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 void 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);
233 if (((value ^ smsr->values[slot].curr) & mask) == 0)
235 smsr->values[slot].curr = value;
236 wrmsrl(shared_msrs_global.msrs[slot], value);
237 if (!smsr->registered) {
238 smsr->urn.on_user_return = kvm_on_user_return;
239 user_return_notifier_register(&smsr->urn);
240 smsr->registered = true;
243 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
245 static void drop_user_return_notifiers(void *ignore)
247 unsigned int cpu = smp_processor_id();
248 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
250 if (smsr->registered)
251 kvm_on_user_return(&smsr->urn);
254 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
256 return vcpu->arch.apic_base;
258 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
260 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
262 /* TODO: reserve bits check */
263 kvm_lapic_set_base(vcpu, data);
265 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
267 asmlinkage void kvm_spurious_fault(void)
269 /* Fault while not rebooting. We want the trace. */
272 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
274 #define EXCPT_BENIGN 0
275 #define EXCPT_CONTRIBUTORY 1
278 static int exception_class(int vector)
288 return EXCPT_CONTRIBUTORY;
295 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
296 unsigned nr, bool has_error, u32 error_code,
302 kvm_make_request(KVM_REQ_EVENT, vcpu);
304 if (!vcpu->arch.exception.pending) {
306 vcpu->arch.exception.pending = true;
307 vcpu->arch.exception.has_error_code = has_error;
308 vcpu->arch.exception.nr = nr;
309 vcpu->arch.exception.error_code = error_code;
310 vcpu->arch.exception.reinject = reinject;
314 /* to check exception */
315 prev_nr = vcpu->arch.exception.nr;
316 if (prev_nr == DF_VECTOR) {
317 /* triple fault -> shutdown */
318 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
321 class1 = exception_class(prev_nr);
322 class2 = exception_class(nr);
323 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
324 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
325 /* generate double fault per SDM Table 5-5 */
326 vcpu->arch.exception.pending = true;
327 vcpu->arch.exception.has_error_code = true;
328 vcpu->arch.exception.nr = DF_VECTOR;
329 vcpu->arch.exception.error_code = 0;
331 /* replace previous exception with a new one in a hope
332 that instruction re-execution will regenerate lost
337 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
339 kvm_multiple_exception(vcpu, nr, false, 0, false);
341 EXPORT_SYMBOL_GPL(kvm_queue_exception);
343 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
345 kvm_multiple_exception(vcpu, nr, false, 0, true);
347 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
349 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
352 kvm_inject_gp(vcpu, 0);
354 kvm_x86_ops->skip_emulated_instruction(vcpu);
356 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
358 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
360 ++vcpu->stat.pf_guest;
361 vcpu->arch.cr2 = fault->address;
362 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
364 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
366 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
368 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
369 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
371 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
374 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
376 atomic_inc(&vcpu->arch.nmi_queued);
377 kvm_make_request(KVM_REQ_NMI, vcpu);
379 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
381 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
383 kvm_multiple_exception(vcpu, nr, true, error_code, false);
385 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
387 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
389 kvm_multiple_exception(vcpu, nr, true, error_code, true);
391 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
394 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
395 * a #GP and return false.
397 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
399 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
401 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
404 EXPORT_SYMBOL_GPL(kvm_require_cpl);
407 * This function will be used to read from the physical memory of the currently
408 * running guest. The difference to kvm_read_guest_page is that this function
409 * can read from guest physical or from the guest's guest physical memory.
411 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
412 gfn_t ngfn, void *data, int offset, int len,
418 ngpa = gfn_to_gpa(ngfn);
419 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
420 if (real_gfn == UNMAPPED_GVA)
423 real_gfn = gpa_to_gfn(real_gfn);
425 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
427 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
429 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
430 void *data, int offset, int len, u32 access)
432 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
433 data, offset, len, access);
437 * Load the pae pdptrs. Return true is they are all valid.
439 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
441 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
442 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
445 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
447 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
448 offset * sizeof(u64), sizeof(pdpte),
449 PFERR_USER_MASK|PFERR_WRITE_MASK);
454 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
455 if (is_present_gpte(pdpte[i]) &&
456 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
463 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
464 __set_bit(VCPU_EXREG_PDPTR,
465 (unsigned long *)&vcpu->arch.regs_avail);
466 __set_bit(VCPU_EXREG_PDPTR,
467 (unsigned long *)&vcpu->arch.regs_dirty);
472 EXPORT_SYMBOL_GPL(load_pdptrs);
474 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
476 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
482 if (is_long_mode(vcpu) || !is_pae(vcpu))
485 if (!test_bit(VCPU_EXREG_PDPTR,
486 (unsigned long *)&vcpu->arch.regs_avail))
489 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
490 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
491 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
492 PFERR_USER_MASK | PFERR_WRITE_MASK);
495 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
501 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
503 unsigned long old_cr0 = kvm_read_cr0(vcpu);
504 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
505 X86_CR0_CD | X86_CR0_NW;
510 if (cr0 & 0xffffffff00000000UL)
514 cr0 &= ~CR0_RESERVED_BITS;
516 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
519 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
522 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
524 if ((vcpu->arch.efer & EFER_LME)) {
529 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
534 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
539 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
542 kvm_x86_ops->set_cr0(vcpu, cr0);
544 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
545 kvm_clear_async_pf_completion_queue(vcpu);
546 kvm_async_pf_hash_reset(vcpu);
549 if ((cr0 ^ old_cr0) & update_bits)
550 kvm_mmu_reset_context(vcpu);
553 EXPORT_SYMBOL_GPL(kvm_set_cr0);
555 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
557 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
559 EXPORT_SYMBOL_GPL(kvm_lmsw);
561 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
563 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
564 !vcpu->guest_xcr0_loaded) {
565 /* kvm_set_xcr() also depends on this */
566 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
567 vcpu->guest_xcr0_loaded = 1;
571 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
573 if (vcpu->guest_xcr0_loaded) {
574 if (vcpu->arch.xcr0 != host_xcr0)
575 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
576 vcpu->guest_xcr0_loaded = 0;
580 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
584 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
585 if (index != XCR_XFEATURE_ENABLED_MASK)
588 if (!(xcr0 & XSTATE_FP))
590 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
592 if (xcr0 & ~host_xcr0)
594 kvm_put_guest_xcr0(vcpu);
595 vcpu->arch.xcr0 = xcr0;
599 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
601 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
602 __kvm_set_xcr(vcpu, index, xcr)) {
603 kvm_inject_gp(vcpu, 0);
608 EXPORT_SYMBOL_GPL(kvm_set_xcr);
610 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
612 unsigned long old_cr4 = kvm_read_cr4(vcpu);
613 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
614 X86_CR4_PAE | X86_CR4_SMEP;
615 if (cr4 & CR4_RESERVED_BITS)
618 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
621 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
624 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_RDWRGSFS))
627 if (is_long_mode(vcpu)) {
628 if (!(cr4 & X86_CR4_PAE))
630 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
631 && ((cr4 ^ old_cr4) & pdptr_bits)
632 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
636 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
637 if (!guest_cpuid_has_pcid(vcpu))
640 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
641 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
645 if (kvm_x86_ops->set_cr4(vcpu, cr4))
648 if (((cr4 ^ old_cr4) & pdptr_bits) ||
649 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
650 kvm_mmu_reset_context(vcpu);
652 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
653 kvm_update_cpuid(vcpu);
657 EXPORT_SYMBOL_GPL(kvm_set_cr4);
659 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
661 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
662 kvm_mmu_sync_roots(vcpu);
663 kvm_mmu_flush_tlb(vcpu);
667 if (is_long_mode(vcpu)) {
668 if (kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)) {
669 if (cr3 & CR3_PCID_ENABLED_RESERVED_BITS)
672 if (cr3 & CR3_L_MODE_RESERVED_BITS)
676 if (cr3 & CR3_PAE_RESERVED_BITS)
678 if (is_paging(vcpu) &&
679 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
683 * We don't check reserved bits in nonpae mode, because
684 * this isn't enforced, and VMware depends on this.
689 * Does the new cr3 value map to physical memory? (Note, we
690 * catch an invalid cr3 even in real-mode, because it would
691 * cause trouble later on when we turn on paging anyway.)
693 * A real CPU would silently accept an invalid cr3 and would
694 * attempt to use it - with largely undefined (and often hard
695 * to debug) behavior on the guest side.
697 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
699 vcpu->arch.cr3 = cr3;
700 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
701 vcpu->arch.mmu.new_cr3(vcpu);
704 EXPORT_SYMBOL_GPL(kvm_set_cr3);
706 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
708 if (cr8 & CR8_RESERVED_BITS)
710 if (irqchip_in_kernel(vcpu->kvm))
711 kvm_lapic_set_tpr(vcpu, cr8);
713 vcpu->arch.cr8 = cr8;
716 EXPORT_SYMBOL_GPL(kvm_set_cr8);
718 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
720 if (irqchip_in_kernel(vcpu->kvm))
721 return kvm_lapic_get_cr8(vcpu);
723 return vcpu->arch.cr8;
725 EXPORT_SYMBOL_GPL(kvm_get_cr8);
727 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
731 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
732 dr7 = vcpu->arch.guest_debug_dr7;
734 dr7 = vcpu->arch.dr7;
735 kvm_x86_ops->set_dr7(vcpu, dr7);
736 vcpu->arch.switch_db_regs = (dr7 & DR7_BP_EN_MASK);
739 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
743 vcpu->arch.db[dr] = val;
744 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
745 vcpu->arch.eff_db[dr] = val;
748 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
752 if (val & 0xffffffff00000000ULL)
754 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
757 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
761 if (val & 0xffffffff00000000ULL)
763 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
764 kvm_update_dr7(vcpu);
771 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
775 res = __kvm_set_dr(vcpu, dr, val);
777 kvm_queue_exception(vcpu, UD_VECTOR);
779 kvm_inject_gp(vcpu, 0);
783 EXPORT_SYMBOL_GPL(kvm_set_dr);
785 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
789 *val = vcpu->arch.db[dr];
792 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
796 *val = vcpu->arch.dr6;
799 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
803 *val = vcpu->arch.dr7;
810 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
812 if (_kvm_get_dr(vcpu, dr, val)) {
813 kvm_queue_exception(vcpu, UD_VECTOR);
818 EXPORT_SYMBOL_GPL(kvm_get_dr);
820 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
822 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
826 err = kvm_pmu_read_pmc(vcpu, ecx, &data);
829 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
830 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
833 EXPORT_SYMBOL_GPL(kvm_rdpmc);
836 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
837 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
839 * This list is modified at module load time to reflect the
840 * capabilities of the host cpu. This capabilities test skips MSRs that are
841 * kvm-specific. Those are put in the beginning of the list.
844 #define KVM_SAVE_MSRS_BEGIN 10
845 static u32 msrs_to_save[] = {
846 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
847 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
848 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
849 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
851 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
854 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
856 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
859 static unsigned num_msrs_to_save;
861 static const u32 emulated_msrs[] = {
863 MSR_IA32_TSCDEADLINE,
864 MSR_IA32_MISC_ENABLE,
869 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
871 if (efer & efer_reserved_bits)
874 if (efer & EFER_FFXSR) {
875 struct kvm_cpuid_entry2 *feat;
877 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
878 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
882 if (efer & EFER_SVME) {
883 struct kvm_cpuid_entry2 *feat;
885 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
886 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
892 EXPORT_SYMBOL_GPL(kvm_valid_efer);
894 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
896 u64 old_efer = vcpu->arch.efer;
898 if (!kvm_valid_efer(vcpu, efer))
902 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
906 efer |= vcpu->arch.efer & EFER_LMA;
908 kvm_x86_ops->set_efer(vcpu, efer);
910 /* Update reserved bits */
911 if ((efer ^ old_efer) & EFER_NX)
912 kvm_mmu_reset_context(vcpu);
917 void kvm_enable_efer_bits(u64 mask)
919 efer_reserved_bits &= ~mask;
921 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
925 * Writes msr value into into the appropriate "register".
926 * Returns 0 on success, non-0 otherwise.
927 * Assumes vcpu_load() was already called.
929 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
931 return kvm_x86_ops->set_msr(vcpu, msr);
935 * Adapt set_msr() to msr_io()'s calling convention
937 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
943 msr.host_initiated = true;
944 return kvm_set_msr(vcpu, &msr);
948 struct pvclock_gtod_data {
951 struct { /* extract of a clocksource struct */
959 /* open coded 'struct timespec' */
960 u64 monotonic_time_snsec;
961 time_t monotonic_time_sec;
964 static struct pvclock_gtod_data pvclock_gtod_data;
966 static void update_pvclock_gtod(struct timekeeper *tk)
968 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
970 write_seqcount_begin(&vdata->seq);
972 /* copy pvclock gtod data */
973 vdata->clock.vclock_mode = tk->clock->archdata.vclock_mode;
974 vdata->clock.cycle_last = tk->clock->cycle_last;
975 vdata->clock.mask = tk->clock->mask;
976 vdata->clock.mult = tk->mult;
977 vdata->clock.shift = tk->shift;
979 vdata->monotonic_time_sec = tk->xtime_sec
980 + tk->wall_to_monotonic.tv_sec;
981 vdata->monotonic_time_snsec = tk->xtime_nsec
982 + (tk->wall_to_monotonic.tv_nsec
984 while (vdata->monotonic_time_snsec >=
985 (((u64)NSEC_PER_SEC) << tk->shift)) {
986 vdata->monotonic_time_snsec -=
987 ((u64)NSEC_PER_SEC) << tk->shift;
988 vdata->monotonic_time_sec++;
991 write_seqcount_end(&vdata->seq);
996 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1000 struct pvclock_wall_clock wc;
1001 struct timespec boot;
1006 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1011 ++version; /* first time write, random junk */
1015 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1018 * The guest calculates current wall clock time by adding
1019 * system time (updated by kvm_guest_time_update below) to the
1020 * wall clock specified here. guest system time equals host
1021 * system time for us, thus we must fill in host boot time here.
1025 if (kvm->arch.kvmclock_offset) {
1026 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1027 boot = timespec_sub(boot, ts);
1029 wc.sec = boot.tv_sec;
1030 wc.nsec = boot.tv_nsec;
1031 wc.version = version;
1033 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1036 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1039 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1041 uint32_t quotient, remainder;
1043 /* Don't try to replace with do_div(), this one calculates
1044 * "(dividend << 32) / divisor" */
1046 : "=a" (quotient), "=d" (remainder)
1047 : "0" (0), "1" (dividend), "r" (divisor) );
1051 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1052 s8 *pshift, u32 *pmultiplier)
1059 tps64 = base_khz * 1000LL;
1060 scaled64 = scaled_khz * 1000LL;
1061 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1066 tps32 = (uint32_t)tps64;
1067 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1068 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1076 *pmultiplier = div_frac(scaled64, tps32);
1078 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1079 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1082 static inline u64 get_kernel_ns(void)
1086 WARN_ON(preemptible());
1088 monotonic_to_bootbased(&ts);
1089 return timespec_to_ns(&ts);
1092 #ifdef CONFIG_X86_64
1093 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1096 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1097 unsigned long max_tsc_khz;
1099 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1101 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1102 vcpu->arch.virtual_tsc_shift);
1105 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1107 u64 v = (u64)khz * (1000000 + ppm);
1112 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1114 u32 thresh_lo, thresh_hi;
1115 int use_scaling = 0;
1117 /* tsc_khz can be zero if TSC calibration fails */
1118 if (this_tsc_khz == 0)
1121 /* Compute a scale to convert nanoseconds in TSC cycles */
1122 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1123 &vcpu->arch.virtual_tsc_shift,
1124 &vcpu->arch.virtual_tsc_mult);
1125 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1128 * Compute the variation in TSC rate which is acceptable
1129 * within the range of tolerance and decide if the
1130 * rate being applied is within that bounds of the hardware
1131 * rate. If so, no scaling or compensation need be done.
1133 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1134 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1135 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1136 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1139 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1142 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1144 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1145 vcpu->arch.virtual_tsc_mult,
1146 vcpu->arch.virtual_tsc_shift);
1147 tsc += vcpu->arch.this_tsc_write;
1151 void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1153 #ifdef CONFIG_X86_64
1155 bool do_request = false;
1156 struct kvm_arch *ka = &vcpu->kvm->arch;
1157 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1159 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1160 atomic_read(&vcpu->kvm->online_vcpus));
1162 if (vcpus_matched && gtod->clock.vclock_mode == VCLOCK_TSC)
1163 if (!ka->use_master_clock)
1166 if (!vcpus_matched && ka->use_master_clock)
1170 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1172 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1173 atomic_read(&vcpu->kvm->online_vcpus),
1174 ka->use_master_clock, gtod->clock.vclock_mode);
1178 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1180 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1181 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1184 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1186 struct kvm *kvm = vcpu->kvm;
1187 u64 offset, ns, elapsed;
1188 unsigned long flags;
1191 u64 data = msr->data;
1193 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1194 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1195 ns = get_kernel_ns();
1196 elapsed = ns - kvm->arch.last_tsc_nsec;
1198 if (vcpu->arch.virtual_tsc_khz) {
1199 /* n.b - signed multiplication and division required */
1200 usdiff = data - kvm->arch.last_tsc_write;
1201 #ifdef CONFIG_X86_64
1202 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1204 /* do_div() only does unsigned */
1205 asm("idivl %2; xor %%edx, %%edx"
1207 : "A"(usdiff * 1000), "rm"(vcpu->arch.virtual_tsc_khz));
1209 do_div(elapsed, 1000);
1214 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1217 * Special case: TSC write with a small delta (1 second) of virtual
1218 * cycle time against real time is interpreted as an attempt to
1219 * synchronize the CPU.
1221 * For a reliable TSC, we can match TSC offsets, and for an unstable
1222 * TSC, we add elapsed time in this computation. We could let the
1223 * compensation code attempt to catch up if we fall behind, but
1224 * it's better to try to match offsets from the beginning.
1226 if (usdiff < USEC_PER_SEC &&
1227 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1228 if (!check_tsc_unstable()) {
1229 offset = kvm->arch.cur_tsc_offset;
1230 pr_debug("kvm: matched tsc offset for %llu\n", data);
1232 u64 delta = nsec_to_cycles(vcpu, elapsed);
1234 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1235 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1240 * We split periods of matched TSC writes into generations.
1241 * For each generation, we track the original measured
1242 * nanosecond time, offset, and write, so if TSCs are in
1243 * sync, we can match exact offset, and if not, we can match
1244 * exact software computation in compute_guest_tsc()
1246 * These values are tracked in kvm->arch.cur_xxx variables.
1248 kvm->arch.cur_tsc_generation++;
1249 kvm->arch.cur_tsc_nsec = ns;
1250 kvm->arch.cur_tsc_write = data;
1251 kvm->arch.cur_tsc_offset = offset;
1253 pr_debug("kvm: new tsc generation %u, clock %llu\n",
1254 kvm->arch.cur_tsc_generation, data);
1258 * We also track th most recent recorded KHZ, write and time to
1259 * allow the matching interval to be extended at each write.
1261 kvm->arch.last_tsc_nsec = ns;
1262 kvm->arch.last_tsc_write = data;
1263 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1265 /* Reset of TSC must disable overshoot protection below */
1266 vcpu->arch.hv_clock.tsc_timestamp = 0;
1267 vcpu->arch.last_guest_tsc = data;
1269 /* Keep track of which generation this VCPU has synchronized to */
1270 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1271 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1272 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1274 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1275 update_ia32_tsc_adjust_msr(vcpu, offset);
1276 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1277 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1279 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1281 kvm->arch.nr_vcpus_matched_tsc++;
1283 kvm->arch.nr_vcpus_matched_tsc = 0;
1285 kvm_track_tsc_matching(vcpu);
1286 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1289 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1291 #ifdef CONFIG_X86_64
1293 static cycle_t read_tsc(void)
1299 * Empirically, a fence (of type that depends on the CPU)
1300 * before rdtsc is enough to ensure that rdtsc is ordered
1301 * with respect to loads. The various CPU manuals are unclear
1302 * as to whether rdtsc can be reordered with later loads,
1303 * but no one has ever seen it happen.
1306 ret = (cycle_t)vget_cycles();
1308 last = pvclock_gtod_data.clock.cycle_last;
1310 if (likely(ret >= last))
1314 * GCC likes to generate cmov here, but this branch is extremely
1315 * predictable (it's just a funciton of time and the likely is
1316 * very likely) and there's a data dependence, so force GCC
1317 * to generate a branch instead. I don't barrier() because
1318 * we don't actually need a barrier, and if this function
1319 * ever gets inlined it will generate worse code.
1325 static inline u64 vgettsc(cycle_t *cycle_now)
1328 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1330 *cycle_now = read_tsc();
1332 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1333 return v * gtod->clock.mult;
1336 static int do_monotonic(struct timespec *ts, cycle_t *cycle_now)
1341 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1345 seq = read_seqcount_begin(>od->seq);
1346 mode = gtod->clock.vclock_mode;
1347 ts->tv_sec = gtod->monotonic_time_sec;
1348 ns = gtod->monotonic_time_snsec;
1349 ns += vgettsc(cycle_now);
1350 ns >>= gtod->clock.shift;
1351 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1352 timespec_add_ns(ts, ns);
1357 /* returns true if host is using tsc clocksource */
1358 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1362 /* checked again under seqlock below */
1363 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1366 if (do_monotonic(&ts, cycle_now) != VCLOCK_TSC)
1369 monotonic_to_bootbased(&ts);
1370 *kernel_ns = timespec_to_ns(&ts);
1378 * Assuming a stable TSC across physical CPUS, and a stable TSC
1379 * across virtual CPUs, the following condition is possible.
1380 * Each numbered line represents an event visible to both
1381 * CPUs at the next numbered event.
1383 * "timespecX" represents host monotonic time. "tscX" represents
1386 * VCPU0 on CPU0 | VCPU1 on CPU1
1388 * 1. read timespec0,tsc0
1389 * 2. | timespec1 = timespec0 + N
1391 * 3. transition to guest | transition to guest
1392 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1393 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1394 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1396 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1399 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1401 * - 0 < N - M => M < N
1403 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1404 * always the case (the difference between two distinct xtime instances
1405 * might be smaller then the difference between corresponding TSC reads,
1406 * when updating guest vcpus pvclock areas).
1408 * To avoid that problem, do not allow visibility of distinct
1409 * system_timestamp/tsc_timestamp values simultaneously: use a master
1410 * copy of host monotonic time values. Update that master copy
1413 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1417 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1419 #ifdef CONFIG_X86_64
1420 struct kvm_arch *ka = &kvm->arch;
1422 bool host_tsc_clocksource, vcpus_matched;
1424 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1425 atomic_read(&kvm->online_vcpus));
1428 * If the host uses TSC clock, then passthrough TSC as stable
1431 host_tsc_clocksource = kvm_get_time_and_clockread(
1432 &ka->master_kernel_ns,
1433 &ka->master_cycle_now);
1435 ka->use_master_clock = host_tsc_clocksource & vcpus_matched;
1437 if (ka->use_master_clock)
1438 atomic_set(&kvm_guest_has_master_clock, 1);
1440 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1441 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1446 static int kvm_guest_time_update(struct kvm_vcpu *v)
1448 unsigned long flags, this_tsc_khz;
1449 struct kvm_vcpu_arch *vcpu = &v->arch;
1450 struct kvm_arch *ka = &v->kvm->arch;
1451 s64 kernel_ns, max_kernel_ns;
1452 u64 tsc_timestamp, host_tsc;
1453 struct pvclock_vcpu_time_info guest_hv_clock;
1455 bool use_master_clock;
1461 * If the host uses TSC clock, then passthrough TSC as stable
1464 spin_lock(&ka->pvclock_gtod_sync_lock);
1465 use_master_clock = ka->use_master_clock;
1466 if (use_master_clock) {
1467 host_tsc = ka->master_cycle_now;
1468 kernel_ns = ka->master_kernel_ns;
1470 spin_unlock(&ka->pvclock_gtod_sync_lock);
1472 /* Keep irq disabled to prevent changes to the clock */
1473 local_irq_save(flags);
1474 this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
1475 if (unlikely(this_tsc_khz == 0)) {
1476 local_irq_restore(flags);
1477 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1480 if (!use_master_clock) {
1481 host_tsc = native_read_tsc();
1482 kernel_ns = get_kernel_ns();
1485 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1488 * We may have to catch up the TSC to match elapsed wall clock
1489 * time for two reasons, even if kvmclock is used.
1490 * 1) CPU could have been running below the maximum TSC rate
1491 * 2) Broken TSC compensation resets the base at each VCPU
1492 * entry to avoid unknown leaps of TSC even when running
1493 * again on the same CPU. This may cause apparent elapsed
1494 * time to disappear, and the guest to stand still or run
1497 if (vcpu->tsc_catchup) {
1498 u64 tsc = compute_guest_tsc(v, kernel_ns);
1499 if (tsc > tsc_timestamp) {
1500 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1501 tsc_timestamp = tsc;
1505 local_irq_restore(flags);
1507 if (!vcpu->pv_time_enabled)
1511 * Time as measured by the TSC may go backwards when resetting the base
1512 * tsc_timestamp. The reason for this is that the TSC resolution is
1513 * higher than the resolution of the other clock scales. Thus, many
1514 * possible measurments of the TSC correspond to one measurement of any
1515 * other clock, and so a spread of values is possible. This is not a
1516 * problem for the computation of the nanosecond clock; with TSC rates
1517 * around 1GHZ, there can only be a few cycles which correspond to one
1518 * nanosecond value, and any path through this code will inevitably
1519 * take longer than that. However, with the kernel_ns value itself,
1520 * the precision may be much lower, down to HZ granularity. If the
1521 * first sampling of TSC against kernel_ns ends in the low part of the
1522 * range, and the second in the high end of the range, we can get:
1524 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1526 * As the sampling errors potentially range in the thousands of cycles,
1527 * it is possible such a time value has already been observed by the
1528 * guest. To protect against this, we must compute the system time as
1529 * observed by the guest and ensure the new system time is greater.
1532 if (vcpu->hv_clock.tsc_timestamp) {
1533 max_kernel_ns = vcpu->last_guest_tsc -
1534 vcpu->hv_clock.tsc_timestamp;
1535 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1536 vcpu->hv_clock.tsc_to_system_mul,
1537 vcpu->hv_clock.tsc_shift);
1538 max_kernel_ns += vcpu->last_kernel_ns;
1541 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1542 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1543 &vcpu->hv_clock.tsc_shift,
1544 &vcpu->hv_clock.tsc_to_system_mul);
1545 vcpu->hw_tsc_khz = this_tsc_khz;
1548 /* with a master <monotonic time, tsc value> tuple,
1549 * pvclock clock reads always increase at the (scaled) rate
1550 * of guest TSC - no need to deal with sampling errors.
1552 if (!use_master_clock) {
1553 if (max_kernel_ns > kernel_ns)
1554 kernel_ns = max_kernel_ns;
1556 /* With all the info we got, fill in the values */
1557 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1558 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1559 vcpu->last_kernel_ns = kernel_ns;
1560 vcpu->last_guest_tsc = tsc_timestamp;
1563 * The interface expects us to write an even number signaling that the
1564 * update is finished. Since the guest won't see the intermediate
1565 * state, we just increase by 2 at the end.
1567 vcpu->hv_clock.version += 2;
1569 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1570 &guest_hv_clock, sizeof(guest_hv_clock))))
1573 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1574 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1576 if (vcpu->pvclock_set_guest_stopped_request) {
1577 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1578 vcpu->pvclock_set_guest_stopped_request = false;
1581 /* If the host uses TSC clocksource, then it is stable */
1582 if (use_master_clock)
1583 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1585 vcpu->hv_clock.flags = pvclock_flags;
1587 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1589 sizeof(vcpu->hv_clock));
1593 static bool msr_mtrr_valid(unsigned msr)
1596 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1597 case MSR_MTRRfix64K_00000:
1598 case MSR_MTRRfix16K_80000:
1599 case MSR_MTRRfix16K_A0000:
1600 case MSR_MTRRfix4K_C0000:
1601 case MSR_MTRRfix4K_C8000:
1602 case MSR_MTRRfix4K_D0000:
1603 case MSR_MTRRfix4K_D8000:
1604 case MSR_MTRRfix4K_E0000:
1605 case MSR_MTRRfix4K_E8000:
1606 case MSR_MTRRfix4K_F0000:
1607 case MSR_MTRRfix4K_F8000:
1608 case MSR_MTRRdefType:
1609 case MSR_IA32_CR_PAT:
1617 static bool valid_pat_type(unsigned t)
1619 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1622 static bool valid_mtrr_type(unsigned t)
1624 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1627 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1631 if (!msr_mtrr_valid(msr))
1634 if (msr == MSR_IA32_CR_PAT) {
1635 for (i = 0; i < 8; i++)
1636 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1639 } else if (msr == MSR_MTRRdefType) {
1642 return valid_mtrr_type(data & 0xff);
1643 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1644 for (i = 0; i < 8 ; i++)
1645 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1650 /* variable MTRRs */
1651 return valid_mtrr_type(data & 0xff);
1654 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1656 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1658 if (!mtrr_valid(vcpu, msr, data))
1661 if (msr == MSR_MTRRdefType) {
1662 vcpu->arch.mtrr_state.def_type = data;
1663 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1664 } else if (msr == MSR_MTRRfix64K_00000)
1666 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1667 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1668 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1669 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1670 else if (msr == MSR_IA32_CR_PAT)
1671 vcpu->arch.pat = data;
1672 else { /* Variable MTRRs */
1673 int idx, is_mtrr_mask;
1676 idx = (msr - 0x200) / 2;
1677 is_mtrr_mask = msr - 0x200 - 2 * idx;
1680 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1683 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1687 kvm_mmu_reset_context(vcpu);
1691 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1693 u64 mcg_cap = vcpu->arch.mcg_cap;
1694 unsigned bank_num = mcg_cap & 0xff;
1697 case MSR_IA32_MCG_STATUS:
1698 vcpu->arch.mcg_status = data;
1700 case MSR_IA32_MCG_CTL:
1701 if (!(mcg_cap & MCG_CTL_P))
1703 if (data != 0 && data != ~(u64)0)
1705 vcpu->arch.mcg_ctl = data;
1708 if (msr >= MSR_IA32_MC0_CTL &&
1709 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1710 u32 offset = msr - MSR_IA32_MC0_CTL;
1711 /* only 0 or all 1s can be written to IA32_MCi_CTL
1712 * some Linux kernels though clear bit 10 in bank 4 to
1713 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1714 * this to avoid an uncatched #GP in the guest
1716 if ((offset & 0x3) == 0 &&
1717 data != 0 && (data | (1 << 10)) != ~(u64)0)
1719 vcpu->arch.mce_banks[offset] = data;
1727 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1729 struct kvm *kvm = vcpu->kvm;
1730 int lm = is_long_mode(vcpu);
1731 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1732 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1733 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1734 : kvm->arch.xen_hvm_config.blob_size_32;
1735 u32 page_num = data & ~PAGE_MASK;
1736 u64 page_addr = data & PAGE_MASK;
1741 if (page_num >= blob_size)
1744 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1749 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1758 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1760 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1763 static bool kvm_hv_msr_partition_wide(u32 msr)
1767 case HV_X64_MSR_GUEST_OS_ID:
1768 case HV_X64_MSR_HYPERCALL:
1776 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1778 struct kvm *kvm = vcpu->kvm;
1781 case HV_X64_MSR_GUEST_OS_ID:
1782 kvm->arch.hv_guest_os_id = data;
1783 /* setting guest os id to zero disables hypercall page */
1784 if (!kvm->arch.hv_guest_os_id)
1785 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1787 case HV_X64_MSR_HYPERCALL: {
1792 /* if guest os id is not set hypercall should remain disabled */
1793 if (!kvm->arch.hv_guest_os_id)
1795 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1796 kvm->arch.hv_hypercall = data;
1799 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1800 addr = gfn_to_hva(kvm, gfn);
1801 if (kvm_is_error_hva(addr))
1803 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1804 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1805 if (__copy_to_user((void __user *)addr, instructions, 4))
1807 kvm->arch.hv_hypercall = data;
1811 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1812 "data 0x%llx\n", msr, data);
1818 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1821 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1824 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1825 vcpu->arch.hv_vapic = data;
1828 addr = gfn_to_hva(vcpu->kvm, data >>
1829 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1830 if (kvm_is_error_hva(addr))
1832 if (__clear_user((void __user *)addr, PAGE_SIZE))
1834 vcpu->arch.hv_vapic = data;
1837 case HV_X64_MSR_EOI:
1838 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1839 case HV_X64_MSR_ICR:
1840 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1841 case HV_X64_MSR_TPR:
1842 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1844 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1845 "data 0x%llx\n", msr, data);
1852 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1854 gpa_t gpa = data & ~0x3f;
1856 /* Bits 2:5 are reserved, Should be zero */
1860 vcpu->arch.apf.msr_val = data;
1862 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1863 kvm_clear_async_pf_completion_queue(vcpu);
1864 kvm_async_pf_hash_reset(vcpu);
1868 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
1872 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1873 kvm_async_pf_wakeup_all(vcpu);
1877 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1879 vcpu->arch.pv_time_enabled = false;
1882 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1886 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1889 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1890 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1891 vcpu->arch.st.accum_steal = delta;
1894 static void record_steal_time(struct kvm_vcpu *vcpu)
1896 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1899 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1900 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1903 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1904 vcpu->arch.st.steal.version += 2;
1905 vcpu->arch.st.accum_steal = 0;
1907 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1908 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1911 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1914 u32 msr = msr_info->index;
1915 u64 data = msr_info->data;
1918 case MSR_AMD64_NB_CFG:
1919 case MSR_IA32_UCODE_REV:
1920 case MSR_IA32_UCODE_WRITE:
1921 case MSR_VM_HSAVE_PA:
1922 case MSR_AMD64_PATCH_LOADER:
1923 case MSR_AMD64_BU_CFG2:
1927 return set_efer(vcpu, data);
1929 data &= ~(u64)0x40; /* ignore flush filter disable */
1930 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1931 data &= ~(u64)0x8; /* ignore TLB cache disable */
1933 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1938 case MSR_FAM10H_MMIO_CONF_BASE:
1940 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1945 case MSR_IA32_DEBUGCTLMSR:
1947 /* We support the non-activated case already */
1949 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1950 /* Values other than LBR and BTF are vendor-specific,
1951 thus reserved and should throw a #GP */
1954 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1957 case 0x200 ... 0x2ff:
1958 return set_msr_mtrr(vcpu, msr, data);
1959 case MSR_IA32_APICBASE:
1960 kvm_set_apic_base(vcpu, data);
1962 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1963 return kvm_x2apic_msr_write(vcpu, msr, data);
1964 case MSR_IA32_TSCDEADLINE:
1965 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1967 case MSR_IA32_TSC_ADJUST:
1968 if (guest_cpuid_has_tsc_adjust(vcpu)) {
1969 if (!msr_info->host_initiated) {
1970 u64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
1971 kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
1973 vcpu->arch.ia32_tsc_adjust_msr = data;
1976 case MSR_IA32_MISC_ENABLE:
1977 vcpu->arch.ia32_misc_enable_msr = data;
1979 case MSR_KVM_WALL_CLOCK_NEW:
1980 case MSR_KVM_WALL_CLOCK:
1981 vcpu->kvm->arch.wall_clock = data;
1982 kvm_write_wall_clock(vcpu->kvm, data);
1984 case MSR_KVM_SYSTEM_TIME_NEW:
1985 case MSR_KVM_SYSTEM_TIME: {
1987 kvmclock_reset(vcpu);
1989 vcpu->arch.time = data;
1990 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1992 /* we verify if the enable bit is set... */
1996 gpa_offset = data & ~(PAGE_MASK | 1);
1998 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
1999 &vcpu->arch.pv_time, data & ~1ULL,
2000 sizeof(struct pvclock_vcpu_time_info)))
2001 vcpu->arch.pv_time_enabled = false;
2003 vcpu->arch.pv_time_enabled = true;
2007 case MSR_KVM_ASYNC_PF_EN:
2008 if (kvm_pv_enable_async_pf(vcpu, data))
2011 case MSR_KVM_STEAL_TIME:
2013 if (unlikely(!sched_info_on()))
2016 if (data & KVM_STEAL_RESERVED_MASK)
2019 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2020 data & KVM_STEAL_VALID_BITS,
2021 sizeof(struct kvm_steal_time)))
2024 vcpu->arch.st.msr_val = data;
2026 if (!(data & KVM_MSR_ENABLED))
2029 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2032 accumulate_steal_time(vcpu);
2035 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2038 case MSR_KVM_PV_EOI_EN:
2039 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2043 case MSR_IA32_MCG_CTL:
2044 case MSR_IA32_MCG_STATUS:
2045 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2046 return set_msr_mce(vcpu, msr, data);
2048 /* Performance counters are not protected by a CPUID bit,
2049 * so we should check all of them in the generic path for the sake of
2050 * cross vendor migration.
2051 * Writing a zero into the event select MSRs disables them,
2052 * which we perfectly emulate ;-). Any other value should be at least
2053 * reported, some guests depend on them.
2055 case MSR_K7_EVNTSEL0:
2056 case MSR_K7_EVNTSEL1:
2057 case MSR_K7_EVNTSEL2:
2058 case MSR_K7_EVNTSEL3:
2060 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2061 "0x%x data 0x%llx\n", msr, data);
2063 /* at least RHEL 4 unconditionally writes to the perfctr registers,
2064 * so we ignore writes to make it happy.
2066 case MSR_K7_PERFCTR0:
2067 case MSR_K7_PERFCTR1:
2068 case MSR_K7_PERFCTR2:
2069 case MSR_K7_PERFCTR3:
2070 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2071 "0x%x data 0x%llx\n", msr, data);
2073 case MSR_P6_PERFCTR0:
2074 case MSR_P6_PERFCTR1:
2076 case MSR_P6_EVNTSEL0:
2077 case MSR_P6_EVNTSEL1:
2078 if (kvm_pmu_msr(vcpu, msr))
2079 return kvm_pmu_set_msr(vcpu, msr_info);
2081 if (pr || data != 0)
2082 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2083 "0x%x data 0x%llx\n", msr, data);
2085 case MSR_K7_CLK_CTL:
2087 * Ignore all writes to this no longer documented MSR.
2088 * Writes are only relevant for old K7 processors,
2089 * all pre-dating SVM, but a recommended workaround from
2090 * AMD for these chips. It is possible to specify the
2091 * affected processor models on the command line, hence
2092 * the need to ignore the workaround.
2095 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2096 if (kvm_hv_msr_partition_wide(msr)) {
2098 mutex_lock(&vcpu->kvm->lock);
2099 r = set_msr_hyperv_pw(vcpu, msr, data);
2100 mutex_unlock(&vcpu->kvm->lock);
2103 return set_msr_hyperv(vcpu, msr, data);
2105 case MSR_IA32_BBL_CR_CTL3:
2106 /* Drop writes to this legacy MSR -- see rdmsr
2107 * counterpart for further detail.
2109 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2111 case MSR_AMD64_OSVW_ID_LENGTH:
2112 if (!guest_cpuid_has_osvw(vcpu))
2114 vcpu->arch.osvw.length = data;
2116 case MSR_AMD64_OSVW_STATUS:
2117 if (!guest_cpuid_has_osvw(vcpu))
2119 vcpu->arch.osvw.status = data;
2122 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2123 return xen_hvm_config(vcpu, data);
2124 if (kvm_pmu_msr(vcpu, msr))
2125 return kvm_pmu_set_msr(vcpu, msr_info);
2127 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2131 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2138 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2142 * Reads an msr value (of 'msr_index') into 'pdata'.
2143 * Returns 0 on success, non-0 otherwise.
2144 * Assumes vcpu_load() was already called.
2146 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2148 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
2151 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2153 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2155 if (!msr_mtrr_valid(msr))
2158 if (msr == MSR_MTRRdefType)
2159 *pdata = vcpu->arch.mtrr_state.def_type +
2160 (vcpu->arch.mtrr_state.enabled << 10);
2161 else if (msr == MSR_MTRRfix64K_00000)
2163 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2164 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2165 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2166 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2167 else if (msr == MSR_IA32_CR_PAT)
2168 *pdata = vcpu->arch.pat;
2169 else { /* Variable MTRRs */
2170 int idx, is_mtrr_mask;
2173 idx = (msr - 0x200) / 2;
2174 is_mtrr_mask = msr - 0x200 - 2 * idx;
2177 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2180 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2187 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2190 u64 mcg_cap = vcpu->arch.mcg_cap;
2191 unsigned bank_num = mcg_cap & 0xff;
2194 case MSR_IA32_P5_MC_ADDR:
2195 case MSR_IA32_P5_MC_TYPE:
2198 case MSR_IA32_MCG_CAP:
2199 data = vcpu->arch.mcg_cap;
2201 case MSR_IA32_MCG_CTL:
2202 if (!(mcg_cap & MCG_CTL_P))
2204 data = vcpu->arch.mcg_ctl;
2206 case MSR_IA32_MCG_STATUS:
2207 data = vcpu->arch.mcg_status;
2210 if (msr >= MSR_IA32_MC0_CTL &&
2211 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
2212 u32 offset = msr - MSR_IA32_MC0_CTL;
2213 data = vcpu->arch.mce_banks[offset];
2222 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2225 struct kvm *kvm = vcpu->kvm;
2228 case HV_X64_MSR_GUEST_OS_ID:
2229 data = kvm->arch.hv_guest_os_id;
2231 case HV_X64_MSR_HYPERCALL:
2232 data = kvm->arch.hv_hypercall;
2235 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2243 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2248 case HV_X64_MSR_VP_INDEX: {
2251 kvm_for_each_vcpu(r, v, vcpu->kvm)
2256 case HV_X64_MSR_EOI:
2257 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2258 case HV_X64_MSR_ICR:
2259 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2260 case HV_X64_MSR_TPR:
2261 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
2262 case HV_X64_MSR_APIC_ASSIST_PAGE:
2263 data = vcpu->arch.hv_vapic;
2266 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2273 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2278 case MSR_IA32_PLATFORM_ID:
2279 case MSR_IA32_EBL_CR_POWERON:
2280 case MSR_IA32_DEBUGCTLMSR:
2281 case MSR_IA32_LASTBRANCHFROMIP:
2282 case MSR_IA32_LASTBRANCHTOIP:
2283 case MSR_IA32_LASTINTFROMIP:
2284 case MSR_IA32_LASTINTTOIP:
2287 case MSR_VM_HSAVE_PA:
2288 case MSR_K7_EVNTSEL0:
2289 case MSR_K7_PERFCTR0:
2290 case MSR_K8_INT_PENDING_MSG:
2291 case MSR_AMD64_NB_CFG:
2292 case MSR_FAM10H_MMIO_CONF_BASE:
2293 case MSR_AMD64_BU_CFG2:
2296 case MSR_P6_PERFCTR0:
2297 case MSR_P6_PERFCTR1:
2298 case MSR_P6_EVNTSEL0:
2299 case MSR_P6_EVNTSEL1:
2300 if (kvm_pmu_msr(vcpu, msr))
2301 return kvm_pmu_get_msr(vcpu, msr, pdata);
2304 case MSR_IA32_UCODE_REV:
2305 data = 0x100000000ULL;
2308 data = 0x500 | KVM_NR_VAR_MTRR;
2310 case 0x200 ... 0x2ff:
2311 return get_msr_mtrr(vcpu, msr, pdata);
2312 case 0xcd: /* fsb frequency */
2316 * MSR_EBC_FREQUENCY_ID
2317 * Conservative value valid for even the basic CPU models.
2318 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2319 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2320 * and 266MHz for model 3, or 4. Set Core Clock
2321 * Frequency to System Bus Frequency Ratio to 1 (bits
2322 * 31:24) even though these are only valid for CPU
2323 * models > 2, however guests may end up dividing or
2324 * multiplying by zero otherwise.
2326 case MSR_EBC_FREQUENCY_ID:
2329 case MSR_IA32_APICBASE:
2330 data = kvm_get_apic_base(vcpu);
2332 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2333 return kvm_x2apic_msr_read(vcpu, msr, pdata);
2335 case MSR_IA32_TSCDEADLINE:
2336 data = kvm_get_lapic_tscdeadline_msr(vcpu);
2338 case MSR_IA32_TSC_ADJUST:
2339 data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2341 case MSR_IA32_MISC_ENABLE:
2342 data = vcpu->arch.ia32_misc_enable_msr;
2344 case MSR_IA32_PERF_STATUS:
2345 /* TSC increment by tick */
2347 /* CPU multiplier */
2348 data |= (((uint64_t)4ULL) << 40);
2351 data = vcpu->arch.efer;
2353 case MSR_KVM_WALL_CLOCK:
2354 case MSR_KVM_WALL_CLOCK_NEW:
2355 data = vcpu->kvm->arch.wall_clock;
2357 case MSR_KVM_SYSTEM_TIME:
2358 case MSR_KVM_SYSTEM_TIME_NEW:
2359 data = vcpu->arch.time;
2361 case MSR_KVM_ASYNC_PF_EN:
2362 data = vcpu->arch.apf.msr_val;
2364 case MSR_KVM_STEAL_TIME:
2365 data = vcpu->arch.st.msr_val;
2367 case MSR_KVM_PV_EOI_EN:
2368 data = vcpu->arch.pv_eoi.msr_val;
2370 case MSR_IA32_P5_MC_ADDR:
2371 case MSR_IA32_P5_MC_TYPE:
2372 case MSR_IA32_MCG_CAP:
2373 case MSR_IA32_MCG_CTL:
2374 case MSR_IA32_MCG_STATUS:
2375 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
2376 return get_msr_mce(vcpu, msr, pdata);
2377 case MSR_K7_CLK_CTL:
2379 * Provide expected ramp-up count for K7. All other
2380 * are set to zero, indicating minimum divisors for
2383 * This prevents guest kernels on AMD host with CPU
2384 * type 6, model 8 and higher from exploding due to
2385 * the rdmsr failing.
2389 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2390 if (kvm_hv_msr_partition_wide(msr)) {
2392 mutex_lock(&vcpu->kvm->lock);
2393 r = get_msr_hyperv_pw(vcpu, msr, pdata);
2394 mutex_unlock(&vcpu->kvm->lock);
2397 return get_msr_hyperv(vcpu, msr, pdata);
2399 case MSR_IA32_BBL_CR_CTL3:
2400 /* This legacy MSR exists but isn't fully documented in current
2401 * silicon. It is however accessed by winxp in very narrow
2402 * scenarios where it sets bit #19, itself documented as
2403 * a "reserved" bit. Best effort attempt to source coherent
2404 * read data here should the balance of the register be
2405 * interpreted by the guest:
2407 * L2 cache control register 3: 64GB range, 256KB size,
2408 * enabled, latency 0x1, configured
2412 case MSR_AMD64_OSVW_ID_LENGTH:
2413 if (!guest_cpuid_has_osvw(vcpu))
2415 data = vcpu->arch.osvw.length;
2417 case MSR_AMD64_OSVW_STATUS:
2418 if (!guest_cpuid_has_osvw(vcpu))
2420 data = vcpu->arch.osvw.status;
2423 if (kvm_pmu_msr(vcpu, msr))
2424 return kvm_pmu_get_msr(vcpu, msr, pdata);
2426 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2429 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2437 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2440 * Read or write a bunch of msrs. All parameters are kernel addresses.
2442 * @return number of msrs set successfully.
2444 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2445 struct kvm_msr_entry *entries,
2446 int (*do_msr)(struct kvm_vcpu *vcpu,
2447 unsigned index, u64 *data))
2451 idx = srcu_read_lock(&vcpu->kvm->srcu);
2452 for (i = 0; i < msrs->nmsrs; ++i)
2453 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2455 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2461 * Read or write a bunch of msrs. Parameters are user addresses.
2463 * @return number of msrs set successfully.
2465 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2466 int (*do_msr)(struct kvm_vcpu *vcpu,
2467 unsigned index, u64 *data),
2470 struct kvm_msrs msrs;
2471 struct kvm_msr_entry *entries;
2476 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2480 if (msrs.nmsrs >= MAX_IO_MSRS)
2483 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2484 entries = memdup_user(user_msrs->entries, size);
2485 if (IS_ERR(entries)) {
2486 r = PTR_ERR(entries);
2490 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2495 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2506 int kvm_dev_ioctl_check_extension(long ext)
2511 case KVM_CAP_IRQCHIP:
2513 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2514 case KVM_CAP_SET_TSS_ADDR:
2515 case KVM_CAP_EXT_CPUID:
2516 case KVM_CAP_CLOCKSOURCE:
2518 case KVM_CAP_NOP_IO_DELAY:
2519 case KVM_CAP_MP_STATE:
2520 case KVM_CAP_SYNC_MMU:
2521 case KVM_CAP_USER_NMI:
2522 case KVM_CAP_REINJECT_CONTROL:
2523 case KVM_CAP_IRQ_INJECT_STATUS:
2525 case KVM_CAP_IOEVENTFD:
2527 case KVM_CAP_PIT_STATE2:
2528 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2529 case KVM_CAP_XEN_HVM:
2530 case KVM_CAP_ADJUST_CLOCK:
2531 case KVM_CAP_VCPU_EVENTS:
2532 case KVM_CAP_HYPERV:
2533 case KVM_CAP_HYPERV_VAPIC:
2534 case KVM_CAP_HYPERV_SPIN:
2535 case KVM_CAP_PCI_SEGMENT:
2536 case KVM_CAP_DEBUGREGS:
2537 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2539 case KVM_CAP_ASYNC_PF:
2540 case KVM_CAP_GET_TSC_KHZ:
2541 case KVM_CAP_KVMCLOCK_CTRL:
2542 case KVM_CAP_READONLY_MEM:
2543 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2544 case KVM_CAP_ASSIGN_DEV_IRQ:
2545 case KVM_CAP_PCI_2_3:
2549 case KVM_CAP_COALESCED_MMIO:
2550 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2553 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2555 case KVM_CAP_NR_VCPUS:
2556 r = KVM_SOFT_MAX_VCPUS;
2558 case KVM_CAP_MAX_VCPUS:
2561 case KVM_CAP_NR_MEMSLOTS:
2562 r = KVM_USER_MEM_SLOTS;
2564 case KVM_CAP_PV_MMU: /* obsolete */
2567 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2569 r = iommu_present(&pci_bus_type);
2573 r = KVM_MAX_MCE_BANKS;
2578 case KVM_CAP_TSC_CONTROL:
2579 r = kvm_has_tsc_control;
2581 case KVM_CAP_TSC_DEADLINE_TIMER:
2582 r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2592 long kvm_arch_dev_ioctl(struct file *filp,
2593 unsigned int ioctl, unsigned long arg)
2595 void __user *argp = (void __user *)arg;
2599 case KVM_GET_MSR_INDEX_LIST: {
2600 struct kvm_msr_list __user *user_msr_list = argp;
2601 struct kvm_msr_list msr_list;
2605 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2608 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2609 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2612 if (n < msr_list.nmsrs)
2615 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2616 num_msrs_to_save * sizeof(u32)))
2618 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2620 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2625 case KVM_GET_SUPPORTED_CPUID: {
2626 struct kvm_cpuid2 __user *cpuid_arg = argp;
2627 struct kvm_cpuid2 cpuid;
2630 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2632 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2633 cpuid_arg->entries);
2638 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2643 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2646 mce_cap = KVM_MCE_CAP_SUPPORTED;
2648 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2660 static void wbinvd_ipi(void *garbage)
2665 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2667 return vcpu->kvm->arch.iommu_domain &&
2668 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2671 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2673 /* Address WBINVD may be executed by guest */
2674 if (need_emulate_wbinvd(vcpu)) {
2675 if (kvm_x86_ops->has_wbinvd_exit())
2676 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2677 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2678 smp_call_function_single(vcpu->cpu,
2679 wbinvd_ipi, NULL, 1);
2682 kvm_x86_ops->vcpu_load(vcpu, cpu);
2684 /* Apply any externally detected TSC adjustments (due to suspend) */
2685 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2686 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2687 vcpu->arch.tsc_offset_adjustment = 0;
2688 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
2691 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2692 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2693 native_read_tsc() - vcpu->arch.last_host_tsc;
2695 mark_tsc_unstable("KVM discovered backwards TSC");
2696 if (check_tsc_unstable()) {
2697 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2698 vcpu->arch.last_guest_tsc);
2699 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2700 vcpu->arch.tsc_catchup = 1;
2703 * On a host with synchronized TSC, there is no need to update
2704 * kvmclock on vcpu->cpu migration
2706 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2707 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2708 if (vcpu->cpu != cpu)
2709 kvm_migrate_timers(vcpu);
2713 accumulate_steal_time(vcpu);
2714 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2717 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2719 kvm_x86_ops->vcpu_put(vcpu);
2720 kvm_put_guest_fpu(vcpu);
2721 vcpu->arch.last_host_tsc = native_read_tsc();
2724 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2725 struct kvm_lapic_state *s)
2727 kvm_x86_ops->sync_pir_to_irr(vcpu);
2728 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2733 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2734 struct kvm_lapic_state *s)
2736 kvm_apic_post_state_restore(vcpu, s);
2737 update_cr8_intercept(vcpu);
2742 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2743 struct kvm_interrupt *irq)
2745 if (irq->irq >= KVM_NR_INTERRUPTS)
2747 if (irqchip_in_kernel(vcpu->kvm))
2750 kvm_queue_interrupt(vcpu, irq->irq, false);
2751 kvm_make_request(KVM_REQ_EVENT, vcpu);
2756 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2758 kvm_inject_nmi(vcpu);
2763 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2764 struct kvm_tpr_access_ctl *tac)
2768 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2772 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2776 unsigned bank_num = mcg_cap & 0xff, bank;
2779 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2781 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2784 vcpu->arch.mcg_cap = mcg_cap;
2785 /* Init IA32_MCG_CTL to all 1s */
2786 if (mcg_cap & MCG_CTL_P)
2787 vcpu->arch.mcg_ctl = ~(u64)0;
2788 /* Init IA32_MCi_CTL to all 1s */
2789 for (bank = 0; bank < bank_num; bank++)
2790 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2795 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2796 struct kvm_x86_mce *mce)
2798 u64 mcg_cap = vcpu->arch.mcg_cap;
2799 unsigned bank_num = mcg_cap & 0xff;
2800 u64 *banks = vcpu->arch.mce_banks;
2802 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2805 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2806 * reporting is disabled
2808 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2809 vcpu->arch.mcg_ctl != ~(u64)0)
2811 banks += 4 * mce->bank;
2813 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2814 * reporting is disabled for the bank
2816 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2818 if (mce->status & MCI_STATUS_UC) {
2819 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2820 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2821 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2824 if (banks[1] & MCI_STATUS_VAL)
2825 mce->status |= MCI_STATUS_OVER;
2826 banks[2] = mce->addr;
2827 banks[3] = mce->misc;
2828 vcpu->arch.mcg_status = mce->mcg_status;
2829 banks[1] = mce->status;
2830 kvm_queue_exception(vcpu, MC_VECTOR);
2831 } else if (!(banks[1] & MCI_STATUS_VAL)
2832 || !(banks[1] & MCI_STATUS_UC)) {
2833 if (banks[1] & MCI_STATUS_VAL)
2834 mce->status |= MCI_STATUS_OVER;
2835 banks[2] = mce->addr;
2836 banks[3] = mce->misc;
2837 banks[1] = mce->status;
2839 banks[1] |= MCI_STATUS_OVER;
2843 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2844 struct kvm_vcpu_events *events)
2847 events->exception.injected =
2848 vcpu->arch.exception.pending &&
2849 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2850 events->exception.nr = vcpu->arch.exception.nr;
2851 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2852 events->exception.pad = 0;
2853 events->exception.error_code = vcpu->arch.exception.error_code;
2855 events->interrupt.injected =
2856 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2857 events->interrupt.nr = vcpu->arch.interrupt.nr;
2858 events->interrupt.soft = 0;
2859 events->interrupt.shadow =
2860 kvm_x86_ops->get_interrupt_shadow(vcpu,
2861 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2863 events->nmi.injected = vcpu->arch.nmi_injected;
2864 events->nmi.pending = vcpu->arch.nmi_pending != 0;
2865 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2866 events->nmi.pad = 0;
2868 events->sipi_vector = 0; /* never valid when reporting to user space */
2870 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2871 | KVM_VCPUEVENT_VALID_SHADOW);
2872 memset(&events->reserved, 0, sizeof(events->reserved));
2875 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2876 struct kvm_vcpu_events *events)
2878 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2879 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2880 | KVM_VCPUEVENT_VALID_SHADOW))
2884 vcpu->arch.exception.pending = events->exception.injected;
2885 vcpu->arch.exception.nr = events->exception.nr;
2886 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2887 vcpu->arch.exception.error_code = events->exception.error_code;
2889 vcpu->arch.interrupt.pending = events->interrupt.injected;
2890 vcpu->arch.interrupt.nr = events->interrupt.nr;
2891 vcpu->arch.interrupt.soft = events->interrupt.soft;
2892 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2893 kvm_x86_ops->set_interrupt_shadow(vcpu,
2894 events->interrupt.shadow);
2896 vcpu->arch.nmi_injected = events->nmi.injected;
2897 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2898 vcpu->arch.nmi_pending = events->nmi.pending;
2899 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2901 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
2902 kvm_vcpu_has_lapic(vcpu))
2903 vcpu->arch.apic->sipi_vector = events->sipi_vector;
2905 kvm_make_request(KVM_REQ_EVENT, vcpu);
2910 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2911 struct kvm_debugregs *dbgregs)
2913 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2914 dbgregs->dr6 = vcpu->arch.dr6;
2915 dbgregs->dr7 = vcpu->arch.dr7;
2917 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2920 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2921 struct kvm_debugregs *dbgregs)
2926 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2927 vcpu->arch.dr6 = dbgregs->dr6;
2928 vcpu->arch.dr7 = dbgregs->dr7;
2933 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2934 struct kvm_xsave *guest_xsave)
2937 memcpy(guest_xsave->region,
2938 &vcpu->arch.guest_fpu.state->xsave,
2941 memcpy(guest_xsave->region,
2942 &vcpu->arch.guest_fpu.state->fxsave,
2943 sizeof(struct i387_fxsave_struct));
2944 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2949 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2950 struct kvm_xsave *guest_xsave)
2953 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2956 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2957 guest_xsave->region, xstate_size);
2959 if (xstate_bv & ~XSTATE_FPSSE)
2961 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2962 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2967 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2968 struct kvm_xcrs *guest_xcrs)
2970 if (!cpu_has_xsave) {
2971 guest_xcrs->nr_xcrs = 0;
2975 guest_xcrs->nr_xcrs = 1;
2976 guest_xcrs->flags = 0;
2977 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2978 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2981 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2982 struct kvm_xcrs *guest_xcrs)
2989 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2992 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2993 /* Only support XCR0 currently */
2994 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2995 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2996 guest_xcrs->xcrs[0].value);
3005 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3006 * stopped by the hypervisor. This function will be called from the host only.
3007 * EINVAL is returned when the host attempts to set the flag for a guest that
3008 * does not support pv clocks.
3010 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3012 if (!vcpu->arch.pv_time_enabled)
3014 vcpu->arch.pvclock_set_guest_stopped_request = true;
3015 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3019 long kvm_arch_vcpu_ioctl(struct file *filp,
3020 unsigned int ioctl, unsigned long arg)
3022 struct kvm_vcpu *vcpu = filp->private_data;
3023 void __user *argp = (void __user *)arg;
3026 struct kvm_lapic_state *lapic;
3027 struct kvm_xsave *xsave;
3028 struct kvm_xcrs *xcrs;
3034 case KVM_GET_LAPIC: {
3036 if (!vcpu->arch.apic)
3038 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3043 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3047 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3052 case KVM_SET_LAPIC: {
3054 if (!vcpu->arch.apic)
3056 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3057 if (IS_ERR(u.lapic))
3058 return PTR_ERR(u.lapic);
3060 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3063 case KVM_INTERRUPT: {
3064 struct kvm_interrupt irq;
3067 if (copy_from_user(&irq, argp, sizeof irq))
3069 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3073 r = kvm_vcpu_ioctl_nmi(vcpu);
3076 case KVM_SET_CPUID: {
3077 struct kvm_cpuid __user *cpuid_arg = argp;
3078 struct kvm_cpuid cpuid;
3081 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3083 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3086 case KVM_SET_CPUID2: {
3087 struct kvm_cpuid2 __user *cpuid_arg = argp;
3088 struct kvm_cpuid2 cpuid;
3091 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3093 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3094 cpuid_arg->entries);
3097 case KVM_GET_CPUID2: {
3098 struct kvm_cpuid2 __user *cpuid_arg = argp;
3099 struct kvm_cpuid2 cpuid;
3102 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3104 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3105 cpuid_arg->entries);
3109 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3115 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3118 r = msr_io(vcpu, argp, do_set_msr, 0);
3120 case KVM_TPR_ACCESS_REPORTING: {
3121 struct kvm_tpr_access_ctl tac;
3124 if (copy_from_user(&tac, argp, sizeof tac))
3126 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3130 if (copy_to_user(argp, &tac, sizeof tac))
3135 case KVM_SET_VAPIC_ADDR: {
3136 struct kvm_vapic_addr va;
3139 if (!irqchip_in_kernel(vcpu->kvm))
3142 if (copy_from_user(&va, argp, sizeof va))
3144 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3147 case KVM_X86_SETUP_MCE: {
3151 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3153 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3156 case KVM_X86_SET_MCE: {
3157 struct kvm_x86_mce mce;
3160 if (copy_from_user(&mce, argp, sizeof mce))
3162 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3165 case KVM_GET_VCPU_EVENTS: {
3166 struct kvm_vcpu_events events;
3168 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3171 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3176 case KVM_SET_VCPU_EVENTS: {
3177 struct kvm_vcpu_events events;
3180 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3183 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3186 case KVM_GET_DEBUGREGS: {
3187 struct kvm_debugregs dbgregs;
3189 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3192 if (copy_to_user(argp, &dbgregs,
3193 sizeof(struct kvm_debugregs)))
3198 case KVM_SET_DEBUGREGS: {
3199 struct kvm_debugregs dbgregs;
3202 if (copy_from_user(&dbgregs, argp,
3203 sizeof(struct kvm_debugregs)))
3206 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3209 case KVM_GET_XSAVE: {
3210 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3215 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3218 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3223 case KVM_SET_XSAVE: {
3224 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3225 if (IS_ERR(u.xsave))
3226 return PTR_ERR(u.xsave);
3228 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3231 case KVM_GET_XCRS: {
3232 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3237 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3240 if (copy_to_user(argp, u.xcrs,
3241 sizeof(struct kvm_xcrs)))
3246 case KVM_SET_XCRS: {
3247 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3249 return PTR_ERR(u.xcrs);
3251 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3254 case KVM_SET_TSC_KHZ: {
3258 user_tsc_khz = (u32)arg;
3260 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3263 if (user_tsc_khz == 0)
3264 user_tsc_khz = tsc_khz;
3266 kvm_set_tsc_khz(vcpu, user_tsc_khz);
3271 case KVM_GET_TSC_KHZ: {
3272 r = vcpu->arch.virtual_tsc_khz;
3275 case KVM_KVMCLOCK_CTRL: {
3276 r = kvm_set_guest_paused(vcpu);
3287 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3289 return VM_FAULT_SIGBUS;
3292 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3296 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3298 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3302 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3305 kvm->arch.ept_identity_map_addr = ident_addr;
3309 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3310 u32 kvm_nr_mmu_pages)
3312 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3315 mutex_lock(&kvm->slots_lock);
3317 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3318 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3320 mutex_unlock(&kvm->slots_lock);
3324 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3326 return kvm->arch.n_max_mmu_pages;
3329 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3334 switch (chip->chip_id) {
3335 case KVM_IRQCHIP_PIC_MASTER:
3336 memcpy(&chip->chip.pic,
3337 &pic_irqchip(kvm)->pics[0],
3338 sizeof(struct kvm_pic_state));
3340 case KVM_IRQCHIP_PIC_SLAVE:
3341 memcpy(&chip->chip.pic,
3342 &pic_irqchip(kvm)->pics[1],
3343 sizeof(struct kvm_pic_state));
3345 case KVM_IRQCHIP_IOAPIC:
3346 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3355 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3360 switch (chip->chip_id) {
3361 case KVM_IRQCHIP_PIC_MASTER:
3362 spin_lock(&pic_irqchip(kvm)->lock);
3363 memcpy(&pic_irqchip(kvm)->pics[0],
3365 sizeof(struct kvm_pic_state));
3366 spin_unlock(&pic_irqchip(kvm)->lock);
3368 case KVM_IRQCHIP_PIC_SLAVE:
3369 spin_lock(&pic_irqchip(kvm)->lock);
3370 memcpy(&pic_irqchip(kvm)->pics[1],
3372 sizeof(struct kvm_pic_state));
3373 spin_unlock(&pic_irqchip(kvm)->lock);
3375 case KVM_IRQCHIP_IOAPIC:
3376 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3382 kvm_pic_update_irq(pic_irqchip(kvm));
3386 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3390 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3391 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3392 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3396 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3400 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3401 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3402 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3403 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3407 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3411 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3412 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3413 sizeof(ps->channels));
3414 ps->flags = kvm->arch.vpit->pit_state.flags;
3415 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3416 memset(&ps->reserved, 0, sizeof(ps->reserved));
3420 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3422 int r = 0, start = 0;
3423 u32 prev_legacy, cur_legacy;
3424 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3425 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3426 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3427 if (!prev_legacy && cur_legacy)
3429 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3430 sizeof(kvm->arch.vpit->pit_state.channels));
3431 kvm->arch.vpit->pit_state.flags = ps->flags;
3432 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3433 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3437 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3438 struct kvm_reinject_control *control)
3440 if (!kvm->arch.vpit)
3442 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3443 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3444 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3449 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3450 * @kvm: kvm instance
3451 * @log: slot id and address to which we copy the log
3453 * We need to keep it in mind that VCPU threads can write to the bitmap
3454 * concurrently. So, to avoid losing data, we keep the following order for
3457 * 1. Take a snapshot of the bit and clear it if needed.
3458 * 2. Write protect the corresponding page.
3459 * 3. Flush TLB's if needed.
3460 * 4. Copy the snapshot to the userspace.
3462 * Between 2 and 3, the guest may write to the page using the remaining TLB
3463 * entry. This is not a problem because the page will be reported dirty at
3464 * step 4 using the snapshot taken before and step 3 ensures that successive
3465 * writes will be logged for the next call.
3467 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3470 struct kvm_memory_slot *memslot;
3472 unsigned long *dirty_bitmap;
3473 unsigned long *dirty_bitmap_buffer;
3474 bool is_dirty = false;
3476 mutex_lock(&kvm->slots_lock);
3479 if (log->slot >= KVM_USER_MEM_SLOTS)
3482 memslot = id_to_memslot(kvm->memslots, log->slot);
3484 dirty_bitmap = memslot->dirty_bitmap;
3489 n = kvm_dirty_bitmap_bytes(memslot);
3491 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
3492 memset(dirty_bitmap_buffer, 0, n);
3494 spin_lock(&kvm->mmu_lock);
3496 for (i = 0; i < n / sizeof(long); i++) {
3500 if (!dirty_bitmap[i])
3505 mask = xchg(&dirty_bitmap[i], 0);
3506 dirty_bitmap_buffer[i] = mask;
3508 offset = i * BITS_PER_LONG;
3509 kvm_mmu_write_protect_pt_masked(kvm, memslot, offset, mask);
3512 kvm_flush_remote_tlbs(kvm);
3514 spin_unlock(&kvm->mmu_lock);
3517 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
3522 mutex_unlock(&kvm->slots_lock);
3526 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3529 if (!irqchip_in_kernel(kvm))
3532 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3533 irq_event->irq, irq_event->level,
3538 long kvm_arch_vm_ioctl(struct file *filp,
3539 unsigned int ioctl, unsigned long arg)
3541 struct kvm *kvm = filp->private_data;
3542 void __user *argp = (void __user *)arg;
3545 * This union makes it completely explicit to gcc-3.x
3546 * that these two variables' stack usage should be
3547 * combined, not added together.
3550 struct kvm_pit_state ps;
3551 struct kvm_pit_state2 ps2;
3552 struct kvm_pit_config pit_config;
3556 case KVM_SET_TSS_ADDR:
3557 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3559 case KVM_SET_IDENTITY_MAP_ADDR: {
3563 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3565 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3568 case KVM_SET_NR_MMU_PAGES:
3569 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3571 case KVM_GET_NR_MMU_PAGES:
3572 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3574 case KVM_CREATE_IRQCHIP: {
3575 struct kvm_pic *vpic;
3577 mutex_lock(&kvm->lock);
3580 goto create_irqchip_unlock;
3582 if (atomic_read(&kvm->online_vcpus))
3583 goto create_irqchip_unlock;
3585 vpic = kvm_create_pic(kvm);
3587 r = kvm_ioapic_init(kvm);
3589 mutex_lock(&kvm->slots_lock);
3590 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3592 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3594 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3596 mutex_unlock(&kvm->slots_lock);
3598 goto create_irqchip_unlock;
3601 goto create_irqchip_unlock;
3603 kvm->arch.vpic = vpic;
3605 r = kvm_setup_default_irq_routing(kvm);
3607 mutex_lock(&kvm->slots_lock);
3608 mutex_lock(&kvm->irq_lock);
3609 kvm_ioapic_destroy(kvm);
3610 kvm_destroy_pic(kvm);
3611 mutex_unlock(&kvm->irq_lock);
3612 mutex_unlock(&kvm->slots_lock);
3614 create_irqchip_unlock:
3615 mutex_unlock(&kvm->lock);
3618 case KVM_CREATE_PIT:
3619 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3621 case KVM_CREATE_PIT2:
3623 if (copy_from_user(&u.pit_config, argp,
3624 sizeof(struct kvm_pit_config)))
3627 mutex_lock(&kvm->slots_lock);
3630 goto create_pit_unlock;
3632 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3636 mutex_unlock(&kvm->slots_lock);
3638 case KVM_GET_IRQCHIP: {
3639 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3640 struct kvm_irqchip *chip;
3642 chip = memdup_user(argp, sizeof(*chip));
3649 if (!irqchip_in_kernel(kvm))
3650 goto get_irqchip_out;
3651 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3653 goto get_irqchip_out;
3655 if (copy_to_user(argp, chip, sizeof *chip))
3656 goto get_irqchip_out;
3662 case KVM_SET_IRQCHIP: {
3663 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3664 struct kvm_irqchip *chip;
3666 chip = memdup_user(argp, sizeof(*chip));
3673 if (!irqchip_in_kernel(kvm))
3674 goto set_irqchip_out;
3675 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3677 goto set_irqchip_out;
3685 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3688 if (!kvm->arch.vpit)
3690 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3694 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3701 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3704 if (!kvm->arch.vpit)
3706 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3709 case KVM_GET_PIT2: {
3711 if (!kvm->arch.vpit)
3713 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3717 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3722 case KVM_SET_PIT2: {
3724 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3727 if (!kvm->arch.vpit)
3729 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3732 case KVM_REINJECT_CONTROL: {
3733 struct kvm_reinject_control control;
3735 if (copy_from_user(&control, argp, sizeof(control)))
3737 r = kvm_vm_ioctl_reinject(kvm, &control);
3740 case KVM_XEN_HVM_CONFIG: {
3742 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3743 sizeof(struct kvm_xen_hvm_config)))
3746 if (kvm->arch.xen_hvm_config.flags)
3751 case KVM_SET_CLOCK: {
3752 struct kvm_clock_data user_ns;
3757 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3765 local_irq_disable();
3766 now_ns = get_kernel_ns();
3767 delta = user_ns.clock - now_ns;
3769 kvm->arch.kvmclock_offset = delta;
3772 case KVM_GET_CLOCK: {
3773 struct kvm_clock_data user_ns;
3776 local_irq_disable();
3777 now_ns = get_kernel_ns();
3778 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3781 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3784 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3797 static void kvm_init_msr_list(void)
3802 /* skip the first msrs in the list. KVM-specific */
3803 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3804 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3807 msrs_to_save[j] = msrs_to_save[i];
3810 num_msrs_to_save = j;
3813 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3821 if (!(vcpu->arch.apic &&
3822 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3823 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3834 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3841 if (!(vcpu->arch.apic &&
3842 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3843 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3845 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3855 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3856 struct kvm_segment *var, int seg)
3858 kvm_x86_ops->set_segment(vcpu, var, seg);
3861 void kvm_get_segment(struct kvm_vcpu *vcpu,
3862 struct kvm_segment *var, int seg)
3864 kvm_x86_ops->get_segment(vcpu, var, seg);
3867 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3870 struct x86_exception exception;
3872 BUG_ON(!mmu_is_nested(vcpu));
3874 /* NPT walks are always user-walks */
3875 access |= PFERR_USER_MASK;
3876 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3881 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3882 struct x86_exception *exception)
3884 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3885 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3888 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3889 struct x86_exception *exception)
3891 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3892 access |= PFERR_FETCH_MASK;
3893 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3896 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3897 struct x86_exception *exception)
3899 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3900 access |= PFERR_WRITE_MASK;
3901 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3904 /* uses this to access any guest's mapped memory without checking CPL */
3905 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3906 struct x86_exception *exception)
3908 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3911 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3912 struct kvm_vcpu *vcpu, u32 access,
3913 struct x86_exception *exception)
3916 int r = X86EMUL_CONTINUE;
3919 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3921 unsigned offset = addr & (PAGE_SIZE-1);
3922 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3925 if (gpa == UNMAPPED_GVA)
3926 return X86EMUL_PROPAGATE_FAULT;
3927 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3929 r = X86EMUL_IO_NEEDED;
3941 /* used for instruction fetching */
3942 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3943 gva_t addr, void *val, unsigned int bytes,
3944 struct x86_exception *exception)
3946 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3947 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3949 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3950 access | PFERR_FETCH_MASK,
3954 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3955 gva_t addr, void *val, unsigned int bytes,
3956 struct x86_exception *exception)
3958 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3959 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3961 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3964 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
3966 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3967 gva_t addr, void *val, unsigned int bytes,
3968 struct x86_exception *exception)
3970 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3971 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3974 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3975 gva_t addr, void *val,
3977 struct x86_exception *exception)
3979 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3981 int r = X86EMUL_CONTINUE;
3984 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3987 unsigned offset = addr & (PAGE_SIZE-1);
3988 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3991 if (gpa == UNMAPPED_GVA)
3992 return X86EMUL_PROPAGATE_FAULT;
3993 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3995 r = X86EMUL_IO_NEEDED;
4006 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4008 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4009 gpa_t *gpa, struct x86_exception *exception,
4012 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4013 | (write ? PFERR_WRITE_MASK : 0);
4015 if (vcpu_match_mmio_gva(vcpu, gva)
4016 && !permission_fault(vcpu->arch.walk_mmu, vcpu->arch.access, access)) {
4017 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4018 (gva & (PAGE_SIZE - 1));
4019 trace_vcpu_match_mmio(gva, *gpa, write, false);
4023 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4025 if (*gpa == UNMAPPED_GVA)
4028 /* For APIC access vmexit */
4029 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4032 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4033 trace_vcpu_match_mmio(gva, *gpa, write, true);
4040 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4041 const void *val, int bytes)
4045 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4048 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4052 struct read_write_emulator_ops {
4053 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4055 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4056 void *val, int bytes);
4057 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4058 int bytes, void *val);
4059 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4060 void *val, int bytes);
4064 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4066 if (vcpu->mmio_read_completed) {
4067 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4068 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4069 vcpu->mmio_read_completed = 0;
4076 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4077 void *val, int bytes)
4079 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4082 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4083 void *val, int bytes)
4085 return emulator_write_phys(vcpu, gpa, val, bytes);
4088 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4090 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4091 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4094 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4095 void *val, int bytes)
4097 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4098 return X86EMUL_IO_NEEDED;
4101 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4102 void *val, int bytes)
4104 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4106 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4107 return X86EMUL_CONTINUE;
4110 static const struct read_write_emulator_ops read_emultor = {
4111 .read_write_prepare = read_prepare,
4112 .read_write_emulate = read_emulate,
4113 .read_write_mmio = vcpu_mmio_read,
4114 .read_write_exit_mmio = read_exit_mmio,
4117 static const struct read_write_emulator_ops write_emultor = {
4118 .read_write_emulate = write_emulate,
4119 .read_write_mmio = write_mmio,
4120 .read_write_exit_mmio = write_exit_mmio,
4124 static int emulator_read_write_onepage(unsigned long addr, void *val,
4126 struct x86_exception *exception,
4127 struct kvm_vcpu *vcpu,
4128 const struct read_write_emulator_ops *ops)
4132 bool write = ops->write;
4133 struct kvm_mmio_fragment *frag;
4135 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4138 return X86EMUL_PROPAGATE_FAULT;
4140 /* For APIC access vmexit */
4144 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4145 return X86EMUL_CONTINUE;
4149 * Is this MMIO handled locally?
4151 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4152 if (handled == bytes)
4153 return X86EMUL_CONTINUE;
4159 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4160 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4164 return X86EMUL_CONTINUE;
4167 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
4168 void *val, unsigned int bytes,
4169 struct x86_exception *exception,
4170 const struct read_write_emulator_ops *ops)
4172 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4176 if (ops->read_write_prepare &&
4177 ops->read_write_prepare(vcpu, val, bytes))
4178 return X86EMUL_CONTINUE;
4180 vcpu->mmio_nr_fragments = 0;
4182 /* Crossing a page boundary? */
4183 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4186 now = -addr & ~PAGE_MASK;
4187 rc = emulator_read_write_onepage(addr, val, now, exception,
4190 if (rc != X86EMUL_CONTINUE)
4197 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4199 if (rc != X86EMUL_CONTINUE)
4202 if (!vcpu->mmio_nr_fragments)
4205 gpa = vcpu->mmio_fragments[0].gpa;
4207 vcpu->mmio_needed = 1;
4208 vcpu->mmio_cur_fragment = 0;
4210 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4211 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4212 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4213 vcpu->run->mmio.phys_addr = gpa;
4215 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4218 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4222 struct x86_exception *exception)
4224 return emulator_read_write(ctxt, addr, val, bytes,
4225 exception, &read_emultor);
4228 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4232 struct x86_exception *exception)
4234 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4235 exception, &write_emultor);
4238 #define CMPXCHG_TYPE(t, ptr, old, new) \
4239 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4241 #ifdef CONFIG_X86_64
4242 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4244 # define CMPXCHG64(ptr, old, new) \
4245 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4248 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4253 struct x86_exception *exception)
4255 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4261 /* guests cmpxchg8b have to be emulated atomically */
4262 if (bytes > 8 || (bytes & (bytes - 1)))
4265 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4267 if (gpa == UNMAPPED_GVA ||
4268 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4271 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4274 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4275 if (is_error_page(page))
4278 kaddr = kmap_atomic(page);
4279 kaddr += offset_in_page(gpa);
4282 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4285 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4288 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4291 exchanged = CMPXCHG64(kaddr, old, new);
4296 kunmap_atomic(kaddr);
4297 kvm_release_page_dirty(page);
4300 return X86EMUL_CMPXCHG_FAILED;
4302 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4304 return X86EMUL_CONTINUE;
4307 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4309 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4312 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4314 /* TODO: String I/O for in kernel device */
4317 if (vcpu->arch.pio.in)
4318 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4319 vcpu->arch.pio.size, pd);
4321 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4322 vcpu->arch.pio.port, vcpu->arch.pio.size,
4327 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4328 unsigned short port, void *val,
4329 unsigned int count, bool in)
4331 trace_kvm_pio(!in, port, size, count);
4333 vcpu->arch.pio.port = port;
4334 vcpu->arch.pio.in = in;
4335 vcpu->arch.pio.count = count;
4336 vcpu->arch.pio.size = size;
4338 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4339 vcpu->arch.pio.count = 0;
4343 vcpu->run->exit_reason = KVM_EXIT_IO;
4344 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4345 vcpu->run->io.size = size;
4346 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4347 vcpu->run->io.count = count;
4348 vcpu->run->io.port = port;
4353 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4354 int size, unsigned short port, void *val,
4357 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4360 if (vcpu->arch.pio.count)
4363 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4366 memcpy(val, vcpu->arch.pio_data, size * count);
4367 vcpu->arch.pio.count = 0;
4374 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4375 int size, unsigned short port,
4376 const void *val, unsigned int count)
4378 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4380 memcpy(vcpu->arch.pio_data, val, size * count);
4381 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4384 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4386 return kvm_x86_ops->get_segment_base(vcpu, seg);
4389 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4391 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4394 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4396 if (!need_emulate_wbinvd(vcpu))
4397 return X86EMUL_CONTINUE;
4399 if (kvm_x86_ops->has_wbinvd_exit()) {
4400 int cpu = get_cpu();
4402 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4403 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4404 wbinvd_ipi, NULL, 1);
4406 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4409 return X86EMUL_CONTINUE;
4411 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4413 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4415 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4418 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4420 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4423 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4426 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4429 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4431 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4434 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4436 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4437 unsigned long value;
4441 value = kvm_read_cr0(vcpu);
4444 value = vcpu->arch.cr2;
4447 value = kvm_read_cr3(vcpu);
4450 value = kvm_read_cr4(vcpu);
4453 value = kvm_get_cr8(vcpu);
4456 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4463 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4465 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4470 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4473 vcpu->arch.cr2 = val;
4476 res = kvm_set_cr3(vcpu, val);
4479 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4482 res = kvm_set_cr8(vcpu, val);
4485 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4492 static void emulator_set_rflags(struct x86_emulate_ctxt *ctxt, ulong val)
4494 kvm_set_rflags(emul_to_vcpu(ctxt), val);
4497 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4499 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4502 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4504 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4507 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4509 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4512 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4514 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4517 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4519 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4522 static unsigned long emulator_get_cached_segment_base(
4523 struct x86_emulate_ctxt *ctxt, int seg)
4525 return get_segment_base(emul_to_vcpu(ctxt), seg);
4528 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4529 struct desc_struct *desc, u32 *base3,
4532 struct kvm_segment var;
4534 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4535 *selector = var.selector;
4538 memset(desc, 0, sizeof(*desc));
4544 set_desc_limit(desc, var.limit);
4545 set_desc_base(desc, (unsigned long)var.base);
4546 #ifdef CONFIG_X86_64
4548 *base3 = var.base >> 32;
4550 desc->type = var.type;
4552 desc->dpl = var.dpl;
4553 desc->p = var.present;
4554 desc->avl = var.avl;
4562 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4563 struct desc_struct *desc, u32 base3,
4566 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4567 struct kvm_segment var;
4569 var.selector = selector;
4570 var.base = get_desc_base(desc);
4571 #ifdef CONFIG_X86_64
4572 var.base |= ((u64)base3) << 32;
4574 var.limit = get_desc_limit(desc);
4576 var.limit = (var.limit << 12) | 0xfff;
4577 var.type = desc->type;
4578 var.present = desc->p;
4579 var.dpl = desc->dpl;
4584 var.avl = desc->avl;
4585 var.present = desc->p;
4586 var.unusable = !var.present;
4589 kvm_set_segment(vcpu, &var, seg);
4593 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4594 u32 msr_index, u64 *pdata)
4596 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4599 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4600 u32 msr_index, u64 data)
4602 struct msr_data msr;
4605 msr.index = msr_index;
4606 msr.host_initiated = false;
4607 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4610 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4611 u32 pmc, u64 *pdata)
4613 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4616 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4618 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4621 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4624 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4626 * CR0.TS may reference the host fpu state, not the guest fpu state,
4627 * so it may be clear at this point.
4632 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4637 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4638 struct x86_instruction_info *info,
4639 enum x86_intercept_stage stage)
4641 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4644 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4645 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4647 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4650 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4652 return kvm_register_read(emul_to_vcpu(ctxt), reg);
4655 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4657 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4660 static const struct x86_emulate_ops emulate_ops = {
4661 .read_gpr = emulator_read_gpr,
4662 .write_gpr = emulator_write_gpr,
4663 .read_std = kvm_read_guest_virt_system,
4664 .write_std = kvm_write_guest_virt_system,
4665 .fetch = kvm_fetch_guest_virt,
4666 .read_emulated = emulator_read_emulated,
4667 .write_emulated = emulator_write_emulated,
4668 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4669 .invlpg = emulator_invlpg,
4670 .pio_in_emulated = emulator_pio_in_emulated,
4671 .pio_out_emulated = emulator_pio_out_emulated,
4672 .get_segment = emulator_get_segment,
4673 .set_segment = emulator_set_segment,
4674 .get_cached_segment_base = emulator_get_cached_segment_base,
4675 .get_gdt = emulator_get_gdt,
4676 .get_idt = emulator_get_idt,
4677 .set_gdt = emulator_set_gdt,
4678 .set_idt = emulator_set_idt,
4679 .get_cr = emulator_get_cr,
4680 .set_cr = emulator_set_cr,
4681 .set_rflags = emulator_set_rflags,
4682 .cpl = emulator_get_cpl,
4683 .get_dr = emulator_get_dr,
4684 .set_dr = emulator_set_dr,
4685 .set_msr = emulator_set_msr,
4686 .get_msr = emulator_get_msr,
4687 .read_pmc = emulator_read_pmc,
4688 .halt = emulator_halt,
4689 .wbinvd = emulator_wbinvd,
4690 .fix_hypercall = emulator_fix_hypercall,
4691 .get_fpu = emulator_get_fpu,
4692 .put_fpu = emulator_put_fpu,
4693 .intercept = emulator_intercept,
4694 .get_cpuid = emulator_get_cpuid,
4697 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4699 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4701 * an sti; sti; sequence only disable interrupts for the first
4702 * instruction. So, if the last instruction, be it emulated or
4703 * not, left the system with the INT_STI flag enabled, it
4704 * means that the last instruction is an sti. We should not
4705 * leave the flag on in this case. The same goes for mov ss
4707 if (!(int_shadow & mask))
4708 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4711 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4713 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4714 if (ctxt->exception.vector == PF_VECTOR)
4715 kvm_propagate_fault(vcpu, &ctxt->exception);
4716 else if (ctxt->exception.error_code_valid)
4717 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4718 ctxt->exception.error_code);
4720 kvm_queue_exception(vcpu, ctxt->exception.vector);
4723 static void init_decode_cache(struct x86_emulate_ctxt *ctxt)
4725 memset(&ctxt->twobyte, 0,
4726 (void *)&ctxt->_regs - (void *)&ctxt->twobyte);
4728 ctxt->fetch.start = 0;
4729 ctxt->fetch.end = 0;
4730 ctxt->io_read.pos = 0;
4731 ctxt->io_read.end = 0;
4732 ctxt->mem_read.pos = 0;
4733 ctxt->mem_read.end = 0;
4736 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4738 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4741 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4743 ctxt->eflags = kvm_get_rflags(vcpu);
4744 ctxt->eip = kvm_rip_read(vcpu);
4745 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4746 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4747 cs_l ? X86EMUL_MODE_PROT64 :
4748 cs_db ? X86EMUL_MODE_PROT32 :
4749 X86EMUL_MODE_PROT16;
4750 ctxt->guest_mode = is_guest_mode(vcpu);
4752 init_decode_cache(ctxt);
4753 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4756 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4758 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4761 init_emulate_ctxt(vcpu);
4765 ctxt->_eip = ctxt->eip + inc_eip;
4766 ret = emulate_int_real(ctxt, irq);
4768 if (ret != X86EMUL_CONTINUE)
4769 return EMULATE_FAIL;
4771 ctxt->eip = ctxt->_eip;
4772 kvm_rip_write(vcpu, ctxt->eip);
4773 kvm_set_rflags(vcpu, ctxt->eflags);
4775 if (irq == NMI_VECTOR)
4776 vcpu->arch.nmi_pending = 0;
4778 vcpu->arch.interrupt.pending = false;
4780 return EMULATE_DONE;
4782 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4784 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4786 int r = EMULATE_DONE;
4788 ++vcpu->stat.insn_emulation_fail;
4789 trace_kvm_emulate_insn_failed(vcpu);
4790 if (!is_guest_mode(vcpu)) {
4791 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4792 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4793 vcpu->run->internal.ndata = 0;
4796 kvm_queue_exception(vcpu, UD_VECTOR);
4801 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
4802 bool write_fault_to_shadow_pgtable,
4808 if (emulation_type & EMULTYPE_NO_REEXECUTE)
4811 if (!vcpu->arch.mmu.direct_map) {
4813 * Write permission should be allowed since only
4814 * write access need to be emulated.
4816 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4819 * If the mapping is invalid in guest, let cpu retry
4820 * it to generate fault.
4822 if (gpa == UNMAPPED_GVA)
4827 * Do not retry the unhandleable instruction if it faults on the
4828 * readonly host memory, otherwise it will goto a infinite loop:
4829 * retry instruction -> write #PF -> emulation fail -> retry
4830 * instruction -> ...
4832 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
4835 * If the instruction failed on the error pfn, it can not be fixed,
4836 * report the error to userspace.
4838 if (is_error_noslot_pfn(pfn))
4841 kvm_release_pfn_clean(pfn);
4843 /* The instructions are well-emulated on direct mmu. */
4844 if (vcpu->arch.mmu.direct_map) {
4845 unsigned int indirect_shadow_pages;
4847 spin_lock(&vcpu->kvm->mmu_lock);
4848 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
4849 spin_unlock(&vcpu->kvm->mmu_lock);
4851 if (indirect_shadow_pages)
4852 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4858 * if emulation was due to access to shadowed page table
4859 * and it failed try to unshadow page and re-enter the
4860 * guest to let CPU execute the instruction.
4862 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4865 * If the access faults on its page table, it can not
4866 * be fixed by unprotecting shadow page and it should
4867 * be reported to userspace.
4869 return !write_fault_to_shadow_pgtable;
4872 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
4873 unsigned long cr2, int emulation_type)
4875 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4876 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
4878 last_retry_eip = vcpu->arch.last_retry_eip;
4879 last_retry_addr = vcpu->arch.last_retry_addr;
4882 * If the emulation is caused by #PF and it is non-page_table
4883 * writing instruction, it means the VM-EXIT is caused by shadow
4884 * page protected, we can zap the shadow page and retry this
4885 * instruction directly.
4887 * Note: if the guest uses a non-page-table modifying instruction
4888 * on the PDE that points to the instruction, then we will unmap
4889 * the instruction and go to an infinite loop. So, we cache the
4890 * last retried eip and the last fault address, if we meet the eip
4891 * and the address again, we can break out of the potential infinite
4894 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
4896 if (!(emulation_type & EMULTYPE_RETRY))
4899 if (x86_page_table_writing_insn(ctxt))
4902 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
4905 vcpu->arch.last_retry_eip = ctxt->eip;
4906 vcpu->arch.last_retry_addr = cr2;
4908 if (!vcpu->arch.mmu.direct_map)
4909 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4911 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
4916 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
4917 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
4919 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4926 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4927 bool writeback = true;
4928 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
4931 * Clear write_fault_to_shadow_pgtable here to ensure it is
4934 vcpu->arch.write_fault_to_shadow_pgtable = false;
4935 kvm_clear_exception_queue(vcpu);
4937 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4938 init_emulate_ctxt(vcpu);
4939 ctxt->interruptibility = 0;
4940 ctxt->have_exception = false;
4941 ctxt->perm_ok = false;
4943 ctxt->only_vendor_specific_insn
4944 = emulation_type & EMULTYPE_TRAP_UD;
4946 r = x86_decode_insn(ctxt, insn, insn_len);
4948 trace_kvm_emulate_insn_start(vcpu);
4949 ++vcpu->stat.insn_emulation;
4950 if (r != EMULATION_OK) {
4951 if (emulation_type & EMULTYPE_TRAP_UD)
4952 return EMULATE_FAIL;
4953 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
4955 return EMULATE_DONE;
4956 if (emulation_type & EMULTYPE_SKIP)
4957 return EMULATE_FAIL;
4958 return handle_emulation_failure(vcpu);
4962 if (emulation_type & EMULTYPE_SKIP) {
4963 kvm_rip_write(vcpu, ctxt->_eip);
4964 return EMULATE_DONE;
4967 if (retry_instruction(ctxt, cr2, emulation_type))
4968 return EMULATE_DONE;
4970 /* this is needed for vmware backdoor interface to work since it
4971 changes registers values during IO operation */
4972 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4973 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4974 emulator_invalidate_register_cache(ctxt);
4978 r = x86_emulate_insn(ctxt);
4980 if (r == EMULATION_INTERCEPTED)
4981 return EMULATE_DONE;
4983 if (r == EMULATION_FAILED) {
4984 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
4986 return EMULATE_DONE;
4988 return handle_emulation_failure(vcpu);
4991 if (ctxt->have_exception) {
4992 inject_emulated_exception(vcpu);
4994 } else if (vcpu->arch.pio.count) {
4995 if (!vcpu->arch.pio.in)
4996 vcpu->arch.pio.count = 0;
4999 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5001 r = EMULATE_DO_MMIO;
5002 } else if (vcpu->mmio_needed) {
5003 if (!vcpu->mmio_is_write)
5005 r = EMULATE_DO_MMIO;
5006 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5007 } else if (r == EMULATION_RESTART)
5013 toggle_interruptibility(vcpu, ctxt->interruptibility);
5014 kvm_set_rflags(vcpu, ctxt->eflags);
5015 kvm_make_request(KVM_REQ_EVENT, vcpu);
5016 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5017 kvm_rip_write(vcpu, ctxt->eip);
5019 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5023 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5025 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5027 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5028 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5029 size, port, &val, 1);
5030 /* do not return to emulator after return from userspace */
5031 vcpu->arch.pio.count = 0;
5034 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5036 static void tsc_bad(void *info)
5038 __this_cpu_write(cpu_tsc_khz, 0);
5041 static void tsc_khz_changed(void *data)
5043 struct cpufreq_freqs *freq = data;
5044 unsigned long khz = 0;
5048 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5049 khz = cpufreq_quick_get(raw_smp_processor_id());
5052 __this_cpu_write(cpu_tsc_khz, khz);
5055 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5058 struct cpufreq_freqs *freq = data;
5060 struct kvm_vcpu *vcpu;
5061 int i, send_ipi = 0;
5064 * We allow guests to temporarily run on slowing clocks,
5065 * provided we notify them after, or to run on accelerating
5066 * clocks, provided we notify them before. Thus time never
5069 * However, we have a problem. We can't atomically update
5070 * the frequency of a given CPU from this function; it is
5071 * merely a notifier, which can be called from any CPU.
5072 * Changing the TSC frequency at arbitrary points in time
5073 * requires a recomputation of local variables related to
5074 * the TSC for each VCPU. We must flag these local variables
5075 * to be updated and be sure the update takes place with the
5076 * new frequency before any guests proceed.
5078 * Unfortunately, the combination of hotplug CPU and frequency
5079 * change creates an intractable locking scenario; the order
5080 * of when these callouts happen is undefined with respect to
5081 * CPU hotplug, and they can race with each other. As such,
5082 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5083 * undefined; you can actually have a CPU frequency change take
5084 * place in between the computation of X and the setting of the
5085 * variable. To protect against this problem, all updates of
5086 * the per_cpu tsc_khz variable are done in an interrupt
5087 * protected IPI, and all callers wishing to update the value
5088 * must wait for a synchronous IPI to complete (which is trivial
5089 * if the caller is on the CPU already). This establishes the
5090 * necessary total order on variable updates.
5092 * Note that because a guest time update may take place
5093 * anytime after the setting of the VCPU's request bit, the
5094 * correct TSC value must be set before the request. However,
5095 * to ensure the update actually makes it to any guest which
5096 * starts running in hardware virtualization between the set
5097 * and the acquisition of the spinlock, we must also ping the
5098 * CPU after setting the request bit.
5102 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5104 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5107 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5109 raw_spin_lock(&kvm_lock);
5110 list_for_each_entry(kvm, &vm_list, vm_list) {
5111 kvm_for_each_vcpu(i, vcpu, kvm) {
5112 if (vcpu->cpu != freq->cpu)
5114 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5115 if (vcpu->cpu != smp_processor_id())
5119 raw_spin_unlock(&kvm_lock);
5121 if (freq->old < freq->new && send_ipi) {
5123 * We upscale the frequency. Must make the guest
5124 * doesn't see old kvmclock values while running with
5125 * the new frequency, otherwise we risk the guest sees
5126 * time go backwards.
5128 * In case we update the frequency for another cpu
5129 * (which might be in guest context) send an interrupt
5130 * to kick the cpu out of guest context. Next time
5131 * guest context is entered kvmclock will be updated,
5132 * so the guest will not see stale values.
5134 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5139 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5140 .notifier_call = kvmclock_cpufreq_notifier
5143 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5144 unsigned long action, void *hcpu)
5146 unsigned int cpu = (unsigned long)hcpu;
5150 case CPU_DOWN_FAILED:
5151 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5153 case CPU_DOWN_PREPARE:
5154 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5160 static struct notifier_block kvmclock_cpu_notifier_block = {
5161 .notifier_call = kvmclock_cpu_notifier,
5162 .priority = -INT_MAX
5165 static void kvm_timer_init(void)
5169 max_tsc_khz = tsc_khz;
5170 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5171 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5172 #ifdef CONFIG_CPU_FREQ
5173 struct cpufreq_policy policy;
5174 memset(&policy, 0, sizeof(policy));
5176 cpufreq_get_policy(&policy, cpu);
5177 if (policy.cpuinfo.max_freq)
5178 max_tsc_khz = policy.cpuinfo.max_freq;
5181 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5182 CPUFREQ_TRANSITION_NOTIFIER);
5184 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5185 for_each_online_cpu(cpu)
5186 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5189 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5191 int kvm_is_in_guest(void)
5193 return __this_cpu_read(current_vcpu) != NULL;
5196 static int kvm_is_user_mode(void)
5200 if (__this_cpu_read(current_vcpu))
5201 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5203 return user_mode != 0;
5206 static unsigned long kvm_get_guest_ip(void)
5208 unsigned long ip = 0;
5210 if (__this_cpu_read(current_vcpu))
5211 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5216 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5217 .is_in_guest = kvm_is_in_guest,
5218 .is_user_mode = kvm_is_user_mode,
5219 .get_guest_ip = kvm_get_guest_ip,
5222 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5224 __this_cpu_write(current_vcpu, vcpu);
5226 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5228 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5230 __this_cpu_write(current_vcpu, NULL);
5232 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5234 static void kvm_set_mmio_spte_mask(void)
5237 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5240 * Set the reserved bits and the present bit of an paging-structure
5241 * entry to generate page fault with PFER.RSV = 1.
5243 mask = ((1ull << (62 - maxphyaddr + 1)) - 1) << maxphyaddr;
5246 #ifdef CONFIG_X86_64
5248 * If reserved bit is not supported, clear the present bit to disable
5251 if (maxphyaddr == 52)
5255 kvm_mmu_set_mmio_spte_mask(mask);
5258 #ifdef CONFIG_X86_64
5259 static void pvclock_gtod_update_fn(struct work_struct *work)
5263 struct kvm_vcpu *vcpu;
5266 raw_spin_lock(&kvm_lock);
5267 list_for_each_entry(kvm, &vm_list, vm_list)
5268 kvm_for_each_vcpu(i, vcpu, kvm)
5269 set_bit(KVM_REQ_MASTERCLOCK_UPDATE, &vcpu->requests);
5270 atomic_set(&kvm_guest_has_master_clock, 0);
5271 raw_spin_unlock(&kvm_lock);
5274 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5277 * Notification about pvclock gtod data update.
5279 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5282 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5283 struct timekeeper *tk = priv;
5285 update_pvclock_gtod(tk);
5287 /* disable master clock if host does not trust, or does not
5288 * use, TSC clocksource
5290 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5291 atomic_read(&kvm_guest_has_master_clock) != 0)
5292 queue_work(system_long_wq, &pvclock_gtod_work);
5297 static struct notifier_block pvclock_gtod_notifier = {
5298 .notifier_call = pvclock_gtod_notify,
5302 int kvm_arch_init(void *opaque)
5305 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
5308 printk(KERN_ERR "kvm: already loaded the other module\n");
5313 if (!ops->cpu_has_kvm_support()) {
5314 printk(KERN_ERR "kvm: no hardware support\n");
5318 if (ops->disabled_by_bios()) {
5319 printk(KERN_ERR "kvm: disabled by bios\n");
5325 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5327 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5331 r = kvm_mmu_module_init();
5333 goto out_free_percpu;
5335 kvm_set_mmio_spte_mask();
5336 kvm_init_msr_list();
5339 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5340 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5344 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5347 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5350 #ifdef CONFIG_X86_64
5351 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5357 free_percpu(shared_msrs);
5362 void kvm_arch_exit(void)
5364 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5366 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5367 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5368 CPUFREQ_TRANSITION_NOTIFIER);
5369 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5370 #ifdef CONFIG_X86_64
5371 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5374 kvm_mmu_module_exit();
5375 free_percpu(shared_msrs);
5378 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5380 ++vcpu->stat.halt_exits;
5381 if (irqchip_in_kernel(vcpu->kvm)) {
5382 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5385 vcpu->run->exit_reason = KVM_EXIT_HLT;
5389 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5391 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5393 u64 param, ingpa, outgpa, ret;
5394 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5395 bool fast, longmode;
5399 * hypercall generates UD from non zero cpl and real mode
5402 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5403 kvm_queue_exception(vcpu, UD_VECTOR);
5407 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5408 longmode = is_long_mode(vcpu) && cs_l == 1;
5411 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5412 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5413 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5414 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5415 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5416 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5418 #ifdef CONFIG_X86_64
5420 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5421 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5422 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5426 code = param & 0xffff;
5427 fast = (param >> 16) & 0x1;
5428 rep_cnt = (param >> 32) & 0xfff;
5429 rep_idx = (param >> 48) & 0xfff;
5431 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5434 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5435 kvm_vcpu_on_spin(vcpu);
5438 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5442 ret = res | (((u64)rep_done & 0xfff) << 32);
5444 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5446 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5447 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5453 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5455 unsigned long nr, a0, a1, a2, a3, ret;
5458 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5459 return kvm_hv_hypercall(vcpu);
5461 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5462 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5463 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5464 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5465 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5467 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5469 if (!is_long_mode(vcpu)) {
5477 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5483 case KVM_HC_VAPIC_POLL_IRQ:
5491 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5492 ++vcpu->stat.hypercalls;
5495 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5497 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5499 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5500 char instruction[3];
5501 unsigned long rip = kvm_rip_read(vcpu);
5504 * Blow out the MMU to ensure that no other VCPU has an active mapping
5505 * to ensure that the updated hypercall appears atomically across all
5508 kvm_mmu_zap_all(vcpu->kvm);
5510 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5512 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5516 * Check if userspace requested an interrupt window, and that the
5517 * interrupt window is open.
5519 * No need to exit to userspace if we already have an interrupt queued.
5521 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5523 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5524 vcpu->run->request_interrupt_window &&
5525 kvm_arch_interrupt_allowed(vcpu));
5528 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5530 struct kvm_run *kvm_run = vcpu->run;
5532 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5533 kvm_run->cr8 = kvm_get_cr8(vcpu);
5534 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5535 if (irqchip_in_kernel(vcpu->kvm))
5536 kvm_run->ready_for_interrupt_injection = 1;
5538 kvm_run->ready_for_interrupt_injection =
5539 kvm_arch_interrupt_allowed(vcpu) &&
5540 !kvm_cpu_has_interrupt(vcpu) &&
5541 !kvm_event_needs_reinjection(vcpu);
5544 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5548 if (!kvm_x86_ops->update_cr8_intercept)
5551 if (!vcpu->arch.apic)
5554 if (!vcpu->arch.apic->vapic_addr)
5555 max_irr = kvm_lapic_find_highest_irr(vcpu);
5562 tpr = kvm_lapic_get_cr8(vcpu);
5564 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5567 static void inject_pending_event(struct kvm_vcpu *vcpu)
5569 /* try to reinject previous events if any */
5570 if (vcpu->arch.exception.pending) {
5571 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5572 vcpu->arch.exception.has_error_code,
5573 vcpu->arch.exception.error_code);
5574 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5575 vcpu->arch.exception.has_error_code,
5576 vcpu->arch.exception.error_code,
5577 vcpu->arch.exception.reinject);
5581 if (vcpu->arch.nmi_injected) {
5582 kvm_x86_ops->set_nmi(vcpu);
5586 if (vcpu->arch.interrupt.pending) {
5587 kvm_x86_ops->set_irq(vcpu);
5591 /* try to inject new event if pending */
5592 if (vcpu->arch.nmi_pending) {
5593 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5594 --vcpu->arch.nmi_pending;
5595 vcpu->arch.nmi_injected = true;
5596 kvm_x86_ops->set_nmi(vcpu);
5598 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
5599 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5600 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5602 kvm_x86_ops->set_irq(vcpu);
5607 static void process_nmi(struct kvm_vcpu *vcpu)
5612 * x86 is limited to one NMI running, and one NMI pending after it.
5613 * If an NMI is already in progress, limit further NMIs to just one.
5614 * Otherwise, allow two (and we'll inject the first one immediately).
5616 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5619 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5620 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5621 kvm_make_request(KVM_REQ_EVENT, vcpu);
5624 static void kvm_gen_update_masterclock(struct kvm *kvm)
5626 #ifdef CONFIG_X86_64
5628 struct kvm_vcpu *vcpu;
5629 struct kvm_arch *ka = &kvm->arch;
5631 spin_lock(&ka->pvclock_gtod_sync_lock);
5632 kvm_make_mclock_inprogress_request(kvm);
5633 /* no guest entries from this point */
5634 pvclock_update_vm_gtod_copy(kvm);
5636 kvm_for_each_vcpu(i, vcpu, kvm)
5637 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
5639 /* guest entries allowed */
5640 kvm_for_each_vcpu(i, vcpu, kvm)
5641 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
5643 spin_unlock(&ka->pvclock_gtod_sync_lock);
5647 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
5649 u64 eoi_exit_bitmap[4];
5652 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
5655 memset(eoi_exit_bitmap, 0, 32);
5658 kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
5659 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
5660 kvm_apic_update_tmr(vcpu, tmr);
5663 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5666 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5667 vcpu->run->request_interrupt_window;
5668 bool req_immediate_exit = false;
5670 if (vcpu->requests) {
5671 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5672 kvm_mmu_unload(vcpu);
5673 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5674 __kvm_migrate_timers(vcpu);
5675 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
5676 kvm_gen_update_masterclock(vcpu->kvm);
5677 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5678 r = kvm_guest_time_update(vcpu);
5682 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5683 kvm_mmu_sync_roots(vcpu);
5684 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5685 kvm_x86_ops->tlb_flush(vcpu);
5686 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5687 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5691 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5692 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5696 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5697 vcpu->fpu_active = 0;
5698 kvm_x86_ops->fpu_deactivate(vcpu);
5700 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5701 /* Page is swapped out. Do synthetic halt */
5702 vcpu->arch.apf.halted = true;
5706 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5707 record_steal_time(vcpu);
5708 if (kvm_check_request(KVM_REQ_NMI, vcpu))
5710 if (kvm_check_request(KVM_REQ_PMU, vcpu))
5711 kvm_handle_pmu_event(vcpu);
5712 if (kvm_check_request(KVM_REQ_PMI, vcpu))
5713 kvm_deliver_pmi(vcpu);
5714 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
5715 vcpu_scan_ioapic(vcpu);
5718 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5719 kvm_apic_accept_events(vcpu);
5720 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
5725 inject_pending_event(vcpu);
5727 /* enable NMI/IRQ window open exits if needed */
5728 if (vcpu->arch.nmi_pending)
5729 req_immediate_exit =
5730 kvm_x86_ops->enable_nmi_window(vcpu) != 0;
5731 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
5732 req_immediate_exit =
5733 kvm_x86_ops->enable_irq_window(vcpu) != 0;
5735 if (kvm_lapic_enabled(vcpu)) {
5737 * Update architecture specific hints for APIC
5738 * virtual interrupt delivery.
5740 if (kvm_x86_ops->hwapic_irr_update)
5741 kvm_x86_ops->hwapic_irr_update(vcpu,
5742 kvm_lapic_find_highest_irr(vcpu));
5743 update_cr8_intercept(vcpu);
5744 kvm_lapic_sync_to_vapic(vcpu);
5748 r = kvm_mmu_reload(vcpu);
5750 goto cancel_injection;
5755 kvm_x86_ops->prepare_guest_switch(vcpu);
5756 if (vcpu->fpu_active)
5757 kvm_load_guest_fpu(vcpu);
5758 kvm_load_guest_xcr0(vcpu);
5760 vcpu->mode = IN_GUEST_MODE;
5762 /* We should set ->mode before check ->requests,
5763 * see the comment in make_all_cpus_request.
5767 local_irq_disable();
5769 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5770 || need_resched() || signal_pending(current)) {
5771 vcpu->mode = OUTSIDE_GUEST_MODE;
5776 goto cancel_injection;
5779 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5781 if (req_immediate_exit)
5782 smp_send_reschedule(vcpu->cpu);
5786 if (unlikely(vcpu->arch.switch_db_regs)) {
5788 set_debugreg(vcpu->arch.eff_db[0], 0);
5789 set_debugreg(vcpu->arch.eff_db[1], 1);
5790 set_debugreg(vcpu->arch.eff_db[2], 2);
5791 set_debugreg(vcpu->arch.eff_db[3], 3);
5794 trace_kvm_entry(vcpu->vcpu_id);
5795 kvm_x86_ops->run(vcpu);
5798 * If the guest has used debug registers, at least dr7
5799 * will be disabled while returning to the host.
5800 * If we don't have active breakpoints in the host, we don't
5801 * care about the messed up debug address registers. But if
5802 * we have some of them active, restore the old state.
5804 if (hw_breakpoint_active())
5805 hw_breakpoint_restore();
5807 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
5810 vcpu->mode = OUTSIDE_GUEST_MODE;
5813 /* Interrupt is enabled by handle_external_intr() */
5814 kvm_x86_ops->handle_external_intr(vcpu);
5819 * We must have an instruction between local_irq_enable() and
5820 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5821 * the interrupt shadow. The stat.exits increment will do nicely.
5822 * But we need to prevent reordering, hence this barrier():
5830 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5833 * Profile KVM exit RIPs:
5835 if (unlikely(prof_on == KVM_PROFILING)) {
5836 unsigned long rip = kvm_rip_read(vcpu);
5837 profile_hit(KVM_PROFILING, (void *)rip);
5840 if (unlikely(vcpu->arch.tsc_always_catchup))
5841 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5843 if (vcpu->arch.apic_attention)
5844 kvm_lapic_sync_from_vapic(vcpu);
5846 r = kvm_x86_ops->handle_exit(vcpu);
5850 kvm_x86_ops->cancel_injection(vcpu);
5851 if (unlikely(vcpu->arch.apic_attention))
5852 kvm_lapic_sync_from_vapic(vcpu);
5858 static int __vcpu_run(struct kvm_vcpu *vcpu)
5861 struct kvm *kvm = vcpu->kvm;
5863 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5867 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5868 !vcpu->arch.apf.halted)
5869 r = vcpu_enter_guest(vcpu);
5871 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5872 kvm_vcpu_block(vcpu);
5873 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5874 if (kvm_check_request(KVM_REQ_UNHALT, vcpu)) {
5875 kvm_apic_accept_events(vcpu);
5876 switch(vcpu->arch.mp_state) {
5877 case KVM_MP_STATE_HALTED:
5878 vcpu->arch.mp_state =
5879 KVM_MP_STATE_RUNNABLE;
5880 case KVM_MP_STATE_RUNNABLE:
5881 vcpu->arch.apf.halted = false;
5883 case KVM_MP_STATE_INIT_RECEIVED:
5895 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5896 if (kvm_cpu_has_pending_timer(vcpu))
5897 kvm_inject_pending_timer_irqs(vcpu);
5899 if (dm_request_for_irq_injection(vcpu)) {
5901 vcpu->run->exit_reason = KVM_EXIT_INTR;
5902 ++vcpu->stat.request_irq_exits;
5905 kvm_check_async_pf_completion(vcpu);
5907 if (signal_pending(current)) {
5909 vcpu->run->exit_reason = KVM_EXIT_INTR;
5910 ++vcpu->stat.signal_exits;
5912 if (need_resched()) {
5913 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5915 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5919 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5924 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
5927 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5928 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5929 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5930 if (r != EMULATE_DONE)
5935 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
5937 BUG_ON(!vcpu->arch.pio.count);
5939 return complete_emulated_io(vcpu);
5943 * Implements the following, as a state machine:
5947 * for each mmio piece in the fragment
5955 * for each mmio piece in the fragment
5960 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
5962 struct kvm_run *run = vcpu->run;
5963 struct kvm_mmio_fragment *frag;
5966 BUG_ON(!vcpu->mmio_needed);
5968 /* Complete previous fragment */
5969 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
5970 len = min(8u, frag->len);
5971 if (!vcpu->mmio_is_write)
5972 memcpy(frag->data, run->mmio.data, len);
5974 if (frag->len <= 8) {
5975 /* Switch to the next fragment. */
5977 vcpu->mmio_cur_fragment++;
5979 /* Go forward to the next mmio piece. */
5985 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
5986 vcpu->mmio_needed = 0;
5987 if (vcpu->mmio_is_write)
5989 vcpu->mmio_read_completed = 1;
5990 return complete_emulated_io(vcpu);
5993 run->exit_reason = KVM_EXIT_MMIO;
5994 run->mmio.phys_addr = frag->gpa;
5995 if (vcpu->mmio_is_write)
5996 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
5997 run->mmio.len = min(8u, frag->len);
5998 run->mmio.is_write = vcpu->mmio_is_write;
5999 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6004 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6009 if (!tsk_used_math(current) && init_fpu(current))
6012 if (vcpu->sigset_active)
6013 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6015 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6016 kvm_vcpu_block(vcpu);
6017 kvm_apic_accept_events(vcpu);
6018 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6023 /* re-sync apic's tpr */
6024 if (!irqchip_in_kernel(vcpu->kvm)) {
6025 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6031 if (unlikely(vcpu->arch.complete_userspace_io)) {
6032 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6033 vcpu->arch.complete_userspace_io = NULL;
6038 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6040 r = __vcpu_run(vcpu);
6043 post_kvm_run_save(vcpu);
6044 if (vcpu->sigset_active)
6045 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6050 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6052 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6054 * We are here if userspace calls get_regs() in the middle of
6055 * instruction emulation. Registers state needs to be copied
6056 * back from emulation context to vcpu. Userspace shouldn't do
6057 * that usually, but some bad designed PV devices (vmware
6058 * backdoor interface) need this to work
6060 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6061 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6063 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6064 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6065 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6066 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6067 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6068 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6069 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6070 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6071 #ifdef CONFIG_X86_64
6072 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6073 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6074 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6075 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6076 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6077 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6078 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6079 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6082 regs->rip = kvm_rip_read(vcpu);
6083 regs->rflags = kvm_get_rflags(vcpu);
6088 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6090 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6091 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6093 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6094 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6095 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6096 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6097 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6098 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6099 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6100 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6101 #ifdef CONFIG_X86_64
6102 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6103 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6104 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6105 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6106 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6107 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6108 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6109 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6112 kvm_rip_write(vcpu, regs->rip);
6113 kvm_set_rflags(vcpu, regs->rflags);
6115 vcpu->arch.exception.pending = false;
6117 kvm_make_request(KVM_REQ_EVENT, vcpu);
6122 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6124 struct kvm_segment cs;
6126 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6130 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6132 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6133 struct kvm_sregs *sregs)
6137 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6138 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6139 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6140 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6141 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6142 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6144 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6145 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6147 kvm_x86_ops->get_idt(vcpu, &dt);
6148 sregs->idt.limit = dt.size;
6149 sregs->idt.base = dt.address;
6150 kvm_x86_ops->get_gdt(vcpu, &dt);
6151 sregs->gdt.limit = dt.size;
6152 sregs->gdt.base = dt.address;
6154 sregs->cr0 = kvm_read_cr0(vcpu);
6155 sregs->cr2 = vcpu->arch.cr2;
6156 sregs->cr3 = kvm_read_cr3(vcpu);
6157 sregs->cr4 = kvm_read_cr4(vcpu);
6158 sregs->cr8 = kvm_get_cr8(vcpu);
6159 sregs->efer = vcpu->arch.efer;
6160 sregs->apic_base = kvm_get_apic_base(vcpu);
6162 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6164 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6165 set_bit(vcpu->arch.interrupt.nr,
6166 (unsigned long *)sregs->interrupt_bitmap);
6171 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6172 struct kvm_mp_state *mp_state)
6174 kvm_apic_accept_events(vcpu);
6175 mp_state->mp_state = vcpu->arch.mp_state;
6179 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6180 struct kvm_mp_state *mp_state)
6182 if (!kvm_vcpu_has_lapic(vcpu) &&
6183 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6186 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6187 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6188 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6190 vcpu->arch.mp_state = mp_state->mp_state;
6191 kvm_make_request(KVM_REQ_EVENT, vcpu);
6195 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6196 int reason, bool has_error_code, u32 error_code)
6198 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6201 init_emulate_ctxt(vcpu);
6203 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6204 has_error_code, error_code);
6207 return EMULATE_FAIL;
6209 kvm_rip_write(vcpu, ctxt->eip);
6210 kvm_set_rflags(vcpu, ctxt->eflags);
6211 kvm_make_request(KVM_REQ_EVENT, vcpu);
6212 return EMULATE_DONE;
6214 EXPORT_SYMBOL_GPL(kvm_task_switch);
6216 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6217 struct kvm_sregs *sregs)
6219 int mmu_reset_needed = 0;
6220 int pending_vec, max_bits, idx;
6223 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6226 dt.size = sregs->idt.limit;
6227 dt.address = sregs->idt.base;
6228 kvm_x86_ops->set_idt(vcpu, &dt);
6229 dt.size = sregs->gdt.limit;
6230 dt.address = sregs->gdt.base;
6231 kvm_x86_ops->set_gdt(vcpu, &dt);
6233 vcpu->arch.cr2 = sregs->cr2;
6234 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6235 vcpu->arch.cr3 = sregs->cr3;
6236 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6238 kvm_set_cr8(vcpu, sregs->cr8);
6240 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6241 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6242 kvm_set_apic_base(vcpu, sregs->apic_base);
6244 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6245 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6246 vcpu->arch.cr0 = sregs->cr0;
6248 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6249 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6250 if (sregs->cr4 & X86_CR4_OSXSAVE)
6251 kvm_update_cpuid(vcpu);
6253 idx = srcu_read_lock(&vcpu->kvm->srcu);
6254 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6255 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6256 mmu_reset_needed = 1;
6258 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6260 if (mmu_reset_needed)
6261 kvm_mmu_reset_context(vcpu);
6263 max_bits = KVM_NR_INTERRUPTS;
6264 pending_vec = find_first_bit(
6265 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6266 if (pending_vec < max_bits) {
6267 kvm_queue_interrupt(vcpu, pending_vec, false);
6268 pr_debug("Set back pending irq %d\n", pending_vec);
6271 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6272 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6273 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6274 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6275 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6276 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6278 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6279 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6281 update_cr8_intercept(vcpu);
6283 /* Older userspace won't unhalt the vcpu on reset. */
6284 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6285 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6287 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6289 kvm_make_request(KVM_REQ_EVENT, vcpu);
6294 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6295 struct kvm_guest_debug *dbg)
6297 unsigned long rflags;
6300 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6302 if (vcpu->arch.exception.pending)
6304 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6305 kvm_queue_exception(vcpu, DB_VECTOR);
6307 kvm_queue_exception(vcpu, BP_VECTOR);
6311 * Read rflags as long as potentially injected trace flags are still
6314 rflags = kvm_get_rflags(vcpu);
6316 vcpu->guest_debug = dbg->control;
6317 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6318 vcpu->guest_debug = 0;
6320 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6321 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6322 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6323 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6325 for (i = 0; i < KVM_NR_DB_REGS; i++)
6326 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6328 kvm_update_dr7(vcpu);
6330 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6331 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6332 get_segment_base(vcpu, VCPU_SREG_CS);
6335 * Trigger an rflags update that will inject or remove the trace
6338 kvm_set_rflags(vcpu, rflags);
6340 kvm_x86_ops->update_db_bp_intercept(vcpu);
6350 * Translate a guest virtual address to a guest physical address.
6352 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6353 struct kvm_translation *tr)
6355 unsigned long vaddr = tr->linear_address;
6359 idx = srcu_read_lock(&vcpu->kvm->srcu);
6360 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6361 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6362 tr->physical_address = gpa;
6363 tr->valid = gpa != UNMAPPED_GVA;
6370 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6372 struct i387_fxsave_struct *fxsave =
6373 &vcpu->arch.guest_fpu.state->fxsave;
6375 memcpy(fpu->fpr, fxsave->st_space, 128);
6376 fpu->fcw = fxsave->cwd;
6377 fpu->fsw = fxsave->swd;
6378 fpu->ftwx = fxsave->twd;
6379 fpu->last_opcode = fxsave->fop;
6380 fpu->last_ip = fxsave->rip;
6381 fpu->last_dp = fxsave->rdp;
6382 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6387 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6389 struct i387_fxsave_struct *fxsave =
6390 &vcpu->arch.guest_fpu.state->fxsave;
6392 memcpy(fxsave->st_space, fpu->fpr, 128);
6393 fxsave->cwd = fpu->fcw;
6394 fxsave->swd = fpu->fsw;
6395 fxsave->twd = fpu->ftwx;
6396 fxsave->fop = fpu->last_opcode;
6397 fxsave->rip = fpu->last_ip;
6398 fxsave->rdp = fpu->last_dp;
6399 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6404 int fx_init(struct kvm_vcpu *vcpu)
6408 err = fpu_alloc(&vcpu->arch.guest_fpu);
6412 fpu_finit(&vcpu->arch.guest_fpu);
6415 * Ensure guest xcr0 is valid for loading
6417 vcpu->arch.xcr0 = XSTATE_FP;
6419 vcpu->arch.cr0 |= X86_CR0_ET;
6423 EXPORT_SYMBOL_GPL(fx_init);
6425 static void fx_free(struct kvm_vcpu *vcpu)
6427 fpu_free(&vcpu->arch.guest_fpu);
6430 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6432 if (vcpu->guest_fpu_loaded)
6436 * Restore all possible states in the guest,
6437 * and assume host would use all available bits.
6438 * Guest xcr0 would be loaded later.
6440 kvm_put_guest_xcr0(vcpu);
6441 vcpu->guest_fpu_loaded = 1;
6442 __kernel_fpu_begin();
6443 fpu_restore_checking(&vcpu->arch.guest_fpu);
6447 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6449 kvm_put_guest_xcr0(vcpu);
6451 if (!vcpu->guest_fpu_loaded)
6454 vcpu->guest_fpu_loaded = 0;
6455 fpu_save_init(&vcpu->arch.guest_fpu);
6457 ++vcpu->stat.fpu_reload;
6458 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6462 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6464 kvmclock_reset(vcpu);
6466 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6468 kvm_x86_ops->vcpu_free(vcpu);
6471 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6474 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6475 printk_once(KERN_WARNING
6476 "kvm: SMP vm created on host with unstable TSC; "
6477 "guest TSC will not be reliable\n");
6478 return kvm_x86_ops->vcpu_create(kvm, id);
6481 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6485 vcpu->arch.mtrr_state.have_fixed = 1;
6486 r = vcpu_load(vcpu);
6489 kvm_vcpu_reset(vcpu);
6490 r = kvm_mmu_setup(vcpu);
6496 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
6499 struct msr_data msr;
6501 r = vcpu_load(vcpu);
6505 msr.index = MSR_IA32_TSC;
6506 msr.host_initiated = true;
6507 kvm_write_tsc(vcpu, &msr);
6513 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6516 vcpu->arch.apf.msr_val = 0;
6518 r = vcpu_load(vcpu);
6520 kvm_mmu_unload(vcpu);
6524 kvm_x86_ops->vcpu_free(vcpu);
6527 void kvm_vcpu_reset(struct kvm_vcpu *vcpu)
6529 atomic_set(&vcpu->arch.nmi_queued, 0);
6530 vcpu->arch.nmi_pending = 0;
6531 vcpu->arch.nmi_injected = false;
6533 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6534 vcpu->arch.dr6 = DR6_FIXED_1;
6535 vcpu->arch.dr7 = DR7_FIXED_1;
6536 kvm_update_dr7(vcpu);
6538 kvm_make_request(KVM_REQ_EVENT, vcpu);
6539 vcpu->arch.apf.msr_val = 0;
6540 vcpu->arch.st.msr_val = 0;
6542 kvmclock_reset(vcpu);
6544 kvm_clear_async_pf_completion_queue(vcpu);
6545 kvm_async_pf_hash_reset(vcpu);
6546 vcpu->arch.apf.halted = false;
6548 kvm_pmu_reset(vcpu);
6550 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
6551 vcpu->arch.regs_avail = ~0;
6552 vcpu->arch.regs_dirty = ~0;
6554 kvm_x86_ops->vcpu_reset(vcpu);
6557 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, unsigned int vector)
6559 struct kvm_segment cs;
6561 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6562 cs.selector = vector << 8;
6563 cs.base = vector << 12;
6564 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6565 kvm_rip_write(vcpu, 0);
6568 int kvm_arch_hardware_enable(void *garbage)
6571 struct kvm_vcpu *vcpu;
6576 bool stable, backwards_tsc = false;
6578 kvm_shared_msr_cpu_online();
6579 ret = kvm_x86_ops->hardware_enable(garbage);
6583 local_tsc = native_read_tsc();
6584 stable = !check_tsc_unstable();
6585 list_for_each_entry(kvm, &vm_list, vm_list) {
6586 kvm_for_each_vcpu(i, vcpu, kvm) {
6587 if (!stable && vcpu->cpu == smp_processor_id())
6588 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
6589 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
6590 backwards_tsc = true;
6591 if (vcpu->arch.last_host_tsc > max_tsc)
6592 max_tsc = vcpu->arch.last_host_tsc;
6598 * Sometimes, even reliable TSCs go backwards. This happens on
6599 * platforms that reset TSC during suspend or hibernate actions, but
6600 * maintain synchronization. We must compensate. Fortunately, we can
6601 * detect that condition here, which happens early in CPU bringup,
6602 * before any KVM threads can be running. Unfortunately, we can't
6603 * bring the TSCs fully up to date with real time, as we aren't yet far
6604 * enough into CPU bringup that we know how much real time has actually
6605 * elapsed; our helper function, get_kernel_ns() will be using boot
6606 * variables that haven't been updated yet.
6608 * So we simply find the maximum observed TSC above, then record the
6609 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
6610 * the adjustment will be applied. Note that we accumulate
6611 * adjustments, in case multiple suspend cycles happen before some VCPU
6612 * gets a chance to run again. In the event that no KVM threads get a
6613 * chance to run, we will miss the entire elapsed period, as we'll have
6614 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
6615 * loose cycle time. This isn't too big a deal, since the loss will be
6616 * uniform across all VCPUs (not to mention the scenario is extremely
6617 * unlikely). It is possible that a second hibernate recovery happens
6618 * much faster than a first, causing the observed TSC here to be
6619 * smaller; this would require additional padding adjustment, which is
6620 * why we set last_host_tsc to the local tsc observed here.
6622 * N.B. - this code below runs only on platforms with reliable TSC,
6623 * as that is the only way backwards_tsc is set above. Also note
6624 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
6625 * have the same delta_cyc adjustment applied if backwards_tsc
6626 * is detected. Note further, this adjustment is only done once,
6627 * as we reset last_host_tsc on all VCPUs to stop this from being
6628 * called multiple times (one for each physical CPU bringup).
6630 * Platforms with unreliable TSCs don't have to deal with this, they
6631 * will be compensated by the logic in vcpu_load, which sets the TSC to
6632 * catchup mode. This will catchup all VCPUs to real time, but cannot
6633 * guarantee that they stay in perfect synchronization.
6635 if (backwards_tsc) {
6636 u64 delta_cyc = max_tsc - local_tsc;
6637 list_for_each_entry(kvm, &vm_list, vm_list) {
6638 kvm_for_each_vcpu(i, vcpu, kvm) {
6639 vcpu->arch.tsc_offset_adjustment += delta_cyc;
6640 vcpu->arch.last_host_tsc = local_tsc;
6641 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
6646 * We have to disable TSC offset matching.. if you were
6647 * booting a VM while issuing an S4 host suspend....
6648 * you may have some problem. Solving this issue is
6649 * left as an exercise to the reader.
6651 kvm->arch.last_tsc_nsec = 0;
6652 kvm->arch.last_tsc_write = 0;
6659 void kvm_arch_hardware_disable(void *garbage)
6661 kvm_x86_ops->hardware_disable(garbage);
6662 drop_user_return_notifiers(garbage);
6665 int kvm_arch_hardware_setup(void)
6667 return kvm_x86_ops->hardware_setup();
6670 void kvm_arch_hardware_unsetup(void)
6672 kvm_x86_ops->hardware_unsetup();
6675 void kvm_arch_check_processor_compat(void *rtn)
6677 kvm_x86_ops->check_processor_compatibility(rtn);
6680 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
6682 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
6685 struct static_key kvm_no_apic_vcpu __read_mostly;
6687 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6693 BUG_ON(vcpu->kvm == NULL);
6696 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6697 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6698 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6700 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6702 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6707 vcpu->arch.pio_data = page_address(page);
6709 kvm_set_tsc_khz(vcpu, max_tsc_khz);
6711 r = kvm_mmu_create(vcpu);
6713 goto fail_free_pio_data;
6715 if (irqchip_in_kernel(kvm)) {
6716 r = kvm_create_lapic(vcpu);
6718 goto fail_mmu_destroy;
6720 static_key_slow_inc(&kvm_no_apic_vcpu);
6722 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6724 if (!vcpu->arch.mce_banks) {
6726 goto fail_free_lapic;
6728 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6730 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
6732 goto fail_free_mce_banks;
6737 goto fail_free_wbinvd_dirty_mask;
6739 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
6740 vcpu->arch.pv_time_enabled = false;
6741 kvm_async_pf_hash_reset(vcpu);
6745 fail_free_wbinvd_dirty_mask:
6746 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6747 fail_free_mce_banks:
6748 kfree(vcpu->arch.mce_banks);
6750 kvm_free_lapic(vcpu);
6752 kvm_mmu_destroy(vcpu);
6754 free_page((unsigned long)vcpu->arch.pio_data);
6759 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6763 kvm_pmu_destroy(vcpu);
6764 kfree(vcpu->arch.mce_banks);
6765 kvm_free_lapic(vcpu);
6766 idx = srcu_read_lock(&vcpu->kvm->srcu);
6767 kvm_mmu_destroy(vcpu);
6768 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6769 free_page((unsigned long)vcpu->arch.pio_data);
6770 if (!irqchip_in_kernel(vcpu->kvm))
6771 static_key_slow_dec(&kvm_no_apic_vcpu);
6774 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
6779 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6780 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6782 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6783 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6784 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
6785 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
6786 &kvm->arch.irq_sources_bitmap);
6788 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6789 mutex_init(&kvm->arch.apic_map_lock);
6790 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
6792 pvclock_update_vm_gtod_copy(kvm);
6797 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6800 r = vcpu_load(vcpu);
6802 kvm_mmu_unload(vcpu);
6806 static void kvm_free_vcpus(struct kvm *kvm)
6809 struct kvm_vcpu *vcpu;
6812 * Unpin any mmu pages first.
6814 kvm_for_each_vcpu(i, vcpu, kvm) {
6815 kvm_clear_async_pf_completion_queue(vcpu);
6816 kvm_unload_vcpu_mmu(vcpu);
6818 kvm_for_each_vcpu(i, vcpu, kvm)
6819 kvm_arch_vcpu_free(vcpu);
6821 mutex_lock(&kvm->lock);
6822 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6823 kvm->vcpus[i] = NULL;
6825 atomic_set(&kvm->online_vcpus, 0);
6826 mutex_unlock(&kvm->lock);
6829 void kvm_arch_sync_events(struct kvm *kvm)
6831 kvm_free_all_assigned_devices(kvm);
6835 void kvm_arch_destroy_vm(struct kvm *kvm)
6837 if (current->mm == kvm->mm) {
6839 * Free memory regions allocated on behalf of userspace,
6840 * unless the the memory map has changed due to process exit
6843 struct kvm_userspace_memory_region mem;
6844 memset(&mem, 0, sizeof(mem));
6845 mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
6846 kvm_set_memory_region(kvm, &mem);
6848 mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
6849 kvm_set_memory_region(kvm, &mem);
6851 mem.slot = TSS_PRIVATE_MEMSLOT;
6852 kvm_set_memory_region(kvm, &mem);
6854 kvm_iommu_unmap_guest(kvm);
6855 kfree(kvm->arch.vpic);
6856 kfree(kvm->arch.vioapic);
6857 kvm_free_vcpus(kvm);
6858 if (kvm->arch.apic_access_page)
6859 put_page(kvm->arch.apic_access_page);
6860 if (kvm->arch.ept_identity_pagetable)
6861 put_page(kvm->arch.ept_identity_pagetable);
6862 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
6865 void kvm_arch_free_memslot(struct kvm_memory_slot *free,
6866 struct kvm_memory_slot *dont)
6870 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6871 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
6872 kvm_kvfree(free->arch.rmap[i]);
6873 free->arch.rmap[i] = NULL;
6878 if (!dont || free->arch.lpage_info[i - 1] !=
6879 dont->arch.lpage_info[i - 1]) {
6880 kvm_kvfree(free->arch.lpage_info[i - 1]);
6881 free->arch.lpage_info[i - 1] = NULL;
6886 int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages)
6890 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6895 lpages = gfn_to_index(slot->base_gfn + npages - 1,
6896 slot->base_gfn, level) + 1;
6898 slot->arch.rmap[i] =
6899 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
6900 if (!slot->arch.rmap[i])
6905 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
6906 sizeof(*slot->arch.lpage_info[i - 1]));
6907 if (!slot->arch.lpage_info[i - 1])
6910 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
6911 slot->arch.lpage_info[i - 1][0].write_count = 1;
6912 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
6913 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
6914 ugfn = slot->userspace_addr >> PAGE_SHIFT;
6916 * If the gfn and userspace address are not aligned wrt each
6917 * other, or if explicitly asked to, disable large page
6918 * support for this slot
6920 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
6921 !kvm_largepages_enabled()) {
6924 for (j = 0; j < lpages; ++j)
6925 slot->arch.lpage_info[i - 1][j].write_count = 1;
6932 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
6933 kvm_kvfree(slot->arch.rmap[i]);
6934 slot->arch.rmap[i] = NULL;
6938 kvm_kvfree(slot->arch.lpage_info[i - 1]);
6939 slot->arch.lpage_info[i - 1] = NULL;
6944 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6945 struct kvm_memory_slot *memslot,
6946 struct kvm_userspace_memory_region *mem,
6947 enum kvm_mr_change change)
6950 * Only private memory slots need to be mapped here since
6951 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
6953 if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
6954 unsigned long userspace_addr;
6957 * MAP_SHARED to prevent internal slot pages from being moved
6960 userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
6961 PROT_READ | PROT_WRITE,
6962 MAP_SHARED | MAP_ANONYMOUS, 0);
6964 if (IS_ERR((void *)userspace_addr))
6965 return PTR_ERR((void *)userspace_addr);
6967 memslot->userspace_addr = userspace_addr;
6973 void kvm_arch_commit_memory_region(struct kvm *kvm,
6974 struct kvm_userspace_memory_region *mem,
6975 const struct kvm_memory_slot *old,
6976 enum kvm_mr_change change)
6979 int nr_mmu_pages = 0;
6981 if ((mem->slot >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_DELETE)) {
6984 ret = vm_munmap(old->userspace_addr,
6985 old->npages * PAGE_SIZE);
6988 "kvm_vm_ioctl_set_memory_region: "
6989 "failed to munmap memory\n");
6992 if (!kvm->arch.n_requested_mmu_pages)
6993 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6996 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6998 * Write protect all pages for dirty logging.
6999 * Existing largepage mappings are destroyed here and new ones will
7000 * not be created until the end of the logging.
7002 if ((change != KVM_MR_DELETE) && (mem->flags & KVM_MEM_LOG_DIRTY_PAGES))
7003 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
7005 * If memory slot is created, or moved, we need to clear all
7008 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
7009 kvm_mmu_zap_mmio_sptes(kvm);
7010 kvm_reload_remote_mmus(kvm);
7014 void kvm_arch_flush_shadow_all(struct kvm *kvm)
7016 kvm_mmu_zap_all(kvm);
7017 kvm_reload_remote_mmus(kvm);
7020 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7021 struct kvm_memory_slot *slot)
7023 kvm_arch_flush_shadow_all(kvm);
7026 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
7028 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7029 !vcpu->arch.apf.halted)
7030 || !list_empty_careful(&vcpu->async_pf.done)
7031 || kvm_apic_has_events(vcpu)
7032 || atomic_read(&vcpu->arch.nmi_queued) ||
7033 (kvm_arch_interrupt_allowed(vcpu) &&
7034 kvm_cpu_has_interrupt(vcpu));
7037 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
7039 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
7042 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7044 return kvm_x86_ops->interrupt_allowed(vcpu);
7047 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7049 unsigned long current_rip = kvm_rip_read(vcpu) +
7050 get_segment_base(vcpu, VCPU_SREG_CS);
7052 return current_rip == linear_rip;
7054 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7056 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7058 unsigned long rflags;
7060 rflags = kvm_x86_ops->get_rflags(vcpu);
7061 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7062 rflags &= ~X86_EFLAGS_TF;
7065 EXPORT_SYMBOL_GPL(kvm_get_rflags);
7067 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7069 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
7070 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
7071 rflags |= X86_EFLAGS_TF;
7072 kvm_x86_ops->set_rflags(vcpu, rflags);
7073 kvm_make_request(KVM_REQ_EVENT, vcpu);
7075 EXPORT_SYMBOL_GPL(kvm_set_rflags);
7077 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7081 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
7082 is_error_page(work->page))
7085 r = kvm_mmu_reload(vcpu);
7089 if (!vcpu->arch.mmu.direct_map &&
7090 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7093 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7096 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7098 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7101 static inline u32 kvm_async_pf_next_probe(u32 key)
7103 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7106 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7108 u32 key = kvm_async_pf_hash_fn(gfn);
7110 while (vcpu->arch.apf.gfns[key] != ~0)
7111 key = kvm_async_pf_next_probe(key);
7113 vcpu->arch.apf.gfns[key] = gfn;
7116 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7119 u32 key = kvm_async_pf_hash_fn(gfn);
7121 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
7122 (vcpu->arch.apf.gfns[key] != gfn &&
7123 vcpu->arch.apf.gfns[key] != ~0); i++)
7124 key = kvm_async_pf_next_probe(key);
7129 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7131 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7134 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7138 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7140 vcpu->arch.apf.gfns[i] = ~0;
7142 j = kvm_async_pf_next_probe(j);
7143 if (vcpu->arch.apf.gfns[j] == ~0)
7145 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7147 * k lies cyclically in ]i,j]
7149 * |....j i.k.| or |.k..j i...|
7151 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7152 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7157 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7160 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7164 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7165 struct kvm_async_pf *work)
7167 struct x86_exception fault;
7169 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
7170 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7172 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
7173 (vcpu->arch.apf.send_user_only &&
7174 kvm_x86_ops->get_cpl(vcpu) == 0))
7175 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7176 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
7177 fault.vector = PF_VECTOR;
7178 fault.error_code_valid = true;
7179 fault.error_code = 0;
7180 fault.nested_page_fault = false;
7181 fault.address = work->arch.token;
7182 kvm_inject_page_fault(vcpu, &fault);
7186 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7187 struct kvm_async_pf *work)
7189 struct x86_exception fault;
7191 trace_kvm_async_pf_ready(work->arch.token, work->gva);
7192 if (is_error_page(work->page))
7193 work->arch.token = ~0; /* broadcast wakeup */
7195 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7197 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7198 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
7199 fault.vector = PF_VECTOR;
7200 fault.error_code_valid = true;
7201 fault.error_code = 0;
7202 fault.nested_page_fault = false;
7203 fault.address = work->arch.token;
7204 kvm_inject_page_fault(vcpu, &fault);
7206 vcpu->arch.apf.halted = false;
7207 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7210 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7212 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7215 return !kvm_event_needs_reinjection(vcpu) &&
7216 kvm_x86_ops->interrupt_allowed(vcpu);
7219 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
7220 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
7221 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
7222 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
7223 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
7224 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
7225 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
7226 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
7227 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
7228 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
7229 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
7230 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);