3706 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3710 mutex_unlock(&kvm->slots_lock);
3712 case KVM_GET_IRQCHIP: {
3713 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3714 struct kvm_irqchip *chip;
3716 chip = memdup_user(argp, sizeof(*chip));
3723 if (!irqchip_in_kernel(kvm))
3724 goto get_irqchip_out;
3725 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3727 goto get_irqchip_out;
3729 if (copy_to_user(argp, chip, sizeof *chip))
3730 goto get_irqchip_out;
3736 case KVM_SET_IRQCHIP: {
3737 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3738 struct kvm_irqchip *chip;
3740 chip = memdup_user(argp, sizeof(*chip));
3747 if (!irqchip_in_kernel(kvm))
3748 goto set_irqchip_out;
3749 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3751 goto set_irqchip_out;
3759 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3762 if (!kvm->arch.vpit)
3764 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3768 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3775 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3778 if (!kvm->arch.vpit)
3780 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3783 case KVM_GET_PIT2: {
3785 if (!kvm->arch.vpit)
3787 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3791 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3796 case KVM_SET_PIT2: {
3798 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3801 if (!kvm->arch.vpit)
3803 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3806 case KVM_REINJECT_CONTROL: {
3807 struct kvm_reinject_control control;
3809 if (copy_from_user(&control, argp, sizeof(control)))
3811 r = kvm_vm_ioctl_reinject(kvm, &control);
3814 case KVM_XEN_HVM_CONFIG: {
3816 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3817 sizeof(struct kvm_xen_hvm_config)))
3820 if (kvm->arch.xen_hvm_config.flags)
3825 case KVM_SET_CLOCK: {
3826 struct kvm_clock_data user_ns;
3831 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3839 local_irq_disable();
3840 now_ns = get_kernel_ns();
3841 delta = user_ns.clock - now_ns;
3843 kvm->arch.kvmclock_offset = delta;
3844 kvm_gen_update_masterclock(kvm);
3847 case KVM_GET_CLOCK: {
3848 struct kvm_clock_data user_ns;
3851 local_irq_disable();
3852 now_ns = get_kernel_ns();
3853 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3856 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3859 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3872 static void kvm_init_msr_list(void)
3877 /* skip the first msrs in the list. KVM-specific */
3878 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3879 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3882 msrs_to_save[j] = msrs_to_save[i];
3885 num_msrs_to_save = j;
3888 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3896 if (!(vcpu->arch.apic &&
3897 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3898 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3909 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3916 if (!(vcpu->arch.apic &&
3917 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3918 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3920 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3930 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3931 struct kvm_segment *var, int seg)
3933 kvm_x86_ops->set_segment(vcpu, var, seg);
3936 void kvm_get_segment(struct kvm_vcpu *vcpu,
3937 struct kvm_segment *var, int seg)
3939 kvm_x86_ops->get_segment(vcpu, var, seg);
3942 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3945 struct x86_exception exception;
3947 BUG_ON(!mmu_is_nested(vcpu));
3949 /* NPT walks are always user-walks */
3950 access |= PFERR_USER_MASK;
3951 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3956 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3957 struct x86_exception *exception)
3959 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3960 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3963 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3964 struct x86_exception *exception)
3966 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3967 access |= PFERR_FETCH_MASK;
3968 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3971 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3972 struct x86_exception *exception)
3974 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3975 access |= PFERR_WRITE_MASK;
3976 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3979 /* uses this to access any guest's mapped memory without checking CPL */
3980 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3981 struct x86_exception *exception)
3983 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3986 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3987 struct kvm_vcpu *vcpu, u32 access,
3988 struct x86_exception *exception)
3991 int r = X86EMUL_CONTINUE;
3994 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3996 unsigned offset = addr & (PAGE_SIZE-1);
3997 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4000 if (gpa == UNMAPPED_GVA)
4001 return X86EMUL_PROPAGATE_FAULT;
4002 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
4004 r = X86EMUL_IO_NEEDED;
4016 /* used for instruction fetching */
4017 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4018 gva_t addr, void *val, unsigned int bytes,
4019 struct x86_exception *exception)
4021 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4022 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4024 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
4025 access | PFERR_FETCH_MASK,
4029 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4030 gva_t addr, void *val, unsigned int bytes,
4031 struct x86_exception *exception)
4033 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4034 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4036 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4039 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4041 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4042 gva_t addr, void *val, unsigned int bytes,
4043 struct x86_exception *exception)
4045 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4046 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4049 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4050 gva_t addr, void *val,
4052 struct x86_exception *exception)
4054 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4056 int r = X86EMUL_CONTINUE;
4059 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4062 unsigned offset = addr & (PAGE_SIZE-1);
4063 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4066 if (gpa == UNMAPPED_GVA)
4067 return X86EMUL_PROPAGATE_FAULT;
4068 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
4070 r = X86EMUL_IO_NEEDED;
4081 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4083 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4084 gpa_t *gpa, struct x86_exception *exception,
4087 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4088 | (write ? PFERR_WRITE_MASK : 0);
4090 if (vcpu_match_mmio_gva(vcpu, gva)
4091 && !permission_fault(vcpu->arch.walk_mmu, vcpu->arch.access, access)) {
4092 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4093 (gva & (PAGE_SIZE - 1));
4094 trace_vcpu_match_mmio(gva, *gpa, write, false);
4098 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4100 if (*gpa == UNMAPPED_GVA)
4103 /* For APIC access vmexit */
4104 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4107 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4108 trace_vcpu_match_mmio(gva, *gpa, write, true);
4115 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4116 const void *val, int bytes)
4120 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4123 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4127 struct read_write_emulator_ops {
4128 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4130 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4131 void *val, int bytes);
4132 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4133 int bytes, void *val);
4134 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4135 void *val, int bytes);
4139 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4141 if (vcpu->mmio_read_completed) {
4142 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4143 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4144 vcpu->mmio_read_completed = 0;
4151 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4152 void *val, int bytes)
4154 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4157 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4158 void *val, int bytes)
4160 return emulator_write_phys(vcpu, gpa, val, bytes);
4163 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4165 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4166 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4169 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4170 void *val, int bytes)
4172 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4173 return X86EMUL_IO_NEEDED;
4176 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4177 void *val, int bytes)
4179 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4181 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4182 return X86EMUL_CONTINUE;
4185 static const struct read_write_emulator_ops read_emultor = {
4186 .read_write_prepare = read_prepare,
4187 .read_write_emulate = read_emulate,
4188 .read_write_mmio = vcpu_mmio_read,
4189 .read_write_exit_mmio = read_exit_mmio,
4192 static const struct read_write_emulator_ops write_emultor = {
4193 .read_write_emulate = write_emulate,
4194 .read_write_mmio = write_mmio,
4195 .read_write_exit_mmio = write_exit_mmio,
4199 static int emulator_read_write_onepage(unsigned long addr, void *val,
4201 struct x86_exception *exception,
4202 struct kvm_vcpu *vcpu,
4203 const struct read_write_emulator_ops *ops)
4207 bool write = ops->write;
4208 struct kvm_mmio_fragment *frag;
4210 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4213 return X86EMUL_PROPAGATE_FAULT;
4215 /* For APIC access vmexit */
4219 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4220 return X86EMUL_CONTINUE;
4224 * Is this MMIO handled locally?
4226 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4227 if (handled == bytes)
4228 return X86EMUL_CONTINUE;
4234 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4235 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4239 return X86EMUL_CONTINUE;
4242 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4243 void *val, unsigned int bytes,
4244 struct x86_exception *exception,
4245 const struct read_write_emulator_ops *ops)
4247 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4251 if (ops->read_write_prepare &&
4252 ops->read_write_prepare(vcpu, val, bytes))
4253 return X86EMUL_CONTINUE;
4255 vcpu->mmio_nr_fragments = 0;
4257 /* Crossing a page boundary? */
4258 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4261 now = -addr & ~PAGE_MASK;
4262 rc = emulator_read_write_onepage(addr, val, now, exception,
4265 if (rc != X86EMUL_CONTINUE)
4272 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4274 if (rc != X86EMUL_CONTINUE)
4277 if (!vcpu->mmio_nr_fragments)
4280 gpa = vcpu->mmio_fragments[0].gpa;
4282 vcpu->mmio_needed = 1;
4283 vcpu->mmio_cur_fragment = 0;
4285 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4286 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4287 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4288 vcpu->run->mmio.phys_addr = gpa;
4290 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4293 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4297 struct x86_exception *exception)
4299 return emulator_read_write(ctxt, addr, val, bytes,
4300 exception, &read_emultor);
4303 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4307 struct x86_exception *exception)
4309 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4310 exception, &write_emultor);
4313 #define CMPXCHG_TYPE(t, ptr, old, new) \
4314 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4316 #ifdef CONFIG_X86_64
4317 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4319 # define CMPXCHG64(ptr, old, new) \
4320 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4323 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4328 struct x86_exception *exception)
4330 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4336 /* guests cmpxchg8b have to be emulated atomically */
4337 if (bytes > 8 || (bytes & (bytes - 1)))
4340 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4342 if (gpa == UNMAPPED_GVA ||
4343 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4346 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4349 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4350 if (is_error_page(page))
4353 kaddr = kmap_atomic(page);
4354 kaddr += offset_in_page(gpa);
4357 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4360 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4363 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4366 exchanged = CMPXCHG64(kaddr, old, new);
4371 kunmap_atomic(kaddr);
4372 kvm_release_page_dirty(page);
4375 return X86EMUL_CMPXCHG_FAILED;
4377 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4379 return X86EMUL_CONTINUE;
4382 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4384 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4387 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4389 /* TODO: String I/O for in kernel device */
4392 if (vcpu->arch.pio.in)
4393 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4394 vcpu->arch.pio.size, pd);
4396 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4397 vcpu->arch.pio.port, vcpu->arch.pio.size,
4402 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4403 unsigned short port, void *val,
4404 unsigned int count, bool in)
4406 trace_kvm_pio(!in, port, size, count);
4408 vcpu->arch.pio.port = port;
4409 vcpu->arch.pio.in = in;
4410 vcpu->arch.pio.count = count;
4411 vcpu->arch.pio.size = size;
4413 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4414 vcpu->arch.pio.count = 0;
4418 vcpu->run->exit_reason = KVM_EXIT_IO;
4419 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4420 vcpu->run->io.size = size;
4421 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4422 vcpu->run->io.count = count;
4423 vcpu->run->io.port = port;
4428 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4429 int size, unsigned short port, void *val,
4432 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4435 if (vcpu->arch.pio.count)
4438 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4441 memcpy(val, vcpu->arch.pio_data, size * count);
4442 vcpu->arch.pio.count = 0;
4449 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4450 int size, unsigned short port,
4451 const void *val, unsigned int count)
4453 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4455 memcpy(vcpu->arch.pio_data, val, size * count);
4456 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4459 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4461 return kvm_x86_ops->get_segment_base(vcpu, seg);
4464 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4466 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4469 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4471 if (!need_emulate_wbinvd(vcpu))
4472 return X86EMUL_CONTINUE;
4474 if (kvm_x86_ops->has_wbinvd_exit()) {
4475 int cpu = get_cpu();
4477 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4478 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4479 wbinvd_ipi, NULL, 1);
4481 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4484 return X86EMUL_CONTINUE;
4486 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4488 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4490 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4493 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4495 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4498 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4501 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4504 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4506 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4509 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4511 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4512 unsigned long value;
4516 value = kvm_read_cr0(vcpu);
4519 value = vcpu->arch.cr2;
4522 value = kvm_read_cr3(vcpu);
4525 value = kvm_read_cr4(vcpu);
4528 value = kvm_get_cr8(vcpu);
4531 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4538 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4540 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4545 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4548 vcpu->arch.cr2 = val;
4551 res = kvm_set_cr3(vcpu, val);
4554 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4557 res = kvm_set_cr8(vcpu, val);
4560 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4567 static void emulator_set_rflags(struct x86_emulate_ctxt *ctxt, ulong val)
4569 kvm_set_rflags(emul_to_vcpu(ctxt), val);
4572 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4574 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4577 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4579 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4582 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4584 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4587 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4589 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4592 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4594 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4597 static unsigned long emulator_get_cached_segment_base(
4598 struct x86_emulate_ctxt *ctxt, int seg)
4600 return get_segment_base(emul_to_vcpu(ctxt), seg);
4603 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4604 struct desc_struct *desc, u32 *base3,
4607 struct kvm_segment var;
4609 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4610 *selector = var.selector;
4613 memset(desc, 0, sizeof(*desc));
4619 set_desc_limit(desc, var.limit);
4620 set_desc_base(desc, (unsigned long)var.base);
4621 #ifdef CONFIG_X86_64
4623 *base3 = var.base >> 32;
4625 desc->type = var.type;
4627 desc->dpl = var.dpl;
4628 desc->p = var.present;
4629 desc->avl = var.avl;
4637 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4638 struct desc_struct *desc, u32 base3,
4641 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4642 struct kvm_segment var;
4644 var.selector = selector;
4645 var.base = get_desc_base(desc);
4646 #ifdef CONFIG_X86_64
4647 var.base |= ((u64)base3) << 32;
4649 var.limit = get_desc_limit(desc);
4651 var.limit = (var.limit << 12) | 0xfff;
4652 var.type = desc->type;
4653 var.present = desc->p;
4654 var.dpl = desc->dpl;
4659 var.avl = desc->avl;
4660 var.present = desc->p;
4661 var.unusable = !var.present;
4664 kvm_set_segment(vcpu, &var, seg);
4668 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4669 u32 msr_index, u64 *pdata)
4671 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4674 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4675 u32 msr_index, u64 data)
4677 struct msr_data msr;
4680 msr.index = msr_index;
4681 msr.host_initiated = false;
4682 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4685 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4686 u32 pmc, u64 *pdata)
4688 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4691 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4693 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4696 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4699 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4701 * CR0.TS may reference the host fpu state, not the guest fpu state,
4702 * so it may be clear at this point.
4707 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4712 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4713 struct x86_instruction_info *info,
4714 enum x86_intercept_stage stage)
4716 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4719 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4720 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4722 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4725 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4727 return kvm_register_read(emul_to_vcpu(ctxt), reg);
4730 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4732 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4735 static const struct x86_emulate_ops emulate_ops = {
4736 .read_gpr = emulator_read_gpr,
4737 .write_gpr = emulator_write_gpr,
4738 .read_std = kvm_read_guest_virt_system,
4739 .write_std = kvm_write_guest_virt_system,
4740 .fetch = kvm_fetch_guest_virt,
4741 .read_emulated = emulator_read_emulated,
4742 .write_emulated = emulator_write_emulated,
4743 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4744 .invlpg = emulator_invlpg,
4745 .pio_in_emulated = emulator_pio_in_emulated,
4746 .pio_out_emulated = emulator_pio_out_emulated,
4747 .get_segment = emulator_get_segment,
4748 .set_segment = emulator_set_segment,
4749 .get_cached_segment_base = emulator_get_cached_segment_base,
4750 .get_gdt = emulator_get_gdt,
4751 .get_idt = emulator_get_idt,
4752 .set_gdt = emulator_set_gdt,
4753 .set_idt = emulator_set_idt,
4754 .get_cr = emulator_get_cr,
4755 .set_cr = emulator_set_cr,
4756 .set_rflags = emulator_set_rflags,
4757 .cpl = emulator_get_cpl,
4758 .get_dr = emulator_get_dr,
4759 .set_dr = emulator_set_dr,
4760 .set_msr = emulator_set_msr,
4761 .get_msr = emulator_get_msr,
4762 .read_pmc = emulator_read_pmc,
4763 .halt = emulator_halt,
4764 .wbinvd = emulator_wbinvd,
4765 .fix_hypercall = emulator_fix_hypercall,
4766 .get_fpu = emulator_get_fpu,
4767 .put_fpu = emulator_put_fpu,
4768 .intercept = emulator_intercept,
4769 .get_cpuid = emulator_get_cpuid,
4772 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4774 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4776 * an sti; sti; sequence only disable interrupts for the first
4777 * instruction. So, if the last instruction, be it emulated or
4778 * not, left the system with the INT_STI flag enabled, it
4779 * means that the last instruction is an sti. We should not
4780 * leave the flag on in this case. The same goes for mov ss
4782 if (!(int_shadow & mask))
4783 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4786 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4788 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4789 if (ctxt->exception.vector == PF_VECTOR)
4790 kvm_propagate_fault(vcpu, &ctxt->exception);
4791 else if (ctxt->exception.error_code_valid)
4792 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4793 ctxt->exception.error_code);
4795 kvm_queue_exception(vcpu, ctxt->exception.vector);
4798 static void init_decode_cache(struct x86_emulate_ctxt *ctxt)
4800 memset(&ctxt->opcode_len, 0,
4801 (void *)&ctxt->_regs - (void *)&ctxt->opcode_len);
4803 ctxt->fetch.start = 0;
4804 ctxt->fetch.end = 0;
4805 ctxt->io_read.pos = 0;
4806 ctxt->io_read.end = 0;
4807 ctxt->mem_read.pos = 0;
4808 ctxt->mem_read.end = 0;
4811 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4813 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4816 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4818 ctxt->eflags = kvm_get_rflags(vcpu);
4819 ctxt->eip = kvm_rip_read(vcpu);
4820 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4821 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4822 cs_l ? X86EMUL_MODE_PROT64 :
4823 cs_db ? X86EMUL_MODE_PROT32 :
4824 X86EMUL_MODE_PROT16;
4825 ctxt->guest_mode = is_guest_mode(vcpu);
4827 init_decode_cache(ctxt);
4828 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4831 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4833 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4836 init_emulate_ctxt(vcpu);
4840 ctxt->_eip = ctxt->eip + inc_eip;
4841 ret = emulate_int_real(ctxt, irq);
4843 if (ret != X86EMUL_CONTINUE)
4844 return EMULATE_FAIL;
4846 ctxt->eip = ctxt->_eip;
4847 kvm_rip_write(vcpu, ctxt->eip);
4848 kvm_set_rflags(vcpu, ctxt->eflags);
4850 if (irq == NMI_VECTOR)
4851 vcpu->arch.nmi_pending = 0;
4853 vcpu->arch.interrupt.pending = false;
4855 return EMULATE_DONE;
4857 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4859 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4861 int r = EMULATE_DONE;
4863 ++vcpu->stat.insn_emulation_fail;
4864 trace_kvm_emulate_insn_failed(vcpu);
4865 if (!is_guest_mode(vcpu)) {
4866 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4867 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4868 vcpu->run->internal.ndata = 0;
4871 kvm_queue_exception(vcpu, UD_VECTOR);
4876 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
4877 bool write_fault_to_shadow_pgtable,
4883 if (emulation_type & EMULTYPE_NO_REEXECUTE)
4886 if (!vcpu->arch.mmu.direct_map) {
4888 * Write permission should be allowed since only
4889 * write access need to be emulated.
4891 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4894 * If the mapping is invalid in guest, let cpu retry
4895 * it to generate fault.
4897 if (gpa == UNMAPPED_GVA)
4902 * Do not retry the unhandleable instruction if it faults on the
4903 * readonly host memory, otherwise it will goto a infinite loop:
4904 * retry instruction -> write #PF -> emulation fail -> retry
4905 * instruction -> ...
4907 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
4910 * If the instruction failed on the error pfn, it can not be fixed,
4911 * report the error to userspace.
4913 if (is_error_noslot_pfn(pfn))
4916 kvm_release_pfn_clean(pfn);
4918 /* The instructions are well-emulated on direct mmu. */
4919 if (vcpu->arch.mmu.direct_map) {
4920 unsigned int indirect_shadow_pages;
4922 spin_lock(&vcpu->kvm->mmu_lock);
4923 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
4924 spin_unlock(&vcpu->kvm->mmu_lock);
4926 if (indirect_shadow_pages)
4927 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4933 * if emulation was due to access to shadowed page table
4934 * and it failed try to unshadow page and re-enter the
4935 * guest to let CPU execute the instruction.
4937 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4940 * If the access faults on its page table, it can not
4941 * be fixed by unprotecting shadow page and it should
4942 * be reported to userspace.
4944 return !write_fault_to_shadow_pgtable;
4947 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
4948 unsigned long cr2, int emulation_type)
4950 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4951 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
4953 last_retry_eip = vcpu->arch.last_retry_eip;
4954 last_retry_addr = vcpu->arch.last_retry_addr;
4957 * If the emulation is caused by #PF and it is non-page_table
4958 * writing instruction, it means the VM-EXIT is caused by shadow
4959 * page protected, we can zap the shadow page and retry this
4960 * instruction directly.
4962 * Note: if the guest uses a non-page-table modifying instruction
4963 * on the PDE that points to the instruction, then we will unmap
4964 * the instruction and go to an infinite loop. So, we cache the
4965 * last retried eip and the last fault address, if we meet the eip
4966 * and the address again, we can break out of the potential infinite
4969 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
4971 if (!(emulation_type & EMULTYPE_RETRY))
4974 if (x86_page_table_writing_insn(ctxt))
4977 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
4980 vcpu->arch.last_retry_eip = ctxt->eip;
4981 vcpu->arch.last_retry_addr = cr2;
4983 if (!vcpu->arch.mmu.direct_map)
4984 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4986 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4991 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
4992 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
4994 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5003 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5004 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5009 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, int *r)
5011 struct kvm_run *kvm_run = vcpu->run;
5014 * Use the "raw" value to see if TF was passed to the processor.
5015 * Note that the new value of the flags has not been saved yet.
5017 * This is correct even for TF set by the guest, because "the
5018 * processor will not generate this exception after the instruction
5019 * that sets the TF flag".
5021 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5023 if (unlikely(rflags & X86_EFLAGS_TF)) {
5024 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5025 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1;
5026 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5027 kvm_run->debug.arch.exception = DB_VECTOR;
5028 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5029 *r = EMULATE_USER_EXIT;
5031 vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5033 * "Certain debug exceptions may clear bit 0-3. The
5034 * remaining contents of the DR6 register are never
5035 * cleared by the processor".
5037 vcpu->arch.dr6 &= ~15;
5038 vcpu->arch.dr6 |= DR6_BS;
5039 kvm_queue_exception(vcpu, DB_VECTOR);
5044 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5046 struct kvm_run *kvm_run = vcpu->run;
5047 unsigned long eip = vcpu->arch.emulate_ctxt.eip;
5050 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5051 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5052 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5053 vcpu->arch.guest_debug_dr7,
5057 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
5058 kvm_run->debug.arch.pc = kvm_rip_read(vcpu) +
5059 get_segment_base(vcpu, VCPU_SREG_CS);
5061 kvm_run->debug.arch.exception = DB_VECTOR;
5062 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5063 *r = EMULATE_USER_EXIT;
5068 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK)) {
5069 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5074 vcpu->arch.dr6 &= ~15;
5075 vcpu->arch.dr6 |= dr6;
5076 kvm_queue_exception(vcpu, DB_VECTOR);
5085 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5092 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5093 bool writeback = true;
5094 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5097 * Clear write_fault_to_shadow_pgtable here to ensure it is
5100 vcpu->arch.write_fault_to_shadow_pgtable = false;
5101 kvm_clear_exception_queue(vcpu);
5103 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5104 init_emulate_ctxt(vcpu);
5107 * We will reenter on the same instruction since
5108 * we do not set complete_userspace_io. This does not
5109 * handle watchpoints yet, those would be handled in
5112 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5115 ctxt->interruptibility = 0;
5116 ctxt->have_exception = false;
5117 ctxt->perm_ok = false;
5119 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5121 r = x86_decode_insn(ctxt, insn, insn_len);
5123 trace_kvm_emulate_insn_start(vcpu);
5124 ++vcpu->stat.insn_emulation;
5125 if (r != EMULATION_OK) {
5126 if (emulation_type & EMULTYPE_TRAP_UD)
5127 return EMULATE_FAIL;
5128 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5130 return EMULATE_DONE;
5131 if (emulation_type & EMULTYPE_SKIP)
5132 return EMULATE_FAIL;
5133 return handle_emulation_failure(vcpu);
5137 if (emulation_type & EMULTYPE_SKIP) {
5138 kvm_rip_write(vcpu, ctxt->_eip);
5139 return EMULATE_DONE;
5142 if (retry_instruction(ctxt, cr2, emulation_type))
5143 return EMULATE_DONE;
5145 /* this is needed for vmware backdoor interface to work since it
5146 changes registers values during IO operation */
5147 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5148 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5149 emulator_invalidate_register_cache(ctxt);
5153 r = x86_emulate_insn(ctxt);
5155 if (r == EMULATION_INTERCEPTED)
5156 return EMULATE_DONE;
5158 if (r == EMULATION_FAILED) {
5159 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5161 return EMULATE_DONE;
5163 return handle_emulation_failure(vcpu);
5166 if (ctxt->have_exception) {
5167 inject_emulated_exception(vcpu);
5169 } else if (vcpu->arch.pio.count) {
5170 if (!vcpu->arch.pio.in) {
5171 /* FIXME: return into emulator if single-stepping. */
5172 vcpu->arch.pio.count = 0;
5175 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5177 r = EMULATE_USER_EXIT;
5178 } else if (vcpu->mmio_needed) {
5179 if (!vcpu->mmio_is_write)
5181 r = EMULATE_USER_EXIT;
5182 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5183 } else if (r == EMULATION_RESTART)
5189 toggle_interruptibility(vcpu, ctxt->interruptibility);
5190 kvm_make_request(KVM_REQ_EVENT, vcpu);
5191 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5192 kvm_rip_write(vcpu, ctxt->eip);
5193 if (r == EMULATE_DONE)
5194 kvm_vcpu_check_singlestep(vcpu, &r);
5195 kvm_set_rflags(vcpu, ctxt->eflags);
5197 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5201 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5203 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5205 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5206 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5207 size, port, &val, 1);
5208 /* do not return to emulator after return from userspace */
5209 vcpu->arch.pio.count = 0;
5212 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5214 static void tsc_bad(void *info)
5216 __this_cpu_write(cpu_tsc_khz, 0);
5219 static void tsc_khz_changed(void *data)
5221 struct cpufreq_freqs *freq = data;
5222 unsigned long khz = 0;
5226 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5227 khz = cpufreq_quick_get(raw_smp_processor_id());
5230 __this_cpu_write(cpu_tsc_khz, khz);
5233 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5236 struct cpufreq_freqs *freq = data;
5238 struct kvm_vcpu *vcpu;
5239 int i, send_ipi = 0;
5242 * We allow guests to temporarily run on slowing clocks,
5243 * provided we notify them after, or to run on accelerating
5244 * clocks, provided we notify them before. Thus time never
5247 * However, we have a problem. We can't atomically update
5248 * the frequency of a given CPU from this function; it is
5249 * merely a notifier, which can be called from any CPU.
5250 * Changing the TSC frequency at arbitrary points in time
5251 * requires a recomputation of local variables related to
5252 * the TSC for each VCPU. We must flag these local variables
5253 * to be updated and be sure the update takes place with the
5254 * new frequency before any guests proceed.
5256 * Unfortunately, the combination of hotplug CPU and frequency
5257 * change creates an intractable locking scenario; the order
5258 * of when these callouts happen is undefined with respect to
5259 * CPU hotplug, and they can race with each other. As such,
5260 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5261 * undefined; you can actually have a CPU frequency change take
5262 * place in between the computation of X and the setting of the
5263 * variable. To protect against this problem, all updates of
5264 * the per_cpu tsc_khz variable are done in an interrupt
5265 * protected IPI, and all callers wishing to update the value
5266 * must wait for a synchronous IPI to complete (which is trivial
5267 * if the caller is on the CPU already). This establishes the
5268 * necessary total order on variable updates.
5270 * Note that because a guest time update may take place
5271 * anytime after the setting of the VCPU's request bit, the
5272 * correct TSC value must be set before the request. However,
5273 * to ensure the update actually makes it to any guest which
5274 * starts running in hardware virtualization between the set
5275 * and the acquisition of the spinlock, we must also ping the
5276 * CPU after setting the request bit.
5280 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5282 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5285 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5287 spin_lock(&kvm_lock);
5288 list_for_each_entry(kvm, &vm_list, vm_list) {
5289 kvm_for_each_vcpu(i, vcpu, kvm) {
5290 if (vcpu->cpu != freq->cpu)
5292 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5293 if (vcpu->cpu != smp_processor_id())
5297 spin_unlock(&kvm_lock);
5299 if (freq->old < freq->new && send_ipi) {
5301 * We upscale the frequency. Must make the guest
5302 * doesn't see old kvmclock values while running with
5303 * the new frequency, otherwise we risk the guest sees
5304 * time go backwards.
5306 * In case we update the frequency for another cpu
5307 * (which might be in guest context) send an interrupt
5308 * to kick the cpu out of guest context. Next time
5309 * guest context is entered kvmclock will be updated,
5310 * so the guest will not see stale values.
5312 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5317 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5318 .notifier_call = kvmclock_cpufreq_notifier
5321 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5322 unsigned long action, void *hcpu)
5324 unsigned int cpu = (unsigned long)hcpu;
5328 case CPU_DOWN_FAILED:
5329 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5331 case CPU_DOWN_PREPARE:
5332 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5338 static struct notifier_block kvmclock_cpu_notifier_block = {
5339 .notifier_call = kvmclock_cpu_notifier,
5340 .priority = -INT_MAX
5343 static void kvm_timer_init(void)
5347 max_tsc_khz = tsc_khz;
5348 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5349 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5350 #ifdef CONFIG_CPU_FREQ
5351 struct cpufreq_policy policy;
5352 memset(&policy, 0, sizeof(policy));
5354 cpufreq_get_policy(&policy, cpu);
5355 if (policy.cpuinfo.max_freq)
5356 max_tsc_khz = policy.cpuinfo.max_freq;
5359 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5360 CPUFREQ_TRANSITION_NOTIFIER);
5362 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5363 for_each_online_cpu(cpu)
5364 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5367 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5369 int kvm_is_in_guest(void)
5371 return __this_cpu_read(current_vcpu) != NULL;
5374 static int kvm_is_user_mode(void)
5378 if (__this_cpu_read(current_vcpu))
5379 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5381 return user_mode != 0;
5384 static unsigned long kvm_get_guest_ip(void)
5386 unsigned long ip = 0;
5388 if (__this_cpu_read(current_vcpu))
5389 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5394 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5395 .is_in_guest = kvm_is_in_guest,
5396 .is_user_mode = kvm_is_user_mode,
5397 .get_guest_ip = kvm_get_guest_ip,
5400 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5402 __this_cpu_write(current_vcpu, vcpu);
5404 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5406 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5408 __this_cpu_write(current_vcpu, NULL);
5410 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5412 static void kvm_set_mmio_spte_mask(void)
5415 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5418 * Set the reserved bits and the present bit of an paging-structure
5419 * entry to generate page fault with PFER.RSV = 1.
5421 /* Mask the reserved physical address bits. */
5422 mask = ((1ull << (51 - maxphyaddr + 1)) - 1) << maxphyaddr;
5424 /* Bit 62 is always reserved for 32bit host. */
5425 mask |= 0x3ull << 62;
5427 /* Set the present bit. */
5430 #ifdef CONFIG_X86_64
5432 * If reserved bit is not supported, clear the present bit to disable
5435 if (maxphyaddr == 52)
5439 kvm_mmu_set_mmio_spte_mask(mask);
5442 #ifdef CONFIG_X86_64
5443 static void pvclock_gtod_update_fn(struct work_struct *work)
5447 struct kvm_vcpu *vcpu;
5450 spin_lock(&kvm_lock);
5451 list_for_each_entry(kvm, &vm_list, vm_list)
5452 kvm_for_each_vcpu(i, vcpu, kvm)
5453 set_bit(KVM_REQ_MASTERCLOCK_UPDATE, &vcpu->requests);
5454 atomic_set(&kvm_guest_has_master_clock, 0);
5455 spin_unlock(&kvm_lock);
5458 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5461 * Notification about pvclock gtod data update.
5463 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5466 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5467 struct timekeeper *tk = priv;
5469 update_pvclock_gtod(tk);
5471 /* disable master clock if host does not trust, or does not
5472 * use, TSC clocksource
5474 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5475 atomic_read(&kvm_guest_has_master_clock) != 0)
5476 queue_work(system_long_wq, &pvclock_gtod_work);
5481 static struct notifier_block pvclock_gtod_notifier = {
5482 .notifier_call = pvclock_gtod_notify,
5486 int kvm_arch_init(void *opaque)
5489 struct kvm_x86_ops *ops = opaque;
5492 printk(KERN_ERR "kvm: already loaded the other module\n");
5497 if (!ops->cpu_has_kvm_support()) {
5498 printk(KERN_ERR "kvm: no hardware support\n");
5502 if (ops->disabled_by_bios()) {
5503 printk(KERN_ERR "kvm: disabled by bios\n");
5509 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5511 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5515 r = kvm_mmu_module_init();
5517 goto out_free_percpu;
5519 kvm_set_mmio_spte_mask();
5520 kvm_init_msr_list();
5523 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5524 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5528 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5531 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5534 #ifdef CONFIG_X86_64
5535 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5541 free_percpu(shared_msrs);
5546 void kvm_arch_exit(void)
5548 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5550 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5551 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5552 CPUFREQ_TRANSITION_NOTIFIER);
5553 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5554 #ifdef CONFIG_X86_64
5555 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5558 kvm_mmu_module_exit();
5559 free_percpu(shared_msrs);
5562 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5564 ++vcpu->stat.halt_exits;
5565 if (irqchip_in_kernel(vcpu->kvm)) {
5566 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5569 vcpu->run->exit_reason = KVM_EXIT_HLT;
5573 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5575 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5577 u64 param, ingpa, outgpa, ret;
5578 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5579 bool fast, longmode;
5583 * hypercall generates UD from non zero cpl and real mode
5586 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5587 kvm_queue_exception(vcpu, UD_VECTOR);
5591 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5592 longmode = is_long_mode(vcpu) && cs_l == 1;
5595 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5596 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5597 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5598 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5599 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5600 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5602 #ifdef CONFIG_X86_64
5604 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5605 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5606 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5610 code = param & 0xffff;
5611 fast = (param >> 16) & 0x1;
5612 rep_cnt = (param >> 32) & 0xfff;
5613 rep_idx = (param >> 48) & 0xfff;
5615 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5618 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5619 kvm_vcpu_on_spin(vcpu);
5622 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5626 ret = res | (((u64)rep_done & 0xfff) << 32);
5628 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5630 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5631 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5638 * kvm_pv_kick_cpu_op: Kick a vcpu.
5640 * @apicid - apicid of vcpu to be kicked.
5642 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5644 struct kvm_lapic_irq lapic_irq;
5646 lapic_irq.shorthand = 0;
5647 lapic_irq.dest_mode = 0;
5648 lapic_irq.dest_id = apicid;
5650 lapic_irq.delivery_mode = APIC_DM_REMRD;
5651 kvm_irq_delivery_to_apic(kvm, 0, &lapic_irq, NULL);
5654 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5656 unsigned long nr, a0, a1, a2, a3, ret;
5659 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5660 return kvm_hv_hypercall(vcpu);
5662 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5663 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5664 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5665 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5666 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5668 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5670 if (!is_long_mode(vcpu)) {
5678 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5684 case KVM_HC_VAPIC_POLL_IRQ:
5687 case KVM_HC_KICK_CPU:
5688 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5696 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5697 ++vcpu->stat.hypercalls;
5700 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5702 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5704 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5705 char instruction[3];
5706 unsigned long rip = kvm_rip_read(vcpu);
5708 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5710 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5714 * Check if userspace requested an interrupt window, and that the
5715 * interrupt window is open.
5717 * No need to exit to userspace if we already have an interrupt queued.
5719 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5721 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5722 vcpu->run->request_interrupt_window &&
5723 kvm_arch_interrupt_allowed(vcpu));
5726 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5728 struct kvm_run *kvm_run = vcpu->run;
5730 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5731 kvm_run->cr8 = kvm_get_cr8(vcpu);
5732 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5733 if (irqchip_in_kernel(vcpu->kvm))
5734 kvm_run->ready_for_interrupt_injection = 1;
5736 kvm_run->ready_for_interrupt_injection =
5737 kvm_arch_interrupt_allowed(vcpu) &&
5738 !kvm_cpu_has_interrupt(vcpu) &&
5739 !kvm_event_needs_reinjection(vcpu);
5742 static int vapic_enter(struct kvm_vcpu *vcpu)
5744 struct kvm_lapic *apic = vcpu->arch.apic;
5747 if (!apic || !apic->vapic_addr)
5750 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5751 if (is_error_page(page))
5754 vcpu->arch.apic->vapic_page = page;
5758 static void vapic_exit(struct kvm_vcpu *vcpu)
5760 struct kvm_lapic *apic = vcpu->arch.apic;
5763 if (!apic || !apic->vapic_addr)
5766 idx = srcu_read_lock(&vcpu->kvm->srcu);
5767 kvm_release_page_dirty(apic->vapic_page);
5768 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5769 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5772 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5776 if (!kvm_x86_ops->update_cr8_intercept)
5779 if (!vcpu->arch.apic)
5782 if (!vcpu->arch.apic->vapic_addr)
5783 max_irr = kvm_lapic_find_highest_irr(vcpu);
5790 tpr = kvm_lapic_get_cr8(vcpu);
5792 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5795 static void inject_pending_event(struct kvm_vcpu *vcpu)
5797 /* try to reinject previous events if any */
5798 if (vcpu->arch.exception.pending) {
5799 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5800 vcpu->arch.exception.has_error_code,
5801 vcpu->arch.exception.error_code);
5802 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5803 vcpu->arch.exception.has_error_code,
5804 vcpu->arch.exception.error_code,
5805 vcpu->arch.exception.reinject);
5809 if (vcpu->arch.nmi_injected) {
5810 kvm_x86_ops->set_nmi(vcpu);
5814 if (vcpu->arch.interrupt.pending) {
5815 kvm_x86_ops->set_irq(vcpu);
5819 /* try to inject new event if pending */
5820 if (vcpu->arch.nmi_pending) {
5821 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5822 --vcpu->arch.nmi_pending;
5823 vcpu->arch.nmi_injected = true;
5824 kvm_x86_ops->set_nmi(vcpu);
5826 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
5827 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5828 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5830 kvm_x86_ops->set_irq(vcpu);
5835 static void process_nmi(struct kvm_vcpu *vcpu)
5840 * x86 is limited to one NMI running, and one NMI pending after it.
5841 * If an NMI is already in progress, limit further NMIs to just one.
5842 * Otherwise, allow two (and we'll inject the first one immediately).
5844 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5847 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5848 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5849 kvm_make_request(KVM_REQ_EVENT, vcpu);
5852 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
5854 u64 eoi_exit_bitmap[4];
5857 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
5860 memset(eoi_exit_bitmap, 0, 32);
5863 kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
5864 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
5865 kvm_apic_update_tmr(vcpu, tmr);
5868 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5871 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5872 vcpu->run->request_interrupt_window;
5873 bool req_immediate_exit = false;
5875 if (vcpu->requests) {
5876 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5877 kvm_mmu_unload(vcpu);
5878 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5879 __kvm_migrate_timers(vcpu);
5880 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
5881 kvm_gen_update_masterclock(vcpu->kvm);
5882 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
5883 kvm_gen_kvmclock_update(vcpu);
5884 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5885 r = kvm_guest_time_update(vcpu);
5889 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5890 kvm_mmu_sync_roots(vcpu);
5891 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5892 kvm_x86_ops->tlb_flush(vcpu);
5893 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5894 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5898 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5899 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5903 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5904 vcpu->fpu_active = 0;
5905 kvm_x86_ops->fpu_deactivate(vcpu);
5907 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5908 /* Page is swapped out. Do synthetic halt */
5909 vcpu->arch.apf.halted = true;
5913 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5914 record_steal_time(vcpu);
5915 if (kvm_check_request(KVM_REQ_NMI, vcpu))
5917 if (kvm_check_request(KVM_REQ_PMU, vcpu))
5918 kvm_handle_pmu_event(vcpu);
5919 if (kvm_check_request(KVM_REQ_PMI, vcpu))
5920 kvm_deliver_pmi(vcpu);
5921 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
5922 vcpu_scan_ioapic(vcpu);
5925 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5926 kvm_apic_accept_events(vcpu);
5927 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
5932 inject_pending_event(vcpu);
5934 /* enable NMI/IRQ window open exits if needed */
5935 if (vcpu->arch.nmi_pending)
5936 req_immediate_exit =
5937 kvm_x86_ops->enable_nmi_window(vcpu) != 0;
5938 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
5939 req_immediate_exit =
5940 kvm_x86_ops->enable_irq_window(vcpu) != 0;
5942 if (kvm_lapic_enabled(vcpu)) {
5944 * Update architecture specific hints for APIC
5945 * virtual interrupt delivery.
5947 if (kvm_x86_ops->hwapic_irr_update)
5948 kvm_x86_ops->hwapic_irr_update(vcpu,
5949 kvm_lapic_find_highest_irr(vcpu));
5950 update_cr8_intercept(vcpu);
5951 kvm_lapic_sync_to_vapic(vcpu);
5955 r = kvm_mmu_reload(vcpu);
5957 goto cancel_injection;
5962 kvm_x86_ops->prepare_guest_switch(vcpu);
5963 if (vcpu->fpu_active)
5964 kvm_load_guest_fpu(vcpu);
5965 kvm_load_guest_xcr0(vcpu);
5967 vcpu->mode = IN_GUEST_MODE;
5969 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5971 /* We should set ->mode before check ->requests,
5972 * see the comment in make_all_cpus_request.
5974 smp_mb__after_srcu_read_unlock();
5976 local_irq_disable();
5978 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5979 || need_resched() || signal_pending(current)) {
5980 vcpu->mode = OUTSIDE_GUEST_MODE;
5984 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5986 goto cancel_injection;
5989 if (req_immediate_exit)
5990 smp_send_reschedule(vcpu->cpu);
5994 if (unlikely(vcpu->arch.switch_db_regs)) {
5996 set_debugreg(vcpu->arch.eff_db[0], 0);
5997 set_debugreg(vcpu->arch.eff_db[1], 1);
5998 set_debugreg(vcpu->arch.eff_db[2], 2);
5999 set_debugreg(vcpu->arch.eff_db[3], 3);
6002 trace_kvm_entry(vcpu->vcpu_id);
6003 kvm_x86_ops->run(vcpu);
6006 * If the guest has used debug registers, at least dr7
6007 * will be disabled while returning to the host.
6008 * If we don't have active breakpoints in the host, we don't
6009 * care about the messed up debug address registers. But if
6010 * we have some of them active, restore the old state.
6012 if (hw_breakpoint_active())
6013 hw_breakpoint_restore();
6015 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
6018 vcpu->mode = OUTSIDE_GUEST_MODE;
6021 /* Interrupt is enabled by handle_external_intr() */
6022 kvm_x86_ops->handle_external_intr(vcpu);
6027 * We must have an instruction between local_irq_enable() and
6028 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6029 * the interrupt shadow. The stat.exits increment will do nicely.
6030 * But we need to prevent reordering, hence this barrier():
6038 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6041 * Profile KVM exit RIPs:
6043 if (unlikely(prof_on == KVM_PROFILING)) {
6044 unsigned long rip = kvm_rip_read(vcpu);
6045 profile_hit(KVM_PROFILING, (void *)rip);
6048 if (unlikely(vcpu->arch.tsc_always_catchup))
6049 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6051 if (vcpu->arch.apic_attention)
6052 kvm_lapic_sync_from_vapic(vcpu);
6054 r = kvm_x86_ops->handle_exit(vcpu);
6058 kvm_x86_ops->cancel_injection(vcpu);
6059 if (unlikely(vcpu->arch.apic_attention))
6060 kvm_lapic_sync_from_vapic(vcpu);
6066 static int __vcpu_run(struct kvm_vcpu *vcpu)
6069 struct kvm *kvm = vcpu->kvm;
6071 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6072 r = vapic_enter(vcpu);
6074 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6080 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6081 !vcpu->arch.apf.halted)
6082 r = vcpu_enter_guest(vcpu);
6084 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6085 kvm_vcpu_block(vcpu);
6086 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6087 if (kvm_check_request(KVM_REQ_UNHALT, vcpu)) {
6088 kvm_apic_accept_events(vcpu);
6089 switch(vcpu->arch.mp_state) {
6090 case KVM_MP_STATE_HALTED:
6091 vcpu->arch.pv.pv_unhalted = false;
6092 vcpu->arch.mp_state =
6093 KVM_MP_STATE_RUNNABLE;
6094 case KVM_MP_STATE_RUNNABLE:
6095 vcpu->arch.apf.halted = false;
6097 case KVM_MP_STATE_INIT_RECEIVED:
6109 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6110 if (kvm_cpu_has_pending_timer(vcpu))
6111 kvm_inject_pending_timer_irqs(vcpu);
6113 if (dm_request_for_irq_injection(vcpu)) {
6115 vcpu->run->exit_reason = KVM_EXIT_INTR;
6116 ++vcpu->stat.request_irq_exits;
6119 kvm_check_async_pf_completion(vcpu);
6121 if (signal_pending(current)) {
6123 vcpu->run->exit_reason = KVM_EXIT_INTR;
6124 ++vcpu->stat.signal_exits;
6126 if (need_resched()) {
6127 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6129 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6133 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6140 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6143 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6144 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6145 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6146 if (r != EMULATE_DONE)
6151 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6153 BUG_ON(!vcpu->arch.pio.count);
6155 return complete_emulated_io(vcpu);
6159 * Implements the following, as a state machine:
6163 * for each mmio piece in the fragment
6171 * for each mmio piece in the fragment
6176 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6178 struct kvm_run *run = vcpu->run;
6179 struct kvm_mmio_fragment *frag;
6182 BUG_ON(!vcpu->mmio_needed);
6184 /* Complete previous fragment */
6185 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6186 len = min(8u, frag->len);
6187 if (!vcpu->mmio_is_write)
6188 memcpy(frag->data, run->mmio.data, len);
6190 if (frag->len <= 8) {
6191 /* Switch to the next fragment. */
6193 vcpu->mmio_cur_fragment++;
6195 /* Go forward to the next mmio piece. */
6201 if (vcpu->mmio_cur_fragment == vcpu->mmio_nr_fragments) {
6202 vcpu->mmio_needed = 0;
6204 /* FIXME: return into emulator if single-stepping. */
6205 if (vcpu->mmio_is_write)
6207 vcpu->mmio_read_completed = 1;
6208 return complete_emulated_io(vcpu);
6211 run->exit_reason = KVM_EXIT_MMIO;
6212 run->mmio.phys_addr = frag->gpa;
6213 if (vcpu->mmio_is_write)
6214 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6215 run->mmio.len = min(8u, frag->len);
6216 run->mmio.is_write = vcpu->mmio_is_write;
6217 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6222 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6227 if (!tsk_used_math(current) && init_fpu(current))
6230 if (vcpu->sigset_active)
6231 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6233 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6234 kvm_vcpu_block(vcpu);
6235 kvm_apic_accept_events(vcpu);
6236 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6241 /* re-sync apic's tpr */
6242 if (!irqchip_in_kernel(vcpu->kvm)) {
6243 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6249 if (unlikely(vcpu->arch.complete_userspace_io)) {
6250 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6251 vcpu->arch.complete_userspace_io = NULL;
6256 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6258 r = __vcpu_run(vcpu);
6261 post_kvm_run_save(vcpu);
6262 if (vcpu->sigset_active)
6263 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6268 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6270 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6272 * We are here if userspace calls get_regs() in the middle of
6273 * instruction emulation. Registers state needs to be copied
6274 * back from emulation context to vcpu. Userspace shouldn't do
6275 * that usually, but some bad designed PV devices (vmware
6276 * backdoor interface) need this to work
6278 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6279 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6281 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6282 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6283 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6284 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6285 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6286 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6287 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6288 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6289 #ifdef CONFIG_X86_64
6290 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6291 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6292 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6293 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6294 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6295 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6296 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6297 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6300 regs->rip = kvm_rip_read(vcpu);
6301 regs->rflags = kvm_get_rflags(vcpu);
6306 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6308 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6309 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6311 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6312 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6313 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6314 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6315 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6316 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6317 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6318 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6319 #ifdef CONFIG_X86_64
6320 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6321 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6322 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6323 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6324 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6325 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6326 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6327 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6330 kvm_rip_write(vcpu, regs->rip);
6331 kvm_set_rflags(vcpu, regs->rflags);
6333 vcpu->arch.exception.pending = false;
6335 kvm_make_request(KVM_REQ_EVENT, vcpu);
6340 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6342 struct kvm_segment cs;
6344 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6348 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6350 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6351 struct kvm_sregs *sregs)
6355 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6356 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6357 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6358 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6359 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6360 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6362 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6363 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6365 kvm_x86_ops->get_idt(vcpu, &dt);
6366 sregs->idt.limit = dt.size;
6367 sregs->idt.base = dt.address;
6368 kvm_x86_ops->get_gdt(vcpu, &dt);
6369 sregs->gdt.limit = dt.size;
6370 sregs->gdt.base = dt.address;
6372 sregs->cr0 = kvm_read_cr0(vcpu);
6373 sregs->cr2 = vcpu->arch.cr2;
6374 sregs->cr3 = kvm_read_cr3(vcpu);
6375 sregs->cr4 = kvm_read_cr4(vcpu);
6376 sregs->cr8 = kvm_get_cr8(vcpu);
6377 sregs->efer = vcpu->arch.efer;
6378 sregs->apic_base = kvm_get_apic_base(vcpu);
6380 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6382 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6383 set_bit(vcpu->arch.interrupt.nr,
6384 (unsigned long *)sregs->interrupt_bitmap);
6389 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6390 struct kvm_mp_state *mp_state)
6392 kvm_apic_accept_events(vcpu);
6393 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
6394 vcpu->arch.pv.pv_unhalted)
6395 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
6397 mp_state->mp_state = vcpu->arch.mp_state;
6402 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6403 struct kvm_mp_state *mp_state)
6405 if (!kvm_vcpu_has_lapic(vcpu) &&
6406 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6409 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6410 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6411 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6413 vcpu->arch.mp_state = mp_state->mp_state;
6414 kvm_make_request(KVM_REQ_EVENT, vcpu);
6418 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6419 int reason, bool has_error_code, u32 error_code)
6421 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6424 init_emulate_ctxt(vcpu);
6426 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6427 has_error_code, error_code);
6430 return EMULATE_FAIL;
6432 kvm_rip_write(vcpu, ctxt->eip);
6433 kvm_set_rflags(vcpu, ctxt->eflags);
6434 kvm_make_request(KVM_REQ_EVENT, vcpu);
6435 return EMULATE_DONE;
6437 EXPORT_SYMBOL_GPL(kvm_task_switch);
6439 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6440 struct kvm_sregs *sregs)
6442 int mmu_reset_needed = 0;
6443 int pending_vec, max_bits, idx;
6446 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6449 dt.size = sregs->idt.limit;
6450 dt.address = sregs->idt.base;
6451 kvm_x86_ops->set_idt(vcpu, &dt);
6452 dt.size = sregs->gdt.limit;
6453 dt.address = sregs->gdt.base;
6454 kvm_x86_ops->set_gdt(vcpu, &dt);
6456 vcpu->arch.cr2 = sregs->cr2;
6457 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6458 vcpu->arch.cr3 = sregs->cr3;
6459 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6461 kvm_set_cr8(vcpu, sregs->cr8);
6463 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6464 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6465 kvm_set_apic_base(vcpu, sregs->apic_base);
6467 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6468 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6469 vcpu->arch.cr0 = sregs->cr0;
6471 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6472 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6473 if (sregs->cr4 & X86_CR4_OSXSAVE)
6474 kvm_update_cpuid(vcpu);
6476 idx = srcu_read_lock(&vcpu->kvm->srcu);
6477 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6478 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6479 mmu_reset_needed = 1;
6481 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6483 if (mmu_reset_needed)
6484 kvm_mmu_reset_context(vcpu);
6486 max_bits = KVM_NR_INTERRUPTS;
6487 pending_vec = find_first_bit(
6488 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6489 if (pending_vec < max_bits) {
6490 kvm_queue_interrupt(vcpu, pending_vec, false);
6491 pr_debug("Set back pending irq %d\n", pending_vec);
6494 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6495 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6496 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6497 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6498 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6499 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6501 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6502 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6504 update_cr8_intercept(vcpu);
6506 /* Older userspace won't unhalt the vcpu on reset. */
6507 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6508 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6510 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6512 kvm_make_request(KVM_REQ_EVENT, vcpu);
6517 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6518 struct kvm_guest_debug *dbg)
6520 unsigned long rflags;
6523 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6525 if (vcpu->arch.exception.pending)
6527 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6528 kvm_queue_exception(vcpu, DB_VECTOR);
6530 kvm_queue_exception(vcpu, BP_VECTOR);
6534 * Read rflags as long as potentially injected trace flags are still
6537 rflags = kvm_get_rflags(vcpu);
6539 vcpu->guest_debug = dbg->control;
6540 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6541 vcpu->guest_debug = 0;
6543 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6544 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6545 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6546 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6548 for (i = 0; i < KVM_NR_DB_REGS; i++)
6549 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6551 kvm_update_dr7(vcpu);
6553 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6554 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6555 get_segment_base(vcpu, VCPU_SREG_CS);
6558 * Trigger an rflags update that will inject or remove the trace
6561 kvm_set_rflags(vcpu, rflags);
6563 kvm_x86_ops->update_db_bp_intercept(vcpu);
6573 * Translate a guest virtual address to a guest physical address.
6575 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6576 struct kvm_translation *tr)
6578 unsigned long vaddr = tr->linear_address;
6582 idx = srcu_read_lock(&vcpu->kvm->srcu);
6583 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6584 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6585 tr->physical_address = gpa;
6586 tr->valid = gpa != UNMAPPED_GVA;
6593 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6595 struct i387_fxsave_struct *fxsave =
6596 &vcpu->arch.guest_fpu.state->fxsave;
6598 memcpy(fpu->fpr, fxsave->st_space, 128);
6599 fpu->fcw = fxsave->cwd;
6600 fpu->fsw = fxsave->swd;
6601 fpu->ftwx = fxsave->twd;
6602 fpu->last_opcode = fxsave->fop;
6603 fpu->last_ip = fxsave->rip;
6604 fpu->last_dp = fxsave->rdp;
6605 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6610 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6612 struct i387_fxsave_struct *fxsave =
6613 &vcpu->arch.guest_fpu.state->fxsave;
6615 memcpy(fxsave->st_space, fpu->fpr, 128);
6616 fxsave->cwd = fpu->fcw;
6617 fxsave->swd = fpu->fsw;
6618 fxsave->twd = fpu->ftwx;
6619 fxsave->fop = fpu->last_opcode;
6620 fxsave->rip = fpu->last_ip;
6621 fxsave->rdp = fpu->last_dp;
6622 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6627 int fx_init(struct kvm_vcpu *vcpu)
6631 err = fpu_alloc(&vcpu->arch.guest_fpu);
6635 fpu_finit(&vcpu->arch.guest_fpu);
6638 * Ensure guest xcr0 is valid for loading
6640 vcpu->arch.xcr0 = XSTATE_FP;
6642 vcpu->arch.cr0 |= X86_CR0_ET;
6646 EXPORT_SYMBOL_GPL(fx_init);
6648 static void fx_free(struct kvm_vcpu *vcpu)
6650 fpu_free(&vcpu->arch.guest_fpu);
6653 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6655 if (vcpu->guest_fpu_loaded)
6659 * Restore all possible states in the guest,
6660 * and assume host would use all available bits.
6661 * Guest xcr0 would be loaded later.
6663 kvm_put_guest_xcr0(vcpu);
6664 vcpu->guest_fpu_loaded = 1;
6665 __kernel_fpu_begin();
6666 fpu_restore_checking(&vcpu->arch.guest_fpu);
6670 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6672 kvm_put_guest_xcr0(vcpu);
6674 if (!vcpu->guest_fpu_loaded)
6677 vcpu->guest_fpu_loaded = 0;
6678 fpu_save_init(&vcpu->arch.guest_fpu);
6680 ++vcpu->stat.fpu_reload;
6681 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6685 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6687 kvmclock_reset(vcpu);
6689 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6691 kvm_x86_ops->vcpu_free(vcpu);
6694 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6697 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6698 printk_once(KERN_WARNING
6699 "kvm: SMP vm created on host with unstable TSC; "
6700 "guest TSC will not be reliable\n");
6701 return kvm_x86_ops->vcpu_create(kvm, id);
6704 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6708 vcpu->arch.mtrr_state.have_fixed = 1;
6709 r = vcpu_load(vcpu);
6712 kvm_vcpu_reset(vcpu);
6713 kvm_mmu_setup(vcpu);
6719 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
6722 struct msr_data msr;
6724 r = vcpu_load(vcpu);
6728 msr.index = MSR_IA32_TSC;
6729 msr.host_initiated = true;
6730 kvm_write_tsc(vcpu, &msr);
6736 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6739 vcpu->arch.apf.msr_val = 0;
6741 r = vcpu_load(vcpu);
6743 kvm_mmu_unload(vcpu);
6747 kvm_x86_ops->vcpu_free(vcpu);
6750 void kvm_vcpu_reset(struct kvm_vcpu *vcpu)
6752 atomic_set(&vcpu->arch.nmi_queued, 0);
6753 vcpu->arch.nmi_pending = 0;
6754 vcpu->arch.nmi_injected = false;
6756 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6757 vcpu->arch.dr6 = DR6_FIXED_1;
6758 vcpu->arch.dr7 = DR7_FIXED_1;
6759 kvm_update_dr7(vcpu);
6761 kvm_make_request(KVM_REQ_EVENT, vcpu);
6762 vcpu->arch.apf.msr_val = 0;
6763 vcpu->arch.st.msr_val = 0;
6765 kvmclock_reset(vcpu);
6767 kvm_clear_async_pf_completion_queue(vcpu);
6768 kvm_async_pf_hash_reset(vcpu);
6769 vcpu->arch.apf.halted = false;
6771 kvm_pmu_reset(vcpu);
6773 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
6774 vcpu->arch.regs_avail = ~0;
6775 vcpu->arch.regs_dirty = ~0;
6777 kvm_x86_ops->vcpu_reset(vcpu);
6780 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, unsigned int vector)
6782 struct kvm_segment cs;
6784 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6785 cs.selector = vector << 8;
6786 cs.base = vector << 12;
6787 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6788 kvm_rip_write(vcpu, 0);
6791 int kvm_arch_hardware_enable(void *garbage)
6794 struct kvm_vcpu *vcpu;
6799 bool stable, backwards_tsc = false;
6801 kvm_shared_msr_cpu_online();
6802 ret = kvm_x86_ops->hardware_enable(garbage);
6806 local_tsc = native_read_tsc();
6807 stable = !check_tsc_unstable();
6808 list_for_each_entry(kvm, &vm_list, vm_list) {
6809 kvm_for_each_vcpu(i, vcpu, kvm) {
6810 if (!stable && vcpu->cpu == smp_processor_id())
6811 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
6812 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
6813 backwards_tsc = true;
6814 if (vcpu->arch.last_host_tsc > max_tsc)
6815 max_tsc = vcpu->arch.last_host_tsc;
6821 * Sometimes, even reliable TSCs go backwards. This happens on
6822 * platforms that reset TSC during suspend or hibernate actions, but
6823 * maintain synchronization. We must compensate. Fortunately, we can
6824 * detect that condition here, which happens early in CPU bringup,
6825 * before any KVM threads can be running. Unfortunately, we can't
6826 * bring the TSCs fully up to date with real time, as we aren't yet far
6827 * enough into CPU bringup that we know how much real time has actually
6828 * elapsed; our helper function, get_kernel_ns() will be using boot
6829 * variables that haven't been updated yet.
6831 * So we simply find the maximum observed TSC above, then record the
6832 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
6833 * the adjustment will be applied. Note that we accumulate
6834 * adjustments, in case multiple suspend cycles happen before some VCPU
6835 * gets a chance to run again. In the event that no KVM threads get a
6836 * chance to run, we will miss the entire elapsed period, as we'll have
6837 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
6838 * loose cycle time. This isn't too big a deal, since the loss will be
6839 * uniform across all VCPUs (not to mention the scenario is extremely
6840 * unlikely). It is possible that a second hibernate recovery happens
6841 * much faster than a first, causing the observed TSC here to be
6842 * smaller; this would require additional padding adjustment, which is
6843 * why we set last_host_tsc to the local tsc observed here.
6845 * N.B. - this code below runs only on platforms with reliable TSC,
6846 * as that is the only way backwards_tsc is set above. Also note
6847 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
6848 * have the same delta_cyc adjustment applied if backwards_tsc
6849 * is detected. Note further, this adjustment is only done once,
6850 * as we reset last_host_tsc on all VCPUs to stop this from being
6851 * called multiple times (one for each physical CPU bringup).
6853 * Platforms with unreliable TSCs don't have to deal with this, they
6854 * will be compensated by the logic in vcpu_load, which sets the TSC to
6855 * catchup mode. This will catchup all VCPUs to real time, but cannot
6856 * guarantee that they stay in perfect synchronization.
6858 if (backwards_tsc) {
6859 u64 delta_cyc = max_tsc - local_tsc;
6860 list_for_each_entry(kvm, &vm_list, vm_list) {
6861 kvm_for_each_vcpu(i, vcpu, kvm) {
6862 vcpu->arch.tsc_offset_adjustment += delta_cyc;
6863 vcpu->arch.last_host_tsc = local_tsc;
6864 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
6869 * We have to disable TSC offset matching.. if you were
6870 * booting a VM while issuing an S4 host suspend....
6871 * you may have some problem. Solving this issue is
6872 * left as an exercise to the reader.
6874 kvm->arch.last_tsc_nsec = 0;
6875 kvm->arch.last_tsc_write = 0;
6882 void kvm_arch_hardware_disable(void *garbage)
6884 kvm_x86_ops->hardware_disable(garbage);
6885 drop_user_return_notifiers(garbage);
6888 int kvm_arch_hardware_setup(void)
6890 return kvm_x86_ops->hardware_setup();
6893 void kvm_arch_hardware_unsetup(void)
6895 kvm_x86_ops->hardware_unsetup();
6898 void kvm_arch_check_processor_compat(void *rtn)
6900 kvm_x86_ops->check_processor_compatibility(rtn);
6903 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
6905 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
6908 struct static_key kvm_no_apic_vcpu __read_mostly;
6910 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6916 BUG_ON(vcpu->kvm == NULL);
6919 vcpu->arch.pv.pv_unhalted = false;
6920 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6921 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6922 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6924 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6926 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6931 vcpu->arch.pio_data = page_address(page);
6933 kvm_set_tsc_khz(vcpu, max_tsc_khz);
6935 r = kvm_mmu_create(vcpu);
6937 goto fail_free_pio_data;
6939 if (irqchip_in_kernel(kvm)) {
6940 r = kvm_create_lapic(vcpu);
6942 goto fail_mmu_destroy;
6944 static_key_slow_inc(&kvm_no_apic_vcpu);
6946 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6948 if (!vcpu->arch.mce_banks) {
6950 goto fail_free_lapic;
6952 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6954 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
6956 goto fail_free_mce_banks;
6961 goto fail_free_wbinvd_dirty_mask;
6963 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
6964 vcpu->arch.pv_time_enabled = false;
6966 vcpu->arch.guest_supported_xcr0 = 0;
6967 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
6969 kvm_async_pf_hash_reset(vcpu);
6973 fail_free_wbinvd_dirty_mask:
6974 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6975 fail_free_mce_banks:
6976 kfree(vcpu->arch.mce_banks);
6978 kvm_free_lapic(vcpu);
6980 kvm_mmu_destroy(vcpu);
6982 free_page((unsigned long)vcpu->arch.pio_data);
6987 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6991 kvm_pmu_destroy(vcpu);
6992 kfree(vcpu->arch.mce_banks);
6993 kvm_free_lapic(vcpu);
6994 idx = srcu_read_lock(&vcpu->kvm->srcu);
6995 kvm_mmu_destroy(vcpu);
6996 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6997 free_page((unsigned long)vcpu->arch.pio_data);
6998 if (!irqchip_in_kernel(vcpu->kvm))
6999 static_key_slow_dec(&kvm_no_apic_vcpu);
7002 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7007 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7008 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7009 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7010 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7012 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7013 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7014 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7015 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7016 &kvm->arch.irq_sources_bitmap);
7018 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7019 mutex_init(&kvm->arch.apic_map_lock);
7020 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7022 pvclock_update_vm_gtod_copy(kvm);
7027 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7030 r = vcpu_load(vcpu);
7032 kvm_mmu_unload(vcpu);
7036 static void kvm_free_vcpus(struct kvm *kvm)
7039 struct kvm_vcpu *vcpu;
7042 * Unpin any mmu pages first.
7044 kvm_for_each_vcpu(i, vcpu, kvm) {
7045 kvm_clear_async_pf_completion_queue(vcpu);
7046 kvm_unload_vcpu_mmu(vcpu);
7048 kvm_for_each_vcpu(i, vcpu, kvm)
7049 kvm_arch_vcpu_free(vcpu);
7051 mutex_lock(&kvm->lock);
7052 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7053 kvm->vcpus[i] = NULL;
7055 atomic_set(&kvm->online_vcpus, 0);
7056 mutex_unlock(&kvm->lock);
7059 void kvm_arch_sync_events(struct kvm *kvm)
7061 kvm_free_all_assigned_devices(kvm);
7065 void kvm_arch_destroy_vm(struct kvm *kvm)
7067 if (current->mm == kvm->mm) {
7069 * Free memory regions allocated on behalf of userspace,
7070 * unless the the memory map has changed due to process exit
7073 struct kvm_userspace_memory_region mem;
7074 memset(&mem, 0, sizeof(mem));
7075 mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
7076 kvm_set_memory_region(kvm, &mem);
7078 mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
7079 kvm_set_memory_region(kvm, &mem);
7081 mem.slot = TSS_PRIVATE_MEMSLOT;
7082 kvm_set_memory_region(kvm, &mem);
7084 kvm_iommu_unmap_guest(kvm);
7085 kfree(kvm->arch.vpic);
7086 kfree(kvm->arch.vioapic);
7087 kvm_free_vcpus(kvm);
7088 if (kvm->arch.apic_access_page)
7089 put_page(kvm->arch.apic_access_page);
7090 if (kvm->arch.ept_identity_pagetable)
7091 put_page(kvm->arch.ept_identity_pagetable);
7092 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7095 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7096 struct kvm_memory_slot *dont)
7100 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7101 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7102 kvm_kvfree(free->arch.rmap[i]);
7103 free->arch.rmap[i] = NULL;
7108 if (!dont || free->arch.lpage_info[i - 1] !=
7109 dont->arch.lpage_info[i - 1]) {
7110 kvm_kvfree(free->arch.lpage_info[i - 1]);
7111 free->arch.lpage_info[i - 1] = NULL;
7116 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7117 unsigned long npages)
7121 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7126 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7127 slot->base_gfn, level) + 1;
7129 slot->arch.rmap[i] =
7130 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7131 if (!slot->arch.rmap[i])
7136 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7137 sizeof(*slot->arch.lpage_info[i - 1]));
7138 if (!slot->arch.lpage_info[i - 1])
7141 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7142 slot->arch.lpage_info[i - 1][0].write_count = 1;
7143 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7144 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7145 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7147 * If the gfn and userspace address are not aligned wrt each
7148 * other, or if explicitly asked to, disable large page
7149 * support for this slot
7151 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7152 !kvm_largepages_enabled()) {
7155 for (j = 0; j < lpages; ++j)
7156 slot->arch.lpage_info[i - 1][j].write_count = 1;
7163 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7164 kvm_kvfree(slot->arch.rmap[i]);
7165 slot->arch.rmap[i] = NULL;
7169 kvm_kvfree(slot->arch.lpage_info[i - 1]);
7170 slot->arch.lpage_info[i - 1] = NULL;
7175 void kvm_arch_memslots_updated(struct kvm *kvm)
7178 * memslots->generation has been incremented.
7179 * mmio generation may have reached its maximum value.
7181 kvm_mmu_invalidate_mmio_sptes(kvm);
7184 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7185 struct kvm_memory_slot *memslot,
7186 struct kvm_userspace_memory_region *mem,
7187 enum kvm_mr_change change)
7190 * Only private memory slots need to be mapped here since
7191 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
7193 if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
7194 unsigned long userspace_addr;
7197 * MAP_SHARED to prevent internal slot pages from being moved
7200 userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
7201 PROT_READ | PROT_WRITE,
7202 MAP_SHARED | MAP_ANONYMOUS, 0);
7204 if (IS_ERR((void *)userspace_addr))
7205 return PTR_ERR((void *)userspace_addr);
7207 memslot->userspace_addr = userspace_addr;
7213 void kvm_arch_commit_memory_region(struct kvm *kvm,
7214 struct kvm_userspace_memory_region *mem,
7215 const struct kvm_memory_slot *old,
7216 enum kvm_mr_change change)
7219 int nr_mmu_pages = 0;
7221 if ((mem->slot >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_DELETE)) {
7224 ret = vm_munmap(old->userspace_addr,
7225 old->npages * PAGE_SIZE);
7228 "kvm_vm_ioctl_set_memory_region: "
7229 "failed to munmap memory\n");
7232 if (!kvm->arch.n_requested_mmu_pages)
7233 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7236 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7238 * Write protect all pages for dirty logging.
7239 * Existing largepage mappings are destroyed here and new ones will
7240 * not be created until the end of the logging.
7242 if ((change != KVM_MR_DELETE) && (mem->flags & KVM_MEM_LOG_DIRTY_PAGES))
7243 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
7246 void kvm_arch_flush_shadow_all(struct kvm *kvm)
7248 kvm_mmu_invalidate_zap_all_pages(kvm);
7251 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7252 struct kvm_memory_slot *slot)
7254 kvm_mmu_invalidate_zap_all_pages(kvm);
7257 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
7259 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7260 !vcpu->arch.apf.halted)
7261 || !list_empty_careful(&vcpu->async_pf.done)
7262 || kvm_apic_has_events(vcpu)
7263 || vcpu->arch.pv.pv_unhalted
7264 || atomic_read(&vcpu->arch.nmi_queued) ||
7265 (kvm_arch_interrupt_allowed(vcpu) &&
7266 kvm_cpu_has_interrupt(vcpu));
7269 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
7271 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
7274 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7276 return kvm_x86_ops->interrupt_allowed(vcpu);
7279 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7281 unsigned long current_rip = kvm_rip_read(vcpu) +
7282 get_segment_base(vcpu, VCPU_SREG_CS);
7284 return current_rip == linear_rip;
7286 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7288 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7290 unsigned long rflags;
7292 rflags = kvm_x86_ops->get_rflags(vcpu);
7293 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7294 rflags &= ~X86_EFLAGS_TF;
7297 EXPORT_SYMBOL_GPL(kvm_get_rflags);
7299 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7301 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
7302 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
7303 rflags |= X86_EFLAGS_TF;
7304 kvm_x86_ops->set_rflags(vcpu, rflags);
7305 kvm_make_request(KVM_REQ_EVENT, vcpu);
7307 EXPORT_SYMBOL_GPL(kvm_set_rflags);
7309 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7313 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
7317 r = kvm_mmu_reload(vcpu);
7321 if (!vcpu->arch.mmu.direct_map &&
7322 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7325 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7328 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7330 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7333 static inline u32 kvm_async_pf_next_probe(u32 key)
7335 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7338 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7340 u32 key = kvm_async_pf_hash_fn(gfn);
7342 while (vcpu->arch.apf.gfns[key] != ~0)
7343 key = kvm_async_pf_next_probe(key);
7345 vcpu->arch.apf.gfns[key] = gfn;
7348 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7351 u32 key = kvm_async_pf_hash_fn(gfn);
7353 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
7354 (vcpu->arch.apf.gfns[key] != gfn &&
7355 vcpu->arch.apf.gfns[key] != ~0); i++)
7356 key = kvm_async_pf_next_probe(key);
7361 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7363 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7366 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7370 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7372 vcpu->arch.apf.gfns[i] = ~0;
7374 j = kvm_async_pf_next_probe(j);
7375 if (vcpu->arch.apf.gfns[j] == ~0)
7377 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7379 * k lies cyclically in ]i,j]
7381 * |....j i.k.| or |.k..j i...|
7383 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7384 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7389 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7392 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7396 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7397 struct kvm_async_pf *work)
7399 struct x86_exception fault;
7401 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
7402 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7404 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
7405 (vcpu->arch.apf.send_user_only &&
7406 kvm_x86_ops->get_cpl(vcpu) == 0))
7407 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7408 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
7409 fault.vector = PF_VECTOR;
7410 fault.error_code_valid = true;
7411 fault.error_code = 0;
7412 fault.nested_page_fault = false;
7413 fault.address = work->arch.token;
7414 kvm_inject_page_fault(vcpu, &fault);
7418 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7419 struct kvm_async_pf *work)
7421 struct x86_exception fault;
7423 trace_kvm_async_pf_ready(work->arch.token, work->gva);
7424 if (work->wakeup_all)
7425 work->arch.token = ~0; /* broadcast wakeup */
7427 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7429 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7430 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
7431 fault.vector = PF_VECTOR;
7432 fault.error_code_valid = true;
7433 fault.error_code = 0;
7434 fault.nested_page_fault = false;
7435 fault.address = work->arch.token;
7436 kvm_inject_page_fault(vcpu, &fault);
7438 vcpu->arch.apf.halted = false;
7439 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7442 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7444 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7447 return !kvm_event_needs_reinjection(vcpu) &&
7448 kvm_x86_ops->interrupt_allowed(vcpu);
7451 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
7453 atomic_inc(&kvm->arch.noncoherent_dma_count);
7455 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
7457 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
7459 atomic_dec(&kvm->arch.noncoherent_dma_count);
7461 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
7463 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
7465 return atomic_read(&kvm->arch.noncoherent_dma_count);
7467 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
7469 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
7470 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
7471 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
7472 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
7473 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
7474 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
7475 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
7476 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
7477 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
7478 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
7479 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
7480 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
7481 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);