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 <trace/events/kvm.h>
51 #define CREATE_TRACE_POINTS
54 #include <asm/debugreg.h>
61 #include <asm/pvclock.h>
62 #include <asm/div64.h>
64 #define MAX_IO_MSRS 256
65 #define KVM_MAX_MCE_BANKS 32
66 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
68 #define emul_to_vcpu(ctxt) \
69 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
72 * - enable syscall per default because its emulated by KVM
73 * - enable LME and LMA per default on 64 bit KVM
77 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
79 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
82 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
83 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
85 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
86 static void process_nmi(struct kvm_vcpu *vcpu);
88 struct kvm_x86_ops *kvm_x86_ops;
89 EXPORT_SYMBOL_GPL(kvm_x86_ops);
91 static bool ignore_msrs = 0;
92 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
94 bool kvm_has_tsc_control;
95 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
96 u32 kvm_max_guest_tsc_khz;
97 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
99 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
100 static u32 tsc_tolerance_ppm = 250;
101 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
103 #define KVM_NR_SHARED_MSRS 16
105 struct kvm_shared_msrs_global {
107 u32 msrs[KVM_NR_SHARED_MSRS];
110 struct kvm_shared_msrs {
111 struct user_return_notifier urn;
113 struct kvm_shared_msr_values {
116 } values[KVM_NR_SHARED_MSRS];
119 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
120 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
122 struct kvm_stats_debugfs_item debugfs_entries[] = {
123 { "pf_fixed", VCPU_STAT(pf_fixed) },
124 { "pf_guest", VCPU_STAT(pf_guest) },
125 { "tlb_flush", VCPU_STAT(tlb_flush) },
126 { "invlpg", VCPU_STAT(invlpg) },
127 { "exits", VCPU_STAT(exits) },
128 { "io_exits", VCPU_STAT(io_exits) },
129 { "mmio_exits", VCPU_STAT(mmio_exits) },
130 { "signal_exits", VCPU_STAT(signal_exits) },
131 { "irq_window", VCPU_STAT(irq_window_exits) },
132 { "nmi_window", VCPU_STAT(nmi_window_exits) },
133 { "halt_exits", VCPU_STAT(halt_exits) },
134 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
135 { "hypercalls", VCPU_STAT(hypercalls) },
136 { "request_irq", VCPU_STAT(request_irq_exits) },
137 { "irq_exits", VCPU_STAT(irq_exits) },
138 { "host_state_reload", VCPU_STAT(host_state_reload) },
139 { "efer_reload", VCPU_STAT(efer_reload) },
140 { "fpu_reload", VCPU_STAT(fpu_reload) },
141 { "insn_emulation", VCPU_STAT(insn_emulation) },
142 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
143 { "irq_injections", VCPU_STAT(irq_injections) },
144 { "nmi_injections", VCPU_STAT(nmi_injections) },
145 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
146 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
147 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
148 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
149 { "mmu_flooded", VM_STAT(mmu_flooded) },
150 { "mmu_recycled", VM_STAT(mmu_recycled) },
151 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
152 { "mmu_unsync", VM_STAT(mmu_unsync) },
153 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
154 { "largepages", VM_STAT(lpages) },
158 u64 __read_mostly host_xcr0;
160 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
162 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
165 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
166 vcpu->arch.apf.gfns[i] = ~0;
169 static void kvm_on_user_return(struct user_return_notifier *urn)
172 struct kvm_shared_msrs *locals
173 = container_of(urn, struct kvm_shared_msrs, urn);
174 struct kvm_shared_msr_values *values;
176 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
177 values = &locals->values[slot];
178 if (values->host != values->curr) {
179 wrmsrl(shared_msrs_global.msrs[slot], values->host);
180 values->curr = values->host;
183 locals->registered = false;
184 user_return_notifier_unregister(urn);
187 static void shared_msr_update(unsigned slot, u32 msr)
189 struct kvm_shared_msrs *smsr;
192 smsr = &__get_cpu_var(shared_msrs);
193 /* only read, and nobody should modify it at this time,
194 * so don't need lock */
195 if (slot >= shared_msrs_global.nr) {
196 printk(KERN_ERR "kvm: invalid MSR slot!");
199 rdmsrl_safe(msr, &value);
200 smsr->values[slot].host = value;
201 smsr->values[slot].curr = value;
204 void kvm_define_shared_msr(unsigned slot, u32 msr)
206 if (slot >= shared_msrs_global.nr)
207 shared_msrs_global.nr = slot + 1;
208 shared_msrs_global.msrs[slot] = msr;
209 /* we need ensured the shared_msr_global have been updated */
212 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
214 static void kvm_shared_msr_cpu_online(void)
218 for (i = 0; i < shared_msrs_global.nr; ++i)
219 shared_msr_update(i, shared_msrs_global.msrs[i]);
222 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
224 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
226 if (((value ^ smsr->values[slot].curr) & mask) == 0)
228 smsr->values[slot].curr = value;
229 wrmsrl(shared_msrs_global.msrs[slot], value);
230 if (!smsr->registered) {
231 smsr->urn.on_user_return = kvm_on_user_return;
232 user_return_notifier_register(&smsr->urn);
233 smsr->registered = true;
236 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
238 static void drop_user_return_notifiers(void *ignore)
240 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
242 if (smsr->registered)
243 kvm_on_user_return(&smsr->urn);
246 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
248 if (irqchip_in_kernel(vcpu->kvm))
249 return vcpu->arch.apic_base;
251 return vcpu->arch.apic_base;
253 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
255 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
257 /* TODO: reserve bits check */
258 if (irqchip_in_kernel(vcpu->kvm))
259 kvm_lapic_set_base(vcpu, data);
261 vcpu->arch.apic_base = data;
263 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
265 #define EXCPT_BENIGN 0
266 #define EXCPT_CONTRIBUTORY 1
269 static int exception_class(int vector)
279 return EXCPT_CONTRIBUTORY;
286 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
287 unsigned nr, bool has_error, u32 error_code,
293 kvm_make_request(KVM_REQ_EVENT, vcpu);
295 if (!vcpu->arch.exception.pending) {
297 vcpu->arch.exception.pending = true;
298 vcpu->arch.exception.has_error_code = has_error;
299 vcpu->arch.exception.nr = nr;
300 vcpu->arch.exception.error_code = error_code;
301 vcpu->arch.exception.reinject = reinject;
305 /* to check exception */
306 prev_nr = vcpu->arch.exception.nr;
307 if (prev_nr == DF_VECTOR) {
308 /* triple fault -> shutdown */
309 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
312 class1 = exception_class(prev_nr);
313 class2 = exception_class(nr);
314 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
315 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
316 /* generate double fault per SDM Table 5-5 */
317 vcpu->arch.exception.pending = true;
318 vcpu->arch.exception.has_error_code = true;
319 vcpu->arch.exception.nr = DF_VECTOR;
320 vcpu->arch.exception.error_code = 0;
322 /* replace previous exception with a new one in a hope
323 that instruction re-execution will regenerate lost
328 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
330 kvm_multiple_exception(vcpu, nr, false, 0, false);
332 EXPORT_SYMBOL_GPL(kvm_queue_exception);
334 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
336 kvm_multiple_exception(vcpu, nr, false, 0, true);
338 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
340 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
343 kvm_inject_gp(vcpu, 0);
345 kvm_x86_ops->skip_emulated_instruction(vcpu);
347 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
349 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
351 ++vcpu->stat.pf_guest;
352 vcpu->arch.cr2 = fault->address;
353 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
355 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
357 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
359 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
360 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
362 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
365 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
367 atomic_inc(&vcpu->arch.nmi_queued);
368 kvm_make_request(KVM_REQ_NMI, vcpu);
370 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
372 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
374 kvm_multiple_exception(vcpu, nr, true, error_code, false);
376 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
378 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
380 kvm_multiple_exception(vcpu, nr, true, error_code, true);
382 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
385 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
386 * a #GP and return false.
388 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
390 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
392 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
395 EXPORT_SYMBOL_GPL(kvm_require_cpl);
398 * This function will be used to read from the physical memory of the currently
399 * running guest. The difference to kvm_read_guest_page is that this function
400 * can read from guest physical or from the guest's guest physical memory.
402 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
403 gfn_t ngfn, void *data, int offset, int len,
409 ngpa = gfn_to_gpa(ngfn);
410 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
411 if (real_gfn == UNMAPPED_GVA)
414 real_gfn = gpa_to_gfn(real_gfn);
416 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
418 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
420 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
421 void *data, int offset, int len, u32 access)
423 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
424 data, offset, len, access);
428 * Load the pae pdptrs. Return true is they are all valid.
430 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
432 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
433 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
436 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
438 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
439 offset * sizeof(u64), sizeof(pdpte),
440 PFERR_USER_MASK|PFERR_WRITE_MASK);
445 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
446 if (is_present_gpte(pdpte[i]) &&
447 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
454 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
455 __set_bit(VCPU_EXREG_PDPTR,
456 (unsigned long *)&vcpu->arch.regs_avail);
457 __set_bit(VCPU_EXREG_PDPTR,
458 (unsigned long *)&vcpu->arch.regs_dirty);
463 EXPORT_SYMBOL_GPL(load_pdptrs);
465 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
467 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
473 if (is_long_mode(vcpu) || !is_pae(vcpu))
476 if (!test_bit(VCPU_EXREG_PDPTR,
477 (unsigned long *)&vcpu->arch.regs_avail))
480 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
481 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
482 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
483 PFERR_USER_MASK | PFERR_WRITE_MASK);
486 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
492 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
494 unsigned long old_cr0 = kvm_read_cr0(vcpu);
495 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
496 X86_CR0_CD | X86_CR0_NW;
501 if (cr0 & 0xffffffff00000000UL)
505 cr0 &= ~CR0_RESERVED_BITS;
507 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
510 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
513 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
515 if ((vcpu->arch.efer & EFER_LME)) {
520 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
525 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
530 kvm_x86_ops->set_cr0(vcpu, cr0);
532 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
533 kvm_clear_async_pf_completion_queue(vcpu);
534 kvm_async_pf_hash_reset(vcpu);
537 if ((cr0 ^ old_cr0) & update_bits)
538 kvm_mmu_reset_context(vcpu);
541 EXPORT_SYMBOL_GPL(kvm_set_cr0);
543 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
545 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
547 EXPORT_SYMBOL_GPL(kvm_lmsw);
549 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
553 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
554 if (index != XCR_XFEATURE_ENABLED_MASK)
557 if (kvm_x86_ops->get_cpl(vcpu) != 0)
559 if (!(xcr0 & XSTATE_FP))
561 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
563 if (xcr0 & ~host_xcr0)
565 vcpu->arch.xcr0 = xcr0;
566 vcpu->guest_xcr0_loaded = 0;
570 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
572 if (__kvm_set_xcr(vcpu, index, xcr)) {
573 kvm_inject_gp(vcpu, 0);
578 EXPORT_SYMBOL_GPL(kvm_set_xcr);
580 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
582 unsigned long old_cr4 = kvm_read_cr4(vcpu);
583 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
584 X86_CR4_PAE | X86_CR4_SMEP;
585 if (cr4 & CR4_RESERVED_BITS)
588 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
591 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
594 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_RDWRGSFS))
597 if (is_long_mode(vcpu)) {
598 if (!(cr4 & X86_CR4_PAE))
600 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
601 && ((cr4 ^ old_cr4) & pdptr_bits)
602 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
606 if (kvm_x86_ops->set_cr4(vcpu, cr4))
609 if ((cr4 ^ old_cr4) & pdptr_bits)
610 kvm_mmu_reset_context(vcpu);
612 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
613 kvm_update_cpuid(vcpu);
617 EXPORT_SYMBOL_GPL(kvm_set_cr4);
619 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
621 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
622 kvm_mmu_sync_roots(vcpu);
623 kvm_mmu_flush_tlb(vcpu);
627 if (is_long_mode(vcpu)) {
628 if (cr3 & CR3_L_MODE_RESERVED_BITS)
632 if (cr3 & CR3_PAE_RESERVED_BITS)
634 if (is_paging(vcpu) &&
635 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
639 * We don't check reserved bits in nonpae mode, because
640 * this isn't enforced, and VMware depends on this.
645 * Does the new cr3 value map to physical memory? (Note, we
646 * catch an invalid cr3 even in real-mode, because it would
647 * cause trouble later on when we turn on paging anyway.)
649 * A real CPU would silently accept an invalid cr3 and would
650 * attempt to use it - with largely undefined (and often hard
651 * to debug) behavior on the guest side.
653 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
655 vcpu->arch.cr3 = cr3;
656 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
657 vcpu->arch.mmu.new_cr3(vcpu);
660 EXPORT_SYMBOL_GPL(kvm_set_cr3);
662 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
664 if (cr8 & CR8_RESERVED_BITS)
666 if (irqchip_in_kernel(vcpu->kvm))
667 kvm_lapic_set_tpr(vcpu, cr8);
669 vcpu->arch.cr8 = cr8;
672 EXPORT_SYMBOL_GPL(kvm_set_cr8);
674 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
676 if (irqchip_in_kernel(vcpu->kvm))
677 return kvm_lapic_get_cr8(vcpu);
679 return vcpu->arch.cr8;
681 EXPORT_SYMBOL_GPL(kvm_get_cr8);
683 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
687 vcpu->arch.db[dr] = val;
688 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
689 vcpu->arch.eff_db[dr] = val;
692 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
696 if (val & 0xffffffff00000000ULL)
698 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
701 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
705 if (val & 0xffffffff00000000ULL)
707 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
708 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
709 kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
710 vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
718 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
722 res = __kvm_set_dr(vcpu, dr, val);
724 kvm_queue_exception(vcpu, UD_VECTOR);
726 kvm_inject_gp(vcpu, 0);
730 EXPORT_SYMBOL_GPL(kvm_set_dr);
732 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
736 *val = vcpu->arch.db[dr];
739 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
743 *val = vcpu->arch.dr6;
746 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
750 *val = vcpu->arch.dr7;
757 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
759 if (_kvm_get_dr(vcpu, dr, val)) {
760 kvm_queue_exception(vcpu, UD_VECTOR);
765 EXPORT_SYMBOL_GPL(kvm_get_dr);
767 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
769 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
773 err = kvm_pmu_read_pmc(vcpu, ecx, &data);
776 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
777 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
780 EXPORT_SYMBOL_GPL(kvm_rdpmc);
783 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
784 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
786 * This list is modified at module load time to reflect the
787 * capabilities of the host cpu. This capabilities test skips MSRs that are
788 * kvm-specific. Those are put in the beginning of the list.
791 #define KVM_SAVE_MSRS_BEGIN 9
792 static u32 msrs_to_save[] = {
793 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
794 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
795 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
796 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
797 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
800 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
802 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
805 static unsigned num_msrs_to_save;
807 static u32 emulated_msrs[] = {
808 MSR_IA32_TSCDEADLINE,
809 MSR_IA32_MISC_ENABLE,
814 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
816 u64 old_efer = vcpu->arch.efer;
818 if (efer & efer_reserved_bits)
822 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
825 if (efer & EFER_FFXSR) {
826 struct kvm_cpuid_entry2 *feat;
828 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
829 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
833 if (efer & EFER_SVME) {
834 struct kvm_cpuid_entry2 *feat;
836 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
837 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
842 efer |= vcpu->arch.efer & EFER_LMA;
844 kvm_x86_ops->set_efer(vcpu, efer);
846 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
848 /* Update reserved bits */
849 if ((efer ^ old_efer) & EFER_NX)
850 kvm_mmu_reset_context(vcpu);
855 void kvm_enable_efer_bits(u64 mask)
857 efer_reserved_bits &= ~mask;
859 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
863 * Writes msr value into into the appropriate "register".
864 * Returns 0 on success, non-0 otherwise.
865 * Assumes vcpu_load() was already called.
867 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
869 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
873 * Adapt set_msr() to msr_io()'s calling convention
875 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
877 return kvm_set_msr(vcpu, index, *data);
880 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
884 struct pvclock_wall_clock wc;
885 struct timespec boot;
890 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
895 ++version; /* first time write, random junk */
899 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
902 * The guest calculates current wall clock time by adding
903 * system time (updated by kvm_guest_time_update below) to the
904 * wall clock specified here. guest system time equals host
905 * system time for us, thus we must fill in host boot time here.
909 wc.sec = boot.tv_sec;
910 wc.nsec = boot.tv_nsec;
911 wc.version = version;
913 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
916 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
919 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
921 uint32_t quotient, remainder;
923 /* Don't try to replace with do_div(), this one calculates
924 * "(dividend << 32) / divisor" */
926 : "=a" (quotient), "=d" (remainder)
927 : "0" (0), "1" (dividend), "r" (divisor) );
931 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
932 s8 *pshift, u32 *pmultiplier)
939 tps64 = base_khz * 1000LL;
940 scaled64 = scaled_khz * 1000LL;
941 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
946 tps32 = (uint32_t)tps64;
947 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
948 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
956 *pmultiplier = div_frac(scaled64, tps32);
958 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
959 __func__, base_khz, scaled_khz, shift, *pmultiplier);
962 static inline u64 get_kernel_ns(void)
966 WARN_ON(preemptible());
968 monotonic_to_bootbased(&ts);
969 return timespec_to_ns(&ts);
972 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
973 unsigned long max_tsc_khz;
975 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
977 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
978 vcpu->arch.virtual_tsc_shift);
981 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
983 u64 v = (u64)khz * (1000000 + ppm);
988 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
990 u32 thresh_lo, thresh_hi;
993 /* Compute a scale to convert nanoseconds in TSC cycles */
994 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
995 &vcpu->arch.virtual_tsc_shift,
996 &vcpu->arch.virtual_tsc_mult);
997 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1000 * Compute the variation in TSC rate which is acceptable
1001 * within the range of tolerance and decide if the
1002 * rate being applied is within that bounds of the hardware
1003 * rate. If so, no scaling or compensation need be done.
1005 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1006 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1007 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1008 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1011 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1014 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1016 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1017 vcpu->arch.virtual_tsc_mult,
1018 vcpu->arch.virtual_tsc_shift);
1019 tsc += vcpu->arch.last_tsc_write;
1023 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1025 struct kvm *kvm = vcpu->kvm;
1026 u64 offset, ns, elapsed;
1027 unsigned long flags;
1030 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1031 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1032 ns = get_kernel_ns();
1033 elapsed = ns - kvm->arch.last_tsc_nsec;
1035 /* n.b - signed multiplication and division required */
1036 nsdiff = data - kvm->arch.last_tsc_write;
1037 #ifdef CONFIG_X86_64
1038 nsdiff = (nsdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1040 /* do_div() only does unsigned */
1041 asm("idivl %2; xor %%edx, %%edx"
1043 : "A"(nsdiff * 1000), "rm"(vcpu->arch.virtual_tsc_khz));
1050 * Special case: TSC write with a small delta (1 second) of virtual
1051 * cycle time against real time is interpreted as an attempt to
1052 * synchronize the CPU.
1054 * For a reliable TSC, we can match TSC offsets, and for an unstable
1055 * TSC, we add elapsed time in this computation. We could let the
1056 * compensation code attempt to catch up if we fall behind, but
1057 * it's better to try to match offsets from the beginning.
1059 if (nsdiff < NSEC_PER_SEC &&
1060 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1061 if (!check_tsc_unstable()) {
1062 offset = kvm->arch.last_tsc_offset;
1063 pr_debug("kvm: matched tsc offset for %llu\n", data);
1065 u64 delta = nsec_to_cycles(vcpu, elapsed);
1067 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1068 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1071 kvm->arch.last_tsc_nsec = ns;
1072 kvm->arch.last_tsc_write = data;
1073 kvm->arch.last_tsc_offset = offset;
1074 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1075 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1076 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1078 /* Reset of TSC must disable overshoot protection below */
1079 vcpu->arch.hv_clock.tsc_timestamp = 0;
1080 vcpu->arch.last_tsc_write = data;
1081 vcpu->arch.last_tsc_nsec = ns;
1082 vcpu->arch.last_guest_tsc = data;
1084 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1086 static int kvm_guest_time_update(struct kvm_vcpu *v)
1088 unsigned long flags;
1089 struct kvm_vcpu_arch *vcpu = &v->arch;
1091 unsigned long this_tsc_khz;
1092 s64 kernel_ns, max_kernel_ns;
1095 /* Keep irq disabled to prevent changes to the clock */
1096 local_irq_save(flags);
1097 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v);
1098 kernel_ns = get_kernel_ns();
1099 this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
1100 if (unlikely(this_tsc_khz == 0)) {
1101 local_irq_restore(flags);
1102 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1107 * We may have to catch up the TSC to match elapsed wall clock
1108 * time for two reasons, even if kvmclock is used.
1109 * 1) CPU could have been running below the maximum TSC rate
1110 * 2) Broken TSC compensation resets the base at each VCPU
1111 * entry to avoid unknown leaps of TSC even when running
1112 * again on the same CPU. This may cause apparent elapsed
1113 * time to disappear, and the guest to stand still or run
1116 if (vcpu->tsc_catchup) {
1117 u64 tsc = compute_guest_tsc(v, kernel_ns);
1118 if (tsc > tsc_timestamp) {
1119 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1120 tsc_timestamp = tsc;
1124 local_irq_restore(flags);
1126 if (!vcpu->time_page)
1130 * Time as measured by the TSC may go backwards when resetting the base
1131 * tsc_timestamp. The reason for this is that the TSC resolution is
1132 * higher than the resolution of the other clock scales. Thus, many
1133 * possible measurments of the TSC correspond to one measurement of any
1134 * other clock, and so a spread of values is possible. This is not a
1135 * problem for the computation of the nanosecond clock; with TSC rates
1136 * around 1GHZ, there can only be a few cycles which correspond to one
1137 * nanosecond value, and any path through this code will inevitably
1138 * take longer than that. However, with the kernel_ns value itself,
1139 * the precision may be much lower, down to HZ granularity. If the
1140 * first sampling of TSC against kernel_ns ends in the low part of the
1141 * range, and the second in the high end of the range, we can get:
1143 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1145 * As the sampling errors potentially range in the thousands of cycles,
1146 * it is possible such a time value has already been observed by the
1147 * guest. To protect against this, we must compute the system time as
1148 * observed by the guest and ensure the new system time is greater.
1151 if (vcpu->hv_clock.tsc_timestamp) {
1152 max_kernel_ns = vcpu->last_guest_tsc -
1153 vcpu->hv_clock.tsc_timestamp;
1154 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1155 vcpu->hv_clock.tsc_to_system_mul,
1156 vcpu->hv_clock.tsc_shift);
1157 max_kernel_ns += vcpu->last_kernel_ns;
1160 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1161 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1162 &vcpu->hv_clock.tsc_shift,
1163 &vcpu->hv_clock.tsc_to_system_mul);
1164 vcpu->hw_tsc_khz = this_tsc_khz;
1167 if (max_kernel_ns > kernel_ns)
1168 kernel_ns = max_kernel_ns;
1170 /* With all the info we got, fill in the values */
1171 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1172 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1173 vcpu->last_kernel_ns = kernel_ns;
1174 vcpu->last_guest_tsc = tsc_timestamp;
1175 vcpu->hv_clock.flags = 0;
1178 * The interface expects us to write an even number signaling that the
1179 * update is finished. Since the guest won't see the intermediate
1180 * state, we just increase by 2 at the end.
1182 vcpu->hv_clock.version += 2;
1184 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1186 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1187 sizeof(vcpu->hv_clock));
1189 kunmap_atomic(shared_kaddr, KM_USER0);
1191 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1195 static bool msr_mtrr_valid(unsigned msr)
1198 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1199 case MSR_MTRRfix64K_00000:
1200 case MSR_MTRRfix16K_80000:
1201 case MSR_MTRRfix16K_A0000:
1202 case MSR_MTRRfix4K_C0000:
1203 case MSR_MTRRfix4K_C8000:
1204 case MSR_MTRRfix4K_D0000:
1205 case MSR_MTRRfix4K_D8000:
1206 case MSR_MTRRfix4K_E0000:
1207 case MSR_MTRRfix4K_E8000:
1208 case MSR_MTRRfix4K_F0000:
1209 case MSR_MTRRfix4K_F8000:
1210 case MSR_MTRRdefType:
1211 case MSR_IA32_CR_PAT:
1219 static bool valid_pat_type(unsigned t)
1221 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1224 static bool valid_mtrr_type(unsigned t)
1226 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1229 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1233 if (!msr_mtrr_valid(msr))
1236 if (msr == MSR_IA32_CR_PAT) {
1237 for (i = 0; i < 8; i++)
1238 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1241 } else if (msr == MSR_MTRRdefType) {
1244 return valid_mtrr_type(data & 0xff);
1245 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1246 for (i = 0; i < 8 ; i++)
1247 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1252 /* variable MTRRs */
1253 return valid_mtrr_type(data & 0xff);
1256 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1258 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1260 if (!mtrr_valid(vcpu, msr, data))
1263 if (msr == MSR_MTRRdefType) {
1264 vcpu->arch.mtrr_state.def_type = data;
1265 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1266 } else if (msr == MSR_MTRRfix64K_00000)
1268 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1269 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1270 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1271 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1272 else if (msr == MSR_IA32_CR_PAT)
1273 vcpu->arch.pat = data;
1274 else { /* Variable MTRRs */
1275 int idx, is_mtrr_mask;
1278 idx = (msr - 0x200) / 2;
1279 is_mtrr_mask = msr - 0x200 - 2 * idx;
1282 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1285 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1289 kvm_mmu_reset_context(vcpu);
1293 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1295 u64 mcg_cap = vcpu->arch.mcg_cap;
1296 unsigned bank_num = mcg_cap & 0xff;
1299 case MSR_IA32_MCG_STATUS:
1300 vcpu->arch.mcg_status = data;
1302 case MSR_IA32_MCG_CTL:
1303 if (!(mcg_cap & MCG_CTL_P))
1305 if (data != 0 && data != ~(u64)0)
1307 vcpu->arch.mcg_ctl = data;
1310 if (msr >= MSR_IA32_MC0_CTL &&
1311 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1312 u32 offset = msr - MSR_IA32_MC0_CTL;
1313 /* only 0 or all 1s can be written to IA32_MCi_CTL
1314 * some Linux kernels though clear bit 10 in bank 4 to
1315 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1316 * this to avoid an uncatched #GP in the guest
1318 if ((offset & 0x3) == 0 &&
1319 data != 0 && (data | (1 << 10)) != ~(u64)0)
1321 vcpu->arch.mce_banks[offset] = data;
1329 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1331 struct kvm *kvm = vcpu->kvm;
1332 int lm = is_long_mode(vcpu);
1333 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1334 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1335 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1336 : kvm->arch.xen_hvm_config.blob_size_32;
1337 u32 page_num = data & ~PAGE_MASK;
1338 u64 page_addr = data & PAGE_MASK;
1343 if (page_num >= blob_size)
1346 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1351 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1360 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1362 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1365 static bool kvm_hv_msr_partition_wide(u32 msr)
1369 case HV_X64_MSR_GUEST_OS_ID:
1370 case HV_X64_MSR_HYPERCALL:
1378 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1380 struct kvm *kvm = vcpu->kvm;
1383 case HV_X64_MSR_GUEST_OS_ID:
1384 kvm->arch.hv_guest_os_id = data;
1385 /* setting guest os id to zero disables hypercall page */
1386 if (!kvm->arch.hv_guest_os_id)
1387 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1389 case HV_X64_MSR_HYPERCALL: {
1394 /* if guest os id is not set hypercall should remain disabled */
1395 if (!kvm->arch.hv_guest_os_id)
1397 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1398 kvm->arch.hv_hypercall = data;
1401 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1402 addr = gfn_to_hva(kvm, gfn);
1403 if (kvm_is_error_hva(addr))
1405 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1406 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1407 if (__copy_to_user((void __user *)addr, instructions, 4))
1409 kvm->arch.hv_hypercall = data;
1413 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1414 "data 0x%llx\n", msr, data);
1420 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1423 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1426 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1427 vcpu->arch.hv_vapic = data;
1430 addr = gfn_to_hva(vcpu->kvm, data >>
1431 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1432 if (kvm_is_error_hva(addr))
1434 if (__clear_user((void __user *)addr, PAGE_SIZE))
1436 vcpu->arch.hv_vapic = data;
1439 case HV_X64_MSR_EOI:
1440 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1441 case HV_X64_MSR_ICR:
1442 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1443 case HV_X64_MSR_TPR:
1444 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1446 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1447 "data 0x%llx\n", msr, data);
1454 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1456 gpa_t gpa = data & ~0x3f;
1458 /* Bits 2:5 are resrved, Should be zero */
1462 vcpu->arch.apf.msr_val = data;
1464 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1465 kvm_clear_async_pf_completion_queue(vcpu);
1466 kvm_async_pf_hash_reset(vcpu);
1470 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1473 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1474 kvm_async_pf_wakeup_all(vcpu);
1478 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1480 if (vcpu->arch.time_page) {
1481 kvm_release_page_dirty(vcpu->arch.time_page);
1482 vcpu->arch.time_page = NULL;
1486 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1490 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1493 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1494 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1495 vcpu->arch.st.accum_steal = delta;
1498 static void record_steal_time(struct kvm_vcpu *vcpu)
1500 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1503 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1504 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1507 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1508 vcpu->arch.st.steal.version += 2;
1509 vcpu->arch.st.accum_steal = 0;
1511 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1512 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1515 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1521 return set_efer(vcpu, data);
1523 data &= ~(u64)0x40; /* ignore flush filter disable */
1524 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1526 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1531 case MSR_FAM10H_MMIO_CONF_BASE:
1533 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1538 case MSR_AMD64_NB_CFG:
1540 case MSR_IA32_DEBUGCTLMSR:
1542 /* We support the non-activated case already */
1544 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1545 /* Values other than LBR and BTF are vendor-specific,
1546 thus reserved and should throw a #GP */
1549 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1552 case MSR_IA32_UCODE_REV:
1553 case MSR_IA32_UCODE_WRITE:
1554 case MSR_VM_HSAVE_PA:
1555 case MSR_AMD64_PATCH_LOADER:
1557 case 0x200 ... 0x2ff:
1558 return set_msr_mtrr(vcpu, msr, data);
1559 case MSR_IA32_APICBASE:
1560 kvm_set_apic_base(vcpu, data);
1562 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1563 return kvm_x2apic_msr_write(vcpu, msr, data);
1564 case MSR_IA32_TSCDEADLINE:
1565 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1567 case MSR_IA32_MISC_ENABLE:
1568 vcpu->arch.ia32_misc_enable_msr = data;
1570 case MSR_KVM_WALL_CLOCK_NEW:
1571 case MSR_KVM_WALL_CLOCK:
1572 vcpu->kvm->arch.wall_clock = data;
1573 kvm_write_wall_clock(vcpu->kvm, data);
1575 case MSR_KVM_SYSTEM_TIME_NEW:
1576 case MSR_KVM_SYSTEM_TIME: {
1577 kvmclock_reset(vcpu);
1579 vcpu->arch.time = data;
1580 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1582 /* we verify if the enable bit is set... */
1586 /* ...but clean it before doing the actual write */
1587 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1589 vcpu->arch.time_page =
1590 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1592 if (is_error_page(vcpu->arch.time_page)) {
1593 kvm_release_page_clean(vcpu->arch.time_page);
1594 vcpu->arch.time_page = NULL;
1598 case MSR_KVM_ASYNC_PF_EN:
1599 if (kvm_pv_enable_async_pf(vcpu, data))
1602 case MSR_KVM_STEAL_TIME:
1604 if (unlikely(!sched_info_on()))
1607 if (data & KVM_STEAL_RESERVED_MASK)
1610 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
1611 data & KVM_STEAL_VALID_BITS))
1614 vcpu->arch.st.msr_val = data;
1616 if (!(data & KVM_MSR_ENABLED))
1619 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1622 accumulate_steal_time(vcpu);
1625 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
1629 case MSR_IA32_MCG_CTL:
1630 case MSR_IA32_MCG_STATUS:
1631 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1632 return set_msr_mce(vcpu, msr, data);
1634 /* Performance counters are not protected by a CPUID bit,
1635 * so we should check all of them in the generic path for the sake of
1636 * cross vendor migration.
1637 * Writing a zero into the event select MSRs disables them,
1638 * which we perfectly emulate ;-). Any other value should be at least
1639 * reported, some guests depend on them.
1641 case MSR_K7_EVNTSEL0:
1642 case MSR_K7_EVNTSEL1:
1643 case MSR_K7_EVNTSEL2:
1644 case MSR_K7_EVNTSEL3:
1646 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1647 "0x%x data 0x%llx\n", msr, data);
1649 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1650 * so we ignore writes to make it happy.
1652 case MSR_K7_PERFCTR0:
1653 case MSR_K7_PERFCTR1:
1654 case MSR_K7_PERFCTR2:
1655 case MSR_K7_PERFCTR3:
1656 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1657 "0x%x data 0x%llx\n", msr, data);
1659 case MSR_P6_PERFCTR0:
1660 case MSR_P6_PERFCTR1:
1662 case MSR_P6_EVNTSEL0:
1663 case MSR_P6_EVNTSEL1:
1664 if (kvm_pmu_msr(vcpu, msr))
1665 return kvm_pmu_set_msr(vcpu, msr, data);
1667 if (pr || data != 0)
1668 pr_unimpl(vcpu, "disabled perfctr wrmsr: "
1669 "0x%x data 0x%llx\n", msr, data);
1671 case MSR_K7_CLK_CTL:
1673 * Ignore all writes to this no longer documented MSR.
1674 * Writes are only relevant for old K7 processors,
1675 * all pre-dating SVM, but a recommended workaround from
1676 * AMD for these chips. It is possible to speicify the
1677 * affected processor models on the command line, hence
1678 * the need to ignore the workaround.
1681 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1682 if (kvm_hv_msr_partition_wide(msr)) {
1684 mutex_lock(&vcpu->kvm->lock);
1685 r = set_msr_hyperv_pw(vcpu, msr, data);
1686 mutex_unlock(&vcpu->kvm->lock);
1689 return set_msr_hyperv(vcpu, msr, data);
1691 case MSR_IA32_BBL_CR_CTL3:
1692 /* Drop writes to this legacy MSR -- see rdmsr
1693 * counterpart for further detail.
1695 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1697 case MSR_AMD64_OSVW_ID_LENGTH:
1698 if (!guest_cpuid_has_osvw(vcpu))
1700 vcpu->arch.osvw.length = data;
1702 case MSR_AMD64_OSVW_STATUS:
1703 if (!guest_cpuid_has_osvw(vcpu))
1705 vcpu->arch.osvw.status = data;
1708 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1709 return xen_hvm_config(vcpu, data);
1710 if (kvm_pmu_msr(vcpu, msr))
1711 return kvm_pmu_set_msr(vcpu, msr, data);
1713 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1717 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1724 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1728 * Reads an msr value (of 'msr_index') into 'pdata'.
1729 * Returns 0 on success, non-0 otherwise.
1730 * Assumes vcpu_load() was already called.
1732 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1734 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1737 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1739 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1741 if (!msr_mtrr_valid(msr))
1744 if (msr == MSR_MTRRdefType)
1745 *pdata = vcpu->arch.mtrr_state.def_type +
1746 (vcpu->arch.mtrr_state.enabled << 10);
1747 else if (msr == MSR_MTRRfix64K_00000)
1749 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1750 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1751 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1752 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1753 else if (msr == MSR_IA32_CR_PAT)
1754 *pdata = vcpu->arch.pat;
1755 else { /* Variable MTRRs */
1756 int idx, is_mtrr_mask;
1759 idx = (msr - 0x200) / 2;
1760 is_mtrr_mask = msr - 0x200 - 2 * idx;
1763 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1766 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1773 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1776 u64 mcg_cap = vcpu->arch.mcg_cap;
1777 unsigned bank_num = mcg_cap & 0xff;
1780 case MSR_IA32_P5_MC_ADDR:
1781 case MSR_IA32_P5_MC_TYPE:
1784 case MSR_IA32_MCG_CAP:
1785 data = vcpu->arch.mcg_cap;
1787 case MSR_IA32_MCG_CTL:
1788 if (!(mcg_cap & MCG_CTL_P))
1790 data = vcpu->arch.mcg_ctl;
1792 case MSR_IA32_MCG_STATUS:
1793 data = vcpu->arch.mcg_status;
1796 if (msr >= MSR_IA32_MC0_CTL &&
1797 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1798 u32 offset = msr - MSR_IA32_MC0_CTL;
1799 data = vcpu->arch.mce_banks[offset];
1808 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1811 struct kvm *kvm = vcpu->kvm;
1814 case HV_X64_MSR_GUEST_OS_ID:
1815 data = kvm->arch.hv_guest_os_id;
1817 case HV_X64_MSR_HYPERCALL:
1818 data = kvm->arch.hv_hypercall;
1821 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1829 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1834 case HV_X64_MSR_VP_INDEX: {
1837 kvm_for_each_vcpu(r, v, vcpu->kvm)
1842 case HV_X64_MSR_EOI:
1843 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1844 case HV_X64_MSR_ICR:
1845 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1846 case HV_X64_MSR_TPR:
1847 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1848 case HV_X64_MSR_APIC_ASSIST_PAGE:
1849 data = vcpu->arch.hv_vapic;
1852 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1859 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1864 case MSR_IA32_PLATFORM_ID:
1865 case MSR_IA32_EBL_CR_POWERON:
1866 case MSR_IA32_DEBUGCTLMSR:
1867 case MSR_IA32_LASTBRANCHFROMIP:
1868 case MSR_IA32_LASTBRANCHTOIP:
1869 case MSR_IA32_LASTINTFROMIP:
1870 case MSR_IA32_LASTINTTOIP:
1873 case MSR_VM_HSAVE_PA:
1874 case MSR_K7_EVNTSEL0:
1875 case MSR_K7_PERFCTR0:
1876 case MSR_K8_INT_PENDING_MSG:
1877 case MSR_AMD64_NB_CFG:
1878 case MSR_FAM10H_MMIO_CONF_BASE:
1881 case MSR_P6_PERFCTR0:
1882 case MSR_P6_PERFCTR1:
1883 case MSR_P6_EVNTSEL0:
1884 case MSR_P6_EVNTSEL1:
1885 if (kvm_pmu_msr(vcpu, msr))
1886 return kvm_pmu_get_msr(vcpu, msr, pdata);
1889 case MSR_IA32_UCODE_REV:
1890 data = 0x100000000ULL;
1893 data = 0x500 | KVM_NR_VAR_MTRR;
1895 case 0x200 ... 0x2ff:
1896 return get_msr_mtrr(vcpu, msr, pdata);
1897 case 0xcd: /* fsb frequency */
1901 * MSR_EBC_FREQUENCY_ID
1902 * Conservative value valid for even the basic CPU models.
1903 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1904 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1905 * and 266MHz for model 3, or 4. Set Core Clock
1906 * Frequency to System Bus Frequency Ratio to 1 (bits
1907 * 31:24) even though these are only valid for CPU
1908 * models > 2, however guests may end up dividing or
1909 * multiplying by zero otherwise.
1911 case MSR_EBC_FREQUENCY_ID:
1914 case MSR_IA32_APICBASE:
1915 data = kvm_get_apic_base(vcpu);
1917 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1918 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1920 case MSR_IA32_TSCDEADLINE:
1921 data = kvm_get_lapic_tscdeadline_msr(vcpu);
1923 case MSR_IA32_MISC_ENABLE:
1924 data = vcpu->arch.ia32_misc_enable_msr;
1926 case MSR_IA32_PERF_STATUS:
1927 /* TSC increment by tick */
1929 /* CPU multiplier */
1930 data |= (((uint64_t)4ULL) << 40);
1933 data = vcpu->arch.efer;
1935 case MSR_KVM_WALL_CLOCK:
1936 case MSR_KVM_WALL_CLOCK_NEW:
1937 data = vcpu->kvm->arch.wall_clock;
1939 case MSR_KVM_SYSTEM_TIME:
1940 case MSR_KVM_SYSTEM_TIME_NEW:
1941 data = vcpu->arch.time;
1943 case MSR_KVM_ASYNC_PF_EN:
1944 data = vcpu->arch.apf.msr_val;
1946 case MSR_KVM_STEAL_TIME:
1947 data = vcpu->arch.st.msr_val;
1949 case MSR_IA32_P5_MC_ADDR:
1950 case MSR_IA32_P5_MC_TYPE:
1951 case MSR_IA32_MCG_CAP:
1952 case MSR_IA32_MCG_CTL:
1953 case MSR_IA32_MCG_STATUS:
1954 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1955 return get_msr_mce(vcpu, msr, pdata);
1956 case MSR_K7_CLK_CTL:
1958 * Provide expected ramp-up count for K7. All other
1959 * are set to zero, indicating minimum divisors for
1962 * This prevents guest kernels on AMD host with CPU
1963 * type 6, model 8 and higher from exploding due to
1964 * the rdmsr failing.
1968 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1969 if (kvm_hv_msr_partition_wide(msr)) {
1971 mutex_lock(&vcpu->kvm->lock);
1972 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1973 mutex_unlock(&vcpu->kvm->lock);
1976 return get_msr_hyperv(vcpu, msr, pdata);
1978 case MSR_IA32_BBL_CR_CTL3:
1979 /* This legacy MSR exists but isn't fully documented in current
1980 * silicon. It is however accessed by winxp in very narrow
1981 * scenarios where it sets bit #19, itself documented as
1982 * a "reserved" bit. Best effort attempt to source coherent
1983 * read data here should the balance of the register be
1984 * interpreted by the guest:
1986 * L2 cache control register 3: 64GB range, 256KB size,
1987 * enabled, latency 0x1, configured
1991 case MSR_AMD64_OSVW_ID_LENGTH:
1992 if (!guest_cpuid_has_osvw(vcpu))
1994 data = vcpu->arch.osvw.length;
1996 case MSR_AMD64_OSVW_STATUS:
1997 if (!guest_cpuid_has_osvw(vcpu))
1999 data = vcpu->arch.osvw.status;
2002 if (kvm_pmu_msr(vcpu, msr))
2003 return kvm_pmu_get_msr(vcpu, msr, pdata);
2005 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2008 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2016 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2019 * Read or write a bunch of msrs. All parameters are kernel addresses.
2021 * @return number of msrs set successfully.
2023 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2024 struct kvm_msr_entry *entries,
2025 int (*do_msr)(struct kvm_vcpu *vcpu,
2026 unsigned index, u64 *data))
2030 idx = srcu_read_lock(&vcpu->kvm->srcu);
2031 for (i = 0; i < msrs->nmsrs; ++i)
2032 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2034 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2040 * Read or write a bunch of msrs. Parameters are user addresses.
2042 * @return number of msrs set successfully.
2044 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2045 int (*do_msr)(struct kvm_vcpu *vcpu,
2046 unsigned index, u64 *data),
2049 struct kvm_msrs msrs;
2050 struct kvm_msr_entry *entries;
2055 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2059 if (msrs.nmsrs >= MAX_IO_MSRS)
2062 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2063 entries = memdup_user(user_msrs->entries, size);
2064 if (IS_ERR(entries)) {
2065 r = PTR_ERR(entries);
2069 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2074 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2085 int kvm_dev_ioctl_check_extension(long ext)
2090 case KVM_CAP_IRQCHIP:
2092 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2093 case KVM_CAP_SET_TSS_ADDR:
2094 case KVM_CAP_EXT_CPUID:
2095 case KVM_CAP_CLOCKSOURCE:
2097 case KVM_CAP_NOP_IO_DELAY:
2098 case KVM_CAP_MP_STATE:
2099 case KVM_CAP_SYNC_MMU:
2100 case KVM_CAP_USER_NMI:
2101 case KVM_CAP_REINJECT_CONTROL:
2102 case KVM_CAP_IRQ_INJECT_STATUS:
2103 case KVM_CAP_ASSIGN_DEV_IRQ:
2105 case KVM_CAP_IOEVENTFD:
2107 case KVM_CAP_PIT_STATE2:
2108 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2109 case KVM_CAP_XEN_HVM:
2110 case KVM_CAP_ADJUST_CLOCK:
2111 case KVM_CAP_VCPU_EVENTS:
2112 case KVM_CAP_HYPERV:
2113 case KVM_CAP_HYPERV_VAPIC:
2114 case KVM_CAP_HYPERV_SPIN:
2115 case KVM_CAP_PCI_SEGMENT:
2116 case KVM_CAP_DEBUGREGS:
2117 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2119 case KVM_CAP_ASYNC_PF:
2120 case KVM_CAP_GET_TSC_KHZ:
2123 case KVM_CAP_COALESCED_MMIO:
2124 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2127 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2129 case KVM_CAP_NR_VCPUS:
2130 r = KVM_SOFT_MAX_VCPUS;
2132 case KVM_CAP_MAX_VCPUS:
2135 case KVM_CAP_NR_MEMSLOTS:
2136 r = KVM_MEMORY_SLOTS;
2138 case KVM_CAP_PV_MMU: /* obsolete */
2142 r = iommu_present(&pci_bus_type);
2145 r = KVM_MAX_MCE_BANKS;
2150 case KVM_CAP_TSC_CONTROL:
2151 r = kvm_has_tsc_control;
2153 case KVM_CAP_TSC_DEADLINE_TIMER:
2154 r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2164 long kvm_arch_dev_ioctl(struct file *filp,
2165 unsigned int ioctl, unsigned long arg)
2167 void __user *argp = (void __user *)arg;
2171 case KVM_GET_MSR_INDEX_LIST: {
2172 struct kvm_msr_list __user *user_msr_list = argp;
2173 struct kvm_msr_list msr_list;
2177 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2180 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2181 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2184 if (n < msr_list.nmsrs)
2187 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2188 num_msrs_to_save * sizeof(u32)))
2190 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2192 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2197 case KVM_GET_SUPPORTED_CPUID: {
2198 struct kvm_cpuid2 __user *cpuid_arg = argp;
2199 struct kvm_cpuid2 cpuid;
2202 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2204 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2205 cpuid_arg->entries);
2210 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2215 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2218 mce_cap = KVM_MCE_CAP_SUPPORTED;
2220 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2232 static void wbinvd_ipi(void *garbage)
2237 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2239 return vcpu->kvm->arch.iommu_domain &&
2240 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2243 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2245 /* Address WBINVD may be executed by guest */
2246 if (need_emulate_wbinvd(vcpu)) {
2247 if (kvm_x86_ops->has_wbinvd_exit())
2248 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2249 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2250 smp_call_function_single(vcpu->cpu,
2251 wbinvd_ipi, NULL, 1);
2254 kvm_x86_ops->vcpu_load(vcpu, cpu);
2256 /* Apply any externally detected TSC adjustments (due to suspend) */
2257 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2258 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2259 vcpu->arch.tsc_offset_adjustment = 0;
2260 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
2263 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2264 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2265 native_read_tsc() - vcpu->arch.last_host_tsc;
2267 mark_tsc_unstable("KVM discovered backwards TSC");
2268 if (check_tsc_unstable()) {
2269 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2270 vcpu->arch.last_guest_tsc);
2271 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2272 vcpu->arch.tsc_catchup = 1;
2274 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2275 if (vcpu->cpu != cpu)
2276 kvm_migrate_timers(vcpu);
2280 accumulate_steal_time(vcpu);
2281 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2284 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2286 kvm_x86_ops->vcpu_put(vcpu);
2287 kvm_put_guest_fpu(vcpu);
2288 vcpu->arch.last_host_tsc = native_read_tsc();
2291 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2292 struct kvm_lapic_state *s)
2294 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2299 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2300 struct kvm_lapic_state *s)
2302 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2303 kvm_apic_post_state_restore(vcpu);
2304 update_cr8_intercept(vcpu);
2309 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2310 struct kvm_interrupt *irq)
2312 if (irq->irq < 0 || irq->irq >= 256)
2314 if (irqchip_in_kernel(vcpu->kvm))
2317 kvm_queue_interrupt(vcpu, irq->irq, false);
2318 kvm_make_request(KVM_REQ_EVENT, vcpu);
2323 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2325 kvm_inject_nmi(vcpu);
2330 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2331 struct kvm_tpr_access_ctl *tac)
2335 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2339 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2343 unsigned bank_num = mcg_cap & 0xff, bank;
2346 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2348 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2351 vcpu->arch.mcg_cap = mcg_cap;
2352 /* Init IA32_MCG_CTL to all 1s */
2353 if (mcg_cap & MCG_CTL_P)
2354 vcpu->arch.mcg_ctl = ~(u64)0;
2355 /* Init IA32_MCi_CTL to all 1s */
2356 for (bank = 0; bank < bank_num; bank++)
2357 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2362 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2363 struct kvm_x86_mce *mce)
2365 u64 mcg_cap = vcpu->arch.mcg_cap;
2366 unsigned bank_num = mcg_cap & 0xff;
2367 u64 *banks = vcpu->arch.mce_banks;
2369 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2372 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2373 * reporting is disabled
2375 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2376 vcpu->arch.mcg_ctl != ~(u64)0)
2378 banks += 4 * mce->bank;
2380 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2381 * reporting is disabled for the bank
2383 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2385 if (mce->status & MCI_STATUS_UC) {
2386 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2387 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2388 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2391 if (banks[1] & MCI_STATUS_VAL)
2392 mce->status |= MCI_STATUS_OVER;
2393 banks[2] = mce->addr;
2394 banks[3] = mce->misc;
2395 vcpu->arch.mcg_status = mce->mcg_status;
2396 banks[1] = mce->status;
2397 kvm_queue_exception(vcpu, MC_VECTOR);
2398 } else if (!(banks[1] & MCI_STATUS_VAL)
2399 || !(banks[1] & MCI_STATUS_UC)) {
2400 if (banks[1] & MCI_STATUS_VAL)
2401 mce->status |= MCI_STATUS_OVER;
2402 banks[2] = mce->addr;
2403 banks[3] = mce->misc;
2404 banks[1] = mce->status;
2406 banks[1] |= MCI_STATUS_OVER;
2410 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2411 struct kvm_vcpu_events *events)
2414 events->exception.injected =
2415 vcpu->arch.exception.pending &&
2416 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2417 events->exception.nr = vcpu->arch.exception.nr;
2418 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2419 events->exception.pad = 0;
2420 events->exception.error_code = vcpu->arch.exception.error_code;
2422 events->interrupt.injected =
2423 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2424 events->interrupt.nr = vcpu->arch.interrupt.nr;
2425 events->interrupt.soft = 0;
2426 events->interrupt.shadow =
2427 kvm_x86_ops->get_interrupt_shadow(vcpu,
2428 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2430 events->nmi.injected = vcpu->arch.nmi_injected;
2431 events->nmi.pending = vcpu->arch.nmi_pending != 0;
2432 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2433 events->nmi.pad = 0;
2435 events->sipi_vector = vcpu->arch.sipi_vector;
2437 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2438 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2439 | KVM_VCPUEVENT_VALID_SHADOW);
2440 memset(&events->reserved, 0, sizeof(events->reserved));
2443 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2444 struct kvm_vcpu_events *events)
2446 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2447 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2448 | KVM_VCPUEVENT_VALID_SHADOW))
2452 vcpu->arch.exception.pending = events->exception.injected;
2453 vcpu->arch.exception.nr = events->exception.nr;
2454 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2455 vcpu->arch.exception.error_code = events->exception.error_code;
2457 vcpu->arch.interrupt.pending = events->interrupt.injected;
2458 vcpu->arch.interrupt.nr = events->interrupt.nr;
2459 vcpu->arch.interrupt.soft = events->interrupt.soft;
2460 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2461 kvm_x86_ops->set_interrupt_shadow(vcpu,
2462 events->interrupt.shadow);
2464 vcpu->arch.nmi_injected = events->nmi.injected;
2465 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2466 vcpu->arch.nmi_pending = events->nmi.pending;
2467 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2469 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2470 vcpu->arch.sipi_vector = events->sipi_vector;
2472 kvm_make_request(KVM_REQ_EVENT, vcpu);
2477 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2478 struct kvm_debugregs *dbgregs)
2480 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2481 dbgregs->dr6 = vcpu->arch.dr6;
2482 dbgregs->dr7 = vcpu->arch.dr7;
2484 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2487 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2488 struct kvm_debugregs *dbgregs)
2493 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2494 vcpu->arch.dr6 = dbgregs->dr6;
2495 vcpu->arch.dr7 = dbgregs->dr7;
2500 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2501 struct kvm_xsave *guest_xsave)
2504 memcpy(guest_xsave->region,
2505 &vcpu->arch.guest_fpu.state->xsave,
2508 memcpy(guest_xsave->region,
2509 &vcpu->arch.guest_fpu.state->fxsave,
2510 sizeof(struct i387_fxsave_struct));
2511 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2516 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2517 struct kvm_xsave *guest_xsave)
2520 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2523 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2524 guest_xsave->region, xstate_size);
2526 if (xstate_bv & ~XSTATE_FPSSE)
2528 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2529 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2534 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2535 struct kvm_xcrs *guest_xcrs)
2537 if (!cpu_has_xsave) {
2538 guest_xcrs->nr_xcrs = 0;
2542 guest_xcrs->nr_xcrs = 1;
2543 guest_xcrs->flags = 0;
2544 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2545 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2548 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2549 struct kvm_xcrs *guest_xcrs)
2556 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2559 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2560 /* Only support XCR0 currently */
2561 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2562 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2563 guest_xcrs->xcrs[0].value);
2571 long kvm_arch_vcpu_ioctl(struct file *filp,
2572 unsigned int ioctl, unsigned long arg)
2574 struct kvm_vcpu *vcpu = filp->private_data;
2575 void __user *argp = (void __user *)arg;
2578 struct kvm_lapic_state *lapic;
2579 struct kvm_xsave *xsave;
2580 struct kvm_xcrs *xcrs;
2586 case KVM_GET_LAPIC: {
2588 if (!vcpu->arch.apic)
2590 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2595 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
2599 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
2604 case KVM_SET_LAPIC: {
2606 if (!vcpu->arch.apic)
2608 u.lapic = memdup_user(argp, sizeof(*u.lapic));
2609 if (IS_ERR(u.lapic)) {
2610 r = PTR_ERR(u.lapic);
2614 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
2620 case KVM_INTERRUPT: {
2621 struct kvm_interrupt irq;
2624 if (copy_from_user(&irq, argp, sizeof irq))
2626 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2633 r = kvm_vcpu_ioctl_nmi(vcpu);
2639 case KVM_SET_CPUID: {
2640 struct kvm_cpuid __user *cpuid_arg = argp;
2641 struct kvm_cpuid cpuid;
2644 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2646 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2651 case KVM_SET_CPUID2: {
2652 struct kvm_cpuid2 __user *cpuid_arg = argp;
2653 struct kvm_cpuid2 cpuid;
2656 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2658 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2659 cpuid_arg->entries);
2664 case KVM_GET_CPUID2: {
2665 struct kvm_cpuid2 __user *cpuid_arg = argp;
2666 struct kvm_cpuid2 cpuid;
2669 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2671 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2672 cpuid_arg->entries);
2676 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2682 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2685 r = msr_io(vcpu, argp, do_set_msr, 0);
2687 case KVM_TPR_ACCESS_REPORTING: {
2688 struct kvm_tpr_access_ctl tac;
2691 if (copy_from_user(&tac, argp, sizeof tac))
2693 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2697 if (copy_to_user(argp, &tac, sizeof tac))
2702 case KVM_SET_VAPIC_ADDR: {
2703 struct kvm_vapic_addr va;
2706 if (!irqchip_in_kernel(vcpu->kvm))
2709 if (copy_from_user(&va, argp, sizeof va))
2712 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2715 case KVM_X86_SETUP_MCE: {
2719 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2721 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2724 case KVM_X86_SET_MCE: {
2725 struct kvm_x86_mce mce;
2728 if (copy_from_user(&mce, argp, sizeof mce))
2730 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2733 case KVM_GET_VCPU_EVENTS: {
2734 struct kvm_vcpu_events events;
2736 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2739 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2744 case KVM_SET_VCPU_EVENTS: {
2745 struct kvm_vcpu_events events;
2748 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2751 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2754 case KVM_GET_DEBUGREGS: {
2755 struct kvm_debugregs dbgregs;
2757 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2760 if (copy_to_user(argp, &dbgregs,
2761 sizeof(struct kvm_debugregs)))
2766 case KVM_SET_DEBUGREGS: {
2767 struct kvm_debugregs dbgregs;
2770 if (copy_from_user(&dbgregs, argp,
2771 sizeof(struct kvm_debugregs)))
2774 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2777 case KVM_GET_XSAVE: {
2778 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2783 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
2786 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
2791 case KVM_SET_XSAVE: {
2792 u.xsave = memdup_user(argp, sizeof(*u.xsave));
2793 if (IS_ERR(u.xsave)) {
2794 r = PTR_ERR(u.xsave);
2798 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
2801 case KVM_GET_XCRS: {
2802 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2807 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
2810 if (copy_to_user(argp, u.xcrs,
2811 sizeof(struct kvm_xcrs)))
2816 case KVM_SET_XCRS: {
2817 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
2818 if (IS_ERR(u.xcrs)) {
2819 r = PTR_ERR(u.xcrs);
2823 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
2826 case KVM_SET_TSC_KHZ: {
2830 user_tsc_khz = (u32)arg;
2832 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
2835 if (user_tsc_khz == 0)
2836 user_tsc_khz = tsc_khz;
2838 kvm_set_tsc_khz(vcpu, user_tsc_khz);
2843 case KVM_GET_TSC_KHZ: {
2844 r = vcpu->arch.virtual_tsc_khz;
2855 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
2857 return VM_FAULT_SIGBUS;
2860 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2864 if (addr > (unsigned int)(-3 * PAGE_SIZE))
2866 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2870 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2873 kvm->arch.ept_identity_map_addr = ident_addr;
2877 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2878 u32 kvm_nr_mmu_pages)
2880 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2883 mutex_lock(&kvm->slots_lock);
2884 spin_lock(&kvm->mmu_lock);
2886 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2887 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2889 spin_unlock(&kvm->mmu_lock);
2890 mutex_unlock(&kvm->slots_lock);
2894 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2896 return kvm->arch.n_max_mmu_pages;
2899 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2904 switch (chip->chip_id) {
2905 case KVM_IRQCHIP_PIC_MASTER:
2906 memcpy(&chip->chip.pic,
2907 &pic_irqchip(kvm)->pics[0],
2908 sizeof(struct kvm_pic_state));
2910 case KVM_IRQCHIP_PIC_SLAVE:
2911 memcpy(&chip->chip.pic,
2912 &pic_irqchip(kvm)->pics[1],
2913 sizeof(struct kvm_pic_state));
2915 case KVM_IRQCHIP_IOAPIC:
2916 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2925 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2930 switch (chip->chip_id) {
2931 case KVM_IRQCHIP_PIC_MASTER:
2932 spin_lock(&pic_irqchip(kvm)->lock);
2933 memcpy(&pic_irqchip(kvm)->pics[0],
2935 sizeof(struct kvm_pic_state));
2936 spin_unlock(&pic_irqchip(kvm)->lock);
2938 case KVM_IRQCHIP_PIC_SLAVE:
2939 spin_lock(&pic_irqchip(kvm)->lock);
2940 memcpy(&pic_irqchip(kvm)->pics[1],
2942 sizeof(struct kvm_pic_state));
2943 spin_unlock(&pic_irqchip(kvm)->lock);
2945 case KVM_IRQCHIP_IOAPIC:
2946 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2952 kvm_pic_update_irq(pic_irqchip(kvm));
2956 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2960 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2961 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2962 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2966 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2970 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2971 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2972 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2973 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2977 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2981 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2982 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2983 sizeof(ps->channels));
2984 ps->flags = kvm->arch.vpit->pit_state.flags;
2985 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2986 memset(&ps->reserved, 0, sizeof(ps->reserved));
2990 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2992 int r = 0, start = 0;
2993 u32 prev_legacy, cur_legacy;
2994 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2995 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2996 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2997 if (!prev_legacy && cur_legacy)
2999 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3000 sizeof(kvm->arch.vpit->pit_state.channels));
3001 kvm->arch.vpit->pit_state.flags = ps->flags;
3002 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3003 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3007 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3008 struct kvm_reinject_control *control)
3010 if (!kvm->arch.vpit)
3012 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3013 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3014 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3019 * write_protect_slot - write protect a slot for dirty logging
3020 * @kvm: the kvm instance
3021 * @memslot: the slot we protect
3022 * @dirty_bitmap: the bitmap indicating which pages are dirty
3023 * @nr_dirty_pages: the number of dirty pages
3025 * We have two ways to find all sptes to protect:
3026 * 1. Use kvm_mmu_slot_remove_write_access() which walks all shadow pages and
3027 * checks ones that have a spte mapping a page in the slot.
3028 * 2. Use kvm_mmu_rmap_write_protect() for each gfn found in the bitmap.
3030 * Generally speaking, if there are not so many dirty pages compared to the
3031 * number of shadow pages, we should use the latter.
3033 * Note that letting others write into a page marked dirty in the old bitmap
3034 * by using the remaining tlb entry is not a problem. That page will become
3035 * write protected again when we flush the tlb and then be reported dirty to
3036 * the user space by copying the old bitmap.
3038 static void write_protect_slot(struct kvm *kvm,
3039 struct kvm_memory_slot *memslot,
3040 unsigned long *dirty_bitmap,
3041 unsigned long nr_dirty_pages)
3043 /* Not many dirty pages compared to # of shadow pages. */
3044 if (nr_dirty_pages < kvm->arch.n_used_mmu_pages) {
3045 unsigned long gfn_offset;
3047 for_each_set_bit(gfn_offset, dirty_bitmap, memslot->npages) {
3048 unsigned long gfn = memslot->base_gfn + gfn_offset;
3050 spin_lock(&kvm->mmu_lock);
3051 kvm_mmu_rmap_write_protect(kvm, gfn, memslot);
3052 spin_unlock(&kvm->mmu_lock);
3054 kvm_flush_remote_tlbs(kvm);
3056 spin_lock(&kvm->mmu_lock);
3057 kvm_mmu_slot_remove_write_access(kvm, memslot->id);
3058 spin_unlock(&kvm->mmu_lock);
3063 * Get (and clear) the dirty memory log for a memory slot.
3065 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3066 struct kvm_dirty_log *log)
3069 struct kvm_memory_slot *memslot;
3070 unsigned long n, nr_dirty_pages;
3072 mutex_lock(&kvm->slots_lock);
3075 if (log->slot >= KVM_MEMORY_SLOTS)
3078 memslot = id_to_memslot(kvm->memslots, log->slot);
3080 if (!memslot->dirty_bitmap)
3083 n = kvm_dirty_bitmap_bytes(memslot);
3084 nr_dirty_pages = memslot->nr_dirty_pages;
3086 /* If nothing is dirty, don't bother messing with page tables. */
3087 if (nr_dirty_pages) {
3088 struct kvm_memslots *slots, *old_slots;
3089 unsigned long *dirty_bitmap, *dirty_bitmap_head;
3091 dirty_bitmap = memslot->dirty_bitmap;
3092 dirty_bitmap_head = memslot->dirty_bitmap_head;
3093 if (dirty_bitmap == dirty_bitmap_head)
3094 dirty_bitmap_head += n / sizeof(long);
3095 memset(dirty_bitmap_head, 0, n);
3098 slots = kmemdup(kvm->memslots, sizeof(*kvm->memslots), GFP_KERNEL);
3102 memslot = id_to_memslot(slots, log->slot);
3103 memslot->nr_dirty_pages = 0;
3104 memslot->dirty_bitmap = dirty_bitmap_head;
3105 update_memslots(slots, NULL);
3107 old_slots = kvm->memslots;
3108 rcu_assign_pointer(kvm->memslots, slots);
3109 synchronize_srcu_expedited(&kvm->srcu);
3112 write_protect_slot(kvm, memslot, dirty_bitmap, nr_dirty_pages);
3115 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
3119 if (clear_user(log->dirty_bitmap, n))
3125 mutex_unlock(&kvm->slots_lock);
3129 long kvm_arch_vm_ioctl(struct file *filp,
3130 unsigned int ioctl, unsigned long arg)
3132 struct kvm *kvm = filp->private_data;
3133 void __user *argp = (void __user *)arg;
3136 * This union makes it completely explicit to gcc-3.x
3137 * that these two variables' stack usage should be
3138 * combined, not added together.
3141 struct kvm_pit_state ps;
3142 struct kvm_pit_state2 ps2;
3143 struct kvm_pit_config pit_config;
3147 case KVM_SET_TSS_ADDR:
3148 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3152 case KVM_SET_IDENTITY_MAP_ADDR: {
3156 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3158 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3163 case KVM_SET_NR_MMU_PAGES:
3164 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3168 case KVM_GET_NR_MMU_PAGES:
3169 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3171 case KVM_CREATE_IRQCHIP: {
3172 struct kvm_pic *vpic;
3174 mutex_lock(&kvm->lock);
3177 goto create_irqchip_unlock;
3179 vpic = kvm_create_pic(kvm);
3181 r = kvm_ioapic_init(kvm);
3183 mutex_lock(&kvm->slots_lock);
3184 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3186 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3188 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3190 mutex_unlock(&kvm->slots_lock);
3192 goto create_irqchip_unlock;
3195 goto create_irqchip_unlock;
3197 kvm->arch.vpic = vpic;
3199 r = kvm_setup_default_irq_routing(kvm);
3201 mutex_lock(&kvm->slots_lock);
3202 mutex_lock(&kvm->irq_lock);
3203 kvm_ioapic_destroy(kvm);
3204 kvm_destroy_pic(kvm);
3205 mutex_unlock(&kvm->irq_lock);
3206 mutex_unlock(&kvm->slots_lock);
3208 create_irqchip_unlock:
3209 mutex_unlock(&kvm->lock);
3212 case KVM_CREATE_PIT:
3213 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3215 case KVM_CREATE_PIT2:
3217 if (copy_from_user(&u.pit_config, argp,
3218 sizeof(struct kvm_pit_config)))
3221 mutex_lock(&kvm->slots_lock);
3224 goto create_pit_unlock;
3226 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3230 mutex_unlock(&kvm->slots_lock);
3232 case KVM_IRQ_LINE_STATUS:
3233 case KVM_IRQ_LINE: {
3234 struct kvm_irq_level irq_event;
3237 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3240 if (irqchip_in_kernel(kvm)) {
3242 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3243 irq_event.irq, irq_event.level);
3244 if (ioctl == KVM_IRQ_LINE_STATUS) {
3246 irq_event.status = status;
3247 if (copy_to_user(argp, &irq_event,
3255 case KVM_GET_IRQCHIP: {
3256 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3257 struct kvm_irqchip *chip;
3259 chip = memdup_user(argp, sizeof(*chip));
3266 if (!irqchip_in_kernel(kvm))
3267 goto get_irqchip_out;
3268 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3270 goto get_irqchip_out;
3272 if (copy_to_user(argp, chip, sizeof *chip))
3273 goto get_irqchip_out;
3281 case KVM_SET_IRQCHIP: {
3282 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3283 struct kvm_irqchip *chip;
3285 chip = memdup_user(argp, sizeof(*chip));
3292 if (!irqchip_in_kernel(kvm))
3293 goto set_irqchip_out;
3294 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3296 goto set_irqchip_out;
3306 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3309 if (!kvm->arch.vpit)
3311 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3315 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3322 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3325 if (!kvm->arch.vpit)
3327 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3333 case KVM_GET_PIT2: {
3335 if (!kvm->arch.vpit)
3337 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3341 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3346 case KVM_SET_PIT2: {
3348 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3351 if (!kvm->arch.vpit)
3353 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3359 case KVM_REINJECT_CONTROL: {
3360 struct kvm_reinject_control control;
3362 if (copy_from_user(&control, argp, sizeof(control)))
3364 r = kvm_vm_ioctl_reinject(kvm, &control);
3370 case KVM_XEN_HVM_CONFIG: {
3372 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3373 sizeof(struct kvm_xen_hvm_config)))
3376 if (kvm->arch.xen_hvm_config.flags)
3381 case KVM_SET_CLOCK: {
3382 struct kvm_clock_data user_ns;
3387 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3395 local_irq_disable();
3396 now_ns = get_kernel_ns();
3397 delta = user_ns.clock - now_ns;
3399 kvm->arch.kvmclock_offset = delta;
3402 case KVM_GET_CLOCK: {
3403 struct kvm_clock_data user_ns;
3406 local_irq_disable();
3407 now_ns = get_kernel_ns();
3408 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3411 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3414 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3427 static void kvm_init_msr_list(void)
3432 /* skip the first msrs in the list. KVM-specific */
3433 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3434 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3437 msrs_to_save[j] = msrs_to_save[i];
3440 num_msrs_to_save = j;
3443 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3451 if (!(vcpu->arch.apic &&
3452 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3453 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3464 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3471 if (!(vcpu->arch.apic &&
3472 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3473 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3475 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3485 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3486 struct kvm_segment *var, int seg)
3488 kvm_x86_ops->set_segment(vcpu, var, seg);
3491 void kvm_get_segment(struct kvm_vcpu *vcpu,
3492 struct kvm_segment *var, int seg)
3494 kvm_x86_ops->get_segment(vcpu, var, seg);
3497 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3500 struct x86_exception exception;
3502 BUG_ON(!mmu_is_nested(vcpu));
3504 /* NPT walks are always user-walks */
3505 access |= PFERR_USER_MASK;
3506 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3511 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3512 struct x86_exception *exception)
3514 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3515 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3518 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3519 struct x86_exception *exception)
3521 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3522 access |= PFERR_FETCH_MASK;
3523 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3526 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3527 struct x86_exception *exception)
3529 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3530 access |= PFERR_WRITE_MASK;
3531 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3534 /* uses this to access any guest's mapped memory without checking CPL */
3535 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3536 struct x86_exception *exception)
3538 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3541 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3542 struct kvm_vcpu *vcpu, u32 access,
3543 struct x86_exception *exception)
3546 int r = X86EMUL_CONTINUE;
3549 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3551 unsigned offset = addr & (PAGE_SIZE-1);
3552 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3555 if (gpa == UNMAPPED_GVA)
3556 return X86EMUL_PROPAGATE_FAULT;
3557 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3559 r = X86EMUL_IO_NEEDED;
3571 /* used for instruction fetching */
3572 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3573 gva_t addr, void *val, unsigned int bytes,
3574 struct x86_exception *exception)
3576 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3577 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3579 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3580 access | PFERR_FETCH_MASK,
3584 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3585 gva_t addr, void *val, unsigned int bytes,
3586 struct x86_exception *exception)
3588 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3589 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3591 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3594 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
3596 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3597 gva_t addr, void *val, unsigned int bytes,
3598 struct x86_exception *exception)
3600 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3601 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3604 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3605 gva_t addr, void *val,
3607 struct x86_exception *exception)
3609 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3611 int r = X86EMUL_CONTINUE;
3614 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3617 unsigned offset = addr & (PAGE_SIZE-1);
3618 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3621 if (gpa == UNMAPPED_GVA)
3622 return X86EMUL_PROPAGATE_FAULT;
3623 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3625 r = X86EMUL_IO_NEEDED;
3636 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
3638 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
3639 gpa_t *gpa, struct x86_exception *exception,
3642 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3644 if (vcpu_match_mmio_gva(vcpu, gva) &&
3645 check_write_user_access(vcpu, write, access,
3646 vcpu->arch.access)) {
3647 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
3648 (gva & (PAGE_SIZE - 1));
3649 trace_vcpu_match_mmio(gva, *gpa, write, false);
3654 access |= PFERR_WRITE_MASK;
3656 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3658 if (*gpa == UNMAPPED_GVA)
3661 /* For APIC access vmexit */
3662 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3665 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
3666 trace_vcpu_match_mmio(gva, *gpa, write, true);
3673 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3674 const void *val, int bytes)
3678 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3681 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
3685 struct read_write_emulator_ops {
3686 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
3688 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
3689 void *val, int bytes);
3690 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
3691 int bytes, void *val);
3692 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
3693 void *val, int bytes);
3697 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
3699 if (vcpu->mmio_read_completed) {
3700 memcpy(val, vcpu->mmio_data, bytes);
3701 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3702 vcpu->mmio_phys_addr, *(u64 *)val);
3703 vcpu->mmio_read_completed = 0;
3710 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
3711 void *val, int bytes)
3713 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
3716 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
3717 void *val, int bytes)
3719 return emulator_write_phys(vcpu, gpa, val, bytes);
3722 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
3724 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3725 return vcpu_mmio_write(vcpu, gpa, bytes, val);
3728 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
3729 void *val, int bytes)
3731 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3732 return X86EMUL_IO_NEEDED;
3735 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
3736 void *val, int bytes)
3738 memcpy(vcpu->mmio_data, val, bytes);
3739 memcpy(vcpu->run->mmio.data, vcpu->mmio_data, 8);
3740 return X86EMUL_CONTINUE;
3743 static struct read_write_emulator_ops read_emultor = {
3744 .read_write_prepare = read_prepare,
3745 .read_write_emulate = read_emulate,
3746 .read_write_mmio = vcpu_mmio_read,
3747 .read_write_exit_mmio = read_exit_mmio,
3750 static struct read_write_emulator_ops write_emultor = {
3751 .read_write_emulate = write_emulate,
3752 .read_write_mmio = write_mmio,
3753 .read_write_exit_mmio = write_exit_mmio,
3757 static int emulator_read_write_onepage(unsigned long addr, void *val,
3759 struct x86_exception *exception,
3760 struct kvm_vcpu *vcpu,
3761 struct read_write_emulator_ops *ops)
3765 bool write = ops->write;
3767 if (ops->read_write_prepare &&
3768 ops->read_write_prepare(vcpu, val, bytes))
3769 return X86EMUL_CONTINUE;
3771 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
3774 return X86EMUL_PROPAGATE_FAULT;
3776 /* For APIC access vmexit */
3780 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
3781 return X86EMUL_CONTINUE;
3785 * Is this MMIO handled locally?
3787 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
3788 if (handled == bytes)
3789 return X86EMUL_CONTINUE;
3795 vcpu->mmio_needed = 1;
3796 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3797 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3798 vcpu->mmio_size = bytes;
3799 vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
3800 vcpu->run->mmio.is_write = vcpu->mmio_is_write = write;
3801 vcpu->mmio_index = 0;
3803 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
3806 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
3807 void *val, unsigned int bytes,
3808 struct x86_exception *exception,
3809 struct read_write_emulator_ops *ops)
3811 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3813 /* Crossing a page boundary? */
3814 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3817 now = -addr & ~PAGE_MASK;
3818 rc = emulator_read_write_onepage(addr, val, now, exception,
3821 if (rc != X86EMUL_CONTINUE)
3828 return emulator_read_write_onepage(addr, val, bytes, exception,
3832 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
3836 struct x86_exception *exception)
3838 return emulator_read_write(ctxt, addr, val, bytes,
3839 exception, &read_emultor);
3842 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
3846 struct x86_exception *exception)
3848 return emulator_read_write(ctxt, addr, (void *)val, bytes,
3849 exception, &write_emultor);
3852 #define CMPXCHG_TYPE(t, ptr, old, new) \
3853 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3855 #ifdef CONFIG_X86_64
3856 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3858 # define CMPXCHG64(ptr, old, new) \
3859 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
3862 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
3867 struct x86_exception *exception)
3869 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3875 /* guests cmpxchg8b have to be emulated atomically */
3876 if (bytes > 8 || (bytes & (bytes - 1)))
3879 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3881 if (gpa == UNMAPPED_GVA ||
3882 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3885 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3888 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3889 if (is_error_page(page)) {
3890 kvm_release_page_clean(page);
3894 kaddr = kmap_atomic(page, KM_USER0);
3895 kaddr += offset_in_page(gpa);
3898 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
3901 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
3904 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
3907 exchanged = CMPXCHG64(kaddr, old, new);
3912 kunmap_atomic(kaddr, KM_USER0);
3913 kvm_release_page_dirty(page);
3916 return X86EMUL_CMPXCHG_FAILED;
3918 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
3920 return X86EMUL_CONTINUE;
3923 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3925 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
3928 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3930 /* TODO: String I/O for in kernel device */
3933 if (vcpu->arch.pio.in)
3934 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3935 vcpu->arch.pio.size, pd);
3937 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3938 vcpu->arch.pio.port, vcpu->arch.pio.size,
3943 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
3944 unsigned short port, void *val,
3945 unsigned int count, bool in)
3947 trace_kvm_pio(!in, port, size, count);
3949 vcpu->arch.pio.port = port;
3950 vcpu->arch.pio.in = in;
3951 vcpu->arch.pio.count = count;
3952 vcpu->arch.pio.size = size;
3954 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3955 vcpu->arch.pio.count = 0;
3959 vcpu->run->exit_reason = KVM_EXIT_IO;
3960 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3961 vcpu->run->io.size = size;
3962 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3963 vcpu->run->io.count = count;
3964 vcpu->run->io.port = port;
3969 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
3970 int size, unsigned short port, void *val,
3973 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3976 if (vcpu->arch.pio.count)
3979 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
3982 memcpy(val, vcpu->arch.pio_data, size * count);
3983 vcpu->arch.pio.count = 0;
3990 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
3991 int size, unsigned short port,
3992 const void *val, unsigned int count)
3994 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3996 memcpy(vcpu->arch.pio_data, val, size * count);
3997 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4000 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4002 return kvm_x86_ops->get_segment_base(vcpu, seg);
4005 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4007 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4010 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4012 if (!need_emulate_wbinvd(vcpu))
4013 return X86EMUL_CONTINUE;
4015 if (kvm_x86_ops->has_wbinvd_exit()) {
4016 int cpu = get_cpu();
4018 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4019 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4020 wbinvd_ipi, NULL, 1);
4022 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4025 return X86EMUL_CONTINUE;
4027 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4029 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4031 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4034 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4036 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4039 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4042 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4045 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4047 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4050 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4052 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4053 unsigned long value;
4057 value = kvm_read_cr0(vcpu);
4060 value = vcpu->arch.cr2;
4063 value = kvm_read_cr3(vcpu);
4066 value = kvm_read_cr4(vcpu);
4069 value = kvm_get_cr8(vcpu);
4072 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4079 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4081 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4086 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4089 vcpu->arch.cr2 = val;
4092 res = kvm_set_cr3(vcpu, val);
4095 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4098 res = kvm_set_cr8(vcpu, val);
4101 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4108 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4110 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4113 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4115 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4118 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4120 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4123 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4125 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4128 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4130 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4133 static unsigned long emulator_get_cached_segment_base(
4134 struct x86_emulate_ctxt *ctxt, int seg)
4136 return get_segment_base(emul_to_vcpu(ctxt), seg);
4139 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4140 struct desc_struct *desc, u32 *base3,
4143 struct kvm_segment var;
4145 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4146 *selector = var.selector;
4153 set_desc_limit(desc, var.limit);
4154 set_desc_base(desc, (unsigned long)var.base);
4155 #ifdef CONFIG_X86_64
4157 *base3 = var.base >> 32;
4159 desc->type = var.type;
4161 desc->dpl = var.dpl;
4162 desc->p = var.present;
4163 desc->avl = var.avl;
4171 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4172 struct desc_struct *desc, u32 base3,
4175 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4176 struct kvm_segment var;
4178 var.selector = selector;
4179 var.base = get_desc_base(desc);
4180 #ifdef CONFIG_X86_64
4181 var.base |= ((u64)base3) << 32;
4183 var.limit = get_desc_limit(desc);
4185 var.limit = (var.limit << 12) | 0xfff;
4186 var.type = desc->type;
4187 var.present = desc->p;
4188 var.dpl = desc->dpl;
4193 var.avl = desc->avl;
4194 var.present = desc->p;
4195 var.unusable = !var.present;
4198 kvm_set_segment(vcpu, &var, seg);
4202 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4203 u32 msr_index, u64 *pdata)
4205 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4208 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4209 u32 msr_index, u64 data)
4211 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
4214 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4215 u32 pmc, u64 *pdata)
4217 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4220 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4222 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4225 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4228 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4230 * CR0.TS may reference the host fpu state, not the guest fpu state,
4231 * so it may be clear at this point.
4236 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4241 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4242 struct x86_instruction_info *info,
4243 enum x86_intercept_stage stage)
4245 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4248 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4249 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4251 struct kvm_cpuid_entry2 *cpuid = NULL;
4254 cpuid = kvm_find_cpuid_entry(emul_to_vcpu(ctxt),
4270 static struct x86_emulate_ops emulate_ops = {
4271 .read_std = kvm_read_guest_virt_system,
4272 .write_std = kvm_write_guest_virt_system,
4273 .fetch = kvm_fetch_guest_virt,
4274 .read_emulated = emulator_read_emulated,
4275 .write_emulated = emulator_write_emulated,
4276 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4277 .invlpg = emulator_invlpg,
4278 .pio_in_emulated = emulator_pio_in_emulated,
4279 .pio_out_emulated = emulator_pio_out_emulated,
4280 .get_segment = emulator_get_segment,
4281 .set_segment = emulator_set_segment,
4282 .get_cached_segment_base = emulator_get_cached_segment_base,
4283 .get_gdt = emulator_get_gdt,
4284 .get_idt = emulator_get_idt,
4285 .set_gdt = emulator_set_gdt,
4286 .set_idt = emulator_set_idt,
4287 .get_cr = emulator_get_cr,
4288 .set_cr = emulator_set_cr,
4289 .cpl = emulator_get_cpl,
4290 .get_dr = emulator_get_dr,
4291 .set_dr = emulator_set_dr,
4292 .set_msr = emulator_set_msr,
4293 .get_msr = emulator_get_msr,
4294 .read_pmc = emulator_read_pmc,
4295 .halt = emulator_halt,
4296 .wbinvd = emulator_wbinvd,
4297 .fix_hypercall = emulator_fix_hypercall,
4298 .get_fpu = emulator_get_fpu,
4299 .put_fpu = emulator_put_fpu,
4300 .intercept = emulator_intercept,
4301 .get_cpuid = emulator_get_cpuid,
4304 static void cache_all_regs(struct kvm_vcpu *vcpu)
4306 kvm_register_read(vcpu, VCPU_REGS_RAX);
4307 kvm_register_read(vcpu, VCPU_REGS_RSP);
4308 kvm_register_read(vcpu, VCPU_REGS_RIP);
4309 vcpu->arch.regs_dirty = ~0;
4312 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4314 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4316 * an sti; sti; sequence only disable interrupts for the first
4317 * instruction. So, if the last instruction, be it emulated or
4318 * not, left the system with the INT_STI flag enabled, it
4319 * means that the last instruction is an sti. We should not
4320 * leave the flag on in this case. The same goes for mov ss
4322 if (!(int_shadow & mask))
4323 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4326 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4328 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4329 if (ctxt->exception.vector == PF_VECTOR)
4330 kvm_propagate_fault(vcpu, &ctxt->exception);
4331 else if (ctxt->exception.error_code_valid)
4332 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4333 ctxt->exception.error_code);
4335 kvm_queue_exception(vcpu, ctxt->exception.vector);
4338 static void init_decode_cache(struct x86_emulate_ctxt *ctxt,
4339 const unsigned long *regs)
4341 memset(&ctxt->twobyte, 0,
4342 (void *)&ctxt->regs - (void *)&ctxt->twobyte);
4343 memcpy(ctxt->regs, regs, sizeof(ctxt->regs));
4345 ctxt->fetch.start = 0;
4346 ctxt->fetch.end = 0;
4347 ctxt->io_read.pos = 0;
4348 ctxt->io_read.end = 0;
4349 ctxt->mem_read.pos = 0;
4350 ctxt->mem_read.end = 0;
4353 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4355 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4359 * TODO: fix emulate.c to use guest_read/write_register
4360 * instead of direct ->regs accesses, can save hundred cycles
4361 * on Intel for instructions that don't read/change RSP, for
4364 cache_all_regs(vcpu);
4366 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4368 ctxt->eflags = kvm_get_rflags(vcpu);
4369 ctxt->eip = kvm_rip_read(vcpu);
4370 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4371 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4372 cs_l ? X86EMUL_MODE_PROT64 :
4373 cs_db ? X86EMUL_MODE_PROT32 :
4374 X86EMUL_MODE_PROT16;
4375 ctxt->guest_mode = is_guest_mode(vcpu);
4377 init_decode_cache(ctxt, vcpu->arch.regs);
4378 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4381 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4383 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4386 init_emulate_ctxt(vcpu);
4390 ctxt->_eip = ctxt->eip + inc_eip;
4391 ret = emulate_int_real(ctxt, irq);
4393 if (ret != X86EMUL_CONTINUE)
4394 return EMULATE_FAIL;
4396 ctxt->eip = ctxt->_eip;
4397 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4398 kvm_rip_write(vcpu, ctxt->eip);
4399 kvm_set_rflags(vcpu, ctxt->eflags);
4401 if (irq == NMI_VECTOR)
4402 vcpu->arch.nmi_pending = 0;
4404 vcpu->arch.interrupt.pending = false;
4406 return EMULATE_DONE;
4408 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4410 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4412 int r = EMULATE_DONE;
4414 ++vcpu->stat.insn_emulation_fail;
4415 trace_kvm_emulate_insn_failed(vcpu);
4416 if (!is_guest_mode(vcpu)) {
4417 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4418 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4419 vcpu->run->internal.ndata = 0;
4422 kvm_queue_exception(vcpu, UD_VECTOR);
4427 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4435 * if emulation was due to access to shadowed page table
4436 * and it failed try to unshadow page and re-entetr the
4437 * guest to let CPU execute the instruction.
4439 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4442 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4444 if (gpa == UNMAPPED_GVA)
4445 return true; /* let cpu generate fault */
4447 if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4453 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
4454 unsigned long cr2, int emulation_type)
4456 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4457 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
4459 last_retry_eip = vcpu->arch.last_retry_eip;
4460 last_retry_addr = vcpu->arch.last_retry_addr;
4463 * If the emulation is caused by #PF and it is non-page_table
4464 * writing instruction, it means the VM-EXIT is caused by shadow
4465 * page protected, we can zap the shadow page and retry this
4466 * instruction directly.
4468 * Note: if the guest uses a non-page-table modifying instruction
4469 * on the PDE that points to the instruction, then we will unmap
4470 * the instruction and go to an infinite loop. So, we cache the
4471 * last retried eip and the last fault address, if we meet the eip
4472 * and the address again, we can break out of the potential infinite
4475 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
4477 if (!(emulation_type & EMULTYPE_RETRY))
4480 if (x86_page_table_writing_insn(ctxt))
4483 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
4486 vcpu->arch.last_retry_eip = ctxt->eip;
4487 vcpu->arch.last_retry_addr = cr2;
4489 if (!vcpu->arch.mmu.direct_map)
4490 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4492 kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4497 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4504 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4505 bool writeback = true;
4507 kvm_clear_exception_queue(vcpu);
4509 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4510 init_emulate_ctxt(vcpu);
4511 ctxt->interruptibility = 0;
4512 ctxt->have_exception = false;
4513 ctxt->perm_ok = false;
4515 ctxt->only_vendor_specific_insn
4516 = emulation_type & EMULTYPE_TRAP_UD;
4518 r = x86_decode_insn(ctxt, insn, insn_len);
4520 trace_kvm_emulate_insn_start(vcpu);
4521 ++vcpu->stat.insn_emulation;
4522 if (r != EMULATION_OK) {
4523 if (emulation_type & EMULTYPE_TRAP_UD)
4524 return EMULATE_FAIL;
4525 if (reexecute_instruction(vcpu, cr2))
4526 return EMULATE_DONE;
4527 if (emulation_type & EMULTYPE_SKIP)
4528 return EMULATE_FAIL;
4529 return handle_emulation_failure(vcpu);
4533 if (emulation_type & EMULTYPE_SKIP) {
4534 kvm_rip_write(vcpu, ctxt->_eip);
4535 return EMULATE_DONE;
4538 if (retry_instruction(ctxt, cr2, emulation_type))
4539 return EMULATE_DONE;
4541 /* this is needed for vmware backdoor interface to work since it
4542 changes registers values during IO operation */
4543 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4544 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4545 memcpy(ctxt->regs, vcpu->arch.regs, sizeof ctxt->regs);
4549 r = x86_emulate_insn(ctxt);
4551 if (r == EMULATION_INTERCEPTED)
4552 return EMULATE_DONE;
4554 if (r == EMULATION_FAILED) {
4555 if (reexecute_instruction(vcpu, cr2))
4556 return EMULATE_DONE;
4558 return handle_emulation_failure(vcpu);
4561 if (ctxt->have_exception) {
4562 inject_emulated_exception(vcpu);
4564 } else if (vcpu->arch.pio.count) {
4565 if (!vcpu->arch.pio.in)
4566 vcpu->arch.pio.count = 0;
4569 r = EMULATE_DO_MMIO;
4570 } else if (vcpu->mmio_needed) {
4571 if (!vcpu->mmio_is_write)
4573 r = EMULATE_DO_MMIO;
4574 } else if (r == EMULATION_RESTART)
4580 toggle_interruptibility(vcpu, ctxt->interruptibility);
4581 kvm_set_rflags(vcpu, ctxt->eflags);
4582 kvm_make_request(KVM_REQ_EVENT, vcpu);
4583 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4584 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4585 kvm_rip_write(vcpu, ctxt->eip);
4587 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
4591 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4593 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4595 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4596 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4597 size, port, &val, 1);
4598 /* do not return to emulator after return from userspace */
4599 vcpu->arch.pio.count = 0;
4602 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4604 static void tsc_bad(void *info)
4606 __this_cpu_write(cpu_tsc_khz, 0);
4609 static void tsc_khz_changed(void *data)
4611 struct cpufreq_freqs *freq = data;
4612 unsigned long khz = 0;
4616 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4617 khz = cpufreq_quick_get(raw_smp_processor_id());
4620 __this_cpu_write(cpu_tsc_khz, khz);
4623 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4626 struct cpufreq_freqs *freq = data;
4628 struct kvm_vcpu *vcpu;
4629 int i, send_ipi = 0;
4632 * We allow guests to temporarily run on slowing clocks,
4633 * provided we notify them after, or to run on accelerating
4634 * clocks, provided we notify them before. Thus time never
4637 * However, we have a problem. We can't atomically update
4638 * the frequency of a given CPU from this function; it is
4639 * merely a notifier, which can be called from any CPU.
4640 * Changing the TSC frequency at arbitrary points in time
4641 * requires a recomputation of local variables related to
4642 * the TSC for each VCPU. We must flag these local variables
4643 * to be updated and be sure the update takes place with the
4644 * new frequency before any guests proceed.
4646 * Unfortunately, the combination of hotplug CPU and frequency
4647 * change creates an intractable locking scenario; the order
4648 * of when these callouts happen is undefined with respect to
4649 * CPU hotplug, and they can race with each other. As such,
4650 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4651 * undefined; you can actually have a CPU frequency change take
4652 * place in between the computation of X and the setting of the
4653 * variable. To protect against this problem, all updates of
4654 * the per_cpu tsc_khz variable are done in an interrupt
4655 * protected IPI, and all callers wishing to update the value
4656 * must wait for a synchronous IPI to complete (which is trivial
4657 * if the caller is on the CPU already). This establishes the
4658 * necessary total order on variable updates.
4660 * Note that because a guest time update may take place
4661 * anytime after the setting of the VCPU's request bit, the
4662 * correct TSC value must be set before the request. However,
4663 * to ensure the update actually makes it to any guest which
4664 * starts running in hardware virtualization between the set
4665 * and the acquisition of the spinlock, we must also ping the
4666 * CPU after setting the request bit.
4670 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4672 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4675 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4677 raw_spin_lock(&kvm_lock);
4678 list_for_each_entry(kvm, &vm_list, vm_list) {
4679 kvm_for_each_vcpu(i, vcpu, kvm) {
4680 if (vcpu->cpu != freq->cpu)
4682 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4683 if (vcpu->cpu != smp_processor_id())
4687 raw_spin_unlock(&kvm_lock);
4689 if (freq->old < freq->new && send_ipi) {
4691 * We upscale the frequency. Must make the guest
4692 * doesn't see old kvmclock values while running with
4693 * the new frequency, otherwise we risk the guest sees
4694 * time go backwards.
4696 * In case we update the frequency for another cpu
4697 * (which might be in guest context) send an interrupt
4698 * to kick the cpu out of guest context. Next time
4699 * guest context is entered kvmclock will be updated,
4700 * so the guest will not see stale values.
4702 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4707 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4708 .notifier_call = kvmclock_cpufreq_notifier
4711 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4712 unsigned long action, void *hcpu)
4714 unsigned int cpu = (unsigned long)hcpu;
4718 case CPU_DOWN_FAILED:
4719 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4721 case CPU_DOWN_PREPARE:
4722 smp_call_function_single(cpu, tsc_bad, NULL, 1);
4728 static struct notifier_block kvmclock_cpu_notifier_block = {
4729 .notifier_call = kvmclock_cpu_notifier,
4730 .priority = -INT_MAX
4733 static void kvm_timer_init(void)
4737 max_tsc_khz = tsc_khz;
4738 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4739 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4740 #ifdef CONFIG_CPU_FREQ
4741 struct cpufreq_policy policy;
4742 memset(&policy, 0, sizeof(policy));
4744 cpufreq_get_policy(&policy, cpu);
4745 if (policy.cpuinfo.max_freq)
4746 max_tsc_khz = policy.cpuinfo.max_freq;
4749 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4750 CPUFREQ_TRANSITION_NOTIFIER);
4752 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
4753 for_each_online_cpu(cpu)
4754 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4757 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4759 int kvm_is_in_guest(void)
4761 return __this_cpu_read(current_vcpu) != NULL;
4764 static int kvm_is_user_mode(void)
4768 if (__this_cpu_read(current_vcpu))
4769 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
4771 return user_mode != 0;
4774 static unsigned long kvm_get_guest_ip(void)
4776 unsigned long ip = 0;
4778 if (__this_cpu_read(current_vcpu))
4779 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
4784 static struct perf_guest_info_callbacks kvm_guest_cbs = {
4785 .is_in_guest = kvm_is_in_guest,
4786 .is_user_mode = kvm_is_user_mode,
4787 .get_guest_ip = kvm_get_guest_ip,
4790 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4792 __this_cpu_write(current_vcpu, vcpu);
4794 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4796 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4798 __this_cpu_write(current_vcpu, NULL);
4800 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4802 static void kvm_set_mmio_spte_mask(void)
4805 int maxphyaddr = boot_cpu_data.x86_phys_bits;
4808 * Set the reserved bits and the present bit of an paging-structure
4809 * entry to generate page fault with PFER.RSV = 1.
4811 mask = ((1ull << (62 - maxphyaddr + 1)) - 1) << maxphyaddr;
4814 #ifdef CONFIG_X86_64
4816 * If reserved bit is not supported, clear the present bit to disable
4819 if (maxphyaddr == 52)
4823 kvm_mmu_set_mmio_spte_mask(mask);
4826 int kvm_arch_init(void *opaque)
4829 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4832 printk(KERN_ERR "kvm: already loaded the other module\n");
4837 if (!ops->cpu_has_kvm_support()) {
4838 printk(KERN_ERR "kvm: no hardware support\n");
4842 if (ops->disabled_by_bios()) {
4843 printk(KERN_ERR "kvm: disabled by bios\n");
4848 r = kvm_mmu_module_init();
4852 kvm_set_mmio_spte_mask();
4853 kvm_init_msr_list();
4856 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4857 PT_DIRTY_MASK, PT64_NX_MASK, 0);
4861 perf_register_guest_info_callbacks(&kvm_guest_cbs);
4864 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
4872 void kvm_arch_exit(void)
4874 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4876 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4877 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4878 CPUFREQ_TRANSITION_NOTIFIER);
4879 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4881 kvm_mmu_module_exit();
4884 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4886 ++vcpu->stat.halt_exits;
4887 if (irqchip_in_kernel(vcpu->kvm)) {
4888 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
4891 vcpu->run->exit_reason = KVM_EXIT_HLT;
4895 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4897 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4899 u64 param, ingpa, outgpa, ret;
4900 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4901 bool fast, longmode;
4905 * hypercall generates UD from non zero cpl and real mode
4908 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
4909 kvm_queue_exception(vcpu, UD_VECTOR);
4913 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4914 longmode = is_long_mode(vcpu) && cs_l == 1;
4917 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4918 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4919 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4920 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4921 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4922 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
4924 #ifdef CONFIG_X86_64
4926 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4927 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4928 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4932 code = param & 0xffff;
4933 fast = (param >> 16) & 0x1;
4934 rep_cnt = (param >> 32) & 0xfff;
4935 rep_idx = (param >> 48) & 0xfff;
4937 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
4940 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
4941 kvm_vcpu_on_spin(vcpu);
4944 res = HV_STATUS_INVALID_HYPERCALL_CODE;
4948 ret = res | (((u64)rep_done & 0xfff) << 32);
4950 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4952 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
4953 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
4959 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
4961 unsigned long nr, a0, a1, a2, a3, ret;
4964 if (kvm_hv_hypercall_enabled(vcpu->kvm))
4965 return kvm_hv_hypercall(vcpu);
4967 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
4968 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
4969 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
4970 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
4971 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
4973 trace_kvm_hypercall(nr, a0, a1, a2, a3);
4975 if (!is_long_mode(vcpu)) {
4983 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
4989 case KVM_HC_VAPIC_POLL_IRQ:
4997 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4998 ++vcpu->stat.hypercalls;
5001 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5003 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5005 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5006 char instruction[3];
5007 unsigned long rip = kvm_rip_read(vcpu);
5010 * Blow out the MMU to ensure that no other VCPU has an active mapping
5011 * to ensure that the updated hypercall appears atomically across all
5014 kvm_mmu_zap_all(vcpu->kvm);
5016 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5018 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5022 * Check if userspace requested an interrupt window, and that the
5023 * interrupt window is open.
5025 * No need to exit to userspace if we already have an interrupt queued.
5027 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5029 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5030 vcpu->run->request_interrupt_window &&
5031 kvm_arch_interrupt_allowed(vcpu));
5034 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5036 struct kvm_run *kvm_run = vcpu->run;
5038 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5039 kvm_run->cr8 = kvm_get_cr8(vcpu);
5040 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5041 if (irqchip_in_kernel(vcpu->kvm))
5042 kvm_run->ready_for_interrupt_injection = 1;
5044 kvm_run->ready_for_interrupt_injection =
5045 kvm_arch_interrupt_allowed(vcpu) &&
5046 !kvm_cpu_has_interrupt(vcpu) &&
5047 !kvm_event_needs_reinjection(vcpu);
5050 static void vapic_enter(struct kvm_vcpu *vcpu)
5052 struct kvm_lapic *apic = vcpu->arch.apic;
5055 if (!apic || !apic->vapic_addr)
5058 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5060 vcpu->arch.apic->vapic_page = page;
5063 static void vapic_exit(struct kvm_vcpu *vcpu)
5065 struct kvm_lapic *apic = vcpu->arch.apic;
5068 if (!apic || !apic->vapic_addr)
5071 idx = srcu_read_lock(&vcpu->kvm->srcu);
5072 kvm_release_page_dirty(apic->vapic_page);
5073 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5074 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5077 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5081 if (!kvm_x86_ops->update_cr8_intercept)
5084 if (!vcpu->arch.apic)
5087 if (!vcpu->arch.apic->vapic_addr)
5088 max_irr = kvm_lapic_find_highest_irr(vcpu);
5095 tpr = kvm_lapic_get_cr8(vcpu);
5097 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5100 static void inject_pending_event(struct kvm_vcpu *vcpu)
5102 /* try to reinject previous events if any */
5103 if (vcpu->arch.exception.pending) {
5104 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5105 vcpu->arch.exception.has_error_code,
5106 vcpu->arch.exception.error_code);
5107 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5108 vcpu->arch.exception.has_error_code,
5109 vcpu->arch.exception.error_code,
5110 vcpu->arch.exception.reinject);
5114 if (vcpu->arch.nmi_injected) {
5115 kvm_x86_ops->set_nmi(vcpu);
5119 if (vcpu->arch.interrupt.pending) {
5120 kvm_x86_ops->set_irq(vcpu);
5124 /* try to inject new event if pending */
5125 if (vcpu->arch.nmi_pending) {
5126 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5127 --vcpu->arch.nmi_pending;
5128 vcpu->arch.nmi_injected = true;
5129 kvm_x86_ops->set_nmi(vcpu);
5131 } else if (kvm_cpu_has_interrupt(vcpu)) {
5132 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5133 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5135 kvm_x86_ops->set_irq(vcpu);
5140 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5142 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5143 !vcpu->guest_xcr0_loaded) {
5144 /* kvm_set_xcr() also depends on this */
5145 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5146 vcpu->guest_xcr0_loaded = 1;
5150 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5152 if (vcpu->guest_xcr0_loaded) {
5153 if (vcpu->arch.xcr0 != host_xcr0)
5154 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5155 vcpu->guest_xcr0_loaded = 0;
5159 static void process_nmi(struct kvm_vcpu *vcpu)
5164 * x86 is limited to one NMI running, and one NMI pending after it.
5165 * If an NMI is already in progress, limit further NMIs to just one.
5166 * Otherwise, allow two (and we'll inject the first one immediately).
5168 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5171 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5172 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5173 kvm_make_request(KVM_REQ_EVENT, vcpu);
5176 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5179 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5180 vcpu->run->request_interrupt_window;
5181 bool req_immediate_exit = 0;
5183 if (vcpu->requests) {
5184 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5185 kvm_mmu_unload(vcpu);
5186 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5187 __kvm_migrate_timers(vcpu);
5188 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5189 r = kvm_guest_time_update(vcpu);
5193 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5194 kvm_mmu_sync_roots(vcpu);
5195 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5196 kvm_x86_ops->tlb_flush(vcpu);
5197 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5198 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5202 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5203 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5207 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5208 vcpu->fpu_active = 0;
5209 kvm_x86_ops->fpu_deactivate(vcpu);
5211 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5212 /* Page is swapped out. Do synthetic halt */
5213 vcpu->arch.apf.halted = true;
5217 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5218 record_steal_time(vcpu);
5219 if (kvm_check_request(KVM_REQ_NMI, vcpu))
5221 req_immediate_exit =
5222 kvm_check_request(KVM_REQ_IMMEDIATE_EXIT, vcpu);
5223 if (kvm_check_request(KVM_REQ_PMU, vcpu))
5224 kvm_handle_pmu_event(vcpu);
5225 if (kvm_check_request(KVM_REQ_PMI, vcpu))
5226 kvm_deliver_pmi(vcpu);
5229 r = kvm_mmu_reload(vcpu);
5233 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5234 inject_pending_event(vcpu);
5236 /* enable NMI/IRQ window open exits if needed */
5237 if (vcpu->arch.nmi_pending)
5238 kvm_x86_ops->enable_nmi_window(vcpu);
5239 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5240 kvm_x86_ops->enable_irq_window(vcpu);
5242 if (kvm_lapic_enabled(vcpu)) {
5243 update_cr8_intercept(vcpu);
5244 kvm_lapic_sync_to_vapic(vcpu);
5250 kvm_x86_ops->prepare_guest_switch(vcpu);
5251 if (vcpu->fpu_active)
5252 kvm_load_guest_fpu(vcpu);
5253 kvm_load_guest_xcr0(vcpu);
5255 vcpu->mode = IN_GUEST_MODE;
5257 /* We should set ->mode before check ->requests,
5258 * see the comment in make_all_cpus_request.
5262 local_irq_disable();
5264 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5265 || need_resched() || signal_pending(current)) {
5266 vcpu->mode = OUTSIDE_GUEST_MODE;
5270 kvm_x86_ops->cancel_injection(vcpu);
5275 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5277 if (req_immediate_exit)
5278 smp_send_reschedule(vcpu->cpu);
5282 if (unlikely(vcpu->arch.switch_db_regs)) {
5284 set_debugreg(vcpu->arch.eff_db[0], 0);
5285 set_debugreg(vcpu->arch.eff_db[1], 1);
5286 set_debugreg(vcpu->arch.eff_db[2], 2);
5287 set_debugreg(vcpu->arch.eff_db[3], 3);
5290 trace_kvm_entry(vcpu->vcpu_id);
5291 kvm_x86_ops->run(vcpu);
5294 * If the guest has used debug registers, at least dr7
5295 * will be disabled while returning to the host.
5296 * If we don't have active breakpoints in the host, we don't
5297 * care about the messed up debug address registers. But if
5298 * we have some of them active, restore the old state.
5300 if (hw_breakpoint_active())
5301 hw_breakpoint_restore();
5303 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu);
5305 vcpu->mode = OUTSIDE_GUEST_MODE;
5312 * We must have an instruction between local_irq_enable() and
5313 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5314 * the interrupt shadow. The stat.exits increment will do nicely.
5315 * But we need to prevent reordering, hence this barrier():
5323 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5326 * Profile KVM exit RIPs:
5328 if (unlikely(prof_on == KVM_PROFILING)) {
5329 unsigned long rip = kvm_rip_read(vcpu);
5330 profile_hit(KVM_PROFILING, (void *)rip);
5333 if (unlikely(vcpu->arch.tsc_always_catchup))
5334 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5336 kvm_lapic_sync_from_vapic(vcpu);
5338 r = kvm_x86_ops->handle_exit(vcpu);
5344 static int __vcpu_run(struct kvm_vcpu *vcpu)
5347 struct kvm *kvm = vcpu->kvm;
5349 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5350 pr_debug("vcpu %d received sipi with vector # %x\n",
5351 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5352 kvm_lapic_reset(vcpu);
5353 r = kvm_arch_vcpu_reset(vcpu);
5356 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5359 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5364 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5365 !vcpu->arch.apf.halted)
5366 r = vcpu_enter_guest(vcpu);
5368 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5369 kvm_vcpu_block(vcpu);
5370 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5371 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5373 switch(vcpu->arch.mp_state) {
5374 case KVM_MP_STATE_HALTED:
5375 vcpu->arch.mp_state =
5376 KVM_MP_STATE_RUNNABLE;
5377 case KVM_MP_STATE_RUNNABLE:
5378 vcpu->arch.apf.halted = false;
5380 case KVM_MP_STATE_SIPI_RECEIVED:
5391 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5392 if (kvm_cpu_has_pending_timer(vcpu))
5393 kvm_inject_pending_timer_irqs(vcpu);
5395 if (dm_request_for_irq_injection(vcpu)) {
5397 vcpu->run->exit_reason = KVM_EXIT_INTR;
5398 ++vcpu->stat.request_irq_exits;
5401 kvm_check_async_pf_completion(vcpu);
5403 if (signal_pending(current)) {
5405 vcpu->run->exit_reason = KVM_EXIT_INTR;
5406 ++vcpu->stat.signal_exits;
5408 if (need_resched()) {
5409 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5411 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5415 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5422 static int complete_mmio(struct kvm_vcpu *vcpu)
5424 struct kvm_run *run = vcpu->run;
5427 if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5430 if (vcpu->mmio_needed) {
5431 vcpu->mmio_needed = 0;
5432 if (!vcpu->mmio_is_write)
5433 memcpy(vcpu->mmio_data + vcpu->mmio_index,
5435 vcpu->mmio_index += 8;
5436 if (vcpu->mmio_index < vcpu->mmio_size) {
5437 run->exit_reason = KVM_EXIT_MMIO;
5438 run->mmio.phys_addr = vcpu->mmio_phys_addr + vcpu->mmio_index;
5439 memcpy(run->mmio.data, vcpu->mmio_data + vcpu->mmio_index, 8);
5440 run->mmio.len = min(vcpu->mmio_size - vcpu->mmio_index, 8);
5441 run->mmio.is_write = vcpu->mmio_is_write;
5442 vcpu->mmio_needed = 1;
5445 if (vcpu->mmio_is_write)
5447 vcpu->mmio_read_completed = 1;
5449 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5450 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5451 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5452 if (r != EMULATE_DONE)
5457 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5462 if (!tsk_used_math(current) && init_fpu(current))
5465 if (vcpu->sigset_active)
5466 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5468 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5469 kvm_vcpu_block(vcpu);
5470 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5475 /* re-sync apic's tpr */
5476 if (!irqchip_in_kernel(vcpu->kvm)) {
5477 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5483 r = complete_mmio(vcpu);
5487 r = __vcpu_run(vcpu);
5490 post_kvm_run_save(vcpu);
5491 if (vcpu->sigset_active)
5492 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5497 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5499 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5501 * We are here if userspace calls get_regs() in the middle of
5502 * instruction emulation. Registers state needs to be copied
5503 * back from emulation context to vcpu. Usrapace shouldn't do
5504 * that usually, but some bad designed PV devices (vmware
5505 * backdoor interface) need this to work
5507 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5508 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
5509 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5511 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5512 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5513 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5514 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5515 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5516 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5517 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5518 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5519 #ifdef CONFIG_X86_64
5520 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5521 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5522 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5523 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5524 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5525 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5526 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5527 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5530 regs->rip = kvm_rip_read(vcpu);
5531 regs->rflags = kvm_get_rflags(vcpu);
5536 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5538 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
5539 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5541 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5542 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5543 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5544 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5545 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5546 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5547 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5548 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5549 #ifdef CONFIG_X86_64
5550 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5551 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5552 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5553 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5554 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5555 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5556 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5557 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5560 kvm_rip_write(vcpu, regs->rip);
5561 kvm_set_rflags(vcpu, regs->rflags);
5563 vcpu->arch.exception.pending = false;
5565 kvm_make_request(KVM_REQ_EVENT, vcpu);
5570 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5572 struct kvm_segment cs;
5574 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
5578 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5580 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5581 struct kvm_sregs *sregs)
5585 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5586 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5587 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5588 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5589 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5590 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5592 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5593 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5595 kvm_x86_ops->get_idt(vcpu, &dt);
5596 sregs->idt.limit = dt.size;
5597 sregs->idt.base = dt.address;
5598 kvm_x86_ops->get_gdt(vcpu, &dt);
5599 sregs->gdt.limit = dt.size;
5600 sregs->gdt.base = dt.address;
5602 sregs->cr0 = kvm_read_cr0(vcpu);
5603 sregs->cr2 = vcpu->arch.cr2;
5604 sregs->cr3 = kvm_read_cr3(vcpu);
5605 sregs->cr4 = kvm_read_cr4(vcpu);
5606 sregs->cr8 = kvm_get_cr8(vcpu);
5607 sregs->efer = vcpu->arch.efer;
5608 sregs->apic_base = kvm_get_apic_base(vcpu);
5610 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
5612 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
5613 set_bit(vcpu->arch.interrupt.nr,
5614 (unsigned long *)sregs->interrupt_bitmap);
5619 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
5620 struct kvm_mp_state *mp_state)
5622 mp_state->mp_state = vcpu->arch.mp_state;
5626 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
5627 struct kvm_mp_state *mp_state)
5629 vcpu->arch.mp_state = mp_state->mp_state;
5630 kvm_make_request(KVM_REQ_EVENT, vcpu);
5634 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
5635 bool has_error_code, u32 error_code)
5637 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5640 init_emulate_ctxt(vcpu);
5642 ret = emulator_task_switch(ctxt, tss_selector, reason,
5643 has_error_code, error_code);
5646 return EMULATE_FAIL;
5648 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
5649 kvm_rip_write(vcpu, ctxt->eip);
5650 kvm_set_rflags(vcpu, ctxt->eflags);
5651 kvm_make_request(KVM_REQ_EVENT, vcpu);
5652 return EMULATE_DONE;
5654 EXPORT_SYMBOL_GPL(kvm_task_switch);
5656 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5657 struct kvm_sregs *sregs)
5659 int mmu_reset_needed = 0;
5660 int pending_vec, max_bits, idx;
5663 dt.size = sregs->idt.limit;
5664 dt.address = sregs->idt.base;
5665 kvm_x86_ops->set_idt(vcpu, &dt);
5666 dt.size = sregs->gdt.limit;
5667 dt.address = sregs->gdt.base;
5668 kvm_x86_ops->set_gdt(vcpu, &dt);
5670 vcpu->arch.cr2 = sregs->cr2;
5671 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
5672 vcpu->arch.cr3 = sregs->cr3;
5673 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
5675 kvm_set_cr8(vcpu, sregs->cr8);
5677 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5678 kvm_x86_ops->set_efer(vcpu, sregs->efer);
5679 kvm_set_apic_base(vcpu, sregs->apic_base);
5681 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5682 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5683 vcpu->arch.cr0 = sregs->cr0;
5685 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5686 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5687 if (sregs->cr4 & X86_CR4_OSXSAVE)
5688 kvm_update_cpuid(vcpu);
5690 idx = srcu_read_lock(&vcpu->kvm->srcu);
5691 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5692 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
5693 mmu_reset_needed = 1;
5695 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5697 if (mmu_reset_needed)
5698 kvm_mmu_reset_context(vcpu);
5700 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5701 pending_vec = find_first_bit(
5702 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5703 if (pending_vec < max_bits) {
5704 kvm_queue_interrupt(vcpu, pending_vec, false);
5705 pr_debug("Set back pending irq %d\n", pending_vec);
5708 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5709 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5710 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5711 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5712 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5713 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5715 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5716 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5718 update_cr8_intercept(vcpu);
5720 /* Older userspace won't unhalt the vcpu on reset. */
5721 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5722 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5724 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5726 kvm_make_request(KVM_REQ_EVENT, vcpu);
5731 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5732 struct kvm_guest_debug *dbg)
5734 unsigned long rflags;
5737 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5739 if (vcpu->arch.exception.pending)
5741 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5742 kvm_queue_exception(vcpu, DB_VECTOR);
5744 kvm_queue_exception(vcpu, BP_VECTOR);
5748 * Read rflags as long as potentially injected trace flags are still
5751 rflags = kvm_get_rflags(vcpu);
5753 vcpu->guest_debug = dbg->control;
5754 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5755 vcpu->guest_debug = 0;
5757 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5758 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5759 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5760 vcpu->arch.switch_db_regs =
5761 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5763 for (i = 0; i < KVM_NR_DB_REGS; i++)
5764 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5765 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5768 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5769 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5770 get_segment_base(vcpu, VCPU_SREG_CS);
5773 * Trigger an rflags update that will inject or remove the trace
5776 kvm_set_rflags(vcpu, rflags);
5778 kvm_x86_ops->set_guest_debug(vcpu, dbg);
5788 * Translate a guest virtual address to a guest physical address.
5790 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5791 struct kvm_translation *tr)
5793 unsigned long vaddr = tr->linear_address;
5797 idx = srcu_read_lock(&vcpu->kvm->srcu);
5798 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5799 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5800 tr->physical_address = gpa;
5801 tr->valid = gpa != UNMAPPED_GVA;
5808 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5810 struct i387_fxsave_struct *fxsave =
5811 &vcpu->arch.guest_fpu.state->fxsave;
5813 memcpy(fpu->fpr, fxsave->st_space, 128);
5814 fpu->fcw = fxsave->cwd;
5815 fpu->fsw = fxsave->swd;
5816 fpu->ftwx = fxsave->twd;
5817 fpu->last_opcode = fxsave->fop;
5818 fpu->last_ip = fxsave->rip;
5819 fpu->last_dp = fxsave->rdp;
5820 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5825 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5827 struct i387_fxsave_struct *fxsave =
5828 &vcpu->arch.guest_fpu.state->fxsave;
5830 memcpy(fxsave->st_space, fpu->fpr, 128);
5831 fxsave->cwd = fpu->fcw;
5832 fxsave->swd = fpu->fsw;
5833 fxsave->twd = fpu->ftwx;
5834 fxsave->fop = fpu->last_opcode;
5835 fxsave->rip = fpu->last_ip;
5836 fxsave->rdp = fpu->last_dp;
5837 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5842 int fx_init(struct kvm_vcpu *vcpu)
5846 err = fpu_alloc(&vcpu->arch.guest_fpu);
5850 fpu_finit(&vcpu->arch.guest_fpu);
5853 * Ensure guest xcr0 is valid for loading
5855 vcpu->arch.xcr0 = XSTATE_FP;
5857 vcpu->arch.cr0 |= X86_CR0_ET;
5861 EXPORT_SYMBOL_GPL(fx_init);
5863 static void fx_free(struct kvm_vcpu *vcpu)
5865 fpu_free(&vcpu->arch.guest_fpu);
5868 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5870 if (vcpu->guest_fpu_loaded)
5874 * Restore all possible states in the guest,
5875 * and assume host would use all available bits.
5876 * Guest xcr0 would be loaded later.
5878 kvm_put_guest_xcr0(vcpu);
5879 vcpu->guest_fpu_loaded = 1;
5880 unlazy_fpu(current);
5881 fpu_restore_checking(&vcpu->arch.guest_fpu);
5885 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5887 kvm_put_guest_xcr0(vcpu);
5889 if (!vcpu->guest_fpu_loaded)
5892 vcpu->guest_fpu_loaded = 0;
5893 fpu_save_init(&vcpu->arch.guest_fpu);
5894 ++vcpu->stat.fpu_reload;
5895 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
5899 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5901 kvmclock_reset(vcpu);
5903 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
5905 kvm_x86_ops->vcpu_free(vcpu);
5908 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5911 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
5912 printk_once(KERN_WARNING
5913 "kvm: SMP vm created on host with unstable TSC; "
5914 "guest TSC will not be reliable\n");
5915 return kvm_x86_ops->vcpu_create(kvm, id);
5918 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5922 vcpu->arch.mtrr_state.have_fixed = 1;
5924 r = kvm_arch_vcpu_reset(vcpu);
5926 r = kvm_mmu_setup(vcpu);
5932 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5934 vcpu->arch.apf.msr_val = 0;
5937 kvm_mmu_unload(vcpu);
5941 kvm_x86_ops->vcpu_free(vcpu);
5944 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5946 atomic_set(&vcpu->arch.nmi_queued, 0);
5947 vcpu->arch.nmi_pending = 0;
5948 vcpu->arch.nmi_injected = false;
5950 vcpu->arch.switch_db_regs = 0;
5951 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5952 vcpu->arch.dr6 = DR6_FIXED_1;
5953 vcpu->arch.dr7 = DR7_FIXED_1;
5955 kvm_make_request(KVM_REQ_EVENT, vcpu);
5956 vcpu->arch.apf.msr_val = 0;
5957 vcpu->arch.st.msr_val = 0;
5959 kvmclock_reset(vcpu);
5961 kvm_clear_async_pf_completion_queue(vcpu);
5962 kvm_async_pf_hash_reset(vcpu);
5963 vcpu->arch.apf.halted = false;
5965 kvm_pmu_reset(vcpu);
5967 return kvm_x86_ops->vcpu_reset(vcpu);
5970 int kvm_arch_hardware_enable(void *garbage)
5973 struct kvm_vcpu *vcpu;
5978 bool stable, backwards_tsc = false;
5980 kvm_shared_msr_cpu_online();
5981 ret = kvm_x86_ops->hardware_enable(garbage);
5985 local_tsc = native_read_tsc();
5986 stable = !check_tsc_unstable();
5987 list_for_each_entry(kvm, &vm_list, vm_list) {
5988 kvm_for_each_vcpu(i, vcpu, kvm) {
5989 if (!stable && vcpu->cpu == smp_processor_id())
5990 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
5991 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
5992 backwards_tsc = true;
5993 if (vcpu->arch.last_host_tsc > max_tsc)
5994 max_tsc = vcpu->arch.last_host_tsc;
6000 * Sometimes, even reliable TSCs go backwards. This happens on
6001 * platforms that reset TSC during suspend or hibernate actions, but
6002 * maintain synchronization. We must compensate. Fortunately, we can
6003 * detect that condition here, which happens early in CPU bringup,
6004 * before any KVM threads can be running. Unfortunately, we can't
6005 * bring the TSCs fully up to date with real time, as we aren't yet far
6006 * enough into CPU bringup that we know how much real time has actually
6007 * elapsed; our helper function, get_kernel_ns() will be using boot
6008 * variables that haven't been updated yet.
6010 * So we simply find the maximum observed TSC above, then record the
6011 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
6012 * the adjustment will be applied. Note that we accumulate
6013 * adjustments, in case multiple suspend cycles happen before some VCPU
6014 * gets a chance to run again. In the event that no KVM threads get a
6015 * chance to run, we will miss the entire elapsed period, as we'll have
6016 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
6017 * loose cycle time. This isn't too big a deal, since the loss will be
6018 * uniform across all VCPUs (not to mention the scenario is extremely
6019 * unlikely). It is possible that a second hibernate recovery happens
6020 * much faster than a first, causing the observed TSC here to be
6021 * smaller; this would require additional padding adjustment, which is
6022 * why we set last_host_tsc to the local tsc observed here.
6024 * N.B. - this code below runs only on platforms with reliable TSC,
6025 * as that is the only way backwards_tsc is set above. Also note
6026 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
6027 * have the same delta_cyc adjustment applied if backwards_tsc
6028 * is detected. Note further, this adjustment is only done once,
6029 * as we reset last_host_tsc on all VCPUs to stop this from being
6030 * called multiple times (one for each physical CPU bringup).
6032 * Platforms with unnreliable TSCs don't have to deal with this, they
6033 * will be compensated by the logic in vcpu_load, which sets the TSC to
6034 * catchup mode. This will catchup all VCPUs to real time, but cannot
6035 * guarantee that they stay in perfect synchronization.
6037 if (backwards_tsc) {
6038 u64 delta_cyc = max_tsc - local_tsc;
6039 list_for_each_entry(kvm, &vm_list, vm_list) {
6040 kvm_for_each_vcpu(i, vcpu, kvm) {
6041 vcpu->arch.tsc_offset_adjustment += delta_cyc;
6042 vcpu->arch.last_host_tsc = local_tsc;
6046 * We have to disable TSC offset matching.. if you were
6047 * booting a VM while issuing an S4 host suspend....
6048 * you may have some problem. Solving this issue is
6049 * left as an exercise to the reader.
6051 kvm->arch.last_tsc_nsec = 0;
6052 kvm->arch.last_tsc_write = 0;
6059 void kvm_arch_hardware_disable(void *garbage)
6061 kvm_x86_ops->hardware_disable(garbage);
6062 drop_user_return_notifiers(garbage);
6065 int kvm_arch_hardware_setup(void)
6067 return kvm_x86_ops->hardware_setup();
6070 void kvm_arch_hardware_unsetup(void)
6072 kvm_x86_ops->hardware_unsetup();
6075 void kvm_arch_check_processor_compat(void *rtn)
6077 kvm_x86_ops->check_processor_compatibility(rtn);
6080 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6086 BUG_ON(vcpu->kvm == NULL);
6089 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6090 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6091 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6093 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6095 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6100 vcpu->arch.pio_data = page_address(page);
6102 kvm_set_tsc_khz(vcpu, max_tsc_khz);
6104 r = kvm_mmu_create(vcpu);
6106 goto fail_free_pio_data;
6108 if (irqchip_in_kernel(kvm)) {
6109 r = kvm_create_lapic(vcpu);
6111 goto fail_mmu_destroy;
6114 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6116 if (!vcpu->arch.mce_banks) {
6118 goto fail_free_lapic;
6120 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6122 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6123 goto fail_free_mce_banks;
6125 kvm_async_pf_hash_reset(vcpu);
6129 fail_free_mce_banks:
6130 kfree(vcpu->arch.mce_banks);
6132 kvm_free_lapic(vcpu);
6134 kvm_mmu_destroy(vcpu);
6136 free_page((unsigned long)vcpu->arch.pio_data);
6141 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6145 kvm_pmu_destroy(vcpu);
6146 kfree(vcpu->arch.mce_banks);
6147 kvm_free_lapic(vcpu);
6148 idx = srcu_read_lock(&vcpu->kvm->srcu);
6149 kvm_mmu_destroy(vcpu);
6150 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6151 free_page((unsigned long)vcpu->arch.pio_data);
6154 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
6159 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6160 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6162 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6163 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6165 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6170 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6173 kvm_mmu_unload(vcpu);
6177 static void kvm_free_vcpus(struct kvm *kvm)
6180 struct kvm_vcpu *vcpu;
6183 * Unpin any mmu pages first.
6185 kvm_for_each_vcpu(i, vcpu, kvm) {
6186 kvm_clear_async_pf_completion_queue(vcpu);
6187 kvm_unload_vcpu_mmu(vcpu);
6189 kvm_for_each_vcpu(i, vcpu, kvm)
6190 kvm_arch_vcpu_free(vcpu);
6192 mutex_lock(&kvm->lock);
6193 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6194 kvm->vcpus[i] = NULL;
6196 atomic_set(&kvm->online_vcpus, 0);
6197 mutex_unlock(&kvm->lock);
6200 void kvm_arch_sync_events(struct kvm *kvm)
6202 kvm_free_all_assigned_devices(kvm);
6206 void kvm_arch_destroy_vm(struct kvm *kvm)
6208 kvm_iommu_unmap_guest(kvm);
6209 kfree(kvm->arch.vpic);
6210 kfree(kvm->arch.vioapic);
6211 kvm_free_vcpus(kvm);
6212 if (kvm->arch.apic_access_page)
6213 put_page(kvm->arch.apic_access_page);
6214 if (kvm->arch.ept_identity_pagetable)
6215 put_page(kvm->arch.ept_identity_pagetable);
6218 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6219 struct kvm_memory_slot *memslot,
6220 struct kvm_memory_slot old,
6221 struct kvm_userspace_memory_region *mem,
6224 int npages = memslot->npages;
6225 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6227 /* Prevent internal slot pages from being moved by fork()/COW. */
6228 if (memslot->id >= KVM_MEMORY_SLOTS)
6229 map_flags = MAP_SHARED | MAP_ANONYMOUS;
6231 /*To keep backward compatibility with older userspace,
6232 *x86 needs to hanlde !user_alloc case.
6235 if (npages && !old.rmap) {
6236 unsigned long userspace_addr;
6238 down_write(¤t->mm->mmap_sem);
6239 userspace_addr = do_mmap(NULL, 0,
6241 PROT_READ | PROT_WRITE,
6244 up_write(¤t->mm->mmap_sem);
6246 if (IS_ERR((void *)userspace_addr))
6247 return PTR_ERR((void *)userspace_addr);
6249 memslot->userspace_addr = userspace_addr;
6257 void kvm_arch_commit_memory_region(struct kvm *kvm,
6258 struct kvm_userspace_memory_region *mem,
6259 struct kvm_memory_slot old,
6263 int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
6265 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6268 down_write(¤t->mm->mmap_sem);
6269 ret = do_munmap(current->mm, old.userspace_addr,
6270 old.npages * PAGE_SIZE);
6271 up_write(¤t->mm->mmap_sem);
6274 "kvm_vm_ioctl_set_memory_region: "
6275 "failed to munmap memory\n");
6278 if (!kvm->arch.n_requested_mmu_pages)
6279 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6281 spin_lock(&kvm->mmu_lock);
6283 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6284 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6285 spin_unlock(&kvm->mmu_lock);
6288 void kvm_arch_flush_shadow(struct kvm *kvm)
6290 kvm_mmu_zap_all(kvm);
6291 kvm_reload_remote_mmus(kvm);
6294 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6296 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6297 !vcpu->arch.apf.halted)
6298 || !list_empty_careful(&vcpu->async_pf.done)
6299 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6300 || atomic_read(&vcpu->arch.nmi_queued) ||
6301 (kvm_arch_interrupt_allowed(vcpu) &&
6302 kvm_cpu_has_interrupt(vcpu));
6305 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6308 int cpu = vcpu->cpu;
6310 if (waitqueue_active(&vcpu->wq)) {
6311 wake_up_interruptible(&vcpu->wq);
6312 ++vcpu->stat.halt_wakeup;
6316 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6317 if (kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE)
6318 smp_send_reschedule(cpu);
6322 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6324 return kvm_x86_ops->interrupt_allowed(vcpu);
6327 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6329 unsigned long current_rip = kvm_rip_read(vcpu) +
6330 get_segment_base(vcpu, VCPU_SREG_CS);
6332 return current_rip == linear_rip;
6334 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6336 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6338 unsigned long rflags;
6340 rflags = kvm_x86_ops->get_rflags(vcpu);
6341 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6342 rflags &= ~X86_EFLAGS_TF;
6345 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6347 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6349 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6350 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6351 rflags |= X86_EFLAGS_TF;
6352 kvm_x86_ops->set_rflags(vcpu, rflags);
6353 kvm_make_request(KVM_REQ_EVENT, vcpu);
6355 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6357 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6361 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6362 is_error_page(work->page))
6365 r = kvm_mmu_reload(vcpu);
6369 if (!vcpu->arch.mmu.direct_map &&
6370 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6373 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6376 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6378 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6381 static inline u32 kvm_async_pf_next_probe(u32 key)
6383 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6386 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6388 u32 key = kvm_async_pf_hash_fn(gfn);
6390 while (vcpu->arch.apf.gfns[key] != ~0)
6391 key = kvm_async_pf_next_probe(key);
6393 vcpu->arch.apf.gfns[key] = gfn;
6396 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6399 u32 key = kvm_async_pf_hash_fn(gfn);
6401 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6402 (vcpu->arch.apf.gfns[key] != gfn &&
6403 vcpu->arch.apf.gfns[key] != ~0); i++)
6404 key = kvm_async_pf_next_probe(key);
6409 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6411 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6414 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6418 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6420 vcpu->arch.apf.gfns[i] = ~0;
6422 j = kvm_async_pf_next_probe(j);
6423 if (vcpu->arch.apf.gfns[j] == ~0)
6425 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6427 * k lies cyclically in ]i,j]
6429 * |....j i.k.| or |.k..j i...|
6431 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6432 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6437 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6440 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6444 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6445 struct kvm_async_pf *work)
6447 struct x86_exception fault;
6449 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6450 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6452 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6453 (vcpu->arch.apf.send_user_only &&
6454 kvm_x86_ops->get_cpl(vcpu) == 0))
6455 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6456 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6457 fault.vector = PF_VECTOR;
6458 fault.error_code_valid = true;
6459 fault.error_code = 0;
6460 fault.nested_page_fault = false;
6461 fault.address = work->arch.token;
6462 kvm_inject_page_fault(vcpu, &fault);
6466 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6467 struct kvm_async_pf *work)
6469 struct x86_exception fault;
6471 trace_kvm_async_pf_ready(work->arch.token, work->gva);
6472 if (is_error_page(work->page))
6473 work->arch.token = ~0; /* broadcast wakeup */
6475 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6477 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6478 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6479 fault.vector = PF_VECTOR;
6480 fault.error_code_valid = true;
6481 fault.error_code = 0;
6482 fault.nested_page_fault = false;
6483 fault.address = work->arch.token;
6484 kvm_inject_page_fault(vcpu, &fault);
6486 vcpu->arch.apf.halted = false;
6489 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6491 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6494 return !kvm_event_needs_reinjection(vcpu) &&
6495 kvm_x86_ops->interrupt_allowed(vcpu);
6498 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6499 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6500 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6501 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6502 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6503 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6504 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6505 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6506 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6507 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6508 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6509 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);