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 <trace/events/kvm.h>
48 #define CREATE_TRACE_POINTS
51 #include <asm/debugreg.h>
58 #include <asm/pvclock.h>
59 #include <asm/div64.h>
61 #define MAX_IO_MSRS 256
62 #define CR0_RESERVED_BITS \
63 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
64 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
65 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
66 #define CR4_RESERVED_BITS \
67 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
68 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
69 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
71 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
73 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
75 #define KVM_MAX_MCE_BANKS 32
76 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
79 * - enable syscall per default because its emulated by KVM
80 * - enable LME and LMA per default on 64 bit KVM
83 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
85 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
88 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
89 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
91 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
92 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
93 struct kvm_cpuid_entry2 __user *entries);
95 struct kvm_x86_ops *kvm_x86_ops;
96 EXPORT_SYMBOL_GPL(kvm_x86_ops);
99 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
101 #define KVM_NR_SHARED_MSRS 16
103 struct kvm_shared_msrs_global {
105 u32 msrs[KVM_NR_SHARED_MSRS];
108 struct kvm_shared_msrs {
109 struct user_return_notifier urn;
111 struct kvm_shared_msr_values {
114 } values[KVM_NR_SHARED_MSRS];
117 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
118 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
120 struct kvm_stats_debugfs_item debugfs_entries[] = {
121 { "pf_fixed", VCPU_STAT(pf_fixed) },
122 { "pf_guest", VCPU_STAT(pf_guest) },
123 { "tlb_flush", VCPU_STAT(tlb_flush) },
124 { "invlpg", VCPU_STAT(invlpg) },
125 { "exits", VCPU_STAT(exits) },
126 { "io_exits", VCPU_STAT(io_exits) },
127 { "mmio_exits", VCPU_STAT(mmio_exits) },
128 { "signal_exits", VCPU_STAT(signal_exits) },
129 { "irq_window", VCPU_STAT(irq_window_exits) },
130 { "nmi_window", VCPU_STAT(nmi_window_exits) },
131 { "halt_exits", VCPU_STAT(halt_exits) },
132 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
133 { "hypercalls", VCPU_STAT(hypercalls) },
134 { "request_irq", VCPU_STAT(request_irq_exits) },
135 { "irq_exits", VCPU_STAT(irq_exits) },
136 { "host_state_reload", VCPU_STAT(host_state_reload) },
137 { "efer_reload", VCPU_STAT(efer_reload) },
138 { "fpu_reload", VCPU_STAT(fpu_reload) },
139 { "insn_emulation", VCPU_STAT(insn_emulation) },
140 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
141 { "irq_injections", VCPU_STAT(irq_injections) },
142 { "nmi_injections", VCPU_STAT(nmi_injections) },
143 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
144 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
145 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
146 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
147 { "mmu_flooded", VM_STAT(mmu_flooded) },
148 { "mmu_recycled", VM_STAT(mmu_recycled) },
149 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
150 { "mmu_unsync", VM_STAT(mmu_unsync) },
151 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
152 { "largepages", VM_STAT(lpages) },
156 u64 __read_mostly host_xcr0;
158 static inline u32 bit(int bitno)
160 return 1 << (bitno & 31);
163 static void kvm_on_user_return(struct user_return_notifier *urn)
166 struct kvm_shared_msrs *locals
167 = container_of(urn, struct kvm_shared_msrs, urn);
168 struct kvm_shared_msr_values *values;
170 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
171 values = &locals->values[slot];
172 if (values->host != values->curr) {
173 wrmsrl(shared_msrs_global.msrs[slot], values->host);
174 values->curr = values->host;
177 locals->registered = false;
178 user_return_notifier_unregister(urn);
181 static void shared_msr_update(unsigned slot, u32 msr)
183 struct kvm_shared_msrs *smsr;
186 smsr = &__get_cpu_var(shared_msrs);
187 /* only read, and nobody should modify it at this time,
188 * so don't need lock */
189 if (slot >= shared_msrs_global.nr) {
190 printk(KERN_ERR "kvm: invalid MSR slot!");
193 rdmsrl_safe(msr, &value);
194 smsr->values[slot].host = value;
195 smsr->values[slot].curr = value;
198 void kvm_define_shared_msr(unsigned slot, u32 msr)
200 if (slot >= shared_msrs_global.nr)
201 shared_msrs_global.nr = slot + 1;
202 shared_msrs_global.msrs[slot] = msr;
203 /* we need ensured the shared_msr_global have been updated */
206 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
208 static void kvm_shared_msr_cpu_online(void)
212 for (i = 0; i < shared_msrs_global.nr; ++i)
213 shared_msr_update(i, shared_msrs_global.msrs[i]);
216 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
218 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
220 if (((value ^ smsr->values[slot].curr) & mask) == 0)
222 smsr->values[slot].curr = value;
223 wrmsrl(shared_msrs_global.msrs[slot], value);
224 if (!smsr->registered) {
225 smsr->urn.on_user_return = kvm_on_user_return;
226 user_return_notifier_register(&smsr->urn);
227 smsr->registered = true;
230 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
232 static void drop_user_return_notifiers(void *ignore)
234 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
236 if (smsr->registered)
237 kvm_on_user_return(&smsr->urn);
240 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
242 if (irqchip_in_kernel(vcpu->kvm))
243 return vcpu->arch.apic_base;
245 return vcpu->arch.apic_base;
247 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
249 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
251 /* TODO: reserve bits check */
252 if (irqchip_in_kernel(vcpu->kvm))
253 kvm_lapic_set_base(vcpu, data);
255 vcpu->arch.apic_base = data;
257 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
259 #define EXCPT_BENIGN 0
260 #define EXCPT_CONTRIBUTORY 1
263 static int exception_class(int vector)
273 return EXCPT_CONTRIBUTORY;
280 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
281 unsigned nr, bool has_error, u32 error_code,
287 kvm_make_request(KVM_REQ_EVENT, vcpu);
289 if (!vcpu->arch.exception.pending) {
291 vcpu->arch.exception.pending = true;
292 vcpu->arch.exception.has_error_code = has_error;
293 vcpu->arch.exception.nr = nr;
294 vcpu->arch.exception.error_code = error_code;
295 vcpu->arch.exception.reinject = reinject;
299 /* to check exception */
300 prev_nr = vcpu->arch.exception.nr;
301 if (prev_nr == DF_VECTOR) {
302 /* triple fault -> shutdown */
303 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
306 class1 = exception_class(prev_nr);
307 class2 = exception_class(nr);
308 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
309 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
310 /* generate double fault per SDM Table 5-5 */
311 vcpu->arch.exception.pending = true;
312 vcpu->arch.exception.has_error_code = true;
313 vcpu->arch.exception.nr = DF_VECTOR;
314 vcpu->arch.exception.error_code = 0;
316 /* replace previous exception with a new one in a hope
317 that instruction re-execution will regenerate lost
322 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
324 kvm_multiple_exception(vcpu, nr, false, 0, false);
326 EXPORT_SYMBOL_GPL(kvm_queue_exception);
328 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
330 kvm_multiple_exception(vcpu, nr, false, 0, true);
332 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
334 void kvm_inject_page_fault(struct kvm_vcpu *vcpu)
336 unsigned error_code = vcpu->arch.fault.error_code;
338 ++vcpu->stat.pf_guest;
339 vcpu->arch.cr2 = vcpu->arch.fault.address;
340 kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
343 void kvm_propagate_fault(struct kvm_vcpu *vcpu)
345 if (mmu_is_nested(vcpu) && !vcpu->arch.fault.nested)
346 vcpu->arch.nested_mmu.inject_page_fault(vcpu);
348 vcpu->arch.mmu.inject_page_fault(vcpu);
350 vcpu->arch.fault.nested = false;
353 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
355 kvm_make_request(KVM_REQ_EVENT, vcpu);
356 vcpu->arch.nmi_pending = 1;
358 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
360 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
362 kvm_multiple_exception(vcpu, nr, true, error_code, false);
364 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
366 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
368 kvm_multiple_exception(vcpu, nr, true, error_code, true);
370 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
373 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
374 * a #GP and return false.
376 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
378 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
380 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
383 EXPORT_SYMBOL_GPL(kvm_require_cpl);
386 * This function will be used to read from the physical memory of the currently
387 * running guest. The difference to kvm_read_guest_page is that this function
388 * can read from guest physical or from the guest's guest physical memory.
390 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
391 gfn_t ngfn, void *data, int offset, int len,
397 ngpa = gfn_to_gpa(ngfn);
398 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
399 if (real_gfn == UNMAPPED_GVA)
402 real_gfn = gpa_to_gfn(real_gfn);
404 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
406 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
408 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
409 void *data, int offset, int len, u32 access)
411 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
412 data, offset, len, access);
416 * Load the pae pdptrs. Return true is they are all valid.
418 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
420 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
421 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
424 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
426 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
427 offset * sizeof(u64), sizeof(pdpte),
428 PFERR_USER_MASK|PFERR_WRITE_MASK);
433 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
434 if (is_present_gpte(pdpte[i]) &&
435 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
442 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
443 __set_bit(VCPU_EXREG_PDPTR,
444 (unsigned long *)&vcpu->arch.regs_avail);
445 __set_bit(VCPU_EXREG_PDPTR,
446 (unsigned long *)&vcpu->arch.regs_dirty);
451 EXPORT_SYMBOL_GPL(load_pdptrs);
453 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
455 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
461 if (is_long_mode(vcpu) || !is_pae(vcpu))
464 if (!test_bit(VCPU_EXREG_PDPTR,
465 (unsigned long *)&vcpu->arch.regs_avail))
468 gfn = (vcpu->arch.cr3 & ~31u) >> PAGE_SHIFT;
469 offset = (vcpu->arch.cr3 & ~31u) & (PAGE_SIZE - 1);
470 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
471 PFERR_USER_MASK | PFERR_WRITE_MASK);
474 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
480 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
482 unsigned long old_cr0 = kvm_read_cr0(vcpu);
483 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
484 X86_CR0_CD | X86_CR0_NW;
489 if (cr0 & 0xffffffff00000000UL)
493 cr0 &= ~CR0_RESERVED_BITS;
495 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
498 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
501 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
503 if ((vcpu->arch.efer & EFER_LME)) {
508 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
513 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
518 kvm_x86_ops->set_cr0(vcpu, cr0);
520 if ((cr0 ^ old_cr0) & update_bits)
521 kvm_mmu_reset_context(vcpu);
524 EXPORT_SYMBOL_GPL(kvm_set_cr0);
526 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
528 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
530 EXPORT_SYMBOL_GPL(kvm_lmsw);
532 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
536 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
537 if (index != XCR_XFEATURE_ENABLED_MASK)
540 if (kvm_x86_ops->get_cpl(vcpu) != 0)
542 if (!(xcr0 & XSTATE_FP))
544 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
546 if (xcr0 & ~host_xcr0)
548 vcpu->arch.xcr0 = xcr0;
549 vcpu->guest_xcr0_loaded = 0;
553 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
555 if (__kvm_set_xcr(vcpu, index, xcr)) {
556 kvm_inject_gp(vcpu, 0);
561 EXPORT_SYMBOL_GPL(kvm_set_xcr);
563 static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
565 struct kvm_cpuid_entry2 *best;
567 best = kvm_find_cpuid_entry(vcpu, 1, 0);
568 return best && (best->ecx & bit(X86_FEATURE_XSAVE));
571 static void update_cpuid(struct kvm_vcpu *vcpu)
573 struct kvm_cpuid_entry2 *best;
575 best = kvm_find_cpuid_entry(vcpu, 1, 0);
579 /* Update OSXSAVE bit */
580 if (cpu_has_xsave && best->function == 0x1) {
581 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
582 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
583 best->ecx |= bit(X86_FEATURE_OSXSAVE);
587 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
589 unsigned long old_cr4 = kvm_read_cr4(vcpu);
590 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
592 if (cr4 & CR4_RESERVED_BITS)
595 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
598 if (is_long_mode(vcpu)) {
599 if (!(cr4 & X86_CR4_PAE))
601 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
602 && ((cr4 ^ old_cr4) & pdptr_bits)
603 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, vcpu->arch.cr3))
606 if (cr4 & X86_CR4_VMXE)
609 kvm_x86_ops->set_cr4(vcpu, cr4);
611 if ((cr4 ^ old_cr4) & pdptr_bits)
612 kvm_mmu_reset_context(vcpu);
614 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
619 EXPORT_SYMBOL_GPL(kvm_set_cr4);
621 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
623 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
624 kvm_mmu_sync_roots(vcpu);
625 kvm_mmu_flush_tlb(vcpu);
629 if (is_long_mode(vcpu)) {
630 if (cr3 & CR3_L_MODE_RESERVED_BITS)
634 if (cr3 & CR3_PAE_RESERVED_BITS)
636 if (is_paging(vcpu) &&
637 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
641 * We don't check reserved bits in nonpae mode, because
642 * this isn't enforced, and VMware depends on this.
647 * Does the new cr3 value map to physical memory? (Note, we
648 * catch an invalid cr3 even in real-mode, because it would
649 * cause trouble later on when we turn on paging anyway.)
651 * A real CPU would silently accept an invalid cr3 and would
652 * attempt to use it - with largely undefined (and often hard
653 * to debug) behavior on the guest side.
655 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
657 vcpu->arch.cr3 = cr3;
658 vcpu->arch.mmu.new_cr3(vcpu);
661 EXPORT_SYMBOL_GPL(kvm_set_cr3);
663 int __kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
665 if (cr8 & CR8_RESERVED_BITS)
667 if (irqchip_in_kernel(vcpu->kvm))
668 kvm_lapic_set_tpr(vcpu, cr8);
670 vcpu->arch.cr8 = cr8;
674 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
676 if (__kvm_set_cr8(vcpu, cr8))
677 kvm_inject_gp(vcpu, 0);
679 EXPORT_SYMBOL_GPL(kvm_set_cr8);
681 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
683 if (irqchip_in_kernel(vcpu->kvm))
684 return kvm_lapic_get_cr8(vcpu);
686 return vcpu->arch.cr8;
688 EXPORT_SYMBOL_GPL(kvm_get_cr8);
690 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
694 vcpu->arch.db[dr] = val;
695 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
696 vcpu->arch.eff_db[dr] = val;
699 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
703 if (val & 0xffffffff00000000ULL)
705 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
708 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
712 if (val & 0xffffffff00000000ULL)
714 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
715 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
716 kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
717 vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
725 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
729 res = __kvm_set_dr(vcpu, dr, val);
731 kvm_queue_exception(vcpu, UD_VECTOR);
733 kvm_inject_gp(vcpu, 0);
737 EXPORT_SYMBOL_GPL(kvm_set_dr);
739 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
743 *val = vcpu->arch.db[dr];
746 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
750 *val = vcpu->arch.dr6;
753 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
757 *val = vcpu->arch.dr7;
764 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
766 if (_kvm_get_dr(vcpu, dr, val)) {
767 kvm_queue_exception(vcpu, UD_VECTOR);
772 EXPORT_SYMBOL_GPL(kvm_get_dr);
775 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
776 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
778 * This list is modified at module load time to reflect the
779 * capabilities of the host cpu. This capabilities test skips MSRs that are
780 * kvm-specific. Those are put in the beginning of the list.
783 #define KVM_SAVE_MSRS_BEGIN 7
784 static u32 msrs_to_save[] = {
785 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
786 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
787 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
788 HV_X64_MSR_APIC_ASSIST_PAGE,
789 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
792 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
794 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
797 static unsigned num_msrs_to_save;
799 static u32 emulated_msrs[] = {
800 MSR_IA32_MISC_ENABLE,
805 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
807 u64 old_efer = vcpu->arch.efer;
809 if (efer & efer_reserved_bits)
813 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
816 if (efer & EFER_FFXSR) {
817 struct kvm_cpuid_entry2 *feat;
819 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
820 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
824 if (efer & EFER_SVME) {
825 struct kvm_cpuid_entry2 *feat;
827 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
828 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
833 efer |= vcpu->arch.efer & EFER_LMA;
835 kvm_x86_ops->set_efer(vcpu, efer);
837 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
838 kvm_mmu_reset_context(vcpu);
840 /* Update reserved bits */
841 if ((efer ^ old_efer) & EFER_NX)
842 kvm_mmu_reset_context(vcpu);
847 void kvm_enable_efer_bits(u64 mask)
849 efer_reserved_bits &= ~mask;
851 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
855 * Writes msr value into into the appropriate "register".
856 * Returns 0 on success, non-0 otherwise.
857 * Assumes vcpu_load() was already called.
859 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
861 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
865 * Adapt set_msr() to msr_io()'s calling convention
867 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
869 return kvm_set_msr(vcpu, index, *data);
872 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
876 struct pvclock_wall_clock wc;
877 struct timespec boot;
882 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
887 ++version; /* first time write, random junk */
891 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
894 * The guest calculates current wall clock time by adding
895 * system time (updated by kvm_guest_time_update below) to the
896 * wall clock specified here. guest system time equals host
897 * system time for us, thus we must fill in host boot time here.
901 wc.sec = boot.tv_sec;
902 wc.nsec = boot.tv_nsec;
903 wc.version = version;
905 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
908 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
911 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
913 uint32_t quotient, remainder;
915 /* Don't try to replace with do_div(), this one calculates
916 * "(dividend << 32) / divisor" */
918 : "=a" (quotient), "=d" (remainder)
919 : "0" (0), "1" (dividend), "r" (divisor) );
923 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
924 s8 *pshift, u32 *pmultiplier)
931 tps64 = base_khz * 1000LL;
932 scaled64 = scaled_khz * 1000LL;
933 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
938 tps32 = (uint32_t)tps64;
939 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
940 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
948 *pmultiplier = div_frac(scaled64, tps32);
950 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
951 __func__, base_khz, scaled_khz, shift, *pmultiplier);
954 static inline u64 get_kernel_ns(void)
958 WARN_ON(preemptible());
960 monotonic_to_bootbased(&ts);
961 return timespec_to_ns(&ts);
964 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
965 unsigned long max_tsc_khz;
967 static inline int kvm_tsc_changes_freq(void)
970 int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
971 cpufreq_quick_get(cpu) != 0;
976 static inline u64 nsec_to_cycles(u64 nsec)
980 WARN_ON(preemptible());
981 if (kvm_tsc_changes_freq())
982 printk_once(KERN_WARNING
983 "kvm: unreliable cycle conversion on adjustable rate TSC\n");
984 ret = nsec * __get_cpu_var(cpu_tsc_khz);
985 do_div(ret, USEC_PER_SEC);
989 static void kvm_arch_set_tsc_khz(struct kvm *kvm, u32 this_tsc_khz)
991 /* Compute a scale to convert nanoseconds in TSC cycles */
992 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
993 &kvm->arch.virtual_tsc_shift,
994 &kvm->arch.virtual_tsc_mult);
995 kvm->arch.virtual_tsc_khz = this_tsc_khz;
998 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1000 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1001 vcpu->kvm->arch.virtual_tsc_mult,
1002 vcpu->kvm->arch.virtual_tsc_shift);
1003 tsc += vcpu->arch.last_tsc_write;
1007 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1009 struct kvm *kvm = vcpu->kvm;
1010 u64 offset, ns, elapsed;
1011 unsigned long flags;
1014 spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1015 offset = data - native_read_tsc();
1016 ns = get_kernel_ns();
1017 elapsed = ns - kvm->arch.last_tsc_nsec;
1018 sdiff = data - kvm->arch.last_tsc_write;
1023 * Special case: close write to TSC within 5 seconds of
1024 * another CPU is interpreted as an attempt to synchronize
1025 * The 5 seconds is to accomodate host load / swapping as
1026 * well as any reset of TSC during the boot process.
1028 * In that case, for a reliable TSC, we can match TSC offsets,
1029 * or make a best guest using elapsed value.
1031 if (sdiff < nsec_to_cycles(5ULL * NSEC_PER_SEC) &&
1032 elapsed < 5ULL * NSEC_PER_SEC) {
1033 if (!check_tsc_unstable()) {
1034 offset = kvm->arch.last_tsc_offset;
1035 pr_debug("kvm: matched tsc offset for %llu\n", data);
1037 u64 delta = nsec_to_cycles(elapsed);
1039 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1041 ns = kvm->arch.last_tsc_nsec;
1043 kvm->arch.last_tsc_nsec = ns;
1044 kvm->arch.last_tsc_write = data;
1045 kvm->arch.last_tsc_offset = offset;
1046 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1047 spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1049 /* Reset of TSC must disable overshoot protection below */
1050 vcpu->arch.hv_clock.tsc_timestamp = 0;
1051 vcpu->arch.last_tsc_write = data;
1052 vcpu->arch.last_tsc_nsec = ns;
1054 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1056 static int kvm_guest_time_update(struct kvm_vcpu *v)
1058 unsigned long flags;
1059 struct kvm_vcpu_arch *vcpu = &v->arch;
1061 unsigned long this_tsc_khz;
1062 s64 kernel_ns, max_kernel_ns;
1065 /* Keep irq disabled to prevent changes to the clock */
1066 local_irq_save(flags);
1067 kvm_get_msr(v, MSR_IA32_TSC, &tsc_timestamp);
1068 kernel_ns = get_kernel_ns();
1069 this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
1071 if (unlikely(this_tsc_khz == 0)) {
1072 local_irq_restore(flags);
1073 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1078 * We may have to catch up the TSC to match elapsed wall clock
1079 * time for two reasons, even if kvmclock is used.
1080 * 1) CPU could have been running below the maximum TSC rate
1081 * 2) Broken TSC compensation resets the base at each VCPU
1082 * entry to avoid unknown leaps of TSC even when running
1083 * again on the same CPU. This may cause apparent elapsed
1084 * time to disappear, and the guest to stand still or run
1087 if (vcpu->tsc_catchup) {
1088 u64 tsc = compute_guest_tsc(v, kernel_ns);
1089 if (tsc > tsc_timestamp) {
1090 kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1091 tsc_timestamp = tsc;
1095 local_irq_restore(flags);
1097 if (!vcpu->time_page)
1101 * Time as measured by the TSC may go backwards when resetting the base
1102 * tsc_timestamp. The reason for this is that the TSC resolution is
1103 * higher than the resolution of the other clock scales. Thus, many
1104 * possible measurments of the TSC correspond to one measurement of any
1105 * other clock, and so a spread of values is possible. This is not a
1106 * problem for the computation of the nanosecond clock; with TSC rates
1107 * around 1GHZ, there can only be a few cycles which correspond to one
1108 * nanosecond value, and any path through this code will inevitably
1109 * take longer than that. However, with the kernel_ns value itself,
1110 * the precision may be much lower, down to HZ granularity. If the
1111 * first sampling of TSC against kernel_ns ends in the low part of the
1112 * range, and the second in the high end of the range, we can get:
1114 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1116 * As the sampling errors potentially range in the thousands of cycles,
1117 * it is possible such a time value has already been observed by the
1118 * guest. To protect against this, we must compute the system time as
1119 * observed by the guest and ensure the new system time is greater.
1122 if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1123 max_kernel_ns = vcpu->last_guest_tsc -
1124 vcpu->hv_clock.tsc_timestamp;
1125 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1126 vcpu->hv_clock.tsc_to_system_mul,
1127 vcpu->hv_clock.tsc_shift);
1128 max_kernel_ns += vcpu->last_kernel_ns;
1131 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1132 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1133 &vcpu->hv_clock.tsc_shift,
1134 &vcpu->hv_clock.tsc_to_system_mul);
1135 vcpu->hw_tsc_khz = this_tsc_khz;
1138 if (max_kernel_ns > kernel_ns)
1139 kernel_ns = max_kernel_ns;
1141 /* With all the info we got, fill in the values */
1142 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1143 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1144 vcpu->last_kernel_ns = kernel_ns;
1145 vcpu->last_guest_tsc = tsc_timestamp;
1146 vcpu->hv_clock.flags = 0;
1149 * The interface expects us to write an even number signaling that the
1150 * update is finished. Since the guest won't see the intermediate
1151 * state, we just increase by 2 at the end.
1153 vcpu->hv_clock.version += 2;
1155 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1157 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1158 sizeof(vcpu->hv_clock));
1160 kunmap_atomic(shared_kaddr, KM_USER0);
1162 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1166 static bool msr_mtrr_valid(unsigned msr)
1169 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1170 case MSR_MTRRfix64K_00000:
1171 case MSR_MTRRfix16K_80000:
1172 case MSR_MTRRfix16K_A0000:
1173 case MSR_MTRRfix4K_C0000:
1174 case MSR_MTRRfix4K_C8000:
1175 case MSR_MTRRfix4K_D0000:
1176 case MSR_MTRRfix4K_D8000:
1177 case MSR_MTRRfix4K_E0000:
1178 case MSR_MTRRfix4K_E8000:
1179 case MSR_MTRRfix4K_F0000:
1180 case MSR_MTRRfix4K_F8000:
1181 case MSR_MTRRdefType:
1182 case MSR_IA32_CR_PAT:
1190 static bool valid_pat_type(unsigned t)
1192 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1195 static bool valid_mtrr_type(unsigned t)
1197 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1200 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1204 if (!msr_mtrr_valid(msr))
1207 if (msr == MSR_IA32_CR_PAT) {
1208 for (i = 0; i < 8; i++)
1209 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1212 } else if (msr == MSR_MTRRdefType) {
1215 return valid_mtrr_type(data & 0xff);
1216 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1217 for (i = 0; i < 8 ; i++)
1218 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1223 /* variable MTRRs */
1224 return valid_mtrr_type(data & 0xff);
1227 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1229 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1231 if (!mtrr_valid(vcpu, msr, data))
1234 if (msr == MSR_MTRRdefType) {
1235 vcpu->arch.mtrr_state.def_type = data;
1236 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1237 } else if (msr == MSR_MTRRfix64K_00000)
1239 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1240 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1241 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1242 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1243 else if (msr == MSR_IA32_CR_PAT)
1244 vcpu->arch.pat = data;
1245 else { /* Variable MTRRs */
1246 int idx, is_mtrr_mask;
1249 idx = (msr - 0x200) / 2;
1250 is_mtrr_mask = msr - 0x200 - 2 * idx;
1253 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1256 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1260 kvm_mmu_reset_context(vcpu);
1264 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1266 u64 mcg_cap = vcpu->arch.mcg_cap;
1267 unsigned bank_num = mcg_cap & 0xff;
1270 case MSR_IA32_MCG_STATUS:
1271 vcpu->arch.mcg_status = data;
1273 case MSR_IA32_MCG_CTL:
1274 if (!(mcg_cap & MCG_CTL_P))
1276 if (data != 0 && data != ~(u64)0)
1278 vcpu->arch.mcg_ctl = data;
1281 if (msr >= MSR_IA32_MC0_CTL &&
1282 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1283 u32 offset = msr - MSR_IA32_MC0_CTL;
1284 /* only 0 or all 1s can be written to IA32_MCi_CTL
1285 * some Linux kernels though clear bit 10 in bank 4 to
1286 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1287 * this to avoid an uncatched #GP in the guest
1289 if ((offset & 0x3) == 0 &&
1290 data != 0 && (data | (1 << 10)) != ~(u64)0)
1292 vcpu->arch.mce_banks[offset] = data;
1300 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1302 struct kvm *kvm = vcpu->kvm;
1303 int lm = is_long_mode(vcpu);
1304 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1305 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1306 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1307 : kvm->arch.xen_hvm_config.blob_size_32;
1308 u32 page_num = data & ~PAGE_MASK;
1309 u64 page_addr = data & PAGE_MASK;
1314 if (page_num >= blob_size)
1317 page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1321 if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1323 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1332 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1334 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1337 static bool kvm_hv_msr_partition_wide(u32 msr)
1341 case HV_X64_MSR_GUEST_OS_ID:
1342 case HV_X64_MSR_HYPERCALL:
1350 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1352 struct kvm *kvm = vcpu->kvm;
1355 case HV_X64_MSR_GUEST_OS_ID:
1356 kvm->arch.hv_guest_os_id = data;
1357 /* setting guest os id to zero disables hypercall page */
1358 if (!kvm->arch.hv_guest_os_id)
1359 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1361 case HV_X64_MSR_HYPERCALL: {
1366 /* if guest os id is not set hypercall should remain disabled */
1367 if (!kvm->arch.hv_guest_os_id)
1369 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1370 kvm->arch.hv_hypercall = data;
1373 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1374 addr = gfn_to_hva(kvm, gfn);
1375 if (kvm_is_error_hva(addr))
1377 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1378 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1379 if (copy_to_user((void __user *)addr, instructions, 4))
1381 kvm->arch.hv_hypercall = data;
1385 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1386 "data 0x%llx\n", msr, data);
1392 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1395 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1398 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1399 vcpu->arch.hv_vapic = data;
1402 addr = gfn_to_hva(vcpu->kvm, data >>
1403 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1404 if (kvm_is_error_hva(addr))
1406 if (clear_user((void __user *)addr, PAGE_SIZE))
1408 vcpu->arch.hv_vapic = data;
1411 case HV_X64_MSR_EOI:
1412 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1413 case HV_X64_MSR_ICR:
1414 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1415 case HV_X64_MSR_TPR:
1416 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1418 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1419 "data 0x%llx\n", msr, data);
1426 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1430 return set_efer(vcpu, data);
1432 data &= ~(u64)0x40; /* ignore flush filter disable */
1433 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1435 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1440 case MSR_FAM10H_MMIO_CONF_BASE:
1442 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1447 case MSR_AMD64_NB_CFG:
1449 case MSR_IA32_DEBUGCTLMSR:
1451 /* We support the non-activated case already */
1453 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1454 /* Values other than LBR and BTF are vendor-specific,
1455 thus reserved and should throw a #GP */
1458 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1461 case MSR_IA32_UCODE_REV:
1462 case MSR_IA32_UCODE_WRITE:
1463 case MSR_VM_HSAVE_PA:
1464 case MSR_AMD64_PATCH_LOADER:
1466 case 0x200 ... 0x2ff:
1467 return set_msr_mtrr(vcpu, msr, data);
1468 case MSR_IA32_APICBASE:
1469 kvm_set_apic_base(vcpu, data);
1471 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1472 return kvm_x2apic_msr_write(vcpu, msr, data);
1473 case MSR_IA32_MISC_ENABLE:
1474 vcpu->arch.ia32_misc_enable_msr = data;
1476 case MSR_KVM_WALL_CLOCK_NEW:
1477 case MSR_KVM_WALL_CLOCK:
1478 vcpu->kvm->arch.wall_clock = data;
1479 kvm_write_wall_clock(vcpu->kvm, data);
1481 case MSR_KVM_SYSTEM_TIME_NEW:
1482 case MSR_KVM_SYSTEM_TIME: {
1483 if (vcpu->arch.time_page) {
1484 kvm_release_page_dirty(vcpu->arch.time_page);
1485 vcpu->arch.time_page = NULL;
1488 vcpu->arch.time = data;
1489 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1491 /* we verify if the enable bit is set... */
1495 /* ...but clean it before doing the actual write */
1496 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1498 vcpu->arch.time_page =
1499 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1501 if (is_error_page(vcpu->arch.time_page)) {
1502 kvm_release_page_clean(vcpu->arch.time_page);
1503 vcpu->arch.time_page = NULL;
1507 case MSR_IA32_MCG_CTL:
1508 case MSR_IA32_MCG_STATUS:
1509 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1510 return set_msr_mce(vcpu, msr, data);
1512 /* Performance counters are not protected by a CPUID bit,
1513 * so we should check all of them in the generic path for the sake of
1514 * cross vendor migration.
1515 * Writing a zero into the event select MSRs disables them,
1516 * which we perfectly emulate ;-). Any other value should be at least
1517 * reported, some guests depend on them.
1519 case MSR_P6_EVNTSEL0:
1520 case MSR_P6_EVNTSEL1:
1521 case MSR_K7_EVNTSEL0:
1522 case MSR_K7_EVNTSEL1:
1523 case MSR_K7_EVNTSEL2:
1524 case MSR_K7_EVNTSEL3:
1526 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1527 "0x%x data 0x%llx\n", msr, data);
1529 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1530 * so we ignore writes to make it happy.
1532 case MSR_P6_PERFCTR0:
1533 case MSR_P6_PERFCTR1:
1534 case MSR_K7_PERFCTR0:
1535 case MSR_K7_PERFCTR1:
1536 case MSR_K7_PERFCTR2:
1537 case MSR_K7_PERFCTR3:
1538 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1539 "0x%x data 0x%llx\n", msr, data);
1541 case MSR_K7_CLK_CTL:
1543 * Ignore all writes to this no longer documented MSR.
1544 * Writes are only relevant for old K7 processors,
1545 * all pre-dating SVM, but a recommended workaround from
1546 * AMD for these chips. It is possible to speicify the
1547 * affected processor models on the command line, hence
1548 * the need to ignore the workaround.
1551 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1552 if (kvm_hv_msr_partition_wide(msr)) {
1554 mutex_lock(&vcpu->kvm->lock);
1555 r = set_msr_hyperv_pw(vcpu, msr, data);
1556 mutex_unlock(&vcpu->kvm->lock);
1559 return set_msr_hyperv(vcpu, msr, data);
1562 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1563 return xen_hvm_config(vcpu, data);
1565 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1569 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1576 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1580 * Reads an msr value (of 'msr_index') into 'pdata'.
1581 * Returns 0 on success, non-0 otherwise.
1582 * Assumes vcpu_load() was already called.
1584 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1586 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1589 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1591 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1593 if (!msr_mtrr_valid(msr))
1596 if (msr == MSR_MTRRdefType)
1597 *pdata = vcpu->arch.mtrr_state.def_type +
1598 (vcpu->arch.mtrr_state.enabled << 10);
1599 else if (msr == MSR_MTRRfix64K_00000)
1601 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1602 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1603 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1604 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1605 else if (msr == MSR_IA32_CR_PAT)
1606 *pdata = vcpu->arch.pat;
1607 else { /* Variable MTRRs */
1608 int idx, is_mtrr_mask;
1611 idx = (msr - 0x200) / 2;
1612 is_mtrr_mask = msr - 0x200 - 2 * idx;
1615 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1618 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1625 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1628 u64 mcg_cap = vcpu->arch.mcg_cap;
1629 unsigned bank_num = mcg_cap & 0xff;
1632 case MSR_IA32_P5_MC_ADDR:
1633 case MSR_IA32_P5_MC_TYPE:
1636 case MSR_IA32_MCG_CAP:
1637 data = vcpu->arch.mcg_cap;
1639 case MSR_IA32_MCG_CTL:
1640 if (!(mcg_cap & MCG_CTL_P))
1642 data = vcpu->arch.mcg_ctl;
1644 case MSR_IA32_MCG_STATUS:
1645 data = vcpu->arch.mcg_status;
1648 if (msr >= MSR_IA32_MC0_CTL &&
1649 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1650 u32 offset = msr - MSR_IA32_MC0_CTL;
1651 data = vcpu->arch.mce_banks[offset];
1660 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1663 struct kvm *kvm = vcpu->kvm;
1666 case HV_X64_MSR_GUEST_OS_ID:
1667 data = kvm->arch.hv_guest_os_id;
1669 case HV_X64_MSR_HYPERCALL:
1670 data = kvm->arch.hv_hypercall;
1673 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1681 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1686 case HV_X64_MSR_VP_INDEX: {
1689 kvm_for_each_vcpu(r, v, vcpu->kvm)
1694 case HV_X64_MSR_EOI:
1695 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1696 case HV_X64_MSR_ICR:
1697 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1698 case HV_X64_MSR_TPR:
1699 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1701 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1708 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1713 case MSR_IA32_PLATFORM_ID:
1714 case MSR_IA32_UCODE_REV:
1715 case MSR_IA32_EBL_CR_POWERON:
1716 case MSR_IA32_DEBUGCTLMSR:
1717 case MSR_IA32_LASTBRANCHFROMIP:
1718 case MSR_IA32_LASTBRANCHTOIP:
1719 case MSR_IA32_LASTINTFROMIP:
1720 case MSR_IA32_LASTINTTOIP:
1723 case MSR_VM_HSAVE_PA:
1724 case MSR_P6_PERFCTR0:
1725 case MSR_P6_PERFCTR1:
1726 case MSR_P6_EVNTSEL0:
1727 case MSR_P6_EVNTSEL1:
1728 case MSR_K7_EVNTSEL0:
1729 case MSR_K7_PERFCTR0:
1730 case MSR_K8_INT_PENDING_MSG:
1731 case MSR_AMD64_NB_CFG:
1732 case MSR_FAM10H_MMIO_CONF_BASE:
1736 data = 0x500 | KVM_NR_VAR_MTRR;
1738 case 0x200 ... 0x2ff:
1739 return get_msr_mtrr(vcpu, msr, pdata);
1740 case 0xcd: /* fsb frequency */
1744 * MSR_EBC_FREQUENCY_ID
1745 * Conservative value valid for even the basic CPU models.
1746 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1747 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1748 * and 266MHz for model 3, or 4. Set Core Clock
1749 * Frequency to System Bus Frequency Ratio to 1 (bits
1750 * 31:24) even though these are only valid for CPU
1751 * models > 2, however guests may end up dividing or
1752 * multiplying by zero otherwise.
1754 case MSR_EBC_FREQUENCY_ID:
1757 case MSR_IA32_APICBASE:
1758 data = kvm_get_apic_base(vcpu);
1760 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1761 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1763 case MSR_IA32_MISC_ENABLE:
1764 data = vcpu->arch.ia32_misc_enable_msr;
1766 case MSR_IA32_PERF_STATUS:
1767 /* TSC increment by tick */
1769 /* CPU multiplier */
1770 data |= (((uint64_t)4ULL) << 40);
1773 data = vcpu->arch.efer;
1775 case MSR_KVM_WALL_CLOCK:
1776 case MSR_KVM_WALL_CLOCK_NEW:
1777 data = vcpu->kvm->arch.wall_clock;
1779 case MSR_KVM_SYSTEM_TIME:
1780 case MSR_KVM_SYSTEM_TIME_NEW:
1781 data = vcpu->arch.time;
1783 case MSR_IA32_P5_MC_ADDR:
1784 case MSR_IA32_P5_MC_TYPE:
1785 case MSR_IA32_MCG_CAP:
1786 case MSR_IA32_MCG_CTL:
1787 case MSR_IA32_MCG_STATUS:
1788 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1789 return get_msr_mce(vcpu, msr, pdata);
1790 case MSR_K7_CLK_CTL:
1792 * Provide expected ramp-up count for K7. All other
1793 * are set to zero, indicating minimum divisors for
1796 * This prevents guest kernels on AMD host with CPU
1797 * type 6, model 8 and higher from exploding due to
1798 * the rdmsr failing.
1802 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1803 if (kvm_hv_msr_partition_wide(msr)) {
1805 mutex_lock(&vcpu->kvm->lock);
1806 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1807 mutex_unlock(&vcpu->kvm->lock);
1810 return get_msr_hyperv(vcpu, msr, pdata);
1814 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1817 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1825 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1828 * Read or write a bunch of msrs. All parameters are kernel addresses.
1830 * @return number of msrs set successfully.
1832 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1833 struct kvm_msr_entry *entries,
1834 int (*do_msr)(struct kvm_vcpu *vcpu,
1835 unsigned index, u64 *data))
1839 idx = srcu_read_lock(&vcpu->kvm->srcu);
1840 for (i = 0; i < msrs->nmsrs; ++i)
1841 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1843 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1849 * Read or write a bunch of msrs. Parameters are user addresses.
1851 * @return number of msrs set successfully.
1853 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1854 int (*do_msr)(struct kvm_vcpu *vcpu,
1855 unsigned index, u64 *data),
1858 struct kvm_msrs msrs;
1859 struct kvm_msr_entry *entries;
1864 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1868 if (msrs.nmsrs >= MAX_IO_MSRS)
1872 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1873 entries = kmalloc(size, GFP_KERNEL);
1878 if (copy_from_user(entries, user_msrs->entries, size))
1881 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1886 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1897 int kvm_dev_ioctl_check_extension(long ext)
1902 case KVM_CAP_IRQCHIP:
1904 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1905 case KVM_CAP_SET_TSS_ADDR:
1906 case KVM_CAP_EXT_CPUID:
1907 case KVM_CAP_CLOCKSOURCE:
1909 case KVM_CAP_NOP_IO_DELAY:
1910 case KVM_CAP_MP_STATE:
1911 case KVM_CAP_SYNC_MMU:
1912 case KVM_CAP_REINJECT_CONTROL:
1913 case KVM_CAP_IRQ_INJECT_STATUS:
1914 case KVM_CAP_ASSIGN_DEV_IRQ:
1916 case KVM_CAP_IOEVENTFD:
1918 case KVM_CAP_PIT_STATE2:
1919 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1920 case KVM_CAP_XEN_HVM:
1921 case KVM_CAP_ADJUST_CLOCK:
1922 case KVM_CAP_VCPU_EVENTS:
1923 case KVM_CAP_HYPERV:
1924 case KVM_CAP_HYPERV_VAPIC:
1925 case KVM_CAP_HYPERV_SPIN:
1926 case KVM_CAP_PCI_SEGMENT:
1927 case KVM_CAP_DEBUGREGS:
1928 case KVM_CAP_X86_ROBUST_SINGLESTEP:
1932 case KVM_CAP_COALESCED_MMIO:
1933 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1936 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1938 case KVM_CAP_NR_VCPUS:
1941 case KVM_CAP_NR_MEMSLOTS:
1942 r = KVM_MEMORY_SLOTS;
1944 case KVM_CAP_PV_MMU: /* obsolete */
1951 r = KVM_MAX_MCE_BANKS;
1964 long kvm_arch_dev_ioctl(struct file *filp,
1965 unsigned int ioctl, unsigned long arg)
1967 void __user *argp = (void __user *)arg;
1971 case KVM_GET_MSR_INDEX_LIST: {
1972 struct kvm_msr_list __user *user_msr_list = argp;
1973 struct kvm_msr_list msr_list;
1977 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1980 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1981 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1984 if (n < msr_list.nmsrs)
1987 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1988 num_msrs_to_save * sizeof(u32)))
1990 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1992 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1997 case KVM_GET_SUPPORTED_CPUID: {
1998 struct kvm_cpuid2 __user *cpuid_arg = argp;
1999 struct kvm_cpuid2 cpuid;
2002 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2004 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2005 cpuid_arg->entries);
2010 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2015 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2018 mce_cap = KVM_MCE_CAP_SUPPORTED;
2020 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2032 static void wbinvd_ipi(void *garbage)
2037 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2039 return vcpu->kvm->arch.iommu_domain &&
2040 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2043 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2045 /* Address WBINVD may be executed by guest */
2046 if (need_emulate_wbinvd(vcpu)) {
2047 if (kvm_x86_ops->has_wbinvd_exit())
2048 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2049 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2050 smp_call_function_single(vcpu->cpu,
2051 wbinvd_ipi, NULL, 1);
2054 kvm_x86_ops->vcpu_load(vcpu, cpu);
2055 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2056 /* Make sure TSC doesn't go backwards */
2057 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2058 native_read_tsc() - vcpu->arch.last_host_tsc;
2060 mark_tsc_unstable("KVM discovered backwards TSC");
2061 if (check_tsc_unstable()) {
2062 kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2063 vcpu->arch.tsc_catchup = 1;
2064 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2066 if (vcpu->cpu != cpu)
2067 kvm_migrate_timers(vcpu);
2072 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2074 kvm_x86_ops->vcpu_put(vcpu);
2075 kvm_put_guest_fpu(vcpu);
2076 vcpu->arch.last_host_tsc = native_read_tsc();
2079 static int is_efer_nx(void)
2081 unsigned long long efer = 0;
2083 rdmsrl_safe(MSR_EFER, &efer);
2084 return efer & EFER_NX;
2087 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2090 struct kvm_cpuid_entry2 *e, *entry;
2093 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2094 e = &vcpu->arch.cpuid_entries[i];
2095 if (e->function == 0x80000001) {
2100 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2101 entry->edx &= ~(1 << 20);
2102 printk(KERN_INFO "kvm: guest NX capability removed\n");
2106 /* when an old userspace process fills a new kernel module */
2107 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2108 struct kvm_cpuid *cpuid,
2109 struct kvm_cpuid_entry __user *entries)
2112 struct kvm_cpuid_entry *cpuid_entries;
2115 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2118 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2122 if (copy_from_user(cpuid_entries, entries,
2123 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2125 for (i = 0; i < cpuid->nent; i++) {
2126 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2127 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2128 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2129 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2130 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2131 vcpu->arch.cpuid_entries[i].index = 0;
2132 vcpu->arch.cpuid_entries[i].flags = 0;
2133 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2134 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2135 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2137 vcpu->arch.cpuid_nent = cpuid->nent;
2138 cpuid_fix_nx_cap(vcpu);
2140 kvm_apic_set_version(vcpu);
2141 kvm_x86_ops->cpuid_update(vcpu);
2145 vfree(cpuid_entries);
2150 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2151 struct kvm_cpuid2 *cpuid,
2152 struct kvm_cpuid_entry2 __user *entries)
2157 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2160 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2161 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2163 vcpu->arch.cpuid_nent = cpuid->nent;
2164 kvm_apic_set_version(vcpu);
2165 kvm_x86_ops->cpuid_update(vcpu);
2173 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2174 struct kvm_cpuid2 *cpuid,
2175 struct kvm_cpuid_entry2 __user *entries)
2180 if (cpuid->nent < vcpu->arch.cpuid_nent)
2183 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2184 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2189 cpuid->nent = vcpu->arch.cpuid_nent;
2193 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2196 entry->function = function;
2197 entry->index = index;
2198 cpuid_count(entry->function, entry->index,
2199 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2203 #define F(x) bit(X86_FEATURE_##x)
2205 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2206 u32 index, int *nent, int maxnent)
2208 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2209 #ifdef CONFIG_X86_64
2210 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2212 unsigned f_lm = F(LM);
2214 unsigned f_gbpages = 0;
2217 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2220 const u32 kvm_supported_word0_x86_features =
2221 F(FPU) | F(VME) | F(DE) | F(PSE) |
2222 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2223 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2224 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2225 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2226 0 /* Reserved, DS, ACPI */ | F(MMX) |
2227 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2228 0 /* HTT, TM, Reserved, PBE */;
2229 /* cpuid 0x80000001.edx */
2230 const u32 kvm_supported_word1_x86_features =
2231 F(FPU) | F(VME) | F(DE) | F(PSE) |
2232 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2233 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2234 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2235 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2236 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2237 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2238 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2240 const u32 kvm_supported_word4_x86_features =
2241 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
2242 0 /* DS-CPL, VMX, SMX, EST */ |
2243 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2244 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2245 0 /* Reserved, DCA */ | F(XMM4_1) |
2246 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2247 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2249 /* cpuid 0x80000001.ecx */
2250 const u32 kvm_supported_word6_x86_features =
2251 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2252 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2253 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2254 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2256 /* all calls to cpuid_count() should be made on the same cpu */
2258 do_cpuid_1_ent(entry, function, index);
2263 entry->eax = min(entry->eax, (u32)0xd);
2266 entry->edx &= kvm_supported_word0_x86_features;
2267 entry->ecx &= kvm_supported_word4_x86_features;
2268 /* we support x2apic emulation even if host does not support
2269 * it since we emulate x2apic in software */
2270 entry->ecx |= F(X2APIC);
2272 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2273 * may return different values. This forces us to get_cpu() before
2274 * issuing the first command, and also to emulate this annoying behavior
2275 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2277 int t, times = entry->eax & 0xff;
2279 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2280 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2281 for (t = 1; t < times && *nent < maxnent; ++t) {
2282 do_cpuid_1_ent(&entry[t], function, 0);
2283 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2288 /* function 4 and 0xb have additional index. */
2292 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2293 /* read more entries until cache_type is zero */
2294 for (i = 1; *nent < maxnent; ++i) {
2295 cache_type = entry[i - 1].eax & 0x1f;
2298 do_cpuid_1_ent(&entry[i], function, i);
2300 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2308 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2309 /* read more entries until level_type is zero */
2310 for (i = 1; *nent < maxnent; ++i) {
2311 level_type = entry[i - 1].ecx & 0xff00;
2314 do_cpuid_1_ent(&entry[i], function, i);
2316 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2324 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2325 for (i = 1; *nent < maxnent; ++i) {
2326 if (entry[i - 1].eax == 0 && i != 2)
2328 do_cpuid_1_ent(&entry[i], function, i);
2330 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2335 case KVM_CPUID_SIGNATURE: {
2336 char signature[12] = "KVMKVMKVM\0\0";
2337 u32 *sigptr = (u32 *)signature;
2339 entry->ebx = sigptr[0];
2340 entry->ecx = sigptr[1];
2341 entry->edx = sigptr[2];
2344 case KVM_CPUID_FEATURES:
2345 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2346 (1 << KVM_FEATURE_NOP_IO_DELAY) |
2347 (1 << KVM_FEATURE_CLOCKSOURCE2) |
2348 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2354 entry->eax = min(entry->eax, 0x8000001a);
2357 entry->edx &= kvm_supported_word1_x86_features;
2358 entry->ecx &= kvm_supported_word6_x86_features;
2362 kvm_x86_ops->set_supported_cpuid(function, entry);
2369 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2370 struct kvm_cpuid_entry2 __user *entries)
2372 struct kvm_cpuid_entry2 *cpuid_entries;
2373 int limit, nent = 0, r = -E2BIG;
2376 if (cpuid->nent < 1)
2378 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2379 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2381 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2385 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2386 limit = cpuid_entries[0].eax;
2387 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2388 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2389 &nent, cpuid->nent);
2391 if (nent >= cpuid->nent)
2394 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2395 limit = cpuid_entries[nent - 1].eax;
2396 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2397 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2398 &nent, cpuid->nent);
2403 if (nent >= cpuid->nent)
2406 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2410 if (nent >= cpuid->nent)
2413 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2417 if (nent >= cpuid->nent)
2421 if (copy_to_user(entries, cpuid_entries,
2422 nent * sizeof(struct kvm_cpuid_entry2)))
2428 vfree(cpuid_entries);
2433 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2434 struct kvm_lapic_state *s)
2436 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2441 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2442 struct kvm_lapic_state *s)
2444 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2445 kvm_apic_post_state_restore(vcpu);
2446 update_cr8_intercept(vcpu);
2451 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2452 struct kvm_interrupt *irq)
2454 if (irq->irq < 0 || irq->irq >= 256)
2456 if (irqchip_in_kernel(vcpu->kvm))
2459 kvm_queue_interrupt(vcpu, irq->irq, false);
2460 kvm_make_request(KVM_REQ_EVENT, vcpu);
2465 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2467 kvm_inject_nmi(vcpu);
2472 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2473 struct kvm_tpr_access_ctl *tac)
2477 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2481 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2485 unsigned bank_num = mcg_cap & 0xff, bank;
2488 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2490 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2493 vcpu->arch.mcg_cap = mcg_cap;
2494 /* Init IA32_MCG_CTL to all 1s */
2495 if (mcg_cap & MCG_CTL_P)
2496 vcpu->arch.mcg_ctl = ~(u64)0;
2497 /* Init IA32_MCi_CTL to all 1s */
2498 for (bank = 0; bank < bank_num; bank++)
2499 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2504 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2505 struct kvm_x86_mce *mce)
2507 u64 mcg_cap = vcpu->arch.mcg_cap;
2508 unsigned bank_num = mcg_cap & 0xff;
2509 u64 *banks = vcpu->arch.mce_banks;
2511 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2514 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2515 * reporting is disabled
2517 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2518 vcpu->arch.mcg_ctl != ~(u64)0)
2520 banks += 4 * mce->bank;
2522 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2523 * reporting is disabled for the bank
2525 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2527 if (mce->status & MCI_STATUS_UC) {
2528 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2529 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2530 printk(KERN_DEBUG "kvm: set_mce: "
2531 "injects mce exception while "
2532 "previous one is in progress!\n");
2533 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2536 if (banks[1] & MCI_STATUS_VAL)
2537 mce->status |= MCI_STATUS_OVER;
2538 banks[2] = mce->addr;
2539 banks[3] = mce->misc;
2540 vcpu->arch.mcg_status = mce->mcg_status;
2541 banks[1] = mce->status;
2542 kvm_queue_exception(vcpu, MC_VECTOR);
2543 } else if (!(banks[1] & MCI_STATUS_VAL)
2544 || !(banks[1] & MCI_STATUS_UC)) {
2545 if (banks[1] & MCI_STATUS_VAL)
2546 mce->status |= MCI_STATUS_OVER;
2547 banks[2] = mce->addr;
2548 banks[3] = mce->misc;
2549 banks[1] = mce->status;
2551 banks[1] |= MCI_STATUS_OVER;
2555 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2556 struct kvm_vcpu_events *events)
2558 events->exception.injected =
2559 vcpu->arch.exception.pending &&
2560 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2561 events->exception.nr = vcpu->arch.exception.nr;
2562 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2563 events->exception.pad = 0;
2564 events->exception.error_code = vcpu->arch.exception.error_code;
2566 events->interrupt.injected =
2567 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2568 events->interrupt.nr = vcpu->arch.interrupt.nr;
2569 events->interrupt.soft = 0;
2570 events->interrupt.shadow =
2571 kvm_x86_ops->get_interrupt_shadow(vcpu,
2572 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2574 events->nmi.injected = vcpu->arch.nmi_injected;
2575 events->nmi.pending = vcpu->arch.nmi_pending;
2576 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2577 events->nmi.pad = 0;
2579 events->sipi_vector = vcpu->arch.sipi_vector;
2581 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2582 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2583 | KVM_VCPUEVENT_VALID_SHADOW);
2584 memset(&events->reserved, 0, sizeof(events->reserved));
2587 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2588 struct kvm_vcpu_events *events)
2590 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2591 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2592 | KVM_VCPUEVENT_VALID_SHADOW))
2595 vcpu->arch.exception.pending = events->exception.injected;
2596 vcpu->arch.exception.nr = events->exception.nr;
2597 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2598 vcpu->arch.exception.error_code = events->exception.error_code;
2600 vcpu->arch.interrupt.pending = events->interrupt.injected;
2601 vcpu->arch.interrupt.nr = events->interrupt.nr;
2602 vcpu->arch.interrupt.soft = events->interrupt.soft;
2603 if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
2604 kvm_pic_clear_isr_ack(vcpu->kvm);
2605 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2606 kvm_x86_ops->set_interrupt_shadow(vcpu,
2607 events->interrupt.shadow);
2609 vcpu->arch.nmi_injected = events->nmi.injected;
2610 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2611 vcpu->arch.nmi_pending = events->nmi.pending;
2612 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2614 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2615 vcpu->arch.sipi_vector = events->sipi_vector;
2617 kvm_make_request(KVM_REQ_EVENT, vcpu);
2622 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2623 struct kvm_debugregs *dbgregs)
2625 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2626 dbgregs->dr6 = vcpu->arch.dr6;
2627 dbgregs->dr7 = vcpu->arch.dr7;
2629 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2632 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2633 struct kvm_debugregs *dbgregs)
2638 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2639 vcpu->arch.dr6 = dbgregs->dr6;
2640 vcpu->arch.dr7 = dbgregs->dr7;
2645 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2646 struct kvm_xsave *guest_xsave)
2649 memcpy(guest_xsave->region,
2650 &vcpu->arch.guest_fpu.state->xsave,
2653 memcpy(guest_xsave->region,
2654 &vcpu->arch.guest_fpu.state->fxsave,
2655 sizeof(struct i387_fxsave_struct));
2656 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2661 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2662 struct kvm_xsave *guest_xsave)
2665 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2668 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2669 guest_xsave->region, xstate_size);
2671 if (xstate_bv & ~XSTATE_FPSSE)
2673 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2674 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2679 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2680 struct kvm_xcrs *guest_xcrs)
2682 if (!cpu_has_xsave) {
2683 guest_xcrs->nr_xcrs = 0;
2687 guest_xcrs->nr_xcrs = 1;
2688 guest_xcrs->flags = 0;
2689 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2690 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2693 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2694 struct kvm_xcrs *guest_xcrs)
2701 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2704 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2705 /* Only support XCR0 currently */
2706 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2707 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2708 guest_xcrs->xcrs[0].value);
2716 long kvm_arch_vcpu_ioctl(struct file *filp,
2717 unsigned int ioctl, unsigned long arg)
2719 struct kvm_vcpu *vcpu = filp->private_data;
2720 void __user *argp = (void __user *)arg;
2723 struct kvm_lapic_state *lapic;
2724 struct kvm_xsave *xsave;
2725 struct kvm_xcrs *xcrs;
2731 case KVM_GET_LAPIC: {
2733 if (!vcpu->arch.apic)
2735 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2740 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
2744 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
2749 case KVM_SET_LAPIC: {
2751 if (!vcpu->arch.apic)
2753 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2758 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
2760 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
2766 case KVM_INTERRUPT: {
2767 struct kvm_interrupt irq;
2770 if (copy_from_user(&irq, argp, sizeof irq))
2772 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2779 r = kvm_vcpu_ioctl_nmi(vcpu);
2785 case KVM_SET_CPUID: {
2786 struct kvm_cpuid __user *cpuid_arg = argp;
2787 struct kvm_cpuid cpuid;
2790 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2792 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2797 case KVM_SET_CPUID2: {
2798 struct kvm_cpuid2 __user *cpuid_arg = argp;
2799 struct kvm_cpuid2 cpuid;
2802 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2804 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2805 cpuid_arg->entries);
2810 case KVM_GET_CPUID2: {
2811 struct kvm_cpuid2 __user *cpuid_arg = argp;
2812 struct kvm_cpuid2 cpuid;
2815 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2817 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2818 cpuid_arg->entries);
2822 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2828 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2831 r = msr_io(vcpu, argp, do_set_msr, 0);
2833 case KVM_TPR_ACCESS_REPORTING: {
2834 struct kvm_tpr_access_ctl tac;
2837 if (copy_from_user(&tac, argp, sizeof tac))
2839 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2843 if (copy_to_user(argp, &tac, sizeof tac))
2848 case KVM_SET_VAPIC_ADDR: {
2849 struct kvm_vapic_addr va;
2852 if (!irqchip_in_kernel(vcpu->kvm))
2855 if (copy_from_user(&va, argp, sizeof va))
2858 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2861 case KVM_X86_SETUP_MCE: {
2865 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2867 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2870 case KVM_X86_SET_MCE: {
2871 struct kvm_x86_mce mce;
2874 if (copy_from_user(&mce, argp, sizeof mce))
2876 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2879 case KVM_GET_VCPU_EVENTS: {
2880 struct kvm_vcpu_events events;
2882 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2885 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2890 case KVM_SET_VCPU_EVENTS: {
2891 struct kvm_vcpu_events events;
2894 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2897 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2900 case KVM_GET_DEBUGREGS: {
2901 struct kvm_debugregs dbgregs;
2903 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2906 if (copy_to_user(argp, &dbgregs,
2907 sizeof(struct kvm_debugregs)))
2912 case KVM_SET_DEBUGREGS: {
2913 struct kvm_debugregs dbgregs;
2916 if (copy_from_user(&dbgregs, argp,
2917 sizeof(struct kvm_debugregs)))
2920 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2923 case KVM_GET_XSAVE: {
2924 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2929 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
2932 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
2937 case KVM_SET_XSAVE: {
2938 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2944 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
2947 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
2950 case KVM_GET_XCRS: {
2951 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2956 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
2959 if (copy_to_user(argp, u.xcrs,
2960 sizeof(struct kvm_xcrs)))
2965 case KVM_SET_XCRS: {
2966 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2972 if (copy_from_user(u.xcrs, argp,
2973 sizeof(struct kvm_xcrs)))
2976 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
2987 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2991 if (addr > (unsigned int)(-3 * PAGE_SIZE))
2993 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2997 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3000 kvm->arch.ept_identity_map_addr = ident_addr;
3004 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3005 u32 kvm_nr_mmu_pages)
3007 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3010 mutex_lock(&kvm->slots_lock);
3011 spin_lock(&kvm->mmu_lock);
3013 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3014 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3016 spin_unlock(&kvm->mmu_lock);
3017 mutex_unlock(&kvm->slots_lock);
3021 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3023 return kvm->arch.n_max_mmu_pages;
3026 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3031 switch (chip->chip_id) {
3032 case KVM_IRQCHIP_PIC_MASTER:
3033 memcpy(&chip->chip.pic,
3034 &pic_irqchip(kvm)->pics[0],
3035 sizeof(struct kvm_pic_state));
3037 case KVM_IRQCHIP_PIC_SLAVE:
3038 memcpy(&chip->chip.pic,
3039 &pic_irqchip(kvm)->pics[1],
3040 sizeof(struct kvm_pic_state));
3042 case KVM_IRQCHIP_IOAPIC:
3043 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3052 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3057 switch (chip->chip_id) {
3058 case KVM_IRQCHIP_PIC_MASTER:
3059 spin_lock(&pic_irqchip(kvm)->lock);
3060 memcpy(&pic_irqchip(kvm)->pics[0],
3062 sizeof(struct kvm_pic_state));
3063 spin_unlock(&pic_irqchip(kvm)->lock);
3065 case KVM_IRQCHIP_PIC_SLAVE:
3066 spin_lock(&pic_irqchip(kvm)->lock);
3067 memcpy(&pic_irqchip(kvm)->pics[1],
3069 sizeof(struct kvm_pic_state));
3070 spin_unlock(&pic_irqchip(kvm)->lock);
3072 case KVM_IRQCHIP_IOAPIC:
3073 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3079 kvm_pic_update_irq(pic_irqchip(kvm));
3083 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3087 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3088 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3089 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3093 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3097 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3098 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3099 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3100 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3104 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3108 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3109 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3110 sizeof(ps->channels));
3111 ps->flags = kvm->arch.vpit->pit_state.flags;
3112 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3113 memset(&ps->reserved, 0, sizeof(ps->reserved));
3117 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3119 int r = 0, start = 0;
3120 u32 prev_legacy, cur_legacy;
3121 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3122 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3123 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3124 if (!prev_legacy && cur_legacy)
3126 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3127 sizeof(kvm->arch.vpit->pit_state.channels));
3128 kvm->arch.vpit->pit_state.flags = ps->flags;
3129 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3130 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3134 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3135 struct kvm_reinject_control *control)
3137 if (!kvm->arch.vpit)
3139 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3140 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3141 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3146 * Get (and clear) the dirty memory log for a memory slot.
3148 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3149 struct kvm_dirty_log *log)
3152 struct kvm_memory_slot *memslot;
3154 unsigned long is_dirty = 0;
3156 mutex_lock(&kvm->slots_lock);
3159 if (log->slot >= KVM_MEMORY_SLOTS)
3162 memslot = &kvm->memslots->memslots[log->slot];
3164 if (!memslot->dirty_bitmap)
3167 n = kvm_dirty_bitmap_bytes(memslot);
3169 for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3170 is_dirty = memslot->dirty_bitmap[i];
3172 /* If nothing is dirty, don't bother messing with page tables. */
3174 struct kvm_memslots *slots, *old_slots;
3175 unsigned long *dirty_bitmap;
3178 dirty_bitmap = vmalloc(n);
3181 memset(dirty_bitmap, 0, n);
3184 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3186 vfree(dirty_bitmap);
3189 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3190 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3192 old_slots = kvm->memslots;
3193 rcu_assign_pointer(kvm->memslots, slots);
3194 synchronize_srcu_expedited(&kvm->srcu);
3195 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3198 spin_lock(&kvm->mmu_lock);
3199 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3200 spin_unlock(&kvm->mmu_lock);
3203 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n)) {
3204 vfree(dirty_bitmap);
3207 vfree(dirty_bitmap);
3210 if (clear_user(log->dirty_bitmap, n))
3216 mutex_unlock(&kvm->slots_lock);
3220 long kvm_arch_vm_ioctl(struct file *filp,
3221 unsigned int ioctl, unsigned long arg)
3223 struct kvm *kvm = filp->private_data;
3224 void __user *argp = (void __user *)arg;
3227 * This union makes it completely explicit to gcc-3.x
3228 * that these two variables' stack usage should be
3229 * combined, not added together.
3232 struct kvm_pit_state ps;
3233 struct kvm_pit_state2 ps2;
3234 struct kvm_pit_config pit_config;
3238 case KVM_SET_TSS_ADDR:
3239 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3243 case KVM_SET_IDENTITY_MAP_ADDR: {
3247 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3249 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3254 case KVM_SET_NR_MMU_PAGES:
3255 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3259 case KVM_GET_NR_MMU_PAGES:
3260 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3262 case KVM_CREATE_IRQCHIP: {
3263 struct kvm_pic *vpic;
3265 mutex_lock(&kvm->lock);
3268 goto create_irqchip_unlock;
3270 vpic = kvm_create_pic(kvm);
3272 r = kvm_ioapic_init(kvm);
3274 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3277 goto create_irqchip_unlock;
3280 goto create_irqchip_unlock;
3282 kvm->arch.vpic = vpic;
3284 r = kvm_setup_default_irq_routing(kvm);
3286 mutex_lock(&kvm->irq_lock);
3287 kvm_ioapic_destroy(kvm);
3288 kvm_destroy_pic(kvm);
3289 mutex_unlock(&kvm->irq_lock);
3291 create_irqchip_unlock:
3292 mutex_unlock(&kvm->lock);
3295 case KVM_CREATE_PIT:
3296 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3298 case KVM_CREATE_PIT2:
3300 if (copy_from_user(&u.pit_config, argp,
3301 sizeof(struct kvm_pit_config)))
3304 mutex_lock(&kvm->slots_lock);
3307 goto create_pit_unlock;
3309 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3313 mutex_unlock(&kvm->slots_lock);
3315 case KVM_IRQ_LINE_STATUS:
3316 case KVM_IRQ_LINE: {
3317 struct kvm_irq_level irq_event;
3320 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3323 if (irqchip_in_kernel(kvm)) {
3325 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3326 irq_event.irq, irq_event.level);
3327 if (ioctl == KVM_IRQ_LINE_STATUS) {
3329 irq_event.status = status;
3330 if (copy_to_user(argp, &irq_event,
3338 case KVM_GET_IRQCHIP: {
3339 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3340 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3346 if (copy_from_user(chip, argp, sizeof *chip))
3347 goto get_irqchip_out;
3349 if (!irqchip_in_kernel(kvm))
3350 goto get_irqchip_out;
3351 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3353 goto get_irqchip_out;
3355 if (copy_to_user(argp, chip, sizeof *chip))
3356 goto get_irqchip_out;
3364 case KVM_SET_IRQCHIP: {
3365 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3366 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3372 if (copy_from_user(chip, argp, sizeof *chip))
3373 goto set_irqchip_out;
3375 if (!irqchip_in_kernel(kvm))
3376 goto set_irqchip_out;
3377 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3379 goto set_irqchip_out;
3389 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3392 if (!kvm->arch.vpit)
3394 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3398 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3405 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3408 if (!kvm->arch.vpit)
3410 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3416 case KVM_GET_PIT2: {
3418 if (!kvm->arch.vpit)
3420 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3424 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3429 case KVM_SET_PIT2: {
3431 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3434 if (!kvm->arch.vpit)
3436 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3442 case KVM_REINJECT_CONTROL: {
3443 struct kvm_reinject_control control;
3445 if (copy_from_user(&control, argp, sizeof(control)))
3447 r = kvm_vm_ioctl_reinject(kvm, &control);
3453 case KVM_XEN_HVM_CONFIG: {
3455 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3456 sizeof(struct kvm_xen_hvm_config)))
3459 if (kvm->arch.xen_hvm_config.flags)
3464 case KVM_SET_CLOCK: {
3465 struct kvm_clock_data user_ns;
3470 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3478 local_irq_disable();
3479 now_ns = get_kernel_ns();
3480 delta = user_ns.clock - now_ns;
3482 kvm->arch.kvmclock_offset = delta;
3485 case KVM_GET_CLOCK: {
3486 struct kvm_clock_data user_ns;
3489 local_irq_disable();
3490 now_ns = get_kernel_ns();
3491 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3494 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3497 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3510 static void kvm_init_msr_list(void)
3515 /* skip the first msrs in the list. KVM-specific */
3516 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3517 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3520 msrs_to_save[j] = msrs_to_save[i];
3523 num_msrs_to_save = j;
3526 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3529 if (vcpu->arch.apic &&
3530 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
3533 return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3536 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3538 if (vcpu->arch.apic &&
3539 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
3542 return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3545 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3546 struct kvm_segment *var, int seg)
3548 kvm_x86_ops->set_segment(vcpu, var, seg);
3551 void kvm_get_segment(struct kvm_vcpu *vcpu,
3552 struct kvm_segment *var, int seg)
3554 kvm_x86_ops->get_segment(vcpu, var, seg);
3557 static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3562 static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3567 BUG_ON(!mmu_is_nested(vcpu));
3569 /* NPT walks are always user-walks */
3570 access |= PFERR_USER_MASK;
3571 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &error);
3572 if (t_gpa == UNMAPPED_GVA)
3573 vcpu->arch.fault.nested = true;
3578 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3580 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3581 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, error);
3584 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3586 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3587 access |= PFERR_FETCH_MASK;
3588 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, error);
3591 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3593 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3594 access |= PFERR_WRITE_MASK;
3595 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, error);
3598 /* uses this to access any guest's mapped memory without checking CPL */
3599 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3601 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, error);
3604 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3605 struct kvm_vcpu *vcpu, u32 access,
3609 int r = X86EMUL_CONTINUE;
3612 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3614 unsigned offset = addr & (PAGE_SIZE-1);
3615 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3618 if (gpa == UNMAPPED_GVA) {
3619 r = X86EMUL_PROPAGATE_FAULT;
3622 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3624 r = X86EMUL_IO_NEEDED;
3636 /* used for instruction fetching */
3637 static int kvm_fetch_guest_virt(gva_t addr, void *val, unsigned int bytes,
3638 struct kvm_vcpu *vcpu, u32 *error)
3640 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3641 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3642 access | PFERR_FETCH_MASK, error);
3645 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3646 struct kvm_vcpu *vcpu, u32 *error)
3648 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3649 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3653 static int kvm_read_guest_virt_system(gva_t addr, void *val, unsigned int bytes,
3654 struct kvm_vcpu *vcpu, u32 *error)
3656 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, error);
3659 static int kvm_write_guest_virt_system(gva_t addr, void *val,
3661 struct kvm_vcpu *vcpu,
3665 int r = X86EMUL_CONTINUE;
3668 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3671 unsigned offset = addr & (PAGE_SIZE-1);
3672 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3675 if (gpa == UNMAPPED_GVA) {
3676 r = X86EMUL_PROPAGATE_FAULT;
3679 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3681 r = X86EMUL_IO_NEEDED;
3693 static int emulator_read_emulated(unsigned long addr,
3696 unsigned int *error_code,
3697 struct kvm_vcpu *vcpu)
3701 if (vcpu->mmio_read_completed) {
3702 memcpy(val, vcpu->mmio_data, bytes);
3703 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3704 vcpu->mmio_phys_addr, *(u64 *)val);
3705 vcpu->mmio_read_completed = 0;
3706 return X86EMUL_CONTINUE;
3709 gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, error_code);
3711 if (gpa == UNMAPPED_GVA)
3712 return X86EMUL_PROPAGATE_FAULT;
3714 /* For APIC access vmexit */
3715 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3718 if (kvm_read_guest_virt(addr, val, bytes, vcpu, NULL)
3719 == X86EMUL_CONTINUE)
3720 return X86EMUL_CONTINUE;
3724 * Is this MMIO handled locally?
3726 if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
3727 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
3728 return X86EMUL_CONTINUE;
3731 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3733 vcpu->mmio_needed = 1;
3734 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3735 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3736 vcpu->run->mmio.len = vcpu->mmio_size = bytes;
3737 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 0;
3739 return X86EMUL_IO_NEEDED;
3742 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3743 const void *val, int bytes)
3747 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3750 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3754 static int emulator_write_emulated_onepage(unsigned long addr,
3757 unsigned int *error_code,
3758 struct kvm_vcpu *vcpu)
3762 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, error_code);
3764 if (gpa == UNMAPPED_GVA)
3765 return X86EMUL_PROPAGATE_FAULT;
3767 /* For APIC access vmexit */
3768 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3771 if (emulator_write_phys(vcpu, gpa, val, bytes))
3772 return X86EMUL_CONTINUE;
3775 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3777 * Is this MMIO handled locally?
3779 if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
3780 return X86EMUL_CONTINUE;
3782 vcpu->mmio_needed = 1;
3783 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3784 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3785 vcpu->run->mmio.len = vcpu->mmio_size = bytes;
3786 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 1;
3787 memcpy(vcpu->run->mmio.data, val, bytes);
3789 return X86EMUL_CONTINUE;
3792 int emulator_write_emulated(unsigned long addr,
3795 unsigned int *error_code,
3796 struct kvm_vcpu *vcpu)
3798 /* Crossing a page boundary? */
3799 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3802 now = -addr & ~PAGE_MASK;
3803 rc = emulator_write_emulated_onepage(addr, val, now, error_code,
3805 if (rc != X86EMUL_CONTINUE)
3811 return emulator_write_emulated_onepage(addr, val, bytes, error_code,
3815 #define CMPXCHG_TYPE(t, ptr, old, new) \
3816 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3818 #ifdef CONFIG_X86_64
3819 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3821 # define CMPXCHG64(ptr, old, new) \
3822 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
3825 static int emulator_cmpxchg_emulated(unsigned long addr,
3829 unsigned int *error_code,
3830 struct kvm_vcpu *vcpu)
3837 /* guests cmpxchg8b have to be emulated atomically */
3838 if (bytes > 8 || (bytes & (bytes - 1)))
3841 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3843 if (gpa == UNMAPPED_GVA ||
3844 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3847 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3850 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3851 if (is_error_page(page)) {
3852 kvm_release_page_clean(page);
3856 kaddr = kmap_atomic(page, KM_USER0);
3857 kaddr += offset_in_page(gpa);
3860 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
3863 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
3866 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
3869 exchanged = CMPXCHG64(kaddr, old, new);
3874 kunmap_atomic(kaddr, KM_USER0);
3875 kvm_release_page_dirty(page);
3878 return X86EMUL_CMPXCHG_FAILED;
3880 kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
3882 return X86EMUL_CONTINUE;
3885 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3887 return emulator_write_emulated(addr, new, bytes, error_code, vcpu);
3890 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3892 /* TODO: String I/O for in kernel device */
3895 if (vcpu->arch.pio.in)
3896 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3897 vcpu->arch.pio.size, pd);
3899 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3900 vcpu->arch.pio.port, vcpu->arch.pio.size,
3906 static int emulator_pio_in_emulated(int size, unsigned short port, void *val,
3907 unsigned int count, struct kvm_vcpu *vcpu)
3909 if (vcpu->arch.pio.count)
3912 trace_kvm_pio(0, port, size, 1);
3914 vcpu->arch.pio.port = port;
3915 vcpu->arch.pio.in = 1;
3916 vcpu->arch.pio.count = count;
3917 vcpu->arch.pio.size = size;
3919 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3921 memcpy(val, vcpu->arch.pio_data, size * count);
3922 vcpu->arch.pio.count = 0;
3926 vcpu->run->exit_reason = KVM_EXIT_IO;
3927 vcpu->run->io.direction = KVM_EXIT_IO_IN;
3928 vcpu->run->io.size = size;
3929 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3930 vcpu->run->io.count = count;
3931 vcpu->run->io.port = port;
3936 static int emulator_pio_out_emulated(int size, unsigned short port,
3937 const void *val, unsigned int count,
3938 struct kvm_vcpu *vcpu)
3940 trace_kvm_pio(1, port, size, 1);
3942 vcpu->arch.pio.port = port;
3943 vcpu->arch.pio.in = 0;
3944 vcpu->arch.pio.count = count;
3945 vcpu->arch.pio.size = size;
3947 memcpy(vcpu->arch.pio_data, val, size * count);
3949 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3950 vcpu->arch.pio.count = 0;
3954 vcpu->run->exit_reason = KVM_EXIT_IO;
3955 vcpu->run->io.direction = KVM_EXIT_IO_OUT;
3956 vcpu->run->io.size = size;
3957 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3958 vcpu->run->io.count = count;
3959 vcpu->run->io.port = port;
3964 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3966 return kvm_x86_ops->get_segment_base(vcpu, seg);
3969 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
3971 kvm_mmu_invlpg(vcpu, address);
3972 return X86EMUL_CONTINUE;
3975 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
3977 if (!need_emulate_wbinvd(vcpu))
3978 return X86EMUL_CONTINUE;
3980 if (kvm_x86_ops->has_wbinvd_exit()) {
3981 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
3982 wbinvd_ipi, NULL, 1);
3983 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
3986 return X86EMUL_CONTINUE;
3988 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
3990 int emulate_clts(struct kvm_vcpu *vcpu)
3992 kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3993 kvm_x86_ops->fpu_activate(vcpu);
3994 return X86EMUL_CONTINUE;
3997 int emulator_get_dr(int dr, unsigned long *dest, struct kvm_vcpu *vcpu)
3999 return _kvm_get_dr(vcpu, dr, dest);
4002 int emulator_set_dr(int dr, unsigned long value, struct kvm_vcpu *vcpu)
4005 return __kvm_set_dr(vcpu, dr, value);
4008 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4010 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4013 static unsigned long emulator_get_cr(int cr, struct kvm_vcpu *vcpu)
4015 unsigned long value;
4019 value = kvm_read_cr0(vcpu);
4022 value = vcpu->arch.cr2;
4025 value = vcpu->arch.cr3;
4028 value = kvm_read_cr4(vcpu);
4031 value = kvm_get_cr8(vcpu);
4034 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4041 static int emulator_set_cr(int cr, unsigned long val, struct kvm_vcpu *vcpu)
4047 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4050 vcpu->arch.cr2 = val;
4053 res = kvm_set_cr3(vcpu, val);
4056 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4059 res = __kvm_set_cr8(vcpu, val & 0xfUL);
4062 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4069 static int emulator_get_cpl(struct kvm_vcpu *vcpu)
4071 return kvm_x86_ops->get_cpl(vcpu);
4074 static void emulator_get_gdt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
4076 kvm_x86_ops->get_gdt(vcpu, dt);
4079 static void emulator_get_idt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
4081 kvm_x86_ops->get_idt(vcpu, dt);
4084 static unsigned long emulator_get_cached_segment_base(int seg,
4085 struct kvm_vcpu *vcpu)
4087 return get_segment_base(vcpu, seg);
4090 static bool emulator_get_cached_descriptor(struct desc_struct *desc, int seg,
4091 struct kvm_vcpu *vcpu)
4093 struct kvm_segment var;
4095 kvm_get_segment(vcpu, &var, seg);
4102 set_desc_limit(desc, var.limit);
4103 set_desc_base(desc, (unsigned long)var.base);
4104 desc->type = var.type;
4106 desc->dpl = var.dpl;
4107 desc->p = var.present;
4108 desc->avl = var.avl;
4116 static void emulator_set_cached_descriptor(struct desc_struct *desc, int seg,
4117 struct kvm_vcpu *vcpu)
4119 struct kvm_segment var;
4121 /* needed to preserve selector */
4122 kvm_get_segment(vcpu, &var, seg);
4124 var.base = get_desc_base(desc);
4125 var.limit = get_desc_limit(desc);
4127 var.limit = (var.limit << 12) | 0xfff;
4128 var.type = desc->type;
4129 var.present = desc->p;
4130 var.dpl = desc->dpl;
4135 var.avl = desc->avl;
4136 var.present = desc->p;
4137 var.unusable = !var.present;
4140 kvm_set_segment(vcpu, &var, seg);
4144 static u16 emulator_get_segment_selector(int seg, struct kvm_vcpu *vcpu)
4146 struct kvm_segment kvm_seg;
4148 kvm_get_segment(vcpu, &kvm_seg, seg);
4149 return kvm_seg.selector;
4152 static void emulator_set_segment_selector(u16 sel, int seg,
4153 struct kvm_vcpu *vcpu)
4155 struct kvm_segment kvm_seg;
4157 kvm_get_segment(vcpu, &kvm_seg, seg);
4158 kvm_seg.selector = sel;
4159 kvm_set_segment(vcpu, &kvm_seg, seg);
4162 static struct x86_emulate_ops emulate_ops = {
4163 .read_std = kvm_read_guest_virt_system,
4164 .write_std = kvm_write_guest_virt_system,
4165 .fetch = kvm_fetch_guest_virt,
4166 .read_emulated = emulator_read_emulated,
4167 .write_emulated = emulator_write_emulated,
4168 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4169 .pio_in_emulated = emulator_pio_in_emulated,
4170 .pio_out_emulated = emulator_pio_out_emulated,
4171 .get_cached_descriptor = emulator_get_cached_descriptor,
4172 .set_cached_descriptor = emulator_set_cached_descriptor,
4173 .get_segment_selector = emulator_get_segment_selector,
4174 .set_segment_selector = emulator_set_segment_selector,
4175 .get_cached_segment_base = emulator_get_cached_segment_base,
4176 .get_gdt = emulator_get_gdt,
4177 .get_idt = emulator_get_idt,
4178 .get_cr = emulator_get_cr,
4179 .set_cr = emulator_set_cr,
4180 .cpl = emulator_get_cpl,
4181 .get_dr = emulator_get_dr,
4182 .set_dr = emulator_set_dr,
4183 .set_msr = kvm_set_msr,
4184 .get_msr = kvm_get_msr,
4187 static void cache_all_regs(struct kvm_vcpu *vcpu)
4189 kvm_register_read(vcpu, VCPU_REGS_RAX);
4190 kvm_register_read(vcpu, VCPU_REGS_RSP);
4191 kvm_register_read(vcpu, VCPU_REGS_RIP);
4192 vcpu->arch.regs_dirty = ~0;
4195 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4197 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4199 * an sti; sti; sequence only disable interrupts for the first
4200 * instruction. So, if the last instruction, be it emulated or
4201 * not, left the system with the INT_STI flag enabled, it
4202 * means that the last instruction is an sti. We should not
4203 * leave the flag on in this case. The same goes for mov ss
4205 if (!(int_shadow & mask))
4206 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4209 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4211 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4212 if (ctxt->exception == PF_VECTOR)
4213 kvm_propagate_fault(vcpu);
4214 else if (ctxt->error_code_valid)
4215 kvm_queue_exception_e(vcpu, ctxt->exception, ctxt->error_code);
4217 kvm_queue_exception(vcpu, ctxt->exception);
4220 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4222 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4225 cache_all_regs(vcpu);
4227 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4229 vcpu->arch.emulate_ctxt.vcpu = vcpu;
4230 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
4231 vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
4232 vcpu->arch.emulate_ctxt.mode =
4233 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4234 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
4235 ? X86EMUL_MODE_VM86 : cs_l
4236 ? X86EMUL_MODE_PROT64 : cs_db
4237 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
4238 memset(c, 0, sizeof(struct decode_cache));
4239 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4242 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq)
4244 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4247 init_emulate_ctxt(vcpu);
4249 vcpu->arch.emulate_ctxt.decode.op_bytes = 2;
4250 vcpu->arch.emulate_ctxt.decode.ad_bytes = 2;
4251 vcpu->arch.emulate_ctxt.decode.eip = vcpu->arch.emulate_ctxt.eip;
4252 ret = emulate_int_real(&vcpu->arch.emulate_ctxt, &emulate_ops, irq);
4254 if (ret != X86EMUL_CONTINUE)
4255 return EMULATE_FAIL;
4257 vcpu->arch.emulate_ctxt.eip = c->eip;
4258 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4259 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4260 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4262 if (irq == NMI_VECTOR)
4263 vcpu->arch.nmi_pending = false;
4265 vcpu->arch.interrupt.pending = false;
4267 return EMULATE_DONE;
4269 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4271 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4273 ++vcpu->stat.insn_emulation_fail;
4274 trace_kvm_emulate_insn_failed(vcpu);
4275 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4276 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4277 vcpu->run->internal.ndata = 0;
4278 kvm_queue_exception(vcpu, UD_VECTOR);
4279 return EMULATE_FAIL;
4282 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4290 * if emulation was due to access to shadowed page table
4291 * and it failed try to unshadow page and re-entetr the
4292 * guest to let CPU execute the instruction.
4294 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4297 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4299 if (gpa == UNMAPPED_GVA)
4300 return true; /* let cpu generate fault */
4302 if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4308 int emulate_instruction(struct kvm_vcpu *vcpu,
4314 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4316 kvm_clear_exception_queue(vcpu);
4317 vcpu->arch.mmio_fault_cr2 = cr2;
4319 * TODO: fix emulate.c to use guest_read/write_register
4320 * instead of direct ->regs accesses, can save hundred cycles
4321 * on Intel for instructions that don't read/change RSP, for
4324 cache_all_regs(vcpu);
4326 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4327 init_emulate_ctxt(vcpu);
4328 vcpu->arch.emulate_ctxt.interruptibility = 0;
4329 vcpu->arch.emulate_ctxt.exception = -1;
4330 vcpu->arch.emulate_ctxt.perm_ok = false;
4332 r = x86_decode_insn(&vcpu->arch.emulate_ctxt);
4333 if (r == X86EMUL_PROPAGATE_FAULT)
4336 trace_kvm_emulate_insn_start(vcpu);
4338 /* Only allow emulation of specific instructions on #UD
4339 * (namely VMMCALL, sysenter, sysexit, syscall)*/
4340 if (emulation_type & EMULTYPE_TRAP_UD) {
4342 return EMULATE_FAIL;
4344 case 0x01: /* VMMCALL */
4345 if (c->modrm_mod != 3 || c->modrm_rm != 1)
4346 return EMULATE_FAIL;
4348 case 0x34: /* sysenter */
4349 case 0x35: /* sysexit */
4350 if (c->modrm_mod != 0 || c->modrm_rm != 0)
4351 return EMULATE_FAIL;
4353 case 0x05: /* syscall */
4354 if (c->modrm_mod != 0 || c->modrm_rm != 0)
4355 return EMULATE_FAIL;
4358 return EMULATE_FAIL;
4361 if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
4362 return EMULATE_FAIL;
4365 ++vcpu->stat.insn_emulation;
4367 if (reexecute_instruction(vcpu, cr2))
4368 return EMULATE_DONE;
4369 if (emulation_type & EMULTYPE_SKIP)
4370 return EMULATE_FAIL;
4371 return handle_emulation_failure(vcpu);
4375 if (emulation_type & EMULTYPE_SKIP) {
4376 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
4377 return EMULATE_DONE;
4380 /* this is needed for vmware backdor interface to work since it
4381 changes registers values during IO operation */
4382 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4385 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt);
4387 if (r == EMULATION_FAILED) {
4388 if (reexecute_instruction(vcpu, cr2))
4389 return EMULATE_DONE;
4391 return handle_emulation_failure(vcpu);
4395 if (vcpu->arch.emulate_ctxt.exception >= 0) {
4396 inject_emulated_exception(vcpu);
4398 } else if (vcpu->arch.pio.count) {
4399 if (!vcpu->arch.pio.in)
4400 vcpu->arch.pio.count = 0;
4401 r = EMULATE_DO_MMIO;
4402 } else if (vcpu->mmio_needed) {
4403 if (vcpu->mmio_is_write)
4404 vcpu->mmio_needed = 0;
4405 r = EMULATE_DO_MMIO;
4406 } else if (r == EMULATION_RESTART)
4411 toggle_interruptibility(vcpu, vcpu->arch.emulate_ctxt.interruptibility);
4412 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4413 kvm_make_request(KVM_REQ_EVENT, vcpu);
4414 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4415 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4419 EXPORT_SYMBOL_GPL(emulate_instruction);
4421 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4423 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4424 int ret = emulator_pio_out_emulated(size, port, &val, 1, vcpu);
4425 /* do not return to emulator after return from userspace */
4426 vcpu->arch.pio.count = 0;
4429 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4431 static void tsc_bad(void *info)
4433 __get_cpu_var(cpu_tsc_khz) = 0;
4436 static void tsc_khz_changed(void *data)
4438 struct cpufreq_freqs *freq = data;
4439 unsigned long khz = 0;
4443 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4444 khz = cpufreq_quick_get(raw_smp_processor_id());
4447 __get_cpu_var(cpu_tsc_khz) = khz;
4450 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4453 struct cpufreq_freqs *freq = data;
4455 struct kvm_vcpu *vcpu;
4456 int i, send_ipi = 0;
4459 * We allow guests to temporarily run on slowing clocks,
4460 * provided we notify them after, or to run on accelerating
4461 * clocks, provided we notify them before. Thus time never
4464 * However, we have a problem. We can't atomically update
4465 * the frequency of a given CPU from this function; it is
4466 * merely a notifier, which can be called from any CPU.
4467 * Changing the TSC frequency at arbitrary points in time
4468 * requires a recomputation of local variables related to
4469 * the TSC for each VCPU. We must flag these local variables
4470 * to be updated and be sure the update takes place with the
4471 * new frequency before any guests proceed.
4473 * Unfortunately, the combination of hotplug CPU and frequency
4474 * change creates an intractable locking scenario; the order
4475 * of when these callouts happen is undefined with respect to
4476 * CPU hotplug, and they can race with each other. As such,
4477 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4478 * undefined; you can actually have a CPU frequency change take
4479 * place in between the computation of X and the setting of the
4480 * variable. To protect against this problem, all updates of
4481 * the per_cpu tsc_khz variable are done in an interrupt
4482 * protected IPI, and all callers wishing to update the value
4483 * must wait for a synchronous IPI to complete (which is trivial
4484 * if the caller is on the CPU already). This establishes the
4485 * necessary total order on variable updates.
4487 * Note that because a guest time update may take place
4488 * anytime after the setting of the VCPU's request bit, the
4489 * correct TSC value must be set before the request. However,
4490 * to ensure the update actually makes it to any guest which
4491 * starts running in hardware virtualization between the set
4492 * and the acquisition of the spinlock, we must also ping the
4493 * CPU after setting the request bit.
4497 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4499 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4502 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4504 spin_lock(&kvm_lock);
4505 list_for_each_entry(kvm, &vm_list, vm_list) {
4506 kvm_for_each_vcpu(i, vcpu, kvm) {
4507 if (vcpu->cpu != freq->cpu)
4509 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4510 if (vcpu->cpu != smp_processor_id())
4514 spin_unlock(&kvm_lock);
4516 if (freq->old < freq->new && send_ipi) {
4518 * We upscale the frequency. Must make the guest
4519 * doesn't see old kvmclock values while running with
4520 * the new frequency, otherwise we risk the guest sees
4521 * time go backwards.
4523 * In case we update the frequency for another cpu
4524 * (which might be in guest context) send an interrupt
4525 * to kick the cpu out of guest context. Next time
4526 * guest context is entered kvmclock will be updated,
4527 * so the guest will not see stale values.
4529 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4534 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4535 .notifier_call = kvmclock_cpufreq_notifier
4538 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4539 unsigned long action, void *hcpu)
4541 unsigned int cpu = (unsigned long)hcpu;
4545 case CPU_DOWN_FAILED:
4546 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4548 case CPU_DOWN_PREPARE:
4549 smp_call_function_single(cpu, tsc_bad, NULL, 1);
4555 static struct notifier_block kvmclock_cpu_notifier_block = {
4556 .notifier_call = kvmclock_cpu_notifier,
4557 .priority = -INT_MAX
4560 static void kvm_timer_init(void)
4564 max_tsc_khz = tsc_khz;
4565 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4566 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4567 #ifdef CONFIG_CPU_FREQ
4568 struct cpufreq_policy policy;
4569 memset(&policy, 0, sizeof(policy));
4570 cpufreq_get_policy(&policy, get_cpu());
4571 if (policy.cpuinfo.max_freq)
4572 max_tsc_khz = policy.cpuinfo.max_freq;
4574 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4575 CPUFREQ_TRANSITION_NOTIFIER);
4577 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
4578 for_each_online_cpu(cpu)
4579 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4582 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4584 static int kvm_is_in_guest(void)
4586 return percpu_read(current_vcpu) != NULL;
4589 static int kvm_is_user_mode(void)
4593 if (percpu_read(current_vcpu))
4594 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
4596 return user_mode != 0;
4599 static unsigned long kvm_get_guest_ip(void)
4601 unsigned long ip = 0;
4603 if (percpu_read(current_vcpu))
4604 ip = kvm_rip_read(percpu_read(current_vcpu));
4609 static struct perf_guest_info_callbacks kvm_guest_cbs = {
4610 .is_in_guest = kvm_is_in_guest,
4611 .is_user_mode = kvm_is_user_mode,
4612 .get_guest_ip = kvm_get_guest_ip,
4615 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4617 percpu_write(current_vcpu, vcpu);
4619 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4621 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4623 percpu_write(current_vcpu, NULL);
4625 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4627 int kvm_arch_init(void *opaque)
4630 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4633 printk(KERN_ERR "kvm: already loaded the other module\n");
4638 if (!ops->cpu_has_kvm_support()) {
4639 printk(KERN_ERR "kvm: no hardware support\n");
4643 if (ops->disabled_by_bios()) {
4644 printk(KERN_ERR "kvm: disabled by bios\n");
4649 r = kvm_mmu_module_init();
4653 kvm_init_msr_list();
4656 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
4657 kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
4658 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4659 PT_DIRTY_MASK, PT64_NX_MASK, 0);
4663 perf_register_guest_info_callbacks(&kvm_guest_cbs);
4666 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
4674 void kvm_arch_exit(void)
4676 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4678 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4679 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4680 CPUFREQ_TRANSITION_NOTIFIER);
4681 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4683 kvm_mmu_module_exit();
4686 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4688 ++vcpu->stat.halt_exits;
4689 if (irqchip_in_kernel(vcpu->kvm)) {
4690 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
4693 vcpu->run->exit_reason = KVM_EXIT_HLT;
4697 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4699 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
4702 if (is_long_mode(vcpu))
4705 return a0 | ((gpa_t)a1 << 32);
4708 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4710 u64 param, ingpa, outgpa, ret;
4711 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4712 bool fast, longmode;
4716 * hypercall generates UD from non zero cpl and real mode
4719 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
4720 kvm_queue_exception(vcpu, UD_VECTOR);
4724 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4725 longmode = is_long_mode(vcpu) && cs_l == 1;
4728 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4729 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4730 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4731 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4732 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4733 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
4735 #ifdef CONFIG_X86_64
4737 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4738 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4739 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4743 code = param & 0xffff;
4744 fast = (param >> 16) & 0x1;
4745 rep_cnt = (param >> 32) & 0xfff;
4746 rep_idx = (param >> 48) & 0xfff;
4748 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
4751 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
4752 kvm_vcpu_on_spin(vcpu);
4755 res = HV_STATUS_INVALID_HYPERCALL_CODE;
4759 ret = res | (((u64)rep_done & 0xfff) << 32);
4761 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4763 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
4764 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
4770 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
4772 unsigned long nr, a0, a1, a2, a3, ret;
4775 if (kvm_hv_hypercall_enabled(vcpu->kvm))
4776 return kvm_hv_hypercall(vcpu);
4778 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
4779 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
4780 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
4781 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
4782 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
4784 trace_kvm_hypercall(nr, a0, a1, a2, a3);
4786 if (!is_long_mode(vcpu)) {
4794 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
4800 case KVM_HC_VAPIC_POLL_IRQ:
4804 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
4811 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4812 ++vcpu->stat.hypercalls;
4815 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
4817 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
4819 char instruction[3];
4820 unsigned long rip = kvm_rip_read(vcpu);
4823 * Blow out the MMU to ensure that no other VCPU has an active mapping
4824 * to ensure that the updated hypercall appears atomically across all
4827 kvm_mmu_zap_all(vcpu->kvm);
4829 kvm_x86_ops->patch_hypercall(vcpu, instruction);
4831 return emulator_write_emulated(rip, instruction, 3, NULL, vcpu);
4834 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4836 struct desc_ptr dt = { limit, base };
4838 kvm_x86_ops->set_gdt(vcpu, &dt);
4841 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4843 struct desc_ptr dt = { limit, base };
4845 kvm_x86_ops->set_idt(vcpu, &dt);
4848 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
4850 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
4851 int j, nent = vcpu->arch.cpuid_nent;
4853 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
4854 /* when no next entry is found, the current entry[i] is reselected */
4855 for (j = i + 1; ; j = (j + 1) % nent) {
4856 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
4857 if (ej->function == e->function) {
4858 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
4862 return 0; /* silence gcc, even though control never reaches here */
4865 /* find an entry with matching function, matching index (if needed), and that
4866 * should be read next (if it's stateful) */
4867 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
4868 u32 function, u32 index)
4870 if (e->function != function)
4872 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
4874 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
4875 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
4880 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
4881 u32 function, u32 index)
4884 struct kvm_cpuid_entry2 *best = NULL;
4886 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
4887 struct kvm_cpuid_entry2 *e;
4889 e = &vcpu->arch.cpuid_entries[i];
4890 if (is_matching_cpuid_entry(e, function, index)) {
4891 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
4892 move_to_next_stateful_cpuid_entry(vcpu, i);
4897 * Both basic or both extended?
4899 if (((e->function ^ function) & 0x80000000) == 0)
4900 if (!best || e->function > best->function)
4905 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
4907 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
4909 struct kvm_cpuid_entry2 *best;
4911 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
4912 if (!best || best->eax < 0x80000008)
4914 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
4916 return best->eax & 0xff;
4921 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
4923 u32 function, index;
4924 struct kvm_cpuid_entry2 *best;
4926 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
4927 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4928 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
4929 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
4930 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
4931 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
4932 best = kvm_find_cpuid_entry(vcpu, function, index);
4934 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
4935 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
4936 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
4937 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
4939 kvm_x86_ops->skip_emulated_instruction(vcpu);
4940 trace_kvm_cpuid(function,
4941 kvm_register_read(vcpu, VCPU_REGS_RAX),
4942 kvm_register_read(vcpu, VCPU_REGS_RBX),
4943 kvm_register_read(vcpu, VCPU_REGS_RCX),
4944 kvm_register_read(vcpu, VCPU_REGS_RDX));
4946 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
4949 * Check if userspace requested an interrupt window, and that the
4950 * interrupt window is open.
4952 * No need to exit to userspace if we already have an interrupt queued.
4954 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
4956 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
4957 vcpu->run->request_interrupt_window &&
4958 kvm_arch_interrupt_allowed(vcpu));
4961 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
4963 struct kvm_run *kvm_run = vcpu->run;
4965 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
4966 kvm_run->cr8 = kvm_get_cr8(vcpu);
4967 kvm_run->apic_base = kvm_get_apic_base(vcpu);
4968 if (irqchip_in_kernel(vcpu->kvm))
4969 kvm_run->ready_for_interrupt_injection = 1;
4971 kvm_run->ready_for_interrupt_injection =
4972 kvm_arch_interrupt_allowed(vcpu) &&
4973 !kvm_cpu_has_interrupt(vcpu) &&
4974 !kvm_event_needs_reinjection(vcpu);
4977 static void vapic_enter(struct kvm_vcpu *vcpu)
4979 struct kvm_lapic *apic = vcpu->arch.apic;
4982 if (!apic || !apic->vapic_addr)
4985 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4987 vcpu->arch.apic->vapic_page = page;
4990 static void vapic_exit(struct kvm_vcpu *vcpu)
4992 struct kvm_lapic *apic = vcpu->arch.apic;
4995 if (!apic || !apic->vapic_addr)
4998 idx = srcu_read_lock(&vcpu->kvm->srcu);
4999 kvm_release_page_dirty(apic->vapic_page);
5000 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5001 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5004 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5008 if (!kvm_x86_ops->update_cr8_intercept)
5011 if (!vcpu->arch.apic)
5014 if (!vcpu->arch.apic->vapic_addr)
5015 max_irr = kvm_lapic_find_highest_irr(vcpu);
5022 tpr = kvm_lapic_get_cr8(vcpu);
5024 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5027 static void inject_pending_event(struct kvm_vcpu *vcpu)
5029 /* try to reinject previous events if any */
5030 if (vcpu->arch.exception.pending) {
5031 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5032 vcpu->arch.exception.has_error_code,
5033 vcpu->arch.exception.error_code);
5034 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5035 vcpu->arch.exception.has_error_code,
5036 vcpu->arch.exception.error_code,
5037 vcpu->arch.exception.reinject);
5041 if (vcpu->arch.nmi_injected) {
5042 kvm_x86_ops->set_nmi(vcpu);
5046 if (vcpu->arch.interrupt.pending) {
5047 kvm_x86_ops->set_irq(vcpu);
5051 /* try to inject new event if pending */
5052 if (vcpu->arch.nmi_pending) {
5053 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5054 vcpu->arch.nmi_pending = false;
5055 vcpu->arch.nmi_injected = true;
5056 kvm_x86_ops->set_nmi(vcpu);
5058 } else if (kvm_cpu_has_interrupt(vcpu)) {
5059 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5060 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5062 kvm_x86_ops->set_irq(vcpu);
5067 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5069 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5070 !vcpu->guest_xcr0_loaded) {
5071 /* kvm_set_xcr() also depends on this */
5072 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5073 vcpu->guest_xcr0_loaded = 1;
5077 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5079 if (vcpu->guest_xcr0_loaded) {
5080 if (vcpu->arch.xcr0 != host_xcr0)
5081 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5082 vcpu->guest_xcr0_loaded = 0;
5086 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5089 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5090 vcpu->run->request_interrupt_window;
5092 if (vcpu->requests) {
5093 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5094 kvm_mmu_unload(vcpu);
5095 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5096 __kvm_migrate_timers(vcpu);
5097 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5098 r = kvm_guest_time_update(vcpu);
5102 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5103 kvm_mmu_sync_roots(vcpu);
5104 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5105 kvm_x86_ops->tlb_flush(vcpu);
5106 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5107 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5111 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5112 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5116 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5117 vcpu->fpu_active = 0;
5118 kvm_x86_ops->fpu_deactivate(vcpu);
5122 r = kvm_mmu_reload(vcpu);
5126 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5127 inject_pending_event(vcpu);
5129 /* enable NMI/IRQ window open exits if needed */
5130 if (vcpu->arch.nmi_pending)
5131 kvm_x86_ops->enable_nmi_window(vcpu);
5132 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5133 kvm_x86_ops->enable_irq_window(vcpu);
5135 if (kvm_lapic_enabled(vcpu)) {
5136 update_cr8_intercept(vcpu);
5137 kvm_lapic_sync_to_vapic(vcpu);
5143 kvm_x86_ops->prepare_guest_switch(vcpu);
5144 if (vcpu->fpu_active)
5145 kvm_load_guest_fpu(vcpu);
5146 kvm_load_guest_xcr0(vcpu);
5148 atomic_set(&vcpu->guest_mode, 1);
5151 local_irq_disable();
5153 if (!atomic_read(&vcpu->guest_mode) || vcpu->requests
5154 || need_resched() || signal_pending(current)) {
5155 atomic_set(&vcpu->guest_mode, 0);
5159 kvm_x86_ops->cancel_injection(vcpu);
5164 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5168 if (unlikely(vcpu->arch.switch_db_regs)) {
5170 set_debugreg(vcpu->arch.eff_db[0], 0);
5171 set_debugreg(vcpu->arch.eff_db[1], 1);
5172 set_debugreg(vcpu->arch.eff_db[2], 2);
5173 set_debugreg(vcpu->arch.eff_db[3], 3);
5176 trace_kvm_entry(vcpu->vcpu_id);
5177 kvm_x86_ops->run(vcpu);
5180 * If the guest has used debug registers, at least dr7
5181 * will be disabled while returning to the host.
5182 * If we don't have active breakpoints in the host, we don't
5183 * care about the messed up debug address registers. But if
5184 * we have some of them active, restore the old state.
5186 if (hw_breakpoint_active())
5187 hw_breakpoint_restore();
5189 kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
5191 atomic_set(&vcpu->guest_mode, 0);
5198 * We must have an instruction between local_irq_enable() and
5199 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5200 * the interrupt shadow. The stat.exits increment will do nicely.
5201 * But we need to prevent reordering, hence this barrier():
5209 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5212 * Profile KVM exit RIPs:
5214 if (unlikely(prof_on == KVM_PROFILING)) {
5215 unsigned long rip = kvm_rip_read(vcpu);
5216 profile_hit(KVM_PROFILING, (void *)rip);
5220 kvm_lapic_sync_from_vapic(vcpu);
5222 r = kvm_x86_ops->handle_exit(vcpu);
5228 static int __vcpu_run(struct kvm_vcpu *vcpu)
5231 struct kvm *kvm = vcpu->kvm;
5233 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5234 pr_debug("vcpu %d received sipi with vector # %x\n",
5235 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5236 kvm_lapic_reset(vcpu);
5237 r = kvm_arch_vcpu_reset(vcpu);
5240 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5243 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5248 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
5249 r = vcpu_enter_guest(vcpu);
5251 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5252 kvm_vcpu_block(vcpu);
5253 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5254 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5256 switch(vcpu->arch.mp_state) {
5257 case KVM_MP_STATE_HALTED:
5258 vcpu->arch.mp_state =
5259 KVM_MP_STATE_RUNNABLE;
5260 case KVM_MP_STATE_RUNNABLE:
5262 case KVM_MP_STATE_SIPI_RECEIVED:
5273 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5274 if (kvm_cpu_has_pending_timer(vcpu))
5275 kvm_inject_pending_timer_irqs(vcpu);
5277 if (dm_request_for_irq_injection(vcpu)) {
5279 vcpu->run->exit_reason = KVM_EXIT_INTR;
5280 ++vcpu->stat.request_irq_exits;
5282 if (signal_pending(current)) {
5284 vcpu->run->exit_reason = KVM_EXIT_INTR;
5285 ++vcpu->stat.signal_exits;
5287 if (need_resched()) {
5288 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5290 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5294 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5301 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5306 if (vcpu->sigset_active)
5307 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5309 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5310 kvm_vcpu_block(vcpu);
5311 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5316 /* re-sync apic's tpr */
5317 if (!irqchip_in_kernel(vcpu->kvm))
5318 kvm_set_cr8(vcpu, kvm_run->cr8);
5320 if (vcpu->arch.pio.count || vcpu->mmio_needed) {
5321 if (vcpu->mmio_needed) {
5322 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
5323 vcpu->mmio_read_completed = 1;
5324 vcpu->mmio_needed = 0;
5326 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5327 r = emulate_instruction(vcpu, 0, 0, EMULTYPE_NO_DECODE);
5328 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5329 if (r != EMULATE_DONE) {
5334 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5335 kvm_register_write(vcpu, VCPU_REGS_RAX,
5336 kvm_run->hypercall.ret);
5338 r = __vcpu_run(vcpu);
5341 post_kvm_run_save(vcpu);
5342 if (vcpu->sigset_active)
5343 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5348 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5350 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5351 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5352 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5353 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5354 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5355 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5356 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5357 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5358 #ifdef CONFIG_X86_64
5359 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5360 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5361 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5362 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5363 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5364 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5365 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5366 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5369 regs->rip = kvm_rip_read(vcpu);
5370 regs->rflags = kvm_get_rflags(vcpu);
5375 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5377 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5378 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5379 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5380 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5381 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5382 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5383 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5384 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5385 #ifdef CONFIG_X86_64
5386 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5387 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5388 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5389 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5390 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5391 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5392 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5393 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5396 kvm_rip_write(vcpu, regs->rip);
5397 kvm_set_rflags(vcpu, regs->rflags);
5399 vcpu->arch.exception.pending = false;
5401 kvm_make_request(KVM_REQ_EVENT, vcpu);
5406 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5408 struct kvm_segment cs;
5410 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
5414 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5416 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5417 struct kvm_sregs *sregs)
5421 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5422 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5423 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5424 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5425 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5426 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5428 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5429 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5431 kvm_x86_ops->get_idt(vcpu, &dt);
5432 sregs->idt.limit = dt.size;
5433 sregs->idt.base = dt.address;
5434 kvm_x86_ops->get_gdt(vcpu, &dt);
5435 sregs->gdt.limit = dt.size;
5436 sregs->gdt.base = dt.address;
5438 sregs->cr0 = kvm_read_cr0(vcpu);
5439 sregs->cr2 = vcpu->arch.cr2;
5440 sregs->cr3 = vcpu->arch.cr3;
5441 sregs->cr4 = kvm_read_cr4(vcpu);
5442 sregs->cr8 = kvm_get_cr8(vcpu);
5443 sregs->efer = vcpu->arch.efer;
5444 sregs->apic_base = kvm_get_apic_base(vcpu);
5446 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
5448 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
5449 set_bit(vcpu->arch.interrupt.nr,
5450 (unsigned long *)sregs->interrupt_bitmap);
5455 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
5456 struct kvm_mp_state *mp_state)
5458 mp_state->mp_state = vcpu->arch.mp_state;
5462 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
5463 struct kvm_mp_state *mp_state)
5465 vcpu->arch.mp_state = mp_state->mp_state;
5466 kvm_make_request(KVM_REQ_EVENT, vcpu);
5470 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
5471 bool has_error_code, u32 error_code)
5473 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5476 init_emulate_ctxt(vcpu);
5478 ret = emulator_task_switch(&vcpu->arch.emulate_ctxt,
5479 tss_selector, reason, has_error_code,
5483 return EMULATE_FAIL;
5485 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5486 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
5487 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
5488 kvm_make_request(KVM_REQ_EVENT, vcpu);
5489 return EMULATE_DONE;
5491 EXPORT_SYMBOL_GPL(kvm_task_switch);
5493 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5494 struct kvm_sregs *sregs)
5496 int mmu_reset_needed = 0;
5497 int pending_vec, max_bits;
5500 dt.size = sregs->idt.limit;
5501 dt.address = sregs->idt.base;
5502 kvm_x86_ops->set_idt(vcpu, &dt);
5503 dt.size = sregs->gdt.limit;
5504 dt.address = sregs->gdt.base;
5505 kvm_x86_ops->set_gdt(vcpu, &dt);
5507 vcpu->arch.cr2 = sregs->cr2;
5508 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
5509 vcpu->arch.cr3 = sregs->cr3;
5511 kvm_set_cr8(vcpu, sregs->cr8);
5513 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5514 kvm_x86_ops->set_efer(vcpu, sregs->efer);
5515 kvm_set_apic_base(vcpu, sregs->apic_base);
5517 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5518 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5519 vcpu->arch.cr0 = sregs->cr0;
5521 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5522 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5523 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5524 load_pdptrs(vcpu, vcpu->arch.walk_mmu, vcpu->arch.cr3);
5525 mmu_reset_needed = 1;
5528 if (mmu_reset_needed)
5529 kvm_mmu_reset_context(vcpu);
5531 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5532 pending_vec = find_first_bit(
5533 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5534 if (pending_vec < max_bits) {
5535 kvm_queue_interrupt(vcpu, pending_vec, false);
5536 pr_debug("Set back pending irq %d\n", pending_vec);
5537 if (irqchip_in_kernel(vcpu->kvm))
5538 kvm_pic_clear_isr_ack(vcpu->kvm);
5541 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5542 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5543 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5544 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5545 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5546 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5548 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5549 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5551 update_cr8_intercept(vcpu);
5553 /* Older userspace won't unhalt the vcpu on reset. */
5554 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5555 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5557 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5559 kvm_make_request(KVM_REQ_EVENT, vcpu);
5564 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5565 struct kvm_guest_debug *dbg)
5567 unsigned long rflags;
5570 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5572 if (vcpu->arch.exception.pending)
5574 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5575 kvm_queue_exception(vcpu, DB_VECTOR);
5577 kvm_queue_exception(vcpu, BP_VECTOR);
5581 * Read rflags as long as potentially injected trace flags are still
5584 rflags = kvm_get_rflags(vcpu);
5586 vcpu->guest_debug = dbg->control;
5587 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5588 vcpu->guest_debug = 0;
5590 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5591 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5592 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5593 vcpu->arch.switch_db_regs =
5594 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5596 for (i = 0; i < KVM_NR_DB_REGS; i++)
5597 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5598 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5601 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5602 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5603 get_segment_base(vcpu, VCPU_SREG_CS);
5606 * Trigger an rflags update that will inject or remove the trace
5609 kvm_set_rflags(vcpu, rflags);
5611 kvm_x86_ops->set_guest_debug(vcpu, dbg);
5621 * Translate a guest virtual address to a guest physical address.
5623 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5624 struct kvm_translation *tr)
5626 unsigned long vaddr = tr->linear_address;
5630 idx = srcu_read_lock(&vcpu->kvm->srcu);
5631 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5632 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5633 tr->physical_address = gpa;
5634 tr->valid = gpa != UNMAPPED_GVA;
5641 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5643 struct i387_fxsave_struct *fxsave =
5644 &vcpu->arch.guest_fpu.state->fxsave;
5646 memcpy(fpu->fpr, fxsave->st_space, 128);
5647 fpu->fcw = fxsave->cwd;
5648 fpu->fsw = fxsave->swd;
5649 fpu->ftwx = fxsave->twd;
5650 fpu->last_opcode = fxsave->fop;
5651 fpu->last_ip = fxsave->rip;
5652 fpu->last_dp = fxsave->rdp;
5653 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5658 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5660 struct i387_fxsave_struct *fxsave =
5661 &vcpu->arch.guest_fpu.state->fxsave;
5663 memcpy(fxsave->st_space, fpu->fpr, 128);
5664 fxsave->cwd = fpu->fcw;
5665 fxsave->swd = fpu->fsw;
5666 fxsave->twd = fpu->ftwx;
5667 fxsave->fop = fpu->last_opcode;
5668 fxsave->rip = fpu->last_ip;
5669 fxsave->rdp = fpu->last_dp;
5670 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5675 int fx_init(struct kvm_vcpu *vcpu)
5679 err = fpu_alloc(&vcpu->arch.guest_fpu);
5683 fpu_finit(&vcpu->arch.guest_fpu);
5686 * Ensure guest xcr0 is valid for loading
5688 vcpu->arch.xcr0 = XSTATE_FP;
5690 vcpu->arch.cr0 |= X86_CR0_ET;
5694 EXPORT_SYMBOL_GPL(fx_init);
5696 static void fx_free(struct kvm_vcpu *vcpu)
5698 fpu_free(&vcpu->arch.guest_fpu);
5701 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5703 if (vcpu->guest_fpu_loaded)
5707 * Restore all possible states in the guest,
5708 * and assume host would use all available bits.
5709 * Guest xcr0 would be loaded later.
5711 kvm_put_guest_xcr0(vcpu);
5712 vcpu->guest_fpu_loaded = 1;
5713 unlazy_fpu(current);
5714 fpu_restore_checking(&vcpu->arch.guest_fpu);
5718 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5720 kvm_put_guest_xcr0(vcpu);
5722 if (!vcpu->guest_fpu_loaded)
5725 vcpu->guest_fpu_loaded = 0;
5726 fpu_save_init(&vcpu->arch.guest_fpu);
5727 ++vcpu->stat.fpu_reload;
5728 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
5732 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5734 if (vcpu->arch.time_page) {
5735 kvm_release_page_dirty(vcpu->arch.time_page);
5736 vcpu->arch.time_page = NULL;
5739 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
5741 kvm_x86_ops->vcpu_free(vcpu);
5744 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5747 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
5748 printk_once(KERN_WARNING
5749 "kvm: SMP vm created on host with unstable TSC; "
5750 "guest TSC will not be reliable\n");
5751 return kvm_x86_ops->vcpu_create(kvm, id);
5754 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5758 vcpu->arch.mtrr_state.have_fixed = 1;
5760 r = kvm_arch_vcpu_reset(vcpu);
5762 r = kvm_mmu_setup(vcpu);
5769 kvm_x86_ops->vcpu_free(vcpu);
5773 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5776 kvm_mmu_unload(vcpu);
5780 kvm_x86_ops->vcpu_free(vcpu);
5783 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5785 vcpu->arch.nmi_pending = false;
5786 vcpu->arch.nmi_injected = false;
5788 vcpu->arch.switch_db_regs = 0;
5789 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5790 vcpu->arch.dr6 = DR6_FIXED_1;
5791 vcpu->arch.dr7 = DR7_FIXED_1;
5793 kvm_make_request(KVM_REQ_EVENT, vcpu);
5795 return kvm_x86_ops->vcpu_reset(vcpu);
5798 int kvm_arch_hardware_enable(void *garbage)
5801 struct kvm_vcpu *vcpu;
5804 kvm_shared_msr_cpu_online();
5805 list_for_each_entry(kvm, &vm_list, vm_list)
5806 kvm_for_each_vcpu(i, vcpu, kvm)
5807 if (vcpu->cpu == smp_processor_id())
5808 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5809 return kvm_x86_ops->hardware_enable(garbage);
5812 void kvm_arch_hardware_disable(void *garbage)
5814 kvm_x86_ops->hardware_disable(garbage);
5815 drop_user_return_notifiers(garbage);
5818 int kvm_arch_hardware_setup(void)
5820 return kvm_x86_ops->hardware_setup();
5823 void kvm_arch_hardware_unsetup(void)
5825 kvm_x86_ops->hardware_unsetup();
5828 void kvm_arch_check_processor_compat(void *rtn)
5830 kvm_x86_ops->check_processor_compatibility(rtn);
5833 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5839 BUG_ON(vcpu->kvm == NULL);
5842 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
5843 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
5844 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5845 vcpu->arch.mmu.translate_gpa = translate_gpa;
5846 vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
5847 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5848 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5850 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5852 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5857 vcpu->arch.pio_data = page_address(page);
5859 if (!kvm->arch.virtual_tsc_khz)
5860 kvm_arch_set_tsc_khz(kvm, max_tsc_khz);
5862 r = kvm_mmu_create(vcpu);
5864 goto fail_free_pio_data;
5866 if (irqchip_in_kernel(kvm)) {
5867 r = kvm_create_lapic(vcpu);
5869 goto fail_mmu_destroy;
5872 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5874 if (!vcpu->arch.mce_banks) {
5876 goto fail_free_lapic;
5878 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5880 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
5881 goto fail_free_mce_banks;
5884 fail_free_mce_banks:
5885 kfree(vcpu->arch.mce_banks);
5887 kvm_free_lapic(vcpu);
5889 kvm_mmu_destroy(vcpu);
5891 free_page((unsigned long)vcpu->arch.pio_data);
5896 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5900 kfree(vcpu->arch.mce_banks);
5901 kvm_free_lapic(vcpu);
5902 idx = srcu_read_lock(&vcpu->kvm->srcu);
5903 kvm_mmu_destroy(vcpu);
5904 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5905 free_page((unsigned long)vcpu->arch.pio_data);
5908 struct kvm *kvm_arch_create_vm(void)
5910 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5913 return ERR_PTR(-ENOMEM);
5915 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5916 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5918 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5919 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5921 spin_lock_init(&kvm->arch.tsc_write_lock);
5926 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5929 kvm_mmu_unload(vcpu);
5933 static void kvm_free_vcpus(struct kvm *kvm)
5936 struct kvm_vcpu *vcpu;
5939 * Unpin any mmu pages first.
5941 kvm_for_each_vcpu(i, vcpu, kvm)
5942 kvm_unload_vcpu_mmu(vcpu);
5943 kvm_for_each_vcpu(i, vcpu, kvm)
5944 kvm_arch_vcpu_free(vcpu);
5946 mutex_lock(&kvm->lock);
5947 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5948 kvm->vcpus[i] = NULL;
5950 atomic_set(&kvm->online_vcpus, 0);
5951 mutex_unlock(&kvm->lock);
5954 void kvm_arch_sync_events(struct kvm *kvm)
5956 kvm_free_all_assigned_devices(kvm);
5960 void kvm_arch_destroy_vm(struct kvm *kvm)
5962 kvm_iommu_unmap_guest(kvm);
5963 kfree(kvm->arch.vpic);
5964 kfree(kvm->arch.vioapic);
5965 kvm_free_vcpus(kvm);
5966 kvm_free_physmem(kvm);
5967 if (kvm->arch.apic_access_page)
5968 put_page(kvm->arch.apic_access_page);
5969 if (kvm->arch.ept_identity_pagetable)
5970 put_page(kvm->arch.ept_identity_pagetable);
5971 cleanup_srcu_struct(&kvm->srcu);
5975 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5976 struct kvm_memory_slot *memslot,
5977 struct kvm_memory_slot old,
5978 struct kvm_userspace_memory_region *mem,
5981 int npages = memslot->npages;
5982 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
5984 /* Prevent internal slot pages from being moved by fork()/COW. */
5985 if (memslot->id >= KVM_MEMORY_SLOTS)
5986 map_flags = MAP_SHARED | MAP_ANONYMOUS;
5988 /*To keep backward compatibility with older userspace,
5989 *x86 needs to hanlde !user_alloc case.
5992 if (npages && !old.rmap) {
5993 unsigned long userspace_addr;
5995 down_write(¤t->mm->mmap_sem);
5996 userspace_addr = do_mmap(NULL, 0,
5998 PROT_READ | PROT_WRITE,
6001 up_write(¤t->mm->mmap_sem);
6003 if (IS_ERR((void *)userspace_addr))
6004 return PTR_ERR((void *)userspace_addr);
6006 memslot->userspace_addr = userspace_addr;
6014 void kvm_arch_commit_memory_region(struct kvm *kvm,
6015 struct kvm_userspace_memory_region *mem,
6016 struct kvm_memory_slot old,
6020 int npages = mem->memory_size >> PAGE_SHIFT;
6022 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6025 down_write(¤t->mm->mmap_sem);
6026 ret = do_munmap(current->mm, old.userspace_addr,
6027 old.npages * PAGE_SIZE);
6028 up_write(¤t->mm->mmap_sem);
6031 "kvm_vm_ioctl_set_memory_region: "
6032 "failed to munmap memory\n");
6035 spin_lock(&kvm->mmu_lock);
6036 if (!kvm->arch.n_requested_mmu_pages) {
6037 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6038 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6041 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6042 spin_unlock(&kvm->mmu_lock);
6045 void kvm_arch_flush_shadow(struct kvm *kvm)
6047 kvm_mmu_zap_all(kvm);
6048 kvm_reload_remote_mmus(kvm);
6051 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6053 return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
6054 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6055 || vcpu->arch.nmi_pending ||
6056 (kvm_arch_interrupt_allowed(vcpu) &&
6057 kvm_cpu_has_interrupt(vcpu));
6060 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6063 int cpu = vcpu->cpu;
6065 if (waitqueue_active(&vcpu->wq)) {
6066 wake_up_interruptible(&vcpu->wq);
6067 ++vcpu->stat.halt_wakeup;
6071 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6072 if (atomic_xchg(&vcpu->guest_mode, 0))
6073 smp_send_reschedule(cpu);
6077 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6079 return kvm_x86_ops->interrupt_allowed(vcpu);
6082 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6084 unsigned long current_rip = kvm_rip_read(vcpu) +
6085 get_segment_base(vcpu, VCPU_SREG_CS);
6087 return current_rip == linear_rip;
6089 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6091 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6093 unsigned long rflags;
6095 rflags = kvm_x86_ops->get_rflags(vcpu);
6096 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6097 rflags &= ~X86_EFLAGS_TF;
6100 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6102 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6104 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6105 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6106 rflags |= X86_EFLAGS_TF;
6107 kvm_x86_ops->set_rflags(vcpu, rflags);
6108 kvm_make_request(KVM_REQ_EVENT, vcpu);
6110 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6112 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6113 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6114 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6115 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6116 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6117 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6118 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6119 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6120 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6121 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6122 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6123 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);