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[] = {
834 MSR_IA32_TSCDEADLINE,
835 MSR_IA32_MISC_ENABLE,
840 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
842 u64 old_efer = vcpu->arch.efer;
844 if (efer & efer_reserved_bits)
848 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
851 if (efer & EFER_FFXSR) {
852 struct kvm_cpuid_entry2 *feat;
854 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
855 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
859 if (efer & EFER_SVME) {
860 struct kvm_cpuid_entry2 *feat;
862 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
863 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
868 efer |= vcpu->arch.efer & EFER_LMA;
870 kvm_x86_ops->set_efer(vcpu, efer);
872 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
874 /* Update reserved bits */
875 if ((efer ^ old_efer) & EFER_NX)
876 kvm_mmu_reset_context(vcpu);
881 void kvm_enable_efer_bits(u64 mask)
883 efer_reserved_bits &= ~mask;
885 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
889 * Writes msr value into into the appropriate "register".
890 * Returns 0 on success, non-0 otherwise.
891 * Assumes vcpu_load() was already called.
893 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
895 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
899 * Adapt set_msr() to msr_io()'s calling convention
901 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
903 return kvm_set_msr(vcpu, index, *data);
907 struct pvclock_gtod_data {
910 struct { /* extract of a clocksource struct */
918 /* open coded 'struct timespec' */
919 u64 monotonic_time_snsec;
920 time_t monotonic_time_sec;
923 static struct pvclock_gtod_data pvclock_gtod_data;
925 static void update_pvclock_gtod(struct timekeeper *tk)
927 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
929 write_seqcount_begin(&vdata->seq);
931 /* copy pvclock gtod data */
932 vdata->clock.vclock_mode = tk->clock->archdata.vclock_mode;
933 vdata->clock.cycle_last = tk->clock->cycle_last;
934 vdata->clock.mask = tk->clock->mask;
935 vdata->clock.mult = tk->mult;
936 vdata->clock.shift = tk->shift;
938 vdata->monotonic_time_sec = tk->xtime_sec
939 + tk->wall_to_monotonic.tv_sec;
940 vdata->monotonic_time_snsec = tk->xtime_nsec
941 + (tk->wall_to_monotonic.tv_nsec
943 while (vdata->monotonic_time_snsec >=
944 (((u64)NSEC_PER_SEC) << tk->shift)) {
945 vdata->monotonic_time_snsec -=
946 ((u64)NSEC_PER_SEC) << tk->shift;
947 vdata->monotonic_time_sec++;
950 write_seqcount_end(&vdata->seq);
955 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
959 struct pvclock_wall_clock wc;
960 struct timespec boot;
965 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
970 ++version; /* first time write, random junk */
974 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
977 * The guest calculates current wall clock time by adding
978 * system time (updated by kvm_guest_time_update below) to the
979 * wall clock specified here. guest system time equals host
980 * system time for us, thus we must fill in host boot time here.
984 if (kvm->arch.kvmclock_offset) {
985 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
986 boot = timespec_sub(boot, ts);
988 wc.sec = boot.tv_sec;
989 wc.nsec = boot.tv_nsec;
990 wc.version = version;
992 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
995 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
998 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1000 uint32_t quotient, remainder;
1002 /* Don't try to replace with do_div(), this one calculates
1003 * "(dividend << 32) / divisor" */
1005 : "=a" (quotient), "=d" (remainder)
1006 : "0" (0), "1" (dividend), "r" (divisor) );
1010 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1011 s8 *pshift, u32 *pmultiplier)
1018 tps64 = base_khz * 1000LL;
1019 scaled64 = scaled_khz * 1000LL;
1020 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1025 tps32 = (uint32_t)tps64;
1026 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1027 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1035 *pmultiplier = div_frac(scaled64, tps32);
1037 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1038 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1041 static inline u64 get_kernel_ns(void)
1045 WARN_ON(preemptible());
1047 monotonic_to_bootbased(&ts);
1048 return timespec_to_ns(&ts);
1051 #ifdef CONFIG_X86_64
1052 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1055 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1056 unsigned long max_tsc_khz;
1058 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1060 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1061 vcpu->arch.virtual_tsc_shift);
1064 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1066 u64 v = (u64)khz * (1000000 + ppm);
1071 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1073 u32 thresh_lo, thresh_hi;
1074 int use_scaling = 0;
1076 /* Compute a scale to convert nanoseconds in TSC cycles */
1077 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1078 &vcpu->arch.virtual_tsc_shift,
1079 &vcpu->arch.virtual_tsc_mult);
1080 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1083 * Compute the variation in TSC rate which is acceptable
1084 * within the range of tolerance and decide if the
1085 * rate being applied is within that bounds of the hardware
1086 * rate. If so, no scaling or compensation need be done.
1088 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1089 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1090 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1091 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1094 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1097 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1099 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1100 vcpu->arch.virtual_tsc_mult,
1101 vcpu->arch.virtual_tsc_shift);
1102 tsc += vcpu->arch.this_tsc_write;
1106 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1108 struct kvm *kvm = vcpu->kvm;
1109 u64 offset, ns, elapsed;
1110 unsigned long flags;
1113 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1114 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1115 ns = get_kernel_ns();
1116 elapsed = ns - kvm->arch.last_tsc_nsec;
1118 /* n.b - signed multiplication and division required */
1119 usdiff = data - kvm->arch.last_tsc_write;
1120 #ifdef CONFIG_X86_64
1121 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1123 /* do_div() only does unsigned */
1124 asm("idivl %2; xor %%edx, %%edx"
1126 : "A"(usdiff * 1000), "rm"(vcpu->arch.virtual_tsc_khz));
1128 do_div(elapsed, 1000);
1134 * Special case: TSC write with a small delta (1 second) of virtual
1135 * cycle time against real time is interpreted as an attempt to
1136 * synchronize the CPU.
1138 * For a reliable TSC, we can match TSC offsets, and for an unstable
1139 * TSC, we add elapsed time in this computation. We could let the
1140 * compensation code attempt to catch up if we fall behind, but
1141 * it's better to try to match offsets from the beginning.
1143 if (usdiff < USEC_PER_SEC &&
1144 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1145 if (!check_tsc_unstable()) {
1146 offset = kvm->arch.cur_tsc_offset;
1147 pr_debug("kvm: matched tsc offset for %llu\n", data);
1149 u64 delta = nsec_to_cycles(vcpu, elapsed);
1151 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1152 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1156 * We split periods of matched TSC writes into generations.
1157 * For each generation, we track the original measured
1158 * nanosecond time, offset, and write, so if TSCs are in
1159 * sync, we can match exact offset, and if not, we can match
1160 * exact software computation in compute_guest_tsc()
1162 * These values are tracked in kvm->arch.cur_xxx variables.
1164 kvm->arch.cur_tsc_generation++;
1165 kvm->arch.cur_tsc_nsec = ns;
1166 kvm->arch.cur_tsc_write = data;
1167 kvm->arch.cur_tsc_offset = offset;
1168 pr_debug("kvm: new tsc generation %u, clock %llu\n",
1169 kvm->arch.cur_tsc_generation, data);
1173 * We also track th most recent recorded KHZ, write and time to
1174 * allow the matching interval to be extended at each write.
1176 kvm->arch.last_tsc_nsec = ns;
1177 kvm->arch.last_tsc_write = data;
1178 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1180 /* Reset of TSC must disable overshoot protection below */
1181 vcpu->arch.hv_clock.tsc_timestamp = 0;
1182 vcpu->arch.last_guest_tsc = data;
1184 /* Keep track of which generation this VCPU has synchronized to */
1185 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1186 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1187 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1189 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1190 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1193 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1195 #ifdef CONFIG_X86_64
1197 static cycle_t read_tsc(void)
1203 * Empirically, a fence (of type that depends on the CPU)
1204 * before rdtsc is enough to ensure that rdtsc is ordered
1205 * with respect to loads. The various CPU manuals are unclear
1206 * as to whether rdtsc can be reordered with later loads,
1207 * but no one has ever seen it happen.
1210 ret = (cycle_t)vget_cycles();
1212 last = pvclock_gtod_data.clock.cycle_last;
1214 if (likely(ret >= last))
1218 * GCC likes to generate cmov here, but this branch is extremely
1219 * predictable (it's just a funciton of time and the likely is
1220 * very likely) and there's a data dependence, so force GCC
1221 * to generate a branch instead. I don't barrier() because
1222 * we don't actually need a barrier, and if this function
1223 * ever gets inlined it will generate worse code.
1229 static inline u64 vgettsc(cycle_t *cycle_now)
1232 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1234 *cycle_now = read_tsc();
1236 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1237 return v * gtod->clock.mult;
1240 static int do_monotonic(struct timespec *ts, cycle_t *cycle_now)
1245 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1249 seq = read_seqcount_begin(>od->seq);
1250 mode = gtod->clock.vclock_mode;
1251 ts->tv_sec = gtod->monotonic_time_sec;
1252 ns = gtod->monotonic_time_snsec;
1253 ns += vgettsc(cycle_now);
1254 ns >>= gtod->clock.shift;
1255 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1256 timespec_add_ns(ts, ns);
1261 /* returns true if host is using tsc clocksource */
1262 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1266 /* checked again under seqlock below */
1267 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1270 if (do_monotonic(&ts, cycle_now) != VCLOCK_TSC)
1273 monotonic_to_bootbased(&ts);
1274 *kernel_ns = timespec_to_ns(&ts);
1282 * Assuming a stable TSC across physical CPUS, the following condition
1283 * is possible. Each numbered line represents an event visible to both
1284 * CPUs at the next numbered event.
1286 * "timespecX" represents host monotonic time. "tscX" represents
1289 * VCPU0 on CPU0 | VCPU1 on CPU1
1291 * 1. read timespec0,tsc0
1292 * 2. | timespec1 = timespec0 + N
1294 * 3. transition to guest | transition to guest
1295 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1296 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1297 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1299 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1302 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1304 * - 0 < N - M => M < N
1306 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1307 * always the case (the difference between two distinct xtime instances
1308 * might be smaller then the difference between corresponding TSC reads,
1309 * when updating guest vcpus pvclock areas).
1311 * To avoid that problem, do not allow visibility of distinct
1312 * system_timestamp/tsc_timestamp values simultaneously: use a master
1313 * copy of host monotonic time values. Update that master copy
1316 * Rely on synchronization of host TSCs for monotonicity.
1320 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1322 #ifdef CONFIG_X86_64
1323 struct kvm_arch *ka = &kvm->arch;
1327 * If the host uses TSC clock, then passthrough TSC as stable
1330 ka->use_master_clock = kvm_get_time_and_clockread(
1331 &ka->master_kernel_ns,
1332 &ka->master_cycle_now);
1334 if (ka->use_master_clock)
1335 atomic_set(&kvm_guest_has_master_clock, 1);
1337 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1338 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode);
1342 static int kvm_guest_time_update(struct kvm_vcpu *v)
1344 unsigned long flags, this_tsc_khz;
1345 struct kvm_vcpu_arch *vcpu = &v->arch;
1346 struct kvm_arch *ka = &v->kvm->arch;
1348 s64 kernel_ns, max_kernel_ns;
1349 u64 tsc_timestamp, host_tsc;
1350 struct pvclock_vcpu_time_info *guest_hv_clock;
1352 bool use_master_clock;
1357 /* Keep irq disabled to prevent changes to the clock */
1358 local_irq_save(flags);
1359 this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
1360 if (unlikely(this_tsc_khz == 0)) {
1361 local_irq_restore(flags);
1362 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1367 * If the host uses TSC clock, then passthrough TSC as stable
1370 spin_lock(&ka->pvclock_gtod_sync_lock);
1371 use_master_clock = ka->use_master_clock;
1372 if (use_master_clock) {
1373 host_tsc = ka->master_cycle_now;
1374 kernel_ns = ka->master_kernel_ns;
1376 spin_unlock(&ka->pvclock_gtod_sync_lock);
1377 if (!use_master_clock) {
1378 host_tsc = native_read_tsc();
1379 kernel_ns = get_kernel_ns();
1382 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1385 * We may have to catch up the TSC to match elapsed wall clock
1386 * time for two reasons, even if kvmclock is used.
1387 * 1) CPU could have been running below the maximum TSC rate
1388 * 2) Broken TSC compensation resets the base at each VCPU
1389 * entry to avoid unknown leaps of TSC even when running
1390 * again on the same CPU. This may cause apparent elapsed
1391 * time to disappear, and the guest to stand still or run
1394 if (vcpu->tsc_catchup) {
1395 u64 tsc = compute_guest_tsc(v, kernel_ns);
1396 if (tsc > tsc_timestamp) {
1397 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1398 tsc_timestamp = tsc;
1402 local_irq_restore(flags);
1404 if (!vcpu->time_page)
1408 * Time as measured by the TSC may go backwards when resetting the base
1409 * tsc_timestamp. The reason for this is that the TSC resolution is
1410 * higher than the resolution of the other clock scales. Thus, many
1411 * possible measurments of the TSC correspond to one measurement of any
1412 * other clock, and so a spread of values is possible. This is not a
1413 * problem for the computation of the nanosecond clock; with TSC rates
1414 * around 1GHZ, there can only be a few cycles which correspond to one
1415 * nanosecond value, and any path through this code will inevitably
1416 * take longer than that. However, with the kernel_ns value itself,
1417 * the precision may be much lower, down to HZ granularity. If the
1418 * first sampling of TSC against kernel_ns ends in the low part of the
1419 * range, and the second in the high end of the range, we can get:
1421 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1423 * As the sampling errors potentially range in the thousands of cycles,
1424 * it is possible such a time value has already been observed by the
1425 * guest. To protect against this, we must compute the system time as
1426 * observed by the guest and ensure the new system time is greater.
1429 if (vcpu->hv_clock.tsc_timestamp) {
1430 max_kernel_ns = vcpu->last_guest_tsc -
1431 vcpu->hv_clock.tsc_timestamp;
1432 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1433 vcpu->hv_clock.tsc_to_system_mul,
1434 vcpu->hv_clock.tsc_shift);
1435 max_kernel_ns += vcpu->last_kernel_ns;
1438 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1439 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1440 &vcpu->hv_clock.tsc_shift,
1441 &vcpu->hv_clock.tsc_to_system_mul);
1442 vcpu->hw_tsc_khz = this_tsc_khz;
1445 /* with a master <monotonic time, tsc value> tuple,
1446 * pvclock clock reads always increase at the (scaled) rate
1447 * of guest TSC - no need to deal with sampling errors.
1449 if (!use_master_clock) {
1450 if (max_kernel_ns > kernel_ns)
1451 kernel_ns = max_kernel_ns;
1453 /* With all the info we got, fill in the values */
1454 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1455 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1456 vcpu->last_kernel_ns = kernel_ns;
1457 vcpu->last_guest_tsc = tsc_timestamp;
1460 * The interface expects us to write an even number signaling that the
1461 * update is finished. Since the guest won't see the intermediate
1462 * state, we just increase by 2 at the end.
1464 vcpu->hv_clock.version += 2;
1466 shared_kaddr = kmap_atomic(vcpu->time_page);
1468 guest_hv_clock = shared_kaddr + vcpu->time_offset;
1470 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1471 pvclock_flags = (guest_hv_clock->flags & PVCLOCK_GUEST_STOPPED);
1473 if (vcpu->pvclock_set_guest_stopped_request) {
1474 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1475 vcpu->pvclock_set_guest_stopped_request = false;
1478 /* If the host uses TSC clocksource, then it is stable */
1479 if (use_master_clock)
1480 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1482 vcpu->hv_clock.flags = pvclock_flags;
1484 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1485 sizeof(vcpu->hv_clock));
1487 kunmap_atomic(shared_kaddr);
1489 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1493 static bool msr_mtrr_valid(unsigned msr)
1496 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1497 case MSR_MTRRfix64K_00000:
1498 case MSR_MTRRfix16K_80000:
1499 case MSR_MTRRfix16K_A0000:
1500 case MSR_MTRRfix4K_C0000:
1501 case MSR_MTRRfix4K_C8000:
1502 case MSR_MTRRfix4K_D0000:
1503 case MSR_MTRRfix4K_D8000:
1504 case MSR_MTRRfix4K_E0000:
1505 case MSR_MTRRfix4K_E8000:
1506 case MSR_MTRRfix4K_F0000:
1507 case MSR_MTRRfix4K_F8000:
1508 case MSR_MTRRdefType:
1509 case MSR_IA32_CR_PAT:
1517 static bool valid_pat_type(unsigned t)
1519 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1522 static bool valid_mtrr_type(unsigned t)
1524 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1527 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1531 if (!msr_mtrr_valid(msr))
1534 if (msr == MSR_IA32_CR_PAT) {
1535 for (i = 0; i < 8; i++)
1536 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1539 } else if (msr == MSR_MTRRdefType) {
1542 return valid_mtrr_type(data & 0xff);
1543 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1544 for (i = 0; i < 8 ; i++)
1545 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1550 /* variable MTRRs */
1551 return valid_mtrr_type(data & 0xff);
1554 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1556 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1558 if (!mtrr_valid(vcpu, msr, data))
1561 if (msr == MSR_MTRRdefType) {
1562 vcpu->arch.mtrr_state.def_type = data;
1563 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1564 } else if (msr == MSR_MTRRfix64K_00000)
1566 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1567 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1568 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1569 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1570 else if (msr == MSR_IA32_CR_PAT)
1571 vcpu->arch.pat = data;
1572 else { /* Variable MTRRs */
1573 int idx, is_mtrr_mask;
1576 idx = (msr - 0x200) / 2;
1577 is_mtrr_mask = msr - 0x200 - 2 * idx;
1580 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1583 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1587 kvm_mmu_reset_context(vcpu);
1591 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1593 u64 mcg_cap = vcpu->arch.mcg_cap;
1594 unsigned bank_num = mcg_cap & 0xff;
1597 case MSR_IA32_MCG_STATUS:
1598 vcpu->arch.mcg_status = data;
1600 case MSR_IA32_MCG_CTL:
1601 if (!(mcg_cap & MCG_CTL_P))
1603 if (data != 0 && data != ~(u64)0)
1605 vcpu->arch.mcg_ctl = data;
1608 if (msr >= MSR_IA32_MC0_CTL &&
1609 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1610 u32 offset = msr - MSR_IA32_MC0_CTL;
1611 /* only 0 or all 1s can be written to IA32_MCi_CTL
1612 * some Linux kernels though clear bit 10 in bank 4 to
1613 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1614 * this to avoid an uncatched #GP in the guest
1616 if ((offset & 0x3) == 0 &&
1617 data != 0 && (data | (1 << 10)) != ~(u64)0)
1619 vcpu->arch.mce_banks[offset] = data;
1627 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1629 struct kvm *kvm = vcpu->kvm;
1630 int lm = is_long_mode(vcpu);
1631 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1632 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1633 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1634 : kvm->arch.xen_hvm_config.blob_size_32;
1635 u32 page_num = data & ~PAGE_MASK;
1636 u64 page_addr = data & PAGE_MASK;
1641 if (page_num >= blob_size)
1644 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1649 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1658 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1660 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1663 static bool kvm_hv_msr_partition_wide(u32 msr)
1667 case HV_X64_MSR_GUEST_OS_ID:
1668 case HV_X64_MSR_HYPERCALL:
1676 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1678 struct kvm *kvm = vcpu->kvm;
1681 case HV_X64_MSR_GUEST_OS_ID:
1682 kvm->arch.hv_guest_os_id = data;
1683 /* setting guest os id to zero disables hypercall page */
1684 if (!kvm->arch.hv_guest_os_id)
1685 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1687 case HV_X64_MSR_HYPERCALL: {
1692 /* if guest os id is not set hypercall should remain disabled */
1693 if (!kvm->arch.hv_guest_os_id)
1695 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1696 kvm->arch.hv_hypercall = data;
1699 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1700 addr = gfn_to_hva(kvm, gfn);
1701 if (kvm_is_error_hva(addr))
1703 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1704 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1705 if (__copy_to_user((void __user *)addr, instructions, 4))
1707 kvm->arch.hv_hypercall = data;
1711 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1712 "data 0x%llx\n", msr, data);
1718 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1721 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1724 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1725 vcpu->arch.hv_vapic = data;
1728 addr = gfn_to_hva(vcpu->kvm, data >>
1729 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1730 if (kvm_is_error_hva(addr))
1732 if (__clear_user((void __user *)addr, PAGE_SIZE))
1734 vcpu->arch.hv_vapic = data;
1737 case HV_X64_MSR_EOI:
1738 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1739 case HV_X64_MSR_ICR:
1740 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1741 case HV_X64_MSR_TPR:
1742 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1744 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1745 "data 0x%llx\n", msr, data);
1752 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1754 gpa_t gpa = data & ~0x3f;
1756 /* Bits 2:5 are reserved, Should be zero */
1760 vcpu->arch.apf.msr_val = data;
1762 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1763 kvm_clear_async_pf_completion_queue(vcpu);
1764 kvm_async_pf_hash_reset(vcpu);
1768 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1771 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1772 kvm_async_pf_wakeup_all(vcpu);
1776 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1778 if (vcpu->arch.time_page) {
1779 kvm_release_page_dirty(vcpu->arch.time_page);
1780 vcpu->arch.time_page = NULL;
1784 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1788 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1791 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1792 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1793 vcpu->arch.st.accum_steal = delta;
1796 static void record_steal_time(struct kvm_vcpu *vcpu)
1798 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1801 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1802 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1805 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1806 vcpu->arch.st.steal.version += 2;
1807 vcpu->arch.st.accum_steal = 0;
1809 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1810 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1813 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1819 return set_efer(vcpu, data);
1821 data &= ~(u64)0x40; /* ignore flush filter disable */
1822 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1823 data &= ~(u64)0x8; /* ignore TLB cache disable */
1825 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1830 case MSR_FAM10H_MMIO_CONF_BASE:
1832 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1837 case MSR_AMD64_NB_CFG:
1839 case MSR_IA32_DEBUGCTLMSR:
1841 /* We support the non-activated case already */
1843 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1844 /* Values other than LBR and BTF are vendor-specific,
1845 thus reserved and should throw a #GP */
1848 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1851 case MSR_IA32_UCODE_REV:
1852 case MSR_IA32_UCODE_WRITE:
1853 case MSR_VM_HSAVE_PA:
1854 case MSR_AMD64_PATCH_LOADER:
1856 case 0x200 ... 0x2ff:
1857 return set_msr_mtrr(vcpu, msr, data);
1858 case MSR_IA32_APICBASE:
1859 kvm_set_apic_base(vcpu, data);
1861 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1862 return kvm_x2apic_msr_write(vcpu, msr, data);
1863 case MSR_IA32_TSCDEADLINE:
1864 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1866 case MSR_IA32_MISC_ENABLE:
1867 vcpu->arch.ia32_misc_enable_msr = data;
1869 case MSR_KVM_WALL_CLOCK_NEW:
1870 case MSR_KVM_WALL_CLOCK:
1871 vcpu->kvm->arch.wall_clock = data;
1872 kvm_write_wall_clock(vcpu->kvm, data);
1874 case MSR_KVM_SYSTEM_TIME_NEW:
1875 case MSR_KVM_SYSTEM_TIME: {
1876 kvmclock_reset(vcpu);
1878 vcpu->arch.time = data;
1879 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1881 /* we verify if the enable bit is set... */
1885 /* ...but clean it before doing the actual write */
1886 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1888 vcpu->arch.time_page =
1889 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1891 if (is_error_page(vcpu->arch.time_page))
1892 vcpu->arch.time_page = NULL;
1896 case MSR_KVM_ASYNC_PF_EN:
1897 if (kvm_pv_enable_async_pf(vcpu, data))
1900 case MSR_KVM_STEAL_TIME:
1902 if (unlikely(!sched_info_on()))
1905 if (data & KVM_STEAL_RESERVED_MASK)
1908 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
1909 data & KVM_STEAL_VALID_BITS))
1912 vcpu->arch.st.msr_val = data;
1914 if (!(data & KVM_MSR_ENABLED))
1917 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1920 accumulate_steal_time(vcpu);
1923 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
1926 case MSR_KVM_PV_EOI_EN:
1927 if (kvm_lapic_enable_pv_eoi(vcpu, data))
1931 case MSR_IA32_MCG_CTL:
1932 case MSR_IA32_MCG_STATUS:
1933 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1934 return set_msr_mce(vcpu, msr, data);
1936 /* Performance counters are not protected by a CPUID bit,
1937 * so we should check all of them in the generic path for the sake of
1938 * cross vendor migration.
1939 * Writing a zero into the event select MSRs disables them,
1940 * which we perfectly emulate ;-). Any other value should be at least
1941 * reported, some guests depend on them.
1943 case MSR_K7_EVNTSEL0:
1944 case MSR_K7_EVNTSEL1:
1945 case MSR_K7_EVNTSEL2:
1946 case MSR_K7_EVNTSEL3:
1948 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1949 "0x%x data 0x%llx\n", msr, data);
1951 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1952 * so we ignore writes to make it happy.
1954 case MSR_K7_PERFCTR0:
1955 case MSR_K7_PERFCTR1:
1956 case MSR_K7_PERFCTR2:
1957 case MSR_K7_PERFCTR3:
1958 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1959 "0x%x data 0x%llx\n", msr, data);
1961 case MSR_P6_PERFCTR0:
1962 case MSR_P6_PERFCTR1:
1964 case MSR_P6_EVNTSEL0:
1965 case MSR_P6_EVNTSEL1:
1966 if (kvm_pmu_msr(vcpu, msr))
1967 return kvm_pmu_set_msr(vcpu, msr, data);
1969 if (pr || data != 0)
1970 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
1971 "0x%x data 0x%llx\n", msr, data);
1973 case MSR_K7_CLK_CTL:
1975 * Ignore all writes to this no longer documented MSR.
1976 * Writes are only relevant for old K7 processors,
1977 * all pre-dating SVM, but a recommended workaround from
1978 * AMD for these chips. It is possible to specify the
1979 * affected processor models on the command line, hence
1980 * the need to ignore the workaround.
1983 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1984 if (kvm_hv_msr_partition_wide(msr)) {
1986 mutex_lock(&vcpu->kvm->lock);
1987 r = set_msr_hyperv_pw(vcpu, msr, data);
1988 mutex_unlock(&vcpu->kvm->lock);
1991 return set_msr_hyperv(vcpu, msr, data);
1993 case MSR_IA32_BBL_CR_CTL3:
1994 /* Drop writes to this legacy MSR -- see rdmsr
1995 * counterpart for further detail.
1997 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1999 case MSR_AMD64_OSVW_ID_LENGTH:
2000 if (!guest_cpuid_has_osvw(vcpu))
2002 vcpu->arch.osvw.length = data;
2004 case MSR_AMD64_OSVW_STATUS:
2005 if (!guest_cpuid_has_osvw(vcpu))
2007 vcpu->arch.osvw.status = data;
2010 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2011 return xen_hvm_config(vcpu, data);
2012 if (kvm_pmu_msr(vcpu, msr))
2013 return kvm_pmu_set_msr(vcpu, msr, data);
2015 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2019 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2026 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2030 * Reads an msr value (of 'msr_index') into 'pdata'.
2031 * Returns 0 on success, non-0 otherwise.
2032 * Assumes vcpu_load() was already called.
2034 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2036 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
2039 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2041 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2043 if (!msr_mtrr_valid(msr))
2046 if (msr == MSR_MTRRdefType)
2047 *pdata = vcpu->arch.mtrr_state.def_type +
2048 (vcpu->arch.mtrr_state.enabled << 10);
2049 else if (msr == MSR_MTRRfix64K_00000)
2051 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2052 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2053 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2054 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2055 else if (msr == MSR_IA32_CR_PAT)
2056 *pdata = vcpu->arch.pat;
2057 else { /* Variable MTRRs */
2058 int idx, is_mtrr_mask;
2061 idx = (msr - 0x200) / 2;
2062 is_mtrr_mask = msr - 0x200 - 2 * idx;
2065 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2068 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2075 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2078 u64 mcg_cap = vcpu->arch.mcg_cap;
2079 unsigned bank_num = mcg_cap & 0xff;
2082 case MSR_IA32_P5_MC_ADDR:
2083 case MSR_IA32_P5_MC_TYPE:
2086 case MSR_IA32_MCG_CAP:
2087 data = vcpu->arch.mcg_cap;
2089 case MSR_IA32_MCG_CTL:
2090 if (!(mcg_cap & MCG_CTL_P))
2092 data = vcpu->arch.mcg_ctl;
2094 case MSR_IA32_MCG_STATUS:
2095 data = vcpu->arch.mcg_status;
2098 if (msr >= MSR_IA32_MC0_CTL &&
2099 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
2100 u32 offset = msr - MSR_IA32_MC0_CTL;
2101 data = vcpu->arch.mce_banks[offset];
2110 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2113 struct kvm *kvm = vcpu->kvm;
2116 case HV_X64_MSR_GUEST_OS_ID:
2117 data = kvm->arch.hv_guest_os_id;
2119 case HV_X64_MSR_HYPERCALL:
2120 data = kvm->arch.hv_hypercall;
2123 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2131 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2136 case HV_X64_MSR_VP_INDEX: {
2139 kvm_for_each_vcpu(r, v, vcpu->kvm)
2144 case HV_X64_MSR_EOI:
2145 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2146 case HV_X64_MSR_ICR:
2147 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2148 case HV_X64_MSR_TPR:
2149 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
2150 case HV_X64_MSR_APIC_ASSIST_PAGE:
2151 data = vcpu->arch.hv_vapic;
2154 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2161 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2166 case MSR_IA32_PLATFORM_ID:
2167 case MSR_IA32_EBL_CR_POWERON:
2168 case MSR_IA32_DEBUGCTLMSR:
2169 case MSR_IA32_LASTBRANCHFROMIP:
2170 case MSR_IA32_LASTBRANCHTOIP:
2171 case MSR_IA32_LASTINTFROMIP:
2172 case MSR_IA32_LASTINTTOIP:
2175 case MSR_VM_HSAVE_PA:
2176 case MSR_K7_EVNTSEL0:
2177 case MSR_K7_PERFCTR0:
2178 case MSR_K8_INT_PENDING_MSG:
2179 case MSR_AMD64_NB_CFG:
2180 case MSR_FAM10H_MMIO_CONF_BASE:
2183 case MSR_P6_PERFCTR0:
2184 case MSR_P6_PERFCTR1:
2185 case MSR_P6_EVNTSEL0:
2186 case MSR_P6_EVNTSEL1:
2187 if (kvm_pmu_msr(vcpu, msr))
2188 return kvm_pmu_get_msr(vcpu, msr, pdata);
2191 case MSR_IA32_UCODE_REV:
2192 data = 0x100000000ULL;
2195 data = 0x500 | KVM_NR_VAR_MTRR;
2197 case 0x200 ... 0x2ff:
2198 return get_msr_mtrr(vcpu, msr, pdata);
2199 case 0xcd: /* fsb frequency */
2203 * MSR_EBC_FREQUENCY_ID
2204 * Conservative value valid for even the basic CPU models.
2205 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2206 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2207 * and 266MHz for model 3, or 4. Set Core Clock
2208 * Frequency to System Bus Frequency Ratio to 1 (bits
2209 * 31:24) even though these are only valid for CPU
2210 * models > 2, however guests may end up dividing or
2211 * multiplying by zero otherwise.
2213 case MSR_EBC_FREQUENCY_ID:
2216 case MSR_IA32_APICBASE:
2217 data = kvm_get_apic_base(vcpu);
2219 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2220 return kvm_x2apic_msr_read(vcpu, msr, pdata);
2222 case MSR_IA32_TSCDEADLINE:
2223 data = kvm_get_lapic_tscdeadline_msr(vcpu);
2225 case MSR_IA32_MISC_ENABLE:
2226 data = vcpu->arch.ia32_misc_enable_msr;
2228 case MSR_IA32_PERF_STATUS:
2229 /* TSC increment by tick */
2231 /* CPU multiplier */
2232 data |= (((uint64_t)4ULL) << 40);
2235 data = vcpu->arch.efer;
2237 case MSR_KVM_WALL_CLOCK:
2238 case MSR_KVM_WALL_CLOCK_NEW:
2239 data = vcpu->kvm->arch.wall_clock;
2241 case MSR_KVM_SYSTEM_TIME:
2242 case MSR_KVM_SYSTEM_TIME_NEW:
2243 data = vcpu->arch.time;
2245 case MSR_KVM_ASYNC_PF_EN:
2246 data = vcpu->arch.apf.msr_val;
2248 case MSR_KVM_STEAL_TIME:
2249 data = vcpu->arch.st.msr_val;
2251 case MSR_KVM_PV_EOI_EN:
2252 data = vcpu->arch.pv_eoi.msr_val;
2254 case MSR_IA32_P5_MC_ADDR:
2255 case MSR_IA32_P5_MC_TYPE:
2256 case MSR_IA32_MCG_CAP:
2257 case MSR_IA32_MCG_CTL:
2258 case MSR_IA32_MCG_STATUS:
2259 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2260 return get_msr_mce(vcpu, msr, pdata);
2261 case MSR_K7_CLK_CTL:
2263 * Provide expected ramp-up count for K7. All other
2264 * are set to zero, indicating minimum divisors for
2267 * This prevents guest kernels on AMD host with CPU
2268 * type 6, model 8 and higher from exploding due to
2269 * the rdmsr failing.
2273 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2274 if (kvm_hv_msr_partition_wide(msr)) {
2276 mutex_lock(&vcpu->kvm->lock);
2277 r = get_msr_hyperv_pw(vcpu, msr, pdata);
2278 mutex_unlock(&vcpu->kvm->lock);
2281 return get_msr_hyperv(vcpu, msr, pdata);
2283 case MSR_IA32_BBL_CR_CTL3:
2284 /* This legacy MSR exists but isn't fully documented in current
2285 * silicon. It is however accessed by winxp in very narrow
2286 * scenarios where it sets bit #19, itself documented as
2287 * a "reserved" bit. Best effort attempt to source coherent
2288 * read data here should the balance of the register be
2289 * interpreted by the guest:
2291 * L2 cache control register 3: 64GB range, 256KB size,
2292 * enabled, latency 0x1, configured
2296 case MSR_AMD64_OSVW_ID_LENGTH:
2297 if (!guest_cpuid_has_osvw(vcpu))
2299 data = vcpu->arch.osvw.length;
2301 case MSR_AMD64_OSVW_STATUS:
2302 if (!guest_cpuid_has_osvw(vcpu))
2304 data = vcpu->arch.osvw.status;
2307 if (kvm_pmu_msr(vcpu, msr))
2308 return kvm_pmu_get_msr(vcpu, msr, pdata);
2310 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2313 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2321 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2324 * Read or write a bunch of msrs. All parameters are kernel addresses.
2326 * @return number of msrs set successfully.
2328 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2329 struct kvm_msr_entry *entries,
2330 int (*do_msr)(struct kvm_vcpu *vcpu,
2331 unsigned index, u64 *data))
2335 idx = srcu_read_lock(&vcpu->kvm->srcu);
2336 for (i = 0; i < msrs->nmsrs; ++i)
2337 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2339 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2345 * Read or write a bunch of msrs. Parameters are user addresses.
2347 * @return number of msrs set successfully.
2349 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2350 int (*do_msr)(struct kvm_vcpu *vcpu,
2351 unsigned index, u64 *data),
2354 struct kvm_msrs msrs;
2355 struct kvm_msr_entry *entries;
2360 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2364 if (msrs.nmsrs >= MAX_IO_MSRS)
2367 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2368 entries = memdup_user(user_msrs->entries, size);
2369 if (IS_ERR(entries)) {
2370 r = PTR_ERR(entries);
2374 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2379 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2390 int kvm_dev_ioctl_check_extension(long ext)
2395 case KVM_CAP_IRQCHIP:
2397 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2398 case KVM_CAP_SET_TSS_ADDR:
2399 case KVM_CAP_EXT_CPUID:
2400 case KVM_CAP_CLOCKSOURCE:
2402 case KVM_CAP_NOP_IO_DELAY:
2403 case KVM_CAP_MP_STATE:
2404 case KVM_CAP_SYNC_MMU:
2405 case KVM_CAP_USER_NMI:
2406 case KVM_CAP_REINJECT_CONTROL:
2407 case KVM_CAP_IRQ_INJECT_STATUS:
2408 case KVM_CAP_ASSIGN_DEV_IRQ:
2410 case KVM_CAP_IOEVENTFD:
2412 case KVM_CAP_PIT_STATE2:
2413 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2414 case KVM_CAP_XEN_HVM:
2415 case KVM_CAP_ADJUST_CLOCK:
2416 case KVM_CAP_VCPU_EVENTS:
2417 case KVM_CAP_HYPERV:
2418 case KVM_CAP_HYPERV_VAPIC:
2419 case KVM_CAP_HYPERV_SPIN:
2420 case KVM_CAP_PCI_SEGMENT:
2421 case KVM_CAP_DEBUGREGS:
2422 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2424 case KVM_CAP_ASYNC_PF:
2425 case KVM_CAP_GET_TSC_KHZ:
2426 case KVM_CAP_PCI_2_3:
2427 case KVM_CAP_KVMCLOCK_CTRL:
2428 case KVM_CAP_READONLY_MEM:
2429 case KVM_CAP_IRQFD_RESAMPLE:
2432 case KVM_CAP_COALESCED_MMIO:
2433 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2436 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2438 case KVM_CAP_NR_VCPUS:
2439 r = KVM_SOFT_MAX_VCPUS;
2441 case KVM_CAP_MAX_VCPUS:
2444 case KVM_CAP_NR_MEMSLOTS:
2445 r = KVM_MEMORY_SLOTS;
2447 case KVM_CAP_PV_MMU: /* obsolete */
2451 r = iommu_present(&pci_bus_type);
2454 r = KVM_MAX_MCE_BANKS;
2459 case KVM_CAP_TSC_CONTROL:
2460 r = kvm_has_tsc_control;
2462 case KVM_CAP_TSC_DEADLINE_TIMER:
2463 r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2473 long kvm_arch_dev_ioctl(struct file *filp,
2474 unsigned int ioctl, unsigned long arg)
2476 void __user *argp = (void __user *)arg;
2480 case KVM_GET_MSR_INDEX_LIST: {
2481 struct kvm_msr_list __user *user_msr_list = argp;
2482 struct kvm_msr_list msr_list;
2486 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2489 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2490 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2493 if (n < msr_list.nmsrs)
2496 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2497 num_msrs_to_save * sizeof(u32)))
2499 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2501 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2506 case KVM_GET_SUPPORTED_CPUID: {
2507 struct kvm_cpuid2 __user *cpuid_arg = argp;
2508 struct kvm_cpuid2 cpuid;
2511 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2513 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2514 cpuid_arg->entries);
2519 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2524 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2527 mce_cap = KVM_MCE_CAP_SUPPORTED;
2529 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2541 static void wbinvd_ipi(void *garbage)
2546 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2548 return vcpu->kvm->arch.iommu_domain &&
2549 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2552 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2554 /* Address WBINVD may be executed by guest */
2555 if (need_emulate_wbinvd(vcpu)) {
2556 if (kvm_x86_ops->has_wbinvd_exit())
2557 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2558 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2559 smp_call_function_single(vcpu->cpu,
2560 wbinvd_ipi, NULL, 1);
2563 kvm_x86_ops->vcpu_load(vcpu, cpu);
2565 /* Apply any externally detected TSC adjustments (due to suspend) */
2566 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2567 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2568 vcpu->arch.tsc_offset_adjustment = 0;
2569 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
2572 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2573 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2574 native_read_tsc() - vcpu->arch.last_host_tsc;
2576 mark_tsc_unstable("KVM discovered backwards TSC");
2577 if (check_tsc_unstable()) {
2578 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2579 vcpu->arch.last_guest_tsc);
2580 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2581 vcpu->arch.tsc_catchup = 1;
2583 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2584 if (vcpu->cpu != cpu)
2585 kvm_migrate_timers(vcpu);
2589 accumulate_steal_time(vcpu);
2590 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2593 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2595 kvm_x86_ops->vcpu_put(vcpu);
2596 kvm_put_guest_fpu(vcpu);
2597 vcpu->arch.last_host_tsc = native_read_tsc();
2600 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2601 struct kvm_lapic_state *s)
2603 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2608 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2609 struct kvm_lapic_state *s)
2611 kvm_apic_post_state_restore(vcpu, s);
2612 update_cr8_intercept(vcpu);
2617 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2618 struct kvm_interrupt *irq)
2620 if (irq->irq < 0 || irq->irq >= KVM_NR_INTERRUPTS)
2622 if (irqchip_in_kernel(vcpu->kvm))
2625 kvm_queue_interrupt(vcpu, irq->irq, false);
2626 kvm_make_request(KVM_REQ_EVENT, vcpu);
2631 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2633 kvm_inject_nmi(vcpu);
2638 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2639 struct kvm_tpr_access_ctl *tac)
2643 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2647 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2651 unsigned bank_num = mcg_cap & 0xff, bank;
2654 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2656 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2659 vcpu->arch.mcg_cap = mcg_cap;
2660 /* Init IA32_MCG_CTL to all 1s */
2661 if (mcg_cap & MCG_CTL_P)
2662 vcpu->arch.mcg_ctl = ~(u64)0;
2663 /* Init IA32_MCi_CTL to all 1s */
2664 for (bank = 0; bank < bank_num; bank++)
2665 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2670 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2671 struct kvm_x86_mce *mce)
2673 u64 mcg_cap = vcpu->arch.mcg_cap;
2674 unsigned bank_num = mcg_cap & 0xff;
2675 u64 *banks = vcpu->arch.mce_banks;
2677 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2680 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2681 * reporting is disabled
2683 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2684 vcpu->arch.mcg_ctl != ~(u64)0)
2686 banks += 4 * mce->bank;
2688 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2689 * reporting is disabled for the bank
2691 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2693 if (mce->status & MCI_STATUS_UC) {
2694 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2695 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2696 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2699 if (banks[1] & MCI_STATUS_VAL)
2700 mce->status |= MCI_STATUS_OVER;
2701 banks[2] = mce->addr;
2702 banks[3] = mce->misc;
2703 vcpu->arch.mcg_status = mce->mcg_status;
2704 banks[1] = mce->status;
2705 kvm_queue_exception(vcpu, MC_VECTOR);
2706 } else if (!(banks[1] & MCI_STATUS_VAL)
2707 || !(banks[1] & MCI_STATUS_UC)) {
2708 if (banks[1] & MCI_STATUS_VAL)
2709 mce->status |= MCI_STATUS_OVER;
2710 banks[2] = mce->addr;
2711 banks[3] = mce->misc;
2712 banks[1] = mce->status;
2714 banks[1] |= MCI_STATUS_OVER;
2718 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2719 struct kvm_vcpu_events *events)
2722 events->exception.injected =
2723 vcpu->arch.exception.pending &&
2724 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2725 events->exception.nr = vcpu->arch.exception.nr;
2726 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2727 events->exception.pad = 0;
2728 events->exception.error_code = vcpu->arch.exception.error_code;
2730 events->interrupt.injected =
2731 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2732 events->interrupt.nr = vcpu->arch.interrupt.nr;
2733 events->interrupt.soft = 0;
2734 events->interrupt.shadow =
2735 kvm_x86_ops->get_interrupt_shadow(vcpu,
2736 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2738 events->nmi.injected = vcpu->arch.nmi_injected;
2739 events->nmi.pending = vcpu->arch.nmi_pending != 0;
2740 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2741 events->nmi.pad = 0;
2743 events->sipi_vector = vcpu->arch.sipi_vector;
2745 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2746 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2747 | KVM_VCPUEVENT_VALID_SHADOW);
2748 memset(&events->reserved, 0, sizeof(events->reserved));
2751 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2752 struct kvm_vcpu_events *events)
2754 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2755 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2756 | KVM_VCPUEVENT_VALID_SHADOW))
2760 vcpu->arch.exception.pending = events->exception.injected;
2761 vcpu->arch.exception.nr = events->exception.nr;
2762 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2763 vcpu->arch.exception.error_code = events->exception.error_code;
2765 vcpu->arch.interrupt.pending = events->interrupt.injected;
2766 vcpu->arch.interrupt.nr = events->interrupt.nr;
2767 vcpu->arch.interrupt.soft = events->interrupt.soft;
2768 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2769 kvm_x86_ops->set_interrupt_shadow(vcpu,
2770 events->interrupt.shadow);
2772 vcpu->arch.nmi_injected = events->nmi.injected;
2773 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2774 vcpu->arch.nmi_pending = events->nmi.pending;
2775 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2777 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2778 vcpu->arch.sipi_vector = events->sipi_vector;
2780 kvm_make_request(KVM_REQ_EVENT, vcpu);
2785 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2786 struct kvm_debugregs *dbgregs)
2788 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2789 dbgregs->dr6 = vcpu->arch.dr6;
2790 dbgregs->dr7 = vcpu->arch.dr7;
2792 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2795 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2796 struct kvm_debugregs *dbgregs)
2801 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2802 vcpu->arch.dr6 = dbgregs->dr6;
2803 vcpu->arch.dr7 = dbgregs->dr7;
2808 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2809 struct kvm_xsave *guest_xsave)
2812 memcpy(guest_xsave->region,
2813 &vcpu->arch.guest_fpu.state->xsave,
2816 memcpy(guest_xsave->region,
2817 &vcpu->arch.guest_fpu.state->fxsave,
2818 sizeof(struct i387_fxsave_struct));
2819 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2824 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2825 struct kvm_xsave *guest_xsave)
2828 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2831 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2832 guest_xsave->region, xstate_size);
2834 if (xstate_bv & ~XSTATE_FPSSE)
2836 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2837 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2842 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2843 struct kvm_xcrs *guest_xcrs)
2845 if (!cpu_has_xsave) {
2846 guest_xcrs->nr_xcrs = 0;
2850 guest_xcrs->nr_xcrs = 1;
2851 guest_xcrs->flags = 0;
2852 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2853 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2856 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2857 struct kvm_xcrs *guest_xcrs)
2864 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2867 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2868 /* Only support XCR0 currently */
2869 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2870 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2871 guest_xcrs->xcrs[0].value);
2880 * kvm_set_guest_paused() indicates to the guest kernel that it has been
2881 * stopped by the hypervisor. This function will be called from the host only.
2882 * EINVAL is returned when the host attempts to set the flag for a guest that
2883 * does not support pv clocks.
2885 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
2887 if (!vcpu->arch.time_page)
2889 vcpu->arch.pvclock_set_guest_stopped_request = true;
2890 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2894 long kvm_arch_vcpu_ioctl(struct file *filp,
2895 unsigned int ioctl, unsigned long arg)
2897 struct kvm_vcpu *vcpu = filp->private_data;
2898 void __user *argp = (void __user *)arg;
2901 struct kvm_lapic_state *lapic;
2902 struct kvm_xsave *xsave;
2903 struct kvm_xcrs *xcrs;
2909 case KVM_GET_LAPIC: {
2911 if (!vcpu->arch.apic)
2913 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2918 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
2922 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
2927 case KVM_SET_LAPIC: {
2928 if (!vcpu->arch.apic)
2930 u.lapic = memdup_user(argp, sizeof(*u.lapic));
2931 if (IS_ERR(u.lapic))
2932 return PTR_ERR(u.lapic);
2934 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
2937 case KVM_INTERRUPT: {
2938 struct kvm_interrupt irq;
2941 if (copy_from_user(&irq, argp, sizeof irq))
2943 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2947 r = kvm_vcpu_ioctl_nmi(vcpu);
2950 case KVM_SET_CPUID: {
2951 struct kvm_cpuid __user *cpuid_arg = argp;
2952 struct kvm_cpuid cpuid;
2955 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2957 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2960 case KVM_SET_CPUID2: {
2961 struct kvm_cpuid2 __user *cpuid_arg = argp;
2962 struct kvm_cpuid2 cpuid;
2965 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2967 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2968 cpuid_arg->entries);
2971 case KVM_GET_CPUID2: {
2972 struct kvm_cpuid2 __user *cpuid_arg = argp;
2973 struct kvm_cpuid2 cpuid;
2976 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2978 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2979 cpuid_arg->entries);
2983 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2989 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2992 r = msr_io(vcpu, argp, do_set_msr, 0);
2994 case KVM_TPR_ACCESS_REPORTING: {
2995 struct kvm_tpr_access_ctl tac;
2998 if (copy_from_user(&tac, argp, sizeof tac))
3000 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3004 if (copy_to_user(argp, &tac, sizeof tac))
3009 case KVM_SET_VAPIC_ADDR: {
3010 struct kvm_vapic_addr va;
3013 if (!irqchip_in_kernel(vcpu->kvm))
3016 if (copy_from_user(&va, argp, sizeof va))
3019 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3022 case KVM_X86_SETUP_MCE: {
3026 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3028 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3031 case KVM_X86_SET_MCE: {
3032 struct kvm_x86_mce mce;
3035 if (copy_from_user(&mce, argp, sizeof mce))
3037 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3040 case KVM_GET_VCPU_EVENTS: {
3041 struct kvm_vcpu_events events;
3043 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3046 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3051 case KVM_SET_VCPU_EVENTS: {
3052 struct kvm_vcpu_events events;
3055 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3058 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3061 case KVM_GET_DEBUGREGS: {
3062 struct kvm_debugregs dbgregs;
3064 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3067 if (copy_to_user(argp, &dbgregs,
3068 sizeof(struct kvm_debugregs)))
3073 case KVM_SET_DEBUGREGS: {
3074 struct kvm_debugregs dbgregs;
3077 if (copy_from_user(&dbgregs, argp,
3078 sizeof(struct kvm_debugregs)))
3081 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3084 case KVM_GET_XSAVE: {
3085 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3090 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3093 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3098 case KVM_SET_XSAVE: {
3099 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3100 if (IS_ERR(u.xsave))
3101 return PTR_ERR(u.xsave);
3103 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3106 case KVM_GET_XCRS: {
3107 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3112 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3115 if (copy_to_user(argp, u.xcrs,
3116 sizeof(struct kvm_xcrs)))
3121 case KVM_SET_XCRS: {
3122 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3124 return PTR_ERR(u.xcrs);
3126 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3129 case KVM_SET_TSC_KHZ: {
3133 user_tsc_khz = (u32)arg;
3135 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3138 if (user_tsc_khz == 0)
3139 user_tsc_khz = tsc_khz;
3141 kvm_set_tsc_khz(vcpu, user_tsc_khz);
3146 case KVM_GET_TSC_KHZ: {
3147 r = vcpu->arch.virtual_tsc_khz;
3150 case KVM_KVMCLOCK_CTRL: {
3151 r = kvm_set_guest_paused(vcpu);
3162 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3164 return VM_FAULT_SIGBUS;
3167 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3171 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3173 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3177 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3180 kvm->arch.ept_identity_map_addr = ident_addr;
3184 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3185 u32 kvm_nr_mmu_pages)
3187 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3190 mutex_lock(&kvm->slots_lock);
3191 spin_lock(&kvm->mmu_lock);
3193 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3194 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3196 spin_unlock(&kvm->mmu_lock);
3197 mutex_unlock(&kvm->slots_lock);
3201 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3203 return kvm->arch.n_max_mmu_pages;
3206 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3211 switch (chip->chip_id) {
3212 case KVM_IRQCHIP_PIC_MASTER:
3213 memcpy(&chip->chip.pic,
3214 &pic_irqchip(kvm)->pics[0],
3215 sizeof(struct kvm_pic_state));
3217 case KVM_IRQCHIP_PIC_SLAVE:
3218 memcpy(&chip->chip.pic,
3219 &pic_irqchip(kvm)->pics[1],
3220 sizeof(struct kvm_pic_state));
3222 case KVM_IRQCHIP_IOAPIC:
3223 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3232 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3237 switch (chip->chip_id) {
3238 case KVM_IRQCHIP_PIC_MASTER:
3239 spin_lock(&pic_irqchip(kvm)->lock);
3240 memcpy(&pic_irqchip(kvm)->pics[0],
3242 sizeof(struct kvm_pic_state));
3243 spin_unlock(&pic_irqchip(kvm)->lock);
3245 case KVM_IRQCHIP_PIC_SLAVE:
3246 spin_lock(&pic_irqchip(kvm)->lock);
3247 memcpy(&pic_irqchip(kvm)->pics[1],
3249 sizeof(struct kvm_pic_state));
3250 spin_unlock(&pic_irqchip(kvm)->lock);
3252 case KVM_IRQCHIP_IOAPIC:
3253 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3259 kvm_pic_update_irq(pic_irqchip(kvm));
3263 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3267 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3268 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3269 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3273 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3277 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3278 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3279 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3280 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3284 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3288 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3289 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3290 sizeof(ps->channels));
3291 ps->flags = kvm->arch.vpit->pit_state.flags;
3292 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3293 memset(&ps->reserved, 0, sizeof(ps->reserved));
3297 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3299 int r = 0, start = 0;
3300 u32 prev_legacy, cur_legacy;
3301 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3302 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3303 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3304 if (!prev_legacy && cur_legacy)
3306 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3307 sizeof(kvm->arch.vpit->pit_state.channels));
3308 kvm->arch.vpit->pit_state.flags = ps->flags;
3309 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3310 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3314 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3315 struct kvm_reinject_control *control)
3317 if (!kvm->arch.vpit)
3319 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3320 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3321 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3326 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3327 * @kvm: kvm instance
3328 * @log: slot id and address to which we copy the log
3330 * We need to keep it in mind that VCPU threads can write to the bitmap
3331 * concurrently. So, to avoid losing data, we keep the following order for
3334 * 1. Take a snapshot of the bit and clear it if needed.
3335 * 2. Write protect the corresponding page.
3336 * 3. Flush TLB's if needed.
3337 * 4. Copy the snapshot to the userspace.
3339 * Between 2 and 3, the guest may write to the page using the remaining TLB
3340 * entry. This is not a problem because the page will be reported dirty at
3341 * step 4 using the snapshot taken before and step 3 ensures that successive
3342 * writes will be logged for the next call.
3344 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3347 struct kvm_memory_slot *memslot;
3349 unsigned long *dirty_bitmap;
3350 unsigned long *dirty_bitmap_buffer;
3351 bool is_dirty = false;
3353 mutex_lock(&kvm->slots_lock);
3356 if (log->slot >= KVM_MEMORY_SLOTS)
3359 memslot = id_to_memslot(kvm->memslots, log->slot);
3361 dirty_bitmap = memslot->dirty_bitmap;
3366 n = kvm_dirty_bitmap_bytes(memslot);
3368 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
3369 memset(dirty_bitmap_buffer, 0, n);
3371 spin_lock(&kvm->mmu_lock);
3373 for (i = 0; i < n / sizeof(long); i++) {
3377 if (!dirty_bitmap[i])
3382 mask = xchg(&dirty_bitmap[i], 0);
3383 dirty_bitmap_buffer[i] = mask;
3385 offset = i * BITS_PER_LONG;
3386 kvm_mmu_write_protect_pt_masked(kvm, memslot, offset, mask);
3389 kvm_flush_remote_tlbs(kvm);
3391 spin_unlock(&kvm->mmu_lock);
3394 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
3399 mutex_unlock(&kvm->slots_lock);
3403 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event)
3405 if (!irqchip_in_kernel(kvm))
3408 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3409 irq_event->irq, irq_event->level);
3413 long kvm_arch_vm_ioctl(struct file *filp,
3414 unsigned int ioctl, unsigned long arg)
3416 struct kvm *kvm = filp->private_data;
3417 void __user *argp = (void __user *)arg;
3420 * This union makes it completely explicit to gcc-3.x
3421 * that these two variables' stack usage should be
3422 * combined, not added together.
3425 struct kvm_pit_state ps;
3426 struct kvm_pit_state2 ps2;
3427 struct kvm_pit_config pit_config;
3431 case KVM_SET_TSS_ADDR:
3432 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3434 case KVM_SET_IDENTITY_MAP_ADDR: {
3438 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3440 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3443 case KVM_SET_NR_MMU_PAGES:
3444 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3446 case KVM_GET_NR_MMU_PAGES:
3447 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3449 case KVM_CREATE_IRQCHIP: {
3450 struct kvm_pic *vpic;
3452 mutex_lock(&kvm->lock);
3455 goto create_irqchip_unlock;
3457 if (atomic_read(&kvm->online_vcpus))
3458 goto create_irqchip_unlock;
3460 vpic = kvm_create_pic(kvm);
3462 r = kvm_ioapic_init(kvm);
3464 mutex_lock(&kvm->slots_lock);
3465 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3467 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3469 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3471 mutex_unlock(&kvm->slots_lock);
3473 goto create_irqchip_unlock;
3476 goto create_irqchip_unlock;
3478 kvm->arch.vpic = vpic;
3480 r = kvm_setup_default_irq_routing(kvm);
3482 mutex_lock(&kvm->slots_lock);
3483 mutex_lock(&kvm->irq_lock);
3484 kvm_ioapic_destroy(kvm);
3485 kvm_destroy_pic(kvm);
3486 mutex_unlock(&kvm->irq_lock);
3487 mutex_unlock(&kvm->slots_lock);
3489 create_irqchip_unlock:
3490 mutex_unlock(&kvm->lock);
3493 case KVM_CREATE_PIT:
3494 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3496 case KVM_CREATE_PIT2:
3498 if (copy_from_user(&u.pit_config, argp,
3499 sizeof(struct kvm_pit_config)))
3502 mutex_lock(&kvm->slots_lock);
3505 goto create_pit_unlock;
3507 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3511 mutex_unlock(&kvm->slots_lock);
3513 case KVM_GET_IRQCHIP: {
3514 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3515 struct kvm_irqchip *chip;
3517 chip = memdup_user(argp, sizeof(*chip));
3524 if (!irqchip_in_kernel(kvm))
3525 goto get_irqchip_out;
3526 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3528 goto get_irqchip_out;
3530 if (copy_to_user(argp, chip, sizeof *chip))
3531 goto get_irqchip_out;
3537 case KVM_SET_IRQCHIP: {
3538 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3539 struct kvm_irqchip *chip;
3541 chip = memdup_user(argp, sizeof(*chip));
3548 if (!irqchip_in_kernel(kvm))
3549 goto set_irqchip_out;
3550 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3552 goto set_irqchip_out;
3560 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3563 if (!kvm->arch.vpit)
3565 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3569 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3576 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3579 if (!kvm->arch.vpit)
3581 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3584 case KVM_GET_PIT2: {
3586 if (!kvm->arch.vpit)
3588 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3592 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3597 case KVM_SET_PIT2: {
3599 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3602 if (!kvm->arch.vpit)
3604 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3607 case KVM_REINJECT_CONTROL: {
3608 struct kvm_reinject_control control;
3610 if (copy_from_user(&control, argp, sizeof(control)))
3612 r = kvm_vm_ioctl_reinject(kvm, &control);
3615 case KVM_XEN_HVM_CONFIG: {
3617 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3618 sizeof(struct kvm_xen_hvm_config)))
3621 if (kvm->arch.xen_hvm_config.flags)
3626 case KVM_SET_CLOCK: {
3627 struct kvm_clock_data user_ns;
3632 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3640 local_irq_disable();
3641 now_ns = get_kernel_ns();
3642 delta = user_ns.clock - now_ns;
3644 kvm->arch.kvmclock_offset = delta;
3647 case KVM_GET_CLOCK: {
3648 struct kvm_clock_data user_ns;
3651 local_irq_disable();
3652 now_ns = get_kernel_ns();
3653 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3656 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3659 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3672 static void kvm_init_msr_list(void)
3677 /* skip the first msrs in the list. KVM-specific */
3678 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3679 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3682 msrs_to_save[j] = msrs_to_save[i];
3685 num_msrs_to_save = j;
3688 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3696 if (!(vcpu->arch.apic &&
3697 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3698 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3709 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3716 if (!(vcpu->arch.apic &&
3717 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3718 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3720 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3730 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3731 struct kvm_segment *var, int seg)
3733 kvm_x86_ops->set_segment(vcpu, var, seg);
3736 void kvm_get_segment(struct kvm_vcpu *vcpu,
3737 struct kvm_segment *var, int seg)
3739 kvm_x86_ops->get_segment(vcpu, var, seg);
3742 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3745 struct x86_exception exception;
3747 BUG_ON(!mmu_is_nested(vcpu));
3749 /* NPT walks are always user-walks */
3750 access |= PFERR_USER_MASK;
3751 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3756 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3757 struct x86_exception *exception)
3759 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3760 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3763 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3764 struct x86_exception *exception)
3766 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3767 access |= PFERR_FETCH_MASK;
3768 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3771 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3772 struct x86_exception *exception)
3774 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3775 access |= PFERR_WRITE_MASK;
3776 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3779 /* uses this to access any guest's mapped memory without checking CPL */
3780 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3781 struct x86_exception *exception)
3783 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3786 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3787 struct kvm_vcpu *vcpu, u32 access,
3788 struct x86_exception *exception)
3791 int r = X86EMUL_CONTINUE;
3794 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3796 unsigned offset = addr & (PAGE_SIZE-1);
3797 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3800 if (gpa == UNMAPPED_GVA)
3801 return X86EMUL_PROPAGATE_FAULT;
3802 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3804 r = X86EMUL_IO_NEEDED;
3816 /* used for instruction fetching */
3817 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3818 gva_t addr, void *val, unsigned int bytes,
3819 struct x86_exception *exception)
3821 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3822 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3824 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3825 access | PFERR_FETCH_MASK,
3829 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3830 gva_t addr, void *val, unsigned int bytes,
3831 struct x86_exception *exception)
3833 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3834 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3836 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3839 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
3841 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3842 gva_t addr, void *val, unsigned int bytes,
3843 struct x86_exception *exception)
3845 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3846 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3849 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3850 gva_t addr, void *val,
3852 struct x86_exception *exception)
3854 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3856 int r = X86EMUL_CONTINUE;
3859 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3862 unsigned offset = addr & (PAGE_SIZE-1);
3863 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3866 if (gpa == UNMAPPED_GVA)
3867 return X86EMUL_PROPAGATE_FAULT;
3868 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3870 r = X86EMUL_IO_NEEDED;
3881 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
3883 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
3884 gpa_t *gpa, struct x86_exception *exception,
3887 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
3888 | (write ? PFERR_WRITE_MASK : 0);
3890 if (vcpu_match_mmio_gva(vcpu, gva)
3891 && !permission_fault(vcpu->arch.walk_mmu, vcpu->arch.access, access)) {
3892 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
3893 (gva & (PAGE_SIZE - 1));
3894 trace_vcpu_match_mmio(gva, *gpa, write, false);
3898 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3900 if (*gpa == UNMAPPED_GVA)
3903 /* For APIC access vmexit */
3904 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3907 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
3908 trace_vcpu_match_mmio(gva, *gpa, write, true);
3915 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3916 const void *val, int bytes)
3920 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3923 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
3927 struct read_write_emulator_ops {
3928 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
3930 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
3931 void *val, int bytes);
3932 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
3933 int bytes, void *val);
3934 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
3935 void *val, int bytes);
3939 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
3941 if (vcpu->mmio_read_completed) {
3942 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3943 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
3944 vcpu->mmio_read_completed = 0;
3951 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
3952 void *val, int bytes)
3954 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
3957 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
3958 void *val, int bytes)
3960 return emulator_write_phys(vcpu, gpa, val, bytes);
3963 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
3965 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3966 return vcpu_mmio_write(vcpu, gpa, bytes, val);
3969 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
3970 void *val, int bytes)
3972 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3973 return X86EMUL_IO_NEEDED;
3976 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
3977 void *val, int bytes)
3979 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
3981 memcpy(vcpu->run->mmio.data, frag->data, frag->len);
3982 return X86EMUL_CONTINUE;
3985 static const struct read_write_emulator_ops read_emultor = {
3986 .read_write_prepare = read_prepare,
3987 .read_write_emulate = read_emulate,
3988 .read_write_mmio = vcpu_mmio_read,
3989 .read_write_exit_mmio = read_exit_mmio,
3992 static const struct read_write_emulator_ops write_emultor = {
3993 .read_write_emulate = write_emulate,
3994 .read_write_mmio = write_mmio,
3995 .read_write_exit_mmio = write_exit_mmio,
3999 static int emulator_read_write_onepage(unsigned long addr, void *val,
4001 struct x86_exception *exception,
4002 struct kvm_vcpu *vcpu,
4003 const struct read_write_emulator_ops *ops)
4007 bool write = ops->write;
4008 struct kvm_mmio_fragment *frag;
4010 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4013 return X86EMUL_PROPAGATE_FAULT;
4015 /* For APIC access vmexit */
4019 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4020 return X86EMUL_CONTINUE;
4024 * Is this MMIO handled locally?
4026 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4027 if (handled == bytes)
4028 return X86EMUL_CONTINUE;
4035 unsigned now = min(bytes, 8U);
4037 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4046 return X86EMUL_CONTINUE;
4049 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4050 void *val, unsigned int bytes,
4051 struct x86_exception *exception,
4052 const struct read_write_emulator_ops *ops)
4054 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4058 if (ops->read_write_prepare &&
4059 ops->read_write_prepare(vcpu, val, bytes))
4060 return X86EMUL_CONTINUE;
4062 vcpu->mmio_nr_fragments = 0;
4064 /* Crossing a page boundary? */
4065 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4068 now = -addr & ~PAGE_MASK;
4069 rc = emulator_read_write_onepage(addr, val, now, exception,
4072 if (rc != X86EMUL_CONTINUE)
4079 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4081 if (rc != X86EMUL_CONTINUE)
4084 if (!vcpu->mmio_nr_fragments)
4087 gpa = vcpu->mmio_fragments[0].gpa;
4089 vcpu->mmio_needed = 1;
4090 vcpu->mmio_cur_fragment = 0;
4092 vcpu->run->mmio.len = vcpu->mmio_fragments[0].len;
4093 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4094 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4095 vcpu->run->mmio.phys_addr = gpa;
4097 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4100 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4104 struct x86_exception *exception)
4106 return emulator_read_write(ctxt, addr, val, bytes,
4107 exception, &read_emultor);
4110 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4114 struct x86_exception *exception)
4116 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4117 exception, &write_emultor);
4120 #define CMPXCHG_TYPE(t, ptr, old, new) \
4121 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4123 #ifdef CONFIG_X86_64
4124 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4126 # define CMPXCHG64(ptr, old, new) \
4127 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4130 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4135 struct x86_exception *exception)
4137 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4143 /* guests cmpxchg8b have to be emulated atomically */
4144 if (bytes > 8 || (bytes & (bytes - 1)))
4147 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4149 if (gpa == UNMAPPED_GVA ||
4150 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4153 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4156 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4157 if (is_error_page(page))
4160 kaddr = kmap_atomic(page);
4161 kaddr += offset_in_page(gpa);
4164 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4167 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4170 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4173 exchanged = CMPXCHG64(kaddr, old, new);
4178 kunmap_atomic(kaddr);
4179 kvm_release_page_dirty(page);
4182 return X86EMUL_CMPXCHG_FAILED;
4184 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4186 return X86EMUL_CONTINUE;
4189 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4191 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4194 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4196 /* TODO: String I/O for in kernel device */
4199 if (vcpu->arch.pio.in)
4200 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4201 vcpu->arch.pio.size, pd);
4203 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4204 vcpu->arch.pio.port, vcpu->arch.pio.size,
4209 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4210 unsigned short port, void *val,
4211 unsigned int count, bool in)
4213 trace_kvm_pio(!in, port, size, count);
4215 vcpu->arch.pio.port = port;
4216 vcpu->arch.pio.in = in;
4217 vcpu->arch.pio.count = count;
4218 vcpu->arch.pio.size = size;
4220 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4221 vcpu->arch.pio.count = 0;
4225 vcpu->run->exit_reason = KVM_EXIT_IO;
4226 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4227 vcpu->run->io.size = size;
4228 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4229 vcpu->run->io.count = count;
4230 vcpu->run->io.port = port;
4235 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4236 int size, unsigned short port, void *val,
4239 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4242 if (vcpu->arch.pio.count)
4245 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4248 memcpy(val, vcpu->arch.pio_data, size * count);
4249 vcpu->arch.pio.count = 0;
4256 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4257 int size, unsigned short port,
4258 const void *val, unsigned int count)
4260 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4262 memcpy(vcpu->arch.pio_data, val, size * count);
4263 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4266 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4268 return kvm_x86_ops->get_segment_base(vcpu, seg);
4271 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4273 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4276 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4278 if (!need_emulate_wbinvd(vcpu))
4279 return X86EMUL_CONTINUE;
4281 if (kvm_x86_ops->has_wbinvd_exit()) {
4282 int cpu = get_cpu();
4284 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4285 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4286 wbinvd_ipi, NULL, 1);
4288 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4291 return X86EMUL_CONTINUE;
4293 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4295 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4297 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4300 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4302 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4305 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4308 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4311 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4313 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4316 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4318 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4319 unsigned long value;
4323 value = kvm_read_cr0(vcpu);
4326 value = vcpu->arch.cr2;
4329 value = kvm_read_cr3(vcpu);
4332 value = kvm_read_cr4(vcpu);
4335 value = kvm_get_cr8(vcpu);
4338 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4345 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4347 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4352 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4355 vcpu->arch.cr2 = val;
4358 res = kvm_set_cr3(vcpu, val);
4361 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4364 res = kvm_set_cr8(vcpu, val);
4367 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4374 static void emulator_set_rflags(struct x86_emulate_ctxt *ctxt, ulong val)
4376 kvm_set_rflags(emul_to_vcpu(ctxt), val);
4379 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4381 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4384 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4386 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4389 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4391 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4394 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4396 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4399 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4401 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4404 static unsigned long emulator_get_cached_segment_base(
4405 struct x86_emulate_ctxt *ctxt, int seg)
4407 return get_segment_base(emul_to_vcpu(ctxt), seg);
4410 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4411 struct desc_struct *desc, u32 *base3,
4414 struct kvm_segment var;
4416 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4417 *selector = var.selector;
4424 set_desc_limit(desc, var.limit);
4425 set_desc_base(desc, (unsigned long)var.base);
4426 #ifdef CONFIG_X86_64
4428 *base3 = var.base >> 32;
4430 desc->type = var.type;
4432 desc->dpl = var.dpl;
4433 desc->p = var.present;
4434 desc->avl = var.avl;
4442 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4443 struct desc_struct *desc, u32 base3,
4446 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4447 struct kvm_segment var;
4449 var.selector = selector;
4450 var.base = get_desc_base(desc);
4451 #ifdef CONFIG_X86_64
4452 var.base |= ((u64)base3) << 32;
4454 var.limit = get_desc_limit(desc);
4456 var.limit = (var.limit << 12) | 0xfff;
4457 var.type = desc->type;
4458 var.present = desc->p;
4459 var.dpl = desc->dpl;
4464 var.avl = desc->avl;
4465 var.present = desc->p;
4466 var.unusable = !var.present;
4469 kvm_set_segment(vcpu, &var, seg);
4473 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4474 u32 msr_index, u64 *pdata)
4476 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4479 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4480 u32 msr_index, u64 data)
4482 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
4485 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4486 u32 pmc, u64 *pdata)
4488 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4491 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4493 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4496 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4499 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4501 * CR0.TS may reference the host fpu state, not the guest fpu state,
4502 * so it may be clear at this point.
4507 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4512 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4513 struct x86_instruction_info *info,
4514 enum x86_intercept_stage stage)
4516 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4519 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4520 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4522 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4525 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4527 return kvm_register_read(emul_to_vcpu(ctxt), reg);
4530 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4532 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4535 static const struct x86_emulate_ops emulate_ops = {
4536 .read_gpr = emulator_read_gpr,
4537 .write_gpr = emulator_write_gpr,
4538 .read_std = kvm_read_guest_virt_system,
4539 .write_std = kvm_write_guest_virt_system,
4540 .fetch = kvm_fetch_guest_virt,
4541 .read_emulated = emulator_read_emulated,
4542 .write_emulated = emulator_write_emulated,
4543 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4544 .invlpg = emulator_invlpg,
4545 .pio_in_emulated = emulator_pio_in_emulated,
4546 .pio_out_emulated = emulator_pio_out_emulated,
4547 .get_segment = emulator_get_segment,
4548 .set_segment = emulator_set_segment,
4549 .get_cached_segment_base = emulator_get_cached_segment_base,
4550 .get_gdt = emulator_get_gdt,
4551 .get_idt = emulator_get_idt,
4552 .set_gdt = emulator_set_gdt,
4553 .set_idt = emulator_set_idt,
4554 .get_cr = emulator_get_cr,
4555 .set_cr = emulator_set_cr,
4556 .set_rflags = emulator_set_rflags,
4557 .cpl = emulator_get_cpl,
4558 .get_dr = emulator_get_dr,
4559 .set_dr = emulator_set_dr,
4560 .set_msr = emulator_set_msr,
4561 .get_msr = emulator_get_msr,
4562 .read_pmc = emulator_read_pmc,
4563 .halt = emulator_halt,
4564 .wbinvd = emulator_wbinvd,
4565 .fix_hypercall = emulator_fix_hypercall,
4566 .get_fpu = emulator_get_fpu,
4567 .put_fpu = emulator_put_fpu,
4568 .intercept = emulator_intercept,
4569 .get_cpuid = emulator_get_cpuid,
4572 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4574 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4576 * an sti; sti; sequence only disable interrupts for the first
4577 * instruction. So, if the last instruction, be it emulated or
4578 * not, left the system with the INT_STI flag enabled, it
4579 * means that the last instruction is an sti. We should not
4580 * leave the flag on in this case. The same goes for mov ss
4582 if (!(int_shadow & mask))
4583 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4586 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4588 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4589 if (ctxt->exception.vector == PF_VECTOR)
4590 kvm_propagate_fault(vcpu, &ctxt->exception);
4591 else if (ctxt->exception.error_code_valid)
4592 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4593 ctxt->exception.error_code);
4595 kvm_queue_exception(vcpu, ctxt->exception.vector);
4598 static void init_decode_cache(struct x86_emulate_ctxt *ctxt)
4600 memset(&ctxt->twobyte, 0,
4601 (void *)&ctxt->_regs - (void *)&ctxt->twobyte);
4603 ctxt->fetch.start = 0;
4604 ctxt->fetch.end = 0;
4605 ctxt->io_read.pos = 0;
4606 ctxt->io_read.end = 0;
4607 ctxt->mem_read.pos = 0;
4608 ctxt->mem_read.end = 0;
4611 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4613 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4616 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4618 ctxt->eflags = kvm_get_rflags(vcpu);
4619 ctxt->eip = kvm_rip_read(vcpu);
4620 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4621 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4622 cs_l ? X86EMUL_MODE_PROT64 :
4623 cs_db ? X86EMUL_MODE_PROT32 :
4624 X86EMUL_MODE_PROT16;
4625 ctxt->guest_mode = is_guest_mode(vcpu);
4627 init_decode_cache(ctxt);
4628 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4631 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4633 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4636 init_emulate_ctxt(vcpu);
4640 ctxt->_eip = ctxt->eip + inc_eip;
4641 ret = emulate_int_real(ctxt, irq);
4643 if (ret != X86EMUL_CONTINUE)
4644 return EMULATE_FAIL;
4646 ctxt->eip = ctxt->_eip;
4647 kvm_rip_write(vcpu, ctxt->eip);
4648 kvm_set_rflags(vcpu, ctxt->eflags);
4650 if (irq == NMI_VECTOR)
4651 vcpu->arch.nmi_pending = 0;
4653 vcpu->arch.interrupt.pending = false;
4655 return EMULATE_DONE;
4657 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4659 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4661 int r = EMULATE_DONE;
4663 ++vcpu->stat.insn_emulation_fail;
4664 trace_kvm_emulate_insn_failed(vcpu);
4665 if (!is_guest_mode(vcpu)) {
4666 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4667 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4668 vcpu->run->internal.ndata = 0;
4671 kvm_queue_exception(vcpu, UD_VECTOR);
4676 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4685 * if emulation was due to access to shadowed page table
4686 * and it failed try to unshadow page and re-enter the
4687 * guest to let CPU execute the instruction.
4689 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4692 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4694 if (gpa == UNMAPPED_GVA)
4695 return true; /* let cpu generate fault */
4698 * Do not retry the unhandleable instruction if it faults on the
4699 * readonly host memory, otherwise it will goto a infinite loop:
4700 * retry instruction -> write #PF -> emulation fail -> retry
4701 * instruction -> ...
4703 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
4704 if (!is_error_noslot_pfn(pfn)) {
4705 kvm_release_pfn_clean(pfn);
4712 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
4713 unsigned long cr2, int emulation_type)
4715 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4716 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
4718 last_retry_eip = vcpu->arch.last_retry_eip;
4719 last_retry_addr = vcpu->arch.last_retry_addr;
4722 * If the emulation is caused by #PF and it is non-page_table
4723 * writing instruction, it means the VM-EXIT is caused by shadow
4724 * page protected, we can zap the shadow page and retry this
4725 * instruction directly.
4727 * Note: if the guest uses a non-page-table modifying instruction
4728 * on the PDE that points to the instruction, then we will unmap
4729 * the instruction and go to an infinite loop. So, we cache the
4730 * last retried eip and the last fault address, if we meet the eip
4731 * and the address again, we can break out of the potential infinite
4734 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
4736 if (!(emulation_type & EMULTYPE_RETRY))
4739 if (x86_page_table_writing_insn(ctxt))
4742 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
4745 vcpu->arch.last_retry_eip = ctxt->eip;
4746 vcpu->arch.last_retry_addr = cr2;
4748 if (!vcpu->arch.mmu.direct_map)
4749 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4751 kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4756 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
4757 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
4759 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4766 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4767 bool writeback = true;
4769 kvm_clear_exception_queue(vcpu);
4771 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4772 init_emulate_ctxt(vcpu);
4773 ctxt->interruptibility = 0;
4774 ctxt->have_exception = false;
4775 ctxt->perm_ok = false;
4777 ctxt->only_vendor_specific_insn
4778 = emulation_type & EMULTYPE_TRAP_UD;
4780 r = x86_decode_insn(ctxt, insn, insn_len);
4782 trace_kvm_emulate_insn_start(vcpu);
4783 ++vcpu->stat.insn_emulation;
4784 if (r != EMULATION_OK) {
4785 if (emulation_type & EMULTYPE_TRAP_UD)
4786 return EMULATE_FAIL;
4787 if (reexecute_instruction(vcpu, cr2))
4788 return EMULATE_DONE;
4789 if (emulation_type & EMULTYPE_SKIP)
4790 return EMULATE_FAIL;
4791 return handle_emulation_failure(vcpu);
4795 if (emulation_type & EMULTYPE_SKIP) {
4796 kvm_rip_write(vcpu, ctxt->_eip);
4797 return EMULATE_DONE;
4800 if (retry_instruction(ctxt, cr2, emulation_type))
4801 return EMULATE_DONE;
4803 /* this is needed for vmware backdoor interface to work since it
4804 changes registers values during IO operation */
4805 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4806 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4807 emulator_invalidate_register_cache(ctxt);
4811 r = x86_emulate_insn(ctxt);
4813 if (r == EMULATION_INTERCEPTED)
4814 return EMULATE_DONE;
4816 if (r == EMULATION_FAILED) {
4817 if (reexecute_instruction(vcpu, cr2))
4818 return EMULATE_DONE;
4820 return handle_emulation_failure(vcpu);
4823 if (ctxt->have_exception) {
4824 inject_emulated_exception(vcpu);
4826 } else if (vcpu->arch.pio.count) {
4827 if (!vcpu->arch.pio.in)
4828 vcpu->arch.pio.count = 0;
4831 vcpu->arch.complete_userspace_io = complete_emulated_pio;
4833 r = EMULATE_DO_MMIO;
4834 } else if (vcpu->mmio_needed) {
4835 if (!vcpu->mmio_is_write)
4837 r = EMULATE_DO_MMIO;
4838 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
4839 } else if (r == EMULATION_RESTART)
4845 toggle_interruptibility(vcpu, ctxt->interruptibility);
4846 kvm_set_rflags(vcpu, ctxt->eflags);
4847 kvm_make_request(KVM_REQ_EVENT, vcpu);
4848 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4849 kvm_rip_write(vcpu, ctxt->eip);
4851 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
4855 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4857 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4859 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4860 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4861 size, port, &val, 1);
4862 /* do not return to emulator after return from userspace */
4863 vcpu->arch.pio.count = 0;
4866 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4868 static void tsc_bad(void *info)
4870 __this_cpu_write(cpu_tsc_khz, 0);
4873 static void tsc_khz_changed(void *data)
4875 struct cpufreq_freqs *freq = data;
4876 unsigned long khz = 0;
4880 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4881 khz = cpufreq_quick_get(raw_smp_processor_id());
4884 __this_cpu_write(cpu_tsc_khz, khz);
4887 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4890 struct cpufreq_freqs *freq = data;
4892 struct kvm_vcpu *vcpu;
4893 int i, send_ipi = 0;
4896 * We allow guests to temporarily run on slowing clocks,
4897 * provided we notify them after, or to run on accelerating
4898 * clocks, provided we notify them before. Thus time never
4901 * However, we have a problem. We can't atomically update
4902 * the frequency of a given CPU from this function; it is
4903 * merely a notifier, which can be called from any CPU.
4904 * Changing the TSC frequency at arbitrary points in time
4905 * requires a recomputation of local variables related to
4906 * the TSC for each VCPU. We must flag these local variables
4907 * to be updated and be sure the update takes place with the
4908 * new frequency before any guests proceed.
4910 * Unfortunately, the combination of hotplug CPU and frequency
4911 * change creates an intractable locking scenario; the order
4912 * of when these callouts happen is undefined with respect to
4913 * CPU hotplug, and they can race with each other. As such,
4914 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4915 * undefined; you can actually have a CPU frequency change take
4916 * place in between the computation of X and the setting of the
4917 * variable. To protect against this problem, all updates of
4918 * the per_cpu tsc_khz variable are done in an interrupt
4919 * protected IPI, and all callers wishing to update the value
4920 * must wait for a synchronous IPI to complete (which is trivial
4921 * if the caller is on the CPU already). This establishes the
4922 * necessary total order on variable updates.
4924 * Note that because a guest time update may take place
4925 * anytime after the setting of the VCPU's request bit, the
4926 * correct TSC value must be set before the request. However,
4927 * to ensure the update actually makes it to any guest which
4928 * starts running in hardware virtualization between the set
4929 * and the acquisition of the spinlock, we must also ping the
4930 * CPU after setting the request bit.
4934 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4936 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4939 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4941 raw_spin_lock(&kvm_lock);
4942 list_for_each_entry(kvm, &vm_list, vm_list) {
4943 kvm_for_each_vcpu(i, vcpu, kvm) {
4944 if (vcpu->cpu != freq->cpu)
4946 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4947 if (vcpu->cpu != smp_processor_id())
4951 raw_spin_unlock(&kvm_lock);
4953 if (freq->old < freq->new && send_ipi) {
4955 * We upscale the frequency. Must make the guest
4956 * doesn't see old kvmclock values while running with
4957 * the new frequency, otherwise we risk the guest sees
4958 * time go backwards.
4960 * In case we update the frequency for another cpu
4961 * (which might be in guest context) send an interrupt
4962 * to kick the cpu out of guest context. Next time
4963 * guest context is entered kvmclock will be updated,
4964 * so the guest will not see stale values.
4966 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4971 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4972 .notifier_call = kvmclock_cpufreq_notifier
4975 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4976 unsigned long action, void *hcpu)
4978 unsigned int cpu = (unsigned long)hcpu;
4982 case CPU_DOWN_FAILED:
4983 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4985 case CPU_DOWN_PREPARE:
4986 smp_call_function_single(cpu, tsc_bad, NULL, 1);
4992 static struct notifier_block kvmclock_cpu_notifier_block = {
4993 .notifier_call = kvmclock_cpu_notifier,
4994 .priority = -INT_MAX
4997 static void kvm_timer_init(void)
5001 max_tsc_khz = tsc_khz;
5002 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5003 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5004 #ifdef CONFIG_CPU_FREQ
5005 struct cpufreq_policy policy;
5006 memset(&policy, 0, sizeof(policy));
5008 cpufreq_get_policy(&policy, cpu);
5009 if (policy.cpuinfo.max_freq)
5010 max_tsc_khz = policy.cpuinfo.max_freq;
5013 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5014 CPUFREQ_TRANSITION_NOTIFIER);
5016 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5017 for_each_online_cpu(cpu)
5018 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5021 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5023 int kvm_is_in_guest(void)
5025 return __this_cpu_read(current_vcpu) != NULL;
5028 static int kvm_is_user_mode(void)
5032 if (__this_cpu_read(current_vcpu))
5033 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5035 return user_mode != 0;
5038 static unsigned long kvm_get_guest_ip(void)
5040 unsigned long ip = 0;
5042 if (__this_cpu_read(current_vcpu))
5043 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5048 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5049 .is_in_guest = kvm_is_in_guest,
5050 .is_user_mode = kvm_is_user_mode,
5051 .get_guest_ip = kvm_get_guest_ip,
5054 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5056 __this_cpu_write(current_vcpu, vcpu);
5058 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5060 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5062 __this_cpu_write(current_vcpu, NULL);
5064 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5066 static void kvm_set_mmio_spte_mask(void)
5069 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5072 * Set the reserved bits and the present bit of an paging-structure
5073 * entry to generate page fault with PFER.RSV = 1.
5075 mask = ((1ull << (62 - maxphyaddr + 1)) - 1) << maxphyaddr;
5078 #ifdef CONFIG_X86_64
5080 * If reserved bit is not supported, clear the present bit to disable
5083 if (maxphyaddr == 52)
5087 kvm_mmu_set_mmio_spte_mask(mask);
5090 #ifdef CONFIG_X86_64
5091 static void pvclock_gtod_update_fn(struct work_struct *work)
5095 struct kvm_vcpu *vcpu;
5098 raw_spin_lock(&kvm_lock);
5099 list_for_each_entry(kvm, &vm_list, vm_list)
5100 kvm_for_each_vcpu(i, vcpu, kvm)
5101 set_bit(KVM_REQ_MASTERCLOCK_UPDATE, &vcpu->requests);
5102 atomic_set(&kvm_guest_has_master_clock, 0);
5103 raw_spin_unlock(&kvm_lock);
5106 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5109 * Notification about pvclock gtod data update.
5111 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5114 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5115 struct timekeeper *tk = priv;
5117 update_pvclock_gtod(tk);
5119 /* disable master clock if host does not trust, or does not
5120 * use, TSC clocksource
5122 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5123 atomic_read(&kvm_guest_has_master_clock) != 0)
5124 queue_work(system_long_wq, &pvclock_gtod_work);
5129 static struct notifier_block pvclock_gtod_notifier = {
5130 .notifier_call = pvclock_gtod_notify,
5134 int kvm_arch_init(void *opaque)
5137 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
5140 printk(KERN_ERR "kvm: already loaded the other module\n");
5145 if (!ops->cpu_has_kvm_support()) {
5146 printk(KERN_ERR "kvm: no hardware support\n");
5150 if (ops->disabled_by_bios()) {
5151 printk(KERN_ERR "kvm: disabled by bios\n");
5156 r = kvm_mmu_module_init();
5160 kvm_set_mmio_spte_mask();
5161 kvm_init_msr_list();
5164 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5165 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5169 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5172 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5175 #ifdef CONFIG_X86_64
5176 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5185 void kvm_arch_exit(void)
5187 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5189 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5190 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5191 CPUFREQ_TRANSITION_NOTIFIER);
5192 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5193 #ifdef CONFIG_X86_64
5194 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5197 kvm_mmu_module_exit();
5200 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5202 ++vcpu->stat.halt_exits;
5203 if (irqchip_in_kernel(vcpu->kvm)) {
5204 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5207 vcpu->run->exit_reason = KVM_EXIT_HLT;
5211 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5213 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5215 u64 param, ingpa, outgpa, ret;
5216 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5217 bool fast, longmode;
5221 * hypercall generates UD from non zero cpl and real mode
5224 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5225 kvm_queue_exception(vcpu, UD_VECTOR);
5229 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5230 longmode = is_long_mode(vcpu) && cs_l == 1;
5233 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5234 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5235 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5236 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5237 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5238 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5240 #ifdef CONFIG_X86_64
5242 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5243 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5244 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5248 code = param & 0xffff;
5249 fast = (param >> 16) & 0x1;
5250 rep_cnt = (param >> 32) & 0xfff;
5251 rep_idx = (param >> 48) & 0xfff;
5253 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5256 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5257 kvm_vcpu_on_spin(vcpu);
5260 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5264 ret = res | (((u64)rep_done & 0xfff) << 32);
5266 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5268 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5269 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5275 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5277 unsigned long nr, a0, a1, a2, a3, ret;
5280 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5281 return kvm_hv_hypercall(vcpu);
5283 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5284 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5285 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5286 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5287 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5289 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5291 if (!is_long_mode(vcpu)) {
5299 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5305 case KVM_HC_VAPIC_POLL_IRQ:
5313 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5314 ++vcpu->stat.hypercalls;
5317 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5319 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5321 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5322 char instruction[3];
5323 unsigned long rip = kvm_rip_read(vcpu);
5326 * Blow out the MMU to ensure that no other VCPU has an active mapping
5327 * to ensure that the updated hypercall appears atomically across all
5330 kvm_mmu_zap_all(vcpu->kvm);
5332 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5334 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5338 * Check if userspace requested an interrupt window, and that the
5339 * interrupt window is open.
5341 * No need to exit to userspace if we already have an interrupt queued.
5343 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5345 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5346 vcpu->run->request_interrupt_window &&
5347 kvm_arch_interrupt_allowed(vcpu));
5350 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5352 struct kvm_run *kvm_run = vcpu->run;
5354 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5355 kvm_run->cr8 = kvm_get_cr8(vcpu);
5356 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5357 if (irqchip_in_kernel(vcpu->kvm))
5358 kvm_run->ready_for_interrupt_injection = 1;
5360 kvm_run->ready_for_interrupt_injection =
5361 kvm_arch_interrupt_allowed(vcpu) &&
5362 !kvm_cpu_has_interrupt(vcpu) &&
5363 !kvm_event_needs_reinjection(vcpu);
5366 static int vapic_enter(struct kvm_vcpu *vcpu)
5368 struct kvm_lapic *apic = vcpu->arch.apic;
5371 if (!apic || !apic->vapic_addr)
5374 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5375 if (is_error_page(page))
5378 vcpu->arch.apic->vapic_page = page;
5382 static void vapic_exit(struct kvm_vcpu *vcpu)
5384 struct kvm_lapic *apic = vcpu->arch.apic;
5387 if (!apic || !apic->vapic_addr)
5390 idx = srcu_read_lock(&vcpu->kvm->srcu);
5391 kvm_release_page_dirty(apic->vapic_page);
5392 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5393 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5396 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5400 if (!kvm_x86_ops->update_cr8_intercept)
5403 if (!vcpu->arch.apic)
5406 if (!vcpu->arch.apic->vapic_addr)
5407 max_irr = kvm_lapic_find_highest_irr(vcpu);
5414 tpr = kvm_lapic_get_cr8(vcpu);
5416 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5419 static void inject_pending_event(struct kvm_vcpu *vcpu)
5421 /* try to reinject previous events if any */
5422 if (vcpu->arch.exception.pending) {
5423 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5424 vcpu->arch.exception.has_error_code,
5425 vcpu->arch.exception.error_code);
5426 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5427 vcpu->arch.exception.has_error_code,
5428 vcpu->arch.exception.error_code,
5429 vcpu->arch.exception.reinject);
5433 if (vcpu->arch.nmi_injected) {
5434 kvm_x86_ops->set_nmi(vcpu);
5438 if (vcpu->arch.interrupt.pending) {
5439 kvm_x86_ops->set_irq(vcpu);
5443 /* try to inject new event if pending */
5444 if (vcpu->arch.nmi_pending) {
5445 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5446 --vcpu->arch.nmi_pending;
5447 vcpu->arch.nmi_injected = true;
5448 kvm_x86_ops->set_nmi(vcpu);
5450 } else if (kvm_cpu_has_interrupt(vcpu)) {
5451 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5452 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5454 kvm_x86_ops->set_irq(vcpu);
5459 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5461 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5462 !vcpu->guest_xcr0_loaded) {
5463 /* kvm_set_xcr() also depends on this */
5464 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5465 vcpu->guest_xcr0_loaded = 1;
5469 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5471 if (vcpu->guest_xcr0_loaded) {
5472 if (vcpu->arch.xcr0 != host_xcr0)
5473 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5474 vcpu->guest_xcr0_loaded = 0;
5478 static void process_nmi(struct kvm_vcpu *vcpu)
5483 * x86 is limited to one NMI running, and one NMI pending after it.
5484 * If an NMI is already in progress, limit further NMIs to just one.
5485 * Otherwise, allow two (and we'll inject the first one immediately).
5487 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5490 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5491 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5492 kvm_make_request(KVM_REQ_EVENT, vcpu);
5495 static void kvm_gen_update_masterclock(struct kvm *kvm)
5497 #ifdef CONFIG_X86_64
5499 struct kvm_vcpu *vcpu;
5500 struct kvm_arch *ka = &kvm->arch;
5502 spin_lock(&ka->pvclock_gtod_sync_lock);
5503 kvm_make_mclock_inprogress_request(kvm);
5504 /* no guest entries from this point */
5505 pvclock_update_vm_gtod_copy(kvm);
5507 kvm_for_each_vcpu(i, vcpu, kvm)
5508 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
5510 /* guest entries allowed */
5511 kvm_for_each_vcpu(i, vcpu, kvm)
5512 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
5514 spin_unlock(&ka->pvclock_gtod_sync_lock);
5518 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5521 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5522 vcpu->run->request_interrupt_window;
5523 bool req_immediate_exit = 0;
5525 if (vcpu->requests) {
5526 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5527 kvm_mmu_unload(vcpu);
5528 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5529 __kvm_migrate_timers(vcpu);
5530 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
5531 kvm_gen_update_masterclock(vcpu->kvm);
5532 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5533 r = kvm_guest_time_update(vcpu);
5537 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5538 kvm_mmu_sync_roots(vcpu);
5539 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5540 kvm_x86_ops->tlb_flush(vcpu);
5541 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5542 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5546 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5547 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5551 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5552 vcpu->fpu_active = 0;
5553 kvm_x86_ops->fpu_deactivate(vcpu);
5555 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5556 /* Page is swapped out. Do synthetic halt */
5557 vcpu->arch.apf.halted = true;
5561 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5562 record_steal_time(vcpu);
5563 if (kvm_check_request(KVM_REQ_NMI, vcpu))
5565 req_immediate_exit =
5566 kvm_check_request(KVM_REQ_IMMEDIATE_EXIT, vcpu);
5567 if (kvm_check_request(KVM_REQ_PMU, vcpu))
5568 kvm_handle_pmu_event(vcpu);
5569 if (kvm_check_request(KVM_REQ_PMI, vcpu))
5570 kvm_deliver_pmi(vcpu);
5573 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5574 inject_pending_event(vcpu);
5576 /* enable NMI/IRQ window open exits if needed */
5577 if (vcpu->arch.nmi_pending)
5578 kvm_x86_ops->enable_nmi_window(vcpu);
5579 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5580 kvm_x86_ops->enable_irq_window(vcpu);
5582 if (kvm_lapic_enabled(vcpu)) {
5583 update_cr8_intercept(vcpu);
5584 kvm_lapic_sync_to_vapic(vcpu);
5588 r = kvm_mmu_reload(vcpu);
5590 goto cancel_injection;
5595 kvm_x86_ops->prepare_guest_switch(vcpu);
5596 if (vcpu->fpu_active)
5597 kvm_load_guest_fpu(vcpu);
5598 kvm_load_guest_xcr0(vcpu);
5600 vcpu->mode = IN_GUEST_MODE;
5602 /* We should set ->mode before check ->requests,
5603 * see the comment in make_all_cpus_request.
5607 local_irq_disable();
5609 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5610 || need_resched() || signal_pending(current)) {
5611 vcpu->mode = OUTSIDE_GUEST_MODE;
5616 goto cancel_injection;
5619 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5621 if (req_immediate_exit)
5622 smp_send_reschedule(vcpu->cpu);
5626 if (unlikely(vcpu->arch.switch_db_regs)) {
5628 set_debugreg(vcpu->arch.eff_db[0], 0);
5629 set_debugreg(vcpu->arch.eff_db[1], 1);
5630 set_debugreg(vcpu->arch.eff_db[2], 2);
5631 set_debugreg(vcpu->arch.eff_db[3], 3);
5634 trace_kvm_entry(vcpu->vcpu_id);
5635 kvm_x86_ops->run(vcpu);
5638 * If the guest has used debug registers, at least dr7
5639 * will be disabled while returning to the host.
5640 * If we don't have active breakpoints in the host, we don't
5641 * care about the messed up debug address registers. But if
5642 * we have some of them active, restore the old state.
5644 if (hw_breakpoint_active())
5645 hw_breakpoint_restore();
5647 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
5650 vcpu->mode = OUTSIDE_GUEST_MODE;
5657 * We must have an instruction between local_irq_enable() and
5658 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5659 * the interrupt shadow. The stat.exits increment will do nicely.
5660 * But we need to prevent reordering, hence this barrier():
5668 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5671 * Profile KVM exit RIPs:
5673 if (unlikely(prof_on == KVM_PROFILING)) {
5674 unsigned long rip = kvm_rip_read(vcpu);
5675 profile_hit(KVM_PROFILING, (void *)rip);
5678 if (unlikely(vcpu->arch.tsc_always_catchup))
5679 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5681 if (vcpu->arch.apic_attention)
5682 kvm_lapic_sync_from_vapic(vcpu);
5684 r = kvm_x86_ops->handle_exit(vcpu);
5688 kvm_x86_ops->cancel_injection(vcpu);
5689 if (unlikely(vcpu->arch.apic_attention))
5690 kvm_lapic_sync_from_vapic(vcpu);
5696 static int __vcpu_run(struct kvm_vcpu *vcpu)
5699 struct kvm *kvm = vcpu->kvm;
5701 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5702 pr_debug("vcpu %d received sipi with vector # %x\n",
5703 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5704 kvm_lapic_reset(vcpu);
5705 r = kvm_vcpu_reset(vcpu);
5708 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5711 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5712 r = vapic_enter(vcpu);
5714 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5720 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5721 !vcpu->arch.apf.halted)
5722 r = vcpu_enter_guest(vcpu);
5724 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5725 kvm_vcpu_block(vcpu);
5726 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5727 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5729 switch(vcpu->arch.mp_state) {
5730 case KVM_MP_STATE_HALTED:
5731 vcpu->arch.mp_state =
5732 KVM_MP_STATE_RUNNABLE;
5733 case KVM_MP_STATE_RUNNABLE:
5734 vcpu->arch.apf.halted = false;
5736 case KVM_MP_STATE_SIPI_RECEIVED:
5747 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5748 if (kvm_cpu_has_pending_timer(vcpu))
5749 kvm_inject_pending_timer_irqs(vcpu);
5751 if (dm_request_for_irq_injection(vcpu)) {
5753 vcpu->run->exit_reason = KVM_EXIT_INTR;
5754 ++vcpu->stat.request_irq_exits;
5757 kvm_check_async_pf_completion(vcpu);
5759 if (signal_pending(current)) {
5761 vcpu->run->exit_reason = KVM_EXIT_INTR;
5762 ++vcpu->stat.signal_exits;
5764 if (need_resched()) {
5765 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5767 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5771 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5778 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
5781 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5782 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5783 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5784 if (r != EMULATE_DONE)
5789 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
5791 BUG_ON(!vcpu->arch.pio.count);
5793 return complete_emulated_io(vcpu);
5797 * Implements the following, as a state machine:
5812 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
5814 struct kvm_run *run = vcpu->run;
5815 struct kvm_mmio_fragment *frag;
5817 BUG_ON(!vcpu->mmio_needed);
5819 /* Complete previous fragment */
5820 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment++];
5821 if (!vcpu->mmio_is_write)
5822 memcpy(frag->data, run->mmio.data, frag->len);
5823 if (vcpu->mmio_cur_fragment == vcpu->mmio_nr_fragments) {
5824 vcpu->mmio_needed = 0;
5825 if (vcpu->mmio_is_write)
5827 vcpu->mmio_read_completed = 1;
5828 return complete_emulated_io(vcpu);
5830 /* Initiate next fragment */
5832 run->exit_reason = KVM_EXIT_MMIO;
5833 run->mmio.phys_addr = frag->gpa;
5834 if (vcpu->mmio_is_write)
5835 memcpy(run->mmio.data, frag->data, frag->len);
5836 run->mmio.len = frag->len;
5837 run->mmio.is_write = vcpu->mmio_is_write;
5838 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5843 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5848 if (!tsk_used_math(current) && init_fpu(current))
5851 if (vcpu->sigset_active)
5852 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5854 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5855 kvm_vcpu_block(vcpu);
5856 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5861 /* re-sync apic's tpr */
5862 if (!irqchip_in_kernel(vcpu->kvm)) {
5863 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5869 if (unlikely(vcpu->arch.complete_userspace_io)) {
5870 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
5871 vcpu->arch.complete_userspace_io = NULL;
5876 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
5878 r = __vcpu_run(vcpu);
5881 post_kvm_run_save(vcpu);
5882 if (vcpu->sigset_active)
5883 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5888 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5890 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5892 * We are here if userspace calls get_regs() in the middle of
5893 * instruction emulation. Registers state needs to be copied
5894 * back from emulation context to vcpu. Userspace shouldn't do
5895 * that usually, but some bad designed PV devices (vmware
5896 * backdoor interface) need this to work
5898 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
5899 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5901 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5902 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5903 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5904 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5905 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5906 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5907 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5908 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5909 #ifdef CONFIG_X86_64
5910 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5911 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5912 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5913 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5914 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5915 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5916 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5917 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5920 regs->rip = kvm_rip_read(vcpu);
5921 regs->rflags = kvm_get_rflags(vcpu);
5926 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5928 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
5929 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5931 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5932 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5933 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5934 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5935 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5936 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5937 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5938 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5939 #ifdef CONFIG_X86_64
5940 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5941 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5942 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5943 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5944 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5945 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5946 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5947 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5950 kvm_rip_write(vcpu, regs->rip);
5951 kvm_set_rflags(vcpu, regs->rflags);
5953 vcpu->arch.exception.pending = false;
5955 kvm_make_request(KVM_REQ_EVENT, vcpu);
5960 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5962 struct kvm_segment cs;
5964 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
5968 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5970 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5971 struct kvm_sregs *sregs)
5975 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5976 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5977 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5978 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5979 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5980 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5982 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5983 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5985 kvm_x86_ops->get_idt(vcpu, &dt);
5986 sregs->idt.limit = dt.size;
5987 sregs->idt.base = dt.address;
5988 kvm_x86_ops->get_gdt(vcpu, &dt);
5989 sregs->gdt.limit = dt.size;
5990 sregs->gdt.base = dt.address;
5992 sregs->cr0 = kvm_read_cr0(vcpu);
5993 sregs->cr2 = vcpu->arch.cr2;
5994 sregs->cr3 = kvm_read_cr3(vcpu);
5995 sregs->cr4 = kvm_read_cr4(vcpu);
5996 sregs->cr8 = kvm_get_cr8(vcpu);
5997 sregs->efer = vcpu->arch.efer;
5998 sregs->apic_base = kvm_get_apic_base(vcpu);
6000 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6002 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6003 set_bit(vcpu->arch.interrupt.nr,
6004 (unsigned long *)sregs->interrupt_bitmap);
6009 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6010 struct kvm_mp_state *mp_state)
6012 mp_state->mp_state = vcpu->arch.mp_state;
6016 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6017 struct kvm_mp_state *mp_state)
6019 vcpu->arch.mp_state = mp_state->mp_state;
6020 kvm_make_request(KVM_REQ_EVENT, vcpu);
6024 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6025 int reason, bool has_error_code, u32 error_code)
6027 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6030 init_emulate_ctxt(vcpu);
6032 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6033 has_error_code, error_code);
6036 return EMULATE_FAIL;
6038 kvm_rip_write(vcpu, ctxt->eip);
6039 kvm_set_rflags(vcpu, ctxt->eflags);
6040 kvm_make_request(KVM_REQ_EVENT, vcpu);
6041 return EMULATE_DONE;
6043 EXPORT_SYMBOL_GPL(kvm_task_switch);
6045 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6046 struct kvm_sregs *sregs)
6048 int mmu_reset_needed = 0;
6049 int pending_vec, max_bits, idx;
6052 dt.size = sregs->idt.limit;
6053 dt.address = sregs->idt.base;
6054 kvm_x86_ops->set_idt(vcpu, &dt);
6055 dt.size = sregs->gdt.limit;
6056 dt.address = sregs->gdt.base;
6057 kvm_x86_ops->set_gdt(vcpu, &dt);
6059 vcpu->arch.cr2 = sregs->cr2;
6060 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6061 vcpu->arch.cr3 = sregs->cr3;
6062 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6064 kvm_set_cr8(vcpu, sregs->cr8);
6066 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6067 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6068 kvm_set_apic_base(vcpu, sregs->apic_base);
6070 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6071 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6072 vcpu->arch.cr0 = sregs->cr0;
6074 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6075 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6076 if (sregs->cr4 & X86_CR4_OSXSAVE)
6077 kvm_update_cpuid(vcpu);
6079 idx = srcu_read_lock(&vcpu->kvm->srcu);
6080 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6081 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6082 mmu_reset_needed = 1;
6084 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6086 if (mmu_reset_needed)
6087 kvm_mmu_reset_context(vcpu);
6089 max_bits = KVM_NR_INTERRUPTS;
6090 pending_vec = find_first_bit(
6091 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6092 if (pending_vec < max_bits) {
6093 kvm_queue_interrupt(vcpu, pending_vec, false);
6094 pr_debug("Set back pending irq %d\n", pending_vec);
6097 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6098 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6099 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6100 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6101 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6102 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6104 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6105 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6107 update_cr8_intercept(vcpu);
6109 /* Older userspace won't unhalt the vcpu on reset. */
6110 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6111 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6113 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6115 kvm_make_request(KVM_REQ_EVENT, vcpu);
6120 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6121 struct kvm_guest_debug *dbg)
6123 unsigned long rflags;
6126 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6128 if (vcpu->arch.exception.pending)
6130 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6131 kvm_queue_exception(vcpu, DB_VECTOR);
6133 kvm_queue_exception(vcpu, BP_VECTOR);
6137 * Read rflags as long as potentially injected trace flags are still
6140 rflags = kvm_get_rflags(vcpu);
6142 vcpu->guest_debug = dbg->control;
6143 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6144 vcpu->guest_debug = 0;
6146 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6147 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6148 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6149 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6151 for (i = 0; i < KVM_NR_DB_REGS; i++)
6152 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6154 kvm_update_dr7(vcpu);
6156 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6157 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6158 get_segment_base(vcpu, VCPU_SREG_CS);
6161 * Trigger an rflags update that will inject or remove the trace
6164 kvm_set_rflags(vcpu, rflags);
6166 kvm_x86_ops->update_db_bp_intercept(vcpu);
6176 * Translate a guest virtual address to a guest physical address.
6178 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6179 struct kvm_translation *tr)
6181 unsigned long vaddr = tr->linear_address;
6185 idx = srcu_read_lock(&vcpu->kvm->srcu);
6186 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6187 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6188 tr->physical_address = gpa;
6189 tr->valid = gpa != UNMAPPED_GVA;
6196 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6198 struct i387_fxsave_struct *fxsave =
6199 &vcpu->arch.guest_fpu.state->fxsave;
6201 memcpy(fpu->fpr, fxsave->st_space, 128);
6202 fpu->fcw = fxsave->cwd;
6203 fpu->fsw = fxsave->swd;
6204 fpu->ftwx = fxsave->twd;
6205 fpu->last_opcode = fxsave->fop;
6206 fpu->last_ip = fxsave->rip;
6207 fpu->last_dp = fxsave->rdp;
6208 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6213 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6215 struct i387_fxsave_struct *fxsave =
6216 &vcpu->arch.guest_fpu.state->fxsave;
6218 memcpy(fxsave->st_space, fpu->fpr, 128);
6219 fxsave->cwd = fpu->fcw;
6220 fxsave->swd = fpu->fsw;
6221 fxsave->twd = fpu->ftwx;
6222 fxsave->fop = fpu->last_opcode;
6223 fxsave->rip = fpu->last_ip;
6224 fxsave->rdp = fpu->last_dp;
6225 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6230 int fx_init(struct kvm_vcpu *vcpu)
6234 err = fpu_alloc(&vcpu->arch.guest_fpu);
6238 fpu_finit(&vcpu->arch.guest_fpu);
6241 * Ensure guest xcr0 is valid for loading
6243 vcpu->arch.xcr0 = XSTATE_FP;
6245 vcpu->arch.cr0 |= X86_CR0_ET;
6249 EXPORT_SYMBOL_GPL(fx_init);
6251 static void fx_free(struct kvm_vcpu *vcpu)
6253 fpu_free(&vcpu->arch.guest_fpu);
6256 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6258 if (vcpu->guest_fpu_loaded)
6262 * Restore all possible states in the guest,
6263 * and assume host would use all available bits.
6264 * Guest xcr0 would be loaded later.
6266 kvm_put_guest_xcr0(vcpu);
6267 vcpu->guest_fpu_loaded = 1;
6268 __kernel_fpu_begin();
6269 fpu_restore_checking(&vcpu->arch.guest_fpu);
6273 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6275 kvm_put_guest_xcr0(vcpu);
6277 if (!vcpu->guest_fpu_loaded)
6280 vcpu->guest_fpu_loaded = 0;
6281 fpu_save_init(&vcpu->arch.guest_fpu);
6283 ++vcpu->stat.fpu_reload;
6284 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6288 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6290 kvmclock_reset(vcpu);
6292 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6294 kvm_x86_ops->vcpu_free(vcpu);
6297 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6300 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6301 printk_once(KERN_WARNING
6302 "kvm: SMP vm created on host with unstable TSC; "
6303 "guest TSC will not be reliable\n");
6304 return kvm_x86_ops->vcpu_create(kvm, id);
6307 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6311 vcpu->arch.mtrr_state.have_fixed = 1;
6312 r = vcpu_load(vcpu);
6315 r = kvm_vcpu_reset(vcpu);
6317 r = kvm_mmu_setup(vcpu);
6323 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
6327 r = vcpu_load(vcpu);
6330 kvm_write_tsc(vcpu, 0);
6336 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6339 vcpu->arch.apf.msr_val = 0;
6341 r = vcpu_load(vcpu);
6343 kvm_mmu_unload(vcpu);
6347 kvm_x86_ops->vcpu_free(vcpu);
6350 static int kvm_vcpu_reset(struct kvm_vcpu *vcpu)
6352 atomic_set(&vcpu->arch.nmi_queued, 0);
6353 vcpu->arch.nmi_pending = 0;
6354 vcpu->arch.nmi_injected = false;
6356 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6357 vcpu->arch.dr6 = DR6_FIXED_1;
6358 vcpu->arch.dr7 = DR7_FIXED_1;
6359 kvm_update_dr7(vcpu);
6361 kvm_make_request(KVM_REQ_EVENT, vcpu);
6362 vcpu->arch.apf.msr_val = 0;
6363 vcpu->arch.st.msr_val = 0;
6365 kvmclock_reset(vcpu);
6367 kvm_clear_async_pf_completion_queue(vcpu);
6368 kvm_async_pf_hash_reset(vcpu);
6369 vcpu->arch.apf.halted = false;
6371 kvm_pmu_reset(vcpu);
6373 return kvm_x86_ops->vcpu_reset(vcpu);
6376 int kvm_arch_hardware_enable(void *garbage)
6379 struct kvm_vcpu *vcpu;
6384 bool stable, backwards_tsc = false;
6386 kvm_shared_msr_cpu_online();
6387 ret = kvm_x86_ops->hardware_enable(garbage);
6391 local_tsc = native_read_tsc();
6392 stable = !check_tsc_unstable();
6393 list_for_each_entry(kvm, &vm_list, vm_list) {
6394 kvm_for_each_vcpu(i, vcpu, kvm) {
6395 if (!stable && vcpu->cpu == smp_processor_id())
6396 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
6397 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
6398 backwards_tsc = true;
6399 if (vcpu->arch.last_host_tsc > max_tsc)
6400 max_tsc = vcpu->arch.last_host_tsc;
6406 * Sometimes, even reliable TSCs go backwards. This happens on
6407 * platforms that reset TSC during suspend or hibernate actions, but
6408 * maintain synchronization. We must compensate. Fortunately, we can
6409 * detect that condition here, which happens early in CPU bringup,
6410 * before any KVM threads can be running. Unfortunately, we can't
6411 * bring the TSCs fully up to date with real time, as we aren't yet far
6412 * enough into CPU bringup that we know how much real time has actually
6413 * elapsed; our helper function, get_kernel_ns() will be using boot
6414 * variables that haven't been updated yet.
6416 * So we simply find the maximum observed TSC above, then record the
6417 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
6418 * the adjustment will be applied. Note that we accumulate
6419 * adjustments, in case multiple suspend cycles happen before some VCPU
6420 * gets a chance to run again. In the event that no KVM threads get a
6421 * chance to run, we will miss the entire elapsed period, as we'll have
6422 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
6423 * loose cycle time. This isn't too big a deal, since the loss will be
6424 * uniform across all VCPUs (not to mention the scenario is extremely
6425 * unlikely). It is possible that a second hibernate recovery happens
6426 * much faster than a first, causing the observed TSC here to be
6427 * smaller; this would require additional padding adjustment, which is
6428 * why we set last_host_tsc to the local tsc observed here.
6430 * N.B. - this code below runs only on platforms with reliable TSC,
6431 * as that is the only way backwards_tsc is set above. Also note
6432 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
6433 * have the same delta_cyc adjustment applied if backwards_tsc
6434 * is detected. Note further, this adjustment is only done once,
6435 * as we reset last_host_tsc on all VCPUs to stop this from being
6436 * called multiple times (one for each physical CPU bringup).
6438 * Platforms with unreliable TSCs don't have to deal with this, they
6439 * will be compensated by the logic in vcpu_load, which sets the TSC to
6440 * catchup mode. This will catchup all VCPUs to real time, but cannot
6441 * guarantee that they stay in perfect synchronization.
6443 if (backwards_tsc) {
6444 u64 delta_cyc = max_tsc - local_tsc;
6445 list_for_each_entry(kvm, &vm_list, vm_list) {
6446 kvm_for_each_vcpu(i, vcpu, kvm) {
6447 vcpu->arch.tsc_offset_adjustment += delta_cyc;
6448 vcpu->arch.last_host_tsc = local_tsc;
6449 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
6454 * We have to disable TSC offset matching.. if you were
6455 * booting a VM while issuing an S4 host suspend....
6456 * you may have some problem. Solving this issue is
6457 * left as an exercise to the reader.
6459 kvm->arch.last_tsc_nsec = 0;
6460 kvm->arch.last_tsc_write = 0;
6467 void kvm_arch_hardware_disable(void *garbage)
6469 kvm_x86_ops->hardware_disable(garbage);
6470 drop_user_return_notifiers(garbage);
6473 int kvm_arch_hardware_setup(void)
6475 return kvm_x86_ops->hardware_setup();
6478 void kvm_arch_hardware_unsetup(void)
6480 kvm_x86_ops->hardware_unsetup();
6483 void kvm_arch_check_processor_compat(void *rtn)
6485 kvm_x86_ops->check_processor_compatibility(rtn);
6488 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
6490 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
6493 struct static_key kvm_no_apic_vcpu __read_mostly;
6495 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6501 BUG_ON(vcpu->kvm == NULL);
6504 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6505 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6506 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6508 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6510 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6515 vcpu->arch.pio_data = page_address(page);
6517 kvm_set_tsc_khz(vcpu, max_tsc_khz);
6519 r = kvm_mmu_create(vcpu);
6521 goto fail_free_pio_data;
6523 if (irqchip_in_kernel(kvm)) {
6524 r = kvm_create_lapic(vcpu);
6526 goto fail_mmu_destroy;
6528 static_key_slow_inc(&kvm_no_apic_vcpu);
6530 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6532 if (!vcpu->arch.mce_banks) {
6534 goto fail_free_lapic;
6536 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6538 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6539 goto fail_free_mce_banks;
6541 kvm_async_pf_hash_reset(vcpu);
6545 fail_free_mce_banks:
6546 kfree(vcpu->arch.mce_banks);
6548 kvm_free_lapic(vcpu);
6550 kvm_mmu_destroy(vcpu);
6552 free_page((unsigned long)vcpu->arch.pio_data);
6557 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6561 kvm_pmu_destroy(vcpu);
6562 kfree(vcpu->arch.mce_banks);
6563 kvm_free_lapic(vcpu);
6564 idx = srcu_read_lock(&vcpu->kvm->srcu);
6565 kvm_mmu_destroy(vcpu);
6566 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6567 free_page((unsigned long)vcpu->arch.pio_data);
6568 if (!irqchip_in_kernel(vcpu->kvm))
6569 static_key_slow_dec(&kvm_no_apic_vcpu);
6572 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
6577 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6578 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6580 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6581 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6582 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
6583 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
6584 &kvm->arch.irq_sources_bitmap);
6586 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6587 mutex_init(&kvm->arch.apic_map_lock);
6588 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
6590 pvclock_update_vm_gtod_copy(kvm);
6595 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6598 r = vcpu_load(vcpu);
6600 kvm_mmu_unload(vcpu);
6604 static void kvm_free_vcpus(struct kvm *kvm)
6607 struct kvm_vcpu *vcpu;
6610 * Unpin any mmu pages first.
6612 kvm_for_each_vcpu(i, vcpu, kvm) {
6613 kvm_clear_async_pf_completion_queue(vcpu);
6614 kvm_unload_vcpu_mmu(vcpu);
6616 kvm_for_each_vcpu(i, vcpu, kvm)
6617 kvm_arch_vcpu_free(vcpu);
6619 mutex_lock(&kvm->lock);
6620 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6621 kvm->vcpus[i] = NULL;
6623 atomic_set(&kvm->online_vcpus, 0);
6624 mutex_unlock(&kvm->lock);
6627 void kvm_arch_sync_events(struct kvm *kvm)
6629 kvm_free_all_assigned_devices(kvm);
6633 void kvm_arch_destroy_vm(struct kvm *kvm)
6635 kvm_iommu_unmap_guest(kvm);
6636 kfree(kvm->arch.vpic);
6637 kfree(kvm->arch.vioapic);
6638 kvm_free_vcpus(kvm);
6639 if (kvm->arch.apic_access_page)
6640 put_page(kvm->arch.apic_access_page);
6641 if (kvm->arch.ept_identity_pagetable)
6642 put_page(kvm->arch.ept_identity_pagetable);
6643 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
6646 void kvm_arch_free_memslot(struct kvm_memory_slot *free,
6647 struct kvm_memory_slot *dont)
6651 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6652 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
6653 kvm_kvfree(free->arch.rmap[i]);
6654 free->arch.rmap[i] = NULL;
6659 if (!dont || free->arch.lpage_info[i - 1] !=
6660 dont->arch.lpage_info[i - 1]) {
6661 kvm_kvfree(free->arch.lpage_info[i - 1]);
6662 free->arch.lpage_info[i - 1] = NULL;
6667 int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages)
6671 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6676 lpages = gfn_to_index(slot->base_gfn + npages - 1,
6677 slot->base_gfn, level) + 1;
6679 slot->arch.rmap[i] =
6680 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
6681 if (!slot->arch.rmap[i])
6686 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
6687 sizeof(*slot->arch.lpage_info[i - 1]));
6688 if (!slot->arch.lpage_info[i - 1])
6691 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
6692 slot->arch.lpage_info[i - 1][0].write_count = 1;
6693 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
6694 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
6695 ugfn = slot->userspace_addr >> PAGE_SHIFT;
6697 * If the gfn and userspace address are not aligned wrt each
6698 * other, or if explicitly asked to, disable large page
6699 * support for this slot
6701 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
6702 !kvm_largepages_enabled()) {
6705 for (j = 0; j < lpages; ++j)
6706 slot->arch.lpage_info[i - 1][j].write_count = 1;
6713 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6714 kvm_kvfree(slot->arch.rmap[i]);
6715 slot->arch.rmap[i] = NULL;
6719 kvm_kvfree(slot->arch.lpage_info[i - 1]);
6720 slot->arch.lpage_info[i - 1] = NULL;
6725 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6726 struct kvm_memory_slot *memslot,
6727 struct kvm_memory_slot old,
6728 struct kvm_userspace_memory_region *mem,
6731 int npages = memslot->npages;
6732 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6734 /* Prevent internal slot pages from being moved by fork()/COW. */
6735 if (memslot->id >= KVM_MEMORY_SLOTS)
6736 map_flags = MAP_SHARED | MAP_ANONYMOUS;
6738 /*To keep backward compatibility with older userspace,
6739 *x86 needs to handle !user_alloc case.
6742 if (npages && !old.npages) {
6743 unsigned long userspace_addr;
6745 userspace_addr = vm_mmap(NULL, 0,
6747 PROT_READ | PROT_WRITE,
6751 if (IS_ERR((void *)userspace_addr))
6752 return PTR_ERR((void *)userspace_addr);
6754 memslot->userspace_addr = userspace_addr;
6762 void kvm_arch_commit_memory_region(struct kvm *kvm,
6763 struct kvm_userspace_memory_region *mem,
6764 struct kvm_memory_slot old,
6768 int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
6770 if (!user_alloc && !old.user_alloc && old.npages && !npages) {
6773 ret = vm_munmap(old.userspace_addr,
6774 old.npages * PAGE_SIZE);
6777 "kvm_vm_ioctl_set_memory_region: "
6778 "failed to munmap memory\n");
6781 if (!kvm->arch.n_requested_mmu_pages)
6782 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6784 spin_lock(&kvm->mmu_lock);
6786 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6787 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6788 spin_unlock(&kvm->mmu_lock);
6790 * If memory slot is created, or moved, we need to clear all
6793 if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT) {
6794 kvm_mmu_zap_all(kvm);
6795 kvm_reload_remote_mmus(kvm);
6799 void kvm_arch_flush_shadow_all(struct kvm *kvm)
6801 kvm_mmu_zap_all(kvm);
6802 kvm_reload_remote_mmus(kvm);
6805 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
6806 struct kvm_memory_slot *slot)
6808 kvm_arch_flush_shadow_all(kvm);
6811 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6813 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6814 !vcpu->arch.apf.halted)
6815 || !list_empty_careful(&vcpu->async_pf.done)
6816 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6817 || atomic_read(&vcpu->arch.nmi_queued) ||
6818 (kvm_arch_interrupt_allowed(vcpu) &&
6819 kvm_cpu_has_interrupt(vcpu));
6822 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
6824 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
6827 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6829 return kvm_x86_ops->interrupt_allowed(vcpu);
6832 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6834 unsigned long current_rip = kvm_rip_read(vcpu) +
6835 get_segment_base(vcpu, VCPU_SREG_CS);
6837 return current_rip == linear_rip;
6839 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6841 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6843 unsigned long rflags;
6845 rflags = kvm_x86_ops->get_rflags(vcpu);
6846 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6847 rflags &= ~X86_EFLAGS_TF;
6850 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6852 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6854 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6855 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6856 rflags |= X86_EFLAGS_TF;
6857 kvm_x86_ops->set_rflags(vcpu, rflags);
6858 kvm_make_request(KVM_REQ_EVENT, vcpu);
6860 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6862 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6866 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6867 is_error_page(work->page))
6870 r = kvm_mmu_reload(vcpu);
6874 if (!vcpu->arch.mmu.direct_map &&
6875 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6878 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6881 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6883 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6886 static inline u32 kvm_async_pf_next_probe(u32 key)
6888 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6891 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6893 u32 key = kvm_async_pf_hash_fn(gfn);
6895 while (vcpu->arch.apf.gfns[key] != ~0)
6896 key = kvm_async_pf_next_probe(key);
6898 vcpu->arch.apf.gfns[key] = gfn;
6901 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6904 u32 key = kvm_async_pf_hash_fn(gfn);
6906 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6907 (vcpu->arch.apf.gfns[key] != gfn &&
6908 vcpu->arch.apf.gfns[key] != ~0); i++)
6909 key = kvm_async_pf_next_probe(key);
6914 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6916 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6919 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6923 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6925 vcpu->arch.apf.gfns[i] = ~0;
6927 j = kvm_async_pf_next_probe(j);
6928 if (vcpu->arch.apf.gfns[j] == ~0)
6930 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6932 * k lies cyclically in ]i,j]
6934 * |....j i.k.| or |.k..j i...|
6936 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6937 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6942 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6945 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6949 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6950 struct kvm_async_pf *work)
6952 struct x86_exception fault;
6954 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6955 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6957 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6958 (vcpu->arch.apf.send_user_only &&
6959 kvm_x86_ops->get_cpl(vcpu) == 0))
6960 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6961 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6962 fault.vector = PF_VECTOR;
6963 fault.error_code_valid = true;
6964 fault.error_code = 0;
6965 fault.nested_page_fault = false;
6966 fault.address = work->arch.token;
6967 kvm_inject_page_fault(vcpu, &fault);
6971 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6972 struct kvm_async_pf *work)
6974 struct x86_exception fault;
6976 trace_kvm_async_pf_ready(work->arch.token, work->gva);
6977 if (is_error_page(work->page))
6978 work->arch.token = ~0; /* broadcast wakeup */
6980 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6982 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6983 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6984 fault.vector = PF_VECTOR;
6985 fault.error_code_valid = true;
6986 fault.error_code = 0;
6987 fault.nested_page_fault = false;
6988 fault.address = work->arch.token;
6989 kvm_inject_page_fault(vcpu, &fault);
6991 vcpu->arch.apf.halted = false;
6992 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6995 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6997 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7000 return !kvm_event_needs_reinjection(vcpu) &&
7001 kvm_x86_ops->interrupt_allowed(vcpu);
7004 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
7005 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
7006 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
7007 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
7008 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
7009 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
7010 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
7011 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
7012 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
7013 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
7014 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
7015 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);