1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * derived from drivers/kvm/kvm_main.c
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright (C) 2008 Qumranet, Inc.
9 * Copyright IBM Corporation, 2008
10 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
13 * Avi Kivity <avi@qumranet.com>
14 * Yaniv Kamay <yaniv@qumranet.com>
15 * Amit Shah <amit.shah@qumranet.com>
16 * Ben-Ami Yassour <benami@il.ibm.com>
19 #include <linux/kvm_host.h>
25 #include "kvm_cache_regs.h"
26 #include "kvm_emulate.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>
57 #include <linux/sched/isolation.h>
58 #include <linux/mem_encrypt.h>
59 #include <linux/entry-kvm.h>
61 #include <trace/events/kvm.h>
63 #include <asm/debugreg.h>
67 #include <linux/kernel_stat.h>
68 #include <asm/fpu/internal.h> /* Ugh! */
69 #include <asm/pvclock.h>
70 #include <asm/div64.h>
71 #include <asm/irq_remapping.h>
72 #include <asm/mshyperv.h>
73 #include <asm/hypervisor.h>
74 #include <asm/tlbflush.h>
75 #include <asm/intel_pt.h>
76 #include <asm/emulate_prefix.h>
77 #include <clocksource/hyperv_timer.h>
79 #define CREATE_TRACE_POINTS
82 #define MAX_IO_MSRS 256
83 #define KVM_MAX_MCE_BANKS 32
84 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
85 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
87 #define emul_to_vcpu(ctxt) \
88 ((struct kvm_vcpu *)(ctxt)->vcpu)
91 * - enable syscall per default because its emulated by KVM
92 * - enable LME and LMA per default on 64 bit KVM
96 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
98 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
101 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
103 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
104 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
106 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
107 static void process_nmi(struct kvm_vcpu *vcpu);
108 static void enter_smm(struct kvm_vcpu *vcpu);
109 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
110 static void store_regs(struct kvm_vcpu *vcpu);
111 static int sync_regs(struct kvm_vcpu *vcpu);
113 struct kvm_x86_ops kvm_x86_ops __read_mostly;
114 EXPORT_SYMBOL_GPL(kvm_x86_ops);
116 static bool __read_mostly ignore_msrs = 0;
117 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
119 static bool __read_mostly report_ignored_msrs = true;
120 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
122 unsigned int min_timer_period_us = 200;
123 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
125 static bool __read_mostly kvmclock_periodic_sync = true;
126 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
128 bool __read_mostly kvm_has_tsc_control;
129 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
130 u32 __read_mostly kvm_max_guest_tsc_khz;
131 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
132 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
133 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
134 u64 __read_mostly kvm_max_tsc_scaling_ratio;
135 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
136 u64 __read_mostly kvm_default_tsc_scaling_ratio;
137 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
139 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
140 static u32 __read_mostly tsc_tolerance_ppm = 250;
141 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
144 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
145 * adaptive tuning starting from default advancment of 1000ns. '0' disables
146 * advancement entirely. Any other value is used as-is and disables adaptive
147 * tuning, i.e. allows priveleged userspace to set an exact advancement time.
149 static int __read_mostly lapic_timer_advance_ns = -1;
150 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
152 static bool __read_mostly vector_hashing = true;
153 module_param(vector_hashing, bool, S_IRUGO);
155 bool __read_mostly enable_vmware_backdoor = false;
156 module_param(enable_vmware_backdoor, bool, S_IRUGO);
157 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
159 static bool __read_mostly force_emulation_prefix = false;
160 module_param(force_emulation_prefix, bool, S_IRUGO);
162 int __read_mostly pi_inject_timer = -1;
163 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
166 * Restoring the host value for MSRs that are only consumed when running in
167 * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
168 * returns to userspace, i.e. the kernel can run with the guest's value.
170 #define KVM_MAX_NR_USER_RETURN_MSRS 16
172 struct kvm_user_return_msrs_global {
174 u32 msrs[KVM_MAX_NR_USER_RETURN_MSRS];
177 struct kvm_user_return_msrs {
178 struct user_return_notifier urn;
180 struct kvm_user_return_msr_values {
183 } values[KVM_MAX_NR_USER_RETURN_MSRS];
186 static struct kvm_user_return_msrs_global __read_mostly user_return_msrs_global;
187 static struct kvm_user_return_msrs __percpu *user_return_msrs;
189 #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
190 | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
191 | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
192 | XFEATURE_MASK_PKRU)
194 u64 __read_mostly host_efer;
195 EXPORT_SYMBOL_GPL(host_efer);
197 bool __read_mostly allow_smaller_maxphyaddr = 0;
198 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
200 static u64 __read_mostly host_xss;
201 u64 __read_mostly supported_xss;
202 EXPORT_SYMBOL_GPL(supported_xss);
204 struct kvm_stats_debugfs_item debugfs_entries[] = {
205 VCPU_STAT("pf_fixed", pf_fixed),
206 VCPU_STAT("pf_guest", pf_guest),
207 VCPU_STAT("tlb_flush", tlb_flush),
208 VCPU_STAT("invlpg", invlpg),
209 VCPU_STAT("exits", exits),
210 VCPU_STAT("io_exits", io_exits),
211 VCPU_STAT("mmio_exits", mmio_exits),
212 VCPU_STAT("signal_exits", signal_exits),
213 VCPU_STAT("irq_window", irq_window_exits),
214 VCPU_STAT("nmi_window", nmi_window_exits),
215 VCPU_STAT("halt_exits", halt_exits),
216 VCPU_STAT("halt_successful_poll", halt_successful_poll),
217 VCPU_STAT("halt_attempted_poll", halt_attempted_poll),
218 VCPU_STAT("halt_poll_invalid", halt_poll_invalid),
219 VCPU_STAT("halt_wakeup", halt_wakeup),
220 VCPU_STAT("hypercalls", hypercalls),
221 VCPU_STAT("request_irq", request_irq_exits),
222 VCPU_STAT("irq_exits", irq_exits),
223 VCPU_STAT("host_state_reload", host_state_reload),
224 VCPU_STAT("fpu_reload", fpu_reload),
225 VCPU_STAT("insn_emulation", insn_emulation),
226 VCPU_STAT("insn_emulation_fail", insn_emulation_fail),
227 VCPU_STAT("irq_injections", irq_injections),
228 VCPU_STAT("nmi_injections", nmi_injections),
229 VCPU_STAT("req_event", req_event),
230 VCPU_STAT("l1d_flush", l1d_flush),
231 VCPU_STAT("halt_poll_success_ns", halt_poll_success_ns),
232 VCPU_STAT("halt_poll_fail_ns", halt_poll_fail_ns),
233 VM_STAT("mmu_shadow_zapped", mmu_shadow_zapped),
234 VM_STAT("mmu_pte_write", mmu_pte_write),
235 VM_STAT("mmu_pte_updated", mmu_pte_updated),
236 VM_STAT("mmu_pde_zapped", mmu_pde_zapped),
237 VM_STAT("mmu_flooded", mmu_flooded),
238 VM_STAT("mmu_recycled", mmu_recycled),
239 VM_STAT("mmu_cache_miss", mmu_cache_miss),
240 VM_STAT("mmu_unsync", mmu_unsync),
241 VM_STAT("remote_tlb_flush", remote_tlb_flush),
242 VM_STAT("largepages", lpages, .mode = 0444),
243 VM_STAT("nx_largepages_splitted", nx_lpage_splits, .mode = 0444),
244 VM_STAT("max_mmu_page_hash_collisions", max_mmu_page_hash_collisions),
248 u64 __read_mostly host_xcr0;
249 u64 __read_mostly supported_xcr0;
250 EXPORT_SYMBOL_GPL(supported_xcr0);
252 static struct kmem_cache *x86_fpu_cache;
254 static struct kmem_cache *x86_emulator_cache;
257 * When called, it means the previous get/set msr reached an invalid msr.
258 * Return true if we want to ignore/silent this failed msr access.
260 static bool kvm_msr_ignored_check(struct kvm_vcpu *vcpu, u32 msr,
261 u64 data, bool write)
263 const char *op = write ? "wrmsr" : "rdmsr";
266 if (report_ignored_msrs)
267 kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
272 kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
278 static struct kmem_cache *kvm_alloc_emulator_cache(void)
280 unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
281 unsigned int size = sizeof(struct x86_emulate_ctxt);
283 return kmem_cache_create_usercopy("x86_emulator", size,
284 __alignof__(struct x86_emulate_ctxt),
285 SLAB_ACCOUNT, useroffset,
286 size - useroffset, NULL);
289 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
291 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
294 for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
295 vcpu->arch.apf.gfns[i] = ~0;
298 static void kvm_on_user_return(struct user_return_notifier *urn)
301 struct kvm_user_return_msrs *msrs
302 = container_of(urn, struct kvm_user_return_msrs, urn);
303 struct kvm_user_return_msr_values *values;
307 * Disabling irqs at this point since the following code could be
308 * interrupted and executed through kvm_arch_hardware_disable()
310 local_irq_save(flags);
311 if (msrs->registered) {
312 msrs->registered = false;
313 user_return_notifier_unregister(urn);
315 local_irq_restore(flags);
316 for (slot = 0; slot < user_return_msrs_global.nr; ++slot) {
317 values = &msrs->values[slot];
318 if (values->host != values->curr) {
319 wrmsrl(user_return_msrs_global.msrs[slot], values->host);
320 values->curr = values->host;
325 void kvm_define_user_return_msr(unsigned slot, u32 msr)
327 BUG_ON(slot >= KVM_MAX_NR_USER_RETURN_MSRS);
328 user_return_msrs_global.msrs[slot] = msr;
329 if (slot >= user_return_msrs_global.nr)
330 user_return_msrs_global.nr = slot + 1;
332 EXPORT_SYMBOL_GPL(kvm_define_user_return_msr);
334 static void kvm_user_return_msr_cpu_online(void)
336 unsigned int cpu = smp_processor_id();
337 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
341 for (i = 0; i < user_return_msrs_global.nr; ++i) {
342 rdmsrl_safe(user_return_msrs_global.msrs[i], &value);
343 msrs->values[i].host = value;
344 msrs->values[i].curr = value;
348 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
350 unsigned int cpu = smp_processor_id();
351 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
354 value = (value & mask) | (msrs->values[slot].host & ~mask);
355 if (value == msrs->values[slot].curr)
357 err = wrmsrl_safe(user_return_msrs_global.msrs[slot], value);
361 msrs->values[slot].curr = value;
362 if (!msrs->registered) {
363 msrs->urn.on_user_return = kvm_on_user_return;
364 user_return_notifier_register(&msrs->urn);
365 msrs->registered = true;
369 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
371 static void drop_user_return_notifiers(void)
373 unsigned int cpu = smp_processor_id();
374 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
376 if (msrs->registered)
377 kvm_on_user_return(&msrs->urn);
380 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
382 return vcpu->arch.apic_base;
384 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
386 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
388 return kvm_apic_mode(kvm_get_apic_base(vcpu));
390 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
392 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
394 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
395 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
396 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
397 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
399 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
401 if (!msr_info->host_initiated) {
402 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
404 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
408 kvm_lapic_set_base(vcpu, msr_info->data);
409 kvm_recalculate_apic_map(vcpu->kvm);
412 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
414 asmlinkage __visible noinstr void kvm_spurious_fault(void)
416 /* Fault while not rebooting. We want the trace. */
417 BUG_ON(!kvm_rebooting);
419 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
421 #define EXCPT_BENIGN 0
422 #define EXCPT_CONTRIBUTORY 1
425 static int exception_class(int vector)
435 return EXCPT_CONTRIBUTORY;
442 #define EXCPT_FAULT 0
444 #define EXCPT_ABORT 2
445 #define EXCPT_INTERRUPT 3
447 static int exception_type(int vector)
451 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
452 return EXCPT_INTERRUPT;
456 /* #DB is trap, as instruction watchpoints are handled elsewhere */
457 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
460 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
463 /* Reserved exceptions will result in fault */
467 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
469 unsigned nr = vcpu->arch.exception.nr;
470 bool has_payload = vcpu->arch.exception.has_payload;
471 unsigned long payload = vcpu->arch.exception.payload;
479 * "Certain debug exceptions may clear bit 0-3. The
480 * remaining contents of the DR6 register are never
481 * cleared by the processor".
483 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
485 * DR6.RTM is set by all #DB exceptions that don't clear it.
487 vcpu->arch.dr6 |= DR6_RTM;
488 vcpu->arch.dr6 |= payload;
490 * Bit 16 should be set in the payload whenever the #DB
491 * exception should clear DR6.RTM. This makes the payload
492 * compatible with the pending debug exceptions under VMX.
493 * Though not currently documented in the SDM, this also
494 * makes the payload compatible with the exit qualification
495 * for #DB exceptions under VMX.
497 vcpu->arch.dr6 ^= payload & DR6_RTM;
500 * The #DB payload is defined as compatible with the 'pending
501 * debug exceptions' field under VMX, not DR6. While bit 12 is
502 * defined in the 'pending debug exceptions' field (enabled
503 * breakpoint), it is reserved and must be zero in DR6.
505 vcpu->arch.dr6 &= ~BIT(12);
508 vcpu->arch.cr2 = payload;
512 vcpu->arch.exception.has_payload = false;
513 vcpu->arch.exception.payload = 0;
515 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
517 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
518 unsigned nr, bool has_error, u32 error_code,
519 bool has_payload, unsigned long payload, bool reinject)
524 kvm_make_request(KVM_REQ_EVENT, vcpu);
526 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
528 if (has_error && !is_protmode(vcpu))
532 * On vmentry, vcpu->arch.exception.pending is only
533 * true if an event injection was blocked by
534 * nested_run_pending. In that case, however,
535 * vcpu_enter_guest requests an immediate exit,
536 * and the guest shouldn't proceed far enough to
539 WARN_ON_ONCE(vcpu->arch.exception.pending);
540 vcpu->arch.exception.injected = true;
541 if (WARN_ON_ONCE(has_payload)) {
543 * A reinjected event has already
544 * delivered its payload.
550 vcpu->arch.exception.pending = true;
551 vcpu->arch.exception.injected = false;
553 vcpu->arch.exception.has_error_code = has_error;
554 vcpu->arch.exception.nr = nr;
555 vcpu->arch.exception.error_code = error_code;
556 vcpu->arch.exception.has_payload = has_payload;
557 vcpu->arch.exception.payload = payload;
558 if (!is_guest_mode(vcpu))
559 kvm_deliver_exception_payload(vcpu);
563 /* to check exception */
564 prev_nr = vcpu->arch.exception.nr;
565 if (prev_nr == DF_VECTOR) {
566 /* triple fault -> shutdown */
567 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
570 class1 = exception_class(prev_nr);
571 class2 = exception_class(nr);
572 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
573 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
575 * Generate double fault per SDM Table 5-5. Set
576 * exception.pending = true so that the double fault
577 * can trigger a nested vmexit.
579 vcpu->arch.exception.pending = true;
580 vcpu->arch.exception.injected = false;
581 vcpu->arch.exception.has_error_code = true;
582 vcpu->arch.exception.nr = DF_VECTOR;
583 vcpu->arch.exception.error_code = 0;
584 vcpu->arch.exception.has_payload = false;
585 vcpu->arch.exception.payload = 0;
587 /* replace previous exception with a new one in a hope
588 that instruction re-execution will regenerate lost
593 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
595 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
597 EXPORT_SYMBOL_GPL(kvm_queue_exception);
599 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
601 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
603 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
605 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
606 unsigned long payload)
608 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
610 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
612 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
613 u32 error_code, unsigned long payload)
615 kvm_multiple_exception(vcpu, nr, true, error_code,
616 true, payload, false);
619 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
622 kvm_inject_gp(vcpu, 0);
624 return kvm_skip_emulated_instruction(vcpu);
628 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
630 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
632 ++vcpu->stat.pf_guest;
633 vcpu->arch.exception.nested_apf =
634 is_guest_mode(vcpu) && fault->async_page_fault;
635 if (vcpu->arch.exception.nested_apf) {
636 vcpu->arch.apf.nested_apf_token = fault->address;
637 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
639 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
643 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
645 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
646 struct x86_exception *fault)
648 struct kvm_mmu *fault_mmu;
649 WARN_ON_ONCE(fault->vector != PF_VECTOR);
651 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
655 * Invalidate the TLB entry for the faulting address, if it exists,
656 * else the access will fault indefinitely (and to emulate hardware).
658 if ((fault->error_code & PFERR_PRESENT_MASK) &&
659 !(fault->error_code & PFERR_RSVD_MASK))
660 kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
661 fault_mmu->root_hpa);
663 fault_mmu->inject_page_fault(vcpu, fault);
664 return fault->nested_page_fault;
666 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
668 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
670 atomic_inc(&vcpu->arch.nmi_queued);
671 kvm_make_request(KVM_REQ_NMI, vcpu);
673 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
675 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
677 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
679 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
681 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
683 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
685 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
688 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
689 * a #GP and return false.
691 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
693 if (kvm_x86_ops.get_cpl(vcpu) <= required_cpl)
695 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
698 EXPORT_SYMBOL_GPL(kvm_require_cpl);
700 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
702 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
705 kvm_queue_exception(vcpu, UD_VECTOR);
708 EXPORT_SYMBOL_GPL(kvm_require_dr);
711 * This function will be used to read from the physical memory of the currently
712 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
713 * can read from guest physical or from the guest's guest physical memory.
715 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
716 gfn_t ngfn, void *data, int offset, int len,
719 struct x86_exception exception;
723 ngpa = gfn_to_gpa(ngfn);
724 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
725 if (real_gfn == UNMAPPED_GVA)
728 real_gfn = gpa_to_gfn(real_gfn);
730 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
732 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
734 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
735 void *data, int offset, int len, u32 access)
737 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
738 data, offset, len, access);
741 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
743 return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
748 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
750 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
752 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
753 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
756 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
758 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
759 offset * sizeof(u64), sizeof(pdpte),
760 PFERR_USER_MASK|PFERR_WRITE_MASK);
765 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
766 if ((pdpte[i] & PT_PRESENT_MASK) &&
767 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
774 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
775 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
781 EXPORT_SYMBOL_GPL(load_pdptrs);
783 bool pdptrs_changed(struct kvm_vcpu *vcpu)
785 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
790 if (!is_pae_paging(vcpu))
793 if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
796 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
797 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
798 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
799 PFERR_USER_MASK | PFERR_WRITE_MASK);
803 return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
805 EXPORT_SYMBOL_GPL(pdptrs_changed);
807 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
809 unsigned long old_cr0 = kvm_read_cr0(vcpu);
810 unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
811 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
816 if (cr0 & 0xffffffff00000000UL)
820 cr0 &= ~CR0_RESERVED_BITS;
822 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
825 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
829 if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
830 (cr0 & X86_CR0_PG)) {
835 kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
840 if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
841 is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
842 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
845 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
848 kvm_x86_ops.set_cr0(vcpu, cr0);
850 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
851 kvm_clear_async_pf_completion_queue(vcpu);
852 kvm_async_pf_hash_reset(vcpu);
855 if ((cr0 ^ old_cr0) & update_bits)
856 kvm_mmu_reset_context(vcpu);
858 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
859 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
860 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
861 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
865 EXPORT_SYMBOL_GPL(kvm_set_cr0);
867 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
869 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
871 EXPORT_SYMBOL_GPL(kvm_lmsw);
873 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
875 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
877 if (vcpu->arch.xcr0 != host_xcr0)
878 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
880 if (vcpu->arch.xsaves_enabled &&
881 vcpu->arch.ia32_xss != host_xss)
882 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
885 if (static_cpu_has(X86_FEATURE_PKU) &&
886 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
887 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
888 vcpu->arch.pkru != vcpu->arch.host_pkru)
889 __write_pkru(vcpu->arch.pkru);
891 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
893 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
895 if (static_cpu_has(X86_FEATURE_PKU) &&
896 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
897 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
898 vcpu->arch.pkru = rdpkru();
899 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
900 __write_pkru(vcpu->arch.host_pkru);
903 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
905 if (vcpu->arch.xcr0 != host_xcr0)
906 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
908 if (vcpu->arch.xsaves_enabled &&
909 vcpu->arch.ia32_xss != host_xss)
910 wrmsrl(MSR_IA32_XSS, host_xss);
914 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
916 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
919 u64 old_xcr0 = vcpu->arch.xcr0;
922 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
923 if (index != XCR_XFEATURE_ENABLED_MASK)
925 if (!(xcr0 & XFEATURE_MASK_FP))
927 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
931 * Do not allow the guest to set bits that we do not support
932 * saving. However, xcr0 bit 0 is always set, even if the
933 * emulated CPU does not support XSAVE (see fx_init).
935 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
936 if (xcr0 & ~valid_bits)
939 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
940 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
943 if (xcr0 & XFEATURE_MASK_AVX512) {
944 if (!(xcr0 & XFEATURE_MASK_YMM))
946 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
949 vcpu->arch.xcr0 = xcr0;
951 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
952 kvm_update_cpuid_runtime(vcpu);
956 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
958 if (kvm_x86_ops.get_cpl(vcpu) != 0 ||
959 __kvm_set_xcr(vcpu, index, xcr)) {
960 kvm_inject_gp(vcpu, 0);
965 EXPORT_SYMBOL_GPL(kvm_set_xcr);
967 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
969 if (cr4 & cr4_reserved_bits)
972 if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
975 return kvm_x86_ops.is_valid_cr4(vcpu, cr4);
977 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
979 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
981 unsigned long old_cr4 = kvm_read_cr4(vcpu);
982 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
984 unsigned long mmu_role_bits = pdptr_bits | X86_CR4_SMAP | X86_CR4_PKE;
986 if (!kvm_is_valid_cr4(vcpu, cr4))
989 if (is_long_mode(vcpu)) {
990 if (!(cr4 & X86_CR4_PAE))
992 if ((cr4 ^ old_cr4) & X86_CR4_LA57)
994 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
995 && ((cr4 ^ old_cr4) & pdptr_bits)
996 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1000 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1001 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1004 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1005 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1009 kvm_x86_ops.set_cr4(vcpu, cr4);
1011 if (((cr4 ^ old_cr4) & mmu_role_bits) ||
1012 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1013 kvm_mmu_reset_context(vcpu);
1015 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
1016 kvm_update_cpuid_runtime(vcpu);
1020 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1022 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1024 bool skip_tlb_flush = false;
1025 #ifdef CONFIG_X86_64
1026 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1029 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1030 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1034 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
1035 if (!skip_tlb_flush) {
1036 kvm_mmu_sync_roots(vcpu);
1037 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1042 if (is_long_mode(vcpu) &&
1043 (cr3 & vcpu->arch.cr3_lm_rsvd_bits))
1045 else if (is_pae_paging(vcpu) &&
1046 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1049 kvm_mmu_new_pgd(vcpu, cr3, skip_tlb_flush, skip_tlb_flush);
1050 vcpu->arch.cr3 = cr3;
1051 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1055 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1057 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1059 if (cr8 & CR8_RESERVED_BITS)
1061 if (lapic_in_kernel(vcpu))
1062 kvm_lapic_set_tpr(vcpu, cr8);
1064 vcpu->arch.cr8 = cr8;
1067 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1069 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1071 if (lapic_in_kernel(vcpu))
1072 return kvm_lapic_get_cr8(vcpu);
1074 return vcpu->arch.cr8;
1076 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1078 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1082 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1083 for (i = 0; i < KVM_NR_DB_REGS; i++)
1084 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1085 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1089 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1093 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1094 dr7 = vcpu->arch.guest_debug_dr7;
1096 dr7 = vcpu->arch.dr7;
1097 kvm_x86_ops.set_dr7(vcpu, dr7);
1098 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1099 if (dr7 & DR7_BP_EN_MASK)
1100 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1102 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1104 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1106 u64 fixed = DR6_FIXED_1;
1108 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1113 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1115 size_t size = ARRAY_SIZE(vcpu->arch.db);
1119 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1120 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1121 vcpu->arch.eff_db[dr] = val;
1125 if (!kvm_dr6_valid(val))
1126 return -1; /* #GP */
1127 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1131 if (!kvm_dr7_valid(val))
1132 return -1; /* #GP */
1133 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1134 kvm_update_dr7(vcpu);
1141 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1143 if (__kvm_set_dr(vcpu, dr, val)) {
1144 kvm_inject_gp(vcpu, 0);
1149 EXPORT_SYMBOL_GPL(kvm_set_dr);
1151 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1153 size_t size = ARRAY_SIZE(vcpu->arch.db);
1157 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1161 *val = vcpu->arch.dr6;
1165 *val = vcpu->arch.dr7;
1170 EXPORT_SYMBOL_GPL(kvm_get_dr);
1172 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1174 u32 ecx = kvm_rcx_read(vcpu);
1178 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1181 kvm_rax_write(vcpu, (u32)data);
1182 kvm_rdx_write(vcpu, data >> 32);
1185 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1188 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1189 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1191 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1192 * extract the supported MSRs from the related const lists.
1193 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1194 * capabilities of the host cpu. This capabilities test skips MSRs that are
1195 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1196 * may depend on host virtualization features rather than host cpu features.
1199 static const u32 msrs_to_save_all[] = {
1200 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1202 #ifdef CONFIG_X86_64
1203 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1205 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1206 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1208 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1209 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1210 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1211 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1212 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1213 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1214 MSR_IA32_UMWAIT_CONTROL,
1216 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1217 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1218 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1219 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1220 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1221 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1222 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1223 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1224 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1225 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1226 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1227 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1228 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1229 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1230 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1231 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1232 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1233 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1234 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1235 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1236 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1237 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1240 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1241 static unsigned num_msrs_to_save;
1243 static const u32 emulated_msrs_all[] = {
1244 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1245 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1246 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1247 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1248 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1249 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1250 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1252 HV_X64_MSR_VP_INDEX,
1253 HV_X64_MSR_VP_RUNTIME,
1254 HV_X64_MSR_SCONTROL,
1255 HV_X64_MSR_STIMER0_CONFIG,
1256 HV_X64_MSR_VP_ASSIST_PAGE,
1257 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1258 HV_X64_MSR_TSC_EMULATION_STATUS,
1259 HV_X64_MSR_SYNDBG_OPTIONS,
1260 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1261 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1262 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1264 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1265 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1267 MSR_IA32_TSC_ADJUST,
1268 MSR_IA32_TSCDEADLINE,
1269 MSR_IA32_ARCH_CAPABILITIES,
1270 MSR_IA32_PERF_CAPABILITIES,
1271 MSR_IA32_MISC_ENABLE,
1272 MSR_IA32_MCG_STATUS,
1274 MSR_IA32_MCG_EXT_CTL,
1278 MSR_MISC_FEATURES_ENABLES,
1279 MSR_AMD64_VIRT_SPEC_CTRL,
1284 * The following list leaves out MSRs whose values are determined
1285 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1286 * We always support the "true" VMX control MSRs, even if the host
1287 * processor does not, so I am putting these registers here rather
1288 * than in msrs_to_save_all.
1291 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1292 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1293 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1294 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1296 MSR_IA32_VMX_CR0_FIXED0,
1297 MSR_IA32_VMX_CR4_FIXED0,
1298 MSR_IA32_VMX_VMCS_ENUM,
1299 MSR_IA32_VMX_PROCBASED_CTLS2,
1300 MSR_IA32_VMX_EPT_VPID_CAP,
1301 MSR_IA32_VMX_VMFUNC,
1304 MSR_KVM_POLL_CONTROL,
1307 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1308 static unsigned num_emulated_msrs;
1311 * List of msr numbers which are used to expose MSR-based features that
1312 * can be used by a hypervisor to validate requested CPU features.
1314 static const u32 msr_based_features_all[] = {
1316 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1317 MSR_IA32_VMX_PINBASED_CTLS,
1318 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1319 MSR_IA32_VMX_PROCBASED_CTLS,
1320 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1321 MSR_IA32_VMX_EXIT_CTLS,
1322 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1323 MSR_IA32_VMX_ENTRY_CTLS,
1325 MSR_IA32_VMX_CR0_FIXED0,
1326 MSR_IA32_VMX_CR0_FIXED1,
1327 MSR_IA32_VMX_CR4_FIXED0,
1328 MSR_IA32_VMX_CR4_FIXED1,
1329 MSR_IA32_VMX_VMCS_ENUM,
1330 MSR_IA32_VMX_PROCBASED_CTLS2,
1331 MSR_IA32_VMX_EPT_VPID_CAP,
1332 MSR_IA32_VMX_VMFUNC,
1336 MSR_IA32_ARCH_CAPABILITIES,
1337 MSR_IA32_PERF_CAPABILITIES,
1340 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1341 static unsigned int num_msr_based_features;
1343 static u64 kvm_get_arch_capabilities(void)
1347 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1348 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1351 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1352 * the nested hypervisor runs with NX huge pages. If it is not,
1353 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1354 * L1 guests, so it need not worry about its own (L2) guests.
1356 data |= ARCH_CAP_PSCHANGE_MC_NO;
1359 * If we're doing cache flushes (either "always" or "cond")
1360 * we will do one whenever the guest does a vmlaunch/vmresume.
1361 * If an outer hypervisor is doing the cache flush for us
1362 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1363 * capability to the guest too, and if EPT is disabled we're not
1364 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1365 * require a nested hypervisor to do a flush of its own.
1367 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1368 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1370 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1371 data |= ARCH_CAP_RDCL_NO;
1372 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1373 data |= ARCH_CAP_SSB_NO;
1374 if (!boot_cpu_has_bug(X86_BUG_MDS))
1375 data |= ARCH_CAP_MDS_NO;
1378 * On TAA affected systems:
1379 * - nothing to do if TSX is disabled on the host.
1380 * - we emulate TSX_CTRL if present on the host.
1381 * This lets the guest use VERW to clear CPU buffers.
1383 if (!boot_cpu_has(X86_FEATURE_RTM))
1384 data &= ~(ARCH_CAP_TAA_NO | ARCH_CAP_TSX_CTRL_MSR);
1385 else if (!boot_cpu_has_bug(X86_BUG_TAA))
1386 data |= ARCH_CAP_TAA_NO;
1391 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1393 switch (msr->index) {
1394 case MSR_IA32_ARCH_CAPABILITIES:
1395 msr->data = kvm_get_arch_capabilities();
1397 case MSR_IA32_UCODE_REV:
1398 rdmsrl_safe(msr->index, &msr->data);
1401 return kvm_x86_ops.get_msr_feature(msr);
1406 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1408 struct kvm_msr_entry msr;
1412 r = kvm_get_msr_feature(&msr);
1414 if (r == KVM_MSR_RET_INVALID) {
1415 /* Unconditionally clear the output for simplicity */
1417 if (kvm_msr_ignored_check(vcpu, index, 0, false))
1429 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1431 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1434 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1437 if (efer & (EFER_LME | EFER_LMA) &&
1438 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1441 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1447 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1449 if (efer & efer_reserved_bits)
1452 return __kvm_valid_efer(vcpu, efer);
1454 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1456 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1458 u64 old_efer = vcpu->arch.efer;
1459 u64 efer = msr_info->data;
1462 if (efer & efer_reserved_bits)
1465 if (!msr_info->host_initiated) {
1466 if (!__kvm_valid_efer(vcpu, efer))
1469 if (is_paging(vcpu) &&
1470 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1475 efer |= vcpu->arch.efer & EFER_LMA;
1477 r = kvm_x86_ops.set_efer(vcpu, efer);
1483 /* Update reserved bits */
1484 if ((efer ^ old_efer) & EFER_NX)
1485 kvm_mmu_reset_context(vcpu);
1490 void kvm_enable_efer_bits(u64 mask)
1492 efer_reserved_bits &= ~mask;
1494 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1496 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1498 struct kvm *kvm = vcpu->kvm;
1499 struct msr_bitmap_range *ranges = kvm->arch.msr_filter.ranges;
1500 u32 count = kvm->arch.msr_filter.count;
1502 bool r = kvm->arch.msr_filter.default_allow;
1505 /* MSR filtering not set up or x2APIC enabled, allow everything */
1506 if (!count || (index >= 0x800 && index <= 0x8ff))
1509 /* Prevent collision with set_msr_filter */
1510 idx = srcu_read_lock(&kvm->srcu);
1512 for (i = 0; i < count; i++) {
1513 u32 start = ranges[i].base;
1514 u32 end = start + ranges[i].nmsrs;
1515 u32 flags = ranges[i].flags;
1516 unsigned long *bitmap = ranges[i].bitmap;
1518 if ((index >= start) && (index < end) && (flags & type)) {
1519 r = !!test_bit(index - start, bitmap);
1524 srcu_read_unlock(&kvm->srcu, idx);
1528 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1531 * Write @data into the MSR specified by @index. Select MSR specific fault
1532 * checks are bypassed if @host_initiated is %true.
1533 * Returns 0 on success, non-0 otherwise.
1534 * Assumes vcpu_load() was already called.
1536 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1537 bool host_initiated)
1539 struct msr_data msr;
1541 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1542 return KVM_MSR_RET_FILTERED;
1547 case MSR_KERNEL_GS_BASE:
1550 if (is_noncanonical_address(data, vcpu))
1553 case MSR_IA32_SYSENTER_EIP:
1554 case MSR_IA32_SYSENTER_ESP:
1556 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1557 * non-canonical address is written on Intel but not on
1558 * AMD (which ignores the top 32-bits, because it does
1559 * not implement 64-bit SYSENTER).
1561 * 64-bit code should hence be able to write a non-canonical
1562 * value on AMD. Making the address canonical ensures that
1563 * vmentry does not fail on Intel after writing a non-canonical
1564 * value, and that something deterministic happens if the guest
1565 * invokes 64-bit SYSENTER.
1567 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1572 msr.host_initiated = host_initiated;
1574 return kvm_x86_ops.set_msr(vcpu, &msr);
1577 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1578 u32 index, u64 data, bool host_initiated)
1580 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1582 if (ret == KVM_MSR_RET_INVALID)
1583 if (kvm_msr_ignored_check(vcpu, index, data, true))
1590 * Read the MSR specified by @index into @data. Select MSR specific fault
1591 * checks are bypassed if @host_initiated is %true.
1592 * Returns 0 on success, non-0 otherwise.
1593 * Assumes vcpu_load() was already called.
1595 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1596 bool host_initiated)
1598 struct msr_data msr;
1601 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1602 return KVM_MSR_RET_FILTERED;
1605 msr.host_initiated = host_initiated;
1607 ret = kvm_x86_ops.get_msr(vcpu, &msr);
1613 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1614 u32 index, u64 *data, bool host_initiated)
1616 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1618 if (ret == KVM_MSR_RET_INVALID) {
1619 /* Unconditionally clear *data for simplicity */
1621 if (kvm_msr_ignored_check(vcpu, index, 0, false))
1628 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1630 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1632 EXPORT_SYMBOL_GPL(kvm_get_msr);
1634 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1636 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1638 EXPORT_SYMBOL_GPL(kvm_set_msr);
1640 static int complete_emulated_msr(struct kvm_vcpu *vcpu, bool is_read)
1642 if (vcpu->run->msr.error) {
1643 kvm_inject_gp(vcpu, 0);
1645 } else if (is_read) {
1646 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1647 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1650 return kvm_skip_emulated_instruction(vcpu);
1653 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1655 return complete_emulated_msr(vcpu, true);
1658 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1660 return complete_emulated_msr(vcpu, false);
1663 static u64 kvm_msr_reason(int r)
1666 case KVM_MSR_RET_INVALID:
1667 return KVM_MSR_EXIT_REASON_UNKNOWN;
1668 case KVM_MSR_RET_FILTERED:
1669 return KVM_MSR_EXIT_REASON_FILTER;
1671 return KVM_MSR_EXIT_REASON_INVAL;
1675 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1676 u32 exit_reason, u64 data,
1677 int (*completion)(struct kvm_vcpu *vcpu),
1680 u64 msr_reason = kvm_msr_reason(r);
1682 /* Check if the user wanted to know about this MSR fault */
1683 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1686 vcpu->run->exit_reason = exit_reason;
1687 vcpu->run->msr.error = 0;
1688 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1689 vcpu->run->msr.reason = msr_reason;
1690 vcpu->run->msr.index = index;
1691 vcpu->run->msr.data = data;
1692 vcpu->arch.complete_userspace_io = completion;
1697 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1699 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1700 complete_emulated_rdmsr, r);
1703 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1705 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1706 complete_emulated_wrmsr, r);
1709 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1711 u32 ecx = kvm_rcx_read(vcpu);
1715 r = kvm_get_msr(vcpu, ecx, &data);
1717 /* MSR read failed? See if we should ask user space */
1718 if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1719 /* Bounce to user space */
1723 /* MSR read failed? Inject a #GP */
1725 trace_kvm_msr_read_ex(ecx);
1726 kvm_inject_gp(vcpu, 0);
1730 trace_kvm_msr_read(ecx, data);
1732 kvm_rax_write(vcpu, data & -1u);
1733 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1734 return kvm_skip_emulated_instruction(vcpu);
1736 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1738 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1740 u32 ecx = kvm_rcx_read(vcpu);
1741 u64 data = kvm_read_edx_eax(vcpu);
1744 r = kvm_set_msr(vcpu, ecx, data);
1746 /* MSR write failed? See if we should ask user space */
1747 if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1748 /* Bounce to user space */
1751 /* Signal all other negative errors to userspace */
1755 /* MSR write failed? Inject a #GP */
1757 trace_kvm_msr_write_ex(ecx, data);
1758 kvm_inject_gp(vcpu, 0);
1762 trace_kvm_msr_write(ecx, data);
1763 return kvm_skip_emulated_instruction(vcpu);
1765 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1767 bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1769 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1770 xfer_to_guest_mode_work_pending();
1772 EXPORT_SYMBOL_GPL(kvm_vcpu_exit_request);
1775 * The fast path for frequent and performance sensitive wrmsr emulation,
1776 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1777 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1778 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1779 * other cases which must be called after interrupts are enabled on the host.
1781 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1783 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1786 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1787 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1788 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1789 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1792 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1793 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1794 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1795 trace_kvm_apic_write(APIC_ICR, (u32)data);
1802 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1804 if (!kvm_can_use_hv_timer(vcpu))
1807 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1811 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1813 u32 msr = kvm_rcx_read(vcpu);
1815 fastpath_t ret = EXIT_FASTPATH_NONE;
1818 case APIC_BASE_MSR + (APIC_ICR >> 4):
1819 data = kvm_read_edx_eax(vcpu);
1820 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1821 kvm_skip_emulated_instruction(vcpu);
1822 ret = EXIT_FASTPATH_EXIT_HANDLED;
1825 case MSR_IA32_TSCDEADLINE:
1826 data = kvm_read_edx_eax(vcpu);
1827 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1828 kvm_skip_emulated_instruction(vcpu);
1829 ret = EXIT_FASTPATH_REENTER_GUEST;
1836 if (ret != EXIT_FASTPATH_NONE)
1837 trace_kvm_msr_write(msr, data);
1841 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1844 * Adapt set_msr() to msr_io()'s calling convention
1846 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1848 return kvm_get_msr_ignored_check(vcpu, index, data, true);
1851 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1853 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
1856 #ifdef CONFIG_X86_64
1857 struct pvclock_clock {
1867 struct pvclock_gtod_data {
1870 struct pvclock_clock clock; /* extract of a clocksource struct */
1871 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1877 static struct pvclock_gtod_data pvclock_gtod_data;
1879 static void update_pvclock_gtod(struct timekeeper *tk)
1881 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1883 write_seqcount_begin(&vdata->seq);
1885 /* copy pvclock gtod data */
1886 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
1887 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1888 vdata->clock.mask = tk->tkr_mono.mask;
1889 vdata->clock.mult = tk->tkr_mono.mult;
1890 vdata->clock.shift = tk->tkr_mono.shift;
1891 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
1892 vdata->clock.offset = tk->tkr_mono.base;
1894 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
1895 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
1896 vdata->raw_clock.mask = tk->tkr_raw.mask;
1897 vdata->raw_clock.mult = tk->tkr_raw.mult;
1898 vdata->raw_clock.shift = tk->tkr_raw.shift;
1899 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
1900 vdata->raw_clock.offset = tk->tkr_raw.base;
1902 vdata->wall_time_sec = tk->xtime_sec;
1904 vdata->offs_boot = tk->offs_boot;
1906 write_seqcount_end(&vdata->seq);
1909 static s64 get_kvmclock_base_ns(void)
1911 /* Count up from boot time, but with the frequency of the raw clock. */
1912 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
1915 static s64 get_kvmclock_base_ns(void)
1917 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
1918 return ktime_get_boottime_ns();
1922 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1926 struct pvclock_wall_clock wc;
1929 kvm->arch.wall_clock = wall_clock;
1934 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1939 ++version; /* first time write, random junk */
1943 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1947 * The guest calculates current wall clock time by adding
1948 * system time (updated by kvm_guest_time_update below) to the
1949 * wall clock specified here. We do the reverse here.
1951 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
1953 wc.nsec = do_div(wall_nsec, 1000000000);
1954 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
1955 wc.version = version;
1957 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1960 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1963 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
1964 bool old_msr, bool host_initiated)
1966 struct kvm_arch *ka = &vcpu->kvm->arch;
1968 if (vcpu->vcpu_id == 0 && !host_initiated) {
1969 if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
1970 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1972 ka->boot_vcpu_runs_old_kvmclock = old_msr;
1975 vcpu->arch.time = system_time;
1976 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
1978 /* we verify if the enable bit is set... */
1979 vcpu->arch.pv_time_enabled = false;
1980 if (!(system_time & 1))
1983 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
1984 &vcpu->arch.pv_time, system_time & ~1ULL,
1985 sizeof(struct pvclock_vcpu_time_info)))
1986 vcpu->arch.pv_time_enabled = true;
1991 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1993 do_shl32_div32(dividend, divisor);
1997 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1998 s8 *pshift, u32 *pmultiplier)
2006 scaled64 = scaled_hz;
2007 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2012 tps32 = (uint32_t)tps64;
2013 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2014 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2022 *pmultiplier = div_frac(scaled64, tps32);
2025 #ifdef CONFIG_X86_64
2026 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2029 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2030 static unsigned long max_tsc_khz;
2032 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2034 u64 v = (u64)khz * (1000000 + ppm);
2039 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2043 /* Guest TSC same frequency as host TSC? */
2045 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2049 /* TSC scaling supported? */
2050 if (!kvm_has_tsc_control) {
2051 if (user_tsc_khz > tsc_khz) {
2052 vcpu->arch.tsc_catchup = 1;
2053 vcpu->arch.tsc_always_catchup = 1;
2056 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2061 /* TSC scaling required - calculate ratio */
2062 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2063 user_tsc_khz, tsc_khz);
2065 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2066 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2071 vcpu->arch.tsc_scaling_ratio = ratio;
2075 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2077 u32 thresh_lo, thresh_hi;
2078 int use_scaling = 0;
2080 /* tsc_khz can be zero if TSC calibration fails */
2081 if (user_tsc_khz == 0) {
2082 /* set tsc_scaling_ratio to a safe value */
2083 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2087 /* Compute a scale to convert nanoseconds in TSC cycles */
2088 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2089 &vcpu->arch.virtual_tsc_shift,
2090 &vcpu->arch.virtual_tsc_mult);
2091 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2094 * Compute the variation in TSC rate which is acceptable
2095 * within the range of tolerance and decide if the
2096 * rate being applied is within that bounds of the hardware
2097 * rate. If so, no scaling or compensation need be done.
2099 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2100 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2101 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2102 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2105 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2108 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2110 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2111 vcpu->arch.virtual_tsc_mult,
2112 vcpu->arch.virtual_tsc_shift);
2113 tsc += vcpu->arch.this_tsc_write;
2117 static inline int gtod_is_based_on_tsc(int mode)
2119 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2122 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2124 #ifdef CONFIG_X86_64
2126 struct kvm_arch *ka = &vcpu->kvm->arch;
2127 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2129 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2130 atomic_read(&vcpu->kvm->online_vcpus));
2133 * Once the masterclock is enabled, always perform request in
2134 * order to update it.
2136 * In order to enable masterclock, the host clocksource must be TSC
2137 * and the vcpus need to have matched TSCs. When that happens,
2138 * perform request to enable masterclock.
2140 if (ka->use_master_clock ||
2141 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2142 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2144 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2145 atomic_read(&vcpu->kvm->online_vcpus),
2146 ka->use_master_clock, gtod->clock.vclock_mode);
2151 * Multiply tsc by a fixed point number represented by ratio.
2153 * The most significant 64-N bits (mult) of ratio represent the
2154 * integral part of the fixed point number; the remaining N bits
2155 * (frac) represent the fractional part, ie. ratio represents a fixed
2156 * point number (mult + frac * 2^(-N)).
2158 * N equals to kvm_tsc_scaling_ratio_frac_bits.
2160 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2162 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2165 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
2168 u64 ratio = vcpu->arch.tsc_scaling_ratio;
2170 if (ratio != kvm_default_tsc_scaling_ratio)
2171 _tsc = __scale_tsc(ratio, tsc);
2175 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2177 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2181 tsc = kvm_scale_tsc(vcpu, rdtsc());
2183 return target_tsc - tsc;
2186 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2188 return vcpu->arch.l1_tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
2190 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2192 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2194 vcpu->arch.l1_tsc_offset = offset;
2195 vcpu->arch.tsc_offset = kvm_x86_ops.write_l1_tsc_offset(vcpu, offset);
2198 static inline bool kvm_check_tsc_unstable(void)
2200 #ifdef CONFIG_X86_64
2202 * TSC is marked unstable when we're running on Hyper-V,
2203 * 'TSC page' clocksource is good.
2205 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2208 return check_tsc_unstable();
2211 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2213 struct kvm *kvm = vcpu->kvm;
2214 u64 offset, ns, elapsed;
2215 unsigned long flags;
2217 bool already_matched;
2218 bool synchronizing = false;
2220 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2221 offset = kvm_compute_tsc_offset(vcpu, data);
2222 ns = get_kvmclock_base_ns();
2223 elapsed = ns - kvm->arch.last_tsc_nsec;
2225 if (vcpu->arch.virtual_tsc_khz) {
2228 * detection of vcpu initialization -- need to sync
2229 * with other vCPUs. This particularly helps to keep
2230 * kvm_clock stable after CPU hotplug
2232 synchronizing = true;
2234 u64 tsc_exp = kvm->arch.last_tsc_write +
2235 nsec_to_cycles(vcpu, elapsed);
2236 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2238 * Special case: TSC write with a small delta (1 second)
2239 * of virtual cycle time against real time is
2240 * interpreted as an attempt to synchronize the CPU.
2242 synchronizing = data < tsc_exp + tsc_hz &&
2243 data + tsc_hz > tsc_exp;
2248 * For a reliable TSC, we can match TSC offsets, and for an unstable
2249 * TSC, we add elapsed time in this computation. We could let the
2250 * compensation code attempt to catch up if we fall behind, but
2251 * it's better to try to match offsets from the beginning.
2253 if (synchronizing &&
2254 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2255 if (!kvm_check_tsc_unstable()) {
2256 offset = kvm->arch.cur_tsc_offset;
2258 u64 delta = nsec_to_cycles(vcpu, elapsed);
2260 offset = kvm_compute_tsc_offset(vcpu, data);
2263 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2266 * We split periods of matched TSC writes into generations.
2267 * For each generation, we track the original measured
2268 * nanosecond time, offset, and write, so if TSCs are in
2269 * sync, we can match exact offset, and if not, we can match
2270 * exact software computation in compute_guest_tsc()
2272 * These values are tracked in kvm->arch.cur_xxx variables.
2274 kvm->arch.cur_tsc_generation++;
2275 kvm->arch.cur_tsc_nsec = ns;
2276 kvm->arch.cur_tsc_write = data;
2277 kvm->arch.cur_tsc_offset = offset;
2282 * We also track th most recent recorded KHZ, write and time to
2283 * allow the matching interval to be extended at each write.
2285 kvm->arch.last_tsc_nsec = ns;
2286 kvm->arch.last_tsc_write = data;
2287 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2289 vcpu->arch.last_guest_tsc = data;
2291 /* Keep track of which generation this VCPU has synchronized to */
2292 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2293 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2294 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2296 kvm_vcpu_write_tsc_offset(vcpu, offset);
2297 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2299 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
2301 kvm->arch.nr_vcpus_matched_tsc = 0;
2302 } else if (!already_matched) {
2303 kvm->arch.nr_vcpus_matched_tsc++;
2306 kvm_track_tsc_matching(vcpu);
2307 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
2310 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2313 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2314 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2317 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2319 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2320 WARN_ON(adjustment < 0);
2321 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2322 adjust_tsc_offset_guest(vcpu, adjustment);
2325 #ifdef CONFIG_X86_64
2327 static u64 read_tsc(void)
2329 u64 ret = (u64)rdtsc_ordered();
2330 u64 last = pvclock_gtod_data.clock.cycle_last;
2332 if (likely(ret >= last))
2336 * GCC likes to generate cmov here, but this branch is extremely
2337 * predictable (it's just a function of time and the likely is
2338 * very likely) and there's a data dependence, so force GCC
2339 * to generate a branch instead. I don't barrier() because
2340 * we don't actually need a barrier, and if this function
2341 * ever gets inlined it will generate worse code.
2347 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2353 switch (clock->vclock_mode) {
2354 case VDSO_CLOCKMODE_HVCLOCK:
2355 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2357 if (tsc_pg_val != U64_MAX) {
2358 /* TSC page valid */
2359 *mode = VDSO_CLOCKMODE_HVCLOCK;
2360 v = (tsc_pg_val - clock->cycle_last) &
2363 /* TSC page invalid */
2364 *mode = VDSO_CLOCKMODE_NONE;
2367 case VDSO_CLOCKMODE_TSC:
2368 *mode = VDSO_CLOCKMODE_TSC;
2369 *tsc_timestamp = read_tsc();
2370 v = (*tsc_timestamp - clock->cycle_last) &
2374 *mode = VDSO_CLOCKMODE_NONE;
2377 if (*mode == VDSO_CLOCKMODE_NONE)
2378 *tsc_timestamp = v = 0;
2380 return v * clock->mult;
2383 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2385 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2391 seq = read_seqcount_begin(>od->seq);
2392 ns = gtod->raw_clock.base_cycles;
2393 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2394 ns >>= gtod->raw_clock.shift;
2395 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2396 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2402 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2404 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2410 seq = read_seqcount_begin(>od->seq);
2411 ts->tv_sec = gtod->wall_time_sec;
2412 ns = gtod->clock.base_cycles;
2413 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2414 ns >>= gtod->clock.shift;
2415 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2417 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2423 /* returns true if host is using TSC based clocksource */
2424 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2426 /* checked again under seqlock below */
2427 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2430 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2434 /* returns true if host is using TSC based clocksource */
2435 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2438 /* checked again under seqlock below */
2439 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2442 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2448 * Assuming a stable TSC across physical CPUS, and a stable TSC
2449 * across virtual CPUs, the following condition is possible.
2450 * Each numbered line represents an event visible to both
2451 * CPUs at the next numbered event.
2453 * "timespecX" represents host monotonic time. "tscX" represents
2456 * VCPU0 on CPU0 | VCPU1 on CPU1
2458 * 1. read timespec0,tsc0
2459 * 2. | timespec1 = timespec0 + N
2461 * 3. transition to guest | transition to guest
2462 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2463 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2464 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2466 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2469 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2471 * - 0 < N - M => M < N
2473 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2474 * always the case (the difference between two distinct xtime instances
2475 * might be smaller then the difference between corresponding TSC reads,
2476 * when updating guest vcpus pvclock areas).
2478 * To avoid that problem, do not allow visibility of distinct
2479 * system_timestamp/tsc_timestamp values simultaneously: use a master
2480 * copy of host monotonic time values. Update that master copy
2483 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2487 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2489 #ifdef CONFIG_X86_64
2490 struct kvm_arch *ka = &kvm->arch;
2492 bool host_tsc_clocksource, vcpus_matched;
2494 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2495 atomic_read(&kvm->online_vcpus));
2498 * If the host uses TSC clock, then passthrough TSC as stable
2501 host_tsc_clocksource = kvm_get_time_and_clockread(
2502 &ka->master_kernel_ns,
2503 &ka->master_cycle_now);
2505 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2506 && !ka->backwards_tsc_observed
2507 && !ka->boot_vcpu_runs_old_kvmclock;
2509 if (ka->use_master_clock)
2510 atomic_set(&kvm_guest_has_master_clock, 1);
2512 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2513 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2518 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2520 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2523 static void kvm_gen_update_masterclock(struct kvm *kvm)
2525 #ifdef CONFIG_X86_64
2527 struct kvm_vcpu *vcpu;
2528 struct kvm_arch *ka = &kvm->arch;
2530 spin_lock(&ka->pvclock_gtod_sync_lock);
2531 kvm_make_mclock_inprogress_request(kvm);
2532 /* no guest entries from this point */
2533 pvclock_update_vm_gtod_copy(kvm);
2535 kvm_for_each_vcpu(i, vcpu, kvm)
2536 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2538 /* guest entries allowed */
2539 kvm_for_each_vcpu(i, vcpu, kvm)
2540 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2542 spin_unlock(&ka->pvclock_gtod_sync_lock);
2546 u64 get_kvmclock_ns(struct kvm *kvm)
2548 struct kvm_arch *ka = &kvm->arch;
2549 struct pvclock_vcpu_time_info hv_clock;
2552 spin_lock(&ka->pvclock_gtod_sync_lock);
2553 if (!ka->use_master_clock) {
2554 spin_unlock(&ka->pvclock_gtod_sync_lock);
2555 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2558 hv_clock.tsc_timestamp = ka->master_cycle_now;
2559 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2560 spin_unlock(&ka->pvclock_gtod_sync_lock);
2562 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2565 if (__this_cpu_read(cpu_tsc_khz)) {
2566 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2567 &hv_clock.tsc_shift,
2568 &hv_clock.tsc_to_system_mul);
2569 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2571 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2578 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2580 struct kvm_vcpu_arch *vcpu = &v->arch;
2581 struct pvclock_vcpu_time_info guest_hv_clock;
2583 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2584 &guest_hv_clock, sizeof(guest_hv_clock))))
2587 /* This VCPU is paused, but it's legal for a guest to read another
2588 * VCPU's kvmclock, so we really have to follow the specification where
2589 * it says that version is odd if data is being modified, and even after
2592 * Version field updates must be kept separate. This is because
2593 * kvm_write_guest_cached might use a "rep movs" instruction, and
2594 * writes within a string instruction are weakly ordered. So there
2595 * are three writes overall.
2597 * As a small optimization, only write the version field in the first
2598 * and third write. The vcpu->pv_time cache is still valid, because the
2599 * version field is the first in the struct.
2601 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2603 if (guest_hv_clock.version & 1)
2604 ++guest_hv_clock.version; /* first time write, random junk */
2606 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2607 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2609 sizeof(vcpu->hv_clock.version));
2613 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2614 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2616 if (vcpu->pvclock_set_guest_stopped_request) {
2617 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2618 vcpu->pvclock_set_guest_stopped_request = false;
2621 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2623 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2625 sizeof(vcpu->hv_clock));
2629 vcpu->hv_clock.version++;
2630 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2632 sizeof(vcpu->hv_clock.version));
2635 static int kvm_guest_time_update(struct kvm_vcpu *v)
2637 unsigned long flags, tgt_tsc_khz;
2638 struct kvm_vcpu_arch *vcpu = &v->arch;
2639 struct kvm_arch *ka = &v->kvm->arch;
2641 u64 tsc_timestamp, host_tsc;
2643 bool use_master_clock;
2649 * If the host uses TSC clock, then passthrough TSC as stable
2652 spin_lock(&ka->pvclock_gtod_sync_lock);
2653 use_master_clock = ka->use_master_clock;
2654 if (use_master_clock) {
2655 host_tsc = ka->master_cycle_now;
2656 kernel_ns = ka->master_kernel_ns;
2658 spin_unlock(&ka->pvclock_gtod_sync_lock);
2660 /* Keep irq disabled to prevent changes to the clock */
2661 local_irq_save(flags);
2662 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2663 if (unlikely(tgt_tsc_khz == 0)) {
2664 local_irq_restore(flags);
2665 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2668 if (!use_master_clock) {
2670 kernel_ns = get_kvmclock_base_ns();
2673 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2676 * We may have to catch up the TSC to match elapsed wall clock
2677 * time for two reasons, even if kvmclock is used.
2678 * 1) CPU could have been running below the maximum TSC rate
2679 * 2) Broken TSC compensation resets the base at each VCPU
2680 * entry to avoid unknown leaps of TSC even when running
2681 * again on the same CPU. This may cause apparent elapsed
2682 * time to disappear, and the guest to stand still or run
2685 if (vcpu->tsc_catchup) {
2686 u64 tsc = compute_guest_tsc(v, kernel_ns);
2687 if (tsc > tsc_timestamp) {
2688 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2689 tsc_timestamp = tsc;
2693 local_irq_restore(flags);
2695 /* With all the info we got, fill in the values */
2697 if (kvm_has_tsc_control)
2698 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2700 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2701 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2702 &vcpu->hv_clock.tsc_shift,
2703 &vcpu->hv_clock.tsc_to_system_mul);
2704 vcpu->hw_tsc_khz = tgt_tsc_khz;
2707 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2708 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2709 vcpu->last_guest_tsc = tsc_timestamp;
2711 /* If the host uses TSC clocksource, then it is stable */
2713 if (use_master_clock)
2714 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2716 vcpu->hv_clock.flags = pvclock_flags;
2718 if (vcpu->pv_time_enabled)
2719 kvm_setup_pvclock_page(v);
2720 if (v == kvm_get_vcpu(v->kvm, 0))
2721 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2726 * kvmclock updates which are isolated to a given vcpu, such as
2727 * vcpu->cpu migration, should not allow system_timestamp from
2728 * the rest of the vcpus to remain static. Otherwise ntp frequency
2729 * correction applies to one vcpu's system_timestamp but not
2732 * So in those cases, request a kvmclock update for all vcpus.
2733 * We need to rate-limit these requests though, as they can
2734 * considerably slow guests that have a large number of vcpus.
2735 * The time for a remote vcpu to update its kvmclock is bound
2736 * by the delay we use to rate-limit the updates.
2739 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2741 static void kvmclock_update_fn(struct work_struct *work)
2744 struct delayed_work *dwork = to_delayed_work(work);
2745 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2746 kvmclock_update_work);
2747 struct kvm *kvm = container_of(ka, struct kvm, arch);
2748 struct kvm_vcpu *vcpu;
2750 kvm_for_each_vcpu(i, vcpu, kvm) {
2751 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2752 kvm_vcpu_kick(vcpu);
2756 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2758 struct kvm *kvm = v->kvm;
2760 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2761 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2762 KVMCLOCK_UPDATE_DELAY);
2765 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2767 static void kvmclock_sync_fn(struct work_struct *work)
2769 struct delayed_work *dwork = to_delayed_work(work);
2770 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2771 kvmclock_sync_work);
2772 struct kvm *kvm = container_of(ka, struct kvm, arch);
2774 if (!kvmclock_periodic_sync)
2777 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2778 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2779 KVMCLOCK_SYNC_PERIOD);
2783 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2785 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2787 /* McStatusWrEn enabled? */
2788 if (guest_cpuid_is_amd_or_hygon(vcpu))
2789 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2794 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2796 u64 mcg_cap = vcpu->arch.mcg_cap;
2797 unsigned bank_num = mcg_cap & 0xff;
2798 u32 msr = msr_info->index;
2799 u64 data = msr_info->data;
2802 case MSR_IA32_MCG_STATUS:
2803 vcpu->arch.mcg_status = data;
2805 case MSR_IA32_MCG_CTL:
2806 if (!(mcg_cap & MCG_CTL_P) &&
2807 (data || !msr_info->host_initiated))
2809 if (data != 0 && data != ~(u64)0)
2811 vcpu->arch.mcg_ctl = data;
2814 if (msr >= MSR_IA32_MC0_CTL &&
2815 msr < MSR_IA32_MCx_CTL(bank_num)) {
2816 u32 offset = array_index_nospec(
2817 msr - MSR_IA32_MC0_CTL,
2818 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2820 /* only 0 or all 1s can be written to IA32_MCi_CTL
2821 * some Linux kernels though clear bit 10 in bank 4 to
2822 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2823 * this to avoid an uncatched #GP in the guest
2825 if ((offset & 0x3) == 0 &&
2826 data != 0 && (data | (1 << 10)) != ~(u64)0)
2830 if (!msr_info->host_initiated &&
2831 (offset & 0x3) == 1 && data != 0) {
2832 if (!can_set_mci_status(vcpu))
2836 vcpu->arch.mce_banks[offset] = data;
2844 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2846 struct kvm *kvm = vcpu->kvm;
2847 int lm = is_long_mode(vcpu);
2848 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2849 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2850 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2851 : kvm->arch.xen_hvm_config.blob_size_32;
2852 u32 page_num = data & ~PAGE_MASK;
2853 u64 page_addr = data & PAGE_MASK;
2856 if (page_num >= blob_size)
2859 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2861 return PTR_ERR(page);
2863 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE)) {
2870 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
2872 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
2874 return (vcpu->arch.apf.msr_en_val & mask) == mask;
2877 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2879 gpa_t gpa = data & ~0x3f;
2881 /* Bits 4:5 are reserved, Should be zero */
2885 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
2886 (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
2889 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
2890 (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
2893 if (!lapic_in_kernel(vcpu))
2894 return data ? 1 : 0;
2896 vcpu->arch.apf.msr_en_val = data;
2898 if (!kvm_pv_async_pf_enabled(vcpu)) {
2899 kvm_clear_async_pf_completion_queue(vcpu);
2900 kvm_async_pf_hash_reset(vcpu);
2904 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2908 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2909 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2911 kvm_async_pf_wakeup_all(vcpu);
2916 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
2918 /* Bits 8-63 are reserved */
2922 if (!lapic_in_kernel(vcpu))
2925 vcpu->arch.apf.msr_int_val = data;
2927 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
2932 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2934 vcpu->arch.pv_time_enabled = false;
2935 vcpu->arch.time = 0;
2938 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
2940 ++vcpu->stat.tlb_flush;
2941 kvm_x86_ops.tlb_flush_all(vcpu);
2944 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
2946 ++vcpu->stat.tlb_flush;
2947 kvm_x86_ops.tlb_flush_guest(vcpu);
2950 static void record_steal_time(struct kvm_vcpu *vcpu)
2952 struct kvm_host_map map;
2953 struct kvm_steal_time *st;
2955 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2958 /* -EAGAIN is returned in atomic context so we can just return. */
2959 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
2960 &map, &vcpu->arch.st.cache, false))
2964 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
2967 * Doing a TLB flush here, on the guest's behalf, can avoid
2970 if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
2971 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2972 st->preempted & KVM_VCPU_FLUSH_TLB);
2973 if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
2974 kvm_vcpu_flush_tlb_guest(vcpu);
2977 vcpu->arch.st.preempted = 0;
2979 if (st->version & 1)
2980 st->version += 1; /* first time write, random junk */
2986 st->steal += current->sched_info.run_delay -
2987 vcpu->arch.st.last_steal;
2988 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2994 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
2997 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3000 u32 msr = msr_info->index;
3001 u64 data = msr_info->data;
3004 case MSR_AMD64_NB_CFG:
3005 case MSR_IA32_UCODE_WRITE:
3006 case MSR_VM_HSAVE_PA:
3007 case MSR_AMD64_PATCH_LOADER:
3008 case MSR_AMD64_BU_CFG2:
3009 case MSR_AMD64_DC_CFG:
3010 case MSR_F15H_EX_CFG:
3013 case MSR_IA32_UCODE_REV:
3014 if (msr_info->host_initiated)
3015 vcpu->arch.microcode_version = data;
3017 case MSR_IA32_ARCH_CAPABILITIES:
3018 if (!msr_info->host_initiated)
3020 vcpu->arch.arch_capabilities = data;
3022 case MSR_IA32_PERF_CAPABILITIES: {
3023 struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3025 if (!msr_info->host_initiated)
3027 if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3029 if (data & ~msr_ent.data)
3032 vcpu->arch.perf_capabilities = data;
3037 return set_efer(vcpu, msr_info);
3039 data &= ~(u64)0x40; /* ignore flush filter disable */
3040 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3041 data &= ~(u64)0x8; /* ignore TLB cache disable */
3043 /* Handle McStatusWrEn */
3044 if (data == BIT_ULL(18)) {
3045 vcpu->arch.msr_hwcr = data;
3046 } else if (data != 0) {
3047 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3052 case MSR_FAM10H_MMIO_CONF_BASE:
3054 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3059 case MSR_IA32_DEBUGCTLMSR:
3061 /* We support the non-activated case already */
3063 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
3064 /* Values other than LBR and BTF are vendor-specific,
3065 thus reserved and should throw a #GP */
3067 } else if (report_ignored_msrs)
3068 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
3071 case 0x200 ... 0x2ff:
3072 return kvm_mtrr_set_msr(vcpu, msr, data);
3073 case MSR_IA32_APICBASE:
3074 return kvm_set_apic_base(vcpu, msr_info);
3075 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3076 return kvm_x2apic_msr_write(vcpu, msr, data);
3077 case MSR_IA32_TSCDEADLINE:
3078 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3080 case MSR_IA32_TSC_ADJUST:
3081 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3082 if (!msr_info->host_initiated) {
3083 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3084 adjust_tsc_offset_guest(vcpu, adj);
3086 vcpu->arch.ia32_tsc_adjust_msr = data;
3089 case MSR_IA32_MISC_ENABLE:
3090 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3091 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3092 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3094 vcpu->arch.ia32_misc_enable_msr = data;
3095 kvm_update_cpuid_runtime(vcpu);
3097 vcpu->arch.ia32_misc_enable_msr = data;
3100 case MSR_IA32_SMBASE:
3101 if (!msr_info->host_initiated)
3103 vcpu->arch.smbase = data;
3105 case MSR_IA32_POWER_CTL:
3106 vcpu->arch.msr_ia32_power_ctl = data;
3109 if (msr_info->host_initiated) {
3110 kvm_synchronize_tsc(vcpu, data);
3112 u64 adj = kvm_compute_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3113 adjust_tsc_offset_guest(vcpu, adj);
3114 vcpu->arch.ia32_tsc_adjust_msr += adj;
3118 if (!msr_info->host_initiated &&
3119 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3122 * KVM supports exposing PT to the guest, but does not support
3123 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3124 * XSAVES/XRSTORS to save/restore PT MSRs.
3126 if (data & ~supported_xss)
3128 vcpu->arch.ia32_xss = data;
3131 if (!msr_info->host_initiated)
3133 vcpu->arch.smi_count = data;
3135 case MSR_KVM_WALL_CLOCK_NEW:
3136 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3139 kvm_write_wall_clock(vcpu->kvm, data);
3141 case MSR_KVM_WALL_CLOCK:
3142 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3145 kvm_write_wall_clock(vcpu->kvm, data);
3147 case MSR_KVM_SYSTEM_TIME_NEW:
3148 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3151 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3153 case MSR_KVM_SYSTEM_TIME:
3154 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3157 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3159 case MSR_KVM_ASYNC_PF_EN:
3160 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3163 if (kvm_pv_enable_async_pf(vcpu, data))
3166 case MSR_KVM_ASYNC_PF_INT:
3167 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3170 if (kvm_pv_enable_async_pf_int(vcpu, data))
3173 case MSR_KVM_ASYNC_PF_ACK:
3174 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3177 vcpu->arch.apf.pageready_pending = false;
3178 kvm_check_async_pf_completion(vcpu);
3181 case MSR_KVM_STEAL_TIME:
3182 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3185 if (unlikely(!sched_info_on()))
3188 if (data & KVM_STEAL_RESERVED_MASK)
3191 vcpu->arch.st.msr_val = data;
3193 if (!(data & KVM_MSR_ENABLED))
3196 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3199 case MSR_KVM_PV_EOI_EN:
3200 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3203 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3207 case MSR_KVM_POLL_CONTROL:
3208 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3211 /* only enable bit supported */
3212 if (data & (-1ULL << 1))
3215 vcpu->arch.msr_kvm_poll_control = data;
3218 case MSR_IA32_MCG_CTL:
3219 case MSR_IA32_MCG_STATUS:
3220 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3221 return set_msr_mce(vcpu, msr_info);
3223 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3224 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3227 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3228 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3229 if (kvm_pmu_is_valid_msr(vcpu, msr))
3230 return kvm_pmu_set_msr(vcpu, msr_info);
3232 if (pr || data != 0)
3233 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3234 "0x%x data 0x%llx\n", msr, data);
3236 case MSR_K7_CLK_CTL:
3238 * Ignore all writes to this no longer documented MSR.
3239 * Writes are only relevant for old K7 processors,
3240 * all pre-dating SVM, but a recommended workaround from
3241 * AMD for these chips. It is possible to specify the
3242 * affected processor models on the command line, hence
3243 * the need to ignore the workaround.
3246 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3247 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3248 case HV_X64_MSR_SYNDBG_OPTIONS:
3249 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3250 case HV_X64_MSR_CRASH_CTL:
3251 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3252 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3253 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3254 case HV_X64_MSR_TSC_EMULATION_STATUS:
3255 return kvm_hv_set_msr_common(vcpu, msr, data,
3256 msr_info->host_initiated);
3257 case MSR_IA32_BBL_CR_CTL3:
3258 /* Drop writes to this legacy MSR -- see rdmsr
3259 * counterpart for further detail.
3261 if (report_ignored_msrs)
3262 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3265 case MSR_AMD64_OSVW_ID_LENGTH:
3266 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3268 vcpu->arch.osvw.length = data;
3270 case MSR_AMD64_OSVW_STATUS:
3271 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3273 vcpu->arch.osvw.status = data;
3275 case MSR_PLATFORM_INFO:
3276 if (!msr_info->host_initiated ||
3277 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3278 cpuid_fault_enabled(vcpu)))
3280 vcpu->arch.msr_platform_info = data;
3282 case MSR_MISC_FEATURES_ENABLES:
3283 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3284 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3285 !supports_cpuid_fault(vcpu)))
3287 vcpu->arch.msr_misc_features_enables = data;
3290 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
3291 return xen_hvm_config(vcpu, data);
3292 if (kvm_pmu_is_valid_msr(vcpu, msr))
3293 return kvm_pmu_set_msr(vcpu, msr_info);
3294 return KVM_MSR_RET_INVALID;
3298 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3300 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3303 u64 mcg_cap = vcpu->arch.mcg_cap;
3304 unsigned bank_num = mcg_cap & 0xff;
3307 case MSR_IA32_P5_MC_ADDR:
3308 case MSR_IA32_P5_MC_TYPE:
3311 case MSR_IA32_MCG_CAP:
3312 data = vcpu->arch.mcg_cap;
3314 case MSR_IA32_MCG_CTL:
3315 if (!(mcg_cap & MCG_CTL_P) && !host)
3317 data = vcpu->arch.mcg_ctl;
3319 case MSR_IA32_MCG_STATUS:
3320 data = vcpu->arch.mcg_status;
3323 if (msr >= MSR_IA32_MC0_CTL &&
3324 msr < MSR_IA32_MCx_CTL(bank_num)) {
3325 u32 offset = array_index_nospec(
3326 msr - MSR_IA32_MC0_CTL,
3327 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3329 data = vcpu->arch.mce_banks[offset];
3338 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3340 switch (msr_info->index) {
3341 case MSR_IA32_PLATFORM_ID:
3342 case MSR_IA32_EBL_CR_POWERON:
3343 case MSR_IA32_DEBUGCTLMSR:
3344 case MSR_IA32_LASTBRANCHFROMIP:
3345 case MSR_IA32_LASTBRANCHTOIP:
3346 case MSR_IA32_LASTINTFROMIP:
3347 case MSR_IA32_LASTINTTOIP:
3349 case MSR_K8_TSEG_ADDR:
3350 case MSR_K8_TSEG_MASK:
3351 case MSR_VM_HSAVE_PA:
3352 case MSR_K8_INT_PENDING_MSG:
3353 case MSR_AMD64_NB_CFG:
3354 case MSR_FAM10H_MMIO_CONF_BASE:
3355 case MSR_AMD64_BU_CFG2:
3356 case MSR_IA32_PERF_CTL:
3357 case MSR_AMD64_DC_CFG:
3358 case MSR_F15H_EX_CFG:
3360 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3361 * limit) MSRs. Just return 0, as we do not want to expose the host
3362 * data here. Do not conditionalize this on CPUID, as KVM does not do
3363 * so for existing CPU-specific MSRs.
3365 case MSR_RAPL_POWER_UNIT:
3366 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3367 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3368 case MSR_PKG_ENERGY_STATUS: /* Total package */
3369 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3372 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3373 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3374 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3375 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3376 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3377 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3378 return kvm_pmu_get_msr(vcpu, msr_info);
3381 case MSR_IA32_UCODE_REV:
3382 msr_info->data = vcpu->arch.microcode_version;
3384 case MSR_IA32_ARCH_CAPABILITIES:
3385 if (!msr_info->host_initiated &&
3386 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3388 msr_info->data = vcpu->arch.arch_capabilities;
3390 case MSR_IA32_PERF_CAPABILITIES:
3391 if (!msr_info->host_initiated &&
3392 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3394 msr_info->data = vcpu->arch.perf_capabilities;
3396 case MSR_IA32_POWER_CTL:
3397 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3399 case MSR_IA32_TSC: {
3401 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3402 * even when not intercepted. AMD manual doesn't explicitly
3403 * state this but appears to behave the same.
3405 * On userspace reads and writes, however, we unconditionally
3406 * return L1's TSC value to ensure backwards-compatible
3407 * behavior for migration.
3409 u64 tsc_offset = msr_info->host_initiated ? vcpu->arch.l1_tsc_offset :
3410 vcpu->arch.tsc_offset;
3412 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + tsc_offset;
3416 case 0x200 ... 0x2ff:
3417 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3418 case 0xcd: /* fsb frequency */
3422 * MSR_EBC_FREQUENCY_ID
3423 * Conservative value valid for even the basic CPU models.
3424 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3425 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3426 * and 266MHz for model 3, or 4. Set Core Clock
3427 * Frequency to System Bus Frequency Ratio to 1 (bits
3428 * 31:24) even though these are only valid for CPU
3429 * models > 2, however guests may end up dividing or
3430 * multiplying by zero otherwise.
3432 case MSR_EBC_FREQUENCY_ID:
3433 msr_info->data = 1 << 24;
3435 case MSR_IA32_APICBASE:
3436 msr_info->data = kvm_get_apic_base(vcpu);
3438 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3439 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3440 case MSR_IA32_TSCDEADLINE:
3441 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3443 case MSR_IA32_TSC_ADJUST:
3444 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3446 case MSR_IA32_MISC_ENABLE:
3447 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3449 case MSR_IA32_SMBASE:
3450 if (!msr_info->host_initiated)
3452 msr_info->data = vcpu->arch.smbase;
3455 msr_info->data = vcpu->arch.smi_count;
3457 case MSR_IA32_PERF_STATUS:
3458 /* TSC increment by tick */
3459 msr_info->data = 1000ULL;
3460 /* CPU multiplier */
3461 msr_info->data |= (((uint64_t)4ULL) << 40);
3464 msr_info->data = vcpu->arch.efer;
3466 case MSR_KVM_WALL_CLOCK:
3467 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3470 msr_info->data = vcpu->kvm->arch.wall_clock;
3472 case MSR_KVM_WALL_CLOCK_NEW:
3473 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3476 msr_info->data = vcpu->kvm->arch.wall_clock;
3478 case MSR_KVM_SYSTEM_TIME:
3479 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3482 msr_info->data = vcpu->arch.time;
3484 case MSR_KVM_SYSTEM_TIME_NEW:
3485 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3488 msr_info->data = vcpu->arch.time;
3490 case MSR_KVM_ASYNC_PF_EN:
3491 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3494 msr_info->data = vcpu->arch.apf.msr_en_val;
3496 case MSR_KVM_ASYNC_PF_INT:
3497 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3500 msr_info->data = vcpu->arch.apf.msr_int_val;
3502 case MSR_KVM_ASYNC_PF_ACK:
3503 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3508 case MSR_KVM_STEAL_TIME:
3509 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3512 msr_info->data = vcpu->arch.st.msr_val;
3514 case MSR_KVM_PV_EOI_EN:
3515 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3518 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3520 case MSR_KVM_POLL_CONTROL:
3521 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3524 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3526 case MSR_IA32_P5_MC_ADDR:
3527 case MSR_IA32_P5_MC_TYPE:
3528 case MSR_IA32_MCG_CAP:
3529 case MSR_IA32_MCG_CTL:
3530 case MSR_IA32_MCG_STATUS:
3531 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3532 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3533 msr_info->host_initiated);
3535 if (!msr_info->host_initiated &&
3536 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3538 msr_info->data = vcpu->arch.ia32_xss;
3540 case MSR_K7_CLK_CTL:
3542 * Provide expected ramp-up count for K7. All other
3543 * are set to zero, indicating minimum divisors for
3546 * This prevents guest kernels on AMD host with CPU
3547 * type 6, model 8 and higher from exploding due to
3548 * the rdmsr failing.
3550 msr_info->data = 0x20000000;
3552 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3553 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3554 case HV_X64_MSR_SYNDBG_OPTIONS:
3555 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3556 case HV_X64_MSR_CRASH_CTL:
3557 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3558 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3559 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3560 case HV_X64_MSR_TSC_EMULATION_STATUS:
3561 return kvm_hv_get_msr_common(vcpu,
3562 msr_info->index, &msr_info->data,
3563 msr_info->host_initiated);
3564 case MSR_IA32_BBL_CR_CTL3:
3565 /* This legacy MSR exists but isn't fully documented in current
3566 * silicon. It is however accessed by winxp in very narrow
3567 * scenarios where it sets bit #19, itself documented as
3568 * a "reserved" bit. Best effort attempt to source coherent
3569 * read data here should the balance of the register be
3570 * interpreted by the guest:
3572 * L2 cache control register 3: 64GB range, 256KB size,
3573 * enabled, latency 0x1, configured
3575 msr_info->data = 0xbe702111;
3577 case MSR_AMD64_OSVW_ID_LENGTH:
3578 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3580 msr_info->data = vcpu->arch.osvw.length;
3582 case MSR_AMD64_OSVW_STATUS:
3583 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3585 msr_info->data = vcpu->arch.osvw.status;
3587 case MSR_PLATFORM_INFO:
3588 if (!msr_info->host_initiated &&
3589 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3591 msr_info->data = vcpu->arch.msr_platform_info;
3593 case MSR_MISC_FEATURES_ENABLES:
3594 msr_info->data = vcpu->arch.msr_misc_features_enables;
3597 msr_info->data = vcpu->arch.msr_hwcr;
3600 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3601 return kvm_pmu_get_msr(vcpu, msr_info);
3602 return KVM_MSR_RET_INVALID;
3606 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3609 * Read or write a bunch of msrs. All parameters are kernel addresses.
3611 * @return number of msrs set successfully.
3613 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3614 struct kvm_msr_entry *entries,
3615 int (*do_msr)(struct kvm_vcpu *vcpu,
3616 unsigned index, u64 *data))
3620 for (i = 0; i < msrs->nmsrs; ++i)
3621 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3628 * Read or write a bunch of msrs. Parameters are user addresses.
3630 * @return number of msrs set successfully.
3632 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3633 int (*do_msr)(struct kvm_vcpu *vcpu,
3634 unsigned index, u64 *data),
3637 struct kvm_msrs msrs;
3638 struct kvm_msr_entry *entries;
3643 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3647 if (msrs.nmsrs >= MAX_IO_MSRS)
3650 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3651 entries = memdup_user(user_msrs->entries, size);
3652 if (IS_ERR(entries)) {
3653 r = PTR_ERR(entries);
3657 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3662 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3673 static inline bool kvm_can_mwait_in_guest(void)
3675 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3676 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3677 boot_cpu_has(X86_FEATURE_ARAT);
3680 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3685 case KVM_CAP_IRQCHIP:
3687 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3688 case KVM_CAP_SET_TSS_ADDR:
3689 case KVM_CAP_EXT_CPUID:
3690 case KVM_CAP_EXT_EMUL_CPUID:
3691 case KVM_CAP_CLOCKSOURCE:
3693 case KVM_CAP_NOP_IO_DELAY:
3694 case KVM_CAP_MP_STATE:
3695 case KVM_CAP_SYNC_MMU:
3696 case KVM_CAP_USER_NMI:
3697 case KVM_CAP_REINJECT_CONTROL:
3698 case KVM_CAP_IRQ_INJECT_STATUS:
3699 case KVM_CAP_IOEVENTFD:
3700 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3702 case KVM_CAP_PIT_STATE2:
3703 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3704 case KVM_CAP_XEN_HVM:
3705 case KVM_CAP_VCPU_EVENTS:
3706 case KVM_CAP_HYPERV:
3707 case KVM_CAP_HYPERV_VAPIC:
3708 case KVM_CAP_HYPERV_SPIN:
3709 case KVM_CAP_HYPERV_SYNIC:
3710 case KVM_CAP_HYPERV_SYNIC2:
3711 case KVM_CAP_HYPERV_VP_INDEX:
3712 case KVM_CAP_HYPERV_EVENTFD:
3713 case KVM_CAP_HYPERV_TLBFLUSH:
3714 case KVM_CAP_HYPERV_SEND_IPI:
3715 case KVM_CAP_HYPERV_CPUID:
3716 case KVM_CAP_PCI_SEGMENT:
3717 case KVM_CAP_DEBUGREGS:
3718 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3720 case KVM_CAP_ASYNC_PF:
3721 case KVM_CAP_ASYNC_PF_INT:
3722 case KVM_CAP_GET_TSC_KHZ:
3723 case KVM_CAP_KVMCLOCK_CTRL:
3724 case KVM_CAP_READONLY_MEM:
3725 case KVM_CAP_HYPERV_TIME:
3726 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3727 case KVM_CAP_TSC_DEADLINE_TIMER:
3728 case KVM_CAP_DISABLE_QUIRKS:
3729 case KVM_CAP_SET_BOOT_CPU_ID:
3730 case KVM_CAP_SPLIT_IRQCHIP:
3731 case KVM_CAP_IMMEDIATE_EXIT:
3732 case KVM_CAP_PMU_EVENT_FILTER:
3733 case KVM_CAP_GET_MSR_FEATURES:
3734 case KVM_CAP_MSR_PLATFORM_INFO:
3735 case KVM_CAP_EXCEPTION_PAYLOAD:
3736 case KVM_CAP_SET_GUEST_DEBUG:
3737 case KVM_CAP_LAST_CPU:
3738 case KVM_CAP_X86_USER_SPACE_MSR:
3739 case KVM_CAP_X86_MSR_FILTER:
3740 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
3743 case KVM_CAP_SYNC_REGS:
3744 r = KVM_SYNC_X86_VALID_FIELDS;
3746 case KVM_CAP_ADJUST_CLOCK:
3747 r = KVM_CLOCK_TSC_STABLE;
3749 case KVM_CAP_X86_DISABLE_EXITS:
3750 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3751 KVM_X86_DISABLE_EXITS_CSTATE;
3752 if(kvm_can_mwait_in_guest())
3753 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3755 case KVM_CAP_X86_SMM:
3756 /* SMBASE is usually relocated above 1M on modern chipsets,
3757 * and SMM handlers might indeed rely on 4G segment limits,
3758 * so do not report SMM to be available if real mode is
3759 * emulated via vm86 mode. Still, do not go to great lengths
3760 * to avoid userspace's usage of the feature, because it is a
3761 * fringe case that is not enabled except via specific settings
3762 * of the module parameters.
3764 r = kvm_x86_ops.has_emulated_msr(MSR_IA32_SMBASE);
3767 r = !kvm_x86_ops.cpu_has_accelerated_tpr();
3769 case KVM_CAP_NR_VCPUS:
3770 r = KVM_SOFT_MAX_VCPUS;
3772 case KVM_CAP_MAX_VCPUS:
3775 case KVM_CAP_MAX_VCPU_ID:
3776 r = KVM_MAX_VCPU_ID;
3778 case KVM_CAP_PV_MMU: /* obsolete */
3782 r = KVM_MAX_MCE_BANKS;
3785 r = boot_cpu_has(X86_FEATURE_XSAVE);
3787 case KVM_CAP_TSC_CONTROL:
3788 r = kvm_has_tsc_control;
3790 case KVM_CAP_X2APIC_API:
3791 r = KVM_X2APIC_API_VALID_FLAGS;
3793 case KVM_CAP_NESTED_STATE:
3794 r = kvm_x86_ops.nested_ops->get_state ?
3795 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
3797 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3798 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
3800 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3801 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
3803 case KVM_CAP_SMALLER_MAXPHYADDR:
3804 r = (int) allow_smaller_maxphyaddr;
3806 case KVM_CAP_STEAL_TIME:
3807 r = sched_info_on();
3816 long kvm_arch_dev_ioctl(struct file *filp,
3817 unsigned int ioctl, unsigned long arg)
3819 void __user *argp = (void __user *)arg;
3823 case KVM_GET_MSR_INDEX_LIST: {
3824 struct kvm_msr_list __user *user_msr_list = argp;
3825 struct kvm_msr_list msr_list;
3829 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3832 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3833 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3836 if (n < msr_list.nmsrs)
3839 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3840 num_msrs_to_save * sizeof(u32)))
3842 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3844 num_emulated_msrs * sizeof(u32)))
3849 case KVM_GET_SUPPORTED_CPUID:
3850 case KVM_GET_EMULATED_CPUID: {
3851 struct kvm_cpuid2 __user *cpuid_arg = argp;
3852 struct kvm_cpuid2 cpuid;
3855 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3858 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3864 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3869 case KVM_X86_GET_MCE_CAP_SUPPORTED:
3871 if (copy_to_user(argp, &kvm_mce_cap_supported,
3872 sizeof(kvm_mce_cap_supported)))
3876 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3877 struct kvm_msr_list __user *user_msr_list = argp;
3878 struct kvm_msr_list msr_list;
3882 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3885 msr_list.nmsrs = num_msr_based_features;
3886 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3889 if (n < msr_list.nmsrs)
3892 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3893 num_msr_based_features * sizeof(u32)))
3899 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3909 static void wbinvd_ipi(void *garbage)
3914 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3916 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3919 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3921 /* Address WBINVD may be executed by guest */
3922 if (need_emulate_wbinvd(vcpu)) {
3923 if (kvm_x86_ops.has_wbinvd_exit())
3924 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3925 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3926 smp_call_function_single(vcpu->cpu,
3927 wbinvd_ipi, NULL, 1);
3930 kvm_x86_ops.vcpu_load(vcpu, cpu);
3932 /* Save host pkru register if supported */
3933 vcpu->arch.host_pkru = read_pkru();
3935 /* Apply any externally detected TSC adjustments (due to suspend) */
3936 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3937 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3938 vcpu->arch.tsc_offset_adjustment = 0;
3939 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3942 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3943 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3944 rdtsc() - vcpu->arch.last_host_tsc;
3946 mark_tsc_unstable("KVM discovered backwards TSC");
3948 if (kvm_check_tsc_unstable()) {
3949 u64 offset = kvm_compute_tsc_offset(vcpu,
3950 vcpu->arch.last_guest_tsc);
3951 kvm_vcpu_write_tsc_offset(vcpu, offset);
3952 vcpu->arch.tsc_catchup = 1;
3955 if (kvm_lapic_hv_timer_in_use(vcpu))
3956 kvm_lapic_restart_hv_timer(vcpu);
3959 * On a host with synchronized TSC, there is no need to update
3960 * kvmclock on vcpu->cpu migration
3962 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3963 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3964 if (vcpu->cpu != cpu)
3965 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3969 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3972 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3974 struct kvm_host_map map;
3975 struct kvm_steal_time *st;
3977 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3980 if (vcpu->arch.st.preempted)
3983 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
3984 &vcpu->arch.st.cache, true))
3988 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3990 st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
3992 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
3995 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3999 if (vcpu->preempted)
4000 vcpu->arch.preempted_in_kernel = !kvm_x86_ops.get_cpl(vcpu);
4003 * Disable page faults because we're in atomic context here.
4004 * kvm_write_guest_offset_cached() would call might_fault()
4005 * that relies on pagefault_disable() to tell if there's a
4006 * bug. NOTE: the write to guest memory may not go through if
4007 * during postcopy live migration or if there's heavy guest
4010 pagefault_disable();
4012 * kvm_memslots() will be called by
4013 * kvm_write_guest_offset_cached() so take the srcu lock.
4015 idx = srcu_read_lock(&vcpu->kvm->srcu);
4016 kvm_steal_time_set_preempted(vcpu);
4017 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4019 kvm_x86_ops.vcpu_put(vcpu);
4020 vcpu->arch.last_host_tsc = rdtsc();
4022 * If userspace has set any breakpoints or watchpoints, dr6 is restored
4023 * on every vmexit, but if not, we might have a stale dr6 from the
4024 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
4029 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4030 struct kvm_lapic_state *s)
4032 if (vcpu->arch.apicv_active)
4033 kvm_x86_ops.sync_pir_to_irr(vcpu);
4035 return kvm_apic_get_state(vcpu, s);
4038 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4039 struct kvm_lapic_state *s)
4043 r = kvm_apic_set_state(vcpu, s);
4046 update_cr8_intercept(vcpu);
4051 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4053 return (!lapic_in_kernel(vcpu) ||
4054 kvm_apic_accept_pic_intr(vcpu));
4058 * if userspace requested an interrupt window, check that the
4059 * interrupt window is open.
4061 * No need to exit to userspace if we already have an interrupt queued.
4063 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4065 return kvm_arch_interrupt_allowed(vcpu) &&
4066 !kvm_cpu_has_interrupt(vcpu) &&
4067 !kvm_event_needs_reinjection(vcpu) &&
4068 kvm_cpu_accept_dm_intr(vcpu);
4071 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4072 struct kvm_interrupt *irq)
4074 if (irq->irq >= KVM_NR_INTERRUPTS)
4077 if (!irqchip_in_kernel(vcpu->kvm)) {
4078 kvm_queue_interrupt(vcpu, irq->irq, false);
4079 kvm_make_request(KVM_REQ_EVENT, vcpu);
4084 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4085 * fail for in-kernel 8259.
4087 if (pic_in_kernel(vcpu->kvm))
4090 if (vcpu->arch.pending_external_vector != -1)
4093 vcpu->arch.pending_external_vector = irq->irq;
4094 kvm_make_request(KVM_REQ_EVENT, vcpu);
4098 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4100 kvm_inject_nmi(vcpu);
4105 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4107 kvm_make_request(KVM_REQ_SMI, vcpu);
4112 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4113 struct kvm_tpr_access_ctl *tac)
4117 vcpu->arch.tpr_access_reporting = !!tac->enabled;
4121 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4125 unsigned bank_num = mcg_cap & 0xff, bank;
4128 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4130 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4133 vcpu->arch.mcg_cap = mcg_cap;
4134 /* Init IA32_MCG_CTL to all 1s */
4135 if (mcg_cap & MCG_CTL_P)
4136 vcpu->arch.mcg_ctl = ~(u64)0;
4137 /* Init IA32_MCi_CTL to all 1s */
4138 for (bank = 0; bank < bank_num; bank++)
4139 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4141 kvm_x86_ops.setup_mce(vcpu);
4146 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4147 struct kvm_x86_mce *mce)
4149 u64 mcg_cap = vcpu->arch.mcg_cap;
4150 unsigned bank_num = mcg_cap & 0xff;
4151 u64 *banks = vcpu->arch.mce_banks;
4153 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4156 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4157 * reporting is disabled
4159 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4160 vcpu->arch.mcg_ctl != ~(u64)0)
4162 banks += 4 * mce->bank;
4164 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4165 * reporting is disabled for the bank
4167 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4169 if (mce->status & MCI_STATUS_UC) {
4170 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4171 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4172 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4175 if (banks[1] & MCI_STATUS_VAL)
4176 mce->status |= MCI_STATUS_OVER;
4177 banks[2] = mce->addr;
4178 banks[3] = mce->misc;
4179 vcpu->arch.mcg_status = mce->mcg_status;
4180 banks[1] = mce->status;
4181 kvm_queue_exception(vcpu, MC_VECTOR);
4182 } else if (!(banks[1] & MCI_STATUS_VAL)
4183 || !(banks[1] & MCI_STATUS_UC)) {
4184 if (banks[1] & MCI_STATUS_VAL)
4185 mce->status |= MCI_STATUS_OVER;
4186 banks[2] = mce->addr;
4187 banks[3] = mce->misc;
4188 banks[1] = mce->status;
4190 banks[1] |= MCI_STATUS_OVER;
4194 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4195 struct kvm_vcpu_events *events)
4200 * In guest mode, payload delivery should be deferred,
4201 * so that the L1 hypervisor can intercept #PF before
4202 * CR2 is modified (or intercept #DB before DR6 is
4203 * modified under nVMX). Unless the per-VM capability,
4204 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4205 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4206 * opportunistically defer the exception payload, deliver it if the
4207 * capability hasn't been requested before processing a
4208 * KVM_GET_VCPU_EVENTS.
4210 if (!vcpu->kvm->arch.exception_payload_enabled &&
4211 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4212 kvm_deliver_exception_payload(vcpu);
4215 * The API doesn't provide the instruction length for software
4216 * exceptions, so don't report them. As long as the guest RIP
4217 * isn't advanced, we should expect to encounter the exception
4220 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4221 events->exception.injected = 0;
4222 events->exception.pending = 0;
4224 events->exception.injected = vcpu->arch.exception.injected;
4225 events->exception.pending = vcpu->arch.exception.pending;
4227 * For ABI compatibility, deliberately conflate
4228 * pending and injected exceptions when
4229 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4231 if (!vcpu->kvm->arch.exception_payload_enabled)
4232 events->exception.injected |=
4233 vcpu->arch.exception.pending;
4235 events->exception.nr = vcpu->arch.exception.nr;
4236 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4237 events->exception.error_code = vcpu->arch.exception.error_code;
4238 events->exception_has_payload = vcpu->arch.exception.has_payload;
4239 events->exception_payload = vcpu->arch.exception.payload;
4241 events->interrupt.injected =
4242 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4243 events->interrupt.nr = vcpu->arch.interrupt.nr;
4244 events->interrupt.soft = 0;
4245 events->interrupt.shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
4247 events->nmi.injected = vcpu->arch.nmi_injected;
4248 events->nmi.pending = vcpu->arch.nmi_pending != 0;
4249 events->nmi.masked = kvm_x86_ops.get_nmi_mask(vcpu);
4250 events->nmi.pad = 0;
4252 events->sipi_vector = 0; /* never valid when reporting to user space */
4254 events->smi.smm = is_smm(vcpu);
4255 events->smi.pending = vcpu->arch.smi_pending;
4256 events->smi.smm_inside_nmi =
4257 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4258 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4260 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4261 | KVM_VCPUEVENT_VALID_SHADOW
4262 | KVM_VCPUEVENT_VALID_SMM);
4263 if (vcpu->kvm->arch.exception_payload_enabled)
4264 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4266 memset(&events->reserved, 0, sizeof(events->reserved));
4269 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
4271 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4272 struct kvm_vcpu_events *events)
4274 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4275 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4276 | KVM_VCPUEVENT_VALID_SHADOW
4277 | KVM_VCPUEVENT_VALID_SMM
4278 | KVM_VCPUEVENT_VALID_PAYLOAD))
4281 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4282 if (!vcpu->kvm->arch.exception_payload_enabled)
4284 if (events->exception.pending)
4285 events->exception.injected = 0;
4287 events->exception_has_payload = 0;
4289 events->exception.pending = 0;
4290 events->exception_has_payload = 0;
4293 if ((events->exception.injected || events->exception.pending) &&
4294 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4297 /* INITs are latched while in SMM */
4298 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4299 (events->smi.smm || events->smi.pending) &&
4300 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4304 vcpu->arch.exception.injected = events->exception.injected;
4305 vcpu->arch.exception.pending = events->exception.pending;
4306 vcpu->arch.exception.nr = events->exception.nr;
4307 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4308 vcpu->arch.exception.error_code = events->exception.error_code;
4309 vcpu->arch.exception.has_payload = events->exception_has_payload;
4310 vcpu->arch.exception.payload = events->exception_payload;
4312 vcpu->arch.interrupt.injected = events->interrupt.injected;
4313 vcpu->arch.interrupt.nr = events->interrupt.nr;
4314 vcpu->arch.interrupt.soft = events->interrupt.soft;
4315 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4316 kvm_x86_ops.set_interrupt_shadow(vcpu,
4317 events->interrupt.shadow);
4319 vcpu->arch.nmi_injected = events->nmi.injected;
4320 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4321 vcpu->arch.nmi_pending = events->nmi.pending;
4322 kvm_x86_ops.set_nmi_mask(vcpu, events->nmi.masked);
4324 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4325 lapic_in_kernel(vcpu))
4326 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4328 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4329 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
4330 if (events->smi.smm)
4331 vcpu->arch.hflags |= HF_SMM_MASK;
4333 vcpu->arch.hflags &= ~HF_SMM_MASK;
4334 kvm_smm_changed(vcpu);
4337 vcpu->arch.smi_pending = events->smi.pending;
4339 if (events->smi.smm) {
4340 if (events->smi.smm_inside_nmi)
4341 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4343 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4346 if (lapic_in_kernel(vcpu)) {
4347 if (events->smi.latched_init)
4348 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4350 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4354 kvm_make_request(KVM_REQ_EVENT, vcpu);
4359 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4360 struct kvm_debugregs *dbgregs)
4364 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4365 kvm_get_dr(vcpu, 6, &val);
4367 dbgregs->dr7 = vcpu->arch.dr7;
4369 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4372 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4373 struct kvm_debugregs *dbgregs)
4378 if (dbgregs->dr6 & ~0xffffffffull)
4380 if (dbgregs->dr7 & ~0xffffffffull)
4383 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4384 kvm_update_dr0123(vcpu);
4385 vcpu->arch.dr6 = dbgregs->dr6;
4386 vcpu->arch.dr7 = dbgregs->dr7;
4387 kvm_update_dr7(vcpu);
4392 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4394 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4396 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4397 u64 xstate_bv = xsave->header.xfeatures;
4401 * Copy legacy XSAVE area, to avoid complications with CPUID
4402 * leaves 0 and 1 in the loop below.
4404 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4407 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4408 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4411 * Copy each region from the possibly compacted offset to the
4412 * non-compacted offset.
4414 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4416 u64 xfeature_mask = valid & -valid;
4417 int xfeature_nr = fls64(xfeature_mask) - 1;
4418 void *src = get_xsave_addr(xsave, xfeature_nr);
4421 u32 size, offset, ecx, edx;
4422 cpuid_count(XSTATE_CPUID, xfeature_nr,
4423 &size, &offset, &ecx, &edx);
4424 if (xfeature_nr == XFEATURE_PKRU)
4425 memcpy(dest + offset, &vcpu->arch.pkru,
4426 sizeof(vcpu->arch.pkru));
4428 memcpy(dest + offset, src, size);
4432 valid -= xfeature_mask;
4436 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4438 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4439 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4443 * Copy legacy XSAVE area, to avoid complications with CPUID
4444 * leaves 0 and 1 in the loop below.
4446 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4448 /* Set XSTATE_BV and possibly XCOMP_BV. */
4449 xsave->header.xfeatures = xstate_bv;
4450 if (boot_cpu_has(X86_FEATURE_XSAVES))
4451 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4454 * Copy each region from the non-compacted offset to the
4455 * possibly compacted offset.
4457 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4459 u64 xfeature_mask = valid & -valid;
4460 int xfeature_nr = fls64(xfeature_mask) - 1;
4461 void *dest = get_xsave_addr(xsave, xfeature_nr);
4464 u32 size, offset, ecx, edx;
4465 cpuid_count(XSTATE_CPUID, xfeature_nr,
4466 &size, &offset, &ecx, &edx);
4467 if (xfeature_nr == XFEATURE_PKRU)
4468 memcpy(&vcpu->arch.pkru, src + offset,
4469 sizeof(vcpu->arch.pkru));
4471 memcpy(dest, src + offset, size);
4474 valid -= xfeature_mask;
4478 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4479 struct kvm_xsave *guest_xsave)
4481 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4482 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4483 fill_xsave((u8 *) guest_xsave->region, vcpu);
4485 memcpy(guest_xsave->region,
4486 &vcpu->arch.guest_fpu->state.fxsave,
4487 sizeof(struct fxregs_state));
4488 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4489 XFEATURE_MASK_FPSSE;
4493 #define XSAVE_MXCSR_OFFSET 24
4495 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4496 struct kvm_xsave *guest_xsave)
4499 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4500 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4502 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4504 * Here we allow setting states that are not present in
4505 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4506 * with old userspace.
4508 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4510 load_xsave(vcpu, (u8 *)guest_xsave->region);
4512 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4513 mxcsr & ~mxcsr_feature_mask)
4515 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4516 guest_xsave->region, sizeof(struct fxregs_state));
4521 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4522 struct kvm_xcrs *guest_xcrs)
4524 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4525 guest_xcrs->nr_xcrs = 0;
4529 guest_xcrs->nr_xcrs = 1;
4530 guest_xcrs->flags = 0;
4531 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4532 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4535 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4536 struct kvm_xcrs *guest_xcrs)
4540 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4543 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4546 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4547 /* Only support XCR0 currently */
4548 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4549 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4550 guest_xcrs->xcrs[i].value);
4559 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4560 * stopped by the hypervisor. This function will be called from the host only.
4561 * EINVAL is returned when the host attempts to set the flag for a guest that
4562 * does not support pv clocks.
4564 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4566 if (!vcpu->arch.pv_time_enabled)
4568 vcpu->arch.pvclock_set_guest_stopped_request = true;
4569 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4573 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4574 struct kvm_enable_cap *cap)
4577 uint16_t vmcs_version;
4578 void __user *user_ptr;
4584 case KVM_CAP_HYPERV_SYNIC2:
4589 case KVM_CAP_HYPERV_SYNIC:
4590 if (!irqchip_in_kernel(vcpu->kvm))
4592 return kvm_hv_activate_synic(vcpu, cap->cap ==
4593 KVM_CAP_HYPERV_SYNIC2);
4594 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4595 if (!kvm_x86_ops.nested_ops->enable_evmcs)
4597 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4599 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4600 if (copy_to_user(user_ptr, &vmcs_version,
4601 sizeof(vmcs_version)))
4605 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4606 if (!kvm_x86_ops.enable_direct_tlbflush)
4609 return kvm_x86_ops.enable_direct_tlbflush(vcpu);
4611 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4612 vcpu->arch.pv_cpuid.enforce = cap->args[0];
4613 if (vcpu->arch.pv_cpuid.enforce)
4614 kvm_update_pv_runtime(vcpu);
4623 long kvm_arch_vcpu_ioctl(struct file *filp,
4624 unsigned int ioctl, unsigned long arg)
4626 struct kvm_vcpu *vcpu = filp->private_data;
4627 void __user *argp = (void __user *)arg;
4630 struct kvm_lapic_state *lapic;
4631 struct kvm_xsave *xsave;
4632 struct kvm_xcrs *xcrs;
4640 case KVM_GET_LAPIC: {
4642 if (!lapic_in_kernel(vcpu))
4644 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4645 GFP_KERNEL_ACCOUNT);
4650 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4654 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4659 case KVM_SET_LAPIC: {
4661 if (!lapic_in_kernel(vcpu))
4663 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4664 if (IS_ERR(u.lapic)) {
4665 r = PTR_ERR(u.lapic);
4669 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4672 case KVM_INTERRUPT: {
4673 struct kvm_interrupt irq;
4676 if (copy_from_user(&irq, argp, sizeof(irq)))
4678 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4682 r = kvm_vcpu_ioctl_nmi(vcpu);
4686 r = kvm_vcpu_ioctl_smi(vcpu);
4689 case KVM_SET_CPUID: {
4690 struct kvm_cpuid __user *cpuid_arg = argp;
4691 struct kvm_cpuid cpuid;
4694 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4696 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4699 case KVM_SET_CPUID2: {
4700 struct kvm_cpuid2 __user *cpuid_arg = argp;
4701 struct kvm_cpuid2 cpuid;
4704 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4706 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4707 cpuid_arg->entries);
4710 case KVM_GET_CPUID2: {
4711 struct kvm_cpuid2 __user *cpuid_arg = argp;
4712 struct kvm_cpuid2 cpuid;
4715 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4717 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4718 cpuid_arg->entries);
4722 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4727 case KVM_GET_MSRS: {
4728 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4729 r = msr_io(vcpu, argp, do_get_msr, 1);
4730 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4733 case KVM_SET_MSRS: {
4734 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4735 r = msr_io(vcpu, argp, do_set_msr, 0);
4736 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4739 case KVM_TPR_ACCESS_REPORTING: {
4740 struct kvm_tpr_access_ctl tac;
4743 if (copy_from_user(&tac, argp, sizeof(tac)))
4745 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4749 if (copy_to_user(argp, &tac, sizeof(tac)))
4754 case KVM_SET_VAPIC_ADDR: {
4755 struct kvm_vapic_addr va;
4759 if (!lapic_in_kernel(vcpu))
4762 if (copy_from_user(&va, argp, sizeof(va)))
4764 idx = srcu_read_lock(&vcpu->kvm->srcu);
4765 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4766 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4769 case KVM_X86_SETUP_MCE: {
4773 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4775 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4778 case KVM_X86_SET_MCE: {
4779 struct kvm_x86_mce mce;
4782 if (copy_from_user(&mce, argp, sizeof(mce)))
4784 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4787 case KVM_GET_VCPU_EVENTS: {
4788 struct kvm_vcpu_events events;
4790 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4793 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4798 case KVM_SET_VCPU_EVENTS: {
4799 struct kvm_vcpu_events events;
4802 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4805 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4808 case KVM_GET_DEBUGREGS: {
4809 struct kvm_debugregs dbgregs;
4811 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4814 if (copy_to_user(argp, &dbgregs,
4815 sizeof(struct kvm_debugregs)))
4820 case KVM_SET_DEBUGREGS: {
4821 struct kvm_debugregs dbgregs;
4824 if (copy_from_user(&dbgregs, argp,
4825 sizeof(struct kvm_debugregs)))
4828 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4831 case KVM_GET_XSAVE: {
4832 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4837 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4840 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4845 case KVM_SET_XSAVE: {
4846 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4847 if (IS_ERR(u.xsave)) {
4848 r = PTR_ERR(u.xsave);
4852 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4855 case KVM_GET_XCRS: {
4856 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4861 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4864 if (copy_to_user(argp, u.xcrs,
4865 sizeof(struct kvm_xcrs)))
4870 case KVM_SET_XCRS: {
4871 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4872 if (IS_ERR(u.xcrs)) {
4873 r = PTR_ERR(u.xcrs);
4877 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4880 case KVM_SET_TSC_KHZ: {
4884 user_tsc_khz = (u32)arg;
4886 if (kvm_has_tsc_control &&
4887 user_tsc_khz >= kvm_max_guest_tsc_khz)
4890 if (user_tsc_khz == 0)
4891 user_tsc_khz = tsc_khz;
4893 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4898 case KVM_GET_TSC_KHZ: {
4899 r = vcpu->arch.virtual_tsc_khz;
4902 case KVM_KVMCLOCK_CTRL: {
4903 r = kvm_set_guest_paused(vcpu);
4906 case KVM_ENABLE_CAP: {
4907 struct kvm_enable_cap cap;
4910 if (copy_from_user(&cap, argp, sizeof(cap)))
4912 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4915 case KVM_GET_NESTED_STATE: {
4916 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4920 if (!kvm_x86_ops.nested_ops->get_state)
4923 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4925 if (get_user(user_data_size, &user_kvm_nested_state->size))
4928 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
4933 if (r > user_data_size) {
4934 if (put_user(r, &user_kvm_nested_state->size))
4944 case KVM_SET_NESTED_STATE: {
4945 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4946 struct kvm_nested_state kvm_state;
4950 if (!kvm_x86_ops.nested_ops->set_state)
4954 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4958 if (kvm_state.size < sizeof(kvm_state))
4961 if (kvm_state.flags &
4962 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4963 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
4964 | KVM_STATE_NESTED_GIF_SET))
4967 /* nested_run_pending implies guest_mode. */
4968 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4969 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4972 idx = srcu_read_lock(&vcpu->kvm->srcu);
4973 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
4974 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4977 case KVM_GET_SUPPORTED_HV_CPUID: {
4978 struct kvm_cpuid2 __user *cpuid_arg = argp;
4979 struct kvm_cpuid2 cpuid;
4982 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4985 r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4986 cpuid_arg->entries);
4991 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
5006 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5008 return VM_FAULT_SIGBUS;
5011 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5015 if (addr > (unsigned int)(-3 * PAGE_SIZE))
5017 ret = kvm_x86_ops.set_tss_addr(kvm, addr);
5021 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5024 return kvm_x86_ops.set_identity_map_addr(kvm, ident_addr);
5027 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5028 unsigned long kvm_nr_mmu_pages)
5030 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5033 mutex_lock(&kvm->slots_lock);
5035 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5036 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5038 mutex_unlock(&kvm->slots_lock);
5042 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5044 return kvm->arch.n_max_mmu_pages;
5047 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5049 struct kvm_pic *pic = kvm->arch.vpic;
5053 switch (chip->chip_id) {
5054 case KVM_IRQCHIP_PIC_MASTER:
5055 memcpy(&chip->chip.pic, &pic->pics[0],
5056 sizeof(struct kvm_pic_state));
5058 case KVM_IRQCHIP_PIC_SLAVE:
5059 memcpy(&chip->chip.pic, &pic->pics[1],
5060 sizeof(struct kvm_pic_state));
5062 case KVM_IRQCHIP_IOAPIC:
5063 kvm_get_ioapic(kvm, &chip->chip.ioapic);
5072 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5074 struct kvm_pic *pic = kvm->arch.vpic;
5078 switch (chip->chip_id) {
5079 case KVM_IRQCHIP_PIC_MASTER:
5080 spin_lock(&pic->lock);
5081 memcpy(&pic->pics[0], &chip->chip.pic,
5082 sizeof(struct kvm_pic_state));
5083 spin_unlock(&pic->lock);
5085 case KVM_IRQCHIP_PIC_SLAVE:
5086 spin_lock(&pic->lock);
5087 memcpy(&pic->pics[1], &chip->chip.pic,
5088 sizeof(struct kvm_pic_state));
5089 spin_unlock(&pic->lock);
5091 case KVM_IRQCHIP_IOAPIC:
5092 kvm_set_ioapic(kvm, &chip->chip.ioapic);
5098 kvm_pic_update_irq(pic);
5102 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5104 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5106 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5108 mutex_lock(&kps->lock);
5109 memcpy(ps, &kps->channels, sizeof(*ps));
5110 mutex_unlock(&kps->lock);
5114 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5117 struct kvm_pit *pit = kvm->arch.vpit;
5119 mutex_lock(&pit->pit_state.lock);
5120 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5121 for (i = 0; i < 3; i++)
5122 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5123 mutex_unlock(&pit->pit_state.lock);
5127 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5129 mutex_lock(&kvm->arch.vpit->pit_state.lock);
5130 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5131 sizeof(ps->channels));
5132 ps->flags = kvm->arch.vpit->pit_state.flags;
5133 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5134 memset(&ps->reserved, 0, sizeof(ps->reserved));
5138 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5142 u32 prev_legacy, cur_legacy;
5143 struct kvm_pit *pit = kvm->arch.vpit;
5145 mutex_lock(&pit->pit_state.lock);
5146 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5147 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5148 if (!prev_legacy && cur_legacy)
5150 memcpy(&pit->pit_state.channels, &ps->channels,
5151 sizeof(pit->pit_state.channels));
5152 pit->pit_state.flags = ps->flags;
5153 for (i = 0; i < 3; i++)
5154 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5156 mutex_unlock(&pit->pit_state.lock);
5160 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5161 struct kvm_reinject_control *control)
5163 struct kvm_pit *pit = kvm->arch.vpit;
5165 /* pit->pit_state.lock was overloaded to prevent userspace from getting
5166 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5167 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
5169 mutex_lock(&pit->pit_state.lock);
5170 kvm_pit_set_reinject(pit, control->pit_reinject);
5171 mutex_unlock(&pit->pit_state.lock);
5176 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5179 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
5181 if (kvm_x86_ops.flush_log_dirty)
5182 kvm_x86_ops.flush_log_dirty(kvm);
5185 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5188 if (!irqchip_in_kernel(kvm))
5191 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5192 irq_event->irq, irq_event->level,
5197 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5198 struct kvm_enable_cap *cap)
5206 case KVM_CAP_DISABLE_QUIRKS:
5207 kvm->arch.disabled_quirks = cap->args[0];
5210 case KVM_CAP_SPLIT_IRQCHIP: {
5211 mutex_lock(&kvm->lock);
5213 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5214 goto split_irqchip_unlock;
5216 if (irqchip_in_kernel(kvm))
5217 goto split_irqchip_unlock;
5218 if (kvm->created_vcpus)
5219 goto split_irqchip_unlock;
5220 r = kvm_setup_empty_irq_routing(kvm);
5222 goto split_irqchip_unlock;
5223 /* Pairs with irqchip_in_kernel. */
5225 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5226 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5228 split_irqchip_unlock:
5229 mutex_unlock(&kvm->lock);
5232 case KVM_CAP_X2APIC_API:
5234 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5237 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5238 kvm->arch.x2apic_format = true;
5239 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5240 kvm->arch.x2apic_broadcast_quirk_disabled = true;
5244 case KVM_CAP_X86_DISABLE_EXITS:
5246 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5249 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5250 kvm_can_mwait_in_guest())
5251 kvm->arch.mwait_in_guest = true;
5252 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5253 kvm->arch.hlt_in_guest = true;
5254 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5255 kvm->arch.pause_in_guest = true;
5256 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5257 kvm->arch.cstate_in_guest = true;
5260 case KVM_CAP_MSR_PLATFORM_INFO:
5261 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5264 case KVM_CAP_EXCEPTION_PAYLOAD:
5265 kvm->arch.exception_payload_enabled = cap->args[0];
5268 case KVM_CAP_X86_USER_SPACE_MSR:
5269 kvm->arch.user_space_msr_mask = cap->args[0];
5279 static void kvm_clear_msr_filter(struct kvm *kvm)
5282 u32 count = kvm->arch.msr_filter.count;
5283 struct msr_bitmap_range ranges[16];
5285 mutex_lock(&kvm->lock);
5286 kvm->arch.msr_filter.count = 0;
5287 memcpy(ranges, kvm->arch.msr_filter.ranges, count * sizeof(ranges[0]));
5288 mutex_unlock(&kvm->lock);
5289 synchronize_srcu(&kvm->srcu);
5291 for (i = 0; i < count; i++)
5292 kfree(ranges[i].bitmap);
5295 static int kvm_add_msr_filter(struct kvm *kvm, struct kvm_msr_filter_range *user_range)
5297 struct msr_bitmap_range *ranges = kvm->arch.msr_filter.ranges;
5298 struct msr_bitmap_range range;
5299 unsigned long *bitmap = NULL;
5303 if (!user_range->nmsrs)
5306 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5307 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5310 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5312 return PTR_ERR(bitmap);
5314 range = (struct msr_bitmap_range) {
5315 .flags = user_range->flags,
5316 .base = user_range->base,
5317 .nmsrs = user_range->nmsrs,
5321 if (range.flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE)) {
5331 /* Everything ok, add this range identifier to our global pool */
5332 ranges[kvm->arch.msr_filter.count] = range;
5333 /* Make sure we filled the array before we tell anyone to walk it */
5335 kvm->arch.msr_filter.count++;
5343 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5345 struct kvm_msr_filter __user *user_msr_filter = argp;
5346 struct kvm_msr_filter filter;
5352 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5355 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5356 empty &= !filter.ranges[i].nmsrs;
5358 default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5359 if (empty && !default_allow)
5362 kvm_clear_msr_filter(kvm);
5364 kvm->arch.msr_filter.default_allow = default_allow;
5367 * Protect from concurrent calls to this function that could trigger
5368 * a TOCTOU violation on kvm->arch.msr_filter.count.
5370 mutex_lock(&kvm->lock);
5371 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5372 r = kvm_add_msr_filter(kvm, &filter.ranges[i]);
5377 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5378 mutex_unlock(&kvm->lock);
5383 long kvm_arch_vm_ioctl(struct file *filp,
5384 unsigned int ioctl, unsigned long arg)
5386 struct kvm *kvm = filp->private_data;
5387 void __user *argp = (void __user *)arg;
5390 * This union makes it completely explicit to gcc-3.x
5391 * that these two variables' stack usage should be
5392 * combined, not added together.
5395 struct kvm_pit_state ps;
5396 struct kvm_pit_state2 ps2;
5397 struct kvm_pit_config pit_config;
5401 case KVM_SET_TSS_ADDR:
5402 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5404 case KVM_SET_IDENTITY_MAP_ADDR: {
5407 mutex_lock(&kvm->lock);
5409 if (kvm->created_vcpus)
5410 goto set_identity_unlock;
5412 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5413 goto set_identity_unlock;
5414 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5415 set_identity_unlock:
5416 mutex_unlock(&kvm->lock);
5419 case KVM_SET_NR_MMU_PAGES:
5420 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5422 case KVM_GET_NR_MMU_PAGES:
5423 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5425 case KVM_CREATE_IRQCHIP: {
5426 mutex_lock(&kvm->lock);
5429 if (irqchip_in_kernel(kvm))
5430 goto create_irqchip_unlock;
5433 if (kvm->created_vcpus)
5434 goto create_irqchip_unlock;
5436 r = kvm_pic_init(kvm);
5438 goto create_irqchip_unlock;
5440 r = kvm_ioapic_init(kvm);
5442 kvm_pic_destroy(kvm);
5443 goto create_irqchip_unlock;
5446 r = kvm_setup_default_irq_routing(kvm);
5448 kvm_ioapic_destroy(kvm);
5449 kvm_pic_destroy(kvm);
5450 goto create_irqchip_unlock;
5452 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5454 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5455 create_irqchip_unlock:
5456 mutex_unlock(&kvm->lock);
5459 case KVM_CREATE_PIT:
5460 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5462 case KVM_CREATE_PIT2:
5464 if (copy_from_user(&u.pit_config, argp,
5465 sizeof(struct kvm_pit_config)))
5468 mutex_lock(&kvm->lock);
5471 goto create_pit_unlock;
5473 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5477 mutex_unlock(&kvm->lock);
5479 case KVM_GET_IRQCHIP: {
5480 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5481 struct kvm_irqchip *chip;
5483 chip = memdup_user(argp, sizeof(*chip));
5490 if (!irqchip_kernel(kvm))
5491 goto get_irqchip_out;
5492 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5494 goto get_irqchip_out;
5496 if (copy_to_user(argp, chip, sizeof(*chip)))
5497 goto get_irqchip_out;
5503 case KVM_SET_IRQCHIP: {
5504 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5505 struct kvm_irqchip *chip;
5507 chip = memdup_user(argp, sizeof(*chip));
5514 if (!irqchip_kernel(kvm))
5515 goto set_irqchip_out;
5516 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5523 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5526 if (!kvm->arch.vpit)
5528 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5532 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5539 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5541 mutex_lock(&kvm->lock);
5543 if (!kvm->arch.vpit)
5545 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5547 mutex_unlock(&kvm->lock);
5550 case KVM_GET_PIT2: {
5552 if (!kvm->arch.vpit)
5554 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5558 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5563 case KVM_SET_PIT2: {
5565 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5567 mutex_lock(&kvm->lock);
5569 if (!kvm->arch.vpit)
5571 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5573 mutex_unlock(&kvm->lock);
5576 case KVM_REINJECT_CONTROL: {
5577 struct kvm_reinject_control control;
5579 if (copy_from_user(&control, argp, sizeof(control)))
5582 if (!kvm->arch.vpit)
5584 r = kvm_vm_ioctl_reinject(kvm, &control);
5587 case KVM_SET_BOOT_CPU_ID:
5589 mutex_lock(&kvm->lock);
5590 if (kvm->created_vcpus)
5593 kvm->arch.bsp_vcpu_id = arg;
5594 mutex_unlock(&kvm->lock);
5596 case KVM_XEN_HVM_CONFIG: {
5597 struct kvm_xen_hvm_config xhc;
5599 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5604 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5608 case KVM_SET_CLOCK: {
5609 struct kvm_clock_data user_ns;
5613 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5622 * TODO: userspace has to take care of races with VCPU_RUN, so
5623 * kvm_gen_update_masterclock() can be cut down to locked
5624 * pvclock_update_vm_gtod_copy().
5626 kvm_gen_update_masterclock(kvm);
5627 now_ns = get_kvmclock_ns(kvm);
5628 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5629 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5632 case KVM_GET_CLOCK: {
5633 struct kvm_clock_data user_ns;
5636 now_ns = get_kvmclock_ns(kvm);
5637 user_ns.clock = now_ns;
5638 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5639 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5642 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5647 case KVM_MEMORY_ENCRYPT_OP: {
5649 if (kvm_x86_ops.mem_enc_op)
5650 r = kvm_x86_ops.mem_enc_op(kvm, argp);
5653 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5654 struct kvm_enc_region region;
5657 if (copy_from_user(®ion, argp, sizeof(region)))
5661 if (kvm_x86_ops.mem_enc_reg_region)
5662 r = kvm_x86_ops.mem_enc_reg_region(kvm, ®ion);
5665 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5666 struct kvm_enc_region region;
5669 if (copy_from_user(®ion, argp, sizeof(region)))
5673 if (kvm_x86_ops.mem_enc_unreg_region)
5674 r = kvm_x86_ops.mem_enc_unreg_region(kvm, ®ion);
5677 case KVM_HYPERV_EVENTFD: {
5678 struct kvm_hyperv_eventfd hvevfd;
5681 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5683 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5686 case KVM_SET_PMU_EVENT_FILTER:
5687 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5689 case KVM_X86_SET_MSR_FILTER:
5690 r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
5699 static void kvm_init_msr_list(void)
5701 struct x86_pmu_capability x86_pmu;
5705 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5706 "Please update the fixed PMCs in msrs_to_saved_all[]");
5708 perf_get_x86_pmu_capability(&x86_pmu);
5710 num_msrs_to_save = 0;
5711 num_emulated_msrs = 0;
5712 num_msr_based_features = 0;
5714 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5715 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5719 * Even MSRs that are valid in the host may not be exposed
5720 * to the guests in some cases.
5722 switch (msrs_to_save_all[i]) {
5723 case MSR_IA32_BNDCFGS:
5724 if (!kvm_mpx_supported())
5728 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
5731 case MSR_IA32_UMWAIT_CONTROL:
5732 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
5735 case MSR_IA32_RTIT_CTL:
5736 case MSR_IA32_RTIT_STATUS:
5737 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
5740 case MSR_IA32_RTIT_CR3_MATCH:
5741 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5742 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5745 case MSR_IA32_RTIT_OUTPUT_BASE:
5746 case MSR_IA32_RTIT_OUTPUT_MASK:
5747 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5748 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5749 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5752 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
5753 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5754 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5755 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5758 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5759 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5760 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5763 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5764 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5765 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5772 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5775 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5776 if (!kvm_x86_ops.has_emulated_msr(emulated_msrs_all[i]))
5779 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5782 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5783 struct kvm_msr_entry msr;
5785 msr.index = msr_based_features_all[i];
5786 if (kvm_get_msr_feature(&msr))
5789 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
5793 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5801 if (!(lapic_in_kernel(vcpu) &&
5802 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5803 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5814 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5821 if (!(lapic_in_kernel(vcpu) &&
5822 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5824 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5826 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5836 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5837 struct kvm_segment *var, int seg)
5839 kvm_x86_ops.set_segment(vcpu, var, seg);
5842 void kvm_get_segment(struct kvm_vcpu *vcpu,
5843 struct kvm_segment *var, int seg)
5845 kvm_x86_ops.get_segment(vcpu, var, seg);
5848 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5849 struct x86_exception *exception)
5853 BUG_ON(!mmu_is_nested(vcpu));
5855 /* NPT walks are always user-walks */
5856 access |= PFERR_USER_MASK;
5857 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5862 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5863 struct x86_exception *exception)
5865 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5866 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5869 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5870 struct x86_exception *exception)
5872 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5873 access |= PFERR_FETCH_MASK;
5874 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5877 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5878 struct x86_exception *exception)
5880 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5881 access |= PFERR_WRITE_MASK;
5882 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5885 /* uses this to access any guest's mapped memory without checking CPL */
5886 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5887 struct x86_exception *exception)
5889 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5892 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5893 struct kvm_vcpu *vcpu, u32 access,
5894 struct x86_exception *exception)
5897 int r = X86EMUL_CONTINUE;
5900 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5902 unsigned offset = addr & (PAGE_SIZE-1);
5903 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5906 if (gpa == UNMAPPED_GVA)
5907 return X86EMUL_PROPAGATE_FAULT;
5908 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5911 r = X86EMUL_IO_NEEDED;
5923 /* used for instruction fetching */
5924 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5925 gva_t addr, void *val, unsigned int bytes,
5926 struct x86_exception *exception)
5928 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5929 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5933 /* Inline kvm_read_guest_virt_helper for speed. */
5934 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5936 if (unlikely(gpa == UNMAPPED_GVA))
5937 return X86EMUL_PROPAGATE_FAULT;
5939 offset = addr & (PAGE_SIZE-1);
5940 if (WARN_ON(offset + bytes > PAGE_SIZE))
5941 bytes = (unsigned)PAGE_SIZE - offset;
5942 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5944 if (unlikely(ret < 0))
5945 return X86EMUL_IO_NEEDED;
5947 return X86EMUL_CONTINUE;
5950 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5951 gva_t addr, void *val, unsigned int bytes,
5952 struct x86_exception *exception)
5954 u32 access = (kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5957 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5958 * is returned, but our callers are not ready for that and they blindly
5959 * call kvm_inject_page_fault. Ensure that they at least do not leak
5960 * uninitialized kernel stack memory into cr2 and error code.
5962 memset(exception, 0, sizeof(*exception));
5963 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5966 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5968 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5969 gva_t addr, void *val, unsigned int bytes,
5970 struct x86_exception *exception, bool system)
5972 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5975 if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
5976 access |= PFERR_USER_MASK;
5978 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5981 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5982 unsigned long addr, void *val, unsigned int bytes)
5984 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5985 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5987 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5990 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5991 struct kvm_vcpu *vcpu, u32 access,
5992 struct x86_exception *exception)
5995 int r = X86EMUL_CONTINUE;
5998 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6001 unsigned offset = addr & (PAGE_SIZE-1);
6002 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6005 if (gpa == UNMAPPED_GVA)
6006 return X86EMUL_PROPAGATE_FAULT;
6007 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6009 r = X86EMUL_IO_NEEDED;
6021 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6022 unsigned int bytes, struct x86_exception *exception,
6025 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6026 u32 access = PFERR_WRITE_MASK;
6028 if (!system && kvm_x86_ops.get_cpl(vcpu) == 3)
6029 access |= PFERR_USER_MASK;
6031 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6035 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6036 unsigned int bytes, struct x86_exception *exception)
6038 /* kvm_write_guest_virt_system can pull in tons of pages. */
6039 vcpu->arch.l1tf_flush_l1d = true;
6041 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6042 PFERR_WRITE_MASK, exception);
6044 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6046 int handle_ud(struct kvm_vcpu *vcpu)
6048 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6049 int emul_type = EMULTYPE_TRAP_UD;
6050 char sig[5]; /* ud2; .ascii "kvm" */
6051 struct x86_exception e;
6053 if (unlikely(!kvm_x86_ops.can_emulate_instruction(vcpu, NULL, 0)))
6056 if (force_emulation_prefix &&
6057 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6058 sig, sizeof(sig), &e) == 0 &&
6059 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6060 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6061 emul_type = EMULTYPE_TRAP_UD_FORCED;
6064 return kvm_emulate_instruction(vcpu, emul_type);
6066 EXPORT_SYMBOL_GPL(handle_ud);
6068 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6069 gpa_t gpa, bool write)
6071 /* For APIC access vmexit */
6072 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6075 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6076 trace_vcpu_match_mmio(gva, gpa, write, true);
6083 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6084 gpa_t *gpa, struct x86_exception *exception,
6087 u32 access = ((kvm_x86_ops.get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
6088 | (write ? PFERR_WRITE_MASK : 0);
6091 * currently PKRU is only applied to ept enabled guest so
6092 * there is no pkey in EPT page table for L1 guest or EPT
6093 * shadow page table for L2 guest.
6095 if (vcpu_match_mmio_gva(vcpu, gva)
6096 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
6097 vcpu->arch.mmio_access, 0, access)) {
6098 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6099 (gva & (PAGE_SIZE - 1));
6100 trace_vcpu_match_mmio(gva, *gpa, write, false);
6104 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6106 if (*gpa == UNMAPPED_GVA)
6109 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6112 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6113 const void *val, int bytes)
6117 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6120 kvm_page_track_write(vcpu, gpa, val, bytes);
6124 struct read_write_emulator_ops {
6125 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6127 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6128 void *val, int bytes);
6129 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6130 int bytes, void *val);
6131 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6132 void *val, int bytes);
6136 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6138 if (vcpu->mmio_read_completed) {
6139 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6140 vcpu->mmio_fragments[0].gpa, val);
6141 vcpu->mmio_read_completed = 0;
6148 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6149 void *val, int bytes)
6151 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6154 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6155 void *val, int bytes)
6157 return emulator_write_phys(vcpu, gpa, val, bytes);
6160 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6162 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6163 return vcpu_mmio_write(vcpu, gpa, bytes, val);
6166 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6167 void *val, int bytes)
6169 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6170 return X86EMUL_IO_NEEDED;
6173 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6174 void *val, int bytes)
6176 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6178 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6179 return X86EMUL_CONTINUE;
6182 static const struct read_write_emulator_ops read_emultor = {
6183 .read_write_prepare = read_prepare,
6184 .read_write_emulate = read_emulate,
6185 .read_write_mmio = vcpu_mmio_read,
6186 .read_write_exit_mmio = read_exit_mmio,
6189 static const struct read_write_emulator_ops write_emultor = {
6190 .read_write_emulate = write_emulate,
6191 .read_write_mmio = write_mmio,
6192 .read_write_exit_mmio = write_exit_mmio,
6196 static int emulator_read_write_onepage(unsigned long addr, void *val,
6198 struct x86_exception *exception,
6199 struct kvm_vcpu *vcpu,
6200 const struct read_write_emulator_ops *ops)
6204 bool write = ops->write;
6205 struct kvm_mmio_fragment *frag;
6206 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6209 * If the exit was due to a NPF we may already have a GPA.
6210 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6211 * Note, this cannot be used on string operations since string
6212 * operation using rep will only have the initial GPA from the NPF
6215 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6216 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6217 gpa = ctxt->gpa_val;
6218 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6220 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6222 return X86EMUL_PROPAGATE_FAULT;
6225 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6226 return X86EMUL_CONTINUE;
6229 * Is this MMIO handled locally?
6231 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6232 if (handled == bytes)
6233 return X86EMUL_CONTINUE;
6239 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6240 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6244 return X86EMUL_CONTINUE;
6247 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6249 void *val, unsigned int bytes,
6250 struct x86_exception *exception,
6251 const struct read_write_emulator_ops *ops)
6253 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6257 if (ops->read_write_prepare &&
6258 ops->read_write_prepare(vcpu, val, bytes))
6259 return X86EMUL_CONTINUE;
6261 vcpu->mmio_nr_fragments = 0;
6263 /* Crossing a page boundary? */
6264 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6267 now = -addr & ~PAGE_MASK;
6268 rc = emulator_read_write_onepage(addr, val, now, exception,
6271 if (rc != X86EMUL_CONTINUE)
6274 if (ctxt->mode != X86EMUL_MODE_PROT64)
6280 rc = emulator_read_write_onepage(addr, val, bytes, exception,
6282 if (rc != X86EMUL_CONTINUE)
6285 if (!vcpu->mmio_nr_fragments)
6288 gpa = vcpu->mmio_fragments[0].gpa;
6290 vcpu->mmio_needed = 1;
6291 vcpu->mmio_cur_fragment = 0;
6293 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6294 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6295 vcpu->run->exit_reason = KVM_EXIT_MMIO;
6296 vcpu->run->mmio.phys_addr = gpa;
6298 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6301 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6305 struct x86_exception *exception)
6307 return emulator_read_write(ctxt, addr, val, bytes,
6308 exception, &read_emultor);
6311 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6315 struct x86_exception *exception)
6317 return emulator_read_write(ctxt, addr, (void *)val, bytes,
6318 exception, &write_emultor);
6321 #define CMPXCHG_TYPE(t, ptr, old, new) \
6322 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6324 #ifdef CONFIG_X86_64
6325 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6327 # define CMPXCHG64(ptr, old, new) \
6328 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6331 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6336 struct x86_exception *exception)
6338 struct kvm_host_map map;
6339 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6345 /* guests cmpxchg8b have to be emulated atomically */
6346 if (bytes > 8 || (bytes & (bytes - 1)))
6349 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6351 if (gpa == UNMAPPED_GVA ||
6352 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6356 * Emulate the atomic as a straight write to avoid #AC if SLD is
6357 * enabled in the host and the access splits a cache line.
6359 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6360 page_line_mask = ~(cache_line_size() - 1);
6362 page_line_mask = PAGE_MASK;
6364 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6367 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6370 kaddr = map.hva + offset_in_page(gpa);
6374 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6377 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6380 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6383 exchanged = CMPXCHG64(kaddr, old, new);
6389 kvm_vcpu_unmap(vcpu, &map, true);
6392 return X86EMUL_CMPXCHG_FAILED;
6394 kvm_page_track_write(vcpu, gpa, new, bytes);
6396 return X86EMUL_CONTINUE;
6399 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6401 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6404 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6408 for (i = 0; i < vcpu->arch.pio.count; i++) {
6409 if (vcpu->arch.pio.in)
6410 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6411 vcpu->arch.pio.size, pd);
6413 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6414 vcpu->arch.pio.port, vcpu->arch.pio.size,
6418 pd += vcpu->arch.pio.size;
6423 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6424 unsigned short port, void *val,
6425 unsigned int count, bool in)
6427 vcpu->arch.pio.port = port;
6428 vcpu->arch.pio.in = in;
6429 vcpu->arch.pio.count = count;
6430 vcpu->arch.pio.size = size;
6432 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6433 vcpu->arch.pio.count = 0;
6437 vcpu->run->exit_reason = KVM_EXIT_IO;
6438 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6439 vcpu->run->io.size = size;
6440 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6441 vcpu->run->io.count = count;
6442 vcpu->run->io.port = port;
6447 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6448 unsigned short port, void *val, unsigned int count)
6452 if (vcpu->arch.pio.count)
6455 memset(vcpu->arch.pio_data, 0, size * count);
6457 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6460 memcpy(val, vcpu->arch.pio_data, size * count);
6461 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6462 vcpu->arch.pio.count = 0;
6469 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6470 int size, unsigned short port, void *val,
6473 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6477 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6478 unsigned short port, const void *val,
6481 memcpy(vcpu->arch.pio_data, val, size * count);
6482 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6483 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6486 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6487 int size, unsigned short port,
6488 const void *val, unsigned int count)
6490 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6493 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6495 return kvm_x86_ops.get_segment_base(vcpu, seg);
6498 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6500 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6503 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6505 if (!need_emulate_wbinvd(vcpu))
6506 return X86EMUL_CONTINUE;
6508 if (kvm_x86_ops.has_wbinvd_exit()) {
6509 int cpu = get_cpu();
6511 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6512 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
6513 wbinvd_ipi, NULL, 1);
6515 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6518 return X86EMUL_CONTINUE;
6521 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6523 kvm_emulate_wbinvd_noskip(vcpu);
6524 return kvm_skip_emulated_instruction(vcpu);
6526 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6530 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6532 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6535 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6536 unsigned long *dest)
6538 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6541 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6542 unsigned long value)
6545 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6548 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6550 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6553 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6555 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6556 unsigned long value;
6560 value = kvm_read_cr0(vcpu);
6563 value = vcpu->arch.cr2;
6566 value = kvm_read_cr3(vcpu);
6569 value = kvm_read_cr4(vcpu);
6572 value = kvm_get_cr8(vcpu);
6575 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6582 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6584 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6589 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6592 vcpu->arch.cr2 = val;
6595 res = kvm_set_cr3(vcpu, val);
6598 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6601 res = kvm_set_cr8(vcpu, val);
6604 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6611 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6613 return kvm_x86_ops.get_cpl(emul_to_vcpu(ctxt));
6616 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6618 kvm_x86_ops.get_gdt(emul_to_vcpu(ctxt), dt);
6621 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6623 kvm_x86_ops.get_idt(emul_to_vcpu(ctxt), dt);
6626 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6628 kvm_x86_ops.set_gdt(emul_to_vcpu(ctxt), dt);
6631 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6633 kvm_x86_ops.set_idt(emul_to_vcpu(ctxt), dt);
6636 static unsigned long emulator_get_cached_segment_base(
6637 struct x86_emulate_ctxt *ctxt, int seg)
6639 return get_segment_base(emul_to_vcpu(ctxt), seg);
6642 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6643 struct desc_struct *desc, u32 *base3,
6646 struct kvm_segment var;
6648 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6649 *selector = var.selector;
6652 memset(desc, 0, sizeof(*desc));
6660 set_desc_limit(desc, var.limit);
6661 set_desc_base(desc, (unsigned long)var.base);
6662 #ifdef CONFIG_X86_64
6664 *base3 = var.base >> 32;
6666 desc->type = var.type;
6668 desc->dpl = var.dpl;
6669 desc->p = var.present;
6670 desc->avl = var.avl;
6678 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6679 struct desc_struct *desc, u32 base3,
6682 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6683 struct kvm_segment var;
6685 var.selector = selector;
6686 var.base = get_desc_base(desc);
6687 #ifdef CONFIG_X86_64
6688 var.base |= ((u64)base3) << 32;
6690 var.limit = get_desc_limit(desc);
6692 var.limit = (var.limit << 12) | 0xfff;
6693 var.type = desc->type;
6694 var.dpl = desc->dpl;
6699 var.avl = desc->avl;
6700 var.present = desc->p;
6701 var.unusable = !var.present;
6704 kvm_set_segment(vcpu, &var, seg);
6708 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6709 u32 msr_index, u64 *pdata)
6711 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6714 r = kvm_get_msr(vcpu, msr_index, pdata);
6716 if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
6717 /* Bounce to user space */
6718 return X86EMUL_IO_NEEDED;
6724 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6725 u32 msr_index, u64 data)
6727 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6730 r = kvm_set_msr(vcpu, msr_index, data);
6732 if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
6733 /* Bounce to user space */
6734 return X86EMUL_IO_NEEDED;
6740 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6742 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6744 return vcpu->arch.smbase;
6747 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6749 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6751 vcpu->arch.smbase = smbase;
6754 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6757 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6760 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6761 u32 pmc, u64 *pdata)
6763 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6766 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6768 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6771 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6772 struct x86_instruction_info *info,
6773 enum x86_intercept_stage stage)
6775 return kvm_x86_ops.check_intercept(emul_to_vcpu(ctxt), info, stage,
6779 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6780 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
6783 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
6786 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
6788 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
6791 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
6793 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
6796 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
6798 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
6801 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6803 return kvm_register_read(emul_to_vcpu(ctxt), reg);
6806 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6808 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6811 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6813 kvm_x86_ops.set_nmi_mask(emul_to_vcpu(ctxt), masked);
6816 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6818 return emul_to_vcpu(ctxt)->arch.hflags;
6821 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6823 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6826 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6827 const char *smstate)
6829 return kvm_x86_ops.pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6832 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6834 kvm_smm_changed(emul_to_vcpu(ctxt));
6837 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6839 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6842 static const struct x86_emulate_ops emulate_ops = {
6843 .read_gpr = emulator_read_gpr,
6844 .write_gpr = emulator_write_gpr,
6845 .read_std = emulator_read_std,
6846 .write_std = emulator_write_std,
6847 .read_phys = kvm_read_guest_phys_system,
6848 .fetch = kvm_fetch_guest_virt,
6849 .read_emulated = emulator_read_emulated,
6850 .write_emulated = emulator_write_emulated,
6851 .cmpxchg_emulated = emulator_cmpxchg_emulated,
6852 .invlpg = emulator_invlpg,
6853 .pio_in_emulated = emulator_pio_in_emulated,
6854 .pio_out_emulated = emulator_pio_out_emulated,
6855 .get_segment = emulator_get_segment,
6856 .set_segment = emulator_set_segment,
6857 .get_cached_segment_base = emulator_get_cached_segment_base,
6858 .get_gdt = emulator_get_gdt,
6859 .get_idt = emulator_get_idt,
6860 .set_gdt = emulator_set_gdt,
6861 .set_idt = emulator_set_idt,
6862 .get_cr = emulator_get_cr,
6863 .set_cr = emulator_set_cr,
6864 .cpl = emulator_get_cpl,
6865 .get_dr = emulator_get_dr,
6866 .set_dr = emulator_set_dr,
6867 .get_smbase = emulator_get_smbase,
6868 .set_smbase = emulator_set_smbase,
6869 .set_msr = emulator_set_msr,
6870 .get_msr = emulator_get_msr,
6871 .check_pmc = emulator_check_pmc,
6872 .read_pmc = emulator_read_pmc,
6873 .halt = emulator_halt,
6874 .wbinvd = emulator_wbinvd,
6875 .fix_hypercall = emulator_fix_hypercall,
6876 .intercept = emulator_intercept,
6877 .get_cpuid = emulator_get_cpuid,
6878 .guest_has_long_mode = emulator_guest_has_long_mode,
6879 .guest_has_movbe = emulator_guest_has_movbe,
6880 .guest_has_fxsr = emulator_guest_has_fxsr,
6881 .set_nmi_mask = emulator_set_nmi_mask,
6882 .get_hflags = emulator_get_hflags,
6883 .set_hflags = emulator_set_hflags,
6884 .pre_leave_smm = emulator_pre_leave_smm,
6885 .post_leave_smm = emulator_post_leave_smm,
6886 .set_xcr = emulator_set_xcr,
6889 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6891 u32 int_shadow = kvm_x86_ops.get_interrupt_shadow(vcpu);
6893 * an sti; sti; sequence only disable interrupts for the first
6894 * instruction. So, if the last instruction, be it emulated or
6895 * not, left the system with the INT_STI flag enabled, it
6896 * means that the last instruction is an sti. We should not
6897 * leave the flag on in this case. The same goes for mov ss
6899 if (int_shadow & mask)
6901 if (unlikely(int_shadow || mask)) {
6902 kvm_x86_ops.set_interrupt_shadow(vcpu, mask);
6904 kvm_make_request(KVM_REQ_EVENT, vcpu);
6908 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6910 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6911 if (ctxt->exception.vector == PF_VECTOR)
6912 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
6914 if (ctxt->exception.error_code_valid)
6915 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6916 ctxt->exception.error_code);
6918 kvm_queue_exception(vcpu, ctxt->exception.vector);
6922 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
6924 struct x86_emulate_ctxt *ctxt;
6926 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
6928 pr_err("kvm: failed to allocate vcpu's emulator\n");
6933 ctxt->ops = &emulate_ops;
6934 vcpu->arch.emulate_ctxt = ctxt;
6939 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6941 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6944 kvm_x86_ops.get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6946 ctxt->gpa_available = false;
6947 ctxt->eflags = kvm_get_rflags(vcpu);
6948 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6950 ctxt->eip = kvm_rip_read(vcpu);
6951 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
6952 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
6953 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
6954 cs_db ? X86EMUL_MODE_PROT32 :
6955 X86EMUL_MODE_PROT16;
6956 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6957 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6958 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6960 init_decode_cache(ctxt);
6961 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6964 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6966 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6969 init_emulate_ctxt(vcpu);
6973 ctxt->_eip = ctxt->eip + inc_eip;
6974 ret = emulate_int_real(ctxt, irq);
6976 if (ret != X86EMUL_CONTINUE) {
6977 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6979 ctxt->eip = ctxt->_eip;
6980 kvm_rip_write(vcpu, ctxt->eip);
6981 kvm_set_rflags(vcpu, ctxt->eflags);
6984 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6986 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6988 ++vcpu->stat.insn_emulation_fail;
6989 trace_kvm_emulate_insn_failed(vcpu);
6991 if (emulation_type & EMULTYPE_VMWARE_GP) {
6992 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6996 if (emulation_type & EMULTYPE_SKIP) {
6997 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6998 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6999 vcpu->run->internal.ndata = 0;
7003 kvm_queue_exception(vcpu, UD_VECTOR);
7005 if (!is_guest_mode(vcpu) && kvm_x86_ops.get_cpl(vcpu) == 0) {
7006 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7007 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7008 vcpu->run->internal.ndata = 0;
7015 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7016 bool write_fault_to_shadow_pgtable,
7019 gpa_t gpa = cr2_or_gpa;
7022 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7025 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7026 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7029 if (!vcpu->arch.mmu->direct_map) {
7031 * Write permission should be allowed since only
7032 * write access need to be emulated.
7034 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7037 * If the mapping is invalid in guest, let cpu retry
7038 * it to generate fault.
7040 if (gpa == UNMAPPED_GVA)
7045 * Do not retry the unhandleable instruction if it faults on the
7046 * readonly host memory, otherwise it will goto a infinite loop:
7047 * retry instruction -> write #PF -> emulation fail -> retry
7048 * instruction -> ...
7050 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7053 * If the instruction failed on the error pfn, it can not be fixed,
7054 * report the error to userspace.
7056 if (is_error_noslot_pfn(pfn))
7059 kvm_release_pfn_clean(pfn);
7061 /* The instructions are well-emulated on direct mmu. */
7062 if (vcpu->arch.mmu->direct_map) {
7063 unsigned int indirect_shadow_pages;
7065 spin_lock(&vcpu->kvm->mmu_lock);
7066 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7067 spin_unlock(&vcpu->kvm->mmu_lock);
7069 if (indirect_shadow_pages)
7070 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7076 * if emulation was due to access to shadowed page table
7077 * and it failed try to unshadow page and re-enter the
7078 * guest to let CPU execute the instruction.
7080 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7083 * If the access faults on its page table, it can not
7084 * be fixed by unprotecting shadow page and it should
7085 * be reported to userspace.
7087 return !write_fault_to_shadow_pgtable;
7090 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7091 gpa_t cr2_or_gpa, int emulation_type)
7093 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7094 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7096 last_retry_eip = vcpu->arch.last_retry_eip;
7097 last_retry_addr = vcpu->arch.last_retry_addr;
7100 * If the emulation is caused by #PF and it is non-page_table
7101 * writing instruction, it means the VM-EXIT is caused by shadow
7102 * page protected, we can zap the shadow page and retry this
7103 * instruction directly.
7105 * Note: if the guest uses a non-page-table modifying instruction
7106 * on the PDE that points to the instruction, then we will unmap
7107 * the instruction and go to an infinite loop. So, we cache the
7108 * last retried eip and the last fault address, if we meet the eip
7109 * and the address again, we can break out of the potential infinite
7112 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7114 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7117 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7118 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7121 if (x86_page_table_writing_insn(ctxt))
7124 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7127 vcpu->arch.last_retry_eip = ctxt->eip;
7128 vcpu->arch.last_retry_addr = cr2_or_gpa;
7130 if (!vcpu->arch.mmu->direct_map)
7131 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7133 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7138 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7139 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7141 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
7143 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
7144 /* This is a good place to trace that we are exiting SMM. */
7145 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
7147 /* Process a latched INIT or SMI, if any. */
7148 kvm_make_request(KVM_REQ_EVENT, vcpu);
7151 kvm_mmu_reset_context(vcpu);
7154 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7163 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7164 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7169 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7171 struct kvm_run *kvm_run = vcpu->run;
7173 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7174 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
7175 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7176 kvm_run->debug.arch.exception = DB_VECTOR;
7177 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7180 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7184 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7186 unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
7189 r = kvm_x86_ops.skip_emulated_instruction(vcpu);
7194 * rflags is the old, "raw" value of the flags. The new value has
7195 * not been saved yet.
7197 * This is correct even for TF set by the guest, because "the
7198 * processor will not generate this exception after the instruction
7199 * that sets the TF flag".
7201 if (unlikely(rflags & X86_EFLAGS_TF))
7202 r = kvm_vcpu_do_singlestep(vcpu);
7205 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7207 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7209 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7210 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7211 struct kvm_run *kvm_run = vcpu->run;
7212 unsigned long eip = kvm_get_linear_rip(vcpu);
7213 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7214 vcpu->arch.guest_debug_dr7,
7218 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
7219 kvm_run->debug.arch.pc = eip;
7220 kvm_run->debug.arch.exception = DB_VECTOR;
7221 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7227 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7228 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7229 unsigned long eip = kvm_get_linear_rip(vcpu);
7230 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7235 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7244 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7246 switch (ctxt->opcode_len) {
7253 case 0xe6: /* OUT */
7257 case 0x6c: /* INS */
7259 case 0x6e: /* OUTS */
7266 case 0x33: /* RDPMC */
7275 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7276 int emulation_type, void *insn, int insn_len)
7279 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7280 bool writeback = true;
7281 bool write_fault_to_spt;
7283 if (unlikely(!kvm_x86_ops.can_emulate_instruction(vcpu, insn, insn_len)))
7286 vcpu->arch.l1tf_flush_l1d = true;
7289 * Clear write_fault_to_shadow_pgtable here to ensure it is
7292 write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7293 vcpu->arch.write_fault_to_shadow_pgtable = false;
7294 kvm_clear_exception_queue(vcpu);
7296 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7297 init_emulate_ctxt(vcpu);
7300 * We will reenter on the same instruction since
7301 * we do not set complete_userspace_io. This does not
7302 * handle watchpoints yet, those would be handled in
7305 if (!(emulation_type & EMULTYPE_SKIP) &&
7306 kvm_vcpu_check_breakpoint(vcpu, &r))
7309 ctxt->interruptibility = 0;
7310 ctxt->have_exception = false;
7311 ctxt->exception.vector = -1;
7312 ctxt->perm_ok = false;
7314 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
7316 r = x86_decode_insn(ctxt, insn, insn_len);
7318 trace_kvm_emulate_insn_start(vcpu);
7319 ++vcpu->stat.insn_emulation;
7320 if (r != EMULATION_OK) {
7321 if ((emulation_type & EMULTYPE_TRAP_UD) ||
7322 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7323 kvm_queue_exception(vcpu, UD_VECTOR);
7326 if (reexecute_instruction(vcpu, cr2_or_gpa,
7330 if (ctxt->have_exception) {
7332 * #UD should result in just EMULATION_FAILED, and trap-like
7333 * exception should not be encountered during decode.
7335 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7336 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7337 inject_emulated_exception(vcpu);
7340 return handle_emulation_failure(vcpu, emulation_type);
7344 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7345 !is_vmware_backdoor_opcode(ctxt)) {
7346 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7351 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7352 * for kvm_skip_emulated_instruction(). The caller is responsible for
7353 * updating interruptibility state and injecting single-step #DBs.
7355 if (emulation_type & EMULTYPE_SKIP) {
7356 kvm_rip_write(vcpu, ctxt->_eip);
7357 if (ctxt->eflags & X86_EFLAGS_RF)
7358 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7362 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7365 /* this is needed for vmware backdoor interface to work since it
7366 changes registers values during IO operation */
7367 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7368 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7369 emulator_invalidate_register_cache(ctxt);
7373 if (emulation_type & EMULTYPE_PF) {
7374 /* Save the faulting GPA (cr2) in the address field */
7375 ctxt->exception.address = cr2_or_gpa;
7377 /* With shadow page tables, cr2 contains a GVA or nGPA. */
7378 if (vcpu->arch.mmu->direct_map) {
7379 ctxt->gpa_available = true;
7380 ctxt->gpa_val = cr2_or_gpa;
7383 /* Sanitize the address out of an abundance of paranoia. */
7384 ctxt->exception.address = 0;
7387 r = x86_emulate_insn(ctxt);
7389 if (r == EMULATION_INTERCEPTED)
7392 if (r == EMULATION_FAILED) {
7393 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7397 return handle_emulation_failure(vcpu, emulation_type);
7400 if (ctxt->have_exception) {
7402 if (inject_emulated_exception(vcpu))
7404 } else if (vcpu->arch.pio.count) {
7405 if (!vcpu->arch.pio.in) {
7406 /* FIXME: return into emulator if single-stepping. */
7407 vcpu->arch.pio.count = 0;
7410 vcpu->arch.complete_userspace_io = complete_emulated_pio;
7413 } else if (vcpu->mmio_needed) {
7414 ++vcpu->stat.mmio_exits;
7416 if (!vcpu->mmio_is_write)
7419 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7420 } else if (r == EMULATION_RESTART)
7426 unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
7427 toggle_interruptibility(vcpu, ctxt->interruptibility);
7428 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7429 if (!ctxt->have_exception ||
7430 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7431 kvm_rip_write(vcpu, ctxt->eip);
7432 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7433 r = kvm_vcpu_do_singlestep(vcpu);
7434 if (kvm_x86_ops.update_emulated_instruction)
7435 kvm_x86_ops.update_emulated_instruction(vcpu);
7436 __kvm_set_rflags(vcpu, ctxt->eflags);
7440 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7441 * do nothing, and it will be requested again as soon as
7442 * the shadow expires. But we still need to check here,
7443 * because POPF has no interrupt shadow.
7445 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7446 kvm_make_request(KVM_REQ_EVENT, vcpu);
7448 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7453 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7455 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7457 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7459 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7460 void *insn, int insn_len)
7462 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7464 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7466 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7468 vcpu->arch.pio.count = 0;
7472 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7474 vcpu->arch.pio.count = 0;
7476 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7479 return kvm_skip_emulated_instruction(vcpu);
7482 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7483 unsigned short port)
7485 unsigned long val = kvm_rax_read(vcpu);
7486 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7492 * Workaround userspace that relies on old KVM behavior of %rip being
7493 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7496 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
7497 vcpu->arch.complete_userspace_io =
7498 complete_fast_pio_out_port_0x7e;
7499 kvm_skip_emulated_instruction(vcpu);
7501 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7502 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
7507 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7511 /* We should only ever be called with arch.pio.count equal to 1 */
7512 BUG_ON(vcpu->arch.pio.count != 1);
7514 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7515 vcpu->arch.pio.count = 0;
7519 /* For size less than 4 we merge, else we zero extend */
7520 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7523 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7524 * the copy and tracing
7526 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7527 kvm_rax_write(vcpu, val);
7529 return kvm_skip_emulated_instruction(vcpu);
7532 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7533 unsigned short port)
7538 /* For size less than 4 we merge, else we zero extend */
7539 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7541 ret = emulator_pio_in(vcpu, size, port, &val, 1);
7543 kvm_rax_write(vcpu, val);
7547 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7548 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7553 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7558 ret = kvm_fast_pio_in(vcpu, size, port);
7560 ret = kvm_fast_pio_out(vcpu, size, port);
7561 return ret && kvm_skip_emulated_instruction(vcpu);
7563 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7565 static int kvmclock_cpu_down_prep(unsigned int cpu)
7567 __this_cpu_write(cpu_tsc_khz, 0);
7571 static void tsc_khz_changed(void *data)
7573 struct cpufreq_freqs *freq = data;
7574 unsigned long khz = 0;
7578 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7579 khz = cpufreq_quick_get(raw_smp_processor_id());
7582 __this_cpu_write(cpu_tsc_khz, khz);
7585 #ifdef CONFIG_X86_64
7586 static void kvm_hyperv_tsc_notifier(void)
7589 struct kvm_vcpu *vcpu;
7592 mutex_lock(&kvm_lock);
7593 list_for_each_entry(kvm, &vm_list, vm_list)
7594 kvm_make_mclock_inprogress_request(kvm);
7596 hyperv_stop_tsc_emulation();
7598 /* TSC frequency always matches when on Hyper-V */
7599 for_each_present_cpu(cpu)
7600 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7601 kvm_max_guest_tsc_khz = tsc_khz;
7603 list_for_each_entry(kvm, &vm_list, vm_list) {
7604 struct kvm_arch *ka = &kvm->arch;
7606 spin_lock(&ka->pvclock_gtod_sync_lock);
7608 pvclock_update_vm_gtod_copy(kvm);
7610 kvm_for_each_vcpu(cpu, vcpu, kvm)
7611 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7613 kvm_for_each_vcpu(cpu, vcpu, kvm)
7614 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7616 spin_unlock(&ka->pvclock_gtod_sync_lock);
7618 mutex_unlock(&kvm_lock);
7622 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7625 struct kvm_vcpu *vcpu;
7626 int i, send_ipi = 0;
7629 * We allow guests to temporarily run on slowing clocks,
7630 * provided we notify them after, or to run on accelerating
7631 * clocks, provided we notify them before. Thus time never
7634 * However, we have a problem. We can't atomically update
7635 * the frequency of a given CPU from this function; it is
7636 * merely a notifier, which can be called from any CPU.
7637 * Changing the TSC frequency at arbitrary points in time
7638 * requires a recomputation of local variables related to
7639 * the TSC for each VCPU. We must flag these local variables
7640 * to be updated and be sure the update takes place with the
7641 * new frequency before any guests proceed.
7643 * Unfortunately, the combination of hotplug CPU and frequency
7644 * change creates an intractable locking scenario; the order
7645 * of when these callouts happen is undefined with respect to
7646 * CPU hotplug, and they can race with each other. As such,
7647 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7648 * undefined; you can actually have a CPU frequency change take
7649 * place in between the computation of X and the setting of the
7650 * variable. To protect against this problem, all updates of
7651 * the per_cpu tsc_khz variable are done in an interrupt
7652 * protected IPI, and all callers wishing to update the value
7653 * must wait for a synchronous IPI to complete (which is trivial
7654 * if the caller is on the CPU already). This establishes the
7655 * necessary total order on variable updates.
7657 * Note that because a guest time update may take place
7658 * anytime after the setting of the VCPU's request bit, the
7659 * correct TSC value must be set before the request. However,
7660 * to ensure the update actually makes it to any guest which
7661 * starts running in hardware virtualization between the set
7662 * and the acquisition of the spinlock, we must also ping the
7663 * CPU after setting the request bit.
7667 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7669 mutex_lock(&kvm_lock);
7670 list_for_each_entry(kvm, &vm_list, vm_list) {
7671 kvm_for_each_vcpu(i, vcpu, kvm) {
7672 if (vcpu->cpu != cpu)
7674 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7675 if (vcpu->cpu != raw_smp_processor_id())
7679 mutex_unlock(&kvm_lock);
7681 if (freq->old < freq->new && send_ipi) {
7683 * We upscale the frequency. Must make the guest
7684 * doesn't see old kvmclock values while running with
7685 * the new frequency, otherwise we risk the guest sees
7686 * time go backwards.
7688 * In case we update the frequency for another cpu
7689 * (which might be in guest context) send an interrupt
7690 * to kick the cpu out of guest context. Next time
7691 * guest context is entered kvmclock will be updated,
7692 * so the guest will not see stale values.
7694 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7698 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7701 struct cpufreq_freqs *freq = data;
7704 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7706 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7709 for_each_cpu(cpu, freq->policy->cpus)
7710 __kvmclock_cpufreq_notifier(freq, cpu);
7715 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7716 .notifier_call = kvmclock_cpufreq_notifier
7719 static int kvmclock_cpu_online(unsigned int cpu)
7721 tsc_khz_changed(NULL);
7725 static void kvm_timer_init(void)
7727 max_tsc_khz = tsc_khz;
7729 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7730 #ifdef CONFIG_CPU_FREQ
7731 struct cpufreq_policy *policy;
7735 policy = cpufreq_cpu_get(cpu);
7737 if (policy->cpuinfo.max_freq)
7738 max_tsc_khz = policy->cpuinfo.max_freq;
7739 cpufreq_cpu_put(policy);
7743 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7744 CPUFREQ_TRANSITION_NOTIFIER);
7747 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7748 kvmclock_cpu_online, kvmclock_cpu_down_prep);
7751 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7752 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7754 int kvm_is_in_guest(void)
7756 return __this_cpu_read(current_vcpu) != NULL;
7759 static int kvm_is_user_mode(void)
7763 if (__this_cpu_read(current_vcpu))
7764 user_mode = kvm_x86_ops.get_cpl(__this_cpu_read(current_vcpu));
7766 return user_mode != 0;
7769 static unsigned long kvm_get_guest_ip(void)
7771 unsigned long ip = 0;
7773 if (__this_cpu_read(current_vcpu))
7774 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7779 static void kvm_handle_intel_pt_intr(void)
7781 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7783 kvm_make_request(KVM_REQ_PMI, vcpu);
7784 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7785 (unsigned long *)&vcpu->arch.pmu.global_status);
7788 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7789 .is_in_guest = kvm_is_in_guest,
7790 .is_user_mode = kvm_is_user_mode,
7791 .get_guest_ip = kvm_get_guest_ip,
7792 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
7795 #ifdef CONFIG_X86_64
7796 static void pvclock_gtod_update_fn(struct work_struct *work)
7800 struct kvm_vcpu *vcpu;
7803 mutex_lock(&kvm_lock);
7804 list_for_each_entry(kvm, &vm_list, vm_list)
7805 kvm_for_each_vcpu(i, vcpu, kvm)
7806 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7807 atomic_set(&kvm_guest_has_master_clock, 0);
7808 mutex_unlock(&kvm_lock);
7811 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7814 * Notification about pvclock gtod data update.
7816 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7819 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7820 struct timekeeper *tk = priv;
7822 update_pvclock_gtod(tk);
7824 /* disable master clock if host does not trust, or does not
7825 * use, TSC based clocksource.
7827 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7828 atomic_read(&kvm_guest_has_master_clock) != 0)
7829 queue_work(system_long_wq, &pvclock_gtod_work);
7834 static struct notifier_block pvclock_gtod_notifier = {
7835 .notifier_call = pvclock_gtod_notify,
7839 int kvm_arch_init(void *opaque)
7841 struct kvm_x86_init_ops *ops = opaque;
7844 if (kvm_x86_ops.hardware_enable) {
7845 printk(KERN_ERR "kvm: already loaded the other module\n");
7850 if (!ops->cpu_has_kvm_support()) {
7851 pr_err_ratelimited("kvm: no hardware support\n");
7855 if (ops->disabled_by_bios()) {
7856 pr_err_ratelimited("kvm: disabled by bios\n");
7862 * KVM explicitly assumes that the guest has an FPU and
7863 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7864 * vCPU's FPU state as a fxregs_state struct.
7866 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7867 printk(KERN_ERR "kvm: inadequate fpu\n");
7873 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7874 __alignof__(struct fpu), SLAB_ACCOUNT,
7876 if (!x86_fpu_cache) {
7877 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7881 x86_emulator_cache = kvm_alloc_emulator_cache();
7882 if (!x86_emulator_cache) {
7883 pr_err("kvm: failed to allocate cache for x86 emulator\n");
7884 goto out_free_x86_fpu_cache;
7887 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
7888 if (!user_return_msrs) {
7889 printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
7890 goto out_free_x86_emulator_cache;
7893 r = kvm_mmu_module_init();
7895 goto out_free_percpu;
7897 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7898 PT_DIRTY_MASK, PT64_NX_MASK, 0,
7899 PT_PRESENT_MASK, 0, sme_me_mask);
7902 perf_register_guest_info_callbacks(&kvm_guest_cbs);
7904 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
7905 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7906 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
7910 if (pi_inject_timer == -1)
7911 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7912 #ifdef CONFIG_X86_64
7913 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7915 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7916 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7922 free_percpu(user_return_msrs);
7923 out_free_x86_emulator_cache:
7924 kmem_cache_destroy(x86_emulator_cache);
7925 out_free_x86_fpu_cache:
7926 kmem_cache_destroy(x86_fpu_cache);
7931 void kvm_arch_exit(void)
7933 #ifdef CONFIG_X86_64
7934 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7935 clear_hv_tscchange_cb();
7938 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7940 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7941 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7942 CPUFREQ_TRANSITION_NOTIFIER);
7943 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7944 #ifdef CONFIG_X86_64
7945 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7947 kvm_x86_ops.hardware_enable = NULL;
7948 kvm_mmu_module_exit();
7949 free_percpu(user_return_msrs);
7950 kmem_cache_destroy(x86_fpu_cache);
7953 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7955 ++vcpu->stat.halt_exits;
7956 if (lapic_in_kernel(vcpu)) {
7957 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7960 vcpu->run->exit_reason = KVM_EXIT_HLT;
7964 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7966 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7968 int ret = kvm_skip_emulated_instruction(vcpu);
7970 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7971 * KVM_EXIT_DEBUG here.
7973 return kvm_vcpu_halt(vcpu) && ret;
7975 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7977 #ifdef CONFIG_X86_64
7978 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7979 unsigned long clock_type)
7981 struct kvm_clock_pairing clock_pairing;
7982 struct timespec64 ts;
7986 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7987 return -KVM_EOPNOTSUPP;
7989 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7990 return -KVM_EOPNOTSUPP;
7992 clock_pairing.sec = ts.tv_sec;
7993 clock_pairing.nsec = ts.tv_nsec;
7994 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7995 clock_pairing.flags = 0;
7996 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7999 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8000 sizeof(struct kvm_clock_pairing)))
8008 * kvm_pv_kick_cpu_op: Kick a vcpu.
8010 * @apicid - apicid of vcpu to be kicked.
8012 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8014 struct kvm_lapic_irq lapic_irq;
8016 lapic_irq.shorthand = APIC_DEST_NOSHORT;
8017 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8018 lapic_irq.level = 0;
8019 lapic_irq.dest_id = apicid;
8020 lapic_irq.msi_redir_hint = false;
8022 lapic_irq.delivery_mode = APIC_DM_REMRD;
8023 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8026 bool kvm_apicv_activated(struct kvm *kvm)
8028 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8030 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8032 void kvm_apicv_init(struct kvm *kvm, bool enable)
8035 clear_bit(APICV_INHIBIT_REASON_DISABLE,
8036 &kvm->arch.apicv_inhibit_reasons);
8038 set_bit(APICV_INHIBIT_REASON_DISABLE,
8039 &kvm->arch.apicv_inhibit_reasons);
8041 EXPORT_SYMBOL_GPL(kvm_apicv_init);
8043 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
8045 struct kvm_vcpu *target = NULL;
8046 struct kvm_apic_map *map;
8049 map = rcu_dereference(kvm->arch.apic_map);
8051 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8052 target = map->phys_map[dest_id]->vcpu;
8056 if (target && READ_ONCE(target->ready))
8057 kvm_vcpu_yield_to(target);
8060 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8062 unsigned long nr, a0, a1, a2, a3, ret;
8065 if (kvm_hv_hypercall_enabled(vcpu->kvm))
8066 return kvm_hv_hypercall(vcpu);
8068 nr = kvm_rax_read(vcpu);
8069 a0 = kvm_rbx_read(vcpu);
8070 a1 = kvm_rcx_read(vcpu);
8071 a2 = kvm_rdx_read(vcpu);
8072 a3 = kvm_rsi_read(vcpu);
8074 trace_kvm_hypercall(nr, a0, a1, a2, a3);
8076 op_64_bit = is_64_bit_mode(vcpu);
8085 if (kvm_x86_ops.get_cpl(vcpu) != 0) {
8093 case KVM_HC_VAPIC_POLL_IRQ:
8096 case KVM_HC_KICK_CPU:
8097 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8100 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8101 kvm_sched_yield(vcpu->kvm, a1);
8104 #ifdef CONFIG_X86_64
8105 case KVM_HC_CLOCK_PAIRING:
8106 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8109 case KVM_HC_SEND_IPI:
8110 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8113 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8115 case KVM_HC_SCHED_YIELD:
8116 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8119 kvm_sched_yield(vcpu->kvm, a0);
8129 kvm_rax_write(vcpu, ret);
8131 ++vcpu->stat.hypercalls;
8132 return kvm_skip_emulated_instruction(vcpu);
8134 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8136 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8138 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8139 char instruction[3];
8140 unsigned long rip = kvm_rip_read(vcpu);
8142 kvm_x86_ops.patch_hypercall(vcpu, instruction);
8144 return emulator_write_emulated(ctxt, rip, instruction, 3,
8148 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8150 return vcpu->run->request_interrupt_window &&
8151 likely(!pic_in_kernel(vcpu->kvm));
8154 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8156 struct kvm_run *kvm_run = vcpu->run;
8158 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8159 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
8160 kvm_run->cr8 = kvm_get_cr8(vcpu);
8161 kvm_run->apic_base = kvm_get_apic_base(vcpu);
8162 kvm_run->ready_for_interrupt_injection =
8163 pic_in_kernel(vcpu->kvm) ||
8164 kvm_vcpu_ready_for_interrupt_injection(vcpu);
8167 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8171 if (!kvm_x86_ops.update_cr8_intercept)
8174 if (!lapic_in_kernel(vcpu))
8177 if (vcpu->arch.apicv_active)
8180 if (!vcpu->arch.apic->vapic_addr)
8181 max_irr = kvm_lapic_find_highest_irr(vcpu);
8188 tpr = kvm_lapic_get_cr8(vcpu);
8190 kvm_x86_ops.update_cr8_intercept(vcpu, tpr, max_irr);
8193 static void inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8196 bool can_inject = true;
8198 /* try to reinject previous events if any */
8200 if (vcpu->arch.exception.injected) {
8201 kvm_x86_ops.queue_exception(vcpu);
8205 * Do not inject an NMI or interrupt if there is a pending
8206 * exception. Exceptions and interrupts are recognized at
8207 * instruction boundaries, i.e. the start of an instruction.
8208 * Trap-like exceptions, e.g. #DB, have higher priority than
8209 * NMIs and interrupts, i.e. traps are recognized before an
8210 * NMI/interrupt that's pending on the same instruction.
8211 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8212 * priority, but are only generated (pended) during instruction
8213 * execution, i.e. a pending fault-like exception means the
8214 * fault occurred on the *previous* instruction and must be
8215 * serviced prior to recognizing any new events in order to
8216 * fully complete the previous instruction.
8218 else if (!vcpu->arch.exception.pending) {
8219 if (vcpu->arch.nmi_injected) {
8220 kvm_x86_ops.set_nmi(vcpu);
8222 } else if (vcpu->arch.interrupt.injected) {
8223 kvm_x86_ops.set_irq(vcpu);
8228 WARN_ON_ONCE(vcpu->arch.exception.injected &&
8229 vcpu->arch.exception.pending);
8232 * Call check_nested_events() even if we reinjected a previous event
8233 * in order for caller to determine if it should require immediate-exit
8234 * from L2 to L1 due to pending L1 events which require exit
8237 if (is_guest_mode(vcpu)) {
8238 r = kvm_x86_ops.nested_ops->check_events(vcpu);
8243 /* try to inject new event if pending */
8244 if (vcpu->arch.exception.pending) {
8245 trace_kvm_inj_exception(vcpu->arch.exception.nr,
8246 vcpu->arch.exception.has_error_code,
8247 vcpu->arch.exception.error_code);
8249 vcpu->arch.exception.pending = false;
8250 vcpu->arch.exception.injected = true;
8252 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8253 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8256 if (vcpu->arch.exception.nr == DB_VECTOR) {
8257 kvm_deliver_exception_payload(vcpu);
8258 if (vcpu->arch.dr7 & DR7_GD) {
8259 vcpu->arch.dr7 &= ~DR7_GD;
8260 kvm_update_dr7(vcpu);
8264 kvm_x86_ops.queue_exception(vcpu);
8269 * Finally, inject interrupt events. If an event cannot be injected
8270 * due to architectural conditions (e.g. IF=0) a window-open exit
8271 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
8272 * and can architecturally be injected, but we cannot do it right now:
8273 * an interrupt could have arrived just now and we have to inject it
8274 * as a vmexit, or there could already an event in the queue, which is
8275 * indicated by can_inject. In that case we request an immediate exit
8276 * in order to make progress and get back here for another iteration.
8277 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8279 if (vcpu->arch.smi_pending) {
8280 r = can_inject ? kvm_x86_ops.smi_allowed(vcpu, true) : -EBUSY;
8284 vcpu->arch.smi_pending = false;
8285 ++vcpu->arch.smi_count;
8289 kvm_x86_ops.enable_smi_window(vcpu);
8292 if (vcpu->arch.nmi_pending) {
8293 r = can_inject ? kvm_x86_ops.nmi_allowed(vcpu, true) : -EBUSY;
8297 --vcpu->arch.nmi_pending;
8298 vcpu->arch.nmi_injected = true;
8299 kvm_x86_ops.set_nmi(vcpu);
8301 WARN_ON(kvm_x86_ops.nmi_allowed(vcpu, true) < 0);
8303 if (vcpu->arch.nmi_pending)
8304 kvm_x86_ops.enable_nmi_window(vcpu);
8307 if (kvm_cpu_has_injectable_intr(vcpu)) {
8308 r = can_inject ? kvm_x86_ops.interrupt_allowed(vcpu, true) : -EBUSY;
8312 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8313 kvm_x86_ops.set_irq(vcpu);
8314 WARN_ON(kvm_x86_ops.interrupt_allowed(vcpu, true) < 0);
8316 if (kvm_cpu_has_injectable_intr(vcpu))
8317 kvm_x86_ops.enable_irq_window(vcpu);
8320 if (is_guest_mode(vcpu) &&
8321 kvm_x86_ops.nested_ops->hv_timer_pending &&
8322 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8323 *req_immediate_exit = true;
8325 WARN_ON(vcpu->arch.exception.pending);
8329 *req_immediate_exit = true;
8333 static void process_nmi(struct kvm_vcpu *vcpu)
8338 * x86 is limited to one NMI running, and one NMI pending after it.
8339 * If an NMI is already in progress, limit further NMIs to just one.
8340 * Otherwise, allow two (and we'll inject the first one immediately).
8342 if (kvm_x86_ops.get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
8345 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8346 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8347 kvm_make_request(KVM_REQ_EVENT, vcpu);
8350 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8353 flags |= seg->g << 23;
8354 flags |= seg->db << 22;
8355 flags |= seg->l << 21;
8356 flags |= seg->avl << 20;
8357 flags |= seg->present << 15;
8358 flags |= seg->dpl << 13;
8359 flags |= seg->s << 12;
8360 flags |= seg->type << 8;
8364 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8366 struct kvm_segment seg;
8369 kvm_get_segment(vcpu, &seg, n);
8370 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
8373 offset = 0x7f84 + n * 12;
8375 offset = 0x7f2c + (n - 3) * 12;
8377 put_smstate(u32, buf, offset + 8, seg.base);
8378 put_smstate(u32, buf, offset + 4, seg.limit);
8379 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
8382 #ifdef CONFIG_X86_64
8383 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
8385 struct kvm_segment seg;
8389 kvm_get_segment(vcpu, &seg, n);
8390 offset = 0x7e00 + n * 16;
8392 flags = enter_smm_get_segment_flags(&seg) >> 8;
8393 put_smstate(u16, buf, offset, seg.selector);
8394 put_smstate(u16, buf, offset + 2, flags);
8395 put_smstate(u32, buf, offset + 4, seg.limit);
8396 put_smstate(u64, buf, offset + 8, seg.base);
8400 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
8403 struct kvm_segment seg;
8407 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
8408 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
8409 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
8410 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
8412 for (i = 0; i < 8; i++)
8413 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
8415 kvm_get_dr(vcpu, 6, &val);
8416 put_smstate(u32, buf, 0x7fcc, (u32)val);
8417 kvm_get_dr(vcpu, 7, &val);
8418 put_smstate(u32, buf, 0x7fc8, (u32)val);
8420 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8421 put_smstate(u32, buf, 0x7fc4, seg.selector);
8422 put_smstate(u32, buf, 0x7f64, seg.base);
8423 put_smstate(u32, buf, 0x7f60, seg.limit);
8424 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
8426 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8427 put_smstate(u32, buf, 0x7fc0, seg.selector);
8428 put_smstate(u32, buf, 0x7f80, seg.base);
8429 put_smstate(u32, buf, 0x7f7c, seg.limit);
8430 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
8432 kvm_x86_ops.get_gdt(vcpu, &dt);
8433 put_smstate(u32, buf, 0x7f74, dt.address);
8434 put_smstate(u32, buf, 0x7f70, dt.size);
8436 kvm_x86_ops.get_idt(vcpu, &dt);
8437 put_smstate(u32, buf, 0x7f58, dt.address);
8438 put_smstate(u32, buf, 0x7f54, dt.size);
8440 for (i = 0; i < 6; i++)
8441 enter_smm_save_seg_32(vcpu, buf, i);
8443 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
8446 put_smstate(u32, buf, 0x7efc, 0x00020000);
8447 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
8450 #ifdef CONFIG_X86_64
8451 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
8454 struct kvm_segment seg;
8458 for (i = 0; i < 16; i++)
8459 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
8461 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
8462 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
8464 kvm_get_dr(vcpu, 6, &val);
8465 put_smstate(u64, buf, 0x7f68, val);
8466 kvm_get_dr(vcpu, 7, &val);
8467 put_smstate(u64, buf, 0x7f60, val);
8469 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
8470 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
8471 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
8473 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
8476 put_smstate(u32, buf, 0x7efc, 0x00020064);
8478 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
8480 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8481 put_smstate(u16, buf, 0x7e90, seg.selector);
8482 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
8483 put_smstate(u32, buf, 0x7e94, seg.limit);
8484 put_smstate(u64, buf, 0x7e98, seg.base);
8486 kvm_x86_ops.get_idt(vcpu, &dt);
8487 put_smstate(u32, buf, 0x7e84, dt.size);
8488 put_smstate(u64, buf, 0x7e88, dt.address);
8490 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8491 put_smstate(u16, buf, 0x7e70, seg.selector);
8492 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
8493 put_smstate(u32, buf, 0x7e74, seg.limit);
8494 put_smstate(u64, buf, 0x7e78, seg.base);
8496 kvm_x86_ops.get_gdt(vcpu, &dt);
8497 put_smstate(u32, buf, 0x7e64, dt.size);
8498 put_smstate(u64, buf, 0x7e68, dt.address);
8500 for (i = 0; i < 6; i++)
8501 enter_smm_save_seg_64(vcpu, buf, i);
8505 static void enter_smm(struct kvm_vcpu *vcpu)
8507 struct kvm_segment cs, ds;
8512 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
8513 memset(buf, 0, 512);
8514 #ifdef CONFIG_X86_64
8515 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8516 enter_smm_save_state_64(vcpu, buf);
8519 enter_smm_save_state_32(vcpu, buf);
8522 * Give pre_enter_smm() a chance to make ISA-specific changes to the
8523 * vCPU state (e.g. leave guest mode) after we've saved the state into
8524 * the SMM state-save area.
8526 kvm_x86_ops.pre_enter_smm(vcpu, buf);
8528 vcpu->arch.hflags |= HF_SMM_MASK;
8529 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
8531 if (kvm_x86_ops.get_nmi_mask(vcpu))
8532 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
8534 kvm_x86_ops.set_nmi_mask(vcpu, true);
8536 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
8537 kvm_rip_write(vcpu, 0x8000);
8539 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
8540 kvm_x86_ops.set_cr0(vcpu, cr0);
8541 vcpu->arch.cr0 = cr0;
8543 kvm_x86_ops.set_cr4(vcpu, 0);
8545 /* Undocumented: IDT limit is set to zero on entry to SMM. */
8546 dt.address = dt.size = 0;
8547 kvm_x86_ops.set_idt(vcpu, &dt);
8549 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
8551 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
8552 cs.base = vcpu->arch.smbase;
8557 cs.limit = ds.limit = 0xffffffff;
8558 cs.type = ds.type = 0x3;
8559 cs.dpl = ds.dpl = 0;
8564 cs.avl = ds.avl = 0;
8565 cs.present = ds.present = 1;
8566 cs.unusable = ds.unusable = 0;
8567 cs.padding = ds.padding = 0;
8569 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8570 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
8571 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
8572 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
8573 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
8574 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
8576 #ifdef CONFIG_X86_64
8577 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8578 kvm_x86_ops.set_efer(vcpu, 0);
8581 kvm_update_cpuid_runtime(vcpu);
8582 kvm_mmu_reset_context(vcpu);
8585 static void process_smi(struct kvm_vcpu *vcpu)
8587 vcpu->arch.smi_pending = true;
8588 kvm_make_request(KVM_REQ_EVENT, vcpu);
8591 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
8592 unsigned long *vcpu_bitmap)
8596 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
8598 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
8599 NULL, vcpu_bitmap, cpus);
8601 free_cpumask_var(cpus);
8604 void kvm_make_scan_ioapic_request(struct kvm *kvm)
8606 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
8609 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
8611 if (!lapic_in_kernel(vcpu))
8614 vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
8615 kvm_apic_update_apicv(vcpu);
8616 kvm_x86_ops.refresh_apicv_exec_ctrl(vcpu);
8618 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
8621 * NOTE: Do not hold any lock prior to calling this.
8623 * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
8624 * locked, because it calls __x86_set_memory_region() which does
8625 * synchronize_srcu(&kvm->srcu).
8627 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
8629 struct kvm_vcpu *except;
8630 unsigned long old, new, expected;
8632 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
8633 !kvm_x86_ops.check_apicv_inhibit_reasons(bit))
8636 old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
8638 expected = new = old;
8640 __clear_bit(bit, &new);
8642 __set_bit(bit, &new);
8645 old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
8646 } while (old != expected);
8651 trace_kvm_apicv_update_request(activate, bit);
8652 if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
8653 kvm_x86_ops.pre_update_apicv_exec_ctrl(kvm, activate);
8656 * Sending request to update APICV for all other vcpus,
8657 * while update the calling vcpu immediately instead of
8658 * waiting for another #VMEXIT to handle the request.
8660 except = kvm_get_running_vcpu();
8661 kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
8664 kvm_vcpu_update_apicv(except);
8666 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
8668 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
8670 if (!kvm_apic_present(vcpu))
8673 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
8675 if (irqchip_split(vcpu->kvm))
8676 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
8678 if (vcpu->arch.apicv_active)
8679 kvm_x86_ops.sync_pir_to_irr(vcpu);
8680 if (ioapic_in_kernel(vcpu->kvm))
8681 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
8684 if (is_guest_mode(vcpu))
8685 vcpu->arch.load_eoi_exitmap_pending = true;
8687 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
8690 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8692 u64 eoi_exit_bitmap[4];
8694 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
8697 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
8698 vcpu_to_synic(vcpu)->vec_bitmap, 256);
8699 kvm_x86_ops.load_eoi_exitmap(vcpu, eoi_exit_bitmap);
8702 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
8703 unsigned long start, unsigned long end)
8705 unsigned long apic_address;
8708 * The physical address of apic access page is stored in the VMCS.
8709 * Update it when it becomes invalid.
8711 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8712 if (start <= apic_address && apic_address < end)
8713 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
8716 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
8718 if (!lapic_in_kernel(vcpu))
8721 if (!kvm_x86_ops.set_apic_access_page_addr)
8724 kvm_x86_ops.set_apic_access_page_addr(vcpu);
8727 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
8729 smp_send_reschedule(vcpu->cpu);
8731 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
8734 * Returns 1 to let vcpu_run() continue the guest execution loop without
8735 * exiting to the userspace. Otherwise, the value will be returned to the
8738 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
8742 dm_request_for_irq_injection(vcpu) &&
8743 kvm_cpu_accept_dm_intr(vcpu);
8744 fastpath_t exit_fastpath;
8746 bool req_immediate_exit = false;
8748 if (kvm_request_pending(vcpu)) {
8749 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
8750 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
8755 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
8756 kvm_mmu_unload(vcpu);
8757 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
8758 __kvm_migrate_timers(vcpu);
8759 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
8760 kvm_gen_update_masterclock(vcpu->kvm);
8761 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
8762 kvm_gen_kvmclock_update(vcpu);
8763 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
8764 r = kvm_guest_time_update(vcpu);
8768 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
8769 kvm_mmu_sync_roots(vcpu);
8770 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
8771 kvm_mmu_load_pgd(vcpu);
8772 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
8773 kvm_vcpu_flush_tlb_all(vcpu);
8775 /* Flushing all ASIDs flushes the current ASID... */
8776 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
8778 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
8779 kvm_vcpu_flush_tlb_current(vcpu);
8780 if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu))
8781 kvm_vcpu_flush_tlb_guest(vcpu);
8783 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
8784 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
8788 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8789 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
8790 vcpu->mmio_needed = 0;
8794 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
8795 /* Page is swapped out. Do synthetic halt */
8796 vcpu->arch.apf.halted = true;
8800 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8801 record_steal_time(vcpu);
8802 if (kvm_check_request(KVM_REQ_SMI, vcpu))
8804 if (kvm_check_request(KVM_REQ_NMI, vcpu))
8806 if (kvm_check_request(KVM_REQ_PMU, vcpu))
8807 kvm_pmu_handle_event(vcpu);
8808 if (kvm_check_request(KVM_REQ_PMI, vcpu))
8809 kvm_pmu_deliver_pmi(vcpu);
8810 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8811 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8812 if (test_bit(vcpu->arch.pending_ioapic_eoi,
8813 vcpu->arch.ioapic_handled_vectors)) {
8814 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8815 vcpu->run->eoi.vector =
8816 vcpu->arch.pending_ioapic_eoi;
8821 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8822 vcpu_scan_ioapic(vcpu);
8823 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8824 vcpu_load_eoi_exitmap(vcpu);
8825 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8826 kvm_vcpu_reload_apic_access_page(vcpu);
8827 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8828 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8829 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8833 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8834 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8835 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8839 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8840 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8841 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8847 * KVM_REQ_HV_STIMER has to be processed after
8848 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8849 * depend on the guest clock being up-to-date
8851 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8852 kvm_hv_process_stimers(vcpu);
8853 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
8854 kvm_vcpu_update_apicv(vcpu);
8855 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
8856 kvm_check_async_pf_completion(vcpu);
8857 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
8858 kvm_x86_ops.msr_filter_changed(vcpu);
8861 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8862 ++vcpu->stat.req_event;
8863 kvm_apic_accept_events(vcpu);
8864 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8869 inject_pending_event(vcpu, &req_immediate_exit);
8871 kvm_x86_ops.enable_irq_window(vcpu);
8873 if (kvm_lapic_enabled(vcpu)) {
8874 update_cr8_intercept(vcpu);
8875 kvm_lapic_sync_to_vapic(vcpu);
8879 r = kvm_mmu_reload(vcpu);
8881 goto cancel_injection;
8886 kvm_x86_ops.prepare_guest_switch(vcpu);
8889 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
8890 * IPI are then delayed after guest entry, which ensures that they
8891 * result in virtual interrupt delivery.
8893 local_irq_disable();
8894 vcpu->mode = IN_GUEST_MODE;
8896 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8899 * 1) We should set ->mode before checking ->requests. Please see
8900 * the comment in kvm_vcpu_exiting_guest_mode().
8902 * 2) For APICv, we should set ->mode before checking PID.ON. This
8903 * pairs with the memory barrier implicit in pi_test_and_set_on
8904 * (see vmx_deliver_posted_interrupt).
8906 * 3) This also orders the write to mode from any reads to the page
8907 * tables done while the VCPU is running. Please see the comment
8908 * in kvm_flush_remote_tlbs.
8910 smp_mb__after_srcu_read_unlock();
8913 * This handles the case where a posted interrupt was
8914 * notified with kvm_vcpu_kick.
8916 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8917 kvm_x86_ops.sync_pir_to_irr(vcpu);
8919 if (kvm_vcpu_exit_request(vcpu)) {
8920 vcpu->mode = OUTSIDE_GUEST_MODE;
8924 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8926 goto cancel_injection;
8929 if (req_immediate_exit) {
8930 kvm_make_request(KVM_REQ_EVENT, vcpu);
8931 kvm_x86_ops.request_immediate_exit(vcpu);
8934 trace_kvm_entry(vcpu);
8936 fpregs_assert_state_consistent();
8937 if (test_thread_flag(TIF_NEED_FPU_LOAD))
8938 switch_fpu_return();
8940 if (unlikely(vcpu->arch.switch_db_regs)) {
8942 set_debugreg(vcpu->arch.eff_db[0], 0);
8943 set_debugreg(vcpu->arch.eff_db[1], 1);
8944 set_debugreg(vcpu->arch.eff_db[2], 2);
8945 set_debugreg(vcpu->arch.eff_db[3], 3);
8946 set_debugreg(vcpu->arch.dr6, 6);
8947 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8950 exit_fastpath = kvm_x86_ops.run(vcpu);
8953 * Do this here before restoring debug registers on the host. And
8954 * since we do this before handling the vmexit, a DR access vmexit
8955 * can (a) read the correct value of the debug registers, (b) set
8956 * KVM_DEBUGREG_WONT_EXIT again.
8958 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8959 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8960 kvm_x86_ops.sync_dirty_debug_regs(vcpu);
8961 kvm_update_dr0123(vcpu);
8962 kvm_update_dr7(vcpu);
8963 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8967 * If the guest has used debug registers, at least dr7
8968 * will be disabled while returning to the host.
8969 * If we don't have active breakpoints in the host, we don't
8970 * care about the messed up debug address registers. But if
8971 * we have some of them active, restore the old state.
8973 if (hw_breakpoint_active())
8974 hw_breakpoint_restore();
8976 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
8977 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8979 vcpu->mode = OUTSIDE_GUEST_MODE;
8982 kvm_x86_ops.handle_exit_irqoff(vcpu);
8985 * Consume any pending interrupts, including the possible source of
8986 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8987 * An instruction is required after local_irq_enable() to fully unblock
8988 * interrupts on processors that implement an interrupt shadow, the
8989 * stat.exits increment will do nicely.
8991 kvm_before_interrupt(vcpu);
8994 local_irq_disable();
8995 kvm_after_interrupt(vcpu);
8997 if (lapic_in_kernel(vcpu)) {
8998 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8999 if (delta != S64_MIN) {
9000 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9001 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9008 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9011 * Profile KVM exit RIPs:
9013 if (unlikely(prof_on == KVM_PROFILING)) {
9014 unsigned long rip = kvm_rip_read(vcpu);
9015 profile_hit(KVM_PROFILING, (void *)rip);
9018 if (unlikely(vcpu->arch.tsc_always_catchup))
9019 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9021 if (vcpu->arch.apic_attention)
9022 kvm_lapic_sync_from_vapic(vcpu);
9024 r = kvm_x86_ops.handle_exit(vcpu, exit_fastpath);
9028 if (req_immediate_exit)
9029 kvm_make_request(KVM_REQ_EVENT, vcpu);
9030 kvm_x86_ops.cancel_injection(vcpu);
9031 if (unlikely(vcpu->arch.apic_attention))
9032 kvm_lapic_sync_from_vapic(vcpu);
9037 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9039 if (!kvm_arch_vcpu_runnable(vcpu) &&
9040 (!kvm_x86_ops.pre_block || kvm_x86_ops.pre_block(vcpu) == 0)) {
9041 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9042 kvm_vcpu_block(vcpu);
9043 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9045 if (kvm_x86_ops.post_block)
9046 kvm_x86_ops.post_block(vcpu);
9048 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9052 kvm_apic_accept_events(vcpu);
9053 switch(vcpu->arch.mp_state) {
9054 case KVM_MP_STATE_HALTED:
9055 vcpu->arch.pv.pv_unhalted = false;
9056 vcpu->arch.mp_state =
9057 KVM_MP_STATE_RUNNABLE;
9059 case KVM_MP_STATE_RUNNABLE:
9060 vcpu->arch.apf.halted = false;
9062 case KVM_MP_STATE_INIT_RECEIVED:
9070 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9072 if (is_guest_mode(vcpu))
9073 kvm_x86_ops.nested_ops->check_events(vcpu);
9075 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9076 !vcpu->arch.apf.halted);
9079 static int vcpu_run(struct kvm_vcpu *vcpu)
9082 struct kvm *kvm = vcpu->kvm;
9084 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9085 vcpu->arch.l1tf_flush_l1d = true;
9088 if (kvm_vcpu_running(vcpu)) {
9089 r = vcpu_enter_guest(vcpu);
9091 r = vcpu_block(kvm, vcpu);
9097 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
9098 if (kvm_cpu_has_pending_timer(vcpu))
9099 kvm_inject_pending_timer_irqs(vcpu);
9101 if (dm_request_for_irq_injection(vcpu) &&
9102 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9104 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9105 ++vcpu->stat.request_irq_exits;
9109 if (__xfer_to_guest_mode_work_pending()) {
9110 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9111 r = xfer_to_guest_mode_handle_work(vcpu);
9114 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9118 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9123 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9127 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9128 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9129 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9133 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9135 BUG_ON(!vcpu->arch.pio.count);
9137 return complete_emulated_io(vcpu);
9141 * Implements the following, as a state machine:
9145 * for each mmio piece in the fragment
9153 * for each mmio piece in the fragment
9158 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9160 struct kvm_run *run = vcpu->run;
9161 struct kvm_mmio_fragment *frag;
9164 BUG_ON(!vcpu->mmio_needed);
9166 /* Complete previous fragment */
9167 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9168 len = min(8u, frag->len);
9169 if (!vcpu->mmio_is_write)
9170 memcpy(frag->data, run->mmio.data, len);
9172 if (frag->len <= 8) {
9173 /* Switch to the next fragment. */
9175 vcpu->mmio_cur_fragment++;
9177 /* Go forward to the next mmio piece. */
9183 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9184 vcpu->mmio_needed = 0;
9186 /* FIXME: return into emulator if single-stepping. */
9187 if (vcpu->mmio_is_write)
9189 vcpu->mmio_read_completed = 1;
9190 return complete_emulated_io(vcpu);
9193 run->exit_reason = KVM_EXIT_MMIO;
9194 run->mmio.phys_addr = frag->gpa;
9195 if (vcpu->mmio_is_write)
9196 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9197 run->mmio.len = min(8u, frag->len);
9198 run->mmio.is_write = vcpu->mmio_is_write;
9199 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9203 static void kvm_save_current_fpu(struct fpu *fpu)
9206 * If the target FPU state is not resident in the CPU registers, just
9207 * memcpy() from current, else save CPU state directly to the target.
9209 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9210 memcpy(&fpu->state, ¤t->thread.fpu.state,
9211 fpu_kernel_xstate_size);
9213 copy_fpregs_to_fpstate(fpu);
9216 /* Swap (qemu) user FPU context for the guest FPU context. */
9217 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9221 kvm_save_current_fpu(vcpu->arch.user_fpu);
9223 /* PKRU is separately restored in kvm_x86_ops.run. */
9224 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
9225 ~XFEATURE_MASK_PKRU);
9227 fpregs_mark_activate();
9233 /* When vcpu_run ends, restore user space FPU context. */
9234 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9238 kvm_save_current_fpu(vcpu->arch.guest_fpu);
9240 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
9242 fpregs_mark_activate();
9245 ++vcpu->stat.fpu_reload;
9249 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9251 struct kvm_run *kvm_run = vcpu->run;
9255 kvm_sigset_activate(vcpu);
9256 kvm_load_guest_fpu(vcpu);
9258 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9259 if (kvm_run->immediate_exit) {
9263 kvm_vcpu_block(vcpu);
9264 kvm_apic_accept_events(vcpu);
9265 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9267 if (signal_pending(current)) {
9269 kvm_run->exit_reason = KVM_EXIT_INTR;
9270 ++vcpu->stat.signal_exits;
9275 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
9280 if (kvm_run->kvm_dirty_regs) {
9281 r = sync_regs(vcpu);
9286 /* re-sync apic's tpr */
9287 if (!lapic_in_kernel(vcpu)) {
9288 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
9294 if (unlikely(vcpu->arch.complete_userspace_io)) {
9295 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
9296 vcpu->arch.complete_userspace_io = NULL;
9301 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
9303 if (kvm_run->immediate_exit)
9309 kvm_put_guest_fpu(vcpu);
9310 if (kvm_run->kvm_valid_regs)
9312 post_kvm_run_save(vcpu);
9313 kvm_sigset_deactivate(vcpu);
9319 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9321 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
9323 * We are here if userspace calls get_regs() in the middle of
9324 * instruction emulation. Registers state needs to be copied
9325 * back from emulation context to vcpu. Userspace shouldn't do
9326 * that usually, but some bad designed PV devices (vmware
9327 * backdoor interface) need this to work
9329 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
9330 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9332 regs->rax = kvm_rax_read(vcpu);
9333 regs->rbx = kvm_rbx_read(vcpu);
9334 regs->rcx = kvm_rcx_read(vcpu);
9335 regs->rdx = kvm_rdx_read(vcpu);
9336 regs->rsi = kvm_rsi_read(vcpu);
9337 regs->rdi = kvm_rdi_read(vcpu);
9338 regs->rsp = kvm_rsp_read(vcpu);
9339 regs->rbp = kvm_rbp_read(vcpu);
9340 #ifdef CONFIG_X86_64
9341 regs->r8 = kvm_r8_read(vcpu);
9342 regs->r9 = kvm_r9_read(vcpu);
9343 regs->r10 = kvm_r10_read(vcpu);
9344 regs->r11 = kvm_r11_read(vcpu);
9345 regs->r12 = kvm_r12_read(vcpu);
9346 regs->r13 = kvm_r13_read(vcpu);
9347 regs->r14 = kvm_r14_read(vcpu);
9348 regs->r15 = kvm_r15_read(vcpu);
9351 regs->rip = kvm_rip_read(vcpu);
9352 regs->rflags = kvm_get_rflags(vcpu);
9355 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9358 __get_regs(vcpu, regs);
9363 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9365 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
9366 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9368 kvm_rax_write(vcpu, regs->rax);
9369 kvm_rbx_write(vcpu, regs->rbx);
9370 kvm_rcx_write(vcpu, regs->rcx);
9371 kvm_rdx_write(vcpu, regs->rdx);
9372 kvm_rsi_write(vcpu, regs->rsi);
9373 kvm_rdi_write(vcpu, regs->rdi);
9374 kvm_rsp_write(vcpu, regs->rsp);
9375 kvm_rbp_write(vcpu, regs->rbp);
9376 #ifdef CONFIG_X86_64
9377 kvm_r8_write(vcpu, regs->r8);
9378 kvm_r9_write(vcpu, regs->r9);
9379 kvm_r10_write(vcpu, regs->r10);
9380 kvm_r11_write(vcpu, regs->r11);
9381 kvm_r12_write(vcpu, regs->r12);
9382 kvm_r13_write(vcpu, regs->r13);
9383 kvm_r14_write(vcpu, regs->r14);
9384 kvm_r15_write(vcpu, regs->r15);
9387 kvm_rip_write(vcpu, regs->rip);
9388 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
9390 vcpu->arch.exception.pending = false;
9392 kvm_make_request(KVM_REQ_EVENT, vcpu);
9395 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9398 __set_regs(vcpu, regs);
9403 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
9405 struct kvm_segment cs;
9407 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9411 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
9413 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9417 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9418 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9419 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9420 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9421 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9422 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9424 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9425 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9427 kvm_x86_ops.get_idt(vcpu, &dt);
9428 sregs->idt.limit = dt.size;
9429 sregs->idt.base = dt.address;
9430 kvm_x86_ops.get_gdt(vcpu, &dt);
9431 sregs->gdt.limit = dt.size;
9432 sregs->gdt.base = dt.address;
9434 sregs->cr0 = kvm_read_cr0(vcpu);
9435 sregs->cr2 = vcpu->arch.cr2;
9436 sregs->cr3 = kvm_read_cr3(vcpu);
9437 sregs->cr4 = kvm_read_cr4(vcpu);
9438 sregs->cr8 = kvm_get_cr8(vcpu);
9439 sregs->efer = vcpu->arch.efer;
9440 sregs->apic_base = kvm_get_apic_base(vcpu);
9442 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
9444 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
9445 set_bit(vcpu->arch.interrupt.nr,
9446 (unsigned long *)sregs->interrupt_bitmap);
9449 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
9450 struct kvm_sregs *sregs)
9453 __get_sregs(vcpu, sregs);
9458 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
9459 struct kvm_mp_state *mp_state)
9462 if (kvm_mpx_supported())
9463 kvm_load_guest_fpu(vcpu);
9465 kvm_apic_accept_events(vcpu);
9466 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
9467 vcpu->arch.pv.pv_unhalted)
9468 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
9470 mp_state->mp_state = vcpu->arch.mp_state;
9472 if (kvm_mpx_supported())
9473 kvm_put_guest_fpu(vcpu);
9478 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
9479 struct kvm_mp_state *mp_state)
9485 if (!lapic_in_kernel(vcpu) &&
9486 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
9490 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
9491 * INIT state; latched init should be reported using
9492 * KVM_SET_VCPU_EVENTS, so reject it here.
9494 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
9495 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
9496 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
9499 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
9500 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
9501 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
9503 vcpu->arch.mp_state = mp_state->mp_state;
9504 kvm_make_request(KVM_REQ_EVENT, vcpu);
9512 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
9513 int reason, bool has_error_code, u32 error_code)
9515 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
9518 init_emulate_ctxt(vcpu);
9520 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9521 has_error_code, error_code);
9523 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
9524 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
9525 vcpu->run->internal.ndata = 0;
9529 kvm_rip_write(vcpu, ctxt->eip);
9530 kvm_set_rflags(vcpu, ctxt->eflags);
9533 EXPORT_SYMBOL_GPL(kvm_task_switch);
9535 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9537 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
9539 * When EFER.LME and CR0.PG are set, the processor is in
9540 * 64-bit mode (though maybe in a 32-bit code segment).
9541 * CR4.PAE and EFER.LMA must be set.
9543 if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
9547 * Not in 64-bit mode: EFER.LMA is clear and the code
9548 * segment cannot be 64-bit.
9550 if (sregs->efer & EFER_LMA || sregs->cs.l)
9554 return kvm_is_valid_cr4(vcpu, sregs->cr4);
9557 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9559 struct msr_data apic_base_msr;
9560 int mmu_reset_needed = 0;
9561 int cpuid_update_needed = 0;
9562 int pending_vec, max_bits, idx;
9566 if (!kvm_is_valid_sregs(vcpu, sregs))
9569 apic_base_msr.data = sregs->apic_base;
9570 apic_base_msr.host_initiated = true;
9571 if (kvm_set_apic_base(vcpu, &apic_base_msr))
9574 dt.size = sregs->idt.limit;
9575 dt.address = sregs->idt.base;
9576 kvm_x86_ops.set_idt(vcpu, &dt);
9577 dt.size = sregs->gdt.limit;
9578 dt.address = sregs->gdt.base;
9579 kvm_x86_ops.set_gdt(vcpu, &dt);
9581 vcpu->arch.cr2 = sregs->cr2;
9582 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
9583 vcpu->arch.cr3 = sregs->cr3;
9584 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
9586 kvm_set_cr8(vcpu, sregs->cr8);
9588 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
9589 kvm_x86_ops.set_efer(vcpu, sregs->efer);
9591 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
9592 kvm_x86_ops.set_cr0(vcpu, sregs->cr0);
9593 vcpu->arch.cr0 = sregs->cr0;
9595 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
9596 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
9597 (X86_CR4_OSXSAVE | X86_CR4_PKE));
9598 kvm_x86_ops.set_cr4(vcpu, sregs->cr4);
9599 if (cpuid_update_needed)
9600 kvm_update_cpuid_runtime(vcpu);
9602 idx = srcu_read_lock(&vcpu->kvm->srcu);
9603 if (is_pae_paging(vcpu)) {
9604 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
9605 mmu_reset_needed = 1;
9607 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9609 if (mmu_reset_needed)
9610 kvm_mmu_reset_context(vcpu);
9612 max_bits = KVM_NR_INTERRUPTS;
9613 pending_vec = find_first_bit(
9614 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
9615 if (pending_vec < max_bits) {
9616 kvm_queue_interrupt(vcpu, pending_vec, false);
9617 pr_debug("Set back pending irq %d\n", pending_vec);
9620 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9621 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9622 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9623 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9624 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9625 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9627 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9628 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9630 update_cr8_intercept(vcpu);
9632 /* Older userspace won't unhalt the vcpu on reset. */
9633 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9634 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
9636 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9638 kvm_make_request(KVM_REQ_EVENT, vcpu);
9645 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
9646 struct kvm_sregs *sregs)
9651 ret = __set_sregs(vcpu, sregs);
9656 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
9657 struct kvm_guest_debug *dbg)
9659 unsigned long rflags;
9664 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
9666 if (vcpu->arch.exception.pending)
9668 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
9669 kvm_queue_exception(vcpu, DB_VECTOR);
9671 kvm_queue_exception(vcpu, BP_VECTOR);
9675 * Read rflags as long as potentially injected trace flags are still
9678 rflags = kvm_get_rflags(vcpu);
9680 vcpu->guest_debug = dbg->control;
9681 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
9682 vcpu->guest_debug = 0;
9684 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
9685 for (i = 0; i < KVM_NR_DB_REGS; ++i)
9686 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
9687 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
9689 for (i = 0; i < KVM_NR_DB_REGS; i++)
9690 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
9692 kvm_update_dr7(vcpu);
9694 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9695 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
9696 get_segment_base(vcpu, VCPU_SREG_CS);
9699 * Trigger an rflags update that will inject or remove the trace
9702 kvm_set_rflags(vcpu, rflags);
9704 kvm_x86_ops.update_exception_bitmap(vcpu);
9714 * Translate a guest virtual address to a guest physical address.
9716 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
9717 struct kvm_translation *tr)
9719 unsigned long vaddr = tr->linear_address;
9725 idx = srcu_read_lock(&vcpu->kvm->srcu);
9726 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
9727 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9728 tr->physical_address = gpa;
9729 tr->valid = gpa != UNMAPPED_GVA;
9737 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9739 struct fxregs_state *fxsave;
9743 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9744 memcpy(fpu->fpr, fxsave->st_space, 128);
9745 fpu->fcw = fxsave->cwd;
9746 fpu->fsw = fxsave->swd;
9747 fpu->ftwx = fxsave->twd;
9748 fpu->last_opcode = fxsave->fop;
9749 fpu->last_ip = fxsave->rip;
9750 fpu->last_dp = fxsave->rdp;
9751 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
9757 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9759 struct fxregs_state *fxsave;
9763 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9765 memcpy(fxsave->st_space, fpu->fpr, 128);
9766 fxsave->cwd = fpu->fcw;
9767 fxsave->swd = fpu->fsw;
9768 fxsave->twd = fpu->ftwx;
9769 fxsave->fop = fpu->last_opcode;
9770 fxsave->rip = fpu->last_ip;
9771 fxsave->rdp = fpu->last_dp;
9772 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
9778 static void store_regs(struct kvm_vcpu *vcpu)
9780 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
9782 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
9783 __get_regs(vcpu, &vcpu->run->s.regs.regs);
9785 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
9786 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
9788 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
9789 kvm_vcpu_ioctl_x86_get_vcpu_events(
9790 vcpu, &vcpu->run->s.regs.events);
9793 static int sync_regs(struct kvm_vcpu *vcpu)
9795 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
9798 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
9799 __set_regs(vcpu, &vcpu->run->s.regs.regs);
9800 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
9802 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
9803 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
9805 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
9807 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
9808 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9809 vcpu, &vcpu->run->s.regs.events))
9811 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9817 static void fx_init(struct kvm_vcpu *vcpu)
9819 fpstate_init(&vcpu->arch.guest_fpu->state);
9820 if (boot_cpu_has(X86_FEATURE_XSAVES))
9821 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9822 host_xcr0 | XSTATE_COMPACTION_ENABLED;
9825 * Ensure guest xcr0 is valid for loading
9827 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9829 vcpu->arch.cr0 |= X86_CR0_ET;
9832 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
9834 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9835 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
9836 "guest TSC will not be reliable\n");
9841 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
9846 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9847 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9849 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9851 kvm_set_tsc_khz(vcpu, max_tsc_khz);
9853 r = kvm_mmu_create(vcpu);
9857 if (irqchip_in_kernel(vcpu->kvm)) {
9858 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9860 goto fail_mmu_destroy;
9861 if (kvm_apicv_activated(vcpu->kvm))
9862 vcpu->arch.apicv_active = true;
9864 static_key_slow_inc(&kvm_no_apic_vcpu);
9868 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9870 goto fail_free_lapic;
9871 vcpu->arch.pio_data = page_address(page);
9873 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9874 GFP_KERNEL_ACCOUNT);
9875 if (!vcpu->arch.mce_banks)
9876 goto fail_free_pio_data;
9877 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9879 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9880 GFP_KERNEL_ACCOUNT))
9881 goto fail_free_mce_banks;
9883 if (!alloc_emulate_ctxt(vcpu))
9884 goto free_wbinvd_dirty_mask;
9886 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
9887 GFP_KERNEL_ACCOUNT);
9888 if (!vcpu->arch.user_fpu) {
9889 pr_err("kvm: failed to allocate userspace's fpu\n");
9890 goto free_emulate_ctxt;
9893 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
9894 GFP_KERNEL_ACCOUNT);
9895 if (!vcpu->arch.guest_fpu) {
9896 pr_err("kvm: failed to allocate vcpu's fpu\n");
9901 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9903 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9905 kvm_async_pf_hash_reset(vcpu);
9908 vcpu->arch.pending_external_vector = -1;
9909 vcpu->arch.preempted_in_kernel = false;
9911 kvm_hv_vcpu_init(vcpu);
9913 r = kvm_x86_ops.vcpu_create(vcpu);
9915 goto free_guest_fpu;
9917 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
9918 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
9919 kvm_vcpu_mtrr_init(vcpu);
9921 kvm_vcpu_reset(vcpu, false);
9922 kvm_init_mmu(vcpu, false);
9927 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9929 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9931 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
9932 free_wbinvd_dirty_mask:
9933 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9934 fail_free_mce_banks:
9935 kfree(vcpu->arch.mce_banks);
9937 free_page((unsigned long)vcpu->arch.pio_data);
9939 kvm_free_lapic(vcpu);
9941 kvm_mmu_destroy(vcpu);
9945 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
9947 struct kvm *kvm = vcpu->kvm;
9949 kvm_hv_vcpu_postcreate(vcpu);
9951 if (mutex_lock_killable(&vcpu->mutex))
9954 kvm_synchronize_tsc(vcpu, 0);
9957 /* poll control enabled by default */
9958 vcpu->arch.msr_kvm_poll_control = 1;
9960 mutex_unlock(&vcpu->mutex);
9962 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
9963 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9964 KVMCLOCK_SYNC_PERIOD);
9967 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9969 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
9972 kvm_release_pfn(cache->pfn, cache->dirty, cache);
9974 kvmclock_reset(vcpu);
9976 kvm_x86_ops.vcpu_free(vcpu);
9978 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
9979 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9980 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9981 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9983 kvm_hv_vcpu_uninit(vcpu);
9984 kvm_pmu_destroy(vcpu);
9985 kfree(vcpu->arch.mce_banks);
9986 kvm_free_lapic(vcpu);
9987 idx = srcu_read_lock(&vcpu->kvm->srcu);
9988 kvm_mmu_destroy(vcpu);
9989 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9990 free_page((unsigned long)vcpu->arch.pio_data);
9991 kvfree(vcpu->arch.cpuid_entries);
9992 if (!lapic_in_kernel(vcpu))
9993 static_key_slow_dec(&kvm_no_apic_vcpu);
9996 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
9998 kvm_lapic_reset(vcpu, init_event);
10000 vcpu->arch.hflags = 0;
10002 vcpu->arch.smi_pending = 0;
10003 vcpu->arch.smi_count = 0;
10004 atomic_set(&vcpu->arch.nmi_queued, 0);
10005 vcpu->arch.nmi_pending = 0;
10006 vcpu->arch.nmi_injected = false;
10007 kvm_clear_interrupt_queue(vcpu);
10008 kvm_clear_exception_queue(vcpu);
10010 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10011 kvm_update_dr0123(vcpu);
10012 vcpu->arch.dr6 = DR6_INIT;
10013 vcpu->arch.dr7 = DR7_FIXED_1;
10014 kvm_update_dr7(vcpu);
10016 vcpu->arch.cr2 = 0;
10018 kvm_make_request(KVM_REQ_EVENT, vcpu);
10019 vcpu->arch.apf.msr_en_val = 0;
10020 vcpu->arch.apf.msr_int_val = 0;
10021 vcpu->arch.st.msr_val = 0;
10023 kvmclock_reset(vcpu);
10025 kvm_clear_async_pf_completion_queue(vcpu);
10026 kvm_async_pf_hash_reset(vcpu);
10027 vcpu->arch.apf.halted = false;
10029 if (kvm_mpx_supported()) {
10030 void *mpx_state_buffer;
10033 * To avoid have the INIT path from kvm_apic_has_events() that be
10034 * called with loaded FPU and does not let userspace fix the state.
10037 kvm_put_guest_fpu(vcpu);
10038 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10040 if (mpx_state_buffer)
10041 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
10042 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10044 if (mpx_state_buffer)
10045 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
10047 kvm_load_guest_fpu(vcpu);
10051 kvm_pmu_reset(vcpu);
10052 vcpu->arch.smbase = 0x30000;
10054 vcpu->arch.msr_misc_features_enables = 0;
10056 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10059 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10060 vcpu->arch.regs_avail = ~0;
10061 vcpu->arch.regs_dirty = ~0;
10063 vcpu->arch.ia32_xss = 0;
10065 kvm_x86_ops.vcpu_reset(vcpu, init_event);
10068 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10070 struct kvm_segment cs;
10072 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10073 cs.selector = vector << 8;
10074 cs.base = vector << 12;
10075 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10076 kvm_rip_write(vcpu, 0);
10079 int kvm_arch_hardware_enable(void)
10082 struct kvm_vcpu *vcpu;
10087 bool stable, backwards_tsc = false;
10089 kvm_user_return_msr_cpu_online();
10090 ret = kvm_x86_ops.hardware_enable();
10094 local_tsc = rdtsc();
10095 stable = !kvm_check_tsc_unstable();
10096 list_for_each_entry(kvm, &vm_list, vm_list) {
10097 kvm_for_each_vcpu(i, vcpu, kvm) {
10098 if (!stable && vcpu->cpu == smp_processor_id())
10099 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10100 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10101 backwards_tsc = true;
10102 if (vcpu->arch.last_host_tsc > max_tsc)
10103 max_tsc = vcpu->arch.last_host_tsc;
10109 * Sometimes, even reliable TSCs go backwards. This happens on
10110 * platforms that reset TSC during suspend or hibernate actions, but
10111 * maintain synchronization. We must compensate. Fortunately, we can
10112 * detect that condition here, which happens early in CPU bringup,
10113 * before any KVM threads can be running. Unfortunately, we can't
10114 * bring the TSCs fully up to date with real time, as we aren't yet far
10115 * enough into CPU bringup that we know how much real time has actually
10116 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10117 * variables that haven't been updated yet.
10119 * So we simply find the maximum observed TSC above, then record the
10120 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
10121 * the adjustment will be applied. Note that we accumulate
10122 * adjustments, in case multiple suspend cycles happen before some VCPU
10123 * gets a chance to run again. In the event that no KVM threads get a
10124 * chance to run, we will miss the entire elapsed period, as we'll have
10125 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10126 * loose cycle time. This isn't too big a deal, since the loss will be
10127 * uniform across all VCPUs (not to mention the scenario is extremely
10128 * unlikely). It is possible that a second hibernate recovery happens
10129 * much faster than a first, causing the observed TSC here to be
10130 * smaller; this would require additional padding adjustment, which is
10131 * why we set last_host_tsc to the local tsc observed here.
10133 * N.B. - this code below runs only on platforms with reliable TSC,
10134 * as that is the only way backwards_tsc is set above. Also note
10135 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10136 * have the same delta_cyc adjustment applied if backwards_tsc
10137 * is detected. Note further, this adjustment is only done once,
10138 * as we reset last_host_tsc on all VCPUs to stop this from being
10139 * called multiple times (one for each physical CPU bringup).
10141 * Platforms with unreliable TSCs don't have to deal with this, they
10142 * will be compensated by the logic in vcpu_load, which sets the TSC to
10143 * catchup mode. This will catchup all VCPUs to real time, but cannot
10144 * guarantee that they stay in perfect synchronization.
10146 if (backwards_tsc) {
10147 u64 delta_cyc = max_tsc - local_tsc;
10148 list_for_each_entry(kvm, &vm_list, vm_list) {
10149 kvm->arch.backwards_tsc_observed = true;
10150 kvm_for_each_vcpu(i, vcpu, kvm) {
10151 vcpu->arch.tsc_offset_adjustment += delta_cyc;
10152 vcpu->arch.last_host_tsc = local_tsc;
10153 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
10157 * We have to disable TSC offset matching.. if you were
10158 * booting a VM while issuing an S4 host suspend....
10159 * you may have some problem. Solving this issue is
10160 * left as an exercise to the reader.
10162 kvm->arch.last_tsc_nsec = 0;
10163 kvm->arch.last_tsc_write = 0;
10170 void kvm_arch_hardware_disable(void)
10172 kvm_x86_ops.hardware_disable();
10173 drop_user_return_notifiers();
10176 int kvm_arch_hardware_setup(void *opaque)
10178 struct kvm_x86_init_ops *ops = opaque;
10181 rdmsrl_safe(MSR_EFER, &host_efer);
10183 if (boot_cpu_has(X86_FEATURE_XSAVES))
10184 rdmsrl(MSR_IA32_XSS, host_xss);
10186 r = ops->hardware_setup();
10190 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
10192 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
10195 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
10196 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
10197 #undef __kvm_cpu_cap_has
10199 if (kvm_has_tsc_control) {
10201 * Make sure the user can only configure tsc_khz values that
10202 * fit into a signed integer.
10203 * A min value is not calculated because it will always
10204 * be 1 on all machines.
10206 u64 max = min(0x7fffffffULL,
10207 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
10208 kvm_max_guest_tsc_khz = max;
10210 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
10213 kvm_init_msr_list();
10217 void kvm_arch_hardware_unsetup(void)
10219 kvm_x86_ops.hardware_unsetup();
10222 int kvm_arch_check_processor_compat(void *opaque)
10224 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
10225 struct kvm_x86_init_ops *ops = opaque;
10227 WARN_ON(!irqs_disabled());
10229 if (__cr4_reserved_bits(cpu_has, c) !=
10230 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
10233 return ops->check_processor_compatibility();
10236 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
10238 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
10240 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
10242 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
10244 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
10247 struct static_key kvm_no_apic_vcpu __read_mostly;
10248 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
10250 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
10252 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
10254 vcpu->arch.l1tf_flush_l1d = true;
10255 if (pmu->version && unlikely(pmu->event_count)) {
10256 pmu->need_cleanup = true;
10257 kvm_make_request(KVM_REQ_PMU, vcpu);
10259 kvm_x86_ops.sched_in(vcpu, cpu);
10262 void kvm_arch_free_vm(struct kvm *kvm)
10264 kfree(kvm->arch.hyperv.hv_pa_pg);
10269 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
10274 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
10275 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
10276 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
10277 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
10278 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
10279 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
10281 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
10282 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
10283 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
10284 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
10285 &kvm->arch.irq_sources_bitmap);
10287 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
10288 mutex_init(&kvm->arch.apic_map_lock);
10289 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
10291 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
10292 pvclock_update_vm_gtod_copy(kvm);
10294 kvm->arch.guest_can_read_msr_platform_info = true;
10296 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
10297 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
10299 kvm_hv_init_vm(kvm);
10300 kvm_page_track_init(kvm);
10301 kvm_mmu_init_vm(kvm);
10303 return kvm_x86_ops.vm_init(kvm);
10306 int kvm_arch_post_init_vm(struct kvm *kvm)
10308 return kvm_mmu_post_init_vm(kvm);
10311 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
10314 kvm_mmu_unload(vcpu);
10318 static void kvm_free_vcpus(struct kvm *kvm)
10321 struct kvm_vcpu *vcpu;
10324 * Unpin any mmu pages first.
10326 kvm_for_each_vcpu(i, vcpu, kvm) {
10327 kvm_clear_async_pf_completion_queue(vcpu);
10328 kvm_unload_vcpu_mmu(vcpu);
10330 kvm_for_each_vcpu(i, vcpu, kvm)
10331 kvm_vcpu_destroy(vcpu);
10333 mutex_lock(&kvm->lock);
10334 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
10335 kvm->vcpus[i] = NULL;
10337 atomic_set(&kvm->online_vcpus, 0);
10338 mutex_unlock(&kvm->lock);
10341 void kvm_arch_sync_events(struct kvm *kvm)
10343 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
10344 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
10348 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
10351 unsigned long hva, old_npages;
10352 struct kvm_memslots *slots = kvm_memslots(kvm);
10353 struct kvm_memory_slot *slot;
10355 /* Called with kvm->slots_lock held. */
10356 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
10359 slot = id_to_memslot(slots, id);
10361 if (slot && slot->npages)
10365 * MAP_SHARED to prevent internal slot pages from being moved
10368 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
10369 MAP_SHARED | MAP_ANONYMOUS, 0);
10370 if (IS_ERR((void *)hva))
10371 return PTR_ERR((void *)hva);
10373 if (!slot || !slot->npages)
10376 old_npages = slot->npages;
10380 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
10381 struct kvm_userspace_memory_region m;
10383 m.slot = id | (i << 16);
10385 m.guest_phys_addr = gpa;
10386 m.userspace_addr = hva;
10387 m.memory_size = size;
10388 r = __kvm_set_memory_region(kvm, &m);
10394 vm_munmap(hva, old_npages * PAGE_SIZE);
10398 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
10400 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
10402 kvm_mmu_pre_destroy_vm(kvm);
10405 void kvm_arch_destroy_vm(struct kvm *kvm)
10409 if (current->mm == kvm->mm) {
10411 * Free memory regions allocated on behalf of userspace,
10412 * unless the the memory map has changed due to process exit
10415 mutex_lock(&kvm->slots_lock);
10416 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
10418 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
10420 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
10421 mutex_unlock(&kvm->slots_lock);
10423 if (kvm_x86_ops.vm_destroy)
10424 kvm_x86_ops.vm_destroy(kvm);
10425 for (i = 0; i < kvm->arch.msr_filter.count; i++)
10426 kfree(kvm->arch.msr_filter.ranges[i].bitmap);
10427 kvm_pic_destroy(kvm);
10428 kvm_ioapic_destroy(kvm);
10429 kvm_free_vcpus(kvm);
10430 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
10431 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
10432 kvm_mmu_uninit_vm(kvm);
10433 kvm_page_track_cleanup(kvm);
10434 kvm_hv_destroy_vm(kvm);
10437 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
10441 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10442 kvfree(slot->arch.rmap[i]);
10443 slot->arch.rmap[i] = NULL;
10448 kvfree(slot->arch.lpage_info[i - 1]);
10449 slot->arch.lpage_info[i - 1] = NULL;
10452 kvm_page_track_free_memslot(slot);
10455 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
10456 unsigned long npages)
10461 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
10462 * old arrays will be freed by __kvm_set_memory_region() if installing
10463 * the new memslot is successful.
10465 memset(&slot->arch, 0, sizeof(slot->arch));
10467 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10468 struct kvm_lpage_info *linfo;
10469 unsigned long ugfn;
10473 lpages = gfn_to_index(slot->base_gfn + npages - 1,
10474 slot->base_gfn, level) + 1;
10476 slot->arch.rmap[i] =
10477 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
10478 GFP_KERNEL_ACCOUNT);
10479 if (!slot->arch.rmap[i])
10484 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
10488 slot->arch.lpage_info[i - 1] = linfo;
10490 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
10491 linfo[0].disallow_lpage = 1;
10492 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
10493 linfo[lpages - 1].disallow_lpage = 1;
10494 ugfn = slot->userspace_addr >> PAGE_SHIFT;
10496 * If the gfn and userspace address are not aligned wrt each
10497 * other, disable large page support for this slot.
10499 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
10502 for (j = 0; j < lpages; ++j)
10503 linfo[j].disallow_lpage = 1;
10507 if (kvm_page_track_create_memslot(slot, npages))
10513 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10514 kvfree(slot->arch.rmap[i]);
10515 slot->arch.rmap[i] = NULL;
10519 kvfree(slot->arch.lpage_info[i - 1]);
10520 slot->arch.lpage_info[i - 1] = NULL;
10525 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
10527 struct kvm_vcpu *vcpu;
10531 * memslots->generation has been incremented.
10532 * mmio generation may have reached its maximum value.
10534 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
10536 /* Force re-initialization of steal_time cache */
10537 kvm_for_each_vcpu(i, vcpu, kvm)
10538 kvm_vcpu_kick(vcpu);
10541 int kvm_arch_prepare_memory_region(struct kvm *kvm,
10542 struct kvm_memory_slot *memslot,
10543 const struct kvm_userspace_memory_region *mem,
10544 enum kvm_mr_change change)
10546 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
10547 return kvm_alloc_memslot_metadata(memslot,
10548 mem->memory_size >> PAGE_SHIFT);
10552 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
10553 struct kvm_memory_slot *old,
10554 struct kvm_memory_slot *new,
10555 enum kvm_mr_change change)
10558 * Nothing to do for RO slots or CREATE/MOVE/DELETE of a slot.
10559 * See comments below.
10561 if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
10565 * Dirty logging tracks sptes in 4k granularity, meaning that large
10566 * sptes have to be split. If live migration is successful, the guest
10567 * in the source machine will be destroyed and large sptes will be
10568 * created in the destination. However, if the guest continues to run
10569 * in the source machine (for example if live migration fails), small
10570 * sptes will remain around and cause bad performance.
10572 * Scan sptes if dirty logging has been stopped, dropping those
10573 * which can be collapsed into a single large-page spte. Later
10574 * page faults will create the large-page sptes.
10576 * There is no need to do this in any of the following cases:
10577 * CREATE: No dirty mappings will already exist.
10578 * MOVE/DELETE: The old mappings will already have been cleaned up by
10579 * kvm_arch_flush_shadow_memslot()
10581 if ((old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
10582 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
10583 kvm_mmu_zap_collapsible_sptes(kvm, new);
10586 * Enable or disable dirty logging for the slot.
10588 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of the old
10589 * slot have been zapped so no dirty logging updates are needed for
10591 * For KVM_MR_CREATE and KVM_MR_MOVE, once the new slot is visible
10592 * any mappings that might be created in it will consume the
10593 * properties of the new slot and do not need to be updated here.
10595 * When PML is enabled, the kvm_x86_ops dirty logging hooks are
10596 * called to enable/disable dirty logging.
10598 * When disabling dirty logging with PML enabled, the D-bit is set
10599 * for sptes in the slot in order to prevent unnecessary GPA
10600 * logging in the PML buffer (and potential PML buffer full VMEXIT).
10601 * This guarantees leaving PML enabled for the guest's lifetime
10602 * won't have any additional overhead from PML when the guest is
10603 * running with dirty logging disabled.
10605 * When enabling dirty logging, large sptes are write-protected
10606 * so they can be split on first write. New large sptes cannot
10607 * be created for this slot until the end of the logging.
10608 * See the comments in fast_page_fault().
10609 * For small sptes, nothing is done if the dirty log is in the
10610 * initial-all-set state. Otherwise, depending on whether pml
10611 * is enabled the D-bit or the W-bit will be cleared.
10613 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
10614 if (kvm_x86_ops.slot_enable_log_dirty) {
10615 kvm_x86_ops.slot_enable_log_dirty(kvm, new);
10618 kvm_dirty_log_manual_protect_and_init_set(kvm) ?
10619 PG_LEVEL_2M : PG_LEVEL_4K;
10622 * If we're with initial-all-set, we don't need
10623 * to write protect any small page because
10624 * they're reported as dirty already. However
10625 * we still need to write-protect huge pages
10626 * so that the page split can happen lazily on
10627 * the first write to the huge page.
10629 kvm_mmu_slot_remove_write_access(kvm, new, level);
10632 if (kvm_x86_ops.slot_disable_log_dirty)
10633 kvm_x86_ops.slot_disable_log_dirty(kvm, new);
10637 void kvm_arch_commit_memory_region(struct kvm *kvm,
10638 const struct kvm_userspace_memory_region *mem,
10639 struct kvm_memory_slot *old,
10640 const struct kvm_memory_slot *new,
10641 enum kvm_mr_change change)
10643 if (!kvm->arch.n_requested_mmu_pages)
10644 kvm_mmu_change_mmu_pages(kvm,
10645 kvm_mmu_calculate_default_mmu_pages(kvm));
10648 * FIXME: const-ify all uses of struct kvm_memory_slot.
10650 kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change);
10652 /* Free the arrays associated with the old memslot. */
10653 if (change == KVM_MR_MOVE)
10654 kvm_arch_free_memslot(kvm, old);
10657 void kvm_arch_flush_shadow_all(struct kvm *kvm)
10659 kvm_mmu_zap_all(kvm);
10662 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
10663 struct kvm_memory_slot *slot)
10665 kvm_page_track_flush_slot(kvm, slot);
10668 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
10670 return (is_guest_mode(vcpu) &&
10671 kvm_x86_ops.guest_apic_has_interrupt &&
10672 kvm_x86_ops.guest_apic_has_interrupt(vcpu));
10675 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
10677 if (!list_empty_careful(&vcpu->async_pf.done))
10680 if (kvm_apic_has_events(vcpu))
10683 if (vcpu->arch.pv.pv_unhalted)
10686 if (vcpu->arch.exception.pending)
10689 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10690 (vcpu->arch.nmi_pending &&
10691 kvm_x86_ops.nmi_allowed(vcpu, false)))
10694 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
10695 (vcpu->arch.smi_pending &&
10696 kvm_x86_ops.smi_allowed(vcpu, false)))
10699 if (kvm_arch_interrupt_allowed(vcpu) &&
10700 (kvm_cpu_has_interrupt(vcpu) ||
10701 kvm_guest_apic_has_interrupt(vcpu)))
10704 if (kvm_hv_has_stimer_pending(vcpu))
10707 if (is_guest_mode(vcpu) &&
10708 kvm_x86_ops.nested_ops->hv_timer_pending &&
10709 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
10715 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
10717 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
10720 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
10722 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
10725 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10726 kvm_test_request(KVM_REQ_SMI, vcpu) ||
10727 kvm_test_request(KVM_REQ_EVENT, vcpu))
10730 if (vcpu->arch.apicv_active && kvm_x86_ops.dy_apicv_has_pending_interrupt(vcpu))
10736 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
10738 return vcpu->arch.preempted_in_kernel;
10741 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
10743 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
10746 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
10748 return kvm_x86_ops.interrupt_allowed(vcpu, false);
10751 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
10753 if (is_64_bit_mode(vcpu))
10754 return kvm_rip_read(vcpu);
10755 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
10756 kvm_rip_read(vcpu));
10758 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
10760 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
10762 return kvm_get_linear_rip(vcpu) == linear_rip;
10764 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
10766 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
10768 unsigned long rflags;
10770 rflags = kvm_x86_ops.get_rflags(vcpu);
10771 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10772 rflags &= ~X86_EFLAGS_TF;
10775 EXPORT_SYMBOL_GPL(kvm_get_rflags);
10777 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10779 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
10780 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
10781 rflags |= X86_EFLAGS_TF;
10782 kvm_x86_ops.set_rflags(vcpu, rflags);
10785 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10787 __kvm_set_rflags(vcpu, rflags);
10788 kvm_make_request(KVM_REQ_EVENT, vcpu);
10790 EXPORT_SYMBOL_GPL(kvm_set_rflags);
10792 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
10796 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
10800 r = kvm_mmu_reload(vcpu);
10804 if (!vcpu->arch.mmu->direct_map &&
10805 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
10808 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
10811 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
10813 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
10815 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
10818 static inline u32 kvm_async_pf_next_probe(u32 key)
10820 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
10823 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10825 u32 key = kvm_async_pf_hash_fn(gfn);
10827 while (vcpu->arch.apf.gfns[key] != ~0)
10828 key = kvm_async_pf_next_probe(key);
10830 vcpu->arch.apf.gfns[key] = gfn;
10833 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
10836 u32 key = kvm_async_pf_hash_fn(gfn);
10838 for (i = 0; i < ASYNC_PF_PER_VCPU &&
10839 (vcpu->arch.apf.gfns[key] != gfn &&
10840 vcpu->arch.apf.gfns[key] != ~0); i++)
10841 key = kvm_async_pf_next_probe(key);
10846 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10848 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
10851 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10855 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
10857 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
10861 vcpu->arch.apf.gfns[i] = ~0;
10863 j = kvm_async_pf_next_probe(j);
10864 if (vcpu->arch.apf.gfns[j] == ~0)
10866 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
10868 * k lies cyclically in ]i,j]
10870 * |....j i.k.| or |.k..j i...|
10872 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
10873 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
10878 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
10880 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
10882 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
10886 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
10888 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
10890 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
10891 &token, offset, sizeof(token));
10894 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
10896 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
10899 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
10900 &val, offset, sizeof(val)))
10906 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
10908 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
10911 if (!kvm_pv_async_pf_enabled(vcpu) ||
10912 (vcpu->arch.apf.send_user_only && kvm_x86_ops.get_cpl(vcpu) == 0))
10918 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
10920 if (unlikely(!lapic_in_kernel(vcpu) ||
10921 kvm_event_needs_reinjection(vcpu) ||
10922 vcpu->arch.exception.pending))
10925 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
10929 * If interrupts are off we cannot even use an artificial
10932 return kvm_arch_interrupt_allowed(vcpu);
10935 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
10936 struct kvm_async_pf *work)
10938 struct x86_exception fault;
10940 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
10941 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
10943 if (kvm_can_deliver_async_pf(vcpu) &&
10944 !apf_put_user_notpresent(vcpu)) {
10945 fault.vector = PF_VECTOR;
10946 fault.error_code_valid = true;
10947 fault.error_code = 0;
10948 fault.nested_page_fault = false;
10949 fault.address = work->arch.token;
10950 fault.async_page_fault = true;
10951 kvm_inject_page_fault(vcpu, &fault);
10955 * It is not possible to deliver a paravirtualized asynchronous
10956 * page fault, but putting the guest in an artificial halt state
10957 * can be beneficial nevertheless: if an interrupt arrives, we
10958 * can deliver it timely and perhaps the guest will schedule
10959 * another process. When the instruction that triggered a page
10960 * fault is retried, hopefully the page will be ready in the host.
10962 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
10967 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
10968 struct kvm_async_pf *work)
10970 struct kvm_lapic_irq irq = {
10971 .delivery_mode = APIC_DM_FIXED,
10972 .vector = vcpu->arch.apf.vec
10975 if (work->wakeup_all)
10976 work->arch.token = ~0; /* broadcast wakeup */
10978 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10979 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
10981 if ((work->wakeup_all || work->notpresent_injected) &&
10982 kvm_pv_async_pf_enabled(vcpu) &&
10983 !apf_put_user_ready(vcpu, work->arch.token)) {
10984 vcpu->arch.apf.pageready_pending = true;
10985 kvm_apic_set_irq(vcpu, &irq, NULL);
10988 vcpu->arch.apf.halted = false;
10989 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10992 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
10994 kvm_make_request(KVM_REQ_APF_READY, vcpu);
10995 if (!vcpu->arch.apf.pageready_pending)
10996 kvm_vcpu_kick(vcpu);
10999 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
11001 if (!kvm_pv_async_pf_enabled(vcpu))
11004 return apf_pageready_slot_free(vcpu);
11007 void kvm_arch_start_assignment(struct kvm *kvm)
11009 atomic_inc(&kvm->arch.assigned_device_count);
11011 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
11013 void kvm_arch_end_assignment(struct kvm *kvm)
11015 atomic_dec(&kvm->arch.assigned_device_count);
11017 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
11019 bool kvm_arch_has_assigned_device(struct kvm *kvm)
11021 return atomic_read(&kvm->arch.assigned_device_count);
11023 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
11025 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
11027 atomic_inc(&kvm->arch.noncoherent_dma_count);
11029 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
11031 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
11033 atomic_dec(&kvm->arch.noncoherent_dma_count);
11035 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
11037 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
11039 return atomic_read(&kvm->arch.noncoherent_dma_count);
11041 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
11043 bool kvm_arch_has_irq_bypass(void)
11048 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
11049 struct irq_bypass_producer *prod)
11051 struct kvm_kernel_irqfd *irqfd =
11052 container_of(cons, struct kvm_kernel_irqfd, consumer);
11055 irqfd->producer = prod;
11056 kvm_arch_start_assignment(irqfd->kvm);
11057 ret = kvm_x86_ops.update_pi_irte(irqfd->kvm,
11058 prod->irq, irqfd->gsi, 1);
11061 kvm_arch_end_assignment(irqfd->kvm);
11066 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
11067 struct irq_bypass_producer *prod)
11070 struct kvm_kernel_irqfd *irqfd =
11071 container_of(cons, struct kvm_kernel_irqfd, consumer);
11073 WARN_ON(irqfd->producer != prod);
11074 irqfd->producer = NULL;
11077 * When producer of consumer is unregistered, we change back to
11078 * remapped mode, so we can re-use the current implementation
11079 * when the irq is masked/disabled or the consumer side (KVM
11080 * int this case doesn't want to receive the interrupts.
11082 ret = kvm_x86_ops.update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
11084 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
11085 " fails: %d\n", irqfd->consumer.token, ret);
11087 kvm_arch_end_assignment(irqfd->kvm);
11090 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
11091 uint32_t guest_irq, bool set)
11093 return kvm_x86_ops.update_pi_irte(kvm, host_irq, guest_irq, set);
11096 bool kvm_vector_hashing_enabled(void)
11098 return vector_hashing;
11101 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
11103 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
11105 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
11108 int kvm_spec_ctrl_test_value(u64 value)
11111 * test that setting IA32_SPEC_CTRL to given value
11112 * is allowed by the host processor
11116 unsigned long flags;
11119 local_irq_save(flags);
11121 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
11123 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
11126 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
11128 local_irq_restore(flags);
11132 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
11134 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
11136 struct x86_exception fault;
11137 u32 access = error_code &
11138 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
11140 if (!(error_code & PFERR_PRESENT_MASK) ||
11141 vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
11143 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
11144 * tables probably do not match the TLB. Just proceed
11145 * with the error code that the processor gave.
11147 fault.vector = PF_VECTOR;
11148 fault.error_code_valid = true;
11149 fault.error_code = error_code;
11150 fault.nested_page_fault = false;
11151 fault.address = gva;
11153 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
11155 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
11158 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
11159 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
11160 * indicates whether exit to userspace is needed.
11162 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
11163 struct x86_exception *e)
11165 if (r == X86EMUL_PROPAGATE_FAULT) {
11166 kvm_inject_emulated_page_fault(vcpu, e);
11171 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
11172 * while handling a VMX instruction KVM could've handled the request
11173 * correctly by exiting to userspace and performing I/O but there
11174 * doesn't seem to be a real use-case behind such requests, just return
11175 * KVM_EXIT_INTERNAL_ERROR for now.
11177 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
11178 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
11179 vcpu->run->internal.ndata = 0;
11183 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
11185 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
11188 struct x86_exception e;
11190 unsigned long roots_to_free = 0;
11197 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
11198 if (r != X86EMUL_CONTINUE)
11199 return kvm_handle_memory_failure(vcpu, r, &e);
11201 if (operand.pcid >> 12 != 0) {
11202 kvm_inject_gp(vcpu, 0);
11206 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
11209 case INVPCID_TYPE_INDIV_ADDR:
11210 if ((!pcid_enabled && (operand.pcid != 0)) ||
11211 is_noncanonical_address(operand.gla, vcpu)) {
11212 kvm_inject_gp(vcpu, 0);
11215 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
11216 return kvm_skip_emulated_instruction(vcpu);
11218 case INVPCID_TYPE_SINGLE_CTXT:
11219 if (!pcid_enabled && (operand.pcid != 0)) {
11220 kvm_inject_gp(vcpu, 0);
11224 if (kvm_get_active_pcid(vcpu) == operand.pcid) {
11225 kvm_mmu_sync_roots(vcpu);
11226 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
11229 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
11230 if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd)
11232 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
11234 kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
11236 * If neither the current cr3 nor any of the prev_roots use the
11237 * given PCID, then nothing needs to be done here because a
11238 * resync will happen anyway before switching to any other CR3.
11241 return kvm_skip_emulated_instruction(vcpu);
11243 case INVPCID_TYPE_ALL_NON_GLOBAL:
11245 * Currently, KVM doesn't mark global entries in the shadow
11246 * page tables, so a non-global flush just degenerates to a
11247 * global flush. If needed, we could optimize this later by
11248 * keeping track of global entries in shadow page tables.
11252 case INVPCID_TYPE_ALL_INCL_GLOBAL:
11253 kvm_mmu_unload(vcpu);
11254 return kvm_skip_emulated_instruction(vcpu);
11257 BUG(); /* We have already checked above that type <= 3 */
11260 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
11262 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
11263 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
11264 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
11265 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
11266 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
11267 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
11268 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
11269 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
11270 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
11271 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
11272 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
11273 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
11274 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
11275 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
11276 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
11277 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
11278 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
11279 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
11280 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
11281 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
11282 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
11283 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);