?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
KVM: x86: missing locking in PIT/IRQCHIP/SET_BSP_CPU ioctl paths
[linux-2.6/btrfs-unstable.git] / arch / x86 / kvm / x86.c
blobe9b0982d2589286dcb52e265d433508e449fefd6
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 #define CREATE_TRACE_POINTS
41 #include "trace.h"
42
43 #include <asm/uaccess.h>
44 #include <asm/msr.h>
45 #include <asm/desc.h>
46 #include <asm/mtrr.h>
47 #include <asm/mce.h>
48
49 #define MAX_IO_MSRS 256
50 #define CR0_RESERVED_BITS \
51 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
52 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
53 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
54 #define CR4_RESERVED_BITS \
55 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
56 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
57 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
58 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
59
60 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
61
62 #define KVM_MAX_MCE_BANKS 32
63 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
64
65 /* EFER defaults:
66 * - enable syscall per default because its emulated by KVM
67 * - enable LME and LMA per default on 64 bit KVM
68 */
69 #ifdef CONFIG_X86_64
70 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
71 #else
72 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
73 #endif
74
75 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
76 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
77
78 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
79 struct kvm_cpuid_entry2 __user *entries);
80 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
81 u32 function, u32 index);
82
83 struct kvm_x86_ops *kvm_x86_ops;
84 EXPORT_SYMBOL_GPL(kvm_x86_ops);
85
86 struct kvm_stats_debugfs_item debugfs_entries[] = {
87 { "pf_fixed", VCPU_STAT(pf_fixed) },
88 { "pf_guest", VCPU_STAT(pf_guest) },
89 { "tlb_flush", VCPU_STAT(tlb_flush) },
90 { "invlpg", VCPU_STAT(invlpg) },
91 { "exits", VCPU_STAT(exits) },
92 { "io_exits", VCPU_STAT(io_exits) },
93 { "mmio_exits", VCPU_STAT(mmio_exits) },
94 { "signal_exits", VCPU_STAT(signal_exits) },
95 { "irq_window", VCPU_STAT(irq_window_exits) },
96 { "nmi_window", VCPU_STAT(nmi_window_exits) },
97 { "halt_exits", VCPU_STAT(halt_exits) },
98 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
99 { "hypercalls", VCPU_STAT(hypercalls) },
100 { "request_irq", VCPU_STAT(request_irq_exits) },
101 { "irq_exits", VCPU_STAT(irq_exits) },
102 { "host_state_reload", VCPU_STAT(host_state_reload) },
103 { "efer_reload", VCPU_STAT(efer_reload) },
104 { "fpu_reload", VCPU_STAT(fpu_reload) },
105 { "insn_emulation", VCPU_STAT(insn_emulation) },
106 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
107 { "irq_injections", VCPU_STAT(irq_injections) },
108 { "nmi_injections", VCPU_STAT(nmi_injections) },
109 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
110 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
111 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
112 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
113 { "mmu_flooded", VM_STAT(mmu_flooded) },
114 { "mmu_recycled", VM_STAT(mmu_recycled) },
115 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
116 { "mmu_unsync", VM_STAT(mmu_unsync) },
117 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
118 { "largepages", VM_STAT(lpages) },
119 { NULL }
120 };
121
122 unsigned long segment_base(u16 selector)
123 {
124 struct descriptor_table gdt;
125 struct desc_struct *d;
126 unsigned long table_base;
127 unsigned long v;
128
129 if (selector == 0)
130 return 0;
131
132 asm("sgdt %0" : "=m"(gdt));
133 table_base = gdt.base;
134
135 if (selector & 4) { /* from ldt */
136 u16 ldt_selector;
137
138 asm("sldt %0" : "=g"(ldt_selector));
139 table_base = segment_base(ldt_selector);
140 }
141 d = (struct desc_struct *)(table_base + (selector & ~7));
142 v = d->base0 | ((unsigned long)d->base1 << 16) |
143 ((unsigned long)d->base2 << 24);
144 #ifdef CONFIG_X86_64
145 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
146 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
147 #endif
148 return v;
149 }
150 EXPORT_SYMBOL_GPL(segment_base);
151
152 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
153 {
154 if (irqchip_in_kernel(vcpu->kvm))
155 return vcpu->arch.apic_base;
156 else
157 return vcpu->arch.apic_base;
158 }
159 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
160
161 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
162 {
163 /* TODO: reserve bits check */
164 if (irqchip_in_kernel(vcpu->kvm))
165 kvm_lapic_set_base(vcpu, data);
166 else
167 vcpu->arch.apic_base = data;
168 }
169 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
170
171 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
172 {
173 WARN_ON(vcpu->arch.exception.pending);
174 vcpu->arch.exception.pending = true;
175 vcpu->arch.exception.has_error_code = false;
176 vcpu->arch.exception.nr = nr;
177 }
178 EXPORT_SYMBOL_GPL(kvm_queue_exception);
179
180 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
181 u32 error_code)
182 {
183 ++vcpu->stat.pf_guest;
184
185 if (vcpu->arch.exception.pending) {
186 switch(vcpu->arch.exception.nr) {
187 case DF_VECTOR:
188 /* triple fault -> shutdown */
189 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
190 return;
191 case PF_VECTOR:
192 vcpu->arch.exception.nr = DF_VECTOR;
193 vcpu->arch.exception.error_code = 0;
194 return;
195 default:
196 /* replace previous exception with a new one in a hope
197 that instruction re-execution will regenerate lost
198 exception */
199 vcpu->arch.exception.pending = false;
200 break;
201 }
202 }
203 vcpu->arch.cr2 = addr;
204 kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
205 }
206
207 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
208 {
209 vcpu->arch.nmi_pending = 1;
210 }
211 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
212
213 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
214 {
215 WARN_ON(vcpu->arch.exception.pending);
216 vcpu->arch.exception.pending = true;
217 vcpu->arch.exception.has_error_code = true;
218 vcpu->arch.exception.nr = nr;
219 vcpu->arch.exception.error_code = error_code;
220 }
221 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
222
223 static void __queue_exception(struct kvm_vcpu *vcpu)
224 {
225 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
226 vcpu->arch.exception.has_error_code,
227 vcpu->arch.exception.error_code);
228 }
229
230 /*
231 * Load the pae pdptrs. Return true is they are all valid.
232 */
233 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
234 {
235 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
236 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
237 int i;
238 int ret;
239 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
240
241 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
242 offset * sizeof(u64), sizeof(pdpte));
243 if (ret < 0) {
244 ret = 0;
245 goto out;
246 }
247 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
248 if (is_present_gpte(pdpte[i]) &&
249 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
250 ret = 0;
251 goto out;
252 }
253 }
254 ret = 1;
255
256 memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
257 __set_bit(VCPU_EXREG_PDPTR,
258 (unsigned long *)&vcpu->arch.regs_avail);
259 __set_bit(VCPU_EXREG_PDPTR,
260 (unsigned long *)&vcpu->arch.regs_dirty);
261 out:
262
263 return ret;
264 }
265 EXPORT_SYMBOL_GPL(load_pdptrs);
266
267 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
268 {
269 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
270 bool changed = true;
271 int r;
272
273 if (is_long_mode(vcpu) || !is_pae(vcpu))
274 return false;
275
276 if (!test_bit(VCPU_EXREG_PDPTR,
277 (unsigned long *)&vcpu->arch.regs_avail))
278 return true;
279
280 r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
281 if (r < 0)
282 goto out;
283 changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
284 out:
285
286 return changed;
287 }
288
289 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
290 {
291 if (cr0 & CR0_RESERVED_BITS) {
292 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
293 cr0, vcpu->arch.cr0);
294 kvm_inject_gp(vcpu, 0);
295 return;
296 }
297
298 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
299 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
300 kvm_inject_gp(vcpu, 0);
301 return;
302 }
303
304 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
305 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
306 "and a clear PE flag\n");
307 kvm_inject_gp(vcpu, 0);
308 return;
309 }
310
311 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
312 #ifdef CONFIG_X86_64
313 if ((vcpu->arch.shadow_efer & EFER_LME)) {
314 int cs_db, cs_l;
315
316 if (!is_pae(vcpu)) {
317 printk(KERN_DEBUG "set_cr0: #GP, start paging "
318 "in long mode while PAE is disabled\n");
319 kvm_inject_gp(vcpu, 0);
320 return;
321 }
322 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
323 if (cs_l) {
324 printk(KERN_DEBUG "set_cr0: #GP, start paging "
325 "in long mode while CS.L == 1\n");
326 kvm_inject_gp(vcpu, 0);
327 return;
328
329 }
330 } else
331 #endif
332 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
333 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
334 "reserved bits\n");
335 kvm_inject_gp(vcpu, 0);
336 return;
337 }
338
339 }
340
341 kvm_x86_ops->set_cr0(vcpu, cr0);
342 vcpu->arch.cr0 = cr0;
343
344 kvm_mmu_reset_context(vcpu);
345 return;
346 }
347 EXPORT_SYMBOL_GPL(kvm_set_cr0);
348
349 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
350 {
351 kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
352 }
353 EXPORT_SYMBOL_GPL(kvm_lmsw);
354
355 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
356 {
357 unsigned long old_cr4 = vcpu->arch.cr4;
358 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
359
360 if (cr4 & CR4_RESERVED_BITS) {
361 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
362 kvm_inject_gp(vcpu, 0);
363 return;
364 }
365
366 if (is_long_mode(vcpu)) {
367 if (!(cr4 & X86_CR4_PAE)) {
368 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
369 "in long mode\n");
370 kvm_inject_gp(vcpu, 0);
371 return;
372 }
373 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
374 && ((cr4 ^ old_cr4) & pdptr_bits)
375 && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
376 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
377 kvm_inject_gp(vcpu, 0);
378 return;
379 }
380
381 if (cr4 & X86_CR4_VMXE) {
382 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
383 kvm_inject_gp(vcpu, 0);
384 return;
385 }
386 kvm_x86_ops->set_cr4(vcpu, cr4);
387 vcpu->arch.cr4 = cr4;
388 vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
389 kvm_mmu_reset_context(vcpu);
390 }
391 EXPORT_SYMBOL_GPL(kvm_set_cr4);
392
393 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
394 {
395 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
396 kvm_mmu_sync_roots(vcpu);
397 kvm_mmu_flush_tlb(vcpu);
398 return;
399 }
400
401 if (is_long_mode(vcpu)) {
402 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
403 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
404 kvm_inject_gp(vcpu, 0);
405 return;
406 }
407 } else {
408 if (is_pae(vcpu)) {
409 if (cr3 & CR3_PAE_RESERVED_BITS) {
410 printk(KERN_DEBUG
411 "set_cr3: #GP, reserved bits\n");
412 kvm_inject_gp(vcpu, 0);
413 return;
414 }
415 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
416 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
417 "reserved bits\n");
418 kvm_inject_gp(vcpu, 0);
419 return;
420 }
421 }
422 /*
423 * We don't check reserved bits in nonpae mode, because
424 * this isn't enforced, and VMware depends on this.
425 */
426 }
427
428 /*
429 * Does the new cr3 value map to physical memory? (Note, we
430 * catch an invalid cr3 even in real-mode, because it would
431 * cause trouble later on when we turn on paging anyway.)
432 *
433 * A real CPU would silently accept an invalid cr3 and would
434 * attempt to use it - with largely undefined (and often hard
435 * to debug) behavior on the guest side.
436 */
437 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
438 kvm_inject_gp(vcpu, 0);
439 else {
440 vcpu->arch.cr3 = cr3;
441 vcpu->arch.mmu.new_cr3(vcpu);
442 }
443 }
444 EXPORT_SYMBOL_GPL(kvm_set_cr3);
445
446 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
447 {
448 if (cr8 & CR8_RESERVED_BITS) {
449 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
450 kvm_inject_gp(vcpu, 0);
451 return;
452 }
453 if (irqchip_in_kernel(vcpu->kvm))
454 kvm_lapic_set_tpr(vcpu, cr8);
455 else
456 vcpu->arch.cr8 = cr8;
457 }
458 EXPORT_SYMBOL_GPL(kvm_set_cr8);
459
460 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
461 {
462 if (irqchip_in_kernel(vcpu->kvm))
463 return kvm_lapic_get_cr8(vcpu);
464 else
465 return vcpu->arch.cr8;
466 }
467 EXPORT_SYMBOL_GPL(kvm_get_cr8);
468
469 static inline u32 bit(int bitno)
470 {
471 return 1 << (bitno & 31);
472 }
473
474 /*
475 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
476 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
477 *
478 * This list is modified at module load time to reflect the
479 * capabilities of the host cpu.
480 */
481 static u32 msrs_to_save[] = {
482 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
483 MSR_K6_STAR,
484 #ifdef CONFIG_X86_64
485 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
486 #endif
487 MSR_IA32_TSC, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
488 MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
489 };
490
491 static unsigned num_msrs_to_save;
492
493 static u32 emulated_msrs[] = {
494 MSR_IA32_MISC_ENABLE,
495 };
496
497 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
498 {
499 if (efer & efer_reserved_bits) {
500 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
501 efer);
502 kvm_inject_gp(vcpu, 0);
503 return;
504 }
505
506 if (is_paging(vcpu)
507 && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
508 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
509 kvm_inject_gp(vcpu, 0);
510 return;
511 }
512
513 if (efer & EFER_FFXSR) {
514 struct kvm_cpuid_entry2 *feat;
515
516 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
517 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
518 printk(KERN_DEBUG "set_efer: #GP, enable FFXSR w/o CPUID capability\n");
519 kvm_inject_gp(vcpu, 0);
520 return;
521 }
522 }
523
524 if (efer & EFER_SVME) {
525 struct kvm_cpuid_entry2 *feat;
526
527 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
528 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
529 printk(KERN_DEBUG "set_efer: #GP, enable SVM w/o SVM\n");
530 kvm_inject_gp(vcpu, 0);
531 return;
532 }
533 }
534
535 kvm_x86_ops->set_efer(vcpu, efer);
536
537 efer &= ~EFER_LMA;
538 efer |= vcpu->arch.shadow_efer & EFER_LMA;
539
540 vcpu->arch.shadow_efer = efer;
541
542 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
543 kvm_mmu_reset_context(vcpu);
544 }
545
546 void kvm_enable_efer_bits(u64 mask)
547 {
548 efer_reserved_bits &= ~mask;
549 }
550 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
551
552
553 /*
554 * Writes msr value into into the appropriate "register".
555 * Returns 0 on success, non-0 otherwise.
556 * Assumes vcpu_load() was already called.
557 */
558 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
559 {
560 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
561 }
562
563 /*
564 * Adapt set_msr() to msr_io()'s calling convention
565 */
566 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
567 {
568 return kvm_set_msr(vcpu, index, *data);
569 }
570
571 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
572 {
573 static int version;
574 struct pvclock_wall_clock wc;
575 struct timespec now, sys, boot;
576
577 if (!wall_clock)
578 return;
579
580 version++;
581
582 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
583
584 /*
585 * The guest calculates current wall clock time by adding
586 * system time (updated by kvm_write_guest_time below) to the
587 * wall clock specified here. guest system time equals host
588 * system time for us, thus we must fill in host boot time here.
589 */
590 now = current_kernel_time();
591 ktime_get_ts(&sys);
592 boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
593
594 wc.sec = boot.tv_sec;
595 wc.nsec = boot.tv_nsec;
596 wc.version = version;
597
598 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
599
600 version++;
601 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
602 }
603
604 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
605 {
606 uint32_t quotient, remainder;
607
608 /* Don't try to replace with do_div(), this one calculates
609 * "(dividend << 32) / divisor" */
610 __asm__ ( "divl %4"
611 : "=a" (quotient), "=d" (remainder)
612 : "0" (0), "1" (dividend), "r" (divisor) );
613 return quotient;
614 }
615
616 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
617 {
618 uint64_t nsecs = 1000000000LL;
619 int32_t shift = 0;
620 uint64_t tps64;
621 uint32_t tps32;
622
623 tps64 = tsc_khz * 1000LL;
624 while (tps64 > nsecs*2) {
625 tps64 >>= 1;
626 shift--;
627 }
628
629 tps32 = (uint32_t)tps64;
630 while (tps32 <= (uint32_t)nsecs) {
631 tps32 <<= 1;
632 shift++;
633 }
634
635 hv_clock->tsc_shift = shift;
636 hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
637
638 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
639 __func__, tsc_khz, hv_clock->tsc_shift,
640 hv_clock->tsc_to_system_mul);
641 }
642
643 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
644
645 static void kvm_write_guest_time(struct kvm_vcpu *v)
646 {
647 struct timespec ts;
648 unsigned long flags;
649 struct kvm_vcpu_arch *vcpu = &v->arch;
650 void *shared_kaddr;
651 unsigned long this_tsc_khz;
652
653 if ((!vcpu->time_page))
654 return;
655
656 this_tsc_khz = get_cpu_var(cpu_tsc_khz);
657 if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
658 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
659 vcpu->hv_clock_tsc_khz = this_tsc_khz;
660 }
661 put_cpu_var(cpu_tsc_khz);
662
663 /* Keep irq disabled to prevent changes to the clock */
664 local_irq_save(flags);
665 kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
666 ktime_get_ts(&ts);
667 local_irq_restore(flags);
668
669 /* With all the info we got, fill in the values */
670
671 vcpu->hv_clock.system_time = ts.tv_nsec +
672 (NSEC_PER_SEC * (u64)ts.tv_sec);
673 /*
674 * The interface expects us to write an even number signaling that the
675 * update is finished. Since the guest won't see the intermediate
676 * state, we just increase by 2 at the end.
677 */
678 vcpu->hv_clock.version += 2;
679
680 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
681
682 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
683 sizeof(vcpu->hv_clock));
684
685 kunmap_atomic(shared_kaddr, KM_USER0);
686
687 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
688 }
689
690 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
691 {
692 struct kvm_vcpu_arch *vcpu = &v->arch;
693
694 if (!vcpu->time_page)
695 return 0;
696 set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
697 return 1;
698 }
699
700 static bool msr_mtrr_valid(unsigned msr)
701 {
702 switch (msr) {
703 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
704 case MSR_MTRRfix64K_00000:
705 case MSR_MTRRfix16K_80000:
706 case MSR_MTRRfix16K_A0000:
707 case MSR_MTRRfix4K_C0000:
708 case MSR_MTRRfix4K_C8000:
709 case MSR_MTRRfix4K_D0000:
710 case MSR_MTRRfix4K_D8000:
711 case MSR_MTRRfix4K_E0000:
712 case MSR_MTRRfix4K_E8000:
713 case MSR_MTRRfix4K_F0000:
714 case MSR_MTRRfix4K_F8000:
715 case MSR_MTRRdefType:
716 case MSR_IA32_CR_PAT:
717 return true;
718 case 0x2f8:
719 return true;
720 }
721 return false;
722 }
723
724 static bool valid_pat_type(unsigned t)
725 {
726 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
727 }
728
729 static bool valid_mtrr_type(unsigned t)
730 {
731 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
732 }
733
734 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
735 {
736 int i;
737
738 if (!msr_mtrr_valid(msr))
739 return false;
740
741 if (msr == MSR_IA32_CR_PAT) {
742 for (i = 0; i < 8; i++)
743 if (!valid_pat_type((data >> (i * 8)) & 0xff))
744 return false;
745 return true;
746 } else if (msr == MSR_MTRRdefType) {
747 if (data & ~0xcff)
748 return false;
749 return valid_mtrr_type(data & 0xff);
750 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
751 for (i = 0; i < 8 ; i++)
752 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
753 return false;
754 return true;
755 }
756
757 /* variable MTRRs */
758 return valid_mtrr_type(data & 0xff);
759 }
760
761 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
762 {
763 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
764
765 if (!mtrr_valid(vcpu, msr, data))
766 return 1;
767
768 if (msr == MSR_MTRRdefType) {
769 vcpu->arch.mtrr_state.def_type = data;
770 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
771 } else if (msr == MSR_MTRRfix64K_00000)
772 p[0] = data;
773 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
774 p[1 + msr - MSR_MTRRfix16K_80000] = data;
775 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
776 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
777 else if (msr == MSR_IA32_CR_PAT)
778 vcpu->arch.pat = data;
779 else { /* Variable MTRRs */
780 int idx, is_mtrr_mask;
781 u64 *pt;
782
783 idx = (msr - 0x200) / 2;
784 is_mtrr_mask = msr - 0x200 - 2 * idx;
785 if (!is_mtrr_mask)
786 pt =
787 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
788 else
789 pt =
790 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
791 *pt = data;
792 }
793
794 kvm_mmu_reset_context(vcpu);
795 return 0;
796 }
797
798 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
799 {
800 u64 mcg_cap = vcpu->arch.mcg_cap;
801 unsigned bank_num = mcg_cap & 0xff;
802
803 switch (msr) {
804 case MSR_IA32_MCG_STATUS:
805 vcpu->arch.mcg_status = data;
806 break;
807 case MSR_IA32_MCG_CTL:
808 if (!(mcg_cap & MCG_CTL_P))
809 return 1;
810 if (data != 0 && data != ~(u64)0)
811 return -1;
812 vcpu->arch.mcg_ctl = data;
813 break;
814 default:
815 if (msr >= MSR_IA32_MC0_CTL &&
816 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
817 u32 offset = msr - MSR_IA32_MC0_CTL;
818 /* only 0 or all 1s can be written to IA32_MCi_CTL */
819 if ((offset & 0x3) == 0 &&
820 data != 0 && data != ~(u64)0)
821 return -1;
822 vcpu->arch.mce_banks[offset] = data;
823 break;
824 }
825 return 1;
826 }
827 return 0;
828 }
829
830 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
831 {
832 switch (msr) {
833 case MSR_EFER:
834 set_efer(vcpu, data);
835 break;
836 case MSR_IA32_DEBUGCTLMSR:
837 if (!data) {
838 /* We support the non-activated case already */
839 break;
840 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
841 /* Values other than LBR and BTF are vendor-specific,
842 thus reserved and should throw a #GP */
843 return 1;
844 }
845 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
846 __func__, data);
847 break;
848 case MSR_IA32_UCODE_REV:
849 case MSR_IA32_UCODE_WRITE:
850 case MSR_VM_HSAVE_PA:
851 break;
852 case 0x200 ... 0x2ff:
853 return set_msr_mtrr(vcpu, msr, data);
854 case MSR_IA32_APICBASE:
855 kvm_set_apic_base(vcpu, data);
856 break;
857 case MSR_IA32_MISC_ENABLE:
858 vcpu->arch.ia32_misc_enable_msr = data;
859 break;
860 case MSR_KVM_WALL_CLOCK:
861 vcpu->kvm->arch.wall_clock = data;
862 kvm_write_wall_clock(vcpu->kvm, data);
863 break;
864 case MSR_KVM_SYSTEM_TIME: {
865 if (vcpu->arch.time_page) {
866 kvm_release_page_dirty(vcpu->arch.time_page);
867 vcpu->arch.time_page = NULL;
868 }
869
870 vcpu->arch.time = data;
871
872 /* we verify if the enable bit is set... */
873 if (!(data & 1))
874 break;
875
876 /* ...but clean it before doing the actual write */
877 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
878
879 vcpu->arch.time_page =
880 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
881
882 if (is_error_page(vcpu->arch.time_page)) {
883 kvm_release_page_clean(vcpu->arch.time_page);
884 vcpu->arch.time_page = NULL;
885 }
886
887 kvm_request_guest_time_update(vcpu);
888 break;
889 }
890 case MSR_IA32_MCG_CTL:
891 case MSR_IA32_MCG_STATUS:
892 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
893 return set_msr_mce(vcpu, msr, data);
894
895 /* Performance counters are not protected by a CPUID bit,
896 * so we should check all of them in the generic path for the sake of
897 * cross vendor migration.
898 * Writing a zero into the event select MSRs disables them,
899 * which we perfectly emulate ;-). Any other value should be at least
900 * reported, some guests depend on them.
901 */
902 case MSR_P6_EVNTSEL0:
903 case MSR_P6_EVNTSEL1:
904 case MSR_K7_EVNTSEL0:
905 case MSR_K7_EVNTSEL1:
906 case MSR_K7_EVNTSEL2:
907 case MSR_K7_EVNTSEL3:
908 if (data != 0)
909 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
910 "0x%x data 0x%llx\n", msr, data);
911 break;
912 /* at least RHEL 4 unconditionally writes to the perfctr registers,
913 * so we ignore writes to make it happy.
914 */
915 case MSR_P6_PERFCTR0:
916 case MSR_P6_PERFCTR1:
917 case MSR_K7_PERFCTR0:
918 case MSR_K7_PERFCTR1:
919 case MSR_K7_PERFCTR2:
920 case MSR_K7_PERFCTR3:
921 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
922 "0x%x data 0x%llx\n", msr, data);
923 break;
924 default:
925 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", msr, data);
926 return 1;
927 }
928 return 0;
929 }
930 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
931
932
933 /*
934 * Reads an msr value (of 'msr_index') into 'pdata'.
935 * Returns 0 on success, non-0 otherwise.
936 * Assumes vcpu_load() was already called.
937 */
938 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
939 {
940 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
941 }
942
943 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
944 {
945 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
946
947 if (!msr_mtrr_valid(msr))
948 return 1;
949
950 if (msr == MSR_MTRRdefType)
951 *pdata = vcpu->arch.mtrr_state.def_type +
952 (vcpu->arch.mtrr_state.enabled << 10);
953 else if (msr == MSR_MTRRfix64K_00000)
954 *pdata = p[0];
955 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
956 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
957 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
958 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
959 else if (msr == MSR_IA32_CR_PAT)
960 *pdata = vcpu->arch.pat;
961 else { /* Variable MTRRs */
962 int idx, is_mtrr_mask;
963 u64 *pt;
964
965 idx = (msr - 0x200) / 2;
966 is_mtrr_mask = msr - 0x200 - 2 * idx;
967 if (!is_mtrr_mask)
968 pt =
969 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
970 else
971 pt =
972 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
973 *pdata = *pt;
974 }
975
976 return 0;
977 }
978
979 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
980 {
981 u64 data;
982 u64 mcg_cap = vcpu->arch.mcg_cap;
983 unsigned bank_num = mcg_cap & 0xff;
984
985 switch (msr) {
986 case MSR_IA32_P5_MC_ADDR:
987 case MSR_IA32_P5_MC_TYPE:
988 data = 0;
989 break;
990 case MSR_IA32_MCG_CAP:
991 data = vcpu->arch.mcg_cap;
992 break;
993 case MSR_IA32_MCG_CTL:
994 if (!(mcg_cap & MCG_CTL_P))
995 return 1;
996 data = vcpu->arch.mcg_ctl;
997 break;
998 case MSR_IA32_MCG_STATUS:
999 data = vcpu->arch.mcg_status;
1000 break;
1001 default:
1002 if (msr >= MSR_IA32_MC0_CTL &&
1003 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1004 u32 offset = msr - MSR_IA32_MC0_CTL;
1005 data = vcpu->arch.mce_banks[offset];
1006 break;
1007 }
1008 return 1;
1009 }
1010 *pdata = data;
1011 return 0;
1012 }
1013
1014 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1015 {
1016 u64 data;
1017
1018 switch (msr) {
1019 case MSR_IA32_PLATFORM_ID:
1020 case MSR_IA32_UCODE_REV:
1021 case MSR_IA32_EBL_CR_POWERON:
1022 case MSR_IA32_DEBUGCTLMSR:
1023 case MSR_IA32_LASTBRANCHFROMIP:
1024 case MSR_IA32_LASTBRANCHTOIP:
1025 case MSR_IA32_LASTINTFROMIP:
1026 case MSR_IA32_LASTINTTOIP:
1027 case MSR_K8_SYSCFG:
1028 case MSR_K7_HWCR:
1029 case MSR_VM_HSAVE_PA:
1030 case MSR_P6_EVNTSEL0:
1031 case MSR_P6_EVNTSEL1:
1032 case MSR_K7_EVNTSEL0:
1033 data = 0;
1034 break;
1035 case MSR_MTRRcap:
1036 data = 0x500 | KVM_NR_VAR_MTRR;
1037 break;
1038 case 0x200 ... 0x2ff:
1039 return get_msr_mtrr(vcpu, msr, pdata);
1040 case 0xcd: /* fsb frequency */
1041 data = 3;
1042 break;
1043 case MSR_IA32_APICBASE:
1044 data = kvm_get_apic_base(vcpu);
1045 break;
1046 case MSR_IA32_MISC_ENABLE:
1047 data = vcpu->arch.ia32_misc_enable_msr;
1048 break;
1049 case MSR_IA32_PERF_STATUS:
1050 /* TSC increment by tick */
1051 data = 1000ULL;
1052 /* CPU multiplier */
1053 data |= (((uint64_t)4ULL) << 40);
1054 break;
1055 case MSR_EFER:
1056 data = vcpu->arch.shadow_efer;
1057 break;
1058 case MSR_KVM_WALL_CLOCK:
1059 data = vcpu->kvm->arch.wall_clock;
1060 break;
1061 case MSR_KVM_SYSTEM_TIME:
1062 data = vcpu->arch.time;
1063 break;
1064 case MSR_IA32_P5_MC_ADDR:
1065 case MSR_IA32_P5_MC_TYPE:
1066 case MSR_IA32_MCG_CAP:
1067 case MSR_IA32_MCG_CTL:
1068 case MSR_IA32_MCG_STATUS:
1069 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1070 return get_msr_mce(vcpu, msr, pdata);
1071 default:
1072 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1073 return 1;
1074 }
1075 *pdata = data;
1076 return 0;
1077 }
1078 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1079
1080 /*
1081 * Read or write a bunch of msrs. All parameters are kernel addresses.
1082 *
1083 * @return number of msrs set successfully.
1084 */
1085 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1086 struct kvm_msr_entry *entries,
1087 int (*do_msr)(struct kvm_vcpu *vcpu,
1088 unsigned index, u64 *data))
1089 {
1090 int i;
1091
1092 vcpu_load(vcpu);
1093
1094 down_read(&vcpu->kvm->slots_lock);
1095 for (i = 0; i < msrs->nmsrs; ++i)
1096 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1097 break;
1098 up_read(&vcpu->kvm->slots_lock);
1099
1100 vcpu_put(vcpu);
1101
1102 return i;
1103 }
1104
1105 /*
1106 * Read or write a bunch of msrs. Parameters are user addresses.
1107 *
1108 * @return number of msrs set successfully.
1109 */
1110 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1111 int (*do_msr)(struct kvm_vcpu *vcpu,
1112 unsigned index, u64 *data),
1113 int writeback)
1114 {
1115 struct kvm_msrs msrs;
1116 struct kvm_msr_entry *entries;
1117 int r, n;
1118 unsigned size;
1119
1120 r = -EFAULT;
1121 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1122 goto out;
1123
1124 r = -E2BIG;
1125 if (msrs.nmsrs >= MAX_IO_MSRS)
1126 goto out;
1127
1128 r = -ENOMEM;
1129 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1130 entries = vmalloc(size);
1131 if (!entries)
1132 goto out;
1133
1134 r = -EFAULT;
1135 if (copy_from_user(entries, user_msrs->entries, size))
1136 goto out_free;
1137
1138 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1139 if (r < 0)
1140 goto out_free;
1141
1142 r = -EFAULT;
1143 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1144 goto out_free;
1145
1146 r = n;
1147
1148 out_free:
1149 vfree(entries);
1150 out:
1151 return r;
1152 }
1153
1154 int kvm_dev_ioctl_check_extension(long ext)
1155 {
1156 int r;
1157
1158 switch (ext) {
1159 case KVM_CAP_IRQCHIP:
1160 case KVM_CAP_HLT:
1161 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1162 case KVM_CAP_SET_TSS_ADDR:
1163 case KVM_CAP_EXT_CPUID:
1164 case KVM_CAP_CLOCKSOURCE:
1165 case KVM_CAP_PIT:
1166 case KVM_CAP_NOP_IO_DELAY:
1167 case KVM_CAP_MP_STATE:
1168 case KVM_CAP_SYNC_MMU:
1169 case KVM_CAP_REINJECT_CONTROL:
1170 case KVM_CAP_IRQ_INJECT_STATUS:
1171 case KVM_CAP_ASSIGN_DEV_IRQ:
1172 case KVM_CAP_IRQFD:
1173 case KVM_CAP_PIT2:
1174 r = 1;
1175 break;
1176 case KVM_CAP_COALESCED_MMIO:
1177 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1178 break;
1179 case KVM_CAP_VAPIC:
1180 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1181 break;
1182 case KVM_CAP_NR_VCPUS:
1183 r = KVM_MAX_VCPUS;
1184 break;
1185 case KVM_CAP_NR_MEMSLOTS:
1186 r = KVM_MEMORY_SLOTS;
1187 break;
1188 case KVM_CAP_PV_MMU:
1189 r = !tdp_enabled;
1190 break;
1191 case KVM_CAP_IOMMU:
1192 r = iommu_found();
1193 break;
1194 case KVM_CAP_MCE:
1195 r = KVM_MAX_MCE_BANKS;
1196 break;
1197 default:
1198 r = 0;
1199 break;
1200 }
1201 return r;
1202
1203 }
1204
1205 long kvm_arch_dev_ioctl(struct file *filp,
1206 unsigned int ioctl, unsigned long arg)
1207 {
1208 void __user *argp = (void __user *)arg;
1209 long r;
1210
1211 switch (ioctl) {
1212 case KVM_GET_MSR_INDEX_LIST: {
1213 struct kvm_msr_list __user *user_msr_list = argp;
1214 struct kvm_msr_list msr_list;
1215 unsigned n;
1216
1217 r = -EFAULT;
1218 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1219 goto out;
1220 n = msr_list.nmsrs;
1221 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1222 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1223 goto out;
1224 r = -E2BIG;
1225 if (n < msr_list.nmsrs)
1226 goto out;
1227 r = -EFAULT;
1228 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1229 num_msrs_to_save * sizeof(u32)))
1230 goto out;
1231 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1232 &emulated_msrs,
1233 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1234 goto out;
1235 r = 0;
1236 break;
1237 }
1238 case KVM_GET_SUPPORTED_CPUID: {
1239 struct kvm_cpuid2 __user *cpuid_arg = argp;
1240 struct kvm_cpuid2 cpuid;
1241
1242 r = -EFAULT;
1243 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1244 goto out;
1245 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1246 cpuid_arg->entries);
1247 if (r)
1248 goto out;
1249
1250 r = -EFAULT;
1251 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1252 goto out;
1253 r = 0;
1254 break;
1255 }
1256 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1257 u64 mce_cap;
1258
1259 mce_cap = KVM_MCE_CAP_SUPPORTED;
1260 r = -EFAULT;
1261 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1262 goto out;
1263 r = 0;
1264 break;
1265 }
1266 default:
1267 r = -EINVAL;
1268 }
1269 out:
1270 return r;
1271 }
1272
1273 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1274 {
1275 kvm_x86_ops->vcpu_load(vcpu, cpu);
1276 kvm_request_guest_time_update(vcpu);
1277 }
1278
1279 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1280 {
1281 kvm_x86_ops->vcpu_put(vcpu);
1282 kvm_put_guest_fpu(vcpu);
1283 }
1284
1285 static int is_efer_nx(void)
1286 {
1287 unsigned long long efer = 0;
1288
1289 rdmsrl_safe(MSR_EFER, &efer);
1290 return efer & EFER_NX;
1291 }
1292
1293 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1294 {
1295 int i;
1296 struct kvm_cpuid_entry2 *e, *entry;
1297
1298 entry = NULL;
1299 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1300 e = &vcpu->arch.cpuid_entries[i];
1301 if (e->function == 0x80000001) {
1302 entry = e;
1303 break;
1304 }
1305 }
1306 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1307 entry->edx &= ~(1 << 20);
1308 printk(KERN_INFO "kvm: guest NX capability removed\n");
1309 }
1310 }
1311
1312 /* when an old userspace process fills a new kernel module */
1313 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1314 struct kvm_cpuid *cpuid,
1315 struct kvm_cpuid_entry __user *entries)
1316 {
1317 int r, i;
1318 struct kvm_cpuid_entry *cpuid_entries;
1319
1320 r = -E2BIG;
1321 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1322 goto out;
1323 r = -ENOMEM;
1324 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1325 if (!cpuid_entries)
1326 goto out;
1327 r = -EFAULT;
1328 if (copy_from_user(cpuid_entries, entries,
1329 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1330 goto out_free;
1331 for (i = 0; i < cpuid->nent; i++) {
1332 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1333 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1334 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1335 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1336 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1337 vcpu->arch.cpuid_entries[i].index = 0;
1338 vcpu->arch.cpuid_entries[i].flags = 0;
1339 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1340 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1341 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1342 }
1343 vcpu->arch.cpuid_nent = cpuid->nent;
1344 cpuid_fix_nx_cap(vcpu);
1345 r = 0;
1346
1347 out_free:
1348 vfree(cpuid_entries);
1349 out:
1350 return r;
1351 }
1352
1353 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1354 struct kvm_cpuid2 *cpuid,
1355 struct kvm_cpuid_entry2 __user *entries)
1356 {
1357 int r;
1358
1359 r = -E2BIG;
1360 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1361 goto out;
1362 r = -EFAULT;
1363 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1364 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1365 goto out;
1366 vcpu->arch.cpuid_nent = cpuid->nent;
1367 return 0;
1368
1369 out:
1370 return r;
1371 }
1372
1373 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1374 struct kvm_cpuid2 *cpuid,
1375 struct kvm_cpuid_entry2 __user *entries)
1376 {
1377 int r;
1378
1379 r = -E2BIG;
1380 if (cpuid->nent < vcpu->arch.cpuid_nent)
1381 goto out;
1382 r = -EFAULT;
1383 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1384 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1385 goto out;
1386 return 0;
1387
1388 out:
1389 cpuid->nent = vcpu->arch.cpuid_nent;
1390 return r;
1391 }
1392
1393 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1394 u32 index)
1395 {
1396 entry->function = function;
1397 entry->index = index;
1398 cpuid_count(entry->function, entry->index,
1399 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1400 entry->flags = 0;
1401 }
1402
1403 #define F(x) bit(X86_FEATURE_##x)
1404
1405 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1406 u32 index, int *nent, int maxnent)
1407 {
1408 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1409 #ifdef CONFIG_X86_64
1410 unsigned f_lm = F(LM);
1411 #else
1412 unsigned f_lm = 0;
1413 #endif
1414
1415 /* cpuid 1.edx */
1416 const u32 kvm_supported_word0_x86_features =
1417 F(FPU) | F(VME) | F(DE) | F(PSE) |
1418 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1419 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1420 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1421 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1422 0 /* Reserved, DS, ACPI */ | F(MMX) |
1423 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1424 0 /* HTT, TM, Reserved, PBE */;
1425 /* cpuid 0x80000001.edx */
1426 const u32 kvm_supported_word1_x86_features =
1427 F(FPU) | F(VME) | F(DE) | F(PSE) |
1428 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1429 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1430 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1431 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1432 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1433 F(FXSR) | F(FXSR_OPT) | 0 /* GBPAGES */ | 0 /* RDTSCP */ |
1434 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1435 /* cpuid 1.ecx */
1436 const u32 kvm_supported_word4_x86_features =
1437 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1438 0 /* DS-CPL, VMX, SMX, EST */ |
1439 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1440 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1441 0 /* Reserved, DCA */ | F(XMM4_1) |
1442 F(XMM4_2) | 0 /* x2APIC */ | F(MOVBE) | F(POPCNT) |
1443 0 /* Reserved, XSAVE, OSXSAVE */;
1444 /* cpuid 0x80000001.ecx */
1445 const u32 kvm_supported_word6_x86_features =
1446 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1447 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1448 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1449 0 /* SKINIT */ | 0 /* WDT */;
1450
1451 /* all calls to cpuid_count() should be made on the same cpu */
1452 get_cpu();
1453 do_cpuid_1_ent(entry, function, index);
1454 ++*nent;
1455
1456 switch (function) {
1457 case 0:
1458 entry->eax = min(entry->eax, (u32)0xb);
1459 break;
1460 case 1:
1461 entry->edx &= kvm_supported_word0_x86_features;
1462 entry->ecx &= kvm_supported_word4_x86_features;
1463 break;
1464 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1465 * may return different values. This forces us to get_cpu() before
1466 * issuing the first command, and also to emulate this annoying behavior
1467 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1468 case 2: {
1469 int t, times = entry->eax & 0xff;
1470
1471 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1472 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1473 for (t = 1; t < times && *nent < maxnent; ++t) {
1474 do_cpuid_1_ent(&entry[t], function, 0);
1475 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1476 ++*nent;
1477 }
1478 break;
1479 }
1480 /* function 4 and 0xb have additional index. */
1481 case 4: {
1482 int i, cache_type;
1483
1484 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1485 /* read more entries until cache_type is zero */
1486 for (i = 1; *nent < maxnent; ++i) {
1487 cache_type = entry[i - 1].eax & 0x1f;
1488 if (!cache_type)
1489 break;
1490 do_cpuid_1_ent(&entry[i], function, i);
1491 entry[i].flags |=
1492 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1493 ++*nent;
1494 }
1495 break;
1496 }
1497 case 0xb: {
1498 int i, level_type;
1499
1500 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1501 /* read more entries until level_type is zero */
1502 for (i = 1; *nent < maxnent; ++i) {
1503 level_type = entry[i - 1].ecx & 0xff00;
1504 if (!level_type)
1505 break;
1506 do_cpuid_1_ent(&entry[i], function, i);
1507 entry[i].flags |=
1508 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1509 ++*nent;
1510 }
1511 break;
1512 }
1513 case 0x80000000:
1514 entry->eax = min(entry->eax, 0x8000001a);
1515 break;
1516 case 0x80000001:
1517 entry->edx &= kvm_supported_word1_x86_features;
1518 entry->ecx &= kvm_supported_word6_x86_features;
1519 break;
1520 }
1521 put_cpu();
1522 }
1523
1524 #undef F
1525
1526 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1527 struct kvm_cpuid_entry2 __user *entries)
1528 {
1529 struct kvm_cpuid_entry2 *cpuid_entries;
1530 int limit, nent = 0, r = -E2BIG;
1531 u32 func;
1532
1533 if (cpuid->nent < 1)
1534 goto out;
1535 r = -ENOMEM;
1536 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1537 if (!cpuid_entries)
1538 goto out;
1539
1540 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1541 limit = cpuid_entries[0].eax;
1542 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1543 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1544 &nent, cpuid->nent);
1545 r = -E2BIG;
1546 if (nent >= cpuid->nent)
1547 goto out_free;
1548
1549 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1550 limit = cpuid_entries[nent - 1].eax;
1551 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1552 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1553 &nent, cpuid->nent);
1554 r = -E2BIG;
1555 if (nent >= cpuid->nent)
1556 goto out_free;
1557
1558 r = -EFAULT;
1559 if (copy_to_user(entries, cpuid_entries,
1560 nent * sizeof(struct kvm_cpuid_entry2)))
1561 goto out_free;
1562 cpuid->nent = nent;
1563 r = 0;
1564
1565 out_free:
1566 vfree(cpuid_entries);
1567 out:
1568 return r;
1569 }
1570
1571 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1572 struct kvm_lapic_state *s)
1573 {
1574 vcpu_load(vcpu);
1575 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1576 vcpu_put(vcpu);
1577
1578 return 0;
1579 }
1580
1581 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1582 struct kvm_lapic_state *s)
1583 {
1584 vcpu_load(vcpu);
1585 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1586 kvm_apic_post_state_restore(vcpu);
1587 vcpu_put(vcpu);
1588
1589 return 0;
1590 }
1591
1592 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1593 struct kvm_interrupt *irq)
1594 {
1595 if (irq->irq < 0 || irq->irq >= 256)
1596 return -EINVAL;
1597 if (irqchip_in_kernel(vcpu->kvm))
1598 return -ENXIO;
1599 vcpu_load(vcpu);
1600
1601 kvm_queue_interrupt(vcpu, irq->irq, false);
1602
1603 vcpu_put(vcpu);
1604
1605 return 0;
1606 }
1607
1608 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1609 {
1610 vcpu_load(vcpu);
1611 kvm_inject_nmi(vcpu);
1612 vcpu_put(vcpu);
1613
1614 return 0;
1615 }
1616
1617 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1618 struct kvm_tpr_access_ctl *tac)
1619 {
1620 if (tac->flags)
1621 return -EINVAL;
1622 vcpu->arch.tpr_access_reporting = !!tac->enabled;
1623 return 0;
1624 }
1625
1626 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
1627 u64 mcg_cap)
1628 {
1629 int r;
1630 unsigned bank_num = mcg_cap & 0xff, bank;
1631
1632 r = -EINVAL;
1633 if (!bank_num)
1634 goto out;
1635 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
1636 goto out;
1637 r = 0;
1638 vcpu->arch.mcg_cap = mcg_cap;
1639 /* Init IA32_MCG_CTL to all 1s */
1640 if (mcg_cap & MCG_CTL_P)
1641 vcpu->arch.mcg_ctl = ~(u64)0;
1642 /* Init IA32_MCi_CTL to all 1s */
1643 for (bank = 0; bank < bank_num; bank++)
1644 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
1645 out:
1646 return r;
1647 }
1648
1649 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
1650 struct kvm_x86_mce *mce)
1651 {
1652 u64 mcg_cap = vcpu->arch.mcg_cap;
1653 unsigned bank_num = mcg_cap & 0xff;
1654 u64 *banks = vcpu->arch.mce_banks;
1655
1656 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
1657 return -EINVAL;
1658 /*
1659 * if IA32_MCG_CTL is not all 1s, the uncorrected error
1660 * reporting is disabled
1661 */
1662 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
1663 vcpu->arch.mcg_ctl != ~(u64)0)
1664 return 0;
1665 banks += 4 * mce->bank;
1666 /*
1667 * if IA32_MCi_CTL is not all 1s, the uncorrected error
1668 * reporting is disabled for the bank
1669 */
1670 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
1671 return 0;
1672 if (mce->status & MCI_STATUS_UC) {
1673 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
1674 !(vcpu->arch.cr4 & X86_CR4_MCE)) {
1675 printk(KERN_DEBUG "kvm: set_mce: "
1676 "injects mce exception while "
1677 "previous one is in progress!\n");
1678 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
1679 return 0;
1680 }
1681 if (banks[1] & MCI_STATUS_VAL)
1682 mce->status |= MCI_STATUS_OVER;
1683 banks[2] = mce->addr;
1684 banks[3] = mce->misc;
1685 vcpu->arch.mcg_status = mce->mcg_status;
1686 banks[1] = mce->status;
1687 kvm_queue_exception(vcpu, MC_VECTOR);
1688 } else if (!(banks[1] & MCI_STATUS_VAL)
1689 || !(banks[1] & MCI_STATUS_UC)) {
1690 if (banks[1] & MCI_STATUS_VAL)
1691 mce->status |= MCI_STATUS_OVER;
1692 banks[2] = mce->addr;
1693 banks[3] = mce->misc;
1694 banks[1] = mce->status;
1695 } else
1696 banks[1] |= MCI_STATUS_OVER;
1697 return 0;
1698 }
1699
1700 long kvm_arch_vcpu_ioctl(struct file *filp,
1701 unsigned int ioctl, unsigned long arg)
1702 {
1703 struct kvm_vcpu *vcpu = filp->private_data;
1704 void __user *argp = (void __user *)arg;
1705 int r;
1706 struct kvm_lapic_state *lapic = NULL;
1707
1708 switch (ioctl) {
1709 case KVM_GET_LAPIC: {
1710 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1711
1712 r = -ENOMEM;
1713 if (!lapic)
1714 goto out;
1715 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1716 if (r)
1717 goto out;
1718 r = -EFAULT;
1719 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1720 goto out;
1721 r = 0;
1722 break;
1723 }
1724 case KVM_SET_LAPIC: {
1725 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1726 r = -ENOMEM;
1727 if (!lapic)
1728 goto out;
1729 r = -EFAULT;
1730 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1731 goto out;
1732 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1733 if (r)
1734 goto out;
1735 r = 0;
1736 break;
1737 }
1738 case KVM_INTERRUPT: {
1739 struct kvm_interrupt irq;
1740
1741 r = -EFAULT;
1742 if (copy_from_user(&irq, argp, sizeof irq))
1743 goto out;
1744 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1745 if (r)
1746 goto out;
1747 r = 0;
1748 break;
1749 }
1750 case KVM_NMI: {
1751 r = kvm_vcpu_ioctl_nmi(vcpu);
1752 if (r)
1753 goto out;
1754 r = 0;
1755 break;
1756 }
1757 case KVM_SET_CPUID: {
1758 struct kvm_cpuid __user *cpuid_arg = argp;
1759 struct kvm_cpuid cpuid;
1760
1761 r = -EFAULT;
1762 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1763 goto out;
1764 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1765 if (r)
1766 goto out;
1767 break;
1768 }
1769 case KVM_SET_CPUID2: {
1770 struct kvm_cpuid2 __user *cpuid_arg = argp;
1771 struct kvm_cpuid2 cpuid;
1772
1773 r = -EFAULT;
1774 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1775 goto out;
1776 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1777 cpuid_arg->entries);
1778 if (r)
1779 goto out;
1780 break;
1781 }
1782 case KVM_GET_CPUID2: {
1783 struct kvm_cpuid2 __user *cpuid_arg = argp;
1784 struct kvm_cpuid2 cpuid;
1785
1786 r = -EFAULT;
1787 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1788 goto out;
1789 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1790 cpuid_arg->entries);
1791 if (r)
1792 goto out;
1793 r = -EFAULT;
1794 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1795 goto out;
1796 r = 0;
1797 break;
1798 }
1799 case KVM_GET_MSRS:
1800 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1801 break;
1802 case KVM_SET_MSRS:
1803 r = msr_io(vcpu, argp, do_set_msr, 0);
1804 break;
1805 case KVM_TPR_ACCESS_REPORTING: {
1806 struct kvm_tpr_access_ctl tac;
1807
1808 r = -EFAULT;
1809 if (copy_from_user(&tac, argp, sizeof tac))
1810 goto out;
1811 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1812 if (r)
1813 goto out;
1814 r = -EFAULT;
1815 if (copy_to_user(argp, &tac, sizeof tac))
1816 goto out;
1817 r = 0;
1818 break;
1819 };
1820 case KVM_SET_VAPIC_ADDR: {
1821 struct kvm_vapic_addr va;
1822
1823 r = -EINVAL;
1824 if (!irqchip_in_kernel(vcpu->kvm))
1825 goto out;
1826 r = -EFAULT;
1827 if (copy_from_user(&va, argp, sizeof va))
1828 goto out;
1829 r = 0;
1830 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1831 break;
1832 }
1833 case KVM_X86_SETUP_MCE: {
1834 u64 mcg_cap;
1835
1836 r = -EFAULT;
1837 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
1838 goto out;
1839 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
1840 break;
1841 }
1842 case KVM_X86_SET_MCE: {
1843 struct kvm_x86_mce mce;
1844
1845 r = -EFAULT;
1846 if (copy_from_user(&mce, argp, sizeof mce))
1847 goto out;
1848 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
1849 break;
1850 }
1851 default:
1852 r = -EINVAL;
1853 }
1854 out:
1855 kfree(lapic);
1856 return r;
1857 }
1858
1859 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1860 {
1861 int ret;
1862
1863 if (addr > (unsigned int)(-3 * PAGE_SIZE))
1864 return -1;
1865 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1866 return ret;
1867 }
1868
1869 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1870 u32 kvm_nr_mmu_pages)
1871 {
1872 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1873 return -EINVAL;
1874
1875 down_write(&kvm->slots_lock);
1876 spin_lock(&kvm->mmu_lock);
1877
1878 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1879 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1880
1881 spin_unlock(&kvm->mmu_lock);
1882 up_write(&kvm->slots_lock);
1883 return 0;
1884 }
1885
1886 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1887 {
1888 return kvm->arch.n_alloc_mmu_pages;
1889 }
1890
1891 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1892 {
1893 int i;
1894 struct kvm_mem_alias *alias;
1895
1896 for (i = 0; i < kvm->arch.naliases; ++i) {
1897 alias = &kvm->arch.aliases[i];
1898 if (gfn >= alias->base_gfn
1899 && gfn < alias->base_gfn + alias->npages)
1900 return alias->target_gfn + gfn - alias->base_gfn;
1901 }
1902 return gfn;
1903 }
1904
1905 /*
1906 * Set a new alias region. Aliases map a portion of physical memory into
1907 * another portion. This is useful for memory windows, for example the PC
1908 * VGA region.
1909 */
1910 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1911 struct kvm_memory_alias *alias)
1912 {
1913 int r, n;
1914 struct kvm_mem_alias *p;
1915
1916 r = -EINVAL;
1917 /* General sanity checks */
1918 if (alias->memory_size & (PAGE_SIZE - 1))
1919 goto out;
1920 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1921 goto out;
1922 if (alias->slot >= KVM_ALIAS_SLOTS)
1923 goto out;
1924 if (alias->guest_phys_addr + alias->memory_size
1925 < alias->guest_phys_addr)
1926 goto out;
1927 if (alias->target_phys_addr + alias->memory_size
1928 < alias->target_phys_addr)
1929 goto out;
1930
1931 down_write(&kvm->slots_lock);
1932 spin_lock(&kvm->mmu_lock);
1933
1934 p = &kvm->arch.aliases[alias->slot];
1935 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
1936 p->npages = alias->memory_size >> PAGE_SHIFT;
1937 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
1938
1939 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
1940 if (kvm->arch.aliases[n - 1].npages)
1941 break;
1942 kvm->arch.naliases = n;
1943
1944 spin_unlock(&kvm->mmu_lock);
1945 kvm_mmu_zap_all(kvm);
1946
1947 up_write(&kvm->slots_lock);
1948
1949 return 0;
1950
1951 out:
1952 return r;
1953 }
1954
1955 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1956 {
1957 int r;
1958
1959 r = 0;
1960 switch (chip->chip_id) {
1961 case KVM_IRQCHIP_PIC_MASTER:
1962 memcpy(&chip->chip.pic,
1963 &pic_irqchip(kvm)->pics[0],
1964 sizeof(struct kvm_pic_state));
1965 break;
1966 case KVM_IRQCHIP_PIC_SLAVE:
1967 memcpy(&chip->chip.pic,
1968 &pic_irqchip(kvm)->pics[1],
1969 sizeof(struct kvm_pic_state));
1970 break;
1971 case KVM_IRQCHIP_IOAPIC:
1972 memcpy(&chip->chip.ioapic,
1973 ioapic_irqchip(kvm),
1974 sizeof(struct kvm_ioapic_state));
1975 break;
1976 default:
1977 r = -EINVAL;
1978 break;
1979 }
1980 return r;
1981 }
1982
1983 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
1984 {
1985 int r;
1986
1987 r = 0;
1988 switch (chip->chip_id) {
1989 case KVM_IRQCHIP_PIC_MASTER:
1990 spin_lock(&pic_irqchip(kvm)->lock);
1991 memcpy(&pic_irqchip(kvm)->pics[0],
1992 &chip->chip.pic,
1993 sizeof(struct kvm_pic_state));
1994 spin_unlock(&pic_irqchip(kvm)->lock);
1995 break;
1996 case KVM_IRQCHIP_PIC_SLAVE:
1997 spin_lock(&pic_irqchip(kvm)->lock);
1998 memcpy(&pic_irqchip(kvm)->pics[1],
1999 &chip->chip.pic,
2000 sizeof(struct kvm_pic_state));
2001 spin_unlock(&pic_irqchip(kvm)->lock);
2002 break;
2003 case KVM_IRQCHIP_IOAPIC:
2004 mutex_lock(&kvm->irq_lock);
2005 memcpy(ioapic_irqchip(kvm),
2006 &chip->chip.ioapic,
2007 sizeof(struct kvm_ioapic_state));
2008 mutex_unlock(&kvm->irq_lock);
2009 break;
2010 default:
2011 r = -EINVAL;
2012 break;
2013 }
2014 kvm_pic_update_irq(pic_irqchip(kvm));
2015 return r;
2016 }
2017
2018 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2019 {
2020 int r = 0;
2021
2022 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2023 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2024 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2025 return r;
2026 }
2027
2028 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2029 {
2030 int r = 0;
2031
2032 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2033 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2034 kvm_pit_load_count(kvm, 0, ps->channels[0].count);
2035 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2036 return r;
2037 }
2038
2039 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2040 struct kvm_reinject_control *control)
2041 {
2042 if (!kvm->arch.vpit)
2043 return -ENXIO;
2044 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2045 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2046 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2047 return 0;
2048 }
2049
2050 /*
2051 * Get (and clear) the dirty memory log for a memory slot.
2052 */
2053 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2054 struct kvm_dirty_log *log)
2055 {
2056 int r;
2057 int n;
2058 struct kvm_memory_slot *memslot;
2059 int is_dirty = 0;
2060
2061 down_write(&kvm->slots_lock);
2062
2063 r = kvm_get_dirty_log(kvm, log, &is_dirty);
2064 if (r)
2065 goto out;
2066
2067 /* If nothing is dirty, don't bother messing with page tables. */
2068 if (is_dirty) {
2069 spin_lock(&kvm->mmu_lock);
2070 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2071 spin_unlock(&kvm->mmu_lock);
2072 kvm_flush_remote_tlbs(kvm);
2073 memslot = &kvm->memslots[log->slot];
2074 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
2075 memset(memslot->dirty_bitmap, 0, n);
2076 }
2077 r = 0;
2078 out:
2079 up_write(&kvm->slots_lock);
2080 return r;
2081 }
2082
2083 long kvm_arch_vm_ioctl(struct file *filp,
2084 unsigned int ioctl, unsigned long arg)
2085 {
2086 struct kvm *kvm = filp->private_data;
2087 void __user *argp = (void __user *)arg;
2088 int r = -EINVAL;
2089 /*
2090 * This union makes it completely explicit to gcc-3.x
2091 * that these two variables' stack usage should be
2092 * combined, not added together.
2093 */
2094 union {
2095 struct kvm_pit_state ps;
2096 struct kvm_memory_alias alias;
2097 struct kvm_pit_config pit_config;
2098 } u;
2099
2100 switch (ioctl) {
2101 case KVM_SET_TSS_ADDR:
2102 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2103 if (r < 0)
2104 goto out;
2105 break;
2106 case KVM_SET_MEMORY_REGION: {
2107 struct kvm_memory_region kvm_mem;
2108 struct kvm_userspace_memory_region kvm_userspace_mem;
2109
2110 r = -EFAULT;
2111 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2112 goto out;
2113 kvm_userspace_mem.slot = kvm_mem.slot;
2114 kvm_userspace_mem.flags = kvm_mem.flags;
2115 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2116 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2117 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2118 if (r)
2119 goto out;
2120 break;
2121 }
2122 case KVM_SET_NR_MMU_PAGES:
2123 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2124 if (r)
2125 goto out;
2126 break;
2127 case KVM_GET_NR_MMU_PAGES:
2128 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2129 break;
2130 case KVM_SET_MEMORY_ALIAS:
2131 r = -EFAULT;
2132 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2133 goto out;
2134 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2135 if (r)
2136 goto out;
2137 break;
2138 case KVM_CREATE_IRQCHIP:
2139 r = -ENOMEM;
2140 kvm->arch.vpic = kvm_create_pic(kvm);
2141 if (kvm->arch.vpic) {
2142 r = kvm_ioapic_init(kvm);
2143 if (r) {
2144 kfree(kvm->arch.vpic);
2145 kvm->arch.vpic = NULL;
2146 goto out;
2147 }
2148 } else
2149 goto out;
2150 r = kvm_setup_default_irq_routing(kvm);
2151 if (r) {
2152 kfree(kvm->arch.vpic);
2153 kfree(kvm->arch.vioapic);
2154 goto out;
2155 }
2156 break;
2157 case KVM_CREATE_PIT:
2158 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2159 goto create_pit;
2160 case KVM_CREATE_PIT2:
2161 r = -EFAULT;
2162 if (copy_from_user(&u.pit_config, argp,
2163 sizeof(struct kvm_pit_config)))
2164 goto out;
2165 create_pit:
2166 mutex_lock(&kvm->lock);
2167 r = -EEXIST;
2168 if (kvm->arch.vpit)
2169 goto create_pit_unlock;
2170 r = -ENOMEM;
2171 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2172 if (kvm->arch.vpit)
2173 r = 0;
2174 create_pit_unlock:
2175 mutex_unlock(&kvm->lock);
2176 break;
2177 case KVM_IRQ_LINE_STATUS:
2178 case KVM_IRQ_LINE: {
2179 struct kvm_irq_level irq_event;
2180
2181 r = -EFAULT;
2182 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2183 goto out;
2184 if (irqchip_in_kernel(kvm)) {
2185 __s32 status;
2186 mutex_lock(&kvm->irq_lock);
2187 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2188 irq_event.irq, irq_event.level);
2189 mutex_unlock(&kvm->irq_lock);
2190 if (ioctl == KVM_IRQ_LINE_STATUS) {
2191 irq_event.status = status;
2192 if (copy_to_user(argp, &irq_event,
2193 sizeof irq_event))
2194 goto out;
2195 }
2196 r = 0;
2197 }
2198 break;
2199 }
2200 case KVM_GET_IRQCHIP: {
2201 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2202 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2203
2204 r = -ENOMEM;
2205 if (!chip)
2206 goto out;
2207 r = -EFAULT;
2208 if (copy_from_user(chip, argp, sizeof *chip))
2209 goto get_irqchip_out;
2210 r = -ENXIO;
2211 if (!irqchip_in_kernel(kvm))
2212 goto get_irqchip_out;
2213 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2214 if (r)
2215 goto get_irqchip_out;
2216 r = -EFAULT;
2217 if (copy_to_user(argp, chip, sizeof *chip))
2218 goto get_irqchip_out;
2219 r = 0;
2220 get_irqchip_out:
2221 kfree(chip);
2222 if (r)
2223 goto out;
2224 break;
2225 }
2226 case KVM_SET_IRQCHIP: {
2227 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2228 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2229
2230 r = -ENOMEM;
2231 if (!chip)
2232 goto out;
2233 r = -EFAULT;
2234 if (copy_from_user(chip, argp, sizeof *chip))
2235 goto set_irqchip_out;
2236 r = -ENXIO;
2237 if (!irqchip_in_kernel(kvm))
2238 goto set_irqchip_out;
2239 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2240 if (r)
2241 goto set_irqchip_out;
2242 r = 0;
2243 set_irqchip_out:
2244 kfree(chip);
2245 if (r)
2246 goto out;
2247 break;
2248 }
2249 case KVM_GET_PIT: {
2250 r = -EFAULT;
2251 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2252 goto out;
2253 r = -ENXIO;
2254 if (!kvm->arch.vpit)
2255 goto out;
2256 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2257 if (r)
2258 goto out;
2259 r = -EFAULT;
2260 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2261 goto out;
2262 r = 0;
2263 break;
2264 }
2265 case KVM_SET_PIT: {
2266 r = -EFAULT;
2267 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2268 goto out;
2269 r = -ENXIO;
2270 if (!kvm->arch.vpit)
2271 goto out;
2272 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2273 if (r)
2274 goto out;
2275 r = 0;
2276 break;
2277 }
2278 case KVM_REINJECT_CONTROL: {
2279 struct kvm_reinject_control control;
2280 r = -EFAULT;
2281 if (copy_from_user(&control, argp, sizeof(control)))
2282 goto out;
2283 r = kvm_vm_ioctl_reinject(kvm, &control);
2284 if (r)
2285 goto out;
2286 r = 0;
2287 break;
2288 }
2289 default:
2290 ;
2291 }
2292 out:
2293 return r;
2294 }
2295
2296 static void kvm_init_msr_list(void)
2297 {
2298 u32 dummy[2];
2299 unsigned i, j;
2300
2301 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2302 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2303 continue;
2304 if (j < i)
2305 msrs_to_save[j] = msrs_to_save[i];
2306 j++;
2307 }
2308 num_msrs_to_save = j;
2309 }
2310
2311 /*
2312 * Only apic need an MMIO device hook, so shortcut now..
2313 */
2314 static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
2315 gpa_t addr, int len,
2316 int is_write)
2317 {
2318 struct kvm_io_device *dev;
2319
2320 if (vcpu->arch.apic) {
2321 dev = &vcpu->arch.apic->dev;
2322 if (kvm_iodevice_in_range(dev, addr, len, is_write))
2323 return dev;
2324 }
2325 return NULL;
2326 }
2327
2328
2329 static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
2330 gpa_t addr, int len,
2331 int is_write)
2332 {
2333 struct kvm_io_device *dev;
2334
2335 dev = vcpu_find_pervcpu_dev(vcpu, addr, len, is_write);
2336 if (dev == NULL)
2337 dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr, len,
2338 is_write);
2339 return dev;
2340 }
2341
2342 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2343 struct kvm_vcpu *vcpu)
2344 {
2345 void *data = val;
2346 int r = X86EMUL_CONTINUE;
2347
2348 while (bytes) {
2349 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2350 unsigned offset = addr & (PAGE_SIZE-1);
2351 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2352 int ret;
2353
2354 if (gpa == UNMAPPED_GVA) {
2355 r = X86EMUL_PROPAGATE_FAULT;
2356 goto out;
2357 }
2358 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2359 if (ret < 0) {
2360 r = X86EMUL_UNHANDLEABLE;
2361 goto out;
2362 }
2363
2364 bytes -= toread;
2365 data += toread;
2366 addr += toread;
2367 }
2368 out:
2369 return r;
2370 }
2371
2372 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2373 struct kvm_vcpu *vcpu)
2374 {
2375 void *data = val;
2376 int r = X86EMUL_CONTINUE;
2377
2378 while (bytes) {
2379 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2380 unsigned offset = addr & (PAGE_SIZE-1);
2381 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2382 int ret;
2383
2384 if (gpa == UNMAPPED_GVA) {
2385 r = X86EMUL_PROPAGATE_FAULT;
2386 goto out;
2387 }
2388 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2389 if (ret < 0) {
2390 r = X86EMUL_UNHANDLEABLE;
2391 goto out;
2392 }
2393
2394 bytes -= towrite;
2395 data += towrite;
2396 addr += towrite;
2397 }
2398 out:
2399 return r;
2400 }
2401
2402
2403 static int emulator_read_emulated(unsigned long addr,
2404 void *val,
2405 unsigned int bytes,
2406 struct kvm_vcpu *vcpu)
2407 {
2408 struct kvm_io_device *mmio_dev;
2409 gpa_t gpa;
2410
2411 if (vcpu->mmio_read_completed) {
2412 memcpy(val, vcpu->mmio_data, bytes);
2413 vcpu->mmio_read_completed = 0;
2414 return X86EMUL_CONTINUE;
2415 }
2416
2417 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2418
2419 /* For APIC access vmexit */
2420 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2421 goto mmio;
2422
2423 if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2424 == X86EMUL_CONTINUE)
2425 return X86EMUL_CONTINUE;
2426 if (gpa == UNMAPPED_GVA)
2427 return X86EMUL_PROPAGATE_FAULT;
2428
2429 mmio:
2430 /*
2431 * Is this MMIO handled locally?
2432 */
2433 mutex_lock(&vcpu->kvm->lock);
2434 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 0);
2435 mutex_unlock(&vcpu->kvm->lock);
2436 if (mmio_dev) {
2437 kvm_iodevice_read(mmio_dev, gpa, bytes, val);
2438 return X86EMUL_CONTINUE;
2439 }
2440
2441 vcpu->mmio_needed = 1;
2442 vcpu->mmio_phys_addr = gpa;
2443 vcpu->mmio_size = bytes;
2444 vcpu->mmio_is_write = 0;
2445
2446 return X86EMUL_UNHANDLEABLE;
2447 }
2448
2449 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2450 const void *val, int bytes)
2451 {
2452 int ret;
2453
2454 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2455 if (ret < 0)
2456 return 0;
2457 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2458 return 1;
2459 }
2460
2461 static int emulator_write_emulated_onepage(unsigned long addr,
2462 const void *val,
2463 unsigned int bytes,
2464 struct kvm_vcpu *vcpu)
2465 {
2466 struct kvm_io_device *mmio_dev;
2467 gpa_t gpa;
2468
2469 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2470
2471 if (gpa == UNMAPPED_GVA) {
2472 kvm_inject_page_fault(vcpu, addr, 2);
2473 return X86EMUL_PROPAGATE_FAULT;
2474 }
2475
2476 /* For APIC access vmexit */
2477 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2478 goto mmio;
2479
2480 if (emulator_write_phys(vcpu, gpa, val, bytes))
2481 return X86EMUL_CONTINUE;
2482
2483 mmio:
2484 /*
2485 * Is this MMIO handled locally?
2486 */
2487 mutex_lock(&vcpu->kvm->lock);
2488 mmio_dev = vcpu_find_mmio_dev(vcpu, gpa, bytes, 1);
2489 mutex_unlock(&vcpu->kvm->lock);
2490 if (mmio_dev) {
2491 kvm_iodevice_write(mmio_dev, gpa, bytes, val);
2492 return X86EMUL_CONTINUE;
2493 }
2494
2495 vcpu->mmio_needed = 1;
2496 vcpu->mmio_phys_addr = gpa;
2497 vcpu->mmio_size = bytes;
2498 vcpu->mmio_is_write = 1;
2499 memcpy(vcpu->mmio_data, val, bytes);
2500
2501 return X86EMUL_CONTINUE;
2502 }
2503
2504 int emulator_write_emulated(unsigned long addr,
2505 const void *val,
2506 unsigned int bytes,
2507 struct kvm_vcpu *vcpu)
2508 {
2509 /* Crossing a page boundary? */
2510 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2511 int rc, now;
2512
2513 now = -addr & ~PAGE_MASK;
2514 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2515 if (rc != X86EMUL_CONTINUE)
2516 return rc;
2517 addr += now;
2518 val += now;
2519 bytes -= now;
2520 }
2521 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2522 }
2523 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2524
2525 static int emulator_cmpxchg_emulated(unsigned long addr,
2526 const void *old,
2527 const void *new,
2528 unsigned int bytes,
2529 struct kvm_vcpu *vcpu)
2530 {
2531 static int reported;
2532
2533 if (!reported) {
2534 reported = 1;
2535 printk(KERN_WARNING "kvm: emulating exchange as write\n");
2536 }
2537 #ifndef CONFIG_X86_64
2538 /* guests cmpxchg8b have to be emulated atomically */
2539 if (bytes == 8) {
2540 gpa_t gpa;
2541 struct page *page;
2542 char *kaddr;
2543 u64 val;
2544
2545 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2546
2547 if (gpa == UNMAPPED_GVA ||
2548 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2549 goto emul_write;
2550
2551 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2552 goto emul_write;
2553
2554 val = *(u64 *)new;
2555
2556 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2557
2558 kaddr = kmap_atomic(page, KM_USER0);
2559 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2560 kunmap_atomic(kaddr, KM_USER0);
2561 kvm_release_page_dirty(page);
2562 }
2563 emul_write:
2564 #endif
2565
2566 return emulator_write_emulated(addr, new, bytes, vcpu);
2567 }
2568
2569 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2570 {
2571 return kvm_x86_ops->get_segment_base(vcpu, seg);
2572 }
2573
2574 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2575 {
2576 kvm_mmu_invlpg(vcpu, address);
2577 return X86EMUL_CONTINUE;
2578 }
2579
2580 int emulate_clts(struct kvm_vcpu *vcpu)
2581 {
2582 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2583 return X86EMUL_CONTINUE;
2584 }
2585
2586 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2587 {
2588 struct kvm_vcpu *vcpu = ctxt->vcpu;
2589
2590 switch (dr) {
2591 case 0 ... 3:
2592 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2593 return X86EMUL_CONTINUE;
2594 default:
2595 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2596 return X86EMUL_UNHANDLEABLE;
2597 }
2598 }
2599
2600 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2601 {
2602 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2603 int exception;
2604
2605 kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2606 if (exception) {
2607 /* FIXME: better handling */
2608 return X86EMUL_UNHANDLEABLE;
2609 }
2610 return X86EMUL_CONTINUE;
2611 }
2612
2613 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2614 {
2615 u8 opcodes[4];
2616 unsigned long rip = kvm_rip_read(vcpu);
2617 unsigned long rip_linear;
2618
2619 if (!printk_ratelimit())
2620 return;
2621
2622 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2623
2624 kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
2625
2626 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2627 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2628 }
2629 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2630
2631 static struct x86_emulate_ops emulate_ops = {
2632 .read_std = kvm_read_guest_virt,
2633 .read_emulated = emulator_read_emulated,
2634 .write_emulated = emulator_write_emulated,
2635 .cmpxchg_emulated = emulator_cmpxchg_emulated,
2636 };
2637
2638 static void cache_all_regs(struct kvm_vcpu *vcpu)
2639 {
2640 kvm_register_read(vcpu, VCPU_REGS_RAX);
2641 kvm_register_read(vcpu, VCPU_REGS_RSP);
2642 kvm_register_read(vcpu, VCPU_REGS_RIP);
2643 vcpu->arch.regs_dirty = ~0;
2644 }
2645
2646 int emulate_instruction(struct kvm_vcpu *vcpu,
2647 struct kvm_run *run,
2648 unsigned long cr2,
2649 u16 error_code,
2650 int emulation_type)
2651 {
2652 int r, shadow_mask;
2653 struct decode_cache *c;
2654
2655 kvm_clear_exception_queue(vcpu);
2656 vcpu->arch.mmio_fault_cr2 = cr2;
2657 /*
2658 * TODO: fix x86_emulate.c to use guest_read/write_register
2659 * instead of direct ->regs accesses, can save hundred cycles
2660 * on Intel for instructions that don't read/change RSP, for
2661 * for example.
2662 */
2663 cache_all_regs(vcpu);
2664
2665 vcpu->mmio_is_write = 0;
2666 vcpu->arch.pio.string = 0;
2667
2668 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2669 int cs_db, cs_l;
2670 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2671
2672 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2673 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2674 vcpu->arch.emulate_ctxt.mode =
2675 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2676 ? X86EMUL_MODE_REAL : cs_l
2677 ? X86EMUL_MODE_PROT64 : cs_db
2678 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2679
2680 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2681
2682 /* Only allow emulation of specific instructions on #UD
2683 * (namely VMMCALL, sysenter, sysexit, syscall)*/
2684 c = &vcpu->arch.emulate_ctxt.decode;
2685 if (emulation_type & EMULTYPE_TRAP_UD) {
2686 if (!c->twobyte)
2687 return EMULATE_FAIL;
2688 switch (c->b) {
2689 case 0x01: /* VMMCALL */
2690 if (c->modrm_mod != 3 || c->modrm_rm != 1)
2691 return EMULATE_FAIL;
2692 break;
2693 case 0x34: /* sysenter */
2694 case 0x35: /* sysexit */
2695 if (c->modrm_mod != 0 || c->modrm_rm != 0)
2696 return EMULATE_FAIL;
2697 break;
2698 case 0x05: /* syscall */
2699 if (c->modrm_mod != 0 || c->modrm_rm != 0)
2700 return EMULATE_FAIL;
2701 break;
2702 default:
2703 return EMULATE_FAIL;
2704 }
2705
2706 if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
2707 return EMULATE_FAIL;
2708 }
2709
2710 ++vcpu->stat.insn_emulation;
2711 if (r) {
2712 ++vcpu->stat.insn_emulation_fail;
2713 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2714 return EMULATE_DONE;
2715 return EMULATE_FAIL;
2716 }
2717 }
2718
2719 if (emulation_type & EMULTYPE_SKIP) {
2720 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
2721 return EMULATE_DONE;
2722 }
2723
2724 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2725 shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
2726
2727 if (r == 0)
2728 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
2729
2730 if (vcpu->arch.pio.string)
2731 return EMULATE_DO_MMIO;
2732
2733 if ((r || vcpu->mmio_is_write) && run) {
2734 run->exit_reason = KVM_EXIT_MMIO;
2735 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2736 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2737 run->mmio.len = vcpu->mmio_size;
2738 run->mmio.is_write = vcpu->mmio_is_write;
2739 }
2740
2741 if (r) {
2742 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2743 return EMULATE_DONE;
2744 if (!vcpu->mmio_needed) {
2745 kvm_report_emulation_failure(vcpu, "mmio");
2746 return EMULATE_FAIL;
2747 }
2748 return EMULATE_DO_MMIO;
2749 }
2750
2751 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2752
2753 if (vcpu->mmio_is_write) {
2754 vcpu->mmio_needed = 0;
2755 return EMULATE_DO_MMIO;
2756 }
2757
2758 return EMULATE_DONE;
2759 }
2760 EXPORT_SYMBOL_GPL(emulate_instruction);
2761
2762 static int pio_copy_data(struct kvm_vcpu *vcpu)
2763 {
2764 void *p = vcpu->arch.pio_data;
2765 gva_t q = vcpu->arch.pio.guest_gva;
2766 unsigned bytes;
2767 int ret;
2768
2769 bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2770 if (vcpu->arch.pio.in)
2771 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
2772 else
2773 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
2774 return ret;
2775 }
2776
2777 int complete_pio(struct kvm_vcpu *vcpu)
2778 {
2779 struct kvm_pio_request *io = &vcpu->arch.pio;
2780 long delta;
2781 int r;
2782 unsigned long val;
2783
2784 if (!io->string) {
2785 if (io->in) {
2786 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2787 memcpy(&val, vcpu->arch.pio_data, io->size);
2788 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2789 }
2790 } else {
2791 if (io->in) {
2792 r = pio_copy_data(vcpu);
2793 if (r)
2794 return r;
2795 }
2796
2797 delta = 1;
2798 if (io->rep) {
2799 delta *= io->cur_count;
2800 /*
2801 * The size of the register should really depend on
2802 * current address size.
2803 */
2804 val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2805 val -= delta;
2806 kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2807 }
2808 if (io->down)
2809 delta = -delta;
2810 delta *= io->size;
2811 if (io->in) {
2812 val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2813 val += delta;
2814 kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2815 } else {
2816 val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2817 val += delta;
2818 kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2819 }
2820 }
2821
2822 io->count -= io->cur_count;
2823 io->cur_count = 0;
2824
2825 return 0;
2826 }
2827
2828 static void kernel_pio(struct kvm_io_device *pio_dev,
2829 struct kvm_vcpu *vcpu,
2830 void *pd)
2831 {
2832 /* TODO: String I/O for in kernel device */
2833
2834 if (vcpu->arch.pio.in)
2835 kvm_iodevice_read(pio_dev, vcpu->arch.pio.port,
2836 vcpu->arch.pio.size,
2837 pd);
2838 else
2839 kvm_iodevice_write(pio_dev, vcpu->arch.pio.port,
2840 vcpu->arch.pio.size,
2841 pd);
2842 }
2843
2844 static void pio_string_write(struct kvm_io_device *pio_dev,
2845 struct kvm_vcpu *vcpu)
2846 {
2847 struct kvm_pio_request *io = &vcpu->arch.pio;
2848 void *pd = vcpu->arch.pio_data;
2849 int i;
2850
2851 for (i = 0; i < io->cur_count; i++) {
2852 kvm_iodevice_write(pio_dev, io->port,
2853 io->size,
2854 pd);
2855 pd += io->size;
2856 }
2857 }
2858
2859 static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
2860 gpa_t addr, int len,
2861 int is_write)
2862 {
2863 return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr, len, is_write);
2864 }
2865
2866 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2867 int size, unsigned port)
2868 {
2869 struct kvm_io_device *pio_dev;
2870 unsigned long val;
2871
2872 vcpu->run->exit_reason = KVM_EXIT_IO;
2873 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2874 vcpu->run->io.size = vcpu->arch.pio.size = size;
2875 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2876 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2877 vcpu->run->io.port = vcpu->arch.pio.port = port;
2878 vcpu->arch.pio.in = in;
2879 vcpu->arch.pio.string = 0;
2880 vcpu->arch.pio.down = 0;
2881 vcpu->arch.pio.rep = 0;
2882
2883 trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
2884 size, 1);
2885
2886 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2887 memcpy(vcpu->arch.pio_data, &val, 4);
2888
2889 mutex_lock(&vcpu->kvm->lock);
2890 pio_dev = vcpu_find_pio_dev(vcpu, port, size, !in);
2891 mutex_unlock(&vcpu->kvm->lock);
2892 if (pio_dev) {
2893 kernel_pio(pio_dev, vcpu, vcpu->arch.pio_data);
2894 complete_pio(vcpu);
2895 return 1;
2896 }
2897 return 0;
2898 }
2899 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
2900
2901 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2902 int size, unsigned long count, int down,
2903 gva_t address, int rep, unsigned port)
2904 {
2905 unsigned now, in_page;
2906 int ret = 0;
2907 struct kvm_io_device *pio_dev;
2908
2909 vcpu->run->exit_reason = KVM_EXIT_IO;
2910 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2911 vcpu->run->io.size = vcpu->arch.pio.size = size;
2912 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2913 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
2914 vcpu->run->io.port = vcpu->arch.pio.port = port;
2915 vcpu->arch.pio.in = in;
2916 vcpu->arch.pio.string = 1;
2917 vcpu->arch.pio.down = down;
2918 vcpu->arch.pio.rep = rep;
2919
2920 trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
2921 size, count);
2922
2923 if (!count) {
2924 kvm_x86_ops->skip_emulated_instruction(vcpu);
2925 return 1;
2926 }
2927
2928 if (!down)
2929 in_page = PAGE_SIZE - offset_in_page(address);
2930 else
2931 in_page = offset_in_page(address) + size;
2932 now = min(count, (unsigned long)in_page / size);
2933 if (!now)
2934 now = 1;
2935 if (down) {
2936 /*
2937 * String I/O in reverse. Yuck. Kill the guest, fix later.
2938 */
2939 pr_unimpl(vcpu, "guest string pio down\n");
2940 kvm_inject_gp(vcpu, 0);
2941 return 1;
2942 }
2943 vcpu->run->io.count = now;
2944 vcpu->arch.pio.cur_count = now;
2945
2946 if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
2947 kvm_x86_ops->skip_emulated_instruction(vcpu);
2948
2949 vcpu->arch.pio.guest_gva = address;
2950
2951 mutex_lock(&vcpu->kvm->lock);
2952 pio_dev = vcpu_find_pio_dev(vcpu, port,
2953 vcpu->arch.pio.cur_count,
2954 !vcpu->arch.pio.in);
2955 mutex_unlock(&vcpu->kvm->lock);
2956
2957 if (!vcpu->arch.pio.in) {
2958 /* string PIO write */
2959 ret = pio_copy_data(vcpu);
2960 if (ret == X86EMUL_PROPAGATE_FAULT) {
2961 kvm_inject_gp(vcpu, 0);
2962 return 1;
2963 }
2964 if (ret == 0 && pio_dev) {
2965 pio_string_write(pio_dev, vcpu);
2966 complete_pio(vcpu);
2967 if (vcpu->arch.pio.count == 0)
2968 ret = 1;
2969 }
2970 } else if (pio_dev)
2971 pr_unimpl(vcpu, "no string pio read support yet, "
2972 "port %x size %d count %ld\n",
2973 port, size, count);
2974
2975 return ret;
2976 }
2977 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
2978
2979 static void bounce_off(void *info)
2980 {
2981 /* nothing */
2982 }
2983
2984 static unsigned int ref_freq;
2985 static unsigned long tsc_khz_ref;
2986
2987 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
2988 void *data)
2989 {
2990 struct cpufreq_freqs *freq = data;
2991 struct kvm *kvm;
2992 struct kvm_vcpu *vcpu;
2993 int i, send_ipi = 0;
2994
2995 if (!ref_freq)
2996 ref_freq = freq->old;
2997
2998 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
2999 return 0;
3000 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3001 return 0;
3002 per_cpu(cpu_tsc_khz, freq->cpu) = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
3003
3004 spin_lock(&kvm_lock);
3005 list_for_each_entry(kvm, &vm_list, vm_list) {
3006 kvm_for_each_vcpu(i, vcpu, kvm) {
3007 if (vcpu->cpu != freq->cpu)
3008 continue;
3009 if (!kvm_request_guest_time_update(vcpu))
3010 continue;
3011 if (vcpu->cpu != smp_processor_id())
3012 send_ipi++;
3013 }
3014 }
3015 spin_unlock(&kvm_lock);
3016
3017 if (freq->old < freq->new && send_ipi) {
3018 /*
3019 * We upscale the frequency. Must make the guest
3020 * doesn't see old kvmclock values while running with
3021 * the new frequency, otherwise we risk the guest sees
3022 * time go backwards.
3023 *
3024 * In case we update the frequency for another cpu
3025 * (which might be in guest context) send an interrupt
3026 * to kick the cpu out of guest context. Next time
3027 * guest context is entered kvmclock will be updated,
3028 * so the guest will not see stale values.
3029 */
3030 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
3031 }
3032 return 0;
3033 }
3034
3035 static struct notifier_block kvmclock_cpufreq_notifier_block = {
3036 .notifier_call = kvmclock_cpufreq_notifier
3037 };
3038
3039 int kvm_arch_init(void *opaque)
3040 {
3041 int r, cpu;
3042 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
3043
3044 if (kvm_x86_ops) {
3045 printk(KERN_ERR "kvm: already loaded the other module\n");
3046 r = -EEXIST;
3047 goto out;
3048 }
3049
3050 if (!ops->cpu_has_kvm_support()) {
3051 printk(KERN_ERR "kvm: no hardware support\n");
3052 r = -EOPNOTSUPP;
3053 goto out;
3054 }
3055 if (ops->disabled_by_bios()) {
3056 printk(KERN_ERR "kvm: disabled by bios\n");
3057 r = -EOPNOTSUPP;
3058 goto out;
3059 }
3060
3061 r = kvm_mmu_module_init();
3062 if (r)
3063 goto out;
3064
3065 kvm_init_msr_list();
3066
3067 kvm_x86_ops = ops;
3068 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3069 kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
3070 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
3071 PT_DIRTY_MASK, PT64_NX_MASK, 0);
3072
3073 for_each_possible_cpu(cpu)
3074 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
3075 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3076 tsc_khz_ref = tsc_khz;
3077 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3078 CPUFREQ_TRANSITION_NOTIFIER);
3079 }
3080
3081 return 0;
3082
3083 out:
3084 return r;
3085 }
3086
3087 void kvm_arch_exit(void)
3088 {
3089 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
3090 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
3091 CPUFREQ_TRANSITION_NOTIFIER);
3092 kvm_x86_ops = NULL;
3093 kvm_mmu_module_exit();
3094 }
3095
3096 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
3097 {
3098 ++vcpu->stat.halt_exits;
3099 if (irqchip_in_kernel(vcpu->kvm)) {
3100 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
3101 return 1;
3102 } else {
3103 vcpu->run->exit_reason = KVM_EXIT_HLT;
3104 return 0;
3105 }
3106 }
3107 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
3108
3109 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
3110 unsigned long a1)
3111 {
3112 if (is_long_mode(vcpu))
3113 return a0;
3114 else
3115 return a0 | ((gpa_t)a1 << 32);
3116 }
3117
3118 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
3119 {
3120 unsigned long nr, a0, a1, a2, a3, ret;
3121 int r = 1;
3122
3123 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
3124 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
3125 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
3126 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
3127 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
3128
3129 trace_kvm_hypercall(nr, a0, a1, a2, a3);
3130
3131 if (!is_long_mode(vcpu)) {
3132 nr &= 0xFFFFFFFF;
3133 a0 &= 0xFFFFFFFF;
3134 a1 &= 0xFFFFFFFF;
3135 a2 &= 0xFFFFFFFF;
3136 a3 &= 0xFFFFFFFF;
3137 }
3138
3139 switch (nr) {
3140 case KVM_HC_VAPIC_POLL_IRQ:
3141 ret = 0;
3142 break;
3143 case KVM_HC_MMU_OP:
3144 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
3145 break;
3146 default:
3147 ret = -KVM_ENOSYS;
3148 break;
3149 }
3150 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3151 ++vcpu->stat.hypercalls;
3152 return r;
3153 }
3154 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
3155
3156 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
3157 {
3158 char instruction[3];
3159 int ret = 0;
3160 unsigned long rip = kvm_rip_read(vcpu);
3161
3162
3163 /*
3164 * Blow out the MMU to ensure that no other VCPU has an active mapping
3165 * to ensure that the updated hypercall appears atomically across all
3166 * VCPUs.
3167 */
3168 kvm_mmu_zap_all(vcpu->kvm);
3169
3170 kvm_x86_ops->patch_hypercall(vcpu, instruction);
3171 if (emulator_write_emulated(rip, instruction, 3, vcpu)
3172 != X86EMUL_CONTINUE)
3173 ret = -EFAULT;
3174
3175 return ret;
3176 }
3177
3178 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3179 {
3180 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3181 }
3182
3183 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3184 {
3185 struct descriptor_table dt = { limit, base };
3186
3187 kvm_x86_ops->set_gdt(vcpu, &dt);
3188 }
3189
3190 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3191 {
3192 struct descriptor_table dt = { limit, base };
3193
3194 kvm_x86_ops->set_idt(vcpu, &dt);
3195 }
3196
3197 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
3198 unsigned long *rflags)
3199 {
3200 kvm_lmsw(vcpu, msw);
3201 *rflags = kvm_x86_ops->get_rflags(vcpu);
3202 }
3203
3204 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
3205 {
3206 unsigned long value;
3207
3208 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3209 switch (cr) {
3210 case 0:
3211 value = vcpu->arch.cr0;
3212 break;
3213 case 2:
3214 value = vcpu->arch.cr2;
3215 break;
3216 case 3:
3217 value = vcpu->arch.cr3;
3218 break;
3219 case 4:
3220 value = vcpu->arch.cr4;
3221 break;
3222 case 8:
3223 value = kvm_get_cr8(vcpu);
3224 break;
3225 default:
3226 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3227 return 0;
3228 }
3229
3230 return value;
3231 }
3232
3233 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
3234 unsigned long *rflags)
3235 {
3236 switch (cr) {
3237 case 0:
3238 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
3239 *rflags = kvm_x86_ops->get_rflags(vcpu);
3240 break;
3241 case 2:
3242 vcpu->arch.cr2 = val;
3243 break;
3244 case 3:
3245 kvm_set_cr3(vcpu, val);
3246 break;
3247 case 4:
3248 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
3249 break;
3250 case 8:
3251 kvm_set_cr8(vcpu, val & 0xfUL);
3252 break;
3253 default:
3254 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3255 }
3256 }
3257
3258 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
3259 {
3260 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
3261 int j, nent = vcpu->arch.cpuid_nent;
3262
3263 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
3264 /* when no next entry is found, the current entry[i] is reselected */
3265 for (j = i + 1; ; j = (j + 1) % nent) {
3266 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
3267 if (ej->function == e->function) {
3268 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
3269 return j;
3270 }
3271 }
3272 return 0; /* silence gcc, even though control never reaches here */
3273 }
3274
3275 /* find an entry with matching function, matching index (if needed), and that
3276 * should be read next (if it's stateful) */
3277 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
3278 u32 function, u32 index)
3279 {
3280 if (e->function != function)
3281 return 0;
3282 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
3283 return 0;
3284 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
3285 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3286 return 0;
3287 return 1;
3288 }
3289
3290 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
3291 u32 function, u32 index)
3292 {
3293 int i;
3294 struct kvm_cpuid_entry2 *best = NULL;
3295
3296 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3297 struct kvm_cpuid_entry2 *e;
3298
3299 e = &vcpu->arch.cpuid_entries[i];
3300 if (is_matching_cpuid_entry(e, function, index)) {
3301 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3302 move_to_next_stateful_cpuid_entry(vcpu, i);
3303 best = e;
3304 break;
3305 }
3306 /*
3307 * Both basic or both extended?
3308 */
3309 if (((e->function ^ function) & 0x80000000) == 0)
3310 if (!best || e->function > best->function)
3311 best = e;
3312 }
3313 return best;
3314 }
3315
3316 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
3317 {
3318 struct kvm_cpuid_entry2 *best;
3319
3320 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
3321 if (best)
3322 return best->eax & 0xff;
3323 return 36;
3324 }
3325
3326 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3327 {
3328 u32 function, index;
3329 struct kvm_cpuid_entry2 *best;
3330
3331 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3332 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3333 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3334 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3335 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3336 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3337 best = kvm_find_cpuid_entry(vcpu, function, index);
3338 if (best) {
3339 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3340 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3341 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3342 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3343 }
3344 kvm_x86_ops->skip_emulated_instruction(vcpu);
3345 trace_kvm_cpuid(function,
3346 kvm_register_read(vcpu, VCPU_REGS_RAX),
3347 kvm_register_read(vcpu, VCPU_REGS_RBX),
3348 kvm_register_read(vcpu, VCPU_REGS_RCX),
3349 kvm_register_read(vcpu, VCPU_REGS_RDX));
3350 }
3351 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3352
3353 /*
3354 * Check if userspace requested an interrupt window, and that the
3355 * interrupt window is open.
3356 *
3357 * No need to exit to userspace if we already have an interrupt queued.
3358 */
3359 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
3360 struct kvm_run *kvm_run)
3361 {
3362 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
3363 kvm_run->request_interrupt_window &&
3364 kvm_arch_interrupt_allowed(vcpu));
3365 }
3366
3367 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
3368 struct kvm_run *kvm_run)
3369 {
3370 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3371 kvm_run->cr8 = kvm_get_cr8(vcpu);
3372 kvm_run->apic_base = kvm_get_apic_base(vcpu);
3373 if (irqchip_in_kernel(vcpu->kvm))
3374 kvm_run->ready_for_interrupt_injection = 1;
3375 else
3376 kvm_run->ready_for_interrupt_injection =
3377 kvm_arch_interrupt_allowed(vcpu) &&
3378 !kvm_cpu_has_interrupt(vcpu) &&
3379 !kvm_event_needs_reinjection(vcpu);
3380 }
3381
3382 static void vapic_enter(struct kvm_vcpu *vcpu)
3383 {
3384 struct kvm_lapic *apic = vcpu->arch.apic;
3385 struct page *page;
3386
3387 if (!apic || !apic->vapic_addr)
3388 return;
3389
3390 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3391
3392 vcpu->arch.apic->vapic_page = page;
3393 }
3394
3395 static void vapic_exit(struct kvm_vcpu *vcpu)
3396 {
3397 struct kvm_lapic *apic = vcpu->arch.apic;
3398
3399 if (!apic || !apic->vapic_addr)
3400 return;
3401
3402 down_read(&vcpu->kvm->slots_lock);
3403 kvm_release_page_dirty(apic->vapic_page);
3404 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3405 up_read(&vcpu->kvm->slots_lock);
3406 }
3407
3408 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
3409 {
3410 int max_irr, tpr;
3411
3412 if (!kvm_x86_ops->update_cr8_intercept)
3413 return;
3414
3415 if (!vcpu->arch.apic->vapic_addr)
3416 max_irr = kvm_lapic_find_highest_irr(vcpu);
3417 else
3418 max_irr = -1;
3419
3420 if (max_irr != -1)
3421 max_irr >>= 4;
3422
3423 tpr = kvm_lapic_get_cr8(vcpu);
3424
3425 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
3426 }
3427
3428 static void inject_pending_irq(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3429 {
3430 /* try to reinject previous events if any */
3431 if (vcpu->arch.nmi_injected) {
3432 kvm_x86_ops->set_nmi(vcpu);
3433 return;
3434 }
3435
3436 if (vcpu->arch.interrupt.pending) {
3437 kvm_x86_ops->set_irq(vcpu);
3438 return;
3439 }
3440
3441 /* try to inject new event if pending */
3442 if (vcpu->arch.nmi_pending) {
3443 if (kvm_x86_ops->nmi_allowed(vcpu)) {
3444 vcpu->arch.nmi_pending = false;
3445 vcpu->arch.nmi_injected = true;
3446 kvm_x86_ops->set_nmi(vcpu);
3447 }
3448 } else if (kvm_cpu_has_interrupt(vcpu)) {
3449 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
3450 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
3451 false);
3452 kvm_x86_ops->set_irq(vcpu);
3453 }
3454 }
3455 }
3456
3457 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3458 {
3459 int r;
3460 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
3461 kvm_run->request_interrupt_window;
3462
3463 if (vcpu->requests)
3464 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3465 kvm_mmu_unload(vcpu);
3466
3467 r = kvm_mmu_reload(vcpu);
3468 if (unlikely(r))
3469 goto out;
3470
3471 if (vcpu->requests) {
3472 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3473 __kvm_migrate_timers(vcpu);
3474 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3475 kvm_write_guest_time(vcpu);
3476 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3477 kvm_mmu_sync_roots(vcpu);
3478 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3479 kvm_x86_ops->tlb_flush(vcpu);
3480 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3481 &vcpu->requests)) {
3482 kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
3483 r = 0;
3484 goto out;
3485 }
3486 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3487 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
3488 r = 0;
3489 goto out;
3490 }
3491 }
3492
3493 preempt_disable();
3494
3495 kvm_x86_ops->prepare_guest_switch(vcpu);
3496 kvm_load_guest_fpu(vcpu);
3497
3498 local_irq_disable();
3499
3500 clear_bit(KVM_REQ_KICK, &vcpu->requests);
3501 smp_mb__after_clear_bit();
3502
3503 if (vcpu->requests || need_resched() || signal_pending(current)) {
3504 local_irq_enable();
3505 preempt_enable();
3506 r = 1;
3507 goto out;
3508 }
3509
3510 if (vcpu->arch.exception.pending)
3511 __queue_exception(vcpu);
3512 else
3513 inject_pending_irq(vcpu, kvm_run);
3514
3515 /* enable NMI/IRQ window open exits if needed */
3516 if (vcpu->arch.nmi_pending)
3517 kvm_x86_ops->enable_nmi_window(vcpu);
3518 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
3519 kvm_x86_ops->enable_irq_window(vcpu);
3520
3521 if (kvm_lapic_enabled(vcpu)) {
3522 update_cr8_intercept(vcpu);
3523 kvm_lapic_sync_to_vapic(vcpu);
3524 }
3525
3526 up_read(&vcpu->kvm->slots_lock);
3527
3528 kvm_guest_enter();
3529
3530 get_debugreg(vcpu->arch.host_dr6, 6);
3531 get_debugreg(vcpu->arch.host_dr7, 7);
3532 if (unlikely(vcpu->arch.switch_db_regs)) {
3533 get_debugreg(vcpu->arch.host_db[0], 0);
3534 get_debugreg(vcpu->arch.host_db[1], 1);
3535 get_debugreg(vcpu->arch.host_db[2], 2);
3536 get_debugreg(vcpu->arch.host_db[3], 3);
3537
3538 set_debugreg(0, 7);
3539 set_debugreg(vcpu->arch.eff_db[0], 0);
3540 set_debugreg(vcpu->arch.eff_db[1], 1);
3541 set_debugreg(vcpu->arch.eff_db[2], 2);
3542 set_debugreg(vcpu->arch.eff_db[3], 3);
3543 }
3544
3545 trace_kvm_entry(vcpu->vcpu_id);
3546 kvm_x86_ops->run(vcpu, kvm_run);
3547
3548 if (unlikely(vcpu->arch.switch_db_regs)) {
3549 set_debugreg(0, 7);
3550 set_debugreg(vcpu->arch.host_db[0], 0);
3551 set_debugreg(vcpu->arch.host_db[1], 1);
3552 set_debugreg(vcpu->arch.host_db[2], 2);
3553 set_debugreg(vcpu->arch.host_db[3], 3);
3554 }
3555 set_debugreg(vcpu->arch.host_dr6, 6);
3556 set_debugreg(vcpu->arch.host_dr7, 7);
3557
3558 set_bit(KVM_REQ_KICK, &vcpu->requests);
3559 local_irq_enable();
3560
3561 ++vcpu->stat.exits;
3562
3563 /*
3564 * We must have an instruction between local_irq_enable() and
3565 * kvm_guest_exit(), so the timer interrupt isn't delayed by
3566 * the interrupt shadow. The stat.exits increment will do nicely.
3567 * But we need to prevent reordering, hence this barrier():
3568 */
3569 barrier();
3570
3571 kvm_guest_exit();
3572
3573 preempt_enable();
3574
3575 down_read(&vcpu->kvm->slots_lock);
3576
3577 /*
3578 * Profile KVM exit RIPs:
3579 */
3580 if (unlikely(prof_on == KVM_PROFILING)) {
3581 unsigned long rip = kvm_rip_read(vcpu);
3582 profile_hit(KVM_PROFILING, (void *)rip);
3583 }
3584
3585
3586 kvm_lapic_sync_from_vapic(vcpu);
3587
3588 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
3589 out:
3590 return r;
3591 }
3592
3593
3594 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3595 {
3596 int r;
3597
3598 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3599 pr_debug("vcpu %d received sipi with vector # %x\n",
3600 vcpu->vcpu_id, vcpu->arch.sipi_vector);
3601 kvm_lapic_reset(vcpu);
3602 r = kvm_arch_vcpu_reset(vcpu);
3603 if (r)
3604 return r;
3605 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3606 }
3607
3608 down_read(&vcpu->kvm->slots_lock);
3609 vapic_enter(vcpu);
3610
3611 r = 1;
3612 while (r > 0) {
3613 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3614 r = vcpu_enter_guest(vcpu, kvm_run);
3615 else {
3616 up_read(&vcpu->kvm->slots_lock);
3617 kvm_vcpu_block(vcpu);
3618 down_read(&vcpu->kvm->slots_lock);
3619 if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3620 {
3621 switch(vcpu->arch.mp_state) {
3622 case KVM_MP_STATE_HALTED:
3623 vcpu->arch.mp_state =
3624 KVM_MP_STATE_RUNNABLE;
3625 case KVM_MP_STATE_RUNNABLE:
3626 break;
3627 case KVM_MP_STATE_SIPI_RECEIVED:
3628 default:
3629 r = -EINTR;
3630 break;
3631 }
3632 }
3633 }
3634
3635 if (r <= 0)
3636 break;
3637
3638 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
3639 if (kvm_cpu_has_pending_timer(vcpu))
3640 kvm_inject_pending_timer_irqs(vcpu);
3641
3642 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3643 r = -EINTR;
3644 kvm_run->exit_reason = KVM_EXIT_INTR;
3645 ++vcpu->stat.request_irq_exits;
3646 }
3647 if (signal_pending(current)) {
3648 r = -EINTR;
3649 kvm_run->exit_reason = KVM_EXIT_INTR;
3650 ++vcpu->stat.signal_exits;
3651 }
3652 if (need_resched()) {
3653 up_read(&vcpu->kvm->slots_lock);
3654 kvm_resched(vcpu);
3655 down_read(&vcpu->kvm->slots_lock);
3656 }
3657 }
3658
3659 up_read(&vcpu->kvm->slots_lock);
3660 post_kvm_run_save(vcpu, kvm_run);
3661
3662 vapic_exit(vcpu);
3663
3664 return r;
3665 }
3666
3667 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3668 {
3669 int r;
3670 sigset_t sigsaved;
3671
3672 vcpu_load(vcpu);
3673
3674 if (vcpu->sigset_active)
3675 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3676
3677 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3678 kvm_vcpu_block(vcpu);
3679 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3680 r = -EAGAIN;
3681 goto out;
3682 }
3683
3684 /* re-sync apic's tpr */
3685 if (!irqchip_in_kernel(vcpu->kvm))
3686 kvm_set_cr8(vcpu, kvm_run->cr8);
3687
3688 if (vcpu->arch.pio.cur_count) {
3689 r = complete_pio(vcpu);
3690 if (r)
3691 goto out;
3692 }
3693 #if CONFIG_HAS_IOMEM
3694 if (vcpu->mmio_needed) {
3695 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3696 vcpu->mmio_read_completed = 1;
3697 vcpu->mmio_needed = 0;
3698
3699 down_read(&vcpu->kvm->slots_lock);
3700 r = emulate_instruction(vcpu, kvm_run,
3701 vcpu->arch.mmio_fault_cr2, 0,
3702 EMULTYPE_NO_DECODE);
3703 up_read(&vcpu->kvm->slots_lock);
3704 if (r == EMULATE_DO_MMIO) {
3705 /*
3706 * Read-modify-write. Back to userspace.
3707 */
3708 r = 0;
3709 goto out;
3710 }
3711 }
3712 #endif
3713 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3714 kvm_register_write(vcpu, VCPU_REGS_RAX,
3715 kvm_run->hypercall.ret);
3716
3717 r = __vcpu_run(vcpu, kvm_run);
3718
3719 out:
3720 if (vcpu->sigset_active)
3721 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3722
3723 vcpu_put(vcpu);
3724 return r;
3725 }
3726
3727 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3728 {
3729 vcpu_load(vcpu);
3730
3731 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3732 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3733 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3734 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3735 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3736 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3737 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3738 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3739 #ifdef CONFIG_X86_64
3740 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3741 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3742 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3743 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3744 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3745 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3746 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3747 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3748 #endif
3749
3750 regs->rip = kvm_rip_read(vcpu);
3751 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3752
3753 /*
3754 * Don't leak debug flags in case they were set for guest debugging
3755 */
3756 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3757 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3758
3759 vcpu_put(vcpu);
3760
3761 return 0;
3762 }
3763
3764 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3765 {
3766 vcpu_load(vcpu);
3767
3768 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3769 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3770 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3771 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3772 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3773 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3774 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3775 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3776 #ifdef CONFIG_X86_64
3777 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3778 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3779 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3780 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3781 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3782 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3783 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3784 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3785
3786 #endif
3787
3788 kvm_rip_write(vcpu, regs->rip);
3789 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3790
3791
3792 vcpu->arch.exception.pending = false;
3793
3794 vcpu_put(vcpu);
3795
3796 return 0;
3797 }
3798
3799 void kvm_get_segment(struct kvm_vcpu *vcpu,
3800 struct kvm_segment *var, int seg)
3801 {
3802 kvm_x86_ops->get_segment(vcpu, var, seg);
3803 }
3804
3805 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3806 {
3807 struct kvm_segment cs;
3808
3809 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3810 *db = cs.db;
3811 *l = cs.l;
3812 }
3813 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3814
3815 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3816 struct kvm_sregs *sregs)
3817 {
3818 struct descriptor_table dt;
3819
3820 vcpu_load(vcpu);
3821
3822 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3823 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3824 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3825 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3826 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3827 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3828
3829 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3830 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3831
3832 kvm_x86_ops->get_idt(vcpu, &dt);
3833 sregs->idt.limit = dt.limit;
3834 sregs->idt.base = dt.base;
3835 kvm_x86_ops->get_gdt(vcpu, &dt);
3836 sregs->gdt.limit = dt.limit;
3837 sregs->gdt.base = dt.base;
3838
3839 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3840 sregs->cr0 = vcpu->arch.cr0;
3841 sregs->cr2 = vcpu->arch.cr2;
3842 sregs->cr3 = vcpu->arch.cr3;
3843 sregs->cr4 = vcpu->arch.cr4;
3844 sregs->cr8 = kvm_get_cr8(vcpu);
3845 sregs->efer = vcpu->arch.shadow_efer;
3846 sregs->apic_base = kvm_get_apic_base(vcpu);
3847
3848 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
3849
3850 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
3851 set_bit(vcpu->arch.interrupt.nr,
3852 (unsigned long *)sregs->interrupt_bitmap);
3853
3854 vcpu_put(vcpu);
3855
3856 return 0;
3857 }
3858
3859 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3860 struct kvm_mp_state *mp_state)
3861 {
3862 vcpu_load(vcpu);
3863 mp_state->mp_state = vcpu->arch.mp_state;
3864 vcpu_put(vcpu);
3865 return 0;
3866 }
3867
3868 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3869 struct kvm_mp_state *mp_state)
3870 {
3871 vcpu_load(vcpu);
3872 vcpu->arch.mp_state = mp_state->mp_state;
3873 vcpu_put(vcpu);
3874 return 0;
3875 }
3876
3877 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3878 struct kvm_segment *var, int seg)
3879 {
3880 kvm_x86_ops->set_segment(vcpu, var, seg);
3881 }
3882
3883 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3884 struct kvm_segment *kvm_desct)
3885 {
3886 kvm_desct->base = seg_desc->base0;
3887 kvm_desct->base |= seg_desc->base1 << 16;
3888 kvm_desct->base |= seg_desc->base2 << 24;
3889 kvm_desct->limit = seg_desc->limit0;
3890 kvm_desct->limit |= seg_desc->limit << 16;
3891 if (seg_desc->g) {
3892 kvm_desct->limit <<= 12;
3893 kvm_desct->limit |= 0xfff;
3894 }
3895 kvm_desct->selector = selector;
3896 kvm_desct->type = seg_desc->type;
3897 kvm_desct->present = seg_desc->p;
3898 kvm_desct->dpl = seg_desc->dpl;
3899 kvm_desct->db = seg_desc->d;
3900 kvm_desct->s = seg_desc->s;
3901 kvm_desct->l = seg_desc->l;
3902 kvm_desct->g = seg_desc->g;
3903 kvm_desct->avl = seg_desc->avl;
3904 if (!selector)
3905 kvm_desct->unusable = 1;
3906 else
3907 kvm_desct->unusable = 0;
3908 kvm_desct->padding = 0;
3909 }
3910
3911 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
3912 u16 selector,
3913 struct descriptor_table *dtable)
3914 {
3915 if (selector & 1 << 2) {
3916 struct kvm_segment kvm_seg;
3917
3918 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
3919
3920 if (kvm_seg.unusable)
3921 dtable->limit = 0;
3922 else
3923 dtable->limit = kvm_seg.limit;
3924 dtable->base = kvm_seg.base;
3925 }
3926 else
3927 kvm_x86_ops->get_gdt(vcpu, dtable);
3928 }
3929
3930 /* allowed just for 8 bytes segments */
3931 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3932 struct desc_struct *seg_desc)
3933 {
3934 gpa_t gpa;
3935 struct descriptor_table dtable;
3936 u16 index = selector >> 3;
3937
3938 get_segment_descriptor_dtable(vcpu, selector, &dtable);
3939
3940 if (dtable.limit < index * 8 + 7) {
3941 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
3942 return 1;
3943 }
3944 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3945 gpa += index * 8;
3946 return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
3947 }
3948
3949 /* allowed just for 8 bytes segments */
3950 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
3951 struct desc_struct *seg_desc)
3952 {
3953 gpa_t gpa;
3954 struct descriptor_table dtable;
3955 u16 index = selector >> 3;
3956
3957 get_segment_descriptor_dtable(vcpu, selector, &dtable);
3958
3959 if (dtable.limit < index * 8 + 7)
3960 return 1;
3961 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
3962 gpa += index * 8;
3963 return kvm_write_guest(vcpu->kvm, gpa, seg_desc, 8);
3964 }
3965
3966 static u32 get_tss_base_addr(struct kvm_vcpu *vcpu,
3967 struct desc_struct *seg_desc)
3968 {
3969 u32 base_addr;
3970
3971 base_addr = seg_desc->base0;
3972 base_addr |= (seg_desc->base1 << 16);
3973 base_addr |= (seg_desc->base2 << 24);
3974
3975 return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
3976 }
3977
3978 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
3979 {
3980 struct kvm_segment kvm_seg;
3981
3982 kvm_get_segment(vcpu, &kvm_seg, seg);
3983 return kvm_seg.selector;
3984 }
3985
3986 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
3987 u16 selector,
3988 struct kvm_segment *kvm_seg)
3989 {
3990 struct desc_struct seg_desc;
3991
3992 if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
3993 return 1;
3994 seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
3995 return 0;
3996 }
3997
3998 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
3999 {
4000 struct kvm_segment segvar = {
4001 .base = selector << 4,
4002 .limit = 0xffff,
4003 .selector = selector,
4004 .type = 3,
4005 .present = 1,
4006 .dpl = 3,
4007 .db = 0,
4008 .s = 1,
4009 .l = 0,
4010 .g = 0,
4011 .avl = 0,
4012 .unusable = 0,
4013 };
4014 kvm_x86_ops->set_segment(vcpu, &segvar, seg);
4015 return 0;
4016 }
4017
4018 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4019 int type_bits, int seg)
4020 {
4021 struct kvm_segment kvm_seg;
4022
4023 if (!(vcpu->arch.cr0 & X86_CR0_PE))
4024 return kvm_load_realmode_segment(vcpu, selector, seg);
4025 if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
4026 return 1;
4027 kvm_seg.type |= type_bits;
4028
4029 if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
4030 seg != VCPU_SREG_LDTR)
4031 if (!kvm_seg.s)
4032 kvm_seg.unusable = 1;
4033
4034 kvm_set_segment(vcpu, &kvm_seg, seg);
4035 return 0;
4036 }
4037
4038 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
4039 struct tss_segment_32 *tss)
4040 {
4041 tss->cr3 = vcpu->arch.cr3;
4042 tss->eip = kvm_rip_read(vcpu);
4043 tss->eflags = kvm_x86_ops->get_rflags(vcpu);
4044 tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4045 tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4046 tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4047 tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4048 tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4049 tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4050 tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4051 tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4052 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4053 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4054 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4055 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4056 tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
4057 tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
4058 tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4059 }
4060
4061 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
4062 struct tss_segment_32 *tss)
4063 {
4064 kvm_set_cr3(vcpu, tss->cr3);
4065
4066 kvm_rip_write(vcpu, tss->eip);
4067 kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
4068
4069 kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
4070 kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
4071 kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
4072 kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
4073 kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
4074 kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
4075 kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
4076 kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
4077
4078 if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
4079 return 1;
4080
4081 if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4082 return 1;
4083
4084 if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4085 return 1;
4086
4087 if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4088 return 1;
4089
4090 if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4091 return 1;
4092
4093 if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
4094 return 1;
4095
4096 if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
4097 return 1;
4098 return 0;
4099 }
4100
4101 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
4102 struct tss_segment_16 *tss)
4103 {
4104 tss->ip = kvm_rip_read(vcpu);
4105 tss->flag = kvm_x86_ops->get_rflags(vcpu);
4106 tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4107 tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4108 tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4109 tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4110 tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4111 tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4112 tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
4113 tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
4114
4115 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4116 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4117 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4118 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4119 tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4120 tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
4121 }
4122
4123 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
4124 struct tss_segment_16 *tss)
4125 {
4126 kvm_rip_write(vcpu, tss->ip);
4127 kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
4128 kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
4129 kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
4130 kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
4131 kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
4132 kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
4133 kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
4134 kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
4135 kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
4136
4137 if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
4138 return 1;
4139
4140 if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4141 return 1;
4142
4143 if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4144 return 1;
4145
4146 if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4147 return 1;
4148
4149 if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4150 return 1;
4151 return 0;
4152 }
4153
4154 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
4155 u16 old_tss_sel, u32 old_tss_base,
4156 struct desc_struct *nseg_desc)
4157 {
4158 struct tss_segment_16 tss_segment_16;
4159 int ret = 0;
4160
4161 if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4162 sizeof tss_segment_16))
4163 goto out;
4164
4165 save_state_to_tss16(vcpu, &tss_segment_16);
4166
4167 if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4168 sizeof tss_segment_16))
4169 goto out;
4170
4171 if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4172 &tss_segment_16, sizeof tss_segment_16))
4173 goto out;
4174
4175 if (old_tss_sel != 0xffff) {
4176 tss_segment_16.prev_task_link = old_tss_sel;
4177
4178 if (kvm_write_guest(vcpu->kvm,
4179 get_tss_base_addr(vcpu, nseg_desc),
4180 &tss_segment_16.prev_task_link,
4181 sizeof tss_segment_16.prev_task_link))
4182 goto out;
4183 }
4184
4185 if (load_state_from_tss16(vcpu, &tss_segment_16))
4186 goto out;
4187
4188 ret = 1;
4189 out:
4190 return ret;
4191 }
4192
4193 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
4194 u16 old_tss_sel, u32 old_tss_base,
4195 struct desc_struct *nseg_desc)
4196 {
4197 struct tss_segment_32 tss_segment_32;
4198 int ret = 0;
4199
4200 if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4201 sizeof tss_segment_32))
4202 goto out;
4203
4204 save_state_to_tss32(vcpu, &tss_segment_32);
4205
4206 if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4207 sizeof tss_segment_32))
4208 goto out;
4209
4210 if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4211 &tss_segment_32, sizeof tss_segment_32))
4212 goto out;
4213
4214 if (old_tss_sel != 0xffff) {
4215 tss_segment_32.prev_task_link = old_tss_sel;
4216
4217 if (kvm_write_guest(vcpu->kvm,
4218 get_tss_base_addr(vcpu, nseg_desc),
4219 &tss_segment_32.prev_task_link,
4220 sizeof tss_segment_32.prev_task_link))
4221 goto out;
4222 }
4223
4224 if (load_state_from_tss32(vcpu, &tss_segment_32))
4225 goto out;
4226
4227 ret = 1;
4228 out:
4229 return ret;
4230 }
4231
4232 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
4233 {
4234 struct kvm_segment tr_seg;
4235 struct desc_struct cseg_desc;
4236 struct desc_struct nseg_desc;
4237 int ret = 0;
4238 u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
4239 u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
4240
4241 old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
4242
4243 /* FIXME: Handle errors. Failure to read either TSS or their
4244 * descriptors should generate a pagefault.
4245 */
4246 if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
4247 goto out;
4248
4249 if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
4250 goto out;
4251
4252 if (reason != TASK_SWITCH_IRET) {
4253 int cpl;
4254
4255 cpl = kvm_x86_ops->get_cpl(vcpu);
4256 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
4257 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4258 return 1;
4259 }
4260 }
4261
4262 if (!nseg_desc.p || (nseg_desc.limit0 | nseg_desc.limit << 16) < 0x67) {
4263 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
4264 return 1;
4265 }
4266
4267 if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
4268 cseg_desc.type &= ~(1 << 1); //clear the B flag
4269 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
4270 }
4271
4272 if (reason == TASK_SWITCH_IRET) {
4273 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4274 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
4275 }
4276
4277 /* set back link to prev task only if NT bit is set in eflags
4278 note that old_tss_sel is not used afetr this point */
4279 if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4280 old_tss_sel = 0xffff;
4281
4282 /* set back link to prev task only if NT bit is set in eflags
4283 note that old_tss_sel is not used afetr this point */
4284 if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4285 old_tss_sel = 0xffff;
4286
4287 if (nseg_desc.type & 8)
4288 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
4289 old_tss_base, &nseg_desc);
4290 else
4291 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
4292 old_tss_base, &nseg_desc);
4293
4294 if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
4295 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4296 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
4297 }
4298
4299 if (reason != TASK_SWITCH_IRET) {
4300 nseg_desc.type |= (1 << 1);
4301 save_guest_segment_descriptor(vcpu, tss_selector,
4302 &nseg_desc);
4303 }
4304
4305 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
4306 seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
4307 tr_seg.type = 11;
4308 kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
4309 out:
4310 return ret;
4311 }
4312 EXPORT_SYMBOL_GPL(kvm_task_switch);
4313
4314 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4315 struct kvm_sregs *sregs)
4316 {
4317 int mmu_reset_needed = 0;
4318 int pending_vec, max_bits;
4319 struct descriptor_table dt;
4320
4321 vcpu_load(vcpu);
4322
4323 dt.limit = sregs->idt.limit;
4324 dt.base = sregs->idt.base;
4325 kvm_x86_ops->set_idt(vcpu, &dt);
4326 dt.limit = sregs->gdt.limit;
4327 dt.base = sregs->gdt.base;
4328 kvm_x86_ops->set_gdt(vcpu, &dt);
4329
4330 vcpu->arch.cr2 = sregs->cr2;
4331 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
4332
4333 down_read(&vcpu->kvm->slots_lock);
4334 if (gfn_to_memslot(vcpu->kvm, sregs->cr3 >> PAGE_SHIFT))
4335 vcpu->arch.cr3 = sregs->cr3;
4336 else
4337 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
4338 up_read(&vcpu->kvm->slots_lock);
4339
4340 kvm_set_cr8(vcpu, sregs->cr8);
4341
4342 mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
4343 kvm_x86_ops->set_efer(vcpu, sregs->efer);
4344 kvm_set_apic_base(vcpu, sregs->apic_base);
4345
4346 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
4347
4348 mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
4349 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
4350 vcpu->arch.cr0 = sregs->cr0;
4351
4352 mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
4353 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
4354 if (!is_long_mode(vcpu) && is_pae(vcpu))
4355 load_pdptrs(vcpu, vcpu->arch.cr3);
4356
4357 if (mmu_reset_needed)
4358 kvm_mmu_reset_context(vcpu);
4359
4360 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
4361 pending_vec = find_first_bit(
4362 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
4363 if (pending_vec < max_bits) {
4364 kvm_queue_interrupt(vcpu, pending_vec, false);
4365 pr_debug("Set back pending irq %d\n", pending_vec);
4366 if (irqchip_in_kernel(vcpu->kvm))
4367 kvm_pic_clear_isr_ack(vcpu->kvm);
4368 }
4369
4370 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4371 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4372 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4373 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4374 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4375 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4376
4377 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4378 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4379
4380 /* Older userspace won't unhalt the vcpu on reset. */
4381 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
4382 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4383 !(vcpu->arch.cr0 & X86_CR0_PE))
4384 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4385
4386 vcpu_put(vcpu);
4387
4388 return 0;
4389 }
4390
4391 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4392 struct kvm_guest_debug *dbg)
4393 {
4394 int i, r;
4395
4396 vcpu_load(vcpu);
4397
4398 if ((dbg->control & (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) ==
4399 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) {
4400 for (i = 0; i < KVM_NR_DB_REGS; ++i)
4401 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4402 vcpu->arch.switch_db_regs =
4403 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4404 } else {
4405 for (i = 0; i < KVM_NR_DB_REGS; i++)
4406 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4407 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4408 }
4409
4410 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
4411
4412 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4413 kvm_queue_exception(vcpu, DB_VECTOR);
4414 else if (dbg->control & KVM_GUESTDBG_INJECT_BP)
4415 kvm_queue_exception(vcpu, BP_VECTOR);
4416
4417 vcpu_put(vcpu);
4418
4419 return r;
4420 }
4421
4422 /*
4423 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
4424 * we have asm/x86/processor.h
4425 */
4426 struct fxsave {
4427 u16 cwd;
4428 u16 swd;
4429 u16 twd;
4430 u16 fop;
4431 u64 rip;
4432 u64 rdp;
4433 u32 mxcsr;
4434 u32 mxcsr_mask;
4435 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
4436 #ifdef CONFIG_X86_64
4437 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
4438 #else
4439 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
4440 #endif
4441 };
4442
4443 /*
4444 * Translate a guest virtual address to a guest physical address.
4445 */
4446 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4447 struct kvm_translation *tr)
4448 {
4449 unsigned long vaddr = tr->linear_address;
4450 gpa_t gpa;
4451
4452 vcpu_load(vcpu);
4453 down_read(&vcpu->kvm->slots_lock);
4454 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4455 up_read(&vcpu->kvm->slots_lock);
4456 tr->physical_address = gpa;
4457 tr->valid = gpa != UNMAPPED_GVA;
4458 tr->writeable = 1;
4459 tr->usermode = 0;
4460 vcpu_put(vcpu);
4461
4462 return 0;
4463 }
4464
4465 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4466 {
4467 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4468
4469 vcpu_load(vcpu);
4470
4471 memcpy(fpu->fpr, fxsave->st_space, 128);
4472 fpu->fcw = fxsave->cwd;
4473 fpu->fsw = fxsave->swd;
4474 fpu->ftwx = fxsave->twd;
4475 fpu->last_opcode = fxsave->fop;
4476 fpu->last_ip = fxsave->rip;
4477 fpu->last_dp = fxsave->rdp;
4478 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4479
4480 vcpu_put(vcpu);
4481
4482 return 0;
4483 }
4484
4485 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4486 {
4487 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4488
4489 vcpu_load(vcpu);
4490
4491 memcpy(fxsave->st_space, fpu->fpr, 128);
4492 fxsave->cwd = fpu->fcw;
4493 fxsave->swd = fpu->fsw;
4494 fxsave->twd = fpu->ftwx;
4495 fxsave->fop = fpu->last_opcode;
4496 fxsave->rip = fpu->last_ip;
4497 fxsave->rdp = fpu->last_dp;
4498 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4499
4500 vcpu_put(vcpu);
4501
4502 return 0;
4503 }
4504
4505 void fx_init(struct kvm_vcpu *vcpu)
4506 {
4507 unsigned after_mxcsr_mask;
4508
4509 /*
4510 * Touch the fpu the first time in non atomic context as if
4511 * this is the first fpu instruction the exception handler
4512 * will fire before the instruction returns and it'll have to
4513 * allocate ram with GFP_KERNEL.
4514 */
4515 if (!used_math())
4516 kvm_fx_save(&vcpu->arch.host_fx_image);
4517
4518 /* Initialize guest FPU by resetting ours and saving into guest's */
4519 preempt_disable();
4520 kvm_fx_save(&vcpu->arch.host_fx_image);
4521 kvm_fx_finit();
4522 kvm_fx_save(&vcpu->arch.guest_fx_image);
4523 kvm_fx_restore(&vcpu->arch.host_fx_image);
4524 preempt_enable();
4525
4526 vcpu->arch.cr0 |= X86_CR0_ET;
4527 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4528 vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4529 memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4530 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4531 }
4532 EXPORT_SYMBOL_GPL(fx_init);
4533
4534 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4535 {
4536 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4537 return;
4538
4539 vcpu->guest_fpu_loaded = 1;
4540 kvm_fx_save(&vcpu->arch.host_fx_image);
4541 kvm_fx_restore(&vcpu->arch.guest_fx_image);
4542 }
4543 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4544
4545 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4546 {
4547 if (!vcpu->guest_fpu_loaded)
4548 return;
4549
4550 vcpu->guest_fpu_loaded = 0;
4551 kvm_fx_save(&vcpu->arch.guest_fx_image);
4552 kvm_fx_restore(&vcpu->arch.host_fx_image);
4553 ++vcpu->stat.fpu_reload;
4554 }
4555 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4556
4557 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4558 {
4559 if (vcpu->arch.time_page) {
4560 kvm_release_page_dirty(vcpu->arch.time_page);
4561 vcpu->arch.time_page = NULL;
4562 }
4563
4564 kvm_x86_ops->vcpu_free(vcpu);
4565 }
4566
4567 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4568 unsigned int id)
4569 {
4570 return kvm_x86_ops->vcpu_create(kvm, id);
4571 }
4572
4573 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4574 {
4575 int r;
4576
4577 /* We do fxsave: this must be aligned. */
4578 BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4579
4580 vcpu->arch.mtrr_state.have_fixed = 1;
4581 vcpu_load(vcpu);
4582 r = kvm_arch_vcpu_reset(vcpu);
4583 if (r == 0)
4584 r = kvm_mmu_setup(vcpu);
4585 vcpu_put(vcpu);
4586 if (r < 0)
4587 goto free_vcpu;
4588
4589 return 0;
4590 free_vcpu:
4591 kvm_x86_ops->vcpu_free(vcpu);
4592 return r;
4593 }
4594
4595 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4596 {
4597 vcpu_load(vcpu);
4598 kvm_mmu_unload(vcpu);
4599 vcpu_put(vcpu);
4600
4601 kvm_x86_ops->vcpu_free(vcpu);
4602 }
4603
4604 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4605 {
4606 vcpu->arch.nmi_pending = false;
4607 vcpu->arch.nmi_injected = false;
4608
4609 vcpu->arch.switch_db_regs = 0;
4610 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
4611 vcpu->arch.dr6 = DR6_FIXED_1;
4612 vcpu->arch.dr7 = DR7_FIXED_1;
4613
4614 return kvm_x86_ops->vcpu_reset(vcpu);
4615 }
4616
4617 void kvm_arch_hardware_enable(void *garbage)
4618 {
4619 kvm_x86_ops->hardware_enable(garbage);
4620 }
4621
4622 void kvm_arch_hardware_disable(void *garbage)
4623 {
4624 kvm_x86_ops->hardware_disable(garbage);
4625 }
4626
4627 int kvm_arch_hardware_setup(void)
4628 {
4629 return kvm_x86_ops->hardware_setup();
4630 }
4631
4632 void kvm_arch_hardware_unsetup(void)
4633 {
4634 kvm_x86_ops->hardware_unsetup();
4635 }
4636
4637 void kvm_arch_check_processor_compat(void *rtn)
4638 {
4639 kvm_x86_ops->check_processor_compatibility(rtn);
4640 }
4641
4642 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4643 {
4644 struct page *page;
4645 struct kvm *kvm;
4646 int r;
4647
4648 BUG_ON(vcpu->kvm == NULL);
4649 kvm = vcpu->kvm;
4650
4651 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4652 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
4653 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4654 else
4655 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4656
4657 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4658 if (!page) {
4659 r = -ENOMEM;
4660 goto fail;
4661 }
4662 vcpu->arch.pio_data = page_address(page);
4663
4664 r = kvm_mmu_create(vcpu);
4665 if (r < 0)
4666 goto fail_free_pio_data;
4667
4668 if (irqchip_in_kernel(kvm)) {
4669 r = kvm_create_lapic(vcpu);
4670 if (r < 0)
4671 goto fail_mmu_destroy;
4672 }
4673
4674 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
4675 GFP_KERNEL);
4676 if (!vcpu->arch.mce_banks) {
4677 r = -ENOMEM;
4678 goto fail_mmu_destroy;
4679 }
4680 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
4681
4682 return 0;
4683
4684 fail_mmu_destroy:
4685 kvm_mmu_destroy(vcpu);
4686 fail_free_pio_data:
4687 free_page((unsigned long)vcpu->arch.pio_data);
4688 fail:
4689 return r;
4690 }
4691
4692 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4693 {
4694 kvm_free_lapic(vcpu);
4695 down_read(&vcpu->kvm->slots_lock);
4696 kvm_mmu_destroy(vcpu);
4697 up_read(&vcpu->kvm->slots_lock);
4698 free_page((unsigned long)vcpu->arch.pio_data);
4699 }
4700
4701 struct kvm *kvm_arch_create_vm(void)
4702 {
4703 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4704
4705 if (!kvm)
4706 return ERR_PTR(-ENOMEM);
4707
4708 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4709 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4710
4711 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4712 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4713
4714 rdtscll(kvm->arch.vm_init_tsc);
4715
4716 return kvm;
4717 }
4718
4719 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4720 {
4721 vcpu_load(vcpu);
4722 kvm_mmu_unload(vcpu);
4723 vcpu_put(vcpu);
4724 }
4725
4726 static void kvm_free_vcpus(struct kvm *kvm)
4727 {
4728 unsigned int i;
4729 struct kvm_vcpu *vcpu;
4730
4731 /*
4732 * Unpin any mmu pages first.
4733 */
4734 kvm_for_each_vcpu(i, vcpu, kvm)
4735 kvm_unload_vcpu_mmu(vcpu);
4736 kvm_for_each_vcpu(i, vcpu, kvm)
4737 kvm_arch_vcpu_free(vcpu);
4738
4739 mutex_lock(&kvm->lock);
4740 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
4741 kvm->vcpus[i] = NULL;
4742
4743 atomic_set(&kvm->online_vcpus, 0);
4744 mutex_unlock(&kvm->lock);
4745 }
4746
4747 void kvm_arch_sync_events(struct kvm *kvm)
4748 {
4749 kvm_free_all_assigned_devices(kvm);
4750 }
4751
4752 void kvm_arch_destroy_vm(struct kvm *kvm)
4753 {
4754 kvm_iommu_unmap_guest(kvm);
4755 kvm_free_pit(kvm);
4756 kfree(kvm->arch.vpic);
4757 kfree(kvm->arch.vioapic);
4758 kvm_free_vcpus(kvm);
4759 kvm_free_physmem(kvm);
4760 if (kvm->arch.apic_access_page)
4761 put_page(kvm->arch.apic_access_page);
4762 if (kvm->arch.ept_identity_pagetable)
4763 put_page(kvm->arch.ept_identity_pagetable);
4764 kfree(kvm);
4765 }
4766
4767 int kvm_arch_set_memory_region(struct kvm *kvm,
4768 struct kvm_userspace_memory_region *mem,
4769 struct kvm_memory_slot old,
4770 int user_alloc)
4771 {
4772 int npages = mem->memory_size >> PAGE_SHIFT;
4773 struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4774
4775 /*To keep backward compatibility with older userspace,
4776 *x86 needs to hanlde !user_alloc case.
4777 */
4778 if (!user_alloc) {
4779 if (npages && !old.rmap) {
4780 unsigned long userspace_addr;
4781
4782 down_write(&current->mm->mmap_sem);
4783 userspace_addr = do_mmap(NULL, 0,
4784 npages * PAGE_SIZE,
4785 PROT_READ | PROT_WRITE,
4786 MAP_PRIVATE | MAP_ANONYMOUS,
4787 0);
4788 up_write(&current->mm->mmap_sem);
4789
4790 if (IS_ERR((void *)userspace_addr))
4791 return PTR_ERR((void *)userspace_addr);
4792
4793 /* set userspace_addr atomically for kvm_hva_to_rmapp */
4794 spin_lock(&kvm->mmu_lock);
4795 memslot->userspace_addr = userspace_addr;
4796 spin_unlock(&kvm->mmu_lock);
4797 } else {
4798 if (!old.user_alloc && old.rmap) {
4799 int ret;
4800
4801 down_write(&current->mm->mmap_sem);
4802 ret = do_munmap(current->mm, old.userspace_addr,
4803 old.npages * PAGE_SIZE);
4804 up_write(&current->mm->mmap_sem);
4805 if (ret < 0)
4806 printk(KERN_WARNING
4807 "kvm_vm_ioctl_set_memory_region: "
4808 "failed to munmap memory\n");
4809 }
4810 }
4811 }
4812
4813 spin_lock(&kvm->mmu_lock);
4814 if (!kvm->arch.n_requested_mmu_pages) {
4815 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4816 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4817 }
4818
4819 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4820 spin_unlock(&kvm->mmu_lock);
4821 kvm_flush_remote_tlbs(kvm);
4822
4823 return 0;
4824 }
4825
4826 void kvm_arch_flush_shadow(struct kvm *kvm)
4827 {
4828 kvm_mmu_zap_all(kvm);
4829 kvm_reload_remote_mmus(kvm);
4830 }
4831
4832 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4833 {
4834 return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4835 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4836 || vcpu->arch.nmi_pending;
4837 }
4838
4839 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4840 {
4841 int me;
4842 int cpu = vcpu->cpu;
4843
4844 if (waitqueue_active(&vcpu->wq)) {
4845 wake_up_interruptible(&vcpu->wq);
4846 ++vcpu->stat.halt_wakeup;
4847 }
4848
4849 me = get_cpu();
4850 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
4851 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
4852 smp_send_reschedule(cpu);
4853 put_cpu();
4854 }
4855
4856 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
4857 {
4858 return kvm_x86_ops->interrupt_allowed(vcpu);
4859 }
4860
4861 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
4862 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
4863 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
4864 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
4865 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
(deprecated) Btrfs devel tree