?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
KVM: Write protect memory after slot swap
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / arch / x86 / kvm / x86.c
blobb0818f6720647db5df3ac42ab9d9577288a3e51e
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 <trace/events/kvm.h>
47
48 #define CREATE_TRACE_POINTS
49 #include "trace.h"
50
51 #include <asm/debugreg.h>
52 #include <asm/msr.h>
53 #include <asm/desc.h>
54 #include <asm/mtrr.h>
55 #include <asm/mce.h>
56 #include <asm/i387.h>
57 #include <asm/xcr.h>
58 #include <asm/pvclock.h>
59 #include <asm/div64.h>
60
61 #define MAX_IO_MSRS 256
62 #define CR0_RESERVED_BITS \
63 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
64 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
65 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
66 #define CR4_RESERVED_BITS \
67 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
68 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
69 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
70 | X86_CR4_OSXSAVE \
71 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
72
73 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
74
75 #define KVM_MAX_MCE_BANKS 32
76 #define KVM_MCE_CAP_SUPPORTED (MCG_CTL_P | MCG_SER_P)
77
78 /* EFER defaults:
79 * - enable syscall per default because its emulated by KVM
80 * - enable LME and LMA per default on 64 bit KVM
81 */
82 #ifdef CONFIG_X86_64
83 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
84 #else
85 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
86 #endif
87
88 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
89 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
90
91 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
92 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
93 struct kvm_cpuid_entry2 __user *entries);
94
95 struct kvm_x86_ops *kvm_x86_ops;
96 EXPORT_SYMBOL_GPL(kvm_x86_ops);
97
98 int ignore_msrs = 0;
99 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
100
101 #define KVM_NR_SHARED_MSRS 16
102
103 struct kvm_shared_msrs_global {
104 int nr;
105 u32 msrs[KVM_NR_SHARED_MSRS];
106 };
107
108 struct kvm_shared_msrs {
109 struct user_return_notifier urn;
110 bool registered;
111 struct kvm_shared_msr_values {
112 u64 host;
113 u64 curr;
114 } values[KVM_NR_SHARED_MSRS];
115 };
116
117 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
118 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
119
120 struct kvm_stats_debugfs_item debugfs_entries[] = {
121 { "pf_fixed", VCPU_STAT(pf_fixed) },
122 { "pf_guest", VCPU_STAT(pf_guest) },
123 { "tlb_flush", VCPU_STAT(tlb_flush) },
124 { "invlpg", VCPU_STAT(invlpg) },
125 { "exits", VCPU_STAT(exits) },
126 { "io_exits", VCPU_STAT(io_exits) },
127 { "mmio_exits", VCPU_STAT(mmio_exits) },
128 { "signal_exits", VCPU_STAT(signal_exits) },
129 { "irq_window", VCPU_STAT(irq_window_exits) },
130 { "nmi_window", VCPU_STAT(nmi_window_exits) },
131 { "halt_exits", VCPU_STAT(halt_exits) },
132 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
133 { "hypercalls", VCPU_STAT(hypercalls) },
134 { "request_irq", VCPU_STAT(request_irq_exits) },
135 { "irq_exits", VCPU_STAT(irq_exits) },
136 { "host_state_reload", VCPU_STAT(host_state_reload) },
137 { "efer_reload", VCPU_STAT(efer_reload) },
138 { "fpu_reload", VCPU_STAT(fpu_reload) },
139 { "insn_emulation", VCPU_STAT(insn_emulation) },
140 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
141 { "irq_injections", VCPU_STAT(irq_injections) },
142 { "nmi_injections", VCPU_STAT(nmi_injections) },
143 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
144 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
145 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
146 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
147 { "mmu_flooded", VM_STAT(mmu_flooded) },
148 { "mmu_recycled", VM_STAT(mmu_recycled) },
149 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
150 { "mmu_unsync", VM_STAT(mmu_unsync) },
151 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
152 { "largepages", VM_STAT(lpages) },
153 { NULL }
154 };
155
156 u64 __read_mostly host_xcr0;
157
158 static inline u32 bit(int bitno)
159 {
160 return 1 << (bitno & 31);
161 }
162
163 static void kvm_on_user_return(struct user_return_notifier *urn)
164 {
165 unsigned slot;
166 struct kvm_shared_msrs *locals
167 = container_of(urn, struct kvm_shared_msrs, urn);
168 struct kvm_shared_msr_values *values;
169
170 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
171 values = &locals->values[slot];
172 if (values->host != values->curr) {
173 wrmsrl(shared_msrs_global.msrs[slot], values->host);
174 values->curr = values->host;
175 }
176 }
177 locals->registered = false;
178 user_return_notifier_unregister(urn);
179 }
180
181 static void shared_msr_update(unsigned slot, u32 msr)
182 {
183 struct kvm_shared_msrs *smsr;
184 u64 value;
185
186 smsr = &__get_cpu_var(shared_msrs);
187 /* only read, and nobody should modify it at this time,
188 * so don't need lock */
189 if (slot >= shared_msrs_global.nr) {
190 printk(KERN_ERR "kvm: invalid MSR slot!");
191 return;
192 }
193 rdmsrl_safe(msr, &value);
194 smsr->values[slot].host = value;
195 smsr->values[slot].curr = value;
196 }
197
198 void kvm_define_shared_msr(unsigned slot, u32 msr)
199 {
200 if (slot >= shared_msrs_global.nr)
201 shared_msrs_global.nr = slot + 1;
202 shared_msrs_global.msrs[slot] = msr;
203 /* we need ensured the shared_msr_global have been updated */
204 smp_wmb();
205 }
206 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
207
208 static void kvm_shared_msr_cpu_online(void)
209 {
210 unsigned i;
211
212 for (i = 0; i < shared_msrs_global.nr; ++i)
213 shared_msr_update(i, shared_msrs_global.msrs[i]);
214 }
215
216 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
217 {
218 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
219
220 if (((value ^ smsr->values[slot].curr) & mask) == 0)
221 return;
222 smsr->values[slot].curr = value;
223 wrmsrl(shared_msrs_global.msrs[slot], value);
224 if (!smsr->registered) {
225 smsr->urn.on_user_return = kvm_on_user_return;
226 user_return_notifier_register(&smsr->urn);
227 smsr->registered = true;
228 }
229 }
230 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
231
232 static void drop_user_return_notifiers(void *ignore)
233 {
234 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
235
236 if (smsr->registered)
237 kvm_on_user_return(&smsr->urn);
238 }
239
240 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
241 {
242 if (irqchip_in_kernel(vcpu->kvm))
243 return vcpu->arch.apic_base;
244 else
245 return vcpu->arch.apic_base;
246 }
247 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
248
249 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
250 {
251 /* TODO: reserve bits check */
252 if (irqchip_in_kernel(vcpu->kvm))
253 kvm_lapic_set_base(vcpu, data);
254 else
255 vcpu->arch.apic_base = data;
256 }
257 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
258
259 #define EXCPT_BENIGN 0
260 #define EXCPT_CONTRIBUTORY 1
261 #define EXCPT_PF 2
262
263 static int exception_class(int vector)
264 {
265 switch (vector) {
266 case PF_VECTOR:
267 return EXCPT_PF;
268 case DE_VECTOR:
269 case TS_VECTOR:
270 case NP_VECTOR:
271 case SS_VECTOR:
272 case GP_VECTOR:
273 return EXCPT_CONTRIBUTORY;
274 default:
275 break;
276 }
277 return EXCPT_BENIGN;
278 }
279
280 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
281 unsigned nr, bool has_error, u32 error_code,
282 bool reinject)
283 {
284 u32 prev_nr;
285 int class1, class2;
286
287 kvm_make_request(KVM_REQ_EVENT, vcpu);
288
289 if (!vcpu->arch.exception.pending) {
290 queue:
291 vcpu->arch.exception.pending = true;
292 vcpu->arch.exception.has_error_code = has_error;
293 vcpu->arch.exception.nr = nr;
294 vcpu->arch.exception.error_code = error_code;
295 vcpu->arch.exception.reinject = reinject;
296 return;
297 }
298
299 /* to check exception */
300 prev_nr = vcpu->arch.exception.nr;
301 if (prev_nr == DF_VECTOR) {
302 /* triple fault -> shutdown */
303 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
304 return;
305 }
306 class1 = exception_class(prev_nr);
307 class2 = exception_class(nr);
308 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
309 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
310 /* generate double fault per SDM Table 5-5 */
311 vcpu->arch.exception.pending = true;
312 vcpu->arch.exception.has_error_code = true;
313 vcpu->arch.exception.nr = DF_VECTOR;
314 vcpu->arch.exception.error_code = 0;
315 } else
316 /* replace previous exception with a new one in a hope
317 that instruction re-execution will regenerate lost
318 exception */
319 goto queue;
320 }
321
322 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
323 {
324 kvm_multiple_exception(vcpu, nr, false, 0, false);
325 }
326 EXPORT_SYMBOL_GPL(kvm_queue_exception);
327
328 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
329 {
330 kvm_multiple_exception(vcpu, nr, false, 0, true);
331 }
332 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
333
334 void kvm_inject_page_fault(struct kvm_vcpu *vcpu)
335 {
336 unsigned error_code = vcpu->arch.fault.error_code;
337
338 ++vcpu->stat.pf_guest;
339 vcpu->arch.cr2 = vcpu->arch.fault.address;
340 kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
341 }
342
343 void kvm_propagate_fault(struct kvm_vcpu *vcpu)
344 {
345 if (mmu_is_nested(vcpu) && !vcpu->arch.fault.nested)
346 vcpu->arch.nested_mmu.inject_page_fault(vcpu);
347 else
348 vcpu->arch.mmu.inject_page_fault(vcpu);
349
350 vcpu->arch.fault.nested = false;
351 }
352
353 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
354 {
355 kvm_make_request(KVM_REQ_EVENT, vcpu);
356 vcpu->arch.nmi_pending = 1;
357 }
358 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
359
360 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
361 {
362 kvm_multiple_exception(vcpu, nr, true, error_code, false);
363 }
364 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
365
366 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
367 {
368 kvm_multiple_exception(vcpu, nr, true, error_code, true);
369 }
370 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
371
372 /*
373 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
374 * a #GP and return false.
375 */
376 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
377 {
378 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
379 return true;
380 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
381 return false;
382 }
383 EXPORT_SYMBOL_GPL(kvm_require_cpl);
384
385 /*
386 * This function will be used to read from the physical memory of the currently
387 * running guest. The difference to kvm_read_guest_page is that this function
388 * can read from guest physical or from the guest's guest physical memory.
389 */
390 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
391 gfn_t ngfn, void *data, int offset, int len,
392 u32 access)
393 {
394 gfn_t real_gfn;
395 gpa_t ngpa;
396
397 ngpa = gfn_to_gpa(ngfn);
398 real_gfn = mmu->translate_gpa(vcpu, ngpa, access);
399 if (real_gfn == UNMAPPED_GVA)
400 return -EFAULT;
401
402 real_gfn = gpa_to_gfn(real_gfn);
403
404 return kvm_read_guest_page(vcpu->kvm, real_gfn, data, offset, len);
405 }
406 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
407
408 int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
409 void *data, int offset, int len, u32 access)
410 {
411 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
412 data, offset, len, access);
413 }
414
415 /*
416 * Load the pae pdptrs. Return true is they are all valid.
417 */
418 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
419 {
420 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
421 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
422 int i;
423 int ret;
424 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
425
426 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
427 offset * sizeof(u64), sizeof(pdpte),
428 PFERR_USER_MASK|PFERR_WRITE_MASK);
429 if (ret < 0) {
430 ret = 0;
431 goto out;
432 }
433 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
434 if (is_present_gpte(pdpte[i]) &&
435 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
436 ret = 0;
437 goto out;
438 }
439 }
440 ret = 1;
441
442 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
443 __set_bit(VCPU_EXREG_PDPTR,
444 (unsigned long *)&vcpu->arch.regs_avail);
445 __set_bit(VCPU_EXREG_PDPTR,
446 (unsigned long *)&vcpu->arch.regs_dirty);
447 out:
448
449 return ret;
450 }
451 EXPORT_SYMBOL_GPL(load_pdptrs);
452
453 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
454 {
455 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
456 bool changed = true;
457 int offset;
458 gfn_t gfn;
459 int r;
460
461 if (is_long_mode(vcpu) || !is_pae(vcpu))
462 return false;
463
464 if (!test_bit(VCPU_EXREG_PDPTR,
465 (unsigned long *)&vcpu->arch.regs_avail))
466 return true;
467
468 gfn = (vcpu->arch.cr3 & ~31u) >> PAGE_SHIFT;
469 offset = (vcpu->arch.cr3 & ~31u) & (PAGE_SIZE - 1);
470 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
471 PFERR_USER_MASK | PFERR_WRITE_MASK);
472 if (r < 0)
473 goto out;
474 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
475 out:
476
477 return changed;
478 }
479
480 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
481 {
482 unsigned long old_cr0 = kvm_read_cr0(vcpu);
483 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP |
484 X86_CR0_CD | X86_CR0_NW;
485
486 cr0 |= X86_CR0_ET;
487
488 #ifdef CONFIG_X86_64
489 if (cr0 & 0xffffffff00000000UL)
490 return 1;
491 #endif
492
493 cr0 &= ~CR0_RESERVED_BITS;
494
495 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
496 return 1;
497
498 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
499 return 1;
500
501 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
502 #ifdef CONFIG_X86_64
503 if ((vcpu->arch.efer & EFER_LME)) {
504 int cs_db, cs_l;
505
506 if (!is_pae(vcpu))
507 return 1;
508 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
509 if (cs_l)
510 return 1;
511 } else
512 #endif
513 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
514 vcpu->arch.cr3))
515 return 1;
516 }
517
518 kvm_x86_ops->set_cr0(vcpu, cr0);
519
520 if ((cr0 ^ old_cr0) & update_bits)
521 kvm_mmu_reset_context(vcpu);
522 return 0;
523 }
524 EXPORT_SYMBOL_GPL(kvm_set_cr0);
525
526 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
527 {
528 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
529 }
530 EXPORT_SYMBOL_GPL(kvm_lmsw);
531
532 int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
533 {
534 u64 xcr0;
535
536 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
537 if (index != XCR_XFEATURE_ENABLED_MASK)
538 return 1;
539 xcr0 = xcr;
540 if (kvm_x86_ops->get_cpl(vcpu) != 0)
541 return 1;
542 if (!(xcr0 & XSTATE_FP))
543 return 1;
544 if ((xcr0 & XSTATE_YMM) && !(xcr0 & XSTATE_SSE))
545 return 1;
546 if (xcr0 & ~host_xcr0)
547 return 1;
548 vcpu->arch.xcr0 = xcr0;
549 vcpu->guest_xcr0_loaded = 0;
550 return 0;
551 }
552
553 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
554 {
555 if (__kvm_set_xcr(vcpu, index, xcr)) {
556 kvm_inject_gp(vcpu, 0);
557 return 1;
558 }
559 return 0;
560 }
561 EXPORT_SYMBOL_GPL(kvm_set_xcr);
562
563 static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
564 {
565 struct kvm_cpuid_entry2 *best;
566
567 best = kvm_find_cpuid_entry(vcpu, 1, 0);
568 return best && (best->ecx & bit(X86_FEATURE_XSAVE));
569 }
570
571 static void update_cpuid(struct kvm_vcpu *vcpu)
572 {
573 struct kvm_cpuid_entry2 *best;
574
575 best = kvm_find_cpuid_entry(vcpu, 1, 0);
576 if (!best)
577 return;
578
579 /* Update OSXSAVE bit */
580 if (cpu_has_xsave && best->function == 0x1) {
581 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
582 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
583 best->ecx |= bit(X86_FEATURE_OSXSAVE);
584 }
585 }
586
587 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
588 {
589 unsigned long old_cr4 = kvm_read_cr4(vcpu);
590 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
591
592 if (cr4 & CR4_RESERVED_BITS)
593 return 1;
594
595 if (!guest_cpuid_has_xsave(vcpu) && (cr4 & X86_CR4_OSXSAVE))
596 return 1;
597
598 if (is_long_mode(vcpu)) {
599 if (!(cr4 & X86_CR4_PAE))
600 return 1;
601 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
602 && ((cr4 ^ old_cr4) & pdptr_bits)
603 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu, vcpu->arch.cr3))
604 return 1;
605
606 if (cr4 & X86_CR4_VMXE)
607 return 1;
608
609 kvm_x86_ops->set_cr4(vcpu, cr4);
610
611 if ((cr4 ^ old_cr4) & pdptr_bits)
612 kvm_mmu_reset_context(vcpu);
613
614 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
615 update_cpuid(vcpu);
616
617 return 0;
618 }
619 EXPORT_SYMBOL_GPL(kvm_set_cr4);
620
621 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
622 {
623 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
624 kvm_mmu_sync_roots(vcpu);
625 kvm_mmu_flush_tlb(vcpu);
626 return 0;
627 }
628
629 if (is_long_mode(vcpu)) {
630 if (cr3 & CR3_L_MODE_RESERVED_BITS)
631 return 1;
632 } else {
633 if (is_pae(vcpu)) {
634 if (cr3 & CR3_PAE_RESERVED_BITS)
635 return 1;
636 if (is_paging(vcpu) &&
637 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
638 return 1;
639 }
640 /*
641 * We don't check reserved bits in nonpae mode, because
642 * this isn't enforced, and VMware depends on this.
643 */
644 }
645
646 /*
647 * Does the new cr3 value map to physical memory? (Note, we
648 * catch an invalid cr3 even in real-mode, because it would
649 * cause trouble later on when we turn on paging anyway.)
650 *
651 * A real CPU would silently accept an invalid cr3 and would
652 * attempt to use it - with largely undefined (and often hard
653 * to debug) behavior on the guest side.
654 */
655 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
656 return 1;
657 vcpu->arch.cr3 = cr3;
658 vcpu->arch.mmu.new_cr3(vcpu);
659 return 0;
660 }
661 EXPORT_SYMBOL_GPL(kvm_set_cr3);
662
663 int __kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
664 {
665 if (cr8 & CR8_RESERVED_BITS)
666 return 1;
667 if (irqchip_in_kernel(vcpu->kvm))
668 kvm_lapic_set_tpr(vcpu, cr8);
669 else
670 vcpu->arch.cr8 = cr8;
671 return 0;
672 }
673
674 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
675 {
676 if (__kvm_set_cr8(vcpu, cr8))
677 kvm_inject_gp(vcpu, 0);
678 }
679 EXPORT_SYMBOL_GPL(kvm_set_cr8);
680
681 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
682 {
683 if (irqchip_in_kernel(vcpu->kvm))
684 return kvm_lapic_get_cr8(vcpu);
685 else
686 return vcpu->arch.cr8;
687 }
688 EXPORT_SYMBOL_GPL(kvm_get_cr8);
689
690 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
691 {
692 switch (dr) {
693 case 0 ... 3:
694 vcpu->arch.db[dr] = val;
695 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
696 vcpu->arch.eff_db[dr] = val;
697 break;
698 case 4:
699 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
700 return 1; /* #UD */
701 /* fall through */
702 case 6:
703 if (val & 0xffffffff00000000ULL)
704 return -1; /* #GP */
705 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
706 break;
707 case 5:
708 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
709 return 1; /* #UD */
710 /* fall through */
711 default: /* 7 */
712 if (val & 0xffffffff00000000ULL)
713 return -1; /* #GP */
714 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
715 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
716 kvm_x86_ops->set_dr7(vcpu, vcpu->arch.dr7);
717 vcpu->arch.switch_db_regs = (val & DR7_BP_EN_MASK);
718 }
719 break;
720 }
721
722 return 0;
723 }
724
725 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
726 {
727 int res;
728
729 res = __kvm_set_dr(vcpu, dr, val);
730 if (res > 0)
731 kvm_queue_exception(vcpu, UD_VECTOR);
732 else if (res < 0)
733 kvm_inject_gp(vcpu, 0);
734
735 return res;
736 }
737 EXPORT_SYMBOL_GPL(kvm_set_dr);
738
739 static int _kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
740 {
741 switch (dr) {
742 case 0 ... 3:
743 *val = vcpu->arch.db[dr];
744 break;
745 case 4:
746 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
747 return 1;
748 /* fall through */
749 case 6:
750 *val = vcpu->arch.dr6;
751 break;
752 case 5:
753 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
754 return 1;
755 /* fall through */
756 default: /* 7 */
757 *val = vcpu->arch.dr7;
758 break;
759 }
760
761 return 0;
762 }
763
764 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
765 {
766 if (_kvm_get_dr(vcpu, dr, val)) {
767 kvm_queue_exception(vcpu, UD_VECTOR);
768 return 1;
769 }
770 return 0;
771 }
772 EXPORT_SYMBOL_GPL(kvm_get_dr);
773
774 /*
775 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
776 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
777 *
778 * This list is modified at module load time to reflect the
779 * capabilities of the host cpu. This capabilities test skips MSRs that are
780 * kvm-specific. Those are put in the beginning of the list.
781 */
782
783 #define KVM_SAVE_MSRS_BEGIN 7
784 static u32 msrs_to_save[] = {
785 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
786 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
787 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
788 HV_X64_MSR_APIC_ASSIST_PAGE,
789 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
790 MSR_STAR,
791 #ifdef CONFIG_X86_64
792 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
793 #endif
794 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
795 };
796
797 static unsigned num_msrs_to_save;
798
799 static u32 emulated_msrs[] = {
800 MSR_IA32_MISC_ENABLE,
801 MSR_IA32_MCG_STATUS,
802 MSR_IA32_MCG_CTL,
803 };
804
805 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
806 {
807 u64 old_efer = vcpu->arch.efer;
808
809 if (efer & efer_reserved_bits)
810 return 1;
811
812 if (is_paging(vcpu)
813 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
814 return 1;
815
816 if (efer & EFER_FFXSR) {
817 struct kvm_cpuid_entry2 *feat;
818
819 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
820 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT)))
821 return 1;
822 }
823
824 if (efer & EFER_SVME) {
825 struct kvm_cpuid_entry2 *feat;
826
827 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
828 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM)))
829 return 1;
830 }
831
832 efer &= ~EFER_LMA;
833 efer |= vcpu->arch.efer & EFER_LMA;
834
835 kvm_x86_ops->set_efer(vcpu, efer);
836
837 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
838 kvm_mmu_reset_context(vcpu);
839
840 /* Update reserved bits */
841 if ((efer ^ old_efer) & EFER_NX)
842 kvm_mmu_reset_context(vcpu);
843
844 return 0;
845 }
846
847 void kvm_enable_efer_bits(u64 mask)
848 {
849 efer_reserved_bits &= ~mask;
850 }
851 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
852
853
854 /*
855 * Writes msr value into into the appropriate "register".
856 * Returns 0 on success, non-0 otherwise.
857 * Assumes vcpu_load() was already called.
858 */
859 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
860 {
861 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
862 }
863
864 /*
865 * Adapt set_msr() to msr_io()'s calling convention
866 */
867 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
868 {
869 return kvm_set_msr(vcpu, index, *data);
870 }
871
872 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
873 {
874 int version;
875 int r;
876 struct pvclock_wall_clock wc;
877 struct timespec boot;
878
879 if (!wall_clock)
880 return;
881
882 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
883 if (r)
884 return;
885
886 if (version & 1)
887 ++version; /* first time write, random junk */
888
889 ++version;
890
891 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
892
893 /*
894 * The guest calculates current wall clock time by adding
895 * system time (updated by kvm_guest_time_update below) to the
896 * wall clock specified here. guest system time equals host
897 * system time for us, thus we must fill in host boot time here.
898 */
899 getboottime(&boot);
900
901 wc.sec = boot.tv_sec;
902 wc.nsec = boot.tv_nsec;
903 wc.version = version;
904
905 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
906
907 version++;
908 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
909 }
910
911 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
912 {
913 uint32_t quotient, remainder;
914
915 /* Don't try to replace with do_div(), this one calculates
916 * "(dividend << 32) / divisor" */
917 __asm__ ( "divl %4"
918 : "=a" (quotient), "=d" (remainder)
919 : "0" (0), "1" (dividend), "r" (divisor) );
920 return quotient;
921 }
922
923 static void kvm_get_time_scale(uint32_t scaled_khz, uint32_t base_khz,
924 s8 *pshift, u32 *pmultiplier)
925 {
926 uint64_t scaled64;
927 int32_t shift = 0;
928 uint64_t tps64;
929 uint32_t tps32;
930
931 tps64 = base_khz * 1000LL;
932 scaled64 = scaled_khz * 1000LL;
933 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
934 tps64 >>= 1;
935 shift--;
936 }
937
938 tps32 = (uint32_t)tps64;
939 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
940 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
941 scaled64 >>= 1;
942 else
943 tps32 <<= 1;
944 shift++;
945 }
946
947 *pshift = shift;
948 *pmultiplier = div_frac(scaled64, tps32);
949
950 pr_debug("%s: base_khz %u => %u, shift %d, mul %u\n",
951 __func__, base_khz, scaled_khz, shift, *pmultiplier);
952 }
953
954 static inline u64 get_kernel_ns(void)
955 {
956 struct timespec ts;
957
958 WARN_ON(preemptible());
959 ktime_get_ts(&ts);
960 monotonic_to_bootbased(&ts);
961 return timespec_to_ns(&ts);
962 }
963
964 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
965 unsigned long max_tsc_khz;
966
967 static inline int kvm_tsc_changes_freq(void)
968 {
969 int cpu = get_cpu();
970 int ret = !boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
971 cpufreq_quick_get(cpu) != 0;
972 put_cpu();
973 return ret;
974 }
975
976 static inline u64 nsec_to_cycles(u64 nsec)
977 {
978 u64 ret;
979
980 WARN_ON(preemptible());
981 if (kvm_tsc_changes_freq())
982 printk_once(KERN_WARNING
983 "kvm: unreliable cycle conversion on adjustable rate TSC\n");
984 ret = nsec * __get_cpu_var(cpu_tsc_khz);
985 do_div(ret, USEC_PER_SEC);
986 return ret;
987 }
988
989 static void kvm_arch_set_tsc_khz(struct kvm *kvm, u32 this_tsc_khz)
990 {
991 /* Compute a scale to convert nanoseconds in TSC cycles */
992 kvm_get_time_scale(this_tsc_khz, NSEC_PER_SEC / 1000,
993 &kvm->arch.virtual_tsc_shift,
994 &kvm->arch.virtual_tsc_mult);
995 kvm->arch.virtual_tsc_khz = this_tsc_khz;
996 }
997
998 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
999 {
1000 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.last_tsc_nsec,
1001 vcpu->kvm->arch.virtual_tsc_mult,
1002 vcpu->kvm->arch.virtual_tsc_shift);
1003 tsc += vcpu->arch.last_tsc_write;
1004 return tsc;
1005 }
1006
1007 void kvm_write_tsc(struct kvm_vcpu *vcpu, u64 data)
1008 {
1009 struct kvm *kvm = vcpu->kvm;
1010 u64 offset, ns, elapsed;
1011 unsigned long flags;
1012 s64 sdiff;
1013
1014 spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1015 offset = data - native_read_tsc();
1016 ns = get_kernel_ns();
1017 elapsed = ns - kvm->arch.last_tsc_nsec;
1018 sdiff = data - kvm->arch.last_tsc_write;
1019 if (sdiff < 0)
1020 sdiff = -sdiff;
1021
1022 /*
1023 * Special case: close write to TSC within 5 seconds of
1024 * another CPU is interpreted as an attempt to synchronize
1025 * The 5 seconds is to accomodate host load / swapping as
1026 * well as any reset of TSC during the boot process.
1027 *
1028 * In that case, for a reliable TSC, we can match TSC offsets,
1029 * or make a best guest using elapsed value.
1030 */
1031 if (sdiff < nsec_to_cycles(5ULL * NSEC_PER_SEC) &&
1032 elapsed < 5ULL * NSEC_PER_SEC) {
1033 if (!check_tsc_unstable()) {
1034 offset = kvm->arch.last_tsc_offset;
1035 pr_debug("kvm: matched tsc offset for %llu\n", data);
1036 } else {
1037 u64 delta = nsec_to_cycles(elapsed);
1038 offset += delta;
1039 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1040 }
1041 ns = kvm->arch.last_tsc_nsec;
1042 }
1043 kvm->arch.last_tsc_nsec = ns;
1044 kvm->arch.last_tsc_write = data;
1045 kvm->arch.last_tsc_offset = offset;
1046 kvm_x86_ops->write_tsc_offset(vcpu, offset);
1047 spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1048
1049 /* Reset of TSC must disable overshoot protection below */
1050 vcpu->arch.hv_clock.tsc_timestamp = 0;
1051 vcpu->arch.last_tsc_write = data;
1052 vcpu->arch.last_tsc_nsec = ns;
1053 }
1054 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1055
1056 static int kvm_guest_time_update(struct kvm_vcpu *v)
1057 {
1058 unsigned long flags;
1059 struct kvm_vcpu_arch *vcpu = &v->arch;
1060 void *shared_kaddr;
1061 unsigned long this_tsc_khz;
1062 s64 kernel_ns, max_kernel_ns;
1063 u64 tsc_timestamp;
1064
1065 /* Keep irq disabled to prevent changes to the clock */
1066 local_irq_save(flags);
1067 kvm_get_msr(v, MSR_IA32_TSC, &tsc_timestamp);
1068 kernel_ns = get_kernel_ns();
1069 this_tsc_khz = __get_cpu_var(cpu_tsc_khz);
1070
1071 if (unlikely(this_tsc_khz == 0)) {
1072 local_irq_restore(flags);
1073 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
1074 return 1;
1075 }
1076
1077 /*
1078 * We may have to catch up the TSC to match elapsed wall clock
1079 * time for two reasons, even if kvmclock is used.
1080 * 1) CPU could have been running below the maximum TSC rate
1081 * 2) Broken TSC compensation resets the base at each VCPU
1082 * entry to avoid unknown leaps of TSC even when running
1083 * again on the same CPU. This may cause apparent elapsed
1084 * time to disappear, and the guest to stand still or run
1085 * very slowly.
1086 */
1087 if (vcpu->tsc_catchup) {
1088 u64 tsc = compute_guest_tsc(v, kernel_ns);
1089 if (tsc > tsc_timestamp) {
1090 kvm_x86_ops->adjust_tsc_offset(v, tsc - tsc_timestamp);
1091 tsc_timestamp = tsc;
1092 }
1093 }
1094
1095 local_irq_restore(flags);
1096
1097 if (!vcpu->time_page)
1098 return 0;
1099
1100 /*
1101 * Time as measured by the TSC may go backwards when resetting the base
1102 * tsc_timestamp. The reason for this is that the TSC resolution is
1103 * higher than the resolution of the other clock scales. Thus, many
1104 * possible measurments of the TSC correspond to one measurement of any
1105 * other clock, and so a spread of values is possible. This is not a
1106 * problem for the computation of the nanosecond clock; with TSC rates
1107 * around 1GHZ, there can only be a few cycles which correspond to one
1108 * nanosecond value, and any path through this code will inevitably
1109 * take longer than that. However, with the kernel_ns value itself,
1110 * the precision may be much lower, down to HZ granularity. If the
1111 * first sampling of TSC against kernel_ns ends in the low part of the
1112 * range, and the second in the high end of the range, we can get:
1113 *
1114 * (TSC - offset_low) * S + kns_old > (TSC - offset_high) * S + kns_new
1115 *
1116 * As the sampling errors potentially range in the thousands of cycles,
1117 * it is possible such a time value has already been observed by the
1118 * guest. To protect against this, we must compute the system time as
1119 * observed by the guest and ensure the new system time is greater.
1120 */
1121 max_kernel_ns = 0;
1122 if (vcpu->hv_clock.tsc_timestamp && vcpu->last_guest_tsc) {
1123 max_kernel_ns = vcpu->last_guest_tsc -
1124 vcpu->hv_clock.tsc_timestamp;
1125 max_kernel_ns = pvclock_scale_delta(max_kernel_ns,
1126 vcpu->hv_clock.tsc_to_system_mul,
1127 vcpu->hv_clock.tsc_shift);
1128 max_kernel_ns += vcpu->last_kernel_ns;
1129 }
1130
1131 if (unlikely(vcpu->hw_tsc_khz != this_tsc_khz)) {
1132 kvm_get_time_scale(NSEC_PER_SEC / 1000, this_tsc_khz,
1133 &vcpu->hv_clock.tsc_shift,
1134 &vcpu->hv_clock.tsc_to_system_mul);
1135 vcpu->hw_tsc_khz = this_tsc_khz;
1136 }
1137
1138 if (max_kernel_ns > kernel_ns)
1139 kernel_ns = max_kernel_ns;
1140
1141 /* With all the info we got, fill in the values */
1142 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
1143 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
1144 vcpu->last_kernel_ns = kernel_ns;
1145 vcpu->last_guest_tsc = tsc_timestamp;
1146 vcpu->hv_clock.flags = 0;
1147
1148 /*
1149 * The interface expects us to write an even number signaling that the
1150 * update is finished. Since the guest won't see the intermediate
1151 * state, we just increase by 2 at the end.
1152 */
1153 vcpu->hv_clock.version += 2;
1154
1155 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
1156
1157 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
1158 sizeof(vcpu->hv_clock));
1159
1160 kunmap_atomic(shared_kaddr, KM_USER0);
1161
1162 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
1163 return 0;
1164 }
1165
1166 static bool msr_mtrr_valid(unsigned msr)
1167 {
1168 switch (msr) {
1169 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
1170 case MSR_MTRRfix64K_00000:
1171 case MSR_MTRRfix16K_80000:
1172 case MSR_MTRRfix16K_A0000:
1173 case MSR_MTRRfix4K_C0000:
1174 case MSR_MTRRfix4K_C8000:
1175 case MSR_MTRRfix4K_D0000:
1176 case MSR_MTRRfix4K_D8000:
1177 case MSR_MTRRfix4K_E0000:
1178 case MSR_MTRRfix4K_E8000:
1179 case MSR_MTRRfix4K_F0000:
1180 case MSR_MTRRfix4K_F8000:
1181 case MSR_MTRRdefType:
1182 case MSR_IA32_CR_PAT:
1183 return true;
1184 case 0x2f8:
1185 return true;
1186 }
1187 return false;
1188 }
1189
1190 static bool valid_pat_type(unsigned t)
1191 {
1192 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
1193 }
1194
1195 static bool valid_mtrr_type(unsigned t)
1196 {
1197 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
1198 }
1199
1200 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1201 {
1202 int i;
1203
1204 if (!msr_mtrr_valid(msr))
1205 return false;
1206
1207 if (msr == MSR_IA32_CR_PAT) {
1208 for (i = 0; i < 8; i++)
1209 if (!valid_pat_type((data >> (i * 8)) & 0xff))
1210 return false;
1211 return true;
1212 } else if (msr == MSR_MTRRdefType) {
1213 if (data & ~0xcff)
1214 return false;
1215 return valid_mtrr_type(data & 0xff);
1216 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
1217 for (i = 0; i < 8 ; i++)
1218 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
1219 return false;
1220 return true;
1221 }
1222
1223 /* variable MTRRs */
1224 return valid_mtrr_type(data & 0xff);
1225 }
1226
1227 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1228 {
1229 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1230
1231 if (!mtrr_valid(vcpu, msr, data))
1232 return 1;
1233
1234 if (msr == MSR_MTRRdefType) {
1235 vcpu->arch.mtrr_state.def_type = data;
1236 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
1237 } else if (msr == MSR_MTRRfix64K_00000)
1238 p[0] = data;
1239 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1240 p[1 + msr - MSR_MTRRfix16K_80000] = data;
1241 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1242 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
1243 else if (msr == MSR_IA32_CR_PAT)
1244 vcpu->arch.pat = data;
1245 else { /* Variable MTRRs */
1246 int idx, is_mtrr_mask;
1247 u64 *pt;
1248
1249 idx = (msr - 0x200) / 2;
1250 is_mtrr_mask = msr - 0x200 - 2 * idx;
1251 if (!is_mtrr_mask)
1252 pt =
1253 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1254 else
1255 pt =
1256 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1257 *pt = data;
1258 }
1259
1260 kvm_mmu_reset_context(vcpu);
1261 return 0;
1262 }
1263
1264 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1265 {
1266 u64 mcg_cap = vcpu->arch.mcg_cap;
1267 unsigned bank_num = mcg_cap & 0xff;
1268
1269 switch (msr) {
1270 case MSR_IA32_MCG_STATUS:
1271 vcpu->arch.mcg_status = data;
1272 break;
1273 case MSR_IA32_MCG_CTL:
1274 if (!(mcg_cap & MCG_CTL_P))
1275 return 1;
1276 if (data != 0 && data != ~(u64)0)
1277 return -1;
1278 vcpu->arch.mcg_ctl = data;
1279 break;
1280 default:
1281 if (msr >= MSR_IA32_MC0_CTL &&
1282 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1283 u32 offset = msr - MSR_IA32_MC0_CTL;
1284 /* only 0 or all 1s can be written to IA32_MCi_CTL
1285 * some Linux kernels though clear bit 10 in bank 4 to
1286 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
1287 * this to avoid an uncatched #GP in the guest
1288 */
1289 if ((offset & 0x3) == 0 &&
1290 data != 0 && (data | (1 << 10)) != ~(u64)0)
1291 return -1;
1292 vcpu->arch.mce_banks[offset] = data;
1293 break;
1294 }
1295 return 1;
1296 }
1297 return 0;
1298 }
1299
1300 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1301 {
1302 struct kvm *kvm = vcpu->kvm;
1303 int lm = is_long_mode(vcpu);
1304 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
1305 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
1306 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
1307 : kvm->arch.xen_hvm_config.blob_size_32;
1308 u32 page_num = data & ~PAGE_MASK;
1309 u64 page_addr = data & PAGE_MASK;
1310 u8 *page;
1311 int r;
1312
1313 r = -E2BIG;
1314 if (page_num >= blob_size)
1315 goto out;
1316 r = -ENOMEM;
1317 page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1318 if (!page)
1319 goto out;
1320 r = -EFAULT;
1321 if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1322 goto out_free;
1323 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1324 goto out_free;
1325 r = 0;
1326 out_free:
1327 kfree(page);
1328 out:
1329 return r;
1330 }
1331
1332 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
1333 {
1334 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
1335 }
1336
1337 static bool kvm_hv_msr_partition_wide(u32 msr)
1338 {
1339 bool r = false;
1340 switch (msr) {
1341 case HV_X64_MSR_GUEST_OS_ID:
1342 case HV_X64_MSR_HYPERCALL:
1343 r = true;
1344 break;
1345 }
1346
1347 return r;
1348 }
1349
1350 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1351 {
1352 struct kvm *kvm = vcpu->kvm;
1353
1354 switch (msr) {
1355 case HV_X64_MSR_GUEST_OS_ID:
1356 kvm->arch.hv_guest_os_id = data;
1357 /* setting guest os id to zero disables hypercall page */
1358 if (!kvm->arch.hv_guest_os_id)
1359 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1360 break;
1361 case HV_X64_MSR_HYPERCALL: {
1362 u64 gfn;
1363 unsigned long addr;
1364 u8 instructions[4];
1365
1366 /* if guest os id is not set hypercall should remain disabled */
1367 if (!kvm->arch.hv_guest_os_id)
1368 break;
1369 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1370 kvm->arch.hv_hypercall = data;
1371 break;
1372 }
1373 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1374 addr = gfn_to_hva(kvm, gfn);
1375 if (kvm_is_error_hva(addr))
1376 return 1;
1377 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1378 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1379 if (copy_to_user((void __user *)addr, instructions, 4))
1380 return 1;
1381 kvm->arch.hv_hypercall = data;
1382 break;
1383 }
1384 default:
1385 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1386 "data 0x%llx\n", msr, data);
1387 return 1;
1388 }
1389 return 0;
1390 }
1391
1392 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1393 {
1394 switch (msr) {
1395 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1396 unsigned long addr;
1397
1398 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1399 vcpu->arch.hv_vapic = data;
1400 break;
1401 }
1402 addr = gfn_to_hva(vcpu->kvm, data >>
1403 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1404 if (kvm_is_error_hva(addr))
1405 return 1;
1406 if (clear_user((void __user *)addr, PAGE_SIZE))
1407 return 1;
1408 vcpu->arch.hv_vapic = data;
1409 break;
1410 }
1411 case HV_X64_MSR_EOI:
1412 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1413 case HV_X64_MSR_ICR:
1414 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1415 case HV_X64_MSR_TPR:
1416 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1417 default:
1418 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1419 "data 0x%llx\n", msr, data);
1420 return 1;
1421 }
1422
1423 return 0;
1424 }
1425
1426 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1427 {
1428 switch (msr) {
1429 case MSR_EFER:
1430 return set_efer(vcpu, data);
1431 case MSR_K7_HWCR:
1432 data &= ~(u64)0x40; /* ignore flush filter disable */
1433 data &= ~(u64)0x100; /* ignore ignne emulation enable */
1434 if (data != 0) {
1435 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1436 data);
1437 return 1;
1438 }
1439 break;
1440 case MSR_FAM10H_MMIO_CONF_BASE:
1441 if (data != 0) {
1442 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1443 "0x%llx\n", data);
1444 return 1;
1445 }
1446 break;
1447 case MSR_AMD64_NB_CFG:
1448 break;
1449 case MSR_IA32_DEBUGCTLMSR:
1450 if (!data) {
1451 /* We support the non-activated case already */
1452 break;
1453 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1454 /* Values other than LBR and BTF are vendor-specific,
1455 thus reserved and should throw a #GP */
1456 return 1;
1457 }
1458 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1459 __func__, data);
1460 break;
1461 case MSR_IA32_UCODE_REV:
1462 case MSR_IA32_UCODE_WRITE:
1463 case MSR_VM_HSAVE_PA:
1464 case MSR_AMD64_PATCH_LOADER:
1465 break;
1466 case 0x200 ... 0x2ff:
1467 return set_msr_mtrr(vcpu, msr, data);
1468 case MSR_IA32_APICBASE:
1469 kvm_set_apic_base(vcpu, data);
1470 break;
1471 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1472 return kvm_x2apic_msr_write(vcpu, msr, data);
1473 case MSR_IA32_MISC_ENABLE:
1474 vcpu->arch.ia32_misc_enable_msr = data;
1475 break;
1476 case MSR_KVM_WALL_CLOCK_NEW:
1477 case MSR_KVM_WALL_CLOCK:
1478 vcpu->kvm->arch.wall_clock = data;
1479 kvm_write_wall_clock(vcpu->kvm, data);
1480 break;
1481 case MSR_KVM_SYSTEM_TIME_NEW:
1482 case MSR_KVM_SYSTEM_TIME: {
1483 if (vcpu->arch.time_page) {
1484 kvm_release_page_dirty(vcpu->arch.time_page);
1485 vcpu->arch.time_page = NULL;
1486 }
1487
1488 vcpu->arch.time = data;
1489 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
1490
1491 /* we verify if the enable bit is set... */
1492 if (!(data & 1))
1493 break;
1494
1495 /* ...but clean it before doing the actual write */
1496 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1497
1498 vcpu->arch.time_page =
1499 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1500
1501 if (is_error_page(vcpu->arch.time_page)) {
1502 kvm_release_page_clean(vcpu->arch.time_page);
1503 vcpu->arch.time_page = NULL;
1504 }
1505 break;
1506 }
1507 case MSR_IA32_MCG_CTL:
1508 case MSR_IA32_MCG_STATUS:
1509 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1510 return set_msr_mce(vcpu, msr, data);
1511
1512 /* Performance counters are not protected by a CPUID bit,
1513 * so we should check all of them in the generic path for the sake of
1514 * cross vendor migration.
1515 * Writing a zero into the event select MSRs disables them,
1516 * which we perfectly emulate ;-). Any other value should be at least
1517 * reported, some guests depend on them.
1518 */
1519 case MSR_P6_EVNTSEL0:
1520 case MSR_P6_EVNTSEL1:
1521 case MSR_K7_EVNTSEL0:
1522 case MSR_K7_EVNTSEL1:
1523 case MSR_K7_EVNTSEL2:
1524 case MSR_K7_EVNTSEL3:
1525 if (data != 0)
1526 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1527 "0x%x data 0x%llx\n", msr, data);
1528 break;
1529 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1530 * so we ignore writes to make it happy.
1531 */
1532 case MSR_P6_PERFCTR0:
1533 case MSR_P6_PERFCTR1:
1534 case MSR_K7_PERFCTR0:
1535 case MSR_K7_PERFCTR1:
1536 case MSR_K7_PERFCTR2:
1537 case MSR_K7_PERFCTR3:
1538 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1539 "0x%x data 0x%llx\n", msr, data);
1540 break;
1541 case MSR_K7_CLK_CTL:
1542 /*
1543 * Ignore all writes to this no longer documented MSR.
1544 * Writes are only relevant for old K7 processors,
1545 * all pre-dating SVM, but a recommended workaround from
1546 * AMD for these chips. It is possible to speicify the
1547 * affected processor models on the command line, hence
1548 * the need to ignore the workaround.
1549 */
1550 break;
1551 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1552 if (kvm_hv_msr_partition_wide(msr)) {
1553 int r;
1554 mutex_lock(&vcpu->kvm->lock);
1555 r = set_msr_hyperv_pw(vcpu, msr, data);
1556 mutex_unlock(&vcpu->kvm->lock);
1557 return r;
1558 } else
1559 return set_msr_hyperv(vcpu, msr, data);
1560 break;
1561 default:
1562 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1563 return xen_hvm_config(vcpu, data);
1564 if (!ignore_msrs) {
1565 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1566 msr, data);
1567 return 1;
1568 } else {
1569 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1570 msr, data);
1571 break;
1572 }
1573 }
1574 return 0;
1575 }
1576 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1577
1578
1579 /*
1580 * Reads an msr value (of 'msr_index') into 'pdata'.
1581 * Returns 0 on success, non-0 otherwise.
1582 * Assumes vcpu_load() was already called.
1583 */
1584 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1585 {
1586 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1587 }
1588
1589 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1590 {
1591 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1592
1593 if (!msr_mtrr_valid(msr))
1594 return 1;
1595
1596 if (msr == MSR_MTRRdefType)
1597 *pdata = vcpu->arch.mtrr_state.def_type +
1598 (vcpu->arch.mtrr_state.enabled << 10);
1599 else if (msr == MSR_MTRRfix64K_00000)
1600 *pdata = p[0];
1601 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1602 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1603 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1604 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1605 else if (msr == MSR_IA32_CR_PAT)
1606 *pdata = vcpu->arch.pat;
1607 else { /* Variable MTRRs */
1608 int idx, is_mtrr_mask;
1609 u64 *pt;
1610
1611 idx = (msr - 0x200) / 2;
1612 is_mtrr_mask = msr - 0x200 - 2 * idx;
1613 if (!is_mtrr_mask)
1614 pt =
1615 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1616 else
1617 pt =
1618 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1619 *pdata = *pt;
1620 }
1621
1622 return 0;
1623 }
1624
1625 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1626 {
1627 u64 data;
1628 u64 mcg_cap = vcpu->arch.mcg_cap;
1629 unsigned bank_num = mcg_cap & 0xff;
1630
1631 switch (msr) {
1632 case MSR_IA32_P5_MC_ADDR:
1633 case MSR_IA32_P5_MC_TYPE:
1634 data = 0;
1635 break;
1636 case MSR_IA32_MCG_CAP:
1637 data = vcpu->arch.mcg_cap;
1638 break;
1639 case MSR_IA32_MCG_CTL:
1640 if (!(mcg_cap & MCG_CTL_P))
1641 return 1;
1642 data = vcpu->arch.mcg_ctl;
1643 break;
1644 case MSR_IA32_MCG_STATUS:
1645 data = vcpu->arch.mcg_status;
1646 break;
1647 default:
1648 if (msr >= MSR_IA32_MC0_CTL &&
1649 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1650 u32 offset = msr - MSR_IA32_MC0_CTL;
1651 data = vcpu->arch.mce_banks[offset];
1652 break;
1653 }
1654 return 1;
1655 }
1656 *pdata = data;
1657 return 0;
1658 }
1659
1660 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1661 {
1662 u64 data = 0;
1663 struct kvm *kvm = vcpu->kvm;
1664
1665 switch (msr) {
1666 case HV_X64_MSR_GUEST_OS_ID:
1667 data = kvm->arch.hv_guest_os_id;
1668 break;
1669 case HV_X64_MSR_HYPERCALL:
1670 data = kvm->arch.hv_hypercall;
1671 break;
1672 default:
1673 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1674 return 1;
1675 }
1676
1677 *pdata = data;
1678 return 0;
1679 }
1680
1681 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1682 {
1683 u64 data = 0;
1684
1685 switch (msr) {
1686 case HV_X64_MSR_VP_INDEX: {
1687 int r;
1688 struct kvm_vcpu *v;
1689 kvm_for_each_vcpu(r, v, vcpu->kvm)
1690 if (v == vcpu)
1691 data = r;
1692 break;
1693 }
1694 case HV_X64_MSR_EOI:
1695 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1696 case HV_X64_MSR_ICR:
1697 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1698 case HV_X64_MSR_TPR:
1699 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1700 default:
1701 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1702 return 1;
1703 }
1704 *pdata = data;
1705 return 0;
1706 }
1707
1708 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1709 {
1710 u64 data;
1711
1712 switch (msr) {
1713 case MSR_IA32_PLATFORM_ID:
1714 case MSR_IA32_UCODE_REV:
1715 case MSR_IA32_EBL_CR_POWERON:
1716 case MSR_IA32_DEBUGCTLMSR:
1717 case MSR_IA32_LASTBRANCHFROMIP:
1718 case MSR_IA32_LASTBRANCHTOIP:
1719 case MSR_IA32_LASTINTFROMIP:
1720 case MSR_IA32_LASTINTTOIP:
1721 case MSR_K8_SYSCFG:
1722 case MSR_K7_HWCR:
1723 case MSR_VM_HSAVE_PA:
1724 case MSR_P6_PERFCTR0:
1725 case MSR_P6_PERFCTR1:
1726 case MSR_P6_EVNTSEL0:
1727 case MSR_P6_EVNTSEL1:
1728 case MSR_K7_EVNTSEL0:
1729 case MSR_K7_PERFCTR0:
1730 case MSR_K8_INT_PENDING_MSG:
1731 case MSR_AMD64_NB_CFG:
1732 case MSR_FAM10H_MMIO_CONF_BASE:
1733 data = 0;
1734 break;
1735 case MSR_MTRRcap:
1736 data = 0x500 | KVM_NR_VAR_MTRR;
1737 break;
1738 case 0x200 ... 0x2ff:
1739 return get_msr_mtrr(vcpu, msr, pdata);
1740 case 0xcd: /* fsb frequency */
1741 data = 3;
1742 break;
1743 /*
1744 * MSR_EBC_FREQUENCY_ID
1745 * Conservative value valid for even the basic CPU models.
1746 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
1747 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
1748 * and 266MHz for model 3, or 4. Set Core Clock
1749 * Frequency to System Bus Frequency Ratio to 1 (bits
1750 * 31:24) even though these are only valid for CPU
1751 * models > 2, however guests may end up dividing or
1752 * multiplying by zero otherwise.
1753 */
1754 case MSR_EBC_FREQUENCY_ID:
1755 data = 1 << 24;
1756 break;
1757 case MSR_IA32_APICBASE:
1758 data = kvm_get_apic_base(vcpu);
1759 break;
1760 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1761 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1762 break;
1763 case MSR_IA32_MISC_ENABLE:
1764 data = vcpu->arch.ia32_misc_enable_msr;
1765 break;
1766 case MSR_IA32_PERF_STATUS:
1767 /* TSC increment by tick */
1768 data = 1000ULL;
1769 /* CPU multiplier */
1770 data |= (((uint64_t)4ULL) << 40);
1771 break;
1772 case MSR_EFER:
1773 data = vcpu->arch.efer;
1774 break;
1775 case MSR_KVM_WALL_CLOCK:
1776 case MSR_KVM_WALL_CLOCK_NEW:
1777 data = vcpu->kvm->arch.wall_clock;
1778 break;
1779 case MSR_KVM_SYSTEM_TIME:
1780 case MSR_KVM_SYSTEM_TIME_NEW:
1781 data = vcpu->arch.time;
1782 break;
1783 case MSR_IA32_P5_MC_ADDR:
1784 case MSR_IA32_P5_MC_TYPE:
1785 case MSR_IA32_MCG_CAP:
1786 case MSR_IA32_MCG_CTL:
1787 case MSR_IA32_MCG_STATUS:
1788 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1789 return get_msr_mce(vcpu, msr, pdata);
1790 case MSR_K7_CLK_CTL:
1791 /*
1792 * Provide expected ramp-up count for K7. All other
1793 * are set to zero, indicating minimum divisors for
1794 * every field.
1795 *
1796 * This prevents guest kernels on AMD host with CPU
1797 * type 6, model 8 and higher from exploding due to
1798 * the rdmsr failing.
1799 */
1800 data = 0x20000000;
1801 break;
1802 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1803 if (kvm_hv_msr_partition_wide(msr)) {
1804 int r;
1805 mutex_lock(&vcpu->kvm->lock);
1806 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1807 mutex_unlock(&vcpu->kvm->lock);
1808 return r;
1809 } else
1810 return get_msr_hyperv(vcpu, msr, pdata);
1811 break;
1812 default:
1813 if (!ignore_msrs) {
1814 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1815 return 1;
1816 } else {
1817 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1818 data = 0;
1819 }
1820 break;
1821 }
1822 *pdata = data;
1823 return 0;
1824 }
1825 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1826
1827 /*
1828 * Read or write a bunch of msrs. All parameters are kernel addresses.
1829 *
1830 * @return number of msrs set successfully.
1831 */
1832 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1833 struct kvm_msr_entry *entries,
1834 int (*do_msr)(struct kvm_vcpu *vcpu,
1835 unsigned index, u64 *data))
1836 {
1837 int i, idx;
1838
1839 idx = srcu_read_lock(&vcpu->kvm->srcu);
1840 for (i = 0; i < msrs->nmsrs; ++i)
1841 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1842 break;
1843 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1844
1845 return i;
1846 }
1847
1848 /*
1849 * Read or write a bunch of msrs. Parameters are user addresses.
1850 *
1851 * @return number of msrs set successfully.
1852 */
1853 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1854 int (*do_msr)(struct kvm_vcpu *vcpu,
1855 unsigned index, u64 *data),
1856 int writeback)
1857 {
1858 struct kvm_msrs msrs;
1859 struct kvm_msr_entry *entries;
1860 int r, n;
1861 unsigned size;
1862
1863 r = -EFAULT;
1864 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1865 goto out;
1866
1867 r = -E2BIG;
1868 if (msrs.nmsrs >= MAX_IO_MSRS)
1869 goto out;
1870
1871 r = -ENOMEM;
1872 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1873 entries = kmalloc(size, GFP_KERNEL);
1874 if (!entries)
1875 goto out;
1876
1877 r = -EFAULT;
1878 if (copy_from_user(entries, user_msrs->entries, size))
1879 goto out_free;
1880
1881 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1882 if (r < 0)
1883 goto out_free;
1884
1885 r = -EFAULT;
1886 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1887 goto out_free;
1888
1889 r = n;
1890
1891 out_free:
1892 kfree(entries);
1893 out:
1894 return r;
1895 }
1896
1897 int kvm_dev_ioctl_check_extension(long ext)
1898 {
1899 int r;
1900
1901 switch (ext) {
1902 case KVM_CAP_IRQCHIP:
1903 case KVM_CAP_HLT:
1904 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1905 case KVM_CAP_SET_TSS_ADDR:
1906 case KVM_CAP_EXT_CPUID:
1907 case KVM_CAP_CLOCKSOURCE:
1908 case KVM_CAP_PIT:
1909 case KVM_CAP_NOP_IO_DELAY:
1910 case KVM_CAP_MP_STATE:
1911 case KVM_CAP_SYNC_MMU:
1912 case KVM_CAP_REINJECT_CONTROL:
1913 case KVM_CAP_IRQ_INJECT_STATUS:
1914 case KVM_CAP_ASSIGN_DEV_IRQ:
1915 case KVM_CAP_IRQFD:
1916 case KVM_CAP_IOEVENTFD:
1917 case KVM_CAP_PIT2:
1918 case KVM_CAP_PIT_STATE2:
1919 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1920 case KVM_CAP_XEN_HVM:
1921 case KVM_CAP_ADJUST_CLOCK:
1922 case KVM_CAP_VCPU_EVENTS:
1923 case KVM_CAP_HYPERV:
1924 case KVM_CAP_HYPERV_VAPIC:
1925 case KVM_CAP_HYPERV_SPIN:
1926 case KVM_CAP_PCI_SEGMENT:
1927 case KVM_CAP_DEBUGREGS:
1928 case KVM_CAP_X86_ROBUST_SINGLESTEP:
1929 case KVM_CAP_XSAVE:
1930 r = 1;
1931 break;
1932 case KVM_CAP_COALESCED_MMIO:
1933 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1934 break;
1935 case KVM_CAP_VAPIC:
1936 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1937 break;
1938 case KVM_CAP_NR_VCPUS:
1939 r = KVM_MAX_VCPUS;
1940 break;
1941 case KVM_CAP_NR_MEMSLOTS:
1942 r = KVM_MEMORY_SLOTS;
1943 break;
1944 case KVM_CAP_PV_MMU: /* obsolete */
1945 r = 0;
1946 break;
1947 case KVM_CAP_IOMMU:
1948 r = iommu_found();
1949 break;
1950 case KVM_CAP_MCE:
1951 r = KVM_MAX_MCE_BANKS;
1952 break;
1953 case KVM_CAP_XCRS:
1954 r = cpu_has_xsave;
1955 break;
1956 default:
1957 r = 0;
1958 break;
1959 }
1960 return r;
1961
1962 }
1963
1964 long kvm_arch_dev_ioctl(struct file *filp,
1965 unsigned int ioctl, unsigned long arg)
1966 {
1967 void __user *argp = (void __user *)arg;
1968 long r;
1969
1970 switch (ioctl) {
1971 case KVM_GET_MSR_INDEX_LIST: {
1972 struct kvm_msr_list __user *user_msr_list = argp;
1973 struct kvm_msr_list msr_list;
1974 unsigned n;
1975
1976 r = -EFAULT;
1977 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1978 goto out;
1979 n = msr_list.nmsrs;
1980 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1981 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1982 goto out;
1983 r = -E2BIG;
1984 if (n < msr_list.nmsrs)
1985 goto out;
1986 r = -EFAULT;
1987 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1988 num_msrs_to_save * sizeof(u32)))
1989 goto out;
1990 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1991 &emulated_msrs,
1992 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1993 goto out;
1994 r = 0;
1995 break;
1996 }
1997 case KVM_GET_SUPPORTED_CPUID: {
1998 struct kvm_cpuid2 __user *cpuid_arg = argp;
1999 struct kvm_cpuid2 cpuid;
2000
2001 r = -EFAULT;
2002 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2003 goto out;
2004 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
2005 cpuid_arg->entries);
2006 if (r)
2007 goto out;
2008
2009 r = -EFAULT;
2010 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2011 goto out;
2012 r = 0;
2013 break;
2014 }
2015 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
2016 u64 mce_cap;
2017
2018 mce_cap = KVM_MCE_CAP_SUPPORTED;
2019 r = -EFAULT;
2020 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
2021 goto out;
2022 r = 0;
2023 break;
2024 }
2025 default:
2026 r = -EINVAL;
2027 }
2028 out:
2029 return r;
2030 }
2031
2032 static void wbinvd_ipi(void *garbage)
2033 {
2034 wbinvd();
2035 }
2036
2037 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
2038 {
2039 return vcpu->kvm->arch.iommu_domain &&
2040 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY);
2041 }
2042
2043 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2044 {
2045 /* Address WBINVD may be executed by guest */
2046 if (need_emulate_wbinvd(vcpu)) {
2047 if (kvm_x86_ops->has_wbinvd_exit())
2048 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
2049 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
2050 smp_call_function_single(vcpu->cpu,
2051 wbinvd_ipi, NULL, 1);
2052 }
2053
2054 kvm_x86_ops->vcpu_load(vcpu, cpu);
2055 if (unlikely(vcpu->cpu != cpu) || check_tsc_unstable()) {
2056 /* Make sure TSC doesn't go backwards */
2057 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
2058 native_read_tsc() - vcpu->arch.last_host_tsc;
2059 if (tsc_delta < 0)
2060 mark_tsc_unstable("KVM discovered backwards TSC");
2061 if (check_tsc_unstable()) {
2062 kvm_x86_ops->adjust_tsc_offset(vcpu, -tsc_delta);
2063 vcpu->arch.tsc_catchup = 1;
2064 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2065 }
2066 if (vcpu->cpu != cpu)
2067 kvm_migrate_timers(vcpu);
2068 vcpu->cpu = cpu;
2069 }
2070 }
2071
2072 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2073 {
2074 kvm_x86_ops->vcpu_put(vcpu);
2075 kvm_put_guest_fpu(vcpu);
2076 vcpu->arch.last_host_tsc = native_read_tsc();
2077 }
2078
2079 static int is_efer_nx(void)
2080 {
2081 unsigned long long efer = 0;
2082
2083 rdmsrl_safe(MSR_EFER, &efer);
2084 return efer & EFER_NX;
2085 }
2086
2087 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2088 {
2089 int i;
2090 struct kvm_cpuid_entry2 *e, *entry;
2091
2092 entry = NULL;
2093 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2094 e = &vcpu->arch.cpuid_entries[i];
2095 if (e->function == 0x80000001) {
2096 entry = e;
2097 break;
2098 }
2099 }
2100 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2101 entry->edx &= ~(1 << 20);
2102 printk(KERN_INFO "kvm: guest NX capability removed\n");
2103 }
2104 }
2105
2106 /* when an old userspace process fills a new kernel module */
2107 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2108 struct kvm_cpuid *cpuid,
2109 struct kvm_cpuid_entry __user *entries)
2110 {
2111 int r, i;
2112 struct kvm_cpuid_entry *cpuid_entries;
2113
2114 r = -E2BIG;
2115 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2116 goto out;
2117 r = -ENOMEM;
2118 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2119 if (!cpuid_entries)
2120 goto out;
2121 r = -EFAULT;
2122 if (copy_from_user(cpuid_entries, entries,
2123 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2124 goto out_free;
2125 for (i = 0; i < cpuid->nent; i++) {
2126 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2127 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2128 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2129 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2130 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2131 vcpu->arch.cpuid_entries[i].index = 0;
2132 vcpu->arch.cpuid_entries[i].flags = 0;
2133 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2134 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2135 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2136 }
2137 vcpu->arch.cpuid_nent = cpuid->nent;
2138 cpuid_fix_nx_cap(vcpu);
2139 r = 0;
2140 kvm_apic_set_version(vcpu);
2141 kvm_x86_ops->cpuid_update(vcpu);
2142 update_cpuid(vcpu);
2143
2144 out_free:
2145 vfree(cpuid_entries);
2146 out:
2147 return r;
2148 }
2149
2150 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2151 struct kvm_cpuid2 *cpuid,
2152 struct kvm_cpuid_entry2 __user *entries)
2153 {
2154 int r;
2155
2156 r = -E2BIG;
2157 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2158 goto out;
2159 r = -EFAULT;
2160 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2161 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2162 goto out;
2163 vcpu->arch.cpuid_nent = cpuid->nent;
2164 kvm_apic_set_version(vcpu);
2165 kvm_x86_ops->cpuid_update(vcpu);
2166 update_cpuid(vcpu);
2167 return 0;
2168
2169 out:
2170 return r;
2171 }
2172
2173 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2174 struct kvm_cpuid2 *cpuid,
2175 struct kvm_cpuid_entry2 __user *entries)
2176 {
2177 int r;
2178
2179 r = -E2BIG;
2180 if (cpuid->nent < vcpu->arch.cpuid_nent)
2181 goto out;
2182 r = -EFAULT;
2183 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2184 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2185 goto out;
2186 return 0;
2187
2188 out:
2189 cpuid->nent = vcpu->arch.cpuid_nent;
2190 return r;
2191 }
2192
2193 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2194 u32 index)
2195 {
2196 entry->function = function;
2197 entry->index = index;
2198 cpuid_count(entry->function, entry->index,
2199 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2200 entry->flags = 0;
2201 }
2202
2203 #define F(x) bit(X86_FEATURE_##x)
2204
2205 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2206 u32 index, int *nent, int maxnent)
2207 {
2208 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2209 #ifdef CONFIG_X86_64
2210 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2211 ? F(GBPAGES) : 0;
2212 unsigned f_lm = F(LM);
2213 #else
2214 unsigned f_gbpages = 0;
2215 unsigned f_lm = 0;
2216 #endif
2217 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2218
2219 /* cpuid 1.edx */
2220 const u32 kvm_supported_word0_x86_features =
2221 F(FPU) | F(VME) | F(DE) | F(PSE) |
2222 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2223 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2224 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2225 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2226 0 /* Reserved, DS, ACPI */ | F(MMX) |
2227 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2228 0 /* HTT, TM, Reserved, PBE */;
2229 /* cpuid 0x80000001.edx */
2230 const u32 kvm_supported_word1_x86_features =
2231 F(FPU) | F(VME) | F(DE) | F(PSE) |
2232 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2233 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2234 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2235 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2236 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2237 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2238 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2239 /* cpuid 1.ecx */
2240 const u32 kvm_supported_word4_x86_features =
2241 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
2242 0 /* DS-CPL, VMX, SMX, EST */ |
2243 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2244 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2245 0 /* Reserved, DCA */ | F(XMM4_1) |
2246 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2247 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2248 F(F16C);
2249 /* cpuid 0x80000001.ecx */
2250 const u32 kvm_supported_word6_x86_features =
2251 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2252 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2253 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2254 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2255
2256 /* all calls to cpuid_count() should be made on the same cpu */
2257 get_cpu();
2258 do_cpuid_1_ent(entry, function, index);
2259 ++*nent;
2260
2261 switch (function) {
2262 case 0:
2263 entry->eax = min(entry->eax, (u32)0xd);
2264 break;
2265 case 1:
2266 entry->edx &= kvm_supported_word0_x86_features;
2267 entry->ecx &= kvm_supported_word4_x86_features;
2268 /* we support x2apic emulation even if host does not support
2269 * it since we emulate x2apic in software */
2270 entry->ecx |= F(X2APIC);
2271 break;
2272 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2273 * may return different values. This forces us to get_cpu() before
2274 * issuing the first command, and also to emulate this annoying behavior
2275 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2276 case 2: {
2277 int t, times = entry->eax & 0xff;
2278
2279 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2280 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2281 for (t = 1; t < times && *nent < maxnent; ++t) {
2282 do_cpuid_1_ent(&entry[t], function, 0);
2283 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2284 ++*nent;
2285 }
2286 break;
2287 }
2288 /* function 4 and 0xb have additional index. */
2289 case 4: {
2290 int i, cache_type;
2291
2292 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2293 /* read more entries until cache_type is zero */
2294 for (i = 1; *nent < maxnent; ++i) {
2295 cache_type = entry[i - 1].eax & 0x1f;
2296 if (!cache_type)
2297 break;
2298 do_cpuid_1_ent(&entry[i], function, i);
2299 entry[i].flags |=
2300 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2301 ++*nent;
2302 }
2303 break;
2304 }
2305 case 0xb: {
2306 int i, level_type;
2307
2308 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2309 /* read more entries until level_type is zero */
2310 for (i = 1; *nent < maxnent; ++i) {
2311 level_type = entry[i - 1].ecx & 0xff00;
2312 if (!level_type)
2313 break;
2314 do_cpuid_1_ent(&entry[i], function, i);
2315 entry[i].flags |=
2316 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2317 ++*nent;
2318 }
2319 break;
2320 }
2321 case 0xd: {
2322 int i;
2323
2324 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2325 for (i = 1; *nent < maxnent; ++i) {
2326 if (entry[i - 1].eax == 0 && i != 2)
2327 break;
2328 do_cpuid_1_ent(&entry[i], function, i);
2329 entry[i].flags |=
2330 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2331 ++*nent;
2332 }
2333 break;
2334 }
2335 case KVM_CPUID_SIGNATURE: {
2336 char signature[12] = "KVMKVMKVM\0\0";
2337 u32 *sigptr = (u32 *)signature;
2338 entry->eax = 0;
2339 entry->ebx = sigptr[0];
2340 entry->ecx = sigptr[1];
2341 entry->edx = sigptr[2];
2342 break;
2343 }
2344 case KVM_CPUID_FEATURES:
2345 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2346 (1 << KVM_FEATURE_NOP_IO_DELAY) |
2347 (1 << KVM_FEATURE_CLOCKSOURCE2) |
2348 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2349 entry->ebx = 0;
2350 entry->ecx = 0;
2351 entry->edx = 0;
2352 break;
2353 case 0x80000000:
2354 entry->eax = min(entry->eax, 0x8000001a);
2355 break;
2356 case 0x80000001:
2357 entry->edx &= kvm_supported_word1_x86_features;
2358 entry->ecx &= kvm_supported_word6_x86_features;
2359 break;
2360 }
2361
2362 kvm_x86_ops->set_supported_cpuid(function, entry);
2363
2364 put_cpu();
2365 }
2366
2367 #undef F
2368
2369 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2370 struct kvm_cpuid_entry2 __user *entries)
2371 {
2372 struct kvm_cpuid_entry2 *cpuid_entries;
2373 int limit, nent = 0, r = -E2BIG;
2374 u32 func;
2375
2376 if (cpuid->nent < 1)
2377 goto out;
2378 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2379 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2380 r = -ENOMEM;
2381 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2382 if (!cpuid_entries)
2383 goto out;
2384
2385 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2386 limit = cpuid_entries[0].eax;
2387 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2388 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2389 &nent, cpuid->nent);
2390 r = -E2BIG;
2391 if (nent >= cpuid->nent)
2392 goto out_free;
2393
2394 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2395 limit = cpuid_entries[nent - 1].eax;
2396 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2397 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2398 &nent, cpuid->nent);
2399
2400
2401
2402 r = -E2BIG;
2403 if (nent >= cpuid->nent)
2404 goto out_free;
2405
2406 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2407 cpuid->nent);
2408
2409 r = -E2BIG;
2410 if (nent >= cpuid->nent)
2411 goto out_free;
2412
2413 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2414 cpuid->nent);
2415
2416 r = -E2BIG;
2417 if (nent >= cpuid->nent)
2418 goto out_free;
2419
2420 r = -EFAULT;
2421 if (copy_to_user(entries, cpuid_entries,
2422 nent * sizeof(struct kvm_cpuid_entry2)))
2423 goto out_free;
2424 cpuid->nent = nent;
2425 r = 0;
2426
2427 out_free:
2428 vfree(cpuid_entries);
2429 out:
2430 return r;
2431 }
2432
2433 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2434 struct kvm_lapic_state *s)
2435 {
2436 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
2437
2438 return 0;
2439 }
2440
2441 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
2442 struct kvm_lapic_state *s)
2443 {
2444 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
2445 kvm_apic_post_state_restore(vcpu);
2446 update_cr8_intercept(vcpu);
2447
2448 return 0;
2449 }
2450
2451 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
2452 struct kvm_interrupt *irq)
2453 {
2454 if (irq->irq < 0 || irq->irq >= 256)
2455 return -EINVAL;
2456 if (irqchip_in_kernel(vcpu->kvm))
2457 return -ENXIO;
2458
2459 kvm_queue_interrupt(vcpu, irq->irq, false);
2460 kvm_make_request(KVM_REQ_EVENT, vcpu);
2461
2462 return 0;
2463 }
2464
2465 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2466 {
2467 kvm_inject_nmi(vcpu);
2468
2469 return 0;
2470 }
2471
2472 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2473 struct kvm_tpr_access_ctl *tac)
2474 {
2475 if (tac->flags)
2476 return -EINVAL;
2477 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2478 return 0;
2479 }
2480
2481 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2482 u64 mcg_cap)
2483 {
2484 int r;
2485 unsigned bank_num = mcg_cap & 0xff, bank;
2486
2487 r = -EINVAL;
2488 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2489 goto out;
2490 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2491 goto out;
2492 r = 0;
2493 vcpu->arch.mcg_cap = mcg_cap;
2494 /* Init IA32_MCG_CTL to all 1s */
2495 if (mcg_cap & MCG_CTL_P)
2496 vcpu->arch.mcg_ctl = ~(u64)0;
2497 /* Init IA32_MCi_CTL to all 1s */
2498 for (bank = 0; bank < bank_num; bank++)
2499 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2500 out:
2501 return r;
2502 }
2503
2504 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2505 struct kvm_x86_mce *mce)
2506 {
2507 u64 mcg_cap = vcpu->arch.mcg_cap;
2508 unsigned bank_num = mcg_cap & 0xff;
2509 u64 *banks = vcpu->arch.mce_banks;
2510
2511 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2512 return -EINVAL;
2513 /*
2514 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2515 * reporting is disabled
2516 */
2517 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2518 vcpu->arch.mcg_ctl != ~(u64)0)
2519 return 0;
2520 banks += 4 * mce->bank;
2521 /*
2522 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2523 * reporting is disabled for the bank
2524 */
2525 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2526 return 0;
2527 if (mce->status & MCI_STATUS_UC) {
2528 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2529 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2530 printk(KERN_DEBUG "kvm: set_mce: "
2531 "injects mce exception while "
2532 "previous one is in progress!\n");
2533 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2534 return 0;
2535 }
2536 if (banks[1] & MCI_STATUS_VAL)
2537 mce->status |= MCI_STATUS_OVER;
2538 banks[2] = mce->addr;
2539 banks[3] = mce->misc;
2540 vcpu->arch.mcg_status = mce->mcg_status;
2541 banks[1] = mce->status;
2542 kvm_queue_exception(vcpu, MC_VECTOR);
2543 } else if (!(banks[1] & MCI_STATUS_VAL)
2544 || !(banks[1] & MCI_STATUS_UC)) {
2545 if (banks[1] & MCI_STATUS_VAL)
2546 mce->status |= MCI_STATUS_OVER;
2547 banks[2] = mce->addr;
2548 banks[3] = mce->misc;
2549 banks[1] = mce->status;
2550 } else
2551 banks[1] |= MCI_STATUS_OVER;
2552 return 0;
2553 }
2554
2555 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2556 struct kvm_vcpu_events *events)
2557 {
2558 events->exception.injected =
2559 vcpu->arch.exception.pending &&
2560 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2561 events->exception.nr = vcpu->arch.exception.nr;
2562 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2563 events->exception.error_code = vcpu->arch.exception.error_code;
2564
2565 events->interrupt.injected =
2566 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2567 events->interrupt.nr = vcpu->arch.interrupt.nr;
2568 events->interrupt.soft = 0;
2569 events->interrupt.shadow =
2570 kvm_x86_ops->get_interrupt_shadow(vcpu,
2571 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2572
2573 events->nmi.injected = vcpu->arch.nmi_injected;
2574 events->nmi.pending = vcpu->arch.nmi_pending;
2575 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2576
2577 events->sipi_vector = vcpu->arch.sipi_vector;
2578
2579 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2580 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2581 | KVM_VCPUEVENT_VALID_SHADOW);
2582 }
2583
2584 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2585 struct kvm_vcpu_events *events)
2586 {
2587 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2588 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2589 | KVM_VCPUEVENT_VALID_SHADOW))
2590 return -EINVAL;
2591
2592 vcpu->arch.exception.pending = events->exception.injected;
2593 vcpu->arch.exception.nr = events->exception.nr;
2594 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2595 vcpu->arch.exception.error_code = events->exception.error_code;
2596
2597 vcpu->arch.interrupt.pending = events->interrupt.injected;
2598 vcpu->arch.interrupt.nr = events->interrupt.nr;
2599 vcpu->arch.interrupt.soft = events->interrupt.soft;
2600 if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
2601 kvm_pic_clear_isr_ack(vcpu->kvm);
2602 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2603 kvm_x86_ops->set_interrupt_shadow(vcpu,
2604 events->interrupt.shadow);
2605
2606 vcpu->arch.nmi_injected = events->nmi.injected;
2607 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2608 vcpu->arch.nmi_pending = events->nmi.pending;
2609 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2610
2611 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2612 vcpu->arch.sipi_vector = events->sipi_vector;
2613
2614 kvm_make_request(KVM_REQ_EVENT, vcpu);
2615
2616 return 0;
2617 }
2618
2619 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
2620 struct kvm_debugregs *dbgregs)
2621 {
2622 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
2623 dbgregs->dr6 = vcpu->arch.dr6;
2624 dbgregs->dr7 = vcpu->arch.dr7;
2625 dbgregs->flags = 0;
2626 }
2627
2628 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
2629 struct kvm_debugregs *dbgregs)
2630 {
2631 if (dbgregs->flags)
2632 return -EINVAL;
2633
2634 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
2635 vcpu->arch.dr6 = dbgregs->dr6;
2636 vcpu->arch.dr7 = dbgregs->dr7;
2637
2638 return 0;
2639 }
2640
2641 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
2642 struct kvm_xsave *guest_xsave)
2643 {
2644 if (cpu_has_xsave)
2645 memcpy(guest_xsave->region,
2646 &vcpu->arch.guest_fpu.state->xsave,
2647 xstate_size);
2648 else {
2649 memcpy(guest_xsave->region,
2650 &vcpu->arch.guest_fpu.state->fxsave,
2651 sizeof(struct i387_fxsave_struct));
2652 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
2653 XSTATE_FPSSE;
2654 }
2655 }
2656
2657 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
2658 struct kvm_xsave *guest_xsave)
2659 {
2660 u64 xstate_bv =
2661 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
2662
2663 if (cpu_has_xsave)
2664 memcpy(&vcpu->arch.guest_fpu.state->xsave,
2665 guest_xsave->region, xstate_size);
2666 else {
2667 if (xstate_bv & ~XSTATE_FPSSE)
2668 return -EINVAL;
2669 memcpy(&vcpu->arch.guest_fpu.state->fxsave,
2670 guest_xsave->region, sizeof(struct i387_fxsave_struct));
2671 }
2672 return 0;
2673 }
2674
2675 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
2676 struct kvm_xcrs *guest_xcrs)
2677 {
2678 if (!cpu_has_xsave) {
2679 guest_xcrs->nr_xcrs = 0;
2680 return;
2681 }
2682
2683 guest_xcrs->nr_xcrs = 1;
2684 guest_xcrs->flags = 0;
2685 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
2686 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
2687 }
2688
2689 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
2690 struct kvm_xcrs *guest_xcrs)
2691 {
2692 int i, r = 0;
2693
2694 if (!cpu_has_xsave)
2695 return -EINVAL;
2696
2697 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
2698 return -EINVAL;
2699
2700 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
2701 /* Only support XCR0 currently */
2702 if (guest_xcrs->xcrs[0].xcr == XCR_XFEATURE_ENABLED_MASK) {
2703 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
2704 guest_xcrs->xcrs[0].value);
2705 break;
2706 }
2707 if (r)
2708 r = -EINVAL;
2709 return r;
2710 }
2711
2712 long kvm_arch_vcpu_ioctl(struct file *filp,
2713 unsigned int ioctl, unsigned long arg)
2714 {
2715 struct kvm_vcpu *vcpu = filp->private_data;
2716 void __user *argp = (void __user *)arg;
2717 int r;
2718 union {
2719 struct kvm_lapic_state *lapic;
2720 struct kvm_xsave *xsave;
2721 struct kvm_xcrs *xcrs;
2722 void *buffer;
2723 } u;
2724
2725 u.buffer = NULL;
2726 switch (ioctl) {
2727 case KVM_GET_LAPIC: {
2728 r = -EINVAL;
2729 if (!vcpu->arch.apic)
2730 goto out;
2731 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2732
2733 r = -ENOMEM;
2734 if (!u.lapic)
2735 goto out;
2736 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
2737 if (r)
2738 goto out;
2739 r = -EFAULT;
2740 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
2741 goto out;
2742 r = 0;
2743 break;
2744 }
2745 case KVM_SET_LAPIC: {
2746 r = -EINVAL;
2747 if (!vcpu->arch.apic)
2748 goto out;
2749 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2750 r = -ENOMEM;
2751 if (!u.lapic)
2752 goto out;
2753 r = -EFAULT;
2754 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
2755 goto out;
2756 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
2757 if (r)
2758 goto out;
2759 r = 0;
2760 break;
2761 }
2762 case KVM_INTERRUPT: {
2763 struct kvm_interrupt irq;
2764
2765 r = -EFAULT;
2766 if (copy_from_user(&irq, argp, sizeof irq))
2767 goto out;
2768 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2769 if (r)
2770 goto out;
2771 r = 0;
2772 break;
2773 }
2774 case KVM_NMI: {
2775 r = kvm_vcpu_ioctl_nmi(vcpu);
2776 if (r)
2777 goto out;
2778 r = 0;
2779 break;
2780 }
2781 case KVM_SET_CPUID: {
2782 struct kvm_cpuid __user *cpuid_arg = argp;
2783 struct kvm_cpuid cpuid;
2784
2785 r = -EFAULT;
2786 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2787 goto out;
2788 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2789 if (r)
2790 goto out;
2791 break;
2792 }
2793 case KVM_SET_CPUID2: {
2794 struct kvm_cpuid2 __user *cpuid_arg = argp;
2795 struct kvm_cpuid2 cpuid;
2796
2797 r = -EFAULT;
2798 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2799 goto out;
2800 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2801 cpuid_arg->entries);
2802 if (r)
2803 goto out;
2804 break;
2805 }
2806 case KVM_GET_CPUID2: {
2807 struct kvm_cpuid2 __user *cpuid_arg = argp;
2808 struct kvm_cpuid2 cpuid;
2809
2810 r = -EFAULT;
2811 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2812 goto out;
2813 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2814 cpuid_arg->entries);
2815 if (r)
2816 goto out;
2817 r = -EFAULT;
2818 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2819 goto out;
2820 r = 0;
2821 break;
2822 }
2823 case KVM_GET_MSRS:
2824 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2825 break;
2826 case KVM_SET_MSRS:
2827 r = msr_io(vcpu, argp, do_set_msr, 0);
2828 break;
2829 case KVM_TPR_ACCESS_REPORTING: {
2830 struct kvm_tpr_access_ctl tac;
2831
2832 r = -EFAULT;
2833 if (copy_from_user(&tac, argp, sizeof tac))
2834 goto out;
2835 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2836 if (r)
2837 goto out;
2838 r = -EFAULT;
2839 if (copy_to_user(argp, &tac, sizeof tac))
2840 goto out;
2841 r = 0;
2842 break;
2843 };
2844 case KVM_SET_VAPIC_ADDR: {
2845 struct kvm_vapic_addr va;
2846
2847 r = -EINVAL;
2848 if (!irqchip_in_kernel(vcpu->kvm))
2849 goto out;
2850 r = -EFAULT;
2851 if (copy_from_user(&va, argp, sizeof va))
2852 goto out;
2853 r = 0;
2854 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2855 break;
2856 }
2857 case KVM_X86_SETUP_MCE: {
2858 u64 mcg_cap;
2859
2860 r = -EFAULT;
2861 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2862 goto out;
2863 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2864 break;
2865 }
2866 case KVM_X86_SET_MCE: {
2867 struct kvm_x86_mce mce;
2868
2869 r = -EFAULT;
2870 if (copy_from_user(&mce, argp, sizeof mce))
2871 goto out;
2872 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2873 break;
2874 }
2875 case KVM_GET_VCPU_EVENTS: {
2876 struct kvm_vcpu_events events;
2877
2878 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2879
2880 r = -EFAULT;
2881 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2882 break;
2883 r = 0;
2884 break;
2885 }
2886 case KVM_SET_VCPU_EVENTS: {
2887 struct kvm_vcpu_events events;
2888
2889 r = -EFAULT;
2890 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2891 break;
2892
2893 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2894 break;
2895 }
2896 case KVM_GET_DEBUGREGS: {
2897 struct kvm_debugregs dbgregs;
2898
2899 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
2900
2901 r = -EFAULT;
2902 if (copy_to_user(argp, &dbgregs,
2903 sizeof(struct kvm_debugregs)))
2904 break;
2905 r = 0;
2906 break;
2907 }
2908 case KVM_SET_DEBUGREGS: {
2909 struct kvm_debugregs dbgregs;
2910
2911 r = -EFAULT;
2912 if (copy_from_user(&dbgregs, argp,
2913 sizeof(struct kvm_debugregs)))
2914 break;
2915
2916 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
2917 break;
2918 }
2919 case KVM_GET_XSAVE: {
2920 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2921 r = -ENOMEM;
2922 if (!u.xsave)
2923 break;
2924
2925 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
2926
2927 r = -EFAULT;
2928 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
2929 break;
2930 r = 0;
2931 break;
2932 }
2933 case KVM_SET_XSAVE: {
2934 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
2935 r = -ENOMEM;
2936 if (!u.xsave)
2937 break;
2938
2939 r = -EFAULT;
2940 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
2941 break;
2942
2943 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
2944 break;
2945 }
2946 case KVM_GET_XCRS: {
2947 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2948 r = -ENOMEM;
2949 if (!u.xcrs)
2950 break;
2951
2952 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
2953
2954 r = -EFAULT;
2955 if (copy_to_user(argp, u.xcrs,
2956 sizeof(struct kvm_xcrs)))
2957 break;
2958 r = 0;
2959 break;
2960 }
2961 case KVM_SET_XCRS: {
2962 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
2963 r = -ENOMEM;
2964 if (!u.xcrs)
2965 break;
2966
2967 r = -EFAULT;
2968 if (copy_from_user(u.xcrs, argp,
2969 sizeof(struct kvm_xcrs)))
2970 break;
2971
2972 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
2973 break;
2974 }
2975 default:
2976 r = -EINVAL;
2977 }
2978 out:
2979 kfree(u.buffer);
2980 return r;
2981 }
2982
2983 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2984 {
2985 int ret;
2986
2987 if (addr > (unsigned int)(-3 * PAGE_SIZE))
2988 return -1;
2989 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2990 return ret;
2991 }
2992
2993 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2994 u64 ident_addr)
2995 {
2996 kvm->arch.ept_identity_map_addr = ident_addr;
2997 return 0;
2998 }
2999
3000 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
3001 u32 kvm_nr_mmu_pages)
3002 {
3003 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
3004 return -EINVAL;
3005
3006 mutex_lock(&kvm->slots_lock);
3007 spin_lock(&kvm->mmu_lock);
3008
3009 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
3010 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
3011
3012 spin_unlock(&kvm->mmu_lock);
3013 mutex_unlock(&kvm->slots_lock);
3014 return 0;
3015 }
3016
3017 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
3018 {
3019 return kvm->arch.n_max_mmu_pages;
3020 }
3021
3022 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3023 {
3024 int r;
3025
3026 r = 0;
3027 switch (chip->chip_id) {
3028 case KVM_IRQCHIP_PIC_MASTER:
3029 memcpy(&chip->chip.pic,
3030 &pic_irqchip(kvm)->pics[0],
3031 sizeof(struct kvm_pic_state));
3032 break;
3033 case KVM_IRQCHIP_PIC_SLAVE:
3034 memcpy(&chip->chip.pic,
3035 &pic_irqchip(kvm)->pics[1],
3036 sizeof(struct kvm_pic_state));
3037 break;
3038 case KVM_IRQCHIP_IOAPIC:
3039 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
3040 break;
3041 default:
3042 r = -EINVAL;
3043 break;
3044 }
3045 return r;
3046 }
3047
3048 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
3049 {
3050 int r;
3051
3052 r = 0;
3053 switch (chip->chip_id) {
3054 case KVM_IRQCHIP_PIC_MASTER:
3055 spin_lock(&pic_irqchip(kvm)->lock);
3056 memcpy(&pic_irqchip(kvm)->pics[0],
3057 &chip->chip.pic,
3058 sizeof(struct kvm_pic_state));
3059 spin_unlock(&pic_irqchip(kvm)->lock);
3060 break;
3061 case KVM_IRQCHIP_PIC_SLAVE:
3062 spin_lock(&pic_irqchip(kvm)->lock);
3063 memcpy(&pic_irqchip(kvm)->pics[1],
3064 &chip->chip.pic,
3065 sizeof(struct kvm_pic_state));
3066 spin_unlock(&pic_irqchip(kvm)->lock);
3067 break;
3068 case KVM_IRQCHIP_IOAPIC:
3069 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
3070 break;
3071 default:
3072 r = -EINVAL;
3073 break;
3074 }
3075 kvm_pic_update_irq(pic_irqchip(kvm));
3076 return r;
3077 }
3078
3079 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3080 {
3081 int r = 0;
3082
3083 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3084 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
3085 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3086 return r;
3087 }
3088
3089 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
3090 {
3091 int r = 0;
3092
3093 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3094 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
3095 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
3096 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3097 return r;
3098 }
3099
3100 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3101 {
3102 int r = 0;
3103
3104 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3105 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
3106 sizeof(ps->channels));
3107 ps->flags = kvm->arch.vpit->pit_state.flags;
3108 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3109 return r;
3110 }
3111
3112 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
3113 {
3114 int r = 0, start = 0;
3115 u32 prev_legacy, cur_legacy;
3116 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3117 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
3118 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
3119 if (!prev_legacy && cur_legacy)
3120 start = 1;
3121 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
3122 sizeof(kvm->arch.vpit->pit_state.channels));
3123 kvm->arch.vpit->pit_state.flags = ps->flags;
3124 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
3125 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3126 return r;
3127 }
3128
3129 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3130 struct kvm_reinject_control *control)
3131 {
3132 if (!kvm->arch.vpit)
3133 return -ENXIO;
3134 mutex_lock(&kvm->arch.vpit->pit_state.lock);
3135 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
3136 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
3137 return 0;
3138 }
3139
3140 /*
3141 * Get (and clear) the dirty memory log for a memory slot.
3142 */
3143 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3144 struct kvm_dirty_log *log)
3145 {
3146 int r, i;
3147 struct kvm_memory_slot *memslot;
3148 unsigned long n;
3149 unsigned long is_dirty = 0;
3150
3151 mutex_lock(&kvm->slots_lock);
3152
3153 r = -EINVAL;
3154 if (log->slot >= KVM_MEMORY_SLOTS)
3155 goto out;
3156
3157 memslot = &kvm->memslots->memslots[log->slot];
3158 r = -ENOENT;
3159 if (!memslot->dirty_bitmap)
3160 goto out;
3161
3162 n = kvm_dirty_bitmap_bytes(memslot);
3163
3164 for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3165 is_dirty = memslot->dirty_bitmap[i];
3166
3167 /* If nothing is dirty, don't bother messing with page tables. */
3168 if (is_dirty) {
3169 struct kvm_memslots *slots, *old_slots;
3170 unsigned long *dirty_bitmap;
3171
3172 r = -ENOMEM;
3173 dirty_bitmap = vmalloc(n);
3174 if (!dirty_bitmap)
3175 goto out;
3176 memset(dirty_bitmap, 0, n);
3177
3178 r = -ENOMEM;
3179 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
3180 if (!slots) {
3181 vfree(dirty_bitmap);
3182 goto out;
3183 }
3184 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
3185 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
3186
3187 old_slots = kvm->memslots;
3188 rcu_assign_pointer(kvm->memslots, slots);
3189 synchronize_srcu_expedited(&kvm->srcu);
3190 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3191 kfree(old_slots);
3192
3193 spin_lock(&kvm->mmu_lock);
3194 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3195 spin_unlock(&kvm->mmu_lock);
3196
3197 r = -EFAULT;
3198 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n)) {
3199 vfree(dirty_bitmap);
3200 goto out;
3201 }
3202 vfree(dirty_bitmap);
3203 } else {
3204 r = -EFAULT;
3205 if (clear_user(log->dirty_bitmap, n))
3206 goto out;
3207 }
3208
3209 r = 0;
3210 out:
3211 mutex_unlock(&kvm->slots_lock);
3212 return r;
3213 }
3214
3215 long kvm_arch_vm_ioctl(struct file *filp,
3216 unsigned int ioctl, unsigned long arg)
3217 {
3218 struct kvm *kvm = filp->private_data;
3219 void __user *argp = (void __user *)arg;
3220 int r = -ENOTTY;
3221 /*
3222 * This union makes it completely explicit to gcc-3.x
3223 * that these two variables' stack usage should be
3224 * combined, not added together.
3225 */
3226 union {
3227 struct kvm_pit_state ps;
3228 struct kvm_pit_state2 ps2;
3229 struct kvm_pit_config pit_config;
3230 } u;
3231
3232 switch (ioctl) {
3233 case KVM_SET_TSS_ADDR:
3234 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
3235 if (r < 0)
3236 goto out;
3237 break;
3238 case KVM_SET_IDENTITY_MAP_ADDR: {
3239 u64 ident_addr;
3240
3241 r = -EFAULT;
3242 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
3243 goto out;
3244 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
3245 if (r < 0)
3246 goto out;
3247 break;
3248 }
3249 case KVM_SET_NR_MMU_PAGES:
3250 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
3251 if (r)
3252 goto out;
3253 break;
3254 case KVM_GET_NR_MMU_PAGES:
3255 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
3256 break;
3257 case KVM_CREATE_IRQCHIP: {
3258 struct kvm_pic *vpic;
3259
3260 mutex_lock(&kvm->lock);
3261 r = -EEXIST;
3262 if (kvm->arch.vpic)
3263 goto create_irqchip_unlock;
3264 r = -ENOMEM;
3265 vpic = kvm_create_pic(kvm);
3266 if (vpic) {
3267 r = kvm_ioapic_init(kvm);
3268 if (r) {
3269 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
3270 &vpic->dev);
3271 kfree(vpic);
3272 goto create_irqchip_unlock;
3273 }
3274 } else
3275 goto create_irqchip_unlock;
3276 smp_wmb();
3277 kvm->arch.vpic = vpic;
3278 smp_wmb();
3279 r = kvm_setup_default_irq_routing(kvm);
3280 if (r) {
3281 mutex_lock(&kvm->irq_lock);
3282 kvm_ioapic_destroy(kvm);
3283 kvm_destroy_pic(kvm);
3284 mutex_unlock(&kvm->irq_lock);
3285 }
3286 create_irqchip_unlock:
3287 mutex_unlock(&kvm->lock);
3288 break;
3289 }
3290 case KVM_CREATE_PIT:
3291 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
3292 goto create_pit;
3293 case KVM_CREATE_PIT2:
3294 r = -EFAULT;
3295 if (copy_from_user(&u.pit_config, argp,
3296 sizeof(struct kvm_pit_config)))
3297 goto out;
3298 create_pit:
3299 mutex_lock(&kvm->slots_lock);
3300 r = -EEXIST;
3301 if (kvm->arch.vpit)
3302 goto create_pit_unlock;
3303 r = -ENOMEM;
3304 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
3305 if (kvm->arch.vpit)
3306 r = 0;
3307 create_pit_unlock:
3308 mutex_unlock(&kvm->slots_lock);
3309 break;
3310 case KVM_IRQ_LINE_STATUS:
3311 case KVM_IRQ_LINE: {
3312 struct kvm_irq_level irq_event;
3313
3314 r = -EFAULT;
3315 if (copy_from_user(&irq_event, argp, sizeof irq_event))
3316 goto out;
3317 r = -ENXIO;
3318 if (irqchip_in_kernel(kvm)) {
3319 __s32 status;
3320 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
3321 irq_event.irq, irq_event.level);
3322 if (ioctl == KVM_IRQ_LINE_STATUS) {
3323 r = -EFAULT;
3324 irq_event.status = status;
3325 if (copy_to_user(argp, &irq_event,
3326 sizeof irq_event))
3327 goto out;
3328 }
3329 r = 0;
3330 }
3331 break;
3332 }
3333 case KVM_GET_IRQCHIP: {
3334 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3335 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3336
3337 r = -ENOMEM;
3338 if (!chip)
3339 goto out;
3340 r = -EFAULT;
3341 if (copy_from_user(chip, argp, sizeof *chip))
3342 goto get_irqchip_out;
3343 r = -ENXIO;
3344 if (!irqchip_in_kernel(kvm))
3345 goto get_irqchip_out;
3346 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
3347 if (r)
3348 goto get_irqchip_out;
3349 r = -EFAULT;
3350 if (copy_to_user(argp, chip, sizeof *chip))
3351 goto get_irqchip_out;
3352 r = 0;
3353 get_irqchip_out:
3354 kfree(chip);
3355 if (r)
3356 goto out;
3357 break;
3358 }
3359 case KVM_SET_IRQCHIP: {
3360 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3361 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
3362
3363 r = -ENOMEM;
3364 if (!chip)
3365 goto out;
3366 r = -EFAULT;
3367 if (copy_from_user(chip, argp, sizeof *chip))
3368 goto set_irqchip_out;
3369 r = -ENXIO;
3370 if (!irqchip_in_kernel(kvm))
3371 goto set_irqchip_out;
3372 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
3373 if (r)
3374 goto set_irqchip_out;
3375 r = 0;
3376 set_irqchip_out:
3377 kfree(chip);
3378 if (r)
3379 goto out;
3380 break;
3381 }
3382 case KVM_GET_PIT: {
3383 r = -EFAULT;
3384 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
3385 goto out;
3386 r = -ENXIO;
3387 if (!kvm->arch.vpit)
3388 goto out;
3389 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
3390 if (r)
3391 goto out;
3392 r = -EFAULT;
3393 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
3394 goto out;
3395 r = 0;
3396 break;
3397 }
3398 case KVM_SET_PIT: {
3399 r = -EFAULT;
3400 if (copy_from_user(&u.ps, argp, sizeof u.ps))
3401 goto out;
3402 r = -ENXIO;
3403 if (!kvm->arch.vpit)
3404 goto out;
3405 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
3406 if (r)
3407 goto out;
3408 r = 0;
3409 break;
3410 }
3411 case KVM_GET_PIT2: {
3412 r = -ENXIO;
3413 if (!kvm->arch.vpit)
3414 goto out;
3415 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
3416 if (r)
3417 goto out;
3418 r = -EFAULT;
3419 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
3420 goto out;
3421 r = 0;
3422 break;
3423 }
3424 case KVM_SET_PIT2: {
3425 r = -EFAULT;
3426 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
3427 goto out;
3428 r = -ENXIO;
3429 if (!kvm->arch.vpit)
3430 goto out;
3431 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
3432 if (r)
3433 goto out;
3434 r = 0;
3435 break;
3436 }
3437 case KVM_REINJECT_CONTROL: {
3438 struct kvm_reinject_control control;
3439 r = -EFAULT;
3440 if (copy_from_user(&control, argp, sizeof(control)))
3441 goto out;
3442 r = kvm_vm_ioctl_reinject(kvm, &control);
3443 if (r)
3444 goto out;
3445 r = 0;
3446 break;
3447 }
3448 case KVM_XEN_HVM_CONFIG: {
3449 r = -EFAULT;
3450 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
3451 sizeof(struct kvm_xen_hvm_config)))
3452 goto out;
3453 r = -EINVAL;
3454 if (kvm->arch.xen_hvm_config.flags)
3455 goto out;
3456 r = 0;
3457 break;
3458 }
3459 case KVM_SET_CLOCK: {
3460 struct kvm_clock_data user_ns;
3461 u64 now_ns;
3462 s64 delta;
3463
3464 r = -EFAULT;
3465 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
3466 goto out;
3467
3468 r = -EINVAL;
3469 if (user_ns.flags)
3470 goto out;
3471
3472 r = 0;
3473 local_irq_disable();
3474 now_ns = get_kernel_ns();
3475 delta = user_ns.clock - now_ns;
3476 local_irq_enable();
3477 kvm->arch.kvmclock_offset = delta;
3478 break;
3479 }
3480 case KVM_GET_CLOCK: {
3481 struct kvm_clock_data user_ns;
3482 u64 now_ns;
3483
3484 local_irq_disable();
3485 now_ns = get_kernel_ns();
3486 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
3487 local_irq_enable();
3488 user_ns.flags = 0;
3489
3490 r = -EFAULT;
3491 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
3492 goto out;
3493 r = 0;
3494 break;
3495 }
3496
3497 default:
3498 ;
3499 }
3500 out:
3501 return r;
3502 }
3503
3504 static void kvm_init_msr_list(void)
3505 {
3506 u32 dummy[2];
3507 unsigned i, j;
3508
3509 /* skip the first msrs in the list. KVM-specific */
3510 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3511 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3512 continue;
3513 if (j < i)
3514 msrs_to_save[j] = msrs_to_save[i];
3515 j++;
3516 }
3517 num_msrs_to_save = j;
3518 }
3519
3520 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3521 const void *v)
3522 {
3523 if (vcpu->arch.apic &&
3524 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
3525 return 0;
3526
3527 return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3528 }
3529
3530 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3531 {
3532 if (vcpu->arch.apic &&
3533 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
3534 return 0;
3535
3536 return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3537 }
3538
3539 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3540 struct kvm_segment *var, int seg)
3541 {
3542 kvm_x86_ops->set_segment(vcpu, var, seg);
3543 }
3544
3545 void kvm_get_segment(struct kvm_vcpu *vcpu,
3546 struct kvm_segment *var, int seg)
3547 {
3548 kvm_x86_ops->get_segment(vcpu, var, seg);
3549 }
3550
3551 static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3552 {
3553 return gpa;
3554 }
3555
3556 static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3557 {
3558 gpa_t t_gpa;
3559 u32 error;
3560
3561 BUG_ON(!mmu_is_nested(vcpu));
3562
3563 /* NPT walks are always user-walks */
3564 access |= PFERR_USER_MASK;
3565 t_gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gpa, access, &error);
3566 if (t_gpa == UNMAPPED_GVA)
3567 vcpu->arch.fault.nested = true;
3568
3569 return t_gpa;
3570 }
3571
3572 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3573 {
3574 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3575 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, error);
3576 }
3577
3578 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3579 {
3580 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3581 access |= PFERR_FETCH_MASK;
3582 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, error);
3583 }
3584
3585 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3586 {
3587 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3588 access |= PFERR_WRITE_MASK;
3589 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, error);
3590 }
3591
3592 /* uses this to access any guest's mapped memory without checking CPL */
3593 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3594 {
3595 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, error);
3596 }
3597
3598 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3599 struct kvm_vcpu *vcpu, u32 access,
3600 u32 *error)
3601 {
3602 void *data = val;
3603 int r = X86EMUL_CONTINUE;
3604
3605 while (bytes) {
3606 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
3607 error);
3608 unsigned offset = addr & (PAGE_SIZE-1);
3609 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3610 int ret;
3611
3612 if (gpa == UNMAPPED_GVA) {
3613 r = X86EMUL_PROPAGATE_FAULT;
3614 goto out;
3615 }
3616 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3617 if (ret < 0) {
3618 r = X86EMUL_IO_NEEDED;
3619 goto out;
3620 }
3621
3622 bytes -= toread;
3623 data += toread;
3624 addr += toread;
3625 }
3626 out:
3627 return r;
3628 }
3629
3630 /* used for instruction fetching */
3631 static int kvm_fetch_guest_virt(gva_t addr, void *val, unsigned int bytes,
3632 struct kvm_vcpu *vcpu, u32 *error)
3633 {
3634 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3635 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3636 access | PFERR_FETCH_MASK, error);
3637 }
3638
3639 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3640 struct kvm_vcpu *vcpu, u32 *error)
3641 {
3642 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3643 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3644 error);
3645 }
3646
3647 static int kvm_read_guest_virt_system(gva_t addr, void *val, unsigned int bytes,
3648 struct kvm_vcpu *vcpu, u32 *error)
3649 {
3650 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, error);
3651 }
3652
3653 static int kvm_write_guest_virt_system(gva_t addr, void *val,
3654 unsigned int bytes,
3655 struct kvm_vcpu *vcpu,
3656 u32 *error)
3657 {
3658 void *data = val;
3659 int r = X86EMUL_CONTINUE;
3660
3661 while (bytes) {
3662 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
3663 PFERR_WRITE_MASK,
3664 error);
3665 unsigned offset = addr & (PAGE_SIZE-1);
3666 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3667 int ret;
3668
3669 if (gpa == UNMAPPED_GVA) {
3670 r = X86EMUL_PROPAGATE_FAULT;
3671 goto out;
3672 }
3673 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3674 if (ret < 0) {
3675 r = X86EMUL_IO_NEEDED;
3676 goto out;
3677 }
3678
3679 bytes -= towrite;
3680 data += towrite;
3681 addr += towrite;
3682 }
3683 out:
3684 return r;
3685 }
3686
3687 static int emulator_read_emulated(unsigned long addr,
3688 void *val,
3689 unsigned int bytes,
3690 unsigned int *error_code,
3691 struct kvm_vcpu *vcpu)
3692 {
3693 gpa_t gpa;
3694
3695 if (vcpu->mmio_read_completed) {
3696 memcpy(val, vcpu->mmio_data, bytes);
3697 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3698 vcpu->mmio_phys_addr, *(u64 *)val);
3699 vcpu->mmio_read_completed = 0;
3700 return X86EMUL_CONTINUE;
3701 }
3702
3703 gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, error_code);
3704
3705 if (gpa == UNMAPPED_GVA)
3706 return X86EMUL_PROPAGATE_FAULT;
3707
3708 /* For APIC access vmexit */
3709 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3710 goto mmio;
3711
3712 if (kvm_read_guest_virt(addr, val, bytes, vcpu, NULL)
3713 == X86EMUL_CONTINUE)
3714 return X86EMUL_CONTINUE;
3715
3716 mmio:
3717 /*
3718 * Is this MMIO handled locally?
3719 */
3720 if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
3721 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
3722 return X86EMUL_CONTINUE;
3723 }
3724
3725 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3726
3727 vcpu->mmio_needed = 1;
3728 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3729 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3730 vcpu->run->mmio.len = vcpu->mmio_size = bytes;
3731 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 0;
3732
3733 return X86EMUL_IO_NEEDED;
3734 }
3735
3736 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3737 const void *val, int bytes)
3738 {
3739 int ret;
3740
3741 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3742 if (ret < 0)
3743 return 0;
3744 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3745 return 1;
3746 }
3747
3748 static int emulator_write_emulated_onepage(unsigned long addr,
3749 const void *val,
3750 unsigned int bytes,
3751 unsigned int *error_code,
3752 struct kvm_vcpu *vcpu)
3753 {
3754 gpa_t gpa;
3755
3756 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, error_code);
3757
3758 if (gpa == UNMAPPED_GVA)
3759 return X86EMUL_PROPAGATE_FAULT;
3760
3761 /* For APIC access vmexit */
3762 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3763 goto mmio;
3764
3765 if (emulator_write_phys(vcpu, gpa, val, bytes))
3766 return X86EMUL_CONTINUE;
3767
3768 mmio:
3769 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3770 /*
3771 * Is this MMIO handled locally?
3772 */
3773 if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
3774 return X86EMUL_CONTINUE;
3775
3776 vcpu->mmio_needed = 1;
3777 vcpu->run->exit_reason = KVM_EXIT_MMIO;
3778 vcpu->run->mmio.phys_addr = vcpu->mmio_phys_addr = gpa;
3779 vcpu->run->mmio.len = vcpu->mmio_size = bytes;
3780 vcpu->run->mmio.is_write = vcpu->mmio_is_write = 1;
3781 memcpy(vcpu->run->mmio.data, val, bytes);
3782
3783 return X86EMUL_CONTINUE;
3784 }
3785
3786 int emulator_write_emulated(unsigned long addr,
3787 const void *val,
3788 unsigned int bytes,
3789 unsigned int *error_code,
3790 struct kvm_vcpu *vcpu)
3791 {
3792 /* Crossing a page boundary? */
3793 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3794 int rc, now;
3795
3796 now = -addr & ~PAGE_MASK;
3797 rc = emulator_write_emulated_onepage(addr, val, now, error_code,
3798 vcpu);
3799 if (rc != X86EMUL_CONTINUE)
3800 return rc;
3801 addr += now;
3802 val += now;
3803 bytes -= now;
3804 }
3805 return emulator_write_emulated_onepage(addr, val, bytes, error_code,
3806 vcpu);
3807 }
3808
3809 #define CMPXCHG_TYPE(t, ptr, old, new) \
3810 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
3811
3812 #ifdef CONFIG_X86_64
3813 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
3814 #else
3815 # define CMPXCHG64(ptr, old, new) \
3816 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
3817 #endif
3818
3819 static int emulator_cmpxchg_emulated(unsigned long addr,
3820 const void *old,
3821 const void *new,
3822 unsigned int bytes,
3823 unsigned int *error_code,
3824 struct kvm_vcpu *vcpu)
3825 {
3826 gpa_t gpa;
3827 struct page *page;
3828 char *kaddr;
3829 bool exchanged;
3830
3831 /* guests cmpxchg8b have to be emulated atomically */
3832 if (bytes > 8 || (bytes & (bytes - 1)))
3833 goto emul_write;
3834
3835 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3836
3837 if (gpa == UNMAPPED_GVA ||
3838 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3839 goto emul_write;
3840
3841 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3842 goto emul_write;
3843
3844 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3845 if (is_error_page(page)) {
3846 kvm_release_page_clean(page);
3847 goto emul_write;
3848 }
3849
3850 kaddr = kmap_atomic(page, KM_USER0);
3851 kaddr += offset_in_page(gpa);
3852 switch (bytes) {
3853 case 1:
3854 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
3855 break;
3856 case 2:
3857 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
3858 break;
3859 case 4:
3860 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
3861 break;
3862 case 8:
3863 exchanged = CMPXCHG64(kaddr, old, new);
3864 break;
3865 default:
3866 BUG();
3867 }
3868 kunmap_atomic(kaddr, KM_USER0);
3869 kvm_release_page_dirty(page);
3870
3871 if (!exchanged)
3872 return X86EMUL_CMPXCHG_FAILED;
3873
3874 kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1);
3875
3876 return X86EMUL_CONTINUE;
3877
3878 emul_write:
3879 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3880
3881 return emulator_write_emulated(addr, new, bytes, error_code, vcpu);
3882 }
3883
3884 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3885 {
3886 /* TODO: String I/O for in kernel device */
3887 int r;
3888
3889 if (vcpu->arch.pio.in)
3890 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3891 vcpu->arch.pio.size, pd);
3892 else
3893 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3894 vcpu->arch.pio.port, vcpu->arch.pio.size,
3895 pd);
3896 return r;
3897 }
3898
3899
3900 static int emulator_pio_in_emulated(int size, unsigned short port, void *val,
3901 unsigned int count, struct kvm_vcpu *vcpu)
3902 {
3903 if (vcpu->arch.pio.count)
3904 goto data_avail;
3905
3906 trace_kvm_pio(0, port, size, 1);
3907
3908 vcpu->arch.pio.port = port;
3909 vcpu->arch.pio.in = 1;
3910 vcpu->arch.pio.count = count;
3911 vcpu->arch.pio.size = size;
3912
3913 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3914 data_avail:
3915 memcpy(val, vcpu->arch.pio_data, size * count);
3916 vcpu->arch.pio.count = 0;
3917 return 1;
3918 }
3919
3920 vcpu->run->exit_reason = KVM_EXIT_IO;
3921 vcpu->run->io.direction = KVM_EXIT_IO_IN;
3922 vcpu->run->io.size = size;
3923 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3924 vcpu->run->io.count = count;
3925 vcpu->run->io.port = port;
3926
3927 return 0;
3928 }
3929
3930 static int emulator_pio_out_emulated(int size, unsigned short port,
3931 const void *val, unsigned int count,
3932 struct kvm_vcpu *vcpu)
3933 {
3934 trace_kvm_pio(1, port, size, 1);
3935
3936 vcpu->arch.pio.port = port;
3937 vcpu->arch.pio.in = 0;
3938 vcpu->arch.pio.count = count;
3939 vcpu->arch.pio.size = size;
3940
3941 memcpy(vcpu->arch.pio_data, val, size * count);
3942
3943 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3944 vcpu->arch.pio.count = 0;
3945 return 1;
3946 }
3947
3948 vcpu->run->exit_reason = KVM_EXIT_IO;
3949 vcpu->run->io.direction = KVM_EXIT_IO_OUT;
3950 vcpu->run->io.size = size;
3951 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3952 vcpu->run->io.count = count;
3953 vcpu->run->io.port = port;
3954
3955 return 0;
3956 }
3957
3958 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3959 {
3960 return kvm_x86_ops->get_segment_base(vcpu, seg);
3961 }
3962
3963 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
3964 {
3965 kvm_mmu_invlpg(vcpu, address);
3966 return X86EMUL_CONTINUE;
3967 }
3968
3969 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
3970 {
3971 if (!need_emulate_wbinvd(vcpu))
3972 return X86EMUL_CONTINUE;
3973
3974 if (kvm_x86_ops->has_wbinvd_exit()) {
3975 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
3976 wbinvd_ipi, NULL, 1);
3977 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
3978 }
3979 wbinvd();
3980 return X86EMUL_CONTINUE;
3981 }
3982 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
3983
3984 int emulate_clts(struct kvm_vcpu *vcpu)
3985 {
3986 kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3987 kvm_x86_ops->fpu_activate(vcpu);
3988 return X86EMUL_CONTINUE;
3989 }
3990
3991 int emulator_get_dr(int dr, unsigned long *dest, struct kvm_vcpu *vcpu)
3992 {
3993 return _kvm_get_dr(vcpu, dr, dest);
3994 }
3995
3996 int emulator_set_dr(int dr, unsigned long value, struct kvm_vcpu *vcpu)
3997 {
3998
3999 return __kvm_set_dr(vcpu, dr, value);
4000 }
4001
4002 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
4003 {
4004 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
4005 }
4006
4007 static unsigned long emulator_get_cr(int cr, struct kvm_vcpu *vcpu)
4008 {
4009 unsigned long value;
4010
4011 switch (cr) {
4012 case 0:
4013 value = kvm_read_cr0(vcpu);
4014 break;
4015 case 2:
4016 value = vcpu->arch.cr2;
4017 break;
4018 case 3:
4019 value = vcpu->arch.cr3;
4020 break;
4021 case 4:
4022 value = kvm_read_cr4(vcpu);
4023 break;
4024 case 8:
4025 value = kvm_get_cr8(vcpu);
4026 break;
4027 default:
4028 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4029 return 0;
4030 }
4031
4032 return value;
4033 }
4034
4035 static int emulator_set_cr(int cr, unsigned long val, struct kvm_vcpu *vcpu)
4036 {
4037 int res = 0;
4038
4039 switch (cr) {
4040 case 0:
4041 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4042 break;
4043 case 2:
4044 vcpu->arch.cr2 = val;
4045 break;
4046 case 3:
4047 res = kvm_set_cr3(vcpu, val);
4048 break;
4049 case 4:
4050 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4051 break;
4052 case 8:
4053 res = __kvm_set_cr8(vcpu, val & 0xfUL);
4054 break;
4055 default:
4056 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4057 res = -1;
4058 }
4059
4060 return res;
4061 }
4062
4063 static int emulator_get_cpl(struct kvm_vcpu *vcpu)
4064 {
4065 return kvm_x86_ops->get_cpl(vcpu);
4066 }
4067
4068 static void emulator_get_gdt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
4069 {
4070 kvm_x86_ops->get_gdt(vcpu, dt);
4071 }
4072
4073 static void emulator_get_idt(struct desc_ptr *dt, struct kvm_vcpu *vcpu)
4074 {
4075 kvm_x86_ops->get_idt(vcpu, dt);
4076 }
4077
4078 static unsigned long emulator_get_cached_segment_base(int seg,
4079 struct kvm_vcpu *vcpu)
4080 {
4081 return get_segment_base(vcpu, seg);
4082 }
4083
4084 static bool emulator_get_cached_descriptor(struct desc_struct *desc, int seg,
4085 struct kvm_vcpu *vcpu)
4086 {
4087 struct kvm_segment var;
4088
4089 kvm_get_segment(vcpu, &var, seg);
4090
4091 if (var.unusable)
4092 return false;
4093
4094 if (var.g)
4095 var.limit >>= 12;
4096 set_desc_limit(desc, var.limit);
4097 set_desc_base(desc, (unsigned long)var.base);
4098 desc->type = var.type;
4099 desc->s = var.s;
4100 desc->dpl = var.dpl;
4101 desc->p = var.present;
4102 desc->avl = var.avl;
4103 desc->l = var.l;
4104 desc->d = var.db;
4105 desc->g = var.g;
4106
4107 return true;
4108 }
4109
4110 static void emulator_set_cached_descriptor(struct desc_struct *desc, int seg,
4111 struct kvm_vcpu *vcpu)
4112 {
4113 struct kvm_segment var;
4114
4115 /* needed to preserve selector */
4116 kvm_get_segment(vcpu, &var, seg);
4117
4118 var.base = get_desc_base(desc);
4119 var.limit = get_desc_limit(desc);
4120 if (desc->g)
4121 var.limit = (var.limit << 12) | 0xfff;
4122 var.type = desc->type;
4123 var.present = desc->p;
4124 var.dpl = desc->dpl;
4125 var.db = desc->d;
4126 var.s = desc->s;
4127 var.l = desc->l;
4128 var.g = desc->g;
4129 var.avl = desc->avl;
4130 var.present = desc->p;
4131 var.unusable = !var.present;
4132 var.padding = 0;
4133
4134 kvm_set_segment(vcpu, &var, seg);
4135 return;
4136 }
4137
4138 static u16 emulator_get_segment_selector(int seg, struct kvm_vcpu *vcpu)
4139 {
4140 struct kvm_segment kvm_seg;
4141
4142 kvm_get_segment(vcpu, &kvm_seg, seg);
4143 return kvm_seg.selector;
4144 }
4145
4146 static void emulator_set_segment_selector(u16 sel, int seg,
4147 struct kvm_vcpu *vcpu)
4148 {
4149 struct kvm_segment kvm_seg;
4150
4151 kvm_get_segment(vcpu, &kvm_seg, seg);
4152 kvm_seg.selector = sel;
4153 kvm_set_segment(vcpu, &kvm_seg, seg);
4154 }
4155
4156 static struct x86_emulate_ops emulate_ops = {
4157 .read_std = kvm_read_guest_virt_system,
4158 .write_std = kvm_write_guest_virt_system,
4159 .fetch = kvm_fetch_guest_virt,
4160 .read_emulated = emulator_read_emulated,
4161 .write_emulated = emulator_write_emulated,
4162 .cmpxchg_emulated = emulator_cmpxchg_emulated,
4163 .pio_in_emulated = emulator_pio_in_emulated,
4164 .pio_out_emulated = emulator_pio_out_emulated,
4165 .get_cached_descriptor = emulator_get_cached_descriptor,
4166 .set_cached_descriptor = emulator_set_cached_descriptor,
4167 .get_segment_selector = emulator_get_segment_selector,
4168 .set_segment_selector = emulator_set_segment_selector,
4169 .get_cached_segment_base = emulator_get_cached_segment_base,
4170 .get_gdt = emulator_get_gdt,
4171 .get_idt = emulator_get_idt,
4172 .get_cr = emulator_get_cr,
4173 .set_cr = emulator_set_cr,
4174 .cpl = emulator_get_cpl,
4175 .get_dr = emulator_get_dr,
4176 .set_dr = emulator_set_dr,
4177 .set_msr = kvm_set_msr,
4178 .get_msr = kvm_get_msr,
4179 };
4180
4181 static void cache_all_regs(struct kvm_vcpu *vcpu)
4182 {
4183 kvm_register_read(vcpu, VCPU_REGS_RAX);
4184 kvm_register_read(vcpu, VCPU_REGS_RSP);
4185 kvm_register_read(vcpu, VCPU_REGS_RIP);
4186 vcpu->arch.regs_dirty = ~0;
4187 }
4188
4189 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
4190 {
4191 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu, mask);
4192 /*
4193 * an sti; sti; sequence only disable interrupts for the first
4194 * instruction. So, if the last instruction, be it emulated or
4195 * not, left the system with the INT_STI flag enabled, it
4196 * means that the last instruction is an sti. We should not
4197 * leave the flag on in this case. The same goes for mov ss
4198 */
4199 if (!(int_shadow & mask))
4200 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
4201 }
4202
4203 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
4204 {
4205 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
4206 if (ctxt->exception == PF_VECTOR)
4207 kvm_propagate_fault(vcpu);
4208 else if (ctxt->error_code_valid)
4209 kvm_queue_exception_e(vcpu, ctxt->exception, ctxt->error_code);
4210 else
4211 kvm_queue_exception(vcpu, ctxt->exception);
4212 }
4213
4214 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
4215 {
4216 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4217 int cs_db, cs_l;
4218
4219 cache_all_regs(vcpu);
4220
4221 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4222
4223 vcpu->arch.emulate_ctxt.vcpu = vcpu;
4224 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
4225 vcpu->arch.emulate_ctxt.eip = kvm_rip_read(vcpu);
4226 vcpu->arch.emulate_ctxt.mode =
4227 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
4228 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
4229 ? X86EMUL_MODE_VM86 : cs_l
4230 ? X86EMUL_MODE_PROT64 : cs_db
4231 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
4232 memset(c, 0, sizeof(struct decode_cache));
4233 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4234 }
4235
4236 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq)
4237 {
4238 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4239 int ret;
4240
4241 init_emulate_ctxt(vcpu);
4242
4243 vcpu->arch.emulate_ctxt.decode.op_bytes = 2;
4244 vcpu->arch.emulate_ctxt.decode.ad_bytes = 2;
4245 vcpu->arch.emulate_ctxt.decode.eip = vcpu->arch.emulate_ctxt.eip;
4246 ret = emulate_int_real(&vcpu->arch.emulate_ctxt, &emulate_ops, irq);
4247
4248 if (ret != X86EMUL_CONTINUE)
4249 return EMULATE_FAIL;
4250
4251 vcpu->arch.emulate_ctxt.eip = c->eip;
4252 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4253 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4254 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4255
4256 if (irq == NMI_VECTOR)
4257 vcpu->arch.nmi_pending = false;
4258 else
4259 vcpu->arch.interrupt.pending = false;
4260
4261 return EMULATE_DONE;
4262 }
4263 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
4264
4265 static int handle_emulation_failure(struct kvm_vcpu *vcpu)
4266 {
4267 ++vcpu->stat.insn_emulation_fail;
4268 trace_kvm_emulate_insn_failed(vcpu);
4269 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4270 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4271 vcpu->run->internal.ndata = 0;
4272 kvm_queue_exception(vcpu, UD_VECTOR);
4273 return EMULATE_FAIL;
4274 }
4275
4276 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4277 {
4278 gpa_t gpa;
4279
4280 if (tdp_enabled)
4281 return false;
4282
4283 /*
4284 * if emulation was due to access to shadowed page table
4285 * and it failed try to unshadow page and re-entetr the
4286 * guest to let CPU execute the instruction.
4287 */
4288 if (kvm_mmu_unprotect_page_virt(vcpu, gva))
4289 return true;
4290
4291 gpa = kvm_mmu_gva_to_gpa_system(vcpu, gva, NULL);
4292
4293 if (gpa == UNMAPPED_GVA)
4294 return true; /* let cpu generate fault */
4295
4296 if (!kvm_is_error_hva(gfn_to_hva(vcpu->kvm, gpa >> PAGE_SHIFT)))
4297 return true;
4298
4299 return false;
4300 }
4301
4302 int emulate_instruction(struct kvm_vcpu *vcpu,
4303 unsigned long cr2,
4304 u16 error_code,
4305 int emulation_type)
4306 {
4307 int r;
4308 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
4309
4310 kvm_clear_exception_queue(vcpu);
4311 vcpu->arch.mmio_fault_cr2 = cr2;
4312 /*
4313 * TODO: fix emulate.c to use guest_read/write_register
4314 * instead of direct ->regs accesses, can save hundred cycles
4315 * on Intel for instructions that don't read/change RSP, for
4316 * for example.
4317 */
4318 cache_all_regs(vcpu);
4319
4320 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
4321 init_emulate_ctxt(vcpu);
4322 vcpu->arch.emulate_ctxt.interruptibility = 0;
4323 vcpu->arch.emulate_ctxt.exception = -1;
4324 vcpu->arch.emulate_ctxt.perm_ok = false;
4325
4326 r = x86_decode_insn(&vcpu->arch.emulate_ctxt);
4327 if (r == X86EMUL_PROPAGATE_FAULT)
4328 goto done;
4329
4330 trace_kvm_emulate_insn_start(vcpu);
4331
4332 /* Only allow emulation of specific instructions on #UD
4333 * (namely VMMCALL, sysenter, sysexit, syscall)*/
4334 if (emulation_type & EMULTYPE_TRAP_UD) {
4335 if (!c->twobyte)
4336 return EMULATE_FAIL;
4337 switch (c->b) {
4338 case 0x01: /* VMMCALL */
4339 if (c->modrm_mod != 3 || c->modrm_rm != 1)
4340 return EMULATE_FAIL;
4341 break;
4342 case 0x34: /* sysenter */
4343 case 0x35: /* sysexit */
4344 if (c->modrm_mod != 0 || c->modrm_rm != 0)
4345 return EMULATE_FAIL;
4346 break;
4347 case 0x05: /* syscall */
4348 if (c->modrm_mod != 0 || c->modrm_rm != 0)
4349 return EMULATE_FAIL;
4350 break;
4351 default:
4352 return EMULATE_FAIL;
4353 }
4354
4355 if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
4356 return EMULATE_FAIL;
4357 }
4358
4359 ++vcpu->stat.insn_emulation;
4360 if (r) {
4361 if (reexecute_instruction(vcpu, cr2))
4362 return EMULATE_DONE;
4363 if (emulation_type & EMULTYPE_SKIP)
4364 return EMULATE_FAIL;
4365 return handle_emulation_failure(vcpu);
4366 }
4367 }
4368
4369 if (emulation_type & EMULTYPE_SKIP) {
4370 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
4371 return EMULATE_DONE;
4372 }
4373
4374 /* this is needed for vmware backdor interface to work since it
4375 changes registers values during IO operation */
4376 memcpy(c->regs, vcpu->arch.regs, sizeof c->regs);
4377
4378 restart:
4379 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt);
4380
4381 if (r == EMULATION_FAILED) {
4382 if (reexecute_instruction(vcpu, cr2))
4383 return EMULATE_DONE;
4384
4385 return handle_emulation_failure(vcpu);
4386 }
4387
4388 done:
4389 if (vcpu->arch.emulate_ctxt.exception >= 0) {
4390 inject_emulated_exception(vcpu);
4391 r = EMULATE_DONE;
4392 } else if (vcpu->arch.pio.count) {
4393 if (!vcpu->arch.pio.in)
4394 vcpu->arch.pio.count = 0;
4395 r = EMULATE_DO_MMIO;
4396 } else if (vcpu->mmio_needed) {
4397 if (vcpu->mmio_is_write)
4398 vcpu->mmio_needed = 0;
4399 r = EMULATE_DO_MMIO;
4400 } else if (r == EMULATION_RESTART)
4401 goto restart;
4402 else
4403 r = EMULATE_DONE;
4404
4405 toggle_interruptibility(vcpu, vcpu->arch.emulate_ctxt.interruptibility);
4406 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
4407 kvm_make_request(KVM_REQ_EVENT, vcpu);
4408 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
4409 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
4410
4411 return r;
4412 }
4413 EXPORT_SYMBOL_GPL(emulate_instruction);
4414
4415 int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size, unsigned short port)
4416 {
4417 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
4418 int ret = emulator_pio_out_emulated(size, port, &val, 1, vcpu);
4419 /* do not return to emulator after return from userspace */
4420 vcpu->arch.pio.count = 0;
4421 return ret;
4422 }
4423 EXPORT_SYMBOL_GPL(kvm_fast_pio_out);
4424
4425 static void tsc_bad(void *info)
4426 {
4427 __get_cpu_var(cpu_tsc_khz) = 0;
4428 }
4429
4430 static void tsc_khz_changed(void *data)
4431 {
4432 struct cpufreq_freqs *freq = data;
4433 unsigned long khz = 0;
4434
4435 if (data)
4436 khz = freq->new;
4437 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4438 khz = cpufreq_quick_get(raw_smp_processor_id());
4439 if (!khz)
4440 khz = tsc_khz;
4441 __get_cpu_var(cpu_tsc_khz) = khz;
4442 }
4443
4444 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
4445 void *data)
4446 {
4447 struct cpufreq_freqs *freq = data;
4448 struct kvm *kvm;
4449 struct kvm_vcpu *vcpu;
4450 int i, send_ipi = 0;
4451
4452 /*
4453 * We allow guests to temporarily run on slowing clocks,
4454 * provided we notify them after, or to run on accelerating
4455 * clocks, provided we notify them before. Thus time never
4456 * goes backwards.
4457 *
4458 * However, we have a problem. We can't atomically update
4459 * the frequency of a given CPU from this function; it is
4460 * merely a notifier, which can be called from any CPU.
4461 * Changing the TSC frequency at arbitrary points in time
4462 * requires a recomputation of local variables related to
4463 * the TSC for each VCPU. We must flag these local variables
4464 * to be updated and be sure the update takes place with the
4465 * new frequency before any guests proceed.
4466 *
4467 * Unfortunately, the combination of hotplug CPU and frequency
4468 * change creates an intractable locking scenario; the order
4469 * of when these callouts happen is undefined with respect to
4470 * CPU hotplug, and they can race with each other. As such,
4471 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
4472 * undefined; you can actually have a CPU frequency change take
4473 * place in between the computation of X and the setting of the
4474 * variable. To protect against this problem, all updates of
4475 * the per_cpu tsc_khz variable are done in an interrupt
4476 * protected IPI, and all callers wishing to update the value
4477 * must wait for a synchronous IPI to complete (which is trivial
4478 * if the caller is on the CPU already). This establishes the
4479 * necessary total order on variable updates.
4480 *
4481 * Note that because a guest time update may take place
4482 * anytime after the setting of the VCPU's request bit, the
4483 * correct TSC value must be set before the request. However,
4484 * to ensure the update actually makes it to any guest which
4485 * starts running in hardware virtualization between the set
4486 * and the acquisition of the spinlock, we must also ping the
4487 * CPU after setting the request bit.
4488 *
4489 */
4490
4491 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
4492 return 0;
4493 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
4494 return 0;
4495
4496 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4497
4498 spin_lock(&kvm_lock);
4499 list_for_each_entry(kvm, &vm_list, vm_list) {
4500 kvm_for_each_vcpu(i, vcpu, kvm) {
4501 if (vcpu->cpu != freq->cpu)
4502 continue;
4503 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4504 if (vcpu->cpu != smp_processor_id())
4505 send_ipi = 1;
4506 }
4507 }
4508 spin_unlock(&kvm_lock);
4509
4510 if (freq->old < freq->new && send_ipi) {
4511 /*
4512 * We upscale the frequency. Must make the guest
4513 * doesn't see old kvmclock values while running with
4514 * the new frequency, otherwise we risk the guest sees
4515 * time go backwards.
4516 *
4517 * In case we update the frequency for another cpu
4518 * (which might be in guest context) send an interrupt
4519 * to kick the cpu out of guest context. Next time
4520 * guest context is entered kvmclock will be updated,
4521 * so the guest will not see stale values.
4522 */
4523 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
4524 }
4525 return 0;
4526 }
4527
4528 static struct notifier_block kvmclock_cpufreq_notifier_block = {
4529 .notifier_call = kvmclock_cpufreq_notifier
4530 };
4531
4532 static int kvmclock_cpu_notifier(struct notifier_block *nfb,
4533 unsigned long action, void *hcpu)
4534 {
4535 unsigned int cpu = (unsigned long)hcpu;
4536
4537 switch (action) {
4538 case CPU_ONLINE:
4539 case CPU_DOWN_FAILED:
4540 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4541 break;
4542 case CPU_DOWN_PREPARE:
4543 smp_call_function_single(cpu, tsc_bad, NULL, 1);
4544 break;
4545 }
4546 return NOTIFY_OK;
4547 }
4548
4549 static struct notifier_block kvmclock_cpu_notifier_block = {
4550 .notifier_call = kvmclock_cpu_notifier,
4551 .priority = -INT_MAX
4552 };
4553
4554 static void kvm_timer_init(void)
4555 {
4556 int cpu;
4557
4558 max_tsc_khz = tsc_khz;
4559 register_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4560 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
4561 #ifdef CONFIG_CPU_FREQ
4562 struct cpufreq_policy policy;
4563 memset(&policy, 0, sizeof(policy));
4564 cpufreq_get_policy(&policy, get_cpu());
4565 if (policy.cpuinfo.max_freq)
4566 max_tsc_khz = policy.cpuinfo.max_freq;
4567 #endif
4568 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
4569 CPUFREQ_TRANSITION_NOTIFIER);
4570 }
4571 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
4572 for_each_online_cpu(cpu)
4573 smp_call_function_single(cpu, tsc_khz_changed, NULL, 1);
4574 }
4575
4576 static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
4577
4578 static int kvm_is_in_guest(void)
4579 {
4580 return percpu_read(current_vcpu) != NULL;
4581 }
4582
4583 static int kvm_is_user_mode(void)
4584 {
4585 int user_mode = 3;
4586
4587 if (percpu_read(current_vcpu))
4588 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu));
4589
4590 return user_mode != 0;
4591 }
4592
4593 static unsigned long kvm_get_guest_ip(void)
4594 {
4595 unsigned long ip = 0;
4596
4597 if (percpu_read(current_vcpu))
4598 ip = kvm_rip_read(percpu_read(current_vcpu));
4599
4600 return ip;
4601 }
4602
4603 static struct perf_guest_info_callbacks kvm_guest_cbs = {
4604 .is_in_guest = kvm_is_in_guest,
4605 .is_user_mode = kvm_is_user_mode,
4606 .get_guest_ip = kvm_get_guest_ip,
4607 };
4608
4609 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
4610 {
4611 percpu_write(current_vcpu, vcpu);
4612 }
4613 EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
4614
4615 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
4616 {
4617 percpu_write(current_vcpu, NULL);
4618 }
4619 EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
4620
4621 int kvm_arch_init(void *opaque)
4622 {
4623 int r;
4624 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
4625
4626 if (kvm_x86_ops) {
4627 printk(KERN_ERR "kvm: already loaded the other module\n");
4628 r = -EEXIST;
4629 goto out;
4630 }
4631
4632 if (!ops->cpu_has_kvm_support()) {
4633 printk(KERN_ERR "kvm: no hardware support\n");
4634 r = -EOPNOTSUPP;
4635 goto out;
4636 }
4637 if (ops->disabled_by_bios()) {
4638 printk(KERN_ERR "kvm: disabled by bios\n");
4639 r = -EOPNOTSUPP;
4640 goto out;
4641 }
4642
4643 r = kvm_mmu_module_init();
4644 if (r)
4645 goto out;
4646
4647 kvm_init_msr_list();
4648
4649 kvm_x86_ops = ops;
4650 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
4651 kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
4652 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
4653 PT_DIRTY_MASK, PT64_NX_MASK, 0);
4654
4655 kvm_timer_init();
4656
4657 perf_register_guest_info_callbacks(&kvm_guest_cbs);
4658
4659 if (cpu_has_xsave)
4660 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
4661
4662 return 0;
4663
4664 out:
4665 return r;
4666 }
4667
4668 void kvm_arch_exit(void)
4669 {
4670 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
4671
4672 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
4673 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
4674 CPUFREQ_TRANSITION_NOTIFIER);
4675 unregister_hotcpu_notifier(&kvmclock_cpu_notifier_block);
4676 kvm_x86_ops = NULL;
4677 kvm_mmu_module_exit();
4678 }
4679
4680 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
4681 {
4682 ++vcpu->stat.halt_exits;
4683 if (irqchip_in_kernel(vcpu->kvm)) {
4684 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
4685 return 1;
4686 } else {
4687 vcpu->run->exit_reason = KVM_EXIT_HLT;
4688 return 0;
4689 }
4690 }
4691 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
4692
4693 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
4694 unsigned long a1)
4695 {
4696 if (is_long_mode(vcpu))
4697 return a0;
4698 else
4699 return a0 | ((gpa_t)a1 << 32);
4700 }
4701
4702 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
4703 {
4704 u64 param, ingpa, outgpa, ret;
4705 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
4706 bool fast, longmode;
4707 int cs_db, cs_l;
4708
4709 /*
4710 * hypercall generates UD from non zero cpl and real mode
4711 * per HYPER-V spec
4712 */
4713 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
4714 kvm_queue_exception(vcpu, UD_VECTOR);
4715 return 0;
4716 }
4717
4718 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
4719 longmode = is_long_mode(vcpu) && cs_l == 1;
4720
4721 if (!longmode) {
4722 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
4723 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
4724 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
4725 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
4726 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
4727 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
4728 }
4729 #ifdef CONFIG_X86_64
4730 else {
4731 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
4732 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
4733 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
4734 }
4735 #endif
4736
4737 code = param & 0xffff;
4738 fast = (param >> 16) & 0x1;
4739 rep_cnt = (param >> 32) & 0xfff;
4740 rep_idx = (param >> 48) & 0xfff;
4741
4742 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
4743
4744 switch (code) {
4745 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
4746 kvm_vcpu_on_spin(vcpu);
4747 break;
4748 default:
4749 res = HV_STATUS_INVALID_HYPERCALL_CODE;
4750 break;
4751 }
4752
4753 ret = res | (((u64)rep_done & 0xfff) << 32);
4754 if (longmode) {
4755 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4756 } else {
4757 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
4758 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
4759 }
4760
4761 return 1;
4762 }
4763
4764 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
4765 {
4766 unsigned long nr, a0, a1, a2, a3, ret;
4767 int r = 1;
4768
4769 if (kvm_hv_hypercall_enabled(vcpu->kvm))
4770 return kvm_hv_hypercall(vcpu);
4771
4772 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
4773 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
4774 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
4775 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
4776 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
4777
4778 trace_kvm_hypercall(nr, a0, a1, a2, a3);
4779
4780 if (!is_long_mode(vcpu)) {
4781 nr &= 0xFFFFFFFF;
4782 a0 &= 0xFFFFFFFF;
4783 a1 &= 0xFFFFFFFF;
4784 a2 &= 0xFFFFFFFF;
4785 a3 &= 0xFFFFFFFF;
4786 }
4787
4788 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
4789 ret = -KVM_EPERM;
4790 goto out;
4791 }
4792
4793 switch (nr) {
4794 case KVM_HC_VAPIC_POLL_IRQ:
4795 ret = 0;
4796 break;
4797 case KVM_HC_MMU_OP:
4798 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
4799 break;
4800 default:
4801 ret = -KVM_ENOSYS;
4802 break;
4803 }
4804 out:
4805 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
4806 ++vcpu->stat.hypercalls;
4807 return r;
4808 }
4809 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
4810
4811 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
4812 {
4813 char instruction[3];
4814 unsigned long rip = kvm_rip_read(vcpu);
4815
4816 /*
4817 * Blow out the MMU to ensure that no other VCPU has an active mapping
4818 * to ensure that the updated hypercall appears atomically across all
4819 * VCPUs.
4820 */
4821 kvm_mmu_zap_all(vcpu->kvm);
4822
4823 kvm_x86_ops->patch_hypercall(vcpu, instruction);
4824
4825 return emulator_write_emulated(rip, instruction, 3, NULL, vcpu);
4826 }
4827
4828 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4829 {
4830 struct desc_ptr dt = { limit, base };
4831
4832 kvm_x86_ops->set_gdt(vcpu, &dt);
4833 }
4834
4835 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
4836 {
4837 struct desc_ptr dt = { limit, base };
4838
4839 kvm_x86_ops->set_idt(vcpu, &dt);
4840 }
4841
4842 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
4843 {
4844 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
4845 int j, nent = vcpu->arch.cpuid_nent;
4846
4847 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
4848 /* when no next entry is found, the current entry[i] is reselected */
4849 for (j = i + 1; ; j = (j + 1) % nent) {
4850 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
4851 if (ej->function == e->function) {
4852 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
4853 return j;
4854 }
4855 }
4856 return 0; /* silence gcc, even though control never reaches here */
4857 }
4858
4859 /* find an entry with matching function, matching index (if needed), and that
4860 * should be read next (if it's stateful) */
4861 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
4862 u32 function, u32 index)
4863 {
4864 if (e->function != function)
4865 return 0;
4866 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
4867 return 0;
4868 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
4869 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
4870 return 0;
4871 return 1;
4872 }
4873
4874 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
4875 u32 function, u32 index)
4876 {
4877 int i;
4878 struct kvm_cpuid_entry2 *best = NULL;
4879
4880 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
4881 struct kvm_cpuid_entry2 *e;
4882
4883 e = &vcpu->arch.cpuid_entries[i];
4884 if (is_matching_cpuid_entry(e, function, index)) {
4885 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
4886 move_to_next_stateful_cpuid_entry(vcpu, i);
4887 best = e;
4888 break;
4889 }
4890 /*
4891 * Both basic or both extended?
4892 */
4893 if (((e->function ^ function) & 0x80000000) == 0)
4894 if (!best || e->function > best->function)
4895 best = e;
4896 }
4897 return best;
4898 }
4899 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
4900
4901 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
4902 {
4903 struct kvm_cpuid_entry2 *best;
4904
4905 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
4906 if (!best || best->eax < 0x80000008)
4907 goto not_found;
4908 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
4909 if (best)
4910 return best->eax & 0xff;
4911 not_found:
4912 return 36;
4913 }
4914
4915 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
4916 {
4917 u32 function, index;
4918 struct kvm_cpuid_entry2 *best;
4919
4920 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
4921 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4922 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
4923 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
4924 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
4925 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
4926 best = kvm_find_cpuid_entry(vcpu, function, index);
4927 if (best) {
4928 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
4929 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
4930 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
4931 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
4932 }
4933 kvm_x86_ops->skip_emulated_instruction(vcpu);
4934 trace_kvm_cpuid(function,
4935 kvm_register_read(vcpu, VCPU_REGS_RAX),
4936 kvm_register_read(vcpu, VCPU_REGS_RBX),
4937 kvm_register_read(vcpu, VCPU_REGS_RCX),
4938 kvm_register_read(vcpu, VCPU_REGS_RDX));
4939 }
4940 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
4941
4942 /*
4943 * Check if userspace requested an interrupt window, and that the
4944 * interrupt window is open.
4945 *
4946 * No need to exit to userspace if we already have an interrupt queued.
4947 */
4948 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
4949 {
4950 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
4951 vcpu->run->request_interrupt_window &&
4952 kvm_arch_interrupt_allowed(vcpu));
4953 }
4954
4955 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
4956 {
4957 struct kvm_run *kvm_run = vcpu->run;
4958
4959 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
4960 kvm_run->cr8 = kvm_get_cr8(vcpu);
4961 kvm_run->apic_base = kvm_get_apic_base(vcpu);
4962 if (irqchip_in_kernel(vcpu->kvm))
4963 kvm_run->ready_for_interrupt_injection = 1;
4964 else
4965 kvm_run->ready_for_interrupt_injection =
4966 kvm_arch_interrupt_allowed(vcpu) &&
4967 !kvm_cpu_has_interrupt(vcpu) &&
4968 !kvm_event_needs_reinjection(vcpu);
4969 }
4970
4971 static void vapic_enter(struct kvm_vcpu *vcpu)
4972 {
4973 struct kvm_lapic *apic = vcpu->arch.apic;
4974 struct page *page;
4975
4976 if (!apic || !apic->vapic_addr)
4977 return;
4978
4979 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4980
4981 vcpu->arch.apic->vapic_page = page;
4982 }
4983
4984 static void vapic_exit(struct kvm_vcpu *vcpu)
4985 {
4986 struct kvm_lapic *apic = vcpu->arch.apic;
4987 int idx;
4988
4989 if (!apic || !apic->vapic_addr)
4990 return;
4991
4992 idx = srcu_read_lock(&vcpu->kvm->srcu);
4993 kvm_release_page_dirty(apic->vapic_page);
4994 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4995 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4996 }
4997
4998 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
4999 {
5000 int max_irr, tpr;
5001
5002 if (!kvm_x86_ops->update_cr8_intercept)
5003 return;
5004
5005 if (!vcpu->arch.apic)
5006 return;
5007
5008 if (!vcpu->arch.apic->vapic_addr)
5009 max_irr = kvm_lapic_find_highest_irr(vcpu);
5010 else
5011 max_irr = -1;
5012
5013 if (max_irr != -1)
5014 max_irr >>= 4;
5015
5016 tpr = kvm_lapic_get_cr8(vcpu);
5017
5018 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
5019 }
5020
5021 static void inject_pending_event(struct kvm_vcpu *vcpu)
5022 {
5023 /* try to reinject previous events if any */
5024 if (vcpu->arch.exception.pending) {
5025 trace_kvm_inj_exception(vcpu->arch.exception.nr,
5026 vcpu->arch.exception.has_error_code,
5027 vcpu->arch.exception.error_code);
5028 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
5029 vcpu->arch.exception.has_error_code,
5030 vcpu->arch.exception.error_code,
5031 vcpu->arch.exception.reinject);
5032 return;
5033 }
5034
5035 if (vcpu->arch.nmi_injected) {
5036 kvm_x86_ops->set_nmi(vcpu);
5037 return;
5038 }
5039
5040 if (vcpu->arch.interrupt.pending) {
5041 kvm_x86_ops->set_irq(vcpu);
5042 return;
5043 }
5044
5045 /* try to inject new event if pending */
5046 if (vcpu->arch.nmi_pending) {
5047 if (kvm_x86_ops->nmi_allowed(vcpu)) {
5048 vcpu->arch.nmi_pending = false;
5049 vcpu->arch.nmi_injected = true;
5050 kvm_x86_ops->set_nmi(vcpu);
5051 }
5052 } else if (kvm_cpu_has_interrupt(vcpu)) {
5053 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
5054 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
5055 false);
5056 kvm_x86_ops->set_irq(vcpu);
5057 }
5058 }
5059 }
5060
5061 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
5062 {
5063 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
5064 !vcpu->guest_xcr0_loaded) {
5065 /* kvm_set_xcr() also depends on this */
5066 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
5067 vcpu->guest_xcr0_loaded = 1;
5068 }
5069 }
5070
5071 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
5072 {
5073 if (vcpu->guest_xcr0_loaded) {
5074 if (vcpu->arch.xcr0 != host_xcr0)
5075 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
5076 vcpu->guest_xcr0_loaded = 0;
5077 }
5078 }
5079
5080 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5081 {
5082 int r;
5083 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5084 vcpu->run->request_interrupt_window;
5085
5086 if (vcpu->requests) {
5087 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
5088 kvm_mmu_unload(vcpu);
5089 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
5090 __kvm_migrate_timers(vcpu);
5091 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
5092 r = kvm_guest_time_update(vcpu);
5093 if (unlikely(r))
5094 goto out;
5095 }
5096 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
5097 kvm_mmu_sync_roots(vcpu);
5098 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
5099 kvm_x86_ops->tlb_flush(vcpu);
5100 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
5101 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
5102 r = 0;
5103 goto out;
5104 }
5105 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
5106 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5107 r = 0;
5108 goto out;
5109 }
5110 if (kvm_check_request(KVM_REQ_DEACTIVATE_FPU, vcpu)) {
5111 vcpu->fpu_active = 0;
5112 kvm_x86_ops->fpu_deactivate(vcpu);
5113 }
5114 }
5115
5116 r = kvm_mmu_reload(vcpu);
5117 if (unlikely(r))
5118 goto out;
5119
5120 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
5121 inject_pending_event(vcpu);
5122
5123 /* enable NMI/IRQ window open exits if needed */
5124 if (vcpu->arch.nmi_pending)
5125 kvm_x86_ops->enable_nmi_window(vcpu);
5126 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
5127 kvm_x86_ops->enable_irq_window(vcpu);
5128
5129 if (kvm_lapic_enabled(vcpu)) {
5130 update_cr8_intercept(vcpu);
5131 kvm_lapic_sync_to_vapic(vcpu);
5132 }
5133 }
5134
5135 preempt_disable();
5136
5137 kvm_x86_ops->prepare_guest_switch(vcpu);
5138 if (vcpu->fpu_active)
5139 kvm_load_guest_fpu(vcpu);
5140 kvm_load_guest_xcr0(vcpu);
5141
5142 atomic_set(&vcpu->guest_mode, 1);
5143 smp_wmb();
5144
5145 local_irq_disable();
5146
5147 if (!atomic_read(&vcpu->guest_mode) || vcpu->requests
5148 || need_resched() || signal_pending(current)) {
5149 atomic_set(&vcpu->guest_mode, 0);
5150 smp_wmb();
5151 local_irq_enable();
5152 preempt_enable();
5153 kvm_x86_ops->cancel_injection(vcpu);
5154 r = 1;
5155 goto out;
5156 }
5157
5158 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5159
5160 kvm_guest_enter();
5161
5162 if (unlikely(vcpu->arch.switch_db_regs)) {
5163 set_debugreg(0, 7);
5164 set_debugreg(vcpu->arch.eff_db[0], 0);
5165 set_debugreg(vcpu->arch.eff_db[1], 1);
5166 set_debugreg(vcpu->arch.eff_db[2], 2);
5167 set_debugreg(vcpu->arch.eff_db[3], 3);
5168 }
5169
5170 trace_kvm_entry(vcpu->vcpu_id);
5171 kvm_x86_ops->run(vcpu);
5172
5173 /*
5174 * If the guest has used debug registers, at least dr7
5175 * will be disabled while returning to the host.
5176 * If we don't have active breakpoints in the host, we don't
5177 * care about the messed up debug address registers. But if
5178 * we have some of them active, restore the old state.
5179 */
5180 if (hw_breakpoint_active())
5181 hw_breakpoint_restore();
5182
5183 kvm_get_msr(vcpu, MSR_IA32_TSC, &vcpu->arch.last_guest_tsc);
5184
5185 atomic_set(&vcpu->guest_mode, 0);
5186 smp_wmb();
5187 local_irq_enable();
5188
5189 ++vcpu->stat.exits;
5190
5191 /*
5192 * We must have an instruction between local_irq_enable() and
5193 * kvm_guest_exit(), so the timer interrupt isn't delayed by
5194 * the interrupt shadow. The stat.exits increment will do nicely.
5195 * But we need to prevent reordering, hence this barrier():
5196 */
5197 barrier();
5198
5199 kvm_guest_exit();
5200
5201 preempt_enable();
5202
5203 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5204
5205 /*
5206 * Profile KVM exit RIPs:
5207 */
5208 if (unlikely(prof_on == KVM_PROFILING)) {
5209 unsigned long rip = kvm_rip_read(vcpu);
5210 profile_hit(KVM_PROFILING, (void *)rip);
5211 }
5212
5213
5214 kvm_lapic_sync_from_vapic(vcpu);
5215
5216 r = kvm_x86_ops->handle_exit(vcpu);
5217 out:
5218 return r;
5219 }
5220
5221
5222 static int __vcpu_run(struct kvm_vcpu *vcpu)
5223 {
5224 int r;
5225 struct kvm *kvm = vcpu->kvm;
5226
5227 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
5228 pr_debug("vcpu %d received sipi with vector # %x\n",
5229 vcpu->vcpu_id, vcpu->arch.sipi_vector);
5230 kvm_lapic_reset(vcpu);
5231 r = kvm_arch_vcpu_reset(vcpu);
5232 if (r)
5233 return r;
5234 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5235 }
5236
5237 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5238 vapic_enter(vcpu);
5239
5240 r = 1;
5241 while (r > 0) {
5242 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
5243 r = vcpu_enter_guest(vcpu);
5244 else {
5245 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5246 kvm_vcpu_block(vcpu);
5247 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5248 if (kvm_check_request(KVM_REQ_UNHALT, vcpu))
5249 {
5250 switch(vcpu->arch.mp_state) {
5251 case KVM_MP_STATE_HALTED:
5252 vcpu->arch.mp_state =
5253 KVM_MP_STATE_RUNNABLE;
5254 case KVM_MP_STATE_RUNNABLE:
5255 break;
5256 case KVM_MP_STATE_SIPI_RECEIVED:
5257 default:
5258 r = -EINTR;
5259 break;
5260 }
5261 }
5262 }
5263
5264 if (r <= 0)
5265 break;
5266
5267 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
5268 if (kvm_cpu_has_pending_timer(vcpu))
5269 kvm_inject_pending_timer_irqs(vcpu);
5270
5271 if (dm_request_for_irq_injection(vcpu)) {
5272 r = -EINTR;
5273 vcpu->run->exit_reason = KVM_EXIT_INTR;
5274 ++vcpu->stat.request_irq_exits;
5275 }
5276 if (signal_pending(current)) {
5277 r = -EINTR;
5278 vcpu->run->exit_reason = KVM_EXIT_INTR;
5279 ++vcpu->stat.signal_exits;
5280 }
5281 if (need_resched()) {
5282 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5283 kvm_resched(vcpu);
5284 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
5285 }
5286 }
5287
5288 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
5289
5290 vapic_exit(vcpu);
5291
5292 return r;
5293 }
5294
5295 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5296 {
5297 int r;
5298 sigset_t sigsaved;
5299
5300 if (vcpu->sigset_active)
5301 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
5302
5303 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
5304 kvm_vcpu_block(vcpu);
5305 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
5306 r = -EAGAIN;
5307 goto out;
5308 }
5309
5310 /* re-sync apic's tpr */
5311 if (!irqchip_in_kernel(vcpu->kvm))
5312 kvm_set_cr8(vcpu, kvm_run->cr8);
5313
5314 if (vcpu->arch.pio.count || vcpu->mmio_needed) {
5315 if (vcpu->mmio_needed) {
5316 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
5317 vcpu->mmio_read_completed = 1;
5318 vcpu->mmio_needed = 0;
5319 }
5320 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
5321 r = emulate_instruction(vcpu, 0, 0, EMULTYPE_NO_DECODE);
5322 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5323 if (r != EMULATE_DONE) {
5324 r = 0;
5325 goto out;
5326 }
5327 }
5328 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5329 kvm_register_write(vcpu, VCPU_REGS_RAX,
5330 kvm_run->hypercall.ret);
5331
5332 r = __vcpu_run(vcpu);
5333
5334 out:
5335 post_kvm_run_save(vcpu);
5336 if (vcpu->sigset_active)
5337 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
5338
5339 return r;
5340 }
5341
5342 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5343 {
5344 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
5345 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
5346 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
5347 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
5348 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
5349 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
5350 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
5351 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
5352 #ifdef CONFIG_X86_64
5353 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
5354 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
5355 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
5356 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
5357 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
5358 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
5359 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
5360 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
5361 #endif
5362
5363 regs->rip = kvm_rip_read(vcpu);
5364 regs->rflags = kvm_get_rflags(vcpu);
5365
5366 return 0;
5367 }
5368
5369 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
5370 {
5371 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
5372 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
5373 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
5374 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
5375 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
5376 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
5377 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
5378 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
5379 #ifdef CONFIG_X86_64
5380 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
5381 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
5382 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
5383 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
5384 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
5385 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
5386 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
5387 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
5388 #endif
5389
5390 kvm_rip_write(vcpu, regs->rip);
5391 kvm_set_rflags(vcpu, regs->rflags);
5392
5393 vcpu->arch.exception.pending = false;
5394
5395 kvm_make_request(KVM_REQ_EVENT, vcpu);
5396
5397 return 0;
5398 }
5399
5400 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
5401 {
5402 struct kvm_segment cs;
5403
5404 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
5405 *db = cs.db;
5406 *l = cs.l;
5407 }
5408 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
5409
5410 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
5411 struct kvm_sregs *sregs)
5412 {
5413 struct desc_ptr dt;
5414
5415 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5416 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5417 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5418 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5419 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5420 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5421
5422 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5423 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5424
5425 kvm_x86_ops->get_idt(vcpu, &dt);
5426 sregs->idt.limit = dt.size;
5427 sregs->idt.base = dt.address;
5428 kvm_x86_ops->get_gdt(vcpu, &dt);
5429 sregs->gdt.limit = dt.size;
5430 sregs->gdt.base = dt.address;
5431
5432 sregs->cr0 = kvm_read_cr0(vcpu);
5433 sregs->cr2 = vcpu->arch.cr2;
5434 sregs->cr3 = vcpu->arch.cr3;
5435 sregs->cr4 = kvm_read_cr4(vcpu);
5436 sregs->cr8 = kvm_get_cr8(vcpu);
5437 sregs->efer = vcpu->arch.efer;
5438 sregs->apic_base = kvm_get_apic_base(vcpu);
5439
5440 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
5441
5442 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
5443 set_bit(vcpu->arch.interrupt.nr,
5444 (unsigned long *)sregs->interrupt_bitmap);
5445
5446 return 0;
5447 }
5448
5449 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
5450 struct kvm_mp_state *mp_state)
5451 {
5452 mp_state->mp_state = vcpu->arch.mp_state;
5453 return 0;
5454 }
5455
5456 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
5457 struct kvm_mp_state *mp_state)
5458 {
5459 vcpu->arch.mp_state = mp_state->mp_state;
5460 kvm_make_request(KVM_REQ_EVENT, vcpu);
5461 return 0;
5462 }
5463
5464 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason,
5465 bool has_error_code, u32 error_code)
5466 {
5467 struct decode_cache *c = &vcpu->arch.emulate_ctxt.decode;
5468 int ret;
5469
5470 init_emulate_ctxt(vcpu);
5471
5472 ret = emulator_task_switch(&vcpu->arch.emulate_ctxt,
5473 tss_selector, reason, has_error_code,
5474 error_code);
5475
5476 if (ret)
5477 return EMULATE_FAIL;
5478
5479 memcpy(vcpu->arch.regs, c->regs, sizeof c->regs);
5480 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.eip);
5481 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
5482 kvm_make_request(KVM_REQ_EVENT, vcpu);
5483 return EMULATE_DONE;
5484 }
5485 EXPORT_SYMBOL_GPL(kvm_task_switch);
5486
5487 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5488 struct kvm_sregs *sregs)
5489 {
5490 int mmu_reset_needed = 0;
5491 int pending_vec, max_bits;
5492 struct desc_ptr dt;
5493
5494 dt.size = sregs->idt.limit;
5495 dt.address = sregs->idt.base;
5496 kvm_x86_ops->set_idt(vcpu, &dt);
5497 dt.size = sregs->gdt.limit;
5498 dt.address = sregs->gdt.base;
5499 kvm_x86_ops->set_gdt(vcpu, &dt);
5500
5501 vcpu->arch.cr2 = sregs->cr2;
5502 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
5503 vcpu->arch.cr3 = sregs->cr3;
5504
5505 kvm_set_cr8(vcpu, sregs->cr8);
5506
5507 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5508 kvm_x86_ops->set_efer(vcpu, sregs->efer);
5509 kvm_set_apic_base(vcpu, sregs->apic_base);
5510
5511 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5512 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5513 vcpu->arch.cr0 = sregs->cr0;
5514
5515 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5516 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5517 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5518 load_pdptrs(vcpu, vcpu->arch.walk_mmu, vcpu->arch.cr3);
5519 mmu_reset_needed = 1;
5520 }
5521
5522 if (mmu_reset_needed)
5523 kvm_mmu_reset_context(vcpu);
5524
5525 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5526 pending_vec = find_first_bit(
5527 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5528 if (pending_vec < max_bits) {
5529 kvm_queue_interrupt(vcpu, pending_vec, false);
5530 pr_debug("Set back pending irq %d\n", pending_vec);
5531 if (irqchip_in_kernel(vcpu->kvm))
5532 kvm_pic_clear_isr_ack(vcpu->kvm);
5533 }
5534
5535 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5536 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5537 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5538 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5539 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5540 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5541
5542 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5543 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5544
5545 update_cr8_intercept(vcpu);
5546
5547 /* Older userspace won't unhalt the vcpu on reset. */
5548 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5549 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5550 !is_protmode(vcpu))
5551 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5552
5553 kvm_make_request(KVM_REQ_EVENT, vcpu);
5554
5555 return 0;
5556 }
5557
5558 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5559 struct kvm_guest_debug *dbg)
5560 {
5561 unsigned long rflags;
5562 int i, r;
5563
5564 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5565 r = -EBUSY;
5566 if (vcpu->arch.exception.pending)
5567 goto out;
5568 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5569 kvm_queue_exception(vcpu, DB_VECTOR);
5570 else
5571 kvm_queue_exception(vcpu, BP_VECTOR);
5572 }
5573
5574 /*
5575 * Read rflags as long as potentially injected trace flags are still
5576 * filtered out.
5577 */
5578 rflags = kvm_get_rflags(vcpu);
5579
5580 vcpu->guest_debug = dbg->control;
5581 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5582 vcpu->guest_debug = 0;
5583
5584 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5585 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5586 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5587 vcpu->arch.switch_db_regs =
5588 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5589 } else {
5590 for (i = 0; i < KVM_NR_DB_REGS; i++)
5591 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5592 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5593 }
5594
5595 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5596 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
5597 get_segment_base(vcpu, VCPU_SREG_CS);
5598
5599 /*
5600 * Trigger an rflags update that will inject or remove the trace
5601 * flags.
5602 */
5603 kvm_set_rflags(vcpu, rflags);
5604
5605 kvm_x86_ops->set_guest_debug(vcpu, dbg);
5606
5607 r = 0;
5608
5609 out:
5610
5611 return r;
5612 }
5613
5614 /*
5615 * Translate a guest virtual address to a guest physical address.
5616 */
5617 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5618 struct kvm_translation *tr)
5619 {
5620 unsigned long vaddr = tr->linear_address;
5621 gpa_t gpa;
5622 int idx;
5623
5624 idx = srcu_read_lock(&vcpu->kvm->srcu);
5625 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5626 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5627 tr->physical_address = gpa;
5628 tr->valid = gpa != UNMAPPED_GVA;
5629 tr->writeable = 1;
5630 tr->usermode = 0;
5631
5632 return 0;
5633 }
5634
5635 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5636 {
5637 struct i387_fxsave_struct *fxsave =
5638 &vcpu->arch.guest_fpu.state->fxsave;
5639
5640 memcpy(fpu->fpr, fxsave->st_space, 128);
5641 fpu->fcw = fxsave->cwd;
5642 fpu->fsw = fxsave->swd;
5643 fpu->ftwx = fxsave->twd;
5644 fpu->last_opcode = fxsave->fop;
5645 fpu->last_ip = fxsave->rip;
5646 fpu->last_dp = fxsave->rdp;
5647 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5648
5649 return 0;
5650 }
5651
5652 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5653 {
5654 struct i387_fxsave_struct *fxsave =
5655 &vcpu->arch.guest_fpu.state->fxsave;
5656
5657 memcpy(fxsave->st_space, fpu->fpr, 128);
5658 fxsave->cwd = fpu->fcw;
5659 fxsave->swd = fpu->fsw;
5660 fxsave->twd = fpu->ftwx;
5661 fxsave->fop = fpu->last_opcode;
5662 fxsave->rip = fpu->last_ip;
5663 fxsave->rdp = fpu->last_dp;
5664 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5665
5666 return 0;
5667 }
5668
5669 int fx_init(struct kvm_vcpu *vcpu)
5670 {
5671 int err;
5672
5673 err = fpu_alloc(&vcpu->arch.guest_fpu);
5674 if (err)
5675 return err;
5676
5677 fpu_finit(&vcpu->arch.guest_fpu);
5678
5679 /*
5680 * Ensure guest xcr0 is valid for loading
5681 */
5682 vcpu->arch.xcr0 = XSTATE_FP;
5683
5684 vcpu->arch.cr0 |= X86_CR0_ET;
5685
5686 return 0;
5687 }
5688 EXPORT_SYMBOL_GPL(fx_init);
5689
5690 static void fx_free(struct kvm_vcpu *vcpu)
5691 {
5692 fpu_free(&vcpu->arch.guest_fpu);
5693 }
5694
5695 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5696 {
5697 if (vcpu->guest_fpu_loaded)
5698 return;
5699
5700 /*
5701 * Restore all possible states in the guest,
5702 * and assume host would use all available bits.
5703 * Guest xcr0 would be loaded later.
5704 */
5705 kvm_put_guest_xcr0(vcpu);
5706 vcpu->guest_fpu_loaded = 1;
5707 unlazy_fpu(current);
5708 fpu_restore_checking(&vcpu->arch.guest_fpu);
5709 trace_kvm_fpu(1);
5710 }
5711
5712 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5713 {
5714 kvm_put_guest_xcr0(vcpu);
5715
5716 if (!vcpu->guest_fpu_loaded)
5717 return;
5718
5719 vcpu->guest_fpu_loaded = 0;
5720 fpu_save_init(&vcpu->arch.guest_fpu);
5721 ++vcpu->stat.fpu_reload;
5722 kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
5723 trace_kvm_fpu(0);
5724 }
5725
5726 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5727 {
5728 if (vcpu->arch.time_page) {
5729 kvm_release_page_dirty(vcpu->arch.time_page);
5730 vcpu->arch.time_page = NULL;
5731 }
5732
5733 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
5734 fx_free(vcpu);
5735 kvm_x86_ops->vcpu_free(vcpu);
5736 }
5737
5738 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5739 unsigned int id)
5740 {
5741 if (check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
5742 printk_once(KERN_WARNING
5743 "kvm: SMP vm created on host with unstable TSC; "
5744 "guest TSC will not be reliable\n");
5745 return kvm_x86_ops->vcpu_create(kvm, id);
5746 }
5747
5748 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5749 {
5750 int r;
5751
5752 vcpu->arch.mtrr_state.have_fixed = 1;
5753 vcpu_load(vcpu);
5754 r = kvm_arch_vcpu_reset(vcpu);
5755 if (r == 0)
5756 r = kvm_mmu_setup(vcpu);
5757 vcpu_put(vcpu);
5758 if (r < 0)
5759 goto free_vcpu;
5760
5761 return 0;
5762 free_vcpu:
5763 kvm_x86_ops->vcpu_free(vcpu);
5764 return r;
5765 }
5766
5767 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5768 {
5769 vcpu_load(vcpu);
5770 kvm_mmu_unload(vcpu);
5771 vcpu_put(vcpu);
5772
5773 fx_free(vcpu);
5774 kvm_x86_ops->vcpu_free(vcpu);
5775 }
5776
5777 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5778 {
5779 vcpu->arch.nmi_pending = false;
5780 vcpu->arch.nmi_injected = false;
5781
5782 vcpu->arch.switch_db_regs = 0;
5783 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5784 vcpu->arch.dr6 = DR6_FIXED_1;
5785 vcpu->arch.dr7 = DR7_FIXED_1;
5786
5787 kvm_make_request(KVM_REQ_EVENT, vcpu);
5788
5789 return kvm_x86_ops->vcpu_reset(vcpu);
5790 }
5791
5792 int kvm_arch_hardware_enable(void *garbage)
5793 {
5794 struct kvm *kvm;
5795 struct kvm_vcpu *vcpu;
5796 int i;
5797
5798 kvm_shared_msr_cpu_online();
5799 list_for_each_entry(kvm, &vm_list, vm_list)
5800 kvm_for_each_vcpu(i, vcpu, kvm)
5801 if (vcpu->cpu == smp_processor_id())
5802 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5803 return kvm_x86_ops->hardware_enable(garbage);
5804 }
5805
5806 void kvm_arch_hardware_disable(void *garbage)
5807 {
5808 kvm_x86_ops->hardware_disable(garbage);
5809 drop_user_return_notifiers(garbage);
5810 }
5811
5812 int kvm_arch_hardware_setup(void)
5813 {
5814 return kvm_x86_ops->hardware_setup();
5815 }
5816
5817 void kvm_arch_hardware_unsetup(void)
5818 {
5819 kvm_x86_ops->hardware_unsetup();
5820 }
5821
5822 void kvm_arch_check_processor_compat(void *rtn)
5823 {
5824 kvm_x86_ops->check_processor_compatibility(rtn);
5825 }
5826
5827 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5828 {
5829 struct page *page;
5830 struct kvm *kvm;
5831 int r;
5832
5833 BUG_ON(vcpu->kvm == NULL);
5834 kvm = vcpu->kvm;
5835
5836 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
5837 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
5838 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5839 vcpu->arch.mmu.translate_gpa = translate_gpa;
5840 vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
5841 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5842 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5843 else
5844 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5845
5846 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5847 if (!page) {
5848 r = -ENOMEM;
5849 goto fail;
5850 }
5851 vcpu->arch.pio_data = page_address(page);
5852
5853 if (!kvm->arch.virtual_tsc_khz)
5854 kvm_arch_set_tsc_khz(kvm, max_tsc_khz);
5855
5856 r = kvm_mmu_create(vcpu);
5857 if (r < 0)
5858 goto fail_free_pio_data;
5859
5860 if (irqchip_in_kernel(kvm)) {
5861 r = kvm_create_lapic(vcpu);
5862 if (r < 0)
5863 goto fail_mmu_destroy;
5864 }
5865
5866 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5867 GFP_KERNEL);
5868 if (!vcpu->arch.mce_banks) {
5869 r = -ENOMEM;
5870 goto fail_free_lapic;
5871 }
5872 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5873
5874 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL))
5875 goto fail_free_mce_banks;
5876
5877 return 0;
5878 fail_free_mce_banks:
5879 kfree(vcpu->arch.mce_banks);
5880 fail_free_lapic:
5881 kvm_free_lapic(vcpu);
5882 fail_mmu_destroy:
5883 kvm_mmu_destroy(vcpu);
5884 fail_free_pio_data:
5885 free_page((unsigned long)vcpu->arch.pio_data);
5886 fail:
5887 return r;
5888 }
5889
5890 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5891 {
5892 int idx;
5893
5894 kfree(vcpu->arch.mce_banks);
5895 kvm_free_lapic(vcpu);
5896 idx = srcu_read_lock(&vcpu->kvm->srcu);
5897 kvm_mmu_destroy(vcpu);
5898 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5899 free_page((unsigned long)vcpu->arch.pio_data);
5900 }
5901
5902 struct kvm *kvm_arch_create_vm(void)
5903 {
5904 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5905
5906 if (!kvm)
5907 return ERR_PTR(-ENOMEM);
5908
5909 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5910 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5911
5912 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5913 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5914
5915 spin_lock_init(&kvm->arch.tsc_write_lock);
5916
5917 return kvm;
5918 }
5919
5920 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5921 {
5922 vcpu_load(vcpu);
5923 kvm_mmu_unload(vcpu);
5924 vcpu_put(vcpu);
5925 }
5926
5927 static void kvm_free_vcpus(struct kvm *kvm)
5928 {
5929 unsigned int i;
5930 struct kvm_vcpu *vcpu;
5931
5932 /*
5933 * Unpin any mmu pages first.
5934 */
5935 kvm_for_each_vcpu(i, vcpu, kvm)
5936 kvm_unload_vcpu_mmu(vcpu);
5937 kvm_for_each_vcpu(i, vcpu, kvm)
5938 kvm_arch_vcpu_free(vcpu);
5939
5940 mutex_lock(&kvm->lock);
5941 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5942 kvm->vcpus[i] = NULL;
5943
5944 atomic_set(&kvm->online_vcpus, 0);
5945 mutex_unlock(&kvm->lock);
5946 }
5947
5948 void kvm_arch_sync_events(struct kvm *kvm)
5949 {
5950 kvm_free_all_assigned_devices(kvm);
5951 kvm_free_pit(kvm);
5952 }
5953
5954 void kvm_arch_destroy_vm(struct kvm *kvm)
5955 {
5956 kvm_iommu_unmap_guest(kvm);
5957 kfree(kvm->arch.vpic);
5958 kfree(kvm->arch.vioapic);
5959 kvm_free_vcpus(kvm);
5960 kvm_free_physmem(kvm);
5961 if (kvm->arch.apic_access_page)
5962 put_page(kvm->arch.apic_access_page);
5963 if (kvm->arch.ept_identity_pagetable)
5964 put_page(kvm->arch.ept_identity_pagetable);
5965 cleanup_srcu_struct(&kvm->srcu);
5966 kfree(kvm);
5967 }
5968
5969 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5970 struct kvm_memory_slot *memslot,
5971 struct kvm_memory_slot old,
5972 struct kvm_userspace_memory_region *mem,
5973 int user_alloc)
5974 {
5975 int npages = memslot->npages;
5976 int map_flags = MAP_PRIVATE | MAP_ANONYMOUS;
5977
5978 /* Prevent internal slot pages from being moved by fork()/COW. */
5979 if (memslot->id >= KVM_MEMORY_SLOTS)
5980 map_flags = MAP_SHARED | MAP_ANONYMOUS;
5981
5982 /*To keep backward compatibility with older userspace,
5983 *x86 needs to hanlde !user_alloc case.
5984 */
5985 if (!user_alloc) {
5986 if (npages && !old.rmap) {
5987 unsigned long userspace_addr;
5988
5989 down_write(&current->mm->mmap_sem);
5990 userspace_addr = do_mmap(NULL, 0,
5991 npages * PAGE_SIZE,
5992 PROT_READ | PROT_WRITE,
5993 map_flags,
5994 0);
5995 up_write(&current->mm->mmap_sem);
5996
5997 if (IS_ERR((void *)userspace_addr))
5998 return PTR_ERR((void *)userspace_addr);
5999
6000 memslot->userspace_addr = userspace_addr;
6001 }
6002 }
6003
6004
6005 return 0;
6006 }
6007
6008 void kvm_arch_commit_memory_region(struct kvm *kvm,
6009 struct kvm_userspace_memory_region *mem,
6010 struct kvm_memory_slot old,
6011 int user_alloc)
6012 {
6013
6014 int npages = mem->memory_size >> PAGE_SHIFT;
6015
6016 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
6017 int ret;
6018
6019 down_write(&current->mm->mmap_sem);
6020 ret = do_munmap(current->mm, old.userspace_addr,
6021 old.npages * PAGE_SIZE);
6022 up_write(&current->mm->mmap_sem);
6023 if (ret < 0)
6024 printk(KERN_WARNING
6025 "kvm_vm_ioctl_set_memory_region: "
6026 "failed to munmap memory\n");
6027 }
6028
6029 spin_lock(&kvm->mmu_lock);
6030 if (!kvm->arch.n_requested_mmu_pages) {
6031 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
6032 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
6033 }
6034
6035 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
6036 spin_unlock(&kvm->mmu_lock);
6037 }
6038
6039 void kvm_arch_flush_shadow(struct kvm *kvm)
6040 {
6041 kvm_mmu_zap_all(kvm);
6042 kvm_reload_remote_mmus(kvm);
6043 }
6044
6045 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
6046 {
6047 return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
6048 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
6049 || vcpu->arch.nmi_pending ||
6050 (kvm_arch_interrupt_allowed(vcpu) &&
6051 kvm_cpu_has_interrupt(vcpu));
6052 }
6053
6054 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
6055 {
6056 int me;
6057 int cpu = vcpu->cpu;
6058
6059 if (waitqueue_active(&vcpu->wq)) {
6060 wake_up_interruptible(&vcpu->wq);
6061 ++vcpu->stat.halt_wakeup;
6062 }
6063
6064 me = get_cpu();
6065 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
6066 if (atomic_xchg(&vcpu->guest_mode, 0))
6067 smp_send_reschedule(cpu);
6068 put_cpu();
6069 }
6070
6071 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
6072 {
6073 return kvm_x86_ops->interrupt_allowed(vcpu);
6074 }
6075
6076 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
6077 {
6078 unsigned long current_rip = kvm_rip_read(vcpu) +
6079 get_segment_base(vcpu, VCPU_SREG_CS);
6080
6081 return current_rip == linear_rip;
6082 }
6083 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
6084
6085 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
6086 {
6087 unsigned long rflags;
6088
6089 rflags = kvm_x86_ops->get_rflags(vcpu);
6090 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6091 rflags &= ~X86_EFLAGS_TF;
6092 return rflags;
6093 }
6094 EXPORT_SYMBOL_GPL(kvm_get_rflags);
6095
6096 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
6097 {
6098 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
6099 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
6100 rflags |= X86_EFLAGS_TF;
6101 kvm_x86_ops->set_rflags(vcpu, rflags);
6102 kvm_make_request(KVM_REQ_EVENT, vcpu);
6103 }
6104 EXPORT_SYMBOL_GPL(kvm_set_rflags);
6105
6106 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
6107 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
6108 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
6109 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
6110 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
6111 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
6112 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
6113 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
6114 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
6115 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
6116 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
6117 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree