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 unsigned int min_timer_period_us = 500;
98 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
100 bool kvm_has_tsc_control;
101 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
102 u32 kvm_max_guest_tsc_khz;
103 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
105 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
106 static u32 tsc_tolerance_ppm = 250;
107 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
109 #define KVM_NR_SHARED_MSRS 16
111 struct kvm_shared_msrs_global {
113 u32 msrs[KVM_NR_SHARED_MSRS];
116 struct kvm_shared_msrs {
117 struct user_return_notifier urn;
119 struct kvm_shared_msr_values {
122 } values[KVM_NR_SHARED_MSRS];
125 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
126 static struct kvm_shared_msrs __percpu *shared_msrs;
128 struct kvm_stats_debugfs_item debugfs_entries[] = {
129 { "pf_fixed", VCPU_STAT(pf_fixed) },
130 { "pf_guest", VCPU_STAT(pf_guest) },
131 { "tlb_flush", VCPU_STAT(tlb_flush) },
132 { "invlpg", VCPU_STAT(invlpg) },
133 { "exits", VCPU_STAT(exits) },
134 { "io_exits", VCPU_STAT(io_exits) },
135 { "mmio_exits", VCPU_STAT(mmio_exits) },
136 { "signal_exits", VCPU_STAT(signal_exits) },
137 { "irq_window", VCPU_STAT(irq_window_exits) },
138 { "nmi_window", VCPU_STAT(nmi_window_exits) },
139 { "halt_exits", VCPU_STAT(halt_exits) },
140 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
141 { "hypercalls", VCPU_STAT(hypercalls) },
142 { "request_irq", VCPU_STAT(request_irq_exits) },
143 { "irq_exits", VCPU_STAT(irq_exits) },
144 { "host_state_reload", VCPU_STAT(host_state_reload) },
145 { "efer_reload", VCPU_STAT(efer_reload) },
146 { "fpu_reload", VCPU_STAT(fpu_reload) },
147 { "insn_emulation", VCPU_STAT(insn_emulation) },
148 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
149 { "irq_injections", VCPU_STAT(irq_injections) },
150 { "nmi_injections", VCPU_STAT(nmi_injections) },
151 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
152 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
153 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
154 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
155 { "mmu_flooded", VM_STAT(mmu_flooded) },
156 { "mmu_recycled", VM_STAT(mmu_recycled) },
157 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
158 { "mmu_unsync", VM_STAT(mmu_unsync) },
159 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
160 { "largepages", VM_STAT(lpages) },
164 u64 __read_mostly host_xcr0;
166 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
168 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
171 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
172 vcpu->arch.apf.gfns[i] = ~0;
175 static void kvm_on_user_return(struct user_return_notifier *urn)
178 struct kvm_shared_msrs *locals
179 = container_of(urn, struct kvm_shared_msrs, urn);
180 struct kvm_shared_msr_values *values;
182 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
183 values = &locals->values[slot];
184 if (values->host != values->curr) {
185 wrmsrl(shared_msrs_global.msrs[slot], values->host);
186 values->curr = values->host;
189 locals->registered = false;
190 user_return_notifier_unregister(urn);
193 static void shared_msr_update(unsigned slot, u32 msr)
196 unsigned int cpu = smp_processor_id();
197 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
199 /* only read, and nobody should modify it at this time,
200 * so don't need lock */
201 if (slot >= shared_msrs_global.nr) {
202 printk(KERN_ERR "kvm: invalid MSR slot!");
205 rdmsrl_safe(msr, &value);
206 smsr->values[slot].host = value;
207 smsr->values[slot].curr = value;
210 void kvm_define_shared_msr(unsigned slot, u32 msr)
212 if (slot >= shared_msrs_global.nr)
213 shared_msrs_global.nr = slot + 1;
214 shared_msrs_global.msrs[slot] = msr;
215 /* we need ensured the shared_msr_global have been updated */
218 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
220 static void kvm_shared_msr_cpu_online(void)
224 for (i = 0; i < shared_msrs_global.nr; ++i)
225 shared_msr_update(i, shared_msrs_global.msrs[i]);
228 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
230 unsigned int cpu = smp_processor_id();
231 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
234 if (((value ^ smsr->values[slot].curr) & mask) == 0)
236 smsr->values[slot].curr = value;
237 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
241 if (!smsr->registered) {
242 smsr->urn.on_user_return = kvm_on_user_return;
243 user_return_notifier_register(&smsr->urn);
244 smsr->registered = true;
248 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
250 static void drop_user_return_notifiers(void *ignore)
252 unsigned int cpu = smp_processor_id();
253 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
255 if (smsr->registered)
256 kvm_on_user_return(&smsr->urn);
259 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
261 return vcpu->arch.apic_base;
263 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
265 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
267 /* TODO: reserve bits check */
268 kvm_lapic_set_base(vcpu, data);
270 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
272 asmlinkage void kvm_spurious_fault(void)
274 /* Fault while not rebooting. We want the trace. */
277 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
279 #define EXCPT_BENIGN 0
280 #define EXCPT_CONTRIBUTORY 1
283 static int exception_class(int vector)
293 return EXCPT_CONTRIBUTORY;
300 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
301 unsigned nr, bool has_error, u32 error_code,
307 kvm_make_request(KVM_REQ_EVENT, vcpu);
309 if (!vcpu->arch.exception.pending) {
311 vcpu->arch.exception.pending = true;
312 vcpu->arch.exception.has_error_code = has_error;
313 vcpu->arch.exception.nr = nr;
314 vcpu->arch.exception.error_code = error_code;
315 vcpu->arch.exception.reinject = reinject;
319 /* to check exception */
320 prev_nr = vcpu->arch.exception.nr;
321 if (prev_nr == DF_VECTOR) {
322 /* triple fault -> shutdown */
323 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
326 class1 = exception_class(prev_nr);
327 class2 = exception_class(nr);
328 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
329 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
330 /* generate double fault per SDM Table 5-5 */
331 vcpu->arch.exception.pending = true;
332 vcpu->arch.exception.has_error_code = true;
333 vcpu->arch.exception.nr = DF_VECTOR;
334 vcpu->arch.exception.error_code = 0;
336 /* replace previous exception with a new one in a hope
337 that instruction re-execution will regenerate lost
342 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
344 kvm_multiple_exception(vcpu, nr, false, 0, false);
346 EXPORT_SYMBOL_GPL(kvm_queue_exception);
348 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
350 kvm_multiple_exception(vcpu, nr, false, 0, true);
352 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
354 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
357 kvm_inject_gp(vcpu, 0);
359 kvm_x86_ops->skip_emulated_instruction(vcpu);
361 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
363 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
365 ++vcpu->stat.pf_guest;
366 vcpu->arch.cr2 = fault->address;
367 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
369 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
371 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
373 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
374 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
376 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
379 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
381 atomic_inc(&vcpu->arch.nmi_queued);
382 kvm_make_request(KVM_REQ_NMI, vcpu);
384 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
386 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
388 kvm_multiple_exception(vcpu, nr, true, error_code, false);
390 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
392 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
394 kvm_multiple_exception(vcpu, nr, true, error_code, true);
396 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
399 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
400 * a #GP and return false.
402 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
404 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
406 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
409 EXPORT_SYMBOL_GPL(kvm_require_cpl);
412 * This function will be used to read from the physical memory of the currently
413 * running guest. The difference to kvm_read_guest_page is that this function
414 * can read from guest physical or from the guest's guest physical memory.
416 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
417 gfn_t ngfn, void *data, int offset, int len,
423 ngpa = gfn_to_gpa(ngfn);
424 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
425 if (real_gfn == UNMAPPED_GVA)
428 real_gfn = gpa_to_gfn(real_gfn);
430 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
432 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
434 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
435 void *data, int offset, int len, u32 access)
437 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
438 data, offset, len, access);
442 * Load the pae pdptrs. Return true is they are all valid.
444 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
446 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
447 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
450 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
452 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
453 offset * sizeof(u64), sizeof(pdpte),
454 PFERR_USER_MASK|PFERR_WRITE_MASK);
459 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
460 if (is_present_gpte(pdpte[i]) &&
461 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
468 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
469 __set_bit(VCPU_EXREG_PDPTR,
470 (unsigned long *)&vcpu->arch.regs_avail);
471 __set_bit(VCPU_EXREG_PDPTR,
472 (unsigned long *)&vcpu->arch.regs_dirty);
477 EXPORT_SYMBOL_GPL(load_pdptrs);
479 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
481 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
487 if (is_long_mode(vcpu) || !is_pae(vcpu))
490 if (!test_bit(VCPU_EXREG_PDPTR,
491 (unsigned long *)&vcpu->arch.regs_avail))
494 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
495 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
496 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
497 PFERR_USER_MASK | PFERR_WRITE_MASK);
500 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
506 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
508 unsigned long old_cr0 = kvm_read_cr0(vcpu);
509 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
510 X86_CR0_CD | X86_CR0_NW;
515 if (cr0 & 0xffffffff00000000UL)
519 cr0 &= ~CR0_RESERVED_BITS;
521 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
524 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
527 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
529 if ((vcpu->arch.efer & EFER_LME)) {
534 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
539 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
544 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
547 kvm_x86_ops->set_cr0(vcpu, cr0);
549 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
550 kvm_clear_async_pf_completion_queue(vcpu);
551 kvm_async_pf_hash_reset(vcpu);
554 if ((cr0 ^ old_cr0) & update_bits)
555 kvm_mmu_reset_context(vcpu);
558 EXPORT_SYMBOL_GPL(kvm_set_cr0);
560 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
562 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
564 EXPORT_SYMBOL_GPL(kvm_lmsw);
566 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
568 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
569 !vcpu->guest_xcr0_loaded) {
570 /* kvm_set_xcr() also depends on this */
571 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
572 vcpu->guest_xcr0_loaded = 1;
576 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
578 if (vcpu->guest_xcr0_loaded) {
579 if (vcpu->arch.xcr0 != host_xcr0)
580 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
581 vcpu->guest_xcr0_loaded = 0;
585 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
589 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
590 if (index != XCR_XFEATURE_ENABLED_MASK)
593 if (!(xcr0 & XSTATE_FP))
595 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
597 if (xcr0 & ~host_xcr0)
599 kvm_put_guest_xcr0(vcpu);
600 vcpu->arch.xcr0 = xcr0;
604 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
606 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
607 __kvm_set_xcr(vcpu, index, xcr)) {
608 kvm_inject_gp(vcpu, 0);
613 EXPORT_SYMBOL_GPL(kvm_set_xcr);
615 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
617 unsigned long old_cr4 = kvm_read_cr4(vcpu);
618 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
619 X86_CR4_PAE | X86_CR4_SMEP;
620 if (cr4 & CR4_RESERVED_BITS)
623 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
626 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
629 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_RDWRGSFS))
632 if (is_long_mode(vcpu)) {
633 if (!(cr4 & X86_CR4_PAE))
635 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
636 && ((cr4 ^ old_cr4) & pdptr_bits)
637 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
641 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
642 if (!guest_cpuid_has_pcid(vcpu))
645 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
646 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
650 if (kvm_x86_ops->set_cr4(vcpu, cr4))
653 if (((cr4 ^ old_cr4) & pdptr_bits) ||
654 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
655 kvm_mmu_reset_context(vcpu);
657 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
658 kvm_update_cpuid(vcpu);
662 EXPORT_SYMBOL_GPL(kvm_set_cr4);
664 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
666 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
667 kvm_mmu_sync_roots(vcpu);
668 kvm_mmu_flush_tlb(vcpu);
672 if (is_long_mode(vcpu)) {
673 if (kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)) {
674 if (cr3 & CR3_PCID_ENABLED_RESERVED_BITS)
677 if (cr3 & CR3_L_MODE_RESERVED_BITS)
681 if (cr3 & CR3_PAE_RESERVED_BITS)
683 if (is_paging(vcpu) &&
684 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
688 * We don't check reserved bits in nonpae mode, because
689 * this isn't enforced, and VMware depends on this.
694 * Does the new cr3 value map to physical memory? (Note, we
695 * catch an invalid cr3 even in real-mode, because it would
696 * cause trouble later on when we turn on paging anyway.)
698 * A real CPU would silently accept an invalid cr3 and would
699 * attempt to use it - with largely undefined (and often hard
700 * to debug) behavior on the guest side.
702 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
704 vcpu->arch.cr3 = cr3;
705 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
706 vcpu->arch.mmu.new_cr3(vcpu);
709 EXPORT_SYMBOL_GPL(kvm_set_cr3);
711 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
713 if (cr8 & CR8_RESERVED_BITS)
715 if (irqchip_in_kernel(vcpu->kvm))
716 kvm_lapic_set_tpr(vcpu, cr8);
718 vcpu->arch.cr8 = cr8;
721 EXPORT_SYMBOL_GPL(kvm_set_cr8);
723 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
725 if (irqchip_in_kernel(vcpu->kvm))
726 return kvm_lapic_get_cr8(vcpu);
728 return vcpu->arch.cr8;
730 EXPORT_SYMBOL_GPL(kvm_get_cr8);
732 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
736 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
737 dr7 = vcpu->arch.guest_debug_dr7;
739 dr7 = vcpu->arch.dr7;
740 kvm_x86_ops->set_dr7(vcpu, dr7);
741 vcpu->arch.switch_db_regs = (dr7 & DR7_BP_EN_MASK);
744 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
748 vcpu->arch.db[dr] = val;
749 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
750 vcpu->arch.eff_db[dr] = val;
753 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
757 if (val & 0xffffffff00000000ULL)
759 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
762 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
766 if (val & 0xffffffff00000000ULL)
768 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
769 kvm_update_dr7(vcpu);
776 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
780 res = __kvm_set_dr(vcpu, dr, val);
782 kvm_queue_exception(vcpu, UD_VECTOR);
784 kvm_inject_gp(vcpu, 0);
788 EXPORT_SYMBOL_GPL(kvm_set_dr);
790 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
794 *val = vcpu->arch.db[dr];
797 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
801 *val = vcpu->arch.dr6;
804 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
808 *val = vcpu->arch.dr7;
815 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
817 if (_kvm_get_dr(vcpu, dr, val)) {
818 kvm_queue_exception(vcpu, UD_VECTOR);
823 EXPORT_SYMBOL_GPL(kvm_get_dr);
825 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
827 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
831 err = kvm_pmu_read_pmc(vcpu, ecx, &data);
834 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
835 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
838 EXPORT_SYMBOL_GPL(kvm_rdpmc);
841 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
842 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
844 * This list is modified at module load time to reflect the
845 * capabilities of the host cpu. This capabilities test skips MSRs that are
846 * kvm-specific. Those are put in the beginning of the list.
849 #define KVM_SAVE_MSRS_BEGIN 10
850 static u32 msrs_to_save[] = {
851 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
852 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
853 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
854 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
856 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
859 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
861 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
864 static unsigned num_msrs_to_save;
866 static const u32 emulated_msrs[] = {
868 MSR_IA32_TSCDEADLINE,
869 MSR_IA32_MISC_ENABLE,
874 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
876 if (efer & efer_reserved_bits)
879 if (efer & EFER_FFXSR) {
880 struct kvm_cpuid_entry2 *feat;
882 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
883 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
887 if (efer & EFER_SVME) {
888 struct kvm_cpuid_entry2 *feat;
890 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
891 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
897 EXPORT_SYMBOL_GPL(kvm_valid_efer);
899 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
901 u64 old_efer = vcpu->arch.efer;
903 if (!kvm_valid_efer(vcpu, efer))
907 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
911 efer |= vcpu->arch.efer & EFER_LMA;
913 kvm_x86_ops->set_efer(vcpu, efer);
915 /* Update reserved bits */
916 if ((efer ^ old_efer) & EFER_NX)
917 kvm_mmu_reset_context(vcpu);
922 void kvm_enable_efer_bits(u64 mask)
924 efer_reserved_bits &= ~mask;
926 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
929 * Writes msr value into into the appropriate "register".
930 * Returns 0 on success, non-0 otherwise.
931 * Assumes vcpu_load() was already called.
933 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
935 switch (msr->index) {
938 case MSR_KERNEL_GS_BASE:
941 if (is_noncanonical_address(msr->data))
944 case MSR_IA32_SYSENTER_EIP:
945 case MSR_IA32_SYSENTER_ESP:
947 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
948 * non-canonical address is written on Intel but not on
949 * AMD (which ignores the top 32-bits, because it does
950 * not implement 64-bit SYSENTER).
952 * 64-bit code should hence be able to write a non-canonical
953 * value on AMD. Making the address canonical ensures that
954 * vmentry does not fail on Intel after writing a non-canonical
955 * value, and that something deterministic happens if the guest
956 * invokes 64-bit SYSENTER.
958 msr->data = get_canonical(msr->data);
960 return kvm_x86_ops->set_msr(vcpu, msr);
962 EXPORT_SYMBOL_GPL(kvm_set_msr);
965 * Adapt set_msr() to msr_io()'s calling convention
967 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
973 msr.host_initiated = true;
974 return kvm_set_msr(vcpu, &msr);
978 struct pvclock_gtod_data {
981 struct { /* extract of a clocksource struct */
989 /* open coded 'struct timespec' */
990 u64 monotonic_time_snsec;
991 time_t monotonic_time_sec;
994 static struct pvclock_gtod_data pvclock_gtod_data;
996 static void update_pvclock_gtod(struct timekeeper *tk)
998 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1000 write_seqcount_begin(&vdata->seq);
1002 /* copy pvclock gtod data */
1003 vdata->clock.vclock_mode = tk->clock->archdata.vclock_mode;
1004 vdata->clock.cycle_last = tk->clock->cycle_last;
1005 vdata->clock.mask = tk->clock->mask;
1006 vdata->clock.mult = tk->mult;
1007 vdata->clock.shift = tk->shift;
1009 vdata->monotonic_time_sec = tk->xtime_sec
1010 + tk->wall_to_monotonic.tv_sec;
1011 vdata->monotonic_time_snsec = tk->xtime_nsec
1012 + (tk->wall_to_monotonic.tv_nsec
1014 while (vdata->monotonic_time_snsec >=
1015 (((u64)NSEC_PER_SEC) << tk->shift)) {
1016 vdata->monotonic_time_snsec -=
1017 ((u64)NSEC_PER_SEC) << tk->shift;
1018 vdata->monotonic_time_sec++;
1021 write_seqcount_end(&vdata->seq);
1026 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1030 struct pvclock_wall_clock wc;
1031 struct timespec boot;
1036 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1041 ++version; /* first time write, random junk */
1045 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1048 * The guest calculates current wall clock time by adding
1049 * system time (updated by kvm_guest_time_update below) to the
1050 * wall clock specified here. guest system time equals host
1051 * system time for us, thus we must fill in host boot time here.
1055 if (kvm->arch.kvmclock_offset) {
1056 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1057 boot = timespec_sub(boot, ts);
1059 wc.sec = boot.tv_sec;
1060 wc.nsec = boot.tv_nsec;
1061 wc.version = version;
1063 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1066 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1069 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1071 uint32_t quotient, remainder;
1073 /* Don't try to replace with do_div(), this one calculates
1074 * "(dividend << 32) / divisor" */
1076 : "=a" (quotient), "=d" (remainder)
1077 : "0" (0), "1" (dividend), "r" (divisor) );
1081 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1082 s8 *pshift, u32 *pmultiplier)
1089 tps64 = base_khz * 1000LL;
1090 scaled64 = scaled_khz * 1000LL;
1091 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1096 tps32 = (uint32_t)tps64;
1097 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1098 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1106 *pmultiplier = div_frac(scaled64, tps32);
1108 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1109 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1112 static inline u64 get_kernel_ns(void)
1116 WARN_ON(preemptible());
1118 monotonic_to_bootbased(&ts);
1119 return timespec_to_ns(&ts);
1122 #ifdef CONFIG_X86_64
1123 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1126 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1127 unsigned long max_tsc_khz;
1129 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1131 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1132 vcpu->arch.virtual_tsc_shift);
1135 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1137 u64 v = (u64)khz * (1000000 + ppm);
1142 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1144 u32 thresh_lo, thresh_hi;
1145 int use_scaling = 0;
1147 /* tsc_khz can be zero if TSC calibration fails */
1148 if (this_tsc_khz == 0)
1151 /* Compute a scale to convert nanoseconds in TSC cycles */
1152 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1153 &vcpu->arch.virtual_tsc_shift,
1154 &vcpu->arch.virtual_tsc_mult);
1155 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1158 * Compute the variation in TSC rate which is acceptable
1159 * within the range of tolerance and decide if the
1160 * rate being applied is within that bounds of the hardware
1161 * rate. If so, no scaling or compensation need be done.
1163 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1164 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1165 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1166 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1169 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1172 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1174 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1175 vcpu->arch.virtual_tsc_mult,
1176 vcpu->arch.virtual_tsc_shift);
1177 tsc += vcpu->arch.this_tsc_write;
1181 void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1183 #ifdef CONFIG_X86_64
1185 bool do_request = false;
1186 struct kvm_arch *ka = &vcpu->kvm->arch;
1187 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1189 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1190 atomic_read(&vcpu->kvm->online_vcpus));
1192 if (vcpus_matched && gtod->clock.vclock_mode == VCLOCK_TSC)
1193 if (!ka->use_master_clock)
1196 if (!vcpus_matched && ka->use_master_clock)
1200 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1202 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1203 atomic_read(&vcpu->kvm->online_vcpus),
1204 ka->use_master_clock, gtod->clock.vclock_mode);
1208 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1210 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1211 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1214 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1216 struct kvm *kvm = vcpu->kvm;
1217 u64 offset, ns, elapsed;
1218 unsigned long flags;
1221 u64 data = msr->data;
1223 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1224 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1225 ns = get_kernel_ns();
1226 elapsed = ns - kvm->arch.last_tsc_nsec;
1228 if (vcpu->arch.virtual_tsc_khz) {
1231 /* n.b - signed multiplication and division required */
1232 usdiff = data - kvm->arch.last_tsc_write;
1233 #ifdef CONFIG_X86_64
1234 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1236 /* do_div() only does unsigned */
1237 asm("1: idivl %[divisor]\n"
1238 "2: xor %%edx, %%edx\n"
1239 " movl $0, %[faulted]\n"
1241 ".section .fixup,\"ax\"\n"
1242 "4: movl $1, %[faulted]\n"
1246 _ASM_EXTABLE(1b, 4b)
1248 : "=A"(usdiff), [faulted] "=r" (faulted)
1249 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1252 do_div(elapsed, 1000);
1257 /* idivl overflow => difference is larger than USEC_PER_SEC */
1259 usdiff = USEC_PER_SEC;
1261 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1264 * Special case: TSC write with a small delta (1 second) of virtual
1265 * cycle time against real time is interpreted as an attempt to
1266 * synchronize the CPU.
1268 * For a reliable TSC, we can match TSC offsets, and for an unstable
1269 * TSC, we add elapsed time in this computation. We could let the
1270 * compensation code attempt to catch up if we fall behind, but
1271 * it's better to try to match offsets from the beginning.
1273 if (usdiff < USEC_PER_SEC &&
1274 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1275 if (!check_tsc_unstable()) {
1276 offset = kvm->arch.cur_tsc_offset;
1277 pr_debug("kvm: matched tsc offset for %llu\n", data);
1279 u64 delta = nsec_to_cycles(vcpu, elapsed);
1281 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1282 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1287 * We split periods of matched TSC writes into generations.
1288 * For each generation, we track the original measured
1289 * nanosecond time, offset, and write, so if TSCs are in
1290 * sync, we can match exact offset, and if not, we can match
1291 * exact software computation in compute_guest_tsc()
1293 * These values are tracked in kvm->arch.cur_xxx variables.
1295 kvm->arch.cur_tsc_generation++;
1296 kvm->arch.cur_tsc_nsec = ns;
1297 kvm->arch.cur_tsc_write = data;
1298 kvm->arch.cur_tsc_offset = offset;
1300 pr_debug("kvm: new tsc generation %u, clock %llu\n",
1301 kvm->arch.cur_tsc_generation, data);
1305 * We also track th most recent recorded KHZ, write and time to
1306 * allow the matching interval to be extended at each write.
1308 kvm->arch.last_tsc_nsec = ns;
1309 kvm->arch.last_tsc_write = data;
1310 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1312 /* Reset of TSC must disable overshoot protection below */
1313 vcpu->arch.hv_clock.tsc_timestamp = 0;
1314 vcpu->arch.last_guest_tsc = data;
1316 /* Keep track of which generation this VCPU has synchronized to */
1317 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1318 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1319 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1321 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1322 update_ia32_tsc_adjust_msr(vcpu, offset);
1323 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1324 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1326 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1328 kvm->arch.nr_vcpus_matched_tsc++;
1330 kvm->arch.nr_vcpus_matched_tsc = 0;
1332 kvm_track_tsc_matching(vcpu);
1333 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1336 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1338 #ifdef CONFIG_X86_64
1340 static cycle_t read_tsc(void)
1346 * Empirically, a fence (of type that depends on the CPU)
1347 * before rdtsc is enough to ensure that rdtsc is ordered
1348 * with respect to loads. The various CPU manuals are unclear
1349 * as to whether rdtsc can be reordered with later loads,
1350 * but no one has ever seen it happen.
1353 ret = (cycle_t)vget_cycles();
1355 last = pvclock_gtod_data.clock.cycle_last;
1357 if (likely(ret >= last))
1361 * GCC likes to generate cmov here, but this branch is extremely
1362 * predictable (it's just a funciton of time and the likely is
1363 * very likely) and there's a data dependence, so force GCC
1364 * to generate a branch instead. I don't barrier() because
1365 * we don't actually need a barrier, and if this function
1366 * ever gets inlined it will generate worse code.
1372 static inline u64 vgettsc(cycle_t *cycle_now)
1375 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1377 *cycle_now = read_tsc();
1379 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1380 return v * gtod->clock.mult;
1383 static int do_monotonic(struct timespec *ts, cycle_t *cycle_now)
1388 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1392 seq = read_seqcount_begin(>od->seq);
1393 mode = gtod->clock.vclock_mode;
1394 ts->tv_sec = gtod->monotonic_time_sec;
1395 ns = gtod->monotonic_time_snsec;
1396 ns += vgettsc(cycle_now);
1397 ns >>= gtod->clock.shift;
1398 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1399 timespec_add_ns(ts, ns);
1404 /* returns true if host is using tsc clocksource */
1405 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1409 /* checked again under seqlock below */
1410 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1413 if (do_monotonic(&ts, cycle_now) != VCLOCK_TSC)
1416 monotonic_to_bootbased(&ts);
1417 *kernel_ns = timespec_to_ns(&ts);
1425 * Assuming a stable TSC across physical CPUS, and a stable TSC
1426 * across virtual CPUs, the following condition is possible.
1427 * Each numbered line represents an event visible to both
1428 * CPUs at the next numbered event.
1430 * "timespecX" represents host monotonic time. "tscX" represents
1433 * VCPU0 on CPU0 | VCPU1 on CPU1
1435 * 1. read timespec0,tsc0
1436 * 2. | timespec1 = timespec0 + N
1438 * 3. transition to guest | transition to guest
1439 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1440 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1441 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1443 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1446 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1448 * - 0 < N - M => M < N
1450 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1451 * always the case (the difference between two distinct xtime instances
1452 * might be smaller then the difference between corresponding TSC reads,
1453 * when updating guest vcpus pvclock areas).
1455 * To avoid that problem, do not allow visibility of distinct
1456 * system_timestamp/tsc_timestamp values simultaneously: use a master
1457 * copy of host monotonic time values. Update that master copy
1460 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1464 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1466 #ifdef CONFIG_X86_64
1467 struct kvm_arch *ka = &kvm->arch;
1469 bool host_tsc_clocksource, vcpus_matched;
1471 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1472 atomic_read(&kvm->online_vcpus));
1475 * If the host uses TSC clock, then passthrough TSC as stable
1478 host_tsc_clocksource = kvm_get_time_and_clockread(
1479 &ka->master_kernel_ns,
1480 &ka->master_cycle_now);
1482 ka->use_master_clock = host_tsc_clocksource & vcpus_matched;
1484 if (ka->use_master_clock)
1485 atomic_set(&kvm_guest_has_master_clock, 1);
1487 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1488 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1493 static int kvm_guest_time_update(struct kvm_vcpu *v)
1495 unsigned long flags, this_tsc_khz;
1496 struct kvm_vcpu_arch *vcpu = &v->arch;
1497 struct kvm_arch *ka = &v->kvm->arch;
1498 s64 kernel_ns, max_kernel_ns;
1499 u64 tsc_timestamp, host_tsc;
1500 struct pvclock_vcpu_time_info guest_hv_clock;
1502 bool use_master_clock;
1508 * If the host uses TSC clock, then passthrough TSC as stable
1511 spin_lock(&ka->pvclock_gtod_sync_lock);
1512 use_master_clock = ka->use_master_clock;
1513 if (use_master_clock) {
1514 host_tsc = ka->master_cycle_now;
1515 kernel_ns = ka->master_kernel_ns;
1517 spin_unlock(&ka->pvclock_gtod_sync_lock);
1519 /* Keep irq disabled to prevent changes to the clock */
1520 local_irq_save(flags);
1521 this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
1522 if (unlikely(this_tsc_khz == 0)) {
1523 local_irq_restore(flags);
1524 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1527 if (!use_master_clock) {
1528 host_tsc = native_read_tsc();
1529 kernel_ns = get_kernel_ns();
1532 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1535 * We may have to catch up the TSC to match elapsed wall clock
1536 * time for two reasons, even if kvmclock is used.
1537 * 1) CPU could have been running below the maximum TSC rate
1538 * 2) Broken TSC compensation resets the base at each VCPU
1539 * entry to avoid unknown leaps of TSC even when running
1540 * again on the same CPU. This may cause apparent elapsed
1541 * time to disappear, and the guest to stand still or run
1544 if (vcpu->tsc_catchup) {
1545 u64 tsc = compute_guest_tsc(v, kernel_ns);
1546 if (tsc > tsc_timestamp) {
1547 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1548 tsc_timestamp = tsc;
1552 local_irq_restore(flags);
1554 if (!vcpu->pv_time_enabled)
1558 * Time as measured by the TSC may go backwards when resetting the base
1559 * tsc_timestamp. The reason for this is that the TSC resolution is
1560 * higher than the resolution of the other clock scales. Thus, many
1561 * possible measurments of the TSC correspond to one measurement of any
1562 * other clock, and so a spread of values is possible. This is not a
1563 * problem for the computation of the nanosecond clock; with TSC rates
1564 * around 1GHZ, there can only be a few cycles which correspond to one
1565 * nanosecond value, and any path through this code will inevitably
1566 * take longer than that. However, with the kernel_ns value itself,
1567 * the precision may be much lower, down to HZ granularity. If the
1568 * first sampling of TSC against kernel_ns ends in the low part of the
1569 * range, and the second in the high end of the range, we can get:
1571 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1573 * As the sampling errors potentially range in the thousands of cycles,
1574 * it is possible such a time value has already been observed by the
1575 * guest. To protect against this, we must compute the system time as
1576 * observed by the guest and ensure the new system time is greater.
1579 if (vcpu->hv_clock.tsc_timestamp) {
1580 max_kernel_ns = vcpu->last_guest_tsc -
1581 vcpu->hv_clock.tsc_timestamp;
1582 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1583 vcpu->hv_clock.tsc_to_system_mul,
1584 vcpu->hv_clock.tsc_shift);
1585 max_kernel_ns += vcpu->last_kernel_ns;
1588 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1589 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1590 &vcpu->hv_clock.tsc_shift,
1591 &vcpu->hv_clock.tsc_to_system_mul);
1592 vcpu->hw_tsc_khz = this_tsc_khz;
1595 /* with a master <monotonic time, tsc value> tuple,
1596 * pvclock clock reads always increase at the (scaled) rate
1597 * of guest TSC - no need to deal with sampling errors.
1599 if (!use_master_clock) {
1600 if (max_kernel_ns > kernel_ns)
1601 kernel_ns = max_kernel_ns;
1603 /* With all the info we got, fill in the values */
1604 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1605 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1606 vcpu->last_kernel_ns = kernel_ns;
1607 vcpu->last_guest_tsc = tsc_timestamp;
1610 * The interface expects us to write an even number signaling that the
1611 * update is finished. Since the guest won't see the intermediate
1612 * state, we just increase by 2 at the end.
1614 vcpu->hv_clock.version += 2;
1616 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1617 &guest_hv_clock, sizeof(guest_hv_clock))))
1620 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1621 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1623 if (vcpu->pvclock_set_guest_stopped_request) {
1624 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1625 vcpu->pvclock_set_guest_stopped_request = false;
1628 /* If the host uses TSC clocksource, then it is stable */
1629 if (use_master_clock)
1630 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1632 vcpu->hv_clock.flags = pvclock_flags;
1634 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1636 sizeof(vcpu->hv_clock));
1640 static bool msr_mtrr_valid(unsigned msr)
1643 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1644 case MSR_MTRRfix64K_00000:
1645 case MSR_MTRRfix16K_80000:
1646 case MSR_MTRRfix16K_A0000:
1647 case MSR_MTRRfix4K_C0000:
1648 case MSR_MTRRfix4K_C8000:
1649 case MSR_MTRRfix4K_D0000:
1650 case MSR_MTRRfix4K_D8000:
1651 case MSR_MTRRfix4K_E0000:
1652 case MSR_MTRRfix4K_E8000:
1653 case MSR_MTRRfix4K_F0000:
1654 case MSR_MTRRfix4K_F8000:
1655 case MSR_MTRRdefType:
1656 case MSR_IA32_CR_PAT:
1664 static bool valid_pat_type(unsigned t)
1666 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1669 static bool valid_mtrr_type(unsigned t)
1671 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1674 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1678 if (!msr_mtrr_valid(msr))
1681 if (msr == MSR_IA32_CR_PAT) {
1682 for (i = 0; i < 8; i++)
1683 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1686 } else if (msr == MSR_MTRRdefType) {
1689 return valid_mtrr_type(data & 0xff);
1690 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1691 for (i = 0; i < 8 ; i++)
1692 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1697 /* variable MTRRs */
1698 return valid_mtrr_type(data & 0xff);
1701 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1703 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1705 if (!mtrr_valid(vcpu, msr, data))
1708 if (msr == MSR_MTRRdefType) {
1709 vcpu->arch.mtrr_state.def_type = data;
1710 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1711 } else if (msr == MSR_MTRRfix64K_00000)
1713 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1714 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1715 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1716 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1717 else if (msr == MSR_IA32_CR_PAT)
1718 vcpu->arch.pat = data;
1719 else { /* Variable MTRRs */
1720 int idx, is_mtrr_mask;
1723 idx = (msr - 0x200) / 2;
1724 is_mtrr_mask = msr - 0x200 - 2 * idx;
1727 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1730 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1734 kvm_mmu_reset_context(vcpu);
1738 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1740 u64 mcg_cap = vcpu->arch.mcg_cap;
1741 unsigned bank_num = mcg_cap & 0xff;
1744 case MSR_IA32_MCG_STATUS:
1745 vcpu->arch.mcg_status = data;
1747 case MSR_IA32_MCG_CTL:
1748 if (!(mcg_cap & MCG_CTL_P))
1750 if (data != 0 && data != ~(u64)0)
1752 vcpu->arch.mcg_ctl = data;
1755 if (msr >= MSR_IA32_MC0_CTL &&
1756 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1757 u32 offset = msr - MSR_IA32_MC0_CTL;
1758 /* only 0 or all 1s can be written to IA32_MCi_CTL
1759 * some Linux kernels though clear bit 10 in bank 4 to
1760 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1761 * this to avoid an uncatched #GP in the guest
1763 if ((offset & 0x3) == 0 &&
1764 data != 0 && (data | (1 << 10)) != ~(u64)0)
1766 vcpu->arch.mce_banks[offset] = data;
1774 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1776 struct kvm *kvm = vcpu->kvm;
1777 int lm = is_long_mode(vcpu);
1778 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1779 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1780 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1781 : kvm->arch.xen_hvm_config.blob_size_32;
1782 u32 page_num = data & ~PAGE_MASK;
1783 u64 page_addr = data & PAGE_MASK;
1788 if (page_num >= blob_size)
1791 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1796 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1805 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1807 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1810 static bool kvm_hv_msr_partition_wide(u32 msr)
1814 case HV_X64_MSR_GUEST_OS_ID:
1815 case HV_X64_MSR_HYPERCALL:
1823 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1825 struct kvm *kvm = vcpu->kvm;
1828 case HV_X64_MSR_GUEST_OS_ID:
1829 kvm->arch.hv_guest_os_id = data;
1830 /* setting guest os id to zero disables hypercall page */
1831 if (!kvm->arch.hv_guest_os_id)
1832 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1834 case HV_X64_MSR_HYPERCALL: {
1839 /* if guest os id is not set hypercall should remain disabled */
1840 if (!kvm->arch.hv_guest_os_id)
1842 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1843 kvm->arch.hv_hypercall = data;
1846 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1847 addr = gfn_to_hva(kvm, gfn);
1848 if (kvm_is_error_hva(addr))
1850 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1851 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1852 if (__copy_to_user((void __user *)addr, instructions, 4))
1854 kvm->arch.hv_hypercall = data;
1858 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1859 "data 0x%llx\n", msr, data);
1865 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1868 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1871 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1872 vcpu->arch.hv_vapic = data;
1875 addr = gfn_to_hva(vcpu->kvm, data >>
1876 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1877 if (kvm_is_error_hva(addr))
1879 if (__clear_user((void __user *)addr, PAGE_SIZE))
1881 vcpu->arch.hv_vapic = data;
1884 case HV_X64_MSR_EOI:
1885 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1886 case HV_X64_MSR_ICR:
1887 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1888 case HV_X64_MSR_TPR:
1889 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1891 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1892 "data 0x%llx\n", msr, data);
1899 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1901 gpa_t gpa = data & ~0x3f;
1903 /* Bits 2:5 are reserved, Should be zero */
1907 vcpu->arch.apf.msr_val = data;
1909 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1910 kvm_clear_async_pf_completion_queue(vcpu);
1911 kvm_async_pf_hash_reset(vcpu);
1915 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1919 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1920 kvm_async_pf_wakeup_all(vcpu);
1924 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1926 vcpu->arch.pv_time_enabled = false;
1929 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1933 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1936 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1937 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1938 vcpu->arch.st.accum_steal = delta;
1941 static void record_steal_time(struct kvm_vcpu *vcpu)
1943 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1946 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1947 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1950 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1951 vcpu->arch.st.steal.version += 2;
1952 vcpu->arch.st.accum_steal = 0;
1954 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1955 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1958 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1961 u32 msr = msr_info->index;
1962 u64 data = msr_info->data;
1965 case MSR_AMD64_NB_CFG:
1966 case MSR_IA32_UCODE_REV:
1967 case MSR_IA32_UCODE_WRITE:
1968 case MSR_VM_HSAVE_PA:
1969 case MSR_AMD64_PATCH_LOADER:
1970 case MSR_AMD64_BU_CFG2:
1974 return set_efer(vcpu, data);
1976 data &= ~(u64)0x40; /* ignore flush filter disable */
1977 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1978 data &= ~(u64)0x8; /* ignore TLB cache disable */
1980 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1985 case MSR_FAM10H_MMIO_CONF_BASE:
1987 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1992 case MSR_IA32_DEBUGCTLMSR:
1994 /* We support the non-activated case already */
1996 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1997 /* Values other than LBR and BTF are vendor-specific,
1998 thus reserved and should throw a #GP */
2001 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2004 case 0x200 ... 0x2ff:
2005 return set_msr_mtrr(vcpu, msr, data);
2006 case MSR_IA32_APICBASE:
2007 kvm_set_apic_base(vcpu, data);
2009 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2010 return kvm_x2apic_msr_write(vcpu, msr, data);
2011 case MSR_IA32_TSCDEADLINE:
2012 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2014 case MSR_IA32_TSC_ADJUST:
2015 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2016 if (!msr_info->host_initiated) {
2017 u64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2018 kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
2020 vcpu->arch.ia32_tsc_adjust_msr = data;
2023 case MSR_IA32_MISC_ENABLE:
2024 vcpu->arch.ia32_misc_enable_msr = data;
2026 case MSR_KVM_WALL_CLOCK_NEW:
2027 case MSR_KVM_WALL_CLOCK:
2028 vcpu->kvm->arch.wall_clock = data;
2029 kvm_write_wall_clock(vcpu->kvm, data);
2031 case MSR_KVM_SYSTEM_TIME_NEW:
2032 case MSR_KVM_SYSTEM_TIME: {
2034 kvmclock_reset(vcpu);
2036 vcpu->arch.time = data;
2037 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2039 /* we verify if the enable bit is set... */
2043 gpa_offset = data & ~(PAGE_MASK | 1);
2045 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2046 &vcpu->arch.pv_time, data & ~1ULL,
2047 sizeof(struct pvclock_vcpu_time_info)))
2048 vcpu->arch.pv_time_enabled = false;
2050 vcpu->arch.pv_time_enabled = true;
2054 case MSR_KVM_ASYNC_PF_EN:
2055 if (kvm_pv_enable_async_pf(vcpu, data))
2058 case MSR_KVM_STEAL_TIME:
2060 if (unlikely(!sched_info_on()))
2063 if (data & KVM_STEAL_RESERVED_MASK)
2066 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2067 data & KVM_STEAL_VALID_BITS,
2068 sizeof(struct kvm_steal_time)))
2071 vcpu->arch.st.msr_val = data;
2073 if (!(data & KVM_MSR_ENABLED))
2076 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2079 accumulate_steal_time(vcpu);
2082 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2085 case MSR_KVM_PV_EOI_EN:
2086 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2090 case MSR_IA32_MCG_CTL:
2091 case MSR_IA32_MCG_STATUS:
2092 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2093 return set_msr_mce(vcpu, msr, data);
2095 /* Performance counters are not protected by a CPUID bit,
2096 * so we should check all of them in the generic path for the sake of
2097 * cross vendor migration.
2098 * Writing a zero into the event select MSRs disables them,
2099 * which we perfectly emulate ;-). Any other value should be at least
2100 * reported, some guests depend on them.
2102 case MSR_K7_EVNTSEL0:
2103 case MSR_K7_EVNTSEL1:
2104 case MSR_K7_EVNTSEL2:
2105 case MSR_K7_EVNTSEL3:
2107 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2108 "0x%x data 0x%llx\n", msr, data);
2110 /* at least RHEL 4 unconditionally writes to the perfctr registers,
2111 * so we ignore writes to make it happy.
2113 case MSR_K7_PERFCTR0:
2114 case MSR_K7_PERFCTR1:
2115 case MSR_K7_PERFCTR2:
2116 case MSR_K7_PERFCTR3:
2117 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2118 "0x%x data 0x%llx\n", msr, data);
2120 case MSR_P6_PERFCTR0:
2121 case MSR_P6_PERFCTR1:
2123 case MSR_P6_EVNTSEL0:
2124 case MSR_P6_EVNTSEL1:
2125 if (kvm_pmu_msr(vcpu, msr))
2126 return kvm_pmu_set_msr(vcpu, msr_info);
2128 if (pr || data != 0)
2129 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2130 "0x%x data 0x%llx\n", msr, data);
2132 case MSR_K7_CLK_CTL:
2134 * Ignore all writes to this no longer documented MSR.
2135 * Writes are only relevant for old K7 processors,
2136 * all pre-dating SVM, but a recommended workaround from
2137 * AMD for these chips. It is possible to specify the
2138 * affected processor models on the command line, hence
2139 * the need to ignore the workaround.
2142 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2143 if (kvm_hv_msr_partition_wide(msr)) {
2145 mutex_lock(&vcpu->kvm->lock);
2146 r = set_msr_hyperv_pw(vcpu, msr, data);
2147 mutex_unlock(&vcpu->kvm->lock);
2150 return set_msr_hyperv(vcpu, msr, data);
2152 case MSR_IA32_BBL_CR_CTL3:
2153 /* Drop writes to this legacy MSR -- see rdmsr
2154 * counterpart for further detail.
2156 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2158 case MSR_AMD64_OSVW_ID_LENGTH:
2159 if (!guest_cpuid_has_osvw(vcpu))
2161 vcpu->arch.osvw.length = data;
2163 case MSR_AMD64_OSVW_STATUS:
2164 if (!guest_cpuid_has_osvw(vcpu))
2166 vcpu->arch.osvw.status = data;
2169 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2170 return xen_hvm_config(vcpu, data);
2171 if (kvm_pmu_msr(vcpu, msr))
2172 return kvm_pmu_set_msr(vcpu, msr_info);
2174 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2178 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2185 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2189 * Reads an msr value (of 'msr_index') into 'pdata'.
2190 * Returns 0 on success, non-0 otherwise.
2191 * Assumes vcpu_load() was already called.
2193 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2195 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
2198 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2200 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2202 if (!msr_mtrr_valid(msr))
2205 if (msr == MSR_MTRRdefType)
2206 *pdata = vcpu->arch.mtrr_state.def_type +
2207 (vcpu->arch.mtrr_state.enabled << 10);
2208 else if (msr == MSR_MTRRfix64K_00000)
2210 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2211 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2212 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2213 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2214 else if (msr == MSR_IA32_CR_PAT)
2215 *pdata = vcpu->arch.pat;
2216 else { /* Variable MTRRs */
2217 int idx, is_mtrr_mask;
2220 idx = (msr - 0x200) / 2;
2221 is_mtrr_mask = msr - 0x200 - 2 * idx;
2224 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2227 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2234 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2237 u64 mcg_cap = vcpu->arch.mcg_cap;
2238 unsigned bank_num = mcg_cap & 0xff;
2241 case MSR_IA32_P5_MC_ADDR:
2242 case MSR_IA32_P5_MC_TYPE:
2245 case MSR_IA32_MCG_CAP:
2246 data = vcpu->arch.mcg_cap;
2248 case MSR_IA32_MCG_CTL:
2249 if (!(mcg_cap & MCG_CTL_P))
2251 data = vcpu->arch.mcg_ctl;
2253 case MSR_IA32_MCG_STATUS:
2254 data = vcpu->arch.mcg_status;
2257 if (msr >= MSR_IA32_MC0_CTL &&
2258 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
2259 u32 offset = msr - MSR_IA32_MC0_CTL;
2260 data = vcpu->arch.mce_banks[offset];
2269 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2272 struct kvm *kvm = vcpu->kvm;
2275 case HV_X64_MSR_GUEST_OS_ID:
2276 data = kvm->arch.hv_guest_os_id;
2278 case HV_X64_MSR_HYPERCALL:
2279 data = kvm->arch.hv_hypercall;
2282 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2290 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2295 case HV_X64_MSR_VP_INDEX: {
2298 kvm_for_each_vcpu(r, v, vcpu->kvm)
2303 case HV_X64_MSR_EOI:
2304 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2305 case HV_X64_MSR_ICR:
2306 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2307 case HV_X64_MSR_TPR:
2308 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
2309 case HV_X64_MSR_APIC_ASSIST_PAGE:
2310 data = vcpu->arch.hv_vapic;
2313 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2320 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2325 case MSR_IA32_PLATFORM_ID:
2326 case MSR_IA32_EBL_CR_POWERON:
2327 case MSR_IA32_DEBUGCTLMSR:
2328 case MSR_IA32_LASTBRANCHFROMIP:
2329 case MSR_IA32_LASTBRANCHTOIP:
2330 case MSR_IA32_LASTINTFROMIP:
2331 case MSR_IA32_LASTINTTOIP:
2334 case MSR_VM_HSAVE_PA:
2335 case MSR_K7_EVNTSEL0:
2336 case MSR_K7_PERFCTR0:
2337 case MSR_K8_INT_PENDING_MSG:
2338 case MSR_AMD64_NB_CFG:
2339 case MSR_FAM10H_MMIO_CONF_BASE:
2340 case MSR_AMD64_BU_CFG2:
2343 case MSR_P6_PERFCTR0:
2344 case MSR_P6_PERFCTR1:
2345 case MSR_P6_EVNTSEL0:
2346 case MSR_P6_EVNTSEL1:
2347 if (kvm_pmu_msr(vcpu, msr))
2348 return kvm_pmu_get_msr(vcpu, msr, pdata);
2351 case MSR_IA32_UCODE_REV:
2352 data = 0x100000000ULL;
2355 data = 0x500 | KVM_NR_VAR_MTRR;
2357 case 0x200 ... 0x2ff:
2358 return get_msr_mtrr(vcpu, msr, pdata);
2359 case 0xcd: /* fsb frequency */
2363 * MSR_EBC_FREQUENCY_ID
2364 * Conservative value valid for even the basic CPU models.
2365 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2366 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2367 * and 266MHz for model 3, or 4. Set Core Clock
2368 * Frequency to System Bus Frequency Ratio to 1 (bits
2369 * 31:24) even though these are only valid for CPU
2370 * models > 2, however guests may end up dividing or
2371 * multiplying by zero otherwise.
2373 case MSR_EBC_FREQUENCY_ID:
2376 case MSR_IA32_APICBASE:
2377 data = kvm_get_apic_base(vcpu);
2379 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2380 return kvm_x2apic_msr_read(vcpu, msr, pdata);
2382 case MSR_IA32_TSCDEADLINE:
2383 data = kvm_get_lapic_tscdeadline_msr(vcpu);
2385 case MSR_IA32_TSC_ADJUST:
2386 data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2388 case MSR_IA32_MISC_ENABLE:
2389 data = vcpu->arch.ia32_misc_enable_msr;
2391 case MSR_IA32_PERF_STATUS:
2392 /* TSC increment by tick */
2394 /* CPU multiplier */
2395 data |= (((uint64_t)4ULL) << 40);
2398 data = vcpu->arch.efer;
2400 case MSR_KVM_WALL_CLOCK:
2401 case MSR_KVM_WALL_CLOCK_NEW:
2402 data = vcpu->kvm->arch.wall_clock;
2404 case MSR_KVM_SYSTEM_TIME:
2405 case MSR_KVM_SYSTEM_TIME_NEW:
2406 data = vcpu->arch.time;
2408 case MSR_KVM_ASYNC_PF_EN:
2409 data = vcpu->arch.apf.msr_val;
2411 case MSR_KVM_STEAL_TIME:
2412 data = vcpu->arch.st.msr_val;
2414 case MSR_KVM_PV_EOI_EN:
2415 data = vcpu->arch.pv_eoi.msr_val;
2417 case MSR_IA32_P5_MC_ADDR:
2418 case MSR_IA32_P5_MC_TYPE:
2419 case MSR_IA32_MCG_CAP:
2420 case MSR_IA32_MCG_CTL:
2421 case MSR_IA32_MCG_STATUS:
2422 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2423 return get_msr_mce(vcpu, msr, pdata);
2424 case MSR_K7_CLK_CTL:
2426 * Provide expected ramp-up count for K7. All other
2427 * are set to zero, indicating minimum divisors for
2430 * This prevents guest kernels on AMD host with CPU
2431 * type 6, model 8 and higher from exploding due to
2432 * the rdmsr failing.
2436 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2437 if (kvm_hv_msr_partition_wide(msr)) {
2439 mutex_lock(&vcpu->kvm->lock);
2440 r = get_msr_hyperv_pw(vcpu, msr, pdata);
2441 mutex_unlock(&vcpu->kvm->lock);
2444 return get_msr_hyperv(vcpu, msr, pdata);
2446 case MSR_IA32_BBL_CR_CTL3:
2447 /* This legacy MSR exists but isn't fully documented in current
2448 * silicon. It is however accessed by winxp in very narrow
2449 * scenarios where it sets bit #19, itself documented as
2450 * a "reserved" bit. Best effort attempt to source coherent
2451 * read data here should the balance of the register be
2452 * interpreted by the guest:
2454 * L2 cache control register 3: 64GB range, 256KB size,
2455 * enabled, latency 0x1, configured
2459 case MSR_AMD64_OSVW_ID_LENGTH:
2460 if (!guest_cpuid_has_osvw(vcpu))
2462 data = vcpu->arch.osvw.length;
2464 case MSR_AMD64_OSVW_STATUS:
2465 if (!guest_cpuid_has_osvw(vcpu))
2467 data = vcpu->arch.osvw.status;
2470 if (kvm_pmu_msr(vcpu, msr))
2471 return kvm_pmu_get_msr(vcpu, msr, pdata);
2473 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2476 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2484 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2487 * Read or write a bunch of msrs. All parameters are kernel addresses.
2489 * @return number of msrs set successfully.
2491 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2492 struct kvm_msr_entry *entries,
2493 int (*do_msr)(struct kvm_vcpu *vcpu,
2494 unsigned index, u64 *data))
2498 idx = srcu_read_lock(&vcpu->kvm->srcu);
2499 for (i = 0; i < msrs->nmsrs; ++i)
2500 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2502 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2508 * Read or write a bunch of msrs. Parameters are user addresses.
2510 * @return number of msrs set successfully.
2512 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2513 int (*do_msr)(struct kvm_vcpu *vcpu,
2514 unsigned index, u64 *data),
2517 struct kvm_msrs msrs;
2518 struct kvm_msr_entry *entries;
2523 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2527 if (msrs.nmsrs >= MAX_IO_MSRS)
2530 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2531 entries = memdup_user(user_msrs->entries, size);
2532 if (IS_ERR(entries)) {
2533 r = PTR_ERR(entries);
2537 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2542 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2553 int kvm_dev_ioctl_check_extension(long ext)
2558 case KVM_CAP_IRQCHIP:
2560 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2561 case KVM_CAP_SET_TSS_ADDR:
2562 case KVM_CAP_EXT_CPUID:
2563 case KVM_CAP_CLOCKSOURCE:
2565 case KVM_CAP_NOP_IO_DELAY:
2566 case KVM_CAP_MP_STATE:
2567 case KVM_CAP_SYNC_MMU:
2568 case KVM_CAP_USER_NMI:
2569 case KVM_CAP_REINJECT_CONTROL:
2570 case KVM_CAP_IRQ_INJECT_STATUS:
2572 case KVM_CAP_IOEVENTFD:
2574 case KVM_CAP_PIT_STATE2:
2575 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2576 case KVM_CAP_XEN_HVM:
2577 case KVM_CAP_ADJUST_CLOCK:
2578 case KVM_CAP_VCPU_EVENTS:
2579 case KVM_CAP_HYPERV:
2580 case KVM_CAP_HYPERV_VAPIC:
2581 case KVM_CAP_HYPERV_SPIN:
2582 case KVM_CAP_PCI_SEGMENT:
2583 case KVM_CAP_DEBUGREGS:
2584 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2586 case KVM_CAP_ASYNC_PF:
2587 case KVM_CAP_GET_TSC_KHZ:
2588 case KVM_CAP_KVMCLOCK_CTRL:
2589 case KVM_CAP_READONLY_MEM:
2590 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2591 case KVM_CAP_ASSIGN_DEV_IRQ:
2592 case KVM_CAP_PCI_2_3:
2596 case KVM_CAP_COALESCED_MMIO:
2597 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2600 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2602 case KVM_CAP_NR_VCPUS:
2603 r = KVM_SOFT_MAX_VCPUS;
2605 case KVM_CAP_MAX_VCPUS:
2608 case KVM_CAP_NR_MEMSLOTS:
2609 r = KVM_USER_MEM_SLOTS;
2611 case KVM_CAP_PV_MMU: /* obsolete */
2614 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2616 r = iommu_present(&pci_bus_type);
2620 r = KVM_MAX_MCE_BANKS;
2625 case KVM_CAP_TSC_CONTROL:
2626 r = kvm_has_tsc_control;
2628 case KVM_CAP_TSC_DEADLINE_TIMER:
2629 r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2639 long kvm_arch_dev_ioctl(struct file *filp,
2640 unsigned int ioctl, unsigned long arg)
2642 void __user *argp = (void __user *)arg;
2646 case KVM_GET_MSR_INDEX_LIST: {
2647 struct kvm_msr_list __user *user_msr_list = argp;
2648 struct kvm_msr_list msr_list;
2652 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2655 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2656 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2659 if (n < msr_list.nmsrs)
2662 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2663 num_msrs_to_save * sizeof(u32)))
2665 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2667 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2672 case KVM_GET_SUPPORTED_CPUID: {
2673 struct kvm_cpuid2 __user *cpuid_arg = argp;
2674 struct kvm_cpuid2 cpuid;
2677 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2679 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2680 cpuid_arg->entries);
2685 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2690 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2693 mce_cap = KVM_MCE_CAP_SUPPORTED;
2695 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2707 static void wbinvd_ipi(void *garbage)
2712 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2714 return vcpu->kvm->arch.iommu_domain &&
2715 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2718 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2720 /* Address WBINVD may be executed by guest */
2721 if (need_emulate_wbinvd(vcpu)) {
2722 if (kvm_x86_ops->has_wbinvd_exit())
2723 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2724 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2725 smp_call_function_single(vcpu->cpu,
2726 wbinvd_ipi, NULL, 1);
2729 kvm_x86_ops->vcpu_load(vcpu, cpu);
2731 /* Apply any externally detected TSC adjustments (due to suspend) */
2732 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2733 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2734 vcpu->arch.tsc_offset_adjustment = 0;
2735 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
2738 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2739 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2740 native_read_tsc() - vcpu->arch.last_host_tsc;
2742 mark_tsc_unstable("KVM discovered backwards TSC");
2743 if (check_tsc_unstable()) {
2744 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2745 vcpu->arch.last_guest_tsc);
2746 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2747 vcpu->arch.tsc_catchup = 1;
2750 * On a host with synchronized TSC, there is no need to update
2751 * kvmclock on vcpu->cpu migration
2753 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2754 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2755 if (vcpu->cpu != cpu)
2756 kvm_migrate_timers(vcpu);
2760 accumulate_steal_time(vcpu);
2761 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2764 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2766 kvm_x86_ops->vcpu_put(vcpu);
2767 kvm_put_guest_fpu(vcpu);
2768 vcpu->arch.last_host_tsc = native_read_tsc();
2771 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2772 struct kvm_lapic_state *s)
2774 kvm_x86_ops->sync_pir_to_irr(vcpu);
2775 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2780 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2781 struct kvm_lapic_state *s)
2783 kvm_apic_post_state_restore(vcpu, s);
2784 update_cr8_intercept(vcpu);
2789 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2790 struct kvm_interrupt *irq)
2792 if (irq->irq >= KVM_NR_INTERRUPTS)
2794 if (irqchip_in_kernel(vcpu->kvm))
2797 kvm_queue_interrupt(vcpu, irq->irq, false);
2798 kvm_make_request(KVM_REQ_EVENT, vcpu);
2803 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2805 kvm_inject_nmi(vcpu);
2810 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2811 struct kvm_tpr_access_ctl *tac)
2815 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2819 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2823 unsigned bank_num = mcg_cap & 0xff, bank;
2826 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2828 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2831 vcpu->arch.mcg_cap = mcg_cap;
2832 /* Init IA32_MCG_CTL to all 1s */
2833 if (mcg_cap & MCG_CTL_P)
2834 vcpu->arch.mcg_ctl = ~(u64)0;
2835 /* Init IA32_MCi_CTL to all 1s */
2836 for (bank = 0; bank < bank_num; bank++)
2837 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2842 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2843 struct kvm_x86_mce *mce)
2845 u64 mcg_cap = vcpu->arch.mcg_cap;
2846 unsigned bank_num = mcg_cap & 0xff;
2847 u64 *banks = vcpu->arch.mce_banks;
2849 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2852 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2853 * reporting is disabled
2855 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2856 vcpu->arch.mcg_ctl != ~(u64)0)
2858 banks += 4 * mce->bank;
2860 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2861 * reporting is disabled for the bank
2863 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2865 if (mce->status & MCI_STATUS_UC) {
2866 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2867 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2868 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2871 if (banks[1] & MCI_STATUS_VAL)
2872 mce->status |= MCI_STATUS_OVER;
2873 banks[2] = mce->addr;
2874 banks[3] = mce->misc;
2875 vcpu->arch.mcg_status = mce->mcg_status;
2876 banks[1] = mce->status;
2877 kvm_queue_exception(vcpu, MC_VECTOR);
2878 } else if (!(banks[1] & MCI_STATUS_VAL)
2879 || !(banks[1] & MCI_STATUS_UC)) {
2880 if (banks[1] & MCI_STATUS_VAL)
2881 mce->status |= MCI_STATUS_OVER;
2882 banks[2] = mce->addr;
2883 banks[3] = mce->misc;
2884 banks[1] = mce->status;
2886 banks[1] |= MCI_STATUS_OVER;
2890 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2891 struct kvm_vcpu_events *events)
2894 events->exception.injected =
2895 vcpu->arch.exception.pending &&
2896 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2897 events->exception.nr = vcpu->arch.exception.nr;
2898 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2899 events->exception.pad = 0;
2900 events->exception.error_code = vcpu->arch.exception.error_code;
2902 events->interrupt.injected =
2903 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2904 events->interrupt.nr = vcpu->arch.interrupt.nr;
2905 events->interrupt.soft = 0;
2906 events->interrupt.shadow =
2907 kvm_x86_ops->get_interrupt_shadow(vcpu,
2908 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2910 events->nmi.injected = vcpu->arch.nmi_injected;
2911 events->nmi.pending = vcpu->arch.nmi_pending != 0;
2912 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2913 events->nmi.pad = 0;
2915 events->sipi_vector = 0; /* never valid when reporting to user space */
2917 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2918 | KVM_VCPUEVENT_VALID_SHADOW);
2919 memset(&events->reserved, 0, sizeof(events->reserved));
2922 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2923 struct kvm_vcpu_events *events)
2925 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2926 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2927 | KVM_VCPUEVENT_VALID_SHADOW))
2931 vcpu->arch.exception.pending = events->exception.injected;
2932 vcpu->arch.exception.nr = events->exception.nr;
2933 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2934 vcpu->arch.exception.error_code = events->exception.error_code;
2936 vcpu->arch.interrupt.pending = events->interrupt.injected;
2937 vcpu->arch.interrupt.nr = events->interrupt.nr;
2938 vcpu->arch.interrupt.soft = events->interrupt.soft;
2939 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2940 kvm_x86_ops->set_interrupt_shadow(vcpu,
2941 events->interrupt.shadow);
2943 vcpu->arch.nmi_injected = events->nmi.injected;
2944 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2945 vcpu->arch.nmi_pending = events->nmi.pending;
2946 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2948 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
2949 kvm_vcpu_has_lapic(vcpu))
2950 vcpu->arch.apic->sipi_vector = events->sipi_vector;
2952 kvm_make_request(KVM_REQ_EVENT, vcpu);
2957 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2958 struct kvm_debugregs *dbgregs)
2960 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2961 dbgregs->dr6 = vcpu->arch.dr6;
2962 dbgregs->dr7 = vcpu->arch.dr7;
2964 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2967 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2968 struct kvm_debugregs *dbgregs)
2973 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2974 vcpu->arch.dr6 = dbgregs->dr6;
2975 vcpu->arch.dr7 = dbgregs->dr7;
2980 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2981 struct kvm_xsave *guest_xsave)
2984 memcpy(guest_xsave->region,
2985 &vcpu->arch.guest_fpu.state->xsave,
2988 memcpy(guest_xsave->region,
2989 &vcpu->arch.guest_fpu.state->fxsave,
2990 sizeof(struct i387_fxsave_struct));
2991 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2996 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2997 struct kvm_xsave *guest_xsave)
3000 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3003 memcpy(&vcpu->arch.guest_fpu.state->xsave,
3004 guest_xsave->region, xstate_size);
3006 if (xstate_bv & ~XSTATE_FPSSE)
3008 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
3009 guest_xsave->region, sizeof(struct i387_fxsave_struct));
3014 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3015 struct kvm_xcrs *guest_xcrs)
3017 if (!cpu_has_xsave) {
3018 guest_xcrs->nr_xcrs = 0;
3022 guest_xcrs->nr_xcrs = 1;
3023 guest_xcrs->flags = 0;
3024 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3025 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3028 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3029 struct kvm_xcrs *guest_xcrs)
3036 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3039 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3040 /* Only support XCR0 currently */
3041 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
3042 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3043 guest_xcrs->xcrs[0].value);
3052 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3053 * stopped by the hypervisor. This function will be called from the host only.
3054 * EINVAL is returned when the host attempts to set the flag for a guest that
3055 * does not support pv clocks.
3057 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3059 if (!vcpu->arch.pv_time_enabled)
3061 vcpu->arch.pvclock_set_guest_stopped_request = true;
3062 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3066 long kvm_arch_vcpu_ioctl(struct file *filp,
3067 unsigned int ioctl, unsigned long arg)
3069 struct kvm_vcpu *vcpu = filp->private_data;
3070 void __user *argp = (void __user *)arg;
3073 struct kvm_lapic_state *lapic;
3074 struct kvm_xsave *xsave;
3075 struct kvm_xcrs *xcrs;
3081 case KVM_GET_LAPIC: {
3083 if (!vcpu->arch.apic)
3085 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3090 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3094 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3099 case KVM_SET_LAPIC: {
3101 if (!vcpu->arch.apic)
3103 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3104 if (IS_ERR(u.lapic))
3105 return PTR_ERR(u.lapic);
3107 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3110 case KVM_INTERRUPT: {
3111 struct kvm_interrupt irq;
3114 if (copy_from_user(&irq, argp, sizeof irq))
3116 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3120 r = kvm_vcpu_ioctl_nmi(vcpu);
3123 case KVM_SET_CPUID: {
3124 struct kvm_cpuid __user *cpuid_arg = argp;
3125 struct kvm_cpuid cpuid;
3128 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3130 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3133 case KVM_SET_CPUID2: {
3134 struct kvm_cpuid2 __user *cpuid_arg = argp;
3135 struct kvm_cpuid2 cpuid;
3138 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3140 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3141 cpuid_arg->entries);
3144 case KVM_GET_CPUID2: {
3145 struct kvm_cpuid2 __user *cpuid_arg = argp;
3146 struct kvm_cpuid2 cpuid;
3149 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3151 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3152 cpuid_arg->entries);
3156 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3162 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3165 r = msr_io(vcpu, argp, do_set_msr, 0);
3167 case KVM_TPR_ACCESS_REPORTING: {
3168 struct kvm_tpr_access_ctl tac;
3171 if (copy_from_user(&tac, argp, sizeof tac))
3173 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3177 if (copy_to_user(argp, &tac, sizeof tac))
3182 case KVM_SET_VAPIC_ADDR: {
3183 struct kvm_vapic_addr va;
3186 if (!irqchip_in_kernel(vcpu->kvm))
3189 if (copy_from_user(&va, argp, sizeof va))
3191 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3194 case KVM_X86_SETUP_MCE: {
3198 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3200 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3203 case KVM_X86_SET_MCE: {
3204 struct kvm_x86_mce mce;
3207 if (copy_from_user(&mce, argp, sizeof mce))
3209 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3212 case KVM_GET_VCPU_EVENTS: {
3213 struct kvm_vcpu_events events;
3215 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3218 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3223 case KVM_SET_VCPU_EVENTS: {
3224 struct kvm_vcpu_events events;
3227 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3230 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3233 case KVM_GET_DEBUGREGS: {
3234 struct kvm_debugregs dbgregs;
3236 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3239 if (copy_to_user(argp, &dbgregs,
3240 sizeof(struct kvm_debugregs)))
3245 case KVM_SET_DEBUGREGS: {
3246 struct kvm_debugregs dbgregs;
3249 if (copy_from_user(&dbgregs, argp,
3250 sizeof(struct kvm_debugregs)))
3253 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3256 case KVM_GET_XSAVE: {
3257 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3262 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3265 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3270 case KVM_SET_XSAVE: {
3271 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3272 if (IS_ERR(u.xsave))
3273 return PTR_ERR(u.xsave);
3275 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3278 case KVM_GET_XCRS: {
3279 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3284 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3287 if (copy_to_user(argp, u.xcrs,
3288 sizeof(struct kvm_xcrs)))
3293 case KVM_SET_XCRS: {
3294 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3296 return PTR_ERR(u.xcrs);
3298 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3301 case KVM_SET_TSC_KHZ: {
3305 user_tsc_khz = (u32)arg;
3307 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3310 if (user_tsc_khz == 0)
3311 user_tsc_khz = tsc_khz;
3313 kvm_set_tsc_khz(vcpu, user_tsc_khz);
3318 case KVM_GET_TSC_KHZ: {
3319 r = vcpu->arch.virtual_tsc_khz;
3322 case KVM_KVMCLOCK_CTRL: {
3323 r = kvm_set_guest_paused(vcpu);
3334 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3336 return VM_FAULT_SIGBUS;
3339 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3343 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3345 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3349 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3352 kvm->arch.ept_identity_map_addr = ident_addr;
3356 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3357 u32 kvm_nr_mmu_pages)
3359 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3362 mutex_lock(&kvm->slots_lock);
3364 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3365 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3367 mutex_unlock(&kvm->slots_lock);
3371 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3373 return kvm->arch.n_max_mmu_pages;
3376 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3381 switch (chip->chip_id) {
3382 case KVM_IRQCHIP_PIC_MASTER:
3383 memcpy(&chip->chip.pic,
3384 &pic_irqchip(kvm)->pics[0],
3385 sizeof(struct kvm_pic_state));
3387 case KVM_IRQCHIP_PIC_SLAVE:
3388 memcpy(&chip->chip.pic,
3389 &pic_irqchip(kvm)->pics[1],
3390 sizeof(struct kvm_pic_state));
3392 case KVM_IRQCHIP_IOAPIC:
3393 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3402 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3407 switch (chip->chip_id) {
3408 case KVM_IRQCHIP_PIC_MASTER:
3409 spin_lock(&pic_irqchip(kvm)->lock);
3410 memcpy(&pic_irqchip(kvm)->pics[0],
3412 sizeof(struct kvm_pic_state));
3413 spin_unlock(&pic_irqchip(kvm)->lock);
3415 case KVM_IRQCHIP_PIC_SLAVE:
3416 spin_lock(&pic_irqchip(kvm)->lock);
3417 memcpy(&pic_irqchip(kvm)->pics[1],
3419 sizeof(struct kvm_pic_state));
3420 spin_unlock(&pic_irqchip(kvm)->lock);
3422 case KVM_IRQCHIP_IOAPIC:
3423 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3429 kvm_pic_update_irq(pic_irqchip(kvm));
3433 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3437 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3438 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3439 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3443 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3447 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3448 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3449 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3450 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3454 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3458 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3459 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3460 sizeof(ps->channels));
3461 ps->flags = kvm->arch.vpit->pit_state.flags;
3462 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3463 memset(&ps->reserved, 0, sizeof(ps->reserved));
3467 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3469 int r = 0, start = 0;
3470 u32 prev_legacy, cur_legacy;
3471 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3472 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3473 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3474 if (!prev_legacy && cur_legacy)
3476 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3477 sizeof(kvm->arch.vpit->pit_state.channels));
3478 kvm->arch.vpit->pit_state.flags = ps->flags;
3479 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3480 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3484 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3485 struct kvm_reinject_control *control)
3487 if (!kvm->arch.vpit)
3489 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3490 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3491 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3496 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3497 * @kvm: kvm instance
3498 * @log: slot id and address to which we copy the log
3500 * We need to keep it in mind that VCPU threads can write to the bitmap
3501 * concurrently. So, to avoid losing data, we keep the following order for
3504 * 1. Take a snapshot of the bit and clear it if needed.
3505 * 2. Write protect the corresponding page.
3506 * 3. Flush TLB's if needed.
3507 * 4. Copy the snapshot to the userspace.
3509 * Between 2 and 3, the guest may write to the page using the remaining TLB
3510 * entry. This is not a problem because the page will be reported dirty at
3511 * step 4 using the snapshot taken before and step 3 ensures that successive
3512 * writes will be logged for the next call.
3514 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3517 struct kvm_memory_slot *memslot;
3519 unsigned long *dirty_bitmap;
3520 unsigned long *dirty_bitmap_buffer;
3521 bool is_dirty = false;
3523 mutex_lock(&kvm->slots_lock);
3526 if (log->slot >= KVM_USER_MEM_SLOTS)
3529 memslot = id_to_memslot(kvm->memslots, log->slot);
3531 dirty_bitmap = memslot->dirty_bitmap;
3536 n = kvm_dirty_bitmap_bytes(memslot);
3538 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
3539 memset(dirty_bitmap_buffer, 0, n);
3541 spin_lock(&kvm->mmu_lock);
3543 for (i = 0; i < n / sizeof(long); i++) {
3547 if (!dirty_bitmap[i])
3552 mask = xchg(&dirty_bitmap[i], 0);
3553 dirty_bitmap_buffer[i] = mask;
3555 offset = i * BITS_PER_LONG;
3556 kvm_mmu_write_protect_pt_masked(kvm, memslot, offset, mask);
3559 kvm_flush_remote_tlbs(kvm);
3561 spin_unlock(&kvm->mmu_lock);
3564 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
3569 mutex_unlock(&kvm->slots_lock);
3573 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3576 if (!irqchip_in_kernel(kvm))
3579 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3580 irq_event->irq, irq_event->level,
3585 long kvm_arch_vm_ioctl(struct file *filp,
3586 unsigned int ioctl, unsigned long arg)
3588 struct kvm *kvm = filp->private_data;
3589 void __user *argp = (void __user *)arg;
3592 * This union makes it completely explicit to gcc-3.x
3593 * that these two variables' stack usage should be
3594 * combined, not added together.
3597 struct kvm_pit_state ps;
3598 struct kvm_pit_state2 ps2;
3599 struct kvm_pit_config pit_config;
3603 case KVM_SET_TSS_ADDR:
3604 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3606 case KVM_SET_IDENTITY_MAP_ADDR: {
3610 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3612 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3615 case KVM_SET_NR_MMU_PAGES:
3616 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3618 case KVM_GET_NR_MMU_PAGES:
3619 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3621 case KVM_CREATE_IRQCHIP: {
3622 struct kvm_pic *vpic;
3624 mutex_lock(&kvm->lock);
3627 goto create_irqchip_unlock;
3629 if (atomic_read(&kvm->online_vcpus))
3630 goto create_irqchip_unlock;
3632 vpic = kvm_create_pic(kvm);
3634 r = kvm_ioapic_init(kvm);
3636 mutex_lock(&kvm->slots_lock);
3637 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3639 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3641 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3643 mutex_unlock(&kvm->slots_lock);
3645 goto create_irqchip_unlock;
3648 goto create_irqchip_unlock;
3650 kvm->arch.vpic = vpic;
3652 r = kvm_setup_default_irq_routing(kvm);
3654 mutex_lock(&kvm->slots_lock);
3655 mutex_lock(&kvm->irq_lock);
3656 kvm_ioapic_destroy(kvm);
3657 kvm_destroy_pic(kvm);
3658 mutex_unlock(&kvm->irq_lock);
3659 mutex_unlock(&kvm->slots_lock);
3661 create_irqchip_unlock:
3662 mutex_unlock(&kvm->lock);
3665 case KVM_CREATE_PIT:
3666 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3668 case KVM_CREATE_PIT2:
3670 if (copy_from_user(&u.pit_config, argp,
3671 sizeof(struct kvm_pit_config)))
3674 mutex_lock(&kvm->slots_lock);
3677 goto create_pit_unlock;
3679 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3683 mutex_unlock(&kvm->slots_lock);
3685 case KVM_GET_IRQCHIP: {
3686 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3687 struct kvm_irqchip *chip;
3689 chip = memdup_user(argp, sizeof(*chip));
3696 if (!irqchip_in_kernel(kvm))
3697 goto get_irqchip_out;
3698 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3700 goto get_irqchip_out;
3702 if (copy_to_user(argp, chip, sizeof *chip))
3703 goto get_irqchip_out;
3709 case KVM_SET_IRQCHIP: {
3710 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3711 struct kvm_irqchip *chip;
3713 chip = memdup_user(argp, sizeof(*chip));
3720 if (!irqchip_in_kernel(kvm))
3721 goto set_irqchip_out;
3722 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3724 goto set_irqchip_out;
3732 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3735 if (!kvm->arch.vpit)
3737 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3741 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3748 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3751 if (!kvm->arch.vpit)
3753 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3756 case KVM_GET_PIT2: {
3758 if (!kvm->arch.vpit)
3760 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3764 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3769 case KVM_SET_PIT2: {
3771 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3774 if (!kvm->arch.vpit)
3776 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3779 case KVM_REINJECT_CONTROL: {
3780 struct kvm_reinject_control control;
3782 if (copy_from_user(&control, argp, sizeof(control)))
3784 r = kvm_vm_ioctl_reinject(kvm, &control);
3787 case KVM_XEN_HVM_CONFIG: {
3789 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3790 sizeof(struct kvm_xen_hvm_config)))
3793 if (kvm->arch.xen_hvm_config.flags)
3798 case KVM_SET_CLOCK: {
3799 struct kvm_clock_data user_ns;
3804 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3812 local_irq_disable();
3813 now_ns = get_kernel_ns();
3814 delta = user_ns.clock - now_ns;
3816 kvm->arch.kvmclock_offset = delta;
3819 case KVM_GET_CLOCK: {
3820 struct kvm_clock_data user_ns;
3823 local_irq_disable();
3824 now_ns = get_kernel_ns();
3825 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3828 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3831 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3844 static void kvm_init_msr_list(void)
3849 /* skip the first msrs in the list. KVM-specific */
3850 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3851 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3854 msrs_to_save[j] = msrs_to_save[i];
3857 num_msrs_to_save = j;
3860 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3868 if (!(vcpu->arch.apic &&
3869 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3870 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3881 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3888 if (!(vcpu->arch.apic &&
3889 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3890 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3892 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3902 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3903 struct kvm_segment *var, int seg)
3905 kvm_x86_ops->set_segment(vcpu, var, seg);
3908 void kvm_get_segment(struct kvm_vcpu *vcpu,
3909 struct kvm_segment *var, int seg)
3911 kvm_x86_ops->get_segment(vcpu, var, seg);
3914 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3917 struct x86_exception exception;
3919 BUG_ON(!mmu_is_nested(vcpu));
3921 /* NPT walks are always user-walks */
3922 access |= PFERR_USER_MASK;
3923 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3928 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3929 struct x86_exception *exception)
3931 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3932 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3935 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3936 struct x86_exception *exception)
3938 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3939 access |= PFERR_FETCH_MASK;
3940 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3943 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3944 struct x86_exception *exception)
3946 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3947 access |= PFERR_WRITE_MASK;
3948 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3951 /* uses this to access any guest's mapped memory without checking CPL */
3952 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3953 struct x86_exception *exception)
3955 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3958 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3959 struct kvm_vcpu *vcpu, u32 access,
3960 struct x86_exception *exception)
3963 int r = X86EMUL_CONTINUE;
3966 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3968 unsigned offset = addr & (PAGE_SIZE-1);
3969 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3972 if (gpa == UNMAPPED_GVA)
3973 return X86EMUL_PROPAGATE_FAULT;
3974 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3976 r = X86EMUL_IO_NEEDED;
3988 /* used for instruction fetching */
3989 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3990 gva_t addr, void *val, unsigned int bytes,
3991 struct x86_exception *exception)
3993 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3994 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3996 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3997 access | PFERR_FETCH_MASK,
4001 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4002 gva_t addr, void *val, unsigned int bytes,
4003 struct x86_exception *exception)
4005 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4006 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4008 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4011 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4013 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4014 gva_t addr, void *val, unsigned int bytes,
4015 struct x86_exception *exception)
4017 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4018 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4021 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4022 gva_t addr, void *val,
4024 struct x86_exception *exception)
4026 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4028 int r = X86EMUL_CONTINUE;
4031 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4034 unsigned offset = addr & (PAGE_SIZE-1);
4035 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4038 if (gpa == UNMAPPED_GVA)
4039 return X86EMUL_PROPAGATE_FAULT;
4040 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
4042 r = X86EMUL_IO_NEEDED;
4053 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4055 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4056 gpa_t *gpa, struct x86_exception *exception,
4059 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4060 | (write ? PFERR_WRITE_MASK : 0);
4062 if (vcpu_match_mmio_gva(vcpu, gva)
4063 && !permission_fault(vcpu->arch.walk_mmu, vcpu->arch.access, access)) {
4064 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4065 (gva & (PAGE_SIZE - 1));
4066 trace_vcpu_match_mmio(gva, *gpa, write, false);
4070 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4072 if (*gpa == UNMAPPED_GVA)
4075 /* For APIC access vmexit */
4076 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4079 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4080 trace_vcpu_match_mmio(gva, *gpa, write, true);
4087 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4088 const void *val, int bytes)
4092 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4095 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4099 struct read_write_emulator_ops {
4100 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4102 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4103 void *val, int bytes);
4104 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4105 int bytes, void *val);
4106 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4107 void *val, int bytes);
4111 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4113 if (vcpu->mmio_read_completed) {
4114 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4115 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4116 vcpu->mmio_read_completed = 0;
4123 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4124 void *val, int bytes)
4126 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4129 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4130 void *val, int bytes)
4132 return emulator_write_phys(vcpu, gpa, val, bytes);
4135 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4137 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4138 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4141 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4142 void *val, int bytes)
4144 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4145 return X86EMUL_IO_NEEDED;
4148 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4149 void *val, int bytes)
4151 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4153 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4154 return X86EMUL_CONTINUE;
4157 static const struct read_write_emulator_ops read_emultor = {
4158 .read_write_prepare = read_prepare,
4159 .read_write_emulate = read_emulate,
4160 .read_write_mmio = vcpu_mmio_read,
4161 .read_write_exit_mmio = read_exit_mmio,
4164 static const struct read_write_emulator_ops write_emultor = {
4165 .read_write_emulate = write_emulate,
4166 .read_write_mmio = write_mmio,
4167 .read_write_exit_mmio = write_exit_mmio,
4171 static int emulator_read_write_onepage(unsigned long addr, void *val,
4173 struct x86_exception *exception,
4174 struct kvm_vcpu *vcpu,
4175 const struct read_write_emulator_ops *ops)
4179 bool write = ops->write;
4180 struct kvm_mmio_fragment *frag;
4182 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4185 return X86EMUL_PROPAGATE_FAULT;
4187 /* For APIC access vmexit */
4191 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4192 return X86EMUL_CONTINUE;
4196 * Is this MMIO handled locally?
4198 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4199 if (handled == bytes)
4200 return X86EMUL_CONTINUE;
4206 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4207 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4211 return X86EMUL_CONTINUE;
4214 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4215 void *val, unsigned int bytes,
4216 struct x86_exception *exception,
4217 const struct read_write_emulator_ops *ops)
4219 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4223 if (ops->read_write_prepare &&
4224 ops->read_write_prepare(vcpu, val, bytes))
4225 return X86EMUL_CONTINUE;
4227 vcpu->mmio_nr_fragments = 0;
4229 /* Crossing a page boundary? */
4230 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4233 now = -addr & ~PAGE_MASK;
4234 rc = emulator_read_write_onepage(addr, val, now, exception,
4237 if (rc != X86EMUL_CONTINUE)
4244 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4246 if (rc != X86EMUL_CONTINUE)
4249 if (!vcpu->mmio_nr_fragments)
4252 gpa = vcpu->mmio_fragments[0].gpa;
4254 vcpu->mmio_needed = 1;
4255 vcpu->mmio_cur_fragment = 0;
4257 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4258 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4259 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4260 vcpu->run->mmio.phys_addr = gpa;
4262 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4265 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4269 struct x86_exception *exception)
4271 return emulator_read_write(ctxt, addr, val, bytes,
4272 exception, &read_emultor);
4275 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4279 struct x86_exception *exception)
4281 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4282 exception, &write_emultor);
4285 #define CMPXCHG_TYPE(t, ptr, old, new) \
4286 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4288 #ifdef CONFIG_X86_64
4289 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4291 # define CMPXCHG64(ptr, old, new) \
4292 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4295 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4300 struct x86_exception *exception)
4302 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4308 /* guests cmpxchg8b have to be emulated atomically */
4309 if (bytes > 8 || (bytes & (bytes - 1)))
4312 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4314 if (gpa == UNMAPPED_GVA ||
4315 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4318 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4321 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4322 if (is_error_page(page))
4325 kaddr = kmap_atomic(page);
4326 kaddr += offset_in_page(gpa);
4329 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4332 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4335 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4338 exchanged = CMPXCHG64(kaddr, old, new);
4343 kunmap_atomic(kaddr);
4344 kvm_release_page_dirty(page);
4347 return X86EMUL_CMPXCHG_FAILED;
4349 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4351 return X86EMUL_CONTINUE;
4354 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4356 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4359 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4361 /* TODO: String I/O for in kernel device */
4364 if (vcpu->arch.pio.in)
4365 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4366 vcpu->arch.pio.size, pd);
4368 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4369 vcpu->arch.pio.port, vcpu->arch.pio.size,
4374 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4375 unsigned short port, void *val,
4376 unsigned int count, bool in)
4378 trace_kvm_pio(!in, port, size, count);
4380 vcpu->arch.pio.port = port;
4381 vcpu->arch.pio.in = in;
4382 vcpu->arch.pio.count = count;
4383 vcpu->arch.pio.size = size;
4385 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4386 vcpu->arch.pio.count = 0;
4390 vcpu->run->exit_reason = KVM_EXIT_IO;
4391 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4392 vcpu->run->io.size = size;
4393 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4394 vcpu->run->io.count = count;
4395 vcpu->run->io.port = port;
4400 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4401 int size, unsigned short port, void *val,
4404 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4407 if (vcpu->arch.pio.count)
4410 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4413 memcpy(val, vcpu->arch.pio_data, size * count);
4414 vcpu->arch.pio.count = 0;
4421 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4422 int size, unsigned short port,
4423 const void *val, unsigned int count)
4425 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4427 memcpy(vcpu->arch.pio_data, val, size * count);
4428 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4431 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4433 return kvm_x86_ops->get_segment_base(vcpu, seg);
4436 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4438 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4441 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4443 if (!need_emulate_wbinvd(vcpu))
4444 return X86EMUL_CONTINUE;
4446 if (kvm_x86_ops->has_wbinvd_exit()) {
4447 int cpu = get_cpu();
4449 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4450 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4451 wbinvd_ipi, NULL, 1);
4453 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4456 return X86EMUL_CONTINUE;
4458 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4460 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4462 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4465 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4467 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4470 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4473 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4476 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4478 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4481 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4483 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4484 unsigned long value;
4488 value = kvm_read_cr0(vcpu);
4491 value = vcpu->arch.cr2;
4494 value = kvm_read_cr3(vcpu);
4497 value = kvm_read_cr4(vcpu);
4500 value = kvm_get_cr8(vcpu);
4503 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4510 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4512 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4517 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4520 vcpu->arch.cr2 = val;
4523 res = kvm_set_cr3(vcpu, val);
4526 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4529 res = kvm_set_cr8(vcpu, val);
4532 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4539 static void emulator_set_rflags(struct x86_emulate_ctxt *ctxt, ulong val)
4541 kvm_set_rflags(emul_to_vcpu(ctxt), val);
4544 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4546 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4549 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4551 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4554 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4556 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4559 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4561 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4564 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4566 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4569 static unsigned long emulator_get_cached_segment_base(
4570 struct x86_emulate_ctxt *ctxt, int seg)
4572 return get_segment_base(emul_to_vcpu(ctxt), seg);
4575 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4576 struct desc_struct *desc, u32 *base3,
4579 struct kvm_segment var;
4581 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4582 *selector = var.selector;
4585 memset(desc, 0, sizeof(*desc));
4591 set_desc_limit(desc, var.limit);
4592 set_desc_base(desc, (unsigned long)var.base);
4593 #ifdef CONFIG_X86_64
4595 *base3 = var.base >> 32;
4597 desc->type = var.type;
4599 desc->dpl = var.dpl;
4600 desc->p = var.present;
4601 desc->avl = var.avl;
4609 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4610 struct desc_struct *desc, u32 base3,
4613 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4614 struct kvm_segment var;
4616 var.selector = selector;
4617 var.base = get_desc_base(desc);
4618 #ifdef CONFIG_X86_64
4619 var.base |= ((u64)base3) << 32;
4621 var.limit = get_desc_limit(desc);
4623 var.limit = (var.limit << 12) | 0xfff;
4624 var.type = desc->type;
4625 var.present = desc->p;
4626 var.dpl = desc->dpl;
4631 var.avl = desc->avl;
4632 var.present = desc->p;
4633 var.unusable = !var.present;
4636 kvm_set_segment(vcpu, &var, seg);
4640 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4641 u32 msr_index, u64 *pdata)
4643 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4646 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4647 u32 msr_index, u64 data)
4649 struct msr_data msr;
4652 msr.index = msr_index;
4653 msr.host_initiated = false;
4654 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4657 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4658 u32 pmc, u64 *pdata)
4660 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4663 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4665 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4668 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4671 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4673 * CR0.TS may reference the host fpu state, not the guest fpu state,
4674 * so it may be clear at this point.
4679 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4684 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4685 struct x86_instruction_info *info,
4686 enum x86_intercept_stage stage)
4688 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4691 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4692 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4694 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4697 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4699 return kvm_register_read(emul_to_vcpu(ctxt), reg);
4702 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4704 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4707 static const struct x86_emulate_ops emulate_ops = {
4708 .read_gpr = emulator_read_gpr,
4709 .write_gpr = emulator_write_gpr,
4710 .read_std = kvm_read_guest_virt_system,
4711 .write_std = kvm_write_guest_virt_system,
4712 .fetch = kvm_fetch_guest_virt,
4713 .read_emulated = emulator_read_emulated,
4714 .write_emulated = emulator_write_emulated,
4715 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4716 .invlpg = emulator_invlpg,
4717 .pio_in_emulated = emulator_pio_in_emulated,
4718 .pio_out_emulated = emulator_pio_out_emulated,
4719 .get_segment = emulator_get_segment,
4720 .set_segment = emulator_set_segment,
4721 .get_cached_segment_base = emulator_get_cached_segment_base,
4722 .get_gdt = emulator_get_gdt,
4723 .get_idt = emulator_get_idt,
4724 .set_gdt = emulator_set_gdt,
4725 .set_idt = emulator_set_idt,
4726 .get_cr = emulator_get_cr,
4727 .set_cr = emulator_set_cr,
4728 .set_rflags = emulator_set_rflags,
4729 .cpl = emulator_get_cpl,
4730 .get_dr = emulator_get_dr,
4731 .set_dr = emulator_set_dr,
4732 .set_msr = emulator_set_msr,
4733 .get_msr = emulator_get_msr,
4734 .read_pmc = emulator_read_pmc,
4735 .halt = emulator_halt,
4736 .wbinvd = emulator_wbinvd,
4737 .fix_hypercall = emulator_fix_hypercall,
4738 .get_fpu = emulator_get_fpu,
4739 .put_fpu = emulator_put_fpu,
4740 .intercept = emulator_intercept,
4741 .get_cpuid = emulator_get_cpuid,
4744 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4746 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4748 * an sti; sti; sequence only disable interrupts for the first
4749 * instruction. So, if the last instruction, be it emulated or
4750 * not, left the system with the INT_STI flag enabled, it
4751 * means that the last instruction is an sti. We should not
4752 * leave the flag on in this case. The same goes for mov ss
4754 if (!(int_shadow & mask))
4755 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4758 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4760 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4761 if (ctxt->exception.vector == PF_VECTOR)
4762 kvm_propagate_fault(vcpu, &ctxt->exception);
4763 else if (ctxt->exception.error_code_valid)
4764 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4765 ctxt->exception.error_code);
4767 kvm_queue_exception(vcpu, ctxt->exception.vector);
4770 static void init_decode_cache(struct x86_emulate_ctxt *ctxt)
4772 memset(&ctxt->twobyte, 0,
4773 (void *)&ctxt->_regs - (void *)&ctxt->twobyte);
4775 ctxt->fetch.start = 0;
4776 ctxt->fetch.end = 0;
4777 ctxt->io_read.pos = 0;
4778 ctxt->io_read.end = 0;
4779 ctxt->mem_read.pos = 0;
4780 ctxt->mem_read.end = 0;
4783 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4785 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4788 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4790 ctxt->eflags = kvm_get_rflags(vcpu);
4791 ctxt->eip = kvm_rip_read(vcpu);
4792 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4793 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4794 cs_l ? X86EMUL_MODE_PROT64 :
4795 cs_db ? X86EMUL_MODE_PROT32 :
4796 X86EMUL_MODE_PROT16;
4797 ctxt->guest_mode = is_guest_mode(vcpu);
4799 init_decode_cache(ctxt);
4800 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4803 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4805 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4808 init_emulate_ctxt(vcpu);
4812 ctxt->_eip = ctxt->eip + inc_eip;
4813 ret = emulate_int_real(ctxt, irq);
4815 if (ret != X86EMUL_CONTINUE)
4816 return EMULATE_FAIL;
4818 ctxt->eip = ctxt->_eip;
4819 kvm_rip_write(vcpu, ctxt->eip);
4820 kvm_set_rflags(vcpu, ctxt->eflags);
4822 if (irq == NMI_VECTOR)
4823 vcpu->arch.nmi_pending = 0;
4825 vcpu->arch.interrupt.pending = false;
4827 return EMULATE_DONE;
4829 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4831 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4833 int r = EMULATE_DONE;
4835 ++vcpu->stat.insn_emulation_fail;
4836 trace_kvm_emulate_insn_failed(vcpu);
4837 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
4838 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4839 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4840 vcpu->run->internal.ndata = 0;
4843 kvm_queue_exception(vcpu, UD_VECTOR);
4848 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
4849 bool write_fault_to_shadow_pgtable,
4855 if (emulation_type & EMULTYPE_NO_REEXECUTE)
4858 if (!vcpu->arch.mmu.direct_map) {
4860 * Write permission should be allowed since only
4861 * write access need to be emulated.
4863 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4866 * If the mapping is invalid in guest, let cpu retry
4867 * it to generate fault.
4869 if (gpa == UNMAPPED_GVA)
4874 * Do not retry the unhandleable instruction if it faults on the
4875 * readonly host memory, otherwise it will goto a infinite loop:
4876 * retry instruction -> write #PF -> emulation fail -> retry
4877 * instruction -> ...
4879 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
4882 * If the instruction failed on the error pfn, it can not be fixed,
4883 * report the error to userspace.
4885 if (is_error_noslot_pfn(pfn))
4888 kvm_release_pfn_clean(pfn);
4890 /* The instructions are well-emulated on direct mmu. */
4891 if (vcpu->arch.mmu.direct_map) {
4892 unsigned int indirect_shadow_pages;
4894 spin_lock(&vcpu->kvm->mmu_lock);
4895 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
4896 spin_unlock(&vcpu->kvm->mmu_lock);
4898 if (indirect_shadow_pages)
4899 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4905 * if emulation was due to access to shadowed page table
4906 * and it failed try to unshadow page and re-enter the
4907 * guest to let CPU execute the instruction.
4909 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4912 * If the access faults on its page table, it can not
4913 * be fixed by unprotecting shadow page and it should
4914 * be reported to userspace.
4916 return !write_fault_to_shadow_pgtable;
4919 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
4920 unsigned long cr2, int emulation_type)
4922 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4923 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
4925 last_retry_eip = vcpu->arch.last_retry_eip;
4926 last_retry_addr = vcpu->arch.last_retry_addr;
4929 * If the emulation is caused by #PF and it is non-page_table
4930 * writing instruction, it means the VM-EXIT is caused by shadow
4931 * page protected, we can zap the shadow page and retry this
4932 * instruction directly.
4934 * Note: if the guest uses a non-page-table modifying instruction
4935 * on the PDE that points to the instruction, then we will unmap
4936 * the instruction and go to an infinite loop. So, we cache the
4937 * last retried eip and the last fault address, if we meet the eip
4938 * and the address again, we can break out of the potential infinite
4941 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
4943 if (!(emulation_type & EMULTYPE_RETRY))
4946 if (x86_page_table_writing_insn(ctxt))
4949 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
4952 vcpu->arch.last_retry_eip = ctxt->eip;
4953 vcpu->arch.last_retry_addr = cr2;
4955 if (!vcpu->arch.mmu.direct_map)
4956 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4958 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4963 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
4964 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
4966 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4973 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4974 bool writeback = true;
4975 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
4978 * Clear write_fault_to_shadow_pgtable here to ensure it is
4981 vcpu->arch.write_fault_to_shadow_pgtable = false;
4982 kvm_clear_exception_queue(vcpu);
4984 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4985 init_emulate_ctxt(vcpu);
4986 ctxt->interruptibility = 0;
4987 ctxt->have_exception = false;
4988 ctxt->perm_ok = false;
4990 ctxt->only_vendor_specific_insn
4991 = emulation_type & EMULTYPE_TRAP_UD;
4993 r = x86_decode_insn(ctxt, insn, insn_len);
4995 trace_kvm_emulate_insn_start(vcpu);
4996 ++vcpu->stat.insn_emulation;
4997 if (r != EMULATION_OK) {
4998 if (emulation_type & EMULTYPE_TRAP_UD)
4999 return EMULATE_FAIL;
5000 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5002 return EMULATE_DONE;
5003 if (emulation_type & EMULTYPE_SKIP)
5004 return EMULATE_FAIL;
5005 return handle_emulation_failure(vcpu);
5009 if (emulation_type & EMULTYPE_SKIP) {
5010 kvm_rip_write(vcpu, ctxt->_eip);
5011 return EMULATE_DONE;
5014 if (retry_instruction(ctxt, cr2, emulation_type))
5015 return EMULATE_DONE;
5017 /* this is needed for vmware backdoor interface to work since it
5018 changes registers values during IO operation */
5019 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5020 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5021 emulator_invalidate_register_cache(ctxt);
5025 r = x86_emulate_insn(ctxt);
5027 if (r == EMULATION_INTERCEPTED)
5028 return EMULATE_DONE;
5030 if (r == EMULATION_FAILED) {
5031 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5033 return EMULATE_DONE;
5035 return handle_emulation_failure(vcpu);
5038 if (ctxt->have_exception) {
5039 inject_emulated_exception(vcpu);
5041 } else if (vcpu->arch.pio.count) {
5042 if (!vcpu->arch.pio.in)
5043 vcpu->arch.pio.count = 0;
5046 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5048 r = EMULATE_DO_MMIO;
5049 } else if (vcpu->mmio_needed) {
5050 if (!vcpu->mmio_is_write)
5052 r = EMULATE_DO_MMIO;
5053 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5054 } else if (r == EMULATION_RESTART)
5060 toggle_interruptibility(vcpu, ctxt->interruptibility);
5061 kvm_set_rflags(vcpu, ctxt->eflags);
5062 kvm_make_request(KVM_REQ_EVENT, vcpu);
5063 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5064 kvm_rip_write(vcpu, ctxt->eip);
5066 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5070 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5072 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5074 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5075 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5076 size, port, &val, 1);
5077 /* do not return to emulator after return from userspace */
5078 vcpu->arch.pio.count = 0;
5081 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5083 static void tsc_bad(void *info)
5085 __this_cpu_write(cpu_tsc_khz, 0);
5088 static void tsc_khz_changed(void *data)
5090 struct cpufreq_freqs *freq = data;
5091 unsigned long khz = 0;
5095 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5096 khz = cpufreq_quick_get(raw_smp_processor_id());
5099 __this_cpu_write(cpu_tsc_khz, khz);
5102 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5105 struct cpufreq_freqs *freq = data;
5107 struct kvm_vcpu *vcpu;
5108 int i, send_ipi = 0;
5111 * We allow guests to temporarily run on slowing clocks,
5112 * provided we notify them after, or to run on accelerating
5113 * clocks, provided we notify them before. Thus time never
5116 * However, we have a problem. We can't atomically update
5117 * the frequency of a given CPU from this function; it is
5118 * merely a notifier, which can be called from any CPU.
5119 * Changing the TSC frequency at arbitrary points in time
5120 * requires a recomputation of local variables related to
5121 * the TSC for each VCPU. We must flag these local variables
5122 * to be updated and be sure the update takes place with the
5123 * new frequency before any guests proceed.
5125 * Unfortunately, the combination of hotplug CPU and frequency
5126 * change creates an intractable locking scenario; the order
5127 * of when these callouts happen is undefined with respect to
5128 * CPU hotplug, and they can race with each other. As such,
5129 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5130 * undefined; you can actually have a CPU frequency change take
5131 * place in between the computation of X and the setting of the
5132 * variable. To protect against this problem, all updates of
5133 * the per_cpu tsc_khz variable are done in an interrupt
5134 * protected IPI, and all callers wishing to update the value
5135 * must wait for a synchronous IPI to complete (which is trivial
5136 * if the caller is on the CPU already). This establishes the
5137 * necessary total order on variable updates.
5139 * Note that because a guest time update may take place
5140 * anytime after the setting of the VCPU's request bit, the
5141 * correct TSC value must be set before the request. However,
5142 * to ensure the update actually makes it to any guest which
5143 * starts running in hardware virtualization between the set
5144 * and the acquisition of the spinlock, we must also ping the
5145 * CPU after setting the request bit.
5149 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5151 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5154 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5156 raw_spin_lock(&kvm_lock);
5157 list_for_each_entry(kvm, &vm_list, vm_list) {
5158 kvm_for_each_vcpu(i, vcpu, kvm) {
5159 if (vcpu->cpu != freq->cpu)
5161 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5162 if (vcpu->cpu != smp_processor_id())
5166 raw_spin_unlock(&kvm_lock);
5168 if (freq->old < freq->new && send_ipi) {
5170 * We upscale the frequency. Must make the guest
5171 * doesn't see old kvmclock values while running with
5172 * the new frequency, otherwise we risk the guest sees
5173 * time go backwards.
5175 * In case we update the frequency for another cpu
5176 * (which might be in guest context) send an interrupt
5177 * to kick the cpu out of guest context. Next time
5178 * guest context is entered kvmclock will be updated,
5179 * so the guest will not see stale values.
5181 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5186 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5187 .notifier_call = kvmclock_cpufreq_notifier
5190 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5191 unsigned long action, void *hcpu)
5193 unsigned int cpu = (unsigned long)hcpu;
5197 case CPU_DOWN_FAILED:
5198 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5200 case CPU_DOWN_PREPARE:
5201 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5207 static struct notifier_block kvmclock_cpu_notifier_block = {
5208 .notifier_call = kvmclock_cpu_notifier,
5209 .priority = -INT_MAX
5212 static void kvm_timer_init(void)
5216 max_tsc_khz = tsc_khz;
5217 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5218 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5219 #ifdef CONFIG_CPU_FREQ
5220 struct cpufreq_policy policy;
5221 memset(&policy, 0, sizeof(policy));
5223 cpufreq_get_policy(&policy, cpu);
5224 if (policy.cpuinfo.max_freq)
5225 max_tsc_khz = policy.cpuinfo.max_freq;
5228 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5229 CPUFREQ_TRANSITION_NOTIFIER);
5231 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5232 for_each_online_cpu(cpu)
5233 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5236 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5238 int kvm_is_in_guest(void)
5240 return __this_cpu_read(current_vcpu) != NULL;
5243 static int kvm_is_user_mode(void)
5247 if (__this_cpu_read(current_vcpu))
5248 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5250 return user_mode != 0;
5253 static unsigned long kvm_get_guest_ip(void)
5255 unsigned long ip = 0;
5257 if (__this_cpu_read(current_vcpu))
5258 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5263 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5264 .is_in_guest = kvm_is_in_guest,
5265 .is_user_mode = kvm_is_user_mode,
5266 .get_guest_ip = kvm_get_guest_ip,
5269 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5271 __this_cpu_write(current_vcpu, vcpu);
5273 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5275 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5277 __this_cpu_write(current_vcpu, NULL);
5279 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5281 static void kvm_set_mmio_spte_mask(void)
5284 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5287 * Set the reserved bits and the present bit of an paging-structure
5288 * entry to generate page fault with PFER.RSV = 1.
5290 mask = ((1ull << (62 - maxphyaddr + 1)) - 1) << maxphyaddr;
5293 #ifdef CONFIG_X86_64
5295 * If reserved bit is not supported, clear the present bit to disable
5298 if (maxphyaddr == 52)
5302 kvm_mmu_set_mmio_spte_mask(mask);
5305 #ifdef CONFIG_X86_64
5306 static void pvclock_gtod_update_fn(struct work_struct *work)
5310 struct kvm_vcpu *vcpu;
5313 raw_spin_lock(&kvm_lock);
5314 list_for_each_entry(kvm, &vm_list, vm_list)
5315 kvm_for_each_vcpu(i, vcpu, kvm)
5316 set_bit(KVM_REQ_MASTERCLOCK_UPDATE, &vcpu->requests);
5317 atomic_set(&kvm_guest_has_master_clock, 0);
5318 raw_spin_unlock(&kvm_lock);
5321 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5324 * Notification about pvclock gtod data update.
5326 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5329 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5330 struct timekeeper *tk = priv;
5332 update_pvclock_gtod(tk);
5334 /* disable master clock if host does not trust, or does not
5335 * use, TSC clocksource
5337 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5338 atomic_read(&kvm_guest_has_master_clock) != 0)
5339 queue_work(system_long_wq, &pvclock_gtod_work);
5344 static struct notifier_block pvclock_gtod_notifier = {
5345 .notifier_call = pvclock_gtod_notify,
5349 int kvm_arch_init(void *opaque)
5352 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
5355 printk(KERN_ERR "kvm: already loaded the other module\n");
5360 if (!ops->cpu_has_kvm_support()) {
5361 printk(KERN_ERR "kvm: no hardware support\n");
5365 if (ops->disabled_by_bios()) {
5366 printk(KERN_ERR "kvm: disabled by bios\n");
5372 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5374 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5378 r = kvm_mmu_module_init();
5380 goto out_free_percpu;
5382 kvm_set_mmio_spte_mask();
5383 kvm_init_msr_list();
5386 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5387 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5391 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5394 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5397 #ifdef CONFIG_X86_64
5398 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5404 free_percpu(shared_msrs);
5409 void kvm_arch_exit(void)
5411 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5413 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5414 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5415 CPUFREQ_TRANSITION_NOTIFIER);
5416 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5417 #ifdef CONFIG_X86_64
5418 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5421 kvm_mmu_module_exit();
5422 free_percpu(shared_msrs);
5425 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5427 ++vcpu->stat.halt_exits;
5428 if (irqchip_in_kernel(vcpu->kvm)) {
5429 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5432 vcpu->run->exit_reason = KVM_EXIT_HLT;
5436 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5438 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5440 u64 param, ingpa, outgpa, ret;
5441 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5442 bool fast, longmode;
5446 * hypercall generates UD from non zero cpl and real mode
5449 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5450 kvm_queue_exception(vcpu, UD_VECTOR);
5454 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5455 longmode = is_long_mode(vcpu) && cs_l == 1;
5458 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5459 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5460 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5461 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5462 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5463 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5465 #ifdef CONFIG_X86_64
5467 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5468 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5469 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5473 code = param & 0xffff;
5474 fast = (param >> 16) & 0x1;
5475 rep_cnt = (param >> 32) & 0xfff;
5476 rep_idx = (param >> 48) & 0xfff;
5478 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5481 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5482 kvm_vcpu_on_spin(vcpu);
5485 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5489 ret = res | (((u64)rep_done & 0xfff) << 32);
5491 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5493 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5494 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5500 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5502 unsigned long nr, a0, a1, a2, a3, ret;
5505 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5506 return kvm_hv_hypercall(vcpu);
5508 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5509 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5510 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5511 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5512 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5514 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5516 if (!is_long_mode(vcpu)) {
5524 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5530 case KVM_HC_VAPIC_POLL_IRQ:
5538 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5539 ++vcpu->stat.hypercalls;
5542 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5544 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5546 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5547 char instruction[3];
5548 unsigned long rip = kvm_rip_read(vcpu);
5551 * Blow out the MMU to ensure that no other VCPU has an active mapping
5552 * to ensure that the updated hypercall appears atomically across all
5555 kvm_mmu_zap_all(vcpu->kvm);
5557 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5559 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5563 * Check if userspace requested an interrupt window, and that the
5564 * interrupt window is open.
5566 * No need to exit to userspace if we already have an interrupt queued.
5568 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5570 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5571 vcpu->run->request_interrupt_window &&
5572 kvm_arch_interrupt_allowed(vcpu));
5575 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5577 struct kvm_run *kvm_run = vcpu->run;
5579 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5580 kvm_run->cr8 = kvm_get_cr8(vcpu);
5581 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5582 if (irqchip_in_kernel(vcpu->kvm))
5583 kvm_run->ready_for_interrupt_injection = 1;
5585 kvm_run->ready_for_interrupt_injection =
5586 kvm_arch_interrupt_allowed(vcpu) &&
5587 !kvm_cpu_has_interrupt(vcpu) &&
5588 !kvm_event_needs_reinjection(vcpu);
5591 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5595 if (!kvm_x86_ops->update_cr8_intercept)
5598 if (!vcpu->arch.apic)
5601 if (!vcpu->arch.apic->vapic_addr)
5602 max_irr = kvm_lapic_find_highest_irr(vcpu);
5609 tpr = kvm_lapic_get_cr8(vcpu);
5611 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5614 static void inject_pending_event(struct kvm_vcpu *vcpu)
5616 /* try to reinject previous events if any */
5617 if (vcpu->arch.exception.pending) {
5618 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5619 vcpu->arch.exception.has_error_code,
5620 vcpu->arch.exception.error_code);
5621 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5622 vcpu->arch.exception.has_error_code,
5623 vcpu->arch.exception.error_code,
5624 vcpu->arch.exception.reinject);
5628 if (vcpu->arch.nmi_injected) {
5629 kvm_x86_ops->set_nmi(vcpu);
5633 if (vcpu->arch.interrupt.pending) {
5634 kvm_x86_ops->set_irq(vcpu);
5638 /* try to inject new event if pending */
5639 if (vcpu->arch.nmi_pending) {
5640 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5641 --vcpu->arch.nmi_pending;
5642 vcpu->arch.nmi_injected = true;
5643 kvm_x86_ops->set_nmi(vcpu);
5645 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
5646 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5647 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5649 kvm_x86_ops->set_irq(vcpu);
5654 static void process_nmi(struct kvm_vcpu *vcpu)
5659 * x86 is limited to one NMI running, and one NMI pending after it.
5660 * If an NMI is already in progress, limit further NMIs to just one.
5661 * Otherwise, allow two (and we'll inject the first one immediately).
5663 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5666 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5667 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5668 kvm_make_request(KVM_REQ_EVENT, vcpu);
5671 static void kvm_gen_update_masterclock(struct kvm *kvm)
5673 #ifdef CONFIG_X86_64
5675 struct kvm_vcpu *vcpu;
5676 struct kvm_arch *ka = &kvm->arch;
5678 spin_lock(&ka->pvclock_gtod_sync_lock);
5679 kvm_make_mclock_inprogress_request(kvm);
5680 /* no guest entries from this point */
5681 pvclock_update_vm_gtod_copy(kvm);
5683 kvm_for_each_vcpu(i, vcpu, kvm)
5684 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
5686 /* guest entries allowed */
5687 kvm_for_each_vcpu(i, vcpu, kvm)
5688 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
5690 spin_unlock(&ka->pvclock_gtod_sync_lock);
5694 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
5696 u64 eoi_exit_bitmap[4];
5699 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
5702 memset(eoi_exit_bitmap, 0, 32);
5705 kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
5706 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
5707 kvm_apic_update_tmr(vcpu, tmr);
5710 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5713 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5714 vcpu->run->request_interrupt_window;
5715 bool req_immediate_exit = false;
5717 if (vcpu->requests) {
5718 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5719 kvm_mmu_unload(vcpu);
5720 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5721 __kvm_migrate_timers(vcpu);
5722 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
5723 kvm_gen_update_masterclock(vcpu->kvm);
5724 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5725 r = kvm_guest_time_update(vcpu);
5729 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5730 kvm_mmu_sync_roots(vcpu);
5731 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5732 kvm_x86_ops->tlb_flush(vcpu);
5733 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5734 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5738 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5739 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5743 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5744 vcpu->fpu_active = 0;
5745 kvm_x86_ops->fpu_deactivate(vcpu);
5747 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5748 /* Page is swapped out. Do synthetic halt */
5749 vcpu->arch.apf.halted = true;
5753 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5754 record_steal_time(vcpu);
5755 if (kvm_check_request(KVM_REQ_NMI, vcpu))
5757 if (kvm_check_request(KVM_REQ_PMU, vcpu))
5758 kvm_handle_pmu_event(vcpu);
5759 if (kvm_check_request(KVM_REQ_PMI, vcpu))
5760 kvm_deliver_pmi(vcpu);
5761 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
5762 vcpu_scan_ioapic(vcpu);
5765 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5766 kvm_apic_accept_events(vcpu);
5767 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
5772 inject_pending_event(vcpu);
5774 /* enable NMI/IRQ window open exits if needed */
5775 if (vcpu->arch.nmi_pending)
5776 req_immediate_exit =
5777 kvm_x86_ops->enable_nmi_window(vcpu) != 0;
5778 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
5779 req_immediate_exit =
5780 kvm_x86_ops->enable_irq_window(vcpu) != 0;
5782 if (kvm_lapic_enabled(vcpu)) {
5784 * Update architecture specific hints for APIC
5785 * virtual interrupt delivery.
5787 if (kvm_x86_ops->hwapic_irr_update)
5788 kvm_x86_ops->hwapic_irr_update(vcpu,
5789 kvm_lapic_find_highest_irr(vcpu));
5790 update_cr8_intercept(vcpu);
5791 kvm_lapic_sync_to_vapic(vcpu);
5795 r = kvm_mmu_reload(vcpu);
5797 goto cancel_injection;
5802 kvm_x86_ops->prepare_guest_switch(vcpu);
5803 if (vcpu->fpu_active)
5804 kvm_load_guest_fpu(vcpu);
5805 kvm_load_guest_xcr0(vcpu);
5807 vcpu->mode = IN_GUEST_MODE;
5809 /* We should set ->mode before check ->requests,
5810 * see the comment in make_all_cpus_request.
5814 local_irq_disable();
5816 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5817 || need_resched() || signal_pending(current)) {
5818 vcpu->mode = OUTSIDE_GUEST_MODE;
5823 goto cancel_injection;
5826 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5828 if (req_immediate_exit)
5829 smp_send_reschedule(vcpu->cpu);
5833 if (unlikely(vcpu->arch.switch_db_regs)) {
5835 set_debugreg(vcpu->arch.eff_db[0], 0);
5836 set_debugreg(vcpu->arch.eff_db[1], 1);
5837 set_debugreg(vcpu->arch.eff_db[2], 2);
5838 set_debugreg(vcpu->arch.eff_db[3], 3);
5841 trace_kvm_entry(vcpu->vcpu_id);
5842 kvm_x86_ops->run(vcpu);
5845 * If the guest has used debug registers, at least dr7
5846 * will be disabled while returning to the host.
5847 * If we don't have active breakpoints in the host, we don't
5848 * care about the messed up debug address registers. But if
5849 * we have some of them active, restore the old state.
5851 if (hw_breakpoint_active())
5852 hw_breakpoint_restore();
5854 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
5857 vcpu->mode = OUTSIDE_GUEST_MODE;
5860 /* Interrupt is enabled by handle_external_intr() */
5861 kvm_x86_ops->handle_external_intr(vcpu);
5866 * We must have an instruction between local_irq_enable() and
5867 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5868 * the interrupt shadow. The stat.exits increment will do nicely.
5869 * But we need to prevent reordering, hence this barrier():
5877 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5880 * Profile KVM exit RIPs:
5882 if (unlikely(prof_on == KVM_PROFILING)) {
5883 unsigned long rip = kvm_rip_read(vcpu);
5884 profile_hit(KVM_PROFILING, (void *)rip);
5887 if (unlikely(vcpu->arch.tsc_always_catchup))
5888 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5890 if (vcpu->arch.apic_attention)
5891 kvm_lapic_sync_from_vapic(vcpu);
5893 r = kvm_x86_ops->handle_exit(vcpu);
5897 kvm_x86_ops->cancel_injection(vcpu);
5898 if (unlikely(vcpu->arch.apic_attention))
5899 kvm_lapic_sync_from_vapic(vcpu);
5905 static int __vcpu_run(struct kvm_vcpu *vcpu)
5908 struct kvm *kvm = vcpu->kvm;
5910 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5914 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5915 !vcpu->arch.apf.halted)
5916 r = vcpu_enter_guest(vcpu);
5918 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5919 kvm_vcpu_block(vcpu);
5920 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5921 if (kvm_check_request(KVM_REQ_UNHALT, vcpu)) {
5922 kvm_apic_accept_events(vcpu);
5923 switch(vcpu->arch.mp_state) {
5924 case KVM_MP_STATE_HALTED:
5925 vcpu->arch.mp_state =
5926 KVM_MP_STATE_RUNNABLE;
5927 case KVM_MP_STATE_RUNNABLE:
5928 vcpu->arch.apf.halted = false;
5930 case KVM_MP_STATE_INIT_RECEIVED:
5942 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5943 if (kvm_cpu_has_pending_timer(vcpu))
5944 kvm_inject_pending_timer_irqs(vcpu);
5946 if (dm_request_for_irq_injection(vcpu)) {
5948 vcpu->run->exit_reason = KVM_EXIT_INTR;
5949 ++vcpu->stat.request_irq_exits;
5952 kvm_check_async_pf_completion(vcpu);
5954 if (signal_pending(current)) {
5956 vcpu->run->exit_reason = KVM_EXIT_INTR;
5957 ++vcpu->stat.signal_exits;
5959 if (need_resched()) {
5960 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5962 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5966 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5971 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
5974 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5975 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5976 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5977 if (r != EMULATE_DONE)
5982 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
5984 BUG_ON(!vcpu->arch.pio.count);
5986 return complete_emulated_io(vcpu);
5990 * Implements the following, as a state machine:
5994 * for each mmio piece in the fragment
6002 * for each mmio piece in the fragment
6007 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6009 struct kvm_run *run = vcpu->run;
6010 struct kvm_mmio_fragment *frag;
6013 BUG_ON(!vcpu->mmio_needed);
6015 /* Complete previous fragment */
6016 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6017 len = min(8u, frag->len);
6018 if (!vcpu->mmio_is_write)
6019 memcpy(frag->data, run->mmio.data, len);
6021 if (frag->len <= 8) {
6022 /* Switch to the next fragment. */
6024 vcpu->mmio_cur_fragment++;
6026 /* Go forward to the next mmio piece. */
6032 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6033 vcpu->mmio_needed = 0;
6034 if (vcpu->mmio_is_write)
6036 vcpu->mmio_read_completed = 1;
6037 return complete_emulated_io(vcpu);
6040 run->exit_reason = KVM_EXIT_MMIO;
6041 run->mmio.phys_addr = frag->gpa;
6042 if (vcpu->mmio_is_write)
6043 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6044 run->mmio.len = min(8u, frag->len);
6045 run->mmio.is_write = vcpu->mmio_is_write;
6046 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6051 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6056 if (!tsk_used_math(current) && init_fpu(current))
6059 if (vcpu->sigset_active)
6060 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6062 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6063 kvm_vcpu_block(vcpu);
6064 kvm_apic_accept_events(vcpu);
6065 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6070 /* re-sync apic's tpr */
6071 if (!irqchip_in_kernel(vcpu->kvm)) {
6072 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6078 if (unlikely(vcpu->arch.complete_userspace_io)) {
6079 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6080 vcpu->arch.complete_userspace_io = NULL;
6085 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6087 r = __vcpu_run(vcpu);
6090 post_kvm_run_save(vcpu);
6091 if (vcpu->sigset_active)
6092 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6097 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6099 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6101 * We are here if userspace calls get_regs() in the middle of
6102 * instruction emulation. Registers state needs to be copied
6103 * back from emulation context to vcpu. Userspace shouldn't do
6104 * that usually, but some bad designed PV devices (vmware
6105 * backdoor interface) need this to work
6107 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6108 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6110 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6111 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6112 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6113 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6114 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6115 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6116 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6117 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6118 #ifdef CONFIG_X86_64
6119 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6120 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6121 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6122 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6123 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6124 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6125 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6126 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6129 regs->rip = kvm_rip_read(vcpu);
6130 regs->rflags = kvm_get_rflags(vcpu);
6135 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6137 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6138 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6140 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6141 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6142 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6143 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6144 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6145 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6146 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6147 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6148 #ifdef CONFIG_X86_64
6149 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6150 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6151 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6152 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6153 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6154 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6155 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6156 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6159 kvm_rip_write(vcpu, regs->rip);
6160 kvm_set_rflags(vcpu, regs->rflags);
6162 vcpu->arch.exception.pending = false;
6164 kvm_make_request(KVM_REQ_EVENT, vcpu);
6169 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6171 struct kvm_segment cs;
6173 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6177 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6179 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6180 struct kvm_sregs *sregs)
6184 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6185 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6186 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6187 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6188 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6189 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6191 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6192 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6194 kvm_x86_ops->get_idt(vcpu, &dt);
6195 sregs->idt.limit = dt.size;
6196 sregs->idt.base = dt.address;
6197 kvm_x86_ops->get_gdt(vcpu, &dt);
6198 sregs->gdt.limit = dt.size;
6199 sregs->gdt.base = dt.address;
6201 sregs->cr0 = kvm_read_cr0(vcpu);
6202 sregs->cr2 = vcpu->arch.cr2;
6203 sregs->cr3 = kvm_read_cr3(vcpu);
6204 sregs->cr4 = kvm_read_cr4(vcpu);
6205 sregs->cr8 = kvm_get_cr8(vcpu);
6206 sregs->efer = vcpu->arch.efer;
6207 sregs->apic_base = kvm_get_apic_base(vcpu);
6209 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6211 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6212 set_bit(vcpu->arch.interrupt.nr,
6213 (unsigned long *)sregs->interrupt_bitmap);
6218 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6219 struct kvm_mp_state *mp_state)
6221 kvm_apic_accept_events(vcpu);
6222 mp_state->mp_state = vcpu->arch.mp_state;
6226 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6227 struct kvm_mp_state *mp_state)
6229 if (!kvm_vcpu_has_lapic(vcpu) &&
6230 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6233 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6234 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6235 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6237 vcpu->arch.mp_state = mp_state->mp_state;
6238 kvm_make_request(KVM_REQ_EVENT, vcpu);
6242 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6243 int reason, bool has_error_code, u32 error_code)
6245 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6248 init_emulate_ctxt(vcpu);
6250 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6251 has_error_code, error_code);
6254 return EMULATE_FAIL;
6256 kvm_rip_write(vcpu, ctxt->eip);
6257 kvm_set_rflags(vcpu, ctxt->eflags);
6258 kvm_make_request(KVM_REQ_EVENT, vcpu);
6259 return EMULATE_DONE;
6261 EXPORT_SYMBOL_GPL(kvm_task_switch);
6263 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6264 struct kvm_sregs *sregs)
6266 int mmu_reset_needed = 0;
6267 int pending_vec, max_bits, idx;
6270 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6273 dt.size = sregs->idt.limit;
6274 dt.address = sregs->idt.base;
6275 kvm_x86_ops->set_idt(vcpu, &dt);
6276 dt.size = sregs->gdt.limit;
6277 dt.address = sregs->gdt.base;
6278 kvm_x86_ops->set_gdt(vcpu, &dt);
6280 vcpu->arch.cr2 = sregs->cr2;
6281 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6282 vcpu->arch.cr3 = sregs->cr3;
6283 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6285 kvm_set_cr8(vcpu, sregs->cr8);
6287 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6288 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6289 kvm_set_apic_base(vcpu, sregs->apic_base);
6291 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6292 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6293 vcpu->arch.cr0 = sregs->cr0;
6295 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6296 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6297 if (sregs->cr4 & X86_CR4_OSXSAVE)
6298 kvm_update_cpuid(vcpu);
6300 idx = srcu_read_lock(&vcpu->kvm->srcu);
6301 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6302 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6303 mmu_reset_needed = 1;
6305 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6307 if (mmu_reset_needed)
6308 kvm_mmu_reset_context(vcpu);
6310 max_bits = KVM_NR_INTERRUPTS;
6311 pending_vec = find_first_bit(
6312 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6313 if (pending_vec < max_bits) {
6314 kvm_queue_interrupt(vcpu, pending_vec, false);
6315 pr_debug("Set back pending irq %d\n", pending_vec);
6318 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6319 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6320 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6321 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6322 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6323 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6325 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6326 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6328 update_cr8_intercept(vcpu);
6330 /* Older userspace won't unhalt the vcpu on reset. */
6331 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6332 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6334 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6336 kvm_make_request(KVM_REQ_EVENT, vcpu);
6341 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6342 struct kvm_guest_debug *dbg)
6344 unsigned long rflags;
6347 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6349 if (vcpu->arch.exception.pending)
6351 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6352 kvm_queue_exception(vcpu, DB_VECTOR);
6354 kvm_queue_exception(vcpu, BP_VECTOR);
6358 * Read rflags as long as potentially injected trace flags are still
6361 rflags = kvm_get_rflags(vcpu);
6363 vcpu->guest_debug = dbg->control;
6364 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6365 vcpu->guest_debug = 0;
6367 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6368 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6369 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6370 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6372 for (i = 0; i < KVM_NR_DB_REGS; i++)
6373 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6375 kvm_update_dr7(vcpu);
6377 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6378 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6379 get_segment_base(vcpu, VCPU_SREG_CS);
6382 * Trigger an rflags update that will inject or remove the trace
6385 kvm_set_rflags(vcpu, rflags);
6387 kvm_x86_ops->update_db_bp_intercept(vcpu);
6397 * Translate a guest virtual address to a guest physical address.
6399 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6400 struct kvm_translation *tr)
6402 unsigned long vaddr = tr->linear_address;
6406 idx = srcu_read_lock(&vcpu->kvm->srcu);
6407 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6408 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6409 tr->physical_address = gpa;
6410 tr->valid = gpa != UNMAPPED_GVA;
6417 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6419 struct i387_fxsave_struct *fxsave =
6420 &vcpu->arch.guest_fpu.state->fxsave;
6422 memcpy(fpu->fpr, fxsave->st_space, 128);
6423 fpu->fcw = fxsave->cwd;
6424 fpu->fsw = fxsave->swd;
6425 fpu->ftwx = fxsave->twd;
6426 fpu->last_opcode = fxsave->fop;
6427 fpu->last_ip = fxsave->rip;
6428 fpu->last_dp = fxsave->rdp;
6429 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6434 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6436 struct i387_fxsave_struct *fxsave =
6437 &vcpu->arch.guest_fpu.state->fxsave;
6439 memcpy(fxsave->st_space, fpu->fpr, 128);
6440 fxsave->cwd = fpu->fcw;
6441 fxsave->swd = fpu->fsw;
6442 fxsave->twd = fpu->ftwx;
6443 fxsave->fop = fpu->last_opcode;
6444 fxsave->rip = fpu->last_ip;
6445 fxsave->rdp = fpu->last_dp;
6446 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6451 int fx_init(struct kvm_vcpu *vcpu)
6455 err = fpu_alloc(&vcpu->arch.guest_fpu);
6459 fpu_finit(&vcpu->arch.guest_fpu);
6462 * Ensure guest xcr0 is valid for loading
6464 vcpu->arch.xcr0 = XSTATE_FP;
6466 vcpu->arch.cr0 |= X86_CR0_ET;
6470 EXPORT_SYMBOL_GPL(fx_init);
6472 static void fx_free(struct kvm_vcpu *vcpu)
6474 fpu_free(&vcpu->arch.guest_fpu);
6477 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6479 if (vcpu->guest_fpu_loaded)
6483 * Restore all possible states in the guest,
6484 * and assume host would use all available bits.
6485 * Guest xcr0 would be loaded later.
6487 kvm_put_guest_xcr0(vcpu);
6488 vcpu->guest_fpu_loaded = 1;
6489 __kernel_fpu_begin();
6490 fpu_restore_checking(&vcpu->arch.guest_fpu);
6494 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6496 kvm_put_guest_xcr0(vcpu);
6498 if (!vcpu->guest_fpu_loaded)
6501 vcpu->guest_fpu_loaded = 0;
6502 fpu_save_init(&vcpu->arch.guest_fpu);
6504 ++vcpu->stat.fpu_reload;
6505 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6509 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6511 kvmclock_reset(vcpu);
6513 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6515 kvm_x86_ops->vcpu_free(vcpu);
6518 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6521 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6522 printk_once(KERN_WARNING
6523 "kvm: SMP vm created on host with unstable TSC; "
6524 "guest TSC will not be reliable\n");
6525 return kvm_x86_ops->vcpu_create(kvm, id);
6528 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6532 vcpu->arch.mtrr_state.have_fixed = 1;
6533 r = vcpu_load(vcpu);
6536 kvm_vcpu_reset(vcpu);
6537 r = kvm_mmu_setup(vcpu);
6543 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
6546 struct msr_data msr;
6548 r = vcpu_load(vcpu);
6552 msr.index = MSR_IA32_TSC;
6553 msr.host_initiated = true;
6554 kvm_write_tsc(vcpu, &msr);
6560 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6563 vcpu->arch.apf.msr_val = 0;
6565 r = vcpu_load(vcpu);
6567 kvm_mmu_unload(vcpu);
6571 kvm_x86_ops->vcpu_free(vcpu);
6574 void kvm_vcpu_reset(struct kvm_vcpu *vcpu)
6576 atomic_set(&vcpu->arch.nmi_queued, 0);
6577 vcpu->arch.nmi_pending = 0;
6578 vcpu->arch.nmi_injected = false;
6580 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6581 vcpu->arch.dr6 = DR6_FIXED_1;
6582 vcpu->arch.dr7 = DR7_FIXED_1;
6583 kvm_update_dr7(vcpu);
6585 kvm_make_request(KVM_REQ_EVENT, vcpu);
6586 vcpu->arch.apf.msr_val = 0;
6587 vcpu->arch.st.msr_val = 0;
6589 kvmclock_reset(vcpu);
6591 kvm_clear_async_pf_completion_queue(vcpu);
6592 kvm_async_pf_hash_reset(vcpu);
6593 vcpu->arch.apf.halted = false;
6595 kvm_pmu_reset(vcpu);
6597 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
6598 vcpu->arch.regs_avail = ~0;
6599 vcpu->arch.regs_dirty = ~0;
6601 kvm_x86_ops->vcpu_reset(vcpu);
6604 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, unsigned int vector)
6606 struct kvm_segment cs;
6608 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6609 cs.selector = vector << 8;
6610 cs.base = vector << 12;
6611 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6612 kvm_rip_write(vcpu, 0);
6615 int kvm_arch_hardware_enable(void *garbage)
6618 struct kvm_vcpu *vcpu;
6623 bool stable, backwards_tsc = false;
6625 kvm_shared_msr_cpu_online();
6626 ret = kvm_x86_ops->hardware_enable(garbage);
6630 local_tsc = native_read_tsc();
6631 stable = !check_tsc_unstable();
6632 list_for_each_entry(kvm, &vm_list, vm_list) {
6633 kvm_for_each_vcpu(i, vcpu, kvm) {
6634 if (!stable && vcpu->cpu == smp_processor_id())
6635 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
6636 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
6637 backwards_tsc = true;
6638 if (vcpu->arch.last_host_tsc > max_tsc)
6639 max_tsc = vcpu->arch.last_host_tsc;
6645 * Sometimes, even reliable TSCs go backwards. This happens on
6646 * platforms that reset TSC during suspend or hibernate actions, but
6647 * maintain synchronization. We must compensate. Fortunately, we can
6648 * detect that condition here, which happens early in CPU bringup,
6649 * before any KVM threads can be running. Unfortunately, we can't
6650 * bring the TSCs fully up to date with real time, as we aren't yet far
6651 * enough into CPU bringup that we know how much real time has actually
6652 * elapsed; our helper function, get_kernel_ns() will be using boot
6653 * variables that haven't been updated yet.
6655 * So we simply find the maximum observed TSC above, then record the
6656 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
6657 * the adjustment will be applied. Note that we accumulate
6658 * adjustments, in case multiple suspend cycles happen before some VCPU
6659 * gets a chance to run again. In the event that no KVM threads get a
6660 * chance to run, we will miss the entire elapsed period, as we'll have
6661 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
6662 * loose cycle time. This isn't too big a deal, since the loss will be
6663 * uniform across all VCPUs (not to mention the scenario is extremely
6664 * unlikely). It is possible that a second hibernate recovery happens
6665 * much faster than a first, causing the observed TSC here to be
6666 * smaller; this would require additional padding adjustment, which is
6667 * why we set last_host_tsc to the local tsc observed here.
6669 * N.B. - this code below runs only on platforms with reliable TSC,
6670 * as that is the only way backwards_tsc is set above. Also note
6671 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
6672 * have the same delta_cyc adjustment applied if backwards_tsc
6673 * is detected. Note further, this adjustment is only done once,
6674 * as we reset last_host_tsc on all VCPUs to stop this from being
6675 * called multiple times (one for each physical CPU bringup).
6677 * Platforms with unreliable TSCs don't have to deal with this, they
6678 * will be compensated by the logic in vcpu_load, which sets the TSC to
6679 * catchup mode. This will catchup all VCPUs to real time, but cannot
6680 * guarantee that they stay in perfect synchronization.
6682 if (backwards_tsc) {
6683 u64 delta_cyc = max_tsc - local_tsc;
6684 list_for_each_entry(kvm, &vm_list, vm_list) {
6685 kvm_for_each_vcpu(i, vcpu, kvm) {
6686 vcpu->arch.tsc_offset_adjustment += delta_cyc;
6687 vcpu->arch.last_host_tsc = local_tsc;
6688 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
6693 * We have to disable TSC offset matching.. if you were
6694 * booting a VM while issuing an S4 host suspend....
6695 * you may have some problem. Solving this issue is
6696 * left as an exercise to the reader.
6698 kvm->arch.last_tsc_nsec = 0;
6699 kvm->arch.last_tsc_write = 0;
6706 void kvm_arch_hardware_disable(void *garbage)
6708 kvm_x86_ops->hardware_disable(garbage);
6709 drop_user_return_notifiers(garbage);
6712 int kvm_arch_hardware_setup(void)
6714 return kvm_x86_ops->hardware_setup();
6717 void kvm_arch_hardware_unsetup(void)
6719 kvm_x86_ops->hardware_unsetup();
6722 void kvm_arch_check_processor_compat(void *rtn)
6724 kvm_x86_ops->check_processor_compatibility(rtn);
6727 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
6729 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
6732 struct static_key kvm_no_apic_vcpu __read_mostly;
6734 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6740 BUG_ON(vcpu->kvm == NULL);
6743 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6744 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6745 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6747 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6749 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6754 vcpu->arch.pio_data = page_address(page);
6756 kvm_set_tsc_khz(vcpu, max_tsc_khz);
6758 r = kvm_mmu_create(vcpu);
6760 goto fail_free_pio_data;
6762 if (irqchip_in_kernel(kvm)) {
6763 r = kvm_create_lapic(vcpu);
6765 goto fail_mmu_destroy;
6767 static_key_slow_inc(&kvm_no_apic_vcpu);
6769 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6771 if (!vcpu->arch.mce_banks) {
6773 goto fail_free_lapic;
6775 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6777 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
6779 goto fail_free_mce_banks;
6784 goto fail_free_wbinvd_dirty_mask;
6786 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
6787 vcpu->arch.pv_time_enabled = false;
6788 kvm_async_pf_hash_reset(vcpu);
6792 fail_free_wbinvd_dirty_mask:
6793 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6794 fail_free_mce_banks:
6795 kfree(vcpu->arch.mce_banks);
6797 kvm_free_lapic(vcpu);
6799 kvm_mmu_destroy(vcpu);
6801 free_page((unsigned long)vcpu->arch.pio_data);
6806 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6810 kvm_pmu_destroy(vcpu);
6811 kfree(vcpu->arch.mce_banks);
6812 kvm_free_lapic(vcpu);
6813 idx = srcu_read_lock(&vcpu->kvm->srcu);
6814 kvm_mmu_destroy(vcpu);
6815 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6816 free_page((unsigned long)vcpu->arch.pio_data);
6817 if (!irqchip_in_kernel(vcpu->kvm))
6818 static_key_slow_dec(&kvm_no_apic_vcpu);
6821 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
6826 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6827 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6829 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6830 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6831 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
6832 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
6833 &kvm->arch.irq_sources_bitmap);
6835 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6836 mutex_init(&kvm->arch.apic_map_lock);
6837 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
6839 pvclock_update_vm_gtod_copy(kvm);
6844 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6847 r = vcpu_load(vcpu);
6849 kvm_mmu_unload(vcpu);
6853 static void kvm_free_vcpus(struct kvm *kvm)
6856 struct kvm_vcpu *vcpu;
6859 * Unpin any mmu pages first.
6861 kvm_for_each_vcpu(i, vcpu, kvm) {
6862 kvm_clear_async_pf_completion_queue(vcpu);
6863 kvm_unload_vcpu_mmu(vcpu);
6865 kvm_for_each_vcpu(i, vcpu, kvm)
6866 kvm_arch_vcpu_free(vcpu);
6868 mutex_lock(&kvm->lock);
6869 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6870 kvm->vcpus[i] = NULL;
6872 atomic_set(&kvm->online_vcpus, 0);
6873 mutex_unlock(&kvm->lock);
6876 void kvm_arch_sync_events(struct kvm *kvm)
6878 kvm_free_all_assigned_devices(kvm);
6882 void kvm_arch_destroy_vm(struct kvm *kvm)
6884 if (current->mm == kvm->mm) {
6886 * Free memory regions allocated on behalf of userspace,
6887 * unless the the memory map has changed due to process exit
6890 struct kvm_userspace_memory_region mem;
6891 memset(&mem, 0, sizeof(mem));
6892 mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
6893 kvm_set_memory_region(kvm, &mem);
6895 mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
6896 kvm_set_memory_region(kvm, &mem);
6898 mem.slot = TSS_PRIVATE_MEMSLOT;
6899 kvm_set_memory_region(kvm, &mem);
6901 kvm_iommu_unmap_guest(kvm);
6902 kfree(kvm->arch.vpic);
6903 kfree(kvm->arch.vioapic);
6904 kvm_free_vcpus(kvm);
6905 if (kvm->arch.apic_access_page)
6906 put_page(kvm->arch.apic_access_page);
6907 if (kvm->arch.ept_identity_pagetable)
6908 put_page(kvm->arch.ept_identity_pagetable);
6909 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
6912 void kvm_arch_free_memslot(struct kvm_memory_slot *free,
6913 struct kvm_memory_slot *dont)
6917 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6918 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
6919 kvm_kvfree(free->arch.rmap[i]);
6920 free->arch.rmap[i] = NULL;
6925 if (!dont || free->arch.lpage_info[i - 1] !=
6926 dont->arch.lpage_info[i - 1]) {
6927 kvm_kvfree(free->arch.lpage_info[i - 1]);
6928 free->arch.lpage_info[i - 1] = NULL;
6933 int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages)
6937 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6942 lpages = gfn_to_index(slot->base_gfn + npages - 1,
6943 slot->base_gfn, level) + 1;
6945 slot->arch.rmap[i] =
6946 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
6947 if (!slot->arch.rmap[i])
6952 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
6953 sizeof(*slot->arch.lpage_info[i - 1]));
6954 if (!slot->arch.lpage_info[i - 1])
6957 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
6958 slot->arch.lpage_info[i - 1][0].write_count = 1;
6959 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
6960 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
6961 ugfn = slot->userspace_addr >> PAGE_SHIFT;
6963 * If the gfn and userspace address are not aligned wrt each
6964 * other, or if explicitly asked to, disable large page
6965 * support for this slot
6967 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
6968 !kvm_largepages_enabled()) {
6971 for (j = 0; j < lpages; ++j)
6972 slot->arch.lpage_info[i - 1][j].write_count = 1;
6979 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6980 kvm_kvfree(slot->arch.rmap[i]);
6981 slot->arch.rmap[i] = NULL;
6985 kvm_kvfree(slot->arch.lpage_info[i - 1]);
6986 slot->arch.lpage_info[i - 1] = NULL;
6991 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6992 struct kvm_memory_slot *memslot,
6993 struct kvm_userspace_memory_region *mem,
6994 enum kvm_mr_change change)
6997 * Only private memory slots need to be mapped here since
6998 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
7000 if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
7001 unsigned long userspace_addr;
7004 * MAP_SHARED to prevent internal slot pages from being moved
7007 userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
7008 PROT_READ | PROT_WRITE,
7009 MAP_SHARED | MAP_ANONYMOUS, 0);
7011 if (IS_ERR((void *)userspace_addr))
7012 return PTR_ERR((void *)userspace_addr);
7014 memslot->userspace_addr = userspace_addr;
7020 void kvm_arch_commit_memory_region(struct kvm *kvm,
7021 struct kvm_userspace_memory_region *mem,
7022 const struct kvm_memory_slot *old,
7023 enum kvm_mr_change change)
7026 int nr_mmu_pages = 0;
7028 if ((mem->slot >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_DELETE)) {
7031 ret = vm_munmap(old->userspace_addr,
7032 old->npages * PAGE_SIZE);
7035 "kvm_vm_ioctl_set_memory_region: "
7036 "failed to munmap memory\n");
7039 if (!kvm->arch.n_requested_mmu_pages)
7040 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7043 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7045 * Write protect all pages for dirty logging.
7046 * Existing largepage mappings are destroyed here and new ones will
7047 * not be created until the end of the logging.
7049 if ((change != KVM_MR_DELETE) && (mem->flags & KVM_MEM_LOG_DIRTY_PAGES))
7050 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
7052 * If memory slot is created, or moved, we need to clear all
7055 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
7056 kvm_mmu_zap_mmio_sptes(kvm);
7057 kvm_reload_remote_mmus(kvm);
7061 void kvm_arch_flush_shadow_all(struct kvm *kvm)
7063 kvm_mmu_zap_all(kvm);
7064 kvm_reload_remote_mmus(kvm);
7067 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7068 struct kvm_memory_slot *slot)
7070 kvm_arch_flush_shadow_all(kvm);
7073 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
7075 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7076 !vcpu->arch.apf.halted)
7077 || !list_empty_careful(&vcpu->async_pf.done)
7078 || kvm_apic_has_events(vcpu)
7079 || atomic_read(&vcpu->arch.nmi_queued) ||
7080 (kvm_arch_interrupt_allowed(vcpu) &&
7081 kvm_cpu_has_interrupt(vcpu));
7084 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
7086 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
7089 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7091 return kvm_x86_ops->interrupt_allowed(vcpu);
7094 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7096 unsigned long current_rip = kvm_rip_read(vcpu) +
7097 get_segment_base(vcpu, VCPU_SREG_CS);
7099 return current_rip == linear_rip;
7101 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7103 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7105 unsigned long rflags;
7107 rflags = kvm_x86_ops->get_rflags(vcpu);
7108 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7109 rflags &= ~X86_EFLAGS_TF;
7112 EXPORT_SYMBOL_GPL(kvm_get_rflags);
7114 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7116 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
7117 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
7118 rflags |= X86_EFLAGS_TF;
7119 kvm_x86_ops->set_rflags(vcpu, rflags);
7120 kvm_make_request(KVM_REQ_EVENT, vcpu);
7122 EXPORT_SYMBOL_GPL(kvm_set_rflags);
7124 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7128 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
7129 is_error_page(work->page))
7132 r = kvm_mmu_reload(vcpu);
7136 if (!vcpu->arch.mmu.direct_map &&
7137 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7140 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7143 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7145 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7148 static inline u32 kvm_async_pf_next_probe(u32 key)
7150 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7153 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7155 u32 key = kvm_async_pf_hash_fn(gfn);
7157 while (vcpu->arch.apf.gfns[key] != ~0)
7158 key = kvm_async_pf_next_probe(key);
7160 vcpu->arch.apf.gfns[key] = gfn;
7163 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7166 u32 key = kvm_async_pf_hash_fn(gfn);
7168 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
7169 (vcpu->arch.apf.gfns[key] != gfn &&
7170 vcpu->arch.apf.gfns[key] != ~0); i++)
7171 key = kvm_async_pf_next_probe(key);
7176 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7178 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7181 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7185 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7187 vcpu->arch.apf.gfns[i] = ~0;
7189 j = kvm_async_pf_next_probe(j);
7190 if (vcpu->arch.apf.gfns[j] == ~0)
7192 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7194 * k lies cyclically in ]i,j]
7196 * |....j i.k.| or |.k..j i...|
7198 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7199 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7204 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7207 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7211 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7212 struct kvm_async_pf *work)
7214 struct x86_exception fault;
7216 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
7217 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7219 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
7220 (vcpu->arch.apf.send_user_only &&
7221 kvm_x86_ops->get_cpl(vcpu) == 0))
7222 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7223 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
7224 fault.vector = PF_VECTOR;
7225 fault.error_code_valid = true;
7226 fault.error_code = 0;
7227 fault.nested_page_fault = false;
7228 fault.address = work->arch.token;
7229 kvm_inject_page_fault(vcpu, &fault);
7233 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7234 struct kvm_async_pf *work)
7236 struct x86_exception fault;
7238 trace_kvm_async_pf_ready(work->arch.token, work->gva);
7239 if (is_error_page(work->page))
7240 work->arch.token = ~0; /* broadcast wakeup */
7242 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7244 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7245 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
7246 fault.vector = PF_VECTOR;
7247 fault.error_code_valid = true;
7248 fault.error_code = 0;
7249 fault.nested_page_fault = false;
7250 fault.address = work->arch.token;
7251 kvm_inject_page_fault(vcpu, &fault);
7253 vcpu->arch.apf.halted = false;
7254 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7257 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7259 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7262 return !kvm_event_needs_reinjection(vcpu) &&
7263 kvm_x86_ops->interrupt_allowed(vcpu);
7266 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
7267 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
7268 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
7269 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
7270 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
7271 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
7272 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
7273 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
7274 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
7275 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
7276 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
7277 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);