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"
33 #include <linux/clocksource.h>
34 #include <linux/interrupt.h>
35 #include <linux/kvm.h>
37 #include <linux/vmalloc.h>
38 #include <linux/export.h>
39 #include <linux/moduleparam.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 <linux/sched/stat.h>
58 #include <trace/events/kvm.h>
60 #include <asm/debugreg.h>
64 #include <linux/kernel_stat.h>
65 #include <asm/fpu/internal.h> /* Ugh! */
66 #include <asm/pvclock.h>
67 #include <asm/div64.h>
68 #include <asm/irq_remapping.h>
70 #define CREATE_TRACE_POINTS
73 #define MAX_IO_MSRS 256
74 #define KVM_MAX_MCE_BANKS 32
75 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
76 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
78 #define emul_to_vcpu(ctxt) \
79 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
82 * - enable syscall per default because its emulated by KVM
83 * - enable LME and LMA per default on 64 bit KVM
87 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
89 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
92 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
93 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
95 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
96 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
98 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
99 static void process_nmi(struct kvm_vcpu *vcpu);
100 static void enter_smm(struct kvm_vcpu *vcpu);
101 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
103 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
104 EXPORT_SYMBOL_GPL(kvm_x86_ops);
106 static bool __read_mostly ignore_msrs = 0;
107 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
109 unsigned int min_timer_period_us = 500;
110 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
112 static bool __read_mostly kvmclock_periodic_sync = true;
113 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
115 bool __read_mostly kvm_has_tsc_control;
116 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
117 u32 __read_mostly kvm_max_guest_tsc_khz;
118 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
119 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
120 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
121 u64 __read_mostly kvm_max_tsc_scaling_ratio;
122 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
123 u64 __read_mostly kvm_default_tsc_scaling_ratio;
124 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
126 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
127 static u32 __read_mostly tsc_tolerance_ppm = 250;
128 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
130 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
131 unsigned int __read_mostly lapic_timer_advance_ns = 0;
132 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
134 static bool __read_mostly vector_hashing = true;
135 module_param(vector_hashing, bool, S_IRUGO);
137 static bool __read_mostly backwards_tsc_observed = false;
139 #define KVM_NR_SHARED_MSRS 16
141 struct kvm_shared_msrs_global {
143 u32 msrs[KVM_NR_SHARED_MSRS];
146 struct kvm_shared_msrs {
147 struct user_return_notifier urn;
149 struct kvm_shared_msr_values {
152 } values[KVM_NR_SHARED_MSRS];
155 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
156 static struct kvm_shared_msrs __percpu *shared_msrs;
158 struct kvm_stats_debugfs_item debugfs_entries[] = {
159 { "pf_fixed", VCPU_STAT(pf_fixed) },
160 { "pf_guest", VCPU_STAT(pf_guest) },
161 { "tlb_flush", VCPU_STAT(tlb_flush) },
162 { "invlpg", VCPU_STAT(invlpg) },
163 { "exits", VCPU_STAT(exits) },
164 { "io_exits", VCPU_STAT(io_exits) },
165 { "mmio_exits", VCPU_STAT(mmio_exits) },
166 { "signal_exits", VCPU_STAT(signal_exits) },
167 { "irq_window", VCPU_STAT(irq_window_exits) },
168 { "nmi_window", VCPU_STAT(nmi_window_exits) },
169 { "halt_exits", VCPU_STAT(halt_exits) },
170 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
171 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
172 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
173 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
174 { "hypercalls", VCPU_STAT(hypercalls) },
175 { "request_irq", VCPU_STAT(request_irq_exits) },
176 { "irq_exits", VCPU_STAT(irq_exits) },
177 { "host_state_reload", VCPU_STAT(host_state_reload) },
178 { "efer_reload", VCPU_STAT(efer_reload) },
179 { "fpu_reload", VCPU_STAT(fpu_reload) },
180 { "insn_emulation", VCPU_STAT(insn_emulation) },
181 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
182 { "irq_injections", VCPU_STAT(irq_injections) },
183 { "nmi_injections", VCPU_STAT(nmi_injections) },
184 { "req_event", VCPU_STAT(req_event) },
185 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
186 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
187 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
188 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
189 { "mmu_flooded", VM_STAT(mmu_flooded) },
190 { "mmu_recycled", VM_STAT(mmu_recycled) },
191 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
192 { "mmu_unsync", VM_STAT(mmu_unsync) },
193 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
194 { "largepages", VM_STAT(lpages) },
195 { "max_mmu_page_hash_collisions",
196 VM_STAT(max_mmu_page_hash_collisions) },
200 u64 __read_mostly host_xcr0;
202 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
204 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
207 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
208 vcpu->arch.apf.gfns[i] = ~0;
211 static void kvm_on_user_return(struct user_return_notifier *urn)
214 struct kvm_shared_msrs *locals
215 = container_of(urn, struct kvm_shared_msrs, urn);
216 struct kvm_shared_msr_values *values;
220 * Disabling irqs at this point since the following code could be
221 * interrupted and executed through kvm_arch_hardware_disable()
223 local_irq_save(flags);
224 if (locals->registered) {
225 locals->registered = false;
226 user_return_notifier_unregister(urn);
228 local_irq_restore(flags);
229 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
230 values = &locals->values[slot];
231 if (values->host != values->curr) {
232 wrmsrl(shared_msrs_global.msrs[slot], values->host);
233 values->curr = values->host;
238 static void shared_msr_update(unsigned slot, u32 msr)
241 unsigned int cpu = smp_processor_id();
242 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
244 /* only read, and nobody should modify it at this time,
245 * so don't need lock */
246 if (slot >= shared_msrs_global.nr) {
247 printk(KERN_ERR "kvm: invalid MSR slot!");
250 rdmsrl_safe(msr, &value);
251 smsr->values[slot].host = value;
252 smsr->values[slot].curr = value;
255 void kvm_define_shared_msr(unsigned slot, u32 msr)
257 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
258 shared_msrs_global.msrs[slot] = msr;
259 if (slot >= shared_msrs_global.nr)
260 shared_msrs_global.nr = slot + 1;
262 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
264 static void kvm_shared_msr_cpu_online(void)
268 for (i = 0; i < shared_msrs_global.nr; ++i)
269 shared_msr_update(i, shared_msrs_global.msrs[i]);
272 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
274 unsigned int cpu = smp_processor_id();
275 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
278 if (((value ^ smsr->values[slot].curr) & mask) == 0)
280 smsr->values[slot].curr = value;
281 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
285 if (!smsr->registered) {
286 smsr->urn.on_user_return = kvm_on_user_return;
287 user_return_notifier_register(&smsr->urn);
288 smsr->registered = true;
292 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
294 static void drop_user_return_notifiers(void)
296 unsigned int cpu = smp_processor_id();
297 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
299 if (smsr->registered)
300 kvm_on_user_return(&smsr->urn);
303 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
305 return vcpu->arch.apic_base;
307 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
309 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
311 u64 old_state = vcpu->arch.apic_base &
312 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
313 u64 new_state = msr_info->data &
314 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
315 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) |
316 0x2ff | (guest_cpuid_has_x2apic(vcpu) ? 0 : X2APIC_ENABLE);
318 if (!msr_info->host_initiated &&
319 ((msr_info->data & reserved_bits) != 0 ||
320 new_state == X2APIC_ENABLE ||
321 (new_state == MSR_IA32_APICBASE_ENABLE &&
322 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
323 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
327 kvm_lapic_set_base(vcpu, msr_info->data);
330 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
332 asmlinkage __visible void kvm_spurious_fault(void)
334 /* Fault while not rebooting. We want the trace. */
337 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
339 #define EXCPT_BENIGN 0
340 #define EXCPT_CONTRIBUTORY 1
343 static int exception_class(int vector)
353 return EXCPT_CONTRIBUTORY;
360 #define EXCPT_FAULT 0
362 #define EXCPT_ABORT 2
363 #define EXCPT_INTERRUPT 3
365 static int exception_type(int vector)
369 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
370 return EXCPT_INTERRUPT;
374 /* #DB is trap, as instruction watchpoints are handled elsewhere */
375 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
378 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
381 /* Reserved exceptions will result in fault */
385 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
386 unsigned nr, bool has_error, u32 error_code,
392 kvm_make_request(KVM_REQ_EVENT, vcpu);
394 if (!vcpu->arch.exception.pending) {
396 if (has_error && !is_protmode(vcpu))
398 vcpu->arch.exception.pending = true;
399 vcpu->arch.exception.has_error_code = has_error;
400 vcpu->arch.exception.nr = nr;
401 vcpu->arch.exception.error_code = error_code;
402 vcpu->arch.exception.reinject = reinject;
406 /* to check exception */
407 prev_nr = vcpu->arch.exception.nr;
408 if (prev_nr == DF_VECTOR) {
409 /* triple fault -> shutdown */
410 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
413 class1 = exception_class(prev_nr);
414 class2 = exception_class(nr);
415 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
416 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
417 /* generate double fault per SDM Table 5-5 */
418 vcpu->arch.exception.pending = true;
419 vcpu->arch.exception.has_error_code = true;
420 vcpu->arch.exception.nr = DF_VECTOR;
421 vcpu->arch.exception.error_code = 0;
423 /* replace previous exception with a new one in a hope
424 that instruction re-execution will regenerate lost
429 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
431 kvm_multiple_exception(vcpu, nr, false, 0, false);
433 EXPORT_SYMBOL_GPL(kvm_queue_exception);
435 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
437 kvm_multiple_exception(vcpu, nr, false, 0, true);
439 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
441 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
444 kvm_inject_gp(vcpu, 0);
446 return kvm_skip_emulated_instruction(vcpu);
450 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
452 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
454 ++vcpu->stat.pf_guest;
455 vcpu->arch.cr2 = fault->address;
456 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
458 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
460 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
462 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
463 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
465 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
467 return fault->nested_page_fault;
470 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
472 atomic_inc(&vcpu->arch.nmi_queued);
473 kvm_make_request(KVM_REQ_NMI, vcpu);
475 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
477 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
479 kvm_multiple_exception(vcpu, nr, true, error_code, false);
481 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
483 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
485 kvm_multiple_exception(vcpu, nr, true, error_code, true);
487 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
490 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
491 * a #GP and return false.
493 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
495 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
497 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
500 EXPORT_SYMBOL_GPL(kvm_require_cpl);
502 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
504 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
507 kvm_queue_exception(vcpu, UD_VECTOR);
510 EXPORT_SYMBOL_GPL(kvm_require_dr);
513 * This function will be used to read from the physical memory of the currently
514 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
515 * can read from guest physical or from the guest's guest physical memory.
517 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
518 gfn_t ngfn, void *data, int offset, int len,
521 struct x86_exception exception;
525 ngpa = gfn_to_gpa(ngfn);
526 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
527 if (real_gfn == UNMAPPED_GVA)
530 real_gfn = gpa_to_gfn(real_gfn);
532 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
534 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
536 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
537 void *data, int offset, int len, u32 access)
539 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
540 data, offset, len, access);
544 * Load the pae pdptrs. Return true is they are all valid.
546 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
548 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
549 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
552 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
554 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
555 offset * sizeof(u64), sizeof(pdpte),
556 PFERR_USER_MASK|PFERR_WRITE_MASK);
561 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
562 if ((pdpte[i] & PT_PRESENT_MASK) &&
564 vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
571 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
572 __set_bit(VCPU_EXREG_PDPTR,
573 (unsigned long *)&vcpu->arch.regs_avail);
574 __set_bit(VCPU_EXREG_PDPTR,
575 (unsigned long *)&vcpu->arch.regs_dirty);
580 EXPORT_SYMBOL_GPL(load_pdptrs);
582 bool pdptrs_changed(struct kvm_vcpu *vcpu)
584 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
590 if (is_long_mode(vcpu) || !is_pae(vcpu))
593 if (!test_bit(VCPU_EXREG_PDPTR,
594 (unsigned long *)&vcpu->arch.regs_avail))
597 gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
598 offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
599 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
600 PFERR_USER_MASK | PFERR_WRITE_MASK);
603 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
608 EXPORT_SYMBOL_GPL(pdptrs_changed);
610 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
612 unsigned long old_cr0 = kvm_read_cr0(vcpu);
613 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
618 if (cr0 & 0xffffffff00000000UL)
622 cr0 &= ~CR0_RESERVED_BITS;
624 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
627 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
630 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
632 if ((vcpu->arch.efer & EFER_LME)) {
637 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
642 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
647 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
650 kvm_x86_ops->set_cr0(vcpu, cr0);
652 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
653 kvm_clear_async_pf_completion_queue(vcpu);
654 kvm_async_pf_hash_reset(vcpu);
657 if ((cr0 ^ old_cr0) & update_bits)
658 kvm_mmu_reset_context(vcpu);
660 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
661 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
662 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
663 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
667 EXPORT_SYMBOL_GPL(kvm_set_cr0);
669 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
671 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
673 EXPORT_SYMBOL_GPL(kvm_lmsw);
675 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
677 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
678 !vcpu->guest_xcr0_loaded) {
679 /* kvm_set_xcr() also depends on this */
680 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
681 vcpu->guest_xcr0_loaded = 1;
685 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
687 if (vcpu->guest_xcr0_loaded) {
688 if (vcpu->arch.xcr0 != host_xcr0)
689 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
690 vcpu->guest_xcr0_loaded = 0;
694 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
697 u64 old_xcr0 = vcpu->arch.xcr0;
700 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
701 if (index != XCR_XFEATURE_ENABLED_MASK)
703 if (!(xcr0 & XFEATURE_MASK_FP))
705 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
709 * Do not allow the guest to set bits that we do not support
710 * saving. However, xcr0 bit 0 is always set, even if the
711 * emulated CPU does not support XSAVE (see fx_init).
713 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
714 if (xcr0 & ~valid_bits)
717 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
718 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
721 if (xcr0 & XFEATURE_MASK_AVX512) {
722 if (!(xcr0 & XFEATURE_MASK_YMM))
724 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
727 vcpu->arch.xcr0 = xcr0;
729 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
730 kvm_update_cpuid(vcpu);
734 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
736 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
737 __kvm_set_xcr(vcpu, index, xcr)) {
738 kvm_inject_gp(vcpu, 0);
743 EXPORT_SYMBOL_GPL(kvm_set_xcr);
745 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
747 unsigned long old_cr4 = kvm_read_cr4(vcpu);
748 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
749 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
751 if (cr4 & CR4_RESERVED_BITS)
754 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
757 if (!guest_cpuid_has_smep(vcpu) && (cr4 & X86_CR4_SMEP))
760 if (!guest_cpuid_has_smap(vcpu) && (cr4 & X86_CR4_SMAP))
763 if (!guest_cpuid_has_fsgsbase(vcpu) && (cr4 & X86_CR4_FSGSBASE))
766 if (!guest_cpuid_has_pku(vcpu) && (cr4 & X86_CR4_PKE))
769 if (is_long_mode(vcpu)) {
770 if (!(cr4 & X86_CR4_PAE))
772 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
773 && ((cr4 ^ old_cr4) & pdptr_bits)
774 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
778 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
779 if (!guest_cpuid_has_pcid(vcpu))
782 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
783 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
787 if (kvm_x86_ops->set_cr4(vcpu, cr4))
790 if (((cr4 ^ old_cr4) & pdptr_bits) ||
791 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
792 kvm_mmu_reset_context(vcpu);
794 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
795 kvm_update_cpuid(vcpu);
799 EXPORT_SYMBOL_GPL(kvm_set_cr4);
801 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
804 cr3 &= ~CR3_PCID_INVD;
807 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
808 kvm_mmu_sync_roots(vcpu);
809 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
813 if (is_long_mode(vcpu)) {
814 if (cr3 & CR3_L_MODE_RESERVED_BITS)
816 } else if (is_pae(vcpu) && is_paging(vcpu) &&
817 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
820 vcpu->arch.cr3 = cr3;
821 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
822 kvm_mmu_new_cr3(vcpu);
825 EXPORT_SYMBOL_GPL(kvm_set_cr3);
827 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
829 if (cr8 & CR8_RESERVED_BITS)
831 if (lapic_in_kernel(vcpu))
832 kvm_lapic_set_tpr(vcpu, cr8);
834 vcpu->arch.cr8 = cr8;
837 EXPORT_SYMBOL_GPL(kvm_set_cr8);
839 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
841 if (lapic_in_kernel(vcpu))
842 return kvm_lapic_get_cr8(vcpu);
844 return vcpu->arch.cr8;
846 EXPORT_SYMBOL_GPL(kvm_get_cr8);
848 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
852 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
853 for (i = 0; i < KVM_NR_DB_REGS; i++)
854 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
855 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
859 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
861 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
862 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
865 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
869 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
870 dr7 = vcpu->arch.guest_debug_dr7;
872 dr7 = vcpu->arch.dr7;
873 kvm_x86_ops->set_dr7(vcpu, dr7);
874 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
875 if (dr7 & DR7_BP_EN_MASK)
876 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
879 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
881 u64 fixed = DR6_FIXED_1;
883 if (!guest_cpuid_has_rtm(vcpu))
888 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
892 vcpu->arch.db[dr] = val;
893 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
894 vcpu->arch.eff_db[dr] = val;
899 if (val & 0xffffffff00000000ULL)
901 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
902 kvm_update_dr6(vcpu);
907 if (val & 0xffffffff00000000ULL)
909 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
910 kvm_update_dr7(vcpu);
917 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
919 if (__kvm_set_dr(vcpu, dr, val)) {
920 kvm_inject_gp(vcpu, 0);
925 EXPORT_SYMBOL_GPL(kvm_set_dr);
927 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
931 *val = vcpu->arch.db[dr];
936 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
937 *val = vcpu->arch.dr6;
939 *val = kvm_x86_ops->get_dr6(vcpu);
944 *val = vcpu->arch.dr7;
949 EXPORT_SYMBOL_GPL(kvm_get_dr);
951 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
953 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
957 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
960 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
961 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
964 EXPORT_SYMBOL_GPL(kvm_rdpmc);
967 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
968 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
970 * This list is modified at module load time to reflect the
971 * capabilities of the host cpu. This capabilities test skips MSRs that are
972 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
973 * may depend on host virtualization features rather than host cpu features.
976 static u32 msrs_to_save[] = {
977 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
980 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
982 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
983 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
986 static unsigned num_msrs_to_save;
988 static u32 emulated_msrs[] = {
989 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
990 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
991 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
992 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
993 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
994 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
997 HV_X64_MSR_VP_RUNTIME,
999 HV_X64_MSR_STIMER0_CONFIG,
1000 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1003 MSR_IA32_TSC_ADJUST,
1004 MSR_IA32_TSCDEADLINE,
1005 MSR_IA32_MISC_ENABLE,
1006 MSR_IA32_MCG_STATUS,
1008 MSR_IA32_MCG_EXT_CTL,
1011 MSR_MISC_FEATURES_ENABLES,
1014 static unsigned num_emulated_msrs;
1016 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1018 if (efer & efer_reserved_bits)
1021 if (efer & EFER_FFXSR) {
1022 struct kvm_cpuid_entry2 *feat;
1024 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1025 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
1029 if (efer & EFER_SVME) {
1030 struct kvm_cpuid_entry2 *feat;
1032 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1033 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
1039 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1041 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1043 u64 old_efer = vcpu->arch.efer;
1045 if (!kvm_valid_efer(vcpu, efer))
1049 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1053 efer |= vcpu->arch.efer & EFER_LMA;
1055 kvm_x86_ops->set_efer(vcpu, efer);
1057 /* Update reserved bits */
1058 if ((efer ^ old_efer) & EFER_NX)
1059 kvm_mmu_reset_context(vcpu);
1064 void kvm_enable_efer_bits(u64 mask)
1066 efer_reserved_bits &= ~mask;
1068 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1071 * Writes msr value into into the appropriate "register".
1072 * Returns 0 on success, non-0 otherwise.
1073 * Assumes vcpu_load() was already called.
1075 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1077 switch (msr->index) {
1080 case MSR_KERNEL_GS_BASE:
1083 if (is_noncanonical_address(msr->data))
1086 case MSR_IA32_SYSENTER_EIP:
1087 case MSR_IA32_SYSENTER_ESP:
1089 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1090 * non-canonical address is written on Intel but not on
1091 * AMD (which ignores the top 32-bits, because it does
1092 * not implement 64-bit SYSENTER).
1094 * 64-bit code should hence be able to write a non-canonical
1095 * value on AMD. Making the address canonical ensures that
1096 * vmentry does not fail on Intel after writing a non-canonical
1097 * value, and that something deterministic happens if the guest
1098 * invokes 64-bit SYSENTER.
1100 msr->data = get_canonical(msr->data);
1102 return kvm_x86_ops->set_msr(vcpu, msr);
1104 EXPORT_SYMBOL_GPL(kvm_set_msr);
1107 * Adapt set_msr() to msr_io()'s calling convention
1109 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1111 struct msr_data msr;
1115 msr.host_initiated = true;
1116 r = kvm_get_msr(vcpu, &msr);
1124 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1126 struct msr_data msr;
1130 msr.host_initiated = true;
1131 return kvm_set_msr(vcpu, &msr);
1134 #ifdef CONFIG_X86_64
1135 struct pvclock_gtod_data {
1138 struct { /* extract of a clocksource struct */
1151 static struct pvclock_gtod_data pvclock_gtod_data;
1153 static void update_pvclock_gtod(struct timekeeper *tk)
1155 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1158 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1160 write_seqcount_begin(&vdata->seq);
1162 /* copy pvclock gtod data */
1163 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1164 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1165 vdata->clock.mask = tk->tkr_mono.mask;
1166 vdata->clock.mult = tk->tkr_mono.mult;
1167 vdata->clock.shift = tk->tkr_mono.shift;
1169 vdata->boot_ns = boot_ns;
1170 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1172 vdata->wall_time_sec = tk->xtime_sec;
1174 write_seqcount_end(&vdata->seq);
1178 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1181 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1182 * vcpu_enter_guest. This function is only called from
1183 * the physical CPU that is running vcpu.
1185 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1188 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1192 struct pvclock_wall_clock wc;
1193 struct timespec64 boot;
1198 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1203 ++version; /* first time write, random junk */
1207 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1211 * The guest calculates current wall clock time by adding
1212 * system time (updated by kvm_guest_time_update below) to the
1213 * wall clock specified here. guest system time equals host
1214 * system time for us, thus we must fill in host boot time here.
1216 getboottime64(&boot);
1218 if (kvm->arch.kvmclock_offset) {
1219 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1220 boot = timespec64_sub(boot, ts);
1222 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1223 wc.nsec = boot.tv_nsec;
1224 wc.version = version;
1226 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1229 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1232 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1234 do_shl32_div32(dividend, divisor);
1238 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1239 s8 *pshift, u32 *pmultiplier)
1247 scaled64 = scaled_hz;
1248 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1253 tps32 = (uint32_t)tps64;
1254 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1255 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1263 *pmultiplier = div_frac(scaled64, tps32);
1265 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1266 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1269 #ifdef CONFIG_X86_64
1270 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1273 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1274 static unsigned long max_tsc_khz;
1276 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1278 u64 v = (u64)khz * (1000000 + ppm);
1283 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1287 /* Guest TSC same frequency as host TSC? */
1289 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1293 /* TSC scaling supported? */
1294 if (!kvm_has_tsc_control) {
1295 if (user_tsc_khz > tsc_khz) {
1296 vcpu->arch.tsc_catchup = 1;
1297 vcpu->arch.tsc_always_catchup = 1;
1300 WARN(1, "user requested TSC rate below hardware speed\n");
1305 /* TSC scaling required - calculate ratio */
1306 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1307 user_tsc_khz, tsc_khz);
1309 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1310 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1315 vcpu->arch.tsc_scaling_ratio = ratio;
1319 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1321 u32 thresh_lo, thresh_hi;
1322 int use_scaling = 0;
1324 /* tsc_khz can be zero if TSC calibration fails */
1325 if (user_tsc_khz == 0) {
1326 /* set tsc_scaling_ratio to a safe value */
1327 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1331 /* Compute a scale to convert nanoseconds in TSC cycles */
1332 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1333 &vcpu->arch.virtual_tsc_shift,
1334 &vcpu->arch.virtual_tsc_mult);
1335 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1338 * Compute the variation in TSC rate which is acceptable
1339 * within the range of tolerance and decide if the
1340 * rate being applied is within that bounds of the hardware
1341 * rate. If so, no scaling or compensation need be done.
1343 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1344 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1345 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1346 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1349 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1352 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1354 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1355 vcpu->arch.virtual_tsc_mult,
1356 vcpu->arch.virtual_tsc_shift);
1357 tsc += vcpu->arch.this_tsc_write;
1361 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1363 #ifdef CONFIG_X86_64
1365 struct kvm_arch *ka = &vcpu->kvm->arch;
1366 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1368 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1369 atomic_read(&vcpu->kvm->online_vcpus));
1372 * Once the masterclock is enabled, always perform request in
1373 * order to update it.
1375 * In order to enable masterclock, the host clocksource must be TSC
1376 * and the vcpus need to have matched TSCs. When that happens,
1377 * perform request to enable masterclock.
1379 if (ka->use_master_clock ||
1380 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1381 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1383 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1384 atomic_read(&vcpu->kvm->online_vcpus),
1385 ka->use_master_clock, gtod->clock.vclock_mode);
1389 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1391 u64 curr_offset = vcpu->arch.tsc_offset;
1392 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1396 * Multiply tsc by a fixed point number represented by ratio.
1398 * The most significant 64-N bits (mult) of ratio represent the
1399 * integral part of the fixed point number; the remaining N bits
1400 * (frac) represent the fractional part, ie. ratio represents a fixed
1401 * point number (mult + frac * 2^(-N)).
1403 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1405 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1407 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1410 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1413 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1415 if (ratio != kvm_default_tsc_scaling_ratio)
1416 _tsc = __scale_tsc(ratio, tsc);
1420 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1422 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1426 tsc = kvm_scale_tsc(vcpu, rdtsc());
1428 return target_tsc - tsc;
1431 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1433 return vcpu->arch.tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1435 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1437 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1439 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1440 vcpu->arch.tsc_offset = offset;
1443 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1445 struct kvm *kvm = vcpu->kvm;
1446 u64 offset, ns, elapsed;
1447 unsigned long flags;
1449 bool already_matched;
1450 u64 data = msr->data;
1451 bool synchronizing = false;
1453 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1454 offset = kvm_compute_tsc_offset(vcpu, data);
1455 ns = ktime_get_boot_ns();
1456 elapsed = ns - kvm->arch.last_tsc_nsec;
1458 if (vcpu->arch.virtual_tsc_khz) {
1459 if (data == 0 && msr->host_initiated) {
1461 * detection of vcpu initialization -- need to sync
1462 * with other vCPUs. This particularly helps to keep
1463 * kvm_clock stable after CPU hotplug
1465 synchronizing = true;
1467 u64 tsc_exp = kvm->arch.last_tsc_write +
1468 nsec_to_cycles(vcpu, elapsed);
1469 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1471 * Special case: TSC write with a small delta (1 second)
1472 * of virtual cycle time against real time is
1473 * interpreted as an attempt to synchronize the CPU.
1475 synchronizing = data < tsc_exp + tsc_hz &&
1476 data + tsc_hz > tsc_exp;
1481 * For a reliable TSC, we can match TSC offsets, and for an unstable
1482 * TSC, we add elapsed time in this computation. We could let the
1483 * compensation code attempt to catch up if we fall behind, but
1484 * it's better to try to match offsets from the beginning.
1486 if (synchronizing &&
1487 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1488 if (!check_tsc_unstable()) {
1489 offset = kvm->arch.cur_tsc_offset;
1490 pr_debug("kvm: matched tsc offset for %llu\n", data);
1492 u64 delta = nsec_to_cycles(vcpu, elapsed);
1494 offset = kvm_compute_tsc_offset(vcpu, data);
1495 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1498 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1501 * We split periods of matched TSC writes into generations.
1502 * For each generation, we track the original measured
1503 * nanosecond time, offset, and write, so if TSCs are in
1504 * sync, we can match exact offset, and if not, we can match
1505 * exact software computation in compute_guest_tsc()
1507 * These values are tracked in kvm->arch.cur_xxx variables.
1509 kvm->arch.cur_tsc_generation++;
1510 kvm->arch.cur_tsc_nsec = ns;
1511 kvm->arch.cur_tsc_write = data;
1512 kvm->arch.cur_tsc_offset = offset;
1514 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1515 kvm->arch.cur_tsc_generation, data);
1519 * We also track th most recent recorded KHZ, write and time to
1520 * allow the matching interval to be extended at each write.
1522 kvm->arch.last_tsc_nsec = ns;
1523 kvm->arch.last_tsc_write = data;
1524 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1526 vcpu->arch.last_guest_tsc = data;
1528 /* Keep track of which generation this VCPU has synchronized to */
1529 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1530 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1531 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1533 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1534 update_ia32_tsc_adjust_msr(vcpu, offset);
1535 kvm_vcpu_write_tsc_offset(vcpu, offset);
1536 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1538 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1540 kvm->arch.nr_vcpus_matched_tsc = 0;
1541 } else if (!already_matched) {
1542 kvm->arch.nr_vcpus_matched_tsc++;
1545 kvm_track_tsc_matching(vcpu);
1546 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1549 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1551 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1554 kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.tsc_offset + adjustment);
1557 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1559 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1560 WARN_ON(adjustment < 0);
1561 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1562 adjust_tsc_offset_guest(vcpu, adjustment);
1565 #ifdef CONFIG_X86_64
1567 static u64 read_tsc(void)
1569 u64 ret = (u64)rdtsc_ordered();
1570 u64 last = pvclock_gtod_data.clock.cycle_last;
1572 if (likely(ret >= last))
1576 * GCC likes to generate cmov here, but this branch is extremely
1577 * predictable (it's just a function of time and the likely is
1578 * very likely) and there's a data dependence, so force GCC
1579 * to generate a branch instead. I don't barrier() because
1580 * we don't actually need a barrier, and if this function
1581 * ever gets inlined it will generate worse code.
1587 static inline u64 vgettsc(u64 *cycle_now)
1590 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1592 *cycle_now = read_tsc();
1594 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1595 return v * gtod->clock.mult;
1598 static int do_monotonic_boot(s64 *t, u64 *cycle_now)
1600 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1606 seq = read_seqcount_begin(>od->seq);
1607 mode = gtod->clock.vclock_mode;
1608 ns = gtod->nsec_base;
1609 ns += vgettsc(cycle_now);
1610 ns >>= gtod->clock.shift;
1611 ns += gtod->boot_ns;
1612 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1618 static int do_realtime(struct timespec *ts, u64 *cycle_now)
1620 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1626 seq = read_seqcount_begin(>od->seq);
1627 mode = gtod->clock.vclock_mode;
1628 ts->tv_sec = gtod->wall_time_sec;
1629 ns = gtod->nsec_base;
1630 ns += vgettsc(cycle_now);
1631 ns >>= gtod->clock.shift;
1632 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1634 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1640 /* returns true if host is using tsc clocksource */
1641 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *cycle_now)
1643 /* checked again under seqlock below */
1644 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1647 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1650 /* returns true if host is using tsc clocksource */
1651 static bool kvm_get_walltime_and_clockread(struct timespec *ts,
1654 /* checked again under seqlock below */
1655 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1658 return do_realtime(ts, cycle_now) == VCLOCK_TSC;
1664 * Assuming a stable TSC across physical CPUS, and a stable TSC
1665 * across virtual CPUs, the following condition is possible.
1666 * Each numbered line represents an event visible to both
1667 * CPUs at the next numbered event.
1669 * "timespecX" represents host monotonic time. "tscX" represents
1672 * VCPU0 on CPU0 | VCPU1 on CPU1
1674 * 1. read timespec0,tsc0
1675 * 2. | timespec1 = timespec0 + N
1677 * 3. transition to guest | transition to guest
1678 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1679 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1680 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1682 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1685 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1687 * - 0 < N - M => M < N
1689 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1690 * always the case (the difference between two distinct xtime instances
1691 * might be smaller then the difference between corresponding TSC reads,
1692 * when updating guest vcpus pvclock areas).
1694 * To avoid that problem, do not allow visibility of distinct
1695 * system_timestamp/tsc_timestamp values simultaneously: use a master
1696 * copy of host monotonic time values. Update that master copy
1699 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1703 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1705 #ifdef CONFIG_X86_64
1706 struct kvm_arch *ka = &kvm->arch;
1708 bool host_tsc_clocksource, vcpus_matched;
1710 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1711 atomic_read(&kvm->online_vcpus));
1714 * If the host uses TSC clock, then passthrough TSC as stable
1717 host_tsc_clocksource = kvm_get_time_and_clockread(
1718 &ka->master_kernel_ns,
1719 &ka->master_cycle_now);
1721 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1722 && !backwards_tsc_observed
1723 && !ka->boot_vcpu_runs_old_kvmclock;
1725 if (ka->use_master_clock)
1726 atomic_set(&kvm_guest_has_master_clock, 1);
1728 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1729 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1734 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1736 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1739 static void kvm_gen_update_masterclock(struct kvm *kvm)
1741 #ifdef CONFIG_X86_64
1743 struct kvm_vcpu *vcpu;
1744 struct kvm_arch *ka = &kvm->arch;
1746 spin_lock(&ka->pvclock_gtod_sync_lock);
1747 kvm_make_mclock_inprogress_request(kvm);
1748 /* no guest entries from this point */
1749 pvclock_update_vm_gtod_copy(kvm);
1751 kvm_for_each_vcpu(i, vcpu, kvm)
1752 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1754 /* guest entries allowed */
1755 kvm_for_each_vcpu(i, vcpu, kvm)
1756 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
1758 spin_unlock(&ka->pvclock_gtod_sync_lock);
1762 u64 get_kvmclock_ns(struct kvm *kvm)
1764 struct kvm_arch *ka = &kvm->arch;
1765 struct pvclock_vcpu_time_info hv_clock;
1767 spin_lock(&ka->pvclock_gtod_sync_lock);
1768 if (!ka->use_master_clock) {
1769 spin_unlock(&ka->pvclock_gtod_sync_lock);
1770 return ktime_get_boot_ns() + ka->kvmclock_offset;
1773 hv_clock.tsc_timestamp = ka->master_cycle_now;
1774 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
1775 spin_unlock(&ka->pvclock_gtod_sync_lock);
1777 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
1778 &hv_clock.tsc_shift,
1779 &hv_clock.tsc_to_system_mul);
1780 return __pvclock_read_cycles(&hv_clock, rdtsc());
1783 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
1785 struct kvm_vcpu_arch *vcpu = &v->arch;
1786 struct pvclock_vcpu_time_info guest_hv_clock;
1788 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1789 &guest_hv_clock, sizeof(guest_hv_clock))))
1792 /* This VCPU is paused, but it's legal for a guest to read another
1793 * VCPU's kvmclock, so we really have to follow the specification where
1794 * it says that version is odd if data is being modified, and even after
1797 * Version field updates must be kept separate. This is because
1798 * kvm_write_guest_cached might use a "rep movs" instruction, and
1799 * writes within a string instruction are weakly ordered. So there
1800 * are three writes overall.
1802 * As a small optimization, only write the version field in the first
1803 * and third write. The vcpu->pv_time cache is still valid, because the
1804 * version field is the first in the struct.
1806 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1808 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1809 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1811 sizeof(vcpu->hv_clock.version));
1815 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1816 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1818 if (vcpu->pvclock_set_guest_stopped_request) {
1819 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
1820 vcpu->pvclock_set_guest_stopped_request = false;
1823 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1825 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1827 sizeof(vcpu->hv_clock));
1831 vcpu->hv_clock.version++;
1832 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1834 sizeof(vcpu->hv_clock.version));
1837 static int kvm_guest_time_update(struct kvm_vcpu *v)
1839 unsigned long flags, tgt_tsc_khz;
1840 struct kvm_vcpu_arch *vcpu = &v->arch;
1841 struct kvm_arch *ka = &v->kvm->arch;
1843 u64 tsc_timestamp, host_tsc;
1845 bool use_master_clock;
1851 * If the host uses TSC clock, then passthrough TSC as stable
1854 spin_lock(&ka->pvclock_gtod_sync_lock);
1855 use_master_clock = ka->use_master_clock;
1856 if (use_master_clock) {
1857 host_tsc = ka->master_cycle_now;
1858 kernel_ns = ka->master_kernel_ns;
1860 spin_unlock(&ka->pvclock_gtod_sync_lock);
1862 /* Keep irq disabled to prevent changes to the clock */
1863 local_irq_save(flags);
1864 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1865 if (unlikely(tgt_tsc_khz == 0)) {
1866 local_irq_restore(flags);
1867 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1870 if (!use_master_clock) {
1872 kernel_ns = ktime_get_boot_ns();
1875 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1878 * We may have to catch up the TSC to match elapsed wall clock
1879 * time for two reasons, even if kvmclock is used.
1880 * 1) CPU could have been running below the maximum TSC rate
1881 * 2) Broken TSC compensation resets the base at each VCPU
1882 * entry to avoid unknown leaps of TSC even when running
1883 * again on the same CPU. This may cause apparent elapsed
1884 * time to disappear, and the guest to stand still or run
1887 if (vcpu->tsc_catchup) {
1888 u64 tsc = compute_guest_tsc(v, kernel_ns);
1889 if (tsc > tsc_timestamp) {
1890 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1891 tsc_timestamp = tsc;
1895 local_irq_restore(flags);
1897 /* With all the info we got, fill in the values */
1899 if (kvm_has_tsc_control)
1900 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
1902 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
1903 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
1904 &vcpu->hv_clock.tsc_shift,
1905 &vcpu->hv_clock.tsc_to_system_mul);
1906 vcpu->hw_tsc_khz = tgt_tsc_khz;
1909 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1910 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1911 vcpu->last_guest_tsc = tsc_timestamp;
1913 /* If the host uses TSC clocksource, then it is stable */
1915 if (use_master_clock)
1916 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1918 vcpu->hv_clock.flags = pvclock_flags;
1920 if (vcpu->pv_time_enabled)
1921 kvm_setup_pvclock_page(v);
1922 if (v == kvm_get_vcpu(v->kvm, 0))
1923 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
1928 * kvmclock updates which are isolated to a given vcpu, such as
1929 * vcpu->cpu migration, should not allow system_timestamp from
1930 * the rest of the vcpus to remain static. Otherwise ntp frequency
1931 * correction applies to one vcpu's system_timestamp but not
1934 * So in those cases, request a kvmclock update for all vcpus.
1935 * We need to rate-limit these requests though, as they can
1936 * considerably slow guests that have a large number of vcpus.
1937 * The time for a remote vcpu to update its kvmclock is bound
1938 * by the delay we use to rate-limit the updates.
1941 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1943 static void kvmclock_update_fn(struct work_struct *work)
1946 struct delayed_work *dwork = to_delayed_work(work);
1947 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1948 kvmclock_update_work);
1949 struct kvm *kvm = container_of(ka, struct kvm, arch);
1950 struct kvm_vcpu *vcpu;
1952 kvm_for_each_vcpu(i, vcpu, kvm) {
1953 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1954 kvm_vcpu_kick(vcpu);
1958 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1960 struct kvm *kvm = v->kvm;
1962 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1963 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1964 KVMCLOCK_UPDATE_DELAY);
1967 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1969 static void kvmclock_sync_fn(struct work_struct *work)
1971 struct delayed_work *dwork = to_delayed_work(work);
1972 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1973 kvmclock_sync_work);
1974 struct kvm *kvm = container_of(ka, struct kvm, arch);
1976 if (!kvmclock_periodic_sync)
1979 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
1980 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
1981 KVMCLOCK_SYNC_PERIOD);
1984 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1986 u64 mcg_cap = vcpu->arch.mcg_cap;
1987 unsigned bank_num = mcg_cap & 0xff;
1990 case MSR_IA32_MCG_STATUS:
1991 vcpu->arch.mcg_status = data;
1993 case MSR_IA32_MCG_CTL:
1994 if (!(mcg_cap & MCG_CTL_P))
1996 if (data != 0 && data != ~(u64)0)
1998 vcpu->arch.mcg_ctl = data;
2001 if (msr >= MSR_IA32_MC0_CTL &&
2002 msr < MSR_IA32_MCx_CTL(bank_num)) {
2003 u32 offset = msr - MSR_IA32_MC0_CTL;
2004 /* only 0 or all 1s can be written to IA32_MCi_CTL
2005 * some Linux kernels though clear bit 10 in bank 4 to
2006 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2007 * this to avoid an uncatched #GP in the guest
2009 if ((offset & 0x3) == 0 &&
2010 data != 0 && (data | (1 << 10)) != ~(u64)0)
2012 vcpu->arch.mce_banks[offset] = data;
2020 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2022 struct kvm *kvm = vcpu->kvm;
2023 int lm = is_long_mode(vcpu);
2024 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2025 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2026 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2027 : kvm->arch.xen_hvm_config.blob_size_32;
2028 u32 page_num = data & ~PAGE_MASK;
2029 u64 page_addr = data & PAGE_MASK;
2034 if (page_num >= blob_size)
2037 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2042 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2051 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2053 gpa_t gpa = data & ~0x3f;
2055 /* Bits 2:5 are reserved, Should be zero */
2059 vcpu->arch.apf.msr_val = data;
2061 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2062 kvm_clear_async_pf_completion_queue(vcpu);
2063 kvm_async_pf_hash_reset(vcpu);
2067 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2071 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2072 kvm_async_pf_wakeup_all(vcpu);
2076 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2078 vcpu->arch.pv_time_enabled = false;
2081 static void record_steal_time(struct kvm_vcpu *vcpu)
2083 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2086 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2087 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2090 vcpu->arch.st.steal.preempted = 0;
2092 if (vcpu->arch.st.steal.version & 1)
2093 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2095 vcpu->arch.st.steal.version += 1;
2097 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2098 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2102 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2103 vcpu->arch.st.last_steal;
2104 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2106 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2107 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2111 vcpu->arch.st.steal.version += 1;
2113 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2114 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2117 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2120 u32 msr = msr_info->index;
2121 u64 data = msr_info->data;
2124 case MSR_AMD64_NB_CFG:
2125 case MSR_IA32_UCODE_REV:
2126 case MSR_IA32_UCODE_WRITE:
2127 case MSR_VM_HSAVE_PA:
2128 case MSR_AMD64_PATCH_LOADER:
2129 case MSR_AMD64_BU_CFG2:
2130 case MSR_AMD64_DC_CFG:
2134 return set_efer(vcpu, data);
2136 data &= ~(u64)0x40; /* ignore flush filter disable */
2137 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2138 data &= ~(u64)0x8; /* ignore TLB cache disable */
2139 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2141 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2146 case MSR_FAM10H_MMIO_CONF_BASE:
2148 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2153 case MSR_IA32_DEBUGCTLMSR:
2155 /* We support the non-activated case already */
2157 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2158 /* Values other than LBR and BTF are vendor-specific,
2159 thus reserved and should throw a #GP */
2162 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2165 case 0x200 ... 0x2ff:
2166 return kvm_mtrr_set_msr(vcpu, msr, data);
2167 case MSR_IA32_APICBASE:
2168 return kvm_set_apic_base(vcpu, msr_info);
2169 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2170 return kvm_x2apic_msr_write(vcpu, msr, data);
2171 case MSR_IA32_TSCDEADLINE:
2172 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2174 case MSR_IA32_TSC_ADJUST:
2175 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2176 if (!msr_info->host_initiated) {
2177 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2178 adjust_tsc_offset_guest(vcpu, adj);
2180 vcpu->arch.ia32_tsc_adjust_msr = data;
2183 case MSR_IA32_MISC_ENABLE:
2184 vcpu->arch.ia32_misc_enable_msr = data;
2186 case MSR_IA32_SMBASE:
2187 if (!msr_info->host_initiated)
2189 vcpu->arch.smbase = data;
2191 case MSR_KVM_WALL_CLOCK_NEW:
2192 case MSR_KVM_WALL_CLOCK:
2193 vcpu->kvm->arch.wall_clock = data;
2194 kvm_write_wall_clock(vcpu->kvm, data);
2196 case MSR_KVM_SYSTEM_TIME_NEW:
2197 case MSR_KVM_SYSTEM_TIME: {
2198 struct kvm_arch *ka = &vcpu->kvm->arch;
2200 kvmclock_reset(vcpu);
2202 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2203 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2205 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2206 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2208 ka->boot_vcpu_runs_old_kvmclock = tmp;
2211 vcpu->arch.time = data;
2212 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2214 /* we verify if the enable bit is set... */
2218 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2219 &vcpu->arch.pv_time, data & ~1ULL,
2220 sizeof(struct pvclock_vcpu_time_info)))
2221 vcpu->arch.pv_time_enabled = false;
2223 vcpu->arch.pv_time_enabled = true;
2227 case MSR_KVM_ASYNC_PF_EN:
2228 if (kvm_pv_enable_async_pf(vcpu, data))
2231 case MSR_KVM_STEAL_TIME:
2233 if (unlikely(!sched_info_on()))
2236 if (data & KVM_STEAL_RESERVED_MASK)
2239 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2240 data & KVM_STEAL_VALID_BITS,
2241 sizeof(struct kvm_steal_time)))
2244 vcpu->arch.st.msr_val = data;
2246 if (!(data & KVM_MSR_ENABLED))
2249 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2252 case MSR_KVM_PV_EOI_EN:
2253 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2257 case MSR_IA32_MCG_CTL:
2258 case MSR_IA32_MCG_STATUS:
2259 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2260 return set_msr_mce(vcpu, msr, data);
2262 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2263 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2264 pr = true; /* fall through */
2265 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2266 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2267 if (kvm_pmu_is_valid_msr(vcpu, msr))
2268 return kvm_pmu_set_msr(vcpu, msr_info);
2270 if (pr || data != 0)
2271 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2272 "0x%x data 0x%llx\n", msr, data);
2274 case MSR_K7_CLK_CTL:
2276 * Ignore all writes to this no longer documented MSR.
2277 * Writes are only relevant for old K7 processors,
2278 * all pre-dating SVM, but a recommended workaround from
2279 * AMD for these chips. It is possible to specify the
2280 * affected processor models on the command line, hence
2281 * the need to ignore the workaround.
2284 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2285 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2286 case HV_X64_MSR_CRASH_CTL:
2287 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2288 return kvm_hv_set_msr_common(vcpu, msr, data,
2289 msr_info->host_initiated);
2290 case MSR_IA32_BBL_CR_CTL3:
2291 /* Drop writes to this legacy MSR -- see rdmsr
2292 * counterpart for further detail.
2294 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n", msr, data);
2296 case MSR_AMD64_OSVW_ID_LENGTH:
2297 if (!guest_cpuid_has_osvw(vcpu))
2299 vcpu->arch.osvw.length = data;
2301 case MSR_AMD64_OSVW_STATUS:
2302 if (!guest_cpuid_has_osvw(vcpu))
2304 vcpu->arch.osvw.status = data;
2306 case MSR_PLATFORM_INFO:
2307 if (!msr_info->host_initiated ||
2308 data & ~MSR_PLATFORM_INFO_CPUID_FAULT ||
2309 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2310 cpuid_fault_enabled(vcpu)))
2312 vcpu->arch.msr_platform_info = data;
2314 case MSR_MISC_FEATURES_ENABLES:
2315 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2316 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2317 !supports_cpuid_fault(vcpu)))
2319 vcpu->arch.msr_misc_features_enables = data;
2322 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2323 return xen_hvm_config(vcpu, data);
2324 if (kvm_pmu_is_valid_msr(vcpu, msr))
2325 return kvm_pmu_set_msr(vcpu, msr_info);
2327 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2331 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2338 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2342 * Reads an msr value (of 'msr_index') into 'pdata'.
2343 * Returns 0 on success, non-0 otherwise.
2344 * Assumes vcpu_load() was already called.
2346 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2348 return kvm_x86_ops->get_msr(vcpu, msr);
2350 EXPORT_SYMBOL_GPL(kvm_get_msr);
2352 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2355 u64 mcg_cap = vcpu->arch.mcg_cap;
2356 unsigned bank_num = mcg_cap & 0xff;
2359 case MSR_IA32_P5_MC_ADDR:
2360 case MSR_IA32_P5_MC_TYPE:
2363 case MSR_IA32_MCG_CAP:
2364 data = vcpu->arch.mcg_cap;
2366 case MSR_IA32_MCG_CTL:
2367 if (!(mcg_cap & MCG_CTL_P))
2369 data = vcpu->arch.mcg_ctl;
2371 case MSR_IA32_MCG_STATUS:
2372 data = vcpu->arch.mcg_status;
2375 if (msr >= MSR_IA32_MC0_CTL &&
2376 msr < MSR_IA32_MCx_CTL(bank_num)) {
2377 u32 offset = msr - MSR_IA32_MC0_CTL;
2378 data = vcpu->arch.mce_banks[offset];
2387 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2389 switch (msr_info->index) {
2390 case MSR_IA32_PLATFORM_ID:
2391 case MSR_IA32_EBL_CR_POWERON:
2392 case MSR_IA32_DEBUGCTLMSR:
2393 case MSR_IA32_LASTBRANCHFROMIP:
2394 case MSR_IA32_LASTBRANCHTOIP:
2395 case MSR_IA32_LASTINTFROMIP:
2396 case MSR_IA32_LASTINTTOIP:
2398 case MSR_K8_TSEG_ADDR:
2399 case MSR_K8_TSEG_MASK:
2401 case MSR_VM_HSAVE_PA:
2402 case MSR_K8_INT_PENDING_MSG:
2403 case MSR_AMD64_NB_CFG:
2404 case MSR_FAM10H_MMIO_CONF_BASE:
2405 case MSR_AMD64_BU_CFG2:
2406 case MSR_IA32_PERF_CTL:
2407 case MSR_AMD64_DC_CFG:
2410 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2411 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2412 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2413 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2414 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2415 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2418 case MSR_IA32_UCODE_REV:
2419 msr_info->data = 0x100000000ULL;
2422 case 0x200 ... 0x2ff:
2423 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2424 case 0xcd: /* fsb frequency */
2428 * MSR_EBC_FREQUENCY_ID
2429 * Conservative value valid for even the basic CPU models.
2430 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2431 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2432 * and 266MHz for model 3, or 4. Set Core Clock
2433 * Frequency to System Bus Frequency Ratio to 1 (bits
2434 * 31:24) even though these are only valid for CPU
2435 * models > 2, however guests may end up dividing or
2436 * multiplying by zero otherwise.
2438 case MSR_EBC_FREQUENCY_ID:
2439 msr_info->data = 1 << 24;
2441 case MSR_IA32_APICBASE:
2442 msr_info->data = kvm_get_apic_base(vcpu);
2444 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2445 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2447 case MSR_IA32_TSCDEADLINE:
2448 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2450 case MSR_IA32_TSC_ADJUST:
2451 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2453 case MSR_IA32_MISC_ENABLE:
2454 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2456 case MSR_IA32_SMBASE:
2457 if (!msr_info->host_initiated)
2459 msr_info->data = vcpu->arch.smbase;
2461 case MSR_IA32_PERF_STATUS:
2462 /* TSC increment by tick */
2463 msr_info->data = 1000ULL;
2464 /* CPU multiplier */
2465 msr_info->data |= (((uint64_t)4ULL) << 40);
2468 msr_info->data = vcpu->arch.efer;
2470 case MSR_KVM_WALL_CLOCK:
2471 case MSR_KVM_WALL_CLOCK_NEW:
2472 msr_info->data = vcpu->kvm->arch.wall_clock;
2474 case MSR_KVM_SYSTEM_TIME:
2475 case MSR_KVM_SYSTEM_TIME_NEW:
2476 msr_info->data = vcpu->arch.time;
2478 case MSR_KVM_ASYNC_PF_EN:
2479 msr_info->data = vcpu->arch.apf.msr_val;
2481 case MSR_KVM_STEAL_TIME:
2482 msr_info->data = vcpu->arch.st.msr_val;
2484 case MSR_KVM_PV_EOI_EN:
2485 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2487 case MSR_IA32_P5_MC_ADDR:
2488 case MSR_IA32_P5_MC_TYPE:
2489 case MSR_IA32_MCG_CAP:
2490 case MSR_IA32_MCG_CTL:
2491 case MSR_IA32_MCG_STATUS:
2492 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2493 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2494 case MSR_K7_CLK_CTL:
2496 * Provide expected ramp-up count for K7. All other
2497 * are set to zero, indicating minimum divisors for
2500 * This prevents guest kernels on AMD host with CPU
2501 * type 6, model 8 and higher from exploding due to
2502 * the rdmsr failing.
2504 msr_info->data = 0x20000000;
2506 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2507 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2508 case HV_X64_MSR_CRASH_CTL:
2509 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2510 return kvm_hv_get_msr_common(vcpu,
2511 msr_info->index, &msr_info->data);
2513 case MSR_IA32_BBL_CR_CTL3:
2514 /* This legacy MSR exists but isn't fully documented in current
2515 * silicon. It is however accessed by winxp in very narrow
2516 * scenarios where it sets bit #19, itself documented as
2517 * a "reserved" bit. Best effort attempt to source coherent
2518 * read data here should the balance of the register be
2519 * interpreted by the guest:
2521 * L2 cache control register 3: 64GB range, 256KB size,
2522 * enabled, latency 0x1, configured
2524 msr_info->data = 0xbe702111;
2526 case MSR_AMD64_OSVW_ID_LENGTH:
2527 if (!guest_cpuid_has_osvw(vcpu))
2529 msr_info->data = vcpu->arch.osvw.length;
2531 case MSR_AMD64_OSVW_STATUS:
2532 if (!guest_cpuid_has_osvw(vcpu))
2534 msr_info->data = vcpu->arch.osvw.status;
2536 case MSR_PLATFORM_INFO:
2537 msr_info->data = vcpu->arch.msr_platform_info;
2539 case MSR_MISC_FEATURES_ENABLES:
2540 msr_info->data = vcpu->arch.msr_misc_features_enables;
2543 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2544 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2546 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2550 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2557 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2560 * Read or write a bunch of msrs. All parameters are kernel addresses.
2562 * @return number of msrs set successfully.
2564 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2565 struct kvm_msr_entry *entries,
2566 int (*do_msr)(struct kvm_vcpu *vcpu,
2567 unsigned index, u64 *data))
2571 idx = srcu_read_lock(&vcpu->kvm->srcu);
2572 for (i = 0; i < msrs->nmsrs; ++i)
2573 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2575 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2581 * Read or write a bunch of msrs. Parameters are user addresses.
2583 * @return number of msrs set successfully.
2585 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2586 int (*do_msr)(struct kvm_vcpu *vcpu,
2587 unsigned index, u64 *data),
2590 struct kvm_msrs msrs;
2591 struct kvm_msr_entry *entries;
2596 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2600 if (msrs.nmsrs >= MAX_IO_MSRS)
2603 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2604 entries = memdup_user(user_msrs->entries, size);
2605 if (IS_ERR(entries)) {
2606 r = PTR_ERR(entries);
2610 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2615 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2626 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2631 case KVM_CAP_IRQCHIP:
2633 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2634 case KVM_CAP_SET_TSS_ADDR:
2635 case KVM_CAP_EXT_CPUID:
2636 case KVM_CAP_EXT_EMUL_CPUID:
2637 case KVM_CAP_CLOCKSOURCE:
2639 case KVM_CAP_NOP_IO_DELAY:
2640 case KVM_CAP_MP_STATE:
2641 case KVM_CAP_SYNC_MMU:
2642 case KVM_CAP_USER_NMI:
2643 case KVM_CAP_REINJECT_CONTROL:
2644 case KVM_CAP_IRQ_INJECT_STATUS:
2645 case KVM_CAP_IOEVENTFD:
2646 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2648 case KVM_CAP_PIT_STATE2:
2649 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2650 case KVM_CAP_XEN_HVM:
2651 case KVM_CAP_VCPU_EVENTS:
2652 case KVM_CAP_HYPERV:
2653 case KVM_CAP_HYPERV_VAPIC:
2654 case KVM_CAP_HYPERV_SPIN:
2655 case KVM_CAP_HYPERV_SYNIC:
2656 case KVM_CAP_PCI_SEGMENT:
2657 case KVM_CAP_DEBUGREGS:
2658 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2660 case KVM_CAP_ASYNC_PF:
2661 case KVM_CAP_GET_TSC_KHZ:
2662 case KVM_CAP_KVMCLOCK_CTRL:
2663 case KVM_CAP_READONLY_MEM:
2664 case KVM_CAP_HYPERV_TIME:
2665 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2666 case KVM_CAP_TSC_DEADLINE_TIMER:
2667 case KVM_CAP_ENABLE_CAP_VM:
2668 case KVM_CAP_DISABLE_QUIRKS:
2669 case KVM_CAP_SET_BOOT_CPU_ID:
2670 case KVM_CAP_SPLIT_IRQCHIP:
2671 case KVM_CAP_IMMEDIATE_EXIT:
2674 case KVM_CAP_ADJUST_CLOCK:
2675 r = KVM_CLOCK_TSC_STABLE;
2677 case KVM_CAP_X86_GUEST_MWAIT:
2678 r = kvm_mwait_in_guest();
2680 case KVM_CAP_X86_SMM:
2681 /* SMBASE is usually relocated above 1M on modern chipsets,
2682 * and SMM handlers might indeed rely on 4G segment limits,
2683 * so do not report SMM to be available if real mode is
2684 * emulated via vm86 mode. Still, do not go to great lengths
2685 * to avoid userspace's usage of the feature, because it is a
2686 * fringe case that is not enabled except via specific settings
2687 * of the module parameters.
2689 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2692 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2694 case KVM_CAP_NR_VCPUS:
2695 r = KVM_SOFT_MAX_VCPUS;
2697 case KVM_CAP_MAX_VCPUS:
2700 case KVM_CAP_NR_MEMSLOTS:
2701 r = KVM_USER_MEM_SLOTS;
2703 case KVM_CAP_PV_MMU: /* obsolete */
2707 r = KVM_MAX_MCE_BANKS;
2710 r = boot_cpu_has(X86_FEATURE_XSAVE);
2712 case KVM_CAP_TSC_CONTROL:
2713 r = kvm_has_tsc_control;
2715 case KVM_CAP_X2APIC_API:
2716 r = KVM_X2APIC_API_VALID_FLAGS;
2726 long kvm_arch_dev_ioctl(struct file *filp,
2727 unsigned int ioctl, unsigned long arg)
2729 void __user *argp = (void __user *)arg;
2733 case KVM_GET_MSR_INDEX_LIST: {
2734 struct kvm_msr_list __user *user_msr_list = argp;
2735 struct kvm_msr_list msr_list;
2739 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2742 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2743 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2746 if (n < msr_list.nmsrs)
2749 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2750 num_msrs_to_save * sizeof(u32)))
2752 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2754 num_emulated_msrs * sizeof(u32)))
2759 case KVM_GET_SUPPORTED_CPUID:
2760 case KVM_GET_EMULATED_CPUID: {
2761 struct kvm_cpuid2 __user *cpuid_arg = argp;
2762 struct kvm_cpuid2 cpuid;
2765 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2768 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2774 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2779 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2781 if (copy_to_user(argp, &kvm_mce_cap_supported,
2782 sizeof(kvm_mce_cap_supported)))
2794 static void wbinvd_ipi(void *garbage)
2799 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2801 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2804 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2806 /* Address WBINVD may be executed by guest */
2807 if (need_emulate_wbinvd(vcpu)) {
2808 if (kvm_x86_ops->has_wbinvd_exit())
2809 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2810 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2811 smp_call_function_single(vcpu->cpu,
2812 wbinvd_ipi, NULL, 1);
2815 kvm_x86_ops->vcpu_load(vcpu, cpu);
2817 /* Apply any externally detected TSC adjustments (due to suspend) */
2818 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2819 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2820 vcpu->arch.tsc_offset_adjustment = 0;
2821 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2824 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2825 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2826 rdtsc() - vcpu->arch.last_host_tsc;
2828 mark_tsc_unstable("KVM discovered backwards TSC");
2830 if (check_tsc_unstable()) {
2831 u64 offset = kvm_compute_tsc_offset(vcpu,
2832 vcpu->arch.last_guest_tsc);
2833 kvm_vcpu_write_tsc_offset(vcpu, offset);
2834 vcpu->arch.tsc_catchup = 1;
2836 if (kvm_lapic_hv_timer_in_use(vcpu) &&
2837 kvm_x86_ops->set_hv_timer(vcpu,
2838 kvm_get_lapic_target_expiration_tsc(vcpu)))
2839 kvm_lapic_switch_to_sw_timer(vcpu);
2841 * On a host with synchronized TSC, there is no need to update
2842 * kvmclock on vcpu->cpu migration
2844 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2845 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2846 if (vcpu->cpu != cpu)
2847 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
2851 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2854 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
2856 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2859 vcpu->arch.st.steal.preempted = 1;
2861 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
2862 &vcpu->arch.st.steal.preempted,
2863 offsetof(struct kvm_steal_time, preempted),
2864 sizeof(vcpu->arch.st.steal.preempted));
2867 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2871 * Disable page faults because we're in atomic context here.
2872 * kvm_write_guest_offset_cached() would call might_fault()
2873 * that relies on pagefault_disable() to tell if there's a
2874 * bug. NOTE: the write to guest memory may not go through if
2875 * during postcopy live migration or if there's heavy guest
2878 pagefault_disable();
2880 * kvm_memslots() will be called by
2881 * kvm_write_guest_offset_cached() so take the srcu lock.
2883 idx = srcu_read_lock(&vcpu->kvm->srcu);
2884 kvm_steal_time_set_preempted(vcpu);
2885 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2887 kvm_x86_ops->vcpu_put(vcpu);
2888 kvm_put_guest_fpu(vcpu);
2889 vcpu->arch.last_host_tsc = rdtsc();
2892 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2893 struct kvm_lapic_state *s)
2895 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
2896 kvm_x86_ops->sync_pir_to_irr(vcpu);
2898 return kvm_apic_get_state(vcpu, s);
2901 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2902 struct kvm_lapic_state *s)
2906 r = kvm_apic_set_state(vcpu, s);
2909 update_cr8_intercept(vcpu);
2914 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2916 return (!lapic_in_kernel(vcpu) ||
2917 kvm_apic_accept_pic_intr(vcpu));
2921 * if userspace requested an interrupt window, check that the
2922 * interrupt window is open.
2924 * No need to exit to userspace if we already have an interrupt queued.
2926 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2928 return kvm_arch_interrupt_allowed(vcpu) &&
2929 !kvm_cpu_has_interrupt(vcpu) &&
2930 !kvm_event_needs_reinjection(vcpu) &&
2931 kvm_cpu_accept_dm_intr(vcpu);
2934 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2935 struct kvm_interrupt *irq)
2937 if (irq->irq >= KVM_NR_INTERRUPTS)
2940 if (!irqchip_in_kernel(vcpu->kvm)) {
2941 kvm_queue_interrupt(vcpu, irq->irq, false);
2942 kvm_make_request(KVM_REQ_EVENT, vcpu);
2947 * With in-kernel LAPIC, we only use this to inject EXTINT, so
2948 * fail for in-kernel 8259.
2950 if (pic_in_kernel(vcpu->kvm))
2953 if (vcpu->arch.pending_external_vector != -1)
2956 vcpu->arch.pending_external_vector = irq->irq;
2957 kvm_make_request(KVM_REQ_EVENT, vcpu);
2961 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2963 kvm_inject_nmi(vcpu);
2968 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2970 kvm_make_request(KVM_REQ_SMI, vcpu);
2975 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2976 struct kvm_tpr_access_ctl *tac)
2980 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2984 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2988 unsigned bank_num = mcg_cap & 0xff, bank;
2991 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2993 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
2996 vcpu->arch.mcg_cap = mcg_cap;
2997 /* Init IA32_MCG_CTL to all 1s */
2998 if (mcg_cap & MCG_CTL_P)
2999 vcpu->arch.mcg_ctl = ~(u64)0;
3000 /* Init IA32_MCi_CTL to all 1s */
3001 for (bank = 0; bank < bank_num; bank++)
3002 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3004 if (kvm_x86_ops->setup_mce)
3005 kvm_x86_ops->setup_mce(vcpu);
3010 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3011 struct kvm_x86_mce *mce)
3013 u64 mcg_cap = vcpu->arch.mcg_cap;
3014 unsigned bank_num = mcg_cap & 0xff;
3015 u64 *banks = vcpu->arch.mce_banks;
3017 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3020 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3021 * reporting is disabled
3023 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3024 vcpu->arch.mcg_ctl != ~(u64)0)
3026 banks += 4 * mce->bank;
3028 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3029 * reporting is disabled for the bank
3031 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3033 if (mce->status & MCI_STATUS_UC) {
3034 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3035 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3036 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3039 if (banks[1] & MCI_STATUS_VAL)
3040 mce->status |= MCI_STATUS_OVER;
3041 banks[2] = mce->addr;
3042 banks[3] = mce->misc;
3043 vcpu->arch.mcg_status = mce->mcg_status;
3044 banks[1] = mce->status;
3045 kvm_queue_exception(vcpu, MC_VECTOR);
3046 } else if (!(banks[1] & MCI_STATUS_VAL)
3047 || !(banks[1] & MCI_STATUS_UC)) {
3048 if (banks[1] & MCI_STATUS_VAL)
3049 mce->status |= MCI_STATUS_OVER;
3050 banks[2] = mce->addr;
3051 banks[3] = mce->misc;
3052 banks[1] = mce->status;
3054 banks[1] |= MCI_STATUS_OVER;
3058 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3059 struct kvm_vcpu_events *events)
3062 events->exception.injected =
3063 vcpu->arch.exception.pending &&
3064 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3065 events->exception.nr = vcpu->arch.exception.nr;
3066 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3067 events->exception.pad = 0;
3068 events->exception.error_code = vcpu->arch.exception.error_code;
3070 events->interrupt.injected =
3071 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3072 events->interrupt.nr = vcpu->arch.interrupt.nr;
3073 events->interrupt.soft = 0;
3074 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3076 events->nmi.injected = vcpu->arch.nmi_injected;
3077 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3078 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3079 events->nmi.pad = 0;
3081 events->sipi_vector = 0; /* never valid when reporting to user space */
3083 events->smi.smm = is_smm(vcpu);
3084 events->smi.pending = vcpu->arch.smi_pending;
3085 events->smi.smm_inside_nmi =
3086 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3087 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3089 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3090 | KVM_VCPUEVENT_VALID_SHADOW
3091 | KVM_VCPUEVENT_VALID_SMM);
3092 memset(&events->reserved, 0, sizeof(events->reserved));
3095 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3097 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3098 struct kvm_vcpu_events *events)
3100 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3101 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3102 | KVM_VCPUEVENT_VALID_SHADOW
3103 | KVM_VCPUEVENT_VALID_SMM))
3106 if (events->exception.injected &&
3107 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR ||
3108 is_guest_mode(vcpu)))
3111 /* INITs are latched while in SMM */
3112 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3113 (events->smi.smm || events->smi.pending) &&
3114 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3118 vcpu->arch.exception.pending = events->exception.injected;
3119 vcpu->arch.exception.nr = events->exception.nr;
3120 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3121 vcpu->arch.exception.error_code = events->exception.error_code;
3123 vcpu->arch.interrupt.pending = events->interrupt.injected;
3124 vcpu->arch.interrupt.nr = events->interrupt.nr;
3125 vcpu->arch.interrupt.soft = events->interrupt.soft;
3126 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3127 kvm_x86_ops->set_interrupt_shadow(vcpu,
3128 events->interrupt.shadow);
3130 vcpu->arch.nmi_injected = events->nmi.injected;
3131 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3132 vcpu->arch.nmi_pending = events->nmi.pending;
3133 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3135 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3136 lapic_in_kernel(vcpu))
3137 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3139 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3140 u32 hflags = vcpu->arch.hflags;
3141 if (events->smi.smm)
3142 hflags |= HF_SMM_MASK;
3144 hflags &= ~HF_SMM_MASK;
3145 kvm_set_hflags(vcpu, hflags);
3147 vcpu->arch.smi_pending = events->smi.pending;
3148 if (events->smi.smm_inside_nmi)
3149 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3151 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3152 if (lapic_in_kernel(vcpu)) {
3153 if (events->smi.latched_init)
3154 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3156 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3160 kvm_make_request(KVM_REQ_EVENT, vcpu);
3165 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3166 struct kvm_debugregs *dbgregs)
3170 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3171 kvm_get_dr(vcpu, 6, &val);
3173 dbgregs->dr7 = vcpu->arch.dr7;
3175 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3178 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3179 struct kvm_debugregs *dbgregs)
3184 if (dbgregs->dr6 & ~0xffffffffull)
3186 if (dbgregs->dr7 & ~0xffffffffull)
3189 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3190 kvm_update_dr0123(vcpu);
3191 vcpu->arch.dr6 = dbgregs->dr6;
3192 kvm_update_dr6(vcpu);
3193 vcpu->arch.dr7 = dbgregs->dr7;
3194 kvm_update_dr7(vcpu);
3199 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3201 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3203 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3204 u64 xstate_bv = xsave->header.xfeatures;
3208 * Copy legacy XSAVE area, to avoid complications with CPUID
3209 * leaves 0 and 1 in the loop below.
3211 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3214 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3215 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3218 * Copy each region from the possibly compacted offset to the
3219 * non-compacted offset.
3221 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3223 u64 feature = valid & -valid;
3224 int index = fls64(feature) - 1;
3225 void *src = get_xsave_addr(xsave, feature);
3228 u32 size, offset, ecx, edx;
3229 cpuid_count(XSTATE_CPUID, index,
3230 &size, &offset, &ecx, &edx);
3231 memcpy(dest + offset, src, size);
3238 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3240 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3241 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3245 * Copy legacy XSAVE area, to avoid complications with CPUID
3246 * leaves 0 and 1 in the loop below.
3248 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3250 /* Set XSTATE_BV and possibly XCOMP_BV. */
3251 xsave->header.xfeatures = xstate_bv;
3252 if (boot_cpu_has(X86_FEATURE_XSAVES))
3253 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3256 * Copy each region from the non-compacted offset to the
3257 * possibly compacted offset.
3259 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3261 u64 feature = valid & -valid;
3262 int index = fls64(feature) - 1;
3263 void *dest = get_xsave_addr(xsave, feature);
3266 u32 size, offset, ecx, edx;
3267 cpuid_count(XSTATE_CPUID, index,
3268 &size, &offset, &ecx, &edx);
3269 memcpy(dest, src + offset, size);
3276 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3277 struct kvm_xsave *guest_xsave)
3279 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3280 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3281 fill_xsave((u8 *) guest_xsave->region, vcpu);
3283 memcpy(guest_xsave->region,
3284 &vcpu->arch.guest_fpu.state.fxsave,
3285 sizeof(struct fxregs_state));
3286 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3287 XFEATURE_MASK_FPSSE;
3291 #define XSAVE_MXCSR_OFFSET 24
3293 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3294 struct kvm_xsave *guest_xsave)
3297 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3298 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3300 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3302 * Here we allow setting states that are not present in
3303 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3304 * with old userspace.
3306 if (xstate_bv & ~kvm_supported_xcr0() ||
3307 mxcsr & ~mxcsr_feature_mask)
3309 load_xsave(vcpu, (u8 *)guest_xsave->region);
3311 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3312 mxcsr & ~mxcsr_feature_mask)
3314 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3315 guest_xsave->region, sizeof(struct fxregs_state));
3320 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3321 struct kvm_xcrs *guest_xcrs)
3323 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3324 guest_xcrs->nr_xcrs = 0;
3328 guest_xcrs->nr_xcrs = 1;
3329 guest_xcrs->flags = 0;
3330 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3331 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3334 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3335 struct kvm_xcrs *guest_xcrs)
3339 if (!boot_cpu_has(X86_FEATURE_XSAVE))
3342 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3345 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3346 /* Only support XCR0 currently */
3347 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3348 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3349 guest_xcrs->xcrs[i].value);
3358 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3359 * stopped by the hypervisor. This function will be called from the host only.
3360 * EINVAL is returned when the host attempts to set the flag for a guest that
3361 * does not support pv clocks.
3363 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3365 if (!vcpu->arch.pv_time_enabled)
3367 vcpu->arch.pvclock_set_guest_stopped_request = true;
3368 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3372 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3373 struct kvm_enable_cap *cap)
3379 case KVM_CAP_HYPERV_SYNIC:
3380 if (!irqchip_in_kernel(vcpu->kvm))
3382 return kvm_hv_activate_synic(vcpu);
3388 long kvm_arch_vcpu_ioctl(struct file *filp,
3389 unsigned int ioctl, unsigned long arg)
3391 struct kvm_vcpu *vcpu = filp->private_data;
3392 void __user *argp = (void __user *)arg;
3395 struct kvm_lapic_state *lapic;
3396 struct kvm_xsave *xsave;
3397 struct kvm_xcrs *xcrs;
3403 case KVM_GET_LAPIC: {
3405 if (!lapic_in_kernel(vcpu))
3407 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3412 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3416 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3421 case KVM_SET_LAPIC: {
3423 if (!lapic_in_kernel(vcpu))
3425 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3426 if (IS_ERR(u.lapic))
3427 return PTR_ERR(u.lapic);
3429 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3432 case KVM_INTERRUPT: {
3433 struct kvm_interrupt irq;
3436 if (copy_from_user(&irq, argp, sizeof irq))
3438 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3442 r = kvm_vcpu_ioctl_nmi(vcpu);
3446 r = kvm_vcpu_ioctl_smi(vcpu);
3449 case KVM_SET_CPUID: {
3450 struct kvm_cpuid __user *cpuid_arg = argp;
3451 struct kvm_cpuid cpuid;
3454 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3456 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3459 case KVM_SET_CPUID2: {
3460 struct kvm_cpuid2 __user *cpuid_arg = argp;
3461 struct kvm_cpuid2 cpuid;
3464 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3466 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3467 cpuid_arg->entries);
3470 case KVM_GET_CPUID2: {
3471 struct kvm_cpuid2 __user *cpuid_arg = argp;
3472 struct kvm_cpuid2 cpuid;
3475 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3477 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3478 cpuid_arg->entries);
3482 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3488 r = msr_io(vcpu, argp, do_get_msr, 1);
3491 r = msr_io(vcpu, argp, do_set_msr, 0);
3493 case KVM_TPR_ACCESS_REPORTING: {
3494 struct kvm_tpr_access_ctl tac;
3497 if (copy_from_user(&tac, argp, sizeof tac))
3499 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3503 if (copy_to_user(argp, &tac, sizeof tac))
3508 case KVM_SET_VAPIC_ADDR: {
3509 struct kvm_vapic_addr va;
3513 if (!lapic_in_kernel(vcpu))
3516 if (copy_from_user(&va, argp, sizeof va))
3518 idx = srcu_read_lock(&vcpu->kvm->srcu);
3519 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3520 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3523 case KVM_X86_SETUP_MCE: {
3527 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3529 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3532 case KVM_X86_SET_MCE: {
3533 struct kvm_x86_mce mce;
3536 if (copy_from_user(&mce, argp, sizeof mce))
3538 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3541 case KVM_GET_VCPU_EVENTS: {
3542 struct kvm_vcpu_events events;
3544 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3547 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3552 case KVM_SET_VCPU_EVENTS: {
3553 struct kvm_vcpu_events events;
3556 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3559 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3562 case KVM_GET_DEBUGREGS: {
3563 struct kvm_debugregs dbgregs;
3565 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3568 if (copy_to_user(argp, &dbgregs,
3569 sizeof(struct kvm_debugregs)))
3574 case KVM_SET_DEBUGREGS: {
3575 struct kvm_debugregs dbgregs;
3578 if (copy_from_user(&dbgregs, argp,
3579 sizeof(struct kvm_debugregs)))
3582 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3585 case KVM_GET_XSAVE: {
3586 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3591 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3594 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3599 case KVM_SET_XSAVE: {
3600 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3601 if (IS_ERR(u.xsave))
3602 return PTR_ERR(u.xsave);
3604 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3607 case KVM_GET_XCRS: {
3608 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3613 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3616 if (copy_to_user(argp, u.xcrs,
3617 sizeof(struct kvm_xcrs)))
3622 case KVM_SET_XCRS: {
3623 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3625 return PTR_ERR(u.xcrs);
3627 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3630 case KVM_SET_TSC_KHZ: {
3634 user_tsc_khz = (u32)arg;
3636 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3639 if (user_tsc_khz == 0)
3640 user_tsc_khz = tsc_khz;
3642 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3647 case KVM_GET_TSC_KHZ: {
3648 r = vcpu->arch.virtual_tsc_khz;
3651 case KVM_KVMCLOCK_CTRL: {
3652 r = kvm_set_guest_paused(vcpu);
3655 case KVM_ENABLE_CAP: {
3656 struct kvm_enable_cap cap;
3659 if (copy_from_user(&cap, argp, sizeof(cap)))
3661 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3672 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3674 return VM_FAULT_SIGBUS;
3677 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3681 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3683 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3687 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3690 kvm->arch.ept_identity_map_addr = ident_addr;
3694 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3695 u32 kvm_nr_mmu_pages)
3697 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3700 mutex_lock(&kvm->slots_lock);
3702 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3703 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3705 mutex_unlock(&kvm->slots_lock);
3709 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3711 return kvm->arch.n_max_mmu_pages;
3714 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3716 struct kvm_pic *pic = kvm->arch.vpic;
3720 switch (chip->chip_id) {
3721 case KVM_IRQCHIP_PIC_MASTER:
3722 memcpy(&chip->chip.pic, &pic->pics[0],
3723 sizeof(struct kvm_pic_state));
3725 case KVM_IRQCHIP_PIC_SLAVE:
3726 memcpy(&chip->chip.pic, &pic->pics[1],
3727 sizeof(struct kvm_pic_state));
3729 case KVM_IRQCHIP_IOAPIC:
3730 kvm_get_ioapic(kvm, &chip->chip.ioapic);
3739 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3741 struct kvm_pic *pic = kvm->arch.vpic;
3745 switch (chip->chip_id) {
3746 case KVM_IRQCHIP_PIC_MASTER:
3747 spin_lock(&pic->lock);
3748 memcpy(&pic->pics[0], &chip->chip.pic,
3749 sizeof(struct kvm_pic_state));
3750 spin_unlock(&pic->lock);
3752 case KVM_IRQCHIP_PIC_SLAVE:
3753 spin_lock(&pic->lock);
3754 memcpy(&pic->pics[1], &chip->chip.pic,
3755 sizeof(struct kvm_pic_state));
3756 spin_unlock(&pic->lock);
3758 case KVM_IRQCHIP_IOAPIC:
3759 kvm_set_ioapic(kvm, &chip->chip.ioapic);
3765 kvm_pic_update_irq(pic);
3769 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3771 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3773 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3775 mutex_lock(&kps->lock);
3776 memcpy(ps, &kps->channels, sizeof(*ps));
3777 mutex_unlock(&kps->lock);
3781 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3784 struct kvm_pit *pit = kvm->arch.vpit;
3786 mutex_lock(&pit->pit_state.lock);
3787 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
3788 for (i = 0; i < 3; i++)
3789 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3790 mutex_unlock(&pit->pit_state.lock);
3794 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3796 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3797 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3798 sizeof(ps->channels));
3799 ps->flags = kvm->arch.vpit->pit_state.flags;
3800 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3801 memset(&ps->reserved, 0, sizeof(ps->reserved));
3805 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3809 u32 prev_legacy, cur_legacy;
3810 struct kvm_pit *pit = kvm->arch.vpit;
3812 mutex_lock(&pit->pit_state.lock);
3813 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3814 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3815 if (!prev_legacy && cur_legacy)
3817 memcpy(&pit->pit_state.channels, &ps->channels,
3818 sizeof(pit->pit_state.channels));
3819 pit->pit_state.flags = ps->flags;
3820 for (i = 0; i < 3; i++)
3821 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
3823 mutex_unlock(&pit->pit_state.lock);
3827 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3828 struct kvm_reinject_control *control)
3830 struct kvm_pit *pit = kvm->arch.vpit;
3835 /* pit->pit_state.lock was overloaded to prevent userspace from getting
3836 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3837 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
3839 mutex_lock(&pit->pit_state.lock);
3840 kvm_pit_set_reinject(pit, control->pit_reinject);
3841 mutex_unlock(&pit->pit_state.lock);
3847 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3848 * @kvm: kvm instance
3849 * @log: slot id and address to which we copy the log
3851 * Steps 1-4 below provide general overview of dirty page logging. See
3852 * kvm_get_dirty_log_protect() function description for additional details.
3854 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3855 * always flush the TLB (step 4) even if previous step failed and the dirty
3856 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3857 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3858 * writes will be marked dirty for next log read.
3860 * 1. Take a snapshot of the bit and clear it if needed.
3861 * 2. Write protect the corresponding page.
3862 * 3. Copy the snapshot to the userspace.
3863 * 4. Flush TLB's if needed.
3865 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3867 bool is_dirty = false;
3870 mutex_lock(&kvm->slots_lock);
3873 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3875 if (kvm_x86_ops->flush_log_dirty)
3876 kvm_x86_ops->flush_log_dirty(kvm);
3878 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3881 * All the TLBs can be flushed out of mmu lock, see the comments in
3882 * kvm_mmu_slot_remove_write_access().
3884 lockdep_assert_held(&kvm->slots_lock);
3886 kvm_flush_remote_tlbs(kvm);
3888 mutex_unlock(&kvm->slots_lock);
3892 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3895 if (!irqchip_in_kernel(kvm))
3898 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3899 irq_event->irq, irq_event->level,
3904 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3905 struct kvm_enable_cap *cap)
3913 case KVM_CAP_DISABLE_QUIRKS:
3914 kvm->arch.disabled_quirks = cap->args[0];
3917 case KVM_CAP_SPLIT_IRQCHIP: {
3918 mutex_lock(&kvm->lock);
3920 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3921 goto split_irqchip_unlock;
3923 if (irqchip_in_kernel(kvm))
3924 goto split_irqchip_unlock;
3925 if (kvm->created_vcpus)
3926 goto split_irqchip_unlock;
3927 r = kvm_setup_empty_irq_routing(kvm);
3929 goto split_irqchip_unlock;
3930 /* Pairs with irqchip_in_kernel. */
3932 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
3933 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
3935 split_irqchip_unlock:
3936 mutex_unlock(&kvm->lock);
3939 case KVM_CAP_X2APIC_API:
3941 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
3944 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
3945 kvm->arch.x2apic_format = true;
3946 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
3947 kvm->arch.x2apic_broadcast_quirk_disabled = true;
3958 long kvm_arch_vm_ioctl(struct file *filp,
3959 unsigned int ioctl, unsigned long arg)
3961 struct kvm *kvm = filp->private_data;
3962 void __user *argp = (void __user *)arg;
3965 * This union makes it completely explicit to gcc-3.x
3966 * that these two variables' stack usage should be
3967 * combined, not added together.
3970 struct kvm_pit_state ps;
3971 struct kvm_pit_state2 ps2;
3972 struct kvm_pit_config pit_config;
3976 case KVM_SET_TSS_ADDR:
3977 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3979 case KVM_SET_IDENTITY_MAP_ADDR: {
3983 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3985 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3988 case KVM_SET_NR_MMU_PAGES:
3989 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3991 case KVM_GET_NR_MMU_PAGES:
3992 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3994 case KVM_CREATE_IRQCHIP: {
3995 mutex_lock(&kvm->lock);
3998 if (irqchip_in_kernel(kvm))
3999 goto create_irqchip_unlock;
4002 if (kvm->created_vcpus)
4003 goto create_irqchip_unlock;
4005 r = kvm_pic_init(kvm);
4007 goto create_irqchip_unlock;
4009 r = kvm_ioapic_init(kvm);
4011 kvm_pic_destroy(kvm);
4012 goto create_irqchip_unlock;
4015 r = kvm_setup_default_irq_routing(kvm);
4017 kvm_ioapic_destroy(kvm);
4018 kvm_pic_destroy(kvm);
4019 goto create_irqchip_unlock;
4021 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4023 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4024 create_irqchip_unlock:
4025 mutex_unlock(&kvm->lock);
4028 case KVM_CREATE_PIT:
4029 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4031 case KVM_CREATE_PIT2:
4033 if (copy_from_user(&u.pit_config, argp,
4034 sizeof(struct kvm_pit_config)))
4037 mutex_lock(&kvm->lock);
4040 goto create_pit_unlock;
4042 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4046 mutex_unlock(&kvm->lock);
4048 case KVM_GET_IRQCHIP: {
4049 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4050 struct kvm_irqchip *chip;
4052 chip = memdup_user(argp, sizeof(*chip));
4059 if (!irqchip_kernel(kvm))
4060 goto get_irqchip_out;
4061 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4063 goto get_irqchip_out;
4065 if (copy_to_user(argp, chip, sizeof *chip))
4066 goto get_irqchip_out;
4072 case KVM_SET_IRQCHIP: {
4073 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4074 struct kvm_irqchip *chip;
4076 chip = memdup_user(argp, sizeof(*chip));
4083 if (!irqchip_kernel(kvm))
4084 goto set_irqchip_out;
4085 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4087 goto set_irqchip_out;
4095 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4098 if (!kvm->arch.vpit)
4100 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4104 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4111 if (copy_from_user(&u.ps, argp, sizeof u.ps))
4114 if (!kvm->arch.vpit)
4116 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4119 case KVM_GET_PIT2: {
4121 if (!kvm->arch.vpit)
4123 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4127 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4132 case KVM_SET_PIT2: {
4134 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4137 if (!kvm->arch.vpit)
4139 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4142 case KVM_REINJECT_CONTROL: {
4143 struct kvm_reinject_control control;
4145 if (copy_from_user(&control, argp, sizeof(control)))
4147 r = kvm_vm_ioctl_reinject(kvm, &control);
4150 case KVM_SET_BOOT_CPU_ID:
4152 mutex_lock(&kvm->lock);
4153 if (kvm->created_vcpus)
4156 kvm->arch.bsp_vcpu_id = arg;
4157 mutex_unlock(&kvm->lock);
4159 case KVM_XEN_HVM_CONFIG: {
4161 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4162 sizeof(struct kvm_xen_hvm_config)))
4165 if (kvm->arch.xen_hvm_config.flags)
4170 case KVM_SET_CLOCK: {
4171 struct kvm_clock_data user_ns;
4175 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4183 now_ns = get_kvmclock_ns(kvm);
4184 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4185 kvm_gen_update_masterclock(kvm);
4188 case KVM_GET_CLOCK: {
4189 struct kvm_clock_data user_ns;
4192 now_ns = get_kvmclock_ns(kvm);
4193 user_ns.clock = now_ns;
4194 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4195 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4198 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4203 case KVM_ENABLE_CAP: {
4204 struct kvm_enable_cap cap;
4207 if (copy_from_user(&cap, argp, sizeof(cap)))
4209 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4219 static void kvm_init_msr_list(void)
4224 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4225 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4229 * Even MSRs that are valid in the host may not be exposed
4230 * to the guests in some cases.
4232 switch (msrs_to_save[i]) {
4233 case MSR_IA32_BNDCFGS:
4234 if (!kvm_x86_ops->mpx_supported())
4238 if (!kvm_x86_ops->rdtscp_supported())
4246 msrs_to_save[j] = msrs_to_save[i];
4249 num_msrs_to_save = j;
4251 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4252 switch (emulated_msrs[i]) {
4253 case MSR_IA32_SMBASE:
4254 if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4262 emulated_msrs[j] = emulated_msrs[i];
4265 num_emulated_msrs = j;
4268 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4276 if (!(lapic_in_kernel(vcpu) &&
4277 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4278 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4289 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4296 if (!(lapic_in_kernel(vcpu) &&
4297 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4299 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4301 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4311 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4312 struct kvm_segment *var, int seg)
4314 kvm_x86_ops->set_segment(vcpu, var, seg);
4317 void kvm_get_segment(struct kvm_vcpu *vcpu,
4318 struct kvm_segment *var, int seg)
4320 kvm_x86_ops->get_segment(vcpu, var, seg);
4323 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4324 struct x86_exception *exception)
4328 BUG_ON(!mmu_is_nested(vcpu));
4330 /* NPT walks are always user-walks */
4331 access |= PFERR_USER_MASK;
4332 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4337 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4338 struct x86_exception *exception)
4340 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4341 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4344 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4345 struct x86_exception *exception)
4347 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4348 access |= PFERR_FETCH_MASK;
4349 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4352 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4353 struct x86_exception *exception)
4355 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4356 access |= PFERR_WRITE_MASK;
4357 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4360 /* uses this to access any guest's mapped memory without checking CPL */
4361 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4362 struct x86_exception *exception)
4364 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4367 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4368 struct kvm_vcpu *vcpu, u32 access,
4369 struct x86_exception *exception)
4372 int r = X86EMUL_CONTINUE;
4375 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4377 unsigned offset = addr & (PAGE_SIZE-1);
4378 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4381 if (gpa == UNMAPPED_GVA)
4382 return X86EMUL_PROPAGATE_FAULT;
4383 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4386 r = X86EMUL_IO_NEEDED;
4398 /* used for instruction fetching */
4399 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4400 gva_t addr, void *val, unsigned int bytes,
4401 struct x86_exception *exception)
4403 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4404 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4408 /* Inline kvm_read_guest_virt_helper for speed. */
4409 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4411 if (unlikely(gpa == UNMAPPED_GVA))
4412 return X86EMUL_PROPAGATE_FAULT;
4414 offset = addr & (PAGE_SIZE-1);
4415 if (WARN_ON(offset + bytes > PAGE_SIZE))
4416 bytes = (unsigned)PAGE_SIZE - offset;
4417 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4419 if (unlikely(ret < 0))
4420 return X86EMUL_IO_NEEDED;
4422 return X86EMUL_CONTINUE;
4425 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4426 gva_t addr, void *val, unsigned int bytes,
4427 struct x86_exception *exception)
4429 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4430 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4432 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4435 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4437 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4438 gva_t addr, void *val, unsigned int bytes,
4439 struct x86_exception *exception)
4441 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4442 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4445 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4446 unsigned long addr, void *val, unsigned int bytes)
4448 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4449 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4451 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4454 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4455 gva_t addr, void *val,
4457 struct x86_exception *exception)
4459 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4461 int r = X86EMUL_CONTINUE;
4464 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4467 unsigned offset = addr & (PAGE_SIZE-1);
4468 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4471 if (gpa == UNMAPPED_GVA)
4472 return X86EMUL_PROPAGATE_FAULT;
4473 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4475 r = X86EMUL_IO_NEEDED;
4486 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4488 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4489 gpa_t gpa, bool write)
4491 /* For APIC access vmexit */
4492 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4495 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
4496 trace_vcpu_match_mmio(gva, gpa, write, true);
4503 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4504 gpa_t *gpa, struct x86_exception *exception,
4507 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4508 | (write ? PFERR_WRITE_MASK : 0);
4511 * currently PKRU is only applied to ept enabled guest so
4512 * there is no pkey in EPT page table for L1 guest or EPT
4513 * shadow page table for L2 guest.
4515 if (vcpu_match_mmio_gva(vcpu, gva)
4516 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4517 vcpu->arch.access, 0, access)) {
4518 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4519 (gva & (PAGE_SIZE - 1));
4520 trace_vcpu_match_mmio(gva, *gpa, write, false);
4524 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4526 if (*gpa == UNMAPPED_GVA)
4529 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
4532 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4533 const void *val, int bytes)
4537 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4540 kvm_page_track_write(vcpu, gpa, val, bytes);
4544 struct read_write_emulator_ops {
4545 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4547 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4548 void *val, int bytes);
4549 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4550 int bytes, void *val);
4551 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4552 void *val, int bytes);
4556 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4558 if (vcpu->mmio_read_completed) {
4559 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4560 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4561 vcpu->mmio_read_completed = 0;
4568 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4569 void *val, int bytes)
4571 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4574 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4575 void *val, int bytes)
4577 return emulator_write_phys(vcpu, gpa, val, bytes);
4580 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4582 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4583 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4586 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4587 void *val, int bytes)
4589 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4590 return X86EMUL_IO_NEEDED;
4593 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4594 void *val, int bytes)
4596 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4598 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4599 return X86EMUL_CONTINUE;
4602 static const struct read_write_emulator_ops read_emultor = {
4603 .read_write_prepare = read_prepare,
4604 .read_write_emulate = read_emulate,
4605 .read_write_mmio = vcpu_mmio_read,
4606 .read_write_exit_mmio = read_exit_mmio,
4609 static const struct read_write_emulator_ops write_emultor = {
4610 .read_write_emulate = write_emulate,
4611 .read_write_mmio = write_mmio,
4612 .read_write_exit_mmio = write_exit_mmio,
4616 static int emulator_read_write_onepage(unsigned long addr, void *val,
4618 struct x86_exception *exception,
4619 struct kvm_vcpu *vcpu,
4620 const struct read_write_emulator_ops *ops)
4624 bool write = ops->write;
4625 struct kvm_mmio_fragment *frag;
4626 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4629 * If the exit was due to a NPF we may already have a GPA.
4630 * If the GPA is present, use it to avoid the GVA to GPA table walk.
4631 * Note, this cannot be used on string operations since string
4632 * operation using rep will only have the initial GPA from the NPF
4635 if (vcpu->arch.gpa_available &&
4636 emulator_can_use_gpa(ctxt) &&
4637 vcpu_is_mmio_gpa(vcpu, addr, exception->address, write) &&
4638 (addr & ~PAGE_MASK) == (exception->address & ~PAGE_MASK)) {
4639 gpa = exception->address;
4643 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4646 return X86EMUL_PROPAGATE_FAULT;
4648 /* For APIC access vmexit */
4652 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4653 return X86EMUL_CONTINUE;
4657 * Is this MMIO handled locally?
4659 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4660 if (handled == bytes)
4661 return X86EMUL_CONTINUE;
4667 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4668 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4672 return X86EMUL_CONTINUE;
4675 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4677 void *val, unsigned int bytes,
4678 struct x86_exception *exception,
4679 const struct read_write_emulator_ops *ops)
4681 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4685 if (ops->read_write_prepare &&
4686 ops->read_write_prepare(vcpu, val, bytes))
4687 return X86EMUL_CONTINUE;
4689 vcpu->mmio_nr_fragments = 0;
4691 /* Crossing a page boundary? */
4692 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4695 now = -addr & ~PAGE_MASK;
4696 rc = emulator_read_write_onepage(addr, val, now, exception,
4699 if (rc != X86EMUL_CONTINUE)
4702 if (ctxt->mode != X86EMUL_MODE_PROT64)
4708 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4710 if (rc != X86EMUL_CONTINUE)
4713 if (!vcpu->mmio_nr_fragments)
4716 gpa = vcpu->mmio_fragments[0].gpa;
4718 vcpu->mmio_needed = 1;
4719 vcpu->mmio_cur_fragment = 0;
4721 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4722 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4723 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4724 vcpu->run->mmio.phys_addr = gpa;
4726 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4729 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4733 struct x86_exception *exception)
4735 return emulator_read_write(ctxt, addr, val, bytes,
4736 exception, &read_emultor);
4739 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4743 struct x86_exception *exception)
4745 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4746 exception, &write_emultor);
4749 #define CMPXCHG_TYPE(t, ptr, old, new) \
4750 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4752 #ifdef CONFIG_X86_64
4753 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4755 # define CMPXCHG64(ptr, old, new) \
4756 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4759 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4764 struct x86_exception *exception)
4766 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4772 /* guests cmpxchg8b have to be emulated atomically */
4773 if (bytes > 8 || (bytes & (bytes - 1)))
4776 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4778 if (gpa == UNMAPPED_GVA ||
4779 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4782 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4785 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4786 if (is_error_page(page))
4789 kaddr = kmap_atomic(page);
4790 kaddr += offset_in_page(gpa);
4793 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4796 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4799 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4802 exchanged = CMPXCHG64(kaddr, old, new);
4807 kunmap_atomic(kaddr);
4808 kvm_release_page_dirty(page);
4811 return X86EMUL_CMPXCHG_FAILED;
4813 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4814 kvm_page_track_write(vcpu, gpa, new, bytes);
4816 return X86EMUL_CONTINUE;
4819 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4821 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4824 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4826 /* TODO: String I/O for in kernel device */
4829 if (vcpu->arch.pio.in)
4830 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4831 vcpu->arch.pio.size, pd);
4833 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4834 vcpu->arch.pio.port, vcpu->arch.pio.size,
4839 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4840 unsigned short port, void *val,
4841 unsigned int count, bool in)
4843 vcpu->arch.pio.port = port;
4844 vcpu->arch.pio.in = in;
4845 vcpu->arch.pio.count = count;
4846 vcpu->arch.pio.size = size;
4848 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4849 vcpu->arch.pio.count = 0;
4853 vcpu->run->exit_reason = KVM_EXIT_IO;
4854 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4855 vcpu->run->io.size = size;
4856 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4857 vcpu->run->io.count = count;
4858 vcpu->run->io.port = port;
4863 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4864 int size, unsigned short port, void *val,
4867 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4870 if (vcpu->arch.pio.count)
4873 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4876 memcpy(val, vcpu->arch.pio_data, size * count);
4877 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4878 vcpu->arch.pio.count = 0;
4885 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4886 int size, unsigned short port,
4887 const void *val, unsigned int count)
4889 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4891 memcpy(vcpu->arch.pio_data, val, size * count);
4892 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4893 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4896 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4898 return kvm_x86_ops->get_segment_base(vcpu, seg);
4901 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4903 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4906 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4908 if (!need_emulate_wbinvd(vcpu))
4909 return X86EMUL_CONTINUE;
4911 if (kvm_x86_ops->has_wbinvd_exit()) {
4912 int cpu = get_cpu();
4914 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4915 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4916 wbinvd_ipi, NULL, 1);
4918 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4921 return X86EMUL_CONTINUE;
4924 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4926 kvm_emulate_wbinvd_noskip(vcpu);
4927 return kvm_skip_emulated_instruction(vcpu);
4929 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4933 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4935 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4938 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4939 unsigned long *dest)
4941 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4944 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4945 unsigned long value)
4948 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4951 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4953 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4956 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4958 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4959 unsigned long value;
4963 value = kvm_read_cr0(vcpu);
4966 value = vcpu->arch.cr2;
4969 value = kvm_read_cr3(vcpu);
4972 value = kvm_read_cr4(vcpu);
4975 value = kvm_get_cr8(vcpu);
4978 kvm_err("%s: unexpected cr %u\n", __func__, cr);
4985 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4987 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4992 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4995 vcpu->arch.cr2 = val;
4998 res = kvm_set_cr3(vcpu, val);
5001 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5004 res = kvm_set_cr8(vcpu, val);
5007 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5014 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5016 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5019 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5021 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5024 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5026 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5029 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5031 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5034 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5036 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5039 static unsigned long emulator_get_cached_segment_base(
5040 struct x86_emulate_ctxt *ctxt, int seg)
5042 return get_segment_base(emul_to_vcpu(ctxt), seg);
5045 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5046 struct desc_struct *desc, u32 *base3,
5049 struct kvm_segment var;
5051 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5052 *selector = var.selector;
5055 memset(desc, 0, sizeof(*desc));
5061 set_desc_limit(desc, var.limit);
5062 set_desc_base(desc, (unsigned long)var.base);
5063 #ifdef CONFIG_X86_64
5065 *base3 = var.base >> 32;
5067 desc->type = var.type;
5069 desc->dpl = var.dpl;
5070 desc->p = var.present;
5071 desc->avl = var.avl;
5079 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5080 struct desc_struct *desc, u32 base3,
5083 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5084 struct kvm_segment var;
5086 var.selector = selector;
5087 var.base = get_desc_base(desc);
5088 #ifdef CONFIG_X86_64
5089 var.base |= ((u64)base3) << 32;
5091 var.limit = get_desc_limit(desc);
5093 var.limit = (var.limit << 12) | 0xfff;
5094 var.type = desc->type;
5095 var.dpl = desc->dpl;
5100 var.avl = desc->avl;
5101 var.present = desc->p;
5102 var.unusable = !var.present;
5105 kvm_set_segment(vcpu, &var, seg);
5109 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5110 u32 msr_index, u64 *pdata)
5112 struct msr_data msr;
5115 msr.index = msr_index;
5116 msr.host_initiated = false;
5117 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5125 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5126 u32 msr_index, u64 data)
5128 struct msr_data msr;
5131 msr.index = msr_index;
5132 msr.host_initiated = false;
5133 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5136 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5138 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5140 return vcpu->arch.smbase;
5143 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5145 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5147 vcpu->arch.smbase = smbase;
5150 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5153 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5156 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5157 u32 pmc, u64 *pdata)
5159 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5162 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5164 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5167 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
5170 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
5173 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
5178 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5179 struct x86_instruction_info *info,
5180 enum x86_intercept_stage stage)
5182 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5185 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5186 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
5188 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
5191 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5193 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5196 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5198 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5201 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5203 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5206 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5208 return emul_to_vcpu(ctxt)->arch.hflags;
5211 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5213 kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5216 static const struct x86_emulate_ops emulate_ops = {
5217 .read_gpr = emulator_read_gpr,
5218 .write_gpr = emulator_write_gpr,
5219 .read_std = kvm_read_guest_virt_system,
5220 .write_std = kvm_write_guest_virt_system,
5221 .read_phys = kvm_read_guest_phys_system,
5222 .fetch = kvm_fetch_guest_virt,
5223 .read_emulated = emulator_read_emulated,
5224 .write_emulated = emulator_write_emulated,
5225 .cmpxchg_emulated = emulator_cmpxchg_emulated,
5226 .invlpg = emulator_invlpg,
5227 .pio_in_emulated = emulator_pio_in_emulated,
5228 .pio_out_emulated = emulator_pio_out_emulated,
5229 .get_segment = emulator_get_segment,
5230 .set_segment = emulator_set_segment,
5231 .get_cached_segment_base = emulator_get_cached_segment_base,
5232 .get_gdt = emulator_get_gdt,
5233 .get_idt = emulator_get_idt,
5234 .set_gdt = emulator_set_gdt,
5235 .set_idt = emulator_set_idt,
5236 .get_cr = emulator_get_cr,
5237 .set_cr = emulator_set_cr,
5238 .cpl = emulator_get_cpl,
5239 .get_dr = emulator_get_dr,
5240 .set_dr = emulator_set_dr,
5241 .get_smbase = emulator_get_smbase,
5242 .set_smbase = emulator_set_smbase,
5243 .set_msr = emulator_set_msr,
5244 .get_msr = emulator_get_msr,
5245 .check_pmc = emulator_check_pmc,
5246 .read_pmc = emulator_read_pmc,
5247 .halt = emulator_halt,
5248 .wbinvd = emulator_wbinvd,
5249 .fix_hypercall = emulator_fix_hypercall,
5250 .get_fpu = emulator_get_fpu,
5251 .put_fpu = emulator_put_fpu,
5252 .intercept = emulator_intercept,
5253 .get_cpuid = emulator_get_cpuid,
5254 .set_nmi_mask = emulator_set_nmi_mask,
5255 .get_hflags = emulator_get_hflags,
5256 .set_hflags = emulator_set_hflags,
5259 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5261 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5263 * an sti; sti; sequence only disable interrupts for the first
5264 * instruction. So, if the last instruction, be it emulated or
5265 * not, left the system with the INT_STI flag enabled, it
5266 * means that the last instruction is an sti. We should not
5267 * leave the flag on in this case. The same goes for mov ss
5269 if (int_shadow & mask)
5271 if (unlikely(int_shadow || mask)) {
5272 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5274 kvm_make_request(KVM_REQ_EVENT, vcpu);
5278 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5280 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5281 if (ctxt->exception.vector == PF_VECTOR)
5282 return kvm_propagate_fault(vcpu, &ctxt->exception);
5284 if (ctxt->exception.error_code_valid)
5285 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5286 ctxt->exception.error_code);
5288 kvm_queue_exception(vcpu, ctxt->exception.vector);
5292 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5294 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5297 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5299 ctxt->eflags = kvm_get_rflags(vcpu);
5300 ctxt->eip = kvm_rip_read(vcpu);
5301 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5302 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
5303 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
5304 cs_db ? X86EMUL_MODE_PROT32 :
5305 X86EMUL_MODE_PROT16;
5306 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5307 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5308 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5310 init_decode_cache(ctxt);
5311 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5314 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5316 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5319 init_emulate_ctxt(vcpu);
5323 ctxt->_eip = ctxt->eip + inc_eip;
5324 ret = emulate_int_real(ctxt, irq);
5326 if (ret != X86EMUL_CONTINUE)
5327 return EMULATE_FAIL;
5329 ctxt->eip = ctxt->_eip;
5330 kvm_rip_write(vcpu, ctxt->eip);
5331 kvm_set_rflags(vcpu, ctxt->eflags);
5333 if (irq == NMI_VECTOR)
5334 vcpu->arch.nmi_pending = 0;
5336 vcpu->arch.interrupt.pending = false;
5338 return EMULATE_DONE;
5340 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5342 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5344 int r = EMULATE_DONE;
5346 ++vcpu->stat.insn_emulation_fail;
5347 trace_kvm_emulate_insn_failed(vcpu);
5348 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5349 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5350 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5351 vcpu->run->internal.ndata = 0;
5354 kvm_queue_exception(vcpu, UD_VECTOR);
5359 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5360 bool write_fault_to_shadow_pgtable,
5366 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5369 if (!vcpu->arch.mmu.direct_map) {
5371 * Write permission should be allowed since only
5372 * write access need to be emulated.
5374 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5377 * If the mapping is invalid in guest, let cpu retry
5378 * it to generate fault.
5380 if (gpa == UNMAPPED_GVA)
5385 * Do not retry the unhandleable instruction if it faults on the
5386 * readonly host memory, otherwise it will goto a infinite loop:
5387 * retry instruction -> write #PF -> emulation fail -> retry
5388 * instruction -> ...
5390 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5393 * If the instruction failed on the error pfn, it can not be fixed,
5394 * report the error to userspace.
5396 if (is_error_noslot_pfn(pfn))
5399 kvm_release_pfn_clean(pfn);
5401 /* The instructions are well-emulated on direct mmu. */
5402 if (vcpu->arch.mmu.direct_map) {
5403 unsigned int indirect_shadow_pages;
5405 spin_lock(&vcpu->kvm->mmu_lock);
5406 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5407 spin_unlock(&vcpu->kvm->mmu_lock);
5409 if (indirect_shadow_pages)
5410 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5416 * if emulation was due to access to shadowed page table
5417 * and it failed try to unshadow page and re-enter the
5418 * guest to let CPU execute the instruction.
5420 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5423 * If the access faults on its page table, it can not
5424 * be fixed by unprotecting shadow page and it should
5425 * be reported to userspace.
5427 return !write_fault_to_shadow_pgtable;
5430 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5431 unsigned long cr2, int emulation_type)
5433 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5434 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5436 last_retry_eip = vcpu->arch.last_retry_eip;
5437 last_retry_addr = vcpu->arch.last_retry_addr;
5440 * If the emulation is caused by #PF and it is non-page_table
5441 * writing instruction, it means the VM-EXIT is caused by shadow
5442 * page protected, we can zap the shadow page and retry this
5443 * instruction directly.
5445 * Note: if the guest uses a non-page-table modifying instruction
5446 * on the PDE that points to the instruction, then we will unmap
5447 * the instruction and go to an infinite loop. So, we cache the
5448 * last retried eip and the last fault address, if we meet the eip
5449 * and the address again, we can break out of the potential infinite
5452 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5454 if (!(emulation_type & EMULTYPE_RETRY))
5457 if (x86_page_table_writing_insn(ctxt))
5460 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5463 vcpu->arch.last_retry_eip = ctxt->eip;
5464 vcpu->arch.last_retry_addr = cr2;
5466 if (!vcpu->arch.mmu.direct_map)
5467 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5469 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5474 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5475 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5477 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5479 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5480 /* This is a good place to trace that we are exiting SMM. */
5481 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5483 /* Process a latched INIT or SMI, if any. */
5484 kvm_make_request(KVM_REQ_EVENT, vcpu);
5487 kvm_mmu_reset_context(vcpu);
5490 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5492 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5494 vcpu->arch.hflags = emul_flags;
5496 if (changed & HF_SMM_MASK)
5497 kvm_smm_changed(vcpu);
5500 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5509 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5510 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5515 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5517 struct kvm_run *kvm_run = vcpu->run;
5520 * rflags is the old, "raw" value of the flags. The new value has
5521 * not been saved yet.
5523 * This is correct even for TF set by the guest, because "the
5524 * processor will not generate this exception after the instruction
5525 * that sets the TF flag".
5527 if (unlikely(rflags & X86_EFLAGS_TF)) {
5528 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5529 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5531 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5532 kvm_run->debug.arch.exception = DB_VECTOR;
5533 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5534 *r = EMULATE_USER_EXIT;
5537 * "Certain debug exceptions may clear bit 0-3. The
5538 * remaining contents of the DR6 register are never
5539 * cleared by the processor".
5541 vcpu->arch.dr6 &= ~15;
5542 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5543 kvm_queue_exception(vcpu, DB_VECTOR);
5548 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
5550 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5551 int r = EMULATE_DONE;
5553 kvm_x86_ops->skip_emulated_instruction(vcpu);
5554 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5555 return r == EMULATE_DONE;
5557 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
5559 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5561 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5562 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5563 struct kvm_run *kvm_run = vcpu->run;
5564 unsigned long eip = kvm_get_linear_rip(vcpu);
5565 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5566 vcpu->arch.guest_debug_dr7,
5570 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5571 kvm_run->debug.arch.pc = eip;
5572 kvm_run->debug.arch.exception = DB_VECTOR;
5573 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5574 *r = EMULATE_USER_EXIT;
5579 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5580 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5581 unsigned long eip = kvm_get_linear_rip(vcpu);
5582 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5587 vcpu->arch.dr6 &= ~15;
5588 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5589 kvm_queue_exception(vcpu, DB_VECTOR);
5598 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5605 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5606 bool writeback = true;
5607 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5610 * Clear write_fault_to_shadow_pgtable here to ensure it is
5613 vcpu->arch.write_fault_to_shadow_pgtable = false;
5614 kvm_clear_exception_queue(vcpu);
5616 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5617 init_emulate_ctxt(vcpu);
5620 * We will reenter on the same instruction since
5621 * we do not set complete_userspace_io. This does not
5622 * handle watchpoints yet, those would be handled in
5625 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5628 ctxt->interruptibility = 0;
5629 ctxt->have_exception = false;
5630 ctxt->exception.vector = -1;
5631 ctxt->perm_ok = false;
5633 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5635 r = x86_decode_insn(ctxt, insn, insn_len);
5637 trace_kvm_emulate_insn_start(vcpu);
5638 ++vcpu->stat.insn_emulation;
5639 if (r != EMULATION_OK) {
5640 if (emulation_type & EMULTYPE_TRAP_UD)
5641 return EMULATE_FAIL;
5642 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5644 return EMULATE_DONE;
5645 if (emulation_type & EMULTYPE_SKIP)
5646 return EMULATE_FAIL;
5647 return handle_emulation_failure(vcpu);
5651 if (emulation_type & EMULTYPE_SKIP) {
5652 kvm_rip_write(vcpu, ctxt->_eip);
5653 if (ctxt->eflags & X86_EFLAGS_RF)
5654 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5655 return EMULATE_DONE;
5658 if (retry_instruction(ctxt, cr2, emulation_type))
5659 return EMULATE_DONE;
5661 /* this is needed for vmware backdoor interface to work since it
5662 changes registers values during IO operation */
5663 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5664 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5665 emulator_invalidate_register_cache(ctxt);
5669 /* Save the faulting GPA (cr2) in the address field */
5670 ctxt->exception.address = cr2;
5672 r = x86_emulate_insn(ctxt);
5674 if (r == EMULATION_INTERCEPTED)
5675 return EMULATE_DONE;
5677 if (r == EMULATION_FAILED) {
5678 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5680 return EMULATE_DONE;
5682 return handle_emulation_failure(vcpu);
5685 if (ctxt->have_exception) {
5687 if (inject_emulated_exception(vcpu))
5689 } else if (vcpu->arch.pio.count) {
5690 if (!vcpu->arch.pio.in) {
5691 /* FIXME: return into emulator if single-stepping. */
5692 vcpu->arch.pio.count = 0;
5695 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5697 r = EMULATE_USER_EXIT;
5698 } else if (vcpu->mmio_needed) {
5699 if (!vcpu->mmio_is_write)
5701 r = EMULATE_USER_EXIT;
5702 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5703 } else if (r == EMULATION_RESTART)
5709 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5710 toggle_interruptibility(vcpu, ctxt->interruptibility);
5711 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5712 kvm_rip_write(vcpu, ctxt->eip);
5713 if (r == EMULATE_DONE)
5714 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5715 if (!ctxt->have_exception ||
5716 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5717 __kvm_set_rflags(vcpu, ctxt->eflags);
5720 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5721 * do nothing, and it will be requested again as soon as
5722 * the shadow expires. But we still need to check here,
5723 * because POPF has no interrupt shadow.
5725 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5726 kvm_make_request(KVM_REQ_EVENT, vcpu);
5728 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5732 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5734 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5736 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5737 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5738 size, port, &val, 1);
5739 /* do not return to emulator after return from userspace */
5740 vcpu->arch.pio.count = 0;
5743 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5745 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
5749 /* We should only ever be called with arch.pio.count equal to 1 */
5750 BUG_ON(vcpu->arch.pio.count != 1);
5752 /* For size less than 4 we merge, else we zero extend */
5753 val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
5757 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
5758 * the copy and tracing
5760 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
5761 vcpu->arch.pio.port, &val, 1);
5762 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5767 int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size, unsigned short port)
5772 /* For size less than 4 we merge, else we zero extend */
5773 val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
5775 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
5778 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5782 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
5786 EXPORT_SYMBOL_GPL(kvm_fast_pio_in);
5788 static int kvmclock_cpu_down_prep(unsigned int cpu)
5790 __this_cpu_write(cpu_tsc_khz, 0);
5794 static void tsc_khz_changed(void *data)
5796 struct cpufreq_freqs *freq = data;
5797 unsigned long khz = 0;
5801 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5802 khz = cpufreq_quick_get(raw_smp_processor_id());
5805 __this_cpu_write(cpu_tsc_khz, khz);
5808 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5811 struct cpufreq_freqs *freq = data;
5813 struct kvm_vcpu *vcpu;
5814 int i, send_ipi = 0;
5817 * We allow guests to temporarily run on slowing clocks,
5818 * provided we notify them after, or to run on accelerating
5819 * clocks, provided we notify them before. Thus time never
5822 * However, we have a problem. We can't atomically update
5823 * the frequency of a given CPU from this function; it is
5824 * merely a notifier, which can be called from any CPU.
5825 * Changing the TSC frequency at arbitrary points in time
5826 * requires a recomputation of local variables related to
5827 * the TSC for each VCPU. We must flag these local variables
5828 * to be updated and be sure the update takes place with the
5829 * new frequency before any guests proceed.
5831 * Unfortunately, the combination of hotplug CPU and frequency
5832 * change creates an intractable locking scenario; the order
5833 * of when these callouts happen is undefined with respect to
5834 * CPU hotplug, and they can race with each other. As such,
5835 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5836 * undefined; you can actually have a CPU frequency change take
5837 * place in between the computation of X and the setting of the
5838 * variable. To protect against this problem, all updates of
5839 * the per_cpu tsc_khz variable are done in an interrupt
5840 * protected IPI, and all callers wishing to update the value
5841 * must wait for a synchronous IPI to complete (which is trivial
5842 * if the caller is on the CPU already). This establishes the
5843 * necessary total order on variable updates.
5845 * Note that because a guest time update may take place
5846 * anytime after the setting of the VCPU's request bit, the
5847 * correct TSC value must be set before the request. However,
5848 * to ensure the update actually makes it to any guest which
5849 * starts running in hardware virtualization between the set
5850 * and the acquisition of the spinlock, we must also ping the
5851 * CPU after setting the request bit.
5855 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5857 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5860 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5862 spin_lock(&kvm_lock);
5863 list_for_each_entry(kvm, &vm_list, vm_list) {
5864 kvm_for_each_vcpu(i, vcpu, kvm) {
5865 if (vcpu->cpu != freq->cpu)
5867 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5868 if (vcpu->cpu != smp_processor_id())
5872 spin_unlock(&kvm_lock);
5874 if (freq->old < freq->new && send_ipi) {
5876 * We upscale the frequency. Must make the guest
5877 * doesn't see old kvmclock values while running with
5878 * the new frequency, otherwise we risk the guest sees
5879 * time go backwards.
5881 * In case we update the frequency for another cpu
5882 * (which might be in guest context) send an interrupt
5883 * to kick the cpu out of guest context. Next time
5884 * guest context is entered kvmclock will be updated,
5885 * so the guest will not see stale values.
5887 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5892 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5893 .notifier_call = kvmclock_cpufreq_notifier
5896 static int kvmclock_cpu_online(unsigned int cpu)
5898 tsc_khz_changed(NULL);
5902 static void kvm_timer_init(void)
5904 max_tsc_khz = tsc_khz;
5906 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5907 #ifdef CONFIG_CPU_FREQ
5908 struct cpufreq_policy policy;
5911 memset(&policy, 0, sizeof(policy));
5913 cpufreq_get_policy(&policy, cpu);
5914 if (policy.cpuinfo.max_freq)
5915 max_tsc_khz = policy.cpuinfo.max_freq;
5918 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5919 CPUFREQ_TRANSITION_NOTIFIER);
5921 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5923 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
5924 kvmclock_cpu_online, kvmclock_cpu_down_prep);
5927 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5929 int kvm_is_in_guest(void)
5931 return __this_cpu_read(current_vcpu) != NULL;
5934 static int kvm_is_user_mode(void)
5938 if (__this_cpu_read(current_vcpu))
5939 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5941 return user_mode != 0;
5944 static unsigned long kvm_get_guest_ip(void)
5946 unsigned long ip = 0;
5948 if (__this_cpu_read(current_vcpu))
5949 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5954 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5955 .is_in_guest = kvm_is_in_guest,
5956 .is_user_mode = kvm_is_user_mode,
5957 .get_guest_ip = kvm_get_guest_ip,
5960 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5962 __this_cpu_write(current_vcpu, vcpu);
5964 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5966 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5968 __this_cpu_write(current_vcpu, NULL);
5970 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5972 static void kvm_set_mmio_spte_mask(void)
5975 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5978 * Set the reserved bits and the present bit of an paging-structure
5979 * entry to generate page fault with PFER.RSV = 1.
5981 /* Mask the reserved physical address bits. */
5982 mask = rsvd_bits(maxphyaddr, 51);
5984 /* Set the present bit. */
5987 #ifdef CONFIG_X86_64
5989 * If reserved bit is not supported, clear the present bit to disable
5992 if (maxphyaddr == 52)
5996 kvm_mmu_set_mmio_spte_mask(mask);
5999 #ifdef CONFIG_X86_64
6000 static void pvclock_gtod_update_fn(struct work_struct *work)
6004 struct kvm_vcpu *vcpu;
6007 spin_lock(&kvm_lock);
6008 list_for_each_entry(kvm, &vm_list, vm_list)
6009 kvm_for_each_vcpu(i, vcpu, kvm)
6010 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6011 atomic_set(&kvm_guest_has_master_clock, 0);
6012 spin_unlock(&kvm_lock);
6015 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6018 * Notification about pvclock gtod data update.
6020 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6023 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6024 struct timekeeper *tk = priv;
6026 update_pvclock_gtod(tk);
6028 /* disable master clock if host does not trust, or does not
6029 * use, TSC clocksource
6031 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
6032 atomic_read(&kvm_guest_has_master_clock) != 0)
6033 queue_work(system_long_wq, &pvclock_gtod_work);
6038 static struct notifier_block pvclock_gtod_notifier = {
6039 .notifier_call = pvclock_gtod_notify,
6043 int kvm_arch_init(void *opaque)
6046 struct kvm_x86_ops *ops = opaque;
6049 printk(KERN_ERR "kvm: already loaded the other module\n");
6054 if (!ops->cpu_has_kvm_support()) {
6055 printk(KERN_ERR "kvm: no hardware support\n");
6059 if (ops->disabled_by_bios()) {
6060 printk(KERN_ERR "kvm: disabled by bios\n");
6066 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6068 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6072 r = kvm_mmu_module_init();
6074 goto out_free_percpu;
6076 kvm_set_mmio_spte_mask();
6080 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6081 PT_DIRTY_MASK, PT64_NX_MASK, 0,
6082 PT_PRESENT_MASK, 0);
6085 perf_register_guest_info_callbacks(&kvm_guest_cbs);
6087 if (boot_cpu_has(X86_FEATURE_XSAVE))
6088 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6091 #ifdef CONFIG_X86_64
6092 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6098 free_percpu(shared_msrs);
6103 void kvm_arch_exit(void)
6106 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6108 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6109 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6110 CPUFREQ_TRANSITION_NOTIFIER);
6111 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6112 #ifdef CONFIG_X86_64
6113 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6116 kvm_mmu_module_exit();
6117 free_percpu(shared_msrs);
6120 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6122 ++vcpu->stat.halt_exits;
6123 if (lapic_in_kernel(vcpu)) {
6124 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6127 vcpu->run->exit_reason = KVM_EXIT_HLT;
6131 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6133 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6135 int ret = kvm_skip_emulated_instruction(vcpu);
6137 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6138 * KVM_EXIT_DEBUG here.
6140 return kvm_vcpu_halt(vcpu) && ret;
6142 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6144 #ifdef CONFIG_X86_64
6145 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6146 unsigned long clock_type)
6148 struct kvm_clock_pairing clock_pairing;
6153 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6154 return -KVM_EOPNOTSUPP;
6156 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6157 return -KVM_EOPNOTSUPP;
6159 clock_pairing.sec = ts.tv_sec;
6160 clock_pairing.nsec = ts.tv_nsec;
6161 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6162 clock_pairing.flags = 0;
6165 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6166 sizeof(struct kvm_clock_pairing)))
6174 * kvm_pv_kick_cpu_op: Kick a vcpu.
6176 * @apicid - apicid of vcpu to be kicked.
6178 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6180 struct kvm_lapic_irq lapic_irq;
6182 lapic_irq.shorthand = 0;
6183 lapic_irq.dest_mode = 0;
6184 lapic_irq.dest_id = apicid;
6185 lapic_irq.msi_redir_hint = false;
6187 lapic_irq.delivery_mode = APIC_DM_REMRD;
6188 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6191 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6193 vcpu->arch.apicv_active = false;
6194 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6197 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6199 unsigned long nr, a0, a1, a2, a3, ret;
6202 r = kvm_skip_emulated_instruction(vcpu);
6204 if (kvm_hv_hypercall_enabled(vcpu->kvm))
6205 return kvm_hv_hypercall(vcpu);
6207 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6208 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6209 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6210 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6211 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
6213 trace_kvm_hypercall(nr, a0, a1, a2, a3);
6215 op_64_bit = is_64_bit_mode(vcpu);
6224 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6230 case KVM_HC_VAPIC_POLL_IRQ:
6233 case KVM_HC_KICK_CPU:
6234 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6237 #ifdef CONFIG_X86_64
6238 case KVM_HC_CLOCK_PAIRING:
6239 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
6249 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6250 ++vcpu->stat.hypercalls;
6253 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6255 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6257 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6258 char instruction[3];
6259 unsigned long rip = kvm_rip_read(vcpu);
6261 kvm_x86_ops->patch_hypercall(vcpu, instruction);
6263 return emulator_write_emulated(ctxt, rip, instruction, 3,
6267 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6269 return vcpu->run->request_interrupt_window &&
6270 likely(!pic_in_kernel(vcpu->kvm));
6273 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6275 struct kvm_run *kvm_run = vcpu->run;
6277 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6278 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6279 kvm_run->cr8 = kvm_get_cr8(vcpu);
6280 kvm_run->apic_base = kvm_get_apic_base(vcpu);
6281 kvm_run->ready_for_interrupt_injection =
6282 pic_in_kernel(vcpu->kvm) ||
6283 kvm_vcpu_ready_for_interrupt_injection(vcpu);
6286 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6290 if (!kvm_x86_ops->update_cr8_intercept)
6293 if (!lapic_in_kernel(vcpu))
6296 if (vcpu->arch.apicv_active)
6299 if (!vcpu->arch.apic->vapic_addr)
6300 max_irr = kvm_lapic_find_highest_irr(vcpu);
6307 tpr = kvm_lapic_get_cr8(vcpu);
6309 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6312 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6316 /* try to reinject previous events if any */
6317 if (vcpu->arch.exception.pending) {
6318 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6319 vcpu->arch.exception.has_error_code,
6320 vcpu->arch.exception.error_code);
6322 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6323 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6326 if (vcpu->arch.exception.nr == DB_VECTOR &&
6327 (vcpu->arch.dr7 & DR7_GD)) {
6328 vcpu->arch.dr7 &= ~DR7_GD;
6329 kvm_update_dr7(vcpu);
6332 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6333 vcpu->arch.exception.has_error_code,
6334 vcpu->arch.exception.error_code,
6335 vcpu->arch.exception.reinject);
6339 if (vcpu->arch.nmi_injected) {
6340 kvm_x86_ops->set_nmi(vcpu);
6344 if (vcpu->arch.interrupt.pending) {
6345 kvm_x86_ops->set_irq(vcpu);
6349 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6350 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6355 /* try to inject new event if pending */
6356 if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
6357 vcpu->arch.smi_pending = false;
6359 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6360 --vcpu->arch.nmi_pending;
6361 vcpu->arch.nmi_injected = true;
6362 kvm_x86_ops->set_nmi(vcpu);
6363 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6365 * Because interrupts can be injected asynchronously, we are
6366 * calling check_nested_events again here to avoid a race condition.
6367 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6368 * proposal and current concerns. Perhaps we should be setting
6369 * KVM_REQ_EVENT only on certain events and not unconditionally?
6371 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6372 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6376 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6377 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6379 kvm_x86_ops->set_irq(vcpu);
6386 static void process_nmi(struct kvm_vcpu *vcpu)
6391 * x86 is limited to one NMI running, and one NMI pending after it.
6392 * If an NMI is already in progress, limit further NMIs to just one.
6393 * Otherwise, allow two (and we'll inject the first one immediately).
6395 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6398 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6399 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6400 kvm_make_request(KVM_REQ_EVENT, vcpu);
6403 #define put_smstate(type, buf, offset, val) \
6404 *(type *)((buf) + (offset) - 0x7e00) = val
6406 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
6409 flags |= seg->g << 23;
6410 flags |= seg->db << 22;
6411 flags |= seg->l << 21;
6412 flags |= seg->avl << 20;
6413 flags |= seg->present << 15;
6414 flags |= seg->dpl << 13;
6415 flags |= seg->s << 12;
6416 flags |= seg->type << 8;
6420 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6422 struct kvm_segment seg;
6425 kvm_get_segment(vcpu, &seg, n);
6426 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6429 offset = 0x7f84 + n * 12;
6431 offset = 0x7f2c + (n - 3) * 12;
6433 put_smstate(u32, buf, offset + 8, seg.base);
6434 put_smstate(u32, buf, offset + 4, seg.limit);
6435 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
6438 #ifdef CONFIG_X86_64
6439 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6441 struct kvm_segment seg;
6445 kvm_get_segment(vcpu, &seg, n);
6446 offset = 0x7e00 + n * 16;
6448 flags = enter_smm_get_segment_flags(&seg) >> 8;
6449 put_smstate(u16, buf, offset, seg.selector);
6450 put_smstate(u16, buf, offset + 2, flags);
6451 put_smstate(u32, buf, offset + 4, seg.limit);
6452 put_smstate(u64, buf, offset + 8, seg.base);
6456 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6459 struct kvm_segment seg;
6463 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6464 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6465 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6466 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6468 for (i = 0; i < 8; i++)
6469 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6471 kvm_get_dr(vcpu, 6, &val);
6472 put_smstate(u32, buf, 0x7fcc, (u32)val);
6473 kvm_get_dr(vcpu, 7, &val);
6474 put_smstate(u32, buf, 0x7fc8, (u32)val);
6476 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6477 put_smstate(u32, buf, 0x7fc4, seg.selector);
6478 put_smstate(u32, buf, 0x7f64, seg.base);
6479 put_smstate(u32, buf, 0x7f60, seg.limit);
6480 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
6482 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6483 put_smstate(u32, buf, 0x7fc0, seg.selector);
6484 put_smstate(u32, buf, 0x7f80, seg.base);
6485 put_smstate(u32, buf, 0x7f7c, seg.limit);
6486 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
6488 kvm_x86_ops->get_gdt(vcpu, &dt);
6489 put_smstate(u32, buf, 0x7f74, dt.address);
6490 put_smstate(u32, buf, 0x7f70, dt.size);
6492 kvm_x86_ops->get_idt(vcpu, &dt);
6493 put_smstate(u32, buf, 0x7f58, dt.address);
6494 put_smstate(u32, buf, 0x7f54, dt.size);
6496 for (i = 0; i < 6; i++)
6497 enter_smm_save_seg_32(vcpu, buf, i);
6499 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6502 put_smstate(u32, buf, 0x7efc, 0x00020000);
6503 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6506 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6508 #ifdef CONFIG_X86_64
6510 struct kvm_segment seg;
6514 for (i = 0; i < 16; i++)
6515 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6517 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6518 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6520 kvm_get_dr(vcpu, 6, &val);
6521 put_smstate(u64, buf, 0x7f68, val);
6522 kvm_get_dr(vcpu, 7, &val);
6523 put_smstate(u64, buf, 0x7f60, val);
6525 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6526 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6527 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6529 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6532 put_smstate(u32, buf, 0x7efc, 0x00020064);
6534 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6536 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6537 put_smstate(u16, buf, 0x7e90, seg.selector);
6538 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
6539 put_smstate(u32, buf, 0x7e94, seg.limit);
6540 put_smstate(u64, buf, 0x7e98, seg.base);
6542 kvm_x86_ops->get_idt(vcpu, &dt);
6543 put_smstate(u32, buf, 0x7e84, dt.size);
6544 put_smstate(u64, buf, 0x7e88, dt.address);
6546 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6547 put_smstate(u16, buf, 0x7e70, seg.selector);
6548 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
6549 put_smstate(u32, buf, 0x7e74, seg.limit);
6550 put_smstate(u64, buf, 0x7e78, seg.base);
6552 kvm_x86_ops->get_gdt(vcpu, &dt);
6553 put_smstate(u32, buf, 0x7e64, dt.size);
6554 put_smstate(u64, buf, 0x7e68, dt.address);
6556 for (i = 0; i < 6; i++)
6557 enter_smm_save_seg_64(vcpu, buf, i);
6563 static void enter_smm(struct kvm_vcpu *vcpu)
6565 struct kvm_segment cs, ds;
6570 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6571 vcpu->arch.hflags |= HF_SMM_MASK;
6572 memset(buf, 0, 512);
6573 if (guest_cpuid_has_longmode(vcpu))
6574 enter_smm_save_state_64(vcpu, buf);
6576 enter_smm_save_state_32(vcpu, buf);
6578 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6580 if (kvm_x86_ops->get_nmi_mask(vcpu))
6581 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6583 kvm_x86_ops->set_nmi_mask(vcpu, true);
6585 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6586 kvm_rip_write(vcpu, 0x8000);
6588 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6589 kvm_x86_ops->set_cr0(vcpu, cr0);
6590 vcpu->arch.cr0 = cr0;
6592 kvm_x86_ops->set_cr4(vcpu, 0);
6594 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6595 dt.address = dt.size = 0;
6596 kvm_x86_ops->set_idt(vcpu, &dt);
6598 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6600 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6601 cs.base = vcpu->arch.smbase;
6606 cs.limit = ds.limit = 0xffffffff;
6607 cs.type = ds.type = 0x3;
6608 cs.dpl = ds.dpl = 0;
6613 cs.avl = ds.avl = 0;
6614 cs.present = ds.present = 1;
6615 cs.unusable = ds.unusable = 0;
6616 cs.padding = ds.padding = 0;
6618 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6619 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6620 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6621 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6622 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6623 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6625 if (guest_cpuid_has_longmode(vcpu))
6626 kvm_x86_ops->set_efer(vcpu, 0);
6628 kvm_update_cpuid(vcpu);
6629 kvm_mmu_reset_context(vcpu);
6632 static void process_smi(struct kvm_vcpu *vcpu)
6634 vcpu->arch.smi_pending = true;
6635 kvm_make_request(KVM_REQ_EVENT, vcpu);
6638 void kvm_make_scan_ioapic_request(struct kvm *kvm)
6640 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6643 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6645 u64 eoi_exit_bitmap[4];
6647 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6650 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
6652 if (irqchip_split(vcpu->kvm))
6653 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
6655 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
6656 kvm_x86_ops->sync_pir_to_irr(vcpu);
6657 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
6659 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6660 vcpu_to_synic(vcpu)->vec_bitmap, 256);
6661 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6664 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6666 ++vcpu->stat.tlb_flush;
6667 kvm_x86_ops->tlb_flush(vcpu);
6670 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6672 struct page *page = NULL;
6674 if (!lapic_in_kernel(vcpu))
6677 if (!kvm_x86_ops->set_apic_access_page_addr)
6680 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6681 if (is_error_page(page))
6683 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6686 * Do not pin apic access page in memory, the MMU notifier
6687 * will call us again if it is migrated or swapped out.
6691 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6693 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6694 unsigned long address)
6697 * The physical address of apic access page is stored in the VMCS.
6698 * Update it when it becomes invalid.
6700 if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6701 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6705 * Returns 1 to let vcpu_run() continue the guest execution loop without
6706 * exiting to the userspace. Otherwise, the value will be returned to the
6709 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6713 dm_request_for_irq_injection(vcpu) &&
6714 kvm_cpu_accept_dm_intr(vcpu);
6716 bool req_immediate_exit = false;
6718 if (vcpu->requests) {
6719 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6720 kvm_mmu_unload(vcpu);
6721 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6722 __kvm_migrate_timers(vcpu);
6723 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6724 kvm_gen_update_masterclock(vcpu->kvm);
6725 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6726 kvm_gen_kvmclock_update(vcpu);
6727 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6728 r = kvm_guest_time_update(vcpu);
6732 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6733 kvm_mmu_sync_roots(vcpu);
6734 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6735 kvm_vcpu_flush_tlb(vcpu);
6736 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6737 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6741 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6742 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6746 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6747 /* Page is swapped out. Do synthetic halt */
6748 vcpu->arch.apf.halted = true;
6752 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6753 record_steal_time(vcpu);
6754 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6756 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6758 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6759 kvm_pmu_handle_event(vcpu);
6760 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6761 kvm_pmu_deliver_pmi(vcpu);
6762 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6763 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6764 if (test_bit(vcpu->arch.pending_ioapic_eoi,
6765 vcpu->arch.ioapic_handled_vectors)) {
6766 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6767 vcpu->run->eoi.vector =
6768 vcpu->arch.pending_ioapic_eoi;
6773 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6774 vcpu_scan_ioapic(vcpu);
6775 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6776 kvm_vcpu_reload_apic_access_page(vcpu);
6777 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6778 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6779 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6783 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6784 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6785 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6789 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6790 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6791 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6797 * KVM_REQ_HV_STIMER has to be processed after
6798 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6799 * depend on the guest clock being up-to-date
6801 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6802 kvm_hv_process_stimers(vcpu);
6805 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6806 ++vcpu->stat.req_event;
6807 kvm_apic_accept_events(vcpu);
6808 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6813 if (inject_pending_event(vcpu, req_int_win) != 0)
6814 req_immediate_exit = true;
6816 /* Enable NMI/IRQ window open exits if needed.
6818 * SMIs have two cases: 1) they can be nested, and
6819 * then there is nothing to do here because RSM will
6820 * cause a vmexit anyway; 2) or the SMI can be pending
6821 * because inject_pending_event has completed the
6822 * injection of an IRQ or NMI from the previous vmexit,
6823 * and then we request an immediate exit to inject the SMI.
6825 if (vcpu->arch.smi_pending && !is_smm(vcpu))
6826 req_immediate_exit = true;
6827 if (vcpu->arch.nmi_pending)
6828 kvm_x86_ops->enable_nmi_window(vcpu);
6829 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6830 kvm_x86_ops->enable_irq_window(vcpu);
6833 if (kvm_lapic_enabled(vcpu)) {
6834 update_cr8_intercept(vcpu);
6835 kvm_lapic_sync_to_vapic(vcpu);
6839 r = kvm_mmu_reload(vcpu);
6841 goto cancel_injection;
6846 kvm_x86_ops->prepare_guest_switch(vcpu);
6847 kvm_load_guest_fpu(vcpu);
6850 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
6851 * IPI are then delayed after guest entry, which ensures that they
6852 * result in virtual interrupt delivery.
6854 local_irq_disable();
6855 vcpu->mode = IN_GUEST_MODE;
6857 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6860 * 1) We should set ->mode before checking ->requests. Please see
6861 * the comment in kvm_vcpu_exiting_guest_mode().
6863 * 2) For APICv, we should set ->mode before checking PIR.ON. This
6864 * pairs with the memory barrier implicit in pi_test_and_set_on
6865 * (see vmx_deliver_posted_interrupt).
6867 * 3) This also orders the write to mode from any reads to the page
6868 * tables done while the VCPU is running. Please see the comment
6869 * in kvm_flush_remote_tlbs.
6871 smp_mb__after_srcu_read_unlock();
6874 * This handles the case where a posted interrupt was
6875 * notified with kvm_vcpu_kick.
6877 if (kvm_lapic_enabled(vcpu)) {
6878 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
6879 kvm_x86_ops->sync_pir_to_irr(vcpu);
6882 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6883 || need_resched() || signal_pending(current)) {
6884 vcpu->mode = OUTSIDE_GUEST_MODE;
6888 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6890 goto cancel_injection;
6893 kvm_load_guest_xcr0(vcpu);
6895 if (req_immediate_exit) {
6896 kvm_make_request(KVM_REQ_EVENT, vcpu);
6897 smp_send_reschedule(vcpu->cpu);
6900 trace_kvm_entry(vcpu->vcpu_id);
6901 wait_lapic_expire(vcpu);
6902 guest_enter_irqoff();
6904 if (unlikely(vcpu->arch.switch_db_regs)) {
6906 set_debugreg(vcpu->arch.eff_db[0], 0);
6907 set_debugreg(vcpu->arch.eff_db[1], 1);
6908 set_debugreg(vcpu->arch.eff_db[2], 2);
6909 set_debugreg(vcpu->arch.eff_db[3], 3);
6910 set_debugreg(vcpu->arch.dr6, 6);
6911 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6914 kvm_x86_ops->run(vcpu);
6917 * Do this here before restoring debug registers on the host. And
6918 * since we do this before handling the vmexit, a DR access vmexit
6919 * can (a) read the correct value of the debug registers, (b) set
6920 * KVM_DEBUGREG_WONT_EXIT again.
6922 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6923 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6924 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6925 kvm_update_dr0123(vcpu);
6926 kvm_update_dr6(vcpu);
6927 kvm_update_dr7(vcpu);
6928 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6932 * If the guest has used debug registers, at least dr7
6933 * will be disabled while returning to the host.
6934 * If we don't have active breakpoints in the host, we don't
6935 * care about the messed up debug address registers. But if
6936 * we have some of them active, restore the old state.
6938 if (hw_breakpoint_active())
6939 hw_breakpoint_restore();
6941 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
6943 vcpu->mode = OUTSIDE_GUEST_MODE;
6946 kvm_put_guest_xcr0(vcpu);
6948 kvm_x86_ops->handle_external_intr(vcpu);
6952 guest_exit_irqoff();
6957 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6960 * Profile KVM exit RIPs:
6962 if (unlikely(prof_on == KVM_PROFILING)) {
6963 unsigned long rip = kvm_rip_read(vcpu);
6964 profile_hit(KVM_PROFILING, (void *)rip);
6967 if (unlikely(vcpu->arch.tsc_always_catchup))
6968 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6970 if (vcpu->arch.apic_attention)
6971 kvm_lapic_sync_from_vapic(vcpu);
6973 r = kvm_x86_ops->handle_exit(vcpu);
6977 kvm_x86_ops->cancel_injection(vcpu);
6978 if (unlikely(vcpu->arch.apic_attention))
6979 kvm_lapic_sync_from_vapic(vcpu);
6984 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6986 if (!kvm_arch_vcpu_runnable(vcpu) &&
6987 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
6988 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
6989 kvm_vcpu_block(vcpu);
6990 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
6992 if (kvm_x86_ops->post_block)
6993 kvm_x86_ops->post_block(vcpu);
6995 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
6999 kvm_apic_accept_events(vcpu);
7000 switch(vcpu->arch.mp_state) {
7001 case KVM_MP_STATE_HALTED:
7002 vcpu->arch.pv.pv_unhalted = false;
7003 vcpu->arch.mp_state =
7004 KVM_MP_STATE_RUNNABLE;
7005 case KVM_MP_STATE_RUNNABLE:
7006 vcpu->arch.apf.halted = false;
7008 case KVM_MP_STATE_INIT_RECEIVED:
7017 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7019 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7020 kvm_x86_ops->check_nested_events(vcpu, false);
7022 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7023 !vcpu->arch.apf.halted);
7026 static int vcpu_run(struct kvm_vcpu *vcpu)
7029 struct kvm *kvm = vcpu->kvm;
7031 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7034 if (kvm_vcpu_running(vcpu)) {
7035 r = vcpu_enter_guest(vcpu);
7037 r = vcpu_block(kvm, vcpu);
7043 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
7044 if (kvm_cpu_has_pending_timer(vcpu))
7045 kvm_inject_pending_timer_irqs(vcpu);
7047 if (dm_request_for_irq_injection(vcpu) &&
7048 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
7050 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
7051 ++vcpu->stat.request_irq_exits;
7055 kvm_check_async_pf_completion(vcpu);
7057 if (signal_pending(current)) {
7059 vcpu->run->exit_reason = KVM_EXIT_INTR;
7060 ++vcpu->stat.signal_exits;
7063 if (need_resched()) {
7064 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7066 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7070 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7075 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7078 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7079 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7080 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7081 if (r != EMULATE_DONE)
7086 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7088 BUG_ON(!vcpu->arch.pio.count);
7090 return complete_emulated_io(vcpu);
7094 * Implements the following, as a state machine:
7098 * for each mmio piece in the fragment
7106 * for each mmio piece in the fragment
7111 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
7113 struct kvm_run *run = vcpu->run;
7114 struct kvm_mmio_fragment *frag;
7117 BUG_ON(!vcpu->mmio_needed);
7119 /* Complete previous fragment */
7120 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7121 len = min(8u, frag->len);
7122 if (!vcpu->mmio_is_write)
7123 memcpy(frag->data, run->mmio.data, len);
7125 if (frag->len <= 8) {
7126 /* Switch to the next fragment. */
7128 vcpu->mmio_cur_fragment++;
7130 /* Go forward to the next mmio piece. */
7136 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
7137 vcpu->mmio_needed = 0;
7139 /* FIXME: return into emulator if single-stepping. */
7140 if (vcpu->mmio_is_write)
7142 vcpu->mmio_read_completed = 1;
7143 return complete_emulated_io(vcpu);
7146 run->exit_reason = KVM_EXIT_MMIO;
7147 run->mmio.phys_addr = frag->gpa;
7148 if (vcpu->mmio_is_write)
7149 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7150 run->mmio.len = min(8u, frag->len);
7151 run->mmio.is_write = vcpu->mmio_is_write;
7152 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7157 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7159 struct fpu *fpu = ¤t->thread.fpu;
7163 fpu__activate_curr(fpu);
7165 if (vcpu->sigset_active)
7166 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
7168 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
7169 kvm_vcpu_block(vcpu);
7170 kvm_apic_accept_events(vcpu);
7171 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
7176 /* re-sync apic's tpr */
7177 if (!lapic_in_kernel(vcpu)) {
7178 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7184 if (unlikely(vcpu->arch.complete_userspace_io)) {
7185 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7186 vcpu->arch.complete_userspace_io = NULL;
7191 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
7193 if (kvm_run->immediate_exit)
7199 post_kvm_run_save(vcpu);
7200 if (vcpu->sigset_active)
7201 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
7206 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7208 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
7210 * We are here if userspace calls get_regs() in the middle of
7211 * instruction emulation. Registers state needs to be copied
7212 * back from emulation context to vcpu. Userspace shouldn't do
7213 * that usually, but some bad designed PV devices (vmware
7214 * backdoor interface) need this to work
7216 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7217 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7219 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
7220 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
7221 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
7222 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
7223 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
7224 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
7225 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7226 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
7227 #ifdef CONFIG_X86_64
7228 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7229 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7230 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7231 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7232 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7233 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7234 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7235 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
7238 regs->rip = kvm_rip_read(vcpu);
7239 regs->rflags = kvm_get_rflags(vcpu);
7244 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7246 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7247 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7249 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7250 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7251 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7252 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7253 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7254 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7255 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7256 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
7257 #ifdef CONFIG_X86_64
7258 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7259 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7260 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7261 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7262 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7263 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7264 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7265 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
7268 kvm_rip_write(vcpu, regs->rip);
7269 kvm_set_rflags(vcpu, regs->rflags);
7271 vcpu->arch.exception.pending = false;
7273 kvm_make_request(KVM_REQ_EVENT, vcpu);
7278 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7280 struct kvm_segment cs;
7282 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7286 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7288 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7289 struct kvm_sregs *sregs)
7293 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7294 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7295 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7296 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7297 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7298 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7300 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7301 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7303 kvm_x86_ops->get_idt(vcpu, &dt);
7304 sregs->idt.limit = dt.size;
7305 sregs->idt.base = dt.address;
7306 kvm_x86_ops->get_gdt(vcpu, &dt);
7307 sregs->gdt.limit = dt.size;
7308 sregs->gdt.base = dt.address;
7310 sregs->cr0 = kvm_read_cr0(vcpu);
7311 sregs->cr2 = vcpu->arch.cr2;
7312 sregs->cr3 = kvm_read_cr3(vcpu);
7313 sregs->cr4 = kvm_read_cr4(vcpu);
7314 sregs->cr8 = kvm_get_cr8(vcpu);
7315 sregs->efer = vcpu->arch.efer;
7316 sregs->apic_base = kvm_get_apic_base(vcpu);
7318 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7320 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
7321 set_bit(vcpu->arch.interrupt.nr,
7322 (unsigned long *)sregs->interrupt_bitmap);
7327 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7328 struct kvm_mp_state *mp_state)
7330 kvm_apic_accept_events(vcpu);
7331 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7332 vcpu->arch.pv.pv_unhalted)
7333 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7335 mp_state->mp_state = vcpu->arch.mp_state;
7340 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7341 struct kvm_mp_state *mp_state)
7343 if (!lapic_in_kernel(vcpu) &&
7344 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7347 /* INITs are latched while in SMM */
7348 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
7349 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
7350 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
7353 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7354 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7355 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7357 vcpu->arch.mp_state = mp_state->mp_state;
7358 kvm_make_request(KVM_REQ_EVENT, vcpu);
7362 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7363 int reason, bool has_error_code, u32 error_code)
7365 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7368 init_emulate_ctxt(vcpu);
7370 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7371 has_error_code, error_code);
7374 return EMULATE_FAIL;
7376 kvm_rip_write(vcpu, ctxt->eip);
7377 kvm_set_rflags(vcpu, ctxt->eflags);
7378 kvm_make_request(KVM_REQ_EVENT, vcpu);
7379 return EMULATE_DONE;
7381 EXPORT_SYMBOL_GPL(kvm_task_switch);
7383 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7384 struct kvm_sregs *sregs)
7386 struct msr_data apic_base_msr;
7387 int mmu_reset_needed = 0;
7388 int pending_vec, max_bits, idx;
7391 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7394 dt.size = sregs->idt.limit;
7395 dt.address = sregs->idt.base;
7396 kvm_x86_ops->set_idt(vcpu, &dt);
7397 dt.size = sregs->gdt.limit;
7398 dt.address = sregs->gdt.base;
7399 kvm_x86_ops->set_gdt(vcpu, &dt);
7401 vcpu->arch.cr2 = sregs->cr2;
7402 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7403 vcpu->arch.cr3 = sregs->cr3;
7404 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7406 kvm_set_cr8(vcpu, sregs->cr8);
7408 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7409 kvm_x86_ops->set_efer(vcpu, sregs->efer);
7410 apic_base_msr.data = sregs->apic_base;
7411 apic_base_msr.host_initiated = true;
7412 kvm_set_apic_base(vcpu, &apic_base_msr);
7414 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7415 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7416 vcpu->arch.cr0 = sregs->cr0;
7418 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7419 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7420 if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
7421 kvm_update_cpuid(vcpu);
7423 idx = srcu_read_lock(&vcpu->kvm->srcu);
7424 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7425 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7426 mmu_reset_needed = 1;
7428 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7430 if (mmu_reset_needed)
7431 kvm_mmu_reset_context(vcpu);
7433 max_bits = KVM_NR_INTERRUPTS;
7434 pending_vec = find_first_bit(
7435 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7436 if (pending_vec < max_bits) {
7437 kvm_queue_interrupt(vcpu, pending_vec, false);
7438 pr_debug("Set back pending irq %d\n", pending_vec);
7441 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7442 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7443 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7444 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7445 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7446 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7448 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7449 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7451 update_cr8_intercept(vcpu);
7453 /* Older userspace won't unhalt the vcpu on reset. */
7454 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7455 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7457 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7459 kvm_make_request(KVM_REQ_EVENT, vcpu);
7464 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7465 struct kvm_guest_debug *dbg)
7467 unsigned long rflags;
7470 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7472 if (vcpu->arch.exception.pending)
7474 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7475 kvm_queue_exception(vcpu, DB_VECTOR);
7477 kvm_queue_exception(vcpu, BP_VECTOR);
7481 * Read rflags as long as potentially injected trace flags are still
7484 rflags = kvm_get_rflags(vcpu);
7486 vcpu->guest_debug = dbg->control;
7487 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7488 vcpu->guest_debug = 0;
7490 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7491 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7492 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7493 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7495 for (i = 0; i < KVM_NR_DB_REGS; i++)
7496 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7498 kvm_update_dr7(vcpu);
7500 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7501 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7502 get_segment_base(vcpu, VCPU_SREG_CS);
7505 * Trigger an rflags update that will inject or remove the trace
7508 kvm_set_rflags(vcpu, rflags);
7510 kvm_x86_ops->update_bp_intercept(vcpu);
7520 * Translate a guest virtual address to a guest physical address.
7522 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7523 struct kvm_translation *tr)
7525 unsigned long vaddr = tr->linear_address;
7529 idx = srcu_read_lock(&vcpu->kvm->srcu);
7530 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7531 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7532 tr->physical_address = gpa;
7533 tr->valid = gpa != UNMAPPED_GVA;
7540 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7542 struct fxregs_state *fxsave =
7543 &vcpu->arch.guest_fpu.state.fxsave;
7545 memcpy(fpu->fpr, fxsave->st_space, 128);
7546 fpu->fcw = fxsave->cwd;
7547 fpu->fsw = fxsave->swd;
7548 fpu->ftwx = fxsave->twd;
7549 fpu->last_opcode = fxsave->fop;
7550 fpu->last_ip = fxsave->rip;
7551 fpu->last_dp = fxsave->rdp;
7552 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7557 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7559 struct fxregs_state *fxsave =
7560 &vcpu->arch.guest_fpu.state.fxsave;
7562 memcpy(fxsave->st_space, fpu->fpr, 128);
7563 fxsave->cwd = fpu->fcw;
7564 fxsave->swd = fpu->fsw;
7565 fxsave->twd = fpu->ftwx;
7566 fxsave->fop = fpu->last_opcode;
7567 fxsave->rip = fpu->last_ip;
7568 fxsave->rdp = fpu->last_dp;
7569 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7574 static void fx_init(struct kvm_vcpu *vcpu)
7576 fpstate_init(&vcpu->arch.guest_fpu.state);
7577 if (boot_cpu_has(X86_FEATURE_XSAVES))
7578 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7579 host_xcr0 | XSTATE_COMPACTION_ENABLED;
7582 * Ensure guest xcr0 is valid for loading
7584 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7586 vcpu->arch.cr0 |= X86_CR0_ET;
7589 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7591 if (vcpu->guest_fpu_loaded)
7595 * Restore all possible states in the guest,
7596 * and assume host would use all available bits.
7597 * Guest xcr0 would be loaded later.
7599 vcpu->guest_fpu_loaded = 1;
7600 __kernel_fpu_begin();
7601 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7605 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7607 if (!vcpu->guest_fpu_loaded)
7610 vcpu->guest_fpu_loaded = 0;
7611 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7613 ++vcpu->stat.fpu_reload;
7617 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7619 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
7621 kvmclock_reset(vcpu);
7623 kvm_x86_ops->vcpu_free(vcpu);
7624 free_cpumask_var(wbinvd_dirty_mask);
7627 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7630 struct kvm_vcpu *vcpu;
7632 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7633 printk_once(KERN_WARNING
7634 "kvm: SMP vm created on host with unstable TSC; "
7635 "guest TSC will not be reliable\n");
7637 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7642 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7646 kvm_vcpu_mtrr_init(vcpu);
7647 r = vcpu_load(vcpu);
7650 kvm_vcpu_reset(vcpu, false);
7651 kvm_mmu_setup(vcpu);
7656 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7658 struct msr_data msr;
7659 struct kvm *kvm = vcpu->kvm;
7661 if (vcpu_load(vcpu))
7664 msr.index = MSR_IA32_TSC;
7665 msr.host_initiated = true;
7666 kvm_write_tsc(vcpu, &msr);
7669 if (!kvmclock_periodic_sync)
7672 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7673 KVMCLOCK_SYNC_PERIOD);
7676 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7679 vcpu->arch.apf.msr_val = 0;
7681 r = vcpu_load(vcpu);
7683 kvm_mmu_unload(vcpu);
7686 kvm_x86_ops->vcpu_free(vcpu);
7689 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7691 vcpu->arch.hflags = 0;
7693 vcpu->arch.smi_pending = 0;
7694 atomic_set(&vcpu->arch.nmi_queued, 0);
7695 vcpu->arch.nmi_pending = 0;
7696 vcpu->arch.nmi_injected = false;
7697 kvm_clear_interrupt_queue(vcpu);
7698 kvm_clear_exception_queue(vcpu);
7700 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7701 kvm_update_dr0123(vcpu);
7702 vcpu->arch.dr6 = DR6_INIT;
7703 kvm_update_dr6(vcpu);
7704 vcpu->arch.dr7 = DR7_FIXED_1;
7705 kvm_update_dr7(vcpu);
7709 kvm_make_request(KVM_REQ_EVENT, vcpu);
7710 vcpu->arch.apf.msr_val = 0;
7711 vcpu->arch.st.msr_val = 0;
7713 kvmclock_reset(vcpu);
7715 kvm_clear_async_pf_completion_queue(vcpu);
7716 kvm_async_pf_hash_reset(vcpu);
7717 vcpu->arch.apf.halted = false;
7720 kvm_pmu_reset(vcpu);
7721 vcpu->arch.smbase = 0x30000;
7723 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
7724 vcpu->arch.msr_misc_features_enables = 0;
7727 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7728 vcpu->arch.regs_avail = ~0;
7729 vcpu->arch.regs_dirty = ~0;
7731 kvm_x86_ops->vcpu_reset(vcpu, init_event);
7734 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7736 struct kvm_segment cs;
7738 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7739 cs.selector = vector << 8;
7740 cs.base = vector << 12;
7741 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7742 kvm_rip_write(vcpu, 0);
7745 int kvm_arch_hardware_enable(void)
7748 struct kvm_vcpu *vcpu;
7753 bool stable, backwards_tsc = false;
7755 kvm_shared_msr_cpu_online();
7756 ret = kvm_x86_ops->hardware_enable();
7760 local_tsc = rdtsc();
7761 stable = !check_tsc_unstable();
7762 list_for_each_entry(kvm, &vm_list, vm_list) {
7763 kvm_for_each_vcpu(i, vcpu, kvm) {
7764 if (!stable && vcpu->cpu == smp_processor_id())
7765 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7766 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7767 backwards_tsc = true;
7768 if (vcpu->arch.last_host_tsc > max_tsc)
7769 max_tsc = vcpu->arch.last_host_tsc;
7775 * Sometimes, even reliable TSCs go backwards. This happens on
7776 * platforms that reset TSC during suspend or hibernate actions, but
7777 * maintain synchronization. We must compensate. Fortunately, we can
7778 * detect that condition here, which happens early in CPU bringup,
7779 * before any KVM threads can be running. Unfortunately, we can't
7780 * bring the TSCs fully up to date with real time, as we aren't yet far
7781 * enough into CPU bringup that we know how much real time has actually
7782 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
7783 * variables that haven't been updated yet.
7785 * So we simply find the maximum observed TSC above, then record the
7786 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7787 * the adjustment will be applied. Note that we accumulate
7788 * adjustments, in case multiple suspend cycles happen before some VCPU
7789 * gets a chance to run again. In the event that no KVM threads get a
7790 * chance to run, we will miss the entire elapsed period, as we'll have
7791 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7792 * loose cycle time. This isn't too big a deal, since the loss will be
7793 * uniform across all VCPUs (not to mention the scenario is extremely
7794 * unlikely). It is possible that a second hibernate recovery happens
7795 * much faster than a first, causing the observed TSC here to be
7796 * smaller; this would require additional padding adjustment, which is
7797 * why we set last_host_tsc to the local tsc observed here.
7799 * N.B. - this code below runs only on platforms with reliable TSC,
7800 * as that is the only way backwards_tsc is set above. Also note
7801 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7802 * have the same delta_cyc adjustment applied if backwards_tsc
7803 * is detected. Note further, this adjustment is only done once,
7804 * as we reset last_host_tsc on all VCPUs to stop this from being
7805 * called multiple times (one for each physical CPU bringup).
7807 * Platforms with unreliable TSCs don't have to deal with this, they
7808 * will be compensated by the logic in vcpu_load, which sets the TSC to
7809 * catchup mode. This will catchup all VCPUs to real time, but cannot
7810 * guarantee that they stay in perfect synchronization.
7812 if (backwards_tsc) {
7813 u64 delta_cyc = max_tsc - local_tsc;
7814 backwards_tsc_observed = true;
7815 list_for_each_entry(kvm, &vm_list, vm_list) {
7816 kvm_for_each_vcpu(i, vcpu, kvm) {
7817 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7818 vcpu->arch.last_host_tsc = local_tsc;
7819 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7823 * We have to disable TSC offset matching.. if you were
7824 * booting a VM while issuing an S4 host suspend....
7825 * you may have some problem. Solving this issue is
7826 * left as an exercise to the reader.
7828 kvm->arch.last_tsc_nsec = 0;
7829 kvm->arch.last_tsc_write = 0;
7836 void kvm_arch_hardware_disable(void)
7838 kvm_x86_ops->hardware_disable();
7839 drop_user_return_notifiers();
7842 int kvm_arch_hardware_setup(void)
7846 r = kvm_x86_ops->hardware_setup();
7850 if (kvm_has_tsc_control) {
7852 * Make sure the user can only configure tsc_khz values that
7853 * fit into a signed integer.
7854 * A min value is not calculated needed because it will always
7855 * be 1 on all machines.
7857 u64 max = min(0x7fffffffULL,
7858 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
7859 kvm_max_guest_tsc_khz = max;
7861 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
7864 kvm_init_msr_list();
7868 void kvm_arch_hardware_unsetup(void)
7870 kvm_x86_ops->hardware_unsetup();
7873 void kvm_arch_check_processor_compat(void *rtn)
7875 kvm_x86_ops->check_processor_compatibility(rtn);
7878 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7880 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7882 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7884 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7886 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7889 struct static_key kvm_no_apic_vcpu __read_mostly;
7890 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
7892 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7898 BUG_ON(vcpu->kvm == NULL);
7901 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv();
7902 vcpu->arch.pv.pv_unhalted = false;
7903 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7904 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7905 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7907 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7909 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7914 vcpu->arch.pio_data = page_address(page);
7916 kvm_set_tsc_khz(vcpu, max_tsc_khz);
7918 r = kvm_mmu_create(vcpu);
7920 goto fail_free_pio_data;
7922 if (irqchip_in_kernel(kvm)) {
7923 r = kvm_create_lapic(vcpu);
7925 goto fail_mmu_destroy;
7927 static_key_slow_inc(&kvm_no_apic_vcpu);
7929 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7931 if (!vcpu->arch.mce_banks) {
7933 goto fail_free_lapic;
7935 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7937 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7939 goto fail_free_mce_banks;
7944 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7945 vcpu->arch.pv_time_enabled = false;
7947 vcpu->arch.guest_supported_xcr0 = 0;
7948 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7950 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7952 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7954 kvm_async_pf_hash_reset(vcpu);
7957 vcpu->arch.pending_external_vector = -1;
7959 kvm_hv_vcpu_init(vcpu);
7963 fail_free_mce_banks:
7964 kfree(vcpu->arch.mce_banks);
7966 kvm_free_lapic(vcpu);
7968 kvm_mmu_destroy(vcpu);
7970 free_page((unsigned long)vcpu->arch.pio_data);
7975 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7979 kvm_hv_vcpu_uninit(vcpu);
7980 kvm_pmu_destroy(vcpu);
7981 kfree(vcpu->arch.mce_banks);
7982 kvm_free_lapic(vcpu);
7983 idx = srcu_read_lock(&vcpu->kvm->srcu);
7984 kvm_mmu_destroy(vcpu);
7985 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7986 free_page((unsigned long)vcpu->arch.pio_data);
7987 if (!lapic_in_kernel(vcpu))
7988 static_key_slow_dec(&kvm_no_apic_vcpu);
7991 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
7993 kvm_x86_ops->sched_in(vcpu, cpu);
7996 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
8001 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
8002 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
8003 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
8004 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
8005 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
8007 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
8008 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
8009 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
8010 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
8011 &kvm->arch.irq_sources_bitmap);
8013 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
8014 mutex_init(&kvm->arch.apic_map_lock);
8015 mutex_init(&kvm->arch.hyperv.hv_lock);
8016 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
8018 kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
8019 pvclock_update_vm_gtod_copy(kvm);
8021 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
8022 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
8024 kvm_page_track_init(kvm);
8025 kvm_mmu_init_vm(kvm);
8027 if (kvm_x86_ops->vm_init)
8028 return kvm_x86_ops->vm_init(kvm);
8033 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
8036 r = vcpu_load(vcpu);
8038 kvm_mmu_unload(vcpu);
8042 static void kvm_free_vcpus(struct kvm *kvm)
8045 struct kvm_vcpu *vcpu;
8048 * Unpin any mmu pages first.
8050 kvm_for_each_vcpu(i, vcpu, kvm) {
8051 kvm_clear_async_pf_completion_queue(vcpu);
8052 kvm_unload_vcpu_mmu(vcpu);
8054 kvm_for_each_vcpu(i, vcpu, kvm)
8055 kvm_arch_vcpu_free(vcpu);
8057 mutex_lock(&kvm->lock);
8058 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
8059 kvm->vcpus[i] = NULL;
8061 atomic_set(&kvm->online_vcpus, 0);
8062 mutex_unlock(&kvm->lock);
8065 void kvm_arch_sync_events(struct kvm *kvm)
8067 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
8068 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
8072 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8076 struct kvm_memslots *slots = kvm_memslots(kvm);
8077 struct kvm_memory_slot *slot, old;
8079 /* Called with kvm->slots_lock held. */
8080 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
8083 slot = id_to_memslot(slots, id);
8089 * MAP_SHARED to prevent internal slot pages from being moved
8092 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
8093 MAP_SHARED | MAP_ANONYMOUS, 0);
8094 if (IS_ERR((void *)hva))
8095 return PTR_ERR((void *)hva);
8104 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
8105 struct kvm_userspace_memory_region m;
8107 m.slot = id | (i << 16);
8109 m.guest_phys_addr = gpa;
8110 m.userspace_addr = hva;
8111 m.memory_size = size;
8112 r = __kvm_set_memory_region(kvm, &m);
8118 r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
8124 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
8126 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8130 mutex_lock(&kvm->slots_lock);
8131 r = __x86_set_memory_region(kvm, id, gpa, size);
8132 mutex_unlock(&kvm->slots_lock);
8136 EXPORT_SYMBOL_GPL(x86_set_memory_region);
8138 void kvm_arch_destroy_vm(struct kvm *kvm)
8140 if (current->mm == kvm->mm) {
8142 * Free memory regions allocated on behalf of userspace,
8143 * unless the the memory map has changed due to process exit
8146 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
8147 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
8148 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
8150 if (kvm_x86_ops->vm_destroy)
8151 kvm_x86_ops->vm_destroy(kvm);
8152 kvm_pic_destroy(kvm);
8153 kvm_ioapic_destroy(kvm);
8154 kvm_free_vcpus(kvm);
8155 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
8156 kvm_mmu_uninit_vm(kvm);
8157 kvm_page_track_cleanup(kvm);
8160 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
8161 struct kvm_memory_slot *dont)
8165 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8166 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
8167 kvfree(free->arch.rmap[i]);
8168 free->arch.rmap[i] = NULL;
8173 if (!dont || free->arch.lpage_info[i - 1] !=
8174 dont->arch.lpage_info[i - 1]) {
8175 kvfree(free->arch.lpage_info[i - 1]);
8176 free->arch.lpage_info[i - 1] = NULL;
8180 kvm_page_track_free_memslot(free, dont);
8183 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
8184 unsigned long npages)
8188 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8189 struct kvm_lpage_info *linfo;
8194 lpages = gfn_to_index(slot->base_gfn + npages - 1,
8195 slot->base_gfn, level) + 1;
8197 slot->arch.rmap[i] =
8198 kvzalloc(lpages * sizeof(*slot->arch.rmap[i]), GFP_KERNEL);
8199 if (!slot->arch.rmap[i])
8204 linfo = kvzalloc(lpages * sizeof(*linfo), GFP_KERNEL);
8208 slot->arch.lpage_info[i - 1] = linfo;
8210 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
8211 linfo[0].disallow_lpage = 1;
8212 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
8213 linfo[lpages - 1].disallow_lpage = 1;
8214 ugfn = slot->userspace_addr >> PAGE_SHIFT;
8216 * If the gfn and userspace address are not aligned wrt each
8217 * other, or if explicitly asked to, disable large page
8218 * support for this slot
8220 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
8221 !kvm_largepages_enabled()) {
8224 for (j = 0; j < lpages; ++j)
8225 linfo[j].disallow_lpage = 1;
8229 if (kvm_page_track_create_memslot(slot, npages))
8235 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8236 kvfree(slot->arch.rmap[i]);
8237 slot->arch.rmap[i] = NULL;
8241 kvfree(slot->arch.lpage_info[i - 1]);
8242 slot->arch.lpage_info[i - 1] = NULL;
8247 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
8250 * memslots->generation has been incremented.
8251 * mmio generation may have reached its maximum value.
8253 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
8256 int kvm_arch_prepare_memory_region(struct kvm *kvm,
8257 struct kvm_memory_slot *memslot,
8258 const struct kvm_userspace_memory_region *mem,
8259 enum kvm_mr_change change)
8264 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8265 struct kvm_memory_slot *new)
8267 /* Still write protect RO slot */
8268 if (new->flags & KVM_MEM_READONLY) {
8269 kvm_mmu_slot_remove_write_access(kvm, new);
8274 * Call kvm_x86_ops dirty logging hooks when they are valid.
8276 * kvm_x86_ops->slot_disable_log_dirty is called when:
8278 * - KVM_MR_CREATE with dirty logging is disabled
8279 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8281 * The reason is, in case of PML, we need to set D-bit for any slots
8282 * with dirty logging disabled in order to eliminate unnecessary GPA
8283 * logging in PML buffer (and potential PML buffer full VMEXT). This
8284 * guarantees leaving PML enabled during guest's lifetime won't have
8285 * any additonal overhead from PML when guest is running with dirty
8286 * logging disabled for memory slots.
8288 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8289 * to dirty logging mode.
8291 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8293 * In case of write protect:
8295 * Write protect all pages for dirty logging.
8297 * All the sptes including the large sptes which point to this
8298 * slot are set to readonly. We can not create any new large
8299 * spte on this slot until the end of the logging.
8301 * See the comments in fast_page_fault().
8303 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8304 if (kvm_x86_ops->slot_enable_log_dirty)
8305 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8307 kvm_mmu_slot_remove_write_access(kvm, new);
8309 if (kvm_x86_ops->slot_disable_log_dirty)
8310 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8314 void kvm_arch_commit_memory_region(struct kvm *kvm,
8315 const struct kvm_userspace_memory_region *mem,
8316 const struct kvm_memory_slot *old,
8317 const struct kvm_memory_slot *new,
8318 enum kvm_mr_change change)
8320 int nr_mmu_pages = 0;
8322 if (!kvm->arch.n_requested_mmu_pages)
8323 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8326 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
8329 * Dirty logging tracks sptes in 4k granularity, meaning that large
8330 * sptes have to be split. If live migration is successful, the guest
8331 * in the source machine will be destroyed and large sptes will be
8332 * created in the destination. However, if the guest continues to run
8333 * in the source machine (for example if live migration fails), small
8334 * sptes will remain around and cause bad performance.
8336 * Scan sptes if dirty logging has been stopped, dropping those
8337 * which can be collapsed into a single large-page spte. Later
8338 * page faults will create the large-page sptes.
8340 if ((change != KVM_MR_DELETE) &&
8341 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8342 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8343 kvm_mmu_zap_collapsible_sptes(kvm, new);
8346 * Set up write protection and/or dirty logging for the new slot.
8348 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8349 * been zapped so no dirty logging staff is needed for old slot. For
8350 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8351 * new and it's also covered when dealing with the new slot.
8353 * FIXME: const-ify all uses of struct kvm_memory_slot.
8355 if (change != KVM_MR_DELETE)
8356 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8359 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8361 kvm_mmu_invalidate_zap_all_pages(kvm);
8364 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8365 struct kvm_memory_slot *slot)
8367 kvm_page_track_flush_slot(kvm, slot);
8370 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8372 if (!list_empty_careful(&vcpu->async_pf.done))
8375 if (kvm_apic_has_events(vcpu))
8378 if (vcpu->arch.pv.pv_unhalted)
8381 if (atomic_read(&vcpu->arch.nmi_queued))
8384 if (kvm_test_request(KVM_REQ_SMI, vcpu))
8387 if (kvm_arch_interrupt_allowed(vcpu) &&
8388 kvm_cpu_has_interrupt(vcpu))
8391 if (kvm_hv_has_stimer_pending(vcpu))
8397 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8399 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8402 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8404 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8407 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8409 return kvm_x86_ops->interrupt_allowed(vcpu);
8412 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8414 if (is_64_bit_mode(vcpu))
8415 return kvm_rip_read(vcpu);
8416 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8417 kvm_rip_read(vcpu));
8419 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8421 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8423 return kvm_get_linear_rip(vcpu) == linear_rip;
8425 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8427 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8429 unsigned long rflags;
8431 rflags = kvm_x86_ops->get_rflags(vcpu);
8432 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8433 rflags &= ~X86_EFLAGS_TF;
8436 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8438 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8440 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8441 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8442 rflags |= X86_EFLAGS_TF;
8443 kvm_x86_ops->set_rflags(vcpu, rflags);
8446 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8448 __kvm_set_rflags(vcpu, rflags);
8449 kvm_make_request(KVM_REQ_EVENT, vcpu);
8451 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8453 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8457 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8461 r = kvm_mmu_reload(vcpu);
8465 if (!vcpu->arch.mmu.direct_map &&
8466 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8469 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8472 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8474 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8477 static inline u32 kvm_async_pf_next_probe(u32 key)
8479 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8482 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8484 u32 key = kvm_async_pf_hash_fn(gfn);
8486 while (vcpu->arch.apf.gfns[key] != ~0)
8487 key = kvm_async_pf_next_probe(key);
8489 vcpu->arch.apf.gfns[key] = gfn;
8492 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8495 u32 key = kvm_async_pf_hash_fn(gfn);
8497 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8498 (vcpu->arch.apf.gfns[key] != gfn &&
8499 vcpu->arch.apf.gfns[key] != ~0); i++)
8500 key = kvm_async_pf_next_probe(key);
8505 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8507 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8510 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8514 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8516 vcpu->arch.apf.gfns[i] = ~0;
8518 j = kvm_async_pf_next_probe(j);
8519 if (vcpu->arch.apf.gfns[j] == ~0)
8521 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8523 * k lies cyclically in ]i,j]
8525 * |....j i.k.| or |.k..j i...|
8527 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8528 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8533 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8536 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8540 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8541 struct kvm_async_pf *work)
8543 struct x86_exception fault;
8545 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8546 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8548 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8549 (vcpu->arch.apf.send_user_only &&
8550 kvm_x86_ops->get_cpl(vcpu) == 0))
8551 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8552 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8553 fault.vector = PF_VECTOR;
8554 fault.error_code_valid = true;
8555 fault.error_code = 0;
8556 fault.nested_page_fault = false;
8557 fault.address = work->arch.token;
8558 kvm_inject_page_fault(vcpu, &fault);
8562 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8563 struct kvm_async_pf *work)
8565 struct x86_exception fault;
8567 if (work->wakeup_all)
8568 work->arch.token = ~0; /* broadcast wakeup */
8570 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8571 trace_kvm_async_pf_ready(work->arch.token, work->gva);
8573 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8574 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8575 fault.vector = PF_VECTOR;
8576 fault.error_code_valid = true;
8577 fault.error_code = 0;
8578 fault.nested_page_fault = false;
8579 fault.address = work->arch.token;
8580 kvm_inject_page_fault(vcpu, &fault);
8582 vcpu->arch.apf.halted = false;
8583 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8586 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8588 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8591 return !kvm_event_needs_reinjection(vcpu) &&
8592 kvm_x86_ops->interrupt_allowed(vcpu);
8595 void kvm_arch_start_assignment(struct kvm *kvm)
8597 atomic_inc(&kvm->arch.assigned_device_count);
8599 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8601 void kvm_arch_end_assignment(struct kvm *kvm)
8603 atomic_dec(&kvm->arch.assigned_device_count);
8605 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8607 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8609 return atomic_read(&kvm->arch.assigned_device_count);
8611 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8613 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8615 atomic_inc(&kvm->arch.noncoherent_dma_count);
8617 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8619 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8621 atomic_dec(&kvm->arch.noncoherent_dma_count);
8623 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8625 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8627 return atomic_read(&kvm->arch.noncoherent_dma_count);
8629 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8631 bool kvm_arch_has_irq_bypass(void)
8633 return kvm_x86_ops->update_pi_irte != NULL;
8636 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8637 struct irq_bypass_producer *prod)
8639 struct kvm_kernel_irqfd *irqfd =
8640 container_of(cons, struct kvm_kernel_irqfd, consumer);
8642 irqfd->producer = prod;
8644 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8645 prod->irq, irqfd->gsi, 1);
8648 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8649 struct irq_bypass_producer *prod)
8652 struct kvm_kernel_irqfd *irqfd =
8653 container_of(cons, struct kvm_kernel_irqfd, consumer);
8655 WARN_ON(irqfd->producer != prod);
8656 irqfd->producer = NULL;
8659 * When producer of consumer is unregistered, we change back to
8660 * remapped mode, so we can re-use the current implementation
8661 * when the irq is masked/disabled or the consumer side (KVM
8662 * int this case doesn't want to receive the interrupts.
8664 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8666 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8667 " fails: %d\n", irqfd->consumer.token, ret);
8670 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8671 uint32_t guest_irq, bool set)
8673 if (!kvm_x86_ops->update_pi_irte)
8676 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8679 bool kvm_vector_hashing_enabled(void)
8681 return vector_hashing;
8683 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
8685 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8686 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8687 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8688 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8689 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8690 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8691 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8692 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8693 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8694 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8695 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8696 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8697 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8698 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8699 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8700 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8701 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
8702 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
8703 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);