iwlwifi: mvm: fix version check for GEO_TX_POWER_LIMIT support
[linux/fpc-iii.git] / arch / x86 / kvm / x86.c
blobe10a7a42449b369a3217bee5c44bd52b568ef864
1 /*
2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Authors:
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
14 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
22 #include <linux/kvm_host.h>
23 #include "irq.h"
24 #include "mmu.h"
25 #include "i8254.h"
26 #include "tss.h"
27 #include "kvm_cache_regs.h"
28 #include "x86.h"
29 #include "cpuid.h"
30 #include "pmu.h"
31 #include "hyperv.h"
33 #include <linux/clocksource.h>
34 #include <linux/interrupt.h>
35 #include <linux/kvm.h>
36 #include <linux/fs.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/mem_encrypt.h>
59 #include <trace/events/kvm.h>
61 #include <asm/debugreg.h>
62 #include <asm/msr.h>
63 #include <asm/desc.h>
64 #include <asm/mce.h>
65 #include <linux/kernel_stat.h>
66 #include <asm/fpu/internal.h> /* Ugh! */
67 #include <asm/pvclock.h>
68 #include <asm/div64.h>
69 #include <asm/irq_remapping.h>
70 #include <asm/mshyperv.h>
71 #include <asm/hypervisor.h>
73 #define CREATE_TRACE_POINTS
74 #include "trace.h"
76 #define MAX_IO_MSRS 256
77 #define KVM_MAX_MCE_BANKS 32
78 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
79 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
81 #define emul_to_vcpu(ctxt) \
82 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
84 /* EFER defaults:
85 * - enable syscall per default because its emulated by KVM
86 * - enable LME and LMA per default on 64 bit KVM
88 #ifdef CONFIG_X86_64
89 static
90 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
91 #else
92 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
93 #endif
95 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
96 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
98 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
99 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
101 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
102 static void process_nmi(struct kvm_vcpu *vcpu);
103 static void enter_smm(struct kvm_vcpu *vcpu);
104 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
105 static void store_regs(struct kvm_vcpu *vcpu);
106 static int sync_regs(struct kvm_vcpu *vcpu);
108 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
109 EXPORT_SYMBOL_GPL(kvm_x86_ops);
111 static bool __read_mostly ignore_msrs = 0;
112 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
114 static bool __read_mostly report_ignored_msrs = true;
115 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
117 unsigned int min_timer_period_us = 200;
118 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
120 static bool __read_mostly kvmclock_periodic_sync = true;
121 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
123 bool __read_mostly kvm_has_tsc_control;
124 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
125 u32 __read_mostly kvm_max_guest_tsc_khz;
126 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
127 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
128 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
129 u64 __read_mostly kvm_max_tsc_scaling_ratio;
130 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
131 u64 __read_mostly kvm_default_tsc_scaling_ratio;
132 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
134 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
135 static u32 __read_mostly tsc_tolerance_ppm = 250;
136 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
138 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
139 unsigned int __read_mostly lapic_timer_advance_ns = 0;
140 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
141 EXPORT_SYMBOL_GPL(lapic_timer_advance_ns);
143 static bool __read_mostly vector_hashing = true;
144 module_param(vector_hashing, bool, S_IRUGO);
146 bool __read_mostly enable_vmware_backdoor = false;
147 module_param(enable_vmware_backdoor, bool, S_IRUGO);
148 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
150 static bool __read_mostly force_emulation_prefix = false;
151 module_param(force_emulation_prefix, bool, S_IRUGO);
153 #define KVM_NR_SHARED_MSRS 16
155 struct kvm_shared_msrs_global {
156 int nr;
157 u32 msrs[KVM_NR_SHARED_MSRS];
160 struct kvm_shared_msrs {
161 struct user_return_notifier urn;
162 bool registered;
163 struct kvm_shared_msr_values {
164 u64 host;
165 u64 curr;
166 } values[KVM_NR_SHARED_MSRS];
169 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
170 static struct kvm_shared_msrs __percpu *shared_msrs;
172 struct kvm_stats_debugfs_item debugfs_entries[] = {
173 { "pf_fixed", VCPU_STAT(pf_fixed) },
174 { "pf_guest", VCPU_STAT(pf_guest) },
175 { "tlb_flush", VCPU_STAT(tlb_flush) },
176 { "invlpg", VCPU_STAT(invlpg) },
177 { "exits", VCPU_STAT(exits) },
178 { "io_exits", VCPU_STAT(io_exits) },
179 { "mmio_exits", VCPU_STAT(mmio_exits) },
180 { "signal_exits", VCPU_STAT(signal_exits) },
181 { "irq_window", VCPU_STAT(irq_window_exits) },
182 { "nmi_window", VCPU_STAT(nmi_window_exits) },
183 { "halt_exits", VCPU_STAT(halt_exits) },
184 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
185 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
186 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
187 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
188 { "hypercalls", VCPU_STAT(hypercalls) },
189 { "request_irq", VCPU_STAT(request_irq_exits) },
190 { "irq_exits", VCPU_STAT(irq_exits) },
191 { "host_state_reload", VCPU_STAT(host_state_reload) },
192 { "fpu_reload", VCPU_STAT(fpu_reload) },
193 { "insn_emulation", VCPU_STAT(insn_emulation) },
194 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
195 { "irq_injections", VCPU_STAT(irq_injections) },
196 { "nmi_injections", VCPU_STAT(nmi_injections) },
197 { "req_event", VCPU_STAT(req_event) },
198 { "l1d_flush", VCPU_STAT(l1d_flush) },
199 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
200 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
201 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
202 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
203 { "mmu_flooded", VM_STAT(mmu_flooded) },
204 { "mmu_recycled", VM_STAT(mmu_recycled) },
205 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
206 { "mmu_unsync", VM_STAT(mmu_unsync) },
207 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
208 { "largepages", VM_STAT(lpages) },
209 { "max_mmu_page_hash_collisions",
210 VM_STAT(max_mmu_page_hash_collisions) },
211 { NULL }
214 u64 __read_mostly host_xcr0;
216 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
218 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
220 int i;
221 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
222 vcpu->arch.apf.gfns[i] = ~0;
225 static void kvm_on_user_return(struct user_return_notifier *urn)
227 unsigned slot;
228 struct kvm_shared_msrs *locals
229 = container_of(urn, struct kvm_shared_msrs, urn);
230 struct kvm_shared_msr_values *values;
231 unsigned long flags;
234 * Disabling irqs at this point since the following code could be
235 * interrupted and executed through kvm_arch_hardware_disable()
237 local_irq_save(flags);
238 if (locals->registered) {
239 locals->registered = false;
240 user_return_notifier_unregister(urn);
242 local_irq_restore(flags);
243 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
244 values = &locals->values[slot];
245 if (values->host != values->curr) {
246 wrmsrl(shared_msrs_global.msrs[slot], values->host);
247 values->curr = values->host;
252 static void shared_msr_update(unsigned slot, u32 msr)
254 u64 value;
255 unsigned int cpu = smp_processor_id();
256 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
258 /* only read, and nobody should modify it at this time,
259 * so don't need lock */
260 if (slot >= shared_msrs_global.nr) {
261 printk(KERN_ERR "kvm: invalid MSR slot!");
262 return;
264 rdmsrl_safe(msr, &value);
265 smsr->values[slot].host = value;
266 smsr->values[slot].curr = value;
269 void kvm_define_shared_msr(unsigned slot, u32 msr)
271 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
272 shared_msrs_global.msrs[slot] = msr;
273 if (slot >= shared_msrs_global.nr)
274 shared_msrs_global.nr = slot + 1;
276 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
278 static void kvm_shared_msr_cpu_online(void)
280 unsigned i;
282 for (i = 0; i < shared_msrs_global.nr; ++i)
283 shared_msr_update(i, shared_msrs_global.msrs[i]);
286 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
288 unsigned int cpu = smp_processor_id();
289 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
290 int err;
292 if (((value ^ smsr->values[slot].curr) & mask) == 0)
293 return 0;
294 smsr->values[slot].curr = value;
295 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
296 if (err)
297 return 1;
299 if (!smsr->registered) {
300 smsr->urn.on_user_return = kvm_on_user_return;
301 user_return_notifier_register(&smsr->urn);
302 smsr->registered = true;
304 return 0;
306 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
308 static void drop_user_return_notifiers(void)
310 unsigned int cpu = smp_processor_id();
311 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
313 if (smsr->registered)
314 kvm_on_user_return(&smsr->urn);
317 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
319 return vcpu->arch.apic_base;
321 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
323 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
325 return kvm_apic_mode(kvm_get_apic_base(vcpu));
327 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
329 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
331 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
332 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
333 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
334 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
336 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
337 return 1;
338 if (!msr_info->host_initiated) {
339 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
340 return 1;
341 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
342 return 1;
345 kvm_lapic_set_base(vcpu, msr_info->data);
346 return 0;
348 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
350 asmlinkage __visible void kvm_spurious_fault(void)
352 /* Fault while not rebooting. We want the trace. */
353 BUG();
355 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
357 #define EXCPT_BENIGN 0
358 #define EXCPT_CONTRIBUTORY 1
359 #define EXCPT_PF 2
361 static int exception_class(int vector)
363 switch (vector) {
364 case PF_VECTOR:
365 return EXCPT_PF;
366 case DE_VECTOR:
367 case TS_VECTOR:
368 case NP_VECTOR:
369 case SS_VECTOR:
370 case GP_VECTOR:
371 return EXCPT_CONTRIBUTORY;
372 default:
373 break;
375 return EXCPT_BENIGN;
378 #define EXCPT_FAULT 0
379 #define EXCPT_TRAP 1
380 #define EXCPT_ABORT 2
381 #define EXCPT_INTERRUPT 3
383 static int exception_type(int vector)
385 unsigned int mask;
387 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
388 return EXCPT_INTERRUPT;
390 mask = 1 << vector;
392 /* #DB is trap, as instruction watchpoints are handled elsewhere */
393 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
394 return EXCPT_TRAP;
396 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
397 return EXCPT_ABORT;
399 /* Reserved exceptions will result in fault */
400 return EXCPT_FAULT;
403 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
404 unsigned nr, bool has_error, u32 error_code,
405 bool reinject)
407 u32 prev_nr;
408 int class1, class2;
410 kvm_make_request(KVM_REQ_EVENT, vcpu);
412 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
413 queue:
414 if (has_error && !is_protmode(vcpu))
415 has_error = false;
416 if (reinject) {
418 * On vmentry, vcpu->arch.exception.pending is only
419 * true if an event injection was blocked by
420 * nested_run_pending. In that case, however,
421 * vcpu_enter_guest requests an immediate exit,
422 * and the guest shouldn't proceed far enough to
423 * need reinjection.
425 WARN_ON_ONCE(vcpu->arch.exception.pending);
426 vcpu->arch.exception.injected = true;
427 } else {
428 vcpu->arch.exception.pending = true;
429 vcpu->arch.exception.injected = false;
431 vcpu->arch.exception.has_error_code = has_error;
432 vcpu->arch.exception.nr = nr;
433 vcpu->arch.exception.error_code = error_code;
434 return;
437 /* to check exception */
438 prev_nr = vcpu->arch.exception.nr;
439 if (prev_nr == DF_VECTOR) {
440 /* triple fault -> shutdown */
441 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
442 return;
444 class1 = exception_class(prev_nr);
445 class2 = exception_class(nr);
446 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
447 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
449 * Generate double fault per SDM Table 5-5. Set
450 * exception.pending = true so that the double fault
451 * can trigger a nested vmexit.
453 vcpu->arch.exception.pending = true;
454 vcpu->arch.exception.injected = false;
455 vcpu->arch.exception.has_error_code = true;
456 vcpu->arch.exception.nr = DF_VECTOR;
457 vcpu->arch.exception.error_code = 0;
458 } else
459 /* replace previous exception with a new one in a hope
460 that instruction re-execution will regenerate lost
461 exception */
462 goto queue;
465 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
467 kvm_multiple_exception(vcpu, nr, false, 0, false);
469 EXPORT_SYMBOL_GPL(kvm_queue_exception);
471 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
473 kvm_multiple_exception(vcpu, nr, false, 0, true);
475 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
477 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
479 if (err)
480 kvm_inject_gp(vcpu, 0);
481 else
482 return kvm_skip_emulated_instruction(vcpu);
484 return 1;
486 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
488 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
490 ++vcpu->stat.pf_guest;
491 vcpu->arch.exception.nested_apf =
492 is_guest_mode(vcpu) && fault->async_page_fault;
493 if (vcpu->arch.exception.nested_apf)
494 vcpu->arch.apf.nested_apf_token = fault->address;
495 else
496 vcpu->arch.cr2 = fault->address;
497 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
499 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
501 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
503 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
504 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
505 else
506 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
508 return fault->nested_page_fault;
511 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
513 atomic_inc(&vcpu->arch.nmi_queued);
514 kvm_make_request(KVM_REQ_NMI, vcpu);
516 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
518 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
520 kvm_multiple_exception(vcpu, nr, true, error_code, false);
522 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
524 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
526 kvm_multiple_exception(vcpu, nr, true, error_code, true);
528 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
531 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
532 * a #GP and return false.
534 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
536 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
537 return true;
538 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
539 return false;
541 EXPORT_SYMBOL_GPL(kvm_require_cpl);
543 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
545 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
546 return true;
548 kvm_queue_exception(vcpu, UD_VECTOR);
549 return false;
551 EXPORT_SYMBOL_GPL(kvm_require_dr);
554 * This function will be used to read from the physical memory of the currently
555 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
556 * can read from guest physical or from the guest's guest physical memory.
558 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
559 gfn_t ngfn, void *data, int offset, int len,
560 u32 access)
562 struct x86_exception exception;
563 gfn_t real_gfn;
564 gpa_t ngpa;
566 ngpa = gfn_to_gpa(ngfn);
567 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
568 if (real_gfn == UNMAPPED_GVA)
569 return -EFAULT;
571 real_gfn = gpa_to_gfn(real_gfn);
573 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
575 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
577 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
578 void *data, int offset, int len, u32 access)
580 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
581 data, offset, len, access);
585 * Load the pae pdptrs. Return true is they are all valid.
587 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
589 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
590 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
591 int i;
592 int ret;
593 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
595 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
596 offset * sizeof(u64), sizeof(pdpte),
597 PFERR_USER_MASK|PFERR_WRITE_MASK);
598 if (ret < 0) {
599 ret = 0;
600 goto out;
602 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
603 if ((pdpte[i] & PT_PRESENT_MASK) &&
604 (pdpte[i] &
605 vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
606 ret = 0;
607 goto out;
610 ret = 1;
612 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
613 __set_bit(VCPU_EXREG_PDPTR,
614 (unsigned long *)&vcpu->arch.regs_avail);
615 __set_bit(VCPU_EXREG_PDPTR,
616 (unsigned long *)&vcpu->arch.regs_dirty);
617 out:
619 return ret;
621 EXPORT_SYMBOL_GPL(load_pdptrs);
623 bool pdptrs_changed(struct kvm_vcpu *vcpu)
625 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
626 bool changed = true;
627 int offset;
628 gfn_t gfn;
629 int r;
631 if (is_long_mode(vcpu) || !is_pae(vcpu) || !is_paging(vcpu))
632 return false;
634 if (!test_bit(VCPU_EXREG_PDPTR,
635 (unsigned long *)&vcpu->arch.regs_avail))
636 return true;
638 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
639 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
640 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
641 PFERR_USER_MASK | PFERR_WRITE_MASK);
642 if (r < 0)
643 goto out;
644 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
645 out:
647 return changed;
649 EXPORT_SYMBOL_GPL(pdptrs_changed);
651 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
653 unsigned long old_cr0 = kvm_read_cr0(vcpu);
654 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
656 cr0 |= X86_CR0_ET;
658 #ifdef CONFIG_X86_64
659 if (cr0 & 0xffffffff00000000UL)
660 return 1;
661 #endif
663 cr0 &= ~CR0_RESERVED_BITS;
665 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
666 return 1;
668 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
669 return 1;
671 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
672 #ifdef CONFIG_X86_64
673 if ((vcpu->arch.efer & EFER_LME)) {
674 int cs_db, cs_l;
676 if (!is_pae(vcpu))
677 return 1;
678 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
679 if (cs_l)
680 return 1;
681 } else
682 #endif
683 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
684 kvm_read_cr3(vcpu)))
685 return 1;
688 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
689 return 1;
691 kvm_x86_ops->set_cr0(vcpu, cr0);
693 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
694 kvm_clear_async_pf_completion_queue(vcpu);
695 kvm_async_pf_hash_reset(vcpu);
698 if ((cr0 ^ old_cr0) & update_bits)
699 kvm_mmu_reset_context(vcpu);
701 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
702 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
703 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
704 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
706 return 0;
708 EXPORT_SYMBOL_GPL(kvm_set_cr0);
710 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
712 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
714 EXPORT_SYMBOL_GPL(kvm_lmsw);
716 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
718 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
719 !vcpu->guest_xcr0_loaded) {
720 /* kvm_set_xcr() also depends on this */
721 if (vcpu->arch.xcr0 != host_xcr0)
722 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
723 vcpu->guest_xcr0_loaded = 1;
727 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
729 if (vcpu->guest_xcr0_loaded) {
730 if (vcpu->arch.xcr0 != host_xcr0)
731 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
732 vcpu->guest_xcr0_loaded = 0;
736 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
738 u64 xcr0 = xcr;
739 u64 old_xcr0 = vcpu->arch.xcr0;
740 u64 valid_bits;
742 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
743 if (index != XCR_XFEATURE_ENABLED_MASK)
744 return 1;
745 if (!(xcr0 & XFEATURE_MASK_FP))
746 return 1;
747 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
748 return 1;
751 * Do not allow the guest to set bits that we do not support
752 * saving. However, xcr0 bit 0 is always set, even if the
753 * emulated CPU does not support XSAVE (see fx_init).
755 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
756 if (xcr0 & ~valid_bits)
757 return 1;
759 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
760 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
761 return 1;
763 if (xcr0 & XFEATURE_MASK_AVX512) {
764 if (!(xcr0 & XFEATURE_MASK_YMM))
765 return 1;
766 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
767 return 1;
769 vcpu->arch.xcr0 = xcr0;
771 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
772 kvm_update_cpuid(vcpu);
773 return 0;
776 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
778 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
779 __kvm_set_xcr(vcpu, index, xcr)) {
780 kvm_inject_gp(vcpu, 0);
781 return 1;
783 return 0;
785 EXPORT_SYMBOL_GPL(kvm_set_xcr);
787 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
789 unsigned long old_cr4 = kvm_read_cr4(vcpu);
790 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
791 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
793 if (cr4 & CR4_RESERVED_BITS)
794 return 1;
796 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
797 return 1;
799 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
800 return 1;
802 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
803 return 1;
805 if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
806 return 1;
808 if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
809 return 1;
811 if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
812 return 1;
814 if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
815 return 1;
817 if (is_long_mode(vcpu)) {
818 if (!(cr4 & X86_CR4_PAE))
819 return 1;
820 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
821 && ((cr4 ^ old_cr4) & pdptr_bits)
822 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
823 kvm_read_cr3(vcpu)))
824 return 1;
826 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
827 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
828 return 1;
830 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
831 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
832 return 1;
835 if (kvm_x86_ops->set_cr4(vcpu, cr4))
836 return 1;
838 if (((cr4 ^ old_cr4) & pdptr_bits) ||
839 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
840 kvm_mmu_reset_context(vcpu);
842 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
843 kvm_update_cpuid(vcpu);
845 return 0;
847 EXPORT_SYMBOL_GPL(kvm_set_cr4);
849 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
851 bool skip_tlb_flush = false;
852 #ifdef CONFIG_X86_64
853 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
855 if (pcid_enabled) {
856 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
857 cr3 &= ~X86_CR3_PCID_NOFLUSH;
859 #endif
861 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
862 if (!skip_tlb_flush) {
863 kvm_mmu_sync_roots(vcpu);
864 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
866 return 0;
869 if (is_long_mode(vcpu) &&
870 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
871 return 1;
872 else if (is_pae(vcpu) && is_paging(vcpu) &&
873 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
874 return 1;
876 kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
877 vcpu->arch.cr3 = cr3;
878 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
880 return 0;
882 EXPORT_SYMBOL_GPL(kvm_set_cr3);
884 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
886 if (cr8 & CR8_RESERVED_BITS)
887 return 1;
888 if (lapic_in_kernel(vcpu))
889 kvm_lapic_set_tpr(vcpu, cr8);
890 else
891 vcpu->arch.cr8 = cr8;
892 return 0;
894 EXPORT_SYMBOL_GPL(kvm_set_cr8);
896 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
898 if (lapic_in_kernel(vcpu))
899 return kvm_lapic_get_cr8(vcpu);
900 else
901 return vcpu->arch.cr8;
903 EXPORT_SYMBOL_GPL(kvm_get_cr8);
905 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
907 int i;
909 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
910 for (i = 0; i < KVM_NR_DB_REGS; i++)
911 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
912 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
916 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
918 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
919 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
922 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
924 unsigned long dr7;
926 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
927 dr7 = vcpu->arch.guest_debug_dr7;
928 else
929 dr7 = vcpu->arch.dr7;
930 kvm_x86_ops->set_dr7(vcpu, dr7);
931 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
932 if (dr7 & DR7_BP_EN_MASK)
933 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
936 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
938 u64 fixed = DR6_FIXED_1;
940 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
941 fixed |= DR6_RTM;
942 return fixed;
945 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
947 switch (dr) {
948 case 0 ... 3:
949 vcpu->arch.db[dr] = val;
950 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
951 vcpu->arch.eff_db[dr] = val;
952 break;
953 case 4:
954 /* fall through */
955 case 6:
956 if (val & 0xffffffff00000000ULL)
957 return -1; /* #GP */
958 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
959 kvm_update_dr6(vcpu);
960 break;
961 case 5:
962 /* fall through */
963 default: /* 7 */
964 if (val & 0xffffffff00000000ULL)
965 return -1; /* #GP */
966 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
967 kvm_update_dr7(vcpu);
968 break;
971 return 0;
974 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
976 if (__kvm_set_dr(vcpu, dr, val)) {
977 kvm_inject_gp(vcpu, 0);
978 return 1;
980 return 0;
982 EXPORT_SYMBOL_GPL(kvm_set_dr);
984 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
986 switch (dr) {
987 case 0 ... 3:
988 *val = vcpu->arch.db[dr];
989 break;
990 case 4:
991 /* fall through */
992 case 6:
993 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
994 *val = vcpu->arch.dr6;
995 else
996 *val = kvm_x86_ops->get_dr6(vcpu);
997 break;
998 case 5:
999 /* fall through */
1000 default: /* 7 */
1001 *val = vcpu->arch.dr7;
1002 break;
1004 return 0;
1006 EXPORT_SYMBOL_GPL(kvm_get_dr);
1008 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1010 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
1011 u64 data;
1012 int err;
1014 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1015 if (err)
1016 return err;
1017 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
1018 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
1019 return err;
1021 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1024 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1025 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1027 * This list is modified at module load time to reflect the
1028 * capabilities of the host cpu. This capabilities test skips MSRs that are
1029 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1030 * may depend on host virtualization features rather than host cpu features.
1033 static u32 msrs_to_save[] = {
1034 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1035 MSR_STAR,
1036 #ifdef CONFIG_X86_64
1037 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1038 #endif
1039 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1040 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1041 MSR_IA32_SPEC_CTRL, MSR_IA32_ARCH_CAPABILITIES
1044 static unsigned num_msrs_to_save;
1046 static u32 emulated_msrs[] = {
1047 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1048 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1049 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1050 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1051 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1052 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1053 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1054 HV_X64_MSR_RESET,
1055 HV_X64_MSR_VP_INDEX,
1056 HV_X64_MSR_VP_RUNTIME,
1057 HV_X64_MSR_SCONTROL,
1058 HV_X64_MSR_STIMER0_CONFIG,
1059 HV_X64_MSR_VP_ASSIST_PAGE,
1060 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1061 HV_X64_MSR_TSC_EMULATION_STATUS,
1063 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1064 MSR_KVM_PV_EOI_EN,
1066 MSR_IA32_TSC_ADJUST,
1067 MSR_IA32_TSCDEADLINE,
1068 MSR_IA32_MISC_ENABLE,
1069 MSR_IA32_MCG_STATUS,
1070 MSR_IA32_MCG_CTL,
1071 MSR_IA32_MCG_EXT_CTL,
1072 MSR_IA32_SMBASE,
1073 MSR_SMI_COUNT,
1074 MSR_PLATFORM_INFO,
1075 MSR_MISC_FEATURES_ENABLES,
1076 MSR_AMD64_VIRT_SPEC_CTRL,
1079 static unsigned num_emulated_msrs;
1082 * List of msr numbers which are used to expose MSR-based features that
1083 * can be used by a hypervisor to validate requested CPU features.
1085 static u32 msr_based_features[] = {
1086 MSR_IA32_VMX_BASIC,
1087 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1088 MSR_IA32_VMX_PINBASED_CTLS,
1089 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1090 MSR_IA32_VMX_PROCBASED_CTLS,
1091 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1092 MSR_IA32_VMX_EXIT_CTLS,
1093 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1094 MSR_IA32_VMX_ENTRY_CTLS,
1095 MSR_IA32_VMX_MISC,
1096 MSR_IA32_VMX_CR0_FIXED0,
1097 MSR_IA32_VMX_CR0_FIXED1,
1098 MSR_IA32_VMX_CR4_FIXED0,
1099 MSR_IA32_VMX_CR4_FIXED1,
1100 MSR_IA32_VMX_VMCS_ENUM,
1101 MSR_IA32_VMX_PROCBASED_CTLS2,
1102 MSR_IA32_VMX_EPT_VPID_CAP,
1103 MSR_IA32_VMX_VMFUNC,
1105 MSR_F10H_DECFG,
1106 MSR_IA32_UCODE_REV,
1107 MSR_IA32_ARCH_CAPABILITIES,
1110 static unsigned int num_msr_based_features;
1112 u64 kvm_get_arch_capabilities(void)
1114 u64 data;
1116 rdmsrl_safe(MSR_IA32_ARCH_CAPABILITIES, &data);
1119 * If we're doing cache flushes (either "always" or "cond")
1120 * we will do one whenever the guest does a vmlaunch/vmresume.
1121 * If an outer hypervisor is doing the cache flush for us
1122 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1123 * capability to the guest too, and if EPT is disabled we're not
1124 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1125 * require a nested hypervisor to do a flush of its own.
1127 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1128 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1130 return data;
1132 EXPORT_SYMBOL_GPL(kvm_get_arch_capabilities);
1134 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1136 switch (msr->index) {
1137 case MSR_IA32_ARCH_CAPABILITIES:
1138 msr->data = kvm_get_arch_capabilities();
1139 break;
1140 case MSR_IA32_UCODE_REV:
1141 rdmsrl_safe(msr->index, &msr->data);
1142 break;
1143 default:
1144 if (kvm_x86_ops->get_msr_feature(msr))
1145 return 1;
1147 return 0;
1150 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1152 struct kvm_msr_entry msr;
1153 int r;
1155 msr.index = index;
1156 r = kvm_get_msr_feature(&msr);
1157 if (r)
1158 return r;
1160 *data = msr.data;
1162 return 0;
1165 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1167 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1168 return false;
1170 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1171 return false;
1173 return true;
1176 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1178 if (efer & efer_reserved_bits)
1179 return false;
1181 return __kvm_valid_efer(vcpu, efer);
1183 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1185 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1187 u64 old_efer = vcpu->arch.efer;
1188 u64 efer = msr_info->data;
1190 if (efer & efer_reserved_bits)
1191 return 1;
1193 if (!msr_info->host_initiated) {
1194 if (!__kvm_valid_efer(vcpu, efer))
1195 return 1;
1197 if (is_paging(vcpu) &&
1198 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1199 return 1;
1202 efer &= ~EFER_LMA;
1203 efer |= vcpu->arch.efer & EFER_LMA;
1205 kvm_x86_ops->set_efer(vcpu, efer);
1207 /* Update reserved bits */
1208 if ((efer ^ old_efer) & EFER_NX)
1209 kvm_mmu_reset_context(vcpu);
1211 return 0;
1214 void kvm_enable_efer_bits(u64 mask)
1216 efer_reserved_bits &= ~mask;
1218 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1221 * Writes msr value into into the appropriate "register".
1222 * Returns 0 on success, non-0 otherwise.
1223 * Assumes vcpu_load() was already called.
1225 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1227 switch (msr->index) {
1228 case MSR_FS_BASE:
1229 case MSR_GS_BASE:
1230 case MSR_KERNEL_GS_BASE:
1231 case MSR_CSTAR:
1232 case MSR_LSTAR:
1233 if (is_noncanonical_address(msr->data, vcpu))
1234 return 1;
1235 break;
1236 case MSR_IA32_SYSENTER_EIP:
1237 case MSR_IA32_SYSENTER_ESP:
1239 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1240 * non-canonical address is written on Intel but not on
1241 * AMD (which ignores the top 32-bits, because it does
1242 * not implement 64-bit SYSENTER).
1244 * 64-bit code should hence be able to write a non-canonical
1245 * value on AMD. Making the address canonical ensures that
1246 * vmentry does not fail on Intel after writing a non-canonical
1247 * value, and that something deterministic happens if the guest
1248 * invokes 64-bit SYSENTER.
1250 msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
1252 return kvm_x86_ops->set_msr(vcpu, msr);
1254 EXPORT_SYMBOL_GPL(kvm_set_msr);
1257 * Adapt set_msr() to msr_io()'s calling convention
1259 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1261 struct msr_data msr;
1262 int r;
1264 msr.index = index;
1265 msr.host_initiated = true;
1266 r = kvm_get_msr(vcpu, &msr);
1267 if (r)
1268 return r;
1270 *data = msr.data;
1271 return 0;
1274 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1276 struct msr_data msr;
1278 msr.data = *data;
1279 msr.index = index;
1280 msr.host_initiated = true;
1281 return kvm_set_msr(vcpu, &msr);
1284 #ifdef CONFIG_X86_64
1285 struct pvclock_gtod_data {
1286 seqcount_t seq;
1288 struct { /* extract of a clocksource struct */
1289 int vclock_mode;
1290 u64 cycle_last;
1291 u64 mask;
1292 u32 mult;
1293 u32 shift;
1294 } clock;
1296 u64 boot_ns;
1297 u64 nsec_base;
1298 u64 wall_time_sec;
1301 static struct pvclock_gtod_data pvclock_gtod_data;
1303 static void update_pvclock_gtod(struct timekeeper *tk)
1305 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1306 u64 boot_ns;
1308 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1310 write_seqcount_begin(&vdata->seq);
1312 /* copy pvclock gtod data */
1313 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1314 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1315 vdata->clock.mask = tk->tkr_mono.mask;
1316 vdata->clock.mult = tk->tkr_mono.mult;
1317 vdata->clock.shift = tk->tkr_mono.shift;
1319 vdata->boot_ns = boot_ns;
1320 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1322 vdata->wall_time_sec = tk->xtime_sec;
1324 write_seqcount_end(&vdata->seq);
1326 #endif
1328 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1331 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1332 * vcpu_enter_guest. This function is only called from
1333 * the physical CPU that is running vcpu.
1335 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1338 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1340 int version;
1341 int r;
1342 struct pvclock_wall_clock wc;
1343 struct timespec64 boot;
1345 if (!wall_clock)
1346 return;
1348 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1349 if (r)
1350 return;
1352 if (version & 1)
1353 ++version; /* first time write, random junk */
1355 ++version;
1357 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1358 return;
1361 * The guest calculates current wall clock time by adding
1362 * system time (updated by kvm_guest_time_update below) to the
1363 * wall clock specified here. guest system time equals host
1364 * system time for us, thus we must fill in host boot time here.
1366 getboottime64(&boot);
1368 if (kvm->arch.kvmclock_offset) {
1369 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1370 boot = timespec64_sub(boot, ts);
1372 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1373 wc.nsec = boot.tv_nsec;
1374 wc.version = version;
1376 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1378 version++;
1379 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1382 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1384 do_shl32_div32(dividend, divisor);
1385 return dividend;
1388 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1389 s8 *pshift, u32 *pmultiplier)
1391 uint64_t scaled64;
1392 int32_t shift = 0;
1393 uint64_t tps64;
1394 uint32_t tps32;
1396 tps64 = base_hz;
1397 scaled64 = scaled_hz;
1398 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1399 tps64 >>= 1;
1400 shift--;
1403 tps32 = (uint32_t)tps64;
1404 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1405 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1406 scaled64 >>= 1;
1407 else
1408 tps32 <<= 1;
1409 shift++;
1412 *pshift = shift;
1413 *pmultiplier = div_frac(scaled64, tps32);
1415 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1416 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1419 #ifdef CONFIG_X86_64
1420 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1421 #endif
1423 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1424 static unsigned long max_tsc_khz;
1426 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1428 u64 v = (u64)khz * (1000000 + ppm);
1429 do_div(v, 1000000);
1430 return v;
1433 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1435 u64 ratio;
1437 /* Guest TSC same frequency as host TSC? */
1438 if (!scale) {
1439 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1440 return 0;
1443 /* TSC scaling supported? */
1444 if (!kvm_has_tsc_control) {
1445 if (user_tsc_khz > tsc_khz) {
1446 vcpu->arch.tsc_catchup = 1;
1447 vcpu->arch.tsc_always_catchup = 1;
1448 return 0;
1449 } else {
1450 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
1451 return -1;
1455 /* TSC scaling required - calculate ratio */
1456 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1457 user_tsc_khz, tsc_khz);
1459 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1460 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1461 user_tsc_khz);
1462 return -1;
1465 vcpu->arch.tsc_scaling_ratio = ratio;
1466 return 0;
1469 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1471 u32 thresh_lo, thresh_hi;
1472 int use_scaling = 0;
1474 /* tsc_khz can be zero if TSC calibration fails */
1475 if (user_tsc_khz == 0) {
1476 /* set tsc_scaling_ratio to a safe value */
1477 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1478 return -1;
1481 /* Compute a scale to convert nanoseconds in TSC cycles */
1482 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1483 &vcpu->arch.virtual_tsc_shift,
1484 &vcpu->arch.virtual_tsc_mult);
1485 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1488 * Compute the variation in TSC rate which is acceptable
1489 * within the range of tolerance and decide if the
1490 * rate being applied is within that bounds of the hardware
1491 * rate. If so, no scaling or compensation need be done.
1493 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1494 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1495 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1496 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1497 use_scaling = 1;
1499 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1502 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1504 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1505 vcpu->arch.virtual_tsc_mult,
1506 vcpu->arch.virtual_tsc_shift);
1507 tsc += vcpu->arch.this_tsc_write;
1508 return tsc;
1511 static inline int gtod_is_based_on_tsc(int mode)
1513 return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1516 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1518 #ifdef CONFIG_X86_64
1519 bool vcpus_matched;
1520 struct kvm_arch *ka = &vcpu->kvm->arch;
1521 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1523 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1524 atomic_read(&vcpu->kvm->online_vcpus));
1527 * Once the masterclock is enabled, always perform request in
1528 * order to update it.
1530 * In order to enable masterclock, the host clocksource must be TSC
1531 * and the vcpus need to have matched TSCs. When that happens,
1532 * perform request to enable masterclock.
1534 if (ka->use_master_clock ||
1535 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1536 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1538 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1539 atomic_read(&vcpu->kvm->online_vcpus),
1540 ka->use_master_clock, gtod->clock.vclock_mode);
1541 #endif
1544 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1546 u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1547 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1551 * Multiply tsc by a fixed point number represented by ratio.
1553 * The most significant 64-N bits (mult) of ratio represent the
1554 * integral part of the fixed point number; the remaining N bits
1555 * (frac) represent the fractional part, ie. ratio represents a fixed
1556 * point number (mult + frac * 2^(-N)).
1558 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1560 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1562 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1565 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1567 u64 _tsc = tsc;
1568 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1570 if (ratio != kvm_default_tsc_scaling_ratio)
1571 _tsc = __scale_tsc(ratio, tsc);
1573 return _tsc;
1575 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1577 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1579 u64 tsc;
1581 tsc = kvm_scale_tsc(vcpu, rdtsc());
1583 return target_tsc - tsc;
1586 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1588 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1590 return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1592 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1594 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1596 vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1599 static inline bool kvm_check_tsc_unstable(void)
1601 #ifdef CONFIG_X86_64
1603 * TSC is marked unstable when we're running on Hyper-V,
1604 * 'TSC page' clocksource is good.
1606 if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1607 return false;
1608 #endif
1609 return check_tsc_unstable();
1612 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1614 struct kvm *kvm = vcpu->kvm;
1615 u64 offset, ns, elapsed;
1616 unsigned long flags;
1617 bool matched;
1618 bool already_matched;
1619 u64 data = msr->data;
1620 bool synchronizing = false;
1622 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1623 offset = kvm_compute_tsc_offset(vcpu, data);
1624 ns = ktime_get_boot_ns();
1625 elapsed = ns - kvm->arch.last_tsc_nsec;
1627 if (vcpu->arch.virtual_tsc_khz) {
1628 if (data == 0 && msr->host_initiated) {
1630 * detection of vcpu initialization -- need to sync
1631 * with other vCPUs. This particularly helps to keep
1632 * kvm_clock stable after CPU hotplug
1634 synchronizing = true;
1635 } else {
1636 u64 tsc_exp = kvm->arch.last_tsc_write +
1637 nsec_to_cycles(vcpu, elapsed);
1638 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1640 * Special case: TSC write with a small delta (1 second)
1641 * of virtual cycle time against real time is
1642 * interpreted as an attempt to synchronize the CPU.
1644 synchronizing = data < tsc_exp + tsc_hz &&
1645 data + tsc_hz > tsc_exp;
1650 * For a reliable TSC, we can match TSC offsets, and for an unstable
1651 * TSC, we add elapsed time in this computation. We could let the
1652 * compensation code attempt to catch up if we fall behind, but
1653 * it's better to try to match offsets from the beginning.
1655 if (synchronizing &&
1656 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1657 if (!kvm_check_tsc_unstable()) {
1658 offset = kvm->arch.cur_tsc_offset;
1659 pr_debug("kvm: matched tsc offset for %llu\n", data);
1660 } else {
1661 u64 delta = nsec_to_cycles(vcpu, elapsed);
1662 data += delta;
1663 offset = kvm_compute_tsc_offset(vcpu, data);
1664 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1666 matched = true;
1667 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1668 } else {
1670 * We split periods of matched TSC writes into generations.
1671 * For each generation, we track the original measured
1672 * nanosecond time, offset, and write, so if TSCs are in
1673 * sync, we can match exact offset, and if not, we can match
1674 * exact software computation in compute_guest_tsc()
1676 * These values are tracked in kvm->arch.cur_xxx variables.
1678 kvm->arch.cur_tsc_generation++;
1679 kvm->arch.cur_tsc_nsec = ns;
1680 kvm->arch.cur_tsc_write = data;
1681 kvm->arch.cur_tsc_offset = offset;
1682 matched = false;
1683 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1684 kvm->arch.cur_tsc_generation, data);
1688 * We also track th most recent recorded KHZ, write and time to
1689 * allow the matching interval to be extended at each write.
1691 kvm->arch.last_tsc_nsec = ns;
1692 kvm->arch.last_tsc_write = data;
1693 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1695 vcpu->arch.last_guest_tsc = data;
1697 /* Keep track of which generation this VCPU has synchronized to */
1698 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1699 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1700 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1702 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1703 update_ia32_tsc_adjust_msr(vcpu, offset);
1705 kvm_vcpu_write_tsc_offset(vcpu, offset);
1706 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1708 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1709 if (!matched) {
1710 kvm->arch.nr_vcpus_matched_tsc = 0;
1711 } else if (!already_matched) {
1712 kvm->arch.nr_vcpus_matched_tsc++;
1715 kvm_track_tsc_matching(vcpu);
1716 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1719 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1721 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1722 s64 adjustment)
1724 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1725 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
1728 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1730 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1731 WARN_ON(adjustment < 0);
1732 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1733 adjust_tsc_offset_guest(vcpu, adjustment);
1736 #ifdef CONFIG_X86_64
1738 static u64 read_tsc(void)
1740 u64 ret = (u64)rdtsc_ordered();
1741 u64 last = pvclock_gtod_data.clock.cycle_last;
1743 if (likely(ret >= last))
1744 return ret;
1747 * GCC likes to generate cmov here, but this branch is extremely
1748 * predictable (it's just a function of time and the likely is
1749 * very likely) and there's a data dependence, so force GCC
1750 * to generate a branch instead. I don't barrier() because
1751 * we don't actually need a barrier, and if this function
1752 * ever gets inlined it will generate worse code.
1754 asm volatile ("");
1755 return last;
1758 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
1760 long v;
1761 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1762 u64 tsc_pg_val;
1764 switch (gtod->clock.vclock_mode) {
1765 case VCLOCK_HVCLOCK:
1766 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
1767 tsc_timestamp);
1768 if (tsc_pg_val != U64_MAX) {
1769 /* TSC page valid */
1770 *mode = VCLOCK_HVCLOCK;
1771 v = (tsc_pg_val - gtod->clock.cycle_last) &
1772 gtod->clock.mask;
1773 } else {
1774 /* TSC page invalid */
1775 *mode = VCLOCK_NONE;
1777 break;
1778 case VCLOCK_TSC:
1779 *mode = VCLOCK_TSC;
1780 *tsc_timestamp = read_tsc();
1781 v = (*tsc_timestamp - gtod->clock.cycle_last) &
1782 gtod->clock.mask;
1783 break;
1784 default:
1785 *mode = VCLOCK_NONE;
1788 if (*mode == VCLOCK_NONE)
1789 *tsc_timestamp = v = 0;
1791 return v * gtod->clock.mult;
1794 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
1796 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1797 unsigned long seq;
1798 int mode;
1799 u64 ns;
1801 do {
1802 seq = read_seqcount_begin(&gtod->seq);
1803 ns = gtod->nsec_base;
1804 ns += vgettsc(tsc_timestamp, &mode);
1805 ns >>= gtod->clock.shift;
1806 ns += gtod->boot_ns;
1807 } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1808 *t = ns;
1810 return mode;
1813 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
1815 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1816 unsigned long seq;
1817 int mode;
1818 u64 ns;
1820 do {
1821 seq = read_seqcount_begin(&gtod->seq);
1822 ts->tv_sec = gtod->wall_time_sec;
1823 ns = gtod->nsec_base;
1824 ns += vgettsc(tsc_timestamp, &mode);
1825 ns >>= gtod->clock.shift;
1826 } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1828 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1829 ts->tv_nsec = ns;
1831 return mode;
1834 /* returns true if host is using TSC based clocksource */
1835 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
1837 /* checked again under seqlock below */
1838 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1839 return false;
1841 return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
1842 tsc_timestamp));
1845 /* returns true if host is using TSC based clocksource */
1846 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
1847 u64 *tsc_timestamp)
1849 /* checked again under seqlock below */
1850 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1851 return false;
1853 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
1855 #endif
1859 * Assuming a stable TSC across physical CPUS, and a stable TSC
1860 * across virtual CPUs, the following condition is possible.
1861 * Each numbered line represents an event visible to both
1862 * CPUs at the next numbered event.
1864 * "timespecX" represents host monotonic time. "tscX" represents
1865 * RDTSC value.
1867 * VCPU0 on CPU0 | VCPU1 on CPU1
1869 * 1. read timespec0,tsc0
1870 * 2. | timespec1 = timespec0 + N
1871 * | tsc1 = tsc0 + M
1872 * 3. transition to guest | transition to guest
1873 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1874 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1875 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1877 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1879 * - ret0 < ret1
1880 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1881 * ...
1882 * - 0 < N - M => M < N
1884 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1885 * always the case (the difference between two distinct xtime instances
1886 * might be smaller then the difference between corresponding TSC reads,
1887 * when updating guest vcpus pvclock areas).
1889 * To avoid that problem, do not allow visibility of distinct
1890 * system_timestamp/tsc_timestamp values simultaneously: use a master
1891 * copy of host monotonic time values. Update that master copy
1892 * in lockstep.
1894 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1898 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1900 #ifdef CONFIG_X86_64
1901 struct kvm_arch *ka = &kvm->arch;
1902 int vclock_mode;
1903 bool host_tsc_clocksource, vcpus_matched;
1905 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1906 atomic_read(&kvm->online_vcpus));
1909 * If the host uses TSC clock, then passthrough TSC as stable
1910 * to the guest.
1912 host_tsc_clocksource = kvm_get_time_and_clockread(
1913 &ka->master_kernel_ns,
1914 &ka->master_cycle_now);
1916 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1917 && !ka->backwards_tsc_observed
1918 && !ka->boot_vcpu_runs_old_kvmclock;
1920 if (ka->use_master_clock)
1921 atomic_set(&kvm_guest_has_master_clock, 1);
1923 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1924 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1925 vcpus_matched);
1926 #endif
1929 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1931 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1934 static void kvm_gen_update_masterclock(struct kvm *kvm)
1936 #ifdef CONFIG_X86_64
1937 int i;
1938 struct kvm_vcpu *vcpu;
1939 struct kvm_arch *ka = &kvm->arch;
1941 spin_lock(&ka->pvclock_gtod_sync_lock);
1942 kvm_make_mclock_inprogress_request(kvm);
1943 /* no guest entries from this point */
1944 pvclock_update_vm_gtod_copy(kvm);
1946 kvm_for_each_vcpu(i, vcpu, kvm)
1947 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1949 /* guest entries allowed */
1950 kvm_for_each_vcpu(i, vcpu, kvm)
1951 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
1953 spin_unlock(&ka->pvclock_gtod_sync_lock);
1954 #endif
1957 u64 get_kvmclock_ns(struct kvm *kvm)
1959 struct kvm_arch *ka = &kvm->arch;
1960 struct pvclock_vcpu_time_info hv_clock;
1961 u64 ret;
1963 spin_lock(&ka->pvclock_gtod_sync_lock);
1964 if (!ka->use_master_clock) {
1965 spin_unlock(&ka->pvclock_gtod_sync_lock);
1966 return ktime_get_boot_ns() + ka->kvmclock_offset;
1969 hv_clock.tsc_timestamp = ka->master_cycle_now;
1970 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
1971 spin_unlock(&ka->pvclock_gtod_sync_lock);
1973 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
1974 get_cpu();
1976 if (__this_cpu_read(cpu_tsc_khz)) {
1977 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
1978 &hv_clock.tsc_shift,
1979 &hv_clock.tsc_to_system_mul);
1980 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
1981 } else
1982 ret = ktime_get_boot_ns() + ka->kvmclock_offset;
1984 put_cpu();
1986 return ret;
1989 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
1991 struct kvm_vcpu_arch *vcpu = &v->arch;
1992 struct pvclock_vcpu_time_info guest_hv_clock;
1994 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1995 &guest_hv_clock, sizeof(guest_hv_clock))))
1996 return;
1998 /* This VCPU is paused, but it's legal for a guest to read another
1999 * VCPU's kvmclock, so we really have to follow the specification where
2000 * it says that version is odd if data is being modified, and even after
2001 * it is consistent.
2003 * Version field updates must be kept separate. This is because
2004 * kvm_write_guest_cached might use a "rep movs" instruction, and
2005 * writes within a string instruction are weakly ordered. So there
2006 * are three writes overall.
2008 * As a small optimization, only write the version field in the first
2009 * and third write. The vcpu->pv_time cache is still valid, because the
2010 * version field is the first in the struct.
2012 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2014 if (guest_hv_clock.version & 1)
2015 ++guest_hv_clock.version; /* first time write, random junk */
2017 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2018 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2019 &vcpu->hv_clock,
2020 sizeof(vcpu->hv_clock.version));
2022 smp_wmb();
2024 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2025 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2027 if (vcpu->pvclock_set_guest_stopped_request) {
2028 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2029 vcpu->pvclock_set_guest_stopped_request = false;
2032 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2034 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2035 &vcpu->hv_clock,
2036 sizeof(vcpu->hv_clock));
2038 smp_wmb();
2040 vcpu->hv_clock.version++;
2041 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2042 &vcpu->hv_clock,
2043 sizeof(vcpu->hv_clock.version));
2046 static int kvm_guest_time_update(struct kvm_vcpu *v)
2048 unsigned long flags, tgt_tsc_khz;
2049 struct kvm_vcpu_arch *vcpu = &v->arch;
2050 struct kvm_arch *ka = &v->kvm->arch;
2051 s64 kernel_ns;
2052 u64 tsc_timestamp, host_tsc;
2053 u8 pvclock_flags;
2054 bool use_master_clock;
2056 kernel_ns = 0;
2057 host_tsc = 0;
2060 * If the host uses TSC clock, then passthrough TSC as stable
2061 * to the guest.
2063 spin_lock(&ka->pvclock_gtod_sync_lock);
2064 use_master_clock = ka->use_master_clock;
2065 if (use_master_clock) {
2066 host_tsc = ka->master_cycle_now;
2067 kernel_ns = ka->master_kernel_ns;
2069 spin_unlock(&ka->pvclock_gtod_sync_lock);
2071 /* Keep irq disabled to prevent changes to the clock */
2072 local_irq_save(flags);
2073 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2074 if (unlikely(tgt_tsc_khz == 0)) {
2075 local_irq_restore(flags);
2076 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2077 return 1;
2079 if (!use_master_clock) {
2080 host_tsc = rdtsc();
2081 kernel_ns = ktime_get_boot_ns();
2084 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2087 * We may have to catch up the TSC to match elapsed wall clock
2088 * time for two reasons, even if kvmclock is used.
2089 * 1) CPU could have been running below the maximum TSC rate
2090 * 2) Broken TSC compensation resets the base at each VCPU
2091 * entry to avoid unknown leaps of TSC even when running
2092 * again on the same CPU. This may cause apparent elapsed
2093 * time to disappear, and the guest to stand still or run
2094 * very slowly.
2096 if (vcpu->tsc_catchup) {
2097 u64 tsc = compute_guest_tsc(v, kernel_ns);
2098 if (tsc > tsc_timestamp) {
2099 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2100 tsc_timestamp = tsc;
2104 local_irq_restore(flags);
2106 /* With all the info we got, fill in the values */
2108 if (kvm_has_tsc_control)
2109 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2111 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2112 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2113 &vcpu->hv_clock.tsc_shift,
2114 &vcpu->hv_clock.tsc_to_system_mul);
2115 vcpu->hw_tsc_khz = tgt_tsc_khz;
2118 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2119 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2120 vcpu->last_guest_tsc = tsc_timestamp;
2122 /* If the host uses TSC clocksource, then it is stable */
2123 pvclock_flags = 0;
2124 if (use_master_clock)
2125 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2127 vcpu->hv_clock.flags = pvclock_flags;
2129 if (vcpu->pv_time_enabled)
2130 kvm_setup_pvclock_page(v);
2131 if (v == kvm_get_vcpu(v->kvm, 0))
2132 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2133 return 0;
2137 * kvmclock updates which are isolated to a given vcpu, such as
2138 * vcpu->cpu migration, should not allow system_timestamp from
2139 * the rest of the vcpus to remain static. Otherwise ntp frequency
2140 * correction applies to one vcpu's system_timestamp but not
2141 * the others.
2143 * So in those cases, request a kvmclock update for all vcpus.
2144 * We need to rate-limit these requests though, as they can
2145 * considerably slow guests that have a large number of vcpus.
2146 * The time for a remote vcpu to update its kvmclock is bound
2147 * by the delay we use to rate-limit the updates.
2150 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2152 static void kvmclock_update_fn(struct work_struct *work)
2154 int i;
2155 struct delayed_work *dwork = to_delayed_work(work);
2156 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2157 kvmclock_update_work);
2158 struct kvm *kvm = container_of(ka, struct kvm, arch);
2159 struct kvm_vcpu *vcpu;
2161 kvm_for_each_vcpu(i, vcpu, kvm) {
2162 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2163 kvm_vcpu_kick(vcpu);
2167 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2169 struct kvm *kvm = v->kvm;
2171 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2172 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2173 KVMCLOCK_UPDATE_DELAY);
2176 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2178 static void kvmclock_sync_fn(struct work_struct *work)
2180 struct delayed_work *dwork = to_delayed_work(work);
2181 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2182 kvmclock_sync_work);
2183 struct kvm *kvm = container_of(ka, struct kvm, arch);
2185 if (!kvmclock_periodic_sync)
2186 return;
2188 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2189 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2190 KVMCLOCK_SYNC_PERIOD);
2193 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2195 u64 mcg_cap = vcpu->arch.mcg_cap;
2196 unsigned bank_num = mcg_cap & 0xff;
2197 u32 msr = msr_info->index;
2198 u64 data = msr_info->data;
2200 switch (msr) {
2201 case MSR_IA32_MCG_STATUS:
2202 vcpu->arch.mcg_status = data;
2203 break;
2204 case MSR_IA32_MCG_CTL:
2205 if (!(mcg_cap & MCG_CTL_P) &&
2206 (data || !msr_info->host_initiated))
2207 return 1;
2208 if (data != 0 && data != ~(u64)0)
2209 return 1;
2210 vcpu->arch.mcg_ctl = data;
2211 break;
2212 default:
2213 if (msr >= MSR_IA32_MC0_CTL &&
2214 msr < MSR_IA32_MCx_CTL(bank_num)) {
2215 u32 offset = msr - MSR_IA32_MC0_CTL;
2216 /* only 0 or all 1s can be written to IA32_MCi_CTL
2217 * some Linux kernels though clear bit 10 in bank 4 to
2218 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2219 * this to avoid an uncatched #GP in the guest
2221 if ((offset & 0x3) == 0 &&
2222 data != 0 && (data | (1 << 10)) != ~(u64)0)
2223 return -1;
2224 if (!msr_info->host_initiated &&
2225 (offset & 0x3) == 1 && data != 0)
2226 return -1;
2227 vcpu->arch.mce_banks[offset] = data;
2228 break;
2230 return 1;
2232 return 0;
2235 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2237 struct kvm *kvm = vcpu->kvm;
2238 int lm = is_long_mode(vcpu);
2239 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2240 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2241 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2242 : kvm->arch.xen_hvm_config.blob_size_32;
2243 u32 page_num = data & ~PAGE_MASK;
2244 u64 page_addr = data & PAGE_MASK;
2245 u8 *page;
2246 int r;
2248 r = -E2BIG;
2249 if (page_num >= blob_size)
2250 goto out;
2251 r = -ENOMEM;
2252 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2253 if (IS_ERR(page)) {
2254 r = PTR_ERR(page);
2255 goto out;
2257 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2258 goto out_free;
2259 r = 0;
2260 out_free:
2261 kfree(page);
2262 out:
2263 return r;
2266 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2268 gpa_t gpa = data & ~0x3f;
2270 /* Bits 3:5 are reserved, Should be zero */
2271 if (data & 0x38)
2272 return 1;
2274 vcpu->arch.apf.msr_val = data;
2276 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2277 kvm_clear_async_pf_completion_queue(vcpu);
2278 kvm_async_pf_hash_reset(vcpu);
2279 return 0;
2282 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2283 sizeof(u32)))
2284 return 1;
2286 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2287 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2288 kvm_async_pf_wakeup_all(vcpu);
2289 return 0;
2292 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2294 vcpu->arch.pv_time_enabled = false;
2297 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2299 ++vcpu->stat.tlb_flush;
2300 kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2303 static void record_steal_time(struct kvm_vcpu *vcpu)
2305 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2306 return;
2308 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2309 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2310 return;
2313 * Doing a TLB flush here, on the guest's behalf, can avoid
2314 * expensive IPIs.
2316 if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2317 kvm_vcpu_flush_tlb(vcpu, false);
2319 if (vcpu->arch.st.steal.version & 1)
2320 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2322 vcpu->arch.st.steal.version += 1;
2324 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2325 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2327 smp_wmb();
2329 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2330 vcpu->arch.st.last_steal;
2331 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2333 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2334 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2336 smp_wmb();
2338 vcpu->arch.st.steal.version += 1;
2340 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2341 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2344 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2346 bool pr = false;
2347 u32 msr = msr_info->index;
2348 u64 data = msr_info->data;
2350 switch (msr) {
2351 case MSR_AMD64_NB_CFG:
2352 case MSR_IA32_UCODE_WRITE:
2353 case MSR_VM_HSAVE_PA:
2354 case MSR_AMD64_PATCH_LOADER:
2355 case MSR_AMD64_BU_CFG2:
2356 case MSR_AMD64_DC_CFG:
2357 case MSR_F15H_EX_CFG:
2358 break;
2360 case MSR_IA32_UCODE_REV:
2361 if (msr_info->host_initiated)
2362 vcpu->arch.microcode_version = data;
2363 break;
2364 case MSR_IA32_ARCH_CAPABILITIES:
2365 if (!msr_info->host_initiated)
2366 return 1;
2367 vcpu->arch.arch_capabilities = data;
2368 break;
2369 case MSR_EFER:
2370 return set_efer(vcpu, msr_info);
2371 case MSR_K7_HWCR:
2372 data &= ~(u64)0x40; /* ignore flush filter disable */
2373 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2374 data &= ~(u64)0x8; /* ignore TLB cache disable */
2375 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2376 if (data != 0) {
2377 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2378 data);
2379 return 1;
2381 break;
2382 case MSR_FAM10H_MMIO_CONF_BASE:
2383 if (data != 0) {
2384 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2385 "0x%llx\n", data);
2386 return 1;
2388 break;
2389 case MSR_IA32_DEBUGCTLMSR:
2390 if (!data) {
2391 /* We support the non-activated case already */
2392 break;
2393 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2394 /* Values other than LBR and BTF are vendor-specific,
2395 thus reserved and should throw a #GP */
2396 return 1;
2398 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2399 __func__, data);
2400 break;
2401 case 0x200 ... 0x2ff:
2402 return kvm_mtrr_set_msr(vcpu, msr, data);
2403 case MSR_IA32_APICBASE:
2404 return kvm_set_apic_base(vcpu, msr_info);
2405 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2406 return kvm_x2apic_msr_write(vcpu, msr, data);
2407 case MSR_IA32_TSCDEADLINE:
2408 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2409 break;
2410 case MSR_IA32_TSC_ADJUST:
2411 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2412 if (!msr_info->host_initiated) {
2413 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2414 adjust_tsc_offset_guest(vcpu, adj);
2416 vcpu->arch.ia32_tsc_adjust_msr = data;
2418 break;
2419 case MSR_IA32_MISC_ENABLE:
2420 vcpu->arch.ia32_misc_enable_msr = data;
2421 break;
2422 case MSR_IA32_SMBASE:
2423 if (!msr_info->host_initiated)
2424 return 1;
2425 vcpu->arch.smbase = data;
2426 break;
2427 case MSR_IA32_TSC:
2428 kvm_write_tsc(vcpu, msr_info);
2429 break;
2430 case MSR_SMI_COUNT:
2431 if (!msr_info->host_initiated)
2432 return 1;
2433 vcpu->arch.smi_count = data;
2434 break;
2435 case MSR_KVM_WALL_CLOCK_NEW:
2436 case MSR_KVM_WALL_CLOCK:
2437 vcpu->kvm->arch.wall_clock = data;
2438 kvm_write_wall_clock(vcpu->kvm, data);
2439 break;
2440 case MSR_KVM_SYSTEM_TIME_NEW:
2441 case MSR_KVM_SYSTEM_TIME: {
2442 struct kvm_arch *ka = &vcpu->kvm->arch;
2444 kvmclock_reset(vcpu);
2446 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2447 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2449 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2450 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2452 ka->boot_vcpu_runs_old_kvmclock = tmp;
2455 vcpu->arch.time = data;
2456 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2458 /* we verify if the enable bit is set... */
2459 if (!(data & 1))
2460 break;
2462 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2463 &vcpu->arch.pv_time, data & ~1ULL,
2464 sizeof(struct pvclock_vcpu_time_info)))
2465 vcpu->arch.pv_time_enabled = false;
2466 else
2467 vcpu->arch.pv_time_enabled = true;
2469 break;
2471 case MSR_KVM_ASYNC_PF_EN:
2472 if (kvm_pv_enable_async_pf(vcpu, data))
2473 return 1;
2474 break;
2475 case MSR_KVM_STEAL_TIME:
2477 if (unlikely(!sched_info_on()))
2478 return 1;
2480 if (data & KVM_STEAL_RESERVED_MASK)
2481 return 1;
2483 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2484 data & KVM_STEAL_VALID_BITS,
2485 sizeof(struct kvm_steal_time)))
2486 return 1;
2488 vcpu->arch.st.msr_val = data;
2490 if (!(data & KVM_MSR_ENABLED))
2491 break;
2493 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2495 break;
2496 case MSR_KVM_PV_EOI_EN:
2497 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2498 return 1;
2499 break;
2501 case MSR_IA32_MCG_CTL:
2502 case MSR_IA32_MCG_STATUS:
2503 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2504 return set_msr_mce(vcpu, msr_info);
2506 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2507 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2508 pr = true; /* fall through */
2509 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2510 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2511 if (kvm_pmu_is_valid_msr(vcpu, msr))
2512 return kvm_pmu_set_msr(vcpu, msr_info);
2514 if (pr || data != 0)
2515 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2516 "0x%x data 0x%llx\n", msr, data);
2517 break;
2518 case MSR_K7_CLK_CTL:
2520 * Ignore all writes to this no longer documented MSR.
2521 * Writes are only relevant for old K7 processors,
2522 * all pre-dating SVM, but a recommended workaround from
2523 * AMD for these chips. It is possible to specify the
2524 * affected processor models on the command line, hence
2525 * the need to ignore the workaround.
2527 break;
2528 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2529 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2530 case HV_X64_MSR_CRASH_CTL:
2531 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2532 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2533 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2534 case HV_X64_MSR_TSC_EMULATION_STATUS:
2535 return kvm_hv_set_msr_common(vcpu, msr, data,
2536 msr_info->host_initiated);
2537 case MSR_IA32_BBL_CR_CTL3:
2538 /* Drop writes to this legacy MSR -- see rdmsr
2539 * counterpart for further detail.
2541 if (report_ignored_msrs)
2542 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2543 msr, data);
2544 break;
2545 case MSR_AMD64_OSVW_ID_LENGTH:
2546 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2547 return 1;
2548 vcpu->arch.osvw.length = data;
2549 break;
2550 case MSR_AMD64_OSVW_STATUS:
2551 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2552 return 1;
2553 vcpu->arch.osvw.status = data;
2554 break;
2555 case MSR_PLATFORM_INFO:
2556 if (!msr_info->host_initiated ||
2557 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2558 cpuid_fault_enabled(vcpu)))
2559 return 1;
2560 vcpu->arch.msr_platform_info = data;
2561 break;
2562 case MSR_MISC_FEATURES_ENABLES:
2563 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2564 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2565 !supports_cpuid_fault(vcpu)))
2566 return 1;
2567 vcpu->arch.msr_misc_features_enables = data;
2568 break;
2569 default:
2570 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2571 return xen_hvm_config(vcpu, data);
2572 if (kvm_pmu_is_valid_msr(vcpu, msr))
2573 return kvm_pmu_set_msr(vcpu, msr_info);
2574 if (!ignore_msrs) {
2575 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2576 msr, data);
2577 return 1;
2578 } else {
2579 if (report_ignored_msrs)
2580 vcpu_unimpl(vcpu,
2581 "ignored wrmsr: 0x%x data 0x%llx\n",
2582 msr, data);
2583 break;
2586 return 0;
2588 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2592 * Reads an msr value (of 'msr_index') into 'pdata'.
2593 * Returns 0 on success, non-0 otherwise.
2594 * Assumes vcpu_load() was already called.
2596 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2598 return kvm_x86_ops->get_msr(vcpu, msr);
2600 EXPORT_SYMBOL_GPL(kvm_get_msr);
2602 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2604 u64 data;
2605 u64 mcg_cap = vcpu->arch.mcg_cap;
2606 unsigned bank_num = mcg_cap & 0xff;
2608 switch (msr) {
2609 case MSR_IA32_P5_MC_ADDR:
2610 case MSR_IA32_P5_MC_TYPE:
2611 data = 0;
2612 break;
2613 case MSR_IA32_MCG_CAP:
2614 data = vcpu->arch.mcg_cap;
2615 break;
2616 case MSR_IA32_MCG_CTL:
2617 if (!(mcg_cap & MCG_CTL_P) && !host)
2618 return 1;
2619 data = vcpu->arch.mcg_ctl;
2620 break;
2621 case MSR_IA32_MCG_STATUS:
2622 data = vcpu->arch.mcg_status;
2623 break;
2624 default:
2625 if (msr >= MSR_IA32_MC0_CTL &&
2626 msr < MSR_IA32_MCx_CTL(bank_num)) {
2627 u32 offset = msr - MSR_IA32_MC0_CTL;
2628 data = vcpu->arch.mce_banks[offset];
2629 break;
2631 return 1;
2633 *pdata = data;
2634 return 0;
2637 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2639 switch (msr_info->index) {
2640 case MSR_IA32_PLATFORM_ID:
2641 case MSR_IA32_EBL_CR_POWERON:
2642 case MSR_IA32_DEBUGCTLMSR:
2643 case MSR_IA32_LASTBRANCHFROMIP:
2644 case MSR_IA32_LASTBRANCHTOIP:
2645 case MSR_IA32_LASTINTFROMIP:
2646 case MSR_IA32_LASTINTTOIP:
2647 case MSR_K8_SYSCFG:
2648 case MSR_K8_TSEG_ADDR:
2649 case MSR_K8_TSEG_MASK:
2650 case MSR_K7_HWCR:
2651 case MSR_VM_HSAVE_PA:
2652 case MSR_K8_INT_PENDING_MSG:
2653 case MSR_AMD64_NB_CFG:
2654 case MSR_FAM10H_MMIO_CONF_BASE:
2655 case MSR_AMD64_BU_CFG2:
2656 case MSR_IA32_PERF_CTL:
2657 case MSR_AMD64_DC_CFG:
2658 case MSR_F15H_EX_CFG:
2659 msr_info->data = 0;
2660 break;
2661 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2662 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2663 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2664 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2665 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2666 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2667 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2668 msr_info->data = 0;
2669 break;
2670 case MSR_IA32_UCODE_REV:
2671 msr_info->data = vcpu->arch.microcode_version;
2672 break;
2673 case MSR_IA32_ARCH_CAPABILITIES:
2674 if (!msr_info->host_initiated &&
2675 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
2676 return 1;
2677 msr_info->data = vcpu->arch.arch_capabilities;
2678 break;
2679 case MSR_IA32_TSC:
2680 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2681 break;
2682 case MSR_MTRRcap:
2683 case 0x200 ... 0x2ff:
2684 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2685 case 0xcd: /* fsb frequency */
2686 msr_info->data = 3;
2687 break;
2689 * MSR_EBC_FREQUENCY_ID
2690 * Conservative value valid for even the basic CPU models.
2691 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2692 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2693 * and 266MHz for model 3, or 4. Set Core Clock
2694 * Frequency to System Bus Frequency Ratio to 1 (bits
2695 * 31:24) even though these are only valid for CPU
2696 * models > 2, however guests may end up dividing or
2697 * multiplying by zero otherwise.
2699 case MSR_EBC_FREQUENCY_ID:
2700 msr_info->data = 1 << 24;
2701 break;
2702 case MSR_IA32_APICBASE:
2703 msr_info->data = kvm_get_apic_base(vcpu);
2704 break;
2705 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2706 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2707 break;
2708 case MSR_IA32_TSCDEADLINE:
2709 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2710 break;
2711 case MSR_IA32_TSC_ADJUST:
2712 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2713 break;
2714 case MSR_IA32_MISC_ENABLE:
2715 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2716 break;
2717 case MSR_IA32_SMBASE:
2718 if (!msr_info->host_initiated)
2719 return 1;
2720 msr_info->data = vcpu->arch.smbase;
2721 break;
2722 case MSR_SMI_COUNT:
2723 msr_info->data = vcpu->arch.smi_count;
2724 break;
2725 case MSR_IA32_PERF_STATUS:
2726 /* TSC increment by tick */
2727 msr_info->data = 1000ULL;
2728 /* CPU multiplier */
2729 msr_info->data |= (((uint64_t)4ULL) << 40);
2730 break;
2731 case MSR_EFER:
2732 msr_info->data = vcpu->arch.efer;
2733 break;
2734 case MSR_KVM_WALL_CLOCK:
2735 case MSR_KVM_WALL_CLOCK_NEW:
2736 msr_info->data = vcpu->kvm->arch.wall_clock;
2737 break;
2738 case MSR_KVM_SYSTEM_TIME:
2739 case MSR_KVM_SYSTEM_TIME_NEW:
2740 msr_info->data = vcpu->arch.time;
2741 break;
2742 case MSR_KVM_ASYNC_PF_EN:
2743 msr_info->data = vcpu->arch.apf.msr_val;
2744 break;
2745 case MSR_KVM_STEAL_TIME:
2746 msr_info->data = vcpu->arch.st.msr_val;
2747 break;
2748 case MSR_KVM_PV_EOI_EN:
2749 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2750 break;
2751 case MSR_IA32_P5_MC_ADDR:
2752 case MSR_IA32_P5_MC_TYPE:
2753 case MSR_IA32_MCG_CAP:
2754 case MSR_IA32_MCG_CTL:
2755 case MSR_IA32_MCG_STATUS:
2756 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2757 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
2758 msr_info->host_initiated);
2759 case MSR_K7_CLK_CTL:
2761 * Provide expected ramp-up count for K7. All other
2762 * are set to zero, indicating minimum divisors for
2763 * every field.
2765 * This prevents guest kernels on AMD host with CPU
2766 * type 6, model 8 and higher from exploding due to
2767 * the rdmsr failing.
2769 msr_info->data = 0x20000000;
2770 break;
2771 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2772 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2773 case HV_X64_MSR_CRASH_CTL:
2774 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2775 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2776 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2777 case HV_X64_MSR_TSC_EMULATION_STATUS:
2778 return kvm_hv_get_msr_common(vcpu,
2779 msr_info->index, &msr_info->data,
2780 msr_info->host_initiated);
2781 break;
2782 case MSR_IA32_BBL_CR_CTL3:
2783 /* This legacy MSR exists but isn't fully documented in current
2784 * silicon. It is however accessed by winxp in very narrow
2785 * scenarios where it sets bit #19, itself documented as
2786 * a "reserved" bit. Best effort attempt to source coherent
2787 * read data here should the balance of the register be
2788 * interpreted by the guest:
2790 * L2 cache control register 3: 64GB range, 256KB size,
2791 * enabled, latency 0x1, configured
2793 msr_info->data = 0xbe702111;
2794 break;
2795 case MSR_AMD64_OSVW_ID_LENGTH:
2796 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2797 return 1;
2798 msr_info->data = vcpu->arch.osvw.length;
2799 break;
2800 case MSR_AMD64_OSVW_STATUS:
2801 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2802 return 1;
2803 msr_info->data = vcpu->arch.osvw.status;
2804 break;
2805 case MSR_PLATFORM_INFO:
2806 if (!msr_info->host_initiated &&
2807 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
2808 return 1;
2809 msr_info->data = vcpu->arch.msr_platform_info;
2810 break;
2811 case MSR_MISC_FEATURES_ENABLES:
2812 msr_info->data = vcpu->arch.msr_misc_features_enables;
2813 break;
2814 default:
2815 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2816 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2817 if (!ignore_msrs) {
2818 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2819 msr_info->index);
2820 return 1;
2821 } else {
2822 if (report_ignored_msrs)
2823 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
2824 msr_info->index);
2825 msr_info->data = 0;
2827 break;
2829 return 0;
2831 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2834 * Read or write a bunch of msrs. All parameters are kernel addresses.
2836 * @return number of msrs set successfully.
2838 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2839 struct kvm_msr_entry *entries,
2840 int (*do_msr)(struct kvm_vcpu *vcpu,
2841 unsigned index, u64 *data))
2843 int i;
2845 for (i = 0; i < msrs->nmsrs; ++i)
2846 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2847 break;
2849 return i;
2853 * Read or write a bunch of msrs. Parameters are user addresses.
2855 * @return number of msrs set successfully.
2857 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2858 int (*do_msr)(struct kvm_vcpu *vcpu,
2859 unsigned index, u64 *data),
2860 int writeback)
2862 struct kvm_msrs msrs;
2863 struct kvm_msr_entry *entries;
2864 int r, n;
2865 unsigned size;
2867 r = -EFAULT;
2868 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2869 goto out;
2871 r = -E2BIG;
2872 if (msrs.nmsrs >= MAX_IO_MSRS)
2873 goto out;
2875 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2876 entries = memdup_user(user_msrs->entries, size);
2877 if (IS_ERR(entries)) {
2878 r = PTR_ERR(entries);
2879 goto out;
2882 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2883 if (r < 0)
2884 goto out_free;
2886 r = -EFAULT;
2887 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2888 goto out_free;
2890 r = n;
2892 out_free:
2893 kfree(entries);
2894 out:
2895 return r;
2898 static inline bool kvm_can_mwait_in_guest(void)
2900 return boot_cpu_has(X86_FEATURE_MWAIT) &&
2901 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
2902 boot_cpu_has(X86_FEATURE_ARAT);
2905 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2907 int r = 0;
2909 switch (ext) {
2910 case KVM_CAP_IRQCHIP:
2911 case KVM_CAP_HLT:
2912 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2913 case KVM_CAP_SET_TSS_ADDR:
2914 case KVM_CAP_EXT_CPUID:
2915 case KVM_CAP_EXT_EMUL_CPUID:
2916 case KVM_CAP_CLOCKSOURCE:
2917 case KVM_CAP_PIT:
2918 case KVM_CAP_NOP_IO_DELAY:
2919 case KVM_CAP_MP_STATE:
2920 case KVM_CAP_SYNC_MMU:
2921 case KVM_CAP_USER_NMI:
2922 case KVM_CAP_REINJECT_CONTROL:
2923 case KVM_CAP_IRQ_INJECT_STATUS:
2924 case KVM_CAP_IOEVENTFD:
2925 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2926 case KVM_CAP_PIT2:
2927 case KVM_CAP_PIT_STATE2:
2928 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2929 case KVM_CAP_XEN_HVM:
2930 case KVM_CAP_VCPU_EVENTS:
2931 case KVM_CAP_HYPERV:
2932 case KVM_CAP_HYPERV_VAPIC:
2933 case KVM_CAP_HYPERV_SPIN:
2934 case KVM_CAP_HYPERV_SYNIC:
2935 case KVM_CAP_HYPERV_SYNIC2:
2936 case KVM_CAP_HYPERV_VP_INDEX:
2937 case KVM_CAP_HYPERV_EVENTFD:
2938 case KVM_CAP_HYPERV_TLBFLUSH:
2939 case KVM_CAP_PCI_SEGMENT:
2940 case KVM_CAP_DEBUGREGS:
2941 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2942 case KVM_CAP_XSAVE:
2943 case KVM_CAP_ASYNC_PF:
2944 case KVM_CAP_GET_TSC_KHZ:
2945 case KVM_CAP_KVMCLOCK_CTRL:
2946 case KVM_CAP_READONLY_MEM:
2947 case KVM_CAP_HYPERV_TIME:
2948 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2949 case KVM_CAP_TSC_DEADLINE_TIMER:
2950 case KVM_CAP_ENABLE_CAP_VM:
2951 case KVM_CAP_DISABLE_QUIRKS:
2952 case KVM_CAP_SET_BOOT_CPU_ID:
2953 case KVM_CAP_SPLIT_IRQCHIP:
2954 case KVM_CAP_IMMEDIATE_EXIT:
2955 case KVM_CAP_GET_MSR_FEATURES:
2956 case KVM_CAP_MSR_PLATFORM_INFO:
2957 r = 1;
2958 break;
2959 case KVM_CAP_SYNC_REGS:
2960 r = KVM_SYNC_X86_VALID_FIELDS;
2961 break;
2962 case KVM_CAP_ADJUST_CLOCK:
2963 r = KVM_CLOCK_TSC_STABLE;
2964 break;
2965 case KVM_CAP_X86_DISABLE_EXITS:
2966 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE;
2967 if(kvm_can_mwait_in_guest())
2968 r |= KVM_X86_DISABLE_EXITS_MWAIT;
2969 break;
2970 case KVM_CAP_X86_SMM:
2971 /* SMBASE is usually relocated above 1M on modern chipsets,
2972 * and SMM handlers might indeed rely on 4G segment limits,
2973 * so do not report SMM to be available if real mode is
2974 * emulated via vm86 mode. Still, do not go to great lengths
2975 * to avoid userspace's usage of the feature, because it is a
2976 * fringe case that is not enabled except via specific settings
2977 * of the module parameters.
2979 r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
2980 break;
2981 case KVM_CAP_VAPIC:
2982 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2983 break;
2984 case KVM_CAP_NR_VCPUS:
2985 r = KVM_SOFT_MAX_VCPUS;
2986 break;
2987 case KVM_CAP_MAX_VCPUS:
2988 r = KVM_MAX_VCPUS;
2989 break;
2990 case KVM_CAP_MAX_VCPU_ID:
2991 r = KVM_MAX_VCPU_ID;
2992 break;
2993 case KVM_CAP_NR_MEMSLOTS:
2994 r = KVM_USER_MEM_SLOTS;
2995 break;
2996 case KVM_CAP_PV_MMU: /* obsolete */
2997 r = 0;
2998 break;
2999 case KVM_CAP_MCE:
3000 r = KVM_MAX_MCE_BANKS;
3001 break;
3002 case KVM_CAP_XCRS:
3003 r = boot_cpu_has(X86_FEATURE_XSAVE);
3004 break;
3005 case KVM_CAP_TSC_CONTROL:
3006 r = kvm_has_tsc_control;
3007 break;
3008 case KVM_CAP_X2APIC_API:
3009 r = KVM_X2APIC_API_VALID_FLAGS;
3010 break;
3011 case KVM_CAP_NESTED_STATE:
3012 r = kvm_x86_ops->get_nested_state ?
3013 kvm_x86_ops->get_nested_state(NULL, 0, 0) : 0;
3014 break;
3015 default:
3016 break;
3018 return r;
3022 long kvm_arch_dev_ioctl(struct file *filp,
3023 unsigned int ioctl, unsigned long arg)
3025 void __user *argp = (void __user *)arg;
3026 long r;
3028 switch (ioctl) {
3029 case KVM_GET_MSR_INDEX_LIST: {
3030 struct kvm_msr_list __user *user_msr_list = argp;
3031 struct kvm_msr_list msr_list;
3032 unsigned n;
3034 r = -EFAULT;
3035 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
3036 goto out;
3037 n = msr_list.nmsrs;
3038 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3039 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
3040 goto out;
3041 r = -E2BIG;
3042 if (n < msr_list.nmsrs)
3043 goto out;
3044 r = -EFAULT;
3045 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3046 num_msrs_to_save * sizeof(u32)))
3047 goto out;
3048 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3049 &emulated_msrs,
3050 num_emulated_msrs * sizeof(u32)))
3051 goto out;
3052 r = 0;
3053 break;
3055 case KVM_GET_SUPPORTED_CPUID:
3056 case KVM_GET_EMULATED_CPUID: {
3057 struct kvm_cpuid2 __user *cpuid_arg = argp;
3058 struct kvm_cpuid2 cpuid;
3060 r = -EFAULT;
3061 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3062 goto out;
3064 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3065 ioctl);
3066 if (r)
3067 goto out;
3069 r = -EFAULT;
3070 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3071 goto out;
3072 r = 0;
3073 break;
3075 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3076 r = -EFAULT;
3077 if (copy_to_user(argp, &kvm_mce_cap_supported,
3078 sizeof(kvm_mce_cap_supported)))
3079 goto out;
3080 r = 0;
3081 break;
3082 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3083 struct kvm_msr_list __user *user_msr_list = argp;
3084 struct kvm_msr_list msr_list;
3085 unsigned int n;
3087 r = -EFAULT;
3088 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3089 goto out;
3090 n = msr_list.nmsrs;
3091 msr_list.nmsrs = num_msr_based_features;
3092 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3093 goto out;
3094 r = -E2BIG;
3095 if (n < msr_list.nmsrs)
3096 goto out;
3097 r = -EFAULT;
3098 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3099 num_msr_based_features * sizeof(u32)))
3100 goto out;
3101 r = 0;
3102 break;
3104 case KVM_GET_MSRS:
3105 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3106 break;
3108 default:
3109 r = -EINVAL;
3111 out:
3112 return r;
3115 static void wbinvd_ipi(void *garbage)
3117 wbinvd();
3120 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3122 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3125 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3127 /* Address WBINVD may be executed by guest */
3128 if (need_emulate_wbinvd(vcpu)) {
3129 if (kvm_x86_ops->has_wbinvd_exit())
3130 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3131 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3132 smp_call_function_single(vcpu->cpu,
3133 wbinvd_ipi, NULL, 1);
3136 kvm_x86_ops->vcpu_load(vcpu, cpu);
3138 /* Apply any externally detected TSC adjustments (due to suspend) */
3139 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3140 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3141 vcpu->arch.tsc_offset_adjustment = 0;
3142 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3145 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3146 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3147 rdtsc() - vcpu->arch.last_host_tsc;
3148 if (tsc_delta < 0)
3149 mark_tsc_unstable("KVM discovered backwards TSC");
3151 if (kvm_check_tsc_unstable()) {
3152 u64 offset = kvm_compute_tsc_offset(vcpu,
3153 vcpu->arch.last_guest_tsc);
3154 kvm_vcpu_write_tsc_offset(vcpu, offset);
3155 vcpu->arch.tsc_catchup = 1;
3158 if (kvm_lapic_hv_timer_in_use(vcpu))
3159 kvm_lapic_restart_hv_timer(vcpu);
3162 * On a host with synchronized TSC, there is no need to update
3163 * kvmclock on vcpu->cpu migration
3165 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3166 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3167 if (vcpu->cpu != cpu)
3168 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3169 vcpu->cpu = cpu;
3172 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3175 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3177 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3178 return;
3180 vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3182 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3183 &vcpu->arch.st.steal.preempted,
3184 offsetof(struct kvm_steal_time, preempted),
3185 sizeof(vcpu->arch.st.steal.preempted));
3188 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3190 int idx;
3192 if (vcpu->preempted)
3193 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3196 * Disable page faults because we're in atomic context here.
3197 * kvm_write_guest_offset_cached() would call might_fault()
3198 * that relies on pagefault_disable() to tell if there's a
3199 * bug. NOTE: the write to guest memory may not go through if
3200 * during postcopy live migration or if there's heavy guest
3201 * paging.
3203 pagefault_disable();
3205 * kvm_memslots() will be called by
3206 * kvm_write_guest_offset_cached() so take the srcu lock.
3208 idx = srcu_read_lock(&vcpu->kvm->srcu);
3209 kvm_steal_time_set_preempted(vcpu);
3210 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3211 pagefault_enable();
3212 kvm_x86_ops->vcpu_put(vcpu);
3213 vcpu->arch.last_host_tsc = rdtsc();
3215 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3216 * on every vmexit, but if not, we might have a stale dr6 from the
3217 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3219 set_debugreg(0, 6);
3222 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3223 struct kvm_lapic_state *s)
3225 if (vcpu->arch.apicv_active)
3226 kvm_x86_ops->sync_pir_to_irr(vcpu);
3228 return kvm_apic_get_state(vcpu, s);
3231 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3232 struct kvm_lapic_state *s)
3234 int r;
3236 r = kvm_apic_set_state(vcpu, s);
3237 if (r)
3238 return r;
3239 update_cr8_intercept(vcpu);
3241 return 0;
3244 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3246 return (!lapic_in_kernel(vcpu) ||
3247 kvm_apic_accept_pic_intr(vcpu));
3251 * if userspace requested an interrupt window, check that the
3252 * interrupt window is open.
3254 * No need to exit to userspace if we already have an interrupt queued.
3256 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3258 return kvm_arch_interrupt_allowed(vcpu) &&
3259 !kvm_cpu_has_interrupt(vcpu) &&
3260 !kvm_event_needs_reinjection(vcpu) &&
3261 kvm_cpu_accept_dm_intr(vcpu);
3264 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3265 struct kvm_interrupt *irq)
3267 if (irq->irq >= KVM_NR_INTERRUPTS)
3268 return -EINVAL;
3270 if (!irqchip_in_kernel(vcpu->kvm)) {
3271 kvm_queue_interrupt(vcpu, irq->irq, false);
3272 kvm_make_request(KVM_REQ_EVENT, vcpu);
3273 return 0;
3277 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3278 * fail for in-kernel 8259.
3280 if (pic_in_kernel(vcpu->kvm))
3281 return -ENXIO;
3283 if (vcpu->arch.pending_external_vector != -1)
3284 return -EEXIST;
3286 vcpu->arch.pending_external_vector = irq->irq;
3287 kvm_make_request(KVM_REQ_EVENT, vcpu);
3288 return 0;
3291 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3293 kvm_inject_nmi(vcpu);
3295 return 0;
3298 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3300 kvm_make_request(KVM_REQ_SMI, vcpu);
3302 return 0;
3305 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3306 struct kvm_tpr_access_ctl *tac)
3308 if (tac->flags)
3309 return -EINVAL;
3310 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3311 return 0;
3314 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3315 u64 mcg_cap)
3317 int r;
3318 unsigned bank_num = mcg_cap & 0xff, bank;
3320 r = -EINVAL;
3321 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3322 goto out;
3323 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3324 goto out;
3325 r = 0;
3326 vcpu->arch.mcg_cap = mcg_cap;
3327 /* Init IA32_MCG_CTL to all 1s */
3328 if (mcg_cap & MCG_CTL_P)
3329 vcpu->arch.mcg_ctl = ~(u64)0;
3330 /* Init IA32_MCi_CTL to all 1s */
3331 for (bank = 0; bank < bank_num; bank++)
3332 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3334 if (kvm_x86_ops->setup_mce)
3335 kvm_x86_ops->setup_mce(vcpu);
3336 out:
3337 return r;
3340 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3341 struct kvm_x86_mce *mce)
3343 u64 mcg_cap = vcpu->arch.mcg_cap;
3344 unsigned bank_num = mcg_cap & 0xff;
3345 u64 *banks = vcpu->arch.mce_banks;
3347 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3348 return -EINVAL;
3350 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3351 * reporting is disabled
3353 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3354 vcpu->arch.mcg_ctl != ~(u64)0)
3355 return 0;
3356 banks += 4 * mce->bank;
3358 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3359 * reporting is disabled for the bank
3361 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3362 return 0;
3363 if (mce->status & MCI_STATUS_UC) {
3364 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3365 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3366 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3367 return 0;
3369 if (banks[1] & MCI_STATUS_VAL)
3370 mce->status |= MCI_STATUS_OVER;
3371 banks[2] = mce->addr;
3372 banks[3] = mce->misc;
3373 vcpu->arch.mcg_status = mce->mcg_status;
3374 banks[1] = mce->status;
3375 kvm_queue_exception(vcpu, MC_VECTOR);
3376 } else if (!(banks[1] & MCI_STATUS_VAL)
3377 || !(banks[1] & MCI_STATUS_UC)) {
3378 if (banks[1] & MCI_STATUS_VAL)
3379 mce->status |= MCI_STATUS_OVER;
3380 banks[2] = mce->addr;
3381 banks[3] = mce->misc;
3382 banks[1] = mce->status;
3383 } else
3384 banks[1] |= MCI_STATUS_OVER;
3385 return 0;
3388 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3389 struct kvm_vcpu_events *events)
3391 process_nmi(vcpu);
3393 * FIXME: pass injected and pending separately. This is only
3394 * needed for nested virtualization, whose state cannot be
3395 * migrated yet. For now we can combine them.
3397 events->exception.injected =
3398 (vcpu->arch.exception.pending ||
3399 vcpu->arch.exception.injected) &&
3400 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3401 events->exception.nr = vcpu->arch.exception.nr;
3402 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3403 events->exception.pad = 0;
3404 events->exception.error_code = vcpu->arch.exception.error_code;
3406 events->interrupt.injected =
3407 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3408 events->interrupt.nr = vcpu->arch.interrupt.nr;
3409 events->interrupt.soft = 0;
3410 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3412 events->nmi.injected = vcpu->arch.nmi_injected;
3413 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3414 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3415 events->nmi.pad = 0;
3417 events->sipi_vector = 0; /* never valid when reporting to user space */
3419 events->smi.smm = is_smm(vcpu);
3420 events->smi.pending = vcpu->arch.smi_pending;
3421 events->smi.smm_inside_nmi =
3422 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3423 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3425 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3426 | KVM_VCPUEVENT_VALID_SHADOW
3427 | KVM_VCPUEVENT_VALID_SMM);
3428 memset(&events->reserved, 0, sizeof(events->reserved));
3431 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3433 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3434 struct kvm_vcpu_events *events)
3436 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3437 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3438 | KVM_VCPUEVENT_VALID_SHADOW
3439 | KVM_VCPUEVENT_VALID_SMM))
3440 return -EINVAL;
3442 if (events->exception.injected &&
3443 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR ||
3444 is_guest_mode(vcpu)))
3445 return -EINVAL;
3447 /* INITs are latched while in SMM */
3448 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3449 (events->smi.smm || events->smi.pending) &&
3450 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3451 return -EINVAL;
3453 process_nmi(vcpu);
3454 vcpu->arch.exception.injected = false;
3455 vcpu->arch.exception.pending = events->exception.injected;
3456 vcpu->arch.exception.nr = events->exception.nr;
3457 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3458 vcpu->arch.exception.error_code = events->exception.error_code;
3460 vcpu->arch.interrupt.injected = events->interrupt.injected;
3461 vcpu->arch.interrupt.nr = events->interrupt.nr;
3462 vcpu->arch.interrupt.soft = events->interrupt.soft;
3463 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3464 kvm_x86_ops->set_interrupt_shadow(vcpu,
3465 events->interrupt.shadow);
3467 vcpu->arch.nmi_injected = events->nmi.injected;
3468 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3469 vcpu->arch.nmi_pending = events->nmi.pending;
3470 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3472 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3473 lapic_in_kernel(vcpu))
3474 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3476 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3477 u32 hflags = vcpu->arch.hflags;
3478 if (events->smi.smm)
3479 hflags |= HF_SMM_MASK;
3480 else
3481 hflags &= ~HF_SMM_MASK;
3482 kvm_set_hflags(vcpu, hflags);
3484 vcpu->arch.smi_pending = events->smi.pending;
3486 if (events->smi.smm) {
3487 if (events->smi.smm_inside_nmi)
3488 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3489 else
3490 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3491 if (lapic_in_kernel(vcpu)) {
3492 if (events->smi.latched_init)
3493 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3494 else
3495 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3500 kvm_make_request(KVM_REQ_EVENT, vcpu);
3502 return 0;
3505 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3506 struct kvm_debugregs *dbgregs)
3508 unsigned long val;
3510 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3511 kvm_get_dr(vcpu, 6, &val);
3512 dbgregs->dr6 = val;
3513 dbgregs->dr7 = vcpu->arch.dr7;
3514 dbgregs->flags = 0;
3515 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3518 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3519 struct kvm_debugregs *dbgregs)
3521 if (dbgregs->flags)
3522 return -EINVAL;
3524 if (dbgregs->dr6 & ~0xffffffffull)
3525 return -EINVAL;
3526 if (dbgregs->dr7 & ~0xffffffffull)
3527 return -EINVAL;
3529 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3530 kvm_update_dr0123(vcpu);
3531 vcpu->arch.dr6 = dbgregs->dr6;
3532 kvm_update_dr6(vcpu);
3533 vcpu->arch.dr7 = dbgregs->dr7;
3534 kvm_update_dr7(vcpu);
3536 return 0;
3539 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3541 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3543 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3544 u64 xstate_bv = xsave->header.xfeatures;
3545 u64 valid;
3548 * Copy legacy XSAVE area, to avoid complications with CPUID
3549 * leaves 0 and 1 in the loop below.
3551 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3553 /* Set XSTATE_BV */
3554 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3555 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3558 * Copy each region from the possibly compacted offset to the
3559 * non-compacted offset.
3561 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3562 while (valid) {
3563 u64 feature = valid & -valid;
3564 int index = fls64(feature) - 1;
3565 void *src = get_xsave_addr(xsave, feature);
3567 if (src) {
3568 u32 size, offset, ecx, edx;
3569 cpuid_count(XSTATE_CPUID, index,
3570 &size, &offset, &ecx, &edx);
3571 if (feature == XFEATURE_MASK_PKRU)
3572 memcpy(dest + offset, &vcpu->arch.pkru,
3573 sizeof(vcpu->arch.pkru));
3574 else
3575 memcpy(dest + offset, src, size);
3579 valid -= feature;
3583 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3585 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3586 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3587 u64 valid;
3590 * Copy legacy XSAVE area, to avoid complications with CPUID
3591 * leaves 0 and 1 in the loop below.
3593 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3595 /* Set XSTATE_BV and possibly XCOMP_BV. */
3596 xsave->header.xfeatures = xstate_bv;
3597 if (boot_cpu_has(X86_FEATURE_XSAVES))
3598 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3601 * Copy each region from the non-compacted offset to the
3602 * possibly compacted offset.
3604 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3605 while (valid) {
3606 u64 feature = valid & -valid;
3607 int index = fls64(feature) - 1;
3608 void *dest = get_xsave_addr(xsave, feature);
3610 if (dest) {
3611 u32 size, offset, ecx, edx;
3612 cpuid_count(XSTATE_CPUID, index,
3613 &size, &offset, &ecx, &edx);
3614 if (feature == XFEATURE_MASK_PKRU)
3615 memcpy(&vcpu->arch.pkru, src + offset,
3616 sizeof(vcpu->arch.pkru));
3617 else
3618 memcpy(dest, src + offset, size);
3621 valid -= feature;
3625 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3626 struct kvm_xsave *guest_xsave)
3628 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3629 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3630 fill_xsave((u8 *) guest_xsave->region, vcpu);
3631 } else {
3632 memcpy(guest_xsave->region,
3633 &vcpu->arch.guest_fpu.state.fxsave,
3634 sizeof(struct fxregs_state));
3635 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3636 XFEATURE_MASK_FPSSE;
3640 #define XSAVE_MXCSR_OFFSET 24
3642 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3643 struct kvm_xsave *guest_xsave)
3645 u64 xstate_bv =
3646 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3647 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3649 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3651 * Here we allow setting states that are not present in
3652 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3653 * with old userspace.
3655 if (xstate_bv & ~kvm_supported_xcr0() ||
3656 mxcsr & ~mxcsr_feature_mask)
3657 return -EINVAL;
3658 load_xsave(vcpu, (u8 *)guest_xsave->region);
3659 } else {
3660 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3661 mxcsr & ~mxcsr_feature_mask)
3662 return -EINVAL;
3663 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3664 guest_xsave->region, sizeof(struct fxregs_state));
3666 return 0;
3669 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3670 struct kvm_xcrs *guest_xcrs)
3672 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3673 guest_xcrs->nr_xcrs = 0;
3674 return;
3677 guest_xcrs->nr_xcrs = 1;
3678 guest_xcrs->flags = 0;
3679 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3680 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3683 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3684 struct kvm_xcrs *guest_xcrs)
3686 int i, r = 0;
3688 if (!boot_cpu_has(X86_FEATURE_XSAVE))
3689 return -EINVAL;
3691 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3692 return -EINVAL;
3694 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3695 /* Only support XCR0 currently */
3696 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3697 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3698 guest_xcrs->xcrs[i].value);
3699 break;
3701 if (r)
3702 r = -EINVAL;
3703 return r;
3707 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3708 * stopped by the hypervisor. This function will be called from the host only.
3709 * EINVAL is returned when the host attempts to set the flag for a guest that
3710 * does not support pv clocks.
3712 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3714 if (!vcpu->arch.pv_time_enabled)
3715 return -EINVAL;
3716 vcpu->arch.pvclock_set_guest_stopped_request = true;
3717 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3718 return 0;
3721 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3722 struct kvm_enable_cap *cap)
3724 if (cap->flags)
3725 return -EINVAL;
3727 switch (cap->cap) {
3728 case KVM_CAP_HYPERV_SYNIC2:
3729 if (cap->args[0])
3730 return -EINVAL;
3731 case KVM_CAP_HYPERV_SYNIC:
3732 if (!irqchip_in_kernel(vcpu->kvm))
3733 return -EINVAL;
3734 return kvm_hv_activate_synic(vcpu, cap->cap ==
3735 KVM_CAP_HYPERV_SYNIC2);
3736 default:
3737 return -EINVAL;
3741 long kvm_arch_vcpu_ioctl(struct file *filp,
3742 unsigned int ioctl, unsigned long arg)
3744 struct kvm_vcpu *vcpu = filp->private_data;
3745 void __user *argp = (void __user *)arg;
3746 int r;
3747 union {
3748 struct kvm_lapic_state *lapic;
3749 struct kvm_xsave *xsave;
3750 struct kvm_xcrs *xcrs;
3751 void *buffer;
3752 } u;
3754 vcpu_load(vcpu);
3756 u.buffer = NULL;
3757 switch (ioctl) {
3758 case KVM_GET_LAPIC: {
3759 r = -EINVAL;
3760 if (!lapic_in_kernel(vcpu))
3761 goto out;
3762 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3764 r = -ENOMEM;
3765 if (!u.lapic)
3766 goto out;
3767 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3768 if (r)
3769 goto out;
3770 r = -EFAULT;
3771 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3772 goto out;
3773 r = 0;
3774 break;
3776 case KVM_SET_LAPIC: {
3777 r = -EINVAL;
3778 if (!lapic_in_kernel(vcpu))
3779 goto out;
3780 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3781 if (IS_ERR(u.lapic)) {
3782 r = PTR_ERR(u.lapic);
3783 goto out_nofree;
3786 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3787 break;
3789 case KVM_INTERRUPT: {
3790 struct kvm_interrupt irq;
3792 r = -EFAULT;
3793 if (copy_from_user(&irq, argp, sizeof irq))
3794 goto out;
3795 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3796 break;
3798 case KVM_NMI: {
3799 r = kvm_vcpu_ioctl_nmi(vcpu);
3800 break;
3802 case KVM_SMI: {
3803 r = kvm_vcpu_ioctl_smi(vcpu);
3804 break;
3806 case KVM_SET_CPUID: {
3807 struct kvm_cpuid __user *cpuid_arg = argp;
3808 struct kvm_cpuid cpuid;
3810 r = -EFAULT;
3811 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3812 goto out;
3813 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3814 break;
3816 case KVM_SET_CPUID2: {
3817 struct kvm_cpuid2 __user *cpuid_arg = argp;
3818 struct kvm_cpuid2 cpuid;
3820 r = -EFAULT;
3821 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3822 goto out;
3823 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3824 cpuid_arg->entries);
3825 break;
3827 case KVM_GET_CPUID2: {
3828 struct kvm_cpuid2 __user *cpuid_arg = argp;
3829 struct kvm_cpuid2 cpuid;
3831 r = -EFAULT;
3832 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3833 goto out;
3834 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3835 cpuid_arg->entries);
3836 if (r)
3837 goto out;
3838 r = -EFAULT;
3839 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3840 goto out;
3841 r = 0;
3842 break;
3844 case KVM_GET_MSRS: {
3845 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3846 r = msr_io(vcpu, argp, do_get_msr, 1);
3847 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3848 break;
3850 case KVM_SET_MSRS: {
3851 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3852 r = msr_io(vcpu, argp, do_set_msr, 0);
3853 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3854 break;
3856 case KVM_TPR_ACCESS_REPORTING: {
3857 struct kvm_tpr_access_ctl tac;
3859 r = -EFAULT;
3860 if (copy_from_user(&tac, argp, sizeof tac))
3861 goto out;
3862 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3863 if (r)
3864 goto out;
3865 r = -EFAULT;
3866 if (copy_to_user(argp, &tac, sizeof tac))
3867 goto out;
3868 r = 0;
3869 break;
3871 case KVM_SET_VAPIC_ADDR: {
3872 struct kvm_vapic_addr va;
3873 int idx;
3875 r = -EINVAL;
3876 if (!lapic_in_kernel(vcpu))
3877 goto out;
3878 r = -EFAULT;
3879 if (copy_from_user(&va, argp, sizeof va))
3880 goto out;
3881 idx = srcu_read_lock(&vcpu->kvm->srcu);
3882 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3883 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3884 break;
3886 case KVM_X86_SETUP_MCE: {
3887 u64 mcg_cap;
3889 r = -EFAULT;
3890 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3891 goto out;
3892 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3893 break;
3895 case KVM_X86_SET_MCE: {
3896 struct kvm_x86_mce mce;
3898 r = -EFAULT;
3899 if (copy_from_user(&mce, argp, sizeof mce))
3900 goto out;
3901 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3902 break;
3904 case KVM_GET_VCPU_EVENTS: {
3905 struct kvm_vcpu_events events;
3907 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3909 r = -EFAULT;
3910 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3911 break;
3912 r = 0;
3913 break;
3915 case KVM_SET_VCPU_EVENTS: {
3916 struct kvm_vcpu_events events;
3918 r = -EFAULT;
3919 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3920 break;
3922 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3923 break;
3925 case KVM_GET_DEBUGREGS: {
3926 struct kvm_debugregs dbgregs;
3928 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3930 r = -EFAULT;
3931 if (copy_to_user(argp, &dbgregs,
3932 sizeof(struct kvm_debugregs)))
3933 break;
3934 r = 0;
3935 break;
3937 case KVM_SET_DEBUGREGS: {
3938 struct kvm_debugregs dbgregs;
3940 r = -EFAULT;
3941 if (copy_from_user(&dbgregs, argp,
3942 sizeof(struct kvm_debugregs)))
3943 break;
3945 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3946 break;
3948 case KVM_GET_XSAVE: {
3949 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3950 r = -ENOMEM;
3951 if (!u.xsave)
3952 break;
3954 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3956 r = -EFAULT;
3957 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3958 break;
3959 r = 0;
3960 break;
3962 case KVM_SET_XSAVE: {
3963 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3964 if (IS_ERR(u.xsave)) {
3965 r = PTR_ERR(u.xsave);
3966 goto out_nofree;
3969 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3970 break;
3972 case KVM_GET_XCRS: {
3973 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3974 r = -ENOMEM;
3975 if (!u.xcrs)
3976 break;
3978 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3980 r = -EFAULT;
3981 if (copy_to_user(argp, u.xcrs,
3982 sizeof(struct kvm_xcrs)))
3983 break;
3984 r = 0;
3985 break;
3987 case KVM_SET_XCRS: {
3988 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3989 if (IS_ERR(u.xcrs)) {
3990 r = PTR_ERR(u.xcrs);
3991 goto out_nofree;
3994 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3995 break;
3997 case KVM_SET_TSC_KHZ: {
3998 u32 user_tsc_khz;
4000 r = -EINVAL;
4001 user_tsc_khz = (u32)arg;
4003 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4004 goto out;
4006 if (user_tsc_khz == 0)
4007 user_tsc_khz = tsc_khz;
4009 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4010 r = 0;
4012 goto out;
4014 case KVM_GET_TSC_KHZ: {
4015 r = vcpu->arch.virtual_tsc_khz;
4016 goto out;
4018 case KVM_KVMCLOCK_CTRL: {
4019 r = kvm_set_guest_paused(vcpu);
4020 goto out;
4022 case KVM_ENABLE_CAP: {
4023 struct kvm_enable_cap cap;
4025 r = -EFAULT;
4026 if (copy_from_user(&cap, argp, sizeof(cap)))
4027 goto out;
4028 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4029 break;
4031 case KVM_GET_NESTED_STATE: {
4032 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4033 u32 user_data_size;
4035 r = -EINVAL;
4036 if (!kvm_x86_ops->get_nested_state)
4037 break;
4039 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4040 r = -EFAULT;
4041 if (get_user(user_data_size, &user_kvm_nested_state->size))
4042 break;
4044 r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4045 user_data_size);
4046 if (r < 0)
4047 break;
4049 if (r > user_data_size) {
4050 if (put_user(r, &user_kvm_nested_state->size))
4051 r = -EFAULT;
4052 else
4053 r = -E2BIG;
4054 break;
4057 r = 0;
4058 break;
4060 case KVM_SET_NESTED_STATE: {
4061 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4062 struct kvm_nested_state kvm_state;
4064 r = -EINVAL;
4065 if (!kvm_x86_ops->set_nested_state)
4066 break;
4068 r = -EFAULT;
4069 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4070 break;
4072 r = -EINVAL;
4073 if (kvm_state.size < sizeof(kvm_state))
4074 break;
4076 if (kvm_state.flags &
4077 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE))
4078 break;
4080 /* nested_run_pending implies guest_mode. */
4081 if (kvm_state.flags == KVM_STATE_NESTED_RUN_PENDING)
4082 break;
4084 r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4085 break;
4087 default:
4088 r = -EINVAL;
4090 out:
4091 kfree(u.buffer);
4092 out_nofree:
4093 vcpu_put(vcpu);
4094 return r;
4097 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4099 return VM_FAULT_SIGBUS;
4102 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4104 int ret;
4106 if (addr > (unsigned int)(-3 * PAGE_SIZE))
4107 return -EINVAL;
4108 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4109 return ret;
4112 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4113 u64 ident_addr)
4115 return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4118 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4119 u32 kvm_nr_mmu_pages)
4121 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4122 return -EINVAL;
4124 mutex_lock(&kvm->slots_lock);
4126 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4127 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4129 mutex_unlock(&kvm->slots_lock);
4130 return 0;
4133 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4135 return kvm->arch.n_max_mmu_pages;
4138 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4140 struct kvm_pic *pic = kvm->arch.vpic;
4141 int r;
4143 r = 0;
4144 switch (chip->chip_id) {
4145 case KVM_IRQCHIP_PIC_MASTER:
4146 memcpy(&chip->chip.pic, &pic->pics[0],
4147 sizeof(struct kvm_pic_state));
4148 break;
4149 case KVM_IRQCHIP_PIC_SLAVE:
4150 memcpy(&chip->chip.pic, &pic->pics[1],
4151 sizeof(struct kvm_pic_state));
4152 break;
4153 case KVM_IRQCHIP_IOAPIC:
4154 kvm_get_ioapic(kvm, &chip->chip.ioapic);
4155 break;
4156 default:
4157 r = -EINVAL;
4158 break;
4160 return r;
4163 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4165 struct kvm_pic *pic = kvm->arch.vpic;
4166 int r;
4168 r = 0;
4169 switch (chip->chip_id) {
4170 case KVM_IRQCHIP_PIC_MASTER:
4171 spin_lock(&pic->lock);
4172 memcpy(&pic->pics[0], &chip->chip.pic,
4173 sizeof(struct kvm_pic_state));
4174 spin_unlock(&pic->lock);
4175 break;
4176 case KVM_IRQCHIP_PIC_SLAVE:
4177 spin_lock(&pic->lock);
4178 memcpy(&pic->pics[1], &chip->chip.pic,
4179 sizeof(struct kvm_pic_state));
4180 spin_unlock(&pic->lock);
4181 break;
4182 case KVM_IRQCHIP_IOAPIC:
4183 kvm_set_ioapic(kvm, &chip->chip.ioapic);
4184 break;
4185 default:
4186 r = -EINVAL;
4187 break;
4189 kvm_pic_update_irq(pic);
4190 return r;
4193 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4195 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4197 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4199 mutex_lock(&kps->lock);
4200 memcpy(ps, &kps->channels, sizeof(*ps));
4201 mutex_unlock(&kps->lock);
4202 return 0;
4205 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4207 int i;
4208 struct kvm_pit *pit = kvm->arch.vpit;
4210 mutex_lock(&pit->pit_state.lock);
4211 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4212 for (i = 0; i < 3; i++)
4213 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4214 mutex_unlock(&pit->pit_state.lock);
4215 return 0;
4218 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4220 mutex_lock(&kvm->arch.vpit->pit_state.lock);
4221 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4222 sizeof(ps->channels));
4223 ps->flags = kvm->arch.vpit->pit_state.flags;
4224 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4225 memset(&ps->reserved, 0, sizeof(ps->reserved));
4226 return 0;
4229 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4231 int start = 0;
4232 int i;
4233 u32 prev_legacy, cur_legacy;
4234 struct kvm_pit *pit = kvm->arch.vpit;
4236 mutex_lock(&pit->pit_state.lock);
4237 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4238 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4239 if (!prev_legacy && cur_legacy)
4240 start = 1;
4241 memcpy(&pit->pit_state.channels, &ps->channels,
4242 sizeof(pit->pit_state.channels));
4243 pit->pit_state.flags = ps->flags;
4244 for (i = 0; i < 3; i++)
4245 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4246 start && i == 0);
4247 mutex_unlock(&pit->pit_state.lock);
4248 return 0;
4251 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4252 struct kvm_reinject_control *control)
4254 struct kvm_pit *pit = kvm->arch.vpit;
4256 if (!pit)
4257 return -ENXIO;
4259 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4260 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4261 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4263 mutex_lock(&pit->pit_state.lock);
4264 kvm_pit_set_reinject(pit, control->pit_reinject);
4265 mutex_unlock(&pit->pit_state.lock);
4267 return 0;
4271 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4272 * @kvm: kvm instance
4273 * @log: slot id and address to which we copy the log
4275 * Steps 1-4 below provide general overview of dirty page logging. See
4276 * kvm_get_dirty_log_protect() function description for additional details.
4278 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4279 * always flush the TLB (step 4) even if previous step failed and the dirty
4280 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4281 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4282 * writes will be marked dirty for next log read.
4284 * 1. Take a snapshot of the bit and clear it if needed.
4285 * 2. Write protect the corresponding page.
4286 * 3. Copy the snapshot to the userspace.
4287 * 4. Flush TLB's if needed.
4289 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4291 bool is_dirty = false;
4292 int r;
4294 mutex_lock(&kvm->slots_lock);
4297 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4299 if (kvm_x86_ops->flush_log_dirty)
4300 kvm_x86_ops->flush_log_dirty(kvm);
4302 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
4305 * All the TLBs can be flushed out of mmu lock, see the comments in
4306 * kvm_mmu_slot_remove_write_access().
4308 lockdep_assert_held(&kvm->slots_lock);
4309 if (is_dirty)
4310 kvm_flush_remote_tlbs(kvm);
4312 mutex_unlock(&kvm->slots_lock);
4313 return r;
4316 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4317 bool line_status)
4319 if (!irqchip_in_kernel(kvm))
4320 return -ENXIO;
4322 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4323 irq_event->irq, irq_event->level,
4324 line_status);
4325 return 0;
4328 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4329 struct kvm_enable_cap *cap)
4331 int r;
4333 if (cap->flags)
4334 return -EINVAL;
4336 switch (cap->cap) {
4337 case KVM_CAP_DISABLE_QUIRKS:
4338 kvm->arch.disabled_quirks = cap->args[0];
4339 r = 0;
4340 break;
4341 case KVM_CAP_SPLIT_IRQCHIP: {
4342 mutex_lock(&kvm->lock);
4343 r = -EINVAL;
4344 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4345 goto split_irqchip_unlock;
4346 r = -EEXIST;
4347 if (irqchip_in_kernel(kvm))
4348 goto split_irqchip_unlock;
4349 if (kvm->created_vcpus)
4350 goto split_irqchip_unlock;
4351 r = kvm_setup_empty_irq_routing(kvm);
4352 if (r)
4353 goto split_irqchip_unlock;
4354 /* Pairs with irqchip_in_kernel. */
4355 smp_wmb();
4356 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4357 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4358 r = 0;
4359 split_irqchip_unlock:
4360 mutex_unlock(&kvm->lock);
4361 break;
4363 case KVM_CAP_X2APIC_API:
4364 r = -EINVAL;
4365 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4366 break;
4368 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4369 kvm->arch.x2apic_format = true;
4370 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4371 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4373 r = 0;
4374 break;
4375 case KVM_CAP_X86_DISABLE_EXITS:
4376 r = -EINVAL;
4377 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4378 break;
4380 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4381 kvm_can_mwait_in_guest())
4382 kvm->arch.mwait_in_guest = true;
4383 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4384 kvm->arch.hlt_in_guest = true;
4385 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4386 kvm->arch.pause_in_guest = true;
4387 r = 0;
4388 break;
4389 case KVM_CAP_MSR_PLATFORM_INFO:
4390 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4391 r = 0;
4392 break;
4393 default:
4394 r = -EINVAL;
4395 break;
4397 return r;
4400 long kvm_arch_vm_ioctl(struct file *filp,
4401 unsigned int ioctl, unsigned long arg)
4403 struct kvm *kvm = filp->private_data;
4404 void __user *argp = (void __user *)arg;
4405 int r = -ENOTTY;
4407 * This union makes it completely explicit to gcc-3.x
4408 * that these two variables' stack usage should be
4409 * combined, not added together.
4411 union {
4412 struct kvm_pit_state ps;
4413 struct kvm_pit_state2 ps2;
4414 struct kvm_pit_config pit_config;
4415 } u;
4417 switch (ioctl) {
4418 case KVM_SET_TSS_ADDR:
4419 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4420 break;
4421 case KVM_SET_IDENTITY_MAP_ADDR: {
4422 u64 ident_addr;
4424 mutex_lock(&kvm->lock);
4425 r = -EINVAL;
4426 if (kvm->created_vcpus)
4427 goto set_identity_unlock;
4428 r = -EFAULT;
4429 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
4430 goto set_identity_unlock;
4431 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4432 set_identity_unlock:
4433 mutex_unlock(&kvm->lock);
4434 break;
4436 case KVM_SET_NR_MMU_PAGES:
4437 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4438 break;
4439 case KVM_GET_NR_MMU_PAGES:
4440 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4441 break;
4442 case KVM_CREATE_IRQCHIP: {
4443 mutex_lock(&kvm->lock);
4445 r = -EEXIST;
4446 if (irqchip_in_kernel(kvm))
4447 goto create_irqchip_unlock;
4449 r = -EINVAL;
4450 if (kvm->created_vcpus)
4451 goto create_irqchip_unlock;
4453 r = kvm_pic_init(kvm);
4454 if (r)
4455 goto create_irqchip_unlock;
4457 r = kvm_ioapic_init(kvm);
4458 if (r) {
4459 kvm_pic_destroy(kvm);
4460 goto create_irqchip_unlock;
4463 r = kvm_setup_default_irq_routing(kvm);
4464 if (r) {
4465 kvm_ioapic_destroy(kvm);
4466 kvm_pic_destroy(kvm);
4467 goto create_irqchip_unlock;
4469 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4470 smp_wmb();
4471 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4472 create_irqchip_unlock:
4473 mutex_unlock(&kvm->lock);
4474 break;
4476 case KVM_CREATE_PIT:
4477 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4478 goto create_pit;
4479 case KVM_CREATE_PIT2:
4480 r = -EFAULT;
4481 if (copy_from_user(&u.pit_config, argp,
4482 sizeof(struct kvm_pit_config)))
4483 goto out;
4484 create_pit:
4485 mutex_lock(&kvm->lock);
4486 r = -EEXIST;
4487 if (kvm->arch.vpit)
4488 goto create_pit_unlock;
4489 r = -ENOMEM;
4490 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4491 if (kvm->arch.vpit)
4492 r = 0;
4493 create_pit_unlock:
4494 mutex_unlock(&kvm->lock);
4495 break;
4496 case KVM_GET_IRQCHIP: {
4497 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4498 struct kvm_irqchip *chip;
4500 chip = memdup_user(argp, sizeof(*chip));
4501 if (IS_ERR(chip)) {
4502 r = PTR_ERR(chip);
4503 goto out;
4506 r = -ENXIO;
4507 if (!irqchip_kernel(kvm))
4508 goto get_irqchip_out;
4509 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4510 if (r)
4511 goto get_irqchip_out;
4512 r = -EFAULT;
4513 if (copy_to_user(argp, chip, sizeof *chip))
4514 goto get_irqchip_out;
4515 r = 0;
4516 get_irqchip_out:
4517 kfree(chip);
4518 break;
4520 case KVM_SET_IRQCHIP: {
4521 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4522 struct kvm_irqchip *chip;
4524 chip = memdup_user(argp, sizeof(*chip));
4525 if (IS_ERR(chip)) {
4526 r = PTR_ERR(chip);
4527 goto out;
4530 r = -ENXIO;
4531 if (!irqchip_kernel(kvm))
4532 goto set_irqchip_out;
4533 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4534 if (r)
4535 goto set_irqchip_out;
4536 r = 0;
4537 set_irqchip_out:
4538 kfree(chip);
4539 break;
4541 case KVM_GET_PIT: {
4542 r = -EFAULT;
4543 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4544 goto out;
4545 r = -ENXIO;
4546 if (!kvm->arch.vpit)
4547 goto out;
4548 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4549 if (r)
4550 goto out;
4551 r = -EFAULT;
4552 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4553 goto out;
4554 r = 0;
4555 break;
4557 case KVM_SET_PIT: {
4558 r = -EFAULT;
4559 if (copy_from_user(&u.ps, argp, sizeof u.ps))
4560 goto out;
4561 r = -ENXIO;
4562 if (!kvm->arch.vpit)
4563 goto out;
4564 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4565 break;
4567 case KVM_GET_PIT2: {
4568 r = -ENXIO;
4569 if (!kvm->arch.vpit)
4570 goto out;
4571 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4572 if (r)
4573 goto out;
4574 r = -EFAULT;
4575 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4576 goto out;
4577 r = 0;
4578 break;
4580 case KVM_SET_PIT2: {
4581 r = -EFAULT;
4582 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4583 goto out;
4584 r = -ENXIO;
4585 if (!kvm->arch.vpit)
4586 goto out;
4587 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4588 break;
4590 case KVM_REINJECT_CONTROL: {
4591 struct kvm_reinject_control control;
4592 r = -EFAULT;
4593 if (copy_from_user(&control, argp, sizeof(control)))
4594 goto out;
4595 r = kvm_vm_ioctl_reinject(kvm, &control);
4596 break;
4598 case KVM_SET_BOOT_CPU_ID:
4599 r = 0;
4600 mutex_lock(&kvm->lock);
4601 if (kvm->created_vcpus)
4602 r = -EBUSY;
4603 else
4604 kvm->arch.bsp_vcpu_id = arg;
4605 mutex_unlock(&kvm->lock);
4606 break;
4607 case KVM_XEN_HVM_CONFIG: {
4608 struct kvm_xen_hvm_config xhc;
4609 r = -EFAULT;
4610 if (copy_from_user(&xhc, argp, sizeof(xhc)))
4611 goto out;
4612 r = -EINVAL;
4613 if (xhc.flags)
4614 goto out;
4615 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
4616 r = 0;
4617 break;
4619 case KVM_SET_CLOCK: {
4620 struct kvm_clock_data user_ns;
4621 u64 now_ns;
4623 r = -EFAULT;
4624 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4625 goto out;
4627 r = -EINVAL;
4628 if (user_ns.flags)
4629 goto out;
4631 r = 0;
4633 * TODO: userspace has to take care of races with VCPU_RUN, so
4634 * kvm_gen_update_masterclock() can be cut down to locked
4635 * pvclock_update_vm_gtod_copy().
4637 kvm_gen_update_masterclock(kvm);
4638 now_ns = get_kvmclock_ns(kvm);
4639 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4640 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4641 break;
4643 case KVM_GET_CLOCK: {
4644 struct kvm_clock_data user_ns;
4645 u64 now_ns;
4647 now_ns = get_kvmclock_ns(kvm);
4648 user_ns.clock = now_ns;
4649 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4650 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4652 r = -EFAULT;
4653 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4654 goto out;
4655 r = 0;
4656 break;
4658 case KVM_ENABLE_CAP: {
4659 struct kvm_enable_cap cap;
4661 r = -EFAULT;
4662 if (copy_from_user(&cap, argp, sizeof(cap)))
4663 goto out;
4664 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4665 break;
4667 case KVM_MEMORY_ENCRYPT_OP: {
4668 r = -ENOTTY;
4669 if (kvm_x86_ops->mem_enc_op)
4670 r = kvm_x86_ops->mem_enc_op(kvm, argp);
4671 break;
4673 case KVM_MEMORY_ENCRYPT_REG_REGION: {
4674 struct kvm_enc_region region;
4676 r = -EFAULT;
4677 if (copy_from_user(&region, argp, sizeof(region)))
4678 goto out;
4680 r = -ENOTTY;
4681 if (kvm_x86_ops->mem_enc_reg_region)
4682 r = kvm_x86_ops->mem_enc_reg_region(kvm, &region);
4683 break;
4685 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
4686 struct kvm_enc_region region;
4688 r = -EFAULT;
4689 if (copy_from_user(&region, argp, sizeof(region)))
4690 goto out;
4692 r = -ENOTTY;
4693 if (kvm_x86_ops->mem_enc_unreg_region)
4694 r = kvm_x86_ops->mem_enc_unreg_region(kvm, &region);
4695 break;
4697 case KVM_HYPERV_EVENTFD: {
4698 struct kvm_hyperv_eventfd hvevfd;
4700 r = -EFAULT;
4701 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
4702 goto out;
4703 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
4704 break;
4706 default:
4707 r = -ENOTTY;
4709 out:
4710 return r;
4713 static void kvm_init_msr_list(void)
4715 u32 dummy[2];
4716 unsigned i, j;
4718 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4719 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4720 continue;
4723 * Even MSRs that are valid in the host may not be exposed
4724 * to the guests in some cases.
4726 switch (msrs_to_save[i]) {
4727 case MSR_IA32_BNDCFGS:
4728 if (!kvm_mpx_supported())
4729 continue;
4730 break;
4731 case MSR_TSC_AUX:
4732 if (!kvm_x86_ops->rdtscp_supported())
4733 continue;
4734 break;
4735 default:
4736 break;
4739 if (j < i)
4740 msrs_to_save[j] = msrs_to_save[i];
4741 j++;
4743 num_msrs_to_save = j;
4745 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4746 if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
4747 continue;
4749 if (j < i)
4750 emulated_msrs[j] = emulated_msrs[i];
4751 j++;
4753 num_emulated_msrs = j;
4755 for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
4756 struct kvm_msr_entry msr;
4758 msr.index = msr_based_features[i];
4759 if (kvm_get_msr_feature(&msr))
4760 continue;
4762 if (j < i)
4763 msr_based_features[j] = msr_based_features[i];
4764 j++;
4766 num_msr_based_features = j;
4769 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4770 const void *v)
4772 int handled = 0;
4773 int n;
4775 do {
4776 n = min(len, 8);
4777 if (!(lapic_in_kernel(vcpu) &&
4778 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4779 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4780 break;
4781 handled += n;
4782 addr += n;
4783 len -= n;
4784 v += n;
4785 } while (len);
4787 return handled;
4790 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4792 int handled = 0;
4793 int n;
4795 do {
4796 n = min(len, 8);
4797 if (!(lapic_in_kernel(vcpu) &&
4798 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4799 addr, n, v))
4800 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4801 break;
4802 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
4803 handled += n;
4804 addr += n;
4805 len -= n;
4806 v += n;
4807 } while (len);
4809 return handled;
4812 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4813 struct kvm_segment *var, int seg)
4815 kvm_x86_ops->set_segment(vcpu, var, seg);
4818 void kvm_get_segment(struct kvm_vcpu *vcpu,
4819 struct kvm_segment *var, int seg)
4821 kvm_x86_ops->get_segment(vcpu, var, seg);
4824 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4825 struct x86_exception *exception)
4827 gpa_t t_gpa;
4829 BUG_ON(!mmu_is_nested(vcpu));
4831 /* NPT walks are always user-walks */
4832 access |= PFERR_USER_MASK;
4833 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4835 return t_gpa;
4838 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4839 struct x86_exception *exception)
4841 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4842 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4845 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4846 struct x86_exception *exception)
4848 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4849 access |= PFERR_FETCH_MASK;
4850 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4853 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4854 struct x86_exception *exception)
4856 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4857 access |= PFERR_WRITE_MASK;
4858 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4861 /* uses this to access any guest's mapped memory without checking CPL */
4862 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4863 struct x86_exception *exception)
4865 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4868 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4869 struct kvm_vcpu *vcpu, u32 access,
4870 struct x86_exception *exception)
4872 void *data = val;
4873 int r = X86EMUL_CONTINUE;
4875 while (bytes) {
4876 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4877 exception);
4878 unsigned offset = addr & (PAGE_SIZE-1);
4879 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4880 int ret;
4882 if (gpa == UNMAPPED_GVA)
4883 return X86EMUL_PROPAGATE_FAULT;
4884 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4885 offset, toread);
4886 if (ret < 0) {
4887 r = X86EMUL_IO_NEEDED;
4888 goto out;
4891 bytes -= toread;
4892 data += toread;
4893 addr += toread;
4895 out:
4896 return r;
4899 /* used for instruction fetching */
4900 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4901 gva_t addr, void *val, unsigned int bytes,
4902 struct x86_exception *exception)
4904 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4905 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4906 unsigned offset;
4907 int ret;
4909 /* Inline kvm_read_guest_virt_helper for speed. */
4910 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4911 exception);
4912 if (unlikely(gpa == UNMAPPED_GVA))
4913 return X86EMUL_PROPAGATE_FAULT;
4915 offset = addr & (PAGE_SIZE-1);
4916 if (WARN_ON(offset + bytes > PAGE_SIZE))
4917 bytes = (unsigned)PAGE_SIZE - offset;
4918 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4919 offset, bytes);
4920 if (unlikely(ret < 0))
4921 return X86EMUL_IO_NEEDED;
4923 return X86EMUL_CONTINUE;
4926 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
4927 gva_t addr, void *val, unsigned int bytes,
4928 struct x86_exception *exception)
4930 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4933 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
4934 * is returned, but our callers are not ready for that and they blindly
4935 * call kvm_inject_page_fault. Ensure that they at least do not leak
4936 * uninitialized kernel stack memory into cr2 and error code.
4938 memset(exception, 0, sizeof(*exception));
4939 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4940 exception);
4942 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4944 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
4945 gva_t addr, void *val, unsigned int bytes,
4946 struct x86_exception *exception, bool system)
4948 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4949 u32 access = 0;
4951 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
4952 access |= PFERR_USER_MASK;
4954 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
4957 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4958 unsigned long addr, void *val, unsigned int bytes)
4960 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4961 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4963 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4966 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4967 struct kvm_vcpu *vcpu, u32 access,
4968 struct x86_exception *exception)
4970 void *data = val;
4971 int r = X86EMUL_CONTINUE;
4973 while (bytes) {
4974 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4975 access,
4976 exception);
4977 unsigned offset = addr & (PAGE_SIZE-1);
4978 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4979 int ret;
4981 if (gpa == UNMAPPED_GVA)
4982 return X86EMUL_PROPAGATE_FAULT;
4983 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4984 if (ret < 0) {
4985 r = X86EMUL_IO_NEEDED;
4986 goto out;
4989 bytes -= towrite;
4990 data += towrite;
4991 addr += towrite;
4993 out:
4994 return r;
4997 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
4998 unsigned int bytes, struct x86_exception *exception,
4999 bool system)
5001 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5002 u32 access = PFERR_WRITE_MASK;
5004 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5005 access |= PFERR_USER_MASK;
5007 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5008 access, exception);
5011 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5012 unsigned int bytes, struct x86_exception *exception)
5014 /* kvm_write_guest_virt_system can pull in tons of pages. */
5015 vcpu->arch.l1tf_flush_l1d = true;
5017 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5018 PFERR_WRITE_MASK, exception);
5020 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5022 int handle_ud(struct kvm_vcpu *vcpu)
5024 int emul_type = EMULTYPE_TRAP_UD;
5025 enum emulation_result er;
5026 char sig[5]; /* ud2; .ascii "kvm" */
5027 struct x86_exception e;
5029 if (force_emulation_prefix &&
5030 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5031 sig, sizeof(sig), &e) == 0 &&
5032 memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
5033 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5034 emul_type = 0;
5037 er = kvm_emulate_instruction(vcpu, emul_type);
5038 if (er == EMULATE_USER_EXIT)
5039 return 0;
5040 if (er != EMULATE_DONE)
5041 kvm_queue_exception(vcpu, UD_VECTOR);
5042 return 1;
5044 EXPORT_SYMBOL_GPL(handle_ud);
5046 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5047 gpa_t gpa, bool write)
5049 /* For APIC access vmexit */
5050 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5051 return 1;
5053 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5054 trace_vcpu_match_mmio(gva, gpa, write, true);
5055 return 1;
5058 return 0;
5061 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5062 gpa_t *gpa, struct x86_exception *exception,
5063 bool write)
5065 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5066 | (write ? PFERR_WRITE_MASK : 0);
5069 * currently PKRU is only applied to ept enabled guest so
5070 * there is no pkey in EPT page table for L1 guest or EPT
5071 * shadow page table for L2 guest.
5073 if (vcpu_match_mmio_gva(vcpu, gva)
5074 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5075 vcpu->arch.access, 0, access)) {
5076 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5077 (gva & (PAGE_SIZE - 1));
5078 trace_vcpu_match_mmio(gva, *gpa, write, false);
5079 return 1;
5082 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5084 if (*gpa == UNMAPPED_GVA)
5085 return -1;
5087 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5090 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5091 const void *val, int bytes)
5093 int ret;
5095 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5096 if (ret < 0)
5097 return 0;
5098 kvm_page_track_write(vcpu, gpa, val, bytes);
5099 return 1;
5102 struct read_write_emulator_ops {
5103 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5104 int bytes);
5105 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5106 void *val, int bytes);
5107 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5108 int bytes, void *val);
5109 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5110 void *val, int bytes);
5111 bool write;
5114 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5116 if (vcpu->mmio_read_completed) {
5117 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5118 vcpu->mmio_fragments[0].gpa, val);
5119 vcpu->mmio_read_completed = 0;
5120 return 1;
5123 return 0;
5126 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5127 void *val, int bytes)
5129 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5132 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5133 void *val, int bytes)
5135 return emulator_write_phys(vcpu, gpa, val, bytes);
5138 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5140 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5141 return vcpu_mmio_write(vcpu, gpa, bytes, val);
5144 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5145 void *val, int bytes)
5147 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5148 return X86EMUL_IO_NEEDED;
5151 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5152 void *val, int bytes)
5154 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5156 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5157 return X86EMUL_CONTINUE;
5160 static const struct read_write_emulator_ops read_emultor = {
5161 .read_write_prepare = read_prepare,
5162 .read_write_emulate = read_emulate,
5163 .read_write_mmio = vcpu_mmio_read,
5164 .read_write_exit_mmio = read_exit_mmio,
5167 static const struct read_write_emulator_ops write_emultor = {
5168 .read_write_emulate = write_emulate,
5169 .read_write_mmio = write_mmio,
5170 .read_write_exit_mmio = write_exit_mmio,
5171 .write = true,
5174 static int emulator_read_write_onepage(unsigned long addr, void *val,
5175 unsigned int bytes,
5176 struct x86_exception *exception,
5177 struct kvm_vcpu *vcpu,
5178 const struct read_write_emulator_ops *ops)
5180 gpa_t gpa;
5181 int handled, ret;
5182 bool write = ops->write;
5183 struct kvm_mmio_fragment *frag;
5184 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5187 * If the exit was due to a NPF we may already have a GPA.
5188 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5189 * Note, this cannot be used on string operations since string
5190 * operation using rep will only have the initial GPA from the NPF
5191 * occurred.
5193 if (vcpu->arch.gpa_available &&
5194 emulator_can_use_gpa(ctxt) &&
5195 (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5196 gpa = vcpu->arch.gpa_val;
5197 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5198 } else {
5199 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5200 if (ret < 0)
5201 return X86EMUL_PROPAGATE_FAULT;
5204 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5205 return X86EMUL_CONTINUE;
5208 * Is this MMIO handled locally?
5210 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5211 if (handled == bytes)
5212 return X86EMUL_CONTINUE;
5214 gpa += handled;
5215 bytes -= handled;
5216 val += handled;
5218 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5219 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5220 frag->gpa = gpa;
5221 frag->data = val;
5222 frag->len = bytes;
5223 return X86EMUL_CONTINUE;
5226 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5227 unsigned long addr,
5228 void *val, unsigned int bytes,
5229 struct x86_exception *exception,
5230 const struct read_write_emulator_ops *ops)
5232 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5233 gpa_t gpa;
5234 int rc;
5236 if (ops->read_write_prepare &&
5237 ops->read_write_prepare(vcpu, val, bytes))
5238 return X86EMUL_CONTINUE;
5240 vcpu->mmio_nr_fragments = 0;
5242 /* Crossing a page boundary? */
5243 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5244 int now;
5246 now = -addr & ~PAGE_MASK;
5247 rc = emulator_read_write_onepage(addr, val, now, exception,
5248 vcpu, ops);
5250 if (rc != X86EMUL_CONTINUE)
5251 return rc;
5252 addr += now;
5253 if (ctxt->mode != X86EMUL_MODE_PROT64)
5254 addr = (u32)addr;
5255 val += now;
5256 bytes -= now;
5259 rc = emulator_read_write_onepage(addr, val, bytes, exception,
5260 vcpu, ops);
5261 if (rc != X86EMUL_CONTINUE)
5262 return rc;
5264 if (!vcpu->mmio_nr_fragments)
5265 return rc;
5267 gpa = vcpu->mmio_fragments[0].gpa;
5269 vcpu->mmio_needed = 1;
5270 vcpu->mmio_cur_fragment = 0;
5272 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5273 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5274 vcpu->run->exit_reason = KVM_EXIT_MMIO;
5275 vcpu->run->mmio.phys_addr = gpa;
5277 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5280 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5281 unsigned long addr,
5282 void *val,
5283 unsigned int bytes,
5284 struct x86_exception *exception)
5286 return emulator_read_write(ctxt, addr, val, bytes,
5287 exception, &read_emultor);
5290 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5291 unsigned long addr,
5292 const void *val,
5293 unsigned int bytes,
5294 struct x86_exception *exception)
5296 return emulator_read_write(ctxt, addr, (void *)val, bytes,
5297 exception, &write_emultor);
5300 #define CMPXCHG_TYPE(t, ptr, old, new) \
5301 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5303 #ifdef CONFIG_X86_64
5304 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5305 #else
5306 # define CMPXCHG64(ptr, old, new) \
5307 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5308 #endif
5310 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5311 unsigned long addr,
5312 const void *old,
5313 const void *new,
5314 unsigned int bytes,
5315 struct x86_exception *exception)
5317 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5318 gpa_t gpa;
5319 struct page *page;
5320 char *kaddr;
5321 bool exchanged;
5323 /* guests cmpxchg8b have to be emulated atomically */
5324 if (bytes > 8 || (bytes & (bytes - 1)))
5325 goto emul_write;
5327 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5329 if (gpa == UNMAPPED_GVA ||
5330 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5331 goto emul_write;
5333 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5334 goto emul_write;
5336 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
5337 if (is_error_page(page))
5338 goto emul_write;
5340 kaddr = kmap_atomic(page);
5341 kaddr += offset_in_page(gpa);
5342 switch (bytes) {
5343 case 1:
5344 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5345 break;
5346 case 2:
5347 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5348 break;
5349 case 4:
5350 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5351 break;
5352 case 8:
5353 exchanged = CMPXCHG64(kaddr, old, new);
5354 break;
5355 default:
5356 BUG();
5358 kunmap_atomic(kaddr);
5359 kvm_release_page_dirty(page);
5361 if (!exchanged)
5362 return X86EMUL_CMPXCHG_FAILED;
5364 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5365 kvm_page_track_write(vcpu, gpa, new, bytes);
5367 return X86EMUL_CONTINUE;
5369 emul_write:
5370 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5372 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5375 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5377 int r = 0, i;
5379 for (i = 0; i < vcpu->arch.pio.count; i++) {
5380 if (vcpu->arch.pio.in)
5381 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5382 vcpu->arch.pio.size, pd);
5383 else
5384 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5385 vcpu->arch.pio.port, vcpu->arch.pio.size,
5386 pd);
5387 if (r)
5388 break;
5389 pd += vcpu->arch.pio.size;
5391 return r;
5394 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5395 unsigned short port, void *val,
5396 unsigned int count, bool in)
5398 vcpu->arch.pio.port = port;
5399 vcpu->arch.pio.in = in;
5400 vcpu->arch.pio.count = count;
5401 vcpu->arch.pio.size = size;
5403 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5404 vcpu->arch.pio.count = 0;
5405 return 1;
5408 vcpu->run->exit_reason = KVM_EXIT_IO;
5409 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5410 vcpu->run->io.size = size;
5411 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5412 vcpu->run->io.count = count;
5413 vcpu->run->io.port = port;
5415 return 0;
5418 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5419 int size, unsigned short port, void *val,
5420 unsigned int count)
5422 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5423 int ret;
5425 if (vcpu->arch.pio.count)
5426 goto data_avail;
5428 memset(vcpu->arch.pio_data, 0, size * count);
5430 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5431 if (ret) {
5432 data_avail:
5433 memcpy(val, vcpu->arch.pio_data, size * count);
5434 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5435 vcpu->arch.pio.count = 0;
5436 return 1;
5439 return 0;
5442 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5443 int size, unsigned short port,
5444 const void *val, unsigned int count)
5446 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5448 memcpy(vcpu->arch.pio_data, val, size * count);
5449 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5450 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5453 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5455 return kvm_x86_ops->get_segment_base(vcpu, seg);
5458 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5460 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5463 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5465 if (!need_emulate_wbinvd(vcpu))
5466 return X86EMUL_CONTINUE;
5468 if (kvm_x86_ops->has_wbinvd_exit()) {
5469 int cpu = get_cpu();
5471 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5472 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5473 wbinvd_ipi, NULL, 1);
5474 put_cpu();
5475 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5476 } else
5477 wbinvd();
5478 return X86EMUL_CONTINUE;
5481 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5483 kvm_emulate_wbinvd_noskip(vcpu);
5484 return kvm_skip_emulated_instruction(vcpu);
5486 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5490 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5492 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5495 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5496 unsigned long *dest)
5498 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5501 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5502 unsigned long value)
5505 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5508 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5510 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5513 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5515 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5516 unsigned long value;
5518 switch (cr) {
5519 case 0:
5520 value = kvm_read_cr0(vcpu);
5521 break;
5522 case 2:
5523 value = vcpu->arch.cr2;
5524 break;
5525 case 3:
5526 value = kvm_read_cr3(vcpu);
5527 break;
5528 case 4:
5529 value = kvm_read_cr4(vcpu);
5530 break;
5531 case 8:
5532 value = kvm_get_cr8(vcpu);
5533 break;
5534 default:
5535 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5536 return 0;
5539 return value;
5542 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5544 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5545 int res = 0;
5547 switch (cr) {
5548 case 0:
5549 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5550 break;
5551 case 2:
5552 vcpu->arch.cr2 = val;
5553 break;
5554 case 3:
5555 res = kvm_set_cr3(vcpu, val);
5556 break;
5557 case 4:
5558 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5559 break;
5560 case 8:
5561 res = kvm_set_cr8(vcpu, val);
5562 break;
5563 default:
5564 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5565 res = -1;
5568 return res;
5571 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5573 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5576 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5578 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5581 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5583 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5586 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5588 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5591 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5593 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5596 static unsigned long emulator_get_cached_segment_base(
5597 struct x86_emulate_ctxt *ctxt, int seg)
5599 return get_segment_base(emul_to_vcpu(ctxt), seg);
5602 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5603 struct desc_struct *desc, u32 *base3,
5604 int seg)
5606 struct kvm_segment var;
5608 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5609 *selector = var.selector;
5611 if (var.unusable) {
5612 memset(desc, 0, sizeof(*desc));
5613 if (base3)
5614 *base3 = 0;
5615 return false;
5618 if (var.g)
5619 var.limit >>= 12;
5620 set_desc_limit(desc, var.limit);
5621 set_desc_base(desc, (unsigned long)var.base);
5622 #ifdef CONFIG_X86_64
5623 if (base3)
5624 *base3 = var.base >> 32;
5625 #endif
5626 desc->type = var.type;
5627 desc->s = var.s;
5628 desc->dpl = var.dpl;
5629 desc->p = var.present;
5630 desc->avl = var.avl;
5631 desc->l = var.l;
5632 desc->d = var.db;
5633 desc->g = var.g;
5635 return true;
5638 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5639 struct desc_struct *desc, u32 base3,
5640 int seg)
5642 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5643 struct kvm_segment var;
5645 var.selector = selector;
5646 var.base = get_desc_base(desc);
5647 #ifdef CONFIG_X86_64
5648 var.base |= ((u64)base3) << 32;
5649 #endif
5650 var.limit = get_desc_limit(desc);
5651 if (desc->g)
5652 var.limit = (var.limit << 12) | 0xfff;
5653 var.type = desc->type;
5654 var.dpl = desc->dpl;
5655 var.db = desc->d;
5656 var.s = desc->s;
5657 var.l = desc->l;
5658 var.g = desc->g;
5659 var.avl = desc->avl;
5660 var.present = desc->p;
5661 var.unusable = !var.present;
5662 var.padding = 0;
5664 kvm_set_segment(vcpu, &var, seg);
5665 return;
5668 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5669 u32 msr_index, u64 *pdata)
5671 struct msr_data msr;
5672 int r;
5674 msr.index = msr_index;
5675 msr.host_initiated = false;
5676 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5677 if (r)
5678 return r;
5680 *pdata = msr.data;
5681 return 0;
5684 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5685 u32 msr_index, u64 data)
5687 struct msr_data msr;
5689 msr.data = data;
5690 msr.index = msr_index;
5691 msr.host_initiated = false;
5692 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5695 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5697 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5699 return vcpu->arch.smbase;
5702 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5704 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5706 vcpu->arch.smbase = smbase;
5709 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5710 u32 pmc)
5712 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5715 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5716 u32 pmc, u64 *pdata)
5718 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5721 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5723 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5726 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5727 struct x86_instruction_info *info,
5728 enum x86_intercept_stage stage)
5730 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5733 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5734 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
5736 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
5739 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5741 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5744 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5746 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5749 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5751 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5754 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5756 return emul_to_vcpu(ctxt)->arch.hflags;
5759 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5761 kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5764 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, u64 smbase)
5766 return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smbase);
5769 static const struct x86_emulate_ops emulate_ops = {
5770 .read_gpr = emulator_read_gpr,
5771 .write_gpr = emulator_write_gpr,
5772 .read_std = emulator_read_std,
5773 .write_std = emulator_write_std,
5774 .read_phys = kvm_read_guest_phys_system,
5775 .fetch = kvm_fetch_guest_virt,
5776 .read_emulated = emulator_read_emulated,
5777 .write_emulated = emulator_write_emulated,
5778 .cmpxchg_emulated = emulator_cmpxchg_emulated,
5779 .invlpg = emulator_invlpg,
5780 .pio_in_emulated = emulator_pio_in_emulated,
5781 .pio_out_emulated = emulator_pio_out_emulated,
5782 .get_segment = emulator_get_segment,
5783 .set_segment = emulator_set_segment,
5784 .get_cached_segment_base = emulator_get_cached_segment_base,
5785 .get_gdt = emulator_get_gdt,
5786 .get_idt = emulator_get_idt,
5787 .set_gdt = emulator_set_gdt,
5788 .set_idt = emulator_set_idt,
5789 .get_cr = emulator_get_cr,
5790 .set_cr = emulator_set_cr,
5791 .cpl = emulator_get_cpl,
5792 .get_dr = emulator_get_dr,
5793 .set_dr = emulator_set_dr,
5794 .get_smbase = emulator_get_smbase,
5795 .set_smbase = emulator_set_smbase,
5796 .set_msr = emulator_set_msr,
5797 .get_msr = emulator_get_msr,
5798 .check_pmc = emulator_check_pmc,
5799 .read_pmc = emulator_read_pmc,
5800 .halt = emulator_halt,
5801 .wbinvd = emulator_wbinvd,
5802 .fix_hypercall = emulator_fix_hypercall,
5803 .intercept = emulator_intercept,
5804 .get_cpuid = emulator_get_cpuid,
5805 .set_nmi_mask = emulator_set_nmi_mask,
5806 .get_hflags = emulator_get_hflags,
5807 .set_hflags = emulator_set_hflags,
5808 .pre_leave_smm = emulator_pre_leave_smm,
5811 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5813 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5815 * an sti; sti; sequence only disable interrupts for the first
5816 * instruction. So, if the last instruction, be it emulated or
5817 * not, left the system with the INT_STI flag enabled, it
5818 * means that the last instruction is an sti. We should not
5819 * leave the flag on in this case. The same goes for mov ss
5821 if (int_shadow & mask)
5822 mask = 0;
5823 if (unlikely(int_shadow || mask)) {
5824 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5825 if (!mask)
5826 kvm_make_request(KVM_REQ_EVENT, vcpu);
5830 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5832 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5833 if (ctxt->exception.vector == PF_VECTOR)
5834 return kvm_propagate_fault(vcpu, &ctxt->exception);
5836 if (ctxt->exception.error_code_valid)
5837 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5838 ctxt->exception.error_code);
5839 else
5840 kvm_queue_exception(vcpu, ctxt->exception.vector);
5841 return false;
5844 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5846 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5847 int cs_db, cs_l;
5849 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5851 ctxt->eflags = kvm_get_rflags(vcpu);
5852 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
5854 ctxt->eip = kvm_rip_read(vcpu);
5855 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5856 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
5857 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
5858 cs_db ? X86EMUL_MODE_PROT32 :
5859 X86EMUL_MODE_PROT16;
5860 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5861 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5862 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5864 init_decode_cache(ctxt);
5865 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5868 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5870 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5871 int ret;
5873 init_emulate_ctxt(vcpu);
5875 ctxt->op_bytes = 2;
5876 ctxt->ad_bytes = 2;
5877 ctxt->_eip = ctxt->eip + inc_eip;
5878 ret = emulate_int_real(ctxt, irq);
5880 if (ret != X86EMUL_CONTINUE)
5881 return EMULATE_FAIL;
5883 ctxt->eip = ctxt->_eip;
5884 kvm_rip_write(vcpu, ctxt->eip);
5885 kvm_set_rflags(vcpu, ctxt->eflags);
5887 return EMULATE_DONE;
5889 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5891 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
5893 int r = EMULATE_DONE;
5895 ++vcpu->stat.insn_emulation_fail;
5896 trace_kvm_emulate_insn_failed(vcpu);
5898 if (emulation_type & EMULTYPE_NO_UD_ON_FAIL)
5899 return EMULATE_FAIL;
5901 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5902 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5903 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5904 vcpu->run->internal.ndata = 0;
5905 r = EMULATE_USER_EXIT;
5908 kvm_queue_exception(vcpu, UD_VECTOR);
5910 return r;
5913 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5914 bool write_fault_to_shadow_pgtable,
5915 int emulation_type)
5917 gpa_t gpa = cr2;
5918 kvm_pfn_t pfn;
5920 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
5921 return false;
5923 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
5924 return false;
5926 if (!vcpu->arch.mmu.direct_map) {
5928 * Write permission should be allowed since only
5929 * write access need to be emulated.
5931 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5934 * If the mapping is invalid in guest, let cpu retry
5935 * it to generate fault.
5937 if (gpa == UNMAPPED_GVA)
5938 return true;
5942 * Do not retry the unhandleable instruction if it faults on the
5943 * readonly host memory, otherwise it will goto a infinite loop:
5944 * retry instruction -> write #PF -> emulation fail -> retry
5945 * instruction -> ...
5947 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5950 * If the instruction failed on the error pfn, it can not be fixed,
5951 * report the error to userspace.
5953 if (is_error_noslot_pfn(pfn))
5954 return false;
5956 kvm_release_pfn_clean(pfn);
5958 /* The instructions are well-emulated on direct mmu. */
5959 if (vcpu->arch.mmu.direct_map) {
5960 unsigned int indirect_shadow_pages;
5962 spin_lock(&vcpu->kvm->mmu_lock);
5963 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5964 spin_unlock(&vcpu->kvm->mmu_lock);
5966 if (indirect_shadow_pages)
5967 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5969 return true;
5973 * if emulation was due to access to shadowed page table
5974 * and it failed try to unshadow page and re-enter the
5975 * guest to let CPU execute the instruction.
5977 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5980 * If the access faults on its page table, it can not
5981 * be fixed by unprotecting shadow page and it should
5982 * be reported to userspace.
5984 return !write_fault_to_shadow_pgtable;
5987 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5988 unsigned long cr2, int emulation_type)
5990 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5991 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5993 last_retry_eip = vcpu->arch.last_retry_eip;
5994 last_retry_addr = vcpu->arch.last_retry_addr;
5997 * If the emulation is caused by #PF and it is non-page_table
5998 * writing instruction, it means the VM-EXIT is caused by shadow
5999 * page protected, we can zap the shadow page and retry this
6000 * instruction directly.
6002 * Note: if the guest uses a non-page-table modifying instruction
6003 * on the PDE that points to the instruction, then we will unmap
6004 * the instruction and go to an infinite loop. So, we cache the
6005 * last retried eip and the last fault address, if we meet the eip
6006 * and the address again, we can break out of the potential infinite
6007 * loop.
6009 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6011 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6012 return false;
6014 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6015 return false;
6017 if (x86_page_table_writing_insn(ctxt))
6018 return false;
6020 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
6021 return false;
6023 vcpu->arch.last_retry_eip = ctxt->eip;
6024 vcpu->arch.last_retry_addr = cr2;
6026 if (!vcpu->arch.mmu.direct_map)
6027 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6029 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6031 return true;
6034 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6035 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6037 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6039 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6040 /* This is a good place to trace that we are exiting SMM. */
6041 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6043 /* Process a latched INIT or SMI, if any. */
6044 kvm_make_request(KVM_REQ_EVENT, vcpu);
6047 kvm_mmu_reset_context(vcpu);
6050 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
6052 unsigned changed = vcpu->arch.hflags ^ emul_flags;
6054 vcpu->arch.hflags = emul_flags;
6056 if (changed & HF_SMM_MASK)
6057 kvm_smm_changed(vcpu);
6060 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6061 unsigned long *db)
6063 u32 dr6 = 0;
6064 int i;
6065 u32 enable, rwlen;
6067 enable = dr7;
6068 rwlen = dr7 >> 16;
6069 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6070 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6071 dr6 |= (1 << i);
6072 return dr6;
6075 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
6077 struct kvm_run *kvm_run = vcpu->run;
6079 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6080 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6081 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6082 kvm_run->debug.arch.exception = DB_VECTOR;
6083 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6084 *r = EMULATE_USER_EXIT;
6085 } else {
6087 * "Certain debug exceptions may clear bit 0-3. The
6088 * remaining contents of the DR6 register are never
6089 * cleared by the processor".
6091 vcpu->arch.dr6 &= ~15;
6092 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
6093 kvm_queue_exception(vcpu, DB_VECTOR);
6097 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6099 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6100 int r = EMULATE_DONE;
6102 kvm_x86_ops->skip_emulated_instruction(vcpu);
6105 * rflags is the old, "raw" value of the flags. The new value has
6106 * not been saved yet.
6108 * This is correct even for TF set by the guest, because "the
6109 * processor will not generate this exception after the instruction
6110 * that sets the TF flag".
6112 if (unlikely(rflags & X86_EFLAGS_TF))
6113 kvm_vcpu_do_singlestep(vcpu, &r);
6114 return r == EMULATE_DONE;
6116 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6118 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6120 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6121 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6122 struct kvm_run *kvm_run = vcpu->run;
6123 unsigned long eip = kvm_get_linear_rip(vcpu);
6124 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6125 vcpu->arch.guest_debug_dr7,
6126 vcpu->arch.eff_db);
6128 if (dr6 != 0) {
6129 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6130 kvm_run->debug.arch.pc = eip;
6131 kvm_run->debug.arch.exception = DB_VECTOR;
6132 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6133 *r = EMULATE_USER_EXIT;
6134 return true;
6138 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6139 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6140 unsigned long eip = kvm_get_linear_rip(vcpu);
6141 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6142 vcpu->arch.dr7,
6143 vcpu->arch.db);
6145 if (dr6 != 0) {
6146 vcpu->arch.dr6 &= ~15;
6147 vcpu->arch.dr6 |= dr6 | DR6_RTM;
6148 kvm_queue_exception(vcpu, DB_VECTOR);
6149 *r = EMULATE_DONE;
6150 return true;
6154 return false;
6157 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6159 switch (ctxt->opcode_len) {
6160 case 1:
6161 switch (ctxt->b) {
6162 case 0xe4: /* IN */
6163 case 0xe5:
6164 case 0xec:
6165 case 0xed:
6166 case 0xe6: /* OUT */
6167 case 0xe7:
6168 case 0xee:
6169 case 0xef:
6170 case 0x6c: /* INS */
6171 case 0x6d:
6172 case 0x6e: /* OUTS */
6173 case 0x6f:
6174 return true;
6176 break;
6177 case 2:
6178 switch (ctxt->b) {
6179 case 0x33: /* RDPMC */
6180 return true;
6182 break;
6185 return false;
6188 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6189 unsigned long cr2,
6190 int emulation_type,
6191 void *insn,
6192 int insn_len)
6194 int r;
6195 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6196 bool writeback = true;
6197 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6199 vcpu->arch.l1tf_flush_l1d = true;
6202 * Clear write_fault_to_shadow_pgtable here to ensure it is
6203 * never reused.
6205 vcpu->arch.write_fault_to_shadow_pgtable = false;
6206 kvm_clear_exception_queue(vcpu);
6208 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6209 init_emulate_ctxt(vcpu);
6212 * We will reenter on the same instruction since
6213 * we do not set complete_userspace_io. This does not
6214 * handle watchpoints yet, those would be handled in
6215 * the emulate_ops.
6217 if (!(emulation_type & EMULTYPE_SKIP) &&
6218 kvm_vcpu_check_breakpoint(vcpu, &r))
6219 return r;
6221 ctxt->interruptibility = 0;
6222 ctxt->have_exception = false;
6223 ctxt->exception.vector = -1;
6224 ctxt->perm_ok = false;
6226 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6228 r = x86_decode_insn(ctxt, insn, insn_len);
6230 trace_kvm_emulate_insn_start(vcpu);
6231 ++vcpu->stat.insn_emulation;
6232 if (r != EMULATION_OK) {
6233 if (emulation_type & EMULTYPE_TRAP_UD)
6234 return EMULATE_FAIL;
6235 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6236 emulation_type))
6237 return EMULATE_DONE;
6238 if (ctxt->have_exception && inject_emulated_exception(vcpu))
6239 return EMULATE_DONE;
6240 if (emulation_type & EMULTYPE_SKIP)
6241 return EMULATE_FAIL;
6242 return handle_emulation_failure(vcpu, emulation_type);
6246 if ((emulation_type & EMULTYPE_VMWARE) &&
6247 !is_vmware_backdoor_opcode(ctxt))
6248 return EMULATE_FAIL;
6250 if (emulation_type & EMULTYPE_SKIP) {
6251 kvm_rip_write(vcpu, ctxt->_eip);
6252 if (ctxt->eflags & X86_EFLAGS_RF)
6253 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6254 return EMULATE_DONE;
6257 if (retry_instruction(ctxt, cr2, emulation_type))
6258 return EMULATE_DONE;
6260 /* this is needed for vmware backdoor interface to work since it
6261 changes registers values during IO operation */
6262 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6263 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6264 emulator_invalidate_register_cache(ctxt);
6267 restart:
6268 /* Save the faulting GPA (cr2) in the address field */
6269 ctxt->exception.address = cr2;
6271 r = x86_emulate_insn(ctxt);
6273 if (r == EMULATION_INTERCEPTED)
6274 return EMULATE_DONE;
6276 if (r == EMULATION_FAILED) {
6277 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6278 emulation_type))
6279 return EMULATE_DONE;
6281 return handle_emulation_failure(vcpu, emulation_type);
6284 if (ctxt->have_exception) {
6285 r = EMULATE_DONE;
6286 if (inject_emulated_exception(vcpu))
6287 return r;
6288 } else if (vcpu->arch.pio.count) {
6289 if (!vcpu->arch.pio.in) {
6290 /* FIXME: return into emulator if single-stepping. */
6291 vcpu->arch.pio.count = 0;
6292 } else {
6293 writeback = false;
6294 vcpu->arch.complete_userspace_io = complete_emulated_pio;
6296 r = EMULATE_USER_EXIT;
6297 } else if (vcpu->mmio_needed) {
6298 if (!vcpu->mmio_is_write)
6299 writeback = false;
6300 r = EMULATE_USER_EXIT;
6301 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6302 } else if (r == EMULATION_RESTART)
6303 goto restart;
6304 else
6305 r = EMULATE_DONE;
6307 if (writeback) {
6308 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6309 toggle_interruptibility(vcpu, ctxt->interruptibility);
6310 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6311 kvm_rip_write(vcpu, ctxt->eip);
6312 if (r == EMULATE_DONE && ctxt->tf)
6313 kvm_vcpu_do_singlestep(vcpu, &r);
6314 if (!ctxt->have_exception ||
6315 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
6316 __kvm_set_rflags(vcpu, ctxt->eflags);
6319 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6320 * do nothing, and it will be requested again as soon as
6321 * the shadow expires. But we still need to check here,
6322 * because POPF has no interrupt shadow.
6324 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6325 kvm_make_request(KVM_REQ_EVENT, vcpu);
6326 } else
6327 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6329 return r;
6332 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6334 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6336 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6338 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6339 void *insn, int insn_len)
6341 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6343 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6345 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
6347 vcpu->arch.pio.count = 0;
6348 return 1;
6351 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
6353 vcpu->arch.pio.count = 0;
6355 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
6356 return 1;
6358 return kvm_skip_emulated_instruction(vcpu);
6361 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6362 unsigned short port)
6364 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
6365 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6366 size, port, &val, 1);
6367 if (ret)
6368 return ret;
6371 * Workaround userspace that relies on old KVM behavior of %rip being
6372 * incremented prior to exiting to userspace to handle "OUT 0x7e".
6374 if (port == 0x7e &&
6375 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
6376 vcpu->arch.complete_userspace_io =
6377 complete_fast_pio_out_port_0x7e;
6378 kvm_skip_emulated_instruction(vcpu);
6379 } else {
6380 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6381 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
6383 return 0;
6386 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6388 unsigned long val;
6390 /* We should only ever be called with arch.pio.count equal to 1 */
6391 BUG_ON(vcpu->arch.pio.count != 1);
6393 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
6394 vcpu->arch.pio.count = 0;
6395 return 1;
6398 /* For size less than 4 we merge, else we zero extend */
6399 val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
6400 : 0;
6403 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6404 * the copy and tracing
6406 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6407 vcpu->arch.pio.port, &val, 1);
6408 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6410 return kvm_skip_emulated_instruction(vcpu);
6413 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6414 unsigned short port)
6416 unsigned long val;
6417 int ret;
6419 /* For size less than 4 we merge, else we zero extend */
6420 val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
6422 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6423 &val, 1);
6424 if (ret) {
6425 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6426 return ret;
6429 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6430 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6432 return 0;
6435 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6437 int ret;
6439 if (in)
6440 ret = kvm_fast_pio_in(vcpu, size, port);
6441 else
6442 ret = kvm_fast_pio_out(vcpu, size, port);
6443 return ret && kvm_skip_emulated_instruction(vcpu);
6445 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6447 static int kvmclock_cpu_down_prep(unsigned int cpu)
6449 __this_cpu_write(cpu_tsc_khz, 0);
6450 return 0;
6453 static void tsc_khz_changed(void *data)
6455 struct cpufreq_freqs *freq = data;
6456 unsigned long khz = 0;
6458 if (data)
6459 khz = freq->new;
6460 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6461 khz = cpufreq_quick_get(raw_smp_processor_id());
6462 if (!khz)
6463 khz = tsc_khz;
6464 __this_cpu_write(cpu_tsc_khz, khz);
6467 #ifdef CONFIG_X86_64
6468 static void kvm_hyperv_tsc_notifier(void)
6470 struct kvm *kvm;
6471 struct kvm_vcpu *vcpu;
6472 int cpu;
6474 spin_lock(&kvm_lock);
6475 list_for_each_entry(kvm, &vm_list, vm_list)
6476 kvm_make_mclock_inprogress_request(kvm);
6478 hyperv_stop_tsc_emulation();
6480 /* TSC frequency always matches when on Hyper-V */
6481 for_each_present_cpu(cpu)
6482 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6483 kvm_max_guest_tsc_khz = tsc_khz;
6485 list_for_each_entry(kvm, &vm_list, vm_list) {
6486 struct kvm_arch *ka = &kvm->arch;
6488 spin_lock(&ka->pvclock_gtod_sync_lock);
6490 pvclock_update_vm_gtod_copy(kvm);
6492 kvm_for_each_vcpu(cpu, vcpu, kvm)
6493 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6495 kvm_for_each_vcpu(cpu, vcpu, kvm)
6496 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6498 spin_unlock(&ka->pvclock_gtod_sync_lock);
6500 spin_unlock(&kvm_lock);
6502 #endif
6504 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
6505 void *data)
6507 struct cpufreq_freqs *freq = data;
6508 struct kvm *kvm;
6509 struct kvm_vcpu *vcpu;
6510 int i, send_ipi = 0;
6513 * We allow guests to temporarily run on slowing clocks,
6514 * provided we notify them after, or to run on accelerating
6515 * clocks, provided we notify them before. Thus time never
6516 * goes backwards.
6518 * However, we have a problem. We can't atomically update
6519 * the frequency of a given CPU from this function; it is
6520 * merely a notifier, which can be called from any CPU.
6521 * Changing the TSC frequency at arbitrary points in time
6522 * requires a recomputation of local variables related to
6523 * the TSC for each VCPU. We must flag these local variables
6524 * to be updated and be sure the update takes place with the
6525 * new frequency before any guests proceed.
6527 * Unfortunately, the combination of hotplug CPU and frequency
6528 * change creates an intractable locking scenario; the order
6529 * of when these callouts happen is undefined with respect to
6530 * CPU hotplug, and they can race with each other. As such,
6531 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6532 * undefined; you can actually have a CPU frequency change take
6533 * place in between the computation of X and the setting of the
6534 * variable. To protect against this problem, all updates of
6535 * the per_cpu tsc_khz variable are done in an interrupt
6536 * protected IPI, and all callers wishing to update the value
6537 * must wait for a synchronous IPI to complete (which is trivial
6538 * if the caller is on the CPU already). This establishes the
6539 * necessary total order on variable updates.
6541 * Note that because a guest time update may take place
6542 * anytime after the setting of the VCPU's request bit, the
6543 * correct TSC value must be set before the request. However,
6544 * to ensure the update actually makes it to any guest which
6545 * starts running in hardware virtualization between the set
6546 * and the acquisition of the spinlock, we must also ping the
6547 * CPU after setting the request bit.
6551 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
6552 return 0;
6553 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
6554 return 0;
6556 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6558 spin_lock(&kvm_lock);
6559 list_for_each_entry(kvm, &vm_list, vm_list) {
6560 kvm_for_each_vcpu(i, vcpu, kvm) {
6561 if (vcpu->cpu != freq->cpu)
6562 continue;
6563 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6564 if (vcpu->cpu != smp_processor_id())
6565 send_ipi = 1;
6568 spin_unlock(&kvm_lock);
6570 if (freq->old < freq->new && send_ipi) {
6572 * We upscale the frequency. Must make the guest
6573 * doesn't see old kvmclock values while running with
6574 * the new frequency, otherwise we risk the guest sees
6575 * time go backwards.
6577 * In case we update the frequency for another cpu
6578 * (which might be in guest context) send an interrupt
6579 * to kick the cpu out of guest context. Next time
6580 * guest context is entered kvmclock will be updated,
6581 * so the guest will not see stale values.
6583 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6585 return 0;
6588 static struct notifier_block kvmclock_cpufreq_notifier_block = {
6589 .notifier_call = kvmclock_cpufreq_notifier
6592 static int kvmclock_cpu_online(unsigned int cpu)
6594 tsc_khz_changed(NULL);
6595 return 0;
6598 static void kvm_timer_init(void)
6600 max_tsc_khz = tsc_khz;
6602 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
6603 #ifdef CONFIG_CPU_FREQ
6604 struct cpufreq_policy policy;
6605 int cpu;
6607 memset(&policy, 0, sizeof(policy));
6608 cpu = get_cpu();
6609 cpufreq_get_policy(&policy, cpu);
6610 if (policy.cpuinfo.max_freq)
6611 max_tsc_khz = policy.cpuinfo.max_freq;
6612 put_cpu();
6613 #endif
6614 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6615 CPUFREQ_TRANSITION_NOTIFIER);
6617 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
6619 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
6620 kvmclock_cpu_online, kvmclock_cpu_down_prep);
6623 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6624 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
6626 int kvm_is_in_guest(void)
6628 return __this_cpu_read(current_vcpu) != NULL;
6631 static int kvm_is_user_mode(void)
6633 int user_mode = 3;
6635 if (__this_cpu_read(current_vcpu))
6636 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
6638 return user_mode != 0;
6641 static unsigned long kvm_get_guest_ip(void)
6643 unsigned long ip = 0;
6645 if (__this_cpu_read(current_vcpu))
6646 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
6648 return ip;
6651 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6652 .is_in_guest = kvm_is_in_guest,
6653 .is_user_mode = kvm_is_user_mode,
6654 .get_guest_ip = kvm_get_guest_ip,
6657 static void kvm_set_mmio_spte_mask(void)
6659 u64 mask;
6660 int maxphyaddr = boot_cpu_data.x86_phys_bits;
6663 * Set the reserved bits and the present bit of an paging-structure
6664 * entry to generate page fault with PFER.RSV = 1.
6668 * Mask the uppermost physical address bit, which would be reserved as
6669 * long as the supported physical address width is less than 52.
6671 mask = 1ull << 51;
6673 /* Set the present bit. */
6674 mask |= 1ull;
6677 * If reserved bit is not supported, clear the present bit to disable
6678 * mmio page fault.
6680 if (IS_ENABLED(CONFIG_X86_64) && maxphyaddr == 52)
6681 mask &= ~1ull;
6683 kvm_mmu_set_mmio_spte_mask(mask, mask);
6686 #ifdef CONFIG_X86_64
6687 static void pvclock_gtod_update_fn(struct work_struct *work)
6689 struct kvm *kvm;
6691 struct kvm_vcpu *vcpu;
6692 int i;
6694 spin_lock(&kvm_lock);
6695 list_for_each_entry(kvm, &vm_list, vm_list)
6696 kvm_for_each_vcpu(i, vcpu, kvm)
6697 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6698 atomic_set(&kvm_guest_has_master_clock, 0);
6699 spin_unlock(&kvm_lock);
6702 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6705 * Notification about pvclock gtod data update.
6707 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6708 void *priv)
6710 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6711 struct timekeeper *tk = priv;
6713 update_pvclock_gtod(tk);
6715 /* disable master clock if host does not trust, or does not
6716 * use, TSC based clocksource.
6718 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
6719 atomic_read(&kvm_guest_has_master_clock) != 0)
6720 queue_work(system_long_wq, &pvclock_gtod_work);
6722 return 0;
6725 static struct notifier_block pvclock_gtod_notifier = {
6726 .notifier_call = pvclock_gtod_notify,
6728 #endif
6730 int kvm_arch_init(void *opaque)
6732 int r;
6733 struct kvm_x86_ops *ops = opaque;
6735 if (kvm_x86_ops) {
6736 printk(KERN_ERR "kvm: already loaded the other module\n");
6737 r = -EEXIST;
6738 goto out;
6741 if (!ops->cpu_has_kvm_support()) {
6742 printk(KERN_ERR "kvm: no hardware support\n");
6743 r = -EOPNOTSUPP;
6744 goto out;
6746 if (ops->disabled_by_bios()) {
6747 printk(KERN_ERR "kvm: disabled by bios\n");
6748 r = -EOPNOTSUPP;
6749 goto out;
6752 r = -ENOMEM;
6753 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6754 if (!shared_msrs) {
6755 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6756 goto out;
6759 r = kvm_mmu_module_init();
6760 if (r)
6761 goto out_free_percpu;
6763 kvm_set_mmio_spte_mask();
6765 kvm_x86_ops = ops;
6767 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6768 PT_DIRTY_MASK, PT64_NX_MASK, 0,
6769 PT_PRESENT_MASK, 0, sme_me_mask);
6770 kvm_timer_init();
6772 perf_register_guest_info_callbacks(&kvm_guest_cbs);
6774 if (boot_cpu_has(X86_FEATURE_XSAVE))
6775 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6777 kvm_lapic_init();
6778 #ifdef CONFIG_X86_64
6779 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6781 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6782 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
6783 #endif
6785 return 0;
6787 out_free_percpu:
6788 free_percpu(shared_msrs);
6789 out:
6790 return r;
6793 void kvm_arch_exit(void)
6795 #ifdef CONFIG_X86_64
6796 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6797 clear_hv_tscchange_cb();
6798 #endif
6799 kvm_lapic_exit();
6800 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6802 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6803 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6804 CPUFREQ_TRANSITION_NOTIFIER);
6805 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6806 #ifdef CONFIG_X86_64
6807 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6808 #endif
6809 kvm_x86_ops = NULL;
6810 kvm_mmu_module_exit();
6811 free_percpu(shared_msrs);
6814 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6816 ++vcpu->stat.halt_exits;
6817 if (lapic_in_kernel(vcpu)) {
6818 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6819 return 1;
6820 } else {
6821 vcpu->run->exit_reason = KVM_EXIT_HLT;
6822 return 0;
6825 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6827 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6829 int ret = kvm_skip_emulated_instruction(vcpu);
6831 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6832 * KVM_EXIT_DEBUG here.
6834 return kvm_vcpu_halt(vcpu) && ret;
6836 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6838 #ifdef CONFIG_X86_64
6839 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6840 unsigned long clock_type)
6842 struct kvm_clock_pairing clock_pairing;
6843 struct timespec64 ts;
6844 u64 cycle;
6845 int ret;
6847 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6848 return -KVM_EOPNOTSUPP;
6850 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6851 return -KVM_EOPNOTSUPP;
6853 clock_pairing.sec = ts.tv_sec;
6854 clock_pairing.nsec = ts.tv_nsec;
6855 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6856 clock_pairing.flags = 0;
6857 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
6859 ret = 0;
6860 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6861 sizeof(struct kvm_clock_pairing)))
6862 ret = -KVM_EFAULT;
6864 return ret;
6866 #endif
6869 * kvm_pv_kick_cpu_op: Kick a vcpu.
6871 * @apicid - apicid of vcpu to be kicked.
6873 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6875 struct kvm_lapic_irq lapic_irq;
6877 lapic_irq.shorthand = 0;
6878 lapic_irq.dest_mode = 0;
6879 lapic_irq.level = 0;
6880 lapic_irq.dest_id = apicid;
6881 lapic_irq.msi_redir_hint = false;
6883 lapic_irq.delivery_mode = APIC_DM_REMRD;
6884 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6887 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6889 vcpu->arch.apicv_active = false;
6890 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6893 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6895 unsigned long nr, a0, a1, a2, a3, ret;
6896 int op_64_bit;
6898 if (kvm_hv_hypercall_enabled(vcpu->kvm))
6899 return kvm_hv_hypercall(vcpu);
6901 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6902 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6903 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6904 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6905 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
6907 trace_kvm_hypercall(nr, a0, a1, a2, a3);
6909 op_64_bit = is_64_bit_mode(vcpu);
6910 if (!op_64_bit) {
6911 nr &= 0xFFFFFFFF;
6912 a0 &= 0xFFFFFFFF;
6913 a1 &= 0xFFFFFFFF;
6914 a2 &= 0xFFFFFFFF;
6915 a3 &= 0xFFFFFFFF;
6918 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6919 ret = -KVM_EPERM;
6920 goto out;
6923 switch (nr) {
6924 case KVM_HC_VAPIC_POLL_IRQ:
6925 ret = 0;
6926 break;
6927 case KVM_HC_KICK_CPU:
6928 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6929 ret = 0;
6930 break;
6931 #ifdef CONFIG_X86_64
6932 case KVM_HC_CLOCK_PAIRING:
6933 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
6934 break;
6935 #endif
6936 case KVM_HC_SEND_IPI:
6937 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
6938 break;
6939 default:
6940 ret = -KVM_ENOSYS;
6941 break;
6943 out:
6944 if (!op_64_bit)
6945 ret = (u32)ret;
6946 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6948 ++vcpu->stat.hypercalls;
6949 return kvm_skip_emulated_instruction(vcpu);
6951 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6953 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6955 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6956 char instruction[3];
6957 unsigned long rip = kvm_rip_read(vcpu);
6959 kvm_x86_ops->patch_hypercall(vcpu, instruction);
6961 return emulator_write_emulated(ctxt, rip, instruction, 3,
6962 &ctxt->exception);
6965 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6967 return vcpu->run->request_interrupt_window &&
6968 likely(!pic_in_kernel(vcpu->kvm));
6971 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6973 struct kvm_run *kvm_run = vcpu->run;
6975 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6976 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6977 kvm_run->cr8 = kvm_get_cr8(vcpu);
6978 kvm_run->apic_base = kvm_get_apic_base(vcpu);
6979 kvm_run->ready_for_interrupt_injection =
6980 pic_in_kernel(vcpu->kvm) ||
6981 kvm_vcpu_ready_for_interrupt_injection(vcpu);
6984 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6986 int max_irr, tpr;
6988 if (!kvm_x86_ops->update_cr8_intercept)
6989 return;
6991 if (!lapic_in_kernel(vcpu))
6992 return;
6994 if (vcpu->arch.apicv_active)
6995 return;
6997 if (!vcpu->arch.apic->vapic_addr)
6998 max_irr = kvm_lapic_find_highest_irr(vcpu);
6999 else
7000 max_irr = -1;
7002 if (max_irr != -1)
7003 max_irr >>= 4;
7005 tpr = kvm_lapic_get_cr8(vcpu);
7007 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7010 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7012 int r;
7014 /* try to reinject previous events if any */
7016 if (vcpu->arch.exception.injected)
7017 kvm_x86_ops->queue_exception(vcpu);
7019 * Do not inject an NMI or interrupt if there is a pending
7020 * exception. Exceptions and interrupts are recognized at
7021 * instruction boundaries, i.e. the start of an instruction.
7022 * Trap-like exceptions, e.g. #DB, have higher priority than
7023 * NMIs and interrupts, i.e. traps are recognized before an
7024 * NMI/interrupt that's pending on the same instruction.
7025 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7026 * priority, but are only generated (pended) during instruction
7027 * execution, i.e. a pending fault-like exception means the
7028 * fault occurred on the *previous* instruction and must be
7029 * serviced prior to recognizing any new events in order to
7030 * fully complete the previous instruction.
7032 else if (!vcpu->arch.exception.pending) {
7033 if (vcpu->arch.nmi_injected)
7034 kvm_x86_ops->set_nmi(vcpu);
7035 else if (vcpu->arch.interrupt.injected)
7036 kvm_x86_ops->set_irq(vcpu);
7040 * Call check_nested_events() even if we reinjected a previous event
7041 * in order for caller to determine if it should require immediate-exit
7042 * from L2 to L1 due to pending L1 events which require exit
7043 * from L2 to L1.
7045 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7046 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7047 if (r != 0)
7048 return r;
7051 /* try to inject new event if pending */
7052 if (vcpu->arch.exception.pending) {
7053 trace_kvm_inj_exception(vcpu->arch.exception.nr,
7054 vcpu->arch.exception.has_error_code,
7055 vcpu->arch.exception.error_code);
7057 WARN_ON_ONCE(vcpu->arch.exception.injected);
7058 vcpu->arch.exception.pending = false;
7059 vcpu->arch.exception.injected = true;
7061 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7062 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7063 X86_EFLAGS_RF);
7065 if (vcpu->arch.exception.nr == DB_VECTOR &&
7066 (vcpu->arch.dr7 & DR7_GD)) {
7067 vcpu->arch.dr7 &= ~DR7_GD;
7068 kvm_update_dr7(vcpu);
7071 kvm_x86_ops->queue_exception(vcpu);
7074 /* Don't consider new event if we re-injected an event */
7075 if (kvm_event_needs_reinjection(vcpu))
7076 return 0;
7078 if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7079 kvm_x86_ops->smi_allowed(vcpu)) {
7080 vcpu->arch.smi_pending = false;
7081 ++vcpu->arch.smi_count;
7082 enter_smm(vcpu);
7083 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7084 --vcpu->arch.nmi_pending;
7085 vcpu->arch.nmi_injected = true;
7086 kvm_x86_ops->set_nmi(vcpu);
7087 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
7089 * Because interrupts can be injected asynchronously, we are
7090 * calling check_nested_events again here to avoid a race condition.
7091 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7092 * proposal and current concerns. Perhaps we should be setting
7093 * KVM_REQ_EVENT only on certain events and not unconditionally?
7095 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7096 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7097 if (r != 0)
7098 return r;
7100 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7101 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7102 false);
7103 kvm_x86_ops->set_irq(vcpu);
7107 return 0;
7110 static void process_nmi(struct kvm_vcpu *vcpu)
7112 unsigned limit = 2;
7115 * x86 is limited to one NMI running, and one NMI pending after it.
7116 * If an NMI is already in progress, limit further NMIs to just one.
7117 * Otherwise, allow two (and we'll inject the first one immediately).
7119 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7120 limit = 1;
7122 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7123 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7124 kvm_make_request(KVM_REQ_EVENT, vcpu);
7127 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7129 u32 flags = 0;
7130 flags |= seg->g << 23;
7131 flags |= seg->db << 22;
7132 flags |= seg->l << 21;
7133 flags |= seg->avl << 20;
7134 flags |= seg->present << 15;
7135 flags |= seg->dpl << 13;
7136 flags |= seg->s << 12;
7137 flags |= seg->type << 8;
7138 return flags;
7141 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7143 struct kvm_segment seg;
7144 int offset;
7146 kvm_get_segment(vcpu, &seg, n);
7147 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7149 if (n < 3)
7150 offset = 0x7f84 + n * 12;
7151 else
7152 offset = 0x7f2c + (n - 3) * 12;
7154 put_smstate(u32, buf, offset + 8, seg.base);
7155 put_smstate(u32, buf, offset + 4, seg.limit);
7156 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7159 #ifdef CONFIG_X86_64
7160 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7162 struct kvm_segment seg;
7163 int offset;
7164 u16 flags;
7166 kvm_get_segment(vcpu, &seg, n);
7167 offset = 0x7e00 + n * 16;
7169 flags = enter_smm_get_segment_flags(&seg) >> 8;
7170 put_smstate(u16, buf, offset, seg.selector);
7171 put_smstate(u16, buf, offset + 2, flags);
7172 put_smstate(u32, buf, offset + 4, seg.limit);
7173 put_smstate(u64, buf, offset + 8, seg.base);
7175 #endif
7177 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7179 struct desc_ptr dt;
7180 struct kvm_segment seg;
7181 unsigned long val;
7182 int i;
7184 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7185 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7186 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7187 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7189 for (i = 0; i < 8; i++)
7190 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7192 kvm_get_dr(vcpu, 6, &val);
7193 put_smstate(u32, buf, 0x7fcc, (u32)val);
7194 kvm_get_dr(vcpu, 7, &val);
7195 put_smstate(u32, buf, 0x7fc8, (u32)val);
7197 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7198 put_smstate(u32, buf, 0x7fc4, seg.selector);
7199 put_smstate(u32, buf, 0x7f64, seg.base);
7200 put_smstate(u32, buf, 0x7f60, seg.limit);
7201 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7203 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7204 put_smstate(u32, buf, 0x7fc0, seg.selector);
7205 put_smstate(u32, buf, 0x7f80, seg.base);
7206 put_smstate(u32, buf, 0x7f7c, seg.limit);
7207 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7209 kvm_x86_ops->get_gdt(vcpu, &dt);
7210 put_smstate(u32, buf, 0x7f74, dt.address);
7211 put_smstate(u32, buf, 0x7f70, dt.size);
7213 kvm_x86_ops->get_idt(vcpu, &dt);
7214 put_smstate(u32, buf, 0x7f58, dt.address);
7215 put_smstate(u32, buf, 0x7f54, dt.size);
7217 for (i = 0; i < 6; i++)
7218 enter_smm_save_seg_32(vcpu, buf, i);
7220 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7222 /* revision id */
7223 put_smstate(u32, buf, 0x7efc, 0x00020000);
7224 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7227 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7229 #ifdef CONFIG_X86_64
7230 struct desc_ptr dt;
7231 struct kvm_segment seg;
7232 unsigned long val;
7233 int i;
7235 for (i = 0; i < 16; i++)
7236 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7238 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7239 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7241 kvm_get_dr(vcpu, 6, &val);
7242 put_smstate(u64, buf, 0x7f68, val);
7243 kvm_get_dr(vcpu, 7, &val);
7244 put_smstate(u64, buf, 0x7f60, val);
7246 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7247 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7248 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7250 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7252 /* revision id */
7253 put_smstate(u32, buf, 0x7efc, 0x00020064);
7255 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7257 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7258 put_smstate(u16, buf, 0x7e90, seg.selector);
7259 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7260 put_smstate(u32, buf, 0x7e94, seg.limit);
7261 put_smstate(u64, buf, 0x7e98, seg.base);
7263 kvm_x86_ops->get_idt(vcpu, &dt);
7264 put_smstate(u32, buf, 0x7e84, dt.size);
7265 put_smstate(u64, buf, 0x7e88, dt.address);
7267 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7268 put_smstate(u16, buf, 0x7e70, seg.selector);
7269 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7270 put_smstate(u32, buf, 0x7e74, seg.limit);
7271 put_smstate(u64, buf, 0x7e78, seg.base);
7273 kvm_x86_ops->get_gdt(vcpu, &dt);
7274 put_smstate(u32, buf, 0x7e64, dt.size);
7275 put_smstate(u64, buf, 0x7e68, dt.address);
7277 for (i = 0; i < 6; i++)
7278 enter_smm_save_seg_64(vcpu, buf, i);
7279 #else
7280 WARN_ON_ONCE(1);
7281 #endif
7284 static void enter_smm(struct kvm_vcpu *vcpu)
7286 struct kvm_segment cs, ds;
7287 struct desc_ptr dt;
7288 char buf[512];
7289 u32 cr0;
7291 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7292 memset(buf, 0, 512);
7293 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7294 enter_smm_save_state_64(vcpu, buf);
7295 else
7296 enter_smm_save_state_32(vcpu, buf);
7299 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7300 * vCPU state (e.g. leave guest mode) after we've saved the state into
7301 * the SMM state-save area.
7303 kvm_x86_ops->pre_enter_smm(vcpu, buf);
7305 vcpu->arch.hflags |= HF_SMM_MASK;
7306 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7308 if (kvm_x86_ops->get_nmi_mask(vcpu))
7309 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7310 else
7311 kvm_x86_ops->set_nmi_mask(vcpu, true);
7313 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7314 kvm_rip_write(vcpu, 0x8000);
7316 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7317 kvm_x86_ops->set_cr0(vcpu, cr0);
7318 vcpu->arch.cr0 = cr0;
7320 kvm_x86_ops->set_cr4(vcpu, 0);
7322 /* Undocumented: IDT limit is set to zero on entry to SMM. */
7323 dt.address = dt.size = 0;
7324 kvm_x86_ops->set_idt(vcpu, &dt);
7326 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7328 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7329 cs.base = vcpu->arch.smbase;
7331 ds.selector = 0;
7332 ds.base = 0;
7334 cs.limit = ds.limit = 0xffffffff;
7335 cs.type = ds.type = 0x3;
7336 cs.dpl = ds.dpl = 0;
7337 cs.db = ds.db = 0;
7338 cs.s = ds.s = 1;
7339 cs.l = ds.l = 0;
7340 cs.g = ds.g = 1;
7341 cs.avl = ds.avl = 0;
7342 cs.present = ds.present = 1;
7343 cs.unusable = ds.unusable = 0;
7344 cs.padding = ds.padding = 0;
7346 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7347 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7348 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7349 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7350 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7351 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7353 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7354 kvm_x86_ops->set_efer(vcpu, 0);
7356 kvm_update_cpuid(vcpu);
7357 kvm_mmu_reset_context(vcpu);
7360 static void process_smi(struct kvm_vcpu *vcpu)
7362 vcpu->arch.smi_pending = true;
7363 kvm_make_request(KVM_REQ_EVENT, vcpu);
7366 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7368 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7371 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7373 if (!kvm_apic_present(vcpu))
7374 return;
7376 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7378 if (irqchip_split(vcpu->kvm))
7379 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7380 else {
7381 if (vcpu->arch.apicv_active)
7382 kvm_x86_ops->sync_pir_to_irr(vcpu);
7383 if (ioapic_in_kernel(vcpu->kvm))
7384 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7387 if (is_guest_mode(vcpu))
7388 vcpu->arch.load_eoi_exitmap_pending = true;
7389 else
7390 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7393 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7395 u64 eoi_exit_bitmap[4];
7397 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7398 return;
7400 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7401 vcpu_to_synic(vcpu)->vec_bitmap, 256);
7402 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7405 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7406 unsigned long start, unsigned long end,
7407 bool blockable)
7409 unsigned long apic_address;
7412 * The physical address of apic access page is stored in the VMCS.
7413 * Update it when it becomes invalid.
7415 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7416 if (start <= apic_address && apic_address < end)
7417 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7419 return 0;
7422 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7424 struct page *page = NULL;
7426 if (!lapic_in_kernel(vcpu))
7427 return;
7429 if (!kvm_x86_ops->set_apic_access_page_addr)
7430 return;
7432 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7433 if (is_error_page(page))
7434 return;
7435 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7438 * Do not pin apic access page in memory, the MMU notifier
7439 * will call us again if it is migrated or swapped out.
7441 put_page(page);
7443 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7445 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
7447 smp_send_reschedule(vcpu->cpu);
7449 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
7452 * Returns 1 to let vcpu_run() continue the guest execution loop without
7453 * exiting to the userspace. Otherwise, the value will be returned to the
7454 * userspace.
7456 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7458 int r;
7459 bool req_int_win =
7460 dm_request_for_irq_injection(vcpu) &&
7461 kvm_cpu_accept_dm_intr(vcpu);
7463 bool req_immediate_exit = false;
7465 if (kvm_request_pending(vcpu)) {
7466 if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu))
7467 kvm_x86_ops->get_vmcs12_pages(vcpu);
7468 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7469 kvm_mmu_unload(vcpu);
7470 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7471 __kvm_migrate_timers(vcpu);
7472 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7473 kvm_gen_update_masterclock(vcpu->kvm);
7474 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7475 kvm_gen_kvmclock_update(vcpu);
7476 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7477 r = kvm_guest_time_update(vcpu);
7478 if (unlikely(r))
7479 goto out;
7481 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7482 kvm_mmu_sync_roots(vcpu);
7483 if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
7484 kvm_mmu_load_cr3(vcpu);
7485 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7486 kvm_vcpu_flush_tlb(vcpu, true);
7487 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7488 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7489 r = 0;
7490 goto out;
7492 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
7493 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
7494 vcpu->mmio_needed = 0;
7495 r = 0;
7496 goto out;
7498 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7499 /* Page is swapped out. Do synthetic halt */
7500 vcpu->arch.apf.halted = true;
7501 r = 1;
7502 goto out;
7504 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
7505 record_steal_time(vcpu);
7506 if (kvm_check_request(KVM_REQ_SMI, vcpu))
7507 process_smi(vcpu);
7508 if (kvm_check_request(KVM_REQ_NMI, vcpu))
7509 process_nmi(vcpu);
7510 if (kvm_check_request(KVM_REQ_PMU, vcpu))
7511 kvm_pmu_handle_event(vcpu);
7512 if (kvm_check_request(KVM_REQ_PMI, vcpu))
7513 kvm_pmu_deliver_pmi(vcpu);
7514 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7515 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7516 if (test_bit(vcpu->arch.pending_ioapic_eoi,
7517 vcpu->arch.ioapic_handled_vectors)) {
7518 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7519 vcpu->run->eoi.vector =
7520 vcpu->arch.pending_ioapic_eoi;
7521 r = 0;
7522 goto out;
7525 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
7526 vcpu_scan_ioapic(vcpu);
7527 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
7528 vcpu_load_eoi_exitmap(vcpu);
7529 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
7530 kvm_vcpu_reload_apic_access_page(vcpu);
7531 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
7532 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7533 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
7534 r = 0;
7535 goto out;
7537 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
7538 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7539 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
7540 r = 0;
7541 goto out;
7543 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
7544 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
7545 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
7546 r = 0;
7547 goto out;
7551 * KVM_REQ_HV_STIMER has to be processed after
7552 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
7553 * depend on the guest clock being up-to-date
7555 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
7556 kvm_hv_process_stimers(vcpu);
7559 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
7560 ++vcpu->stat.req_event;
7561 kvm_apic_accept_events(vcpu);
7562 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
7563 r = 1;
7564 goto out;
7567 if (inject_pending_event(vcpu, req_int_win) != 0)
7568 req_immediate_exit = true;
7569 else {
7570 /* Enable SMI/NMI/IRQ window open exits if needed.
7572 * SMIs have three cases:
7573 * 1) They can be nested, and then there is nothing to
7574 * do here because RSM will cause a vmexit anyway.
7575 * 2) There is an ISA-specific reason why SMI cannot be
7576 * injected, and the moment when this changes can be
7577 * intercepted.
7578 * 3) Or the SMI can be pending because
7579 * inject_pending_event has completed the injection
7580 * of an IRQ or NMI from the previous vmexit, and
7581 * then we request an immediate exit to inject the
7582 * SMI.
7584 if (vcpu->arch.smi_pending && !is_smm(vcpu))
7585 if (!kvm_x86_ops->enable_smi_window(vcpu))
7586 req_immediate_exit = true;
7587 if (vcpu->arch.nmi_pending)
7588 kvm_x86_ops->enable_nmi_window(vcpu);
7589 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
7590 kvm_x86_ops->enable_irq_window(vcpu);
7591 WARN_ON(vcpu->arch.exception.pending);
7594 if (kvm_lapic_enabled(vcpu)) {
7595 update_cr8_intercept(vcpu);
7596 kvm_lapic_sync_to_vapic(vcpu);
7600 r = kvm_mmu_reload(vcpu);
7601 if (unlikely(r)) {
7602 goto cancel_injection;
7605 preempt_disable();
7607 kvm_x86_ops->prepare_guest_switch(vcpu);
7610 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
7611 * IPI are then delayed after guest entry, which ensures that they
7612 * result in virtual interrupt delivery.
7614 local_irq_disable();
7615 vcpu->mode = IN_GUEST_MODE;
7617 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7620 * 1) We should set ->mode before checking ->requests. Please see
7621 * the comment in kvm_vcpu_exiting_guest_mode().
7623 * 2) For APICv, we should set ->mode before checking PIR.ON. This
7624 * pairs with the memory barrier implicit in pi_test_and_set_on
7625 * (see vmx_deliver_posted_interrupt).
7627 * 3) This also orders the write to mode from any reads to the page
7628 * tables done while the VCPU is running. Please see the comment
7629 * in kvm_flush_remote_tlbs.
7631 smp_mb__after_srcu_read_unlock();
7634 * This handles the case where a posted interrupt was
7635 * notified with kvm_vcpu_kick.
7637 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
7638 kvm_x86_ops->sync_pir_to_irr(vcpu);
7640 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
7641 || need_resched() || signal_pending(current)) {
7642 vcpu->mode = OUTSIDE_GUEST_MODE;
7643 smp_wmb();
7644 local_irq_enable();
7645 preempt_enable();
7646 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7647 r = 1;
7648 goto cancel_injection;
7651 kvm_load_guest_xcr0(vcpu);
7653 if (req_immediate_exit) {
7654 kvm_make_request(KVM_REQ_EVENT, vcpu);
7655 kvm_x86_ops->request_immediate_exit(vcpu);
7658 trace_kvm_entry(vcpu->vcpu_id);
7659 if (lapic_timer_advance_ns)
7660 wait_lapic_expire(vcpu);
7661 guest_enter_irqoff();
7663 if (unlikely(vcpu->arch.switch_db_regs)) {
7664 set_debugreg(0, 7);
7665 set_debugreg(vcpu->arch.eff_db[0], 0);
7666 set_debugreg(vcpu->arch.eff_db[1], 1);
7667 set_debugreg(vcpu->arch.eff_db[2], 2);
7668 set_debugreg(vcpu->arch.eff_db[3], 3);
7669 set_debugreg(vcpu->arch.dr6, 6);
7670 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7673 kvm_x86_ops->run(vcpu);
7676 * Do this here before restoring debug registers on the host. And
7677 * since we do this before handling the vmexit, a DR access vmexit
7678 * can (a) read the correct value of the debug registers, (b) set
7679 * KVM_DEBUGREG_WONT_EXIT again.
7681 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
7682 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
7683 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
7684 kvm_update_dr0123(vcpu);
7685 kvm_update_dr6(vcpu);
7686 kvm_update_dr7(vcpu);
7687 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7691 * If the guest has used debug registers, at least dr7
7692 * will be disabled while returning to the host.
7693 * If we don't have active breakpoints in the host, we don't
7694 * care about the messed up debug address registers. But if
7695 * we have some of them active, restore the old state.
7697 if (hw_breakpoint_active())
7698 hw_breakpoint_restore();
7700 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
7702 vcpu->mode = OUTSIDE_GUEST_MODE;
7703 smp_wmb();
7705 kvm_put_guest_xcr0(vcpu);
7707 kvm_before_interrupt(vcpu);
7708 kvm_x86_ops->handle_external_intr(vcpu);
7709 kvm_after_interrupt(vcpu);
7711 ++vcpu->stat.exits;
7713 guest_exit_irqoff();
7715 local_irq_enable();
7716 preempt_enable();
7718 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7721 * Profile KVM exit RIPs:
7723 if (unlikely(prof_on == KVM_PROFILING)) {
7724 unsigned long rip = kvm_rip_read(vcpu);
7725 profile_hit(KVM_PROFILING, (void *)rip);
7728 if (unlikely(vcpu->arch.tsc_always_catchup))
7729 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7731 if (vcpu->arch.apic_attention)
7732 kvm_lapic_sync_from_vapic(vcpu);
7734 vcpu->arch.gpa_available = false;
7735 r = kvm_x86_ops->handle_exit(vcpu);
7736 return r;
7738 cancel_injection:
7739 kvm_x86_ops->cancel_injection(vcpu);
7740 if (unlikely(vcpu->arch.apic_attention))
7741 kvm_lapic_sync_from_vapic(vcpu);
7742 out:
7743 return r;
7746 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7748 if (!kvm_arch_vcpu_runnable(vcpu) &&
7749 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7750 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7751 kvm_vcpu_block(vcpu);
7752 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7754 if (kvm_x86_ops->post_block)
7755 kvm_x86_ops->post_block(vcpu);
7757 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7758 return 1;
7761 kvm_apic_accept_events(vcpu);
7762 switch(vcpu->arch.mp_state) {
7763 case KVM_MP_STATE_HALTED:
7764 vcpu->arch.pv.pv_unhalted = false;
7765 vcpu->arch.mp_state =
7766 KVM_MP_STATE_RUNNABLE;
7767 case KVM_MP_STATE_RUNNABLE:
7768 vcpu->arch.apf.halted = false;
7769 break;
7770 case KVM_MP_STATE_INIT_RECEIVED:
7771 break;
7772 default:
7773 return -EINTR;
7774 break;
7776 return 1;
7779 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7781 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7782 kvm_x86_ops->check_nested_events(vcpu, false);
7784 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7785 !vcpu->arch.apf.halted);
7788 static int vcpu_run(struct kvm_vcpu *vcpu)
7790 int r;
7791 struct kvm *kvm = vcpu->kvm;
7793 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7794 vcpu->arch.l1tf_flush_l1d = true;
7796 for (;;) {
7797 if (kvm_vcpu_running(vcpu)) {
7798 r = vcpu_enter_guest(vcpu);
7799 } else {
7800 r = vcpu_block(kvm, vcpu);
7803 if (r <= 0)
7804 break;
7806 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
7807 if (kvm_cpu_has_pending_timer(vcpu))
7808 kvm_inject_pending_timer_irqs(vcpu);
7810 if (dm_request_for_irq_injection(vcpu) &&
7811 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
7812 r = 0;
7813 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
7814 ++vcpu->stat.request_irq_exits;
7815 break;
7818 kvm_check_async_pf_completion(vcpu);
7820 if (signal_pending(current)) {
7821 r = -EINTR;
7822 vcpu->run->exit_reason = KVM_EXIT_INTR;
7823 ++vcpu->stat.signal_exits;
7824 break;
7826 if (need_resched()) {
7827 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7828 cond_resched();
7829 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7833 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7835 return r;
7838 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7840 int r;
7841 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7842 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7843 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7844 if (r != EMULATE_DONE)
7845 return 0;
7846 return 1;
7849 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7851 BUG_ON(!vcpu->arch.pio.count);
7853 return complete_emulated_io(vcpu);
7857 * Implements the following, as a state machine:
7859 * read:
7860 * for each fragment
7861 * for each mmio piece in the fragment
7862 * write gpa, len
7863 * exit
7864 * copy data
7865 * execute insn
7867 * write:
7868 * for each fragment
7869 * for each mmio piece in the fragment
7870 * write gpa, len
7871 * copy data
7872 * exit
7874 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
7876 struct kvm_run *run = vcpu->run;
7877 struct kvm_mmio_fragment *frag;
7878 unsigned len;
7880 BUG_ON(!vcpu->mmio_needed);
7882 /* Complete previous fragment */
7883 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7884 len = min(8u, frag->len);
7885 if (!vcpu->mmio_is_write)
7886 memcpy(frag->data, run->mmio.data, len);
7888 if (frag->len <= 8) {
7889 /* Switch to the next fragment. */
7890 frag++;
7891 vcpu->mmio_cur_fragment++;
7892 } else {
7893 /* Go forward to the next mmio piece. */
7894 frag->data += len;
7895 frag->gpa += len;
7896 frag->len -= len;
7899 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
7900 vcpu->mmio_needed = 0;
7902 /* FIXME: return into emulator if single-stepping. */
7903 if (vcpu->mmio_is_write)
7904 return 1;
7905 vcpu->mmio_read_completed = 1;
7906 return complete_emulated_io(vcpu);
7909 run->exit_reason = KVM_EXIT_MMIO;
7910 run->mmio.phys_addr = frag->gpa;
7911 if (vcpu->mmio_is_write)
7912 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7913 run->mmio.len = min(8u, frag->len);
7914 run->mmio.is_write = vcpu->mmio_is_write;
7915 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7916 return 0;
7919 /* Swap (qemu) user FPU context for the guest FPU context. */
7920 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7922 preempt_disable();
7923 copy_fpregs_to_fpstate(&vcpu->arch.user_fpu);
7924 /* PKRU is separately restored in kvm_x86_ops->run. */
7925 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state,
7926 ~XFEATURE_MASK_PKRU);
7927 preempt_enable();
7928 trace_kvm_fpu(1);
7931 /* When vcpu_run ends, restore user space FPU context. */
7932 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7934 preempt_disable();
7935 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7936 copy_kernel_to_fpregs(&vcpu->arch.user_fpu.state);
7937 preempt_enable();
7938 ++vcpu->stat.fpu_reload;
7939 trace_kvm_fpu(0);
7942 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7944 int r;
7946 vcpu_load(vcpu);
7947 kvm_sigset_activate(vcpu);
7948 kvm_load_guest_fpu(vcpu);
7950 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
7951 if (kvm_run->immediate_exit) {
7952 r = -EINTR;
7953 goto out;
7955 kvm_vcpu_block(vcpu);
7956 kvm_apic_accept_events(vcpu);
7957 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
7958 r = -EAGAIN;
7959 if (signal_pending(current)) {
7960 r = -EINTR;
7961 vcpu->run->exit_reason = KVM_EXIT_INTR;
7962 ++vcpu->stat.signal_exits;
7964 goto out;
7967 if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
7968 r = -EINVAL;
7969 goto out;
7972 if (vcpu->run->kvm_dirty_regs) {
7973 r = sync_regs(vcpu);
7974 if (r != 0)
7975 goto out;
7978 /* re-sync apic's tpr */
7979 if (!lapic_in_kernel(vcpu)) {
7980 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7981 r = -EINVAL;
7982 goto out;
7986 if (unlikely(vcpu->arch.complete_userspace_io)) {
7987 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7988 vcpu->arch.complete_userspace_io = NULL;
7989 r = cui(vcpu);
7990 if (r <= 0)
7991 goto out;
7992 } else
7993 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
7995 if (kvm_run->immediate_exit)
7996 r = -EINTR;
7997 else
7998 r = vcpu_run(vcpu);
8000 out:
8001 kvm_put_guest_fpu(vcpu);
8002 if (vcpu->run->kvm_valid_regs)
8003 store_regs(vcpu);
8004 post_kvm_run_save(vcpu);
8005 kvm_sigset_deactivate(vcpu);
8007 vcpu_put(vcpu);
8008 return r;
8011 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8013 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8015 * We are here if userspace calls get_regs() in the middle of
8016 * instruction emulation. Registers state needs to be copied
8017 * back from emulation context to vcpu. Userspace shouldn't do
8018 * that usually, but some bad designed PV devices (vmware
8019 * backdoor interface) need this to work
8021 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8022 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8024 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
8025 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
8026 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
8027 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
8028 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
8029 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
8030 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
8031 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
8032 #ifdef CONFIG_X86_64
8033 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
8034 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
8035 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
8036 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
8037 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
8038 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
8039 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
8040 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
8041 #endif
8043 regs->rip = kvm_rip_read(vcpu);
8044 regs->rflags = kvm_get_rflags(vcpu);
8047 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8049 vcpu_load(vcpu);
8050 __get_regs(vcpu, regs);
8051 vcpu_put(vcpu);
8052 return 0;
8055 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8057 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8058 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8060 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
8061 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
8062 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
8063 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
8064 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
8065 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
8066 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
8067 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
8068 #ifdef CONFIG_X86_64
8069 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
8070 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
8071 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
8072 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
8073 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
8074 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
8075 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
8076 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
8077 #endif
8079 kvm_rip_write(vcpu, regs->rip);
8080 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8082 vcpu->arch.exception.pending = false;
8084 kvm_make_request(KVM_REQ_EVENT, vcpu);
8087 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8089 vcpu_load(vcpu);
8090 __set_regs(vcpu, regs);
8091 vcpu_put(vcpu);
8092 return 0;
8095 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8097 struct kvm_segment cs;
8099 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8100 *db = cs.db;
8101 *l = cs.l;
8103 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8105 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8107 struct desc_ptr dt;
8109 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8110 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8111 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8112 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8113 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8114 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8116 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8117 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8119 kvm_x86_ops->get_idt(vcpu, &dt);
8120 sregs->idt.limit = dt.size;
8121 sregs->idt.base = dt.address;
8122 kvm_x86_ops->get_gdt(vcpu, &dt);
8123 sregs->gdt.limit = dt.size;
8124 sregs->gdt.base = dt.address;
8126 sregs->cr0 = kvm_read_cr0(vcpu);
8127 sregs->cr2 = vcpu->arch.cr2;
8128 sregs->cr3 = kvm_read_cr3(vcpu);
8129 sregs->cr4 = kvm_read_cr4(vcpu);
8130 sregs->cr8 = kvm_get_cr8(vcpu);
8131 sregs->efer = vcpu->arch.efer;
8132 sregs->apic_base = kvm_get_apic_base(vcpu);
8134 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
8136 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8137 set_bit(vcpu->arch.interrupt.nr,
8138 (unsigned long *)sregs->interrupt_bitmap);
8141 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8142 struct kvm_sregs *sregs)
8144 vcpu_load(vcpu);
8145 __get_sregs(vcpu, sregs);
8146 vcpu_put(vcpu);
8147 return 0;
8150 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8151 struct kvm_mp_state *mp_state)
8153 vcpu_load(vcpu);
8155 kvm_apic_accept_events(vcpu);
8156 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8157 vcpu->arch.pv.pv_unhalted)
8158 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8159 else
8160 mp_state->mp_state = vcpu->arch.mp_state;
8162 vcpu_put(vcpu);
8163 return 0;
8166 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8167 struct kvm_mp_state *mp_state)
8169 int ret = -EINVAL;
8171 vcpu_load(vcpu);
8173 if (!lapic_in_kernel(vcpu) &&
8174 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8175 goto out;
8177 /* INITs are latched while in SMM */
8178 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
8179 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8180 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8181 goto out;
8183 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8184 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8185 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8186 } else
8187 vcpu->arch.mp_state = mp_state->mp_state;
8188 kvm_make_request(KVM_REQ_EVENT, vcpu);
8190 ret = 0;
8191 out:
8192 vcpu_put(vcpu);
8193 return ret;
8196 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8197 int reason, bool has_error_code, u32 error_code)
8199 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8200 int ret;
8202 init_emulate_ctxt(vcpu);
8204 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8205 has_error_code, error_code);
8207 if (ret)
8208 return EMULATE_FAIL;
8210 kvm_rip_write(vcpu, ctxt->eip);
8211 kvm_set_rflags(vcpu, ctxt->eflags);
8212 kvm_make_request(KVM_REQ_EVENT, vcpu);
8213 return EMULATE_DONE;
8215 EXPORT_SYMBOL_GPL(kvm_task_switch);
8217 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8219 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
8220 (sregs->cr4 & X86_CR4_OSXSAVE))
8221 return -EINVAL;
8223 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8225 * When EFER.LME and CR0.PG are set, the processor is in
8226 * 64-bit mode (though maybe in a 32-bit code segment).
8227 * CR4.PAE and EFER.LMA must be set.
8229 if (!(sregs->cr4 & X86_CR4_PAE)
8230 || !(sregs->efer & EFER_LMA))
8231 return -EINVAL;
8232 } else {
8234 * Not in 64-bit mode: EFER.LMA is clear and the code
8235 * segment cannot be 64-bit.
8237 if (sregs->efer & EFER_LMA || sregs->cs.l)
8238 return -EINVAL;
8241 return 0;
8244 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8246 struct msr_data apic_base_msr;
8247 int mmu_reset_needed = 0;
8248 int cpuid_update_needed = 0;
8249 int pending_vec, max_bits, idx;
8250 struct desc_ptr dt;
8251 int ret = -EINVAL;
8253 if (kvm_valid_sregs(vcpu, sregs))
8254 goto out;
8256 apic_base_msr.data = sregs->apic_base;
8257 apic_base_msr.host_initiated = true;
8258 if (kvm_set_apic_base(vcpu, &apic_base_msr))
8259 goto out;
8261 dt.size = sregs->idt.limit;
8262 dt.address = sregs->idt.base;
8263 kvm_x86_ops->set_idt(vcpu, &dt);
8264 dt.size = sregs->gdt.limit;
8265 dt.address = sregs->gdt.base;
8266 kvm_x86_ops->set_gdt(vcpu, &dt);
8268 vcpu->arch.cr2 = sregs->cr2;
8269 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8270 vcpu->arch.cr3 = sregs->cr3;
8271 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8273 kvm_set_cr8(vcpu, sregs->cr8);
8275 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8276 kvm_x86_ops->set_efer(vcpu, sregs->efer);
8278 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8279 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8280 vcpu->arch.cr0 = sregs->cr0;
8282 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8283 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8284 (X86_CR4_OSXSAVE | X86_CR4_PKE));
8285 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8286 if (cpuid_update_needed)
8287 kvm_update_cpuid(vcpu);
8289 idx = srcu_read_lock(&vcpu->kvm->srcu);
8290 if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu)) {
8291 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8292 mmu_reset_needed = 1;
8294 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8296 if (mmu_reset_needed)
8297 kvm_mmu_reset_context(vcpu);
8299 max_bits = KVM_NR_INTERRUPTS;
8300 pending_vec = find_first_bit(
8301 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
8302 if (pending_vec < max_bits) {
8303 kvm_queue_interrupt(vcpu, pending_vec, false);
8304 pr_debug("Set back pending irq %d\n", pending_vec);
8307 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8308 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8309 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8310 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8311 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8312 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8314 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8315 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8317 update_cr8_intercept(vcpu);
8319 /* Older userspace won't unhalt the vcpu on reset. */
8320 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8321 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8322 !is_protmode(vcpu))
8323 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8325 kvm_make_request(KVM_REQ_EVENT, vcpu);
8327 ret = 0;
8328 out:
8329 return ret;
8332 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8333 struct kvm_sregs *sregs)
8335 int ret;
8337 vcpu_load(vcpu);
8338 ret = __set_sregs(vcpu, sregs);
8339 vcpu_put(vcpu);
8340 return ret;
8343 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8344 struct kvm_guest_debug *dbg)
8346 unsigned long rflags;
8347 int i, r;
8349 vcpu_load(vcpu);
8351 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8352 r = -EBUSY;
8353 if (vcpu->arch.exception.pending)
8354 goto out;
8355 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8356 kvm_queue_exception(vcpu, DB_VECTOR);
8357 else
8358 kvm_queue_exception(vcpu, BP_VECTOR);
8362 * Read rflags as long as potentially injected trace flags are still
8363 * filtered out.
8365 rflags = kvm_get_rflags(vcpu);
8367 vcpu->guest_debug = dbg->control;
8368 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8369 vcpu->guest_debug = 0;
8371 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8372 for (i = 0; i < KVM_NR_DB_REGS; ++i)
8373 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8374 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8375 } else {
8376 for (i = 0; i < KVM_NR_DB_REGS; i++)
8377 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8379 kvm_update_dr7(vcpu);
8381 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8382 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8383 get_segment_base(vcpu, VCPU_SREG_CS);
8386 * Trigger an rflags update that will inject or remove the trace
8387 * flags.
8389 kvm_set_rflags(vcpu, rflags);
8391 kvm_x86_ops->update_bp_intercept(vcpu);
8393 r = 0;
8395 out:
8396 vcpu_put(vcpu);
8397 return r;
8401 * Translate a guest virtual address to a guest physical address.
8403 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8404 struct kvm_translation *tr)
8406 unsigned long vaddr = tr->linear_address;
8407 gpa_t gpa;
8408 int idx;
8410 vcpu_load(vcpu);
8412 idx = srcu_read_lock(&vcpu->kvm->srcu);
8413 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8414 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8415 tr->physical_address = gpa;
8416 tr->valid = gpa != UNMAPPED_GVA;
8417 tr->writeable = 1;
8418 tr->usermode = 0;
8420 vcpu_put(vcpu);
8421 return 0;
8424 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8426 struct fxregs_state *fxsave;
8428 vcpu_load(vcpu);
8430 fxsave = &vcpu->arch.guest_fpu.state.fxsave;
8431 memcpy(fpu->fpr, fxsave->st_space, 128);
8432 fpu->fcw = fxsave->cwd;
8433 fpu->fsw = fxsave->swd;
8434 fpu->ftwx = fxsave->twd;
8435 fpu->last_opcode = fxsave->fop;
8436 fpu->last_ip = fxsave->rip;
8437 fpu->last_dp = fxsave->rdp;
8438 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
8440 vcpu_put(vcpu);
8441 return 0;
8444 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8446 struct fxregs_state *fxsave;
8448 vcpu_load(vcpu);
8450 fxsave = &vcpu->arch.guest_fpu.state.fxsave;
8452 memcpy(fxsave->st_space, fpu->fpr, 128);
8453 fxsave->cwd = fpu->fcw;
8454 fxsave->swd = fpu->fsw;
8455 fxsave->twd = fpu->ftwx;
8456 fxsave->fop = fpu->last_opcode;
8457 fxsave->rip = fpu->last_ip;
8458 fxsave->rdp = fpu->last_dp;
8459 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
8461 vcpu_put(vcpu);
8462 return 0;
8465 static void store_regs(struct kvm_vcpu *vcpu)
8467 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8469 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8470 __get_regs(vcpu, &vcpu->run->s.regs.regs);
8472 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8473 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8475 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8476 kvm_vcpu_ioctl_x86_get_vcpu_events(
8477 vcpu, &vcpu->run->s.regs.events);
8480 static int sync_regs(struct kvm_vcpu *vcpu)
8482 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
8483 return -EINVAL;
8485 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
8486 __set_regs(vcpu, &vcpu->run->s.regs.regs);
8487 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
8489 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
8490 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
8491 return -EINVAL;
8492 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
8494 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
8495 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
8496 vcpu, &vcpu->run->s.regs.events))
8497 return -EINVAL;
8498 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
8501 return 0;
8504 static void fx_init(struct kvm_vcpu *vcpu)
8506 fpstate_init(&vcpu->arch.guest_fpu.state);
8507 if (boot_cpu_has(X86_FEATURE_XSAVES))
8508 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
8509 host_xcr0 | XSTATE_COMPACTION_ENABLED;
8512 * Ensure guest xcr0 is valid for loading
8514 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8516 vcpu->arch.cr0 |= X86_CR0_ET;
8519 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
8521 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
8523 kvmclock_reset(vcpu);
8525 kvm_x86_ops->vcpu_free(vcpu);
8526 free_cpumask_var(wbinvd_dirty_mask);
8529 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
8530 unsigned int id)
8532 struct kvm_vcpu *vcpu;
8534 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
8535 printk_once(KERN_WARNING
8536 "kvm: SMP vm created on host with unstable TSC; "
8537 "guest TSC will not be reliable\n");
8539 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
8541 return vcpu;
8544 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
8546 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
8547 kvm_vcpu_mtrr_init(vcpu);
8548 vcpu_load(vcpu);
8549 kvm_vcpu_reset(vcpu, false);
8550 kvm_mmu_setup(vcpu);
8551 vcpu_put(vcpu);
8552 return 0;
8555 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
8557 struct msr_data msr;
8558 struct kvm *kvm = vcpu->kvm;
8560 kvm_hv_vcpu_postcreate(vcpu);
8562 if (mutex_lock_killable(&vcpu->mutex))
8563 return;
8564 vcpu_load(vcpu);
8565 msr.data = 0x0;
8566 msr.index = MSR_IA32_TSC;
8567 msr.host_initiated = true;
8568 kvm_write_tsc(vcpu, &msr);
8569 vcpu_put(vcpu);
8570 mutex_unlock(&vcpu->mutex);
8572 if (!kvmclock_periodic_sync)
8573 return;
8575 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
8576 KVMCLOCK_SYNC_PERIOD);
8579 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
8581 vcpu->arch.apf.msr_val = 0;
8583 vcpu_load(vcpu);
8584 kvm_mmu_unload(vcpu);
8585 vcpu_put(vcpu);
8587 kvm_x86_ops->vcpu_free(vcpu);
8590 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
8592 kvm_lapic_reset(vcpu, init_event);
8594 vcpu->arch.hflags = 0;
8596 vcpu->arch.smi_pending = 0;
8597 vcpu->arch.smi_count = 0;
8598 atomic_set(&vcpu->arch.nmi_queued, 0);
8599 vcpu->arch.nmi_pending = 0;
8600 vcpu->arch.nmi_injected = false;
8601 kvm_clear_interrupt_queue(vcpu);
8602 kvm_clear_exception_queue(vcpu);
8603 vcpu->arch.exception.pending = false;
8605 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
8606 kvm_update_dr0123(vcpu);
8607 vcpu->arch.dr6 = DR6_INIT;
8608 kvm_update_dr6(vcpu);
8609 vcpu->arch.dr7 = DR7_FIXED_1;
8610 kvm_update_dr7(vcpu);
8612 vcpu->arch.cr2 = 0;
8614 kvm_make_request(KVM_REQ_EVENT, vcpu);
8615 vcpu->arch.apf.msr_val = 0;
8616 vcpu->arch.st.msr_val = 0;
8618 kvmclock_reset(vcpu);
8620 kvm_clear_async_pf_completion_queue(vcpu);
8621 kvm_async_pf_hash_reset(vcpu);
8622 vcpu->arch.apf.halted = false;
8624 if (kvm_mpx_supported()) {
8625 void *mpx_state_buffer;
8628 * To avoid have the INIT path from kvm_apic_has_events() that be
8629 * called with loaded FPU and does not let userspace fix the state.
8631 if (init_event)
8632 kvm_put_guest_fpu(vcpu);
8633 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
8634 XFEATURE_MASK_BNDREGS);
8635 if (mpx_state_buffer)
8636 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
8637 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
8638 XFEATURE_MASK_BNDCSR);
8639 if (mpx_state_buffer)
8640 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
8641 if (init_event)
8642 kvm_load_guest_fpu(vcpu);
8645 if (!init_event) {
8646 kvm_pmu_reset(vcpu);
8647 vcpu->arch.smbase = 0x30000;
8649 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
8650 vcpu->arch.msr_misc_features_enables = 0;
8652 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8655 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
8656 vcpu->arch.regs_avail = ~0;
8657 vcpu->arch.regs_dirty = ~0;
8659 vcpu->arch.ia32_xss = 0;
8661 kvm_x86_ops->vcpu_reset(vcpu, init_event);
8664 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
8666 struct kvm_segment cs;
8668 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8669 cs.selector = vector << 8;
8670 cs.base = vector << 12;
8671 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8672 kvm_rip_write(vcpu, 0);
8675 int kvm_arch_hardware_enable(void)
8677 struct kvm *kvm;
8678 struct kvm_vcpu *vcpu;
8679 int i;
8680 int ret;
8681 u64 local_tsc;
8682 u64 max_tsc = 0;
8683 bool stable, backwards_tsc = false;
8685 kvm_shared_msr_cpu_online();
8686 ret = kvm_x86_ops->hardware_enable();
8687 if (ret != 0)
8688 return ret;
8690 local_tsc = rdtsc();
8691 stable = !kvm_check_tsc_unstable();
8692 list_for_each_entry(kvm, &vm_list, vm_list) {
8693 kvm_for_each_vcpu(i, vcpu, kvm) {
8694 if (!stable && vcpu->cpu == smp_processor_id())
8695 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8696 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
8697 backwards_tsc = true;
8698 if (vcpu->arch.last_host_tsc > max_tsc)
8699 max_tsc = vcpu->arch.last_host_tsc;
8705 * Sometimes, even reliable TSCs go backwards. This happens on
8706 * platforms that reset TSC during suspend or hibernate actions, but
8707 * maintain synchronization. We must compensate. Fortunately, we can
8708 * detect that condition here, which happens early in CPU bringup,
8709 * before any KVM threads can be running. Unfortunately, we can't
8710 * bring the TSCs fully up to date with real time, as we aren't yet far
8711 * enough into CPU bringup that we know how much real time has actually
8712 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
8713 * variables that haven't been updated yet.
8715 * So we simply find the maximum observed TSC above, then record the
8716 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
8717 * the adjustment will be applied. Note that we accumulate
8718 * adjustments, in case multiple suspend cycles happen before some VCPU
8719 * gets a chance to run again. In the event that no KVM threads get a
8720 * chance to run, we will miss the entire elapsed period, as we'll have
8721 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
8722 * loose cycle time. This isn't too big a deal, since the loss will be
8723 * uniform across all VCPUs (not to mention the scenario is extremely
8724 * unlikely). It is possible that a second hibernate recovery happens
8725 * much faster than a first, causing the observed TSC here to be
8726 * smaller; this would require additional padding adjustment, which is
8727 * why we set last_host_tsc to the local tsc observed here.
8729 * N.B. - this code below runs only on platforms with reliable TSC,
8730 * as that is the only way backwards_tsc is set above. Also note
8731 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
8732 * have the same delta_cyc adjustment applied if backwards_tsc
8733 * is detected. Note further, this adjustment is only done once,
8734 * as we reset last_host_tsc on all VCPUs to stop this from being
8735 * called multiple times (one for each physical CPU bringup).
8737 * Platforms with unreliable TSCs don't have to deal with this, they
8738 * will be compensated by the logic in vcpu_load, which sets the TSC to
8739 * catchup mode. This will catchup all VCPUs to real time, but cannot
8740 * guarantee that they stay in perfect synchronization.
8742 if (backwards_tsc) {
8743 u64 delta_cyc = max_tsc - local_tsc;
8744 list_for_each_entry(kvm, &vm_list, vm_list) {
8745 kvm->arch.backwards_tsc_observed = true;
8746 kvm_for_each_vcpu(i, vcpu, kvm) {
8747 vcpu->arch.tsc_offset_adjustment += delta_cyc;
8748 vcpu->arch.last_host_tsc = local_tsc;
8749 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8753 * We have to disable TSC offset matching.. if you were
8754 * booting a VM while issuing an S4 host suspend....
8755 * you may have some problem. Solving this issue is
8756 * left as an exercise to the reader.
8758 kvm->arch.last_tsc_nsec = 0;
8759 kvm->arch.last_tsc_write = 0;
8763 return 0;
8766 void kvm_arch_hardware_disable(void)
8768 kvm_x86_ops->hardware_disable();
8769 drop_user_return_notifiers();
8772 int kvm_arch_hardware_setup(void)
8774 int r;
8776 r = kvm_x86_ops->hardware_setup();
8777 if (r != 0)
8778 return r;
8780 if (kvm_has_tsc_control) {
8782 * Make sure the user can only configure tsc_khz values that
8783 * fit into a signed integer.
8784 * A min value is not calculated because it will always
8785 * be 1 on all machines.
8787 u64 max = min(0x7fffffffULL,
8788 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
8789 kvm_max_guest_tsc_khz = max;
8791 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
8794 kvm_init_msr_list();
8795 return 0;
8798 void kvm_arch_hardware_unsetup(void)
8800 kvm_x86_ops->hardware_unsetup();
8803 void kvm_arch_check_processor_compat(void *rtn)
8805 kvm_x86_ops->check_processor_compatibility(rtn);
8808 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
8810 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
8812 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
8814 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
8816 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
8819 struct static_key kvm_no_apic_vcpu __read_mostly;
8820 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
8822 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
8824 struct page *page;
8825 int r;
8827 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
8828 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
8829 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
8830 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8831 else
8832 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
8834 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
8835 if (!page) {
8836 r = -ENOMEM;
8837 goto fail;
8839 vcpu->arch.pio_data = page_address(page);
8841 kvm_set_tsc_khz(vcpu, max_tsc_khz);
8843 r = kvm_mmu_create(vcpu);
8844 if (r < 0)
8845 goto fail_free_pio_data;
8847 if (irqchip_in_kernel(vcpu->kvm)) {
8848 r = kvm_create_lapic(vcpu);
8849 if (r < 0)
8850 goto fail_mmu_destroy;
8851 } else
8852 static_key_slow_inc(&kvm_no_apic_vcpu);
8854 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
8855 GFP_KERNEL);
8856 if (!vcpu->arch.mce_banks) {
8857 r = -ENOMEM;
8858 goto fail_free_lapic;
8860 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
8862 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
8863 r = -ENOMEM;
8864 goto fail_free_mce_banks;
8867 fx_init(vcpu);
8869 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
8871 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
8873 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
8875 kvm_async_pf_hash_reset(vcpu);
8876 kvm_pmu_init(vcpu);
8878 vcpu->arch.pending_external_vector = -1;
8879 vcpu->arch.preempted_in_kernel = false;
8881 kvm_hv_vcpu_init(vcpu);
8883 return 0;
8885 fail_free_mce_banks:
8886 kfree(vcpu->arch.mce_banks);
8887 fail_free_lapic:
8888 kvm_free_lapic(vcpu);
8889 fail_mmu_destroy:
8890 kvm_mmu_destroy(vcpu);
8891 fail_free_pio_data:
8892 free_page((unsigned long)vcpu->arch.pio_data);
8893 fail:
8894 return r;
8897 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
8899 int idx;
8901 kvm_hv_vcpu_uninit(vcpu);
8902 kvm_pmu_destroy(vcpu);
8903 kfree(vcpu->arch.mce_banks);
8904 kvm_free_lapic(vcpu);
8905 idx = srcu_read_lock(&vcpu->kvm->srcu);
8906 kvm_mmu_destroy(vcpu);
8907 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8908 free_page((unsigned long)vcpu->arch.pio_data);
8909 if (!lapic_in_kernel(vcpu))
8910 static_key_slow_dec(&kvm_no_apic_vcpu);
8913 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
8915 vcpu->arch.l1tf_flush_l1d = true;
8916 kvm_x86_ops->sched_in(vcpu, cpu);
8919 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
8921 if (type)
8922 return -EINVAL;
8924 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
8925 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
8926 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
8927 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
8928 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
8930 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
8931 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
8932 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
8933 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
8934 &kvm->arch.irq_sources_bitmap);
8936 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
8937 mutex_init(&kvm->arch.apic_map_lock);
8938 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
8940 kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
8941 pvclock_update_vm_gtod_copy(kvm);
8943 kvm->arch.guest_can_read_msr_platform_info = true;
8945 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
8946 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
8948 kvm_hv_init_vm(kvm);
8949 kvm_page_track_init(kvm);
8950 kvm_mmu_init_vm(kvm);
8952 if (kvm_x86_ops->vm_init)
8953 return kvm_x86_ops->vm_init(kvm);
8955 return 0;
8958 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
8960 vcpu_load(vcpu);
8961 kvm_mmu_unload(vcpu);
8962 vcpu_put(vcpu);
8965 static void kvm_free_vcpus(struct kvm *kvm)
8967 unsigned int i;
8968 struct kvm_vcpu *vcpu;
8971 * Unpin any mmu pages first.
8973 kvm_for_each_vcpu(i, vcpu, kvm) {
8974 kvm_clear_async_pf_completion_queue(vcpu);
8975 kvm_unload_vcpu_mmu(vcpu);
8977 kvm_for_each_vcpu(i, vcpu, kvm)
8978 kvm_arch_vcpu_free(vcpu);
8980 mutex_lock(&kvm->lock);
8981 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
8982 kvm->vcpus[i] = NULL;
8984 atomic_set(&kvm->online_vcpus, 0);
8985 mutex_unlock(&kvm->lock);
8988 void kvm_arch_sync_events(struct kvm *kvm)
8990 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
8991 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
8992 kvm_free_pit(kvm);
8995 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8997 int i, r;
8998 unsigned long hva;
8999 struct kvm_memslots *slots = kvm_memslots(kvm);
9000 struct kvm_memory_slot *slot, old;
9002 /* Called with kvm->slots_lock held. */
9003 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9004 return -EINVAL;
9006 slot = id_to_memslot(slots, id);
9007 if (size) {
9008 if (slot->npages)
9009 return -EEXIST;
9012 * MAP_SHARED to prevent internal slot pages from being moved
9013 * by fork()/COW.
9015 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9016 MAP_SHARED | MAP_ANONYMOUS, 0);
9017 if (IS_ERR((void *)hva))
9018 return PTR_ERR((void *)hva);
9019 } else {
9020 if (!slot->npages)
9021 return 0;
9023 hva = 0;
9026 old = *slot;
9027 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9028 struct kvm_userspace_memory_region m;
9030 m.slot = id | (i << 16);
9031 m.flags = 0;
9032 m.guest_phys_addr = gpa;
9033 m.userspace_addr = hva;
9034 m.memory_size = size;
9035 r = __kvm_set_memory_region(kvm, &m);
9036 if (r < 0)
9037 return r;
9040 if (!size)
9041 vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9043 return 0;
9045 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9047 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9049 int r;
9051 mutex_lock(&kvm->slots_lock);
9052 r = __x86_set_memory_region(kvm, id, gpa, size);
9053 mutex_unlock(&kvm->slots_lock);
9055 return r;
9057 EXPORT_SYMBOL_GPL(x86_set_memory_region);
9059 void kvm_arch_destroy_vm(struct kvm *kvm)
9061 if (current->mm == kvm->mm) {
9063 * Free memory regions allocated on behalf of userspace,
9064 * unless the the memory map has changed due to process exit
9065 * or fd copying.
9067 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9068 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9069 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9071 if (kvm_x86_ops->vm_destroy)
9072 kvm_x86_ops->vm_destroy(kvm);
9073 kvm_pic_destroy(kvm);
9074 kvm_ioapic_destroy(kvm);
9075 kvm_free_vcpus(kvm);
9076 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9077 kvm_mmu_uninit_vm(kvm);
9078 kvm_page_track_cleanup(kvm);
9079 kvm_hv_destroy_vm(kvm);
9082 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9083 struct kvm_memory_slot *dont)
9085 int i;
9087 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9088 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9089 kvfree(free->arch.rmap[i]);
9090 free->arch.rmap[i] = NULL;
9092 if (i == 0)
9093 continue;
9095 if (!dont || free->arch.lpage_info[i - 1] !=
9096 dont->arch.lpage_info[i - 1]) {
9097 kvfree(free->arch.lpage_info[i - 1]);
9098 free->arch.lpage_info[i - 1] = NULL;
9102 kvm_page_track_free_memslot(free, dont);
9105 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9106 unsigned long npages)
9108 int i;
9110 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9111 struct kvm_lpage_info *linfo;
9112 unsigned long ugfn;
9113 int lpages;
9114 int level = i + 1;
9116 lpages = gfn_to_index(slot->base_gfn + npages - 1,
9117 slot->base_gfn, level) + 1;
9119 slot->arch.rmap[i] =
9120 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9121 GFP_KERNEL);
9122 if (!slot->arch.rmap[i])
9123 goto out_free;
9124 if (i == 0)
9125 continue;
9127 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL);
9128 if (!linfo)
9129 goto out_free;
9131 slot->arch.lpage_info[i - 1] = linfo;
9133 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9134 linfo[0].disallow_lpage = 1;
9135 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9136 linfo[lpages - 1].disallow_lpage = 1;
9137 ugfn = slot->userspace_addr >> PAGE_SHIFT;
9139 * If the gfn and userspace address are not aligned wrt each
9140 * other, or if explicitly asked to, disable large page
9141 * support for this slot
9143 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9144 !kvm_largepages_enabled()) {
9145 unsigned long j;
9147 for (j = 0; j < lpages; ++j)
9148 linfo[j].disallow_lpage = 1;
9152 if (kvm_page_track_create_memslot(slot, npages))
9153 goto out_free;
9155 return 0;
9157 out_free:
9158 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9159 kvfree(slot->arch.rmap[i]);
9160 slot->arch.rmap[i] = NULL;
9161 if (i == 0)
9162 continue;
9164 kvfree(slot->arch.lpage_info[i - 1]);
9165 slot->arch.lpage_info[i - 1] = NULL;
9167 return -ENOMEM;
9170 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
9173 * memslots->generation has been incremented.
9174 * mmio generation may have reached its maximum value.
9176 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
9179 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9180 struct kvm_memory_slot *memslot,
9181 const struct kvm_userspace_memory_region *mem,
9182 enum kvm_mr_change change)
9184 return 0;
9187 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9188 struct kvm_memory_slot *new)
9190 /* Still write protect RO slot */
9191 if (new->flags & KVM_MEM_READONLY) {
9192 kvm_mmu_slot_remove_write_access(kvm, new);
9193 return;
9197 * Call kvm_x86_ops dirty logging hooks when they are valid.
9199 * kvm_x86_ops->slot_disable_log_dirty is called when:
9201 * - KVM_MR_CREATE with dirty logging is disabled
9202 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9204 * The reason is, in case of PML, we need to set D-bit for any slots
9205 * with dirty logging disabled in order to eliminate unnecessary GPA
9206 * logging in PML buffer (and potential PML buffer full VMEXT). This
9207 * guarantees leaving PML enabled during guest's lifetime won't have
9208 * any additonal overhead from PML when guest is running with dirty
9209 * logging disabled for memory slots.
9211 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9212 * to dirty logging mode.
9214 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9216 * In case of write protect:
9218 * Write protect all pages for dirty logging.
9220 * All the sptes including the large sptes which point to this
9221 * slot are set to readonly. We can not create any new large
9222 * spte on this slot until the end of the logging.
9224 * See the comments in fast_page_fault().
9226 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9227 if (kvm_x86_ops->slot_enable_log_dirty)
9228 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9229 else
9230 kvm_mmu_slot_remove_write_access(kvm, new);
9231 } else {
9232 if (kvm_x86_ops->slot_disable_log_dirty)
9233 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9237 void kvm_arch_commit_memory_region(struct kvm *kvm,
9238 const struct kvm_userspace_memory_region *mem,
9239 const struct kvm_memory_slot *old,
9240 const struct kvm_memory_slot *new,
9241 enum kvm_mr_change change)
9243 int nr_mmu_pages = 0;
9245 if (!kvm->arch.n_requested_mmu_pages)
9246 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
9248 if (nr_mmu_pages)
9249 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
9252 * Dirty logging tracks sptes in 4k granularity, meaning that large
9253 * sptes have to be split. If live migration is successful, the guest
9254 * in the source machine will be destroyed and large sptes will be
9255 * created in the destination. However, if the guest continues to run
9256 * in the source machine (for example if live migration fails), small
9257 * sptes will remain around and cause bad performance.
9259 * Scan sptes if dirty logging has been stopped, dropping those
9260 * which can be collapsed into a single large-page spte. Later
9261 * page faults will create the large-page sptes.
9263 if ((change != KVM_MR_DELETE) &&
9264 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9265 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9266 kvm_mmu_zap_collapsible_sptes(kvm, new);
9269 * Set up write protection and/or dirty logging for the new slot.
9271 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9272 * been zapped so no dirty logging staff is needed for old slot. For
9273 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9274 * new and it's also covered when dealing with the new slot.
9276 * FIXME: const-ify all uses of struct kvm_memory_slot.
9278 if (change != KVM_MR_DELETE)
9279 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9282 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9284 kvm_mmu_invalidate_zap_all_pages(kvm);
9287 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9288 struct kvm_memory_slot *slot)
9290 kvm_page_track_flush_slot(kvm, slot);
9293 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9295 return (is_guest_mode(vcpu) &&
9296 kvm_x86_ops->guest_apic_has_interrupt &&
9297 kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9300 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9302 if (!list_empty_careful(&vcpu->async_pf.done))
9303 return true;
9305 if (kvm_apic_has_events(vcpu))
9306 return true;
9308 if (vcpu->arch.pv.pv_unhalted)
9309 return true;
9311 if (vcpu->arch.exception.pending)
9312 return true;
9314 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9315 (vcpu->arch.nmi_pending &&
9316 kvm_x86_ops->nmi_allowed(vcpu)))
9317 return true;
9319 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9320 (vcpu->arch.smi_pending && !is_smm(vcpu)))
9321 return true;
9323 if (kvm_arch_interrupt_allowed(vcpu) &&
9324 (kvm_cpu_has_interrupt(vcpu) ||
9325 kvm_guest_apic_has_interrupt(vcpu)))
9326 return true;
9328 if (kvm_hv_has_stimer_pending(vcpu))
9329 return true;
9331 return false;
9334 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9336 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9339 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
9341 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
9342 return true;
9344 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9345 kvm_test_request(KVM_REQ_SMI, vcpu) ||
9346 kvm_test_request(KVM_REQ_EVENT, vcpu))
9347 return true;
9349 if (vcpu->arch.apicv_active && kvm_x86_ops->dy_apicv_has_pending_interrupt(vcpu))
9350 return true;
9352 return false;
9355 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9357 return vcpu->arch.preempted_in_kernel;
9360 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9362 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9365 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9367 return kvm_x86_ops->interrupt_allowed(vcpu);
9370 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9372 if (is_64_bit_mode(vcpu))
9373 return kvm_rip_read(vcpu);
9374 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9375 kvm_rip_read(vcpu));
9377 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9379 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9381 return kvm_get_linear_rip(vcpu) == linear_rip;
9383 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9385 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9387 unsigned long rflags;
9389 rflags = kvm_x86_ops->get_rflags(vcpu);
9390 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9391 rflags &= ~X86_EFLAGS_TF;
9392 return rflags;
9394 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9396 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9398 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9399 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9400 rflags |= X86_EFLAGS_TF;
9401 kvm_x86_ops->set_rflags(vcpu, rflags);
9404 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9406 __kvm_set_rflags(vcpu, rflags);
9407 kvm_make_request(KVM_REQ_EVENT, vcpu);
9409 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9411 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9413 int r;
9415 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
9416 work->wakeup_all)
9417 return;
9419 r = kvm_mmu_reload(vcpu);
9420 if (unlikely(r))
9421 return;
9423 if (!vcpu->arch.mmu.direct_map &&
9424 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
9425 return;
9427 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
9430 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9432 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9435 static inline u32 kvm_async_pf_next_probe(u32 key)
9437 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9440 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9442 u32 key = kvm_async_pf_hash_fn(gfn);
9444 while (vcpu->arch.apf.gfns[key] != ~0)
9445 key = kvm_async_pf_next_probe(key);
9447 vcpu->arch.apf.gfns[key] = gfn;
9450 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9452 int i;
9453 u32 key = kvm_async_pf_hash_fn(gfn);
9455 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9456 (vcpu->arch.apf.gfns[key] != gfn &&
9457 vcpu->arch.apf.gfns[key] != ~0); i++)
9458 key = kvm_async_pf_next_probe(key);
9460 return key;
9463 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9465 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9468 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9470 u32 i, j, k;
9472 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9473 while (true) {
9474 vcpu->arch.apf.gfns[i] = ~0;
9475 do {
9476 j = kvm_async_pf_next_probe(j);
9477 if (vcpu->arch.apf.gfns[j] == ~0)
9478 return;
9479 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
9481 * k lies cyclically in ]i,j]
9482 * | i.k.j |
9483 * |....j i.k.| or |.k..j i...|
9485 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
9486 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
9487 i = j;
9491 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
9494 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
9495 sizeof(val));
9498 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
9501 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
9502 sizeof(u32));
9505 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
9506 struct kvm_async_pf *work)
9508 struct x86_exception fault;
9510 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
9511 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
9513 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
9514 (vcpu->arch.apf.send_user_only &&
9515 kvm_x86_ops->get_cpl(vcpu) == 0))
9516 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
9517 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
9518 fault.vector = PF_VECTOR;
9519 fault.error_code_valid = true;
9520 fault.error_code = 0;
9521 fault.nested_page_fault = false;
9522 fault.address = work->arch.token;
9523 fault.async_page_fault = true;
9524 kvm_inject_page_fault(vcpu, &fault);
9528 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
9529 struct kvm_async_pf *work)
9531 struct x86_exception fault;
9532 u32 val;
9534 if (work->wakeup_all)
9535 work->arch.token = ~0; /* broadcast wakeup */
9536 else
9537 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
9538 trace_kvm_async_pf_ready(work->arch.token, work->gva);
9540 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
9541 !apf_get_user(vcpu, &val)) {
9542 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
9543 vcpu->arch.exception.pending &&
9544 vcpu->arch.exception.nr == PF_VECTOR &&
9545 !apf_put_user(vcpu, 0)) {
9546 vcpu->arch.exception.injected = false;
9547 vcpu->arch.exception.pending = false;
9548 vcpu->arch.exception.nr = 0;
9549 vcpu->arch.exception.has_error_code = false;
9550 vcpu->arch.exception.error_code = 0;
9551 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
9552 fault.vector = PF_VECTOR;
9553 fault.error_code_valid = true;
9554 fault.error_code = 0;
9555 fault.nested_page_fault = false;
9556 fault.address = work->arch.token;
9557 fault.async_page_fault = true;
9558 kvm_inject_page_fault(vcpu, &fault);
9561 vcpu->arch.apf.halted = false;
9562 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9565 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
9567 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
9568 return true;
9569 else
9570 return kvm_can_do_async_pf(vcpu);
9573 void kvm_arch_start_assignment(struct kvm *kvm)
9575 atomic_inc(&kvm->arch.assigned_device_count);
9577 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
9579 void kvm_arch_end_assignment(struct kvm *kvm)
9581 atomic_dec(&kvm->arch.assigned_device_count);
9583 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
9585 bool kvm_arch_has_assigned_device(struct kvm *kvm)
9587 return atomic_read(&kvm->arch.assigned_device_count);
9589 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
9591 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
9593 atomic_inc(&kvm->arch.noncoherent_dma_count);
9595 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
9597 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
9599 atomic_dec(&kvm->arch.noncoherent_dma_count);
9601 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
9603 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
9605 return atomic_read(&kvm->arch.noncoherent_dma_count);
9607 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
9609 bool kvm_arch_has_irq_bypass(void)
9611 return kvm_x86_ops->update_pi_irte != NULL;
9614 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
9615 struct irq_bypass_producer *prod)
9617 struct kvm_kernel_irqfd *irqfd =
9618 container_of(cons, struct kvm_kernel_irqfd, consumer);
9620 irqfd->producer = prod;
9622 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
9623 prod->irq, irqfd->gsi, 1);
9626 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
9627 struct irq_bypass_producer *prod)
9629 int ret;
9630 struct kvm_kernel_irqfd *irqfd =
9631 container_of(cons, struct kvm_kernel_irqfd, consumer);
9633 WARN_ON(irqfd->producer != prod);
9634 irqfd->producer = NULL;
9637 * When producer of consumer is unregistered, we change back to
9638 * remapped mode, so we can re-use the current implementation
9639 * when the irq is masked/disabled or the consumer side (KVM
9640 * int this case doesn't want to receive the interrupts.
9642 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
9643 if (ret)
9644 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
9645 " fails: %d\n", irqfd->consumer.token, ret);
9648 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
9649 uint32_t guest_irq, bool set)
9651 if (!kvm_x86_ops->update_pi_irte)
9652 return -EINVAL;
9654 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
9657 bool kvm_vector_hashing_enabled(void)
9659 return vector_hashing;
9661 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
9663 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
9664 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
9665 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
9666 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
9667 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
9668 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
9669 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
9670 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
9671 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
9672 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
9673 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
9674 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
9675 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
9676 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
9677 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
9678 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
9679 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
9680 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
9681 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);