?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
KVM: x86: Save&restore interrupt shadow mask
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / arch / x86 / kvm / x86.c
blob84ffd95ee1987752de521753566cb088445447a3
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 *
10 * Authors:
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 * Amit Shah <amit.shah@qumranet.com>
14 * Ben-Ami Yassour <benami@il.ibm.com>
15 *
16 * This work is licensed under the terms of the GNU GPL, version 2. See
17 * the COPYING file in the top-level directory.
18 *
19 */
20
21 #include <linux/kvm_host.h>
22 #include "irq.h"
23 #include "mmu.h"
24 #include "i8254.h"
25 #include "tss.h"
26 #include "kvm_cache_regs.h"
27 #include "x86.h"
28
29 #include <linux/clocksource.h>
30 #include <linux/interrupt.h>
31 #include <linux/kvm.h>
32 #include <linux/fs.h>
33 #include <linux/vmalloc.h>
34 #include <linux/module.h>
35 #include <linux/mman.h>
36 #include <linux/highmem.h>
37 #include <linux/iommu.h>
38 #include <linux/intel-iommu.h>
39 #include <linux/cpufreq.h>
40 #include <linux/user-return-notifier.h>
41 #include <linux/srcu.h>
42 #include <linux/slab.h>
43 #include <trace/events/kvm.h>
44 #undef TRACE_INCLUDE_FILE
45 #define CREATE_TRACE_POINTS
46 #include "trace.h"
47
48 #include <asm/debugreg.h>
49 #include <asm/uaccess.h>
50 #include <asm/msr.h>
51 #include <asm/desc.h>
52 #include <asm/mtrr.h>
53 #include <asm/mce.h>
54
55 #define MAX_IO_MSRS 256
56 #define CR0_RESERVED_BITS \
57 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
58 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
59 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
60 #define CR4_RESERVED_BITS \
61 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
62 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
63 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
64 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
65
66 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
67
68 #define KVM_MAX_MCE_BANKS 32
69 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
70
71 /* EFER defaults:
72 * - enable syscall per default because its emulated by KVM
73 * - enable LME and LMA per default on 64 bit KVM
74 */
75 #ifdef CONFIG_X86_64
76 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
77 #else
78 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
79 #endif
80
81 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
82 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
83
84 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
85 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
86 struct kvm_cpuid_entry2 __user *entries);
87
88 struct kvm_x86_ops *kvm_x86_ops;
89 EXPORT_SYMBOL_GPL(kvm_x86_ops);
90
91 int ignore_msrs = 0;
92 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
93
94 #define KVM_NR_SHARED_MSRS 16
95
96 struct kvm_shared_msrs_global {
97 int nr;
98 u32 msrs[KVM_NR_SHARED_MSRS];
99 };
100
101 struct kvm_shared_msrs {
102 struct user_return_notifier urn;
103 bool registered;
104 struct kvm_shared_msr_values {
105 u64 host;
106 u64 curr;
107 } values[KVM_NR_SHARED_MSRS];
108 };
109
110 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
111 static DEFINE_PER_CPU(struct kvm_shared_msrs, shared_msrs);
112
113 struct kvm_stats_debugfs_item debugfs_entries[] = {
114 { "pf_fixed", VCPU_STAT(pf_fixed) },
115 { "pf_guest", VCPU_STAT(pf_guest) },
116 { "tlb_flush", VCPU_STAT(tlb_flush) },
117 { "invlpg", VCPU_STAT(invlpg) },
118 { "exits", VCPU_STAT(exits) },
119 { "io_exits", VCPU_STAT(io_exits) },
120 { "mmio_exits", VCPU_STAT(mmio_exits) },
121 { "signal_exits", VCPU_STAT(signal_exits) },
122 { "irq_window", VCPU_STAT(irq_window_exits) },
123 { "nmi_window", VCPU_STAT(nmi_window_exits) },
124 { "halt_exits", VCPU_STAT(halt_exits) },
125 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
126 { "hypercalls", VCPU_STAT(hypercalls) },
127 { "request_irq", VCPU_STAT(request_irq_exits) },
128 { "irq_exits", VCPU_STAT(irq_exits) },
129 { "host_state_reload", VCPU_STAT(host_state_reload) },
130 { "efer_reload", VCPU_STAT(efer_reload) },
131 { "fpu_reload", VCPU_STAT(fpu_reload) },
132 { "insn_emulation", VCPU_STAT(insn_emulation) },
133 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
134 { "irq_injections", VCPU_STAT(irq_injections) },
135 { "nmi_injections", VCPU_STAT(nmi_injections) },
136 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
137 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
138 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
139 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
140 { "mmu_flooded", VM_STAT(mmu_flooded) },
141 { "mmu_recycled", VM_STAT(mmu_recycled) },
142 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
143 { "mmu_unsync", VM_STAT(mmu_unsync) },
144 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
145 { "largepages", VM_STAT(lpages) },
146 { NULL }
147 };
148
149 static void kvm_on_user_return(struct user_return_notifier *urn)
150 {
151 unsigned slot;
152 struct kvm_shared_msrs *locals
153 = container_of(urn, struct kvm_shared_msrs, urn);
154 struct kvm_shared_msr_values *values;
155
156 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
157 values = &locals->values[slot];
158 if (values->host != values->curr) {
159 wrmsrl(shared_msrs_global.msrs[slot], values->host);
160 values->curr = values->host;
161 }
162 }
163 locals->registered = false;
164 user_return_notifier_unregister(urn);
165 }
166
167 static void shared_msr_update(unsigned slot, u32 msr)
168 {
169 struct kvm_shared_msrs *smsr;
170 u64 value;
171
172 smsr = &__get_cpu_var(shared_msrs);
173 /* only read, and nobody should modify it at this time,
174 * so don't need lock */
175 if (slot >= shared_msrs_global.nr) {
176 printk(KERN_ERR "kvm: invalid MSR slot!");
177 return;
178 }
179 rdmsrl_safe(msr, &value);
180 smsr->values[slot].host = value;
181 smsr->values[slot].curr = value;
182 }
183
184 void kvm_define_shared_msr(unsigned slot, u32 msr)
185 {
186 if (slot >= shared_msrs_global.nr)
187 shared_msrs_global.nr = slot + 1;
188 shared_msrs_global.msrs[slot] = msr;
189 /* we need ensured the shared_msr_global have been updated */
190 smp_wmb();
191 }
192 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
193
194 static void kvm_shared_msr_cpu_online(void)
195 {
196 unsigned i;
197
198 for (i = 0; i < shared_msrs_global.nr; ++i)
199 shared_msr_update(i, shared_msrs_global.msrs[i]);
200 }
201
202 void kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
203 {
204 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
205
206 if (((value ^ smsr->values[slot].curr) & mask) == 0)
207 return;
208 smsr->values[slot].curr = value;
209 wrmsrl(shared_msrs_global.msrs[slot], value);
210 if (!smsr->registered) {
211 smsr->urn.on_user_return = kvm_on_user_return;
212 user_return_notifier_register(&smsr->urn);
213 smsr->registered = true;
214 }
215 }
216 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
217
218 static void drop_user_return_notifiers(void *ignore)
219 {
220 struct kvm_shared_msrs *smsr = &__get_cpu_var(shared_msrs);
221
222 if (smsr->registered)
223 kvm_on_user_return(&smsr->urn);
224 }
225
226 unsigned long segment_base(u16 selector)
227 {
228 struct desc_ptr gdt;
229 struct desc_struct *d;
230 unsigned long table_base;
231 unsigned long v;
232
233 if (selector == 0)
234 return 0;
235
236 kvm_get_gdt(&gdt);
237 table_base = gdt.address;
238
239 if (selector & 4) { /* from ldt */
240 u16 ldt_selector = kvm_read_ldt();
241
242 table_base = segment_base(ldt_selector);
243 }
244 d = (struct desc_struct *)(table_base + (selector & ~7));
245 v = get_desc_base(d);
246 #ifdef CONFIG_X86_64
247 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
248 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
249 #endif
250 return v;
251 }
252 EXPORT_SYMBOL_GPL(segment_base);
253
254 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
255 {
256 if (irqchip_in_kernel(vcpu->kvm))
257 return vcpu->arch.apic_base;
258 else
259 return vcpu->arch.apic_base;
260 }
261 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
262
263 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
264 {
265 /* TODO: reserve bits check */
266 if (irqchip_in_kernel(vcpu->kvm))
267 kvm_lapic_set_base(vcpu, data);
268 else
269 vcpu->arch.apic_base = data;
270 }
271 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
272
273 #define EXCPT_BENIGN 0
274 #define EXCPT_CONTRIBUTORY 1
275 #define EXCPT_PF 2
276
277 static int exception_class(int vector)
278 {
279 switch (vector) {
280 case PF_VECTOR:
281 return EXCPT_PF;
282 case DE_VECTOR:
283 case TS_VECTOR:
284 case NP_VECTOR:
285 case SS_VECTOR:
286 case GP_VECTOR:
287 return EXCPT_CONTRIBUTORY;
288 default:
289 break;
290 }
291 return EXCPT_BENIGN;
292 }
293
294 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
295 unsigned nr, bool has_error, u32 error_code)
296 {
297 u32 prev_nr;
298 int class1, class2;
299
300 if (!vcpu->arch.exception.pending) {
301 queue:
302 vcpu->arch.exception.pending = true;
303 vcpu->arch.exception.has_error_code = has_error;
304 vcpu->arch.exception.nr = nr;
305 vcpu->arch.exception.error_code = error_code;
306 return;
307 }
308
309 /* to check exception */
310 prev_nr = vcpu->arch.exception.nr;
311 if (prev_nr == DF_VECTOR) {
312 /* triple fault -> shutdown */
313 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
314 return;
315 }
316 class1 = exception_class(prev_nr);
317 class2 = exception_class(nr);
318 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
319 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
320 /* generate double fault per SDM Table 5-5 */
321 vcpu->arch.exception.pending = true;
322 vcpu->arch.exception.has_error_code = true;
323 vcpu->arch.exception.nr = DF_VECTOR;
324 vcpu->arch.exception.error_code = 0;
325 } else
326 /* replace previous exception with a new one in a hope
327 that instruction re-execution will regenerate lost
328 exception */
329 goto queue;
330 }
331
332 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
333 {
334 kvm_multiple_exception(vcpu, nr, false, 0);
335 }
336 EXPORT_SYMBOL_GPL(kvm_queue_exception);
337
338 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
339 u32 error_code)
340 {
341 ++vcpu->stat.pf_guest;
342 vcpu->arch.cr2 = addr;
343 kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
344 }
345
346 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
347 {
348 vcpu->arch.nmi_pending = 1;
349 }
350 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
351
352 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
353 {
354 kvm_multiple_exception(vcpu, nr, true, error_code);
355 }
356 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
357
358 /*
359 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
360 * a #GP and return false.
361 */
362 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
363 {
364 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
365 return true;
366 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
367 return false;
368 }
369 EXPORT_SYMBOL_GPL(kvm_require_cpl);
370
371 /*
372 * Load the pae pdptrs. Return true is they are all valid.
373 */
374 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
375 {
376 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
377 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
378 int i;
379 int ret;
380 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
381
382 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
383 offset * sizeof(u64), sizeof(pdpte));
384 if (ret < 0) {
385 ret = 0;
386 goto out;
387 }
388 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
389 if (is_present_gpte(pdpte[i]) &&
390 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
391 ret = 0;
392 goto out;
393 }
394 }
395 ret = 1;
396
397 memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
398 __set_bit(VCPU_EXREG_PDPTR,
399 (unsigned long *)&vcpu->arch.regs_avail);
400 __set_bit(VCPU_EXREG_PDPTR,
401 (unsigned long *)&vcpu->arch.regs_dirty);
402 out:
403
404 return ret;
405 }
406 EXPORT_SYMBOL_GPL(load_pdptrs);
407
408 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
409 {
410 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
411 bool changed = true;
412 int r;
413
414 if (is_long_mode(vcpu) || !is_pae(vcpu))
415 return false;
416
417 if (!test_bit(VCPU_EXREG_PDPTR,
418 (unsigned long *)&vcpu->arch.regs_avail))
419 return true;
420
421 r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
422 if (r < 0)
423 goto out;
424 changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
425 out:
426
427 return changed;
428 }
429
430 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
431 {
432 cr0 |= X86_CR0_ET;
433
434 #ifdef CONFIG_X86_64
435 if (cr0 & 0xffffffff00000000UL) {
436 kvm_inject_gp(vcpu, 0);
437 return;
438 }
439 #endif
440
441 cr0 &= ~CR0_RESERVED_BITS;
442
443 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
444 kvm_inject_gp(vcpu, 0);
445 return;
446 }
447
448 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
449 kvm_inject_gp(vcpu, 0);
450 return;
451 }
452
453 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
454 #ifdef CONFIG_X86_64
455 if ((vcpu->arch.efer & EFER_LME)) {
456 int cs_db, cs_l;
457
458 if (!is_pae(vcpu)) {
459 kvm_inject_gp(vcpu, 0);
460 return;
461 }
462 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
463 if (cs_l) {
464 kvm_inject_gp(vcpu, 0);
465 return;
466
467 }
468 } else
469 #endif
470 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
471 kvm_inject_gp(vcpu, 0);
472 return;
473 }
474
475 }
476
477 kvm_x86_ops->set_cr0(vcpu, cr0);
478 vcpu->arch.cr0 = cr0;
479
480 kvm_mmu_reset_context(vcpu);
481 return;
482 }
483 EXPORT_SYMBOL_GPL(kvm_set_cr0);
484
485 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
486 {
487 kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0ful) | (msw & 0x0f));
488 }
489 EXPORT_SYMBOL_GPL(kvm_lmsw);
490
491 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
492 {
493 unsigned long old_cr4 = kvm_read_cr4(vcpu);
494 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
495
496 if (cr4 & CR4_RESERVED_BITS) {
497 kvm_inject_gp(vcpu, 0);
498 return;
499 }
500
501 if (is_long_mode(vcpu)) {
502 if (!(cr4 & X86_CR4_PAE)) {
503 kvm_inject_gp(vcpu, 0);
504 return;
505 }
506 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
507 && ((cr4 ^ old_cr4) & pdptr_bits)
508 && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
509 kvm_inject_gp(vcpu, 0);
510 return;
511 }
512
513 if (cr4 & X86_CR4_VMXE) {
514 kvm_inject_gp(vcpu, 0);
515 return;
516 }
517 kvm_x86_ops->set_cr4(vcpu, cr4);
518 vcpu->arch.cr4 = cr4;
519 vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
520 kvm_mmu_reset_context(vcpu);
521 }
522 EXPORT_SYMBOL_GPL(kvm_set_cr4);
523
524 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
525 {
526 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
527 kvm_mmu_sync_roots(vcpu);
528 kvm_mmu_flush_tlb(vcpu);
529 return;
530 }
531
532 if (is_long_mode(vcpu)) {
533 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
534 kvm_inject_gp(vcpu, 0);
535 return;
536 }
537 } else {
538 if (is_pae(vcpu)) {
539 if (cr3 & CR3_PAE_RESERVED_BITS) {
540 kvm_inject_gp(vcpu, 0);
541 return;
542 }
543 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
544 kvm_inject_gp(vcpu, 0);
545 return;
546 }
547 }
548 /*
549 * We don't check reserved bits in nonpae mode, because
550 * this isn't enforced, and VMware depends on this.
551 */
552 }
553
554 /*
555 * Does the new cr3 value map to physical memory? (Note, we
556 * catch an invalid cr3 even in real-mode, because it would
557 * cause trouble later on when we turn on paging anyway.)
558 *
559 * A real CPU would silently accept an invalid cr3 and would
560 * attempt to use it - with largely undefined (and often hard
561 * to debug) behavior on the guest side.
562 */
563 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
564 kvm_inject_gp(vcpu, 0);
565 else {
566 vcpu->arch.cr3 = cr3;
567 vcpu->arch.mmu.new_cr3(vcpu);
568 }
569 }
570 EXPORT_SYMBOL_GPL(kvm_set_cr3);
571
572 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
573 {
574 if (cr8 & CR8_RESERVED_BITS) {
575 kvm_inject_gp(vcpu, 0);
576 return;
577 }
578 if (irqchip_in_kernel(vcpu->kvm))
579 kvm_lapic_set_tpr(vcpu, cr8);
580 else
581 vcpu->arch.cr8 = cr8;
582 }
583 EXPORT_SYMBOL_GPL(kvm_set_cr8);
584
585 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
586 {
587 if (irqchip_in_kernel(vcpu->kvm))
588 return kvm_lapic_get_cr8(vcpu);
589 else
590 return vcpu->arch.cr8;
591 }
592 EXPORT_SYMBOL_GPL(kvm_get_cr8);
593
594 static inline u32 bit(int bitno)
595 {
596 return 1 << (bitno & 31);
597 }
598
599 /*
600 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
601 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
602 *
603 * This list is modified at module load time to reflect the
604 * capabilities of the host cpu. This capabilities test skips MSRs that are
605 * kvm-specific. Those are put in the beginning of the list.
606 */
607
608 #define KVM_SAVE_MSRS_BEGIN 5
609 static u32 msrs_to_save[] = {
610 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
611 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
612 HV_X64_MSR_APIC_ASSIST_PAGE,
613 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
614 MSR_K6_STAR,
615 #ifdef CONFIG_X86_64
616 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
617 #endif
618 MSR_IA32_TSC, MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
619 };
620
621 static unsigned num_msrs_to_save;
622
623 static u32 emulated_msrs[] = {
624 MSR_IA32_MISC_ENABLE,
625 };
626
627 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
628 {
629 if (efer & efer_reserved_bits) {
630 kvm_inject_gp(vcpu, 0);
631 return;
632 }
633
634 if (is_paging(vcpu)
635 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME)) {
636 kvm_inject_gp(vcpu, 0);
637 return;
638 }
639
640 if (efer & EFER_FFXSR) {
641 struct kvm_cpuid_entry2 *feat;
642
643 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
644 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
645 kvm_inject_gp(vcpu, 0);
646 return;
647 }
648 }
649
650 if (efer & EFER_SVME) {
651 struct kvm_cpuid_entry2 *feat;
652
653 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
654 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
655 kvm_inject_gp(vcpu, 0);
656 return;
657 }
658 }
659
660 kvm_x86_ops->set_efer(vcpu, efer);
661
662 efer &= ~EFER_LMA;
663 efer |= vcpu->arch.efer & EFER_LMA;
664
665 vcpu->arch.efer = efer;
666
667 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
668 kvm_mmu_reset_context(vcpu);
669 }
670
671 void kvm_enable_efer_bits(u64 mask)
672 {
673 efer_reserved_bits &= ~mask;
674 }
675 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
676
677
678 /*
679 * Writes msr value into into the appropriate "register".
680 * Returns 0 on success, non-0 otherwise.
681 * Assumes vcpu_load() was already called.
682 */
683 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
684 {
685 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
686 }
687
688 /*
689 * Adapt set_msr() to msr_io()'s calling convention
690 */
691 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
692 {
693 return kvm_set_msr(vcpu, index, *data);
694 }
695
696 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
697 {
698 static int version;
699 struct pvclock_wall_clock wc;
700 struct timespec boot;
701
702 if (!wall_clock)
703 return;
704
705 version++;
706
707 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
708
709 /*
710 * The guest calculates current wall clock time by adding
711 * system time (updated by kvm_write_guest_time below) to the
712 * wall clock specified here. guest system time equals host
713 * system time for us, thus we must fill in host boot time here.
714 */
715 getboottime(&boot);
716
717 wc.sec = boot.tv_sec;
718 wc.nsec = boot.tv_nsec;
719 wc.version = version;
720
721 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
722
723 version++;
724 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
725 }
726
727 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
728 {
729 uint32_t quotient, remainder;
730
731 /* Don't try to replace with do_div(), this one calculates
732 * "(dividend << 32) / divisor" */
733 __asm__ ( "divl %4"
734 : "=a" (quotient), "=d" (remainder)
735 : "0" (0), "1" (dividend), "r" (divisor) );
736 return quotient;
737 }
738
739 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
740 {
741 uint64_t nsecs = 1000000000LL;
742 int32_t shift = 0;
743 uint64_t tps64;
744 uint32_t tps32;
745
746 tps64 = tsc_khz * 1000LL;
747 while (tps64 > nsecs*2) {
748 tps64 >>= 1;
749 shift--;
750 }
751
752 tps32 = (uint32_t)tps64;
753 while (tps32 <= (uint32_t)nsecs) {
754 tps32 <<= 1;
755 shift++;
756 }
757
758 hv_clock->tsc_shift = shift;
759 hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
760
761 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
762 __func__, tsc_khz, hv_clock->tsc_shift,
763 hv_clock->tsc_to_system_mul);
764 }
765
766 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
767
768 static void kvm_write_guest_time(struct kvm_vcpu *v)
769 {
770 struct timespec ts;
771 unsigned long flags;
772 struct kvm_vcpu_arch *vcpu = &v->arch;
773 void *shared_kaddr;
774 unsigned long this_tsc_khz;
775
776 if ((!vcpu->time_page))
777 return;
778
779 this_tsc_khz = get_cpu_var(cpu_tsc_khz);
780 if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
781 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
782 vcpu->hv_clock_tsc_khz = this_tsc_khz;
783 }
784 put_cpu_var(cpu_tsc_khz);
785
786 /* Keep irq disabled to prevent changes to the clock */
787 local_irq_save(flags);
788 kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
789 ktime_get_ts(&ts);
790 monotonic_to_bootbased(&ts);
791 local_irq_restore(flags);
792
793 /* With all the info we got, fill in the values */
794
795 vcpu->hv_clock.system_time = ts.tv_nsec +
796 (NSEC_PER_SEC * (u64)ts.tv_sec) + v->kvm->arch.kvmclock_offset;
797
798 /*
799 * The interface expects us to write an even number signaling that the
800 * update is finished. Since the guest won't see the intermediate
801 * state, we just increase by 2 at the end.
802 */
803 vcpu->hv_clock.version += 2;
804
805 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
806
807 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
808 sizeof(vcpu->hv_clock));
809
810 kunmap_atomic(shared_kaddr, KM_USER0);
811
812 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
813 }
814
815 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
816 {
817 struct kvm_vcpu_arch *vcpu = &v->arch;
818
819 if (!vcpu->time_page)
820 return 0;
821 set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
822 return 1;
823 }
824
825 static bool msr_mtrr_valid(unsigned msr)
826 {
827 switch (msr) {
828 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
829 case MSR_MTRRfix64K_00000:
830 case MSR_MTRRfix16K_80000:
831 case MSR_MTRRfix16K_A0000:
832 case MSR_MTRRfix4K_C0000:
833 case MSR_MTRRfix4K_C8000:
834 case MSR_MTRRfix4K_D0000:
835 case MSR_MTRRfix4K_D8000:
836 case MSR_MTRRfix4K_E0000:
837 case MSR_MTRRfix4K_E8000:
838 case MSR_MTRRfix4K_F0000:
839 case MSR_MTRRfix4K_F8000:
840 case MSR_MTRRdefType:
841 case MSR_IA32_CR_PAT:
842 return true;
843 case 0x2f8:
844 return true;
845 }
846 return false;
847 }
848
849 static bool valid_pat_type(unsigned t)
850 {
851 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
852 }
853
854 static bool valid_mtrr_type(unsigned t)
855 {
856 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
857 }
858
859 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
860 {
861 int i;
862
863 if (!msr_mtrr_valid(msr))
864 return false;
865
866 if (msr == MSR_IA32_CR_PAT) {
867 for (i = 0; i < 8; i++)
868 if (!valid_pat_type((data >> (i * 8)) & 0xff))
869 return false;
870 return true;
871 } else if (msr == MSR_MTRRdefType) {
872 if (data & ~0xcff)
873 return false;
874 return valid_mtrr_type(data & 0xff);
875 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
876 for (i = 0; i < 8 ; i++)
877 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
878 return false;
879 return true;
880 }
881
882 /* variable MTRRs */
883 return valid_mtrr_type(data & 0xff);
884 }
885
886 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
887 {
888 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
889
890 if (!mtrr_valid(vcpu, msr, data))
891 return 1;
892
893 if (msr == MSR_MTRRdefType) {
894 vcpu->arch.mtrr_state.def_type = data;
895 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
896 } else if (msr == MSR_MTRRfix64K_00000)
897 p[0] = data;
898 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
899 p[1 + msr - MSR_MTRRfix16K_80000] = data;
900 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
901 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
902 else if (msr == MSR_IA32_CR_PAT)
903 vcpu->arch.pat = data;
904 else { /* Variable MTRRs */
905 int idx, is_mtrr_mask;
906 u64 *pt;
907
908 idx = (msr - 0x200) / 2;
909 is_mtrr_mask = msr - 0x200 - 2 * idx;
910 if (!is_mtrr_mask)
911 pt =
912 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
913 else
914 pt =
915 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
916 *pt = data;
917 }
918
919 kvm_mmu_reset_context(vcpu);
920 return 0;
921 }
922
923 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
924 {
925 u64 mcg_cap = vcpu->arch.mcg_cap;
926 unsigned bank_num = mcg_cap & 0xff;
927
928 switch (msr) {
929 case MSR_IA32_MCG_STATUS:
930 vcpu->arch.mcg_status = data;
931 break;
932 case MSR_IA32_MCG_CTL:
933 if (!(mcg_cap & MCG_CTL_P))
934 return 1;
935 if (data != 0 && data != ~(u64)0)
936 return -1;
937 vcpu->arch.mcg_ctl = data;
938 break;
939 default:
940 if (msr >= MSR_IA32_MC0_CTL &&
941 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
942 u32 offset = msr - MSR_IA32_MC0_CTL;
943 /* only 0 or all 1s can be written to IA32_MCi_CTL
944 * some Linux kernels though clear bit 10 in bank 4 to
945 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
946 * this to avoid an uncatched #GP in the guest
947 */
948 if ((offset & 0x3) == 0 &&
949 data != 0 && (data | (1 << 10)) != ~(u64)0)
950 return -1;
951 vcpu->arch.mce_banks[offset] = data;
952 break;
953 }
954 return 1;
955 }
956 return 0;
957 }
958
959 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
960 {
961 struct kvm *kvm = vcpu->kvm;
962 int lm = is_long_mode(vcpu);
963 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
964 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
965 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
966 : kvm->arch.xen_hvm_config.blob_size_32;
967 u32 page_num = data & ~PAGE_MASK;
968 u64 page_addr = data & PAGE_MASK;
969 u8 *page;
970 int r;
971
972 r = -E2BIG;
973 if (page_num >= blob_size)
974 goto out;
975 r = -ENOMEM;
976 page = kzalloc(PAGE_SIZE, GFP_KERNEL);
977 if (!page)
978 goto out;
979 r = -EFAULT;
980 if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
981 goto out_free;
982 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
983 goto out_free;
984 r = 0;
985 out_free:
986 kfree(page);
987 out:
988 return r;
989 }
990
991 static bool kvm_hv_hypercall_enabled(struct kvm *kvm)
992 {
993 return kvm->arch.hv_hypercall & HV_X64_MSR_HYPERCALL_ENABLE;
994 }
995
996 static bool kvm_hv_msr_partition_wide(u32 msr)
997 {
998 bool r = false;
999 switch (msr) {
1000 case HV_X64_MSR_GUEST_OS_ID:
1001 case HV_X64_MSR_HYPERCALL:
1002 r = true;
1003 break;
1004 }
1005
1006 return r;
1007 }
1008
1009 static int set_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1010 {
1011 struct kvm *kvm = vcpu->kvm;
1012
1013 switch (msr) {
1014 case HV_X64_MSR_GUEST_OS_ID:
1015 kvm->arch.hv_guest_os_id = data;
1016 /* setting guest os id to zero disables hypercall page */
1017 if (!kvm->arch.hv_guest_os_id)
1018 kvm->arch.hv_hypercall &= ~HV_X64_MSR_HYPERCALL_ENABLE;
1019 break;
1020 case HV_X64_MSR_HYPERCALL: {
1021 u64 gfn;
1022 unsigned long addr;
1023 u8 instructions[4];
1024
1025 /* if guest os id is not set hypercall should remain disabled */
1026 if (!kvm->arch.hv_guest_os_id)
1027 break;
1028 if (!(data & HV_X64_MSR_HYPERCALL_ENABLE)) {
1029 kvm->arch.hv_hypercall = data;
1030 break;
1031 }
1032 gfn = data >> HV_X64_MSR_HYPERCALL_PAGE_ADDRESS_SHIFT;
1033 addr = gfn_to_hva(kvm, gfn);
1034 if (kvm_is_error_hva(addr))
1035 return 1;
1036 kvm_x86_ops->patch_hypercall(vcpu, instructions);
1037 ((unsigned char *)instructions)[3] = 0xc3; /* ret */
1038 if (copy_to_user((void __user *)addr, instructions, 4))
1039 return 1;
1040 kvm->arch.hv_hypercall = data;
1041 break;
1042 }
1043 default:
1044 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1045 "data 0x%llx\n", msr, data);
1046 return 1;
1047 }
1048 return 0;
1049 }
1050
1051 static int set_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1052 {
1053 switch (msr) {
1054 case HV_X64_MSR_APIC_ASSIST_PAGE: {
1055 unsigned long addr;
1056
1057 if (!(data & HV_X64_MSR_APIC_ASSIST_PAGE_ENABLE)) {
1058 vcpu->arch.hv_vapic = data;
1059 break;
1060 }
1061 addr = gfn_to_hva(vcpu->kvm, data >>
1062 HV_X64_MSR_APIC_ASSIST_PAGE_ADDRESS_SHIFT);
1063 if (kvm_is_error_hva(addr))
1064 return 1;
1065 if (clear_user((void __user *)addr, PAGE_SIZE))
1066 return 1;
1067 vcpu->arch.hv_vapic = data;
1068 break;
1069 }
1070 case HV_X64_MSR_EOI:
1071 return kvm_hv_vapic_msr_write(vcpu, APIC_EOI, data);
1072 case HV_X64_MSR_ICR:
1073 return kvm_hv_vapic_msr_write(vcpu, APIC_ICR, data);
1074 case HV_X64_MSR_TPR:
1075 return kvm_hv_vapic_msr_write(vcpu, APIC_TASKPRI, data);
1076 default:
1077 pr_unimpl(vcpu, "HYPER-V unimplemented wrmsr: 0x%x "
1078 "data 0x%llx\n", msr, data);
1079 return 1;
1080 }
1081
1082 return 0;
1083 }
1084
1085 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1086 {
1087 switch (msr) {
1088 case MSR_EFER:
1089 set_efer(vcpu, data);
1090 break;
1091 case MSR_K7_HWCR:
1092 data &= ~(u64)0x40; /* ignore flush filter disable */
1093 if (data != 0) {
1094 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
1095 data);
1096 return 1;
1097 }
1098 break;
1099 case MSR_FAM10H_MMIO_CONF_BASE:
1100 if (data != 0) {
1101 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
1102 "0x%llx\n", data);
1103 return 1;
1104 }
1105 break;
1106 case MSR_AMD64_NB_CFG:
1107 break;
1108 case MSR_IA32_DEBUGCTLMSR:
1109 if (!data) {
1110 /* We support the non-activated case already */
1111 break;
1112 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
1113 /* Values other than LBR and BTF are vendor-specific,
1114 thus reserved and should throw a #GP */
1115 return 1;
1116 }
1117 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
1118 __func__, data);
1119 break;
1120 case MSR_IA32_UCODE_REV:
1121 case MSR_IA32_UCODE_WRITE:
1122 case MSR_VM_HSAVE_PA:
1123 case MSR_AMD64_PATCH_LOADER:
1124 break;
1125 case 0x200 ... 0x2ff:
1126 return set_msr_mtrr(vcpu, msr, data);
1127 case MSR_IA32_APICBASE:
1128 kvm_set_apic_base(vcpu, data);
1129 break;
1130 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1131 return kvm_x2apic_msr_write(vcpu, msr, data);
1132 case MSR_IA32_MISC_ENABLE:
1133 vcpu->arch.ia32_misc_enable_msr = data;
1134 break;
1135 case MSR_KVM_WALL_CLOCK:
1136 vcpu->kvm->arch.wall_clock = data;
1137 kvm_write_wall_clock(vcpu->kvm, data);
1138 break;
1139 case MSR_KVM_SYSTEM_TIME: {
1140 if (vcpu->arch.time_page) {
1141 kvm_release_page_dirty(vcpu->arch.time_page);
1142 vcpu->arch.time_page = NULL;
1143 }
1144
1145 vcpu->arch.time = data;
1146
1147 /* we verify if the enable bit is set... */
1148 if (!(data & 1))
1149 break;
1150
1151 /* ...but clean it before doing the actual write */
1152 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
1153
1154 vcpu->arch.time_page =
1155 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
1156
1157 if (is_error_page(vcpu->arch.time_page)) {
1158 kvm_release_page_clean(vcpu->arch.time_page);
1159 vcpu->arch.time_page = NULL;
1160 }
1161
1162 kvm_request_guest_time_update(vcpu);
1163 break;
1164 }
1165 case MSR_IA32_MCG_CTL:
1166 case MSR_IA32_MCG_STATUS:
1167 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1168 return set_msr_mce(vcpu, msr, data);
1169
1170 /* Performance counters are not protected by a CPUID bit,
1171 * so we should check all of them in the generic path for the sake of
1172 * cross vendor migration.
1173 * Writing a zero into the event select MSRs disables them,
1174 * which we perfectly emulate ;-). Any other value should be at least
1175 * reported, some guests depend on them.
1176 */
1177 case MSR_P6_EVNTSEL0:
1178 case MSR_P6_EVNTSEL1:
1179 case MSR_K7_EVNTSEL0:
1180 case MSR_K7_EVNTSEL1:
1181 case MSR_K7_EVNTSEL2:
1182 case MSR_K7_EVNTSEL3:
1183 if (data != 0)
1184 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1185 "0x%x data 0x%llx\n", msr, data);
1186 break;
1187 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1188 * so we ignore writes to make it happy.
1189 */
1190 case MSR_P6_PERFCTR0:
1191 case MSR_P6_PERFCTR1:
1192 case MSR_K7_PERFCTR0:
1193 case MSR_K7_PERFCTR1:
1194 case MSR_K7_PERFCTR2:
1195 case MSR_K7_PERFCTR3:
1196 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
1197 "0x%x data 0x%llx\n", msr, data);
1198 break;
1199 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1200 if (kvm_hv_msr_partition_wide(msr)) {
1201 int r;
1202 mutex_lock(&vcpu->kvm->lock);
1203 r = set_msr_hyperv_pw(vcpu, msr, data);
1204 mutex_unlock(&vcpu->kvm->lock);
1205 return r;
1206 } else
1207 return set_msr_hyperv(vcpu, msr, data);
1208 break;
1209 default:
1210 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1211 return xen_hvm_config(vcpu, data);
1212 if (!ignore_msrs) {
1213 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1214 msr, data);
1215 return 1;
1216 } else {
1217 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
1218 msr, data);
1219 break;
1220 }
1221 }
1222 return 0;
1223 }
1224 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
1225
1226
1227 /*
1228 * Reads an msr value (of 'msr_index') into 'pdata'.
1229 * Returns 0 on success, non-0 otherwise.
1230 * Assumes vcpu_load() was already called.
1231 */
1232 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1233 {
1234 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
1235 }
1236
1237 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1238 {
1239 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
1240
1241 if (!msr_mtrr_valid(msr))
1242 return 1;
1243
1244 if (msr == MSR_MTRRdefType)
1245 *pdata = vcpu->arch.mtrr_state.def_type +
1246 (vcpu->arch.mtrr_state.enabled << 10);
1247 else if (msr == MSR_MTRRfix64K_00000)
1248 *pdata = p[0];
1249 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
1250 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
1251 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
1252 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
1253 else if (msr == MSR_IA32_CR_PAT)
1254 *pdata = vcpu->arch.pat;
1255 else { /* Variable MTRRs */
1256 int idx, is_mtrr_mask;
1257 u64 *pt;
1258
1259 idx = (msr - 0x200) / 2;
1260 is_mtrr_mask = msr - 0x200 - 2 * idx;
1261 if (!is_mtrr_mask)
1262 pt =
1263 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1264 else
1265 pt =
1266 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1267 *pdata = *pt;
1268 }
1269
1270 return 0;
1271 }
1272
1273 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1274 {
1275 u64 data;
1276 u64 mcg_cap = vcpu->arch.mcg_cap;
1277 unsigned bank_num = mcg_cap & 0xff;
1278
1279 switch (msr) {
1280 case MSR_IA32_P5_MC_ADDR:
1281 case MSR_IA32_P5_MC_TYPE:
1282 data = 0;
1283 break;
1284 case MSR_IA32_MCG_CAP:
1285 data = vcpu->arch.mcg_cap;
1286 break;
1287 case MSR_IA32_MCG_CTL:
1288 if (!(mcg_cap & MCG_CTL_P))
1289 return 1;
1290 data = vcpu->arch.mcg_ctl;
1291 break;
1292 case MSR_IA32_MCG_STATUS:
1293 data = vcpu->arch.mcg_status;
1294 break;
1295 default:
1296 if (msr >= MSR_IA32_MC0_CTL &&
1297 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1298 u32 offset = msr - MSR_IA32_MC0_CTL;
1299 data = vcpu->arch.mce_banks[offset];
1300 break;
1301 }
1302 return 1;
1303 }
1304 *pdata = data;
1305 return 0;
1306 }
1307
1308 static int get_msr_hyperv_pw(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1309 {
1310 u64 data = 0;
1311 struct kvm *kvm = vcpu->kvm;
1312
1313 switch (msr) {
1314 case HV_X64_MSR_GUEST_OS_ID:
1315 data = kvm->arch.hv_guest_os_id;
1316 break;
1317 case HV_X64_MSR_HYPERCALL:
1318 data = kvm->arch.hv_hypercall;
1319 break;
1320 default:
1321 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1322 return 1;
1323 }
1324
1325 *pdata = data;
1326 return 0;
1327 }
1328
1329 static int get_msr_hyperv(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1330 {
1331 u64 data = 0;
1332
1333 switch (msr) {
1334 case HV_X64_MSR_VP_INDEX: {
1335 int r;
1336 struct kvm_vcpu *v;
1337 kvm_for_each_vcpu(r, v, vcpu->kvm)
1338 if (v == vcpu)
1339 data = r;
1340 break;
1341 }
1342 case HV_X64_MSR_EOI:
1343 return kvm_hv_vapic_msr_read(vcpu, APIC_EOI, pdata);
1344 case HV_X64_MSR_ICR:
1345 return kvm_hv_vapic_msr_read(vcpu, APIC_ICR, pdata);
1346 case HV_X64_MSR_TPR:
1347 return kvm_hv_vapic_msr_read(vcpu, APIC_TASKPRI, pdata);
1348 default:
1349 pr_unimpl(vcpu, "Hyper-V unhandled rdmsr: 0x%x\n", msr);
1350 return 1;
1351 }
1352 *pdata = data;
1353 return 0;
1354 }
1355
1356 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1357 {
1358 u64 data;
1359
1360 switch (msr) {
1361 case MSR_IA32_PLATFORM_ID:
1362 case MSR_IA32_UCODE_REV:
1363 case MSR_IA32_EBL_CR_POWERON:
1364 case MSR_IA32_DEBUGCTLMSR:
1365 case MSR_IA32_LASTBRANCHFROMIP:
1366 case MSR_IA32_LASTBRANCHTOIP:
1367 case MSR_IA32_LASTINTFROMIP:
1368 case MSR_IA32_LASTINTTOIP:
1369 case MSR_K8_SYSCFG:
1370 case MSR_K7_HWCR:
1371 case MSR_VM_HSAVE_PA:
1372 case MSR_P6_PERFCTR0:
1373 case MSR_P6_PERFCTR1:
1374 case MSR_P6_EVNTSEL0:
1375 case MSR_P6_EVNTSEL1:
1376 case MSR_K7_EVNTSEL0:
1377 case MSR_K7_PERFCTR0:
1378 case MSR_K8_INT_PENDING_MSG:
1379 case MSR_AMD64_NB_CFG:
1380 case MSR_FAM10H_MMIO_CONF_BASE:
1381 data = 0;
1382 break;
1383 case MSR_MTRRcap:
1384 data = 0x500 | KVM_NR_VAR_MTRR;
1385 break;
1386 case 0x200 ... 0x2ff:
1387 return get_msr_mtrr(vcpu, msr, pdata);
1388 case 0xcd: /* fsb frequency */
1389 data = 3;
1390 break;
1391 case MSR_IA32_APICBASE:
1392 data = kvm_get_apic_base(vcpu);
1393 break;
1394 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1395 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1396 break;
1397 case MSR_IA32_MISC_ENABLE:
1398 data = vcpu->arch.ia32_misc_enable_msr;
1399 break;
1400 case MSR_IA32_PERF_STATUS:
1401 /* TSC increment by tick */
1402 data = 1000ULL;
1403 /* CPU multiplier */
1404 data |= (((uint64_t)4ULL) << 40);
1405 break;
1406 case MSR_EFER:
1407 data = vcpu->arch.efer;
1408 break;
1409 case MSR_KVM_WALL_CLOCK:
1410 data = vcpu->kvm->arch.wall_clock;
1411 break;
1412 case MSR_KVM_SYSTEM_TIME:
1413 data = vcpu->arch.time;
1414 break;
1415 case MSR_IA32_P5_MC_ADDR:
1416 case MSR_IA32_P5_MC_TYPE:
1417 case MSR_IA32_MCG_CAP:
1418 case MSR_IA32_MCG_CTL:
1419 case MSR_IA32_MCG_STATUS:
1420 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1421 return get_msr_mce(vcpu, msr, pdata);
1422 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
1423 if (kvm_hv_msr_partition_wide(msr)) {
1424 int r;
1425 mutex_lock(&vcpu->kvm->lock);
1426 r = get_msr_hyperv_pw(vcpu, msr, pdata);
1427 mutex_unlock(&vcpu->kvm->lock);
1428 return r;
1429 } else
1430 return get_msr_hyperv(vcpu, msr, pdata);
1431 break;
1432 default:
1433 if (!ignore_msrs) {
1434 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1435 return 1;
1436 } else {
1437 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1438 data = 0;
1439 }
1440 break;
1441 }
1442 *pdata = data;
1443 return 0;
1444 }
1445 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1446
1447 /*
1448 * Read or write a bunch of msrs. All parameters are kernel addresses.
1449 *
1450 * @return number of msrs set successfully.
1451 */
1452 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1453 struct kvm_msr_entry *entries,
1454 int (*do_msr)(struct kvm_vcpu *vcpu,
1455 unsigned index, u64 *data))
1456 {
1457 int i, idx;
1458
1459 vcpu_load(vcpu);
1460
1461 idx = srcu_read_lock(&vcpu->kvm->srcu);
1462 for (i = 0; i < msrs->nmsrs; ++i)
1463 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1464 break;
1465 srcu_read_unlock(&vcpu->kvm->srcu, idx);
1466
1467 vcpu_put(vcpu);
1468
1469 return i;
1470 }
1471
1472 /*
1473 * Read or write a bunch of msrs. Parameters are user addresses.
1474 *
1475 * @return number of msrs set successfully.
1476 */
1477 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1478 int (*do_msr)(struct kvm_vcpu *vcpu,
1479 unsigned index, u64 *data),
1480 int writeback)
1481 {
1482 struct kvm_msrs msrs;
1483 struct kvm_msr_entry *entries;
1484 int r, n;
1485 unsigned size;
1486
1487 r = -EFAULT;
1488 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1489 goto out;
1490
1491 r = -E2BIG;
1492 if (msrs.nmsrs >= MAX_IO_MSRS)
1493 goto out;
1494
1495 r = -ENOMEM;
1496 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1497 entries = vmalloc(size);
1498 if (!entries)
1499 goto out;
1500
1501 r = -EFAULT;
1502 if (copy_from_user(entries, user_msrs->entries, size))
1503 goto out_free;
1504
1505 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1506 if (r < 0)
1507 goto out_free;
1508
1509 r = -EFAULT;
1510 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1511 goto out_free;
1512
1513 r = n;
1514
1515 out_free:
1516 vfree(entries);
1517 out:
1518 return r;
1519 }
1520
1521 int kvm_dev_ioctl_check_extension(long ext)
1522 {
1523 int r;
1524
1525 switch (ext) {
1526 case KVM_CAP_IRQCHIP:
1527 case KVM_CAP_HLT:
1528 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1529 case KVM_CAP_SET_TSS_ADDR:
1530 case KVM_CAP_EXT_CPUID:
1531 case KVM_CAP_CLOCKSOURCE:
1532 case KVM_CAP_PIT:
1533 case KVM_CAP_NOP_IO_DELAY:
1534 case KVM_CAP_MP_STATE:
1535 case KVM_CAP_SYNC_MMU:
1536 case KVM_CAP_REINJECT_CONTROL:
1537 case KVM_CAP_IRQ_INJECT_STATUS:
1538 case KVM_CAP_ASSIGN_DEV_IRQ:
1539 case KVM_CAP_IRQFD:
1540 case KVM_CAP_IOEVENTFD:
1541 case KVM_CAP_PIT2:
1542 case KVM_CAP_PIT_STATE2:
1543 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1544 case KVM_CAP_XEN_HVM:
1545 case KVM_CAP_ADJUST_CLOCK:
1546 case KVM_CAP_VCPU_EVENTS:
1547 case KVM_CAP_HYPERV:
1548 case KVM_CAP_HYPERV_VAPIC:
1549 case KVM_CAP_HYPERV_SPIN:
1550 case KVM_CAP_PCI_SEGMENT:
1551 case KVM_CAP_X86_ROBUST_SINGLESTEP:
1552 r = 1;
1553 break;
1554 case KVM_CAP_COALESCED_MMIO:
1555 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1556 break;
1557 case KVM_CAP_VAPIC:
1558 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1559 break;
1560 case KVM_CAP_NR_VCPUS:
1561 r = KVM_MAX_VCPUS;
1562 break;
1563 case KVM_CAP_NR_MEMSLOTS:
1564 r = KVM_MEMORY_SLOTS;
1565 break;
1566 case KVM_CAP_PV_MMU: /* obsolete */
1567 r = 0;
1568 break;
1569 case KVM_CAP_IOMMU:
1570 r = iommu_found();
1571 break;
1572 case KVM_CAP_MCE:
1573 r = KVM_MAX_MCE_BANKS;
1574 break;
1575 default:
1576 r = 0;
1577 break;
1578 }
1579 return r;
1580
1581 }
1582
1583 long kvm_arch_dev_ioctl(struct file *filp,
1584 unsigned int ioctl, unsigned long arg)
1585 {
1586 void __user *argp = (void __user *)arg;
1587 long r;
1588
1589 switch (ioctl) {
1590 case KVM_GET_MSR_INDEX_LIST: {
1591 struct kvm_msr_list __user *user_msr_list = argp;
1592 struct kvm_msr_list msr_list;
1593 unsigned n;
1594
1595 r = -EFAULT;
1596 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1597 goto out;
1598 n = msr_list.nmsrs;
1599 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1600 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1601 goto out;
1602 r = -E2BIG;
1603 if (n < msr_list.nmsrs)
1604 goto out;
1605 r = -EFAULT;
1606 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1607 num_msrs_to_save * sizeof(u32)))
1608 goto out;
1609 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1610 &emulated_msrs,
1611 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1612 goto out;
1613 r = 0;
1614 break;
1615 }
1616 case KVM_GET_SUPPORTED_CPUID: {
1617 struct kvm_cpuid2 __user *cpuid_arg = argp;
1618 struct kvm_cpuid2 cpuid;
1619
1620 r = -EFAULT;
1621 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1622 goto out;
1623 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1624 cpuid_arg->entries);
1625 if (r)
1626 goto out;
1627
1628 r = -EFAULT;
1629 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1630 goto out;
1631 r = 0;
1632 break;
1633 }
1634 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1635 u64 mce_cap;
1636
1637 mce_cap = KVM_MCE_CAP_SUPPORTED;
1638 r = -EFAULT;
1639 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1640 goto out;
1641 r = 0;
1642 break;
1643 }
1644 default:
1645 r = -EINVAL;
1646 }
1647 out:
1648 return r;
1649 }
1650
1651 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1652 {
1653 kvm_x86_ops->vcpu_load(vcpu, cpu);
1654 if (unlikely(per_cpu(cpu_tsc_khz, cpu) == 0)) {
1655 unsigned long khz = cpufreq_quick_get(cpu);
1656 if (!khz)
1657 khz = tsc_khz;
1658 per_cpu(cpu_tsc_khz, cpu) = khz;
1659 }
1660 kvm_request_guest_time_update(vcpu);
1661 }
1662
1663 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1664 {
1665 kvm_put_guest_fpu(vcpu);
1666 kvm_x86_ops->vcpu_put(vcpu);
1667 }
1668
1669 static int is_efer_nx(void)
1670 {
1671 unsigned long long efer = 0;
1672
1673 rdmsrl_safe(MSR_EFER, &efer);
1674 return efer & EFER_NX;
1675 }
1676
1677 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1678 {
1679 int i;
1680 struct kvm_cpuid_entry2 *e, *entry;
1681
1682 entry = NULL;
1683 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1684 e = &vcpu->arch.cpuid_entries[i];
1685 if (e->function == 0x80000001) {
1686 entry = e;
1687 break;
1688 }
1689 }
1690 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1691 entry->edx &= ~(1 << 20);
1692 printk(KERN_INFO "kvm: guest NX capability removed\n");
1693 }
1694 }
1695
1696 /* when an old userspace process fills a new kernel module */
1697 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1698 struct kvm_cpuid *cpuid,
1699 struct kvm_cpuid_entry __user *entries)
1700 {
1701 int r, i;
1702 struct kvm_cpuid_entry *cpuid_entries;
1703
1704 r = -E2BIG;
1705 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1706 goto out;
1707 r = -ENOMEM;
1708 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1709 if (!cpuid_entries)
1710 goto out;
1711 r = -EFAULT;
1712 if (copy_from_user(cpuid_entries, entries,
1713 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1714 goto out_free;
1715 for (i = 0; i < cpuid->nent; i++) {
1716 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1717 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1718 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1719 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1720 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1721 vcpu->arch.cpuid_entries[i].index = 0;
1722 vcpu->arch.cpuid_entries[i].flags = 0;
1723 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1724 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1725 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1726 }
1727 vcpu->arch.cpuid_nent = cpuid->nent;
1728 cpuid_fix_nx_cap(vcpu);
1729 r = 0;
1730 kvm_apic_set_version(vcpu);
1731 kvm_x86_ops->cpuid_update(vcpu);
1732
1733 out_free:
1734 vfree(cpuid_entries);
1735 out:
1736 return r;
1737 }
1738
1739 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1740 struct kvm_cpuid2 *cpuid,
1741 struct kvm_cpuid_entry2 __user *entries)
1742 {
1743 int r;
1744
1745 r = -E2BIG;
1746 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1747 goto out;
1748 r = -EFAULT;
1749 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1750 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1751 goto out;
1752 vcpu->arch.cpuid_nent = cpuid->nent;
1753 kvm_apic_set_version(vcpu);
1754 kvm_x86_ops->cpuid_update(vcpu);
1755 return 0;
1756
1757 out:
1758 return r;
1759 }
1760
1761 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1762 struct kvm_cpuid2 *cpuid,
1763 struct kvm_cpuid_entry2 __user *entries)
1764 {
1765 int r;
1766
1767 r = -E2BIG;
1768 if (cpuid->nent < vcpu->arch.cpuid_nent)
1769 goto out;
1770 r = -EFAULT;
1771 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1772 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1773 goto out;
1774 return 0;
1775
1776 out:
1777 cpuid->nent = vcpu->arch.cpuid_nent;
1778 return r;
1779 }
1780
1781 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1782 u32 index)
1783 {
1784 entry->function = function;
1785 entry->index = index;
1786 cpuid_count(entry->function, entry->index,
1787 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1788 entry->flags = 0;
1789 }
1790
1791 #define F(x) bit(X86_FEATURE_##x)
1792
1793 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1794 u32 index, int *nent, int maxnent)
1795 {
1796 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1797 #ifdef CONFIG_X86_64
1798 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
1799 ? F(GBPAGES) : 0;
1800 unsigned f_lm = F(LM);
1801 #else
1802 unsigned f_gbpages = 0;
1803 unsigned f_lm = 0;
1804 #endif
1805 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
1806
1807 /* cpuid 1.edx */
1808 const u32 kvm_supported_word0_x86_features =
1809 F(FPU) | F(VME) | F(DE) | F(PSE) |
1810 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1811 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1812 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1813 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1814 0 /* Reserved, DS, ACPI */ | F(MMX) |
1815 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1816 0 /* HTT, TM, Reserved, PBE */;
1817 /* cpuid 0x80000001.edx */
1818 const u32 kvm_supported_word1_x86_features =
1819 F(FPU) | F(VME) | F(DE) | F(PSE) |
1820 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1821 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1822 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1823 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1824 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1825 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
1826 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1827 /* cpuid 1.ecx */
1828 const u32 kvm_supported_word4_x86_features =
1829 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1830 0 /* DS-CPL, VMX, SMX, EST */ |
1831 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1832 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1833 0 /* Reserved, DCA */ | F(XMM4_1) |
1834 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1835 0 /* Reserved, XSAVE, OSXSAVE */;
1836 /* cpuid 0x80000001.ecx */
1837 const u32 kvm_supported_word6_x86_features =
1838 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1839 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1840 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1841 0 /* SKINIT */ | 0 /* WDT */;
1842
1843 /* all calls to cpuid_count() should be made on the same cpu */
1844 get_cpu();
1845 do_cpuid_1_ent(entry, function, index);
1846 ++*nent;
1847
1848 switch (function) {
1849 case 0:
1850 entry->eax = min(entry->eax, (u32)0xb);
1851 break;
1852 case 1:
1853 entry->edx &= kvm_supported_word0_x86_features;
1854 entry->ecx &= kvm_supported_word4_x86_features;
1855 /* we support x2apic emulation even if host does not support
1856 * it since we emulate x2apic in software */
1857 entry->ecx |= F(X2APIC);
1858 break;
1859 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1860 * may return different values. This forces us to get_cpu() before
1861 * issuing the first command, and also to emulate this annoying behavior
1862 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1863 case 2: {
1864 int t, times = entry->eax & 0xff;
1865
1866 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1867 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1868 for (t = 1; t < times && *nent < maxnent; ++t) {
1869 do_cpuid_1_ent(&entry[t], function, 0);
1870 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1871 ++*nent;
1872 }
1873 break;
1874 }
1875 /* function 4 and 0xb have additional index. */
1876 case 4: {
1877 int i, cache_type;
1878
1879 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1880 /* read more entries until cache_type is zero */
1881 for (i = 1; *nent < maxnent; ++i) {
1882 cache_type = entry[i - 1].eax & 0x1f;
1883 if (!cache_type)
1884 break;
1885 do_cpuid_1_ent(&entry[i], function, i);
1886 entry[i].flags |=
1887 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1888 ++*nent;
1889 }
1890 break;
1891 }
1892 case 0xb: {
1893 int i, level_type;
1894
1895 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1896 /* read more entries until level_type is zero */
1897 for (i = 1; *nent < maxnent; ++i) {
1898 level_type = entry[i - 1].ecx & 0xff00;
1899 if (!level_type)
1900 break;
1901 do_cpuid_1_ent(&entry[i], function, i);
1902 entry[i].flags |=
1903 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1904 ++*nent;
1905 }
1906 break;
1907 }
1908 case 0x80000000:
1909 entry->eax = min(entry->eax, 0x8000001a);
1910 break;
1911 case 0x80000001:
1912 entry->edx &= kvm_supported_word1_x86_features;
1913 entry->ecx &= kvm_supported_word6_x86_features;
1914 break;
1915 }
1916 put_cpu();
1917 }
1918
1919 #undef F
1920
1921 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1922 struct kvm_cpuid_entry2 __user *entries)
1923 {
1924 struct kvm_cpuid_entry2 *cpuid_entries;
1925 int limit, nent = 0, r = -E2BIG;
1926 u32 func;
1927
1928 if (cpuid->nent < 1)
1929 goto out;
1930 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1931 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1932 r = -ENOMEM;
1933 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1934 if (!cpuid_entries)
1935 goto out;
1936
1937 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1938 limit = cpuid_entries[0].eax;
1939 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1940 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1941 &nent, cpuid->nent);
1942 r = -E2BIG;
1943 if (nent >= cpuid->nent)
1944 goto out_free;
1945
1946 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1947 limit = cpuid_entries[nent - 1].eax;
1948 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1949 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1950 &nent, cpuid->nent);
1951 r = -E2BIG;
1952 if (nent >= cpuid->nent)
1953 goto out_free;
1954
1955 r = -EFAULT;
1956 if (copy_to_user(entries, cpuid_entries,
1957 nent * sizeof(struct kvm_cpuid_entry2)))
1958 goto out_free;
1959 cpuid->nent = nent;
1960 r = 0;
1961
1962 out_free:
1963 vfree(cpuid_entries);
1964 out:
1965 return r;
1966 }
1967
1968 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1969 struct kvm_lapic_state *s)
1970 {
1971 vcpu_load(vcpu);
1972 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1973 vcpu_put(vcpu);
1974
1975 return 0;
1976 }
1977
1978 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1979 struct kvm_lapic_state *s)
1980 {
1981 vcpu_load(vcpu);
1982 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1983 kvm_apic_post_state_restore(vcpu);
1984 update_cr8_intercept(vcpu);
1985 vcpu_put(vcpu);
1986
1987 return 0;
1988 }
1989
1990 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1991 struct kvm_interrupt *irq)
1992 {
1993 if (irq->irq < 0 || irq->irq >= 256)
1994 return -EINVAL;
1995 if (irqchip_in_kernel(vcpu->kvm))
1996 return -ENXIO;
1997 vcpu_load(vcpu);
1998
1999 kvm_queue_interrupt(vcpu, irq->irq, false);
2000
2001 vcpu_put(vcpu);
2002
2003 return 0;
2004 }
2005
2006 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
2007 {
2008 vcpu_load(vcpu);
2009 kvm_inject_nmi(vcpu);
2010 vcpu_put(vcpu);
2011
2012 return 0;
2013 }
2014
2015 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
2016 struct kvm_tpr_access_ctl *tac)
2017 {
2018 if (tac->flags)
2019 return -EINVAL;
2020 vcpu->arch.tpr_access_reporting = !!tac->enabled;
2021 return 0;
2022 }
2023
2024 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
2025 u64 mcg_cap)
2026 {
2027 int r;
2028 unsigned bank_num = mcg_cap & 0xff, bank;
2029
2030 r = -EINVAL;
2031 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
2032 goto out;
2033 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
2034 goto out;
2035 r = 0;
2036 vcpu->arch.mcg_cap = mcg_cap;
2037 /* Init IA32_MCG_CTL to all 1s */
2038 if (mcg_cap & MCG_CTL_P)
2039 vcpu->arch.mcg_ctl = ~(u64)0;
2040 /* Init IA32_MCi_CTL to all 1s */
2041 for (bank = 0; bank < bank_num; bank++)
2042 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
2043 out:
2044 return r;
2045 }
2046
2047 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
2048 struct kvm_x86_mce *mce)
2049 {
2050 u64 mcg_cap = vcpu->arch.mcg_cap;
2051 unsigned bank_num = mcg_cap & 0xff;
2052 u64 *banks = vcpu->arch.mce_banks;
2053
2054 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
2055 return -EINVAL;
2056 /*
2057 * if IA32_MCG_CTL is not all 1s, the uncorrected error
2058 * reporting is disabled
2059 */
2060 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
2061 vcpu->arch.mcg_ctl != ~(u64)0)
2062 return 0;
2063 banks += 4 * mce->bank;
2064 /*
2065 * if IA32_MCi_CTL is not all 1s, the uncorrected error
2066 * reporting is disabled for the bank
2067 */
2068 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
2069 return 0;
2070 if (mce->status & MCI_STATUS_UC) {
2071 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
2072 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
2073 printk(KERN_DEBUG "kvm: set_mce: "
2074 "injects mce exception while "
2075 "previous one is in progress!\n");
2076 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
2077 return 0;
2078 }
2079 if (banks[1] & MCI_STATUS_VAL)
2080 mce->status |= MCI_STATUS_OVER;
2081 banks[2] = mce->addr;
2082 banks[3] = mce->misc;
2083 vcpu->arch.mcg_status = mce->mcg_status;
2084 banks[1] = mce->status;
2085 kvm_queue_exception(vcpu, MC_VECTOR);
2086 } else if (!(banks[1] & MCI_STATUS_VAL)
2087 || !(banks[1] & MCI_STATUS_UC)) {
2088 if (banks[1] & MCI_STATUS_VAL)
2089 mce->status |= MCI_STATUS_OVER;
2090 banks[2] = mce->addr;
2091 banks[3] = mce->misc;
2092 banks[1] = mce->status;
2093 } else
2094 banks[1] |= MCI_STATUS_OVER;
2095 return 0;
2096 }
2097
2098 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
2099 struct kvm_vcpu_events *events)
2100 {
2101 vcpu_load(vcpu);
2102
2103 events->exception.injected =
2104 vcpu->arch.exception.pending &&
2105 !kvm_exception_is_soft(vcpu->arch.exception.nr);
2106 events->exception.nr = vcpu->arch.exception.nr;
2107 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
2108 events->exception.error_code = vcpu->arch.exception.error_code;
2109
2110 events->interrupt.injected =
2111 vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft;
2112 events->interrupt.nr = vcpu->arch.interrupt.nr;
2113 events->interrupt.soft = 0;
2114 events->interrupt.shadow =
2115 kvm_x86_ops->get_interrupt_shadow(vcpu,
2116 KVM_X86_SHADOW_INT_MOV_SS | KVM_X86_SHADOW_INT_STI);
2117
2118 events->nmi.injected = vcpu->arch.nmi_injected;
2119 events->nmi.pending = vcpu->arch.nmi_pending;
2120 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
2121
2122 events->sipi_vector = vcpu->arch.sipi_vector;
2123
2124 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
2125 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2126 | KVM_VCPUEVENT_VALID_SHADOW);
2127
2128 vcpu_put(vcpu);
2129 }
2130
2131 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
2132 struct kvm_vcpu_events *events)
2133 {
2134 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
2135 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
2136 | KVM_VCPUEVENT_VALID_SHADOW))
2137 return -EINVAL;
2138
2139 vcpu_load(vcpu);
2140
2141 vcpu->arch.exception.pending = events->exception.injected;
2142 vcpu->arch.exception.nr = events->exception.nr;
2143 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
2144 vcpu->arch.exception.error_code = events->exception.error_code;
2145
2146 vcpu->arch.interrupt.pending = events->interrupt.injected;
2147 vcpu->arch.interrupt.nr = events->interrupt.nr;
2148 vcpu->arch.interrupt.soft = events->interrupt.soft;
2149 if (vcpu->arch.interrupt.pending && irqchip_in_kernel(vcpu->kvm))
2150 kvm_pic_clear_isr_ack(vcpu->kvm);
2151 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
2152 kvm_x86_ops->set_interrupt_shadow(vcpu,
2153 events->interrupt.shadow);
2154
2155 vcpu->arch.nmi_injected = events->nmi.injected;
2156 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
2157 vcpu->arch.nmi_pending = events->nmi.pending;
2158 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
2159
2160 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR)
2161 vcpu->arch.sipi_vector = events->sipi_vector;
2162
2163 vcpu_put(vcpu);
2164
2165 return 0;
2166 }
2167
2168 long kvm_arch_vcpu_ioctl(struct file *filp,
2169 unsigned int ioctl, unsigned long arg)
2170 {
2171 struct kvm_vcpu *vcpu = filp->private_data;
2172 void __user *argp = (void __user *)arg;
2173 int r;
2174 struct kvm_lapic_state *lapic = NULL;
2175
2176 switch (ioctl) {
2177 case KVM_GET_LAPIC: {
2178 r = -EINVAL;
2179 if (!vcpu->arch.apic)
2180 goto out;
2181 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2182
2183 r = -ENOMEM;
2184 if (!lapic)
2185 goto out;
2186 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
2187 if (r)
2188 goto out;
2189 r = -EFAULT;
2190 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
2191 goto out;
2192 r = 0;
2193 break;
2194 }
2195 case KVM_SET_LAPIC: {
2196 r = -EINVAL;
2197 if (!vcpu->arch.apic)
2198 goto out;
2199 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
2200 r = -ENOMEM;
2201 if (!lapic)
2202 goto out;
2203 r = -EFAULT;
2204 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
2205 goto out;
2206 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
2207 if (r)
2208 goto out;
2209 r = 0;
2210 break;
2211 }
2212 case KVM_INTERRUPT: {
2213 struct kvm_interrupt irq;
2214
2215 r = -EFAULT;
2216 if (copy_from_user(&irq, argp, sizeof irq))
2217 goto out;
2218 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
2219 if (r)
2220 goto out;
2221 r = 0;
2222 break;
2223 }
2224 case KVM_NMI: {
2225 r = kvm_vcpu_ioctl_nmi(vcpu);
2226 if (r)
2227 goto out;
2228 r = 0;
2229 break;
2230 }
2231 case KVM_SET_CPUID: {
2232 struct kvm_cpuid __user *cpuid_arg = argp;
2233 struct kvm_cpuid cpuid;
2234
2235 r = -EFAULT;
2236 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2237 goto out;
2238 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
2239 if (r)
2240 goto out;
2241 break;
2242 }
2243 case KVM_SET_CPUID2: {
2244 struct kvm_cpuid2 __user *cpuid_arg = argp;
2245 struct kvm_cpuid2 cpuid;
2246
2247 r = -EFAULT;
2248 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2249 goto out;
2250 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
2251 cpuid_arg->entries);
2252 if (r)
2253 goto out;
2254 break;
2255 }
2256 case KVM_GET_CPUID2: {
2257 struct kvm_cpuid2 __user *cpuid_arg = argp;
2258 struct kvm_cpuid2 cpuid;
2259
2260 r = -EFAULT;
2261 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
2262 goto out;
2263 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
2264 cpuid_arg->entries);
2265 if (r)
2266 goto out;
2267 r = -EFAULT;
2268 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
2269 goto out;
2270 r = 0;
2271 break;
2272 }
2273 case KVM_GET_MSRS:
2274 r = msr_io(vcpu, argp, kvm_get_msr, 1);
2275 break;
2276 case KVM_SET_MSRS:
2277 r = msr_io(vcpu, argp, do_set_msr, 0);
2278 break;
2279 case KVM_TPR_ACCESS_REPORTING: {
2280 struct kvm_tpr_access_ctl tac;
2281
2282 r = -EFAULT;
2283 if (copy_from_user(&tac, argp, sizeof tac))
2284 goto out;
2285 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
2286 if (r)
2287 goto out;
2288 r = -EFAULT;
2289 if (copy_to_user(argp, &tac, sizeof tac))
2290 goto out;
2291 r = 0;
2292 break;
2293 };
2294 case KVM_SET_VAPIC_ADDR: {
2295 struct kvm_vapic_addr va;
2296
2297 r = -EINVAL;
2298 if (!irqchip_in_kernel(vcpu->kvm))
2299 goto out;
2300 r = -EFAULT;
2301 if (copy_from_user(&va, argp, sizeof va))
2302 goto out;
2303 r = 0;
2304 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
2305 break;
2306 }
2307 case KVM_X86_SETUP_MCE: {
2308 u64 mcg_cap;
2309
2310 r = -EFAULT;
2311 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
2312 goto out;
2313 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
2314 break;
2315 }
2316 case KVM_X86_SET_MCE: {
2317 struct kvm_x86_mce mce;
2318
2319 r = -EFAULT;
2320 if (copy_from_user(&mce, argp, sizeof mce))
2321 goto out;
2322 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
2323 break;
2324 }
2325 case KVM_GET_VCPU_EVENTS: {
2326 struct kvm_vcpu_events events;
2327
2328 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
2329
2330 r = -EFAULT;
2331 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
2332 break;
2333 r = 0;
2334 break;
2335 }
2336 case KVM_SET_VCPU_EVENTS: {
2337 struct kvm_vcpu_events events;
2338
2339 r = -EFAULT;
2340 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
2341 break;
2342
2343 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
2344 break;
2345 }
2346 default:
2347 r = -EINVAL;
2348 }
2349 out:
2350 kfree(lapic);
2351 return r;
2352 }
2353
2354 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
2355 {
2356 int ret;
2357
2358 if (addr > (unsigned int)(-3 * PAGE_SIZE))
2359 return -1;
2360 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
2361 return ret;
2362 }
2363
2364 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
2365 u64 ident_addr)
2366 {
2367 kvm->arch.ept_identity_map_addr = ident_addr;
2368 return 0;
2369 }
2370
2371 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
2372 u32 kvm_nr_mmu_pages)
2373 {
2374 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
2375 return -EINVAL;
2376
2377 mutex_lock(&kvm->slots_lock);
2378 spin_lock(&kvm->mmu_lock);
2379
2380 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
2381 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
2382
2383 spin_unlock(&kvm->mmu_lock);
2384 mutex_unlock(&kvm->slots_lock);
2385 return 0;
2386 }
2387
2388 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
2389 {
2390 return kvm->arch.n_alloc_mmu_pages;
2391 }
2392
2393 gfn_t unalias_gfn_instantiation(struct kvm *kvm, gfn_t gfn)
2394 {
2395 int i;
2396 struct kvm_mem_alias *alias;
2397 struct kvm_mem_aliases *aliases;
2398
2399 aliases = rcu_dereference(kvm->arch.aliases);
2400
2401 for (i = 0; i < aliases->naliases; ++i) {
2402 alias = &aliases->aliases[i];
2403 if (alias->flags & KVM_ALIAS_INVALID)
2404 continue;
2405 if (gfn >= alias->base_gfn
2406 && gfn < alias->base_gfn + alias->npages)
2407 return alias->target_gfn + gfn - alias->base_gfn;
2408 }
2409 return gfn;
2410 }
2411
2412 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
2413 {
2414 int i;
2415 struct kvm_mem_alias *alias;
2416 struct kvm_mem_aliases *aliases;
2417
2418 aliases = rcu_dereference(kvm->arch.aliases);
2419
2420 for (i = 0; i < aliases->naliases; ++i) {
2421 alias = &aliases->aliases[i];
2422 if (gfn >= alias->base_gfn
2423 && gfn < alias->base_gfn + alias->npages)
2424 return alias->target_gfn + gfn - alias->base_gfn;
2425 }
2426 return gfn;
2427 }
2428
2429 /*
2430 * Set a new alias region. Aliases map a portion of physical memory into
2431 * another portion. This is useful for memory windows, for example the PC
2432 * VGA region.
2433 */
2434 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
2435 struct kvm_memory_alias *alias)
2436 {
2437 int r, n;
2438 struct kvm_mem_alias *p;
2439 struct kvm_mem_aliases *aliases, *old_aliases;
2440
2441 r = -EINVAL;
2442 /* General sanity checks */
2443 if (alias->memory_size & (PAGE_SIZE - 1))
2444 goto out;
2445 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
2446 goto out;
2447 if (alias->slot >= KVM_ALIAS_SLOTS)
2448 goto out;
2449 if (alias->guest_phys_addr + alias->memory_size
2450 < alias->guest_phys_addr)
2451 goto out;
2452 if (alias->target_phys_addr + alias->memory_size
2453 < alias->target_phys_addr)
2454 goto out;
2455
2456 r = -ENOMEM;
2457 aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2458 if (!aliases)
2459 goto out;
2460
2461 mutex_lock(&kvm->slots_lock);
2462
2463 /* invalidate any gfn reference in case of deletion/shrinking */
2464 memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2465 aliases->aliases[alias->slot].flags |= KVM_ALIAS_INVALID;
2466 old_aliases = kvm->arch.aliases;
2467 rcu_assign_pointer(kvm->arch.aliases, aliases);
2468 synchronize_srcu_expedited(&kvm->srcu);
2469 kvm_mmu_zap_all(kvm);
2470 kfree(old_aliases);
2471
2472 r = -ENOMEM;
2473 aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
2474 if (!aliases)
2475 goto out_unlock;
2476
2477 memcpy(aliases, kvm->arch.aliases, sizeof(struct kvm_mem_aliases));
2478
2479 p = &aliases->aliases[alias->slot];
2480 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2481 p->npages = alias->memory_size >> PAGE_SHIFT;
2482 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2483 p->flags &= ~(KVM_ALIAS_INVALID);
2484
2485 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2486 if (aliases->aliases[n - 1].npages)
2487 break;
2488 aliases->naliases = n;
2489
2490 old_aliases = kvm->arch.aliases;
2491 rcu_assign_pointer(kvm->arch.aliases, aliases);
2492 synchronize_srcu_expedited(&kvm->srcu);
2493 kfree(old_aliases);
2494 r = 0;
2495
2496 out_unlock:
2497 mutex_unlock(&kvm->slots_lock);
2498 out:
2499 return r;
2500 }
2501
2502 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2503 {
2504 int r;
2505
2506 r = 0;
2507 switch (chip->chip_id) {
2508 case KVM_IRQCHIP_PIC_MASTER:
2509 memcpy(&chip->chip.pic,
2510 &pic_irqchip(kvm)->pics[0],
2511 sizeof(struct kvm_pic_state));
2512 break;
2513 case KVM_IRQCHIP_PIC_SLAVE:
2514 memcpy(&chip->chip.pic,
2515 &pic_irqchip(kvm)->pics[1],
2516 sizeof(struct kvm_pic_state));
2517 break;
2518 case KVM_IRQCHIP_IOAPIC:
2519 r = kvm_get_ioapic(kvm, &chip->chip.ioapic);
2520 break;
2521 default:
2522 r = -EINVAL;
2523 break;
2524 }
2525 return r;
2526 }
2527
2528 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2529 {
2530 int r;
2531
2532 r = 0;
2533 switch (chip->chip_id) {
2534 case KVM_IRQCHIP_PIC_MASTER:
2535 raw_spin_lock(&pic_irqchip(kvm)->lock);
2536 memcpy(&pic_irqchip(kvm)->pics[0],
2537 &chip->chip.pic,
2538 sizeof(struct kvm_pic_state));
2539 raw_spin_unlock(&pic_irqchip(kvm)->lock);
2540 break;
2541 case KVM_IRQCHIP_PIC_SLAVE:
2542 raw_spin_lock(&pic_irqchip(kvm)->lock);
2543 memcpy(&pic_irqchip(kvm)->pics[1],
2544 &chip->chip.pic,
2545 sizeof(struct kvm_pic_state));
2546 raw_spin_unlock(&pic_irqchip(kvm)->lock);
2547 break;
2548 case KVM_IRQCHIP_IOAPIC:
2549 r = kvm_set_ioapic(kvm, &chip->chip.ioapic);
2550 break;
2551 default:
2552 r = -EINVAL;
2553 break;
2554 }
2555 kvm_pic_update_irq(pic_irqchip(kvm));
2556 return r;
2557 }
2558
2559 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2560 {
2561 int r = 0;
2562
2563 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2564 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2565 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2566 return r;
2567 }
2568
2569 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2570 {
2571 int r = 0;
2572
2573 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2574 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2575 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2576 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2577 return r;
2578 }
2579
2580 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2581 {
2582 int r = 0;
2583
2584 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2585 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2586 sizeof(ps->channels));
2587 ps->flags = kvm->arch.vpit->pit_state.flags;
2588 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2589 return r;
2590 }
2591
2592 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2593 {
2594 int r = 0, start = 0;
2595 u32 prev_legacy, cur_legacy;
2596 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2597 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2598 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2599 if (!prev_legacy && cur_legacy)
2600 start = 1;
2601 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2602 sizeof(kvm->arch.vpit->pit_state.channels));
2603 kvm->arch.vpit->pit_state.flags = ps->flags;
2604 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2605 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2606 return r;
2607 }
2608
2609 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2610 struct kvm_reinject_control *control)
2611 {
2612 if (!kvm->arch.vpit)
2613 return -ENXIO;
2614 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2615 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2616 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2617 return 0;
2618 }
2619
2620 /*
2621 * Get (and clear) the dirty memory log for a memory slot.
2622 */
2623 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2624 struct kvm_dirty_log *log)
2625 {
2626 int r, i;
2627 struct kvm_memory_slot *memslot;
2628 unsigned long n;
2629 unsigned long is_dirty = 0;
2630 unsigned long *dirty_bitmap = NULL;
2631
2632 mutex_lock(&kvm->slots_lock);
2633
2634 r = -EINVAL;
2635 if (log->slot >= KVM_MEMORY_SLOTS)
2636 goto out;
2637
2638 memslot = &kvm->memslots->memslots[log->slot];
2639 r = -ENOENT;
2640 if (!memslot->dirty_bitmap)
2641 goto out;
2642
2643 n = kvm_dirty_bitmap_bytes(memslot);
2644
2645 r = -ENOMEM;
2646 dirty_bitmap = vmalloc(n);
2647 if (!dirty_bitmap)
2648 goto out;
2649 memset(dirty_bitmap, 0, n);
2650
2651 for (i = 0; !is_dirty && i < n/sizeof(long); i++)
2652 is_dirty = memslot->dirty_bitmap[i];
2653
2654 /* If nothing is dirty, don't bother messing with page tables. */
2655 if (is_dirty) {
2656 struct kvm_memslots *slots, *old_slots;
2657
2658 spin_lock(&kvm->mmu_lock);
2659 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2660 spin_unlock(&kvm->mmu_lock);
2661
2662 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
2663 if (!slots)
2664 goto out_free;
2665
2666 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
2667 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap;
2668
2669 old_slots = kvm->memslots;
2670 rcu_assign_pointer(kvm->memslots, slots);
2671 synchronize_srcu_expedited(&kvm->srcu);
2672 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
2673 kfree(old_slots);
2674 }
2675
2676 r = 0;
2677 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
2678 r = -EFAULT;
2679 out_free:
2680 vfree(dirty_bitmap);
2681 out:
2682 mutex_unlock(&kvm->slots_lock);
2683 return r;
2684 }
2685
2686 long kvm_arch_vm_ioctl(struct file *filp,
2687 unsigned int ioctl, unsigned long arg)
2688 {
2689 struct kvm *kvm = filp->private_data;
2690 void __user *argp = (void __user *)arg;
2691 int r = -ENOTTY;
2692 /*
2693 * This union makes it completely explicit to gcc-3.x
2694 * that these two variables' stack usage should be
2695 * combined, not added together.
2696 */
2697 union {
2698 struct kvm_pit_state ps;
2699 struct kvm_pit_state2 ps2;
2700 struct kvm_memory_alias alias;
2701 struct kvm_pit_config pit_config;
2702 } u;
2703
2704 switch (ioctl) {
2705 case KVM_SET_TSS_ADDR:
2706 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2707 if (r < 0)
2708 goto out;
2709 break;
2710 case KVM_SET_IDENTITY_MAP_ADDR: {
2711 u64 ident_addr;
2712
2713 r = -EFAULT;
2714 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2715 goto out;
2716 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2717 if (r < 0)
2718 goto out;
2719 break;
2720 }
2721 case KVM_SET_MEMORY_REGION: {
2722 struct kvm_memory_region kvm_mem;
2723 struct kvm_userspace_memory_region kvm_userspace_mem;
2724
2725 r = -EFAULT;
2726 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2727 goto out;
2728 kvm_userspace_mem.slot = kvm_mem.slot;
2729 kvm_userspace_mem.flags = kvm_mem.flags;
2730 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2731 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2732 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2733 if (r)
2734 goto out;
2735 break;
2736 }
2737 case KVM_SET_NR_MMU_PAGES:
2738 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2739 if (r)
2740 goto out;
2741 break;
2742 case KVM_GET_NR_MMU_PAGES:
2743 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2744 break;
2745 case KVM_SET_MEMORY_ALIAS:
2746 r = -EFAULT;
2747 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2748 goto out;
2749 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2750 if (r)
2751 goto out;
2752 break;
2753 case KVM_CREATE_IRQCHIP: {
2754 struct kvm_pic *vpic;
2755
2756 mutex_lock(&kvm->lock);
2757 r = -EEXIST;
2758 if (kvm->arch.vpic)
2759 goto create_irqchip_unlock;
2760 r = -ENOMEM;
2761 vpic = kvm_create_pic(kvm);
2762 if (vpic) {
2763 r = kvm_ioapic_init(kvm);
2764 if (r) {
2765 kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS,
2766 &vpic->dev);
2767 kfree(vpic);
2768 goto create_irqchip_unlock;
2769 }
2770 } else
2771 goto create_irqchip_unlock;
2772 smp_wmb();
2773 kvm->arch.vpic = vpic;
2774 smp_wmb();
2775 r = kvm_setup_default_irq_routing(kvm);
2776 if (r) {
2777 mutex_lock(&kvm->irq_lock);
2778 kvm_ioapic_destroy(kvm);
2779 kvm_destroy_pic(kvm);
2780 mutex_unlock(&kvm->irq_lock);
2781 }
2782 create_irqchip_unlock:
2783 mutex_unlock(&kvm->lock);
2784 break;
2785 }
2786 case KVM_CREATE_PIT:
2787 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2788 goto create_pit;
2789 case KVM_CREATE_PIT2:
2790 r = -EFAULT;
2791 if (copy_from_user(&u.pit_config, argp,
2792 sizeof(struct kvm_pit_config)))
2793 goto out;
2794 create_pit:
2795 mutex_lock(&kvm->slots_lock);
2796 r = -EEXIST;
2797 if (kvm->arch.vpit)
2798 goto create_pit_unlock;
2799 r = -ENOMEM;
2800 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2801 if (kvm->arch.vpit)
2802 r = 0;
2803 create_pit_unlock:
2804 mutex_unlock(&kvm->slots_lock);
2805 break;
2806 case KVM_IRQ_LINE_STATUS:
2807 case KVM_IRQ_LINE: {
2808 struct kvm_irq_level irq_event;
2809
2810 r = -EFAULT;
2811 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2812 goto out;
2813 if (irqchip_in_kernel(kvm)) {
2814 __s32 status;
2815 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2816 irq_event.irq, irq_event.level);
2817 if (ioctl == KVM_IRQ_LINE_STATUS) {
2818 irq_event.status = status;
2819 if (copy_to_user(argp, &irq_event,
2820 sizeof irq_event))
2821 goto out;
2822 }
2823 r = 0;
2824 }
2825 break;
2826 }
2827 case KVM_GET_IRQCHIP: {
2828 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2829 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2830
2831 r = -ENOMEM;
2832 if (!chip)
2833 goto out;
2834 r = -EFAULT;
2835 if (copy_from_user(chip, argp, sizeof *chip))
2836 goto get_irqchip_out;
2837 r = -ENXIO;
2838 if (!irqchip_in_kernel(kvm))
2839 goto get_irqchip_out;
2840 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2841 if (r)
2842 goto get_irqchip_out;
2843 r = -EFAULT;
2844 if (copy_to_user(argp, chip, sizeof *chip))
2845 goto get_irqchip_out;
2846 r = 0;
2847 get_irqchip_out:
2848 kfree(chip);
2849 if (r)
2850 goto out;
2851 break;
2852 }
2853 case KVM_SET_IRQCHIP: {
2854 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2855 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2856
2857 r = -ENOMEM;
2858 if (!chip)
2859 goto out;
2860 r = -EFAULT;
2861 if (copy_from_user(chip, argp, sizeof *chip))
2862 goto set_irqchip_out;
2863 r = -ENXIO;
2864 if (!irqchip_in_kernel(kvm))
2865 goto set_irqchip_out;
2866 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2867 if (r)
2868 goto set_irqchip_out;
2869 r = 0;
2870 set_irqchip_out:
2871 kfree(chip);
2872 if (r)
2873 goto out;
2874 break;
2875 }
2876 case KVM_GET_PIT: {
2877 r = -EFAULT;
2878 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2879 goto out;
2880 r = -ENXIO;
2881 if (!kvm->arch.vpit)
2882 goto out;
2883 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2884 if (r)
2885 goto out;
2886 r = -EFAULT;
2887 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2888 goto out;
2889 r = 0;
2890 break;
2891 }
2892 case KVM_SET_PIT: {
2893 r = -EFAULT;
2894 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2895 goto out;
2896 r = -ENXIO;
2897 if (!kvm->arch.vpit)
2898 goto out;
2899 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2900 if (r)
2901 goto out;
2902 r = 0;
2903 break;
2904 }
2905 case KVM_GET_PIT2: {
2906 r = -ENXIO;
2907 if (!kvm->arch.vpit)
2908 goto out;
2909 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
2910 if (r)
2911 goto out;
2912 r = -EFAULT;
2913 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
2914 goto out;
2915 r = 0;
2916 break;
2917 }
2918 case KVM_SET_PIT2: {
2919 r = -EFAULT;
2920 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
2921 goto out;
2922 r = -ENXIO;
2923 if (!kvm->arch.vpit)
2924 goto out;
2925 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
2926 if (r)
2927 goto out;
2928 r = 0;
2929 break;
2930 }
2931 case KVM_REINJECT_CONTROL: {
2932 struct kvm_reinject_control control;
2933 r = -EFAULT;
2934 if (copy_from_user(&control, argp, sizeof(control)))
2935 goto out;
2936 r = kvm_vm_ioctl_reinject(kvm, &control);
2937 if (r)
2938 goto out;
2939 r = 0;
2940 break;
2941 }
2942 case KVM_XEN_HVM_CONFIG: {
2943 r = -EFAULT;
2944 if (copy_from_user(&kvm->arch.xen_hvm_config, argp,
2945 sizeof(struct kvm_xen_hvm_config)))
2946 goto out;
2947 r = -EINVAL;
2948 if (kvm->arch.xen_hvm_config.flags)
2949 goto out;
2950 r = 0;
2951 break;
2952 }
2953 case KVM_SET_CLOCK: {
2954 struct timespec now;
2955 struct kvm_clock_data user_ns;
2956 u64 now_ns;
2957 s64 delta;
2958
2959 r = -EFAULT;
2960 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
2961 goto out;
2962
2963 r = -EINVAL;
2964 if (user_ns.flags)
2965 goto out;
2966
2967 r = 0;
2968 ktime_get_ts(&now);
2969 now_ns = timespec_to_ns(&now);
2970 delta = user_ns.clock - now_ns;
2971 kvm->arch.kvmclock_offset = delta;
2972 break;
2973 }
2974 case KVM_GET_CLOCK: {
2975 struct timespec now;
2976 struct kvm_clock_data user_ns;
2977 u64 now_ns;
2978
2979 ktime_get_ts(&now);
2980 now_ns = timespec_to_ns(&now);
2981 user_ns.clock = kvm->arch.kvmclock_offset + now_ns;
2982 user_ns.flags = 0;
2983
2984 r = -EFAULT;
2985 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
2986 goto out;
2987 r = 0;
2988 break;
2989 }
2990
2991 default:
2992 ;
2993 }
2994 out:
2995 return r;
2996 }
2997
2998 static void kvm_init_msr_list(void)
2999 {
3000 u32 dummy[2];
3001 unsigned i, j;
3002
3003 /* skip the first msrs in the list. KVM-specific */
3004 for (i = j = KVM_SAVE_MSRS_BEGIN; i < ARRAY_SIZE(msrs_to_save); i++) {
3005 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
3006 continue;
3007 if (j < i)
3008 msrs_to_save[j] = msrs_to_save[i];
3009 j++;
3010 }
3011 num_msrs_to_save = j;
3012 }
3013
3014 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
3015 const void *v)
3016 {
3017 if (vcpu->arch.apic &&
3018 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
3019 return 0;
3020
3021 return kvm_io_bus_write(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3022 }
3023
3024 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
3025 {
3026 if (vcpu->arch.apic &&
3027 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
3028 return 0;
3029
3030 return kvm_io_bus_read(vcpu->kvm, KVM_MMIO_BUS, addr, len, v);
3031 }
3032
3033 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3034 {
3035 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3036 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3037 }
3038
3039 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3040 {
3041 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3042 access |= PFERR_FETCH_MASK;
3043 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3044 }
3045
3046 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3047 {
3048 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3049 access |= PFERR_WRITE_MASK;
3050 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, access, error);
3051 }
3052
3053 /* uses this to access any guest's mapped memory without checking CPL */
3054 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva, u32 *error)
3055 {
3056 return vcpu->arch.mmu.gva_to_gpa(vcpu, gva, 0, error);
3057 }
3058
3059 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
3060 struct kvm_vcpu *vcpu, u32 access,
3061 u32 *error)
3062 {
3063 void *data = val;
3064 int r = X86EMUL_CONTINUE;
3065
3066 while (bytes) {
3067 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr, access, error);
3068 unsigned offset = addr & (PAGE_SIZE-1);
3069 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
3070 int ret;
3071
3072 if (gpa == UNMAPPED_GVA) {
3073 r = X86EMUL_PROPAGATE_FAULT;
3074 goto out;
3075 }
3076 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
3077 if (ret < 0) {
3078 r = X86EMUL_UNHANDLEABLE;
3079 goto out;
3080 }
3081
3082 bytes -= toread;
3083 data += toread;
3084 addr += toread;
3085 }
3086 out:
3087 return r;
3088 }
3089
3090 /* used for instruction fetching */
3091 static int kvm_fetch_guest_virt(gva_t addr, void *val, unsigned int bytes,
3092 struct kvm_vcpu *vcpu, u32 *error)
3093 {
3094 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3095 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu,
3096 access | PFERR_FETCH_MASK, error);
3097 }
3098
3099 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
3100 struct kvm_vcpu *vcpu, u32 *error)
3101 {
3102 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
3103 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
3104 error);
3105 }
3106
3107 static int kvm_read_guest_virt_system(gva_t addr, void *val, unsigned int bytes,
3108 struct kvm_vcpu *vcpu, u32 *error)
3109 {
3110 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, 0, error);
3111 }
3112
3113 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
3114 struct kvm_vcpu *vcpu, u32 *error)
3115 {
3116 void *data = val;
3117 int r = X86EMUL_CONTINUE;
3118
3119 while (bytes) {
3120 gpa_t gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, error);
3121 unsigned offset = addr & (PAGE_SIZE-1);
3122 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
3123 int ret;
3124
3125 if (gpa == UNMAPPED_GVA) {
3126 r = X86EMUL_PROPAGATE_FAULT;
3127 goto out;
3128 }
3129 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
3130 if (ret < 0) {
3131 r = X86EMUL_UNHANDLEABLE;
3132 goto out;
3133 }
3134
3135 bytes -= towrite;
3136 data += towrite;
3137 addr += towrite;
3138 }
3139 out:
3140 return r;
3141 }
3142
3143
3144 static int emulator_read_emulated(unsigned long addr,
3145 void *val,
3146 unsigned int bytes,
3147 struct kvm_vcpu *vcpu)
3148 {
3149 gpa_t gpa;
3150 u32 error_code;
3151
3152 if (vcpu->mmio_read_completed) {
3153 memcpy(val, vcpu->mmio_data, bytes);
3154 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
3155 vcpu->mmio_phys_addr, *(u64 *)val);
3156 vcpu->mmio_read_completed = 0;
3157 return X86EMUL_CONTINUE;
3158 }
3159
3160 gpa = kvm_mmu_gva_to_gpa_read(vcpu, addr, &error_code);
3161
3162 if (gpa == UNMAPPED_GVA) {
3163 kvm_inject_page_fault(vcpu, addr, error_code);
3164 return X86EMUL_PROPAGATE_FAULT;
3165 }
3166
3167 /* For APIC access vmexit */
3168 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3169 goto mmio;
3170
3171 if (kvm_read_guest_virt(addr, val, bytes, vcpu, NULL)
3172 == X86EMUL_CONTINUE)
3173 return X86EMUL_CONTINUE;
3174
3175 mmio:
3176 /*
3177 * Is this MMIO handled locally?
3178 */
3179 if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
3180 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
3181 return X86EMUL_CONTINUE;
3182 }
3183
3184 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
3185
3186 vcpu->mmio_needed = 1;
3187 vcpu->mmio_phys_addr = gpa;
3188 vcpu->mmio_size = bytes;
3189 vcpu->mmio_is_write = 0;
3190
3191 return X86EMUL_UNHANDLEABLE;
3192 }
3193
3194 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
3195 const void *val, int bytes)
3196 {
3197 int ret;
3198
3199 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
3200 if (ret < 0)
3201 return 0;
3202 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
3203 return 1;
3204 }
3205
3206 static int emulator_write_emulated_onepage(unsigned long addr,
3207 const void *val,
3208 unsigned int bytes,
3209 struct kvm_vcpu *vcpu)
3210 {
3211 gpa_t gpa;
3212 u32 error_code;
3213
3214 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, &error_code);
3215
3216 if (gpa == UNMAPPED_GVA) {
3217 kvm_inject_page_fault(vcpu, addr, error_code);
3218 return X86EMUL_PROPAGATE_FAULT;
3219 }
3220
3221 /* For APIC access vmexit */
3222 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3223 goto mmio;
3224
3225 if (emulator_write_phys(vcpu, gpa, val, bytes))
3226 return X86EMUL_CONTINUE;
3227
3228 mmio:
3229 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
3230 /*
3231 * Is this MMIO handled locally?
3232 */
3233 if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
3234 return X86EMUL_CONTINUE;
3235
3236 vcpu->mmio_needed = 1;
3237 vcpu->mmio_phys_addr = gpa;
3238 vcpu->mmio_size = bytes;
3239 vcpu->mmio_is_write = 1;
3240 memcpy(vcpu->mmio_data, val, bytes);
3241
3242 return X86EMUL_CONTINUE;
3243 }
3244
3245 int emulator_write_emulated(unsigned long addr,
3246 const void *val,
3247 unsigned int bytes,
3248 struct kvm_vcpu *vcpu)
3249 {
3250 /* Crossing a page boundary? */
3251 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
3252 int rc, now;
3253
3254 now = -addr & ~PAGE_MASK;
3255 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
3256 if (rc != X86EMUL_CONTINUE)
3257 return rc;
3258 addr += now;
3259 val += now;
3260 bytes -= now;
3261 }
3262 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
3263 }
3264 EXPORT_SYMBOL_GPL(emulator_write_emulated);
3265
3266 static int emulator_cmpxchg_emulated(unsigned long addr,
3267 const void *old,
3268 const void *new,
3269 unsigned int bytes,
3270 struct kvm_vcpu *vcpu)
3271 {
3272 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
3273 #ifndef CONFIG_X86_64
3274 /* guests cmpxchg8b have to be emulated atomically */
3275 if (bytes == 8) {
3276 gpa_t gpa;
3277 struct page *page;
3278 char *kaddr;
3279 u64 val;
3280
3281 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
3282
3283 if (gpa == UNMAPPED_GVA ||
3284 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
3285 goto emul_write;
3286
3287 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
3288 goto emul_write;
3289
3290 val = *(u64 *)new;
3291
3292 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3293
3294 kaddr = kmap_atomic(page, KM_USER0);
3295 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
3296 kunmap_atomic(kaddr, KM_USER0);
3297 kvm_release_page_dirty(page);
3298 }
3299 emul_write:
3300 #endif
3301
3302 return emulator_write_emulated(addr, new, bytes, vcpu);
3303 }
3304
3305 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
3306 {
3307 return kvm_x86_ops->get_segment_base(vcpu, seg);
3308 }
3309
3310 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
3311 {
3312 kvm_mmu_invlpg(vcpu, address);
3313 return X86EMUL_CONTINUE;
3314 }
3315
3316 int emulate_clts(struct kvm_vcpu *vcpu)
3317 {
3318 kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3319 kvm_x86_ops->fpu_activate(vcpu);
3320 return X86EMUL_CONTINUE;
3321 }
3322
3323 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
3324 {
3325 return kvm_x86_ops->get_dr(ctxt->vcpu, dr, dest);
3326 }
3327
3328 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
3329 {
3330 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
3331
3332 return kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask);
3333 }
3334
3335 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
3336 {
3337 u8 opcodes[4];
3338 unsigned long rip = kvm_rip_read(vcpu);
3339 unsigned long rip_linear;
3340
3341 if (!printk_ratelimit())
3342 return;
3343
3344 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
3345
3346 kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu, NULL);
3347
3348 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
3349 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
3350 }
3351 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
3352
3353 static struct x86_emulate_ops emulate_ops = {
3354 .read_std = kvm_read_guest_virt_system,
3355 .fetch = kvm_fetch_guest_virt,
3356 .read_emulated = emulator_read_emulated,
3357 .write_emulated = emulator_write_emulated,
3358 .cmpxchg_emulated = emulator_cmpxchg_emulated,
3359 };
3360
3361 static void cache_all_regs(struct kvm_vcpu *vcpu)
3362 {
3363 kvm_register_read(vcpu, VCPU_REGS_RAX);
3364 kvm_register_read(vcpu, VCPU_REGS_RSP);
3365 kvm_register_read(vcpu, VCPU_REGS_RIP);
3366 vcpu->arch.regs_dirty = ~0;
3367 }
3368
3369 int emulate_instruction(struct kvm_vcpu *vcpu,
3370 unsigned long cr2,
3371 u16 error_code,
3372 int emulation_type)
3373 {
3374 int r, shadow_mask;
3375 struct decode_cache *c;
3376 struct kvm_run *run = vcpu->run;
3377
3378 kvm_clear_exception_queue(vcpu);
3379 vcpu->arch.mmio_fault_cr2 = cr2;
3380 /*
3381 * TODO: fix emulate.c to use guest_read/write_register
3382 * instead of direct ->regs accesses, can save hundred cycles
3383 * on Intel for instructions that don't read/change RSP, for
3384 * for example.
3385 */
3386 cache_all_regs(vcpu);
3387
3388 vcpu->mmio_is_write = 0;
3389 vcpu->arch.pio.string = 0;
3390
3391 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
3392 int cs_db, cs_l;
3393 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3394
3395 vcpu->arch.emulate_ctxt.vcpu = vcpu;
3396 vcpu->arch.emulate_ctxt.eflags = kvm_get_rflags(vcpu);
3397 vcpu->arch.emulate_ctxt.mode =
3398 (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
3399 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
3400 ? X86EMUL_MODE_VM86 : cs_l
3401 ? X86EMUL_MODE_PROT64 : cs_db
3402 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
3403
3404 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3405
3406 /* Only allow emulation of specific instructions on #UD
3407 * (namely VMMCALL, sysenter, sysexit, syscall)*/
3408 c = &vcpu->arch.emulate_ctxt.decode;
3409 if (emulation_type & EMULTYPE_TRAP_UD) {
3410 if (!c->twobyte)
3411 return EMULATE_FAIL;
3412 switch (c->b) {
3413 case 0x01: /* VMMCALL */
3414 if (c->modrm_mod != 3 || c->modrm_rm != 1)
3415 return EMULATE_FAIL;
3416 break;
3417 case 0x34: /* sysenter */
3418 case 0x35: /* sysexit */
3419 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3420 return EMULATE_FAIL;
3421 break;
3422 case 0x05: /* syscall */
3423 if (c->modrm_mod != 0 || c->modrm_rm != 0)
3424 return EMULATE_FAIL;
3425 break;
3426 default:
3427 return EMULATE_FAIL;
3428 }
3429
3430 if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
3431 return EMULATE_FAIL;
3432 }
3433
3434 ++vcpu->stat.insn_emulation;
3435 if (r) {
3436 ++vcpu->stat.insn_emulation_fail;
3437 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3438 return EMULATE_DONE;
3439 return EMULATE_FAIL;
3440 }
3441 }
3442
3443 if (emulation_type & EMULTYPE_SKIP) {
3444 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
3445 return EMULATE_DONE;
3446 }
3447
3448 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
3449 shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
3450
3451 if (r == 0)
3452 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
3453
3454 if (vcpu->arch.pio.string)
3455 return EMULATE_DO_MMIO;
3456
3457 if (r || vcpu->mmio_is_write) {
3458 run->exit_reason = KVM_EXIT_MMIO;
3459 run->mmio.phys_addr = vcpu->mmio_phys_addr;
3460 memcpy(run->mmio.data, vcpu->mmio_data, 8);
3461 run->mmio.len = vcpu->mmio_size;
3462 run->mmio.is_write = vcpu->mmio_is_write;
3463 }
3464
3465 if (r) {
3466 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
3467 return EMULATE_DONE;
3468 if (!vcpu->mmio_needed) {
3469 kvm_report_emulation_failure(vcpu, "mmio");
3470 return EMULATE_FAIL;
3471 }
3472 return EMULATE_DO_MMIO;
3473 }
3474
3475 kvm_set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
3476
3477 if (vcpu->mmio_is_write) {
3478 vcpu->mmio_needed = 0;
3479 return EMULATE_DO_MMIO;
3480 }
3481
3482 return EMULATE_DONE;
3483 }
3484 EXPORT_SYMBOL_GPL(emulate_instruction);
3485
3486 static int pio_copy_data(struct kvm_vcpu *vcpu)
3487 {
3488 void *p = vcpu->arch.pio_data;
3489 gva_t q = vcpu->arch.pio.guest_gva;
3490 unsigned bytes;
3491 int ret;
3492 u32 error_code;
3493
3494 bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
3495 if (vcpu->arch.pio.in)
3496 ret = kvm_write_guest_virt(q, p, bytes, vcpu, &error_code);
3497 else
3498 ret = kvm_read_guest_virt(q, p, bytes, vcpu, &error_code);
3499
3500 if (ret == X86EMUL_PROPAGATE_FAULT)
3501 kvm_inject_page_fault(vcpu, q, error_code);
3502
3503 return ret;
3504 }
3505
3506 int complete_pio(struct kvm_vcpu *vcpu)
3507 {
3508 struct kvm_pio_request *io = &vcpu->arch.pio;
3509 long delta;
3510 int r;
3511 unsigned long val;
3512
3513 if (!io->string) {
3514 if (io->in) {
3515 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3516 memcpy(&val, vcpu->arch.pio_data, io->size);
3517 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
3518 }
3519 } else {
3520 if (io->in) {
3521 r = pio_copy_data(vcpu);
3522 if (r)
3523 goto out;
3524 }
3525
3526 delta = 1;
3527 if (io->rep) {
3528 delta *= io->cur_count;
3529 /*
3530 * The size of the register should really depend on
3531 * current address size.
3532 */
3533 val = kvm_register_read(vcpu, VCPU_REGS_RCX);
3534 val -= delta;
3535 kvm_register_write(vcpu, VCPU_REGS_RCX, val);
3536 }
3537 if (io->down)
3538 delta = -delta;
3539 delta *= io->size;
3540 if (io->in) {
3541 val = kvm_register_read(vcpu, VCPU_REGS_RDI);
3542 val += delta;
3543 kvm_register_write(vcpu, VCPU_REGS_RDI, val);
3544 } else {
3545 val = kvm_register_read(vcpu, VCPU_REGS_RSI);
3546 val += delta;
3547 kvm_register_write(vcpu, VCPU_REGS_RSI, val);
3548 }
3549 }
3550 out:
3551 io->count -= io->cur_count;
3552 io->cur_count = 0;
3553
3554 return 0;
3555 }
3556
3557 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
3558 {
3559 /* TODO: String I/O for in kernel device */
3560 int r;
3561
3562 if (vcpu->arch.pio.in)
3563 r = kvm_io_bus_read(vcpu->kvm, KVM_PIO_BUS, vcpu->arch.pio.port,
3564 vcpu->arch.pio.size, pd);
3565 else
3566 r = kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3567 vcpu->arch.pio.port, vcpu->arch.pio.size,
3568 pd);
3569 return r;
3570 }
3571
3572 static int pio_string_write(struct kvm_vcpu *vcpu)
3573 {
3574 struct kvm_pio_request *io = &vcpu->arch.pio;
3575 void *pd = vcpu->arch.pio_data;
3576 int i, r = 0;
3577
3578 for (i = 0; i < io->cur_count; i++) {
3579 if (kvm_io_bus_write(vcpu->kvm, KVM_PIO_BUS,
3580 io->port, io->size, pd)) {
3581 r = -EOPNOTSUPP;
3582 break;
3583 }
3584 pd += io->size;
3585 }
3586 return r;
3587 }
3588
3589 int kvm_emulate_pio(struct kvm_vcpu *vcpu, int in, int size, unsigned port)
3590 {
3591 unsigned long val;
3592
3593 trace_kvm_pio(!in, port, size, 1);
3594
3595 vcpu->run->exit_reason = KVM_EXIT_IO;
3596 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3597 vcpu->run->io.size = vcpu->arch.pio.size = size;
3598 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3599 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
3600 vcpu->run->io.port = vcpu->arch.pio.port = port;
3601 vcpu->arch.pio.in = in;
3602 vcpu->arch.pio.string = 0;
3603 vcpu->arch.pio.down = 0;
3604 vcpu->arch.pio.rep = 0;
3605
3606 if (!vcpu->arch.pio.in) {
3607 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
3608 memcpy(vcpu->arch.pio_data, &val, 4);
3609 }
3610
3611 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
3612 complete_pio(vcpu);
3613 return 1;
3614 }
3615 return 0;
3616 }
3617 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
3618
3619 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, int in,
3620 int size, unsigned long count, int down,
3621 gva_t address, int rep, unsigned port)
3622 {
3623 unsigned now, in_page;
3624 int ret = 0;
3625
3626 trace_kvm_pio(!in, port, size, count);
3627
3628 vcpu->run->exit_reason = KVM_EXIT_IO;
3629 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3630 vcpu->run->io.size = vcpu->arch.pio.size = size;
3631 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3632 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
3633 vcpu->run->io.port = vcpu->arch.pio.port = port;
3634 vcpu->arch.pio.in = in;
3635 vcpu->arch.pio.string = 1;
3636 vcpu->arch.pio.down = down;
3637 vcpu->arch.pio.rep = rep;
3638
3639 if (!count) {
3640 kvm_x86_ops->skip_emulated_instruction(vcpu);
3641 return 1;
3642 }
3643
3644 if (!down)
3645 in_page = PAGE_SIZE - offset_in_page(address);
3646 else
3647 in_page = offset_in_page(address) + size;
3648 now = min(count, (unsigned long)in_page / size);
3649 if (!now)
3650 now = 1;
3651 if (down) {
3652 /*
3653 * String I/O in reverse. Yuck. Kill the guest, fix later.
3654 */
3655 pr_unimpl(vcpu, "guest string pio down\n");
3656 kvm_inject_gp(vcpu, 0);
3657 return 1;
3658 }
3659 vcpu->run->io.count = now;
3660 vcpu->arch.pio.cur_count = now;
3661
3662 if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
3663 kvm_x86_ops->skip_emulated_instruction(vcpu);
3664
3665 vcpu->arch.pio.guest_gva = address;
3666
3667 if (!vcpu->arch.pio.in) {
3668 /* string PIO write */
3669 ret = pio_copy_data(vcpu);
3670 if (ret == X86EMUL_PROPAGATE_FAULT)
3671 return 1;
3672 if (ret == 0 && !pio_string_write(vcpu)) {
3673 complete_pio(vcpu);
3674 if (vcpu->arch.pio.count == 0)
3675 ret = 1;
3676 }
3677 }
3678 /* no string PIO read support yet */
3679
3680 return ret;
3681 }
3682 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
3683
3684 static void bounce_off(void *info)
3685 {
3686 /* nothing */
3687 }
3688
3689 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3690 void *data)
3691 {
3692 struct cpufreq_freqs *freq = data;
3693 struct kvm *kvm;
3694 struct kvm_vcpu *vcpu;
3695 int i, send_ipi = 0;
3696
3697 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3698 return 0;
3699 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3700 return 0;
3701 per_cpu(cpu_tsc_khz, freq->cpu) = freq->new;
3702
3703 spin_lock(&kvm_lock);
3704 list_for_each_entry(kvm, &vm_list, vm_list) {
3705 kvm_for_each_vcpu(i, vcpu, kvm) {
3706 if (vcpu->cpu != freq->cpu)
3707 continue;
3708 if (!kvm_request_guest_time_update(vcpu))
3709 continue;
3710 if (vcpu->cpu != smp_processor_id())
3711 send_ipi++;
3712 }
3713 }
3714 spin_unlock(&kvm_lock);
3715
3716 if (freq->old < freq->new && send_ipi) {
3717 /*
3718 * We upscale the frequency. Must make the guest
3719 * doesn't see old kvmclock values while running with
3720 * the new frequency, otherwise we risk the guest sees
3721 * time go backwards.
3722 *
3723 * In case we update the frequency for another cpu
3724 * (which might be in guest context) send an interrupt
3725 * to kick the cpu out of guest context. Next time
3726 * guest context is entered kvmclock will be updated,
3727 * so the guest will not see stale values.
3728 */
3729 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
3730 }
3731 return 0;
3732 }
3733
3734 static struct notifier_block kvmclock_cpufreq_notifier_block = {
3735 .notifier_call = kvmclock_cpufreq_notifier
3736 };
3737
3738 static void kvm_timer_init(void)
3739 {
3740 int cpu;
3741
3742 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3743 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3744 CPUFREQ_TRANSITION_NOTIFIER);
3745 for_each_online_cpu(cpu) {
3746 unsigned long khz = cpufreq_get(cpu);
3747 if (!khz)
3748 khz = tsc_khz;
3749 per_cpu(cpu_tsc_khz, cpu) = khz;
3750 }
3751 } else {
3752 for_each_possible_cpu(cpu)
3753 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
3754 }
3755 }
3756
3757 int kvm_arch_init(void *opaque)
3758 {
3759 int r;
3760 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
3761
3762 if (kvm_x86_ops) {
3763 printk(KERN_ERR "kvm: already loaded the other module\n");
3764 r = -EEXIST;
3765 goto out;
3766 }
3767
3768 if (!ops->cpu_has_kvm_support()) {
3769 printk(KERN_ERR "kvm: no hardware support\n");
3770 r = -EOPNOTSUPP;
3771 goto out;
3772 }
3773 if (ops->disabled_by_bios()) {
3774 printk(KERN_ERR "kvm: disabled by bios\n");
3775 r = -EOPNOTSUPP;
3776 goto out;
3777 }
3778
3779 r = kvm_mmu_module_init();
3780 if (r)
3781 goto out;
3782
3783 kvm_init_msr_list();
3784
3785 kvm_x86_ops = ops;
3786 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3787 kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
3788 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
3789 PT_DIRTY_MASK, PT64_NX_MASK, 0);
3790
3791 kvm_timer_init();
3792
3793 return 0;
3794
3795 out:
3796 return r;
3797 }
3798
3799 void kvm_arch_exit(void)
3800 {
3801 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
3802 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
3803 CPUFREQ_TRANSITION_NOTIFIER);
3804 kvm_x86_ops = NULL;
3805 kvm_mmu_module_exit();
3806 }
3807
3808 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
3809 {
3810 ++vcpu->stat.halt_exits;
3811 if (irqchip_in_kernel(vcpu->kvm)) {
3812 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
3813 return 1;
3814 } else {
3815 vcpu->run->exit_reason = KVM_EXIT_HLT;
3816 return 0;
3817 }
3818 }
3819 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
3820
3821 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
3822 unsigned long a1)
3823 {
3824 if (is_long_mode(vcpu))
3825 return a0;
3826 else
3827 return a0 | ((gpa_t)a1 << 32);
3828 }
3829
3830 int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
3831 {
3832 u64 param, ingpa, outgpa, ret;
3833 uint16_t code, rep_idx, rep_cnt, res = HV_STATUS_SUCCESS, rep_done = 0;
3834 bool fast, longmode;
3835 int cs_db, cs_l;
3836
3837 /*
3838 * hypercall generates UD from non zero cpl and real mode
3839 * per HYPER-V spec
3840 */
3841 if (kvm_x86_ops->get_cpl(vcpu) != 0 || !is_protmode(vcpu)) {
3842 kvm_queue_exception(vcpu, UD_VECTOR);
3843 return 0;
3844 }
3845
3846 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
3847 longmode = is_long_mode(vcpu) && cs_l == 1;
3848
3849 if (!longmode) {
3850 param = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDX) << 32) |
3851 (kvm_register_read(vcpu, VCPU_REGS_RAX) & 0xffffffff);
3852 ingpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RBX) << 32) |
3853 (kvm_register_read(vcpu, VCPU_REGS_RCX) & 0xffffffff);
3854 outgpa = ((u64)kvm_register_read(vcpu, VCPU_REGS_RDI) << 32) |
3855 (kvm_register_read(vcpu, VCPU_REGS_RSI) & 0xffffffff);
3856 }
3857 #ifdef CONFIG_X86_64
3858 else {
3859 param = kvm_register_read(vcpu, VCPU_REGS_RCX);
3860 ingpa = kvm_register_read(vcpu, VCPU_REGS_RDX);
3861 outgpa = kvm_register_read(vcpu, VCPU_REGS_R8);
3862 }
3863 #endif
3864
3865 code = param & 0xffff;
3866 fast = (param >> 16) & 0x1;
3867 rep_cnt = (param >> 32) & 0xfff;
3868 rep_idx = (param >> 48) & 0xfff;
3869
3870 trace_kvm_hv_hypercall(code, fast, rep_cnt, rep_idx, ingpa, outgpa);
3871
3872 switch (code) {
3873 case HV_X64_HV_NOTIFY_LONG_SPIN_WAIT:
3874 kvm_vcpu_on_spin(vcpu);
3875 break;
3876 default:
3877 res = HV_STATUS_INVALID_HYPERCALL_CODE;
3878 break;
3879 }
3880
3881 ret = res | (((u64)rep_done & 0xfff) << 32);
3882 if (longmode) {
3883 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3884 } else {
3885 kvm_register_write(vcpu, VCPU_REGS_RDX, ret >> 32);
3886 kvm_register_write(vcpu, VCPU_REGS_RAX, ret & 0xffffffff);
3887 }
3888
3889 return 1;
3890 }
3891
3892 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
3893 {
3894 unsigned long nr, a0, a1, a2, a3, ret;
3895 int r = 1;
3896
3897 if (kvm_hv_hypercall_enabled(vcpu->kvm))
3898 return kvm_hv_hypercall(vcpu);
3899
3900 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
3901 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
3902 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
3903 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
3904 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
3905
3906 trace_kvm_hypercall(nr, a0, a1, a2, a3);
3907
3908 if (!is_long_mode(vcpu)) {
3909 nr &= 0xFFFFFFFF;
3910 a0 &= 0xFFFFFFFF;
3911 a1 &= 0xFFFFFFFF;
3912 a2 &= 0xFFFFFFFF;
3913 a3 &= 0xFFFFFFFF;
3914 }
3915
3916 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
3917 ret = -KVM_EPERM;
3918 goto out;
3919 }
3920
3921 switch (nr) {
3922 case KVM_HC_VAPIC_POLL_IRQ:
3923 ret = 0;
3924 break;
3925 case KVM_HC_MMU_OP:
3926 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
3927 break;
3928 default:
3929 ret = -KVM_ENOSYS;
3930 break;
3931 }
3932 out:
3933 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3934 ++vcpu->stat.hypercalls;
3935 return r;
3936 }
3937 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
3938
3939 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
3940 {
3941 char instruction[3];
3942 unsigned long rip = kvm_rip_read(vcpu);
3943
3944 /*
3945 * Blow out the MMU to ensure that no other VCPU has an active mapping
3946 * to ensure that the updated hypercall appears atomically across all
3947 * VCPUs.
3948 */
3949 kvm_mmu_zap_all(vcpu->kvm);
3950
3951 kvm_x86_ops->patch_hypercall(vcpu, instruction);
3952
3953 return emulator_write_emulated(rip, instruction, 3, vcpu);
3954 }
3955
3956 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3957 {
3958 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3959 }
3960
3961 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3962 {
3963 struct desc_ptr dt = { limit, base };
3964
3965 kvm_x86_ops->set_gdt(vcpu, &dt);
3966 }
3967
3968 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3969 {
3970 struct desc_ptr dt = { limit, base };
3971
3972 kvm_x86_ops->set_idt(vcpu, &dt);
3973 }
3974
3975 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
3976 unsigned long *rflags)
3977 {
3978 kvm_lmsw(vcpu, msw);
3979 *rflags = kvm_get_rflags(vcpu);
3980 }
3981
3982 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
3983 {
3984 unsigned long value;
3985
3986 switch (cr) {
3987 case 0:
3988 value = kvm_read_cr0(vcpu);
3989 break;
3990 case 2:
3991 value = vcpu->arch.cr2;
3992 break;
3993 case 3:
3994 value = vcpu->arch.cr3;
3995 break;
3996 case 4:
3997 value = kvm_read_cr4(vcpu);
3998 break;
3999 case 8:
4000 value = kvm_get_cr8(vcpu);
4001 break;
4002 default:
4003 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4004 return 0;
4005 }
4006
4007 return value;
4008 }
4009
4010 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
4011 unsigned long *rflags)
4012 {
4013 switch (cr) {
4014 case 0:
4015 kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
4016 *rflags = kvm_get_rflags(vcpu);
4017 break;
4018 case 2:
4019 vcpu->arch.cr2 = val;
4020 break;
4021 case 3:
4022 kvm_set_cr3(vcpu, val);
4023 break;
4024 case 4:
4025 kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
4026 break;
4027 case 8:
4028 kvm_set_cr8(vcpu, val & 0xfUL);
4029 break;
4030 default:
4031 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
4032 }
4033 }
4034
4035 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
4036 {
4037 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
4038 int j, nent = vcpu->arch.cpuid_nent;
4039
4040 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
4041 /* when no next entry is found, the current entry[i] is reselected */
4042 for (j = i + 1; ; j = (j + 1) % nent) {
4043 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
4044 if (ej->function == e->function) {
4045 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
4046 return j;
4047 }
4048 }
4049 return 0; /* silence gcc, even though control never reaches here */
4050 }
4051
4052 /* find an entry with matching function, matching index (if needed), and that
4053 * should be read next (if it's stateful) */
4054 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
4055 u32 function, u32 index)
4056 {
4057 if (e->function != function)
4058 return 0;
4059 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
4060 return 0;
4061 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
4062 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
4063 return 0;
4064 return 1;
4065 }
4066
4067 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
4068 u32 function, u32 index)
4069 {
4070 int i;
4071 struct kvm_cpuid_entry2 *best = NULL;
4072
4073 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
4074 struct kvm_cpuid_entry2 *e;
4075
4076 e = &vcpu->arch.cpuid_entries[i];
4077 if (is_matching_cpuid_entry(e, function, index)) {
4078 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
4079 move_to_next_stateful_cpuid_entry(vcpu, i);
4080 best = e;
4081 break;
4082 }
4083 /*
4084 * Both basic or both extended?
4085 */
4086 if (((e->function ^ function) & 0x80000000) == 0)
4087 if (!best || e->function > best->function)
4088 best = e;
4089 }
4090 return best;
4091 }
4092 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
4093
4094 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
4095 {
4096 struct kvm_cpuid_entry2 *best;
4097
4098 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
4099 if (best)
4100 return best->eax & 0xff;
4101 return 36;
4102 }
4103
4104 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
4105 {
4106 u32 function, index;
4107 struct kvm_cpuid_entry2 *best;
4108
4109 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
4110 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4111 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
4112 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
4113 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
4114 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
4115 best = kvm_find_cpuid_entry(vcpu, function, index);
4116 if (best) {
4117 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
4118 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
4119 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
4120 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
4121 }
4122 kvm_x86_ops->skip_emulated_instruction(vcpu);
4123 trace_kvm_cpuid(function,
4124 kvm_register_read(vcpu, VCPU_REGS_RAX),
4125 kvm_register_read(vcpu, VCPU_REGS_RBX),
4126 kvm_register_read(vcpu, VCPU_REGS_RCX),
4127 kvm_register_read(vcpu, VCPU_REGS_RDX));
4128 }
4129 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
4130
4131 /*
4132 * Check if userspace requested an interrupt window, and that the
4133 * interrupt window is open.
4134 *
4135 * No need to exit to userspace if we already have an interrupt queued.
4136 */
4137 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
4138 {
4139 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
4140 vcpu->run->request_interrupt_window &&
4141 kvm_arch_interrupt_allowed(vcpu));
4142 }
4143
4144 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
4145 {
4146 struct kvm_run *kvm_run = vcpu->run;
4147
4148 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
4149 kvm_run->cr8 = kvm_get_cr8(vcpu);
4150 kvm_run->apic_base = kvm_get_apic_base(vcpu);
4151 if (irqchip_in_kernel(vcpu->kvm))
4152 kvm_run->ready_for_interrupt_injection = 1;
4153 else
4154 kvm_run->ready_for_interrupt_injection =
4155 kvm_arch_interrupt_allowed(vcpu) &&
4156 !kvm_cpu_has_interrupt(vcpu) &&
4157 !kvm_event_needs_reinjection(vcpu);
4158 }
4159
4160 static void vapic_enter(struct kvm_vcpu *vcpu)
4161 {
4162 struct kvm_lapic *apic = vcpu->arch.apic;
4163 struct page *page;
4164
4165 if (!apic || !apic->vapic_addr)
4166 return;
4167
4168 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4169
4170 vcpu->arch.apic->vapic_page = page;
4171 }
4172
4173 static void vapic_exit(struct kvm_vcpu *vcpu)
4174 {
4175 struct kvm_lapic *apic = vcpu->arch.apic;
4176 int idx;
4177
4178 if (!apic || !apic->vapic_addr)
4179 return;
4180
4181 idx = srcu_read_lock(&vcpu->kvm->srcu);
4182 kvm_release_page_dirty(apic->vapic_page);
4183 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
4184 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4185 }
4186
4187 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
4188 {
4189 int max_irr, tpr;
4190
4191 if (!kvm_x86_ops->update_cr8_intercept)
4192 return;
4193
4194 if (!vcpu->arch.apic)
4195 return;
4196
4197 if (!vcpu->arch.apic->vapic_addr)
4198 max_irr = kvm_lapic_find_highest_irr(vcpu);
4199 else
4200 max_irr = -1;
4201
4202 if (max_irr != -1)
4203 max_irr >>= 4;
4204
4205 tpr = kvm_lapic_get_cr8(vcpu);
4206
4207 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
4208 }
4209
4210 static void inject_pending_event(struct kvm_vcpu *vcpu)
4211 {
4212 /* try to reinject previous events if any */
4213 if (vcpu->arch.exception.pending) {
4214 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
4215 vcpu->arch.exception.has_error_code,
4216 vcpu->arch.exception.error_code);
4217 return;
4218 }
4219
4220 if (vcpu->arch.nmi_injected) {
4221 kvm_x86_ops->set_nmi(vcpu);
4222 return;
4223 }
4224
4225 if (vcpu->arch.interrupt.pending) {
4226 kvm_x86_ops->set_irq(vcpu);
4227 return;
4228 }
4229
4230 /* try to inject new event if pending */
4231 if (vcpu->arch.nmi_pending) {
4232 if (kvm_x86_ops->nmi_allowed(vcpu)) {
4233 vcpu->arch.nmi_pending = false;
4234 vcpu->arch.nmi_injected = true;
4235 kvm_x86_ops->set_nmi(vcpu);
4236 }
4237 } else if (kvm_cpu_has_interrupt(vcpu)) {
4238 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
4239 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
4240 false);
4241 kvm_x86_ops->set_irq(vcpu);
4242 }
4243 }
4244 }
4245
4246 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
4247 {
4248 int r;
4249 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
4250 vcpu->run->request_interrupt_window;
4251
4252 if (vcpu->requests)
4253 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
4254 kvm_mmu_unload(vcpu);
4255
4256 r = kvm_mmu_reload(vcpu);
4257 if (unlikely(r))
4258 goto out;
4259
4260 if (vcpu->requests) {
4261 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
4262 __kvm_migrate_timers(vcpu);
4263 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
4264 kvm_write_guest_time(vcpu);
4265 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
4266 kvm_mmu_sync_roots(vcpu);
4267 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
4268 kvm_x86_ops->tlb_flush(vcpu);
4269 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
4270 &vcpu->requests)) {
4271 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
4272 r = 0;
4273 goto out;
4274 }
4275 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
4276 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4277 r = 0;
4278 goto out;
4279 }
4280 if (test_and_clear_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests)) {
4281 vcpu->fpu_active = 0;
4282 kvm_x86_ops->fpu_deactivate(vcpu);
4283 }
4284 }
4285
4286 preempt_disable();
4287
4288 kvm_x86_ops->prepare_guest_switch(vcpu);
4289 if (vcpu->fpu_active)
4290 kvm_load_guest_fpu(vcpu);
4291
4292 local_irq_disable();
4293
4294 clear_bit(KVM_REQ_KICK, &vcpu->requests);
4295 smp_mb__after_clear_bit();
4296
4297 if (vcpu->requests || need_resched() || signal_pending(current)) {
4298 set_bit(KVM_REQ_KICK, &vcpu->requests);
4299 local_irq_enable();
4300 preempt_enable();
4301 r = 1;
4302 goto out;
4303 }
4304
4305 inject_pending_event(vcpu);
4306
4307 /* enable NMI/IRQ window open exits if needed */
4308 if (vcpu->arch.nmi_pending)
4309 kvm_x86_ops->enable_nmi_window(vcpu);
4310 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
4311 kvm_x86_ops->enable_irq_window(vcpu);
4312
4313 if (kvm_lapic_enabled(vcpu)) {
4314 update_cr8_intercept(vcpu);
4315 kvm_lapic_sync_to_vapic(vcpu);
4316 }
4317
4318 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4319
4320 kvm_guest_enter();
4321
4322 if (unlikely(vcpu->arch.switch_db_regs)) {
4323 set_debugreg(0, 7);
4324 set_debugreg(vcpu->arch.eff_db[0], 0);
4325 set_debugreg(vcpu->arch.eff_db[1], 1);
4326 set_debugreg(vcpu->arch.eff_db[2], 2);
4327 set_debugreg(vcpu->arch.eff_db[3], 3);
4328 }
4329
4330 trace_kvm_entry(vcpu->vcpu_id);
4331 kvm_x86_ops->run(vcpu);
4332
4333 /*
4334 * If the guest has used debug registers, at least dr7
4335 * will be disabled while returning to the host.
4336 * If we don't have active breakpoints in the host, we don't
4337 * care about the messed up debug address registers. But if
4338 * we have some of them active, restore the old state.
4339 */
4340 if (hw_breakpoint_active())
4341 hw_breakpoint_restore();
4342
4343 set_bit(KVM_REQ_KICK, &vcpu->requests);
4344 local_irq_enable();
4345
4346 ++vcpu->stat.exits;
4347
4348 /*
4349 * We must have an instruction between local_irq_enable() and
4350 * kvm_guest_exit(), so the timer interrupt isn't delayed by
4351 * the interrupt shadow. The stat.exits increment will do nicely.
4352 * But we need to prevent reordering, hence this barrier():
4353 */
4354 barrier();
4355
4356 kvm_guest_exit();
4357
4358 preempt_enable();
4359
4360 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4361
4362 /*
4363 * Profile KVM exit RIPs:
4364 */
4365 if (unlikely(prof_on == KVM_PROFILING)) {
4366 unsigned long rip = kvm_rip_read(vcpu);
4367 profile_hit(KVM_PROFILING, (void *)rip);
4368 }
4369
4370
4371 kvm_lapic_sync_from_vapic(vcpu);
4372
4373 r = kvm_x86_ops->handle_exit(vcpu);
4374 out:
4375 return r;
4376 }
4377
4378
4379 static int __vcpu_run(struct kvm_vcpu *vcpu)
4380 {
4381 int r;
4382 struct kvm *kvm = vcpu->kvm;
4383
4384 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
4385 pr_debug("vcpu %d received sipi with vector # %x\n",
4386 vcpu->vcpu_id, vcpu->arch.sipi_vector);
4387 kvm_lapic_reset(vcpu);
4388 r = kvm_arch_vcpu_reset(vcpu);
4389 if (r)
4390 return r;
4391 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4392 }
4393
4394 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4395 vapic_enter(vcpu);
4396
4397 r = 1;
4398 while (r > 0) {
4399 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
4400 r = vcpu_enter_guest(vcpu);
4401 else {
4402 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4403 kvm_vcpu_block(vcpu);
4404 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4405 if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
4406 {
4407 switch(vcpu->arch.mp_state) {
4408 case KVM_MP_STATE_HALTED:
4409 vcpu->arch.mp_state =
4410 KVM_MP_STATE_RUNNABLE;
4411 case KVM_MP_STATE_RUNNABLE:
4412 break;
4413 case KVM_MP_STATE_SIPI_RECEIVED:
4414 default:
4415 r = -EINTR;
4416 break;
4417 }
4418 }
4419 }
4420
4421 if (r <= 0)
4422 break;
4423
4424 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
4425 if (kvm_cpu_has_pending_timer(vcpu))
4426 kvm_inject_pending_timer_irqs(vcpu);
4427
4428 if (dm_request_for_irq_injection(vcpu)) {
4429 r = -EINTR;
4430 vcpu->run->exit_reason = KVM_EXIT_INTR;
4431 ++vcpu->stat.request_irq_exits;
4432 }
4433 if (signal_pending(current)) {
4434 r = -EINTR;
4435 vcpu->run->exit_reason = KVM_EXIT_INTR;
4436 ++vcpu->stat.signal_exits;
4437 }
4438 if (need_resched()) {
4439 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4440 kvm_resched(vcpu);
4441 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
4442 }
4443 }
4444
4445 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
4446 post_kvm_run_save(vcpu);
4447
4448 vapic_exit(vcpu);
4449
4450 return r;
4451 }
4452
4453 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
4454 {
4455 int r;
4456 sigset_t sigsaved;
4457
4458 vcpu_load(vcpu);
4459
4460 if (vcpu->sigset_active)
4461 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
4462
4463 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
4464 kvm_vcpu_block(vcpu);
4465 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
4466 r = -EAGAIN;
4467 goto out;
4468 }
4469
4470 /* re-sync apic's tpr */
4471 if (!irqchip_in_kernel(vcpu->kvm))
4472 kvm_set_cr8(vcpu, kvm_run->cr8);
4473
4474 if (vcpu->arch.pio.cur_count) {
4475 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4476 r = complete_pio(vcpu);
4477 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4478 if (r)
4479 goto out;
4480 }
4481 if (vcpu->mmio_needed) {
4482 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
4483 vcpu->mmio_read_completed = 1;
4484 vcpu->mmio_needed = 0;
4485
4486 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4487 r = emulate_instruction(vcpu, vcpu->arch.mmio_fault_cr2, 0,
4488 EMULTYPE_NO_DECODE);
4489 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4490 if (r == EMULATE_DO_MMIO) {
4491 /*
4492 * Read-modify-write. Back to userspace.
4493 */
4494 r = 0;
4495 goto out;
4496 }
4497 }
4498 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
4499 kvm_register_write(vcpu, VCPU_REGS_RAX,
4500 kvm_run->hypercall.ret);
4501
4502 r = __vcpu_run(vcpu);
4503
4504 out:
4505 if (vcpu->sigset_active)
4506 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
4507
4508 vcpu_put(vcpu);
4509 return r;
4510 }
4511
4512 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4513 {
4514 vcpu_load(vcpu);
4515
4516 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4517 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4518 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4519 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4520 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4521 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4522 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4523 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4524 #ifdef CONFIG_X86_64
4525 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
4526 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
4527 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
4528 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
4529 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
4530 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
4531 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
4532 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
4533 #endif
4534
4535 regs->rip = kvm_rip_read(vcpu);
4536 regs->rflags = kvm_get_rflags(vcpu);
4537
4538 vcpu_put(vcpu);
4539
4540 return 0;
4541 }
4542
4543 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
4544 {
4545 vcpu_load(vcpu);
4546
4547 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
4548 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
4549 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
4550 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
4551 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
4552 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
4553 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
4554 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
4555 #ifdef CONFIG_X86_64
4556 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
4557 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
4558 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
4559 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
4560 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
4561 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
4562 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
4563 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
4564 #endif
4565
4566 kvm_rip_write(vcpu, regs->rip);
4567 kvm_set_rflags(vcpu, regs->rflags);
4568
4569 vcpu->arch.exception.pending = false;
4570
4571 vcpu_put(vcpu);
4572
4573 return 0;
4574 }
4575
4576 void kvm_get_segment(struct kvm_vcpu *vcpu,
4577 struct kvm_segment *var, int seg)
4578 {
4579 kvm_x86_ops->get_segment(vcpu, var, seg);
4580 }
4581
4582 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4583 {
4584 struct kvm_segment cs;
4585
4586 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
4587 *db = cs.db;
4588 *l = cs.l;
4589 }
4590 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
4591
4592 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
4593 struct kvm_sregs *sregs)
4594 {
4595 struct desc_ptr dt;
4596
4597 vcpu_load(vcpu);
4598
4599 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4600 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4601 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4602 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4603 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4604 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4605
4606 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4607 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4608
4609 kvm_x86_ops->get_idt(vcpu, &dt);
4610 sregs->idt.limit = dt.size;
4611 sregs->idt.base = dt.address;
4612 kvm_x86_ops->get_gdt(vcpu, &dt);
4613 sregs->gdt.limit = dt.size;
4614 sregs->gdt.base = dt.address;
4615
4616 sregs->cr0 = kvm_read_cr0(vcpu);
4617 sregs->cr2 = vcpu->arch.cr2;
4618 sregs->cr3 = vcpu->arch.cr3;
4619 sregs->cr4 = kvm_read_cr4(vcpu);
4620 sregs->cr8 = kvm_get_cr8(vcpu);
4621 sregs->efer = vcpu->arch.efer;
4622 sregs->apic_base = kvm_get_apic_base(vcpu);
4623
4624 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
4625
4626 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
4627 set_bit(vcpu->arch.interrupt.nr,
4628 (unsigned long *)sregs->interrupt_bitmap);
4629
4630 vcpu_put(vcpu);
4631
4632 return 0;
4633 }
4634
4635 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
4636 struct kvm_mp_state *mp_state)
4637 {
4638 vcpu_load(vcpu);
4639 mp_state->mp_state = vcpu->arch.mp_state;
4640 vcpu_put(vcpu);
4641 return 0;
4642 }
4643
4644 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
4645 struct kvm_mp_state *mp_state)
4646 {
4647 vcpu_load(vcpu);
4648 vcpu->arch.mp_state = mp_state->mp_state;
4649 vcpu_put(vcpu);
4650 return 0;
4651 }
4652
4653 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4654 struct kvm_segment *var, int seg)
4655 {
4656 kvm_x86_ops->set_segment(vcpu, var, seg);
4657 }
4658
4659 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
4660 struct kvm_segment *kvm_desct)
4661 {
4662 kvm_desct->base = get_desc_base(seg_desc);
4663 kvm_desct->limit = get_desc_limit(seg_desc);
4664 if (seg_desc->g) {
4665 kvm_desct->limit <<= 12;
4666 kvm_desct->limit |= 0xfff;
4667 }
4668 kvm_desct->selector = selector;
4669 kvm_desct->type = seg_desc->type;
4670 kvm_desct->present = seg_desc->p;
4671 kvm_desct->dpl = seg_desc->dpl;
4672 kvm_desct->db = seg_desc->d;
4673 kvm_desct->s = seg_desc->s;
4674 kvm_desct->l = seg_desc->l;
4675 kvm_desct->g = seg_desc->g;
4676 kvm_desct->avl = seg_desc->avl;
4677 if (!selector)
4678 kvm_desct->unusable = 1;
4679 else
4680 kvm_desct->unusable = 0;
4681 kvm_desct->padding = 0;
4682 }
4683
4684 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
4685 u16 selector,
4686 struct desc_ptr *dtable)
4687 {
4688 if (selector & 1 << 2) {
4689 struct kvm_segment kvm_seg;
4690
4691 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
4692
4693 if (kvm_seg.unusable)
4694 dtable->size = 0;
4695 else
4696 dtable->size = kvm_seg.limit;
4697 dtable->address = kvm_seg.base;
4698 }
4699 else
4700 kvm_x86_ops->get_gdt(vcpu, dtable);
4701 }
4702
4703 /* allowed just for 8 bytes segments */
4704 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4705 struct desc_struct *seg_desc)
4706 {
4707 struct desc_ptr dtable;
4708 u16 index = selector >> 3;
4709 int ret;
4710 u32 err;
4711 gva_t addr;
4712
4713 get_segment_descriptor_dtable(vcpu, selector, &dtable);
4714
4715 if (dtable.size < index * 8 + 7) {
4716 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
4717 return X86EMUL_PROPAGATE_FAULT;
4718 }
4719 addr = dtable.base + index * 8;
4720 ret = kvm_read_guest_virt_system(addr, seg_desc, sizeof(*seg_desc),
4721 vcpu, &err);
4722 if (ret == X86EMUL_PROPAGATE_FAULT)
4723 kvm_inject_page_fault(vcpu, addr, err);
4724
4725 return ret;
4726 }
4727
4728 /* allowed just for 8 bytes segments */
4729 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4730 struct desc_struct *seg_desc)
4731 {
4732 struct desc_ptr dtable;
4733 u16 index = selector >> 3;
4734
4735 get_segment_descriptor_dtable(vcpu, selector, &dtable);
4736
4737 if (dtable.size < index * 8 + 7)
4738 return 1;
4739 return kvm_write_guest_virt(dtable.address + index*8, seg_desc, sizeof(*seg_desc), vcpu, NULL);
4740 }
4741
4742 static gpa_t get_tss_base_addr_write(struct kvm_vcpu *vcpu,
4743 struct desc_struct *seg_desc)
4744 {
4745 u32 base_addr = get_desc_base(seg_desc);
4746
4747 return kvm_mmu_gva_to_gpa_write(vcpu, base_addr, NULL);
4748 }
4749
4750 static gpa_t get_tss_base_addr_read(struct kvm_vcpu *vcpu,
4751 struct desc_struct *seg_desc)
4752 {
4753 u32 base_addr = get_desc_base(seg_desc);
4754
4755 return kvm_mmu_gva_to_gpa_read(vcpu, base_addr, NULL);
4756 }
4757
4758 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
4759 {
4760 struct kvm_segment kvm_seg;
4761
4762 kvm_get_segment(vcpu, &kvm_seg, seg);
4763 return kvm_seg.selector;
4764 }
4765
4766 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
4767 {
4768 struct kvm_segment segvar = {
4769 .base = selector << 4,
4770 .limit = 0xffff,
4771 .selector = selector,
4772 .type = 3,
4773 .present = 1,
4774 .dpl = 3,
4775 .db = 0,
4776 .s = 1,
4777 .l = 0,
4778 .g = 0,
4779 .avl = 0,
4780 .unusable = 0,
4781 };
4782 kvm_x86_ops->set_segment(vcpu, &segvar, seg);
4783 return X86EMUL_CONTINUE;
4784 }
4785
4786 static int is_vm86_segment(struct kvm_vcpu *vcpu, int seg)
4787 {
4788 return (seg != VCPU_SREG_LDTR) &&
4789 (seg != VCPU_SREG_TR) &&
4790 (kvm_get_rflags(vcpu) & X86_EFLAGS_VM);
4791 }
4792
4793 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector, int seg)
4794 {
4795 struct kvm_segment kvm_seg;
4796 struct desc_struct seg_desc;
4797 u8 dpl, rpl, cpl;
4798 unsigned err_vec = GP_VECTOR;
4799 u32 err_code = 0;
4800 bool null_selector = !(selector & ~0x3); /* 0000-0003 are null */
4801 int ret;
4802
4803 if (is_vm86_segment(vcpu, seg) || !is_protmode(vcpu))
4804 return kvm_load_realmode_segment(vcpu, selector, seg);
4805
4806 /* NULL selector is not valid for TR, CS and SS */
4807 if ((seg == VCPU_SREG_CS || seg == VCPU_SREG_SS || seg == VCPU_SREG_TR)
4808 && null_selector)
4809 goto exception;
4810
4811 /* TR should be in GDT only */
4812 if (seg == VCPU_SREG_TR && (selector & (1 << 2)))
4813 goto exception;
4814
4815 ret = load_guest_segment_descriptor(vcpu, selector, &seg_desc);
4816 if (ret)
4817 return ret;
4818
4819 seg_desct_to_kvm_desct(&seg_desc, selector, &kvm_seg);
4820
4821 if (null_selector) { /* for NULL selector skip all following checks */
4822 kvm_seg.unusable = 1;
4823 goto load;
4824 }
4825
4826 err_code = selector & 0xfffc;
4827 err_vec = GP_VECTOR;
4828
4829 /* can't load system descriptor into segment selecor */
4830 if (seg <= VCPU_SREG_GS && !kvm_seg.s)
4831 goto exception;
4832
4833 if (!kvm_seg.present) {
4834 err_vec = (seg == VCPU_SREG_SS) ? SS_VECTOR : NP_VECTOR;
4835 goto exception;
4836 }
4837
4838 rpl = selector & 3;
4839 dpl = kvm_seg.dpl;
4840 cpl = kvm_x86_ops->get_cpl(vcpu);
4841
4842 switch (seg) {
4843 case VCPU_SREG_SS:
4844 /*
4845 * segment is not a writable data segment or segment
4846 * selector's RPL != CPL or segment selector's RPL != CPL
4847 */
4848 if (rpl != cpl || (kvm_seg.type & 0xa) != 0x2 || dpl != cpl)
4849 goto exception;
4850 break;
4851 case VCPU_SREG_CS:
4852 if (!(kvm_seg.type & 8))
4853 goto exception;
4854
4855 if (kvm_seg.type & 4) {
4856 /* conforming */
4857 if (dpl > cpl)
4858 goto exception;
4859 } else {
4860 /* nonconforming */
4861 if (rpl > cpl || dpl != cpl)
4862 goto exception;
4863 }
4864 /* CS(RPL) <- CPL */
4865 selector = (selector & 0xfffc) | cpl;
4866 break;
4867 case VCPU_SREG_TR:
4868 if (kvm_seg.s || (kvm_seg.type != 1 && kvm_seg.type != 9))
4869 goto exception;
4870 break;
4871 case VCPU_SREG_LDTR:
4872 if (kvm_seg.s || kvm_seg.type != 2)
4873 goto exception;
4874 break;
4875 default: /* DS, ES, FS, or GS */
4876 /*
4877 * segment is not a data or readable code segment or
4878 * ((segment is a data or nonconforming code segment)
4879 * and (both RPL and CPL > DPL))
4880 */
4881 if ((kvm_seg.type & 0xa) == 0x8 ||
4882 (((kvm_seg.type & 0xc) != 0xc) && (rpl > dpl && cpl > dpl)))
4883 goto exception;
4884 break;
4885 }
4886
4887 if (!kvm_seg.unusable && kvm_seg.s) {
4888 /* mark segment as accessed */
4889 kvm_seg.type |= 1;
4890 seg_desc.type |= 1;
4891 save_guest_segment_descriptor(vcpu, selector, &seg_desc);
4892 }
4893 load:
4894 kvm_set_segment(vcpu, &kvm_seg, seg);
4895 return X86EMUL_CONTINUE;
4896 exception:
4897 kvm_queue_exception_e(vcpu, err_vec, err_code);
4898 return X86EMUL_PROPAGATE_FAULT;
4899 }
4900
4901 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
4902 struct tss_segment_32 *tss)
4903 {
4904 tss->cr3 = vcpu->arch.cr3;
4905 tss->eip = kvm_rip_read(vcpu);
4906 tss->eflags = kvm_get_rflags(vcpu);
4907 tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4908 tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4909 tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4910 tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4911 tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4912 tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4913 tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4914 tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4915 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4916 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4917 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4918 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4919 tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
4920 tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
4921 tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4922 }
4923
4924 static void kvm_load_segment_selector(struct kvm_vcpu *vcpu, u16 sel, int seg)
4925 {
4926 struct kvm_segment kvm_seg;
4927 kvm_get_segment(vcpu, &kvm_seg, seg);
4928 kvm_seg.selector = sel;
4929 kvm_set_segment(vcpu, &kvm_seg, seg);
4930 }
4931
4932 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
4933 struct tss_segment_32 *tss)
4934 {
4935 kvm_set_cr3(vcpu, tss->cr3);
4936
4937 kvm_rip_write(vcpu, tss->eip);
4938 kvm_set_rflags(vcpu, tss->eflags | 2);
4939
4940 kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
4941 kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
4942 kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
4943 kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
4944 kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
4945 kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
4946 kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
4947 kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
4948
4949 /*
4950 * SDM says that segment selectors are loaded before segment
4951 * descriptors
4952 */
4953 kvm_load_segment_selector(vcpu, tss->ldt_selector, VCPU_SREG_LDTR);
4954 kvm_load_segment_selector(vcpu, tss->es, VCPU_SREG_ES);
4955 kvm_load_segment_selector(vcpu, tss->cs, VCPU_SREG_CS);
4956 kvm_load_segment_selector(vcpu, tss->ss, VCPU_SREG_SS);
4957 kvm_load_segment_selector(vcpu, tss->ds, VCPU_SREG_DS);
4958 kvm_load_segment_selector(vcpu, tss->fs, VCPU_SREG_FS);
4959 kvm_load_segment_selector(vcpu, tss->gs, VCPU_SREG_GS);
4960
4961 /*
4962 * Now load segment descriptors. If fault happenes at this stage
4963 * it is handled in a context of new task
4964 */
4965 if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, VCPU_SREG_LDTR))
4966 return 1;
4967
4968 if (kvm_load_segment_descriptor(vcpu, tss->es, VCPU_SREG_ES))
4969 return 1;
4970
4971 if (kvm_load_segment_descriptor(vcpu, tss->cs, VCPU_SREG_CS))
4972 return 1;
4973
4974 if (kvm_load_segment_descriptor(vcpu, tss->ss, VCPU_SREG_SS))
4975 return 1;
4976
4977 if (kvm_load_segment_descriptor(vcpu, tss->ds, VCPU_SREG_DS))
4978 return 1;
4979
4980 if (kvm_load_segment_descriptor(vcpu, tss->fs, VCPU_SREG_FS))
4981 return 1;
4982
4983 if (kvm_load_segment_descriptor(vcpu, tss->gs, VCPU_SREG_GS))
4984 return 1;
4985 return 0;
4986 }
4987
4988 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
4989 struct tss_segment_16 *tss)
4990 {
4991 tss->ip = kvm_rip_read(vcpu);
4992 tss->flag = kvm_get_rflags(vcpu);
4993 tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4994 tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4995 tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4996 tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4997 tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4998 tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4999 tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
5000 tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
5001
5002 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
5003 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
5004 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
5005 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
5006 tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
5007 }
5008
5009 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
5010 struct tss_segment_16 *tss)
5011 {
5012 kvm_rip_write(vcpu, tss->ip);
5013 kvm_set_rflags(vcpu, tss->flag | 2);
5014 kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
5015 kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
5016 kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
5017 kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
5018 kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
5019 kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
5020 kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
5021 kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
5022
5023 /*
5024 * SDM says that segment selectors are loaded before segment
5025 * descriptors
5026 */
5027 kvm_load_segment_selector(vcpu, tss->ldt, VCPU_SREG_LDTR);
5028 kvm_load_segment_selector(vcpu, tss->es, VCPU_SREG_ES);
5029 kvm_load_segment_selector(vcpu, tss->cs, VCPU_SREG_CS);
5030 kvm_load_segment_selector(vcpu, tss->ss, VCPU_SREG_SS);
5031 kvm_load_segment_selector(vcpu, tss->ds, VCPU_SREG_DS);
5032
5033 /*
5034 * Now load segment descriptors. If fault happenes at this stage
5035 * it is handled in a context of new task
5036 */
5037 if (kvm_load_segment_descriptor(vcpu, tss->ldt, VCPU_SREG_LDTR))
5038 return 1;
5039
5040 if (kvm_load_segment_descriptor(vcpu, tss->es, VCPU_SREG_ES))
5041 return 1;
5042
5043 if (kvm_load_segment_descriptor(vcpu, tss->cs, VCPU_SREG_CS))
5044 return 1;
5045
5046 if (kvm_load_segment_descriptor(vcpu, tss->ss, VCPU_SREG_SS))
5047 return 1;
5048
5049 if (kvm_load_segment_descriptor(vcpu, tss->ds, VCPU_SREG_DS))
5050 return 1;
5051 return 0;
5052 }
5053
5054 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
5055 u16 old_tss_sel, u32 old_tss_base,
5056 struct desc_struct *nseg_desc)
5057 {
5058 struct tss_segment_16 tss_segment_16;
5059 int ret = 0;
5060
5061 if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
5062 sizeof tss_segment_16))
5063 goto out;
5064
5065 save_state_to_tss16(vcpu, &tss_segment_16);
5066
5067 if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
5068 sizeof tss_segment_16))
5069 goto out;
5070
5071 if (kvm_read_guest(vcpu->kvm, get_tss_base_addr_read(vcpu, nseg_desc),
5072 &tss_segment_16, sizeof tss_segment_16))
5073 goto out;
5074
5075 if (old_tss_sel != 0xffff) {
5076 tss_segment_16.prev_task_link = old_tss_sel;
5077
5078 if (kvm_write_guest(vcpu->kvm,
5079 get_tss_base_addr_write(vcpu, nseg_desc),
5080 &tss_segment_16.prev_task_link,
5081 sizeof tss_segment_16.prev_task_link))
5082 goto out;
5083 }
5084
5085 if (load_state_from_tss16(vcpu, &tss_segment_16))
5086 goto out;
5087
5088 ret = 1;
5089 out:
5090 return ret;
5091 }
5092
5093 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
5094 u16 old_tss_sel, u32 old_tss_base,
5095 struct desc_struct *nseg_desc)
5096 {
5097 struct tss_segment_32 tss_segment_32;
5098 int ret = 0;
5099
5100 if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
5101 sizeof tss_segment_32))
5102 goto out;
5103
5104 save_state_to_tss32(vcpu, &tss_segment_32);
5105
5106 if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
5107 sizeof tss_segment_32))
5108 goto out;
5109
5110 if (kvm_read_guest(vcpu->kvm, get_tss_base_addr_read(vcpu, nseg_desc),
5111 &tss_segment_32, sizeof tss_segment_32))
5112 goto out;
5113
5114 if (old_tss_sel != 0xffff) {
5115 tss_segment_32.prev_task_link = old_tss_sel;
5116
5117 if (kvm_write_guest(vcpu->kvm,
5118 get_tss_base_addr_write(vcpu, nseg_desc),
5119 &tss_segment_32.prev_task_link,
5120 sizeof tss_segment_32.prev_task_link))
5121 goto out;
5122 }
5123
5124 if (load_state_from_tss32(vcpu, &tss_segment_32))
5125 goto out;
5126
5127 ret = 1;
5128 out:
5129 return ret;
5130 }
5131
5132 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
5133 {
5134 struct kvm_segment tr_seg;
5135 struct desc_struct cseg_desc;
5136 struct desc_struct nseg_desc;
5137 int ret = 0;
5138 u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
5139 u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
5140 u32 desc_limit;
5141
5142 old_tss_base = kvm_mmu_gva_to_gpa_write(vcpu, old_tss_base, NULL);
5143
5144 /* FIXME: Handle errors. Failure to read either TSS or their
5145 * descriptors should generate a pagefault.
5146 */
5147 if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
5148 goto out;
5149
5150 if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
5151 goto out;
5152
5153 if (reason != TASK_SWITCH_IRET) {
5154 int cpl;
5155
5156 cpl = kvm_x86_ops->get_cpl(vcpu);
5157 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
5158 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
5159 return 1;
5160 }
5161 }
5162
5163 desc_limit = get_desc_limit(&nseg_desc);
5164 if (!nseg_desc.p ||
5165 ((desc_limit < 0x67 && (nseg_desc.type & 8)) ||
5166 desc_limit < 0x2b)) {
5167 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
5168 return 1;
5169 }
5170
5171 if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
5172 cseg_desc.type &= ~(1 << 1); //clear the B flag
5173 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
5174 }
5175
5176 if (reason == TASK_SWITCH_IRET) {
5177 u32 eflags = kvm_get_rflags(vcpu);
5178 kvm_set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
5179 }
5180
5181 /* set back link to prev task only if NT bit is set in eflags
5182 note that old_tss_sel is not used afetr this point */
5183 if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
5184 old_tss_sel = 0xffff;
5185
5186 if (nseg_desc.type & 8)
5187 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
5188 old_tss_base, &nseg_desc);
5189 else
5190 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
5191 old_tss_base, &nseg_desc);
5192
5193 if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
5194 u32 eflags = kvm_get_rflags(vcpu);
5195 kvm_set_rflags(vcpu, eflags | X86_EFLAGS_NT);
5196 }
5197
5198 if (reason != TASK_SWITCH_IRET) {
5199 nseg_desc.type |= (1 << 1);
5200 save_guest_segment_descriptor(vcpu, tss_selector,
5201 &nseg_desc);
5202 }
5203
5204 kvm_x86_ops->set_cr0(vcpu, kvm_read_cr0(vcpu) | X86_CR0_TS);
5205 seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
5206 tr_seg.type = 11;
5207 kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
5208 out:
5209 return ret;
5210 }
5211 EXPORT_SYMBOL_GPL(kvm_task_switch);
5212
5213 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
5214 struct kvm_sregs *sregs)
5215 {
5216 int mmu_reset_needed = 0;
5217 int pending_vec, max_bits;
5218 struct desc_ptr dt;
5219
5220 vcpu_load(vcpu);
5221
5222 dt.size = sregs->idt.limit;
5223 dt.address = sregs->idt.base;
5224 kvm_x86_ops->set_idt(vcpu, &dt);
5225 dt.size = sregs->gdt.limit;
5226 dt.address = sregs->gdt.base;
5227 kvm_x86_ops->set_gdt(vcpu, &dt);
5228
5229 vcpu->arch.cr2 = sregs->cr2;
5230 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
5231 vcpu->arch.cr3 = sregs->cr3;
5232
5233 kvm_set_cr8(vcpu, sregs->cr8);
5234
5235 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
5236 kvm_x86_ops->set_efer(vcpu, sregs->efer);
5237 kvm_set_apic_base(vcpu, sregs->apic_base);
5238
5239 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
5240 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
5241 vcpu->arch.cr0 = sregs->cr0;
5242
5243 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
5244 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
5245 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
5246 load_pdptrs(vcpu, vcpu->arch.cr3);
5247 mmu_reset_needed = 1;
5248 }
5249
5250 if (mmu_reset_needed)
5251 kvm_mmu_reset_context(vcpu);
5252
5253 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
5254 pending_vec = find_first_bit(
5255 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
5256 if (pending_vec < max_bits) {
5257 kvm_queue_interrupt(vcpu, pending_vec, false);
5258 pr_debug("Set back pending irq %d\n", pending_vec);
5259 if (irqchip_in_kernel(vcpu->kvm))
5260 kvm_pic_clear_isr_ack(vcpu->kvm);
5261 }
5262
5263 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
5264 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
5265 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
5266 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
5267 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
5268 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
5269
5270 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
5271 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
5272
5273 update_cr8_intercept(vcpu);
5274
5275 /* Older userspace won't unhalt the vcpu on reset. */
5276 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
5277 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
5278 !is_protmode(vcpu))
5279 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5280
5281 vcpu_put(vcpu);
5282
5283 return 0;
5284 }
5285
5286 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
5287 struct kvm_guest_debug *dbg)
5288 {
5289 unsigned long rflags;
5290 int i, r;
5291
5292 vcpu_load(vcpu);
5293
5294 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
5295 r = -EBUSY;
5296 if (vcpu->arch.exception.pending)
5297 goto unlock_out;
5298 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
5299 kvm_queue_exception(vcpu, DB_VECTOR);
5300 else
5301 kvm_queue_exception(vcpu, BP_VECTOR);
5302 }
5303
5304 /*
5305 * Read rflags as long as potentially injected trace flags are still
5306 * filtered out.
5307 */
5308 rflags = kvm_get_rflags(vcpu);
5309
5310 vcpu->guest_debug = dbg->control;
5311 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
5312 vcpu->guest_debug = 0;
5313
5314 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5315 for (i = 0; i < KVM_NR_DB_REGS; ++i)
5316 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
5317 vcpu->arch.switch_db_regs =
5318 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
5319 } else {
5320 for (i = 0; i < KVM_NR_DB_REGS; i++)
5321 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
5322 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
5323 }
5324
5325 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
5326 vcpu->arch.singlestep_cs =
5327 get_segment_selector(vcpu, VCPU_SREG_CS);
5328 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu);
5329 }
5330
5331 /*
5332 * Trigger an rflags update that will inject or remove the trace
5333 * flags.
5334 */
5335 kvm_set_rflags(vcpu, rflags);
5336
5337 kvm_x86_ops->set_guest_debug(vcpu, dbg);
5338
5339 r = 0;
5340
5341 unlock_out:
5342 vcpu_put(vcpu);
5343
5344 return r;
5345 }
5346
5347 /*
5348 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
5349 * we have asm/x86/processor.h
5350 */
5351 struct fxsave {
5352 u16 cwd;
5353 u16 swd;
5354 u16 twd;
5355 u16 fop;
5356 u64 rip;
5357 u64 rdp;
5358 u32 mxcsr;
5359 u32 mxcsr_mask;
5360 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
5361 #ifdef CONFIG_X86_64
5362 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
5363 #else
5364 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
5365 #endif
5366 };
5367
5368 /*
5369 * Translate a guest virtual address to a guest physical address.
5370 */
5371 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
5372 struct kvm_translation *tr)
5373 {
5374 unsigned long vaddr = tr->linear_address;
5375 gpa_t gpa;
5376 int idx;
5377
5378 vcpu_load(vcpu);
5379 idx = srcu_read_lock(&vcpu->kvm->srcu);
5380 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
5381 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5382 tr->physical_address = gpa;
5383 tr->valid = gpa != UNMAPPED_GVA;
5384 tr->writeable = 1;
5385 tr->usermode = 0;
5386 vcpu_put(vcpu);
5387
5388 return 0;
5389 }
5390
5391 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5392 {
5393 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5394
5395 vcpu_load(vcpu);
5396
5397 memcpy(fpu->fpr, fxsave->st_space, 128);
5398 fpu->fcw = fxsave->cwd;
5399 fpu->fsw = fxsave->swd;
5400 fpu->ftwx = fxsave->twd;
5401 fpu->last_opcode = fxsave->fop;
5402 fpu->last_ip = fxsave->rip;
5403 fpu->last_dp = fxsave->rdp;
5404 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
5405
5406 vcpu_put(vcpu);
5407
5408 return 0;
5409 }
5410
5411 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
5412 {
5413 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
5414
5415 vcpu_load(vcpu);
5416
5417 memcpy(fxsave->st_space, fpu->fpr, 128);
5418 fxsave->cwd = fpu->fcw;
5419 fxsave->swd = fpu->fsw;
5420 fxsave->twd = fpu->ftwx;
5421 fxsave->fop = fpu->last_opcode;
5422 fxsave->rip = fpu->last_ip;
5423 fxsave->rdp = fpu->last_dp;
5424 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
5425
5426 vcpu_put(vcpu);
5427
5428 return 0;
5429 }
5430
5431 void fx_init(struct kvm_vcpu *vcpu)
5432 {
5433 unsigned after_mxcsr_mask;
5434
5435 /*
5436 * Touch the fpu the first time in non atomic context as if
5437 * this is the first fpu instruction the exception handler
5438 * will fire before the instruction returns and it'll have to
5439 * allocate ram with GFP_KERNEL.
5440 */
5441 if (!used_math())
5442 kvm_fx_save(&vcpu->arch.host_fx_image);
5443
5444 /* Initialize guest FPU by resetting ours and saving into guest's */
5445 preempt_disable();
5446 kvm_fx_save(&vcpu->arch.host_fx_image);
5447 kvm_fx_finit();
5448 kvm_fx_save(&vcpu->arch.guest_fx_image);
5449 kvm_fx_restore(&vcpu->arch.host_fx_image);
5450 preempt_enable();
5451
5452 vcpu->arch.cr0 |= X86_CR0_ET;
5453 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
5454 vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
5455 memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
5456 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
5457 }
5458 EXPORT_SYMBOL_GPL(fx_init);
5459
5460 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
5461 {
5462 if (vcpu->guest_fpu_loaded)
5463 return;
5464
5465 vcpu->guest_fpu_loaded = 1;
5466 kvm_fx_save(&vcpu->arch.host_fx_image);
5467 kvm_fx_restore(&vcpu->arch.guest_fx_image);
5468 trace_kvm_fpu(1);
5469 }
5470
5471 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
5472 {
5473 if (!vcpu->guest_fpu_loaded)
5474 return;
5475
5476 vcpu->guest_fpu_loaded = 0;
5477 kvm_fx_save(&vcpu->arch.guest_fx_image);
5478 kvm_fx_restore(&vcpu->arch.host_fx_image);
5479 ++vcpu->stat.fpu_reload;
5480 set_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests);
5481 trace_kvm_fpu(0);
5482 }
5483
5484 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
5485 {
5486 if (vcpu->arch.time_page) {
5487 kvm_release_page_dirty(vcpu->arch.time_page);
5488 vcpu->arch.time_page = NULL;
5489 }
5490
5491 kvm_x86_ops->vcpu_free(vcpu);
5492 }
5493
5494 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
5495 unsigned int id)
5496 {
5497 return kvm_x86_ops->vcpu_create(kvm, id);
5498 }
5499
5500 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
5501 {
5502 int r;
5503
5504 /* We do fxsave: this must be aligned. */
5505 BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
5506
5507 vcpu->arch.mtrr_state.have_fixed = 1;
5508 vcpu_load(vcpu);
5509 r = kvm_arch_vcpu_reset(vcpu);
5510 if (r == 0)
5511 r = kvm_mmu_setup(vcpu);
5512 vcpu_put(vcpu);
5513 if (r < 0)
5514 goto free_vcpu;
5515
5516 return 0;
5517 free_vcpu:
5518 kvm_x86_ops->vcpu_free(vcpu);
5519 return r;
5520 }
5521
5522 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
5523 {
5524 vcpu_load(vcpu);
5525 kvm_mmu_unload(vcpu);
5526 vcpu_put(vcpu);
5527
5528 kvm_x86_ops->vcpu_free(vcpu);
5529 }
5530
5531 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
5532 {
5533 vcpu->arch.nmi_pending = false;
5534 vcpu->arch.nmi_injected = false;
5535
5536 vcpu->arch.switch_db_regs = 0;
5537 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
5538 vcpu->arch.dr6 = DR6_FIXED_1;
5539 vcpu->arch.dr7 = DR7_FIXED_1;
5540
5541 return kvm_x86_ops->vcpu_reset(vcpu);
5542 }
5543
5544 int kvm_arch_hardware_enable(void *garbage)
5545 {
5546 /*
5547 * Since this may be called from a hotplug notifcation,
5548 * we can't get the CPU frequency directly.
5549 */
5550 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
5551 int cpu = raw_smp_processor_id();
5552 per_cpu(cpu_tsc_khz, cpu) = 0;
5553 }
5554
5555 kvm_shared_msr_cpu_online();
5556
5557 return kvm_x86_ops->hardware_enable(garbage);
5558 }
5559
5560 void kvm_arch_hardware_disable(void *garbage)
5561 {
5562 kvm_x86_ops->hardware_disable(garbage);
5563 drop_user_return_notifiers(garbage);
5564 }
5565
5566 int kvm_arch_hardware_setup(void)
5567 {
5568 return kvm_x86_ops->hardware_setup();
5569 }
5570
5571 void kvm_arch_hardware_unsetup(void)
5572 {
5573 kvm_x86_ops->hardware_unsetup();
5574 }
5575
5576 void kvm_arch_check_processor_compat(void *rtn)
5577 {
5578 kvm_x86_ops->check_processor_compatibility(rtn);
5579 }
5580
5581 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
5582 {
5583 struct page *page;
5584 struct kvm *kvm;
5585 int r;
5586
5587 BUG_ON(vcpu->kvm == NULL);
5588 kvm = vcpu->kvm;
5589
5590 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
5591 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
5592 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
5593 else
5594 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
5595
5596 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
5597 if (!page) {
5598 r = -ENOMEM;
5599 goto fail;
5600 }
5601 vcpu->arch.pio_data = page_address(page);
5602
5603 r = kvm_mmu_create(vcpu);
5604 if (r < 0)
5605 goto fail_free_pio_data;
5606
5607 if (irqchip_in_kernel(kvm)) {
5608 r = kvm_create_lapic(vcpu);
5609 if (r < 0)
5610 goto fail_mmu_destroy;
5611 }
5612
5613 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
5614 GFP_KERNEL);
5615 if (!vcpu->arch.mce_banks) {
5616 r = -ENOMEM;
5617 goto fail_free_lapic;
5618 }
5619 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
5620
5621 return 0;
5622 fail_free_lapic:
5623 kvm_free_lapic(vcpu);
5624 fail_mmu_destroy:
5625 kvm_mmu_destroy(vcpu);
5626 fail_free_pio_data:
5627 free_page((unsigned long)vcpu->arch.pio_data);
5628 fail:
5629 return r;
5630 }
5631
5632 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
5633 {
5634 int idx;
5635
5636 kfree(vcpu->arch.mce_banks);
5637 kvm_free_lapic(vcpu);
5638 idx = srcu_read_lock(&vcpu->kvm->srcu);
5639 kvm_mmu_destroy(vcpu);
5640 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5641 free_page((unsigned long)vcpu->arch.pio_data);
5642 }
5643
5644 struct kvm *kvm_arch_create_vm(void)
5645 {
5646 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
5647
5648 if (!kvm)
5649 return ERR_PTR(-ENOMEM);
5650
5651 kvm->arch.aliases = kzalloc(sizeof(struct kvm_mem_aliases), GFP_KERNEL);
5652 if (!kvm->arch.aliases) {
5653 kfree(kvm);
5654 return ERR_PTR(-ENOMEM);
5655 }
5656
5657 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
5658 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
5659
5660 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
5661 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
5662
5663 rdtscll(kvm->arch.vm_init_tsc);
5664
5665 return kvm;
5666 }
5667
5668 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
5669 {
5670 vcpu_load(vcpu);
5671 kvm_mmu_unload(vcpu);
5672 vcpu_put(vcpu);
5673 }
5674
5675 static void kvm_free_vcpus(struct kvm *kvm)
5676 {
5677 unsigned int i;
5678 struct kvm_vcpu *vcpu;
5679
5680 /*
5681 * Unpin any mmu pages first.
5682 */
5683 kvm_for_each_vcpu(i, vcpu, kvm)
5684 kvm_unload_vcpu_mmu(vcpu);
5685 kvm_for_each_vcpu(i, vcpu, kvm)
5686 kvm_arch_vcpu_free(vcpu);
5687
5688 mutex_lock(&kvm->lock);
5689 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
5690 kvm->vcpus[i] = NULL;
5691
5692 atomic_set(&kvm->online_vcpus, 0);
5693 mutex_unlock(&kvm->lock);
5694 }
5695
5696 void kvm_arch_sync_events(struct kvm *kvm)
5697 {
5698 kvm_free_all_assigned_devices(kvm);
5699 }
5700
5701 void kvm_arch_destroy_vm(struct kvm *kvm)
5702 {
5703 kvm_iommu_unmap_guest(kvm);
5704 kvm_free_pit(kvm);
5705 kfree(kvm->arch.vpic);
5706 kfree(kvm->arch.vioapic);
5707 kvm_free_vcpus(kvm);
5708 kvm_free_physmem(kvm);
5709 if (kvm->arch.apic_access_page)
5710 put_page(kvm->arch.apic_access_page);
5711 if (kvm->arch.ept_identity_pagetable)
5712 put_page(kvm->arch.ept_identity_pagetable);
5713 cleanup_srcu_struct(&kvm->srcu);
5714 kfree(kvm->arch.aliases);
5715 kfree(kvm);
5716 }
5717
5718 int kvm_arch_prepare_memory_region(struct kvm *kvm,
5719 struct kvm_memory_slot *memslot,
5720 struct kvm_memory_slot old,
5721 struct kvm_userspace_memory_region *mem,
5722 int user_alloc)
5723 {
5724 int npages = memslot->npages;
5725
5726 /*To keep backward compatibility with older userspace,
5727 *x86 needs to hanlde !user_alloc case.
5728 */
5729 if (!user_alloc) {
5730 if (npages && !old.rmap) {
5731 unsigned long userspace_addr;
5732
5733 down_write(&current->mm->mmap_sem);
5734 userspace_addr = do_mmap(NULL, 0,
5735 npages * PAGE_SIZE,
5736 PROT_READ | PROT_WRITE,
5737 MAP_PRIVATE | MAP_ANONYMOUS,
5738 0);
5739 up_write(&current->mm->mmap_sem);
5740
5741 if (IS_ERR((void *)userspace_addr))
5742 return PTR_ERR((void *)userspace_addr);
5743
5744 memslot->userspace_addr = userspace_addr;
5745 }
5746 }
5747
5748
5749 return 0;
5750 }
5751
5752 void kvm_arch_commit_memory_region(struct kvm *kvm,
5753 struct kvm_userspace_memory_region *mem,
5754 struct kvm_memory_slot old,
5755 int user_alloc)
5756 {
5757
5758 int npages = mem->memory_size >> PAGE_SHIFT;
5759
5760 if (!user_alloc && !old.user_alloc && old.rmap && !npages) {
5761 int ret;
5762
5763 down_write(&current->mm->mmap_sem);
5764 ret = do_munmap(current->mm, old.userspace_addr,
5765 old.npages * PAGE_SIZE);
5766 up_write(&current->mm->mmap_sem);
5767 if (ret < 0)
5768 printk(KERN_WARNING
5769 "kvm_vm_ioctl_set_memory_region: "
5770 "failed to munmap memory\n");
5771 }
5772
5773 spin_lock(&kvm->mmu_lock);
5774 if (!kvm->arch.n_requested_mmu_pages) {
5775 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
5776 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
5777 }
5778
5779 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
5780 spin_unlock(&kvm->mmu_lock);
5781 }
5782
5783 void kvm_arch_flush_shadow(struct kvm *kvm)
5784 {
5785 kvm_mmu_zap_all(kvm);
5786 kvm_reload_remote_mmus(kvm);
5787 }
5788
5789 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
5790 {
5791 return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
5792 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
5793 || vcpu->arch.nmi_pending ||
5794 (kvm_arch_interrupt_allowed(vcpu) &&
5795 kvm_cpu_has_interrupt(vcpu));
5796 }
5797
5798 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
5799 {
5800 int me;
5801 int cpu = vcpu->cpu;
5802
5803 if (waitqueue_active(&vcpu->wq)) {
5804 wake_up_interruptible(&vcpu->wq);
5805 ++vcpu->stat.halt_wakeup;
5806 }
5807
5808 me = get_cpu();
5809 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
5810 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
5811 smp_send_reschedule(cpu);
5812 put_cpu();
5813 }
5814
5815 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
5816 {
5817 return kvm_x86_ops->interrupt_allowed(vcpu);
5818 }
5819
5820 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
5821 {
5822 unsigned long rflags;
5823
5824 rflags = kvm_x86_ops->get_rflags(vcpu);
5825 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5826 rflags &= ~(unsigned long)(X86_EFLAGS_TF | X86_EFLAGS_RF);
5827 return rflags;
5828 }
5829 EXPORT_SYMBOL_GPL(kvm_get_rflags);
5830
5831 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
5832 {
5833 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
5834 vcpu->arch.singlestep_cs ==
5835 get_segment_selector(vcpu, VCPU_SREG_CS) &&
5836 vcpu->arch.singlestep_rip == kvm_rip_read(vcpu))
5837 rflags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
5838 kvm_x86_ops->set_rflags(vcpu, rflags);
5839 }
5840 EXPORT_SYMBOL_GPL(kvm_set_rflags);
5841
5842 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
5843 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
5844 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
5845 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
5846 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
5847 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
5848 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
5849 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
5850 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
5851 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
5852 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree