2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
14 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
22 #include <linux/kvm_host.h>
27 #include "kvm_cache_regs.h"
30 #include <linux/clocksource.h>
31 #include <linux/interrupt.h>
32 #include <linux/kvm.h>
34 #include <linux/vmalloc.h>
35 #include <linux/module.h>
36 #include <linux/mman.h>
37 #include <linux/highmem.h>
38 #include <linux/iommu.h>
39 #include <linux/intel-iommu.h>
40 #include <linux/cpufreq.h>
41 #include <linux/user-return-notifier.h>
42 #include <linux/srcu.h>
43 #include <linux/slab.h>
44 #include <linux/perf_event.h>
45 #include <linux/uaccess.h>
46 #include <linux/hash.h>
47 #include <trace/events/kvm.h>
49 #define CREATE_TRACE_POINTS
52 #include <asm/debugreg.h>
59 #include <asm/pvclock.h>
60 #include <asm/div64.h>
62 #define MAX_IO_MSRS 256
63 #define CR0_RESERVED_BITS \
64 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
65 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
66 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
67 #define CR4_RESERVED_BITS \
68 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
69 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
70 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
72 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
74 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
76 #define KVM_MAX_MCE_BANKS 32
77 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
80 * - enable syscall per default because its emulated by KVM
81 * - enable LME and LMA per default on 64 bit KVM
84 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
86 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
89 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
90 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
92 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
93 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
94 struct kvm_cpuid_entry2 __user *entries);
96 struct kvm_x86_ops *kvm_x86_ops;
97 EXPORT_SYMBOL_GPL(kvm_x86_ops);
100 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
102 #define KVM_NR_SHARED_MSRS 16
104 struct kvm_shared_msrs_global {
106 u32 msrs[KVM_NR_SHARED_MSRS];
109 struct kvm_shared_msrs {
110 struct user_return_notifier urn;
112 struct kvm_shared_msr_values {
115 } values[KVM_NR_SHARED_MSRS];
118 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
119 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
121 struct kvm_stats_debugfs_item debugfs_entries[] = {
122 { "pf_fixed", VCPU_STAT(pf_fixed) },
123 { "pf_guest", VCPU_STAT(pf_guest) },
124 { "tlb_flush", VCPU_STAT(tlb_flush) },
125 { "invlpg", VCPU_STAT(invlpg) },
126 { "exits", VCPU_STAT(exits) },
127 { "io_exits", VCPU_STAT(io_exits) },
128 { "mmio_exits", VCPU_STAT(mmio_exits) },
129 { "signal_exits", VCPU_STAT(signal_exits) },
130 { "irq_window", VCPU_STAT(irq_window_exits) },
131 { "nmi_window", VCPU_STAT(nmi_window_exits) },
132 { "halt_exits", VCPU_STAT(halt_exits) },
133 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
134 { "hypercalls", VCPU_STAT(hypercalls) },
135 { "request_irq", VCPU_STAT(request_irq_exits) },
136 { "irq_exits", VCPU_STAT(irq_exits) },
137 { "host_state_reload", VCPU_STAT(host_state_reload) },
138 { "efer_reload", VCPU_STAT(efer_reload) },
139 { "fpu_reload", VCPU_STAT(fpu_reload) },
140 { "insn_emulation", VCPU_STAT(insn_emulation) },
141 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
142 { "irq_injections", VCPU_STAT(irq_injections) },
143 { "nmi_injections", VCPU_STAT(nmi_injections) },
144 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
145 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
146 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
147 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
148 { "mmu_flooded", VM_STAT(mmu_flooded) },
149 { "mmu_recycled", VM_STAT(mmu_recycled) },
150 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
151 { "mmu_unsync", VM_STAT(mmu_unsync) },
152 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
153 { "largepages", VM_STAT(lpages) },
157 u64 __read_mostly host_xcr0;
159 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
162 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
163 vcpu->arch.apf.gfns[i] = ~0;
166 static void kvm_on_user_return(struct user_return_notifier *urn)
169 struct kvm_shared_msrs *locals
170 = container_of(urn, struct kvm_shared_msrs, urn);
171 struct kvm_shared_msr_values *values;
173 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
174 values = &locals->values[slot];
175 if (values->host != values->curr) {
176 wrmsrl(shared_msrs_global.msrs[slot], values->host);
177 values->curr = values->host;
180 locals->registered = false;
181 user_return_notifier_unregister(urn);
184 static void shared_msr_update(unsigned slot, u32 msr)
186 struct kvm_shared_msrs *smsr;
189 smsr = &__get_cpu_var(shared_msrs);
190 /* only read, and nobody should modify it at this time,
191 * so don't need lock */
192 if (slot >= shared_msrs_global.nr) {
193 printk(KERN_ERR "kvm: invalid MSR slot!");
196 rdmsrl_safe(msr, &value);
197 smsr->values[slot].host = value;
198 smsr->values[slot].curr = value;
201 void kvm_define_shared_msr(unsigned slot, u32 msr)
203 if (slot >= shared_msrs_global.nr)
204 shared_msrs_global.nr = slot + 1;
205 shared_msrs_global.msrs[slot] = msr;
206 /* we need ensured the shared_msr_global have been updated */
209 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
211 static void kvm_shared_msr_cpu_online(void)
215 for (i = 0; i < shared_msrs_global.nr; ++i)
216 shared_msr_update(i, shared_msrs_global.msrs[i]);
219 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
221 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
223 if (((value ^ smsr->values[slot].curr) & mask) == 0)
225 smsr->values[slot].curr = value;
226 wrmsrl(shared_msrs_global.msrs[slot], value);
227 if (!smsr->registered) {
228 smsr->urn.on_user_return = kvm_on_user_return;
229 user_return_notifier_register(&smsr->urn);
230 smsr->registered = true;
233 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
235 static void drop_user_return_notifiers(void *ignore)
237 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
239 if (smsr->registered)
240 kvm_on_user_return(&smsr->urn);
243 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
245 if (irqchip_in_kernel(vcpu->kvm))
246 return vcpu->arch.apic_base;
248 return vcpu->arch.apic_base;
250 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
252 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
254 /* TODO: reserve bits check */
255 if (irqchip_in_kernel(vcpu->kvm))
256 kvm_lapic_set_base(vcpu, data);
258 vcpu->arch.apic_base = data;
260 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
262 #define EXCPT_BENIGN 0
263 #define EXCPT_CONTRIBUTORY 1
266 static int exception_class(int vector)
276 return EXCPT_CONTRIBUTORY;
283 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
284 unsigned nr, bool has_error, u32 error_code,
290 kvm_make_request(KVM_REQ_EVENT, vcpu);
292 if (!vcpu->arch.exception.pending) {
294 vcpu->arch.exception.pending = true;
295 vcpu->arch.exception.has_error_code = has_error;
296 vcpu->arch.exception.nr = nr;
297 vcpu->arch.exception.error_code = error_code;
298 vcpu->arch.exception.reinject = reinject;
302 /* to check exception */
303 prev_nr = vcpu->arch.exception.nr;
304 if (prev_nr == DF_VECTOR) {
305 /* triple fault -> shutdown */
306 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
309 class1 = exception_class(prev_nr);
310 class2 = exception_class(nr);
311 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
312 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
313 /* generate double fault per SDM Table 5-5 */
314 vcpu->arch.exception.pending = true;
315 vcpu->arch.exception.has_error_code = true;
316 vcpu->arch.exception.nr = DF_VECTOR;
317 vcpu->arch.exception.error_code = 0;
319 /* replace previous exception with a new one in a hope
320 that instruction re-execution will regenerate lost
325 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
327 kvm_multiple_exception(vcpu, nr, false, 0, false);
329 EXPORT_SYMBOL_GPL(kvm_queue_exception);
331 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
333 kvm_multiple_exception(vcpu, nr, false, 0, true);
335 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
337 void kvm_inject_page_fault(struct kvm_vcpu *vcpu)
339 unsigned error_code = vcpu->arch.fault.error_code;
341 ++vcpu->stat.pf_guest;
342 vcpu->arch.cr2 = vcpu->arch.fault.address;
343 kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
346 void kvm_propagate_fault(struct kvm_vcpu *vcpu)
348 if (mmu_is_nested(vcpu) && !vcpu->arch.fault.nested)
349 vcpu->arch.nested_mmu.inject_page_fault(vcpu);
351 vcpu->arch.mmu.inject_page_fault(vcpu);
353 vcpu->arch.fault.nested = false;
356 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
358 kvm_make_request(KVM_REQ_EVENT, vcpu);
359 vcpu->arch.nmi_pending = 1;
361 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
363 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
365 kvm_multiple_exception(vcpu, nr, true, error_code, false);
367 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
369 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
371 kvm_multiple_exception(vcpu, nr, true, error_code, true);
373 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
376 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
377 * a #GP and return false.
379 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
381 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
383 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
386 EXPORT_SYMBOL_GPL(kvm_require_cpl);
389 * This function will be used to read from the physical memory of the currently
390 * running guest. The difference to kvm_read_guest_page is that this function
391 * can read from guest physical or from the guest's guest physical memory.
393 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
394 gfn_t ngfn, void *data, int offset, int len,
400 ngpa = gfn_to_gpa(ngfn);
401 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
402 if (real_gfn == UNMAPPED_GVA)
405 real_gfn = gpa_to_gfn(real_gfn);
407 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
409 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
411 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
412 void *data, int offset, int len, u32 access)
414 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
415 data, offset, len, access);
419 * Load the pae pdptrs. Return true is they are all valid.
421 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
423 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
424 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
427 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
429 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
430 offset * sizeof(u64), sizeof(pdpte),
431 PFERR_USER_MASK|PFERR_WRITE_MASK);
436 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
437 if (is_present_gpte(pdpte[i]) &&
438 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
445 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
446 __set_bit(VCPU_EXREG_PDPTR,
447 (unsigned long *)&vcpu->arch.regs_avail);
448 __set_bit(VCPU_EXREG_PDPTR,
449 (unsigned long *)&vcpu->arch.regs_dirty);
454 EXPORT_SYMBOL_GPL(load_pdptrs);
456 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
458 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
464 if (is_long_mode(vcpu) || !is_pae(vcpu))
467 if (!test_bit(VCPU_EXREG_PDPTR,
468 (unsigned long *)&vcpu->arch.regs_avail))
471 gfn = (vcpu->arch.cr3 & ~31u) >> PAGE_SHIFT;
472 offset = (vcpu->arch.cr3 & ~31u) & (PAGE_SIZE - 1);
473 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
474 PFERR_USER_MASK | PFERR_WRITE_MASK);
477 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
483 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
485 unsigned long old_cr0 = kvm_read_cr0(vcpu);
486 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
487 X86_CR0_CD | X86_CR0_NW;
492 if (cr0 & 0xffffffff00000000UL)
496 cr0 &= ~CR0_RESERVED_BITS;
498 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
501 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
504 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
506 if ((vcpu->arch.efer & EFER_LME)) {
511 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
516 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
521 kvm_x86_ops->set_cr0(vcpu, cr0);
523 if ((cr0 ^ old_cr0) & update_bits)
524 kvm_mmu_reset_context(vcpu);
527 EXPORT_SYMBOL_GPL(kvm_set_cr0);
529 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
531 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
533 EXPORT_SYMBOL_GPL(kvm_lmsw);
535 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
539 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
540 if (index != XCR_XFEATURE_ENABLED_MASK)
543 if (kvm_x86_ops->get_cpl(vcpu) != 0)
545 if (!(xcr0 & XSTATE_FP))
547 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
549 if (xcr0 & ~host_xcr0)
551 vcpu->arch.xcr0 = xcr0;
552 vcpu->guest_xcr0_loaded = 0;
556 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
558 if (__kvm_set_xcr(vcpu, index, xcr)) {
559 kvm_inject_gp(vcpu, 0);
564 EXPORT_SYMBOL_GPL(kvm_set_xcr);
566 static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
568 struct kvm_cpuid_entry2 *best;
570 best = kvm_find_cpuid_entry(vcpu, 1, 0);
571 return best && (best->ecx & bit(X86_FEATURE_XSAVE));
574 static void update_cpuid(struct kvm_vcpu *vcpu)
576 struct kvm_cpuid_entry2 *best;
578 best = kvm_find_cpuid_entry(vcpu, 1, 0);
582 /* Update OSXSAVE bit */
583 if (cpu_has_xsave && best->function == 0x1) {
584 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
585 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
586 best->ecx |= bit(X86_FEATURE_OSXSAVE);
590 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
592 unsigned long old_cr4 = kvm_read_cr4(vcpu);
593 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
595 if (cr4 & CR4_RESERVED_BITS)
598 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
601 if (is_long_mode(vcpu)) {
602 if (!(cr4 & X86_CR4_PAE))
604 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
605 && ((cr4 ^ old_cr4) & pdptr_bits)
606 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, vcpu->arch.cr3))
609 if (cr4 & X86_CR4_VMXE)
612 kvm_x86_ops->set_cr4(vcpu, cr4);
614 if ((cr4 ^ old_cr4) & pdptr_bits)
615 kvm_mmu_reset_context(vcpu);
617 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
622 EXPORT_SYMBOL_GPL(kvm_set_cr4);
624 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
626 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
627 kvm_mmu_sync_roots(vcpu);
628 kvm_mmu_flush_tlb(vcpu);
632 if (is_long_mode(vcpu)) {
633 if (cr3 & CR3_L_MODE_RESERVED_BITS)
637 if (cr3 & CR3_PAE_RESERVED_BITS)
639 if (is_paging(vcpu) &&
640 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
644 * We don't check reserved bits in nonpae mode, because
645 * this isn't enforced, and VMware depends on this.
650 * Does the new cr3 value map to physical memory? (Note, we
651 * catch an invalid cr3 even in real-mode, because it would
652 * cause trouble later on when we turn on paging anyway.)
654 * A real CPU would silently accept an invalid cr3 and would
655 * attempt to use it - with largely undefined (and often hard
656 * to debug) behavior on the guest side.
658 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
660 vcpu->arch.cr3 = cr3;
661 vcpu->arch.mmu.new_cr3(vcpu);
664 EXPORT_SYMBOL_GPL(kvm_set_cr3);
666 int __kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
668 if (cr8 & CR8_RESERVED_BITS)
670 if (irqchip_in_kernel(vcpu->kvm))
671 kvm_lapic_set_tpr(vcpu, cr8);
673 vcpu->arch.cr8 = cr8;
677 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
679 if (__kvm_set_cr8(vcpu, cr8))
680 kvm_inject_gp(vcpu, 0);
682 EXPORT_SYMBOL_GPL(kvm_set_cr8);
684 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
686 if (irqchip_in_kernel(vcpu->kvm))
687 return kvm_lapic_get_cr8(vcpu);
689 return vcpu->arch.cr8;
691 EXPORT_SYMBOL_GPL(kvm_get_cr8);
693 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
697 vcpu->arch.db[dr] = val;
698 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
699 vcpu->arch.eff_db[dr] = val;
702 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
706 if (val & 0xffffffff00000000ULL)
708 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
711 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
715 if (val & 0xffffffff00000000ULL)
717 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
718 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
719 kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
720 vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
728 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
732 res = __kvm_set_dr(vcpu, dr, val);
734 kvm_queue_exception(vcpu, UD_VECTOR);
736 kvm_inject_gp(vcpu, 0);
740 EXPORT_SYMBOL_GPL(kvm_set_dr);
742 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
746 *val = vcpu->arch.db[dr];
749 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
753 *val = vcpu->arch.dr6;
756 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
760 *val = vcpu->arch.dr7;
767 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
769 if (_kvm_get_dr(vcpu, dr, val)) {
770 kvm_queue_exception(vcpu, UD_VECTOR);
775 EXPORT_SYMBOL_GPL(kvm_get_dr);
778 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
779 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
781 * This list is modified at module load time to reflect the
782 * capabilities of the host cpu. This capabilities test skips MSRs that are
783 * kvm-specific. Those are put in the beginning of the list.
786 #define KVM_SAVE_MSRS_BEGIN 8
787 static u32 msrs_to_save[] = {
788 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
789 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
790 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
791 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN,
792 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
795 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
797 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
800 static unsigned num_msrs_to_save;
802 static u32 emulated_msrs[] = {
803 MSR_IA32_MISC_ENABLE,
808 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
810 u64 old_efer = vcpu->arch.efer;
812 if (efer & efer_reserved_bits)
816 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
819 if (efer & EFER_FFXSR) {
820 struct kvm_cpuid_entry2 *feat;
822 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
823 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
827 if (efer & EFER_SVME) {
828 struct kvm_cpuid_entry2 *feat;
830 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
831 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
836 efer |= vcpu->arch.efer & EFER_LMA;
838 kvm_x86_ops->set_efer(vcpu, efer);
840 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
842 /* Update reserved bits */
843 if ((efer ^ old_efer) & EFER_NX)
844 kvm_mmu_reset_context(vcpu);
849 void kvm_enable_efer_bits(u64 mask)
851 efer_reserved_bits &= ~mask;
853 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
857 * Writes msr value into into the appropriate "register".
858 * Returns 0 on success, non-0 otherwise.
859 * Assumes vcpu_load() was already called.
861 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
863 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
867 * Adapt set_msr() to msr_io()'s calling convention
869 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
871 return kvm_set_msr(vcpu, index, *data);
874 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
878 struct pvclock_wall_clock wc;
879 struct timespec boot;
884 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
889 ++version; /* first time write, random junk */
893 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
896 * The guest calculates current wall clock time by adding
897 * system time (updated by kvm_guest_time_update below) to the
898 * wall clock specified here. guest system time equals host
899 * system time for us, thus we must fill in host boot time here.
903 wc.sec = boot.tv_sec;
904 wc.nsec = boot.tv_nsec;
905 wc.version = version;
907 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
910 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
913 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
915 uint32_t quotient, remainder;
917 /* Don't try to replace with do_div(), this one calculates
918 * "(dividend << 32) / divisor" */
920 : "=a" (quotient), "=d" (remainder)
921 : "0" (0), "1" (dividend), "r" (divisor) );
925 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
926 s8 *pshift, u32 *pmultiplier)
933 tps64 = base_khz * 1000LL;
934 scaled64 = scaled_khz * 1000LL;
935 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
940 tps32 = (uint32_t)tps64;
941 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
942 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
950 *pmultiplier = div_frac(scaled64, tps32);
952 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
953 __func__, base_khz, scaled_khz, shift, *pmultiplier);
956 static inline u64 get_kernel_ns(void)
960 WARN_ON(preemptible());
962 monotonic_to_bootbased(&ts);
963 return timespec_to_ns(&ts);
966 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
967 unsigned long max_tsc_khz;
969 static inline int kvm_tsc_changes_freq(void)
972 int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
973 cpufreq_quick_get(cpu) != 0;
978 static inline u64 nsec_to_cycles(u64 nsec)
982 WARN_ON(preemptible());
983 if (kvm_tsc_changes_freq())
984 printk_once(KERN_WARNING
985 "kvm: unreliable cycle conversion on adjustable rate TSC\n");
986 ret = nsec * __get_cpu_var(cpu_tsc_khz);
987 do_div(ret, USEC_PER_SEC);
991 static void kvm_arch_set_tsc_khz(struct kvm *kvm, u32 this_tsc_khz)
993 /* Compute a scale to convert nanoseconds in TSC cycles */
994 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
995 &kvm->arch.virtual_tsc_shift,
996 &kvm->arch.virtual_tsc_mult);
997 kvm->arch.virtual_tsc_khz = this_tsc_khz;
1000 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1002 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1003 vcpu->kvm->arch.virtual_tsc_mult,
1004 vcpu->kvm->arch.virtual_tsc_shift);
1005 tsc += vcpu->arch.last_tsc_write;
1009 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1011 struct kvm *kvm = vcpu->kvm;
1012 u64 offset, ns, elapsed;
1013 unsigned long flags;
1016 spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1017 offset = data - native_read_tsc();
1018 ns = get_kernel_ns();
1019 elapsed = ns - kvm->arch.last_tsc_nsec;
1020 sdiff = data - kvm->arch.last_tsc_write;
1025 * Special case: close write to TSC within 5 seconds of
1026 * another CPU is interpreted as an attempt to synchronize
1027 * The 5 seconds is to accomodate host load / swapping as
1028 * well as any reset of TSC during the boot process.
1030 * In that case, for a reliable TSC, we can match TSC offsets,
1031 * or make a best guest using elapsed value.
1033 if (sdiff < nsec_to_cycles(5ULL * NSEC_PER_SEC) &&
1034 elapsed < 5ULL * NSEC_PER_SEC) {
1035 if (!check_tsc_unstable()) {
1036 offset = kvm->arch.last_tsc_offset;
1037 pr_debug("kvm: matched tsc offset for %llu\n", data);
1039 u64 delta = nsec_to_cycles(elapsed);
1041 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1043 ns = kvm->arch.last_tsc_nsec;
1045 kvm->arch.last_tsc_nsec = ns;
1046 kvm->arch.last_tsc_write = data;
1047 kvm->arch.last_tsc_offset = offset;
1048 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1049 spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1051 /* Reset of TSC must disable overshoot protection below */
1052 vcpu->arch.hv_clock.tsc_timestamp = 0;
1053 vcpu->arch.last_tsc_write = data;
1054 vcpu->arch.last_tsc_nsec = ns;
1056 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1058 static int kvm_guest_time_update(struct kvm_vcpu *v)
1060 unsigned long flags;
1061 struct kvm_vcpu_arch *vcpu = &v->arch;
1063 unsigned long this_tsc_khz;
1064 s64 kernel_ns, max_kernel_ns;
1067 /* Keep irq disabled to prevent changes to the clock */
1068 local_irq_save(flags);
1069 kvm_get_msr(v, MSR_IA32_TSC, &tsc_timestamp);
1070 kernel_ns = get_kernel_ns();
1071 this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
1073 if (unlikely(this_tsc_khz == 0)) {
1074 local_irq_restore(flags);
1075 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1080 * We may have to catch up the TSC to match elapsed wall clock
1081 * time for two reasons, even if kvmclock is used.
1082 * 1) CPU could have been running below the maximum TSC rate
1083 * 2) Broken TSC compensation resets the base at each VCPU
1084 * entry to avoid unknown leaps of TSC even when running
1085 * again on the same CPU. This may cause apparent elapsed
1086 * time to disappear, and the guest to stand still or run
1089 if (vcpu->tsc_catchup) {
1090 u64 tsc = compute_guest_tsc(v, kernel_ns);
1091 if (tsc > tsc_timestamp) {
1092 kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1093 tsc_timestamp = tsc;
1097 local_irq_restore(flags);
1099 if (!vcpu->time_page)
1103 * Time as measured by the TSC may go backwards when resetting the base
1104 * tsc_timestamp. The reason for this is that the TSC resolution is
1105 * higher than the resolution of the other clock scales. Thus, many
1106 * possible measurments of the TSC correspond to one measurement of any
1107 * other clock, and so a spread of values is possible. This is not a
1108 * problem for the computation of the nanosecond clock; with TSC rates
1109 * around 1GHZ, there can only be a few cycles which correspond to one
1110 * nanosecond value, and any path through this code will inevitably
1111 * take longer than that. However, with the kernel_ns value itself,
1112 * the precision may be much lower, down to HZ granularity. If the
1113 * first sampling of TSC against kernel_ns ends in the low part of the
1114 * range, and the second in the high end of the range, we can get:
1116 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1118 * As the sampling errors potentially range in the thousands of cycles,
1119 * it is possible such a time value has already been observed by the
1120 * guest. To protect against this, we must compute the system time as
1121 * observed by the guest and ensure the new system time is greater.
1124 if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1125 max_kernel_ns = vcpu->last_guest_tsc -
1126 vcpu->hv_clock.tsc_timestamp;
1127 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1128 vcpu->hv_clock.tsc_to_system_mul,
1129 vcpu->hv_clock.tsc_shift);
1130 max_kernel_ns += vcpu->last_kernel_ns;
1133 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1134 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1135 &vcpu->hv_clock.tsc_shift,
1136 &vcpu->hv_clock.tsc_to_system_mul);
1137 vcpu->hw_tsc_khz = this_tsc_khz;
1140 if (max_kernel_ns > kernel_ns)
1141 kernel_ns = max_kernel_ns;
1143 /* With all the info we got, fill in the values */
1144 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1145 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1146 vcpu->last_kernel_ns = kernel_ns;
1147 vcpu->last_guest_tsc = tsc_timestamp;
1148 vcpu->hv_clock.flags = 0;
1151 * The interface expects us to write an even number signaling that the
1152 * update is finished. Since the guest won't see the intermediate
1153 * state, we just increase by 2 at the end.
1155 vcpu->hv_clock.version += 2;
1157 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1159 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1160 sizeof(vcpu->hv_clock));
1162 kunmap_atomic(shared_kaddr, KM_USER0);
1164 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1168 static bool msr_mtrr_valid(unsigned msr)
1171 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1172 case MSR_MTRRfix64K_00000:
1173 case MSR_MTRRfix16K_80000:
1174 case MSR_MTRRfix16K_A0000:
1175 case MSR_MTRRfix4K_C0000:
1176 case MSR_MTRRfix4K_C8000:
1177 case MSR_MTRRfix4K_D0000:
1178 case MSR_MTRRfix4K_D8000:
1179 case MSR_MTRRfix4K_E0000:
1180 case MSR_MTRRfix4K_E8000:
1181 case MSR_MTRRfix4K_F0000:
1182 case MSR_MTRRfix4K_F8000:
1183 case MSR_MTRRdefType:
1184 case MSR_IA32_CR_PAT:
1192 static bool valid_pat_type(unsigned t)
1194 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1197 static bool valid_mtrr_type(unsigned t)
1199 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1202 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1206 if (!msr_mtrr_valid(msr))
1209 if (msr == MSR_IA32_CR_PAT) {
1210 for (i = 0; i < 8; i++)
1211 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1214 } else if (msr == MSR_MTRRdefType) {
1217 return valid_mtrr_type(data & 0xff);
1218 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1219 for (i = 0; i < 8 ; i++)
1220 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1225 /* variable MTRRs */
1226 return valid_mtrr_type(data & 0xff);
1229 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1231 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1233 if (!mtrr_valid(vcpu, msr, data))
1236 if (msr == MSR_MTRRdefType) {
1237 vcpu->arch.mtrr_state.def_type = data;
1238 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1239 } else if (msr == MSR_MTRRfix64K_00000)
1241 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1242 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1243 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1244 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1245 else if (msr == MSR_IA32_CR_PAT)
1246 vcpu->arch.pat = data;
1247 else { /* Variable MTRRs */
1248 int idx, is_mtrr_mask;
1251 idx = (msr - 0x200) / 2;
1252 is_mtrr_mask = msr - 0x200 - 2 * idx;
1255 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1258 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1262 kvm_mmu_reset_context(vcpu);
1266 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1268 u64 mcg_cap = vcpu->arch.mcg_cap;
1269 unsigned bank_num = mcg_cap & 0xff;
1272 case MSR_IA32_MCG_STATUS:
1273 vcpu->arch.mcg_status = data;
1275 case MSR_IA32_MCG_CTL:
1276 if (!(mcg_cap & MCG_CTL_P))
1278 if (data != 0 && data != ~(u64)0)
1280 vcpu->arch.mcg_ctl = data;
1283 if (msr >= MSR_IA32_MC0_CTL &&
1284 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1285 u32 offset = msr - MSR_IA32_MC0_CTL;
1286 /* only 0 or all 1s can be written to IA32_MCi_CTL
1287 * some Linux kernels though clear bit 10 in bank 4 to
1288 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1289 * this to avoid an uncatched #GP in the guest
1291 if ((offset & 0x3) == 0 &&
1292 data != 0 && (data | (1 << 10)) != ~(u64)0)
1294 vcpu->arch.mce_banks[offset] = data;
1302 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1304 struct kvm *kvm = vcpu->kvm;
1305 int lm = is_long_mode(vcpu);
1306 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1307 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1308 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1309 : kvm->arch.xen_hvm_config.blob_size_32;
1310 u32 page_num = data & ~PAGE_MASK;
1311 u64 page_addr = data & PAGE_MASK;
1316 if (page_num >= blob_size)
1319 page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1323 if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1325 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1334 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1336 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1339 static bool kvm_hv_msr_partition_wide(u32 msr)
1343 case HV_X64_MSR_GUEST_OS_ID:
1344 case HV_X64_MSR_HYPERCALL:
1352 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1354 struct kvm *kvm = vcpu->kvm;
1357 case HV_X64_MSR_GUEST_OS_ID:
1358 kvm->arch.hv_guest_os_id = data;
1359 /* setting guest os id to zero disables hypercall page */
1360 if (!kvm->arch.hv_guest_os_id)
1361 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1363 case HV_X64_MSR_HYPERCALL: {
1368 /* if guest os id is not set hypercall should remain disabled */
1369 if (!kvm->arch.hv_guest_os_id)
1371 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1372 kvm->arch.hv_hypercall = data;
1375 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1376 addr = gfn_to_hva(kvm, gfn);
1377 if (kvm_is_error_hva(addr))
1379 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1380 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1381 if (copy_to_user((void __user *)addr, instructions, 4))
1383 kvm->arch.hv_hypercall = data;
1387 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1388 "data 0x%llx\n", msr, data);
1394 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1397 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1400 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1401 vcpu->arch.hv_vapic = data;
1404 addr = gfn_to_hva(vcpu->kvm, data >>
1405 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1406 if (kvm_is_error_hva(addr))
1408 if (clear_user((void __user *)addr, PAGE_SIZE))
1410 vcpu->arch.hv_vapic = data;
1413 case HV_X64_MSR_EOI:
1414 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1415 case HV_X64_MSR_ICR:
1416 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1417 case HV_X64_MSR_TPR:
1418 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1420 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1421 "data 0x%llx\n", msr, data);
1428 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1430 gpa_t gpa = data & ~0x3f;
1432 /* Bits 1:5 are resrved, Should be zero */
1436 vcpu->arch.apf.msr_val = data;
1438 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1439 kvm_clear_async_pf_completion_queue(vcpu);
1440 kvm_async_pf_hash_reset(vcpu);
1444 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1447 kvm_async_pf_wakeup_all(vcpu);
1451 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1455 return set_efer(vcpu, data);
1457 data &= ~(u64)0x40; /* ignore flush filter disable */
1458 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1460 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1465 case MSR_FAM10H_MMIO_CONF_BASE:
1467 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1472 case MSR_AMD64_NB_CFG:
1474 case MSR_IA32_DEBUGCTLMSR:
1476 /* We support the non-activated case already */
1478 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1479 /* Values other than LBR and BTF are vendor-specific,
1480 thus reserved and should throw a #GP */
1483 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1486 case MSR_IA32_UCODE_REV:
1487 case MSR_IA32_UCODE_WRITE:
1488 case MSR_VM_HSAVE_PA:
1489 case MSR_AMD64_PATCH_LOADER:
1491 case 0x200 ... 0x2ff:
1492 return set_msr_mtrr(vcpu, msr, data);
1493 case MSR_IA32_APICBASE:
1494 kvm_set_apic_base(vcpu, data);
1496 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1497 return kvm_x2apic_msr_write(vcpu, msr, data);
1498 case MSR_IA32_MISC_ENABLE:
1499 vcpu->arch.ia32_misc_enable_msr = data;
1501 case MSR_KVM_WALL_CLOCK_NEW:
1502 case MSR_KVM_WALL_CLOCK:
1503 vcpu->kvm->arch.wall_clock = data;
1504 kvm_write_wall_clock(vcpu->kvm, data);
1506 case MSR_KVM_SYSTEM_TIME_NEW:
1507 case MSR_KVM_SYSTEM_TIME: {
1508 if (vcpu->arch.time_page) {
1509 kvm_release_page_dirty(vcpu->arch.time_page);
1510 vcpu->arch.time_page = NULL;
1513 vcpu->arch.time = data;
1514 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1516 /* we verify if the enable bit is set... */
1520 /* ...but clean it before doing the actual write */
1521 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1523 vcpu->arch.time_page =
1524 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1526 if (is_error_page(vcpu->arch.time_page)) {
1527 kvm_release_page_clean(vcpu->arch.time_page);
1528 vcpu->arch.time_page = NULL;
1532 case MSR_KVM_ASYNC_PF_EN:
1533 if (kvm_pv_enable_async_pf(vcpu, data))
1536 case MSR_IA32_MCG_CTL:
1537 case MSR_IA32_MCG_STATUS:
1538 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1539 return set_msr_mce(vcpu, msr, data);
1541 /* Performance counters are not protected by a CPUID bit,
1542 * so we should check all of them in the generic path for the sake of
1543 * cross vendor migration.
1544 * Writing a zero into the event select MSRs disables them,
1545 * which we perfectly emulate ;-). Any other value should be at least
1546 * reported, some guests depend on them.
1548 case MSR_P6_EVNTSEL0:
1549 case MSR_P6_EVNTSEL1:
1550 case MSR_K7_EVNTSEL0:
1551 case MSR_K7_EVNTSEL1:
1552 case MSR_K7_EVNTSEL2:
1553 case MSR_K7_EVNTSEL3:
1555 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1556 "0x%x data 0x%llx\n", msr, data);
1558 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1559 * so we ignore writes to make it happy.
1561 case MSR_P6_PERFCTR0:
1562 case MSR_P6_PERFCTR1:
1563 case MSR_K7_PERFCTR0:
1564 case MSR_K7_PERFCTR1:
1565 case MSR_K7_PERFCTR2:
1566 case MSR_K7_PERFCTR3:
1567 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1568 "0x%x data 0x%llx\n", msr, data);
1570 case MSR_K7_CLK_CTL:
1572 * Ignore all writes to this no longer documented MSR.
1573 * Writes are only relevant for old K7 processors,
1574 * all pre-dating SVM, but a recommended workaround from
1575 * AMD for these chips. It is possible to speicify the
1576 * affected processor models on the command line, hence
1577 * the need to ignore the workaround.
1580 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1581 if (kvm_hv_msr_partition_wide(msr)) {
1583 mutex_lock(&vcpu->kvm->lock);
1584 r = set_msr_hyperv_pw(vcpu, msr, data);
1585 mutex_unlock(&vcpu->kvm->lock);
1588 return set_msr_hyperv(vcpu, msr, data);
1591 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1592 return xen_hvm_config(vcpu, data);
1594 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1598 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1605 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1609 * Reads an msr value (of 'msr_index') into 'pdata'.
1610 * Returns 0 on success, non-0 otherwise.
1611 * Assumes vcpu_load() was already called.
1613 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1615 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1618 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1620 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1622 if (!msr_mtrr_valid(msr))
1625 if (msr == MSR_MTRRdefType)
1626 *pdata = vcpu->arch.mtrr_state.def_type +
1627 (vcpu->arch.mtrr_state.enabled << 10);
1628 else if (msr == MSR_MTRRfix64K_00000)
1630 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1631 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1632 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1633 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1634 else if (msr == MSR_IA32_CR_PAT)
1635 *pdata = vcpu->arch.pat;
1636 else { /* Variable MTRRs */
1637 int idx, is_mtrr_mask;
1640 idx = (msr - 0x200) / 2;
1641 is_mtrr_mask = msr - 0x200 - 2 * idx;
1644 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1647 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1654 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1657 u64 mcg_cap = vcpu->arch.mcg_cap;
1658 unsigned bank_num = mcg_cap & 0xff;
1661 case MSR_IA32_P5_MC_ADDR:
1662 case MSR_IA32_P5_MC_TYPE:
1665 case MSR_IA32_MCG_CAP:
1666 data = vcpu->arch.mcg_cap;
1668 case MSR_IA32_MCG_CTL:
1669 if (!(mcg_cap & MCG_CTL_P))
1671 data = vcpu->arch.mcg_ctl;
1673 case MSR_IA32_MCG_STATUS:
1674 data = vcpu->arch.mcg_status;
1677 if (msr >= MSR_IA32_MC0_CTL &&
1678 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1679 u32 offset = msr - MSR_IA32_MC0_CTL;
1680 data = vcpu->arch.mce_banks[offset];
1689 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1692 struct kvm *kvm = vcpu->kvm;
1695 case HV_X64_MSR_GUEST_OS_ID:
1696 data = kvm->arch.hv_guest_os_id;
1698 case HV_X64_MSR_HYPERCALL:
1699 data = kvm->arch.hv_hypercall;
1702 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1710 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1715 case HV_X64_MSR_VP_INDEX: {
1718 kvm_for_each_vcpu(r, v, vcpu->kvm)
1723 case HV_X64_MSR_EOI:
1724 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1725 case HV_X64_MSR_ICR:
1726 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1727 case HV_X64_MSR_TPR:
1728 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1730 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1737 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1742 case MSR_IA32_PLATFORM_ID:
1743 case MSR_IA32_UCODE_REV:
1744 case MSR_IA32_EBL_CR_POWERON:
1745 case MSR_IA32_DEBUGCTLMSR:
1746 case MSR_IA32_LASTBRANCHFROMIP:
1747 case MSR_IA32_LASTBRANCHTOIP:
1748 case MSR_IA32_LASTINTFROMIP:
1749 case MSR_IA32_LASTINTTOIP:
1752 case MSR_VM_HSAVE_PA:
1753 case MSR_P6_PERFCTR0:
1754 case MSR_P6_PERFCTR1:
1755 case MSR_P6_EVNTSEL0:
1756 case MSR_P6_EVNTSEL1:
1757 case MSR_K7_EVNTSEL0:
1758 case MSR_K7_PERFCTR0:
1759 case MSR_K8_INT_PENDING_MSG:
1760 case MSR_AMD64_NB_CFG:
1761 case MSR_FAM10H_MMIO_CONF_BASE:
1765 data = 0x500 | KVM_NR_VAR_MTRR;
1767 case 0x200 ... 0x2ff:
1768 return get_msr_mtrr(vcpu, msr, pdata);
1769 case 0xcd: /* fsb frequency */
1773 * MSR_EBC_FREQUENCY_ID
1774 * Conservative value valid for even the basic CPU models.
1775 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1776 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1777 * and 266MHz for model 3, or 4. Set Core Clock
1778 * Frequency to System Bus Frequency Ratio to 1 (bits
1779 * 31:24) even though these are only valid for CPU
1780 * models > 2, however guests may end up dividing or
1781 * multiplying by zero otherwise.
1783 case MSR_EBC_FREQUENCY_ID:
1786 case MSR_IA32_APICBASE:
1787 data = kvm_get_apic_base(vcpu);
1789 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1790 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1792 case MSR_IA32_MISC_ENABLE:
1793 data = vcpu->arch.ia32_misc_enable_msr;
1795 case MSR_IA32_PERF_STATUS:
1796 /* TSC increment by tick */
1798 /* CPU multiplier */
1799 data |= (((uint64_t)4ULL) << 40);
1802 data = vcpu->arch.efer;
1804 case MSR_KVM_WALL_CLOCK:
1805 case MSR_KVM_WALL_CLOCK_NEW:
1806 data = vcpu->kvm->arch.wall_clock;
1808 case MSR_KVM_SYSTEM_TIME:
1809 case MSR_KVM_SYSTEM_TIME_NEW:
1810 data = vcpu->arch.time;
1812 case MSR_KVM_ASYNC_PF_EN:
1813 data = vcpu->arch.apf.msr_val;
1815 case MSR_IA32_P5_MC_ADDR:
1816 case MSR_IA32_P5_MC_TYPE:
1817 case MSR_IA32_MCG_CAP:
1818 case MSR_IA32_MCG_CTL:
1819 case MSR_IA32_MCG_STATUS:
1820 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1821 return get_msr_mce(vcpu, msr, pdata);
1822 case MSR_K7_CLK_CTL:
1824 * Provide expected ramp-up count for K7. All other
1825 * are set to zero, indicating minimum divisors for
1828 * This prevents guest kernels on AMD host with CPU
1829 * type 6, model 8 and higher from exploding due to
1830 * the rdmsr failing.
1834 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1835 if (kvm_hv_msr_partition_wide(msr)) {
1837 mutex_lock(&vcpu->kvm->lock);
1838 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1839 mutex_unlock(&vcpu->kvm->lock);
1842 return get_msr_hyperv(vcpu, msr, pdata);
1846 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1849 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1857 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1860 * Read or write a bunch of msrs. All parameters are kernel addresses.
1862 * @return number of msrs set successfully.
1864 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1865 struct kvm_msr_entry *entries,
1866 int (*do_msr)(struct kvm_vcpu *vcpu,
1867 unsigned index, u64 *data))
1871 idx = srcu_read_lock(&vcpu->kvm->srcu);
1872 for (i = 0; i < msrs->nmsrs; ++i)
1873 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1875 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1881 * Read or write a bunch of msrs. Parameters are user addresses.
1883 * @return number of msrs set successfully.
1885 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1886 int (*do_msr)(struct kvm_vcpu *vcpu,
1887 unsigned index, u64 *data),
1890 struct kvm_msrs msrs;
1891 struct kvm_msr_entry *entries;
1896 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1900 if (msrs.nmsrs >= MAX_IO_MSRS)
1904 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1905 entries = kmalloc(size, GFP_KERNEL);
1910 if (copy_from_user(entries, user_msrs->entries, size))
1913 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1918 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1929 int kvm_dev_ioctl_check_extension(long ext)
1934 case KVM_CAP_IRQCHIP:
1936 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1937 case KVM_CAP_SET_TSS_ADDR:
1938 case KVM_CAP_EXT_CPUID:
1939 case KVM_CAP_CLOCKSOURCE:
1941 case KVM_CAP_NOP_IO_DELAY:
1942 case KVM_CAP_MP_STATE:
1943 case KVM_CAP_SYNC_MMU:
1944 case KVM_CAP_REINJECT_CONTROL:
1945 case KVM_CAP_IRQ_INJECT_STATUS:
1946 case KVM_CAP_ASSIGN_DEV_IRQ:
1948 case KVM_CAP_IOEVENTFD:
1950 case KVM_CAP_PIT_STATE2:
1951 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1952 case KVM_CAP_XEN_HVM:
1953 case KVM_CAP_ADJUST_CLOCK:
1954 case KVM_CAP_VCPU_EVENTS:
1955 case KVM_CAP_HYPERV:
1956 case KVM_CAP_HYPERV_VAPIC:
1957 case KVM_CAP_HYPERV_SPIN:
1958 case KVM_CAP_PCI_SEGMENT:
1959 case KVM_CAP_DEBUGREGS:
1960 case KVM_CAP_X86_ROBUST_SINGLESTEP:
1962 case KVM_CAP_ASYNC_PF:
1965 case KVM_CAP_COALESCED_MMIO:
1966 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1969 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1971 case KVM_CAP_NR_VCPUS:
1974 case KVM_CAP_NR_MEMSLOTS:
1975 r = KVM_MEMORY_SLOTS;
1977 case KVM_CAP_PV_MMU: /* obsolete */
1984 r = KVM_MAX_MCE_BANKS;
1997 long kvm_arch_dev_ioctl(struct file *filp,
1998 unsigned int ioctl, unsigned long arg)
2000 void __user *argp = (void __user *)arg;
2004 case KVM_GET_MSR_INDEX_LIST: {
2005 struct kvm_msr_list __user *user_msr_list = argp;
2006 struct kvm_msr_list msr_list;
2010 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2013 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2014 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2017 if (n < msr_list.nmsrs)
2020 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2021 num_msrs_to_save * sizeof(u32)))
2023 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2025 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2030 case KVM_GET_SUPPORTED_CPUID: {
2031 struct kvm_cpuid2 __user *cpuid_arg = argp;
2032 struct kvm_cpuid2 cpuid;
2035 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2037 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2038 cpuid_arg->entries);
2043 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2048 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2051 mce_cap = KVM_MCE_CAP_SUPPORTED;
2053 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2065 static void wbinvd_ipi(void *garbage)
2070 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2072 return vcpu->kvm->arch.iommu_domain &&
2073 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2076 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2078 /* Address WBINVD may be executed by guest */
2079 if (need_emulate_wbinvd(vcpu)) {
2080 if (kvm_x86_ops->has_wbinvd_exit())
2081 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2082 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2083 smp_call_function_single(vcpu->cpu,
2084 wbinvd_ipi, NULL, 1);
2087 kvm_x86_ops->vcpu_load(vcpu, cpu);
2088 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2089 /* Make sure TSC doesn't go backwards */
2090 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2091 native_read_tsc() - vcpu->arch.last_host_tsc;
2093 mark_tsc_unstable("KVM discovered backwards TSC");
2094 if (check_tsc_unstable()) {
2095 kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2096 vcpu->arch.tsc_catchup = 1;
2097 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2099 if (vcpu->cpu != cpu)
2100 kvm_migrate_timers(vcpu);
2105 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2107 kvm_x86_ops->vcpu_put(vcpu);
2108 kvm_put_guest_fpu(vcpu);
2109 vcpu->arch.last_host_tsc = native_read_tsc();
2112 static int is_efer_nx(void)
2114 unsigned long long efer = 0;
2116 rdmsrl_safe(MSR_EFER, &efer);
2117 return efer & EFER_NX;
2120 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2123 struct kvm_cpuid_entry2 *e, *entry;
2126 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2127 e = &vcpu->arch.cpuid_entries[i];
2128 if (e->function == 0x80000001) {
2133 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2134 entry->edx &= ~(1 << 20);
2135 printk(KERN_INFO "kvm: guest NX capability removed\n");
2139 /* when an old userspace process fills a new kernel module */
2140 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2141 struct kvm_cpuid *cpuid,
2142 struct kvm_cpuid_entry __user *entries)
2145 struct kvm_cpuid_entry *cpuid_entries;
2148 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2151 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2155 if (copy_from_user(cpuid_entries, entries,
2156 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2158 for (i = 0; i < cpuid->nent; i++) {
2159 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2160 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2161 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2162 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2163 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2164 vcpu->arch.cpuid_entries[i].index = 0;
2165 vcpu->arch.cpuid_entries[i].flags = 0;
2166 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2167 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2168 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2170 vcpu->arch.cpuid_nent = cpuid->nent;
2171 cpuid_fix_nx_cap(vcpu);
2173 kvm_apic_set_version(vcpu);
2174 kvm_x86_ops->cpuid_update(vcpu);
2178 vfree(cpuid_entries);
2183 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2184 struct kvm_cpuid2 *cpuid,
2185 struct kvm_cpuid_entry2 __user *entries)
2190 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2193 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2194 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2196 vcpu->arch.cpuid_nent = cpuid->nent;
2197 kvm_apic_set_version(vcpu);
2198 kvm_x86_ops->cpuid_update(vcpu);
2206 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2207 struct kvm_cpuid2 *cpuid,
2208 struct kvm_cpuid_entry2 __user *entries)
2213 if (cpuid->nent < vcpu->arch.cpuid_nent)
2216 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2217 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2222 cpuid->nent = vcpu->arch.cpuid_nent;
2226 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2229 entry->function = function;
2230 entry->index = index;
2231 cpuid_count(entry->function, entry->index,
2232 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2236 #define F(x) bit(X86_FEATURE_##x)
2238 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2239 u32 index, int *nent, int maxnent)
2241 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2242 #ifdef CONFIG_X86_64
2243 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2245 unsigned f_lm = F(LM);
2247 unsigned f_gbpages = 0;
2250 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2253 const u32 kvm_supported_word0_x86_features =
2254 F(FPU) | F(VME) | F(DE) | F(PSE) |
2255 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2256 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2257 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2258 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2259 0 /* Reserved, DS, ACPI */ | F(MMX) |
2260 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2261 0 /* HTT, TM, Reserved, PBE */;
2262 /* cpuid 0x80000001.edx */
2263 const u32 kvm_supported_word1_x86_features =
2264 F(FPU) | F(VME) | F(DE) | F(PSE) |
2265 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2266 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2267 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2268 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2269 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2270 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2271 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2273 const u32 kvm_supported_word4_x86_features =
2274 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
2275 0 /* DS-CPL, VMX, SMX, EST */ |
2276 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2277 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2278 0 /* Reserved, DCA */ | F(XMM4_1) |
2279 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2280 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2282 /* cpuid 0x80000001.ecx */
2283 const u32 kvm_supported_word6_x86_features =
2284 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2285 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2286 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2287 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2289 /* all calls to cpuid_count() should be made on the same cpu */
2291 do_cpuid_1_ent(entry, function, index);
2296 entry->eax = min(entry->eax, (u32)0xd);
2299 entry->edx &= kvm_supported_word0_x86_features;
2300 entry->ecx &= kvm_supported_word4_x86_features;
2301 /* we support x2apic emulation even if host does not support
2302 * it since we emulate x2apic in software */
2303 entry->ecx |= F(X2APIC);
2305 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2306 * may return different values. This forces us to get_cpu() before
2307 * issuing the first command, and also to emulate this annoying behavior
2308 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2310 int t, times = entry->eax & 0xff;
2312 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2313 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2314 for (t = 1; t < times && *nent < maxnent; ++t) {
2315 do_cpuid_1_ent(&entry[t], function, 0);
2316 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2321 /* function 4 and 0xb have additional index. */
2325 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2326 /* read more entries until cache_type is zero */
2327 for (i = 1; *nent < maxnent; ++i) {
2328 cache_type = entry[i - 1].eax & 0x1f;
2331 do_cpuid_1_ent(&entry[i], function, i);
2333 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2341 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2342 /* read more entries until level_type is zero */
2343 for (i = 1; *nent < maxnent; ++i) {
2344 level_type = entry[i - 1].ecx & 0xff00;
2347 do_cpuid_1_ent(&entry[i], function, i);
2349 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2357 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2358 for (i = 1; *nent < maxnent; ++i) {
2359 if (entry[i - 1].eax == 0 && i != 2)
2361 do_cpuid_1_ent(&entry[i], function, i);
2363 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2368 case KVM_CPUID_SIGNATURE: {
2369 char signature[12] = "KVMKVMKVM\0\0";
2370 u32 *sigptr = (u32 *)signature;
2372 entry->ebx = sigptr[0];
2373 entry->ecx = sigptr[1];
2374 entry->edx = sigptr[2];
2377 case KVM_CPUID_FEATURES:
2378 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2379 (1 << KVM_FEATURE_NOP_IO_DELAY) |
2380 (1 << KVM_FEATURE_CLOCKSOURCE2) |
2381 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2387 entry->eax = min(entry->eax, 0x8000001a);
2390 entry->edx &= kvm_supported_word1_x86_features;
2391 entry->ecx &= kvm_supported_word6_x86_features;
2395 kvm_x86_ops->set_supported_cpuid(function, entry);
2402 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2403 struct kvm_cpuid_entry2 __user *entries)
2405 struct kvm_cpuid_entry2 *cpuid_entries;
2406 int limit, nent = 0, r = -E2BIG;
2409 if (cpuid->nent < 1)
2411 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2412 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2414 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2418 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2419 limit = cpuid_entries[0].eax;
2420 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2421 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2422 &nent, cpuid->nent);
2424 if (nent >= cpuid->nent)
2427 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2428 limit = cpuid_entries[nent - 1].eax;
2429 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2430 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2431 &nent, cpuid->nent);
2436 if (nent >= cpuid->nent)
2439 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2443 if (nent >= cpuid->nent)
2446 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2450 if (nent >= cpuid->nent)
2454 if (copy_to_user(entries, cpuid_entries,
2455 nent * sizeof(struct kvm_cpuid_entry2)))
2461 vfree(cpuid_entries);
2466 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2467 struct kvm_lapic_state *s)
2469 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2474 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2475 struct kvm_lapic_state *s)
2477 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2478 kvm_apic_post_state_restore(vcpu);
2479 update_cr8_intercept(vcpu);
2484 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2485 struct kvm_interrupt *irq)
2487 if (irq->irq < 0 || irq->irq >= 256)
2489 if (irqchip_in_kernel(vcpu->kvm))
2492 kvm_queue_interrupt(vcpu, irq->irq, false);
2493 kvm_make_request(KVM_REQ_EVENT, vcpu);
2498 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2500 kvm_inject_nmi(vcpu);
2505 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2506 struct kvm_tpr_access_ctl *tac)
2510 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2514 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2518 unsigned bank_num = mcg_cap & 0xff, bank;
2521 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2523 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2526 vcpu->arch.mcg_cap = mcg_cap;
2527 /* Init IA32_MCG_CTL to all 1s */
2528 if (mcg_cap & MCG_CTL_P)
2529 vcpu->arch.mcg_ctl = ~(u64)0;
2530 /* Init IA32_MCi_CTL to all 1s */
2531 for (bank = 0; bank < bank_num; bank++)
2532 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2537 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2538 struct kvm_x86_mce *mce)
2540 u64 mcg_cap = vcpu->arch.mcg_cap;
2541 unsigned bank_num = mcg_cap & 0xff;
2542 u64 *banks = vcpu->arch.mce_banks;
2544 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2547 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2548 * reporting is disabled
2550 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2551 vcpu->arch.mcg_ctl != ~(u64)0)
2553 banks += 4 * mce->bank;
2555 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2556 * reporting is disabled for the bank
2558 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2560 if (mce->status & MCI_STATUS_UC) {
2561 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2562 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2563 printk(KERN_DEBUG "kvm: set_mce: "
2564 "injects mce exception while "
2565 "previous one is in progress!\n");
2566 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2569 if (banks[1] & MCI_STATUS_VAL)
2570 mce->status |= MCI_STATUS_OVER;
2571 banks[2] = mce->addr;
2572 banks[3] = mce->misc;
2573 vcpu->arch.mcg_status = mce->mcg_status;
2574 banks[1] = mce->status;
2575 kvm_queue_exception(vcpu, MC_VECTOR);
2576 } else if (!(banks[1] & MCI_STATUS_VAL)
2577 || !(banks[1] & MCI_STATUS_UC)) {
2578 if (banks[1] & MCI_STATUS_VAL)
2579 mce->status |= MCI_STATUS_OVER;
2580 banks[2] = mce->addr;
2581 banks[3] = mce->misc;
2582 banks[1] = mce->status;
2584 banks[1] |= MCI_STATUS_OVER;
2588 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2589 struct kvm_vcpu_events *events)
2591 events->exception.injected =
2592 vcpu->arch.exception.pending &&
2593 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2594 events->exception.nr = vcpu->arch.exception.nr;
2595 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2596 events->exception.pad = 0;
2597 events->exception.error_code = vcpu->arch.exception.error_code;
2599 events->interrupt.injected =
2600 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2601 events->interrupt.nr = vcpu->arch.interrupt.nr;
2602 events->interrupt.soft = 0;
2603 events->interrupt.shadow =
2604 kvm_x86_ops->get_interrupt_shadow(vcpu,
2605 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2607 events->nmi.injected = vcpu->arch.nmi_injected;
2608 events->nmi.pending = vcpu->arch.nmi_pending;
2609 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2610 events->nmi.pad = 0;
2612 events->sipi_vector = vcpu->arch.sipi_vector;
2614 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2615 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2616 | KVM_VCPUEVENT_VALID_SHADOW);
2617 memset(&events->reserved, 0, sizeof(events->reserved));
2620 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2621 struct kvm_vcpu_events *events)
2623 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2624 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2625 | KVM_VCPUEVENT_VALID_SHADOW))
2628 vcpu->arch.exception.pending = events->exception.injected;
2629 vcpu->arch.exception.nr = events->exception.nr;
2630 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2631 vcpu->arch.exception.error_code = events->exception.error_code;
2633 vcpu->arch.interrupt.pending = events->interrupt.injected;
2634 vcpu->arch.interrupt.nr = events->interrupt.nr;
2635 vcpu->arch.interrupt.soft = events->interrupt.soft;
2636 if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
2637 kvm_pic_clear_isr_ack(vcpu->kvm);
2638 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2639 kvm_x86_ops->set_interrupt_shadow(vcpu,
2640 events->interrupt.shadow);
2642 vcpu->arch.nmi_injected = events->nmi.injected;
2643 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2644 vcpu->arch.nmi_pending = events->nmi.pending;
2645 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2647 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2648 vcpu->arch.sipi_vector = events->sipi_vector;
2650 kvm_make_request(KVM_REQ_EVENT, vcpu);
2655 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2656 struct kvm_debugregs *dbgregs)
2658 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2659 dbgregs->dr6 = vcpu->arch.dr6;
2660 dbgregs->dr7 = vcpu->arch.dr7;
2662 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2665 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2666 struct kvm_debugregs *dbgregs)
2671 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2672 vcpu->arch.dr6 = dbgregs->dr6;
2673 vcpu->arch.dr7 = dbgregs->dr7;
2678 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2679 struct kvm_xsave *guest_xsave)
2682 memcpy(guest_xsave->region,
2683 &vcpu->arch.guest_fpu.state->xsave,
2686 memcpy(guest_xsave->region,
2687 &vcpu->arch.guest_fpu.state->fxsave,
2688 sizeof(struct i387_fxsave_struct));
2689 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2694 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2695 struct kvm_xsave *guest_xsave)
2698 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2701 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2702 guest_xsave->region, xstate_size);
2704 if (xstate_bv & ~XSTATE_FPSSE)
2706 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2707 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2712 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2713 struct kvm_xcrs *guest_xcrs)
2715 if (!cpu_has_xsave) {
2716 guest_xcrs->nr_xcrs = 0;
2720 guest_xcrs->nr_xcrs = 1;
2721 guest_xcrs->flags = 0;
2722 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2723 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2726 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2727 struct kvm_xcrs *guest_xcrs)
2734 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2737 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2738 /* Only support XCR0 currently */
2739 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2740 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2741 guest_xcrs->xcrs[0].value);
2749 long kvm_arch_vcpu_ioctl(struct file *filp,
2750 unsigned int ioctl, unsigned long arg)
2752 struct kvm_vcpu *vcpu = filp->private_data;
2753 void __user *argp = (void __user *)arg;
2756 struct kvm_lapic_state *lapic;
2757 struct kvm_xsave *xsave;
2758 struct kvm_xcrs *xcrs;
2764 case KVM_GET_LAPIC: {
2766 if (!vcpu->arch.apic)
2768 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2773 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
2777 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
2782 case KVM_SET_LAPIC: {
2784 if (!vcpu->arch.apic)
2786 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2791 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
2793 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
2799 case KVM_INTERRUPT: {
2800 struct kvm_interrupt irq;
2803 if (copy_from_user(&irq, argp, sizeof irq))
2805 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2812 r = kvm_vcpu_ioctl_nmi(vcpu);
2818 case KVM_SET_CPUID: {
2819 struct kvm_cpuid __user *cpuid_arg = argp;
2820 struct kvm_cpuid cpuid;
2823 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2825 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2830 case KVM_SET_CPUID2: {
2831 struct kvm_cpuid2 __user *cpuid_arg = argp;
2832 struct kvm_cpuid2 cpuid;
2835 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2837 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2838 cpuid_arg->entries);
2843 case KVM_GET_CPUID2: {
2844 struct kvm_cpuid2 __user *cpuid_arg = argp;
2845 struct kvm_cpuid2 cpuid;
2848 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2850 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2851 cpuid_arg->entries);
2855 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2861 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2864 r = msr_io(vcpu, argp, do_set_msr, 0);
2866 case KVM_TPR_ACCESS_REPORTING: {
2867 struct kvm_tpr_access_ctl tac;
2870 if (copy_from_user(&tac, argp, sizeof tac))
2872 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2876 if (copy_to_user(argp, &tac, sizeof tac))
2881 case KVM_SET_VAPIC_ADDR: {
2882 struct kvm_vapic_addr va;
2885 if (!irqchip_in_kernel(vcpu->kvm))
2888 if (copy_from_user(&va, argp, sizeof va))
2891 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2894 case KVM_X86_SETUP_MCE: {
2898 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2900 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2903 case KVM_X86_SET_MCE: {
2904 struct kvm_x86_mce mce;
2907 if (copy_from_user(&mce, argp, sizeof mce))
2909 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2912 case KVM_GET_VCPU_EVENTS: {
2913 struct kvm_vcpu_events events;
2915 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2918 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2923 case KVM_SET_VCPU_EVENTS: {
2924 struct kvm_vcpu_events events;
2927 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2930 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2933 case KVM_GET_DEBUGREGS: {
2934 struct kvm_debugregs dbgregs;
2936 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2939 if (copy_to_user(argp, &dbgregs,
2940 sizeof(struct kvm_debugregs)))
2945 case KVM_SET_DEBUGREGS: {
2946 struct kvm_debugregs dbgregs;
2949 if (copy_from_user(&dbgregs, argp,
2950 sizeof(struct kvm_debugregs)))
2953 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2956 case KVM_GET_XSAVE: {
2957 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2962 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
2965 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
2970 case KVM_SET_XSAVE: {
2971 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2977 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
2980 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
2983 case KVM_GET_XCRS: {
2984 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2989 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
2992 if (copy_to_user(argp, u.xcrs,
2993 sizeof(struct kvm_xcrs)))
2998 case KVM_SET_XCRS: {
2999 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3005 if (copy_from_user(u.xcrs, argp,
3006 sizeof(struct kvm_xcrs)))
3009 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3020 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3024 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3026 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3030 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3033 kvm->arch.ept_identity_map_addr = ident_addr;
3037 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3038 u32 kvm_nr_mmu_pages)
3040 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3043 mutex_lock(&kvm->slots_lock);
3044 spin_lock(&kvm->mmu_lock);
3046 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3047 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3049 spin_unlock(&kvm->mmu_lock);
3050 mutex_unlock(&kvm->slots_lock);
3054 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3056 return kvm->arch.n_max_mmu_pages;
3059 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3064 switch (chip->chip_id) {
3065 case KVM_IRQCHIP_PIC_MASTER:
3066 memcpy(&chip->chip.pic,
3067 &pic_irqchip(kvm)->pics[0],
3068 sizeof(struct kvm_pic_state));
3070 case KVM_IRQCHIP_PIC_SLAVE:
3071 memcpy(&chip->chip.pic,
3072 &pic_irqchip(kvm)->pics[1],
3073 sizeof(struct kvm_pic_state));
3075 case KVM_IRQCHIP_IOAPIC:
3076 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3085 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3090 switch (chip->chip_id) {
3091 case KVM_IRQCHIP_PIC_MASTER:
3092 spin_lock(&pic_irqchip(kvm)->lock);
3093 memcpy(&pic_irqchip(kvm)->pics[0],
3095 sizeof(struct kvm_pic_state));
3096 spin_unlock(&pic_irqchip(kvm)->lock);
3098 case KVM_IRQCHIP_PIC_SLAVE:
3099 spin_lock(&pic_irqchip(kvm)->lock);
3100 memcpy(&pic_irqchip(kvm)->pics[1],
3102 sizeof(struct kvm_pic_state));
3103 spin_unlock(&pic_irqchip(kvm)->lock);
3105 case KVM_IRQCHIP_IOAPIC:
3106 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3112 kvm_pic_update_irq(pic_irqchip(kvm));
3116 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3120 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3121 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3122 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3126 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3130 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3131 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3132 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3133 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3137 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3141 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3142 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3143 sizeof(ps->channels));
3144 ps->flags = kvm->arch.vpit->pit_state.flags;
3145 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3146 memset(&ps->reserved, 0, sizeof(ps->reserved));
3150 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3152 int r = 0, start = 0;
3153 u32 prev_legacy, cur_legacy;
3154 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3155 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3156 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3157 if (!prev_legacy && cur_legacy)
3159 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3160 sizeof(kvm->arch.vpit->pit_state.channels));
3161 kvm->arch.vpit->pit_state.flags = ps->flags;
3162 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3163 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3167 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3168 struct kvm_reinject_control *control)
3170 if (!kvm->arch.vpit)
3172 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3173 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3174 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3179 * Get (and clear) the dirty memory log for a memory slot.
3181 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3182 struct kvm_dirty_log *log)
3185 struct kvm_memory_slot *memslot;
3187 unsigned long is_dirty = 0;
3189 mutex_lock(&kvm->slots_lock);
3192 if (log->slot >= KVM_MEMORY_SLOTS)
3195 memslot = &kvm->memslots->memslots[log->slot];
3197 if (!memslot->dirty_bitmap)
3200 n = kvm_dirty_bitmap_bytes(memslot);
3202 for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3203 is_dirty = memslot->dirty_bitmap[i];
3205 /* If nothing is dirty, don't bother messing with page tables. */
3207 struct kvm_memslots *slots, *old_slots;
3208 unsigned long *dirty_bitmap;
3211 dirty_bitmap = vmalloc(n);
3214 memset(dirty_bitmap, 0, n);
3217 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3219 vfree(dirty_bitmap);
3222 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3223 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3224 slots->generation++;
3226 old_slots = kvm->memslots;
3227 rcu_assign_pointer(kvm->memslots, slots);
3228 synchronize_srcu_expedited(&kvm->srcu);
3229 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3232 spin_lock(&kvm->mmu_lock);
3233 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3234 spin_unlock(&kvm->mmu_lock);
3237 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n)) {
3238 vfree(dirty_bitmap);
3241 vfree(dirty_bitmap);
3244 if (clear_user(log->dirty_bitmap, n))
3250 mutex_unlock(&kvm->slots_lock);
3254 long kvm_arch_vm_ioctl(struct file *filp,
3255 unsigned int ioctl, unsigned long arg)
3257 struct kvm *kvm = filp->private_data;
3258 void __user *argp = (void __user *)arg;
3261 * This union makes it completely explicit to gcc-3.x
3262 * that these two variables' stack usage should be
3263 * combined, not added together.
3266 struct kvm_pit_state ps;
3267 struct kvm_pit_state2 ps2;
3268 struct kvm_pit_config pit_config;
3272 case KVM_SET_TSS_ADDR:
3273 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3277 case KVM_SET_IDENTITY_MAP_ADDR: {
3281 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3283 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3288 case KVM_SET_NR_MMU_PAGES:
3289 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3293 case KVM_GET_NR_MMU_PAGES:
3294 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3296 case KVM_CREATE_IRQCHIP: {
3297 struct kvm_pic *vpic;
3299 mutex_lock(&kvm->lock);
3302 goto create_irqchip_unlock;
3304 vpic = kvm_create_pic(kvm);
3306 r = kvm_ioapic_init(kvm);
3308 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3311 goto create_irqchip_unlock;
3314 goto create_irqchip_unlock;
3316 kvm->arch.vpic = vpic;
3318 r = kvm_setup_default_irq_routing(kvm);
3320 mutex_lock(&kvm->irq_lock);
3321 kvm_ioapic_destroy(kvm);
3322 kvm_destroy_pic(kvm);
3323 mutex_unlock(&kvm->irq_lock);
3325 create_irqchip_unlock:
3326 mutex_unlock(&kvm->lock);
3329 case KVM_CREATE_PIT:
3330 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3332 case KVM_CREATE_PIT2:
3334 if (copy_from_user(&u.pit_config, argp,
3335 sizeof(struct kvm_pit_config)))
3338 mutex_lock(&kvm->slots_lock);
3341 goto create_pit_unlock;
3343 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3347 mutex_unlock(&kvm->slots_lock);
3349 case KVM_IRQ_LINE_STATUS:
3350 case KVM_IRQ_LINE: {
3351 struct kvm_irq_level irq_event;
3354 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3357 if (irqchip_in_kernel(kvm)) {
3359 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3360 irq_event.irq, irq_event.level);
3361 if (ioctl == KVM_IRQ_LINE_STATUS) {
3363 irq_event.status = status;
3364 if (copy_to_user(argp, &irq_event,
3372 case KVM_GET_IRQCHIP: {
3373 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3374 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3380 if (copy_from_user(chip, argp, sizeof *chip))
3381 goto get_irqchip_out;
3383 if (!irqchip_in_kernel(kvm))
3384 goto get_irqchip_out;
3385 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3387 goto get_irqchip_out;
3389 if (copy_to_user(argp, chip, sizeof *chip))
3390 goto get_irqchip_out;
3398 case KVM_SET_IRQCHIP: {
3399 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3400 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3406 if (copy_from_user(chip, argp, sizeof *chip))
3407 goto set_irqchip_out;
3409 if (!irqchip_in_kernel(kvm))
3410 goto set_irqchip_out;
3411 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3413 goto set_irqchip_out;
3423 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3426 if (!kvm->arch.vpit)
3428 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3432 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3439 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3442 if (!kvm->arch.vpit)
3444 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3450 case KVM_GET_PIT2: {
3452 if (!kvm->arch.vpit)
3454 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3458 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3463 case KVM_SET_PIT2: {
3465 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3468 if (!kvm->arch.vpit)
3470 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3476 case KVM_REINJECT_CONTROL: {
3477 struct kvm_reinject_control control;
3479 if (copy_from_user(&control, argp, sizeof(control)))
3481 r = kvm_vm_ioctl_reinject(kvm, &control);
3487 case KVM_XEN_HVM_CONFIG: {
3489 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3490 sizeof(struct kvm_xen_hvm_config)))
3493 if (kvm->arch.xen_hvm_config.flags)
3498 case KVM_SET_CLOCK: {
3499 struct kvm_clock_data user_ns;
3504 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3512 local_irq_disable();
3513 now_ns = get_kernel_ns();
3514 delta = user_ns.clock - now_ns;
3516 kvm->arch.kvmclock_offset = delta;
3519 case KVM_GET_CLOCK: {
3520 struct kvm_clock_data user_ns;
3523 local_irq_disable();
3524 now_ns = get_kernel_ns();
3525 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3528 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3531 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3544 static void kvm_init_msr_list(void)
3549 /* skip the first msrs in the list. KVM-specific */
3550 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3551 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3554 msrs_to_save[j] = msrs_to_save[i];
3557 num_msrs_to_save = j;
3560 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3563 if (vcpu->arch.apic &&
3564 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
3567 return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3570 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3572 if (vcpu->arch.apic &&
3573 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
3576 return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3579 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3580 struct kvm_segment *var, int seg)
3582 kvm_x86_ops->set_segment(vcpu, var, seg);
3585 void kvm_get_segment(struct kvm_vcpu *vcpu,
3586 struct kvm_segment *var, int seg)
3588 kvm_x86_ops->get_segment(vcpu, var, seg);
3591 static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3596 static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3601 BUG_ON(!mmu_is_nested(vcpu));
3603 /* NPT walks are always user-walks */
3604 access |= PFERR_USER_MASK;
3605 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &error);
3606 if (t_gpa == UNMAPPED_GVA)
3607 vcpu->arch.fault.nested = true;
3612 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3614 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3615 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, error);
3618 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3620 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3621 access |= PFERR_FETCH_MASK;
3622 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, error);
3625 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3627 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3628 access |= PFERR_WRITE_MASK;
3629 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, error);
3632 /* uses this to access any guest's mapped memory without checking CPL */
3633 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3635 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, error);
3638 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3639 struct kvm_vcpu *vcpu, u32 access,
3643 int r = X86EMUL_CONTINUE;
3646 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3648 unsigned offset = addr & (PAGE_SIZE-1);
3649 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3652 if (gpa == UNMAPPED_GVA) {
3653 r = X86EMUL_PROPAGATE_FAULT;
3656 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3658 r = X86EMUL_IO_NEEDED;
3670 /* used for instruction fetching */
3671 static int kvm_fetch_guest_virt(gva_t addr, void *val, unsigned int bytes,
3672 struct kvm_vcpu *vcpu, u32 *error)
3674 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3675 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3676 access | PFERR_FETCH_MASK, error);
3679 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3680 struct kvm_vcpu *vcpu, u32 *error)
3682 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3683 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3687 static int kvm_read_guest_virt_system(gva_t addr, void *val, unsigned int bytes,
3688 struct kvm_vcpu *vcpu, u32 *error)
3690 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, error);
3693 static int kvm_write_guest_virt_system(gva_t addr, void *val,
3695 struct kvm_vcpu *vcpu,
3699 int r = X86EMUL_CONTINUE;
3702 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3705 unsigned offset = addr & (PAGE_SIZE-1);
3706 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3709 if (gpa == UNMAPPED_GVA) {
3710 r = X86EMUL_PROPAGATE_FAULT;
3713 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3715 r = X86EMUL_IO_NEEDED;
3727 static int emulator_read_emulated(unsigned long addr,
3730 unsigned int *error_code,
3731 struct kvm_vcpu *vcpu)
3735 if (vcpu->mmio_read_completed) {
3736 memcpy(val, vcpu->mmio_data, bytes);
3737 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3738 vcpu->mmio_phys_addr, *(u64 *)val);
3739 vcpu->mmio_read_completed = 0;
3740 return X86EMUL_CONTINUE;
3743 gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, error_code);
3745 if (gpa == UNMAPPED_GVA)
3746 return X86EMUL_PROPAGATE_FAULT;
3748 /* For APIC access vmexit */
3749 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3752 if (kvm_read_guest_virt(addr, val, bytes, vcpu, NULL)
3753 == X86EMUL_CONTINUE)
3754 return X86EMUL_CONTINUE;
3758 * Is this MMIO handled locally?
3760 if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
3761 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
3762 return X86EMUL_CONTINUE;
3765 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3767 vcpu->mmio_needed = 1;
3768 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3769 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3770 vcpu->run->mmio.len = vcpu->mmio_size = bytes;
3771 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 0;
3773 return X86EMUL_IO_NEEDED;
3776 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3777 const void *val, int bytes)
3781 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3784 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3788 static int emulator_write_emulated_onepage(unsigned long addr,
3791 unsigned int *error_code,
3792 struct kvm_vcpu *vcpu)
3796 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, error_code);
3798 if (gpa == UNMAPPED_GVA)
3799 return X86EMUL_PROPAGATE_FAULT;
3801 /* For APIC access vmexit */
3802 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3805 if (emulator_write_phys(vcpu, gpa, val, bytes))
3806 return X86EMUL_CONTINUE;
3809 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3811 * Is this MMIO handled locally?
3813 if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
3814 return X86EMUL_CONTINUE;
3816 vcpu->mmio_needed = 1;
3817 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3818 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3819 vcpu->run->mmio.len = vcpu->mmio_size = bytes;
3820 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 1;
3821 memcpy(vcpu->run->mmio.data, val, bytes);
3823 return X86EMUL_CONTINUE;
3826 int emulator_write_emulated(unsigned long addr,
3829 unsigned int *error_code,
3830 struct kvm_vcpu *vcpu)
3832 /* Crossing a page boundary? */
3833 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3836 now = -addr & ~PAGE_MASK;
3837 rc = emulator_write_emulated_onepage(addr, val, now, error_code,
3839 if (rc != X86EMUL_CONTINUE)
3845 return emulator_write_emulated_onepage(addr, val, bytes, error_code,
3849 #define CMPXCHG_TYPE(t, ptr, old, new) \
3850 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3852 #ifdef CONFIG_X86_64
3853 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3855 # define CMPXCHG64(ptr, old, new) \
3856 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
3859 static int emulator_cmpxchg_emulated(unsigned long addr,
3863 unsigned int *error_code,
3864 struct kvm_vcpu *vcpu)
3871 /* guests cmpxchg8b have to be emulated atomically */
3872 if (bytes > 8 || (bytes & (bytes - 1)))
3875 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3877 if (gpa == UNMAPPED_GVA ||
3878 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3881 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3884 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3885 if (is_error_page(page)) {
3886 kvm_release_page_clean(page);
3890 kaddr = kmap_atomic(page, KM_USER0);
3891 kaddr += offset_in_page(gpa);
3894 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
3897 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
3900 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
3903 exchanged = CMPXCHG64(kaddr, old, new);
3908 kunmap_atomic(kaddr, KM_USER0);
3909 kvm_release_page_dirty(page);
3912 return X86EMUL_CMPXCHG_FAILED;
3914 kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
3916 return X86EMUL_CONTINUE;
3919 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3921 return emulator_write_emulated(addr, new, bytes, error_code, vcpu);
3924 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3926 /* TODO: String I/O for in kernel device */
3929 if (vcpu->arch.pio.in)
3930 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3931 vcpu->arch.pio.size, pd);
3933 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3934 vcpu->arch.pio.port, vcpu->arch.pio.size,
3940 static int emulator_pio_in_emulated(int size, unsigned short port, void *val,
3941 unsigned int count, struct kvm_vcpu *vcpu)
3943 if (vcpu->arch.pio.count)
3946 trace_kvm_pio(0, port, size, 1);
3948 vcpu->arch.pio.port = port;
3949 vcpu->arch.pio.in = 1;
3950 vcpu->arch.pio.count = count;
3951 vcpu->arch.pio.size = size;
3953 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3955 memcpy(val, vcpu->arch.pio_data, size * count);
3956 vcpu->arch.pio.count = 0;
3960 vcpu->run->exit_reason = KVM_EXIT_IO;
3961 vcpu->run->io.direction = KVM_EXIT_IO_IN;
3962 vcpu->run->io.size = size;
3963 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3964 vcpu->run->io.count = count;
3965 vcpu->run->io.port = port;
3970 static int emulator_pio_out_emulated(int size, unsigned short port,
3971 const void *val, unsigned int count,
3972 struct kvm_vcpu *vcpu)
3974 trace_kvm_pio(1, port, size, 1);
3976 vcpu->arch.pio.port = port;
3977 vcpu->arch.pio.in = 0;
3978 vcpu->arch.pio.count = count;
3979 vcpu->arch.pio.size = size;
3981 memcpy(vcpu->arch.pio_data, val, size * count);
3983 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3984 vcpu->arch.pio.count = 0;
3988 vcpu->run->exit_reason = KVM_EXIT_IO;
3989 vcpu->run->io.direction = KVM_EXIT_IO_OUT;
3990 vcpu->run->io.size = size;
3991 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3992 vcpu->run->io.count = count;
3993 vcpu->run->io.port = port;
3998 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4000 return kvm_x86_ops->get_segment_base(vcpu, seg);
4003 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
4005 kvm_mmu_invlpg(vcpu, address);
4006 return X86EMUL_CONTINUE;
4009 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4011 if (!need_emulate_wbinvd(vcpu))
4012 return X86EMUL_CONTINUE;
4014 if (kvm_x86_ops->has_wbinvd_exit()) {
4016 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4017 wbinvd_ipi, NULL, 1);
4019 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4022 return X86EMUL_CONTINUE;
4024 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4026 int emulate_clts(struct kvm_vcpu *vcpu)
4028 kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4029 kvm_x86_ops->fpu_activate(vcpu);
4030 return X86EMUL_CONTINUE;
4033 int emulator_get_dr(int dr, unsigned long *dest, struct kvm_vcpu *vcpu)
4035 return _kvm_get_dr(vcpu, dr, dest);
4038 int emulator_set_dr(int dr, unsigned long value, struct kvm_vcpu *vcpu)
4041 return __kvm_set_dr(vcpu, dr, value);
4044 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4046 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4049 static unsigned long emulator_get_cr(int cr, struct kvm_vcpu *vcpu)
4051 unsigned long value;
4055 value = kvm_read_cr0(vcpu);
4058 value = vcpu->arch.cr2;
4061 value = vcpu->arch.cr3;
4064 value = kvm_read_cr4(vcpu);
4067 value = kvm_get_cr8(vcpu);
4070 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4077 static int emulator_set_cr(int cr, unsigned long val, struct kvm_vcpu *vcpu)
4083 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4086 vcpu->arch.cr2 = val;
4089 res = kvm_set_cr3(vcpu, val);
4092 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4095 res = __kvm_set_cr8(vcpu, val & 0xfUL);
4098 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4105 static int emulator_get_cpl(struct kvm_vcpu *vcpu)
4107 return kvm_x86_ops->get_cpl(vcpu);
4110 static void emulator_get_gdt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
4112 kvm_x86_ops->get_gdt(vcpu, dt);
4115 static void emulator_get_idt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
4117 kvm_x86_ops->get_idt(vcpu, dt);
4120 static unsigned long emulator_get_cached_segment_base(int seg,
4121 struct kvm_vcpu *vcpu)
4123 return get_segment_base(vcpu, seg);
4126 static bool emulator_get_cached_descriptor(struct desc_struct *desc, int seg,
4127 struct kvm_vcpu *vcpu)
4129 struct kvm_segment var;
4131 kvm_get_segment(vcpu, &var, seg);
4138 set_desc_limit(desc, var.limit);
4139 set_desc_base(desc, (unsigned long)var.base);
4140 desc->type = var.type;
4142 desc->dpl = var.dpl;
4143 desc->p = var.present;
4144 desc->avl = var.avl;
4152 static void emulator_set_cached_descriptor(struct desc_struct *desc, int seg,
4153 struct kvm_vcpu *vcpu)
4155 struct kvm_segment var;
4157 /* needed to preserve selector */
4158 kvm_get_segment(vcpu, &var, seg);
4160 var.base = get_desc_base(desc);
4161 var.limit = get_desc_limit(desc);
4163 var.limit = (var.limit << 12) | 0xfff;
4164 var.type = desc->type;
4165 var.present = desc->p;
4166 var.dpl = desc->dpl;
4171 var.avl = desc->avl;
4172 var.present = desc->p;
4173 var.unusable = !var.present;
4176 kvm_set_segment(vcpu, &var, seg);
4180 static u16 emulator_get_segment_selector(int seg, struct kvm_vcpu *vcpu)
4182 struct kvm_segment kvm_seg;
4184 kvm_get_segment(vcpu, &kvm_seg, seg);
4185 return kvm_seg.selector;
4188 static void emulator_set_segment_selector(u16 sel, int seg,
4189 struct kvm_vcpu *vcpu)
4191 struct kvm_segment kvm_seg;
4193 kvm_get_segment(vcpu, &kvm_seg, seg);
4194 kvm_seg.selector = sel;
4195 kvm_set_segment(vcpu, &kvm_seg, seg);
4198 static struct x86_emulate_ops emulate_ops = {
4199 .read_std = kvm_read_guest_virt_system,
4200 .write_std = kvm_write_guest_virt_system,
4201 .fetch = kvm_fetch_guest_virt,
4202 .read_emulated = emulator_read_emulated,
4203 .write_emulated = emulator_write_emulated,
4204 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4205 .pio_in_emulated = emulator_pio_in_emulated,
4206 .pio_out_emulated = emulator_pio_out_emulated,
4207 .get_cached_descriptor = emulator_get_cached_descriptor,
4208 .set_cached_descriptor = emulator_set_cached_descriptor,
4209 .get_segment_selector = emulator_get_segment_selector,
4210 .set_segment_selector = emulator_set_segment_selector,
4211 .get_cached_segment_base = emulator_get_cached_segment_base,
4212 .get_gdt = emulator_get_gdt,
4213 .get_idt = emulator_get_idt,
4214 .get_cr = emulator_get_cr,
4215 .set_cr = emulator_set_cr,
4216 .cpl = emulator_get_cpl,
4217 .get_dr = emulator_get_dr,
4218 .set_dr = emulator_set_dr,
4219 .set_msr = kvm_set_msr,
4220 .get_msr = kvm_get_msr,
4223 static void cache_all_regs(struct kvm_vcpu *vcpu)
4225 kvm_register_read(vcpu, VCPU_REGS_RAX);
4226 kvm_register_read(vcpu, VCPU_REGS_RSP);
4227 kvm_register_read(vcpu, VCPU_REGS_RIP);
4228 vcpu->arch.regs_dirty = ~0;
4231 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4233 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4235 * an sti; sti; sequence only disable interrupts for the first
4236 * instruction. So, if the last instruction, be it emulated or
4237 * not, left the system with the INT_STI flag enabled, it
4238 * means that the last instruction is an sti. We should not
4239 * leave the flag on in this case. The same goes for mov ss
4241 if (!(int_shadow & mask))
4242 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4245 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4247 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4248 if (ctxt->exception == PF_VECTOR)
4249 kvm_propagate_fault(vcpu);
4250 else if (ctxt->error_code_valid)
4251 kvm_queue_exception_e(vcpu, ctxt->exception, ctxt->error_code);
4253 kvm_queue_exception(vcpu, ctxt->exception);
4256 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4258 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4261 cache_all_regs(vcpu);
4263 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4265 vcpu->arch.emulate_ctxt.vcpu = vcpu;
4266 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
4267 vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
4268 vcpu->arch.emulate_ctxt.mode =
4269 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4270 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
4271 ? X86EMUL_MODE_VM86 : cs_l
4272 ? X86EMUL_MODE_PROT64 : cs_db
4273 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
4274 memset(c, 0, sizeof(struct decode_cache));
4275 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4278 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq)
4280 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4283 init_emulate_ctxt(vcpu);
4285 vcpu->arch.emulate_ctxt.decode.op_bytes = 2;
4286 vcpu->arch.emulate_ctxt.decode.ad_bytes = 2;
4287 vcpu->arch.emulate_ctxt.decode.eip = vcpu->arch.emulate_ctxt.eip;
4288 ret = emulate_int_real(&vcpu->arch.emulate_ctxt, &emulate_ops, irq);
4290 if (ret != X86EMUL_CONTINUE)
4291 return EMULATE_FAIL;
4293 vcpu->arch.emulate_ctxt.eip = c->eip;
4294 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4295 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4296 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4298 if (irq == NMI_VECTOR)
4299 vcpu->arch.nmi_pending = false;
4301 vcpu->arch.interrupt.pending = false;
4303 return EMULATE_DONE;
4305 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4307 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4309 ++vcpu->stat.insn_emulation_fail;
4310 trace_kvm_emulate_insn_failed(vcpu);
4311 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4312 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4313 vcpu->run->internal.ndata = 0;
4314 kvm_queue_exception(vcpu, UD_VECTOR);
4315 return EMULATE_FAIL;
4318 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4326 * if emulation was due to access to shadowed page table
4327 * and it failed try to unshadow page and re-entetr the
4328 * guest to let CPU execute the instruction.
4330 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4333 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4335 if (gpa == UNMAPPED_GVA)
4336 return true; /* let cpu generate fault */
4338 if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4344 int emulate_instruction(struct kvm_vcpu *vcpu,
4350 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4352 kvm_clear_exception_queue(vcpu);
4353 vcpu->arch.mmio_fault_cr2 = cr2;
4355 * TODO: fix emulate.c to use guest_read/write_register
4356 * instead of direct ->regs accesses, can save hundred cycles
4357 * on Intel for instructions that don't read/change RSP, for
4360 cache_all_regs(vcpu);
4362 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4363 init_emulate_ctxt(vcpu);
4364 vcpu->arch.emulate_ctxt.interruptibility = 0;
4365 vcpu->arch.emulate_ctxt.exception = -1;
4366 vcpu->arch.emulate_ctxt.perm_ok = false;
4368 r = x86_decode_insn(&vcpu->arch.emulate_ctxt);
4369 if (r == X86EMUL_PROPAGATE_FAULT)
4372 trace_kvm_emulate_insn_start(vcpu);
4374 /* Only allow emulation of specific instructions on #UD
4375 * (namely VMMCALL, sysenter, sysexit, syscall)*/
4376 if (emulation_type & EMULTYPE_TRAP_UD) {
4378 return EMULATE_FAIL;
4380 case 0x01: /* VMMCALL */
4381 if (c->modrm_mod != 3 || c->modrm_rm != 1)
4382 return EMULATE_FAIL;
4384 case 0x34: /* sysenter */
4385 case 0x35: /* sysexit */
4386 if (c->modrm_mod != 0 || c->modrm_rm != 0)
4387 return EMULATE_FAIL;
4389 case 0x05: /* syscall */
4390 if (c->modrm_mod != 0 || c->modrm_rm != 0)
4391 return EMULATE_FAIL;
4394 return EMULATE_FAIL;
4397 if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
4398 return EMULATE_FAIL;
4401 ++vcpu->stat.insn_emulation;
4403 if (reexecute_instruction(vcpu, cr2))
4404 return EMULATE_DONE;
4405 if (emulation_type & EMULTYPE_SKIP)
4406 return EMULATE_FAIL;
4407 return handle_emulation_failure(vcpu);
4411 if (emulation_type & EMULTYPE_SKIP) {
4412 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
4413 return EMULATE_DONE;
4416 /* this is needed for vmware backdor interface to work since it
4417 changes registers values during IO operation */
4418 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4421 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt);
4423 if (r == EMULATION_FAILED) {
4424 if (reexecute_instruction(vcpu, cr2))
4425 return EMULATE_DONE;
4427 return handle_emulation_failure(vcpu);
4431 if (vcpu->arch.emulate_ctxt.exception >= 0) {
4432 inject_emulated_exception(vcpu);
4434 } else if (vcpu->arch.pio.count) {
4435 if (!vcpu->arch.pio.in)
4436 vcpu->arch.pio.count = 0;
4437 r = EMULATE_DO_MMIO;
4438 } else if (vcpu->mmio_needed) {
4439 if (vcpu->mmio_is_write)
4440 vcpu->mmio_needed = 0;
4441 r = EMULATE_DO_MMIO;
4442 } else if (r == EMULATION_RESTART)
4447 toggle_interruptibility(vcpu, vcpu->arch.emulate_ctxt.interruptibility);
4448 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4449 kvm_make_request(KVM_REQ_EVENT, vcpu);
4450 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4451 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4455 EXPORT_SYMBOL_GPL(emulate_instruction);
4457 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4459 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4460 int ret = emulator_pio_out_emulated(size, port, &val, 1, vcpu);
4461 /* do not return to emulator after return from userspace */
4462 vcpu->arch.pio.count = 0;
4465 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4467 static void tsc_bad(void *info)
4469 __get_cpu_var(cpu_tsc_khz) = 0;
4472 static void tsc_khz_changed(void *data)
4474 struct cpufreq_freqs *freq = data;
4475 unsigned long khz = 0;
4479 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4480 khz = cpufreq_quick_get(raw_smp_processor_id());
4483 __get_cpu_var(cpu_tsc_khz) = khz;
4486 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4489 struct cpufreq_freqs *freq = data;
4491 struct kvm_vcpu *vcpu;
4492 int i, send_ipi = 0;
4495 * We allow guests to temporarily run on slowing clocks,
4496 * provided we notify them after, or to run on accelerating
4497 * clocks, provided we notify them before. Thus time never
4500 * However, we have a problem. We can't atomically update
4501 * the frequency of a given CPU from this function; it is
4502 * merely a notifier, which can be called from any CPU.
4503 * Changing the TSC frequency at arbitrary points in time
4504 * requires a recomputation of local variables related to
4505 * the TSC for each VCPU. We must flag these local variables
4506 * to be updated and be sure the update takes place with the
4507 * new frequency before any guests proceed.
4509 * Unfortunately, the combination of hotplug CPU and frequency
4510 * change creates an intractable locking scenario; the order
4511 * of when these callouts happen is undefined with respect to
4512 * CPU hotplug, and they can race with each other. As such,
4513 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4514 * undefined; you can actually have a CPU frequency change take
4515 * place in between the computation of X and the setting of the
4516 * variable. To protect against this problem, all updates of
4517 * the per_cpu tsc_khz variable are done in an interrupt
4518 * protected IPI, and all callers wishing to update the value
4519 * must wait for a synchronous IPI to complete (which is trivial
4520 * if the caller is on the CPU already). This establishes the
4521 * necessary total order on variable updates.
4523 * Note that because a guest time update may take place
4524 * anytime after the setting of the VCPU's request bit, the
4525 * correct TSC value must be set before the request. However,
4526 * to ensure the update actually makes it to any guest which
4527 * starts running in hardware virtualization between the set
4528 * and the acquisition of the spinlock, we must also ping the
4529 * CPU after setting the request bit.
4533 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4535 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4538 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4540 spin_lock(&kvm_lock);
4541 list_for_each_entry(kvm, &vm_list, vm_list) {
4542 kvm_for_each_vcpu(i, vcpu, kvm) {
4543 if (vcpu->cpu != freq->cpu)
4545 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4546 if (vcpu->cpu != smp_processor_id())
4550 spin_unlock(&kvm_lock);
4552 if (freq->old < freq->new && send_ipi) {
4554 * We upscale the frequency. Must make the guest
4555 * doesn't see old kvmclock values while running with
4556 * the new frequency, otherwise we risk the guest sees
4557 * time go backwards.
4559 * In case we update the frequency for another cpu
4560 * (which might be in guest context) send an interrupt
4561 * to kick the cpu out of guest context. Next time
4562 * guest context is entered kvmclock will be updated,
4563 * so the guest will not see stale values.
4565 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4570 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4571 .notifier_call = kvmclock_cpufreq_notifier
4574 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4575 unsigned long action, void *hcpu)
4577 unsigned int cpu = (unsigned long)hcpu;
4581 case CPU_DOWN_FAILED:
4582 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4584 case CPU_DOWN_PREPARE:
4585 smp_call_function_single(cpu, tsc_bad, NULL, 1);
4591 static struct notifier_block kvmclock_cpu_notifier_block = {
4592 .notifier_call = kvmclock_cpu_notifier,
4593 .priority = -INT_MAX
4596 static void kvm_timer_init(void)
4600 max_tsc_khz = tsc_khz;
4601 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4602 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4603 #ifdef CONFIG_CPU_FREQ
4604 struct cpufreq_policy policy;
4605 memset(&policy, 0, sizeof(policy));
4607 cpufreq_get_policy(&policy, cpu);
4608 if (policy.cpuinfo.max_freq)
4609 max_tsc_khz = policy.cpuinfo.max_freq;
4612 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4613 CPUFREQ_TRANSITION_NOTIFIER);
4615 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
4616 for_each_online_cpu(cpu)
4617 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4620 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4622 static int kvm_is_in_guest(void)
4624 return percpu_read(current_vcpu) != NULL;
4627 static int kvm_is_user_mode(void)
4631 if (percpu_read(current_vcpu))
4632 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
4634 return user_mode != 0;
4637 static unsigned long kvm_get_guest_ip(void)
4639 unsigned long ip = 0;
4641 if (percpu_read(current_vcpu))
4642 ip = kvm_rip_read(percpu_read(current_vcpu));
4647 static struct perf_guest_info_callbacks kvm_guest_cbs = {
4648 .is_in_guest = kvm_is_in_guest,
4649 .is_user_mode = kvm_is_user_mode,
4650 .get_guest_ip = kvm_get_guest_ip,
4653 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4655 percpu_write(current_vcpu, vcpu);
4657 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4659 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4661 percpu_write(current_vcpu, NULL);
4663 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4665 int kvm_arch_init(void *opaque)
4668 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4671 printk(KERN_ERR "kvm: already loaded the other module\n");
4676 if (!ops->cpu_has_kvm_support()) {
4677 printk(KERN_ERR "kvm: no hardware support\n");
4681 if (ops->disabled_by_bios()) {
4682 printk(KERN_ERR "kvm: disabled by bios\n");
4687 r = kvm_mmu_module_init();
4691 kvm_init_msr_list();
4694 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
4695 kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
4696 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4697 PT_DIRTY_MASK, PT64_NX_MASK, 0);
4701 perf_register_guest_info_callbacks(&kvm_guest_cbs);
4704 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
4712 void kvm_arch_exit(void)
4714 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4716 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4717 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4718 CPUFREQ_TRANSITION_NOTIFIER);
4719 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4721 kvm_mmu_module_exit();
4724 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4726 ++vcpu->stat.halt_exits;
4727 if (irqchip_in_kernel(vcpu->kvm)) {
4728 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
4731 vcpu->run->exit_reason = KVM_EXIT_HLT;
4735 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4737 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
4740 if (is_long_mode(vcpu))
4743 return a0 | ((gpa_t)a1 << 32);
4746 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4748 u64 param, ingpa, outgpa, ret;
4749 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4750 bool fast, longmode;
4754 * hypercall generates UD from non zero cpl and real mode
4757 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
4758 kvm_queue_exception(vcpu, UD_VECTOR);
4762 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4763 longmode = is_long_mode(vcpu) && cs_l == 1;
4766 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4767 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4768 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4769 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4770 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4771 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
4773 #ifdef CONFIG_X86_64
4775 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4776 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4777 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4781 code = param & 0xffff;
4782 fast = (param >> 16) & 0x1;
4783 rep_cnt = (param >> 32) & 0xfff;
4784 rep_idx = (param >> 48) & 0xfff;
4786 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
4789 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
4790 kvm_vcpu_on_spin(vcpu);
4793 res = HV_STATUS_INVALID_HYPERCALL_CODE;
4797 ret = res | (((u64)rep_done & 0xfff) << 32);
4799 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4801 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
4802 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
4808 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
4810 unsigned long nr, a0, a1, a2, a3, ret;
4813 if (kvm_hv_hypercall_enabled(vcpu->kvm))
4814 return kvm_hv_hypercall(vcpu);
4816 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
4817 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
4818 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
4819 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
4820 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
4822 trace_kvm_hypercall(nr, a0, a1, a2, a3);
4824 if (!is_long_mode(vcpu)) {
4832 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
4838 case KVM_HC_VAPIC_POLL_IRQ:
4842 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
4849 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4850 ++vcpu->stat.hypercalls;
4853 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
4855 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
4857 char instruction[3];
4858 unsigned long rip = kvm_rip_read(vcpu);
4861 * Blow out the MMU to ensure that no other VCPU has an active mapping
4862 * to ensure that the updated hypercall appears atomically across all
4865 kvm_mmu_zap_all(vcpu->kvm);
4867 kvm_x86_ops->patch_hypercall(vcpu, instruction);
4869 return emulator_write_emulated(rip, instruction, 3, NULL, vcpu);
4872 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4874 struct desc_ptr dt = { limit, base };
4876 kvm_x86_ops->set_gdt(vcpu, &dt);
4879 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4881 struct desc_ptr dt = { limit, base };
4883 kvm_x86_ops->set_idt(vcpu, &dt);
4886 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
4888 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
4889 int j, nent = vcpu->arch.cpuid_nent;
4891 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
4892 /* when no next entry is found, the current entry[i] is reselected */
4893 for (j = i + 1; ; j = (j + 1) % nent) {
4894 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
4895 if (ej->function == e->function) {
4896 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
4900 return 0; /* silence gcc, even though control never reaches here */
4903 /* find an entry with matching function, matching index (if needed), and that
4904 * should be read next (if it's stateful) */
4905 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
4906 u32 function, u32 index)
4908 if (e->function != function)
4910 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
4912 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
4913 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
4918 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
4919 u32 function, u32 index)
4922 struct kvm_cpuid_entry2 *best = NULL;
4924 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
4925 struct kvm_cpuid_entry2 *e;
4927 e = &vcpu->arch.cpuid_entries[i];
4928 if (is_matching_cpuid_entry(e, function, index)) {
4929 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
4930 move_to_next_stateful_cpuid_entry(vcpu, i);
4935 * Both basic or both extended?
4937 if (((e->function ^ function) & 0x80000000) == 0)
4938 if (!best || e->function > best->function)
4943 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
4945 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
4947 struct kvm_cpuid_entry2 *best;
4949 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
4950 if (!best || best->eax < 0x80000008)
4952 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
4954 return best->eax & 0xff;
4959 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
4961 u32 function, index;
4962 struct kvm_cpuid_entry2 *best;
4964 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
4965 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4966 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
4967 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
4968 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
4969 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
4970 best = kvm_find_cpuid_entry(vcpu, function, index);
4972 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
4973 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
4974 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
4975 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
4977 kvm_x86_ops->skip_emulated_instruction(vcpu);
4978 trace_kvm_cpuid(function,
4979 kvm_register_read(vcpu, VCPU_REGS_RAX),
4980 kvm_register_read(vcpu, VCPU_REGS_RBX),
4981 kvm_register_read(vcpu, VCPU_REGS_RCX),
4982 kvm_register_read(vcpu, VCPU_REGS_RDX));
4984 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
4987 * Check if userspace requested an interrupt window, and that the
4988 * interrupt window is open.
4990 * No need to exit to userspace if we already have an interrupt queued.
4992 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
4994 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
4995 vcpu->run->request_interrupt_window &&
4996 kvm_arch_interrupt_allowed(vcpu));
4999 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5001 struct kvm_run *kvm_run = vcpu->run;
5003 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5004 kvm_run->cr8 = kvm_get_cr8(vcpu);
5005 kvm_run->apic_base = kvm_get_apic_base(vcpu);
5006 if (irqchip_in_kernel(vcpu->kvm))
5007 kvm_run->ready_for_interrupt_injection = 1;
5009 kvm_run->ready_for_interrupt_injection =
5010 kvm_arch_interrupt_allowed(vcpu) &&
5011 !kvm_cpu_has_interrupt(vcpu) &&
5012 !kvm_event_needs_reinjection(vcpu);
5015 static void vapic_enter(struct kvm_vcpu *vcpu)
5017 struct kvm_lapic *apic = vcpu->arch.apic;
5020 if (!apic || !apic->vapic_addr)
5023 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5025 vcpu->arch.apic->vapic_page = page;
5028 static void vapic_exit(struct kvm_vcpu *vcpu)
5030 struct kvm_lapic *apic = vcpu->arch.apic;
5033 if (!apic || !apic->vapic_addr)
5036 idx = srcu_read_lock(&vcpu->kvm->srcu);
5037 kvm_release_page_dirty(apic->vapic_page);
5038 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5039 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5042 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5046 if (!kvm_x86_ops->update_cr8_intercept)
5049 if (!vcpu->arch.apic)
5052 if (!vcpu->arch.apic->vapic_addr)
5053 max_irr = kvm_lapic_find_highest_irr(vcpu);
5060 tpr = kvm_lapic_get_cr8(vcpu);
5062 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5065 static void inject_pending_event(struct kvm_vcpu *vcpu)
5067 /* try to reinject previous events if any */
5068 if (vcpu->arch.exception.pending) {
5069 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5070 vcpu->arch.exception.has_error_code,
5071 vcpu->arch.exception.error_code);
5072 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5073 vcpu->arch.exception.has_error_code,
5074 vcpu->arch.exception.error_code,
5075 vcpu->arch.exception.reinject);
5079 if (vcpu->arch.nmi_injected) {
5080 kvm_x86_ops->set_nmi(vcpu);
5084 if (vcpu->arch.interrupt.pending) {
5085 kvm_x86_ops->set_irq(vcpu);
5089 /* try to inject new event if pending */
5090 if (vcpu->arch.nmi_pending) {
5091 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5092 vcpu->arch.nmi_pending = false;
5093 vcpu->arch.nmi_injected = true;
5094 kvm_x86_ops->set_nmi(vcpu);
5096 } else if (kvm_cpu_has_interrupt(vcpu)) {
5097 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5098 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5100 kvm_x86_ops->set_irq(vcpu);
5105 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5107 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5108 !vcpu->guest_xcr0_loaded) {
5109 /* kvm_set_xcr() also depends on this */
5110 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5111 vcpu->guest_xcr0_loaded = 1;
5115 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5117 if (vcpu->guest_xcr0_loaded) {
5118 if (vcpu->arch.xcr0 != host_xcr0)
5119 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5120 vcpu->guest_xcr0_loaded = 0;
5124 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5127 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5128 vcpu->run->request_interrupt_window;
5130 if (vcpu->requests) {
5131 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5132 kvm_mmu_unload(vcpu);
5133 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5134 __kvm_migrate_timers(vcpu);
5135 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5136 r = kvm_guest_time_update(vcpu);
5140 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5141 kvm_mmu_sync_roots(vcpu);
5142 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5143 kvm_x86_ops->tlb_flush(vcpu);
5144 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5145 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5149 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5150 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5154 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5155 vcpu->fpu_active = 0;
5156 kvm_x86_ops->fpu_deactivate(vcpu);
5158 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5159 /* Page is swapped out. Do synthetic halt */
5160 vcpu->arch.apf.halted = true;
5166 r = kvm_mmu_reload(vcpu);
5170 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5171 inject_pending_event(vcpu);
5173 /* enable NMI/IRQ window open exits if needed */
5174 if (vcpu->arch.nmi_pending)
5175 kvm_x86_ops->enable_nmi_window(vcpu);
5176 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5177 kvm_x86_ops->enable_irq_window(vcpu);
5179 if (kvm_lapic_enabled(vcpu)) {
5180 update_cr8_intercept(vcpu);
5181 kvm_lapic_sync_to_vapic(vcpu);
5187 kvm_x86_ops->prepare_guest_switch(vcpu);
5188 if (vcpu->fpu_active)
5189 kvm_load_guest_fpu(vcpu);
5190 kvm_load_guest_xcr0(vcpu);
5192 atomic_set(&vcpu->guest_mode, 1);
5195 local_irq_disable();
5197 if (!atomic_read(&vcpu->guest_mode) || vcpu->requests
5198 || need_resched() || signal_pending(current)) {
5199 atomic_set(&vcpu->guest_mode, 0);
5203 kvm_x86_ops->cancel_injection(vcpu);
5208 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5212 if (unlikely(vcpu->arch.switch_db_regs)) {
5214 set_debugreg(vcpu->arch.eff_db[0], 0);
5215 set_debugreg(vcpu->arch.eff_db[1], 1);
5216 set_debugreg(vcpu->arch.eff_db[2], 2);
5217 set_debugreg(vcpu->arch.eff_db[3], 3);
5220 trace_kvm_entry(vcpu->vcpu_id);
5221 kvm_x86_ops->run(vcpu);
5224 * If the guest has used debug registers, at least dr7
5225 * will be disabled while returning to the host.
5226 * If we don't have active breakpoints in the host, we don't
5227 * care about the messed up debug address registers. But if
5228 * we have some of them active, restore the old state.
5230 if (hw_breakpoint_active())
5231 hw_breakpoint_restore();
5233 kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
5235 atomic_set(&vcpu->guest_mode, 0);
5242 * We must have an instruction between local_irq_enable() and
5243 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5244 * the interrupt shadow. The stat.exits increment will do nicely.
5245 * But we need to prevent reordering, hence this barrier():
5253 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5256 * Profile KVM exit RIPs:
5258 if (unlikely(prof_on == KVM_PROFILING)) {
5259 unsigned long rip = kvm_rip_read(vcpu);
5260 profile_hit(KVM_PROFILING, (void *)rip);
5264 kvm_lapic_sync_from_vapic(vcpu);
5266 r = kvm_x86_ops->handle_exit(vcpu);
5272 static int __vcpu_run(struct kvm_vcpu *vcpu)
5275 struct kvm *kvm = vcpu->kvm;
5277 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5278 pr_debug("vcpu %d received sipi with vector # %x\n",
5279 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5280 kvm_lapic_reset(vcpu);
5281 r = kvm_arch_vcpu_reset(vcpu);
5284 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5287 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5292 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5293 !vcpu->arch.apf.halted)
5294 r = vcpu_enter_guest(vcpu);
5296 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5297 kvm_vcpu_block(vcpu);
5298 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5299 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5301 switch(vcpu->arch.mp_state) {
5302 case KVM_MP_STATE_HALTED:
5303 vcpu->arch.mp_state =
5304 KVM_MP_STATE_RUNNABLE;
5305 case KVM_MP_STATE_RUNNABLE:
5306 vcpu->arch.apf.halted = false;
5308 case KVM_MP_STATE_SIPI_RECEIVED:
5319 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5320 if (kvm_cpu_has_pending_timer(vcpu))
5321 kvm_inject_pending_timer_irqs(vcpu);
5323 if (dm_request_for_irq_injection(vcpu)) {
5325 vcpu->run->exit_reason = KVM_EXIT_INTR;
5326 ++vcpu->stat.request_irq_exits;
5329 kvm_check_async_pf_completion(vcpu);
5331 if (signal_pending(current)) {
5333 vcpu->run->exit_reason = KVM_EXIT_INTR;
5334 ++vcpu->stat.signal_exits;
5336 if (need_resched()) {
5337 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5339 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5343 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5350 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5355 if (vcpu->sigset_active)
5356 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5358 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5359 kvm_vcpu_block(vcpu);
5360 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5365 /* re-sync apic's tpr */
5366 if (!irqchip_in_kernel(vcpu->kvm))
5367 kvm_set_cr8(vcpu, kvm_run->cr8);
5369 if (vcpu->arch.pio.count || vcpu->mmio_needed) {
5370 if (vcpu->mmio_needed) {
5371 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
5372 vcpu->mmio_read_completed = 1;
5373 vcpu->mmio_needed = 0;
5375 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5376 r = emulate_instruction(vcpu, 0, 0, EMULTYPE_NO_DECODE);
5377 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5378 if (r != EMULATE_DONE) {
5383 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5384 kvm_register_write(vcpu, VCPU_REGS_RAX,
5385 kvm_run->hypercall.ret);
5387 r = __vcpu_run(vcpu);
5390 post_kvm_run_save(vcpu);
5391 if (vcpu->sigset_active)
5392 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5397 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5399 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5400 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5401 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5402 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5403 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5404 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5405 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5406 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5407 #ifdef CONFIG_X86_64
5408 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5409 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5410 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5411 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5412 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5413 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5414 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5415 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5418 regs->rip = kvm_rip_read(vcpu);
5419 regs->rflags = kvm_get_rflags(vcpu);
5424 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5426 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5427 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5428 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5429 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5430 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5431 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5432 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5433 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5434 #ifdef CONFIG_X86_64
5435 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5436 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5437 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5438 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5439 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5440 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5441 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5442 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5445 kvm_rip_write(vcpu, regs->rip);
5446 kvm_set_rflags(vcpu, regs->rflags);
5448 vcpu->arch.exception.pending = false;
5450 kvm_make_request(KVM_REQ_EVENT, vcpu);
5455 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5457 struct kvm_segment cs;
5459 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
5463 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5465 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5466 struct kvm_sregs *sregs)
5470 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5471 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5472 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5473 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5474 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5475 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5477 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5478 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5480 kvm_x86_ops->get_idt(vcpu, &dt);
5481 sregs->idt.limit = dt.size;
5482 sregs->idt.base = dt.address;
5483 kvm_x86_ops->get_gdt(vcpu, &dt);
5484 sregs->gdt.limit = dt.size;
5485 sregs->gdt.base = dt.address;
5487 sregs->cr0 = kvm_read_cr0(vcpu);
5488 sregs->cr2 = vcpu->arch.cr2;
5489 sregs->cr3 = vcpu->arch.cr3;
5490 sregs->cr4 = kvm_read_cr4(vcpu);
5491 sregs->cr8 = kvm_get_cr8(vcpu);
5492 sregs->efer = vcpu->arch.efer;
5493 sregs->apic_base = kvm_get_apic_base(vcpu);
5495 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
5497 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
5498 set_bit(vcpu->arch.interrupt.nr,
5499 (unsigned long *)sregs->interrupt_bitmap);
5504 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
5505 struct kvm_mp_state *mp_state)
5507 mp_state->mp_state = vcpu->arch.mp_state;
5511 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
5512 struct kvm_mp_state *mp_state)
5514 vcpu->arch.mp_state = mp_state->mp_state;
5515 kvm_make_request(KVM_REQ_EVENT, vcpu);
5519 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
5520 bool has_error_code, u32 error_code)
5522 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5525 init_emulate_ctxt(vcpu);
5527 ret = emulator_task_switch(&vcpu->arch.emulate_ctxt,
5528 tss_selector, reason, has_error_code,
5532 return EMULATE_FAIL;
5534 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5535 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
5536 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
5537 kvm_make_request(KVM_REQ_EVENT, vcpu);
5538 return EMULATE_DONE;
5540 EXPORT_SYMBOL_GPL(kvm_task_switch);
5542 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5543 struct kvm_sregs *sregs)
5545 int mmu_reset_needed = 0;
5546 int pending_vec, max_bits;
5549 dt.size = sregs->idt.limit;
5550 dt.address = sregs->idt.base;
5551 kvm_x86_ops->set_idt(vcpu, &dt);
5552 dt.size = sregs->gdt.limit;
5553 dt.address = sregs->gdt.base;
5554 kvm_x86_ops->set_gdt(vcpu, &dt);
5556 vcpu->arch.cr2 = sregs->cr2;
5557 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
5558 vcpu->arch.cr3 = sregs->cr3;
5560 kvm_set_cr8(vcpu, sregs->cr8);
5562 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5563 kvm_x86_ops->set_efer(vcpu, sregs->efer);
5564 kvm_set_apic_base(vcpu, sregs->apic_base);
5566 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5567 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5568 vcpu->arch.cr0 = sregs->cr0;
5570 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5571 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5572 if (sregs->cr4 & X86_CR4_OSXSAVE)
5574 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5575 load_pdptrs(vcpu, vcpu->arch.walk_mmu, vcpu->arch.cr3);
5576 mmu_reset_needed = 1;
5579 if (mmu_reset_needed)
5580 kvm_mmu_reset_context(vcpu);
5582 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5583 pending_vec = find_first_bit(
5584 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5585 if (pending_vec < max_bits) {
5586 kvm_queue_interrupt(vcpu, pending_vec, false);
5587 pr_debug("Set back pending irq %d\n", pending_vec);
5588 if (irqchip_in_kernel(vcpu->kvm))
5589 kvm_pic_clear_isr_ack(vcpu->kvm);
5592 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5593 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5594 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5595 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5596 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5597 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5599 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5600 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5602 update_cr8_intercept(vcpu);
5604 /* Older userspace won't unhalt the vcpu on reset. */
5605 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5606 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5608 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5610 kvm_make_request(KVM_REQ_EVENT, vcpu);
5615 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5616 struct kvm_guest_debug *dbg)
5618 unsigned long rflags;
5621 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5623 if (vcpu->arch.exception.pending)
5625 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5626 kvm_queue_exception(vcpu, DB_VECTOR);
5628 kvm_queue_exception(vcpu, BP_VECTOR);
5632 * Read rflags as long as potentially injected trace flags are still
5635 rflags = kvm_get_rflags(vcpu);
5637 vcpu->guest_debug = dbg->control;
5638 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5639 vcpu->guest_debug = 0;
5641 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5642 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5643 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5644 vcpu->arch.switch_db_regs =
5645 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5647 for (i = 0; i < KVM_NR_DB_REGS; i++)
5648 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5649 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5652 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5653 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5654 get_segment_base(vcpu, VCPU_SREG_CS);
5657 * Trigger an rflags update that will inject or remove the trace
5660 kvm_set_rflags(vcpu, rflags);
5662 kvm_x86_ops->set_guest_debug(vcpu, dbg);
5672 * Translate a guest virtual address to a guest physical address.
5674 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5675 struct kvm_translation *tr)
5677 unsigned long vaddr = tr->linear_address;
5681 idx = srcu_read_lock(&vcpu->kvm->srcu);
5682 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5683 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5684 tr->physical_address = gpa;
5685 tr->valid = gpa != UNMAPPED_GVA;
5692 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5694 struct i387_fxsave_struct *fxsave =
5695 &vcpu->arch.guest_fpu.state->fxsave;
5697 memcpy(fpu->fpr, fxsave->st_space, 128);
5698 fpu->fcw = fxsave->cwd;
5699 fpu->fsw = fxsave->swd;
5700 fpu->ftwx = fxsave->twd;
5701 fpu->last_opcode = fxsave->fop;
5702 fpu->last_ip = fxsave->rip;
5703 fpu->last_dp = fxsave->rdp;
5704 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5709 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5711 struct i387_fxsave_struct *fxsave =
5712 &vcpu->arch.guest_fpu.state->fxsave;
5714 memcpy(fxsave->st_space, fpu->fpr, 128);
5715 fxsave->cwd = fpu->fcw;
5716 fxsave->swd = fpu->fsw;
5717 fxsave->twd = fpu->ftwx;
5718 fxsave->fop = fpu->last_opcode;
5719 fxsave->rip = fpu->last_ip;
5720 fxsave->rdp = fpu->last_dp;
5721 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5726 int fx_init(struct kvm_vcpu *vcpu)
5730 err = fpu_alloc(&vcpu->arch.guest_fpu);
5734 fpu_finit(&vcpu->arch.guest_fpu);
5737 * Ensure guest xcr0 is valid for loading
5739 vcpu->arch.xcr0 = XSTATE_FP;
5741 vcpu->arch.cr0 |= X86_CR0_ET;
5745 EXPORT_SYMBOL_GPL(fx_init);
5747 static void fx_free(struct kvm_vcpu *vcpu)
5749 fpu_free(&vcpu->arch.guest_fpu);
5752 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5754 if (vcpu->guest_fpu_loaded)
5758 * Restore all possible states in the guest,
5759 * and assume host would use all available bits.
5760 * Guest xcr0 would be loaded later.
5762 kvm_put_guest_xcr0(vcpu);
5763 vcpu->guest_fpu_loaded = 1;
5764 unlazy_fpu(current);
5765 fpu_restore_checking(&vcpu->arch.guest_fpu);
5769 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5771 kvm_put_guest_xcr0(vcpu);
5773 if (!vcpu->guest_fpu_loaded)
5776 vcpu->guest_fpu_loaded = 0;
5777 fpu_save_init(&vcpu->arch.guest_fpu);
5778 ++vcpu->stat.fpu_reload;
5779 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
5783 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5785 if (vcpu->arch.time_page) {
5786 kvm_release_page_dirty(vcpu->arch.time_page);
5787 vcpu->arch.time_page = NULL;
5790 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
5792 kvm_x86_ops->vcpu_free(vcpu);
5795 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5798 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
5799 printk_once(KERN_WARNING
5800 "kvm: SMP vm created on host with unstable TSC; "
5801 "guest TSC will not be reliable\n");
5802 return kvm_x86_ops->vcpu_create(kvm, id);
5805 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5809 vcpu->arch.mtrr_state.have_fixed = 1;
5811 r = kvm_arch_vcpu_reset(vcpu);
5813 r = kvm_mmu_setup(vcpu);
5820 kvm_x86_ops->vcpu_free(vcpu);
5824 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5826 vcpu->arch.apf.msr_val = 0;
5829 kvm_mmu_unload(vcpu);
5833 kvm_x86_ops->vcpu_free(vcpu);
5836 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5838 vcpu->arch.nmi_pending = false;
5839 vcpu->arch.nmi_injected = false;
5841 vcpu->arch.switch_db_regs = 0;
5842 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5843 vcpu->arch.dr6 = DR6_FIXED_1;
5844 vcpu->arch.dr7 = DR7_FIXED_1;
5846 kvm_make_request(KVM_REQ_EVENT, vcpu);
5847 vcpu->arch.apf.msr_val = 0;
5849 kvm_clear_async_pf_completion_queue(vcpu);
5850 kvm_async_pf_hash_reset(vcpu);
5851 vcpu->arch.apf.halted = false;
5853 return kvm_x86_ops->vcpu_reset(vcpu);
5856 int kvm_arch_hardware_enable(void *garbage)
5859 struct kvm_vcpu *vcpu;
5862 kvm_shared_msr_cpu_online();
5863 list_for_each_entry(kvm, &vm_list, vm_list)
5864 kvm_for_each_vcpu(i, vcpu, kvm)
5865 if (vcpu->cpu == smp_processor_id())
5866 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5867 return kvm_x86_ops->hardware_enable(garbage);
5870 void kvm_arch_hardware_disable(void *garbage)
5872 kvm_x86_ops->hardware_disable(garbage);
5873 drop_user_return_notifiers(garbage);
5876 int kvm_arch_hardware_setup(void)
5878 return kvm_x86_ops->hardware_setup();
5881 void kvm_arch_hardware_unsetup(void)
5883 kvm_x86_ops->hardware_unsetup();
5886 void kvm_arch_check_processor_compat(void *rtn)
5888 kvm_x86_ops->check_processor_compatibility(rtn);
5891 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5897 BUG_ON(vcpu->kvm == NULL);
5900 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
5901 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
5902 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5903 vcpu->arch.mmu.translate_gpa = translate_gpa;
5904 vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
5905 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5906 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5908 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5910 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5915 vcpu->arch.pio_data = page_address(page);
5917 if (!kvm->arch.virtual_tsc_khz)
5918 kvm_arch_set_tsc_khz(kvm, max_tsc_khz);
5920 r = kvm_mmu_create(vcpu);
5922 goto fail_free_pio_data;
5924 if (irqchip_in_kernel(kvm)) {
5925 r = kvm_create_lapic(vcpu);
5927 goto fail_mmu_destroy;
5930 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5932 if (!vcpu->arch.mce_banks) {
5934 goto fail_free_lapic;
5936 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5938 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
5939 goto fail_free_mce_banks;
5941 kvm_async_pf_hash_reset(vcpu);
5944 fail_free_mce_banks:
5945 kfree(vcpu->arch.mce_banks);
5947 kvm_free_lapic(vcpu);
5949 kvm_mmu_destroy(vcpu);
5951 free_page((unsigned long)vcpu->arch.pio_data);
5956 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5960 kfree(vcpu->arch.mce_banks);
5961 kvm_free_lapic(vcpu);
5962 idx = srcu_read_lock(&vcpu->kvm->srcu);
5963 kvm_mmu_destroy(vcpu);
5964 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5965 free_page((unsigned long)vcpu->arch.pio_data);
5968 struct kvm *kvm_arch_create_vm(void)
5970 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5973 return ERR_PTR(-ENOMEM);
5975 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5976 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5978 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5979 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5981 spin_lock_init(&kvm->arch.tsc_write_lock);
5986 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5989 kvm_mmu_unload(vcpu);
5993 static void kvm_free_vcpus(struct kvm *kvm)
5996 struct kvm_vcpu *vcpu;
5999 * Unpin any mmu pages first.
6001 kvm_for_each_vcpu(i, vcpu, kvm) {
6002 kvm_clear_async_pf_completion_queue(vcpu);
6003 kvm_unload_vcpu_mmu(vcpu);
6005 kvm_for_each_vcpu(i, vcpu, kvm)
6006 kvm_arch_vcpu_free(vcpu);
6008 mutex_lock(&kvm->lock);
6009 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
6010 kvm->vcpus[i] = NULL;
6012 atomic_set(&kvm->online_vcpus, 0);
6013 mutex_unlock(&kvm->lock);
6016 void kvm_arch_sync_events(struct kvm *kvm)
6018 kvm_free_all_assigned_devices(kvm);
6022 void kvm_arch_destroy_vm(struct kvm *kvm)
6024 kvm_iommu_unmap_guest(kvm);
6025 kfree(kvm->arch.vpic);
6026 kfree(kvm->arch.vioapic);
6027 kvm_free_vcpus(kvm);
6028 kvm_free_physmem(kvm);
6029 if (kvm->arch.apic_access_page)
6030 put_page(kvm->arch.apic_access_page);
6031 if (kvm->arch.ept_identity_pagetable)
6032 put_page(kvm->arch.ept_identity_pagetable);
6033 cleanup_srcu_struct(&kvm->srcu);
6037 int kvm_arch_prepare_memory_region(struct kvm *kvm,
6038 struct kvm_memory_slot *memslot,
6039 struct kvm_memory_slot old,
6040 struct kvm_userspace_memory_region *mem,
6043 int npages = memslot->npages;
6044 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
6046 /* Prevent internal slot pages from being moved by fork()/COW. */
6047 if (memslot->id >= KVM_MEMORY_SLOTS)
6048 map_flags = MAP_SHARED | MAP_ANONYMOUS;
6050 /*To keep backward compatibility with older userspace,
6051 *x86 needs to hanlde !user_alloc case.
6054 if (npages && !old.rmap) {
6055 unsigned long userspace_addr;
6057 down_write(¤t->mm->mmap_sem);
6058 userspace_addr = do_mmap(NULL, 0,
6060 PROT_READ | PROT_WRITE,
6063 up_write(¤t->mm->mmap_sem);
6065 if (IS_ERR((void *)userspace_addr))
6066 return PTR_ERR((void *)userspace_addr);
6068 memslot->userspace_addr = userspace_addr;
6076 void kvm_arch_commit_memory_region(struct kvm *kvm,
6077 struct kvm_userspace_memory_region *mem,
6078 struct kvm_memory_slot old,
6082 int npages = mem->memory_size >> PAGE_SHIFT;
6084 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6087 down_write(¤t->mm->mmap_sem);
6088 ret = do_munmap(current->mm, old.userspace_addr,
6089 old.npages * PAGE_SIZE);
6090 up_write(¤t->mm->mmap_sem);
6093 "kvm_vm_ioctl_set_memory_region: "
6094 "failed to munmap memory\n");
6097 spin_lock(&kvm->mmu_lock);
6098 if (!kvm->arch.n_requested_mmu_pages) {
6099 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6100 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6103 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6104 spin_unlock(&kvm->mmu_lock);
6107 void kvm_arch_flush_shadow(struct kvm *kvm)
6109 kvm_mmu_zap_all(kvm);
6110 kvm_reload_remote_mmus(kvm);
6113 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6115 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6116 !vcpu->arch.apf.halted)
6117 || !list_empty_careful(&vcpu->async_pf.done)
6118 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6119 || vcpu->arch.nmi_pending ||
6120 (kvm_arch_interrupt_allowed(vcpu) &&
6121 kvm_cpu_has_interrupt(vcpu));
6124 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6127 int cpu = vcpu->cpu;
6129 if (waitqueue_active(&vcpu->wq)) {
6130 wake_up_interruptible(&vcpu->wq);
6131 ++vcpu->stat.halt_wakeup;
6135 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6136 if (atomic_xchg(&vcpu->guest_mode, 0))
6137 smp_send_reschedule(cpu);
6141 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6143 return kvm_x86_ops->interrupt_allowed(vcpu);
6146 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6148 unsigned long current_rip = kvm_rip_read(vcpu) +
6149 get_segment_base(vcpu, VCPU_SREG_CS);
6151 return current_rip == linear_rip;
6153 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6155 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6157 unsigned long rflags;
6159 rflags = kvm_x86_ops->get_rflags(vcpu);
6160 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6161 rflags &= ~X86_EFLAGS_TF;
6164 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6166 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6168 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6169 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6170 rflags |= X86_EFLAGS_TF;
6171 kvm_x86_ops->set_rflags(vcpu, rflags);
6172 kvm_make_request(KVM_REQ_EVENT, vcpu);
6174 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6176 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
6180 if (!vcpu->arch.mmu.direct_map || is_error_page(work->page))
6183 r = kvm_mmu_reload(vcpu);
6187 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
6190 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
6192 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
6195 static inline u32 kvm_async_pf_next_probe(u32 key)
6197 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
6200 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6202 u32 key = kvm_async_pf_hash_fn(gfn);
6204 while (vcpu->arch.apf.gfns[key] != ~0)
6205 key = kvm_async_pf_next_probe(key);
6207 vcpu->arch.apf.gfns[key] = gfn;
6210 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
6213 u32 key = kvm_async_pf_hash_fn(gfn);
6215 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
6216 (vcpu->arch.apf.gfns[key] != gfn ||
6217 vcpu->arch.apf.gfns[key] == ~0); i++)
6218 key = kvm_async_pf_next_probe(key);
6223 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6225 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
6228 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
6232 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
6234 vcpu->arch.apf.gfns[i] = ~0;
6236 j = kvm_async_pf_next_probe(j);
6237 if (vcpu->arch.apf.gfns[j] == ~0)
6239 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
6241 * k lies cyclically in ]i,j]
6243 * |....j i.k.| or |.k..j i...|
6245 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
6246 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
6251 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
6254 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
6258 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
6259 struct kvm_async_pf *work)
6261 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
6262 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
6264 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
6265 kvm_x86_ops->get_cpl(vcpu) == 0)
6266 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
6267 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
6268 vcpu->arch.fault.error_code = 0;
6269 vcpu->arch.fault.address = work->arch.token;
6270 kvm_inject_page_fault(vcpu);
6274 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
6275 struct kvm_async_pf *work)
6277 trace_kvm_async_pf_ready(work->arch.token, work->gva);
6278 if (is_error_page(work->page))
6279 work->arch.token = ~0; /* broadcast wakeup */
6281 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
6283 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
6284 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
6285 vcpu->arch.fault.error_code = 0;
6286 vcpu->arch.fault.address = work->arch.token;
6287 kvm_inject_page_fault(vcpu);
6291 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
6293 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
6296 return !kvm_event_needs_reinjection(vcpu) &&
6297 kvm_x86_ops->interrupt_allowed(vcpu);
6300 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6301 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6302 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6303 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6304 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6305 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6306 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6307 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6308 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6309 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6310 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6311 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);