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 DEFINE_PER_CPU(struct kvm_shared_msrs, 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 int kvm_vcpu_reset(struct kvm_vcpu *vcpu);
167 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
170 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
171 vcpu->arch.apf.gfns[i] = ~0;
174 static void kvm_on_user_return(struct user_return_notifier *urn)
177 struct kvm_shared_msrs *locals
178 = container_of(urn, struct kvm_shared_msrs, urn);
179 struct kvm_shared_msr_values *values;
181 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
182 values = &locals->values[slot];
183 if (values->host != values->curr) {
184 wrmsrl(shared_msrs_global.msrs[slot], values->host);
185 values->curr = values->host;
188 locals->registered = false;
189 user_return_notifier_unregister(urn);
192 static void shared_msr_update(unsigned slot, u32 msr)
194 struct kvm_shared_msrs *smsr;
197 smsr = &__get_cpu_var(shared_msrs);
198 /* only read, and nobody should modify it at this time,
199 * so don't need lock */
200 if (slot >= shared_msrs_global.nr) {
201 printk(KERN_ERR "kvm: invalid MSR slot!");
204 rdmsrl_safe(msr, &value);
205 smsr->values[slot].host = value;
206 smsr->values[slot].curr = value;
209 void kvm_define_shared_msr(unsigned slot, u32 msr)
211 if (slot >= shared_msrs_global.nr)
212 shared_msrs_global.nr = slot + 1;
213 shared_msrs_global.msrs[slot] = msr;
214 /* we need ensured the shared_msr_global have been updated */
217 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
219 static void kvm_shared_msr_cpu_online(void)
223 for (i = 0; i < shared_msrs_global.nr; ++i)
224 shared_msr_update(i, shared_msrs_global.msrs[i]);
227 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
229 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
231 if (((value ^ smsr->values[slot].curr) & mask) == 0)
233 smsr->values[slot].curr = value;
234 wrmsrl(shared_msrs_global.msrs[slot], value);
235 if (!smsr->registered) {
236 smsr->urn.on_user_return = kvm_on_user_return;
237 user_return_notifier_register(&smsr->urn);
238 smsr->registered = true;
241 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
243 static void drop_user_return_notifiers(void *ignore)
245 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
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 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
875 /* Update reserved bits */
876 if ((efer ^ old_efer) & EFER_NX)
877 kvm_mmu_reset_context(vcpu);
882 void kvm_enable_efer_bits(u64 mask)
884 efer_reserved_bits &= ~mask;
886 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
890 * Writes msr value into into the appropriate "register".
891 * Returns 0 on success, non-0 otherwise.
892 * Assumes vcpu_load() was already called.
894 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
896 return kvm_x86_ops->set_msr(vcpu, msr);
900 * Adapt set_msr() to msr_io()'s calling convention
902 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
908 msr.host_initiated = true;
909 return kvm_set_msr(vcpu, &msr);
913 struct pvclock_gtod_data {
916 struct { /* extract of a clocksource struct */
924 /* open coded 'struct timespec' */
925 u64 monotonic_time_snsec;
926 time_t monotonic_time_sec;
929 static struct pvclock_gtod_data pvclock_gtod_data;
931 static void update_pvclock_gtod(struct timekeeper *tk)
933 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
935 write_seqcount_begin(&vdata->seq);
937 /* copy pvclock gtod data */
938 vdata->clock.vclock_mode = tk->clock->archdata.vclock_mode;
939 vdata->clock.cycle_last = tk->clock->cycle_last;
940 vdata->clock.mask = tk->clock->mask;
941 vdata->clock.mult = tk->mult;
942 vdata->clock.shift = tk->shift;
944 vdata->monotonic_time_sec = tk->xtime_sec
945 + tk->wall_to_monotonic.tv_sec;
946 vdata->monotonic_time_snsec = tk->xtime_nsec
947 + (tk->wall_to_monotonic.tv_nsec
949 while (vdata->monotonic_time_snsec >=
950 (((u64)NSEC_PER_SEC) << tk->shift)) {
951 vdata->monotonic_time_snsec -=
952 ((u64)NSEC_PER_SEC) << tk->shift;
953 vdata->monotonic_time_sec++;
956 write_seqcount_end(&vdata->seq);
961 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
965 struct pvclock_wall_clock wc;
966 struct timespec boot;
971 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
976 ++version; /* first time write, random junk */
980 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
983 * The guest calculates current wall clock time by adding
984 * system time (updated by kvm_guest_time_update below) to the
985 * wall clock specified here. guest system time equals host
986 * system time for us, thus we must fill in host boot time here.
990 if (kvm->arch.kvmclock_offset) {
991 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
992 boot = timespec_sub(boot, ts);
994 wc.sec = boot.tv_sec;
995 wc.nsec = boot.tv_nsec;
996 wc.version = version;
998 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1001 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1004 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1006 uint32_t quotient, remainder;
1008 /* Don't try to replace with do_div(), this one calculates
1009 * "(dividend << 32) / divisor" */
1011 : "=a" (quotient), "=d" (remainder)
1012 : "0" (0), "1" (dividend), "r" (divisor) );
1016 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1017 s8 *pshift, u32 *pmultiplier)
1024 tps64 = base_khz * 1000LL;
1025 scaled64 = scaled_khz * 1000LL;
1026 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1031 tps32 = (uint32_t)tps64;
1032 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1033 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1041 *pmultiplier = div_frac(scaled64, tps32);
1043 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1044 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1047 static inline u64 get_kernel_ns(void)
1051 WARN_ON(preemptible());
1053 monotonic_to_bootbased(&ts);
1054 return timespec_to_ns(&ts);
1057 #ifdef CONFIG_X86_64
1058 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1061 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1062 unsigned long max_tsc_khz;
1064 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1066 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1067 vcpu->arch.virtual_tsc_shift);
1070 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1072 u64 v = (u64)khz * (1000000 + ppm);
1077 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1079 u32 thresh_lo, thresh_hi;
1080 int use_scaling = 0;
1082 /* Compute a scale to convert nanoseconds in TSC cycles */
1083 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1084 &vcpu->arch.virtual_tsc_shift,
1085 &vcpu->arch.virtual_tsc_mult);
1086 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1089 * Compute the variation in TSC rate which is acceptable
1090 * within the range of tolerance and decide if the
1091 * rate being applied is within that bounds of the hardware
1092 * rate. If so, no scaling or compensation need be done.
1094 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1095 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1096 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1097 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1100 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1103 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1105 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1106 vcpu->arch.virtual_tsc_mult,
1107 vcpu->arch.virtual_tsc_shift);
1108 tsc += vcpu->arch.this_tsc_write;
1112 void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1114 #ifdef CONFIG_X86_64
1116 bool do_request = false;
1117 struct kvm_arch *ka = &vcpu->kvm->arch;
1118 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1120 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1121 atomic_read(&vcpu->kvm->online_vcpus));
1123 if (vcpus_matched && gtod->clock.vclock_mode == VCLOCK_TSC)
1124 if (!ka->use_master_clock)
1127 if (!vcpus_matched && ka->use_master_clock)
1131 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1133 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1134 atomic_read(&vcpu->kvm->online_vcpus),
1135 ka->use_master_clock, gtod->clock.vclock_mode);
1139 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1141 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1142 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1145 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1147 struct kvm *kvm = vcpu->kvm;
1148 u64 offset, ns, elapsed;
1149 unsigned long flags;
1152 u64 data = msr->data;
1154 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1155 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1156 ns = get_kernel_ns();
1157 elapsed = ns - kvm->arch.last_tsc_nsec;
1159 /* n.b - signed multiplication and division required */
1160 usdiff = data - kvm->arch.last_tsc_write;
1161 #ifdef CONFIG_X86_64
1162 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1164 /* do_div() only does unsigned */
1165 asm("idivl %2; xor %%edx, %%edx"
1167 : "A"(usdiff * 1000), "rm"(vcpu->arch.virtual_tsc_khz));
1169 do_div(elapsed, 1000);
1175 * Special case: TSC write with a small delta (1 second) of virtual
1176 * cycle time against real time is interpreted as an attempt to
1177 * synchronize the CPU.
1179 * For a reliable TSC, we can match TSC offsets, and for an unstable
1180 * TSC, we add elapsed time in this computation. We could let the
1181 * compensation code attempt to catch up if we fall behind, but
1182 * it's better to try to match offsets from the beginning.
1184 if (usdiff < USEC_PER_SEC &&
1185 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1186 if (!check_tsc_unstable()) {
1187 offset = kvm->arch.cur_tsc_offset;
1188 pr_debug("kvm: matched tsc offset for %llu\n", data);
1190 u64 delta = nsec_to_cycles(vcpu, elapsed);
1192 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1193 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1198 * We split periods of matched TSC writes into generations.
1199 * For each generation, we track the original measured
1200 * nanosecond time, offset, and write, so if TSCs are in
1201 * sync, we can match exact offset, and if not, we can match
1202 * exact software computation in compute_guest_tsc()
1204 * These values are tracked in kvm->arch.cur_xxx variables.
1206 kvm->arch.cur_tsc_generation++;
1207 kvm->arch.cur_tsc_nsec = ns;
1208 kvm->arch.cur_tsc_write = data;
1209 kvm->arch.cur_tsc_offset = offset;
1211 pr_debug("kvm: new tsc generation %u, clock %llu\n",
1212 kvm->arch.cur_tsc_generation, data);
1216 * We also track th most recent recorded KHZ, write and time to
1217 * allow the matching interval to be extended at each write.
1219 kvm->arch.last_tsc_nsec = ns;
1220 kvm->arch.last_tsc_write = data;
1221 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1223 /* Reset of TSC must disable overshoot protection below */
1224 vcpu->arch.hv_clock.tsc_timestamp = 0;
1225 vcpu->arch.last_guest_tsc = data;
1227 /* Keep track of which generation this VCPU has synchronized to */
1228 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1229 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1230 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1232 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1233 update_ia32_tsc_adjust_msr(vcpu, offset);
1234 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1235 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1237 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1239 kvm->arch.nr_vcpus_matched_tsc++;
1241 kvm->arch.nr_vcpus_matched_tsc = 0;
1243 kvm_track_tsc_matching(vcpu);
1244 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1247 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1249 #ifdef CONFIG_X86_64
1251 static cycle_t read_tsc(void)
1257 * Empirically, a fence (of type that depends on the CPU)
1258 * before rdtsc is enough to ensure that rdtsc is ordered
1259 * with respect to loads. The various CPU manuals are unclear
1260 * as to whether rdtsc can be reordered with later loads,
1261 * but no one has ever seen it happen.
1264 ret = (cycle_t)vget_cycles();
1266 last = pvclock_gtod_data.clock.cycle_last;
1268 if (likely(ret >= last))
1272 * GCC likes to generate cmov here, but this branch is extremely
1273 * predictable (it's just a funciton of time and the likely is
1274 * very likely) and there's a data dependence, so force GCC
1275 * to generate a branch instead. I don't barrier() because
1276 * we don't actually need a barrier, and if this function
1277 * ever gets inlined it will generate worse code.
1283 static inline u64 vgettsc(cycle_t *cycle_now)
1286 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1288 *cycle_now = read_tsc();
1290 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1291 return v * gtod->clock.mult;
1294 static int do_monotonic(struct timespec *ts, cycle_t *cycle_now)
1299 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1303 seq = read_seqcount_begin(>od->seq);
1304 mode = gtod->clock.vclock_mode;
1305 ts->tv_sec = gtod->monotonic_time_sec;
1306 ns = gtod->monotonic_time_snsec;
1307 ns += vgettsc(cycle_now);
1308 ns >>= gtod->clock.shift;
1309 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1310 timespec_add_ns(ts, ns);
1315 /* returns true if host is using tsc clocksource */
1316 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1320 /* checked again under seqlock below */
1321 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1324 if (do_monotonic(&ts, cycle_now) != VCLOCK_TSC)
1327 monotonic_to_bootbased(&ts);
1328 *kernel_ns = timespec_to_ns(&ts);
1336 * Assuming a stable TSC across physical CPUS, and a stable TSC
1337 * across virtual CPUs, the following condition is possible.
1338 * Each numbered line represents an event visible to both
1339 * CPUs at the next numbered event.
1341 * "timespecX" represents host monotonic time. "tscX" represents
1344 * VCPU0 on CPU0 | VCPU1 on CPU1
1346 * 1. read timespec0,tsc0
1347 * 2. | timespec1 = timespec0 + N
1349 * 3. transition to guest | transition to guest
1350 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1351 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1352 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1354 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1357 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1359 * - 0 < N - M => M < N
1361 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1362 * always the case (the difference between two distinct xtime instances
1363 * might be smaller then the difference between corresponding TSC reads,
1364 * when updating guest vcpus pvclock areas).
1366 * To avoid that problem, do not allow visibility of distinct
1367 * system_timestamp/tsc_timestamp values simultaneously: use a master
1368 * copy of host monotonic time values. Update that master copy
1371 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1375 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1377 #ifdef CONFIG_X86_64
1378 struct kvm_arch *ka = &kvm->arch;
1380 bool host_tsc_clocksource, vcpus_matched;
1382 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1383 atomic_read(&kvm->online_vcpus));
1386 * If the host uses TSC clock, then passthrough TSC as stable
1389 host_tsc_clocksource = kvm_get_time_and_clockread(
1390 &ka->master_kernel_ns,
1391 &ka->master_cycle_now);
1393 ka->use_master_clock = host_tsc_clocksource & vcpus_matched;
1395 if (ka->use_master_clock)
1396 atomic_set(&kvm_guest_has_master_clock, 1);
1398 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1399 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1404 static int kvm_guest_time_update(struct kvm_vcpu *v)
1406 unsigned long flags, this_tsc_khz;
1407 struct kvm_vcpu_arch *vcpu = &v->arch;
1408 struct kvm_arch *ka = &v->kvm->arch;
1410 s64 kernel_ns, max_kernel_ns;
1411 u64 tsc_timestamp, host_tsc;
1412 struct pvclock_vcpu_time_info *guest_hv_clock;
1414 bool use_master_clock;
1419 /* Keep irq disabled to prevent changes to the clock */
1420 local_irq_save(flags);
1421 this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
1422 if (unlikely(this_tsc_khz == 0)) {
1423 local_irq_restore(flags);
1424 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1429 * If the host uses TSC clock, then passthrough TSC as stable
1432 spin_lock(&ka->pvclock_gtod_sync_lock);
1433 use_master_clock = ka->use_master_clock;
1434 if (use_master_clock) {
1435 host_tsc = ka->master_cycle_now;
1436 kernel_ns = ka->master_kernel_ns;
1438 spin_unlock(&ka->pvclock_gtod_sync_lock);
1439 if (!use_master_clock) {
1440 host_tsc = native_read_tsc();
1441 kernel_ns = get_kernel_ns();
1444 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1447 * We may have to catch up the TSC to match elapsed wall clock
1448 * time for two reasons, even if kvmclock is used.
1449 * 1) CPU could have been running below the maximum TSC rate
1450 * 2) Broken TSC compensation resets the base at each VCPU
1451 * entry to avoid unknown leaps of TSC even when running
1452 * again on the same CPU. This may cause apparent elapsed
1453 * time to disappear, and the guest to stand still or run
1456 if (vcpu->tsc_catchup) {
1457 u64 tsc = compute_guest_tsc(v, kernel_ns);
1458 if (tsc > tsc_timestamp) {
1459 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1460 tsc_timestamp = tsc;
1464 local_irq_restore(flags);
1466 if (!vcpu->time_page)
1470 * Time as measured by the TSC may go backwards when resetting the base
1471 * tsc_timestamp. The reason for this is that the TSC resolution is
1472 * higher than the resolution of the other clock scales. Thus, many
1473 * possible measurments of the TSC correspond to one measurement of any
1474 * other clock, and so a spread of values is possible. This is not a
1475 * problem for the computation of the nanosecond clock; with TSC rates
1476 * around 1GHZ, there can only be a few cycles which correspond to one
1477 * nanosecond value, and any path through this code will inevitably
1478 * take longer than that. However, with the kernel_ns value itself,
1479 * the precision may be much lower, down to HZ granularity. If the
1480 * first sampling of TSC against kernel_ns ends in the low part of the
1481 * range, and the second in the high end of the range, we can get:
1483 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1485 * As the sampling errors potentially range in the thousands of cycles,
1486 * it is possible such a time value has already been observed by the
1487 * guest. To protect against this, we must compute the system time as
1488 * observed by the guest and ensure the new system time is greater.
1491 if (vcpu->hv_clock.tsc_timestamp) {
1492 max_kernel_ns = vcpu->last_guest_tsc -
1493 vcpu->hv_clock.tsc_timestamp;
1494 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1495 vcpu->hv_clock.tsc_to_system_mul,
1496 vcpu->hv_clock.tsc_shift);
1497 max_kernel_ns += vcpu->last_kernel_ns;
1500 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1501 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1502 &vcpu->hv_clock.tsc_shift,
1503 &vcpu->hv_clock.tsc_to_system_mul);
1504 vcpu->hw_tsc_khz = this_tsc_khz;
1507 /* with a master <monotonic time, tsc value> tuple,
1508 * pvclock clock reads always increase at the (scaled) rate
1509 * of guest TSC - no need to deal with sampling errors.
1511 if (!use_master_clock) {
1512 if (max_kernel_ns > kernel_ns)
1513 kernel_ns = max_kernel_ns;
1515 /* With all the info we got, fill in the values */
1516 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1517 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1518 vcpu->last_kernel_ns = kernel_ns;
1519 vcpu->last_guest_tsc = tsc_timestamp;
1522 * The interface expects us to write an even number signaling that the
1523 * update is finished. Since the guest won't see the intermediate
1524 * state, we just increase by 2 at the end.
1526 vcpu->hv_clock.version += 2;
1528 shared_kaddr = kmap_atomic(vcpu->time_page);
1530 guest_hv_clock = shared_kaddr + vcpu->time_offset;
1532 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1533 pvclock_flags = (guest_hv_clock->flags & PVCLOCK_GUEST_STOPPED);
1535 if (vcpu->pvclock_set_guest_stopped_request) {
1536 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1537 vcpu->pvclock_set_guest_stopped_request = false;
1540 /* If the host uses TSC clocksource, then it is stable */
1541 if (use_master_clock)
1542 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1544 vcpu->hv_clock.flags = pvclock_flags;
1546 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1547 sizeof(vcpu->hv_clock));
1549 kunmap_atomic(shared_kaddr);
1551 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1555 static bool msr_mtrr_valid(unsigned msr)
1558 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1559 case MSR_MTRRfix64K_00000:
1560 case MSR_MTRRfix16K_80000:
1561 case MSR_MTRRfix16K_A0000:
1562 case MSR_MTRRfix4K_C0000:
1563 case MSR_MTRRfix4K_C8000:
1564 case MSR_MTRRfix4K_D0000:
1565 case MSR_MTRRfix4K_D8000:
1566 case MSR_MTRRfix4K_E0000:
1567 case MSR_MTRRfix4K_E8000:
1568 case MSR_MTRRfix4K_F0000:
1569 case MSR_MTRRfix4K_F8000:
1570 case MSR_MTRRdefType:
1571 case MSR_IA32_CR_PAT:
1579 static bool valid_pat_type(unsigned t)
1581 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1584 static bool valid_mtrr_type(unsigned t)
1586 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1589 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1593 if (!msr_mtrr_valid(msr))
1596 if (msr == MSR_IA32_CR_PAT) {
1597 for (i = 0; i < 8; i++)
1598 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1601 } else if (msr == MSR_MTRRdefType) {
1604 return valid_mtrr_type(data & 0xff);
1605 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1606 for (i = 0; i < 8 ; i++)
1607 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1612 /* variable MTRRs */
1613 return valid_mtrr_type(data & 0xff);
1616 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1618 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1620 if (!mtrr_valid(vcpu, msr, data))
1623 if (msr == MSR_MTRRdefType) {
1624 vcpu->arch.mtrr_state.def_type = data;
1625 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1626 } else if (msr == MSR_MTRRfix64K_00000)
1628 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1629 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1630 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1631 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1632 else if (msr == MSR_IA32_CR_PAT)
1633 vcpu->arch.pat = data;
1634 else { /* Variable MTRRs */
1635 int idx, is_mtrr_mask;
1638 idx = (msr - 0x200) / 2;
1639 is_mtrr_mask = msr - 0x200 - 2 * idx;
1642 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1645 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1649 kvm_mmu_reset_context(vcpu);
1653 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1655 u64 mcg_cap = vcpu->arch.mcg_cap;
1656 unsigned bank_num = mcg_cap & 0xff;
1659 case MSR_IA32_MCG_STATUS:
1660 vcpu->arch.mcg_status = data;
1662 case MSR_IA32_MCG_CTL:
1663 if (!(mcg_cap & MCG_CTL_P))
1665 if (data != 0 && data != ~(u64)0)
1667 vcpu->arch.mcg_ctl = data;
1670 if (msr >= MSR_IA32_MC0_CTL &&
1671 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1672 u32 offset = msr - MSR_IA32_MC0_CTL;
1673 /* only 0 or all 1s can be written to IA32_MCi_CTL
1674 * some Linux kernels though clear bit 10 in bank 4 to
1675 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1676 * this to avoid an uncatched #GP in the guest
1678 if ((offset & 0x3) == 0 &&
1679 data != 0 && (data | (1 << 10)) != ~(u64)0)
1681 vcpu->arch.mce_banks[offset] = data;
1689 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1691 struct kvm *kvm = vcpu->kvm;
1692 int lm = is_long_mode(vcpu);
1693 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1694 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1695 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1696 : kvm->arch.xen_hvm_config.blob_size_32;
1697 u32 page_num = data & ~PAGE_MASK;
1698 u64 page_addr = data & PAGE_MASK;
1703 if (page_num >= blob_size)
1706 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1711 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1720 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1722 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1725 static bool kvm_hv_msr_partition_wide(u32 msr)
1729 case HV_X64_MSR_GUEST_OS_ID:
1730 case HV_X64_MSR_HYPERCALL:
1738 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1740 struct kvm *kvm = vcpu->kvm;
1743 case HV_X64_MSR_GUEST_OS_ID:
1744 kvm->arch.hv_guest_os_id = data;
1745 /* setting guest os id to zero disables hypercall page */
1746 if (!kvm->arch.hv_guest_os_id)
1747 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1749 case HV_X64_MSR_HYPERCALL: {
1754 /* if guest os id is not set hypercall should remain disabled */
1755 if (!kvm->arch.hv_guest_os_id)
1757 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1758 kvm->arch.hv_hypercall = data;
1761 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1762 addr = gfn_to_hva(kvm, gfn);
1763 if (kvm_is_error_hva(addr))
1765 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1766 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1767 if (__copy_to_user((void __user *)addr, instructions, 4))
1769 kvm->arch.hv_hypercall = data;
1773 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1774 "data 0x%llx\n", msr, data);
1780 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1783 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1786 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1787 vcpu->arch.hv_vapic = data;
1790 addr = gfn_to_hva(vcpu->kvm, data >>
1791 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1792 if (kvm_is_error_hva(addr))
1794 if (__clear_user((void __user *)addr, PAGE_SIZE))
1796 vcpu->arch.hv_vapic = data;
1799 case HV_X64_MSR_EOI:
1800 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1801 case HV_X64_MSR_ICR:
1802 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1803 case HV_X64_MSR_TPR:
1804 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1806 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1807 "data 0x%llx\n", msr, data);
1814 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1816 gpa_t gpa = data & ~0x3f;
1818 /* Bits 2:5 are reserved, Should be zero */
1822 vcpu->arch.apf.msr_val = data;
1824 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1825 kvm_clear_async_pf_completion_queue(vcpu);
1826 kvm_async_pf_hash_reset(vcpu);
1830 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1833 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1834 kvm_async_pf_wakeup_all(vcpu);
1838 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1840 if (vcpu->arch.time_page) {
1841 kvm_release_page_dirty(vcpu->arch.time_page);
1842 vcpu->arch.time_page = NULL;
1846 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1850 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1853 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1854 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1855 vcpu->arch.st.accum_steal = delta;
1858 static void record_steal_time(struct kvm_vcpu *vcpu)
1860 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1863 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1864 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1867 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1868 vcpu->arch.st.steal.version += 2;
1869 vcpu->arch.st.accum_steal = 0;
1871 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1872 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1875 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1878 u32 msr = msr_info->index;
1879 u64 data = msr_info->data;
1883 return set_efer(vcpu, data);
1885 data &= ~(u64)0x40; /* ignore flush filter disable */
1886 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1887 data &= ~(u64)0x8; /* ignore TLB cache disable */
1889 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1894 case MSR_FAM10H_MMIO_CONF_BASE:
1896 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1901 case MSR_AMD64_NB_CFG:
1903 case MSR_IA32_DEBUGCTLMSR:
1905 /* We support the non-activated case already */
1907 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1908 /* Values other than LBR and BTF are vendor-specific,
1909 thus reserved and should throw a #GP */
1912 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1915 case MSR_IA32_UCODE_REV:
1916 case MSR_IA32_UCODE_WRITE:
1917 case MSR_VM_HSAVE_PA:
1918 case MSR_AMD64_PATCH_LOADER:
1920 case 0x200 ... 0x2ff:
1921 return set_msr_mtrr(vcpu, msr, data);
1922 case MSR_IA32_APICBASE:
1923 kvm_set_apic_base(vcpu, data);
1925 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1926 return kvm_x2apic_msr_write(vcpu, msr, data);
1927 case MSR_IA32_TSCDEADLINE:
1928 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1930 case MSR_IA32_TSC_ADJUST:
1931 if (guest_cpuid_has_tsc_adjust(vcpu)) {
1932 if (!msr_info->host_initiated) {
1933 u64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
1934 kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
1936 vcpu->arch.ia32_tsc_adjust_msr = data;
1939 case MSR_IA32_MISC_ENABLE:
1940 vcpu->arch.ia32_misc_enable_msr = data;
1942 case MSR_KVM_WALL_CLOCK_NEW:
1943 case MSR_KVM_WALL_CLOCK:
1944 vcpu->kvm->arch.wall_clock = data;
1945 kvm_write_wall_clock(vcpu->kvm, data);
1947 case MSR_KVM_SYSTEM_TIME_NEW:
1948 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 /* ...but clean it before doing the actual write */
1959 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1961 vcpu->arch.time_page =
1962 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1964 if (is_error_page(vcpu->arch.time_page))
1965 vcpu->arch.time_page = NULL;
1969 case MSR_KVM_ASYNC_PF_EN:
1970 if (kvm_pv_enable_async_pf(vcpu, data))
1973 case MSR_KVM_STEAL_TIME:
1975 if (unlikely(!sched_info_on()))
1978 if (data & KVM_STEAL_RESERVED_MASK)
1981 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
1982 data & KVM_STEAL_VALID_BITS))
1985 vcpu->arch.st.msr_val = data;
1987 if (!(data & KVM_MSR_ENABLED))
1990 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1993 accumulate_steal_time(vcpu);
1996 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
1999 case MSR_KVM_PV_EOI_EN:
2000 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2004 case MSR_IA32_MCG_CTL:
2005 case MSR_IA32_MCG_STATUS:
2006 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2007 return set_msr_mce(vcpu, msr, data);
2009 /* Performance counters are not protected by a CPUID bit,
2010 * so we should check all of them in the generic path for the sake of
2011 * cross vendor migration.
2012 * Writing a zero into the event select MSRs disables them,
2013 * which we perfectly emulate ;-). Any other value should be at least
2014 * reported, some guests depend on them.
2016 case MSR_K7_EVNTSEL0:
2017 case MSR_K7_EVNTSEL1:
2018 case MSR_K7_EVNTSEL2:
2019 case MSR_K7_EVNTSEL3:
2021 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2022 "0x%x data 0x%llx\n", msr, data);
2024 /* at least RHEL 4 unconditionally writes to the perfctr registers,
2025 * so we ignore writes to make it happy.
2027 case MSR_K7_PERFCTR0:
2028 case MSR_K7_PERFCTR1:
2029 case MSR_K7_PERFCTR2:
2030 case MSR_K7_PERFCTR3:
2031 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2032 "0x%x data 0x%llx\n", msr, data);
2034 case MSR_P6_PERFCTR0:
2035 case MSR_P6_PERFCTR1:
2037 case MSR_P6_EVNTSEL0:
2038 case MSR_P6_EVNTSEL1:
2039 if (kvm_pmu_msr(vcpu, msr))
2040 return kvm_pmu_set_msr(vcpu, msr, data);
2042 if (pr || data != 0)
2043 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2044 "0x%x data 0x%llx\n", msr, data);
2046 case MSR_K7_CLK_CTL:
2048 * Ignore all writes to this no longer documented MSR.
2049 * Writes are only relevant for old K7 processors,
2050 * all pre-dating SVM, but a recommended workaround from
2051 * AMD for these chips. It is possible to specify the
2052 * affected processor models on the command line, hence
2053 * the need to ignore the workaround.
2056 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2057 if (kvm_hv_msr_partition_wide(msr)) {
2059 mutex_lock(&vcpu->kvm->lock);
2060 r = set_msr_hyperv_pw(vcpu, msr, data);
2061 mutex_unlock(&vcpu->kvm->lock);
2064 return set_msr_hyperv(vcpu, msr, data);
2066 case MSR_IA32_BBL_CR_CTL3:
2067 /* Drop writes to this legacy MSR -- see rdmsr
2068 * counterpart for further detail.
2070 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2072 case MSR_AMD64_OSVW_ID_LENGTH:
2073 if (!guest_cpuid_has_osvw(vcpu))
2075 vcpu->arch.osvw.length = data;
2077 case MSR_AMD64_OSVW_STATUS:
2078 if (!guest_cpuid_has_osvw(vcpu))
2080 vcpu->arch.osvw.status = data;
2083 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2084 return xen_hvm_config(vcpu, data);
2085 if (kvm_pmu_msr(vcpu, msr))
2086 return kvm_pmu_set_msr(vcpu, msr, data);
2088 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2092 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2099 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2103 * Reads an msr value (of 'msr_index') into 'pdata'.
2104 * Returns 0 on success, non-0 otherwise.
2105 * Assumes vcpu_load() was already called.
2107 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2109 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
2112 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2114 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2116 if (!msr_mtrr_valid(msr))
2119 if (msr == MSR_MTRRdefType)
2120 *pdata = vcpu->arch.mtrr_state.def_type +
2121 (vcpu->arch.mtrr_state.enabled << 10);
2122 else if (msr == MSR_MTRRfix64K_00000)
2124 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2125 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2126 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2127 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2128 else if (msr == MSR_IA32_CR_PAT)
2129 *pdata = vcpu->arch.pat;
2130 else { /* Variable MTRRs */
2131 int idx, is_mtrr_mask;
2134 idx = (msr - 0x200) / 2;
2135 is_mtrr_mask = msr - 0x200 - 2 * idx;
2138 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2141 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2148 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2151 u64 mcg_cap = vcpu->arch.mcg_cap;
2152 unsigned bank_num = mcg_cap & 0xff;
2155 case MSR_IA32_P5_MC_ADDR:
2156 case MSR_IA32_P5_MC_TYPE:
2159 case MSR_IA32_MCG_CAP:
2160 data = vcpu->arch.mcg_cap;
2162 case MSR_IA32_MCG_CTL:
2163 if (!(mcg_cap & MCG_CTL_P))
2165 data = vcpu->arch.mcg_ctl;
2167 case MSR_IA32_MCG_STATUS:
2168 data = vcpu->arch.mcg_status;
2171 if (msr >= MSR_IA32_MC0_CTL &&
2172 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
2173 u32 offset = msr - MSR_IA32_MC0_CTL;
2174 data = vcpu->arch.mce_banks[offset];
2183 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2186 struct kvm *kvm = vcpu->kvm;
2189 case HV_X64_MSR_GUEST_OS_ID:
2190 data = kvm->arch.hv_guest_os_id;
2192 case HV_X64_MSR_HYPERCALL:
2193 data = kvm->arch.hv_hypercall;
2196 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2204 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2209 case HV_X64_MSR_VP_INDEX: {
2212 kvm_for_each_vcpu(r, v, vcpu->kvm)
2217 case HV_X64_MSR_EOI:
2218 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2219 case HV_X64_MSR_ICR:
2220 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2221 case HV_X64_MSR_TPR:
2222 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
2223 case HV_X64_MSR_APIC_ASSIST_PAGE:
2224 data = vcpu->arch.hv_vapic;
2227 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2234 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2239 case MSR_IA32_PLATFORM_ID:
2240 case MSR_IA32_EBL_CR_POWERON:
2241 case MSR_IA32_DEBUGCTLMSR:
2242 case MSR_IA32_LASTBRANCHFROMIP:
2243 case MSR_IA32_LASTBRANCHTOIP:
2244 case MSR_IA32_LASTINTFROMIP:
2245 case MSR_IA32_LASTINTTOIP:
2248 case MSR_VM_HSAVE_PA:
2249 case MSR_K7_EVNTSEL0:
2250 case MSR_K7_PERFCTR0:
2251 case MSR_K8_INT_PENDING_MSG:
2252 case MSR_AMD64_NB_CFG:
2253 case MSR_FAM10H_MMIO_CONF_BASE:
2256 case MSR_P6_PERFCTR0:
2257 case MSR_P6_PERFCTR1:
2258 case MSR_P6_EVNTSEL0:
2259 case MSR_P6_EVNTSEL1:
2260 if (kvm_pmu_msr(vcpu, msr))
2261 return kvm_pmu_get_msr(vcpu, msr, pdata);
2264 case MSR_IA32_UCODE_REV:
2265 data = 0x100000000ULL;
2268 data = 0x500 | KVM_NR_VAR_MTRR;
2270 case 0x200 ... 0x2ff:
2271 return get_msr_mtrr(vcpu, msr, pdata);
2272 case 0xcd: /* fsb frequency */
2276 * MSR_EBC_FREQUENCY_ID
2277 * Conservative value valid for even the basic CPU models.
2278 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2279 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2280 * and 266MHz for model 3, or 4. Set Core Clock
2281 * Frequency to System Bus Frequency Ratio to 1 (bits
2282 * 31:24) even though these are only valid for CPU
2283 * models > 2, however guests may end up dividing or
2284 * multiplying by zero otherwise.
2286 case MSR_EBC_FREQUENCY_ID:
2289 case MSR_IA32_APICBASE:
2290 data = kvm_get_apic_base(vcpu);
2292 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2293 return kvm_x2apic_msr_read(vcpu, msr, pdata);
2295 case MSR_IA32_TSCDEADLINE:
2296 data = kvm_get_lapic_tscdeadline_msr(vcpu);
2298 case MSR_IA32_TSC_ADJUST:
2299 data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2301 case MSR_IA32_MISC_ENABLE:
2302 data = vcpu->arch.ia32_misc_enable_msr;
2304 case MSR_IA32_PERF_STATUS:
2305 /* TSC increment by tick */
2307 /* CPU multiplier */
2308 data |= (((uint64_t)4ULL) << 40);
2311 data = vcpu->arch.efer;
2313 case MSR_KVM_WALL_CLOCK:
2314 case MSR_KVM_WALL_CLOCK_NEW:
2315 data = vcpu->kvm->arch.wall_clock;
2317 case MSR_KVM_SYSTEM_TIME:
2318 case MSR_KVM_SYSTEM_TIME_NEW:
2319 data = vcpu->arch.time;
2321 case MSR_KVM_ASYNC_PF_EN:
2322 data = vcpu->arch.apf.msr_val;
2324 case MSR_KVM_STEAL_TIME:
2325 data = vcpu->arch.st.msr_val;
2327 case MSR_KVM_PV_EOI_EN:
2328 data = vcpu->arch.pv_eoi.msr_val;
2330 case MSR_IA32_P5_MC_ADDR:
2331 case MSR_IA32_P5_MC_TYPE:
2332 case MSR_IA32_MCG_CAP:
2333 case MSR_IA32_MCG_CTL:
2334 case MSR_IA32_MCG_STATUS:
2335 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2336 return get_msr_mce(vcpu, msr, pdata);
2337 case MSR_K7_CLK_CTL:
2339 * Provide expected ramp-up count for K7. All other
2340 * are set to zero, indicating minimum divisors for
2343 * This prevents guest kernels on AMD host with CPU
2344 * type 6, model 8 and higher from exploding due to
2345 * the rdmsr failing.
2349 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2350 if (kvm_hv_msr_partition_wide(msr)) {
2352 mutex_lock(&vcpu->kvm->lock);
2353 r = get_msr_hyperv_pw(vcpu, msr, pdata);
2354 mutex_unlock(&vcpu->kvm->lock);
2357 return get_msr_hyperv(vcpu, msr, pdata);
2359 case MSR_IA32_BBL_CR_CTL3:
2360 /* This legacy MSR exists but isn't fully documented in current
2361 * silicon. It is however accessed by winxp in very narrow
2362 * scenarios where it sets bit #19, itself documented as
2363 * a "reserved" bit. Best effort attempt to source coherent
2364 * read data here should the balance of the register be
2365 * interpreted by the guest:
2367 * L2 cache control register 3: 64GB range, 256KB size,
2368 * enabled, latency 0x1, configured
2372 case MSR_AMD64_OSVW_ID_LENGTH:
2373 if (!guest_cpuid_has_osvw(vcpu))
2375 data = vcpu->arch.osvw.length;
2377 case MSR_AMD64_OSVW_STATUS:
2378 if (!guest_cpuid_has_osvw(vcpu))
2380 data = vcpu->arch.osvw.status;
2383 if (kvm_pmu_msr(vcpu, msr))
2384 return kvm_pmu_get_msr(vcpu, msr, pdata);
2386 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2389 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2397 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2400 * Read or write a bunch of msrs. All parameters are kernel addresses.
2402 * @return number of msrs set successfully.
2404 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2405 struct kvm_msr_entry *entries,
2406 int (*do_msr)(struct kvm_vcpu *vcpu,
2407 unsigned index, u64 *data))
2411 idx = srcu_read_lock(&vcpu->kvm->srcu);
2412 for (i = 0; i < msrs->nmsrs; ++i)
2413 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2415 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2421 * Read or write a bunch of msrs. Parameters are user addresses.
2423 * @return number of msrs set successfully.
2425 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2426 int (*do_msr)(struct kvm_vcpu *vcpu,
2427 unsigned index, u64 *data),
2430 struct kvm_msrs msrs;
2431 struct kvm_msr_entry *entries;
2436 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2440 if (msrs.nmsrs >= MAX_IO_MSRS)
2443 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2444 entries = memdup_user(user_msrs->entries, size);
2445 if (IS_ERR(entries)) {
2446 r = PTR_ERR(entries);
2450 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2455 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2466 int kvm_dev_ioctl_check_extension(long ext)
2471 case KVM_CAP_IRQCHIP:
2473 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2474 case KVM_CAP_SET_TSS_ADDR:
2475 case KVM_CAP_EXT_CPUID:
2476 case KVM_CAP_CLOCKSOURCE:
2478 case KVM_CAP_NOP_IO_DELAY:
2479 case KVM_CAP_MP_STATE:
2480 case KVM_CAP_SYNC_MMU:
2481 case KVM_CAP_USER_NMI:
2482 case KVM_CAP_REINJECT_CONTROL:
2483 case KVM_CAP_IRQ_INJECT_STATUS:
2484 case KVM_CAP_ASSIGN_DEV_IRQ:
2486 case KVM_CAP_IOEVENTFD:
2488 case KVM_CAP_PIT_STATE2:
2489 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2490 case KVM_CAP_XEN_HVM:
2491 case KVM_CAP_ADJUST_CLOCK:
2492 case KVM_CAP_VCPU_EVENTS:
2493 case KVM_CAP_HYPERV:
2494 case KVM_CAP_HYPERV_VAPIC:
2495 case KVM_CAP_HYPERV_SPIN:
2496 case KVM_CAP_PCI_SEGMENT:
2497 case KVM_CAP_DEBUGREGS:
2498 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2500 case KVM_CAP_ASYNC_PF:
2501 case KVM_CAP_GET_TSC_KHZ:
2502 case KVM_CAP_PCI_2_3:
2503 case KVM_CAP_KVMCLOCK_CTRL:
2504 case KVM_CAP_READONLY_MEM:
2505 case KVM_CAP_IRQFD_RESAMPLE:
2508 case KVM_CAP_COALESCED_MMIO:
2509 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2512 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2514 case KVM_CAP_NR_VCPUS:
2515 r = KVM_SOFT_MAX_VCPUS;
2517 case KVM_CAP_MAX_VCPUS:
2520 case KVM_CAP_NR_MEMSLOTS:
2521 r = KVM_USER_MEM_SLOTS;
2523 case KVM_CAP_PV_MMU: /* obsolete */
2527 r = iommu_present(&pci_bus_type);
2530 r = KVM_MAX_MCE_BANKS;
2535 case KVM_CAP_TSC_CONTROL:
2536 r = kvm_has_tsc_control;
2538 case KVM_CAP_TSC_DEADLINE_TIMER:
2539 r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2549 long kvm_arch_dev_ioctl(struct file *filp,
2550 unsigned int ioctl, unsigned long arg)
2552 void __user *argp = (void __user *)arg;
2556 case KVM_GET_MSR_INDEX_LIST: {
2557 struct kvm_msr_list __user *user_msr_list = argp;
2558 struct kvm_msr_list msr_list;
2562 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2565 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2566 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2569 if (n < msr_list.nmsrs)
2572 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2573 num_msrs_to_save * sizeof(u32)))
2575 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2577 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2582 case KVM_GET_SUPPORTED_CPUID: {
2583 struct kvm_cpuid2 __user *cpuid_arg = argp;
2584 struct kvm_cpuid2 cpuid;
2587 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2589 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2590 cpuid_arg->entries);
2595 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2600 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2603 mce_cap = KVM_MCE_CAP_SUPPORTED;
2605 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2617 static void wbinvd_ipi(void *garbage)
2622 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2624 return vcpu->kvm->arch.iommu_domain &&
2625 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2628 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2630 /* Address WBINVD may be executed by guest */
2631 if (need_emulate_wbinvd(vcpu)) {
2632 if (kvm_x86_ops->has_wbinvd_exit())
2633 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2634 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2635 smp_call_function_single(vcpu->cpu,
2636 wbinvd_ipi, NULL, 1);
2639 kvm_x86_ops->vcpu_load(vcpu, cpu);
2641 /* Apply any externally detected TSC adjustments (due to suspend) */
2642 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2643 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2644 vcpu->arch.tsc_offset_adjustment = 0;
2645 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
2648 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2649 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2650 native_read_tsc() - vcpu->arch.last_host_tsc;
2652 mark_tsc_unstable("KVM discovered backwards TSC");
2653 if (check_tsc_unstable()) {
2654 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2655 vcpu->arch.last_guest_tsc);
2656 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2657 vcpu->arch.tsc_catchup = 1;
2660 * On a host with synchronized TSC, there is no need to update
2661 * kvmclock on vcpu->cpu migration
2663 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2664 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2665 if (vcpu->cpu != cpu)
2666 kvm_migrate_timers(vcpu);
2670 accumulate_steal_time(vcpu);
2671 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2674 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2676 kvm_x86_ops->vcpu_put(vcpu);
2677 kvm_put_guest_fpu(vcpu);
2678 vcpu->arch.last_host_tsc = native_read_tsc();
2681 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2682 struct kvm_lapic_state *s)
2684 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2689 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2690 struct kvm_lapic_state *s)
2692 kvm_apic_post_state_restore(vcpu, s);
2693 update_cr8_intercept(vcpu);
2698 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2699 struct kvm_interrupt *irq)
2701 if (irq->irq < 0 || irq->irq >= KVM_NR_INTERRUPTS)
2703 if (irqchip_in_kernel(vcpu->kvm))
2706 kvm_queue_interrupt(vcpu, irq->irq, false);
2707 kvm_make_request(KVM_REQ_EVENT, vcpu);
2712 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2714 kvm_inject_nmi(vcpu);
2719 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2720 struct kvm_tpr_access_ctl *tac)
2724 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2728 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2732 unsigned bank_num = mcg_cap & 0xff, bank;
2735 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2737 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2740 vcpu->arch.mcg_cap = mcg_cap;
2741 /* Init IA32_MCG_CTL to all 1s */
2742 if (mcg_cap & MCG_CTL_P)
2743 vcpu->arch.mcg_ctl = ~(u64)0;
2744 /* Init IA32_MCi_CTL to all 1s */
2745 for (bank = 0; bank < bank_num; bank++)
2746 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2751 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2752 struct kvm_x86_mce *mce)
2754 u64 mcg_cap = vcpu->arch.mcg_cap;
2755 unsigned bank_num = mcg_cap & 0xff;
2756 u64 *banks = vcpu->arch.mce_banks;
2758 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2761 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2762 * reporting is disabled
2764 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2765 vcpu->arch.mcg_ctl != ~(u64)0)
2767 banks += 4 * mce->bank;
2769 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2770 * reporting is disabled for the bank
2772 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2774 if (mce->status & MCI_STATUS_UC) {
2775 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2776 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2777 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2780 if (banks[1] & MCI_STATUS_VAL)
2781 mce->status |= MCI_STATUS_OVER;
2782 banks[2] = mce->addr;
2783 banks[3] = mce->misc;
2784 vcpu->arch.mcg_status = mce->mcg_status;
2785 banks[1] = mce->status;
2786 kvm_queue_exception(vcpu, MC_VECTOR);
2787 } else if (!(banks[1] & MCI_STATUS_VAL)
2788 || !(banks[1] & MCI_STATUS_UC)) {
2789 if (banks[1] & MCI_STATUS_VAL)
2790 mce->status |= MCI_STATUS_OVER;
2791 banks[2] = mce->addr;
2792 banks[3] = mce->misc;
2793 banks[1] = mce->status;
2795 banks[1] |= MCI_STATUS_OVER;
2799 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2800 struct kvm_vcpu_events *events)
2803 events->exception.injected =
2804 vcpu->arch.exception.pending &&
2805 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2806 events->exception.nr = vcpu->arch.exception.nr;
2807 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2808 events->exception.pad = 0;
2809 events->exception.error_code = vcpu->arch.exception.error_code;
2811 events->interrupt.injected =
2812 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2813 events->interrupt.nr = vcpu->arch.interrupt.nr;
2814 events->interrupt.soft = 0;
2815 events->interrupt.shadow =
2816 kvm_x86_ops->get_interrupt_shadow(vcpu,
2817 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2819 events->nmi.injected = vcpu->arch.nmi_injected;
2820 events->nmi.pending = vcpu->arch.nmi_pending != 0;
2821 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2822 events->nmi.pad = 0;
2824 events->sipi_vector = vcpu->arch.sipi_vector;
2826 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2827 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2828 | KVM_VCPUEVENT_VALID_SHADOW);
2829 memset(&events->reserved, 0, sizeof(events->reserved));
2832 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2833 struct kvm_vcpu_events *events)
2835 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2836 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2837 | KVM_VCPUEVENT_VALID_SHADOW))
2841 vcpu->arch.exception.pending = events->exception.injected;
2842 vcpu->arch.exception.nr = events->exception.nr;
2843 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2844 vcpu->arch.exception.error_code = events->exception.error_code;
2846 vcpu->arch.interrupt.pending = events->interrupt.injected;
2847 vcpu->arch.interrupt.nr = events->interrupt.nr;
2848 vcpu->arch.interrupt.soft = events->interrupt.soft;
2849 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2850 kvm_x86_ops->set_interrupt_shadow(vcpu,
2851 events->interrupt.shadow);
2853 vcpu->arch.nmi_injected = events->nmi.injected;
2854 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2855 vcpu->arch.nmi_pending = events->nmi.pending;
2856 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2858 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2859 vcpu->arch.sipi_vector = events->sipi_vector;
2861 kvm_make_request(KVM_REQ_EVENT, vcpu);
2866 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2867 struct kvm_debugregs *dbgregs)
2869 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2870 dbgregs->dr6 = vcpu->arch.dr6;
2871 dbgregs->dr7 = vcpu->arch.dr7;
2873 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2876 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2877 struct kvm_debugregs *dbgregs)
2882 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2883 vcpu->arch.dr6 = dbgregs->dr6;
2884 vcpu->arch.dr7 = dbgregs->dr7;
2889 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2890 struct kvm_xsave *guest_xsave)
2893 memcpy(guest_xsave->region,
2894 &vcpu->arch.guest_fpu.state->xsave,
2897 memcpy(guest_xsave->region,
2898 &vcpu->arch.guest_fpu.state->fxsave,
2899 sizeof(struct i387_fxsave_struct));
2900 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2905 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2906 struct kvm_xsave *guest_xsave)
2909 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2912 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2913 guest_xsave->region, xstate_size);
2915 if (xstate_bv & ~XSTATE_FPSSE)
2917 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2918 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2923 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2924 struct kvm_xcrs *guest_xcrs)
2926 if (!cpu_has_xsave) {
2927 guest_xcrs->nr_xcrs = 0;
2931 guest_xcrs->nr_xcrs = 1;
2932 guest_xcrs->flags = 0;
2933 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2934 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2937 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2938 struct kvm_xcrs *guest_xcrs)
2945 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2948 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2949 /* Only support XCR0 currently */
2950 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2951 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2952 guest_xcrs->xcrs[0].value);
2961 * kvm_set_guest_paused() indicates to the guest kernel that it has been
2962 * stopped by the hypervisor. This function will be called from the host only.
2963 * EINVAL is returned when the host attempts to set the flag for a guest that
2964 * does not support pv clocks.
2966 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
2968 if (!vcpu->arch.time_page)
2970 vcpu->arch.pvclock_set_guest_stopped_request = true;
2971 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2975 long kvm_arch_vcpu_ioctl(struct file *filp,
2976 unsigned int ioctl, unsigned long arg)
2978 struct kvm_vcpu *vcpu = filp->private_data;
2979 void __user *argp = (void __user *)arg;
2982 struct kvm_lapic_state *lapic;
2983 struct kvm_xsave *xsave;
2984 struct kvm_xcrs *xcrs;
2990 case KVM_GET_LAPIC: {
2992 if (!vcpu->arch.apic)
2994 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2999 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3003 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3008 case KVM_SET_LAPIC: {
3010 if (!vcpu->arch.apic)
3012 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3013 if (IS_ERR(u.lapic))
3014 return PTR_ERR(u.lapic);
3016 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3019 case KVM_INTERRUPT: {
3020 struct kvm_interrupt irq;
3023 if (copy_from_user(&irq, argp, sizeof irq))
3025 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3029 r = kvm_vcpu_ioctl_nmi(vcpu);
3032 case KVM_SET_CPUID: {
3033 struct kvm_cpuid __user *cpuid_arg = argp;
3034 struct kvm_cpuid cpuid;
3037 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3039 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3042 case KVM_SET_CPUID2: {
3043 struct kvm_cpuid2 __user *cpuid_arg = argp;
3044 struct kvm_cpuid2 cpuid;
3047 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3049 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3050 cpuid_arg->entries);
3053 case KVM_GET_CPUID2: {
3054 struct kvm_cpuid2 __user *cpuid_arg = argp;
3055 struct kvm_cpuid2 cpuid;
3058 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3060 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3061 cpuid_arg->entries);
3065 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3071 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3074 r = msr_io(vcpu, argp, do_set_msr, 0);
3076 case KVM_TPR_ACCESS_REPORTING: {
3077 struct kvm_tpr_access_ctl tac;
3080 if (copy_from_user(&tac, argp, sizeof tac))
3082 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3086 if (copy_to_user(argp, &tac, sizeof tac))
3091 case KVM_SET_VAPIC_ADDR: {
3092 struct kvm_vapic_addr va;
3095 if (!irqchip_in_kernel(vcpu->kvm))
3098 if (copy_from_user(&va, argp, sizeof va))
3101 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3104 case KVM_X86_SETUP_MCE: {
3108 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3110 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3113 case KVM_X86_SET_MCE: {
3114 struct kvm_x86_mce mce;
3117 if (copy_from_user(&mce, argp, sizeof mce))
3119 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3122 case KVM_GET_VCPU_EVENTS: {
3123 struct kvm_vcpu_events events;
3125 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3128 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3133 case KVM_SET_VCPU_EVENTS: {
3134 struct kvm_vcpu_events events;
3137 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3140 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3143 case KVM_GET_DEBUGREGS: {
3144 struct kvm_debugregs dbgregs;
3146 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3149 if (copy_to_user(argp, &dbgregs,
3150 sizeof(struct kvm_debugregs)))
3155 case KVM_SET_DEBUGREGS: {
3156 struct kvm_debugregs dbgregs;
3159 if (copy_from_user(&dbgregs, argp,
3160 sizeof(struct kvm_debugregs)))
3163 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3166 case KVM_GET_XSAVE: {
3167 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3172 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3175 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3180 case KVM_SET_XSAVE: {
3181 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3182 if (IS_ERR(u.xsave))
3183 return PTR_ERR(u.xsave);
3185 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3188 case KVM_GET_XCRS: {
3189 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3194 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3197 if (copy_to_user(argp, u.xcrs,
3198 sizeof(struct kvm_xcrs)))
3203 case KVM_SET_XCRS: {
3204 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3206 return PTR_ERR(u.xcrs);
3208 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3211 case KVM_SET_TSC_KHZ: {
3215 user_tsc_khz = (u32)arg;
3217 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3220 if (user_tsc_khz == 0)
3221 user_tsc_khz = tsc_khz;
3223 kvm_set_tsc_khz(vcpu, user_tsc_khz);
3228 case KVM_GET_TSC_KHZ: {
3229 r = vcpu->arch.virtual_tsc_khz;
3232 case KVM_KVMCLOCK_CTRL: {
3233 r = kvm_set_guest_paused(vcpu);
3244 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3246 return VM_FAULT_SIGBUS;
3249 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3253 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3255 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3259 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3262 kvm->arch.ept_identity_map_addr = ident_addr;
3266 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3267 u32 kvm_nr_mmu_pages)
3269 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3272 mutex_lock(&kvm->slots_lock);
3274 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3275 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3277 mutex_unlock(&kvm->slots_lock);
3281 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3283 return kvm->arch.n_max_mmu_pages;
3286 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3291 switch (chip->chip_id) {
3292 case KVM_IRQCHIP_PIC_MASTER:
3293 memcpy(&chip->chip.pic,
3294 &pic_irqchip(kvm)->pics[0],
3295 sizeof(struct kvm_pic_state));
3297 case KVM_IRQCHIP_PIC_SLAVE:
3298 memcpy(&chip->chip.pic,
3299 &pic_irqchip(kvm)->pics[1],
3300 sizeof(struct kvm_pic_state));
3302 case KVM_IRQCHIP_IOAPIC:
3303 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3312 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3317 switch (chip->chip_id) {
3318 case KVM_IRQCHIP_PIC_MASTER:
3319 spin_lock(&pic_irqchip(kvm)->lock);
3320 memcpy(&pic_irqchip(kvm)->pics[0],
3322 sizeof(struct kvm_pic_state));
3323 spin_unlock(&pic_irqchip(kvm)->lock);
3325 case KVM_IRQCHIP_PIC_SLAVE:
3326 spin_lock(&pic_irqchip(kvm)->lock);
3327 memcpy(&pic_irqchip(kvm)->pics[1],
3329 sizeof(struct kvm_pic_state));
3330 spin_unlock(&pic_irqchip(kvm)->lock);
3332 case KVM_IRQCHIP_IOAPIC:
3333 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3339 kvm_pic_update_irq(pic_irqchip(kvm));
3343 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3347 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3348 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3349 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3353 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3357 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3358 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3359 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3360 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3364 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3368 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3369 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3370 sizeof(ps->channels));
3371 ps->flags = kvm->arch.vpit->pit_state.flags;
3372 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3373 memset(&ps->reserved, 0, sizeof(ps->reserved));
3377 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3379 int r = 0, start = 0;
3380 u32 prev_legacy, cur_legacy;
3381 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3382 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3383 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3384 if (!prev_legacy && cur_legacy)
3386 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3387 sizeof(kvm->arch.vpit->pit_state.channels));
3388 kvm->arch.vpit->pit_state.flags = ps->flags;
3389 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3390 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3394 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3395 struct kvm_reinject_control *control)
3397 if (!kvm->arch.vpit)
3399 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3400 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3401 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3406 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3407 * @kvm: kvm instance
3408 * @log: slot id and address to which we copy the log
3410 * We need to keep it in mind that VCPU threads can write to the bitmap
3411 * concurrently. So, to avoid losing data, we keep the following order for
3414 * 1. Take a snapshot of the bit and clear it if needed.
3415 * 2. Write protect the corresponding page.
3416 * 3. Flush TLB's if needed.
3417 * 4. Copy the snapshot to the userspace.
3419 * Between 2 and 3, the guest may write to the page using the remaining TLB
3420 * entry. This is not a problem because the page will be reported dirty at
3421 * step 4 using the snapshot taken before and step 3 ensures that successive
3422 * writes will be logged for the next call.
3424 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3427 struct kvm_memory_slot *memslot;
3429 unsigned long *dirty_bitmap;
3430 unsigned long *dirty_bitmap_buffer;
3431 bool is_dirty = false;
3433 mutex_lock(&kvm->slots_lock);
3436 if (log->slot >= KVM_USER_MEM_SLOTS)
3439 memslot = id_to_memslot(kvm->memslots, log->slot);
3441 dirty_bitmap = memslot->dirty_bitmap;
3446 n = kvm_dirty_bitmap_bytes(memslot);
3448 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
3449 memset(dirty_bitmap_buffer, 0, n);
3451 spin_lock(&kvm->mmu_lock);
3453 for (i = 0; i < n / sizeof(long); i++) {
3457 if (!dirty_bitmap[i])
3462 mask = xchg(&dirty_bitmap[i], 0);
3463 dirty_bitmap_buffer[i] = mask;
3465 offset = i * BITS_PER_LONG;
3466 kvm_mmu_write_protect_pt_masked(kvm, memslot, offset, mask);
3469 kvm_flush_remote_tlbs(kvm);
3471 spin_unlock(&kvm->mmu_lock);
3474 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
3479 mutex_unlock(&kvm->slots_lock);
3483 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event)
3485 if (!irqchip_in_kernel(kvm))
3488 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3489 irq_event->irq, irq_event->level);
3493 long kvm_arch_vm_ioctl(struct file *filp,
3494 unsigned int ioctl, unsigned long arg)
3496 struct kvm *kvm = filp->private_data;
3497 void __user *argp = (void __user *)arg;
3500 * This union makes it completely explicit to gcc-3.x
3501 * that these two variables' stack usage should be
3502 * combined, not added together.
3505 struct kvm_pit_state ps;
3506 struct kvm_pit_state2 ps2;
3507 struct kvm_pit_config pit_config;
3511 case KVM_SET_TSS_ADDR:
3512 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3514 case KVM_SET_IDENTITY_MAP_ADDR: {
3518 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3520 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3523 case KVM_SET_NR_MMU_PAGES:
3524 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3526 case KVM_GET_NR_MMU_PAGES:
3527 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3529 case KVM_CREATE_IRQCHIP: {
3530 struct kvm_pic *vpic;
3532 mutex_lock(&kvm->lock);
3535 goto create_irqchip_unlock;
3537 if (atomic_read(&kvm->online_vcpus))
3538 goto create_irqchip_unlock;
3540 vpic = kvm_create_pic(kvm);
3542 r = kvm_ioapic_init(kvm);
3544 mutex_lock(&kvm->slots_lock);
3545 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3547 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3549 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3551 mutex_unlock(&kvm->slots_lock);
3553 goto create_irqchip_unlock;
3556 goto create_irqchip_unlock;
3558 kvm->arch.vpic = vpic;
3560 r = kvm_setup_default_irq_routing(kvm);
3562 mutex_lock(&kvm->slots_lock);
3563 mutex_lock(&kvm->irq_lock);
3564 kvm_ioapic_destroy(kvm);
3565 kvm_destroy_pic(kvm);
3566 mutex_unlock(&kvm->irq_lock);
3567 mutex_unlock(&kvm->slots_lock);
3569 create_irqchip_unlock:
3570 mutex_unlock(&kvm->lock);
3573 case KVM_CREATE_PIT:
3574 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3576 case KVM_CREATE_PIT2:
3578 if (copy_from_user(&u.pit_config, argp,
3579 sizeof(struct kvm_pit_config)))
3582 mutex_lock(&kvm->slots_lock);
3585 goto create_pit_unlock;
3587 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3591 mutex_unlock(&kvm->slots_lock);
3593 case KVM_GET_IRQCHIP: {
3594 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3595 struct kvm_irqchip *chip;
3597 chip = memdup_user(argp, sizeof(*chip));
3604 if (!irqchip_in_kernel(kvm))
3605 goto get_irqchip_out;
3606 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3608 goto get_irqchip_out;
3610 if (copy_to_user(argp, chip, sizeof *chip))
3611 goto get_irqchip_out;
3617 case KVM_SET_IRQCHIP: {
3618 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3619 struct kvm_irqchip *chip;
3621 chip = memdup_user(argp, sizeof(*chip));
3628 if (!irqchip_in_kernel(kvm))
3629 goto set_irqchip_out;
3630 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3632 goto set_irqchip_out;
3640 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3643 if (!kvm->arch.vpit)
3645 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3649 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3656 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3659 if (!kvm->arch.vpit)
3661 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3664 case KVM_GET_PIT2: {
3666 if (!kvm->arch.vpit)
3668 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3672 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3677 case KVM_SET_PIT2: {
3679 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3682 if (!kvm->arch.vpit)
3684 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3687 case KVM_REINJECT_CONTROL: {
3688 struct kvm_reinject_control control;
3690 if (copy_from_user(&control, argp, sizeof(control)))
3692 r = kvm_vm_ioctl_reinject(kvm, &control);
3695 case KVM_XEN_HVM_CONFIG: {
3697 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3698 sizeof(struct kvm_xen_hvm_config)))
3701 if (kvm->arch.xen_hvm_config.flags)
3706 case KVM_SET_CLOCK: {
3707 struct kvm_clock_data user_ns;
3712 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3720 local_irq_disable();
3721 now_ns = get_kernel_ns();
3722 delta = user_ns.clock - now_ns;
3724 kvm->arch.kvmclock_offset = delta;
3727 case KVM_GET_CLOCK: {
3728 struct kvm_clock_data user_ns;
3731 local_irq_disable();
3732 now_ns = get_kernel_ns();
3733 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3736 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3739 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3752 static void kvm_init_msr_list(void)
3757 /* skip the first msrs in the list. KVM-specific */
3758 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3759 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3762 msrs_to_save[j] = msrs_to_save[i];
3765 num_msrs_to_save = j;
3768 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3776 if (!(vcpu->arch.apic &&
3777 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3778 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3789 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3796 if (!(vcpu->arch.apic &&
3797 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3798 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3800 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3810 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3811 struct kvm_segment *var, int seg)
3813 kvm_x86_ops->set_segment(vcpu, var, seg);
3816 void kvm_get_segment(struct kvm_vcpu *vcpu,
3817 struct kvm_segment *var, int seg)
3819 kvm_x86_ops->get_segment(vcpu, var, seg);
3822 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3825 struct x86_exception exception;
3827 BUG_ON(!mmu_is_nested(vcpu));
3829 /* NPT walks are always user-walks */
3830 access |= PFERR_USER_MASK;
3831 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3836 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3837 struct x86_exception *exception)
3839 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3840 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3843 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3844 struct x86_exception *exception)
3846 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3847 access |= PFERR_FETCH_MASK;
3848 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3851 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3852 struct x86_exception *exception)
3854 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3855 access |= PFERR_WRITE_MASK;
3856 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3859 /* uses this to access any guest's mapped memory without checking CPL */
3860 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3861 struct x86_exception *exception)
3863 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3866 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3867 struct kvm_vcpu *vcpu, u32 access,
3868 struct x86_exception *exception)
3871 int r = X86EMUL_CONTINUE;
3874 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3876 unsigned offset = addr & (PAGE_SIZE-1);
3877 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3880 if (gpa == UNMAPPED_GVA)
3881 return X86EMUL_PROPAGATE_FAULT;
3882 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3884 r = X86EMUL_IO_NEEDED;
3896 /* used for instruction fetching */
3897 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3898 gva_t addr, void *val, unsigned int bytes,
3899 struct x86_exception *exception)
3901 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3902 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3904 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3905 access | PFERR_FETCH_MASK,
3909 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3910 gva_t addr, void *val, unsigned int bytes,
3911 struct x86_exception *exception)
3913 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3914 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3916 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3919 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
3921 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3922 gva_t addr, void *val, unsigned int bytes,
3923 struct x86_exception *exception)
3925 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3926 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3929 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3930 gva_t addr, void *val,
3932 struct x86_exception *exception)
3934 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3936 int r = X86EMUL_CONTINUE;
3939 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3942 unsigned offset = addr & (PAGE_SIZE-1);
3943 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3946 if (gpa == UNMAPPED_GVA)
3947 return X86EMUL_PROPAGATE_FAULT;
3948 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3950 r = X86EMUL_IO_NEEDED;
3961 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
3963 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
3964 gpa_t *gpa, struct x86_exception *exception,
3967 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
3968 | (write ? PFERR_WRITE_MASK : 0);
3970 if (vcpu_match_mmio_gva(vcpu, gva)
3971 && !permission_fault(vcpu->arch.walk_mmu, vcpu->arch.access, access)) {
3972 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
3973 (gva & (PAGE_SIZE - 1));
3974 trace_vcpu_match_mmio(gva, *gpa, write, false);
3978 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3980 if (*gpa == UNMAPPED_GVA)
3983 /* For APIC access vmexit */
3984 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3987 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
3988 trace_vcpu_match_mmio(gva, *gpa, write, true);
3995 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3996 const void *val, int bytes)
4000 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4003 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4007 struct read_write_emulator_ops {
4008 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4010 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4011 void *val, int bytes);
4012 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4013 int bytes, void *val);
4014 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4015 void *val, int bytes);
4019 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4021 if (vcpu->mmio_read_completed) {
4022 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4023 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4024 vcpu->mmio_read_completed = 0;
4031 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4032 void *val, int bytes)
4034 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4037 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4038 void *val, int bytes)
4040 return emulator_write_phys(vcpu, gpa, val, bytes);
4043 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4045 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4046 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4049 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4050 void *val, int bytes)
4052 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4053 return X86EMUL_IO_NEEDED;
4056 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4057 void *val, int bytes)
4059 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4061 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4062 return X86EMUL_CONTINUE;
4065 static const struct read_write_emulator_ops read_emultor = {
4066 .read_write_prepare = read_prepare,
4067 .read_write_emulate = read_emulate,
4068 .read_write_mmio = vcpu_mmio_read,
4069 .read_write_exit_mmio = read_exit_mmio,
4072 static const struct read_write_emulator_ops write_emultor = {
4073 .read_write_emulate = write_emulate,
4074 .read_write_mmio = write_mmio,
4075 .read_write_exit_mmio = write_exit_mmio,
4079 static int emulator_read_write_onepage(unsigned long addr, void *val,
4081 struct x86_exception *exception,
4082 struct kvm_vcpu *vcpu,
4083 const struct read_write_emulator_ops *ops)
4087 bool write = ops->write;
4088 struct kvm_mmio_fragment *frag;
4090 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4093 return X86EMUL_PROPAGATE_FAULT;
4095 /* For APIC access vmexit */
4099 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4100 return X86EMUL_CONTINUE;
4104 * Is this MMIO handled locally?
4106 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4107 if (handled == bytes)
4108 return X86EMUL_CONTINUE;
4114 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4115 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4119 return X86EMUL_CONTINUE;
4122 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4123 void *val, unsigned int bytes,
4124 struct x86_exception *exception,
4125 const struct read_write_emulator_ops *ops)
4127 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4131 if (ops->read_write_prepare &&
4132 ops->read_write_prepare(vcpu, val, bytes))
4133 return X86EMUL_CONTINUE;
4135 vcpu->mmio_nr_fragments = 0;
4137 /* Crossing a page boundary? */
4138 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4141 now = -addr & ~PAGE_MASK;
4142 rc = emulator_read_write_onepage(addr, val, now, exception,
4145 if (rc != X86EMUL_CONTINUE)
4152 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4154 if (rc != X86EMUL_CONTINUE)
4157 if (!vcpu->mmio_nr_fragments)
4160 gpa = vcpu->mmio_fragments[0].gpa;
4162 vcpu->mmio_needed = 1;
4163 vcpu->mmio_cur_fragment = 0;
4165 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4166 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4167 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4168 vcpu->run->mmio.phys_addr = gpa;
4170 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4173 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4177 struct x86_exception *exception)
4179 return emulator_read_write(ctxt, addr, val, bytes,
4180 exception, &read_emultor);
4183 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4187 struct x86_exception *exception)
4189 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4190 exception, &write_emultor);
4193 #define CMPXCHG_TYPE(t, ptr, old, new) \
4194 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4196 #ifdef CONFIG_X86_64
4197 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4199 # define CMPXCHG64(ptr, old, new) \
4200 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4203 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4208 struct x86_exception *exception)
4210 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4216 /* guests cmpxchg8b have to be emulated atomically */
4217 if (bytes > 8 || (bytes & (bytes - 1)))
4220 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4222 if (gpa == UNMAPPED_GVA ||
4223 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4226 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4229 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4230 if (is_error_page(page))
4233 kaddr = kmap_atomic(page);
4234 kaddr += offset_in_page(gpa);
4237 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4240 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4243 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4246 exchanged = CMPXCHG64(kaddr, old, new);
4251 kunmap_atomic(kaddr);
4252 kvm_release_page_dirty(page);
4255 return X86EMUL_CMPXCHG_FAILED;
4257 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4259 return X86EMUL_CONTINUE;
4262 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4264 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4267 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4269 /* TODO: String I/O for in kernel device */
4272 if (vcpu->arch.pio.in)
4273 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4274 vcpu->arch.pio.size, pd);
4276 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4277 vcpu->arch.pio.port, vcpu->arch.pio.size,
4282 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4283 unsigned short port, void *val,
4284 unsigned int count, bool in)
4286 trace_kvm_pio(!in, port, size, count);
4288 vcpu->arch.pio.port = port;
4289 vcpu->arch.pio.in = in;
4290 vcpu->arch.pio.count = count;
4291 vcpu->arch.pio.size = size;
4293 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4294 vcpu->arch.pio.count = 0;
4298 vcpu->run->exit_reason = KVM_EXIT_IO;
4299 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4300 vcpu->run->io.size = size;
4301 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4302 vcpu->run->io.count = count;
4303 vcpu->run->io.port = port;
4308 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4309 int size, unsigned short port, void *val,
4312 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4315 if (vcpu->arch.pio.count)
4318 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4321 memcpy(val, vcpu->arch.pio_data, size * count);
4322 vcpu->arch.pio.count = 0;
4329 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4330 int size, unsigned short port,
4331 const void *val, unsigned int count)
4333 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4335 memcpy(vcpu->arch.pio_data, val, size * count);
4336 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4339 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4341 return kvm_x86_ops->get_segment_base(vcpu, seg);
4344 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4346 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4349 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4351 if (!need_emulate_wbinvd(vcpu))
4352 return X86EMUL_CONTINUE;
4354 if (kvm_x86_ops->has_wbinvd_exit()) {
4355 int cpu = get_cpu();
4357 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4358 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4359 wbinvd_ipi, NULL, 1);
4361 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4364 return X86EMUL_CONTINUE;
4366 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4368 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4370 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4373 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4375 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4378 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4381 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4384 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4386 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4389 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4391 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4392 unsigned long value;
4396 value = kvm_read_cr0(vcpu);
4399 value = vcpu->arch.cr2;
4402 value = kvm_read_cr3(vcpu);
4405 value = kvm_read_cr4(vcpu);
4408 value = kvm_get_cr8(vcpu);
4411 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4418 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4420 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4425 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4428 vcpu->arch.cr2 = val;
4431 res = kvm_set_cr3(vcpu, val);
4434 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4437 res = kvm_set_cr8(vcpu, val);
4440 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4447 static void emulator_set_rflags(struct x86_emulate_ctxt *ctxt, ulong val)
4449 kvm_set_rflags(emul_to_vcpu(ctxt), val);
4452 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4454 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4457 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4459 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4462 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4464 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4467 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4469 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4472 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4474 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4477 static unsigned long emulator_get_cached_segment_base(
4478 struct x86_emulate_ctxt *ctxt, int seg)
4480 return get_segment_base(emul_to_vcpu(ctxt), seg);
4483 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4484 struct desc_struct *desc, u32 *base3,
4487 struct kvm_segment var;
4489 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4490 *selector = var.selector;
4497 set_desc_limit(desc, var.limit);
4498 set_desc_base(desc, (unsigned long)var.base);
4499 #ifdef CONFIG_X86_64
4501 *base3 = var.base >> 32;
4503 desc->type = var.type;
4505 desc->dpl = var.dpl;
4506 desc->p = var.present;
4507 desc->avl = var.avl;
4515 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4516 struct desc_struct *desc, u32 base3,
4519 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4520 struct kvm_segment var;
4522 var.selector = selector;
4523 var.base = get_desc_base(desc);
4524 #ifdef CONFIG_X86_64
4525 var.base |= ((u64)base3) << 32;
4527 var.limit = get_desc_limit(desc);
4529 var.limit = (var.limit << 12) | 0xfff;
4530 var.type = desc->type;
4531 var.present = desc->p;
4532 var.dpl = desc->dpl;
4537 var.avl = desc->avl;
4538 var.present = desc->p;
4539 var.unusable = !var.present;
4542 kvm_set_segment(vcpu, &var, seg);
4546 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4547 u32 msr_index, u64 *pdata)
4549 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4552 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4553 u32 msr_index, u64 data)
4555 struct msr_data msr;
4558 msr.index = msr_index;
4559 msr.host_initiated = false;
4560 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4563 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4564 u32 pmc, u64 *pdata)
4566 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4569 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4571 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4574 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4577 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4579 * CR0.TS may reference the host fpu state, not the guest fpu state,
4580 * so it may be clear at this point.
4585 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4590 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4591 struct x86_instruction_info *info,
4592 enum x86_intercept_stage stage)
4594 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4597 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4598 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4600 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4603 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4605 return kvm_register_read(emul_to_vcpu(ctxt), reg);
4608 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4610 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4613 static const struct x86_emulate_ops emulate_ops = {
4614 .read_gpr = emulator_read_gpr,
4615 .write_gpr = emulator_write_gpr,
4616 .read_std = kvm_read_guest_virt_system,
4617 .write_std = kvm_write_guest_virt_system,
4618 .fetch = kvm_fetch_guest_virt,
4619 .read_emulated = emulator_read_emulated,
4620 .write_emulated = emulator_write_emulated,
4621 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4622 .invlpg = emulator_invlpg,
4623 .pio_in_emulated = emulator_pio_in_emulated,
4624 .pio_out_emulated = emulator_pio_out_emulated,
4625 .get_segment = emulator_get_segment,
4626 .set_segment = emulator_set_segment,
4627 .get_cached_segment_base = emulator_get_cached_segment_base,
4628 .get_gdt = emulator_get_gdt,
4629 .get_idt = emulator_get_idt,
4630 .set_gdt = emulator_set_gdt,
4631 .set_idt = emulator_set_idt,
4632 .get_cr = emulator_get_cr,
4633 .set_cr = emulator_set_cr,
4634 .set_rflags = emulator_set_rflags,
4635 .cpl = emulator_get_cpl,
4636 .get_dr = emulator_get_dr,
4637 .set_dr = emulator_set_dr,
4638 .set_msr = emulator_set_msr,
4639 .get_msr = emulator_get_msr,
4640 .read_pmc = emulator_read_pmc,
4641 .halt = emulator_halt,
4642 .wbinvd = emulator_wbinvd,
4643 .fix_hypercall = emulator_fix_hypercall,
4644 .get_fpu = emulator_get_fpu,
4645 .put_fpu = emulator_put_fpu,
4646 .intercept = emulator_intercept,
4647 .get_cpuid = emulator_get_cpuid,
4650 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4652 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4654 * an sti; sti; sequence only disable interrupts for the first
4655 * instruction. So, if the last instruction, be it emulated or
4656 * not, left the system with the INT_STI flag enabled, it
4657 * means that the last instruction is an sti. We should not
4658 * leave the flag on in this case. The same goes for mov ss
4660 if (!(int_shadow & mask))
4661 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4664 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4666 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4667 if (ctxt->exception.vector == PF_VECTOR)
4668 kvm_propagate_fault(vcpu, &ctxt->exception);
4669 else if (ctxt->exception.error_code_valid)
4670 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4671 ctxt->exception.error_code);
4673 kvm_queue_exception(vcpu, ctxt->exception.vector);
4676 static void init_decode_cache(struct x86_emulate_ctxt *ctxt)
4678 memset(&ctxt->twobyte, 0,
4679 (void *)&ctxt->_regs - (void *)&ctxt->twobyte);
4681 ctxt->fetch.start = 0;
4682 ctxt->fetch.end = 0;
4683 ctxt->io_read.pos = 0;
4684 ctxt->io_read.end = 0;
4685 ctxt->mem_read.pos = 0;
4686 ctxt->mem_read.end = 0;
4689 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4691 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4694 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4696 ctxt->eflags = kvm_get_rflags(vcpu);
4697 ctxt->eip = kvm_rip_read(vcpu);
4698 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4699 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4700 cs_l ? X86EMUL_MODE_PROT64 :
4701 cs_db ? X86EMUL_MODE_PROT32 :
4702 X86EMUL_MODE_PROT16;
4703 ctxt->guest_mode = is_guest_mode(vcpu);
4705 init_decode_cache(ctxt);
4706 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4709 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4711 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4714 init_emulate_ctxt(vcpu);
4718 ctxt->_eip = ctxt->eip + inc_eip;
4719 ret = emulate_int_real(ctxt, irq);
4721 if (ret != X86EMUL_CONTINUE)
4722 return EMULATE_FAIL;
4724 ctxt->eip = ctxt->_eip;
4725 kvm_rip_write(vcpu, ctxt->eip);
4726 kvm_set_rflags(vcpu, ctxt->eflags);
4728 if (irq == NMI_VECTOR)
4729 vcpu->arch.nmi_pending = 0;
4731 vcpu->arch.interrupt.pending = false;
4733 return EMULATE_DONE;
4735 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4737 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4739 int r = EMULATE_DONE;
4741 ++vcpu->stat.insn_emulation_fail;
4742 trace_kvm_emulate_insn_failed(vcpu);
4743 if (!is_guest_mode(vcpu)) {
4744 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4745 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4746 vcpu->run->internal.ndata = 0;
4749 kvm_queue_exception(vcpu, UD_VECTOR);
4754 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4763 * if emulation was due to access to shadowed page table
4764 * and it failed try to unshadow page and re-enter the
4765 * guest to let CPU execute the instruction.
4767 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4770 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4772 if (gpa == UNMAPPED_GVA)
4773 return true; /* let cpu generate fault */
4776 * Do not retry the unhandleable instruction if it faults on the
4777 * readonly host memory, otherwise it will goto a infinite loop:
4778 * retry instruction -> write #PF -> emulation fail -> retry
4779 * instruction -> ...
4781 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
4782 if (!is_error_noslot_pfn(pfn)) {
4783 kvm_release_pfn_clean(pfn);
4790 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
4791 unsigned long cr2, int emulation_type)
4793 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4794 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
4796 last_retry_eip = vcpu->arch.last_retry_eip;
4797 last_retry_addr = vcpu->arch.last_retry_addr;
4800 * If the emulation is caused by #PF and it is non-page_table
4801 * writing instruction, it means the VM-EXIT is caused by shadow
4802 * page protected, we can zap the shadow page and retry this
4803 * instruction directly.
4805 * Note: if the guest uses a non-page-table modifying instruction
4806 * on the PDE that points to the instruction, then we will unmap
4807 * the instruction and go to an infinite loop. So, we cache the
4808 * last retried eip and the last fault address, if we meet the eip
4809 * and the address again, we can break out of the potential infinite
4812 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
4814 if (!(emulation_type & EMULTYPE_RETRY))
4817 if (x86_page_table_writing_insn(ctxt))
4820 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
4823 vcpu->arch.last_retry_eip = ctxt->eip;
4824 vcpu->arch.last_retry_addr = cr2;
4826 if (!vcpu->arch.mmu.direct_map)
4827 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4829 kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4834 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
4835 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
4837 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4844 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4845 bool writeback = true;
4847 kvm_clear_exception_queue(vcpu);
4849 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4850 init_emulate_ctxt(vcpu);
4851 ctxt->interruptibility = 0;
4852 ctxt->have_exception = false;
4853 ctxt->perm_ok = false;
4855 ctxt->only_vendor_specific_insn
4856 = emulation_type & EMULTYPE_TRAP_UD;
4858 r = x86_decode_insn(ctxt, insn, insn_len);
4860 trace_kvm_emulate_insn_start(vcpu);
4861 ++vcpu->stat.insn_emulation;
4862 if (r != EMULATION_OK) {
4863 if (emulation_type & EMULTYPE_TRAP_UD)
4864 return EMULATE_FAIL;
4865 if (reexecute_instruction(vcpu, cr2))
4866 return EMULATE_DONE;
4867 if (emulation_type & EMULTYPE_SKIP)
4868 return EMULATE_FAIL;
4869 return handle_emulation_failure(vcpu);
4873 if (emulation_type & EMULTYPE_SKIP) {
4874 kvm_rip_write(vcpu, ctxt->_eip);
4875 return EMULATE_DONE;
4878 if (retry_instruction(ctxt, cr2, emulation_type))
4879 return EMULATE_DONE;
4881 /* this is needed for vmware backdoor interface to work since it
4882 changes registers values during IO operation */
4883 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4884 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4885 emulator_invalidate_register_cache(ctxt);
4889 r = x86_emulate_insn(ctxt);
4891 if (r == EMULATION_INTERCEPTED)
4892 return EMULATE_DONE;
4894 if (r == EMULATION_FAILED) {
4895 if (reexecute_instruction(vcpu, cr2))
4896 return EMULATE_DONE;
4898 return handle_emulation_failure(vcpu);
4901 if (ctxt->have_exception) {
4902 inject_emulated_exception(vcpu);
4904 } else if (vcpu->arch.pio.count) {
4905 if (!vcpu->arch.pio.in)
4906 vcpu->arch.pio.count = 0;
4909 vcpu->arch.complete_userspace_io = complete_emulated_pio;
4911 r = EMULATE_DO_MMIO;
4912 } else if (vcpu->mmio_needed) {
4913 if (!vcpu->mmio_is_write)
4915 r = EMULATE_DO_MMIO;
4916 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
4917 } else if (r == EMULATION_RESTART)
4923 toggle_interruptibility(vcpu, ctxt->interruptibility);
4924 kvm_set_rflags(vcpu, ctxt->eflags);
4925 kvm_make_request(KVM_REQ_EVENT, vcpu);
4926 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4927 kvm_rip_write(vcpu, ctxt->eip);
4929 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
4933 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4935 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4937 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4938 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4939 size, port, &val, 1);
4940 /* do not return to emulator after return from userspace */
4941 vcpu->arch.pio.count = 0;
4944 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4946 static void tsc_bad(void *info)
4948 __this_cpu_write(cpu_tsc_khz, 0);
4951 static void tsc_khz_changed(void *data)
4953 struct cpufreq_freqs *freq = data;
4954 unsigned long khz = 0;
4958 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4959 khz = cpufreq_quick_get(raw_smp_processor_id());
4962 __this_cpu_write(cpu_tsc_khz, khz);
4965 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4968 struct cpufreq_freqs *freq = data;
4970 struct kvm_vcpu *vcpu;
4971 int i, send_ipi = 0;
4974 * We allow guests to temporarily run on slowing clocks,
4975 * provided we notify them after, or to run on accelerating
4976 * clocks, provided we notify them before. Thus time never
4979 * However, we have a problem. We can't atomically update
4980 * the frequency of a given CPU from this function; it is
4981 * merely a notifier, which can be called from any CPU.
4982 * Changing the TSC frequency at arbitrary points in time
4983 * requires a recomputation of local variables related to
4984 * the TSC for each VCPU. We must flag these local variables
4985 * to be updated and be sure the update takes place with the
4986 * new frequency before any guests proceed.
4988 * Unfortunately, the combination of hotplug CPU and frequency
4989 * change creates an intractable locking scenario; the order
4990 * of when these callouts happen is undefined with respect to
4991 * CPU hotplug, and they can race with each other. As such,
4992 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4993 * undefined; you can actually have a CPU frequency change take
4994 * place in between the computation of X and the setting of the
4995 * variable. To protect against this problem, all updates of
4996 * the per_cpu tsc_khz variable are done in an interrupt
4997 * protected IPI, and all callers wishing to update the value
4998 * must wait for a synchronous IPI to complete (which is trivial
4999 * if the caller is on the CPU already). This establishes the
5000 * necessary total order on variable updates.
5002 * Note that because a guest time update may take place
5003 * anytime after the setting of the VCPU's request bit, the
5004 * correct TSC value must be set before the request. However,
5005 * to ensure the update actually makes it to any guest which
5006 * starts running in hardware virtualization between the set
5007 * and the acquisition of the spinlock, we must also ping the
5008 * CPU after setting the request bit.
5012 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5014 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5017 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5019 raw_spin_lock(&kvm_lock);
5020 list_for_each_entry(kvm, &vm_list, vm_list) {
5021 kvm_for_each_vcpu(i, vcpu, kvm) {
5022 if (vcpu->cpu != freq->cpu)
5024 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5025 if (vcpu->cpu != smp_processor_id())
5029 raw_spin_unlock(&kvm_lock);
5031 if (freq->old < freq->new && send_ipi) {
5033 * We upscale the frequency. Must make the guest
5034 * doesn't see old kvmclock values while running with
5035 * the new frequency, otherwise we risk the guest sees
5036 * time go backwards.
5038 * In case we update the frequency for another cpu
5039 * (which might be in guest context) send an interrupt
5040 * to kick the cpu out of guest context. Next time
5041 * guest context is entered kvmclock will be updated,
5042 * so the guest will not see stale values.
5044 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5049 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5050 .notifier_call = kvmclock_cpufreq_notifier
5053 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5054 unsigned long action, void *hcpu)
5056 unsigned int cpu = (unsigned long)hcpu;
5060 case CPU_DOWN_FAILED:
5061 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5063 case CPU_DOWN_PREPARE:
5064 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5070 static struct notifier_block kvmclock_cpu_notifier_block = {
5071 .notifier_call = kvmclock_cpu_notifier,
5072 .priority = -INT_MAX
5075 static void kvm_timer_init(void)
5079 max_tsc_khz = tsc_khz;
5080 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5081 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5082 #ifdef CONFIG_CPU_FREQ
5083 struct cpufreq_policy policy;
5084 memset(&policy, 0, sizeof(policy));
5086 cpufreq_get_policy(&policy, cpu);
5087 if (policy.cpuinfo.max_freq)
5088 max_tsc_khz = policy.cpuinfo.max_freq;
5091 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5092 CPUFREQ_TRANSITION_NOTIFIER);
5094 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5095 for_each_online_cpu(cpu)
5096 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5099 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5101 int kvm_is_in_guest(void)
5103 return __this_cpu_read(current_vcpu) != NULL;
5106 static int kvm_is_user_mode(void)
5110 if (__this_cpu_read(current_vcpu))
5111 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5113 return user_mode != 0;
5116 static unsigned long kvm_get_guest_ip(void)
5118 unsigned long ip = 0;
5120 if (__this_cpu_read(current_vcpu))
5121 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5126 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5127 .is_in_guest = kvm_is_in_guest,
5128 .is_user_mode = kvm_is_user_mode,
5129 .get_guest_ip = kvm_get_guest_ip,
5132 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5134 __this_cpu_write(current_vcpu, vcpu);
5136 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5138 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5140 __this_cpu_write(current_vcpu, NULL);
5142 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5144 static void kvm_set_mmio_spte_mask(void)
5147 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5150 * Set the reserved bits and the present bit of an paging-structure
5151 * entry to generate page fault with PFER.RSV = 1.
5153 mask = ((1ull << (62 - maxphyaddr + 1)) - 1) << maxphyaddr;
5156 #ifdef CONFIG_X86_64
5158 * If reserved bit is not supported, clear the present bit to disable
5161 if (maxphyaddr == 52)
5165 kvm_mmu_set_mmio_spte_mask(mask);
5168 #ifdef CONFIG_X86_64
5169 static void pvclock_gtod_update_fn(struct work_struct *work)
5173 struct kvm_vcpu *vcpu;
5176 raw_spin_lock(&kvm_lock);
5177 list_for_each_entry(kvm, &vm_list, vm_list)
5178 kvm_for_each_vcpu(i, vcpu, kvm)
5179 set_bit(KVM_REQ_MASTERCLOCK_UPDATE, &vcpu->requests);
5180 atomic_set(&kvm_guest_has_master_clock, 0);
5181 raw_spin_unlock(&kvm_lock);
5184 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5187 * Notification about pvclock gtod data update.
5189 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5192 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5193 struct timekeeper *tk = priv;
5195 update_pvclock_gtod(tk);
5197 /* disable master clock if host does not trust, or does not
5198 * use, TSC clocksource
5200 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5201 atomic_read(&kvm_guest_has_master_clock) != 0)
5202 queue_work(system_long_wq, &pvclock_gtod_work);
5207 static struct notifier_block pvclock_gtod_notifier = {
5208 .notifier_call = pvclock_gtod_notify,
5212 int kvm_arch_init(void *opaque)
5215 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
5218 printk(KERN_ERR "kvm: already loaded the other module\n");
5223 if (!ops->cpu_has_kvm_support()) {
5224 printk(KERN_ERR "kvm: no hardware support\n");
5228 if (ops->disabled_by_bios()) {
5229 printk(KERN_ERR "kvm: disabled by bios\n");
5234 r = kvm_mmu_module_init();
5238 kvm_set_mmio_spte_mask();
5239 kvm_init_msr_list();
5242 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5243 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5247 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5250 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5253 #ifdef CONFIG_X86_64
5254 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5263 void kvm_arch_exit(void)
5265 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5267 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5268 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5269 CPUFREQ_TRANSITION_NOTIFIER);
5270 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5271 #ifdef CONFIG_X86_64
5272 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5275 kvm_mmu_module_exit();
5278 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5280 ++vcpu->stat.halt_exits;
5281 if (irqchip_in_kernel(vcpu->kvm)) {
5282 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5285 vcpu->run->exit_reason = KVM_EXIT_HLT;
5289 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5291 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5293 u64 param, ingpa, outgpa, ret;
5294 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5295 bool fast, longmode;
5299 * hypercall generates UD from non zero cpl and real mode
5302 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5303 kvm_queue_exception(vcpu, UD_VECTOR);
5307 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5308 longmode = is_long_mode(vcpu) && cs_l == 1;
5311 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5312 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5313 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5314 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5315 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5316 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5318 #ifdef CONFIG_X86_64
5320 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5321 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5322 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5326 code = param & 0xffff;
5327 fast = (param >> 16) & 0x1;
5328 rep_cnt = (param >> 32) & 0xfff;
5329 rep_idx = (param >> 48) & 0xfff;
5331 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5334 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5335 kvm_vcpu_on_spin(vcpu);
5338 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5342 ret = res | (((u64)rep_done & 0xfff) << 32);
5344 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5346 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5347 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5353 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5355 unsigned long nr, a0, a1, a2, a3, ret;
5358 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5359 return kvm_hv_hypercall(vcpu);
5361 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5362 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5363 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5364 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5365 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5367 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5369 if (!is_long_mode(vcpu)) {
5377 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5383 case KVM_HC_VAPIC_POLL_IRQ:
5391 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5392 ++vcpu->stat.hypercalls;
5395 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5397 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5399 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5400 char instruction[3];
5401 unsigned long rip = kvm_rip_read(vcpu);
5404 * Blow out the MMU to ensure that no other VCPU has an active mapping
5405 * to ensure that the updated hypercall appears atomically across all
5408 kvm_mmu_zap_all(vcpu->kvm);
5410 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5412 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5416 * Check if userspace requested an interrupt window, and that the
5417 * interrupt window is open.
5419 * No need to exit to userspace if we already have an interrupt queued.
5421 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5423 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5424 vcpu->run->request_interrupt_window &&
5425 kvm_arch_interrupt_allowed(vcpu));
5428 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5430 struct kvm_run *kvm_run = vcpu->run;
5432 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5433 kvm_run->cr8 = kvm_get_cr8(vcpu);
5434 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5435 if (irqchip_in_kernel(vcpu->kvm))
5436 kvm_run->ready_for_interrupt_injection = 1;
5438 kvm_run->ready_for_interrupt_injection =
5439 kvm_arch_interrupt_allowed(vcpu) &&
5440 !kvm_cpu_has_interrupt(vcpu) &&
5441 !kvm_event_needs_reinjection(vcpu);
5444 static int vapic_enter(struct kvm_vcpu *vcpu)
5446 struct kvm_lapic *apic = vcpu->arch.apic;
5449 if (!apic || !apic->vapic_addr)
5452 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5453 if (is_error_page(page))
5456 vcpu->arch.apic->vapic_page = page;
5460 static void vapic_exit(struct kvm_vcpu *vcpu)
5462 struct kvm_lapic *apic = vcpu->arch.apic;
5465 if (!apic || !apic->vapic_addr)
5468 idx = srcu_read_lock(&vcpu->kvm->srcu);
5469 kvm_release_page_dirty(apic->vapic_page);
5470 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5471 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5474 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5478 if (!kvm_x86_ops->update_cr8_intercept)
5481 if (!vcpu->arch.apic)
5484 if (!vcpu->arch.apic->vapic_addr)
5485 max_irr = kvm_lapic_find_highest_irr(vcpu);
5492 tpr = kvm_lapic_get_cr8(vcpu);
5494 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5497 static void inject_pending_event(struct kvm_vcpu *vcpu)
5499 /* try to reinject previous events if any */
5500 if (vcpu->arch.exception.pending) {
5501 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5502 vcpu->arch.exception.has_error_code,
5503 vcpu->arch.exception.error_code);
5504 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5505 vcpu->arch.exception.has_error_code,
5506 vcpu->arch.exception.error_code,
5507 vcpu->arch.exception.reinject);
5511 if (vcpu->arch.nmi_injected) {
5512 kvm_x86_ops->set_nmi(vcpu);
5516 if (vcpu->arch.interrupt.pending) {
5517 kvm_x86_ops->set_irq(vcpu);
5521 /* try to inject new event if pending */
5522 if (vcpu->arch.nmi_pending) {
5523 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5524 --vcpu->arch.nmi_pending;
5525 vcpu->arch.nmi_injected = true;
5526 kvm_x86_ops->set_nmi(vcpu);
5528 } else if (kvm_cpu_has_interrupt(vcpu)) {
5529 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5530 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5532 kvm_x86_ops->set_irq(vcpu);
5537 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5539 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5540 !vcpu->guest_xcr0_loaded) {
5541 /* kvm_set_xcr() also depends on this */
5542 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5543 vcpu->guest_xcr0_loaded = 1;
5547 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5549 if (vcpu->guest_xcr0_loaded) {
5550 if (vcpu->arch.xcr0 != host_xcr0)
5551 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5552 vcpu->guest_xcr0_loaded = 0;
5556 static void process_nmi(struct kvm_vcpu *vcpu)
5561 * x86 is limited to one NMI running, and one NMI pending after it.
5562 * If an NMI is already in progress, limit further NMIs to just one.
5563 * Otherwise, allow two (and we'll inject the first one immediately).
5565 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5568 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5569 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5570 kvm_make_request(KVM_REQ_EVENT, vcpu);
5573 static void kvm_gen_update_masterclock(struct kvm *kvm)
5575 #ifdef CONFIG_X86_64
5577 struct kvm_vcpu *vcpu;
5578 struct kvm_arch *ka = &kvm->arch;
5580 spin_lock(&ka->pvclock_gtod_sync_lock);
5581 kvm_make_mclock_inprogress_request(kvm);
5582 /* no guest entries from this point */
5583 pvclock_update_vm_gtod_copy(kvm);
5585 kvm_for_each_vcpu(i, vcpu, kvm)
5586 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
5588 /* guest entries allowed */
5589 kvm_for_each_vcpu(i, vcpu, kvm)
5590 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
5592 spin_unlock(&ka->pvclock_gtod_sync_lock);
5596 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5599 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5600 vcpu->run->request_interrupt_window;
5601 bool req_immediate_exit = 0;
5603 if (vcpu->requests) {
5604 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5605 kvm_mmu_unload(vcpu);
5606 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5607 __kvm_migrate_timers(vcpu);
5608 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
5609 kvm_gen_update_masterclock(vcpu->kvm);
5610 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5611 r = kvm_guest_time_update(vcpu);
5615 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5616 kvm_mmu_sync_roots(vcpu);
5617 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5618 kvm_x86_ops->tlb_flush(vcpu);
5619 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5620 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5624 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5625 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5629 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5630 vcpu->fpu_active = 0;
5631 kvm_x86_ops->fpu_deactivate(vcpu);
5633 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5634 /* Page is swapped out. Do synthetic halt */
5635 vcpu->arch.apf.halted = true;
5639 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5640 record_steal_time(vcpu);
5641 if (kvm_check_request(KVM_REQ_NMI, vcpu))
5643 req_immediate_exit =
5644 kvm_check_request(KVM_REQ_IMMEDIATE_EXIT, vcpu);
5645 if (kvm_check_request(KVM_REQ_PMU, vcpu))
5646 kvm_handle_pmu_event(vcpu);
5647 if (kvm_check_request(KVM_REQ_PMI, vcpu))
5648 kvm_deliver_pmi(vcpu);
5651 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5652 inject_pending_event(vcpu);
5654 /* enable NMI/IRQ window open exits if needed */
5655 if (vcpu->arch.nmi_pending)
5656 kvm_x86_ops->enable_nmi_window(vcpu);
5657 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5658 kvm_x86_ops->enable_irq_window(vcpu);
5660 if (kvm_lapic_enabled(vcpu)) {
5661 update_cr8_intercept(vcpu);
5662 kvm_lapic_sync_to_vapic(vcpu);
5666 r = kvm_mmu_reload(vcpu);
5668 goto cancel_injection;
5673 kvm_x86_ops->prepare_guest_switch(vcpu);
5674 if (vcpu->fpu_active)
5675 kvm_load_guest_fpu(vcpu);
5676 kvm_load_guest_xcr0(vcpu);
5678 vcpu->mode = IN_GUEST_MODE;
5680 /* We should set ->mode before check ->requests,
5681 * see the comment in make_all_cpus_request.
5685 local_irq_disable();
5687 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5688 || need_resched() || signal_pending(current)) {
5689 vcpu->mode = OUTSIDE_GUEST_MODE;
5694 goto cancel_injection;
5697 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5699 if (req_immediate_exit)
5700 smp_send_reschedule(vcpu->cpu);
5704 if (unlikely(vcpu->arch.switch_db_regs)) {
5706 set_debugreg(vcpu->arch.eff_db[0], 0);
5707 set_debugreg(vcpu->arch.eff_db[1], 1);
5708 set_debugreg(vcpu->arch.eff_db[2], 2);
5709 set_debugreg(vcpu->arch.eff_db[3], 3);
5712 trace_kvm_entry(vcpu->vcpu_id);
5713 kvm_x86_ops->run(vcpu);
5716 * If the guest has used debug registers, at least dr7
5717 * will be disabled while returning to the host.
5718 * If we don't have active breakpoints in the host, we don't
5719 * care about the messed up debug address registers. But if
5720 * we have some of them active, restore the old state.
5722 if (hw_breakpoint_active())
5723 hw_breakpoint_restore();
5725 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
5728 vcpu->mode = OUTSIDE_GUEST_MODE;
5735 * We must have an instruction between local_irq_enable() and
5736 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5737 * the interrupt shadow. The stat.exits increment will do nicely.
5738 * But we need to prevent reordering, hence this barrier():
5746 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5749 * Profile KVM exit RIPs:
5751 if (unlikely(prof_on == KVM_PROFILING)) {
5752 unsigned long rip = kvm_rip_read(vcpu);
5753 profile_hit(KVM_PROFILING, (void *)rip);
5756 if (unlikely(vcpu->arch.tsc_always_catchup))
5757 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5759 if (vcpu->arch.apic_attention)
5760 kvm_lapic_sync_from_vapic(vcpu);
5762 r = kvm_x86_ops->handle_exit(vcpu);
5766 kvm_x86_ops->cancel_injection(vcpu);
5767 if (unlikely(vcpu->arch.apic_attention))
5768 kvm_lapic_sync_from_vapic(vcpu);
5774 static int __vcpu_run(struct kvm_vcpu *vcpu)
5777 struct kvm *kvm = vcpu->kvm;
5779 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5780 pr_debug("vcpu %d received sipi with vector # %x\n",
5781 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5782 kvm_lapic_reset(vcpu);
5783 r = kvm_vcpu_reset(vcpu);
5786 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5789 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5790 r = vapic_enter(vcpu);
5792 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5798 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5799 !vcpu->arch.apf.halted)
5800 r = vcpu_enter_guest(vcpu);
5802 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5803 kvm_vcpu_block(vcpu);
5804 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5805 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5807 switch(vcpu->arch.mp_state) {
5808 case KVM_MP_STATE_HALTED:
5809 vcpu->arch.mp_state =
5810 KVM_MP_STATE_RUNNABLE;
5811 case KVM_MP_STATE_RUNNABLE:
5812 vcpu->arch.apf.halted = false;
5814 case KVM_MP_STATE_SIPI_RECEIVED:
5825 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5826 if (kvm_cpu_has_pending_timer(vcpu))
5827 kvm_inject_pending_timer_irqs(vcpu);
5829 if (dm_request_for_irq_injection(vcpu)) {
5831 vcpu->run->exit_reason = KVM_EXIT_INTR;
5832 ++vcpu->stat.request_irq_exits;
5835 kvm_check_async_pf_completion(vcpu);
5837 if (signal_pending(current)) {
5839 vcpu->run->exit_reason = KVM_EXIT_INTR;
5840 ++vcpu->stat.signal_exits;
5842 if (need_resched()) {
5843 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5845 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5849 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5856 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
5859 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5860 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5861 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5862 if (r != EMULATE_DONE)
5867 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
5869 BUG_ON(!vcpu->arch.pio.count);
5871 return complete_emulated_io(vcpu);
5875 * Implements the following, as a state machine:
5879 * for each mmio piece in the fragment
5887 * for each mmio piece in the fragment
5892 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
5894 struct kvm_run *run = vcpu->run;
5895 struct kvm_mmio_fragment *frag;
5898 BUG_ON(!vcpu->mmio_needed);
5900 /* Complete previous fragment */
5901 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
5902 len = min(8u, frag->len);
5903 if (!vcpu->mmio_is_write)
5904 memcpy(frag->data, run->mmio.data, len);
5906 if (frag->len <= 8) {
5907 /* Switch to the next fragment. */
5909 vcpu->mmio_cur_fragment++;
5911 /* Go forward to the next mmio piece. */
5917 if (vcpu->mmio_cur_fragment == vcpu->mmio_nr_fragments) {
5918 vcpu->mmio_needed = 0;
5919 if (vcpu->mmio_is_write)
5921 vcpu->mmio_read_completed = 1;
5922 return complete_emulated_io(vcpu);
5925 run->exit_reason = KVM_EXIT_MMIO;
5926 run->mmio.phys_addr = frag->gpa;
5927 if (vcpu->mmio_is_write)
5928 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
5929 run->mmio.len = min(8u, frag->len);
5930 run->mmio.is_write = vcpu->mmio_is_write;
5931 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5936 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5941 if (!tsk_used_math(current) && init_fpu(current))
5944 if (vcpu->sigset_active)
5945 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5947 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5948 kvm_vcpu_block(vcpu);
5949 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5954 /* re-sync apic's tpr */
5955 if (!irqchip_in_kernel(vcpu->kvm)) {
5956 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5962 if (unlikely(vcpu->arch.complete_userspace_io)) {
5963 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
5964 vcpu->arch.complete_userspace_io = NULL;
5969 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
5971 r = __vcpu_run(vcpu);
5974 post_kvm_run_save(vcpu);
5975 if (vcpu->sigset_active)
5976 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5981 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5983 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5985 * We are here if userspace calls get_regs() in the middle of
5986 * instruction emulation. Registers state needs to be copied
5987 * back from emulation context to vcpu. Userspace shouldn't do
5988 * that usually, but some bad designed PV devices (vmware
5989 * backdoor interface) need this to work
5991 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
5992 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5994 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5995 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5996 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5997 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5998 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5999 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6000 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6001 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6002 #ifdef CONFIG_X86_64
6003 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6004 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6005 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6006 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6007 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6008 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6009 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6010 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6013 regs->rip = kvm_rip_read(vcpu);
6014 regs->rflags = kvm_get_rflags(vcpu);
6019 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6021 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6022 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6024 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6025 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6026 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6027 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6028 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6029 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6030 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6031 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6032 #ifdef CONFIG_X86_64
6033 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6034 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6035 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6036 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6037 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6038 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6039 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6040 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6043 kvm_rip_write(vcpu, regs->rip);
6044 kvm_set_rflags(vcpu, regs->rflags);
6046 vcpu->arch.exception.pending = false;
6048 kvm_make_request(KVM_REQ_EVENT, vcpu);
6053 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6055 struct kvm_segment cs;
6057 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6061 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6063 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6064 struct kvm_sregs *sregs)
6068 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6069 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6070 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6071 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6072 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6073 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6075 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6076 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6078 kvm_x86_ops->get_idt(vcpu, &dt);
6079 sregs->idt.limit = dt.size;
6080 sregs->idt.base = dt.address;
6081 kvm_x86_ops->get_gdt(vcpu, &dt);
6082 sregs->gdt.limit = dt.size;
6083 sregs->gdt.base = dt.address;
6085 sregs->cr0 = kvm_read_cr0(vcpu);
6086 sregs->cr2 = vcpu->arch.cr2;
6087 sregs->cr3 = kvm_read_cr3(vcpu);
6088 sregs->cr4 = kvm_read_cr4(vcpu);
6089 sregs->cr8 = kvm_get_cr8(vcpu);
6090 sregs->efer = vcpu->arch.efer;
6091 sregs->apic_base = kvm_get_apic_base(vcpu);
6093 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6095 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6096 set_bit(vcpu->arch.interrupt.nr,
6097 (unsigned long *)sregs->interrupt_bitmap);
6102 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6103 struct kvm_mp_state *mp_state)
6105 mp_state->mp_state = vcpu->arch.mp_state;
6109 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6110 struct kvm_mp_state *mp_state)
6112 vcpu->arch.mp_state = mp_state->mp_state;
6113 kvm_make_request(KVM_REQ_EVENT, vcpu);
6117 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6118 int reason, bool has_error_code, u32 error_code)
6120 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6123 init_emulate_ctxt(vcpu);
6125 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6126 has_error_code, error_code);
6129 return EMULATE_FAIL;
6131 kvm_rip_write(vcpu, ctxt->eip);
6132 kvm_set_rflags(vcpu, ctxt->eflags);
6133 kvm_make_request(KVM_REQ_EVENT, vcpu);
6134 return EMULATE_DONE;
6136 EXPORT_SYMBOL_GPL(kvm_task_switch);
6138 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6139 struct kvm_sregs *sregs)
6141 int mmu_reset_needed = 0;
6142 int pending_vec, max_bits, idx;
6145 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6148 dt.size = sregs->idt.limit;
6149 dt.address = sregs->idt.base;
6150 kvm_x86_ops->set_idt(vcpu, &dt);
6151 dt.size = sregs->gdt.limit;
6152 dt.address = sregs->gdt.base;
6153 kvm_x86_ops->set_gdt(vcpu, &dt);
6155 vcpu->arch.cr2 = sregs->cr2;
6156 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6157 vcpu->arch.cr3 = sregs->cr3;
6158 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6160 kvm_set_cr8(vcpu, sregs->cr8);
6162 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6163 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6164 kvm_set_apic_base(vcpu, sregs->apic_base);
6166 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6167 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6168 vcpu->arch.cr0 = sregs->cr0;
6170 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6171 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6172 if (sregs->cr4 & X86_CR4_OSXSAVE)
6173 kvm_update_cpuid(vcpu);
6175 idx = srcu_read_lock(&vcpu->kvm->srcu);
6176 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6177 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6178 mmu_reset_needed = 1;
6180 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6182 if (mmu_reset_needed)
6183 kvm_mmu_reset_context(vcpu);
6185 max_bits = KVM_NR_INTERRUPTS;
6186 pending_vec = find_first_bit(
6187 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6188 if (pending_vec < max_bits) {
6189 kvm_queue_interrupt(vcpu, pending_vec, false);
6190 pr_debug("Set back pending irq %d\n", pending_vec);
6193 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6194 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6195 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6196 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6197 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6198 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6200 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6201 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6203 update_cr8_intercept(vcpu);
6205 /* Older userspace won't unhalt the vcpu on reset. */
6206 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6207 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6209 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6211 kvm_make_request(KVM_REQ_EVENT, vcpu);
6216 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6217 struct kvm_guest_debug *dbg)
6219 unsigned long rflags;
6222 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6224 if (vcpu->arch.exception.pending)
6226 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6227 kvm_queue_exception(vcpu, DB_VECTOR);
6229 kvm_queue_exception(vcpu, BP_VECTOR);
6233 * Read rflags as long as potentially injected trace flags are still
6236 rflags = kvm_get_rflags(vcpu);
6238 vcpu->guest_debug = dbg->control;
6239 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6240 vcpu->guest_debug = 0;
6242 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6243 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6244 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6245 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6247 for (i = 0; i < KVM_NR_DB_REGS; i++)
6248 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6250 kvm_update_dr7(vcpu);
6252 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6253 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6254 get_segment_base(vcpu, VCPU_SREG_CS);
6257 * Trigger an rflags update that will inject or remove the trace
6260 kvm_set_rflags(vcpu, rflags);
6262 kvm_x86_ops->update_db_bp_intercept(vcpu);
6272 * Translate a guest virtual address to a guest physical address.
6274 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6275 struct kvm_translation *tr)
6277 unsigned long vaddr = tr->linear_address;
6281 idx = srcu_read_lock(&vcpu->kvm->srcu);
6282 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6283 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6284 tr->physical_address = gpa;
6285 tr->valid = gpa != UNMAPPED_GVA;
6292 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6294 struct i387_fxsave_struct *fxsave =
6295 &vcpu->arch.guest_fpu.state->fxsave;
6297 memcpy(fpu->fpr, fxsave->st_space, 128);
6298 fpu->fcw = fxsave->cwd;
6299 fpu->fsw = fxsave->swd;
6300 fpu->ftwx = fxsave->twd;
6301 fpu->last_opcode = fxsave->fop;
6302 fpu->last_ip = fxsave->rip;
6303 fpu->last_dp = fxsave->rdp;
6304 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6309 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6311 struct i387_fxsave_struct *fxsave =
6312 &vcpu->arch.guest_fpu.state->fxsave;
6314 memcpy(fxsave->st_space, fpu->fpr, 128);
6315 fxsave->cwd = fpu->fcw;
6316 fxsave->swd = fpu->fsw;
6317 fxsave->twd = fpu->ftwx;
6318 fxsave->fop = fpu->last_opcode;
6319 fxsave->rip = fpu->last_ip;
6320 fxsave->rdp = fpu->last_dp;
6321 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6326 int fx_init(struct kvm_vcpu *vcpu)
6330 err = fpu_alloc(&vcpu->arch.guest_fpu);
6334 fpu_finit(&vcpu->arch.guest_fpu);
6337 * Ensure guest xcr0 is valid for loading
6339 vcpu->arch.xcr0 = XSTATE_FP;
6341 vcpu->arch.cr0 |= X86_CR0_ET;
6345 EXPORT_SYMBOL_GPL(fx_init);
6347 static void fx_free(struct kvm_vcpu *vcpu)
6349 fpu_free(&vcpu->arch.guest_fpu);
6352 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6354 if (vcpu->guest_fpu_loaded)
6358 * Restore all possible states in the guest,
6359 * and assume host would use all available bits.
6360 * Guest xcr0 would be loaded later.
6362 kvm_put_guest_xcr0(vcpu);
6363 vcpu->guest_fpu_loaded = 1;
6364 __kernel_fpu_begin();
6365 fpu_restore_checking(&vcpu->arch.guest_fpu);
6369 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6371 kvm_put_guest_xcr0(vcpu);
6373 if (!vcpu->guest_fpu_loaded)
6376 vcpu->guest_fpu_loaded = 0;
6377 fpu_save_init(&vcpu->arch.guest_fpu);
6379 ++vcpu->stat.fpu_reload;
6380 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6384 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6386 kvmclock_reset(vcpu);
6388 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6390 kvm_x86_ops->vcpu_free(vcpu);
6393 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6396 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6397 printk_once(KERN_WARNING
6398 "kvm: SMP vm created on host with unstable TSC; "
6399 "guest TSC will not be reliable\n");
6400 return kvm_x86_ops->vcpu_create(kvm, id);
6403 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6407 vcpu->arch.mtrr_state.have_fixed = 1;
6408 r = vcpu_load(vcpu);
6411 r = kvm_vcpu_reset(vcpu);
6413 r = kvm_mmu_setup(vcpu);
6419 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
6422 struct msr_data msr;
6424 r = vcpu_load(vcpu);
6428 msr.index = MSR_IA32_TSC;
6429 msr.host_initiated = true;
6430 kvm_write_tsc(vcpu, &msr);
6436 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6439 vcpu->arch.apf.msr_val = 0;
6441 r = vcpu_load(vcpu);
6443 kvm_mmu_unload(vcpu);
6447 kvm_x86_ops->vcpu_free(vcpu);
6450 static int kvm_vcpu_reset(struct kvm_vcpu *vcpu)
6452 atomic_set(&vcpu->arch.nmi_queued, 0);
6453 vcpu->arch.nmi_pending = 0;
6454 vcpu->arch.nmi_injected = false;
6456 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6457 vcpu->arch.dr6 = DR6_FIXED_1;
6458 vcpu->arch.dr7 = DR7_FIXED_1;
6459 kvm_update_dr7(vcpu);
6461 kvm_make_request(KVM_REQ_EVENT, vcpu);
6462 vcpu->arch.apf.msr_val = 0;
6463 vcpu->arch.st.msr_val = 0;
6465 kvmclock_reset(vcpu);
6467 kvm_clear_async_pf_completion_queue(vcpu);
6468 kvm_async_pf_hash_reset(vcpu);
6469 vcpu->arch.apf.halted = false;
6471 kvm_pmu_reset(vcpu);
6473 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
6474 vcpu->arch.regs_avail = ~0;
6475 vcpu->arch.regs_dirty = ~0;
6477 return kvm_x86_ops->vcpu_reset(vcpu);
6480 int kvm_arch_hardware_enable(void *garbage)
6483 struct kvm_vcpu *vcpu;
6488 bool stable, backwards_tsc = false;
6490 kvm_shared_msr_cpu_online();
6491 ret = kvm_x86_ops->hardware_enable(garbage);
6495 local_tsc = native_read_tsc();
6496 stable = !check_tsc_unstable();
6497 list_for_each_entry(kvm, &vm_list, vm_list) {
6498 kvm_for_each_vcpu(i, vcpu, kvm) {
6499 if (!stable && vcpu->cpu == smp_processor_id())
6500 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
6501 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
6502 backwards_tsc = true;
6503 if (vcpu->arch.last_host_tsc > max_tsc)
6504 max_tsc = vcpu->arch.last_host_tsc;
6510 * Sometimes, even reliable TSCs go backwards. This happens on
6511 * platforms that reset TSC during suspend or hibernate actions, but
6512 * maintain synchronization. We must compensate. Fortunately, we can
6513 * detect that condition here, which happens early in CPU bringup,
6514 * before any KVM threads can be running. Unfortunately, we can't
6515 * bring the TSCs fully up to date with real time, as we aren't yet far
6516 * enough into CPU bringup that we know how much real time has actually
6517 * elapsed; our helper function, get_kernel_ns() will be using boot
6518 * variables that haven't been updated yet.
6520 * So we simply find the maximum observed TSC above, then record the
6521 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
6522 * the adjustment will be applied. Note that we accumulate
6523 * adjustments, in case multiple suspend cycles happen before some VCPU
6524 * gets a chance to run again. In the event that no KVM threads get a
6525 * chance to run, we will miss the entire elapsed period, as we'll have
6526 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
6527 * loose cycle time. This isn't too big a deal, since the loss will be
6528 * uniform across all VCPUs (not to mention the scenario is extremely
6529 * unlikely). It is possible that a second hibernate recovery happens
6530 * much faster than a first, causing the observed TSC here to be
6531 * smaller; this would require additional padding adjustment, which is
6532 * why we set last_host_tsc to the local tsc observed here.
6534 * N.B. - this code below runs only on platforms with reliable TSC,
6535 * as that is the only way backwards_tsc is set above. Also note
6536 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
6537 * have the same delta_cyc adjustment applied if backwards_tsc
6538 * is detected. Note further, this adjustment is only done once,
6539 * as we reset last_host_tsc on all VCPUs to stop this from being
6540 * called multiple times (one for each physical CPU bringup).
6542 * Platforms with unreliable TSCs don't have to deal with this, they
6543 * will be compensated by the logic in vcpu_load, which sets the TSC to
6544 * catchup mode. This will catchup all VCPUs to real time, but cannot
6545 * guarantee that they stay in perfect synchronization.
6547 if (backwards_tsc) {
6548 u64 delta_cyc = max_tsc - local_tsc;
6549 list_for_each_entry(kvm, &vm_list, vm_list) {
6550 kvm_for_each_vcpu(i, vcpu, kvm) {
6551 vcpu->arch.tsc_offset_adjustment += delta_cyc;
6552 vcpu->arch.last_host_tsc = local_tsc;
6553 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
6558 * We have to disable TSC offset matching.. if you were
6559 * booting a VM while issuing an S4 host suspend....
6560 * you may have some problem. Solving this issue is
6561 * left as an exercise to the reader.
6563 kvm->arch.last_tsc_nsec = 0;
6564 kvm->arch.last_tsc_write = 0;
6571 void kvm_arch_hardware_disable(void *garbage)
6573 kvm_x86_ops->hardware_disable(garbage);
6574 drop_user_return_notifiers(garbage);
6577 int kvm_arch_hardware_setup(void)
6579 return kvm_x86_ops->hardware_setup();
6582 void kvm_arch_hardware_unsetup(void)
6584 kvm_x86_ops->hardware_unsetup();
6587 void kvm_arch_check_processor_compat(void *rtn)
6589 kvm_x86_ops->check_processor_compatibility(rtn);
6592 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
6594 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
6597 struct static_key kvm_no_apic_vcpu __read_mostly;
6599 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6605 BUG_ON(vcpu->kvm == NULL);
6608 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6609 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6610 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6612 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6614 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6619 vcpu->arch.pio_data = page_address(page);
6621 kvm_set_tsc_khz(vcpu, max_tsc_khz);
6623 r = kvm_mmu_create(vcpu);
6625 goto fail_free_pio_data;
6627 if (irqchip_in_kernel(kvm)) {
6628 r = kvm_create_lapic(vcpu);
6630 goto fail_mmu_destroy;
6632 static_key_slow_inc(&kvm_no_apic_vcpu);
6634 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6636 if (!vcpu->arch.mce_banks) {
6638 goto fail_free_lapic;
6640 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6642 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6643 goto fail_free_mce_banks;
6647 goto fail_free_wbinvd_dirty_mask;
6649 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
6650 kvm_async_pf_hash_reset(vcpu);
6654 fail_free_wbinvd_dirty_mask:
6655 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6656 fail_free_mce_banks:
6657 kfree(vcpu->arch.mce_banks);
6659 kvm_free_lapic(vcpu);
6661 kvm_mmu_destroy(vcpu);
6663 free_page((unsigned long)vcpu->arch.pio_data);
6668 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6672 kvm_pmu_destroy(vcpu);
6673 kfree(vcpu->arch.mce_banks);
6674 kvm_free_lapic(vcpu);
6675 idx = srcu_read_lock(&vcpu->kvm->srcu);
6676 kvm_mmu_destroy(vcpu);
6677 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6678 free_page((unsigned long)vcpu->arch.pio_data);
6679 if (!irqchip_in_kernel(vcpu->kvm))
6680 static_key_slow_dec(&kvm_no_apic_vcpu);
6683 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
6688 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6689 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6691 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6692 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6693 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
6694 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
6695 &kvm->arch.irq_sources_bitmap);
6697 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6698 mutex_init(&kvm->arch.apic_map_lock);
6699 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
6701 pvclock_update_vm_gtod_copy(kvm);
6706 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6709 r = vcpu_load(vcpu);
6711 kvm_mmu_unload(vcpu);
6715 static void kvm_free_vcpus(struct kvm *kvm)
6718 struct kvm_vcpu *vcpu;
6721 * Unpin any mmu pages first.
6723 kvm_for_each_vcpu(i, vcpu, kvm) {
6724 kvm_clear_async_pf_completion_queue(vcpu);
6725 kvm_unload_vcpu_mmu(vcpu);
6727 kvm_for_each_vcpu(i, vcpu, kvm)
6728 kvm_arch_vcpu_free(vcpu);
6730 mutex_lock(&kvm->lock);
6731 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6732 kvm->vcpus[i] = NULL;
6734 atomic_set(&kvm->online_vcpus, 0);
6735 mutex_unlock(&kvm->lock);
6738 void kvm_arch_sync_events(struct kvm *kvm)
6740 kvm_free_all_assigned_devices(kvm);
6744 void kvm_arch_destroy_vm(struct kvm *kvm)
6746 kvm_iommu_unmap_guest(kvm);
6747 kfree(kvm->arch.vpic);
6748 kfree(kvm->arch.vioapic);
6749 kvm_free_vcpus(kvm);
6750 if (kvm->arch.apic_access_page)
6751 put_page(kvm->arch.apic_access_page);
6752 if (kvm->arch.ept_identity_pagetable)
6753 put_page(kvm->arch.ept_identity_pagetable);
6754 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
6757 void kvm_arch_free_memslot(struct kvm_memory_slot *free,
6758 struct kvm_memory_slot *dont)
6762 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6763 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
6764 kvm_kvfree(free->arch.rmap[i]);
6765 free->arch.rmap[i] = NULL;
6770 if (!dont || free->arch.lpage_info[i - 1] !=
6771 dont->arch.lpage_info[i - 1]) {
6772 kvm_kvfree(free->arch.lpage_info[i - 1]);
6773 free->arch.lpage_info[i - 1] = NULL;
6778 int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages)
6782 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6787 lpages = gfn_to_index(slot->base_gfn + npages - 1,
6788 slot->base_gfn, level) + 1;
6790 slot->arch.rmap[i] =
6791 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
6792 if (!slot->arch.rmap[i])
6797 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
6798 sizeof(*slot->arch.lpage_info[i - 1]));
6799 if (!slot->arch.lpage_info[i - 1])
6802 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
6803 slot->arch.lpage_info[i - 1][0].write_count = 1;
6804 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
6805 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
6806 ugfn = slot->userspace_addr >> PAGE_SHIFT;
6808 * If the gfn and userspace address are not aligned wrt each
6809 * other, or if explicitly asked to, disable large page
6810 * support for this slot
6812 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
6813 !kvm_largepages_enabled()) {
6816 for (j = 0; j < lpages; ++j)
6817 slot->arch.lpage_info[i - 1][j].write_count = 1;
6824 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6825 kvm_kvfree(slot->arch.rmap[i]);
6826 slot->arch.rmap[i] = NULL;
6830 kvm_kvfree(slot->arch.lpage_info[i - 1]);
6831 slot->arch.lpage_info[i - 1] = NULL;
6836 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6837 struct kvm_memory_slot *memslot,
6838 struct kvm_memory_slot old,
6839 struct kvm_userspace_memory_region *mem,
6842 int npages = memslot->npages;
6843 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6845 /* Prevent internal slot pages from being moved by fork()/COW. */
6846 if (memslot->id >= KVM_USER_MEM_SLOTS)
6847 map_flags = MAP_SHARED | MAP_ANONYMOUS;
6849 /*To keep backward compatibility with older userspace,
6850 *x86 needs to handle !user_alloc case.
6853 if (npages && !old.npages) {
6854 unsigned long userspace_addr;
6856 userspace_addr = vm_mmap(NULL, 0,
6858 PROT_READ | PROT_WRITE,
6862 if (IS_ERR((void *)userspace_addr))
6863 return PTR_ERR((void *)userspace_addr);
6865 memslot->userspace_addr = userspace_addr;
6873 void kvm_arch_commit_memory_region(struct kvm *kvm,
6874 struct kvm_userspace_memory_region *mem,
6875 struct kvm_memory_slot old,
6879 int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
6881 if (!user_alloc && !old.user_alloc && old.npages && !npages) {
6884 ret = vm_munmap(old.userspace_addr,
6885 old.npages * PAGE_SIZE);
6888 "kvm_vm_ioctl_set_memory_region: "
6889 "failed to munmap memory\n");
6892 if (!kvm->arch.n_requested_mmu_pages)
6893 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6896 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6898 * Write protect all pages for dirty logging.
6899 * Existing largepage mappings are destroyed here and new ones will
6900 * not be created until the end of the logging.
6902 if (npages && (mem->flags & KVM_MEM_LOG_DIRTY_PAGES)) {
6903 spin_lock(&kvm->mmu_lock);
6904 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6905 spin_unlock(&kvm->mmu_lock);
6908 * If memory slot is created, or moved, we need to clear all
6911 if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT) {
6912 kvm_mmu_zap_all(kvm);
6913 kvm_reload_remote_mmus(kvm);
6917 void kvm_arch_flush_shadow_all(struct kvm *kvm)
6919 kvm_mmu_zap_all(kvm);
6920 kvm_reload_remote_mmus(kvm);
6923 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
6924 struct kvm_memory_slot *slot)
6926 kvm_arch_flush_shadow_all(kvm);
6929 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6931 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6932 !vcpu->arch.apf.halted)
6933 || !list_empty_careful(&vcpu->async_pf.done)
6934 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6935 || atomic_read(&vcpu->arch.nmi_queued) ||
6936 (kvm_arch_interrupt_allowed(vcpu) &&
6937 kvm_cpu_has_interrupt(vcpu));
6940 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
6942 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
6945 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6947 return kvm_x86_ops->interrupt_allowed(vcpu);
6950 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6952 unsigned long current_rip = kvm_rip_read(vcpu) +
6953 get_segment_base(vcpu, VCPU_SREG_CS);
6955 return current_rip == linear_rip;
6957 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6959 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6961 unsigned long rflags;
6963 rflags = kvm_x86_ops->get_rflags(vcpu);
6964 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6965 rflags &= ~X86_EFLAGS_TF;
6968 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6970 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6972 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6973 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6974 rflags |= X86_EFLAGS_TF;
6975 kvm_x86_ops->set_rflags(vcpu, rflags);
6976 kvm_make_request(KVM_REQ_EVENT, vcpu);
6978 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6980 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6984 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6985 is_error_page(work->page))
6988 r = kvm_mmu_reload(vcpu);
6992 if (!vcpu->arch.mmu.direct_map &&
6993 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6996 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6999 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7001 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7004 static inline u32 kvm_async_pf_next_probe(u32 key)
7006 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7009 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7011 u32 key = kvm_async_pf_hash_fn(gfn);
7013 while (vcpu->arch.apf.gfns[key] != ~0)
7014 key = kvm_async_pf_next_probe(key);
7016 vcpu->arch.apf.gfns[key] = gfn;
7019 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7022 u32 key = kvm_async_pf_hash_fn(gfn);
7024 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
7025 (vcpu->arch.apf.gfns[key] != gfn &&
7026 vcpu->arch.apf.gfns[key] != ~0); i++)
7027 key = kvm_async_pf_next_probe(key);
7032 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7034 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7037 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7041 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7043 vcpu->arch.apf.gfns[i] = ~0;
7045 j = kvm_async_pf_next_probe(j);
7046 if (vcpu->arch.apf.gfns[j] == ~0)
7048 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7050 * k lies cyclically in ]i,j]
7052 * |....j i.k.| or |.k..j i...|
7054 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7055 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7060 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7063 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7067 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7068 struct kvm_async_pf *work)
7070 struct x86_exception fault;
7072 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
7073 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7075 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
7076 (vcpu->arch.apf.send_user_only &&
7077 kvm_x86_ops->get_cpl(vcpu) == 0))
7078 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7079 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
7080 fault.vector = PF_VECTOR;
7081 fault.error_code_valid = true;
7082 fault.error_code = 0;
7083 fault.nested_page_fault = false;
7084 fault.address = work->arch.token;
7085 kvm_inject_page_fault(vcpu, &fault);
7089 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7090 struct kvm_async_pf *work)
7092 struct x86_exception fault;
7094 trace_kvm_async_pf_ready(work->arch.token, work->gva);
7095 if (is_error_page(work->page))
7096 work->arch.token = ~0; /* broadcast wakeup */
7098 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7100 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7101 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
7102 fault.vector = PF_VECTOR;
7103 fault.error_code_valid = true;
7104 fault.error_code = 0;
7105 fault.nested_page_fault = false;
7106 fault.address = work->arch.token;
7107 kvm_inject_page_fault(vcpu, &fault);
7109 vcpu->arch.apf.halted = false;
7110 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7113 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7115 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7118 return !kvm_event_needs_reinjection(vcpu) &&
7119 kvm_x86_ops->interrupt_allowed(vcpu);
7122 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
7123 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
7124 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
7125 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
7126 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
7127 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
7128 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
7129 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
7130 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
7131 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
7132 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
7133 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);