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.this_tsc_nsec,
1017 vcpu->arch.virtual_tsc_mult,
1018 vcpu->arch.virtual_tsc_shift);
1019 tsc += vcpu->arch.this_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.cur_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);
1072 * We split periods of matched TSC writes into generations.
1073 * For each generation, we track the original measured
1074 * nanosecond time, offset, and write, so if TSCs are in
1075 * sync, we can match exact offset, and if not, we can match
1076 * exact software computaion in compute_guest_tsc()
1078 * These values are tracked in kvm->arch.cur_xxx variables.
1080 kvm->arch.cur_tsc_generation++;
1081 kvm->arch.cur_tsc_nsec = ns;
1082 kvm->arch.cur_tsc_write = data;
1083 kvm->arch.cur_tsc_offset = offset;
1084 pr_debug("kvm: new tsc generation %u, clock %llu\n",
1085 kvm->arch.cur_tsc_generation, data);
1089 * We also track th most recent recorded KHZ, write and time to
1090 * allow the matching interval to be extended at each write.
1092 kvm->arch.last_tsc_nsec = ns;
1093 kvm->arch.last_tsc_write = data;
1094 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1096 /* Reset of TSC must disable overshoot protection below */
1097 vcpu->arch.hv_clock.tsc_timestamp = 0;
1098 vcpu->arch.last_guest_tsc = data;
1100 /* Keep track of which generation this VCPU has synchronized to */
1101 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1102 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1103 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1105 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1106 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1109 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1111 static int kvm_guest_time_update(struct kvm_vcpu *v)
1113 unsigned long flags;
1114 struct kvm_vcpu_arch *vcpu = &v->arch;
1116 unsigned long this_tsc_khz;
1117 s64 kernel_ns, max_kernel_ns;
1120 /* Keep irq disabled to prevent changes to the clock */
1121 local_irq_save(flags);
1122 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v);
1123 kernel_ns = get_kernel_ns();
1124 this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
1125 if (unlikely(this_tsc_khz == 0)) {
1126 local_irq_restore(flags);
1127 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1132 * We may have to catch up the TSC to match elapsed wall clock
1133 * time for two reasons, even if kvmclock is used.
1134 * 1) CPU could have been running below the maximum TSC rate
1135 * 2) Broken TSC compensation resets the base at each VCPU
1136 * entry to avoid unknown leaps of TSC even when running
1137 * again on the same CPU. This may cause apparent elapsed
1138 * time to disappear, and the guest to stand still or run
1141 if (vcpu->tsc_catchup) {
1142 u64 tsc = compute_guest_tsc(v, kernel_ns);
1143 if (tsc > tsc_timestamp) {
1144 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1145 tsc_timestamp = tsc;
1149 local_irq_restore(flags);
1151 if (!vcpu->time_page)
1155 * Time as measured by the TSC may go backwards when resetting the base
1156 * tsc_timestamp. The reason for this is that the TSC resolution is
1157 * higher than the resolution of the other clock scales. Thus, many
1158 * possible measurments of the TSC correspond to one measurement of any
1159 * other clock, and so a spread of values is possible. This is not a
1160 * problem for the computation of the nanosecond clock; with TSC rates
1161 * around 1GHZ, there can only be a few cycles which correspond to one
1162 * nanosecond value, and any path through this code will inevitably
1163 * take longer than that. However, with the kernel_ns value itself,
1164 * the precision may be much lower, down to HZ granularity. If the
1165 * first sampling of TSC against kernel_ns ends in the low part of the
1166 * range, and the second in the high end of the range, we can get:
1168 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1170 * As the sampling errors potentially range in the thousands of cycles,
1171 * it is possible such a time value has already been observed by the
1172 * guest. To protect against this, we must compute the system time as
1173 * observed by the guest and ensure the new system time is greater.
1176 if (vcpu->hv_clock.tsc_timestamp) {
1177 max_kernel_ns = vcpu->last_guest_tsc -
1178 vcpu->hv_clock.tsc_timestamp;
1179 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1180 vcpu->hv_clock.tsc_to_system_mul,
1181 vcpu->hv_clock.tsc_shift);
1182 max_kernel_ns += vcpu->last_kernel_ns;
1185 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1186 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1187 &vcpu->hv_clock.tsc_shift,
1188 &vcpu->hv_clock.tsc_to_system_mul);
1189 vcpu->hw_tsc_khz = this_tsc_khz;
1192 if (max_kernel_ns > kernel_ns)
1193 kernel_ns = max_kernel_ns;
1195 /* With all the info we got, fill in the values */
1196 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1197 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1198 vcpu->last_kernel_ns = kernel_ns;
1199 vcpu->last_guest_tsc = tsc_timestamp;
1200 vcpu->hv_clock.flags = 0;
1203 * The interface expects us to write an even number signaling that the
1204 * update is finished. Since the guest won't see the intermediate
1205 * state, we just increase by 2 at the end.
1207 vcpu->hv_clock.version += 2;
1209 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1211 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1212 sizeof(vcpu->hv_clock));
1214 kunmap_atomic(shared_kaddr, KM_USER0);
1216 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1220 static bool msr_mtrr_valid(unsigned msr)
1223 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1224 case MSR_MTRRfix64K_00000:
1225 case MSR_MTRRfix16K_80000:
1226 case MSR_MTRRfix16K_A0000:
1227 case MSR_MTRRfix4K_C0000:
1228 case MSR_MTRRfix4K_C8000:
1229 case MSR_MTRRfix4K_D0000:
1230 case MSR_MTRRfix4K_D8000:
1231 case MSR_MTRRfix4K_E0000:
1232 case MSR_MTRRfix4K_E8000:
1233 case MSR_MTRRfix4K_F0000:
1234 case MSR_MTRRfix4K_F8000:
1235 case MSR_MTRRdefType:
1236 case MSR_IA32_CR_PAT:
1244 static bool valid_pat_type(unsigned t)
1246 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1249 static bool valid_mtrr_type(unsigned t)
1251 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1254 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1258 if (!msr_mtrr_valid(msr))
1261 if (msr == MSR_IA32_CR_PAT) {
1262 for (i = 0; i < 8; i++)
1263 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1266 } else if (msr == MSR_MTRRdefType) {
1269 return valid_mtrr_type(data & 0xff);
1270 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1271 for (i = 0; i < 8 ; i++)
1272 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1277 /* variable MTRRs */
1278 return valid_mtrr_type(data & 0xff);
1281 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1283 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1285 if (!mtrr_valid(vcpu, msr, data))
1288 if (msr == MSR_MTRRdefType) {
1289 vcpu->arch.mtrr_state.def_type = data;
1290 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1291 } else if (msr == MSR_MTRRfix64K_00000)
1293 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1294 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1295 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1296 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1297 else if (msr == MSR_IA32_CR_PAT)
1298 vcpu->arch.pat = data;
1299 else { /* Variable MTRRs */
1300 int idx, is_mtrr_mask;
1303 idx = (msr - 0x200) / 2;
1304 is_mtrr_mask = msr - 0x200 - 2 * idx;
1307 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1310 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1314 kvm_mmu_reset_context(vcpu);
1318 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1320 u64 mcg_cap = vcpu->arch.mcg_cap;
1321 unsigned bank_num = mcg_cap & 0xff;
1324 case MSR_IA32_MCG_STATUS:
1325 vcpu->arch.mcg_status = data;
1327 case MSR_IA32_MCG_CTL:
1328 if (!(mcg_cap & MCG_CTL_P))
1330 if (data != 0 && data != ~(u64)0)
1332 vcpu->arch.mcg_ctl = data;
1335 if (msr >= MSR_IA32_MC0_CTL &&
1336 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1337 u32 offset = msr - MSR_IA32_MC0_CTL;
1338 /* only 0 or all 1s can be written to IA32_MCi_CTL
1339 * some Linux kernels though clear bit 10 in bank 4 to
1340 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1341 * this to avoid an uncatched #GP in the guest
1343 if ((offset & 0x3) == 0 &&
1344 data != 0 && (data | (1 << 10)) != ~(u64)0)
1346 vcpu->arch.mce_banks[offset] = data;
1354 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1356 struct kvm *kvm = vcpu->kvm;
1357 int lm = is_long_mode(vcpu);
1358 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1359 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1360 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1361 : kvm->arch.xen_hvm_config.blob_size_32;
1362 u32 page_num = data & ~PAGE_MASK;
1363 u64 page_addr = data & PAGE_MASK;
1368 if (page_num >= blob_size)
1371 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1376 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1385 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1387 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1390 static bool kvm_hv_msr_partition_wide(u32 msr)
1394 case HV_X64_MSR_GUEST_OS_ID:
1395 case HV_X64_MSR_HYPERCALL:
1403 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1405 struct kvm *kvm = vcpu->kvm;
1408 case HV_X64_MSR_GUEST_OS_ID:
1409 kvm->arch.hv_guest_os_id = data;
1410 /* setting guest os id to zero disables hypercall page */
1411 if (!kvm->arch.hv_guest_os_id)
1412 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1414 case HV_X64_MSR_HYPERCALL: {
1419 /* if guest os id is not set hypercall should remain disabled */
1420 if (!kvm->arch.hv_guest_os_id)
1422 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1423 kvm->arch.hv_hypercall = data;
1426 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1427 addr = gfn_to_hva(kvm, gfn);
1428 if (kvm_is_error_hva(addr))
1430 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1431 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1432 if (__copy_to_user((void __user *)addr, instructions, 4))
1434 kvm->arch.hv_hypercall = data;
1438 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1439 "data 0x%llx\n", msr, data);
1445 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1448 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1451 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1452 vcpu->arch.hv_vapic = data;
1455 addr = gfn_to_hva(vcpu->kvm, data >>
1456 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1457 if (kvm_is_error_hva(addr))
1459 if (__clear_user((void __user *)addr, PAGE_SIZE))
1461 vcpu->arch.hv_vapic = data;
1464 case HV_X64_MSR_EOI:
1465 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1466 case HV_X64_MSR_ICR:
1467 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1468 case HV_X64_MSR_TPR:
1469 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1471 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1472 "data 0x%llx\n", msr, data);
1479 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1481 gpa_t gpa = data & ~0x3f;
1483 /* Bits 2:5 are resrved, Should be zero */
1487 vcpu->arch.apf.msr_val = data;
1489 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1490 kvm_clear_async_pf_completion_queue(vcpu);
1491 kvm_async_pf_hash_reset(vcpu);
1495 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1498 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1499 kvm_async_pf_wakeup_all(vcpu);
1503 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1505 if (vcpu->arch.time_page) {
1506 kvm_release_page_dirty(vcpu->arch.time_page);
1507 vcpu->arch.time_page = NULL;
1511 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
1515 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1518 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
1519 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1520 vcpu->arch.st.accum_steal = delta;
1523 static void record_steal_time(struct kvm_vcpu *vcpu)
1525 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
1528 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1529 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
1532 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
1533 vcpu->arch.st.steal.version += 2;
1534 vcpu->arch.st.accum_steal = 0;
1536 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
1537 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
1540 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1546 return set_efer(vcpu, data);
1548 data &= ~(u64)0x40; /* ignore flush filter disable */
1549 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1551 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1556 case MSR_FAM10H_MMIO_CONF_BASE:
1558 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1563 case MSR_AMD64_NB_CFG:
1565 case MSR_IA32_DEBUGCTLMSR:
1567 /* We support the non-activated case already */
1569 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1570 /* Values other than LBR and BTF are vendor-specific,
1571 thus reserved and should throw a #GP */
1574 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1577 case MSR_IA32_UCODE_REV:
1578 case MSR_IA32_UCODE_WRITE:
1579 case MSR_VM_HSAVE_PA:
1580 case MSR_AMD64_PATCH_LOADER:
1582 case 0x200 ... 0x2ff:
1583 return set_msr_mtrr(vcpu, msr, data);
1584 case MSR_IA32_APICBASE:
1585 kvm_set_apic_base(vcpu, data);
1587 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1588 return kvm_x2apic_msr_write(vcpu, msr, data);
1589 case MSR_IA32_TSCDEADLINE:
1590 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1592 case MSR_IA32_MISC_ENABLE:
1593 vcpu->arch.ia32_misc_enable_msr = data;
1595 case MSR_KVM_WALL_CLOCK_NEW:
1596 case MSR_KVM_WALL_CLOCK:
1597 vcpu->kvm->arch.wall_clock = data;
1598 kvm_write_wall_clock(vcpu->kvm, data);
1600 case MSR_KVM_SYSTEM_TIME_NEW:
1601 case MSR_KVM_SYSTEM_TIME: {
1602 kvmclock_reset(vcpu);
1604 vcpu->arch.time = data;
1605 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1607 /* we verify if the enable bit is set... */
1611 /* ...but clean it before doing the actual write */
1612 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1614 vcpu->arch.time_page =
1615 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1617 if (is_error_page(vcpu->arch.time_page)) {
1618 kvm_release_page_clean(vcpu->arch.time_page);
1619 vcpu->arch.time_page = NULL;
1623 case MSR_KVM_ASYNC_PF_EN:
1624 if (kvm_pv_enable_async_pf(vcpu, data))
1627 case MSR_KVM_STEAL_TIME:
1629 if (unlikely(!sched_info_on()))
1632 if (data & KVM_STEAL_RESERVED_MASK)
1635 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
1636 data & KVM_STEAL_VALID_BITS))
1639 vcpu->arch.st.msr_val = data;
1641 if (!(data & KVM_MSR_ENABLED))
1644 vcpu->arch.st.last_steal = current->sched_info.run_delay;
1647 accumulate_steal_time(vcpu);
1650 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
1654 case MSR_IA32_MCG_CTL:
1655 case MSR_IA32_MCG_STATUS:
1656 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1657 return set_msr_mce(vcpu, msr, data);
1659 /* Performance counters are not protected by a CPUID bit,
1660 * so we should check all of them in the generic path for the sake of
1661 * cross vendor migration.
1662 * Writing a zero into the event select MSRs disables them,
1663 * which we perfectly emulate ;-). Any other value should be at least
1664 * reported, some guests depend on them.
1666 case MSR_K7_EVNTSEL0:
1667 case MSR_K7_EVNTSEL1:
1668 case MSR_K7_EVNTSEL2:
1669 case MSR_K7_EVNTSEL3:
1671 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1672 "0x%x data 0x%llx\n", msr, data);
1674 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1675 * so we ignore writes to make it happy.
1677 case MSR_K7_PERFCTR0:
1678 case MSR_K7_PERFCTR1:
1679 case MSR_K7_PERFCTR2:
1680 case MSR_K7_PERFCTR3:
1681 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1682 "0x%x data 0x%llx\n", msr, data);
1684 case MSR_P6_PERFCTR0:
1685 case MSR_P6_PERFCTR1:
1687 case MSR_P6_EVNTSEL0:
1688 case MSR_P6_EVNTSEL1:
1689 if (kvm_pmu_msr(vcpu, msr))
1690 return kvm_pmu_set_msr(vcpu, msr, data);
1692 if (pr || data != 0)
1693 pr_unimpl(vcpu, "disabled perfctr wrmsr: "
1694 "0x%x data 0x%llx\n", msr, data);
1696 case MSR_K7_CLK_CTL:
1698 * Ignore all writes to this no longer documented MSR.
1699 * Writes are only relevant for old K7 processors,
1700 * all pre-dating SVM, but a recommended workaround from
1701 * AMD for these chips. It is possible to speicify the
1702 * affected processor models on the command line, hence
1703 * the need to ignore the workaround.
1706 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1707 if (kvm_hv_msr_partition_wide(msr)) {
1709 mutex_lock(&vcpu->kvm->lock);
1710 r = set_msr_hyperv_pw(vcpu, msr, data);
1711 mutex_unlock(&vcpu->kvm->lock);
1714 return set_msr_hyperv(vcpu, msr, data);
1716 case MSR_IA32_BBL_CR_CTL3:
1717 /* Drop writes to this legacy MSR -- see rdmsr
1718 * counterpart for further detail.
1720 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1722 case MSR_AMD64_OSVW_ID_LENGTH:
1723 if (!guest_cpuid_has_osvw(vcpu))
1725 vcpu->arch.osvw.length = data;
1727 case MSR_AMD64_OSVW_STATUS:
1728 if (!guest_cpuid_has_osvw(vcpu))
1730 vcpu->arch.osvw.status = data;
1733 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1734 return xen_hvm_config(vcpu, data);
1735 if (kvm_pmu_msr(vcpu, msr))
1736 return kvm_pmu_set_msr(vcpu, msr, data);
1738 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1742 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1749 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1753 * Reads an msr value (of 'msr_index') into 'pdata'.
1754 * Returns 0 on success, non-0 otherwise.
1755 * Assumes vcpu_load() was already called.
1757 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1759 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1762 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1764 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1766 if (!msr_mtrr_valid(msr))
1769 if (msr == MSR_MTRRdefType)
1770 *pdata = vcpu->arch.mtrr_state.def_type +
1771 (vcpu->arch.mtrr_state.enabled << 10);
1772 else if (msr == MSR_MTRRfix64K_00000)
1774 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1775 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1776 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1777 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1778 else if (msr == MSR_IA32_CR_PAT)
1779 *pdata = vcpu->arch.pat;
1780 else { /* Variable MTRRs */
1781 int idx, is_mtrr_mask;
1784 idx = (msr - 0x200) / 2;
1785 is_mtrr_mask = msr - 0x200 - 2 * idx;
1788 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1791 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1798 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1801 u64 mcg_cap = vcpu->arch.mcg_cap;
1802 unsigned bank_num = mcg_cap & 0xff;
1805 case MSR_IA32_P5_MC_ADDR:
1806 case MSR_IA32_P5_MC_TYPE:
1809 case MSR_IA32_MCG_CAP:
1810 data = vcpu->arch.mcg_cap;
1812 case MSR_IA32_MCG_CTL:
1813 if (!(mcg_cap & MCG_CTL_P))
1815 data = vcpu->arch.mcg_ctl;
1817 case MSR_IA32_MCG_STATUS:
1818 data = vcpu->arch.mcg_status;
1821 if (msr >= MSR_IA32_MC0_CTL &&
1822 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1823 u32 offset = msr - MSR_IA32_MC0_CTL;
1824 data = vcpu->arch.mce_banks[offset];
1833 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1836 struct kvm *kvm = vcpu->kvm;
1839 case HV_X64_MSR_GUEST_OS_ID:
1840 data = kvm->arch.hv_guest_os_id;
1842 case HV_X64_MSR_HYPERCALL:
1843 data = kvm->arch.hv_hypercall;
1846 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1854 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1859 case HV_X64_MSR_VP_INDEX: {
1862 kvm_for_each_vcpu(r, v, vcpu->kvm)
1867 case HV_X64_MSR_EOI:
1868 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1869 case HV_X64_MSR_ICR:
1870 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1871 case HV_X64_MSR_TPR:
1872 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1873 case HV_X64_MSR_APIC_ASSIST_PAGE:
1874 data = vcpu->arch.hv_vapic;
1877 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1884 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1889 case MSR_IA32_PLATFORM_ID:
1890 case MSR_IA32_EBL_CR_POWERON:
1891 case MSR_IA32_DEBUGCTLMSR:
1892 case MSR_IA32_LASTBRANCHFROMIP:
1893 case MSR_IA32_LASTBRANCHTOIP:
1894 case MSR_IA32_LASTINTFROMIP:
1895 case MSR_IA32_LASTINTTOIP:
1898 case MSR_VM_HSAVE_PA:
1899 case MSR_K7_EVNTSEL0:
1900 case MSR_K7_PERFCTR0:
1901 case MSR_K8_INT_PENDING_MSG:
1902 case MSR_AMD64_NB_CFG:
1903 case MSR_FAM10H_MMIO_CONF_BASE:
1906 case MSR_P6_PERFCTR0:
1907 case MSR_P6_PERFCTR1:
1908 case MSR_P6_EVNTSEL0:
1909 case MSR_P6_EVNTSEL1:
1910 if (kvm_pmu_msr(vcpu, msr))
1911 return kvm_pmu_get_msr(vcpu, msr, pdata);
1914 case MSR_IA32_UCODE_REV:
1915 data = 0x100000000ULL;
1918 data = 0x500 | KVM_NR_VAR_MTRR;
1920 case 0x200 ... 0x2ff:
1921 return get_msr_mtrr(vcpu, msr, pdata);
1922 case 0xcd: /* fsb frequency */
1926 * MSR_EBC_FREQUENCY_ID
1927 * Conservative value valid for even the basic CPU models.
1928 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1929 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1930 * and 266MHz for model 3, or 4. Set Core Clock
1931 * Frequency to System Bus Frequency Ratio to 1 (bits
1932 * 31:24) even though these are only valid for CPU
1933 * models > 2, however guests may end up dividing or
1934 * multiplying by zero otherwise.
1936 case MSR_EBC_FREQUENCY_ID:
1939 case MSR_IA32_APICBASE:
1940 data = kvm_get_apic_base(vcpu);
1942 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1943 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1945 case MSR_IA32_TSCDEADLINE:
1946 data = kvm_get_lapic_tscdeadline_msr(vcpu);
1948 case MSR_IA32_MISC_ENABLE:
1949 data = vcpu->arch.ia32_misc_enable_msr;
1951 case MSR_IA32_PERF_STATUS:
1952 /* TSC increment by tick */
1954 /* CPU multiplier */
1955 data |= (((uint64_t)4ULL) << 40);
1958 data = vcpu->arch.efer;
1960 case MSR_KVM_WALL_CLOCK:
1961 case MSR_KVM_WALL_CLOCK_NEW:
1962 data = vcpu->kvm->arch.wall_clock;
1964 case MSR_KVM_SYSTEM_TIME:
1965 case MSR_KVM_SYSTEM_TIME_NEW:
1966 data = vcpu->arch.time;
1968 case MSR_KVM_ASYNC_PF_EN:
1969 data = vcpu->arch.apf.msr_val;
1971 case MSR_KVM_STEAL_TIME:
1972 data = vcpu->arch.st.msr_val;
1974 case MSR_IA32_P5_MC_ADDR:
1975 case MSR_IA32_P5_MC_TYPE:
1976 case MSR_IA32_MCG_CAP:
1977 case MSR_IA32_MCG_CTL:
1978 case MSR_IA32_MCG_STATUS:
1979 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1980 return get_msr_mce(vcpu, msr, pdata);
1981 case MSR_K7_CLK_CTL:
1983 * Provide expected ramp-up count for K7. All other
1984 * are set to zero, indicating minimum divisors for
1987 * This prevents guest kernels on AMD host with CPU
1988 * type 6, model 8 and higher from exploding due to
1989 * the rdmsr failing.
1993 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1994 if (kvm_hv_msr_partition_wide(msr)) {
1996 mutex_lock(&vcpu->kvm->lock);
1997 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1998 mutex_unlock(&vcpu->kvm->lock);
2001 return get_msr_hyperv(vcpu, msr, pdata);
2003 case MSR_IA32_BBL_CR_CTL3:
2004 /* This legacy MSR exists but isn't fully documented in current
2005 * silicon. It is however accessed by winxp in very narrow
2006 * scenarios where it sets bit #19, itself documented as
2007 * a "reserved" bit. Best effort attempt to source coherent
2008 * read data here should the balance of the register be
2009 * interpreted by the guest:
2011 * L2 cache control register 3: 64GB range, 256KB size,
2012 * enabled, latency 0x1, configured
2016 case MSR_AMD64_OSVW_ID_LENGTH:
2017 if (!guest_cpuid_has_osvw(vcpu))
2019 data = vcpu->arch.osvw.length;
2021 case MSR_AMD64_OSVW_STATUS:
2022 if (!guest_cpuid_has_osvw(vcpu))
2024 data = vcpu->arch.osvw.status;
2027 if (kvm_pmu_msr(vcpu, msr))
2028 return kvm_pmu_get_msr(vcpu, msr, pdata);
2030 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2033 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2041 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2044 * Read or write a bunch of msrs. All parameters are kernel addresses.
2046 * @return number of msrs set successfully.
2048 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2049 struct kvm_msr_entry *entries,
2050 int (*do_msr)(struct kvm_vcpu *vcpu,
2051 unsigned index, u64 *data))
2055 idx = srcu_read_lock(&vcpu->kvm->srcu);
2056 for (i = 0; i < msrs->nmsrs; ++i)
2057 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2059 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2065 * Read or write a bunch of msrs. Parameters are user addresses.
2067 * @return number of msrs set successfully.
2069 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2070 int (*do_msr)(struct kvm_vcpu *vcpu,
2071 unsigned index, u64 *data),
2074 struct kvm_msrs msrs;
2075 struct kvm_msr_entry *entries;
2080 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2084 if (msrs.nmsrs >= MAX_IO_MSRS)
2087 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2088 entries = memdup_user(user_msrs->entries, size);
2089 if (IS_ERR(entries)) {
2090 r = PTR_ERR(entries);
2094 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2099 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2110 int kvm_dev_ioctl_check_extension(long ext)
2115 case KVM_CAP_IRQCHIP:
2117 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2118 case KVM_CAP_SET_TSS_ADDR:
2119 case KVM_CAP_EXT_CPUID:
2120 case KVM_CAP_CLOCKSOURCE:
2122 case KVM_CAP_NOP_IO_DELAY:
2123 case KVM_CAP_MP_STATE:
2124 case KVM_CAP_SYNC_MMU:
2125 case KVM_CAP_USER_NMI:
2126 case KVM_CAP_REINJECT_CONTROL:
2127 case KVM_CAP_IRQ_INJECT_STATUS:
2128 case KVM_CAP_ASSIGN_DEV_IRQ:
2130 case KVM_CAP_IOEVENTFD:
2132 case KVM_CAP_PIT_STATE2:
2133 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2134 case KVM_CAP_XEN_HVM:
2135 case KVM_CAP_ADJUST_CLOCK:
2136 case KVM_CAP_VCPU_EVENTS:
2137 case KVM_CAP_HYPERV:
2138 case KVM_CAP_HYPERV_VAPIC:
2139 case KVM_CAP_HYPERV_SPIN:
2140 case KVM_CAP_PCI_SEGMENT:
2141 case KVM_CAP_DEBUGREGS:
2142 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2144 case KVM_CAP_ASYNC_PF:
2145 case KVM_CAP_GET_TSC_KHZ:
2148 case KVM_CAP_COALESCED_MMIO:
2149 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2152 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2154 case KVM_CAP_NR_VCPUS:
2155 r = KVM_SOFT_MAX_VCPUS;
2157 case KVM_CAP_MAX_VCPUS:
2160 case KVM_CAP_NR_MEMSLOTS:
2161 r = KVM_MEMORY_SLOTS;
2163 case KVM_CAP_PV_MMU: /* obsolete */
2167 r = iommu_present(&pci_bus_type);
2170 r = KVM_MAX_MCE_BANKS;
2175 case KVM_CAP_TSC_CONTROL:
2176 r = kvm_has_tsc_control;
2178 case KVM_CAP_TSC_DEADLINE_TIMER:
2179 r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2189 long kvm_arch_dev_ioctl(struct file *filp,
2190 unsigned int ioctl, unsigned long arg)
2192 void __user *argp = (void __user *)arg;
2196 case KVM_GET_MSR_INDEX_LIST: {
2197 struct kvm_msr_list __user *user_msr_list = argp;
2198 struct kvm_msr_list msr_list;
2202 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2205 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2206 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2209 if (n < msr_list.nmsrs)
2212 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2213 num_msrs_to_save * sizeof(u32)))
2215 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2217 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2222 case KVM_GET_SUPPORTED_CPUID: {
2223 struct kvm_cpuid2 __user *cpuid_arg = argp;
2224 struct kvm_cpuid2 cpuid;
2227 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2229 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2230 cpuid_arg->entries);
2235 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2240 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2243 mce_cap = KVM_MCE_CAP_SUPPORTED;
2245 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2257 static void wbinvd_ipi(void *garbage)
2262 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2264 return vcpu->kvm->arch.iommu_domain &&
2265 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2268 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2270 /* Address WBINVD may be executed by guest */
2271 if (need_emulate_wbinvd(vcpu)) {
2272 if (kvm_x86_ops->has_wbinvd_exit())
2273 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2274 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2275 smp_call_function_single(vcpu->cpu,
2276 wbinvd_ipi, NULL, 1);
2279 kvm_x86_ops->vcpu_load(vcpu, cpu);
2281 /* Apply any externally detected TSC adjustments (due to suspend) */
2282 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2283 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2284 vcpu->arch.tsc_offset_adjustment = 0;
2285 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
2288 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2289 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2290 native_read_tsc() - vcpu->arch.last_host_tsc;
2292 mark_tsc_unstable("KVM discovered backwards TSC");
2293 if (check_tsc_unstable()) {
2294 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2295 vcpu->arch.last_guest_tsc);
2296 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2297 vcpu->arch.tsc_catchup = 1;
2299 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2300 if (vcpu->cpu != cpu)
2301 kvm_migrate_timers(vcpu);
2305 accumulate_steal_time(vcpu);
2306 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2309 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2311 kvm_x86_ops->vcpu_put(vcpu);
2312 kvm_put_guest_fpu(vcpu);
2313 vcpu->arch.last_host_tsc = native_read_tsc();
2316 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2317 struct kvm_lapic_state *s)
2319 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2324 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2325 struct kvm_lapic_state *s)
2327 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2328 kvm_apic_post_state_restore(vcpu);
2329 update_cr8_intercept(vcpu);
2334 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2335 struct kvm_interrupt *irq)
2337 if (irq->irq < 0 || irq->irq >= 256)
2339 if (irqchip_in_kernel(vcpu->kvm))
2342 kvm_queue_interrupt(vcpu, irq->irq, false);
2343 kvm_make_request(KVM_REQ_EVENT, vcpu);
2348 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2350 kvm_inject_nmi(vcpu);
2355 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2356 struct kvm_tpr_access_ctl *tac)
2360 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2364 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2368 unsigned bank_num = mcg_cap & 0xff, bank;
2371 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2373 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2376 vcpu->arch.mcg_cap = mcg_cap;
2377 /* Init IA32_MCG_CTL to all 1s */
2378 if (mcg_cap & MCG_CTL_P)
2379 vcpu->arch.mcg_ctl = ~(u64)0;
2380 /* Init IA32_MCi_CTL to all 1s */
2381 for (bank = 0; bank < bank_num; bank++)
2382 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2387 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2388 struct kvm_x86_mce *mce)
2390 u64 mcg_cap = vcpu->arch.mcg_cap;
2391 unsigned bank_num = mcg_cap & 0xff;
2392 u64 *banks = vcpu->arch.mce_banks;
2394 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2397 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2398 * reporting is disabled
2400 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2401 vcpu->arch.mcg_ctl != ~(u64)0)
2403 banks += 4 * mce->bank;
2405 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2406 * reporting is disabled for the bank
2408 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2410 if (mce->status & MCI_STATUS_UC) {
2411 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2412 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2413 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2416 if (banks[1] & MCI_STATUS_VAL)
2417 mce->status |= MCI_STATUS_OVER;
2418 banks[2] = mce->addr;
2419 banks[3] = mce->misc;
2420 vcpu->arch.mcg_status = mce->mcg_status;
2421 banks[1] = mce->status;
2422 kvm_queue_exception(vcpu, MC_VECTOR);
2423 } else if (!(banks[1] & MCI_STATUS_VAL)
2424 || !(banks[1] & MCI_STATUS_UC)) {
2425 if (banks[1] & MCI_STATUS_VAL)
2426 mce->status |= MCI_STATUS_OVER;
2427 banks[2] = mce->addr;
2428 banks[3] = mce->misc;
2429 banks[1] = mce->status;
2431 banks[1] |= MCI_STATUS_OVER;
2435 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2436 struct kvm_vcpu_events *events)
2439 events->exception.injected =
2440 vcpu->arch.exception.pending &&
2441 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2442 events->exception.nr = vcpu->arch.exception.nr;
2443 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2444 events->exception.pad = 0;
2445 events->exception.error_code = vcpu->arch.exception.error_code;
2447 events->interrupt.injected =
2448 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2449 events->interrupt.nr = vcpu->arch.interrupt.nr;
2450 events->interrupt.soft = 0;
2451 events->interrupt.shadow =
2452 kvm_x86_ops->get_interrupt_shadow(vcpu,
2453 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2455 events->nmi.injected = vcpu->arch.nmi_injected;
2456 events->nmi.pending = vcpu->arch.nmi_pending != 0;
2457 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2458 events->nmi.pad = 0;
2460 events->sipi_vector = vcpu->arch.sipi_vector;
2462 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2463 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2464 | KVM_VCPUEVENT_VALID_SHADOW);
2465 memset(&events->reserved, 0, sizeof(events->reserved));
2468 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2469 struct kvm_vcpu_events *events)
2471 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2472 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2473 | KVM_VCPUEVENT_VALID_SHADOW))
2477 vcpu->arch.exception.pending = events->exception.injected;
2478 vcpu->arch.exception.nr = events->exception.nr;
2479 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2480 vcpu->arch.exception.error_code = events->exception.error_code;
2482 vcpu->arch.interrupt.pending = events->interrupt.injected;
2483 vcpu->arch.interrupt.nr = events->interrupt.nr;
2484 vcpu->arch.interrupt.soft = events->interrupt.soft;
2485 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2486 kvm_x86_ops->set_interrupt_shadow(vcpu,
2487 events->interrupt.shadow);
2489 vcpu->arch.nmi_injected = events->nmi.injected;
2490 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2491 vcpu->arch.nmi_pending = events->nmi.pending;
2492 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2494 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2495 vcpu->arch.sipi_vector = events->sipi_vector;
2497 kvm_make_request(KVM_REQ_EVENT, vcpu);
2502 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2503 struct kvm_debugregs *dbgregs)
2505 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2506 dbgregs->dr6 = vcpu->arch.dr6;
2507 dbgregs->dr7 = vcpu->arch.dr7;
2509 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2512 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2513 struct kvm_debugregs *dbgregs)
2518 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2519 vcpu->arch.dr6 = dbgregs->dr6;
2520 vcpu->arch.dr7 = dbgregs->dr7;
2525 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2526 struct kvm_xsave *guest_xsave)
2529 memcpy(guest_xsave->region,
2530 &vcpu->arch.guest_fpu.state->xsave,
2533 memcpy(guest_xsave->region,
2534 &vcpu->arch.guest_fpu.state->fxsave,
2535 sizeof(struct i387_fxsave_struct));
2536 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2541 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2542 struct kvm_xsave *guest_xsave)
2545 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2548 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2549 guest_xsave->region, xstate_size);
2551 if (xstate_bv & ~XSTATE_FPSSE)
2553 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2554 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2559 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2560 struct kvm_xcrs *guest_xcrs)
2562 if (!cpu_has_xsave) {
2563 guest_xcrs->nr_xcrs = 0;
2567 guest_xcrs->nr_xcrs = 1;
2568 guest_xcrs->flags = 0;
2569 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2570 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2573 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2574 struct kvm_xcrs *guest_xcrs)
2581 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2584 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2585 /* Only support XCR0 currently */
2586 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2587 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2588 guest_xcrs->xcrs[0].value);
2596 long kvm_arch_vcpu_ioctl(struct file *filp,
2597 unsigned int ioctl, unsigned long arg)
2599 struct kvm_vcpu *vcpu = filp->private_data;
2600 void __user *argp = (void __user *)arg;
2603 struct kvm_lapic_state *lapic;
2604 struct kvm_xsave *xsave;
2605 struct kvm_xcrs *xcrs;
2611 case KVM_GET_LAPIC: {
2613 if (!vcpu->arch.apic)
2615 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2620 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
2624 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
2629 case KVM_SET_LAPIC: {
2631 if (!vcpu->arch.apic)
2633 u.lapic = memdup_user(argp, sizeof(*u.lapic));
2634 if (IS_ERR(u.lapic)) {
2635 r = PTR_ERR(u.lapic);
2639 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
2645 case KVM_INTERRUPT: {
2646 struct kvm_interrupt irq;
2649 if (copy_from_user(&irq, argp, sizeof irq))
2651 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2658 r = kvm_vcpu_ioctl_nmi(vcpu);
2664 case KVM_SET_CPUID: {
2665 struct kvm_cpuid __user *cpuid_arg = argp;
2666 struct kvm_cpuid cpuid;
2669 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2671 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2676 case KVM_SET_CPUID2: {
2677 struct kvm_cpuid2 __user *cpuid_arg = argp;
2678 struct kvm_cpuid2 cpuid;
2681 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2683 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2684 cpuid_arg->entries);
2689 case KVM_GET_CPUID2: {
2690 struct kvm_cpuid2 __user *cpuid_arg = argp;
2691 struct kvm_cpuid2 cpuid;
2694 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2696 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2697 cpuid_arg->entries);
2701 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2707 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2710 r = msr_io(vcpu, argp, do_set_msr, 0);
2712 case KVM_TPR_ACCESS_REPORTING: {
2713 struct kvm_tpr_access_ctl tac;
2716 if (copy_from_user(&tac, argp, sizeof tac))
2718 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2722 if (copy_to_user(argp, &tac, sizeof tac))
2727 case KVM_SET_VAPIC_ADDR: {
2728 struct kvm_vapic_addr va;
2731 if (!irqchip_in_kernel(vcpu->kvm))
2734 if (copy_from_user(&va, argp, sizeof va))
2737 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2740 case KVM_X86_SETUP_MCE: {
2744 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2746 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2749 case KVM_X86_SET_MCE: {
2750 struct kvm_x86_mce mce;
2753 if (copy_from_user(&mce, argp, sizeof mce))
2755 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2758 case KVM_GET_VCPU_EVENTS: {
2759 struct kvm_vcpu_events events;
2761 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2764 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2769 case KVM_SET_VCPU_EVENTS: {
2770 struct kvm_vcpu_events events;
2773 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2776 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2779 case KVM_GET_DEBUGREGS: {
2780 struct kvm_debugregs dbgregs;
2782 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2785 if (copy_to_user(argp, &dbgregs,
2786 sizeof(struct kvm_debugregs)))
2791 case KVM_SET_DEBUGREGS: {
2792 struct kvm_debugregs dbgregs;
2795 if (copy_from_user(&dbgregs, argp,
2796 sizeof(struct kvm_debugregs)))
2799 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2802 case KVM_GET_XSAVE: {
2803 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2808 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
2811 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
2816 case KVM_SET_XSAVE: {
2817 u.xsave = memdup_user(argp, sizeof(*u.xsave));
2818 if (IS_ERR(u.xsave)) {
2819 r = PTR_ERR(u.xsave);
2823 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
2826 case KVM_GET_XCRS: {
2827 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2832 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
2835 if (copy_to_user(argp, u.xcrs,
2836 sizeof(struct kvm_xcrs)))
2841 case KVM_SET_XCRS: {
2842 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
2843 if (IS_ERR(u.xcrs)) {
2844 r = PTR_ERR(u.xcrs);
2848 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
2851 case KVM_SET_TSC_KHZ: {
2855 user_tsc_khz = (u32)arg;
2857 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
2860 if (user_tsc_khz == 0)
2861 user_tsc_khz = tsc_khz;
2863 kvm_set_tsc_khz(vcpu, user_tsc_khz);
2868 case KVM_GET_TSC_KHZ: {
2869 r = vcpu->arch.virtual_tsc_khz;
2880 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
2882 return VM_FAULT_SIGBUS;
2885 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2889 if (addr > (unsigned int)(-3 * PAGE_SIZE))
2891 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2895 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2898 kvm->arch.ept_identity_map_addr = ident_addr;
2902 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2903 u32 kvm_nr_mmu_pages)
2905 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2908 mutex_lock(&kvm->slots_lock);
2909 spin_lock(&kvm->mmu_lock);
2911 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2912 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2914 spin_unlock(&kvm->mmu_lock);
2915 mutex_unlock(&kvm->slots_lock);
2919 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2921 return kvm->arch.n_max_mmu_pages;
2924 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2929 switch (chip->chip_id) {
2930 case KVM_IRQCHIP_PIC_MASTER:
2931 memcpy(&chip->chip.pic,
2932 &pic_irqchip(kvm)->pics[0],
2933 sizeof(struct kvm_pic_state));
2935 case KVM_IRQCHIP_PIC_SLAVE:
2936 memcpy(&chip->chip.pic,
2937 &pic_irqchip(kvm)->pics[1],
2938 sizeof(struct kvm_pic_state));
2940 case KVM_IRQCHIP_IOAPIC:
2941 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2950 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2955 switch (chip->chip_id) {
2956 case KVM_IRQCHIP_PIC_MASTER:
2957 spin_lock(&pic_irqchip(kvm)->lock);
2958 memcpy(&pic_irqchip(kvm)->pics[0],
2960 sizeof(struct kvm_pic_state));
2961 spin_unlock(&pic_irqchip(kvm)->lock);
2963 case KVM_IRQCHIP_PIC_SLAVE:
2964 spin_lock(&pic_irqchip(kvm)->lock);
2965 memcpy(&pic_irqchip(kvm)->pics[1],
2967 sizeof(struct kvm_pic_state));
2968 spin_unlock(&pic_irqchip(kvm)->lock);
2970 case KVM_IRQCHIP_IOAPIC:
2971 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2977 kvm_pic_update_irq(pic_irqchip(kvm));
2981 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2985 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2986 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2987 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2991 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2995 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2996 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2997 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2998 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3002 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3006 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3007 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3008 sizeof(ps->channels));
3009 ps->flags = kvm->arch.vpit->pit_state.flags;
3010 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3011 memset(&ps->reserved, 0, sizeof(ps->reserved));
3015 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3017 int r = 0, start = 0;
3018 u32 prev_legacy, cur_legacy;
3019 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3020 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3021 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3022 if (!prev_legacy && cur_legacy)
3024 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3025 sizeof(kvm->arch.vpit->pit_state.channels));
3026 kvm->arch.vpit->pit_state.flags = ps->flags;
3027 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3028 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3032 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3033 struct kvm_reinject_control *control)
3035 if (!kvm->arch.vpit)
3037 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3038 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3039 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3044 * write_protect_slot - write protect a slot for dirty logging
3045 * @kvm: the kvm instance
3046 * @memslot: the slot we protect
3047 * @dirty_bitmap: the bitmap indicating which pages are dirty
3048 * @nr_dirty_pages: the number of dirty pages
3050 * We have two ways to find all sptes to protect:
3051 * 1. Use kvm_mmu_slot_remove_write_access() which walks all shadow pages and
3052 * checks ones that have a spte mapping a page in the slot.
3053 * 2. Use kvm_mmu_rmap_write_protect() for each gfn found in the bitmap.
3055 * Generally speaking, if there are not so many dirty pages compared to the
3056 * number of shadow pages, we should use the latter.
3058 * Note that letting others write into a page marked dirty in the old bitmap
3059 * by using the remaining tlb entry is not a problem. That page will become
3060 * write protected again when we flush the tlb and then be reported dirty to
3061 * the user space by copying the old bitmap.
3063 static void write_protect_slot(struct kvm *kvm,
3064 struct kvm_memory_slot *memslot,
3065 unsigned long *dirty_bitmap,
3066 unsigned long nr_dirty_pages)
3068 spin_lock(&kvm->mmu_lock);
3070 /* Not many dirty pages compared to # of shadow pages. */
3071 if (nr_dirty_pages < kvm->arch.n_used_mmu_pages) {
3072 unsigned long gfn_offset;
3074 for_each_set_bit(gfn_offset, dirty_bitmap, memslot->npages) {
3075 unsigned long gfn = memslot->base_gfn + gfn_offset;
3077 kvm_mmu_rmap_write_protect(kvm, gfn, memslot);
3079 kvm_flush_remote_tlbs(kvm);
3081 kvm_mmu_slot_remove_write_access(kvm, memslot->id);
3083 spin_unlock(&kvm->mmu_lock);
3087 * Get (and clear) the dirty memory log for a memory slot.
3089 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3090 struct kvm_dirty_log *log)
3093 struct kvm_memory_slot *memslot;
3094 unsigned long n, nr_dirty_pages;
3096 mutex_lock(&kvm->slots_lock);
3099 if (log->slot >= KVM_MEMORY_SLOTS)
3102 memslot = id_to_memslot(kvm->memslots, log->slot);
3104 if (!memslot->dirty_bitmap)
3107 n = kvm_dirty_bitmap_bytes(memslot);
3108 nr_dirty_pages = memslot->nr_dirty_pages;
3110 /* If nothing is dirty, don't bother messing with page tables. */
3111 if (nr_dirty_pages) {
3112 struct kvm_memslots *slots, *old_slots;
3113 unsigned long *dirty_bitmap, *dirty_bitmap_head;
3115 dirty_bitmap = memslot->dirty_bitmap;
3116 dirty_bitmap_head = memslot->dirty_bitmap_head;
3117 if (dirty_bitmap == dirty_bitmap_head)
3118 dirty_bitmap_head += n / sizeof(long);
3119 memset(dirty_bitmap_head, 0, n);
3122 slots = kmemdup(kvm->memslots, sizeof(*kvm->memslots), GFP_KERNEL);
3126 memslot = id_to_memslot(slots, log->slot);
3127 memslot->nr_dirty_pages = 0;
3128 memslot->dirty_bitmap = dirty_bitmap_head;
3129 update_memslots(slots, NULL);
3131 old_slots = kvm->memslots;
3132 rcu_assign_pointer(kvm->memslots, slots);
3133 synchronize_srcu_expedited(&kvm->srcu);
3136 write_protect_slot(kvm, memslot, dirty_bitmap, nr_dirty_pages);
3139 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
3143 if (clear_user(log->dirty_bitmap, n))
3149 mutex_unlock(&kvm->slots_lock);
3153 long kvm_arch_vm_ioctl(struct file *filp,
3154 unsigned int ioctl, unsigned long arg)
3156 struct kvm *kvm = filp->private_data;
3157 void __user *argp = (void __user *)arg;
3160 * This union makes it completely explicit to gcc-3.x
3161 * that these two variables' stack usage should be
3162 * combined, not added together.
3165 struct kvm_pit_state ps;
3166 struct kvm_pit_state2 ps2;
3167 struct kvm_pit_config pit_config;
3171 case KVM_SET_TSS_ADDR:
3172 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3176 case KVM_SET_IDENTITY_MAP_ADDR: {
3180 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3182 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3187 case KVM_SET_NR_MMU_PAGES:
3188 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3192 case KVM_GET_NR_MMU_PAGES:
3193 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3195 case KVM_CREATE_IRQCHIP: {
3196 struct kvm_pic *vpic;
3198 mutex_lock(&kvm->lock);
3201 goto create_irqchip_unlock;
3203 vpic = kvm_create_pic(kvm);
3205 r = kvm_ioapic_init(kvm);
3207 mutex_lock(&kvm->slots_lock);
3208 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3210 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3212 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3214 mutex_unlock(&kvm->slots_lock);
3216 goto create_irqchip_unlock;
3219 goto create_irqchip_unlock;
3221 kvm->arch.vpic = vpic;
3223 r = kvm_setup_default_irq_routing(kvm);
3225 mutex_lock(&kvm->slots_lock);
3226 mutex_lock(&kvm->irq_lock);
3227 kvm_ioapic_destroy(kvm);
3228 kvm_destroy_pic(kvm);
3229 mutex_unlock(&kvm->irq_lock);
3230 mutex_unlock(&kvm->slots_lock);
3232 create_irqchip_unlock:
3233 mutex_unlock(&kvm->lock);
3236 case KVM_CREATE_PIT:
3237 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3239 case KVM_CREATE_PIT2:
3241 if (copy_from_user(&u.pit_config, argp,
3242 sizeof(struct kvm_pit_config)))
3245 mutex_lock(&kvm->slots_lock);
3248 goto create_pit_unlock;
3250 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3254 mutex_unlock(&kvm->slots_lock);
3256 case KVM_IRQ_LINE_STATUS:
3257 case KVM_IRQ_LINE: {
3258 struct kvm_irq_level irq_event;
3261 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3264 if (irqchip_in_kernel(kvm)) {
3266 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3267 irq_event.irq, irq_event.level);
3268 if (ioctl == KVM_IRQ_LINE_STATUS) {
3270 irq_event.status = status;
3271 if (copy_to_user(argp, &irq_event,
3279 case KVM_GET_IRQCHIP: {
3280 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3281 struct kvm_irqchip *chip;
3283 chip = memdup_user(argp, sizeof(*chip));
3290 if (!irqchip_in_kernel(kvm))
3291 goto get_irqchip_out;
3292 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3294 goto get_irqchip_out;
3296 if (copy_to_user(argp, chip, sizeof *chip))
3297 goto get_irqchip_out;
3305 case KVM_SET_IRQCHIP: {
3306 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3307 struct kvm_irqchip *chip;
3309 chip = memdup_user(argp, sizeof(*chip));
3316 if (!irqchip_in_kernel(kvm))
3317 goto set_irqchip_out;
3318 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3320 goto set_irqchip_out;
3330 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3333 if (!kvm->arch.vpit)
3335 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3339 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3346 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3349 if (!kvm->arch.vpit)
3351 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3357 case KVM_GET_PIT2: {
3359 if (!kvm->arch.vpit)
3361 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3365 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3370 case KVM_SET_PIT2: {
3372 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3375 if (!kvm->arch.vpit)
3377 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3383 case KVM_REINJECT_CONTROL: {
3384 struct kvm_reinject_control control;
3386 if (copy_from_user(&control, argp, sizeof(control)))
3388 r = kvm_vm_ioctl_reinject(kvm, &control);
3394 case KVM_XEN_HVM_CONFIG: {
3396 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3397 sizeof(struct kvm_xen_hvm_config)))
3400 if (kvm->arch.xen_hvm_config.flags)
3405 case KVM_SET_CLOCK: {
3406 struct kvm_clock_data user_ns;
3411 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3419 local_irq_disable();
3420 now_ns = get_kernel_ns();
3421 delta = user_ns.clock - now_ns;
3423 kvm->arch.kvmclock_offset = delta;
3426 case KVM_GET_CLOCK: {
3427 struct kvm_clock_data user_ns;
3430 local_irq_disable();
3431 now_ns = get_kernel_ns();
3432 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3435 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3438 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3451 static void kvm_init_msr_list(void)
3456 /* skip the first msrs in the list. KVM-specific */
3457 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3458 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3461 msrs_to_save[j] = msrs_to_save[i];
3464 num_msrs_to_save = j;
3467 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3475 if (!(vcpu->arch.apic &&
3476 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3477 && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3488 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3495 if (!(vcpu->arch.apic &&
3496 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3497 && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3499 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3509 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3510 struct kvm_segment *var, int seg)
3512 kvm_x86_ops->set_segment(vcpu, var, seg);
3515 void kvm_get_segment(struct kvm_vcpu *vcpu,
3516 struct kvm_segment *var, int seg)
3518 kvm_x86_ops->get_segment(vcpu, var, seg);
3521 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3524 struct x86_exception exception;
3526 BUG_ON(!mmu_is_nested(vcpu));
3528 /* NPT walks are always user-walks */
3529 access |= PFERR_USER_MASK;
3530 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3535 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3536 struct x86_exception *exception)
3538 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3539 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3542 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3543 struct x86_exception *exception)
3545 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3546 access |= PFERR_FETCH_MASK;
3547 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3550 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3551 struct x86_exception *exception)
3553 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3554 access |= PFERR_WRITE_MASK;
3555 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3558 /* uses this to access any guest's mapped memory without checking CPL */
3559 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3560 struct x86_exception *exception)
3562 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3565 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3566 struct kvm_vcpu *vcpu, u32 access,
3567 struct x86_exception *exception)
3570 int r = X86EMUL_CONTINUE;
3573 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3575 unsigned offset = addr & (PAGE_SIZE-1);
3576 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3579 if (gpa == UNMAPPED_GVA)
3580 return X86EMUL_PROPAGATE_FAULT;
3581 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3583 r = X86EMUL_IO_NEEDED;
3595 /* used for instruction fetching */
3596 static int kvm_fetch_guest_virt(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 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3603 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3604 access | PFERR_FETCH_MASK,
3608 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3609 gva_t addr, void *val, unsigned int bytes,
3610 struct x86_exception *exception)
3612 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3613 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3615 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3618 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
3620 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3621 gva_t addr, void *val, unsigned int bytes,
3622 struct x86_exception *exception)
3624 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3625 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3628 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3629 gva_t addr, void *val,
3631 struct x86_exception *exception)
3633 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3635 int r = X86EMUL_CONTINUE;
3638 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3641 unsigned offset = addr & (PAGE_SIZE-1);
3642 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3645 if (gpa == UNMAPPED_GVA)
3646 return X86EMUL_PROPAGATE_FAULT;
3647 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3649 r = X86EMUL_IO_NEEDED;
3660 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
3662 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
3663 gpa_t *gpa, struct x86_exception *exception,
3666 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3668 if (vcpu_match_mmio_gva(vcpu, gva) &&
3669 check_write_user_access(vcpu, write, access,
3670 vcpu->arch.access)) {
3671 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
3672 (gva & (PAGE_SIZE - 1));
3673 trace_vcpu_match_mmio(gva, *gpa, write, false);
3678 access |= PFERR_WRITE_MASK;
3680 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3682 if (*gpa == UNMAPPED_GVA)
3685 /* For APIC access vmexit */
3686 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3689 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
3690 trace_vcpu_match_mmio(gva, *gpa, write, true);
3697 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3698 const void *val, int bytes)
3702 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3705 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
3709 struct read_write_emulator_ops {
3710 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
3712 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
3713 void *val, int bytes);
3714 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
3715 int bytes, void *val);
3716 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
3717 void *val, int bytes);
3721 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
3723 if (vcpu->mmio_read_completed) {
3724 memcpy(val, vcpu->mmio_data, bytes);
3725 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3726 vcpu->mmio_phys_addr, *(u64 *)val);
3727 vcpu->mmio_read_completed = 0;
3734 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
3735 void *val, int bytes)
3737 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
3740 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
3741 void *val, int bytes)
3743 return emulator_write_phys(vcpu, gpa, val, bytes);
3746 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
3748 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3749 return vcpu_mmio_write(vcpu, gpa, bytes, val);
3752 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
3753 void *val, int bytes)
3755 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3756 return X86EMUL_IO_NEEDED;
3759 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
3760 void *val, int bytes)
3762 memcpy(vcpu->mmio_data, val, bytes);
3763 memcpy(vcpu->run->mmio.data, vcpu->mmio_data, 8);
3764 return X86EMUL_CONTINUE;
3767 static struct read_write_emulator_ops read_emultor = {
3768 .read_write_prepare = read_prepare,
3769 .read_write_emulate = read_emulate,
3770 .read_write_mmio = vcpu_mmio_read,
3771 .read_write_exit_mmio = read_exit_mmio,
3774 static struct read_write_emulator_ops write_emultor = {
3775 .read_write_emulate = write_emulate,
3776 .read_write_mmio = write_mmio,
3777 .read_write_exit_mmio = write_exit_mmio,
3781 static int emulator_read_write_onepage(unsigned long addr, void *val,
3783 struct x86_exception *exception,
3784 struct kvm_vcpu *vcpu,
3785 struct read_write_emulator_ops *ops)
3789 bool write = ops->write;
3791 if (ops->read_write_prepare &&
3792 ops->read_write_prepare(vcpu, val, bytes))
3793 return X86EMUL_CONTINUE;
3795 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
3798 return X86EMUL_PROPAGATE_FAULT;
3800 /* For APIC access vmexit */
3804 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
3805 return X86EMUL_CONTINUE;
3809 * Is this MMIO handled locally?
3811 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
3812 if (handled == bytes)
3813 return X86EMUL_CONTINUE;
3819 vcpu->mmio_needed = 1;
3820 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3821 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3822 vcpu->mmio_size = bytes;
3823 vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
3824 vcpu->run->mmio.is_write = vcpu->mmio_is_write = write;
3825 vcpu->mmio_index = 0;
3827 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
3830 int emulator_read_write(struct x86_emulate_ctxt *ctxt, unsigned long addr,
3831 void *val, unsigned int bytes,
3832 struct x86_exception *exception,
3833 struct read_write_emulator_ops *ops)
3835 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3837 /* Crossing a page boundary? */
3838 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3841 now = -addr & ~PAGE_MASK;
3842 rc = emulator_read_write_onepage(addr, val, now, exception,
3845 if (rc != X86EMUL_CONTINUE)
3852 return emulator_read_write_onepage(addr, val, bytes, exception,
3856 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
3860 struct x86_exception *exception)
3862 return emulator_read_write(ctxt, addr, val, bytes,
3863 exception, &read_emultor);
3866 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
3870 struct x86_exception *exception)
3872 return emulator_read_write(ctxt, addr, (void *)val, bytes,
3873 exception, &write_emultor);
3876 #define CMPXCHG_TYPE(t, ptr, old, new) \
3877 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3879 #ifdef CONFIG_X86_64
3880 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3882 # define CMPXCHG64(ptr, old, new) \
3883 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
3886 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
3891 struct x86_exception *exception)
3893 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3899 /* guests cmpxchg8b have to be emulated atomically */
3900 if (bytes > 8 || (bytes & (bytes - 1)))
3903 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3905 if (gpa == UNMAPPED_GVA ||
3906 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3909 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3912 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3913 if (is_error_page(page)) {
3914 kvm_release_page_clean(page);
3918 kaddr = kmap_atomic(page, KM_USER0);
3919 kaddr += offset_in_page(gpa);
3922 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
3925 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
3928 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
3931 exchanged = CMPXCHG64(kaddr, old, new);
3936 kunmap_atomic(kaddr, KM_USER0);
3937 kvm_release_page_dirty(page);
3940 return X86EMUL_CMPXCHG_FAILED;
3942 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
3944 return X86EMUL_CONTINUE;
3947 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3949 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
3952 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3954 /* TODO: String I/O for in kernel device */
3957 if (vcpu->arch.pio.in)
3958 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3959 vcpu->arch.pio.size, pd);
3961 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3962 vcpu->arch.pio.port, vcpu->arch.pio.size,
3967 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
3968 unsigned short port, void *val,
3969 unsigned int count, bool in)
3971 trace_kvm_pio(!in, port, size, count);
3973 vcpu->arch.pio.port = port;
3974 vcpu->arch.pio.in = in;
3975 vcpu->arch.pio.count = count;
3976 vcpu->arch.pio.size = size;
3978 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3979 vcpu->arch.pio.count = 0;
3983 vcpu->run->exit_reason = KVM_EXIT_IO;
3984 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3985 vcpu->run->io.size = size;
3986 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3987 vcpu->run->io.count = count;
3988 vcpu->run->io.port = port;
3993 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
3994 int size, unsigned short port, void *val,
3997 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4000 if (vcpu->arch.pio.count)
4003 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4006 memcpy(val, vcpu->arch.pio_data, size * count);
4007 vcpu->arch.pio.count = 0;
4014 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4015 int size, unsigned short port,
4016 const void *val, unsigned int count)
4018 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4020 memcpy(vcpu->arch.pio_data, val, size * count);
4021 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4024 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4026 return kvm_x86_ops->get_segment_base(vcpu, seg);
4029 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4031 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4034 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4036 if (!need_emulate_wbinvd(vcpu))
4037 return X86EMUL_CONTINUE;
4039 if (kvm_x86_ops->has_wbinvd_exit()) {
4040 int cpu = get_cpu();
4042 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4043 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4044 wbinvd_ipi, NULL, 1);
4046 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4049 return X86EMUL_CONTINUE;
4051 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4053 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4055 kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4058 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4060 return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4063 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4066 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4069 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4071 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4074 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4076 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4077 unsigned long value;
4081 value = kvm_read_cr0(vcpu);
4084 value = vcpu->arch.cr2;
4087 value = kvm_read_cr3(vcpu);
4090 value = kvm_read_cr4(vcpu);
4093 value = kvm_get_cr8(vcpu);
4096 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4103 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4105 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4110 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4113 vcpu->arch.cr2 = val;
4116 res = kvm_set_cr3(vcpu, val);
4119 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4122 res = kvm_set_cr8(vcpu, val);
4125 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4132 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4134 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4137 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4139 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4142 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4144 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4147 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4149 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4152 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4154 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4157 static unsigned long emulator_get_cached_segment_base(
4158 struct x86_emulate_ctxt *ctxt, int seg)
4160 return get_segment_base(emul_to_vcpu(ctxt), seg);
4163 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4164 struct desc_struct *desc, u32 *base3,
4167 struct kvm_segment var;
4169 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4170 *selector = var.selector;
4177 set_desc_limit(desc, var.limit);
4178 set_desc_base(desc, (unsigned long)var.base);
4179 #ifdef CONFIG_X86_64
4181 *base3 = var.base >> 32;
4183 desc->type = var.type;
4185 desc->dpl = var.dpl;
4186 desc->p = var.present;
4187 desc->avl = var.avl;
4195 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4196 struct desc_struct *desc, u32 base3,
4199 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4200 struct kvm_segment var;
4202 var.selector = selector;
4203 var.base = get_desc_base(desc);
4204 #ifdef CONFIG_X86_64
4205 var.base |= ((u64)base3) << 32;
4207 var.limit = get_desc_limit(desc);
4209 var.limit = (var.limit << 12) | 0xfff;
4210 var.type = desc->type;
4211 var.present = desc->p;
4212 var.dpl = desc->dpl;
4217 var.avl = desc->avl;
4218 var.present = desc->p;
4219 var.unusable = !var.present;
4222 kvm_set_segment(vcpu, &var, seg);
4226 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4227 u32 msr_index, u64 *pdata)
4229 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4232 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4233 u32 msr_index, u64 data)
4235 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
4238 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4239 u32 pmc, u64 *pdata)
4241 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4244 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4246 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4249 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4252 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4254 * CR0.TS may reference the host fpu state, not the guest fpu state,
4255 * so it may be clear at this point.
4260 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4265 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4266 struct x86_instruction_info *info,
4267 enum x86_intercept_stage stage)
4269 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4272 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4273 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4275 struct kvm_cpuid_entry2 *cpuid = NULL;
4278 cpuid = kvm_find_cpuid_entry(emul_to_vcpu(ctxt),
4294 static struct x86_emulate_ops emulate_ops = {
4295 .read_std = kvm_read_guest_virt_system,
4296 .write_std = kvm_write_guest_virt_system,
4297 .fetch = kvm_fetch_guest_virt,
4298 .read_emulated = emulator_read_emulated,
4299 .write_emulated = emulator_write_emulated,
4300 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4301 .invlpg = emulator_invlpg,
4302 .pio_in_emulated = emulator_pio_in_emulated,
4303 .pio_out_emulated = emulator_pio_out_emulated,
4304 .get_segment = emulator_get_segment,
4305 .set_segment = emulator_set_segment,
4306 .get_cached_segment_base = emulator_get_cached_segment_base,
4307 .get_gdt = emulator_get_gdt,
4308 .get_idt = emulator_get_idt,
4309 .set_gdt = emulator_set_gdt,
4310 .set_idt = emulator_set_idt,
4311 .get_cr = emulator_get_cr,
4312 .set_cr = emulator_set_cr,
4313 .cpl = emulator_get_cpl,
4314 .get_dr = emulator_get_dr,
4315 .set_dr = emulator_set_dr,
4316 .set_msr = emulator_set_msr,
4317 .get_msr = emulator_get_msr,
4318 .read_pmc = emulator_read_pmc,
4319 .halt = emulator_halt,
4320 .wbinvd = emulator_wbinvd,
4321 .fix_hypercall = emulator_fix_hypercall,
4322 .get_fpu = emulator_get_fpu,
4323 .put_fpu = emulator_put_fpu,
4324 .intercept = emulator_intercept,
4325 .get_cpuid = emulator_get_cpuid,
4328 static void cache_all_regs(struct kvm_vcpu *vcpu)
4330 kvm_register_read(vcpu, VCPU_REGS_RAX);
4331 kvm_register_read(vcpu, VCPU_REGS_RSP);
4332 kvm_register_read(vcpu, VCPU_REGS_RIP);
4333 vcpu->arch.regs_dirty = ~0;
4336 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4338 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4340 * an sti; sti; sequence only disable interrupts for the first
4341 * instruction. So, if the last instruction, be it emulated or
4342 * not, left the system with the INT_STI flag enabled, it
4343 * means that the last instruction is an sti. We should not
4344 * leave the flag on in this case. The same goes for mov ss
4346 if (!(int_shadow & mask))
4347 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4350 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4352 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4353 if (ctxt->exception.vector == PF_VECTOR)
4354 kvm_propagate_fault(vcpu, &ctxt->exception);
4355 else if (ctxt->exception.error_code_valid)
4356 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4357 ctxt->exception.error_code);
4359 kvm_queue_exception(vcpu, ctxt->exception.vector);
4362 static void init_decode_cache(struct x86_emulate_ctxt *ctxt,
4363 const unsigned long *regs)
4365 memset(&ctxt->twobyte, 0,
4366 (void *)&ctxt->regs - (void *)&ctxt->twobyte);
4367 memcpy(ctxt->regs, regs, sizeof(ctxt->regs));
4369 ctxt->fetch.start = 0;
4370 ctxt->fetch.end = 0;
4371 ctxt->io_read.pos = 0;
4372 ctxt->io_read.end = 0;
4373 ctxt->mem_read.pos = 0;
4374 ctxt->mem_read.end = 0;
4377 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4379 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4383 * TODO: fix emulate.c to use guest_read/write_register
4384 * instead of direct ->regs accesses, can save hundred cycles
4385 * on Intel for instructions that don't read/change RSP, for
4388 cache_all_regs(vcpu);
4390 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4392 ctxt->eflags = kvm_get_rflags(vcpu);
4393 ctxt->eip = kvm_rip_read(vcpu);
4394 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4395 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
4396 cs_l ? X86EMUL_MODE_PROT64 :
4397 cs_db ? X86EMUL_MODE_PROT32 :
4398 X86EMUL_MODE_PROT16;
4399 ctxt->guest_mode = is_guest_mode(vcpu);
4401 init_decode_cache(ctxt, vcpu->arch.regs);
4402 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4405 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4407 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4410 init_emulate_ctxt(vcpu);
4414 ctxt->_eip = ctxt->eip + inc_eip;
4415 ret = emulate_int_real(ctxt, irq);
4417 if (ret != X86EMUL_CONTINUE)
4418 return EMULATE_FAIL;
4420 ctxt->eip = ctxt->_eip;
4421 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4422 kvm_rip_write(vcpu, ctxt->eip);
4423 kvm_set_rflags(vcpu, ctxt->eflags);
4425 if (irq == NMI_VECTOR)
4426 vcpu->arch.nmi_pending = 0;
4428 vcpu->arch.interrupt.pending = false;
4430 return EMULATE_DONE;
4432 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4434 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4436 int r = EMULATE_DONE;
4438 ++vcpu->stat.insn_emulation_fail;
4439 trace_kvm_emulate_insn_failed(vcpu);
4440 if (!is_guest_mode(vcpu)) {
4441 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4442 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4443 vcpu->run->internal.ndata = 0;
4446 kvm_queue_exception(vcpu, UD_VECTOR);
4451 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4459 * if emulation was due to access to shadowed page table
4460 * and it failed try to unshadow page and re-entetr the
4461 * guest to let CPU execute the instruction.
4463 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4466 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4468 if (gpa == UNMAPPED_GVA)
4469 return true; /* let cpu generate fault */
4471 if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4477 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
4478 unsigned long cr2, int emulation_type)
4480 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4481 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
4483 last_retry_eip = vcpu->arch.last_retry_eip;
4484 last_retry_addr = vcpu->arch.last_retry_addr;
4487 * If the emulation is caused by #PF and it is non-page_table
4488 * writing instruction, it means the VM-EXIT is caused by shadow
4489 * page protected, we can zap the shadow page and retry this
4490 * instruction directly.
4492 * Note: if the guest uses a non-page-table modifying instruction
4493 * on the PDE that points to the instruction, then we will unmap
4494 * the instruction and go to an infinite loop. So, we cache the
4495 * last retried eip and the last fault address, if we meet the eip
4496 * and the address again, we can break out of the potential infinite
4499 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
4501 if (!(emulation_type & EMULTYPE_RETRY))
4504 if (x86_page_table_writing_insn(ctxt))
4507 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
4510 vcpu->arch.last_retry_eip = ctxt->eip;
4511 vcpu->arch.last_retry_addr = cr2;
4513 if (!vcpu->arch.mmu.direct_map)
4514 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4516 kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4521 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4528 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4529 bool writeback = true;
4531 kvm_clear_exception_queue(vcpu);
4533 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4534 init_emulate_ctxt(vcpu);
4535 ctxt->interruptibility = 0;
4536 ctxt->have_exception = false;
4537 ctxt->perm_ok = false;
4539 ctxt->only_vendor_specific_insn
4540 = emulation_type & EMULTYPE_TRAP_UD;
4542 r = x86_decode_insn(ctxt, insn, insn_len);
4544 trace_kvm_emulate_insn_start(vcpu);
4545 ++vcpu->stat.insn_emulation;
4546 if (r != EMULATION_OK) {
4547 if (emulation_type & EMULTYPE_TRAP_UD)
4548 return EMULATE_FAIL;
4549 if (reexecute_instruction(vcpu, cr2))
4550 return EMULATE_DONE;
4551 if (emulation_type & EMULTYPE_SKIP)
4552 return EMULATE_FAIL;
4553 return handle_emulation_failure(vcpu);
4557 if (emulation_type & EMULTYPE_SKIP) {
4558 kvm_rip_write(vcpu, ctxt->_eip);
4559 return EMULATE_DONE;
4562 if (retry_instruction(ctxt, cr2, emulation_type))
4563 return EMULATE_DONE;
4565 /* this is needed for vmware backdoor interface to work since it
4566 changes registers values during IO operation */
4567 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4568 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4569 memcpy(ctxt->regs, vcpu->arch.regs, sizeof ctxt->regs);
4573 r = x86_emulate_insn(ctxt);
4575 if (r == EMULATION_INTERCEPTED)
4576 return EMULATE_DONE;
4578 if (r == EMULATION_FAILED) {
4579 if (reexecute_instruction(vcpu, cr2))
4580 return EMULATE_DONE;
4582 return handle_emulation_failure(vcpu);
4585 if (ctxt->have_exception) {
4586 inject_emulated_exception(vcpu);
4588 } else if (vcpu->arch.pio.count) {
4589 if (!vcpu->arch.pio.in)
4590 vcpu->arch.pio.count = 0;
4593 r = EMULATE_DO_MMIO;
4594 } else if (vcpu->mmio_needed) {
4595 if (!vcpu->mmio_is_write)
4597 r = EMULATE_DO_MMIO;
4598 } else if (r == EMULATION_RESTART)
4604 toggle_interruptibility(vcpu, ctxt->interruptibility);
4605 kvm_set_rflags(vcpu, ctxt->eflags);
4606 kvm_make_request(KVM_REQ_EVENT, vcpu);
4607 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
4608 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4609 kvm_rip_write(vcpu, ctxt->eip);
4611 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
4615 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4617 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4619 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4620 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4621 size, port, &val, 1);
4622 /* do not return to emulator after return from userspace */
4623 vcpu->arch.pio.count = 0;
4626 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4628 static void tsc_bad(void *info)
4630 __this_cpu_write(cpu_tsc_khz, 0);
4633 static void tsc_khz_changed(void *data)
4635 struct cpufreq_freqs *freq = data;
4636 unsigned long khz = 0;
4640 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4641 khz = cpufreq_quick_get(raw_smp_processor_id());
4644 __this_cpu_write(cpu_tsc_khz, khz);
4647 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4650 struct cpufreq_freqs *freq = data;
4652 struct kvm_vcpu *vcpu;
4653 int i, send_ipi = 0;
4656 * We allow guests to temporarily run on slowing clocks,
4657 * provided we notify them after, or to run on accelerating
4658 * clocks, provided we notify them before. Thus time never
4661 * However, we have a problem. We can't atomically update
4662 * the frequency of a given CPU from this function; it is
4663 * merely a notifier, which can be called from any CPU.
4664 * Changing the TSC frequency at arbitrary points in time
4665 * requires a recomputation of local variables related to
4666 * the TSC for each VCPU. We must flag these local variables
4667 * to be updated and be sure the update takes place with the
4668 * new frequency before any guests proceed.
4670 * Unfortunately, the combination of hotplug CPU and frequency
4671 * change creates an intractable locking scenario; the order
4672 * of when these callouts happen is undefined with respect to
4673 * CPU hotplug, and they can race with each other. As such,
4674 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4675 * undefined; you can actually have a CPU frequency change take
4676 * place in between the computation of X and the setting of the
4677 * variable. To protect against this problem, all updates of
4678 * the per_cpu tsc_khz variable are done in an interrupt
4679 * protected IPI, and all callers wishing to update the value
4680 * must wait for a synchronous IPI to complete (which is trivial
4681 * if the caller is on the CPU already). This establishes the
4682 * necessary total order on variable updates.
4684 * Note that because a guest time update may take place
4685 * anytime after the setting of the VCPU's request bit, the
4686 * correct TSC value must be set before the request. However,
4687 * to ensure the update actually makes it to any guest which
4688 * starts running in hardware virtualization between the set
4689 * and the acquisition of the spinlock, we must also ping the
4690 * CPU after setting the request bit.
4694 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4696 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4699 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4701 raw_spin_lock(&kvm_lock);
4702 list_for_each_entry(kvm, &vm_list, vm_list) {
4703 kvm_for_each_vcpu(i, vcpu, kvm) {
4704 if (vcpu->cpu != freq->cpu)
4706 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4707 if (vcpu->cpu != smp_processor_id())
4711 raw_spin_unlock(&kvm_lock);
4713 if (freq->old < freq->new && send_ipi) {
4715 * We upscale the frequency. Must make the guest
4716 * doesn't see old kvmclock values while running with
4717 * the new frequency, otherwise we risk the guest sees
4718 * time go backwards.
4720 * In case we update the frequency for another cpu
4721 * (which might be in guest context) send an interrupt
4722 * to kick the cpu out of guest context. Next time
4723 * guest context is entered kvmclock will be updated,
4724 * so the guest will not see stale values.
4726 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4731 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4732 .notifier_call = kvmclock_cpufreq_notifier
4735 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4736 unsigned long action, void *hcpu)
4738 unsigned int cpu = (unsigned long)hcpu;
4742 case CPU_DOWN_FAILED:
4743 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4745 case CPU_DOWN_PREPARE:
4746 smp_call_function_single(cpu, tsc_bad, NULL, 1);
4752 static struct notifier_block kvmclock_cpu_notifier_block = {
4753 .notifier_call = kvmclock_cpu_notifier,
4754 .priority = -INT_MAX
4757 static void kvm_timer_init(void)
4761 max_tsc_khz = tsc_khz;
4762 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4763 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4764 #ifdef CONFIG_CPU_FREQ
4765 struct cpufreq_policy policy;
4766 memset(&policy, 0, sizeof(policy));
4768 cpufreq_get_policy(&policy, cpu);
4769 if (policy.cpuinfo.max_freq)
4770 max_tsc_khz = policy.cpuinfo.max_freq;
4773 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4774 CPUFREQ_TRANSITION_NOTIFIER);
4776 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
4777 for_each_online_cpu(cpu)
4778 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4781 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4783 int kvm_is_in_guest(void)
4785 return __this_cpu_read(current_vcpu) != NULL;
4788 static int kvm_is_user_mode(void)
4792 if (__this_cpu_read(current_vcpu))
4793 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
4795 return user_mode != 0;
4798 static unsigned long kvm_get_guest_ip(void)
4800 unsigned long ip = 0;
4802 if (__this_cpu_read(current_vcpu))
4803 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
4808 static struct perf_guest_info_callbacks kvm_guest_cbs = {
4809 .is_in_guest = kvm_is_in_guest,
4810 .is_user_mode = kvm_is_user_mode,
4811 .get_guest_ip = kvm_get_guest_ip,
4814 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4816 __this_cpu_write(current_vcpu, vcpu);
4818 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4820 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4822 __this_cpu_write(current_vcpu, NULL);
4824 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4826 static void kvm_set_mmio_spte_mask(void)
4829 int maxphyaddr = boot_cpu_data.x86_phys_bits;
4832 * Set the reserved bits and the present bit of an paging-structure
4833 * entry to generate page fault with PFER.RSV = 1.
4835 mask = ((1ull << (62 - maxphyaddr + 1)) - 1) << maxphyaddr;
4838 #ifdef CONFIG_X86_64
4840 * If reserved bit is not supported, clear the present bit to disable
4843 if (maxphyaddr == 52)
4847 kvm_mmu_set_mmio_spte_mask(mask);
4850 int kvm_arch_init(void *opaque)
4853 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4856 printk(KERN_ERR "kvm: already loaded the other module\n");
4861 if (!ops->cpu_has_kvm_support()) {
4862 printk(KERN_ERR "kvm: no hardware support\n");
4866 if (ops->disabled_by_bios()) {
4867 printk(KERN_ERR "kvm: disabled by bios\n");
4872 r = kvm_mmu_module_init();
4876 kvm_set_mmio_spte_mask();
4877 kvm_init_msr_list();
4880 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4881 PT_DIRTY_MASK, PT64_NX_MASK, 0);
4885 perf_register_guest_info_callbacks(&kvm_guest_cbs);
4888 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
4896 void kvm_arch_exit(void)
4898 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4900 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4901 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4902 CPUFREQ_TRANSITION_NOTIFIER);
4903 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4905 kvm_mmu_module_exit();
4908 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4910 ++vcpu->stat.halt_exits;
4911 if (irqchip_in_kernel(vcpu->kvm)) {
4912 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
4915 vcpu->run->exit_reason = KVM_EXIT_HLT;
4919 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4921 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4923 u64 param, ingpa, outgpa, ret;
4924 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4925 bool fast, longmode;
4929 * hypercall generates UD from non zero cpl and real mode
4932 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
4933 kvm_queue_exception(vcpu, UD_VECTOR);
4937 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4938 longmode = is_long_mode(vcpu) && cs_l == 1;
4941 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4942 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4943 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4944 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4945 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4946 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
4948 #ifdef CONFIG_X86_64
4950 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4951 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4952 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4956 code = param & 0xffff;
4957 fast = (param >> 16) & 0x1;
4958 rep_cnt = (param >> 32) & 0xfff;
4959 rep_idx = (param >> 48) & 0xfff;
4961 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
4964 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
4965 kvm_vcpu_on_spin(vcpu);
4968 res = HV_STATUS_INVALID_HYPERCALL_CODE;
4972 ret = res | (((u64)rep_done & 0xfff) << 32);
4974 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4976 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
4977 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
4983 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
4985 unsigned long nr, a0, a1, a2, a3, ret;
4988 if (kvm_hv_hypercall_enabled(vcpu->kvm))
4989 return kvm_hv_hypercall(vcpu);
4991 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
4992 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
4993 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
4994 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
4995 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
4997 trace_kvm_hypercall(nr, a0, a1, a2, a3);
4999 if (!is_long_mode(vcpu)) {
5007 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5013 case KVM_HC_VAPIC_POLL_IRQ:
5021 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5022 ++vcpu->stat.hypercalls;
5025 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5027 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5029 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5030 char instruction[3];
5031 unsigned long rip = kvm_rip_read(vcpu);
5034 * Blow out the MMU to ensure that no other VCPU has an active mapping
5035 * to ensure that the updated hypercall appears atomically across all
5038 kvm_mmu_zap_all(vcpu->kvm);
5040 kvm_x86_ops->patch_hypercall(vcpu, instruction);
5042 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5046 * Check if userspace requested an interrupt window, and that the
5047 * interrupt window is open.
5049 * No need to exit to userspace if we already have an interrupt queued.
5051 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5053 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5054 vcpu->run->request_interrupt_window &&
5055 kvm_arch_interrupt_allowed(vcpu));
5058 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5060 struct kvm_run *kvm_run = vcpu->run;
5062 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5063 kvm_run->cr8 = kvm_get_cr8(vcpu);
5064 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5065 if (irqchip_in_kernel(vcpu->kvm))
5066 kvm_run->ready_for_interrupt_injection = 1;
5068 kvm_run->ready_for_interrupt_injection =
5069 kvm_arch_interrupt_allowed(vcpu) &&
5070 !kvm_cpu_has_interrupt(vcpu) &&
5071 !kvm_event_needs_reinjection(vcpu);
5074 static void vapic_enter(struct kvm_vcpu *vcpu)
5076 struct kvm_lapic *apic = vcpu->arch.apic;
5079 if (!apic || !apic->vapic_addr)
5082 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5084 vcpu->arch.apic->vapic_page = page;
5087 static void vapic_exit(struct kvm_vcpu *vcpu)
5089 struct kvm_lapic *apic = vcpu->arch.apic;
5092 if (!apic || !apic->vapic_addr)
5095 idx = srcu_read_lock(&vcpu->kvm->srcu);
5096 kvm_release_page_dirty(apic->vapic_page);
5097 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5098 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5101 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5105 if (!kvm_x86_ops->update_cr8_intercept)
5108 if (!vcpu->arch.apic)
5111 if (!vcpu->arch.apic->vapic_addr)
5112 max_irr = kvm_lapic_find_highest_irr(vcpu);
5119 tpr = kvm_lapic_get_cr8(vcpu);
5121 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5124 static void inject_pending_event(struct kvm_vcpu *vcpu)
5126 /* try to reinject previous events if any */
5127 if (vcpu->arch.exception.pending) {
5128 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5129 vcpu->arch.exception.has_error_code,
5130 vcpu->arch.exception.error_code);
5131 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5132 vcpu->arch.exception.has_error_code,
5133 vcpu->arch.exception.error_code,
5134 vcpu->arch.exception.reinject);
5138 if (vcpu->arch.nmi_injected) {
5139 kvm_x86_ops->set_nmi(vcpu);
5143 if (vcpu->arch.interrupt.pending) {
5144 kvm_x86_ops->set_irq(vcpu);
5148 /* try to inject new event if pending */
5149 if (vcpu->arch.nmi_pending) {
5150 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5151 --vcpu->arch.nmi_pending;
5152 vcpu->arch.nmi_injected = true;
5153 kvm_x86_ops->set_nmi(vcpu);
5155 } else if (kvm_cpu_has_interrupt(vcpu)) {
5156 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5157 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5159 kvm_x86_ops->set_irq(vcpu);
5164 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5166 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5167 !vcpu->guest_xcr0_loaded) {
5168 /* kvm_set_xcr() also depends on this */
5169 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5170 vcpu->guest_xcr0_loaded = 1;
5174 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5176 if (vcpu->guest_xcr0_loaded) {
5177 if (vcpu->arch.xcr0 != host_xcr0)
5178 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5179 vcpu->guest_xcr0_loaded = 0;
5183 static void process_nmi(struct kvm_vcpu *vcpu)
5188 * x86 is limited to one NMI running, and one NMI pending after it.
5189 * If an NMI is already in progress, limit further NMIs to just one.
5190 * Otherwise, allow two (and we'll inject the first one immediately).
5192 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
5195 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
5196 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
5197 kvm_make_request(KVM_REQ_EVENT, vcpu);
5200 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5203 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5204 vcpu->run->request_interrupt_window;
5205 bool req_immediate_exit = 0;
5207 if (vcpu->requests) {
5208 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5209 kvm_mmu_unload(vcpu);
5210 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5211 __kvm_migrate_timers(vcpu);
5212 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5213 r = kvm_guest_time_update(vcpu);
5217 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5218 kvm_mmu_sync_roots(vcpu);
5219 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5220 kvm_x86_ops->tlb_flush(vcpu);
5221 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5222 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5226 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5227 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5231 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5232 vcpu->fpu_active = 0;
5233 kvm_x86_ops->fpu_deactivate(vcpu);
5235 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5236 /* Page is swapped out. Do synthetic halt */
5237 vcpu->arch.apf.halted = true;
5241 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
5242 record_steal_time(vcpu);
5243 if (kvm_check_request(KVM_REQ_NMI, vcpu))
5245 req_immediate_exit =
5246 kvm_check_request(KVM_REQ_IMMEDIATE_EXIT, vcpu);
5247 if (kvm_check_request(KVM_REQ_PMU, vcpu))
5248 kvm_handle_pmu_event(vcpu);
5249 if (kvm_check_request(KVM_REQ_PMI, vcpu))
5250 kvm_deliver_pmi(vcpu);
5253 r = kvm_mmu_reload(vcpu);
5257 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5258 inject_pending_event(vcpu);
5260 /* enable NMI/IRQ window open exits if needed */
5261 if (vcpu->arch.nmi_pending)
5262 kvm_x86_ops->enable_nmi_window(vcpu);
5263 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5264 kvm_x86_ops->enable_irq_window(vcpu);
5266 if (kvm_lapic_enabled(vcpu)) {
5267 update_cr8_intercept(vcpu);
5268 kvm_lapic_sync_to_vapic(vcpu);
5274 kvm_x86_ops->prepare_guest_switch(vcpu);
5275 if (vcpu->fpu_active)
5276 kvm_load_guest_fpu(vcpu);
5277 kvm_load_guest_xcr0(vcpu);
5279 vcpu->mode = IN_GUEST_MODE;
5281 /* We should set ->mode before check ->requests,
5282 * see the comment in make_all_cpus_request.
5286 local_irq_disable();
5288 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5289 || need_resched() || signal_pending(current)) {
5290 vcpu->mode = OUTSIDE_GUEST_MODE;
5294 kvm_x86_ops->cancel_injection(vcpu);
5299 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5301 if (req_immediate_exit)
5302 smp_send_reschedule(vcpu->cpu);
5306 if (unlikely(vcpu->arch.switch_db_regs)) {
5308 set_debugreg(vcpu->arch.eff_db[0], 0);
5309 set_debugreg(vcpu->arch.eff_db[1], 1);
5310 set_debugreg(vcpu->arch.eff_db[2], 2);
5311 set_debugreg(vcpu->arch.eff_db[3], 3);
5314 trace_kvm_entry(vcpu->vcpu_id);
5315 kvm_x86_ops->run(vcpu);
5318 * If the guest has used debug registers, at least dr7
5319 * will be disabled while returning to the host.
5320 * If we don't have active breakpoints in the host, we don't
5321 * care about the messed up debug address registers. But if
5322 * we have some of them active, restore the old state.
5324 if (hw_breakpoint_active())
5325 hw_breakpoint_restore();
5327 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu);
5329 vcpu->mode = OUTSIDE_GUEST_MODE;
5336 * We must have an instruction between local_irq_enable() and
5337 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5338 * the interrupt shadow. The stat.exits increment will do nicely.
5339 * But we need to prevent reordering, hence this barrier():
5347 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5350 * Profile KVM exit RIPs:
5352 if (unlikely(prof_on == KVM_PROFILING)) {
5353 unsigned long rip = kvm_rip_read(vcpu);
5354 profile_hit(KVM_PROFILING, (void *)rip);
5357 if (unlikely(vcpu->arch.tsc_always_catchup))
5358 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5360 kvm_lapic_sync_from_vapic(vcpu);
5362 r = kvm_x86_ops->handle_exit(vcpu);
5368 static int __vcpu_run(struct kvm_vcpu *vcpu)
5371 struct kvm *kvm = vcpu->kvm;
5373 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5374 pr_debug("vcpu %d received sipi with vector # %x\n",
5375 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5376 kvm_lapic_reset(vcpu);
5377 r = kvm_arch_vcpu_reset(vcpu);
5380 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5383 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5388 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5389 !vcpu->arch.apf.halted)
5390 r = vcpu_enter_guest(vcpu);
5392 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5393 kvm_vcpu_block(vcpu);
5394 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5395 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5397 switch(vcpu->arch.mp_state) {
5398 case KVM_MP_STATE_HALTED:
5399 vcpu->arch.mp_state =
5400 KVM_MP_STATE_RUNNABLE;
5401 case KVM_MP_STATE_RUNNABLE:
5402 vcpu->arch.apf.halted = false;
5404 case KVM_MP_STATE_SIPI_RECEIVED:
5415 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5416 if (kvm_cpu_has_pending_timer(vcpu))
5417 kvm_inject_pending_timer_irqs(vcpu);
5419 if (dm_request_for_irq_injection(vcpu)) {
5421 vcpu->run->exit_reason = KVM_EXIT_INTR;
5422 ++vcpu->stat.request_irq_exits;
5425 kvm_check_async_pf_completion(vcpu);
5427 if (signal_pending(current)) {
5429 vcpu->run->exit_reason = KVM_EXIT_INTR;
5430 ++vcpu->stat.signal_exits;
5432 if (need_resched()) {
5433 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5435 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5439 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5446 static int complete_mmio(struct kvm_vcpu *vcpu)
5448 struct kvm_run *run = vcpu->run;
5451 if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5454 if (vcpu->mmio_needed) {
5455 vcpu->mmio_needed = 0;
5456 if (!vcpu->mmio_is_write)
5457 memcpy(vcpu->mmio_data + vcpu->mmio_index,
5459 vcpu->mmio_index += 8;
5460 if (vcpu->mmio_index < vcpu->mmio_size) {
5461 run->exit_reason = KVM_EXIT_MMIO;
5462 run->mmio.phys_addr = vcpu->mmio_phys_addr + vcpu->mmio_index;
5463 memcpy(run->mmio.data, vcpu->mmio_data + vcpu->mmio_index, 8);
5464 run->mmio.len = min(vcpu->mmio_size - vcpu->mmio_index, 8);
5465 run->mmio.is_write = vcpu->mmio_is_write;
5466 vcpu->mmio_needed = 1;
5469 if (vcpu->mmio_is_write)
5471 vcpu->mmio_read_completed = 1;
5473 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5474 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5475 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5476 if (r != EMULATE_DONE)
5481 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5486 if (!tsk_used_math(current) && init_fpu(current))
5489 if (vcpu->sigset_active)
5490 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5492 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5493 kvm_vcpu_block(vcpu);
5494 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5499 /* re-sync apic's tpr */
5500 if (!irqchip_in_kernel(vcpu->kvm)) {
5501 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5507 r = complete_mmio(vcpu);
5511 r = __vcpu_run(vcpu);
5514 post_kvm_run_save(vcpu);
5515 if (vcpu->sigset_active)
5516 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5521 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5523 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5525 * We are here if userspace calls get_regs() in the middle of
5526 * instruction emulation. Registers state needs to be copied
5527 * back from emulation context to vcpu. Usrapace shouldn't do
5528 * that usually, but some bad designed PV devices (vmware
5529 * backdoor interface) need this to work
5531 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5532 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
5533 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5535 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5536 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5537 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5538 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5539 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5540 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5541 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5542 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5543 #ifdef CONFIG_X86_64
5544 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5545 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5546 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5547 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5548 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5549 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5550 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5551 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5554 regs->rip = kvm_rip_read(vcpu);
5555 regs->rflags = kvm_get_rflags(vcpu);
5560 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5562 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
5563 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5565 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5566 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5567 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5568 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5569 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5570 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5571 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5572 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5573 #ifdef CONFIG_X86_64
5574 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5575 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5576 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5577 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5578 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5579 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5580 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5581 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5584 kvm_rip_write(vcpu, regs->rip);
5585 kvm_set_rflags(vcpu, regs->rflags);
5587 vcpu->arch.exception.pending = false;
5589 kvm_make_request(KVM_REQ_EVENT, vcpu);
5594 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5596 struct kvm_segment cs;
5598 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
5602 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5604 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5605 struct kvm_sregs *sregs)
5609 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5610 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5611 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5612 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5613 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5614 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5616 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5617 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5619 kvm_x86_ops->get_idt(vcpu, &dt);
5620 sregs->idt.limit = dt.size;
5621 sregs->idt.base = dt.address;
5622 kvm_x86_ops->get_gdt(vcpu, &dt);
5623 sregs->gdt.limit = dt.size;
5624 sregs->gdt.base = dt.address;
5626 sregs->cr0 = kvm_read_cr0(vcpu);
5627 sregs->cr2 = vcpu->arch.cr2;
5628 sregs->cr3 = kvm_read_cr3(vcpu);
5629 sregs->cr4 = kvm_read_cr4(vcpu);
5630 sregs->cr8 = kvm_get_cr8(vcpu);
5631 sregs->efer = vcpu->arch.efer;
5632 sregs->apic_base = kvm_get_apic_base(vcpu);
5634 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
5636 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
5637 set_bit(vcpu->arch.interrupt.nr,
5638 (unsigned long *)sregs->interrupt_bitmap);
5643 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
5644 struct kvm_mp_state *mp_state)
5646 mp_state->mp_state = vcpu->arch.mp_state;
5650 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
5651 struct kvm_mp_state *mp_state)
5653 vcpu->arch.mp_state = mp_state->mp_state;
5654 kvm_make_request(KVM_REQ_EVENT, vcpu);
5658 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
5659 int reason, bool has_error_code, u32 error_code)
5661 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5664 init_emulate_ctxt(vcpu);
5666 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
5667 has_error_code, error_code);
5670 return EMULATE_FAIL;
5672 memcpy(vcpu->arch.regs, ctxt->regs, sizeof ctxt->regs);
5673 kvm_rip_write(vcpu, ctxt->eip);
5674 kvm_set_rflags(vcpu, ctxt->eflags);
5675 kvm_make_request(KVM_REQ_EVENT, vcpu);
5676 return EMULATE_DONE;
5678 EXPORT_SYMBOL_GPL(kvm_task_switch);
5680 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5681 struct kvm_sregs *sregs)
5683 int mmu_reset_needed = 0;
5684 int pending_vec, max_bits, idx;
5687 dt.size = sregs->idt.limit;
5688 dt.address = sregs->idt.base;
5689 kvm_x86_ops->set_idt(vcpu, &dt);
5690 dt.size = sregs->gdt.limit;
5691 dt.address = sregs->gdt.base;
5692 kvm_x86_ops->set_gdt(vcpu, &dt);
5694 vcpu->arch.cr2 = sregs->cr2;
5695 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
5696 vcpu->arch.cr3 = sregs->cr3;
5697 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
5699 kvm_set_cr8(vcpu, sregs->cr8);
5701 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5702 kvm_x86_ops->set_efer(vcpu, sregs->efer);
5703 kvm_set_apic_base(vcpu, sregs->apic_base);
5705 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5706 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5707 vcpu->arch.cr0 = sregs->cr0;
5709 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5710 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5711 if (sregs->cr4 & X86_CR4_OSXSAVE)
5712 kvm_update_cpuid(vcpu);
5714 idx = srcu_read_lock(&vcpu->kvm->srcu);
5715 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5716 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
5717 mmu_reset_needed = 1;
5719 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5721 if (mmu_reset_needed)
5722 kvm_mmu_reset_context(vcpu);
5724 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5725 pending_vec = find_first_bit(
5726 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5727 if (pending_vec < max_bits) {
5728 kvm_queue_interrupt(vcpu, pending_vec, false);
5729 pr_debug("Set back pending irq %d\n", pending_vec);
5732 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5733 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5734 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5735 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5736 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5737 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5739 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5740 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5742 update_cr8_intercept(vcpu);
5744 /* Older userspace won't unhalt the vcpu on reset. */
5745 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5746 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5748 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5750 kvm_make_request(KVM_REQ_EVENT, vcpu);
5755 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5756 struct kvm_guest_debug *dbg)
5758 unsigned long rflags;
5761 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5763 if (vcpu->arch.exception.pending)
5765 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5766 kvm_queue_exception(vcpu, DB_VECTOR);
5768 kvm_queue_exception(vcpu, BP_VECTOR);
5772 * Read rflags as long as potentially injected trace flags are still
5775 rflags = kvm_get_rflags(vcpu);
5777 vcpu->guest_debug = dbg->control;
5778 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5779 vcpu->guest_debug = 0;
5781 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5782 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5783 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5784 vcpu->arch.switch_db_regs =
5785 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5787 for (i = 0; i < KVM_NR_DB_REGS; i++)
5788 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5789 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5792 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5793 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5794 get_segment_base(vcpu, VCPU_SREG_CS);
5797 * Trigger an rflags update that will inject or remove the trace
5800 kvm_set_rflags(vcpu, rflags);
5802 kvm_x86_ops->set_guest_debug(vcpu, dbg);
5812 * Translate a guest virtual address to a guest physical address.
5814 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5815 struct kvm_translation *tr)
5817 unsigned long vaddr = tr->linear_address;
5821 idx = srcu_read_lock(&vcpu->kvm->srcu);
5822 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5823 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5824 tr->physical_address = gpa;
5825 tr->valid = gpa != UNMAPPED_GVA;
5832 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5834 struct i387_fxsave_struct *fxsave =
5835 &vcpu->arch.guest_fpu.state->fxsave;
5837 memcpy(fpu->fpr, fxsave->st_space, 128);
5838 fpu->fcw = fxsave->cwd;
5839 fpu->fsw = fxsave->swd;
5840 fpu->ftwx = fxsave->twd;
5841 fpu->last_opcode = fxsave->fop;
5842 fpu->last_ip = fxsave->rip;
5843 fpu->last_dp = fxsave->rdp;
5844 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5849 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5851 struct i387_fxsave_struct *fxsave =
5852 &vcpu->arch.guest_fpu.state->fxsave;
5854 memcpy(fxsave->st_space, fpu->fpr, 128);
5855 fxsave->cwd = fpu->fcw;
5856 fxsave->swd = fpu->fsw;
5857 fxsave->twd = fpu->ftwx;
5858 fxsave->fop = fpu->last_opcode;
5859 fxsave->rip = fpu->last_ip;
5860 fxsave->rdp = fpu->last_dp;
5861 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5866 int fx_init(struct kvm_vcpu *vcpu)
5870 err = fpu_alloc(&vcpu->arch.guest_fpu);
5874 fpu_finit(&vcpu->arch.guest_fpu);
5877 * Ensure guest xcr0 is valid for loading
5879 vcpu->arch.xcr0 = XSTATE_FP;
5881 vcpu->arch.cr0 |= X86_CR0_ET;
5885 EXPORT_SYMBOL_GPL(fx_init);
5887 static void fx_free(struct kvm_vcpu *vcpu)
5889 fpu_free(&vcpu->arch.guest_fpu);
5892 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5894 if (vcpu->guest_fpu_loaded)
5898 * Restore all possible states in the guest,
5899 * and assume host would use all available bits.
5900 * Guest xcr0 would be loaded later.
5902 kvm_put_guest_xcr0(vcpu);
5903 vcpu->guest_fpu_loaded = 1;
5904 unlazy_fpu(current);
5905 fpu_restore_checking(&vcpu->arch.guest_fpu);
5909 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5911 kvm_put_guest_xcr0(vcpu);
5913 if (!vcpu->guest_fpu_loaded)
5916 vcpu->guest_fpu_loaded = 0;
5917 fpu_save_init(&vcpu->arch.guest_fpu);
5918 ++vcpu->stat.fpu_reload;
5919 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
5923 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5925 kvmclock_reset(vcpu);
5927 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
5929 kvm_x86_ops->vcpu_free(vcpu);
5932 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5935 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
5936 printk_once(KERN_WARNING
5937 "kvm: SMP vm created on host with unstable TSC; "
5938 "guest TSC will not be reliable\n");
5939 return kvm_x86_ops->vcpu_create(kvm, id);
5942 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5946 vcpu->arch.mtrr_state.have_fixed = 1;
5948 r = kvm_arch_vcpu_reset(vcpu);
5950 r = kvm_mmu_setup(vcpu);
5956 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5958 vcpu->arch.apf.msr_val = 0;
5961 kvm_mmu_unload(vcpu);
5965 kvm_x86_ops->vcpu_free(vcpu);
5968 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5970 atomic_set(&vcpu->arch.nmi_queued, 0);
5971 vcpu->arch.nmi_pending = 0;
5972 vcpu->arch.nmi_injected = false;
5974 vcpu->arch.switch_db_regs = 0;
5975 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5976 vcpu->arch.dr6 = DR6_FIXED_1;
5977 vcpu->arch.dr7 = DR7_FIXED_1;
5979 kvm_make_request(KVM_REQ_EVENT, vcpu);
5980 vcpu->arch.apf.msr_val = 0;
5981 vcpu->arch.st.msr_val = 0;
5983 kvmclock_reset(vcpu);
5985 kvm_clear_async_pf_completion_queue(vcpu);
5986 kvm_async_pf_hash_reset(vcpu);
5987 vcpu->arch.apf.halted = false;
5989 kvm_pmu_reset(vcpu);
5991 return kvm_x86_ops->vcpu_reset(vcpu);
5994 int kvm_arch_hardware_enable(void *garbage)
5997 struct kvm_vcpu *vcpu;
6002 bool stable, backwards_tsc = false;
6004 kvm_shared_msr_cpu_online();
6005 ret = kvm_x86_ops->hardware_enable(garbage);
6009 local_tsc = native_read_tsc();
6010 stable = !check_tsc_unstable();
6011 list_for_each_entry(kvm, &vm_list, vm_list) {
6012 kvm_for_each_vcpu(i, vcpu, kvm) {
6013 if (!stable && vcpu->cpu == smp_processor_id())
6014 set_bit(KVM_REQ_CLOCK_UPDATE, &vcpu->requests);
6015 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
6016 backwards_tsc = true;
6017 if (vcpu->arch.last_host_tsc > max_tsc)
6018 max_tsc = vcpu->arch.last_host_tsc;
6024 * Sometimes, even reliable TSCs go backwards. This happens on
6025 * platforms that reset TSC during suspend or hibernate actions, but
6026 * maintain synchronization. We must compensate. Fortunately, we can
6027 * detect that condition here, which happens early in CPU bringup,
6028 * before any KVM threads can be running. Unfortunately, we can't
6029 * bring the TSCs fully up to date with real time, as we aren't yet far
6030 * enough into CPU bringup that we know how much real time has actually
6031 * elapsed; our helper function, get_kernel_ns() will be using boot
6032 * variables that haven't been updated yet.
6034 * So we simply find the maximum observed TSC above, then record the
6035 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
6036 * the adjustment will be applied. Note that we accumulate
6037 * adjustments, in case multiple suspend cycles happen before some VCPU
6038 * gets a chance to run again. In the event that no KVM threads get a
6039 * chance to run, we will miss the entire elapsed period, as we'll have
6040 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
6041 * loose cycle time. This isn't too big a deal, since the loss will be
6042 * uniform across all VCPUs (not to mention the scenario is extremely
6043 * unlikely). It is possible that a second hibernate recovery happens
6044 * much faster than a first, causing the observed TSC here to be
6045 * smaller; this would require additional padding adjustment, which is
6046 * why we set last_host_tsc to the local tsc observed here.
6048 * N.B. - this code below runs only on platforms with reliable TSC,
6049 * as that is the only way backwards_tsc is set above. Also note
6050 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
6051 * have the same delta_cyc adjustment applied if backwards_tsc
6052 * is detected. Note further, this adjustment is only done once,
6053 * as we reset last_host_tsc on all VCPUs to stop this from being
6054 * called multiple times (one for each physical CPU bringup).
6056 * Platforms with unnreliable TSCs don't have to deal with this, they
6057 * will be compensated by the logic in vcpu_load, which sets the TSC to
6058 * catchup mode. This will catchup all VCPUs to real time, but cannot
6059 * guarantee that they stay in perfect synchronization.
6061 if (backwards_tsc) {
6062 u64 delta_cyc = max_tsc - local_tsc;
6063 list_for_each_entry(kvm, &vm_list, vm_list) {
6064 kvm_for_each_vcpu(i, vcpu, kvm) {
6065 vcpu->arch.tsc_offset_adjustment += delta_cyc;
6066 vcpu->arch.last_host_tsc = local_tsc;
6070 * We have to disable TSC offset matching.. if you were
6071 * booting a VM while issuing an S4 host suspend....
6072 * you may have some problem. Solving this issue is
6073 * left as an exercise to the reader.
6075 kvm->arch.last_tsc_nsec = 0;
6076 kvm->arch.last_tsc_write = 0;
6083 void kvm_arch_hardware_disable(void *garbage)
6085 kvm_x86_ops->hardware_disable(garbage);
6086 drop_user_return_notifiers(garbage);
6089 int kvm_arch_hardware_setup(void)
6091 return kvm_x86_ops->hardware_setup();
6094 void kvm_arch_hardware_unsetup(void)
6096 kvm_x86_ops->hardware_unsetup();
6099 void kvm_arch_check_processor_compat(void *rtn)
6101 kvm_x86_ops->check_processor_compatibility(rtn);
6104 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6110 BUG_ON(vcpu->kvm == NULL);
6113 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6114 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6115 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6117 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6119 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6124 vcpu->arch.pio_data = page_address(page);
6126 kvm_set_tsc_khz(vcpu, max_tsc_khz);
6128 r = kvm_mmu_create(vcpu);
6130 goto fail_free_pio_data;
6132 if (irqchip_in_kernel(kvm)) {
6133 r = kvm_create_lapic(vcpu);
6135 goto fail_mmu_destroy;
6138 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6140 if (!vcpu->arch.mce_banks) {
6142 goto fail_free_lapic;
6144 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6146 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6147 goto fail_free_mce_banks;
6149 kvm_async_pf_hash_reset(vcpu);
6153 fail_free_mce_banks:
6154 kfree(vcpu->arch.mce_banks);
6156 kvm_free_lapic(vcpu);
6158 kvm_mmu_destroy(vcpu);
6160 free_page((unsigned long)vcpu->arch.pio_data);
6165 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6169 kvm_pmu_destroy(vcpu);
6170 kfree(vcpu->arch.mce_banks);
6171 kvm_free_lapic(vcpu);
6172 idx = srcu_read_lock(&vcpu->kvm->srcu);
6173 kvm_mmu_destroy(vcpu);
6174 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6175 free_page((unsigned long)vcpu->arch.pio_data);
6178 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
6183 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
6184 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
6186 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
6187 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6189 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
6194 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6197 kvm_mmu_unload(vcpu);
6201 static void kvm_free_vcpus(struct kvm *kvm)
6204 struct kvm_vcpu *vcpu;
6207 * Unpin any mmu pages first.
6209 kvm_for_each_vcpu(i, vcpu, kvm) {
6210 kvm_clear_async_pf_completion_queue(vcpu);
6211 kvm_unload_vcpu_mmu(vcpu);
6213 kvm_for_each_vcpu(i, vcpu, kvm)
6214 kvm_arch_vcpu_free(vcpu);
6216 mutex_lock(&kvm->lock);
6217 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6218 kvm->vcpus[i] = NULL;
6220 atomic_set(&kvm->online_vcpus, 0);
6221 mutex_unlock(&kvm->lock);
6224 void kvm_arch_sync_events(struct kvm *kvm)
6226 kvm_free_all_assigned_devices(kvm);
6230 void kvm_arch_destroy_vm(struct kvm *kvm)
6232 kvm_iommu_unmap_guest(kvm);
6233 kfree(kvm->arch.vpic);
6234 kfree(kvm->arch.vioapic);
6235 kvm_free_vcpus(kvm);
6236 if (kvm->arch.apic_access_page)
6237 put_page(kvm->arch.apic_access_page);
6238 if (kvm->arch.ept_identity_pagetable)
6239 put_page(kvm->arch.ept_identity_pagetable);
6242 void kvm_arch_free_memslot(struct kvm_memory_slot *free,
6243 struct kvm_memory_slot *dont)
6247 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
6248 if (!dont || free->arch.lpage_info[i] != dont->arch.lpage_info[i]) {
6249 vfree(free->arch.lpage_info[i]);
6250 free->arch.lpage_info[i] = NULL;
6255 int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages)
6259 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
6264 lpages = gfn_to_index(slot->base_gfn + npages - 1,
6265 slot->base_gfn, level) + 1;
6267 slot->arch.lpage_info[i] =
6268 vzalloc(lpages * sizeof(*slot->arch.lpage_info[i]));
6269 if (!slot->arch.lpage_info[i])
6272 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
6273 slot->arch.lpage_info[i][0].write_count = 1;
6274 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
6275 slot->arch.lpage_info[i][lpages - 1].write_count = 1;
6276 ugfn = slot->userspace_addr >> PAGE_SHIFT;
6278 * If the gfn and userspace address are not aligned wrt each
6279 * other, or if explicitly asked to, disable large page
6280 * support for this slot
6282 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
6283 !kvm_largepages_enabled()) {
6286 for (j = 0; j < lpages; ++j)
6287 slot->arch.lpage_info[i][j].write_count = 1;
6294 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
6295 vfree(slot->arch.lpage_info[i]);
6296 slot->arch.lpage_info[i] = NULL;
6301 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6302 struct kvm_memory_slot *memslot,
6303 struct kvm_memory_slot old,
6304 struct kvm_userspace_memory_region *mem,
6307 int npages = memslot->npages;
6308 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6310 /* Prevent internal slot pages from being moved by fork()/COW. */
6311 if (memslot->id >= KVM_MEMORY_SLOTS)
6312 map_flags = MAP_SHARED | MAP_ANONYMOUS;
6314 /*To keep backward compatibility with older userspace,
6315 *x86 needs to hanlde !user_alloc case.
6318 if (npages && !old.rmap) {
6319 unsigned long userspace_addr;
6321 down_write(¤t->mm->mmap_sem);
6322 userspace_addr = do_mmap(NULL, 0,
6324 PROT_READ | PROT_WRITE,
6327 up_write(¤t->mm->mmap_sem);
6329 if (IS_ERR((void *)userspace_addr))
6330 return PTR_ERR((void *)userspace_addr);
6332 memslot->userspace_addr = userspace_addr;
6340 void kvm_arch_commit_memory_region(struct kvm *kvm,
6341 struct kvm_userspace_memory_region *mem,
6342 struct kvm_memory_slot old,
6346 int nr_mmu_pages = 0, npages = mem->memory_size >> PAGE_SHIFT;
6348 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6351 down_write(¤t->mm->mmap_sem);
6352 ret = do_munmap(current->mm, old.userspace_addr,
6353 old.npages * PAGE_SIZE);
6354 up_write(¤t->mm->mmap_sem);
6357 "kvm_vm_ioctl_set_memory_region: "
6358 "failed to munmap memory\n");
6361 if (!kvm->arch.n_requested_mmu_pages)
6362 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6364 spin_lock(&kvm->mmu_lock);
6366 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6367 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6368 spin_unlock(&kvm->mmu_lock);
6371 void kvm_arch_flush_shadow(struct kvm *kvm)
6373 kvm_mmu_zap_all(kvm);
6374 kvm_reload_remote_mmus(kvm);
6377 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6379 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6380 !vcpu->arch.apf.halted)
6381 || !list_empty_careful(&vcpu->async_pf.done)
6382 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6383 || atomic_read(&vcpu->arch.nmi_queued) ||
6384 (kvm_arch_interrupt_allowed(vcpu) &&
6385 kvm_cpu_has_interrupt(vcpu));
6388 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6391 int cpu = vcpu->cpu;
6393 if (waitqueue_active(&vcpu->wq)) {
6394 wake_up_interruptible(&vcpu->wq);
6395 ++vcpu->stat.halt_wakeup;
6399 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6400 if (kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE)
6401 smp_send_reschedule(cpu);
6405 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6407 return kvm_x86_ops->interrupt_allowed(vcpu);
6410 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6412 unsigned long current_rip = kvm_rip_read(vcpu) +
6413 get_segment_base(vcpu, VCPU_SREG_CS);
6415 return current_rip == linear_rip;
6417 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6419 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6421 unsigned long rflags;
6423 rflags = kvm_x86_ops->get_rflags(vcpu);
6424 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6425 rflags &= ~X86_EFLAGS_TF;
6428 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6430 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6432 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6433 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6434 rflags |= X86_EFLAGS_TF;
6435 kvm_x86_ops->set_rflags(vcpu, rflags);
6436 kvm_make_request(KVM_REQ_EVENT, vcpu);
6438 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6440 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6444 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6445 is_error_page(work->page))
6448 r = kvm_mmu_reload(vcpu);
6452 if (!vcpu->arch.mmu.direct_map &&
6453 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6456 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6459 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6461 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6464 static inline u32 kvm_async_pf_next_probe(u32 key)
6466 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6469 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6471 u32 key = kvm_async_pf_hash_fn(gfn);
6473 while (vcpu->arch.apf.gfns[key] != ~0)
6474 key = kvm_async_pf_next_probe(key);
6476 vcpu->arch.apf.gfns[key] = gfn;
6479 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6482 u32 key = kvm_async_pf_hash_fn(gfn);
6484 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6485 (vcpu->arch.apf.gfns[key] != gfn &&
6486 vcpu->arch.apf.gfns[key] != ~0); i++)
6487 key = kvm_async_pf_next_probe(key);
6492 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6494 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6497 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6501 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6503 vcpu->arch.apf.gfns[i] = ~0;
6505 j = kvm_async_pf_next_probe(j);
6506 if (vcpu->arch.apf.gfns[j] == ~0)
6508 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6510 * k lies cyclically in ]i,j]
6512 * |....j i.k.| or |.k..j i...|
6514 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6515 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6520 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6523 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6527 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6528 struct kvm_async_pf *work)
6530 struct x86_exception fault;
6532 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6533 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6535 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6536 (vcpu->arch.apf.send_user_only &&
6537 kvm_x86_ops->get_cpl(vcpu) == 0))
6538 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6539 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6540 fault.vector = PF_VECTOR;
6541 fault.error_code_valid = true;
6542 fault.error_code = 0;
6543 fault.nested_page_fault = false;
6544 fault.address = work->arch.token;
6545 kvm_inject_page_fault(vcpu, &fault);
6549 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6550 struct kvm_async_pf *work)
6552 struct x86_exception fault;
6554 trace_kvm_async_pf_ready(work->arch.token, work->gva);
6555 if (is_error_page(work->page))
6556 work->arch.token = ~0; /* broadcast wakeup */
6558 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6560 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6561 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6562 fault.vector = PF_VECTOR;
6563 fault.error_code_valid = true;
6564 fault.error_code = 0;
6565 fault.nested_page_fault = false;
6566 fault.address = work->arch.token;
6567 kvm_inject_page_fault(vcpu, &fault);
6569 vcpu->arch.apf.halted = false;
6572 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6574 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6577 return !kvm_event_needs_reinjection(vcpu) &&
6578 kvm_x86_ops->interrupt_allowed(vcpu);
6581 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6582 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6583 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6584 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6585 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6586 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6587 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6588 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6589 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6590 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6591 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6592 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);