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 "assigned-dev.h"
32 #include <linux/clocksource.h>
33 #include <linux/interrupt.h>
34 #include <linux/kvm.h>
36 #include <linux/vmalloc.h>
37 #include <linux/module.h>
38 #include <linux/mman.h>
39 #include <linux/highmem.h>
40 #include <linux/iommu.h>
41 #include <linux/intel-iommu.h>
42 #include <linux/cpufreq.h>
43 #include <linux/user-return-notifier.h>
44 #include <linux/srcu.h>
45 #include <linux/slab.h>
46 #include <linux/perf_event.h>
47 #include <linux/uaccess.h>
48 #include <linux/hash.h>
49 #include <linux/pci.h>
50 #include <linux/timekeeper_internal.h>
51 #include <linux/pvclock_gtod.h>
52 #include <trace/events/kvm.h>
54 #define CREATE_TRACE_POINTS
57 #include <asm/debugreg.h>
63 #include <asm/fpu-internal.h> /* Ugh! */
65 #include <asm/pvclock.h>
66 #include <asm/div64.h>
68 #define MAX_IO_MSRS 256
69 #define KVM_MAX_MCE_BANKS 32
70 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
72 #define emul_to_vcpu(ctxt) \
73 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
76 * - enable syscall per default because its emulated by KVM
77 * - enable LME and LMA per default on 64 bit KVM
81 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
83 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
86 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
87 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
89 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
90 static void process_nmi(struct kvm_vcpu *vcpu);
91 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
93 struct kvm_x86_ops *kvm_x86_ops;
94 EXPORT_SYMBOL_GPL(kvm_x86_ops);
96 static bool ignore_msrs = 0;
97 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
99 unsigned int min_timer_period_us = 500;
100 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
102 bool kvm_has_tsc_control;
103 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
104 u32 kvm_max_guest_tsc_khz;
105 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
107 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
108 static u32 tsc_tolerance_ppm = 250;
109 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
111 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
112 unsigned int lapic_timer_advance_ns = 0;
113 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
115 static bool backwards_tsc_observed = false;
117 #define KVM_NR_SHARED_MSRS 16
119 struct kvm_shared_msrs_global {
121 u32 msrs[KVM_NR_SHARED_MSRS];
124 struct kvm_shared_msrs {
125 struct user_return_notifier urn;
127 struct kvm_shared_msr_values {
130 } values[KVM_NR_SHARED_MSRS];
133 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
134 static struct kvm_shared_msrs __percpu *shared_msrs;
136 struct kvm_stats_debugfs_item debugfs_entries[] = {
137 { "pf_fixed", VCPU_STAT(pf_fixed) },
138 { "pf_guest", VCPU_STAT(pf_guest) },
139 { "tlb_flush", VCPU_STAT(tlb_flush) },
140 { "invlpg", VCPU_STAT(invlpg) },
141 { "exits", VCPU_STAT(exits) },
142 { "io_exits", VCPU_STAT(io_exits) },
143 { "mmio_exits", VCPU_STAT(mmio_exits) },
144 { "signal_exits", VCPU_STAT(signal_exits) },
145 { "irq_window", VCPU_STAT(irq_window_exits) },
146 { "nmi_window", VCPU_STAT(nmi_window_exits) },
147 { "halt_exits", VCPU_STAT(halt_exits) },
148 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
149 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
150 { "hypercalls", VCPU_STAT(hypercalls) },
151 { "request_irq", VCPU_STAT(request_irq_exits) },
152 { "irq_exits", VCPU_STAT(irq_exits) },
153 { "host_state_reload", VCPU_STAT(host_state_reload) },
154 { "efer_reload", VCPU_STAT(efer_reload) },
155 { "fpu_reload", VCPU_STAT(fpu_reload) },
156 { "insn_emulation", VCPU_STAT(insn_emulation) },
157 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
158 { "irq_injections", VCPU_STAT(irq_injections) },
159 { "nmi_injections", VCPU_STAT(nmi_injections) },
160 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
161 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
162 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
163 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
164 { "mmu_flooded", VM_STAT(mmu_flooded) },
165 { "mmu_recycled", VM_STAT(mmu_recycled) },
166 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
167 { "mmu_unsync", VM_STAT(mmu_unsync) },
168 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
169 { "largepages", VM_STAT(lpages) },
173 u64 __read_mostly host_xcr0;
175 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
177 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
180 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
181 vcpu->arch.apf.gfns[i] = ~0;
184 static void kvm_on_user_return(struct user_return_notifier *urn)
187 struct kvm_shared_msrs *locals
188 = container_of(urn, struct kvm_shared_msrs, urn);
189 struct kvm_shared_msr_values *values;
191 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
192 values = &locals->values[slot];
193 if (values->host != values->curr) {
194 wrmsrl(shared_msrs_global.msrs[slot], values->host);
195 values->curr = values->host;
198 locals->registered = false;
199 user_return_notifier_unregister(urn);
202 static void shared_msr_update(unsigned slot, u32 msr)
205 unsigned int cpu = smp_processor_id();
206 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
208 /* only read, and nobody should modify it at this time,
209 * so don't need lock */
210 if (slot >= shared_msrs_global.nr) {
211 printk(KERN_ERR "kvm: invalid MSR slot!");
214 rdmsrl_safe(msr, &value);
215 smsr->values[slot].host = value;
216 smsr->values[slot].curr = value;
219 void kvm_define_shared_msr(unsigned slot, u32 msr)
221 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
222 if (slot >= shared_msrs_global.nr)
223 shared_msrs_global.nr = slot + 1;
224 shared_msrs_global.msrs[slot] = msr;
225 /* we need ensured the shared_msr_global have been updated */
228 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
230 static void kvm_shared_msr_cpu_online(void)
234 for (i = 0; i < shared_msrs_global.nr; ++i)
235 shared_msr_update(i, shared_msrs_global.msrs[i]);
238 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
240 unsigned int cpu = smp_processor_id();
241 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
244 if (((value ^ smsr->values[slot].curr) & mask) == 0)
246 smsr->values[slot].curr = value;
247 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
251 if (!smsr->registered) {
252 smsr->urn.on_user_return = kvm_on_user_return;
253 user_return_notifier_register(&smsr->urn);
254 smsr->registered = true;
258 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
260 static void drop_user_return_notifiers(void)
262 unsigned int cpu = smp_processor_id();
263 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
265 if (smsr->registered)
266 kvm_on_user_return(&smsr->urn);
269 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
271 return vcpu->arch.apic_base;
273 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
275 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
277 u64 old_state = vcpu->arch.apic_base &
278 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
279 u64 new_state = msr_info->data &
280 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
281 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
282 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
284 if (!msr_info->host_initiated &&
285 ((msr_info->data & reserved_bits) != 0 ||
286 new_state == X2APIC_ENABLE ||
287 (new_state == MSR_IA32_APICBASE_ENABLE &&
288 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
289 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
293 kvm_lapic_set_base(vcpu, msr_info->data);
296 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
298 asmlinkage __visible void kvm_spurious_fault(void)
300 /* Fault while not rebooting. We want the trace. */
303 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
305 #define EXCPT_BENIGN 0
306 #define EXCPT_CONTRIBUTORY 1
309 static int exception_class(int vector)
319 return EXCPT_CONTRIBUTORY;
326 #define EXCPT_FAULT 0
328 #define EXCPT_ABORT 2
329 #define EXCPT_INTERRUPT 3
331 static int exception_type(int vector)
335 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
336 return EXCPT_INTERRUPT;
340 /* #DB is trap, as instruction watchpoints are handled elsewhere */
341 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
344 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
347 /* Reserved exceptions will result in fault */
351 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
352 unsigned nr, bool has_error, u32 error_code,
358 kvm_make_request(KVM_REQ_EVENT, vcpu);
360 if (!vcpu->arch.exception.pending) {
362 if (has_error && !is_protmode(vcpu))
364 vcpu->arch.exception.pending = true;
365 vcpu->arch.exception.has_error_code = has_error;
366 vcpu->arch.exception.nr = nr;
367 vcpu->arch.exception.error_code = error_code;
368 vcpu->arch.exception.reinject = reinject;
372 /* to check exception */
373 prev_nr = vcpu->arch.exception.nr;
374 if (prev_nr == DF_VECTOR) {
375 /* triple fault -> shutdown */
376 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
379 class1 = exception_class(prev_nr);
380 class2 = exception_class(nr);
381 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
382 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
383 /* generate double fault per SDM Table 5-5 */
384 vcpu->arch.exception.pending = true;
385 vcpu->arch.exception.has_error_code = true;
386 vcpu->arch.exception.nr = DF_VECTOR;
387 vcpu->arch.exception.error_code = 0;
389 /* replace previous exception with a new one in a hope
390 that instruction re-execution will regenerate lost
395 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
397 kvm_multiple_exception(vcpu, nr, false, 0, false);
399 EXPORT_SYMBOL_GPL(kvm_queue_exception);
401 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
403 kvm_multiple_exception(vcpu, nr, false, 0, true);
405 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
407 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
410 kvm_inject_gp(vcpu, 0);
412 kvm_x86_ops->skip_emulated_instruction(vcpu);
414 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
416 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
418 ++vcpu->stat.pf_guest;
419 vcpu->arch.cr2 = fault->address;
420 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
422 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
424 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
426 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
427 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
429 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
431 return fault->nested_page_fault;
434 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
436 atomic_inc(&vcpu->arch.nmi_queued);
437 kvm_make_request(KVM_REQ_NMI, vcpu);
439 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
441 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
443 kvm_multiple_exception(vcpu, nr, true, error_code, false);
445 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
447 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
449 kvm_multiple_exception(vcpu, nr, true, error_code, true);
451 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
454 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
455 * a #GP and return false.
457 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
459 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
461 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
464 EXPORT_SYMBOL_GPL(kvm_require_cpl);
466 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
468 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
471 kvm_queue_exception(vcpu, UD_VECTOR);
474 EXPORT_SYMBOL_GPL(kvm_require_dr);
477 * This function will be used to read from the physical memory of the currently
478 * running guest. The difference to kvm_read_guest_page is that this function
479 * can read from guest physical or from the guest's guest physical memory.
481 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
482 gfn_t ngfn, void *data, int offset, int len,
485 struct x86_exception exception;
489 ngpa = gfn_to_gpa(ngfn);
490 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
491 if (real_gfn == UNMAPPED_GVA)
494 real_gfn = gpa_to_gfn(real_gfn);
496 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
498 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
500 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
501 void *data, int offset, int len, u32 access)
503 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
504 data, offset, len, access);
508 * Load the pae pdptrs. Return true is they are all valid.
510 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
512 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
513 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
516 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
518 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
519 offset * sizeof(u64), sizeof(pdpte),
520 PFERR_USER_MASK|PFERR_WRITE_MASK);
525 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
526 if (is_present_gpte(pdpte[i]) &&
527 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
534 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
535 __set_bit(VCPU_EXREG_PDPTR,
536 (unsigned long *)&vcpu->arch.regs_avail);
537 __set_bit(VCPU_EXREG_PDPTR,
538 (unsigned long *)&vcpu->arch.regs_dirty);
543 EXPORT_SYMBOL_GPL(load_pdptrs);
545 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
547 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
553 if (is_long_mode(vcpu) || !is_pae(vcpu))
556 if (!test_bit(VCPU_EXREG_PDPTR,
557 (unsigned long *)&vcpu->arch.regs_avail))
560 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
561 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
562 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
563 PFERR_USER_MASK | PFERR_WRITE_MASK);
566 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
572 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
574 unsigned long old_cr0 = kvm_read_cr0(vcpu);
575 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
576 X86_CR0_CD | X86_CR0_NW;
581 if (cr0 & 0xffffffff00000000UL)
585 cr0 &= ~CR0_RESERVED_BITS;
587 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
590 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
593 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
595 if ((vcpu->arch.efer & EFER_LME)) {
600 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
605 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
610 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
613 kvm_x86_ops->set_cr0(vcpu, cr0);
615 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
616 kvm_clear_async_pf_completion_queue(vcpu);
617 kvm_async_pf_hash_reset(vcpu);
620 if ((cr0 ^ old_cr0) & update_bits)
621 kvm_mmu_reset_context(vcpu);
624 EXPORT_SYMBOL_GPL(kvm_set_cr0);
626 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
628 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
630 EXPORT_SYMBOL_GPL(kvm_lmsw);
632 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
634 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
635 !vcpu->guest_xcr0_loaded) {
636 /* kvm_set_xcr() also depends on this */
637 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
638 vcpu->guest_xcr0_loaded = 1;
642 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
644 if (vcpu->guest_xcr0_loaded) {
645 if (vcpu->arch.xcr0 != host_xcr0)
646 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
647 vcpu->guest_xcr0_loaded = 0;
651 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
654 u64 old_xcr0 = vcpu->arch.xcr0;
657 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
658 if (index != XCR_XFEATURE_ENABLED_MASK)
660 if (!(xcr0 & XSTATE_FP))
662 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
666 * Do not allow the guest to set bits that we do not support
667 * saving. However, xcr0 bit 0 is always set, even if the
668 * emulated CPU does not support XSAVE (see fx_init).
670 valid_bits = vcpu->arch.guest_supported_xcr0 | XSTATE_FP;
671 if (xcr0 & ~valid_bits)
674 if ((!(xcr0 & XSTATE_BNDREGS)) != (!(xcr0 & XSTATE_BNDCSR)))
677 if (xcr0 & XSTATE_AVX512) {
678 if (!(xcr0 & XSTATE_YMM))
680 if ((xcr0 & XSTATE_AVX512) != XSTATE_AVX512)
683 kvm_put_guest_xcr0(vcpu);
684 vcpu->arch.xcr0 = xcr0;
686 if ((xcr0 ^ old_xcr0) & XSTATE_EXTEND_MASK)
687 kvm_update_cpuid(vcpu);
691 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
693 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
694 __kvm_set_xcr(vcpu, index, xcr)) {
695 kvm_inject_gp(vcpu, 0);
700 EXPORT_SYMBOL_GPL(kvm_set_xcr);
702 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
704 unsigned long old_cr4 = kvm_read_cr4(vcpu);
705 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE |
706 X86_CR4_PAE | X86_CR4_SMEP;
707 if (cr4 & CR4_RESERVED_BITS)
710 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
713 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
716 if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
719 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
722 if (is_long_mode(vcpu)) {
723 if (!(cr4 & X86_CR4_PAE))
725 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
726 && ((cr4 ^ old_cr4) & pdptr_bits)
727 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
731 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
732 if (!guest_cpuid_has_pcid(vcpu))
735 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
736 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
740 if (kvm_x86_ops->set_cr4(vcpu, cr4))
743 if (((cr4 ^ old_cr4) & pdptr_bits) ||
744 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
745 kvm_mmu_reset_context(vcpu);
747 if ((cr4 ^ old_cr4) & X86_CR4_SMAP)
748 update_permission_bitmask(vcpu, vcpu->arch.walk_mmu, false);
750 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
751 kvm_update_cpuid(vcpu);
755 EXPORT_SYMBOL_GPL(kvm_set_cr4);
757 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
760 cr3 &= ~CR3_PCID_INVD;
763 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
764 kvm_mmu_sync_roots(vcpu);
765 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
769 if (is_long_mode(vcpu)) {
770 if (cr3 & CR3_L_MODE_RESERVED_BITS)
772 } else if (is_pae(vcpu) && is_paging(vcpu) &&
773 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
776 vcpu->arch.cr3 = cr3;
777 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
778 kvm_mmu_new_cr3(vcpu);
781 EXPORT_SYMBOL_GPL(kvm_set_cr3);
783 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
785 if (cr8 & CR8_RESERVED_BITS)
787 if (irqchip_in_kernel(vcpu->kvm))
788 kvm_lapic_set_tpr(vcpu, cr8);
790 vcpu->arch.cr8 = cr8;
793 EXPORT_SYMBOL_GPL(kvm_set_cr8);
795 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
797 if (irqchip_in_kernel(vcpu->kvm))
798 return kvm_lapic_get_cr8(vcpu);
800 return vcpu->arch.cr8;
802 EXPORT_SYMBOL_GPL(kvm_get_cr8);
804 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
808 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
809 for (i = 0; i < KVM_NR_DB_REGS; i++)
810 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
811 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
815 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
817 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
818 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
821 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
825 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
826 dr7 = vcpu->arch.guest_debug_dr7;
828 dr7 = vcpu->arch.dr7;
829 kvm_x86_ops->set_dr7(vcpu, dr7);
830 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
831 if (dr7 & DR7_BP_EN_MASK)
832 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
835 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
837 u64 fixed = DR6_FIXED_1;
839 if (!guest_cpuid_has_rtm(vcpu))
844 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
848 vcpu->arch.db[dr] = val;
849 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
850 vcpu->arch.eff_db[dr] = val;
855 if (val & 0xffffffff00000000ULL)
857 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
858 kvm_update_dr6(vcpu);
863 if (val & 0xffffffff00000000ULL)
865 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
866 kvm_update_dr7(vcpu);
873 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
875 if (__kvm_set_dr(vcpu, dr, val)) {
876 kvm_inject_gp(vcpu, 0);
881 EXPORT_SYMBOL_GPL(kvm_set_dr);
883 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
887 *val = vcpu->arch.db[dr];
892 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
893 *val = vcpu->arch.dr6;
895 *val = kvm_x86_ops->get_dr6(vcpu);
900 *val = vcpu->arch.dr7;
905 EXPORT_SYMBOL_GPL(kvm_get_dr);
907 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
909 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
913 err = kvm_pmu_read_pmc(vcpu, ecx, &data);
916 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
917 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
920 EXPORT_SYMBOL_GPL(kvm_rdpmc);
923 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
924 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
926 * This list is modified at module load time to reflect the
927 * capabilities of the host cpu. This capabilities test skips MSRs that are
928 * kvm-specific. Those are put in the beginning of the list.
931 #define KVM_SAVE_MSRS_BEGIN 12
932 static u32 msrs_to_save[] = {
933 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
934 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
935 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
936 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
937 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
939 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
942 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
944 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
945 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS
948 static unsigned num_msrs_to_save;
950 static const u32 emulated_msrs[] = {
952 MSR_IA32_TSCDEADLINE,
953 MSR_IA32_MISC_ENABLE,
958 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
960 if (efer & efer_reserved_bits)
963 if (efer & EFER_FFXSR) {
964 struct kvm_cpuid_entry2 *feat;
966 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
967 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
971 if (efer & EFER_SVME) {
972 struct kvm_cpuid_entry2 *feat;
974 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
975 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
981 EXPORT_SYMBOL_GPL(kvm_valid_efer);
983 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
985 u64 old_efer = vcpu->arch.efer;
987 if (!kvm_valid_efer(vcpu, efer))
991 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
995 efer |= vcpu->arch.efer & EFER_LMA;
997 kvm_x86_ops->set_efer(vcpu, efer);
999 /* Update reserved bits */
1000 if ((efer ^ old_efer) & EFER_NX)
1001 kvm_mmu_reset_context(vcpu);
1006 void kvm_enable_efer_bits(u64 mask)
1008 efer_reserved_bits &= ~mask;
1010 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1013 * Writes msr value into into the appropriate "register".
1014 * Returns 0 on success, non-0 otherwise.
1015 * Assumes vcpu_load() was already called.
1017 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1019 switch (msr->index) {
1022 case MSR_KERNEL_GS_BASE:
1025 if (is_noncanonical_address(msr->data))
1028 case MSR_IA32_SYSENTER_EIP:
1029 case MSR_IA32_SYSENTER_ESP:
1031 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1032 * non-canonical address is written on Intel but not on
1033 * AMD (which ignores the top 32-bits, because it does
1034 * not implement 64-bit SYSENTER).
1036 * 64-bit code should hence be able to write a non-canonical
1037 * value on AMD. Making the address canonical ensures that
1038 * vmentry does not fail on Intel after writing a non-canonical
1039 * value, and that something deterministic happens if the guest
1040 * invokes 64-bit SYSENTER.
1042 msr->data = get_canonical(msr->data);
1044 return kvm_x86_ops->set_msr(vcpu, msr);
1046 EXPORT_SYMBOL_GPL(kvm_set_msr);
1049 * Adapt set_msr() to msr_io()'s calling convention
1051 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1053 struct msr_data msr;
1057 msr.host_initiated = true;
1058 return kvm_set_msr(vcpu, &msr);
1061 #ifdef CONFIG_X86_64
1062 struct pvclock_gtod_data {
1065 struct { /* extract of a clocksource struct */
1077 static struct pvclock_gtod_data pvclock_gtod_data;
1079 static void update_pvclock_gtod(struct timekeeper *tk)
1081 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1084 boot_ns = ktime_to_ns(ktime_add(tk->tkr.base_mono, tk->offs_boot));
1086 write_seqcount_begin(&vdata->seq);
1088 /* copy pvclock gtod data */
1089 vdata->clock.vclock_mode = tk->tkr.clock->archdata.vclock_mode;
1090 vdata->clock.cycle_last = tk->tkr.cycle_last;
1091 vdata->clock.mask = tk->tkr.mask;
1092 vdata->clock.mult = tk->tkr.mult;
1093 vdata->clock.shift = tk->tkr.shift;
1095 vdata->boot_ns = boot_ns;
1096 vdata->nsec_base = tk->tkr.xtime_nsec;
1098 write_seqcount_end(&vdata->seq);
1102 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1105 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1106 * vcpu_enter_guest. This function is only called from
1107 * the physical CPU that is running vcpu.
1109 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1112 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1116 struct pvclock_wall_clock wc;
1117 struct timespec boot;
1122 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1127 ++version; /* first time write, random junk */
1131 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1134 * The guest calculates current wall clock time by adding
1135 * system time (updated by kvm_guest_time_update below) to the
1136 * wall clock specified here. guest system time equals host
1137 * system time for us, thus we must fill in host boot time here.
1141 if (kvm->arch.kvmclock_offset) {
1142 struct timespec ts = ns_to_timespec(kvm->arch.kvmclock_offset);
1143 boot = timespec_sub(boot, ts);
1145 wc.sec = boot.tv_sec;
1146 wc.nsec = boot.tv_nsec;
1147 wc.version = version;
1149 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1152 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1155 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1157 uint32_t quotient, remainder;
1159 /* Don't try to replace with do_div(), this one calculates
1160 * "(dividend << 32) / divisor" */
1162 : "=a" (quotient), "=d" (remainder)
1163 : "0" (0), "1" (dividend), "r" (divisor) );
1167 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
1168 s8 *pshift, u32 *pmultiplier)
1175 tps64 = base_khz * 1000LL;
1176 scaled64 = scaled_khz * 1000LL;
1177 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1182 tps32 = (uint32_t)tps64;
1183 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1184 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1192 *pmultiplier = div_frac(scaled64, tps32);
1194 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
1195 __func__, base_khz, scaled_khz, shift, *pmultiplier);
1198 static inline u64 get_kernel_ns(void)
1200 return ktime_get_boot_ns();
1203 #ifdef CONFIG_X86_64
1204 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1207 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1208 static unsigned long max_tsc_khz;
1210 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1212 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1213 vcpu->arch.virtual_tsc_shift);
1216 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1218 u64 v = (u64)khz * (1000000 + ppm);
1223 static void kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1225 u32 thresh_lo, thresh_hi;
1226 int use_scaling = 0;
1228 /* tsc_khz can be zero if TSC calibration fails */
1229 if (this_tsc_khz == 0)
1232 /* Compute a scale to convert nanoseconds in TSC cycles */
1233 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1234 &vcpu->arch.virtual_tsc_shift,
1235 &vcpu->arch.virtual_tsc_mult);
1236 vcpu->arch.virtual_tsc_khz = this_tsc_khz;
1239 * Compute the variation in TSC rate which is acceptable
1240 * within the range of tolerance and decide if the
1241 * rate being applied is within that bounds of the hardware
1242 * rate. If so, no scaling or compensation need be done.
1244 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1245 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1246 if (this_tsc_khz < thresh_lo || this_tsc_khz > thresh_hi) {
1247 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", this_tsc_khz, thresh_lo, thresh_hi);
1250 kvm_x86_ops->set_tsc_khz(vcpu, this_tsc_khz, use_scaling);
1253 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1255 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1256 vcpu->arch.virtual_tsc_mult,
1257 vcpu->arch.virtual_tsc_shift);
1258 tsc += vcpu->arch.this_tsc_write;
1262 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1264 #ifdef CONFIG_X86_64
1266 struct kvm_arch *ka = &vcpu->kvm->arch;
1267 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1269 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1270 atomic_read(&vcpu->kvm->online_vcpus));
1273 * Once the masterclock is enabled, always perform request in
1274 * order to update it.
1276 * In order to enable masterclock, the host clocksource must be TSC
1277 * and the vcpus need to have matched TSCs. When that happens,
1278 * perform request to enable masterclock.
1280 if (ka->use_master_clock ||
1281 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1282 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1284 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1285 atomic_read(&vcpu->kvm->online_vcpus),
1286 ka->use_master_clock, gtod->clock.vclock_mode);
1290 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1292 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1293 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1296 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1298 struct kvm *kvm = vcpu->kvm;
1299 u64 offset, ns, elapsed;
1300 unsigned long flags;
1303 bool already_matched;
1304 u64 data = msr->data;
1306 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1307 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1308 ns = get_kernel_ns();
1309 elapsed = ns - kvm->arch.last_tsc_nsec;
1311 if (vcpu->arch.virtual_tsc_khz) {
1314 /* n.b - signed multiplication and division required */
1315 usdiff = data - kvm->arch.last_tsc_write;
1316 #ifdef CONFIG_X86_64
1317 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1319 /* do_div() only does unsigned */
1320 asm("1: idivl %[divisor]\n"
1321 "2: xor %%edx, %%edx\n"
1322 " movl $0, %[faulted]\n"
1324 ".section .fixup,\"ax\"\n"
1325 "4: movl $1, %[faulted]\n"
1329 _ASM_EXTABLE(1b, 4b)
1331 : "=A"(usdiff), [faulted] "=r" (faulted)
1332 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1335 do_div(elapsed, 1000);
1340 /* idivl overflow => difference is larger than USEC_PER_SEC */
1342 usdiff = USEC_PER_SEC;
1344 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1347 * Special case: TSC write with a small delta (1 second) of virtual
1348 * cycle time against real time is interpreted as an attempt to
1349 * synchronize the CPU.
1351 * For a reliable TSC, we can match TSC offsets, and for an unstable
1352 * TSC, we add elapsed time in this computation. We could let the
1353 * compensation code attempt to catch up if we fall behind, but
1354 * it's better to try to match offsets from the beginning.
1356 if (usdiff < USEC_PER_SEC &&
1357 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1358 if (!check_tsc_unstable()) {
1359 offset = kvm->arch.cur_tsc_offset;
1360 pr_debug("kvm: matched tsc offset for %llu\n", data);
1362 u64 delta = nsec_to_cycles(vcpu, elapsed);
1364 offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1365 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1368 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1371 * We split periods of matched TSC writes into generations.
1372 * For each generation, we track the original measured
1373 * nanosecond time, offset, and write, so if TSCs are in
1374 * sync, we can match exact offset, and if not, we can match
1375 * exact software computation in compute_guest_tsc()
1377 * These values are tracked in kvm->arch.cur_xxx variables.
1379 kvm->arch.cur_tsc_generation++;
1380 kvm->arch.cur_tsc_nsec = ns;
1381 kvm->arch.cur_tsc_write = data;
1382 kvm->arch.cur_tsc_offset = offset;
1384 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1385 kvm->arch.cur_tsc_generation, data);
1389 * We also track th most recent recorded KHZ, write and time to
1390 * allow the matching interval to be extended at each write.
1392 kvm->arch.last_tsc_nsec = ns;
1393 kvm->arch.last_tsc_write = data;
1394 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1396 vcpu->arch.last_guest_tsc = data;
1398 /* Keep track of which generation this VCPU has synchronized to */
1399 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1400 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1401 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1403 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1404 update_ia32_tsc_adjust_msr(vcpu, offset);
1405 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1406 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1408 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1410 kvm->arch.nr_vcpus_matched_tsc = 0;
1411 } else if (!already_matched) {
1412 kvm->arch.nr_vcpus_matched_tsc++;
1415 kvm_track_tsc_matching(vcpu);
1416 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1419 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1421 #ifdef CONFIG_X86_64
1423 static cycle_t read_tsc(void)
1429 * Empirically, a fence (of type that depends on the CPU)
1430 * before rdtsc is enough to ensure that rdtsc is ordered
1431 * with respect to loads. The various CPU manuals are unclear
1432 * as to whether rdtsc can be reordered with later loads,
1433 * but no one has ever seen it happen.
1436 ret = (cycle_t)vget_cycles();
1438 last = pvclock_gtod_data.clock.cycle_last;
1440 if (likely(ret >= last))
1444 * GCC likes to generate cmov here, but this branch is extremely
1445 * predictable (it's just a funciton of time and the likely is
1446 * very likely) and there's a data dependence, so force GCC
1447 * to generate a branch instead. I don't barrier() because
1448 * we don't actually need a barrier, and if this function
1449 * ever gets inlined it will generate worse code.
1455 static inline u64 vgettsc(cycle_t *cycle_now)
1458 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1460 *cycle_now = read_tsc();
1462 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1463 return v * gtod->clock.mult;
1466 static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1468 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1474 seq = read_seqcount_begin(>od->seq);
1475 mode = gtod->clock.vclock_mode;
1476 ns = gtod->nsec_base;
1477 ns += vgettsc(cycle_now);
1478 ns >>= gtod->clock.shift;
1479 ns += gtod->boot_ns;
1480 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1486 /* returns true if host is using tsc clocksource */
1487 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1489 /* checked again under seqlock below */
1490 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1493 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1499 * Assuming a stable TSC across physical CPUS, and a stable TSC
1500 * across virtual CPUs, the following condition is possible.
1501 * Each numbered line represents an event visible to both
1502 * CPUs at the next numbered event.
1504 * "timespecX" represents host monotonic time. "tscX" represents
1507 * VCPU0 on CPU0 | VCPU1 on CPU1
1509 * 1. read timespec0,tsc0
1510 * 2. | timespec1 = timespec0 + N
1512 * 3. transition to guest | transition to guest
1513 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1514 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1515 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1517 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1520 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1522 * - 0 < N - M => M < N
1524 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1525 * always the case (the difference between two distinct xtime instances
1526 * might be smaller then the difference between corresponding TSC reads,
1527 * when updating guest vcpus pvclock areas).
1529 * To avoid that problem, do not allow visibility of distinct
1530 * system_timestamp/tsc_timestamp values simultaneously: use a master
1531 * copy of host monotonic time values. Update that master copy
1534 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1538 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1540 #ifdef CONFIG_X86_64
1541 struct kvm_arch *ka = &kvm->arch;
1543 bool host_tsc_clocksource, vcpus_matched;
1545 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1546 atomic_read(&kvm->online_vcpus));
1549 * If the host uses TSC clock, then passthrough TSC as stable
1552 host_tsc_clocksource = kvm_get_time_and_clockread(
1553 &ka->master_kernel_ns,
1554 &ka->master_cycle_now);
1556 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1557 && !backwards_tsc_observed
1558 && !ka->boot_vcpu_runs_old_kvmclock;
1560 if (ka->use_master_clock)
1561 atomic_set(&kvm_guest_has_master_clock, 1);
1563 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1564 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1569 static void kvm_gen_update_masterclock(struct kvm *kvm)
1571 #ifdef CONFIG_X86_64
1573 struct kvm_vcpu *vcpu;
1574 struct kvm_arch *ka = &kvm->arch;
1576 spin_lock(&ka->pvclock_gtod_sync_lock);
1577 kvm_make_mclock_inprogress_request(kvm);
1578 /* no guest entries from this point */
1579 pvclock_update_vm_gtod_copy(kvm);
1581 kvm_for_each_vcpu(i, vcpu, kvm)
1582 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1584 /* guest entries allowed */
1585 kvm_for_each_vcpu(i, vcpu, kvm)
1586 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1588 spin_unlock(&ka->pvclock_gtod_sync_lock);
1592 static int kvm_guest_time_update(struct kvm_vcpu *v)
1594 unsigned long flags, this_tsc_khz;
1595 struct kvm_vcpu_arch *vcpu = &v->arch;
1596 struct kvm_arch *ka = &v->kvm->arch;
1598 u64 tsc_timestamp, host_tsc;
1599 struct pvclock_vcpu_time_info guest_hv_clock;
1601 bool use_master_clock;
1607 * If the host uses TSC clock, then passthrough TSC as stable
1610 spin_lock(&ka->pvclock_gtod_sync_lock);
1611 use_master_clock = ka->use_master_clock;
1612 if (use_master_clock) {
1613 host_tsc = ka->master_cycle_now;
1614 kernel_ns = ka->master_kernel_ns;
1616 spin_unlock(&ka->pvclock_gtod_sync_lock);
1618 /* Keep irq disabled to prevent changes to the clock */
1619 local_irq_save(flags);
1620 this_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1621 if (unlikely(this_tsc_khz == 0)) {
1622 local_irq_restore(flags);
1623 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1626 if (!use_master_clock) {
1627 host_tsc = native_read_tsc();
1628 kernel_ns = get_kernel_ns();
1631 tsc_timestamp = kvm_x86_ops->read_l1_tsc(v, host_tsc);
1634 * We may have to catch up the TSC to match elapsed wall clock
1635 * time for two reasons, even if kvmclock is used.
1636 * 1) CPU could have been running below the maximum TSC rate
1637 * 2) Broken TSC compensation resets the base at each VCPU
1638 * entry to avoid unknown leaps of TSC even when running
1639 * again on the same CPU. This may cause apparent elapsed
1640 * time to disappear, and the guest to stand still or run
1643 if (vcpu->tsc_catchup) {
1644 u64 tsc = compute_guest_tsc(v, kernel_ns);
1645 if (tsc > tsc_timestamp) {
1646 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1647 tsc_timestamp = tsc;
1651 local_irq_restore(flags);
1653 if (!vcpu->pv_time_enabled)
1656 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1657 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1658 &vcpu->hv_clock.tsc_shift,
1659 &vcpu->hv_clock.tsc_to_system_mul);
1660 vcpu->hw_tsc_khz = this_tsc_khz;
1663 /* With all the info we got, fill in the values */
1664 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1665 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1666 vcpu->last_guest_tsc = tsc_timestamp;
1668 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1669 &guest_hv_clock, sizeof(guest_hv_clock))))
1673 * The interface expects us to write an even number signaling that the
1674 * update is finished. Since the guest won't see the intermediate
1675 * state, we just increase by 2 at the end.
1677 vcpu->hv_clock.version = guest_hv_clock.version + 2;
1679 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1680 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1682 if (vcpu->pvclock_set_guest_stopped_request) {
1683 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1684 vcpu->pvclock_set_guest_stopped_request = false;
1687 /* If the host uses TSC clocksource, then it is stable */
1688 if (use_master_clock)
1689 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1691 vcpu->hv_clock.flags = pvclock_flags;
1693 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1695 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1697 sizeof(vcpu->hv_clock));
1702 * kvmclock updates which are isolated to a given vcpu, such as
1703 * vcpu->cpu migration, should not allow system_timestamp from
1704 * the rest of the vcpus to remain static. Otherwise ntp frequency
1705 * correction applies to one vcpu's system_timestamp but not
1708 * So in those cases, request a kvmclock update for all vcpus.
1709 * We need to rate-limit these requests though, as they can
1710 * considerably slow guests that have a large number of vcpus.
1711 * The time for a remote vcpu to update its kvmclock is bound
1712 * by the delay we use to rate-limit the updates.
1715 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1717 static void kvmclock_update_fn(struct work_struct *work)
1720 struct delayed_work *dwork = to_delayed_work(work);
1721 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1722 kvmclock_update_work);
1723 struct kvm *kvm = container_of(ka, struct kvm, arch);
1724 struct kvm_vcpu *vcpu;
1726 kvm_for_each_vcpu(i, vcpu, kvm) {
1727 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1728 kvm_vcpu_kick(vcpu);
1732 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1734 struct kvm *kvm = v->kvm;
1736 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1737 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1738 KVMCLOCK_UPDATE_DELAY);
1741 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1743 static void kvmclock_sync_fn(struct work_struct *work)
1745 struct delayed_work *dwork = to_delayed_work(work);
1746 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1747 kvmclock_sync_work);
1748 struct kvm *kvm = container_of(ka, struct kvm, arch);
1750 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1751 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1752 KVMCLOCK_SYNC_PERIOD);
1755 static bool msr_mtrr_valid(unsigned msr)
1758 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1759 case MSR_MTRRfix64K_00000:
1760 case MSR_MTRRfix16K_80000:
1761 case MSR_MTRRfix16K_A0000:
1762 case MSR_MTRRfix4K_C0000:
1763 case MSR_MTRRfix4K_C8000:
1764 case MSR_MTRRfix4K_D0000:
1765 case MSR_MTRRfix4K_D8000:
1766 case MSR_MTRRfix4K_E0000:
1767 case MSR_MTRRfix4K_E8000:
1768 case MSR_MTRRfix4K_F0000:
1769 case MSR_MTRRfix4K_F8000:
1770 case MSR_MTRRdefType:
1771 case MSR_IA32_CR_PAT:
1779 static bool valid_pat_type(unsigned t)
1781 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1784 static bool valid_mtrr_type(unsigned t)
1786 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1789 bool kvm_mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1794 if (!msr_mtrr_valid(msr))
1797 if (msr == MSR_IA32_CR_PAT) {
1798 for (i = 0; i < 8; i++)
1799 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1802 } else if (msr == MSR_MTRRdefType) {
1805 return valid_mtrr_type(data & 0xff);
1806 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1807 for (i = 0; i < 8 ; i++)
1808 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1813 /* variable MTRRs */
1814 WARN_ON(!(msr >= 0x200 && msr < 0x200 + 2 * KVM_NR_VAR_MTRR));
1816 mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
1817 if ((msr & 1) == 0) {
1819 if (!valid_mtrr_type(data & 0xff))
1826 kvm_inject_gp(vcpu, 0);
1832 EXPORT_SYMBOL_GPL(kvm_mtrr_valid);
1834 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1836 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1838 if (!kvm_mtrr_valid(vcpu, msr, data))
1841 if (msr == MSR_MTRRdefType) {
1842 vcpu->arch.mtrr_state.def_type = data;
1843 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1844 } else if (msr == MSR_MTRRfix64K_00000)
1846 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1847 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1848 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1849 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1850 else if (msr == MSR_IA32_CR_PAT)
1851 vcpu->arch.pat = data;
1852 else { /* Variable MTRRs */
1853 int idx, is_mtrr_mask;
1856 idx = (msr - 0x200) / 2;
1857 is_mtrr_mask = msr - 0x200 - 2 * idx;
1860 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1863 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1867 kvm_mmu_reset_context(vcpu);
1871 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1873 u64 mcg_cap = vcpu->arch.mcg_cap;
1874 unsigned bank_num = mcg_cap & 0xff;
1877 case MSR_IA32_MCG_STATUS:
1878 vcpu->arch.mcg_status = data;
1880 case MSR_IA32_MCG_CTL:
1881 if (!(mcg_cap & MCG_CTL_P))
1883 if (data != 0 && data != ~(u64)0)
1885 vcpu->arch.mcg_ctl = data;
1888 if (msr >= MSR_IA32_MC0_CTL &&
1889 msr < MSR_IA32_MCx_CTL(bank_num)) {
1890 u32 offset = msr - MSR_IA32_MC0_CTL;
1891 /* only 0 or all 1s can be written to IA32_MCi_CTL
1892 * some Linux kernels though clear bit 10 in bank 4 to
1893 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1894 * this to avoid an uncatched #GP in the guest
1896 if ((offset & 0x3) == 0 &&
1897 data != 0 && (data | (1 << 10)) != ~(u64)0)
1899 vcpu->arch.mce_banks[offset] = data;
1907 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1909 struct kvm *kvm = vcpu->kvm;
1910 int lm = is_long_mode(vcpu);
1911 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1912 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1913 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1914 : kvm->arch.xen_hvm_config.blob_size_32;
1915 u32 page_num = data & ~PAGE_MASK;
1916 u64 page_addr = data & PAGE_MASK;
1921 if (page_num >= blob_size)
1924 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1929 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1938 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1940 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1943 static bool kvm_hv_msr_partition_wide(u32 msr)
1947 case HV_X64_MSR_GUEST_OS_ID:
1948 case HV_X64_MSR_HYPERCALL:
1949 case HV_X64_MSR_REFERENCE_TSC:
1950 case HV_X64_MSR_TIME_REF_COUNT:
1958 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1960 struct kvm *kvm = vcpu->kvm;
1963 case HV_X64_MSR_GUEST_OS_ID:
1964 kvm->arch.hv_guest_os_id = data;
1965 /* setting guest os id to zero disables hypercall page */
1966 if (!kvm->arch.hv_guest_os_id)
1967 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1969 case HV_X64_MSR_HYPERCALL: {
1974 /* if guest os id is not set hypercall should remain disabled */
1975 if (!kvm->arch.hv_guest_os_id)
1977 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1978 kvm->arch.hv_hypercall = data;
1981 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1982 addr = gfn_to_hva(kvm, gfn);
1983 if (kvm_is_error_hva(addr))
1985 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1986 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1987 if (__copy_to_user((void __user *)addr, instructions, 4))
1989 kvm->arch.hv_hypercall = data;
1990 mark_page_dirty(kvm, gfn);
1993 case HV_X64_MSR_REFERENCE_TSC: {
1995 HV_REFERENCE_TSC_PAGE tsc_ref;
1996 memset(&tsc_ref, 0, sizeof(tsc_ref));
1997 kvm->arch.hv_tsc_page = data;
1998 if (!(data & HV_X64_MSR_TSC_REFERENCE_ENABLE))
2000 gfn = data >> HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT;
2001 if (kvm_write_guest(kvm, gfn << HV_X64_MSR_TSC_REFERENCE_ADDRESS_SHIFT,
2002 &tsc_ref, sizeof(tsc_ref)))
2004 mark_page_dirty(kvm, gfn);
2008 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
2009 "data 0x%llx\n", msr, data);
2015 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2018 case HV_X64_MSR_APIC_ASSIST_PAGE: {
2022 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
2023 vcpu->arch.hv_vapic = data;
2024 if (kvm_lapic_enable_pv_eoi(vcpu, 0))
2028 gfn = data >> HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT;
2029 addr = gfn_to_hva(vcpu->kvm, gfn);
2030 if (kvm_is_error_hva(addr))
2032 if (__clear_user((void __user *)addr, PAGE_SIZE))
2034 vcpu->arch.hv_vapic = data;
2035 mark_page_dirty(vcpu->kvm, gfn);
2036 if (kvm_lapic_enable_pv_eoi(vcpu, gfn_to_gpa(gfn) | KVM_MSR_ENABLED))
2040 case HV_X64_MSR_EOI:
2041 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
2042 case HV_X64_MSR_ICR:
2043 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
2044 case HV_X64_MSR_TPR:
2045 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
2047 vcpu_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
2048 "data 0x%llx\n", msr, data);
2055 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2057 gpa_t gpa = data & ~0x3f;
2059 /* Bits 2:5 are reserved, Should be zero */
2063 vcpu->arch.apf.msr_val = data;
2065 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2066 kvm_clear_async_pf_completion_queue(vcpu);
2067 kvm_async_pf_hash_reset(vcpu);
2071 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2075 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2076 kvm_async_pf_wakeup_all(vcpu);
2080 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2082 vcpu->arch.pv_time_enabled = false;
2085 static void accumulate_steal_time(struct kvm_vcpu *vcpu)
2089 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2092 delta = current->sched_info.run_delay - vcpu->arch.st.last_steal;
2093 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2094 vcpu->arch.st.accum_steal = delta;
2097 static void record_steal_time(struct kvm_vcpu *vcpu)
2099 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2102 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2103 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2106 vcpu->arch.st.steal.steal += vcpu->arch.st.accum_steal;
2107 vcpu->arch.st.steal.version += 2;
2108 vcpu->arch.st.accum_steal = 0;
2110 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2111 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2114 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2117 u32 msr = msr_info->index;
2118 u64 data = msr_info->data;
2121 case MSR_AMD64_NB_CFG:
2122 case MSR_IA32_UCODE_REV:
2123 case MSR_IA32_UCODE_WRITE:
2124 case MSR_VM_HSAVE_PA:
2125 case MSR_AMD64_PATCH_LOADER:
2126 case MSR_AMD64_BU_CFG2:
2130 return set_efer(vcpu, data);
2132 data &= ~(u64)0x40; /* ignore flush filter disable */
2133 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2134 data &= ~(u64)0x8; /* ignore TLB cache disable */
2135 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2137 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2142 case MSR_FAM10H_MMIO_CONF_BASE:
2144 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2149 case MSR_IA32_DEBUGCTLMSR:
2151 /* We support the non-activated case already */
2153 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2154 /* Values other than LBR and BTF are vendor-specific,
2155 thus reserved and should throw a #GP */
2158 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2161 case 0x200 ... 0x2ff:
2162 return set_msr_mtrr(vcpu, msr, data);
2163 case MSR_IA32_APICBASE:
2164 return kvm_set_apic_base(vcpu, msr_info);
2165 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2166 return kvm_x2apic_msr_write(vcpu, msr, data);
2167 case MSR_IA32_TSCDEADLINE:
2168 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2170 case MSR_IA32_TSC_ADJUST:
2171 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2172 if (!msr_info->host_initiated) {
2173 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2174 kvm_x86_ops->adjust_tsc_offset(vcpu, adj, true);
2176 vcpu->arch.ia32_tsc_adjust_msr = data;
2179 case MSR_IA32_MISC_ENABLE:
2180 vcpu->arch.ia32_misc_enable_msr = data;
2182 case MSR_KVM_WALL_CLOCK_NEW:
2183 case MSR_KVM_WALL_CLOCK:
2184 vcpu->kvm->arch.wall_clock = data;
2185 kvm_write_wall_clock(vcpu->kvm, data);
2187 case MSR_KVM_SYSTEM_TIME_NEW:
2188 case MSR_KVM_SYSTEM_TIME: {
2190 struct kvm_arch *ka = &vcpu->kvm->arch;
2192 kvmclock_reset(vcpu);
2194 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2195 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2197 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2198 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2201 ka->boot_vcpu_runs_old_kvmclock = tmp;
2204 vcpu->arch.time = data;
2205 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2207 /* we verify if the enable bit is set... */
2211 gpa_offset = data & ~(PAGE_MASK | 1);
2213 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2214 &vcpu->arch.pv_time, data & ~1ULL,
2215 sizeof(struct pvclock_vcpu_time_info)))
2216 vcpu->arch.pv_time_enabled = false;
2218 vcpu->arch.pv_time_enabled = true;
2222 case MSR_KVM_ASYNC_PF_EN:
2223 if (kvm_pv_enable_async_pf(vcpu, data))
2226 case MSR_KVM_STEAL_TIME:
2228 if (unlikely(!sched_info_on()))
2231 if (data & KVM_STEAL_RESERVED_MASK)
2234 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2235 data & KVM_STEAL_VALID_BITS,
2236 sizeof(struct kvm_steal_time)))
2239 vcpu->arch.st.msr_val = data;
2241 if (!(data & KVM_MSR_ENABLED))
2244 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2247 accumulate_steal_time(vcpu);
2250 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2253 case MSR_KVM_PV_EOI_EN:
2254 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2258 case MSR_IA32_MCG_CTL:
2259 case MSR_IA32_MCG_STATUS:
2260 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2261 return set_msr_mce(vcpu, msr, data);
2263 /* Performance counters are not protected by a CPUID bit,
2264 * so we should check all of them in the generic path for the sake of
2265 * cross vendor migration.
2266 * Writing a zero into the event select MSRs disables them,
2267 * which we perfectly emulate ;-). Any other value should be at least
2268 * reported, some guests depend on them.
2270 case MSR_K7_EVNTSEL0:
2271 case MSR_K7_EVNTSEL1:
2272 case MSR_K7_EVNTSEL2:
2273 case MSR_K7_EVNTSEL3:
2275 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2276 "0x%x data 0x%llx\n", msr, data);
2278 /* at least RHEL 4 unconditionally writes to the perfctr registers,
2279 * so we ignore writes to make it happy.
2281 case MSR_K7_PERFCTR0:
2282 case MSR_K7_PERFCTR1:
2283 case MSR_K7_PERFCTR2:
2284 case MSR_K7_PERFCTR3:
2285 vcpu_unimpl(vcpu, "unimplemented perfctr wrmsr: "
2286 "0x%x data 0x%llx\n", msr, data);
2288 case MSR_P6_PERFCTR0:
2289 case MSR_P6_PERFCTR1:
2291 case MSR_P6_EVNTSEL0:
2292 case MSR_P6_EVNTSEL1:
2293 if (kvm_pmu_msr(vcpu, msr))
2294 return kvm_pmu_set_msr(vcpu, msr_info);
2296 if (pr || data != 0)
2297 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2298 "0x%x data 0x%llx\n", msr, data);
2300 case MSR_K7_CLK_CTL:
2302 * Ignore all writes to this no longer documented MSR.
2303 * Writes are only relevant for old K7 processors,
2304 * all pre-dating SVM, but a recommended workaround from
2305 * AMD for these chips. It is possible to specify the
2306 * affected processor models on the command line, hence
2307 * the need to ignore the workaround.
2310 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2311 if (kvm_hv_msr_partition_wide(msr)) {
2313 mutex_lock(&vcpu->kvm->lock);
2314 r = set_msr_hyperv_pw(vcpu, msr, data);
2315 mutex_unlock(&vcpu->kvm->lock);
2318 return set_msr_hyperv(vcpu, msr, data);
2320 case MSR_IA32_BBL_CR_CTL3:
2321 /* Drop writes to this legacy MSR -- see rdmsr
2322 * counterpart for further detail.
2324 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2326 case MSR_AMD64_OSVW_ID_LENGTH:
2327 if (!guest_cpuid_has_osvw(vcpu))
2329 vcpu->arch.osvw.length = data;
2331 case MSR_AMD64_OSVW_STATUS:
2332 if (!guest_cpuid_has_osvw(vcpu))
2334 vcpu->arch.osvw.status = data;
2337 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2338 return xen_hvm_config(vcpu, data);
2339 if (kvm_pmu_msr(vcpu, msr))
2340 return kvm_pmu_set_msr(vcpu, msr_info);
2342 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2346 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2353 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2357 * Reads an msr value (of 'msr_index') into 'pdata'.
2358 * Returns 0 on success, non-0 otherwise.
2359 * Assumes vcpu_load() was already called.
2361 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2363 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
2365 EXPORT_SYMBOL_GPL(kvm_get_msr);
2367 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2369 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
2371 if (!msr_mtrr_valid(msr))
2374 if (msr == MSR_MTRRdefType)
2375 *pdata = vcpu->arch.mtrr_state.def_type +
2376 (vcpu->arch.mtrr_state.enabled << 10);
2377 else if (msr == MSR_MTRRfix64K_00000)
2379 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
2380 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
2381 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
2382 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
2383 else if (msr == MSR_IA32_CR_PAT)
2384 *pdata = vcpu->arch.pat;
2385 else { /* Variable MTRRs */
2386 int idx, is_mtrr_mask;
2389 idx = (msr - 0x200) / 2;
2390 is_mtrr_mask = msr - 0x200 - 2 * idx;
2393 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
2396 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
2403 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2406 u64 mcg_cap = vcpu->arch.mcg_cap;
2407 unsigned bank_num = mcg_cap & 0xff;
2410 case MSR_IA32_P5_MC_ADDR:
2411 case MSR_IA32_P5_MC_TYPE:
2414 case MSR_IA32_MCG_CAP:
2415 data = vcpu->arch.mcg_cap;
2417 case MSR_IA32_MCG_CTL:
2418 if (!(mcg_cap & MCG_CTL_P))
2420 data = vcpu->arch.mcg_ctl;
2422 case MSR_IA32_MCG_STATUS:
2423 data = vcpu->arch.mcg_status;
2426 if (msr >= MSR_IA32_MC0_CTL &&
2427 msr < MSR_IA32_MCx_CTL(bank_num)) {
2428 u32 offset = msr - MSR_IA32_MC0_CTL;
2429 data = vcpu->arch.mce_banks[offset];
2438 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2441 struct kvm *kvm = vcpu->kvm;
2444 case HV_X64_MSR_GUEST_OS_ID:
2445 data = kvm->arch.hv_guest_os_id;
2447 case HV_X64_MSR_HYPERCALL:
2448 data = kvm->arch.hv_hypercall;
2450 case HV_X64_MSR_TIME_REF_COUNT: {
2452 div_u64(get_kernel_ns() + kvm->arch.kvmclock_offset, 100);
2455 case HV_X64_MSR_REFERENCE_TSC:
2456 data = kvm->arch.hv_tsc_page;
2459 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2467 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2472 case HV_X64_MSR_VP_INDEX: {
2475 kvm_for_each_vcpu(r, v, vcpu->kvm) {
2483 case HV_X64_MSR_EOI:
2484 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
2485 case HV_X64_MSR_ICR:
2486 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
2487 case HV_X64_MSR_TPR:
2488 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
2489 case HV_X64_MSR_APIC_ASSIST_PAGE:
2490 data = vcpu->arch.hv_vapic;
2493 vcpu_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
2500 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2505 case MSR_IA32_PLATFORM_ID:
2506 case MSR_IA32_EBL_CR_POWERON:
2507 case MSR_IA32_DEBUGCTLMSR:
2508 case MSR_IA32_LASTBRANCHFROMIP:
2509 case MSR_IA32_LASTBRANCHTOIP:
2510 case MSR_IA32_LASTINTFROMIP:
2511 case MSR_IA32_LASTINTTOIP:
2514 case MSR_VM_HSAVE_PA:
2515 case MSR_K7_EVNTSEL0:
2516 case MSR_K7_EVNTSEL1:
2517 case MSR_K7_EVNTSEL2:
2518 case MSR_K7_EVNTSEL3:
2519 case MSR_K7_PERFCTR0:
2520 case MSR_K7_PERFCTR1:
2521 case MSR_K7_PERFCTR2:
2522 case MSR_K7_PERFCTR3:
2523 case MSR_K8_INT_PENDING_MSG:
2524 case MSR_AMD64_NB_CFG:
2525 case MSR_FAM10H_MMIO_CONF_BASE:
2526 case MSR_AMD64_BU_CFG2:
2529 case MSR_P6_PERFCTR0:
2530 case MSR_P6_PERFCTR1:
2531 case MSR_P6_EVNTSEL0:
2532 case MSR_P6_EVNTSEL1:
2533 if (kvm_pmu_msr(vcpu, msr))
2534 return kvm_pmu_get_msr(vcpu, msr, pdata);
2537 case MSR_IA32_UCODE_REV:
2538 data = 0x100000000ULL;
2541 data = 0x500 | KVM_NR_VAR_MTRR;
2543 case 0x200 ... 0x2ff:
2544 return get_msr_mtrr(vcpu, msr, pdata);
2545 case 0xcd: /* fsb frequency */
2549 * MSR_EBC_FREQUENCY_ID
2550 * Conservative value valid for even the basic CPU models.
2551 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2552 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2553 * and 266MHz for model 3, or 4. Set Core Clock
2554 * Frequency to System Bus Frequency Ratio to 1 (bits
2555 * 31:24) even though these are only valid for CPU
2556 * models > 2, however guests may end up dividing or
2557 * multiplying by zero otherwise.
2559 case MSR_EBC_FREQUENCY_ID:
2562 case MSR_IA32_APICBASE:
2563 data = kvm_get_apic_base(vcpu);
2565 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2566 return kvm_x2apic_msr_read(vcpu, msr, pdata);
2568 case MSR_IA32_TSCDEADLINE:
2569 data = kvm_get_lapic_tscdeadline_msr(vcpu);
2571 case MSR_IA32_TSC_ADJUST:
2572 data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2574 case MSR_IA32_MISC_ENABLE:
2575 data = vcpu->arch.ia32_misc_enable_msr;
2577 case MSR_IA32_PERF_STATUS:
2578 /* TSC increment by tick */
2580 /* CPU multiplier */
2581 data |= (((uint64_t)4ULL) << 40);
2584 data = vcpu->arch.efer;
2586 case MSR_KVM_WALL_CLOCK:
2587 case MSR_KVM_WALL_CLOCK_NEW:
2588 data = vcpu->kvm->arch.wall_clock;
2590 case MSR_KVM_SYSTEM_TIME:
2591 case MSR_KVM_SYSTEM_TIME_NEW:
2592 data = vcpu->arch.time;
2594 case MSR_KVM_ASYNC_PF_EN:
2595 data = vcpu->arch.apf.msr_val;
2597 case MSR_KVM_STEAL_TIME:
2598 data = vcpu->arch.st.msr_val;
2600 case MSR_KVM_PV_EOI_EN:
2601 data = vcpu->arch.pv_eoi.msr_val;
2603 case MSR_IA32_P5_MC_ADDR:
2604 case MSR_IA32_P5_MC_TYPE:
2605 case MSR_IA32_MCG_CAP:
2606 case MSR_IA32_MCG_CTL:
2607 case MSR_IA32_MCG_STATUS:
2608 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2609 return get_msr_mce(vcpu, msr, pdata);
2610 case MSR_K7_CLK_CTL:
2612 * Provide expected ramp-up count for K7. All other
2613 * are set to zero, indicating minimum divisors for
2616 * This prevents guest kernels on AMD host with CPU
2617 * type 6, model 8 and higher from exploding due to
2618 * the rdmsr failing.
2622 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2623 if (kvm_hv_msr_partition_wide(msr)) {
2625 mutex_lock(&vcpu->kvm->lock);
2626 r = get_msr_hyperv_pw(vcpu, msr, pdata);
2627 mutex_unlock(&vcpu->kvm->lock);
2630 return get_msr_hyperv(vcpu, msr, pdata);
2632 case MSR_IA32_BBL_CR_CTL3:
2633 /* This legacy MSR exists but isn't fully documented in current
2634 * silicon. It is however accessed by winxp in very narrow
2635 * scenarios where it sets bit #19, itself documented as
2636 * a "reserved" bit. Best effort attempt to source coherent
2637 * read data here should the balance of the register be
2638 * interpreted by the guest:
2640 * L2 cache control register 3: 64GB range, 256KB size,
2641 * enabled, latency 0x1, configured
2645 case MSR_AMD64_OSVW_ID_LENGTH:
2646 if (!guest_cpuid_has_osvw(vcpu))
2648 data = vcpu->arch.osvw.length;
2650 case MSR_AMD64_OSVW_STATUS:
2651 if (!guest_cpuid_has_osvw(vcpu))
2653 data = vcpu->arch.osvw.status;
2656 if (kvm_pmu_msr(vcpu, msr))
2657 return kvm_pmu_get_msr(vcpu, msr, pdata);
2659 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
2662 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
2670 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2673 * Read or write a bunch of msrs. All parameters are kernel addresses.
2675 * @return number of msrs set successfully.
2677 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2678 struct kvm_msr_entry *entries,
2679 int (*do_msr)(struct kvm_vcpu *vcpu,
2680 unsigned index, u64 *data))
2684 idx = srcu_read_lock(&vcpu->kvm->srcu);
2685 for (i = 0; i < msrs->nmsrs; ++i)
2686 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2688 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2694 * Read or write a bunch of msrs. Parameters are user addresses.
2696 * @return number of msrs set successfully.
2698 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2699 int (*do_msr)(struct kvm_vcpu *vcpu,
2700 unsigned index, u64 *data),
2703 struct kvm_msrs msrs;
2704 struct kvm_msr_entry *entries;
2709 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2713 if (msrs.nmsrs >= MAX_IO_MSRS)
2716 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2717 entries = memdup_user(user_msrs->entries, size);
2718 if (IS_ERR(entries)) {
2719 r = PTR_ERR(entries);
2723 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2728 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2739 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2744 case KVM_CAP_IRQCHIP:
2746 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2747 case KVM_CAP_SET_TSS_ADDR:
2748 case KVM_CAP_EXT_CPUID:
2749 case KVM_CAP_EXT_EMUL_CPUID:
2750 case KVM_CAP_CLOCKSOURCE:
2752 case KVM_CAP_NOP_IO_DELAY:
2753 case KVM_CAP_MP_STATE:
2754 case KVM_CAP_SYNC_MMU:
2755 case KVM_CAP_USER_NMI:
2756 case KVM_CAP_REINJECT_CONTROL:
2757 case KVM_CAP_IRQ_INJECT_STATUS:
2759 case KVM_CAP_IOEVENTFD:
2760 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2762 case KVM_CAP_PIT_STATE2:
2763 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2764 case KVM_CAP_XEN_HVM:
2765 case KVM_CAP_ADJUST_CLOCK:
2766 case KVM_CAP_VCPU_EVENTS:
2767 case KVM_CAP_HYPERV:
2768 case KVM_CAP_HYPERV_VAPIC:
2769 case KVM_CAP_HYPERV_SPIN:
2770 case KVM_CAP_PCI_SEGMENT:
2771 case KVM_CAP_DEBUGREGS:
2772 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2774 case KVM_CAP_ASYNC_PF:
2775 case KVM_CAP_GET_TSC_KHZ:
2776 case KVM_CAP_KVMCLOCK_CTRL:
2777 case KVM_CAP_READONLY_MEM:
2778 case KVM_CAP_HYPERV_TIME:
2779 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2780 case KVM_CAP_TSC_DEADLINE_TIMER:
2781 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2782 case KVM_CAP_ASSIGN_DEV_IRQ:
2783 case KVM_CAP_PCI_2_3:
2787 case KVM_CAP_COALESCED_MMIO:
2788 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2791 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2793 case KVM_CAP_NR_VCPUS:
2794 r = KVM_SOFT_MAX_VCPUS;
2796 case KVM_CAP_MAX_VCPUS:
2799 case KVM_CAP_NR_MEMSLOTS:
2800 r = KVM_USER_MEM_SLOTS;
2802 case KVM_CAP_PV_MMU: /* obsolete */
2805 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2807 r = iommu_present(&pci_bus_type);
2811 r = KVM_MAX_MCE_BANKS;
2816 case KVM_CAP_TSC_CONTROL:
2817 r = kvm_has_tsc_control;
2827 long kvm_arch_dev_ioctl(struct file *filp,
2828 unsigned int ioctl, unsigned long arg)
2830 void __user *argp = (void __user *)arg;
2834 case KVM_GET_MSR_INDEX_LIST: {
2835 struct kvm_msr_list __user *user_msr_list = argp;
2836 struct kvm_msr_list msr_list;
2840 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2843 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2844 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2847 if (n < msr_list.nmsrs)
2850 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2851 num_msrs_to_save * sizeof(u32)))
2853 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2855 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2860 case KVM_GET_SUPPORTED_CPUID:
2861 case KVM_GET_EMULATED_CPUID: {
2862 struct kvm_cpuid2 __user *cpuid_arg = argp;
2863 struct kvm_cpuid2 cpuid;
2866 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2869 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2875 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2880 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2883 mce_cap = KVM_MCE_CAP_SUPPORTED;
2885 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2897 static void wbinvd_ipi(void *garbage)
2902 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2904 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2907 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2909 /* Address WBINVD may be executed by guest */
2910 if (need_emulate_wbinvd(vcpu)) {
2911 if (kvm_x86_ops->has_wbinvd_exit())
2912 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2913 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2914 smp_call_function_single(vcpu->cpu,
2915 wbinvd_ipi, NULL, 1);
2918 kvm_x86_ops->vcpu_load(vcpu, cpu);
2920 /* Apply any externally detected TSC adjustments (due to suspend) */
2921 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2922 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2923 vcpu->arch.tsc_offset_adjustment = 0;
2924 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2927 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2928 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2929 native_read_tsc() - vcpu->arch.last_host_tsc;
2931 mark_tsc_unstable("KVM discovered backwards TSC");
2932 if (check_tsc_unstable()) {
2933 u64 offset = kvm_x86_ops->compute_tsc_offset(vcpu,
2934 vcpu->arch.last_guest_tsc);
2935 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2936 vcpu->arch.tsc_catchup = 1;
2939 * On a host with synchronized TSC, there is no need to update
2940 * kvmclock on vcpu->cpu migration
2942 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2943 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2944 if (vcpu->cpu != cpu)
2945 kvm_migrate_timers(vcpu);
2949 accumulate_steal_time(vcpu);
2950 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2953 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2955 kvm_x86_ops->vcpu_put(vcpu);
2956 kvm_put_guest_fpu(vcpu);
2957 vcpu->arch.last_host_tsc = native_read_tsc();
2960 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2961 struct kvm_lapic_state *s)
2963 kvm_x86_ops->sync_pir_to_irr(vcpu);
2964 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2969 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2970 struct kvm_lapic_state *s)
2972 kvm_apic_post_state_restore(vcpu, s);
2973 update_cr8_intercept(vcpu);
2978 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2979 struct kvm_interrupt *irq)
2981 if (irq->irq >= KVM_NR_INTERRUPTS)
2983 if (irqchip_in_kernel(vcpu->kvm))
2986 kvm_queue_interrupt(vcpu, irq->irq, false);
2987 kvm_make_request(KVM_REQ_EVENT, vcpu);
2992 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2994 kvm_inject_nmi(vcpu);
2999 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3000 struct kvm_tpr_access_ctl *tac)
3004 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3008 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3012 unsigned bank_num = mcg_cap & 0xff, bank;
3015 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3017 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
3020 vcpu->arch.mcg_cap = mcg_cap;
3021 /* Init IA32_MCG_CTL to all 1s */
3022 if (mcg_cap & MCG_CTL_P)
3023 vcpu->arch.mcg_ctl = ~(u64)0;
3024 /* Init IA32_MCi_CTL to all 1s */
3025 for (bank = 0; bank < bank_num; bank++)
3026 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3031 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3032 struct kvm_x86_mce *mce)
3034 u64 mcg_cap = vcpu->arch.mcg_cap;
3035 unsigned bank_num = mcg_cap & 0xff;
3036 u64 *banks = vcpu->arch.mce_banks;
3038 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3041 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3042 * reporting is disabled
3044 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3045 vcpu->arch.mcg_ctl != ~(u64)0)
3047 banks += 4 * mce->bank;
3049 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3050 * reporting is disabled for the bank
3052 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3054 if (mce->status & MCI_STATUS_UC) {
3055 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3056 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3057 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3060 if (banks[1] & MCI_STATUS_VAL)
3061 mce->status |= MCI_STATUS_OVER;
3062 banks[2] = mce->addr;
3063 banks[3] = mce->misc;
3064 vcpu->arch.mcg_status = mce->mcg_status;
3065 banks[1] = mce->status;
3066 kvm_queue_exception(vcpu, MC_VECTOR);
3067 } else if (!(banks[1] & MCI_STATUS_VAL)
3068 || !(banks[1] & MCI_STATUS_UC)) {
3069 if (banks[1] & MCI_STATUS_VAL)
3070 mce->status |= MCI_STATUS_OVER;
3071 banks[2] = mce->addr;
3072 banks[3] = mce->misc;
3073 banks[1] = mce->status;
3075 banks[1] |= MCI_STATUS_OVER;
3079 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3080 struct kvm_vcpu_events *events)
3083 events->exception.injected =
3084 vcpu->arch.exception.pending &&
3085 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3086 events->exception.nr = vcpu->arch.exception.nr;
3087 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3088 events->exception.pad = 0;
3089 events->exception.error_code = vcpu->arch.exception.error_code;
3091 events->interrupt.injected =
3092 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3093 events->interrupt.nr = vcpu->arch.interrupt.nr;
3094 events->interrupt.soft = 0;
3095 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3097 events->nmi.injected = vcpu->arch.nmi_injected;
3098 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3099 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3100 events->nmi.pad = 0;
3102 events->sipi_vector = 0; /* never valid when reporting to user space */
3104 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3105 | KVM_VCPUEVENT_VALID_SHADOW);
3106 memset(&events->reserved, 0, sizeof(events->reserved));
3109 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3110 struct kvm_vcpu_events *events)
3112 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3113 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3114 | KVM_VCPUEVENT_VALID_SHADOW))
3118 vcpu->arch.exception.pending = events->exception.injected;
3119 vcpu->arch.exception.nr = events->exception.nr;
3120 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3121 vcpu->arch.exception.error_code = events->exception.error_code;
3123 vcpu->arch.interrupt.pending = events->interrupt.injected;
3124 vcpu->arch.interrupt.nr = events->interrupt.nr;
3125 vcpu->arch.interrupt.soft = events->interrupt.soft;
3126 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3127 kvm_x86_ops->set_interrupt_shadow(vcpu,
3128 events->interrupt.shadow);
3130 vcpu->arch.nmi_injected = events->nmi.injected;
3131 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3132 vcpu->arch.nmi_pending = events->nmi.pending;
3133 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3135 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3136 kvm_vcpu_has_lapic(vcpu))
3137 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3139 kvm_make_request(KVM_REQ_EVENT, vcpu);
3144 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3145 struct kvm_debugregs *dbgregs)
3149 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3150 kvm_get_dr(vcpu, 6, &val);
3152 dbgregs->dr7 = vcpu->arch.dr7;
3154 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3157 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3158 struct kvm_debugregs *dbgregs)
3163 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3164 kvm_update_dr0123(vcpu);
3165 vcpu->arch.dr6 = dbgregs->dr6;
3166 kvm_update_dr6(vcpu);
3167 vcpu->arch.dr7 = dbgregs->dr7;
3168 kvm_update_dr7(vcpu);
3173 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3175 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3177 struct xsave_struct *xsave = &vcpu->arch.guest_fpu.state->xsave;
3178 u64 xstate_bv = xsave->xsave_hdr.xstate_bv;
3182 * Copy legacy XSAVE area, to avoid complications with CPUID
3183 * leaves 0 and 1 in the loop below.
3185 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3188 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3191 * Copy each region from the possibly compacted offset to the
3192 * non-compacted offset.
3194 valid = xstate_bv & ~XSTATE_FPSSE;
3196 u64 feature = valid & -valid;
3197 int index = fls64(feature) - 1;
3198 void *src = get_xsave_addr(xsave, feature);
3201 u32 size, offset, ecx, edx;
3202 cpuid_count(XSTATE_CPUID, index,
3203 &size, &offset, &ecx, &edx);
3204 memcpy(dest + offset, src, size);
3211 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3213 struct xsave_struct *xsave = &vcpu->arch.guest_fpu.state->xsave;
3214 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3218 * Copy legacy XSAVE area, to avoid complications with CPUID
3219 * leaves 0 and 1 in the loop below.
3221 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3223 /* Set XSTATE_BV and possibly XCOMP_BV. */
3224 xsave->xsave_hdr.xstate_bv = xstate_bv;
3226 xsave->xsave_hdr.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3229 * Copy each region from the non-compacted offset to the
3230 * possibly compacted offset.
3232 valid = xstate_bv & ~XSTATE_FPSSE;
3234 u64 feature = valid & -valid;
3235 int index = fls64(feature) - 1;
3236 void *dest = get_xsave_addr(xsave, feature);
3239 u32 size, offset, ecx, edx;
3240 cpuid_count(XSTATE_CPUID, index,
3241 &size, &offset, &ecx, &edx);
3242 memcpy(dest, src + offset, size);
3250 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3251 struct kvm_xsave *guest_xsave)
3253 if (cpu_has_xsave) {
3254 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3255 fill_xsave((u8 *) guest_xsave->region, vcpu);
3257 memcpy(guest_xsave->region,
3258 &vcpu->arch.guest_fpu.state->fxsave,
3259 sizeof(struct i387_fxsave_struct));
3260 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3265 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3266 struct kvm_xsave *guest_xsave)
3269 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3271 if (cpu_has_xsave) {
3273 * Here we allow setting states that are not present in
3274 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3275 * with old userspace.
3277 if (xstate_bv & ~kvm_supported_xcr0())
3279 load_xsave(vcpu, (u8 *)guest_xsave->region);
3281 if (xstate_bv & ~XSTATE_FPSSE)
3283 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
3284 guest_xsave->region, sizeof(struct i387_fxsave_struct));
3289 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3290 struct kvm_xcrs *guest_xcrs)
3292 if (!cpu_has_xsave) {
3293 guest_xcrs->nr_xcrs = 0;
3297 guest_xcrs->nr_xcrs = 1;
3298 guest_xcrs->flags = 0;
3299 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3300 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3303 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3304 struct kvm_xcrs *guest_xcrs)
3311 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3314 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3315 /* Only support XCR0 currently */
3316 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3317 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3318 guest_xcrs->xcrs[i].value);
3327 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3328 * stopped by the hypervisor. This function will be called from the host only.
3329 * EINVAL is returned when the host attempts to set the flag for a guest that
3330 * does not support pv clocks.
3332 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3334 if (!vcpu->arch.pv_time_enabled)
3336 vcpu->arch.pvclock_set_guest_stopped_request = true;
3337 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3341 long kvm_arch_vcpu_ioctl(struct file *filp,
3342 unsigned int ioctl, unsigned long arg)
3344 struct kvm_vcpu *vcpu = filp->private_data;
3345 void __user *argp = (void __user *)arg;
3348 struct kvm_lapic_state *lapic;
3349 struct kvm_xsave *xsave;
3350 struct kvm_xcrs *xcrs;
3356 case KVM_GET_LAPIC: {
3358 if (!vcpu->arch.apic)
3360 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3365 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3369 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3374 case KVM_SET_LAPIC: {
3376 if (!vcpu->arch.apic)
3378 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3379 if (IS_ERR(u.lapic))
3380 return PTR_ERR(u.lapic);
3382 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3385 case KVM_INTERRUPT: {
3386 struct kvm_interrupt irq;
3389 if (copy_from_user(&irq, argp, sizeof irq))
3391 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3395 r = kvm_vcpu_ioctl_nmi(vcpu);
3398 case KVM_SET_CPUID: {
3399 struct kvm_cpuid __user *cpuid_arg = argp;
3400 struct kvm_cpuid cpuid;
3403 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3405 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3408 case KVM_SET_CPUID2: {
3409 struct kvm_cpuid2 __user *cpuid_arg = argp;
3410 struct kvm_cpuid2 cpuid;
3413 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3415 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3416 cpuid_arg->entries);
3419 case KVM_GET_CPUID2: {
3420 struct kvm_cpuid2 __user *cpuid_arg = argp;
3421 struct kvm_cpuid2 cpuid;
3424 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3426 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3427 cpuid_arg->entries);
3431 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3437 r = msr_io(vcpu, argp, kvm_get_msr, 1);
3440 r = msr_io(vcpu, argp, do_set_msr, 0);
3442 case KVM_TPR_ACCESS_REPORTING: {
3443 struct kvm_tpr_access_ctl tac;
3446 if (copy_from_user(&tac, argp, sizeof tac))
3448 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3452 if (copy_to_user(argp, &tac, sizeof tac))
3457 case KVM_SET_VAPIC_ADDR: {
3458 struct kvm_vapic_addr va;
3461 if (!irqchip_in_kernel(vcpu->kvm))
3464 if (copy_from_user(&va, argp, sizeof va))
3466 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3469 case KVM_X86_SETUP_MCE: {
3473 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3475 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3478 case KVM_X86_SET_MCE: {
3479 struct kvm_x86_mce mce;
3482 if (copy_from_user(&mce, argp, sizeof mce))
3484 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3487 case KVM_GET_VCPU_EVENTS: {
3488 struct kvm_vcpu_events events;
3490 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3493 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3498 case KVM_SET_VCPU_EVENTS: {
3499 struct kvm_vcpu_events events;
3502 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3505 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3508 case KVM_GET_DEBUGREGS: {
3509 struct kvm_debugregs dbgregs;
3511 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3514 if (copy_to_user(argp, &dbgregs,
3515 sizeof(struct kvm_debugregs)))
3520 case KVM_SET_DEBUGREGS: {
3521 struct kvm_debugregs dbgregs;
3524 if (copy_from_user(&dbgregs, argp,
3525 sizeof(struct kvm_debugregs)))
3528 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3531 case KVM_GET_XSAVE: {
3532 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3537 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3540 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3545 case KVM_SET_XSAVE: {
3546 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3547 if (IS_ERR(u.xsave))
3548 return PTR_ERR(u.xsave);
3550 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3553 case KVM_GET_XCRS: {
3554 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3559 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3562 if (copy_to_user(argp, u.xcrs,
3563 sizeof(struct kvm_xcrs)))
3568 case KVM_SET_XCRS: {
3569 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3571 return PTR_ERR(u.xcrs);
3573 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3576 case KVM_SET_TSC_KHZ: {
3580 user_tsc_khz = (u32)arg;
3582 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3585 if (user_tsc_khz == 0)
3586 user_tsc_khz = tsc_khz;
3588 kvm_set_tsc_khz(vcpu, user_tsc_khz);
3593 case KVM_GET_TSC_KHZ: {
3594 r = vcpu->arch.virtual_tsc_khz;
3597 case KVM_KVMCLOCK_CTRL: {
3598 r = kvm_set_guest_paused(vcpu);
3609 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3611 return VM_FAULT_SIGBUS;
3614 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3618 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3620 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3624 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3627 kvm->arch.ept_identity_map_addr = ident_addr;
3631 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3632 u32 kvm_nr_mmu_pages)
3634 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3637 mutex_lock(&kvm->slots_lock);
3639 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3640 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3642 mutex_unlock(&kvm->slots_lock);
3646 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3648 return kvm->arch.n_max_mmu_pages;
3651 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3656 switch (chip->chip_id) {
3657 case KVM_IRQCHIP_PIC_MASTER:
3658 memcpy(&chip->chip.pic,
3659 &pic_irqchip(kvm)->pics[0],
3660 sizeof(struct kvm_pic_state));
3662 case KVM_IRQCHIP_PIC_SLAVE:
3663 memcpy(&chip->chip.pic,
3664 &pic_irqchip(kvm)->pics[1],
3665 sizeof(struct kvm_pic_state));
3667 case KVM_IRQCHIP_IOAPIC:
3668 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3677 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3682 switch (chip->chip_id) {
3683 case KVM_IRQCHIP_PIC_MASTER:
3684 spin_lock(&pic_irqchip(kvm)->lock);
3685 memcpy(&pic_irqchip(kvm)->pics[0],
3687 sizeof(struct kvm_pic_state));
3688 spin_unlock(&pic_irqchip(kvm)->lock);
3690 case KVM_IRQCHIP_PIC_SLAVE:
3691 spin_lock(&pic_irqchip(kvm)->lock);
3692 memcpy(&pic_irqchip(kvm)->pics[1],
3694 sizeof(struct kvm_pic_state));
3695 spin_unlock(&pic_irqchip(kvm)->lock);
3697 case KVM_IRQCHIP_IOAPIC:
3698 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3704 kvm_pic_update_irq(pic_irqchip(kvm));
3708 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3712 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3713 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3714 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3718 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3722 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3723 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3724 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3725 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3729 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3733 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3734 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3735 sizeof(ps->channels));
3736 ps->flags = kvm->arch.vpit->pit_state.flags;
3737 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3738 memset(&ps->reserved, 0, sizeof(ps->reserved));
3742 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3744 int r = 0, start = 0;
3745 u32 prev_legacy, cur_legacy;
3746 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3747 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3748 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3749 if (!prev_legacy && cur_legacy)
3751 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3752 sizeof(kvm->arch.vpit->pit_state.channels));
3753 kvm->arch.vpit->pit_state.flags = ps->flags;
3754 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3755 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3759 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3760 struct kvm_reinject_control *control)
3762 if (!kvm->arch.vpit)
3764 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3765 kvm->arch.vpit->pit_state.reinject = control->pit_reinject;
3766 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3771 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3772 * @kvm: kvm instance
3773 * @log: slot id and address to which we copy the log
3775 * Steps 1-4 below provide general overview of dirty page logging. See
3776 * kvm_get_dirty_log_protect() function description for additional details.
3778 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3779 * always flush the TLB (step 4) even if previous step failed and the dirty
3780 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3781 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3782 * writes will be marked dirty for next log read.
3784 * 1. Take a snapshot of the bit and clear it if needed.
3785 * 2. Write protect the corresponding page.
3786 * 3. Copy the snapshot to the userspace.
3787 * 4. Flush TLB's if needed.
3789 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3791 bool is_dirty = false;
3794 mutex_lock(&kvm->slots_lock);
3797 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3799 if (kvm_x86_ops->flush_log_dirty)
3800 kvm_x86_ops->flush_log_dirty(kvm);
3802 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3805 * All the TLBs can be flushed out of mmu lock, see the comments in
3806 * kvm_mmu_slot_remove_write_access().
3808 lockdep_assert_held(&kvm->slots_lock);
3810 kvm_flush_remote_tlbs(kvm);
3812 mutex_unlock(&kvm->slots_lock);
3816 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3819 if (!irqchip_in_kernel(kvm))
3822 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3823 irq_event->irq, irq_event->level,
3828 long kvm_arch_vm_ioctl(struct file *filp,
3829 unsigned int ioctl, unsigned long arg)
3831 struct kvm *kvm = filp->private_data;
3832 void __user *argp = (void __user *)arg;
3835 * This union makes it completely explicit to gcc-3.x
3836 * that these two variables' stack usage should be
3837 * combined, not added together.
3840 struct kvm_pit_state ps;
3841 struct kvm_pit_state2 ps2;
3842 struct kvm_pit_config pit_config;
3846 case KVM_SET_TSS_ADDR:
3847 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3849 case KVM_SET_IDENTITY_MAP_ADDR: {
3853 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3855 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3858 case KVM_SET_NR_MMU_PAGES:
3859 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3861 case KVM_GET_NR_MMU_PAGES:
3862 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3864 case KVM_CREATE_IRQCHIP: {
3865 struct kvm_pic *vpic;
3867 mutex_lock(&kvm->lock);
3870 goto create_irqchip_unlock;
3872 if (atomic_read(&kvm->online_vcpus))
3873 goto create_irqchip_unlock;
3875 vpic = kvm_create_pic(kvm);
3877 r = kvm_ioapic_init(kvm);
3879 mutex_lock(&kvm->slots_lock);
3880 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3882 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3884 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3886 mutex_unlock(&kvm->slots_lock);
3888 goto create_irqchip_unlock;
3891 goto create_irqchip_unlock;
3893 kvm->arch.vpic = vpic;
3895 r = kvm_setup_default_irq_routing(kvm);
3897 mutex_lock(&kvm->slots_lock);
3898 mutex_lock(&kvm->irq_lock);
3899 kvm_ioapic_destroy(kvm);
3900 kvm_destroy_pic(kvm);
3901 mutex_unlock(&kvm->irq_lock);
3902 mutex_unlock(&kvm->slots_lock);
3904 create_irqchip_unlock:
3905 mutex_unlock(&kvm->lock);
3908 case KVM_CREATE_PIT:
3909 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3911 case KVM_CREATE_PIT2:
3913 if (copy_from_user(&u.pit_config, argp,
3914 sizeof(struct kvm_pit_config)))
3917 mutex_lock(&kvm->slots_lock);
3920 goto create_pit_unlock;
3922 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3926 mutex_unlock(&kvm->slots_lock);
3928 case KVM_GET_IRQCHIP: {
3929 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3930 struct kvm_irqchip *chip;
3932 chip = memdup_user(argp, sizeof(*chip));
3939 if (!irqchip_in_kernel(kvm))
3940 goto get_irqchip_out;
3941 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3943 goto get_irqchip_out;
3945 if (copy_to_user(argp, chip, sizeof *chip))
3946 goto get_irqchip_out;
3952 case KVM_SET_IRQCHIP: {
3953 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3954 struct kvm_irqchip *chip;
3956 chip = memdup_user(argp, sizeof(*chip));
3963 if (!irqchip_in_kernel(kvm))
3964 goto set_irqchip_out;
3965 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3967 goto set_irqchip_out;
3975 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3978 if (!kvm->arch.vpit)
3980 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3984 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3991 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3994 if (!kvm->arch.vpit)
3996 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3999 case KVM_GET_PIT2: {
4001 if (!kvm->arch.vpit)
4003 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4007 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4012 case KVM_SET_PIT2: {
4014 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4017 if (!kvm->arch.vpit)
4019 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4022 case KVM_REINJECT_CONTROL: {
4023 struct kvm_reinject_control control;
4025 if (copy_from_user(&control, argp, sizeof(control)))
4027 r = kvm_vm_ioctl_reinject(kvm, &control);
4030 case KVM_XEN_HVM_CONFIG: {
4032 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4033 sizeof(struct kvm_xen_hvm_config)))
4036 if (kvm->arch.xen_hvm_config.flags)
4041 case KVM_SET_CLOCK: {
4042 struct kvm_clock_data user_ns;
4047 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4055 local_irq_disable();
4056 now_ns = get_kernel_ns();
4057 delta = user_ns.clock - now_ns;
4059 kvm->arch.kvmclock_offset = delta;
4060 kvm_gen_update_masterclock(kvm);
4063 case KVM_GET_CLOCK: {
4064 struct kvm_clock_data user_ns;
4067 local_irq_disable();
4068 now_ns = get_kernel_ns();
4069 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
4072 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4075 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4082 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
4088 static void kvm_init_msr_list(void)
4093 /* skip the first msrs in the list. KVM-specific */
4094 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
4095 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4099 * Even MSRs that are valid in the host may not be exposed
4100 * to the guests in some cases. We could work around this
4101 * in VMX with the generic MSR save/load machinery, but it
4102 * is not really worthwhile since it will really only
4103 * happen with nested virtualization.
4105 switch (msrs_to_save[i]) {
4106 case MSR_IA32_BNDCFGS:
4107 if (!kvm_x86_ops->mpx_supported())
4115 msrs_to_save[j] = msrs_to_save[i];
4118 num_msrs_to_save = j;
4121 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4129 if (!(vcpu->arch.apic &&
4130 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4131 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4142 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4149 if (!(vcpu->arch.apic &&
4150 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4152 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4154 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4164 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4165 struct kvm_segment *var, int seg)
4167 kvm_x86_ops->set_segment(vcpu, var, seg);
4170 void kvm_get_segment(struct kvm_vcpu *vcpu,
4171 struct kvm_segment *var, int seg)
4173 kvm_x86_ops->get_segment(vcpu, var, seg);
4176 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4177 struct x86_exception *exception)
4181 BUG_ON(!mmu_is_nested(vcpu));
4183 /* NPT walks are always user-walks */
4184 access |= PFERR_USER_MASK;
4185 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4190 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4191 struct x86_exception *exception)
4193 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4194 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4197 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4198 struct x86_exception *exception)
4200 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4201 access |= PFERR_FETCH_MASK;
4202 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4205 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4206 struct x86_exception *exception)
4208 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4209 access |= PFERR_WRITE_MASK;
4210 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4213 /* uses this to access any guest's mapped memory without checking CPL */
4214 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4215 struct x86_exception *exception)
4217 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4220 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4221 struct kvm_vcpu *vcpu, u32 access,
4222 struct x86_exception *exception)
4225 int r = X86EMUL_CONTINUE;
4228 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4230 unsigned offset = addr & (PAGE_SIZE-1);
4231 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4234 if (gpa == UNMAPPED_GVA)
4235 return X86EMUL_PROPAGATE_FAULT;
4236 ret = kvm_read_guest_page(vcpu->kvm, gpa >> PAGE_SHIFT, data,
4239 r = X86EMUL_IO_NEEDED;
4251 /* used for instruction fetching */
4252 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4253 gva_t addr, void *val, unsigned int bytes,
4254 struct x86_exception *exception)
4256 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4257 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4261 /* Inline kvm_read_guest_virt_helper for speed. */
4262 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4264 if (unlikely(gpa == UNMAPPED_GVA))
4265 return X86EMUL_PROPAGATE_FAULT;
4267 offset = addr & (PAGE_SIZE-1);
4268 if (WARN_ON(offset + bytes > PAGE_SIZE))
4269 bytes = (unsigned)PAGE_SIZE - offset;
4270 ret = kvm_read_guest_page(vcpu->kvm, gpa >> PAGE_SHIFT, val,
4272 if (unlikely(ret < 0))
4273 return X86EMUL_IO_NEEDED;
4275 return X86EMUL_CONTINUE;
4278 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4279 gva_t addr, void *val, unsigned int bytes,
4280 struct x86_exception *exception)
4282 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4283 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4285 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4288 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4290 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4291 gva_t addr, void *val, unsigned int bytes,
4292 struct x86_exception *exception)
4294 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4295 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4298 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4299 gva_t addr, void *val,
4301 struct x86_exception *exception)
4303 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4305 int r = X86EMUL_CONTINUE;
4308 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4311 unsigned offset = addr & (PAGE_SIZE-1);
4312 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4315 if (gpa == UNMAPPED_GVA)
4316 return X86EMUL_PROPAGATE_FAULT;
4317 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
4319 r = X86EMUL_IO_NEEDED;
4330 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4332 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4333 gpa_t *gpa, struct x86_exception *exception,
4336 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4337 | (write ? PFERR_WRITE_MASK : 0);
4339 if (vcpu_match_mmio_gva(vcpu, gva)
4340 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4341 vcpu->arch.access, access)) {
4342 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4343 (gva & (PAGE_SIZE - 1));
4344 trace_vcpu_match_mmio(gva, *gpa, write, false);
4348 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4350 if (*gpa == UNMAPPED_GVA)
4353 /* For APIC access vmexit */
4354 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4357 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4358 trace_vcpu_match_mmio(gva, *gpa, write, true);
4365 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4366 const void *val, int bytes)
4370 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4373 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4377 struct read_write_emulator_ops {
4378 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4380 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4381 void *val, int bytes);
4382 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4383 int bytes, void *val);
4384 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4385 void *val, int bytes);
4389 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4391 if (vcpu->mmio_read_completed) {
4392 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4393 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4394 vcpu->mmio_read_completed = 0;
4401 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4402 void *val, int bytes)
4404 return !kvm_read_guest(vcpu->kvm, gpa, val, bytes);
4407 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4408 void *val, int bytes)
4410 return emulator_write_phys(vcpu, gpa, val, bytes);
4413 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4415 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4416 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4419 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4420 void *val, int bytes)
4422 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4423 return X86EMUL_IO_NEEDED;
4426 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4427 void *val, int bytes)
4429 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4431 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4432 return X86EMUL_CONTINUE;
4435 static const struct read_write_emulator_ops read_emultor = {
4436 .read_write_prepare = read_prepare,
4437 .read_write_emulate = read_emulate,
4438 .read_write_mmio = vcpu_mmio_read,
4439 .read_write_exit_mmio = read_exit_mmio,
4442 static const struct read_write_emulator_ops write_emultor = {
4443 .read_write_emulate = write_emulate,
4444 .read_write_mmio = write_mmio,
4445 .read_write_exit_mmio = write_exit_mmio,
4449 static int emulator_read_write_onepage(unsigned long addr, void *val,
4451 struct x86_exception *exception,
4452 struct kvm_vcpu *vcpu,
4453 const struct read_write_emulator_ops *ops)
4457 bool write = ops->write;
4458 struct kvm_mmio_fragment *frag;
4460 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4463 return X86EMUL_PROPAGATE_FAULT;
4465 /* For APIC access vmexit */
4469 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4470 return X86EMUL_CONTINUE;
4474 * Is this MMIO handled locally?
4476 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4477 if (handled == bytes)
4478 return X86EMUL_CONTINUE;
4484 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4485 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4489 return X86EMUL_CONTINUE;
4492 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4494 void *val, unsigned int bytes,
4495 struct x86_exception *exception,
4496 const struct read_write_emulator_ops *ops)
4498 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4502 if (ops->read_write_prepare &&
4503 ops->read_write_prepare(vcpu, val, bytes))
4504 return X86EMUL_CONTINUE;
4506 vcpu->mmio_nr_fragments = 0;
4508 /* Crossing a page boundary? */
4509 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4512 now = -addr & ~PAGE_MASK;
4513 rc = emulator_read_write_onepage(addr, val, now, exception,
4516 if (rc != X86EMUL_CONTINUE)
4519 if (ctxt->mode != X86EMUL_MODE_PROT64)
4525 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4527 if (rc != X86EMUL_CONTINUE)
4530 if (!vcpu->mmio_nr_fragments)
4533 gpa = vcpu->mmio_fragments[0].gpa;
4535 vcpu->mmio_needed = 1;
4536 vcpu->mmio_cur_fragment = 0;
4538 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4539 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4540 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4541 vcpu->run->mmio.phys_addr = gpa;
4543 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4546 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4550 struct x86_exception *exception)
4552 return emulator_read_write(ctxt, addr, val, bytes,
4553 exception, &read_emultor);
4556 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4560 struct x86_exception *exception)
4562 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4563 exception, &write_emultor);
4566 #define CMPXCHG_TYPE(t, ptr, old, new) \
4567 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4569 #ifdef CONFIG_X86_64
4570 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4572 # define CMPXCHG64(ptr, old, new) \
4573 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4576 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4581 struct x86_exception *exception)
4583 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4589 /* guests cmpxchg8b have to be emulated atomically */
4590 if (bytes > 8 || (bytes & (bytes - 1)))
4593 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4595 if (gpa == UNMAPPED_GVA ||
4596 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4599 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4602 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4603 if (is_error_page(page))
4606 kaddr = kmap_atomic(page);
4607 kaddr += offset_in_page(gpa);
4610 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4613 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4616 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4619 exchanged = CMPXCHG64(kaddr, old, new);
4624 kunmap_atomic(kaddr);
4625 kvm_release_page_dirty(page);
4628 return X86EMUL_CMPXCHG_FAILED;
4630 mark_page_dirty(vcpu->kvm, gpa >> PAGE_SHIFT);
4631 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4633 return X86EMUL_CONTINUE;
4636 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4638 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4641 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4643 /* TODO: String I/O for in kernel device */
4646 if (vcpu->arch.pio.in)
4647 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4648 vcpu->arch.pio.size, pd);
4650 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4651 vcpu->arch.pio.port, vcpu->arch.pio.size,
4656 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4657 unsigned short port, void *val,
4658 unsigned int count, bool in)
4660 vcpu->arch.pio.port = port;
4661 vcpu->arch.pio.in = in;
4662 vcpu->arch.pio.count = count;
4663 vcpu->arch.pio.size = size;
4665 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4666 vcpu->arch.pio.count = 0;
4670 vcpu->run->exit_reason = KVM_EXIT_IO;
4671 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4672 vcpu->run->io.size = size;
4673 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4674 vcpu->run->io.count = count;
4675 vcpu->run->io.port = port;
4680 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4681 int size, unsigned short port, void *val,
4684 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4687 if (vcpu->arch.pio.count)
4690 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4693 memcpy(val, vcpu->arch.pio_data, size * count);
4694 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4695 vcpu->arch.pio.count = 0;
4702 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4703 int size, unsigned short port,
4704 const void *val, unsigned int count)
4706 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4708 memcpy(vcpu->arch.pio_data, val, size * count);
4709 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4710 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4713 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4715 return kvm_x86_ops->get_segment_base(vcpu, seg);
4718 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4720 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4723 int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4725 if (!need_emulate_wbinvd(vcpu))
4726 return X86EMUL_CONTINUE;
4728 if (kvm_x86_ops->has_wbinvd_exit()) {
4729 int cpu = get_cpu();
4731 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4732 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4733 wbinvd_ipi, NULL, 1);
4735 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4738 return X86EMUL_CONTINUE;
4741 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4743 kvm_x86_ops->skip_emulated_instruction(vcpu);
4744 return kvm_emulate_wbinvd_noskip(vcpu);
4746 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4750 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4752 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4755 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4756 unsigned long *dest)
4758 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4761 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4762 unsigned long value)
4765 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4768 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4770 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4773 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4775 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4776 unsigned long value;
4780 value = kvm_read_cr0(vcpu);
4783 value = vcpu->arch.cr2;
4786 value = kvm_read_cr3(vcpu);
4789 value = kvm_read_cr4(vcpu);
4792 value = kvm_get_cr8(vcpu);
4795 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4802 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4804 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4809 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4812 vcpu->arch.cr2 = val;
4815 res = kvm_set_cr3(vcpu, val);
4818 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4821 res = kvm_set_cr8(vcpu, val);
4824 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4831 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4833 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4836 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4838 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4841 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4843 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4846 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4848 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4851 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4853 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4856 static unsigned long emulator_get_cached_segment_base(
4857 struct x86_emulate_ctxt *ctxt, int seg)
4859 return get_segment_base(emul_to_vcpu(ctxt), seg);
4862 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4863 struct desc_struct *desc, u32 *base3,
4866 struct kvm_segment var;
4868 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4869 *selector = var.selector;
4872 memset(desc, 0, sizeof(*desc));
4878 set_desc_limit(desc, var.limit);
4879 set_desc_base(desc, (unsigned long)var.base);
4880 #ifdef CONFIG_X86_64
4882 *base3 = var.base >> 32;
4884 desc->type = var.type;
4886 desc->dpl = var.dpl;
4887 desc->p = var.present;
4888 desc->avl = var.avl;
4896 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4897 struct desc_struct *desc, u32 base3,
4900 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4901 struct kvm_segment var;
4903 var.selector = selector;
4904 var.base = get_desc_base(desc);
4905 #ifdef CONFIG_X86_64
4906 var.base |= ((u64)base3) << 32;
4908 var.limit = get_desc_limit(desc);
4910 var.limit = (var.limit << 12) | 0xfff;
4911 var.type = desc->type;
4912 var.dpl = desc->dpl;
4917 var.avl = desc->avl;
4918 var.present = desc->p;
4919 var.unusable = !var.present;
4922 kvm_set_segment(vcpu, &var, seg);
4926 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4927 u32 msr_index, u64 *pdata)
4929 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4932 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4933 u32 msr_index, u64 data)
4935 struct msr_data msr;
4938 msr.index = msr_index;
4939 msr.host_initiated = false;
4940 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4943 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
4946 return kvm_pmu_check_pmc(emul_to_vcpu(ctxt), pmc);
4949 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4950 u32 pmc, u64 *pdata)
4952 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4955 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4957 emul_to_vcpu(ctxt)->arch.halt_request = 1;
4960 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4963 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4965 * CR0.TS may reference the host fpu state, not the guest fpu state,
4966 * so it may be clear at this point.
4971 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4976 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4977 struct x86_instruction_info *info,
4978 enum x86_intercept_stage stage)
4980 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4983 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
4984 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
4986 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
4989 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
4991 return kvm_register_read(emul_to_vcpu(ctxt), reg);
4994 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
4996 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
4999 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5001 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5004 static const struct x86_emulate_ops emulate_ops = {
5005 .read_gpr = emulator_read_gpr,
5006 .write_gpr = emulator_write_gpr,
5007 .read_std = kvm_read_guest_virt_system,
5008 .write_std = kvm_write_guest_virt_system,
5009 .fetch = kvm_fetch_guest_virt,
5010 .read_emulated = emulator_read_emulated,
5011 .write_emulated = emulator_write_emulated,
5012 .cmpxchg_emulated = emulator_cmpxchg_emulated,
5013 .invlpg = emulator_invlpg,
5014 .pio_in_emulated = emulator_pio_in_emulated,
5015 .pio_out_emulated = emulator_pio_out_emulated,
5016 .get_segment = emulator_get_segment,
5017 .set_segment = emulator_set_segment,
5018 .get_cached_segment_base = emulator_get_cached_segment_base,
5019 .get_gdt = emulator_get_gdt,
5020 .get_idt = emulator_get_idt,
5021 .set_gdt = emulator_set_gdt,
5022 .set_idt = emulator_set_idt,
5023 .get_cr = emulator_get_cr,
5024 .set_cr = emulator_set_cr,
5025 .cpl = emulator_get_cpl,
5026 .get_dr = emulator_get_dr,
5027 .set_dr = emulator_set_dr,
5028 .set_msr = emulator_set_msr,
5029 .get_msr = emulator_get_msr,
5030 .check_pmc = emulator_check_pmc,
5031 .read_pmc = emulator_read_pmc,
5032 .halt = emulator_halt,
5033 .wbinvd = emulator_wbinvd,
5034 .fix_hypercall = emulator_fix_hypercall,
5035 .get_fpu = emulator_get_fpu,
5036 .put_fpu = emulator_put_fpu,
5037 .intercept = emulator_intercept,
5038 .get_cpuid = emulator_get_cpuid,
5039 .set_nmi_mask = emulator_set_nmi_mask,
5042 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5044 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5046 * an sti; sti; sequence only disable interrupts for the first
5047 * instruction. So, if the last instruction, be it emulated or
5048 * not, left the system with the INT_STI flag enabled, it
5049 * means that the last instruction is an sti. We should not
5050 * leave the flag on in this case. The same goes for mov ss
5052 if (int_shadow & mask)
5054 if (unlikely(int_shadow || mask)) {
5055 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5057 kvm_make_request(KVM_REQ_EVENT, vcpu);
5061 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5063 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5064 if (ctxt->exception.vector == PF_VECTOR)
5065 return kvm_propagate_fault(vcpu, &ctxt->exception);
5067 if (ctxt->exception.error_code_valid)
5068 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5069 ctxt->exception.error_code);
5071 kvm_queue_exception(vcpu, ctxt->exception.vector);
5075 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5077 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5080 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5082 ctxt->eflags = kvm_get_rflags(vcpu);
5083 ctxt->eip = kvm_rip_read(vcpu);
5084 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5085 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
5086 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
5087 cs_db ? X86EMUL_MODE_PROT32 :
5088 X86EMUL_MODE_PROT16;
5089 ctxt->guest_mode = is_guest_mode(vcpu);
5091 init_decode_cache(ctxt);
5092 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5095 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5097 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5100 init_emulate_ctxt(vcpu);
5104 ctxt->_eip = ctxt->eip + inc_eip;
5105 ret = emulate_int_real(ctxt, irq);
5107 if (ret != X86EMUL_CONTINUE)
5108 return EMULATE_FAIL;
5110 ctxt->eip = ctxt->_eip;
5111 kvm_rip_write(vcpu, ctxt->eip);
5112 kvm_set_rflags(vcpu, ctxt->eflags);
5114 if (irq == NMI_VECTOR)
5115 vcpu->arch.nmi_pending = 0;
5117 vcpu->arch.interrupt.pending = false;
5119 return EMULATE_DONE;
5121 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5123 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5125 int r = EMULATE_DONE;
5127 ++vcpu->stat.insn_emulation_fail;
5128 trace_kvm_emulate_insn_failed(vcpu);
5129 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5130 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5131 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5132 vcpu->run->internal.ndata = 0;
5135 kvm_queue_exception(vcpu, UD_VECTOR);
5140 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5141 bool write_fault_to_shadow_pgtable,
5147 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5150 if (!vcpu->arch.mmu.direct_map) {
5152 * Write permission should be allowed since only
5153 * write access need to be emulated.
5155 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5158 * If the mapping is invalid in guest, let cpu retry
5159 * it to generate fault.
5161 if (gpa == UNMAPPED_GVA)
5166 * Do not retry the unhandleable instruction if it faults on the
5167 * readonly host memory, otherwise it will goto a infinite loop:
5168 * retry instruction -> write #PF -> emulation fail -> retry
5169 * instruction -> ...
5171 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5174 * If the instruction failed on the error pfn, it can not be fixed,
5175 * report the error to userspace.
5177 if (is_error_noslot_pfn(pfn))
5180 kvm_release_pfn_clean(pfn);
5182 /* The instructions are well-emulated on direct mmu. */
5183 if (vcpu->arch.mmu.direct_map) {
5184 unsigned int indirect_shadow_pages;
5186 spin_lock(&vcpu->kvm->mmu_lock);
5187 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5188 spin_unlock(&vcpu->kvm->mmu_lock);
5190 if (indirect_shadow_pages)
5191 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5197 * if emulation was due to access to shadowed page table
5198 * and it failed try to unshadow page and re-enter the
5199 * guest to let CPU execute the instruction.
5201 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5204 * If the access faults on its page table, it can not
5205 * be fixed by unprotecting shadow page and it should
5206 * be reported to userspace.
5208 return !write_fault_to_shadow_pgtable;
5211 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5212 unsigned long cr2, int emulation_type)
5214 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5215 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5217 last_retry_eip = vcpu->arch.last_retry_eip;
5218 last_retry_addr = vcpu->arch.last_retry_addr;
5221 * If the emulation is caused by #PF and it is non-page_table
5222 * writing instruction, it means the VM-EXIT is caused by shadow
5223 * page protected, we can zap the shadow page and retry this
5224 * instruction directly.
5226 * Note: if the guest uses a non-page-table modifying instruction
5227 * on the PDE that points to the instruction, then we will unmap
5228 * the instruction and go to an infinite loop. So, we cache the
5229 * last retried eip and the last fault address, if we meet the eip
5230 * and the address again, we can break out of the potential infinite
5233 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5235 if (!(emulation_type & EMULTYPE_RETRY))
5238 if (x86_page_table_writing_insn(ctxt))
5241 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5244 vcpu->arch.last_retry_eip = ctxt->eip;
5245 vcpu->arch.last_retry_addr = cr2;
5247 if (!vcpu->arch.mmu.direct_map)
5248 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5250 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5255 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5256 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5258 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5267 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5268 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5273 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5275 struct kvm_run *kvm_run = vcpu->run;
5278 * rflags is the old, "raw" value of the flags. The new value has
5279 * not been saved yet.
5281 * This is correct even for TF set by the guest, because "the
5282 * processor will not generate this exception after the instruction
5283 * that sets the TF flag".
5285 if (unlikely(rflags & X86_EFLAGS_TF)) {
5286 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5287 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5289 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5290 kvm_run->debug.arch.exception = DB_VECTOR;
5291 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5292 *r = EMULATE_USER_EXIT;
5294 vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5296 * "Certain debug exceptions may clear bit 0-3. The
5297 * remaining contents of the DR6 register are never
5298 * cleared by the processor".
5300 vcpu->arch.dr6 &= ~15;
5301 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5302 kvm_queue_exception(vcpu, DB_VECTOR);
5307 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5309 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5310 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5311 struct kvm_run *kvm_run = vcpu->run;
5312 unsigned long eip = kvm_get_linear_rip(vcpu);
5313 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5314 vcpu->arch.guest_debug_dr7,
5318 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5319 kvm_run->debug.arch.pc = eip;
5320 kvm_run->debug.arch.exception = DB_VECTOR;
5321 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5322 *r = EMULATE_USER_EXIT;
5327 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5328 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5329 unsigned long eip = kvm_get_linear_rip(vcpu);
5330 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5335 vcpu->arch.dr6 &= ~15;
5336 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5337 kvm_queue_exception(vcpu, DB_VECTOR);
5346 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5353 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5354 bool writeback = true;
5355 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5358 * Clear write_fault_to_shadow_pgtable here to ensure it is
5361 vcpu->arch.write_fault_to_shadow_pgtable = false;
5362 kvm_clear_exception_queue(vcpu);
5364 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5365 init_emulate_ctxt(vcpu);
5368 * We will reenter on the same instruction since
5369 * we do not set complete_userspace_io. This does not
5370 * handle watchpoints yet, those would be handled in
5373 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5376 ctxt->interruptibility = 0;
5377 ctxt->have_exception = false;
5378 ctxt->exception.vector = -1;
5379 ctxt->perm_ok = false;
5381 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5383 r = x86_decode_insn(ctxt, insn, insn_len);
5385 trace_kvm_emulate_insn_start(vcpu);
5386 ++vcpu->stat.insn_emulation;
5387 if (r != EMULATION_OK) {
5388 if (emulation_type & EMULTYPE_TRAP_UD)
5389 return EMULATE_FAIL;
5390 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5392 return EMULATE_DONE;
5393 if (emulation_type & EMULTYPE_SKIP)
5394 return EMULATE_FAIL;
5395 return handle_emulation_failure(vcpu);
5399 if (emulation_type & EMULTYPE_SKIP) {
5400 kvm_rip_write(vcpu, ctxt->_eip);
5401 if (ctxt->eflags & X86_EFLAGS_RF)
5402 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5403 return EMULATE_DONE;
5406 if (retry_instruction(ctxt, cr2, emulation_type))
5407 return EMULATE_DONE;
5409 /* this is needed for vmware backdoor interface to work since it
5410 changes registers values during IO operation */
5411 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5412 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5413 emulator_invalidate_register_cache(ctxt);
5417 r = x86_emulate_insn(ctxt);
5419 if (r == EMULATION_INTERCEPTED)
5420 return EMULATE_DONE;
5422 if (r == EMULATION_FAILED) {
5423 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5425 return EMULATE_DONE;
5427 return handle_emulation_failure(vcpu);
5430 if (ctxt->have_exception) {
5432 if (inject_emulated_exception(vcpu))
5434 } else if (vcpu->arch.pio.count) {
5435 if (!vcpu->arch.pio.in) {
5436 /* FIXME: return into emulator if single-stepping. */
5437 vcpu->arch.pio.count = 0;
5440 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5442 r = EMULATE_USER_EXIT;
5443 } else if (vcpu->mmio_needed) {
5444 if (!vcpu->mmio_is_write)
5446 r = EMULATE_USER_EXIT;
5447 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5448 } else if (r == EMULATION_RESTART)
5454 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5455 toggle_interruptibility(vcpu, ctxt->interruptibility);
5456 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5457 kvm_rip_write(vcpu, ctxt->eip);
5458 if (r == EMULATE_DONE)
5459 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5460 if (!ctxt->have_exception ||
5461 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5462 __kvm_set_rflags(vcpu, ctxt->eflags);
5465 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5466 * do nothing, and it will be requested again as soon as
5467 * the shadow expires. But we still need to check here,
5468 * because POPF has no interrupt shadow.
5470 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5471 kvm_make_request(KVM_REQ_EVENT, vcpu);
5473 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5477 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5479 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5481 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5482 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5483 size, port, &val, 1);
5484 /* do not return to emulator after return from userspace */
5485 vcpu->arch.pio.count = 0;
5488 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5490 static void tsc_bad(void *info)
5492 __this_cpu_write(cpu_tsc_khz, 0);
5495 static void tsc_khz_changed(void *data)
5497 struct cpufreq_freqs *freq = data;
5498 unsigned long khz = 0;
5502 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5503 khz = cpufreq_quick_get(raw_smp_processor_id());
5506 __this_cpu_write(cpu_tsc_khz, khz);
5509 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5512 struct cpufreq_freqs *freq = data;
5514 struct kvm_vcpu *vcpu;
5515 int i, send_ipi = 0;
5518 * We allow guests to temporarily run on slowing clocks,
5519 * provided we notify them after, or to run on accelerating
5520 * clocks, provided we notify them before. Thus time never
5523 * However, we have a problem. We can't atomically update
5524 * the frequency of a given CPU from this function; it is
5525 * merely a notifier, which can be called from any CPU.
5526 * Changing the TSC frequency at arbitrary points in time
5527 * requires a recomputation of local variables related to
5528 * the TSC for each VCPU. We must flag these local variables
5529 * to be updated and be sure the update takes place with the
5530 * new frequency before any guests proceed.
5532 * Unfortunately, the combination of hotplug CPU and frequency
5533 * change creates an intractable locking scenario; the order
5534 * of when these callouts happen is undefined with respect to
5535 * CPU hotplug, and they can race with each other. As such,
5536 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5537 * undefined; you can actually have a CPU frequency change take
5538 * place in between the computation of X and the setting of the
5539 * variable. To protect against this problem, all updates of
5540 * the per_cpu tsc_khz variable are done in an interrupt
5541 * protected IPI, and all callers wishing to update the value
5542 * must wait for a synchronous IPI to complete (which is trivial
5543 * if the caller is on the CPU already). This establishes the
5544 * necessary total order on variable updates.
5546 * Note that because a guest time update may take place
5547 * anytime after the setting of the VCPU's request bit, the
5548 * correct TSC value must be set before the request. However,
5549 * to ensure the update actually makes it to any guest which
5550 * starts running in hardware virtualization between the set
5551 * and the acquisition of the spinlock, we must also ping the
5552 * CPU after setting the request bit.
5556 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5558 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5561 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5563 spin_lock(&kvm_lock);
5564 list_for_each_entry(kvm, &vm_list, vm_list) {
5565 kvm_for_each_vcpu(i, vcpu, kvm) {
5566 if (vcpu->cpu != freq->cpu)
5568 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5569 if (vcpu->cpu != smp_processor_id())
5573 spin_unlock(&kvm_lock);
5575 if (freq->old < freq->new && send_ipi) {
5577 * We upscale the frequency. Must make the guest
5578 * doesn't see old kvmclock values while running with
5579 * the new frequency, otherwise we risk the guest sees
5580 * time go backwards.
5582 * In case we update the frequency for another cpu
5583 * (which might be in guest context) send an interrupt
5584 * to kick the cpu out of guest context. Next time
5585 * guest context is entered kvmclock will be updated,
5586 * so the guest will not see stale values.
5588 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5593 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5594 .notifier_call = kvmclock_cpufreq_notifier
5597 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5598 unsigned long action, void *hcpu)
5600 unsigned int cpu = (unsigned long)hcpu;
5604 case CPU_DOWN_FAILED:
5605 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5607 case CPU_DOWN_PREPARE:
5608 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5614 static struct notifier_block kvmclock_cpu_notifier_block = {
5615 .notifier_call = kvmclock_cpu_notifier,
5616 .priority = -INT_MAX
5619 static void kvm_timer_init(void)
5623 max_tsc_khz = tsc_khz;
5625 cpu_notifier_register_begin();
5626 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5627 #ifdef CONFIG_CPU_FREQ
5628 struct cpufreq_policy policy;
5629 memset(&policy, 0, sizeof(policy));
5631 cpufreq_get_policy(&policy, cpu);
5632 if (policy.cpuinfo.max_freq)
5633 max_tsc_khz = policy.cpuinfo.max_freq;
5636 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5637 CPUFREQ_TRANSITION_NOTIFIER);
5639 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5640 for_each_online_cpu(cpu)
5641 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5643 __register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5644 cpu_notifier_register_done();
5648 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5650 int kvm_is_in_guest(void)
5652 return __this_cpu_read(current_vcpu) != NULL;
5655 static int kvm_is_user_mode(void)
5659 if (__this_cpu_read(current_vcpu))
5660 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5662 return user_mode != 0;
5665 static unsigned long kvm_get_guest_ip(void)
5667 unsigned long ip = 0;
5669 if (__this_cpu_read(current_vcpu))
5670 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5675 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5676 .is_in_guest = kvm_is_in_guest,
5677 .is_user_mode = kvm_is_user_mode,
5678 .get_guest_ip = kvm_get_guest_ip,
5681 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5683 __this_cpu_write(current_vcpu, vcpu);
5685 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5687 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5689 __this_cpu_write(current_vcpu, NULL);
5691 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5693 static void kvm_set_mmio_spte_mask(void)
5696 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5699 * Set the reserved bits and the present bit of an paging-structure
5700 * entry to generate page fault with PFER.RSV = 1.
5702 /* Mask the reserved physical address bits. */
5703 mask = rsvd_bits(maxphyaddr, 51);
5705 /* Bit 62 is always reserved for 32bit host. */
5706 mask |= 0x3ull << 62;
5708 /* Set the present bit. */
5711 #ifdef CONFIG_X86_64
5713 * If reserved bit is not supported, clear the present bit to disable
5716 if (maxphyaddr == 52)
5720 kvm_mmu_set_mmio_spte_mask(mask);
5723 #ifdef CONFIG_X86_64
5724 static void pvclock_gtod_update_fn(struct work_struct *work)
5728 struct kvm_vcpu *vcpu;
5731 spin_lock(&kvm_lock);
5732 list_for_each_entry(kvm, &vm_list, vm_list)
5733 kvm_for_each_vcpu(i, vcpu, kvm)
5734 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5735 atomic_set(&kvm_guest_has_master_clock, 0);
5736 spin_unlock(&kvm_lock);
5739 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5742 * Notification about pvclock gtod data update.
5744 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5747 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5748 struct timekeeper *tk = priv;
5750 update_pvclock_gtod(tk);
5752 /* disable master clock if host does not trust, or does not
5753 * use, TSC clocksource
5755 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5756 atomic_read(&kvm_guest_has_master_clock) != 0)
5757 queue_work(system_long_wq, &pvclock_gtod_work);
5762 static struct notifier_block pvclock_gtod_notifier = {
5763 .notifier_call = pvclock_gtod_notify,
5767 int kvm_arch_init(void *opaque)
5770 struct kvm_x86_ops *ops = opaque;
5773 printk(KERN_ERR "kvm: already loaded the other module\n");
5778 if (!ops->cpu_has_kvm_support()) {
5779 printk(KERN_ERR "kvm: no hardware support\n");
5783 if (ops->disabled_by_bios()) {
5784 printk(KERN_ERR "kvm: disabled by bios\n");
5790 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5792 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5796 r = kvm_mmu_module_init();
5798 goto out_free_percpu;
5800 kvm_set_mmio_spte_mask();
5803 kvm_init_msr_list();
5805 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5806 PT_DIRTY_MASK, PT64_NX_MASK, 0);
5810 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5813 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5816 #ifdef CONFIG_X86_64
5817 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5823 free_percpu(shared_msrs);
5828 void kvm_arch_exit(void)
5830 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5832 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5833 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5834 CPUFREQ_TRANSITION_NOTIFIER);
5835 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5836 #ifdef CONFIG_X86_64
5837 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5840 kvm_mmu_module_exit();
5841 free_percpu(shared_msrs);
5844 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
5846 ++vcpu->stat.halt_exits;
5847 if (irqchip_in_kernel(vcpu->kvm)) {
5848 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5851 vcpu->run->exit_reason = KVM_EXIT_HLT;
5855 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5857 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5859 kvm_x86_ops->skip_emulated_instruction(vcpu);
5860 return kvm_vcpu_halt(vcpu);
5862 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5864 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5866 u64 param, ingpa, outgpa, ret;
5867 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
5868 bool fast, longmode;
5871 * hypercall generates UD from non zero cpl and real mode
5874 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
5875 kvm_queue_exception(vcpu, UD_VECTOR);
5879 longmode = is_64_bit_mode(vcpu);
5882 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
5883 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
5884 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
5885 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
5886 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
5887 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
5889 #ifdef CONFIG_X86_64
5891 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
5892 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
5893 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
5897 code = param & 0xffff;
5898 fast = (param >> 16) & 0x1;
5899 rep_cnt = (param >> 32) & 0xfff;
5900 rep_idx = (param >> 48) & 0xfff;
5902 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5905 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5906 kvm_vcpu_on_spin(vcpu);
5909 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5913 ret = res | (((u64)rep_done & 0xfff) << 32);
5915 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5917 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5918 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5925 * kvm_pv_kick_cpu_op: Kick a vcpu.
5927 * @apicid - apicid of vcpu to be kicked.
5929 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5931 struct kvm_lapic_irq lapic_irq;
5933 lapic_irq.shorthand = 0;
5934 lapic_irq.dest_mode = 0;
5935 lapic_irq.dest_id = apicid;
5937 lapic_irq.delivery_mode = APIC_DM_REMRD;
5938 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
5941 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5943 unsigned long nr, a0, a1, a2, a3, ret;
5944 int op_64_bit, r = 1;
5946 kvm_x86_ops->skip_emulated_instruction(vcpu);
5948 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5949 return kvm_hv_hypercall(vcpu);
5951 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5952 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5953 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5954 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5955 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5957 trace_kvm_hypercall(nr, a0, a1, a2, a3);
5959 op_64_bit = is_64_bit_mode(vcpu);
5968 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5974 case KVM_HC_VAPIC_POLL_IRQ:
5977 case KVM_HC_KICK_CPU:
5978 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
5988 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5989 ++vcpu->stat.hypercalls;
5992 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5994 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5996 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5997 char instruction[3];
5998 unsigned long rip = kvm_rip_read(vcpu);
6000 kvm_x86_ops->patch_hypercall(vcpu, instruction);
6002 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
6006 * Check if userspace requested an interrupt window, and that the
6007 * interrupt window is open.
6009 * No need to exit to userspace if we already have an interrupt queued.
6011 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6013 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
6014 vcpu->run->request_interrupt_window &&
6015 kvm_arch_interrupt_allowed(vcpu));
6018 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6020 struct kvm_run *kvm_run = vcpu->run;
6022 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6023 kvm_run->cr8 = kvm_get_cr8(vcpu);
6024 kvm_run->apic_base = kvm_get_apic_base(vcpu);
6025 if (irqchip_in_kernel(vcpu->kvm))
6026 kvm_run->ready_for_interrupt_injection = 1;
6028 kvm_run->ready_for_interrupt_injection =
6029 kvm_arch_interrupt_allowed(vcpu) &&
6030 !kvm_cpu_has_interrupt(vcpu) &&
6031 !kvm_event_needs_reinjection(vcpu);
6034 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6038 if (!kvm_x86_ops->update_cr8_intercept)
6041 if (!vcpu->arch.apic)
6044 if (!vcpu->arch.apic->vapic_addr)
6045 max_irr = kvm_lapic_find_highest_irr(vcpu);
6052 tpr = kvm_lapic_get_cr8(vcpu);
6054 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6057 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6061 /* try to reinject previous events if any */
6062 if (vcpu->arch.exception.pending) {
6063 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6064 vcpu->arch.exception.has_error_code,
6065 vcpu->arch.exception.error_code);
6067 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6068 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6071 if (vcpu->arch.exception.nr == DB_VECTOR &&
6072 (vcpu->arch.dr7 & DR7_GD)) {
6073 vcpu->arch.dr7 &= ~DR7_GD;
6074 kvm_update_dr7(vcpu);
6077 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6078 vcpu->arch.exception.has_error_code,
6079 vcpu->arch.exception.error_code,
6080 vcpu->arch.exception.reinject);
6084 if (vcpu->arch.nmi_injected) {
6085 kvm_x86_ops->set_nmi(vcpu);
6089 if (vcpu->arch.interrupt.pending) {
6090 kvm_x86_ops->set_irq(vcpu);
6094 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6095 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6100 /* try to inject new event if pending */
6101 if (vcpu->arch.nmi_pending) {
6102 if (kvm_x86_ops->nmi_allowed(vcpu)) {
6103 --vcpu->arch.nmi_pending;
6104 vcpu->arch.nmi_injected = true;
6105 kvm_x86_ops->set_nmi(vcpu);
6107 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6109 * Because interrupts can be injected asynchronously, we are
6110 * calling check_nested_events again here to avoid a race condition.
6111 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6112 * proposal and current concerns. Perhaps we should be setting
6113 * KVM_REQ_EVENT only on certain events and not unconditionally?
6115 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6116 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6120 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6121 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6123 kvm_x86_ops->set_irq(vcpu);
6129 static void process_nmi(struct kvm_vcpu *vcpu)
6134 * x86 is limited to one NMI running, and one NMI pending after it.
6135 * If an NMI is already in progress, limit further NMIs to just one.
6136 * Otherwise, allow two (and we'll inject the first one immediately).
6138 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6141 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6142 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6143 kvm_make_request(KVM_REQ_EVENT, vcpu);
6146 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6148 u64 eoi_exit_bitmap[4];
6151 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6154 memset(eoi_exit_bitmap, 0, 32);
6157 kvm_ioapic_scan_entry(vcpu, eoi_exit_bitmap, tmr);
6158 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6159 kvm_apic_update_tmr(vcpu, tmr);
6162 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6164 ++vcpu->stat.tlb_flush;
6165 kvm_x86_ops->tlb_flush(vcpu);
6168 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6170 struct page *page = NULL;
6172 if (!irqchip_in_kernel(vcpu->kvm))
6175 if (!kvm_x86_ops->set_apic_access_page_addr)
6178 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6179 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6182 * Do not pin apic access page in memory, the MMU notifier
6183 * will call us again if it is migrated or swapped out.
6187 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6189 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6190 unsigned long address)
6193 * The physical address of apic access page is stored in the VMCS.
6194 * Update it when it becomes invalid.
6196 if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6197 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6201 * Returns 1 to let vcpu_run() continue the guest execution loop without
6202 * exiting to the userspace. Otherwise, the value will be returned to the
6205 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6208 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
6209 vcpu->run->request_interrupt_window;
6210 bool req_immediate_exit = false;
6212 if (vcpu->requests) {
6213 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6214 kvm_mmu_unload(vcpu);
6215 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6216 __kvm_migrate_timers(vcpu);
6217 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6218 kvm_gen_update_masterclock(vcpu->kvm);
6219 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6220 kvm_gen_kvmclock_update(vcpu);
6221 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6222 r = kvm_guest_time_update(vcpu);
6226 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6227 kvm_mmu_sync_roots(vcpu);
6228 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6229 kvm_vcpu_flush_tlb(vcpu);
6230 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6231 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6235 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6236 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6240 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6241 vcpu->fpu_active = 0;
6242 kvm_x86_ops->fpu_deactivate(vcpu);
6244 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6245 /* Page is swapped out. Do synthetic halt */
6246 vcpu->arch.apf.halted = true;
6250 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6251 record_steal_time(vcpu);
6252 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6254 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6255 kvm_handle_pmu_event(vcpu);
6256 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6257 kvm_deliver_pmi(vcpu);
6258 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6259 vcpu_scan_ioapic(vcpu);
6260 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6261 kvm_vcpu_reload_apic_access_page(vcpu);
6264 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6265 kvm_apic_accept_events(vcpu);
6266 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6271 if (inject_pending_event(vcpu, req_int_win) != 0)
6272 req_immediate_exit = true;
6273 /* enable NMI/IRQ window open exits if needed */
6274 else if (vcpu->arch.nmi_pending)
6275 kvm_x86_ops->enable_nmi_window(vcpu);
6276 else if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6277 kvm_x86_ops->enable_irq_window(vcpu);
6279 if (kvm_lapic_enabled(vcpu)) {
6281 * Update architecture specific hints for APIC
6282 * virtual interrupt delivery.
6284 if (kvm_x86_ops->hwapic_irr_update)
6285 kvm_x86_ops->hwapic_irr_update(vcpu,
6286 kvm_lapic_find_highest_irr(vcpu));
6287 update_cr8_intercept(vcpu);
6288 kvm_lapic_sync_to_vapic(vcpu);
6292 r = kvm_mmu_reload(vcpu);
6294 goto cancel_injection;
6299 kvm_x86_ops->prepare_guest_switch(vcpu);
6300 if (vcpu->fpu_active)
6301 kvm_load_guest_fpu(vcpu);
6302 kvm_load_guest_xcr0(vcpu);
6304 vcpu->mode = IN_GUEST_MODE;
6306 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6308 /* We should set ->mode before check ->requests,
6309 * see the comment in make_all_cpus_request.
6311 smp_mb__after_srcu_read_unlock();
6313 local_irq_disable();
6315 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6316 || need_resched() || signal_pending(current)) {
6317 vcpu->mode = OUTSIDE_GUEST_MODE;
6321 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6323 goto cancel_injection;
6326 if (req_immediate_exit)
6327 smp_send_reschedule(vcpu->cpu);
6331 if (unlikely(vcpu->arch.switch_db_regs)) {
6333 set_debugreg(vcpu->arch.eff_db[0], 0);
6334 set_debugreg(vcpu->arch.eff_db[1], 1);
6335 set_debugreg(vcpu->arch.eff_db[2], 2);
6336 set_debugreg(vcpu->arch.eff_db[3], 3);
6337 set_debugreg(vcpu->arch.dr6, 6);
6338 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6341 trace_kvm_entry(vcpu->vcpu_id);
6342 wait_lapic_expire(vcpu);
6343 kvm_x86_ops->run(vcpu);
6346 * Do this here before restoring debug registers on the host. And
6347 * since we do this before handling the vmexit, a DR access vmexit
6348 * can (a) read the correct value of the debug registers, (b) set
6349 * KVM_DEBUGREG_WONT_EXIT again.
6351 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6354 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6355 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6356 for (i = 0; i < KVM_NR_DB_REGS; i++)
6357 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6361 * If the guest has used debug registers, at least dr7
6362 * will be disabled while returning to the host.
6363 * If we don't have active breakpoints in the host, we don't
6364 * care about the messed up debug address registers. But if
6365 * we have some of them active, restore the old state.
6367 if (hw_breakpoint_active())
6368 hw_breakpoint_restore();
6370 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu,
6373 vcpu->mode = OUTSIDE_GUEST_MODE;
6376 /* Interrupt is enabled by handle_external_intr() */
6377 kvm_x86_ops->handle_external_intr(vcpu);
6382 * We must have an instruction between local_irq_enable() and
6383 * kvm_guest_exit(), so the timer interrupt isn't delayed by
6384 * the interrupt shadow. The stat.exits increment will do nicely.
6385 * But we need to prevent reordering, hence this barrier():
6393 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6396 * Profile KVM exit RIPs:
6398 if (unlikely(prof_on == KVM_PROFILING)) {
6399 unsigned long rip = kvm_rip_read(vcpu);
6400 profile_hit(KVM_PROFILING, (void *)rip);
6403 if (unlikely(vcpu->arch.tsc_always_catchup))
6404 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6406 if (vcpu->arch.apic_attention)
6407 kvm_lapic_sync_from_vapic(vcpu);
6409 r = kvm_x86_ops->handle_exit(vcpu);
6413 kvm_x86_ops->cancel_injection(vcpu);
6414 if (unlikely(vcpu->arch.apic_attention))
6415 kvm_lapic_sync_from_vapic(vcpu);
6420 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6422 if (!kvm_arch_vcpu_runnable(vcpu)) {
6423 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6424 kvm_vcpu_block(vcpu);
6425 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6426 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6430 kvm_apic_accept_events(vcpu);
6431 switch(vcpu->arch.mp_state) {
6432 case KVM_MP_STATE_HALTED:
6433 vcpu->arch.pv.pv_unhalted = false;
6434 vcpu->arch.mp_state =
6435 KVM_MP_STATE_RUNNABLE;
6436 case KVM_MP_STATE_RUNNABLE:
6437 vcpu->arch.apf.halted = false;
6439 case KVM_MP_STATE_INIT_RECEIVED:
6448 static int vcpu_run(struct kvm_vcpu *vcpu)
6451 struct kvm *kvm = vcpu->kvm;
6453 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6456 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6457 !vcpu->arch.apf.halted)
6458 r = vcpu_enter_guest(vcpu);
6460 r = vcpu_block(kvm, vcpu);
6464 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6465 if (kvm_cpu_has_pending_timer(vcpu))
6466 kvm_inject_pending_timer_irqs(vcpu);
6468 if (dm_request_for_irq_injection(vcpu)) {
6470 vcpu->run->exit_reason = KVM_EXIT_INTR;
6471 ++vcpu->stat.request_irq_exits;
6475 kvm_check_async_pf_completion(vcpu);
6477 if (signal_pending(current)) {
6479 vcpu->run->exit_reason = KVM_EXIT_INTR;
6480 ++vcpu->stat.signal_exits;
6483 if (need_resched()) {
6484 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6486 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6490 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6495 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6498 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6499 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6500 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6501 if (r != EMULATE_DONE)
6506 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6508 BUG_ON(!vcpu->arch.pio.count);
6510 return complete_emulated_io(vcpu);
6514 * Implements the following, as a state machine:
6518 * for each mmio piece in the fragment
6526 * for each mmio piece in the fragment
6531 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6533 struct kvm_run *run = vcpu->run;
6534 struct kvm_mmio_fragment *frag;
6537 BUG_ON(!vcpu->mmio_needed);
6539 /* Complete previous fragment */
6540 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6541 len = min(8u, frag->len);
6542 if (!vcpu->mmio_is_write)
6543 memcpy(frag->data, run->mmio.data, len);
6545 if (frag->len <= 8) {
6546 /* Switch to the next fragment. */
6548 vcpu->mmio_cur_fragment++;
6550 /* Go forward to the next mmio piece. */
6556 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6557 vcpu->mmio_needed = 0;
6559 /* FIXME: return into emulator if single-stepping. */
6560 if (vcpu->mmio_is_write)
6562 vcpu->mmio_read_completed = 1;
6563 return complete_emulated_io(vcpu);
6566 run->exit_reason = KVM_EXIT_MMIO;
6567 run->mmio.phys_addr = frag->gpa;
6568 if (vcpu->mmio_is_write)
6569 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6570 run->mmio.len = min(8u, frag->len);
6571 run->mmio.is_write = vcpu->mmio_is_write;
6572 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6577 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6582 if (!tsk_used_math(current) && init_fpu(current))
6585 if (vcpu->sigset_active)
6586 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6588 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6589 kvm_vcpu_block(vcpu);
6590 kvm_apic_accept_events(vcpu);
6591 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6596 /* re-sync apic's tpr */
6597 if (!irqchip_in_kernel(vcpu->kvm)) {
6598 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6604 if (unlikely(vcpu->arch.complete_userspace_io)) {
6605 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6606 vcpu->arch.complete_userspace_io = NULL;
6611 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6616 post_kvm_run_save(vcpu);
6617 if (vcpu->sigset_active)
6618 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6623 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6625 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6627 * We are here if userspace calls get_regs() in the middle of
6628 * instruction emulation. Registers state needs to be copied
6629 * back from emulation context to vcpu. Userspace shouldn't do
6630 * that usually, but some bad designed PV devices (vmware
6631 * backdoor interface) need this to work
6633 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6634 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6636 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6637 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6638 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6639 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6640 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6641 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6642 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6643 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
6644 #ifdef CONFIG_X86_64
6645 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
6646 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
6647 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
6648 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
6649 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
6650 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
6651 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
6652 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
6655 regs->rip = kvm_rip_read(vcpu);
6656 regs->rflags = kvm_get_rflags(vcpu);
6661 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6663 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
6664 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6666 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
6667 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
6668 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
6669 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
6670 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
6671 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
6672 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
6673 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
6674 #ifdef CONFIG_X86_64
6675 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
6676 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
6677 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
6678 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
6679 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
6680 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
6681 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
6682 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
6685 kvm_rip_write(vcpu, regs->rip);
6686 kvm_set_rflags(vcpu, regs->rflags);
6688 vcpu->arch.exception.pending = false;
6690 kvm_make_request(KVM_REQ_EVENT, vcpu);
6695 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
6697 struct kvm_segment cs;
6699 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
6703 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
6705 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
6706 struct kvm_sregs *sregs)
6710 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6711 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6712 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6713 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6714 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6715 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6717 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6718 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6720 kvm_x86_ops->get_idt(vcpu, &dt);
6721 sregs->idt.limit = dt.size;
6722 sregs->idt.base = dt.address;
6723 kvm_x86_ops->get_gdt(vcpu, &dt);
6724 sregs->gdt.limit = dt.size;
6725 sregs->gdt.base = dt.address;
6727 sregs->cr0 = kvm_read_cr0(vcpu);
6728 sregs->cr2 = vcpu->arch.cr2;
6729 sregs->cr3 = kvm_read_cr3(vcpu);
6730 sregs->cr4 = kvm_read_cr4(vcpu);
6731 sregs->cr8 = kvm_get_cr8(vcpu);
6732 sregs->efer = vcpu->arch.efer;
6733 sregs->apic_base = kvm_get_apic_base(vcpu);
6735 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
6737 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
6738 set_bit(vcpu->arch.interrupt.nr,
6739 (unsigned long *)sregs->interrupt_bitmap);
6744 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
6745 struct kvm_mp_state *mp_state)
6747 kvm_apic_accept_events(vcpu);
6748 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
6749 vcpu->arch.pv.pv_unhalted)
6750 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
6752 mp_state->mp_state = vcpu->arch.mp_state;
6757 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
6758 struct kvm_mp_state *mp_state)
6760 if (!kvm_vcpu_has_lapic(vcpu) &&
6761 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
6764 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
6765 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
6766 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
6768 vcpu->arch.mp_state = mp_state->mp_state;
6769 kvm_make_request(KVM_REQ_EVENT, vcpu);
6773 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
6774 int reason, bool has_error_code, u32 error_code)
6776 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6779 init_emulate_ctxt(vcpu);
6781 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
6782 has_error_code, error_code);
6785 return EMULATE_FAIL;
6787 kvm_rip_write(vcpu, ctxt->eip);
6788 kvm_set_rflags(vcpu, ctxt->eflags);
6789 kvm_make_request(KVM_REQ_EVENT, vcpu);
6790 return EMULATE_DONE;
6792 EXPORT_SYMBOL_GPL(kvm_task_switch);
6794 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6795 struct kvm_sregs *sregs)
6797 struct msr_data apic_base_msr;
6798 int mmu_reset_needed = 0;
6799 int pending_vec, max_bits, idx;
6802 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
6805 dt.size = sregs->idt.limit;
6806 dt.address = sregs->idt.base;
6807 kvm_x86_ops->set_idt(vcpu, &dt);
6808 dt.size = sregs->gdt.limit;
6809 dt.address = sregs->gdt.base;
6810 kvm_x86_ops->set_gdt(vcpu, &dt);
6812 vcpu->arch.cr2 = sregs->cr2;
6813 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
6814 vcpu->arch.cr3 = sregs->cr3;
6815 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
6817 kvm_set_cr8(vcpu, sregs->cr8);
6819 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
6820 kvm_x86_ops->set_efer(vcpu, sregs->efer);
6821 apic_base_msr.data = sregs->apic_base;
6822 apic_base_msr.host_initiated = true;
6823 kvm_set_apic_base(vcpu, &apic_base_msr);
6825 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
6826 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
6827 vcpu->arch.cr0 = sregs->cr0;
6829 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6830 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6831 if (sregs->cr4 & X86_CR4_OSXSAVE)
6832 kvm_update_cpuid(vcpu);
6834 idx = srcu_read_lock(&vcpu->kvm->srcu);
6835 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
6836 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
6837 mmu_reset_needed = 1;
6839 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6841 if (mmu_reset_needed)
6842 kvm_mmu_reset_context(vcpu);
6844 max_bits = KVM_NR_INTERRUPTS;
6845 pending_vec = find_first_bit(
6846 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
6847 if (pending_vec < max_bits) {
6848 kvm_queue_interrupt(vcpu, pending_vec, false);
6849 pr_debug("Set back pending irq %d\n", pending_vec);
6852 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
6853 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
6854 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
6855 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
6856 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
6857 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
6859 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
6860 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
6862 update_cr8_intercept(vcpu);
6864 /* Older userspace won't unhalt the vcpu on reset. */
6865 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
6866 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
6868 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6870 kvm_make_request(KVM_REQ_EVENT, vcpu);
6875 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
6876 struct kvm_guest_debug *dbg)
6878 unsigned long rflags;
6881 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
6883 if (vcpu->arch.exception.pending)
6885 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
6886 kvm_queue_exception(vcpu, DB_VECTOR);
6888 kvm_queue_exception(vcpu, BP_VECTOR);
6892 * Read rflags as long as potentially injected trace flags are still
6895 rflags = kvm_get_rflags(vcpu);
6897 vcpu->guest_debug = dbg->control;
6898 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
6899 vcpu->guest_debug = 0;
6901 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6902 for (i = 0; i < KVM_NR_DB_REGS; ++i)
6903 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
6904 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
6906 for (i = 0; i < KVM_NR_DB_REGS; i++)
6907 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
6909 kvm_update_dr7(vcpu);
6911 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6912 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
6913 get_segment_base(vcpu, VCPU_SREG_CS);
6916 * Trigger an rflags update that will inject or remove the trace
6919 kvm_set_rflags(vcpu, rflags);
6921 kvm_x86_ops->update_db_bp_intercept(vcpu);
6931 * Translate a guest virtual address to a guest physical address.
6933 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
6934 struct kvm_translation *tr)
6936 unsigned long vaddr = tr->linear_address;
6940 idx = srcu_read_lock(&vcpu->kvm->srcu);
6941 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
6942 srcu_read_unlock(&vcpu->kvm->srcu, idx);
6943 tr->physical_address = gpa;
6944 tr->valid = gpa != UNMAPPED_GVA;
6951 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6953 struct i387_fxsave_struct *fxsave =
6954 &vcpu->arch.guest_fpu.state->fxsave;
6956 memcpy(fpu->fpr, fxsave->st_space, 128);
6957 fpu->fcw = fxsave->cwd;
6958 fpu->fsw = fxsave->swd;
6959 fpu->ftwx = fxsave->twd;
6960 fpu->last_opcode = fxsave->fop;
6961 fpu->last_ip = fxsave->rip;
6962 fpu->last_dp = fxsave->rdp;
6963 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
6968 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
6970 struct i387_fxsave_struct *fxsave =
6971 &vcpu->arch.guest_fpu.state->fxsave;
6973 memcpy(fxsave->st_space, fpu->fpr, 128);
6974 fxsave->cwd = fpu->fcw;
6975 fxsave->swd = fpu->fsw;
6976 fxsave->twd = fpu->ftwx;
6977 fxsave->fop = fpu->last_opcode;
6978 fxsave->rip = fpu->last_ip;
6979 fxsave->rdp = fpu->last_dp;
6980 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6985 int fx_init(struct kvm_vcpu *vcpu)
6989 err = fpu_alloc(&vcpu->arch.guest_fpu);
6993 fpu_finit(&vcpu->arch.guest_fpu);
6995 vcpu->arch.guest_fpu.state->xsave.xsave_hdr.xcomp_bv =
6996 host_xcr0 | XSTATE_COMPACTION_ENABLED;
6999 * Ensure guest xcr0 is valid for loading
7001 vcpu->arch.xcr0 = XSTATE_FP;
7003 vcpu->arch.cr0 |= X86_CR0_ET;
7007 EXPORT_SYMBOL_GPL(fx_init);
7009 static void fx_free(struct kvm_vcpu *vcpu)
7011 fpu_free(&vcpu->arch.guest_fpu);
7014 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7016 if (vcpu->guest_fpu_loaded)
7020 * Restore all possible states in the guest,
7021 * and assume host would use all available bits.
7022 * Guest xcr0 would be loaded later.
7024 kvm_put_guest_xcr0(vcpu);
7025 vcpu->guest_fpu_loaded = 1;
7026 __kernel_fpu_begin();
7027 fpu_restore_checking(&vcpu->arch.guest_fpu);
7031 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7033 kvm_put_guest_xcr0(vcpu);
7035 if (!vcpu->guest_fpu_loaded)
7038 vcpu->guest_fpu_loaded = 0;
7039 fpu_save_init(&vcpu->arch.guest_fpu);
7041 ++vcpu->stat.fpu_reload;
7042 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7046 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7048 kvmclock_reset(vcpu);
7050 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7052 kvm_x86_ops->vcpu_free(vcpu);
7055 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7058 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7059 printk_once(KERN_WARNING
7060 "kvm: SMP vm created on host with unstable TSC; "
7061 "guest TSC will not be reliable\n");
7062 return kvm_x86_ops->vcpu_create(kvm, id);
7065 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7069 vcpu->arch.mtrr_state.have_fixed = 1;
7070 r = vcpu_load(vcpu);
7073 kvm_vcpu_reset(vcpu);
7074 kvm_mmu_setup(vcpu);
7080 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7082 struct msr_data msr;
7083 struct kvm *kvm = vcpu->kvm;
7085 if (vcpu_load(vcpu))
7088 msr.index = MSR_IA32_TSC;
7089 msr.host_initiated = true;
7090 kvm_write_tsc(vcpu, &msr);
7093 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7094 KVMCLOCK_SYNC_PERIOD);
7097 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7100 vcpu->arch.apf.msr_val = 0;
7102 r = vcpu_load(vcpu);
7104 kvm_mmu_unload(vcpu);
7108 kvm_x86_ops->vcpu_free(vcpu);
7111 void kvm_vcpu_reset(struct kvm_vcpu *vcpu)
7113 atomic_set(&vcpu->arch.nmi_queued, 0);
7114 vcpu->arch.nmi_pending = 0;
7115 vcpu->arch.nmi_injected = false;
7116 kvm_clear_interrupt_queue(vcpu);
7117 kvm_clear_exception_queue(vcpu);
7119 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7120 kvm_update_dr0123(vcpu);
7121 vcpu->arch.dr6 = DR6_INIT;
7122 kvm_update_dr6(vcpu);
7123 vcpu->arch.dr7 = DR7_FIXED_1;
7124 kvm_update_dr7(vcpu);
7126 kvm_make_request(KVM_REQ_EVENT, vcpu);
7127 vcpu->arch.apf.msr_val = 0;
7128 vcpu->arch.st.msr_val = 0;
7130 kvmclock_reset(vcpu);
7132 kvm_clear_async_pf_completion_queue(vcpu);
7133 kvm_async_pf_hash_reset(vcpu);
7134 vcpu->arch.apf.halted = false;
7136 kvm_pmu_reset(vcpu);
7138 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7139 vcpu->arch.regs_avail = ~0;
7140 vcpu->arch.regs_dirty = ~0;
7142 kvm_x86_ops->vcpu_reset(vcpu);
7145 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7147 struct kvm_segment cs;
7149 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7150 cs.selector = vector << 8;
7151 cs.base = vector << 12;
7152 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7153 kvm_rip_write(vcpu, 0);
7156 int kvm_arch_hardware_enable(void)
7159 struct kvm_vcpu *vcpu;
7164 bool stable, backwards_tsc = false;
7166 kvm_shared_msr_cpu_online();
7167 ret = kvm_x86_ops->hardware_enable();
7171 local_tsc = native_read_tsc();
7172 stable = !check_tsc_unstable();
7173 list_for_each_entry(kvm, &vm_list, vm_list) {
7174 kvm_for_each_vcpu(i, vcpu, kvm) {
7175 if (!stable && vcpu->cpu == smp_processor_id())
7176 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7177 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7178 backwards_tsc = true;
7179 if (vcpu->arch.last_host_tsc > max_tsc)
7180 max_tsc = vcpu->arch.last_host_tsc;
7186 * Sometimes, even reliable TSCs go backwards. This happens on
7187 * platforms that reset TSC during suspend or hibernate actions, but
7188 * maintain synchronization. We must compensate. Fortunately, we can
7189 * detect that condition here, which happens early in CPU bringup,
7190 * before any KVM threads can be running. Unfortunately, we can't
7191 * bring the TSCs fully up to date with real time, as we aren't yet far
7192 * enough into CPU bringup that we know how much real time has actually
7193 * elapsed; our helper function, get_kernel_ns() will be using boot
7194 * variables that haven't been updated yet.
7196 * So we simply find the maximum observed TSC above, then record the
7197 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7198 * the adjustment will be applied. Note that we accumulate
7199 * adjustments, in case multiple suspend cycles happen before some VCPU
7200 * gets a chance to run again. In the event that no KVM threads get a
7201 * chance to run, we will miss the entire elapsed period, as we'll have
7202 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7203 * loose cycle time. This isn't too big a deal, since the loss will be
7204 * uniform across all VCPUs (not to mention the scenario is extremely
7205 * unlikely). It is possible that a second hibernate recovery happens
7206 * much faster than a first, causing the observed TSC here to be
7207 * smaller; this would require additional padding adjustment, which is
7208 * why we set last_host_tsc to the local tsc observed here.
7210 * N.B. - this code below runs only on platforms with reliable TSC,
7211 * as that is the only way backwards_tsc is set above. Also note
7212 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7213 * have the same delta_cyc adjustment applied if backwards_tsc
7214 * is detected. Note further, this adjustment is only done once,
7215 * as we reset last_host_tsc on all VCPUs to stop this from being
7216 * called multiple times (one for each physical CPU bringup).
7218 * Platforms with unreliable TSCs don't have to deal with this, they
7219 * will be compensated by the logic in vcpu_load, which sets the TSC to
7220 * catchup mode. This will catchup all VCPUs to real time, but cannot
7221 * guarantee that they stay in perfect synchronization.
7223 if (backwards_tsc) {
7224 u64 delta_cyc = max_tsc - local_tsc;
7225 backwards_tsc_observed = true;
7226 list_for_each_entry(kvm, &vm_list, vm_list) {
7227 kvm_for_each_vcpu(i, vcpu, kvm) {
7228 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7229 vcpu->arch.last_host_tsc = local_tsc;
7230 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7234 * We have to disable TSC offset matching.. if you were
7235 * booting a VM while issuing an S4 host suspend....
7236 * you may have some problem. Solving this issue is
7237 * left as an exercise to the reader.
7239 kvm->arch.last_tsc_nsec = 0;
7240 kvm->arch.last_tsc_write = 0;
7247 void kvm_arch_hardware_disable(void)
7249 kvm_x86_ops->hardware_disable();
7250 drop_user_return_notifiers();
7253 int kvm_arch_hardware_setup(void)
7255 return kvm_x86_ops->hardware_setup();
7258 void kvm_arch_hardware_unsetup(void)
7260 kvm_x86_ops->hardware_unsetup();
7263 void kvm_arch_check_processor_compat(void *rtn)
7265 kvm_x86_ops->check_processor_compatibility(rtn);
7268 bool kvm_vcpu_compatible(struct kvm_vcpu *vcpu)
7270 return irqchip_in_kernel(vcpu->kvm) == (vcpu->arch.apic != NULL);
7273 struct static_key kvm_no_apic_vcpu __read_mostly;
7275 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7281 BUG_ON(vcpu->kvm == NULL);
7284 vcpu->arch.pv.pv_unhalted = false;
7285 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7286 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7287 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7289 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7291 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7296 vcpu->arch.pio_data = page_address(page);
7298 kvm_set_tsc_khz(vcpu, max_tsc_khz);
7300 r = kvm_mmu_create(vcpu);
7302 goto fail_free_pio_data;
7304 if (irqchip_in_kernel(kvm)) {
7305 r = kvm_create_lapic(vcpu);
7307 goto fail_mmu_destroy;
7309 static_key_slow_inc(&kvm_no_apic_vcpu);
7311 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7313 if (!vcpu->arch.mce_banks) {
7315 goto fail_free_lapic;
7317 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7319 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7321 goto fail_free_mce_banks;
7326 goto fail_free_wbinvd_dirty_mask;
7328 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7329 vcpu->arch.pv_time_enabled = false;
7331 vcpu->arch.guest_supported_xcr0 = 0;
7332 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7334 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7336 kvm_async_pf_hash_reset(vcpu);
7340 fail_free_wbinvd_dirty_mask:
7341 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7342 fail_free_mce_banks:
7343 kfree(vcpu->arch.mce_banks);
7345 kvm_free_lapic(vcpu);
7347 kvm_mmu_destroy(vcpu);
7349 free_page((unsigned long)vcpu->arch.pio_data);
7354 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7358 kvm_pmu_destroy(vcpu);
7359 kfree(vcpu->arch.mce_banks);
7360 kvm_free_lapic(vcpu);
7361 idx = srcu_read_lock(&vcpu->kvm->srcu);
7362 kvm_mmu_destroy(vcpu);
7363 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7364 free_page((unsigned long)vcpu->arch.pio_data);
7365 if (!irqchip_in_kernel(vcpu->kvm))
7366 static_key_slow_dec(&kvm_no_apic_vcpu);
7369 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7371 kvm_x86_ops->sched_in(vcpu, cpu);
7374 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7379 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7380 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7381 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7382 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7383 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7385 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7386 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7387 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7388 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7389 &kvm->arch.irq_sources_bitmap);
7391 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7392 mutex_init(&kvm->arch.apic_map_lock);
7393 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7395 pvclock_update_vm_gtod_copy(kvm);
7397 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7398 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7403 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7406 r = vcpu_load(vcpu);
7408 kvm_mmu_unload(vcpu);
7412 static void kvm_free_vcpus(struct kvm *kvm)
7415 struct kvm_vcpu *vcpu;
7418 * Unpin any mmu pages first.
7420 kvm_for_each_vcpu(i, vcpu, kvm) {
7421 kvm_clear_async_pf_completion_queue(vcpu);
7422 kvm_unload_vcpu_mmu(vcpu);
7424 kvm_for_each_vcpu(i, vcpu, kvm)
7425 kvm_arch_vcpu_free(vcpu);
7427 mutex_lock(&kvm->lock);
7428 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7429 kvm->vcpus[i] = NULL;
7431 atomic_set(&kvm->online_vcpus, 0);
7432 mutex_unlock(&kvm->lock);
7435 void kvm_arch_sync_events(struct kvm *kvm)
7437 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7438 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7439 kvm_free_all_assigned_devices(kvm);
7443 void kvm_arch_destroy_vm(struct kvm *kvm)
7445 if (current->mm == kvm->mm) {
7447 * Free memory regions allocated on behalf of userspace,
7448 * unless the the memory map has changed due to process exit
7451 struct kvm_userspace_memory_region mem;
7452 memset(&mem, 0, sizeof(mem));
7453 mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
7454 kvm_set_memory_region(kvm, &mem);
7456 mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
7457 kvm_set_memory_region(kvm, &mem);
7459 mem.slot = TSS_PRIVATE_MEMSLOT;
7460 kvm_set_memory_region(kvm, &mem);
7462 kvm_iommu_unmap_guest(kvm);
7463 kfree(kvm->arch.vpic);
7464 kfree(kvm->arch.vioapic);
7465 kvm_free_vcpus(kvm);
7466 kfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7469 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7470 struct kvm_memory_slot *dont)
7474 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7475 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7476 kvfree(free->arch.rmap[i]);
7477 free->arch.rmap[i] = NULL;
7482 if (!dont || free->arch.lpage_info[i - 1] !=
7483 dont->arch.lpage_info[i - 1]) {
7484 kvfree(free->arch.lpage_info[i - 1]);
7485 free->arch.lpage_info[i - 1] = NULL;
7490 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7491 unsigned long npages)
7495 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7500 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7501 slot->base_gfn, level) + 1;
7503 slot->arch.rmap[i] =
7504 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7505 if (!slot->arch.rmap[i])
7510 slot->arch.lpage_info[i - 1] = kvm_kvzalloc(lpages *
7511 sizeof(*slot->arch.lpage_info[i - 1]));
7512 if (!slot->arch.lpage_info[i - 1])
7515 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7516 slot->arch.lpage_info[i - 1][0].write_count = 1;
7517 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7518 slot->arch.lpage_info[i - 1][lpages - 1].write_count = 1;
7519 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7521 * If the gfn and userspace address are not aligned wrt each
7522 * other, or if explicitly asked to, disable large page
7523 * support for this slot
7525 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7526 !kvm_largepages_enabled()) {
7529 for (j = 0; j < lpages; ++j)
7530 slot->arch.lpage_info[i - 1][j].write_count = 1;
7537 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7538 kvfree(slot->arch.rmap[i]);
7539 slot->arch.rmap[i] = NULL;
7543 kvfree(slot->arch.lpage_info[i - 1]);
7544 slot->arch.lpage_info[i - 1] = NULL;
7549 void kvm_arch_memslots_updated(struct kvm *kvm)
7552 * memslots->generation has been incremented.
7553 * mmio generation may have reached its maximum value.
7555 kvm_mmu_invalidate_mmio_sptes(kvm);
7558 int kvm_arch_prepare_memory_region(struct kvm *kvm,
7559 struct kvm_memory_slot *memslot,
7560 struct kvm_userspace_memory_region *mem,
7561 enum kvm_mr_change change)
7564 * Only private memory slots need to be mapped here since
7565 * KVM_SET_MEMORY_REGION ioctl is no longer supported.
7567 if ((memslot->id >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_CREATE)) {
7568 unsigned long userspace_addr;
7571 * MAP_SHARED to prevent internal slot pages from being moved
7574 userspace_addr = vm_mmap(NULL, 0, memslot->npages * PAGE_SIZE,
7575 PROT_READ | PROT_WRITE,
7576 MAP_SHARED | MAP_ANONYMOUS, 0);
7578 if (IS_ERR((void *)userspace_addr))
7579 return PTR_ERR((void *)userspace_addr);
7581 memslot->userspace_addr = userspace_addr;
7587 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
7588 struct kvm_memory_slot *new)
7590 /* Still write protect RO slot */
7591 if (new->flags & KVM_MEM_READONLY) {
7592 kvm_mmu_slot_remove_write_access(kvm, new);
7597 * Call kvm_x86_ops dirty logging hooks when they are valid.
7599 * kvm_x86_ops->slot_disable_log_dirty is called when:
7601 * - KVM_MR_CREATE with dirty logging is disabled
7602 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
7604 * The reason is, in case of PML, we need to set D-bit for any slots
7605 * with dirty logging disabled in order to eliminate unnecessary GPA
7606 * logging in PML buffer (and potential PML buffer full VMEXT). This
7607 * guarantees leaving PML enabled during guest's lifetime won't have
7608 * any additonal overhead from PML when guest is running with dirty
7609 * logging disabled for memory slots.
7611 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
7612 * to dirty logging mode.
7614 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
7616 * In case of write protect:
7618 * Write protect all pages for dirty logging.
7620 * All the sptes including the large sptes which point to this
7621 * slot are set to readonly. We can not create any new large
7622 * spte on this slot until the end of the logging.
7624 * See the comments in fast_page_fault().
7626 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
7627 if (kvm_x86_ops->slot_enable_log_dirty)
7628 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
7630 kvm_mmu_slot_remove_write_access(kvm, new);
7632 if (kvm_x86_ops->slot_disable_log_dirty)
7633 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
7637 void kvm_arch_commit_memory_region(struct kvm *kvm,
7638 struct kvm_userspace_memory_region *mem,
7639 const struct kvm_memory_slot *old,
7640 enum kvm_mr_change change)
7642 struct kvm_memory_slot *new;
7643 int nr_mmu_pages = 0;
7645 if ((mem->slot >= KVM_USER_MEM_SLOTS) && (change == KVM_MR_DELETE)) {
7648 ret = vm_munmap(old->userspace_addr,
7649 old->npages * PAGE_SIZE);
7652 "kvm_vm_ioctl_set_memory_region: "
7653 "failed to munmap memory\n");
7656 if (!kvm->arch.n_requested_mmu_pages)
7657 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
7660 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
7662 /* It's OK to get 'new' slot here as it has already been installed */
7663 new = id_to_memslot(kvm->memslots, mem->slot);
7666 * Set up write protection and/or dirty logging for the new slot.
7668 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
7669 * been zapped so no dirty logging staff is needed for old slot. For
7670 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
7671 * new and it's also covered when dealing with the new slot.
7673 if (change != KVM_MR_DELETE)
7674 kvm_mmu_slot_apply_flags(kvm, new);
7677 void kvm_arch_flush_shadow_all(struct kvm *kvm)
7679 kvm_mmu_invalidate_zap_all_pages(kvm);
7682 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
7683 struct kvm_memory_slot *slot)
7685 kvm_mmu_invalidate_zap_all_pages(kvm);
7688 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
7690 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7691 kvm_x86_ops->check_nested_events(vcpu, false);
7693 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7694 !vcpu->arch.apf.halted)
7695 || !list_empty_careful(&vcpu->async_pf.done)
7696 || kvm_apic_has_events(vcpu)
7697 || vcpu->arch.pv.pv_unhalted
7698 || atomic_read(&vcpu->arch.nmi_queued) ||
7699 (kvm_arch_interrupt_allowed(vcpu) &&
7700 kvm_cpu_has_interrupt(vcpu));
7703 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
7705 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
7708 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
7710 return kvm_x86_ops->interrupt_allowed(vcpu);
7713 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
7715 if (is_64_bit_mode(vcpu))
7716 return kvm_rip_read(vcpu);
7717 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
7718 kvm_rip_read(vcpu));
7720 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
7722 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
7724 return kvm_get_linear_rip(vcpu) == linear_rip;
7726 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
7728 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
7730 unsigned long rflags;
7732 rflags = kvm_x86_ops->get_rflags(vcpu);
7733 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7734 rflags &= ~X86_EFLAGS_TF;
7737 EXPORT_SYMBOL_GPL(kvm_get_rflags);
7739 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7741 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
7742 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
7743 rflags |= X86_EFLAGS_TF;
7744 kvm_x86_ops->set_rflags(vcpu, rflags);
7747 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
7749 __kvm_set_rflags(vcpu, rflags);
7750 kvm_make_request(KVM_REQ_EVENT, vcpu);
7752 EXPORT_SYMBOL_GPL(kvm_set_rflags);
7754 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
7758 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
7762 r = kvm_mmu_reload(vcpu);
7766 if (!vcpu->arch.mmu.direct_map &&
7767 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
7770 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
7773 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
7775 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
7778 static inline u32 kvm_async_pf_next_probe(u32 key)
7780 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
7783 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7785 u32 key = kvm_async_pf_hash_fn(gfn);
7787 while (vcpu->arch.apf.gfns[key] != ~0)
7788 key = kvm_async_pf_next_probe(key);
7790 vcpu->arch.apf.gfns[key] = gfn;
7793 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
7796 u32 key = kvm_async_pf_hash_fn(gfn);
7798 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
7799 (vcpu->arch.apf.gfns[key] != gfn &&
7800 vcpu->arch.apf.gfns[key] != ~0); i++)
7801 key = kvm_async_pf_next_probe(key);
7806 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7808 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
7811 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
7815 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
7817 vcpu->arch.apf.gfns[i] = ~0;
7819 j = kvm_async_pf_next_probe(j);
7820 if (vcpu->arch.apf.gfns[j] == ~0)
7822 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
7824 * k lies cyclically in ]i,j]
7826 * |....j i.k.| or |.k..j i...|
7828 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
7829 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
7834 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
7837 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
7841 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
7842 struct kvm_async_pf *work)
7844 struct x86_exception fault;
7846 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
7847 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
7849 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
7850 (vcpu->arch.apf.send_user_only &&
7851 kvm_x86_ops->get_cpl(vcpu) == 0))
7852 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
7853 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
7854 fault.vector = PF_VECTOR;
7855 fault.error_code_valid = true;
7856 fault.error_code = 0;
7857 fault.nested_page_fault = false;
7858 fault.address = work->arch.token;
7859 kvm_inject_page_fault(vcpu, &fault);
7863 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
7864 struct kvm_async_pf *work)
7866 struct x86_exception fault;
7868 trace_kvm_async_pf_ready(work->arch.token, work->gva);
7869 if (work->wakeup_all)
7870 work->arch.token = ~0; /* broadcast wakeup */
7872 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
7874 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
7875 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
7876 fault.vector = PF_VECTOR;
7877 fault.error_code_valid = true;
7878 fault.error_code = 0;
7879 fault.nested_page_fault = false;
7880 fault.address = work->arch.token;
7881 kvm_inject_page_fault(vcpu, &fault);
7883 vcpu->arch.apf.halted = false;
7884 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7887 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
7889 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
7892 return !kvm_event_needs_reinjection(vcpu) &&
7893 kvm_x86_ops->interrupt_allowed(vcpu);
7896 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
7898 atomic_inc(&kvm->arch.noncoherent_dma_count);
7900 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
7902 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
7904 atomic_dec(&kvm->arch.noncoherent_dma_count);
7906 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
7908 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
7910 return atomic_read(&kvm->arch.noncoherent_dma_count);
7912 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
7914 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
7915 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
7916 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
7917 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
7918 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
7919 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
7920 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
7921 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
7922 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
7923 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
7924 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
7925 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
7926 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
7927 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
7928 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);