51df0b6c891ab3dc3c0da09bbd68a8759560dd19
[linux-2.6.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
30 #include <linux/clocksource.h>
31 #include <linux/interrupt.h>
32 #include <linux/kvm.h>
33 #include <linux/fs.h>
34 #include <linux/vmalloc.h>
35 #include <linux/module.h>
36 #include <linux/mman.h>
37 #include <linux/highmem.h>
38 #include <linux/iommu.h>
39 #include <linux/intel-iommu.h>
40 #include <linux/cpufreq.h>
41 #include <linux/user-return-notifier.h>
42 #include <linux/srcu.h>
43 #include <linux/slab.h>
44 #include <linux/perf_event.h>
45 #include <linux/uaccess.h>
46 #include <linux/hash.h>
47 #include <trace/events/kvm.h>
48
49 #define CREATE_TRACE_POINTS
50 #include "trace.h"
51
52 #include <asm/debugreg.h>
53 #include <asm/msr.h>
54 #include <asm/desc.h>
55 #include <asm/mtrr.h>
56 #include <asm/mce.h>
57 #include <asm/i387.h>
58 #include <asm/xcr.h>
59 #include <asm/pvclock.h>
60 #include <asm/div64.h>
61
62 #define MAX_IO_MSRS 256
63 #define KVM_MAX_MCE_BANKS 32
64 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
65
66 #define emul_to_vcpu(ctxt) \
67         container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
68
69 /* EFER defaults:
70  * - enable syscall per default because its emulated by KVM
71  * - enable LME and LMA per default on 64 bit KVM
72  */
73 #ifdef CONFIG_X86_64
74 static
75 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
76 #else
77 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
78 #endif
79
80 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
81 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
82
83 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
84 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
85                                     struct kvm_cpuid_entry2 __user *entries);
86
87 struct kvm_x86_ops *kvm_x86_ops;
88 EXPORT_SYMBOL_GPL(kvm_x86_ops);
89
90 int ignore_msrs = 0;
91 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
92
93 bool kvm_has_tsc_control;
94 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
95 u32  kvm_max_guest_tsc_khz;
96 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
97
98 #define KVM_NR_SHARED_MSRS 16
99
100 struct kvm_shared_msrs_global {
101         int nr;
102         u32 msrs[KVM_NR_SHARED_MSRS];
103 };
104
105 struct kvm_shared_msrs {
106         struct user_return_notifier urn;
107         bool registered;
108         struct kvm_shared_msr_values {
109                 u64 host;
110                 u64 curr;
111         } values[KVM_NR_SHARED_MSRS];
112 };
113
114 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
115 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
116
117 struct kvm_stats_debugfs_item debugfs_entries[] = {
118         { "pf_fixed", VCPU_STAT(pf_fixed) },
119         { "pf_guest", VCPU_STAT(pf_guest) },
120         { "tlb_flush", VCPU_STAT(tlb_flush) },
121         { "invlpg", VCPU_STAT(invlpg) },
122         { "exits", VCPU_STAT(exits) },
123         { "io_exits", VCPU_STAT(io_exits) },
124         { "mmio_exits", VCPU_STAT(mmio_exits) },
125         { "signal_exits", VCPU_STAT(signal_exits) },
126         { "irq_window", VCPU_STAT(irq_window_exits) },
127         { "nmi_window", VCPU_STAT(nmi_window_exits) },
128         { "halt_exits", VCPU_STAT(halt_exits) },
129         { "halt_wakeup", VCPU_STAT(halt_wakeup) },
130         { "hypercalls", VCPU_STAT(hypercalls) },
131         { "request_irq", VCPU_STAT(request_irq_exits) },
132         { "irq_exits", VCPU_STAT(irq_exits) },
133         { "host_state_reload", VCPU_STAT(host_state_reload) },
134         { "efer_reload", VCPU_STAT(efer_reload) },
135         { "fpu_reload", VCPU_STAT(fpu_reload) },
136         { "insn_emulation", VCPU_STAT(insn_emulation) },
137         { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
138         { "irq_injections", VCPU_STAT(irq_injections) },
139         { "nmi_injections", VCPU_STAT(nmi_injections) },
140         { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
141         { "mmu_pte_write", VM_STAT(mmu_pte_write) },
142         { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
143         { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
144         { "mmu_flooded", VM_STAT(mmu_flooded) },
145         { "mmu_recycled", VM_STAT(mmu_recycled) },
146         { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
147         { "mmu_unsync", VM_STAT(mmu_unsync) },
148         { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
149         { "largepages", VM_STAT(lpages) },
150         { NULL }
151 };
152
153 u64 __read_mostly host_xcr0;
154
155 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
156
157 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
158 {
159         int i;
160         for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
161                 vcpu->arch.apf.gfns[i] = ~0;
162 }
163
164 static void kvm_on_user_return(struct user_return_notifier *urn)
165 {
166         unsigned slot;
167         struct kvm_shared_msrs *locals
168                 = container_of(urn, struct kvm_shared_msrs, urn);
169         struct kvm_shared_msr_values *values;
170
171         for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
172                 values = &locals->values[slot];
173                 if (values->host != values->curr) {
174                         wrmsrl(shared_msrs_global.msrs[slot], values->host);
175                         values->curr = values->host;
176                 }
177         }
178         locals->registered = false;
179         user_return_notifier_unregister(urn);
180 }
181
182 static void shared_msr_update(unsigned slot, u32 msr)
183 {
184         struct kvm_shared_msrs *smsr;
185         u64 value;
186
187         smsr = &__get_cpu_var(shared_msrs);
188         /* only read, and nobody should modify it at this time,
189          * so don't need lock */
190         if (slot >= shared_msrs_global.nr) {
191                 printk(KERN_ERR "kvm: invalid MSR slot!");
192                 return;
193         }
194         rdmsrl_safe(msr, &value);
195         smsr->values[slot].host = value;
196         smsr->values[slot].curr = value;
197 }
198
199 void kvm_define_shared_msr(unsigned slot, u32 msr)
200 {
201         if (slot >= shared_msrs_global.nr)
202                 shared_msrs_global.nr = slot + 1;
203         shared_msrs_global.msrs[slot] = msr;
204         /* we need ensured the shared_msr_global have been updated */
205         smp_wmb();
206 }
207 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
208
209 static void kvm_shared_msr_cpu_online(void)
210 {
211         unsigned i;
212
213         for (i = 0; i < shared_msrs_global.nr; ++i)
214                 shared_msr_update(i, shared_msrs_global.msrs[i]);
215 }
216
217 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
218 {
219         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
220
221         if (((value ^ smsr->values[slot].curr) & mask) == 0)
222                 return;
223         smsr->values[slot].curr = value;
224         wrmsrl(shared_msrs_global.msrs[slot], value);
225         if (!smsr->registered) {
226                 smsr->urn.on_user_return = kvm_on_user_return;
227                 user_return_notifier_register(&smsr->urn);
228                 smsr->registered = true;
229         }
230 }
231 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
232
233 static void drop_user_return_notifiers(void *ignore)
234 {
235         struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
236
237         if (smsr->registered)
238                 kvm_on_user_return(&smsr->urn);
239 }
240
241 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
242 {
243         if (irqchip_in_kernel(vcpu->kvm))
244                 return vcpu->arch.apic_base;
245         else
246                 return vcpu->arch.apic_base;
247 }
248 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
249
250 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
251 {
252         /* TODO: reserve bits check */
253         if (irqchip_in_kernel(vcpu->kvm))
254                 kvm_lapic_set_base(vcpu, data);
255         else
256                 vcpu->arch.apic_base = data;
257 }
258 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
259
260 #define EXCPT_BENIGN            0
261 #define EXCPT_CONTRIBUTORY      1
262 #define EXCPT_PF                2
263
264 static int exception_class(int vector)
265 {
266         switch (vector) {
267         case PF_VECTOR:
268                 return EXCPT_PF;
269         case DE_VECTOR:
270         case TS_VECTOR:
271         case NP_VECTOR:
272         case SS_VECTOR:
273         case GP_VECTOR:
274                 return EXCPT_CONTRIBUTORY;
275         default:
276                 break;
277         }
278         return EXCPT_BENIGN;
279 }
280
281 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
282                 unsigned nr, bool has_error, u32 error_code,
283                 bool reinject)
284 {
285         u32 prev_nr;
286         int class1, class2;
287
288         kvm_make_request(KVM_REQ_EVENT, vcpu);
289
290         if (!vcpu->arch.exception.pending) {
291         queue:
292                 vcpu->arch.exception.pending = true;
293                 vcpu->arch.exception.has_error_code = has_error;
294                 vcpu->arch.exception.nr = nr;
295                 vcpu->arch.exception.error_code = error_code;
296                 vcpu->arch.exception.reinject = reinject;
297                 return;
298         }
299
300         /* to check exception */
301         prev_nr = vcpu->arch.exception.nr;
302         if (prev_nr == DF_VECTOR) {
303                 /* triple fault -> shutdown */
304                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
305                 return;
306         }
307         class1 = exception_class(prev_nr);
308         class2 = exception_class(nr);
309         if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
310                 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
311                 /* generate double fault per SDM Table 5-5 */
312                 vcpu->arch.exception.pending = true;
313                 vcpu->arch.exception.has_error_code = true;
314                 vcpu->arch.exception.nr = DF_VECTOR;
315                 vcpu->arch.exception.error_code = 0;
316         } else
317                 /* replace previous exception with a new one in a hope
318                    that instruction re-execution will regenerate lost
319                    exception */
320                 goto queue;
321 }
322
323 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
324 {
325         kvm_multiple_exception(vcpu, nr, false, 0, false);
326 }
327 EXPORT_SYMBOL_GPL(kvm_queue_exception);
328
329 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
330 {
331         kvm_multiple_exception(vcpu, nr, false, 0, true);
332 }
333 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
334
335 void kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
336 {
337         if (err)
338                 kvm_inject_gp(vcpu, 0);
339         else
340                 kvm_x86_ops->skip_emulated_instruction(vcpu);
341 }
342 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
343
344 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
345 {
346         ++vcpu->stat.pf_guest;
347         vcpu->arch.cr2 = fault->address;
348         kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
349 }
350
351 void kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
352 {
353         if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
354                 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
355         else
356                 vcpu->arch.mmu.inject_page_fault(vcpu, fault);
357 }
358
359 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
360 {
361         kvm_make_request(KVM_REQ_EVENT, vcpu);
362         vcpu->arch.nmi_pending = 1;
363 }
364 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
365
366 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
367 {
368         kvm_multiple_exception(vcpu, nr, true, error_code, false);
369 }
370 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
371
372 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
373 {
374         kvm_multiple_exception(vcpu, nr, true, error_code, true);
375 }
376 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
377
378 /*
379  * Checks if cpl <= required_cpl; if true, return true.  Otherwise queue
380  * a #GP and return false.
381  */
382 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
383 {
384         if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
385                 return true;
386         kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
387         return false;
388 }
389 EXPORT_SYMBOL_GPL(kvm_require_cpl);
390
391 /*
392  * This function will be used to read from the physical memory of the currently
393  * running guest. The difference to kvm_read_guest_page is that this function
394  * can read from guest physical or from the guest's guest physical memory.
395  */
396 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
397                             gfn_t ngfn, void *data, int offset, int len,
398                             u32 access)
399 {
400         gfn_t real_gfn;
401         gpa_t ngpa;
402
403         ngpa     = gfn_to_gpa(ngfn);
404         real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
405         if (real_gfn == UNMAPPED_GVA)
406                 return -EFAULT;
407
408         real_gfn = gpa_to_gfn(real_gfn);
409
410         return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
411 }
412 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
413
414 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
415                                void *data, int offset, int len, u32 access)
416 {
417         return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
418                                        data, offset, len, access);
419 }
420
421 /*
422  * Load the pae pdptrs.  Return true is they are all valid.
423  */
424 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
425 {
426         gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
427         unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
428         int i;
429         int ret;
430         u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
431
432         ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
433                                       offset * sizeof(u64), sizeof(pdpte),
434                                       PFERR_USER_MASK|PFERR_WRITE_MASK);
435         if (ret < 0) {
436                 ret = 0;
437                 goto out;
438         }
439         for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
440                 if (is_present_gpte(pdpte[i]) &&
441                     (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
442                         ret = 0;
443                         goto out;
444                 }
445         }
446         ret = 1;
447
448         memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
449         __set_bit(VCPU_EXREG_PDPTR,
450                   (unsigned long *)&vcpu->arch.regs_avail);
451         __set_bit(VCPU_EXREG_PDPTR,
452                   (unsigned long *)&vcpu->arch.regs_dirty);
453 out:
454
455         return ret;
456 }
457 EXPORT_SYMBOL_GPL(load_pdptrs);
458
459 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
460 {
461         u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
462         bool changed = true;
463         int offset;
464         gfn_t gfn;
465         int r;
466
467         if (is_long_mode(vcpu) || !is_pae(vcpu))
468                 return false;
469
470         if (!test_bit(VCPU_EXREG_PDPTR,
471                       (unsigned long *)&vcpu->arch.regs_avail))
472                 return true;
473
474         gfn = (kvm_read_cr3(vcpu) & ~31u) >> PAGE_SHIFT;
475         offset = (kvm_read_cr3(vcpu) & ~31u) & (PAGE_SIZE - 1);
476         r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
477                                        PFERR_USER_MASK | PFERR_WRITE_MASK);
478         if (r < 0)
479                 goto out;
480         changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
481 out:
482
483         return changed;
484 }
485
486 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
487 {
488         unsigned long old_cr0 = kvm_read_cr0(vcpu);
489         unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
490                                     X86_CR0_CD | X86_CR0_NW;
491
492         cr0 |= X86_CR0_ET;
493
494 #ifdef CONFIG_X86_64
495         if (cr0 & 0xffffffff00000000UL)
496                 return 1;
497 #endif
498
499         cr0 &= ~CR0_RESERVED_BITS;
500
501         if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
502                 return 1;
503
504         if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
505                 return 1;
506
507         if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
508 #ifdef CONFIG_X86_64
509                 if ((vcpu->arch.efer & EFER_LME)) {
510                         int cs_db, cs_l;
511
512                         if (!is_pae(vcpu))
513                                 return 1;
514                         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
515                         if (cs_l)
516                                 return 1;
517                 } else
518 #endif
519                 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
520                                                  kvm_read_cr3(vcpu)))
521                         return 1;
522         }
523
524         kvm_x86_ops->set_cr0(vcpu, cr0);
525
526         if ((cr0 ^ old_cr0) & X86_CR0_PG) {
527                 kvm_clear_async_pf_completion_queue(vcpu);
528                 kvm_async_pf_hash_reset(vcpu);
529         }
530
531         if ((cr0 ^ old_cr0) & update_bits)
532                 kvm_mmu_reset_context(vcpu);
533         return 0;
534 }
535 EXPORT_SYMBOL_GPL(kvm_set_cr0);
536
537 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
538 {
539         (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
540 }
541 EXPORT_SYMBOL_GPL(kvm_lmsw);
542
543 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
544 {
545         u64 xcr0;
546
547         /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now  */
548         if (index != XCR_XFEATURE_ENABLED_MASK)
549                 return 1;
550         xcr0 = xcr;
551         if (kvm_x86_ops->get_cpl(vcpu) != 0)
552                 return 1;
553         if (!(xcr0 & XSTATE_FP))
554                 return 1;
555         if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
556                 return 1;
557         if (xcr0 & ~host_xcr0)
558                 return 1;
559         vcpu->arch.xcr0 = xcr0;
560         vcpu->guest_xcr0_loaded = 0;
561         return 0;
562 }
563
564 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
565 {
566         if (__kvm_set_xcr(vcpu, index, xcr)) {
567                 kvm_inject_gp(vcpu, 0);
568                 return 1;
569         }
570         return 0;
571 }
572 EXPORT_SYMBOL_GPL(kvm_set_xcr);
573
574 static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
575 {
576         struct kvm_cpuid_entry2 *best;
577
578         best = kvm_find_cpuid_entry(vcpu, 1, 0);
579         return best && (best->ecx & bit(X86_FEATURE_XSAVE));
580 }
581
582 static void update_cpuid(struct kvm_vcpu *vcpu)
583 {
584         struct kvm_cpuid_entry2 *best;
585
586         best = kvm_find_cpuid_entry(vcpu, 1, 0);
587         if (!best)
588                 return;
589
590         /* Update OSXSAVE bit */
591         if (cpu_has_xsave && best->function == 0x1) {
592                 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
593                 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
594                         best->ecx |= bit(X86_FEATURE_OSXSAVE);
595         }
596 }
597
598 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
599 {
600         unsigned long old_cr4 = kvm_read_cr4(vcpu);
601         unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
602
603         if (cr4 & CR4_RESERVED_BITS)
604                 return 1;
605
606         if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
607                 return 1;
608
609         if (is_long_mode(vcpu)) {
610                 if (!(cr4 & X86_CR4_PAE))
611                         return 1;
612         } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
613                    && ((cr4 ^ old_cr4) & pdptr_bits)
614                    && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
615                                    kvm_read_cr3(vcpu)))
616                 return 1;
617
618         if (cr4 & X86_CR4_VMXE)
619                 return 1;
620
621         kvm_x86_ops->set_cr4(vcpu, cr4);
622
623         if ((cr4 ^ old_cr4) & pdptr_bits)
624                 kvm_mmu_reset_context(vcpu);
625
626         if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
627                 update_cpuid(vcpu);
628
629         return 0;
630 }
631 EXPORT_SYMBOL_GPL(kvm_set_cr4);
632
633 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
634 {
635         if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
636                 kvm_mmu_sync_roots(vcpu);
637                 kvm_mmu_flush_tlb(vcpu);
638                 return 0;
639         }
640
641         if (is_long_mode(vcpu)) {
642                 if (cr3 & CR3_L_MODE_RESERVED_BITS)
643                         return 1;
644         } else {
645                 if (is_pae(vcpu)) {
646                         if (cr3 & CR3_PAE_RESERVED_BITS)
647                                 return 1;
648                         if (is_paging(vcpu) &&
649                             !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
650                                 return 1;
651                 }
652                 /*
653                  * We don't check reserved bits in nonpae mode, because
654                  * this isn't enforced, and VMware depends on this.
655                  */
656         }
657
658         /*
659          * Does the new cr3 value map to physical memory? (Note, we
660          * catch an invalid cr3 even in real-mode, because it would
661          * cause trouble later on when we turn on paging anyway.)
662          *
663          * A real CPU would silently accept an invalid cr3 and would
664          * attempt to use it - with largely undefined (and often hard
665          * to debug) behavior on the guest side.
666          */
667         if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
668                 return 1;
669         vcpu->arch.cr3 = cr3;
670         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
671         vcpu->arch.mmu.new_cr3(vcpu);
672         return 0;
673 }
674 EXPORT_SYMBOL_GPL(kvm_set_cr3);
675
676 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
677 {
678         if (cr8 & CR8_RESERVED_BITS)
679                 return 1;
680         if (irqchip_in_kernel(vcpu->kvm))
681                 kvm_lapic_set_tpr(vcpu, cr8);
682         else
683                 vcpu->arch.cr8 = cr8;
684         return 0;
685 }
686 EXPORT_SYMBOL_GPL(kvm_set_cr8);
687
688 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
689 {
690         if (irqchip_in_kernel(vcpu->kvm))
691                 return kvm_lapic_get_cr8(vcpu);
692         else
693                 return vcpu->arch.cr8;
694 }
695 EXPORT_SYMBOL_GPL(kvm_get_cr8);
696
697 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
698 {
699         switch (dr) {
700         case 0 ... 3:
701                 vcpu->arch.db[dr] = val;
702                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
703                         vcpu->arch.eff_db[dr] = val;
704                 break;
705         case 4:
706                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
707                         return 1; /* #UD */
708                 /* fall through */
709         case 6:
710                 if (val & 0xffffffff00000000ULL)
711                         return -1; /* #GP */
712                 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
713                 break;
714         case 5:
715                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
716                         return 1; /* #UD */
717                 /* fall through */
718         default: /* 7 */
719                 if (val & 0xffffffff00000000ULL)
720                         return -1; /* #GP */
721                 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
722                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
723                         kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
724                         vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
725                 }
726                 break;
727         }
728
729         return 0;
730 }
731
732 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
733 {
734         int res;
735
736         res = __kvm_set_dr(vcpu, dr, val);
737         if (res > 0)
738                 kvm_queue_exception(vcpu, UD_VECTOR);
739         else if (res < 0)
740                 kvm_inject_gp(vcpu, 0);
741
742         return res;
743 }
744 EXPORT_SYMBOL_GPL(kvm_set_dr);
745
746 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
747 {
748         switch (dr) {
749         case 0 ... 3:
750                 *val = vcpu->arch.db[dr];
751                 break;
752         case 4:
753                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
754                         return 1;
755                 /* fall through */
756         case 6:
757                 *val = vcpu->arch.dr6;
758                 break;
759         case 5:
760                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
761                         return 1;
762                 /* fall through */
763         default: /* 7 */
764                 *val = vcpu->arch.dr7;
765                 break;
766         }
767
768         return 0;
769 }
770
771 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
772 {
773         if (_kvm_get_dr(vcpu, dr, val)) {
774                 kvm_queue_exception(vcpu, UD_VECTOR);
775                 return 1;
776         }
777         return 0;
778 }
779 EXPORT_SYMBOL_GPL(kvm_get_dr);
780
781 /*
782  * List of msr numbers which we expose to userspace through KVM_GET_MSRS
783  * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
784  *
785  * This list is modified at module load time to reflect the
786  * capabilities of the host cpu. This capabilities test skips MSRs that are
787  * kvm-specific. Those are put in the beginning of the list.
788  */
789
790 #define KVM_SAVE_MSRS_BEGIN     8
791 static u32 msrs_to_save[] = {
792         MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
793         MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
794         HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
795         HV_X64_MSR_APIC_ASSIST_PAGE, MSR_KVM_ASYNC_PF_EN,
796         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
797         MSR_STAR,
798 #ifdef CONFIG_X86_64
799         MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
800 #endif
801         MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
802 };
803
804 static unsigned num_msrs_to_save;
805
806 static u32 emulated_msrs[] = {
807         MSR_IA32_MISC_ENABLE,
808         MSR_IA32_MCG_STATUS,
809         MSR_IA32_MCG_CTL,
810 };
811
812 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
813 {
814         u64 old_efer = vcpu->arch.efer;
815
816         if (efer & efer_reserved_bits)
817                 return 1;
818
819         if (is_paging(vcpu)
820             && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
821                 return 1;
822
823         if (efer & EFER_FFXSR) {
824                 struct kvm_cpuid_entry2 *feat;
825
826                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
827                 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
828                         return 1;
829         }
830
831         if (efer & EFER_SVME) {
832                 struct kvm_cpuid_entry2 *feat;
833
834                 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
835                 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
836                         return 1;
837         }
838
839         efer &= ~EFER_LMA;
840         efer |= vcpu->arch.efer & EFER_LMA;
841
842         kvm_x86_ops->set_efer(vcpu, efer);
843
844         vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
845
846         /* Update reserved bits */
847         if ((efer ^ old_efer) & EFER_NX)
848                 kvm_mmu_reset_context(vcpu);
849
850         return 0;
851 }
852
853 void kvm_enable_efer_bits(u64 mask)
854 {
855        efer_reserved_bits &= ~mask;
856 }
857 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
858
859
860 /*
861  * Writes msr value into into the appropriate "register".
862  * Returns 0 on success, non-0 otherwise.
863  * Assumes vcpu_load() was already called.
864  */
865 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
866 {
867         return kvm_x86_ops->set_msr(vcpu, msr_index, data);
868 }
869
870 /*
871  * Adapt set_msr() to msr_io()'s calling convention
872  */
873 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
874 {
875         return kvm_set_msr(vcpu, index, *data);
876 }
877
878 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
879 {
880         int version;
881         int r;
882         struct pvclock_wall_clock wc;
883         struct timespec boot;
884
885         if (!wall_clock)
886                 return;
887
888         r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
889         if (r)
890                 return;
891
892         if (version & 1)
893                 ++version;  /* first time write, random junk */
894
895         ++version;
896
897         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
898
899         /*
900          * The guest calculates current wall clock time by adding
901          * system time (updated by kvm_guest_time_update below) to the
902          * wall clock specified here.  guest system time equals host
903          * system time for us, thus we must fill in host boot time here.
904          */
905         getboottime(&boot);
906
907         wc.sec = boot.tv_sec;
908         wc.nsec = boot.tv_nsec;
909         wc.version = version;
910
911         kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
912
913         version++;
914         kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
915 }
916
917 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
918 {
919         uint32_t quotient, remainder;
920
921         /* Don't try to replace with do_div(), this one calculates
922          * "(dividend << 32) / divisor" */
923         __asm__ ( "divl %4"
924                   : "=a" (quotient), "=d" (remainder)
925                   : "0" (0), "1" (dividend), "r" (divisor) );
926         return quotient;
927 }
928
929 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
930                                s8 *pshift, u32 *pmultiplier)
931 {
932         uint64_t scaled64;
933         int32_t  shift = 0;
934         uint64_t tps64;
935         uint32_t tps32;
936
937         tps64 = base_khz * 1000LL;
938         scaled64 = scaled_khz * 1000LL;
939         while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
940                 tps64 >>= 1;
941                 shift--;
942         }
943
944         tps32 = (uint32_t)tps64;
945         while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
946                 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
947                         scaled64 >>= 1;
948                 else
949                         tps32 <<= 1;
950                 shift++;
951         }
952
953         *pshift = shift;
954         *pmultiplier = div_frac(scaled64, tps32);
955
956         pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
957                  __func__, base_khz, scaled_khz, shift, *pmultiplier);
958 }
959
960 static inline u64 get_kernel_ns(void)
961 {
962         struct timespec ts;
963
964         WARN_ON(preemptible());
965         ktime_get_ts(&ts);
966         monotonic_to_bootbased(&ts);
967         return timespec_to_ns(&ts);
968 }
969
970 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
971 unsigned long max_tsc_khz;
972
973 static inline int kvm_tsc_changes_freq(void)
974 {
975         int cpu = get_cpu();
976         int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
977                   cpufreq_quick_get(cpu) != 0;
978         put_cpu();
979         return ret;
980 }
981
982 static u64 vcpu_tsc_khz(struct kvm_vcpu *vcpu)
983 {
984         if (vcpu->arch.virtual_tsc_khz)
985                 return vcpu->arch.virtual_tsc_khz;
986         else
987                 return __this_cpu_read(cpu_tsc_khz);
988 }
989
990 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
991 {
992         u64 ret;
993
994         WARN_ON(preemptible());
995         if (kvm_tsc_changes_freq())
996                 printk_once(KERN_WARNING
997                  "kvm: unreliable cycle conversion on adjustable rate TSC\n");
998         ret = nsec * vcpu_tsc_khz(vcpu);
999         do_div(ret, USEC_PER_SEC);
1000         return ret;
1001 }
1002
1003 static void kvm_init_tsc_catchup(struct kvm_vcpu *vcpu, u32 this_tsc_khz)
1004 {
1005         /* Compute a scale to convert nanoseconds in TSC cycles */
1006         kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
1007                            &vcpu->arch.tsc_catchup_shift,
1008                            &vcpu->arch.tsc_catchup_mult);
1009 }
1010
1011 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1012 {
1013         u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1014                                       vcpu->arch.tsc_catchup_mult,
1015                                       vcpu->arch.tsc_catchup_shift);
1016         tsc += vcpu->arch.last_tsc_write;
1017         return tsc;
1018 }
1019
1020 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1021 {
1022         struct kvm *kvm = vcpu->kvm;
1023         u64 offset, ns, elapsed;
1024         unsigned long flags;
1025         s64 sdiff;
1026
1027         raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1028         offset = kvm_x86_ops->compute_tsc_offset(vcpu, data);
1029         ns = get_kernel_ns();
1030         elapsed = ns - kvm->arch.last_tsc_nsec;
1031         sdiff = data - kvm->arch.last_tsc_write;
1032         if (sdiff < 0)
1033                 sdiff = -sdiff;
1034
1035         /*
1036          * Special case: close write to TSC within 5 seconds of
1037          * another CPU is interpreted as an attempt to synchronize
1038          * The 5 seconds is to accommodate host load / swapping as
1039          * well as any reset of TSC during the boot process.
1040          *
1041          * In that case, for a reliable TSC, we can match TSC offsets,
1042          * or make a best guest using elapsed value.
1043          */
1044         if (sdiff < nsec_to_cycles(vcpu, 5ULL * NSEC_PER_SEC) &&
1045             elapsed < 5ULL * NSEC_PER_SEC) {
1046                 if (!check_tsc_unstable()) {
1047                         offset = kvm->arch.last_tsc_offset;
1048                         pr_debug("kvm: matched tsc offset for %llu\n", data);
1049                 } else {
1050                         u64 delta = nsec_to_cycles(vcpu, elapsed);
1051                         offset += delta;
1052                         pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1053                 }
1054                 ns = kvm->arch.last_tsc_nsec;
1055         }
1056         kvm->arch.last_tsc_nsec = ns;
1057         kvm->arch.last_tsc_write = data;
1058         kvm->arch.last_tsc_offset = offset;
1059         kvm_x86_ops->write_tsc_offset(vcpu, offset);
1060         raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1061
1062         /* Reset of TSC must disable overshoot protection below */
1063         vcpu->arch.hv_clock.tsc_timestamp = 0;
1064         vcpu->arch.last_tsc_write = data;
1065         vcpu->arch.last_tsc_nsec = ns;
1066 }
1067 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1068
1069 static int kvm_guest_time_update(struct kvm_vcpu *v)
1070 {
1071         unsigned long flags;
1072         struct kvm_vcpu_arch *vcpu = &v->arch;
1073         void *shared_kaddr;
1074         unsigned long this_tsc_khz;
1075         s64 kernel_ns, max_kernel_ns;
1076         u64 tsc_timestamp;
1077
1078         /* Keep irq disabled to prevent changes to the clock */
1079         local_irq_save(flags);
1080         kvm_get_msr(v, MSR_IA32_TSC, &tsc_timestamp);
1081         kernel_ns = get_kernel_ns();
1082         this_tsc_khz = vcpu_tsc_khz(v);
1083         if (unlikely(this_tsc_khz == 0)) {
1084                 local_irq_restore(flags);
1085                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1086                 return 1;
1087         }
1088
1089         /*
1090          * We may have to catch up the TSC to match elapsed wall clock
1091          * time for two reasons, even if kvmclock is used.
1092          *   1) CPU could have been running below the maximum TSC rate
1093          *   2) Broken TSC compensation resets the base at each VCPU
1094          *      entry to avoid unknown leaps of TSC even when running
1095          *      again on the same CPU.  This may cause apparent elapsed
1096          *      time to disappear, and the guest to stand still or run
1097          *      very slowly.
1098          */
1099         if (vcpu->tsc_catchup) {
1100                 u64 tsc = compute_guest_tsc(v, kernel_ns);
1101                 if (tsc > tsc_timestamp) {
1102                         kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1103                         tsc_timestamp = tsc;
1104                 }
1105         }
1106
1107         local_irq_restore(flags);
1108
1109         if (!vcpu->time_page)
1110                 return 0;
1111
1112         /*
1113          * Time as measured by the TSC may go backwards when resetting the base
1114          * tsc_timestamp.  The reason for this is that the TSC resolution is
1115          * higher than the resolution of the other clock scales.  Thus, many
1116          * possible measurments of the TSC correspond to one measurement of any
1117          * other clock, and so a spread of values is possible.  This is not a
1118          * problem for the computation of the nanosecond clock; with TSC rates
1119          * around 1GHZ, there can only be a few cycles which correspond to one
1120          * nanosecond value, and any path through this code will inevitably
1121          * take longer than that.  However, with the kernel_ns value itself,
1122          * the precision may be much lower, down to HZ granularity.  If the
1123          * first sampling of TSC against kernel_ns ends in the low part of the
1124          * range, and the second in the high end of the range, we can get:
1125          *
1126          * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1127          *
1128          * As the sampling errors potentially range in the thousands of cycles,
1129          * it is possible such a time value has already been observed by the
1130          * guest.  To protect against this, we must compute the system time as
1131          * observed by the guest and ensure the new system time is greater.
1132          */
1133         max_kernel_ns = 0;
1134         if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1135                 max_kernel_ns = vcpu->last_guest_tsc -
1136                                 vcpu->hv_clock.tsc_timestamp;
1137                 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1138                                     vcpu->hv_clock.tsc_to_system_mul,
1139                                     vcpu->hv_clock.tsc_shift);
1140                 max_kernel_ns += vcpu->last_kernel_ns;
1141         }
1142
1143         if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1144                 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1145                                    &vcpu->hv_clock.tsc_shift,
1146                                    &vcpu->hv_clock.tsc_to_system_mul);
1147                 vcpu->hw_tsc_khz = this_tsc_khz;
1148         }
1149
1150         if (max_kernel_ns > kernel_ns)
1151                 kernel_ns = max_kernel_ns;
1152
1153         /* With all the info we got, fill in the values */
1154         vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1155         vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1156         vcpu->last_kernel_ns = kernel_ns;
1157         vcpu->last_guest_tsc = tsc_timestamp;
1158         vcpu->hv_clock.flags = 0;
1159
1160         /*
1161          * The interface expects us to write an even number signaling that the
1162          * update is finished. Since the guest won't see the intermediate
1163          * state, we just increase by 2 at the end.
1164          */
1165         vcpu->hv_clock.version += 2;
1166
1167         shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1168
1169         memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1170                sizeof(vcpu->hv_clock));
1171
1172         kunmap_atomic(shared_kaddr, KM_USER0);
1173
1174         mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1175         return 0;
1176 }
1177
1178 static bool msr_mtrr_valid(unsigned msr)
1179 {
1180         switch (msr) {
1181         case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1182         case MSR_MTRRfix64K_00000:
1183         case MSR_MTRRfix16K_80000:
1184         case MSR_MTRRfix16K_A0000:
1185         case MSR_MTRRfix4K_C0000:
1186         case MSR_MTRRfix4K_C8000:
1187         case MSR_MTRRfix4K_D0000:
1188         case MSR_MTRRfix4K_D8000:
1189         case MSR_MTRRfix4K_E0000:
1190         case MSR_MTRRfix4K_E8000:
1191         case MSR_MTRRfix4K_F0000:
1192         case MSR_MTRRfix4K_F8000:
1193         case MSR_MTRRdefType:
1194         case MSR_IA32_CR_PAT:
1195                 return true;
1196         case 0x2f8:
1197                 return true;
1198         }
1199         return false;
1200 }
1201
1202 static bool valid_pat_type(unsigned t)
1203 {
1204         return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1205 }
1206
1207 static bool valid_mtrr_type(unsigned t)
1208 {
1209         return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1210 }
1211
1212 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1213 {
1214         int i;
1215
1216         if (!msr_mtrr_valid(msr))
1217                 return false;
1218
1219         if (msr == MSR_IA32_CR_PAT) {
1220                 for (i = 0; i < 8; i++)
1221                         if (!valid_pat_type((data >> (i * 8)) & 0xff))
1222                                 return false;
1223                 return true;
1224         } else if (msr == MSR_MTRRdefType) {
1225                 if (data & ~0xcff)
1226                         return false;
1227                 return valid_mtrr_type(data & 0xff);
1228         } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1229                 for (i = 0; i < 8 ; i++)
1230                         if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1231                                 return false;
1232                 return true;
1233         }
1234
1235         /* variable MTRRs */
1236         return valid_mtrr_type(data & 0xff);
1237 }
1238
1239 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1240 {
1241         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1242
1243         if (!mtrr_valid(vcpu, msr, data))
1244                 return 1;
1245
1246         if (msr == MSR_MTRRdefType) {
1247                 vcpu->arch.mtrr_state.def_type = data;
1248                 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1249         } else if (msr == MSR_MTRRfix64K_00000)
1250                 p[0] = data;
1251         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1252                 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1253         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1254                 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1255         else if (msr == MSR_IA32_CR_PAT)
1256                 vcpu->arch.pat = data;
1257         else {  /* Variable MTRRs */
1258                 int idx, is_mtrr_mask;
1259                 u64 *pt;
1260
1261                 idx = (msr - 0x200) / 2;
1262                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1263                 if (!is_mtrr_mask)
1264                         pt =
1265                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1266                 else
1267                         pt =
1268                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1269                 *pt = data;
1270         }
1271
1272         kvm_mmu_reset_context(vcpu);
1273         return 0;
1274 }
1275
1276 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1277 {
1278         u64 mcg_cap = vcpu->arch.mcg_cap;
1279         unsigned bank_num = mcg_cap & 0xff;
1280
1281         switch (msr) {
1282         case MSR_IA32_MCG_STATUS:
1283                 vcpu->arch.mcg_status = data;
1284                 break;
1285         case MSR_IA32_MCG_CTL:
1286                 if (!(mcg_cap & MCG_CTL_P))
1287                         return 1;
1288                 if (data != 0 && data != ~(u64)0)
1289                         return -1;
1290                 vcpu->arch.mcg_ctl = data;
1291                 break;
1292         default:
1293                 if (msr >= MSR_IA32_MC0_CTL &&
1294                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1295                         u32 offset = msr - MSR_IA32_MC0_CTL;
1296                         /* only 0 or all 1s can be written to IA32_MCi_CTL
1297                          * some Linux kernels though clear bit 10 in bank 4 to
1298                          * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1299                          * this to avoid an uncatched #GP in the guest
1300                          */
1301                         if ((offset & 0x3) == 0 &&
1302                             data != 0 && (data | (1 << 10)) != ~(u64)0)
1303                                 return -1;
1304                         vcpu->arch.mce_banks[offset] = data;
1305                         break;
1306                 }
1307                 return 1;
1308         }
1309         return 0;
1310 }
1311
1312 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1313 {
1314         struct kvm *kvm = vcpu->kvm;
1315         int lm = is_long_mode(vcpu);
1316         u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1317                 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1318         u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1319                 : kvm->arch.xen_hvm_config.blob_size_32;
1320         u32 page_num = data & ~PAGE_MASK;
1321         u64 page_addr = data & PAGE_MASK;
1322         u8 *page;
1323         int r;
1324
1325         r = -E2BIG;
1326         if (page_num >= blob_size)
1327                 goto out;
1328         r = -ENOMEM;
1329         page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1330         if (!page)
1331                 goto out;
1332         r = -EFAULT;
1333         if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1334                 goto out_free;
1335         if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1336                 goto out_free;
1337         r = 0;
1338 out_free:
1339         kfree(page);
1340 out:
1341         return r;
1342 }
1343
1344 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1345 {
1346         return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1347 }
1348
1349 static bool kvm_hv_msr_partition_wide(u32 msr)
1350 {
1351         bool r = false;
1352         switch (msr) {
1353         case HV_X64_MSR_GUEST_OS_ID:
1354         case HV_X64_MSR_HYPERCALL:
1355                 r = true;
1356                 break;
1357         }
1358
1359         return r;
1360 }
1361
1362 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1363 {
1364         struct kvm *kvm = vcpu->kvm;
1365
1366         switch (msr) {
1367         case HV_X64_MSR_GUEST_OS_ID:
1368                 kvm->arch.hv_guest_os_id = data;
1369                 /* setting guest os id to zero disables hypercall page */
1370                 if (!kvm->arch.hv_guest_os_id)
1371                         kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1372                 break;
1373         case HV_X64_MSR_HYPERCALL: {
1374                 u64 gfn;
1375                 unsigned long addr;
1376                 u8 instructions[4];
1377
1378                 /* if guest os id is not set hypercall should remain disabled */
1379                 if (!kvm->arch.hv_guest_os_id)
1380                         break;
1381                 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1382                         kvm->arch.hv_hypercall = data;
1383                         break;
1384                 }
1385                 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1386                 addr = gfn_to_hva(kvm, gfn);
1387                 if (kvm_is_error_hva(addr))
1388                         return 1;
1389                 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1390                 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1391                 if (copy_to_user((void __user *)addr, instructions, 4))
1392                         return 1;
1393                 kvm->arch.hv_hypercall = data;
1394                 break;
1395         }
1396         default:
1397                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1398                           "data 0x%llx\n", msr, data);
1399                 return 1;
1400         }
1401         return 0;
1402 }
1403
1404 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1405 {
1406         switch (msr) {
1407         case HV_X64_MSR_APIC_ASSIST_PAGE: {
1408                 unsigned long addr;
1409
1410                 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1411                         vcpu->arch.hv_vapic = data;
1412                         break;
1413                 }
1414                 addr = gfn_to_hva(vcpu->kvm, data >>
1415                                   HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1416                 if (kvm_is_error_hva(addr))
1417                         return 1;
1418                 if (clear_user((void __user *)addr, PAGE_SIZE))
1419                         return 1;
1420                 vcpu->arch.hv_vapic = data;
1421                 break;
1422         }
1423         case HV_X64_MSR_EOI:
1424                 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1425         case HV_X64_MSR_ICR:
1426                 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1427         case HV_X64_MSR_TPR:
1428                 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1429         default:
1430                 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1431                           "data 0x%llx\n", msr, data);
1432                 return 1;
1433         }
1434
1435         return 0;
1436 }
1437
1438 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
1439 {
1440         gpa_t gpa = data & ~0x3f;
1441
1442         /* Bits 2:5 are resrved, Should be zero */
1443         if (data & 0x3c)
1444                 return 1;
1445
1446         vcpu->arch.apf.msr_val = data;
1447
1448         if (!(data & KVM_ASYNC_PF_ENABLED)) {
1449                 kvm_clear_async_pf_completion_queue(vcpu);
1450                 kvm_async_pf_hash_reset(vcpu);
1451                 return 0;
1452         }
1453
1454         if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa))
1455                 return 1;
1456
1457         vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
1458         kvm_async_pf_wakeup_all(vcpu);
1459         return 0;
1460 }
1461
1462 static void kvmclock_reset(struct kvm_vcpu *vcpu)
1463 {
1464         if (vcpu->arch.time_page) {
1465                 kvm_release_page_dirty(vcpu->arch.time_page);
1466                 vcpu->arch.time_page = NULL;
1467         }
1468 }
1469
1470 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1471 {
1472         switch (msr) {
1473         case MSR_EFER:
1474                 return set_efer(vcpu, data);
1475         case MSR_K7_HWCR:
1476                 data &= ~(u64)0x40;     /* ignore flush filter disable */
1477                 data &= ~(u64)0x100;    /* ignore ignne emulation enable */
1478                 if (data != 0) {
1479                         pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1480                                 data);
1481                         return 1;
1482                 }
1483                 break;
1484         case MSR_FAM10H_MMIO_CONF_BASE:
1485                 if (data != 0) {
1486                         pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1487                                 "0x%llx\n", data);
1488                         return 1;
1489                 }
1490                 break;
1491         case MSR_AMD64_NB_CFG:
1492                 break;
1493         case MSR_IA32_DEBUGCTLMSR:
1494                 if (!data) {
1495                         /* We support the non-activated case already */
1496                         break;
1497                 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1498                         /* Values other than LBR and BTF are vendor-specific,
1499                            thus reserved and should throw a #GP */
1500                         return 1;
1501                 }
1502                 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1503                         __func__, data);
1504                 break;
1505         case MSR_IA32_UCODE_REV:
1506         case MSR_IA32_UCODE_WRITE:
1507         case MSR_VM_HSAVE_PA:
1508         case MSR_AMD64_PATCH_LOADER:
1509                 break;
1510         case 0x200 ... 0x2ff:
1511                 return set_msr_mtrr(vcpu, msr, data);
1512         case MSR_IA32_APICBASE:
1513                 kvm_set_apic_base(vcpu, data);
1514                 break;
1515         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1516                 return kvm_x2apic_msr_write(vcpu, msr, data);
1517         case MSR_IA32_MISC_ENABLE:
1518                 vcpu->arch.ia32_misc_enable_msr = data;
1519                 break;
1520         case MSR_KVM_WALL_CLOCK_NEW:
1521         case MSR_KVM_WALL_CLOCK:
1522                 vcpu->kvm->arch.wall_clock = data;
1523                 kvm_write_wall_clock(vcpu->kvm, data);
1524                 break;
1525         case MSR_KVM_SYSTEM_TIME_NEW:
1526         case MSR_KVM_SYSTEM_TIME: {
1527                 kvmclock_reset(vcpu);
1528
1529                 vcpu->arch.time = data;
1530                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1531
1532                 /* we verify if the enable bit is set... */
1533                 if (!(data & 1))
1534                         break;
1535
1536                 /* ...but clean it before doing the actual write */
1537                 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1538
1539                 vcpu->arch.time_page =
1540                                 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1541
1542                 if (is_error_page(vcpu->arch.time_page)) {
1543                         kvm_release_page_clean(vcpu->arch.time_page);
1544                         vcpu->arch.time_page = NULL;
1545                 }
1546                 break;
1547         }
1548         case MSR_KVM_ASYNC_PF_EN:
1549                 if (kvm_pv_enable_async_pf(vcpu, data))
1550                         return 1;
1551                 break;
1552         case MSR_IA32_MCG_CTL:
1553         case MSR_IA32_MCG_STATUS:
1554         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1555                 return set_msr_mce(vcpu, msr, data);
1556
1557         /* Performance counters are not protected by a CPUID bit,
1558          * so we should check all of them in the generic path for the sake of
1559          * cross vendor migration.
1560          * Writing a zero into the event select MSRs disables them,
1561          * which we perfectly emulate ;-). Any other value should be at least
1562          * reported, some guests depend on them.
1563          */
1564         case MSR_P6_EVNTSEL0:
1565         case MSR_P6_EVNTSEL1:
1566         case MSR_K7_EVNTSEL0:
1567         case MSR_K7_EVNTSEL1:
1568         case MSR_K7_EVNTSEL2:
1569         case MSR_K7_EVNTSEL3:
1570                 if (data != 0)
1571                         pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1572                                 "0x%x data 0x%llx\n", msr, data);
1573                 break;
1574         /* at least RHEL 4 unconditionally writes to the perfctr registers,
1575          * so we ignore writes to make it happy.
1576          */
1577         case MSR_P6_PERFCTR0:
1578         case MSR_P6_PERFCTR1:
1579         case MSR_K7_PERFCTR0:
1580         case MSR_K7_PERFCTR1:
1581         case MSR_K7_PERFCTR2:
1582         case MSR_K7_PERFCTR3:
1583                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1584                         "0x%x data 0x%llx\n", msr, data);
1585                 break;
1586         case MSR_K7_CLK_CTL:
1587                 /*
1588                  * Ignore all writes to this no longer documented MSR.
1589                  * Writes are only relevant for old K7 processors,
1590                  * all pre-dating SVM, but a recommended workaround from
1591                  * AMD for these chips. It is possible to speicify the
1592                  * affected processor models on the command line, hence
1593                  * the need to ignore the workaround.
1594                  */
1595                 break;
1596         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1597                 if (kvm_hv_msr_partition_wide(msr)) {
1598                         int r;
1599                         mutex_lock(&vcpu->kvm->lock);
1600                         r = set_msr_hyperv_pw(vcpu, msr, data);
1601                         mutex_unlock(&vcpu->kvm->lock);
1602                         return r;
1603                 } else
1604                         return set_msr_hyperv(vcpu, msr, data);
1605                 break;
1606         case MSR_IA32_BBL_CR_CTL3:
1607                 /* Drop writes to this legacy MSR -- see rdmsr
1608                  * counterpart for further detail.
1609                  */
1610                 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n", msr, data);
1611                 break;
1612         default:
1613                 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1614                         return xen_hvm_config(vcpu, data);
1615                 if (!ignore_msrs) {
1616                         pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1617                                 msr, data);
1618                         return 1;
1619                 } else {
1620                         pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1621                                 msr, data);
1622                         break;
1623                 }
1624         }
1625         return 0;
1626 }
1627 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1628
1629
1630 /*
1631  * Reads an msr value (of 'msr_index') into 'pdata'.
1632  * Returns 0 on success, non-0 otherwise.
1633  * Assumes vcpu_load() was already called.
1634  */
1635 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1636 {
1637         return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1638 }
1639
1640 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1641 {
1642         u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1643
1644         if (!msr_mtrr_valid(msr))
1645                 return 1;
1646
1647         if (msr == MSR_MTRRdefType)
1648                 *pdata = vcpu->arch.mtrr_state.def_type +
1649                          (vcpu->arch.mtrr_state.enabled << 10);
1650         else if (msr == MSR_MTRRfix64K_00000)
1651                 *pdata = p[0];
1652         else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1653                 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1654         else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1655                 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1656         else if (msr == MSR_IA32_CR_PAT)
1657                 *pdata = vcpu->arch.pat;
1658         else {  /* Variable MTRRs */
1659                 int idx, is_mtrr_mask;
1660                 u64 *pt;
1661
1662                 idx = (msr - 0x200) / 2;
1663                 is_mtrr_mask = msr - 0x200 - 2 * idx;
1664                 if (!is_mtrr_mask)
1665                         pt =
1666                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1667                 else
1668                         pt =
1669                           (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1670                 *pdata = *pt;
1671         }
1672
1673         return 0;
1674 }
1675
1676 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1677 {
1678         u64 data;
1679         u64 mcg_cap = vcpu->arch.mcg_cap;
1680         unsigned bank_num = mcg_cap & 0xff;
1681
1682         switch (msr) {
1683         case MSR_IA32_P5_MC_ADDR:
1684         case MSR_IA32_P5_MC_TYPE:
1685                 data = 0;
1686                 break;
1687         case MSR_IA32_MCG_CAP:
1688                 data = vcpu->arch.mcg_cap;
1689                 break;
1690         case MSR_IA32_MCG_CTL:
1691                 if (!(mcg_cap & MCG_CTL_P))
1692                         return 1;
1693                 data = vcpu->arch.mcg_ctl;
1694                 break;
1695         case MSR_IA32_MCG_STATUS:
1696                 data = vcpu->arch.mcg_status;
1697                 break;
1698         default:
1699                 if (msr >= MSR_IA32_MC0_CTL &&
1700                     msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1701                         u32 offset = msr - MSR_IA32_MC0_CTL;
1702                         data = vcpu->arch.mce_banks[offset];
1703                         break;
1704                 }
1705                 return 1;
1706         }
1707         *pdata = data;
1708         return 0;
1709 }
1710
1711 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1712 {
1713         u64 data = 0;
1714         struct kvm *kvm = vcpu->kvm;
1715
1716         switch (msr) {
1717         case HV_X64_MSR_GUEST_OS_ID:
1718                 data = kvm->arch.hv_guest_os_id;
1719                 break;
1720         case HV_X64_MSR_HYPERCALL:
1721                 data = kvm->arch.hv_hypercall;
1722                 break;
1723         default:
1724                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1725                 return 1;
1726         }
1727
1728         *pdata = data;
1729         return 0;
1730 }
1731
1732 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1733 {
1734         u64 data = 0;
1735
1736         switch (msr) {
1737         case HV_X64_MSR_VP_INDEX: {
1738                 int r;
1739                 struct kvm_vcpu *v;
1740                 kvm_for_each_vcpu(r, v, vcpu->kvm)
1741                         if (v == vcpu)
1742                                 data = r;
1743                 break;
1744         }
1745         case HV_X64_MSR_EOI:
1746                 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1747         case HV_X64_MSR_ICR:
1748                 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1749         case HV_X64_MSR_TPR:
1750                 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1751         default:
1752                 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1753                 return 1;
1754         }
1755         *pdata = data;
1756         return 0;
1757 }
1758
1759 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1760 {
1761         u64 data;
1762
1763         switch (msr) {
1764         case MSR_IA32_PLATFORM_ID:
1765         case MSR_IA32_UCODE_REV:
1766         case MSR_IA32_EBL_CR_POWERON:
1767         case MSR_IA32_DEBUGCTLMSR:
1768         case MSR_IA32_LASTBRANCHFROMIP:
1769         case MSR_IA32_LASTBRANCHTOIP:
1770         case MSR_IA32_LASTINTFROMIP:
1771         case MSR_IA32_LASTINTTOIP:
1772         case MSR_K8_SYSCFG:
1773         case MSR_K7_HWCR:
1774         case MSR_VM_HSAVE_PA:
1775         case MSR_P6_PERFCTR0:
1776         case MSR_P6_PERFCTR1:
1777         case MSR_P6_EVNTSEL0:
1778         case MSR_P6_EVNTSEL1:
1779         case MSR_K7_EVNTSEL0:
1780         case MSR_K7_PERFCTR0:
1781         case MSR_K8_INT_PENDING_MSG:
1782         case MSR_AMD64_NB_CFG:
1783         case MSR_FAM10H_MMIO_CONF_BASE:
1784                 data = 0;
1785                 break;
1786         case MSR_MTRRcap:
1787                 data = 0x500 | KVM_NR_VAR_MTRR;
1788                 break;
1789         case 0x200 ... 0x2ff:
1790                 return get_msr_mtrr(vcpu, msr, pdata);
1791         case 0xcd: /* fsb frequency */
1792                 data = 3;
1793                 break;
1794                 /*
1795                  * MSR_EBC_FREQUENCY_ID
1796                  * Conservative value valid for even the basic CPU models.
1797                  * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1798                  * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1799                  * and 266MHz for model 3, or 4. Set Core Clock
1800                  * Frequency to System Bus Frequency Ratio to 1 (bits
1801                  * 31:24) even though these are only valid for CPU
1802                  * models > 2, however guests may end up dividing or
1803                  * multiplying by zero otherwise.
1804                  */
1805         case MSR_EBC_FREQUENCY_ID:
1806                 data = 1 << 24;
1807                 break;
1808         case MSR_IA32_APICBASE:
1809                 data = kvm_get_apic_base(vcpu);
1810                 break;
1811         case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1812                 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1813                 break;
1814         case MSR_IA32_MISC_ENABLE:
1815                 data = vcpu->arch.ia32_misc_enable_msr;
1816                 break;
1817         case MSR_IA32_PERF_STATUS:
1818                 /* TSC increment by tick */
1819                 data = 1000ULL;
1820                 /* CPU multiplier */
1821                 data |= (((uint64_t)4ULL) << 40);
1822                 break;
1823         case MSR_EFER:
1824                 data = vcpu->arch.efer;
1825                 break;
1826         case MSR_KVM_WALL_CLOCK:
1827         case MSR_KVM_WALL_CLOCK_NEW:
1828                 data = vcpu->kvm->arch.wall_clock;
1829                 break;
1830         case MSR_KVM_SYSTEM_TIME:
1831         case MSR_KVM_SYSTEM_TIME_NEW:
1832                 data = vcpu->arch.time;
1833                 break;
1834         case MSR_KVM_ASYNC_PF_EN:
1835                 data = vcpu->arch.apf.msr_val;
1836                 break;
1837         case MSR_IA32_P5_MC_ADDR:
1838         case MSR_IA32_P5_MC_TYPE:
1839         case MSR_IA32_MCG_CAP:
1840         case MSR_IA32_MCG_CTL:
1841         case MSR_IA32_MCG_STATUS:
1842         case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1843                 return get_msr_mce(vcpu, msr, pdata);
1844         case MSR_K7_CLK_CTL:
1845                 /*
1846                  * Provide expected ramp-up count for K7. All other
1847                  * are set to zero, indicating minimum divisors for
1848                  * every field.
1849                  *
1850                  * This prevents guest kernels on AMD host with CPU
1851                  * type 6, model 8 and higher from exploding due to
1852                  * the rdmsr failing.
1853                  */
1854                 data = 0x20000000;
1855                 break;
1856         case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1857                 if (kvm_hv_msr_partition_wide(msr)) {
1858                         int r;
1859                         mutex_lock(&vcpu->kvm->lock);
1860                         r = get_msr_hyperv_pw(vcpu, msr, pdata);
1861                         mutex_unlock(&vcpu->kvm->lock);
1862                         return r;
1863                 } else
1864                         return get_msr_hyperv(vcpu, msr, pdata);
1865                 break;
1866         case MSR_IA32_BBL_CR_CTL3:
1867                 /* This legacy MSR exists but isn't fully documented in current
1868                  * silicon.  It is however accessed by winxp in very narrow
1869                  * scenarios where it sets bit #19, itself documented as
1870                  * a "reserved" bit.  Best effort attempt to source coherent
1871                  * read data here should the balance of the register be
1872                  * interpreted by the guest:
1873                  *
1874                  * L2 cache control register 3: 64GB range, 256KB size,
1875                  * enabled, latency 0x1, configured
1876                  */
1877                 data = 0xbe702111;
1878                 break;
1879         default:
1880                 if (!ignore_msrs) {
1881                         pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1882                         return 1;
1883                 } else {
1884                         pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1885                         data = 0;
1886                 }
1887                 break;
1888         }
1889         *pdata = data;
1890         return 0;
1891 }
1892 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1893
1894 /*
1895  * Read or write a bunch of msrs. All parameters are kernel addresses.
1896  *
1897  * @return number of msrs set successfully.
1898  */
1899 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1900                     struct kvm_msr_entry *entries,
1901                     int (*do_msr)(struct kvm_vcpu *vcpu,
1902                                   unsigned index, u64 *data))
1903 {
1904         int i, idx;
1905
1906         idx = srcu_read_lock(&vcpu->kvm->srcu);
1907         for (i = 0; i < msrs->nmsrs; ++i)
1908                 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1909                         break;
1910         srcu_read_unlock(&vcpu->kvm->srcu, idx);
1911
1912         return i;
1913 }
1914
1915 /*
1916  * Read or write a bunch of msrs. Parameters are user addresses.
1917  *
1918  * @return number of msrs set successfully.
1919  */
1920 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1921                   int (*do_msr)(struct kvm_vcpu *vcpu,
1922                                 unsigned index, u64 *data),
1923                   int writeback)
1924 {
1925         struct kvm_msrs msrs;
1926         struct kvm_msr_entry *entries;
1927         int r, n;
1928         unsigned size;
1929
1930         r = -EFAULT;
1931         if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1932                 goto out;
1933
1934         r = -E2BIG;
1935         if (msrs.nmsrs >= MAX_IO_MSRS)
1936                 goto out;
1937
1938         r = -ENOMEM;
1939         size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1940         entries = kmalloc(size, GFP_KERNEL);
1941         if (!entries)
1942                 goto out;
1943
1944         r = -EFAULT;
1945         if (copy_from_user(entries, user_msrs->entries, size))
1946                 goto out_free;
1947
1948         r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1949         if (r < 0)
1950                 goto out_free;
1951
1952         r = -EFAULT;
1953         if (writeback && copy_to_user(user_msrs->entries, entries, size))
1954                 goto out_free;
1955
1956         r = n;
1957
1958 out_free:
1959         kfree(entries);
1960 out:
1961         return r;
1962 }
1963
1964 int kvm_dev_ioctl_check_extension(long ext)
1965 {
1966         int r;
1967
1968         switch (ext) {
1969         case KVM_CAP_IRQCHIP:
1970         case KVM_CAP_HLT:
1971         case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1972         case KVM_CAP_SET_TSS_ADDR:
1973         case KVM_CAP_EXT_CPUID:
1974         case KVM_CAP_CLOCKSOURCE:
1975         case KVM_CAP_PIT:
1976         case KVM_CAP_NOP_IO_DELAY:
1977         case KVM_CAP_MP_STATE:
1978         case KVM_CAP_SYNC_MMU:
1979         case KVM_CAP_USER_NMI:
1980         case KVM_CAP_REINJECT_CONTROL:
1981         case KVM_CAP_IRQ_INJECT_STATUS:
1982         case KVM_CAP_ASSIGN_DEV_IRQ:
1983         case KVM_CAP_IRQFD:
1984         case KVM_CAP_IOEVENTFD:
1985         case KVM_CAP_PIT2:
1986         case KVM_CAP_PIT_STATE2:
1987         case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1988         case KVM_CAP_XEN_HVM:
1989         case KVM_CAP_ADJUST_CLOCK:
1990         case KVM_CAP_VCPU_EVENTS:
1991         case KVM_CAP_HYPERV:
1992         case KVM_CAP_HYPERV_VAPIC:
1993         case KVM_CAP_HYPERV_SPIN:
1994         case KVM_CAP_PCI_SEGMENT:
1995         case KVM_CAP_DEBUGREGS:
1996         case KVM_CAP_X86_ROBUST_SINGLESTEP:
1997         case KVM_CAP_XSAVE:
1998         case KVM_CAP_ASYNC_PF:
1999         case KVM_CAP_GET_TSC_KHZ:
2000                 r = 1;
2001                 break;
2002         case KVM_CAP_COALESCED_MMIO:
2003                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
2004                 break;
2005         case KVM_CAP_VAPIC:
2006                 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
2007                 break;
2008         case KVM_CAP_NR_VCPUS:
2009                 r = KVM_MAX_VCPUS;
2010                 break;
2011         case KVM_CAP_NR_MEMSLOTS:
2012                 r = KVM_MEMORY_SLOTS;
2013                 break;
2014         case KVM_CAP_PV_MMU:    /* obsolete */
2015                 r = 0;
2016                 break;
2017         case KVM_CAP_IOMMU:
2018                 r = iommu_found();
2019                 break;
2020         case KVM_CAP_MCE:
2021                 r = KVM_MAX_MCE_BANKS;
2022                 break;
2023         case KVM_CAP_XCRS:
2024                 r = cpu_has_xsave;
2025                 break;
2026         case KVM_CAP_TSC_CONTROL:
2027                 r = kvm_has_tsc_control;
2028                 break;
2029         default:
2030                 r = 0;
2031                 break;
2032         }
2033         return r;
2034
2035 }
2036
2037 long kvm_arch_dev_ioctl(struct file *filp,
2038                         unsigned int ioctl, unsigned long arg)
2039 {
2040         void __user *argp = (void __user *)arg;
2041         long r;
2042
2043         switch (ioctl) {
2044         case KVM_GET_MSR_INDEX_LIST: {
2045                 struct kvm_msr_list __user *user_msr_list = argp;
2046                 struct kvm_msr_list msr_list;
2047                 unsigned n;
2048
2049                 r = -EFAULT;
2050                 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
2051                         goto out;
2052                 n = msr_list.nmsrs;
2053                 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
2054                 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
2055                         goto out;
2056                 r = -E2BIG;
2057                 if (n < msr_list.nmsrs)
2058                         goto out;
2059                 r = -EFAULT;
2060                 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
2061                                  num_msrs_to_save * sizeof(u32)))
2062                         goto out;
2063                 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
2064                                  &emulated_msrs,
2065                                  ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
2066                         goto out;
2067                 r = 0;
2068                 break;
2069         }
2070         case KVM_GET_SUPPORTED_CPUID: {
2071                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2072                 struct kvm_cpuid2 cpuid;
2073
2074                 r = -EFAULT;
2075                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2076                         goto out;
2077                 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2078                                                       cpuid_arg->entries);
2079                 if (r)
2080                         goto out;
2081
2082                 r = -EFAULT;
2083                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2084                         goto out;
2085                 r = 0;
2086                 break;
2087         }
2088         case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2089                 u64 mce_cap;
2090
2091                 mce_cap = KVM_MCE_CAP_SUPPORTED;
2092                 r = -EFAULT;
2093                 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2094                         goto out;
2095                 r = 0;
2096                 break;
2097         }
2098         default:
2099                 r = -EINVAL;
2100         }
2101 out:
2102         return r;
2103 }
2104
2105 static void wbinvd_ipi(void *garbage)
2106 {
2107         wbinvd();
2108 }
2109
2110 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2111 {
2112         return vcpu->kvm->arch.iommu_domain &&
2113                 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2114 }
2115
2116 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2117 {
2118         /* Address WBINVD may be executed by guest */
2119         if (need_emulate_wbinvd(vcpu)) {
2120                 if (kvm_x86_ops->has_wbinvd_exit())
2121                         cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2122                 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2123                         smp_call_function_single(vcpu->cpu,
2124                                         wbinvd_ipi, NULL, 1);
2125         }
2126
2127         kvm_x86_ops->vcpu_load(vcpu, cpu);
2128         if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2129                 /* Make sure TSC doesn't go backwards */
2130                 s64 tsc_delta;
2131                 u64 tsc;
2132
2133                 kvm_get_msr(vcpu, MSR_IA32_TSC, &tsc);
2134                 tsc_delta = !vcpu->arch.last_guest_tsc ? 0 :
2135                              tsc - vcpu->arch.last_guest_tsc;
2136
2137                 if (tsc_delta < 0)
2138                         mark_tsc_unstable("KVM discovered backwards TSC");
2139                 if (check_tsc_unstable()) {
2140                         kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2141                         vcpu->arch.tsc_catchup = 1;
2142                 }
2143                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2144                 if (vcpu->cpu != cpu)
2145                         kvm_migrate_timers(vcpu);
2146                 vcpu->cpu = cpu;
2147         }
2148 }
2149
2150 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2151 {
2152         kvm_x86_ops->vcpu_put(vcpu);
2153         kvm_put_guest_fpu(vcpu);
2154         kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
2155 }
2156
2157 static int is_efer_nx(void)
2158 {
2159         unsigned long long efer = 0;
2160
2161         rdmsrl_safe(MSR_EFER, &efer);
2162         return efer & EFER_NX;
2163 }
2164
2165 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2166 {
2167         int i;
2168         struct kvm_cpuid_entry2 *e, *entry;
2169
2170         entry = NULL;
2171         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2172                 e = &vcpu->arch.cpuid_entries[i];
2173                 if (e->function == 0x80000001) {
2174                         entry = e;
2175                         break;
2176                 }
2177         }
2178         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2179                 entry->edx &= ~(1 << 20);
2180                 printk(KERN_INFO "kvm: guest NX capability removed\n");
2181         }
2182 }
2183
2184 /* when an old userspace process fills a new kernel module */
2185 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2186                                     struct kvm_cpuid *cpuid,
2187                                     struct kvm_cpuid_entry __user *entries)
2188 {
2189         int r, i;
2190         struct kvm_cpuid_entry *cpuid_entries;
2191
2192         r = -E2BIG;
2193         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2194                 goto out;
2195         r = -ENOMEM;
2196         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2197         if (!cpuid_entries)
2198                 goto out;
2199         r = -EFAULT;
2200         if (copy_from_user(cpuid_entries, entries,
2201                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2202                 goto out_free;
2203         for (i = 0; i < cpuid->nent; i++) {
2204                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2205                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2206                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2207                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2208                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2209                 vcpu->arch.cpuid_entries[i].index = 0;
2210                 vcpu->arch.cpuid_entries[i].flags = 0;
2211                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2212                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2213                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2214         }
2215         vcpu->arch.cpuid_nent = cpuid->nent;
2216         cpuid_fix_nx_cap(vcpu);
2217         r = 0;
2218         kvm_apic_set_version(vcpu);
2219         kvm_x86_ops->cpuid_update(vcpu);
2220         update_cpuid(vcpu);
2221
2222 out_free:
2223         vfree(cpuid_entries);
2224 out:
2225         return r;
2226 }
2227
2228 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2229                                      struct kvm_cpuid2 *cpuid,
2230                                      struct kvm_cpuid_entry2 __user *entries)
2231 {
2232         int r;
2233
2234         r = -E2BIG;
2235         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2236                 goto out;
2237         r = -EFAULT;
2238         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2239                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2240                 goto out;
2241         vcpu->arch.cpuid_nent = cpuid->nent;
2242         kvm_apic_set_version(vcpu);
2243         kvm_x86_ops->cpuid_update(vcpu);
2244         update_cpuid(vcpu);
2245         return 0;
2246
2247 out:
2248         return r;
2249 }
2250
2251 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2252                                      struct kvm_cpuid2 *cpuid,
2253                                      struct kvm_cpuid_entry2 __user *entries)
2254 {
2255         int r;
2256
2257         r = -E2BIG;
2258         if (cpuid->nent < vcpu->arch.cpuid_nent)
2259                 goto out;
2260         r = -EFAULT;
2261         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2262                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2263                 goto out;
2264         return 0;
2265
2266 out:
2267         cpuid->nent = vcpu->arch.cpuid_nent;
2268         return r;
2269 }
2270
2271 static void cpuid_mask(u32 *word, int wordnum)
2272 {
2273         *word &= boot_cpu_data.x86_capability[wordnum];
2274 }
2275
2276 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2277                            u32 index)
2278 {
2279         entry->function = function;
2280         entry->index = index;
2281         cpuid_count(entry->function, entry->index,
2282                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2283         entry->flags = 0;
2284 }
2285
2286 #define F(x) bit(X86_FEATURE_##x)
2287
2288 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2289                          u32 index, int *nent, int maxnent)
2290 {
2291         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2292 #ifdef CONFIG_X86_64
2293         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2294                                 ? F(GBPAGES) : 0;
2295         unsigned f_lm = F(LM);
2296 #else
2297         unsigned f_gbpages = 0;
2298         unsigned f_lm = 0;
2299 #endif
2300         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2301
2302         /* cpuid 1.edx */
2303         const u32 kvm_supported_word0_x86_features =
2304                 F(FPU) | F(VME) | F(DE) | F(PSE) |
2305                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2306                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2307                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2308                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2309                 0 /* Reserved, DS, ACPI */ | F(MMX) |
2310                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2311                 0 /* HTT, TM, Reserved, PBE */;
2312         /* cpuid 0x80000001.edx */
2313         const u32 kvm_supported_word1_x86_features =
2314                 F(FPU) | F(VME) | F(DE) | F(PSE) |
2315                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2316                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2317                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2318                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2319                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2320                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2321                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2322         /* cpuid 1.ecx */
2323         const u32 kvm_supported_word4_x86_features =
2324                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
2325                 0 /* DS-CPL, VMX, SMX, EST */ |
2326                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2327                 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2328                 0 /* Reserved, DCA */ | F(XMM4_1) |
2329                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2330                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2331                 F(F16C);
2332         /* cpuid 0x80000001.ecx */
2333         const u32 kvm_supported_word6_x86_features =
2334                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2335                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2336                 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2337                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2338
2339         /* cpuid 0xC0000001.edx */
2340         const u32 kvm_supported_word5_x86_features =
2341                 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
2342                 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
2343                 F(PMM) | F(PMM_EN);
2344
2345         /* all calls to cpuid_count() should be made on the same cpu */
2346         get_cpu();
2347         do_cpuid_1_ent(entry, function, index);
2348         ++*nent;
2349
2350         switch (function) {
2351         case 0:
2352                 entry->eax = min(entry->eax, (u32)0xd);
2353                 break;
2354         case 1:
2355                 entry->edx &= kvm_supported_word0_x86_features;
2356                 cpuid_mask(&entry->edx, 0);
2357                 entry->ecx &= kvm_supported_word4_x86_features;
2358                 cpuid_mask(&entry->ecx, 4);
2359                 /* we support x2apic emulation even if host does not support
2360                  * it since we emulate x2apic in software */
2361                 entry->ecx |= F(X2APIC);
2362                 break;
2363         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2364          * may return different values. This forces us to get_cpu() before
2365          * issuing the first command, and also to emulate this annoying behavior
2366          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2367         case 2: {
2368                 int t, times = entry->eax & 0xff;
2369
2370                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2371                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2372                 for (t = 1; t < times && *nent < maxnent; ++t) {
2373                         do_cpuid_1_ent(&entry[t], function, 0);
2374                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2375                         ++*nent;
2376                 }
2377                 break;
2378         }
2379         /* function 4 and 0xb have additional index. */
2380         case 4: {
2381                 int i, cache_type;
2382
2383                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2384                 /* read more entries until cache_type is zero */
2385                 for (i = 1; *nent < maxnent; ++i) {
2386                         cache_type = entry[i - 1].eax & 0x1f;
2387                         if (!cache_type)
2388                                 break;
2389                         do_cpuid_1_ent(&entry[i], function, i);
2390                         entry[i].flags |=
2391                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2392                         ++*nent;
2393                 }
2394                 break;
2395         }
2396         case 0xb: {
2397                 int i, level_type;
2398
2399                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2400                 /* read more entries until level_type is zero */
2401                 for (i = 1; *nent < maxnent; ++i) {
2402                         level_type = entry[i - 1].ecx & 0xff00;
2403                         if (!level_type)
2404                                 break;
2405                         do_cpuid_1_ent(&entry[i], function, i);
2406                         entry[i].flags |=
2407                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2408                         ++*nent;
2409                 }
2410                 break;
2411         }
2412         case 0xd: {
2413                 int i;
2414
2415                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2416                 for (i = 1; *nent < maxnent && i < 64; ++i) {
2417                         if (entry[i].eax == 0)
2418                                 continue;
2419                         do_cpuid_1_ent(&entry[i], function, i);
2420                         entry[i].flags |=
2421                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2422                         ++*nent;
2423                 }
2424                 break;
2425         }
2426         case KVM_CPUID_SIGNATURE: {
2427                 char signature[12] = "KVMKVMKVM\0\0";
2428                 u32 *sigptr = (u32 *)signature;
2429                 entry->eax = 0;
2430                 entry->ebx = sigptr[0];
2431                 entry->ecx = sigptr[1];
2432                 entry->edx = sigptr[2];
2433                 break;
2434         }
2435         case KVM_CPUID_FEATURES:
2436                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2437                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
2438                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
2439                              (1 << KVM_FEATURE_ASYNC_PF) |
2440                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2441                 entry->ebx = 0;
2442                 entry->ecx = 0;
2443                 entry->edx = 0;
2444                 break;
2445         case 0x80000000:
2446                 entry->eax = min(entry->eax, 0x8000001a);
2447                 break;
2448         case 0x80000001:
2449                 entry->edx &= kvm_supported_word1_x86_features;
2450                 cpuid_mask(&entry->edx, 1);
2451                 entry->ecx &= kvm_supported_word6_x86_features;
2452                 cpuid_mask(&entry->ecx, 6);
2453                 break;
2454         /*Add support for Centaur's CPUID instruction*/
2455         case 0xC0000000:
2456                 /*Just support up to 0xC0000004 now*/
2457                 entry->eax = min(entry->eax, 0xC0000004);
2458                 break;
2459         case 0xC0000001:
2460                 entry->edx &= kvm_supported_word5_x86_features;
2461                 cpuid_mask(&entry->edx, 5);
2462                 break;
2463         case 0xC0000002:
2464         case 0xC0000003:
2465         case 0xC0000004:
2466                 /*Now nothing to do, reserved for the future*/
2467                 break;
2468         }
2469
2470         kvm_x86_ops->set_supported_cpuid(function, entry);
2471
2472         put_cpu();
2473 }
2474
2475 #undef F
2476
2477 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2478                                      struct kvm_cpuid_entry2 __user *entries)
2479 {
2480         struct kvm_cpuid_entry2 *cpuid_entries;
2481         int limit, nent = 0, r = -E2BIG;
2482         u32 func;
2483
2484         if (cpuid->nent < 1)
2485                 goto out;
2486         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2487                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2488         r = -ENOMEM;
2489         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2490         if (!cpuid_entries)
2491                 goto out;
2492
2493         do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2494         limit = cpuid_entries[0].eax;
2495         for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2496                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2497                              &nent, cpuid->nent);
2498         r = -E2BIG;
2499         if (nent >= cpuid->nent)
2500                 goto out_free;
2501
2502         do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2503         limit = cpuid_entries[nent - 1].eax;
2504         for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2505                 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2506                              &nent, cpuid->nent);
2507
2508
2509
2510         r = -E2BIG;
2511         if (nent >= cpuid->nent)
2512                 goto out_free;
2513
2514         /* Add support for Centaur's CPUID instruction. */
2515         if (boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR) {
2516                 do_cpuid_ent(&cpuid_entries[nent], 0xC0000000, 0,
2517                                 &nent, cpuid->nent);
2518
2519                 r = -E2BIG;
2520                 if (nent >= cpuid->nent)
2521                         goto out_free;
2522
2523                 limit = cpuid_entries[nent - 1].eax;
2524                 for (func = 0xC0000001;
2525                         func <= limit && nent < cpuid->nent; ++func)
2526                         do_cpuid_ent(&cpuid_entries[nent], func, 0,
2527                                         &nent, cpuid->nent);
2528
2529                 r = -E2BIG;
2530                 if (nent >= cpuid->nent)
2531                         goto out_free;
2532         }
2533
2534         do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2535                      cpuid->nent);
2536
2537         r = -E2BIG;
2538         if (nent >= cpuid->nent)
2539                 goto out_free;
2540
2541         do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2542                      cpuid->nent);
2543
2544         r = -E2BIG;
2545         if (nent >= cpuid->nent)
2546                 goto out_free;
2547
2548         r = -EFAULT;
2549         if (copy_to_user(entries, cpuid_entries,
2550                          nent * sizeof(struct kvm_cpuid_entry2)))
2551                 goto out_free;
2552         cpuid->nent = nent;
2553         r = 0;
2554
2555 out_free:
2556         vfree(cpuid_entries);
2557 out:
2558         return r;
2559 }
2560
2561 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2562                                     struct kvm_lapic_state *s)
2563 {
2564         memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2565
2566         return 0;
2567 }
2568
2569 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2570                                     struct kvm_lapic_state *s)
2571 {
2572         memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2573         kvm_apic_post_state_restore(vcpu);
2574         update_cr8_intercept(vcpu);
2575
2576         return 0;
2577 }
2578
2579 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2580                                     struct kvm_interrupt *irq)
2581 {
2582         if (irq->irq < 0 || irq->irq >= 256)
2583                 return -EINVAL;
2584         if (irqchip_in_kernel(vcpu->kvm))
2585                 return -ENXIO;
2586
2587         kvm_queue_interrupt(vcpu, irq->irq, false);
2588         kvm_make_request(KVM_REQ_EVENT, vcpu);
2589
2590         return 0;
2591 }
2592
2593 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2594 {
2595         kvm_inject_nmi(vcpu);
2596
2597         return 0;
2598 }
2599
2600 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2601                                            struct kvm_tpr_access_ctl *tac)
2602 {
2603         if (tac->flags)
2604                 return -EINVAL;
2605         vcpu->arch.tpr_access_reporting = !!tac->enabled;
2606         return 0;
2607 }
2608
2609 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2610                                         u64 mcg_cap)
2611 {
2612         int r;
2613         unsigned bank_num = mcg_cap & 0xff, bank;
2614
2615         r = -EINVAL;
2616         if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2617                 goto out;
2618         if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2619                 goto out;
2620         r = 0;
2621         vcpu->arch.mcg_cap = mcg_cap;
2622         /* Init IA32_MCG_CTL to all 1s */
2623         if (mcg_cap & MCG_CTL_P)
2624                 vcpu->arch.mcg_ctl = ~(u64)0;
2625         /* Init IA32_MCi_CTL to all 1s */
2626         for (bank = 0; bank < bank_num; bank++)
2627                 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2628 out:
2629         return r;
2630 }
2631
2632 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2633                                       struct kvm_x86_mce *mce)
2634 {
2635         u64 mcg_cap = vcpu->arch.mcg_cap;
2636         unsigned bank_num = mcg_cap & 0xff;
2637         u64 *banks = vcpu->arch.mce_banks;
2638
2639         if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2640                 return -EINVAL;
2641         /*
2642          * if IA32_MCG_CTL is not all 1s, the uncorrected error
2643          * reporting is disabled
2644          */
2645         if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2646             vcpu->arch.mcg_ctl != ~(u64)0)
2647                 return 0;
2648         banks += 4 * mce->bank;
2649         /*
2650          * if IA32_MCi_CTL is not all 1s, the uncorrected error
2651          * reporting is disabled for the bank
2652          */
2653         if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2654                 return 0;
2655         if (mce->status & MCI_STATUS_UC) {
2656                 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2657                     !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2658                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2659                         return 0;
2660                 }
2661                 if (banks[1] & MCI_STATUS_VAL)
2662                         mce->status |= MCI_STATUS_OVER;
2663                 banks[2] = mce->addr;
2664                 banks[3] = mce->misc;
2665                 vcpu->arch.mcg_status = mce->mcg_status;
2666                 banks[1] = mce->status;
2667                 kvm_queue_exception(vcpu, MC_VECTOR);
2668         } else if (!(banks[1] & MCI_STATUS_VAL)
2669                    || !(banks[1] & MCI_STATUS_UC)) {
2670                 if (banks[1] & MCI_STATUS_VAL)
2671                         mce->status |= MCI_STATUS_OVER;
2672                 banks[2] = mce->addr;
2673                 banks[3] = mce->misc;
2674                 banks[1] = mce->status;
2675         } else
2676                 banks[1] |= MCI_STATUS_OVER;
2677         return 0;
2678 }
2679
2680 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2681                                                struct kvm_vcpu_events *events)
2682 {
2683         events->exception.injected =
2684                 vcpu->arch.exception.pending &&
2685                 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2686         events->exception.nr = vcpu->arch.exception.nr;
2687         events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2688         events->exception.pad = 0;
2689         events->exception.error_code = vcpu->arch.exception.error_code;
2690
2691         events->interrupt.injected =
2692                 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2693         events->interrupt.nr = vcpu->arch.interrupt.nr;
2694         events->interrupt.soft = 0;
2695         events->interrupt.shadow =
2696                 kvm_x86_ops->get_interrupt_shadow(vcpu,
2697                         KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2698
2699         events->nmi.injected = vcpu->arch.nmi_injected;
2700         events->nmi.pending = vcpu->arch.nmi_pending;
2701         events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2702         events->nmi.pad = 0;
2703
2704         events->sipi_vector = vcpu->arch.sipi_vector;
2705
2706         events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2707                          | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2708                          | KVM_VCPUEVENT_VALID_SHADOW);
2709         memset(&events->reserved, 0, sizeof(events->reserved));
2710 }
2711
2712 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2713                                               struct kvm_vcpu_events *events)
2714 {
2715         if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2716                               | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2717                               | KVM_VCPUEVENT_VALID_SHADOW))
2718                 return -EINVAL;
2719
2720         vcpu->arch.exception.pending = events->exception.injected;
2721         vcpu->arch.exception.nr = events->exception.nr;
2722         vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2723         vcpu->arch.exception.error_code = events->exception.error_code;
2724
2725         vcpu->arch.interrupt.pending = events->interrupt.injected;
2726         vcpu->arch.interrupt.nr = events->interrupt.nr;
2727         vcpu->arch.interrupt.soft = events->interrupt.soft;
2728         if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2729                 kvm_x86_ops->set_interrupt_shadow(vcpu,
2730                                                   events->interrupt.shadow);
2731
2732         vcpu->arch.nmi_injected = events->nmi.injected;
2733         if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2734                 vcpu->arch.nmi_pending = events->nmi.pending;
2735         kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2736
2737         if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2738                 vcpu->arch.sipi_vector = events->sipi_vector;
2739
2740         kvm_make_request(KVM_REQ_EVENT, vcpu);
2741
2742         return 0;
2743 }
2744
2745 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2746                                              struct kvm_debugregs *dbgregs)
2747 {
2748         memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2749         dbgregs->dr6 = vcpu->arch.dr6;
2750         dbgregs->dr7 = vcpu->arch.dr7;
2751         dbgregs->flags = 0;
2752         memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
2753 }
2754
2755 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2756                                             struct kvm_debugregs *dbgregs)
2757 {
2758         if (dbgregs->flags)
2759                 return -EINVAL;
2760
2761         memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2762         vcpu->arch.dr6 = dbgregs->dr6;
2763         vcpu->arch.dr7 = dbgregs->dr7;
2764
2765         return 0;
2766 }
2767
2768 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2769                                          struct kvm_xsave *guest_xsave)
2770 {
2771         if (cpu_has_xsave)
2772                 memcpy(guest_xsave->region,
2773                         &vcpu->arch.guest_fpu.state->xsave,
2774                         xstate_size);
2775         else {
2776                 memcpy(guest_xsave->region,
2777                         &vcpu->arch.guest_fpu.state->fxsave,
2778                         sizeof(struct i387_fxsave_struct));
2779                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2780                         XSTATE_FPSSE;
2781         }
2782 }
2783
2784 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2785                                         struct kvm_xsave *guest_xsave)
2786 {
2787         u64 xstate_bv =
2788                 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2789
2790         if (cpu_has_xsave)
2791                 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2792                         guest_xsave->region, xstate_size);
2793         else {
2794                 if (xstate_bv & ~XSTATE_FPSSE)
2795                         return -EINVAL;
2796                 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2797                         guest_xsave->region, sizeof(struct i387_fxsave_struct));
2798         }
2799         return 0;
2800 }
2801
2802 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2803                                         struct kvm_xcrs *guest_xcrs)
2804 {
2805         if (!cpu_has_xsave) {
2806                 guest_xcrs->nr_xcrs = 0;
2807                 return;
2808         }
2809
2810         guest_xcrs->nr_xcrs = 1;
2811         guest_xcrs->flags = 0;
2812         guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2813         guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2814 }
2815
2816 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2817                                        struct kvm_xcrs *guest_xcrs)
2818 {
2819         int i, r = 0;
2820
2821         if (!cpu_has_xsave)
2822                 return -EINVAL;
2823
2824         if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2825                 return -EINVAL;
2826
2827         for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2828                 /* Only support XCR0 currently */
2829                 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2830                         r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2831                                 guest_xcrs->xcrs[0].value);
2832                         break;
2833                 }
2834         if (r)
2835                 r = -EINVAL;
2836         return r;
2837 }
2838
2839 long kvm_arch_vcpu_ioctl(struct file *filp,
2840                          unsigned int ioctl, unsigned long arg)
2841 {
2842         struct kvm_vcpu *vcpu = filp->private_data;
2843         void __user *argp = (void __user *)arg;
2844         int r;
2845         union {
2846                 struct kvm_lapic_state *lapic;
2847                 struct kvm_xsave *xsave;
2848                 struct kvm_xcrs *xcrs;
2849                 void *buffer;
2850         } u;
2851
2852         u.buffer = NULL;
2853         switch (ioctl) {
2854         case KVM_GET_LAPIC: {
2855                 r = -EINVAL;
2856                 if (!vcpu->arch.apic)
2857                         goto out;
2858                 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2859
2860                 r = -ENOMEM;
2861                 if (!u.lapic)
2862                         goto out;
2863                 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
2864                 if (r)
2865                         goto out;
2866                 r = -EFAULT;
2867                 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
2868                         goto out;
2869                 r = 0;
2870                 break;
2871         }
2872         case KVM_SET_LAPIC: {
2873                 r = -EINVAL;
2874                 if (!vcpu->arch.apic)
2875                         goto out;
2876                 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2877                 r = -ENOMEM;
2878                 if (!u.lapic)
2879                         goto out;
2880                 r = -EFAULT;
2881                 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
2882                         goto out;
2883                 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
2884                 if (r)
2885                         goto out;
2886                 r = 0;
2887                 break;
2888         }
2889         case KVM_INTERRUPT: {
2890                 struct kvm_interrupt irq;
2891
2892                 r = -EFAULT;
2893                 if (copy_from_user(&irq, argp, sizeof irq))
2894                         goto out;
2895                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2896                 if (r)
2897                         goto out;
2898                 r = 0;
2899                 break;
2900         }
2901         case KVM_NMI: {
2902                 r = kvm_vcpu_ioctl_nmi(vcpu);
2903                 if (r)
2904                         goto out;
2905                 r = 0;
2906                 break;
2907         }
2908         case KVM_SET_CPUID: {
2909                 struct kvm_cpuid __user *cpuid_arg = argp;
2910                 struct kvm_cpuid cpuid;
2911
2912                 r = -EFAULT;
2913                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2914                         goto out;
2915                 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2916                 if (r)
2917                         goto out;
2918                 break;
2919         }
2920         case KVM_SET_CPUID2: {
2921                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2922                 struct kvm_cpuid2 cpuid;
2923
2924                 r = -EFAULT;
2925                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2926                         goto out;
2927                 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2928                                               cpuid_arg->entries);
2929                 if (r)
2930                         goto out;
2931                 break;
2932         }
2933         case KVM_GET_CPUID2: {
2934                 struct kvm_cpuid2 __user *cpuid_arg = argp;
2935                 struct kvm_cpuid2 cpuid;
2936
2937                 r = -EFAULT;
2938                 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2939                         goto out;
2940                 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2941                                               cpuid_arg->entries);
2942                 if (r)
2943                         goto out;
2944                 r = -EFAULT;
2945                 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2946                         goto out;
2947                 r = 0;
2948                 break;
2949         }
2950         case KVM_GET_MSRS:
2951                 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2952                 break;
2953         case KVM_SET_MSRS:
2954                 r = msr_io(vcpu, argp, do_set_msr, 0);
2955                 break;
2956         case KVM_TPR_ACCESS_REPORTING: {
2957                 struct kvm_tpr_access_ctl tac;
2958
2959                 r = -EFAULT;
2960                 if (copy_from_user(&tac, argp, sizeof tac))
2961                         goto out;
2962                 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2963                 if (r)
2964                         goto out;
2965                 r = -EFAULT;
2966                 if (copy_to_user(argp, &tac, sizeof tac))
2967                         goto out;
2968                 r = 0;
2969                 break;
2970         };
2971         case KVM_SET_VAPIC_ADDR: {
2972                 struct kvm_vapic_addr va;
2973
2974                 r = -EINVAL;
2975                 if (!irqchip_in_kernel(vcpu->kvm))
2976                         goto out;
2977                 r = -EFAULT;
2978                 if (copy_from_user(&va, argp, sizeof va))
2979                         goto out;
2980                 r = 0;
2981                 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2982                 break;
2983         }
2984         case KVM_X86_SETUP_MCE: {
2985                 u64 mcg_cap;
2986
2987                 r = -EFAULT;
2988                 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2989                         goto out;
2990                 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2991                 break;
2992         }
2993         case KVM_X86_SET_MCE: {
2994                 struct kvm_x86_mce mce;
2995
2996                 r = -EFAULT;
2997                 if (copy_from_user(&mce, argp, sizeof mce))
2998                         goto out;
2999                 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
3000                 break;
3001         }
3002         case KVM_GET_VCPU_EVENTS: {
3003                 struct kvm_vcpu_events events;
3004
3005                 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
3006
3007                 r = -EFAULT;
3008                 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
3009                         break;
3010                 r = 0;
3011                 break;
3012         }
3013         case KVM_SET_VCPU_EVENTS: {
3014                 struct kvm_vcpu_events events;
3015
3016                 r = -EFAULT;
3017                 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
3018                         break;
3019
3020                 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
3021                 break;
3022         }
3023         case KVM_GET_DEBUGREGS: {
3024                 struct kvm_debugregs dbgregs;
3025
3026                 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
3027
3028                 r = -EFAULT;
3029                 if (copy_to_user(argp, &dbgregs,
3030                                  sizeof(struct kvm_debugregs)))
3031                         break;
3032                 r = 0;
3033                 break;
3034         }
3035         case KVM_SET_DEBUGREGS: {
3036                 struct kvm_debugregs dbgregs;
3037
3038                 r = -EFAULT;
3039                 if (copy_from_user(&dbgregs, argp,
3040                                    sizeof(struct kvm_debugregs)))
3041                         break;
3042
3043                 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
3044                 break;
3045         }
3046         case KVM_GET_XSAVE: {
3047                 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3048                 r = -ENOMEM;
3049                 if (!u.xsave)
3050                         break;
3051
3052                 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
3053
3054                 r = -EFAULT;
3055                 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
3056                         break;
3057                 r = 0;
3058                 break;
3059         }
3060         case KVM_SET_XSAVE: {
3061                 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
3062                 r = -ENOMEM;
3063                 if (!u.xsave)
3064                         break;
3065
3066                 r = -EFAULT;
3067                 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
3068                         break;
3069
3070                 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3071                 break;
3072         }
3073         case KVM_GET_XCRS: {
3074                 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3075                 r = -ENOMEM;
3076                 if (!u.xcrs)
3077                         break;
3078
3079                 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
3080
3081                 r = -EFAULT;
3082                 if (copy_to_user(argp, u.xcrs,
3083                                  sizeof(struct kvm_xcrs)))
3084                         break;
3085                 r = 0;
3086                 break;
3087         }
3088         case KVM_SET_XCRS: {
3089                 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
3090                 r = -ENOMEM;
3091                 if (!u.xcrs)
3092                         break;
3093
3094                 r = -EFAULT;
3095                 if (copy_from_user(u.xcrs, argp,
3096                                    sizeof(struct kvm_xcrs)))
3097                         break;
3098
3099                 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3100                 break;
3101         }
3102         case KVM_SET_TSC_KHZ: {
3103                 u32 user_tsc_khz;
3104
3105                 r = -EINVAL;
3106                 if (!kvm_has_tsc_control)
3107                         break;
3108
3109                 user_tsc_khz = (u32)arg;
3110
3111                 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
3112                         goto out;
3113
3114                 kvm_x86_ops->set_tsc_khz(vcpu, user_tsc_khz);
3115
3116                 r = 0;
3117                 goto out;
3118         }
3119         case KVM_GET_TSC_KHZ: {
3120                 r = -EIO;
3121                 if (check_tsc_unstable())
3122                         goto out;
3123
3124                 r = vcpu_tsc_khz(vcpu);
3125
3126                 goto out;
3127         }
3128         default:
3129                 r = -EINVAL;
3130         }
3131 out:
3132         kfree(u.buffer);
3133         return r;
3134 }
3135
3136 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
3137 {
3138         int ret;
3139
3140         if (addr > (unsigned int)(-3 * PAGE_SIZE))
3141                 return -1;
3142         ret = kvm_x86_ops->set_tss_addr(kvm, addr);
3143         return ret;
3144 }
3145
3146 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
3147                                               u64 ident_addr)
3148 {
3149         kvm->arch.ept_identity_map_addr = ident_addr;
3150         return 0;
3151 }
3152
3153 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3154                                           u32 kvm_nr_mmu_pages)
3155 {
3156         if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3157                 return -EINVAL;
3158
3159         mutex_lock(&kvm->slots_lock);
3160         spin_lock(&kvm->mmu_lock);
3161
3162         kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3163         kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3164
3165         spin_unlock(&kvm->mmu_lock);
3166         mutex_unlock(&kvm->slots_lock);
3167         return 0;
3168 }
3169
3170 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3171 {
3172         return kvm->arch.n_max_mmu_pages;
3173 }
3174
3175 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3176 {
3177         int r;
3178
3179         r = 0;
3180         switch (chip->chip_id) {
3181         case KVM_IRQCHIP_PIC_MASTER:
3182                 memcpy(&chip->chip.pic,
3183                         &pic_irqchip(kvm)->pics[0],
3184                         sizeof(struct kvm_pic_state));
3185                 break;
3186         case KVM_IRQCHIP_PIC_SLAVE:
3187                 memcpy(&chip->chip.pic,
3188                         &pic_irqchip(kvm)->pics[1],
3189                         sizeof(struct kvm_pic_state));
3190                 break;
3191         case KVM_IRQCHIP_IOAPIC:
3192                 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3193                 break;
3194         default:
3195                 r = -EINVAL;
3196                 break;
3197         }
3198         return r;
3199 }
3200
3201 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3202 {
3203         int r;
3204
3205         r = 0;
3206         switch (chip->chip_id) {
3207         case KVM_IRQCHIP_PIC_MASTER:
3208                 spin_lock(&pic_irqchip(kvm)->lock);
3209                 memcpy(&pic_irqchip(kvm)->pics[0],
3210                         &chip->chip.pic,
3211                         sizeof(struct kvm_pic_state));
3212                 spin_unlock(&pic_irqchip(kvm)->lock);
3213                 break;
3214         case KVM_IRQCHIP_PIC_SLAVE:
3215                 spin_lock(&pic_irqchip(kvm)->lock);
3216                 memcpy(&pic_irqchip(kvm)->pics[1],
3217                         &chip->chip.pic,
3218                         sizeof(struct kvm_pic_state));
3219                 spin_unlock(&pic_irqchip(kvm)->lock);
3220                 break;
3221         case KVM_IRQCHIP_IOAPIC:
3222                 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3223                 break;
3224         default:
3225                 r = -EINVAL;
3226                 break;
3227         }
3228         kvm_pic_update_irq(pic_irqchip(kvm));
3229         return r;
3230 }
3231
3232 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3233 {
3234         int r = 0;
3235
3236         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3237         memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3238         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3239         return r;
3240 }
3241
3242 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3243 {
3244         int r = 0;
3245
3246         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3247         memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3248         kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3249         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3250         return r;
3251 }
3252
3253 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3254 {
3255         int r = 0;
3256
3257         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3258         memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3259                 sizeof(ps->channels));
3260         ps->flags = kvm->arch.vpit->pit_state.flags;
3261         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3262         memset(&ps->reserved, 0, sizeof(ps->reserved));
3263         return r;
3264 }
3265
3266 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3267 {
3268         int r = 0, start = 0;
3269         u32 prev_legacy, cur_legacy;
3270         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3271         prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3272         cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3273         if (!prev_legacy && cur_legacy)
3274                 start = 1;
3275         memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3276                sizeof(kvm->arch.vpit->pit_state.channels));
3277         kvm->arch.vpit->pit_state.flags = ps->flags;
3278         kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3279         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3280         return r;
3281 }
3282
3283 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3284                                  struct kvm_reinject_control *control)
3285 {
3286         if (!kvm->arch.vpit)
3287                 return -ENXIO;
3288         mutex_lock(&kvm->arch.vpit->pit_state.lock);
3289         kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3290         mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3291         return 0;
3292 }
3293
3294 /*
3295  * Get (and clear) the dirty memory log for a memory slot.
3296  */
3297 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3298                                       struct kvm_dirty_log *log)
3299 {
3300         int r, i;
3301         struct kvm_memory_slot *memslot;
3302         unsigned long n;
3303         unsigned long is_dirty = 0;
3304
3305         mutex_lock(&kvm->slots_lock);
3306
3307         r = -EINVAL;
3308         if (log->slot >= KVM_MEMORY_SLOTS)
3309                 goto out;
3310
3311         memslot = &kvm->memslots->memslots[log->slot];
3312         r = -ENOENT;
3313         if (!memslot->dirty_bitmap)
3314                 goto out;
3315
3316         n = kvm_dirty_bitmap_bytes(memslot);
3317
3318         for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3319                 is_dirty = memslot->dirty_bitmap[i];
3320
3321         /* If nothing is dirty, don't bother messing with page tables. */
3322         if (is_dirty) {
3323                 struct kvm_memslots *slots, *old_slots;
3324                 unsigned long *dirty_bitmap;
3325
3326                 dirty_bitmap = memslot->dirty_bitmap_head;
3327                 if (memslot->dirty_bitmap == dirty_bitmap)
3328                         dirty_bitmap += n / sizeof(long);
3329                 memset(dirty_bitmap, 0, n);
3330
3331                 r = -ENOMEM;
3332                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3333                 if (!slots)
3334                         goto out;
3335                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3336                 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3337                 slots->generation++;
3338
3339                 old_slots = kvm->memslots;
3340                 rcu_assign_pointer(kvm->memslots, slots);
3341                 synchronize_srcu_expedited(&kvm->srcu);
3342                 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3343                 kfree(old_slots);
3344
3345                 spin_lock(&kvm->mmu_lock);
3346                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3347                 spin_unlock(&kvm->mmu_lock);
3348
3349                 r = -EFAULT;
3350                 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
3351                         goto out;
3352         } else {
3353                 r = -EFAULT;
3354                 if (clear_user(log->dirty_bitmap, n))
3355                         goto out;
3356         }
3357
3358         r = 0;
3359 out:
3360         mutex_unlock(&kvm->slots_lock);
3361         return r;
3362 }
3363
3364 long kvm_arch_vm_ioctl(struct file *filp,
3365                        unsigned int ioctl, unsigned long arg)
3366 {
3367         struct kvm *kvm = filp->private_data;
3368         void __user *argp = (void __user *)arg;
3369         int r = -ENOTTY;
3370         /*
3371          * This union makes it completely explicit to gcc-3.x
3372          * that these two variables' stack usage should be
3373          * combined, not added together.
3374          */
3375         union {
3376                 struct kvm_pit_state ps;
3377                 struct kvm_pit_state2 ps2;
3378                 struct kvm_pit_config pit_config;
3379         } u;
3380
3381         switch (ioctl) {
3382         case KVM_SET_TSS_ADDR:
3383                 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3384                 if (r < 0)
3385                         goto out;
3386                 break;
3387         case KVM_SET_IDENTITY_MAP_ADDR: {
3388                 u64 ident_addr;
3389
3390                 r = -EFAULT;
3391                 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3392                         goto out;
3393                 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3394                 if (r < 0)
3395                         goto out;
3396                 break;
3397         }
3398         case KVM_SET_NR_MMU_PAGES:
3399                 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3400                 if (r)
3401                         goto out;
3402                 break;
3403         case KVM_GET_NR_MMU_PAGES:
3404                 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3405                 break;
3406         case KVM_CREATE_IRQCHIP: {
3407                 struct kvm_pic *vpic;
3408
3409                 mutex_lock(&kvm->lock);
3410                 r = -EEXIST;
3411                 if (kvm->arch.vpic)
3412                         goto create_irqchip_unlock;
3413                 r = -ENOMEM;
3414                 vpic = kvm_create_pic(kvm);
3415                 if (vpic) {
3416                         r = kvm_ioapic_init(kvm);
3417                         if (r) {
3418                                 mutex_lock(&kvm->slots_lock);
3419                                 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3420                                                           &vpic->dev);
3421                                 mutex_unlock(&kvm->slots_lock);
3422                                 kfree(vpic);
3423                                 goto create_irqchip_unlock;
3424                         }
3425                 } else
3426                         goto create_irqchip_unlock;
3427                 smp_wmb();
3428                 kvm->arch.vpic = vpic;
3429                 smp_wmb();
3430                 r = kvm_setup_default_irq_routing(kvm);
3431                 if (r) {
3432                         mutex_lock(&kvm->slots_lock);
3433                         mutex_lock(&kvm->irq_lock);
3434                         kvm_ioapic_destroy(kvm);
3435                         kvm_destroy_pic(kvm);
3436                         mutex_unlock(&kvm->irq_lock);
3437                         mutex_unlock(&kvm->slots_lock);
3438                 }
3439         create_irqchip_unlock:
3440                 mutex_unlock(&kvm->lock);
3441                 break;
3442         }
3443         case KVM_CREATE_PIT:
3444                 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3445                 goto create_pit;
3446         case KVM_CREATE_PIT2:
3447                 r = -EFAULT;
3448                 if (copy_from_user(&u.pit_config, argp,
3449                                    sizeof(struct kvm_pit_config)))
3450                         goto out;
3451         create_pit:
3452                 mutex_lock(&kvm->slots_lock);
3453                 r = -EEXIST;
3454                 if (kvm->arch.vpit)
3455                         goto create_pit_unlock;
3456                 r = -ENOMEM;
3457                 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3458                 if (kvm->arch.vpit)
3459                         r = 0;
3460         create_pit_unlock:
3461                 mutex_unlock(&kvm->slots_lock);
3462                 break;
3463         case KVM_IRQ_LINE_STATUS:
3464         case KVM_IRQ_LINE: {
3465                 struct kvm_irq_level irq_event;
3466
3467                 r = -EFAULT;
3468                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3469                         goto out;
3470                 r = -ENXIO;
3471                 if (irqchip_in_kernel(kvm)) {
3472                         __s32 status;
3473                         status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3474                                         irq_event.irq, irq_event.level);
3475                         if (ioctl == KVM_IRQ_LINE_STATUS) {
3476                                 r = -EFAULT;
3477                                 irq_event.status = status;
3478                                 if (copy_to_user(argp, &irq_event,
3479                                                         sizeof irq_event))
3480                                         goto out;
3481                         }
3482                         r = 0;
3483                 }
3484                 break;
3485         }
3486         case KVM_GET_IRQCHIP: {
3487                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3488                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3489
3490                 r = -ENOMEM;
3491                 if (!chip)
3492                         goto out;
3493                 r = -EFAULT;
3494                 if (copy_from_user(chip, argp, sizeof *chip))
3495                         goto get_irqchip_out;
3496                 r = -ENXIO;
3497                 if (!irqchip_in_kernel(kvm))
3498                         goto get_irqchip_out;
3499                 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3500                 if (r)
3501                         goto get_irqchip_out;
3502                 r = -EFAULT;
3503                 if (copy_to_user(argp, chip, sizeof *chip))
3504                         goto get_irqchip_out;
3505                 r = 0;
3506         get_irqchip_out:
3507                 kfree(chip);
3508                 if (r)
3509                         goto out;
3510                 break;
3511         }
3512         case KVM_SET_IRQCHIP: {
3513                 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3514                 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3515
3516                 r = -ENOMEM;
3517                 if (!chip)
3518                         goto out;
3519                 r = -EFAULT;
3520                 if (copy_from_user(chip, argp, sizeof *chip))
3521                         goto set_irqchip_out;
3522                 r = -ENXIO;
3523                 if (!irqchip_in_kernel(kvm))
3524                         goto set_irqchip_out;
3525                 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3526                 if (r)
3527                         goto set_irqchip_out;
3528                 r = 0;
3529         set_irqchip_out:
3530                 kfree(chip);
3531                 if (r)
3532                         goto out;
3533                 break;
3534         }
3535         case KVM_GET_PIT: {
3536                 r = -EFAULT;
3537                 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3538                         goto out;
3539                 r = -ENXIO;
3540                 if (!kvm->arch.vpit)
3541                         goto out;
3542                 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3543                 if (r)
3544                         goto out;
3545                 r = -EFAULT;
3546                 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3547                         goto out;
3548                 r = 0;
3549                 break;
3550         }
3551         case KVM_SET_PIT: {
3552                 r = -EFAULT;
3553                 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3554                         goto out;
3555                 r = -ENXIO;
3556                 if (!kvm->arch.vpit)
3557                         goto out;
3558                 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3559                 if (r)
3560                         goto out;
3561                 r = 0;
3562                 break;
3563         }
3564         case KVM_GET_PIT2: {
3565                 r = -ENXIO;
3566                 if (!kvm->arch.vpit)
3567                         goto out;
3568                 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3569                 if (r)
3570                         goto out;
3571                 r = -EFAULT;
3572                 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3573                         goto out;
3574                 r = 0;
3575                 break;
3576         }
3577         case KVM_SET_PIT2: {
3578                 r = -EFAULT;
3579                 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3580                         goto out;
3581                 r = -ENXIO;
3582                 if (!kvm->arch.vpit)
3583                         goto out;
3584                 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3585                 if (r)
3586                         goto out;
3587                 r = 0;
3588                 break;
3589         }
3590         case KVM_REINJECT_CONTROL: {
3591                 struct kvm_reinject_control control;
3592                 r =  -EFAULT;
3593                 if (copy_from_user(&control, argp, sizeof(control)))
3594                         goto out;
3595                 r = kvm_vm_ioctl_reinject(kvm, &control);
3596                 if (r)
3597                         goto out;
3598                 r = 0;
3599                 break;
3600         }
3601         case KVM_XEN_HVM_CONFIG: {
3602                 r = -EFAULT;
3603                 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3604                                    sizeof(struct kvm_xen_hvm_config)))
3605                         goto out;
3606                 r = -EINVAL;
3607                 if (kvm->arch.xen_hvm_config.flags)
3608                         goto out;
3609                 r = 0;
3610                 break;
3611         }
3612         case KVM_SET_CLOCK: {
3613                 struct kvm_clock_data user_ns;
3614                 u64 now_ns;
3615                 s64 delta;
3616
3617                 r = -EFAULT;
3618                 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3619                         goto out;
3620
3621                 r = -EINVAL;
3622                 if (user_ns.flags)
3623                         goto out;
3624
3625                 r = 0;
3626                 local_irq_disable();
3627                 now_ns = get_kernel_ns();
3628                 delta = user_ns.clock - now_ns;
3629                 local_irq_enable();
3630                 kvm->arch.kvmclock_offset = delta;
3631                 break;
3632         }
3633         case KVM_GET_CLOCK: {
3634                 struct kvm_clock_data user_ns;
3635                 u64 now_ns;
3636
3637                 local_irq_disable();
3638                 now_ns = get_kernel_ns();
3639                 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3640                 local_irq_enable();
3641                 user_ns.flags = 0;
3642                 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
3643
3644                 r = -EFAULT;
3645                 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3646                         goto out;
3647                 r = 0;
3648                 break;
3649         }
3650
3651         default:
3652                 ;
3653         }
3654 out:
3655         return r;
3656 }
3657
3658 static void kvm_init_msr_list(void)
3659 {
3660         u32 dummy[2];
3661         unsigned i, j;
3662
3663         /* skip the first msrs in the list. KVM-specific */
3664         for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3665                 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3666                         continue;
3667                 if (j < i)
3668                         msrs_to_save[j] = msrs_to_save[i];
3669                 j++;
3670         }
3671         num_msrs_to_save = j;
3672 }
3673
3674 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3675                            const void *v)
3676 {
3677         int handled = 0;
3678         int n;
3679
3680         do {
3681                 n = min(len, 8);
3682                 if (!(vcpu->arch.apic &&
3683                       !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, n, v))
3684                     && kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3685                         break;
3686                 handled += n;
3687                 addr += n;
3688                 len -= n;
3689                 v += n;
3690         } while (len);
3691
3692         return handled;
3693 }
3694
3695 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3696 {
3697         int handled = 0;
3698         int n;
3699
3700         do {
3701                 n = min(len, 8);
3702                 if (!(vcpu->arch.apic &&
3703                       !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, n, v))
3704                     && kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, n, v))
3705                         break;
3706                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, *(u64 *)v);
3707                 handled += n;
3708                 addr += n;
3709                 len -= n;
3710                 v += n;
3711         } while (len);
3712
3713         return handled;
3714 }
3715
3716 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3717                         struct kvm_segment *var, int seg)
3718 {
3719         kvm_x86_ops->set_segment(vcpu, var, seg);
3720 }
3721
3722 void kvm_get_segment(struct kvm_vcpu *vcpu,
3723                      struct kvm_segment *var, int seg)
3724 {
3725         kvm_x86_ops->get_segment(vcpu, var, seg);
3726 }
3727
3728 static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3729 {
3730         return gpa;
3731 }
3732
3733 static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3734 {
3735         gpa_t t_gpa;
3736         struct x86_exception exception;
3737
3738         BUG_ON(!mmu_is_nested(vcpu));
3739
3740         /* NPT walks are always user-walks */
3741         access |= PFERR_USER_MASK;
3742         t_gpa  = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &exception);
3743
3744         return t_gpa;
3745 }
3746
3747 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
3748                               struct x86_exception *exception)
3749 {
3750         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3751         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3752 }
3753
3754  gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
3755                                 struct x86_exception *exception)
3756 {
3757         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3758         access |= PFERR_FETCH_MASK;
3759         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3760 }
3761
3762 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
3763                                struct x86_exception *exception)
3764 {
3765         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3766         access |= PFERR_WRITE_MASK;
3767         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
3768 }
3769
3770 /* uses this to access any guest's mapped memory without checking CPL */
3771 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
3772                                 struct x86_exception *exception)
3773 {
3774         return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
3775 }
3776
3777 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3778                                       struct kvm_vcpu *vcpu, u32 access,
3779                                       struct x86_exception *exception)
3780 {
3781         void *data = val;
3782         int r = X86EMUL_CONTINUE;
3783
3784         while (bytes) {
3785                 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3786                                                             exception);
3787                 unsigned offset = addr & (PAGE_SIZE-1);
3788                 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3789                 int ret;
3790
3791                 if (gpa == UNMAPPED_GVA)
3792                         return X86EMUL_PROPAGATE_FAULT;
3793                 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3794                 if (ret < 0) {
3795                         r = X86EMUL_IO_NEEDED;
3796                         goto out;
3797                 }
3798
3799                 bytes -= toread;
3800                 data += toread;
3801                 addr += toread;
3802         }
3803 out:
3804         return r;
3805 }
3806
3807 /* used for instruction fetching */
3808 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
3809                                 gva_t addr, void *val, unsigned int bytes,
3810                                 struct x86_exception *exception)
3811 {
3812         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3813         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3814
3815         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3816                                           access | PFERR_FETCH_MASK,
3817                                           exception);
3818 }
3819
3820 static int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
3821                                gva_t addr, void *val, unsigned int bytes,
3822                                struct x86_exception *exception)
3823 {
3824         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3825         u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3826
3827         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3828                                           exception);
3829 }
3830
3831 static int kvm_read_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3832                                       gva_t addr, void *val, unsigned int bytes,
3833                                       struct x86_exception *exception)
3834 {
3835         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3836         return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, exception);
3837 }
3838
3839 static int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
3840                                        gva_t addr, void *val,
3841                                        unsigned int bytes,
3842                                        struct x86_exception *exception)
3843 {
3844         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3845         void *data = val;
3846         int r = X86EMUL_CONTINUE;
3847
3848         while (bytes) {
3849                 gpa_t gpa =  vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3850                                                              PFERR_WRITE_MASK,
3851                                                              exception);
3852                 unsigned offset = addr & (PAGE_SIZE-1);
3853                 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3854                 int ret;
3855
3856                 if (gpa == UNMAPPED_GVA)
3857                         return X86EMUL_PROPAGATE_FAULT;
3858                 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3859                 if (ret < 0) {
3860                         r = X86EMUL_IO_NEEDED;
3861                         goto out;
3862                 }
3863
3864                 bytes -= towrite;
3865                 data += towrite;
3866                 addr += towrite;
3867         }
3868 out:
3869         return r;
3870 }
3871
3872 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
3873                                   unsigned long addr,
3874                                   void *val,
3875                                   unsigned int bytes,
3876                                   struct x86_exception *exception)
3877 {
3878         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3879         gpa_t                 gpa;
3880         int handled;
3881
3882         if (vcpu->mmio_read_completed) {
3883                 memcpy(val, vcpu->mmio_data, bytes);
3884                 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3885                                vcpu->mmio_phys_addr, *(u64 *)val);
3886                 vcpu->mmio_read_completed = 0;
3887                 return X86EMUL_CONTINUE;
3888         }
3889
3890         gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, exception);
3891
3892         if (gpa == UNMAPPED_GVA)
3893                 return X86EMUL_PROPAGATE_FAULT;
3894
3895         /* For APIC access vmexit */
3896         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3897                 goto mmio;
3898
3899         if (kvm_read_guest_virt(ctxt, addr, val, bytes, exception)
3900             == X86EMUL_CONTINUE)
3901                 return X86EMUL_CONTINUE;
3902
3903 mmio:
3904         /*
3905          * Is this MMIO handled locally?
3906          */
3907         handled = vcpu_mmio_read(vcpu, gpa, bytes, val);
3908
3909         if (handled == bytes)
3910                 return X86EMUL_CONTINUE;
3911
3912         gpa += handled;
3913         bytes -= handled;
3914         val += handled;
3915
3916         trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3917
3918         vcpu->mmio_needed = 1;
3919         vcpu->run->exit_reason = KVM_EXIT_MMIO;
3920         vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3921         vcpu->mmio_size = bytes;
3922         vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
3923         vcpu->run->mmio.is_write = vcpu->mmio_is_write = 0;
3924         vcpu->mmio_index = 0;
3925
3926         return X86EMUL_IO_NEEDED;
3927 }
3928
3929 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3930                         const void *val, int bytes)
3931 {
3932         int ret;
3933
3934         ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3935         if (ret < 0)
3936                 return 0;
3937         kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3938         return 1;
3939 }
3940
3941 static int emulator_write_emulated_onepage(unsigned long addr,
3942                                            const void *val,
3943                                            unsigned int bytes,
3944                                            struct x86_exception *exception,
3945                                            struct kvm_vcpu *vcpu)
3946 {
3947         gpa_t                 gpa;
3948         int handled;
3949
3950         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, exception);
3951
3952         if (gpa == UNMAPPED_GVA)
3953                 return X86EMUL_PROPAGATE_FAULT;
3954
3955         /* For APIC access vmexit */
3956         if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3957                 goto mmio;
3958
3959         if (emulator_write_phys(vcpu, gpa, val, bytes))
3960                 return X86EMUL_CONTINUE;
3961
3962 mmio:
3963         trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3964         /*
3965          * Is this MMIO handled locally?
3966          */
3967         handled = vcpu_mmio_write(vcpu, gpa, bytes, val);
3968         if (handled == bytes)
3969                 return X86EMUL_CONTINUE;
3970
3971         gpa += handled;
3972         bytes -= handled;
3973         val += handled;
3974
3975         vcpu->mmio_needed = 1;
3976         memcpy(vcpu->mmio_data, val, bytes);
3977         vcpu->run->exit_reason = KVM_EXIT_MMIO;
3978         vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3979         vcpu->mmio_size = bytes;
3980         vcpu->run->mmio.len = min(vcpu->mmio_size, 8);
3981         vcpu->run->mmio.is_write = vcpu->mmio_is_write = 1;
3982         memcpy(vcpu->run->mmio.data, vcpu->mmio_data, 8);
3983         vcpu->mmio_index = 0;
3984
3985         return X86EMUL_CONTINUE;
3986 }
3987
3988 int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
3989                             unsigned long addr,
3990                             const void *val,
3991                             unsigned int bytes,
3992                             struct x86_exception *exception)
3993 {
3994         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3995
3996         /* Crossing a page boundary? */
3997         if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3998                 int rc, now;
3999
4000                 now = -addr & ~PAGE_MASK;
4001                 rc = emulator_write_emulated_onepage(addr, val, now, exception,
4002                                                      vcpu);
4003                 if (rc != X86EMUL_CONTINUE)
4004                         return rc;
4005                 addr += now;
4006                 val += now;
4007                 bytes -= now;
4008         }
4009         return emulator_write_emulated_onepage(addr, val, bytes, exception,
4010                                                vcpu);
4011 }
4012
4013 #define CMPXCHG_TYPE(t, ptr, old, new) \
4014         (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
4015
4016 #ifdef CONFIG_X86_64
4017 #  define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
4018 #else
4019 #  define CMPXCHG64(ptr, old, new) \
4020         (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
4021 #endif
4022
4023 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4024                                      unsigned long addr,
4025                                      const void *old,
4026                                      const void *new,
4027                                      unsigned int bytes,
4028                                      struct x86_exception *exception)
4029 {
4030         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4031         gpa_t gpa;
4032         struct page *page;
4033         char *kaddr;
4034         bool exchanged;
4035
4036         /* guests cmpxchg8b have to be emulated atomically */
4037         if (bytes > 8 || (bytes & (bytes - 1)))
4038                 goto emul_write;
4039
4040         gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
4041
4042         if (gpa == UNMAPPED_GVA ||
4043             (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
4044                 goto emul_write;
4045
4046         if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
4047                 goto emul_write;
4048
4049         page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4050         if (is_error_page(page)) {
4051                 kvm_release_page_clean(page);
4052                 goto emul_write;
4053         }
4054
4055         kaddr = kmap_atomic(page, KM_USER0);
4056         kaddr += offset_in_page(gpa);
4057         switch (bytes) {
4058         case 1:
4059                 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
4060                 break;
4061         case 2:
4062                 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
4063                 break;
4064         case 4:
4065                 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
4066                 break;
4067         case 8:
4068                 exchanged = CMPXCHG64(kaddr, old, new);
4069                 break;
4070         default:
4071                 BUG();
4072         }
4073         kunmap_atomic(kaddr, KM_USER0);
4074         kvm_release_page_dirty(page);
4075
4076         if (!exchanged)
4077                 return X86EMUL_CMPXCHG_FAILED;
4078
4079         kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
4080
4081         return X86EMUL_CONTINUE;
4082
4083 emul_write:
4084         printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
4085
4086         return emulator_write_emulated(ctxt, addr, new, bytes, exception);
4087 }
4088
4089 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4090 {
4091         /* TODO: String I/O for in kernel device */
4092         int r;
4093
4094         if (vcpu->arch.pio.in)
4095                 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
4096                                     vcpu->arch.pio.size, pd);
4097         else
4098                 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
4099                                      vcpu->arch.pio.port, vcpu->arch.pio.size,
4100                                      pd);
4101         return r;
4102 }
4103
4104
4105 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4106                                     int size, unsigned short port, void *val,
4107                                     unsigned int count)
4108 {
4109         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4110
4111         if (vcpu->arch.pio.count)
4112                 goto data_avail;
4113
4114         trace_kvm_pio(0, port, size, count);
4115
4116         vcpu->arch.pio.port = port;
4117         vcpu->arch.pio.in = 1;
4118         vcpu->arch.pio.count  = count;
4119         vcpu->arch.pio.size = size;
4120
4121         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4122         data_avail:
4123                 memcpy(val, vcpu->arch.pio_data, size * count);
4124                 vcpu->arch.pio.count = 0;
4125                 return 1;
4126         }
4127
4128         vcpu->run->exit_reason = KVM_EXIT_IO;
4129         vcpu->run->io.direction = KVM_EXIT_IO_IN;
4130         vcpu->run->io.size = size;
4131         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4132         vcpu->run->io.count = count;
4133         vcpu->run->io.port = port;
4134
4135         return 0;
4136 }
4137
4138 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
4139                                      int size, unsigned short port,
4140                                      const void *val, unsigned int count)
4141 {
4142         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4143
4144         trace_kvm_pio(1, port, size, count);
4145
4146         vcpu->arch.pio.port = port;
4147         vcpu->arch.pio.in = 0;
4148         vcpu->arch.pio.count = count;
4149         vcpu->arch.pio.size = size;
4150
4151         memcpy(vcpu->arch.pio_data, val, size * count);
4152
4153         if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4154                 vcpu->arch.pio.count = 0;
4155                 return 1;
4156         }
4157
4158         vcpu->run->exit_reason = KVM_EXIT_IO;
4159         vcpu->run->io.direction = KVM_EXIT_IO_OUT;
4160         vcpu->run->io.size = size;
4161         vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4162         vcpu->run->io.count = count;
4163         vcpu->run->io.port = port;
4164
4165         return 0;
4166 }
4167
4168 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4169 {
4170         return kvm_x86_ops->get_segment_base(vcpu, seg);
4171 }
4172
4173 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
4174 {
4175         kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
4176 }
4177
4178 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
4179 {
4180         if (!need_emulate_wbinvd(vcpu))
4181                 return X86EMUL_CONTINUE;
4182
4183         if (kvm_x86_ops->has_wbinvd_exit()) {
4184                 int cpu = get_cpu();
4185
4186                 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4187                 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
4188                                 wbinvd_ipi, NULL, 1);
4189                 put_cpu();
4190                 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
4191         } else
4192                 wbinvd();
4193         return X86EMUL_CONTINUE;
4194 }
4195 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
4196
4197 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
4198 {
4199         kvm_emulate_wbinvd(emul_to_vcpu(ctxt));
4200 }
4201
4202 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
4203 {
4204         return _kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
4205 }
4206
4207 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
4208 {
4209
4210         return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
4211 }
4212
4213 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4214 {
4215         return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4216 }
4217
4218 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
4219 {
4220         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4221         unsigned long value;
4222
4223         switch (cr) {
4224         case 0:
4225                 value = kvm_read_cr0(vcpu);
4226                 break;
4227         case 2:
4228                 value = vcpu->arch.cr2;
4229                 break;
4230         case 3:
4231                 value = kvm_read_cr3(vcpu);
4232                 break;
4233         case 4:
4234                 value = kvm_read_cr4(vcpu);
4235                 break;
4236         case 8:
4237                 value = kvm_get_cr8(vcpu);
4238                 break;
4239         default:
4240                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4241                 return 0;
4242         }
4243
4244         return value;
4245 }
4246
4247 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
4248 {
4249         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4250         int res = 0;
4251
4252         switch (cr) {
4253         case 0:
4254                 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4255                 break;
4256         case 2:
4257                 vcpu->arch.cr2 = val;
4258                 break;
4259         case 3:
4260                 res = kvm_set_cr3(vcpu, val);
4261                 break;
4262         case 4:
4263                 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4264                 break;
4265         case 8:
4266                 res = kvm_set_cr8(vcpu, val);
4267                 break;
4268         default:
4269                 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4270                 res = -1;
4271         }
4272
4273         return res;
4274 }
4275
4276 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
4277 {
4278         return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
4279 }
4280
4281 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4282 {
4283         kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
4284 }
4285
4286 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4287 {
4288         kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
4289 }
4290
4291 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4292 {
4293         kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
4294 }
4295
4296 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
4297 {
4298         kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
4299 }
4300
4301 static unsigned long emulator_get_cached_segment_base(
4302         struct x86_emulate_ctxt *ctxt, int seg)
4303 {
4304         return get_segment_base(emul_to_vcpu(ctxt), seg);
4305 }
4306
4307 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
4308                                  struct desc_struct *desc, u32 *base3,
4309                                  int seg)
4310 {
4311         struct kvm_segment var;
4312
4313         kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
4314         *selector = var.selector;
4315
4316         if (var.unusable)
4317                 return false;
4318
4319         if (var.g)
4320                 var.limit >>= 12;
4321         set_desc_limit(desc, var.limit);
4322         set_desc_base(desc, (unsigned long)var.base);
4323 #ifdef CONFIG_X86_64
4324         if (base3)
4325                 *base3 = var.base >> 32;
4326 #endif
4327         desc->type = var.type;
4328         desc->s = var.s;
4329         desc->dpl = var.dpl;
4330         desc->p = var.present;
4331         desc->avl = var.avl;
4332         desc->l = var.l;
4333         desc->d = var.db;
4334         desc->g = var.g;
4335
4336         return true;
4337 }
4338
4339 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
4340                                  struct desc_struct *desc, u32 base3,
4341                                  int seg)
4342 {
4343         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4344         struct kvm_segment var;
4345
4346         var.selector = selector;
4347         var.base = get_desc_base(desc);
4348 #ifdef CONFIG_X86_64
4349         var.base |= ((u64)base3) << 32;
4350 #endif
4351         var.limit = get_desc_limit(desc);
4352         if (desc->g)
4353                 var.limit = (var.limit << 12) | 0xfff;
4354         var.type = desc->type;
4355         var.present = desc->p;
4356         var.dpl = desc->dpl;
4357         var.db = desc->d;
4358         var.s = desc->s;
4359         var.l = desc->l;
4360         var.g = desc->g;
4361         var.avl = desc->avl;
4362         var.present = desc->p;
4363         var.unusable = !var.present;
4364         var.padding = 0;
4365
4366         kvm_set_segment(vcpu, &var, seg);
4367         return;
4368 }
4369
4370 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
4371                             u32 msr_index, u64 *pdata)
4372 {
4373         return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
4374 }
4375
4376 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4377                             u32 msr_index, u64 data)
4378 {
4379         return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
4380 }
4381
4382 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4383 {
4384         emul_to_vcpu(ctxt)->arch.halt_request = 1;
4385 }
4386
4387 static void emulator_get_fpu(struct x86_emulate_ctxt *ctxt)
4388 {
4389         preempt_disable();
4390         kvm_load_guest_fpu(emul_to_vcpu(ctxt));
4391         /*
4392          * CR0.TS may reference the host fpu state, not the guest fpu state,
4393          * so it may be clear at this point.
4394          */
4395         clts();
4396 }
4397
4398 static void emulator_put_fpu(struct x86_emulate_ctxt *ctxt)
4399 {
4400         preempt_enable();
4401 }
4402
4403 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
4404                               struct x86_instruction_info *info,
4405                               enum x86_intercept_stage stage)
4406 {
4407         return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
4408 }
4409
4410 static struct x86_emulate_ops emulate_ops = {
4411         .read_std            = kvm_read_guest_virt_system,
4412         .write_std           = kvm_write_guest_virt_system,
4413         .fetch               = kvm_fetch_guest_virt,
4414         .read_emulated       = emulator_read_emulated,
4415         .write_emulated      = emulator_write_emulated,
4416         .cmpxchg_emulated    = emulator_cmpxchg_emulated,
4417         .invlpg              = emulator_invlpg,
4418         .pio_in_emulated     = emulator_pio_in_emulated,
4419         .pio_out_emulated    = emulator_pio_out_emulated,
4420         .get_segment         = emulator_get_segment,
4421         .set_segment         = emulator_set_segment,
4422         .get_cached_segment_base = emulator_get_cached_segment_base,
4423         .get_gdt             = emulator_get_gdt,
4424         .get_idt             = emulator_get_idt,
4425         .set_gdt             = emulator_set_gdt,
4426         .set_idt             = emulator_set_idt,
4427         .get_cr              = emulator_get_cr,
4428         .set_cr              = emulator_set_cr,
4429         .cpl                 = emulator_get_cpl,
4430         .get_dr              = emulator_get_dr,
4431         .set_dr              = emulator_set_dr,
4432         .set_msr             = emulator_set_msr,
4433         .get_msr             = emulator_get_msr,
4434         .halt                = emulator_halt,
4435         .wbinvd              = emulator_wbinvd,
4436         .fix_hypercall       = emulator_fix_hypercall,
4437         .get_fpu             = emulator_get_fpu,
4438         .put_fpu             = emulator_put_fpu,
4439         .intercept           = emulator_intercept,
4440 };
4441
4442 static void cache_all_regs(struct kvm_vcpu *vcpu)
4443 {
4444         kvm_register_read(vcpu, VCPU_REGS_RAX);
4445         kvm_register_read(vcpu, VCPU_REGS_RSP);
4446         kvm_register_read(vcpu, VCPU_REGS_RIP);
4447         vcpu->arch.regs_dirty = ~0;
4448 }
4449
4450 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4451 {
4452         u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4453         /*
4454          * an sti; sti; sequence only disable interrupts for the first
4455          * instruction. So, if the last instruction, be it emulated or
4456          * not, left the system with the INT_STI flag enabled, it
4457          * means that the last instruction is an sti. We should not
4458          * leave the flag on in this case. The same goes for mov ss
4459          */
4460         if (!(int_shadow & mask))
4461                 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4462 }
4463
4464 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4465 {
4466         struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4467         if (ctxt->exception.vector == PF_VECTOR)
4468                 kvm_propagate_fault(vcpu, &ctxt->exception);
4469         else if (ctxt->exception.error_code_valid)
4470                 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
4471                                       ctxt->exception.error_code);
4472         else
4473                 kvm_queue_exception(vcpu, ctxt->exception.vector);
4474 }
4475
4476 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4477 {
4478         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4479         int cs_db, cs_l;
4480
4481         /*
4482          * TODO: fix emulate.c to use guest_read/write_register
4483          * instead of direct ->regs accesses, can save hundred cycles
4484          * on Intel for instructions that don't read/change RSP, for
4485          * for example.
4486          */
4487         cache_all_regs(vcpu);
4488
4489         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4490
4491         vcpu->arch.emulate_ctxt.eflags = kvm_get_rflags(vcpu);
4492         vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
4493         vcpu->arch.emulate_ctxt.mode =
4494                 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4495                 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
4496                 ? X86EMUL_MODE_VM86 : cs_l
4497                 ? X86EMUL_MODE_PROT64 : cs_db
4498                 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
4499         vcpu->arch.emulate_ctxt.guest_mode = is_guest_mode(vcpu);
4500         memset(c, 0, sizeof(struct decode_cache));
4501         memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4502         vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4503 }
4504
4505 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
4506 {
4507         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4508         int ret;
4509
4510         init_emulate_ctxt(vcpu);
4511
4512         vcpu->arch.emulate_ctxt.decode.op_bytes = 2;
4513         vcpu->arch.emulate_ctxt.decode.ad_bytes = 2;
4514         vcpu->arch.emulate_ctxt.decode.eip = vcpu->arch.emulate_ctxt.eip +
4515                                                                  inc_eip;
4516         ret = emulate_int_real(&vcpu->arch.emulate_ctxt, irq);
4517
4518         if (ret != X86EMUL_CONTINUE)
4519                 return EMULATE_FAIL;
4520
4521         vcpu->arch.emulate_ctxt.eip = c->eip;
4522         memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4523         kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4524         kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4525
4526         if (irq == NMI_VECTOR)
4527                 vcpu->arch.nmi_pending = false;
4528         else
4529                 vcpu->arch.interrupt.pending = false;
4530
4531         return EMULATE_DONE;
4532 }
4533 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4534
4535 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4536 {
4537         int r = EMULATE_DONE;
4538
4539         ++vcpu->stat.insn_emulation_fail;
4540         trace_kvm_emulate_insn_failed(vcpu);
4541         if (!is_guest_mode(vcpu)) {
4542                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4543                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4544                 vcpu->run->internal.ndata = 0;
4545                 r = EMULATE_FAIL;
4546         }
4547         kvm_queue_exception(vcpu, UD_VECTOR);
4548
4549         return r;
4550 }
4551
4552 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4553 {
4554         gpa_t gpa;
4555
4556         if (tdp_enabled)
4557                 return false;
4558
4559         /*
4560          * if emulation was due to access to shadowed page table
4561          * and it failed try to unshadow page and re-entetr the
4562          * guest to let CPU execute the instruction.
4563          */
4564         if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4565                 return true;
4566
4567         gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4568
4569         if (gpa == UNMAPPED_GVA)
4570                 return true; /* let cpu generate fault */
4571
4572         if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4573                 return true;
4574
4575         return false;
4576 }
4577
4578 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4579                             unsigned long cr2,
4580                             int emulation_type,
4581                             void *insn,
4582                             int insn_len)
4583 {
4584         int r;
4585         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4586         bool writeback = true;
4587
4588         kvm_clear_exception_queue(vcpu);
4589
4590         if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4591                 init_emulate_ctxt(vcpu);
4592                 vcpu->arch.emulate_ctxt.interruptibility = 0;
4593                 vcpu->arch.emulate_ctxt.have_exception = false;
4594                 vcpu->arch.emulate_ctxt.perm_ok = false;
4595
4596                 vcpu->arch.emulate_ctxt.only_vendor_specific_insn
4597                         = emulation_type & EMULTYPE_TRAP_UD;
4598
4599                 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, insn, insn_len);
4600
4601                 trace_kvm_emulate_insn_start(vcpu);
4602                 ++vcpu->stat.insn_emulation;
4603                 if (r)  {
4604                         if (emulation_type & EMULTYPE_TRAP_UD)
4605                                 return EMULATE_FAIL;
4606                         if (reexecute_instruction(vcpu, cr2))
4607                                 return EMULATE_DONE;
4608                         if (emulation_type & EMULTYPE_SKIP)
4609                                 return EMULATE_FAIL;
4610                         return handle_emulation_failure(vcpu);
4611                 }
4612         }
4613
4614         if (emulation_type & EMULTYPE_SKIP) {
4615                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
4616                 return EMULATE_DONE;
4617         }
4618
4619         /* this is needed for vmware backdoor interface to work since it
4620            changes registers values  during IO operation */
4621         if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
4622                 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
4623                 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4624         }
4625
4626 restart:
4627         r = x86_emulate_insn(&vcpu->arch.emulate_ctxt);
4628
4629         if (r == EMULATION_INTERCEPTED)
4630                 return EMULATE_DONE;
4631
4632         if (r == EMULATION_FAILED) {
4633                 if (reexecute_instruction(vcpu, cr2))
4634                         return EMULATE_DONE;
4635
4636                 return handle_emulation_failure(vcpu);
4637         }
4638
4639         if (vcpu->arch.emulate_ctxt.have_exception) {
4640                 inject_emulated_exception(vcpu);
4641                 r = EMULATE_DONE;
4642         } else if (vcpu->arch.pio.count) {
4643                 if (!vcpu->arch.pio.in)
4644                         vcpu->arch.pio.count = 0;
4645                 else
4646                         writeback = false;
4647                 r = EMULATE_DO_MMIO;
4648         } else if (vcpu->mmio_needed) {
4649                 if (!vcpu->mmio_is_write)
4650                         writeback = false;
4651                 r = EMULATE_DO_MMIO;
4652         } else if (r == EMULATION_RESTART)
4653                 goto restart;
4654         else
4655                 r = EMULATE_DONE;
4656
4657         if (writeback) {
4658                 toggle_interruptibility(vcpu,
4659                                 vcpu->arch.emulate_ctxt.interruptibility);
4660                 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4661                 kvm_make_request(KVM_REQ_EVENT, vcpu);
4662                 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4663                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
4664                 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4665         } else
4666                 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
4667
4668         return r;
4669 }
4670 EXPORT_SYMBOL_GPL(x86_emulate_instruction);
4671
4672 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4673 {
4674         unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4675         int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
4676                                             size, port, &val, 1);
4677         /* do not return to emulator after return from userspace */
4678         vcpu->arch.pio.count = 0;
4679         return ret;
4680 }
4681 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4682
4683 static void tsc_bad(void *info)
4684 {
4685         __this_cpu_write(cpu_tsc_khz, 0);
4686 }
4687
4688 static void tsc_khz_changed(void *data)
4689 {
4690         struct cpufreq_freqs *freq = data;
4691         unsigned long khz = 0;
4692
4693         if (data)
4694                 khz = freq->new;
4695         else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4696                 khz = cpufreq_quick_get(raw_smp_processor_id());
4697         if (!khz)
4698                 khz = tsc_khz;
4699         __this_cpu_write(cpu_tsc_khz, khz);
4700 }
4701
4702 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4703                                      void *data)
4704 {
4705         struct cpufreq_freqs *freq = data;
4706         struct kvm *kvm;
4707         struct kvm_vcpu *vcpu;
4708         int i, send_ipi = 0;
4709
4710         /*
4711          * We allow guests to temporarily run on slowing clocks,
4712          * provided we notify them after, or to run on accelerating
4713          * clocks, provided we notify them before.  Thus time never
4714          * goes backwards.
4715          *
4716          * However, we have a problem.  We can't atomically update
4717          * the frequency of a given CPU from this function; it is
4718          * merely a notifier, which can be called from any CPU.
4719          * Changing the TSC frequency at arbitrary points in time
4720          * requires a recomputation of local variables related to
4721          * the TSC for each VCPU.  We must flag these local variables
4722          * to be updated and be sure the update takes place with the
4723          * new frequency before any guests proceed.
4724          *
4725          * Unfortunately, the combination of hotplug CPU and frequency
4726          * change creates an intractable locking scenario; the order
4727          * of when these callouts happen is undefined with respect to
4728          * CPU hotplug, and they can race with each other.  As such,
4729          * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4730          * undefined; you can actually have a CPU frequency change take
4731          * place in between the computation of X and the setting of the
4732          * variable.  To protect against this problem, all updates of
4733          * the per_cpu tsc_khz variable are done in an interrupt
4734          * protected IPI, and all callers wishing to update the value
4735          * must wait for a synchronous IPI to complete (which is trivial
4736          * if the caller is on the CPU already).  This establishes the
4737          * necessary total order on variable updates.
4738          *
4739          * Note that because a guest time update may take place
4740          * anytime after the setting of the VCPU's request bit, the
4741          * correct TSC value must be set before the request.  However,
4742          * to ensure the update actually makes it to any guest which
4743          * starts running in hardware virtualization between the set
4744          * and the acquisition of the spinlock, we must also ping the
4745          * CPU after setting the request bit.
4746          *
4747          */
4748
4749         if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4750                 return 0;
4751         if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4752                 return 0;
4753
4754         smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4755
4756         raw_spin_lock(&kvm_lock);
4757         list_for_each_entry(kvm, &vm_list, vm_list) {
4758                 kvm_for_each_vcpu(i, vcpu, kvm) {
4759                         if (vcpu->cpu != freq->cpu)
4760                                 continue;
4761                         kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4762                         if (vcpu->cpu != smp_processor_id())
4763                                 send_ipi = 1;
4764                 }
4765         }
4766         raw_spin_unlock(&kvm_lock);
4767
4768         if (freq->old < freq->new && send_ipi) {
4769                 /*
4770                  * We upscale the frequency.  Must make the guest
4771                  * doesn't see old kvmclock values while running with
4772                  * the new frequency, otherwise we risk the guest sees
4773                  * time go backwards.
4774                  *
4775                  * In case we update the frequency for another cpu
4776                  * (which might be in guest context) send an interrupt
4777                  * to kick the cpu out of guest context.  Next time
4778                  * guest context is entered kvmclock will be updated,
4779                  * so the guest will not see stale values.
4780                  */
4781                 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4782         }
4783         return 0;
4784 }
4785
4786 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4787         .notifier_call  = kvmclock_cpufreq_notifier
4788 };
4789
4790 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4791                                         unsigned long action, void *hcpu)
4792 {
4793         unsigned int cpu = (unsigned long)hcpu;
4794
4795         switch (action) {
4796                 case CPU_ONLINE:
4797                 case CPU_DOWN_FAILED:
4798                         smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4799                         break;
4800                 case CPU_DOWN_PREPARE:
4801                         smp_call_function_single(cpu, tsc_bad, NULL, 1);
4802                         break;
4803         }
4804         return NOTIFY_OK;
4805 }
4806
4807 static struct notifier_block kvmclock_cpu_notifier_block = {
4808         .notifier_call  = kvmclock_cpu_notifier,
4809         .priority = -INT_MAX
4810 };
4811
4812 static void kvm_timer_init(void)
4813 {
4814         int cpu;
4815
4816         max_tsc_khz = tsc_khz;
4817         register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4818         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4819 #ifdef CONFIG_CPU_FREQ
4820                 struct cpufreq_policy policy;
4821                 memset(&policy, 0, sizeof(policy));
4822                 cpu = get_cpu();
4823                 cpufreq_get_policy(&policy, cpu);
4824                 if (policy.cpuinfo.max_freq)
4825                         max_tsc_khz = policy.cpuinfo.max_freq;
4826                 put_cpu();
4827 #endif
4828                 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4829                                           CPUFREQ_TRANSITION_NOTIFIER);
4830         }
4831         pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
4832         for_each_online_cpu(cpu)
4833                 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4834 }
4835
4836 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4837
4838 static int kvm_is_in_guest(void)
4839 {
4840         return percpu_read(current_vcpu) != NULL;
4841 }
4842
4843 static int kvm_is_user_mode(void)
4844 {
4845         int user_mode = 3;
4846
4847         if (percpu_read(current_vcpu))
4848                 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
4849
4850         return user_mode != 0;
4851 }
4852
4853 static unsigned long kvm_get_guest_ip(void)
4854 {
4855         unsigned long ip = 0;
4856
4857         if (percpu_read(current_vcpu))
4858                 ip = kvm_rip_read(percpu_read(current_vcpu));
4859
4860         return ip;
4861 }
4862
4863 static struct perf_guest_info_callbacks kvm_guest_cbs = {
4864         .is_in_guest            = kvm_is_in_guest,
4865         .is_user_mode           = kvm_is_user_mode,
4866         .get_guest_ip           = kvm_get_guest_ip,
4867 };
4868
4869 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4870 {
4871         percpu_write(current_vcpu, vcpu);
4872 }
4873 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4874
4875 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4876 {
4877         percpu_write(current_vcpu, NULL);
4878 }
4879 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4880
4881 int kvm_arch_init(void *opaque)
4882 {
4883         int r;
4884         struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4885
4886         if (kvm_x86_ops) {
4887                 printk(KERN_ERR "kvm: already loaded the other module\n");
4888                 r = -EEXIST;
4889                 goto out;
4890         }
4891
4892         if (!ops->cpu_has_kvm_support()) {
4893                 printk(KERN_ERR "kvm: no hardware support\n");
4894                 r = -EOPNOTSUPP;
4895                 goto out;
4896         }
4897         if (ops->disabled_by_bios()) {
4898                 printk(KERN_ERR "kvm: disabled by bios\n");
4899                 r = -EOPNOTSUPP;
4900                 goto out;
4901         }
4902
4903         r = kvm_mmu_module_init();
4904         if (r)
4905                 goto out;
4906
4907         kvm_init_msr_list();
4908
4909         kvm_x86_ops = ops;
4910         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
4911         kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4912                         PT_DIRTY_MASK, PT64_NX_MASK, 0);
4913
4914         kvm_timer_init();
4915
4916         perf_register_guest_info_callbacks(&kvm_guest_cbs);
4917
4918         if (cpu_has_xsave)
4919                 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
4920
4921         return 0;
4922
4923 out:
4924         return r;
4925 }
4926
4927 void kvm_arch_exit(void)
4928 {
4929         perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4930
4931         if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4932                 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4933                                             CPUFREQ_TRANSITION_NOTIFIER);
4934         unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4935         kvm_x86_ops = NULL;
4936         kvm_mmu_module_exit();
4937 }
4938
4939 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4940 {
4941         ++vcpu->stat.halt_exits;
4942         if (irqchip_in_kernel(vcpu->kvm)) {
4943                 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
4944                 return 1;
4945         } else {
4946                 vcpu->run->exit_reason = KVM_EXIT_HLT;
4947                 return 0;
4948         }
4949 }
4950 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4951
4952 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
4953                            unsigned long a1)
4954 {
4955         if (is_long_mode(vcpu))
4956                 return a0;
4957         else
4958                 return a0 | ((gpa_t)a1 << 32);
4959 }
4960
4961 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4962 {
4963         u64 param, ingpa, outgpa, ret;
4964         uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4965         bool fast, longmode;
4966         int cs_db, cs_l;
4967
4968         /*
4969          * hypercall generates UD from non zero cpl and real mode
4970          * per HYPER-V spec
4971          */
4972         if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
4973                 kvm_queue_exception(vcpu, UD_VECTOR);
4974                 return 0;
4975         }
4976
4977         kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4978         longmode = is_long_mode(vcpu) && cs_l == 1;
4979
4980         if (!longmode) {
4981                 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4982                         (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4983                 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4984                         (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4985                 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4986                         (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
4987         }
4988 #ifdef CONFIG_X86_64
4989         else {
4990                 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4991                 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4992                 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4993         }
4994 #endif
4995
4996         code = param & 0xffff;
4997         fast = (param >> 16) & 0x1;
4998         rep_cnt = (param >> 32) & 0xfff;
4999         rep_idx = (param >> 48) & 0xfff;
5000
5001         trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
5002
5003         switch (code) {
5004         case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
5005                 kvm_vcpu_on_spin(vcpu);
5006                 break;
5007         default:
5008                 res = HV_STATUS_INVALID_HYPERCALL_CODE;
5009                 break;
5010         }
5011
5012         ret = res | (((u64)rep_done & 0xfff) << 32);
5013         if (longmode) {
5014                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5015         } else {
5016                 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
5017                 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
5018         }
5019
5020         return 1;
5021 }
5022
5023 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5024 {
5025         unsigned long nr, a0, a1, a2, a3, ret;
5026         int r = 1;
5027
5028         if (kvm_hv_hypercall_enabled(vcpu->kvm))
5029                 return kvm_hv_hypercall(vcpu);
5030
5031         nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
5032         a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
5033         a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
5034         a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
5035         a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
5036
5037         trace_kvm_hypercall(nr, a0, a1, a2, a3);
5038
5039         if (!is_long_mode(vcpu)) {
5040                 nr &= 0xFFFFFFFF;
5041                 a0 &= 0xFFFFFFFF;
5042                 a1 &= 0xFFFFFFFF;
5043                 a2 &= 0xFFFFFFFF;
5044                 a3 &= 0xFFFFFFFF;
5045         }
5046
5047         if (kvm_x86_ops->get_cpl(vcpu) != 0) {
5048                 ret = -KVM_EPERM;
5049                 goto out;
5050         }
5051
5052         switch (nr) {
5053         case KVM_HC_VAPIC_POLL_IRQ:
5054                 ret = 0;
5055                 break;
5056         case KVM_HC_MMU_OP:
5057                 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
5058                 break;
5059         default:
5060                 ret = -KVM_ENOSYS;
5061                 break;
5062         }
5063 out:
5064         kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
5065         ++vcpu->stat.hypercalls;
5066         return r;
5067 }
5068 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
5069
5070 int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5071 {
5072         struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5073         char instruction[3];
5074         unsigned long rip = kvm_rip_read(vcpu);
5075
5076         /*
5077          * Blow out the MMU to ensure that no other VCPU has an active mapping
5078          * to ensure that the updated hypercall appears atomically across all
5079          * VCPUs.
5080          */
5081         kvm_mmu_zap_all(vcpu->kvm);
5082
5083         kvm_x86_ops->patch_hypercall(vcpu, instruction);
5084
5085         return emulator_write_emulated(&vcpu->arch.emulate_ctxt,
5086                                        rip, instruction, 3, NULL);
5087 }
5088
5089 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
5090 {
5091         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
5092         int j, nent = vcpu->arch.cpuid_nent;
5093
5094         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
5095         /* when no next entry is found, the current entry[i] is reselected */
5096         for (j = i + 1; ; j = (j + 1) % nent) {
5097                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
5098                 if (ej->function == e->function) {
5099                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
5100                         return j;
5101                 }
5102         }
5103         return 0; /* silence gcc, even though control never reaches here */
5104 }
5105
5106 /* find an entry with matching function, matching index (if needed), and that
5107  * should be read next (if it's stateful) */
5108 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
5109         u32 function, u32 index)
5110 {
5111         if (e->function != function)
5112                 return 0;
5113         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
5114                 return 0;
5115         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
5116             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
5117                 return 0;
5118         return 1;
5119 }
5120
5121 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
5122                                               u32 function, u32 index)
5123 {
5124         int i;
5125         struct kvm_cpuid_entry2 *best = NULL;
5126
5127         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
5128                 struct kvm_cpuid_entry2 *e;
5129
5130                 e = &vcpu->arch.cpuid_entries[i];
5131                 if (is_matching_cpuid_entry(e, function, index)) {
5132                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
5133                                 move_to_next_stateful_cpuid_entry(vcpu, i);
5134                         best = e;
5135                         break;
5136                 }
5137         }
5138         return best;
5139 }
5140 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
5141
5142 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
5143 {
5144         struct kvm_cpuid_entry2 *best;
5145
5146         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
5147         if (!best || best->eax < 0x80000008)
5148                 goto not_found;
5149         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
5150         if (best)
5151                 return best->eax & 0xff;
5152 not_found:
5153         return 36;
5154 }
5155
5156 /*
5157  * If no match is found, check whether we exceed the vCPU's limit
5158  * and return the content of the highest valid _standard_ leaf instead.
5159  * This is to satisfy the CPUID specification.
5160  */
5161 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
5162                                                   u32 function, u32 index)
5163 {
5164         struct kvm_cpuid_entry2 *maxlevel;
5165
5166         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
5167         if (!maxlevel || maxlevel->eax >= function)
5168                 return NULL;
5169         if (function & 0x80000000) {
5170                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
5171                 if (!maxlevel)
5172                         return NULL;
5173         }
5174         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
5175 }
5176
5177 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
5178 {
5179         u32 function, index;
5180         struct kvm_cpuid_entry2 *best;
5181
5182         function = kvm_register_read(vcpu, VCPU_REGS_RAX);
5183         index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5184         kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
5185         kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
5186         kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
5187         kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
5188         best = kvm_find_cpuid_entry(vcpu, function, index);
5189
5190         if (!best)
5191                 best = check_cpuid_limit(vcpu, function, index);
5192
5193         if (best) {
5194                 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
5195                 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
5196                 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
5197                 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
5198         }
5199         kvm_x86_ops->skip_emulated_instruction(vcpu);
5200         trace_kvm_cpuid(function,
5201                         kvm_register_read(vcpu, VCPU_REGS_RAX),
5202                         kvm_register_read(vcpu, VCPU_REGS_RBX),
5203                         kvm_register_read(vcpu, VCPU_REGS_RCX),
5204                         kvm_register_read(vcpu, VCPU_REGS_RDX));
5205 }
5206 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
5207
5208 /*
5209  * Check if userspace requested an interrupt window, and that the
5210  * interrupt window is open.
5211  *
5212  * No need to exit to userspace if we already have an interrupt queued.
5213  */
5214 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
5215 {
5216         return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
5217                 vcpu->run->request_interrupt_window &&
5218                 kvm_arch_interrupt_allowed(vcpu));
5219 }
5220
5221 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
5222 {
5223         struct kvm_run *kvm_run = vcpu->run;
5224
5225         kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
5226         kvm_run->cr8 = kvm_get_cr8(vcpu);
5227         kvm_run->apic_base = kvm_get_apic_base(vcpu);
5228         if (irqchip_in_kernel(vcpu->kvm))
5229                 kvm_run->ready_for_interrupt_injection = 1;
5230         else
5231                 kvm_run->ready_for_interrupt_injection =
5232                         kvm_arch_interrupt_allowed(vcpu) &&
5233                         !kvm_cpu_has_interrupt(vcpu) &&
5234                         !kvm_event_needs_reinjection(vcpu);
5235 }
5236
5237 static void vapic_enter(struct kvm_vcpu *vcpu)
5238 {
5239         struct kvm_lapic *apic = vcpu->arch.apic;
5240         struct page *page;
5241
5242         if (!apic || !apic->vapic_addr)
5243                 return;
5244
5245         page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5246
5247         vcpu->arch.apic->vapic_page = page;
5248 }
5249
5250 static void vapic_exit(struct kvm_vcpu *vcpu)
5251 {
5252         struct kvm_lapic *apic = vcpu->arch.apic;
5253         int idx;
5254
5255         if (!apic || !apic->vapic_addr)
5256                 return;
5257
5258         idx = srcu_read_lock(&vcpu->kvm->srcu);
5259         kvm_release_page_dirty(apic->vapic_page);
5260         mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
5261         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5262 }
5263
5264 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
5265 {
5266         int max_irr, tpr;
5267
5268         if (!kvm_x86_ops->update_cr8_intercept)
5269                 return;
5270
5271         if (!vcpu->arch.apic)
5272                 return;
5273
5274         if (!vcpu->arch.apic->vapic_addr)
5275                 max_irr = kvm_lapic_find_highest_irr(vcpu);
5276         else
5277                 max_irr = -1;
5278
5279         if (max_irr != -1)
5280                 max_irr >>= 4;
5281
5282         tpr = kvm_lapic_get_cr8(vcpu);
5283
5284         kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5285 }
5286
5287 static void inject_pending_event(struct kvm_vcpu *vcpu)
5288 {
5289         /* try to reinject previous events if any */
5290         if (vcpu->arch.exception.pending) {
5291                 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5292                                         vcpu->arch.exception.has_error_code,
5293                                         vcpu->arch.exception.error_code);
5294                 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5295                                           vcpu->arch.exception.has_error_code,
5296                                           vcpu->arch.exception.error_code,
5297                                           vcpu->arch.exception.reinject);
5298                 return;
5299         }
5300
5301         if (vcpu->arch.nmi_injected) {
5302                 kvm_x86_ops->set_nmi(vcpu);
5303                 return;
5304         }
5305
5306         if (vcpu->arch.interrupt.pending) {
5307                 kvm_x86_ops->set_irq(vcpu);
5308                 return;
5309         }
5310
5311         /* try to inject new event if pending */
5312         if (vcpu->arch.nmi_pending) {
5313                 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5314                         vcpu->arch.nmi_pending = false;
5315                         vcpu->arch.nmi_injected = true;
5316                         kvm_x86_ops->set_nmi(vcpu);
5317                 }
5318         } else if (kvm_cpu_has_interrupt(vcpu)) {
5319                 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5320                         kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5321                                             false);
5322                         kvm_x86_ops->set_irq(vcpu);
5323                 }
5324         }
5325 }
5326
5327 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5328 {
5329         if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5330                         !vcpu->guest_xcr0_loaded) {
5331                 /* kvm_set_xcr() also depends on this */
5332                 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5333                 vcpu->guest_xcr0_loaded = 1;
5334         }
5335 }
5336
5337 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5338 {
5339         if (vcpu->guest_xcr0_loaded) {
5340                 if (vcpu->arch.xcr0 != host_xcr0)
5341                         xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5342                 vcpu->guest_xcr0_loaded = 0;
5343         }
5344 }
5345
5346 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5347 {
5348         int r;
5349         bool nmi_pending;
5350         bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5351                 vcpu->run->request_interrupt_window;
5352
5353         if (vcpu->requests) {
5354                 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5355                         kvm_mmu_unload(vcpu);
5356                 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5357                         __kvm_migrate_timers(vcpu);
5358                 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5359                         r = kvm_guest_time_update(vcpu);
5360                         if (unlikely(r))
5361                                 goto out;
5362                 }
5363                 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5364                         kvm_mmu_sync_roots(vcpu);
5365                 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5366                         kvm_x86_ops->tlb_flush(vcpu);
5367                 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5368                         vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5369                         r = 0;
5370                         goto out;
5371                 }
5372                 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5373                         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5374                         r = 0;
5375                         goto out;
5376                 }
5377                 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5378                         vcpu->fpu_active = 0;
5379                         kvm_x86_ops->fpu_deactivate(vcpu);
5380                 }
5381                 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
5382                         /* Page is swapped out. Do synthetic halt */
5383                         vcpu->arch.apf.halted = true;
5384                         r = 1;
5385                         goto out;
5386                 }
5387         }
5388
5389         r = kvm_mmu_reload(vcpu);
5390         if (unlikely(r))
5391                 goto out;
5392
5393         /*
5394          * An NMI can be injected between local nmi_pending read and
5395          * vcpu->arch.nmi_pending read inside inject_pending_event().
5396          * But in that case, KVM_REQ_EVENT will be set, which makes
5397          * the race described above benign.
5398          */
5399         nmi_pending = ACCESS_ONCE(vcpu->arch.nmi_pending);
5400
5401         if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5402                 inject_pending_event(vcpu);
5403
5404                 /* enable NMI/IRQ window open exits if needed */
5405                 if (nmi_pending)
5406                         kvm_x86_ops->enable_nmi_window(vcpu);
5407                 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5408                         kvm_x86_ops->enable_irq_window(vcpu);
5409
5410                 if (kvm_lapic_enabled(vcpu)) {
5411                         update_cr8_intercept(vcpu);
5412                         kvm_lapic_sync_to_vapic(vcpu);
5413                 }
5414         }
5415
5416         preempt_disable();
5417
5418         kvm_x86_ops->prepare_guest_switch(vcpu);
5419         if (vcpu->fpu_active)
5420                 kvm_load_guest_fpu(vcpu);
5421         kvm_load_guest_xcr0(vcpu);
5422
5423         vcpu->mode = IN_GUEST_MODE;
5424
5425         /* We should set ->mode before check ->requests,
5426          * see the comment in make_all_cpus_request.
5427          */
5428         smp_mb();
5429
5430         local_irq_disable();
5431
5432         if (vcpu->mode == EXITING_GUEST_MODE || vcpu->requests
5433             || need_resched() || signal_pending(current)) {
5434                 vcpu->mode = OUTSIDE_GUEST_MODE;
5435                 smp_wmb();
5436                 local_irq_enable();
5437                 preempt_enable();
5438                 kvm_x86_ops->cancel_injection(vcpu);
5439                 r = 1;
5440                 goto out;
5441         }
5442
5443         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5444
5445         kvm_guest_enter();
5446
5447         if (unlikely(vcpu->arch.switch_db_regs)) {
5448                 set_debugreg(0, 7);
5449                 set_debugreg(vcpu->arch.eff_db[0], 0);
5450                 set_debugreg(vcpu->arch.eff_db[1], 1);
5451                 set_debugreg(vcpu->arch.eff_db[2], 2);
5452                 set_debugreg(vcpu->arch.eff_db[3], 3);
5453         }
5454
5455         trace_kvm_entry(vcpu->vcpu_id);
5456         kvm_x86_ops->run(vcpu);
5457
5458         /*
5459          * If the guest has used debug registers, at least dr7
5460          * will be disabled while returning to the host.
5461          * If we don't have active breakpoints in the host, we don't
5462          * care about the messed up debug address registers. But if
5463          * we have some of them active, restore the old state.
5464          */
5465         if (hw_breakpoint_active())
5466                 hw_breakpoint_restore();
5467
5468         kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
5469
5470         vcpu->mode = OUTSIDE_GUEST_MODE;
5471         smp_wmb();
5472         local_irq_enable();
5473
5474         ++vcpu->stat.exits;
5475
5476         /*
5477          * We must have an instruction between local_irq_enable() and
5478          * kvm_guest_exit(), so the timer interrupt isn't delayed by
5479          * the interrupt shadow.  The stat.exits increment will do nicely.
5480          * But we need to prevent reordering, hence this barrier():
5481          */
5482         barrier();
5483
5484         kvm_guest_exit();
5485
5486         preempt_enable();
5487
5488         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5489
5490         /*
5491          * Profile KVM exit RIPs:
5492          */
5493         if (unlikely(prof_on == KVM_PROFILING)) {
5494                 unsigned long rip = kvm_rip_read(vcpu);
5495                 profile_hit(KVM_PROFILING, (void *)rip);
5496         }
5497
5498
5499         kvm_lapic_sync_from_vapic(vcpu);
5500
5501         r = kvm_x86_ops->handle_exit(vcpu);
5502 out:
5503         return r;
5504 }
5505
5506
5507 static int __vcpu_run(struct kvm_vcpu *vcpu)
5508 {
5509         int r;
5510         struct kvm *kvm = vcpu->kvm;
5511
5512         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5513                 pr_debug("vcpu %d received sipi with vector # %x\n",
5514                          vcpu->vcpu_id, vcpu->arch.sipi_vector);
5515                 kvm_lapic_reset(vcpu);
5516                 r = kvm_arch_vcpu_reset(vcpu);
5517                 if (r)
5518                         return r;
5519                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5520         }
5521
5522         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5523         vapic_enter(vcpu);
5524
5525         r = 1;
5526         while (r > 0) {
5527                 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
5528                     !vcpu->arch.apf.halted)
5529                         r = vcpu_enter_guest(vcpu);
5530                 else {
5531                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5532                         kvm_vcpu_block(vcpu);
5533                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5534                         if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5535                         {
5536                                 switch(vcpu->arch.mp_state) {
5537                                 case KVM_MP_STATE_HALTED:
5538                                         vcpu->arch.mp_state =
5539                                                 KVM_MP_STATE_RUNNABLE;
5540                                 case KVM_MP_STATE_RUNNABLE:
5541                                         vcpu->arch.apf.halted = false;
5542                                         break;
5543                                 case KVM_MP_STATE_SIPI_RECEIVED:
5544                                 default:
5545                                         r = -EINTR;
5546                                         break;
5547                                 }
5548                         }
5549                 }
5550
5551                 if (r <= 0)
5552                         break;
5553
5554                 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5555                 if (kvm_cpu_has_pending_timer(vcpu))
5556                         kvm_inject_pending_timer_irqs(vcpu);
5557
5558                 if (dm_request_for_irq_injection(vcpu)) {
5559                         r = -EINTR;
5560                         vcpu->run->exit_reason = KVM_EXIT_INTR;
5561                         ++vcpu->stat.request_irq_exits;
5562                 }
5563
5564                 kvm_check_async_pf_completion(vcpu);
5565
5566                 if (signal_pending(current)) {
5567                         r = -EINTR;
5568                         vcpu->run->exit_reason = KVM_EXIT_INTR;
5569                         ++vcpu->stat.signal_exits;
5570                 }
5571                 if (need_resched()) {
5572                         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5573                         kvm_resched(vcpu);
5574                         vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5575                 }
5576         }
5577
5578         srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5579
5580         vapic_exit(vcpu);
5581
5582         return r;
5583 }
5584
5585 static int complete_mmio(struct kvm_vcpu *vcpu)
5586 {
5587         struct kvm_run *run = vcpu->run;
5588         int r;
5589
5590         if (!(vcpu->arch.pio.count || vcpu->mmio_needed))
5591                 return 1;
5592
5593         if (vcpu->mmio_needed) {
5594                 vcpu->mmio_needed = 0;
5595                 if (!vcpu->mmio_is_write)
5596                         memcpy(vcpu->mmio_data + vcpu->mmio_index,
5597                                run->mmio.data, 8);
5598                 vcpu->mmio_index += 8;
5599                 if (vcpu->mmio_index < vcpu->mmio_size) {
5600                         run->exit_reason = KVM_EXIT_MMIO;
5601                         run->mmio.phys_addr = vcpu->mmio_phys_addr + vcpu->mmio_index;
5602                         memcpy(run->mmio.data, vcpu->mmio_data + vcpu->mmio_index, 8);
5603                         run->mmio.len = min(vcpu->mmio_size - vcpu->mmio_index, 8);
5604                         run->mmio.is_write = vcpu->mmio_is_write;
5605                         vcpu->mmio_needed = 1;
5606                         return 0;
5607                 }
5608                 if (vcpu->mmio_is_write)
5609                         return 1;
5610                 vcpu->mmio_read_completed = 1;
5611         }
5612         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5613         r = emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
5614         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5615         if (r != EMULATE_DONE)
5616                 return 0;
5617         return 1;
5618 }
5619
5620 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5621 {
5622         int r;
5623         sigset_t sigsaved;
5624
5625         if (!tsk_used_math(current) && init_fpu(current))
5626                 return -ENOMEM;
5627
5628         if (vcpu->sigset_active)
5629                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5630
5631         if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5632                 kvm_vcpu_block(vcpu);
5633                 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5634                 r = -EAGAIN;
5635                 goto out;
5636         }
5637
5638         /* re-sync apic's tpr */
5639         if (!irqchip_in_kernel(vcpu->kvm)) {
5640                 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
5641                         r = -EINVAL;
5642                         goto out;
5643                 }
5644         }
5645
5646         r = complete_mmio(vcpu);
5647         if (r <= 0)
5648                 goto out;
5649
5650         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5651                 kvm_register_write(vcpu, VCPU_REGS_RAX,
5652                                      kvm_run->hypercall.ret);
5653
5654         r = __vcpu_run(vcpu);
5655
5656 out:
5657         post_kvm_run_save(vcpu);
5658         if (vcpu->sigset_active)
5659                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5660
5661         return r;
5662 }
5663
5664 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5665 {
5666         if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
5667                 /*
5668                  * We are here if userspace calls get_regs() in the middle of
5669                  * instruction emulation. Registers state needs to be copied
5670                  * back from emulation context to vcpu. Usrapace shouldn't do
5671                  * that usually, but some bad designed PV devices (vmware
5672                  * backdoor interface) need this to work
5673                  */
5674                 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5675                 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5676                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5677         }
5678         regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5679         regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5680         regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5681         regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5682         regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5683         regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5684         regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5685         regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5686 #ifdef CONFIG_X86_64
5687         regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5688         regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5689         regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5690         regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5691         regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5692         regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5693         regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5694         regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5695 #endif
5696
5697         regs->rip = kvm_rip_read(vcpu);
5698         regs->rflags = kvm_get_rflags(vcpu);
5699
5700         return 0;
5701 }
5702
5703 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5704 {
5705         vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
5706         vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
5707
5708         kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5709         kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5710         kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5711         kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5712         kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5713         kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5714         kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5715         kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5716 #ifdef CONFIG_X86_64
5717         kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5718         kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5719         kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5720         kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5721         kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5722         kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5723         kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5724         kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5725 #endif
5726
5727         kvm_rip_write(vcpu, regs->rip);
5728         kvm_set_rflags(vcpu, regs->rflags);
5729
5730         vcpu->arch.exception.pending = false;
5731
5732         kvm_make_request(KVM_REQ_EVENT, vcpu);
5733
5734         return 0;
5735 }
5736
5737 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5738 {
5739         struct kvm_segment cs;
5740
5741         kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
5742         *db = cs.db;
5743         *l = cs.l;
5744 }
5745 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5746
5747 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5748                                   struct kvm_sregs *sregs)
5749 {
5750         struct desc_ptr dt;
5751
5752         kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5753         kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5754         kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5755         kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5756         kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5757         kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5758
5759         kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5760         kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5761
5762         kvm_x86_ops->get_idt(vcpu, &dt);
5763         sregs->idt.limit = dt.size;
5764         sregs->idt.base = dt.address;
5765         kvm_x86_ops->get_gdt(vcpu, &dt);
5766         sregs->gdt.limit = dt.size;
5767         sregs->gdt.base = dt.address;
5768
5769         sregs->cr0 = kvm_read_cr0(vcpu);
5770         sregs->cr2 = vcpu->arch.cr2;
5771         sregs->cr3 = kvm_read_cr3(vcpu);
5772         sregs->cr4 = kvm_read_cr4(vcpu);
5773         sregs->cr8 = kvm_get_cr8(vcpu);
5774         sregs->efer = vcpu->arch.efer;
5775         sregs->apic_base = kvm_get_apic_base(vcpu);
5776
5777         memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
5778
5779         if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
5780                 set_bit(vcpu->arch.interrupt.nr,
5781                         (unsigned long *)sregs->interrupt_bitmap);
5782
5783         return 0;
5784 }
5785
5786 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
5787                                     struct kvm_mp_state *mp_state)
5788 {
5789         mp_state->mp_state = vcpu->arch.mp_state;
5790         return 0;
5791 }
5792
5793 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
5794                                     struct kvm_mp_state *mp_state)
5795 {
5796         vcpu->arch.mp_state = mp_state->mp_state;
5797         kvm_make_request(KVM_REQ_EVENT, vcpu);
5798         return 0;
5799 }
5800
5801 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
5802                     bool has_error_code, u32 error_code)
5803 {
5804         struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5805         int ret;
5806
5807         init_emulate_ctxt(vcpu);
5808
5809         ret = emulator_task_switch(&vcpu->arch.emulate_ctxt,
5810                                    tss_selector, reason, has_error_code,
5811                                    error_code);
5812
5813         if (ret)
5814                 return EMULATE_FAIL;
5815
5816         memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5817         kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
5818         kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
5819         kvm_make_request(KVM_REQ_EVENT, vcpu);
5820         return EMULATE_DONE;
5821 }
5822 EXPORT_SYMBOL_GPL(kvm_task_switch);
5823
5824 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5825                                   struct kvm_sregs *sregs)
5826 {
5827         int mmu_reset_needed = 0;
5828         int pending_vec, max_bits, idx;
5829         struct desc_ptr dt;
5830
5831         dt.size = sregs->idt.limit;
5832         dt.address = sregs->idt.base;
5833         kvm_x86_ops->set_idt(vcpu, &dt);
5834         dt.size = sregs->gdt.limit;
5835         dt.address = sregs->gdt.base;
5836         kvm_x86_ops->set_gdt(vcpu, &dt);
5837
5838         vcpu->arch.cr2 = sregs->cr2;
5839         mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
5840         vcpu->arch.cr3 = sregs->cr3;
5841         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
5842
5843         kvm_set_cr8(vcpu, sregs->cr8);
5844
5845         mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5846         kvm_x86_ops->set_efer(vcpu, sregs->efer);
5847         kvm_set_apic_base(vcpu, sregs->apic_base);
5848
5849         mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5850         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5851         vcpu->arch.cr0 = sregs->cr0;
5852
5853         mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5854         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5855         if (sregs->cr4 & X86_CR4_OSXSAVE)
5856                 update_cpuid(vcpu);
5857
5858         idx = srcu_read_lock(&vcpu->kvm->srcu);
5859         if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5860                 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
5861                 mmu_reset_needed = 1;
5862         }
5863         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5864
5865         if (mmu_reset_needed)
5866                 kvm_mmu_reset_context(vcpu);
5867
5868         max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5869         pending_vec = find_first_bit(
5870                 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5871         if (pending_vec < max_bits) {
5872                 kvm_queue_interrupt(vcpu, pending_vec, false);
5873                 pr_debug("Set back pending irq %d\n", pending_vec);
5874         }
5875
5876         kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5877         kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5878         kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5879         kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5880         kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5881         kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5882
5883         kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5884         kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5885
5886         update_cr8_intercept(vcpu);
5887
5888         /* Older userspace won't unhalt the vcpu on reset. */
5889         if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5890             sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5891             !is_protmode(vcpu))
5892                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5893
5894         kvm_make_request(KVM_REQ_EVENT, vcpu);
5895
5896         return 0;
5897 }
5898
5899 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5900                                         struct kvm_guest_debug *dbg)
5901 {
5902         unsigned long rflags;
5903         int i, r;
5904
5905         if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5906                 r = -EBUSY;
5907                 if (vcpu->arch.exception.pending)
5908                         goto out;
5909                 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5910                         kvm_queue_exception(vcpu, DB_VECTOR);
5911                 else
5912                         kvm_queue_exception(vcpu, BP_VECTOR);
5913         }
5914
5915         /*
5916          * Read rflags as long as potentially injected trace flags are still
5917          * filtered out.
5918          */
5919         rflags = kvm_get_rflags(vcpu);
5920
5921         vcpu->guest_debug = dbg->control;
5922         if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5923                 vcpu->guest_debug = 0;
5924
5925         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5926                 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5927                         vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5928                 vcpu->arch.switch_db_regs =
5929                         (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5930         } else {
5931                 for (i = 0; i < KVM_NR_DB_REGS; i++)
5932                         vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5933                 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5934         }
5935
5936         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5937                 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5938                         get_segment_base(vcpu, VCPU_SREG_CS);
5939
5940         /*
5941          * Trigger an rflags update that will inject or remove the trace
5942          * flags.
5943          */
5944         kvm_set_rflags(vcpu, rflags);
5945
5946         kvm_x86_ops->set_guest_debug(vcpu, dbg);
5947
5948         r = 0;
5949
5950 out:
5951
5952         return r;
5953 }
5954
5955 /*
5956  * Translate a guest virtual address to a guest physical address.
5957  */
5958 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5959                                     struct kvm_translation *tr)
5960 {
5961         unsigned long vaddr = tr->linear_address;
5962         gpa_t gpa;
5963         int idx;
5964
5965         idx = srcu_read_lock(&vcpu->kvm->srcu);
5966         gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5967         srcu_read_unlock(&vcpu->kvm->srcu, idx);
5968         tr->physical_address = gpa;
5969         tr->valid = gpa != UNMAPPED_GVA;
5970         tr->writeable = 1;
5971         tr->usermode = 0;
5972
5973         return 0;
5974 }
5975
5976 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5977 {
5978         struct i387_fxsave_struct *fxsave =
5979                         &vcpu->arch.guest_fpu.state->fxsave;
5980
5981         memcpy(fpu->fpr, fxsave->st_space, 128);
5982         fpu->fcw = fxsave->cwd;
5983         fpu->fsw = fxsave->swd;
5984         fpu->ftwx = fxsave->twd;
5985         fpu->last_opcode = fxsave->fop;
5986         fpu->last_ip = fxsave->rip;
5987         fpu->last_dp = fxsave->rdp;
5988         memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5989
5990         return 0;
5991 }
5992
5993 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5994 {
5995         struct i387_fxsave_struct *fxsave =
5996                         &vcpu->arch.guest_fpu.state->fxsave;
5997
5998         memcpy(fxsave->st_space, fpu->fpr, 128);
5999         fxsave->cwd = fpu->fcw;
6000         fxsave->swd = fpu->fsw;
6001         fxsave->twd = fpu->ftwx;
6002         fxsave->fop = fpu->last_opcode;
6003         fxsave->rip = fpu->last_ip;
6004         fxsave->rdp = fpu->last_dp;
6005         memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
6006
6007         return 0;
6008 }
6009
6010 int fx_init(struct kvm_vcpu *vcpu)
6011 {
6012         int err;
6013
6014         err = fpu_alloc(&vcpu->arch.guest_fpu);
6015         if (err)
6016                 return err;
6017
6018         fpu_finit(&vcpu->arch.guest_fpu);
6019
6020         /*
6021          * Ensure guest xcr0 is valid for loading
6022          */
6023         vcpu->arch.xcr0 = XSTATE_FP;
6024
6025         vcpu->arch.cr0 |= X86_CR0_ET;
6026
6027         return 0;
6028 }
6029 EXPORT_SYMBOL_GPL(fx_init);
6030
6031 static void fx_free(struct kvm_vcpu *vcpu)
6032 {
6033         fpu_free(&vcpu->arch.guest_fpu);
6034 }
6035
6036 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
6037 {
6038         if (vcpu->guest_fpu_loaded)
6039                 return;
6040
6041         /*
6042          * Restore all possible states in the guest,
6043          * and assume host would use all available bits.
6044          * Guest xcr0 would be loaded later.
6045          */
6046         kvm_put_guest_xcr0(vcpu);
6047         vcpu->guest_fpu_loaded = 1;
6048         unlazy_fpu(current);
6049         fpu_restore_checking(&vcpu->arch.guest_fpu);
6050         trace_kvm_fpu(1);
6051 }
6052
6053 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
6054 {
6055         kvm_put_guest_xcr0(vcpu);
6056
6057         if (!vcpu->guest_fpu_loaded)
6058                 return;
6059
6060         vcpu->guest_fpu_loaded = 0;
6061         fpu_save_init(&vcpu->arch.guest_fpu);
6062         ++vcpu->stat.fpu_reload;
6063         kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
6064         trace_kvm_fpu(0);
6065 }
6066
6067 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
6068 {
6069         kvmclock_reset(vcpu);
6070
6071         free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
6072         fx_free(vcpu);
6073         kvm_x86_ops->vcpu_free(vcpu);
6074 }
6075
6076 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
6077                                                 unsigned int id)
6078 {
6079         if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
6080                 printk_once(KERN_WARNING
6081                 "kvm: SMP vm created on host with unstable TSC; "
6082                 "guest TSC will not be reliable\n");
6083         return kvm_x86_ops->vcpu_create(kvm, id);
6084 }
6085
6086 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
6087 {
6088         int r;
6089
6090         vcpu->arch.mtrr_state.have_fixed = 1;
6091         vcpu_load(vcpu);
6092         r = kvm_arch_vcpu_reset(vcpu);
6093         if (r == 0)
6094                 r = kvm_mmu_setup(vcpu);
6095         vcpu_put(vcpu);
6096         if (r < 0)
6097                 goto free_vcpu;
6098
6099         return 0;
6100 free_vcpu:
6101         kvm_x86_ops->vcpu_free(vcpu);
6102         return r;
6103 }
6104
6105 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
6106 {
6107         vcpu->arch.apf.msr_val = 0;
6108
6109         vcpu_load(vcpu);
6110         kvm_mmu_unload(vcpu);
6111         vcpu_put(vcpu);
6112
6113         fx_free(vcpu);
6114         kvm_x86_ops->vcpu_free(vcpu);
6115 }
6116
6117 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
6118 {
6119         vcpu->arch.nmi_pending = false;
6120         vcpu->arch.nmi_injected = false;
6121
6122         vcpu->arch.switch_db_regs = 0;
6123         memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
6124         vcpu->arch.dr6 = DR6_FIXED_1;
6125         vcpu->arch.dr7 = DR7_FIXED_1;
6126
6127         kvm_make_request(KVM_REQ_EVENT, vcpu);
6128         vcpu->arch.apf.msr_val = 0;
6129
6130         kvmclock_reset(vcpu);
6131
6132         kvm_clear_async_pf_completion_queue(vcpu);
6133         kvm_async_pf_hash_reset(vcpu);
6134         vcpu->arch.apf.halted = false;
6135
6136         return kvm_x86_ops->vcpu_reset(vcpu);
6137 }
6138
6139 int kvm_arch_hardware_enable(void *garbage)
6140 {
6141         struct kvm *kvm;
6142         struct kvm_vcpu *vcpu;
6143         int i;
6144
6145         kvm_shared_msr_cpu_online();
6146         list_for_each_entry(kvm, &vm_list, vm_list)
6147                 kvm_for_each_vcpu(i, vcpu, kvm)
6148                         if (vcpu->cpu == smp_processor_id())
6149                                 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6150         return kvm_x86_ops->hardware_enable(garbage);
6151 }
6152
6153 void kvm_arch_hardware_disable(void *garbage)
6154 {
6155         kvm_x86_ops->hardware_disable(garbage);
6156         drop_user_return_notifiers(garbage);
6157 }
6158
6159 int kvm_arch_hardware_setup(void)
6160 {
6161         return kvm_x86_ops->hardware_setup();
6162 }
6163
6164 void kvm_arch_hardware_unsetup(void)
6165 {
6166         kvm_x86_ops->hardware_unsetup();
6167 }
6168
6169 void kvm_arch_check_processor_compat(void *rtn)
6170 {
6171         kvm_x86_ops->check_processor_compatibility(rtn);
6172 }
6173
6174 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6175 {
6176         struct page *page;
6177         struct kvm *kvm;
6178         int r;
6179
6180         BUG_ON(vcpu->kvm == NULL);
6181         kvm = vcpu->kvm;
6182
6183         vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6184         vcpu->arch.walk_mmu = &vcpu->arch.mmu;
6185         vcpu->arch.mmu.root_hpa = INVALID_PAGE;
6186         vcpu->arch.mmu.translate_gpa = translate_gpa;
6187         vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
6188         if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6189                 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6190         else
6191                 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
6192
6193         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
6194         if (!page) {
6195                 r = -ENOMEM;
6196                 goto fail;
6197         }
6198         vcpu->arch.pio_data = page_address(page);
6199
6200         kvm_init_tsc_catchup(vcpu, max_tsc_khz);
6201
6202         r = kvm_mmu_create(vcpu);
6203         if (r < 0)
6204                 goto fail_free_pio_data;
6205
6206         if (irqchip_in_kernel(kvm)) {
6207                 r = kvm_create_lapic(vcpu);
6208                 if (r < 0)
6209                         goto fail_mmu_destroy;
6210         }
6211
6212         vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
6213                                        GFP_KERNEL);
6214         if (!vcpu->arch.mce_banks) {
6215                 r = -ENOMEM;
6216                 goto fail_free_lapic;
6217         }
6218         vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
6219
6220         if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
6221                 goto fail_free_mce_banks;
6222
6223         kvm_async_pf_hash_reset(vcpu);
6224
6225         return 0;
6226 fail_free_mce_banks:
6227         kfree(vcpu->arch.mce_banks);
6228 fail_free_lapic:
6229         kvm_free_lapic(vcpu);
6230 fail_mmu_destroy:
6231         kvm_mmu_destroy(vcpu);
6232 fail_free_pio_data:
6233         free_page((unsigned long)vcpu->arch.pio_data);
6234 fail:
6235         return r;