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,
1012 static unsigned num_emulated_msrs;
1014 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1016 if (efer & efer_reserved_bits)
1019 if (efer & EFER_FFXSR) {
1020 struct kvm_cpuid_entry2 *feat;
1022 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1023 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
1027 if (efer & EFER_SVME) {
1028 struct kvm_cpuid_entry2 *feat;
1030 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
1031 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
1037 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1039 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1041 u64 old_efer = vcpu->arch.efer;
1043 if (!kvm_valid_efer(vcpu, efer))
1047 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1051 efer |= vcpu->arch.efer & EFER_LMA;
1053 kvm_x86_ops->set_efer(vcpu, efer);
1055 /* Update reserved bits */
1056 if ((efer ^ old_efer) & EFER_NX)
1057 kvm_mmu_reset_context(vcpu);
1062 void kvm_enable_efer_bits(u64 mask)
1064 efer_reserved_bits &= ~mask;
1066 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1069 * Writes msr value into into the appropriate "register".
1070 * Returns 0 on success, non-0 otherwise.
1071 * Assumes vcpu_load() was already called.
1073 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1075 switch (msr->index) {
1078 case MSR_KERNEL_GS_BASE:
1081 if (is_noncanonical_address(msr->data))
1084 case MSR_IA32_SYSENTER_EIP:
1085 case MSR_IA32_SYSENTER_ESP:
1087 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1088 * non-canonical address is written on Intel but not on
1089 * AMD (which ignores the top 32-bits, because it does
1090 * not implement 64-bit SYSENTER).
1092 * 64-bit code should hence be able to write a non-canonical
1093 * value on AMD. Making the address canonical ensures that
1094 * vmentry does not fail on Intel after writing a non-canonical
1095 * value, and that something deterministic happens if the guest
1096 * invokes 64-bit SYSENTER.
1098 msr->data = get_canonical(msr->data);
1100 return kvm_x86_ops->set_msr(vcpu, msr);
1102 EXPORT_SYMBOL_GPL(kvm_set_msr);
1105 * Adapt set_msr() to msr_io()'s calling convention
1107 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1109 struct msr_data msr;
1113 msr.host_initiated = true;
1114 r = kvm_get_msr(vcpu, &msr);
1122 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1124 struct msr_data msr;
1128 msr.host_initiated = true;
1129 return kvm_set_msr(vcpu, &msr);
1132 #ifdef CONFIG_X86_64
1133 struct pvclock_gtod_data {
1136 struct { /* extract of a clocksource struct */
1149 static struct pvclock_gtod_data pvclock_gtod_data;
1151 static void update_pvclock_gtod(struct timekeeper *tk)
1153 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1156 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1158 write_seqcount_begin(&vdata->seq);
1160 /* copy pvclock gtod data */
1161 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1162 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1163 vdata->clock.mask = tk->tkr_mono.mask;
1164 vdata->clock.mult = tk->tkr_mono.mult;
1165 vdata->clock.shift = tk->tkr_mono.shift;
1167 vdata->boot_ns = boot_ns;
1168 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1170 vdata->wall_time_sec = tk->xtime_sec;
1172 write_seqcount_end(&vdata->seq);
1176 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1179 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1180 * vcpu_enter_guest. This function is only called from
1181 * the physical CPU that is running vcpu.
1183 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1186 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1190 struct pvclock_wall_clock wc;
1191 struct timespec64 boot;
1196 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1201 ++version; /* first time write, random junk */
1205 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1209 * The guest calculates current wall clock time by adding
1210 * system time (updated by kvm_guest_time_update below) to the
1211 * wall clock specified here. guest system time equals host
1212 * system time for us, thus we must fill in host boot time here.
1214 getboottime64(&boot);
1216 if (kvm->arch.kvmclock_offset) {
1217 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1218 boot = timespec64_sub(boot, ts);
1220 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1221 wc.nsec = boot.tv_nsec;
1222 wc.version = version;
1224 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1227 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1230 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1232 do_shl32_div32(dividend, divisor);
1236 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1237 s8 *pshift, u32 *pmultiplier)
1245 scaled64 = scaled_hz;
1246 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1251 tps32 = (uint32_t)tps64;
1252 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1253 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1261 *pmultiplier = div_frac(scaled64, tps32);
1263 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1264 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1267 #ifdef CONFIG_X86_64
1268 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1271 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1272 static unsigned long max_tsc_khz;
1274 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1276 u64 v = (u64)khz * (1000000 + ppm);
1281 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1285 /* Guest TSC same frequency as host TSC? */
1287 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1291 /* TSC scaling supported? */
1292 if (!kvm_has_tsc_control) {
1293 if (user_tsc_khz > tsc_khz) {
1294 vcpu->arch.tsc_catchup = 1;
1295 vcpu->arch.tsc_always_catchup = 1;
1298 WARN(1, "user requested TSC rate below hardware speed\n");
1303 /* TSC scaling required - calculate ratio */
1304 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1305 user_tsc_khz, tsc_khz);
1307 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1308 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1313 vcpu->arch.tsc_scaling_ratio = ratio;
1317 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1319 u32 thresh_lo, thresh_hi;
1320 int use_scaling = 0;
1322 /* tsc_khz can be zero if TSC calibration fails */
1323 if (user_tsc_khz == 0) {
1324 /* set tsc_scaling_ratio to a safe value */
1325 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1329 /* Compute a scale to convert nanoseconds in TSC cycles */
1330 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1331 &vcpu->arch.virtual_tsc_shift,
1332 &vcpu->arch.virtual_tsc_mult);
1333 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1336 * Compute the variation in TSC rate which is acceptable
1337 * within the range of tolerance and decide if the
1338 * rate being applied is within that bounds of the hardware
1339 * rate. If so, no scaling or compensation need be done.
1341 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1342 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1343 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1344 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1347 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1350 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1352 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1353 vcpu->arch.virtual_tsc_mult,
1354 vcpu->arch.virtual_tsc_shift);
1355 tsc += vcpu->arch.this_tsc_write;
1359 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1361 #ifdef CONFIG_X86_64
1363 struct kvm_arch *ka = &vcpu->kvm->arch;
1364 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1366 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1367 atomic_read(&vcpu->kvm->online_vcpus));
1370 * Once the masterclock is enabled, always perform request in
1371 * order to update it.
1373 * In order to enable masterclock, the host clocksource must be TSC
1374 * and the vcpus need to have matched TSCs. When that happens,
1375 * perform request to enable masterclock.
1377 if (ka->use_master_clock ||
1378 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1379 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1381 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1382 atomic_read(&vcpu->kvm->online_vcpus),
1383 ka->use_master_clock, gtod->clock.vclock_mode);
1387 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1389 u64 curr_offset = vcpu->arch.tsc_offset;
1390 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1394 * Multiply tsc by a fixed point number represented by ratio.
1396 * The most significant 64-N bits (mult) of ratio represent the
1397 * integral part of the fixed point number; the remaining N bits
1398 * (frac) represent the fractional part, ie. ratio represents a fixed
1399 * point number (mult + frac * 2^(-N)).
1401 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1403 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1405 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1408 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1411 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1413 if (ratio != kvm_default_tsc_scaling_ratio)
1414 _tsc = __scale_tsc(ratio, tsc);
1418 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1420 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1424 tsc = kvm_scale_tsc(vcpu, rdtsc());
1426 return target_tsc - tsc;
1429 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1431 return vcpu->arch.tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1433 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1435 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1437 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1438 vcpu->arch.tsc_offset = offset;
1441 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1443 struct kvm *kvm = vcpu->kvm;
1444 u64 offset, ns, elapsed;
1445 unsigned long flags;
1448 bool already_matched;
1449 u64 data = msr->data;
1451 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1452 offset = kvm_compute_tsc_offset(vcpu, data);
1453 ns = ktime_get_boot_ns();
1454 elapsed = ns - kvm->arch.last_tsc_nsec;
1456 if (vcpu->arch.virtual_tsc_khz) {
1459 /* n.b - signed multiplication and division required */
1460 usdiff = data - kvm->arch.last_tsc_write;
1461 #ifdef CONFIG_X86_64
1462 usdiff = (usdiff * 1000) / vcpu->arch.virtual_tsc_khz;
1464 /* do_div() only does unsigned */
1465 asm("1: idivl %[divisor]\n"
1466 "2: xor %%edx, %%edx\n"
1467 " movl $0, %[faulted]\n"
1469 ".section .fixup,\"ax\"\n"
1470 "4: movl $1, %[faulted]\n"
1474 _ASM_EXTABLE(1b, 4b)
1476 : "=A"(usdiff), [faulted] "=r" (faulted)
1477 : "A"(usdiff * 1000), [divisor] "rm"(vcpu->arch.virtual_tsc_khz));
1480 do_div(elapsed, 1000);
1485 /* idivl overflow => difference is larger than USEC_PER_SEC */
1487 usdiff = USEC_PER_SEC;
1489 usdiff = USEC_PER_SEC; /* disable TSC match window below */
1492 * Special case: TSC write with a small delta (1 second) of virtual
1493 * cycle time against real time is interpreted as an attempt to
1494 * synchronize the CPU.
1496 * For a reliable TSC, we can match TSC offsets, and for an unstable
1497 * TSC, we add elapsed time in this computation. We could let the
1498 * compensation code attempt to catch up if we fall behind, but
1499 * it's better to try to match offsets from the beginning.
1501 if (usdiff < USEC_PER_SEC &&
1502 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1503 if (!check_tsc_unstable()) {
1504 offset = kvm->arch.cur_tsc_offset;
1505 pr_debug("kvm: matched tsc offset for %llu\n", data);
1507 u64 delta = nsec_to_cycles(vcpu, elapsed);
1509 offset = kvm_compute_tsc_offset(vcpu, data);
1510 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1513 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1516 * We split periods of matched TSC writes into generations.
1517 * For each generation, we track the original measured
1518 * nanosecond time, offset, and write, so if TSCs are in
1519 * sync, we can match exact offset, and if not, we can match
1520 * exact software computation in compute_guest_tsc()
1522 * These values are tracked in kvm->arch.cur_xxx variables.
1524 kvm->arch.cur_tsc_generation++;
1525 kvm->arch.cur_tsc_nsec = ns;
1526 kvm->arch.cur_tsc_write = data;
1527 kvm->arch.cur_tsc_offset = offset;
1529 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1530 kvm->arch.cur_tsc_generation, data);
1534 * We also track th most recent recorded KHZ, write and time to
1535 * allow the matching interval to be extended at each write.
1537 kvm->arch.last_tsc_nsec = ns;
1538 kvm->arch.last_tsc_write = data;
1539 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1541 vcpu->arch.last_guest_tsc = data;
1543 /* Keep track of which generation this VCPU has synchronized to */
1544 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1545 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1546 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1548 if (guest_cpuid_has_tsc_adjust(vcpu) && !msr->host_initiated)
1549 update_ia32_tsc_adjust_msr(vcpu, offset);
1550 kvm_vcpu_write_tsc_offset(vcpu, offset);
1551 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1553 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1555 kvm->arch.nr_vcpus_matched_tsc = 0;
1556 } else if (!already_matched) {
1557 kvm->arch.nr_vcpus_matched_tsc++;
1560 kvm_track_tsc_matching(vcpu);
1561 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1564 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1566 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1569 kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.tsc_offset + adjustment);
1572 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1574 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1575 WARN_ON(adjustment < 0);
1576 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1577 adjust_tsc_offset_guest(vcpu, adjustment);
1580 #ifdef CONFIG_X86_64
1582 static u64 read_tsc(void)
1584 u64 ret = (u64)rdtsc_ordered();
1585 u64 last = pvclock_gtod_data.clock.cycle_last;
1587 if (likely(ret >= last))
1591 * GCC likes to generate cmov here, but this branch is extremely
1592 * predictable (it's just a function of time and the likely is
1593 * very likely) and there's a data dependence, so force GCC
1594 * to generate a branch instead. I don't barrier() because
1595 * we don't actually need a barrier, and if this function
1596 * ever gets inlined it will generate worse code.
1602 static inline u64 vgettsc(u64 *cycle_now)
1605 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1607 *cycle_now = read_tsc();
1609 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1610 return v * gtod->clock.mult;
1613 static int do_monotonic_boot(s64 *t, u64 *cycle_now)
1615 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1621 seq = read_seqcount_begin(>od->seq);
1622 mode = gtod->clock.vclock_mode;
1623 ns = gtod->nsec_base;
1624 ns += vgettsc(cycle_now);
1625 ns >>= gtod->clock.shift;
1626 ns += gtod->boot_ns;
1627 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1633 static int do_realtime(struct timespec *ts, u64 *cycle_now)
1635 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1641 seq = read_seqcount_begin(>od->seq);
1642 mode = gtod->clock.vclock_mode;
1643 ts->tv_sec = gtod->wall_time_sec;
1644 ns = gtod->nsec_base;
1645 ns += vgettsc(cycle_now);
1646 ns >>= gtod->clock.shift;
1647 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1649 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1655 /* returns true if host is using tsc clocksource */
1656 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *cycle_now)
1658 /* checked again under seqlock below */
1659 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1662 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1665 /* returns true if host is using tsc clocksource */
1666 static bool kvm_get_walltime_and_clockread(struct timespec *ts,
1669 /* checked again under seqlock below */
1670 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1673 return do_realtime(ts, cycle_now) == VCLOCK_TSC;
1679 * Assuming a stable TSC across physical CPUS, and a stable TSC
1680 * across virtual CPUs, the following condition is possible.
1681 * Each numbered line represents an event visible to both
1682 * CPUs at the next numbered event.
1684 * "timespecX" represents host monotonic time. "tscX" represents
1687 * VCPU0 on CPU0 | VCPU1 on CPU1
1689 * 1. read timespec0,tsc0
1690 * 2. | timespec1 = timespec0 + N
1692 * 3. transition to guest | transition to guest
1693 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1694 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1695 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1697 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1700 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1702 * - 0 < N - M => M < N
1704 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1705 * always the case (the difference between two distinct xtime instances
1706 * might be smaller then the difference between corresponding TSC reads,
1707 * when updating guest vcpus pvclock areas).
1709 * To avoid that problem, do not allow visibility of distinct
1710 * system_timestamp/tsc_timestamp values simultaneously: use a master
1711 * copy of host monotonic time values. Update that master copy
1714 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1718 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1720 #ifdef CONFIG_X86_64
1721 struct kvm_arch *ka = &kvm->arch;
1723 bool host_tsc_clocksource, vcpus_matched;
1725 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1726 atomic_read(&kvm->online_vcpus));
1729 * If the host uses TSC clock, then passthrough TSC as stable
1732 host_tsc_clocksource = kvm_get_time_and_clockread(
1733 &ka->master_kernel_ns,
1734 &ka->master_cycle_now);
1736 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1737 && !backwards_tsc_observed
1738 && !ka->boot_vcpu_runs_old_kvmclock;
1740 if (ka->use_master_clock)
1741 atomic_set(&kvm_guest_has_master_clock, 1);
1743 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1744 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1749 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1751 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1754 static void kvm_gen_update_masterclock(struct kvm *kvm)
1756 #ifdef CONFIG_X86_64
1758 struct kvm_vcpu *vcpu;
1759 struct kvm_arch *ka = &kvm->arch;
1761 spin_lock(&ka->pvclock_gtod_sync_lock);
1762 kvm_make_mclock_inprogress_request(kvm);
1763 /* no guest entries from this point */
1764 pvclock_update_vm_gtod_copy(kvm);
1766 kvm_for_each_vcpu(i, vcpu, kvm)
1767 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1769 /* guest entries allowed */
1770 kvm_for_each_vcpu(i, vcpu, kvm)
1771 clear_bit(KVM_REQ_MCLOCK_INPROGRESS, &vcpu->requests);
1773 spin_unlock(&ka->pvclock_gtod_sync_lock);
1777 static u64 __get_kvmclock_ns(struct kvm *kvm)
1779 struct kvm_arch *ka = &kvm->arch;
1780 struct pvclock_vcpu_time_info hv_clock;
1782 spin_lock(&ka->pvclock_gtod_sync_lock);
1783 if (!ka->use_master_clock) {
1784 spin_unlock(&ka->pvclock_gtod_sync_lock);
1785 return ktime_get_boot_ns() + ka->kvmclock_offset;
1788 hv_clock.tsc_timestamp = ka->master_cycle_now;
1789 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
1790 spin_unlock(&ka->pvclock_gtod_sync_lock);
1792 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
1793 &hv_clock.tsc_shift,
1794 &hv_clock.tsc_to_system_mul);
1795 return __pvclock_read_cycles(&hv_clock, rdtsc());
1798 u64 get_kvmclock_ns(struct kvm *kvm)
1800 unsigned long flags;
1803 local_irq_save(flags);
1804 ns = __get_kvmclock_ns(kvm);
1805 local_irq_restore(flags);
1810 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
1812 struct kvm_vcpu_arch *vcpu = &v->arch;
1813 struct pvclock_vcpu_time_info guest_hv_clock;
1815 if (unlikely(kvm_vcpu_read_guest_cached(v, &vcpu->pv_time,
1816 &guest_hv_clock, sizeof(guest_hv_clock))))
1819 /* This VCPU is paused, but it's legal for a guest to read another
1820 * VCPU's kvmclock, so we really have to follow the specification where
1821 * it says that version is odd if data is being modified, and even after
1824 * Version field updates must be kept separate. This is because
1825 * kvm_write_guest_cached might use a "rep movs" instruction, and
1826 * writes within a string instruction are weakly ordered. So there
1827 * are three writes overall.
1829 * As a small optimization, only write the version field in the first
1830 * and third write. The vcpu->pv_time cache is still valid, because the
1831 * version field is the first in the struct.
1833 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1835 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1836 kvm_vcpu_write_guest_cached(v, &vcpu->pv_time,
1838 sizeof(vcpu->hv_clock.version));
1842 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1843 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1845 if (vcpu->pvclock_set_guest_stopped_request) {
1846 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
1847 vcpu->pvclock_set_guest_stopped_request = false;
1850 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1852 kvm_vcpu_write_guest_cached(v, &vcpu->pv_time,
1854 sizeof(vcpu->hv_clock));
1858 vcpu->hv_clock.version++;
1859 kvm_vcpu_write_guest_cached(v, &vcpu->pv_time,
1861 sizeof(vcpu->hv_clock.version));
1864 static int kvm_guest_time_update(struct kvm_vcpu *v)
1866 unsigned long flags, tgt_tsc_khz;
1867 struct kvm_vcpu_arch *vcpu = &v->arch;
1868 struct kvm_arch *ka = &v->kvm->arch;
1870 u64 tsc_timestamp, host_tsc;
1872 bool use_master_clock;
1878 * If the host uses TSC clock, then passthrough TSC as stable
1881 spin_lock(&ka->pvclock_gtod_sync_lock);
1882 use_master_clock = ka->use_master_clock;
1883 if (use_master_clock) {
1884 host_tsc = ka->master_cycle_now;
1885 kernel_ns = ka->master_kernel_ns;
1887 spin_unlock(&ka->pvclock_gtod_sync_lock);
1889 /* Keep irq disabled to prevent changes to the clock */
1890 local_irq_save(flags);
1891 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1892 if (unlikely(tgt_tsc_khz == 0)) {
1893 local_irq_restore(flags);
1894 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1897 if (!use_master_clock) {
1899 kernel_ns = ktime_get_boot_ns();
1902 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1905 * We may have to catch up the TSC to match elapsed wall clock
1906 * time for two reasons, even if kvmclock is used.
1907 * 1) CPU could have been running below the maximum TSC rate
1908 * 2) Broken TSC compensation resets the base at each VCPU
1909 * entry to avoid unknown leaps of TSC even when running
1910 * again on the same CPU. This may cause apparent elapsed
1911 * time to disappear, and the guest to stand still or run
1914 if (vcpu->tsc_catchup) {
1915 u64 tsc = compute_guest_tsc(v, kernel_ns);
1916 if (tsc > tsc_timestamp) {
1917 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1918 tsc_timestamp = tsc;
1922 local_irq_restore(flags);
1924 /* With all the info we got, fill in the values */
1926 if (kvm_has_tsc_control)
1927 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
1929 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
1930 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
1931 &vcpu->hv_clock.tsc_shift,
1932 &vcpu->hv_clock.tsc_to_system_mul);
1933 vcpu->hw_tsc_khz = tgt_tsc_khz;
1936 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1937 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1938 vcpu->last_guest_tsc = tsc_timestamp;
1940 /* If the host uses TSC clocksource, then it is stable */
1942 if (use_master_clock)
1943 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
1945 vcpu->hv_clock.flags = pvclock_flags;
1947 if (vcpu->pv_time_enabled)
1948 kvm_setup_pvclock_page(v);
1949 if (v == kvm_get_vcpu(v->kvm, 0))
1950 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
1955 * kvmclock updates which are isolated to a given vcpu, such as
1956 * vcpu->cpu migration, should not allow system_timestamp from
1957 * the rest of the vcpus to remain static. Otherwise ntp frequency
1958 * correction applies to one vcpu's system_timestamp but not
1961 * So in those cases, request a kvmclock update for all vcpus.
1962 * We need to rate-limit these requests though, as they can
1963 * considerably slow guests that have a large number of vcpus.
1964 * The time for a remote vcpu to update its kvmclock is bound
1965 * by the delay we use to rate-limit the updates.
1968 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
1970 static void kvmclock_update_fn(struct work_struct *work)
1973 struct delayed_work *dwork = to_delayed_work(work);
1974 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
1975 kvmclock_update_work);
1976 struct kvm *kvm = container_of(ka, struct kvm, arch);
1977 struct kvm_vcpu *vcpu;
1979 kvm_for_each_vcpu(i, vcpu, kvm) {
1980 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1981 kvm_vcpu_kick(vcpu);
1985 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
1987 struct kvm *kvm = v->kvm;
1989 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1990 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
1991 KVMCLOCK_UPDATE_DELAY);
1994 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
1996 static void kvmclock_sync_fn(struct work_struct *work)
1998 struct delayed_work *dwork = to_delayed_work(work);
1999 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2000 kvmclock_sync_work);
2001 struct kvm *kvm = container_of(ka, struct kvm, arch);
2003 if (!kvmclock_periodic_sync)
2006 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2007 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2008 KVMCLOCK_SYNC_PERIOD);
2011 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
2013 u64 mcg_cap = vcpu->arch.mcg_cap;
2014 unsigned bank_num = mcg_cap & 0xff;
2017 case MSR_IA32_MCG_STATUS:
2018 vcpu->arch.mcg_status = data;
2020 case MSR_IA32_MCG_CTL:
2021 if (!(mcg_cap & MCG_CTL_P))
2023 if (data != 0 && data != ~(u64)0)
2025 vcpu->arch.mcg_ctl = data;
2028 if (msr >= MSR_IA32_MC0_CTL &&
2029 msr < MSR_IA32_MCx_CTL(bank_num)) {
2030 u32 offset = msr - MSR_IA32_MC0_CTL;
2031 /* only 0 or all 1s can be written to IA32_MCi_CTL
2032 * some Linux kernels though clear bit 10 in bank 4 to
2033 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2034 * this to avoid an uncatched #GP in the guest
2036 if ((offset & 0x3) == 0 &&
2037 data != 0 && (data | (1 << 10)) != ~(u64)0)
2039 vcpu->arch.mce_banks[offset] = data;
2047 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2049 struct kvm *kvm = vcpu->kvm;
2050 int lm = is_long_mode(vcpu);
2051 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2052 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2053 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2054 : kvm->arch.xen_hvm_config.blob_size_32;
2055 u32 page_num = data & ~PAGE_MASK;
2056 u64 page_addr = data & PAGE_MASK;
2061 if (page_num >= blob_size)
2064 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2069 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2078 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2080 gpa_t gpa = data & ~0x3f;
2082 /* Bits 2:5 are reserved, Should be zero */
2086 vcpu->arch.apf.msr_val = data;
2088 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2089 kvm_clear_async_pf_completion_queue(vcpu);
2090 kvm_async_pf_hash_reset(vcpu);
2094 if (kvm_vcpu_gfn_to_hva_cache_init(vcpu, &vcpu->arch.apf.data, gpa,
2098 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2099 kvm_async_pf_wakeup_all(vcpu);
2103 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2105 vcpu->arch.pv_time_enabled = false;
2108 static void record_steal_time(struct kvm_vcpu *vcpu)
2110 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2113 if (unlikely(kvm_vcpu_read_guest_cached(vcpu, &vcpu->arch.st.stime,
2114 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2117 vcpu->arch.st.steal.preempted = 0;
2119 if (vcpu->arch.st.steal.version & 1)
2120 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2122 vcpu->arch.st.steal.version += 1;
2124 kvm_vcpu_write_guest_cached(vcpu, &vcpu->arch.st.stime,
2125 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2129 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2130 vcpu->arch.st.last_steal;
2131 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2133 kvm_vcpu_write_guest_cached(vcpu, &vcpu->arch.st.stime,
2134 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2138 vcpu->arch.st.steal.version += 1;
2140 kvm_vcpu_write_guest_cached(vcpu, &vcpu->arch.st.stime,
2141 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2144 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2147 u32 msr = msr_info->index;
2148 u64 data = msr_info->data;
2151 case MSR_AMD64_NB_CFG:
2152 case MSR_IA32_UCODE_REV:
2153 case MSR_IA32_UCODE_WRITE:
2154 case MSR_VM_HSAVE_PA:
2155 case MSR_AMD64_PATCH_LOADER:
2156 case MSR_AMD64_BU_CFG2:
2160 return set_efer(vcpu, data);
2162 data &= ~(u64)0x40; /* ignore flush filter disable */
2163 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2164 data &= ~(u64)0x8; /* ignore TLB cache disable */
2165 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2167 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2172 case MSR_FAM10H_MMIO_CONF_BASE:
2174 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2179 case MSR_IA32_DEBUGCTLMSR:
2181 /* We support the non-activated case already */
2183 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2184 /* Values other than LBR and BTF are vendor-specific,
2185 thus reserved and should throw a #GP */
2188 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2191 case 0x200 ... 0x2ff:
2192 return kvm_mtrr_set_msr(vcpu, msr, data);
2193 case MSR_IA32_APICBASE:
2194 return kvm_set_apic_base(vcpu, msr_info);
2195 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2196 return kvm_x2apic_msr_write(vcpu, msr, data);
2197 case MSR_IA32_TSCDEADLINE:
2198 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2200 case MSR_IA32_TSC_ADJUST:
2201 if (guest_cpuid_has_tsc_adjust(vcpu)) {
2202 if (!msr_info->host_initiated) {
2203 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2204 adjust_tsc_offset_guest(vcpu, adj);
2206 vcpu->arch.ia32_tsc_adjust_msr = data;
2209 case MSR_IA32_MISC_ENABLE:
2210 vcpu->arch.ia32_misc_enable_msr = data;
2212 case MSR_IA32_SMBASE:
2213 if (!msr_info->host_initiated)
2215 vcpu->arch.smbase = data;
2217 case MSR_KVM_WALL_CLOCK_NEW:
2218 case MSR_KVM_WALL_CLOCK:
2219 vcpu->kvm->arch.wall_clock = data;
2220 kvm_write_wall_clock(vcpu->kvm, data);
2222 case MSR_KVM_SYSTEM_TIME_NEW:
2223 case MSR_KVM_SYSTEM_TIME: {
2224 struct kvm_arch *ka = &vcpu->kvm->arch;
2226 kvmclock_reset(vcpu);
2228 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2229 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2231 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2232 set_bit(KVM_REQ_MASTERCLOCK_UPDATE,
2235 ka->boot_vcpu_runs_old_kvmclock = tmp;
2238 vcpu->arch.time = data;
2239 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2241 /* we verify if the enable bit is set... */
2245 if (kvm_vcpu_gfn_to_hva_cache_init(vcpu,
2246 &vcpu->arch.pv_time, data & ~1ULL,
2247 sizeof(struct pvclock_vcpu_time_info)))
2248 vcpu->arch.pv_time_enabled = false;
2250 vcpu->arch.pv_time_enabled = true;
2254 case MSR_KVM_ASYNC_PF_EN:
2255 if (kvm_pv_enable_async_pf(vcpu, data))
2258 case MSR_KVM_STEAL_TIME:
2260 if (unlikely(!sched_info_on()))
2263 if (data & KVM_STEAL_RESERVED_MASK)
2266 if (kvm_vcpu_gfn_to_hva_cache_init(vcpu, &vcpu->arch.st.stime,
2267 data & KVM_STEAL_VALID_BITS,
2268 sizeof(struct kvm_steal_time)))
2271 vcpu->arch.st.msr_val = data;
2273 if (!(data & KVM_MSR_ENABLED))
2276 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2279 case MSR_KVM_PV_EOI_EN:
2280 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2284 case MSR_IA32_MCG_CTL:
2285 case MSR_IA32_MCG_STATUS:
2286 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2287 return set_msr_mce(vcpu, msr, data);
2289 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2290 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2291 pr = true; /* fall through */
2292 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2293 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2294 if (kvm_pmu_is_valid_msr(vcpu, msr))
2295 return kvm_pmu_set_msr(vcpu, msr_info);
2297 if (pr || data != 0)
2298 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2299 "0x%x data 0x%llx\n", msr, data);
2301 case MSR_K7_CLK_CTL:
2303 * Ignore all writes to this no longer documented MSR.
2304 * Writes are only relevant for old K7 processors,
2305 * all pre-dating SVM, but a recommended workaround from
2306 * AMD for these chips. It is possible to specify the
2307 * affected processor models on the command line, hence
2308 * the need to ignore the workaround.
2311 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2312 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2313 case HV_X64_MSR_CRASH_CTL:
2314 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2315 return kvm_hv_set_msr_common(vcpu, msr, data,
2316 msr_info->host_initiated);
2317 case MSR_IA32_BBL_CR_CTL3:
2318 /* Drop writes to this legacy MSR -- see rdmsr
2319 * counterpart for further detail.
2321 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n", msr, data);
2323 case MSR_AMD64_OSVW_ID_LENGTH:
2324 if (!guest_cpuid_has_osvw(vcpu))
2326 vcpu->arch.osvw.length = data;
2328 case MSR_AMD64_OSVW_STATUS:
2329 if (!guest_cpuid_has_osvw(vcpu))
2331 vcpu->arch.osvw.status = data;
2334 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2335 return xen_hvm_config(vcpu, data);
2336 if (kvm_pmu_is_valid_msr(vcpu, msr))
2337 return kvm_pmu_set_msr(vcpu, msr_info);
2339 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2343 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2350 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2354 * Reads an msr value (of 'msr_index') into 'pdata'.
2355 * Returns 0 on success, non-0 otherwise.
2356 * Assumes vcpu_load() was already called.
2358 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2360 return kvm_x86_ops->get_msr(vcpu, msr);
2362 EXPORT_SYMBOL_GPL(kvm_get_msr);
2364 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2367 u64 mcg_cap = vcpu->arch.mcg_cap;
2368 unsigned bank_num = mcg_cap & 0xff;
2371 case MSR_IA32_P5_MC_ADDR:
2372 case MSR_IA32_P5_MC_TYPE:
2375 case MSR_IA32_MCG_CAP:
2376 data = vcpu->arch.mcg_cap;
2378 case MSR_IA32_MCG_CTL:
2379 if (!(mcg_cap & MCG_CTL_P))
2381 data = vcpu->arch.mcg_ctl;
2383 case MSR_IA32_MCG_STATUS:
2384 data = vcpu->arch.mcg_status;
2387 if (msr >= MSR_IA32_MC0_CTL &&
2388 msr < MSR_IA32_MCx_CTL(bank_num)) {
2389 u32 offset = msr - MSR_IA32_MC0_CTL;
2390 data = vcpu->arch.mce_banks[offset];
2399 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2401 switch (msr_info->index) {
2402 case MSR_IA32_PLATFORM_ID:
2403 case MSR_IA32_EBL_CR_POWERON:
2404 case MSR_IA32_DEBUGCTLMSR:
2405 case MSR_IA32_LASTBRANCHFROMIP:
2406 case MSR_IA32_LASTBRANCHTOIP:
2407 case MSR_IA32_LASTINTFROMIP:
2408 case MSR_IA32_LASTINTTOIP:
2410 case MSR_K8_TSEG_ADDR:
2411 case MSR_K8_TSEG_MASK:
2413 case MSR_VM_HSAVE_PA:
2414 case MSR_K8_INT_PENDING_MSG:
2415 case MSR_AMD64_NB_CFG:
2416 case MSR_FAM10H_MMIO_CONF_BASE:
2417 case MSR_AMD64_BU_CFG2:
2418 case MSR_IA32_PERF_CTL:
2421 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2422 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2423 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2424 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2425 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2426 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2429 case MSR_IA32_UCODE_REV:
2430 msr_info->data = 0x100000000ULL;
2433 case 0x200 ... 0x2ff:
2434 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2435 case 0xcd: /* fsb frequency */
2439 * MSR_EBC_FREQUENCY_ID
2440 * Conservative value valid for even the basic CPU models.
2441 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2442 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2443 * and 266MHz for model 3, or 4. Set Core Clock
2444 * Frequency to System Bus Frequency Ratio to 1 (bits
2445 * 31:24) even though these are only valid for CPU
2446 * models > 2, however guests may end up dividing or
2447 * multiplying by zero otherwise.
2449 case MSR_EBC_FREQUENCY_ID:
2450 msr_info->data = 1 << 24;
2452 case MSR_IA32_APICBASE:
2453 msr_info->data = kvm_get_apic_base(vcpu);
2455 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2456 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2458 case MSR_IA32_TSCDEADLINE:
2459 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2461 case MSR_IA32_TSC_ADJUST:
2462 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2464 case MSR_IA32_MISC_ENABLE:
2465 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2467 case MSR_IA32_SMBASE:
2468 if (!msr_info->host_initiated)
2470 msr_info->data = vcpu->arch.smbase;
2472 case MSR_IA32_PERF_STATUS:
2473 /* TSC increment by tick */
2474 msr_info->data = 1000ULL;
2475 /* CPU multiplier */
2476 msr_info->data |= (((uint64_t)4ULL) << 40);
2479 msr_info->data = vcpu->arch.efer;
2481 case MSR_KVM_WALL_CLOCK:
2482 case MSR_KVM_WALL_CLOCK_NEW:
2483 msr_info->data = vcpu->kvm->arch.wall_clock;
2485 case MSR_KVM_SYSTEM_TIME:
2486 case MSR_KVM_SYSTEM_TIME_NEW:
2487 msr_info->data = vcpu->arch.time;
2489 case MSR_KVM_ASYNC_PF_EN:
2490 msr_info->data = vcpu->arch.apf.msr_val;
2492 case MSR_KVM_STEAL_TIME:
2493 msr_info->data = vcpu->arch.st.msr_val;
2495 case MSR_KVM_PV_EOI_EN:
2496 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2498 case MSR_IA32_P5_MC_ADDR:
2499 case MSR_IA32_P5_MC_TYPE:
2500 case MSR_IA32_MCG_CAP:
2501 case MSR_IA32_MCG_CTL:
2502 case MSR_IA32_MCG_STATUS:
2503 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2504 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2505 case MSR_K7_CLK_CTL:
2507 * Provide expected ramp-up count for K7. All other
2508 * are set to zero, indicating minimum divisors for
2511 * This prevents guest kernels on AMD host with CPU
2512 * type 6, model 8 and higher from exploding due to
2513 * the rdmsr failing.
2515 msr_info->data = 0x20000000;
2517 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2518 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2519 case HV_X64_MSR_CRASH_CTL:
2520 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2521 return kvm_hv_get_msr_common(vcpu,
2522 msr_info->index, &msr_info->data);
2524 case MSR_IA32_BBL_CR_CTL3:
2525 /* This legacy MSR exists but isn't fully documented in current
2526 * silicon. It is however accessed by winxp in very narrow
2527 * scenarios where it sets bit #19, itself documented as
2528 * a "reserved" bit. Best effort attempt to source coherent
2529 * read data here should the balance of the register be
2530 * interpreted by the guest:
2532 * L2 cache control register 3: 64GB range, 256KB size,
2533 * enabled, latency 0x1, configured
2535 msr_info->data = 0xbe702111;
2537 case MSR_AMD64_OSVW_ID_LENGTH:
2538 if (!guest_cpuid_has_osvw(vcpu))
2540 msr_info->data = vcpu->arch.osvw.length;
2542 case MSR_AMD64_OSVW_STATUS:
2543 if (!guest_cpuid_has_osvw(vcpu))
2545 msr_info->data = vcpu->arch.osvw.status;
2548 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2549 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2551 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2555 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr_info->index);
2562 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2565 * Read or write a bunch of msrs. All parameters are kernel addresses.
2567 * @return number of msrs set successfully.
2569 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2570 struct kvm_msr_entry *entries,
2571 int (*do_msr)(struct kvm_vcpu *vcpu,
2572 unsigned index, u64 *data))
2576 idx = srcu_read_lock(&vcpu->kvm->srcu);
2577 for (i = 0; i < msrs->nmsrs; ++i)
2578 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2580 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2586 * Read or write a bunch of msrs. Parameters are user addresses.
2588 * @return number of msrs set successfully.
2590 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2591 int (*do_msr)(struct kvm_vcpu *vcpu,
2592 unsigned index, u64 *data),
2595 struct kvm_msrs msrs;
2596 struct kvm_msr_entry *entries;
2601 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2605 if (msrs.nmsrs >= MAX_IO_MSRS)
2608 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2609 entries = memdup_user(user_msrs->entries, size);
2610 if (IS_ERR(entries)) {
2611 r = PTR_ERR(entries);
2615 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2620 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2631 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2636 case KVM_CAP_IRQCHIP:
2638 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2639 case KVM_CAP_SET_TSS_ADDR:
2640 case KVM_CAP_EXT_CPUID:
2641 case KVM_CAP_EXT_EMUL_CPUID:
2642 case KVM_CAP_CLOCKSOURCE:
2644 case KVM_CAP_NOP_IO_DELAY:
2645 case KVM_CAP_MP_STATE:
2646 case KVM_CAP_SYNC_MMU:
2647 case KVM_CAP_USER_NMI:
2648 case KVM_CAP_REINJECT_CONTROL:
2649 case KVM_CAP_IRQ_INJECT_STATUS:
2650 case KVM_CAP_IOEVENTFD:
2651 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2653 case KVM_CAP_PIT_STATE2:
2654 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2655 case KVM_CAP_XEN_HVM:
2656 case KVM_CAP_VCPU_EVENTS:
2657 case KVM_CAP_HYPERV:
2658 case KVM_CAP_HYPERV_VAPIC:
2659 case KVM_CAP_HYPERV_SPIN:
2660 case KVM_CAP_HYPERV_SYNIC:
2661 case KVM_CAP_PCI_SEGMENT:
2662 case KVM_CAP_DEBUGREGS:
2663 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2665 case KVM_CAP_ASYNC_PF:
2666 case KVM_CAP_GET_TSC_KHZ:
2667 case KVM_CAP_KVMCLOCK_CTRL:
2668 case KVM_CAP_READONLY_MEM:
2669 case KVM_CAP_HYPERV_TIME:
2670 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2671 case KVM_CAP_TSC_DEADLINE_TIMER:
2672 case KVM_CAP_ENABLE_CAP_VM:
2673 case KVM_CAP_DISABLE_QUIRKS:
2674 case KVM_CAP_SET_BOOT_CPU_ID:
2675 case KVM_CAP_SPLIT_IRQCHIP:
2676 case KVM_CAP_IMMEDIATE_EXIT:
2679 case KVM_CAP_ADJUST_CLOCK:
2680 r = KVM_CLOCK_TSC_STABLE;
2682 case KVM_CAP_X86_SMM:
2683 /* SMBASE is usually relocated above 1M on modern chipsets,
2684 * and SMM handlers might indeed rely on 4G segment limits,
2685 * so do not report SMM to be available if real mode is
2686 * emulated via vm86 mode. Still, do not go to great lengths
2687 * to avoid userspace's usage of the feature, because it is a
2688 * fringe case that is not enabled except via specific settings
2689 * of the module parameters.
2691 r = kvm_x86_ops->cpu_has_high_real_mode_segbase();
2694 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2696 case KVM_CAP_NR_VCPUS:
2697 r = KVM_SOFT_MAX_VCPUS;
2699 case KVM_CAP_MAX_VCPUS:
2702 case KVM_CAP_NR_MEMSLOTS:
2703 r = KVM_USER_MEM_SLOTS;
2705 case KVM_CAP_PV_MMU: /* obsolete */
2709 r = KVM_MAX_MCE_BANKS;
2712 r = boot_cpu_has(X86_FEATURE_XSAVE);
2714 case KVM_CAP_TSC_CONTROL:
2715 r = kvm_has_tsc_control;
2717 case KVM_CAP_X2APIC_API:
2718 r = KVM_X2APIC_API_VALID_FLAGS;
2728 long kvm_arch_dev_ioctl(struct file *filp,
2729 unsigned int ioctl, unsigned long arg)
2731 void __user *argp = (void __user *)arg;
2735 case KVM_GET_MSR_INDEX_LIST: {
2736 struct kvm_msr_list __user *user_msr_list = argp;
2737 struct kvm_msr_list msr_list;
2741 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2744 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2745 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2748 if (n < msr_list.nmsrs)
2751 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2752 num_msrs_to_save * sizeof(u32)))
2754 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2756 num_emulated_msrs * sizeof(u32)))
2761 case KVM_GET_SUPPORTED_CPUID:
2762 case KVM_GET_EMULATED_CPUID: {
2763 struct kvm_cpuid2 __user *cpuid_arg = argp;
2764 struct kvm_cpuid2 cpuid;
2767 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2770 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2776 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2781 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2783 if (copy_to_user(argp, &kvm_mce_cap_supported,
2784 sizeof(kvm_mce_cap_supported)))
2796 static void wbinvd_ipi(void *garbage)
2801 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2803 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2806 static inline void kvm_migrate_timers(struct kvm_vcpu *vcpu)
2808 set_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests);
2811 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2813 /* Address WBINVD may be executed by guest */
2814 if (need_emulate_wbinvd(vcpu)) {
2815 if (kvm_x86_ops->has_wbinvd_exit())
2816 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2817 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2818 smp_call_function_single(vcpu->cpu,
2819 wbinvd_ipi, NULL, 1);
2822 kvm_x86_ops->vcpu_load(vcpu, cpu);
2824 /* Apply any externally detected TSC adjustments (due to suspend) */
2825 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2826 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2827 vcpu->arch.tsc_offset_adjustment = 0;
2828 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2831 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2832 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2833 rdtsc() - vcpu->arch.last_host_tsc;
2835 mark_tsc_unstable("KVM discovered backwards TSC");
2837 if (check_tsc_unstable()) {
2838 u64 offset = kvm_compute_tsc_offset(vcpu,
2839 vcpu->arch.last_guest_tsc);
2840 kvm_vcpu_write_tsc_offset(vcpu, offset);
2841 vcpu->arch.tsc_catchup = 1;
2843 if (kvm_lapic_hv_timer_in_use(vcpu) &&
2844 kvm_x86_ops->set_hv_timer(vcpu,
2845 kvm_get_lapic_target_expiration_tsc(vcpu)))
2846 kvm_lapic_switch_to_sw_timer(vcpu);
2848 * On a host with synchronized TSC, there is no need to update
2849 * kvmclock on vcpu->cpu migration
2851 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2852 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2853 if (vcpu->cpu != cpu)
2854 kvm_migrate_timers(vcpu);
2858 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2861 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
2863 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2866 vcpu->arch.st.steal.preempted = 1;
2868 kvm_vcpu_write_guest_offset_cached(vcpu, &vcpu->arch.st.stime,
2869 &vcpu->arch.st.steal.preempted,
2870 offsetof(struct kvm_steal_time, preempted),
2871 sizeof(vcpu->arch.st.steal.preempted));
2874 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2878 * Disable page faults because we're in atomic context here.
2879 * kvm_write_guest_offset_cached() would call might_fault()
2880 * that relies on pagefault_disable() to tell if there's a
2881 * bug. NOTE: the write to guest memory may not go through if
2882 * during postcopy live migration or if there's heavy guest
2885 pagefault_disable();
2887 * kvm_memslots() will be called by
2888 * kvm_write_guest_offset_cached() so take the srcu lock.
2890 idx = srcu_read_lock(&vcpu->kvm->srcu);
2891 kvm_steal_time_set_preempted(vcpu);
2892 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2894 kvm_x86_ops->vcpu_put(vcpu);
2895 kvm_put_guest_fpu(vcpu);
2896 vcpu->arch.last_host_tsc = rdtsc();
2899 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2900 struct kvm_lapic_state *s)
2902 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
2903 kvm_x86_ops->sync_pir_to_irr(vcpu);
2905 return kvm_apic_get_state(vcpu, s);
2908 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2909 struct kvm_lapic_state *s)
2913 r = kvm_apic_set_state(vcpu, s);
2916 update_cr8_intercept(vcpu);
2921 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
2923 return (!lapic_in_kernel(vcpu) ||
2924 kvm_apic_accept_pic_intr(vcpu));
2928 * if userspace requested an interrupt window, check that the
2929 * interrupt window is open.
2931 * No need to exit to userspace if we already have an interrupt queued.
2933 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
2935 return kvm_arch_interrupt_allowed(vcpu) &&
2936 !kvm_cpu_has_interrupt(vcpu) &&
2937 !kvm_event_needs_reinjection(vcpu) &&
2938 kvm_cpu_accept_dm_intr(vcpu);
2941 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2942 struct kvm_interrupt *irq)
2944 if (irq->irq >= KVM_NR_INTERRUPTS)
2947 if (!irqchip_in_kernel(vcpu->kvm)) {
2948 kvm_queue_interrupt(vcpu, irq->irq, false);
2949 kvm_make_request(KVM_REQ_EVENT, vcpu);
2954 * With in-kernel LAPIC, we only use this to inject EXTINT, so
2955 * fail for in-kernel 8259.
2957 if (pic_in_kernel(vcpu->kvm))
2960 if (vcpu->arch.pending_external_vector != -1)
2963 vcpu->arch.pending_external_vector = irq->irq;
2964 kvm_make_request(KVM_REQ_EVENT, vcpu);
2968 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2970 kvm_inject_nmi(vcpu);
2975 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
2977 kvm_make_request(KVM_REQ_SMI, vcpu);
2982 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2983 struct kvm_tpr_access_ctl *tac)
2987 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2991 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2995 unsigned bank_num = mcg_cap & 0xff, bank;
2998 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3000 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3003 vcpu->arch.mcg_cap = mcg_cap;
3004 /* Init IA32_MCG_CTL to all 1s */
3005 if (mcg_cap & MCG_CTL_P)
3006 vcpu->arch.mcg_ctl = ~(u64)0;
3007 /* Init IA32_MCi_CTL to all 1s */
3008 for (bank = 0; bank < bank_num; bank++)
3009 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3011 if (kvm_x86_ops->setup_mce)
3012 kvm_x86_ops->setup_mce(vcpu);
3017 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3018 struct kvm_x86_mce *mce)
3020 u64 mcg_cap = vcpu->arch.mcg_cap;
3021 unsigned bank_num = mcg_cap & 0xff;
3022 u64 *banks = vcpu->arch.mce_banks;
3024 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3027 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3028 * reporting is disabled
3030 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3031 vcpu->arch.mcg_ctl != ~(u64)0)
3033 banks += 4 * mce->bank;
3035 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3036 * reporting is disabled for the bank
3038 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3040 if (mce->status & MCI_STATUS_UC) {
3041 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3042 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3043 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3046 if (banks[1] & MCI_STATUS_VAL)
3047 mce->status |= MCI_STATUS_OVER;
3048 banks[2] = mce->addr;
3049 banks[3] = mce->misc;
3050 vcpu->arch.mcg_status = mce->mcg_status;
3051 banks[1] = mce->status;
3052 kvm_queue_exception(vcpu, MC_VECTOR);
3053 } else if (!(banks[1] & MCI_STATUS_VAL)
3054 || !(banks[1] & MCI_STATUS_UC)) {
3055 if (banks[1] & MCI_STATUS_VAL)
3056 mce->status |= MCI_STATUS_OVER;
3057 banks[2] = mce->addr;
3058 banks[3] = mce->misc;
3059 banks[1] = mce->status;
3061 banks[1] |= MCI_STATUS_OVER;
3065 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3066 struct kvm_vcpu_events *events)
3069 events->exception.injected =
3070 vcpu->arch.exception.pending &&
3071 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3072 events->exception.nr = vcpu->arch.exception.nr;
3073 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3074 events->exception.pad = 0;
3075 events->exception.error_code = vcpu->arch.exception.error_code;
3077 events->interrupt.injected =
3078 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3079 events->interrupt.nr = vcpu->arch.interrupt.nr;
3080 events->interrupt.soft = 0;
3081 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3083 events->nmi.injected = vcpu->arch.nmi_injected;
3084 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3085 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3086 events->nmi.pad = 0;
3088 events->sipi_vector = 0; /* never valid when reporting to user space */
3090 events->smi.smm = is_smm(vcpu);
3091 events->smi.pending = vcpu->arch.smi_pending;
3092 events->smi.smm_inside_nmi =
3093 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3094 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3096 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3097 | KVM_VCPUEVENT_VALID_SHADOW
3098 | KVM_VCPUEVENT_VALID_SMM);
3099 memset(&events->reserved, 0, sizeof(events->reserved));
3102 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3104 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3105 struct kvm_vcpu_events *events)
3107 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3108 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3109 | KVM_VCPUEVENT_VALID_SHADOW
3110 | KVM_VCPUEVENT_VALID_SMM))
3113 if (events->exception.injected &&
3114 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR ||
3115 is_guest_mode(vcpu)))
3118 /* INITs are latched while in SMM */
3119 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3120 (events->smi.smm || events->smi.pending) &&
3121 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3125 vcpu->arch.exception.pending = events->exception.injected;
3126 vcpu->arch.exception.nr = events->exception.nr;
3127 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3128 vcpu->arch.exception.error_code = events->exception.error_code;
3130 vcpu->arch.interrupt.pending = events->interrupt.injected;
3131 vcpu->arch.interrupt.nr = events->interrupt.nr;
3132 vcpu->arch.interrupt.soft = events->interrupt.soft;
3133 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3134 kvm_x86_ops->set_interrupt_shadow(vcpu,
3135 events->interrupt.shadow);
3137 vcpu->arch.nmi_injected = events->nmi.injected;
3138 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3139 vcpu->arch.nmi_pending = events->nmi.pending;
3140 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3142 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3143 lapic_in_kernel(vcpu))
3144 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3146 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3147 u32 hflags = vcpu->arch.hflags;
3148 if (events->smi.smm)
3149 hflags |= HF_SMM_MASK;
3151 hflags &= ~HF_SMM_MASK;
3152 kvm_set_hflags(vcpu, hflags);
3154 vcpu->arch.smi_pending = events->smi.pending;
3155 if (events->smi.smm_inside_nmi)
3156 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3158 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3159 if (lapic_in_kernel(vcpu)) {
3160 if (events->smi.latched_init)
3161 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3163 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3167 kvm_make_request(KVM_REQ_EVENT, vcpu);
3172 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3173 struct kvm_debugregs *dbgregs)
3177 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3178 kvm_get_dr(vcpu, 6, &val);
3180 dbgregs->dr7 = vcpu->arch.dr7;
3182 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3185 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3186 struct kvm_debugregs *dbgregs)
3191 if (dbgregs->dr6 & ~0xffffffffull)
3193 if (dbgregs->dr7 & ~0xffffffffull)
3196 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3197 kvm_update_dr0123(vcpu);
3198 vcpu->arch.dr6 = dbgregs->dr6;
3199 kvm_update_dr6(vcpu);
3200 vcpu->arch.dr7 = dbgregs->dr7;
3201 kvm_update_dr7(vcpu);
3206 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3208 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3210 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3211 u64 xstate_bv = xsave->header.xfeatures;
3215 * Copy legacy XSAVE area, to avoid complications with CPUID
3216 * leaves 0 and 1 in the loop below.
3218 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3221 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3222 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3225 * Copy each region from the possibly compacted offset to the
3226 * non-compacted offset.
3228 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3230 u64 feature = valid & -valid;
3231 int index = fls64(feature) - 1;
3232 void *src = get_xsave_addr(xsave, feature);
3235 u32 size, offset, ecx, edx;
3236 cpuid_count(XSTATE_CPUID, index,
3237 &size, &offset, &ecx, &edx);
3238 memcpy(dest + offset, src, size);
3245 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3247 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3248 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3252 * Copy legacy XSAVE area, to avoid complications with CPUID
3253 * leaves 0 and 1 in the loop below.
3255 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3257 /* Set XSTATE_BV and possibly XCOMP_BV. */
3258 xsave->header.xfeatures = xstate_bv;
3259 if (boot_cpu_has(X86_FEATURE_XSAVES))
3260 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3263 * Copy each region from the non-compacted offset to the
3264 * possibly compacted offset.
3266 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3268 u64 feature = valid & -valid;
3269 int index = fls64(feature) - 1;
3270 void *dest = get_xsave_addr(xsave, feature);
3273 u32 size, offset, ecx, edx;
3274 cpuid_count(XSTATE_CPUID, index,
3275 &size, &offset, &ecx, &edx);
3276 memcpy(dest, src + offset, size);
3283 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3284 struct kvm_xsave *guest_xsave)
3286 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3287 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3288 fill_xsave((u8 *) guest_xsave->region, vcpu);
3290 memcpy(guest_xsave->region,
3291 &vcpu->arch.guest_fpu.state.fxsave,
3292 sizeof(struct fxregs_state));
3293 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3294 XFEATURE_MASK_FPSSE;
3298 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3299 struct kvm_xsave *guest_xsave)
3302 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3304 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3306 * Here we allow setting states that are not present in
3307 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3308 * with old userspace.
3310 if (xstate_bv & ~kvm_supported_xcr0())
3312 load_xsave(vcpu, (u8 *)guest_xsave->region);
3314 if (xstate_bv & ~XFEATURE_MASK_FPSSE)
3316 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3317 guest_xsave->region, sizeof(struct fxregs_state));
3322 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3323 struct kvm_xcrs *guest_xcrs)
3325 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3326 guest_xcrs->nr_xcrs = 0;
3330 guest_xcrs->nr_xcrs = 1;
3331 guest_xcrs->flags = 0;
3332 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3333 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3336 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3337 struct kvm_xcrs *guest_xcrs)
3341 if (!boot_cpu_has(X86_FEATURE_XSAVE))
3344 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3347 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3348 /* Only support XCR0 currently */
3349 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3350 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3351 guest_xcrs->xcrs[i].value);
3360 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3361 * stopped by the hypervisor. This function will be called from the host only.
3362 * EINVAL is returned when the host attempts to set the flag for a guest that
3363 * does not support pv clocks.
3365 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3367 if (!vcpu->arch.pv_time_enabled)
3369 vcpu->arch.pvclock_set_guest_stopped_request = true;
3370 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3374 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3375 struct kvm_enable_cap *cap)
3381 case KVM_CAP_HYPERV_SYNIC:
3382 if (!irqchip_in_kernel(vcpu->kvm))
3384 return kvm_hv_activate_synic(vcpu);
3390 long kvm_arch_vcpu_ioctl(struct file *filp,
3391 unsigned int ioctl, unsigned long arg)
3393 struct kvm_vcpu *vcpu = filp->private_data;
3394 void __user *argp = (void __user *)arg;
3397 struct kvm_lapic_state *lapic;
3398 struct kvm_xsave *xsave;
3399 struct kvm_xcrs *xcrs;
3405 case KVM_GET_LAPIC: {
3407 if (!lapic_in_kernel(vcpu))
3409 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3414 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3418 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3423 case KVM_SET_LAPIC: {
3425 if (!lapic_in_kernel(vcpu))
3427 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3428 if (IS_ERR(u.lapic))
3429 return PTR_ERR(u.lapic);
3431 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3434 case KVM_INTERRUPT: {
3435 struct kvm_interrupt irq;
3438 if (copy_from_user(&irq, argp, sizeof irq))
3440 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3444 r = kvm_vcpu_ioctl_nmi(vcpu);
3448 r = kvm_vcpu_ioctl_smi(vcpu);
3451 case KVM_SET_CPUID: {
3452 struct kvm_cpuid __user *cpuid_arg = argp;
3453 struct kvm_cpuid cpuid;
3456 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3458 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3461 case KVM_SET_CPUID2: {
3462 struct kvm_cpuid2 __user *cpuid_arg = argp;
3463 struct kvm_cpuid2 cpuid;
3466 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3468 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3469 cpuid_arg->entries);
3472 case KVM_GET_CPUID2: {
3473 struct kvm_cpuid2 __user *cpuid_arg = argp;
3474 struct kvm_cpuid2 cpuid;
3477 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3479 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3480 cpuid_arg->entries);
3484 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3490 r = msr_io(vcpu, argp, do_get_msr, 1);
3493 r = msr_io(vcpu, argp, do_set_msr, 0);
3495 case KVM_TPR_ACCESS_REPORTING: {
3496 struct kvm_tpr_access_ctl tac;
3499 if (copy_from_user(&tac, argp, sizeof tac))
3501 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3505 if (copy_to_user(argp, &tac, sizeof tac))
3510 case KVM_SET_VAPIC_ADDR: {
3511 struct kvm_vapic_addr va;
3515 if (!lapic_in_kernel(vcpu))
3518 if (copy_from_user(&va, argp, sizeof va))
3520 idx = srcu_read_lock(&vcpu->kvm->srcu);
3521 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3522 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3525 case KVM_X86_SETUP_MCE: {
3529 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3531 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3534 case KVM_X86_SET_MCE: {
3535 struct kvm_x86_mce mce;
3538 if (copy_from_user(&mce, argp, sizeof mce))
3540 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3543 case KVM_GET_VCPU_EVENTS: {
3544 struct kvm_vcpu_events events;
3546 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3549 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3554 case KVM_SET_VCPU_EVENTS: {
3555 struct kvm_vcpu_events events;
3558 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3561 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3564 case KVM_GET_DEBUGREGS: {
3565 struct kvm_debugregs dbgregs;
3567 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3570 if (copy_to_user(argp, &dbgregs,
3571 sizeof(struct kvm_debugregs)))
3576 case KVM_SET_DEBUGREGS: {
3577 struct kvm_debugregs dbgregs;
3580 if (copy_from_user(&dbgregs, argp,
3581 sizeof(struct kvm_debugregs)))
3584 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3587 case KVM_GET_XSAVE: {
3588 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3593 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3596 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3601 case KVM_SET_XSAVE: {
3602 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3603 if (IS_ERR(u.xsave))
3604 return PTR_ERR(u.xsave);
3606 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3609 case KVM_GET_XCRS: {
3610 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3615 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3618 if (copy_to_user(argp, u.xcrs,
3619 sizeof(struct kvm_xcrs)))
3624 case KVM_SET_XCRS: {
3625 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3627 return PTR_ERR(u.xcrs);
3629 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3632 case KVM_SET_TSC_KHZ: {
3636 user_tsc_khz = (u32)arg;
3638 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3641 if (user_tsc_khz == 0)
3642 user_tsc_khz = tsc_khz;
3644 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3649 case KVM_GET_TSC_KHZ: {
3650 r = vcpu->arch.virtual_tsc_khz;
3653 case KVM_KVMCLOCK_CTRL: {
3654 r = kvm_set_guest_paused(vcpu);
3657 case KVM_ENABLE_CAP: {
3658 struct kvm_enable_cap cap;
3661 if (copy_from_user(&cap, argp, sizeof(cap)))
3663 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3674 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3676 return VM_FAULT_SIGBUS;
3679 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3683 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3685 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3689 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3692 kvm->arch.ept_identity_map_addr = ident_addr;
3696 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3697 u32 kvm_nr_mmu_pages)
3699 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3702 mutex_lock(&kvm->slots_lock);
3704 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3705 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3707 mutex_unlock(&kvm->slots_lock);
3711 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3713 return kvm->arch.n_max_mmu_pages;
3716 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3718 struct kvm_pic *pic = kvm->arch.vpic;
3722 switch (chip->chip_id) {
3723 case KVM_IRQCHIP_PIC_MASTER:
3724 memcpy(&chip->chip.pic, &pic->pics[0],
3725 sizeof(struct kvm_pic_state));
3727 case KVM_IRQCHIP_PIC_SLAVE:
3728 memcpy(&chip->chip.pic, &pic->pics[1],
3729 sizeof(struct kvm_pic_state));
3731 case KVM_IRQCHIP_IOAPIC:
3732 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3741 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3743 struct kvm_pic *pic = kvm->arch.vpic;
3747 switch (chip->chip_id) {
3748 case KVM_IRQCHIP_PIC_MASTER:
3749 spin_lock(&pic->lock);
3750 memcpy(&pic->pics[0], &chip->chip.pic,
3751 sizeof(struct kvm_pic_state));
3752 spin_unlock(&pic->lock);
3754 case KVM_IRQCHIP_PIC_SLAVE:
3755 spin_lock(&pic->lock);
3756 memcpy(&pic->pics[1], &chip->chip.pic,
3757 sizeof(struct kvm_pic_state));
3758 spin_unlock(&pic->lock);
3760 case KVM_IRQCHIP_IOAPIC:
3761 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3767 kvm_pic_update_irq(pic);
3771 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3773 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3775 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3777 mutex_lock(&kps->lock);
3778 memcpy(ps, &kps->channels, sizeof(*ps));
3779 mutex_unlock(&kps->lock);
3783 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3786 struct kvm_pit *pit = kvm->arch.vpit;
3788 mutex_lock(&pit->pit_state.lock);
3789 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
3790 for (i = 0; i < 3; i++)
3791 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3792 mutex_unlock(&pit->pit_state.lock);
3796 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3798 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3799 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3800 sizeof(ps->channels));
3801 ps->flags = kvm->arch.vpit->pit_state.flags;
3802 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3803 memset(&ps->reserved, 0, sizeof(ps->reserved));
3807 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3811 u32 prev_legacy, cur_legacy;
3812 struct kvm_pit *pit = kvm->arch.vpit;
3814 mutex_lock(&pit->pit_state.lock);
3815 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3816 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3817 if (!prev_legacy && cur_legacy)
3819 memcpy(&pit->pit_state.channels, &ps->channels,
3820 sizeof(pit->pit_state.channels));
3821 pit->pit_state.flags = ps->flags;
3822 for (i = 0; i < 3; i++)
3823 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
3825 mutex_unlock(&pit->pit_state.lock);
3829 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3830 struct kvm_reinject_control *control)
3832 struct kvm_pit *pit = kvm->arch.vpit;
3837 /* pit->pit_state.lock was overloaded to prevent userspace from getting
3838 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
3839 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
3841 mutex_lock(&pit->pit_state.lock);
3842 kvm_pit_set_reinject(pit, control->pit_reinject);
3843 mutex_unlock(&pit->pit_state.lock);
3849 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
3850 * @kvm: kvm instance
3851 * @log: slot id and address to which we copy the log
3853 * Steps 1-4 below provide general overview of dirty page logging. See
3854 * kvm_get_dirty_log_protect() function description for additional details.
3856 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
3857 * always flush the TLB (step 4) even if previous step failed and the dirty
3858 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
3859 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
3860 * writes will be marked dirty for next log read.
3862 * 1. Take a snapshot of the bit and clear it if needed.
3863 * 2. Write protect the corresponding page.
3864 * 3. Copy the snapshot to the userspace.
3865 * 4. Flush TLB's if needed.
3867 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
3869 bool is_dirty = false;
3872 mutex_lock(&kvm->slots_lock);
3875 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
3877 if (kvm_x86_ops->flush_log_dirty)
3878 kvm_x86_ops->flush_log_dirty(kvm);
3880 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
3883 * All the TLBs can be flushed out of mmu lock, see the comments in
3884 * kvm_mmu_slot_remove_write_access().
3886 lockdep_assert_held(&kvm->slots_lock);
3888 kvm_flush_remote_tlbs(kvm);
3890 mutex_unlock(&kvm->slots_lock);
3894 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
3897 if (!irqchip_in_kernel(kvm))
3900 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3901 irq_event->irq, irq_event->level,
3906 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3907 struct kvm_enable_cap *cap)
3915 case KVM_CAP_DISABLE_QUIRKS:
3916 kvm->arch.disabled_quirks = cap->args[0];
3919 case KVM_CAP_SPLIT_IRQCHIP: {
3920 mutex_lock(&kvm->lock);
3922 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
3923 goto split_irqchip_unlock;
3925 if (irqchip_in_kernel(kvm))
3926 goto split_irqchip_unlock;
3927 if (kvm->created_vcpus)
3928 goto split_irqchip_unlock;
3929 kvm->arch.irqchip_mode = KVM_IRQCHIP_INIT_IN_PROGRESS;
3930 r = kvm_setup_empty_irq_routing(kvm);
3932 kvm->arch.irqchip_mode = KVM_IRQCHIP_NONE;
3933 /* Pairs with smp_rmb() when reading irqchip_mode */
3935 goto split_irqchip_unlock;
3937 /* Pairs with irqchip_in_kernel. */
3939 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
3940 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
3942 split_irqchip_unlock:
3943 mutex_unlock(&kvm->lock);
3946 case KVM_CAP_X2APIC_API:
3948 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
3951 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
3952 kvm->arch.x2apic_format = true;
3953 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
3954 kvm->arch.x2apic_broadcast_quirk_disabled = true;
3965 long kvm_arch_vm_ioctl(struct file *filp,
3966 unsigned int ioctl, unsigned long arg)
3968 struct kvm *kvm = filp->private_data;
3969 void __user *argp = (void __user *)arg;
3972 * This union makes it completely explicit to gcc-3.x
3973 * that these two variables' stack usage should be
3974 * combined, not added together.
3977 struct kvm_pit_state ps;
3978 struct kvm_pit_state2 ps2;
3979 struct kvm_pit_config pit_config;
3983 case KVM_SET_TSS_ADDR:
3984 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3986 case KVM_SET_IDENTITY_MAP_ADDR: {
3990 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3992 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3995 case KVM_SET_NR_MMU_PAGES:
3996 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3998 case KVM_GET_NR_MMU_PAGES:
3999 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4001 case KVM_CREATE_IRQCHIP: {
4002 mutex_lock(&kvm->lock);
4005 if (irqchip_in_kernel(kvm))
4006 goto create_irqchip_unlock;
4009 if (kvm->created_vcpus)
4010 goto create_irqchip_unlock;
4012 r = kvm_pic_init(kvm);
4014 goto create_irqchip_unlock;
4016 r = kvm_ioapic_init(kvm);
4018 mutex_lock(&kvm->slots_lock);
4019 kvm_pic_destroy(kvm);
4020 mutex_unlock(&kvm->slots_lock);
4021 goto create_irqchip_unlock;
4024 kvm->arch.irqchip_mode = KVM_IRQCHIP_INIT_IN_PROGRESS;
4025 r = kvm_setup_default_irq_routing(kvm);
4027 kvm->arch.irqchip_mode = KVM_IRQCHIP_NONE;
4028 /* Pairs with smp_rmb() when reading irqchip_mode */
4030 mutex_lock(&kvm->slots_lock);
4031 mutex_lock(&kvm->irq_lock);
4032 kvm_ioapic_destroy(kvm);
4033 kvm_pic_destroy(kvm);
4034 mutex_unlock(&kvm->irq_lock);
4035 mutex_unlock(&kvm->slots_lock);
4036 goto create_irqchip_unlock;
4038 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4040 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4041 create_irqchip_unlock:
4042 mutex_unlock(&kvm->lock);
4045 case KVM_CREATE_PIT:
4046 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4048 case KVM_CREATE_PIT2:
4050 if (copy_from_user(&u.pit_config, argp,
4051 sizeof(struct kvm_pit_config)))
4054 mutex_lock(&kvm->lock);
4057 goto create_pit_unlock;
4059 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4063 mutex_unlock(&kvm->lock);
4065 case KVM_GET_IRQCHIP: {
4066 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4067 struct kvm_irqchip *chip;
4069 chip = memdup_user(argp, sizeof(*chip));
4076 if (!irqchip_kernel(kvm))
4077 goto get_irqchip_out;
4078 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4080 goto get_irqchip_out;
4082 if (copy_to_user(argp, chip, sizeof *chip))
4083 goto get_irqchip_out;
4089 case KVM_SET_IRQCHIP: {
4090 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4091 struct kvm_irqchip *chip;
4093 chip = memdup_user(argp, sizeof(*chip));
4100 if (!irqchip_kernel(kvm))
4101 goto set_irqchip_out;
4102 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4104 goto set_irqchip_out;
4112 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4115 if (!kvm->arch.vpit)
4117 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4121 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4128 if (copy_from_user(&u.ps, argp, sizeof u.ps))
4131 if (!kvm->arch.vpit)
4133 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4136 case KVM_GET_PIT2: {
4138 if (!kvm->arch.vpit)
4140 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4144 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4149 case KVM_SET_PIT2: {
4151 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4154 if (!kvm->arch.vpit)
4156 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4159 case KVM_REINJECT_CONTROL: {
4160 struct kvm_reinject_control control;
4162 if (copy_from_user(&control, argp, sizeof(control)))
4164 r = kvm_vm_ioctl_reinject(kvm, &control);
4167 case KVM_SET_BOOT_CPU_ID:
4169 mutex_lock(&kvm->lock);
4170 if (kvm->created_vcpus)
4173 kvm->arch.bsp_vcpu_id = arg;
4174 mutex_unlock(&kvm->lock);
4176 case KVM_XEN_HVM_CONFIG: {
4178 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
4179 sizeof(struct kvm_xen_hvm_config)))
4182 if (kvm->arch.xen_hvm_config.flags)
4187 case KVM_SET_CLOCK: {
4188 struct kvm_clock_data user_ns;
4192 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4200 local_irq_disable();
4201 now_ns = __get_kvmclock_ns(kvm);
4202 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4204 kvm_gen_update_masterclock(kvm);
4207 case KVM_GET_CLOCK: {
4208 struct kvm_clock_data user_ns;
4211 local_irq_disable();
4212 now_ns = __get_kvmclock_ns(kvm);
4213 user_ns.clock = now_ns;
4214 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4216 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4219 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4224 case KVM_ENABLE_CAP: {
4225 struct kvm_enable_cap cap;
4228 if (copy_from_user(&cap, argp, sizeof(cap)))
4230 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4240 static void kvm_init_msr_list(void)
4245 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4246 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4250 * Even MSRs that are valid in the host may not be exposed
4251 * to the guests in some cases.
4253 switch (msrs_to_save[i]) {
4254 case MSR_IA32_BNDCFGS:
4255 if (!kvm_x86_ops->mpx_supported())
4259 if (!kvm_x86_ops->rdtscp_supported())
4267 msrs_to_save[j] = msrs_to_save[i];
4270 num_msrs_to_save = j;
4272 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4273 switch (emulated_msrs[i]) {
4274 case MSR_IA32_SMBASE:
4275 if (!kvm_x86_ops->cpu_has_high_real_mode_segbase())
4283 emulated_msrs[j] = emulated_msrs[i];
4286 num_emulated_msrs = j;
4289 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4297 if (!(lapic_in_kernel(vcpu) &&
4298 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4299 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4310 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4317 if (!(lapic_in_kernel(vcpu) &&
4318 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4320 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4322 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
4332 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4333 struct kvm_segment *var, int seg)
4335 kvm_x86_ops->set_segment(vcpu, var, seg);
4338 void kvm_get_segment(struct kvm_vcpu *vcpu,
4339 struct kvm_segment *var, int seg)
4341 kvm_x86_ops->get_segment(vcpu, var, seg);
4344 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4345 struct x86_exception *exception)
4349 BUG_ON(!mmu_is_nested(vcpu));
4351 /* NPT walks are always user-walks */
4352 access |= PFERR_USER_MASK;
4353 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4358 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4359 struct x86_exception *exception)
4361 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4362 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4365 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4366 struct x86_exception *exception)
4368 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4369 access |= PFERR_FETCH_MASK;
4370 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4373 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4374 struct x86_exception *exception)
4376 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4377 access |= PFERR_WRITE_MASK;
4378 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4381 /* uses this to access any guest's mapped memory without checking CPL */
4382 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4383 struct x86_exception *exception)
4385 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4388 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4389 struct kvm_vcpu *vcpu, u32 access,
4390 struct x86_exception *exception)
4393 int r = X86EMUL_CONTINUE;
4396 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4398 unsigned offset = addr & (PAGE_SIZE-1);
4399 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4402 if (gpa == UNMAPPED_GVA)
4403 return X86EMUL_PROPAGATE_FAULT;
4404 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4407 r = X86EMUL_IO_NEEDED;
4419 /* used for instruction fetching */
4420 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4421 gva_t addr, void *val, unsigned int bytes,
4422 struct x86_exception *exception)
4424 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4425 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4429 /* Inline kvm_read_guest_virt_helper for speed. */
4430 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4432 if (unlikely(gpa == UNMAPPED_GVA))
4433 return X86EMUL_PROPAGATE_FAULT;
4435 offset = addr & (PAGE_SIZE-1);
4436 if (WARN_ON(offset + bytes > PAGE_SIZE))
4437 bytes = (unsigned)PAGE_SIZE - offset;
4438 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4440 if (unlikely(ret < 0))
4441 return X86EMUL_IO_NEEDED;
4443 return X86EMUL_CONTINUE;
4446 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
4447 gva_t addr, void *val, unsigned int bytes,
4448 struct x86_exception *exception)
4450 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4451 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4453 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4456 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4458 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4459 gva_t addr, void *val, unsigned int bytes,
4460 struct x86_exception *exception)
4462 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4463 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4466 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4467 unsigned long addr, void *val, unsigned int bytes)
4469 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4470 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4472 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4475 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
4476 gva_t addr, void *val,
4478 struct x86_exception *exception)
4480 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4482 int r = X86EMUL_CONTINUE;
4485 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4488 unsigned offset = addr & (PAGE_SIZE-1);
4489 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4492 if (gpa == UNMAPPED_GVA)
4493 return X86EMUL_PROPAGATE_FAULT;
4494 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4496 r = X86EMUL_IO_NEEDED;
4507 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4509 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4510 gpa_t gpa, bool write)
4512 /* For APIC access vmexit */
4513 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4516 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
4517 trace_vcpu_match_mmio(gva, gpa, write, true);
4524 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4525 gpa_t *gpa, struct x86_exception *exception,
4528 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4529 | (write ? PFERR_WRITE_MASK : 0);
4532 * currently PKRU is only applied to ept enabled guest so
4533 * there is no pkey in EPT page table for L1 guest or EPT
4534 * shadow page table for L2 guest.
4536 if (vcpu_match_mmio_gva(vcpu, gva)
4537 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4538 vcpu->arch.access, 0, access)) {
4539 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4540 (gva & (PAGE_SIZE - 1));
4541 trace_vcpu_match_mmio(gva, *gpa, write, false);
4545 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4547 if (*gpa == UNMAPPED_GVA)
4550 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
4553 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4554 const void *val, int bytes)
4558 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4561 kvm_page_track_write(vcpu, gpa, val, bytes);
4565 struct read_write_emulator_ops {
4566 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4568 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4569 void *val, int bytes);
4570 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4571 int bytes, void *val);
4572 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4573 void *val, int bytes);
4577 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4579 if (vcpu->mmio_read_completed) {
4580 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4581 vcpu->mmio_fragments[0].gpa, *(u64 *)val);
4582 vcpu->mmio_read_completed = 0;
4589 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4590 void *val, int bytes)
4592 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4595 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4596 void *val, int bytes)
4598 return emulator_write_phys(vcpu, gpa, val, bytes);
4601 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4603 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
4604 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4607 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4608 void *val, int bytes)
4610 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
4611 return X86EMUL_IO_NEEDED;
4614 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4615 void *val, int bytes)
4617 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4619 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4620 return X86EMUL_CONTINUE;
4623 static const struct read_write_emulator_ops read_emultor = {
4624 .read_write_prepare = read_prepare,
4625 .read_write_emulate = read_emulate,
4626 .read_write_mmio = vcpu_mmio_read,
4627 .read_write_exit_mmio = read_exit_mmio,
4630 static const struct read_write_emulator_ops write_emultor = {
4631 .read_write_emulate = write_emulate,
4632 .read_write_mmio = write_mmio,
4633 .read_write_exit_mmio = write_exit_mmio,
4637 static int emulator_read_write_onepage(unsigned long addr, void *val,
4639 struct x86_exception *exception,
4640 struct kvm_vcpu *vcpu,
4641 const struct read_write_emulator_ops *ops)
4645 bool write = ops->write;
4646 struct kvm_mmio_fragment *frag;
4647 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4650 * If the exit was due to a NPF we may already have a GPA.
4651 * If the GPA is present, use it to avoid the GVA to GPA table walk.
4652 * Note, this cannot be used on string operations since string
4653 * operation using rep will only have the initial GPA from the NPF
4656 if (vcpu->arch.gpa_available &&
4657 emulator_can_use_gpa(ctxt) &&
4658 vcpu_is_mmio_gpa(vcpu, addr, exception->address, write) &&
4659 (addr & ~PAGE_MASK) == (exception->address & ~PAGE_MASK)) {
4660 gpa = exception->address;
4664 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4667 return X86EMUL_PROPAGATE_FAULT;
4669 /* For APIC access vmexit */
4673 if (ops->read_write_emulate(vcpu, gpa, val, bytes))
4674 return X86EMUL_CONTINUE;
4678 * Is this MMIO handled locally?
4680 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4681 if (handled == bytes)
4682 return X86EMUL_CONTINUE;
4688 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4689 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4693 return X86EMUL_CONTINUE;
4696 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4698 void *val, unsigned int bytes,
4699 struct x86_exception *exception,
4700 const struct read_write_emulator_ops *ops)
4702 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4706 if (ops->read_write_prepare &&
4707 ops->read_write_prepare(vcpu, val, bytes))
4708 return X86EMUL_CONTINUE;
4710 vcpu->mmio_nr_fragments = 0;
4712 /* Crossing a page boundary? */
4713 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4716 now = -addr & ~PAGE_MASK;
4717 rc = emulator_read_write_onepage(addr, val, now, exception,
4720 if (rc != X86EMUL_CONTINUE)
4723 if (ctxt->mode != X86EMUL_MODE_PROT64)
4729 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4731 if (rc != X86EMUL_CONTINUE)
4734 if (!vcpu->mmio_nr_fragments)
4737 gpa = vcpu->mmio_fragments[0].gpa;
4739 vcpu->mmio_needed = 1;
4740 vcpu->mmio_cur_fragment = 0;
4742 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4743 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4744 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4745 vcpu->run->mmio.phys_addr = gpa;
4747 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4750 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4754 struct x86_exception *exception)
4756 return emulator_read_write(ctxt, addr, val, bytes,
4757 exception, &read_emultor);
4760 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4764 struct x86_exception *exception)
4766 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4767 exception, &write_emultor);
4770 #define CMPXCHG_TYPE(t, ptr, old, new) \
4771 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4773 #ifdef CONFIG_X86_64
4774 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4776 # define CMPXCHG64(ptr, old, new) \
4777 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4780 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4785 struct x86_exception *exception)
4787 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4793 /* guests cmpxchg8b have to be emulated atomically */
4794 if (bytes > 8 || (bytes & (bytes - 1)))
4797 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4799 if (gpa == UNMAPPED_GVA ||
4800 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4803 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4806 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4807 if (is_error_page(page))
4810 kaddr = kmap_atomic(page);
4811 kaddr += offset_in_page(gpa);
4814 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4817 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4820 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4823 exchanged = CMPXCHG64(kaddr, old, new);
4828 kunmap_atomic(kaddr);
4829 kvm_release_page_dirty(page);
4832 return X86EMUL_CMPXCHG_FAILED;
4834 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
4835 kvm_page_track_write(vcpu, gpa, new, bytes);
4837 return X86EMUL_CONTINUE;
4840 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4842 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4845 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4847 /* TODO: String I/O for in kernel device */
4850 if (vcpu->arch.pio.in)
4851 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
4852 vcpu->arch.pio.size, pd);
4854 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
4855 vcpu->arch.pio.port, vcpu->arch.pio.size,
4860 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4861 unsigned short port, void *val,
4862 unsigned int count, bool in)
4864 vcpu->arch.pio.port = port;
4865 vcpu->arch.pio.in = in;
4866 vcpu->arch.pio.count = count;
4867 vcpu->arch.pio.size = size;
4869 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4870 vcpu->arch.pio.count = 0;
4874 vcpu->run->exit_reason = KVM_EXIT_IO;
4875 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4876 vcpu->run->io.size = size;
4877 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4878 vcpu->run->io.count = count;
4879 vcpu->run->io.port = port;
4884 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4885 int size, unsigned short port, void *val,
4888 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4891 if (vcpu->arch.pio.count)
4894 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
4897 memcpy(val, vcpu->arch.pio_data, size * count);
4898 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
4899 vcpu->arch.pio.count = 0;
4906 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4907 int size, unsigned short port,
4908 const void *val, unsigned int count)
4910 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4912 memcpy(vcpu->arch.pio_data, val, size * count);
4913 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
4914 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
4917 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4919 return kvm_x86_ops->get_segment_base(vcpu, seg);
4922 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4924 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4927 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
4929 if (!need_emulate_wbinvd(vcpu))
4930 return X86EMUL_CONTINUE;
4932 if (kvm_x86_ops->has_wbinvd_exit()) {
4933 int cpu = get_cpu();
4935 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4936 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4937 wbinvd_ipi, NULL, 1);
4939 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4942 return X86EMUL_CONTINUE;
4945 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4947 kvm_emulate_wbinvd_noskip(vcpu);
4948 return kvm_skip_emulated_instruction(vcpu);
4950 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4954 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4956 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
4959 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
4960 unsigned long *dest)
4962 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4965 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
4966 unsigned long value)
4969 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4972 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4974 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4977 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4979 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4980 unsigned long value;
4984 value = kvm_read_cr0(vcpu);
4987 value = vcpu->arch.cr2;
4990 value = kvm_read_cr3(vcpu);
4993 value = kvm_read_cr4(vcpu);
4996 value = kvm_get_cr8(vcpu);
4999 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5006 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5008 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5013 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5016 vcpu->arch.cr2 = val;
5019 res = kvm_set_cr3(vcpu, val);
5022 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5025 res = kvm_set_cr8(vcpu, val);
5028 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5035 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5037 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5040 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5042 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5045 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5047 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5050 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5052 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5055 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5057 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5060 static unsigned long emulator_get_cached_segment_base(
5061 struct x86_emulate_ctxt *ctxt, int seg)
5063 return get_segment_base(emul_to_vcpu(ctxt), seg);
5066 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5067 struct desc_struct *desc, u32 *base3,
5070 struct kvm_segment var;
5072 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5073 *selector = var.selector;
5076 memset(desc, 0, sizeof(*desc));
5082 set_desc_limit(desc, var.limit);
5083 set_desc_base(desc, (unsigned long)var.base);
5084 #ifdef CONFIG_X86_64
5086 *base3 = var.base >> 32;
5088 desc->type = var.type;
5090 desc->dpl = var.dpl;
5091 desc->p = var.present;
5092 desc->avl = var.avl;
5100 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5101 struct desc_struct *desc, u32 base3,
5104 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5105 struct kvm_segment var;
5107 var.selector = selector;
5108 var.base = get_desc_base(desc);
5109 #ifdef CONFIG_X86_64
5110 var.base |= ((u64)base3) << 32;
5112 var.limit = get_desc_limit(desc);
5114 var.limit = (var.limit << 12) | 0xfff;
5115 var.type = desc->type;
5116 var.dpl = desc->dpl;
5121 var.avl = desc->avl;
5122 var.present = desc->p;
5123 var.unusable = !var.present;
5126 kvm_set_segment(vcpu, &var, seg);
5130 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5131 u32 msr_index, u64 *pdata)
5133 struct msr_data msr;
5136 msr.index = msr_index;
5137 msr.host_initiated = false;
5138 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5146 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5147 u32 msr_index, u64 data)
5149 struct msr_data msr;
5152 msr.index = msr_index;
5153 msr.host_initiated = false;
5154 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5157 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5159 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5161 return vcpu->arch.smbase;
5164 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5166 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5168 vcpu->arch.smbase = smbase;
5171 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5174 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5177 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5178 u32 pmc, u64 *pdata)
5180 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5183 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5185 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5188 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
5191 kvm_load_guest_fpu(emul_to_vcpu(ctxt));
5194 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
5199 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5200 struct x86_instruction_info *info,
5201 enum x86_intercept_stage stage)
5203 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5206 static void emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5207 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
5209 kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx);
5212 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5214 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5217 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5219 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5222 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5224 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5227 static const struct x86_emulate_ops emulate_ops = {
5228 .read_gpr = emulator_read_gpr,
5229 .write_gpr = emulator_write_gpr,
5230 .read_std = kvm_read_guest_virt_system,
5231 .write_std = kvm_write_guest_virt_system,
5232 .read_phys = kvm_read_guest_phys_system,
5233 .fetch = kvm_fetch_guest_virt,
5234 .read_emulated = emulator_read_emulated,
5235 .write_emulated = emulator_write_emulated,
5236 .cmpxchg_emulated = emulator_cmpxchg_emulated,
5237 .invlpg = emulator_invlpg,
5238 .pio_in_emulated = emulator_pio_in_emulated,
5239 .pio_out_emulated = emulator_pio_out_emulated,
5240 .get_segment = emulator_get_segment,
5241 .set_segment = emulator_set_segment,
5242 .get_cached_segment_base = emulator_get_cached_segment_base,
5243 .get_gdt = emulator_get_gdt,
5244 .get_idt = emulator_get_idt,
5245 .set_gdt = emulator_set_gdt,
5246 .set_idt = emulator_set_idt,
5247 .get_cr = emulator_get_cr,
5248 .set_cr = emulator_set_cr,
5249 .cpl = emulator_get_cpl,
5250 .get_dr = emulator_get_dr,
5251 .set_dr = emulator_set_dr,
5252 .get_smbase = emulator_get_smbase,
5253 .set_smbase = emulator_set_smbase,
5254 .set_msr = emulator_set_msr,
5255 .get_msr = emulator_get_msr,
5256 .check_pmc = emulator_check_pmc,
5257 .read_pmc = emulator_read_pmc,
5258 .halt = emulator_halt,
5259 .wbinvd = emulator_wbinvd,
5260 .fix_hypercall = emulator_fix_hypercall,
5261 .get_fpu = emulator_get_fpu,
5262 .put_fpu = emulator_put_fpu,
5263 .intercept = emulator_intercept,
5264 .get_cpuid = emulator_get_cpuid,
5265 .set_nmi_mask = emulator_set_nmi_mask,
5268 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5270 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5272 * an sti; sti; sequence only disable interrupts for the first
5273 * instruction. So, if the last instruction, be it emulated or
5274 * not, left the system with the INT_STI flag enabled, it
5275 * means that the last instruction is an sti. We should not
5276 * leave the flag on in this case. The same goes for mov ss
5278 if (int_shadow & mask)
5280 if (unlikely(int_shadow || mask)) {
5281 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5283 kvm_make_request(KVM_REQ_EVENT, vcpu);
5287 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5289 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5290 if (ctxt->exception.vector == PF_VECTOR)
5291 return kvm_propagate_fault(vcpu, &ctxt->exception);
5293 if (ctxt->exception.error_code_valid)
5294 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5295 ctxt->exception.error_code);
5297 kvm_queue_exception(vcpu, ctxt->exception.vector);
5301 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5303 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5306 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5308 ctxt->eflags = kvm_get_rflags(vcpu);
5309 ctxt->eip = kvm_rip_read(vcpu);
5310 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5311 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
5312 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
5313 cs_db ? X86EMUL_MODE_PROT32 :
5314 X86EMUL_MODE_PROT16;
5315 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5316 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5317 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5318 ctxt->emul_flags = vcpu->arch.hflags;
5320 init_decode_cache(ctxt);
5321 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5324 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5326 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5329 init_emulate_ctxt(vcpu);
5333 ctxt->_eip = ctxt->eip + inc_eip;
5334 ret = emulate_int_real(ctxt, irq);
5336 if (ret != X86EMUL_CONTINUE)
5337 return EMULATE_FAIL;
5339 ctxt->eip = ctxt->_eip;
5340 kvm_rip_write(vcpu, ctxt->eip);
5341 kvm_set_rflags(vcpu, ctxt->eflags);
5343 if (irq == NMI_VECTOR)
5344 vcpu->arch.nmi_pending = 0;
5346 vcpu->arch.interrupt.pending = false;
5348 return EMULATE_DONE;
5350 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5352 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5354 int r = EMULATE_DONE;
5356 ++vcpu->stat.insn_emulation_fail;
5357 trace_kvm_emulate_insn_failed(vcpu);
5358 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5359 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5360 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5361 vcpu->run->internal.ndata = 0;
5364 kvm_queue_exception(vcpu, UD_VECTOR);
5369 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5370 bool write_fault_to_shadow_pgtable,
5376 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5379 if (!vcpu->arch.mmu.direct_map) {
5381 * Write permission should be allowed since only
5382 * write access need to be emulated.
5384 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5387 * If the mapping is invalid in guest, let cpu retry
5388 * it to generate fault.
5390 if (gpa == UNMAPPED_GVA)
5395 * Do not retry the unhandleable instruction if it faults on the
5396 * readonly host memory, otherwise it will goto a infinite loop:
5397 * retry instruction -> write #PF -> emulation fail -> retry
5398 * instruction -> ...
5400 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5403 * If the instruction failed on the error pfn, it can not be fixed,
5404 * report the error to userspace.
5406 if (is_error_noslot_pfn(pfn))
5409 kvm_release_pfn_clean(pfn);
5411 /* The instructions are well-emulated on direct mmu. */
5412 if (vcpu->arch.mmu.direct_map) {
5413 unsigned int indirect_shadow_pages;
5415 spin_lock(&vcpu->kvm->mmu_lock);
5416 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5417 spin_unlock(&vcpu->kvm->mmu_lock);
5419 if (indirect_shadow_pages)
5420 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5426 * if emulation was due to access to shadowed page table
5427 * and it failed try to unshadow page and re-enter the
5428 * guest to let CPU execute the instruction.
5430 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5433 * If the access faults on its page table, it can not
5434 * be fixed by unprotecting shadow page and it should
5435 * be reported to userspace.
5437 return !write_fault_to_shadow_pgtable;
5440 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5441 unsigned long cr2, int emulation_type)
5443 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5444 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5446 last_retry_eip = vcpu->arch.last_retry_eip;
5447 last_retry_addr = vcpu->arch.last_retry_addr;
5450 * If the emulation is caused by #PF and it is non-page_table
5451 * writing instruction, it means the VM-EXIT is caused by shadow
5452 * page protected, we can zap the shadow page and retry this
5453 * instruction directly.
5455 * Note: if the guest uses a non-page-table modifying instruction
5456 * on the PDE that points to the instruction, then we will unmap
5457 * the instruction and go to an infinite loop. So, we cache the
5458 * last retried eip and the last fault address, if we meet the eip
5459 * and the address again, we can break out of the potential infinite
5462 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5464 if (!(emulation_type & EMULTYPE_RETRY))
5467 if (x86_page_table_writing_insn(ctxt))
5470 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5473 vcpu->arch.last_retry_eip = ctxt->eip;
5474 vcpu->arch.last_retry_addr = cr2;
5476 if (!vcpu->arch.mmu.direct_map)
5477 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5479 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5484 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5485 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5487 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5489 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5490 /* This is a good place to trace that we are exiting SMM. */
5491 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5493 /* Process a latched INIT or SMI, if any. */
5494 kvm_make_request(KVM_REQ_EVENT, vcpu);
5497 kvm_mmu_reset_context(vcpu);
5500 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5502 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5504 vcpu->arch.hflags = emul_flags;
5506 if (changed & HF_SMM_MASK)
5507 kvm_smm_changed(vcpu);
5510 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5519 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5520 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5525 static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
5527 struct kvm_run *kvm_run = vcpu->run;
5530 * rflags is the old, "raw" value of the flags. The new value has
5531 * not been saved yet.
5533 * This is correct even for TF set by the guest, because "the
5534 * processor will not generate this exception after the instruction
5535 * that sets the TF flag".
5537 if (unlikely(rflags & X86_EFLAGS_TF)) {
5538 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5539 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
5541 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5542 kvm_run->debug.arch.exception = DB_VECTOR;
5543 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5544 *r = EMULATE_USER_EXIT;
5547 * "Certain debug exceptions may clear bit 0-3. The
5548 * remaining contents of the DR6 register are never
5549 * cleared by the processor".
5551 vcpu->arch.dr6 &= ~15;
5552 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5553 kvm_queue_exception(vcpu, DB_VECTOR);
5558 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
5560 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5561 int r = EMULATE_DONE;
5563 kvm_x86_ops->skip_emulated_instruction(vcpu);
5564 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5565 return r == EMULATE_DONE;
5567 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
5569 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5571 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5572 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5573 struct kvm_run *kvm_run = vcpu->run;
5574 unsigned long eip = kvm_get_linear_rip(vcpu);
5575 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5576 vcpu->arch.guest_debug_dr7,
5580 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5581 kvm_run->debug.arch.pc = eip;
5582 kvm_run->debug.arch.exception = DB_VECTOR;
5583 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5584 *r = EMULATE_USER_EXIT;
5589 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5590 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5591 unsigned long eip = kvm_get_linear_rip(vcpu);
5592 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5597 vcpu->arch.dr6 &= ~15;
5598 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5599 kvm_queue_exception(vcpu, DB_VECTOR);
5608 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5615 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5616 bool writeback = true;
5617 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5620 * Clear write_fault_to_shadow_pgtable here to ensure it is
5623 vcpu->arch.write_fault_to_shadow_pgtable = false;
5624 kvm_clear_exception_queue(vcpu);
5626 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5627 init_emulate_ctxt(vcpu);
5630 * We will reenter on the same instruction since
5631 * we do not set complete_userspace_io. This does not
5632 * handle watchpoints yet, those would be handled in
5635 if (kvm_vcpu_check_breakpoint(vcpu, &r))
5638 ctxt->interruptibility = 0;
5639 ctxt->have_exception = false;
5640 ctxt->exception.vector = -1;
5641 ctxt->perm_ok = false;
5643 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5645 r = x86_decode_insn(ctxt, insn, insn_len);
5647 trace_kvm_emulate_insn_start(vcpu);
5648 ++vcpu->stat.insn_emulation;
5649 if (r != EMULATION_OK) {
5650 if (emulation_type & EMULTYPE_TRAP_UD)
5651 return EMULATE_FAIL;
5652 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5654 return EMULATE_DONE;
5655 if (emulation_type & EMULTYPE_SKIP)
5656 return EMULATE_FAIL;
5657 return handle_emulation_failure(vcpu);
5661 if (emulation_type & EMULTYPE_SKIP) {
5662 kvm_rip_write(vcpu, ctxt->_eip);
5663 if (ctxt->eflags & X86_EFLAGS_RF)
5664 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5665 return EMULATE_DONE;
5668 if (retry_instruction(ctxt, cr2, emulation_type))
5669 return EMULATE_DONE;
5671 /* this is needed for vmware backdoor interface to work since it
5672 changes registers values during IO operation */
5673 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5674 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5675 emulator_invalidate_register_cache(ctxt);
5679 /* Save the faulting GPA (cr2) in the address field */
5680 ctxt->exception.address = cr2;
5682 r = x86_emulate_insn(ctxt);
5684 if (r == EMULATION_INTERCEPTED)
5685 return EMULATE_DONE;
5687 if (r == EMULATION_FAILED) {
5688 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5690 return EMULATE_DONE;
5692 return handle_emulation_failure(vcpu);
5695 if (ctxt->have_exception) {
5697 if (inject_emulated_exception(vcpu))
5699 } else if (vcpu->arch.pio.count) {
5700 if (!vcpu->arch.pio.in) {
5701 /* FIXME: return into emulator if single-stepping. */
5702 vcpu->arch.pio.count = 0;
5705 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5707 r = EMULATE_USER_EXIT;
5708 } else if (vcpu->mmio_needed) {
5709 if (!vcpu->mmio_is_write)
5711 r = EMULATE_USER_EXIT;
5712 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5713 } else if (r == EMULATION_RESTART)
5719 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5720 toggle_interruptibility(vcpu, ctxt->interruptibility);
5721 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5722 if (vcpu->arch.hflags != ctxt->emul_flags)
5723 kvm_set_hflags(vcpu, ctxt->emul_flags);
5724 kvm_rip_write(vcpu, ctxt->eip);
5725 if (r == EMULATE_DONE)
5726 kvm_vcpu_check_singlestep(vcpu, rflags, &r);
5727 if (!ctxt->have_exception ||
5728 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5729 __kvm_set_rflags(vcpu, ctxt->eflags);
5732 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5733 * do nothing, and it will be requested again as soon as
5734 * the shadow expires. But we still need to check here,
5735 * because POPF has no interrupt shadow.
5737 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5738 kvm_make_request(KVM_REQ_EVENT, vcpu);
5740 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5744 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5746 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5748 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5749 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5750 size, port, &val, 1);
5751 /* do not return to emulator after return from userspace */
5752 vcpu->arch.pio.count = 0;
5755 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5757 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
5761 /* We should only ever be called with arch.pio.count equal to 1 */
5762 BUG_ON(vcpu->arch.pio.count != 1);
5764 /* For size less than 4 we merge, else we zero extend */
5765 val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
5769 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
5770 * the copy and tracing
5772 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
5773 vcpu->arch.pio.port, &val, 1);
5774 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5779 int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size, unsigned short port)
5784 /* For size less than 4 we merge, else we zero extend */
5785 val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
5787 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
5790 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5794 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
5798 EXPORT_SYMBOL_GPL(kvm_fast_pio_in);
5800 static int kvmclock_cpu_down_prep(unsigned int cpu)
5802 __this_cpu_write(cpu_tsc_khz, 0);
5806 static void tsc_khz_changed(void *data)
5808 struct cpufreq_freqs *freq = data;
5809 unsigned long khz = 0;
5813 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
5814 khz = cpufreq_quick_get(raw_smp_processor_id());
5817 __this_cpu_write(cpu_tsc_khz, khz);
5820 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
5823 struct cpufreq_freqs *freq = data;
5825 struct kvm_vcpu *vcpu;
5826 int i, send_ipi = 0;
5829 * We allow guests to temporarily run on slowing clocks,
5830 * provided we notify them after, or to run on accelerating
5831 * clocks, provided we notify them before. Thus time never
5834 * However, we have a problem. We can't atomically update
5835 * the frequency of a given CPU from this function; it is
5836 * merely a notifier, which can be called from any CPU.
5837 * Changing the TSC frequency at arbitrary points in time
5838 * requires a recomputation of local variables related to
5839 * the TSC for each VCPU. We must flag these local variables
5840 * to be updated and be sure the update takes place with the
5841 * new frequency before any guests proceed.
5843 * Unfortunately, the combination of hotplug CPU and frequency
5844 * change creates an intractable locking scenario; the order
5845 * of when these callouts happen is undefined with respect to
5846 * CPU hotplug, and they can race with each other. As such,
5847 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
5848 * undefined; you can actually have a CPU frequency change take
5849 * place in between the computation of X and the setting of the
5850 * variable. To protect against this problem, all updates of
5851 * the per_cpu tsc_khz variable are done in an interrupt
5852 * protected IPI, and all callers wishing to update the value
5853 * must wait for a synchronous IPI to complete (which is trivial
5854 * if the caller is on the CPU already). This establishes the
5855 * necessary total order on variable updates.
5857 * Note that because a guest time update may take place
5858 * anytime after the setting of the VCPU's request bit, the
5859 * correct TSC value must be set before the request. However,
5860 * to ensure the update actually makes it to any guest which
5861 * starts running in hardware virtualization between the set
5862 * and the acquisition of the spinlock, we must also ping the
5863 * CPU after setting the request bit.
5867 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
5869 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
5872 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5874 spin_lock(&kvm_lock);
5875 list_for_each_entry(kvm, &vm_list, vm_list) {
5876 kvm_for_each_vcpu(i, vcpu, kvm) {
5877 if (vcpu->cpu != freq->cpu)
5879 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5880 if (vcpu->cpu != smp_processor_id())
5884 spin_unlock(&kvm_lock);
5886 if (freq->old < freq->new && send_ipi) {
5888 * We upscale the frequency. Must make the guest
5889 * doesn't see old kvmclock values while running with
5890 * the new frequency, otherwise we risk the guest sees
5891 * time go backwards.
5893 * In case we update the frequency for another cpu
5894 * (which might be in guest context) send an interrupt
5895 * to kick the cpu out of guest context. Next time
5896 * guest context is entered kvmclock will be updated,
5897 * so the guest will not see stale values.
5899 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
5904 static struct notifier_block kvmclock_cpufreq_notifier_block = {
5905 .notifier_call = kvmclock_cpufreq_notifier
5908 static int kvmclock_cpu_online(unsigned int cpu)
5910 tsc_khz_changed(NULL);
5914 static void kvm_timer_init(void)
5916 max_tsc_khz = tsc_khz;
5918 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5919 #ifdef CONFIG_CPU_FREQ
5920 struct cpufreq_policy policy;
5923 memset(&policy, 0, sizeof(policy));
5925 cpufreq_get_policy(&policy, cpu);
5926 if (policy.cpuinfo.max_freq)
5927 max_tsc_khz = policy.cpuinfo.max_freq;
5930 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
5931 CPUFREQ_TRANSITION_NOTIFIER);
5933 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
5935 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
5936 kvmclock_cpu_online, kvmclock_cpu_down_prep);
5939 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5941 int kvm_is_in_guest(void)
5943 return __this_cpu_read(current_vcpu) != NULL;
5946 static int kvm_is_user_mode(void)
5950 if (__this_cpu_read(current_vcpu))
5951 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5953 return user_mode != 0;
5956 static unsigned long kvm_get_guest_ip(void)
5958 unsigned long ip = 0;
5960 if (__this_cpu_read(current_vcpu))
5961 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5966 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5967 .is_in_guest = kvm_is_in_guest,
5968 .is_user_mode = kvm_is_user_mode,
5969 .get_guest_ip = kvm_get_guest_ip,
5972 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5974 __this_cpu_write(current_vcpu, vcpu);
5976 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5978 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5980 __this_cpu_write(current_vcpu, NULL);
5982 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5984 static void kvm_set_mmio_spte_mask(void)
5987 int maxphyaddr = boot_cpu_data.x86_phys_bits;
5990 * Set the reserved bits and the present bit of an paging-structure
5991 * entry to generate page fault with PFER.RSV = 1.
5993 /* Mask the reserved physical address bits. */
5994 mask = rsvd_bits(maxphyaddr, 51);
5996 /* Set the present bit. */
5999 #ifdef CONFIG_X86_64
6001 * If reserved bit is not supported, clear the present bit to disable
6004 if (maxphyaddr == 52)
6008 kvm_mmu_set_mmio_spte_mask(mask);
6011 #ifdef CONFIG_X86_64
6012 static void pvclock_gtod_update_fn(struct work_struct *work)
6016 struct kvm_vcpu *vcpu;
6019 spin_lock(&kvm_lock);
6020 list_for_each_entry(kvm, &vm_list, vm_list)
6021 kvm_for_each_vcpu(i, vcpu, kvm)
6022 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6023 atomic_set(&kvm_guest_has_master_clock, 0);
6024 spin_unlock(&kvm_lock);
6027 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6030 * Notification about pvclock gtod data update.
6032 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6035 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6036 struct timekeeper *tk = priv;
6038 update_pvclock_gtod(tk);
6040 /* disable master clock if host does not trust, or does not
6041 * use, TSC clocksource
6043 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
6044 atomic_read(&kvm_guest_has_master_clock) != 0)
6045 queue_work(system_long_wq, &pvclock_gtod_work);
6050 static struct notifier_block pvclock_gtod_notifier = {
6051 .notifier_call = pvclock_gtod_notify,
6055 int kvm_arch_init(void *opaque)
6058 struct kvm_x86_ops *ops = opaque;
6061 printk(KERN_ERR "kvm: already loaded the other module\n");
6066 if (!ops->cpu_has_kvm_support()) {
6067 printk(KERN_ERR "kvm: no hardware support\n");
6071 if (ops->disabled_by_bios()) {
6072 printk(KERN_ERR "kvm: disabled by bios\n");
6078 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6080 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6084 r = kvm_mmu_module_init();
6086 goto out_free_percpu;
6088 kvm_set_mmio_spte_mask();
6092 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6093 PT_DIRTY_MASK, PT64_NX_MASK, 0,
6094 PT_PRESENT_MASK, 0);
6097 perf_register_guest_info_callbacks(&kvm_guest_cbs);
6099 if (boot_cpu_has(X86_FEATURE_XSAVE))
6100 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6103 #ifdef CONFIG_X86_64
6104 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6110 free_percpu(shared_msrs);
6115 void kvm_arch_exit(void)
6118 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6120 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6121 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6122 CPUFREQ_TRANSITION_NOTIFIER);
6123 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6124 #ifdef CONFIG_X86_64
6125 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6128 kvm_mmu_module_exit();
6129 free_percpu(shared_msrs);
6132 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6134 ++vcpu->stat.halt_exits;
6135 if (lapic_in_kernel(vcpu)) {
6136 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6139 vcpu->run->exit_reason = KVM_EXIT_HLT;
6143 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6145 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6147 int ret = kvm_skip_emulated_instruction(vcpu);
6149 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6150 * KVM_EXIT_DEBUG here.
6152 return kvm_vcpu_halt(vcpu) && ret;
6154 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6156 #ifdef CONFIG_X86_64
6157 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6158 unsigned long clock_type)
6160 struct kvm_clock_pairing clock_pairing;
6165 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6166 return -KVM_EOPNOTSUPP;
6168 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6169 return -KVM_EOPNOTSUPP;
6171 clock_pairing.sec = ts.tv_sec;
6172 clock_pairing.nsec = ts.tv_nsec;
6173 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6174 clock_pairing.flags = 0;
6177 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6178 sizeof(struct kvm_clock_pairing)))
6186 * kvm_pv_kick_cpu_op: Kick a vcpu.
6188 * @apicid - apicid of vcpu to be kicked.
6190 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6192 struct kvm_lapic_irq lapic_irq;
6194 lapic_irq.shorthand = 0;
6195 lapic_irq.dest_mode = 0;
6196 lapic_irq.dest_id = apicid;
6197 lapic_irq.msi_redir_hint = false;
6199 lapic_irq.delivery_mode = APIC_DM_REMRD;
6200 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6203 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6205 vcpu->arch.apicv_active = false;
6206 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6209 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6211 unsigned long nr, a0, a1, a2, a3, ret;
6214 r = kvm_skip_emulated_instruction(vcpu);
6216 if (kvm_hv_hypercall_enabled(vcpu->kvm))
6217 return kvm_hv_hypercall(vcpu);
6219 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6220 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6221 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6222 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6223 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
6225 trace_kvm_hypercall(nr, a0, a1, a2, a3);
6227 op_64_bit = is_64_bit_mode(vcpu);
6236 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6242 case KVM_HC_VAPIC_POLL_IRQ:
6245 case KVM_HC_KICK_CPU:
6246 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6249 #ifdef CONFIG_X86_64
6250 case KVM_HC_CLOCK_PAIRING:
6251 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
6261 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6262 ++vcpu->stat.hypercalls;
6265 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6267 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6269 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6270 char instruction[3];
6271 unsigned long rip = kvm_rip_read(vcpu);
6273 kvm_x86_ops->patch_hypercall(vcpu, instruction);
6275 return emulator_write_emulated(ctxt, rip, instruction, 3,
6279 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6281 return vcpu->run->request_interrupt_window &&
6282 likely(!pic_in_kernel(vcpu->kvm));
6285 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6287 struct kvm_run *kvm_run = vcpu->run;
6289 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6290 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6291 kvm_run->cr8 = kvm_get_cr8(vcpu);
6292 kvm_run->apic_base = kvm_get_apic_base(vcpu);
6293 kvm_run->ready_for_interrupt_injection =
6294 pic_in_kernel(vcpu->kvm) ||
6295 kvm_vcpu_ready_for_interrupt_injection(vcpu);
6298 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6302 if (!kvm_x86_ops->update_cr8_intercept)
6305 if (!lapic_in_kernel(vcpu))
6308 if (vcpu->arch.apicv_active)
6311 if (!vcpu->arch.apic->vapic_addr)
6312 max_irr = kvm_lapic_find_highest_irr(vcpu);
6319 tpr = kvm_lapic_get_cr8(vcpu);
6321 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6324 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6328 /* try to reinject previous events if any */
6329 if (vcpu->arch.exception.pending) {
6330 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6331 vcpu->arch.exception.has_error_code,
6332 vcpu->arch.exception.error_code);
6334 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6335 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6338 if (vcpu->arch.exception.nr == DB_VECTOR &&
6339 (vcpu->arch.dr7 & DR7_GD)) {
6340 vcpu->arch.dr7 &= ~DR7_GD;
6341 kvm_update_dr7(vcpu);
6344 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
6345 vcpu->arch.exception.has_error_code,
6346 vcpu->arch.exception.error_code,
6347 vcpu->arch.exception.reinject);
6351 if (vcpu->arch.nmi_injected) {
6352 kvm_x86_ops->set_nmi(vcpu);
6356 if (vcpu->arch.interrupt.pending) {
6357 kvm_x86_ops->set_irq(vcpu);
6361 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6362 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6367 /* try to inject new event if pending */
6368 if (vcpu->arch.smi_pending && !is_smm(vcpu)) {
6369 vcpu->arch.smi_pending = false;
6371 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6372 --vcpu->arch.nmi_pending;
6373 vcpu->arch.nmi_injected = true;
6374 kvm_x86_ops->set_nmi(vcpu);
6375 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6377 * Because interrupts can be injected asynchronously, we are
6378 * calling check_nested_events again here to avoid a race condition.
6379 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6380 * proposal and current concerns. Perhaps we should be setting
6381 * KVM_REQ_EVENT only on certain events and not unconditionally?
6383 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6384 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6388 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6389 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6391 kvm_x86_ops->set_irq(vcpu);
6398 static void process_nmi(struct kvm_vcpu *vcpu)
6403 * x86 is limited to one NMI running, and one NMI pending after it.
6404 * If an NMI is already in progress, limit further NMIs to just one.
6405 * Otherwise, allow two (and we'll inject the first one immediately).
6407 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6410 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6411 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6412 kvm_make_request(KVM_REQ_EVENT, vcpu);
6415 #define put_smstate(type, buf, offset, val) \
6416 *(type *)((buf) + (offset) - 0x7e00) = val
6418 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
6421 flags |= seg->g << 23;
6422 flags |= seg->db << 22;
6423 flags |= seg->l << 21;
6424 flags |= seg->avl << 20;
6425 flags |= seg->present << 15;
6426 flags |= seg->dpl << 13;
6427 flags |= seg->s << 12;
6428 flags |= seg->type << 8;
6432 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6434 struct kvm_segment seg;
6437 kvm_get_segment(vcpu, &seg, n);
6438 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6441 offset = 0x7f84 + n * 12;
6443 offset = 0x7f2c + (n - 3) * 12;
6445 put_smstate(u32, buf, offset + 8, seg.base);
6446 put_smstate(u32, buf, offset + 4, seg.limit);
6447 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
6450 #ifdef CONFIG_X86_64
6451 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6453 struct kvm_segment seg;
6457 kvm_get_segment(vcpu, &seg, n);
6458 offset = 0x7e00 + n * 16;
6460 flags = enter_smm_get_segment_flags(&seg) >> 8;
6461 put_smstate(u16, buf, offset, seg.selector);
6462 put_smstate(u16, buf, offset + 2, flags);
6463 put_smstate(u32, buf, offset + 4, seg.limit);
6464 put_smstate(u64, buf, offset + 8, seg.base);
6468 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6471 struct kvm_segment seg;
6475 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6476 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6477 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6478 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6480 for (i = 0; i < 8; i++)
6481 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6483 kvm_get_dr(vcpu, 6, &val);
6484 put_smstate(u32, buf, 0x7fcc, (u32)val);
6485 kvm_get_dr(vcpu, 7, &val);
6486 put_smstate(u32, buf, 0x7fc8, (u32)val);
6488 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6489 put_smstate(u32, buf, 0x7fc4, seg.selector);
6490 put_smstate(u32, buf, 0x7f64, seg.base);
6491 put_smstate(u32, buf, 0x7f60, seg.limit);
6492 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
6494 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6495 put_smstate(u32, buf, 0x7fc0, seg.selector);
6496 put_smstate(u32, buf, 0x7f80, seg.base);
6497 put_smstate(u32, buf, 0x7f7c, seg.limit);
6498 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
6500 kvm_x86_ops->get_gdt(vcpu, &dt);
6501 put_smstate(u32, buf, 0x7f74, dt.address);
6502 put_smstate(u32, buf, 0x7f70, dt.size);
6504 kvm_x86_ops->get_idt(vcpu, &dt);
6505 put_smstate(u32, buf, 0x7f58, dt.address);
6506 put_smstate(u32, buf, 0x7f54, dt.size);
6508 for (i = 0; i < 6; i++)
6509 enter_smm_save_seg_32(vcpu, buf, i);
6511 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6514 put_smstate(u32, buf, 0x7efc, 0x00020000);
6515 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6518 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6520 #ifdef CONFIG_X86_64
6522 struct kvm_segment seg;
6526 for (i = 0; i < 16; i++)
6527 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6529 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6530 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6532 kvm_get_dr(vcpu, 6, &val);
6533 put_smstate(u64, buf, 0x7f68, val);
6534 kvm_get_dr(vcpu, 7, &val);
6535 put_smstate(u64, buf, 0x7f60, val);
6537 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6538 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6539 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6541 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6544 put_smstate(u32, buf, 0x7efc, 0x00020064);
6546 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6548 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6549 put_smstate(u16, buf, 0x7e90, seg.selector);
6550 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
6551 put_smstate(u32, buf, 0x7e94, seg.limit);
6552 put_smstate(u64, buf, 0x7e98, seg.base);
6554 kvm_x86_ops->get_idt(vcpu, &dt);
6555 put_smstate(u32, buf, 0x7e84, dt.size);
6556 put_smstate(u64, buf, 0x7e88, dt.address);
6558 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6559 put_smstate(u16, buf, 0x7e70, seg.selector);
6560 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
6561 put_smstate(u32, buf, 0x7e74, seg.limit);
6562 put_smstate(u64, buf, 0x7e78, seg.base);
6564 kvm_x86_ops->get_gdt(vcpu, &dt);
6565 put_smstate(u32, buf, 0x7e64, dt.size);
6566 put_smstate(u64, buf, 0x7e68, dt.address);
6568 for (i = 0; i < 6; i++)
6569 enter_smm_save_seg_64(vcpu, buf, i);
6575 static void enter_smm(struct kvm_vcpu *vcpu)
6577 struct kvm_segment cs, ds;
6582 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6583 vcpu->arch.hflags |= HF_SMM_MASK;
6584 memset(buf, 0, 512);
6585 if (guest_cpuid_has_longmode(vcpu))
6586 enter_smm_save_state_64(vcpu, buf);
6588 enter_smm_save_state_32(vcpu, buf);
6590 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6592 if (kvm_x86_ops->get_nmi_mask(vcpu))
6593 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6595 kvm_x86_ops->set_nmi_mask(vcpu, true);
6597 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6598 kvm_rip_write(vcpu, 0x8000);
6600 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6601 kvm_x86_ops->set_cr0(vcpu, cr0);
6602 vcpu->arch.cr0 = cr0;
6604 kvm_x86_ops->set_cr4(vcpu, 0);
6606 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6607 dt.address = dt.size = 0;
6608 kvm_x86_ops->set_idt(vcpu, &dt);
6610 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6612 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6613 cs.base = vcpu->arch.smbase;
6618 cs.limit = ds.limit = 0xffffffff;
6619 cs.type = ds.type = 0x3;
6620 cs.dpl = ds.dpl = 0;
6625 cs.avl = ds.avl = 0;
6626 cs.present = ds.present = 1;
6627 cs.unusable = ds.unusable = 0;
6628 cs.padding = ds.padding = 0;
6630 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6631 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6632 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6633 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6634 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6635 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6637 if (guest_cpuid_has_longmode(vcpu))
6638 kvm_x86_ops->set_efer(vcpu, 0);
6640 kvm_update_cpuid(vcpu);
6641 kvm_mmu_reset_context(vcpu);
6644 static void process_smi(struct kvm_vcpu *vcpu)
6646 vcpu->arch.smi_pending = true;
6647 kvm_make_request(KVM_REQ_EVENT, vcpu);
6650 void kvm_make_scan_ioapic_request(struct kvm *kvm)
6652 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6655 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6657 u64 eoi_exit_bitmap[4];
6659 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6662 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
6664 if (irqchip_split(vcpu->kvm))
6665 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
6667 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
6668 kvm_x86_ops->sync_pir_to_irr(vcpu);
6669 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
6671 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6672 vcpu_to_synic(vcpu)->vec_bitmap, 256);
6673 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6676 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6678 ++vcpu->stat.tlb_flush;
6679 kvm_x86_ops->tlb_flush(vcpu);
6682 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6684 struct page *page = NULL;
6686 if (!lapic_in_kernel(vcpu))
6689 if (!kvm_x86_ops->set_apic_access_page_addr)
6692 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6693 if (is_error_page(page))
6695 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6698 * Do not pin apic access page in memory, the MMU notifier
6699 * will call us again if it is migrated or swapped out.
6703 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6705 void kvm_arch_mmu_notifier_invalidate_page(struct kvm *kvm,
6706 unsigned long address)
6709 * The physical address of apic access page is stored in the VMCS.
6710 * Update it when it becomes invalid.
6712 if (address == gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT))
6713 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6717 * Returns 1 to let vcpu_run() continue the guest execution loop without
6718 * exiting to the userspace. Otherwise, the value will be returned to the
6721 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6725 dm_request_for_irq_injection(vcpu) &&
6726 kvm_cpu_accept_dm_intr(vcpu);
6728 bool req_immediate_exit = false;
6730 if (vcpu->requests) {
6731 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6732 kvm_mmu_unload(vcpu);
6733 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6734 __kvm_migrate_timers(vcpu);
6735 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6736 kvm_gen_update_masterclock(vcpu->kvm);
6737 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6738 kvm_gen_kvmclock_update(vcpu);
6739 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6740 r = kvm_guest_time_update(vcpu);
6744 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6745 kvm_mmu_sync_roots(vcpu);
6746 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6747 kvm_vcpu_flush_tlb(vcpu);
6748 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6749 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6753 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6754 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6758 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6759 /* Page is swapped out. Do synthetic halt */
6760 vcpu->arch.apf.halted = true;
6764 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6765 record_steal_time(vcpu);
6766 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6768 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6770 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6771 kvm_pmu_handle_event(vcpu);
6772 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6773 kvm_pmu_deliver_pmi(vcpu);
6774 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6775 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6776 if (test_bit(vcpu->arch.pending_ioapic_eoi,
6777 vcpu->arch.ioapic_handled_vectors)) {
6778 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6779 vcpu->run->eoi.vector =
6780 vcpu->arch.pending_ioapic_eoi;
6785 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
6786 vcpu_scan_ioapic(vcpu);
6787 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
6788 kvm_vcpu_reload_apic_access_page(vcpu);
6789 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
6790 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6791 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
6795 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
6796 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
6797 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
6801 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
6802 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
6803 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
6809 * KVM_REQ_HV_STIMER has to be processed after
6810 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
6811 * depend on the guest clock being up-to-date
6813 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
6814 kvm_hv_process_stimers(vcpu);
6817 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
6818 ++vcpu->stat.req_event;
6819 kvm_apic_accept_events(vcpu);
6820 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
6825 if (inject_pending_event(vcpu, req_int_win) != 0)
6826 req_immediate_exit = true;
6828 /* Enable NMI/IRQ window open exits if needed.
6830 * SMIs have two cases: 1) they can be nested, and
6831 * then there is nothing to do here because RSM will
6832 * cause a vmexit anyway; 2) or the SMI can be pending
6833 * because inject_pending_event has completed the
6834 * injection of an IRQ or NMI from the previous vmexit,
6835 * and then we request an immediate exit to inject the SMI.
6837 if (vcpu->arch.smi_pending && !is_smm(vcpu))
6838 req_immediate_exit = true;
6839 if (vcpu->arch.nmi_pending)
6840 kvm_x86_ops->enable_nmi_window(vcpu);
6841 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
6842 kvm_x86_ops->enable_irq_window(vcpu);
6845 if (kvm_lapic_enabled(vcpu)) {
6846 update_cr8_intercept(vcpu);
6847 kvm_lapic_sync_to_vapic(vcpu);
6851 r = kvm_mmu_reload(vcpu);
6853 goto cancel_injection;
6858 kvm_x86_ops->prepare_guest_switch(vcpu);
6859 kvm_load_guest_fpu(vcpu);
6862 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
6863 * IPI are then delayed after guest entry, which ensures that they
6864 * result in virtual interrupt delivery.
6866 local_irq_disable();
6867 vcpu->mode = IN_GUEST_MODE;
6869 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
6872 * 1) We should set ->mode before checking ->requests. Please see
6873 * the comment in kvm_make_all_cpus_request.
6875 * 2) For APICv, we should set ->mode before checking PIR.ON. This
6876 * pairs with the memory barrier implicit in pi_test_and_set_on
6877 * (see vmx_deliver_posted_interrupt).
6879 * 3) This also orders the write to mode from any reads to the page
6880 * tables done while the VCPU is running. Please see the comment
6881 * in kvm_flush_remote_tlbs.
6883 smp_mb__after_srcu_read_unlock();
6886 * This handles the case where a posted interrupt was
6887 * notified with kvm_vcpu_kick.
6889 if (kvm_lapic_enabled(vcpu)) {
6890 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
6891 kvm_x86_ops->sync_pir_to_irr(vcpu);
6894 if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
6895 || need_resched() || signal_pending(current)) {
6896 vcpu->mode = OUTSIDE_GUEST_MODE;
6900 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6902 goto cancel_injection;
6905 kvm_load_guest_xcr0(vcpu);
6907 if (req_immediate_exit) {
6908 kvm_make_request(KVM_REQ_EVENT, vcpu);
6909 smp_send_reschedule(vcpu->cpu);
6912 trace_kvm_entry(vcpu->vcpu_id);
6913 wait_lapic_expire(vcpu);
6914 guest_enter_irqoff();
6916 if (unlikely(vcpu->arch.switch_db_regs)) {
6918 set_debugreg(vcpu->arch.eff_db[0], 0);
6919 set_debugreg(vcpu->arch.eff_db[1], 1);
6920 set_debugreg(vcpu->arch.eff_db[2], 2);
6921 set_debugreg(vcpu->arch.eff_db[3], 3);
6922 set_debugreg(vcpu->arch.dr6, 6);
6923 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6926 kvm_x86_ops->run(vcpu);
6929 * Do this here before restoring debug registers on the host. And
6930 * since we do this before handling the vmexit, a DR access vmexit
6931 * can (a) read the correct value of the debug registers, (b) set
6932 * KVM_DEBUGREG_WONT_EXIT again.
6934 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
6935 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
6936 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
6937 kvm_update_dr0123(vcpu);
6938 kvm_update_dr6(vcpu);
6939 kvm_update_dr7(vcpu);
6940 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
6944 * If the guest has used debug registers, at least dr7
6945 * will be disabled while returning to the host.
6946 * If we don't have active breakpoints in the host, we don't
6947 * care about the messed up debug address registers. But if
6948 * we have some of them active, restore the old state.
6950 if (hw_breakpoint_active())
6951 hw_breakpoint_restore();
6953 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
6955 vcpu->mode = OUTSIDE_GUEST_MODE;
6958 kvm_put_guest_xcr0(vcpu);
6960 kvm_x86_ops->handle_external_intr(vcpu);
6964 guest_exit_irqoff();
6969 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
6972 * Profile KVM exit RIPs:
6974 if (unlikely(prof_on == KVM_PROFILING)) {
6975 unsigned long rip = kvm_rip_read(vcpu);
6976 profile_hit(KVM_PROFILING, (void *)rip);
6979 if (unlikely(vcpu->arch.tsc_always_catchup))
6980 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6982 if (vcpu->arch.apic_attention)
6983 kvm_lapic_sync_from_vapic(vcpu);
6985 r = kvm_x86_ops->handle_exit(vcpu);
6989 kvm_x86_ops->cancel_injection(vcpu);
6990 if (unlikely(vcpu->arch.apic_attention))
6991 kvm_lapic_sync_from_vapic(vcpu);
6996 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
6998 if (!kvm_arch_vcpu_runnable(vcpu) &&
6999 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7000 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7001 kvm_vcpu_block(vcpu);
7002 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7004 if (kvm_x86_ops->post_block)
7005 kvm_x86_ops->post_block(vcpu);
7007 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7011 kvm_apic_accept_events(vcpu);
7012 switch(vcpu->arch.mp_state) {
7013 case KVM_MP_STATE_HALTED:
7014 vcpu->arch.pv.pv_unhalted = false;
7015 vcpu->arch.mp_state =
7016 KVM_MP_STATE_RUNNABLE;
7017 case KVM_MP_STATE_RUNNABLE:
7018 vcpu->arch.apf.halted = false;
7020 case KVM_MP_STATE_INIT_RECEIVED:
7029 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7031 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7032 kvm_x86_ops->check_nested_events(vcpu, false);
7034 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7035 !vcpu->arch.apf.halted);
7038 static int vcpu_run(struct kvm_vcpu *vcpu)
7041 struct kvm *kvm = vcpu->kvm;
7043 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7046 if (kvm_vcpu_running(vcpu)) {
7047 r = vcpu_enter_guest(vcpu);
7049 r = vcpu_block(kvm, vcpu);
7055 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
7056 if (kvm_cpu_has_pending_timer(vcpu))
7057 kvm_inject_pending_timer_irqs(vcpu);
7059 if (dm_request_for_irq_injection(vcpu) &&
7060 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
7062 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
7063 ++vcpu->stat.request_irq_exits;
7067 kvm_check_async_pf_completion(vcpu);
7069 if (signal_pending(current)) {
7071 vcpu->run->exit_reason = KVM_EXIT_INTR;
7072 ++vcpu->stat.signal_exits;
7075 if (need_resched()) {
7076 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7078 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7082 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7087 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7090 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7091 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7092 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7093 if (r != EMULATE_DONE)
7098 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7100 BUG_ON(!vcpu->arch.pio.count);
7102 return complete_emulated_io(vcpu);
7106 * Implements the following, as a state machine:
7110 * for each mmio piece in the fragment
7118 * for each mmio piece in the fragment
7123 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
7125 struct kvm_run *run = vcpu->run;
7126 struct kvm_mmio_fragment *frag;
7129 BUG_ON(!vcpu->mmio_needed);
7131 /* Complete previous fragment */
7132 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7133 len = min(8u, frag->len);
7134 if (!vcpu->mmio_is_write)
7135 memcpy(frag->data, run->mmio.data, len);
7137 if (frag->len <= 8) {
7138 /* Switch to the next fragment. */
7140 vcpu->mmio_cur_fragment++;
7142 /* Go forward to the next mmio piece. */
7148 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
7149 vcpu->mmio_needed = 0;
7151 /* FIXME: return into emulator if single-stepping. */
7152 if (vcpu->mmio_is_write)
7154 vcpu->mmio_read_completed = 1;
7155 return complete_emulated_io(vcpu);
7158 run->exit_reason = KVM_EXIT_MMIO;
7159 run->mmio.phys_addr = frag->gpa;
7160 if (vcpu->mmio_is_write)
7161 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7162 run->mmio.len = min(8u, frag->len);
7163 run->mmio.is_write = vcpu->mmio_is_write;
7164 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7169 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7171 struct fpu *fpu = ¤t->thread.fpu;
7175 fpu__activate_curr(fpu);
7177 if (vcpu->sigset_active)
7178 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
7180 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
7181 kvm_vcpu_block(vcpu);
7182 kvm_apic_accept_events(vcpu);
7183 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
7188 /* re-sync apic's tpr */
7189 if (!lapic_in_kernel(vcpu)) {
7190 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7196 if (unlikely(vcpu->arch.complete_userspace_io)) {
7197 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7198 vcpu->arch.complete_userspace_io = NULL;
7203 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
7205 if (kvm_run->immediate_exit)
7211 post_kvm_run_save(vcpu);
7212 if (vcpu->sigset_active)
7213 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
7218 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7220 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
7222 * We are here if userspace calls get_regs() in the middle of
7223 * instruction emulation. Registers state needs to be copied
7224 * back from emulation context to vcpu. Userspace shouldn't do
7225 * that usually, but some bad designed PV devices (vmware
7226 * backdoor interface) need this to work
7228 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7229 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7231 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
7232 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
7233 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
7234 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
7235 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
7236 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
7237 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7238 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
7239 #ifdef CONFIG_X86_64
7240 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7241 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7242 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7243 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7244 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7245 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7246 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7247 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
7250 regs->rip = kvm_rip_read(vcpu);
7251 regs->rflags = kvm_get_rflags(vcpu);
7256 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7258 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7259 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7261 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7262 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7263 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7264 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7265 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7266 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7267 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7268 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
7269 #ifdef CONFIG_X86_64
7270 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7271 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7272 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7273 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7274 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7275 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7276 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7277 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
7280 kvm_rip_write(vcpu, regs->rip);
7281 kvm_set_rflags(vcpu, regs->rflags);
7283 vcpu->arch.exception.pending = false;
7285 kvm_make_request(KVM_REQ_EVENT, vcpu);
7290 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7292 struct kvm_segment cs;
7294 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7298 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7300 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7301 struct kvm_sregs *sregs)
7305 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7306 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7307 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7308 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7309 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7310 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7312 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7313 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7315 kvm_x86_ops->get_idt(vcpu, &dt);
7316 sregs->idt.limit = dt.size;
7317 sregs->idt.base = dt.address;
7318 kvm_x86_ops->get_gdt(vcpu, &dt);
7319 sregs->gdt.limit = dt.size;
7320 sregs->gdt.base = dt.address;
7322 sregs->cr0 = kvm_read_cr0(vcpu);
7323 sregs->cr2 = vcpu->arch.cr2;
7324 sregs->cr3 = kvm_read_cr3(vcpu);
7325 sregs->cr4 = kvm_read_cr4(vcpu);
7326 sregs->cr8 = kvm_get_cr8(vcpu);
7327 sregs->efer = vcpu->arch.efer;
7328 sregs->apic_base = kvm_get_apic_base(vcpu);
7330 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7332 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
7333 set_bit(vcpu->arch.interrupt.nr,
7334 (unsigned long *)sregs->interrupt_bitmap);
7339 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7340 struct kvm_mp_state *mp_state)
7342 kvm_apic_accept_events(vcpu);
7343 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7344 vcpu->arch.pv.pv_unhalted)
7345 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7347 mp_state->mp_state = vcpu->arch.mp_state;
7352 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7353 struct kvm_mp_state *mp_state)
7355 if (!lapic_in_kernel(vcpu) &&
7356 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7359 /* INITs are latched while in SMM */
7360 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
7361 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
7362 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
7365 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7366 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7367 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7369 vcpu->arch.mp_state = mp_state->mp_state;
7370 kvm_make_request(KVM_REQ_EVENT, vcpu);
7374 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7375 int reason, bool has_error_code, u32 error_code)
7377 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7380 init_emulate_ctxt(vcpu);
7382 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7383 has_error_code, error_code);
7386 return EMULATE_FAIL;
7388 kvm_rip_write(vcpu, ctxt->eip);
7389 kvm_set_rflags(vcpu, ctxt->eflags);
7390 kvm_make_request(KVM_REQ_EVENT, vcpu);
7391 return EMULATE_DONE;
7393 EXPORT_SYMBOL_GPL(kvm_task_switch);
7395 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7396 struct kvm_sregs *sregs)
7398 struct msr_data apic_base_msr;
7399 int mmu_reset_needed = 0;
7400 int pending_vec, max_bits, idx;
7403 if (!guest_cpuid_has_xsave(vcpu) && (sregs->cr4 & X86_CR4_OSXSAVE))
7406 dt.size = sregs->idt.limit;
7407 dt.address = sregs->idt.base;
7408 kvm_x86_ops->set_idt(vcpu, &dt);
7409 dt.size = sregs->gdt.limit;
7410 dt.address = sregs->gdt.base;
7411 kvm_x86_ops->set_gdt(vcpu, &dt);
7413 vcpu->arch.cr2 = sregs->cr2;
7414 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7415 vcpu->arch.cr3 = sregs->cr3;
7416 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7418 kvm_set_cr8(vcpu, sregs->cr8);
7420 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7421 kvm_x86_ops->set_efer(vcpu, sregs->efer);
7422 apic_base_msr.data = sregs->apic_base;
7423 apic_base_msr.host_initiated = true;
7424 kvm_set_apic_base(vcpu, &apic_base_msr);
7426 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7427 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7428 vcpu->arch.cr0 = sregs->cr0;
7430 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7431 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7432 if (sregs->cr4 & (X86_CR4_OSXSAVE | X86_CR4_PKE))
7433 kvm_update_cpuid(vcpu);
7435 idx = srcu_read_lock(&vcpu->kvm->srcu);
7436 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7437 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7438 mmu_reset_needed = 1;
7440 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7442 if (mmu_reset_needed)
7443 kvm_mmu_reset_context(vcpu);
7445 max_bits = KVM_NR_INTERRUPTS;
7446 pending_vec = find_first_bit(
7447 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7448 if (pending_vec < max_bits) {
7449 kvm_queue_interrupt(vcpu, pending_vec, false);
7450 pr_debug("Set back pending irq %d\n", pending_vec);
7453 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7454 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7455 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7456 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7457 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7458 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7460 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7461 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7463 update_cr8_intercept(vcpu);
7465 /* Older userspace won't unhalt the vcpu on reset. */
7466 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7467 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7469 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7471 kvm_make_request(KVM_REQ_EVENT, vcpu);
7476 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7477 struct kvm_guest_debug *dbg)
7479 unsigned long rflags;
7482 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7484 if (vcpu->arch.exception.pending)
7486 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7487 kvm_queue_exception(vcpu, DB_VECTOR);
7489 kvm_queue_exception(vcpu, BP_VECTOR);
7493 * Read rflags as long as potentially injected trace flags are still
7496 rflags = kvm_get_rflags(vcpu);
7498 vcpu->guest_debug = dbg->control;
7499 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7500 vcpu->guest_debug = 0;
7502 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7503 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7504 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7505 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7507 for (i = 0; i < KVM_NR_DB_REGS; i++)
7508 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7510 kvm_update_dr7(vcpu);
7512 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7513 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7514 get_segment_base(vcpu, VCPU_SREG_CS);
7517 * Trigger an rflags update that will inject or remove the trace
7520 kvm_set_rflags(vcpu, rflags);
7522 kvm_x86_ops->update_bp_intercept(vcpu);
7532 * Translate a guest virtual address to a guest physical address.
7534 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7535 struct kvm_translation *tr)
7537 unsigned long vaddr = tr->linear_address;
7541 idx = srcu_read_lock(&vcpu->kvm->srcu);
7542 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7543 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7544 tr->physical_address = gpa;
7545 tr->valid = gpa != UNMAPPED_GVA;
7552 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7554 struct fxregs_state *fxsave =
7555 &vcpu->arch.guest_fpu.state.fxsave;
7557 memcpy(fpu->fpr, fxsave->st_space, 128);
7558 fpu->fcw = fxsave->cwd;
7559 fpu->fsw = fxsave->swd;
7560 fpu->ftwx = fxsave->twd;
7561 fpu->last_opcode = fxsave->fop;
7562 fpu->last_ip = fxsave->rip;
7563 fpu->last_dp = fxsave->rdp;
7564 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7569 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7571 struct fxregs_state *fxsave =
7572 &vcpu->arch.guest_fpu.state.fxsave;
7574 memcpy(fxsave->st_space, fpu->fpr, 128);
7575 fxsave->cwd = fpu->fcw;
7576 fxsave->swd = fpu->fsw;
7577 fxsave->twd = fpu->ftwx;
7578 fxsave->fop = fpu->last_opcode;
7579 fxsave->rip = fpu->last_ip;
7580 fxsave->rdp = fpu->last_dp;
7581 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7586 static void fx_init(struct kvm_vcpu *vcpu)
7588 fpstate_init(&vcpu->arch.guest_fpu.state);
7589 if (boot_cpu_has(X86_FEATURE_XSAVES))
7590 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7591 host_xcr0 | XSTATE_COMPACTION_ENABLED;
7594 * Ensure guest xcr0 is valid for loading
7596 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7598 vcpu->arch.cr0 |= X86_CR0_ET;
7601 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7603 if (vcpu->guest_fpu_loaded)
7607 * Restore all possible states in the guest,
7608 * and assume host would use all available bits.
7609 * Guest xcr0 would be loaded later.
7611 vcpu->guest_fpu_loaded = 1;
7612 __kernel_fpu_begin();
7613 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state);
7617 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7619 if (!vcpu->guest_fpu_loaded)
7622 vcpu->guest_fpu_loaded = 0;
7623 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7625 ++vcpu->stat.fpu_reload;
7629 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7631 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
7633 kvmclock_reset(vcpu);
7635 kvm_x86_ops->vcpu_free(vcpu);
7636 free_cpumask_var(wbinvd_dirty_mask);
7639 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7642 struct kvm_vcpu *vcpu;
7644 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7645 printk_once(KERN_WARNING
7646 "kvm: SMP vm created on host with unstable TSC; "
7647 "guest TSC will not be reliable\n");
7649 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7654 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7658 kvm_vcpu_mtrr_init(vcpu);
7659 r = vcpu_load(vcpu);
7662 kvm_vcpu_reset(vcpu, false);
7663 kvm_mmu_setup(vcpu);
7668 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7670 struct msr_data msr;
7671 struct kvm *kvm = vcpu->kvm;
7673 if (vcpu_load(vcpu))
7676 msr.index = MSR_IA32_TSC;
7677 msr.host_initiated = true;
7678 kvm_write_tsc(vcpu, &msr);
7681 if (!kvmclock_periodic_sync)
7684 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7685 KVMCLOCK_SYNC_PERIOD);
7688 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7691 vcpu->arch.apf.msr_val = 0;
7693 r = vcpu_load(vcpu);
7695 kvm_mmu_unload(vcpu);
7698 kvm_x86_ops->vcpu_free(vcpu);
7701 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7703 vcpu->arch.hflags = 0;
7705 vcpu->arch.smi_pending = 0;
7706 atomic_set(&vcpu->arch.nmi_queued, 0);
7707 vcpu->arch.nmi_pending = 0;
7708 vcpu->arch.nmi_injected = false;
7709 kvm_clear_interrupt_queue(vcpu);
7710 kvm_clear_exception_queue(vcpu);
7712 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7713 kvm_update_dr0123(vcpu);
7714 vcpu->arch.dr6 = DR6_INIT;
7715 kvm_update_dr6(vcpu);
7716 vcpu->arch.dr7 = DR7_FIXED_1;
7717 kvm_update_dr7(vcpu);
7721 kvm_make_request(KVM_REQ_EVENT, vcpu);
7722 vcpu->arch.apf.msr_val = 0;
7723 vcpu->arch.st.msr_val = 0;
7725 kvmclock_reset(vcpu);
7727 kvm_clear_async_pf_completion_queue(vcpu);
7728 kvm_async_pf_hash_reset(vcpu);
7729 vcpu->arch.apf.halted = false;
7732 kvm_pmu_reset(vcpu);
7733 vcpu->arch.smbase = 0x30000;
7736 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
7737 vcpu->arch.regs_avail = ~0;
7738 vcpu->arch.regs_dirty = ~0;
7740 kvm_x86_ops->vcpu_reset(vcpu, init_event);
7743 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
7745 struct kvm_segment cs;
7747 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7748 cs.selector = vector << 8;
7749 cs.base = vector << 12;
7750 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7751 kvm_rip_write(vcpu, 0);
7754 int kvm_arch_hardware_enable(void)
7757 struct kvm_vcpu *vcpu;
7762 bool stable, backwards_tsc = false;
7764 kvm_shared_msr_cpu_online();
7765 ret = kvm_x86_ops->hardware_enable();
7769 local_tsc = rdtsc();
7770 stable = !check_tsc_unstable();
7771 list_for_each_entry(kvm, &vm_list, vm_list) {
7772 kvm_for_each_vcpu(i, vcpu, kvm) {
7773 if (!stable && vcpu->cpu == smp_processor_id())
7774 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7775 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
7776 backwards_tsc = true;
7777 if (vcpu->arch.last_host_tsc > max_tsc)
7778 max_tsc = vcpu->arch.last_host_tsc;
7784 * Sometimes, even reliable TSCs go backwards. This happens on
7785 * platforms that reset TSC during suspend or hibernate actions, but
7786 * maintain synchronization. We must compensate. Fortunately, we can
7787 * detect that condition here, which happens early in CPU bringup,
7788 * before any KVM threads can be running. Unfortunately, we can't
7789 * bring the TSCs fully up to date with real time, as we aren't yet far
7790 * enough into CPU bringup that we know how much real time has actually
7791 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
7792 * variables that haven't been updated yet.
7794 * So we simply find the maximum observed TSC above, then record the
7795 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
7796 * the adjustment will be applied. Note that we accumulate
7797 * adjustments, in case multiple suspend cycles happen before some VCPU
7798 * gets a chance to run again. In the event that no KVM threads get a
7799 * chance to run, we will miss the entire elapsed period, as we'll have
7800 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
7801 * loose cycle time. This isn't too big a deal, since the loss will be
7802 * uniform across all VCPUs (not to mention the scenario is extremely
7803 * unlikely). It is possible that a second hibernate recovery happens
7804 * much faster than a first, causing the observed TSC here to be
7805 * smaller; this would require additional padding adjustment, which is
7806 * why we set last_host_tsc to the local tsc observed here.
7808 * N.B. - this code below runs only on platforms with reliable TSC,
7809 * as that is the only way backwards_tsc is set above. Also note
7810 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
7811 * have the same delta_cyc adjustment applied if backwards_tsc
7812 * is detected. Note further, this adjustment is only done once,
7813 * as we reset last_host_tsc on all VCPUs to stop this from being
7814 * called multiple times (one for each physical CPU bringup).
7816 * Platforms with unreliable TSCs don't have to deal with this, they
7817 * will be compensated by the logic in vcpu_load, which sets the TSC to
7818 * catchup mode. This will catchup all VCPUs to real time, but cannot
7819 * guarantee that they stay in perfect synchronization.
7821 if (backwards_tsc) {
7822 u64 delta_cyc = max_tsc - local_tsc;
7823 backwards_tsc_observed = true;
7824 list_for_each_entry(kvm, &vm_list, vm_list) {
7825 kvm_for_each_vcpu(i, vcpu, kvm) {
7826 vcpu->arch.tsc_offset_adjustment += delta_cyc;
7827 vcpu->arch.last_host_tsc = local_tsc;
7828 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7832 * We have to disable TSC offset matching.. if you were
7833 * booting a VM while issuing an S4 host suspend....
7834 * you may have some problem. Solving this issue is
7835 * left as an exercise to the reader.
7837 kvm->arch.last_tsc_nsec = 0;
7838 kvm->arch.last_tsc_write = 0;
7845 void kvm_arch_hardware_disable(void)
7847 kvm_x86_ops->hardware_disable();
7848 drop_user_return_notifiers();
7851 int kvm_arch_hardware_setup(void)
7855 r = kvm_x86_ops->hardware_setup();
7859 if (kvm_has_tsc_control) {
7861 * Make sure the user can only configure tsc_khz values that
7862 * fit into a signed integer.
7863 * A min value is not calculated needed because it will always
7864 * be 1 on all machines.
7866 u64 max = min(0x7fffffffULL,
7867 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
7868 kvm_max_guest_tsc_khz = max;
7870 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
7873 kvm_init_msr_list();
7877 void kvm_arch_hardware_unsetup(void)
7879 kvm_x86_ops->hardware_unsetup();
7882 void kvm_arch_check_processor_compat(void *rtn)
7884 kvm_x86_ops->check_processor_compatibility(rtn);
7887 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
7889 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
7891 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
7893 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
7895 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
7898 struct static_key kvm_no_apic_vcpu __read_mostly;
7899 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
7901 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
7907 BUG_ON(vcpu->kvm == NULL);
7910 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv();
7911 vcpu->arch.pv.pv_unhalted = false;
7912 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
7913 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_reset_bsp(vcpu))
7914 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7916 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
7918 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
7923 vcpu->arch.pio_data = page_address(page);
7925 kvm_set_tsc_khz(vcpu, max_tsc_khz);
7927 r = kvm_mmu_create(vcpu);
7929 goto fail_free_pio_data;
7931 if (irqchip_in_kernel(kvm)) {
7932 r = kvm_create_lapic(vcpu);
7934 goto fail_mmu_destroy;
7936 static_key_slow_inc(&kvm_no_apic_vcpu);
7938 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
7940 if (!vcpu->arch.mce_banks) {
7942 goto fail_free_lapic;
7944 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
7946 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
7948 goto fail_free_mce_banks;
7953 vcpu->arch.ia32_tsc_adjust_msr = 0x0;
7954 vcpu->arch.pv_time_enabled = false;
7956 vcpu->arch.guest_supported_xcr0 = 0;
7957 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
7959 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
7961 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
7963 kvm_async_pf_hash_reset(vcpu);
7966 vcpu->arch.pending_external_vector = -1;
7968 kvm_hv_vcpu_init(vcpu);
7972 fail_free_mce_banks:
7973 kfree(vcpu->arch.mce_banks);
7975 kvm_free_lapic(vcpu);
7977 kvm_mmu_destroy(vcpu);
7979 free_page((unsigned long)vcpu->arch.pio_data);
7984 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
7988 kvm_hv_vcpu_uninit(vcpu);
7989 kvm_pmu_destroy(vcpu);
7990 kfree(vcpu->arch.mce_banks);
7991 kvm_free_lapic(vcpu);
7992 idx = srcu_read_lock(&vcpu->kvm->srcu);
7993 kvm_mmu_destroy(vcpu);
7994 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7995 free_page((unsigned long)vcpu->arch.pio_data);
7996 if (!lapic_in_kernel(vcpu))
7997 static_key_slow_dec(&kvm_no_apic_vcpu);
8000 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
8002 kvm_x86_ops->sched_in(vcpu, cpu);
8005 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
8010 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
8011 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
8012 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
8013 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
8014 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
8016 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
8017 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
8018 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
8019 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
8020 &kvm->arch.irq_sources_bitmap);
8022 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
8023 mutex_init(&kvm->arch.apic_map_lock);
8024 mutex_init(&kvm->arch.hyperv.hv_lock);
8025 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
8027 kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
8028 pvclock_update_vm_gtod_copy(kvm);
8030 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
8031 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
8033 kvm_page_track_init(kvm);
8034 kvm_mmu_init_vm(kvm);
8036 if (kvm_x86_ops->vm_init)
8037 return kvm_x86_ops->vm_init(kvm);
8042 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
8045 r = vcpu_load(vcpu);
8047 kvm_mmu_unload(vcpu);
8051 static void kvm_free_vcpus(struct kvm *kvm)
8054 struct kvm_vcpu *vcpu;
8057 * Unpin any mmu pages first.
8059 kvm_for_each_vcpu(i, vcpu, kvm) {
8060 kvm_clear_async_pf_completion_queue(vcpu);
8061 kvm_unload_vcpu_mmu(vcpu);
8063 kvm_for_each_vcpu(i, vcpu, kvm)
8064 kvm_arch_vcpu_free(vcpu);
8066 mutex_lock(&kvm->lock);
8067 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
8068 kvm->vcpus[i] = NULL;
8070 atomic_set(&kvm->online_vcpus, 0);
8071 mutex_unlock(&kvm->lock);
8074 void kvm_arch_sync_events(struct kvm *kvm)
8076 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
8077 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
8081 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8085 struct kvm_memslots *slots = kvm_memslots(kvm);
8086 struct kvm_memory_slot *slot, old;
8088 /* Called with kvm->slots_lock held. */
8089 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
8092 slot = id_to_memslot(slots, id);
8098 * MAP_SHARED to prevent internal slot pages from being moved
8101 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
8102 MAP_SHARED | MAP_ANONYMOUS, 0);
8103 if (IS_ERR((void *)hva))
8104 return PTR_ERR((void *)hva);
8113 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
8114 struct kvm_userspace_memory_region m;
8116 m.slot = id | (i << 16);
8118 m.guest_phys_addr = gpa;
8119 m.userspace_addr = hva;
8120 m.memory_size = size;
8121 r = __kvm_set_memory_region(kvm, &m);
8127 r = vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
8133 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
8135 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8139 mutex_lock(&kvm->slots_lock);
8140 r = __x86_set_memory_region(kvm, id, gpa, size);
8141 mutex_unlock(&kvm->slots_lock);
8145 EXPORT_SYMBOL_GPL(x86_set_memory_region);
8147 void kvm_arch_destroy_vm(struct kvm *kvm)
8149 if (current->mm == kvm->mm) {
8151 * Free memory regions allocated on behalf of userspace,
8152 * unless the the memory map has changed due to process exit
8155 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
8156 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
8157 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
8159 if (kvm_x86_ops->vm_destroy)
8160 kvm_x86_ops->vm_destroy(kvm);
8161 kvm_pic_destroy(kvm);
8162 kvm_ioapic_destroy(kvm);
8163 kvm_free_vcpus(kvm);
8164 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
8165 kvm_mmu_uninit_vm(kvm);
8166 kvm_page_track_cleanup(kvm);
8169 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
8170 struct kvm_memory_slot *dont)
8174 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8175 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
8176 kvfree(free->arch.rmap[i]);
8177 free->arch.rmap[i] = NULL;
8182 if (!dont || free->arch.lpage_info[i - 1] !=
8183 dont->arch.lpage_info[i - 1]) {
8184 kvfree(free->arch.lpage_info[i - 1]);
8185 free->arch.lpage_info[i - 1] = NULL;
8189 kvm_page_track_free_memslot(free, dont);
8192 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
8193 unsigned long npages)
8197 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8198 struct kvm_lpage_info *linfo;
8203 lpages = gfn_to_index(slot->base_gfn + npages - 1,
8204 slot->base_gfn, level) + 1;
8206 slot->arch.rmap[i] =
8207 kvm_kvzalloc(lpages * sizeof(*slot->arch.rmap[i]));
8208 if (!slot->arch.rmap[i])
8213 linfo = kvm_kvzalloc(lpages * sizeof(*linfo));
8217 slot->arch.lpage_info[i - 1] = linfo;
8219 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
8220 linfo[0].disallow_lpage = 1;
8221 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
8222 linfo[lpages - 1].disallow_lpage = 1;
8223 ugfn = slot->userspace_addr >> PAGE_SHIFT;
8225 * If the gfn and userspace address are not aligned wrt each
8226 * other, or if explicitly asked to, disable large page
8227 * support for this slot
8229 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
8230 !kvm_largepages_enabled()) {
8233 for (j = 0; j < lpages; ++j)
8234 linfo[j].disallow_lpage = 1;
8238 if (kvm_page_track_create_memslot(slot, npages))
8244 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8245 kvfree(slot->arch.rmap[i]);
8246 slot->arch.rmap[i] = NULL;
8250 kvfree(slot->arch.lpage_info[i - 1]);
8251 slot->arch.lpage_info[i - 1] = NULL;
8256 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
8259 * memslots->generation has been incremented.
8260 * mmio generation may have reached its maximum value.
8262 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
8265 int kvm_arch_prepare_memory_region(struct kvm *kvm,
8266 struct kvm_memory_slot *memslot,
8267 const struct kvm_userspace_memory_region *mem,
8268 enum kvm_mr_change change)
8273 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8274 struct kvm_memory_slot *new)
8276 /* Still write protect RO slot */
8277 if (new->flags & KVM_MEM_READONLY) {
8278 kvm_mmu_slot_remove_write_access(kvm, new);
8283 * Call kvm_x86_ops dirty logging hooks when they are valid.
8285 * kvm_x86_ops->slot_disable_log_dirty is called when:
8287 * - KVM_MR_CREATE with dirty logging is disabled
8288 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8290 * The reason is, in case of PML, we need to set D-bit for any slots
8291 * with dirty logging disabled in order to eliminate unnecessary GPA
8292 * logging in PML buffer (and potential PML buffer full VMEXT). This
8293 * guarantees leaving PML enabled during guest's lifetime won't have
8294 * any additonal overhead from PML when guest is running with dirty
8295 * logging disabled for memory slots.
8297 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8298 * to dirty logging mode.
8300 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8302 * In case of write protect:
8304 * Write protect all pages for dirty logging.
8306 * All the sptes including the large sptes which point to this
8307 * slot are set to readonly. We can not create any new large
8308 * spte on this slot until the end of the logging.
8310 * See the comments in fast_page_fault().
8312 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8313 if (kvm_x86_ops->slot_enable_log_dirty)
8314 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8316 kvm_mmu_slot_remove_write_access(kvm, new);
8318 if (kvm_x86_ops->slot_disable_log_dirty)
8319 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8323 void kvm_arch_commit_memory_region(struct kvm *kvm,
8324 const struct kvm_userspace_memory_region *mem,
8325 const struct kvm_memory_slot *old,
8326 const struct kvm_memory_slot *new,
8327 enum kvm_mr_change change)
8329 int nr_mmu_pages = 0;
8331 if (!kvm->arch.n_requested_mmu_pages)
8332 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8335 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
8338 * Dirty logging tracks sptes in 4k granularity, meaning that large
8339 * sptes have to be split. If live migration is successful, the guest
8340 * in the source machine will be destroyed and large sptes will be
8341 * created in the destination. However, if the guest continues to run
8342 * in the source machine (for example if live migration fails), small
8343 * sptes will remain around and cause bad performance.
8345 * Scan sptes if dirty logging has been stopped, dropping those
8346 * which can be collapsed into a single large-page spte. Later
8347 * page faults will create the large-page sptes.
8349 if ((change != KVM_MR_DELETE) &&
8350 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8351 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8352 kvm_mmu_zap_collapsible_sptes(kvm, new);
8355 * Set up write protection and/or dirty logging for the new slot.
8357 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8358 * been zapped so no dirty logging staff is needed for old slot. For
8359 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8360 * new and it's also covered when dealing with the new slot.
8362 * FIXME: const-ify all uses of struct kvm_memory_slot.
8364 if (change != KVM_MR_DELETE)
8365 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8368 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8370 kvm_mmu_invalidate_zap_all_pages(kvm);
8373 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8374 struct kvm_memory_slot *slot)
8376 kvm_page_track_flush_slot(kvm, slot);
8379 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8381 if (!list_empty_careful(&vcpu->async_pf.done))
8384 if (kvm_apic_has_events(vcpu))
8387 if (vcpu->arch.pv.pv_unhalted)
8390 if (atomic_read(&vcpu->arch.nmi_queued))
8393 if (test_bit(KVM_REQ_SMI, &vcpu->requests))
8396 if (kvm_arch_interrupt_allowed(vcpu) &&
8397 kvm_cpu_has_interrupt(vcpu))
8400 if (kvm_hv_has_stimer_pending(vcpu))
8406 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8408 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8411 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8413 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8416 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8418 return kvm_x86_ops->interrupt_allowed(vcpu);
8421 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8423 if (is_64_bit_mode(vcpu))
8424 return kvm_rip_read(vcpu);
8425 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8426 kvm_rip_read(vcpu));
8428 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8430 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8432 return kvm_get_linear_rip(vcpu) == linear_rip;
8434 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8436 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8438 unsigned long rflags;
8440 rflags = kvm_x86_ops->get_rflags(vcpu);
8441 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8442 rflags &= ~X86_EFLAGS_TF;
8445 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8447 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8449 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8450 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8451 rflags |= X86_EFLAGS_TF;
8452 kvm_x86_ops->set_rflags(vcpu, rflags);
8455 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8457 __kvm_set_rflags(vcpu, rflags);
8458 kvm_make_request(KVM_REQ_EVENT, vcpu);
8460 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8462 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8466 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8470 r = kvm_mmu_reload(vcpu);
8474 if (!vcpu->arch.mmu.direct_map &&
8475 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8478 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8481 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8483 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8486 static inline u32 kvm_async_pf_next_probe(u32 key)
8488 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8491 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8493 u32 key = kvm_async_pf_hash_fn(gfn);
8495 while (vcpu->arch.apf.gfns[key] != ~0)
8496 key = kvm_async_pf_next_probe(key);
8498 vcpu->arch.apf.gfns[key] = gfn;
8501 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8504 u32 key = kvm_async_pf_hash_fn(gfn);
8506 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8507 (vcpu->arch.apf.gfns[key] != gfn &&
8508 vcpu->arch.apf.gfns[key] != ~0); i++)
8509 key = kvm_async_pf_next_probe(key);
8514 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8516 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8519 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8523 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8525 vcpu->arch.apf.gfns[i] = ~0;
8527 j = kvm_async_pf_next_probe(j);
8528 if (vcpu->arch.apf.gfns[j] == ~0)
8530 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8532 * k lies cyclically in ]i,j]
8534 * |....j i.k.| or |.k..j i...|
8536 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8537 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8542 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8544 return kvm_vcpu_write_guest_cached(vcpu, &vcpu->arch.apf.data, &val,
8548 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8549 struct kvm_async_pf *work)
8551 struct x86_exception fault;
8553 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8554 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8556 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8557 (vcpu->arch.apf.send_user_only &&
8558 kvm_x86_ops->get_cpl(vcpu) == 0))
8559 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8560 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8561 fault.vector = PF_VECTOR;
8562 fault.error_code_valid = true;
8563 fault.error_code = 0;
8564 fault.nested_page_fault = false;
8565 fault.address = work->arch.token;
8566 kvm_inject_page_fault(vcpu, &fault);
8570 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8571 struct kvm_async_pf *work)
8573 struct x86_exception fault;
8575 if (work->wakeup_all)
8576 work->arch.token = ~0; /* broadcast wakeup */
8578 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8579 trace_kvm_async_pf_ready(work->arch.token, work->gva);
8581 if ((vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) &&
8582 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8583 fault.vector = PF_VECTOR;
8584 fault.error_code_valid = true;
8585 fault.error_code = 0;
8586 fault.nested_page_fault = false;
8587 fault.address = work->arch.token;
8588 kvm_inject_page_fault(vcpu, &fault);
8590 vcpu->arch.apf.halted = false;
8591 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8594 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8596 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8599 return !kvm_event_needs_reinjection(vcpu) &&
8600 kvm_x86_ops->interrupt_allowed(vcpu);
8603 void kvm_arch_start_assignment(struct kvm *kvm)
8605 atomic_inc(&kvm->arch.assigned_device_count);
8607 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8609 void kvm_arch_end_assignment(struct kvm *kvm)
8611 atomic_dec(&kvm->arch.assigned_device_count);
8613 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8615 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8617 return atomic_read(&kvm->arch.assigned_device_count);
8619 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8621 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8623 atomic_inc(&kvm->arch.noncoherent_dma_count);
8625 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8627 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8629 atomic_dec(&kvm->arch.noncoherent_dma_count);
8631 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8633 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8635 return atomic_read(&kvm->arch.noncoherent_dma_count);
8637 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8639 bool kvm_arch_has_irq_bypass(void)
8641 return kvm_x86_ops->update_pi_irte != NULL;
8644 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8645 struct irq_bypass_producer *prod)
8647 struct kvm_kernel_irqfd *irqfd =
8648 container_of(cons, struct kvm_kernel_irqfd, consumer);
8650 irqfd->producer = prod;
8652 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8653 prod->irq, irqfd->gsi, 1);
8656 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8657 struct irq_bypass_producer *prod)
8660 struct kvm_kernel_irqfd *irqfd =
8661 container_of(cons, struct kvm_kernel_irqfd, consumer);
8663 WARN_ON(irqfd->producer != prod);
8664 irqfd->producer = NULL;
8667 * When producer of consumer is unregistered, we change back to
8668 * remapped mode, so we can re-use the current implementation
8669 * when the irq is masked/disabled or the consumer side (KVM
8670 * int this case doesn't want to receive the interrupts.
8672 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8674 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8675 " fails: %d\n", irqfd->consumer.token, ret);
8678 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8679 uint32_t guest_irq, bool set)
8681 if (!kvm_x86_ops->update_pi_irte)
8684 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
8687 bool kvm_vector_hashing_enabled(void)
8689 return vector_hashing;
8691 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
8693 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
8694 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
8695 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
8696 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
8697 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
8698 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
8699 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
8700 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
8701 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
8702 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
8703 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
8704 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
8705 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
8706 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
8707 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
8708 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
8709 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
8710 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
8711 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);