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"
34 #include <linux/clocksource.h>
35 #include <linux/interrupt.h>
36 #include <linux/kvm.h>
38 #include <linux/vmalloc.h>
39 #include <linux/module.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <trace/events/kvm.h>
58 #define CREATE_TRACE_POINTS
61 #include <asm/debugreg.h>
65 #include <linux/kernel_stat.h>
66 #include <asm/fpu/internal.h> /* Ugh! */
67 #include <asm/pvclock.h>
68 #include <asm/div64.h>
69 #include <asm/irq_remapping.h>
71 #define MAX_IO_MSRS 256
72 #define KVM_MAX_MCE_BANKS 32
73 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
74 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
76 #define emul_to_vcpu(ctxt) \
77 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
80 * - enable syscall per default because its emulated by KVM
81 * - enable LME and LMA per default on 64 bit KVM
85 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
87 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
90 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
91 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
93 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
94 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
96 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
97 static void process_nmi(struct kvm_vcpu *vcpu);
98 static void enter_smm(struct kvm_vcpu *vcpu);
99 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
101 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
102 EXPORT_SYMBOL_GPL(kvm_x86_ops);
104 static bool __read_mostly ignore_msrs = 0;
105 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
107 unsigned int min_timer_period_us = 500;
108 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
110 static bool __read_mostly kvmclock_periodic_sync = true;
111 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
113 bool __read_mostly kvm_has_tsc_control;
114 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
115 u32 __read_mostly kvm_max_guest_tsc_khz;
116 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
117 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
118 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
119 u64 __read_mostly kvm_max_tsc_scaling_ratio;
120 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
121 u64 __read_mostly kvm_default_tsc_scaling_ratio;
122 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
124 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
125 static u32 __read_mostly tsc_tolerance_ppm = 250;
126 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
128 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
129 unsigned int __read_mostly lapic_timer_advance_ns = 0;
130 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
132 static bool __read_mostly vector_hashing = true;
133 module_param(vector_hashing, bool, S_IRUGO);
135 static bool __read_mostly backwards_tsc_observed = false;
137 #define KVM_NR_SHARED_MSRS 16
139 struct kvm_shared_msrs_global {
141 u32 msrs[KVM_NR_SHARED_MSRS];
144 struct kvm_shared_msrs {
145 struct user_return_notifier urn;
147 struct kvm_shared_msr_values {
150 } values[KVM_NR_SHARED_MSRS];
153 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
154 static struct kvm_shared_msrs __percpu *shared_msrs;
156 struct kvm_stats_debugfs_item debugfs_entries[] = {
157 { "pf_fixed", VCPU_STAT(pf_fixed) },
158 { "pf_guest", VCPU_STAT(pf_guest) },
159 { "tlb_flush", VCPU_STAT(tlb_flush) },
160 { "invlpg", VCPU_STAT(invlpg) },
161 { "exits", VCPU_STAT(exits) },
162 { "io_exits", VCPU_STAT(io_exits) },
163 { "mmio_exits", VCPU_STAT(mmio_exits) },
164 { "signal_exits", VCPU_STAT(signal_exits) },
165 { "irq_window", VCPU_STAT(irq_window_exits) },
166 { "nmi_window", VCPU_STAT(nmi_window_exits) },
167 { "halt_exits", VCPU_STAT(halt_exits) },
168 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
169 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
170 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
171 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
172 { "hypercalls", VCPU_STAT(hypercalls) },
173 { "request_irq", VCPU_STAT(request_irq_exits) },
174 { "irq_exits", VCPU_STAT(irq_exits) },
175 { "host_state_reload", VCPU_STAT(host_state_reload) },
176 { "efer_reload", VCPU_STAT(efer_reload) },
177 { "fpu_reload", VCPU_STAT(fpu_reload) },
178 { "insn_emulation", VCPU_STAT(insn_emulation) },
179 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
180 { "irq_injections", VCPU_STAT(irq_injections) },
181 { "nmi_injections", VCPU_STAT(nmi_injections) },
182 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
183 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
184 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
185 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
186 { "mmu_flooded", VM_STAT(mmu_flooded) },
187 { "mmu_recycled", VM_STAT(mmu_recycled) },
188 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
189 { "mmu_unsync", VM_STAT(mmu_unsync) },
190 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
191 { "largepages", VM_STAT(lpages) },
195 u64 __read_mostly host_xcr0;
197 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
199 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
202 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
203 vcpu->arch.apf.gfns[i] = ~0;
206 static void kvm_on_user_return(struct user_return_notifier *urn)
209 struct kvm_shared_msrs *locals
210 = container_of(urn, struct kvm_shared_msrs, urn);
211 struct kvm_shared_msr_values *values;
213 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
214 values = &locals->values[slot];
215 if (values->host != values->curr) {
216 wrmsrl(shared_msrs_global.msrs[slot], values->host);
217 values->curr = values->host;
220 locals->registered = false;
221 user_return_notifier_unregister(urn);
224 static void shared_msr_update(unsigned slot, u32 msr)
227 unsigned int cpu = smp_processor_id();
228 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
230 /* only read, and nobody should modify it at this time,
231 * so don't need lock */
232 if (slot >= shared_msrs_global.nr) {
233 printk(KERN_ERR "kvm: invalid MSR slot!");
236 rdmsrl_safe(msr, &value);
237 smsr->values[slot].host = value;
238 smsr->values[slot].curr = value;
241 void kvm_define_shared_msr(unsigned slot, u32 msr)
243 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
244 shared_msrs_global.msrs[slot] = msr;
245 if (slot >= shared_msrs_global.nr)
246 shared_msrs_global.nr = slot + 1;
248 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
250 static void kvm_shared_msr_cpu_online(void)
254 for (i = 0; i < shared_msrs_global.nr; ++i)
255 shared_msr_update(i, shared_msrs_global.msrs[i]);
258 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
260 unsigned int cpu = smp_processor_id();
261 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
264 if (((value ^ smsr->values[slot].curr) & mask) == 0)
266 smsr->values[slot].curr = value;
267 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
271 if (!smsr->registered) {
272 smsr->urn.on_user_return = kvm_on_user_return;
273 user_return_notifier_register(&smsr->urn);
274 smsr->registered = true;
278 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
280 static void drop_user_return_notifiers(void)
282 unsigned int cpu = smp_processor_id();
283 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
285 if (smsr->registered)
286 kvm_on_user_return(&smsr->urn);
289 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
291 return vcpu->arch.apic_base;
293 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
295 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
297 u64 old_state = vcpu->arch.apic_base &
298 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
299 u64 new_state = msr_info->data &
300 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
301 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
302 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
304 if (!msr_info->host_initiated &&
305 ((msr_info->data & reserved_bits) != 0 ||
306 new_state == X2APIC_ENABLE ||
307 (new_state == MSR_IA32_APICBASE_ENABLE &&
308 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
309 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
313 kvm_lapic_set_base(vcpu, msr_info->data);
316 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
318 asmlinkage __visible void kvm_spurious_fault(void)
320 /* Fault while not rebooting. We want the trace. */
323 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
325 #define EXCPT_BENIGN 0
326 #define EXCPT_CONTRIBUTORY 1
329 static int exception_class(int vector)
339 return EXCPT_CONTRIBUTORY;
346 #define EXCPT_FAULT 0
348 #define EXCPT_ABORT 2
349 #define EXCPT_INTERRUPT 3
351 static int exception_type(int vector)
355 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
356 return EXCPT_INTERRUPT;
360 /* #DB is trap, as instruction watchpoints are handled elsewhere */
361 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
364 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
367 /* Reserved exceptions will result in fault */
371 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
372 unsigned nr, bool has_error, u32 error_code,
378 kvm_make_request(KVM_REQ_EVENT, vcpu);
380 if (!vcpu->arch.exception.pending) {
382 if (has_error && !is_protmode(vcpu))
384 vcpu->arch.exception.pending = true;
385 vcpu->arch.exception.has_error_code = has_error;
386 vcpu->arch.exception.nr = nr;
387 vcpu->arch.exception.error_code = error_code;
388 vcpu->arch.exception.reinject = reinject;
392 /* to check exception */
393 prev_nr = vcpu->arch.exception.nr;
394 if (prev_nr == DF_VECTOR) {
395 /* triple fault -> shutdown */
396 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
399 class1 = exception_class(prev_nr);
400 class2 = exception_class(nr);
401 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
402 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
403 /* generate double fault per SDM Table 5-5 */
404 vcpu->arch.exception.pending = true;
405 vcpu->arch.exception.has_error_code = true;
406 vcpu->arch.exception.nr = DF_VECTOR;
407 vcpu->arch.exception.error_code = 0;
409 /* replace previous exception with a new one in a hope
410 that instruction re-execution will regenerate lost
415 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
417 kvm_multiple_exception(vcpu, nr, false, 0, false);
419 EXPORT_SYMBOL_GPL(kvm_queue_exception);
421 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
423 kvm_multiple_exception(vcpu, nr, false, 0, true);
425 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
427 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
430 kvm_inject_gp(vcpu, 0);
432 kvm_x86_ops->skip_emulated_instruction(vcpu);
434 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
436 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
438 ++vcpu->stat.pf_guest;
439 vcpu->arch.cr2 = fault->address;
440 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
442 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
444 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
446 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
447 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
449 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
451 return fault->nested_page_fault;
454 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
456 atomic_inc(&vcpu->arch.nmi_queued);
457 kvm_make_request(KVM_REQ_NMI, vcpu);
459 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
461 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
463 kvm_multiple_exception(vcpu, nr, true, error_code, false);
465 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
467 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
469 kvm_multiple_exception(vcpu, nr, true, error_code, true);
471 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
474 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
475 * a #GP and return false.
477 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
479 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
481 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
484 EXPORT_SYMBOL_GPL(kvm_require_cpl);
486 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
488 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
491 kvm_queue_exception(vcpu, UD_VECTOR);
494 EXPORT_SYMBOL_GPL(kvm_require_dr);
497 * This function will be used to read from the physical memory of the currently
498 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
499 * can read from guest physical or from the guest's guest physical memory.
501 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
502 gfn_t ngfn, void *data, int offset, int len,
505 struct x86_exception exception;
509 ngpa = gfn_to_gpa(ngfn);
510 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
511 if (real_gfn == UNMAPPED_GVA)
514 real_gfn = gpa_to_gfn(real_gfn);
516 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
518 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
520 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
521 void *data, int offset, int len, u32 access)
523 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
524 data, offset, len, access);
528 * Load the pae pdptrs. Return true is they are all valid.
530 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
532 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
533 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
536 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
538 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
539 offset * sizeof(u64), sizeof(pdpte),
540 PFERR_USER_MASK|PFERR_WRITE_MASK);
545 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
546 if ((pdpte[i] & PT_PRESENT_MASK) &&
548 vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
555 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
556 __set_bit(VCPU_EXREG_PDPTR,
557 (unsigned long *)&vcpu->arch.regs_avail);
558 __set_bit(VCPU_EXREG_PDPTR,
559 (unsigned long *)&vcpu->arch.regs_dirty);
564 EXPORT_SYMBOL_GPL(load_pdptrs);
566 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
568 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
574 if (is_long_mode(vcpu) || !is_pae(vcpu))
577 if (!test_bit(VCPU_EXREG_PDPTR,
578 (unsigned long *)&vcpu->arch.regs_avail))
581 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
582 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
583 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
584 PFERR_USER_MASK | PFERR_WRITE_MASK);
587 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
593 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
595 unsigned long old_cr0 = kvm_read_cr0(vcpu);
596 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
601 if (cr0 & 0xffffffff00000000UL)
605 cr0 &= ~CR0_RESERVED_BITS;
607 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
610 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
613 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
615 if ((vcpu->arch.efer & EFER_LME)) {
620 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
625 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
630 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
633 kvm_x86_ops->set_cr0(vcpu, cr0);
635 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
636 kvm_clear_async_pf_completion_queue(vcpu);
637 kvm_async_pf_hash_reset(vcpu);
640 if ((cr0 ^ old_cr0) & update_bits)
641 kvm_mmu_reset_context(vcpu);
643 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
644 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
645 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
646 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
650 EXPORT_SYMBOL_GPL(kvm_set_cr0);
652 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
654 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
656 EXPORT_SYMBOL_GPL(kvm_lmsw);
658 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
660 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
661 !vcpu->guest_xcr0_loaded) {
662 /* kvm_set_xcr() also depends on this */
663 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
664 vcpu->guest_xcr0_loaded = 1;
668 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
670 if (vcpu->guest_xcr0_loaded) {
671 if (vcpu->arch.xcr0 != host_xcr0)
672 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
673 vcpu->guest_xcr0_loaded = 0;
677 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
680 u64 old_xcr0 = vcpu->arch.xcr0;
683 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
684 if (index != XCR_XFEATURE_ENABLED_MASK)
686 if (!(xcr0 & XFEATURE_MASK_FP))
688 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
692 * Do not allow the guest to set bits that we do not support
693 * saving. However, xcr0 bit 0 is always set, even if the
694 * emulated CPU does not support XSAVE (see fx_init).
696 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
697 if (xcr0 & ~valid_bits)
700 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
701 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
704 if (xcr0 & XFEATURE_MASK_AVX512) {
705 if (!(xcr0 & XFEATURE_MASK_YMM))
707 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
710 vcpu->arch.xcr0 = xcr0;
712 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
713 kvm_update_cpuid(vcpu);
717 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
719 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
720 __kvm_set_xcr(vcpu, index, xcr)) {
721 kvm_inject_gp(vcpu, 0);
726 EXPORT_SYMBOL_GPL(kvm_set_xcr);
728 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
730 unsigned long old_cr4 = kvm_read_cr4(vcpu);
731 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
732 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
734 if (cr4 & CR4_RESERVED_BITS)
737 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
740 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
743 if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
746 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
749 if (!guest_cpuid_has_pku(vcpu) && (cr4 & X86_CR4_PKE))
752 if (is_long_mode(vcpu)) {
753 if (!(cr4 & X86_CR4_PAE))
755 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
756 && ((cr4 ^ old_cr4) & pdptr_bits)
757 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
761 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
762 if (!guest_cpuid_has_pcid(vcpu))
765 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
766 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
770 if (kvm_x86_ops->set_cr4(vcpu, cr4))
773 if (((cr4 ^ old_cr4) & pdptr_bits) ||
774 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
775 kvm_mmu_reset_context(vcpu);
777 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
778 kvm_update_cpuid(vcpu);
782 EXPORT_SYMBOL_GPL(kvm_set_cr4);
784 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
787 cr3 &= ~CR3_PCID_INVD;
790 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
791 kvm_mmu_sync_roots(vcpu);
792 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
796 if (is_long_mode(vcpu)) {
797 if (cr3 & CR3_L_MODE_RESERVED_BITS)
799 } else if (is_pae(vcpu) && is_paging(vcpu) &&
800 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
803 vcpu->arch.cr3 = cr3;
804 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
805 kvm_mmu_new_cr3(vcpu);
808 EXPORT_SYMBOL_GPL(kvm_set_cr3);
810 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
812 if (cr8 & CR8_RESERVED_BITS)
814 if (lapic_in_kernel(vcpu))
815 kvm_lapic_set_tpr(vcpu, cr8);
817 vcpu->arch.cr8 = cr8;
820 EXPORT_SYMBOL_GPL(kvm_set_cr8);
822 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
824 if (lapic_in_kernel(vcpu))
825 return kvm_lapic_get_cr8(vcpu);
827 return vcpu->arch.cr8;
829 EXPORT_SYMBOL_GPL(kvm_get_cr8);
831 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
835 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
836 for (i = 0; i < KVM_NR_DB_REGS; i++)
837 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
838 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
842 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
844 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
845 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
848 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
852 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
853 dr7 = vcpu->arch.guest_debug_dr7;
855 dr7 = vcpu->arch.dr7;
856 kvm_x86_ops->set_dr7(vcpu, dr7);
857 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
858 if (dr7 & DR7_BP_EN_MASK)
859 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
862 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
864 u64 fixed = DR6_FIXED_1;
866 if (!guest_cpuid_has_rtm(vcpu))
871 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
875 vcpu->arch.db[dr] = val;
876 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
877 vcpu->arch.eff_db[dr] = val;
882 if (val & 0xffffffff00000000ULL)
884 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
885 kvm_update_dr6(vcpu);
890 if (val & 0xffffffff00000000ULL)
892 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
893 kvm_update_dr7(vcpu);
900 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
902 if (__kvm_set_dr(vcpu, dr, val)) {
903 kvm_inject_gp(vcpu, 0);
908 EXPORT_SYMBOL_GPL(kvm_set_dr);
910 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
914 *val = vcpu->arch.db[dr];
919 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
920 *val = vcpu->arch.dr6;
922 *val = kvm_x86_ops->get_dr6(vcpu);
927 *val = vcpu->arch.dr7;
932 EXPORT_SYMBOL_GPL(kvm_get_dr);
934 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
936 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
940 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
943 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
944 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
947 EXPORT_SYMBOL_GPL(kvm_rdpmc);
950 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
951 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
953 * This list is modified at module load time to reflect the
954 * capabilities of the host cpu. This capabilities test skips MSRs that are
955 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
956 * may depend on host virtualization features rather than host cpu features.
959 static u32 msrs_to_save[] = {
960 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
963 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
965 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
966 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
969 static unsigned num_msrs_to_save;
971 static u32 emulated_msrs[] = {
972 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
973 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
974 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
975 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
976 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
977 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
980 HV_X64_MSR_VP_RUNTIME,
982 HV_X64_MSR_STIMER0_CONFIG,
983 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
987 MSR_IA32_TSCDEADLINE,
988 MSR_IA32_MISC_ENABLE,
991 MSR_IA32_MCG_EXT_CTL,
995 static unsigned num_emulated_msrs;
997 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
999 if (efer & efer_reserved_bits)
1002 if (efer & EFER_FFXSR) {
1003 struct kvm_cpuid_entry2 *feat;
1005 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1006 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
1010 if (efer & EFER_SVME) {
1011 struct kvm_cpuid_entry2 *feat;
1013 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1014 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
1020 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1022 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1024 u64 old_efer = vcpu->arch.efer;
1026 if (!kvm_valid_efer(vcpu, efer))
1030 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1034 efer |= vcpu->arch.efer & EFER_LMA;
1036 kvm_x86_ops->set_efer(vcpu, efer);
1038 /* Update reserved bits */
1039 if ((efer ^ old_efer) & EFER_NX)
1040 kvm_mmu_reset_context(vcpu);
1045 void kvm_enable_efer_bits(u64 mask)
1047 efer_reserved_bits &= ~mask;
1049 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1052 * Writes msr value into into the appropriate "register".
1053 * Returns 0 on success, non-0 otherwise.
1054 * Assumes vcpu_load() was already called.
1056 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1058 switch (msr->index) {
1061 case MSR_KERNEL_GS_BASE:
1064 if (is_noncanonical_address(msr->data))
1067 case MSR_IA32_SYSENTER_EIP:
1068 case MSR_IA32_SYSENTER_ESP:
1070 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1071 * non-canonical address is written on Intel but not on
1072 * AMD (which ignores the top 32-bits, because it does
1073 * not implement 64-bit SYSENTER).
1075 * 64-bit code should hence be able to write a non-canonical
1076 * value on AMD. Making the address canonical ensures that
1077 * vmentry does not fail on Intel after writing a non-canonical
1078 * value, and that something deterministic happens if the guest
1079 * invokes 64-bit SYSENTER.
1081 msr->data = get_canonical(msr->data);
1083 return kvm_x86_ops->set_msr(vcpu, msr);
1085 EXPORT_SYMBOL_GPL(kvm_set_msr);
1088 * Adapt set_msr() to msr_io()'s calling convention
1090 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1092 struct msr_data msr;
1096 msr.host_initiated = true;
1097 r = kvm_get_msr(vcpu, &msr);
1105 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1107 struct msr_data msr;
1111 msr.host_initiated = true;
1112 return kvm_set_msr(vcpu, &msr);
1115 #ifdef CONFIG_X86_64
1116 struct pvclock_gtod_data {
1119 struct { /* extract of a clocksource struct */
1131 static struct pvclock_gtod_data pvclock_gtod_data;
1133 static void update_pvclock_gtod(struct timekeeper *tk)
1135 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1138 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1140 write_seqcount_begin(&vdata->seq);
1142 /* copy pvclock gtod data */
1143 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1144 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1145 vdata->clock.mask = tk->tkr_mono.mask;
1146 vdata->clock.mult = tk->tkr_mono.mult;
1147 vdata->clock.shift = tk->tkr_mono.shift;
1149 vdata->boot_ns = boot_ns;
1150 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1152 write_seqcount_end(&vdata->seq);
1156 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1159 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1160 * vcpu_enter_guest. This function is only called from
1161 * the physical CPU that is running vcpu.
1163 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1166 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1170 struct pvclock_wall_clock wc;
1171 struct timespec64 boot;
1176 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1181 ++version; /* first time write, random junk */
1185 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1189 * The guest calculates current wall clock time by adding
1190 * system time (updated by kvm_guest_time_update below) to the
1191 * wall clock specified here. guest system time equals host
1192 * system time for us, thus we must fill in host boot time here.
1194 getboottime64(&boot);
1196 if (kvm->arch.kvmclock_offset) {
1197 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1198 boot = timespec64_sub(boot, ts);
1200 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1201 wc.nsec = boot.tv_nsec;
1202 wc.version = version;
1204 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1207 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1210 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1212 do_shl32_div32(dividend, divisor);
1216 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1217 s8 *pshift, u32 *pmultiplier)
1225 scaled64 = scaled_hz;
1226 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1231 tps32 = (uint32_t)tps64;
1232 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1233 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1241 *pmultiplier = div_frac(scaled64, tps32);
1243 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1244 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1247 #ifdef CONFIG_X86_64
1248 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1251 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1252 static unsigned long max_tsc_khz;
1254 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
1256 return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
1257 vcpu->arch.virtual_tsc_shift);
1260 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1262 u64 v = (u64)khz * (1000000 + ppm);
1267 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1271 /* Guest TSC same frequency as host TSC? */
1273 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1277 /* TSC scaling supported? */
1278 if (!kvm_has_tsc_control) {
1279 if (user_tsc_khz > tsc_khz) {
1280 vcpu->arch.tsc_catchup = 1;
1281 vcpu->arch.tsc_always_catchup = 1;
1284 WARN(1, "user requested TSC rate below hardware speed\n");
1289 /* TSC scaling required - calculate ratio */
1290 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1291 user_tsc_khz, tsc_khz);
1293 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1294 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1299 vcpu->arch.tsc_scaling_ratio = ratio;
1303 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1305 u32 thresh_lo, thresh_hi;
1306 int use_scaling = 0;
1308 /* tsc_khz can be zero if TSC calibration fails */
1309 if (user_tsc_khz == 0) {
1310 /* set tsc_scaling_ratio to a safe value */
1311 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1315 /* Compute a scale to convert nanoseconds in TSC cycles */
1316 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1317 &vcpu->arch.virtual_tsc_shift,
1318 &vcpu->arch.virtual_tsc_mult);
1319 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1322 * Compute the variation in TSC rate which is acceptable
1323 * within the range of tolerance and decide if the
1324 * rate being applied is within that bounds of the hardware
1325 * rate. If so, no scaling or compensation need be done.
1327 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1328 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1329 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1330 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1333 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1336 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1338 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1339 vcpu->arch.virtual_tsc_mult,
1340 vcpu->arch.virtual_tsc_shift);
1341 tsc += vcpu->arch.this_tsc_write;
1345 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1347 #ifdef CONFIG_X86_64
1349 struct kvm_arch *ka = &vcpu->kvm->arch;
1350 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1352 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1353 atomic_read(&vcpu->kvm->online_vcpus));
1356 * Once the masterclock is enabled, always perform request in
1357 * order to update it.
1359 * In order to enable masterclock, the host clocksource must be TSC
1360 * and the vcpus need to have matched TSCs. When that happens,
1361 * perform request to enable masterclock.
1363 if (ka->use_master_clock ||
1364 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1365 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1367 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1368 atomic_read(&vcpu->kvm->online_vcpus),
1369 ka->use_master_clock, gtod->clock.vclock_mode);
1373 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1375 u64 curr_offset = kvm_x86_ops->read_tsc_offset(vcpu);
1376 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1380 * Multiply tsc by a fixed point number represented by ratio.
1382 * The most significant 64-N bits (mult) of ratio represent the
1383 * integral part of the fixed point number; the remaining N bits
1384 * (frac) represent the fractional part, ie. ratio represents a fixed
1385 * point number (mult + frac * 2^(-N)).
1387 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1389 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1391 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1394 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1397 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1399 if (ratio != kvm_default_tsc_scaling_ratio)
1400 _tsc = __scale_tsc(ratio, tsc);
1404 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1406 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1410 tsc = kvm_scale_tsc(vcpu, rdtsc());
1412 return target_tsc - tsc;
1415 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1417 return kvm_x86_ops->read_l1_tsc(vcpu, kvm_scale_tsc(vcpu, host_tsc));
1419 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1421 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1423 struct kvm *kvm = vcpu->kvm;
1424 u64 offset, ns, elapsed;
1425 unsigned long flags;
1428 bool already_matched;
1429 u64 data = msr->data;
1431 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1432 offset = kvm_compute_tsc_offset(vcpu, data);
1433 ns = get_kernel_ns();
1434 elapsed = ns - kvm->arch.last_tsc_nsec;
1436 if (vcpu->arch.virtual_tsc_khz) {
1439 /* n.b - signed multiplication and division required */
1440 usdiff = data - kvm->arch.last_tsc_write;
1441 #ifdef CONFIG_X86_64
1442 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1444 /* do_div() only does unsigned */
1445 asm("1: idivl %[divisor]\n"
1446 "2: xor %%edx, %%edx\n"
1447 " movl $0, %[faulted]\n"
1449 ".section .fixup,\"ax\"\n"
1450 "4: movl $1, %[faulted]\n"
1454 _ASM_EXTABLE(1b, 4b)
1456 : "=A"(usdiff), [faulted] "=r" (faulted)
1457 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1460 do_div(elapsed, 1000);
1465 /* idivl overflow => difference is larger than USEC_PER_SEC */
1467 usdiff = USEC_PER_SEC;
1469 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1472 * Special case: TSC write with a small delta (1 second) of virtual
1473 * cycle time against real time is interpreted as an attempt to
1474 * synchronize the CPU.
1476 * For a reliable TSC, we can match TSC offsets, and for an unstable
1477 * TSC, we add elapsed time in this computation. We could let the
1478 * compensation code attempt to catch up if we fall behind, but
1479 * it's better to try to match offsets from the beginning.
1481 if (usdiff < USEC_PER_SEC &&
1482 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1483 if (!check_tsc_unstable()) {
1484 offset = kvm->arch.cur_tsc_offset;
1485 pr_debug("kvm: matched tsc offset for %llu\n", data);
1487 u64 delta = nsec_to_cycles(vcpu, elapsed);
1489 offset = kvm_compute_tsc_offset(vcpu, data);
1490 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1493 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1496 * We split periods of matched TSC writes into generations.
1497 * For each generation, we track the original measured
1498 * nanosecond time, offset, and write, so if TSCs are in
1499 * sync, we can match exact offset, and if not, we can match
1500 * exact software computation in compute_guest_tsc()
1502 * These values are tracked in kvm->arch.cur_xxx variables.
1504 kvm->arch.cur_tsc_generation++;
1505 kvm->arch.cur_tsc_nsec = ns;
1506 kvm->arch.cur_tsc_write = data;
1507 kvm->arch.cur_tsc_offset = offset;
1509 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1510 kvm->arch.cur_tsc_generation, data);
1514 * We also track th most recent recorded KHZ, write and time to
1515 * allow the matching interval to be extended at each write.
1517 kvm->arch.last_tsc_nsec = ns;
1518 kvm->arch.last_tsc_write = data;
1519 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1521 vcpu->arch.last_guest_tsc = data;
1523 /* Keep track of which generation this VCPU has synchronized to */
1524 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1525 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1526 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1528 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1529 update_ia32_tsc_adjust_msr(vcpu, offset);
1530 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1531 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1533 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1535 kvm->arch.nr_vcpus_matched_tsc = 0;
1536 } else if (!already_matched) {
1537 kvm->arch.nr_vcpus_matched_tsc++;
1540 kvm_track_tsc_matching(vcpu);
1541 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1544 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1546 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1549 kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1552 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1554 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1555 WARN_ON(adjustment < 0);
1556 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1557 kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
1560 #ifdef CONFIG_X86_64
1562 static cycle_t read_tsc(void)
1564 cycle_t ret = (cycle_t)rdtsc_ordered();
1565 u64 last = pvclock_gtod_data.clock.cycle_last;
1567 if (likely(ret >= last))
1571 * GCC likes to generate cmov here, but this branch is extremely
1572 * predictable (it's just a function of time and the likely is
1573 * very likely) and there's a data dependence, so force GCC
1574 * to generate a branch instead. I don't barrier() because
1575 * we don't actually need a barrier, and if this function
1576 * ever gets inlined it will generate worse code.
1582 static inline u64 vgettsc(cycle_t *cycle_now)
1585 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1587 *cycle_now = read_tsc();
1589 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1590 return v * gtod->clock.mult;
1593 static int do_monotonic_boot(s64 *t, cycle_t *cycle_now)
1595 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1601 seq = read_seqcount_begin(>od->seq);
1602 mode = gtod->clock.vclock_mode;
1603 ns = gtod->nsec_base;
1604 ns += vgettsc(cycle_now);
1605 ns >>= gtod->clock.shift;
1606 ns += gtod->boot_ns;
1607 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1613 /* returns true if host is using tsc clocksource */
1614 static bool kvm_get_time_and_clockread(s64 *kernel_ns, cycle_t *cycle_now)
1616 /* checked again under seqlock below */
1617 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1620 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1626 * Assuming a stable TSC across physical CPUS, and a stable TSC
1627 * across virtual CPUs, the following condition is possible.
1628 * Each numbered line represents an event visible to both
1629 * CPUs at the next numbered event.
1631 * "timespecX" represents host monotonic time. "tscX" represents
1634 * VCPU0 on CPU0 | VCPU1 on CPU1
1636 * 1. read timespec0,tsc0
1637 * 2. | timespec1 = timespec0 + N
1639 * 3. transition to guest | transition to guest
1640 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1641 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1642 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1644 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1647 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1649 * - 0 < N - M => M < N
1651 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1652 * always the case (the difference between two distinct xtime instances
1653 * might be smaller then the difference between corresponding TSC reads,
1654 * when updating guest vcpus pvclock areas).
1656 * To avoid that problem, do not allow visibility of distinct
1657 * system_timestamp/tsc_timestamp values simultaneously: use a master
1658 * copy of host monotonic time values. Update that master copy
1661 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1665 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1667 #ifdef CONFIG_X86_64
1668 struct kvm_arch *ka = &kvm->arch;
1670 bool host_tsc_clocksource, vcpus_matched;
1672 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1673 atomic_read(&kvm->online_vcpus));
1676 * If the host uses TSC clock, then passthrough TSC as stable
1679 host_tsc_clocksource = kvm_get_time_and_clockread(
1680 &ka->master_kernel_ns,
1681 &ka->master_cycle_now);
1683 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1684 && !backwards_tsc_observed
1685 && !ka->boot_vcpu_runs_old_kvmclock;
1687 if (ka->use_master_clock)
1688 atomic_set(&kvm_guest_has_master_clock, 1);
1690 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1691 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1696 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1698 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1701 static void kvm_gen_update_masterclock(struct kvm *kvm)
1703 #ifdef CONFIG_X86_64
1705 struct kvm_vcpu *vcpu;
1706 struct kvm_arch *ka = &kvm->arch;
1708 spin_lock(&ka->pvclock_gtod_sync_lock);
1709 kvm_make_mclock_inprogress_request(kvm);
1710 /* no guest entries from this point */
1711 pvclock_update_vm_gtod_copy(kvm);
1713 kvm_for_each_vcpu(i, vcpu, kvm)
1714 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1716 /* guest entries allowed */
1717 kvm_for_each_vcpu(i, vcpu, kvm)
1718 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1720 spin_unlock(&ka->pvclock_gtod_sync_lock);
1724 static int kvm_guest_time_update(struct kvm_vcpu *v)
1726 unsigned long flags, tgt_tsc_khz;
1727 struct kvm_vcpu_arch *vcpu = &v->arch;
1728 struct kvm_arch *ka = &v->kvm->arch;
1730 u64 tsc_timestamp, host_tsc;
1731 struct pvclock_vcpu_time_info guest_hv_clock;
1733 bool use_master_clock;
1739 * If the host uses TSC clock, then passthrough TSC as stable
1742 spin_lock(&ka->pvclock_gtod_sync_lock);
1743 use_master_clock = ka->use_master_clock;
1744 if (use_master_clock) {
1745 host_tsc = ka->master_cycle_now;
1746 kernel_ns = ka->master_kernel_ns;
1748 spin_unlock(&ka->pvclock_gtod_sync_lock);
1750 /* Keep irq disabled to prevent changes to the clock */
1751 local_irq_save(flags);
1752 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1753 if (unlikely(tgt_tsc_khz == 0)) {
1754 local_irq_restore(flags);
1755 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1758 if (!use_master_clock) {
1760 kernel_ns = get_kernel_ns();
1763 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1766 * We may have to catch up the TSC to match elapsed wall clock
1767 * time for two reasons, even if kvmclock is used.
1768 * 1) CPU could have been running below the maximum TSC rate
1769 * 2) Broken TSC compensation resets the base at each VCPU
1770 * entry to avoid unknown leaps of TSC even when running
1771 * again on the same CPU. This may cause apparent elapsed
1772 * time to disappear, and the guest to stand still or run
1775 if (vcpu->tsc_catchup) {
1776 u64 tsc = compute_guest_tsc(v, kernel_ns);
1777 if (tsc > tsc_timestamp) {
1778 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1779 tsc_timestamp = tsc;
1783 local_irq_restore(flags);
1785 if (!vcpu->pv_time_enabled)
1788 if (kvm_has_tsc_control)
1789 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
1791 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
1792 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
1793 &vcpu->hv_clock.tsc_shift,
1794 &vcpu->hv_clock.tsc_to_system_mul);
1795 vcpu->hw_tsc_khz = tgt_tsc_khz;
1798 /* With all the info we got, fill in the values */
1799 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1800 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1801 vcpu->last_guest_tsc = tsc_timestamp;
1803 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1804 &guest_hv_clock, sizeof(guest_hv_clock))))
1807 /* This VCPU is paused, but it's legal for a guest to read another
1808 * VCPU's kvmclock, so we really have to follow the specification where
1809 * it says that version is odd if data is being modified, and even after
1812 * Version field updates must be kept separate. This is because
1813 * kvm_write_guest_cached might use a "rep movs" instruction, and
1814 * writes within a string instruction are weakly ordered. So there
1815 * are three writes overall.
1817 * As a small optimization, only write the version field in the first
1818 * and third write. The vcpu->pv_time cache is still valid, because the
1819 * version field is the first in the struct.
1821 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1823 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1824 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1826 sizeof(vcpu->hv_clock.version));
1830 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1831 pvclock_flags = (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1833 if (vcpu->pvclock_set_guest_stopped_request) {
1834 pvclock_flags |= PVCLOCK_GUEST_STOPPED;
1835 vcpu->pvclock_set_guest_stopped_request = false;
1838 /* If the host uses TSC clocksource, then it is stable */
1839 if (use_master_clock)
1840 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1842 vcpu->hv_clock.flags = pvclock_flags;
1844 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1846 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1848 sizeof(vcpu->hv_clock));
1852 vcpu->hv_clock.version++;
1853 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1855 sizeof(vcpu->hv_clock.version));
1860 * kvmclock updates which are isolated to a given vcpu, such as
1861 * vcpu->cpu migration, should not allow system_timestamp from
1862 * the rest of the vcpus to remain static. Otherwise ntp frequency
1863 * correction applies to one vcpu's system_timestamp but not
1866 * So in those cases, request a kvmclock update for all vcpus.
1867 * We need to rate-limit these requests though, as they can
1868 * considerably slow guests that have a large number of vcpus.
1869 * The time for a remote vcpu to update its kvmclock is bound
1870 * by the delay we use to rate-limit the updates.
1873 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1875 static void kvmclock_update_fn(struct work_struct *work)
1878 struct delayed_work *dwork = to_delayed_work(work);
1879 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1880 kvmclock_update_work);
1881 struct kvm *kvm = container_of(ka, struct kvm, arch);
1882 struct kvm_vcpu *vcpu;
1884 kvm_for_each_vcpu(i, vcpu, kvm) {
1885 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1886 kvm_vcpu_kick(vcpu);
1890 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1892 struct kvm *kvm = v->kvm;
1894 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1895 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1896 KVMCLOCK_UPDATE_DELAY);
1899 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1901 static void kvmclock_sync_fn(struct work_struct *work)
1903 struct delayed_work *dwork = to_delayed_work(work);
1904 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1905 kvmclock_sync_work);
1906 struct kvm *kvm = container_of(ka, struct kvm, arch);
1908 if (!kvmclock_periodic_sync)
1911 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1912 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1913 KVMCLOCK_SYNC_PERIOD);
1916 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1918 u64 mcg_cap = vcpu->arch.mcg_cap;
1919 unsigned bank_num = mcg_cap & 0xff;
1922 case MSR_IA32_MCG_STATUS:
1923 vcpu->arch.mcg_status = data;
1925 case MSR_IA32_MCG_CTL:
1926 if (!(mcg_cap & MCG_CTL_P))
1928 if (data != 0 && data != ~(u64)0)
1930 vcpu->arch.mcg_ctl = data;
1933 if (msr >= MSR_IA32_MC0_CTL &&
1934 msr < MSR_IA32_MCx_CTL(bank_num)) {
1935 u32 offset = msr - MSR_IA32_MC0_CTL;
1936 /* only 0 or all 1s can be written to IA32_MCi_CTL
1937 * some Linux kernels though clear bit 10 in bank 4 to
1938 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1939 * this to avoid an uncatched #GP in the guest
1941 if ((offset & 0x3) == 0 &&
1942 data != 0 && (data | (1 << 10)) != ~(u64)0)
1944 vcpu->arch.mce_banks[offset] = data;
1952 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1954 struct kvm *kvm = vcpu->kvm;
1955 int lm = is_long_mode(vcpu);
1956 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1957 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1958 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1959 : kvm->arch.xen_hvm_config.blob_size_32;
1960 u32 page_num = data & ~PAGE_MASK;
1961 u64 page_addr = data & PAGE_MASK;
1966 if (page_num >= blob_size)
1969 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1974 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
1983 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1985 gpa_t gpa = data & ~0x3f;
1987 /* Bits 2:5 are reserved, Should be zero */
1991 vcpu->arch.apf.msr_val = data;
1993 if (!(data & KVM_ASYNC_PF_ENABLED)) {
1994 kvm_clear_async_pf_completion_queue(vcpu);
1995 kvm_async_pf_hash_reset(vcpu);
1999 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2003 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2004 kvm_async_pf_wakeup_all(vcpu);
2008 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2010 vcpu->arch.pv_time_enabled = false;
2013 static void record_steal_time(struct kvm_vcpu *vcpu)
2015 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2018 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2019 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2022 if (vcpu->arch.st.steal.version & 1)
2023 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2025 vcpu->arch.st.steal.version += 1;
2027 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2028 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2032 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2033 vcpu->arch.st.last_steal;
2034 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2036 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2037 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2041 vcpu->arch.st.steal.version += 1;
2043 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2044 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2047 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2050 u32 msr = msr_info->index;
2051 u64 data = msr_info->data;
2054 case MSR_AMD64_NB_CFG:
2055 case MSR_IA32_UCODE_REV:
2056 case MSR_IA32_UCODE_WRITE:
2057 case MSR_VM_HSAVE_PA:
2058 case MSR_AMD64_PATCH_LOADER:
2059 case MSR_AMD64_BU_CFG2:
2063 return set_efer(vcpu, data);
2065 data &= ~(u64)0x40; /* ignore flush filter disable */
2066 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2067 data &= ~(u64)0x8; /* ignore TLB cache disable */
2068 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2070 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2075 case MSR_FAM10H_MMIO_CONF_BASE:
2077 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2082 case MSR_IA32_DEBUGCTLMSR:
2084 /* We support the non-activated case already */
2086 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2087 /* Values other than LBR and BTF are vendor-specific,
2088 thus reserved and should throw a #GP */
2091 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2094 case 0x200 ... 0x2ff:
2095 return kvm_mtrr_set_msr(vcpu, msr, data);
2096 case MSR_IA32_APICBASE:
2097 return kvm_set_apic_base(vcpu, msr_info);
2098 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2099 return kvm_x2apic_msr_write(vcpu, msr, data);
2100 case MSR_IA32_TSCDEADLINE:
2101 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2103 case MSR_IA32_TSC_ADJUST:
2104 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2105 if (!msr_info->host_initiated) {
2106 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2107 adjust_tsc_offset_guest(vcpu, adj);
2109 vcpu->arch.ia32_tsc_adjust_msr = data;
2112 case MSR_IA32_MISC_ENABLE:
2113 vcpu->arch.ia32_misc_enable_msr = data;
2115 case MSR_IA32_SMBASE:
2116 if (!msr_info->host_initiated)
2118 vcpu->arch.smbase = data;
2120 case MSR_KVM_WALL_CLOCK_NEW:
2121 case MSR_KVM_WALL_CLOCK:
2122 vcpu->kvm->arch.wall_clock = data;
2123 kvm_write_wall_clock(vcpu->kvm, data);
2125 case MSR_KVM_SYSTEM_TIME_NEW:
2126 case MSR_KVM_SYSTEM_TIME: {
2128 struct kvm_arch *ka = &vcpu->kvm->arch;
2130 kvmclock_reset(vcpu);
2132 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2133 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2135 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2136 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2139 ka->boot_vcpu_runs_old_kvmclock = tmp;
2142 vcpu->arch.time = data;
2143 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2145 /* we verify if the enable bit is set... */
2149 gpa_offset = data & ~(PAGE_MASK | 1);
2151 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2152 &vcpu->arch.pv_time, data & ~1ULL,
2153 sizeof(struct pvclock_vcpu_time_info)))
2154 vcpu->arch.pv_time_enabled = false;
2156 vcpu->arch.pv_time_enabled = true;
2160 case MSR_KVM_ASYNC_PF_EN:
2161 if (kvm_pv_enable_async_pf(vcpu, data))
2164 case MSR_KVM_STEAL_TIME:
2166 if (unlikely(!sched_info_on()))
2169 if (data & KVM_STEAL_RESERVED_MASK)
2172 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2173 data & KVM_STEAL_VALID_BITS,
2174 sizeof(struct kvm_steal_time)))
2177 vcpu->arch.st.msr_val = data;
2179 if (!(data & KVM_MSR_ENABLED))
2182 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2185 case MSR_KVM_PV_EOI_EN:
2186 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2190 case MSR_IA32_MCG_CTL:
2191 case MSR_IA32_MCG_STATUS:
2192 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2193 return set_msr_mce(vcpu, msr, data);
2195 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2196 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2197 pr = true; /* fall through */
2198 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2199 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2200 if (kvm_pmu_is_valid_msr(vcpu, msr))
2201 return kvm_pmu_set_msr(vcpu, msr_info);
2203 if (pr || data != 0)
2204 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2205 "0x%x data 0x%llx\n", msr, data);
2207 case MSR_K7_CLK_CTL:
2209 * Ignore all writes to this no longer documented MSR.
2210 * Writes are only relevant for old K7 processors,
2211 * all pre-dating SVM, but a recommended workaround from
2212 * AMD for these chips. It is possible to specify the
2213 * affected processor models on the command line, hence
2214 * the need to ignore the workaround.
2217 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2218 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2219 case HV_X64_MSR_CRASH_CTL:
2220 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2221 return kvm_hv_set_msr_common(vcpu, msr, data,
2222 msr_info->host_initiated);
2223 case MSR_IA32_BBL_CR_CTL3:
2224 /* Drop writes to this legacy MSR -- see rdmsr
2225 * counterpart for further detail.
2227 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
2229 case MSR_AMD64_OSVW_ID_LENGTH:
2230 if (!guest_cpuid_has_osvw(vcpu))
2232 vcpu->arch.osvw.length = data;
2234 case MSR_AMD64_OSVW_STATUS:
2235 if (!guest_cpuid_has_osvw(vcpu))
2237 vcpu->arch.osvw.status = data;
2240 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2241 return xen_hvm_config(vcpu, data);
2242 if (kvm_pmu_is_valid_msr(vcpu, msr))
2243 return kvm_pmu_set_msr(vcpu, msr_info);
2245 vcpu_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
2249 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
2256 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2260 * Reads an msr value (of 'msr_index') into 'pdata'.
2261 * Returns 0 on success, non-0 otherwise.
2262 * Assumes vcpu_load() was already called.
2264 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2266 return kvm_x86_ops->get_msr(vcpu, msr);
2268 EXPORT_SYMBOL_GPL(kvm_get_msr);
2270 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2273 u64 mcg_cap = vcpu->arch.mcg_cap;
2274 unsigned bank_num = mcg_cap & 0xff;
2277 case MSR_IA32_P5_MC_ADDR:
2278 case MSR_IA32_P5_MC_TYPE:
2281 case MSR_IA32_MCG_CAP:
2282 data = vcpu->arch.mcg_cap;
2284 case MSR_IA32_MCG_CTL:
2285 if (!(mcg_cap & MCG_CTL_P))
2287 data = vcpu->arch.mcg_ctl;
2289 case MSR_IA32_MCG_STATUS:
2290 data = vcpu->arch.mcg_status;
2293 if (msr >= MSR_IA32_MC0_CTL &&
2294 msr < MSR_IA32_MCx_CTL(bank_num)) {
2295 u32 offset = msr - MSR_IA32_MC0_CTL;
2296 data = vcpu->arch.mce_banks[offset];
2305 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2307 switch (msr_info->index) {
2308 case MSR_IA32_PLATFORM_ID:
2309 case MSR_IA32_EBL_CR_POWERON:
2310 case MSR_IA32_DEBUGCTLMSR:
2311 case MSR_IA32_LASTBRANCHFROMIP:
2312 case MSR_IA32_LASTBRANCHTOIP:
2313 case MSR_IA32_LASTINTFROMIP:
2314 case MSR_IA32_LASTINTTOIP:
2316 case MSR_K8_TSEG_ADDR:
2317 case MSR_K8_TSEG_MASK:
2319 case MSR_VM_HSAVE_PA:
2320 case MSR_K8_INT_PENDING_MSG:
2321 case MSR_AMD64_NB_CFG:
2322 case MSR_FAM10H_MMIO_CONF_BASE:
2323 case MSR_AMD64_BU_CFG2:
2324 case MSR_IA32_PERF_CTL:
2327 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2328 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2329 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2330 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2331 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2332 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2335 case MSR_IA32_UCODE_REV:
2336 msr_info->data = 0x100000000ULL;
2339 case 0x200 ... 0x2ff:
2340 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2341 case 0xcd: /* fsb frequency */
2345 * MSR_EBC_FREQUENCY_ID
2346 * Conservative value valid for even the basic CPU models.
2347 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2348 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2349 * and 266MHz for model 3, or 4. Set Core Clock
2350 * Frequency to System Bus Frequency Ratio to 1 (bits
2351 * 31:24) even though these are only valid for CPU
2352 * models > 2, however guests may end up dividing or
2353 * multiplying by zero otherwise.
2355 case MSR_EBC_FREQUENCY_ID:
2356 msr_info->data = 1 << 24;
2358 case MSR_IA32_APICBASE:
2359 msr_info->data = kvm_get_apic_base(vcpu);
2361 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2362 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2364 case MSR_IA32_TSCDEADLINE:
2365 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2367 case MSR_IA32_TSC_ADJUST:
2368 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2370 case MSR_IA32_MISC_ENABLE:
2371 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2373 case MSR_IA32_SMBASE:
2374 if (!msr_info->host_initiated)
2376 msr_info->data = vcpu->arch.smbase;
2378 case MSR_IA32_PERF_STATUS:
2379 /* TSC increment by tick */
2380 msr_info->data = 1000ULL;
2381 /* CPU multiplier */
2382 msr_info->data |= (((uint64_t)4ULL) << 40);
2385 msr_info->data = vcpu->arch.efer;
2387 case MSR_KVM_WALL_CLOCK:
2388 case MSR_KVM_WALL_CLOCK_NEW:
2389 msr_info->data = vcpu->kvm->arch.wall_clock;
2391 case MSR_KVM_SYSTEM_TIME:
2392 case MSR_KVM_SYSTEM_TIME_NEW:
2393 msr_info->data = vcpu->arch.time;
2395 case MSR_KVM_ASYNC_PF_EN:
2396 msr_info->data = vcpu->arch.apf.msr_val;
2398 case MSR_KVM_STEAL_TIME:
2399 msr_info->data = vcpu->arch.st.msr_val;
2401 case MSR_KVM_PV_EOI_EN:
2402 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2404 case MSR_IA32_P5_MC_ADDR:
2405 case MSR_IA32_P5_MC_TYPE:
2406 case MSR_IA32_MCG_CAP:
2407 case MSR_IA32_MCG_CTL:
2408 case MSR_IA32_MCG_STATUS:
2409 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2410 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2411 case MSR_K7_CLK_CTL:
2413 * Provide expected ramp-up count for K7. All other
2414 * are set to zero, indicating minimum divisors for
2417 * This prevents guest kernels on AMD host with CPU
2418 * type 6, model 8 and higher from exploding due to
2419 * the rdmsr failing.
2421 msr_info->data = 0x20000000;
2423 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2424 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2425 case HV_X64_MSR_CRASH_CTL:
2426 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2427 return kvm_hv_get_msr_common(vcpu,
2428 msr_info->index, &msr_info->data);
2430 case MSR_IA32_BBL_CR_CTL3:
2431 /* This legacy MSR exists but isn't fully documented in current
2432 * silicon. It is however accessed by winxp in very narrow
2433 * scenarios where it sets bit #19, itself documented as
2434 * a "reserved" bit. Best effort attempt to source coherent
2435 * read data here should the balance of the register be
2436 * interpreted by the guest:
2438 * L2 cache control register 3: 64GB range, 256KB size,
2439 * enabled, latency 0x1, configured
2441 msr_info->data = 0xbe702111;
2443 case MSR_AMD64_OSVW_ID_LENGTH:
2444 if (!guest_cpuid_has_osvw(vcpu))
2446 msr_info->data = vcpu->arch.osvw.length;
2448 case MSR_AMD64_OSVW_STATUS:
2449 if (!guest_cpuid_has_osvw(vcpu))
2451 msr_info->data = vcpu->arch.osvw.status;
2454 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2455 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2457 vcpu_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr_info->index);
2460 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2467 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2470 * Read or write a bunch of msrs. All parameters are kernel addresses.
2472 * @return number of msrs set successfully.
2474 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2475 struct kvm_msr_entry *entries,
2476 int (*do_msr)(struct kvm_vcpu *vcpu,
2477 unsigned index, u64 *data))
2481 idx = srcu_read_lock(&vcpu->kvm->srcu);
2482 for (i = 0; i < msrs->nmsrs; ++i)
2483 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2485 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2491 * Read or write a bunch of msrs. Parameters are user addresses.
2493 * @return number of msrs set successfully.
2495 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2496 int (*do_msr)(struct kvm_vcpu *vcpu,
2497 unsigned index, u64 *data),
2500 struct kvm_msrs msrs;
2501 struct kvm_msr_entry *entries;
2506 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2510 if (msrs.nmsrs >= MAX_IO_MSRS)
2513 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2514 entries = memdup_user(user_msrs->entries, size);
2515 if (IS_ERR(entries)) {
2516 r = PTR_ERR(entries);
2520 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2525 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2536 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2541 case KVM_CAP_IRQCHIP:
2543 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2544 case KVM_CAP_SET_TSS_ADDR:
2545 case KVM_CAP_EXT_CPUID:
2546 case KVM_CAP_EXT_EMUL_CPUID:
2547 case KVM_CAP_CLOCKSOURCE:
2549 case KVM_CAP_NOP_IO_DELAY:
2550 case KVM_CAP_MP_STATE:
2551 case KVM_CAP_SYNC_MMU:
2552 case KVM_CAP_USER_NMI:
2553 case KVM_CAP_REINJECT_CONTROL:
2554 case KVM_CAP_IRQ_INJECT_STATUS:
2555 case KVM_CAP_IOEVENTFD:
2556 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2558 case KVM_CAP_PIT_STATE2:
2559 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2560 case KVM_CAP_XEN_HVM:
2561 case KVM_CAP_ADJUST_CLOCK:
2562 case KVM_CAP_VCPU_EVENTS:
2563 case KVM_CAP_HYPERV:
2564 case KVM_CAP_HYPERV_VAPIC:
2565 case KVM_CAP_HYPERV_SPIN:
2566 case KVM_CAP_HYPERV_SYNIC:
2567 case KVM_CAP_PCI_SEGMENT:
2568 case KVM_CAP_DEBUGREGS:
2569 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2571 case KVM_CAP_ASYNC_PF:
2572 case KVM_CAP_GET_TSC_KHZ:
2573 case KVM_CAP_KVMCLOCK_CTRL:
2574 case KVM_CAP_READONLY_MEM:
2575 case KVM_CAP_HYPERV_TIME:
2576 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2577 case KVM_CAP_TSC_DEADLINE_TIMER:
2578 case KVM_CAP_ENABLE_CAP_VM:
2579 case KVM_CAP_DISABLE_QUIRKS:
2580 case KVM_CAP_SET_BOOT_CPU_ID:
2581 case KVM_CAP_SPLIT_IRQCHIP:
2582 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2583 case KVM_CAP_ASSIGN_DEV_IRQ:
2584 case KVM_CAP_PCI_2_3:
2588 case KVM_CAP_X86_SMM:
2589 /* SMBASE is usually relocated above 1M on modern chipsets,
2590 * and SMM handlers might indeed rely on 4G segment limits,
2591 * so do not report SMM to be available if real mode is
2592 * emulated via vm86 mode. Still, do not go to great lengths
2593 * to avoid userspace's usage of the feature, because it is a
2594 * fringe case that is not enabled except via specific settings
2595 * of the module parameters.
2597 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2599 case KVM_CAP_COALESCED_MMIO:
2600 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2603 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2605 case KVM_CAP_NR_VCPUS:
2606 r = KVM_SOFT_MAX_VCPUS;
2608 case KVM_CAP_MAX_VCPUS:
2611 case KVM_CAP_NR_MEMSLOTS:
2612 r = KVM_USER_MEM_SLOTS;
2614 case KVM_CAP_PV_MMU: /* obsolete */
2617 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
2619 r = iommu_present(&pci_bus_type);
2623 r = KVM_MAX_MCE_BANKS;
2626 r = boot_cpu_has(X86_FEATURE_XSAVE);
2628 case KVM_CAP_TSC_CONTROL:
2629 r = kvm_has_tsc_control;
2631 case KVM_CAP_X2APIC_API:
2632 r = KVM_X2APIC_API_VALID_FLAGS;
2642 long kvm_arch_dev_ioctl(struct file *filp,
2643 unsigned int ioctl, unsigned long arg)
2645 void __user *argp = (void __user *)arg;
2649 case KVM_GET_MSR_INDEX_LIST: {
2650 struct kvm_msr_list __user *user_msr_list = argp;
2651 struct kvm_msr_list msr_list;
2655 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2658 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2659 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2662 if (n < msr_list.nmsrs)
2665 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2666 num_msrs_to_save * sizeof(u32)))
2668 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2670 num_emulated_msrs * sizeof(u32)))
2675 case KVM_GET_SUPPORTED_CPUID:
2676 case KVM_GET_EMULATED_CPUID: {
2677 struct kvm_cpuid2 __user *cpuid_arg = argp;
2678 struct kvm_cpuid2 cpuid;
2681 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2684 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2690 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2695 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2697 if (copy_to_user(argp, &kvm_mce_cap_supported,
2698 sizeof(kvm_mce_cap_supported)))
2710 static void wbinvd_ipi(void *garbage)
2715 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2717 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2720 static inline void kvm_migrate_timers(struct kvm_vcpu *vcpu)
2722 set_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests);
2725 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2727 /* Address WBINVD may be executed by guest */
2728 if (need_emulate_wbinvd(vcpu)) {
2729 if (kvm_x86_ops->has_wbinvd_exit())
2730 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2731 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2732 smp_call_function_single(vcpu->cpu,
2733 wbinvd_ipi, NULL, 1);
2736 kvm_x86_ops->vcpu_load(vcpu, cpu);
2738 /* Apply any externally detected TSC adjustments (due to suspend) */
2739 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2740 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2741 vcpu->arch.tsc_offset_adjustment = 0;
2742 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2745 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2746 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2747 rdtsc() - vcpu->arch.last_host_tsc;
2749 mark_tsc_unstable("KVM discovered backwards TSC");
2751 if (kvm_lapic_hv_timer_in_use(vcpu) &&
2752 kvm_x86_ops->set_hv_timer(vcpu,
2753 kvm_get_lapic_tscdeadline_msr(vcpu)))
2754 kvm_lapic_switch_to_sw_timer(vcpu);
2755 if (check_tsc_unstable()) {
2756 u64 offset = kvm_compute_tsc_offset(vcpu,
2757 vcpu->arch.last_guest_tsc);
2758 kvm_x86_ops->write_tsc_offset(vcpu, offset);
2759 vcpu->arch.tsc_catchup = 1;
2762 * On a host with synchronized TSC, there is no need to update
2763 * kvmclock on vcpu->cpu migration
2765 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2766 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2767 if (vcpu->cpu != cpu)
2768 kvm_migrate_timers(vcpu);
2772 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2775 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2777 kvm_x86_ops->vcpu_put(vcpu);
2778 kvm_put_guest_fpu(vcpu);
2779 vcpu->arch.last_host_tsc = rdtsc();
2782 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2783 struct kvm_lapic_state *s)
2785 if (vcpu->arch.apicv_active)
2786 kvm_x86_ops->sync_pir_to_irr(vcpu);
2788 return kvm_apic_get_state(vcpu, s);
2791 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2792 struct kvm_lapic_state *s)
2796 r = kvm_apic_set_state(vcpu, s);
2799 update_cr8_intercept(vcpu);
2804 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2806 return (!lapic_in_kernel(vcpu) ||
2807 kvm_apic_accept_pic_intr(vcpu));
2811 * if userspace requested an interrupt window, check that the
2812 * interrupt window is open.
2814 * No need to exit to userspace if we already have an interrupt queued.
2816 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2818 return kvm_arch_interrupt_allowed(vcpu) &&
2819 !kvm_cpu_has_interrupt(vcpu) &&
2820 !kvm_event_needs_reinjection(vcpu) &&
2821 kvm_cpu_accept_dm_intr(vcpu);
2824 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2825 struct kvm_interrupt *irq)
2827 if (irq->irq >= KVM_NR_INTERRUPTS)
2830 if (!irqchip_in_kernel(vcpu->kvm)) {
2831 kvm_queue_interrupt(vcpu, irq->irq, false);
2832 kvm_make_request(KVM_REQ_EVENT, vcpu);
2837 * With in-kernel LAPIC, we only use this to inject EXTINT, so
2838 * fail for in-kernel 8259.
2840 if (pic_in_kernel(vcpu->kvm))
2843 if (vcpu->arch.pending_external_vector != -1)
2846 vcpu->arch.pending_external_vector = irq->irq;
2847 kvm_make_request(KVM_REQ_EVENT, vcpu);
2851 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2853 kvm_inject_nmi(vcpu);
2858 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2860 kvm_make_request(KVM_REQ_SMI, vcpu);
2865 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2866 struct kvm_tpr_access_ctl *tac)
2870 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2874 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2878 unsigned bank_num = mcg_cap & 0xff, bank;
2881 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2883 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
2886 vcpu->arch.mcg_cap = mcg_cap;
2887 /* Init IA32_MCG_CTL to all 1s */
2888 if (mcg_cap & MCG_CTL_P)
2889 vcpu->arch.mcg_ctl = ~(u64)0;
2890 /* Init IA32_MCi_CTL to all 1s */
2891 for (bank = 0; bank < bank_num; bank++)
2892 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2894 if (kvm_x86_ops->setup_mce)
2895 kvm_x86_ops->setup_mce(vcpu);
2900 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2901 struct kvm_x86_mce *mce)
2903 u64 mcg_cap = vcpu->arch.mcg_cap;
2904 unsigned bank_num = mcg_cap & 0xff;
2905 u64 *banks = vcpu->arch.mce_banks;
2907 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2910 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2911 * reporting is disabled
2913 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2914 vcpu->arch.mcg_ctl != ~(u64)0)
2916 banks += 4 * mce->bank;
2918 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2919 * reporting is disabled for the bank
2921 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2923 if (mce->status & MCI_STATUS_UC) {
2924 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2925 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2926 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2929 if (banks[1] & MCI_STATUS_VAL)
2930 mce->status |= MCI_STATUS_OVER;
2931 banks[2] = mce->addr;
2932 banks[3] = mce->misc;
2933 vcpu->arch.mcg_status = mce->mcg_status;
2934 banks[1] = mce->status;
2935 kvm_queue_exception(vcpu, MC_VECTOR);
2936 } else if (!(banks[1] & MCI_STATUS_VAL)
2937 || !(banks[1] & MCI_STATUS_UC)) {
2938 if (banks[1] & MCI_STATUS_VAL)
2939 mce->status |= MCI_STATUS_OVER;
2940 banks[2] = mce->addr;
2941 banks[3] = mce->misc;
2942 banks[1] = mce->status;
2944 banks[1] |= MCI_STATUS_OVER;
2948 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2949 struct kvm_vcpu_events *events)
2952 events->exception.injected =
2953 vcpu->arch.exception.pending &&
2954 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2955 events->exception.nr = vcpu->arch.exception.nr;
2956 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2957 events->exception.pad = 0;
2958 events->exception.error_code = vcpu->arch.exception.error_code;
2960 events->interrupt.injected =
2961 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2962 events->interrupt.nr = vcpu->arch.interrupt.nr;
2963 events->interrupt.soft = 0;
2964 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
2966 events->nmi.injected = vcpu->arch.nmi_injected;
2967 events->nmi.pending = vcpu->arch.nmi_pending != 0;
2968 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2969 events->nmi.pad = 0;
2971 events->sipi_vector = 0; /* never valid when reporting to user space */
2973 events->smi.smm = is_smm(vcpu);
2974 events->smi.pending = vcpu->arch.smi_pending;
2975 events->smi.smm_inside_nmi =
2976 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
2977 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
2979 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2980 | KVM_VCPUEVENT_VALID_SHADOW
2981 | KVM_VCPUEVENT_VALID_SMM);
2982 memset(&events->reserved, 0, sizeof(events->reserved));
2985 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2986 struct kvm_vcpu_events *events)
2988 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2989 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2990 | KVM_VCPUEVENT_VALID_SHADOW
2991 | KVM_VCPUEVENT_VALID_SMM))
2994 if (events->exception.injected &&
2995 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
2999 vcpu->arch.exception.pending = events->exception.injected;
3000 vcpu->arch.exception.nr = events->exception.nr;
3001 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3002 vcpu->arch.exception.error_code = events->exception.error_code;
3004 vcpu->arch.interrupt.pending = events->interrupt.injected;
3005 vcpu->arch.interrupt.nr = events->interrupt.nr;
3006 vcpu->arch.interrupt.soft = events->interrupt.soft;
3007 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3008 kvm_x86_ops->set_interrupt_shadow(vcpu,
3009 events->interrupt.shadow);
3011 vcpu->arch.nmi_injected = events->nmi.injected;
3012 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3013 vcpu->arch.nmi_pending = events->nmi.pending;
3014 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3016 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3017 lapic_in_kernel(vcpu))
3018 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3020 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3021 if (events->smi.smm)
3022 vcpu->arch.hflags |= HF_SMM_MASK;
3024 vcpu->arch.hflags &= ~HF_SMM_MASK;
3025 vcpu->arch.smi_pending = events->smi.pending;
3026 if (events->smi.smm_inside_nmi)
3027 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3029 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3030 if (lapic_in_kernel(vcpu)) {
3031 if (events->smi.latched_init)
3032 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3034 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3038 kvm_make_request(KVM_REQ_EVENT, vcpu);
3043 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3044 struct kvm_debugregs *dbgregs)
3048 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3049 kvm_get_dr(vcpu, 6, &val);
3051 dbgregs->dr7 = vcpu->arch.dr7;
3053 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3056 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3057 struct kvm_debugregs *dbgregs)
3062 if (dbgregs->dr6 & ~0xffffffffull)
3064 if (dbgregs->dr7 & ~0xffffffffull)
3067 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3068 kvm_update_dr0123(vcpu);
3069 vcpu->arch.dr6 = dbgregs->dr6;
3070 kvm_update_dr6(vcpu);
3071 vcpu->arch.dr7 = dbgregs->dr7;
3072 kvm_update_dr7(vcpu);
3077 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3079 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3081 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3082 u64 xstate_bv = xsave->header.xfeatures;
3086 * Copy legacy XSAVE area, to avoid complications with CPUID
3087 * leaves 0 and 1 in the loop below.
3089 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3092 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3095 * Copy each region from the possibly compacted offset to the
3096 * non-compacted offset.
3098 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3100 u64 feature = valid & -valid;
3101 int index = fls64(feature) - 1;
3102 void *src = get_xsave_addr(xsave, feature);
3105 u32 size, offset, ecx, edx;
3106 cpuid_count(XSTATE_CPUID, index,
3107 &size, &offset, &ecx, &edx);
3108 memcpy(dest + offset, src, size);
3115 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3117 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3118 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3122 * Copy legacy XSAVE area, to avoid complications with CPUID
3123 * leaves 0 and 1 in the loop below.
3125 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3127 /* Set XSTATE_BV and possibly XCOMP_BV. */
3128 xsave->header.xfeatures = xstate_bv;
3129 if (boot_cpu_has(X86_FEATURE_XSAVES))
3130 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3133 * Copy each region from the non-compacted offset to the
3134 * possibly compacted offset.
3136 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3138 u64 feature = valid & -valid;
3139 int index = fls64(feature) - 1;
3140 void *dest = get_xsave_addr(xsave, feature);
3143 u32 size, offset, ecx, edx;
3144 cpuid_count(XSTATE_CPUID, index,
3145 &size, &offset, &ecx, &edx);
3146 memcpy(dest, src + offset, size);
3153 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3154 struct kvm_xsave *guest_xsave)
3156 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3157 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3158 fill_xsave((u8 *) guest_xsave->region, vcpu);
3160 memcpy(guest_xsave->region,
3161 &vcpu->arch.guest_fpu.state.fxsave,
3162 sizeof(struct fxregs_state));
3163 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3164 XFEATURE_MASK_FPSSE;
3168 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3169 struct kvm_xsave *guest_xsave)
3172 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3174 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3176 * Here we allow setting states that are not present in
3177 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3178 * with old userspace.
3180 if (xstate_bv & ~kvm_supported_xcr0())
3182 load_xsave(vcpu, (u8 *)guest_xsave->region);
3184 if (xstate_bv & ~XFEATURE_MASK_FPSSE)
3186 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3187 guest_xsave->region, sizeof(struct fxregs_state));
3192 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3193 struct kvm_xcrs *guest_xcrs)
3195 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3196 guest_xcrs->nr_xcrs = 0;
3200 guest_xcrs->nr_xcrs = 1;
3201 guest_xcrs->flags = 0;
3202 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3203 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3206 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3207 struct kvm_xcrs *guest_xcrs)
3211 if (!boot_cpu_has(X86_FEATURE_XSAVE))
3214 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3217 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3218 /* Only support XCR0 currently */
3219 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3220 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3221 guest_xcrs->xcrs[i].value);
3230 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3231 * stopped by the hypervisor. This function will be called from the host only.
3232 * EINVAL is returned when the host attempts to set the flag for a guest that
3233 * does not support pv clocks.
3235 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3237 if (!vcpu->arch.pv_time_enabled)
3239 vcpu->arch.pvclock_set_guest_stopped_request = true;
3240 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3244 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3245 struct kvm_enable_cap *cap)
3251 case KVM_CAP_HYPERV_SYNIC:
3252 return kvm_hv_activate_synic(vcpu);
3258 long kvm_arch_vcpu_ioctl(struct file *filp,
3259 unsigned int ioctl, unsigned long arg)
3261 struct kvm_vcpu *vcpu = filp->private_data;
3262 void __user *argp = (void __user *)arg;
3265 struct kvm_lapic_state *lapic;
3266 struct kvm_xsave *xsave;
3267 struct kvm_xcrs *xcrs;
3273 case KVM_GET_LAPIC: {
3275 if (!lapic_in_kernel(vcpu))
3277 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3282 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3286 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3291 case KVM_SET_LAPIC: {
3293 if (!lapic_in_kernel(vcpu))
3295 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3296 if (IS_ERR(u.lapic))
3297 return PTR_ERR(u.lapic);
3299 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3302 case KVM_INTERRUPT: {
3303 struct kvm_interrupt irq;
3306 if (copy_from_user(&irq, argp, sizeof irq))
3308 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3312 r = kvm_vcpu_ioctl_nmi(vcpu);
3316 r = kvm_vcpu_ioctl_smi(vcpu);
3319 case KVM_SET_CPUID: {
3320 struct kvm_cpuid __user *cpuid_arg = argp;
3321 struct kvm_cpuid cpuid;
3324 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3326 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3329 case KVM_SET_CPUID2: {
3330 struct kvm_cpuid2 __user *cpuid_arg = argp;
3331 struct kvm_cpuid2 cpuid;
3334 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3336 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3337 cpuid_arg->entries);
3340 case KVM_GET_CPUID2: {
3341 struct kvm_cpuid2 __user *cpuid_arg = argp;
3342 struct kvm_cpuid2 cpuid;
3345 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3347 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3348 cpuid_arg->entries);
3352 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3358 r = msr_io(vcpu, argp, do_get_msr, 1);
3361 r = msr_io(vcpu, argp, do_set_msr, 0);
3363 case KVM_TPR_ACCESS_REPORTING: {
3364 struct kvm_tpr_access_ctl tac;
3367 if (copy_from_user(&tac, argp, sizeof tac))
3369 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3373 if (copy_to_user(argp, &tac, sizeof tac))
3378 case KVM_SET_VAPIC_ADDR: {
3379 struct kvm_vapic_addr va;
3382 if (!lapic_in_kernel(vcpu))
3385 if (copy_from_user(&va, argp, sizeof va))
3387 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3390 case KVM_X86_SETUP_MCE: {
3394 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3396 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3399 case KVM_X86_SET_MCE: {
3400 struct kvm_x86_mce mce;
3403 if (copy_from_user(&mce, argp, sizeof mce))
3405 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3408 case KVM_GET_VCPU_EVENTS: {
3409 struct kvm_vcpu_events events;
3411 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3414 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3419 case KVM_SET_VCPU_EVENTS: {
3420 struct kvm_vcpu_events events;
3423 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3426 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3429 case KVM_GET_DEBUGREGS: {
3430 struct kvm_debugregs dbgregs;
3432 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3435 if (copy_to_user(argp, &dbgregs,
3436 sizeof(struct kvm_debugregs)))
3441 case KVM_SET_DEBUGREGS: {
3442 struct kvm_debugregs dbgregs;
3445 if (copy_from_user(&dbgregs, argp,
3446 sizeof(struct kvm_debugregs)))
3449 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3452 case KVM_GET_XSAVE: {
3453 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3458 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3461 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3466 case KVM_SET_XSAVE: {
3467 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3468 if (IS_ERR(u.xsave))
3469 return PTR_ERR(u.xsave);
3471 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3474 case KVM_GET_XCRS: {
3475 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3480 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3483 if (copy_to_user(argp, u.xcrs,
3484 sizeof(struct kvm_xcrs)))
3489 case KVM_SET_XCRS: {
3490 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3492 return PTR_ERR(u.xcrs);
3494 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3497 case KVM_SET_TSC_KHZ: {
3501 user_tsc_khz = (u32)arg;
3503 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3506 if (user_tsc_khz == 0)
3507 user_tsc_khz = tsc_khz;
3509 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3514 case KVM_GET_TSC_KHZ: {
3515 r = vcpu->arch.virtual_tsc_khz;
3518 case KVM_KVMCLOCK_CTRL: {
3519 r = kvm_set_guest_paused(vcpu);
3522 case KVM_ENABLE_CAP: {
3523 struct kvm_enable_cap cap;
3526 if (copy_from_user(&cap, argp, sizeof(cap)))
3528 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3539 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3541 return VM_FAULT_SIGBUS;
3544 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3548 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3550 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3554 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3557 kvm->arch.ept_identity_map_addr = ident_addr;
3561 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3562 u32 kvm_nr_mmu_pages)
3564 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3567 mutex_lock(&kvm->slots_lock);
3569 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3570 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3572 mutex_unlock(&kvm->slots_lock);
3576 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3578 return kvm->arch.n_max_mmu_pages;
3581 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3586 switch (chip->chip_id) {
3587 case KVM_IRQCHIP_PIC_MASTER:
3588 memcpy(&chip->chip.pic,
3589 &pic_irqchip(kvm)->pics[0],
3590 sizeof(struct kvm_pic_state));
3592 case KVM_IRQCHIP_PIC_SLAVE:
3593 memcpy(&chip->chip.pic,
3594 &pic_irqchip(kvm)->pics[1],
3595 sizeof(struct kvm_pic_state));
3597 case KVM_IRQCHIP_IOAPIC:
3598 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3607 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3612 switch (chip->chip_id) {
3613 case KVM_IRQCHIP_PIC_MASTER:
3614 spin_lock(&pic_irqchip(kvm)->lock);
3615 memcpy(&pic_irqchip(kvm)->pics[0],
3617 sizeof(struct kvm_pic_state));
3618 spin_unlock(&pic_irqchip(kvm)->lock);
3620 case KVM_IRQCHIP_PIC_SLAVE:
3621 spin_lock(&pic_irqchip(kvm)->lock);
3622 memcpy(&pic_irqchip(kvm)->pics[1],
3624 sizeof(struct kvm_pic_state));
3625 spin_unlock(&pic_irqchip(kvm)->lock);
3627 case KVM_IRQCHIP_IOAPIC:
3628 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3634 kvm_pic_update_irq(pic_irqchip(kvm));
3638 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3640 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3642 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3644 mutex_lock(&kps->lock);
3645 memcpy(ps, &kps->channels, sizeof(*ps));
3646 mutex_unlock(&kps->lock);
3650 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3653 struct kvm_pit *pit = kvm->arch.vpit;
3655 mutex_lock(&pit->pit_state.lock);
3656 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
3657 for (i = 0; i < 3; i++)
3658 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3659 mutex_unlock(&pit->pit_state.lock);
3663 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3665 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3666 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3667 sizeof(ps->channels));
3668 ps->flags = kvm->arch.vpit->pit_state.flags;
3669 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3670 memset(&ps->reserved, 0, sizeof(ps->reserved));
3674 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3678 u32 prev_legacy, cur_legacy;
3679 struct kvm_pit *pit = kvm->arch.vpit;
3681 mutex_lock(&pit->pit_state.lock);
3682 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3683 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3684 if (!prev_legacy && cur_legacy)
3686 memcpy(&pit->pit_state.channels, &ps->channels,
3687 sizeof(pit->pit_state.channels));
3688 pit->pit_state.flags = ps->flags;
3689 for (i = 0; i < 3; i++)
3690 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
3692 mutex_unlock(&pit->pit_state.lock);
3696 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3697 struct kvm_reinject_control *control)
3699 struct kvm_pit *pit = kvm->arch.vpit;
3704 /* pit->pit_state.lock was overloaded to prevent userspace from getting
3705 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3706 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
3708 mutex_lock(&pit->pit_state.lock);
3709 kvm_pit_set_reinject(pit, control->pit_reinject);
3710 mutex_unlock(&pit->pit_state.lock);
3716 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3717 * @kvm: kvm instance
3718 * @log: slot id and address to which we copy the log
3720 * Steps 1-4 below provide general overview of dirty page logging. See
3721 * kvm_get_dirty_log_protect() function description for additional details.
3723 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3724 * always flush the TLB (step 4) even if previous step failed and the dirty
3725 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3726 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3727 * writes will be marked dirty for next log read.
3729 * 1. Take a snapshot of the bit and clear it if needed.
3730 * 2. Write protect the corresponding page.
3731 * 3. Copy the snapshot to the userspace.
3732 * 4. Flush TLB's if needed.
3734 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3736 bool is_dirty = false;
3739 mutex_lock(&kvm->slots_lock);
3742 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3744 if (kvm_x86_ops->flush_log_dirty)
3745 kvm_x86_ops->flush_log_dirty(kvm);
3747 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3750 * All the TLBs can be flushed out of mmu lock, see the comments in
3751 * kvm_mmu_slot_remove_write_access().
3753 lockdep_assert_held(&kvm->slots_lock);
3755 kvm_flush_remote_tlbs(kvm);
3757 mutex_unlock(&kvm->slots_lock);
3761 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3764 if (!irqchip_in_kernel(kvm))
3767 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3768 irq_event->irq, irq_event->level,
3773 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3774 struct kvm_enable_cap *cap)
3782 case KVM_CAP_DISABLE_QUIRKS:
3783 kvm->arch.disabled_quirks = cap->args[0];
3786 case KVM_CAP_SPLIT_IRQCHIP: {
3787 mutex_lock(&kvm->lock);
3789 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3790 goto split_irqchip_unlock;
3792 if (irqchip_in_kernel(kvm))
3793 goto split_irqchip_unlock;
3794 if (kvm->created_vcpus)
3795 goto split_irqchip_unlock;
3796 r = kvm_setup_empty_irq_routing(kvm);
3798 goto split_irqchip_unlock;
3799 /* Pairs with irqchip_in_kernel. */
3801 kvm->arch.irqchip_split = true;
3802 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
3804 split_irqchip_unlock:
3805 mutex_unlock(&kvm->lock);
3808 case KVM_CAP_X2APIC_API:
3810 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
3813 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
3814 kvm->arch.x2apic_format = true;
3815 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
3816 kvm->arch.x2apic_broadcast_quirk_disabled = true;
3827 long kvm_arch_vm_ioctl(struct file *filp,
3828 unsigned int ioctl, unsigned long arg)
3830 struct kvm *kvm = filp->private_data;
3831 void __user *argp = (void __user *)arg;
3834 * This union makes it completely explicit to gcc-3.x
3835 * that these two variables' stack usage should be
3836 * combined, not added together.
3839 struct kvm_pit_state ps;
3840 struct kvm_pit_state2 ps2;
3841 struct kvm_pit_config pit_config;
3845 case KVM_SET_TSS_ADDR:
3846 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3848 case KVM_SET_IDENTITY_MAP_ADDR: {
3852 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3854 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3857 case KVM_SET_NR_MMU_PAGES:
3858 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3860 case KVM_GET_NR_MMU_PAGES:
3861 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3863 case KVM_CREATE_IRQCHIP: {
3864 struct kvm_pic *vpic;
3866 mutex_lock(&kvm->lock);
3869 goto create_irqchip_unlock;
3871 if (kvm->created_vcpus)
3872 goto create_irqchip_unlock;
3874 vpic = kvm_create_pic(kvm);
3876 r = kvm_ioapic_init(kvm);
3878 mutex_lock(&kvm->slots_lock);
3879 kvm_destroy_pic(vpic);
3880 mutex_unlock(&kvm->slots_lock);
3881 goto create_irqchip_unlock;
3884 goto create_irqchip_unlock;
3885 r = kvm_setup_default_irq_routing(kvm);
3887 mutex_lock(&kvm->slots_lock);
3888 mutex_lock(&kvm->irq_lock);
3889 kvm_ioapic_destroy(kvm);
3890 kvm_destroy_pic(vpic);
3891 mutex_unlock(&kvm->irq_lock);
3892 mutex_unlock(&kvm->slots_lock);
3893 goto create_irqchip_unlock;
3895 /* Write kvm->irq_routing before kvm->arch.vpic. */
3897 kvm->arch.vpic = vpic;
3898 create_irqchip_unlock:
3899 mutex_unlock(&kvm->lock);
3902 case KVM_CREATE_PIT:
3903 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3905 case KVM_CREATE_PIT2:
3907 if (copy_from_user(&u.pit_config, argp,
3908 sizeof(struct kvm_pit_config)))
3911 mutex_lock(&kvm->lock);
3914 goto create_pit_unlock;
3916 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3920 mutex_unlock(&kvm->lock);
3922 case KVM_GET_IRQCHIP: {
3923 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3924 struct kvm_irqchip *chip;
3926 chip = memdup_user(argp, sizeof(*chip));
3933 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3934 goto get_irqchip_out;
3935 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3937 goto get_irqchip_out;
3939 if (copy_to_user(argp, chip, sizeof *chip))
3940 goto get_irqchip_out;
3946 case KVM_SET_IRQCHIP: {
3947 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3948 struct kvm_irqchip *chip;
3950 chip = memdup_user(argp, sizeof(*chip));
3957 if (!irqchip_in_kernel(kvm) || irqchip_split(kvm))
3958 goto set_irqchip_out;
3959 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3961 goto set_irqchip_out;
3969 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3972 if (!kvm->arch.vpit)
3974 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3978 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3985 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3988 if (!kvm->arch.vpit)
3990 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3993 case KVM_GET_PIT2: {
3995 if (!kvm->arch.vpit)
3997 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4001 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4006 case KVM_SET_PIT2: {
4008 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4011 if (!kvm->arch.vpit)
4013 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4016 case KVM_REINJECT_CONTROL: {
4017 struct kvm_reinject_control control;
4019 if (copy_from_user(&control, argp, sizeof(control)))
4021 r = kvm_vm_ioctl_reinject(kvm, &control);
4024 case KVM_SET_BOOT_CPU_ID:
4026 mutex_lock(&kvm->lock);
4027 if (kvm->created_vcpus)
4030 kvm->arch.bsp_vcpu_id = arg;
4031 mutex_unlock(&kvm->lock);
4033 case KVM_XEN_HVM_CONFIG: {
4035 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4036 sizeof(struct kvm_xen_hvm_config)))
4039 if (kvm->arch.xen_hvm_config.flags)
4044 case KVM_SET_CLOCK: {
4045 struct kvm_clock_data user_ns;
4050 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4058 local_irq_disable();
4059 now_ns = get_kernel_ns();
4060 delta = user_ns.clock - now_ns;
4062 kvm->arch.kvmclock_offset = delta;
4063 kvm_gen_update_masterclock(kvm);
4066 case KVM_GET_CLOCK: {
4067 struct kvm_clock_data user_ns;
4070 local_irq_disable();
4071 now_ns = get_kernel_ns();
4072 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
4075 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4078 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4083 case KVM_ENABLE_CAP: {
4084 struct kvm_enable_cap cap;
4087 if (copy_from_user(&cap, argp, sizeof(cap)))
4089 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4093 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
4099 static void kvm_init_msr_list(void)
4104 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4105 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4109 * Even MSRs that are valid in the host may not be exposed
4110 * to the guests in some cases.
4112 switch (msrs_to_save[i]) {
4113 case MSR_IA32_BNDCFGS:
4114 if (!kvm_x86_ops->mpx_supported())
4118 if (!kvm_x86_ops->rdtscp_supported())
4126 msrs_to_save[j] = msrs_to_save[i];
4129 num_msrs_to_save = j;
4131 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4132 switch (emulated_msrs[i]) {
4133 case MSR_IA32_SMBASE:
4134 if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4142 emulated_msrs[j] = emulated_msrs[i];
4145 num_emulated_msrs = j;
4148 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4156 if (!(lapic_in_kernel(vcpu) &&
4157 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4158 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4169 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4176 if (!(lapic_in_kernel(vcpu) &&
4177 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4179 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4181 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4191 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4192 struct kvm_segment *var, int seg)
4194 kvm_x86_ops->set_segment(vcpu, var, seg);
4197 void kvm_get_segment(struct kvm_vcpu *vcpu,
4198 struct kvm_segment *var, int seg)
4200 kvm_x86_ops->get_segment(vcpu, var, seg);
4203 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4204 struct x86_exception *exception)
4208 BUG_ON(!mmu_is_nested(vcpu));
4210 /* NPT walks are always user-walks */
4211 access |= PFERR_USER_MASK;
4212 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4217 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4218 struct x86_exception *exception)
4220 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4221 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4224 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4225 struct x86_exception *exception)
4227 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4228 access |= PFERR_FETCH_MASK;
4229 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4232 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4233 struct x86_exception *exception)
4235 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4236 access |= PFERR_WRITE_MASK;
4237 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4240 /* uses this to access any guest's mapped memory without checking CPL */
4241 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4242 struct x86_exception *exception)
4244 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4247 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4248 struct kvm_vcpu *vcpu, u32 access,
4249 struct x86_exception *exception)
4252 int r = X86EMUL_CONTINUE;
4255 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4257 unsigned offset = addr & (PAGE_SIZE-1);
4258 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4261 if (gpa == UNMAPPED_GVA)
4262 return X86EMUL_PROPAGATE_FAULT;
4263 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4266 r = X86EMUL_IO_NEEDED;
4278 /* used for instruction fetching */
4279 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4280 gva_t addr, void *val, unsigned int bytes,
4281 struct x86_exception *exception)
4283 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4284 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4288 /* Inline kvm_read_guest_virt_helper for speed. */
4289 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4291 if (unlikely(gpa == UNMAPPED_GVA))
4292 return X86EMUL_PROPAGATE_FAULT;
4294 offset = addr & (PAGE_SIZE-1);
4295 if (WARN_ON(offset + bytes > PAGE_SIZE))
4296 bytes = (unsigned)PAGE_SIZE - offset;
4297 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4299 if (unlikely(ret < 0))
4300 return X86EMUL_IO_NEEDED;
4302 return X86EMUL_CONTINUE;
4305 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4306 gva_t addr, void *val, unsigned int bytes,
4307 struct x86_exception *exception)
4309 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4310 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4312 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4315 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4317 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4318 gva_t addr, void *val, unsigned int bytes,
4319 struct x86_exception *exception)
4321 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4322 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4325 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4326 unsigned long addr, void *val, unsigned int bytes)
4328 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4329 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4331 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4334 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4335 gva_t addr, void *val,
4337 struct x86_exception *exception)
4339 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4341 int r = X86EMUL_CONTINUE;
4344 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4347 unsigned offset = addr & (PAGE_SIZE-1);
4348 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4351 if (gpa == UNMAPPED_GVA)
4352 return X86EMUL_PROPAGATE_FAULT;
4353 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4355 r = X86EMUL_IO_NEEDED;
4366 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4368 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4369 gpa_t *gpa, struct x86_exception *exception,
4372 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4373 | (write ? PFERR_WRITE_MASK : 0);
4376 * currently PKRU is only applied to ept enabled guest so
4377 * there is no pkey in EPT page table for L1 guest or EPT
4378 * shadow page table for L2 guest.
4380 if (vcpu_match_mmio_gva(vcpu, gva)
4381 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4382 vcpu->arch.access, 0, access)) {
4383 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4384 (gva & (PAGE_SIZE - 1));
4385 trace_vcpu_match_mmio(gva, *gpa, write, false);
4389 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4391 if (*gpa == UNMAPPED_GVA)
4394 /* For APIC access vmexit */
4395 if ((*gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4398 if (vcpu_match_mmio_gpa(vcpu, *gpa)) {
4399 trace_vcpu_match_mmio(gva, *gpa, write, true);
4406 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4407 const void *val, int bytes)
4411 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4414 kvm_page_track_write(vcpu, gpa, val, bytes);
4418 struct read_write_emulator_ops {
4419 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4421 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4422 void *val, int bytes);
4423 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4424 int bytes, void *val);
4425 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4426 void *val, int bytes);
4430 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4432 if (vcpu->mmio_read_completed) {
4433 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4434 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4435 vcpu->mmio_read_completed = 0;
4442 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4443 void *val, int bytes)
4445 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4448 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4449 void *val, int bytes)
4451 return emulator_write_phys(vcpu, gpa, val, bytes);
4454 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4456 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4457 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4460 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4461 void *val, int bytes)
4463 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4464 return X86EMUL_IO_NEEDED;
4467 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4468 void *val, int bytes)
4470 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4472 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4473 return X86EMUL_CONTINUE;
4476 static const struct read_write_emulator_ops read_emultor = {
4477 .read_write_prepare = read_prepare,
4478 .read_write_emulate = read_emulate,
4479 .read_write_mmio = vcpu_mmio_read,
4480 .read_write_exit_mmio = read_exit_mmio,
4483 static const struct read_write_emulator_ops write_emultor = {
4484 .read_write_emulate = write_emulate,
4485 .read_write_mmio = write_mmio,
4486 .read_write_exit_mmio = write_exit_mmio,
4490 static int emulator_read_write_onepage(unsigned long addr, void *val,
4492 struct x86_exception *exception,
4493 struct kvm_vcpu *vcpu,
4494 const struct read_write_emulator_ops *ops)
4498 bool write = ops->write;
4499 struct kvm_mmio_fragment *frag;
4501 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4504 return X86EMUL_PROPAGATE_FAULT;
4506 /* For APIC access vmexit */
4510 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4511 return X86EMUL_CONTINUE;
4515 * Is this MMIO handled locally?
4517 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4518 if (handled == bytes)
4519 return X86EMUL_CONTINUE;
4525 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4526 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4530 return X86EMUL_CONTINUE;
4533 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4535 void *val, unsigned int bytes,
4536 struct x86_exception *exception,
4537 const struct read_write_emulator_ops *ops)
4539 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4543 if (ops->read_write_prepare &&
4544 ops->read_write_prepare(vcpu, val, bytes))
4545 return X86EMUL_CONTINUE;
4547 vcpu->mmio_nr_fragments = 0;
4549 /* Crossing a page boundary? */
4550 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4553 now = -addr & ~PAGE_MASK;
4554 rc = emulator_read_write_onepage(addr, val, now, exception,
4557 if (rc != X86EMUL_CONTINUE)
4560 if (ctxt->mode != X86EMUL_MODE_PROT64)
4566 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4568 if (rc != X86EMUL_CONTINUE)
4571 if (!vcpu->mmio_nr_fragments)
4574 gpa = vcpu->mmio_fragments[0].gpa;
4576 vcpu->mmio_needed = 1;
4577 vcpu->mmio_cur_fragment = 0;
4579 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4580 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4581 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4582 vcpu->run->mmio.phys_addr = gpa;
4584 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4587 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4591 struct x86_exception *exception)
4593 return emulator_read_write(ctxt, addr, val, bytes,
4594 exception, &read_emultor);
4597 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4601 struct x86_exception *exception)
4603 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4604 exception, &write_emultor);
4607 #define CMPXCHG_TYPE(t, ptr, old, new) \
4608 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4610 #ifdef CONFIG_X86_64
4611 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4613 # define CMPXCHG64(ptr, old, new) \
4614 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4617 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4622 struct x86_exception *exception)
4624 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4630 /* guests cmpxchg8b have to be emulated atomically */
4631 if (bytes > 8 || (bytes & (bytes - 1)))
4634 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4636 if (gpa == UNMAPPED_GVA ||
4637 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4640 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4643 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4644 if (is_error_page(page))
4647 kaddr = kmap_atomic(page);
4648 kaddr += offset_in_page(gpa);
4651 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4654 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4657 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4660 exchanged = CMPXCHG64(kaddr, old, new);
4665 kunmap_atomic(kaddr);
4666 kvm_release_page_dirty(page);
4669 return X86EMUL_CMPXCHG_FAILED;
4671 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4672 kvm_page_track_write(vcpu, gpa, new, bytes);
4674 return X86EMUL_CONTINUE;
4677 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4679 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4682 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4684 /* TODO: String I/O for in kernel device */
4687 if (vcpu->arch.pio.in)
4688 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4689 vcpu->arch.pio.size, pd);
4691 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4692 vcpu->arch.pio.port, vcpu->arch.pio.size,
4697 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4698 unsigned short port, void *val,
4699 unsigned int count, bool in)
4701 vcpu->arch.pio.port = port;
4702 vcpu->arch.pio.in = in;
4703 vcpu->arch.pio.count = count;
4704 vcpu->arch.pio.size = size;
4706 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4707 vcpu->arch.pio.count = 0;
4711 vcpu->run->exit_reason = KVM_EXIT_IO;
4712 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4713 vcpu->run->io.size = size;
4714 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4715 vcpu->run->io.count = count;
4716 vcpu->run->io.port = port;
4721 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4722 int size, unsigned short port, void *val,
4725 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4728 if (vcpu->arch.pio.count)
4731 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4734 memcpy(val, vcpu->arch.pio_data, size * count);
4735 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4736 vcpu->arch.pio.count = 0;
4743 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4744 int size, unsigned short port,
4745 const void *val, unsigned int count)
4747 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4749 memcpy(vcpu->arch.pio_data, val, size * count);
4750 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4751 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4754 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4756 return kvm_x86_ops->get_segment_base(vcpu, seg);
4759 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4761 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4764 int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4766 if (!need_emulate_wbinvd(vcpu))
4767 return X86EMUL_CONTINUE;
4769 if (kvm_x86_ops->has_wbinvd_exit()) {
4770 int cpu = get_cpu();
4772 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4773 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4774 wbinvd_ipi, NULL, 1);
4776 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4779 return X86EMUL_CONTINUE;
4782 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4784 kvm_x86_ops->skip_emulated_instruction(vcpu);
4785 return kvm_emulate_wbinvd_noskip(vcpu);
4787 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4791 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4793 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4796 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4797 unsigned long *dest)
4799 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4802 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4803 unsigned long value)
4806 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4809 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4811 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4814 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4816 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4817 unsigned long value;
4821 value = kvm_read_cr0(vcpu);
4824 value = vcpu->arch.cr2;
4827 value = kvm_read_cr3(vcpu);
4830 value = kvm_read_cr4(vcpu);
4833 value = kvm_get_cr8(vcpu);
4836 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4843 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4845 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4850 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4853 vcpu->arch.cr2 = val;
4856 res = kvm_set_cr3(vcpu, val);
4859 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4862 res = kvm_set_cr8(vcpu, val);
4865 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4872 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4874 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4877 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4879 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4882 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4884 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4887 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4889 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4892 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4894 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4897 static unsigned long emulator_get_cached_segment_base(
4898 struct x86_emulate_ctxt *ctxt, int seg)
4900 return get_segment_base(emul_to_vcpu(ctxt), seg);
4903 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4904 struct desc_struct *desc, u32 *base3,
4907 struct kvm_segment var;
4909 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4910 *selector = var.selector;
4913 memset(desc, 0, sizeof(*desc));
4919 set_desc_limit(desc, var.limit);
4920 set_desc_base(desc, (unsigned long)var.base);
4921 #ifdef CONFIG_X86_64
4923 *base3 = var.base >> 32;
4925 desc->type = var.type;
4927 desc->dpl = var.dpl;
4928 desc->p = var.present;
4929 desc->avl = var.avl;
4937 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4938 struct desc_struct *desc, u32 base3,
4941 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4942 struct kvm_segment var;
4944 var.selector = selector;
4945 var.base = get_desc_base(desc);
4946 #ifdef CONFIG_X86_64
4947 var.base |= ((u64)base3) << 32;
4949 var.limit = get_desc_limit(desc);
4951 var.limit = (var.limit << 12) | 0xfff;
4952 var.type = desc->type;
4953 var.dpl = desc->dpl;
4958 var.avl = desc->avl;
4959 var.present = desc->p;
4960 var.unusable = !var.present;
4963 kvm_set_segment(vcpu, &var, seg);
4967 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4968 u32 msr_index, u64 *pdata)
4970 struct msr_data msr;
4973 msr.index = msr_index;
4974 msr.host_initiated = false;
4975 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
4983 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4984 u32 msr_index, u64 data)
4986 struct msr_data msr;
4989 msr.index = msr_index;
4990 msr.host_initiated = false;
4991 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
4994 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
4996 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4998 return vcpu->arch.smbase;
5001 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5003 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5005 vcpu->arch.smbase = smbase;
5008 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5011 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5014 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5015 u32 pmc, u64 *pdata)
5017 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5020 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5022 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5025 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
5028 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
5030 * CR0.TS may reference the host fpu state, not the guest fpu state,
5031 * so it may be clear at this point.
5036 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
5041 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5042 struct x86_instruction_info *info,
5043 enum x86_intercept_stage stage)
5045 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5048 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5049 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
5051 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
5054 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5056 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5059 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5061 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5064 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5066 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5069 static const struct x86_emulate_ops emulate_ops = {
5070 .read_gpr = emulator_read_gpr,
5071 .write_gpr = emulator_write_gpr,
5072 .read_std = kvm_read_guest_virt_system,
5073 .write_std = kvm_write_guest_virt_system,
5074 .read_phys = kvm_read_guest_phys_system,
5075 .fetch = kvm_fetch_guest_virt,
5076 .read_emulated = emulator_read_emulated,
5077 .write_emulated = emulator_write_emulated,
5078 .cmpxchg_emulated = emulator_cmpxchg_emulated,
5079 .invlpg = emulator_invlpg,
5080 .pio_in_emulated = emulator_pio_in_emulated,
5081 .pio_out_emulated = emulator_pio_out_emulated,
5082 .get_segment = emulator_get_segment,
5083 .set_segment = emulator_set_segment,
5084 .get_cached_segment_base = emulator_get_cached_segment_base,
5085 .get_gdt = emulator_get_gdt,
5086 .get_idt = emulator_get_idt,
5087 .set_gdt = emulator_set_gdt,
5088 .set_idt = emulator_set_idt,
5089 .get_cr = emulator_get_cr,
5090 .set_cr = emulator_set_cr,
5091 .cpl = emulator_get_cpl,
5092 .get_dr = emulator_get_dr,
5093 .set_dr = emulator_set_dr,
5094 .get_smbase = emulator_get_smbase,
5095 .set_smbase = emulator_set_smbase,
5096 .set_msr = emulator_set_msr,
5097 .get_msr = emulator_get_msr,
5098 .check_pmc = emulator_check_pmc,
5099 .read_pmc = emulator_read_pmc,
5100 .halt = emulator_halt,
5101 .wbinvd = emulator_wbinvd,
5102 .fix_hypercall = emulator_fix_hypercall,
5103 .get_fpu = emulator_get_fpu,
5104 .put_fpu = emulator_put_fpu,
5105 .intercept = emulator_intercept,
5106 .get_cpuid = emulator_get_cpuid,
5107 .set_nmi_mask = emulator_set_nmi_mask,
5110 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5112 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5114 * an sti; sti; sequence only disable interrupts for the first
5115 * instruction. So, if the last instruction, be it emulated or
5116 * not, left the system with the INT_STI flag enabled, it
5117 * means that the last instruction is an sti. We should not
5118 * leave the flag on in this case. The same goes for mov ss
5120 if (int_shadow & mask)
5122 if (unlikely(int_shadow || mask)) {
5123 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5125 kvm_make_request(KVM_REQ_EVENT, vcpu);
5129 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5131 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5132 if (ctxt->exception.vector == PF_VECTOR)
5133 return kvm_propagate_fault(vcpu, &ctxt->exception);
5135 if (ctxt->exception.error_code_valid)
5136 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5137 ctxt->exception.error_code);
5139 kvm_queue_exception(vcpu, ctxt->exception.vector);
5143 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5145 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5148 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5150 ctxt->eflags = kvm_get_rflags(vcpu);
5151 ctxt->eip = kvm_rip_read(vcpu);
5152 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5153 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
5154 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
5155 cs_db ? X86EMUL_MODE_PROT32 :
5156 X86EMUL_MODE_PROT16;
5157 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5158 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5159 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5160 ctxt->emul_flags = vcpu->arch.hflags;
5162 init_decode_cache(ctxt);
5163 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5166 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5168 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5171 init_emulate_ctxt(vcpu);
5175 ctxt->_eip = ctxt->eip + inc_eip;
5176 ret = emulate_int_real(ctxt, irq);
5178 if (ret != X86EMUL_CONTINUE)
5179 return EMULATE_FAIL;
5181 ctxt->eip = ctxt->_eip;
5182 kvm_rip_write(vcpu, ctxt->eip);
5183 kvm_set_rflags(vcpu, ctxt->eflags);
5185 if (irq == NMI_VECTOR)
5186 vcpu->arch.nmi_pending = 0;
5188 vcpu->arch.interrupt.pending = false;
5190 return EMULATE_DONE;
5192 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5194 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5196 int r = EMULATE_DONE;
5198 ++vcpu->stat.insn_emulation_fail;
5199 trace_kvm_emulate_insn_failed(vcpu);
5200 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5201 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5202 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5203 vcpu->run->internal.ndata = 0;
5206 kvm_queue_exception(vcpu, UD_VECTOR);
5211 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5212 bool write_fault_to_shadow_pgtable,
5218 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5221 if (!vcpu->arch.mmu.direct_map) {
5223 * Write permission should be allowed since only
5224 * write access need to be emulated.
5226 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5229 * If the mapping is invalid in guest, let cpu retry
5230 * it to generate fault.
5232 if (gpa == UNMAPPED_GVA)
5237 * Do not retry the unhandleable instruction if it faults on the
5238 * readonly host memory, otherwise it will goto a infinite loop:
5239 * retry instruction -> write #PF -> emulation fail -> retry
5240 * instruction -> ...
5242 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5245 * If the instruction failed on the error pfn, it can not be fixed,
5246 * report the error to userspace.
5248 if (is_error_noslot_pfn(pfn))
5251 kvm_release_pfn_clean(pfn);
5253 /* The instructions are well-emulated on direct mmu. */
5254 if (vcpu->arch.mmu.direct_map) {
5255 unsigned int indirect_shadow_pages;
5257 spin_lock(&vcpu->kvm->mmu_lock);
5258 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5259 spin_unlock(&vcpu->kvm->mmu_lock);
5261 if (indirect_shadow_pages)
5262 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5268 * if emulation was due to access to shadowed page table
5269 * and it failed try to unshadow page and re-enter the
5270 * guest to let CPU execute the instruction.
5272 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5275 * If the access faults on its page table, it can not
5276 * be fixed by unprotecting shadow page and it should
5277 * be reported to userspace.
5279 return !write_fault_to_shadow_pgtable;
5282 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5283 unsigned long cr2, int emulation_type)
5285 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5286 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5288 last_retry_eip = vcpu->arch.last_retry_eip;
5289 last_retry_addr = vcpu->arch.last_retry_addr;
5292 * If the emulation is caused by #PF and it is non-page_table
5293 * writing instruction, it means the VM-EXIT is caused by shadow
5294 * page protected, we can zap the shadow page and retry this
5295 * instruction directly.
5297 * Note: if the guest uses a non-page-table modifying instruction
5298 * on the PDE that points to the instruction, then we will unmap
5299 * the instruction and go to an infinite loop. So, we cache the
5300 * last retried eip and the last fault address, if we meet the eip
5301 * and the address again, we can break out of the potential infinite
5304 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5306 if (!(emulation_type & EMULTYPE_RETRY))
5309 if (x86_page_table_writing_insn(ctxt))
5312 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5315 vcpu->arch.last_retry_eip = ctxt->eip;
5316 vcpu->arch.last_retry_addr = cr2;
5318 if (!vcpu->arch.mmu.direct_map)
5319 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5321 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5326 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5327 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5329 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5331 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5332 /* This is a good place to trace that we are exiting SMM. */
5333 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5335 /* Process a latched INIT or SMI, if any. */
5336 kvm_make_request(KVM_REQ_EVENT, vcpu);
5339 kvm_mmu_reset_context(vcpu);
5342 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5344 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5346 vcpu->arch.hflags = emul_flags;
5348 if (changed & HF_SMM_MASK)
5349 kvm_smm_changed(vcpu);
5352 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5361 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5362 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5367 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5369 struct kvm_run *kvm_run = vcpu->run;
5372 * rflags is the old, "raw" value of the flags. The new value has
5373 * not been saved yet.
5375 * This is correct even for TF set by the guest, because "the
5376 * processor will not generate this exception after the instruction
5377 * that sets the TF flag".
5379 if (unlikely(rflags & X86_EFLAGS_TF)) {
5380 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5381 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5383 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5384 kvm_run->debug.arch.exception = DB_VECTOR;
5385 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5386 *r = EMULATE_USER_EXIT;
5388 vcpu->arch.emulate_ctxt.eflags &= ~X86_EFLAGS_TF;
5390 * "Certain debug exceptions may clear bit 0-3. The
5391 * remaining contents of the DR6 register are never
5392 * cleared by the processor".
5394 vcpu->arch.dr6 &= ~15;
5395 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5396 kvm_queue_exception(vcpu, DB_VECTOR);
5401 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5403 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5404 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5405 struct kvm_run *kvm_run = vcpu->run;
5406 unsigned long eip = kvm_get_linear_rip(vcpu);
5407 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5408 vcpu->arch.guest_debug_dr7,
5412 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5413 kvm_run->debug.arch.pc = eip;
5414 kvm_run->debug.arch.exception = DB_VECTOR;
5415 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5416 *r = EMULATE_USER_EXIT;
5421 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5422 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5423 unsigned long eip = kvm_get_linear_rip(vcpu);
5424 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5429 vcpu->arch.dr6 &= ~15;
5430 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5431 kvm_queue_exception(vcpu, DB_VECTOR);
5440 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5447 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5448 bool writeback = true;
5449 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5452 * Clear write_fault_to_shadow_pgtable here to ensure it is
5455 vcpu->arch.write_fault_to_shadow_pgtable = false;
5456 kvm_clear_exception_queue(vcpu);
5458 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5459 init_emulate_ctxt(vcpu);
5462 * We will reenter on the same instruction since
5463 * we do not set complete_userspace_io. This does not
5464 * handle watchpoints yet, those would be handled in
5467 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5470 ctxt->interruptibility = 0;
5471 ctxt->have_exception = false;
5472 ctxt->exception.vector = -1;
5473 ctxt->perm_ok = false;
5475 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5477 r = x86_decode_insn(ctxt, insn, insn_len);
5479 trace_kvm_emulate_insn_start(vcpu);
5480 ++vcpu->stat.insn_emulation;
5481 if (r != EMULATION_OK) {
5482 if (emulation_type & EMULTYPE_TRAP_UD)
5483 return EMULATE_FAIL;
5484 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5486 return EMULATE_DONE;
5487 if (emulation_type & EMULTYPE_SKIP)
5488 return EMULATE_FAIL;
5489 return handle_emulation_failure(vcpu);
5493 if (emulation_type & EMULTYPE_SKIP) {
5494 kvm_rip_write(vcpu, ctxt->_eip);
5495 if (ctxt->eflags & X86_EFLAGS_RF)
5496 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5497 return EMULATE_DONE;
5500 if (retry_instruction(ctxt, cr2, emulation_type))
5501 return EMULATE_DONE;
5503 /* this is needed for vmware backdoor interface to work since it
5504 changes registers values during IO operation */
5505 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5506 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5507 emulator_invalidate_register_cache(ctxt);
5511 r = x86_emulate_insn(ctxt);
5513 if (r == EMULATION_INTERCEPTED)
5514 return EMULATE_DONE;
5516 if (r == EMULATION_FAILED) {
5517 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5519 return EMULATE_DONE;
5521 return handle_emulation_failure(vcpu);
5524 if (ctxt->have_exception) {
5526 if (inject_emulated_exception(vcpu))
5528 } else if (vcpu->arch.pio.count) {
5529 if (!vcpu->arch.pio.in) {
5530 /* FIXME: return into emulator if single-stepping. */
5531 vcpu->arch.pio.count = 0;
5534 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5536 r = EMULATE_USER_EXIT;
5537 } else if (vcpu->mmio_needed) {
5538 if (!vcpu->mmio_is_write)
5540 r = EMULATE_USER_EXIT;
5541 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5542 } else if (r == EMULATION_RESTART)
5548 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5549 toggle_interruptibility(vcpu, ctxt->interruptibility);
5550 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5551 if (vcpu->arch.hflags != ctxt->emul_flags)
5552 kvm_set_hflags(vcpu, ctxt->emul_flags);
5553 kvm_rip_write(vcpu, ctxt->eip);
5554 if (r == EMULATE_DONE)
5555 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5556 if (!ctxt->have_exception ||
5557 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5558 __kvm_set_rflags(vcpu, ctxt->eflags);
5561 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5562 * do nothing, and it will be requested again as soon as
5563 * the shadow expires. But we still need to check here,
5564 * because POPF has no interrupt shadow.
5566 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5567 kvm_make_request(KVM_REQ_EVENT, vcpu);
5569 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5573 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5575 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5577 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5578 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5579 size, port, &val, 1);
5580 /* do not return to emulator after return from userspace */
5581 vcpu->arch.pio.count = 0;
5584 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5586 static void tsc_bad(void *info)
5588 __this_cpu_write(cpu_tsc_khz, 0);
5591 static void tsc_khz_changed(void *data)
5593 struct cpufreq_freqs *freq = data;
5594 unsigned long khz = 0;
5598 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5599 khz = cpufreq_quick_get(raw_smp_processor_id());
5602 __this_cpu_write(cpu_tsc_khz, khz);
5605 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5608 struct cpufreq_freqs *freq = data;
5610 struct kvm_vcpu *vcpu;
5611 int i, send_ipi = 0;
5614 * We allow guests to temporarily run on slowing clocks,
5615 * provided we notify them after, or to run on accelerating
5616 * clocks, provided we notify them before. Thus time never
5619 * However, we have a problem. We can't atomically update
5620 * the frequency of a given CPU from this function; it is
5621 * merely a notifier, which can be called from any CPU.
5622 * Changing the TSC frequency at arbitrary points in time
5623 * requires a recomputation of local variables related to
5624 * the TSC for each VCPU. We must flag these local variables
5625 * to be updated and be sure the update takes place with the
5626 * new frequency before any guests proceed.
5628 * Unfortunately, the combination of hotplug CPU and frequency
5629 * change creates an intractable locking scenario; the order
5630 * of when these callouts happen is undefined with respect to
5631 * CPU hotplug, and they can race with each other. As such,
5632 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5633 * undefined; you can actually have a CPU frequency change take
5634 * place in between the computation of X and the setting of the
5635 * variable. To protect against this problem, all updates of
5636 * the per_cpu tsc_khz variable are done in an interrupt
5637 * protected IPI, and all callers wishing to update the value
5638 * must wait for a synchronous IPI to complete (which is trivial
5639 * if the caller is on the CPU already). This establishes the
5640 * necessary total order on variable updates.
5642 * Note that because a guest time update may take place
5643 * anytime after the setting of the VCPU's request bit, the
5644 * correct TSC value must be set before the request. However,
5645 * to ensure the update actually makes it to any guest which
5646 * starts running in hardware virtualization between the set
5647 * and the acquisition of the spinlock, we must also ping the
5648 * CPU after setting the request bit.
5652 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5654 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5657 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5659 spin_lock(&kvm_lock);
5660 list_for_each_entry(kvm, &vm_list, vm_list) {
5661 kvm_for_each_vcpu(i, vcpu, kvm) {
5662 if (vcpu->cpu != freq->cpu)
5664 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5665 if (vcpu->cpu != smp_processor_id())
5669 spin_unlock(&kvm_lock);
5671 if (freq->old < freq->new && send_ipi) {
5673 * We upscale the frequency. Must make the guest
5674 * doesn't see old kvmclock values while running with
5675 * the new frequency, otherwise we risk the guest sees
5676 * time go backwards.
5678 * In case we update the frequency for another cpu
5679 * (which might be in guest context) send an interrupt
5680 * to kick the cpu out of guest context. Next time
5681 * guest context is entered kvmclock will be updated,
5682 * so the guest will not see stale values.
5684 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5689 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5690 .notifier_call = kvmclock_cpufreq_notifier
5693 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
5694 unsigned long action, void *hcpu)
5696 unsigned int cpu = (unsigned long)hcpu;
5700 case CPU_DOWN_FAILED:
5701 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5703 case CPU_DOWN_PREPARE:
5704 smp_call_function_single(cpu, tsc_bad, NULL, 1);
5710 static struct notifier_block kvmclock_cpu_notifier_block = {
5711 .notifier_call = kvmclock_cpu_notifier,
5712 .priority = -INT_MAX
5715 static void kvm_timer_init(void)
5719 max_tsc_khz = tsc_khz;
5721 cpu_notifier_register_begin();
5722 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5723 #ifdef CONFIG_CPU_FREQ
5724 struct cpufreq_policy policy;
5725 memset(&policy, 0, sizeof(policy));
5727 cpufreq_get_policy(&policy, cpu);
5728 if (policy.cpuinfo.max_freq)
5729 max_tsc_khz = policy.cpuinfo.max_freq;
5732 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5733 CPUFREQ_TRANSITION_NOTIFIER);
5735 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5736 for_each_online_cpu(cpu)
5737 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
5739 __register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5740 cpu_notifier_register_done();
5744 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5746 int kvm_is_in_guest(void)
5748 return __this_cpu_read(current_vcpu) != NULL;
5751 static int kvm_is_user_mode(void)
5755 if (__this_cpu_read(current_vcpu))
5756 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5758 return user_mode != 0;
5761 static unsigned long kvm_get_guest_ip(void)
5763 unsigned long ip = 0;
5765 if (__this_cpu_read(current_vcpu))
5766 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5771 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5772 .is_in_guest = kvm_is_in_guest,
5773 .is_user_mode = kvm_is_user_mode,
5774 .get_guest_ip = kvm_get_guest_ip,
5777 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5779 __this_cpu_write(current_vcpu, vcpu);
5781 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5783 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5785 __this_cpu_write(current_vcpu, NULL);
5787 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5789 static void kvm_set_mmio_spte_mask(void)
5792 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5795 * Set the reserved bits and the present bit of an paging-structure
5796 * entry to generate page fault with PFER.RSV = 1.
5798 /* Mask the reserved physical address bits. */
5799 mask = rsvd_bits(maxphyaddr, 51);
5801 /* Bit 62 is always reserved for 32bit host. */
5802 mask |= 0x3ull << 62;
5804 /* Set the present bit. */
5807 #ifdef CONFIG_X86_64
5809 * If reserved bit is not supported, clear the present bit to disable
5812 if (maxphyaddr == 52)
5816 kvm_mmu_set_mmio_spte_mask(mask);
5819 #ifdef CONFIG_X86_64
5820 static void pvclock_gtod_update_fn(struct work_struct *work)
5824 struct kvm_vcpu *vcpu;
5827 spin_lock(&kvm_lock);
5828 list_for_each_entry(kvm, &vm_list, vm_list)
5829 kvm_for_each_vcpu(i, vcpu, kvm)
5830 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
5831 atomic_set(&kvm_guest_has_master_clock, 0);
5832 spin_unlock(&kvm_lock);
5835 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
5838 * Notification about pvclock gtod data update.
5840 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
5843 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
5844 struct timekeeper *tk = priv;
5846 update_pvclock_gtod(tk);
5848 /* disable master clock if host does not trust, or does not
5849 * use, TSC clocksource
5851 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
5852 atomic_read(&kvm_guest_has_master_clock) != 0)
5853 queue_work(system_long_wq, &pvclock_gtod_work);
5858 static struct notifier_block pvclock_gtod_notifier = {
5859 .notifier_call = pvclock_gtod_notify,
5863 int kvm_arch_init(void *opaque)
5866 struct kvm_x86_ops *ops = opaque;
5869 printk(KERN_ERR "kvm: already loaded the other module\n");
5874 if (!ops->cpu_has_kvm_support()) {
5875 printk(KERN_ERR "kvm: no hardware support\n");
5879 if (ops->disabled_by_bios()) {
5880 printk(KERN_ERR "kvm: disabled by bios\n");
5886 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
5888 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
5892 r = kvm_mmu_module_init();
5894 goto out_free_percpu;
5896 kvm_set_mmio_spte_mask();
5900 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
5901 PT_DIRTY_MASK, PT64_NX_MASK, 0,
5905 perf_register_guest_info_callbacks(&kvm_guest_cbs);
5907 if (boot_cpu_has(X86_FEATURE_XSAVE))
5908 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
5911 #ifdef CONFIG_X86_64
5912 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
5918 free_percpu(shared_msrs);
5923 void kvm_arch_exit(void)
5925 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5927 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5928 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
5929 CPUFREQ_TRANSITION_NOTIFIER);
5930 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
5931 #ifdef CONFIG_X86_64
5932 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
5935 kvm_mmu_module_exit();
5936 free_percpu(shared_msrs);
5939 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
5941 ++vcpu->stat.halt_exits;
5942 if (lapic_in_kernel(vcpu)) {
5943 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
5946 vcpu->run->exit_reason = KVM_EXIT_HLT;
5950 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
5952 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5954 kvm_x86_ops->skip_emulated_instruction(vcpu);
5955 return kvm_vcpu_halt(vcpu);
5957 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5960 * kvm_pv_kick_cpu_op: Kick a vcpu.
5962 * @apicid - apicid of vcpu to be kicked.
5964 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
5966 struct kvm_lapic_irq lapic_irq;
5968 lapic_irq.shorthand = 0;
5969 lapic_irq.dest_mode = 0;
5970 lapic_irq.dest_id = apicid;
5971 lapic_irq.msi_redir_hint = false;
5973 lapic_irq.delivery_mode = APIC_DM_REMRD;
5974 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
5977 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
5979 vcpu->arch.apicv_active = false;
5980 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
5983 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5985 unsigned long nr, a0, a1, a2, a3, ret;
5986 int op_64_bit, r = 1;
5988 kvm_x86_ops->skip_emulated_instruction(vcpu);
5990 if (kvm_hv_hypercall_enabled(vcpu->kvm))
5991 return kvm_hv_hypercall(vcpu);
5993 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5994 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5995 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5996 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5997 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5999 trace_kvm_hypercall(nr, a0, a1, a2, a3);
6001 op_64_bit = is_64_bit_mode(vcpu);
6010 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6016 case KVM_HC_VAPIC_POLL_IRQ:
6019 case KVM_HC_KICK_CPU:
6020 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6030 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6031 ++vcpu->stat.hypercalls;
6034 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6036 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6038 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6039 char instruction[3];
6040 unsigned long rip = kvm_rip_read(vcpu);
6042 kvm_x86_ops->patch_hypercall(vcpu, instruction);
6044 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
6047 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6049 return vcpu->run->request_interrupt_window &&
6050 likely(!pic_in_kernel(vcpu->kvm));
6053 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6055 struct kvm_run *kvm_run = vcpu->run;
6057 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6058 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6059 kvm_run->cr8 = kvm_get_cr8(vcpu);
6060 kvm_run->apic_base = kvm_get_apic_base(vcpu);
6061 kvm_run->ready_for_interrupt_injection =
6062 pic_in_kernel(vcpu->kvm) ||
6063 kvm_vcpu_ready_for_interrupt_injection(vcpu);
6066 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6070 if (!kvm_x86_ops->update_cr8_intercept)
6073 if (!lapic_in_kernel(vcpu))
6076 if (vcpu->arch.apicv_active)
6079 if (!vcpu->arch.apic->vapic_addr)
6080 max_irr = kvm_lapic_find_highest_irr(vcpu);
6087 tpr = kvm_lapic_get_cr8(vcpu);
6089 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6092 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6096 /* try to reinject previous events if any */
6097 if (vcpu->arch.exception.pending) {
6098 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6099 vcpu->arch.exception.has_error_code,
6100 vcpu->arch.exception.error_code);
6102 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6103 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6106 if (vcpu->arch.exception.nr == DB_VECTOR &&
6107 (vcpu->arch.dr7 & DR7_GD)) {
6108 vcpu->arch.dr7 &= ~DR7_GD;
6109 kvm_update_dr7(vcpu);
6112 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6113 vcpu->arch.exception.has_error_code,
6114 vcpu->arch.exception.error_code,
6115 vcpu->arch.exception.reinject);
6119 if (vcpu->arch.nmi_injected) {
6120 kvm_x86_ops->set_nmi(vcpu);
6124 if (vcpu->arch.interrupt.pending) {
6125 kvm_x86_ops->set_irq(vcpu);
6129 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6130 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6135 /* try to inject new event if pending */
6136 if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
6137 vcpu->arch.smi_pending = false;
6139 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6140 --vcpu->arch.nmi_pending;
6141 vcpu->arch.nmi_injected = true;
6142 kvm_x86_ops->set_nmi(vcpu);
6143 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6145 * Because interrupts can be injected asynchronously, we are
6146 * calling check_nested_events again here to avoid a race condition.
6147 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6148 * proposal and current concerns. Perhaps we should be setting
6149 * KVM_REQ_EVENT only on certain events and not unconditionally?
6151 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6152 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6156 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6157 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6159 kvm_x86_ops->set_irq(vcpu);
6166 static void process_nmi(struct kvm_vcpu *vcpu)
6171 * x86 is limited to one NMI running, and one NMI pending after it.
6172 * If an NMI is already in progress, limit further NMIs to just one.
6173 * Otherwise, allow two (and we'll inject the first one immediately).
6175 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6178 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6179 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6180 kvm_make_request(KVM_REQ_EVENT, vcpu);
6183 #define put_smstate(type, buf, offset, val) \
6184 *(type *)((buf) + (offset) - 0x7e00) = val
6186 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
6189 flags |= seg->g << 23;
6190 flags |= seg->db << 22;
6191 flags |= seg->l << 21;
6192 flags |= seg->avl << 20;
6193 flags |= seg->present << 15;
6194 flags |= seg->dpl << 13;
6195 flags |= seg->s << 12;
6196 flags |= seg->type << 8;
6200 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6202 struct kvm_segment seg;
6205 kvm_get_segment(vcpu, &seg, n);
6206 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6209 offset = 0x7f84 + n * 12;
6211 offset = 0x7f2c + (n - 3) * 12;
6213 put_smstate(u32, buf, offset + 8, seg.base);
6214 put_smstate(u32, buf, offset + 4, seg.limit);
6215 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
6218 #ifdef CONFIG_X86_64
6219 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6221 struct kvm_segment seg;
6225 kvm_get_segment(vcpu, &seg, n);
6226 offset = 0x7e00 + n * 16;
6228 flags = enter_smm_get_segment_flags(&seg) >> 8;
6229 put_smstate(u16, buf, offset, seg.selector);
6230 put_smstate(u16, buf, offset + 2, flags);
6231 put_smstate(u32, buf, offset + 4, seg.limit);
6232 put_smstate(u64, buf, offset + 8, seg.base);
6236 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6239 struct kvm_segment seg;
6243 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6244 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6245 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6246 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6248 for (i = 0; i < 8; i++)
6249 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6251 kvm_get_dr(vcpu, 6, &val);
6252 put_smstate(u32, buf, 0x7fcc, (u32)val);
6253 kvm_get_dr(vcpu, 7, &val);
6254 put_smstate(u32, buf, 0x7fc8, (u32)val);
6256 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6257 put_smstate(u32, buf, 0x7fc4, seg.selector);
6258 put_smstate(u32, buf, 0x7f64, seg.base);
6259 put_smstate(u32, buf, 0x7f60, seg.limit);
6260 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
6262 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6263 put_smstate(u32, buf, 0x7fc0, seg.selector);
6264 put_smstate(u32, buf, 0x7f80, seg.base);
6265 put_smstate(u32, buf, 0x7f7c, seg.limit);
6266 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
6268 kvm_x86_ops->get_gdt(vcpu, &dt);
6269 put_smstate(u32, buf, 0x7f74, dt.address);
6270 put_smstate(u32, buf, 0x7f70, dt.size);
6272 kvm_x86_ops->get_idt(vcpu, &dt);
6273 put_smstate(u32, buf, 0x7f58, dt.address);
6274 put_smstate(u32, buf, 0x7f54, dt.size);
6276 for (i = 0; i < 6; i++)
6277 enter_smm_save_seg_32(vcpu, buf, i);
6279 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6282 put_smstate(u32, buf, 0x7efc, 0x00020000);
6283 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6286 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6288 #ifdef CONFIG_X86_64
6290 struct kvm_segment seg;
6294 for (i = 0; i < 16; i++)
6295 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6297 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6298 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6300 kvm_get_dr(vcpu, 6, &val);
6301 put_smstate(u64, buf, 0x7f68, val);
6302 kvm_get_dr(vcpu, 7, &val);
6303 put_smstate(u64, buf, 0x7f60, val);
6305 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6306 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6307 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6309 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6312 put_smstate(u32, buf, 0x7efc, 0x00020064);
6314 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6316 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6317 put_smstate(u16, buf, 0x7e90, seg.selector);
6318 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
6319 put_smstate(u32, buf, 0x7e94, seg.limit);
6320 put_smstate(u64, buf, 0x7e98, seg.base);
6322 kvm_x86_ops->get_idt(vcpu, &dt);
6323 put_smstate(u32, buf, 0x7e84, dt.size);
6324 put_smstate(u64, buf, 0x7e88, dt.address);
6326 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6327 put_smstate(u16, buf, 0x7e70, seg.selector);
6328 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
6329 put_smstate(u32, buf, 0x7e74, seg.limit);
6330 put_smstate(u64, buf, 0x7e78, seg.base);
6332 kvm_x86_ops->get_gdt(vcpu, &dt);
6333 put_smstate(u32, buf, 0x7e64, dt.size);
6334 put_smstate(u64, buf, 0x7e68, dt.address);
6336 for (i = 0; i < 6; i++)
6337 enter_smm_save_seg_64(vcpu, buf, i);
6343 static void enter_smm(struct kvm_vcpu *vcpu)
6345 struct kvm_segment cs, ds;
6350 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6351 vcpu->arch.hflags |= HF_SMM_MASK;
6352 memset(buf, 0, 512);
6353 if (guest_cpuid_has_longmode(vcpu))
6354 enter_smm_save_state_64(vcpu, buf);
6356 enter_smm_save_state_32(vcpu, buf);
6358 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6360 if (kvm_x86_ops->get_nmi_mask(vcpu))
6361 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6363 kvm_x86_ops->set_nmi_mask(vcpu, true);
6365 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6366 kvm_rip_write(vcpu, 0x8000);
6368 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6369 kvm_x86_ops->set_cr0(vcpu, cr0);
6370 vcpu->arch.cr0 = cr0;
6372 kvm_x86_ops->set_cr4(vcpu, 0);
6374 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6375 dt.address = dt.size = 0;
6376 kvm_x86_ops->set_idt(vcpu, &dt);
6378 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6380 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6381 cs.base = vcpu->arch.smbase;
6386 cs.limit = ds.limit = 0xffffffff;
6387 cs.type = ds.type = 0x3;
6388 cs.dpl = ds.dpl = 0;
6393 cs.avl = ds.avl = 0;
6394 cs.present = ds.present = 1;
6395 cs.unusable = ds.unusable = 0;
6396 cs.padding = ds.padding = 0;
6398 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6399 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6400 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6401 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6402 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6403 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6405 if (guest_cpuid_has_longmode(vcpu))
6406 kvm_x86_ops->set_efer(vcpu, 0);
6408 kvm_update_cpuid(vcpu);
6409 kvm_mmu_reset_context(vcpu);
6412 static void process_smi(struct kvm_vcpu *vcpu)
6414 vcpu->arch.smi_pending = true;
6415 kvm_make_request(KVM_REQ_EVENT, vcpu);
6418 void kvm_make_scan_ioapic_request(struct kvm *kvm)
6420 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6423 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6425 u64 eoi_exit_bitmap[4];
6427 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6430 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
6432 if (irqchip_split(vcpu->kvm))
6433 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
6435 if (vcpu->arch.apicv_active)
6436 kvm_x86_ops->sync_pir_to_irr(vcpu);
6437 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
6439 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6440 vcpu_to_synic(vcpu)->vec_bitmap, 256);
6441 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6444 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6446 ++vcpu->stat.tlb_flush;
6447 kvm_x86_ops->tlb_flush(vcpu);
6450 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6452 struct page *page = NULL;
6454 if (!lapic_in_kernel(vcpu))
6457 if (!kvm_x86_ops->set_apic_access_page_addr)
6460 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6461 if (is_error_page(page))
6463 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6466 * Do not pin apic access page in memory, the MMU notifier
6467 * will call us again if it is migrated or swapped out.
6471 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6473 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6474 unsigned long address)
6477 * The physical address of apic access page is stored in the VMCS.
6478 * Update it when it becomes invalid.
6480 if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6481 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6485 * Returns 1 to let vcpu_run() continue the guest execution loop without
6486 * exiting to the userspace. Otherwise, the value will be returned to the
6489 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6493 dm_request_for_irq_injection(vcpu) &&
6494 kvm_cpu_accept_dm_intr(vcpu);
6496 bool req_immediate_exit = false;
6498 if (vcpu->requests) {
6499 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6500 kvm_mmu_unload(vcpu);
6501 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6502 __kvm_migrate_timers(vcpu);
6503 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6504 kvm_gen_update_masterclock(vcpu->kvm);
6505 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6506 kvm_gen_kvmclock_update(vcpu);
6507 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6508 r = kvm_guest_time_update(vcpu);
6512 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6513 kvm_mmu_sync_roots(vcpu);
6514 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6515 kvm_vcpu_flush_tlb(vcpu);
6516 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6517 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6521 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6522 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6526 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
6527 vcpu->fpu_active = 0;
6528 kvm_x86_ops->fpu_deactivate(vcpu);
6530 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6531 /* Page is swapped out. Do synthetic halt */
6532 vcpu->arch.apf.halted = true;
6536 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6537 record_steal_time(vcpu);
6538 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6540 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6542 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6543 kvm_pmu_handle_event(vcpu);
6544 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6545 kvm_pmu_deliver_pmi(vcpu);
6546 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6547 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6548 if (test_bit(vcpu->arch.pending_ioapic_eoi,
6549 vcpu->arch.ioapic_handled_vectors)) {
6550 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6551 vcpu->run->eoi.vector =
6552 vcpu->arch.pending_ioapic_eoi;
6557 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6558 vcpu_scan_ioapic(vcpu);
6559 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6560 kvm_vcpu_reload_apic_access_page(vcpu);
6561 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6562 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6563 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6567 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6568 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6569 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6573 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6574 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6575 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6581 * KVM_REQ_HV_STIMER has to be processed after
6582 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6583 * depend on the guest clock being up-to-date
6585 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6586 kvm_hv_process_stimers(vcpu);
6590 * KVM_REQ_EVENT is not set when posted interrupts are set by
6591 * VT-d hardware, so we have to update RVI unconditionally.
6593 if (kvm_lapic_enabled(vcpu)) {
6595 * Update architecture specific hints for APIC
6596 * virtual interrupt delivery.
6598 if (vcpu->arch.apicv_active)
6599 kvm_x86_ops->hwapic_irr_update(vcpu,
6600 kvm_lapic_find_highest_irr(vcpu));
6603 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6604 kvm_apic_accept_events(vcpu);
6605 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6610 if (inject_pending_event(vcpu, req_int_win) != 0)
6611 req_immediate_exit = true;
6613 /* Enable NMI/IRQ window open exits if needed.
6615 * SMIs have two cases: 1) they can be nested, and
6616 * then there is nothing to do here because RSM will
6617 * cause a vmexit anyway; 2) or the SMI can be pending
6618 * because inject_pending_event has completed the
6619 * injection of an IRQ or NMI from the previous vmexit,
6620 * and then we request an immediate exit to inject the SMI.
6622 if (vcpu->arch.smi_pending && !is_smm(vcpu))
6623 req_immediate_exit = true;
6624 if (vcpu->arch.nmi_pending)
6625 kvm_x86_ops->enable_nmi_window(vcpu);
6626 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6627 kvm_x86_ops->enable_irq_window(vcpu);
6630 if (kvm_lapic_enabled(vcpu)) {
6631 update_cr8_intercept(vcpu);
6632 kvm_lapic_sync_to_vapic(vcpu);
6636 r = kvm_mmu_reload(vcpu);
6638 goto cancel_injection;
6643 kvm_x86_ops->prepare_guest_switch(vcpu);
6644 if (vcpu->fpu_active)
6645 kvm_load_guest_fpu(vcpu);
6646 vcpu->mode = IN_GUEST_MODE;
6648 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6651 * We should set ->mode before check ->requests,
6652 * Please see the comment in kvm_make_all_cpus_request.
6653 * This also orders the write to mode from any reads
6654 * to the page tables done while the VCPU is running.
6655 * Please see the comment in kvm_flush_remote_tlbs.
6657 smp_mb__after_srcu_read_unlock();
6659 local_irq_disable();
6661 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6662 || need_resched() || signal_pending(current)) {
6663 vcpu->mode = OUTSIDE_GUEST_MODE;
6667 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6669 goto cancel_injection;
6672 kvm_load_guest_xcr0(vcpu);
6674 if (req_immediate_exit) {
6675 kvm_make_request(KVM_REQ_EVENT, vcpu);
6676 smp_send_reschedule(vcpu->cpu);
6679 trace_kvm_entry(vcpu->vcpu_id);
6680 wait_lapic_expire(vcpu);
6681 guest_enter_irqoff();
6683 if (unlikely(vcpu->arch.switch_db_regs)) {
6685 set_debugreg(vcpu->arch.eff_db[0], 0);
6686 set_debugreg(vcpu->arch.eff_db[1], 1);
6687 set_debugreg(vcpu->arch.eff_db[2], 2);
6688 set_debugreg(vcpu->arch.eff_db[3], 3);
6689 set_debugreg(vcpu->arch.dr6, 6);
6690 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6693 kvm_x86_ops->run(vcpu);
6696 * Do this here before restoring debug registers on the host. And
6697 * since we do this before handling the vmexit, a DR access vmexit
6698 * can (a) read the correct value of the debug registers, (b) set
6699 * KVM_DEBUGREG_WONT_EXIT again.
6701 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6702 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6703 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6704 kvm_update_dr0123(vcpu);
6705 kvm_update_dr6(vcpu);
6706 kvm_update_dr7(vcpu);
6707 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6711 * If the guest has used debug registers, at least dr7
6712 * will be disabled while returning to the host.
6713 * If we don't have active breakpoints in the host, we don't
6714 * care about the messed up debug address registers. But if
6715 * we have some of them active, restore the old state.
6717 if (hw_breakpoint_active())
6718 hw_breakpoint_restore();
6720 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
6722 vcpu->mode = OUTSIDE_GUEST_MODE;
6725 kvm_put_guest_xcr0(vcpu);
6727 /* Interrupt is enabled by handle_external_intr() */
6728 kvm_x86_ops->handle_external_intr(vcpu);
6732 guest_exit_irqoff();
6737 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6740 * Profile KVM exit RIPs:
6742 if (unlikely(prof_on == KVM_PROFILING)) {
6743 unsigned long rip = kvm_rip_read(vcpu);
6744 profile_hit(KVM_PROFILING, (void *)rip);
6747 if (unlikely(vcpu->arch.tsc_always_catchup))
6748 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6750 if (vcpu->arch.apic_attention)
6751 kvm_lapic_sync_from_vapic(vcpu);
6753 r = kvm_x86_ops->handle_exit(vcpu);
6757 kvm_x86_ops->cancel_injection(vcpu);
6758 if (unlikely(vcpu->arch.apic_attention))
6759 kvm_lapic_sync_from_vapic(vcpu);
6764 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6766 if (!kvm_arch_vcpu_runnable(vcpu) &&
6767 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
6768 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6769 kvm_vcpu_block(vcpu);
6770 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6772 if (kvm_x86_ops->post_block)
6773 kvm_x86_ops->post_block(vcpu);
6775 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6779 kvm_apic_accept_events(vcpu);
6780 switch(vcpu->arch.mp_state) {
6781 case KVM_MP_STATE_HALTED:
6782 vcpu->arch.pv.pv_unhalted = false;
6783 vcpu->arch.mp_state =
6784 KVM_MP_STATE_RUNNABLE;
6785 case KVM_MP_STATE_RUNNABLE:
6786 vcpu->arch.apf.halted = false;
6788 case KVM_MP_STATE_INIT_RECEIVED:
6797 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
6799 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
6800 !vcpu->arch.apf.halted);
6803 static int vcpu_run(struct kvm_vcpu *vcpu)
6806 struct kvm *kvm = vcpu->kvm;
6808 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6811 if (kvm_vcpu_running(vcpu)) {
6812 r = vcpu_enter_guest(vcpu);
6814 r = vcpu_block(kvm, vcpu);
6820 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
6821 if (kvm_cpu_has_pending_timer(vcpu))
6822 kvm_inject_pending_timer_irqs(vcpu);
6824 if (dm_request_for_irq_injection(vcpu) &&
6825 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
6827 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
6828 ++vcpu->stat.request_irq_exits;
6832 kvm_check_async_pf_completion(vcpu);
6834 if (signal_pending(current)) {
6836 vcpu->run->exit_reason = KVM_EXIT_INTR;
6837 ++vcpu->stat.signal_exits;
6840 if (need_resched()) {
6841 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6843 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6847 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6852 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
6855 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6856 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
6857 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6858 if (r != EMULATE_DONE)
6863 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
6865 BUG_ON(!vcpu->arch.pio.count);
6867 return complete_emulated_io(vcpu);
6871 * Implements the following, as a state machine:
6875 * for each mmio piece in the fragment
6883 * for each mmio piece in the fragment
6888 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
6890 struct kvm_run *run = vcpu->run;
6891 struct kvm_mmio_fragment *frag;
6894 BUG_ON(!vcpu->mmio_needed);
6896 /* Complete previous fragment */
6897 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
6898 len = min(8u, frag->len);
6899 if (!vcpu->mmio_is_write)
6900 memcpy(frag->data, run->mmio.data, len);
6902 if (frag->len <= 8) {
6903 /* Switch to the next fragment. */
6905 vcpu->mmio_cur_fragment++;
6907 /* Go forward to the next mmio piece. */
6913 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
6914 vcpu->mmio_needed = 0;
6916 /* FIXME: return into emulator if single-stepping. */
6917 if (vcpu->mmio_is_write)
6919 vcpu->mmio_read_completed = 1;
6920 return complete_emulated_io(vcpu);
6923 run->exit_reason = KVM_EXIT_MMIO;
6924 run->mmio.phys_addr = frag->gpa;
6925 if (vcpu->mmio_is_write)
6926 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
6927 run->mmio.len = min(8u, frag->len);
6928 run->mmio.is_write = vcpu->mmio_is_write;
6929 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6934 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
6936 struct fpu *fpu = ¤t->thread.fpu;
6940 fpu__activate_curr(fpu);
6942 if (vcpu->sigset_active)
6943 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
6945 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
6946 kvm_vcpu_block(vcpu);
6947 kvm_apic_accept_events(vcpu);
6948 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
6953 /* re-sync apic's tpr */
6954 if (!lapic_in_kernel(vcpu)) {
6955 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
6961 if (unlikely(vcpu->arch.complete_userspace_io)) {
6962 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
6963 vcpu->arch.complete_userspace_io = NULL;
6968 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
6973 post_kvm_run_save(vcpu);
6974 if (vcpu->sigset_active)
6975 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
6980 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
6982 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
6984 * We are here if userspace calls get_regs() in the middle of
6985 * instruction emulation. Registers state needs to be copied
6986 * back from emulation context to vcpu. Userspace shouldn't do
6987 * that usually, but some bad designed PV devices (vmware
6988 * backdoor interface) need this to work
6990 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
6991 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6993 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
6994 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
6995 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
6996 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
6997 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
6998 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
6999 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7000 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
7001 #ifdef CONFIG_X86_64
7002 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7003 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7004 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7005 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7006 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7007 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7008 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7009 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
7012 regs->rip = kvm_rip_read(vcpu);
7013 regs->rflags = kvm_get_rflags(vcpu);
7018 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7020 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7021 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7023 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7024 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7025 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7026 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7027 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7028 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7029 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7030 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
7031 #ifdef CONFIG_X86_64
7032 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7033 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7034 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7035 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7036 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7037 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7038 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7039 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
7042 kvm_rip_write(vcpu, regs->rip);
7043 kvm_set_rflags(vcpu, regs->rflags);
7045 vcpu->arch.exception.pending = false;
7047 kvm_make_request(KVM_REQ_EVENT, vcpu);
7052 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7054 struct kvm_segment cs;
7056 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7060 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7062 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7063 struct kvm_sregs *sregs)
7067 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7068 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7069 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7070 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7071 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7072 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7074 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7075 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7077 kvm_x86_ops->get_idt(vcpu, &dt);
7078 sregs->idt.limit = dt.size;
7079 sregs->idt.base = dt.address;
7080 kvm_x86_ops->get_gdt(vcpu, &dt);
7081 sregs->gdt.limit = dt.size;
7082 sregs->gdt.base = dt.address;
7084 sregs->cr0 = kvm_read_cr0(vcpu);
7085 sregs->cr2 = vcpu->arch.cr2;
7086 sregs->cr3 = kvm_read_cr3(vcpu);
7087 sregs->cr4 = kvm_read_cr4(vcpu);
7088 sregs->cr8 = kvm_get_cr8(vcpu);
7089 sregs->efer = vcpu->arch.efer;
7090 sregs->apic_base = kvm_get_apic_base(vcpu);
7092 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7094 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
7095 set_bit(vcpu->arch.interrupt.nr,
7096 (unsigned long *)sregs->interrupt_bitmap);
7101 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7102 struct kvm_mp_state *mp_state)
7104 kvm_apic_accept_events(vcpu);
7105 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7106 vcpu->arch.pv.pv_unhalted)
7107 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7109 mp_state->mp_state = vcpu->arch.mp_state;
7114 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7115 struct kvm_mp_state *mp_state)
7117 if (!lapic_in_kernel(vcpu) &&
7118 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7121 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7122 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7123 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7125 vcpu->arch.mp_state = mp_state->mp_state;
7126 kvm_make_request(KVM_REQ_EVENT, vcpu);
7130 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7131 int reason, bool has_error_code, u32 error_code)
7133 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7136 init_emulate_ctxt(vcpu);
7138 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7139 has_error_code, error_code);
7142 return EMULATE_FAIL;
7144 kvm_rip_write(vcpu, ctxt->eip);
7145 kvm_set_rflags(vcpu, ctxt->eflags);
7146 kvm_make_request(KVM_REQ_EVENT, vcpu);
7147 return EMULATE_DONE;
7149 EXPORT_SYMBOL_GPL(kvm_task_switch);
7151 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7152 struct kvm_sregs *sregs)
7154 struct msr_data apic_base_msr;
7155 int mmu_reset_needed = 0;
7156 int pending_vec, max_bits, idx;
7159 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7162 dt.size = sregs->idt.limit;
7163 dt.address = sregs->idt.base;
7164 kvm_x86_ops->set_idt(vcpu, &dt);
7165 dt.size = sregs->gdt.limit;
7166 dt.address = sregs->gdt.base;
7167 kvm_x86_ops->set_gdt(vcpu, &dt);
7169 vcpu->arch.cr2 = sregs->cr2;
7170 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7171 vcpu->arch.cr3 = sregs->cr3;
7172 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7174 kvm_set_cr8(vcpu, sregs->cr8);
7176 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7177 kvm_x86_ops->set_efer(vcpu, sregs->efer);
7178 apic_base_msr.data = sregs->apic_base;
7179 apic_base_msr.host_initiated = true;
7180 kvm_set_apic_base(vcpu, &apic_base_msr);
7182 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7183 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7184 vcpu->arch.cr0 = sregs->cr0;
7186 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7187 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7188 if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
7189 kvm_update_cpuid(vcpu);
7191 idx = srcu_read_lock(&vcpu->kvm->srcu);
7192 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7193 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7194 mmu_reset_needed = 1;
7196 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7198 if (mmu_reset_needed)
7199 kvm_mmu_reset_context(vcpu);
7201 max_bits = KVM_NR_INTERRUPTS;
7202 pending_vec = find_first_bit(
7203 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7204 if (pending_vec < max_bits) {
7205 kvm_queue_interrupt(vcpu, pending_vec, false);
7206 pr_debug("Set back pending irq %d\n", pending_vec);
7209 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7210 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7211 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7212 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7213 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7214 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7216 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7217 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7219 update_cr8_intercept(vcpu);
7221 /* Older userspace won't unhalt the vcpu on reset. */
7222 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7223 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7225 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7227 kvm_make_request(KVM_REQ_EVENT, vcpu);
7232 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7233 struct kvm_guest_debug *dbg)
7235 unsigned long rflags;
7238 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7240 if (vcpu->arch.exception.pending)
7242 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7243 kvm_queue_exception(vcpu, DB_VECTOR);
7245 kvm_queue_exception(vcpu, BP_VECTOR);
7249 * Read rflags as long as potentially injected trace flags are still
7252 rflags = kvm_get_rflags(vcpu);
7254 vcpu->guest_debug = dbg->control;
7255 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7256 vcpu->guest_debug = 0;
7258 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7259 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7260 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7261 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7263 for (i = 0; i < KVM_NR_DB_REGS; i++)
7264 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7266 kvm_update_dr7(vcpu);
7268 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7269 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7270 get_segment_base(vcpu, VCPU_SREG_CS);
7273 * Trigger an rflags update that will inject or remove the trace
7276 kvm_set_rflags(vcpu, rflags);
7278 kvm_x86_ops->update_bp_intercept(vcpu);
7288 * Translate a guest virtual address to a guest physical address.
7290 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7291 struct kvm_translation *tr)
7293 unsigned long vaddr = tr->linear_address;
7297 idx = srcu_read_lock(&vcpu->kvm->srcu);
7298 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7299 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7300 tr->physical_address = gpa;
7301 tr->valid = gpa != UNMAPPED_GVA;
7308 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7310 struct fxregs_state *fxsave =
7311 &vcpu->arch.guest_fpu.state.fxsave;
7313 memcpy(fpu->fpr, fxsave->st_space, 128);
7314 fpu->fcw = fxsave->cwd;
7315 fpu->fsw = fxsave->swd;
7316 fpu->ftwx = fxsave->twd;
7317 fpu->last_opcode = fxsave->fop;
7318 fpu->last_ip = fxsave->rip;
7319 fpu->last_dp = fxsave->rdp;
7320 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7325 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7327 struct fxregs_state *fxsave =
7328 &vcpu->arch.guest_fpu.state.fxsave;
7330 memcpy(fxsave->st_space, fpu->fpr, 128);
7331 fxsave->cwd = fpu->fcw;
7332 fxsave->swd = fpu->fsw;
7333 fxsave->twd = fpu->ftwx;
7334 fxsave->fop = fpu->last_opcode;
7335 fxsave->rip = fpu->last_ip;
7336 fxsave->rdp = fpu->last_dp;
7337 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7342 static void fx_init(struct kvm_vcpu *vcpu)
7344 fpstate_init(&vcpu->arch.guest_fpu.state);
7345 if (boot_cpu_has(X86_FEATURE_XSAVES))
7346 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7347 host_xcr0 | XSTATE_COMPACTION_ENABLED;
7350 * Ensure guest xcr0 is valid for loading
7352 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7354 vcpu->arch.cr0 |= X86_CR0_ET;
7357 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7359 if (vcpu->guest_fpu_loaded)
7363 * Restore all possible states in the guest,
7364 * and assume host would use all available bits.
7365 * Guest xcr0 would be loaded later.
7367 vcpu->guest_fpu_loaded = 1;
7368 __kernel_fpu_begin();
7369 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7373 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7375 if (!vcpu->guest_fpu_loaded) {
7376 vcpu->fpu_counter = 0;
7380 vcpu->guest_fpu_loaded = 0;
7381 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7383 ++vcpu->stat.fpu_reload;
7385 * If using eager FPU mode, or if the guest is a frequent user
7386 * of the FPU, just leave the FPU active for next time.
7387 * Every 255 times fpu_counter rolls over to 0; a guest that uses
7388 * the FPU in bursts will revert to loading it on demand.
7390 if (!use_eager_fpu()) {
7391 if (++vcpu->fpu_counter < 5)
7392 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
7397 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7399 kvmclock_reset(vcpu);
7401 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
7402 kvm_x86_ops->vcpu_free(vcpu);
7405 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7408 struct kvm_vcpu *vcpu;
7410 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7411 printk_once(KERN_WARNING
7412 "kvm: SMP vm created on host with unstable TSC; "
7413 "guest TSC will not be reliable\n");
7415 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7420 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7424 kvm_vcpu_mtrr_init(vcpu);
7425 r = vcpu_load(vcpu);
7428 kvm_vcpu_reset(vcpu, false);
7429 kvm_mmu_setup(vcpu);
7434 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7436 struct msr_data msr;
7437 struct kvm *kvm = vcpu->kvm;
7439 if (vcpu_load(vcpu))
7442 msr.index = MSR_IA32_TSC;
7443 msr.host_initiated = true;
7444 kvm_write_tsc(vcpu, &msr);
7447 if (!kvmclock_periodic_sync)
7450 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7451 KVMCLOCK_SYNC_PERIOD);
7454 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7457 vcpu->arch.apf.msr_val = 0;
7459 r = vcpu_load(vcpu);
7461 kvm_mmu_unload(vcpu);
7464 kvm_x86_ops->vcpu_free(vcpu);
7467 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7469 vcpu->arch.hflags = 0;
7471 vcpu->arch.smi_pending = 0;
7472 atomic_set(&vcpu->arch.nmi_queued, 0);
7473 vcpu->arch.nmi_pending = 0;
7474 vcpu->arch.nmi_injected = false;
7475 kvm_clear_interrupt_queue(vcpu);
7476 kvm_clear_exception_queue(vcpu);
7478 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7479 kvm_update_dr0123(vcpu);
7480 vcpu->arch.dr6 = DR6_INIT;
7481 kvm_update_dr6(vcpu);
7482 vcpu->arch.dr7 = DR7_FIXED_1;
7483 kvm_update_dr7(vcpu);
7487 kvm_make_request(KVM_REQ_EVENT, vcpu);
7488 vcpu->arch.apf.msr_val = 0;
7489 vcpu->arch.st.msr_val = 0;
7491 kvmclock_reset(vcpu);
7493 kvm_clear_async_pf_completion_queue(vcpu);
7494 kvm_async_pf_hash_reset(vcpu);
7495 vcpu->arch.apf.halted = false;
7498 kvm_pmu_reset(vcpu);
7499 vcpu->arch.smbase = 0x30000;
7502 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7503 vcpu->arch.regs_avail = ~0;
7504 vcpu->arch.regs_dirty = ~0;
7506 kvm_x86_ops->vcpu_reset(vcpu, init_event);
7509 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7511 struct kvm_segment cs;
7513 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7514 cs.selector = vector << 8;
7515 cs.base = vector << 12;
7516 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7517 kvm_rip_write(vcpu, 0);
7520 int kvm_arch_hardware_enable(void)
7523 struct kvm_vcpu *vcpu;
7528 bool stable, backwards_tsc = false;
7530 kvm_shared_msr_cpu_online();
7531 ret = kvm_x86_ops->hardware_enable();
7535 local_tsc = rdtsc();
7536 stable = !check_tsc_unstable();
7537 list_for_each_entry(kvm, &vm_list, vm_list) {
7538 kvm_for_each_vcpu(i, vcpu, kvm) {
7539 if (!stable && vcpu->cpu == smp_processor_id())
7540 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7541 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7542 backwards_tsc = true;
7543 if (vcpu->arch.last_host_tsc > max_tsc)
7544 max_tsc = vcpu->arch.last_host_tsc;
7550 * Sometimes, even reliable TSCs go backwards. This happens on
7551 * platforms that reset TSC during suspend or hibernate actions, but
7552 * maintain synchronization. We must compensate. Fortunately, we can
7553 * detect that condition here, which happens early in CPU bringup,
7554 * before any KVM threads can be running. Unfortunately, we can't
7555 * bring the TSCs fully up to date with real time, as we aren't yet far
7556 * enough into CPU bringup that we know how much real time has actually
7557 * elapsed; our helper function, get_kernel_ns() will be using boot
7558 * variables that haven't been updated yet.
7560 * So we simply find the maximum observed TSC above, then record the
7561 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7562 * the adjustment will be applied. Note that we accumulate
7563 * adjustments, in case multiple suspend cycles happen before some VCPU
7564 * gets a chance to run again. In the event that no KVM threads get a
7565 * chance to run, we will miss the entire elapsed period, as we'll have
7566 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7567 * loose cycle time. This isn't too big a deal, since the loss will be
7568 * uniform across all VCPUs (not to mention the scenario is extremely
7569 * unlikely). It is possible that a second hibernate recovery happens
7570 * much faster than a first, causing the observed TSC here to be
7571 * smaller; this would require additional padding adjustment, which is
7572 * why we set last_host_tsc to the local tsc observed here.
7574 * N.B. - this code below runs only on platforms with reliable TSC,
7575 * as that is the only way backwards_tsc is set above. Also note
7576 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7577 * have the same delta_cyc adjustment applied if backwards_tsc
7578 * is detected. Note further, this adjustment is only done once,
7579 * as we reset last_host_tsc on all VCPUs to stop this from being
7580 * called multiple times (one for each physical CPU bringup).
7582 * Platforms with unreliable TSCs don't have to deal with this, they
7583 * will be compensated by the logic in vcpu_load, which sets the TSC to
7584 * catchup mode. This will catchup all VCPUs to real time, but cannot
7585 * guarantee that they stay in perfect synchronization.
7587 if (backwards_tsc) {
7588 u64 delta_cyc = max_tsc - local_tsc;
7589 backwards_tsc_observed = true;
7590 list_for_each_entry(kvm, &vm_list, vm_list) {
7591 kvm_for_each_vcpu(i, vcpu, kvm) {
7592 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7593 vcpu->arch.last_host_tsc = local_tsc;
7594 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7598 * We have to disable TSC offset matching.. if you were
7599 * booting a VM while issuing an S4 host suspend....
7600 * you may have some problem. Solving this issue is
7601 * left as an exercise to the reader.
7603 kvm->arch.last_tsc_nsec = 0;
7604 kvm->arch.last_tsc_write = 0;
7611 void kvm_arch_hardware_disable(void)
7613 kvm_x86_ops->hardware_disable();
7614 drop_user_return_notifiers();
7617 int kvm_arch_hardware_setup(void)
7621 r = kvm_x86_ops->hardware_setup();
7625 if (kvm_has_tsc_control) {
7627 * Make sure the user can only configure tsc_khz values that
7628 * fit into a signed integer.
7629 * A min value is not calculated needed because it will always
7630 * be 1 on all machines.
7632 u64 max = min(0x7fffffffULL,
7633 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
7634 kvm_max_guest_tsc_khz = max;
7636 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
7639 kvm_init_msr_list();
7643 void kvm_arch_hardware_unsetup(void)
7645 kvm_x86_ops->hardware_unsetup();
7648 void kvm_arch_check_processor_compat(void *rtn)
7650 kvm_x86_ops->check_processor_compatibility(rtn);
7653 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7655 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7657 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7659 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7661 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7664 struct static_key kvm_no_apic_vcpu __read_mostly;
7665 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
7667 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7673 BUG_ON(vcpu->kvm == NULL);
7676 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv();
7677 vcpu->arch.pv.pv_unhalted = false;
7678 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7679 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7680 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7682 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7684 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7689 vcpu->arch.pio_data = page_address(page);
7691 kvm_set_tsc_khz(vcpu, max_tsc_khz);
7693 r = kvm_mmu_create(vcpu);
7695 goto fail_free_pio_data;
7697 if (irqchip_in_kernel(kvm)) {
7698 r = kvm_create_lapic(vcpu);
7700 goto fail_mmu_destroy;
7702 static_key_slow_inc(&kvm_no_apic_vcpu);
7704 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7706 if (!vcpu->arch.mce_banks) {
7708 goto fail_free_lapic;
7710 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7712 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7714 goto fail_free_mce_banks;
7719 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7720 vcpu->arch.pv_time_enabled = false;
7722 vcpu->arch.guest_supported_xcr0 = 0;
7723 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7725 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7727 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7729 kvm_async_pf_hash_reset(vcpu);
7732 vcpu->arch.pending_external_vector = -1;
7734 kvm_hv_vcpu_init(vcpu);
7738 fail_free_mce_banks:
7739 kfree(vcpu->arch.mce_banks);
7741 kvm_free_lapic(vcpu);
7743 kvm_mmu_destroy(vcpu);
7745 free_page((unsigned long)vcpu->arch.pio_data);
7750 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7754 kvm_hv_vcpu_uninit(vcpu);
7755 kvm_pmu_destroy(vcpu);
7756 kfree(vcpu->arch.mce_banks);
7757 kvm_free_lapic(vcpu);
7758 idx = srcu_read_lock(&vcpu->kvm->srcu);
7759 kvm_mmu_destroy(vcpu);
7760 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7761 free_page((unsigned long)vcpu->arch.pio_data);
7762 if (!lapic_in_kernel(vcpu))
7763 static_key_slow_dec(&kvm_no_apic_vcpu);
7766 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7768 kvm_x86_ops->sched_in(vcpu, cpu);
7771 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
7776 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
7777 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
7778 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
7779 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
7780 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
7782 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
7783 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
7784 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
7785 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
7786 &kvm->arch.irq_sources_bitmap);
7788 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
7789 mutex_init(&kvm->arch.apic_map_lock);
7790 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
7792 pvclock_update_vm_gtod_copy(kvm);
7794 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
7795 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
7797 kvm_page_track_init(kvm);
7798 kvm_mmu_init_vm(kvm);
7800 if (kvm_x86_ops->vm_init)
7801 return kvm_x86_ops->vm_init(kvm);
7806 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
7809 r = vcpu_load(vcpu);
7811 kvm_mmu_unload(vcpu);
7815 static void kvm_free_vcpus(struct kvm *kvm)
7818 struct kvm_vcpu *vcpu;
7821 * Unpin any mmu pages first.
7823 kvm_for_each_vcpu(i, vcpu, kvm) {
7824 kvm_clear_async_pf_completion_queue(vcpu);
7825 kvm_unload_vcpu_mmu(vcpu);
7827 kvm_for_each_vcpu(i, vcpu, kvm)
7828 kvm_arch_vcpu_free(vcpu);
7830 mutex_lock(&kvm->lock);
7831 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
7832 kvm->vcpus[i] = NULL;
7834 atomic_set(&kvm->online_vcpus, 0);
7835 mutex_unlock(&kvm->lock);
7838 void kvm_arch_sync_events(struct kvm *kvm)
7840 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
7841 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
7842 kvm_free_all_assigned_devices(kvm);
7846 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7850 struct kvm_memslots *slots = kvm_memslots(kvm);
7851 struct kvm_memory_slot *slot, old;
7853 /* Called with kvm->slots_lock held. */
7854 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
7857 slot = id_to_memslot(slots, id);
7863 * MAP_SHARED to prevent internal slot pages from being moved
7866 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
7867 MAP_SHARED | MAP_ANONYMOUS, 0);
7868 if (IS_ERR((void *)hva))
7869 return PTR_ERR((void *)hva);
7878 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
7879 struct kvm_userspace_memory_region m;
7881 m.slot = id | (i << 16);
7883 m.guest_phys_addr = gpa;
7884 m.userspace_addr = hva;
7885 m.memory_size = size;
7886 r = __kvm_set_memory_region(kvm, &m);
7892 r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
7898 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
7900 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
7904 mutex_lock(&kvm->slots_lock);
7905 r = __x86_set_memory_region(kvm, id, gpa, size);
7906 mutex_unlock(&kvm->slots_lock);
7910 EXPORT_SYMBOL_GPL(x86_set_memory_region);
7912 void kvm_arch_destroy_vm(struct kvm *kvm)
7914 if (current->mm == kvm->mm) {
7916 * Free memory regions allocated on behalf of userspace,
7917 * unless the the memory map has changed due to process exit
7920 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
7921 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
7922 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
7924 if (kvm_x86_ops->vm_destroy)
7925 kvm_x86_ops->vm_destroy(kvm);
7926 kvm_iommu_unmap_guest(kvm);
7927 kfree(kvm->arch.vpic);
7928 kfree(kvm->arch.vioapic);
7929 kvm_free_vcpus(kvm);
7930 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
7931 kvm_mmu_uninit_vm(kvm);
7934 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
7935 struct kvm_memory_slot *dont)
7939 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7940 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
7941 kvfree(free->arch.rmap[i]);
7942 free->arch.rmap[i] = NULL;
7947 if (!dont || free->arch.lpage_info[i - 1] !=
7948 dont->arch.lpage_info[i - 1]) {
7949 kvfree(free->arch.lpage_info[i - 1]);
7950 free->arch.lpage_info[i - 1] = NULL;
7954 kvm_page_track_free_memslot(free, dont);
7957 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
7958 unsigned long npages)
7962 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
7963 struct kvm_lpage_info *linfo;
7968 lpages = gfn_to_index(slot->base_gfn + npages - 1,
7969 slot->base_gfn, level) + 1;
7971 slot->arch.rmap[i] =
7972 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
7973 if (!slot->arch.rmap[i])
7978 linfo = kvm_kvzalloc(lpages * sizeof(*linfo));
7982 slot->arch.lpage_info[i - 1] = linfo;
7984 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
7985 linfo[0].disallow_lpage = 1;
7986 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
7987 linfo[lpages - 1].disallow_lpage = 1;
7988 ugfn = slot->userspace_addr >> PAGE_SHIFT;
7990 * If the gfn and userspace address are not aligned wrt each
7991 * other, or if explicitly asked to, disable large page
7992 * support for this slot
7994 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
7995 !kvm_largepages_enabled()) {
7998 for (j = 0; j < lpages; ++j)
7999 linfo[j].disallow_lpage = 1;
8003 if (kvm_page_track_create_memslot(slot, npages))
8009 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8010 kvfree(slot->arch.rmap[i]);
8011 slot->arch.rmap[i] = NULL;
8015 kvfree(slot->arch.lpage_info[i - 1]);
8016 slot->arch.lpage_info[i - 1] = NULL;
8021 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
8024 * memslots->generation has been incremented.
8025 * mmio generation may have reached its maximum value.
8027 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
8030 int kvm_arch_prepare_memory_region(struct kvm *kvm,
8031 struct kvm_memory_slot *memslot,
8032 const struct kvm_userspace_memory_region *mem,
8033 enum kvm_mr_change change)
8038 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8039 struct kvm_memory_slot *new)
8041 /* Still write protect RO slot */
8042 if (new->flags & KVM_MEM_READONLY) {
8043 kvm_mmu_slot_remove_write_access(kvm, new);
8048 * Call kvm_x86_ops dirty logging hooks when they are valid.
8050 * kvm_x86_ops->slot_disable_log_dirty is called when:
8052 * - KVM_MR_CREATE with dirty logging is disabled
8053 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8055 * The reason is, in case of PML, we need to set D-bit for any slots
8056 * with dirty logging disabled in order to eliminate unnecessary GPA
8057 * logging in PML buffer (and potential PML buffer full VMEXT). This
8058 * guarantees leaving PML enabled during guest's lifetime won't have
8059 * any additonal overhead from PML when guest is running with dirty
8060 * logging disabled for memory slots.
8062 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8063 * to dirty logging mode.
8065 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8067 * In case of write protect:
8069 * Write protect all pages for dirty logging.
8071 * All the sptes including the large sptes which point to this
8072 * slot are set to readonly. We can not create any new large
8073 * spte on this slot until the end of the logging.
8075 * See the comments in fast_page_fault().
8077 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8078 if (kvm_x86_ops->slot_enable_log_dirty)
8079 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8081 kvm_mmu_slot_remove_write_access(kvm, new);
8083 if (kvm_x86_ops->slot_disable_log_dirty)
8084 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8088 void kvm_arch_commit_memory_region(struct kvm *kvm,
8089 const struct kvm_userspace_memory_region *mem,
8090 const struct kvm_memory_slot *old,
8091 const struct kvm_memory_slot *new,
8092 enum kvm_mr_change change)
8094 int nr_mmu_pages = 0;
8096 if (!kvm->arch.n_requested_mmu_pages)
8097 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8100 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
8103 * Dirty logging tracks sptes in 4k granularity, meaning that large
8104 * sptes have to be split. If live migration is successful, the guest
8105 * in the source machine will be destroyed and large sptes will be
8106 * created in the destination. However, if the guest continues to run
8107 * in the source machine (for example if live migration fails), small
8108 * sptes will remain around and cause bad performance.
8110 * Scan sptes if dirty logging has been stopped, dropping those
8111 * which can be collapsed into a single large-page spte. Later
8112 * page faults will create the large-page sptes.
8114 if ((change != KVM_MR_DELETE) &&
8115 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8116 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8117 kvm_mmu_zap_collapsible_sptes(kvm, new);
8120 * Set up write protection and/or dirty logging for the new slot.
8122 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8123 * been zapped so no dirty logging staff is needed for old slot. For
8124 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8125 * new and it's also covered when dealing with the new slot.
8127 * FIXME: const-ify all uses of struct kvm_memory_slot.
8129 if (change != KVM_MR_DELETE)
8130 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8133 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8135 kvm_mmu_invalidate_zap_all_pages(kvm);
8138 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8139 struct kvm_memory_slot *slot)
8141 kvm_mmu_invalidate_zap_all_pages(kvm);
8144 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8146 if (!list_empty_careful(&vcpu->async_pf.done))
8149 if (kvm_apic_has_events(vcpu))
8152 if (vcpu->arch.pv.pv_unhalted)
8155 if (atomic_read(&vcpu->arch.nmi_queued))
8158 if (test_bit(KVM_REQ_SMI, &vcpu->requests))
8161 if (kvm_arch_interrupt_allowed(vcpu) &&
8162 kvm_cpu_has_interrupt(vcpu))
8165 if (kvm_hv_has_stimer_pending(vcpu))
8171 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8173 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8174 kvm_x86_ops->check_nested_events(vcpu, false);
8176 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8179 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8181 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8184 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8186 return kvm_x86_ops->interrupt_allowed(vcpu);
8189 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8191 if (is_64_bit_mode(vcpu))
8192 return kvm_rip_read(vcpu);
8193 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8194 kvm_rip_read(vcpu));
8196 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8198 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8200 return kvm_get_linear_rip(vcpu) == linear_rip;
8202 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8204 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8206 unsigned long rflags;
8208 rflags = kvm_x86_ops->get_rflags(vcpu);
8209 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8210 rflags &= ~X86_EFLAGS_TF;
8213 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8215 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8217 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8218 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8219 rflags |= X86_EFLAGS_TF;
8220 kvm_x86_ops->set_rflags(vcpu, rflags);
8223 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8225 __kvm_set_rflags(vcpu, rflags);
8226 kvm_make_request(KVM_REQ_EVENT, vcpu);
8228 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8230 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8234 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8238 r = kvm_mmu_reload(vcpu);
8242 if (!vcpu->arch.mmu.direct_map &&
8243 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8246 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8249 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8251 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8254 static inline u32 kvm_async_pf_next_probe(u32 key)
8256 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8259 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8261 u32 key = kvm_async_pf_hash_fn(gfn);
8263 while (vcpu->arch.apf.gfns[key] != ~0)
8264 key = kvm_async_pf_next_probe(key);
8266 vcpu->arch.apf.gfns[key] = gfn;
8269 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8272 u32 key = kvm_async_pf_hash_fn(gfn);
8274 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8275 (vcpu->arch.apf.gfns[key] != gfn &&
8276 vcpu->arch.apf.gfns[key] != ~0); i++)
8277 key = kvm_async_pf_next_probe(key);
8282 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8284 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8287 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8291 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8293 vcpu->arch.apf.gfns[i] = ~0;
8295 j = kvm_async_pf_next_probe(j);
8296 if (vcpu->arch.apf.gfns[j] == ~0)
8298 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8300 * k lies cyclically in ]i,j]
8302 * |....j i.k.| or |.k..j i...|
8304 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8305 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8310 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8313 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8317 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8318 struct kvm_async_pf *work)
8320 struct x86_exception fault;
8322 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8323 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8325 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8326 (vcpu->arch.apf.send_user_only &&
8327 kvm_x86_ops->get_cpl(vcpu) == 0))
8328 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8329 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8330 fault.vector = PF_VECTOR;
8331 fault.error_code_valid = true;
8332 fault.error_code = 0;
8333 fault.nested_page_fault = false;
8334 fault.address = work->arch.token;
8335 kvm_inject_page_fault(vcpu, &fault);
8339 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8340 struct kvm_async_pf *work)
8342 struct x86_exception fault;
8344 trace_kvm_async_pf_ready(work->arch.token, work->gva);
8345 if (work->wakeup_all)
8346 work->arch.token = ~0; /* broadcast wakeup */
8348 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8350 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8351 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8352 fault.vector = PF_VECTOR;
8353 fault.error_code_valid = true;
8354 fault.error_code = 0;
8355 fault.nested_page_fault = false;
8356 fault.address = work->arch.token;
8357 kvm_inject_page_fault(vcpu, &fault);
8359 vcpu->arch.apf.halted = false;
8360 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8363 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8365 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8368 return !kvm_event_needs_reinjection(vcpu) &&
8369 kvm_x86_ops->interrupt_allowed(vcpu);
8372 void kvm_arch_start_assignment(struct kvm *kvm)
8374 atomic_inc(&kvm->arch.assigned_device_count);
8376 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8378 void kvm_arch_end_assignment(struct kvm *kvm)
8380 atomic_dec(&kvm->arch.assigned_device_count);
8382 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8384 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8386 return atomic_read(&kvm->arch.assigned_device_count);
8388 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8390 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8392 atomic_inc(&kvm->arch.noncoherent_dma_count);
8394 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8396 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8398 atomic_dec(&kvm->arch.noncoherent_dma_count);
8400 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8402 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8404 return atomic_read(&kvm->arch.noncoherent_dma_count);
8406 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8408 bool kvm_arch_has_irq_bypass(void)
8410 return kvm_x86_ops->update_pi_irte != NULL;
8413 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8414 struct irq_bypass_producer *prod)
8416 struct kvm_kernel_irqfd *irqfd =
8417 container_of(cons, struct kvm_kernel_irqfd, consumer);
8419 irqfd->producer = prod;
8421 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8422 prod->irq, irqfd->gsi, 1);
8425 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8426 struct irq_bypass_producer *prod)
8429 struct kvm_kernel_irqfd *irqfd =
8430 container_of(cons, struct kvm_kernel_irqfd, consumer);
8432 WARN_ON(irqfd->producer != prod);
8433 irqfd->producer = NULL;
8436 * When producer of consumer is unregistered, we change back to
8437 * remapped mode, so we can re-use the current implementation
8438 * when the irq is masked/disabled or the consumer side (KVM
8439 * int this case doesn't want to receive the interrupts.
8441 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8443 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8444 " fails: %d\n", irqfd->consumer.token, ret);
8447 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8448 uint32_t guest_irq, bool set)
8450 if (!kvm_x86_ops->update_pi_irte)
8453 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8456 bool kvm_vector_hashing_enabled(void)
8458 return vector_hashing;
8460 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
8462 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8463 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8464 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8465 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8466 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8467 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8468 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8469 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8470 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8471 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8472 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8473 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8474 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8475 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8476 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8477 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8478 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
8479 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
8480 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);