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 #define EXCPT_BENIGN 0
265 #define EXCPT_CONTRIBUTORY 1
268 static int exception_class(int vector)
278 return EXCPT_CONTRIBUTORY;
285 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
286 unsigned nr, bool has_error, u32 error_code,
292 kvm_make_request(KVM_REQ_EVENT, vcpu);
294 if (!vcpu->arch.exception.pending) {
296 vcpu->arch.exception.pending = true;
297 vcpu->arch.exception.has_error_code = has_error;
298 vcpu->arch.exception.nr = nr;
299 vcpu->arch.exception.error_code = error_code;
300 vcpu->arch.exception.reinject = reinject;
304 /* to check exception */
305 prev_nr = vcpu->arch.exception.nr;
306 if (prev_nr == DF_VECTOR) {
307 /* triple fault -> shutdown */
308 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
311 class1 = exception_class(prev_nr);
312 class2 = exception_class(nr);
313 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
314 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
315 /* generate double fault per SDM Table 5-5 */
316 vcpu->arch.exception.pending = true;
317 vcpu->arch.exception.has_error_code = true;
318 vcpu->arch.exception.nr = DF_VECTOR;
319 vcpu->arch.exception.error_code = 0;
321 /* replace previous exception with a new one in a hope
322 that instruction re-execution will regenerate lost
327 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
329 kvm_multiple_exception(vcpu, nr, false, 0, false);
331 EXPORT_SYMBOL_GPL(kvm_queue_exception);
333 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
335 kvm_multiple_exception(vcpu, nr, false, 0, true);
337 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
339 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
342 kvm_inject_gp(vcpu, 0);
344 kvm_x86_ops->skip_emulated_instruction(vcpu);
346 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
348 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
350 ++vcpu->stat.pf_guest;
351 vcpu->arch.cr2 = fault->address;
352 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
354 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
356 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
358 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
359 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
361 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
364 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
366 atomic_inc(&vcpu->arch.nmi_queued);
367 kvm_make_request(KVM_REQ_NMI, vcpu);
369 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
371 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
373 kvm_multiple_exception(vcpu, nr, true, error_code, false);
375 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
377 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
379 kvm_multiple_exception(vcpu, nr, true, error_code, true);
381 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
384 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
385 * a #GP and return false.
387 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
389 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
391 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
394 EXPORT_SYMBOL_GPL(kvm_require_cpl);
397 * This function will be used to read from the physical memory of the currently
398 * running guest. The difference to kvm_read_guest_page is that this function
399 * can read from guest physical or from the guest's guest physical memory.
401 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
402 gfn_t ngfn, void *data, int offset, int len,
408 ngpa = gfn_to_gpa(ngfn);
409 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
410 if (real_gfn == UNMAPPED_GVA)
413 real_gfn = gpa_to_gfn(real_gfn);
415 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
417 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
419 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
420 void *data, int offset, int len, u32 access)
422 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
423 data, offset, len, access);
427 * Load the pae pdptrs. Return true is they are all valid.
429 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
431 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
432 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
435 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
437 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
438 offset * sizeof(u64), sizeof(pdpte),
439 PFERR_USER_MASK|PFERR_WRITE_MASK);
444 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
445 if (is_present_gpte(pdpte[i]) &&
446 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
453 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
454 __set_bit(VCPU_EXREG_PDPTR,
455 (unsigned long *)&vcpu->arch.regs_avail);
456 __set_bit(VCPU_EXREG_PDPTR,
457 (unsigned long *)&vcpu->arch.regs_dirty);
462 EXPORT_SYMBOL_GPL(load_pdptrs);
464 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
466 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
472 if (is_long_mode(vcpu) || !is_pae(vcpu))
475 if (!test_bit(VCPU_EXREG_PDPTR,
476 (unsigned long *)&vcpu->arch.regs_avail))
479 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
480 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
481 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
482 PFERR_USER_MASK | PFERR_WRITE_MASK);
485 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
491 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
493 unsigned long old_cr0 = kvm_read_cr0(vcpu);
494 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
495 X86_CR0_CD | X86_CR0_NW;
500 if (cr0 & 0xffffffff00000000UL)
504 cr0 &= ~CR0_RESERVED_BITS;
506 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
509 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
512 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
514 if ((vcpu->arch.efer & EFER_LME)) {
519 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
524 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
529 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
532 kvm_x86_ops->set_cr0(vcpu, cr0);
534 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
535 kvm_clear_async_pf_completion_queue(vcpu);
536 kvm_async_pf_hash_reset(vcpu);
539 if ((cr0 ^ old_cr0) & update_bits)
540 kvm_mmu_reset_context(vcpu);
543 EXPORT_SYMBOL_GPL(kvm_set_cr0);
545 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
547 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
549 EXPORT_SYMBOL_GPL(kvm_lmsw);
551 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
555 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
556 if (index != XCR_XFEATURE_ENABLED_MASK)
559 if (kvm_x86_ops->get_cpl(vcpu) != 0)
561 if (!(xcr0 & XSTATE_FP))
563 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
565 if (xcr0 & ~host_xcr0)
567 vcpu->arch.xcr0 = xcr0;
568 vcpu->guest_xcr0_loaded = 0;
572 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
574 if (__kvm_set_xcr(vcpu, index, xcr)) {
575 kvm_inject_gp(vcpu, 0);
580 EXPORT_SYMBOL_GPL(kvm_set_xcr);
582 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
584 unsigned long old_cr4 = kvm_read_cr4(vcpu);
585 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
586 X86_CR4_PAE | X86_CR4_SMEP;
587 if (cr4 & CR4_RESERVED_BITS)
590 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
593 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
596 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_RDWRGSFS))
599 if (is_long_mode(vcpu)) {
600 if (!(cr4 & X86_CR4_PAE))
602 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
603 && ((cr4 ^ old_cr4) & pdptr_bits)
604 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
608 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
609 if (!guest_cpuid_has_pcid(vcpu))
612 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
613 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
617 if (kvm_x86_ops->set_cr4(vcpu, cr4))
620 if (((cr4 ^ old_cr4) & pdptr_bits) ||
621 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
622 kvm_mmu_reset_context(vcpu);
624 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
625 kvm_update_cpuid(vcpu);
629 EXPORT_SYMBOL_GPL(kvm_set_cr4);
631 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
633 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
634 kvm_mmu_sync_roots(vcpu);
635 kvm_mmu_flush_tlb(vcpu);
639 if (is_long_mode(vcpu)) {
640 if (kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)) {
641 if (cr3 & CR3_PCID_ENABLED_RESERVED_BITS)
644 if (cr3 & CR3_L_MODE_RESERVED_BITS)
648 if (cr3 & CR3_PAE_RESERVED_BITS)
650 if (is_paging(vcpu) &&
651 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
655 * We don't check reserved bits in nonpae mode, because
656 * this isn't enforced, and VMware depends on this.
661 * Does the new cr3 value map to physical memory? (Note, we
662 * catch an invalid cr3 even in real-mode, because it would
663 * cause trouble later on when we turn on paging anyway.)
665 * A real CPU would silently accept an invalid cr3 and would
666 * attempt to use it - with largely undefined (and often hard
667 * to debug) behavior on the guest side.
669 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
671 vcpu->arch.cr3 = cr3;
672 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
673 vcpu->arch.mmu.new_cr3(vcpu);
676 EXPORT_SYMBOL_GPL(kvm_set_cr3);
678 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
680 if (cr8 & CR8_RESERVED_BITS)
682 if (irqchip_in_kernel(vcpu->kvm))
683 kvm_lapic_set_tpr(vcpu, cr8);
685 vcpu->arch.cr8 = cr8;
688 EXPORT_SYMBOL_GPL(kvm_set_cr8);
690 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
692 if (irqchip_in_kernel(vcpu->kvm))
693 return kvm_lapic_get_cr8(vcpu);
695 return vcpu->arch.cr8;
697 EXPORT_SYMBOL_GPL(kvm_get_cr8);
699 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
703 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
704 dr7 = vcpu->arch.guest_debug_dr7;
706 dr7 = vcpu->arch.dr7;
707 kvm_x86_ops->set_dr7(vcpu, dr7);
708 vcpu->arch.switch_db_regs = (dr7 & DR7_BP_EN_MASK);
711 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
715 vcpu->arch.db[dr] = val;
716 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
717 vcpu->arch.eff_db[dr] = val;
720 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
724 if (val & 0xffffffff00000000ULL)
726 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
729 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
733 if (val & 0xffffffff00000000ULL)
735 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
736 kvm_update_dr7(vcpu);
743 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
747 res = __kvm_set_dr(vcpu, dr, val);
749 kvm_queue_exception(vcpu, UD_VECTOR);
751 kvm_inject_gp(vcpu, 0);
755 EXPORT_SYMBOL_GPL(kvm_set_dr);
757 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
761 *val = vcpu->arch.db[dr];
764 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
768 *val = vcpu->arch.dr6;
771 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
775 *val = vcpu->arch.dr7;
782 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
784 if (_kvm_get_dr(vcpu, dr, val)) {
785 kvm_queue_exception(vcpu, UD_VECTOR);
790 EXPORT_SYMBOL_GPL(kvm_get_dr);
792 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
794 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
798 err = kvm_pmu_read_pmc(vcpu, ecx, &data);
801 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
802 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
805 EXPORT_SYMBOL_GPL(kvm_rdpmc);
808 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
809 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
811 * This list is modified at module load time to reflect the
812 * capabilities of the host cpu. This capabilities test skips MSRs that are
813 * kvm-specific. Those are put in the beginning of the list.
816 #define KVM_SAVE_MSRS_BEGIN 10
817 static u32 msrs_to_save[] = {
818 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
819 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
820 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
821 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
823 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
826 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
828 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
831 static unsigned num_msrs_to_save;
833 static const u32 emulated_msrs[] = {
835 MSR_IA32_TSCDEADLINE,
836 MSR_IA32_MISC_ENABLE,
841 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
843 u64 old_efer = vcpu->arch.efer;
845 if (efer & efer_reserved_bits)
849 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
852 if (efer & EFER_FFXSR) {
853 struct kvm_cpuid_entry2 *feat;
855 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
856 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
860 if (efer & EFER_SVME) {
861 struct kvm_cpuid_entry2 *feat;
863 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
864 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
869 efer |= vcpu->arch.efer & EFER_LMA;
871 kvm_x86_ops->set_efer(vcpu, efer);
873 /* Update reserved bits */
874 if ((efer ^ old_efer) & EFER_NX)
875 kvm_mmu_reset_context(vcpu);
880 void kvm_enable_efer_bits(u64 mask)
882 efer_reserved_bits &= ~mask;
884 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
888 * Writes msr value into into the appropriate "register".
889 * Returns 0 on success, non-0 otherwise.
890 * Assumes vcpu_load() was already called.
892 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
894 return kvm_x86_ops->set_msr(vcpu, msr);
898 * Adapt set_msr() to msr_io()'s calling convention
900 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
906 msr.host_initiated = true;
907 return kvm_set_msr(vcpu, &msr);
911 struct pvclock_gtod_data {
914 struct { /* extract of a clocksource struct */
922 /* open coded 'struct timespec' */
923 u64 monotonic_time_snsec;
924 time_t monotonic_time_sec;
927 static struct pvclock_gtod_data pvclock_gtod_data;
929 static void update_pvclock_gtod(struct timekeeper *tk)
931 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
933 write_seqcount_begin(&vdata->seq);
935 /* copy pvclock gtod data */
936 vdata->clock.vclock_mode = tk->clock->archdata.vclock_mode;
937 vdata->clock.cycle_last = tk->clock->cycle_last;
938 vdata->clock.mask = tk->clock->mask;
939 vdata->clock.mult = tk->mult;
940 vdata->clock.shift = tk->shift;
942 vdata->monotonic_time_sec = tk->xtime_sec
943 + tk->wall_to_monotonic.tv_sec;
944 vdata->monotonic_time_snsec = tk->xtime_nsec
945 + (tk->wall_to_monotonic.tv_nsec
947 while (vdata->monotonic_time_snsec >=
948 (((u64)NSEC_PER_SEC) << tk->shift)) {
949 vdata->monotonic_time_snsec -=
950 ((u64)NSEC_PER_SEC) << tk->shift;
951 vdata->monotonic_time_sec++;
954 write_seqcount_end(&vdata->seq);
959 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
963 struct pvclock_wall_clock wc;
964 struct timespec boot;
969 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
974 ++version; /* first time write, random junk */
978 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
981 * The guest calculates current wall clock time by adding
982 * system time (updated by kvm_guest_time_update below) to the
983 * wall clock specified here. guest system time equals host
984 * system time for us, thus we must fill in host boot time here.
988 if (kvm->arch.kvmclock_offset) {
989 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
990 boot = timespec_sub(boot, ts);
992 wc.sec = boot.tv_sec;
993 wc.nsec = boot.tv_nsec;
994 wc.version = version;
996 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
999 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1002 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1004 uint32_t quotient, remainder;
1006 /* Don't try to replace with do_div(), this one calculates
1007 * "(dividend << 32) / divisor" */
1009 : "=a" (quotient), "=d" (remainder)
1010 : "0" (0), "1" (dividend), "r" (divisor) );
1014 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1015 s8 *pshift, u32 *pmultiplier)
1022 tps64 = base_khz * 1000LL;
1023 scaled64 = scaled_khz * 1000LL;
1024 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1029 tps32 = (uint32_t)tps64;
1030 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1031 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1039 *pmultiplier = div_frac(scaled64, tps32);
1041 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1042 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1045 static inline u64 get_kernel_ns(void)
1049 WARN_ON(preemptible());
1051 monotonic_to_bootbased(&ts);
1052 return timespec_to_ns(&ts);
1055 #ifdef CONFIG_X86_64
1056 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1059 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1060 unsigned long max_tsc_khz;
1062 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1064 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1065 vcpu->arch.virtual_tsc_shift);
1068 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1070 u64 v = (u64)khz * (1000000 + ppm);
1075 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1077 u32 thresh_lo, thresh_hi;
1078 int use_scaling = 0;
1080 /* tsc_khz can be zero if TSC calibration fails */
1081 if (this_tsc_khz == 0)
1084 /* Compute a scale to convert nanoseconds in TSC cycles */
1085 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1086 &vcpu->arch.virtual_tsc_shift,
1087 &vcpu->arch.virtual_tsc_mult);
1088 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1091 * Compute the variation in TSC rate which is acceptable
1092 * within the range of tolerance and decide if the
1093 * rate being applied is within that bounds of the hardware
1094 * rate. If so, no scaling or compensation need be done.
1096 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1097 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1098 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1099 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1102 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1105 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1107 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1108 vcpu->arch.virtual_tsc_mult,
1109 vcpu->arch.virtual_tsc_shift);
1110 tsc += vcpu->arch.this_tsc_write;
1114 void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1116 #ifdef CONFIG_X86_64
1118 bool do_request = false;
1119 struct kvm_arch *ka = &vcpu->kvm->arch;
1120 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1122 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1123 atomic_read(&vcpu->kvm->online_vcpus));
1125 if (vcpus_matched && gtod->clock.vclock_mode == VCLOCK_TSC)
1126 if (!ka->use_master_clock)
1129 if (!vcpus_matched && ka->use_master_clock)
1133 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1135 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1136 atomic_read(&vcpu->kvm->online_vcpus),
1137 ka->use_master_clock, gtod->clock.vclock_mode);
1141 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1143 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1144 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1147 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1149 struct kvm *kvm = vcpu->kvm;
1150 u64 offset, ns, elapsed;
1151 unsigned long flags;
1154 u64 data = msr->data;
1156 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1157 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1158 ns = get_kernel_ns();
1159 elapsed = ns - kvm->arch.last_tsc_nsec;
1161 if (vcpu->arch.virtual_tsc_khz) {
1162 /* n.b - signed multiplication and division required */
1163 usdiff = data - kvm->arch.last_tsc_write;
1164 #ifdef CONFIG_X86_64
1165 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1167 /* do_div() only does unsigned */
1168 asm("idivl %2; xor %%edx, %%edx"
1170 : "A"(usdiff * 1000), "rm"(vcpu->arch.virtual_tsc_khz));
1172 do_div(elapsed, 1000);
1177 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1180 * Special case: TSC write with a small delta (1 second) of virtual
1181 * cycle time against real time is interpreted as an attempt to
1182 * synchronize the CPU.
1184 * For a reliable TSC, we can match TSC offsets, and for an unstable
1185 * TSC, we add elapsed time in this computation. We could let the
1186 * compensation code attempt to catch up if we fall behind, but
1187 * it's better to try to match offsets from the beginning.
1189 if (usdiff < USEC_PER_SEC &&
1190 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1191 if (!check_tsc_unstable()) {
1192 offset = kvm->arch.cur_tsc_offset;
1193 pr_debug("kvm: matched tsc offset for %llu\n", data);
1195 u64 delta = nsec_to_cycles(vcpu, elapsed);
1197 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1198 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1203 * We split periods of matched TSC writes into generations.
1204 * For each generation, we track the original measured
1205 * nanosecond time, offset, and write, so if TSCs are in
1206 * sync, we can match exact offset, and if not, we can match
1207 * exact software computation in compute_guest_tsc()
1209 * These values are tracked in kvm->arch.cur_xxx variables.
1211 kvm->arch.cur_tsc_generation++;
1212 kvm->arch.cur_tsc_nsec = ns;
1213 kvm->arch.cur_tsc_write = data;
1214 kvm->arch.cur_tsc_offset = offset;
1216 pr_debug("kvm: new tsc generation %u, clock %llu\n",
1217 kvm->arch.cur_tsc_generation, data);
1221 * We also track th most recent recorded KHZ, write and time to
1222 * allow the matching interval to be extended at each write.
1224 kvm->arch.last_tsc_nsec = ns;
1225 kvm->arch.last_tsc_write = data;
1226 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1228 /* Reset of TSC must disable overshoot protection below */
1229 vcpu->arch.hv_clock.tsc_timestamp = 0;
1230 vcpu->arch.last_guest_tsc = data;
1232 /* Keep track of which generation this VCPU has synchronized to */
1233 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1234 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1235 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1237 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1238 update_ia32_tsc_adjust_msr(vcpu, offset);
1239 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1240 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1242 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1244 kvm->arch.nr_vcpus_matched_tsc++;
1246 kvm->arch.nr_vcpus_matched_tsc = 0;
1248 kvm_track_tsc_matching(vcpu);
1249 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1252 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1254 #ifdef CONFIG_X86_64
1256 static cycle_t read_tsc(void)
1262 * Empirically, a fence (of type that depends on the CPU)
1263 * before rdtsc is enough to ensure that rdtsc is ordered
1264 * with respect to loads. The various CPU manuals are unclear
1265 * as to whether rdtsc can be reordered with later loads,
1266 * but no one has ever seen it happen.
1269 ret = (cycle_t)vget_cycles();
1271 last = pvclock_gtod_data.clock.cycle_last;
1273 if (likely(ret >= last))
1277 * GCC likes to generate cmov here, but this branch is extremely
1278 * predictable (it's just a funciton of time and the likely is
1279 * very likely) and there's a data dependence, so force GCC
1280 * to generate a branch instead. I don't barrier() because
1281 * we don't actually need a barrier, and if this function
1282 * ever gets inlined it will generate worse code.
1288 static inline u64 vgettsc(cycle_t *cycle_now)
1291 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1293 *cycle_now = read_tsc();
1295 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1296 return v * gtod->clock.mult;
1299 static int do_monotonic(struct timespec *ts, cycle_t *cycle_now)
1304 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1308 seq = read_seqcount_begin(>od->seq);
1309 mode = gtod->clock.vclock_mode;
1310 ts->tv_sec = gtod->monotonic_time_sec;
1311 ns = gtod->monotonic_time_snsec;
1312 ns += vgettsc(cycle_now);
1313 ns >>= gtod->clock.shift;
1314 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1315 timespec_add_ns(ts, ns);
1320 /* returns true if host is using tsc clocksource */
1321 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1325 /* checked again under seqlock below */
1326 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1329 if (do_monotonic(&ts, cycle_now) != VCLOCK_TSC)
1332 monotonic_to_bootbased(&ts);
1333 *kernel_ns = timespec_to_ns(&ts);
1341 * Assuming a stable TSC across physical CPUS, and a stable TSC
1342 * across virtual CPUs, the following condition is possible.
1343 * Each numbered line represents an event visible to both
1344 * CPUs at the next numbered event.
1346 * "timespecX" represents host monotonic time. "tscX" represents
1349 * VCPU0 on CPU0 | VCPU1 on CPU1
1351 * 1. read timespec0,tsc0
1352 * 2. | timespec1 = timespec0 + N
1354 * 3. transition to guest | transition to guest
1355 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1356 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1357 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1359 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1362 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1364 * - 0 < N - M => M < N
1366 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1367 * always the case (the difference between two distinct xtime instances
1368 * might be smaller then the difference between corresponding TSC reads,
1369 * when updating guest vcpus pvclock areas).
1371 * To avoid that problem, do not allow visibility of distinct
1372 * system_timestamp/tsc_timestamp values simultaneously: use a master
1373 * copy of host monotonic time values. Update that master copy
1376 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1380 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1382 #ifdef CONFIG_X86_64
1383 struct kvm_arch *ka = &kvm->arch;
1385 bool host_tsc_clocksource, vcpus_matched;
1387 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1388 atomic_read(&kvm->online_vcpus));
1391 * If the host uses TSC clock, then passthrough TSC as stable
1394 host_tsc_clocksource = kvm_get_time_and_clockread(
1395 &ka->master_kernel_ns,
1396 &ka->master_cycle_now);
1398 ka->use_master_clock = host_tsc_clocksource & vcpus_matched;
1400 if (ka->use_master_clock)
1401 atomic_set(&kvm_guest_has_master_clock, 1);
1403 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1404 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1409 static int kvm_guest_time_update(struct kvm_vcpu *v)
1411 unsigned long flags, this_tsc_khz;
1412 struct kvm_vcpu_arch *vcpu = &v->arch;
1413 struct kvm_arch *ka = &v->kvm->arch;
1414 s64 kernel_ns, max_kernel_ns;
1415 u64 tsc_timestamp, host_tsc;
1416 struct pvclock_vcpu_time_info guest_hv_clock;
1418 bool use_master_clock;
1424 * If the host uses TSC clock, then passthrough TSC as stable
1427 spin_lock(&ka->pvclock_gtod_sync_lock);
1428 use_master_clock = ka->use_master_clock;
1429 if (use_master_clock) {
1430 host_tsc = ka->master_cycle_now;
1431 kernel_ns = ka->master_kernel_ns;
1433 spin_unlock(&ka->pvclock_gtod_sync_lock);
1435 /* Keep irq disabled to prevent changes to the clock */
1436 local_irq_save(flags);
1437 this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
1438 if (unlikely(this_tsc_khz == 0)) {
1439 local_irq_restore(flags);
1440 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1443 if (!use_master_clock) {
1444 host_tsc = native_read_tsc();
1445 kernel_ns = get_kernel_ns();
1448 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1451 * We may have to catch up the TSC to match elapsed wall clock
1452 * time for two reasons, even if kvmclock is used.
1453 * 1) CPU could have been running below the maximum TSC rate
1454 * 2) Broken TSC compensation resets the base at each VCPU
1455 * entry to avoid unknown leaps of TSC even when running
1456 * again on the same CPU. This may cause apparent elapsed
1457 * time to disappear, and the guest to stand still or run
1460 if (vcpu->tsc_catchup) {
1461 u64 tsc = compute_guest_tsc(v, kernel_ns);
1462 if (tsc > tsc_timestamp) {
1463 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1464 tsc_timestamp = tsc;
1468 local_irq_restore(flags);
1470 if (!vcpu->pv_time_enabled)
1474 * Time as measured by the TSC may go backwards when resetting the base
1475 * tsc_timestamp. The reason for this is that the TSC resolution is
1476 * higher than the resolution of the other clock scales. Thus, many
1477 * possible measurments of the TSC correspond to one measurement of any
1478 * other clock, and so a spread of values is possible. This is not a
1479 * problem for the computation of the nanosecond clock; with TSC rates
1480 * around 1GHZ, there can only be a few cycles which correspond to one
1481 * nanosecond value, and any path through this code will inevitably
1482 * take longer than that. However, with the kernel_ns value itself,
1483 * the precision may be much lower, down to HZ granularity. If the
1484 * first sampling of TSC against kernel_ns ends in the low part of the
1485 * range, and the second in the high end of the range, we can get:
1487 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1489 * As the sampling errors potentially range in the thousands of cycles,
1490 * it is possible such a time value has already been observed by the
1491 * guest. To protect against this, we must compute the system time as
1492 * observed by the guest and ensure the new system time is greater.
1495 if (vcpu->hv_clock.tsc_timestamp) {
1496 max_kernel_ns = vcpu->last_guest_tsc -
1497 vcpu->hv_clock.tsc_timestamp;
1498 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1499 vcpu->hv_clock.tsc_to_system_mul,
1500 vcpu->hv_clock.tsc_shift);
1501 max_kernel_ns += vcpu->last_kernel_ns;
1504 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1505 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1506 &vcpu->hv_clock.tsc_shift,
1507 &vcpu->hv_clock.tsc_to_system_mul);
1508 vcpu->hw_tsc_khz = this_tsc_khz;
1511 /* with a master <monotonic time, tsc value> tuple,
1512 * pvclock clock reads always increase at the (scaled) rate
1513 * of guest TSC - no need to deal with sampling errors.
1515 if (!use_master_clock) {
1516 if (max_kernel_ns > kernel_ns)
1517 kernel_ns = max_kernel_ns;
1519 /* With all the info we got, fill in the values */
1520 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1521 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1522 vcpu->last_kernel_ns = kernel_ns;
1523 vcpu->last_guest_tsc = tsc_timestamp;
1526 * The interface expects us to write an even number signaling that the
1527 * update is finished. Since the guest won't see the intermediate
1528 * state, we just increase by 2 at the end.
1530 vcpu->hv_clock.version += 2;
1532 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1533 &guest_hv_clock, sizeof(guest_hv_clock))))
1536 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1537 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1539 if (vcpu->pvclock_set_guest_stopped_request) {
1540 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1541 vcpu->pvclock_set_guest_stopped_request = false;
1544 /* If the host uses TSC clocksource, then it is stable */
1545 if (use_master_clock)
1546 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1548 vcpu->hv_clock.flags = pvclock_flags;
1550 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1552 sizeof(vcpu->hv_clock));
1556 static bool msr_mtrr_valid(unsigned msr)
1559 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1560 case MSR_MTRRfix64K_00000:
1561 case MSR_MTRRfix16K_80000:
1562 case MSR_MTRRfix16K_A0000:
1563 case MSR_MTRRfix4K_C0000:
1564 case MSR_MTRRfix4K_C8000:
1565 case MSR_MTRRfix4K_D0000:
1566 case MSR_MTRRfix4K_D8000:
1567 case MSR_MTRRfix4K_E0000:
1568 case MSR_MTRRfix4K_E8000:
1569 case MSR_MTRRfix4K_F0000:
1570 case MSR_MTRRfix4K_F8000:
1571 case MSR_MTRRdefType:
1572 case MSR_IA32_CR_PAT:
1580 static bool valid_pat_type(unsigned t)
1582 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1585 static bool valid_mtrr_type(unsigned t)
1587 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1590 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1594 if (!msr_mtrr_valid(msr))
1597 if (msr == MSR_IA32_CR_PAT) {
1598 for (i = 0; i < 8; i++)
1599 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1602 } else if (msr == MSR_MTRRdefType) {
1605 return valid_mtrr_type(data & 0xff);
1606 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1607 for (i = 0; i < 8 ; i++)
1608 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1613 /* variable MTRRs */
1614 return valid_mtrr_type(data & 0xff);
1617 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1619 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1621 if (!mtrr_valid(vcpu, msr, data))
1624 if (msr == MSR_MTRRdefType) {
1625 vcpu->arch.mtrr_state.def_type = data;
1626 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1627 } else if (msr == MSR_MTRRfix64K_00000)
1629 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1630 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1631 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1632 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1633 else if (msr == MSR_IA32_CR_PAT)
1634 vcpu->arch.pat = data;
1635 else { /* Variable MTRRs */
1636 int idx, is_mtrr_mask;
1639 idx = (msr - 0x200) / 2;
1640 is_mtrr_mask = msr - 0x200 - 2 * idx;
1643 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1646 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1650 kvm_mmu_reset_context(vcpu);
1654 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1656 u64 mcg_cap = vcpu->arch.mcg_cap;
1657 unsigned bank_num = mcg_cap & 0xff;
1660 case MSR_IA32_MCG_STATUS:
1661 vcpu->arch.mcg_status = data;
1663 case MSR_IA32_MCG_CTL:
1664 if (!(mcg_cap & MCG_CTL_P))
1666 if (data != 0 && data != ~(u64)0)
1668 vcpu->arch.mcg_ctl = data;
1671 if (msr >= MSR_IA32_MC0_CTL &&
1672 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1673 u32 offset = msr - MSR_IA32_MC0_CTL;
1674 /* only 0 or all 1s can be written to IA32_MCi_CTL
1675 * some Linux kernels though clear bit 10 in bank 4 to
1676 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1677 * this to avoid an uncatched #GP in the guest
1679 if ((offset & 0x3) == 0 &&
1680 data != 0 && (data | (1 << 10)) != ~(u64)0)
1682 vcpu->arch.mce_banks[offset] = data;
1690 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1692 struct kvm *kvm = vcpu->kvm;
1693 int lm = is_long_mode(vcpu);
1694 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1695 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1696 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1697 : kvm->arch.xen_hvm_config.blob_size_32;
1698 u32 page_num = data & ~PAGE_MASK;
1699 u64 page_addr = data & PAGE_MASK;
1704 if (page_num >= blob_size)
1707 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1712 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1721 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1723 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1726 static bool kvm_hv_msr_partition_wide(u32 msr)
1730 case HV_X64_MSR_GUEST_OS_ID:
1731 case HV_X64_MSR_HYPERCALL:
1739 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1741 struct kvm *kvm = vcpu->kvm;
1744 case HV_X64_MSR_GUEST_OS_ID:
1745 kvm->arch.hv_guest_os_id = data;
1746 /* setting guest os id to zero disables hypercall page */
1747 if (!kvm->arch.hv_guest_os_id)
1748 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1750 case HV_X64_MSR_HYPERCALL: {
1755 /* if guest os id is not set hypercall should remain disabled */
1756 if (!kvm->arch.hv_guest_os_id)
1758 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1759 kvm->arch.hv_hypercall = data;
1762 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1763 addr = gfn_to_hva(kvm, gfn);
1764 if (kvm_is_error_hva(addr))
1766 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1767 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1768 if (__copy_to_user((void __user *)addr, instructions, 4))
1770 kvm->arch.hv_hypercall = data;
1774 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1775 "data 0x%llx\n", msr, data);
1781 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1784 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1787 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1788 vcpu->arch.hv_vapic = data;
1791 addr = gfn_to_hva(vcpu->kvm, data >>
1792 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1793 if (kvm_is_error_hva(addr))
1795 if (__clear_user((void __user *)addr, PAGE_SIZE))
1797 vcpu->arch.hv_vapic = data;
1800 case HV_X64_MSR_EOI:
1801 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1802 case HV_X64_MSR_ICR:
1803 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1804 case HV_X64_MSR_TPR:
1805 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1807 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1808 "data 0x%llx\n", msr, data);
1815 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1817 gpa_t gpa = data & ~0x3f;
1819 /* Bits 2:5 are reserved, Should be zero */
1823 vcpu->arch.apf.msr_val = data;
1825 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1826 kvm_clear_async_pf_completion_queue(vcpu);
1827 kvm_async_pf_hash_reset(vcpu);
1831 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1834 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1835 kvm_async_pf_wakeup_all(vcpu);
1839 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1841 vcpu->arch.pv_time_enabled = false;
1844 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1848 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1851 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1852 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1853 vcpu->arch.st.accum_steal = delta;
1856 static void record_steal_time(struct kvm_vcpu *vcpu)
1858 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1861 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1862 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1865 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1866 vcpu->arch.st.steal.version += 2;
1867 vcpu->arch.st.accum_steal = 0;
1869 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1870 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1873 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1876 u32 msr = msr_info->index;
1877 u64 data = msr_info->data;
1880 case MSR_AMD64_NB_CFG:
1881 case MSR_IA32_UCODE_REV:
1882 case MSR_IA32_UCODE_WRITE:
1883 case MSR_VM_HSAVE_PA:
1884 case MSR_AMD64_PATCH_LOADER:
1885 case MSR_AMD64_BU_CFG2:
1889 return set_efer(vcpu, data);
1891 data &= ~(u64)0x40; /* ignore flush filter disable */
1892 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1893 data &= ~(u64)0x8; /* ignore TLB cache disable */
1895 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1900 case MSR_FAM10H_MMIO_CONF_BASE:
1902 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1907 case MSR_IA32_DEBUGCTLMSR:
1909 /* We support the non-activated case already */
1911 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1912 /* Values other than LBR and BTF are vendor-specific,
1913 thus reserved and should throw a #GP */
1916 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1919 case 0x200 ... 0x2ff:
1920 return set_msr_mtrr(vcpu, msr, data);
1921 case MSR_IA32_APICBASE:
1922 kvm_set_apic_base(vcpu, data);
1924 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1925 return kvm_x2apic_msr_write(vcpu, msr, data);
1926 case MSR_IA32_TSCDEADLINE:
1927 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1929 case MSR_IA32_TSC_ADJUST:
1930 if (guest_cpuid_has_tsc_adjust(vcpu)) {
1931 if (!msr_info->host_initiated) {
1932 u64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
1933 kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
1935 vcpu->arch.ia32_tsc_adjust_msr = data;
1938 case MSR_IA32_MISC_ENABLE:
1939 vcpu->arch.ia32_misc_enable_msr = data;
1941 case MSR_KVM_WALL_CLOCK_NEW:
1942 case MSR_KVM_WALL_CLOCK:
1943 vcpu->kvm->arch.wall_clock = data;
1944 kvm_write_wall_clock(vcpu->kvm, data);
1946 case MSR_KVM_SYSTEM_TIME_NEW:
1947 case MSR_KVM_SYSTEM_TIME: {
1949 kvmclock_reset(vcpu);
1951 vcpu->arch.time = data;
1952 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1954 /* we verify if the enable bit is set... */
1958 gpa_offset = data & ~(PAGE_MASK | 1);
1960 /* Check that the address is 32-byte aligned. */
1961 if (gpa_offset & (sizeof(struct pvclock_vcpu_time_info) - 1))
1964 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
1965 &vcpu->arch.pv_time, data & ~1ULL))
1966 vcpu->arch.pv_time_enabled = false;
1968 vcpu->arch.pv_time_enabled = true;
1972 case MSR_KVM_ASYNC_PF_EN:
1973 if (kvm_pv_enable_async_pf(vcpu, data))
1976 case MSR_KVM_STEAL_TIME:
1978 if (unlikely(!sched_info_on()))
1981 if (data & KVM_STEAL_RESERVED_MASK)
1984 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
1985 data & KVM_STEAL_VALID_BITS))
1988 vcpu->arch.st.msr_val = data;
1990 if (!(data & KVM_MSR_ENABLED))
1993 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1996 accumulate_steal_time(vcpu);
1999 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2002 case MSR_KVM_PV_EOI_EN:
2003 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2007 case MSR_IA32_MCG_CTL:
2008 case MSR_IA32_MCG_STATUS:
2009 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2010 return set_msr_mce(vcpu, msr, data);
2012 /* Performance counters are not protected by a CPUID bit,
2013 * so we should check all of them in the generic path for the sake of
2014 * cross vendor migration.
2015 * Writing a zero into the event select MSRs disables them,
2016 * which we perfectly emulate ;-). Any other value should be at least
2017 * reported, some guests depend on them.
2019 case MSR_K7_EVNTSEL0:
2020 case MSR_K7_EVNTSEL1:
2021 case MSR_K7_EVNTSEL2:
2022 case MSR_K7_EVNTSEL3:
2024 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2025 "0x%x data 0x%llx\n", msr, data);
2027 /* at least RHEL 4 unconditionally writes to the perfctr registers,
2028 * so we ignore writes to make it happy.
2030 case MSR_K7_PERFCTR0:
2031 case MSR_K7_PERFCTR1:
2032 case MSR_K7_PERFCTR2:
2033 case MSR_K7_PERFCTR3:
2034 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2035 "0x%x data 0x%llx\n", msr, data);
2037 case MSR_P6_PERFCTR0:
2038 case MSR_P6_PERFCTR1:
2040 case MSR_P6_EVNTSEL0:
2041 case MSR_P6_EVNTSEL1:
2042 if (kvm_pmu_msr(vcpu, msr))
2043 return kvm_pmu_set_msr(vcpu, msr_info);
2045 if (pr || data != 0)
2046 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2047 "0x%x data 0x%llx\n", msr, data);
2049 case MSR_K7_CLK_CTL:
2051 * Ignore all writes to this no longer documented MSR.
2052 * Writes are only relevant for old K7 processors,
2053 * all pre-dating SVM, but a recommended workaround from
2054 * AMD for these chips. It is possible to specify the
2055 * affected processor models on the command line, hence
2056 * the need to ignore the workaround.
2059 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2060 if (kvm_hv_msr_partition_wide(msr)) {
2062 mutex_lock(&vcpu->kvm->lock);
2063 r = set_msr_hyperv_pw(vcpu, msr, data);
2064 mutex_unlock(&vcpu->kvm->lock);
2067 return set_msr_hyperv(vcpu, msr, data);
2069 case MSR_IA32_BBL_CR_CTL3:
2070 /* Drop writes to this legacy MSR -- see rdmsr
2071 * counterpart for further detail.
2073 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2075 case MSR_AMD64_OSVW_ID_LENGTH:
2076 if (!guest_cpuid_has_osvw(vcpu))
2078 vcpu->arch.osvw.length = data;
2080 case MSR_AMD64_OSVW_STATUS:
2081 if (!guest_cpuid_has_osvw(vcpu))
2083 vcpu->arch.osvw.status = data;
2086 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2087 return xen_hvm_config(vcpu, data);
2088 if (kvm_pmu_msr(vcpu, msr))
2089 return kvm_pmu_set_msr(vcpu, msr_info);
2091 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2095 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2102 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2106 * Reads an msr value (of 'msr_index') into 'pdata'.
2107 * Returns 0 on success, non-0 otherwise.
2108 * Assumes vcpu_load() was already called.
2110 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2112 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
2115 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2117 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2119 if (!msr_mtrr_valid(msr))
2122 if (msr == MSR_MTRRdefType)
2123 *pdata = vcpu->arch.mtrr_state.def_type +
2124 (vcpu->arch.mtrr_state.enabled << 10);
2125 else if (msr == MSR_MTRRfix64K_00000)
2127 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2128 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2129 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2130 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2131 else if (msr == MSR_IA32_CR_PAT)
2132 *pdata = vcpu->arch.pat;
2133 else { /* Variable MTRRs */
2134 int idx, is_mtrr_mask;
2137 idx = (msr - 0x200) / 2;
2138 is_mtrr_mask = msr - 0x200 - 2 * idx;
2141 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2144 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2151 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2154 u64 mcg_cap = vcpu->arch.mcg_cap;
2155 unsigned bank_num = mcg_cap & 0xff;
2158 case MSR_IA32_P5_MC_ADDR:
2159 case MSR_IA32_P5_MC_TYPE:
2162 case MSR_IA32_MCG_CAP:
2163 data = vcpu->arch.mcg_cap;
2165 case MSR_IA32_MCG_CTL:
2166 if (!(mcg_cap & MCG_CTL_P))
2168 data = vcpu->arch.mcg_ctl;
2170 case MSR_IA32_MCG_STATUS:
2171 data = vcpu->arch.mcg_status;
2174 if (msr >= MSR_IA32_MC0_CTL &&
2175 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
2176 u32 offset = msr - MSR_IA32_MC0_CTL;
2177 data = vcpu->arch.mce_banks[offset];
2186 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2189 struct kvm *kvm = vcpu->kvm;
2192 case HV_X64_MSR_GUEST_OS_ID:
2193 data = kvm->arch.hv_guest_os_id;
2195 case HV_X64_MSR_HYPERCALL:
2196 data = kvm->arch.hv_hypercall;
2199 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2207 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2212 case HV_X64_MSR_VP_INDEX: {
2215 kvm_for_each_vcpu(r, v, vcpu->kvm)
2220 case HV_X64_MSR_EOI:
2221 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2222 case HV_X64_MSR_ICR:
2223 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2224 case HV_X64_MSR_TPR:
2225 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
2226 case HV_X64_MSR_APIC_ASSIST_PAGE:
2227 data = vcpu->arch.hv_vapic;
2230 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2237 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2242 case MSR_IA32_PLATFORM_ID:
2243 case MSR_IA32_EBL_CR_POWERON:
2244 case MSR_IA32_DEBUGCTLMSR:
2245 case MSR_IA32_LASTBRANCHFROMIP:
2246 case MSR_IA32_LASTBRANCHTOIP:
2247 case MSR_IA32_LASTINTFROMIP:
2248 case MSR_IA32_LASTINTTOIP:
2251 case MSR_VM_HSAVE_PA:
2252 case MSR_K7_EVNTSEL0:
2253 case MSR_K7_PERFCTR0:
2254 case MSR_K8_INT_PENDING_MSG:
2255 case MSR_AMD64_NB_CFG:
2256 case MSR_FAM10H_MMIO_CONF_BASE:
2257 case MSR_AMD64_BU_CFG2:
2260 case MSR_P6_PERFCTR0:
2261 case MSR_P6_PERFCTR1:
2262 case MSR_P6_EVNTSEL0:
2263 case MSR_P6_EVNTSEL1:
2264 if (kvm_pmu_msr(vcpu, msr))
2265 return kvm_pmu_get_msr(vcpu, msr, pdata);
2268 case MSR_IA32_UCODE_REV:
2269 data = 0x100000000ULL;
2272 data = 0x500 | KVM_NR_VAR_MTRR;
2274 case 0x200 ... 0x2ff:
2275 return get_msr_mtrr(vcpu, msr, pdata);
2276 case 0xcd: /* fsb frequency */
2280 * MSR_EBC_FREQUENCY_ID
2281 * Conservative value valid for even the basic CPU models.
2282 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2283 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2284 * and 266MHz for model 3, or 4. Set Core Clock
2285 * Frequency to System Bus Frequency Ratio to 1 (bits
2286 * 31:24) even though these are only valid for CPU
2287 * models > 2, however guests may end up dividing or
2288 * multiplying by zero otherwise.
2290 case MSR_EBC_FREQUENCY_ID:
2293 case MSR_IA32_APICBASE:
2294 data = kvm_get_apic_base(vcpu);
2296 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2297 return kvm_x2apic_msr_read(vcpu, msr, pdata);
2299 case MSR_IA32_TSCDEADLINE:
2300 data = kvm_get_lapic_tscdeadline_msr(vcpu);
2302 case MSR_IA32_TSC_ADJUST:
2303 data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2305 case MSR_IA32_MISC_ENABLE:
2306 data = vcpu->arch.ia32_misc_enable_msr;
2308 case MSR_IA32_PERF_STATUS:
2309 /* TSC increment by tick */
2311 /* CPU multiplier */
2312 data |= (((uint64_t)4ULL) << 40);
2315 data = vcpu->arch.efer;
2317 case MSR_KVM_WALL_CLOCK:
2318 case MSR_KVM_WALL_CLOCK_NEW:
2319 data = vcpu->kvm->arch.wall_clock;
2321 case MSR_KVM_SYSTEM_TIME:
2322 case MSR_KVM_SYSTEM_TIME_NEW:
2323 data = vcpu->arch.time;
2325 case MSR_KVM_ASYNC_PF_EN:
2326 data = vcpu->arch.apf.msr_val;
2328 case MSR_KVM_STEAL_TIME:
2329 data = vcpu->arch.st.msr_val;
2331 case MSR_KVM_PV_EOI_EN:
2332 data = vcpu->arch.pv_eoi.msr_val;
2334 case MSR_IA32_P5_MC_ADDR:
2335 case MSR_IA32_P5_MC_TYPE:
2336 case MSR_IA32_MCG_CAP:
2337 case MSR_IA32_MCG_CTL:
2338 case MSR_IA32_MCG_STATUS:
2339 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2340 return get_msr_mce(vcpu, msr, pdata);
2341 case MSR_K7_CLK_CTL:
2343 * Provide expected ramp-up count for K7. All other
2344 * are set to zero, indicating minimum divisors for
2347 * This prevents guest kernels on AMD host with CPU
2348 * type 6, model 8 and higher from exploding due to
2349 * the rdmsr failing.
2353 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2354 if (kvm_hv_msr_partition_wide(msr)) {
2356 mutex_lock(&vcpu->kvm->lock);
2357 r = get_msr_hyperv_pw(vcpu, msr, pdata);
2358 mutex_unlock(&vcpu->kvm->lock);
2361 return get_msr_hyperv(vcpu, msr, pdata);
2363 case MSR_IA32_BBL_CR_CTL3:
2364 /* This legacy MSR exists but isn't fully documented in current
2365 * silicon. It is however accessed by winxp in very narrow
2366 * scenarios where it sets bit #19, itself documented as
2367 * a "reserved" bit. Best effort attempt to source coherent
2368 * read data here should the balance of the register be
2369 * interpreted by the guest:
2371 * L2 cache control register 3: 64GB range, 256KB size,
2372 * enabled, latency 0x1, configured
2376 case MSR_AMD64_OSVW_ID_LENGTH:
2377 if (!guest_cpuid_has_osvw(vcpu))
2379 data = vcpu->arch.osvw.length;
2381 case MSR_AMD64_OSVW_STATUS:
2382 if (!guest_cpuid_has_osvw(vcpu))
2384 data = vcpu->arch.osvw.status;
2387 if (kvm_pmu_msr(vcpu, msr))
2388 return kvm_pmu_get_msr(vcpu, msr, pdata);
2390 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2393 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2401 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2404 * Read or write a bunch of msrs. All parameters are kernel addresses.
2406 * @return number of msrs set successfully.
2408 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2409 struct kvm_msr_entry *entries,
2410 int (*do_msr)(struct kvm_vcpu *vcpu,
2411 unsigned index, u64 *data))
2415 idx = srcu_read_lock(&vcpu->kvm->srcu);
2416 for (i = 0; i < msrs->nmsrs; ++i)
2417 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2419 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2425 * Read or write a bunch of msrs. Parameters are user addresses.
2427 * @return number of msrs set successfully.
2429 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2430 int (*do_msr)(struct kvm_vcpu *vcpu,
2431 unsigned index, u64 *data),
2434 struct kvm_msrs msrs;
2435 struct kvm_msr_entry *entries;
2440 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2444 if (msrs.nmsrs >= MAX_IO_MSRS)
2447 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2448 entries = memdup_user(user_msrs->entries, size);
2449 if (IS_ERR(entries)) {
2450 r = PTR_ERR(entries);
2454 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2459 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2470 int kvm_dev_ioctl_check_extension(long ext)
2475 case KVM_CAP_IRQCHIP:
2477 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2478 case KVM_CAP_SET_TSS_ADDR:
2479 case KVM_CAP_EXT_CPUID:
2480 case KVM_CAP_CLOCKSOURCE:
2482 case KVM_CAP_NOP_IO_DELAY:
2483 case KVM_CAP_MP_STATE:
2484 case KVM_CAP_SYNC_MMU:
2485 case KVM_CAP_USER_NMI:
2486 case KVM_CAP_REINJECT_CONTROL:
2487 case KVM_CAP_IRQ_INJECT_STATUS:
2488 case KVM_CAP_ASSIGN_DEV_IRQ:
2490 case KVM_CAP_IOEVENTFD:
2492 case KVM_CAP_PIT_STATE2:
2493 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2494 case KVM_CAP_XEN_HVM:
2495 case KVM_CAP_ADJUST_CLOCK:
2496 case KVM_CAP_VCPU_EVENTS:
2497 case KVM_CAP_HYPERV:
2498 case KVM_CAP_HYPERV_VAPIC:
2499 case KVM_CAP_HYPERV_SPIN:
2500 case KVM_CAP_PCI_SEGMENT:
2501 case KVM_CAP_DEBUGREGS:
2502 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2504 case KVM_CAP_ASYNC_PF:
2505 case KVM_CAP_GET_TSC_KHZ:
2506 case KVM_CAP_PCI_2_3:
2507 case KVM_CAP_KVMCLOCK_CTRL:
2508 case KVM_CAP_READONLY_MEM:
2509 case KVM_CAP_IRQFD_RESAMPLE:
2512 case KVM_CAP_COALESCED_MMIO:
2513 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2516 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2518 case KVM_CAP_NR_VCPUS:
2519 r = KVM_SOFT_MAX_VCPUS;
2521 case KVM_CAP_MAX_VCPUS:
2524 case KVM_CAP_NR_MEMSLOTS:
2525 r = KVM_USER_MEM_SLOTS;
2527 case KVM_CAP_PV_MMU: /* obsolete */
2531 r = iommu_present(&pci_bus_type);
2534 r = KVM_MAX_MCE_BANKS;
2539 case KVM_CAP_TSC_CONTROL:
2540 r = kvm_has_tsc_control;
2542 case KVM_CAP_TSC_DEADLINE_TIMER:
2543 r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2553 long kvm_arch_dev_ioctl(struct file *filp,
2554 unsigned int ioctl, unsigned long arg)
2556 void __user *argp = (void __user *)arg;
2560 case KVM_GET_MSR_INDEX_LIST: {
2561 struct kvm_msr_list __user *user_msr_list = argp;
2562 struct kvm_msr_list msr_list;
2566 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2569 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2570 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2573 if (n < msr_list.nmsrs)
2576 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2577 num_msrs_to_save * sizeof(u32)))
2579 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2581 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2586 case KVM_GET_SUPPORTED_CPUID: {
2587 struct kvm_cpuid2 __user *cpuid_arg = argp;
2588 struct kvm_cpuid2 cpuid;
2591 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2593 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2594 cpuid_arg->entries);
2599 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2604 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2607 mce_cap = KVM_MCE_CAP_SUPPORTED;
2609 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2621 static void wbinvd_ipi(void *garbage)
2626 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2628 return vcpu->kvm->arch.iommu_domain &&
2629 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2632 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2634 /* Address WBINVD may be executed by guest */
2635 if (need_emulate_wbinvd(vcpu)) {
2636 if (kvm_x86_ops->has_wbinvd_exit())
2637 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2638 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2639 smp_call_function_single(vcpu->cpu,
2640 wbinvd_ipi, NULL, 1);
2643 kvm_x86_ops->vcpu_load(vcpu, cpu);
2645 /* Apply any externally detected TSC adjustments (due to suspend) */
2646 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2647 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2648 vcpu->arch.tsc_offset_adjustment = 0;
2649 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
2652 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2653 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2654 native_read_tsc() - vcpu->arch.last_host_tsc;
2656 mark_tsc_unstable("KVM discovered backwards TSC");
2657 if (check_tsc_unstable()) {
2658 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2659 vcpu->arch.last_guest_tsc);
2660 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2661 vcpu->arch.tsc_catchup = 1;
2664 * On a host with synchronized TSC, there is no need to update
2665 * kvmclock on vcpu->cpu migration
2667 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2668 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2669 if (vcpu->cpu != cpu)
2670 kvm_migrate_timers(vcpu);
2674 accumulate_steal_time(vcpu);
2675 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2678 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2680 kvm_x86_ops->vcpu_put(vcpu);
2681 kvm_put_guest_fpu(vcpu);
2682 vcpu->arch.last_host_tsc = native_read_tsc();
2685 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2686 struct kvm_lapic_state *s)
2688 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2693 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2694 struct kvm_lapic_state *s)
2696 kvm_apic_post_state_restore(vcpu, s);
2697 update_cr8_intercept(vcpu);
2702 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2703 struct kvm_interrupt *irq)
2705 if (irq->irq >= KVM_NR_INTERRUPTS)
2707 if (irqchip_in_kernel(vcpu->kvm))
2710 kvm_queue_interrupt(vcpu, irq->irq, false);
2711 kvm_make_request(KVM_REQ_EVENT, vcpu);
2716 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2718 kvm_inject_nmi(vcpu);
2723 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2724 struct kvm_tpr_access_ctl *tac)
2728 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2732 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2736 unsigned bank_num = mcg_cap & 0xff, bank;
2739 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2741 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2744 vcpu->arch.mcg_cap = mcg_cap;
2745 /* Init IA32_MCG_CTL to all 1s */
2746 if (mcg_cap & MCG_CTL_P)
2747 vcpu->arch.mcg_ctl = ~(u64)0;
2748 /* Init IA32_MCi_CTL to all 1s */
2749 for (bank = 0; bank < bank_num; bank++)
2750 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2755 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2756 struct kvm_x86_mce *mce)
2758 u64 mcg_cap = vcpu->arch.mcg_cap;
2759 unsigned bank_num = mcg_cap & 0xff;
2760 u64 *banks = vcpu->arch.mce_banks;
2762 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2765 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2766 * reporting is disabled
2768 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2769 vcpu->arch.mcg_ctl != ~(u64)0)
2771 banks += 4 * mce->bank;
2773 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2774 * reporting is disabled for the bank
2776 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2778 if (mce->status & MCI_STATUS_UC) {
2779 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2780 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2781 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2784 if (banks[1] & MCI_STATUS_VAL)
2785 mce->status |= MCI_STATUS_OVER;
2786 banks[2] = mce->addr;
2787 banks[3] = mce->misc;
2788 vcpu->arch.mcg_status = mce->mcg_status;
2789 banks[1] = mce->status;
2790 kvm_queue_exception(vcpu, MC_VECTOR);
2791 } else if (!(banks[1] & MCI_STATUS_VAL)
2792 || !(banks[1] & MCI_STATUS_UC)) {
2793 if (banks[1] & MCI_STATUS_VAL)
2794 mce->status |= MCI_STATUS_OVER;
2795 banks[2] = mce->addr;
2796 banks[3] = mce->misc;
2797 banks[1] = mce->status;
2799 banks[1] |= MCI_STATUS_OVER;
2803 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2804 struct kvm_vcpu_events *events)
2807 events->exception.injected =
2808 vcpu->arch.exception.pending &&
2809 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2810 events->exception.nr = vcpu->arch.exception.nr;
2811 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2812 events->exception.pad = 0;
2813 events->exception.error_code = vcpu->arch.exception.error_code;
2815 events->interrupt.injected =
2816 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2817 events->interrupt.nr = vcpu->arch.interrupt.nr;
2818 events->interrupt.soft = 0;
2819 events->interrupt.shadow =
2820 kvm_x86_ops->get_interrupt_shadow(vcpu,
2821 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2823 events->nmi.injected = vcpu->arch.nmi_injected;
2824 events->nmi.pending = vcpu->arch.nmi_pending != 0;
2825 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2826 events->nmi.pad = 0;
2828 events->sipi_vector = 0; /* never valid when reporting to user space */
2830 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2831 | KVM_VCPUEVENT_VALID_SHADOW);
2832 memset(&events->reserved, 0, sizeof(events->reserved));
2835 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2836 struct kvm_vcpu_events *events)
2838 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2839 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2840 | KVM_VCPUEVENT_VALID_SHADOW))
2844 vcpu->arch.exception.pending = events->exception.injected;
2845 vcpu->arch.exception.nr = events->exception.nr;
2846 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2847 vcpu->arch.exception.error_code = events->exception.error_code;
2849 vcpu->arch.interrupt.pending = events->interrupt.injected;
2850 vcpu->arch.interrupt.nr = events->interrupt.nr;
2851 vcpu->arch.interrupt.soft = events->interrupt.soft;
2852 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2853 kvm_x86_ops->set_interrupt_shadow(vcpu,
2854 events->interrupt.shadow);
2856 vcpu->arch.nmi_injected = events->nmi.injected;
2857 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2858 vcpu->arch.nmi_pending = events->nmi.pending;
2859 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2861 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
2862 kvm_vcpu_has_lapic(vcpu))
2863 vcpu->arch.apic->sipi_vector = events->sipi_vector;
2865 kvm_make_request(KVM_REQ_EVENT, vcpu);
2870 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2871 struct kvm_debugregs *dbgregs)
2873 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2874 dbgregs->dr6 = vcpu->arch.dr6;
2875 dbgregs->dr7 = vcpu->arch.dr7;
2877 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2880 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2881 struct kvm_debugregs *dbgregs)
2886 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2887 vcpu->arch.dr6 = dbgregs->dr6;
2888 vcpu->arch.dr7 = dbgregs->dr7;
2893 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2894 struct kvm_xsave *guest_xsave)
2897 memcpy(guest_xsave->region,
2898 &vcpu->arch.guest_fpu.state->xsave,
2901 memcpy(guest_xsave->region,
2902 &vcpu->arch.guest_fpu.state->fxsave,
2903 sizeof(struct i387_fxsave_struct));
2904 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2909 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2910 struct kvm_xsave *guest_xsave)
2913 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2916 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2917 guest_xsave->region, xstate_size);
2919 if (xstate_bv & ~XSTATE_FPSSE)
2921 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2922 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2927 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2928 struct kvm_xcrs *guest_xcrs)
2930 if (!cpu_has_xsave) {
2931 guest_xcrs->nr_xcrs = 0;
2935 guest_xcrs->nr_xcrs = 1;
2936 guest_xcrs->flags = 0;
2937 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2938 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2941 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2942 struct kvm_xcrs *guest_xcrs)
2949 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2952 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2953 /* Only support XCR0 currently */
2954 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2955 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2956 guest_xcrs->xcrs[0].value);
2965 * kvm_set_guest_paused() indicates to the guest kernel that it has been
2966 * stopped by the hypervisor. This function will be called from the host only.
2967 * EINVAL is returned when the host attempts to set the flag for a guest that
2968 * does not support pv clocks.
2970 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
2972 if (!vcpu->arch.pv_time_enabled)
2974 vcpu->arch.pvclock_set_guest_stopped_request = true;
2975 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2979 long kvm_arch_vcpu_ioctl(struct file *filp,
2980 unsigned int ioctl, unsigned long arg)
2982 struct kvm_vcpu *vcpu = filp->private_data;
2983 void __user *argp = (void __user *)arg;
2986 struct kvm_lapic_state *lapic;
2987 struct kvm_xsave *xsave;
2988 struct kvm_xcrs *xcrs;
2994 case KVM_GET_LAPIC: {
2996 if (!vcpu->arch.apic)
2998 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3003 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3007 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3012 case KVM_SET_LAPIC: {
3014 if (!vcpu->arch.apic)
3016 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3017 if (IS_ERR(u.lapic))
3018 return PTR_ERR(u.lapic);
3020 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3023 case KVM_INTERRUPT: {
3024 struct kvm_interrupt irq;
3027 if (copy_from_user(&irq, argp, sizeof irq))
3029 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3033 r = kvm_vcpu_ioctl_nmi(vcpu);
3036 case KVM_SET_CPUID: {
3037 struct kvm_cpuid __user *cpuid_arg = argp;
3038 struct kvm_cpuid cpuid;
3041 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3043 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3046 case KVM_SET_CPUID2: {
3047 struct kvm_cpuid2 __user *cpuid_arg = argp;
3048 struct kvm_cpuid2 cpuid;
3051 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3053 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3054 cpuid_arg->entries);
3057 case KVM_GET_CPUID2: {
3058 struct kvm_cpuid2 __user *cpuid_arg = argp;
3059 struct kvm_cpuid2 cpuid;
3062 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3064 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3065 cpuid_arg->entries);
3069 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3075 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3078 r = msr_io(vcpu, argp, do_set_msr, 0);
3080 case KVM_TPR_ACCESS_REPORTING: {
3081 struct kvm_tpr_access_ctl tac;
3084 if (copy_from_user(&tac, argp, sizeof tac))
3086 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3090 if (copy_to_user(argp, &tac, sizeof tac))
3095 case KVM_SET_VAPIC_ADDR: {
3096 struct kvm_vapic_addr va;
3099 if (!irqchip_in_kernel(vcpu->kvm))
3102 if (copy_from_user(&va, argp, sizeof va))
3105 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3108 case KVM_X86_SETUP_MCE: {
3112 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3114 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3117 case KVM_X86_SET_MCE: {
3118 struct kvm_x86_mce mce;
3121 if (copy_from_user(&mce, argp, sizeof mce))
3123 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3126 case KVM_GET_VCPU_EVENTS: {
3127 struct kvm_vcpu_events events;
3129 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3132 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3137 case KVM_SET_VCPU_EVENTS: {
3138 struct kvm_vcpu_events events;
3141 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3144 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3147 case KVM_GET_DEBUGREGS: {
3148 struct kvm_debugregs dbgregs;
3150 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3153 if (copy_to_user(argp, &dbgregs,
3154 sizeof(struct kvm_debugregs)))
3159 case KVM_SET_DEBUGREGS: {
3160 struct kvm_debugregs dbgregs;
3163 if (copy_from_user(&dbgregs, argp,
3164 sizeof(struct kvm_debugregs)))
3167 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3170 case KVM_GET_XSAVE: {
3171 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3176 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3179 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3184 case KVM_SET_XSAVE: {
3185 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3186 if (IS_ERR(u.xsave))
3187 return PTR_ERR(u.xsave);
3189 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3192 case KVM_GET_XCRS: {
3193 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3198 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3201 if (copy_to_user(argp, u.xcrs,
3202 sizeof(struct kvm_xcrs)))
3207 case KVM_SET_XCRS: {
3208 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3210 return PTR_ERR(u.xcrs);
3212 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3215 case KVM_SET_TSC_KHZ: {
3219 user_tsc_khz = (u32)arg;
3221 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3224 if (user_tsc_khz == 0)
3225 user_tsc_khz = tsc_khz;
3227 kvm_set_tsc_khz(vcpu, user_tsc_khz);
3232 case KVM_GET_TSC_KHZ: {
3233 r = vcpu->arch.virtual_tsc_khz;
3236 case KVM_KVMCLOCK_CTRL: {
3237 r = kvm_set_guest_paused(vcpu);
3248 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3250 return VM_FAULT_SIGBUS;
3253 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3257 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3259 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3263 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3266 kvm->arch.ept_identity_map_addr = ident_addr;
3270 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3271 u32 kvm_nr_mmu_pages)
3273 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3276 mutex_lock(&kvm->slots_lock);
3278 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3279 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3281 mutex_unlock(&kvm->slots_lock);
3285 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3287 return kvm->arch.n_max_mmu_pages;
3290 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3295 switch (chip->chip_id) {
3296 case KVM_IRQCHIP_PIC_MASTER:
3297 memcpy(&chip->chip.pic,
3298 &pic_irqchip(kvm)->pics[0],
3299 sizeof(struct kvm_pic_state));
3301 case KVM_IRQCHIP_PIC_SLAVE:
3302 memcpy(&chip->chip.pic,
3303 &pic_irqchip(kvm)->pics[1],
3304 sizeof(struct kvm_pic_state));
3306 case KVM_IRQCHIP_IOAPIC:
3307 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3316 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3321 switch (chip->chip_id) {
3322 case KVM_IRQCHIP_PIC_MASTER:
3323 spin_lock(&pic_irqchip(kvm)->lock);
3324 memcpy(&pic_irqchip(kvm)->pics[0],
3326 sizeof(struct kvm_pic_state));
3327 spin_unlock(&pic_irqchip(kvm)->lock);
3329 case KVM_IRQCHIP_PIC_SLAVE:
3330 spin_lock(&pic_irqchip(kvm)->lock);
3331 memcpy(&pic_irqchip(kvm)->pics[1],
3333 sizeof(struct kvm_pic_state));
3334 spin_unlock(&pic_irqchip(kvm)->lock);
3336 case KVM_IRQCHIP_IOAPIC:
3337 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3343 kvm_pic_update_irq(pic_irqchip(kvm));
3347 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3351 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3352 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3353 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3357 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3361 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3362 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3363 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3364 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3368 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3372 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3373 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3374 sizeof(ps->channels));
3375 ps->flags = kvm->arch.vpit->pit_state.flags;
3376 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3377 memset(&ps->reserved, 0, sizeof(ps->reserved));
3381 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3383 int r = 0, start = 0;
3384 u32 prev_legacy, cur_legacy;
3385 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3386 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3387 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3388 if (!prev_legacy && cur_legacy)
3390 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3391 sizeof(kvm->arch.vpit->pit_state.channels));
3392 kvm->arch.vpit->pit_state.flags = ps->flags;
3393 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3394 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3398 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3399 struct kvm_reinject_control *control)
3401 if (!kvm->arch.vpit)
3403 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3404 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3405 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3410 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3411 * @kvm: kvm instance
3412 * @log: slot id and address to which we copy the log
3414 * We need to keep it in mind that VCPU threads can write to the bitmap
3415 * concurrently. So, to avoid losing data, we keep the following order for
3418 * 1. Take a snapshot of the bit and clear it if needed.
3419 * 2. Write protect the corresponding page.
3420 * 3. Flush TLB's if needed.
3421 * 4. Copy the snapshot to the userspace.
3423 * Between 2 and 3, the guest may write to the page using the remaining TLB
3424 * entry. This is not a problem because the page will be reported dirty at
3425 * step 4 using the snapshot taken before and step 3 ensures that successive
3426 * writes will be logged for the next call.
3428 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3431 struct kvm_memory_slot *memslot;
3433 unsigned long *dirty_bitmap;
3434 unsigned long *dirty_bitmap_buffer;
3435 bool is_dirty = false;
3437 mutex_lock(&kvm->slots_lock);
3440 if (log->slot >= KVM_USER_MEM_SLOTS)
3443 memslot = id_to_memslot(kvm->memslots, log->slot);
3445 dirty_bitmap = memslot->dirty_bitmap;
3450 n = kvm_dirty_bitmap_bytes(memslot);
3452 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
3453 memset(dirty_bitmap_buffer, 0, n);
3455 spin_lock(&kvm->mmu_lock);
3457 for (i = 0; i < n / sizeof(long); i++) {
3461 if (!dirty_bitmap[i])
3466 mask = xchg(&dirty_bitmap[i], 0);
3467 dirty_bitmap_buffer[i] = mask;
3469 offset = i * BITS_PER_LONG;
3470 kvm_mmu_write_protect_pt_masked(kvm, memslot, offset, mask);
3473 kvm_flush_remote_tlbs(kvm);
3475 spin_unlock(&kvm->mmu_lock);
3478 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
3483 mutex_unlock(&kvm->slots_lock);
3487 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event)
3489 if (!irqchip_in_kernel(kvm))
3492 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3493 irq_event->irq, irq_event->level);
3497 long kvm_arch_vm_ioctl(struct file *filp,
3498 unsigned int ioctl, unsigned long arg)
3500 struct kvm *kvm = filp->private_data;
3501 void __user *argp = (void __user *)arg;
3504 * This union makes it completely explicit to gcc-3.x
3505 * that these two variables' stack usage should be
3506 * combined, not added together.
3509 struct kvm_pit_state ps;
3510 struct kvm_pit_state2 ps2;
3511 struct kvm_pit_config pit_config;
3515 case KVM_SET_TSS_ADDR:
3516 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3518 case KVM_SET_IDENTITY_MAP_ADDR: {
3522 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3524 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3527 case KVM_SET_NR_MMU_PAGES:
3528 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3530 case KVM_GET_NR_MMU_PAGES:
3531 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3533 case KVM_CREATE_IRQCHIP: {
3534 struct kvm_pic *vpic;
3536 mutex_lock(&kvm->lock);
3539 goto create_irqchip_unlock;
3541 if (atomic_read(&kvm->online_vcpus))
3542 goto create_irqchip_unlock;
3544 vpic = kvm_create_pic(kvm);
3546 r = kvm_ioapic_init(kvm);
3548 mutex_lock(&kvm->slots_lock);
3549 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3551 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3553 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3555 mutex_unlock(&kvm->slots_lock);
3557 goto create_irqchip_unlock;
3560 goto create_irqchip_unlock;
3562 kvm->arch.vpic = vpic;
3564 r = kvm_setup_default_irq_routing(kvm);
3566 mutex_lock(&kvm->slots_lock);
3567 mutex_lock(&kvm->irq_lock);
3568 kvm_ioapic_destroy(kvm);
3569 kvm_destroy_pic(kvm);
3570 mutex_unlock(&kvm->irq_lock);
3571 mutex_unlock(&kvm->slots_lock);
3573 create_irqchip_unlock:
3574 mutex_unlock(&kvm->lock);
3577 case KVM_CREATE_PIT:
3578 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3580 case KVM_CREATE_PIT2:
3582 if (copy_from_user(&u.pit_config, argp,
3583 sizeof(struct kvm_pit_config)))
3586 mutex_lock(&kvm->slots_lock);
3589 goto create_pit_unlock;
3591 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3595 mutex_unlock(&kvm->slots_lock);
3597 case KVM_GET_IRQCHIP: {
3598 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3599 struct kvm_irqchip *chip;
3601 chip = memdup_user(argp, sizeof(*chip));
3608 if (!irqchip_in_kernel(kvm))
3609 goto get_irqchip_out;
3610 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3612 goto get_irqchip_out;
3614 if (copy_to_user(argp, chip, sizeof *chip))
3615 goto get_irqchip_out;
3621 case KVM_SET_IRQCHIP: {
3622 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3623 struct kvm_irqchip *chip;
3625 chip = memdup_user(argp, sizeof(*chip));
3632 if (!irqchip_in_kernel(kvm))
3633 goto set_irqchip_out;
3634 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3636 goto set_irqchip_out;
3644 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3647 if (!kvm->arch.vpit)
3649 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3653 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3660 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3663 if (!kvm->arch.vpit)
3665 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3668 case KVM_GET_PIT2: {
3670 if (!kvm->arch.vpit)
3672 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3676 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3681 case KVM_SET_PIT2: {
3683 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3686 if (!kvm->arch.vpit)
3688 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3691 case KVM_REINJECT_CONTROL: {
3692 struct kvm_reinject_control control;
3694 if (copy_from_user(&control, argp, sizeof(control)))
3696 r = kvm_vm_ioctl_reinject(kvm, &control);
3699 case KVM_XEN_HVM_CONFIG: {
3701 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3702 sizeof(struct kvm_xen_hvm_config)))
3705 if (kvm->arch.xen_hvm_config.flags)
3710 case KVM_SET_CLOCK: {
3711 struct kvm_clock_data user_ns;
3716 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3724 local_irq_disable();
3725 now_ns = get_kernel_ns();
3726 delta = user_ns.clock - now_ns;
3728 kvm->arch.kvmclock_offset = delta;
3731 case KVM_GET_CLOCK: {
3732 struct kvm_clock_data user_ns;
3735 local_irq_disable();
3736 now_ns = get_kernel_ns();
3737 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3740 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3743 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3756 static void kvm_init_msr_list(void)
3761 /* skip the first msrs in the list. KVM-specific */
3762 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3763 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3766 msrs_to_save[j] = msrs_to_save[i];
3769 num_msrs_to_save = j;
3772 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3780 if (!(vcpu->arch.apic &&
3781 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3782 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3793 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3800 if (!(vcpu->arch.apic &&
3801 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3802 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3804 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3814 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3815 struct kvm_segment *var, int seg)
3817 kvm_x86_ops->set_segment(vcpu, var, seg);
3820 void kvm_get_segment(struct kvm_vcpu *vcpu,
3821 struct kvm_segment *var, int seg)
3823 kvm_x86_ops->get_segment(vcpu, var, seg);
3826 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3829 struct x86_exception exception;
3831 BUG_ON(!mmu_is_nested(vcpu));
3833 /* NPT walks are always user-walks */
3834 access |= PFERR_USER_MASK;
3835 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3840 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3841 struct x86_exception *exception)
3843 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3844 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3847 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3848 struct x86_exception *exception)
3850 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3851 access |= PFERR_FETCH_MASK;
3852 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3855 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3856 struct x86_exception *exception)
3858 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3859 access |= PFERR_WRITE_MASK;
3860 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3863 /* uses this to access any guest's mapped memory without checking CPL */
3864 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3865 struct x86_exception *exception)
3867 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3870 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3871 struct kvm_vcpu *vcpu, u32 access,
3872 struct x86_exception *exception)
3875 int r = X86EMUL_CONTINUE;
3878 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3880 unsigned offset = addr & (PAGE_SIZE-1);
3881 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3884 if (gpa == UNMAPPED_GVA)
3885 return X86EMUL_PROPAGATE_FAULT;
3886 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3888 r = X86EMUL_IO_NEEDED;
3900 /* used for instruction fetching */
3901 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3902 gva_t addr, void *val, unsigned int bytes,
3903 struct x86_exception *exception)
3905 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3906 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3908 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3909 access | PFERR_FETCH_MASK,
3913 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3914 gva_t addr, void *val, unsigned int bytes,
3915 struct x86_exception *exception)
3917 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3918 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3920 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3923 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
3925 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3926 gva_t addr, void *val, unsigned int bytes,
3927 struct x86_exception *exception)
3929 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3930 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3933 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3934 gva_t addr, void *val,
3936 struct x86_exception *exception)
3938 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3940 int r = X86EMUL_CONTINUE;
3943 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3946 unsigned offset = addr & (PAGE_SIZE-1);
3947 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3950 if (gpa == UNMAPPED_GVA)
3951 return X86EMUL_PROPAGATE_FAULT;
3952 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3954 r = X86EMUL_IO_NEEDED;
3965 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
3967 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
3968 gpa_t *gpa, struct x86_exception *exception,
3971 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
3972 | (write ? PFERR_WRITE_MASK : 0);
3974 if (vcpu_match_mmio_gva(vcpu, gva)
3975 && !permission_fault(vcpu->arch.walk_mmu, vcpu->arch.access, access)) {
3976 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
3977 (gva & (PAGE_SIZE - 1));
3978 trace_vcpu_match_mmio(gva, *gpa, write, false);
3982 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3984 if (*gpa == UNMAPPED_GVA)
3987 /* For APIC access vmexit */
3988 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3991 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
3992 trace_vcpu_match_mmio(gva, *gpa, write, true);
3999 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4000 const void *val, int bytes)
4004 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4007 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4011 struct read_write_emulator_ops {
4012 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4014 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4015 void *val, int bytes);
4016 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4017 int bytes, void *val);
4018 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4019 void *val, int bytes);
4023 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4025 if (vcpu->mmio_read_completed) {
4026 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4027 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4028 vcpu->mmio_read_completed = 0;
4035 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4036 void *val, int bytes)
4038 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4041 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4042 void *val, int bytes)
4044 return emulator_write_phys(vcpu, gpa, val, bytes);
4047 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4049 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4050 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4053 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4054 void *val, int bytes)
4056 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4057 return X86EMUL_IO_NEEDED;
4060 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4061 void *val, int bytes)
4063 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4065 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4066 return X86EMUL_CONTINUE;
4069 static const struct read_write_emulator_ops read_emultor = {
4070 .read_write_prepare = read_prepare,
4071 .read_write_emulate = read_emulate,
4072 .read_write_mmio = vcpu_mmio_read,
4073 .read_write_exit_mmio = read_exit_mmio,
4076 static const struct read_write_emulator_ops write_emultor = {
4077 .read_write_emulate = write_emulate,
4078 .read_write_mmio = write_mmio,
4079 .read_write_exit_mmio = write_exit_mmio,
4083 static int emulator_read_write_onepage(unsigned long addr, void *val,
4085 struct x86_exception *exception,
4086 struct kvm_vcpu *vcpu,
4087 const struct read_write_emulator_ops *ops)
4091 bool write = ops->write;
4092 struct kvm_mmio_fragment *frag;
4094 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4097 return X86EMUL_PROPAGATE_FAULT;
4099 /* For APIC access vmexit */
4103 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4104 return X86EMUL_CONTINUE;
4108 * Is this MMIO handled locally?
4110 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4111 if (handled == bytes)
4112 return X86EMUL_CONTINUE;
4118 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4119 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4123 return X86EMUL_CONTINUE;
4126 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4127 void *val, unsigned int bytes,
4128 struct x86_exception *exception,
4129 const struct read_write_emulator_ops *ops)
4131 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4135 if (ops->read_write_prepare &&
4136 ops->read_write_prepare(vcpu, val, bytes))
4137 return X86EMUL_CONTINUE;
4139 vcpu->mmio_nr_fragments = 0;
4141 /* Crossing a page boundary? */
4142 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4145 now = -addr & ~PAGE_MASK;
4146 rc = emulator_read_write_onepage(addr, val, now, exception,
4149 if (rc != X86EMUL_CONTINUE)
4156 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4158 if (rc != X86EMUL_CONTINUE)
4161 if (!vcpu->mmio_nr_fragments)
4164 gpa = vcpu->mmio_fragments[0].gpa;
4166 vcpu->mmio_needed = 1;
4167 vcpu->mmio_cur_fragment = 0;
4169 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4170 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4171 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4172 vcpu->run->mmio.phys_addr = gpa;
4174 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4177 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4181 struct x86_exception *exception)
4183 return emulator_read_write(ctxt, addr, val, bytes,
4184 exception, &read_emultor);
4187 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4191 struct x86_exception *exception)
4193 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4194 exception, &write_emultor);
4197 #define CMPXCHG_TYPE(t, ptr, old, new) \
4198 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4200 #ifdef CONFIG_X86_64
4201 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4203 # define CMPXCHG64(ptr, old, new) \
4204 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4207 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4212 struct x86_exception *exception)
4214 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4220 /* guests cmpxchg8b have to be emulated atomically */
4221 if (bytes > 8 || (bytes & (bytes - 1)))
4224 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4226 if (gpa == UNMAPPED_GVA ||
4227 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4230 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4233 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4234 if (is_error_page(page))
4237 kaddr = kmap_atomic(page);
4238 kaddr += offset_in_page(gpa);
4241 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4244 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4247 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4250 exchanged = CMPXCHG64(kaddr, old, new);
4255 kunmap_atomic(kaddr);
4256 kvm_release_page_dirty(page);
4259 return X86EMUL_CMPXCHG_FAILED;
4261 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4263 return X86EMUL_CONTINUE;
4266 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4268 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4271 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4273 /* TODO: String I/O for in kernel device */
4276 if (vcpu->arch.pio.in)
4277 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4278 vcpu->arch.pio.size, pd);
4280 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4281 vcpu->arch.pio.port, vcpu->arch.pio.size,
4286 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4287 unsigned short port, void *val,
4288 unsigned int count, bool in)
4290 trace_kvm_pio(!in, port, size, count);
4292 vcpu->arch.pio.port = port;
4293 vcpu->arch.pio.in = in;
4294 vcpu->arch.pio.count = count;
4295 vcpu->arch.pio.size = size;
4297 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4298 vcpu->arch.pio.count = 0;
4302 vcpu->run->exit_reason = KVM_EXIT_IO;
4303 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4304 vcpu->run->io.size = size;
4305 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4306 vcpu->run->io.count = count;
4307 vcpu->run->io.port = port;
4312 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4313 int size, unsigned short port, void *val,
4316 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4319 if (vcpu->arch.pio.count)
4322 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4325 memcpy(val, vcpu->arch.pio_data, size * count);
4326 vcpu->arch.pio.count = 0;
4333 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4334 int size, unsigned short port,
4335 const void *val, unsigned int count)
4337 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4339 memcpy(vcpu->arch.pio_data, val, size * count);
4340 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4343 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4345 return kvm_x86_ops->get_segment_base(vcpu, seg);
4348 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4350 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4353 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4355 if (!need_emulate_wbinvd(vcpu))
4356 return X86EMUL_CONTINUE;
4358 if (kvm_x86_ops->has_wbinvd_exit()) {
4359 int cpu = get_cpu();
4361 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4362 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4363 wbinvd_ipi, NULL, 1);
4365 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4368 return X86EMUL_CONTINUE;
4370 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4372 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4374 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4377 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4379 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4382 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4385 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4388 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4390 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4393 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4395 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4396 unsigned long value;
4400 value = kvm_read_cr0(vcpu);
4403 value = vcpu->arch.cr2;
4406 value = kvm_read_cr3(vcpu);
4409 value = kvm_read_cr4(vcpu);
4412 value = kvm_get_cr8(vcpu);
4415 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4422 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4424 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4429 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4432 vcpu->arch.cr2 = val;
4435 res = kvm_set_cr3(vcpu, val);
4438 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4441 res = kvm_set_cr8(vcpu, val);
4444 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4451 static void emulator_set_rflags(struct x86_emulate_ctxt *ctxt, ulong val)
4453 kvm_set_rflags(emul_to_vcpu(ctxt), val);
4456 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4458 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4461 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4463 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4466 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4468 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4471 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4473 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4476 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4478 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4481 static unsigned long emulator_get_cached_segment_base(
4482 struct x86_emulate_ctxt *ctxt, int seg)
4484 return get_segment_base(emul_to_vcpu(ctxt), seg);
4487 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4488 struct desc_struct *desc, u32 *base3,
4491 struct kvm_segment var;
4493 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4494 *selector = var.selector;
4497 memset(desc, 0, sizeof(*desc));
4503 set_desc_limit(desc, var.limit);
4504 set_desc_base(desc, (unsigned long)var.base);
4505 #ifdef CONFIG_X86_64
4507 *base3 = var.base >> 32;
4509 desc->type = var.type;
4511 desc->dpl = var.dpl;
4512 desc->p = var.present;
4513 desc->avl = var.avl;
4521 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4522 struct desc_struct *desc, u32 base3,
4525 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4526 struct kvm_segment var;
4528 var.selector = selector;
4529 var.base = get_desc_base(desc);
4530 #ifdef CONFIG_X86_64
4531 var.base |= ((u64)base3) << 32;
4533 var.limit = get_desc_limit(desc);
4535 var.limit = (var.limit << 12) | 0xfff;
4536 var.type = desc->type;
4537 var.present = desc->p;
4538 var.dpl = desc->dpl;
4543 var.avl = desc->avl;
4544 var.present = desc->p;
4545 var.unusable = !var.present;
4548 kvm_set_segment(vcpu, &var, seg);
4552 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4553 u32 msr_index, u64 *pdata)
4555 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4558 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4559 u32 msr_index, u64 data)
4561 struct msr_data msr;
4564 msr.index = msr_index;
4565 msr.host_initiated = false;
4566 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4569 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4570 u32 pmc, u64 *pdata)
4572 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4575 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4577 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4580 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4583 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4585 * CR0.TS may reference the host fpu state, not the guest fpu state,
4586 * so it may be clear at this point.
4591 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4596 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4597 struct x86_instruction_info *info,
4598 enum x86_intercept_stage stage)
4600 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4603 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4604 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4606 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4609 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4611 return kvm_register_read(emul_to_vcpu(ctxt), reg);
4614 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4616 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4619 static const struct x86_emulate_ops emulate_ops = {
4620 .read_gpr = emulator_read_gpr,
4621 .write_gpr = emulator_write_gpr,
4622 .read_std = kvm_read_guest_virt_system,
4623 .write_std = kvm_write_guest_virt_system,
4624 .fetch = kvm_fetch_guest_virt,
4625 .read_emulated = emulator_read_emulated,
4626 .write_emulated = emulator_write_emulated,
4627 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4628 .invlpg = emulator_invlpg,
4629 .pio_in_emulated = emulator_pio_in_emulated,
4630 .pio_out_emulated = emulator_pio_out_emulated,
4631 .get_segment = emulator_get_segment,
4632 .set_segment = emulator_set_segment,
4633 .get_cached_segment_base = emulator_get_cached_segment_base,
4634 .get_gdt = emulator_get_gdt,
4635 .get_idt = emulator_get_idt,
4636 .set_gdt = emulator_set_gdt,
4637 .set_idt = emulator_set_idt,
4638 .get_cr = emulator_get_cr,
4639 .set_cr = emulator_set_cr,
4640 .set_rflags = emulator_set_rflags,
4641 .cpl = emulator_get_cpl,
4642 .get_dr = emulator_get_dr,
4643 .set_dr = emulator_set_dr,
4644 .set_msr = emulator_set_msr,
4645 .get_msr = emulator_get_msr,
4646 .read_pmc = emulator_read_pmc,
4647 .halt = emulator_halt,
4648 .wbinvd = emulator_wbinvd,
4649 .fix_hypercall = emulator_fix_hypercall,
4650 .get_fpu = emulator_get_fpu,
4651 .put_fpu = emulator_put_fpu,
4652 .intercept = emulator_intercept,
4653 .get_cpuid = emulator_get_cpuid,
4656 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4658 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4660 * an sti; sti; sequence only disable interrupts for the first
4661 * instruction. So, if the last instruction, be it emulated or
4662 * not, left the system with the INT_STI flag enabled, it
4663 * means that the last instruction is an sti. We should not
4664 * leave the flag on in this case. The same goes for mov ss
4666 if (!(int_shadow & mask))
4667 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4670 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4672 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4673 if (ctxt->exception.vector == PF_VECTOR)
4674 kvm_propagate_fault(vcpu, &ctxt->exception);
4675 else if (ctxt->exception.error_code_valid)
4676 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4677 ctxt->exception.error_code);
4679 kvm_queue_exception(vcpu, ctxt->exception.vector);
4682 static void init_decode_cache(struct x86_emulate_ctxt *ctxt)
4684 memset(&ctxt->twobyte, 0,
4685 (void *)&ctxt->_regs - (void *)&ctxt->twobyte);
4687 ctxt->fetch.start = 0;
4688 ctxt->fetch.end = 0;
4689 ctxt->io_read.pos = 0;
4690 ctxt->io_read.end = 0;
4691 ctxt->mem_read.pos = 0;
4692 ctxt->mem_read.end = 0;
4695 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4697 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4700 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4702 ctxt->eflags = kvm_get_rflags(vcpu);
4703 ctxt->eip = kvm_rip_read(vcpu);
4704 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4705 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4706 cs_l ? X86EMUL_MODE_PROT64 :
4707 cs_db ? X86EMUL_MODE_PROT32 :
4708 X86EMUL_MODE_PROT16;
4709 ctxt->guest_mode = is_guest_mode(vcpu);
4711 init_decode_cache(ctxt);
4712 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4715 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4717 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4720 init_emulate_ctxt(vcpu);
4724 ctxt->_eip = ctxt->eip + inc_eip;
4725 ret = emulate_int_real(ctxt, irq);
4727 if (ret != X86EMUL_CONTINUE)
4728 return EMULATE_FAIL;
4730 ctxt->eip = ctxt->_eip;
4731 kvm_rip_write(vcpu, ctxt->eip);
4732 kvm_set_rflags(vcpu, ctxt->eflags);
4734 if (irq == NMI_VECTOR)
4735 vcpu->arch.nmi_pending = 0;
4737 vcpu->arch.interrupt.pending = false;
4739 return EMULATE_DONE;
4741 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4743 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4745 int r = EMULATE_DONE;
4747 ++vcpu->stat.insn_emulation_fail;
4748 trace_kvm_emulate_insn_failed(vcpu);
4749 if (!is_guest_mode(vcpu)) {
4750 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4751 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4752 vcpu->run->internal.ndata = 0;
4755 kvm_queue_exception(vcpu, UD_VECTOR);
4760 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
4761 bool write_fault_to_shadow_pgtable)
4766 if (!vcpu->arch.mmu.direct_map) {
4768 * Write permission should be allowed since only
4769 * write access need to be emulated.
4771 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4774 * If the mapping is invalid in guest, let cpu retry
4775 * it to generate fault.
4777 if (gpa == UNMAPPED_GVA)
4782 * Do not retry the unhandleable instruction if it faults on the
4783 * readonly host memory, otherwise it will goto a infinite loop:
4784 * retry instruction -> write #PF -> emulation fail -> retry
4785 * instruction -> ...
4787 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
4790 * If the instruction failed on the error pfn, it can not be fixed,
4791 * report the error to userspace.
4793 if (is_error_noslot_pfn(pfn))
4796 kvm_release_pfn_clean(pfn);
4798 /* The instructions are well-emulated on direct mmu. */
4799 if (vcpu->arch.mmu.direct_map) {
4800 unsigned int indirect_shadow_pages;
4802 spin_lock(&vcpu->kvm->mmu_lock);
4803 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
4804 spin_unlock(&vcpu->kvm->mmu_lock);
4806 if (indirect_shadow_pages)
4807 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4813 * if emulation was due to access to shadowed page table
4814 * and it failed try to unshadow page and re-enter the
4815 * guest to let CPU execute the instruction.
4817 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4820 * If the access faults on its page table, it can not
4821 * be fixed by unprotecting shadow page and it should
4822 * be reported to userspace.
4824 return !write_fault_to_shadow_pgtable;
4827 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
4828 unsigned long cr2, int emulation_type)
4830 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4831 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
4833 last_retry_eip = vcpu->arch.last_retry_eip;
4834 last_retry_addr = vcpu->arch.last_retry_addr;
4837 * If the emulation is caused by #PF and it is non-page_table
4838 * writing instruction, it means the VM-EXIT is caused by shadow
4839 * page protected, we can zap the shadow page and retry this
4840 * instruction directly.
4842 * Note: if the guest uses a non-page-table modifying instruction
4843 * on the PDE that points to the instruction, then we will unmap
4844 * the instruction and go to an infinite loop. So, we cache the
4845 * last retried eip and the last fault address, if we meet the eip
4846 * and the address again, we can break out of the potential infinite
4849 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
4851 if (!(emulation_type & EMULTYPE_RETRY))
4854 if (x86_page_table_writing_insn(ctxt))
4857 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
4860 vcpu->arch.last_retry_eip = ctxt->eip;
4861 vcpu->arch.last_retry_addr = cr2;
4863 if (!vcpu->arch.mmu.direct_map)
4864 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4866 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4871 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
4872 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
4874 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4881 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4882 bool writeback = true;
4883 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
4886 * Clear write_fault_to_shadow_pgtable here to ensure it is
4889 vcpu->arch.write_fault_to_shadow_pgtable = false;
4890 kvm_clear_exception_queue(vcpu);
4892 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4893 init_emulate_ctxt(vcpu);
4894 ctxt->interruptibility = 0;
4895 ctxt->have_exception = false;
4896 ctxt->perm_ok = false;
4898 ctxt->only_vendor_specific_insn
4899 = emulation_type & EMULTYPE_TRAP_UD;
4901 r = x86_decode_insn(ctxt, insn, insn_len);
4903 trace_kvm_emulate_insn_start(vcpu);
4904 ++vcpu->stat.insn_emulation;
4905 if (r != EMULATION_OK) {
4906 if (emulation_type & EMULTYPE_TRAP_UD)
4907 return EMULATE_FAIL;
4908 if (reexecute_instruction(vcpu, cr2,
4909 write_fault_to_spt))
4910 return EMULATE_DONE;
4911 if (emulation_type & EMULTYPE_SKIP)
4912 return EMULATE_FAIL;
4913 return handle_emulation_failure(vcpu);
4917 if (emulation_type & EMULTYPE_SKIP) {
4918 kvm_rip_write(vcpu, ctxt->_eip);
4919 return EMULATE_DONE;
4922 if (retry_instruction(ctxt, cr2, emulation_type))
4923 return EMULATE_DONE;
4925 /* this is needed for vmware backdoor interface to work since it
4926 changes registers values during IO operation */
4927 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4928 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4929 emulator_invalidate_register_cache(ctxt);
4933 r = x86_emulate_insn(ctxt);
4935 if (r == EMULATION_INTERCEPTED)
4936 return EMULATE_DONE;
4938 if (r == EMULATION_FAILED) {
4939 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt))
4940 return EMULATE_DONE;
4942 return handle_emulation_failure(vcpu);
4945 if (ctxt->have_exception) {
4946 inject_emulated_exception(vcpu);
4948 } else if (vcpu->arch.pio.count) {
4949 if (!vcpu->arch.pio.in)
4950 vcpu->arch.pio.count = 0;
4953 vcpu->arch.complete_userspace_io = complete_emulated_pio;
4955 r = EMULATE_DO_MMIO;
4956 } else if (vcpu->mmio_needed) {
4957 if (!vcpu->mmio_is_write)
4959 r = EMULATE_DO_MMIO;
4960 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
4961 } else if (r == EMULATION_RESTART)
4967 toggle_interruptibility(vcpu, ctxt->interruptibility);
4968 kvm_set_rflags(vcpu, ctxt->eflags);
4969 kvm_make_request(KVM_REQ_EVENT, vcpu);
4970 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4971 kvm_rip_write(vcpu, ctxt->eip);
4973 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
4977 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4979 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4981 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4982 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4983 size, port, &val, 1);
4984 /* do not return to emulator after return from userspace */
4985 vcpu->arch.pio.count = 0;
4988 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4990 static void tsc_bad(void *info)
4992 __this_cpu_write(cpu_tsc_khz, 0);
4995 static void tsc_khz_changed(void *data)
4997 struct cpufreq_freqs *freq = data;
4998 unsigned long khz = 0;
5002 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5003 khz = cpufreq_quick_get(raw_smp_processor_id());
5006 __this_cpu_write(cpu_tsc_khz, khz);
5009 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5012 struct cpufreq_freqs *freq = data;
5014 struct kvm_vcpu *vcpu;
5015 int i, send_ipi = 0;
5018 * We allow guests to temporarily run on slowing clocks,
5019 * provided we notify them after, or to run on accelerating
5020 * clocks, provided we notify them before. Thus time never
5023 * However, we have a problem. We can't atomically update
5024 * the frequency of a given CPU from this function; it is
5025 * merely a notifier, which can be called from any CPU.
5026 * Changing the TSC frequency at arbitrary points in time
5027 * requires a recomputation of local variables related to
5028 * the TSC for each VCPU. We must flag these local variables
5029 * to be updated and be sure the update takes place with the
5030 * new frequency before any guests proceed.
5032 * Unfortunately, the combination of hotplug CPU and frequency
5033 * change creates an intractable locking scenario; the order
5034 * of when these callouts happen is undefined with respect to
5035 * CPU hotplug, and they can race with each other. As such,
5036 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5037 * undefined; you can actually have a CPU frequency change take
5038 * place in between the computation of X and the setting of the
5039 * variable. To protect against this problem, all updates of
5040 * the per_cpu tsc_khz variable are done in an interrupt
5041 * protected IPI, and all callers wishing to update the value
5042 * must wait for a synchronous IPI to complete (which is trivial
5043 * if the caller is on the CPU already). This establishes the
5044 * necessary total order on variable updates.
5046 * Note that because a guest time update may take place
5047 * anytime after the setting of the VCPU's request bit, the
5048 * correct TSC value must be set before the request. However,
5049 * to ensure the update actually makes it to any guest which
5050 * starts running in hardware virtualization between the set
5051 * and the acquisition of the spinlock, we must also ping the
5052 * CPU after setting the request bit.
5056 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5058 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5061 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5063 raw_spin_lock(&kvm_lock);
5064 list_for_each_entry(kvm, &vm_list, vm_list) {
5065 kvm_for_each_vcpu(i, vcpu, kvm) {
5066 if (vcpu->cpu != freq->cpu)
5068 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5069 if (vcpu->cpu != smp_processor_id())
5073 raw_spin_unlock(&kvm_lock);
5075 if (freq->old < freq->new && send_ipi) {
5077 * We upscale the frequency. Must make the guest
5078 * doesn't see old kvmclock values while running with
5079 * the new frequency, otherwise we risk the guest sees
5080 * time go backwards.
5082 * In case we update the frequency for another cpu
5083 * (which might be in guest context) send an interrupt
5084 * to kick the cpu out of guest context. Next time
5085 * guest context is entered kvmclock will be updated,
5086 * so the guest will not see stale values.
5088 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5093 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5094 .notifier_call = kvmclock_cpufreq_notifier
5097 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5098 unsigned long action, void *hcpu)
5100 unsigned int cpu = (unsigned long)hcpu;
5104 case CPU_DOWN_FAILED:
5105 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5107 case CPU_DOWN_PREPARE:
5108 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5114 static struct notifier_block kvmclock_cpu_notifier_block = {
5115 .notifier_call = kvmclock_cpu_notifier,
5116 .priority = -INT_MAX
5119 static void kvm_timer_init(void)
5123 max_tsc_khz = tsc_khz;
5124 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5125 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5126 #ifdef CONFIG_CPU_FREQ
5127 struct cpufreq_policy policy;
5128 memset(&policy, 0, sizeof(policy));
5130 cpufreq_get_policy(&policy, cpu);
5131 if (policy.cpuinfo.max_freq)
5132 max_tsc_khz = policy.cpuinfo.max_freq;
5135 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5136 CPUFREQ_TRANSITION_NOTIFIER);
5138 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5139 for_each_online_cpu(cpu)
5140 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5143 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5145 int kvm_is_in_guest(void)
5147 return __this_cpu_read(current_vcpu) != NULL;
5150 static int kvm_is_user_mode(void)
5154 if (__this_cpu_read(current_vcpu))
5155 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5157 return user_mode != 0;
5160 static unsigned long kvm_get_guest_ip(void)
5162 unsigned long ip = 0;
5164 if (__this_cpu_read(current_vcpu))
5165 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5170 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5171 .is_in_guest = kvm_is_in_guest,
5172 .is_user_mode = kvm_is_user_mode,
5173 .get_guest_ip = kvm_get_guest_ip,
5176 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5178 __this_cpu_write(current_vcpu, vcpu);
5180 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5182 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5184 __this_cpu_write(current_vcpu, NULL);
5186 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5188 static void kvm_set_mmio_spte_mask(void)
5191 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5194 * Set the reserved bits and the present bit of an paging-structure
5195 * entry to generate page fault with PFER.RSV = 1.
5197 mask = ((1ull << (62 - maxphyaddr + 1)) - 1) << maxphyaddr;
5200 #ifdef CONFIG_X86_64
5202 * If reserved bit is not supported, clear the present bit to disable
5205 if (maxphyaddr == 52)
5209 kvm_mmu_set_mmio_spte_mask(mask);
5212 #ifdef CONFIG_X86_64
5213 static void pvclock_gtod_update_fn(struct work_struct *work)
5217 struct kvm_vcpu *vcpu;
5220 raw_spin_lock(&kvm_lock);
5221 list_for_each_entry(kvm, &vm_list, vm_list)
5222 kvm_for_each_vcpu(i, vcpu, kvm)
5223 set_bit(KVM_REQ_MASTERCLOCK_UPDATE, &vcpu->requests);
5224 atomic_set(&kvm_guest_has_master_clock, 0);
5225 raw_spin_unlock(&kvm_lock);
5228 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5231 * Notification about pvclock gtod data update.
5233 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5236 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5237 struct timekeeper *tk = priv;
5239 update_pvclock_gtod(tk);
5241 /* disable master clock if host does not trust, or does not
5242 * use, TSC clocksource
5244 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5245 atomic_read(&kvm_guest_has_master_clock) != 0)
5246 queue_work(system_long_wq, &pvclock_gtod_work);
5251 static struct notifier_block pvclock_gtod_notifier = {
5252 .notifier_call = pvclock_gtod_notify,
5256 int kvm_arch_init(void *opaque)
5259 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
5262 printk(KERN_ERR "kvm: already loaded the other module\n");
5267 if (!ops->cpu_has_kvm_support()) {
5268 printk(KERN_ERR "kvm: no hardware support\n");
5272 if (ops->disabled_by_bios()) {
5273 printk(KERN_ERR "kvm: disabled by bios\n");
5279 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5281 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5285 r = kvm_mmu_module_init();
5287 goto out_free_percpu;
5289 kvm_set_mmio_spte_mask();
5290 kvm_init_msr_list();
5293 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5294 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5298 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5301 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5304 #ifdef CONFIG_X86_64
5305 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5311 free_percpu(shared_msrs);
5316 void kvm_arch_exit(void)
5318 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5320 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5321 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5322 CPUFREQ_TRANSITION_NOTIFIER);
5323 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5324 #ifdef CONFIG_X86_64
5325 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5328 kvm_mmu_module_exit();
5329 free_percpu(shared_msrs);
5332 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5334 ++vcpu->stat.halt_exits;
5335 if (irqchip_in_kernel(vcpu->kvm)) {
5336 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5339 vcpu->run->exit_reason = KVM_EXIT_HLT;
5343 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5345 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5347 u64 param, ingpa, outgpa, ret;
5348 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5349 bool fast, longmode;
5353 * hypercall generates UD from non zero cpl and real mode
5356 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5357 kvm_queue_exception(vcpu, UD_VECTOR);
5361 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5362 longmode = is_long_mode(vcpu) && cs_l == 1;
5365 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5366 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5367 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5368 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5369 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5370 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5372 #ifdef CONFIG_X86_64
5374 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5375 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5376 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5380 code = param & 0xffff;
5381 fast = (param >> 16) & 0x1;
5382 rep_cnt = (param >> 32) & 0xfff;
5383 rep_idx = (param >> 48) & 0xfff;
5385 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5388 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5389 kvm_vcpu_on_spin(vcpu);
5392 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5396 ret = res | (((u64)rep_done & 0xfff) << 32);
5398 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5400 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5401 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5407 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5409 unsigned long nr, a0, a1, a2, a3, ret;
5412 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5413 return kvm_hv_hypercall(vcpu);
5415 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5416 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5417 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5418 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5419 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5421 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5423 if (!is_long_mode(vcpu)) {
5431 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5437 case KVM_HC_VAPIC_POLL_IRQ:
5445 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5446 ++vcpu->stat.hypercalls;
5449 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5451 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5453 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5454 char instruction[3];
5455 unsigned long rip = kvm_rip_read(vcpu);
5458 * Blow out the MMU to ensure that no other VCPU has an active mapping
5459 * to ensure that the updated hypercall appears atomically across all
5462 kvm_mmu_zap_all(vcpu->kvm);
5464 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5466 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5470 * Check if userspace requested an interrupt window, and that the
5471 * interrupt window is open.
5473 * No need to exit to userspace if we already have an interrupt queued.
5475 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5477 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5478 vcpu->run->request_interrupt_window &&
5479 kvm_arch_interrupt_allowed(vcpu));
5482 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5484 struct kvm_run *kvm_run = vcpu->run;
5486 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5487 kvm_run->cr8 = kvm_get_cr8(vcpu);
5488 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5489 if (irqchip_in_kernel(vcpu->kvm))
5490 kvm_run->ready_for_interrupt_injection = 1;
5492 kvm_run->ready_for_interrupt_injection =
5493 kvm_arch_interrupt_allowed(vcpu) &&
5494 !kvm_cpu_has_interrupt(vcpu) &&
5495 !kvm_event_needs_reinjection(vcpu);
5498 static int vapic_enter(struct kvm_vcpu *vcpu)
5500 struct kvm_lapic *apic = vcpu->arch.apic;
5503 if (!apic || !apic->vapic_addr)
5506 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5507 if (is_error_page(page))
5510 vcpu->arch.apic->vapic_page = page;
5514 static void vapic_exit(struct kvm_vcpu *vcpu)
5516 struct kvm_lapic *apic = vcpu->arch.apic;
5519 if (!apic || !apic->vapic_addr)
5522 idx = srcu_read_lock(&vcpu->kvm->srcu);
5523 kvm_release_page_dirty(apic->vapic_page);
5524 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5525 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5528 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5532 if (!kvm_x86_ops->update_cr8_intercept)
5535 if (!vcpu->arch.apic)
5538 if (!vcpu->arch.apic->vapic_addr)
5539 max_irr = kvm_lapic_find_highest_irr(vcpu);
5546 tpr = kvm_lapic_get_cr8(vcpu);
5548 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5551 static void inject_pending_event(struct kvm_vcpu *vcpu)
5553 /* try to reinject previous events if any */
5554 if (vcpu->arch.exception.pending) {
5555 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5556 vcpu->arch.exception.has_error_code,
5557 vcpu->arch.exception.error_code);
5558 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5559 vcpu->arch.exception.has_error_code,
5560 vcpu->arch.exception.error_code,
5561 vcpu->arch.exception.reinject);
5565 if (vcpu->arch.nmi_injected) {
5566 kvm_x86_ops->set_nmi(vcpu);
5570 if (vcpu->arch.interrupt.pending) {
5571 kvm_x86_ops->set_irq(vcpu);
5575 /* try to inject new event if pending */
5576 if (vcpu->arch.nmi_pending) {
5577 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5578 --vcpu->arch.nmi_pending;
5579 vcpu->arch.nmi_injected = true;
5580 kvm_x86_ops->set_nmi(vcpu);
5582 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
5583 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5584 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5586 kvm_x86_ops->set_irq(vcpu);
5591 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5593 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5594 !vcpu->guest_xcr0_loaded) {
5595 /* kvm_set_xcr() also depends on this */
5596 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5597 vcpu->guest_xcr0_loaded = 1;
5601 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5603 if (vcpu->guest_xcr0_loaded) {
5604 if (vcpu->arch.xcr0 != host_xcr0)
5605 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5606 vcpu->guest_xcr0_loaded = 0;
5610 static void process_nmi(struct kvm_vcpu *vcpu)
5615 * x86 is limited to one NMI running, and one NMI pending after it.
5616 * If an NMI is already in progress, limit further NMIs to just one.
5617 * Otherwise, allow two (and we'll inject the first one immediately).
5619 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5622 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5623 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5624 kvm_make_request(KVM_REQ_EVENT, vcpu);
5627 static void kvm_gen_update_masterclock(struct kvm *kvm)
5629 #ifdef CONFIG_X86_64
5631 struct kvm_vcpu *vcpu;
5632 struct kvm_arch *ka = &kvm->arch;
5634 spin_lock(&ka->pvclock_gtod_sync_lock);
5635 kvm_make_mclock_inprogress_request(kvm);
5636 /* no guest entries from this point */
5637 pvclock_update_vm_gtod_copy(kvm);
5639 kvm_for_each_vcpu(i, vcpu, kvm)
5640 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
5642 /* guest entries allowed */
5643 kvm_for_each_vcpu(i, vcpu, kvm)
5644 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
5646 spin_unlock(&ka->pvclock_gtod_sync_lock);
5650 static void update_eoi_exitmap(struct kvm_vcpu *vcpu)
5652 u64 eoi_exit_bitmap[4];
5654 memset(eoi_exit_bitmap, 0, 32);
5656 kvm_ioapic_calculate_eoi_exitmap(vcpu, eoi_exit_bitmap);
5657 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
5660 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5663 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5664 vcpu->run->request_interrupt_window;
5665 bool req_immediate_exit = 0;
5667 if (vcpu->requests) {
5668 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5669 kvm_mmu_unload(vcpu);
5670 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5671 __kvm_migrate_timers(vcpu);
5672 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
5673 kvm_gen_update_masterclock(vcpu->kvm);
5674 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5675 r = kvm_guest_time_update(vcpu);
5679 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5680 kvm_mmu_sync_roots(vcpu);
5681 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5682 kvm_x86_ops->tlb_flush(vcpu);
5683 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5684 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5688 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5689 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5693 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5694 vcpu->fpu_active = 0;
5695 kvm_x86_ops->fpu_deactivate(vcpu);
5697 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5698 /* Page is swapped out. Do synthetic halt */
5699 vcpu->arch.apf.halted = true;
5703 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5704 record_steal_time(vcpu);
5705 if (kvm_check_request(KVM_REQ_NMI, vcpu))
5707 req_immediate_exit =
5708 kvm_check_request(KVM_REQ_IMMEDIATE_EXIT, vcpu);
5709 if (kvm_check_request(KVM_REQ_PMU, vcpu))
5710 kvm_handle_pmu_event(vcpu);
5711 if (kvm_check_request(KVM_REQ_PMI, vcpu))
5712 kvm_deliver_pmi(vcpu);
5713 if (kvm_check_request(KVM_REQ_EOIBITMAP, vcpu))
5714 update_eoi_exitmap(vcpu);
5717 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5718 kvm_apic_accept_events(vcpu);
5719 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
5724 inject_pending_event(vcpu);
5726 /* enable NMI/IRQ window open exits if needed */
5727 if (vcpu->arch.nmi_pending)
5728 kvm_x86_ops->enable_nmi_window(vcpu);
5729 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
5730 kvm_x86_ops->enable_irq_window(vcpu);
5732 if (kvm_lapic_enabled(vcpu)) {
5734 * Update architecture specific hints for APIC
5735 * virtual interrupt delivery.
5737 if (kvm_x86_ops->hwapic_irr_update)
5738 kvm_x86_ops->hwapic_irr_update(vcpu,
5739 kvm_lapic_find_highest_irr(vcpu));
5740 update_cr8_intercept(vcpu);
5741 kvm_lapic_sync_to_vapic(vcpu);
5745 r = kvm_mmu_reload(vcpu);
5747 goto cancel_injection;
5752 kvm_x86_ops->prepare_guest_switch(vcpu);
5753 if (vcpu->fpu_active)
5754 kvm_load_guest_fpu(vcpu);
5755 kvm_load_guest_xcr0(vcpu);
5757 vcpu->mode = IN_GUEST_MODE;
5759 /* We should set ->mode before check ->requests,
5760 * see the comment in make_all_cpus_request.
5764 local_irq_disable();
5766 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5767 || need_resched() || signal_pending(current)) {
5768 vcpu->mode = OUTSIDE_GUEST_MODE;
5773 goto cancel_injection;
5776 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5778 if (req_immediate_exit)
5779 smp_send_reschedule(vcpu->cpu);
5783 if (unlikely(vcpu->arch.switch_db_regs)) {
5785 set_debugreg(vcpu->arch.eff_db[0], 0);
5786 set_debugreg(vcpu->arch.eff_db[1], 1);
5787 set_debugreg(vcpu->arch.eff_db[2], 2);
5788 set_debugreg(vcpu->arch.eff_db[3], 3);
5791 trace_kvm_entry(vcpu->vcpu_id);
5792 kvm_x86_ops->run(vcpu);
5795 * If the guest has used debug registers, at least dr7
5796 * will be disabled while returning to the host.
5797 * If we don't have active breakpoints in the host, we don't
5798 * care about the messed up debug address registers. But if
5799 * we have some of them active, restore the old state.
5801 if (hw_breakpoint_active())
5802 hw_breakpoint_restore();
5804 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
5807 vcpu->mode = OUTSIDE_GUEST_MODE;
5814 * We must have an instruction between local_irq_enable() and
5815 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5816 * the interrupt shadow. The stat.exits increment will do nicely.
5817 * But we need to prevent reordering, hence this barrier():
5825 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5828 * Profile KVM exit RIPs:
5830 if (unlikely(prof_on == KVM_PROFILING)) {
5831 unsigned long rip = kvm_rip_read(vcpu);
5832 profile_hit(KVM_PROFILING, (void *)rip);
5835 if (unlikely(vcpu->arch.tsc_always_catchup))
5836 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5838 if (vcpu->arch.apic_attention)
5839 kvm_lapic_sync_from_vapic(vcpu);
5841 r = kvm_x86_ops->handle_exit(vcpu);
5845 kvm_x86_ops->cancel_injection(vcpu);
5846 if (unlikely(vcpu->arch.apic_attention))
5847 kvm_lapic_sync_from_vapic(vcpu);
5853 static int __vcpu_run(struct kvm_vcpu *vcpu)
5856 struct kvm *kvm = vcpu->kvm;
5858 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5859 r = vapic_enter(vcpu);
5861 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5867 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5868 !vcpu->arch.apf.halted)
5869 r = vcpu_enter_guest(vcpu);
5871 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5872 kvm_vcpu_block(vcpu);
5873 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5874 if (kvm_check_request(KVM_REQ_UNHALT, vcpu)) {
5875 kvm_apic_accept_events(vcpu);
5876 switch(vcpu->arch.mp_state) {
5877 case KVM_MP_STATE_HALTED:
5878 vcpu->arch.mp_state =
5879 KVM_MP_STATE_RUNNABLE;
5880 case KVM_MP_STATE_RUNNABLE:
5881 vcpu->arch.apf.halted = false;
5883 case KVM_MP_STATE_INIT_RECEIVED:
5895 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5896 if (kvm_cpu_has_pending_timer(vcpu))
5897 kvm_inject_pending_timer_irqs(vcpu);
5899 if (dm_request_for_irq_injection(vcpu)) {
5901 vcpu->run->exit_reason = KVM_EXIT_INTR;
5902 ++vcpu->stat.request_irq_exits;
5905 kvm_check_async_pf_completion(vcpu);
5907 if (signal_pending(current)) {
5909 vcpu->run->exit_reason = KVM_EXIT_INTR;
5910 ++vcpu->stat.signal_exits;
5912 if (need_resched()) {
5913 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5915 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5919 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5926 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
5929 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5930 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5931 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5932 if (r != EMULATE_DONE)
5937 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
5939 BUG_ON(!vcpu->arch.pio.count);
5941 return complete_emulated_io(vcpu);
5945 * Implements the following, as a state machine:
5949 * for each mmio piece in the fragment
5957 * for each mmio piece in the fragment
5962 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
5964 struct kvm_run *run = vcpu->run;
5965 struct kvm_mmio_fragment *frag;
5968 BUG_ON(!vcpu->mmio_needed);
5970 /* Complete previous fragment */
5971 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
5972 len = min(8u, frag->len);
5973 if (!vcpu->mmio_is_write)
5974 memcpy(frag->data, run->mmio.data, len);
5976 if (frag->len <= 8) {
5977 /* Switch to the next fragment. */
5979 vcpu->mmio_cur_fragment++;
5981 /* Go forward to the next mmio piece. */
5987 if (vcpu->mmio_cur_fragment == vcpu->mmio_nr_fragments) {
5988 vcpu->mmio_needed = 0;
5989 if (vcpu->mmio_is_write)
5991 vcpu->mmio_read_completed = 1;
5992 return complete_emulated_io(vcpu);
5995 run->exit_reason = KVM_EXIT_MMIO;
5996 run->mmio.phys_addr = frag->gpa;
5997 if (vcpu->mmio_is_write)
5998 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
5999 run->mmio.len = min(8u, frag->len);
6000 run->mmio.is_write = vcpu->mmio_is_write;
6001 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6006 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6011 if (!tsk_used_math(current) && init_fpu(current))
6014 if (vcpu->sigset_active)
6015 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6017 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6018 kvm_vcpu_block(vcpu);
6019 kvm_apic_accept_events(vcpu);
6020 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6025 /* re-sync apic's tpr */
6026 if (!irqchip_in_kernel(vcpu->kvm)) {
6027 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6033 if (unlikely(vcpu->arch.complete_userspace_io)) {
6034 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6035 vcpu->arch.complete_userspace_io = NULL;
6040 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6042 r = __vcpu_run(vcpu);
6045 post_kvm_run_save(vcpu);
6046 if (vcpu->sigset_active)
6047 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6052 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6054 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6056 * We are here if userspace calls get_regs() in the middle of
6057 * instruction emulation. Registers state needs to be copied
6058 * back from emulation context to vcpu. Userspace shouldn't do
6059 * that usually, but some bad designed PV devices (vmware
6060 * backdoor interface) need this to work
6062 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6063 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6065 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6066 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6067 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6068 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6069 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6070 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6071 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6072 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6073 #ifdef CONFIG_X86_64
6074 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6075 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6076 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6077 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6078 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6079 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6080 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6081 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6084 regs->rip = kvm_rip_read(vcpu);
6085 regs->rflags = kvm_get_rflags(vcpu);
6090 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6092 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6093 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6095 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6096 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6097 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6098 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6099 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6100 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6101 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6102 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6103 #ifdef CONFIG_X86_64
6104 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6105 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6106 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6107 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6108 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6109 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6110 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6111 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6114 kvm_rip_write(vcpu, regs->rip);
6115 kvm_set_rflags(vcpu, regs->rflags);
6117 vcpu->arch.exception.pending = false;
6119 kvm_make_request(KVM_REQ_EVENT, vcpu);
6124 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6126 struct kvm_segment cs;
6128 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6132 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6134 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6135 struct kvm_sregs *sregs)
6139 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6140 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6141 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6142 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6143 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6144 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6146 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6147 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6149 kvm_x86_ops->get_idt(vcpu, &dt);
6150 sregs->idt.limit = dt.size;
6151 sregs->idt.base = dt.address;
6152 kvm_x86_ops->get_gdt(vcpu, &dt);
6153 sregs->gdt.limit = dt.size;
6154 sregs->gdt.base = dt.address;
6156 sregs->cr0 = kvm_read_cr0(vcpu);
6157 sregs->cr2 = vcpu->arch.cr2;
6158 sregs->cr3 = kvm_read_cr3(vcpu);
6159 sregs->cr4 = kvm_read_cr4(vcpu);
6160 sregs->cr8 = kvm_get_cr8(vcpu);
6161 sregs->efer = vcpu->arch.efer;
6162 sregs->apic_base = kvm_get_apic_base(vcpu);
6164 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6166 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6167 set_bit(vcpu->arch.interrupt.nr,
6168 (unsigned long *)sregs->interrupt_bitmap);
6173 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6174 struct kvm_mp_state *mp_state)
6176 kvm_apic_accept_events(vcpu);
6177 mp_state->mp_state = vcpu->arch.mp_state;
6181 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6182 struct kvm_mp_state *mp_state)
6184 if (!kvm_vcpu_has_lapic(vcpu) &&
6185 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6188 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6189 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6190 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6192 vcpu->arch.mp_state = mp_state->mp_state;
6193 kvm_make_request(KVM_REQ_EVENT, vcpu);
6197 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6198 int reason, bool has_error_code, u32 error_code)
6200 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6203 init_emulate_ctxt(vcpu);
6205 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6206 has_error_code, error_code);
6209 return EMULATE_FAIL;
6211 kvm_rip_write(vcpu, ctxt->eip);
6212 kvm_set_rflags(vcpu, ctxt->eflags);
6213 kvm_make_request(KVM_REQ_EVENT, vcpu);
6214 return EMULATE_DONE;
6216 EXPORT_SYMBOL_GPL(kvm_task_switch);
6218 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6219 struct kvm_sregs *sregs)
6221 int mmu_reset_needed = 0;
6222 int pending_vec, max_bits, idx;
6225 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6228 dt.size = sregs->idt.limit;
6229 dt.address = sregs->idt.base;
6230 kvm_x86_ops->set_idt(vcpu, &dt);
6231 dt.size = sregs->gdt.limit;
6232 dt.address = sregs->gdt.base;
6233 kvm_x86_ops->set_gdt(vcpu, &dt);
6235 vcpu->arch.cr2 = sregs->cr2;
6236 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6237 vcpu->arch.cr3 = sregs->cr3;
6238 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6240 kvm_set_cr8(vcpu, sregs->cr8);
6242 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6243 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6244 kvm_set_apic_base(vcpu, sregs->apic_base);
6246 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6247 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6248 vcpu->arch.cr0 = sregs->cr0;
6250 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6251 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6252 if (sregs->cr4 & X86_CR4_OSXSAVE)
6253 kvm_update_cpuid(vcpu);
6255 idx = srcu_read_lock(&vcpu->kvm->srcu);
6256 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6257 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6258 mmu_reset_needed = 1;
6260 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6262 if (mmu_reset_needed)
6263 kvm_mmu_reset_context(vcpu);
6265 max_bits = KVM_NR_INTERRUPTS;
6266 pending_vec = find_first_bit(
6267 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6268 if (pending_vec < max_bits) {
6269 kvm_queue_interrupt(vcpu, pending_vec, false);
6270 pr_debug("Set back pending irq %d\n", pending_vec);
6273 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6274 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6275 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6276 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6277 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6278 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6280 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6281 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6283 update_cr8_intercept(vcpu);
6285 /* Older userspace won't unhalt the vcpu on reset. */
6286 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6287 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6289 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6291 kvm_make_request(KVM_REQ_EVENT, vcpu);
6296 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6297 struct kvm_guest_debug *dbg)
6299 unsigned long rflags;
6302 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6304 if (vcpu->arch.exception.pending)
6306 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6307 kvm_queue_exception(vcpu, DB_VECTOR);
6309 kvm_queue_exception(vcpu, BP_VECTOR);
6313 * Read rflags as long as potentially injected trace flags are still
6316 rflags = kvm_get_rflags(vcpu);
6318 vcpu->guest_debug = dbg->control;
6319 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6320 vcpu->guest_debug = 0;
6322 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6323 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6324 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6325 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6327 for (i = 0; i < KVM_NR_DB_REGS; i++)
6328 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6330 kvm_update_dr7(vcpu);
6332 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6333 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6334 get_segment_base(vcpu, VCPU_SREG_CS);
6337 * Trigger an rflags update that will inject or remove the trace
6340 kvm_set_rflags(vcpu, rflags);
6342 kvm_x86_ops->update_db_bp_intercept(vcpu);
6352 * Translate a guest virtual address to a guest physical address.
6354 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6355 struct kvm_translation *tr)
6357 unsigned long vaddr = tr->linear_address;
6361 idx = srcu_read_lock(&vcpu->kvm->srcu);
6362 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6363 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6364 tr->physical_address = gpa;
6365 tr->valid = gpa != UNMAPPED_GVA;
6372 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6374 struct i387_fxsave_struct *fxsave =
6375 &vcpu->arch.guest_fpu.state->fxsave;
6377 memcpy(fpu->fpr, fxsave->st_space, 128);
6378 fpu->fcw = fxsave->cwd;
6379 fpu->fsw = fxsave->swd;
6380 fpu->ftwx = fxsave->twd;
6381 fpu->last_opcode = fxsave->fop;
6382 fpu->last_ip = fxsave->rip;
6383 fpu->last_dp = fxsave->rdp;
6384 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6389 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6391 struct i387_fxsave_struct *fxsave =
6392 &vcpu->arch.guest_fpu.state->fxsave;
6394 memcpy(fxsave->st_space, fpu->fpr, 128);
6395 fxsave->cwd = fpu->fcw;
6396 fxsave->swd = fpu->fsw;
6397 fxsave->twd = fpu->ftwx;
6398 fxsave->fop = fpu->last_opcode;
6399 fxsave->rip = fpu->last_ip;
6400 fxsave->rdp = fpu->last_dp;
6401 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6406 int fx_init(struct kvm_vcpu *vcpu)
6410 err = fpu_alloc(&vcpu->arch.guest_fpu);
6414 fpu_finit(&vcpu->arch.guest_fpu);
6417 * Ensure guest xcr0 is valid for loading
6419 vcpu->arch.xcr0 = XSTATE_FP;
6421 vcpu->arch.cr0 |= X86_CR0_ET;
6425 EXPORT_SYMBOL_GPL(fx_init);
6427 static void fx_free(struct kvm_vcpu *vcpu)
6429 fpu_free(&vcpu->arch.guest_fpu);
6432 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6434 if (vcpu->guest_fpu_loaded)
6438 * Restore all possible states in the guest,
6439 * and assume host would use all available bits.
6440 * Guest xcr0 would be loaded later.
6442 kvm_put_guest_xcr0(vcpu);
6443 vcpu->guest_fpu_loaded = 1;
6444 __kernel_fpu_begin();
6445 fpu_restore_checking(&vcpu->arch.guest_fpu);
6449 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6451 kvm_put_guest_xcr0(vcpu);
6453 if (!vcpu->guest_fpu_loaded)
6456 vcpu->guest_fpu_loaded = 0;
6457 fpu_save_init(&vcpu->arch.guest_fpu);
6459 ++vcpu->stat.fpu_reload;
6460 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6464 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6466 kvmclock_reset(vcpu);
6468 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6470 kvm_x86_ops->vcpu_free(vcpu);
6473 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6476 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6477 printk_once(KERN_WARNING
6478 "kvm: SMP vm created on host with unstable TSC; "
6479 "guest TSC will not be reliable\n");
6480 return kvm_x86_ops->vcpu_create(kvm, id);
6483 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6487 vcpu->arch.mtrr_state.have_fixed = 1;
6488 r = vcpu_load(vcpu);
6491 kvm_vcpu_reset(vcpu);
6492 r = kvm_mmu_setup(vcpu);
6498 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
6501 struct msr_data msr;
6503 r = vcpu_load(vcpu);
6507 msr.index = MSR_IA32_TSC;
6508 msr.host_initiated = true;
6509 kvm_write_tsc(vcpu, &msr);
6515 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6518 vcpu->arch.apf.msr_val = 0;
6520 r = vcpu_load(vcpu);
6522 kvm_mmu_unload(vcpu);
6526 kvm_x86_ops->vcpu_free(vcpu);
6529 void kvm_vcpu_reset(struct kvm_vcpu *vcpu)
6531 atomic_set(&vcpu->arch.nmi_queued, 0);
6532 vcpu->arch.nmi_pending = 0;
6533 vcpu->arch.nmi_injected = false;
6535 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6536 vcpu->arch.dr6 = DR6_FIXED_1;
6537 vcpu->arch.dr7 = DR7_FIXED_1;
6538 kvm_update_dr7(vcpu);
6540 kvm_make_request(KVM_REQ_EVENT, vcpu);
6541 vcpu->arch.apf.msr_val = 0;
6542 vcpu->arch.st.msr_val = 0;
6544 kvmclock_reset(vcpu);
6546 kvm_clear_async_pf_completion_queue(vcpu);
6547 kvm_async_pf_hash_reset(vcpu);
6548 vcpu->arch.apf.halted = false;
6550 kvm_pmu_reset(vcpu);
6552 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
6553 vcpu->arch.regs_avail = ~0;
6554 vcpu->arch.regs_dirty = ~0;
6556 kvm_x86_ops->vcpu_reset(vcpu);
6559 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, unsigned int vector)
6561 struct kvm_segment cs;
6563 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6564 cs.selector = vector << 8;
6565 cs.base = vector << 12;
6566 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6567 kvm_rip_write(vcpu, 0);
6570 int kvm_arch_hardware_enable(void *garbage)
6573 struct kvm_vcpu *vcpu;
6578 bool stable, backwards_tsc = false;
6580 kvm_shared_msr_cpu_online();
6581 ret = kvm_x86_ops->hardware_enable(garbage);
6585 local_tsc = native_read_tsc();
6586 stable = !check_tsc_unstable();
6587 list_for_each_entry(kvm, &vm_list, vm_list) {
6588 kvm_for_each_vcpu(i, vcpu, kvm) {
6589 if (!stable && vcpu->cpu == smp_processor_id())
6590 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
6591 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
6592 backwards_tsc = true;
6593 if (vcpu->arch.last_host_tsc > max_tsc)
6594 max_tsc = vcpu->arch.last_host_tsc;
6600 * Sometimes, even reliable TSCs go backwards. This happens on
6601 * platforms that reset TSC during suspend or hibernate actions, but
6602 * maintain synchronization. We must compensate. Fortunately, we can
6603 * detect that condition here, which happens early in CPU bringup,
6604 * before any KVM threads can be running. Unfortunately, we can't
6605 * bring the TSCs fully up to date with real time, as we aren't yet far
6606 * enough into CPU bringup that we know how much real time has actually
6607 * elapsed; our helper function, get_kernel_ns() will be using boot
6608 * variables that haven't been updated yet.
6610 * So we simply find the maximum observed TSC above, then record the
6611 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
6612 * the adjustment will be applied. Note that we accumulate
6613 * adjustments, in case multiple suspend cycles happen before some VCPU
6614 * gets a chance to run again. In the event that no KVM threads get a
6615 * chance to run, we will miss the entire elapsed period, as we'll have
6616 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
6617 * loose cycle time. This isn't too big a deal, since the loss will be
6618 * uniform across all VCPUs (not to mention the scenario is extremely
6619 * unlikely). It is possible that a second hibernate recovery happens
6620 * much faster than a first, causing the observed TSC here to be
6621 * smaller; this would require additional padding adjustment, which is
6622 * why we set last_host_tsc to the local tsc observed here.
6624 * N.B. - this code below runs only on platforms with reliable TSC,
6625 * as that is the only way backwards_tsc is set above. Also note
6626 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
6627 * have the same delta_cyc adjustment applied if backwards_tsc
6628 * is detected. Note further, this adjustment is only done once,
6629 * as we reset last_host_tsc on all VCPUs to stop this from being
6630 * called multiple times (one for each physical CPU bringup).
6632 * Platforms with unreliable TSCs don't have to deal with this, they
6633 * will be compensated by the logic in vcpu_load, which sets the TSC to
6634 * catchup mode. This will catchup all VCPUs to real time, but cannot
6635 * guarantee that they stay in perfect synchronization.
6637 if (backwards_tsc) {
6638 u64 delta_cyc = max_tsc - local_tsc;
6639 list_for_each_entry(kvm, &vm_list, vm_list) {
6640 kvm_for_each_vcpu(i, vcpu, kvm) {
6641 vcpu->arch.tsc_offset_adjustment += delta_cyc;
6642 vcpu->arch.last_host_tsc = local_tsc;
6643 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
6648 * We have to disable TSC offset matching.. if you were
6649 * booting a VM while issuing an S4 host suspend....
6650 * you may have some problem. Solving this issue is
6651 * left as an exercise to the reader.
6653 kvm->arch.last_tsc_nsec = 0;
6654 kvm->arch.last_tsc_write = 0;
6661 void kvm_arch_hardware_disable(void *garbage)
6663 kvm_x86_ops->hardware_disable(garbage);
6664 drop_user_return_notifiers(garbage);
6667 int kvm_arch_hardware_setup(void)
6669 return kvm_x86_ops->hardware_setup();
6672 void kvm_arch_hardware_unsetup(void)
6674 kvm_x86_ops->hardware_unsetup();
6677 void kvm_arch_check_processor_compat(void *rtn)
6679 kvm_x86_ops->check_processor_compatibility(rtn);
6682 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
6684 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
6687 struct static_key kvm_no_apic_vcpu __read_mostly;
6689 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6695 BUG_ON(vcpu->kvm == NULL);
6698 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6699 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6700 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6702 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6704 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6709 vcpu->arch.pio_data = page_address(page);
6711 kvm_set_tsc_khz(vcpu, max_tsc_khz);
6713 r = kvm_mmu_create(vcpu);
6715 goto fail_free_pio_data;
6717 if (irqchip_in_kernel(kvm)) {
6718 r = kvm_create_lapic(vcpu);
6720 goto fail_mmu_destroy;
6722 static_key_slow_inc(&kvm_no_apic_vcpu);
6724 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6726 if (!vcpu->arch.mce_banks) {
6728 goto fail_free_lapic;
6730 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6732 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6733 goto fail_free_mce_banks;
6737 goto fail_free_wbinvd_dirty_mask;
6739 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
6740 vcpu->arch.pv_time_enabled = false;
6741 kvm_async_pf_hash_reset(vcpu);
6745 fail_free_wbinvd_dirty_mask:
6746 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6747 fail_free_mce_banks:
6748 kfree(vcpu->arch.mce_banks);
6750 kvm_free_lapic(vcpu);
6752 kvm_mmu_destroy(vcpu);
6754 free_page((unsigned long)vcpu->arch.pio_data);
6759 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6763 kvm_pmu_destroy(vcpu);
6764 kfree(vcpu->arch.mce_banks);
6765 kvm_free_lapic(vcpu);
6766 idx = srcu_read_lock(&vcpu->kvm->srcu);
6767 kvm_mmu_destroy(vcpu);
6768 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6769 free_page((unsigned long)vcpu->arch.pio_data);
6770 if (!irqchip_in_kernel(vcpu->kvm))
6771 static_key_slow_dec(&kvm_no_apic_vcpu);
6774 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
6779 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6780 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6782 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6783 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6784 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
6785 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
6786 &kvm->arch.irq_sources_bitmap);
6788 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6789 mutex_init(&kvm->arch.apic_map_lock);
6790 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
6792 pvclock_update_vm_gtod_copy(kvm);
6797 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6800 r = vcpu_load(vcpu);
6802 kvm_mmu_unload(vcpu);
6806 static void kvm_free_vcpus(struct kvm *kvm)
6809 struct kvm_vcpu *vcpu;
6812 * Unpin any mmu pages first.
6814 kvm_for_each_vcpu(i, vcpu, kvm) {
6815 kvm_clear_async_pf_completion_queue(vcpu);
6816 kvm_unload_vcpu_mmu(vcpu);
6818 kvm_for_each_vcpu(i, vcpu, kvm)
6819 kvm_arch_vcpu_free(vcpu);
6821 mutex_lock(&kvm->lock);
6822 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6823 kvm->vcpus[i] = NULL;
6825 atomic_set(&kvm->online_vcpus, 0);
6826 mutex_unlock(&kvm->lock);
6829 void kvm_arch_sync_events(struct kvm *kvm)
6831 kvm_free_all_assigned_devices(kvm);
6835 void kvm_arch_destroy_vm(struct kvm *kvm)
6837 kvm_iommu_unmap_guest(kvm);
6838 kfree(kvm->arch.vpic);
6839 kfree(kvm->arch.vioapic);
6840 kvm_free_vcpus(kvm);
6841 if (kvm->arch.apic_access_page)
6842 put_page(kvm->arch.apic_access_page);
6843 if (kvm->arch.ept_identity_pagetable)
6844 put_page(kvm->arch.ept_identity_pagetable);
6845 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
6848 void kvm_arch_free_memslot(struct kvm_memory_slot *free,
6849 struct kvm_memory_slot *dont)
6853 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6854 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
6855 kvm_kvfree(free->arch.rmap[i]);
6856 free->arch.rmap[i] = NULL;
6861 if (!dont || free->arch.lpage_info[i - 1] !=
6862 dont->arch.lpage_info[i - 1]) {
6863 kvm_kvfree(free->arch.lpage_info[i - 1]);
6864 free->arch.lpage_info[i - 1] = NULL;
6869 int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages)
6873 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6878 lpages = gfn_to_index(slot->base_gfn + npages - 1,
6879 slot->base_gfn, level) + 1;
6881 slot->arch.rmap[i] =
6882 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
6883 if (!slot->arch.rmap[i])
6888 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
6889 sizeof(*slot->arch.lpage_info[i - 1]));
6890 if (!slot->arch.lpage_info[i - 1])
6893 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
6894 slot->arch.lpage_info[i - 1][0].write_count = 1;
6895 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
6896 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
6897 ugfn = slot->userspace_addr >> PAGE_SHIFT;
6899 * If the gfn and userspace address are not aligned wrt each
6900 * other, or if explicitly asked to, disable large page
6901 * support for this slot
6903 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
6904 !kvm_largepages_enabled()) {
6907 for (j = 0; j < lpages; ++j)
6908 slot->arch.lpage_info[i - 1][j].write_count = 1;
6915 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6916 kvm_kvfree(slot->arch.rmap[i]);
6917 slot->arch.rmap[i] = NULL;
6921 kvm_kvfree(slot->arch.lpage_info[i - 1]);
6922 slot->arch.lpage_info[i - 1] = NULL;
6927 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6928 struct kvm_memory_slot *memslot,
6929 struct kvm_userspace_memory_region *mem,
6930 enum kvm_mr_change change)
6933 * Only private memory slots need to be mapped here since
6934 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
6936 if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
6937 unsigned long userspace_addr;
6940 * MAP_SHARED to prevent internal slot pages from being moved
6943 userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
6944 PROT_READ | PROT_WRITE,
6945 MAP_SHARED | MAP_ANONYMOUS, 0);
6947 if (IS_ERR((void *)userspace_addr))
6948 return PTR_ERR((void *)userspace_addr);
6950 memslot->userspace_addr = userspace_addr;
6956 void kvm_arch_commit_memory_region(struct kvm *kvm,
6957 struct kvm_userspace_memory_region *mem,
6958 const struct kvm_memory_slot *old,
6959 enum kvm_mr_change change)
6962 int nr_mmu_pages = 0;
6964 if ((mem->slot >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_DELETE)) {
6967 ret = vm_munmap(old->userspace_addr,
6968 old->npages * PAGE_SIZE);
6971 "kvm_vm_ioctl_set_memory_region: "
6972 "failed to munmap memory\n");
6975 if (!kvm->arch.n_requested_mmu_pages)
6976 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6979 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6981 * Write protect all pages for dirty logging.
6982 * Existing largepage mappings are destroyed here and new ones will
6983 * not be created until the end of the logging.
6985 if ((change != KVM_MR_DELETE) && (mem->flags & KVM_MEM_LOG_DIRTY_PAGES))
6986 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6988 * If memory slot is created, or moved, we need to clear all
6991 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
6992 kvm_mmu_zap_mmio_sptes(kvm);
6993 kvm_reload_remote_mmus(kvm);
6997 void kvm_arch_flush_shadow_all(struct kvm *kvm)
6999 kvm_mmu_zap_all(kvm);
7000 kvm_reload_remote_mmus(kvm);
7003 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7004 struct kvm_memory_slot *slot)
7006 kvm_arch_flush_shadow_all(kvm);
7009 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
7011 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7012 !vcpu->arch.apf.halted)
7013 || !list_empty_careful(&vcpu->async_pf.done)
7014 || kvm_apic_has_events(vcpu)
7015 || atomic_read(&vcpu->arch.nmi_queued) ||
7016 (kvm_arch_interrupt_allowed(vcpu) &&
7017 kvm_cpu_has_interrupt(vcpu));
7020 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
7022 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
7025 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7027 return kvm_x86_ops->interrupt_allowed(vcpu);
7030 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7032 unsigned long current_rip = kvm_rip_read(vcpu) +
7033 get_segment_base(vcpu, VCPU_SREG_CS);
7035 return current_rip == linear_rip;
7037 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7039 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7041 unsigned long rflags;
7043 rflags = kvm_x86_ops->get_rflags(vcpu);
7044 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7045 rflags &= ~X86_EFLAGS_TF;
7048 EXPORT_SYMBOL_GPL(kvm_get_rflags);
7050 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7052 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
7053 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
7054 rflags |= X86_EFLAGS_TF;
7055 kvm_x86_ops->set_rflags(vcpu, rflags);
7056 kvm_make_request(KVM_REQ_EVENT, vcpu);
7058 EXPORT_SYMBOL_GPL(kvm_set_rflags);
7060 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7064 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
7065 is_error_page(work->page))
7068 r = kvm_mmu_reload(vcpu);
7072 if (!vcpu->arch.mmu.direct_map &&
7073 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7076 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7079 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7081 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7084 static inline u32 kvm_async_pf_next_probe(u32 key)
7086 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7089 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7091 u32 key = kvm_async_pf_hash_fn(gfn);
7093 while (vcpu->arch.apf.gfns[key] != ~0)
7094 key = kvm_async_pf_next_probe(key);
7096 vcpu->arch.apf.gfns[key] = gfn;
7099 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7102 u32 key = kvm_async_pf_hash_fn(gfn);
7104 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
7105 (vcpu->arch.apf.gfns[key] != gfn &&
7106 vcpu->arch.apf.gfns[key] != ~0); i++)
7107 key = kvm_async_pf_next_probe(key);
7112 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7114 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7117 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7121 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7123 vcpu->arch.apf.gfns[i] = ~0;
7125 j = kvm_async_pf_next_probe(j);
7126 if (vcpu->arch.apf.gfns[j] == ~0)
7128 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7130 * k lies cyclically in ]i,j]
7132 * |....j i.k.| or |.k..j i...|
7134 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7135 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7140 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7143 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7147 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7148 struct kvm_async_pf *work)
7150 struct x86_exception fault;
7152 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
7153 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7155 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
7156 (vcpu->arch.apf.send_user_only &&
7157 kvm_x86_ops->get_cpl(vcpu) == 0))
7158 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7159 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
7160 fault.vector = PF_VECTOR;
7161 fault.error_code_valid = true;
7162 fault.error_code = 0;
7163 fault.nested_page_fault = false;
7164 fault.address = work->arch.token;
7165 kvm_inject_page_fault(vcpu, &fault);
7169 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7170 struct kvm_async_pf *work)
7172 struct x86_exception fault;
7174 trace_kvm_async_pf_ready(work->arch.token, work->gva);
7175 if (is_error_page(work->page))
7176 work->arch.token = ~0; /* broadcast wakeup */
7178 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7180 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7181 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
7182 fault.vector = PF_VECTOR;
7183 fault.error_code_valid = true;
7184 fault.error_code = 0;
7185 fault.nested_page_fault = false;
7186 fault.address = work->arch.token;
7187 kvm_inject_page_fault(vcpu, &fault);
7189 vcpu->arch.apf.halted = false;
7190 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7193 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7195 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7198 return !kvm_event_needs_reinjection(vcpu) &&
7199 kvm_x86_ops->interrupt_allowed(vcpu);
7202 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
7203 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
7204 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
7205 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
7206 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
7207 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
7208 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
7209 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
7210 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
7211 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
7212 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
7213 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);