]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - arch/x86/kvm/x86.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
KVM: x86: pass kvm_vcpu to kvm_read_guest_virt and kvm_write_guest_virt_system
[mirror_ubuntu-bionic-kernel.git] / arch / x86 / kvm / x86.c
1 /*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * derived from drivers/kvm/kvm_main.c
5 *
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.
10 *
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>
16 *
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
19 *
20 */
21
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"
32
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>
58
59 #include <trace/events/kvm.h>
60
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
71 #define CREATE_TRACE_POINTS
72 #include "trace.h"
73
74 #define MAX_IO_MSRS 256
75 #define KVM_MAX_MCE_BANKS 32
76 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
77 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
78
79 #define emul_to_vcpu(ctxt) \
80 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
81
82 /* EFER defaults:
83 * - enable syscall per default because its emulated by KVM
84 * - enable LME and LMA per default on 64 bit KVM
85 */
86 #ifdef CONFIG_X86_64
87 static
88 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
89 #else
90 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
91 #endif
92
93 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
94 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
95
96 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
97 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
98
99 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
100 static void process_nmi(struct kvm_vcpu *vcpu);
101 static void enter_smm(struct kvm_vcpu *vcpu);
102 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
103
104 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
105 EXPORT_SYMBOL_GPL(kvm_x86_ops);
106
107 static bool __read_mostly ignore_msrs = 0;
108 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
109
110 static bool __read_mostly report_ignored_msrs = true;
111 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
112
113 unsigned int min_timer_period_us = 500;
114 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
115
116 static bool __read_mostly kvmclock_periodic_sync = true;
117 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
118
119 bool __read_mostly kvm_has_tsc_control;
120 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
121 u32 __read_mostly kvm_max_guest_tsc_khz;
122 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
123 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
124 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
125 u64 __read_mostly kvm_max_tsc_scaling_ratio;
126 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
127 u64 __read_mostly kvm_default_tsc_scaling_ratio;
128 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
129
130 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
131 static u32 __read_mostly tsc_tolerance_ppm = 250;
132 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
133
134 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
135 unsigned int __read_mostly lapic_timer_advance_ns = 0;
136 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
137
138 static bool __read_mostly vector_hashing = true;
139 module_param(vector_hashing, bool, S_IRUGO);
140
141 #define KVM_NR_SHARED_MSRS 16
142
143 struct kvm_shared_msrs_global {
144 int nr;
145 u32 msrs[KVM_NR_SHARED_MSRS];
146 };
147
148 struct kvm_shared_msrs {
149 struct user_return_notifier urn;
150 bool registered;
151 struct kvm_shared_msr_values {
152 u64 host;
153 u64 curr;
154 } values[KVM_NR_SHARED_MSRS];
155 };
156
157 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
158 static struct kvm_shared_msrs __percpu *shared_msrs;
159
160 struct kvm_stats_debugfs_item debugfs_entries[] = {
161 { "pf_fixed", VCPU_STAT(pf_fixed) },
162 { "pf_guest", VCPU_STAT(pf_guest) },
163 { "tlb_flush", VCPU_STAT(tlb_flush) },
164 { "invlpg", VCPU_STAT(invlpg) },
165 { "exits", VCPU_STAT(exits) },
166 { "io_exits", VCPU_STAT(io_exits) },
167 { "mmio_exits", VCPU_STAT(mmio_exits) },
168 { "signal_exits", VCPU_STAT(signal_exits) },
169 { "irq_window", VCPU_STAT(irq_window_exits) },
170 { "nmi_window", VCPU_STAT(nmi_window_exits) },
171 { "halt_exits", VCPU_STAT(halt_exits) },
172 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
173 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
174 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
175 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
176 { "hypercalls", VCPU_STAT(hypercalls) },
177 { "request_irq", VCPU_STAT(request_irq_exits) },
178 { "irq_exits", VCPU_STAT(irq_exits) },
179 { "host_state_reload", VCPU_STAT(host_state_reload) },
180 { "efer_reload", VCPU_STAT(efer_reload) },
181 { "fpu_reload", VCPU_STAT(fpu_reload) },
182 { "insn_emulation", VCPU_STAT(insn_emulation) },
183 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
184 { "irq_injections", VCPU_STAT(irq_injections) },
185 { "nmi_injections", VCPU_STAT(nmi_injections) },
186 { "req_event", VCPU_STAT(req_event) },
187 { "l1d_flush", VCPU_STAT(l1d_flush) },
188 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
189 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
190 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
191 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
192 { "mmu_flooded", VM_STAT(mmu_flooded) },
193 { "mmu_recycled", VM_STAT(mmu_recycled) },
194 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
195 { "mmu_unsync", VM_STAT(mmu_unsync) },
196 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
197 { "largepages", VM_STAT(lpages) },
198 { "max_mmu_page_hash_collisions",
199 VM_STAT(max_mmu_page_hash_collisions) },
200 { NULL }
201 };
202
203 u64 __read_mostly host_xcr0;
204
205 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
206
207 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
208 {
209 int i;
210 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
211 vcpu->arch.apf.gfns[i] = ~0;
212 }
213
214 static void kvm_on_user_return(struct user_return_notifier *urn)
215 {
216 unsigned slot;
217 struct kvm_shared_msrs *locals
218 = container_of(urn, struct kvm_shared_msrs, urn);
219 struct kvm_shared_msr_values *values;
220 unsigned long flags;
221
222 /*
223 * Disabling irqs at this point since the following code could be
224 * interrupted and executed through kvm_arch_hardware_disable()
225 */
226 local_irq_save(flags);
227 if (locals->registered) {
228 locals->registered = false;
229 user_return_notifier_unregister(urn);
230 }
231 local_irq_restore(flags);
232 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
233 values = &locals->values[slot];
234 if (values->host != values->curr) {
235 wrmsrl(shared_msrs_global.msrs[slot], values->host);
236 values->curr = values->host;
237 }
238 }
239 }
240
241 static void shared_msr_update(unsigned slot, u32 msr)
242 {
243 u64 value;
244 unsigned int cpu = smp_processor_id();
245 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
246
247 /* only read, and nobody should modify it at this time,
248 * so don't need lock */
249 if (slot >= shared_msrs_global.nr) {
250 printk(KERN_ERR "kvm: invalid MSR slot!");
251 return;
252 }
253 rdmsrl_safe(msr, &value);
254 smsr->values[slot].host = value;
255 smsr->values[slot].curr = value;
256 }
257
258 void kvm_define_shared_msr(unsigned slot, u32 msr)
259 {
260 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
261 shared_msrs_global.msrs[slot] = msr;
262 if (slot >= shared_msrs_global.nr)
263 shared_msrs_global.nr = slot + 1;
264 }
265 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
266
267 static void kvm_shared_msr_cpu_online(void)
268 {
269 unsigned i;
270
271 for (i = 0; i < shared_msrs_global.nr; ++i)
272 shared_msr_update(i, shared_msrs_global.msrs[i]);
273 }
274
275 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
276 {
277 unsigned int cpu = smp_processor_id();
278 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
279 int err;
280
281 if (((value ^ smsr->values[slot].curr) & mask) == 0)
282 return 0;
283 smsr->values[slot].curr = value;
284 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
285 if (err)
286 return 1;
287
288 if (!smsr->registered) {
289 smsr->urn.on_user_return = kvm_on_user_return;
290 user_return_notifier_register(&smsr->urn);
291 smsr->registered = true;
292 }
293 return 0;
294 }
295 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
296
297 static void drop_user_return_notifiers(void)
298 {
299 unsigned int cpu = smp_processor_id();
300 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
301
302 if (smsr->registered)
303 kvm_on_user_return(&smsr->urn);
304 }
305
306 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
307 {
308 return vcpu->arch.apic_base;
309 }
310 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
311
312 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
313 {
314 u64 old_state = vcpu->arch.apic_base &
315 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
316 u64 new_state = msr_info->data &
317 (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE);
318 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
319 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
320
321 if ((msr_info->data & reserved_bits) || new_state == X2APIC_ENABLE)
322 return 1;
323 if (!msr_info->host_initiated &&
324 ((new_state == MSR_IA32_APICBASE_ENABLE &&
325 old_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE)) ||
326 (new_state == (MSR_IA32_APICBASE_ENABLE | X2APIC_ENABLE) &&
327 old_state == 0)))
328 return 1;
329
330 kvm_lapic_set_base(vcpu, msr_info->data);
331 return 0;
332 }
333 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
334
335 asmlinkage __visible void kvm_spurious_fault(void)
336 {
337 /* Fault while not rebooting. We want the trace. */
338 BUG();
339 }
340 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
341
342 #define EXCPT_BENIGN 0
343 #define EXCPT_CONTRIBUTORY 1
344 #define EXCPT_PF 2
345
346 static int exception_class(int vector)
347 {
348 switch (vector) {
349 case PF_VECTOR:
350 return EXCPT_PF;
351 case DE_VECTOR:
352 case TS_VECTOR:
353 case NP_VECTOR:
354 case SS_VECTOR:
355 case GP_VECTOR:
356 return EXCPT_CONTRIBUTORY;
357 default:
358 break;
359 }
360 return EXCPT_BENIGN;
361 }
362
363 #define EXCPT_FAULT 0
364 #define EXCPT_TRAP 1
365 #define EXCPT_ABORT 2
366 #define EXCPT_INTERRUPT 3
367
368 static int exception_type(int vector)
369 {
370 unsigned int mask;
371
372 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
373 return EXCPT_INTERRUPT;
374
375 mask = 1 << vector;
376
377 /* #DB is trap, as instruction watchpoints are handled elsewhere */
378 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
379 return EXCPT_TRAP;
380
381 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
382 return EXCPT_ABORT;
383
384 /* Reserved exceptions will result in fault */
385 return EXCPT_FAULT;
386 }
387
388 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
389 unsigned nr, bool has_error, u32 error_code,
390 bool reinject)
391 {
392 u32 prev_nr;
393 int class1, class2;
394
395 kvm_make_request(KVM_REQ_EVENT, vcpu);
396
397 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
398 queue:
399 if (has_error && !is_protmode(vcpu))
400 has_error = false;
401 if (reinject) {
402 /*
403 * On vmentry, vcpu->arch.exception.pending is only
404 * true if an event injection was blocked by
405 * nested_run_pending. In that case, however,
406 * vcpu_enter_guest requests an immediate exit,
407 * and the guest shouldn't proceed far enough to
408 * need reinjection.
409 */
410 WARN_ON_ONCE(vcpu->arch.exception.pending);
411 vcpu->arch.exception.injected = true;
412 } else {
413 vcpu->arch.exception.pending = true;
414 vcpu->arch.exception.injected = false;
415 }
416 vcpu->arch.exception.has_error_code = has_error;
417 vcpu->arch.exception.nr = nr;
418 vcpu->arch.exception.error_code = error_code;
419 return;
420 }
421
422 /* to check exception */
423 prev_nr = vcpu->arch.exception.nr;
424 if (prev_nr == DF_VECTOR) {
425 /* triple fault -> shutdown */
426 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
427 return;
428 }
429 class1 = exception_class(prev_nr);
430 class2 = exception_class(nr);
431 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
432 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
433 /*
434 * Generate double fault per SDM Table 5-5. Set
435 * exception.pending = true so that the double fault
436 * can trigger a nested vmexit.
437 */
438 vcpu->arch.exception.pending = true;
439 vcpu->arch.exception.injected = false;
440 vcpu->arch.exception.has_error_code = true;
441 vcpu->arch.exception.nr = DF_VECTOR;
442 vcpu->arch.exception.error_code = 0;
443 } else
444 /* replace previous exception with a new one in a hope
445 that instruction re-execution will regenerate lost
446 exception */
447 goto queue;
448 }
449
450 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
451 {
452 kvm_multiple_exception(vcpu, nr, false, 0, false);
453 }
454 EXPORT_SYMBOL_GPL(kvm_queue_exception);
455
456 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
457 {
458 kvm_multiple_exception(vcpu, nr, false, 0, true);
459 }
460 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
461
462 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
463 {
464 if (err)
465 kvm_inject_gp(vcpu, 0);
466 else
467 return kvm_skip_emulated_instruction(vcpu);
468
469 return 1;
470 }
471 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
472
473 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
474 {
475 ++vcpu->stat.pf_guest;
476 vcpu->arch.exception.nested_apf =
477 is_guest_mode(vcpu) && fault->async_page_fault;
478 if (vcpu->arch.exception.nested_apf)
479 vcpu->arch.apf.nested_apf_token = fault->address;
480 else
481 vcpu->arch.cr2 = fault->address;
482 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
483 }
484 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
485
486 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
487 {
488 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
489 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
490 else
491 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
492
493 return fault->nested_page_fault;
494 }
495
496 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
497 {
498 atomic_inc(&vcpu->arch.nmi_queued);
499 kvm_make_request(KVM_REQ_NMI, vcpu);
500 }
501 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
502
503 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
504 {
505 kvm_multiple_exception(vcpu, nr, true, error_code, false);
506 }
507 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
508
509 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
510 {
511 kvm_multiple_exception(vcpu, nr, true, error_code, true);
512 }
513 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
514
515 /*
516 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
517 * a #GP and return false.
518 */
519 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
520 {
521 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
522 return true;
523 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
524 return false;
525 }
526 EXPORT_SYMBOL_GPL(kvm_require_cpl);
527
528 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
529 {
530 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
531 return true;
532
533 kvm_queue_exception(vcpu, UD_VECTOR);
534 return false;
535 }
536 EXPORT_SYMBOL_GPL(kvm_require_dr);
537
538 /*
539 * This function will be used to read from the physical memory of the currently
540 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
541 * can read from guest physical or from the guest's guest physical memory.
542 */
543 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
544 gfn_t ngfn, void *data, int offset, int len,
545 u32 access)
546 {
547 struct x86_exception exception;
548 gfn_t real_gfn;
549 gpa_t ngpa;
550
551 ngpa = gfn_to_gpa(ngfn);
552 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
553 if (real_gfn == UNMAPPED_GVA)
554 return -EFAULT;
555
556 real_gfn = gpa_to_gfn(real_gfn);
557
558 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
559 }
560 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
561
562 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
563 void *data, int offset, int len, u32 access)
564 {
565 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
566 data, offset, len, access);
567 }
568
569 /*
570 * Load the pae pdptrs. Return true is they are all valid.
571 */
572 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
573 {
574 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
575 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
576 int i;
577 int ret;
578 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
579
580 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
581 offset * sizeof(u64), sizeof(pdpte),
582 PFERR_USER_MASK|PFERR_WRITE_MASK);
583 if (ret < 0) {
584 ret = 0;
585 goto out;
586 }
587 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
588 if ((pdpte[i] & PT_PRESENT_MASK) &&
589 (pdpte[i] &
590 vcpu->arch.mmu.guest_rsvd_check.rsvd_bits_mask[0][2])) {
591 ret = 0;
592 goto out;
593 }
594 }
595 ret = 1;
596
597 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
598 __set_bit(VCPU_EXREG_PDPTR,
599 (unsigned long *)&vcpu->arch.regs_avail);
600 __set_bit(VCPU_EXREG_PDPTR,
601 (unsigned long *)&vcpu->arch.regs_dirty);
602 out:
603
604 return ret;
605 }
606 EXPORT_SYMBOL_GPL(load_pdptrs);
607
608 bool pdptrs_changed(struct kvm_vcpu *vcpu)
609 {
610 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
611 bool changed = true;
612 int offset;
613 gfn_t gfn;
614 int r;
615
616 if (is_long_mode(vcpu) || !is_pae(vcpu))
617 return false;
618
619 if (!test_bit(VCPU_EXREG_PDPTR,
620 (unsigned long *)&vcpu->arch.regs_avail))
621 return true;
622
623 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
624 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
625 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
626 PFERR_USER_MASK | PFERR_WRITE_MASK);
627 if (r < 0)
628 goto out;
629 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
630 out:
631
632 return changed;
633 }
634 EXPORT_SYMBOL_GPL(pdptrs_changed);
635
636 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
637 {
638 unsigned long old_cr0 = kvm_read_cr0(vcpu);
639 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
640
641 cr0 |= X86_CR0_ET;
642
643 #ifdef CONFIG_X86_64
644 if (cr0 & 0xffffffff00000000UL)
645 return 1;
646 #endif
647
648 cr0 &= ~CR0_RESERVED_BITS;
649
650 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
651 return 1;
652
653 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
654 return 1;
655
656 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
657 #ifdef CONFIG_X86_64
658 if ((vcpu->arch.efer & EFER_LME)) {
659 int cs_db, cs_l;
660
661 if (!is_pae(vcpu))
662 return 1;
663 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
664 if (cs_l)
665 return 1;
666 } else
667 #endif
668 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
669 kvm_read_cr3(vcpu)))
670 return 1;
671 }
672
673 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
674 return 1;
675
676 kvm_x86_ops->set_cr0(vcpu, cr0);
677
678 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
679 kvm_clear_async_pf_completion_queue(vcpu);
680 kvm_async_pf_hash_reset(vcpu);
681 }
682
683 if ((cr0 ^ old_cr0) & update_bits)
684 kvm_mmu_reset_context(vcpu);
685
686 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
687 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
688 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
689 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
690
691 return 0;
692 }
693 EXPORT_SYMBOL_GPL(kvm_set_cr0);
694
695 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
696 {
697 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
698 }
699 EXPORT_SYMBOL_GPL(kvm_lmsw);
700
701 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
702 {
703 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
704 !vcpu->guest_xcr0_loaded) {
705 /* kvm_set_xcr() also depends on this */
706 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
707 vcpu->guest_xcr0_loaded = 1;
708 }
709 }
710
711 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
712 {
713 if (vcpu->guest_xcr0_loaded) {
714 if (vcpu->arch.xcr0 != host_xcr0)
715 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
716 vcpu->guest_xcr0_loaded = 0;
717 }
718 }
719
720 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
721 {
722 u64 xcr0 = xcr;
723 u64 old_xcr0 = vcpu->arch.xcr0;
724 u64 valid_bits;
725
726 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
727 if (index != XCR_XFEATURE_ENABLED_MASK)
728 return 1;
729 if (!(xcr0 & XFEATURE_MASK_FP))
730 return 1;
731 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
732 return 1;
733
734 /*
735 * Do not allow the guest to set bits that we do not support
736 * saving. However, xcr0 bit 0 is always set, even if the
737 * emulated CPU does not support XSAVE (see fx_init).
738 */
739 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
740 if (xcr0 & ~valid_bits)
741 return 1;
742
743 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
744 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
745 return 1;
746
747 if (xcr0 & XFEATURE_MASK_AVX512) {
748 if (!(xcr0 & XFEATURE_MASK_YMM))
749 return 1;
750 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
751 return 1;
752 }
753 vcpu->arch.xcr0 = xcr0;
754
755 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
756 kvm_update_cpuid(vcpu);
757 return 0;
758 }
759
760 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
761 {
762 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
763 __kvm_set_xcr(vcpu, index, xcr)) {
764 kvm_inject_gp(vcpu, 0);
765 return 1;
766 }
767 return 0;
768 }
769 EXPORT_SYMBOL_GPL(kvm_set_xcr);
770
771 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
772 {
773 unsigned long old_cr4 = kvm_read_cr4(vcpu);
774 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
775 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
776
777 if (cr4 & CR4_RESERVED_BITS)
778 return 1;
779
780 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
781 return 1;
782
783 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
784 return 1;
785
786 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
787 return 1;
788
789 if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
790 return 1;
791
792 if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
793 return 1;
794
795 if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
796 return 1;
797
798 if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
799 return 1;
800
801 if (is_long_mode(vcpu)) {
802 if (!(cr4 & X86_CR4_PAE))
803 return 1;
804 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
805 && ((cr4 ^ old_cr4) & pdptr_bits)
806 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
807 kvm_read_cr3(vcpu)))
808 return 1;
809
810 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
811 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
812 return 1;
813
814 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
815 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
816 return 1;
817 }
818
819 if (kvm_x86_ops->set_cr4(vcpu, cr4))
820 return 1;
821
822 if (((cr4 ^ old_cr4) & pdptr_bits) ||
823 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
824 kvm_mmu_reset_context(vcpu);
825
826 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
827 kvm_update_cpuid(vcpu);
828
829 return 0;
830 }
831 EXPORT_SYMBOL_GPL(kvm_set_cr4);
832
833 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
834 {
835 #ifdef CONFIG_X86_64
836 cr3 &= ~CR3_PCID_INVD;
837 #endif
838
839 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
840 kvm_mmu_sync_roots(vcpu);
841 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
842 return 0;
843 }
844
845 if (is_long_mode(vcpu) &&
846 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
847 return 1;
848 else if (is_pae(vcpu) && is_paging(vcpu) &&
849 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
850 return 1;
851
852 vcpu->arch.cr3 = cr3;
853 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
854 kvm_mmu_new_cr3(vcpu);
855 return 0;
856 }
857 EXPORT_SYMBOL_GPL(kvm_set_cr3);
858
859 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
860 {
861 if (cr8 & CR8_RESERVED_BITS)
862 return 1;
863 if (lapic_in_kernel(vcpu))
864 kvm_lapic_set_tpr(vcpu, cr8);
865 else
866 vcpu->arch.cr8 = cr8;
867 return 0;
868 }
869 EXPORT_SYMBOL_GPL(kvm_set_cr8);
870
871 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
872 {
873 if (lapic_in_kernel(vcpu))
874 return kvm_lapic_get_cr8(vcpu);
875 else
876 return vcpu->arch.cr8;
877 }
878 EXPORT_SYMBOL_GPL(kvm_get_cr8);
879
880 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
881 {
882 int i;
883
884 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
885 for (i = 0; i < KVM_NR_DB_REGS; i++)
886 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
887 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
888 }
889 }
890
891 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
892 {
893 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
894 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
895 }
896
897 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
898 {
899 unsigned long dr7;
900
901 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
902 dr7 = vcpu->arch.guest_debug_dr7;
903 else
904 dr7 = vcpu->arch.dr7;
905 kvm_x86_ops->set_dr7(vcpu, dr7);
906 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
907 if (dr7 & DR7_BP_EN_MASK)
908 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
909 }
910
911 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
912 {
913 u64 fixed = DR6_FIXED_1;
914
915 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
916 fixed |= DR6_RTM;
917 return fixed;
918 }
919
920 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
921 {
922 switch (dr) {
923 case 0 ... 3:
924 vcpu->arch.db[dr] = val;
925 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
926 vcpu->arch.eff_db[dr] = val;
927 break;
928 case 4:
929 /* fall through */
930 case 6:
931 if (val & 0xffffffff00000000ULL)
932 return -1; /* #GP */
933 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
934 kvm_update_dr6(vcpu);
935 break;
936 case 5:
937 /* fall through */
938 default: /* 7 */
939 if (val & 0xffffffff00000000ULL)
940 return -1; /* #GP */
941 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
942 kvm_update_dr7(vcpu);
943 break;
944 }
945
946 return 0;
947 }
948
949 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
950 {
951 if (__kvm_set_dr(vcpu, dr, val)) {
952 kvm_inject_gp(vcpu, 0);
953 return 1;
954 }
955 return 0;
956 }
957 EXPORT_SYMBOL_GPL(kvm_set_dr);
958
959 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
960 {
961 switch (dr) {
962 case 0 ... 3:
963 *val = vcpu->arch.db[dr];
964 break;
965 case 4:
966 /* fall through */
967 case 6:
968 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
969 *val = vcpu->arch.dr6;
970 else
971 *val = kvm_x86_ops->get_dr6(vcpu);
972 break;
973 case 5:
974 /* fall through */
975 default: /* 7 */
976 *val = vcpu->arch.dr7;
977 break;
978 }
979 return 0;
980 }
981 EXPORT_SYMBOL_GPL(kvm_get_dr);
982
983 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
984 {
985 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
986 u64 data;
987 int err;
988
989 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
990 if (err)
991 return err;
992 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
993 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
994 return err;
995 }
996 EXPORT_SYMBOL_GPL(kvm_rdpmc);
997
998 /*
999 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1000 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1001 *
1002 * This list is modified at module load time to reflect the
1003 * capabilities of the host cpu. This capabilities test skips MSRs that are
1004 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1005 * may depend on host virtualization features rather than host cpu features.
1006 */
1007
1008 static u32 msrs_to_save[] = {
1009 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1010 MSR_STAR,
1011 #ifdef CONFIG_X86_64
1012 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1013 #endif
1014 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1015 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1016 MSR_IA32_SPEC_CTRL, MSR_IA32_ARCH_CAPABILITIES
1017 };
1018
1019 static unsigned num_msrs_to_save;
1020
1021 static u32 emulated_msrs[] = {
1022 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1023 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1024 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1025 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1026 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1027 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1028 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1029 HV_X64_MSR_RESET,
1030 HV_X64_MSR_VP_INDEX,
1031 HV_X64_MSR_VP_RUNTIME,
1032 HV_X64_MSR_SCONTROL,
1033 HV_X64_MSR_STIMER0_CONFIG,
1034 HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1035 MSR_KVM_PV_EOI_EN,
1036
1037 MSR_IA32_TSC_ADJUST,
1038 MSR_IA32_TSCDEADLINE,
1039 MSR_IA32_MISC_ENABLE,
1040 MSR_IA32_MCG_STATUS,
1041 MSR_IA32_MCG_CTL,
1042 MSR_IA32_MCG_EXT_CTL,
1043 MSR_IA32_SMBASE,
1044 MSR_PLATFORM_INFO,
1045 MSR_MISC_FEATURES_ENABLES,
1046 MSR_AMD64_VIRT_SPEC_CTRL,
1047 };
1048
1049 static unsigned num_emulated_msrs;
1050
1051 /*
1052 * List of msr numbers which are used to expose MSR-based features that
1053 * can be used by a hypervisor to validate requested CPU features.
1054 */
1055 static u32 msr_based_features[] = {
1056 MSR_IA32_ARCH_CAPABILITIES,
1057 };
1058
1059 static unsigned int num_msr_based_features;
1060
1061 u64 kvm_get_arch_capabilities(void)
1062 {
1063 u64 data;
1064
1065 rdmsrl_safe(MSR_IA32_ARCH_CAPABILITIES, &data);
1066
1067 /*
1068 * If we're doing cache flushes (either "always" or "cond")
1069 * we will do one whenever the guest does a vmlaunch/vmresume.
1070 * If an outer hypervisor is doing the cache flush for us
1071 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1072 * capability to the guest too, and if EPT is disabled we're not
1073 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1074 * require a nested hypervisor to do a flush of its own.
1075 */
1076 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1077 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1078
1079 return data;
1080 }
1081 EXPORT_SYMBOL_GPL(kvm_get_arch_capabilities);
1082
1083 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1084 {
1085 switch (msr->index) {
1086 case MSR_IA32_ARCH_CAPABILITIES:
1087 msr->data = kvm_get_arch_capabilities();
1088 break;
1089 default:
1090 if (kvm_x86_ops->get_msr_feature(msr))
1091 return 1;
1092 }
1093 return 0;
1094 }
1095
1096 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1097 {
1098 struct kvm_msr_entry msr;
1099 int r;
1100
1101 msr.index = index;
1102 r = kvm_get_msr_feature(&msr);
1103 if (r)
1104 return r;
1105
1106 *data = msr.data;
1107
1108 return 0;
1109 }
1110
1111 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1112 {
1113 if (efer & efer_reserved_bits)
1114 return false;
1115
1116 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1117 return false;
1118
1119 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1120 return false;
1121
1122 return true;
1123 }
1124 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1125
1126 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1127 {
1128 u64 old_efer = vcpu->arch.efer;
1129
1130 if (!kvm_valid_efer(vcpu, efer))
1131 return 1;
1132
1133 if (is_paging(vcpu)
1134 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1135 return 1;
1136
1137 efer &= ~EFER_LMA;
1138 efer |= vcpu->arch.efer & EFER_LMA;
1139
1140 kvm_x86_ops->set_efer(vcpu, efer);
1141
1142 /* Update reserved bits */
1143 if ((efer ^ old_efer) & EFER_NX)
1144 kvm_mmu_reset_context(vcpu);
1145
1146 return 0;
1147 }
1148
1149 void kvm_enable_efer_bits(u64 mask)
1150 {
1151 efer_reserved_bits &= ~mask;
1152 }
1153 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1154
1155 /*
1156 * Writes msr value into into the appropriate "register".
1157 * Returns 0 on success, non-0 otherwise.
1158 * Assumes vcpu_load() was already called.
1159 */
1160 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1161 {
1162 switch (msr->index) {
1163 case MSR_FS_BASE:
1164 case MSR_GS_BASE:
1165 case MSR_KERNEL_GS_BASE:
1166 case MSR_CSTAR:
1167 case MSR_LSTAR:
1168 if (is_noncanonical_address(msr->data, vcpu))
1169 return 1;
1170 break;
1171 case MSR_IA32_SYSENTER_EIP:
1172 case MSR_IA32_SYSENTER_ESP:
1173 /*
1174 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1175 * non-canonical address is written on Intel but not on
1176 * AMD (which ignores the top 32-bits, because it does
1177 * not implement 64-bit SYSENTER).
1178 *
1179 * 64-bit code should hence be able to write a non-canonical
1180 * value on AMD. Making the address canonical ensures that
1181 * vmentry does not fail on Intel after writing a non-canonical
1182 * value, and that something deterministic happens if the guest
1183 * invokes 64-bit SYSENTER.
1184 */
1185 msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
1186 }
1187 return kvm_x86_ops->set_msr(vcpu, msr);
1188 }
1189 EXPORT_SYMBOL_GPL(kvm_set_msr);
1190
1191 /*
1192 * Adapt set_msr() to msr_io()'s calling convention
1193 */
1194 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1195 {
1196 struct msr_data msr;
1197 int r;
1198
1199 msr.index = index;
1200 msr.host_initiated = true;
1201 r = kvm_get_msr(vcpu, &msr);
1202 if (r)
1203 return r;
1204
1205 *data = msr.data;
1206 return 0;
1207 }
1208
1209 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1210 {
1211 struct msr_data msr;
1212
1213 msr.data = *data;
1214 msr.index = index;
1215 msr.host_initiated = true;
1216 return kvm_set_msr(vcpu, &msr);
1217 }
1218
1219 #ifdef CONFIG_X86_64
1220 struct pvclock_gtod_data {
1221 seqcount_t seq;
1222
1223 struct { /* extract of a clocksource struct */
1224 int vclock_mode;
1225 u64 cycle_last;
1226 u64 mask;
1227 u32 mult;
1228 u32 shift;
1229 } clock;
1230
1231 u64 boot_ns;
1232 u64 nsec_base;
1233 u64 wall_time_sec;
1234 };
1235
1236 static struct pvclock_gtod_data pvclock_gtod_data;
1237
1238 static void update_pvclock_gtod(struct timekeeper *tk)
1239 {
1240 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1241 u64 boot_ns;
1242
1243 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1244
1245 write_seqcount_begin(&vdata->seq);
1246
1247 /* copy pvclock gtod data */
1248 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1249 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1250 vdata->clock.mask = tk->tkr_mono.mask;
1251 vdata->clock.mult = tk->tkr_mono.mult;
1252 vdata->clock.shift = tk->tkr_mono.shift;
1253
1254 vdata->boot_ns = boot_ns;
1255 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1256
1257 vdata->wall_time_sec = tk->xtime_sec;
1258
1259 write_seqcount_end(&vdata->seq);
1260 }
1261 #endif
1262
1263 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1264 {
1265 /*
1266 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1267 * vcpu_enter_guest. This function is only called from
1268 * the physical CPU that is running vcpu.
1269 */
1270 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1271 }
1272
1273 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1274 {
1275 int version;
1276 int r;
1277 struct pvclock_wall_clock wc;
1278 struct timespec64 boot;
1279
1280 if (!wall_clock)
1281 return;
1282
1283 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1284 if (r)
1285 return;
1286
1287 if (version & 1)
1288 ++version; /* first time write, random junk */
1289
1290 ++version;
1291
1292 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1293 return;
1294
1295 /*
1296 * The guest calculates current wall clock time by adding
1297 * system time (updated by kvm_guest_time_update below) to the
1298 * wall clock specified here. guest system time equals host
1299 * system time for us, thus we must fill in host boot time here.
1300 */
1301 getboottime64(&boot);
1302
1303 if (kvm->arch.kvmclock_offset) {
1304 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1305 boot = timespec64_sub(boot, ts);
1306 }
1307 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1308 wc.nsec = boot.tv_nsec;
1309 wc.version = version;
1310
1311 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1312
1313 version++;
1314 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1315 }
1316
1317 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1318 {
1319 do_shl32_div32(dividend, divisor);
1320 return dividend;
1321 }
1322
1323 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1324 s8 *pshift, u32 *pmultiplier)
1325 {
1326 uint64_t scaled64;
1327 int32_t shift = 0;
1328 uint64_t tps64;
1329 uint32_t tps32;
1330
1331 tps64 = base_hz;
1332 scaled64 = scaled_hz;
1333 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1334 tps64 >>= 1;
1335 shift--;
1336 }
1337
1338 tps32 = (uint32_t)tps64;
1339 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1340 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1341 scaled64 >>= 1;
1342 else
1343 tps32 <<= 1;
1344 shift++;
1345 }
1346
1347 *pshift = shift;
1348 *pmultiplier = div_frac(scaled64, tps32);
1349
1350 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1351 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1352 }
1353
1354 #ifdef CONFIG_X86_64
1355 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1356 #endif
1357
1358 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1359 static unsigned long max_tsc_khz;
1360
1361 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1362 {
1363 u64 v = (u64)khz * (1000000 + ppm);
1364 do_div(v, 1000000);
1365 return v;
1366 }
1367
1368 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1369 {
1370 u64 ratio;
1371
1372 /* Guest TSC same frequency as host TSC? */
1373 if (!scale) {
1374 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1375 return 0;
1376 }
1377
1378 /* TSC scaling supported? */
1379 if (!kvm_has_tsc_control) {
1380 if (user_tsc_khz > tsc_khz) {
1381 vcpu->arch.tsc_catchup = 1;
1382 vcpu->arch.tsc_always_catchup = 1;
1383 return 0;
1384 } else {
1385 WARN(1, "user requested TSC rate below hardware speed\n");
1386 return -1;
1387 }
1388 }
1389
1390 /* TSC scaling required - calculate ratio */
1391 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1392 user_tsc_khz, tsc_khz);
1393
1394 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1395 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1396 user_tsc_khz);
1397 return -1;
1398 }
1399
1400 vcpu->arch.tsc_scaling_ratio = ratio;
1401 return 0;
1402 }
1403
1404 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1405 {
1406 u32 thresh_lo, thresh_hi;
1407 int use_scaling = 0;
1408
1409 /* tsc_khz can be zero if TSC calibration fails */
1410 if (user_tsc_khz == 0) {
1411 /* set tsc_scaling_ratio to a safe value */
1412 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1413 return -1;
1414 }
1415
1416 /* Compute a scale to convert nanoseconds in TSC cycles */
1417 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1418 &vcpu->arch.virtual_tsc_shift,
1419 &vcpu->arch.virtual_tsc_mult);
1420 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1421
1422 /*
1423 * Compute the variation in TSC rate which is acceptable
1424 * within the range of tolerance and decide if the
1425 * rate being applied is within that bounds of the hardware
1426 * rate. If so, no scaling or compensation need be done.
1427 */
1428 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1429 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1430 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1431 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1432 use_scaling = 1;
1433 }
1434 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1435 }
1436
1437 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1438 {
1439 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1440 vcpu->arch.virtual_tsc_mult,
1441 vcpu->arch.virtual_tsc_shift);
1442 tsc += vcpu->arch.this_tsc_write;
1443 return tsc;
1444 }
1445
1446 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1447 {
1448 #ifdef CONFIG_X86_64
1449 bool vcpus_matched;
1450 struct kvm_arch *ka = &vcpu->kvm->arch;
1451 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1452
1453 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1454 atomic_read(&vcpu->kvm->online_vcpus));
1455
1456 /*
1457 * Once the masterclock is enabled, always perform request in
1458 * order to update it.
1459 *
1460 * In order to enable masterclock, the host clocksource must be TSC
1461 * and the vcpus need to have matched TSCs. When that happens,
1462 * perform request to enable masterclock.
1463 */
1464 if (ka->use_master_clock ||
1465 (gtod->clock.vclock_mode == VCLOCK_TSC && vcpus_matched))
1466 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1467
1468 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1469 atomic_read(&vcpu->kvm->online_vcpus),
1470 ka->use_master_clock, gtod->clock.vclock_mode);
1471 #endif
1472 }
1473
1474 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1475 {
1476 u64 curr_offset = vcpu->arch.tsc_offset;
1477 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1478 }
1479
1480 /*
1481 * Multiply tsc by a fixed point number represented by ratio.
1482 *
1483 * The most significant 64-N bits (mult) of ratio represent the
1484 * integral part of the fixed point number; the remaining N bits
1485 * (frac) represent the fractional part, ie. ratio represents a fixed
1486 * point number (mult + frac * 2^(-N)).
1487 *
1488 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1489 */
1490 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1491 {
1492 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1493 }
1494
1495 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1496 {
1497 u64 _tsc = tsc;
1498 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1499
1500 if (ratio != kvm_default_tsc_scaling_ratio)
1501 _tsc = __scale_tsc(ratio, tsc);
1502
1503 return _tsc;
1504 }
1505 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1506
1507 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1508 {
1509 u64 tsc;
1510
1511 tsc = kvm_scale_tsc(vcpu, rdtsc());
1512
1513 return target_tsc - tsc;
1514 }
1515
1516 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1517 {
1518 return vcpu->arch.tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1519 }
1520 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1521
1522 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1523 {
1524 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1525 vcpu->arch.tsc_offset = offset;
1526 }
1527
1528 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1529 {
1530 struct kvm *kvm = vcpu->kvm;
1531 u64 offset, ns, elapsed;
1532 unsigned long flags;
1533 bool matched;
1534 bool already_matched;
1535 u64 data = msr->data;
1536 bool synchronizing = false;
1537
1538 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1539 offset = kvm_compute_tsc_offset(vcpu, data);
1540 ns = ktime_get_boot_ns();
1541 elapsed = ns - kvm->arch.last_tsc_nsec;
1542
1543 if (vcpu->arch.virtual_tsc_khz) {
1544 if (data == 0 && msr->host_initiated) {
1545 /*
1546 * detection of vcpu initialization -- need to sync
1547 * with other vCPUs. This particularly helps to keep
1548 * kvm_clock stable after CPU hotplug
1549 */
1550 synchronizing = true;
1551 } else {
1552 u64 tsc_exp = kvm->arch.last_tsc_write +
1553 nsec_to_cycles(vcpu, elapsed);
1554 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1555 /*
1556 * Special case: TSC write with a small delta (1 second)
1557 * of virtual cycle time against real time is
1558 * interpreted as an attempt to synchronize the CPU.
1559 */
1560 synchronizing = data < tsc_exp + tsc_hz &&
1561 data + tsc_hz > tsc_exp;
1562 }
1563 }
1564
1565 /*
1566 * For a reliable TSC, we can match TSC offsets, and for an unstable
1567 * TSC, we add elapsed time in this computation. We could let the
1568 * compensation code attempt to catch up if we fall behind, but
1569 * it's better to try to match offsets from the beginning.
1570 */
1571 if (synchronizing &&
1572 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1573 if (!check_tsc_unstable()) {
1574 offset = kvm->arch.cur_tsc_offset;
1575 pr_debug("kvm: matched tsc offset for %llu\n", data);
1576 } else {
1577 u64 delta = nsec_to_cycles(vcpu, elapsed);
1578 data += delta;
1579 offset = kvm_compute_tsc_offset(vcpu, data);
1580 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1581 }
1582 matched = true;
1583 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1584 } else {
1585 /*
1586 * We split periods of matched TSC writes into generations.
1587 * For each generation, we track the original measured
1588 * nanosecond time, offset, and write, so if TSCs are in
1589 * sync, we can match exact offset, and if not, we can match
1590 * exact software computation in compute_guest_tsc()
1591 *
1592 * These values are tracked in kvm->arch.cur_xxx variables.
1593 */
1594 kvm->arch.cur_tsc_generation++;
1595 kvm->arch.cur_tsc_nsec = ns;
1596 kvm->arch.cur_tsc_write = data;
1597 kvm->arch.cur_tsc_offset = offset;
1598 matched = false;
1599 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1600 kvm->arch.cur_tsc_generation, data);
1601 }
1602
1603 /*
1604 * We also track th most recent recorded KHZ, write and time to
1605 * allow the matching interval to be extended at each write.
1606 */
1607 kvm->arch.last_tsc_nsec = ns;
1608 kvm->arch.last_tsc_write = data;
1609 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1610
1611 vcpu->arch.last_guest_tsc = data;
1612
1613 /* Keep track of which generation this VCPU has synchronized to */
1614 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1615 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1616 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1617
1618 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1619 update_ia32_tsc_adjust_msr(vcpu, offset);
1620
1621 kvm_vcpu_write_tsc_offset(vcpu, offset);
1622 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1623
1624 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1625 if (!matched) {
1626 kvm->arch.nr_vcpus_matched_tsc = 0;
1627 } else if (!already_matched) {
1628 kvm->arch.nr_vcpus_matched_tsc++;
1629 }
1630
1631 kvm_track_tsc_matching(vcpu);
1632 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1633 }
1634
1635 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1636
1637 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1638 s64 adjustment)
1639 {
1640 kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.tsc_offset + adjustment);
1641 }
1642
1643 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1644 {
1645 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1646 WARN_ON(adjustment < 0);
1647 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1648 adjust_tsc_offset_guest(vcpu, adjustment);
1649 }
1650
1651 #ifdef CONFIG_X86_64
1652
1653 static u64 read_tsc(void)
1654 {
1655 u64 ret = (u64)rdtsc_ordered();
1656 u64 last = pvclock_gtod_data.clock.cycle_last;
1657
1658 if (likely(ret >= last))
1659 return ret;
1660
1661 /*
1662 * GCC likes to generate cmov here, but this branch is extremely
1663 * predictable (it's just a function of time and the likely is
1664 * very likely) and there's a data dependence, so force GCC
1665 * to generate a branch instead. I don't barrier() because
1666 * we don't actually need a barrier, and if this function
1667 * ever gets inlined it will generate worse code.
1668 */
1669 asm volatile ("");
1670 return last;
1671 }
1672
1673 static inline u64 vgettsc(u64 *cycle_now)
1674 {
1675 long v;
1676 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1677
1678 *cycle_now = read_tsc();
1679
1680 v = (*cycle_now - gtod->clock.cycle_last) & gtod->clock.mask;
1681 return v * gtod->clock.mult;
1682 }
1683
1684 static int do_monotonic_boot(s64 *t, u64 *cycle_now)
1685 {
1686 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1687 unsigned long seq;
1688 int mode;
1689 u64 ns;
1690
1691 do {
1692 seq = read_seqcount_begin(&gtod->seq);
1693 mode = gtod->clock.vclock_mode;
1694 ns = gtod->nsec_base;
1695 ns += vgettsc(cycle_now);
1696 ns >>= gtod->clock.shift;
1697 ns += gtod->boot_ns;
1698 } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1699 *t = ns;
1700
1701 return mode;
1702 }
1703
1704 static int do_realtime(struct timespec *ts, u64 *cycle_now)
1705 {
1706 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1707 unsigned long seq;
1708 int mode;
1709 u64 ns;
1710
1711 do {
1712 seq = read_seqcount_begin(&gtod->seq);
1713 mode = gtod->clock.vclock_mode;
1714 ts->tv_sec = gtod->wall_time_sec;
1715 ns = gtod->nsec_base;
1716 ns += vgettsc(cycle_now);
1717 ns >>= gtod->clock.shift;
1718 } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
1719
1720 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1721 ts->tv_nsec = ns;
1722
1723 return mode;
1724 }
1725
1726 /* returns true if host is using tsc clocksource */
1727 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *cycle_now)
1728 {
1729 /* checked again under seqlock below */
1730 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1731 return false;
1732
1733 return do_monotonic_boot(kernel_ns, cycle_now) == VCLOCK_TSC;
1734 }
1735
1736 /* returns true if host is using tsc clocksource */
1737 static bool kvm_get_walltime_and_clockread(struct timespec *ts,
1738 u64 *cycle_now)
1739 {
1740 /* checked again under seqlock below */
1741 if (pvclock_gtod_data.clock.vclock_mode != VCLOCK_TSC)
1742 return false;
1743
1744 return do_realtime(ts, cycle_now) == VCLOCK_TSC;
1745 }
1746 #endif
1747
1748 /*
1749 *
1750 * Assuming a stable TSC across physical CPUS, and a stable TSC
1751 * across virtual CPUs, the following condition is possible.
1752 * Each numbered line represents an event visible to both
1753 * CPUs at the next numbered event.
1754 *
1755 * "timespecX" represents host monotonic time. "tscX" represents
1756 * RDTSC value.
1757 *
1758 * VCPU0 on CPU0 | VCPU1 on CPU1
1759 *
1760 * 1. read timespec0,tsc0
1761 * 2. | timespec1 = timespec0 + N
1762 * | tsc1 = tsc0 + M
1763 * 3. transition to guest | transition to guest
1764 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1765 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1766 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1767 *
1768 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1769 *
1770 * - ret0 < ret1
1771 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1772 * ...
1773 * - 0 < N - M => M < N
1774 *
1775 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1776 * always the case (the difference between two distinct xtime instances
1777 * might be smaller then the difference between corresponding TSC reads,
1778 * when updating guest vcpus pvclock areas).
1779 *
1780 * To avoid that problem, do not allow visibility of distinct
1781 * system_timestamp/tsc_timestamp values simultaneously: use a master
1782 * copy of host monotonic time values. Update that master copy
1783 * in lockstep.
1784 *
1785 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1786 *
1787 */
1788
1789 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1790 {
1791 #ifdef CONFIG_X86_64
1792 struct kvm_arch *ka = &kvm->arch;
1793 int vclock_mode;
1794 bool host_tsc_clocksource, vcpus_matched;
1795
1796 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1797 atomic_read(&kvm->online_vcpus));
1798
1799 /*
1800 * If the host uses TSC clock, then passthrough TSC as stable
1801 * to the guest.
1802 */
1803 host_tsc_clocksource = kvm_get_time_and_clockread(
1804 &ka->master_kernel_ns,
1805 &ka->master_cycle_now);
1806
1807 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1808 && !ka->backwards_tsc_observed
1809 && !ka->boot_vcpu_runs_old_kvmclock;
1810
1811 if (ka->use_master_clock)
1812 atomic_set(&kvm_guest_has_master_clock, 1);
1813
1814 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1815 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
1816 vcpus_matched);
1817 #endif
1818 }
1819
1820 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
1821 {
1822 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
1823 }
1824
1825 static void kvm_gen_update_masterclock(struct kvm *kvm)
1826 {
1827 #ifdef CONFIG_X86_64
1828 int i;
1829 struct kvm_vcpu *vcpu;
1830 struct kvm_arch *ka = &kvm->arch;
1831
1832 spin_lock(&ka->pvclock_gtod_sync_lock);
1833 kvm_make_mclock_inprogress_request(kvm);
1834 /* no guest entries from this point */
1835 pvclock_update_vm_gtod_copy(kvm);
1836
1837 kvm_for_each_vcpu(i, vcpu, kvm)
1838 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1839
1840 /* guest entries allowed */
1841 kvm_for_each_vcpu(i, vcpu, kvm)
1842 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
1843
1844 spin_unlock(&ka->pvclock_gtod_sync_lock);
1845 #endif
1846 }
1847
1848 u64 get_kvmclock_ns(struct kvm *kvm)
1849 {
1850 struct kvm_arch *ka = &kvm->arch;
1851 struct pvclock_vcpu_time_info hv_clock;
1852 u64 ret;
1853
1854 spin_lock(&ka->pvclock_gtod_sync_lock);
1855 if (!ka->use_master_clock) {
1856 spin_unlock(&ka->pvclock_gtod_sync_lock);
1857 return ktime_get_boot_ns() + ka->kvmclock_offset;
1858 }
1859
1860 hv_clock.tsc_timestamp = ka->master_cycle_now;
1861 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
1862 spin_unlock(&ka->pvclock_gtod_sync_lock);
1863
1864 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
1865 get_cpu();
1866
1867 if (__this_cpu_read(cpu_tsc_khz)) {
1868 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
1869 &hv_clock.tsc_shift,
1870 &hv_clock.tsc_to_system_mul);
1871 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
1872 } else
1873 ret = ktime_get_boot_ns() + ka->kvmclock_offset;
1874
1875 put_cpu();
1876
1877 return ret;
1878 }
1879
1880 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
1881 {
1882 struct kvm_vcpu_arch *vcpu = &v->arch;
1883 struct pvclock_vcpu_time_info guest_hv_clock;
1884
1885 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
1886 &guest_hv_clock, sizeof(guest_hv_clock))))
1887 return;
1888
1889 /* This VCPU is paused, but it's legal for a guest to read another
1890 * VCPU's kvmclock, so we really have to follow the specification where
1891 * it says that version is odd if data is being modified, and even after
1892 * it is consistent.
1893 *
1894 * Version field updates must be kept separate. This is because
1895 * kvm_write_guest_cached might use a "rep movs" instruction, and
1896 * writes within a string instruction are weakly ordered. So there
1897 * are three writes overall.
1898 *
1899 * As a small optimization, only write the version field in the first
1900 * and third write. The vcpu->pv_time cache is still valid, because the
1901 * version field is the first in the struct.
1902 */
1903 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
1904
1905 if (guest_hv_clock.version & 1)
1906 ++guest_hv_clock.version; /* first time write, random junk */
1907
1908 vcpu->hv_clock.version = guest_hv_clock.version + 1;
1909 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1910 &vcpu->hv_clock,
1911 sizeof(vcpu->hv_clock.version));
1912
1913 smp_wmb();
1914
1915 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
1916 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
1917
1918 if (vcpu->pvclock_set_guest_stopped_request) {
1919 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
1920 vcpu->pvclock_set_guest_stopped_request = false;
1921 }
1922
1923 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
1924
1925 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1926 &vcpu->hv_clock,
1927 sizeof(vcpu->hv_clock));
1928
1929 smp_wmb();
1930
1931 vcpu->hv_clock.version++;
1932 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
1933 &vcpu->hv_clock,
1934 sizeof(vcpu->hv_clock.version));
1935 }
1936
1937 static int kvm_guest_time_update(struct kvm_vcpu *v)
1938 {
1939 unsigned long flags, tgt_tsc_khz;
1940 struct kvm_vcpu_arch *vcpu = &v->arch;
1941 struct kvm_arch *ka = &v->kvm->arch;
1942 s64 kernel_ns;
1943 u64 tsc_timestamp, host_tsc;
1944 u8 pvclock_flags;
1945 bool use_master_clock;
1946
1947 kernel_ns = 0;
1948 host_tsc = 0;
1949
1950 /*
1951 * If the host uses TSC clock, then passthrough TSC as stable
1952 * to the guest.
1953 */
1954 spin_lock(&ka->pvclock_gtod_sync_lock);
1955 use_master_clock = ka->use_master_clock;
1956 if (use_master_clock) {
1957 host_tsc = ka->master_cycle_now;
1958 kernel_ns = ka->master_kernel_ns;
1959 }
1960 spin_unlock(&ka->pvclock_gtod_sync_lock);
1961
1962 /* Keep irq disabled to prevent changes to the clock */
1963 local_irq_save(flags);
1964 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
1965 if (unlikely(tgt_tsc_khz == 0)) {
1966 local_irq_restore(flags);
1967 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1968 return 1;
1969 }
1970 if (!use_master_clock) {
1971 host_tsc = rdtsc();
1972 kernel_ns = ktime_get_boot_ns();
1973 }
1974
1975 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
1976
1977 /*
1978 * We may have to catch up the TSC to match elapsed wall clock
1979 * time for two reasons, even if kvmclock is used.
1980 * 1) CPU could have been running below the maximum TSC rate
1981 * 2) Broken TSC compensation resets the base at each VCPU
1982 * entry to avoid unknown leaps of TSC even when running
1983 * again on the same CPU. This may cause apparent elapsed
1984 * time to disappear, and the guest to stand still or run
1985 * very slowly.
1986 */
1987 if (vcpu->tsc_catchup) {
1988 u64 tsc = compute_guest_tsc(v, kernel_ns);
1989 if (tsc > tsc_timestamp) {
1990 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
1991 tsc_timestamp = tsc;
1992 }
1993 }
1994
1995 local_irq_restore(flags);
1996
1997 /* With all the info we got, fill in the values */
1998
1999 if (kvm_has_tsc_control)
2000 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2001
2002 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2003 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2004 &vcpu->hv_clock.tsc_shift,
2005 &vcpu->hv_clock.tsc_to_system_mul);
2006 vcpu->hw_tsc_khz = tgt_tsc_khz;
2007 }
2008
2009 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2010 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2011 vcpu->last_guest_tsc = tsc_timestamp;
2012
2013 /* If the host uses TSC clocksource, then it is stable */
2014 pvclock_flags = 0;
2015 if (use_master_clock)
2016 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2017
2018 vcpu->hv_clock.flags = pvclock_flags;
2019
2020 if (vcpu->pv_time_enabled)
2021 kvm_setup_pvclock_page(v);
2022 if (v == kvm_get_vcpu(v->kvm, 0))
2023 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2024 return 0;
2025 }
2026
2027 /*
2028 * kvmclock updates which are isolated to a given vcpu, such as
2029 * vcpu->cpu migration, should not allow system_timestamp from
2030 * the rest of the vcpus to remain static. Otherwise ntp frequency
2031 * correction applies to one vcpu's system_timestamp but not
2032 * the others.
2033 *
2034 * So in those cases, request a kvmclock update for all vcpus.
2035 * We need to rate-limit these requests though, as they can
2036 * considerably slow guests that have a large number of vcpus.
2037 * The time for a remote vcpu to update its kvmclock is bound
2038 * by the delay we use to rate-limit the updates.
2039 */
2040
2041 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2042
2043 static void kvmclock_update_fn(struct work_struct *work)
2044 {
2045 int i;
2046 struct delayed_work *dwork = to_delayed_work(work);
2047 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2048 kvmclock_update_work);
2049 struct kvm *kvm = container_of(ka, struct kvm, arch);
2050 struct kvm_vcpu *vcpu;
2051
2052 kvm_for_each_vcpu(i, vcpu, kvm) {
2053 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2054 kvm_vcpu_kick(vcpu);
2055 }
2056 }
2057
2058 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2059 {
2060 struct kvm *kvm = v->kvm;
2061
2062 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2063 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2064 KVMCLOCK_UPDATE_DELAY);
2065 }
2066
2067 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2068
2069 static void kvmclock_sync_fn(struct work_struct *work)
2070 {
2071 struct delayed_work *dwork = to_delayed_work(work);
2072 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2073 kvmclock_sync_work);
2074 struct kvm *kvm = container_of(ka, struct kvm, arch);
2075
2076 if (!kvmclock_periodic_sync)
2077 return;
2078
2079 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2080 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2081 KVMCLOCK_SYNC_PERIOD);
2082 }
2083
2084 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2085 {
2086 u64 mcg_cap = vcpu->arch.mcg_cap;
2087 unsigned bank_num = mcg_cap & 0xff;
2088 u32 msr = msr_info->index;
2089 u64 data = msr_info->data;
2090
2091 switch (msr) {
2092 case MSR_IA32_MCG_STATUS:
2093 vcpu->arch.mcg_status = data;
2094 break;
2095 case MSR_IA32_MCG_CTL:
2096 if (!(mcg_cap & MCG_CTL_P))
2097 return 1;
2098 if (data != 0 && data != ~(u64)0)
2099 return -1;
2100 vcpu->arch.mcg_ctl = data;
2101 break;
2102 default:
2103 if (msr >= MSR_IA32_MC0_CTL &&
2104 msr < MSR_IA32_MCx_CTL(bank_num)) {
2105 u32 offset = msr - MSR_IA32_MC0_CTL;
2106 /* only 0 or all 1s can be written to IA32_MCi_CTL
2107 * some Linux kernels though clear bit 10 in bank 4 to
2108 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2109 * this to avoid an uncatched #GP in the guest
2110 */
2111 if ((offset & 0x3) == 0 &&
2112 data != 0 && (data | (1 << 10)) != ~(u64)0)
2113 return -1;
2114 if (!msr_info->host_initiated &&
2115 (offset & 0x3) == 1 && data != 0)
2116 return -1;
2117 vcpu->arch.mce_banks[offset] = data;
2118 break;
2119 }
2120 return 1;
2121 }
2122 return 0;
2123 }
2124
2125 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2126 {
2127 struct kvm *kvm = vcpu->kvm;
2128 int lm = is_long_mode(vcpu);
2129 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2130 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2131 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2132 : kvm->arch.xen_hvm_config.blob_size_32;
2133 u32 page_num = data & ~PAGE_MASK;
2134 u64 page_addr = data & PAGE_MASK;
2135 u8 *page;
2136 int r;
2137
2138 r = -E2BIG;
2139 if (page_num >= blob_size)
2140 goto out;
2141 r = -ENOMEM;
2142 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2143 if (IS_ERR(page)) {
2144 r = PTR_ERR(page);
2145 goto out;
2146 }
2147 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2148 goto out_free;
2149 r = 0;
2150 out_free:
2151 kfree(page);
2152 out:
2153 return r;
2154 }
2155
2156 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2157 {
2158 gpa_t gpa = data & ~0x3f;
2159
2160 /* Bits 3:5 are reserved, Should be zero */
2161 if (data & 0x38)
2162 return 1;
2163
2164 vcpu->arch.apf.msr_val = data;
2165
2166 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2167 kvm_clear_async_pf_completion_queue(vcpu);
2168 kvm_async_pf_hash_reset(vcpu);
2169 return 0;
2170 }
2171
2172 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2173 sizeof(u32)))
2174 return 1;
2175
2176 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2177 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2178 kvm_async_pf_wakeup_all(vcpu);
2179 return 0;
2180 }
2181
2182 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2183 {
2184 vcpu->arch.pv_time_enabled = false;
2185 }
2186
2187 static void record_steal_time(struct kvm_vcpu *vcpu)
2188 {
2189 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2190 return;
2191
2192 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2193 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2194 return;
2195
2196 vcpu->arch.st.steal.preempted = 0;
2197
2198 if (vcpu->arch.st.steal.version & 1)
2199 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2200
2201 vcpu->arch.st.steal.version += 1;
2202
2203 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2204 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2205
2206 smp_wmb();
2207
2208 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2209 vcpu->arch.st.last_steal;
2210 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2211
2212 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2213 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2214
2215 smp_wmb();
2216
2217 vcpu->arch.st.steal.version += 1;
2218
2219 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2220 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2221 }
2222
2223 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2224 {
2225 bool pr = false;
2226 u32 msr = msr_info->index;
2227 u64 data = msr_info->data;
2228
2229 switch (msr) {
2230 case MSR_AMD64_NB_CFG:
2231 case MSR_IA32_UCODE_REV:
2232 case MSR_IA32_UCODE_WRITE:
2233 case MSR_VM_HSAVE_PA:
2234 case MSR_AMD64_PATCH_LOADER:
2235 case MSR_AMD64_BU_CFG2:
2236 case MSR_AMD64_DC_CFG:
2237 break;
2238
2239 case MSR_EFER:
2240 return set_efer(vcpu, data);
2241 case MSR_K7_HWCR:
2242 data &= ~(u64)0x40; /* ignore flush filter disable */
2243 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2244 data &= ~(u64)0x8; /* ignore TLB cache disable */
2245 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2246 if (data != 0) {
2247 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2248 data);
2249 return 1;
2250 }
2251 break;
2252 case MSR_FAM10H_MMIO_CONF_BASE:
2253 if (data != 0) {
2254 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2255 "0x%llx\n", data);
2256 return 1;
2257 }
2258 break;
2259 case MSR_IA32_DEBUGCTLMSR:
2260 if (!data) {
2261 /* We support the non-activated case already */
2262 break;
2263 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2264 /* Values other than LBR and BTF are vendor-specific,
2265 thus reserved and should throw a #GP */
2266 return 1;
2267 }
2268 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2269 __func__, data);
2270 break;
2271 case 0x200 ... 0x2ff:
2272 return kvm_mtrr_set_msr(vcpu, msr, data);
2273 case MSR_IA32_APICBASE:
2274 return kvm_set_apic_base(vcpu, msr_info);
2275 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2276 return kvm_x2apic_msr_write(vcpu, msr, data);
2277 case MSR_IA32_TSCDEADLINE:
2278 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2279 break;
2280 case MSR_IA32_TSC_ADJUST:
2281 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2282 if (!msr_info->host_initiated) {
2283 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2284 adjust_tsc_offset_guest(vcpu, adj);
2285 }
2286 vcpu->arch.ia32_tsc_adjust_msr = data;
2287 }
2288 break;
2289 case MSR_IA32_MISC_ENABLE:
2290 vcpu->arch.ia32_misc_enable_msr = data;
2291 break;
2292 case MSR_IA32_SMBASE:
2293 if (!msr_info->host_initiated)
2294 return 1;
2295 vcpu->arch.smbase = data;
2296 break;
2297 case MSR_KVM_WALL_CLOCK_NEW:
2298 case MSR_KVM_WALL_CLOCK:
2299 vcpu->kvm->arch.wall_clock = data;
2300 kvm_write_wall_clock(vcpu->kvm, data);
2301 break;
2302 case MSR_KVM_SYSTEM_TIME_NEW:
2303 case MSR_KVM_SYSTEM_TIME: {
2304 struct kvm_arch *ka = &vcpu->kvm->arch;
2305
2306 kvmclock_reset(vcpu);
2307
2308 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2309 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2310
2311 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2312 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2313
2314 ka->boot_vcpu_runs_old_kvmclock = tmp;
2315 }
2316
2317 vcpu->arch.time = data;
2318 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2319
2320 /* we verify if the enable bit is set... */
2321 if (!(data & 1))
2322 break;
2323
2324 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2325 &vcpu->arch.pv_time, data & ~1ULL,
2326 sizeof(struct pvclock_vcpu_time_info)))
2327 vcpu->arch.pv_time_enabled = false;
2328 else
2329 vcpu->arch.pv_time_enabled = true;
2330
2331 break;
2332 }
2333 case MSR_KVM_ASYNC_PF_EN:
2334 if (kvm_pv_enable_async_pf(vcpu, data))
2335 return 1;
2336 break;
2337 case MSR_KVM_STEAL_TIME:
2338
2339 if (unlikely(!sched_info_on()))
2340 return 1;
2341
2342 if (data & KVM_STEAL_RESERVED_MASK)
2343 return 1;
2344
2345 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2346 data & KVM_STEAL_VALID_BITS,
2347 sizeof(struct kvm_steal_time)))
2348 return 1;
2349
2350 vcpu->arch.st.msr_val = data;
2351
2352 if (!(data & KVM_MSR_ENABLED))
2353 break;
2354
2355 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2356
2357 break;
2358 case MSR_KVM_PV_EOI_EN:
2359 if (kvm_lapic_enable_pv_eoi(vcpu, data))
2360 return 1;
2361 break;
2362
2363 case MSR_IA32_MCG_CTL:
2364 case MSR_IA32_MCG_STATUS:
2365 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2366 return set_msr_mce(vcpu, msr_info);
2367
2368 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2369 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2370 pr = true; /* fall through */
2371 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2372 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2373 if (kvm_pmu_is_valid_msr(vcpu, msr))
2374 return kvm_pmu_set_msr(vcpu, msr_info);
2375
2376 if (pr || data != 0)
2377 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2378 "0x%x data 0x%llx\n", msr, data);
2379 break;
2380 case MSR_K7_CLK_CTL:
2381 /*
2382 * Ignore all writes to this no longer documented MSR.
2383 * Writes are only relevant for old K7 processors,
2384 * all pre-dating SVM, but a recommended workaround from
2385 * AMD for these chips. It is possible to specify the
2386 * affected processor models on the command line, hence
2387 * the need to ignore the workaround.
2388 */
2389 break;
2390 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2391 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2392 case HV_X64_MSR_CRASH_CTL:
2393 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2394 return kvm_hv_set_msr_common(vcpu, msr, data,
2395 msr_info->host_initiated);
2396 case MSR_IA32_BBL_CR_CTL3:
2397 /* Drop writes to this legacy MSR -- see rdmsr
2398 * counterpart for further detail.
2399 */
2400 if (report_ignored_msrs)
2401 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2402 msr, data);
2403 break;
2404 case MSR_AMD64_OSVW_ID_LENGTH:
2405 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2406 return 1;
2407 vcpu->arch.osvw.length = data;
2408 break;
2409 case MSR_AMD64_OSVW_STATUS:
2410 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2411 return 1;
2412 vcpu->arch.osvw.status = data;
2413 break;
2414 case MSR_PLATFORM_INFO:
2415 if (!msr_info->host_initiated ||
2416 data & ~MSR_PLATFORM_INFO_CPUID_FAULT ||
2417 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2418 cpuid_fault_enabled(vcpu)))
2419 return 1;
2420 vcpu->arch.msr_platform_info = data;
2421 break;
2422 case MSR_MISC_FEATURES_ENABLES:
2423 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2424 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2425 !supports_cpuid_fault(vcpu)))
2426 return 1;
2427 vcpu->arch.msr_misc_features_enables = data;
2428 break;
2429 default:
2430 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2431 return xen_hvm_config(vcpu, data);
2432 if (kvm_pmu_is_valid_msr(vcpu, msr))
2433 return kvm_pmu_set_msr(vcpu, msr_info);
2434 if (!ignore_msrs) {
2435 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2436 msr, data);
2437 return 1;
2438 } else {
2439 if (report_ignored_msrs)
2440 vcpu_unimpl(vcpu,
2441 "ignored wrmsr: 0x%x data 0x%llx\n",
2442 msr, data);
2443 break;
2444 }
2445 }
2446 return 0;
2447 }
2448 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2449
2450
2451 /*
2452 * Reads an msr value (of 'msr_index') into 'pdata'.
2453 * Returns 0 on success, non-0 otherwise.
2454 * Assumes vcpu_load() was already called.
2455 */
2456 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2457 {
2458 return kvm_x86_ops->get_msr(vcpu, msr);
2459 }
2460 EXPORT_SYMBOL_GPL(kvm_get_msr);
2461
2462 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
2463 {
2464 u64 data;
2465 u64 mcg_cap = vcpu->arch.mcg_cap;
2466 unsigned bank_num = mcg_cap & 0xff;
2467
2468 switch (msr) {
2469 case MSR_IA32_P5_MC_ADDR:
2470 case MSR_IA32_P5_MC_TYPE:
2471 data = 0;
2472 break;
2473 case MSR_IA32_MCG_CAP:
2474 data = vcpu->arch.mcg_cap;
2475 break;
2476 case MSR_IA32_MCG_CTL:
2477 if (!(mcg_cap & MCG_CTL_P))
2478 return 1;
2479 data = vcpu->arch.mcg_ctl;
2480 break;
2481 case MSR_IA32_MCG_STATUS:
2482 data = vcpu->arch.mcg_status;
2483 break;
2484 default:
2485 if (msr >= MSR_IA32_MC0_CTL &&
2486 msr < MSR_IA32_MCx_CTL(bank_num)) {
2487 u32 offset = msr - MSR_IA32_MC0_CTL;
2488 data = vcpu->arch.mce_banks[offset];
2489 break;
2490 }
2491 return 1;
2492 }
2493 *pdata = data;
2494 return 0;
2495 }
2496
2497 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2498 {
2499 switch (msr_info->index) {
2500 case MSR_IA32_PLATFORM_ID:
2501 case MSR_IA32_EBL_CR_POWERON:
2502 case MSR_IA32_DEBUGCTLMSR:
2503 case MSR_IA32_LASTBRANCHFROMIP:
2504 case MSR_IA32_LASTBRANCHTOIP:
2505 case MSR_IA32_LASTINTFROMIP:
2506 case MSR_IA32_LASTINTTOIP:
2507 case MSR_K8_SYSCFG:
2508 case MSR_K8_TSEG_ADDR:
2509 case MSR_K8_TSEG_MASK:
2510 case MSR_K7_HWCR:
2511 case MSR_VM_HSAVE_PA:
2512 case MSR_K8_INT_PENDING_MSG:
2513 case MSR_AMD64_NB_CFG:
2514 case MSR_FAM10H_MMIO_CONF_BASE:
2515 case MSR_AMD64_BU_CFG2:
2516 case MSR_IA32_PERF_CTL:
2517 case MSR_AMD64_DC_CFG:
2518 msr_info->data = 0;
2519 break;
2520 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2521 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2522 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2523 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2524 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2525 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2526 msr_info->data = 0;
2527 break;
2528 case MSR_IA32_UCODE_REV:
2529 msr_info->data = 0x100000000ULL;
2530 break;
2531 case MSR_MTRRcap:
2532 case 0x200 ... 0x2ff:
2533 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2534 case 0xcd: /* fsb frequency */
2535 msr_info->data = 3;
2536 break;
2537 /*
2538 * MSR_EBC_FREQUENCY_ID
2539 * Conservative value valid for even the basic CPU models.
2540 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2541 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2542 * and 266MHz for model 3, or 4. Set Core Clock
2543 * Frequency to System Bus Frequency Ratio to 1 (bits
2544 * 31:24) even though these are only valid for CPU
2545 * models > 2, however guests may end up dividing or
2546 * multiplying by zero otherwise.
2547 */
2548 case MSR_EBC_FREQUENCY_ID:
2549 msr_info->data = 1 << 24;
2550 break;
2551 case MSR_IA32_APICBASE:
2552 msr_info->data = kvm_get_apic_base(vcpu);
2553 break;
2554 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2555 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2556 break;
2557 case MSR_IA32_TSCDEADLINE:
2558 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2559 break;
2560 case MSR_IA32_TSC_ADJUST:
2561 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2562 break;
2563 case MSR_IA32_MISC_ENABLE:
2564 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2565 break;
2566 case MSR_IA32_SMBASE:
2567 if (!msr_info->host_initiated)
2568 return 1;
2569 msr_info->data = vcpu->arch.smbase;
2570 break;
2571 case MSR_IA32_PERF_STATUS:
2572 /* TSC increment by tick */
2573 msr_info->data = 1000ULL;
2574 /* CPU multiplier */
2575 msr_info->data |= (((uint64_t)4ULL) << 40);
2576 break;
2577 case MSR_EFER:
2578 msr_info->data = vcpu->arch.efer;
2579 break;
2580 case MSR_KVM_WALL_CLOCK:
2581 case MSR_KVM_WALL_CLOCK_NEW:
2582 msr_info->data = vcpu->kvm->arch.wall_clock;
2583 break;
2584 case MSR_KVM_SYSTEM_TIME:
2585 case MSR_KVM_SYSTEM_TIME_NEW:
2586 msr_info->data = vcpu->arch.time;
2587 break;
2588 case MSR_KVM_ASYNC_PF_EN:
2589 msr_info->data = vcpu->arch.apf.msr_val;
2590 break;
2591 case MSR_KVM_STEAL_TIME:
2592 msr_info->data = vcpu->arch.st.msr_val;
2593 break;
2594 case MSR_KVM_PV_EOI_EN:
2595 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2596 break;
2597 case MSR_IA32_P5_MC_ADDR:
2598 case MSR_IA32_P5_MC_TYPE:
2599 case MSR_IA32_MCG_CAP:
2600 case MSR_IA32_MCG_CTL:
2601 case MSR_IA32_MCG_STATUS:
2602 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2603 return get_msr_mce(vcpu, msr_info->index, &msr_info->data);
2604 case MSR_K7_CLK_CTL:
2605 /*
2606 * Provide expected ramp-up count for K7. All other
2607 * are set to zero, indicating minimum divisors for
2608 * every field.
2609 *
2610 * This prevents guest kernels on AMD host with CPU
2611 * type 6, model 8 and higher from exploding due to
2612 * the rdmsr failing.
2613 */
2614 msr_info->data = 0x20000000;
2615 break;
2616 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2617 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2618 case HV_X64_MSR_CRASH_CTL:
2619 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2620 return kvm_hv_get_msr_common(vcpu,
2621 msr_info->index, &msr_info->data);
2622 break;
2623 case MSR_IA32_BBL_CR_CTL3:
2624 /* This legacy MSR exists but isn't fully documented in current
2625 * silicon. It is however accessed by winxp in very narrow
2626 * scenarios where it sets bit #19, itself documented as
2627 * a "reserved" bit. Best effort attempt to source coherent
2628 * read data here should the balance of the register be
2629 * interpreted by the guest:
2630 *
2631 * L2 cache control register 3: 64GB range, 256KB size,
2632 * enabled, latency 0x1, configured
2633 */
2634 msr_info->data = 0xbe702111;
2635 break;
2636 case MSR_AMD64_OSVW_ID_LENGTH:
2637 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2638 return 1;
2639 msr_info->data = vcpu->arch.osvw.length;
2640 break;
2641 case MSR_AMD64_OSVW_STATUS:
2642 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2643 return 1;
2644 msr_info->data = vcpu->arch.osvw.status;
2645 break;
2646 case MSR_PLATFORM_INFO:
2647 msr_info->data = vcpu->arch.msr_platform_info;
2648 break;
2649 case MSR_MISC_FEATURES_ENABLES:
2650 msr_info->data = vcpu->arch.msr_misc_features_enables;
2651 break;
2652 default:
2653 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2654 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2655 if (!ignore_msrs) {
2656 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2657 msr_info->index);
2658 return 1;
2659 } else {
2660 if (report_ignored_msrs)
2661 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
2662 msr_info->index);
2663 msr_info->data = 0;
2664 }
2665 break;
2666 }
2667 return 0;
2668 }
2669 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2670
2671 /*
2672 * Read or write a bunch of msrs. All parameters are kernel addresses.
2673 *
2674 * @return number of msrs set successfully.
2675 */
2676 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2677 struct kvm_msr_entry *entries,
2678 int (*do_msr)(struct kvm_vcpu *vcpu,
2679 unsigned index, u64 *data))
2680 {
2681 int i;
2682
2683 for (i = 0; i < msrs->nmsrs; ++i)
2684 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2685 break;
2686
2687 return i;
2688 }
2689
2690 /*
2691 * Read or write a bunch of msrs. Parameters are user addresses.
2692 *
2693 * @return number of msrs set successfully.
2694 */
2695 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2696 int (*do_msr)(struct kvm_vcpu *vcpu,
2697 unsigned index, u64 *data),
2698 int writeback)
2699 {
2700 struct kvm_msrs msrs;
2701 struct kvm_msr_entry *entries;
2702 int r, n;
2703 unsigned size;
2704
2705 r = -EFAULT;
2706 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
2707 goto out;
2708
2709 r = -E2BIG;
2710 if (msrs.nmsrs >= MAX_IO_MSRS)
2711 goto out;
2712
2713 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2714 entries = memdup_user(user_msrs->entries, size);
2715 if (IS_ERR(entries)) {
2716 r = PTR_ERR(entries);
2717 goto out;
2718 }
2719
2720 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2721 if (r < 0)
2722 goto out_free;
2723
2724 r = -EFAULT;
2725 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2726 goto out_free;
2727
2728 r = n;
2729
2730 out_free:
2731 kfree(entries);
2732 out:
2733 return r;
2734 }
2735
2736 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2737 {
2738 int r;
2739
2740 switch (ext) {
2741 case KVM_CAP_IRQCHIP:
2742 case KVM_CAP_HLT:
2743 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2744 case KVM_CAP_SET_TSS_ADDR:
2745 case KVM_CAP_EXT_CPUID:
2746 case KVM_CAP_EXT_EMUL_CPUID:
2747 case KVM_CAP_CLOCKSOURCE:
2748 case KVM_CAP_PIT:
2749 case KVM_CAP_NOP_IO_DELAY:
2750 case KVM_CAP_MP_STATE:
2751 case KVM_CAP_SYNC_MMU:
2752 case KVM_CAP_USER_NMI:
2753 case KVM_CAP_REINJECT_CONTROL:
2754 case KVM_CAP_IRQ_INJECT_STATUS:
2755 case KVM_CAP_IOEVENTFD:
2756 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2757 case KVM_CAP_PIT2:
2758 case KVM_CAP_PIT_STATE2:
2759 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2760 case KVM_CAP_XEN_HVM:
2761 case KVM_CAP_VCPU_EVENTS:
2762 case KVM_CAP_HYPERV:
2763 case KVM_CAP_HYPERV_VAPIC:
2764 case KVM_CAP_HYPERV_SPIN:
2765 case KVM_CAP_HYPERV_SYNIC:
2766 case KVM_CAP_HYPERV_SYNIC2:
2767 case KVM_CAP_HYPERV_VP_INDEX:
2768 case KVM_CAP_PCI_SEGMENT:
2769 case KVM_CAP_DEBUGREGS:
2770 case KVM_CAP_X86_ROBUST_SINGLESTEP:
2771 case KVM_CAP_XSAVE:
2772 case KVM_CAP_ASYNC_PF:
2773 case KVM_CAP_GET_TSC_KHZ:
2774 case KVM_CAP_KVMCLOCK_CTRL:
2775 case KVM_CAP_READONLY_MEM:
2776 case KVM_CAP_HYPERV_TIME:
2777 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
2778 case KVM_CAP_TSC_DEADLINE_TIMER:
2779 case KVM_CAP_ENABLE_CAP_VM:
2780 case KVM_CAP_DISABLE_QUIRKS:
2781 case KVM_CAP_SET_BOOT_CPU_ID:
2782 case KVM_CAP_SPLIT_IRQCHIP:
2783 case KVM_CAP_IMMEDIATE_EXIT:
2784 case KVM_CAP_GET_MSR_FEATURES:
2785 r = 1;
2786 break;
2787 case KVM_CAP_ADJUST_CLOCK:
2788 r = KVM_CLOCK_TSC_STABLE;
2789 break;
2790 case KVM_CAP_X86_GUEST_MWAIT:
2791 r = kvm_mwait_in_guest();
2792 break;
2793 case KVM_CAP_X86_SMM:
2794 /* SMBASE is usually relocated above 1M on modern chipsets,
2795 * and SMM handlers might indeed rely on 4G segment limits,
2796 * so do not report SMM to be available if real mode is
2797 * emulated via vm86 mode. Still, do not go to great lengths
2798 * to avoid userspace's usage of the feature, because it is a
2799 * fringe case that is not enabled except via specific settings
2800 * of the module parameters.
2801 */
2802 r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
2803 break;
2804 case KVM_CAP_VAPIC:
2805 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2806 break;
2807 case KVM_CAP_NR_VCPUS:
2808 r = KVM_SOFT_MAX_VCPUS;
2809 break;
2810 case KVM_CAP_MAX_VCPUS:
2811 r = KVM_MAX_VCPUS;
2812 break;
2813 case KVM_CAP_NR_MEMSLOTS:
2814 r = KVM_USER_MEM_SLOTS;
2815 break;
2816 case KVM_CAP_PV_MMU: /* obsolete */
2817 r = 0;
2818 break;
2819 case KVM_CAP_MCE:
2820 r = KVM_MAX_MCE_BANKS;
2821 break;
2822 case KVM_CAP_XCRS:
2823 r = boot_cpu_has(X86_FEATURE_XSAVE);
2824 break;
2825 case KVM_CAP_TSC_CONTROL:
2826 r = kvm_has_tsc_control;
2827 break;
2828 case KVM_CAP_X2APIC_API:
2829 r = KVM_X2APIC_API_VALID_FLAGS;
2830 break;
2831 default:
2832 r = 0;
2833 break;
2834 }
2835 return r;
2836
2837 }
2838
2839 long kvm_arch_dev_ioctl(struct file *filp,
2840 unsigned int ioctl, unsigned long arg)
2841 {
2842 void __user *argp = (void __user *)arg;
2843 long r;
2844
2845 switch (ioctl) {
2846 case KVM_GET_MSR_INDEX_LIST: {
2847 struct kvm_msr_list __user *user_msr_list = argp;
2848 struct kvm_msr_list msr_list;
2849 unsigned n;
2850
2851 r = -EFAULT;
2852 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2853 goto out;
2854 n = msr_list.nmsrs;
2855 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
2856 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2857 goto out;
2858 r = -E2BIG;
2859 if (n < msr_list.nmsrs)
2860 goto out;
2861 r = -EFAULT;
2862 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2863 num_msrs_to_save * sizeof(u32)))
2864 goto out;
2865 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2866 &emulated_msrs,
2867 num_emulated_msrs * sizeof(u32)))
2868 goto out;
2869 r = 0;
2870 break;
2871 }
2872 case KVM_GET_SUPPORTED_CPUID:
2873 case KVM_GET_EMULATED_CPUID: {
2874 struct kvm_cpuid2 __user *cpuid_arg = argp;
2875 struct kvm_cpuid2 cpuid;
2876
2877 r = -EFAULT;
2878 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2879 goto out;
2880
2881 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
2882 ioctl);
2883 if (r)
2884 goto out;
2885
2886 r = -EFAULT;
2887 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2888 goto out;
2889 r = 0;
2890 break;
2891 }
2892 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2893 r = -EFAULT;
2894 if (copy_to_user(argp, &kvm_mce_cap_supported,
2895 sizeof(kvm_mce_cap_supported)))
2896 goto out;
2897 r = 0;
2898 break;
2899 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
2900 struct kvm_msr_list __user *user_msr_list = argp;
2901 struct kvm_msr_list msr_list;
2902 unsigned int n;
2903
2904 r = -EFAULT;
2905 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
2906 goto out;
2907 n = msr_list.nmsrs;
2908 msr_list.nmsrs = num_msr_based_features;
2909 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
2910 goto out;
2911 r = -E2BIG;
2912 if (n < msr_list.nmsrs)
2913 goto out;
2914 r = -EFAULT;
2915 if (copy_to_user(user_msr_list->indices, &msr_based_features,
2916 num_msr_based_features * sizeof(u32)))
2917 goto out;
2918 r = 0;
2919 break;
2920 }
2921 case KVM_GET_MSRS:
2922 r = msr_io(NULL, argp, do_get_msr_feature, 1);
2923 break;
2924 }
2925 default:
2926 r = -EINVAL;
2927 }
2928 out:
2929 return r;
2930 }
2931
2932 static void wbinvd_ipi(void *garbage)
2933 {
2934 wbinvd();
2935 }
2936
2937 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2938 {
2939 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
2940 }
2941
2942 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2943 {
2944 /* Address WBINVD may be executed by guest */
2945 if (need_emulate_wbinvd(vcpu)) {
2946 if (kvm_x86_ops->has_wbinvd_exit())
2947 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2948 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2949 smp_call_function_single(vcpu->cpu,
2950 wbinvd_ipi, NULL, 1);
2951 }
2952
2953 kvm_x86_ops->vcpu_load(vcpu, cpu);
2954
2955 /* Apply any externally detected TSC adjustments (due to suspend) */
2956 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
2957 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
2958 vcpu->arch.tsc_offset_adjustment = 0;
2959 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2960 }
2961
2962 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2963 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2964 rdtsc() - vcpu->arch.last_host_tsc;
2965 if (tsc_delta < 0)
2966 mark_tsc_unstable("KVM discovered backwards TSC");
2967
2968 if (check_tsc_unstable()) {
2969 u64 offset = kvm_compute_tsc_offset(vcpu,
2970 vcpu->arch.last_guest_tsc);
2971 kvm_vcpu_write_tsc_offset(vcpu, offset);
2972 vcpu->arch.tsc_catchup = 1;
2973 }
2974
2975 if (kvm_lapic_hv_timer_in_use(vcpu))
2976 kvm_lapic_restart_hv_timer(vcpu);
2977
2978 /*
2979 * On a host with synchronized TSC, there is no need to update
2980 * kvmclock on vcpu->cpu migration
2981 */
2982 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
2983 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2984 if (vcpu->cpu != cpu)
2985 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
2986 vcpu->cpu = cpu;
2987 }
2988
2989 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2990 }
2991
2992 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
2993 {
2994 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2995 return;
2996
2997 vcpu->arch.st.steal.preempted = 1;
2998
2999 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3000 &vcpu->arch.st.steal.preempted,
3001 offsetof(struct kvm_steal_time, preempted),
3002 sizeof(vcpu->arch.st.steal.preempted));
3003 }
3004
3005 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3006 {
3007 int idx;
3008
3009 if (vcpu->preempted)
3010 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3011
3012 /*
3013 * Disable page faults because we're in atomic context here.
3014 * kvm_write_guest_offset_cached() would call might_fault()
3015 * that relies on pagefault_disable() to tell if there's a
3016 * bug. NOTE: the write to guest memory may not go through if
3017 * during postcopy live migration or if there's heavy guest
3018 * paging.
3019 */
3020 pagefault_disable();
3021 /*
3022 * kvm_memslots() will be called by
3023 * kvm_write_guest_offset_cached() so take the srcu lock.
3024 */
3025 idx = srcu_read_lock(&vcpu->kvm->srcu);
3026 kvm_steal_time_set_preempted(vcpu);
3027 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3028 pagefault_enable();
3029 kvm_x86_ops->vcpu_put(vcpu);
3030 vcpu->arch.last_host_tsc = rdtsc();
3031 /*
3032 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3033 * on every vmexit, but if not, we might have a stale dr6 from the
3034 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3035 */
3036 set_debugreg(0, 6);
3037 }
3038
3039 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3040 struct kvm_lapic_state *s)
3041 {
3042 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
3043 kvm_x86_ops->sync_pir_to_irr(vcpu);
3044
3045 return kvm_apic_get_state(vcpu, s);
3046 }
3047
3048 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3049 struct kvm_lapic_state *s)
3050 {
3051 int r;
3052
3053 r = kvm_apic_set_state(vcpu, s);
3054 if (r)
3055 return r;
3056 update_cr8_intercept(vcpu);
3057
3058 return 0;
3059 }
3060
3061 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3062 {
3063 return (!lapic_in_kernel(vcpu) ||
3064 kvm_apic_accept_pic_intr(vcpu));
3065 }
3066
3067 /*
3068 * if userspace requested an interrupt window, check that the
3069 * interrupt window is open.
3070 *
3071 * No need to exit to userspace if we already have an interrupt queued.
3072 */
3073 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3074 {
3075 return kvm_arch_interrupt_allowed(vcpu) &&
3076 !kvm_cpu_has_interrupt(vcpu) &&
3077 !kvm_event_needs_reinjection(vcpu) &&
3078 kvm_cpu_accept_dm_intr(vcpu);
3079 }
3080
3081 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3082 struct kvm_interrupt *irq)
3083 {
3084 if (irq->irq >= KVM_NR_INTERRUPTS)
3085 return -EINVAL;
3086
3087 if (!irqchip_in_kernel(vcpu->kvm)) {
3088 kvm_queue_interrupt(vcpu, irq->irq, false);
3089 kvm_make_request(KVM_REQ_EVENT, vcpu);
3090 return 0;
3091 }
3092
3093 /*
3094 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3095 * fail for in-kernel 8259.
3096 */
3097 if (pic_in_kernel(vcpu->kvm))
3098 return -ENXIO;
3099
3100 if (vcpu->arch.pending_external_vector != -1)
3101 return -EEXIST;
3102
3103 vcpu->arch.pending_external_vector = irq->irq;
3104 kvm_make_request(KVM_REQ_EVENT, vcpu);
3105 return 0;
3106 }
3107
3108 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3109 {
3110 kvm_inject_nmi(vcpu);
3111
3112 return 0;
3113 }
3114
3115 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3116 {
3117 kvm_make_request(KVM_REQ_SMI, vcpu);
3118
3119 return 0;
3120 }
3121
3122 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3123 struct kvm_tpr_access_ctl *tac)
3124 {
3125 if (tac->flags)
3126 return -EINVAL;
3127 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3128 return 0;
3129 }
3130
3131 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3132 u64 mcg_cap)
3133 {
3134 int r;
3135 unsigned bank_num = mcg_cap & 0xff, bank;
3136
3137 r = -EINVAL;
3138 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3139 goto out;
3140 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3141 goto out;
3142 r = 0;
3143 vcpu->arch.mcg_cap = mcg_cap;
3144 /* Init IA32_MCG_CTL to all 1s */
3145 if (mcg_cap & MCG_CTL_P)
3146 vcpu->arch.mcg_ctl = ~(u64)0;
3147 /* Init IA32_MCi_CTL to all 1s */
3148 for (bank = 0; bank < bank_num; bank++)
3149 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3150
3151 if (kvm_x86_ops->setup_mce)
3152 kvm_x86_ops->setup_mce(vcpu);
3153 out:
3154 return r;
3155 }
3156
3157 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3158 struct kvm_x86_mce *mce)
3159 {
3160 u64 mcg_cap = vcpu->arch.mcg_cap;
3161 unsigned bank_num = mcg_cap & 0xff;
3162 u64 *banks = vcpu->arch.mce_banks;
3163
3164 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3165 return -EINVAL;
3166 /*
3167 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3168 * reporting is disabled
3169 */
3170 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3171 vcpu->arch.mcg_ctl != ~(u64)0)
3172 return 0;
3173 banks += 4 * mce->bank;
3174 /*
3175 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3176 * reporting is disabled for the bank
3177 */
3178 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3179 return 0;
3180 if (mce->status & MCI_STATUS_UC) {
3181 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3182 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3183 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3184 return 0;
3185 }
3186 if (banks[1] & MCI_STATUS_VAL)
3187 mce->status |= MCI_STATUS_OVER;
3188 banks[2] = mce->addr;
3189 banks[3] = mce->misc;
3190 vcpu->arch.mcg_status = mce->mcg_status;
3191 banks[1] = mce->status;
3192 kvm_queue_exception(vcpu, MC_VECTOR);
3193 } else if (!(banks[1] & MCI_STATUS_VAL)
3194 || !(banks[1] & MCI_STATUS_UC)) {
3195 if (banks[1] & MCI_STATUS_VAL)
3196 mce->status |= MCI_STATUS_OVER;
3197 banks[2] = mce->addr;
3198 banks[3] = mce->misc;
3199 banks[1] = mce->status;
3200 } else
3201 banks[1] |= MCI_STATUS_OVER;
3202 return 0;
3203 }
3204
3205 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3206 struct kvm_vcpu_events *events)
3207 {
3208 process_nmi(vcpu);
3209 /*
3210 * FIXME: pass injected and pending separately. This is only
3211 * needed for nested virtualization, whose state cannot be
3212 * migrated yet. For now we can combine them.
3213 */
3214 events->exception.injected =
3215 (vcpu->arch.exception.pending ||
3216 vcpu->arch.exception.injected) &&
3217 !kvm_exception_is_soft(vcpu->arch.exception.nr);
3218 events->exception.nr = vcpu->arch.exception.nr;
3219 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3220 events->exception.pad = 0;
3221 events->exception.error_code = vcpu->arch.exception.error_code;
3222
3223 events->interrupt.injected =
3224 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
3225 events->interrupt.nr = vcpu->arch.interrupt.nr;
3226 events->interrupt.soft = 0;
3227 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3228
3229 events->nmi.injected = vcpu->arch.nmi_injected;
3230 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3231 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3232 events->nmi.pad = 0;
3233
3234 events->sipi_vector = 0; /* never valid when reporting to user space */
3235
3236 events->smi.smm = is_smm(vcpu);
3237 events->smi.pending = vcpu->arch.smi_pending;
3238 events->smi.smm_inside_nmi =
3239 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3240 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3241
3242 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3243 | KVM_VCPUEVENT_VALID_SHADOW
3244 | KVM_VCPUEVENT_VALID_SMM);
3245 memset(&events->reserved, 0, sizeof(events->reserved));
3246 }
3247
3248 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3249
3250 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3251 struct kvm_vcpu_events *events)
3252 {
3253 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3254 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3255 | KVM_VCPUEVENT_VALID_SHADOW
3256 | KVM_VCPUEVENT_VALID_SMM))
3257 return -EINVAL;
3258
3259 if (events->exception.injected &&
3260 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR ||
3261 is_guest_mode(vcpu)))
3262 return -EINVAL;
3263
3264 /* INITs are latched while in SMM */
3265 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3266 (events->smi.smm || events->smi.pending) &&
3267 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3268 return -EINVAL;
3269
3270 process_nmi(vcpu);
3271 vcpu->arch.exception.injected = false;
3272 vcpu->arch.exception.pending = events->exception.injected;
3273 vcpu->arch.exception.nr = events->exception.nr;
3274 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3275 vcpu->arch.exception.error_code = events->exception.error_code;
3276
3277 vcpu->arch.interrupt.pending = events->interrupt.injected;
3278 vcpu->arch.interrupt.nr = events->interrupt.nr;
3279 vcpu->arch.interrupt.soft = events->interrupt.soft;
3280 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3281 kvm_x86_ops->set_interrupt_shadow(vcpu,
3282 events->interrupt.shadow);
3283
3284 vcpu->arch.nmi_injected = events->nmi.injected;
3285 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3286 vcpu->arch.nmi_pending = events->nmi.pending;
3287 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3288
3289 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3290 lapic_in_kernel(vcpu))
3291 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3292
3293 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3294 u32 hflags = vcpu->arch.hflags;
3295 if (events->smi.smm)
3296 hflags |= HF_SMM_MASK;
3297 else
3298 hflags &= ~HF_SMM_MASK;
3299 kvm_set_hflags(vcpu, hflags);
3300
3301 vcpu->arch.smi_pending = events->smi.pending;
3302
3303 if (events->smi.smm) {
3304 if (events->smi.smm_inside_nmi)
3305 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3306 else
3307 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3308 if (lapic_in_kernel(vcpu)) {
3309 if (events->smi.latched_init)
3310 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3311 else
3312 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3313 }
3314 }
3315 }
3316
3317 kvm_make_request(KVM_REQ_EVENT, vcpu);
3318
3319 return 0;
3320 }
3321
3322 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3323 struct kvm_debugregs *dbgregs)
3324 {
3325 unsigned long val;
3326
3327 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3328 kvm_get_dr(vcpu, 6, &val);
3329 dbgregs->dr6 = val;
3330 dbgregs->dr7 = vcpu->arch.dr7;
3331 dbgregs->flags = 0;
3332 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3333 }
3334
3335 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3336 struct kvm_debugregs *dbgregs)
3337 {
3338 if (dbgregs->flags)
3339 return -EINVAL;
3340
3341 if (dbgregs->dr6 & ~0xffffffffull)
3342 return -EINVAL;
3343 if (dbgregs->dr7 & ~0xffffffffull)
3344 return -EINVAL;
3345
3346 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3347 kvm_update_dr0123(vcpu);
3348 vcpu->arch.dr6 = dbgregs->dr6;
3349 kvm_update_dr6(vcpu);
3350 vcpu->arch.dr7 = dbgregs->dr7;
3351 kvm_update_dr7(vcpu);
3352
3353 return 0;
3354 }
3355
3356 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3357
3358 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3359 {
3360 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3361 u64 xstate_bv = xsave->header.xfeatures;
3362 u64 valid;
3363
3364 /*
3365 * Copy legacy XSAVE area, to avoid complications with CPUID
3366 * leaves 0 and 1 in the loop below.
3367 */
3368 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3369
3370 /* Set XSTATE_BV */
3371 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3372 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3373
3374 /*
3375 * Copy each region from the possibly compacted offset to the
3376 * non-compacted offset.
3377 */
3378 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3379 while (valid) {
3380 u64 feature = valid & -valid;
3381 int index = fls64(feature) - 1;
3382 void *src = get_xsave_addr(xsave, feature);
3383
3384 if (src) {
3385 u32 size, offset, ecx, edx;
3386 cpuid_count(XSTATE_CPUID, index,
3387 &size, &offset, &ecx, &edx);
3388 if (feature == XFEATURE_MASK_PKRU)
3389 memcpy(dest + offset, &vcpu->arch.pkru,
3390 sizeof(vcpu->arch.pkru));
3391 else
3392 memcpy(dest + offset, src, size);
3393
3394 }
3395
3396 valid -= feature;
3397 }
3398 }
3399
3400 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3401 {
3402 struct xregs_state *xsave = &vcpu->arch.guest_fpu.state.xsave;
3403 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3404 u64 valid;
3405
3406 /*
3407 * Copy legacy XSAVE area, to avoid complications with CPUID
3408 * leaves 0 and 1 in the loop below.
3409 */
3410 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3411
3412 /* Set XSTATE_BV and possibly XCOMP_BV. */
3413 xsave->header.xfeatures = xstate_bv;
3414 if (boot_cpu_has(X86_FEATURE_XSAVES))
3415 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3416
3417 /*
3418 * Copy each region from the non-compacted offset to the
3419 * possibly compacted offset.
3420 */
3421 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3422 while (valid) {
3423 u64 feature = valid & -valid;
3424 int index = fls64(feature) - 1;
3425 void *dest = get_xsave_addr(xsave, feature);
3426
3427 if (dest) {
3428 u32 size, offset, ecx, edx;
3429 cpuid_count(XSTATE_CPUID, index,
3430 &size, &offset, &ecx, &edx);
3431 if (feature == XFEATURE_MASK_PKRU)
3432 memcpy(&vcpu->arch.pkru, src + offset,
3433 sizeof(vcpu->arch.pkru));
3434 else
3435 memcpy(dest, src + offset, size);
3436 }
3437
3438 valid -= feature;
3439 }
3440 }
3441
3442 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3443 struct kvm_xsave *guest_xsave)
3444 {
3445 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3446 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3447 fill_xsave((u8 *) guest_xsave->region, vcpu);
3448 } else {
3449 memcpy(guest_xsave->region,
3450 &vcpu->arch.guest_fpu.state.fxsave,
3451 sizeof(struct fxregs_state));
3452 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3453 XFEATURE_MASK_FPSSE;
3454 }
3455 }
3456
3457 #define XSAVE_MXCSR_OFFSET 24
3458
3459 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3460 struct kvm_xsave *guest_xsave)
3461 {
3462 u64 xstate_bv =
3463 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3464 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3465
3466 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3467 /*
3468 * Here we allow setting states that are not present in
3469 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3470 * with old userspace.
3471 */
3472 if (xstate_bv & ~kvm_supported_xcr0() ||
3473 mxcsr & ~mxcsr_feature_mask)
3474 return -EINVAL;
3475 load_xsave(vcpu, (u8 *)guest_xsave->region);
3476 } else {
3477 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3478 mxcsr & ~mxcsr_feature_mask)
3479 return -EINVAL;
3480 memcpy(&vcpu->arch.guest_fpu.state.fxsave,
3481 guest_xsave->region, sizeof(struct fxregs_state));
3482 }
3483 return 0;
3484 }
3485
3486 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3487 struct kvm_xcrs *guest_xcrs)
3488 {
3489 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3490 guest_xcrs->nr_xcrs = 0;
3491 return;
3492 }
3493
3494 guest_xcrs->nr_xcrs = 1;
3495 guest_xcrs->flags = 0;
3496 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3497 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3498 }
3499
3500 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3501 struct kvm_xcrs *guest_xcrs)
3502 {
3503 int i, r = 0;
3504
3505 if (!boot_cpu_has(X86_FEATURE_XSAVE))
3506 return -EINVAL;
3507
3508 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3509 return -EINVAL;
3510
3511 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3512 /* Only support XCR0 currently */
3513 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3514 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3515 guest_xcrs->xcrs[i].value);
3516 break;
3517 }
3518 if (r)
3519 r = -EINVAL;
3520 return r;
3521 }
3522
3523 /*
3524 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3525 * stopped by the hypervisor. This function will be called from the host only.
3526 * EINVAL is returned when the host attempts to set the flag for a guest that
3527 * does not support pv clocks.
3528 */
3529 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3530 {
3531 if (!vcpu->arch.pv_time_enabled)
3532 return -EINVAL;
3533 vcpu->arch.pvclock_set_guest_stopped_request = true;
3534 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3535 return 0;
3536 }
3537
3538 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3539 struct kvm_enable_cap *cap)
3540 {
3541 if (cap->flags)
3542 return -EINVAL;
3543
3544 switch (cap->cap) {
3545 case KVM_CAP_HYPERV_SYNIC2:
3546 if (cap->args[0])
3547 return -EINVAL;
3548 case KVM_CAP_HYPERV_SYNIC:
3549 if (!irqchip_in_kernel(vcpu->kvm))
3550 return -EINVAL;
3551 return kvm_hv_activate_synic(vcpu, cap->cap ==
3552 KVM_CAP_HYPERV_SYNIC2);
3553 default:
3554 return -EINVAL;
3555 }
3556 }
3557
3558 long kvm_arch_vcpu_ioctl(struct file *filp,
3559 unsigned int ioctl, unsigned long arg)
3560 {
3561 struct kvm_vcpu *vcpu = filp->private_data;
3562 void __user *argp = (void __user *)arg;
3563 int r;
3564 union {
3565 struct kvm_lapic_state *lapic;
3566 struct kvm_xsave *xsave;
3567 struct kvm_xcrs *xcrs;
3568 void *buffer;
3569 } u;
3570
3571 u.buffer = NULL;
3572 switch (ioctl) {
3573 case KVM_GET_LAPIC: {
3574 r = -EINVAL;
3575 if (!lapic_in_kernel(vcpu))
3576 goto out;
3577 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3578
3579 r = -ENOMEM;
3580 if (!u.lapic)
3581 goto out;
3582 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3583 if (r)
3584 goto out;
3585 r = -EFAULT;
3586 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3587 goto out;
3588 r = 0;
3589 break;
3590 }
3591 case KVM_SET_LAPIC: {
3592 r = -EINVAL;
3593 if (!lapic_in_kernel(vcpu))
3594 goto out;
3595 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3596 if (IS_ERR(u.lapic))
3597 return PTR_ERR(u.lapic);
3598
3599 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3600 break;
3601 }
3602 case KVM_INTERRUPT: {
3603 struct kvm_interrupt irq;
3604
3605 r = -EFAULT;
3606 if (copy_from_user(&irq, argp, sizeof irq))
3607 goto out;
3608 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3609 break;
3610 }
3611 case KVM_NMI: {
3612 r = kvm_vcpu_ioctl_nmi(vcpu);
3613 break;
3614 }
3615 case KVM_SMI: {
3616 r = kvm_vcpu_ioctl_smi(vcpu);
3617 break;
3618 }
3619 case KVM_SET_CPUID: {
3620 struct kvm_cpuid __user *cpuid_arg = argp;
3621 struct kvm_cpuid cpuid;
3622
3623 r = -EFAULT;
3624 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3625 goto out;
3626 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3627 break;
3628 }
3629 case KVM_SET_CPUID2: {
3630 struct kvm_cpuid2 __user *cpuid_arg = argp;
3631 struct kvm_cpuid2 cpuid;
3632
3633 r = -EFAULT;
3634 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3635 goto out;
3636 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3637 cpuid_arg->entries);
3638 break;
3639 }
3640 case KVM_GET_CPUID2: {
3641 struct kvm_cpuid2 __user *cpuid_arg = argp;
3642 struct kvm_cpuid2 cpuid;
3643
3644 r = -EFAULT;
3645 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
3646 goto out;
3647 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3648 cpuid_arg->entries);
3649 if (r)
3650 goto out;
3651 r = -EFAULT;
3652 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
3653 goto out;
3654 r = 0;
3655 break;
3656 }
3657 case KVM_GET_MSRS: {
3658 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3659 r = msr_io(vcpu, argp, do_get_msr, 1);
3660 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3661 break;
3662 }
3663 case KVM_SET_MSRS: {
3664 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3665 r = msr_io(vcpu, argp, do_set_msr, 0);
3666 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3667 break;
3668 }
3669 case KVM_TPR_ACCESS_REPORTING: {
3670 struct kvm_tpr_access_ctl tac;
3671
3672 r = -EFAULT;
3673 if (copy_from_user(&tac, argp, sizeof tac))
3674 goto out;
3675 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3676 if (r)
3677 goto out;
3678 r = -EFAULT;
3679 if (copy_to_user(argp, &tac, sizeof tac))
3680 goto out;
3681 r = 0;
3682 break;
3683 };
3684 case KVM_SET_VAPIC_ADDR: {
3685 struct kvm_vapic_addr va;
3686 int idx;
3687
3688 r = -EINVAL;
3689 if (!lapic_in_kernel(vcpu))
3690 goto out;
3691 r = -EFAULT;
3692 if (copy_from_user(&va, argp, sizeof va))
3693 goto out;
3694 idx = srcu_read_lock(&vcpu->kvm->srcu);
3695 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3696 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3697 break;
3698 }
3699 case KVM_X86_SETUP_MCE: {
3700 u64 mcg_cap;
3701
3702 r = -EFAULT;
3703 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
3704 goto out;
3705 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
3706 break;
3707 }
3708 case KVM_X86_SET_MCE: {
3709 struct kvm_x86_mce mce;
3710
3711 r = -EFAULT;
3712 if (copy_from_user(&mce, argp, sizeof mce))
3713 goto out;
3714 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3715 break;
3716 }
3717 case KVM_GET_VCPU_EVENTS: {
3718 struct kvm_vcpu_events events;
3719
3720 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3721
3722 r = -EFAULT;
3723 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3724 break;
3725 r = 0;
3726 break;
3727 }
3728 case KVM_SET_VCPU_EVENTS: {
3729 struct kvm_vcpu_events events;
3730
3731 r = -EFAULT;
3732 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3733 break;
3734
3735 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3736 break;
3737 }
3738 case KVM_GET_DEBUGREGS: {
3739 struct kvm_debugregs dbgregs;
3740
3741 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3742
3743 r = -EFAULT;
3744 if (copy_to_user(argp, &dbgregs,
3745 sizeof(struct kvm_debugregs)))
3746 break;
3747 r = 0;
3748 break;
3749 }
3750 case KVM_SET_DEBUGREGS: {
3751 struct kvm_debugregs dbgregs;
3752
3753 r = -EFAULT;
3754 if (copy_from_user(&dbgregs, argp,
3755 sizeof(struct kvm_debugregs)))
3756 break;
3757
3758 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3759 break;
3760 }
3761 case KVM_GET_XSAVE: {
3762 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3763 r = -ENOMEM;
3764 if (!u.xsave)
3765 break;
3766
3767 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3768
3769 r = -EFAULT;
3770 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3771 break;
3772 r = 0;
3773 break;
3774 }
3775 case KVM_SET_XSAVE: {
3776 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3777 if (IS_ERR(u.xsave))
3778 return PTR_ERR(u.xsave);
3779
3780 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3781 break;
3782 }
3783 case KVM_GET_XCRS: {
3784 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3785 r = -ENOMEM;
3786 if (!u.xcrs)
3787 break;
3788
3789 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3790
3791 r = -EFAULT;
3792 if (copy_to_user(argp, u.xcrs,
3793 sizeof(struct kvm_xcrs)))
3794 break;
3795 r = 0;
3796 break;
3797 }
3798 case KVM_SET_XCRS: {
3799 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3800 if (IS_ERR(u.xcrs))
3801 return PTR_ERR(u.xcrs);
3802
3803 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3804 break;
3805 }
3806 case KVM_SET_TSC_KHZ: {
3807 u32 user_tsc_khz;
3808
3809 r = -EINVAL;
3810 user_tsc_khz = (u32)arg;
3811
3812 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3813 goto out;
3814
3815 if (user_tsc_khz == 0)
3816 user_tsc_khz = tsc_khz;
3817
3818 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
3819 r = 0;
3820
3821 goto out;
3822 }
3823 case KVM_GET_TSC_KHZ: {
3824 r = vcpu->arch.virtual_tsc_khz;
3825 goto out;
3826 }
3827 case KVM_KVMCLOCK_CTRL: {
3828 r = kvm_set_guest_paused(vcpu);
3829 goto out;
3830 }
3831 case KVM_ENABLE_CAP: {
3832 struct kvm_enable_cap cap;
3833
3834 r = -EFAULT;
3835 if (copy_from_user(&cap, argp, sizeof(cap)))
3836 goto out;
3837 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
3838 break;
3839 }
3840 default:
3841 r = -EINVAL;
3842 }
3843 out:
3844 kfree(u.buffer);
3845 return r;
3846 }
3847
3848 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
3849 {
3850 return VM_FAULT_SIGBUS;
3851 }
3852
3853 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3854 {
3855 int ret;
3856
3857 if (addr > (unsigned int)(-3 * PAGE_SIZE))
3858 return -EINVAL;
3859 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3860 return ret;
3861 }
3862
3863 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3864 u64 ident_addr)
3865 {
3866 kvm->arch.ept_identity_map_addr = ident_addr;
3867 return 0;
3868 }
3869
3870 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3871 u32 kvm_nr_mmu_pages)
3872 {
3873 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3874 return -EINVAL;
3875
3876 mutex_lock(&kvm->slots_lock);
3877
3878 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3879 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3880
3881 mutex_unlock(&kvm->slots_lock);
3882 return 0;
3883 }
3884
3885 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3886 {
3887 return kvm->arch.n_max_mmu_pages;
3888 }
3889
3890 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3891 {
3892 struct kvm_pic *pic = kvm->arch.vpic;
3893 int r;
3894
3895 r = 0;
3896 switch (chip->chip_id) {
3897 case KVM_IRQCHIP_PIC_MASTER:
3898 memcpy(&chip->chip.pic, &pic->pics[0],
3899 sizeof(struct kvm_pic_state));
3900 break;
3901 case KVM_IRQCHIP_PIC_SLAVE:
3902 memcpy(&chip->chip.pic, &pic->pics[1],
3903 sizeof(struct kvm_pic_state));
3904 break;
3905 case KVM_IRQCHIP_IOAPIC:
3906 kvm_get_ioapic(kvm, &chip->chip.ioapic);
3907 break;
3908 default:
3909 r = -EINVAL;
3910 break;
3911 }
3912 return r;
3913 }
3914
3915 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3916 {
3917 struct kvm_pic *pic = kvm->arch.vpic;
3918 int r;
3919
3920 r = 0;
3921 switch (chip->chip_id) {
3922 case KVM_IRQCHIP_PIC_MASTER:
3923 spin_lock(&pic->lock);
3924 memcpy(&pic->pics[0], &chip->chip.pic,
3925 sizeof(struct kvm_pic_state));
3926 spin_unlock(&pic->lock);
3927 break;
3928 case KVM_IRQCHIP_PIC_SLAVE:
3929 spin_lock(&pic->lock);
3930 memcpy(&pic->pics[1], &chip->chip.pic,
3931 sizeof(struct kvm_pic_state));
3932 spin_unlock(&pic->lock);
3933 break;
3934 case KVM_IRQCHIP_IOAPIC:
3935 kvm_set_ioapic(kvm, &chip->chip.ioapic);
3936 break;
3937 default:
3938 r = -EINVAL;
3939 break;
3940 }
3941 kvm_pic_update_irq(pic);
3942 return r;
3943 }
3944
3945 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3946 {
3947 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
3948
3949 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
3950
3951 mutex_lock(&kps->lock);
3952 memcpy(ps, &kps->channels, sizeof(*ps));
3953 mutex_unlock(&kps->lock);
3954 return 0;
3955 }
3956
3957 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3958 {
3959 int i;
3960 struct kvm_pit *pit = kvm->arch.vpit;
3961
3962 mutex_lock(&pit->pit_state.lock);
3963 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
3964 for (i = 0; i < 3; i++)
3965 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
3966 mutex_unlock(&pit->pit_state.lock);
3967 return 0;
3968 }
3969
3970 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3971 {
3972 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3973 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3974 sizeof(ps->channels));
3975 ps->flags = kvm->arch.vpit->pit_state.flags;
3976 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3977 memset(&ps->reserved, 0, sizeof(ps->reserved));
3978 return 0;
3979 }
3980
3981 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3982 {
3983 int start = 0;
3984 int i;
3985 u32 prev_legacy, cur_legacy;
3986 struct kvm_pit *pit = kvm->arch.vpit;
3987
3988 mutex_lock(&pit->pit_state.lock);
3989 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3990 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3991 if (!prev_legacy && cur_legacy)
3992 start = 1;
3993 memcpy(&pit->pit_state.channels, &ps->channels,
3994 sizeof(pit->pit_state.channels));
3995 pit->pit_state.flags = ps->flags;
3996 for (i = 0; i < 3; i++)
3997 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
3998 start && i == 0);
3999 mutex_unlock(&pit->pit_state.lock);
4000 return 0;
4001 }
4002
4003 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4004 struct kvm_reinject_control *control)
4005 {
4006 struct kvm_pit *pit = kvm->arch.vpit;
4007
4008 if (!pit)
4009 return -ENXIO;
4010
4011 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4012 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4013 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4014 */
4015 mutex_lock(&pit->pit_state.lock);
4016 kvm_pit_set_reinject(pit, control->pit_reinject);
4017 mutex_unlock(&pit->pit_state.lock);
4018
4019 return 0;
4020 }
4021
4022 /**
4023 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4024 * @kvm: kvm instance
4025 * @log: slot id and address to which we copy the log
4026 *
4027 * Steps 1-4 below provide general overview of dirty page logging. See
4028 * kvm_get_dirty_log_protect() function description for additional details.
4029 *
4030 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4031 * always flush the TLB (step 4) even if previous step failed and the dirty
4032 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4033 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4034 * writes will be marked dirty for next log read.
4035 *
4036 * 1. Take a snapshot of the bit and clear it if needed.
4037 * 2. Write protect the corresponding page.
4038 * 3. Copy the snapshot to the userspace.
4039 * 4. Flush TLB's if needed.
4040 */
4041 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4042 {
4043 bool is_dirty = false;
4044 int r;
4045
4046 mutex_lock(&kvm->slots_lock);
4047
4048 /*
4049 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4050 */
4051 if (kvm_x86_ops->flush_log_dirty)
4052 kvm_x86_ops->flush_log_dirty(kvm);
4053
4054 r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
4055
4056 /*
4057 * All the TLBs can be flushed out of mmu lock, see the comments in
4058 * kvm_mmu_slot_remove_write_access().
4059 */
4060 lockdep_assert_held(&kvm->slots_lock);
4061 if (is_dirty)
4062 kvm_flush_remote_tlbs(kvm);
4063
4064 mutex_unlock(&kvm->slots_lock);
4065 return r;
4066 }
4067
4068 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4069 bool line_status)
4070 {
4071 if (!irqchip_in_kernel(kvm))
4072 return -ENXIO;
4073
4074 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4075 irq_event->irq, irq_event->level,
4076 line_status);
4077 return 0;
4078 }
4079
4080 static int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4081 struct kvm_enable_cap *cap)
4082 {
4083 int r;
4084
4085 if (cap->flags)
4086 return -EINVAL;
4087
4088 switch (cap->cap) {
4089 case KVM_CAP_DISABLE_QUIRKS:
4090 kvm->arch.disabled_quirks = cap->args[0];
4091 r = 0;
4092 break;
4093 case KVM_CAP_SPLIT_IRQCHIP: {
4094 mutex_lock(&kvm->lock);
4095 r = -EINVAL;
4096 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4097 goto split_irqchip_unlock;
4098 r = -EEXIST;
4099 if (irqchip_in_kernel(kvm))
4100 goto split_irqchip_unlock;
4101 if (kvm->created_vcpus)
4102 goto split_irqchip_unlock;
4103 r = kvm_setup_empty_irq_routing(kvm);
4104 if (r)
4105 goto split_irqchip_unlock;
4106 /* Pairs with irqchip_in_kernel. */
4107 smp_wmb();
4108 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4109 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4110 r = 0;
4111 split_irqchip_unlock:
4112 mutex_unlock(&kvm->lock);
4113 break;
4114 }
4115 case KVM_CAP_X2APIC_API:
4116 r = -EINVAL;
4117 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4118 break;
4119
4120 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4121 kvm->arch.x2apic_format = true;
4122 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4123 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4124
4125 r = 0;
4126 break;
4127 default:
4128 r = -EINVAL;
4129 break;
4130 }
4131 return r;
4132 }
4133
4134 long kvm_arch_vm_ioctl(struct file *filp,
4135 unsigned int ioctl, unsigned long arg)
4136 {
4137 struct kvm *kvm = filp->private_data;
4138 void __user *argp = (void __user *)arg;
4139 int r = -ENOTTY;
4140 /*
4141 * This union makes it completely explicit to gcc-3.x
4142 * that these two variables' stack usage should be
4143 * combined, not added together.
4144 */
4145 union {
4146 struct kvm_pit_state ps;
4147 struct kvm_pit_state2 ps2;
4148 struct kvm_pit_config pit_config;
4149 } u;
4150
4151 switch (ioctl) {
4152 case KVM_SET_TSS_ADDR:
4153 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4154 break;
4155 case KVM_SET_IDENTITY_MAP_ADDR: {
4156 u64 ident_addr;
4157
4158 mutex_lock(&kvm->lock);
4159 r = -EINVAL;
4160 if (kvm->created_vcpus)
4161 goto set_identity_unlock;
4162 r = -EFAULT;
4163 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
4164 goto set_identity_unlock;
4165 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4166 set_identity_unlock:
4167 mutex_unlock(&kvm->lock);
4168 break;
4169 }
4170 case KVM_SET_NR_MMU_PAGES:
4171 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4172 break;
4173 case KVM_GET_NR_MMU_PAGES:
4174 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4175 break;
4176 case KVM_CREATE_IRQCHIP: {
4177 mutex_lock(&kvm->lock);
4178
4179 r = -EEXIST;
4180 if (irqchip_in_kernel(kvm))
4181 goto create_irqchip_unlock;
4182
4183 r = -EINVAL;
4184 if (kvm->created_vcpus)
4185 goto create_irqchip_unlock;
4186
4187 r = kvm_pic_init(kvm);
4188 if (r)
4189 goto create_irqchip_unlock;
4190
4191 r = kvm_ioapic_init(kvm);
4192 if (r) {
4193 kvm_pic_destroy(kvm);
4194 goto create_irqchip_unlock;
4195 }
4196
4197 r = kvm_setup_default_irq_routing(kvm);
4198 if (r) {
4199 kvm_ioapic_destroy(kvm);
4200 kvm_pic_destroy(kvm);
4201 goto create_irqchip_unlock;
4202 }
4203 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4204 smp_wmb();
4205 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4206 create_irqchip_unlock:
4207 mutex_unlock(&kvm->lock);
4208 break;
4209 }
4210 case KVM_CREATE_PIT:
4211 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4212 goto create_pit;
4213 case KVM_CREATE_PIT2:
4214 r = -EFAULT;
4215 if (copy_from_user(&u.pit_config, argp,
4216 sizeof(struct kvm_pit_config)))
4217 goto out;
4218 create_pit:
4219 mutex_lock(&kvm->lock);
4220 r = -EEXIST;
4221 if (kvm->arch.vpit)
4222 goto create_pit_unlock;
4223 r = -ENOMEM;
4224 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4225 if (kvm->arch.vpit)
4226 r = 0;
4227 create_pit_unlock:
4228 mutex_unlock(&kvm->lock);
4229 break;
4230 case KVM_GET_IRQCHIP: {
4231 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4232 struct kvm_irqchip *chip;
4233
4234 chip = memdup_user(argp, sizeof(*chip));
4235 if (IS_ERR(chip)) {
4236 r = PTR_ERR(chip);
4237 goto out;
4238 }
4239
4240 r = -ENXIO;
4241 if (!irqchip_kernel(kvm))
4242 goto get_irqchip_out;
4243 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4244 if (r)
4245 goto get_irqchip_out;
4246 r = -EFAULT;
4247 if (copy_to_user(argp, chip, sizeof *chip))
4248 goto get_irqchip_out;
4249 r = 0;
4250 get_irqchip_out:
4251 kfree(chip);
4252 break;
4253 }
4254 case KVM_SET_IRQCHIP: {
4255 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4256 struct kvm_irqchip *chip;
4257
4258 chip = memdup_user(argp, sizeof(*chip));
4259 if (IS_ERR(chip)) {
4260 r = PTR_ERR(chip);
4261 goto out;
4262 }
4263
4264 r = -ENXIO;
4265 if (!irqchip_kernel(kvm))
4266 goto set_irqchip_out;
4267 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4268 if (r)
4269 goto set_irqchip_out;
4270 r = 0;
4271 set_irqchip_out:
4272 kfree(chip);
4273 break;
4274 }
4275 case KVM_GET_PIT: {
4276 r = -EFAULT;
4277 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4278 goto out;
4279 r = -ENXIO;
4280 if (!kvm->arch.vpit)
4281 goto out;
4282 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4283 if (r)
4284 goto out;
4285 r = -EFAULT;
4286 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4287 goto out;
4288 r = 0;
4289 break;
4290 }
4291 case KVM_SET_PIT: {
4292 r = -EFAULT;
4293 if (copy_from_user(&u.ps, argp, sizeof u.ps))
4294 goto out;
4295 r = -ENXIO;
4296 if (!kvm->arch.vpit)
4297 goto out;
4298 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4299 break;
4300 }
4301 case KVM_GET_PIT2: {
4302 r = -ENXIO;
4303 if (!kvm->arch.vpit)
4304 goto out;
4305 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4306 if (r)
4307 goto out;
4308 r = -EFAULT;
4309 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4310 goto out;
4311 r = 0;
4312 break;
4313 }
4314 case KVM_SET_PIT2: {
4315 r = -EFAULT;
4316 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4317 goto out;
4318 r = -ENXIO;
4319 if (!kvm->arch.vpit)
4320 goto out;
4321 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4322 break;
4323 }
4324 case KVM_REINJECT_CONTROL: {
4325 struct kvm_reinject_control control;
4326 r = -EFAULT;
4327 if (copy_from_user(&control, argp, sizeof(control)))
4328 goto out;
4329 r = kvm_vm_ioctl_reinject(kvm, &control);
4330 break;
4331 }
4332 case KVM_SET_BOOT_CPU_ID:
4333 r = 0;
4334 mutex_lock(&kvm->lock);
4335 if (kvm->created_vcpus)
4336 r = -EBUSY;
4337 else
4338 kvm->arch.bsp_vcpu_id = arg;
4339 mutex_unlock(&kvm->lock);
4340 break;
4341 case KVM_XEN_HVM_CONFIG: {
4342 struct kvm_xen_hvm_config xhc;
4343 r = -EFAULT;
4344 if (copy_from_user(&xhc, argp, sizeof(xhc)))
4345 goto out;
4346 r = -EINVAL;
4347 if (xhc.flags)
4348 goto out;
4349 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
4350 r = 0;
4351 break;
4352 }
4353 case KVM_SET_CLOCK: {
4354 struct kvm_clock_data user_ns;
4355 u64 now_ns;
4356
4357 r = -EFAULT;
4358 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4359 goto out;
4360
4361 r = -EINVAL;
4362 if (user_ns.flags)
4363 goto out;
4364
4365 r = 0;
4366 /*
4367 * TODO: userspace has to take care of races with VCPU_RUN, so
4368 * kvm_gen_update_masterclock() can be cut down to locked
4369 * pvclock_update_vm_gtod_copy().
4370 */
4371 kvm_gen_update_masterclock(kvm);
4372 now_ns = get_kvmclock_ns(kvm);
4373 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4374 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4375 break;
4376 }
4377 case KVM_GET_CLOCK: {
4378 struct kvm_clock_data user_ns;
4379 u64 now_ns;
4380
4381 now_ns = get_kvmclock_ns(kvm);
4382 user_ns.clock = now_ns;
4383 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4384 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4385
4386 r = -EFAULT;
4387 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4388 goto out;
4389 r = 0;
4390 break;
4391 }
4392 case KVM_ENABLE_CAP: {
4393 struct kvm_enable_cap cap;
4394
4395 r = -EFAULT;
4396 if (copy_from_user(&cap, argp, sizeof(cap)))
4397 goto out;
4398 r = kvm_vm_ioctl_enable_cap(kvm, &cap);
4399 break;
4400 }
4401 default:
4402 r = -ENOTTY;
4403 }
4404 out:
4405 return r;
4406 }
4407
4408 static void kvm_init_msr_list(void)
4409 {
4410 u32 dummy[2];
4411 unsigned i, j;
4412
4413 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4414 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4415 continue;
4416
4417 /*
4418 * Even MSRs that are valid in the host may not be exposed
4419 * to the guests in some cases.
4420 */
4421 switch (msrs_to_save[i]) {
4422 case MSR_IA32_BNDCFGS:
4423 if (!kvm_x86_ops->mpx_supported())
4424 continue;
4425 break;
4426 case MSR_TSC_AUX:
4427 if (!kvm_x86_ops->rdtscp_supported())
4428 continue;
4429 break;
4430 default:
4431 break;
4432 }
4433
4434 if (j < i)
4435 msrs_to_save[j] = msrs_to_save[i];
4436 j++;
4437 }
4438 num_msrs_to_save = j;
4439
4440 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4441 if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
4442 continue;
4443
4444 if (j < i)
4445 emulated_msrs[j] = emulated_msrs[i];
4446 j++;
4447 }
4448 num_emulated_msrs = j;
4449
4450 for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
4451 struct kvm_msr_entry msr;
4452
4453 msr.index = msr_based_features[i];
4454 if (kvm_get_msr_feature(&msr))
4455 continue;
4456
4457 if (j < i)
4458 msr_based_features[j] = msr_based_features[i];
4459 j++;
4460 }
4461 num_msr_based_features = j;
4462 }
4463
4464 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4465 const void *v)
4466 {
4467 int handled = 0;
4468 int n;
4469
4470 do {
4471 n = min(len, 8);
4472 if (!(lapic_in_kernel(vcpu) &&
4473 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4474 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4475 break;
4476 handled += n;
4477 addr += n;
4478 len -= n;
4479 v += n;
4480 } while (len);
4481
4482 return handled;
4483 }
4484
4485 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4486 {
4487 int handled = 0;
4488 int n;
4489
4490 do {
4491 n = min(len, 8);
4492 if (!(lapic_in_kernel(vcpu) &&
4493 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4494 addr, n, v))
4495 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4496 break;
4497 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
4498 handled += n;
4499 addr += n;
4500 len -= n;
4501 v += n;
4502 } while (len);
4503
4504 return handled;
4505 }
4506
4507 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4508 struct kvm_segment *var, int seg)
4509 {
4510 kvm_x86_ops->set_segment(vcpu, var, seg);
4511 }
4512
4513 void kvm_get_segment(struct kvm_vcpu *vcpu,
4514 struct kvm_segment *var, int seg)
4515 {
4516 kvm_x86_ops->get_segment(vcpu, var, seg);
4517 }
4518
4519 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4520 struct x86_exception *exception)
4521 {
4522 gpa_t t_gpa;
4523
4524 BUG_ON(!mmu_is_nested(vcpu));
4525
4526 /* NPT walks are always user-walks */
4527 access |= PFERR_USER_MASK;
4528 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, exception);
4529
4530 return t_gpa;
4531 }
4532
4533 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4534 struct x86_exception *exception)
4535 {
4536 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4537 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4538 }
4539
4540 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4541 struct x86_exception *exception)
4542 {
4543 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4544 access |= PFERR_FETCH_MASK;
4545 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4546 }
4547
4548 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
4549 struct x86_exception *exception)
4550 {
4551 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4552 access |= PFERR_WRITE_MASK;
4553 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4554 }
4555
4556 /* uses this to access any guest's mapped memory without checking CPL */
4557 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
4558 struct x86_exception *exception)
4559 {
4560 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
4561 }
4562
4563 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4564 struct kvm_vcpu *vcpu, u32 access,
4565 struct x86_exception *exception)
4566 {
4567 void *data = val;
4568 int r = X86EMUL_CONTINUE;
4569
4570 while (bytes) {
4571 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
4572 exception);
4573 unsigned offset = addr & (PAGE_SIZE-1);
4574 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
4575 int ret;
4576
4577 if (gpa == UNMAPPED_GVA)
4578 return X86EMUL_PROPAGATE_FAULT;
4579 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
4580 offset, toread);
4581 if (ret < 0) {
4582 r = X86EMUL_IO_NEEDED;
4583 goto out;
4584 }
4585
4586 bytes -= toread;
4587 data += toread;
4588 addr += toread;
4589 }
4590 out:
4591 return r;
4592 }
4593
4594 /* used for instruction fetching */
4595 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
4596 gva_t addr, void *val, unsigned int bytes,
4597 struct x86_exception *exception)
4598 {
4599 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4600 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4601 unsigned offset;
4602 int ret;
4603
4604 /* Inline kvm_read_guest_virt_helper for speed. */
4605 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
4606 exception);
4607 if (unlikely(gpa == UNMAPPED_GVA))
4608 return X86EMUL_PROPAGATE_FAULT;
4609
4610 offset = addr & (PAGE_SIZE-1);
4611 if (WARN_ON(offset + bytes > PAGE_SIZE))
4612 bytes = (unsigned)PAGE_SIZE - offset;
4613 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
4614 offset, bytes);
4615 if (unlikely(ret < 0))
4616 return X86EMUL_IO_NEEDED;
4617
4618 return X86EMUL_CONTINUE;
4619 }
4620
4621 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
4622 gva_t addr, void *val, unsigned int bytes,
4623 struct x86_exception *exception)
4624 {
4625 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4626
4627 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
4628 exception);
4629 }
4630 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
4631
4632 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
4633 gva_t addr, void *val, unsigned int bytes,
4634 struct x86_exception *exception)
4635 {
4636 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4637 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
4638 }
4639
4640 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
4641 unsigned long addr, void *val, unsigned int bytes)
4642 {
4643 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4644 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
4645
4646 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
4647 }
4648
4649 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
4650 struct kvm_vcpu *vcpu, u32 access,
4651 struct x86_exception *exception)
4652 {
4653 void *data = val;
4654 int r = X86EMUL_CONTINUE;
4655
4656 /* kvm_write_guest_virt_system can pull in tons of pages. */
4657 vcpu->arch.l1tf_flush_l1d = true;
4658
4659 while (bytes) {
4660 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
4661 access,
4662 exception);
4663 unsigned offset = addr & (PAGE_SIZE-1);
4664 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
4665 int ret;
4666
4667 if (gpa == UNMAPPED_GVA)
4668 return X86EMUL_PROPAGATE_FAULT;
4669 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
4670 if (ret < 0) {
4671 r = X86EMUL_IO_NEEDED;
4672 goto out;
4673 }
4674
4675 bytes -= towrite;
4676 data += towrite;
4677 addr += towrite;
4678 }
4679 out:
4680 return r;
4681 }
4682
4683 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
4684 unsigned int bytes, struct x86_exception *exception)
4685 {
4686 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4687
4688 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
4689 PFERR_WRITE_MASK, exception);
4690 }
4691
4692 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
4693 unsigned int bytes, struct x86_exception *exception)
4694 {
4695 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
4696 PFERR_WRITE_MASK, exception);
4697 }
4698 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
4699
4700 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4701 gpa_t gpa, bool write)
4702 {
4703 /* For APIC access vmexit */
4704 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4705 return 1;
4706
4707 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
4708 trace_vcpu_match_mmio(gva, gpa, write, true);
4709 return 1;
4710 }
4711
4712 return 0;
4713 }
4714
4715 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
4716 gpa_t *gpa, struct x86_exception *exception,
4717 bool write)
4718 {
4719 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
4720 | (write ? PFERR_WRITE_MASK : 0);
4721
4722 /*
4723 * currently PKRU is only applied to ept enabled guest so
4724 * there is no pkey in EPT page table for L1 guest or EPT
4725 * shadow page table for L2 guest.
4726 */
4727 if (vcpu_match_mmio_gva(vcpu, gva)
4728 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
4729 vcpu->arch.access, 0, access)) {
4730 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
4731 (gva & (PAGE_SIZE - 1));
4732 trace_vcpu_match_mmio(gva, *gpa, write, false);
4733 return 1;
4734 }
4735
4736 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4737
4738 if (*gpa == UNMAPPED_GVA)
4739 return -1;
4740
4741 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
4742 }
4743
4744 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4745 const void *val, int bytes)
4746 {
4747 int ret;
4748
4749 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
4750 if (ret < 0)
4751 return 0;
4752 kvm_page_track_write(vcpu, gpa, val, bytes);
4753 return 1;
4754 }
4755
4756 struct read_write_emulator_ops {
4757 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
4758 int bytes);
4759 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
4760 void *val, int bytes);
4761 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4762 int bytes, void *val);
4763 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
4764 void *val, int bytes);
4765 bool write;
4766 };
4767
4768 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
4769 {
4770 if (vcpu->mmio_read_completed) {
4771 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
4772 vcpu->mmio_fragments[0].gpa, val);
4773 vcpu->mmio_read_completed = 0;
4774 return 1;
4775 }
4776
4777 return 0;
4778 }
4779
4780 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4781 void *val, int bytes)
4782 {
4783 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
4784 }
4785
4786 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
4787 void *val, int bytes)
4788 {
4789 return emulator_write_phys(vcpu, gpa, val, bytes);
4790 }
4791
4792 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
4793 {
4794 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
4795 return vcpu_mmio_write(vcpu, gpa, bytes, val);
4796 }
4797
4798 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4799 void *val, int bytes)
4800 {
4801 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
4802 return X86EMUL_IO_NEEDED;
4803 }
4804
4805 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
4806 void *val, int bytes)
4807 {
4808 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
4809
4810 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
4811 return X86EMUL_CONTINUE;
4812 }
4813
4814 static const struct read_write_emulator_ops read_emultor = {
4815 .read_write_prepare = read_prepare,
4816 .read_write_emulate = read_emulate,
4817 .read_write_mmio = vcpu_mmio_read,
4818 .read_write_exit_mmio = read_exit_mmio,
4819 };
4820
4821 static const struct read_write_emulator_ops write_emultor = {
4822 .read_write_emulate = write_emulate,
4823 .read_write_mmio = write_mmio,
4824 .read_write_exit_mmio = write_exit_mmio,
4825 .write = true,
4826 };
4827
4828 static int emulator_read_write_onepage(unsigned long addr, void *val,
4829 unsigned int bytes,
4830 struct x86_exception *exception,
4831 struct kvm_vcpu *vcpu,
4832 const struct read_write_emulator_ops *ops)
4833 {
4834 gpa_t gpa;
4835 int handled, ret;
4836 bool write = ops->write;
4837 struct kvm_mmio_fragment *frag;
4838 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4839
4840 /*
4841 * If the exit was due to a NPF we may already have a GPA.
4842 * If the GPA is present, use it to avoid the GVA to GPA table walk.
4843 * Note, this cannot be used on string operations since string
4844 * operation using rep will only have the initial GPA from the NPF
4845 * occurred.
4846 */
4847 if (vcpu->arch.gpa_available &&
4848 emulator_can_use_gpa(ctxt) &&
4849 (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
4850 gpa = vcpu->arch.gpa_val;
4851 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
4852 } else {
4853 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
4854 if (ret < 0)
4855 return X86EMUL_PROPAGATE_FAULT;
4856 }
4857
4858 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
4859 return X86EMUL_CONTINUE;
4860
4861 /*
4862 * Is this MMIO handled locally?
4863 */
4864 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
4865 if (handled == bytes)
4866 return X86EMUL_CONTINUE;
4867
4868 gpa += handled;
4869 bytes -= handled;
4870 val += handled;
4871
4872 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
4873 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
4874 frag->gpa = gpa;
4875 frag->data = val;
4876 frag->len = bytes;
4877 return X86EMUL_CONTINUE;
4878 }
4879
4880 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
4881 unsigned long addr,
4882 void *val, unsigned int bytes,
4883 struct x86_exception *exception,
4884 const struct read_write_emulator_ops *ops)
4885 {
4886 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4887 gpa_t gpa;
4888 int rc;
4889
4890 if (ops->read_write_prepare &&
4891 ops->read_write_prepare(vcpu, val, bytes))
4892 return X86EMUL_CONTINUE;
4893
4894 vcpu->mmio_nr_fragments = 0;
4895
4896 /* Crossing a page boundary? */
4897 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
4898 int now;
4899
4900 now = -addr & ~PAGE_MASK;
4901 rc = emulator_read_write_onepage(addr, val, now, exception,
4902 vcpu, ops);
4903
4904 if (rc != X86EMUL_CONTINUE)
4905 return rc;
4906 addr += now;
4907 if (ctxt->mode != X86EMUL_MODE_PROT64)
4908 addr = (u32)addr;
4909 val += now;
4910 bytes -= now;
4911 }
4912
4913 rc = emulator_read_write_onepage(addr, val, bytes, exception,
4914 vcpu, ops);
4915 if (rc != X86EMUL_CONTINUE)
4916 return rc;
4917
4918 if (!vcpu->mmio_nr_fragments)
4919 return rc;
4920
4921 gpa = vcpu->mmio_fragments[0].gpa;
4922
4923 vcpu->mmio_needed = 1;
4924 vcpu->mmio_cur_fragment = 0;
4925
4926 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
4927 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
4928 vcpu->run->exit_reason = KVM_EXIT_MMIO;
4929 vcpu->run->mmio.phys_addr = gpa;
4930
4931 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
4932 }
4933
4934 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
4935 unsigned long addr,
4936 void *val,
4937 unsigned int bytes,
4938 struct x86_exception *exception)
4939 {
4940 return emulator_read_write(ctxt, addr, val, bytes,
4941 exception, &read_emultor);
4942 }
4943
4944 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
4945 unsigned long addr,
4946 const void *val,
4947 unsigned int bytes,
4948 struct x86_exception *exception)
4949 {
4950 return emulator_read_write(ctxt, addr, (void *)val, bytes,
4951 exception, &write_emultor);
4952 }
4953
4954 #define CMPXCHG_TYPE(t, ptr, old, new) \
4955 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4956
4957 #ifdef CONFIG_X86_64
4958 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4959 #else
4960 # define CMPXCHG64(ptr, old, new) \
4961 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4962 #endif
4963
4964 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4965 unsigned long addr,
4966 const void *old,
4967 const void *new,
4968 unsigned int bytes,
4969 struct x86_exception *exception)
4970 {
4971 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4972 gpa_t gpa;
4973 struct page *page;
4974 char *kaddr;
4975 bool exchanged;
4976
4977 /* guests cmpxchg8b have to be emulated atomically */
4978 if (bytes > 8 || (bytes & (bytes - 1)))
4979 goto emul_write;
4980
4981 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4982
4983 if (gpa == UNMAPPED_GVA ||
4984 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4985 goto emul_write;
4986
4987 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4988 goto emul_write;
4989
4990 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
4991 if (is_error_page(page))
4992 goto emul_write;
4993
4994 kaddr = kmap_atomic(page);
4995 kaddr += offset_in_page(gpa);
4996 switch (bytes) {
4997 case 1:
4998 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4999 break;
5000 case 2:
5001 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5002 break;
5003 case 4:
5004 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5005 break;
5006 case 8:
5007 exchanged = CMPXCHG64(kaddr, old, new);
5008 break;
5009 default:
5010 BUG();
5011 }
5012 kunmap_atomic(kaddr);
5013 kvm_release_page_dirty(page);
5014
5015 if (!exchanged)
5016 return X86EMUL_CMPXCHG_FAILED;
5017
5018 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5019 kvm_page_track_write(vcpu, gpa, new, bytes);
5020
5021 return X86EMUL_CONTINUE;
5022
5023 emul_write:
5024 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5025
5026 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5027 }
5028
5029 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5030 {
5031 int r = 0, i;
5032
5033 for (i = 0; i < vcpu->arch.pio.count; i++) {
5034 if (vcpu->arch.pio.in)
5035 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5036 vcpu->arch.pio.size, pd);
5037 else
5038 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5039 vcpu->arch.pio.port, vcpu->arch.pio.size,
5040 pd);
5041 if (r)
5042 break;
5043 pd += vcpu->arch.pio.size;
5044 }
5045 return r;
5046 }
5047
5048 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5049 unsigned short port, void *val,
5050 unsigned int count, bool in)
5051 {
5052 vcpu->arch.pio.port = port;
5053 vcpu->arch.pio.in = in;
5054 vcpu->arch.pio.count = count;
5055 vcpu->arch.pio.size = size;
5056
5057 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5058 vcpu->arch.pio.count = 0;
5059 return 1;
5060 }
5061
5062 vcpu->run->exit_reason = KVM_EXIT_IO;
5063 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5064 vcpu->run->io.size = size;
5065 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5066 vcpu->run->io.count = count;
5067 vcpu->run->io.port = port;
5068
5069 return 0;
5070 }
5071
5072 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5073 int size, unsigned short port, void *val,
5074 unsigned int count)
5075 {
5076 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5077 int ret;
5078
5079 if (vcpu->arch.pio.count)
5080 goto data_avail;
5081
5082 memset(vcpu->arch.pio_data, 0, size * count);
5083
5084 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5085 if (ret) {
5086 data_avail:
5087 memcpy(val, vcpu->arch.pio_data, size * count);
5088 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5089 vcpu->arch.pio.count = 0;
5090 return 1;
5091 }
5092
5093 return 0;
5094 }
5095
5096 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5097 int size, unsigned short port,
5098 const void *val, unsigned int count)
5099 {
5100 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5101
5102 memcpy(vcpu->arch.pio_data, val, size * count);
5103 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5104 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5105 }
5106
5107 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5108 {
5109 return kvm_x86_ops->get_segment_base(vcpu, seg);
5110 }
5111
5112 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5113 {
5114 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5115 }
5116
5117 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5118 {
5119 if (!need_emulate_wbinvd(vcpu))
5120 return X86EMUL_CONTINUE;
5121
5122 if (kvm_x86_ops->has_wbinvd_exit()) {
5123 int cpu = get_cpu();
5124
5125 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5126 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5127 wbinvd_ipi, NULL, 1);
5128 put_cpu();
5129 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5130 } else
5131 wbinvd();
5132 return X86EMUL_CONTINUE;
5133 }
5134
5135 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5136 {
5137 kvm_emulate_wbinvd_noskip(vcpu);
5138 return kvm_skip_emulated_instruction(vcpu);
5139 }
5140 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5141
5142
5143
5144 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5145 {
5146 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5147 }
5148
5149 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5150 unsigned long *dest)
5151 {
5152 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5153 }
5154
5155 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5156 unsigned long value)
5157 {
5158
5159 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5160 }
5161
5162 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5163 {
5164 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5165 }
5166
5167 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5168 {
5169 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5170 unsigned long value;
5171
5172 switch (cr) {
5173 case 0:
5174 value = kvm_read_cr0(vcpu);
5175 break;
5176 case 2:
5177 value = vcpu->arch.cr2;
5178 break;
5179 case 3:
5180 value = kvm_read_cr3(vcpu);
5181 break;
5182 case 4:
5183 value = kvm_read_cr4(vcpu);
5184 break;
5185 case 8:
5186 value = kvm_get_cr8(vcpu);
5187 break;
5188 default:
5189 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5190 return 0;
5191 }
5192
5193 return value;
5194 }
5195
5196 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5197 {
5198 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5199 int res = 0;
5200
5201 switch (cr) {
5202 case 0:
5203 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5204 break;
5205 case 2:
5206 vcpu->arch.cr2 = val;
5207 break;
5208 case 3:
5209 res = kvm_set_cr3(vcpu, val);
5210 break;
5211 case 4:
5212 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5213 break;
5214 case 8:
5215 res = kvm_set_cr8(vcpu, val);
5216 break;
5217 default:
5218 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5219 res = -1;
5220 }
5221
5222 return res;
5223 }
5224
5225 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5226 {
5227 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5228 }
5229
5230 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5231 {
5232 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5233 }
5234
5235 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5236 {
5237 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5238 }
5239
5240 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5241 {
5242 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5243 }
5244
5245 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5246 {
5247 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5248 }
5249
5250 static unsigned long emulator_get_cached_segment_base(
5251 struct x86_emulate_ctxt *ctxt, int seg)
5252 {
5253 return get_segment_base(emul_to_vcpu(ctxt), seg);
5254 }
5255
5256 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5257 struct desc_struct *desc, u32 *base3,
5258 int seg)
5259 {
5260 struct kvm_segment var;
5261
5262 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5263 *selector = var.selector;
5264
5265 if (var.unusable) {
5266 memset(desc, 0, sizeof(*desc));
5267 if (base3)
5268 *base3 = 0;
5269 return false;
5270 }
5271
5272 if (var.g)
5273 var.limit >>= 12;
5274 set_desc_limit(desc, var.limit);
5275 set_desc_base(desc, (unsigned long)var.base);
5276 #ifdef CONFIG_X86_64
5277 if (base3)
5278 *base3 = var.base >> 32;
5279 #endif
5280 desc->type = var.type;
5281 desc->s = var.s;
5282 desc->dpl = var.dpl;
5283 desc->p = var.present;
5284 desc->avl = var.avl;
5285 desc->l = var.l;
5286 desc->d = var.db;
5287 desc->g = var.g;
5288
5289 return true;
5290 }
5291
5292 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5293 struct desc_struct *desc, u32 base3,
5294 int seg)
5295 {
5296 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5297 struct kvm_segment var;
5298
5299 var.selector = selector;
5300 var.base = get_desc_base(desc);
5301 #ifdef CONFIG_X86_64
5302 var.base |= ((u64)base3) << 32;
5303 #endif
5304 var.limit = get_desc_limit(desc);
5305 if (desc->g)
5306 var.limit = (var.limit << 12) | 0xfff;
5307 var.type = desc->type;
5308 var.dpl = desc->dpl;
5309 var.db = desc->d;
5310 var.s = desc->s;
5311 var.l = desc->l;
5312 var.g = desc->g;
5313 var.avl = desc->avl;
5314 var.present = desc->p;
5315 var.unusable = !var.present;
5316 var.padding = 0;
5317
5318 kvm_set_segment(vcpu, &var, seg);
5319 return;
5320 }
5321
5322 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5323 u32 msr_index, u64 *pdata)
5324 {
5325 struct msr_data msr;
5326 int r;
5327
5328 msr.index = msr_index;
5329 msr.host_initiated = false;
5330 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5331 if (r)
5332 return r;
5333
5334 *pdata = msr.data;
5335 return 0;
5336 }
5337
5338 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5339 u32 msr_index, u64 data)
5340 {
5341 struct msr_data msr;
5342
5343 msr.data = data;
5344 msr.index = msr_index;
5345 msr.host_initiated = false;
5346 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5347 }
5348
5349 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5350 {
5351 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5352
5353 return vcpu->arch.smbase;
5354 }
5355
5356 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5357 {
5358 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5359
5360 vcpu->arch.smbase = smbase;
5361 }
5362
5363 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5364 u32 pmc)
5365 {
5366 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5367 }
5368
5369 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5370 u32 pmc, u64 *pdata)
5371 {
5372 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5373 }
5374
5375 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5376 {
5377 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5378 }
5379
5380 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5381 struct x86_instruction_info *info,
5382 enum x86_intercept_stage stage)
5383 {
5384 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5385 }
5386
5387 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5388 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
5389 {
5390 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
5391 }
5392
5393 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5394 {
5395 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5396 }
5397
5398 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5399 {
5400 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5401 }
5402
5403 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5404 {
5405 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5406 }
5407
5408 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5409 {
5410 return emul_to_vcpu(ctxt)->arch.hflags;
5411 }
5412
5413 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5414 {
5415 kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5416 }
5417
5418 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, u64 smbase)
5419 {
5420 return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smbase);
5421 }
5422
5423 static const struct x86_emulate_ops emulate_ops = {
5424 .read_gpr = emulator_read_gpr,
5425 .write_gpr = emulator_write_gpr,
5426 .read_std = emulator_read_std,
5427 .write_std = emulator_write_std,
5428 .read_phys = kvm_read_guest_phys_system,
5429 .fetch = kvm_fetch_guest_virt,
5430 .read_emulated = emulator_read_emulated,
5431 .write_emulated = emulator_write_emulated,
5432 .cmpxchg_emulated = emulator_cmpxchg_emulated,
5433 .invlpg = emulator_invlpg,
5434 .pio_in_emulated = emulator_pio_in_emulated,
5435 .pio_out_emulated = emulator_pio_out_emulated,
5436 .get_segment = emulator_get_segment,
5437 .set_segment = emulator_set_segment,
5438 .get_cached_segment_base = emulator_get_cached_segment_base,
5439 .get_gdt = emulator_get_gdt,
5440 .get_idt = emulator_get_idt,
5441 .set_gdt = emulator_set_gdt,
5442 .set_idt = emulator_set_idt,
5443 .get_cr = emulator_get_cr,
5444 .set_cr = emulator_set_cr,
5445 .cpl = emulator_get_cpl,
5446 .get_dr = emulator_get_dr,
5447 .set_dr = emulator_set_dr,
5448 .get_smbase = emulator_get_smbase,
5449 .set_smbase = emulator_set_smbase,
5450 .set_msr = emulator_set_msr,
5451 .get_msr = emulator_get_msr,
5452 .check_pmc = emulator_check_pmc,
5453 .read_pmc = emulator_read_pmc,
5454 .halt = emulator_halt,
5455 .wbinvd = emulator_wbinvd,
5456 .fix_hypercall = emulator_fix_hypercall,
5457 .intercept = emulator_intercept,
5458 .get_cpuid = emulator_get_cpuid,
5459 .set_nmi_mask = emulator_set_nmi_mask,
5460 .get_hflags = emulator_get_hflags,
5461 .set_hflags = emulator_set_hflags,
5462 .pre_leave_smm = emulator_pre_leave_smm,
5463 };
5464
5465 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5466 {
5467 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5468 /*
5469 * an sti; sti; sequence only disable interrupts for the first
5470 * instruction. So, if the last instruction, be it emulated or
5471 * not, left the system with the INT_STI flag enabled, it
5472 * means that the last instruction is an sti. We should not
5473 * leave the flag on in this case. The same goes for mov ss
5474 */
5475 if (int_shadow & mask)
5476 mask = 0;
5477 if (unlikely(int_shadow || mask)) {
5478 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5479 if (!mask)
5480 kvm_make_request(KVM_REQ_EVENT, vcpu);
5481 }
5482 }
5483
5484 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5485 {
5486 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5487 if (ctxt->exception.vector == PF_VECTOR)
5488 return kvm_propagate_fault(vcpu, &ctxt->exception);
5489
5490 if (ctxt->exception.error_code_valid)
5491 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5492 ctxt->exception.error_code);
5493 else
5494 kvm_queue_exception(vcpu, ctxt->exception.vector);
5495 return false;
5496 }
5497
5498 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5499 {
5500 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5501 int cs_db, cs_l;
5502
5503 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5504
5505 ctxt->eflags = kvm_get_rflags(vcpu);
5506 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
5507
5508 ctxt->eip = kvm_rip_read(vcpu);
5509 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
5510 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
5511 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
5512 cs_db ? X86EMUL_MODE_PROT32 :
5513 X86EMUL_MODE_PROT16;
5514 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
5515 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
5516 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
5517
5518 init_decode_cache(ctxt);
5519 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5520 }
5521
5522 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
5523 {
5524 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5525 int ret;
5526
5527 init_emulate_ctxt(vcpu);
5528
5529 ctxt->op_bytes = 2;
5530 ctxt->ad_bytes = 2;
5531 ctxt->_eip = ctxt->eip + inc_eip;
5532 ret = emulate_int_real(ctxt, irq);
5533
5534 if (ret != X86EMUL_CONTINUE)
5535 return EMULATE_FAIL;
5536
5537 ctxt->eip = ctxt->_eip;
5538 kvm_rip_write(vcpu, ctxt->eip);
5539 kvm_set_rflags(vcpu, ctxt->eflags);
5540
5541 if (irq == NMI_VECTOR)
5542 vcpu->arch.nmi_pending = 0;
5543 else
5544 vcpu->arch.interrupt.pending = false;
5545
5546 return EMULATE_DONE;
5547 }
5548 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
5549
5550 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
5551 {
5552 int r = EMULATE_DONE;
5553
5554 ++vcpu->stat.insn_emulation_fail;
5555 trace_kvm_emulate_insn_failed(vcpu);
5556 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
5557 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5558 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5559 vcpu->run->internal.ndata = 0;
5560 r = EMULATE_USER_EXIT;
5561 }
5562 kvm_queue_exception(vcpu, UD_VECTOR);
5563
5564 return r;
5565 }
5566
5567 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
5568 bool write_fault_to_shadow_pgtable,
5569 int emulation_type)
5570 {
5571 gpa_t gpa = cr2;
5572 kvm_pfn_t pfn;
5573
5574 if (emulation_type & EMULTYPE_NO_REEXECUTE)
5575 return false;
5576
5577 if (!vcpu->arch.mmu.direct_map) {
5578 /*
5579 * Write permission should be allowed since only
5580 * write access need to be emulated.
5581 */
5582 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5583
5584 /*
5585 * If the mapping is invalid in guest, let cpu retry
5586 * it to generate fault.
5587 */
5588 if (gpa == UNMAPPED_GVA)
5589 return true;
5590 }
5591
5592 /*
5593 * Do not retry the unhandleable instruction if it faults on the
5594 * readonly host memory, otherwise it will goto a infinite loop:
5595 * retry instruction -> write #PF -> emulation fail -> retry
5596 * instruction -> ...
5597 */
5598 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
5599
5600 /*
5601 * If the instruction failed on the error pfn, it can not be fixed,
5602 * report the error to userspace.
5603 */
5604 if (is_error_noslot_pfn(pfn))
5605 return false;
5606
5607 kvm_release_pfn_clean(pfn);
5608
5609 /* The instructions are well-emulated on direct mmu. */
5610 if (vcpu->arch.mmu.direct_map) {
5611 unsigned int indirect_shadow_pages;
5612
5613 spin_lock(&vcpu->kvm->mmu_lock);
5614 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
5615 spin_unlock(&vcpu->kvm->mmu_lock);
5616
5617 if (indirect_shadow_pages)
5618 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5619
5620 return true;
5621 }
5622
5623 /*
5624 * if emulation was due to access to shadowed page table
5625 * and it failed try to unshadow page and re-enter the
5626 * guest to let CPU execute the instruction.
5627 */
5628 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5629
5630 /*
5631 * If the access faults on its page table, it can not
5632 * be fixed by unprotecting shadow page and it should
5633 * be reported to userspace.
5634 */
5635 return !write_fault_to_shadow_pgtable;
5636 }
5637
5638 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
5639 unsigned long cr2, int emulation_type)
5640 {
5641 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5642 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
5643
5644 last_retry_eip = vcpu->arch.last_retry_eip;
5645 last_retry_addr = vcpu->arch.last_retry_addr;
5646
5647 /*
5648 * If the emulation is caused by #PF and it is non-page_table
5649 * writing instruction, it means the VM-EXIT is caused by shadow
5650 * page protected, we can zap the shadow page and retry this
5651 * instruction directly.
5652 *
5653 * Note: if the guest uses a non-page-table modifying instruction
5654 * on the PDE that points to the instruction, then we will unmap
5655 * the instruction and go to an infinite loop. So, we cache the
5656 * last retried eip and the last fault address, if we meet the eip
5657 * and the address again, we can break out of the potential infinite
5658 * loop.
5659 */
5660 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
5661
5662 if (!(emulation_type & EMULTYPE_RETRY))
5663 return false;
5664
5665 if (x86_page_table_writing_insn(ctxt))
5666 return false;
5667
5668 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
5669 return false;
5670
5671 vcpu->arch.last_retry_eip = ctxt->eip;
5672 vcpu->arch.last_retry_addr = cr2;
5673
5674 if (!vcpu->arch.mmu.direct_map)
5675 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
5676
5677 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
5678
5679 return true;
5680 }
5681
5682 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
5683 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
5684
5685 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
5686 {
5687 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
5688 /* This is a good place to trace that we are exiting SMM. */
5689 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
5690
5691 /* Process a latched INIT or SMI, if any. */
5692 kvm_make_request(KVM_REQ_EVENT, vcpu);
5693 }
5694
5695 kvm_mmu_reset_context(vcpu);
5696 }
5697
5698 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
5699 {
5700 unsigned changed = vcpu->arch.hflags ^ emul_flags;
5701
5702 vcpu->arch.hflags = emul_flags;
5703
5704 if (changed & HF_SMM_MASK)
5705 kvm_smm_changed(vcpu);
5706 }
5707
5708 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
5709 unsigned long *db)
5710 {
5711 u32 dr6 = 0;
5712 int i;
5713 u32 enable, rwlen;
5714
5715 enable = dr7;
5716 rwlen = dr7 >> 16;
5717 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
5718 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
5719 dr6 |= (1 << i);
5720 return dr6;
5721 }
5722
5723 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
5724 {
5725 struct kvm_run *kvm_run = vcpu->run;
5726
5727 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5728 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
5729 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
5730 kvm_run->debug.arch.exception = DB_VECTOR;
5731 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5732 *r = EMULATE_USER_EXIT;
5733 } else {
5734 /*
5735 * "Certain debug exceptions may clear bit 0-3. The
5736 * remaining contents of the DR6 register are never
5737 * cleared by the processor".
5738 */
5739 vcpu->arch.dr6 &= ~15;
5740 vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
5741 kvm_queue_exception(vcpu, DB_VECTOR);
5742 }
5743 }
5744
5745 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
5746 {
5747 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5748 int r = EMULATE_DONE;
5749
5750 kvm_x86_ops->skip_emulated_instruction(vcpu);
5751
5752 /*
5753 * rflags is the old, "raw" value of the flags. The new value has
5754 * not been saved yet.
5755 *
5756 * This is correct even for TF set by the guest, because "the
5757 * processor will not generate this exception after the instruction
5758 * that sets the TF flag".
5759 */
5760 if (unlikely(rflags & X86_EFLAGS_TF))
5761 kvm_vcpu_do_singlestep(vcpu, &r);
5762 return r == EMULATE_DONE;
5763 }
5764 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
5765
5766 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
5767 {
5768 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
5769 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
5770 struct kvm_run *kvm_run = vcpu->run;
5771 unsigned long eip = kvm_get_linear_rip(vcpu);
5772 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5773 vcpu->arch.guest_debug_dr7,
5774 vcpu->arch.eff_db);
5775
5776 if (dr6 != 0) {
5777 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
5778 kvm_run->debug.arch.pc = eip;
5779 kvm_run->debug.arch.exception = DB_VECTOR;
5780 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5781 *r = EMULATE_USER_EXIT;
5782 return true;
5783 }
5784 }
5785
5786 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
5787 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
5788 unsigned long eip = kvm_get_linear_rip(vcpu);
5789 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
5790 vcpu->arch.dr7,
5791 vcpu->arch.db);
5792
5793 if (dr6 != 0) {
5794 vcpu->arch.dr6 &= ~15;
5795 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5796 kvm_queue_exception(vcpu, DB_VECTOR);
5797 *r = EMULATE_DONE;
5798 return true;
5799 }
5800 }
5801
5802 return false;
5803 }
5804
5805 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
5806 unsigned long cr2,
5807 int emulation_type,
5808 void *insn,
5809 int insn_len)
5810 {
5811 int r;
5812 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5813 bool writeback = true;
5814 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
5815
5816 vcpu->arch.l1tf_flush_l1d = true;
5817
5818 /*
5819 * Clear write_fault_to_shadow_pgtable here to ensure it is
5820 * never reused.
5821 */
5822 vcpu->arch.write_fault_to_shadow_pgtable = false;
5823 kvm_clear_exception_queue(vcpu);
5824
5825 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
5826 init_emulate_ctxt(vcpu);
5827
5828 /*
5829 * We will reenter on the same instruction since
5830 * we do not set complete_userspace_io. This does not
5831 * handle watchpoints yet, those would be handled in
5832 * the emulate_ops.
5833 */
5834 if (!(emulation_type & EMULTYPE_SKIP) &&
5835 kvm_vcpu_check_breakpoint(vcpu, &r))
5836 return r;
5837
5838 ctxt->interruptibility = 0;
5839 ctxt->have_exception = false;
5840 ctxt->exception.vector = -1;
5841 ctxt->perm_ok = false;
5842
5843 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
5844
5845 r = x86_decode_insn(ctxt, insn, insn_len);
5846
5847 trace_kvm_emulate_insn_start(vcpu);
5848 ++vcpu->stat.insn_emulation;
5849 if (r != EMULATION_OK) {
5850 if (emulation_type & EMULTYPE_TRAP_UD)
5851 return EMULATE_FAIL;
5852 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5853 emulation_type))
5854 return EMULATE_DONE;
5855 if (ctxt->have_exception && inject_emulated_exception(vcpu))
5856 return EMULATE_DONE;
5857 if (emulation_type & EMULTYPE_SKIP)
5858 return EMULATE_FAIL;
5859 return handle_emulation_failure(vcpu);
5860 }
5861 }
5862
5863 if (emulation_type & EMULTYPE_SKIP) {
5864 kvm_rip_write(vcpu, ctxt->_eip);
5865 if (ctxt->eflags & X86_EFLAGS_RF)
5866 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
5867 return EMULATE_DONE;
5868 }
5869
5870 if (retry_instruction(ctxt, cr2, emulation_type))
5871 return EMULATE_DONE;
5872
5873 /* this is needed for vmware backdoor interface to work since it
5874 changes registers values during IO operation */
5875 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
5876 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
5877 emulator_invalidate_register_cache(ctxt);
5878 }
5879
5880 restart:
5881 /* Save the faulting GPA (cr2) in the address field */
5882 ctxt->exception.address = cr2;
5883
5884 r = x86_emulate_insn(ctxt);
5885
5886 if (r == EMULATION_INTERCEPTED)
5887 return EMULATE_DONE;
5888
5889 if (r == EMULATION_FAILED) {
5890 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
5891 emulation_type))
5892 return EMULATE_DONE;
5893
5894 return handle_emulation_failure(vcpu);
5895 }
5896
5897 if (ctxt->have_exception) {
5898 r = EMULATE_DONE;
5899 if (inject_emulated_exception(vcpu))
5900 return r;
5901 } else if (vcpu->arch.pio.count) {
5902 if (!vcpu->arch.pio.in) {
5903 /* FIXME: return into emulator if single-stepping. */
5904 vcpu->arch.pio.count = 0;
5905 } else {
5906 writeback = false;
5907 vcpu->arch.complete_userspace_io = complete_emulated_pio;
5908 }
5909 r = EMULATE_USER_EXIT;
5910 } else if (vcpu->mmio_needed) {
5911 if (!vcpu->mmio_is_write)
5912 writeback = false;
5913 r = EMULATE_USER_EXIT;
5914 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
5915 } else if (r == EMULATION_RESTART)
5916 goto restart;
5917 else
5918 r = EMULATE_DONE;
5919
5920 if (writeback) {
5921 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
5922 toggle_interruptibility(vcpu, ctxt->interruptibility);
5923 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5924 kvm_rip_write(vcpu, ctxt->eip);
5925 if (r == EMULATE_DONE &&
5926 (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
5927 kvm_vcpu_do_singlestep(vcpu, &r);
5928 if (!ctxt->have_exception ||
5929 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
5930 __kvm_set_rflags(vcpu, ctxt->eflags);
5931
5932 /*
5933 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
5934 * do nothing, and it will be requested again as soon as
5935 * the shadow expires. But we still need to check here,
5936 * because POPF has no interrupt shadow.
5937 */
5938 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
5939 kvm_make_request(KVM_REQ_EVENT, vcpu);
5940 } else
5941 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
5942
5943 return r;
5944 }
5945 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
5946
5947 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
5948 {
5949 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
5950 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
5951 size, port, &val, 1);
5952 /* do not return to emulator after return from userspace */
5953 vcpu->arch.pio.count = 0;
5954 return ret;
5955 }
5956 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
5957
5958 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
5959 {
5960 unsigned long val;
5961
5962 /* We should only ever be called with arch.pio.count equal to 1 */
5963 BUG_ON(vcpu->arch.pio.count != 1);
5964
5965 /* For size less than 4 we merge, else we zero extend */
5966 val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
5967 : 0;
5968
5969 /*
5970 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
5971 * the copy and tracing
5972 */
5973 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
5974 vcpu->arch.pio.port, &val, 1);
5975 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5976
5977 return 1;
5978 }
5979
5980 int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size, unsigned short port)
5981 {
5982 unsigned long val;
5983 int ret;
5984
5985 /* For size less than 4 we merge, else we zero extend */
5986 val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
5987
5988 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
5989 &val, 1);
5990 if (ret) {
5991 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
5992 return ret;
5993 }
5994
5995 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
5996
5997 return 0;
5998 }
5999 EXPORT_SYMBOL_GPL(kvm_fast_pio_in);
6000
6001 static int kvmclock_cpu_down_prep(unsigned int cpu)
6002 {
6003 __this_cpu_write(cpu_tsc_khz, 0);
6004 return 0;
6005 }
6006
6007 static void tsc_khz_changed(void *data)
6008 {
6009 struct cpufreq_freqs *freq = data;
6010 unsigned long khz = 0;
6011
6012 if (data)
6013 khz = freq->new;
6014 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6015 khz = cpufreq_quick_get(raw_smp_processor_id());
6016 if (!khz)
6017 khz = tsc_khz;
6018 __this_cpu_write(cpu_tsc_khz, khz);
6019 }
6020
6021 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
6022 void *data)
6023 {
6024 struct cpufreq_freqs *freq = data;
6025 struct kvm *kvm;
6026 struct kvm_vcpu *vcpu;
6027 int i, send_ipi = 0;
6028
6029 /*
6030 * We allow guests to temporarily run on slowing clocks,
6031 * provided we notify them after, or to run on accelerating
6032 * clocks, provided we notify them before. Thus time never
6033 * goes backwards.
6034 *
6035 * However, we have a problem. We can't atomically update
6036 * the frequency of a given CPU from this function; it is
6037 * merely a notifier, which can be called from any CPU.
6038 * Changing the TSC frequency at arbitrary points in time
6039 * requires a recomputation of local variables related to
6040 * the TSC for each VCPU. We must flag these local variables
6041 * to be updated and be sure the update takes place with the
6042 * new frequency before any guests proceed.
6043 *
6044 * Unfortunately, the combination of hotplug CPU and frequency
6045 * change creates an intractable locking scenario; the order
6046 * of when these callouts happen is undefined with respect to
6047 * CPU hotplug, and they can race with each other. As such,
6048 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6049 * undefined; you can actually have a CPU frequency change take
6050 * place in between the computation of X and the setting of the
6051 * variable. To protect against this problem, all updates of
6052 * the per_cpu tsc_khz variable are done in an interrupt
6053 * protected IPI, and all callers wishing to update the value
6054 * must wait for a synchronous IPI to complete (which is trivial
6055 * if the caller is on the CPU already). This establishes the
6056 * necessary total order on variable updates.
6057 *
6058 * Note that because a guest time update may take place
6059 * anytime after the setting of the VCPU's request bit, the
6060 * correct TSC value must be set before the request. However,
6061 * to ensure the update actually makes it to any guest which
6062 * starts running in hardware virtualization between the set
6063 * and the acquisition of the spinlock, we must also ping the
6064 * CPU after setting the request bit.
6065 *
6066 */
6067
6068 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
6069 return 0;
6070 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
6071 return 0;
6072
6073 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6074
6075 spin_lock(&kvm_lock);
6076 list_for_each_entry(kvm, &vm_list, vm_list) {
6077 kvm_for_each_vcpu(i, vcpu, kvm) {
6078 if (vcpu->cpu != freq->cpu)
6079 continue;
6080 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6081 if (vcpu->cpu != smp_processor_id())
6082 send_ipi = 1;
6083 }
6084 }
6085 spin_unlock(&kvm_lock);
6086
6087 if (freq->old < freq->new && send_ipi) {
6088 /*
6089 * We upscale the frequency. Must make the guest
6090 * doesn't see old kvmclock values while running with
6091 * the new frequency, otherwise we risk the guest sees
6092 * time go backwards.
6093 *
6094 * In case we update the frequency for another cpu
6095 * (which might be in guest context) send an interrupt
6096 * to kick the cpu out of guest context. Next time
6097 * guest context is entered kvmclock will be updated,
6098 * so the guest will not see stale values.
6099 */
6100 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6101 }
6102 return 0;
6103 }
6104
6105 static struct notifier_block kvmclock_cpufreq_notifier_block = {
6106 .notifier_call = kvmclock_cpufreq_notifier
6107 };
6108
6109 static int kvmclock_cpu_online(unsigned int cpu)
6110 {
6111 tsc_khz_changed(NULL);
6112 return 0;
6113 }
6114
6115 static void kvm_timer_init(void)
6116 {
6117 max_tsc_khz = tsc_khz;
6118
6119 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
6120 #ifdef CONFIG_CPU_FREQ
6121 struct cpufreq_policy policy;
6122 int cpu;
6123
6124 memset(&policy, 0, sizeof(policy));
6125 cpu = get_cpu();
6126 cpufreq_get_policy(&policy, cpu);
6127 if (policy.cpuinfo.max_freq)
6128 max_tsc_khz = policy.cpuinfo.max_freq;
6129 put_cpu();
6130 #endif
6131 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6132 CPUFREQ_TRANSITION_NOTIFIER);
6133 }
6134 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
6135
6136 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
6137 kvmclock_cpu_online, kvmclock_cpu_down_prep);
6138 }
6139
6140 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6141
6142 int kvm_is_in_guest(void)
6143 {
6144 return __this_cpu_read(current_vcpu) != NULL;
6145 }
6146
6147 static int kvm_is_user_mode(void)
6148 {
6149 int user_mode = 3;
6150
6151 if (__this_cpu_read(current_vcpu))
6152 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
6153
6154 return user_mode != 0;
6155 }
6156
6157 static unsigned long kvm_get_guest_ip(void)
6158 {
6159 unsigned long ip = 0;
6160
6161 if (__this_cpu_read(current_vcpu))
6162 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
6163
6164 return ip;
6165 }
6166
6167 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6168 .is_in_guest = kvm_is_in_guest,
6169 .is_user_mode = kvm_is_user_mode,
6170 .get_guest_ip = kvm_get_guest_ip,
6171 };
6172
6173 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
6174 {
6175 __this_cpu_write(current_vcpu, vcpu);
6176 }
6177 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
6178
6179 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
6180 {
6181 __this_cpu_write(current_vcpu, NULL);
6182 }
6183 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
6184
6185 static void kvm_set_mmio_spte_mask(void)
6186 {
6187 u64 mask;
6188 int maxphyaddr = boot_cpu_data.x86_phys_bits;
6189
6190 /*
6191 * Set the reserved bits and the present bit of an paging-structure
6192 * entry to generate page fault with PFER.RSV = 1.
6193 */
6194 /* Mask the reserved physical address bits. */
6195 mask = rsvd_bits(maxphyaddr, 51);
6196
6197 /* Set the present bit. */
6198 mask |= 1ull;
6199
6200 #ifdef CONFIG_X86_64
6201 /*
6202 * If reserved bit is not supported, clear the present bit to disable
6203 * mmio page fault.
6204 */
6205 if (maxphyaddr == 52)
6206 mask &= ~1ull;
6207 #endif
6208
6209 kvm_mmu_set_mmio_spte_mask(mask, mask);
6210 }
6211
6212 #ifdef CONFIG_X86_64
6213 static void pvclock_gtod_update_fn(struct work_struct *work)
6214 {
6215 struct kvm *kvm;
6216
6217 struct kvm_vcpu *vcpu;
6218 int i;
6219
6220 spin_lock(&kvm_lock);
6221 list_for_each_entry(kvm, &vm_list, vm_list)
6222 kvm_for_each_vcpu(i, vcpu, kvm)
6223 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6224 atomic_set(&kvm_guest_has_master_clock, 0);
6225 spin_unlock(&kvm_lock);
6226 }
6227
6228 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6229
6230 /*
6231 * Notification about pvclock gtod data update.
6232 */
6233 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6234 void *priv)
6235 {
6236 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6237 struct timekeeper *tk = priv;
6238
6239 update_pvclock_gtod(tk);
6240
6241 /* disable master clock if host does not trust, or does not
6242 * use, TSC clocksource
6243 */
6244 if (gtod->clock.vclock_mode != VCLOCK_TSC &&
6245 atomic_read(&kvm_guest_has_master_clock) != 0)
6246 queue_work(system_long_wq, &pvclock_gtod_work);
6247
6248 return 0;
6249 }
6250
6251 static struct notifier_block pvclock_gtod_notifier = {
6252 .notifier_call = pvclock_gtod_notify,
6253 };
6254 #endif
6255
6256 int kvm_arch_init(void *opaque)
6257 {
6258 int r;
6259 struct kvm_x86_ops *ops = opaque;
6260
6261 if (kvm_x86_ops) {
6262 printk(KERN_ERR "kvm: already loaded the other module\n");
6263 r = -EEXIST;
6264 goto out;
6265 }
6266
6267 if (!ops->cpu_has_kvm_support()) {
6268 printk(KERN_ERR "kvm: no hardware support\n");
6269 r = -EOPNOTSUPP;
6270 goto out;
6271 }
6272 if (ops->disabled_by_bios()) {
6273 printk(KERN_WARNING "kvm: disabled by bios\n");
6274 r = -EOPNOTSUPP;
6275 goto out;
6276 }
6277
6278 r = -ENOMEM;
6279 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6280 if (!shared_msrs) {
6281 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6282 goto out;
6283 }
6284
6285 r = kvm_mmu_module_init();
6286 if (r)
6287 goto out_free_percpu;
6288
6289 kvm_set_mmio_spte_mask();
6290
6291 kvm_x86_ops = ops;
6292
6293 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6294 PT_DIRTY_MASK, PT64_NX_MASK, 0,
6295 PT_PRESENT_MASK, 0, sme_me_mask);
6296 kvm_timer_init();
6297
6298 perf_register_guest_info_callbacks(&kvm_guest_cbs);
6299
6300 if (boot_cpu_has(X86_FEATURE_XSAVE))
6301 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6302
6303 kvm_lapic_init();
6304 #ifdef CONFIG_X86_64
6305 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6306 #endif
6307
6308 return 0;
6309
6310 out_free_percpu:
6311 free_percpu(shared_msrs);
6312 out:
6313 return r;
6314 }
6315
6316 void kvm_arch_exit(void)
6317 {
6318 kvm_lapic_exit();
6319 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6320
6321 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6322 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6323 CPUFREQ_TRANSITION_NOTIFIER);
6324 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6325 #ifdef CONFIG_X86_64
6326 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6327 #endif
6328 kvm_x86_ops = NULL;
6329 kvm_mmu_module_exit();
6330 free_percpu(shared_msrs);
6331 }
6332
6333 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6334 {
6335 ++vcpu->stat.halt_exits;
6336 if (lapic_in_kernel(vcpu)) {
6337 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6338 return 1;
6339 } else {
6340 vcpu->run->exit_reason = KVM_EXIT_HLT;
6341 return 0;
6342 }
6343 }
6344 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6345
6346 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6347 {
6348 int ret = kvm_skip_emulated_instruction(vcpu);
6349 /*
6350 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6351 * KVM_EXIT_DEBUG here.
6352 */
6353 return kvm_vcpu_halt(vcpu) && ret;
6354 }
6355 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6356
6357 #ifdef CONFIG_X86_64
6358 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6359 unsigned long clock_type)
6360 {
6361 struct kvm_clock_pairing clock_pairing;
6362 struct timespec ts;
6363 u64 cycle;
6364 int ret;
6365
6366 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6367 return -KVM_EOPNOTSUPP;
6368
6369 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6370 return -KVM_EOPNOTSUPP;
6371
6372 clock_pairing.sec = ts.tv_sec;
6373 clock_pairing.nsec = ts.tv_nsec;
6374 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6375 clock_pairing.flags = 0;
6376
6377 ret = 0;
6378 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6379 sizeof(struct kvm_clock_pairing)))
6380 ret = -KVM_EFAULT;
6381
6382 return ret;
6383 }
6384 #endif
6385
6386 /*
6387 * kvm_pv_kick_cpu_op: Kick a vcpu.
6388 *
6389 * @apicid - apicid of vcpu to be kicked.
6390 */
6391 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
6392 {
6393 struct kvm_lapic_irq lapic_irq;
6394
6395 lapic_irq.shorthand = 0;
6396 lapic_irq.dest_mode = 0;
6397 lapic_irq.level = 0;
6398 lapic_irq.dest_id = apicid;
6399 lapic_irq.msi_redir_hint = false;
6400
6401 lapic_irq.delivery_mode = APIC_DM_REMRD;
6402 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
6403 }
6404
6405 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
6406 {
6407 vcpu->arch.apicv_active = false;
6408 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
6409 }
6410
6411 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
6412 {
6413 unsigned long nr, a0, a1, a2, a3, ret;
6414 int op_64_bit, r;
6415
6416 r = kvm_skip_emulated_instruction(vcpu);
6417
6418 if (kvm_hv_hypercall_enabled(vcpu->kvm))
6419 return kvm_hv_hypercall(vcpu);
6420
6421 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
6422 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
6423 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
6424 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
6425 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
6426
6427 trace_kvm_hypercall(nr, a0, a1, a2, a3);
6428
6429 op_64_bit = is_64_bit_mode(vcpu);
6430 if (!op_64_bit) {
6431 nr &= 0xFFFFFFFF;
6432 a0 &= 0xFFFFFFFF;
6433 a1 &= 0xFFFFFFFF;
6434 a2 &= 0xFFFFFFFF;
6435 a3 &= 0xFFFFFFFF;
6436 }
6437
6438 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
6439 ret = -KVM_EPERM;
6440 goto out;
6441 }
6442
6443 switch (nr) {
6444 case KVM_HC_VAPIC_POLL_IRQ:
6445 ret = 0;
6446 break;
6447 case KVM_HC_KICK_CPU:
6448 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
6449 ret = 0;
6450 break;
6451 #ifdef CONFIG_X86_64
6452 case KVM_HC_CLOCK_PAIRING:
6453 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
6454 break;
6455 #endif
6456 default:
6457 ret = -KVM_ENOSYS;
6458 break;
6459 }
6460 out:
6461 if (!op_64_bit)
6462 ret = (u32)ret;
6463 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
6464 ++vcpu->stat.hypercalls;
6465 return r;
6466 }
6467 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
6468
6469 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
6470 {
6471 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6472 char instruction[3];
6473 unsigned long rip = kvm_rip_read(vcpu);
6474
6475 kvm_x86_ops->patch_hypercall(vcpu, instruction);
6476
6477 return emulator_write_emulated(ctxt, rip, instruction, 3,
6478 &ctxt->exception);
6479 }
6480
6481 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
6482 {
6483 return vcpu->run->request_interrupt_window &&
6484 likely(!pic_in_kernel(vcpu->kvm));
6485 }
6486
6487 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
6488 {
6489 struct kvm_run *kvm_run = vcpu->run;
6490
6491 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
6492 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
6493 kvm_run->cr8 = kvm_get_cr8(vcpu);
6494 kvm_run->apic_base = kvm_get_apic_base(vcpu);
6495 kvm_run->ready_for_interrupt_injection =
6496 pic_in_kernel(vcpu->kvm) ||
6497 kvm_vcpu_ready_for_interrupt_injection(vcpu);
6498 }
6499
6500 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
6501 {
6502 int max_irr, tpr;
6503
6504 if (!kvm_x86_ops->update_cr8_intercept)
6505 return;
6506
6507 if (!lapic_in_kernel(vcpu))
6508 return;
6509
6510 if (vcpu->arch.apicv_active)
6511 return;
6512
6513 if (!vcpu->arch.apic->vapic_addr)
6514 max_irr = kvm_lapic_find_highest_irr(vcpu);
6515 else
6516 max_irr = -1;
6517
6518 if (max_irr != -1)
6519 max_irr >>= 4;
6520
6521 tpr = kvm_lapic_get_cr8(vcpu);
6522
6523 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
6524 }
6525
6526 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
6527 {
6528 int r;
6529
6530 /* try to reinject previous events if any */
6531 if (vcpu->arch.exception.injected) {
6532 kvm_x86_ops->queue_exception(vcpu);
6533 return 0;
6534 }
6535
6536 /*
6537 * Exceptions must be injected immediately, or the exception
6538 * frame will have the address of the NMI or interrupt handler.
6539 */
6540 if (!vcpu->arch.exception.pending) {
6541 if (vcpu->arch.nmi_injected) {
6542 kvm_x86_ops->set_nmi(vcpu);
6543 return 0;
6544 }
6545
6546 if (vcpu->arch.interrupt.pending) {
6547 kvm_x86_ops->set_irq(vcpu);
6548 return 0;
6549 }
6550 }
6551
6552 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6553 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6554 if (r != 0)
6555 return r;
6556 }
6557
6558 /* try to inject new event if pending */
6559 if (vcpu->arch.exception.pending) {
6560 trace_kvm_inj_exception(vcpu->arch.exception.nr,
6561 vcpu->arch.exception.has_error_code,
6562 vcpu->arch.exception.error_code);
6563
6564 vcpu->arch.exception.pending = false;
6565 vcpu->arch.exception.injected = true;
6566
6567 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
6568 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
6569 X86_EFLAGS_RF);
6570
6571 if (vcpu->arch.exception.nr == DB_VECTOR &&
6572 (vcpu->arch.dr7 & DR7_GD)) {
6573 vcpu->arch.dr7 &= ~DR7_GD;
6574 kvm_update_dr7(vcpu);
6575 }
6576
6577 kvm_x86_ops->queue_exception(vcpu);
6578 } else if (vcpu->arch.smi_pending && !is_smm(vcpu) && kvm_x86_ops->smi_allowed(vcpu)) {
6579 vcpu->arch.smi_pending = false;
6580 enter_smm(vcpu);
6581 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
6582 --vcpu->arch.nmi_pending;
6583 vcpu->arch.nmi_injected = true;
6584 kvm_x86_ops->set_nmi(vcpu);
6585 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
6586 /*
6587 * Because interrupts can be injected asynchronously, we are
6588 * calling check_nested_events again here to avoid a race condition.
6589 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
6590 * proposal and current concerns. Perhaps we should be setting
6591 * KVM_REQ_EVENT only on certain events and not unconditionally?
6592 */
6593 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
6594 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
6595 if (r != 0)
6596 return r;
6597 }
6598 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
6599 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
6600 false);
6601 kvm_x86_ops->set_irq(vcpu);
6602 }
6603 }
6604
6605 return 0;
6606 }
6607
6608 static void process_nmi(struct kvm_vcpu *vcpu)
6609 {
6610 unsigned limit = 2;
6611
6612 /*
6613 * x86 is limited to one NMI running, and one NMI pending after it.
6614 * If an NMI is already in progress, limit further NMIs to just one.
6615 * Otherwise, allow two (and we'll inject the first one immediately).
6616 */
6617 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
6618 limit = 1;
6619
6620 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
6621 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
6622 kvm_make_request(KVM_REQ_EVENT, vcpu);
6623 }
6624
6625 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
6626 {
6627 u32 flags = 0;
6628 flags |= seg->g << 23;
6629 flags |= seg->db << 22;
6630 flags |= seg->l << 21;
6631 flags |= seg->avl << 20;
6632 flags |= seg->present << 15;
6633 flags |= seg->dpl << 13;
6634 flags |= seg->s << 12;
6635 flags |= seg->type << 8;
6636 return flags;
6637 }
6638
6639 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
6640 {
6641 struct kvm_segment seg;
6642 int offset;
6643
6644 kvm_get_segment(vcpu, &seg, n);
6645 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
6646
6647 if (n < 3)
6648 offset = 0x7f84 + n * 12;
6649 else
6650 offset = 0x7f2c + (n - 3) * 12;
6651
6652 put_smstate(u32, buf, offset + 8, seg.base);
6653 put_smstate(u32, buf, offset + 4, seg.limit);
6654 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
6655 }
6656
6657 #ifdef CONFIG_X86_64
6658 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
6659 {
6660 struct kvm_segment seg;
6661 int offset;
6662 u16 flags;
6663
6664 kvm_get_segment(vcpu, &seg, n);
6665 offset = 0x7e00 + n * 16;
6666
6667 flags = enter_smm_get_segment_flags(&seg) >> 8;
6668 put_smstate(u16, buf, offset, seg.selector);
6669 put_smstate(u16, buf, offset + 2, flags);
6670 put_smstate(u32, buf, offset + 4, seg.limit);
6671 put_smstate(u64, buf, offset + 8, seg.base);
6672 }
6673 #endif
6674
6675 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
6676 {
6677 struct desc_ptr dt;
6678 struct kvm_segment seg;
6679 unsigned long val;
6680 int i;
6681
6682 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
6683 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
6684 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
6685 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
6686
6687 for (i = 0; i < 8; i++)
6688 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
6689
6690 kvm_get_dr(vcpu, 6, &val);
6691 put_smstate(u32, buf, 0x7fcc, (u32)val);
6692 kvm_get_dr(vcpu, 7, &val);
6693 put_smstate(u32, buf, 0x7fc8, (u32)val);
6694
6695 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6696 put_smstate(u32, buf, 0x7fc4, seg.selector);
6697 put_smstate(u32, buf, 0x7f64, seg.base);
6698 put_smstate(u32, buf, 0x7f60, seg.limit);
6699 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
6700
6701 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6702 put_smstate(u32, buf, 0x7fc0, seg.selector);
6703 put_smstate(u32, buf, 0x7f80, seg.base);
6704 put_smstate(u32, buf, 0x7f7c, seg.limit);
6705 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
6706
6707 kvm_x86_ops->get_gdt(vcpu, &dt);
6708 put_smstate(u32, buf, 0x7f74, dt.address);
6709 put_smstate(u32, buf, 0x7f70, dt.size);
6710
6711 kvm_x86_ops->get_idt(vcpu, &dt);
6712 put_smstate(u32, buf, 0x7f58, dt.address);
6713 put_smstate(u32, buf, 0x7f54, dt.size);
6714
6715 for (i = 0; i < 6; i++)
6716 enter_smm_save_seg_32(vcpu, buf, i);
6717
6718 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
6719
6720 /* revision id */
6721 put_smstate(u32, buf, 0x7efc, 0x00020000);
6722 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
6723 }
6724
6725 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
6726 {
6727 #ifdef CONFIG_X86_64
6728 struct desc_ptr dt;
6729 struct kvm_segment seg;
6730 unsigned long val;
6731 int i;
6732
6733 for (i = 0; i < 16; i++)
6734 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
6735
6736 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
6737 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
6738
6739 kvm_get_dr(vcpu, 6, &val);
6740 put_smstate(u64, buf, 0x7f68, val);
6741 kvm_get_dr(vcpu, 7, &val);
6742 put_smstate(u64, buf, 0x7f60, val);
6743
6744 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
6745 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
6746 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
6747
6748 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
6749
6750 /* revision id */
6751 put_smstate(u32, buf, 0x7efc, 0x00020064);
6752
6753 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
6754
6755 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
6756 put_smstate(u16, buf, 0x7e90, seg.selector);
6757 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
6758 put_smstate(u32, buf, 0x7e94, seg.limit);
6759 put_smstate(u64, buf, 0x7e98, seg.base);
6760
6761 kvm_x86_ops->get_idt(vcpu, &dt);
6762 put_smstate(u32, buf, 0x7e84, dt.size);
6763 put_smstate(u64, buf, 0x7e88, dt.address);
6764
6765 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
6766 put_smstate(u16, buf, 0x7e70, seg.selector);
6767 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
6768 put_smstate(u32, buf, 0x7e74, seg.limit);
6769 put_smstate(u64, buf, 0x7e78, seg.base);
6770
6771 kvm_x86_ops->get_gdt(vcpu, &dt);
6772 put_smstate(u32, buf, 0x7e64, dt.size);
6773 put_smstate(u64, buf, 0x7e68, dt.address);
6774
6775 for (i = 0; i < 6; i++)
6776 enter_smm_save_seg_64(vcpu, buf, i);
6777 #else
6778 WARN_ON_ONCE(1);
6779 #endif
6780 }
6781
6782 static void enter_smm(struct kvm_vcpu *vcpu)
6783 {
6784 struct kvm_segment cs, ds;
6785 struct desc_ptr dt;
6786 char buf[512];
6787 u32 cr0;
6788
6789 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
6790 memset(buf, 0, 512);
6791 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
6792 enter_smm_save_state_64(vcpu, buf);
6793 else
6794 enter_smm_save_state_32(vcpu, buf);
6795
6796 /*
6797 * Give pre_enter_smm() a chance to make ISA-specific changes to the
6798 * vCPU state (e.g. leave guest mode) after we've saved the state into
6799 * the SMM state-save area.
6800 */
6801 kvm_x86_ops->pre_enter_smm(vcpu, buf);
6802
6803 vcpu->arch.hflags |= HF_SMM_MASK;
6804 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
6805
6806 if (kvm_x86_ops->get_nmi_mask(vcpu))
6807 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
6808 else
6809 kvm_x86_ops->set_nmi_mask(vcpu, true);
6810
6811 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
6812 kvm_rip_write(vcpu, 0x8000);
6813
6814 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
6815 kvm_x86_ops->set_cr0(vcpu, cr0);
6816 vcpu->arch.cr0 = cr0;
6817
6818 kvm_x86_ops->set_cr4(vcpu, 0);
6819
6820 /* Undocumented: IDT limit is set to zero on entry to SMM. */
6821 dt.address = dt.size = 0;
6822 kvm_x86_ops->set_idt(vcpu, &dt);
6823
6824 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
6825
6826 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
6827 cs.base = vcpu->arch.smbase;
6828
6829 ds.selector = 0;
6830 ds.base = 0;
6831
6832 cs.limit = ds.limit = 0xffffffff;
6833 cs.type = ds.type = 0x3;
6834 cs.dpl = ds.dpl = 0;
6835 cs.db = ds.db = 0;
6836 cs.s = ds.s = 1;
6837 cs.l = ds.l = 0;
6838 cs.g = ds.g = 1;
6839 cs.avl = ds.avl = 0;
6840 cs.present = ds.present = 1;
6841 cs.unusable = ds.unusable = 0;
6842 cs.padding = ds.padding = 0;
6843
6844 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
6845 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
6846 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
6847 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
6848 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
6849 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
6850
6851 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
6852 kvm_x86_ops->set_efer(vcpu, 0);
6853
6854 kvm_update_cpuid(vcpu);
6855 kvm_mmu_reset_context(vcpu);
6856 }
6857
6858 static void process_smi(struct kvm_vcpu *vcpu)
6859 {
6860 vcpu->arch.smi_pending = true;
6861 kvm_make_request(KVM_REQ_EVENT, vcpu);
6862 }
6863
6864 void kvm_make_scan_ioapic_request(struct kvm *kvm)
6865 {
6866 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
6867 }
6868
6869 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
6870 {
6871 u64 eoi_exit_bitmap[4];
6872
6873 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
6874 return;
6875
6876 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
6877
6878 if (irqchip_split(vcpu->kvm))
6879 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
6880 else {
6881 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
6882 kvm_x86_ops->sync_pir_to_irr(vcpu);
6883 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
6884 }
6885 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
6886 vcpu_to_synic(vcpu)->vec_bitmap, 256);
6887 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
6888 }
6889
6890 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu)
6891 {
6892 ++vcpu->stat.tlb_flush;
6893 kvm_x86_ops->tlb_flush(vcpu);
6894 }
6895
6896 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
6897 unsigned long start, unsigned long end)
6898 {
6899 unsigned long apic_address;
6900
6901 /*
6902 * The physical address of apic access page is stored in the VMCS.
6903 * Update it when it becomes invalid.
6904 */
6905 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6906 if (start <= apic_address && apic_address < end)
6907 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
6908 }
6909
6910 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
6911 {
6912 struct page *page = NULL;
6913
6914 if (!lapic_in_kernel(vcpu))
6915 return;
6916
6917 if (!kvm_x86_ops->set_apic_access_page_addr)
6918 return;
6919
6920 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6921 if (is_error_page(page))
6922 return;
6923 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
6924
6925 /*
6926 * Do not pin apic access page in memory, the MMU notifier
6927 * will call us again if it is migrated or swapped out.
6928 */
6929 put_page(page);
6930 }
6931 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
6932
6933 /*
6934 * Returns 1 to let vcpu_run() continue the guest execution loop without
6935 * exiting to the userspace. Otherwise, the value will be returned to the
6936 * userspace.
6937 */
6938 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
6939 {
6940 int r;
6941 bool req_int_win =
6942 dm_request_for_irq_injection(vcpu) &&
6943 kvm_cpu_accept_dm_intr(vcpu);
6944
6945 bool req_immediate_exit = false;
6946
6947 if (kvm_request_pending(vcpu)) {
6948 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
6949 kvm_mmu_unload(vcpu);
6950 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
6951 __kvm_migrate_timers(vcpu);
6952 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
6953 kvm_gen_update_masterclock(vcpu->kvm);
6954 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
6955 kvm_gen_kvmclock_update(vcpu);
6956 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
6957 r = kvm_guest_time_update(vcpu);
6958 if (unlikely(r))
6959 goto out;
6960 }
6961 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
6962 kvm_mmu_sync_roots(vcpu);
6963 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
6964 kvm_vcpu_flush_tlb(vcpu);
6965 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
6966 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
6967 r = 0;
6968 goto out;
6969 }
6970 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
6971 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6972 vcpu->mmio_needed = 0;
6973 r = 0;
6974 goto out;
6975 }
6976 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
6977 /* Page is swapped out. Do synthetic halt */
6978 vcpu->arch.apf.halted = true;
6979 r = 1;
6980 goto out;
6981 }
6982 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
6983 record_steal_time(vcpu);
6984 if (kvm_check_request(KVM_REQ_SMI, vcpu))
6985 process_smi(vcpu);
6986 if (kvm_check_request(KVM_REQ_NMI, vcpu))
6987 process_nmi(vcpu);
6988 if (kvm_check_request(KVM_REQ_PMU, vcpu))
6989 kvm_pmu_handle_event(vcpu);
6990 if (kvm_check_request(KVM_REQ_PMI, vcpu))
6991 kvm_pmu_deliver_pmi(vcpu);
6992 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
6993 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
6994 if (test_bit(vcpu->arch.pending_ioapic_eoi,
6995 vcpu->arch.ioapic_handled_vectors)) {
6996 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
6997 vcpu->run->eoi.vector =
6998 vcpu->arch.pending_ioapic_eoi;
6999 r = 0;
7000 goto out;
7001 }
7002 }
7003 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
7004 vcpu_scan_ioapic(vcpu);
7005 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
7006 kvm_vcpu_reload_apic_access_page(vcpu);
7007 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
7008 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7009 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
7010 r = 0;
7011 goto out;
7012 }
7013 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
7014 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7015 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
7016 r = 0;
7017 goto out;
7018 }
7019 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
7020 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
7021 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
7022 r = 0;
7023 goto out;
7024 }
7025
7026 /*
7027 * KVM_REQ_HV_STIMER has to be processed after
7028 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
7029 * depend on the guest clock being up-to-date
7030 */
7031 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
7032 kvm_hv_process_stimers(vcpu);
7033 }
7034
7035 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
7036 ++vcpu->stat.req_event;
7037 kvm_apic_accept_events(vcpu);
7038 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
7039 r = 1;
7040 goto out;
7041 }
7042
7043 if (inject_pending_event(vcpu, req_int_win) != 0)
7044 req_immediate_exit = true;
7045 else {
7046 /* Enable SMI/NMI/IRQ window open exits if needed.
7047 *
7048 * SMIs have three cases:
7049 * 1) They can be nested, and then there is nothing to
7050 * do here because RSM will cause a vmexit anyway.
7051 * 2) There is an ISA-specific reason why SMI cannot be
7052 * injected, and the moment when this changes can be
7053 * intercepted.
7054 * 3) Or the SMI can be pending because
7055 * inject_pending_event has completed the injection
7056 * of an IRQ or NMI from the previous vmexit, and
7057 * then we request an immediate exit to inject the
7058 * SMI.
7059 */
7060 if (vcpu->arch.smi_pending && !is_smm(vcpu))
7061 if (!kvm_x86_ops->enable_smi_window(vcpu))
7062 req_immediate_exit = true;
7063 if (vcpu->arch.nmi_pending)
7064 kvm_x86_ops->enable_nmi_window(vcpu);
7065 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
7066 kvm_x86_ops->enable_irq_window(vcpu);
7067 WARN_ON(vcpu->arch.exception.pending);
7068 }
7069
7070 if (kvm_lapic_enabled(vcpu)) {
7071 update_cr8_intercept(vcpu);
7072 kvm_lapic_sync_to_vapic(vcpu);
7073 }
7074 }
7075
7076 r = kvm_mmu_reload(vcpu);
7077 if (unlikely(r)) {
7078 goto cancel_injection;
7079 }
7080
7081 preempt_disable();
7082
7083 kvm_x86_ops->prepare_guest_switch(vcpu);
7084
7085 /*
7086 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
7087 * IPI are then delayed after guest entry, which ensures that they
7088 * result in virtual interrupt delivery.
7089 */
7090 local_irq_disable();
7091 vcpu->mode = IN_GUEST_MODE;
7092
7093 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7094
7095 /*
7096 * 1) We should set ->mode before checking ->requests. Please see
7097 * the comment in kvm_vcpu_exiting_guest_mode().
7098 *
7099 * 2) For APICv, we should set ->mode before checking PIR.ON. This
7100 * pairs with the memory barrier implicit in pi_test_and_set_on
7101 * (see vmx_deliver_posted_interrupt).
7102 *
7103 * 3) This also orders the write to mode from any reads to the page
7104 * tables done while the VCPU is running. Please see the comment
7105 * in kvm_flush_remote_tlbs.
7106 */
7107 smp_mb__after_srcu_read_unlock();
7108
7109 /*
7110 * This handles the case where a posted interrupt was
7111 * notified with kvm_vcpu_kick.
7112 */
7113 if (kvm_lapic_enabled(vcpu)) {
7114 if (kvm_x86_ops->sync_pir_to_irr && vcpu->arch.apicv_active)
7115 kvm_x86_ops->sync_pir_to_irr(vcpu);
7116 }
7117
7118 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
7119 || need_resched() || signal_pending(current)) {
7120 vcpu->mode = OUTSIDE_GUEST_MODE;
7121 smp_wmb();
7122 local_irq_enable();
7123 preempt_enable();
7124 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7125 r = 1;
7126 goto cancel_injection;
7127 }
7128
7129 kvm_load_guest_xcr0(vcpu);
7130
7131 if (req_immediate_exit) {
7132 kvm_make_request(KVM_REQ_EVENT, vcpu);
7133 smp_send_reschedule(vcpu->cpu);
7134 }
7135
7136 trace_kvm_entry(vcpu->vcpu_id);
7137 wait_lapic_expire(vcpu);
7138 guest_enter_irqoff();
7139
7140 if (unlikely(vcpu->arch.switch_db_regs)) {
7141 set_debugreg(0, 7);
7142 set_debugreg(vcpu->arch.eff_db[0], 0);
7143 set_debugreg(vcpu->arch.eff_db[1], 1);
7144 set_debugreg(vcpu->arch.eff_db[2], 2);
7145 set_debugreg(vcpu->arch.eff_db[3], 3);
7146 set_debugreg(vcpu->arch.dr6, 6);
7147 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7148 }
7149
7150 kvm_x86_ops->run(vcpu);
7151
7152 /*
7153 * Do this here before restoring debug registers on the host. And
7154 * since we do this before handling the vmexit, a DR access vmexit
7155 * can (a) read the correct value of the debug registers, (b) set
7156 * KVM_DEBUGREG_WONT_EXIT again.
7157 */
7158 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
7159 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
7160 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
7161 kvm_update_dr0123(vcpu);
7162 kvm_update_dr6(vcpu);
7163 kvm_update_dr7(vcpu);
7164 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7165 }
7166
7167 /*
7168 * If the guest has used debug registers, at least dr7
7169 * will be disabled while returning to the host.
7170 * If we don't have active breakpoints in the host, we don't
7171 * care about the messed up debug address registers. But if
7172 * we have some of them active, restore the old state.
7173 */
7174 if (hw_breakpoint_active())
7175 hw_breakpoint_restore();
7176
7177 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
7178
7179 vcpu->mode = OUTSIDE_GUEST_MODE;
7180 smp_wmb();
7181
7182 kvm_put_guest_xcr0(vcpu);
7183
7184 kvm_x86_ops->handle_external_intr(vcpu);
7185
7186 ++vcpu->stat.exits;
7187
7188 guest_exit_irqoff();
7189
7190 local_irq_enable();
7191 preempt_enable();
7192
7193 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7194
7195 /*
7196 * Profile KVM exit RIPs:
7197 */
7198 if (unlikely(prof_on == KVM_PROFILING)) {
7199 unsigned long rip = kvm_rip_read(vcpu);
7200 profile_hit(KVM_PROFILING, (void *)rip);
7201 }
7202
7203 if (unlikely(vcpu->arch.tsc_always_catchup))
7204 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7205
7206 if (vcpu->arch.apic_attention)
7207 kvm_lapic_sync_from_vapic(vcpu);
7208
7209 vcpu->arch.gpa_available = false;
7210 r = kvm_x86_ops->handle_exit(vcpu);
7211 return r;
7212
7213 cancel_injection:
7214 kvm_x86_ops->cancel_injection(vcpu);
7215 if (unlikely(vcpu->arch.apic_attention))
7216 kvm_lapic_sync_from_vapic(vcpu);
7217 out:
7218 return r;
7219 }
7220
7221 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7222 {
7223 if (!kvm_arch_vcpu_runnable(vcpu) &&
7224 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7225 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7226 kvm_vcpu_block(vcpu);
7227 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7228
7229 if (kvm_x86_ops->post_block)
7230 kvm_x86_ops->post_block(vcpu);
7231
7232 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7233 return 1;
7234 }
7235
7236 kvm_apic_accept_events(vcpu);
7237 switch(vcpu->arch.mp_state) {
7238 case KVM_MP_STATE_HALTED:
7239 vcpu->arch.pv.pv_unhalted = false;
7240 vcpu->arch.mp_state =
7241 KVM_MP_STATE_RUNNABLE;
7242 case KVM_MP_STATE_RUNNABLE:
7243 vcpu->arch.apf.halted = false;
7244 break;
7245 case KVM_MP_STATE_INIT_RECEIVED:
7246 break;
7247 default:
7248 return -EINTR;
7249 break;
7250 }
7251 return 1;
7252 }
7253
7254 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7255 {
7256 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7257 kvm_x86_ops->check_nested_events(vcpu, false);
7258
7259 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7260 !vcpu->arch.apf.halted);
7261 }
7262
7263 static int vcpu_run(struct kvm_vcpu *vcpu)
7264 {
7265 int r;
7266 struct kvm *kvm = vcpu->kvm;
7267
7268 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7269 vcpu->arch.l1tf_flush_l1d = true;
7270
7271 for (;;) {
7272 if (kvm_vcpu_running(vcpu)) {
7273 r = vcpu_enter_guest(vcpu);
7274 } else {
7275 r = vcpu_block(kvm, vcpu);
7276 }
7277
7278 if (r <= 0)
7279 break;
7280
7281 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
7282 if (kvm_cpu_has_pending_timer(vcpu))
7283 kvm_inject_pending_timer_irqs(vcpu);
7284
7285 if (dm_request_for_irq_injection(vcpu) &&
7286 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
7287 r = 0;
7288 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
7289 ++vcpu->stat.request_irq_exits;
7290 break;
7291 }
7292
7293 kvm_check_async_pf_completion(vcpu);
7294
7295 if (signal_pending(current)) {
7296 r = -EINTR;
7297 vcpu->run->exit_reason = KVM_EXIT_INTR;
7298 ++vcpu->stat.signal_exits;
7299 break;
7300 }
7301 if (need_resched()) {
7302 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7303 cond_resched();
7304 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7305 }
7306 }
7307
7308 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7309
7310 return r;
7311 }
7312
7313 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7314 {
7315 int r;
7316 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7317 r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7318 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7319 if (r != EMULATE_DONE)
7320 return 0;
7321 return 1;
7322 }
7323
7324 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7325 {
7326 BUG_ON(!vcpu->arch.pio.count);
7327
7328 return complete_emulated_io(vcpu);
7329 }
7330
7331 /*
7332 * Implements the following, as a state machine:
7333 *
7334 * read:
7335 * for each fragment
7336 * for each mmio piece in the fragment
7337 * write gpa, len
7338 * exit
7339 * copy data
7340 * execute insn
7341 *
7342 * write:
7343 * for each fragment
7344 * for each mmio piece in the fragment
7345 * write gpa, len
7346 * copy data
7347 * exit
7348 */
7349 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
7350 {
7351 struct kvm_run *run = vcpu->run;
7352 struct kvm_mmio_fragment *frag;
7353 unsigned len;
7354
7355 BUG_ON(!vcpu->mmio_needed);
7356
7357 /* Complete previous fragment */
7358 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
7359 len = min(8u, frag->len);
7360 if (!vcpu->mmio_is_write)
7361 memcpy(frag->data, run->mmio.data, len);
7362
7363 if (frag->len <= 8) {
7364 /* Switch to the next fragment. */
7365 frag++;
7366 vcpu->mmio_cur_fragment++;
7367 } else {
7368 /* Go forward to the next mmio piece. */
7369 frag->data += len;
7370 frag->gpa += len;
7371 frag->len -= len;
7372 }
7373
7374 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
7375 vcpu->mmio_needed = 0;
7376
7377 /* FIXME: return into emulator if single-stepping. */
7378 if (vcpu->mmio_is_write)
7379 return 1;
7380 vcpu->mmio_read_completed = 1;
7381 return complete_emulated_io(vcpu);
7382 }
7383
7384 run->exit_reason = KVM_EXIT_MMIO;
7385 run->mmio.phys_addr = frag->gpa;
7386 if (vcpu->mmio_is_write)
7387 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
7388 run->mmio.len = min(8u, frag->len);
7389 run->mmio.is_write = vcpu->mmio_is_write;
7390 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7391 return 0;
7392 }
7393
7394
7395 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
7396 {
7397 int r;
7398
7399 kvm_sigset_activate(vcpu);
7400
7401 kvm_load_guest_fpu(vcpu);
7402
7403 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
7404 if (kvm_run->immediate_exit) {
7405 r = -EINTR;
7406 goto out;
7407 }
7408 kvm_vcpu_block(vcpu);
7409 kvm_apic_accept_events(vcpu);
7410 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
7411 r = -EAGAIN;
7412 if (signal_pending(current)) {
7413 r = -EINTR;
7414 vcpu->run->exit_reason = KVM_EXIT_INTR;
7415 ++vcpu->stat.signal_exits;
7416 }
7417 goto out;
7418 }
7419
7420 /* re-sync apic's tpr */
7421 if (!lapic_in_kernel(vcpu)) {
7422 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
7423 r = -EINVAL;
7424 goto out;
7425 }
7426 }
7427
7428 if (unlikely(vcpu->arch.complete_userspace_io)) {
7429 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
7430 vcpu->arch.complete_userspace_io = NULL;
7431 r = cui(vcpu);
7432 if (r <= 0)
7433 goto out;
7434 } else
7435 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
7436
7437 if (kvm_run->immediate_exit)
7438 r = -EINTR;
7439 else
7440 r = vcpu_run(vcpu);
7441
7442 out:
7443 kvm_put_guest_fpu(vcpu);
7444 post_kvm_run_save(vcpu);
7445 kvm_sigset_deactivate(vcpu);
7446
7447 return r;
7448 }
7449
7450 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7451 {
7452 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
7453 /*
7454 * We are here if userspace calls get_regs() in the middle of
7455 * instruction emulation. Registers state needs to be copied
7456 * back from emulation context to vcpu. Userspace shouldn't do
7457 * that usually, but some bad designed PV devices (vmware
7458 * backdoor interface) need this to work
7459 */
7460 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
7461 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7462 }
7463 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
7464 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
7465 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
7466 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
7467 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
7468 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
7469 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7470 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
7471 #ifdef CONFIG_X86_64
7472 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
7473 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
7474 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
7475 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
7476 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
7477 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
7478 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
7479 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
7480 #endif
7481
7482 regs->rip = kvm_rip_read(vcpu);
7483 regs->rflags = kvm_get_rflags(vcpu);
7484
7485 return 0;
7486 }
7487
7488 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
7489 {
7490 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
7491 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7492
7493 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
7494 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
7495 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
7496 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
7497 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
7498 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
7499 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
7500 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
7501 #ifdef CONFIG_X86_64
7502 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
7503 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
7504 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
7505 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
7506 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
7507 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
7508 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
7509 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
7510 #endif
7511
7512 kvm_rip_write(vcpu, regs->rip);
7513 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
7514
7515 vcpu->arch.exception.pending = false;
7516
7517 kvm_make_request(KVM_REQ_EVENT, vcpu);
7518
7519 return 0;
7520 }
7521
7522 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
7523 {
7524 struct kvm_segment cs;
7525
7526 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
7527 *db = cs.db;
7528 *l = cs.l;
7529 }
7530 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
7531
7532 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
7533 struct kvm_sregs *sregs)
7534 {
7535 struct desc_ptr dt;
7536
7537 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7538 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7539 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7540 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7541 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7542 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7543
7544 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7545 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7546
7547 kvm_x86_ops->get_idt(vcpu, &dt);
7548 sregs->idt.limit = dt.size;
7549 sregs->idt.base = dt.address;
7550 kvm_x86_ops->get_gdt(vcpu, &dt);
7551 sregs->gdt.limit = dt.size;
7552 sregs->gdt.base = dt.address;
7553
7554 sregs->cr0 = kvm_read_cr0(vcpu);
7555 sregs->cr2 = vcpu->arch.cr2;
7556 sregs->cr3 = kvm_read_cr3(vcpu);
7557 sregs->cr4 = kvm_read_cr4(vcpu);
7558 sregs->cr8 = kvm_get_cr8(vcpu);
7559 sregs->efer = vcpu->arch.efer;
7560 sregs->apic_base = kvm_get_apic_base(vcpu);
7561
7562 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
7563
7564 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
7565 set_bit(vcpu->arch.interrupt.nr,
7566 (unsigned long *)sregs->interrupt_bitmap);
7567
7568 return 0;
7569 }
7570
7571 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
7572 struct kvm_mp_state *mp_state)
7573 {
7574 kvm_apic_accept_events(vcpu);
7575 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
7576 vcpu->arch.pv.pv_unhalted)
7577 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
7578 else
7579 mp_state->mp_state = vcpu->arch.mp_state;
7580
7581 return 0;
7582 }
7583
7584 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
7585 struct kvm_mp_state *mp_state)
7586 {
7587 if (!lapic_in_kernel(vcpu) &&
7588 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
7589 return -EINVAL;
7590
7591 /* INITs are latched while in SMM */
7592 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
7593 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
7594 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
7595 return -EINVAL;
7596
7597 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
7598 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
7599 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
7600 } else
7601 vcpu->arch.mp_state = mp_state->mp_state;
7602 kvm_make_request(KVM_REQ_EVENT, vcpu);
7603 return 0;
7604 }
7605
7606 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
7607 int reason, bool has_error_code, u32 error_code)
7608 {
7609 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7610 int ret;
7611
7612 init_emulate_ctxt(vcpu);
7613
7614 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
7615 has_error_code, error_code);
7616
7617 if (ret)
7618 return EMULATE_FAIL;
7619
7620 kvm_rip_write(vcpu, ctxt->eip);
7621 kvm_set_rflags(vcpu, ctxt->eflags);
7622 kvm_make_request(KVM_REQ_EVENT, vcpu);
7623 return EMULATE_DONE;
7624 }
7625 EXPORT_SYMBOL_GPL(kvm_task_switch);
7626
7627 int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
7628 {
7629 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
7630 /*
7631 * When EFER.LME and CR0.PG are set, the processor is in
7632 * 64-bit mode (though maybe in a 32-bit code segment).
7633 * CR4.PAE and EFER.LMA must be set.
7634 */
7635 if (!(sregs->cr4 & X86_CR4_PAE)
7636 || !(sregs->efer & EFER_LMA))
7637 return -EINVAL;
7638 } else {
7639 /*
7640 * Not in 64-bit mode: EFER.LMA is clear and the code
7641 * segment cannot be 64-bit.
7642 */
7643 if (sregs->efer & EFER_LMA || sregs->cs.l)
7644 return -EINVAL;
7645 }
7646
7647 return 0;
7648 }
7649
7650 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
7651 struct kvm_sregs *sregs)
7652 {
7653 struct msr_data apic_base_msr;
7654 int mmu_reset_needed = 0;
7655 int cpuid_update_needed = 0;
7656 int pending_vec, max_bits, idx;
7657 struct desc_ptr dt;
7658
7659 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
7660 (sregs->cr4 & X86_CR4_OSXSAVE))
7661 return -EINVAL;
7662
7663 if (kvm_valid_sregs(vcpu, sregs))
7664 return -EINVAL;
7665
7666 apic_base_msr.data = sregs->apic_base;
7667 apic_base_msr.host_initiated = true;
7668 if (kvm_set_apic_base(vcpu, &apic_base_msr))
7669 return -EINVAL;
7670
7671 dt.size = sregs->idt.limit;
7672 dt.address = sregs->idt.base;
7673 kvm_x86_ops->set_idt(vcpu, &dt);
7674 dt.size = sregs->gdt.limit;
7675 dt.address = sregs->gdt.base;
7676 kvm_x86_ops->set_gdt(vcpu, &dt);
7677
7678 vcpu->arch.cr2 = sregs->cr2;
7679 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
7680 vcpu->arch.cr3 = sregs->cr3;
7681 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
7682
7683 kvm_set_cr8(vcpu, sregs->cr8);
7684
7685 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
7686 kvm_x86_ops->set_efer(vcpu, sregs->efer);
7687
7688 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
7689 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
7690 vcpu->arch.cr0 = sregs->cr0;
7691
7692 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
7693 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
7694 (X86_CR4_OSXSAVE | X86_CR4_PKE));
7695 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
7696 if (cpuid_update_needed)
7697 kvm_update_cpuid(vcpu);
7698
7699 idx = srcu_read_lock(&vcpu->kvm->srcu);
7700 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
7701 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
7702 mmu_reset_needed = 1;
7703 }
7704 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7705
7706 if (mmu_reset_needed)
7707 kvm_mmu_reset_context(vcpu);
7708
7709 max_bits = KVM_NR_INTERRUPTS;
7710 pending_vec = find_first_bit(
7711 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
7712 if (pending_vec < max_bits) {
7713 kvm_queue_interrupt(vcpu, pending_vec, false);
7714 pr_debug("Set back pending irq %d\n", pending_vec);
7715 }
7716
7717 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
7718 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
7719 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
7720 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
7721 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
7722 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
7723
7724 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
7725 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
7726
7727 update_cr8_intercept(vcpu);
7728
7729 /* Older userspace won't unhalt the vcpu on reset. */
7730 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
7731 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
7732 !is_protmode(vcpu))
7733 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
7734
7735 kvm_make_request(KVM_REQ_EVENT, vcpu);
7736
7737 return 0;
7738 }
7739
7740 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
7741 struct kvm_guest_debug *dbg)
7742 {
7743 unsigned long rflags;
7744 int i, r;
7745
7746 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
7747 r = -EBUSY;
7748 if (vcpu->arch.exception.pending)
7749 goto out;
7750 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
7751 kvm_queue_exception(vcpu, DB_VECTOR);
7752 else
7753 kvm_queue_exception(vcpu, BP_VECTOR);
7754 }
7755
7756 /*
7757 * Read rflags as long as potentially injected trace flags are still
7758 * filtered out.
7759 */
7760 rflags = kvm_get_rflags(vcpu);
7761
7762 vcpu->guest_debug = dbg->control;
7763 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
7764 vcpu->guest_debug = 0;
7765
7766 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
7767 for (i = 0; i < KVM_NR_DB_REGS; ++i)
7768 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
7769 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
7770 } else {
7771 for (i = 0; i < KVM_NR_DB_REGS; i++)
7772 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
7773 }
7774 kvm_update_dr7(vcpu);
7775
7776 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7777 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
7778 get_segment_base(vcpu, VCPU_SREG_CS);
7779
7780 /*
7781 * Trigger an rflags update that will inject or remove the trace
7782 * flags.
7783 */
7784 kvm_set_rflags(vcpu, rflags);
7785
7786 kvm_x86_ops->update_bp_intercept(vcpu);
7787
7788 r = 0;
7789
7790 out:
7791
7792 return r;
7793 }
7794
7795 /*
7796 * Translate a guest virtual address to a guest physical address.
7797 */
7798 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
7799 struct kvm_translation *tr)
7800 {
7801 unsigned long vaddr = tr->linear_address;
7802 gpa_t gpa;
7803 int idx;
7804
7805 idx = srcu_read_lock(&vcpu->kvm->srcu);
7806 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
7807 srcu_read_unlock(&vcpu->kvm->srcu, idx);
7808 tr->physical_address = gpa;
7809 tr->valid = gpa != UNMAPPED_GVA;
7810 tr->writeable = 1;
7811 tr->usermode = 0;
7812
7813 return 0;
7814 }
7815
7816 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7817 {
7818 struct fxregs_state *fxsave =
7819 &vcpu->arch.guest_fpu.state.fxsave;
7820
7821 memcpy(fpu->fpr, fxsave->st_space, 128);
7822 fpu->fcw = fxsave->cwd;
7823 fpu->fsw = fxsave->swd;
7824 fpu->ftwx = fxsave->twd;
7825 fpu->last_opcode = fxsave->fop;
7826 fpu->last_ip = fxsave->rip;
7827 fpu->last_dp = fxsave->rdp;
7828 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
7829
7830 return 0;
7831 }
7832
7833 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
7834 {
7835 struct fxregs_state *fxsave =
7836 &vcpu->arch.guest_fpu.state.fxsave;
7837
7838 memcpy(fxsave->st_space, fpu->fpr, 128);
7839 fxsave->cwd = fpu->fcw;
7840 fxsave->swd = fpu->fsw;
7841 fxsave->twd = fpu->ftwx;
7842 fxsave->fop = fpu->last_opcode;
7843 fxsave->rip = fpu->last_ip;
7844 fxsave->rdp = fpu->last_dp;
7845 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
7846
7847 return 0;
7848 }
7849
7850 static void fx_init(struct kvm_vcpu *vcpu)
7851 {
7852 fpstate_init(&vcpu->arch.guest_fpu.state);
7853 if (boot_cpu_has(X86_FEATURE_XSAVES))
7854 vcpu->arch.guest_fpu.state.xsave.header.xcomp_bv =
7855 host_xcr0 | XSTATE_COMPACTION_ENABLED;
7856
7857 /*
7858 * Ensure guest xcr0 is valid for loading
7859 */
7860 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
7861
7862 vcpu->arch.cr0 |= X86_CR0_ET;
7863 }
7864
7865 /* Swap (qemu) user FPU context for the guest FPU context. */
7866 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
7867 {
7868 preempt_disable();
7869 copy_fpregs_to_fpstate(&vcpu->arch.user_fpu);
7870 /* PKRU is separately restored in kvm_x86_ops->run. */
7871 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu.state,
7872 ~XFEATURE_MASK_PKRU);
7873 preempt_enable();
7874 trace_kvm_fpu(1);
7875 }
7876
7877 /* When vcpu_run ends, restore user space FPU context. */
7878 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
7879 {
7880 preempt_disable();
7881 copy_fpregs_to_fpstate(&vcpu->arch.guest_fpu);
7882 copy_kernel_to_fpregs(&vcpu->arch.user_fpu.state);
7883 preempt_enable();
7884 ++vcpu->stat.fpu_reload;
7885 trace_kvm_fpu(0);
7886 }
7887
7888 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
7889 {
7890 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
7891
7892 kvmclock_reset(vcpu);
7893
7894 kvm_x86_ops->vcpu_free(vcpu);
7895 free_cpumask_var(wbinvd_dirty_mask);
7896 }
7897
7898 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
7899 unsigned int id)
7900 {
7901 struct kvm_vcpu *vcpu;
7902
7903 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
7904 printk_once(KERN_WARNING
7905 "kvm: SMP vm created on host with unstable TSC; "
7906 "guest TSC will not be reliable\n");
7907
7908 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
7909
7910 return vcpu;
7911 }
7912
7913 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
7914 {
7915 int r;
7916
7917 kvm_vcpu_mtrr_init(vcpu);
7918 r = vcpu_load(vcpu);
7919 if (r)
7920 return r;
7921 kvm_vcpu_reset(vcpu, false);
7922 kvm_mmu_setup(vcpu);
7923 vcpu_put(vcpu);
7924 return r;
7925 }
7926
7927 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
7928 {
7929 struct msr_data msr;
7930 struct kvm *kvm = vcpu->kvm;
7931
7932 kvm_hv_vcpu_postcreate(vcpu);
7933
7934 if (vcpu_load(vcpu))
7935 return;
7936 msr.data = 0x0;
7937 msr.index = MSR_IA32_TSC;
7938 msr.host_initiated = true;
7939 kvm_write_tsc(vcpu, &msr);
7940 vcpu_put(vcpu);
7941
7942 if (!kvmclock_periodic_sync)
7943 return;
7944
7945 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
7946 KVMCLOCK_SYNC_PERIOD);
7947 }
7948
7949 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
7950 {
7951 int r;
7952 vcpu->arch.apf.msr_val = 0;
7953
7954 r = vcpu_load(vcpu);
7955 BUG_ON(r);
7956 kvm_mmu_unload(vcpu);
7957 vcpu_put(vcpu);
7958
7959 kvm_x86_ops->vcpu_free(vcpu);
7960 }
7961
7962 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
7963 {
7964 kvm_lapic_reset(vcpu, init_event);
7965
7966 vcpu->arch.hflags = 0;
7967
7968 vcpu->arch.smi_pending = 0;
7969 atomic_set(&vcpu->arch.nmi_queued, 0);
7970 vcpu->arch.nmi_pending = 0;
7971 vcpu->arch.nmi_injected = false;
7972 kvm_clear_interrupt_queue(vcpu);
7973 kvm_clear_exception_queue(vcpu);
7974 vcpu->arch.exception.pending = false;
7975
7976 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
7977 kvm_update_dr0123(vcpu);
7978 vcpu->arch.dr6 = DR6_INIT;
7979 kvm_update_dr6(vcpu);
7980 vcpu->arch.dr7 = DR7_FIXED_1;
7981 kvm_update_dr7(vcpu);
7982
7983 vcpu->arch.cr2 = 0;
7984
7985 kvm_make_request(KVM_REQ_EVENT, vcpu);
7986 vcpu->arch.apf.msr_val = 0;
7987 vcpu->arch.st.msr_val = 0;
7988
7989 kvmclock_reset(vcpu);
7990
7991 kvm_clear_async_pf_completion_queue(vcpu);
7992 kvm_async_pf_hash_reset(vcpu);
7993 vcpu->arch.apf.halted = false;
7994
7995 if (kvm_mpx_supported()) {
7996 void *mpx_state_buffer;
7997
7998 /*
7999 * To avoid have the INIT path from kvm_apic_has_events() that be
8000 * called with loaded FPU and does not let userspace fix the state.
8001 */
8002 if (init_event)
8003 kvm_put_guest_fpu(vcpu);
8004 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
8005 XFEATURE_MASK_BNDREGS);
8006 if (mpx_state_buffer)
8007 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
8008 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu.state.xsave,
8009 XFEATURE_MASK_BNDCSR);
8010 if (mpx_state_buffer)
8011 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
8012 if (init_event)
8013 kvm_load_guest_fpu(vcpu);
8014 }
8015
8016 if (!init_event) {
8017 kvm_pmu_reset(vcpu);
8018 vcpu->arch.smbase = 0x30000;
8019
8020 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
8021 vcpu->arch.msr_misc_features_enables = 0;
8022
8023 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8024 }
8025
8026 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
8027 vcpu->arch.regs_avail = ~0;
8028 vcpu->arch.regs_dirty = ~0;
8029
8030 vcpu->arch.ia32_xss = 0;
8031
8032 kvm_x86_ops->vcpu_reset(vcpu, init_event);
8033 }
8034
8035 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
8036 {
8037 struct kvm_segment cs;
8038
8039 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8040 cs.selector = vector << 8;
8041 cs.base = vector << 12;
8042 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8043 kvm_rip_write(vcpu, 0);
8044 }
8045
8046 int kvm_arch_hardware_enable(void)
8047 {
8048 struct kvm *kvm;
8049 struct kvm_vcpu *vcpu;
8050 int i;
8051 int ret;
8052 u64 local_tsc;
8053 u64 max_tsc = 0;
8054 bool stable, backwards_tsc = false;
8055
8056 kvm_shared_msr_cpu_online();
8057 ret = kvm_x86_ops->hardware_enable();
8058 if (ret != 0)
8059 return ret;
8060
8061 local_tsc = rdtsc();
8062 stable = !check_tsc_unstable();
8063 list_for_each_entry(kvm, &vm_list, vm_list) {
8064 kvm_for_each_vcpu(i, vcpu, kvm) {
8065 if (!stable && vcpu->cpu == smp_processor_id())
8066 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8067 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
8068 backwards_tsc = true;
8069 if (vcpu->arch.last_host_tsc > max_tsc)
8070 max_tsc = vcpu->arch.last_host_tsc;
8071 }
8072 }
8073 }
8074
8075 /*
8076 * Sometimes, even reliable TSCs go backwards. This happens on
8077 * platforms that reset TSC during suspend or hibernate actions, but
8078 * maintain synchronization. We must compensate. Fortunately, we can
8079 * detect that condition here, which happens early in CPU bringup,
8080 * before any KVM threads can be running. Unfortunately, we can't
8081 * bring the TSCs fully up to date with real time, as we aren't yet far
8082 * enough into CPU bringup that we know how much real time has actually
8083 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
8084 * variables that haven't been updated yet.
8085 *
8086 * So we simply find the maximum observed TSC above, then record the
8087 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
8088 * the adjustment will be applied. Note that we accumulate
8089 * adjustments, in case multiple suspend cycles happen before some VCPU
8090 * gets a chance to run again. In the event that no KVM threads get a
8091 * chance to run, we will miss the entire elapsed period, as we'll have
8092 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
8093 * loose cycle time. This isn't too big a deal, since the loss will be
8094 * uniform across all VCPUs (not to mention the scenario is extremely
8095 * unlikely). It is possible that a second hibernate recovery happens
8096 * much faster than a first, causing the observed TSC here to be
8097 * smaller; this would require additional padding adjustment, which is
8098 * why we set last_host_tsc to the local tsc observed here.
8099 *
8100 * N.B. - this code below runs only on platforms with reliable TSC,
8101 * as that is the only way backwards_tsc is set above. Also note
8102 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
8103 * have the same delta_cyc adjustment applied if backwards_tsc
8104 * is detected. Note further, this adjustment is only done once,
8105 * as we reset last_host_tsc on all VCPUs to stop this from being
8106 * called multiple times (one for each physical CPU bringup).
8107 *
8108 * Platforms with unreliable TSCs don't have to deal with this, they
8109 * will be compensated by the logic in vcpu_load, which sets the TSC to
8110 * catchup mode. This will catchup all VCPUs to real time, but cannot
8111 * guarantee that they stay in perfect synchronization.
8112 */
8113 if (backwards_tsc) {
8114 u64 delta_cyc = max_tsc - local_tsc;
8115 list_for_each_entry(kvm, &vm_list, vm_list) {
8116 kvm->arch.backwards_tsc_observed = true;
8117 kvm_for_each_vcpu(i, vcpu, kvm) {
8118 vcpu->arch.tsc_offset_adjustment += delta_cyc;
8119 vcpu->arch.last_host_tsc = local_tsc;
8120 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8121 }
8122
8123 /*
8124 * We have to disable TSC offset matching.. if you were
8125 * booting a VM while issuing an S4 host suspend....
8126 * you may have some problem. Solving this issue is
8127 * left as an exercise to the reader.
8128 */
8129 kvm->arch.last_tsc_nsec = 0;
8130 kvm->arch.last_tsc_write = 0;
8131 }
8132
8133 }
8134 return 0;
8135 }
8136
8137 void kvm_arch_hardware_disable(void)
8138 {
8139 kvm_x86_ops->hardware_disable();
8140 drop_user_return_notifiers();
8141 }
8142
8143 int kvm_arch_hardware_setup(void)
8144 {
8145 int r;
8146
8147 r = kvm_x86_ops->hardware_setup();
8148 if (r != 0)
8149 return r;
8150
8151 if (kvm_has_tsc_control) {
8152 /*
8153 * Make sure the user can only configure tsc_khz values that
8154 * fit into a signed integer.
8155 * A min value is not calculated needed because it will always
8156 * be 1 on all machines.
8157 */
8158 u64 max = min(0x7fffffffULL,
8159 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
8160 kvm_max_guest_tsc_khz = max;
8161
8162 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
8163 }
8164
8165 kvm_init_msr_list();
8166 return 0;
8167 }
8168
8169 void kvm_arch_hardware_unsetup(void)
8170 {
8171 kvm_x86_ops->hardware_unsetup();
8172 }
8173
8174 void kvm_arch_check_processor_compat(void *rtn)
8175 {
8176 kvm_x86_ops->check_processor_compatibility(rtn);
8177 }
8178
8179 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
8180 {
8181 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
8182 }
8183 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
8184
8185 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
8186 {
8187 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
8188 }
8189
8190 struct static_key kvm_no_apic_vcpu __read_mostly;
8191 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
8192
8193 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
8194 {
8195 struct page *page;
8196 int r;
8197
8198 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
8199 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
8200 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
8201 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8202 else
8203 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
8204
8205 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
8206 if (!page) {
8207 r = -ENOMEM;
8208 goto fail;
8209 }
8210 vcpu->arch.pio_data = page_address(page);
8211
8212 kvm_set_tsc_khz(vcpu, max_tsc_khz);
8213
8214 r = kvm_mmu_create(vcpu);
8215 if (r < 0)
8216 goto fail_free_pio_data;
8217
8218 if (irqchip_in_kernel(vcpu->kvm)) {
8219 r = kvm_create_lapic(vcpu);
8220 if (r < 0)
8221 goto fail_mmu_destroy;
8222 } else
8223 static_key_slow_inc(&kvm_no_apic_vcpu);
8224
8225 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
8226 GFP_KERNEL);
8227 if (!vcpu->arch.mce_banks) {
8228 r = -ENOMEM;
8229 goto fail_free_lapic;
8230 }
8231 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
8232
8233 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
8234 r = -ENOMEM;
8235 goto fail_free_mce_banks;
8236 }
8237
8238 fx_init(vcpu);
8239
8240 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
8241
8242 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
8243
8244 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
8245
8246 kvm_async_pf_hash_reset(vcpu);
8247 kvm_pmu_init(vcpu);
8248
8249 vcpu->arch.pending_external_vector = -1;
8250 vcpu->arch.preempted_in_kernel = false;
8251
8252 kvm_hv_vcpu_init(vcpu);
8253
8254 return 0;
8255
8256 fail_free_mce_banks:
8257 kfree(vcpu->arch.mce_banks);
8258 fail_free_lapic:
8259 kvm_free_lapic(vcpu);
8260 fail_mmu_destroy:
8261 kvm_mmu_destroy(vcpu);
8262 fail_free_pio_data:
8263 free_page((unsigned long)vcpu->arch.pio_data);
8264 fail:
8265 return r;
8266 }
8267
8268 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
8269 {
8270 int idx;
8271
8272 kvm_hv_vcpu_uninit(vcpu);
8273 kvm_pmu_destroy(vcpu);
8274 kfree(vcpu->arch.mce_banks);
8275 kvm_free_lapic(vcpu);
8276 idx = srcu_read_lock(&vcpu->kvm->srcu);
8277 kvm_mmu_destroy(vcpu);
8278 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8279 free_page((unsigned long)vcpu->arch.pio_data);
8280 if (!lapic_in_kernel(vcpu))
8281 static_key_slow_dec(&kvm_no_apic_vcpu);
8282 }
8283
8284 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
8285 {
8286 vcpu->arch.l1tf_flush_l1d = true;
8287 kvm_x86_ops->sched_in(vcpu, cpu);
8288 }
8289
8290 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
8291 {
8292 if (type)
8293 return -EINVAL;
8294
8295 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
8296 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
8297 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
8298 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
8299 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
8300
8301 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
8302 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
8303 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
8304 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
8305 &kvm->arch.irq_sources_bitmap);
8306
8307 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
8308 mutex_init(&kvm->arch.apic_map_lock);
8309 mutex_init(&kvm->arch.hyperv.hv_lock);
8310 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
8311
8312 kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
8313 pvclock_update_vm_gtod_copy(kvm);
8314
8315 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
8316 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
8317
8318 kvm_page_track_init(kvm);
8319 kvm_mmu_init_vm(kvm);
8320
8321 if (kvm_x86_ops->vm_init)
8322 return kvm_x86_ops->vm_init(kvm);
8323
8324 return 0;
8325 }
8326
8327 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
8328 {
8329 int r;
8330 r = vcpu_load(vcpu);
8331 BUG_ON(r);
8332 kvm_mmu_unload(vcpu);
8333 vcpu_put(vcpu);
8334 }
8335
8336 static void kvm_free_vcpus(struct kvm *kvm)
8337 {
8338 unsigned int i;
8339 struct kvm_vcpu *vcpu;
8340
8341 /*
8342 * Unpin any mmu pages first.
8343 */
8344 kvm_for_each_vcpu(i, vcpu, kvm) {
8345 kvm_clear_async_pf_completion_queue(vcpu);
8346 kvm_unload_vcpu_mmu(vcpu);
8347 }
8348 kvm_for_each_vcpu(i, vcpu, kvm)
8349 kvm_arch_vcpu_free(vcpu);
8350
8351 mutex_lock(&kvm->lock);
8352 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
8353 kvm->vcpus[i] = NULL;
8354
8355 atomic_set(&kvm->online_vcpus, 0);
8356 mutex_unlock(&kvm->lock);
8357 }
8358
8359 void kvm_arch_sync_events(struct kvm *kvm)
8360 {
8361 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
8362 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
8363 kvm_free_pit(kvm);
8364 }
8365
8366 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8367 {
8368 int i, r;
8369 unsigned long hva;
8370 struct kvm_memslots *slots = kvm_memslots(kvm);
8371 struct kvm_memory_slot *slot, old;
8372
8373 /* Called with kvm->slots_lock held. */
8374 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
8375 return -EINVAL;
8376
8377 slot = id_to_memslot(slots, id);
8378 if (size) {
8379 if (slot->npages)
8380 return -EEXIST;
8381
8382 /*
8383 * MAP_SHARED to prevent internal slot pages from being moved
8384 * by fork()/COW.
8385 */
8386 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
8387 MAP_SHARED | MAP_ANONYMOUS, 0);
8388 if (IS_ERR((void *)hva))
8389 return PTR_ERR((void *)hva);
8390 } else {
8391 if (!slot->npages)
8392 return 0;
8393
8394 hva = 0;
8395 }
8396
8397 old = *slot;
8398 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
8399 struct kvm_userspace_memory_region m;
8400
8401 m.slot = id | (i << 16);
8402 m.flags = 0;
8403 m.guest_phys_addr = gpa;
8404 m.userspace_addr = hva;
8405 m.memory_size = size;
8406 r = __kvm_set_memory_region(kvm, &m);
8407 if (r < 0)
8408 return r;
8409 }
8410
8411 if (!size)
8412 vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
8413
8414 return 0;
8415 }
8416 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
8417
8418 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
8419 {
8420 int r;
8421
8422 mutex_lock(&kvm->slots_lock);
8423 r = __x86_set_memory_region(kvm, id, gpa, size);
8424 mutex_unlock(&kvm->slots_lock);
8425
8426 return r;
8427 }
8428 EXPORT_SYMBOL_GPL(x86_set_memory_region);
8429
8430 void kvm_arch_destroy_vm(struct kvm *kvm)
8431 {
8432 if (current->mm == kvm->mm) {
8433 /*
8434 * Free memory regions allocated on behalf of userspace,
8435 * unless the the memory map has changed due to process exit
8436 * or fd copying.
8437 */
8438 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
8439 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
8440 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
8441 }
8442 if (kvm_x86_ops->vm_destroy)
8443 kvm_x86_ops->vm_destroy(kvm);
8444 kvm_pic_destroy(kvm);
8445 kvm_ioapic_destroy(kvm);
8446 kvm_free_vcpus(kvm);
8447 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
8448 kvm_mmu_uninit_vm(kvm);
8449 kvm_page_track_cleanup(kvm);
8450 }
8451
8452 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
8453 struct kvm_memory_slot *dont)
8454 {
8455 int i;
8456
8457 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8458 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
8459 kvfree(free->arch.rmap[i]);
8460 free->arch.rmap[i] = NULL;
8461 }
8462 if (i == 0)
8463 continue;
8464
8465 if (!dont || free->arch.lpage_info[i - 1] !=
8466 dont->arch.lpage_info[i - 1]) {
8467 kvfree(free->arch.lpage_info[i - 1]);
8468 free->arch.lpage_info[i - 1] = NULL;
8469 }
8470 }
8471
8472 kvm_page_track_free_memslot(free, dont);
8473 }
8474
8475 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
8476 unsigned long npages)
8477 {
8478 int i;
8479
8480 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8481 struct kvm_lpage_info *linfo;
8482 unsigned long ugfn;
8483 int lpages;
8484 int level = i + 1;
8485
8486 lpages = gfn_to_index(slot->base_gfn + npages - 1,
8487 slot->base_gfn, level) + 1;
8488
8489 slot->arch.rmap[i] =
8490 kvzalloc(lpages * sizeof(*slot->arch.rmap[i]), GFP_KERNEL);
8491 if (!slot->arch.rmap[i])
8492 goto out_free;
8493 if (i == 0)
8494 continue;
8495
8496 linfo = kvzalloc(lpages * sizeof(*linfo), GFP_KERNEL);
8497 if (!linfo)
8498 goto out_free;
8499
8500 slot->arch.lpage_info[i - 1] = linfo;
8501
8502 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
8503 linfo[0].disallow_lpage = 1;
8504 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
8505 linfo[lpages - 1].disallow_lpage = 1;
8506 ugfn = slot->userspace_addr >> PAGE_SHIFT;
8507 /*
8508 * If the gfn and userspace address are not aligned wrt each
8509 * other, or if explicitly asked to, disable large page
8510 * support for this slot
8511 */
8512 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
8513 !kvm_largepages_enabled()) {
8514 unsigned long j;
8515
8516 for (j = 0; j < lpages; ++j)
8517 linfo[j].disallow_lpage = 1;
8518 }
8519 }
8520
8521 if (kvm_page_track_create_memslot(slot, npages))
8522 goto out_free;
8523
8524 return 0;
8525
8526 out_free:
8527 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
8528 kvfree(slot->arch.rmap[i]);
8529 slot->arch.rmap[i] = NULL;
8530 if (i == 0)
8531 continue;
8532
8533 kvfree(slot->arch.lpage_info[i - 1]);
8534 slot->arch.lpage_info[i - 1] = NULL;
8535 }
8536 return -ENOMEM;
8537 }
8538
8539 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
8540 {
8541 /*
8542 * memslots->generation has been incremented.
8543 * mmio generation may have reached its maximum value.
8544 */
8545 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
8546 }
8547
8548 int kvm_arch_prepare_memory_region(struct kvm *kvm,
8549 struct kvm_memory_slot *memslot,
8550 const struct kvm_userspace_memory_region *mem,
8551 enum kvm_mr_change change)
8552 {
8553 return 0;
8554 }
8555
8556 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
8557 struct kvm_memory_slot *new)
8558 {
8559 /* Still write protect RO slot */
8560 if (new->flags & KVM_MEM_READONLY) {
8561 kvm_mmu_slot_remove_write_access(kvm, new);
8562 return;
8563 }
8564
8565 /*
8566 * Call kvm_x86_ops dirty logging hooks when they are valid.
8567 *
8568 * kvm_x86_ops->slot_disable_log_dirty is called when:
8569 *
8570 * - KVM_MR_CREATE with dirty logging is disabled
8571 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
8572 *
8573 * The reason is, in case of PML, we need to set D-bit for any slots
8574 * with dirty logging disabled in order to eliminate unnecessary GPA
8575 * logging in PML buffer (and potential PML buffer full VMEXT). This
8576 * guarantees leaving PML enabled during guest's lifetime won't have
8577 * any additonal overhead from PML when guest is running with dirty
8578 * logging disabled for memory slots.
8579 *
8580 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
8581 * to dirty logging mode.
8582 *
8583 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
8584 *
8585 * In case of write protect:
8586 *
8587 * Write protect all pages for dirty logging.
8588 *
8589 * All the sptes including the large sptes which point to this
8590 * slot are set to readonly. We can not create any new large
8591 * spte on this slot until the end of the logging.
8592 *
8593 * See the comments in fast_page_fault().
8594 */
8595 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
8596 if (kvm_x86_ops->slot_enable_log_dirty)
8597 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
8598 else
8599 kvm_mmu_slot_remove_write_access(kvm, new);
8600 } else {
8601 if (kvm_x86_ops->slot_disable_log_dirty)
8602 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
8603 }
8604 }
8605
8606 void kvm_arch_commit_memory_region(struct kvm *kvm,
8607 const struct kvm_userspace_memory_region *mem,
8608 const struct kvm_memory_slot *old,
8609 const struct kvm_memory_slot *new,
8610 enum kvm_mr_change change)
8611 {
8612 int nr_mmu_pages = 0;
8613
8614 if (!kvm->arch.n_requested_mmu_pages)
8615 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
8616
8617 if (nr_mmu_pages)
8618 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
8619
8620 /*
8621 * Dirty logging tracks sptes in 4k granularity, meaning that large
8622 * sptes have to be split. If live migration is successful, the guest
8623 * in the source machine will be destroyed and large sptes will be
8624 * created in the destination. However, if the guest continues to run
8625 * in the source machine (for example if live migration fails), small
8626 * sptes will remain around and cause bad performance.
8627 *
8628 * Scan sptes if dirty logging has been stopped, dropping those
8629 * which can be collapsed into a single large-page spte. Later
8630 * page faults will create the large-page sptes.
8631 */
8632 if ((change != KVM_MR_DELETE) &&
8633 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
8634 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
8635 kvm_mmu_zap_collapsible_sptes(kvm, new);
8636
8637 /*
8638 * Set up write protection and/or dirty logging for the new slot.
8639 *
8640 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
8641 * been zapped so no dirty logging staff is needed for old slot. For
8642 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
8643 * new and it's also covered when dealing with the new slot.
8644 *
8645 * FIXME: const-ify all uses of struct kvm_memory_slot.
8646 */
8647 if (change != KVM_MR_DELETE)
8648 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
8649 }
8650
8651 void kvm_arch_flush_shadow_all(struct kvm *kvm)
8652 {
8653 kvm_mmu_invalidate_zap_all_pages(kvm);
8654 }
8655
8656 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
8657 struct kvm_memory_slot *slot)
8658 {
8659 kvm_page_track_flush_slot(kvm, slot);
8660 }
8661
8662 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
8663 {
8664 if (!list_empty_careful(&vcpu->async_pf.done))
8665 return true;
8666
8667 if (kvm_apic_has_events(vcpu))
8668 return true;
8669
8670 if (vcpu->arch.pv.pv_unhalted)
8671 return true;
8672
8673 if (vcpu->arch.exception.pending)
8674 return true;
8675
8676 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
8677 (vcpu->arch.nmi_pending &&
8678 kvm_x86_ops->nmi_allowed(vcpu)))
8679 return true;
8680
8681 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
8682 (vcpu->arch.smi_pending && !is_smm(vcpu)))
8683 return true;
8684
8685 if (kvm_arch_interrupt_allowed(vcpu) &&
8686 kvm_cpu_has_interrupt(vcpu))
8687 return true;
8688
8689 if (kvm_hv_has_stimer_pending(vcpu))
8690 return true;
8691
8692 return false;
8693 }
8694
8695 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
8696 {
8697 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
8698 }
8699
8700 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
8701 {
8702 return vcpu->arch.preempted_in_kernel;
8703 }
8704
8705 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
8706 {
8707 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
8708 }
8709
8710 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
8711 {
8712 return kvm_x86_ops->interrupt_allowed(vcpu);
8713 }
8714
8715 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
8716 {
8717 if (is_64_bit_mode(vcpu))
8718 return kvm_rip_read(vcpu);
8719 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
8720 kvm_rip_read(vcpu));
8721 }
8722 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
8723
8724 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
8725 {
8726 return kvm_get_linear_rip(vcpu) == linear_rip;
8727 }
8728 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
8729
8730 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
8731 {
8732 unsigned long rflags;
8733
8734 rflags = kvm_x86_ops->get_rflags(vcpu);
8735 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8736 rflags &= ~X86_EFLAGS_TF;
8737 return rflags;
8738 }
8739 EXPORT_SYMBOL_GPL(kvm_get_rflags);
8740
8741 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8742 {
8743 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
8744 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
8745 rflags |= X86_EFLAGS_TF;
8746 kvm_x86_ops->set_rflags(vcpu, rflags);
8747 }
8748
8749 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
8750 {
8751 __kvm_set_rflags(vcpu, rflags);
8752 kvm_make_request(KVM_REQ_EVENT, vcpu);
8753 }
8754 EXPORT_SYMBOL_GPL(kvm_set_rflags);
8755
8756 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
8757 {
8758 int r;
8759
8760 if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
8761 work->wakeup_all)
8762 return;
8763
8764 r = kvm_mmu_reload(vcpu);
8765 if (unlikely(r))
8766 return;
8767
8768 if (!vcpu->arch.mmu.direct_map &&
8769 work->arch.cr3 != vcpu->arch.mmu.get_cr3(vcpu))
8770 return;
8771
8772 vcpu->arch.mmu.page_fault(vcpu, work->gva, 0, true);
8773 }
8774
8775 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
8776 {
8777 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
8778 }
8779
8780 static inline u32 kvm_async_pf_next_probe(u32 key)
8781 {
8782 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
8783 }
8784
8785 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8786 {
8787 u32 key = kvm_async_pf_hash_fn(gfn);
8788
8789 while (vcpu->arch.apf.gfns[key] != ~0)
8790 key = kvm_async_pf_next_probe(key);
8791
8792 vcpu->arch.apf.gfns[key] = gfn;
8793 }
8794
8795 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
8796 {
8797 int i;
8798 u32 key = kvm_async_pf_hash_fn(gfn);
8799
8800 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
8801 (vcpu->arch.apf.gfns[key] != gfn &&
8802 vcpu->arch.apf.gfns[key] != ~0); i++)
8803 key = kvm_async_pf_next_probe(key);
8804
8805 return key;
8806 }
8807
8808 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8809 {
8810 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
8811 }
8812
8813 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
8814 {
8815 u32 i, j, k;
8816
8817 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
8818 while (true) {
8819 vcpu->arch.apf.gfns[i] = ~0;
8820 do {
8821 j = kvm_async_pf_next_probe(j);
8822 if (vcpu->arch.apf.gfns[j] == ~0)
8823 return;
8824 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
8825 /*
8826 * k lies cyclically in ]i,j]
8827 * | i.k.j |
8828 * |....j i.k.| or |.k..j i...|
8829 */
8830 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
8831 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
8832 i = j;
8833 }
8834 }
8835
8836 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
8837 {
8838
8839 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
8840 sizeof(val));
8841 }
8842
8843 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
8844 {
8845
8846 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
8847 sizeof(u32));
8848 }
8849
8850 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
8851 struct kvm_async_pf *work)
8852 {
8853 struct x86_exception fault;
8854
8855 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
8856 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
8857
8858 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
8859 (vcpu->arch.apf.send_user_only &&
8860 kvm_x86_ops->get_cpl(vcpu) == 0))
8861 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
8862 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
8863 fault.vector = PF_VECTOR;
8864 fault.error_code_valid = true;
8865 fault.error_code = 0;
8866 fault.nested_page_fault = false;
8867 fault.address = work->arch.token;
8868 fault.async_page_fault = true;
8869 kvm_inject_page_fault(vcpu, &fault);
8870 }
8871 }
8872
8873 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
8874 struct kvm_async_pf *work)
8875 {
8876 struct x86_exception fault;
8877 u32 val;
8878
8879 if (work->wakeup_all)
8880 work->arch.token = ~0; /* broadcast wakeup */
8881 else
8882 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
8883 trace_kvm_async_pf_ready(work->arch.token, work->gva);
8884
8885 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
8886 !apf_get_user(vcpu, &val)) {
8887 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
8888 vcpu->arch.exception.pending &&
8889 vcpu->arch.exception.nr == PF_VECTOR &&
8890 !apf_put_user(vcpu, 0)) {
8891 vcpu->arch.exception.injected = false;
8892 vcpu->arch.exception.pending = false;
8893 vcpu->arch.exception.nr = 0;
8894 vcpu->arch.exception.has_error_code = false;
8895 vcpu->arch.exception.error_code = 0;
8896 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
8897 fault.vector = PF_VECTOR;
8898 fault.error_code_valid = true;
8899 fault.error_code = 0;
8900 fault.nested_page_fault = false;
8901 fault.address = work->arch.token;
8902 fault.async_page_fault = true;
8903 kvm_inject_page_fault(vcpu, &fault);
8904 }
8905 }
8906 vcpu->arch.apf.halted = false;
8907 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8908 }
8909
8910 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
8911 {
8912 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
8913 return true;
8914 else
8915 return kvm_can_do_async_pf(vcpu);
8916 }
8917
8918 void kvm_arch_start_assignment(struct kvm *kvm)
8919 {
8920 atomic_inc(&kvm->arch.assigned_device_count);
8921 }
8922 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
8923
8924 void kvm_arch_end_assignment(struct kvm *kvm)
8925 {
8926 atomic_dec(&kvm->arch.assigned_device_count);
8927 }
8928 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
8929
8930 bool kvm_arch_has_assigned_device(struct kvm *kvm)
8931 {
8932 return atomic_read(&kvm->arch.assigned_device_count);
8933 }
8934 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
8935
8936 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
8937 {
8938 atomic_inc(&kvm->arch.noncoherent_dma_count);
8939 }
8940 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
8941
8942 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
8943 {
8944 atomic_dec(&kvm->arch.noncoherent_dma_count);
8945 }
8946 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
8947
8948 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
8949 {
8950 return atomic_read(&kvm->arch.noncoherent_dma_count);
8951 }
8952 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
8953
8954 bool kvm_arch_has_irq_bypass(void)
8955 {
8956 return kvm_x86_ops->update_pi_irte != NULL;
8957 }
8958
8959 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
8960 struct irq_bypass_producer *prod)
8961 {
8962 struct kvm_kernel_irqfd *irqfd =
8963 container_of(cons, struct kvm_kernel_irqfd, consumer);
8964
8965 irqfd->producer = prod;
8966
8967 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
8968 prod->irq, irqfd->gsi, 1);
8969 }
8970
8971 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
8972 struct irq_bypass_producer *prod)
8973 {
8974 int ret;
8975 struct kvm_kernel_irqfd *irqfd =
8976 container_of(cons, struct kvm_kernel_irqfd, consumer);
8977
8978 WARN_ON(irqfd->producer != prod);
8979 irqfd->producer = NULL;
8980
8981 /*
8982 * When producer of consumer is unregistered, we change back to
8983 * remapped mode, so we can re-use the current implementation
8984 * when the irq is masked/disabled or the consumer side (KVM
8985 * int this case doesn't want to receive the interrupts.
8986 */
8987 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
8988 if (ret)
8989 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
8990 " fails: %d\n", irqfd->consumer.token, ret);
8991 }
8992
8993 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
8994 uint32_t guest_irq, bool set)
8995 {
8996 if (!kvm_x86_ops->update_pi_irte)
8997 return -EINVAL;
8998
8999 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
9000 }
9001
9002 bool kvm_vector_hashing_enabled(void)
9003 {
9004 return vector_hashing;
9005 }
9006 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
9007
9008 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
9009 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
9010 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
9011 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
9012 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
9013 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
9014 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
9015 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
9016 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
9017 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
9018 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
9019 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
9020 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
9021 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
9022 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
9023 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
9024 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
9025 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
9026 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
ubuntu-bionic-kernel mirror