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_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
340 kvm_inject_gp(vcpu, 0);
342 kvm_x86_ops->skip_emulated_instruction(vcpu);
344 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
346 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
348 ++vcpu->stat.pf_guest;
349 vcpu->arch.cr2 = fault->address;
350 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
353 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
355 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
356 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
358 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
361 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
363 kvm_make_request(KVM_REQ_EVENT, vcpu);
364 vcpu->arch.nmi_pending = 1;
366 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
368 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
370 kvm_multiple_exception(vcpu, nr, true, error_code, false);
372 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
374 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
376 kvm_multiple_exception(vcpu, nr, true, error_code, true);
378 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
381 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
382 * a #GP and return false.
384 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
386 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
388 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
391 EXPORT_SYMBOL_GPL(kvm_require_cpl);
394 * This function will be used to read from the physical memory of the currently
395 * running guest. The difference to kvm_read_guest_page is that this function
396 * can read from guest physical or from the guest's guest physical memory.
398 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
399 gfn_t ngfn, void *data, int offset, int len,
405 ngpa = gfn_to_gpa(ngfn);
406 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
407 if (real_gfn == UNMAPPED_GVA)
410 real_gfn = gpa_to_gfn(real_gfn);
412 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
414 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
416 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
417 void *data, int offset, int len, u32 access)
419 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
420 data, offset, len, access);
424 * Load the pae pdptrs. Return true is they are all valid.
426 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
428 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
429 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
432 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
434 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
435 offset * sizeof(u64), sizeof(pdpte),
436 PFERR_USER_MASK|PFERR_WRITE_MASK);
441 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
442 if (is_present_gpte(pdpte[i]) &&
443 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
450 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
451 __set_bit(VCPU_EXREG_PDPTR,
452 (unsigned long *)&vcpu->arch.regs_avail);
453 __set_bit(VCPU_EXREG_PDPTR,
454 (unsigned long *)&vcpu->arch.regs_dirty);
459 EXPORT_SYMBOL_GPL(load_pdptrs);
461 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
463 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
469 if (is_long_mode(vcpu) || !is_pae(vcpu))
472 if (!test_bit(VCPU_EXREG_PDPTR,
473 (unsigned long *)&vcpu->arch.regs_avail))
476 gfn = (vcpu->arch.cr3 & ~31u) >> PAGE_SHIFT;
477 offset = (vcpu->arch.cr3 & ~31u) & (PAGE_SIZE - 1);
478 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
479 PFERR_USER_MASK | PFERR_WRITE_MASK);
482 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
488 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
490 unsigned long old_cr0 = kvm_read_cr0(vcpu);
491 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
492 X86_CR0_CD | X86_CR0_NW;
497 if (cr0 & 0xffffffff00000000UL)
501 cr0 &= ~CR0_RESERVED_BITS;
503 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
506 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
509 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
511 if ((vcpu->arch.efer & EFER_LME)) {
516 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
521 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
526 kvm_x86_ops->set_cr0(vcpu, cr0);
528 if ((cr0 ^ old_cr0) & X86_CR0_PG)
529 kvm_clear_async_pf_completion_queue(vcpu);
531 if ((cr0 ^ old_cr0) & update_bits)
532 kvm_mmu_reset_context(vcpu);
535 EXPORT_SYMBOL_GPL(kvm_set_cr0);
537 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
539 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
541 EXPORT_SYMBOL_GPL(kvm_lmsw);
543 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
547 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
548 if (index != XCR_XFEATURE_ENABLED_MASK)
551 if (kvm_x86_ops->get_cpl(vcpu) != 0)
553 if (!(xcr0 & XSTATE_FP))
555 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
557 if (xcr0 & ~host_xcr0)
559 vcpu->arch.xcr0 = xcr0;
560 vcpu->guest_xcr0_loaded = 0;
564 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
566 if (__kvm_set_xcr(vcpu, index, xcr)) {
567 kvm_inject_gp(vcpu, 0);
572 EXPORT_SYMBOL_GPL(kvm_set_xcr);
574 static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
576 struct kvm_cpuid_entry2 *best;
578 best = kvm_find_cpuid_entry(vcpu, 1, 0);
579 return best && (best->ecx & bit(X86_FEATURE_XSAVE));
582 static void update_cpuid(struct kvm_vcpu *vcpu)
584 struct kvm_cpuid_entry2 *best;
586 best = kvm_find_cpuid_entry(vcpu, 1, 0);
590 /* Update OSXSAVE bit */
591 if (cpu_has_xsave && best->function == 0x1) {
592 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
593 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
594 best->ecx |= bit(X86_FEATURE_OSXSAVE);
598 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
600 unsigned long old_cr4 = kvm_read_cr4(vcpu);
601 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
603 if (cr4 & CR4_RESERVED_BITS)
606 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
609 if (is_long_mode(vcpu)) {
610 if (!(cr4 & X86_CR4_PAE))
612 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
613 && ((cr4 ^ old_cr4) & pdptr_bits)
614 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, vcpu->arch.cr3))
617 if (cr4 & X86_CR4_VMXE)
620 kvm_x86_ops->set_cr4(vcpu, cr4);
622 if ((cr4 ^ old_cr4) & pdptr_bits)
623 kvm_mmu_reset_context(vcpu);
625 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
630 EXPORT_SYMBOL_GPL(kvm_set_cr4);
632 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
634 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
635 kvm_mmu_sync_roots(vcpu);
636 kvm_mmu_flush_tlb(vcpu);
640 if (is_long_mode(vcpu)) {
641 if (cr3 & CR3_L_MODE_RESERVED_BITS)
645 if (cr3 & CR3_PAE_RESERVED_BITS)
647 if (is_paging(vcpu) &&
648 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
652 * We don't check reserved bits in nonpae mode, because
653 * this isn't enforced, and VMware depends on this.
658 * Does the new cr3 value map to physical memory? (Note, we
659 * catch an invalid cr3 even in real-mode, because it would
660 * cause trouble later on when we turn on paging anyway.)
662 * A real CPU would silently accept an invalid cr3 and would
663 * attempt to use it - with largely undefined (and often hard
664 * to debug) behavior on the guest side.
666 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
668 vcpu->arch.cr3 = cr3;
669 vcpu->arch.mmu.new_cr3(vcpu);
672 EXPORT_SYMBOL_GPL(kvm_set_cr3);
674 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
676 if (cr8 & CR8_RESERVED_BITS)
678 if (irqchip_in_kernel(vcpu->kvm))
679 kvm_lapic_set_tpr(vcpu, cr8);
681 vcpu->arch.cr8 = cr8;
684 EXPORT_SYMBOL_GPL(kvm_set_cr8);
686 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
688 if (irqchip_in_kernel(vcpu->kvm))
689 return kvm_lapic_get_cr8(vcpu);
691 return vcpu->arch.cr8;
693 EXPORT_SYMBOL_GPL(kvm_get_cr8);
695 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
699 vcpu->arch.db[dr] = val;
700 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
701 vcpu->arch.eff_db[dr] = val;
704 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
708 if (val & 0xffffffff00000000ULL)
710 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
713 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
717 if (val & 0xffffffff00000000ULL)
719 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
720 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
721 kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
722 vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
730 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
734 res = __kvm_set_dr(vcpu, dr, val);
736 kvm_queue_exception(vcpu, UD_VECTOR);
738 kvm_inject_gp(vcpu, 0);
742 EXPORT_SYMBOL_GPL(kvm_set_dr);
744 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
748 *val = vcpu->arch.db[dr];
751 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
755 *val = vcpu->arch.dr6;
758 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
762 *val = vcpu->arch.dr7;
769 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
771 if (_kvm_get_dr(vcpu, dr, val)) {
772 kvm_queue_exception(vcpu, UD_VECTOR);
777 EXPORT_SYMBOL_GPL(kvm_get_dr);
780 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
781 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
783 * This list is modified at module load time to reflect the
784 * capabilities of the host cpu. This capabilities test skips MSRs that are
785 * kvm-specific. Those are put in the beginning of the list.
788 #define KVM_SAVE_MSRS_BEGIN 8
789 static u32 msrs_to_save[] = {
790 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
791 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
792 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
793 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN,
794 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
797 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
799 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
802 static unsigned num_msrs_to_save;
804 static u32 emulated_msrs[] = {
805 MSR_IA32_MISC_ENABLE,
810 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
812 u64 old_efer = vcpu->arch.efer;
814 if (efer & efer_reserved_bits)
818 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
821 if (efer & EFER_FFXSR) {
822 struct kvm_cpuid_entry2 *feat;
824 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
825 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
829 if (efer & EFER_SVME) {
830 struct kvm_cpuid_entry2 *feat;
832 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
833 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
838 efer |= vcpu->arch.efer & EFER_LMA;
840 kvm_x86_ops->set_efer(vcpu, efer);
842 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
844 /* Update reserved bits */
845 if ((efer ^ old_efer) & EFER_NX)
846 kvm_mmu_reset_context(vcpu);
851 void kvm_enable_efer_bits(u64 mask)
853 efer_reserved_bits &= ~mask;
855 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
859 * Writes msr value into into the appropriate "register".
860 * Returns 0 on success, non-0 otherwise.
861 * Assumes vcpu_load() was already called.
863 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
865 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
869 * Adapt set_msr() to msr_io()'s calling convention
871 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
873 return kvm_set_msr(vcpu, index, *data);
876 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
880 struct pvclock_wall_clock wc;
881 struct timespec boot;
886 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
891 ++version; /* first time write, random junk */
895 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
898 * The guest calculates current wall clock time by adding
899 * system time (updated by kvm_guest_time_update below) to the
900 * wall clock specified here. guest system time equals host
901 * system time for us, thus we must fill in host boot time here.
905 wc.sec = boot.tv_sec;
906 wc.nsec = boot.tv_nsec;
907 wc.version = version;
909 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
912 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
915 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
917 uint32_t quotient, remainder;
919 /* Don't try to replace with do_div(), this one calculates
920 * "(dividend << 32) / divisor" */
922 : "=a" (quotient), "=d" (remainder)
923 : "0" (0), "1" (dividend), "r" (divisor) );
927 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
928 s8 *pshift, u32 *pmultiplier)
935 tps64 = base_khz * 1000LL;
936 scaled64 = scaled_khz * 1000LL;
937 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
942 tps32 = (uint32_t)tps64;
943 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
944 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
952 *pmultiplier = div_frac(scaled64, tps32);
954 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
955 __func__, base_khz, scaled_khz, shift, *pmultiplier);
958 static inline u64 get_kernel_ns(void)
962 WARN_ON(preemptible());
964 monotonic_to_bootbased(&ts);
965 return timespec_to_ns(&ts);
968 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
969 unsigned long max_tsc_khz;
971 static inline int kvm_tsc_changes_freq(void)
974 int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
975 cpufreq_quick_get(cpu) != 0;
980 static inline u64 nsec_to_cycles(u64 nsec)
984 WARN_ON(preemptible());
985 if (kvm_tsc_changes_freq())
986 printk_once(KERN_WARNING
987 "kvm: unreliable cycle conversion on adjustable rate TSC\n");
988 ret = nsec * __get_cpu_var(cpu_tsc_khz);
989 do_div(ret, USEC_PER_SEC);
993 static void kvm_arch_set_tsc_khz(struct kvm *kvm, u32 this_tsc_khz)
995 /* Compute a scale to convert nanoseconds in TSC cycles */
996 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
997 &kvm->arch.virtual_tsc_shift,
998 &kvm->arch.virtual_tsc_mult);
999 kvm->arch.virtual_tsc_khz = this_tsc_khz;
1002 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1004 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1005 vcpu->kvm->arch.virtual_tsc_mult,
1006 vcpu->kvm->arch.virtual_tsc_shift);
1007 tsc += vcpu->arch.last_tsc_write;
1011 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1013 struct kvm *kvm = vcpu->kvm;
1014 u64 offset, ns, elapsed;
1015 unsigned long flags;
1018 spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1019 offset = data - native_read_tsc();
1020 ns = get_kernel_ns();
1021 elapsed = ns - kvm->arch.last_tsc_nsec;
1022 sdiff = data - kvm->arch.last_tsc_write;
1027 * Special case: close write to TSC within 5 seconds of
1028 * another CPU is interpreted as an attempt to synchronize
1029 * The 5 seconds is to accomodate host load / swapping as
1030 * well as any reset of TSC during the boot process.
1032 * In that case, for a reliable TSC, we can match TSC offsets,
1033 * or make a best guest using elapsed value.
1035 if (sdiff < nsec_to_cycles(5ULL * NSEC_PER_SEC) &&
1036 elapsed < 5ULL * NSEC_PER_SEC) {
1037 if (!check_tsc_unstable()) {
1038 offset = kvm->arch.last_tsc_offset;
1039 pr_debug("kvm: matched tsc offset for %llu\n", data);
1041 u64 delta = nsec_to_cycles(elapsed);
1043 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1045 ns = kvm->arch.last_tsc_nsec;
1047 kvm->arch.last_tsc_nsec = ns;
1048 kvm->arch.last_tsc_write = data;
1049 kvm->arch.last_tsc_offset = offset;
1050 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1051 spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1053 /* Reset of TSC must disable overshoot protection below */
1054 vcpu->arch.hv_clock.tsc_timestamp = 0;
1055 vcpu->arch.last_tsc_write = data;
1056 vcpu->arch.last_tsc_nsec = ns;
1058 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1060 static int kvm_guest_time_update(struct kvm_vcpu *v)
1062 unsigned long flags;
1063 struct kvm_vcpu_arch *vcpu = &v->arch;
1065 unsigned long this_tsc_khz;
1066 s64 kernel_ns, max_kernel_ns;
1069 /* Keep irq disabled to prevent changes to the clock */
1070 local_irq_save(flags);
1071 kvm_get_msr(v, MSR_IA32_TSC, &tsc_timestamp);
1072 kernel_ns = get_kernel_ns();
1073 this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
1075 if (unlikely(this_tsc_khz == 0)) {
1076 local_irq_restore(flags);
1077 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1082 * We may have to catch up the TSC to match elapsed wall clock
1083 * time for two reasons, even if kvmclock is used.
1084 * 1) CPU could have been running below the maximum TSC rate
1085 * 2) Broken TSC compensation resets the base at each VCPU
1086 * entry to avoid unknown leaps of TSC even when running
1087 * again on the same CPU. This may cause apparent elapsed
1088 * time to disappear, and the guest to stand still or run
1091 if (vcpu->tsc_catchup) {
1092 u64 tsc = compute_guest_tsc(v, kernel_ns);
1093 if (tsc > tsc_timestamp) {
1094 kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1095 tsc_timestamp = tsc;
1099 local_irq_restore(flags);
1101 if (!vcpu->time_page)
1105 * Time as measured by the TSC may go backwards when resetting the base
1106 * tsc_timestamp. The reason for this is that the TSC resolution is
1107 * higher than the resolution of the other clock scales. Thus, many
1108 * possible measurments of the TSC correspond to one measurement of any
1109 * other clock, and so a spread of values is possible. This is not a
1110 * problem for the computation of the nanosecond clock; with TSC rates
1111 * around 1GHZ, there can only be a few cycles which correspond to one
1112 * nanosecond value, and any path through this code will inevitably
1113 * take longer than that. However, with the kernel_ns value itself,
1114 * the precision may be much lower, down to HZ granularity. If the
1115 * first sampling of TSC against kernel_ns ends in the low part of the
1116 * range, and the second in the high end of the range, we can get:
1118 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1120 * As the sampling errors potentially range in the thousands of cycles,
1121 * it is possible such a time value has already been observed by the
1122 * guest. To protect against this, we must compute the system time as
1123 * observed by the guest and ensure the new system time is greater.
1126 if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1127 max_kernel_ns = vcpu->last_guest_tsc -
1128 vcpu->hv_clock.tsc_timestamp;
1129 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1130 vcpu->hv_clock.tsc_to_system_mul,
1131 vcpu->hv_clock.tsc_shift);
1132 max_kernel_ns += vcpu->last_kernel_ns;
1135 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1136 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1137 &vcpu->hv_clock.tsc_shift,
1138 &vcpu->hv_clock.tsc_to_system_mul);
1139 vcpu->hw_tsc_khz = this_tsc_khz;
1142 if (max_kernel_ns > kernel_ns)
1143 kernel_ns = max_kernel_ns;
1145 /* With all the info we got, fill in the values */
1146 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1147 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1148 vcpu->last_kernel_ns = kernel_ns;
1149 vcpu->last_guest_tsc = tsc_timestamp;
1150 vcpu->hv_clock.flags = 0;
1153 * The interface expects us to write an even number signaling that the
1154 * update is finished. Since the guest won't see the intermediate
1155 * state, we just increase by 2 at the end.
1157 vcpu->hv_clock.version += 2;
1159 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1161 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1162 sizeof(vcpu->hv_clock));
1164 kunmap_atomic(shared_kaddr, KM_USER0);
1166 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1170 static bool msr_mtrr_valid(unsigned msr)
1173 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1174 case MSR_MTRRfix64K_00000:
1175 case MSR_MTRRfix16K_80000:
1176 case MSR_MTRRfix16K_A0000:
1177 case MSR_MTRRfix4K_C0000:
1178 case MSR_MTRRfix4K_C8000:
1179 case MSR_MTRRfix4K_D0000:
1180 case MSR_MTRRfix4K_D8000:
1181 case MSR_MTRRfix4K_E0000:
1182 case MSR_MTRRfix4K_E8000:
1183 case MSR_MTRRfix4K_F0000:
1184 case MSR_MTRRfix4K_F8000:
1185 case MSR_MTRRdefType:
1186 case MSR_IA32_CR_PAT:
1194 static bool valid_pat_type(unsigned t)
1196 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1199 static bool valid_mtrr_type(unsigned t)
1201 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1204 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1208 if (!msr_mtrr_valid(msr))
1211 if (msr == MSR_IA32_CR_PAT) {
1212 for (i = 0; i < 8; i++)
1213 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1216 } else if (msr == MSR_MTRRdefType) {
1219 return valid_mtrr_type(data & 0xff);
1220 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1221 for (i = 0; i < 8 ; i++)
1222 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1227 /* variable MTRRs */
1228 return valid_mtrr_type(data & 0xff);
1231 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1233 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1235 if (!mtrr_valid(vcpu, msr, data))
1238 if (msr == MSR_MTRRdefType) {
1239 vcpu->arch.mtrr_state.def_type = data;
1240 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1241 } else if (msr == MSR_MTRRfix64K_00000)
1243 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1244 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1245 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1246 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1247 else if (msr == MSR_IA32_CR_PAT)
1248 vcpu->arch.pat = data;
1249 else { /* Variable MTRRs */
1250 int idx, is_mtrr_mask;
1253 idx = (msr - 0x200) / 2;
1254 is_mtrr_mask = msr - 0x200 - 2 * idx;
1257 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1260 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1264 kvm_mmu_reset_context(vcpu);
1268 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1270 u64 mcg_cap = vcpu->arch.mcg_cap;
1271 unsigned bank_num = mcg_cap & 0xff;
1274 case MSR_IA32_MCG_STATUS:
1275 vcpu->arch.mcg_status = data;
1277 case MSR_IA32_MCG_CTL:
1278 if (!(mcg_cap & MCG_CTL_P))
1280 if (data != 0 && data != ~(u64)0)
1282 vcpu->arch.mcg_ctl = data;
1285 if (msr >= MSR_IA32_MC0_CTL &&
1286 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1287 u32 offset = msr - MSR_IA32_MC0_CTL;
1288 /* only 0 or all 1s can be written to IA32_MCi_CTL
1289 * some Linux kernels though clear bit 10 in bank 4 to
1290 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1291 * this to avoid an uncatched #GP in the guest
1293 if ((offset & 0x3) == 0 &&
1294 data != 0 && (data | (1 << 10)) != ~(u64)0)
1296 vcpu->arch.mce_banks[offset] = data;
1304 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1306 struct kvm *kvm = vcpu->kvm;
1307 int lm = is_long_mode(vcpu);
1308 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1309 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1310 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1311 : kvm->arch.xen_hvm_config.blob_size_32;
1312 u32 page_num = data & ~PAGE_MASK;
1313 u64 page_addr = data & PAGE_MASK;
1318 if (page_num >= blob_size)
1321 page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1325 if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1327 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1336 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1338 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1341 static bool kvm_hv_msr_partition_wide(u32 msr)
1345 case HV_X64_MSR_GUEST_OS_ID:
1346 case HV_X64_MSR_HYPERCALL:
1354 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1356 struct kvm *kvm = vcpu->kvm;
1359 case HV_X64_MSR_GUEST_OS_ID:
1360 kvm->arch.hv_guest_os_id = data;
1361 /* setting guest os id to zero disables hypercall page */
1362 if (!kvm->arch.hv_guest_os_id)
1363 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1365 case HV_X64_MSR_HYPERCALL: {
1370 /* if guest os id is not set hypercall should remain disabled */
1371 if (!kvm->arch.hv_guest_os_id)
1373 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1374 kvm->arch.hv_hypercall = data;
1377 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1378 addr = gfn_to_hva(kvm, gfn);
1379 if (kvm_is_error_hva(addr))
1381 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1382 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1383 if (copy_to_user((void __user *)addr, instructions, 4))
1385 kvm->arch.hv_hypercall = data;
1389 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1390 "data 0x%llx\n", msr, data);
1396 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1399 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1402 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1403 vcpu->arch.hv_vapic = data;
1406 addr = gfn_to_hva(vcpu->kvm, data >>
1407 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1408 if (kvm_is_error_hva(addr))
1410 if (clear_user((void __user *)addr, PAGE_SIZE))
1412 vcpu->arch.hv_vapic = data;
1415 case HV_X64_MSR_EOI:
1416 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1417 case HV_X64_MSR_ICR:
1418 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1419 case HV_X64_MSR_TPR:
1420 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1422 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1423 "data 0x%llx\n", msr, data);
1430 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1432 gpa_t gpa = data & ~0x3f;
1434 /* Bits 2:5 are resrved, Should be zero */
1438 vcpu->arch.apf.msr_val = data;
1440 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1441 kvm_clear_async_pf_completion_queue(vcpu);
1442 kvm_async_pf_hash_reset(vcpu);
1446 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1449 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1450 kvm_async_pf_wakeup_all(vcpu);
1454 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1458 return set_efer(vcpu, data);
1460 data &= ~(u64)0x40; /* ignore flush filter disable */
1461 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1463 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1468 case MSR_FAM10H_MMIO_CONF_BASE:
1470 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1475 case MSR_AMD64_NB_CFG:
1477 case MSR_IA32_DEBUGCTLMSR:
1479 /* We support the non-activated case already */
1481 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1482 /* Values other than LBR and BTF are vendor-specific,
1483 thus reserved and should throw a #GP */
1486 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1489 case MSR_IA32_UCODE_REV:
1490 case MSR_IA32_UCODE_WRITE:
1491 case MSR_VM_HSAVE_PA:
1492 case MSR_AMD64_PATCH_LOADER:
1494 case 0x200 ... 0x2ff:
1495 return set_msr_mtrr(vcpu, msr, data);
1496 case MSR_IA32_APICBASE:
1497 kvm_set_apic_base(vcpu, data);
1499 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1500 return kvm_x2apic_msr_write(vcpu, msr, data);
1501 case MSR_IA32_MISC_ENABLE:
1502 vcpu->arch.ia32_misc_enable_msr = data;
1504 case MSR_KVM_WALL_CLOCK_NEW:
1505 case MSR_KVM_WALL_CLOCK:
1506 vcpu->kvm->arch.wall_clock = data;
1507 kvm_write_wall_clock(vcpu->kvm, data);
1509 case MSR_KVM_SYSTEM_TIME_NEW:
1510 case MSR_KVM_SYSTEM_TIME: {
1511 if (vcpu->arch.time_page) {
1512 kvm_release_page_dirty(vcpu->arch.time_page);
1513 vcpu->arch.time_page = NULL;
1516 vcpu->arch.time = data;
1517 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1519 /* we verify if the enable bit is set... */
1523 /* ...but clean it before doing the actual write */
1524 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1526 vcpu->arch.time_page =
1527 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1529 if (is_error_page(vcpu->arch.time_page)) {
1530 kvm_release_page_clean(vcpu->arch.time_page);
1531 vcpu->arch.time_page = NULL;
1535 case MSR_KVM_ASYNC_PF_EN:
1536 if (kvm_pv_enable_async_pf(vcpu, data))
1539 case MSR_IA32_MCG_CTL:
1540 case MSR_IA32_MCG_STATUS:
1541 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1542 return set_msr_mce(vcpu, msr, data);
1544 /* Performance counters are not protected by a CPUID bit,
1545 * so we should check all of them in the generic path for the sake of
1546 * cross vendor migration.
1547 * Writing a zero into the event select MSRs disables them,
1548 * which we perfectly emulate ;-). Any other value should be at least
1549 * reported, some guests depend on them.
1551 case MSR_P6_EVNTSEL0:
1552 case MSR_P6_EVNTSEL1:
1553 case MSR_K7_EVNTSEL0:
1554 case MSR_K7_EVNTSEL1:
1555 case MSR_K7_EVNTSEL2:
1556 case MSR_K7_EVNTSEL3:
1558 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1559 "0x%x data 0x%llx\n", msr, data);
1561 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1562 * so we ignore writes to make it happy.
1564 case MSR_P6_PERFCTR0:
1565 case MSR_P6_PERFCTR1:
1566 case MSR_K7_PERFCTR0:
1567 case MSR_K7_PERFCTR1:
1568 case MSR_K7_PERFCTR2:
1569 case MSR_K7_PERFCTR3:
1570 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1571 "0x%x data 0x%llx\n", msr, data);
1573 case MSR_K7_CLK_CTL:
1575 * Ignore all writes to this no longer documented MSR.
1576 * Writes are only relevant for old K7 processors,
1577 * all pre-dating SVM, but a recommended workaround from
1578 * AMD for these chips. It is possible to speicify the
1579 * affected processor models on the command line, hence
1580 * the need to ignore the workaround.
1583 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1584 if (kvm_hv_msr_partition_wide(msr)) {
1586 mutex_lock(&vcpu->kvm->lock);
1587 r = set_msr_hyperv_pw(vcpu, msr, data);
1588 mutex_unlock(&vcpu->kvm->lock);
1591 return set_msr_hyperv(vcpu, msr, data);
1594 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1595 return xen_hvm_config(vcpu, data);
1597 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1601 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1608 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1612 * Reads an msr value (of 'msr_index') into 'pdata'.
1613 * Returns 0 on success, non-0 otherwise.
1614 * Assumes vcpu_load() was already called.
1616 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1618 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1621 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1623 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1625 if (!msr_mtrr_valid(msr))
1628 if (msr == MSR_MTRRdefType)
1629 *pdata = vcpu->arch.mtrr_state.def_type +
1630 (vcpu->arch.mtrr_state.enabled << 10);
1631 else if (msr == MSR_MTRRfix64K_00000)
1633 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1634 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1635 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1636 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1637 else if (msr == MSR_IA32_CR_PAT)
1638 *pdata = vcpu->arch.pat;
1639 else { /* Variable MTRRs */
1640 int idx, is_mtrr_mask;
1643 idx = (msr - 0x200) / 2;
1644 is_mtrr_mask = msr - 0x200 - 2 * idx;
1647 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1650 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1657 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1660 u64 mcg_cap = vcpu->arch.mcg_cap;
1661 unsigned bank_num = mcg_cap & 0xff;
1664 case MSR_IA32_P5_MC_ADDR:
1665 case MSR_IA32_P5_MC_TYPE:
1668 case MSR_IA32_MCG_CAP:
1669 data = vcpu->arch.mcg_cap;
1671 case MSR_IA32_MCG_CTL:
1672 if (!(mcg_cap & MCG_CTL_P))
1674 data = vcpu->arch.mcg_ctl;
1676 case MSR_IA32_MCG_STATUS:
1677 data = vcpu->arch.mcg_status;
1680 if (msr >= MSR_IA32_MC0_CTL &&
1681 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1682 u32 offset = msr - MSR_IA32_MC0_CTL;
1683 data = vcpu->arch.mce_banks[offset];
1692 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1695 struct kvm *kvm = vcpu->kvm;
1698 case HV_X64_MSR_GUEST_OS_ID:
1699 data = kvm->arch.hv_guest_os_id;
1701 case HV_X64_MSR_HYPERCALL:
1702 data = kvm->arch.hv_hypercall;
1705 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1713 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1718 case HV_X64_MSR_VP_INDEX: {
1721 kvm_for_each_vcpu(r, v, vcpu->kvm)
1726 case HV_X64_MSR_EOI:
1727 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1728 case HV_X64_MSR_ICR:
1729 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1730 case HV_X64_MSR_TPR:
1731 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1733 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1740 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1745 case MSR_IA32_PLATFORM_ID:
1746 case MSR_IA32_UCODE_REV:
1747 case MSR_IA32_EBL_CR_POWERON:
1748 case MSR_IA32_DEBUGCTLMSR:
1749 case MSR_IA32_LASTBRANCHFROMIP:
1750 case MSR_IA32_LASTBRANCHTOIP:
1751 case MSR_IA32_LASTINTFROMIP:
1752 case MSR_IA32_LASTINTTOIP:
1755 case MSR_VM_HSAVE_PA:
1756 case MSR_P6_PERFCTR0:
1757 case MSR_P6_PERFCTR1:
1758 case MSR_P6_EVNTSEL0:
1759 case MSR_P6_EVNTSEL1:
1760 case MSR_K7_EVNTSEL0:
1761 case MSR_K7_PERFCTR0:
1762 case MSR_K8_INT_PENDING_MSG:
1763 case MSR_AMD64_NB_CFG:
1764 case MSR_FAM10H_MMIO_CONF_BASE:
1768 data = 0x500 | KVM_NR_VAR_MTRR;
1770 case 0x200 ... 0x2ff:
1771 return get_msr_mtrr(vcpu, msr, pdata);
1772 case 0xcd: /* fsb frequency */
1776 * MSR_EBC_FREQUENCY_ID
1777 * Conservative value valid for even the basic CPU models.
1778 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1779 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1780 * and 266MHz for model 3, or 4. Set Core Clock
1781 * Frequency to System Bus Frequency Ratio to 1 (bits
1782 * 31:24) even though these are only valid for CPU
1783 * models > 2, however guests may end up dividing or
1784 * multiplying by zero otherwise.
1786 case MSR_EBC_FREQUENCY_ID:
1789 case MSR_IA32_APICBASE:
1790 data = kvm_get_apic_base(vcpu);
1792 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1793 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1795 case MSR_IA32_MISC_ENABLE:
1796 data = vcpu->arch.ia32_misc_enable_msr;
1798 case MSR_IA32_PERF_STATUS:
1799 /* TSC increment by tick */
1801 /* CPU multiplier */
1802 data |= (((uint64_t)4ULL) << 40);
1805 data = vcpu->arch.efer;
1807 case MSR_KVM_WALL_CLOCK:
1808 case MSR_KVM_WALL_CLOCK_NEW:
1809 data = vcpu->kvm->arch.wall_clock;
1811 case MSR_KVM_SYSTEM_TIME:
1812 case MSR_KVM_SYSTEM_TIME_NEW:
1813 data = vcpu->arch.time;
1815 case MSR_KVM_ASYNC_PF_EN:
1816 data = vcpu->arch.apf.msr_val;
1818 case MSR_IA32_P5_MC_ADDR:
1819 case MSR_IA32_P5_MC_TYPE:
1820 case MSR_IA32_MCG_CAP:
1821 case MSR_IA32_MCG_CTL:
1822 case MSR_IA32_MCG_STATUS:
1823 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1824 return get_msr_mce(vcpu, msr, pdata);
1825 case MSR_K7_CLK_CTL:
1827 * Provide expected ramp-up count for K7. All other
1828 * are set to zero, indicating minimum divisors for
1831 * This prevents guest kernels on AMD host with CPU
1832 * type 6, model 8 and higher from exploding due to
1833 * the rdmsr failing.
1837 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1838 if (kvm_hv_msr_partition_wide(msr)) {
1840 mutex_lock(&vcpu->kvm->lock);
1841 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1842 mutex_unlock(&vcpu->kvm->lock);
1845 return get_msr_hyperv(vcpu, msr, pdata);
1849 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1852 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1860 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1863 * Read or write a bunch of msrs. All parameters are kernel addresses.
1865 * @return number of msrs set successfully.
1867 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1868 struct kvm_msr_entry *entries,
1869 int (*do_msr)(struct kvm_vcpu *vcpu,
1870 unsigned index, u64 *data))
1874 idx = srcu_read_lock(&vcpu->kvm->srcu);
1875 for (i = 0; i < msrs->nmsrs; ++i)
1876 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1878 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1884 * Read or write a bunch of msrs. Parameters are user addresses.
1886 * @return number of msrs set successfully.
1888 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1889 int (*do_msr)(struct kvm_vcpu *vcpu,
1890 unsigned index, u64 *data),
1893 struct kvm_msrs msrs;
1894 struct kvm_msr_entry *entries;
1899 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1903 if (msrs.nmsrs >= MAX_IO_MSRS)
1907 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1908 entries = kmalloc(size, GFP_KERNEL);
1913 if (copy_from_user(entries, user_msrs->entries, size))
1916 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1921 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1932 int kvm_dev_ioctl_check_extension(long ext)
1937 case KVM_CAP_IRQCHIP:
1939 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1940 case KVM_CAP_SET_TSS_ADDR:
1941 case KVM_CAP_EXT_CPUID:
1942 case KVM_CAP_CLOCKSOURCE:
1944 case KVM_CAP_NOP_IO_DELAY:
1945 case KVM_CAP_MP_STATE:
1946 case KVM_CAP_SYNC_MMU:
1947 case KVM_CAP_USER_NMI:
1948 case KVM_CAP_REINJECT_CONTROL:
1949 case KVM_CAP_IRQ_INJECT_STATUS:
1950 case KVM_CAP_ASSIGN_DEV_IRQ:
1952 case KVM_CAP_IOEVENTFD:
1954 case KVM_CAP_PIT_STATE2:
1955 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1956 case KVM_CAP_XEN_HVM:
1957 case KVM_CAP_ADJUST_CLOCK:
1958 case KVM_CAP_VCPU_EVENTS:
1959 case KVM_CAP_HYPERV:
1960 case KVM_CAP_HYPERV_VAPIC:
1961 case KVM_CAP_HYPERV_SPIN:
1962 case KVM_CAP_PCI_SEGMENT:
1963 case KVM_CAP_DEBUGREGS:
1964 case KVM_CAP_X86_ROBUST_SINGLESTEP:
1966 case KVM_CAP_ASYNC_PF:
1969 case KVM_CAP_COALESCED_MMIO:
1970 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1973 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1975 case KVM_CAP_NR_VCPUS:
1978 case KVM_CAP_NR_MEMSLOTS:
1979 r = KVM_MEMORY_SLOTS;
1981 case KVM_CAP_PV_MMU: /* obsolete */
1988 r = KVM_MAX_MCE_BANKS;
2001 long kvm_arch_dev_ioctl(struct file *filp,
2002 unsigned int ioctl, unsigned long arg)
2004 void __user *argp = (void __user *)arg;
2008 case KVM_GET_MSR_INDEX_LIST: {
2009 struct kvm_msr_list __user *user_msr_list = argp;
2010 struct kvm_msr_list msr_list;
2014 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2017 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2018 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2021 if (n < msr_list.nmsrs)
2024 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2025 num_msrs_to_save * sizeof(u32)))
2027 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2029 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2034 case KVM_GET_SUPPORTED_CPUID: {
2035 struct kvm_cpuid2 __user *cpuid_arg = argp;
2036 struct kvm_cpuid2 cpuid;
2039 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2041 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2042 cpuid_arg->entries);
2047 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2052 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2055 mce_cap = KVM_MCE_CAP_SUPPORTED;
2057 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2069 static void wbinvd_ipi(void *garbage)
2074 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2076 return vcpu->kvm->arch.iommu_domain &&
2077 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2080 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2082 /* Address WBINVD may be executed by guest */
2083 if (need_emulate_wbinvd(vcpu)) {
2084 if (kvm_x86_ops->has_wbinvd_exit())
2085 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2086 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2087 smp_call_function_single(vcpu->cpu,
2088 wbinvd_ipi, NULL, 1);
2091 kvm_x86_ops->vcpu_load(vcpu, cpu);
2092 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2093 /* Make sure TSC doesn't go backwards */
2094 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2095 native_read_tsc() - vcpu->arch.last_host_tsc;
2097 mark_tsc_unstable("KVM discovered backwards TSC");
2098 if (check_tsc_unstable()) {
2099 kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2100 vcpu->arch.tsc_catchup = 1;
2101 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2103 if (vcpu->cpu != cpu)
2104 kvm_migrate_timers(vcpu);
2109 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2111 kvm_x86_ops->vcpu_put(vcpu);
2112 kvm_put_guest_fpu(vcpu);
2113 vcpu->arch.last_host_tsc = native_read_tsc();
2116 static int is_efer_nx(void)
2118 unsigned long long efer = 0;
2120 rdmsrl_safe(MSR_EFER, &efer);
2121 return efer & EFER_NX;
2124 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2127 struct kvm_cpuid_entry2 *e, *entry;
2130 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2131 e = &vcpu->arch.cpuid_entries[i];
2132 if (e->function == 0x80000001) {
2137 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2138 entry->edx &= ~(1 << 20);
2139 printk(KERN_INFO "kvm: guest NX capability removed\n");
2143 /* when an old userspace process fills a new kernel module */
2144 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2145 struct kvm_cpuid *cpuid,
2146 struct kvm_cpuid_entry __user *entries)
2149 struct kvm_cpuid_entry *cpuid_entries;
2152 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2155 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2159 if (copy_from_user(cpuid_entries, entries,
2160 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2162 for (i = 0; i < cpuid->nent; i++) {
2163 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2164 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2165 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2166 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2167 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2168 vcpu->arch.cpuid_entries[i].index = 0;
2169 vcpu->arch.cpuid_entries[i].flags = 0;
2170 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2171 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2172 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2174 vcpu->arch.cpuid_nent = cpuid->nent;
2175 cpuid_fix_nx_cap(vcpu);
2177 kvm_apic_set_version(vcpu);
2178 kvm_x86_ops->cpuid_update(vcpu);
2182 vfree(cpuid_entries);
2187 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2188 struct kvm_cpuid2 *cpuid,
2189 struct kvm_cpuid_entry2 __user *entries)
2194 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2197 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2198 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2200 vcpu->arch.cpuid_nent = cpuid->nent;
2201 kvm_apic_set_version(vcpu);
2202 kvm_x86_ops->cpuid_update(vcpu);
2210 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2211 struct kvm_cpuid2 *cpuid,
2212 struct kvm_cpuid_entry2 __user *entries)
2217 if (cpuid->nent < vcpu->arch.cpuid_nent)
2220 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2221 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2226 cpuid->nent = vcpu->arch.cpuid_nent;
2230 static void cpuid_mask(u32 *word, int wordnum)
2232 *word &= boot_cpu_data.x86_capability[wordnum];
2235 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2238 entry->function = function;
2239 entry->index = index;
2240 cpuid_count(entry->function, entry->index,
2241 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2245 #define F(x) bit(X86_FEATURE_##x)
2247 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2248 u32 index, int *nent, int maxnent)
2250 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2251 #ifdef CONFIG_X86_64
2252 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2254 unsigned f_lm = F(LM);
2256 unsigned f_gbpages = 0;
2259 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2262 const u32 kvm_supported_word0_x86_features =
2263 F(FPU) | F(VME) | F(DE) | F(PSE) |
2264 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2265 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2266 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2267 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2268 0 /* Reserved, DS, ACPI */ | F(MMX) |
2269 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2270 0 /* HTT, TM, Reserved, PBE */;
2271 /* cpuid 0x80000001.edx */
2272 const u32 kvm_supported_word1_x86_features =
2273 F(FPU) | F(VME) | F(DE) | F(PSE) |
2274 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2275 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2276 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2277 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2278 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2279 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2280 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2282 const u32 kvm_supported_word4_x86_features =
2283 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
2284 0 /* DS-CPL, VMX, SMX, EST */ |
2285 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2286 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2287 0 /* Reserved, DCA */ | F(XMM4_1) |
2288 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2289 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2291 /* cpuid 0x80000001.ecx */
2292 const u32 kvm_supported_word6_x86_features =
2293 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2294 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2295 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2296 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2298 /* all calls to cpuid_count() should be made on the same cpu */
2300 do_cpuid_1_ent(entry, function, index);
2305 entry->eax = min(entry->eax, (u32)0xd);
2308 entry->edx &= kvm_supported_word0_x86_features;
2309 cpuid_mask(&entry->edx, 0);
2310 entry->ecx &= kvm_supported_word4_x86_features;
2311 cpuid_mask(&entry->ecx, 4);
2312 /* we support x2apic emulation even if host does not support
2313 * it since we emulate x2apic in software */
2314 entry->ecx |= F(X2APIC);
2316 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2317 * may return different values. This forces us to get_cpu() before
2318 * issuing the first command, and also to emulate this annoying behavior
2319 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2321 int t, times = entry->eax & 0xff;
2323 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2324 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2325 for (t = 1; t < times && *nent < maxnent; ++t) {
2326 do_cpuid_1_ent(&entry[t], function, 0);
2327 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2332 /* function 4 and 0xb have additional index. */
2336 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2337 /* read more entries until cache_type is zero */
2338 for (i = 1; *nent < maxnent; ++i) {
2339 cache_type = entry[i - 1].eax & 0x1f;
2342 do_cpuid_1_ent(&entry[i], function, i);
2344 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2352 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2353 /* read more entries until level_type is zero */
2354 for (i = 1; *nent < maxnent; ++i) {
2355 level_type = entry[i - 1].ecx & 0xff00;
2358 do_cpuid_1_ent(&entry[i], function, i);
2360 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2368 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2369 for (i = 1; *nent < maxnent; ++i) {
2370 if (entry[i - 1].eax == 0 && i != 2)
2372 do_cpuid_1_ent(&entry[i], function, i);
2374 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2379 case KVM_CPUID_SIGNATURE: {
2380 char signature[12] = "KVMKVMKVM\0\0";
2381 u32 *sigptr = (u32 *)signature;
2383 entry->ebx = sigptr[0];
2384 entry->ecx = sigptr[1];
2385 entry->edx = sigptr[2];
2388 case KVM_CPUID_FEATURES:
2389 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2390 (1 << KVM_FEATURE_NOP_IO_DELAY) |
2391 (1 << KVM_FEATURE_CLOCKSOURCE2) |
2392 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2398 entry->eax = min(entry->eax, 0x8000001a);
2401 entry->edx &= kvm_supported_word1_x86_features;
2402 cpuid_mask(&entry->edx, 1);
2403 entry->ecx &= kvm_supported_word6_x86_features;
2404 cpuid_mask(&entry->ecx, 6);
2408 kvm_x86_ops->set_supported_cpuid(function, entry);
2415 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2416 struct kvm_cpuid_entry2 __user *entries)
2418 struct kvm_cpuid_entry2 *cpuid_entries;
2419 int limit, nent = 0, r = -E2BIG;
2422 if (cpuid->nent < 1)
2424 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2425 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2427 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2431 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2432 limit = cpuid_entries[0].eax;
2433 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2434 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2435 &nent, cpuid->nent);
2437 if (nent >= cpuid->nent)
2440 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2441 limit = cpuid_entries[nent - 1].eax;
2442 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2443 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2444 &nent, cpuid->nent);
2449 if (nent >= cpuid->nent)
2452 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2456 if (nent >= cpuid->nent)
2459 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2463 if (nent >= cpuid->nent)
2467 if (copy_to_user(entries, cpuid_entries,
2468 nent * sizeof(struct kvm_cpuid_entry2)))
2474 vfree(cpuid_entries);
2479 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2480 struct kvm_lapic_state *s)
2482 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2487 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2488 struct kvm_lapic_state *s)
2490 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2491 kvm_apic_post_state_restore(vcpu);
2492 update_cr8_intercept(vcpu);
2497 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2498 struct kvm_interrupt *irq)
2500 if (irq->irq < 0 || irq->irq >= 256)
2502 if (irqchip_in_kernel(vcpu->kvm))
2505 kvm_queue_interrupt(vcpu, irq->irq, false);
2506 kvm_make_request(KVM_REQ_EVENT, vcpu);
2511 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2513 kvm_inject_nmi(vcpu);
2518 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2519 struct kvm_tpr_access_ctl *tac)
2523 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2527 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2531 unsigned bank_num = mcg_cap & 0xff, bank;
2534 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2536 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2539 vcpu->arch.mcg_cap = mcg_cap;
2540 /* Init IA32_MCG_CTL to all 1s */
2541 if (mcg_cap & MCG_CTL_P)
2542 vcpu->arch.mcg_ctl = ~(u64)0;
2543 /* Init IA32_MCi_CTL to all 1s */
2544 for (bank = 0; bank < bank_num; bank++)
2545 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2550 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2551 struct kvm_x86_mce *mce)
2553 u64 mcg_cap = vcpu->arch.mcg_cap;
2554 unsigned bank_num = mcg_cap & 0xff;
2555 u64 *banks = vcpu->arch.mce_banks;
2557 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2560 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2561 * reporting is disabled
2563 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2564 vcpu->arch.mcg_ctl != ~(u64)0)
2566 banks += 4 * mce->bank;
2568 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2569 * reporting is disabled for the bank
2571 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2573 if (mce->status & MCI_STATUS_UC) {
2574 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2575 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2576 printk(KERN_DEBUG "kvm: set_mce: "
2577 "injects mce exception while "
2578 "previous one is in progress!\n");
2579 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2582 if (banks[1] & MCI_STATUS_VAL)
2583 mce->status |= MCI_STATUS_OVER;
2584 banks[2] = mce->addr;
2585 banks[3] = mce->misc;
2586 vcpu->arch.mcg_status = mce->mcg_status;
2587 banks[1] = mce->status;
2588 kvm_queue_exception(vcpu, MC_VECTOR);
2589 } else if (!(banks[1] & MCI_STATUS_VAL)
2590 || !(banks[1] & MCI_STATUS_UC)) {
2591 if (banks[1] & MCI_STATUS_VAL)
2592 mce->status |= MCI_STATUS_OVER;
2593 banks[2] = mce->addr;
2594 banks[3] = mce->misc;
2595 banks[1] = mce->status;
2597 banks[1] |= MCI_STATUS_OVER;
2601 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2602 struct kvm_vcpu_events *events)
2604 events->exception.injected =
2605 vcpu->arch.exception.pending &&
2606 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2607 events->exception.nr = vcpu->arch.exception.nr;
2608 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2609 events->exception.pad = 0;
2610 events->exception.error_code = vcpu->arch.exception.error_code;
2612 events->interrupt.injected =
2613 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2614 events->interrupt.nr = vcpu->arch.interrupt.nr;
2615 events->interrupt.soft = 0;
2616 events->interrupt.shadow =
2617 kvm_x86_ops->get_interrupt_shadow(vcpu,
2618 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2620 events->nmi.injected = vcpu->arch.nmi_injected;
2621 events->nmi.pending = vcpu->arch.nmi_pending;
2622 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2623 events->nmi.pad = 0;
2625 events->sipi_vector = vcpu->arch.sipi_vector;
2627 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2628 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2629 | KVM_VCPUEVENT_VALID_SHADOW);
2630 memset(&events->reserved, 0, sizeof(events->reserved));
2633 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2634 struct kvm_vcpu_events *events)
2636 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2637 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2638 | KVM_VCPUEVENT_VALID_SHADOW))
2641 vcpu->arch.exception.pending = events->exception.injected;
2642 vcpu->arch.exception.nr = events->exception.nr;
2643 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2644 vcpu->arch.exception.error_code = events->exception.error_code;
2646 vcpu->arch.interrupt.pending = events->interrupt.injected;
2647 vcpu->arch.interrupt.nr = events->interrupt.nr;
2648 vcpu->arch.interrupt.soft = events->interrupt.soft;
2649 if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
2650 kvm_pic_clear_isr_ack(vcpu->kvm);
2651 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2652 kvm_x86_ops->set_interrupt_shadow(vcpu,
2653 events->interrupt.shadow);
2655 vcpu->arch.nmi_injected = events->nmi.injected;
2656 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2657 vcpu->arch.nmi_pending = events->nmi.pending;
2658 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2660 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2661 vcpu->arch.sipi_vector = events->sipi_vector;
2663 kvm_make_request(KVM_REQ_EVENT, vcpu);
2668 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2669 struct kvm_debugregs *dbgregs)
2671 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2672 dbgregs->dr6 = vcpu->arch.dr6;
2673 dbgregs->dr7 = vcpu->arch.dr7;
2675 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2678 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2679 struct kvm_debugregs *dbgregs)
2684 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2685 vcpu->arch.dr6 = dbgregs->dr6;
2686 vcpu->arch.dr7 = dbgregs->dr7;
2691 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2692 struct kvm_xsave *guest_xsave)
2695 memcpy(guest_xsave->region,
2696 &vcpu->arch.guest_fpu.state->xsave,
2699 memcpy(guest_xsave->region,
2700 &vcpu->arch.guest_fpu.state->fxsave,
2701 sizeof(struct i387_fxsave_struct));
2702 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2707 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2708 struct kvm_xsave *guest_xsave)
2711 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2714 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2715 guest_xsave->region, xstate_size);
2717 if (xstate_bv & ~XSTATE_FPSSE)
2719 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2720 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2725 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2726 struct kvm_xcrs *guest_xcrs)
2728 if (!cpu_has_xsave) {
2729 guest_xcrs->nr_xcrs = 0;
2733 guest_xcrs->nr_xcrs = 1;
2734 guest_xcrs->flags = 0;
2735 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2736 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2739 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2740 struct kvm_xcrs *guest_xcrs)
2747 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2750 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2751 /* Only support XCR0 currently */
2752 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2753 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2754 guest_xcrs->xcrs[0].value);
2762 long kvm_arch_vcpu_ioctl(struct file *filp,
2763 unsigned int ioctl, unsigned long arg)
2765 struct kvm_vcpu *vcpu = filp->private_data;
2766 void __user *argp = (void __user *)arg;
2769 struct kvm_lapic_state *lapic;
2770 struct kvm_xsave *xsave;
2771 struct kvm_xcrs *xcrs;
2777 case KVM_GET_LAPIC: {
2779 if (!vcpu->arch.apic)
2781 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2786 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
2790 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
2795 case KVM_SET_LAPIC: {
2797 if (!vcpu->arch.apic)
2799 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2804 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
2806 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
2812 case KVM_INTERRUPT: {
2813 struct kvm_interrupt irq;
2816 if (copy_from_user(&irq, argp, sizeof irq))
2818 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2825 r = kvm_vcpu_ioctl_nmi(vcpu);
2831 case KVM_SET_CPUID: {
2832 struct kvm_cpuid __user *cpuid_arg = argp;
2833 struct kvm_cpuid cpuid;
2836 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2838 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2843 case KVM_SET_CPUID2: {
2844 struct kvm_cpuid2 __user *cpuid_arg = argp;
2845 struct kvm_cpuid2 cpuid;
2848 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2850 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2851 cpuid_arg->entries);
2856 case KVM_GET_CPUID2: {
2857 struct kvm_cpuid2 __user *cpuid_arg = argp;
2858 struct kvm_cpuid2 cpuid;
2861 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2863 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2864 cpuid_arg->entries);
2868 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2874 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2877 r = msr_io(vcpu, argp, do_set_msr, 0);
2879 case KVM_TPR_ACCESS_REPORTING: {
2880 struct kvm_tpr_access_ctl tac;
2883 if (copy_from_user(&tac, argp, sizeof tac))
2885 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2889 if (copy_to_user(argp, &tac, sizeof tac))
2894 case KVM_SET_VAPIC_ADDR: {
2895 struct kvm_vapic_addr va;
2898 if (!irqchip_in_kernel(vcpu->kvm))
2901 if (copy_from_user(&va, argp, sizeof va))
2904 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2907 case KVM_X86_SETUP_MCE: {
2911 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2913 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2916 case KVM_X86_SET_MCE: {
2917 struct kvm_x86_mce mce;
2920 if (copy_from_user(&mce, argp, sizeof mce))
2922 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2925 case KVM_GET_VCPU_EVENTS: {
2926 struct kvm_vcpu_events events;
2928 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2931 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2936 case KVM_SET_VCPU_EVENTS: {
2937 struct kvm_vcpu_events events;
2940 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2943 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2946 case KVM_GET_DEBUGREGS: {
2947 struct kvm_debugregs dbgregs;
2949 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2952 if (copy_to_user(argp, &dbgregs,
2953 sizeof(struct kvm_debugregs)))
2958 case KVM_SET_DEBUGREGS: {
2959 struct kvm_debugregs dbgregs;
2962 if (copy_from_user(&dbgregs, argp,
2963 sizeof(struct kvm_debugregs)))
2966 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2969 case KVM_GET_XSAVE: {
2970 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2975 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
2978 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
2983 case KVM_SET_XSAVE: {
2984 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2990 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
2993 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
2996 case KVM_GET_XCRS: {
2997 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3002 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3005 if (copy_to_user(argp, u.xcrs,
3006 sizeof(struct kvm_xcrs)))
3011 case KVM_SET_XCRS: {
3012 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3018 if (copy_from_user(u.xcrs, argp,
3019 sizeof(struct kvm_xcrs)))
3022 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3033 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3037 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3039 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3043 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3046 kvm->arch.ept_identity_map_addr = ident_addr;
3050 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3051 u32 kvm_nr_mmu_pages)
3053 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3056 mutex_lock(&kvm->slots_lock);
3057 spin_lock(&kvm->mmu_lock);
3059 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3060 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3062 spin_unlock(&kvm->mmu_lock);
3063 mutex_unlock(&kvm->slots_lock);
3067 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3069 return kvm->arch.n_max_mmu_pages;
3072 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3077 switch (chip->chip_id) {
3078 case KVM_IRQCHIP_PIC_MASTER:
3079 memcpy(&chip->chip.pic,
3080 &pic_irqchip(kvm)->pics[0],
3081 sizeof(struct kvm_pic_state));
3083 case KVM_IRQCHIP_PIC_SLAVE:
3084 memcpy(&chip->chip.pic,
3085 &pic_irqchip(kvm)->pics[1],
3086 sizeof(struct kvm_pic_state));
3088 case KVM_IRQCHIP_IOAPIC:
3089 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3098 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3103 switch (chip->chip_id) {
3104 case KVM_IRQCHIP_PIC_MASTER:
3105 spin_lock(&pic_irqchip(kvm)->lock);
3106 memcpy(&pic_irqchip(kvm)->pics[0],
3108 sizeof(struct kvm_pic_state));
3109 spin_unlock(&pic_irqchip(kvm)->lock);
3111 case KVM_IRQCHIP_PIC_SLAVE:
3112 spin_lock(&pic_irqchip(kvm)->lock);
3113 memcpy(&pic_irqchip(kvm)->pics[1],
3115 sizeof(struct kvm_pic_state));
3116 spin_unlock(&pic_irqchip(kvm)->lock);
3118 case KVM_IRQCHIP_IOAPIC:
3119 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3125 kvm_pic_update_irq(pic_irqchip(kvm));
3129 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3133 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3134 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3135 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3139 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3143 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3144 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3145 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3146 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3150 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3154 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3155 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3156 sizeof(ps->channels));
3157 ps->flags = kvm->arch.vpit->pit_state.flags;
3158 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3159 memset(&ps->reserved, 0, sizeof(ps->reserved));
3163 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3165 int r = 0, start = 0;
3166 u32 prev_legacy, cur_legacy;
3167 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3168 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3169 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3170 if (!prev_legacy && cur_legacy)
3172 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3173 sizeof(kvm->arch.vpit->pit_state.channels));
3174 kvm->arch.vpit->pit_state.flags = ps->flags;
3175 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3176 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3180 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3181 struct kvm_reinject_control *control)
3183 if (!kvm->arch.vpit)
3185 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3186 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3187 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3192 * Get (and clear) the dirty memory log for a memory slot.
3194 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3195 struct kvm_dirty_log *log)
3198 struct kvm_memory_slot *memslot;
3200 unsigned long is_dirty = 0;
3202 mutex_lock(&kvm->slots_lock);
3205 if (log->slot >= KVM_MEMORY_SLOTS)
3208 memslot = &kvm->memslots->memslots[log->slot];
3210 if (!memslot->dirty_bitmap)
3213 n = kvm_dirty_bitmap_bytes(memslot);
3215 for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3216 is_dirty = memslot->dirty_bitmap[i];
3218 /* If nothing is dirty, don't bother messing with page tables. */
3220 struct kvm_memslots *slots, *old_slots;
3221 unsigned long *dirty_bitmap;
3223 dirty_bitmap = memslot->dirty_bitmap_head;
3224 if (memslot->dirty_bitmap == dirty_bitmap)
3225 dirty_bitmap += n / sizeof(long);
3226 memset(dirty_bitmap, 0, n);
3229 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3232 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3233 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3234 slots->generation++;
3236 old_slots = kvm->memslots;
3237 rcu_assign_pointer(kvm->memslots, slots);
3238 synchronize_srcu_expedited(&kvm->srcu);
3239 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3242 spin_lock(&kvm->mmu_lock);
3243 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3244 spin_unlock(&kvm->mmu_lock);
3247 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
3251 if (clear_user(log->dirty_bitmap, n))
3257 mutex_unlock(&kvm->slots_lock);
3261 long kvm_arch_vm_ioctl(struct file *filp,
3262 unsigned int ioctl, unsigned long arg)
3264 struct kvm *kvm = filp->private_data;
3265 void __user *argp = (void __user *)arg;
3268 * This union makes it completely explicit to gcc-3.x
3269 * that these two variables' stack usage should be
3270 * combined, not added together.
3273 struct kvm_pit_state ps;
3274 struct kvm_pit_state2 ps2;
3275 struct kvm_pit_config pit_config;
3279 case KVM_SET_TSS_ADDR:
3280 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3284 case KVM_SET_IDENTITY_MAP_ADDR: {
3288 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3290 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3295 case KVM_SET_NR_MMU_PAGES:
3296 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3300 case KVM_GET_NR_MMU_PAGES:
3301 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3303 case KVM_CREATE_IRQCHIP: {
3304 struct kvm_pic *vpic;
3306 mutex_lock(&kvm->lock);
3309 goto create_irqchip_unlock;
3311 vpic = kvm_create_pic(kvm);
3313 r = kvm_ioapic_init(kvm);
3315 mutex_lock(&kvm->slots_lock);
3316 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3318 mutex_unlock(&kvm->slots_lock);
3320 goto create_irqchip_unlock;
3323 goto create_irqchip_unlock;
3325 kvm->arch.vpic = vpic;
3327 r = kvm_setup_default_irq_routing(kvm);
3329 mutex_lock(&kvm->slots_lock);
3330 mutex_lock(&kvm->irq_lock);
3331 kvm_ioapic_destroy(kvm);
3332 kvm_destroy_pic(kvm);
3333 mutex_unlock(&kvm->irq_lock);
3334 mutex_unlock(&kvm->slots_lock);
3336 create_irqchip_unlock:
3337 mutex_unlock(&kvm->lock);
3340 case KVM_CREATE_PIT:
3341 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3343 case KVM_CREATE_PIT2:
3345 if (copy_from_user(&u.pit_config, argp,
3346 sizeof(struct kvm_pit_config)))
3349 mutex_lock(&kvm->slots_lock);
3352 goto create_pit_unlock;
3354 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3358 mutex_unlock(&kvm->slots_lock);
3360 case KVM_IRQ_LINE_STATUS:
3361 case KVM_IRQ_LINE: {
3362 struct kvm_irq_level irq_event;
3365 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3368 if (irqchip_in_kernel(kvm)) {
3370 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3371 irq_event.irq, irq_event.level);
3372 if (ioctl == KVM_IRQ_LINE_STATUS) {
3374 irq_event.status = status;
3375 if (copy_to_user(argp, &irq_event,
3383 case KVM_GET_IRQCHIP: {
3384 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3385 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3391 if (copy_from_user(chip, argp, sizeof *chip))
3392 goto get_irqchip_out;
3394 if (!irqchip_in_kernel(kvm))
3395 goto get_irqchip_out;
3396 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3398 goto get_irqchip_out;
3400 if (copy_to_user(argp, chip, sizeof *chip))
3401 goto get_irqchip_out;
3409 case KVM_SET_IRQCHIP: {
3410 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3411 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3417 if (copy_from_user(chip, argp, sizeof *chip))
3418 goto set_irqchip_out;
3420 if (!irqchip_in_kernel(kvm))
3421 goto set_irqchip_out;
3422 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3424 goto set_irqchip_out;
3434 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3437 if (!kvm->arch.vpit)
3439 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3443 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3450 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3453 if (!kvm->arch.vpit)
3455 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3461 case KVM_GET_PIT2: {
3463 if (!kvm->arch.vpit)
3465 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3469 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3474 case KVM_SET_PIT2: {
3476 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3479 if (!kvm->arch.vpit)
3481 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3487 case KVM_REINJECT_CONTROL: {
3488 struct kvm_reinject_control control;
3490 if (copy_from_user(&control, argp, sizeof(control)))
3492 r = kvm_vm_ioctl_reinject(kvm, &control);
3498 case KVM_XEN_HVM_CONFIG: {
3500 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3501 sizeof(struct kvm_xen_hvm_config)))
3504 if (kvm->arch.xen_hvm_config.flags)
3509 case KVM_SET_CLOCK: {
3510 struct kvm_clock_data user_ns;
3515 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3523 local_irq_disable();
3524 now_ns = get_kernel_ns();
3525 delta = user_ns.clock - now_ns;
3527 kvm->arch.kvmclock_offset = delta;
3530 case KVM_GET_CLOCK: {
3531 struct kvm_clock_data user_ns;
3534 local_irq_disable();
3535 now_ns = get_kernel_ns();
3536 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3539 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3542 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3555 static void kvm_init_msr_list(void)
3560 /* skip the first msrs in the list. KVM-specific */
3561 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3562 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3565 msrs_to_save[j] = msrs_to_save[i];
3568 num_msrs_to_save = j;
3571 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3574 if (vcpu->arch.apic &&
3575 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
3578 return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3581 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3583 if (vcpu->arch.apic &&
3584 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
3587 return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3590 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3591 struct kvm_segment *var, int seg)
3593 kvm_x86_ops->set_segment(vcpu, var, seg);
3596 void kvm_get_segment(struct kvm_vcpu *vcpu,
3597 struct kvm_segment *var, int seg)
3599 kvm_x86_ops->get_segment(vcpu, var, seg);
3602 static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3607 static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3610 struct x86_exception exception;
3612 BUG_ON(!mmu_is_nested(vcpu));
3614 /* NPT walks are always user-walks */
3615 access |= PFERR_USER_MASK;
3616 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3621 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3622 struct x86_exception *exception)
3624 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3625 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3628 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3629 struct x86_exception *exception)
3631 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3632 access |= PFERR_FETCH_MASK;
3633 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3636 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3637 struct x86_exception *exception)
3639 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3640 access |= PFERR_WRITE_MASK;
3641 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3644 /* uses this to access any guest's mapped memory without checking CPL */
3645 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3646 struct x86_exception *exception)
3648 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3651 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3652 struct kvm_vcpu *vcpu, u32 access,
3653 struct x86_exception *exception)
3656 int r = X86EMUL_CONTINUE;
3659 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3661 unsigned offset = addr & (PAGE_SIZE-1);
3662 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3665 if (gpa == UNMAPPED_GVA)
3666 return X86EMUL_PROPAGATE_FAULT;
3667 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3669 r = X86EMUL_IO_NEEDED;
3681 /* used for instruction fetching */
3682 static int kvm_fetch_guest_virt(gva_t addr, void *val, unsigned int bytes,
3683 struct kvm_vcpu *vcpu,
3684 struct x86_exception *exception)
3686 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3687 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3688 access | PFERR_FETCH_MASK,
3692 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3693 struct kvm_vcpu *vcpu,
3694 struct x86_exception *exception)
3696 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3697 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3701 static int kvm_read_guest_virt_system(gva_t addr, void *val, unsigned int bytes,
3702 struct kvm_vcpu *vcpu,
3703 struct x86_exception *exception)
3705 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3708 static int kvm_write_guest_virt_system(gva_t addr, void *val,
3710 struct kvm_vcpu *vcpu,
3711 struct x86_exception *exception)
3714 int r = X86EMUL_CONTINUE;
3717 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3720 unsigned offset = addr & (PAGE_SIZE-1);
3721 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3724 if (gpa == UNMAPPED_GVA)
3725 return X86EMUL_PROPAGATE_FAULT;
3726 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3728 r = X86EMUL_IO_NEEDED;
3740 static int emulator_read_emulated(unsigned long addr,
3743 struct x86_exception *exception,
3744 struct kvm_vcpu *vcpu)
3748 if (vcpu->mmio_read_completed) {
3749 memcpy(val, vcpu->mmio_data, bytes);
3750 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3751 vcpu->mmio_phys_addr, *(u64 *)val);
3752 vcpu->mmio_read_completed = 0;
3753 return X86EMUL_CONTINUE;
3756 gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, exception);
3758 if (gpa == UNMAPPED_GVA)
3759 return X86EMUL_PROPAGATE_FAULT;
3761 /* For APIC access vmexit */
3762 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3765 if (kvm_read_guest_virt(addr, val, bytes, vcpu, exception)
3766 == X86EMUL_CONTINUE)
3767 return X86EMUL_CONTINUE;
3771 * Is this MMIO handled locally?
3773 if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
3774 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
3775 return X86EMUL_CONTINUE;
3778 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3780 vcpu->mmio_needed = 1;
3781 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3782 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3783 vcpu->run->mmio.len = vcpu->mmio_size = bytes;
3784 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 0;
3786 return X86EMUL_IO_NEEDED;
3789 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3790 const void *val, int bytes)
3794 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3797 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3801 static int emulator_write_emulated_onepage(unsigned long addr,
3804 struct x86_exception *exception,
3805 struct kvm_vcpu *vcpu)
3809 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, exception);
3811 if (gpa == UNMAPPED_GVA)
3812 return X86EMUL_PROPAGATE_FAULT;
3814 /* For APIC access vmexit */
3815 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3818 if (emulator_write_phys(vcpu, gpa, val, bytes))
3819 return X86EMUL_CONTINUE;
3822 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3824 * Is this MMIO handled locally?
3826 if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
3827 return X86EMUL_CONTINUE;
3829 vcpu->mmio_needed = 1;
3830 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3831 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3832 vcpu->run->mmio.len = vcpu->mmio_size = bytes;
3833 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 1;
3834 memcpy(vcpu->run->mmio.data, val, bytes);
3836 return X86EMUL_CONTINUE;
3839 int emulator_write_emulated(unsigned long addr,
3842 struct x86_exception *exception,
3843 struct kvm_vcpu *vcpu)
3845 /* Crossing a page boundary? */
3846 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3849 now = -addr & ~PAGE_MASK;
3850 rc = emulator_write_emulated_onepage(addr, val, now, exception,
3852 if (rc != X86EMUL_CONTINUE)
3858 return emulator_write_emulated_onepage(addr, val, bytes, exception,
3862 #define CMPXCHG_TYPE(t, ptr, old, new) \
3863 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3865 #ifdef CONFIG_X86_64
3866 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3868 # define CMPXCHG64(ptr, old, new) \
3869 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
3872 static int emulator_cmpxchg_emulated(unsigned long addr,
3876 struct x86_exception *exception,
3877 struct kvm_vcpu *vcpu)
3884 /* guests cmpxchg8b have to be emulated atomically */
3885 if (bytes > 8 || (bytes & (bytes - 1)))
3888 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3890 if (gpa == UNMAPPED_GVA ||
3891 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3894 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3897 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3898 if (is_error_page(page)) {
3899 kvm_release_page_clean(page);
3903 kaddr = kmap_atomic(page, KM_USER0);
3904 kaddr += offset_in_page(gpa);
3907 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
3910 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
3913 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
3916 exchanged = CMPXCHG64(kaddr, old, new);
3921 kunmap_atomic(kaddr, KM_USER0);
3922 kvm_release_page_dirty(page);
3925 return X86EMUL_CMPXCHG_FAILED;
3927 kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
3929 return X86EMUL_CONTINUE;
3932 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3934 return emulator_write_emulated(addr, new, bytes, exception, vcpu);
3937 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3939 /* TODO: String I/O for in kernel device */
3942 if (vcpu->arch.pio.in)
3943 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3944 vcpu->arch.pio.size, pd);
3946 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3947 vcpu->arch.pio.port, vcpu->arch.pio.size,
3953 static int emulator_pio_in_emulated(int size, unsigned short port, void *val,
3954 unsigned int count, struct kvm_vcpu *vcpu)
3956 if (vcpu->arch.pio.count)
3959 trace_kvm_pio(0, port, size, count);
3961 vcpu->arch.pio.port = port;
3962 vcpu->arch.pio.in = 1;
3963 vcpu->arch.pio.count = count;
3964 vcpu->arch.pio.size = size;
3966 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3968 memcpy(val, vcpu->arch.pio_data, size * count);
3969 vcpu->arch.pio.count = 0;
3973 vcpu->run->exit_reason = KVM_EXIT_IO;
3974 vcpu->run->io.direction = KVM_EXIT_IO_IN;
3975 vcpu->run->io.size = size;
3976 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3977 vcpu->run->io.count = count;
3978 vcpu->run->io.port = port;
3983 static int emulator_pio_out_emulated(int size, unsigned short port,
3984 const void *val, unsigned int count,
3985 struct kvm_vcpu *vcpu)
3987 trace_kvm_pio(1, port, size, count);
3989 vcpu->arch.pio.port = port;
3990 vcpu->arch.pio.in = 0;
3991 vcpu->arch.pio.count = count;
3992 vcpu->arch.pio.size = size;
3994 memcpy(vcpu->arch.pio_data, val, size * count);
3996 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3997 vcpu->arch.pio.count = 0;
4001 vcpu->run->exit_reason = KVM_EXIT_IO;
4002 vcpu->run->io.direction = KVM_EXIT_IO_OUT;
4003 vcpu->run->io.size = size;
4004 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4005 vcpu->run->io.count = count;
4006 vcpu->run->io.port = port;
4011 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4013 return kvm_x86_ops->get_segment_base(vcpu, seg);
4016 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
4018 kvm_mmu_invlpg(vcpu, address);
4019 return X86EMUL_CONTINUE;
4022 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4024 if (!need_emulate_wbinvd(vcpu))
4025 return X86EMUL_CONTINUE;
4027 if (kvm_x86_ops->has_wbinvd_exit()) {
4028 int cpu = get_cpu();
4030 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4031 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4032 wbinvd_ipi, NULL, 1);
4034 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4037 return X86EMUL_CONTINUE;
4039 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4041 int emulate_clts(struct kvm_vcpu *vcpu)
4043 kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4044 kvm_x86_ops->fpu_activate(vcpu);
4045 return X86EMUL_CONTINUE;
4048 int emulator_get_dr(int dr, unsigned long *dest, struct kvm_vcpu *vcpu)
4050 return _kvm_get_dr(vcpu, dr, dest);
4053 int emulator_set_dr(int dr, unsigned long value, struct kvm_vcpu *vcpu)
4056 return __kvm_set_dr(vcpu, dr, value);
4059 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4061 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4064 static unsigned long emulator_get_cr(int cr, struct kvm_vcpu *vcpu)
4066 unsigned long value;
4070 value = kvm_read_cr0(vcpu);
4073 value = vcpu->arch.cr2;
4076 value = vcpu->arch.cr3;
4079 value = kvm_read_cr4(vcpu);
4082 value = kvm_get_cr8(vcpu);
4085 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4092 static int emulator_set_cr(int cr, unsigned long val, struct kvm_vcpu *vcpu)
4098 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4101 vcpu->arch.cr2 = val;
4104 res = kvm_set_cr3(vcpu, val);
4107 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4110 res = kvm_set_cr8(vcpu, val);
4113 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4120 static int emulator_get_cpl(struct kvm_vcpu *vcpu)
4122 return kvm_x86_ops->get_cpl(vcpu);
4125 static void emulator_get_gdt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
4127 kvm_x86_ops->get_gdt(vcpu, dt);
4130 static void emulator_get_idt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
4132 kvm_x86_ops->get_idt(vcpu, dt);
4135 static unsigned long emulator_get_cached_segment_base(int seg,
4136 struct kvm_vcpu *vcpu)
4138 return get_segment_base(vcpu, seg);
4141 static bool emulator_get_cached_descriptor(struct desc_struct *desc, int seg,
4142 struct kvm_vcpu *vcpu)
4144 struct kvm_segment var;
4146 kvm_get_segment(vcpu, &var, seg);
4153 set_desc_limit(desc, var.limit);
4154 set_desc_base(desc, (unsigned long)var.base);
4155 desc->type = var.type;
4157 desc->dpl = var.dpl;
4158 desc->p = var.present;
4159 desc->avl = var.avl;
4167 static void emulator_set_cached_descriptor(struct desc_struct *desc, int seg,
4168 struct kvm_vcpu *vcpu)
4170 struct kvm_segment var;
4172 /* needed to preserve selector */
4173 kvm_get_segment(vcpu, &var, seg);
4175 var.base = get_desc_base(desc);
4176 var.limit = get_desc_limit(desc);
4178 var.limit = (var.limit << 12) | 0xfff;
4179 var.type = desc->type;
4180 var.present = desc->p;
4181 var.dpl = desc->dpl;
4186 var.avl = desc->avl;
4187 var.present = desc->p;
4188 var.unusable = !var.present;
4191 kvm_set_segment(vcpu, &var, seg);
4195 static u16 emulator_get_segment_selector(int seg, struct kvm_vcpu *vcpu)
4197 struct kvm_segment kvm_seg;
4199 kvm_get_segment(vcpu, &kvm_seg, seg);
4200 return kvm_seg.selector;
4203 static void emulator_set_segment_selector(u16 sel, int seg,
4204 struct kvm_vcpu *vcpu)
4206 struct kvm_segment kvm_seg;
4208 kvm_get_segment(vcpu, &kvm_seg, seg);
4209 kvm_seg.selector = sel;
4210 kvm_set_segment(vcpu, &kvm_seg, seg);
4213 static struct x86_emulate_ops emulate_ops = {
4214 .read_std = kvm_read_guest_virt_system,
4215 .write_std = kvm_write_guest_virt_system,
4216 .fetch = kvm_fetch_guest_virt,
4217 .read_emulated = emulator_read_emulated,
4218 .write_emulated = emulator_write_emulated,
4219 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4220 .pio_in_emulated = emulator_pio_in_emulated,
4221 .pio_out_emulated = emulator_pio_out_emulated,
4222 .get_cached_descriptor = emulator_get_cached_descriptor,
4223 .set_cached_descriptor = emulator_set_cached_descriptor,
4224 .get_segment_selector = emulator_get_segment_selector,
4225 .set_segment_selector = emulator_set_segment_selector,
4226 .get_cached_segment_base = emulator_get_cached_segment_base,
4227 .get_gdt = emulator_get_gdt,
4228 .get_idt = emulator_get_idt,
4229 .get_cr = emulator_get_cr,
4230 .set_cr = emulator_set_cr,
4231 .cpl = emulator_get_cpl,
4232 .get_dr = emulator_get_dr,
4233 .set_dr = emulator_set_dr,
4234 .set_msr = kvm_set_msr,
4235 .get_msr = kvm_get_msr,
4238 static void cache_all_regs(struct kvm_vcpu *vcpu)
4240 kvm_register_read(vcpu, VCPU_REGS_RAX);
4241 kvm_register_read(vcpu, VCPU_REGS_RSP);
4242 kvm_register_read(vcpu, VCPU_REGS_RIP);
4243 vcpu->arch.regs_dirty = ~0;
4246 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4248 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4250 * an sti; sti; sequence only disable interrupts for the first
4251 * instruction. So, if the last instruction, be it emulated or
4252 * not, left the system with the INT_STI flag enabled, it
4253 * means that the last instruction is an sti. We should not
4254 * leave the flag on in this case. The same goes for mov ss
4256 if (!(int_shadow & mask))
4257 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4260 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4262 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4263 if (ctxt->exception.vector == PF_VECTOR)
4264 kvm_propagate_fault(vcpu, &ctxt->exception);
4265 else if (ctxt->exception.error_code_valid)
4266 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4267 ctxt->exception.error_code);
4269 kvm_queue_exception(vcpu, ctxt->exception.vector);
4272 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4274 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4277 cache_all_regs(vcpu);
4279 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4281 vcpu->arch.emulate_ctxt.vcpu = vcpu;
4282 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
4283 vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
4284 vcpu->arch.emulate_ctxt.mode =
4285 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4286 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
4287 ? X86EMUL_MODE_VM86 : cs_l
4288 ? X86EMUL_MODE_PROT64 : cs_db
4289 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
4290 memset(c, 0, sizeof(struct decode_cache));
4291 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4294 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq)
4296 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4299 init_emulate_ctxt(vcpu);
4301 vcpu->arch.emulate_ctxt.decode.op_bytes = 2;
4302 vcpu->arch.emulate_ctxt.decode.ad_bytes = 2;
4303 vcpu->arch.emulate_ctxt.decode.eip = vcpu->arch.emulate_ctxt.eip;
4304 ret = emulate_int_real(&vcpu->arch.emulate_ctxt, &emulate_ops, irq);
4306 if (ret != X86EMUL_CONTINUE)
4307 return EMULATE_FAIL;
4309 vcpu->arch.emulate_ctxt.eip = c->eip;
4310 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4311 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4312 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4314 if (irq == NMI_VECTOR)
4315 vcpu->arch.nmi_pending = false;
4317 vcpu->arch.interrupt.pending = false;
4319 return EMULATE_DONE;
4321 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4323 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4325 int r = EMULATE_DONE;
4327 ++vcpu->stat.insn_emulation_fail;
4328 trace_kvm_emulate_insn_failed(vcpu);
4329 if (!is_guest_mode(vcpu)) {
4330 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4331 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4332 vcpu->run->internal.ndata = 0;
4335 kvm_queue_exception(vcpu, UD_VECTOR);
4340 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4348 * if emulation was due to access to shadowed page table
4349 * and it failed try to unshadow page and re-entetr the
4350 * guest to let CPU execute the instruction.
4352 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4355 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4357 if (gpa == UNMAPPED_GVA)
4358 return true; /* let cpu generate fault */
4360 if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4366 int emulate_instruction(struct kvm_vcpu *vcpu,
4372 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4374 kvm_clear_exception_queue(vcpu);
4375 vcpu->arch.mmio_fault_cr2 = cr2;
4377 * TODO: fix emulate.c to use guest_read/write_register
4378 * instead of direct ->regs accesses, can save hundred cycles
4379 * on Intel for instructions that don't read/change RSP, for
4382 cache_all_regs(vcpu);
4384 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4385 init_emulate_ctxt(vcpu);
4386 vcpu->arch.emulate_ctxt.interruptibility = 0;
4387 vcpu->arch.emulate_ctxt.have_exception = false;
4388 vcpu->arch.emulate_ctxt.perm_ok = false;
4390 r = x86_decode_insn(&vcpu->arch.emulate_ctxt);
4391 if (r == X86EMUL_PROPAGATE_FAULT)
4394 trace_kvm_emulate_insn_start(vcpu);
4396 /* Only allow emulation of specific instructions on #UD
4397 * (namely VMMCALL, sysenter, sysexit, syscall)*/
4398 if (emulation_type & EMULTYPE_TRAP_UD) {
4400 return EMULATE_FAIL;
4402 case 0x01: /* VMMCALL */
4403 if (c->modrm_mod != 3 || c->modrm_rm != 1)
4404 return EMULATE_FAIL;
4406 case 0x34: /* sysenter */
4407 case 0x35: /* sysexit */
4408 if (c->modrm_mod != 0 || c->modrm_rm != 0)
4409 return EMULATE_FAIL;
4411 case 0x05: /* syscall */
4412 if (c->modrm_mod != 0 || c->modrm_rm != 0)
4413 return EMULATE_FAIL;
4416 return EMULATE_FAIL;
4419 if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
4420 return EMULATE_FAIL;
4423 ++vcpu->stat.insn_emulation;
4425 if (reexecute_instruction(vcpu, cr2))
4426 return EMULATE_DONE;
4427 if (emulation_type & EMULTYPE_SKIP)
4428 return EMULATE_FAIL;
4429 return handle_emulation_failure(vcpu);
4433 if (emulation_type & EMULTYPE_SKIP) {
4434 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
4435 return EMULATE_DONE;
4438 /* this is needed for vmware backdor interface to work since it
4439 changes registers values during IO operation */
4440 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4443 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt);
4445 if (r == EMULATION_FAILED) {
4446 if (reexecute_instruction(vcpu, cr2))
4447 return EMULATE_DONE;
4449 return handle_emulation_failure(vcpu);
4453 if (vcpu->arch.emulate_ctxt.have_exception) {
4454 inject_emulated_exception(vcpu);
4456 } else if (vcpu->arch.pio.count) {
4457 if (!vcpu->arch.pio.in)
4458 vcpu->arch.pio.count = 0;
4459 r = EMULATE_DO_MMIO;
4460 } else if (vcpu->mmio_needed) {
4461 if (vcpu->mmio_is_write)
4462 vcpu->mmio_needed = 0;
4463 r = EMULATE_DO_MMIO;
4464 } else if (r == EMULATION_RESTART)
4469 toggle_interruptibility(vcpu, vcpu->arch.emulate_ctxt.interruptibility);
4470 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4471 kvm_make_request(KVM_REQ_EVENT, vcpu);
4472 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4473 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4477 EXPORT_SYMBOL_GPL(emulate_instruction);
4479 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4481 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4482 int ret = emulator_pio_out_emulated(size, port, &val, 1, vcpu);
4483 /* do not return to emulator after return from userspace */
4484 vcpu->arch.pio.count = 0;
4487 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4489 static void tsc_bad(void *info)
4491 __get_cpu_var(cpu_tsc_khz) = 0;
4494 static void tsc_khz_changed(void *data)
4496 struct cpufreq_freqs *freq = data;
4497 unsigned long khz = 0;
4501 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4502 khz = cpufreq_quick_get(raw_smp_processor_id());
4505 __get_cpu_var(cpu_tsc_khz) = khz;
4508 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4511 struct cpufreq_freqs *freq = data;
4513 struct kvm_vcpu *vcpu;
4514 int i, send_ipi = 0;
4517 * We allow guests to temporarily run on slowing clocks,
4518 * provided we notify them after, or to run on accelerating
4519 * clocks, provided we notify them before. Thus time never
4522 * However, we have a problem. We can't atomically update
4523 * the frequency of a given CPU from this function; it is
4524 * merely a notifier, which can be called from any CPU.
4525 * Changing the TSC frequency at arbitrary points in time
4526 * requires a recomputation of local variables related to
4527 * the TSC for each VCPU. We must flag these local variables
4528 * to be updated and be sure the update takes place with the
4529 * new frequency before any guests proceed.
4531 * Unfortunately, the combination of hotplug CPU and frequency
4532 * change creates an intractable locking scenario; the order
4533 * of when these callouts happen is undefined with respect to
4534 * CPU hotplug, and they can race with each other. As such,
4535 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4536 * undefined; you can actually have a CPU frequency change take
4537 * place in between the computation of X and the setting of the
4538 * variable. To protect against this problem, all updates of
4539 * the per_cpu tsc_khz variable are done in an interrupt
4540 * protected IPI, and all callers wishing to update the value
4541 * must wait for a synchronous IPI to complete (which is trivial
4542 * if the caller is on the CPU already). This establishes the
4543 * necessary total order on variable updates.
4545 * Note that because a guest time update may take place
4546 * anytime after the setting of the VCPU's request bit, the
4547 * correct TSC value must be set before the request. However,
4548 * to ensure the update actually makes it to any guest which
4549 * starts running in hardware virtualization between the set
4550 * and the acquisition of the spinlock, we must also ping the
4551 * CPU after setting the request bit.
4555 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4557 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4560 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4562 spin_lock(&kvm_lock);
4563 list_for_each_entry(kvm, &vm_list, vm_list) {
4564 kvm_for_each_vcpu(i, vcpu, kvm) {
4565 if (vcpu->cpu != freq->cpu)
4567 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4568 if (vcpu->cpu != smp_processor_id())
4572 spin_unlock(&kvm_lock);
4574 if (freq->old < freq->new && send_ipi) {
4576 * We upscale the frequency. Must make the guest
4577 * doesn't see old kvmclock values while running with
4578 * the new frequency, otherwise we risk the guest sees
4579 * time go backwards.
4581 * In case we update the frequency for another cpu
4582 * (which might be in guest context) send an interrupt
4583 * to kick the cpu out of guest context. Next time
4584 * guest context is entered kvmclock will be updated,
4585 * so the guest will not see stale values.
4587 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4592 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4593 .notifier_call = kvmclock_cpufreq_notifier
4596 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4597 unsigned long action, void *hcpu)
4599 unsigned int cpu = (unsigned long)hcpu;
4603 case CPU_DOWN_FAILED:
4604 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4606 case CPU_DOWN_PREPARE:
4607 smp_call_function_single(cpu, tsc_bad, NULL, 1);
4613 static struct notifier_block kvmclock_cpu_notifier_block = {
4614 .notifier_call = kvmclock_cpu_notifier,
4615 .priority = -INT_MAX
4618 static void kvm_timer_init(void)
4622 max_tsc_khz = tsc_khz;
4623 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4624 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4625 #ifdef CONFIG_CPU_FREQ
4626 struct cpufreq_policy policy;
4627 memset(&policy, 0, sizeof(policy));
4629 cpufreq_get_policy(&policy, cpu);
4630 if (policy.cpuinfo.max_freq)
4631 max_tsc_khz = policy.cpuinfo.max_freq;
4634 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4635 CPUFREQ_TRANSITION_NOTIFIER);
4637 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
4638 for_each_online_cpu(cpu)
4639 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4642 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4644 static int kvm_is_in_guest(void)
4646 return percpu_read(current_vcpu) != NULL;
4649 static int kvm_is_user_mode(void)
4653 if (percpu_read(current_vcpu))
4654 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
4656 return user_mode != 0;
4659 static unsigned long kvm_get_guest_ip(void)
4661 unsigned long ip = 0;
4663 if (percpu_read(current_vcpu))
4664 ip = kvm_rip_read(percpu_read(current_vcpu));
4669 static struct perf_guest_info_callbacks kvm_guest_cbs = {
4670 .is_in_guest = kvm_is_in_guest,
4671 .is_user_mode = kvm_is_user_mode,
4672 .get_guest_ip = kvm_get_guest_ip,
4675 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4677 percpu_write(current_vcpu, vcpu);
4679 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4681 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4683 percpu_write(current_vcpu, NULL);
4685 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4687 int kvm_arch_init(void *opaque)
4690 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4693 printk(KERN_ERR "kvm: already loaded the other module\n");
4698 if (!ops->cpu_has_kvm_support()) {
4699 printk(KERN_ERR "kvm: no hardware support\n");
4703 if (ops->disabled_by_bios()) {
4704 printk(KERN_ERR "kvm: disabled by bios\n");
4709 r = kvm_mmu_module_init();
4713 kvm_init_msr_list();
4716 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
4717 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4718 PT_DIRTY_MASK, PT64_NX_MASK, 0);
4722 perf_register_guest_info_callbacks(&kvm_guest_cbs);
4725 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
4733 void kvm_arch_exit(void)
4735 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4737 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4738 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4739 CPUFREQ_TRANSITION_NOTIFIER);
4740 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4742 kvm_mmu_module_exit();
4745 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4747 ++vcpu->stat.halt_exits;
4748 if (irqchip_in_kernel(vcpu->kvm)) {
4749 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
4752 vcpu->run->exit_reason = KVM_EXIT_HLT;
4756 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4758 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
4761 if (is_long_mode(vcpu))
4764 return a0 | ((gpa_t)a1 << 32);
4767 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4769 u64 param, ingpa, outgpa, ret;
4770 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4771 bool fast, longmode;
4775 * hypercall generates UD from non zero cpl and real mode
4778 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
4779 kvm_queue_exception(vcpu, UD_VECTOR);
4783 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4784 longmode = is_long_mode(vcpu) && cs_l == 1;
4787 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4788 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4789 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4790 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4791 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4792 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
4794 #ifdef CONFIG_X86_64
4796 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4797 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4798 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4802 code = param & 0xffff;
4803 fast = (param >> 16) & 0x1;
4804 rep_cnt = (param >> 32) & 0xfff;
4805 rep_idx = (param >> 48) & 0xfff;
4807 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
4810 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
4811 kvm_vcpu_on_spin(vcpu);
4814 res = HV_STATUS_INVALID_HYPERCALL_CODE;
4818 ret = res | (((u64)rep_done & 0xfff) << 32);
4820 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4822 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
4823 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
4829 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
4831 unsigned long nr, a0, a1, a2, a3, ret;
4834 if (kvm_hv_hypercall_enabled(vcpu->kvm))
4835 return kvm_hv_hypercall(vcpu);
4837 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
4838 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
4839 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
4840 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
4841 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
4843 trace_kvm_hypercall(nr, a0, a1, a2, a3);
4845 if (!is_long_mode(vcpu)) {
4853 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
4859 case KVM_HC_VAPIC_POLL_IRQ:
4863 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
4870 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4871 ++vcpu->stat.hypercalls;
4874 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
4876 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
4878 char instruction[3];
4879 unsigned long rip = kvm_rip_read(vcpu);
4882 * Blow out the MMU to ensure that no other VCPU has an active mapping
4883 * to ensure that the updated hypercall appears atomically across all
4886 kvm_mmu_zap_all(vcpu->kvm);
4888 kvm_x86_ops->patch_hypercall(vcpu, instruction);
4890 return emulator_write_emulated(rip, instruction, 3, NULL, vcpu);
4893 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4895 struct desc_ptr dt = { limit, base };
4897 kvm_x86_ops->set_gdt(vcpu, &dt);
4900 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4902 struct desc_ptr dt = { limit, base };
4904 kvm_x86_ops->set_idt(vcpu, &dt);
4907 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
4909 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
4910 int j, nent = vcpu->arch.cpuid_nent;
4912 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
4913 /* when no next entry is found, the current entry[i] is reselected */
4914 for (j = i + 1; ; j = (j + 1) % nent) {
4915 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
4916 if (ej->function == e->function) {
4917 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
4921 return 0; /* silence gcc, even though control never reaches here */
4924 /* find an entry with matching function, matching index (if needed), and that
4925 * should be read next (if it's stateful) */
4926 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
4927 u32 function, u32 index)
4929 if (e->function != function)
4931 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
4933 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
4934 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
4939 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
4940 u32 function, u32 index)
4943 struct kvm_cpuid_entry2 *best = NULL;
4945 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
4946 struct kvm_cpuid_entry2 *e;
4948 e = &vcpu->arch.cpuid_entries[i];
4949 if (is_matching_cpuid_entry(e, function, index)) {
4950 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
4951 move_to_next_stateful_cpuid_entry(vcpu, i);
4956 * Both basic or both extended?
4958 if (((e->function ^ function) & 0x80000000) == 0)
4959 if (!best || e->function > best->function)
4964 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
4966 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
4968 struct kvm_cpuid_entry2 *best;
4970 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
4971 if (!best || best->eax < 0x80000008)
4973 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
4975 return best->eax & 0xff;
4980 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
4982 u32 function, index;
4983 struct kvm_cpuid_entry2 *best;
4985 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
4986 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4987 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
4988 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
4989 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
4990 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
4991 best = kvm_find_cpuid_entry(vcpu, function, index);
4993 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
4994 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
4995 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
4996 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
4998 kvm_x86_ops->skip_emulated_instruction(vcpu);
4999 trace_kvm_cpuid(function,
5000 kvm_register_read(vcpu, VCPU_REGS_RAX),
5001 kvm_register_read(vcpu, VCPU_REGS_RBX),
5002 kvm_register_read(vcpu, VCPU_REGS_RCX),
5003 kvm_register_read(vcpu, VCPU_REGS_RDX));
5005 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
5008 * Check if userspace requested an interrupt window, and that the
5009 * interrupt window is open.
5011 * No need to exit to userspace if we already have an interrupt queued.
5013 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5015 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5016 vcpu->run->request_interrupt_window &&
5017 kvm_arch_interrupt_allowed(vcpu));
5020 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5022 struct kvm_run *kvm_run = vcpu->run;
5024 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5025 kvm_run->cr8 = kvm_get_cr8(vcpu);
5026 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5027 if (irqchip_in_kernel(vcpu->kvm))
5028 kvm_run->ready_for_interrupt_injection = 1;
5030 kvm_run->ready_for_interrupt_injection =
5031 kvm_arch_interrupt_allowed(vcpu) &&
5032 !kvm_cpu_has_interrupt(vcpu) &&
5033 !kvm_event_needs_reinjection(vcpu);
5036 static void vapic_enter(struct kvm_vcpu *vcpu)
5038 struct kvm_lapic *apic = vcpu->arch.apic;
5041 if (!apic || !apic->vapic_addr)
5044 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5046 vcpu->arch.apic->vapic_page = page;
5049 static void vapic_exit(struct kvm_vcpu *vcpu)
5051 struct kvm_lapic *apic = vcpu->arch.apic;
5054 if (!apic || !apic->vapic_addr)
5057 idx = srcu_read_lock(&vcpu->kvm->srcu);
5058 kvm_release_page_dirty(apic->vapic_page);
5059 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5060 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5063 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5067 if (!kvm_x86_ops->update_cr8_intercept)
5070 if (!vcpu->arch.apic)
5073 if (!vcpu->arch.apic->vapic_addr)
5074 max_irr = kvm_lapic_find_highest_irr(vcpu);
5081 tpr = kvm_lapic_get_cr8(vcpu);
5083 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5086 static void inject_pending_event(struct kvm_vcpu *vcpu)
5088 /* try to reinject previous events if any */
5089 if (vcpu->arch.exception.pending) {
5090 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5091 vcpu->arch.exception.has_error_code,
5092 vcpu->arch.exception.error_code);
5093 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5094 vcpu->arch.exception.has_error_code,
5095 vcpu->arch.exception.error_code,
5096 vcpu->arch.exception.reinject);
5100 if (vcpu->arch.nmi_injected) {
5101 kvm_x86_ops->set_nmi(vcpu);
5105 if (vcpu->arch.interrupt.pending) {
5106 kvm_x86_ops->set_irq(vcpu);
5110 /* try to inject new event if pending */
5111 if (vcpu->arch.nmi_pending) {
5112 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5113 vcpu->arch.nmi_pending = false;
5114 vcpu->arch.nmi_injected = true;
5115 kvm_x86_ops->set_nmi(vcpu);
5117 } else if (kvm_cpu_has_interrupt(vcpu)) {
5118 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5119 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5121 kvm_x86_ops->set_irq(vcpu);
5126 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5128 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5129 !vcpu->guest_xcr0_loaded) {
5130 /* kvm_set_xcr() also depends on this */
5131 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5132 vcpu->guest_xcr0_loaded = 1;
5136 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5138 if (vcpu->guest_xcr0_loaded) {
5139 if (vcpu->arch.xcr0 != host_xcr0)
5140 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5141 vcpu->guest_xcr0_loaded = 0;
5145 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5148 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5149 vcpu->run->request_interrupt_window;
5151 if (vcpu->requests) {
5152 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5153 kvm_mmu_unload(vcpu);
5154 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5155 __kvm_migrate_timers(vcpu);
5156 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5157 r = kvm_guest_time_update(vcpu);
5161 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5162 kvm_mmu_sync_roots(vcpu);
5163 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5164 kvm_x86_ops->tlb_flush(vcpu);
5165 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5166 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5170 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5171 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5175 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5176 vcpu->fpu_active = 0;
5177 kvm_x86_ops->fpu_deactivate(vcpu);
5179 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5180 /* Page is swapped out. Do synthetic halt */
5181 vcpu->arch.apf.halted = true;
5187 r = kvm_mmu_reload(vcpu);
5191 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5192 inject_pending_event(vcpu);
5194 /* enable NMI/IRQ window open exits if needed */
5195 if (vcpu->arch.nmi_pending)
5196 kvm_x86_ops->enable_nmi_window(vcpu);
5197 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5198 kvm_x86_ops->enable_irq_window(vcpu);
5200 if (kvm_lapic_enabled(vcpu)) {
5201 update_cr8_intercept(vcpu);
5202 kvm_lapic_sync_to_vapic(vcpu);
5208 kvm_x86_ops->prepare_guest_switch(vcpu);
5209 if (vcpu->fpu_active)
5210 kvm_load_guest_fpu(vcpu);
5211 kvm_load_guest_xcr0(vcpu);
5213 atomic_set(&vcpu->guest_mode, 1);
5216 local_irq_disable();
5218 if (!atomic_read(&vcpu->guest_mode) || vcpu->requests
5219 || need_resched() || signal_pending(current)) {
5220 atomic_set(&vcpu->guest_mode, 0);
5224 kvm_x86_ops->cancel_injection(vcpu);
5229 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5233 if (unlikely(vcpu->arch.switch_db_regs)) {
5235 set_debugreg(vcpu->arch.eff_db[0], 0);
5236 set_debugreg(vcpu->arch.eff_db[1], 1);
5237 set_debugreg(vcpu->arch.eff_db[2], 2);
5238 set_debugreg(vcpu->arch.eff_db[3], 3);
5241 trace_kvm_entry(vcpu->vcpu_id);
5242 kvm_x86_ops->run(vcpu);
5245 * If the guest has used debug registers, at least dr7
5246 * will be disabled while returning to the host.
5247 * If we don't have active breakpoints in the host, we don't
5248 * care about the messed up debug address registers. But if
5249 * we have some of them active, restore the old state.
5251 if (hw_breakpoint_active())
5252 hw_breakpoint_restore();
5254 kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
5256 atomic_set(&vcpu->guest_mode, 0);
5263 * We must have an instruction between local_irq_enable() and
5264 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5265 * the interrupt shadow. The stat.exits increment will do nicely.
5266 * But we need to prevent reordering, hence this barrier():
5274 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5277 * Profile KVM exit RIPs:
5279 if (unlikely(prof_on == KVM_PROFILING)) {
5280 unsigned long rip = kvm_rip_read(vcpu);
5281 profile_hit(KVM_PROFILING, (void *)rip);
5285 kvm_lapic_sync_from_vapic(vcpu);
5287 r = kvm_x86_ops->handle_exit(vcpu);
5293 static int __vcpu_run(struct kvm_vcpu *vcpu)
5296 struct kvm *kvm = vcpu->kvm;
5298 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5299 pr_debug("vcpu %d received sipi with vector # %x\n",
5300 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5301 kvm_lapic_reset(vcpu);
5302 r = kvm_arch_vcpu_reset(vcpu);
5305 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5308 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5313 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5314 !vcpu->arch.apf.halted)
5315 r = vcpu_enter_guest(vcpu);
5317 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5318 kvm_vcpu_block(vcpu);
5319 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5320 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5322 switch(vcpu->arch.mp_state) {
5323 case KVM_MP_STATE_HALTED:
5324 vcpu->arch.mp_state =
5325 KVM_MP_STATE_RUNNABLE;
5326 case KVM_MP_STATE_RUNNABLE:
5327 vcpu->arch.apf.halted = false;
5329 case KVM_MP_STATE_SIPI_RECEIVED:
5340 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5341 if (kvm_cpu_has_pending_timer(vcpu))
5342 kvm_inject_pending_timer_irqs(vcpu);
5344 if (dm_request_for_irq_injection(vcpu)) {
5346 vcpu->run->exit_reason = KVM_EXIT_INTR;
5347 ++vcpu->stat.request_irq_exits;
5350 kvm_check_async_pf_completion(vcpu);
5352 if (signal_pending(current)) {
5354 vcpu->run->exit_reason = KVM_EXIT_INTR;
5355 ++vcpu->stat.signal_exits;
5357 if (need_resched()) {
5358 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5360 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5364 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5371 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5376 if (vcpu->sigset_active)
5377 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5379 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5380 kvm_vcpu_block(vcpu);
5381 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5386 /* re-sync apic's tpr */
5387 if (!irqchip_in_kernel(vcpu->kvm)) {
5388 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5394 if (vcpu->arch.pio.count || vcpu->mmio_needed) {
5395 if (vcpu->mmio_needed) {
5396 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
5397 vcpu->mmio_read_completed = 1;
5398 vcpu->mmio_needed = 0;
5400 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5401 r = emulate_instruction(vcpu, 0, 0, EMULTYPE_NO_DECODE);
5402 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5403 if (r != EMULATE_DONE) {
5408 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5409 kvm_register_write(vcpu, VCPU_REGS_RAX,
5410 kvm_run->hypercall.ret);
5412 r = __vcpu_run(vcpu);
5415 post_kvm_run_save(vcpu);
5416 if (vcpu->sigset_active)
5417 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5422 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5424 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5425 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5426 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5427 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5428 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5429 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5430 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5431 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5432 #ifdef CONFIG_X86_64
5433 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5434 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5435 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5436 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5437 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5438 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5439 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5440 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5443 regs->rip = kvm_rip_read(vcpu);
5444 regs->rflags = kvm_get_rflags(vcpu);
5449 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5451 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5452 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5453 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5454 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5455 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5456 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5457 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5458 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5459 #ifdef CONFIG_X86_64
5460 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5461 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5462 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5463 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5464 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5465 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5466 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5467 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5470 kvm_rip_write(vcpu, regs->rip);
5471 kvm_set_rflags(vcpu, regs->rflags);
5473 vcpu->arch.exception.pending = false;
5475 kvm_make_request(KVM_REQ_EVENT, vcpu);
5480 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5482 struct kvm_segment cs;
5484 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
5488 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5490 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5491 struct kvm_sregs *sregs)
5495 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5496 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5497 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5498 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5499 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5500 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5502 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5503 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5505 kvm_x86_ops->get_idt(vcpu, &dt);
5506 sregs->idt.limit = dt.size;
5507 sregs->idt.base = dt.address;
5508 kvm_x86_ops->get_gdt(vcpu, &dt);
5509 sregs->gdt.limit = dt.size;
5510 sregs->gdt.base = dt.address;
5512 sregs->cr0 = kvm_read_cr0(vcpu);
5513 sregs->cr2 = vcpu->arch.cr2;
5514 sregs->cr3 = vcpu->arch.cr3;
5515 sregs->cr4 = kvm_read_cr4(vcpu);
5516 sregs->cr8 = kvm_get_cr8(vcpu);
5517 sregs->efer = vcpu->arch.efer;
5518 sregs->apic_base = kvm_get_apic_base(vcpu);
5520 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
5522 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
5523 set_bit(vcpu->arch.interrupt.nr,
5524 (unsigned long *)sregs->interrupt_bitmap);
5529 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
5530 struct kvm_mp_state *mp_state)
5532 mp_state->mp_state = vcpu->arch.mp_state;
5536 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
5537 struct kvm_mp_state *mp_state)
5539 vcpu->arch.mp_state = mp_state->mp_state;
5540 kvm_make_request(KVM_REQ_EVENT, vcpu);
5544 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
5545 bool has_error_code, u32 error_code)
5547 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5550 init_emulate_ctxt(vcpu);
5552 ret = emulator_task_switch(&vcpu->arch.emulate_ctxt,
5553 tss_selector, reason, has_error_code,
5557 return EMULATE_FAIL;
5559 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5560 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
5561 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
5562 kvm_make_request(KVM_REQ_EVENT, vcpu);
5563 return EMULATE_DONE;
5565 EXPORT_SYMBOL_GPL(kvm_task_switch);
5567 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5568 struct kvm_sregs *sregs)
5570 int mmu_reset_needed = 0;
5571 int pending_vec, max_bits;
5574 dt.size = sregs->idt.limit;
5575 dt.address = sregs->idt.base;
5576 kvm_x86_ops->set_idt(vcpu, &dt);
5577 dt.size = sregs->gdt.limit;
5578 dt.address = sregs->gdt.base;
5579 kvm_x86_ops->set_gdt(vcpu, &dt);
5581 vcpu->arch.cr2 = sregs->cr2;
5582 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
5583 vcpu->arch.cr3 = sregs->cr3;
5585 kvm_set_cr8(vcpu, sregs->cr8);
5587 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5588 kvm_x86_ops->set_efer(vcpu, sregs->efer);
5589 kvm_set_apic_base(vcpu, sregs->apic_base);
5591 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5592 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5593 vcpu->arch.cr0 = sregs->cr0;
5595 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5596 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5597 if (sregs->cr4 & X86_CR4_OSXSAVE)
5599 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5600 load_pdptrs(vcpu, vcpu->arch.walk_mmu, vcpu->arch.cr3);
5601 mmu_reset_needed = 1;
5604 if (mmu_reset_needed)
5605 kvm_mmu_reset_context(vcpu);
5607 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5608 pending_vec = find_first_bit(
5609 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5610 if (pending_vec < max_bits) {
5611 kvm_queue_interrupt(vcpu, pending_vec, false);
5612 pr_debug("Set back pending irq %d\n", pending_vec);
5613 if (irqchip_in_kernel(vcpu->kvm))
5614 kvm_pic_clear_isr_ack(vcpu->kvm);
5617 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5618 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5619 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5620 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5621 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5622 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5624 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5625 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5627 update_cr8_intercept(vcpu);
5629 /* Older userspace won't unhalt the vcpu on reset. */
5630 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5631 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5633 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5635 kvm_make_request(KVM_REQ_EVENT, vcpu);
5640 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5641 struct kvm_guest_debug *dbg)
5643 unsigned long rflags;
5646 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5648 if (vcpu->arch.exception.pending)
5650 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5651 kvm_queue_exception(vcpu, DB_VECTOR);
5653 kvm_queue_exception(vcpu, BP_VECTOR);
5657 * Read rflags as long as potentially injected trace flags are still
5660 rflags = kvm_get_rflags(vcpu);
5662 vcpu->guest_debug = dbg->control;
5663 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5664 vcpu->guest_debug = 0;
5666 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5667 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5668 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5669 vcpu->arch.switch_db_regs =
5670 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5672 for (i = 0; i < KVM_NR_DB_REGS; i++)
5673 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5674 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5677 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5678 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5679 get_segment_base(vcpu, VCPU_SREG_CS);
5682 * Trigger an rflags update that will inject or remove the trace
5685 kvm_set_rflags(vcpu, rflags);
5687 kvm_x86_ops->set_guest_debug(vcpu, dbg);
5697 * Translate a guest virtual address to a guest physical address.
5699 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5700 struct kvm_translation *tr)
5702 unsigned long vaddr = tr->linear_address;
5706 idx = srcu_read_lock(&vcpu->kvm->srcu);
5707 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5708 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5709 tr->physical_address = gpa;
5710 tr->valid = gpa != UNMAPPED_GVA;
5717 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5719 struct i387_fxsave_struct *fxsave =
5720 &vcpu->arch.guest_fpu.state->fxsave;
5722 memcpy(fpu->fpr, fxsave->st_space, 128);
5723 fpu->fcw = fxsave->cwd;
5724 fpu->fsw = fxsave->swd;
5725 fpu->ftwx = fxsave->twd;
5726 fpu->last_opcode = fxsave->fop;
5727 fpu->last_ip = fxsave->rip;
5728 fpu->last_dp = fxsave->rdp;
5729 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5734 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5736 struct i387_fxsave_struct *fxsave =
5737 &vcpu->arch.guest_fpu.state->fxsave;
5739 memcpy(fxsave->st_space, fpu->fpr, 128);
5740 fxsave->cwd = fpu->fcw;
5741 fxsave->swd = fpu->fsw;
5742 fxsave->twd = fpu->ftwx;
5743 fxsave->fop = fpu->last_opcode;
5744 fxsave->rip = fpu->last_ip;
5745 fxsave->rdp = fpu->last_dp;
5746 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5751 int fx_init(struct kvm_vcpu *vcpu)
5755 err = fpu_alloc(&vcpu->arch.guest_fpu);
5759 fpu_finit(&vcpu->arch.guest_fpu);
5762 * Ensure guest xcr0 is valid for loading
5764 vcpu->arch.xcr0 = XSTATE_FP;
5766 vcpu->arch.cr0 |= X86_CR0_ET;
5770 EXPORT_SYMBOL_GPL(fx_init);
5772 static void fx_free(struct kvm_vcpu *vcpu)
5774 fpu_free(&vcpu->arch.guest_fpu);
5777 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5779 if (vcpu->guest_fpu_loaded)
5783 * Restore all possible states in the guest,
5784 * and assume host would use all available bits.
5785 * Guest xcr0 would be loaded later.
5787 kvm_put_guest_xcr0(vcpu);
5788 vcpu->guest_fpu_loaded = 1;
5789 unlazy_fpu(current);
5790 fpu_restore_checking(&vcpu->arch.guest_fpu);
5794 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5796 kvm_put_guest_xcr0(vcpu);
5798 if (!vcpu->guest_fpu_loaded)
5801 vcpu->guest_fpu_loaded = 0;
5802 fpu_save_init(&vcpu->arch.guest_fpu);
5803 ++vcpu->stat.fpu_reload;
5804 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
5808 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5810 if (vcpu->arch.time_page) {
5811 kvm_release_page_dirty(vcpu->arch.time_page);
5812 vcpu->arch.time_page = NULL;
5815 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
5817 kvm_x86_ops->vcpu_free(vcpu);
5820 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5823 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
5824 printk_once(KERN_WARNING
5825 "kvm: SMP vm created on host with unstable TSC; "
5826 "guest TSC will not be reliable\n");
5827 return kvm_x86_ops->vcpu_create(kvm, id);
5830 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5834 vcpu->arch.mtrr_state.have_fixed = 1;
5836 r = kvm_arch_vcpu_reset(vcpu);
5838 r = kvm_mmu_setup(vcpu);
5845 kvm_x86_ops->vcpu_free(vcpu);
5849 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5851 vcpu->arch.apf.msr_val = 0;
5854 kvm_mmu_unload(vcpu);
5858 kvm_x86_ops->vcpu_free(vcpu);
5861 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5863 vcpu->arch.nmi_pending = false;
5864 vcpu->arch.nmi_injected = false;
5866 vcpu->arch.switch_db_regs = 0;
5867 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5868 vcpu->arch.dr6 = DR6_FIXED_1;
5869 vcpu->arch.dr7 = DR7_FIXED_1;
5871 kvm_make_request(KVM_REQ_EVENT, vcpu);
5872 vcpu->arch.apf.msr_val = 0;
5874 kvm_clear_async_pf_completion_queue(vcpu);
5875 kvm_async_pf_hash_reset(vcpu);
5876 vcpu->arch.apf.halted = false;
5878 return kvm_x86_ops->vcpu_reset(vcpu);
5881 int kvm_arch_hardware_enable(void *garbage)
5884 struct kvm_vcpu *vcpu;
5887 kvm_shared_msr_cpu_online();
5888 list_for_each_entry(kvm, &vm_list, vm_list)
5889 kvm_for_each_vcpu(i, vcpu, kvm)
5890 if (vcpu->cpu == smp_processor_id())
5891 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5892 return kvm_x86_ops->hardware_enable(garbage);
5895 void kvm_arch_hardware_disable(void *garbage)
5897 kvm_x86_ops->hardware_disable(garbage);
5898 drop_user_return_notifiers(garbage);
5901 int kvm_arch_hardware_setup(void)
5903 return kvm_x86_ops->hardware_setup();
5906 void kvm_arch_hardware_unsetup(void)
5908 kvm_x86_ops->hardware_unsetup();
5911 void kvm_arch_check_processor_compat(void *rtn)
5913 kvm_x86_ops->check_processor_compatibility(rtn);
5916 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5922 BUG_ON(vcpu->kvm == NULL);
5925 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
5926 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
5927 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5928 vcpu->arch.mmu.translate_gpa = translate_gpa;
5929 vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
5930 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5931 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5933 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5935 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5940 vcpu->arch.pio_data = page_address(page);
5942 if (!kvm->arch.virtual_tsc_khz)
5943 kvm_arch_set_tsc_khz(kvm, max_tsc_khz);
5945 r = kvm_mmu_create(vcpu);
5947 goto fail_free_pio_data;
5949 if (irqchip_in_kernel(kvm)) {
5950 r = kvm_create_lapic(vcpu);
5952 goto fail_mmu_destroy;
5955 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5957 if (!vcpu->arch.mce_banks) {
5959 goto fail_free_lapic;
5961 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5963 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
5964 goto fail_free_mce_banks;
5966 kvm_async_pf_hash_reset(vcpu);
5969 fail_free_mce_banks:
5970 kfree(vcpu->arch.mce_banks);
5972 kvm_free_lapic(vcpu);
5974 kvm_mmu_destroy(vcpu);
5976 free_page((unsigned long)vcpu->arch.pio_data);
5981 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5985 kfree(vcpu->arch.mce_banks);
5986 kvm_free_lapic(vcpu);
5987 idx = srcu_read_lock(&vcpu->kvm->srcu);
5988 kvm_mmu_destroy(vcpu);
5989 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5990 free_page((unsigned long)vcpu->arch.pio_data);
5993 int kvm_arch_init_vm(struct kvm *kvm)
5995 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5996 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5998 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5999 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
6001 spin_lock_init(&kvm->arch.tsc_write_lock);
6006 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
6009 kvm_mmu_unload(vcpu);
6013 static void kvm_free_vcpus(struct kvm *kvm)
6016 struct kvm_vcpu *vcpu;
6019 * Unpin any mmu pages first.
6021 kvm_for_each_vcpu(i, vcpu, kvm) {
6022 kvm_clear_async_pf_completion_queue(vcpu);
6023 kvm_unload_vcpu_mmu(vcpu);
6025 kvm_for_each_vcpu(i, vcpu, kvm)
6026 kvm_arch_vcpu_free(vcpu);
6028 mutex_lock(&kvm->lock);
6029 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6030 kvm->vcpus[i] = NULL;
6032 atomic_set(&kvm->online_vcpus, 0);
6033 mutex_unlock(&kvm->lock);
6036 void kvm_arch_sync_events(struct kvm *kvm)
6038 kvm_free_all_assigned_devices(kvm);
6042 void kvm_arch_destroy_vm(struct kvm *kvm)
6044 kvm_iommu_unmap_guest(kvm);
6045 kfree(kvm->arch.vpic);
6046 kfree(kvm->arch.vioapic);
6047 kvm_free_vcpus(kvm);
6048 if (kvm->arch.apic_access_page)
6049 put_page(kvm->arch.apic_access_page);
6050 if (kvm->arch.ept_identity_pagetable)
6051 put_page(kvm->arch.ept_identity_pagetable);
6054 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6055 struct kvm_memory_slot *memslot,
6056 struct kvm_memory_slot old,
6057 struct kvm_userspace_memory_region *mem,
6060 int npages = memslot->npages;
6061 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6063 /* Prevent internal slot pages from being moved by fork()/COW. */
6064 if (memslot->id >= KVM_MEMORY_SLOTS)
6065 map_flags = MAP_SHARED | MAP_ANONYMOUS;
6067 /*To keep backward compatibility with older userspace,
6068 *x86 needs to hanlde !user_alloc case.
6071 if (npages && !old.rmap) {
6072 unsigned long userspace_addr;
6074 down_write(¤t->mm->mmap_sem);
6075 userspace_addr = do_mmap(NULL, 0,
6077 PROT_READ | PROT_WRITE,
6080 up_write(¤t->mm->mmap_sem);
6082 if (IS_ERR((void *)userspace_addr))
6083 return PTR_ERR((void *)userspace_addr);
6085 memslot->userspace_addr = userspace_addr;
6093 void kvm_arch_commit_memory_region(struct kvm *kvm,
6094 struct kvm_userspace_memory_region *mem,
6095 struct kvm_memory_slot old,
6099 int npages = mem->memory_size >> PAGE_SHIFT;
6101 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6104 down_write(¤t->mm->mmap_sem);
6105 ret = do_munmap(current->mm, old.userspace_addr,
6106 old.npages * PAGE_SIZE);
6107 up_write(¤t->mm->mmap_sem);
6110 "kvm_vm_ioctl_set_memory_region: "
6111 "failed to munmap memory\n");
6114 spin_lock(&kvm->mmu_lock);
6115 if (!kvm->arch.n_requested_mmu_pages) {
6116 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6117 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6120 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6121 spin_unlock(&kvm->mmu_lock);
6124 void kvm_arch_flush_shadow(struct kvm *kvm)
6126 kvm_mmu_zap_all(kvm);
6127 kvm_reload_remote_mmus(kvm);
6130 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6132 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6133 !vcpu->arch.apf.halted)
6134 || !list_empty_careful(&vcpu->async_pf.done)
6135 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6136 || vcpu->arch.nmi_pending ||
6137 (kvm_arch_interrupt_allowed(vcpu) &&
6138 kvm_cpu_has_interrupt(vcpu));
6141 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6144 int cpu = vcpu->cpu;
6146 if (waitqueue_active(&vcpu->wq)) {
6147 wake_up_interruptible(&vcpu->wq);
6148 ++vcpu->stat.halt_wakeup;
6152 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6153 if (atomic_xchg(&vcpu->guest_mode, 0))
6154 smp_send_reschedule(cpu);
6158 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6160 return kvm_x86_ops->interrupt_allowed(vcpu);
6163 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6165 unsigned long current_rip = kvm_rip_read(vcpu) +
6166 get_segment_base(vcpu, VCPU_SREG_CS);
6168 return current_rip == linear_rip;
6170 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6172 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6174 unsigned long rflags;
6176 rflags = kvm_x86_ops->get_rflags(vcpu);
6177 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6178 rflags &= ~X86_EFLAGS_TF;
6181 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6183 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6185 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6186 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6187 rflags |= X86_EFLAGS_TF;
6188 kvm_x86_ops->set_rflags(vcpu, rflags);
6189 kvm_make_request(KVM_REQ_EVENT, vcpu);
6191 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6193 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6197 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
6198 is_error_page(work->page))
6201 r = kvm_mmu_reload(vcpu);
6205 if (!vcpu->arch.mmu.direct_map &&
6206 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
6209 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6212 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6214 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6217 static inline u32 kvm_async_pf_next_probe(u32 key)
6219 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6222 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6224 u32 key = kvm_async_pf_hash_fn(gfn);
6226 while (vcpu->arch.apf.gfns[key] != ~0)
6227 key = kvm_async_pf_next_probe(key);
6229 vcpu->arch.apf.gfns[key] = gfn;
6232 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6235 u32 key = kvm_async_pf_hash_fn(gfn);
6237 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6238 (vcpu->arch.apf.gfns[key] != gfn &&
6239 vcpu->arch.apf.gfns[key] != ~0); i++)
6240 key = kvm_async_pf_next_probe(key);
6245 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6247 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6250 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6254 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6256 vcpu->arch.apf.gfns[i] = ~0;
6258 j = kvm_async_pf_next_probe(j);
6259 if (vcpu->arch.apf.gfns[j] == ~0)
6261 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6263 * k lies cyclically in ]i,j]
6265 * |....j i.k.| or |.k..j i...|
6267 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6268 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6273 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6276 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6280 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6281 struct kvm_async_pf *work)
6283 struct x86_exception fault;
6285 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6286 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6288 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6289 (vcpu->arch.apf.send_user_only &&
6290 kvm_x86_ops->get_cpl(vcpu) == 0))
6291 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6292 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6293 fault.vector = PF_VECTOR;
6294 fault.error_code_valid = true;
6295 fault.error_code = 0;
6296 fault.nested_page_fault = false;
6297 fault.address = work->arch.token;
6298 kvm_inject_page_fault(vcpu, &fault);
6302 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6303 struct kvm_async_pf *work)
6305 struct x86_exception fault;
6307 trace_kvm_async_pf_ready(work->arch.token, work->gva);
6308 if (is_error_page(work->page))
6309 work->arch.token = ~0; /* broadcast wakeup */
6311 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6313 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6314 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6315 fault.vector = PF_VECTOR;
6316 fault.error_code_valid = true;
6317 fault.error_code = 0;
6318 fault.nested_page_fault = false;
6319 fault.address = work->arch.token;
6320 kvm_inject_page_fault(vcpu, &fault);
6322 vcpu->arch.apf.halted = false;
6325 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6327 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6330 return !kvm_event_needs_reinjection(vcpu) &&
6331 kvm_x86_ops->interrupt_allowed(vcpu);
6334 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6335 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6336 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6337 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6338 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6339 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6340 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6341 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6342 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6343 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6344 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6345 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);