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"
30 #include <linux/clocksource.h>
31 #include <linux/interrupt.h>
32 #include <linux/kvm.h>
34 #include <linux/vmalloc.h>
35 #include <linux/module.h>
36 #include <linux/mman.h>
37 #include <linux/highmem.h>
38 #include <linux/iommu.h>
39 #include <linux/intel-iommu.h>
40 #include <linux/cpufreq.h>
41 #include <linux/user-return-notifier.h>
42 #include <linux/srcu.h>
43 #include <linux/slab.h>
44 #include <linux/perf_event.h>
45 #include <linux/uaccess.h>
46 #include <linux/hash.h>
47 #include <trace/events/kvm.h>
49 #define CREATE_TRACE_POINTS
52 #include <asm/debugreg.h>
59 #include <asm/pvclock.h>
60 #include <asm/div64.h>
62 #define MAX_IO_MSRS 256
63 #define CR0_RESERVED_BITS \
64 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
65 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
66 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
67 #define CR4_RESERVED_BITS \
68 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
69 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
70 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
72 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
74 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
76 #define KVM_MAX_MCE_BANKS 32
77 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
80 * - enable syscall per default because its emulated by KVM
81 * - enable LME and LMA per default on 64 bit KVM
84 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
86 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
89 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
90 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
92 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
93 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
94 struct kvm_cpuid_entry2 __user *entries);
96 struct kvm_x86_ops *kvm_x86_ops;
97 EXPORT_SYMBOL_GPL(kvm_x86_ops);
100 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
102 #define KVM_NR_SHARED_MSRS 16
104 struct kvm_shared_msrs_global {
106 u32 msrs[KVM_NR_SHARED_MSRS];
109 struct kvm_shared_msrs {
110 struct user_return_notifier urn;
112 struct kvm_shared_msr_values {
115 } values[KVM_NR_SHARED_MSRS];
118 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
119 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
121 struct kvm_stats_debugfs_item debugfs_entries[] = {
122 { "pf_fixed", VCPU_STAT(pf_fixed) },
123 { "pf_guest", VCPU_STAT(pf_guest) },
124 { "tlb_flush", VCPU_STAT(tlb_flush) },
125 { "invlpg", VCPU_STAT(invlpg) },
126 { "exits", VCPU_STAT(exits) },
127 { "io_exits", VCPU_STAT(io_exits) },
128 { "mmio_exits", VCPU_STAT(mmio_exits) },
129 { "signal_exits", VCPU_STAT(signal_exits) },
130 { "irq_window", VCPU_STAT(irq_window_exits) },
131 { "nmi_window", VCPU_STAT(nmi_window_exits) },
132 { "halt_exits", VCPU_STAT(halt_exits) },
133 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
134 { "hypercalls", VCPU_STAT(hypercalls) },
135 { "request_irq", VCPU_STAT(request_irq_exits) },
136 { "irq_exits", VCPU_STAT(irq_exits) },
137 { "host_state_reload", VCPU_STAT(host_state_reload) },
138 { "efer_reload", VCPU_STAT(efer_reload) },
139 { "fpu_reload", VCPU_STAT(fpu_reload) },
140 { "insn_emulation", VCPU_STAT(insn_emulation) },
141 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
142 { "irq_injections", VCPU_STAT(irq_injections) },
143 { "nmi_injections", VCPU_STAT(nmi_injections) },
144 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
145 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
146 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
147 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
148 { "mmu_flooded", VM_STAT(mmu_flooded) },
149 { "mmu_recycled", VM_STAT(mmu_recycled) },
150 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
151 { "mmu_unsync", VM_STAT(mmu_unsync) },
152 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
153 { "largepages", VM_STAT(lpages) },
157 u64 __read_mostly host_xcr0;
159 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
162 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
163 vcpu->arch.apf.gfns[i] = ~0;
166 static void kvm_on_user_return(struct user_return_notifier *urn)
169 struct kvm_shared_msrs *locals
170 = container_of(urn, struct kvm_shared_msrs, urn);
171 struct kvm_shared_msr_values *values;
173 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
174 values = &locals->values[slot];
175 if (values->host != values->curr) {
176 wrmsrl(shared_msrs_global.msrs[slot], values->host);
177 values->curr = values->host;
180 locals->registered = false;
181 user_return_notifier_unregister(urn);
184 static void shared_msr_update(unsigned slot, u32 msr)
186 struct kvm_shared_msrs *smsr;
189 smsr = &__get_cpu_var(shared_msrs);
190 /* only read, and nobody should modify it at this time,
191 * so don't need lock */
192 if (slot >= shared_msrs_global.nr) {
193 printk(KERN_ERR "kvm: invalid MSR slot!");
196 rdmsrl_safe(msr, &value);
197 smsr->values[slot].host = value;
198 smsr->values[slot].curr = value;
201 void kvm_define_shared_msr(unsigned slot, u32 msr)
203 if (slot >= shared_msrs_global.nr)
204 shared_msrs_global.nr = slot + 1;
205 shared_msrs_global.msrs[slot] = msr;
206 /* we need ensured the shared_msr_global have been updated */
209 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
211 static void kvm_shared_msr_cpu_online(void)
215 for (i = 0; i < shared_msrs_global.nr; ++i)
216 shared_msr_update(i, shared_msrs_global.msrs[i]);
219 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
221 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
223 if (((value ^ smsr->values[slot].curr) & mask) == 0)
225 smsr->values[slot].curr = value;
226 wrmsrl(shared_msrs_global.msrs[slot], value);
227 if (!smsr->registered) {
228 smsr->urn.on_user_return = kvm_on_user_return;
229 user_return_notifier_register(&smsr->urn);
230 smsr->registered = true;
233 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
235 static void drop_user_return_notifiers(void *ignore)
237 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
239 if (smsr->registered)
240 kvm_on_user_return(&smsr->urn);
243 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
245 if (irqchip_in_kernel(vcpu->kvm))
246 return vcpu->arch.apic_base;
248 return vcpu->arch.apic_base;
250 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
252 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
254 /* TODO: reserve bits check */
255 if (irqchip_in_kernel(vcpu->kvm))
256 kvm_lapic_set_base(vcpu, data);
258 vcpu->arch.apic_base = data;
260 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
262 #define EXCPT_BENIGN 0
263 #define EXCPT_CONTRIBUTORY 1
266 static int exception_class(int vector)
276 return EXCPT_CONTRIBUTORY;
283 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
284 unsigned nr, bool has_error, u32 error_code,
290 kvm_make_request(KVM_REQ_EVENT, vcpu);
292 if (!vcpu->arch.exception.pending) {
294 vcpu->arch.exception.pending = true;
295 vcpu->arch.exception.has_error_code = has_error;
296 vcpu->arch.exception.nr = nr;
297 vcpu->arch.exception.error_code = error_code;
298 vcpu->arch.exception.reinject = reinject;
302 /* to check exception */
303 prev_nr = vcpu->arch.exception.nr;
304 if (prev_nr == DF_VECTOR) {
305 /* triple fault -> shutdown */
306 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
309 class1 = exception_class(prev_nr);
310 class2 = exception_class(nr);
311 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
312 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
313 /* generate double fault per SDM Table 5-5 */
314 vcpu->arch.exception.pending = true;
315 vcpu->arch.exception.has_error_code = true;
316 vcpu->arch.exception.nr = DF_VECTOR;
317 vcpu->arch.exception.error_code = 0;
319 /* replace previous exception with a new one in a hope
320 that instruction re-execution will regenerate lost
325 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
327 kvm_multiple_exception(vcpu, nr, false, 0, false);
329 EXPORT_SYMBOL_GPL(kvm_queue_exception);
331 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
333 kvm_multiple_exception(vcpu, nr, false, 0, true);
335 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
337 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
339 ++vcpu->stat.pf_guest;
340 vcpu->arch.cr2 = fault->address;
341 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
344 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
346 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
347 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
349 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
352 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
354 kvm_make_request(KVM_REQ_EVENT, vcpu);
355 vcpu->arch.nmi_pending = 1;
357 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
359 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
361 kvm_multiple_exception(vcpu, nr, true, error_code, false);
363 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
365 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
367 kvm_multiple_exception(vcpu, nr, true, error_code, true);
369 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
372 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
373 * a #GP and return false.
375 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
377 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
379 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
382 EXPORT_SYMBOL_GPL(kvm_require_cpl);
385 * This function will be used to read from the physical memory of the currently
386 * running guest. The difference to kvm_read_guest_page is that this function
387 * can read from guest physical or from the guest's guest physical memory.
389 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
390 gfn_t ngfn, void *data, int offset, int len,
396 ngpa = gfn_to_gpa(ngfn);
397 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
398 if (real_gfn == UNMAPPED_GVA)
401 real_gfn = gpa_to_gfn(real_gfn);
403 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
405 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
407 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
408 void *data, int offset, int len, u32 access)
410 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
411 data, offset, len, access);
415 * Load the pae pdptrs. Return true is they are all valid.
417 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
419 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
420 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
423 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
425 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
426 offset * sizeof(u64), sizeof(pdpte),
427 PFERR_USER_MASK|PFERR_WRITE_MASK);
432 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
433 if (is_present_gpte(pdpte[i]) &&
434 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
441 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
442 __set_bit(VCPU_EXREG_PDPTR,
443 (unsigned long *)&vcpu->arch.regs_avail);
444 __set_bit(VCPU_EXREG_PDPTR,
445 (unsigned long *)&vcpu->arch.regs_dirty);
450 EXPORT_SYMBOL_GPL(load_pdptrs);
452 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
454 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
460 if (is_long_mode(vcpu) || !is_pae(vcpu))
463 if (!test_bit(VCPU_EXREG_PDPTR,
464 (unsigned long *)&vcpu->arch.regs_avail))
467 gfn = (vcpu->arch.cr3 & ~31u) >> PAGE_SHIFT;
468 offset = (vcpu->arch.cr3 & ~31u) & (PAGE_SIZE - 1);
469 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
470 PFERR_USER_MASK | PFERR_WRITE_MASK);
473 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
479 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
481 unsigned long old_cr0 = kvm_read_cr0(vcpu);
482 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
483 X86_CR0_CD | X86_CR0_NW;
488 if (cr0 & 0xffffffff00000000UL)
492 cr0 &= ~CR0_RESERVED_BITS;
494 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
497 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
500 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
502 if ((vcpu->arch.efer & EFER_LME)) {
507 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
512 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
517 kvm_x86_ops->set_cr0(vcpu, cr0);
519 if ((cr0 ^ old_cr0) & X86_CR0_PG)
520 kvm_clear_async_pf_completion_queue(vcpu);
522 if ((cr0 ^ old_cr0) & update_bits)
523 kvm_mmu_reset_context(vcpu);
526 EXPORT_SYMBOL_GPL(kvm_set_cr0);
528 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
530 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
532 EXPORT_SYMBOL_GPL(kvm_lmsw);
534 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
538 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
539 if (index != XCR_XFEATURE_ENABLED_MASK)
542 if (kvm_x86_ops->get_cpl(vcpu) != 0)
544 if (!(xcr0 & XSTATE_FP))
546 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
548 if (xcr0 & ~host_xcr0)
550 vcpu->arch.xcr0 = xcr0;
551 vcpu->guest_xcr0_loaded = 0;
555 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
557 if (__kvm_set_xcr(vcpu, index, xcr)) {
558 kvm_inject_gp(vcpu, 0);
563 EXPORT_SYMBOL_GPL(kvm_set_xcr);
565 static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
567 struct kvm_cpuid_entry2 *best;
569 best = kvm_find_cpuid_entry(vcpu, 1, 0);
570 return best && (best->ecx & bit(X86_FEATURE_XSAVE));
573 static void update_cpuid(struct kvm_vcpu *vcpu)
575 struct kvm_cpuid_entry2 *best;
577 best = kvm_find_cpuid_entry(vcpu, 1, 0);
581 /* Update OSXSAVE bit */
582 if (cpu_has_xsave && best->function == 0x1) {
583 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
584 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
585 best->ecx |= bit(X86_FEATURE_OSXSAVE);
589 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
591 unsigned long old_cr4 = kvm_read_cr4(vcpu);
592 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
594 if (cr4 & CR4_RESERVED_BITS)
597 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
600 if (is_long_mode(vcpu)) {
601 if (!(cr4 & X86_CR4_PAE))
603 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
604 && ((cr4 ^ old_cr4) & pdptr_bits)
605 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, vcpu->arch.cr3))
608 if (cr4 & X86_CR4_VMXE)
611 kvm_x86_ops->set_cr4(vcpu, cr4);
613 if ((cr4 ^ old_cr4) & pdptr_bits)
614 kvm_mmu_reset_context(vcpu);
616 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
621 EXPORT_SYMBOL_GPL(kvm_set_cr4);
623 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
625 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
626 kvm_mmu_sync_roots(vcpu);
627 kvm_mmu_flush_tlb(vcpu);
631 if (is_long_mode(vcpu)) {
632 if (cr3 & CR3_L_MODE_RESERVED_BITS)
636 if (cr3 & CR3_PAE_RESERVED_BITS)
638 if (is_paging(vcpu) &&
639 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
643 * We don't check reserved bits in nonpae mode, because
644 * this isn't enforced, and VMware depends on this.
649 * Does the new cr3 value map to physical memory? (Note, we
650 * catch an invalid cr3 even in real-mode, because it would
651 * cause trouble later on when we turn on paging anyway.)
653 * A real CPU would silently accept an invalid cr3 and would
654 * attempt to use it - with largely undefined (and often hard
655 * to debug) behavior on the guest side.
657 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
659 vcpu->arch.cr3 = cr3;
660 vcpu->arch.mmu.new_cr3(vcpu);
663 EXPORT_SYMBOL_GPL(kvm_set_cr3);
665 int __kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
667 if (cr8 & CR8_RESERVED_BITS)
669 if (irqchip_in_kernel(vcpu->kvm))
670 kvm_lapic_set_tpr(vcpu, cr8);
672 vcpu->arch.cr8 = cr8;
676 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
678 if (__kvm_set_cr8(vcpu, cr8))
679 kvm_inject_gp(vcpu, 0);
681 EXPORT_SYMBOL_GPL(kvm_set_cr8);
683 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
685 if (irqchip_in_kernel(vcpu->kvm))
686 return kvm_lapic_get_cr8(vcpu);
688 return vcpu->arch.cr8;
690 EXPORT_SYMBOL_GPL(kvm_get_cr8);
692 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
696 vcpu->arch.db[dr] = val;
697 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
698 vcpu->arch.eff_db[dr] = val;
701 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
705 if (val & 0xffffffff00000000ULL)
707 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
710 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
714 if (val & 0xffffffff00000000ULL)
716 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
717 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
718 kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
719 vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
727 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
731 res = __kvm_set_dr(vcpu, dr, val);
733 kvm_queue_exception(vcpu, UD_VECTOR);
735 kvm_inject_gp(vcpu, 0);
739 EXPORT_SYMBOL_GPL(kvm_set_dr);
741 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
745 *val = vcpu->arch.db[dr];
748 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
752 *val = vcpu->arch.dr6;
755 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
759 *val = vcpu->arch.dr7;
766 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
768 if (_kvm_get_dr(vcpu, dr, val)) {
769 kvm_queue_exception(vcpu, UD_VECTOR);
774 EXPORT_SYMBOL_GPL(kvm_get_dr);
777 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
778 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
780 * This list is modified at module load time to reflect the
781 * capabilities of the host cpu. This capabilities test skips MSRs that are
782 * kvm-specific. Those are put in the beginning of the list.
785 #define KVM_SAVE_MSRS_BEGIN 8
786 static u32 msrs_to_save[] = {
787 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
788 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
789 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
790 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN,
791 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
794 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
796 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
799 static unsigned num_msrs_to_save;
801 static u32 emulated_msrs[] = {
802 MSR_IA32_MISC_ENABLE,
807 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
809 u64 old_efer = vcpu->arch.efer;
811 if (efer & efer_reserved_bits)
815 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
818 if (efer & EFER_FFXSR) {
819 struct kvm_cpuid_entry2 *feat;
821 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
822 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
826 if (efer & EFER_SVME) {
827 struct kvm_cpuid_entry2 *feat;
829 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
830 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
835 efer |= vcpu->arch.efer & EFER_LMA;
837 kvm_x86_ops->set_efer(vcpu, efer);
839 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
841 /* Update reserved bits */
842 if ((efer ^ old_efer) & EFER_NX)
843 kvm_mmu_reset_context(vcpu);
848 void kvm_enable_efer_bits(u64 mask)
850 efer_reserved_bits &= ~mask;
852 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
856 * Writes msr value into into the appropriate "register".
857 * Returns 0 on success, non-0 otherwise.
858 * Assumes vcpu_load() was already called.
860 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
862 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
866 * Adapt set_msr() to msr_io()'s calling convention
868 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
870 return kvm_set_msr(vcpu, index, *data);
873 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
877 struct pvclock_wall_clock wc;
878 struct timespec boot;
883 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
888 ++version; /* first time write, random junk */
892 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
895 * The guest calculates current wall clock time by adding
896 * system time (updated by kvm_guest_time_update below) to the
897 * wall clock specified here. guest system time equals host
898 * system time for us, thus we must fill in host boot time here.
902 wc.sec = boot.tv_sec;
903 wc.nsec = boot.tv_nsec;
904 wc.version = version;
906 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
909 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
912 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
914 uint32_t quotient, remainder;
916 /* Don't try to replace with do_div(), this one calculates
917 * "(dividend << 32) / divisor" */
919 : "=a" (quotient), "=d" (remainder)
920 : "0" (0), "1" (dividend), "r" (divisor) );
924 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
925 s8 *pshift, u32 *pmultiplier)
932 tps64 = base_khz * 1000LL;
933 scaled64 = scaled_khz * 1000LL;
934 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
939 tps32 = (uint32_t)tps64;
940 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
941 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
949 *pmultiplier = div_frac(scaled64, tps32);
951 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
952 __func__, base_khz, scaled_khz, shift, *pmultiplier);
955 static inline u64 get_kernel_ns(void)
959 WARN_ON(preemptible());
961 monotonic_to_bootbased(&ts);
962 return timespec_to_ns(&ts);
965 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
966 unsigned long max_tsc_khz;
968 static inline int kvm_tsc_changes_freq(void)
971 int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
972 cpufreq_quick_get(cpu) != 0;
977 static inline u64 nsec_to_cycles(u64 nsec)
981 WARN_ON(preemptible());
982 if (kvm_tsc_changes_freq())
983 printk_once(KERN_WARNING
984 "kvm: unreliable cycle conversion on adjustable rate TSC\n");
985 ret = nsec * __get_cpu_var(cpu_tsc_khz);
986 do_div(ret, USEC_PER_SEC);
990 static void kvm_arch_set_tsc_khz(struct kvm *kvm, u32 this_tsc_khz)
992 /* Compute a scale to convert nanoseconds in TSC cycles */
993 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
994 &kvm->arch.virtual_tsc_shift,
995 &kvm->arch.virtual_tsc_mult);
996 kvm->arch.virtual_tsc_khz = this_tsc_khz;
999 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1001 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1002 vcpu->kvm->arch.virtual_tsc_mult,
1003 vcpu->kvm->arch.virtual_tsc_shift);
1004 tsc += vcpu->arch.last_tsc_write;
1008 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1010 struct kvm *kvm = vcpu->kvm;
1011 u64 offset, ns, elapsed;
1012 unsigned long flags;
1015 spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1016 offset = data - native_read_tsc();
1017 ns = get_kernel_ns();
1018 elapsed = ns - kvm->arch.last_tsc_nsec;
1019 sdiff = data - kvm->arch.last_tsc_write;
1024 * Special case: close write to TSC within 5 seconds of
1025 * another CPU is interpreted as an attempt to synchronize
1026 * The 5 seconds is to accomodate host load / swapping as
1027 * well as any reset of TSC during the boot process.
1029 * In that case, for a reliable TSC, we can match TSC offsets,
1030 * or make a best guest using elapsed value.
1032 if (sdiff < nsec_to_cycles(5ULL * NSEC_PER_SEC) &&
1033 elapsed < 5ULL * NSEC_PER_SEC) {
1034 if (!check_tsc_unstable()) {
1035 offset = kvm->arch.last_tsc_offset;
1036 pr_debug("kvm: matched tsc offset for %llu\n", data);
1038 u64 delta = nsec_to_cycles(elapsed);
1040 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1042 ns = kvm->arch.last_tsc_nsec;
1044 kvm->arch.last_tsc_nsec = ns;
1045 kvm->arch.last_tsc_write = data;
1046 kvm->arch.last_tsc_offset = offset;
1047 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1048 spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1050 /* Reset of TSC must disable overshoot protection below */
1051 vcpu->arch.hv_clock.tsc_timestamp = 0;
1052 vcpu->arch.last_tsc_write = data;
1053 vcpu->arch.last_tsc_nsec = ns;
1055 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1057 static int kvm_guest_time_update(struct kvm_vcpu *v)
1059 unsigned long flags;
1060 struct kvm_vcpu_arch *vcpu = &v->arch;
1062 unsigned long this_tsc_khz;
1063 s64 kernel_ns, max_kernel_ns;
1066 /* Keep irq disabled to prevent changes to the clock */
1067 local_irq_save(flags);
1068 kvm_get_msr(v, MSR_IA32_TSC, &tsc_timestamp);
1069 kernel_ns = get_kernel_ns();
1070 this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
1072 if (unlikely(this_tsc_khz == 0)) {
1073 local_irq_restore(flags);
1074 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1079 * We may have to catch up the TSC to match elapsed wall clock
1080 * time for two reasons, even if kvmclock is used.
1081 * 1) CPU could have been running below the maximum TSC rate
1082 * 2) Broken TSC compensation resets the base at each VCPU
1083 * entry to avoid unknown leaps of TSC even when running
1084 * again on the same CPU. This may cause apparent elapsed
1085 * time to disappear, and the guest to stand still or run
1088 if (vcpu->tsc_catchup) {
1089 u64 tsc = compute_guest_tsc(v, kernel_ns);
1090 if (tsc > tsc_timestamp) {
1091 kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1092 tsc_timestamp = tsc;
1096 local_irq_restore(flags);
1098 if (!vcpu->time_page)
1102 * Time as measured by the TSC may go backwards when resetting the base
1103 * tsc_timestamp. The reason for this is that the TSC resolution is
1104 * higher than the resolution of the other clock scales. Thus, many
1105 * possible measurments of the TSC correspond to one measurement of any
1106 * other clock, and so a spread of values is possible. This is not a
1107 * problem for the computation of the nanosecond clock; with TSC rates
1108 * around 1GHZ, there can only be a few cycles which correspond to one
1109 * nanosecond value, and any path through this code will inevitably
1110 * take longer than that. However, with the kernel_ns value itself,
1111 * the precision may be much lower, down to HZ granularity. If the
1112 * first sampling of TSC against kernel_ns ends in the low part of the
1113 * range, and the second in the high end of the range, we can get:
1115 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1117 * As the sampling errors potentially range in the thousands of cycles,
1118 * it is possible such a time value has already been observed by the
1119 * guest. To protect against this, we must compute the system time as
1120 * observed by the guest and ensure the new system time is greater.
1123 if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1124 max_kernel_ns = vcpu->last_guest_tsc -
1125 vcpu->hv_clock.tsc_timestamp;
1126 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1127 vcpu->hv_clock.tsc_to_system_mul,
1128 vcpu->hv_clock.tsc_shift);
1129 max_kernel_ns += vcpu->last_kernel_ns;
1132 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1133 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1134 &vcpu->hv_clock.tsc_shift,
1135 &vcpu->hv_clock.tsc_to_system_mul);
1136 vcpu->hw_tsc_khz = this_tsc_khz;
1139 if (max_kernel_ns > kernel_ns)
1140 kernel_ns = max_kernel_ns;
1142 /* With all the info we got, fill in the values */
1143 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1144 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1145 vcpu->last_kernel_ns = kernel_ns;
1146 vcpu->last_guest_tsc = tsc_timestamp;
1147 vcpu->hv_clock.flags = 0;
1150 * The interface expects us to write an even number signaling that the
1151 * update is finished. Since the guest won't see the intermediate
1152 * state, we just increase by 2 at the end.
1154 vcpu->hv_clock.version += 2;
1156 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1158 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1159 sizeof(vcpu->hv_clock));
1161 kunmap_atomic(shared_kaddr, KM_USER0);
1163 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1167 static bool msr_mtrr_valid(unsigned msr)
1170 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1171 case MSR_MTRRfix64K_00000:
1172 case MSR_MTRRfix16K_80000:
1173 case MSR_MTRRfix16K_A0000:
1174 case MSR_MTRRfix4K_C0000:
1175 case MSR_MTRRfix4K_C8000:
1176 case MSR_MTRRfix4K_D0000:
1177 case MSR_MTRRfix4K_D8000:
1178 case MSR_MTRRfix4K_E0000:
1179 case MSR_MTRRfix4K_E8000:
1180 case MSR_MTRRfix4K_F0000:
1181 case MSR_MTRRfix4K_F8000:
1182 case MSR_MTRRdefType:
1183 case MSR_IA32_CR_PAT:
1191 static bool valid_pat_type(unsigned t)
1193 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1196 static bool valid_mtrr_type(unsigned t)
1198 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1201 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1205 if (!msr_mtrr_valid(msr))
1208 if (msr == MSR_IA32_CR_PAT) {
1209 for (i = 0; i < 8; i++)
1210 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1213 } else if (msr == MSR_MTRRdefType) {
1216 return valid_mtrr_type(data & 0xff);
1217 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1218 for (i = 0; i < 8 ; i++)
1219 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1224 /* variable MTRRs */
1225 return valid_mtrr_type(data & 0xff);
1228 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1230 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1232 if (!mtrr_valid(vcpu, msr, data))
1235 if (msr == MSR_MTRRdefType) {
1236 vcpu->arch.mtrr_state.def_type = data;
1237 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1238 } else if (msr == MSR_MTRRfix64K_00000)
1240 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1241 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1242 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1243 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1244 else if (msr == MSR_IA32_CR_PAT)
1245 vcpu->arch.pat = data;
1246 else { /* Variable MTRRs */
1247 int idx, is_mtrr_mask;
1250 idx = (msr - 0x200) / 2;
1251 is_mtrr_mask = msr - 0x200 - 2 * idx;
1254 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1257 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1261 kvm_mmu_reset_context(vcpu);
1265 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1267 u64 mcg_cap = vcpu->arch.mcg_cap;
1268 unsigned bank_num = mcg_cap & 0xff;
1271 case MSR_IA32_MCG_STATUS:
1272 vcpu->arch.mcg_status = data;
1274 case MSR_IA32_MCG_CTL:
1275 if (!(mcg_cap & MCG_CTL_P))
1277 if (data != 0 && data != ~(u64)0)
1279 vcpu->arch.mcg_ctl = data;
1282 if (msr >= MSR_IA32_MC0_CTL &&
1283 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1284 u32 offset = msr - MSR_IA32_MC0_CTL;
1285 /* only 0 or all 1s can be written to IA32_MCi_CTL
1286 * some Linux kernels though clear bit 10 in bank 4 to
1287 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1288 * this to avoid an uncatched #GP in the guest
1290 if ((offset & 0x3) == 0 &&
1291 data != 0 && (data | (1 << 10)) != ~(u64)0)
1293 vcpu->arch.mce_banks[offset] = data;
1301 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1303 struct kvm *kvm = vcpu->kvm;
1304 int lm = is_long_mode(vcpu);
1305 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1306 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1307 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1308 : kvm->arch.xen_hvm_config.blob_size_32;
1309 u32 page_num = data & ~PAGE_MASK;
1310 u64 page_addr = data & PAGE_MASK;
1315 if (page_num >= blob_size)
1318 page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1322 if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1324 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1333 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1335 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1338 static bool kvm_hv_msr_partition_wide(u32 msr)
1342 case HV_X64_MSR_GUEST_OS_ID:
1343 case HV_X64_MSR_HYPERCALL:
1351 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1353 struct kvm *kvm = vcpu->kvm;
1356 case HV_X64_MSR_GUEST_OS_ID:
1357 kvm->arch.hv_guest_os_id = data;
1358 /* setting guest os id to zero disables hypercall page */
1359 if (!kvm->arch.hv_guest_os_id)
1360 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1362 case HV_X64_MSR_HYPERCALL: {
1367 /* if guest os id is not set hypercall should remain disabled */
1368 if (!kvm->arch.hv_guest_os_id)
1370 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1371 kvm->arch.hv_hypercall = data;
1374 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1375 addr = gfn_to_hva(kvm, gfn);
1376 if (kvm_is_error_hva(addr))
1378 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1379 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1380 if (copy_to_user((void __user *)addr, instructions, 4))
1382 kvm->arch.hv_hypercall = data;
1386 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1387 "data 0x%llx\n", msr, data);
1393 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1396 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1399 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1400 vcpu->arch.hv_vapic = data;
1403 addr = gfn_to_hva(vcpu->kvm, data >>
1404 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1405 if (kvm_is_error_hva(addr))
1407 if (clear_user((void __user *)addr, PAGE_SIZE))
1409 vcpu->arch.hv_vapic = data;
1412 case HV_X64_MSR_EOI:
1413 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1414 case HV_X64_MSR_ICR:
1415 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1416 case HV_X64_MSR_TPR:
1417 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1419 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1420 "data 0x%llx\n", msr, data);
1427 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1429 gpa_t gpa = data & ~0x3f;
1431 /* Bits 2:5 are resrved, Should be zero */
1435 vcpu->arch.apf.msr_val = data;
1437 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1438 kvm_clear_async_pf_completion_queue(vcpu);
1439 kvm_async_pf_hash_reset(vcpu);
1443 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1446 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1447 kvm_async_pf_wakeup_all(vcpu);
1451 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1455 return set_efer(vcpu, data);
1457 data &= ~(u64)0x40; /* ignore flush filter disable */
1458 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1460 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1465 case MSR_FAM10H_MMIO_CONF_BASE:
1467 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1472 case MSR_AMD64_NB_CFG:
1474 case MSR_IA32_DEBUGCTLMSR:
1476 /* We support the non-activated case already */
1478 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1479 /* Values other than LBR and BTF are vendor-specific,
1480 thus reserved and should throw a #GP */
1483 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1486 case MSR_IA32_UCODE_REV:
1487 case MSR_IA32_UCODE_WRITE:
1488 case MSR_VM_HSAVE_PA:
1489 case MSR_AMD64_PATCH_LOADER:
1491 case 0x200 ... 0x2ff:
1492 return set_msr_mtrr(vcpu, msr, data);
1493 case MSR_IA32_APICBASE:
1494 kvm_set_apic_base(vcpu, data);
1496 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1497 return kvm_x2apic_msr_write(vcpu, msr, data);
1498 case MSR_IA32_MISC_ENABLE:
1499 vcpu->arch.ia32_misc_enable_msr = data;
1501 case MSR_KVM_WALL_CLOCK_NEW:
1502 case MSR_KVM_WALL_CLOCK:
1503 vcpu->kvm->arch.wall_clock = data;
1504 kvm_write_wall_clock(vcpu->kvm, data);
1506 case MSR_KVM_SYSTEM_TIME_NEW:
1507 case MSR_KVM_SYSTEM_TIME: {
1508 if (vcpu->arch.time_page) {
1509 kvm_release_page_dirty(vcpu->arch.time_page);
1510 vcpu->arch.time_page = NULL;
1513 vcpu->arch.time = data;
1514 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1516 /* we verify if the enable bit is set... */
1520 /* ...but clean it before doing the actual write */
1521 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1523 vcpu->arch.time_page =
1524 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1526 if (is_error_page(vcpu->arch.time_page)) {
1527 kvm_release_page_clean(vcpu->arch.time_page);
1528 vcpu->arch.time_page = NULL;
1532 case MSR_KVM_ASYNC_PF_EN:
1533 if (kvm_pv_enable_async_pf(vcpu, data))
1536 case MSR_IA32_MCG_CTL:
1537 case MSR_IA32_MCG_STATUS:
1538 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1539 return set_msr_mce(vcpu, msr, data);
1541 /* Performance counters are not protected by a CPUID bit,
1542 * so we should check all of them in the generic path for the sake of
1543 * cross vendor migration.
1544 * Writing a zero into the event select MSRs disables them,
1545 * which we perfectly emulate ;-). Any other value should be at least
1546 * reported, some guests depend on them.
1548 case MSR_P6_EVNTSEL0:
1549 case MSR_P6_EVNTSEL1:
1550 case MSR_K7_EVNTSEL0:
1551 case MSR_K7_EVNTSEL1:
1552 case MSR_K7_EVNTSEL2:
1553 case MSR_K7_EVNTSEL3:
1555 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1556 "0x%x data 0x%llx\n", msr, data);
1558 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1559 * so we ignore writes to make it happy.
1561 case MSR_P6_PERFCTR0:
1562 case MSR_P6_PERFCTR1:
1563 case MSR_K7_PERFCTR0:
1564 case MSR_K7_PERFCTR1:
1565 case MSR_K7_PERFCTR2:
1566 case MSR_K7_PERFCTR3:
1567 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1568 "0x%x data 0x%llx\n", msr, data);
1570 case MSR_K7_CLK_CTL:
1572 * Ignore all writes to this no longer documented MSR.
1573 * Writes are only relevant for old K7 processors,
1574 * all pre-dating SVM, but a recommended workaround from
1575 * AMD for these chips. It is possible to speicify the
1576 * affected processor models on the command line, hence
1577 * the need to ignore the workaround.
1580 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1581 if (kvm_hv_msr_partition_wide(msr)) {
1583 mutex_lock(&vcpu->kvm->lock);
1584 r = set_msr_hyperv_pw(vcpu, msr, data);
1585 mutex_unlock(&vcpu->kvm->lock);
1588 return set_msr_hyperv(vcpu, msr, data);
1591 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1592 return xen_hvm_config(vcpu, data);
1594 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1598 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1605 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1609 * Reads an msr value (of 'msr_index') into 'pdata'.
1610 * Returns 0 on success, non-0 otherwise.
1611 * Assumes vcpu_load() was already called.
1613 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1615 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1618 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1620 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1622 if (!msr_mtrr_valid(msr))
1625 if (msr == MSR_MTRRdefType)
1626 *pdata = vcpu->arch.mtrr_state.def_type +
1627 (vcpu->arch.mtrr_state.enabled << 10);
1628 else if (msr == MSR_MTRRfix64K_00000)
1630 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1631 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1632 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1633 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1634 else if (msr == MSR_IA32_CR_PAT)
1635 *pdata = vcpu->arch.pat;
1636 else { /* Variable MTRRs */
1637 int idx, is_mtrr_mask;
1640 idx = (msr - 0x200) / 2;
1641 is_mtrr_mask = msr - 0x200 - 2 * idx;
1644 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1647 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1654 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1657 u64 mcg_cap = vcpu->arch.mcg_cap;
1658 unsigned bank_num = mcg_cap & 0xff;
1661 case MSR_IA32_P5_MC_ADDR:
1662 case MSR_IA32_P5_MC_TYPE:
1665 case MSR_IA32_MCG_CAP:
1666 data = vcpu->arch.mcg_cap;
1668 case MSR_IA32_MCG_CTL:
1669 if (!(mcg_cap & MCG_CTL_P))
1671 data = vcpu->arch.mcg_ctl;
1673 case MSR_IA32_MCG_STATUS:
1674 data = vcpu->arch.mcg_status;
1677 if (msr >= MSR_IA32_MC0_CTL &&
1678 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1679 u32 offset = msr - MSR_IA32_MC0_CTL;
1680 data = vcpu->arch.mce_banks[offset];
1689 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1692 struct kvm *kvm = vcpu->kvm;
1695 case HV_X64_MSR_GUEST_OS_ID:
1696 data = kvm->arch.hv_guest_os_id;
1698 case HV_X64_MSR_HYPERCALL:
1699 data = kvm->arch.hv_hypercall;
1702 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1710 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1715 case HV_X64_MSR_VP_INDEX: {
1718 kvm_for_each_vcpu(r, v, vcpu->kvm)
1723 case HV_X64_MSR_EOI:
1724 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1725 case HV_X64_MSR_ICR:
1726 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1727 case HV_X64_MSR_TPR:
1728 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1730 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1737 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1742 case MSR_IA32_PLATFORM_ID:
1743 case MSR_IA32_UCODE_REV:
1744 case MSR_IA32_EBL_CR_POWERON:
1745 case MSR_IA32_DEBUGCTLMSR:
1746 case MSR_IA32_LASTBRANCHFROMIP:
1747 case MSR_IA32_LASTBRANCHTOIP:
1748 case MSR_IA32_LASTINTFROMIP:
1749 case MSR_IA32_LASTINTTOIP:
1752 case MSR_VM_HSAVE_PA:
1753 case MSR_P6_PERFCTR0:
1754 case MSR_P6_PERFCTR1:
1755 case MSR_P6_EVNTSEL0:
1756 case MSR_P6_EVNTSEL1:
1757 case MSR_K7_EVNTSEL0:
1758 case MSR_K7_PERFCTR0:
1759 case MSR_K8_INT_PENDING_MSG:
1760 case MSR_AMD64_NB_CFG:
1761 case MSR_FAM10H_MMIO_CONF_BASE:
1765 data = 0x500 | KVM_NR_VAR_MTRR;
1767 case 0x200 ... 0x2ff:
1768 return get_msr_mtrr(vcpu, msr, pdata);
1769 case 0xcd: /* fsb frequency */
1773 * MSR_EBC_FREQUENCY_ID
1774 * Conservative value valid for even the basic CPU models.
1775 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1776 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1777 * and 266MHz for model 3, or 4. Set Core Clock
1778 * Frequency to System Bus Frequency Ratio to 1 (bits
1779 * 31:24) even though these are only valid for CPU
1780 * models > 2, however guests may end up dividing or
1781 * multiplying by zero otherwise.
1783 case MSR_EBC_FREQUENCY_ID:
1786 case MSR_IA32_APICBASE:
1787 data = kvm_get_apic_base(vcpu);
1789 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1790 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1792 case MSR_IA32_MISC_ENABLE:
1793 data = vcpu->arch.ia32_misc_enable_msr;
1795 case MSR_IA32_PERF_STATUS:
1796 /* TSC increment by tick */
1798 /* CPU multiplier */
1799 data |= (((uint64_t)4ULL) << 40);
1802 data = vcpu->arch.efer;
1804 case MSR_KVM_WALL_CLOCK:
1805 case MSR_KVM_WALL_CLOCK_NEW:
1806 data = vcpu->kvm->arch.wall_clock;
1808 case MSR_KVM_SYSTEM_TIME:
1809 case MSR_KVM_SYSTEM_TIME_NEW:
1810 data = vcpu->arch.time;
1812 case MSR_KVM_ASYNC_PF_EN:
1813 data = vcpu->arch.apf.msr_val;
1815 case MSR_IA32_P5_MC_ADDR:
1816 case MSR_IA32_P5_MC_TYPE:
1817 case MSR_IA32_MCG_CAP:
1818 case MSR_IA32_MCG_CTL:
1819 case MSR_IA32_MCG_STATUS:
1820 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1821 return get_msr_mce(vcpu, msr, pdata);
1822 case MSR_K7_CLK_CTL:
1824 * Provide expected ramp-up count for K7. All other
1825 * are set to zero, indicating minimum divisors for
1828 * This prevents guest kernels on AMD host with CPU
1829 * type 6, model 8 and higher from exploding due to
1830 * the rdmsr failing.
1834 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1835 if (kvm_hv_msr_partition_wide(msr)) {
1837 mutex_lock(&vcpu->kvm->lock);
1838 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1839 mutex_unlock(&vcpu->kvm->lock);
1842 return get_msr_hyperv(vcpu, msr, pdata);
1846 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1849 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1857 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1860 * Read or write a bunch of msrs. All parameters are kernel addresses.
1862 * @return number of msrs set successfully.
1864 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1865 struct kvm_msr_entry *entries,
1866 int (*do_msr)(struct kvm_vcpu *vcpu,
1867 unsigned index, u64 *data))
1871 idx = srcu_read_lock(&vcpu->kvm->srcu);
1872 for (i = 0; i < msrs->nmsrs; ++i)
1873 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1875 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1881 * Read or write a bunch of msrs. Parameters are user addresses.
1883 * @return number of msrs set successfully.
1885 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1886 int (*do_msr)(struct kvm_vcpu *vcpu,
1887 unsigned index, u64 *data),
1890 struct kvm_msrs msrs;
1891 struct kvm_msr_entry *entries;
1896 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1900 if (msrs.nmsrs >= MAX_IO_MSRS)
1904 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1905 entries = kmalloc(size, GFP_KERNEL);
1910 if (copy_from_user(entries, user_msrs->entries, size))
1913 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1918 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1929 int kvm_dev_ioctl_check_extension(long ext)
1934 case KVM_CAP_IRQCHIP:
1936 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1937 case KVM_CAP_SET_TSS_ADDR:
1938 case KVM_CAP_EXT_CPUID:
1939 case KVM_CAP_CLOCKSOURCE:
1941 case KVM_CAP_NOP_IO_DELAY:
1942 case KVM_CAP_MP_STATE:
1943 case KVM_CAP_SYNC_MMU:
1944 case KVM_CAP_USER_NMI:
1945 case KVM_CAP_REINJECT_CONTROL:
1946 case KVM_CAP_IRQ_INJECT_STATUS:
1947 case KVM_CAP_ASSIGN_DEV_IRQ:
1949 case KVM_CAP_IOEVENTFD:
1951 case KVM_CAP_PIT_STATE2:
1952 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1953 case KVM_CAP_XEN_HVM:
1954 case KVM_CAP_ADJUST_CLOCK:
1955 case KVM_CAP_VCPU_EVENTS:
1956 case KVM_CAP_HYPERV:
1957 case KVM_CAP_HYPERV_VAPIC:
1958 case KVM_CAP_HYPERV_SPIN:
1959 case KVM_CAP_PCI_SEGMENT:
1960 case KVM_CAP_DEBUGREGS:
1961 case KVM_CAP_X86_ROBUST_SINGLESTEP:
1963 case KVM_CAP_ASYNC_PF:
1966 case KVM_CAP_COALESCED_MMIO:
1967 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1970 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1972 case KVM_CAP_NR_VCPUS:
1975 case KVM_CAP_NR_MEMSLOTS:
1976 r = KVM_MEMORY_SLOTS;
1978 case KVM_CAP_PV_MMU: /* obsolete */
1985 r = KVM_MAX_MCE_BANKS;
1998 long kvm_arch_dev_ioctl(struct file *filp,
1999 unsigned int ioctl, unsigned long arg)
2001 void __user *argp = (void __user *)arg;
2005 case KVM_GET_MSR_INDEX_LIST: {
2006 struct kvm_msr_list __user *user_msr_list = argp;
2007 struct kvm_msr_list msr_list;
2011 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2014 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2015 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2018 if (n < msr_list.nmsrs)
2021 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2022 num_msrs_to_save * sizeof(u32)))
2024 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2026 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2031 case KVM_GET_SUPPORTED_CPUID: {
2032 struct kvm_cpuid2 __user *cpuid_arg = argp;
2033 struct kvm_cpuid2 cpuid;
2036 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2038 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2039 cpuid_arg->entries);
2044 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2049 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2052 mce_cap = KVM_MCE_CAP_SUPPORTED;
2054 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2066 static void wbinvd_ipi(void *garbage)
2071 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2073 return vcpu->kvm->arch.iommu_domain &&
2074 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2077 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2079 /* Address WBINVD may be executed by guest */
2080 if (need_emulate_wbinvd(vcpu)) {
2081 if (kvm_x86_ops->has_wbinvd_exit())
2082 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2083 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2084 smp_call_function_single(vcpu->cpu,
2085 wbinvd_ipi, NULL, 1);
2088 kvm_x86_ops->vcpu_load(vcpu, cpu);
2089 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2090 /* Make sure TSC doesn't go backwards */
2091 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2092 native_read_tsc() - vcpu->arch.last_host_tsc;
2094 mark_tsc_unstable("KVM discovered backwards TSC");
2095 if (check_tsc_unstable()) {
2096 kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2097 vcpu->arch.tsc_catchup = 1;
2098 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2100 if (vcpu->cpu != cpu)
2101 kvm_migrate_timers(vcpu);
2106 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2108 kvm_x86_ops->vcpu_put(vcpu);
2109 kvm_put_guest_fpu(vcpu);
2110 vcpu->arch.last_host_tsc = native_read_tsc();
2113 static int is_efer_nx(void)
2115 unsigned long long efer = 0;
2117 rdmsrl_safe(MSR_EFER, &efer);
2118 return efer & EFER_NX;
2121 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2124 struct kvm_cpuid_entry2 *e, *entry;
2127 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2128 e = &vcpu->arch.cpuid_entries[i];
2129 if (e->function == 0x80000001) {
2134 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2135 entry->edx &= ~(1 << 20);
2136 printk(KERN_INFO "kvm: guest NX capability removed\n");
2140 /* when an old userspace process fills a new kernel module */
2141 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2142 struct kvm_cpuid *cpuid,
2143 struct kvm_cpuid_entry __user *entries)
2146 struct kvm_cpuid_entry *cpuid_entries;
2149 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2152 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2156 if (copy_from_user(cpuid_entries, entries,
2157 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2159 for (i = 0; i < cpuid->nent; i++) {
2160 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2161 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2162 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2163 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2164 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2165 vcpu->arch.cpuid_entries[i].index = 0;
2166 vcpu->arch.cpuid_entries[i].flags = 0;
2167 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2168 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2169 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2171 vcpu->arch.cpuid_nent = cpuid->nent;
2172 cpuid_fix_nx_cap(vcpu);
2174 kvm_apic_set_version(vcpu);
2175 kvm_x86_ops->cpuid_update(vcpu);
2179 vfree(cpuid_entries);
2184 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2185 struct kvm_cpuid2 *cpuid,
2186 struct kvm_cpuid_entry2 __user *entries)
2191 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2194 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2195 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2197 vcpu->arch.cpuid_nent = cpuid->nent;
2198 kvm_apic_set_version(vcpu);
2199 kvm_x86_ops->cpuid_update(vcpu);
2207 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2208 struct kvm_cpuid2 *cpuid,
2209 struct kvm_cpuid_entry2 __user *entries)
2214 if (cpuid->nent < vcpu->arch.cpuid_nent)
2217 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2218 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2223 cpuid->nent = vcpu->arch.cpuid_nent;
2227 static void cpuid_mask(u32 *word, int wordnum)
2229 *word &= boot_cpu_data.x86_capability[wordnum];
2232 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2235 entry->function = function;
2236 entry->index = index;
2237 cpuid_count(entry->function, entry->index,
2238 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2242 #define F(x) bit(X86_FEATURE_##x)
2244 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2245 u32 index, int *nent, int maxnent)
2247 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2248 #ifdef CONFIG_X86_64
2249 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2251 unsigned f_lm = F(LM);
2253 unsigned f_gbpages = 0;
2256 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2259 const u32 kvm_supported_word0_x86_features =
2260 F(FPU) | F(VME) | F(DE) | F(PSE) |
2261 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2262 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2263 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2264 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2265 0 /* Reserved, DS, ACPI */ | F(MMX) |
2266 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2267 0 /* HTT, TM, Reserved, PBE */;
2268 /* cpuid 0x80000001.edx */
2269 const u32 kvm_supported_word1_x86_features =
2270 F(FPU) | F(VME) | F(DE) | F(PSE) |
2271 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2272 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2273 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2274 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2275 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2276 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2277 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2279 const u32 kvm_supported_word4_x86_features =
2280 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
2281 0 /* DS-CPL, VMX, SMX, EST */ |
2282 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2283 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2284 0 /* Reserved, DCA */ | F(XMM4_1) |
2285 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2286 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2288 /* cpuid 0x80000001.ecx */
2289 const u32 kvm_supported_word6_x86_features =
2290 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2291 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2292 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2293 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2295 /* all calls to cpuid_count() should be made on the same cpu */
2297 do_cpuid_1_ent(entry, function, index);
2302 entry->eax = min(entry->eax, (u32)0xd);
2305 entry->edx &= kvm_supported_word0_x86_features;
2306 cpuid_mask(&entry->edx, 0);
2307 entry->ecx &= kvm_supported_word4_x86_features;
2308 cpuid_mask(&entry->ecx, 4);
2309 /* we support x2apic emulation even if host does not support
2310 * it since we emulate x2apic in software */
2311 entry->ecx |= F(X2APIC);
2313 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2314 * may return different values. This forces us to get_cpu() before
2315 * issuing the first command, and also to emulate this annoying behavior
2316 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2318 int t, times = entry->eax & 0xff;
2320 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2321 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2322 for (t = 1; t < times && *nent < maxnent; ++t) {
2323 do_cpuid_1_ent(&entry[t], function, 0);
2324 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2329 /* function 4 and 0xb have additional index. */
2333 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2334 /* read more entries until cache_type is zero */
2335 for (i = 1; *nent < maxnent; ++i) {
2336 cache_type = entry[i - 1].eax & 0x1f;
2339 do_cpuid_1_ent(&entry[i], function, i);
2341 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2349 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2350 /* read more entries until level_type is zero */
2351 for (i = 1; *nent < maxnent; ++i) {
2352 level_type = entry[i - 1].ecx & 0xff00;
2355 do_cpuid_1_ent(&entry[i], function, i);
2357 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2365 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2366 for (i = 1; *nent < maxnent; ++i) {
2367 if (entry[i - 1].eax == 0 && i != 2)
2369 do_cpuid_1_ent(&entry[i], function, i);
2371 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2376 case KVM_CPUID_SIGNATURE: {
2377 char signature[12] = "KVMKVMKVM\0\0";
2378 u32 *sigptr = (u32 *)signature;
2380 entry->ebx = sigptr[0];
2381 entry->ecx = sigptr[1];
2382 entry->edx = sigptr[2];
2385 case KVM_CPUID_FEATURES:
2386 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2387 (1 << KVM_FEATURE_NOP_IO_DELAY) |
2388 (1 << KVM_FEATURE_CLOCKSOURCE2) |
2389 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2395 entry->eax = min(entry->eax, 0x8000001a);
2398 entry->edx &= kvm_supported_word1_x86_features;
2399 cpuid_mask(&entry->edx, 1);
2400 entry->ecx &= kvm_supported_word6_x86_features;
2401 cpuid_mask(&entry->ecx, 6);
2405 kvm_x86_ops->set_supported_cpuid(function, entry);
2412 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2413 struct kvm_cpuid_entry2 __user *entries)
2415 struct kvm_cpuid_entry2 *cpuid_entries;
2416 int limit, nent = 0, r = -E2BIG;
2419 if (cpuid->nent < 1)
2421 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2422 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2424 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2428 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2429 limit = cpuid_entries[0].eax;
2430 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2431 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2432 &nent, cpuid->nent);
2434 if (nent >= cpuid->nent)
2437 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2438 limit = cpuid_entries[nent - 1].eax;
2439 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2440 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2441 &nent, cpuid->nent);
2446 if (nent >= cpuid->nent)
2449 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2453 if (nent >= cpuid->nent)
2456 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2460 if (nent >= cpuid->nent)
2464 if (copy_to_user(entries, cpuid_entries,
2465 nent * sizeof(struct kvm_cpuid_entry2)))
2471 vfree(cpuid_entries);
2476 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2477 struct kvm_lapic_state *s)
2479 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2484 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2485 struct kvm_lapic_state *s)
2487 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2488 kvm_apic_post_state_restore(vcpu);
2489 update_cr8_intercept(vcpu);
2494 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2495 struct kvm_interrupt *irq)
2497 if (irq->irq < 0 || irq->irq >= 256)
2499 if (irqchip_in_kernel(vcpu->kvm))
2502 kvm_queue_interrupt(vcpu, irq->irq, false);
2503 kvm_make_request(KVM_REQ_EVENT, vcpu);
2508 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2510 kvm_inject_nmi(vcpu);
2515 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2516 struct kvm_tpr_access_ctl *tac)
2520 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2524 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2528 unsigned bank_num = mcg_cap & 0xff, bank;
2531 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2533 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2536 vcpu->arch.mcg_cap = mcg_cap;
2537 /* Init IA32_MCG_CTL to all 1s */
2538 if (mcg_cap & MCG_CTL_P)
2539 vcpu->arch.mcg_ctl = ~(u64)0;
2540 /* Init IA32_MCi_CTL to all 1s */
2541 for (bank = 0; bank < bank_num; bank++)
2542 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2547 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2548 struct kvm_x86_mce *mce)
2550 u64 mcg_cap = vcpu->arch.mcg_cap;
2551 unsigned bank_num = mcg_cap & 0xff;
2552 u64 *banks = vcpu->arch.mce_banks;
2554 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2557 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2558 * reporting is disabled
2560 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2561 vcpu->arch.mcg_ctl != ~(u64)0)
2563 banks += 4 * mce->bank;
2565 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2566 * reporting is disabled for the bank
2568 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2570 if (mce->status & MCI_STATUS_UC) {
2571 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2572 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2573 printk(KERN_DEBUG "kvm: set_mce: "
2574 "injects mce exception while "
2575 "previous one is in progress!\n");
2576 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2579 if (banks[1] & MCI_STATUS_VAL)
2580 mce->status |= MCI_STATUS_OVER;
2581 banks[2] = mce->addr;
2582 banks[3] = mce->misc;
2583 vcpu->arch.mcg_status = mce->mcg_status;
2584 banks[1] = mce->status;
2585 kvm_queue_exception(vcpu, MC_VECTOR);
2586 } else if (!(banks[1] & MCI_STATUS_VAL)
2587 || !(banks[1] & MCI_STATUS_UC)) {
2588 if (banks[1] & MCI_STATUS_VAL)
2589 mce->status |= MCI_STATUS_OVER;
2590 banks[2] = mce->addr;
2591 banks[3] = mce->misc;
2592 banks[1] = mce->status;
2594 banks[1] |= MCI_STATUS_OVER;
2598 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2599 struct kvm_vcpu_events *events)
2601 events->exception.injected =
2602 vcpu->arch.exception.pending &&
2603 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2604 events->exception.nr = vcpu->arch.exception.nr;
2605 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2606 events->exception.pad = 0;
2607 events->exception.error_code = vcpu->arch.exception.error_code;
2609 events->interrupt.injected =
2610 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2611 events->interrupt.nr = vcpu->arch.interrupt.nr;
2612 events->interrupt.soft = 0;
2613 events->interrupt.shadow =
2614 kvm_x86_ops->get_interrupt_shadow(vcpu,
2615 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2617 events->nmi.injected = vcpu->arch.nmi_injected;
2618 events->nmi.pending = vcpu->arch.nmi_pending;
2619 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2620 events->nmi.pad = 0;
2622 events->sipi_vector = vcpu->arch.sipi_vector;
2624 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2625 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2626 | KVM_VCPUEVENT_VALID_SHADOW);
2627 memset(&events->reserved, 0, sizeof(events->reserved));
2630 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2631 struct kvm_vcpu_events *events)
2633 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2634 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2635 | KVM_VCPUEVENT_VALID_SHADOW))
2638 vcpu->arch.exception.pending = events->exception.injected;
2639 vcpu->arch.exception.nr = events->exception.nr;
2640 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2641 vcpu->arch.exception.error_code = events->exception.error_code;
2643 vcpu->arch.interrupt.pending = events->interrupt.injected;
2644 vcpu->arch.interrupt.nr = events->interrupt.nr;
2645 vcpu->arch.interrupt.soft = events->interrupt.soft;
2646 if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
2647 kvm_pic_clear_isr_ack(vcpu->kvm);
2648 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2649 kvm_x86_ops->set_interrupt_shadow(vcpu,
2650 events->interrupt.shadow);
2652 vcpu->arch.nmi_injected = events->nmi.injected;
2653 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2654 vcpu->arch.nmi_pending = events->nmi.pending;
2655 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2657 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2658 vcpu->arch.sipi_vector = events->sipi_vector;
2660 kvm_make_request(KVM_REQ_EVENT, vcpu);
2665 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2666 struct kvm_debugregs *dbgregs)
2668 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2669 dbgregs->dr6 = vcpu->arch.dr6;
2670 dbgregs->dr7 = vcpu->arch.dr7;
2672 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2675 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2676 struct kvm_debugregs *dbgregs)
2681 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2682 vcpu->arch.dr6 = dbgregs->dr6;
2683 vcpu->arch.dr7 = dbgregs->dr7;
2688 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2689 struct kvm_xsave *guest_xsave)
2692 memcpy(guest_xsave->region,
2693 &vcpu->arch.guest_fpu.state->xsave,
2696 memcpy(guest_xsave->region,
2697 &vcpu->arch.guest_fpu.state->fxsave,
2698 sizeof(struct i387_fxsave_struct));
2699 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2704 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2705 struct kvm_xsave *guest_xsave)
2708 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2711 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2712 guest_xsave->region, xstate_size);
2714 if (xstate_bv & ~XSTATE_FPSSE)
2716 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2717 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2722 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2723 struct kvm_xcrs *guest_xcrs)
2725 if (!cpu_has_xsave) {
2726 guest_xcrs->nr_xcrs = 0;
2730 guest_xcrs->nr_xcrs = 1;
2731 guest_xcrs->flags = 0;
2732 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2733 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2736 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2737 struct kvm_xcrs *guest_xcrs)
2744 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2747 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2748 /* Only support XCR0 currently */
2749 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2750 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2751 guest_xcrs->xcrs[0].value);
2759 long kvm_arch_vcpu_ioctl(struct file *filp,
2760 unsigned int ioctl, unsigned long arg)
2762 struct kvm_vcpu *vcpu = filp->private_data;
2763 void __user *argp = (void __user *)arg;
2766 struct kvm_lapic_state *lapic;
2767 struct kvm_xsave *xsave;
2768 struct kvm_xcrs *xcrs;
2774 case KVM_GET_LAPIC: {
2776 if (!vcpu->arch.apic)
2778 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2783 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
2787 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
2792 case KVM_SET_LAPIC: {
2794 if (!vcpu->arch.apic)
2796 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2801 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
2803 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
2809 case KVM_INTERRUPT: {
2810 struct kvm_interrupt irq;
2813 if (copy_from_user(&irq, argp, sizeof irq))
2815 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2822 r = kvm_vcpu_ioctl_nmi(vcpu);
2828 case KVM_SET_CPUID: {
2829 struct kvm_cpuid __user *cpuid_arg = argp;
2830 struct kvm_cpuid cpuid;
2833 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2835 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2840 case KVM_SET_CPUID2: {
2841 struct kvm_cpuid2 __user *cpuid_arg = argp;
2842 struct kvm_cpuid2 cpuid;
2845 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2847 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2848 cpuid_arg->entries);
2853 case KVM_GET_CPUID2: {
2854 struct kvm_cpuid2 __user *cpuid_arg = argp;
2855 struct kvm_cpuid2 cpuid;
2858 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2860 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2861 cpuid_arg->entries);
2865 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2871 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2874 r = msr_io(vcpu, argp, do_set_msr, 0);
2876 case KVM_TPR_ACCESS_REPORTING: {
2877 struct kvm_tpr_access_ctl tac;
2880 if (copy_from_user(&tac, argp, sizeof tac))
2882 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2886 if (copy_to_user(argp, &tac, sizeof tac))
2891 case KVM_SET_VAPIC_ADDR: {
2892 struct kvm_vapic_addr va;
2895 if (!irqchip_in_kernel(vcpu->kvm))
2898 if (copy_from_user(&va, argp, sizeof va))
2901 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2904 case KVM_X86_SETUP_MCE: {
2908 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2910 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2913 case KVM_X86_SET_MCE: {
2914 struct kvm_x86_mce mce;
2917 if (copy_from_user(&mce, argp, sizeof mce))
2919 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2922 case KVM_GET_VCPU_EVENTS: {
2923 struct kvm_vcpu_events events;
2925 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2928 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2933 case KVM_SET_VCPU_EVENTS: {
2934 struct kvm_vcpu_events events;
2937 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2940 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2943 case KVM_GET_DEBUGREGS: {
2944 struct kvm_debugregs dbgregs;
2946 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2949 if (copy_to_user(argp, &dbgregs,
2950 sizeof(struct kvm_debugregs)))
2955 case KVM_SET_DEBUGREGS: {
2956 struct kvm_debugregs dbgregs;
2959 if (copy_from_user(&dbgregs, argp,
2960 sizeof(struct kvm_debugregs)))
2963 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2966 case KVM_GET_XSAVE: {
2967 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2972 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
2975 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
2980 case KVM_SET_XSAVE: {
2981 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2987 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
2990 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
2993 case KVM_GET_XCRS: {
2994 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2999 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3002 if (copy_to_user(argp, u.xcrs,
3003 sizeof(struct kvm_xcrs)))
3008 case KVM_SET_XCRS: {
3009 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3015 if (copy_from_user(u.xcrs, argp,
3016 sizeof(struct kvm_xcrs)))
3019 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3030 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3034 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3036 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3040 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3043 kvm->arch.ept_identity_map_addr = ident_addr;
3047 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3048 u32 kvm_nr_mmu_pages)
3050 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3053 mutex_lock(&kvm->slots_lock);
3054 spin_lock(&kvm->mmu_lock);
3056 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3057 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3059 spin_unlock(&kvm->mmu_lock);
3060 mutex_unlock(&kvm->slots_lock);
3064 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3066 return kvm->arch.n_max_mmu_pages;
3069 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3074 switch (chip->chip_id) {
3075 case KVM_IRQCHIP_PIC_MASTER:
3076 memcpy(&chip->chip.pic,
3077 &pic_irqchip(kvm)->pics[0],
3078 sizeof(struct kvm_pic_state));
3080 case KVM_IRQCHIP_PIC_SLAVE:
3081 memcpy(&chip->chip.pic,
3082 &pic_irqchip(kvm)->pics[1],
3083 sizeof(struct kvm_pic_state));
3085 case KVM_IRQCHIP_IOAPIC:
3086 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3095 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3100 switch (chip->chip_id) {
3101 case KVM_IRQCHIP_PIC_MASTER:
3102 spin_lock(&pic_irqchip(kvm)->lock);
3103 memcpy(&pic_irqchip(kvm)->pics[0],
3105 sizeof(struct kvm_pic_state));
3106 spin_unlock(&pic_irqchip(kvm)->lock);
3108 case KVM_IRQCHIP_PIC_SLAVE:
3109 spin_lock(&pic_irqchip(kvm)->lock);
3110 memcpy(&pic_irqchip(kvm)->pics[1],
3112 sizeof(struct kvm_pic_state));
3113 spin_unlock(&pic_irqchip(kvm)->lock);
3115 case KVM_IRQCHIP_IOAPIC:
3116 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3122 kvm_pic_update_irq(pic_irqchip(kvm));
3126 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3130 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3131 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3132 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3136 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3140 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3141 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3142 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3143 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3147 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3151 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3152 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3153 sizeof(ps->channels));
3154 ps->flags = kvm->arch.vpit->pit_state.flags;
3155 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3156 memset(&ps->reserved, 0, sizeof(ps->reserved));
3160 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3162 int r = 0, start = 0;
3163 u32 prev_legacy, cur_legacy;
3164 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3165 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3166 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3167 if (!prev_legacy && cur_legacy)
3169 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3170 sizeof(kvm->arch.vpit->pit_state.channels));
3171 kvm->arch.vpit->pit_state.flags = ps->flags;
3172 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3173 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3177 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3178 struct kvm_reinject_control *control)
3180 if (!kvm->arch.vpit)
3182 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3183 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3184 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3189 * Get (and clear) the dirty memory log for a memory slot.
3191 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3192 struct kvm_dirty_log *log)
3195 struct kvm_memory_slot *memslot;
3197 unsigned long is_dirty = 0;
3199 mutex_lock(&kvm->slots_lock);
3202 if (log->slot >= KVM_MEMORY_SLOTS)
3205 memslot = &kvm->memslots->memslots[log->slot];
3207 if (!memslot->dirty_bitmap)
3210 n = kvm_dirty_bitmap_bytes(memslot);
3212 for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3213 is_dirty = memslot->dirty_bitmap[i];
3215 /* If nothing is dirty, don't bother messing with page tables. */
3217 struct kvm_memslots *slots, *old_slots;
3218 unsigned long *dirty_bitmap;
3220 dirty_bitmap = memslot->dirty_bitmap_head;
3221 if (memslot->dirty_bitmap == dirty_bitmap)
3222 dirty_bitmap += n / sizeof(long);
3223 memset(dirty_bitmap, 0, n);
3226 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3229 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3230 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3231 slots->generation++;
3233 old_slots = kvm->memslots;
3234 rcu_assign_pointer(kvm->memslots, slots);
3235 synchronize_srcu_expedited(&kvm->srcu);
3236 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3239 spin_lock(&kvm->mmu_lock);
3240 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3241 spin_unlock(&kvm->mmu_lock);
3244 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
3248 if (clear_user(log->dirty_bitmap, n))
3254 mutex_unlock(&kvm->slots_lock);
3258 long kvm_arch_vm_ioctl(struct file *filp,
3259 unsigned int ioctl, unsigned long arg)
3261 struct kvm *kvm = filp->private_data;
3262 void __user *argp = (void __user *)arg;
3265 * This union makes it completely explicit to gcc-3.x
3266 * that these two variables' stack usage should be
3267 * combined, not added together.
3270 struct kvm_pit_state ps;
3271 struct kvm_pit_state2 ps2;
3272 struct kvm_pit_config pit_config;
3276 case KVM_SET_TSS_ADDR:
3277 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3281 case KVM_SET_IDENTITY_MAP_ADDR: {
3285 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3287 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3292 case KVM_SET_NR_MMU_PAGES:
3293 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3297 case KVM_GET_NR_MMU_PAGES:
3298 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3300 case KVM_CREATE_IRQCHIP: {
3301 struct kvm_pic *vpic;
3303 mutex_lock(&kvm->lock);
3306 goto create_irqchip_unlock;
3308 vpic = kvm_create_pic(kvm);
3310 r = kvm_ioapic_init(kvm);
3312 mutex_lock(&kvm->slots_lock);
3313 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3315 mutex_unlock(&kvm->slots_lock);
3317 goto create_irqchip_unlock;
3320 goto create_irqchip_unlock;
3322 kvm->arch.vpic = vpic;
3324 r = kvm_setup_default_irq_routing(kvm);
3326 mutex_lock(&kvm->slots_lock);
3327 mutex_lock(&kvm->irq_lock);
3328 kvm_ioapic_destroy(kvm);
3329 kvm_destroy_pic(kvm);
3330 mutex_unlock(&kvm->irq_lock);
3331 mutex_unlock(&kvm->slots_lock);
3333 create_irqchip_unlock:
3334 mutex_unlock(&kvm->lock);
3337 case KVM_CREATE_PIT:
3338 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3340 case KVM_CREATE_PIT2:
3342 if (copy_from_user(&u.pit_config, argp,
3343 sizeof(struct kvm_pit_config)))
3346 mutex_lock(&kvm->slots_lock);
3349 goto create_pit_unlock;
3351 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3355 mutex_unlock(&kvm->slots_lock);
3357 case KVM_IRQ_LINE_STATUS:
3358 case KVM_IRQ_LINE: {
3359 struct kvm_irq_level irq_event;
3362 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3365 if (irqchip_in_kernel(kvm)) {
3367 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3368 irq_event.irq, irq_event.level);
3369 if (ioctl == KVM_IRQ_LINE_STATUS) {
3371 irq_event.status = status;
3372 if (copy_to_user(argp, &irq_event,
3380 case KVM_GET_IRQCHIP: {
3381 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3382 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3388 if (copy_from_user(chip, argp, sizeof *chip))
3389 goto get_irqchip_out;
3391 if (!irqchip_in_kernel(kvm))
3392 goto get_irqchip_out;
3393 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3395 goto get_irqchip_out;
3397 if (copy_to_user(argp, chip, sizeof *chip))
3398 goto get_irqchip_out;
3406 case KVM_SET_IRQCHIP: {
3407 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3408 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3414 if (copy_from_user(chip, argp, sizeof *chip))
3415 goto set_irqchip_out;
3417 if (!irqchip_in_kernel(kvm))
3418 goto set_irqchip_out;
3419 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3421 goto set_irqchip_out;
3431 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3434 if (!kvm->arch.vpit)
3436 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3440 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3447 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3450 if (!kvm->arch.vpit)
3452 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3458 case KVM_GET_PIT2: {
3460 if (!kvm->arch.vpit)
3462 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3466 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3471 case KVM_SET_PIT2: {
3473 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3476 if (!kvm->arch.vpit)
3478 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3484 case KVM_REINJECT_CONTROL: {
3485 struct kvm_reinject_control control;
3487 if (copy_from_user(&control, argp, sizeof(control)))
3489 r = kvm_vm_ioctl_reinject(kvm, &control);
3495 case KVM_XEN_HVM_CONFIG: {
3497 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3498 sizeof(struct kvm_xen_hvm_config)))
3501 if (kvm->arch.xen_hvm_config.flags)
3506 case KVM_SET_CLOCK: {
3507 struct kvm_clock_data user_ns;
3512 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3520 local_irq_disable();
3521 now_ns = get_kernel_ns();
3522 delta = user_ns.clock - now_ns;
3524 kvm->arch.kvmclock_offset = delta;
3527 case KVM_GET_CLOCK: {
3528 struct kvm_clock_data user_ns;
3531 local_irq_disable();
3532 now_ns = get_kernel_ns();
3533 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3536 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3539 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3552 static void kvm_init_msr_list(void)
3557 /* skip the first msrs in the list. KVM-specific */
3558 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3559 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3562 msrs_to_save[j] = msrs_to_save[i];
3565 num_msrs_to_save = j;
3568 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3571 if (vcpu->arch.apic &&
3572 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
3575 return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3578 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3580 if (vcpu->arch.apic &&
3581 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
3584 return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3587 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3588 struct kvm_segment *var, int seg)
3590 kvm_x86_ops->set_segment(vcpu, var, seg);
3593 void kvm_get_segment(struct kvm_vcpu *vcpu,
3594 struct kvm_segment *var, int seg)
3596 kvm_x86_ops->get_segment(vcpu, var, seg);
3599 static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3604 static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3607 struct x86_exception exception;
3609 BUG_ON(!mmu_is_nested(vcpu));
3611 /* NPT walks are always user-walks */
3612 access |= PFERR_USER_MASK;
3613 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3618 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3619 struct x86_exception *exception)
3621 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3622 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3625 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3626 struct x86_exception *exception)
3628 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3629 access |= PFERR_FETCH_MASK;
3630 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3633 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3634 struct x86_exception *exception)
3636 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3637 access |= PFERR_WRITE_MASK;
3638 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3641 /* uses this to access any guest's mapped memory without checking CPL */
3642 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3643 struct x86_exception *exception)
3645 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3648 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3649 struct kvm_vcpu *vcpu, u32 access,
3650 struct x86_exception *exception)
3653 int r = X86EMUL_CONTINUE;
3656 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3658 unsigned offset = addr & (PAGE_SIZE-1);
3659 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3662 if (gpa == UNMAPPED_GVA)
3663 return X86EMUL_PROPAGATE_FAULT;
3664 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3666 r = X86EMUL_IO_NEEDED;
3678 /* used for instruction fetching */
3679 static int kvm_fetch_guest_virt(gva_t addr, void *val, unsigned int bytes,
3680 struct kvm_vcpu *vcpu,
3681 struct x86_exception *exception)
3683 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3684 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3685 access | PFERR_FETCH_MASK,
3689 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3690 struct kvm_vcpu *vcpu,
3691 struct x86_exception *exception)
3693 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3694 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3698 static int kvm_read_guest_virt_system(gva_t addr, void *val, unsigned int bytes,
3699 struct kvm_vcpu *vcpu,
3700 struct x86_exception *exception)
3702 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3705 static int kvm_write_guest_virt_system(gva_t addr, void *val,
3707 struct kvm_vcpu *vcpu,
3708 struct x86_exception *exception)
3711 int r = X86EMUL_CONTINUE;
3714 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3717 unsigned offset = addr & (PAGE_SIZE-1);
3718 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3721 if (gpa == UNMAPPED_GVA)
3722 return X86EMUL_PROPAGATE_FAULT;
3723 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3725 r = X86EMUL_IO_NEEDED;
3737 static int emulator_read_emulated(unsigned long addr,
3740 struct x86_exception *exception,
3741 struct kvm_vcpu *vcpu)
3745 if (vcpu->mmio_read_completed) {
3746 memcpy(val, vcpu->mmio_data, bytes);
3747 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3748 vcpu->mmio_phys_addr, *(u64 *)val);
3749 vcpu->mmio_read_completed = 0;
3750 return X86EMUL_CONTINUE;
3753 gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, exception);
3755 if (gpa == UNMAPPED_GVA)
3756 return X86EMUL_PROPAGATE_FAULT;
3758 /* For APIC access vmexit */
3759 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3762 if (kvm_read_guest_virt(addr, val, bytes, vcpu, exception)
3763 == X86EMUL_CONTINUE)
3764 return X86EMUL_CONTINUE;
3768 * Is this MMIO handled locally?
3770 if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
3771 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
3772 return X86EMUL_CONTINUE;
3775 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3777 vcpu->mmio_needed = 1;
3778 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3779 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3780 vcpu->run->mmio.len = vcpu->mmio_size = bytes;
3781 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 0;
3783 return X86EMUL_IO_NEEDED;
3786 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3787 const void *val, int bytes)
3791 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3794 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3798 static int emulator_write_emulated_onepage(unsigned long addr,
3801 struct x86_exception *exception,
3802 struct kvm_vcpu *vcpu)
3806 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, exception);
3808 if (gpa == UNMAPPED_GVA)
3809 return X86EMUL_PROPAGATE_FAULT;
3811 /* For APIC access vmexit */
3812 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3815 if (emulator_write_phys(vcpu, gpa, val, bytes))
3816 return X86EMUL_CONTINUE;
3819 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3821 * Is this MMIO handled locally?
3823 if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
3824 return X86EMUL_CONTINUE;
3826 vcpu->mmio_needed = 1;
3827 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3828 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3829 vcpu->run->mmio.len = vcpu->mmio_size = bytes;
3830 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 1;
3831 memcpy(vcpu->run->mmio.data, val, bytes);
3833 return X86EMUL_CONTINUE;
3836 int emulator_write_emulated(unsigned long addr,
3839 struct x86_exception *exception,
3840 struct kvm_vcpu *vcpu)
3842 /* Crossing a page boundary? */
3843 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3846 now = -addr & ~PAGE_MASK;
3847 rc = emulator_write_emulated_onepage(addr, val, now, exception,
3849 if (rc != X86EMUL_CONTINUE)
3855 return emulator_write_emulated_onepage(addr, val, bytes, exception,
3859 #define CMPXCHG_TYPE(t, ptr, old, new) \
3860 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3862 #ifdef CONFIG_X86_64
3863 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3865 # define CMPXCHG64(ptr, old, new) \
3866 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
3869 static int emulator_cmpxchg_emulated(unsigned long addr,
3873 struct x86_exception *exception,
3874 struct kvm_vcpu *vcpu)
3881 /* guests cmpxchg8b have to be emulated atomically */
3882 if (bytes > 8 || (bytes & (bytes - 1)))
3885 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3887 if (gpa == UNMAPPED_GVA ||
3888 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3891 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3894 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3895 if (is_error_page(page)) {
3896 kvm_release_page_clean(page);
3900 kaddr = kmap_atomic(page, KM_USER0);
3901 kaddr += offset_in_page(gpa);
3904 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
3907 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
3910 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
3913 exchanged = CMPXCHG64(kaddr, old, new);
3918 kunmap_atomic(kaddr, KM_USER0);
3919 kvm_release_page_dirty(page);
3922 return X86EMUL_CMPXCHG_FAILED;
3924 kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
3926 return X86EMUL_CONTINUE;
3929 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3931 return emulator_write_emulated(addr, new, bytes, exception, vcpu);
3934 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3936 /* TODO: String I/O for in kernel device */
3939 if (vcpu->arch.pio.in)
3940 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3941 vcpu->arch.pio.size, pd);
3943 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3944 vcpu->arch.pio.port, vcpu->arch.pio.size,
3950 static int emulator_pio_in_emulated(int size, unsigned short port, void *val,
3951 unsigned int count, struct kvm_vcpu *vcpu)
3953 if (vcpu->arch.pio.count)
3956 trace_kvm_pio(0, port, size, count);
3958 vcpu->arch.pio.port = port;
3959 vcpu->arch.pio.in = 1;
3960 vcpu->arch.pio.count = count;
3961 vcpu->arch.pio.size = size;
3963 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3965 memcpy(val, vcpu->arch.pio_data, size * count);
3966 vcpu->arch.pio.count = 0;
3970 vcpu->run->exit_reason = KVM_EXIT_IO;
3971 vcpu->run->io.direction = KVM_EXIT_IO_IN;
3972 vcpu->run->io.size = size;
3973 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3974 vcpu->run->io.count = count;
3975 vcpu->run->io.port = port;
3980 static int emulator_pio_out_emulated(int size, unsigned short port,
3981 const void *val, unsigned int count,
3982 struct kvm_vcpu *vcpu)
3984 trace_kvm_pio(1, port, size, count);
3986 vcpu->arch.pio.port = port;
3987 vcpu->arch.pio.in = 0;
3988 vcpu->arch.pio.count = count;
3989 vcpu->arch.pio.size = size;
3991 memcpy(vcpu->arch.pio_data, val, size * count);
3993 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3994 vcpu->arch.pio.count = 0;
3998 vcpu->run->exit_reason = KVM_EXIT_IO;
3999 vcpu->run->io.direction = KVM_EXIT_IO_OUT;
4000 vcpu->run->io.size = size;
4001 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4002 vcpu->run->io.count = count;
4003 vcpu->run->io.port = port;
4008 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4010 return kvm_x86_ops->get_segment_base(vcpu, seg);
4013 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
4015 kvm_mmu_invlpg(vcpu, address);
4016 return X86EMUL_CONTINUE;
4019 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4021 if (!need_emulate_wbinvd(vcpu))
4022 return X86EMUL_CONTINUE;
4024 if (kvm_x86_ops->has_wbinvd_exit()) {
4025 int cpu = get_cpu();
4027 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4028 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4029 wbinvd_ipi, NULL, 1);
4031 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4034 return X86EMUL_CONTINUE;
4036 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4038 int emulate_clts(struct kvm_vcpu *vcpu)
4040 kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4041 kvm_x86_ops->fpu_activate(vcpu);
4042 return X86EMUL_CONTINUE;
4045 int emulator_get_dr(int dr, unsigned long *dest, struct kvm_vcpu *vcpu)
4047 return _kvm_get_dr(vcpu, dr, dest);
4050 int emulator_set_dr(int dr, unsigned long value, struct kvm_vcpu *vcpu)
4053 return __kvm_set_dr(vcpu, dr, value);
4056 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4058 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4061 static unsigned long emulator_get_cr(int cr, struct kvm_vcpu *vcpu)
4063 unsigned long value;
4067 value = kvm_read_cr0(vcpu);
4070 value = vcpu->arch.cr2;
4073 value = vcpu->arch.cr3;
4076 value = kvm_read_cr4(vcpu);
4079 value = kvm_get_cr8(vcpu);
4082 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4089 static int emulator_set_cr(int cr, unsigned long val, struct kvm_vcpu *vcpu)
4095 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4098 vcpu->arch.cr2 = val;
4101 res = kvm_set_cr3(vcpu, val);
4104 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4107 res = __kvm_set_cr8(vcpu, val & 0xfUL);
4110 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4117 static int emulator_get_cpl(struct kvm_vcpu *vcpu)
4119 return kvm_x86_ops->get_cpl(vcpu);
4122 static void emulator_get_gdt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
4124 kvm_x86_ops->get_gdt(vcpu, dt);
4127 static void emulator_get_idt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
4129 kvm_x86_ops->get_idt(vcpu, dt);
4132 static unsigned long emulator_get_cached_segment_base(int seg,
4133 struct kvm_vcpu *vcpu)
4135 return get_segment_base(vcpu, seg);
4138 static bool emulator_get_cached_descriptor(struct desc_struct *desc, int seg,
4139 struct kvm_vcpu *vcpu)
4141 struct kvm_segment var;
4143 kvm_get_segment(vcpu, &var, seg);
4150 set_desc_limit(desc, var.limit);
4151 set_desc_base(desc, (unsigned long)var.base);
4152 desc->type = var.type;
4154 desc->dpl = var.dpl;
4155 desc->p = var.present;
4156 desc->avl = var.avl;
4164 static void emulator_set_cached_descriptor(struct desc_struct *desc, int seg,
4165 struct kvm_vcpu *vcpu)
4167 struct kvm_segment var;
4169 /* needed to preserve selector */
4170 kvm_get_segment(vcpu, &var, seg);
4172 var.base = get_desc_base(desc);
4173 var.limit = get_desc_limit(desc);
4175 var.limit = (var.limit << 12) | 0xfff;
4176 var.type = desc->type;
4177 var.present = desc->p;
4178 var.dpl = desc->dpl;
4183 var.avl = desc->avl;
4184 var.present = desc->p;
4185 var.unusable = !var.present;
4188 kvm_set_segment(vcpu, &var, seg);
4192 static u16 emulator_get_segment_selector(int seg, struct kvm_vcpu *vcpu)
4194 struct kvm_segment kvm_seg;
4196 kvm_get_segment(vcpu, &kvm_seg, seg);
4197 return kvm_seg.selector;
4200 static void emulator_set_segment_selector(u16 sel, int seg,
4201 struct kvm_vcpu *vcpu)
4203 struct kvm_segment kvm_seg;
4205 kvm_get_segment(vcpu, &kvm_seg, seg);
4206 kvm_seg.selector = sel;
4207 kvm_set_segment(vcpu, &kvm_seg, seg);
4210 static struct x86_emulate_ops emulate_ops = {
4211 .read_std = kvm_read_guest_virt_system,
4212 .write_std = kvm_write_guest_virt_system,
4213 .fetch = kvm_fetch_guest_virt,
4214 .read_emulated = emulator_read_emulated,
4215 .write_emulated = emulator_write_emulated,
4216 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4217 .pio_in_emulated = emulator_pio_in_emulated,
4218 .pio_out_emulated = emulator_pio_out_emulated,
4219 .get_cached_descriptor = emulator_get_cached_descriptor,
4220 .set_cached_descriptor = emulator_set_cached_descriptor,
4221 .get_segment_selector = emulator_get_segment_selector,
4222 .set_segment_selector = emulator_set_segment_selector,
4223 .get_cached_segment_base = emulator_get_cached_segment_base,
4224 .get_gdt = emulator_get_gdt,
4225 .get_idt = emulator_get_idt,
4226 .get_cr = emulator_get_cr,
4227 .set_cr = emulator_set_cr,
4228 .cpl = emulator_get_cpl,
4229 .get_dr = emulator_get_dr,
4230 .set_dr = emulator_set_dr,
4231 .set_msr = kvm_set_msr,
4232 .get_msr = kvm_get_msr,
4235 static void cache_all_regs(struct kvm_vcpu *vcpu)
4237 kvm_register_read(vcpu, VCPU_REGS_RAX);
4238 kvm_register_read(vcpu, VCPU_REGS_RSP);
4239 kvm_register_read(vcpu, VCPU_REGS_RIP);
4240 vcpu->arch.regs_dirty = ~0;
4243 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4245 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4247 * an sti; sti; sequence only disable interrupts for the first
4248 * instruction. So, if the last instruction, be it emulated or
4249 * not, left the system with the INT_STI flag enabled, it
4250 * means that the last instruction is an sti. We should not
4251 * leave the flag on in this case. The same goes for mov ss
4253 if (!(int_shadow & mask))
4254 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4257 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4259 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4260 if (ctxt->exception.vector == PF_VECTOR)
4261 kvm_propagate_fault(vcpu, &ctxt->exception);
4262 else if (ctxt->exception.error_code_valid)
4263 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4264 ctxt->exception.error_code);
4266 kvm_queue_exception(vcpu, ctxt->exception.vector);
4269 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4271 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4274 cache_all_regs(vcpu);
4276 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4278 vcpu->arch.emulate_ctxt.vcpu = vcpu;
4279 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
4280 vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
4281 vcpu->arch.emulate_ctxt.mode =
4282 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4283 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
4284 ? X86EMUL_MODE_VM86 : cs_l
4285 ? X86EMUL_MODE_PROT64 : cs_db
4286 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
4287 memset(c, 0, sizeof(struct decode_cache));
4288 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4291 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq)
4293 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4296 init_emulate_ctxt(vcpu);
4298 vcpu->arch.emulate_ctxt.decode.op_bytes = 2;
4299 vcpu->arch.emulate_ctxt.decode.ad_bytes = 2;
4300 vcpu->arch.emulate_ctxt.decode.eip = vcpu->arch.emulate_ctxt.eip;
4301 ret = emulate_int_real(&vcpu->arch.emulate_ctxt, &emulate_ops, irq);
4303 if (ret != X86EMUL_CONTINUE)
4304 return EMULATE_FAIL;
4306 vcpu->arch.emulate_ctxt.eip = c->eip;
4307 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4308 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4309 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4311 if (irq == NMI_VECTOR)
4312 vcpu->arch.nmi_pending = false;
4314 vcpu->arch.interrupt.pending = false;
4316 return EMULATE_DONE;
4318 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4320 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4322 int r = EMULATE_DONE;
4324 ++vcpu->stat.insn_emulation_fail;
4325 trace_kvm_emulate_insn_failed(vcpu);
4326 if (!is_guest_mode(vcpu)) {
4327 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4328 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4329 vcpu->run->internal.ndata = 0;
4332 kvm_queue_exception(vcpu, UD_VECTOR);
4337 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4345 * if emulation was due to access to shadowed page table
4346 * and it failed try to unshadow page and re-entetr the
4347 * guest to let CPU execute the instruction.
4349 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4352 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4354 if (gpa == UNMAPPED_GVA)
4355 return true; /* let cpu generate fault */
4357 if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4363 int emulate_instruction(struct kvm_vcpu *vcpu,
4369 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4371 kvm_clear_exception_queue(vcpu);
4372 vcpu->arch.mmio_fault_cr2 = cr2;
4374 * TODO: fix emulate.c to use guest_read/write_register
4375 * instead of direct ->regs accesses, can save hundred cycles
4376 * on Intel for instructions that don't read/change RSP, for
4379 cache_all_regs(vcpu);
4381 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4382 init_emulate_ctxt(vcpu);
4383 vcpu->arch.emulate_ctxt.interruptibility = 0;
4384 vcpu->arch.emulate_ctxt.have_exception = false;
4385 vcpu->arch.emulate_ctxt.perm_ok = false;
4387 r = x86_decode_insn(&vcpu->arch.emulate_ctxt);
4388 if (r == X86EMUL_PROPAGATE_FAULT)
4391 trace_kvm_emulate_insn_start(vcpu);
4393 /* Only allow emulation of specific instructions on #UD
4394 * (namely VMMCALL, sysenter, sysexit, syscall)*/
4395 if (emulation_type & EMULTYPE_TRAP_UD) {
4397 return EMULATE_FAIL;
4399 case 0x01: /* VMMCALL */
4400 if (c->modrm_mod != 3 || c->modrm_rm != 1)
4401 return EMULATE_FAIL;
4403 case 0x34: /* sysenter */
4404 case 0x35: /* sysexit */
4405 if (c->modrm_mod != 0 || c->modrm_rm != 0)
4406 return EMULATE_FAIL;
4408 case 0x05: /* syscall */
4409 if (c->modrm_mod != 0 || c->modrm_rm != 0)
4410 return EMULATE_FAIL;
4413 return EMULATE_FAIL;
4416 if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
4417 return EMULATE_FAIL;
4420 ++vcpu->stat.insn_emulation;
4422 if (reexecute_instruction(vcpu, cr2))
4423 return EMULATE_DONE;
4424 if (emulation_type & EMULTYPE_SKIP)
4425 return EMULATE_FAIL;
4426 return handle_emulation_failure(vcpu);
4430 if (emulation_type & EMULTYPE_SKIP) {
4431 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
4432 return EMULATE_DONE;
4435 /* this is needed for vmware backdor interface to work since it
4436 changes registers values during IO operation */
4437 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4440 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt);
4442 if (r == EMULATION_FAILED) {
4443 if (reexecute_instruction(vcpu, cr2))
4444 return EMULATE_DONE;
4446 return handle_emulation_failure(vcpu);
4450 if (vcpu->arch.emulate_ctxt.have_exception) {
4451 inject_emulated_exception(vcpu);
4453 } else if (vcpu->arch.pio.count) {
4454 if (!vcpu->arch.pio.in)
4455 vcpu->arch.pio.count = 0;
4456 r = EMULATE_DO_MMIO;
4457 } else if (vcpu->mmio_needed) {
4458 if (vcpu->mmio_is_write)
4459 vcpu->mmio_needed = 0;
4460 r = EMULATE_DO_MMIO;
4461 } else if (r == EMULATION_RESTART)
4466 toggle_interruptibility(vcpu, vcpu->arch.emulate_ctxt.interruptibility);
4467 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4468 kvm_make_request(KVM_REQ_EVENT, vcpu);
4469 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4470 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4474 EXPORT_SYMBOL_GPL(emulate_instruction);
4476 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4478 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4479 int ret = emulator_pio_out_emulated(size, port, &val, 1, vcpu);
4480 /* do not return to emulator after return from userspace */
4481 vcpu->arch.pio.count = 0;
4484 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4486 static void tsc_bad(void *info)
4488 __get_cpu_var(cpu_tsc_khz) = 0;
4491 static void tsc_khz_changed(void *data)
4493 struct cpufreq_freqs *freq = data;
4494 unsigned long khz = 0;
4498 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4499 khz = cpufreq_quick_get(raw_smp_processor_id());
4502 __get_cpu_var(cpu_tsc_khz) = khz;
4505 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4508 struct cpufreq_freqs *freq = data;
4510 struct kvm_vcpu *vcpu;
4511 int i, send_ipi = 0;
4514 * We allow guests to temporarily run on slowing clocks,
4515 * provided we notify them after, or to run on accelerating
4516 * clocks, provided we notify them before. Thus time never
4519 * However, we have a problem. We can't atomically update
4520 * the frequency of a given CPU from this function; it is
4521 * merely a notifier, which can be called from any CPU.
4522 * Changing the TSC frequency at arbitrary points in time
4523 * requires a recomputation of local variables related to
4524 * the TSC for each VCPU. We must flag these local variables
4525 * to be updated and be sure the update takes place with the
4526 * new frequency before any guests proceed.
4528 * Unfortunately, the combination of hotplug CPU and frequency
4529 * change creates an intractable locking scenario; the order
4530 * of when these callouts happen is undefined with respect to
4531 * CPU hotplug, and they can race with each other. As such,
4532 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4533 * undefined; you can actually have a CPU frequency change take
4534 * place in between the computation of X and the setting of the
4535 * variable. To protect against this problem, all updates of
4536 * the per_cpu tsc_khz variable are done in an interrupt
4537 * protected IPI, and all callers wishing to update the value
4538 * must wait for a synchronous IPI to complete (which is trivial
4539 * if the caller is on the CPU already). This establishes the
4540 * necessary total order on variable updates.
4542 * Note that because a guest time update may take place
4543 * anytime after the setting of the VCPU's request bit, the
4544 * correct TSC value must be set before the request. However,
4545 * to ensure the update actually makes it to any guest which
4546 * starts running in hardware virtualization between the set
4547 * and the acquisition of the spinlock, we must also ping the
4548 * CPU after setting the request bit.
4552 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4554 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4557 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4559 spin_lock(&kvm_lock);
4560 list_for_each_entry(kvm, &vm_list, vm_list) {
4561 kvm_for_each_vcpu(i, vcpu, kvm) {
4562 if (vcpu->cpu != freq->cpu)
4564 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4565 if (vcpu->cpu != smp_processor_id())
4569 spin_unlock(&kvm_lock);
4571 if (freq->old < freq->new && send_ipi) {
4573 * We upscale the frequency. Must make the guest
4574 * doesn't see old kvmclock values while running with
4575 * the new frequency, otherwise we risk the guest sees
4576 * time go backwards.
4578 * In case we update the frequency for another cpu
4579 * (which might be in guest context) send an interrupt
4580 * to kick the cpu out of guest context. Next time
4581 * guest context is entered kvmclock will be updated,
4582 * so the guest will not see stale values.
4584 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4589 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4590 .notifier_call = kvmclock_cpufreq_notifier
4593 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4594 unsigned long action, void *hcpu)
4596 unsigned int cpu = (unsigned long)hcpu;
4600 case CPU_DOWN_FAILED:
4601 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4603 case CPU_DOWN_PREPARE:
4604 smp_call_function_single(cpu, tsc_bad, NULL, 1);
4610 static struct notifier_block kvmclock_cpu_notifier_block = {
4611 .notifier_call = kvmclock_cpu_notifier,
4612 .priority = -INT_MAX
4615 static void kvm_timer_init(void)
4619 max_tsc_khz = tsc_khz;
4620 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4621 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4622 #ifdef CONFIG_CPU_FREQ
4623 struct cpufreq_policy policy;
4624 memset(&policy, 0, sizeof(policy));
4626 cpufreq_get_policy(&policy, cpu);
4627 if (policy.cpuinfo.max_freq)
4628 max_tsc_khz = policy.cpuinfo.max_freq;
4631 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4632 CPUFREQ_TRANSITION_NOTIFIER);
4634 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
4635 for_each_online_cpu(cpu)
4636 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4639 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4641 static int kvm_is_in_guest(void)
4643 return percpu_read(current_vcpu) != NULL;
4646 static int kvm_is_user_mode(void)
4650 if (percpu_read(current_vcpu))
4651 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
4653 return user_mode != 0;
4656 static unsigned long kvm_get_guest_ip(void)
4658 unsigned long ip = 0;
4660 if (percpu_read(current_vcpu))
4661 ip = kvm_rip_read(percpu_read(current_vcpu));
4666 static struct perf_guest_info_callbacks kvm_guest_cbs = {
4667 .is_in_guest = kvm_is_in_guest,
4668 .is_user_mode = kvm_is_user_mode,
4669 .get_guest_ip = kvm_get_guest_ip,
4672 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4674 percpu_write(current_vcpu, vcpu);
4676 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4678 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4680 percpu_write(current_vcpu, NULL);
4682 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4684 int kvm_arch_init(void *opaque)
4687 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4690 printk(KERN_ERR "kvm: already loaded the other module\n");
4695 if (!ops->cpu_has_kvm_support()) {
4696 printk(KERN_ERR "kvm: no hardware support\n");
4700 if (ops->disabled_by_bios()) {
4701 printk(KERN_ERR "kvm: disabled by bios\n");
4706 r = kvm_mmu_module_init();
4710 kvm_init_msr_list();
4713 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
4714 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4715 PT_DIRTY_MASK, PT64_NX_MASK, 0);
4719 perf_register_guest_info_callbacks(&kvm_guest_cbs);
4722 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
4730 void kvm_arch_exit(void)
4732 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4734 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4735 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4736 CPUFREQ_TRANSITION_NOTIFIER);
4737 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4739 kvm_mmu_module_exit();
4742 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4744 ++vcpu->stat.halt_exits;
4745 if (irqchip_in_kernel(vcpu->kvm)) {
4746 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
4749 vcpu->run->exit_reason = KVM_EXIT_HLT;
4753 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4755 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
4758 if (is_long_mode(vcpu))
4761 return a0 | ((gpa_t)a1 << 32);
4764 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4766 u64 param, ingpa, outgpa, ret;
4767 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4768 bool fast, longmode;
4772 * hypercall generates UD from non zero cpl and real mode
4775 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
4776 kvm_queue_exception(vcpu, UD_VECTOR);
4780 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4781 longmode = is_long_mode(vcpu) && cs_l == 1;
4784 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4785 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4786 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4787 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4788 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4789 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
4791 #ifdef CONFIG_X86_64
4793 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4794 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4795 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4799 code = param & 0xffff;
4800 fast = (param >> 16) & 0x1;
4801 rep_cnt = (param >> 32) & 0xfff;
4802 rep_idx = (param >> 48) & 0xfff;
4804 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
4807 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
4808 kvm_vcpu_on_spin(vcpu);
4811 res = HV_STATUS_INVALID_HYPERCALL_CODE;
4815 ret = res | (((u64)rep_done & 0xfff) << 32);
4817 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4819 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
4820 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
4826 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
4828 unsigned long nr, a0, a1, a2, a3, ret;
4831 if (kvm_hv_hypercall_enabled(vcpu->kvm))
4832 return kvm_hv_hypercall(vcpu);
4834 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
4835 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
4836 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
4837 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
4838 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
4840 trace_kvm_hypercall(nr, a0, a1, a2, a3);
4842 if (!is_long_mode(vcpu)) {
4850 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
4856 case KVM_HC_VAPIC_POLL_IRQ:
4860 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
4867 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4868 ++vcpu->stat.hypercalls;
4871 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
4873 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
4875 char instruction[3];
4876 unsigned long rip = kvm_rip_read(vcpu);
4879 * Blow out the MMU to ensure that no other VCPU has an active mapping
4880 * to ensure that the updated hypercall appears atomically across all
4883 kvm_mmu_zap_all(vcpu->kvm);
4885 kvm_x86_ops->patch_hypercall(vcpu, instruction);
4887 return emulator_write_emulated(rip, instruction, 3, NULL, vcpu);
4890 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4892 struct desc_ptr dt = { limit, base };
4894 kvm_x86_ops->set_gdt(vcpu, &dt);
4897 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4899 struct desc_ptr dt = { limit, base };
4901 kvm_x86_ops->set_idt(vcpu, &dt);
4904 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
4906 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
4907 int j, nent = vcpu->arch.cpuid_nent;
4909 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
4910 /* when no next entry is found, the current entry[i] is reselected */
4911 for (j = i + 1; ; j = (j + 1) % nent) {
4912 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
4913 if (ej->function == e->function) {
4914 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
4918 return 0; /* silence gcc, even though control never reaches here */
4921 /* find an entry with matching function, matching index (if needed), and that
4922 * should be read next (if it's stateful) */
4923 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
4924 u32 function, u32 index)
4926 if (e->function != function)
4928 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
4930 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
4931 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
4936 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
4937 u32 function, u32 index)
4940 struct kvm_cpuid_entry2 *best = NULL;
4942 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
4943 struct kvm_cpuid_entry2 *e;
4945 e = &vcpu->arch.cpuid_entries[i];
4946 if (is_matching_cpuid_entry(e, function, index)) {
4947 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
4948 move_to_next_stateful_cpuid_entry(vcpu, i);
4953 * Both basic or both extended?
4955 if (((e->function ^ function) & 0x80000000) == 0)
4956 if (!best || e->function > best->function)
4961 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
4963 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
4965 struct kvm_cpuid_entry2 *best;
4967 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
4968 if (!best || best->eax < 0x80000008)
4970 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
4972 return best->eax & 0xff;
4977 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
4979 u32 function, index;
4980 struct kvm_cpuid_entry2 *best;
4982 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
4983 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4984 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
4985 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
4986 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
4987 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
4988 best = kvm_find_cpuid_entry(vcpu, function, index);
4990 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
4991 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
4992 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
4993 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
4995 kvm_x86_ops->skip_emulated_instruction(vcpu);
4996 trace_kvm_cpuid(function,
4997 kvm_register_read(vcpu, VCPU_REGS_RAX),
4998 kvm_register_read(vcpu, VCPU_REGS_RBX),
4999 kvm_register_read(vcpu, VCPU_REGS_RCX),
5000 kvm_register_read(vcpu, VCPU_REGS_RDX));
5002 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
5005 * Check if userspace requested an interrupt window, and that the
5006 * interrupt window is open.
5008 * No need to exit to userspace if we already have an interrupt queued.
5010 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5012 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5013 vcpu->run->request_interrupt_window &&
5014 kvm_arch_interrupt_allowed(vcpu));
5017 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5019 struct kvm_run *kvm_run = vcpu->run;
5021 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5022 kvm_run->cr8 = kvm_get_cr8(vcpu);
5023 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5024 if (irqchip_in_kernel(vcpu->kvm))
5025 kvm_run->ready_for_interrupt_injection = 1;
5027 kvm_run->ready_for_interrupt_injection =
5028 kvm_arch_interrupt_allowed(vcpu) &&
5029 !kvm_cpu_has_interrupt(vcpu) &&
5030 !kvm_event_needs_reinjection(vcpu);
5033 static void vapic_enter(struct kvm_vcpu *vcpu)
5035 struct kvm_lapic *apic = vcpu->arch.apic;
5038 if (!apic || !apic->vapic_addr)
5041 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5043 vcpu->arch.apic->vapic_page = page;
5046 static void vapic_exit(struct kvm_vcpu *vcpu)
5048 struct kvm_lapic *apic = vcpu->arch.apic;
5051 if (!apic || !apic->vapic_addr)
5054 idx = srcu_read_lock(&vcpu->kvm->srcu);
5055 kvm_release_page_dirty(apic->vapic_page);
5056 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5057 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5060 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5064 if (!kvm_x86_ops->update_cr8_intercept)
5067 if (!vcpu->arch.apic)
5070 if (!vcpu->arch.apic->vapic_addr)
5071 max_irr = kvm_lapic_find_highest_irr(vcpu);
5078 tpr = kvm_lapic_get_cr8(vcpu);
5080 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5083 static void inject_pending_event(struct kvm_vcpu *vcpu)
5085 /* try to reinject previous events if any */
5086 if (vcpu->arch.exception.pending) {
5087 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5088 vcpu->arch.exception.has_error_code,
5089 vcpu->arch.exception.error_code);
5090 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5091 vcpu->arch.exception.has_error_code,
5092 vcpu->arch.exception.error_code,
5093 vcpu->arch.exception.reinject);
5097 if (vcpu->arch.nmi_injected) {
5098 kvm_x86_ops->set_nmi(vcpu);
5102 if (vcpu->arch.interrupt.pending) {
5103 kvm_x86_ops->set_irq(vcpu);
5107 /* try to inject new event if pending */
5108 if (vcpu->arch.nmi_pending) {
5109 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5110 vcpu->arch.nmi_pending = false;
5111 vcpu->arch.nmi_injected = true;
5112 kvm_x86_ops->set_nmi(vcpu);
5114 } else if (kvm_cpu_has_interrupt(vcpu)) {
5115 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5116 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5118 kvm_x86_ops->set_irq(vcpu);
5123 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5125 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5126 !vcpu->guest_xcr0_loaded) {
5127 /* kvm_set_xcr() also depends on this */
5128 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5129 vcpu->guest_xcr0_loaded = 1;
5133 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5135 if (vcpu->guest_xcr0_loaded) {
5136 if (vcpu->arch.xcr0 != host_xcr0)
5137 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5138 vcpu->guest_xcr0_loaded = 0;
5142 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5145 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5146 vcpu->run->request_interrupt_window;
5148 if (vcpu->requests) {
5149 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5150 kvm_mmu_unload(vcpu);
5151 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5152 __kvm_migrate_timers(vcpu);
5153 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5154 r = kvm_guest_time_update(vcpu);
5158 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5159 kvm_mmu_sync_roots(vcpu);
5160 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5161 kvm_x86_ops->tlb_flush(vcpu);
5162 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5163 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5167 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5168 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5172 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5173 vcpu->fpu_active = 0;
5174 kvm_x86_ops->fpu_deactivate(vcpu);
5176 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5177 /* Page is swapped out. Do synthetic halt */
5178 vcpu->arch.apf.halted = true;
5184 r = kvm_mmu_reload(vcpu);
5188 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5189 inject_pending_event(vcpu);
5191 /* enable NMI/IRQ window open exits if needed */
5192 if (vcpu->arch.nmi_pending)
5193 kvm_x86_ops->enable_nmi_window(vcpu);
5194 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5195 kvm_x86_ops->enable_irq_window(vcpu);
5197 if (kvm_lapic_enabled(vcpu)) {
5198 update_cr8_intercept(vcpu);
5199 kvm_lapic_sync_to_vapic(vcpu);
5205 kvm_x86_ops->prepare_guest_switch(vcpu);
5206 if (vcpu->fpu_active)
5207 kvm_load_guest_fpu(vcpu);
5208 kvm_load_guest_xcr0(vcpu);
5210 atomic_set(&vcpu->guest_mode, 1);
5213 local_irq_disable();
5215 if (!atomic_read(&vcpu->guest_mode) || vcpu->requests
5216 || need_resched() || signal_pending(current)) {
5217 atomic_set(&vcpu->guest_mode, 0);
5221 kvm_x86_ops->cancel_injection(vcpu);
5226 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5230 if (unlikely(vcpu->arch.switch_db_regs)) {
5232 set_debugreg(vcpu->arch.eff_db[0], 0);
5233 set_debugreg(vcpu->arch.eff_db[1], 1);
5234 set_debugreg(vcpu->arch.eff_db[2], 2);
5235 set_debugreg(vcpu->arch.eff_db[3], 3);
5238 trace_kvm_entry(vcpu->vcpu_id);
5239 kvm_x86_ops->run(vcpu);
5242 * If the guest has used debug registers, at least dr7
5243 * will be disabled while returning to the host.
5244 * If we don't have active breakpoints in the host, we don't
5245 * care about the messed up debug address registers. But if
5246 * we have some of them active, restore the old state.
5248 if (hw_breakpoint_active())
5249 hw_breakpoint_restore();
5251 kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
5253 atomic_set(&vcpu->guest_mode, 0);
5260 * We must have an instruction between local_irq_enable() and
5261 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5262 * the interrupt shadow. The stat.exits increment will do nicely.
5263 * But we need to prevent reordering, hence this barrier():
5271 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5274 * Profile KVM exit RIPs:
5276 if (unlikely(prof_on == KVM_PROFILING)) {
5277 unsigned long rip = kvm_rip_read(vcpu);
5278 profile_hit(KVM_PROFILING, (void *)rip);
5282 kvm_lapic_sync_from_vapic(vcpu);
5284 r = kvm_x86_ops->handle_exit(vcpu);
5290 static int __vcpu_run(struct kvm_vcpu *vcpu)
5293 struct kvm *kvm = vcpu->kvm;
5295 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5296 pr_debug("vcpu %d received sipi with vector # %x\n",
5297 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5298 kvm_lapic_reset(vcpu);
5299 r = kvm_arch_vcpu_reset(vcpu);
5302 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5305 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5310 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5311 !vcpu->arch.apf.halted)
5312 r = vcpu_enter_guest(vcpu);
5314 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5315 kvm_vcpu_block(vcpu);
5316 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5317 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5319 switch(vcpu->arch.mp_state) {
5320 case KVM_MP_STATE_HALTED:
5321 vcpu->arch.mp_state =
5322 KVM_MP_STATE_RUNNABLE;
5323 case KVM_MP_STATE_RUNNABLE:
5324 vcpu->arch.apf.halted = false;
5326 case KVM_MP_STATE_SIPI_RECEIVED:
5337 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5338 if (kvm_cpu_has_pending_timer(vcpu))
5339 kvm_inject_pending_timer_irqs(vcpu);
5341 if (dm_request_for_irq_injection(vcpu)) {
5343 vcpu->run->exit_reason = KVM_EXIT_INTR;
5344 ++vcpu->stat.request_irq_exits;
5347 kvm_check_async_pf_completion(vcpu);
5349 if (signal_pending(current)) {
5351 vcpu->run->exit_reason = KVM_EXIT_INTR;
5352 ++vcpu->stat.signal_exits;
5354 if (need_resched()) {
5355 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5357 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5361 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5368 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5373 if (vcpu->sigset_active)
5374 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5376 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5377 kvm_vcpu_block(vcpu);
5378 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5383 /* re-sync apic's tpr */
5384 if (!irqchip_in_kernel(vcpu->kvm))
5385 kvm_set_cr8(vcpu, kvm_run->cr8);
5387 if (vcpu->arch.pio.count || vcpu->mmio_needed) {
5388 if (vcpu->mmio_needed) {
5389 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
5390 vcpu->mmio_read_completed = 1;
5391 vcpu->mmio_needed = 0;
5393 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5394 r = emulate_instruction(vcpu, 0, 0, EMULTYPE_NO_DECODE);
5395 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5396 if (r != EMULATE_DONE) {
5401 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5402 kvm_register_write(vcpu, VCPU_REGS_RAX,
5403 kvm_run->hypercall.ret);
5405 r = __vcpu_run(vcpu);
5408 post_kvm_run_save(vcpu);
5409 if (vcpu->sigset_active)
5410 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5415 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5417 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5418 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5419 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5420 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5421 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5422 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5423 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5424 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5425 #ifdef CONFIG_X86_64
5426 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5427 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5428 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5429 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5430 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5431 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5432 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5433 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5436 regs->rip = kvm_rip_read(vcpu);
5437 regs->rflags = kvm_get_rflags(vcpu);
5442 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5444 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5445 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5446 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5447 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5448 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5449 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5450 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5451 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5452 #ifdef CONFIG_X86_64
5453 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5454 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5455 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5456 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5457 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5458 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5459 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5460 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5463 kvm_rip_write(vcpu, regs->rip);
5464 kvm_set_rflags(vcpu, regs->rflags);
5466 vcpu->arch.exception.pending = false;
5468 kvm_make_request(KVM_REQ_EVENT, vcpu);
5473 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5475 struct kvm_segment cs;
5477 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
5481 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5483 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5484 struct kvm_sregs *sregs)
5488 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5489 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5490 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5491 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5492 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5493 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5495 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5496 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5498 kvm_x86_ops->get_idt(vcpu, &dt);
5499 sregs->idt.limit = dt.size;
5500 sregs->idt.base = dt.address;
5501 kvm_x86_ops->get_gdt(vcpu, &dt);
5502 sregs->gdt.limit = dt.size;
5503 sregs->gdt.base = dt.address;
5505 sregs->cr0 = kvm_read_cr0(vcpu);
5506 sregs->cr2 = vcpu->arch.cr2;
5507 sregs->cr3 = vcpu->arch.cr3;
5508 sregs->cr4 = kvm_read_cr4(vcpu);
5509 sregs->cr8 = kvm_get_cr8(vcpu);
5510 sregs->efer = vcpu->arch.efer;
5511 sregs->apic_base = kvm_get_apic_base(vcpu);
5513 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
5515 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
5516 set_bit(vcpu->arch.interrupt.nr,
5517 (unsigned long *)sregs->interrupt_bitmap);
5522 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
5523 struct kvm_mp_state *mp_state)
5525 mp_state->mp_state = vcpu->arch.mp_state;
5529 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
5530 struct kvm_mp_state *mp_state)
5532 vcpu->arch.mp_state = mp_state->mp_state;
5533 kvm_make_request(KVM_REQ_EVENT, vcpu);
5537 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
5538 bool has_error_code, u32 error_code)
5540 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5543 init_emulate_ctxt(vcpu);
5545 ret = emulator_task_switch(&vcpu->arch.emulate_ctxt,
5546 tss_selector, reason, has_error_code,
5550 return EMULATE_FAIL;
5552 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5553 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
5554 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
5555 kvm_make_request(KVM_REQ_EVENT, vcpu);
5556 return EMULATE_DONE;
5558 EXPORT_SYMBOL_GPL(kvm_task_switch);
5560 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5561 struct kvm_sregs *sregs)
5563 int mmu_reset_needed = 0;
5564 int pending_vec, max_bits;
5567 dt.size = sregs->idt.limit;
5568 dt.address = sregs->idt.base;
5569 kvm_x86_ops->set_idt(vcpu, &dt);
5570 dt.size = sregs->gdt.limit;
5571 dt.address = sregs->gdt.base;
5572 kvm_x86_ops->set_gdt(vcpu, &dt);
5574 vcpu->arch.cr2 = sregs->cr2;
5575 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
5576 vcpu->arch.cr3 = sregs->cr3;
5578 kvm_set_cr8(vcpu, sregs->cr8);
5580 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5581 kvm_x86_ops->set_efer(vcpu, sregs->efer);
5582 kvm_set_apic_base(vcpu, sregs->apic_base);
5584 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5585 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5586 vcpu->arch.cr0 = sregs->cr0;
5588 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5589 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5590 if (sregs->cr4 & X86_CR4_OSXSAVE)
5592 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5593 load_pdptrs(vcpu, vcpu->arch.walk_mmu, vcpu->arch.cr3);
5594 mmu_reset_needed = 1;
5597 if (mmu_reset_needed)
5598 kvm_mmu_reset_context(vcpu);
5600 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5601 pending_vec = find_first_bit(
5602 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5603 if (pending_vec < max_bits) {
5604 kvm_queue_interrupt(vcpu, pending_vec, false);
5605 pr_debug("Set back pending irq %d\n", pending_vec);
5606 if (irqchip_in_kernel(vcpu->kvm))
5607 kvm_pic_clear_isr_ack(vcpu->kvm);
5610 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5611 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5612 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5613 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5614 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5615 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5617 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5618 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5620 update_cr8_intercept(vcpu);
5622 /* Older userspace won't unhalt the vcpu on reset. */
5623 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5624 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5626 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5628 kvm_make_request(KVM_REQ_EVENT, vcpu);
5633 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5634 struct kvm_guest_debug *dbg)
5636 unsigned long rflags;
5639 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5641 if (vcpu->arch.exception.pending)
5643 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5644 kvm_queue_exception(vcpu, DB_VECTOR);
5646 kvm_queue_exception(vcpu, BP_VECTOR);
5650 * Read rflags as long as potentially injected trace flags are still
5653 rflags = kvm_get_rflags(vcpu);
5655 vcpu->guest_debug = dbg->control;
5656 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5657 vcpu->guest_debug = 0;
5659 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5660 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5661 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5662 vcpu->arch.switch_db_regs =
5663 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5665 for (i = 0; i < KVM_NR_DB_REGS; i++)
5666 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5667 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5670 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5671 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5672 get_segment_base(vcpu, VCPU_SREG_CS);
5675 * Trigger an rflags update that will inject or remove the trace
5678 kvm_set_rflags(vcpu, rflags);
5680 kvm_x86_ops->set_guest_debug(vcpu, dbg);
5690 * Translate a guest virtual address to a guest physical address.
5692 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5693 struct kvm_translation *tr)
5695 unsigned long vaddr = tr->linear_address;
5699 idx = srcu_read_lock(&vcpu->kvm->srcu);
5700 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5701 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5702 tr->physical_address = gpa;
5703 tr->valid = gpa != UNMAPPED_GVA;
5710 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5712 struct i387_fxsave_struct *fxsave =
5713 &vcpu->arch.guest_fpu.state->fxsave;
5715 memcpy(fpu->fpr, fxsave->st_space, 128);
5716 fpu->fcw = fxsave->cwd;
5717 fpu->fsw = fxsave->swd;
5718 fpu->ftwx = fxsave->twd;
5719 fpu->last_opcode = fxsave->fop;
5720 fpu->last_ip = fxsave->rip;
5721 fpu->last_dp = fxsave->rdp;
5722 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5727 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5729 struct i387_fxsave_struct *fxsave =
5730 &vcpu->arch.guest_fpu.state->fxsave;
5732 memcpy(fxsave->st_space, fpu->fpr, 128);
5733 fxsave->cwd = fpu->fcw;
5734 fxsave->swd = fpu->fsw;
5735 fxsave->twd = fpu->ftwx;
5736 fxsave->fop = fpu->last_opcode;
5737 fxsave->rip = fpu->last_ip;
5738 fxsave->rdp = fpu->last_dp;
5739 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5744 int fx_init(struct kvm_vcpu *vcpu)
5748 err = fpu_alloc(&vcpu->arch.guest_fpu);
5752 fpu_finit(&vcpu->arch.guest_fpu);
5755 * Ensure guest xcr0 is valid for loading
5757 vcpu->arch.xcr0 = XSTATE_FP;
5759 vcpu->arch.cr0 |= X86_CR0_ET;
5763 EXPORT_SYMBOL_GPL(fx_init);
5765 static void fx_free(struct kvm_vcpu *vcpu)
5767 fpu_free(&vcpu->arch.guest_fpu);
5770 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5772 if (vcpu->guest_fpu_loaded)
5776 * Restore all possible states in the guest,
5777 * and assume host would use all available bits.
5778 * Guest xcr0 would be loaded later.
5780 kvm_put_guest_xcr0(vcpu);
5781 vcpu->guest_fpu_loaded = 1;
5782 unlazy_fpu(current);
5783 fpu_restore_checking(&vcpu->arch.guest_fpu);
5787 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5789 kvm_put_guest_xcr0(vcpu);
5791 if (!vcpu->guest_fpu_loaded)
5794 vcpu->guest_fpu_loaded = 0;
5795 fpu_save_init(&vcpu->arch.guest_fpu);
5796 ++vcpu->stat.fpu_reload;
5797 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
5801 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5803 if (vcpu->arch.time_page) {
5804 kvm_release_page_dirty(vcpu->arch.time_page);
5805 vcpu->arch.time_page = NULL;
5808 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
5810 kvm_x86_ops->vcpu_free(vcpu);
5813 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5816 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
5817 printk_once(KERN_WARNING
5818 "kvm: SMP vm created on host with unstable TSC; "
5819 "guest TSC will not be reliable\n");
5820 return kvm_x86_ops->vcpu_create(kvm, id);
5823 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5827 vcpu->arch.mtrr_state.have_fixed = 1;
5829 r = kvm_arch_vcpu_reset(vcpu);
5831 r = kvm_mmu_setup(vcpu);
5838 kvm_x86_ops->vcpu_free(vcpu);
5842 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5844 vcpu->arch.apf.msr_val = 0;
5847 kvm_mmu_unload(vcpu);
5851 kvm_x86_ops->vcpu_free(vcpu);
5854 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5856 vcpu->arch.nmi_pending = false;
5857 vcpu->arch.nmi_injected = false;
5859 vcpu->arch.switch_db_regs = 0;
5860 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5861 vcpu->arch.dr6 = DR6_FIXED_1;
5862 vcpu->arch.dr7 = DR7_FIXED_1;
5864 kvm_make_request(KVM_REQ_EVENT, vcpu);
5865 vcpu->arch.apf.msr_val = 0;
5867 kvm_clear_async_pf_completion_queue(vcpu);
5868 kvm_async_pf_hash_reset(vcpu);
5869 vcpu->arch.apf.halted = false;
5871 return kvm_x86_ops->vcpu_reset(vcpu);
5874 int kvm_arch_hardware_enable(void *garbage)
5877 struct kvm_vcpu *vcpu;
5880 kvm_shared_msr_cpu_online();
5881 list_for_each_entry(kvm, &vm_list, vm_list)
5882 kvm_for_each_vcpu(i, vcpu, kvm)
5883 if (vcpu->cpu == smp_processor_id())
5884 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5885 return kvm_x86_ops->hardware_enable(garbage);
5888 void kvm_arch_hardware_disable(void *garbage)
5890 kvm_x86_ops->hardware_disable(garbage);
5891 drop_user_return_notifiers(garbage);
5894 int kvm_arch_hardware_setup(void)
5896 return kvm_x86_ops->hardware_setup();
5899 void kvm_arch_hardware_unsetup(void)
5901 kvm_x86_ops->hardware_unsetup();
5904 void kvm_arch_check_processor_compat(void *rtn)
5906 kvm_x86_ops->check_processor_compatibility(rtn);
5909 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5915 BUG_ON(vcpu->kvm == NULL);
5918 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
5919 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
5920 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5921 vcpu->arch.mmu.translate_gpa = translate_gpa;
5922 vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
5923 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5924 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5926 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5928 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5933 vcpu->arch.pio_data = page_address(page);
5935 if (!kvm->arch.virtual_tsc_khz)
5936 kvm_arch_set_tsc_khz(kvm, max_tsc_khz);
5938 r = kvm_mmu_create(vcpu);
5940 goto fail_free_pio_data;
5942 if (irqchip_in_kernel(kvm)) {
5943 r = kvm_create_lapic(vcpu);
5945 goto fail_mmu_destroy;
5948 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5950 if (!vcpu->arch.mce_banks) {
5952 goto fail_free_lapic;
5954 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5956 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
5957 goto fail_free_mce_banks;
5959 kvm_async_pf_hash_reset(vcpu);
5962 fail_free_mce_banks:
5963 kfree(vcpu->arch.mce_banks);
5965 kvm_free_lapic(vcpu);
5967 kvm_mmu_destroy(vcpu);
5969 free_page((unsigned long)vcpu->arch.pio_data);
5974 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5978 kfree(vcpu->arch.mce_banks);
5979 kvm_free_lapic(vcpu);
5980 idx = srcu_read_lock(&vcpu->kvm->srcu);
5981 kvm_mmu_destroy(vcpu);
5982 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5983 free_page((unsigned long)vcpu->arch.pio_data);
5986 int kvm_arch_init_vm(struct kvm *kvm)
5988 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5989 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5991 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5992 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5994 spin_lock_init(&kvm->arch.tsc_write_lock);
5999 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6002 kvm_mmu_unload(vcpu);
6006 static void kvm_free_vcpus(struct kvm *kvm)
6009 struct kvm_vcpu *vcpu;
6012 * Unpin any mmu pages first.
6014 kvm_for_each_vcpu(i, vcpu, kvm) {
6015 kvm_clear_async_pf_completion_queue(vcpu);
6016 kvm_unload_vcpu_mmu(vcpu);
6018 kvm_for_each_vcpu(i, vcpu, kvm)
6019 kvm_arch_vcpu_free(vcpu);
6021 mutex_lock(&kvm->lock);
6022 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6023 kvm->vcpus[i] = NULL;
6025 atomic_set(&kvm->online_vcpus, 0);
6026 mutex_unlock(&kvm->lock);
6029 void kvm_arch_sync_events(struct kvm *kvm)
6031 kvm_free_all_assigned_devices(kvm);
6035 void kvm_arch_destroy_vm(struct kvm *kvm)
6037 kvm_iommu_unmap_guest(kvm);
6038 kfree(kvm->arch.vpic);
6039 kfree(kvm->arch.vioapic);
6040 kvm_free_vcpus(kvm);
6041 if (kvm->arch.apic_access_page)
6042 put_page(kvm->arch.apic_access_page);
6043 if (kvm->arch.ept_identity_pagetable)
6044 put_page(kvm->arch.ept_identity_pagetable);
6047 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6048 struct kvm_memory_slot *memslot,
6049 struct kvm_memory_slot old,
6050 struct kvm_userspace_memory_region *mem,
6053 int npages = memslot->npages;
6054 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6056 /* Prevent internal slot pages from being moved by fork()/COW. */
6057 if (memslot->id >= KVM_MEMORY_SLOTS)
6058 map_flags = MAP_SHARED | MAP_ANONYMOUS;
6060 /*To keep backward compatibility with older userspace,
6061 *x86 needs to hanlde !user_alloc case.
6064 if (npages && !old.rmap) {
6065 unsigned long userspace_addr;
6067 down_write(¤t->mm->mmap_sem);
6068 userspace_addr = do_mmap(NULL, 0,
6070 PROT_READ | PROT_WRITE,
6073 up_write(¤t->mm->mmap_sem);
6075 if (IS_ERR((void *)userspace_addr))
6076 return PTR_ERR((void *)userspace_addr);
6078 memslot->userspace_addr = userspace_addr;
6086 void kvm_arch_commit_memory_region(struct kvm *kvm,
6087 struct kvm_userspace_memory_region *mem,
6088 struct kvm_memory_slot old,
6092 int npages = mem->memory_size >> PAGE_SHIFT;
6094 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6097 down_write(¤t->mm->mmap_sem);
6098 ret = do_munmap(current->mm, old.userspace_addr,
6099 old.npages * PAGE_SIZE);
6100 up_write(¤t->mm->mmap_sem);
6103 "kvm_vm_ioctl_set_memory_region: "
6104 "failed to munmap memory\n");
6107 spin_lock(&kvm->mmu_lock);
6108 if (!kvm->arch.n_requested_mmu_pages) {
6109 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6110 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6113 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6114 spin_unlock(&kvm->mmu_lock);
6117 void kvm_arch_flush_shadow(struct kvm *kvm)
6119 kvm_mmu_zap_all(kvm);
6120 kvm_reload_remote_mmus(kvm);
6123 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6125 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6126 !vcpu->arch.apf.halted)
6127 || !list_empty_careful(&vcpu->async_pf.done)
6128 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6129 || vcpu->arch.nmi_pending ||
6130 (kvm_arch_interrupt_allowed(vcpu) &&
6131 kvm_cpu_has_interrupt(vcpu));
6134 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6137 int cpu = vcpu->cpu;
6139 if (waitqueue_active(&vcpu->wq)) {
6140 wake_up_interruptible(&vcpu->wq);
6141 ++vcpu->stat.halt_wakeup;
6145 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6146 if (atomic_xchg(&vcpu->guest_mode, 0))
6147 smp_send_reschedule(cpu);
6151 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6153 return kvm_x86_ops->interrupt_allowed(vcpu);
6156 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6158 unsigned long current_rip = kvm_rip_read(vcpu) +
6159 get_segment_base(vcpu, VCPU_SREG_CS);
6161 return current_rip == linear_rip;
6163 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6165 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6167 unsigned long rflags;
6169 rflags = kvm_x86_ops->get_rflags(vcpu);
6170 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6171 rflags &= ~X86_EFLAGS_TF;
6174 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6176 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6178 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6179 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6180 rflags |= X86_EFLAGS_TF;
6181 kvm_x86_ops->set_rflags(vcpu, rflags);
6182 kvm_make_request(KVM_REQ_EVENT, vcpu);
6184 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6186 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6190 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6191 is_error_page(work->page))
6194 r = kvm_mmu_reload(vcpu);
6198 if (!vcpu->arch.mmu.direct_map &&
6199 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6202 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6205 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6207 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6210 static inline u32 kvm_async_pf_next_probe(u32 key)
6212 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6215 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6217 u32 key = kvm_async_pf_hash_fn(gfn);
6219 while (vcpu->arch.apf.gfns[key] != ~0)
6220 key = kvm_async_pf_next_probe(key);
6222 vcpu->arch.apf.gfns[key] = gfn;
6225 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6228 u32 key = kvm_async_pf_hash_fn(gfn);
6230 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6231 (vcpu->arch.apf.gfns[key] != gfn &&
6232 vcpu->arch.apf.gfns[key] != ~0); i++)
6233 key = kvm_async_pf_next_probe(key);
6238 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6240 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6243 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6247 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6249 vcpu->arch.apf.gfns[i] = ~0;
6251 j = kvm_async_pf_next_probe(j);
6252 if (vcpu->arch.apf.gfns[j] == ~0)
6254 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6256 * k lies cyclically in ]i,j]
6258 * |....j i.k.| or |.k..j i...|
6260 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6261 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6266 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6269 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6273 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6274 struct kvm_async_pf *work)
6276 struct x86_exception fault;
6278 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6279 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6281 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6282 (vcpu->arch.apf.send_user_only &&
6283 kvm_x86_ops->get_cpl(vcpu) == 0))
6284 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6285 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6286 fault.vector = PF_VECTOR;
6287 fault.error_code_valid = true;
6288 fault.error_code = 0;
6289 fault.nested_page_fault = false;
6290 fault.address = work->arch.token;
6291 kvm_inject_page_fault(vcpu, &fault);
6295 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6296 struct kvm_async_pf *work)
6298 struct x86_exception fault;
6300 trace_kvm_async_pf_ready(work->arch.token, work->gva);
6301 if (is_error_page(work->page))
6302 work->arch.token = ~0; /* broadcast wakeup */
6304 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6306 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6307 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6308 fault.vector = PF_VECTOR;
6309 fault.error_code_valid = true;
6310 fault.error_code = 0;
6311 fault.nested_page_fault = false;
6312 fault.address = work->arch.token;
6313 kvm_inject_page_fault(vcpu, &fault);
6315 vcpu->arch.apf.halted = false;
6318 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6320 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6323 return !kvm_event_needs_reinjection(vcpu) &&
6324 kvm_x86_ops->interrupt_allowed(vcpu);
6327 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6328 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6329 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6330 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6331 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6332 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6333 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6334 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6335 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6336 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6337 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6338 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);