?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
x86/amd-iommu: Add per IOMMU reference counting
[linux/fpc-iii.git] / arch / x86 / kvm / x86.c
blobae07d261527cba458ed1682118b19295bc997847
1 /*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * derived from drivers/kvm/kvm_main.c
5 *
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9 *
10 * Authors:
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 * Amit Shah <amit.shah@qumranet.com>
14 * Ben-Ami Yassour <benami@il.ibm.com>
15 *
16 * This work is licensed under the terms of the GNU GPL, version 2. See
17 * the COPYING file in the top-level directory.
18 *
19 */
20
21 #include <linux/kvm_host.h>
22 #include "irq.h"
23 #include "mmu.h"
24 #include "i8254.h"
25 #include "tss.h"
26 #include "kvm_cache_regs.h"
27 #include "x86.h"
28
29 #include <linux/clocksource.h>
30 #include <linux/interrupt.h>
31 #include <linux/kvm.h>
32 #include <linux/fs.h>
33 #include <linux/vmalloc.h>
34 #include <linux/module.h>
35 #include <linux/mman.h>
36 #include <linux/highmem.h>
37 #include <linux/iommu.h>
38 #include <linux/intel-iommu.h>
39 #include <linux/cpufreq.h>
40 #include <trace/events/kvm.h>
41 #undef TRACE_INCLUDE_FILE
42 #define CREATE_TRACE_POINTS
43 #include "trace.h"
44
45 #include <asm/uaccess.h>
46 #include <asm/msr.h>
47 #include <asm/desc.h>
48 #include <asm/mtrr.h>
49 #include <asm/mce.h>
50
51 #define MAX_IO_MSRS 256
52 #define CR0_RESERVED_BITS \
53 (~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
54 | X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
55 | X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
56 #define CR4_RESERVED_BITS \
57 (~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
58 | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
59 | X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
60 | X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
61
62 #define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
63
64 #define KVM_MAX_MCE_BANKS 32
65 #define KVM_MCE_CAP_SUPPORTED MCG_CTL_P
66
67 /* EFER defaults:
68 * - enable syscall per default because its emulated by KVM
69 * - enable LME and LMA per default on 64 bit KVM
70 */
71 #ifdef CONFIG_X86_64
72 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffafeULL;
73 #else
74 static u64 __read_mostly efer_reserved_bits = 0xfffffffffffffffeULL;
75 #endif
76
77 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
78 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
79
80 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
81 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
82 struct kvm_cpuid_entry2 __user *entries);
83
84 struct kvm_x86_ops *kvm_x86_ops;
85 EXPORT_SYMBOL_GPL(kvm_x86_ops);
86
87 int ignore_msrs = 0;
88 module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR);
89
90 struct kvm_stats_debugfs_item debugfs_entries[] = {
91 { "pf_fixed", VCPU_STAT(pf_fixed) },
92 { "pf_guest", VCPU_STAT(pf_guest) },
93 { "tlb_flush", VCPU_STAT(tlb_flush) },
94 { "invlpg", VCPU_STAT(invlpg) },
95 { "exits", VCPU_STAT(exits) },
96 { "io_exits", VCPU_STAT(io_exits) },
97 { "mmio_exits", VCPU_STAT(mmio_exits) },
98 { "signal_exits", VCPU_STAT(signal_exits) },
99 { "irq_window", VCPU_STAT(irq_window_exits) },
100 { "nmi_window", VCPU_STAT(nmi_window_exits) },
101 { "halt_exits", VCPU_STAT(halt_exits) },
102 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
103 { "hypercalls", VCPU_STAT(hypercalls) },
104 { "request_irq", VCPU_STAT(request_irq_exits) },
105 { "irq_exits", VCPU_STAT(irq_exits) },
106 { "host_state_reload", VCPU_STAT(host_state_reload) },
107 { "efer_reload", VCPU_STAT(efer_reload) },
108 { "fpu_reload", VCPU_STAT(fpu_reload) },
109 { "insn_emulation", VCPU_STAT(insn_emulation) },
110 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
111 { "irq_injections", VCPU_STAT(irq_injections) },
112 { "nmi_injections", VCPU_STAT(nmi_injections) },
113 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
114 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
115 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
116 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
117 { "mmu_flooded", VM_STAT(mmu_flooded) },
118 { "mmu_recycled", VM_STAT(mmu_recycled) },
119 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
120 { "mmu_unsync", VM_STAT(mmu_unsync) },
121 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
122 { "largepages", VM_STAT(lpages) },
123 { NULL }
124 };
125
126 unsigned long segment_base(u16 selector)
127 {
128 struct descriptor_table gdt;
129 struct desc_struct *d;
130 unsigned long table_base;
131 unsigned long v;
132
133 if (selector == 0)
134 return 0;
135
136 kvm_get_gdt(&gdt);
137 table_base = gdt.base;
138
139 if (selector & 4) { /* from ldt */
140 u16 ldt_selector = kvm_read_ldt();
141
142 table_base = segment_base(ldt_selector);
143 }
144 d = (struct desc_struct *)(table_base + (selector & ~7));
145 v = get_desc_base(d);
146 #ifdef CONFIG_X86_64
147 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
148 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
149 #endif
150 return v;
151 }
152 EXPORT_SYMBOL_GPL(segment_base);
153
154 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
155 {
156 if (irqchip_in_kernel(vcpu->kvm))
157 return vcpu->arch.apic_base;
158 else
159 return vcpu->arch.apic_base;
160 }
161 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
162
163 void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
164 {
165 /* TODO: reserve bits check */
166 if (irqchip_in_kernel(vcpu->kvm))
167 kvm_lapic_set_base(vcpu, data);
168 else
169 vcpu->arch.apic_base = data;
170 }
171 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
172
173 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
174 {
175 WARN_ON(vcpu->arch.exception.pending);
176 vcpu->arch.exception.pending = true;
177 vcpu->arch.exception.has_error_code = false;
178 vcpu->arch.exception.nr = nr;
179 }
180 EXPORT_SYMBOL_GPL(kvm_queue_exception);
181
182 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long addr,
183 u32 error_code)
184 {
185 ++vcpu->stat.pf_guest;
186
187 if (vcpu->arch.exception.pending) {
188 switch(vcpu->arch.exception.nr) {
189 case DF_VECTOR:
190 /* triple fault -> shutdown */
191 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
192 return;
193 case PF_VECTOR:
194 vcpu->arch.exception.nr = DF_VECTOR;
195 vcpu->arch.exception.error_code = 0;
196 return;
197 default:
198 /* replace previous exception with a new one in a hope
199 that instruction re-execution will regenerate lost
200 exception */
201 vcpu->arch.exception.pending = false;
202 break;
203 }
204 }
205 vcpu->arch.cr2 = addr;
206 kvm_queue_exception_e(vcpu, PF_VECTOR, error_code);
207 }
208
209 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
210 {
211 vcpu->arch.nmi_pending = 1;
212 }
213 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
214
215 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
216 {
217 WARN_ON(vcpu->arch.exception.pending);
218 vcpu->arch.exception.pending = true;
219 vcpu->arch.exception.has_error_code = true;
220 vcpu->arch.exception.nr = nr;
221 vcpu->arch.exception.error_code = error_code;
222 }
223 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
224
225 /*
226 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
227 * a #GP and return false.
228 */
229 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
230 {
231 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
232 return true;
233 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
234 return false;
235 }
236 EXPORT_SYMBOL_GPL(kvm_require_cpl);
237
238 /*
239 * Load the pae pdptrs. Return true is they are all valid.
240 */
241 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
242 {
243 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
244 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
245 int i;
246 int ret;
247 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
248
249 ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
250 offset * sizeof(u64), sizeof(pdpte));
251 if (ret < 0) {
252 ret = 0;
253 goto out;
254 }
255 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
256 if (is_present_gpte(pdpte[i]) &&
257 (pdpte[i] & vcpu->arch.mmu.rsvd_bits_mask[0][2])) {
258 ret = 0;
259 goto out;
260 }
261 }
262 ret = 1;
263
264 memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
265 __set_bit(VCPU_EXREG_PDPTR,
266 (unsigned long *)&vcpu->arch.regs_avail);
267 __set_bit(VCPU_EXREG_PDPTR,
268 (unsigned long *)&vcpu->arch.regs_dirty);
269 out:
270
271 return ret;
272 }
273 EXPORT_SYMBOL_GPL(load_pdptrs);
274
275 static bool pdptrs_changed(struct kvm_vcpu *vcpu)
276 {
277 u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
278 bool changed = true;
279 int r;
280
281 if (is_long_mode(vcpu) || !is_pae(vcpu))
282 return false;
283
284 if (!test_bit(VCPU_EXREG_PDPTR,
285 (unsigned long *)&vcpu->arch.regs_avail))
286 return true;
287
288 r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
289 if (r < 0)
290 goto out;
291 changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
292 out:
293
294 return changed;
295 }
296
297 void kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
298 {
299 if (cr0 & CR0_RESERVED_BITS) {
300 printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
301 cr0, vcpu->arch.cr0);
302 kvm_inject_gp(vcpu, 0);
303 return;
304 }
305
306 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
307 printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
308 kvm_inject_gp(vcpu, 0);
309 return;
310 }
311
312 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
313 printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
314 "and a clear PE flag\n");
315 kvm_inject_gp(vcpu, 0);
316 return;
317 }
318
319 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
320 #ifdef CONFIG_X86_64
321 if ((vcpu->arch.shadow_efer & EFER_LME)) {
322 int cs_db, cs_l;
323
324 if (!is_pae(vcpu)) {
325 printk(KERN_DEBUG "set_cr0: #GP, start paging "
326 "in long mode while PAE is disabled\n");
327 kvm_inject_gp(vcpu, 0);
328 return;
329 }
330 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
331 if (cs_l) {
332 printk(KERN_DEBUG "set_cr0: #GP, start paging "
333 "in long mode while CS.L == 1\n");
334 kvm_inject_gp(vcpu, 0);
335 return;
336
337 }
338 } else
339 #endif
340 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
341 printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
342 "reserved bits\n");
343 kvm_inject_gp(vcpu, 0);
344 return;
345 }
346
347 }
348
349 kvm_x86_ops->set_cr0(vcpu, cr0);
350 vcpu->arch.cr0 = cr0;
351
352 kvm_mmu_reset_context(vcpu);
353 return;
354 }
355 EXPORT_SYMBOL_GPL(kvm_set_cr0);
356
357 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
358 {
359 kvm_set_cr0(vcpu, (vcpu->arch.cr0 & ~0x0ful) | (msw & 0x0f));
360 }
361 EXPORT_SYMBOL_GPL(kvm_lmsw);
362
363 void kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
364 {
365 unsigned long old_cr4 = vcpu->arch.cr4;
366 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE;
367
368 if (cr4 & CR4_RESERVED_BITS) {
369 printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
370 kvm_inject_gp(vcpu, 0);
371 return;
372 }
373
374 if (is_long_mode(vcpu)) {
375 if (!(cr4 & X86_CR4_PAE)) {
376 printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
377 "in long mode\n");
378 kvm_inject_gp(vcpu, 0);
379 return;
380 }
381 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
382 && ((cr4 ^ old_cr4) & pdptr_bits)
383 && !load_pdptrs(vcpu, vcpu->arch.cr3)) {
384 printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
385 kvm_inject_gp(vcpu, 0);
386 return;
387 }
388
389 if (cr4 & X86_CR4_VMXE) {
390 printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
391 kvm_inject_gp(vcpu, 0);
392 return;
393 }
394 kvm_x86_ops->set_cr4(vcpu, cr4);
395 vcpu->arch.cr4 = cr4;
396 vcpu->arch.mmu.base_role.cr4_pge = (cr4 & X86_CR4_PGE) && !tdp_enabled;
397 kvm_mmu_reset_context(vcpu);
398 }
399 EXPORT_SYMBOL_GPL(kvm_set_cr4);
400
401 void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
402 {
403 if (cr3 == vcpu->arch.cr3 && !pdptrs_changed(vcpu)) {
404 kvm_mmu_sync_roots(vcpu);
405 kvm_mmu_flush_tlb(vcpu);
406 return;
407 }
408
409 if (is_long_mode(vcpu)) {
410 if (cr3 & CR3_L_MODE_RESERVED_BITS) {
411 printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
412 kvm_inject_gp(vcpu, 0);
413 return;
414 }
415 } else {
416 if (is_pae(vcpu)) {
417 if (cr3 & CR3_PAE_RESERVED_BITS) {
418 printk(KERN_DEBUG
419 "set_cr3: #GP, reserved bits\n");
420 kvm_inject_gp(vcpu, 0);
421 return;
422 }
423 if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
424 printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
425 "reserved bits\n");
426 kvm_inject_gp(vcpu, 0);
427 return;
428 }
429 }
430 /*
431 * We don't check reserved bits in nonpae mode, because
432 * this isn't enforced, and VMware depends on this.
433 */
434 }
435
436 /*
437 * Does the new cr3 value map to physical memory? (Note, we
438 * catch an invalid cr3 even in real-mode, because it would
439 * cause trouble later on when we turn on paging anyway.)
440 *
441 * A real CPU would silently accept an invalid cr3 and would
442 * attempt to use it - with largely undefined (and often hard
443 * to debug) behavior on the guest side.
444 */
445 if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
446 kvm_inject_gp(vcpu, 0);
447 else {
448 vcpu->arch.cr3 = cr3;
449 vcpu->arch.mmu.new_cr3(vcpu);
450 }
451 }
452 EXPORT_SYMBOL_GPL(kvm_set_cr3);
453
454 void kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
455 {
456 if (cr8 & CR8_RESERVED_BITS) {
457 printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
458 kvm_inject_gp(vcpu, 0);
459 return;
460 }
461 if (irqchip_in_kernel(vcpu->kvm))
462 kvm_lapic_set_tpr(vcpu, cr8);
463 else
464 vcpu->arch.cr8 = cr8;
465 }
466 EXPORT_SYMBOL_GPL(kvm_set_cr8);
467
468 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
469 {
470 if (irqchip_in_kernel(vcpu->kvm))
471 return kvm_lapic_get_cr8(vcpu);
472 else
473 return vcpu->arch.cr8;
474 }
475 EXPORT_SYMBOL_GPL(kvm_get_cr8);
476
477 static inline u32 bit(int bitno)
478 {
479 return 1 << (bitno & 31);
480 }
481
482 /*
483 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
484 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
485 *
486 * This list is modified at module load time to reflect the
487 * capabilities of the host cpu.
488 */
489 static u32 msrs_to_save[] = {
490 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
491 MSR_K6_STAR,
492 #ifdef CONFIG_X86_64
493 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
494 #endif
495 MSR_IA32_TSC, MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
496 MSR_IA32_PERF_STATUS, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA
497 };
498
499 static unsigned num_msrs_to_save;
500
501 static u32 emulated_msrs[] = {
502 MSR_IA32_MISC_ENABLE,
503 };
504
505 static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
506 {
507 if (efer & efer_reserved_bits) {
508 printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
509 efer);
510 kvm_inject_gp(vcpu, 0);
511 return;
512 }
513
514 if (is_paging(vcpu)
515 && (vcpu->arch.shadow_efer & EFER_LME) != (efer & EFER_LME)) {
516 printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
517 kvm_inject_gp(vcpu, 0);
518 return;
519 }
520
521 if (efer & EFER_FFXSR) {
522 struct kvm_cpuid_entry2 *feat;
523
524 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
525 if (!feat || !(feat->edx & bit(X86_FEATURE_FXSR_OPT))) {
526 printk(KERN_DEBUG "set_efer: #GP, enable FFXSR w/o CPUID capability\n");
527 kvm_inject_gp(vcpu, 0);
528 return;
529 }
530 }
531
532 if (efer & EFER_SVME) {
533 struct kvm_cpuid_entry2 *feat;
534
535 feat = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
536 if (!feat || !(feat->ecx & bit(X86_FEATURE_SVM))) {
537 printk(KERN_DEBUG "set_efer: #GP, enable SVM w/o SVM\n");
538 kvm_inject_gp(vcpu, 0);
539 return;
540 }
541 }
542
543 kvm_x86_ops->set_efer(vcpu, efer);
544
545 efer &= ~EFER_LMA;
546 efer |= vcpu->arch.shadow_efer & EFER_LMA;
547
548 vcpu->arch.shadow_efer = efer;
549
550 vcpu->arch.mmu.base_role.nxe = (efer & EFER_NX) && !tdp_enabled;
551 kvm_mmu_reset_context(vcpu);
552 }
553
554 void kvm_enable_efer_bits(u64 mask)
555 {
556 efer_reserved_bits &= ~mask;
557 }
558 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
559
560
561 /*
562 * Writes msr value into into the appropriate "register".
563 * Returns 0 on success, non-0 otherwise.
564 * Assumes vcpu_load() was already called.
565 */
566 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
567 {
568 return kvm_x86_ops->set_msr(vcpu, msr_index, data);
569 }
570
571 /*
572 * Adapt set_msr() to msr_io()'s calling convention
573 */
574 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
575 {
576 return kvm_set_msr(vcpu, index, *data);
577 }
578
579 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
580 {
581 static int version;
582 struct pvclock_wall_clock wc;
583 struct timespec now, sys, boot;
584
585 if (!wall_clock)
586 return;
587
588 version++;
589
590 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
591
592 /*
593 * The guest calculates current wall clock time by adding
594 * system time (updated by kvm_write_guest_time below) to the
595 * wall clock specified here. guest system time equals host
596 * system time for us, thus we must fill in host boot time here.
597 */
598 now = current_kernel_time();
599 ktime_get_ts(&sys);
600 boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
601
602 wc.sec = boot.tv_sec;
603 wc.nsec = boot.tv_nsec;
604 wc.version = version;
605
606 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
607
608 version++;
609 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
610 }
611
612 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
613 {
614 uint32_t quotient, remainder;
615
616 /* Don't try to replace with do_div(), this one calculates
617 * "(dividend << 32) / divisor" */
618 __asm__ ( "divl %4"
619 : "=a" (quotient), "=d" (remainder)
620 : "0" (0), "1" (dividend), "r" (divisor) );
621 return quotient;
622 }
623
624 static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
625 {
626 uint64_t nsecs = 1000000000LL;
627 int32_t shift = 0;
628 uint64_t tps64;
629 uint32_t tps32;
630
631 tps64 = tsc_khz * 1000LL;
632 while (tps64 > nsecs*2) {
633 tps64 >>= 1;
634 shift--;
635 }
636
637 tps32 = (uint32_t)tps64;
638 while (tps32 <= (uint32_t)nsecs) {
639 tps32 <<= 1;
640 shift++;
641 }
642
643 hv_clock->tsc_shift = shift;
644 hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
645
646 pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
647 __func__, tsc_khz, hv_clock->tsc_shift,
648 hv_clock->tsc_to_system_mul);
649 }
650
651 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
652
653 static void kvm_write_guest_time(struct kvm_vcpu *v)
654 {
655 struct timespec ts;
656 unsigned long flags;
657 struct kvm_vcpu_arch *vcpu = &v->arch;
658 void *shared_kaddr;
659 unsigned long this_tsc_khz;
660
661 if ((!vcpu->time_page))
662 return;
663
664 this_tsc_khz = get_cpu_var(cpu_tsc_khz);
665 if (unlikely(vcpu->hv_clock_tsc_khz != this_tsc_khz)) {
666 kvm_set_time_scale(this_tsc_khz, &vcpu->hv_clock);
667 vcpu->hv_clock_tsc_khz = this_tsc_khz;
668 }
669 put_cpu_var(cpu_tsc_khz);
670
671 /* Keep irq disabled to prevent changes to the clock */
672 local_irq_save(flags);
673 kvm_get_msr(v, MSR_IA32_TSC, &vcpu->hv_clock.tsc_timestamp);
674 ktime_get_ts(&ts);
675 local_irq_restore(flags);
676
677 /* With all the info we got, fill in the values */
678
679 vcpu->hv_clock.system_time = ts.tv_nsec +
680 (NSEC_PER_SEC * (u64)ts.tv_sec);
681 /*
682 * The interface expects us to write an even number signaling that the
683 * update is finished. Since the guest won't see the intermediate
684 * state, we just increase by 2 at the end.
685 */
686 vcpu->hv_clock.version += 2;
687
688 shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
689
690 memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
691 sizeof(vcpu->hv_clock));
692
693 kunmap_atomic(shared_kaddr, KM_USER0);
694
695 mark_page_dirty(v->kvm, vcpu->time >> PAGE_SHIFT);
696 }
697
698 static int kvm_request_guest_time_update(struct kvm_vcpu *v)
699 {
700 struct kvm_vcpu_arch *vcpu = &v->arch;
701
702 if (!vcpu->time_page)
703 return 0;
704 set_bit(KVM_REQ_KVMCLOCK_UPDATE, &v->requests);
705 return 1;
706 }
707
708 static bool msr_mtrr_valid(unsigned msr)
709 {
710 switch (msr) {
711 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
712 case MSR_MTRRfix64K_00000:
713 case MSR_MTRRfix16K_80000:
714 case MSR_MTRRfix16K_A0000:
715 case MSR_MTRRfix4K_C0000:
716 case MSR_MTRRfix4K_C8000:
717 case MSR_MTRRfix4K_D0000:
718 case MSR_MTRRfix4K_D8000:
719 case MSR_MTRRfix4K_E0000:
720 case MSR_MTRRfix4K_E8000:
721 case MSR_MTRRfix4K_F0000:
722 case MSR_MTRRfix4K_F8000:
723 case MSR_MTRRdefType:
724 case MSR_IA32_CR_PAT:
725 return true;
726 case 0x2f8:
727 return true;
728 }
729 return false;
730 }
731
732 static bool valid_pat_type(unsigned t)
733 {
734 return t < 8 && (1 << t) & 0xf3; /* 0, 1, 4, 5, 6, 7 */
735 }
736
737 static bool valid_mtrr_type(unsigned t)
738 {
739 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
740 }
741
742 static bool mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
743 {
744 int i;
745
746 if (!msr_mtrr_valid(msr))
747 return false;
748
749 if (msr == MSR_IA32_CR_PAT) {
750 for (i = 0; i < 8; i++)
751 if (!valid_pat_type((data >> (i * 8)) & 0xff))
752 return false;
753 return true;
754 } else if (msr == MSR_MTRRdefType) {
755 if (data & ~0xcff)
756 return false;
757 return valid_mtrr_type(data & 0xff);
758 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
759 for (i = 0; i < 8 ; i++)
760 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
761 return false;
762 return true;
763 }
764
765 /* variable MTRRs */
766 return valid_mtrr_type(data & 0xff);
767 }
768
769 static int set_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
770 {
771 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
772
773 if (!mtrr_valid(vcpu, msr, data))
774 return 1;
775
776 if (msr == MSR_MTRRdefType) {
777 vcpu->arch.mtrr_state.def_type = data;
778 vcpu->arch.mtrr_state.enabled = (data & 0xc00) >> 10;
779 } else if (msr == MSR_MTRRfix64K_00000)
780 p[0] = data;
781 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
782 p[1 + msr - MSR_MTRRfix16K_80000] = data;
783 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
784 p[3 + msr - MSR_MTRRfix4K_C0000] = data;
785 else if (msr == MSR_IA32_CR_PAT)
786 vcpu->arch.pat = data;
787 else { /* Variable MTRRs */
788 int idx, is_mtrr_mask;
789 u64 *pt;
790
791 idx = (msr - 0x200) / 2;
792 is_mtrr_mask = msr - 0x200 - 2 * idx;
793 if (!is_mtrr_mask)
794 pt =
795 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
796 else
797 pt =
798 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
799 *pt = data;
800 }
801
802 kvm_mmu_reset_context(vcpu);
803 return 0;
804 }
805
806 static int set_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 data)
807 {
808 u64 mcg_cap = vcpu->arch.mcg_cap;
809 unsigned bank_num = mcg_cap & 0xff;
810
811 switch (msr) {
812 case MSR_IA32_MCG_STATUS:
813 vcpu->arch.mcg_status = data;
814 break;
815 case MSR_IA32_MCG_CTL:
816 if (!(mcg_cap & MCG_CTL_P))
817 return 1;
818 if (data != 0 && data != ~(u64)0)
819 return -1;
820 vcpu->arch.mcg_ctl = data;
821 break;
822 default:
823 if (msr >= MSR_IA32_MC0_CTL &&
824 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
825 u32 offset = msr - MSR_IA32_MC0_CTL;
826 /* only 0 or all 1s can be written to IA32_MCi_CTL */
827 if ((offset & 0x3) == 0 &&
828 data != 0 && data != ~(u64)0)
829 return -1;
830 vcpu->arch.mce_banks[offset] = data;
831 break;
832 }
833 return 1;
834 }
835 return 0;
836 }
837
838 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
839 {
840 switch (msr) {
841 case MSR_EFER:
842 set_efer(vcpu, data);
843 break;
844 case MSR_K7_HWCR:
845 data &= ~(u64)0x40; /* ignore flush filter disable */
846 if (data != 0) {
847 pr_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
848 data);
849 return 1;
850 }
851 break;
852 case MSR_FAM10H_MMIO_CONF_BASE:
853 if (data != 0) {
854 pr_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
855 "0x%llx\n", data);
856 return 1;
857 }
858 break;
859 case MSR_AMD64_NB_CFG:
860 break;
861 case MSR_IA32_DEBUGCTLMSR:
862 if (!data) {
863 /* We support the non-activated case already */
864 break;
865 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
866 /* Values other than LBR and BTF are vendor-specific,
867 thus reserved and should throw a #GP */
868 return 1;
869 }
870 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
871 __func__, data);
872 break;
873 case MSR_IA32_UCODE_REV:
874 case MSR_IA32_UCODE_WRITE:
875 case MSR_VM_HSAVE_PA:
876 case MSR_AMD64_PATCH_LOADER:
877 break;
878 case 0x200 ... 0x2ff:
879 return set_msr_mtrr(vcpu, msr, data);
880 case MSR_IA32_APICBASE:
881 kvm_set_apic_base(vcpu, data);
882 break;
883 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
884 return kvm_x2apic_msr_write(vcpu, msr, data);
885 case MSR_IA32_MISC_ENABLE:
886 vcpu->arch.ia32_misc_enable_msr = data;
887 break;
888 case MSR_KVM_WALL_CLOCK:
889 vcpu->kvm->arch.wall_clock = data;
890 kvm_write_wall_clock(vcpu->kvm, data);
891 break;
892 case MSR_KVM_SYSTEM_TIME: {
893 if (vcpu->arch.time_page) {
894 kvm_release_page_dirty(vcpu->arch.time_page);
895 vcpu->arch.time_page = NULL;
896 }
897
898 vcpu->arch.time = data;
899
900 /* we verify if the enable bit is set... */
901 if (!(data & 1))
902 break;
903
904 /* ...but clean it before doing the actual write */
905 vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
906
907 vcpu->arch.time_page =
908 gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
909
910 if (is_error_page(vcpu->arch.time_page)) {
911 kvm_release_page_clean(vcpu->arch.time_page);
912 vcpu->arch.time_page = NULL;
913 }
914
915 kvm_request_guest_time_update(vcpu);
916 break;
917 }
918 case MSR_IA32_MCG_CTL:
919 case MSR_IA32_MCG_STATUS:
920 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
921 return set_msr_mce(vcpu, msr, data);
922
923 /* Performance counters are not protected by a CPUID bit,
924 * so we should check all of them in the generic path for the sake of
925 * cross vendor migration.
926 * Writing a zero into the event select MSRs disables them,
927 * which we perfectly emulate ;-). Any other value should be at least
928 * reported, some guests depend on them.
929 */
930 case MSR_P6_EVNTSEL0:
931 case MSR_P6_EVNTSEL1:
932 case MSR_K7_EVNTSEL0:
933 case MSR_K7_EVNTSEL1:
934 case MSR_K7_EVNTSEL2:
935 case MSR_K7_EVNTSEL3:
936 if (data != 0)
937 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
938 "0x%x data 0x%llx\n", msr, data);
939 break;
940 /* at least RHEL 4 unconditionally writes to the perfctr registers,
941 * so we ignore writes to make it happy.
942 */
943 case MSR_P6_PERFCTR0:
944 case MSR_P6_PERFCTR1:
945 case MSR_K7_PERFCTR0:
946 case MSR_K7_PERFCTR1:
947 case MSR_K7_PERFCTR2:
948 case MSR_K7_PERFCTR3:
949 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: "
950 "0x%x data 0x%llx\n", msr, data);
951 break;
952 default:
953 if (!ignore_msrs) {
954 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
955 msr, data);
956 return 1;
957 } else {
958 pr_unimpl(vcpu, "ignored wrmsr: 0x%x data %llx\n",
959 msr, data);
960 break;
961 }
962 }
963 return 0;
964 }
965 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
966
967
968 /*
969 * Reads an msr value (of 'msr_index') into 'pdata'.
970 * Returns 0 on success, non-0 otherwise.
971 * Assumes vcpu_load() was already called.
972 */
973 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
974 {
975 return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
976 }
977
978 static int get_msr_mtrr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
979 {
980 u64 *p = (u64 *)&vcpu->arch.mtrr_state.fixed_ranges;
981
982 if (!msr_mtrr_valid(msr))
983 return 1;
984
985 if (msr == MSR_MTRRdefType)
986 *pdata = vcpu->arch.mtrr_state.def_type +
987 (vcpu->arch.mtrr_state.enabled << 10);
988 else if (msr == MSR_MTRRfix64K_00000)
989 *pdata = p[0];
990 else if (msr == MSR_MTRRfix16K_80000 || msr == MSR_MTRRfix16K_A0000)
991 *pdata = p[1 + msr - MSR_MTRRfix16K_80000];
992 else if (msr >= MSR_MTRRfix4K_C0000 && msr <= MSR_MTRRfix4K_F8000)
993 *pdata = p[3 + msr - MSR_MTRRfix4K_C0000];
994 else if (msr == MSR_IA32_CR_PAT)
995 *pdata = vcpu->arch.pat;
996 else { /* Variable MTRRs */
997 int idx, is_mtrr_mask;
998 u64 *pt;
999
1000 idx = (msr - 0x200) / 2;
1001 is_mtrr_mask = msr - 0x200 - 2 * idx;
1002 if (!is_mtrr_mask)
1003 pt =
1004 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].base_lo;
1005 else
1006 pt =
1007 (u64 *)&vcpu->arch.mtrr_state.var_ranges[idx].mask_lo;
1008 *pdata = *pt;
1009 }
1010
1011 return 0;
1012 }
1013
1014 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1015 {
1016 u64 data;
1017 u64 mcg_cap = vcpu->arch.mcg_cap;
1018 unsigned bank_num = mcg_cap & 0xff;
1019
1020 switch (msr) {
1021 case MSR_IA32_P5_MC_ADDR:
1022 case MSR_IA32_P5_MC_TYPE:
1023 data = 0;
1024 break;
1025 case MSR_IA32_MCG_CAP:
1026 data = vcpu->arch.mcg_cap;
1027 break;
1028 case MSR_IA32_MCG_CTL:
1029 if (!(mcg_cap & MCG_CTL_P))
1030 return 1;
1031 data = vcpu->arch.mcg_ctl;
1032 break;
1033 case MSR_IA32_MCG_STATUS:
1034 data = vcpu->arch.mcg_status;
1035 break;
1036 default:
1037 if (msr >= MSR_IA32_MC0_CTL &&
1038 msr < MSR_IA32_MC0_CTL + 4 * bank_num) {
1039 u32 offset = msr - MSR_IA32_MC0_CTL;
1040 data = vcpu->arch.mce_banks[offset];
1041 break;
1042 }
1043 return 1;
1044 }
1045 *pdata = data;
1046 return 0;
1047 }
1048
1049 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1050 {
1051 u64 data;
1052
1053 switch (msr) {
1054 case MSR_IA32_PLATFORM_ID:
1055 case MSR_IA32_UCODE_REV:
1056 case MSR_IA32_EBL_CR_POWERON:
1057 case MSR_IA32_DEBUGCTLMSR:
1058 case MSR_IA32_LASTBRANCHFROMIP:
1059 case MSR_IA32_LASTBRANCHTOIP:
1060 case MSR_IA32_LASTINTFROMIP:
1061 case MSR_IA32_LASTINTTOIP:
1062 case MSR_K8_SYSCFG:
1063 case MSR_K7_HWCR:
1064 case MSR_VM_HSAVE_PA:
1065 case MSR_P6_PERFCTR0:
1066 case MSR_P6_PERFCTR1:
1067 case MSR_P6_EVNTSEL0:
1068 case MSR_P6_EVNTSEL1:
1069 case MSR_K7_EVNTSEL0:
1070 case MSR_K7_PERFCTR0:
1071 case MSR_K8_INT_PENDING_MSG:
1072 case MSR_AMD64_NB_CFG:
1073 case MSR_FAM10H_MMIO_CONF_BASE:
1074 data = 0;
1075 break;
1076 case MSR_MTRRcap:
1077 data = 0x500 | KVM_NR_VAR_MTRR;
1078 break;
1079 case 0x200 ... 0x2ff:
1080 return get_msr_mtrr(vcpu, msr, pdata);
1081 case 0xcd: /* fsb frequency */
1082 data = 3;
1083 break;
1084 case MSR_IA32_APICBASE:
1085 data = kvm_get_apic_base(vcpu);
1086 break;
1087 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
1088 return kvm_x2apic_msr_read(vcpu, msr, pdata);
1089 break;
1090 case MSR_IA32_MISC_ENABLE:
1091 data = vcpu->arch.ia32_misc_enable_msr;
1092 break;
1093 case MSR_IA32_PERF_STATUS:
1094 /* TSC increment by tick */
1095 data = 1000ULL;
1096 /* CPU multiplier */
1097 data |= (((uint64_t)4ULL) << 40);
1098 break;
1099 case MSR_EFER:
1100 data = vcpu->arch.shadow_efer;
1101 break;
1102 case MSR_KVM_WALL_CLOCK:
1103 data = vcpu->kvm->arch.wall_clock;
1104 break;
1105 case MSR_KVM_SYSTEM_TIME:
1106 data = vcpu->arch.time;
1107 break;
1108 case MSR_IA32_P5_MC_ADDR:
1109 case MSR_IA32_P5_MC_TYPE:
1110 case MSR_IA32_MCG_CAP:
1111 case MSR_IA32_MCG_CTL:
1112 case MSR_IA32_MCG_STATUS:
1113 case MSR_IA32_MC0_CTL ... MSR_IA32_MC0_CTL + 4 * KVM_MAX_MCE_BANKS - 1:
1114 return get_msr_mce(vcpu, msr, pdata);
1115 default:
1116 if (!ignore_msrs) {
1117 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1118 return 1;
1119 } else {
1120 pr_unimpl(vcpu, "ignored rdmsr: 0x%x\n", msr);
1121 data = 0;
1122 }
1123 break;
1124 }
1125 *pdata = data;
1126 return 0;
1127 }
1128 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
1129
1130 /*
1131 * Read or write a bunch of msrs. All parameters are kernel addresses.
1132 *
1133 * @return number of msrs set successfully.
1134 */
1135 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
1136 struct kvm_msr_entry *entries,
1137 int (*do_msr)(struct kvm_vcpu *vcpu,
1138 unsigned index, u64 *data))
1139 {
1140 int i;
1141
1142 vcpu_load(vcpu);
1143
1144 down_read(&vcpu->kvm->slots_lock);
1145 for (i = 0; i < msrs->nmsrs; ++i)
1146 if (do_msr(vcpu, entries[i].index, &entries[i].data))
1147 break;
1148 up_read(&vcpu->kvm->slots_lock);
1149
1150 vcpu_put(vcpu);
1151
1152 return i;
1153 }
1154
1155 /*
1156 * Read or write a bunch of msrs. Parameters are user addresses.
1157 *
1158 * @return number of msrs set successfully.
1159 */
1160 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
1161 int (*do_msr)(struct kvm_vcpu *vcpu,
1162 unsigned index, u64 *data),
1163 int writeback)
1164 {
1165 struct kvm_msrs msrs;
1166 struct kvm_msr_entry *entries;
1167 int r, n;
1168 unsigned size;
1169
1170 r = -EFAULT;
1171 if (copy_from_user(&msrs, user_msrs, sizeof msrs))
1172 goto out;
1173
1174 r = -E2BIG;
1175 if (msrs.nmsrs >= MAX_IO_MSRS)
1176 goto out;
1177
1178 r = -ENOMEM;
1179 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
1180 entries = vmalloc(size);
1181 if (!entries)
1182 goto out;
1183
1184 r = -EFAULT;
1185 if (copy_from_user(entries, user_msrs->entries, size))
1186 goto out_free;
1187
1188 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
1189 if (r < 0)
1190 goto out_free;
1191
1192 r = -EFAULT;
1193 if (writeback && copy_to_user(user_msrs->entries, entries, size))
1194 goto out_free;
1195
1196 r = n;
1197
1198 out_free:
1199 vfree(entries);
1200 out:
1201 return r;
1202 }
1203
1204 int kvm_dev_ioctl_check_extension(long ext)
1205 {
1206 int r;
1207
1208 switch (ext) {
1209 case KVM_CAP_IRQCHIP:
1210 case KVM_CAP_HLT:
1211 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1212 case KVM_CAP_SET_TSS_ADDR:
1213 case KVM_CAP_EXT_CPUID:
1214 case KVM_CAP_CLOCKSOURCE:
1215 case KVM_CAP_PIT:
1216 case KVM_CAP_NOP_IO_DELAY:
1217 case KVM_CAP_MP_STATE:
1218 case KVM_CAP_SYNC_MMU:
1219 case KVM_CAP_REINJECT_CONTROL:
1220 case KVM_CAP_IRQ_INJECT_STATUS:
1221 case KVM_CAP_ASSIGN_DEV_IRQ:
1222 case KVM_CAP_IRQFD:
1223 case KVM_CAP_IOEVENTFD:
1224 case KVM_CAP_PIT2:
1225 case KVM_CAP_PIT_STATE2:
1226 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
1227 r = 1;
1228 break;
1229 case KVM_CAP_COALESCED_MMIO:
1230 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
1231 break;
1232 case KVM_CAP_VAPIC:
1233 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
1234 break;
1235 case KVM_CAP_NR_VCPUS:
1236 r = KVM_MAX_VCPUS;
1237 break;
1238 case KVM_CAP_NR_MEMSLOTS:
1239 r = KVM_MEMORY_SLOTS;
1240 break;
1241 case KVM_CAP_PV_MMU:
1242 r = !tdp_enabled;
1243 break;
1244 case KVM_CAP_IOMMU:
1245 r = iommu_found();
1246 break;
1247 case KVM_CAP_MCE:
1248 r = KVM_MAX_MCE_BANKS;
1249 break;
1250 default:
1251 r = 0;
1252 break;
1253 }
1254 return r;
1255
1256 }
1257
1258 long kvm_arch_dev_ioctl(struct file *filp,
1259 unsigned int ioctl, unsigned long arg)
1260 {
1261 void __user *argp = (void __user *)arg;
1262 long r;
1263
1264 switch (ioctl) {
1265 case KVM_GET_MSR_INDEX_LIST: {
1266 struct kvm_msr_list __user *user_msr_list = argp;
1267 struct kvm_msr_list msr_list;
1268 unsigned n;
1269
1270 r = -EFAULT;
1271 if (copy_from_user(&msr_list, user_msr_list, sizeof msr_list))
1272 goto out;
1273 n = msr_list.nmsrs;
1274 msr_list.nmsrs = num_msrs_to_save + ARRAY_SIZE(emulated_msrs);
1275 if (copy_to_user(user_msr_list, &msr_list, sizeof msr_list))
1276 goto out;
1277 r = -E2BIG;
1278 if (n < msr_list.nmsrs)
1279 goto out;
1280 r = -EFAULT;
1281 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
1282 num_msrs_to_save * sizeof(u32)))
1283 goto out;
1284 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
1285 &emulated_msrs,
1286 ARRAY_SIZE(emulated_msrs) * sizeof(u32)))
1287 goto out;
1288 r = 0;
1289 break;
1290 }
1291 case KVM_GET_SUPPORTED_CPUID: {
1292 struct kvm_cpuid2 __user *cpuid_arg = argp;
1293 struct kvm_cpuid2 cpuid;
1294
1295 r = -EFAULT;
1296 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1297 goto out;
1298 r = kvm_dev_ioctl_get_supported_cpuid(&cpuid,
1299 cpuid_arg->entries);
1300 if (r)
1301 goto out;
1302
1303 r = -EFAULT;
1304 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1305 goto out;
1306 r = 0;
1307 break;
1308 }
1309 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
1310 u64 mce_cap;
1311
1312 mce_cap = KVM_MCE_CAP_SUPPORTED;
1313 r = -EFAULT;
1314 if (copy_to_user(argp, &mce_cap, sizeof mce_cap))
1315 goto out;
1316 r = 0;
1317 break;
1318 }
1319 default:
1320 r = -EINVAL;
1321 }
1322 out:
1323 return r;
1324 }
1325
1326 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1327 {
1328 kvm_x86_ops->vcpu_load(vcpu, cpu);
1329 kvm_request_guest_time_update(vcpu);
1330 }
1331
1332 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
1333 {
1334 kvm_x86_ops->vcpu_put(vcpu);
1335 kvm_put_guest_fpu(vcpu);
1336 }
1337
1338 static int is_efer_nx(void)
1339 {
1340 unsigned long long efer = 0;
1341
1342 rdmsrl_safe(MSR_EFER, &efer);
1343 return efer & EFER_NX;
1344 }
1345
1346 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
1347 {
1348 int i;
1349 struct kvm_cpuid_entry2 *e, *entry;
1350
1351 entry = NULL;
1352 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
1353 e = &vcpu->arch.cpuid_entries[i];
1354 if (e->function == 0x80000001) {
1355 entry = e;
1356 break;
1357 }
1358 }
1359 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
1360 entry->edx &= ~(1 << 20);
1361 printk(KERN_INFO "kvm: guest NX capability removed\n");
1362 }
1363 }
1364
1365 /* when an old userspace process fills a new kernel module */
1366 static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
1367 struct kvm_cpuid *cpuid,
1368 struct kvm_cpuid_entry __user *entries)
1369 {
1370 int r, i;
1371 struct kvm_cpuid_entry *cpuid_entries;
1372
1373 r = -E2BIG;
1374 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1375 goto out;
1376 r = -ENOMEM;
1377 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
1378 if (!cpuid_entries)
1379 goto out;
1380 r = -EFAULT;
1381 if (copy_from_user(cpuid_entries, entries,
1382 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
1383 goto out_free;
1384 for (i = 0; i < cpuid->nent; i++) {
1385 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
1386 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
1387 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
1388 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
1389 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
1390 vcpu->arch.cpuid_entries[i].index = 0;
1391 vcpu->arch.cpuid_entries[i].flags = 0;
1392 vcpu->arch.cpuid_entries[i].padding[0] = 0;
1393 vcpu->arch.cpuid_entries[i].padding[1] = 0;
1394 vcpu->arch.cpuid_entries[i].padding[2] = 0;
1395 }
1396 vcpu->arch.cpuid_nent = cpuid->nent;
1397 cpuid_fix_nx_cap(vcpu);
1398 r = 0;
1399 kvm_apic_set_version(vcpu);
1400
1401 out_free:
1402 vfree(cpuid_entries);
1403 out:
1404 return r;
1405 }
1406
1407 static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
1408 struct kvm_cpuid2 *cpuid,
1409 struct kvm_cpuid_entry2 __user *entries)
1410 {
1411 int r;
1412
1413 r = -E2BIG;
1414 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1415 goto out;
1416 r = -EFAULT;
1417 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
1418 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
1419 goto out;
1420 vcpu->arch.cpuid_nent = cpuid->nent;
1421 kvm_apic_set_version(vcpu);
1422 return 0;
1423
1424 out:
1425 return r;
1426 }
1427
1428 static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
1429 struct kvm_cpuid2 *cpuid,
1430 struct kvm_cpuid_entry2 __user *entries)
1431 {
1432 int r;
1433
1434 r = -E2BIG;
1435 if (cpuid->nent < vcpu->arch.cpuid_nent)
1436 goto out;
1437 r = -EFAULT;
1438 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
1439 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
1440 goto out;
1441 return 0;
1442
1443 out:
1444 cpuid->nent = vcpu->arch.cpuid_nent;
1445 return r;
1446 }
1447
1448 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1449 u32 index)
1450 {
1451 entry->function = function;
1452 entry->index = index;
1453 cpuid_count(entry->function, entry->index,
1454 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
1455 entry->flags = 0;
1456 }
1457
1458 #define F(x) bit(X86_FEATURE_##x)
1459
1460 static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
1461 u32 index, int *nent, int maxnent)
1462 {
1463 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
1464 unsigned f_gbpages = kvm_x86_ops->gb_page_enable() ? F(GBPAGES) : 0;
1465 #ifdef CONFIG_X86_64
1466 unsigned f_lm = F(LM);
1467 #else
1468 unsigned f_lm = 0;
1469 #endif
1470
1471 /* cpuid 1.edx */
1472 const u32 kvm_supported_word0_x86_features =
1473 F(FPU) | F(VME) | F(DE) | F(PSE) |
1474 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1475 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
1476 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1477 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
1478 0 /* Reserved, DS, ACPI */ | F(MMX) |
1479 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
1480 0 /* HTT, TM, Reserved, PBE */;
1481 /* cpuid 0x80000001.edx */
1482 const u32 kvm_supported_word1_x86_features =
1483 F(FPU) | F(VME) | F(DE) | F(PSE) |
1484 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
1485 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
1486 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
1487 F(PAT) | F(PSE36) | 0 /* Reserved */ |
1488 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
1489 F(FXSR) | F(FXSR_OPT) | f_gbpages | 0 /* RDTSCP */ |
1490 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
1491 /* cpuid 1.ecx */
1492 const u32 kvm_supported_word4_x86_features =
1493 F(XMM3) | 0 /* Reserved, DTES64, MONITOR */ |
1494 0 /* DS-CPL, VMX, SMX, EST */ |
1495 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
1496 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
1497 0 /* Reserved, DCA */ | F(XMM4_1) |
1498 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
1499 0 /* Reserved, XSAVE, OSXSAVE */;
1500 /* cpuid 0x80000001.ecx */
1501 const u32 kvm_supported_word6_x86_features =
1502 F(LAHF_LM) | F(CMP_LEGACY) | F(SVM) | 0 /* ExtApicSpace */ |
1503 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
1504 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(SSE5) |
1505 0 /* SKINIT */ | 0 /* WDT */;
1506
1507 /* all calls to cpuid_count() should be made on the same cpu */
1508 get_cpu();
1509 do_cpuid_1_ent(entry, function, index);
1510 ++*nent;
1511
1512 switch (function) {
1513 case 0:
1514 entry->eax = min(entry->eax, (u32)0xb);
1515 break;
1516 case 1:
1517 entry->edx &= kvm_supported_word0_x86_features;
1518 entry->ecx &= kvm_supported_word4_x86_features;
1519 /* we support x2apic emulation even if host does not support
1520 * it since we emulate x2apic in software */
1521 entry->ecx |= F(X2APIC);
1522 break;
1523 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
1524 * may return different values. This forces us to get_cpu() before
1525 * issuing the first command, and also to emulate this annoying behavior
1526 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
1527 case 2: {
1528 int t, times = entry->eax & 0xff;
1529
1530 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1531 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
1532 for (t = 1; t < times && *nent < maxnent; ++t) {
1533 do_cpuid_1_ent(&entry[t], function, 0);
1534 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
1535 ++*nent;
1536 }
1537 break;
1538 }
1539 /* function 4 and 0xb have additional index. */
1540 case 4: {
1541 int i, cache_type;
1542
1543 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1544 /* read more entries until cache_type is zero */
1545 for (i = 1; *nent < maxnent; ++i) {
1546 cache_type = entry[i - 1].eax & 0x1f;
1547 if (!cache_type)
1548 break;
1549 do_cpuid_1_ent(&entry[i], function, i);
1550 entry[i].flags |=
1551 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1552 ++*nent;
1553 }
1554 break;
1555 }
1556 case 0xb: {
1557 int i, level_type;
1558
1559 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1560 /* read more entries until level_type is zero */
1561 for (i = 1; *nent < maxnent; ++i) {
1562 level_type = entry[i - 1].ecx & 0xff00;
1563 if (!level_type)
1564 break;
1565 do_cpuid_1_ent(&entry[i], function, i);
1566 entry[i].flags |=
1567 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
1568 ++*nent;
1569 }
1570 break;
1571 }
1572 case 0x80000000:
1573 entry->eax = min(entry->eax, 0x8000001a);
1574 break;
1575 case 0x80000001:
1576 entry->edx &= kvm_supported_word1_x86_features;
1577 entry->ecx &= kvm_supported_word6_x86_features;
1578 break;
1579 }
1580 put_cpu();
1581 }
1582
1583 #undef F
1584
1585 static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
1586 struct kvm_cpuid_entry2 __user *entries)
1587 {
1588 struct kvm_cpuid_entry2 *cpuid_entries;
1589 int limit, nent = 0, r = -E2BIG;
1590 u32 func;
1591
1592 if (cpuid->nent < 1)
1593 goto out;
1594 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1595 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1596 r = -ENOMEM;
1597 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
1598 if (!cpuid_entries)
1599 goto out;
1600
1601 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
1602 limit = cpuid_entries[0].eax;
1603 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
1604 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1605 &nent, cpuid->nent);
1606 r = -E2BIG;
1607 if (nent >= cpuid->nent)
1608 goto out_free;
1609
1610 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
1611 limit = cpuid_entries[nent - 1].eax;
1612 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
1613 do_cpuid_ent(&cpuid_entries[nent], func, 0,
1614 &nent, cpuid->nent);
1615 r = -E2BIG;
1616 if (nent >= cpuid->nent)
1617 goto out_free;
1618
1619 r = -EFAULT;
1620 if (copy_to_user(entries, cpuid_entries,
1621 nent * sizeof(struct kvm_cpuid_entry2)))
1622 goto out_free;
1623 cpuid->nent = nent;
1624 r = 0;
1625
1626 out_free:
1627 vfree(cpuid_entries);
1628 out:
1629 return r;
1630 }
1631
1632 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
1633 struct kvm_lapic_state *s)
1634 {
1635 vcpu_load(vcpu);
1636 memcpy(s->regs, vcpu->arch.apic->regs, sizeof *s);
1637 vcpu_put(vcpu);
1638
1639 return 0;
1640 }
1641
1642 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
1643 struct kvm_lapic_state *s)
1644 {
1645 vcpu_load(vcpu);
1646 memcpy(vcpu->arch.apic->regs, s->regs, sizeof *s);
1647 kvm_apic_post_state_restore(vcpu);
1648 update_cr8_intercept(vcpu);
1649 vcpu_put(vcpu);
1650
1651 return 0;
1652 }
1653
1654 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1655 struct kvm_interrupt *irq)
1656 {
1657 if (irq->irq < 0 || irq->irq >= 256)
1658 return -EINVAL;
1659 if (irqchip_in_kernel(vcpu->kvm))
1660 return -ENXIO;
1661 vcpu_load(vcpu);
1662
1663 kvm_queue_interrupt(vcpu, irq->irq, false);
1664
1665 vcpu_put(vcpu);
1666
1667 return 0;
1668 }
1669
1670 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
1671 {
1672 vcpu_load(vcpu);
1673 kvm_inject_nmi(vcpu);
1674 vcpu_put(vcpu);
1675
1676 return 0;
1677 }
1678
1679 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
1680 struct kvm_tpr_access_ctl *tac)
1681 {
1682 if (tac->flags)
1683 return -EINVAL;
1684 vcpu->arch.tpr_access_reporting = !!tac->enabled;
1685 return 0;
1686 }
1687
1688 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
1689 u64 mcg_cap)
1690 {
1691 int r;
1692 unsigned bank_num = mcg_cap & 0xff, bank;
1693
1694 r = -EINVAL;
1695 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
1696 goto out;
1697 if (mcg_cap & ~(KVM_MCE_CAP_SUPPORTED | 0xff | 0xff0000))
1698 goto out;
1699 r = 0;
1700 vcpu->arch.mcg_cap = mcg_cap;
1701 /* Init IA32_MCG_CTL to all 1s */
1702 if (mcg_cap & MCG_CTL_P)
1703 vcpu->arch.mcg_ctl = ~(u64)0;
1704 /* Init IA32_MCi_CTL to all 1s */
1705 for (bank = 0; bank < bank_num; bank++)
1706 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
1707 out:
1708 return r;
1709 }
1710
1711 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
1712 struct kvm_x86_mce *mce)
1713 {
1714 u64 mcg_cap = vcpu->arch.mcg_cap;
1715 unsigned bank_num = mcg_cap & 0xff;
1716 u64 *banks = vcpu->arch.mce_banks;
1717
1718 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
1719 return -EINVAL;
1720 /*
1721 * if IA32_MCG_CTL is not all 1s, the uncorrected error
1722 * reporting is disabled
1723 */
1724 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
1725 vcpu->arch.mcg_ctl != ~(u64)0)
1726 return 0;
1727 banks += 4 * mce->bank;
1728 /*
1729 * if IA32_MCi_CTL is not all 1s, the uncorrected error
1730 * reporting is disabled for the bank
1731 */
1732 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
1733 return 0;
1734 if (mce->status & MCI_STATUS_UC) {
1735 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
1736 !(vcpu->arch.cr4 & X86_CR4_MCE)) {
1737 printk(KERN_DEBUG "kvm: set_mce: "
1738 "injects mce exception while "
1739 "previous one is in progress!\n");
1740 set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
1741 return 0;
1742 }
1743 if (banks[1] & MCI_STATUS_VAL)
1744 mce->status |= MCI_STATUS_OVER;
1745 banks[2] = mce->addr;
1746 banks[3] = mce->misc;
1747 vcpu->arch.mcg_status = mce->mcg_status;
1748 banks[1] = mce->status;
1749 kvm_queue_exception(vcpu, MC_VECTOR);
1750 } else if (!(banks[1] & MCI_STATUS_VAL)
1751 || !(banks[1] & MCI_STATUS_UC)) {
1752 if (banks[1] & MCI_STATUS_VAL)
1753 mce->status |= MCI_STATUS_OVER;
1754 banks[2] = mce->addr;
1755 banks[3] = mce->misc;
1756 banks[1] = mce->status;
1757 } else
1758 banks[1] |= MCI_STATUS_OVER;
1759 return 0;
1760 }
1761
1762 long kvm_arch_vcpu_ioctl(struct file *filp,
1763 unsigned int ioctl, unsigned long arg)
1764 {
1765 struct kvm_vcpu *vcpu = filp->private_data;
1766 void __user *argp = (void __user *)arg;
1767 int r;
1768 struct kvm_lapic_state *lapic = NULL;
1769
1770 switch (ioctl) {
1771 case KVM_GET_LAPIC: {
1772 lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1773
1774 r = -ENOMEM;
1775 if (!lapic)
1776 goto out;
1777 r = kvm_vcpu_ioctl_get_lapic(vcpu, lapic);
1778 if (r)
1779 goto out;
1780 r = -EFAULT;
1781 if (copy_to_user(argp, lapic, sizeof(struct kvm_lapic_state)))
1782 goto out;
1783 r = 0;
1784 break;
1785 }
1786 case KVM_SET_LAPIC: {
1787 lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
1788 r = -ENOMEM;
1789 if (!lapic)
1790 goto out;
1791 r = -EFAULT;
1792 if (copy_from_user(lapic, argp, sizeof(struct kvm_lapic_state)))
1793 goto out;
1794 r = kvm_vcpu_ioctl_set_lapic(vcpu, lapic);
1795 if (r)
1796 goto out;
1797 r = 0;
1798 break;
1799 }
1800 case KVM_INTERRUPT: {
1801 struct kvm_interrupt irq;
1802
1803 r = -EFAULT;
1804 if (copy_from_user(&irq, argp, sizeof irq))
1805 goto out;
1806 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1807 if (r)
1808 goto out;
1809 r = 0;
1810 break;
1811 }
1812 case KVM_NMI: {
1813 r = kvm_vcpu_ioctl_nmi(vcpu);
1814 if (r)
1815 goto out;
1816 r = 0;
1817 break;
1818 }
1819 case KVM_SET_CPUID: {
1820 struct kvm_cpuid __user *cpuid_arg = argp;
1821 struct kvm_cpuid cpuid;
1822
1823 r = -EFAULT;
1824 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1825 goto out;
1826 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
1827 if (r)
1828 goto out;
1829 break;
1830 }
1831 case KVM_SET_CPUID2: {
1832 struct kvm_cpuid2 __user *cpuid_arg = argp;
1833 struct kvm_cpuid2 cpuid;
1834
1835 r = -EFAULT;
1836 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1837 goto out;
1838 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
1839 cpuid_arg->entries);
1840 if (r)
1841 goto out;
1842 break;
1843 }
1844 case KVM_GET_CPUID2: {
1845 struct kvm_cpuid2 __user *cpuid_arg = argp;
1846 struct kvm_cpuid2 cpuid;
1847
1848 r = -EFAULT;
1849 if (copy_from_user(&cpuid, cpuid_arg, sizeof cpuid))
1850 goto out;
1851 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
1852 cpuid_arg->entries);
1853 if (r)
1854 goto out;
1855 r = -EFAULT;
1856 if (copy_to_user(cpuid_arg, &cpuid, sizeof cpuid))
1857 goto out;
1858 r = 0;
1859 break;
1860 }
1861 case KVM_GET_MSRS:
1862 r = msr_io(vcpu, argp, kvm_get_msr, 1);
1863 break;
1864 case KVM_SET_MSRS:
1865 r = msr_io(vcpu, argp, do_set_msr, 0);
1866 break;
1867 case KVM_TPR_ACCESS_REPORTING: {
1868 struct kvm_tpr_access_ctl tac;
1869
1870 r = -EFAULT;
1871 if (copy_from_user(&tac, argp, sizeof tac))
1872 goto out;
1873 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
1874 if (r)
1875 goto out;
1876 r = -EFAULT;
1877 if (copy_to_user(argp, &tac, sizeof tac))
1878 goto out;
1879 r = 0;
1880 break;
1881 };
1882 case KVM_SET_VAPIC_ADDR: {
1883 struct kvm_vapic_addr va;
1884
1885 r = -EINVAL;
1886 if (!irqchip_in_kernel(vcpu->kvm))
1887 goto out;
1888 r = -EFAULT;
1889 if (copy_from_user(&va, argp, sizeof va))
1890 goto out;
1891 r = 0;
1892 kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
1893 break;
1894 }
1895 case KVM_X86_SETUP_MCE: {
1896 u64 mcg_cap;
1897
1898 r = -EFAULT;
1899 if (copy_from_user(&mcg_cap, argp, sizeof mcg_cap))
1900 goto out;
1901 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
1902 break;
1903 }
1904 case KVM_X86_SET_MCE: {
1905 struct kvm_x86_mce mce;
1906
1907 r = -EFAULT;
1908 if (copy_from_user(&mce, argp, sizeof mce))
1909 goto out;
1910 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
1911 break;
1912 }
1913 default:
1914 r = -EINVAL;
1915 }
1916 out:
1917 kfree(lapic);
1918 return r;
1919 }
1920
1921 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
1922 {
1923 int ret;
1924
1925 if (addr > (unsigned int)(-3 * PAGE_SIZE))
1926 return -1;
1927 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
1928 return ret;
1929 }
1930
1931 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
1932 u64 ident_addr)
1933 {
1934 kvm->arch.ept_identity_map_addr = ident_addr;
1935 return 0;
1936 }
1937
1938 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
1939 u32 kvm_nr_mmu_pages)
1940 {
1941 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
1942 return -EINVAL;
1943
1944 down_write(&kvm->slots_lock);
1945 spin_lock(&kvm->mmu_lock);
1946
1947 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
1948 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
1949
1950 spin_unlock(&kvm->mmu_lock);
1951 up_write(&kvm->slots_lock);
1952 return 0;
1953 }
1954
1955 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
1956 {
1957 return kvm->arch.n_alloc_mmu_pages;
1958 }
1959
1960 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
1961 {
1962 int i;
1963 struct kvm_mem_alias *alias;
1964
1965 for (i = 0; i < kvm->arch.naliases; ++i) {
1966 alias = &kvm->arch.aliases[i];
1967 if (gfn >= alias->base_gfn
1968 && gfn < alias->base_gfn + alias->npages)
1969 return alias->target_gfn + gfn - alias->base_gfn;
1970 }
1971 return gfn;
1972 }
1973
1974 /*
1975 * Set a new alias region. Aliases map a portion of physical memory into
1976 * another portion. This is useful for memory windows, for example the PC
1977 * VGA region.
1978 */
1979 static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
1980 struct kvm_memory_alias *alias)
1981 {
1982 int r, n;
1983 struct kvm_mem_alias *p;
1984
1985 r = -EINVAL;
1986 /* General sanity checks */
1987 if (alias->memory_size & (PAGE_SIZE - 1))
1988 goto out;
1989 if (alias->guest_phys_addr & (PAGE_SIZE - 1))
1990 goto out;
1991 if (alias->slot >= KVM_ALIAS_SLOTS)
1992 goto out;
1993 if (alias->guest_phys_addr + alias->memory_size
1994 < alias->guest_phys_addr)
1995 goto out;
1996 if (alias->target_phys_addr + alias->memory_size
1997 < alias->target_phys_addr)
1998 goto out;
1999
2000 down_write(&kvm->slots_lock);
2001 spin_lock(&kvm->mmu_lock);
2002
2003 p = &kvm->arch.aliases[alias->slot];
2004 p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
2005 p->npages = alias->memory_size >> PAGE_SHIFT;
2006 p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
2007
2008 for (n = KVM_ALIAS_SLOTS; n > 0; --n)
2009 if (kvm->arch.aliases[n - 1].npages)
2010 break;
2011 kvm->arch.naliases = n;
2012
2013 spin_unlock(&kvm->mmu_lock);
2014 kvm_mmu_zap_all(kvm);
2015
2016 up_write(&kvm->slots_lock);
2017
2018 return 0;
2019
2020 out:
2021 return r;
2022 }
2023
2024 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2025 {
2026 int r;
2027
2028 r = 0;
2029 switch (chip->chip_id) {
2030 case KVM_IRQCHIP_PIC_MASTER:
2031 memcpy(&chip->chip.pic,
2032 &pic_irqchip(kvm)->pics[0],
2033 sizeof(struct kvm_pic_state));
2034 break;
2035 case KVM_IRQCHIP_PIC_SLAVE:
2036 memcpy(&chip->chip.pic,
2037 &pic_irqchip(kvm)->pics[1],
2038 sizeof(struct kvm_pic_state));
2039 break;
2040 case KVM_IRQCHIP_IOAPIC:
2041 memcpy(&chip->chip.ioapic,
2042 ioapic_irqchip(kvm),
2043 sizeof(struct kvm_ioapic_state));
2044 break;
2045 default:
2046 r = -EINVAL;
2047 break;
2048 }
2049 return r;
2050 }
2051
2052 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
2053 {
2054 int r;
2055
2056 r = 0;
2057 switch (chip->chip_id) {
2058 case KVM_IRQCHIP_PIC_MASTER:
2059 spin_lock(&pic_irqchip(kvm)->lock);
2060 memcpy(&pic_irqchip(kvm)->pics[0],
2061 &chip->chip.pic,
2062 sizeof(struct kvm_pic_state));
2063 spin_unlock(&pic_irqchip(kvm)->lock);
2064 break;
2065 case KVM_IRQCHIP_PIC_SLAVE:
2066 spin_lock(&pic_irqchip(kvm)->lock);
2067 memcpy(&pic_irqchip(kvm)->pics[1],
2068 &chip->chip.pic,
2069 sizeof(struct kvm_pic_state));
2070 spin_unlock(&pic_irqchip(kvm)->lock);
2071 break;
2072 case KVM_IRQCHIP_IOAPIC:
2073 mutex_lock(&kvm->irq_lock);
2074 memcpy(ioapic_irqchip(kvm),
2075 &chip->chip.ioapic,
2076 sizeof(struct kvm_ioapic_state));
2077 mutex_unlock(&kvm->irq_lock);
2078 break;
2079 default:
2080 r = -EINVAL;
2081 break;
2082 }
2083 kvm_pic_update_irq(pic_irqchip(kvm));
2084 return r;
2085 }
2086
2087 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2088 {
2089 int r = 0;
2090
2091 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2092 memcpy(ps, &kvm->arch.vpit->pit_state, sizeof(struct kvm_pit_state));
2093 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2094 return r;
2095 }
2096
2097 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
2098 {
2099 int r = 0;
2100
2101 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2102 memcpy(&kvm->arch.vpit->pit_state, ps, sizeof(struct kvm_pit_state));
2103 kvm_pit_load_count(kvm, 0, ps->channels[0].count, 0);
2104 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2105 return r;
2106 }
2107
2108 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2109 {
2110 int r = 0;
2111
2112 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2113 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
2114 sizeof(ps->channels));
2115 ps->flags = kvm->arch.vpit->pit_state.flags;
2116 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2117 return r;
2118 }
2119
2120 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
2121 {
2122 int r = 0, start = 0;
2123 u32 prev_legacy, cur_legacy;
2124 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2125 prev_legacy = kvm->arch.vpit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
2126 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
2127 if (!prev_legacy && cur_legacy)
2128 start = 1;
2129 memcpy(&kvm->arch.vpit->pit_state.channels, &ps->channels,
2130 sizeof(kvm->arch.vpit->pit_state.channels));
2131 kvm->arch.vpit->pit_state.flags = ps->flags;
2132 kvm_pit_load_count(kvm, 0, kvm->arch.vpit->pit_state.channels[0].count, start);
2133 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2134 return r;
2135 }
2136
2137 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
2138 struct kvm_reinject_control *control)
2139 {
2140 if (!kvm->arch.vpit)
2141 return -ENXIO;
2142 mutex_lock(&kvm->arch.vpit->pit_state.lock);
2143 kvm->arch.vpit->pit_state.pit_timer.reinject = control->pit_reinject;
2144 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
2145 return 0;
2146 }
2147
2148 /*
2149 * Get (and clear) the dirty memory log for a memory slot.
2150 */
2151 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2152 struct kvm_dirty_log *log)
2153 {
2154 int r;
2155 int n;
2156 struct kvm_memory_slot *memslot;
2157 int is_dirty = 0;
2158
2159 down_write(&kvm->slots_lock);
2160
2161 r = kvm_get_dirty_log(kvm, log, &is_dirty);
2162 if (r)
2163 goto out;
2164
2165 /* If nothing is dirty, don't bother messing with page tables. */
2166 if (is_dirty) {
2167 spin_lock(&kvm->mmu_lock);
2168 kvm_mmu_slot_remove_write_access(kvm, log->slot);
2169 spin_unlock(&kvm->mmu_lock);
2170 memslot = &kvm->memslots[log->slot];
2171 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
2172 memset(memslot->dirty_bitmap, 0, n);
2173 }
2174 r = 0;
2175 out:
2176 up_write(&kvm->slots_lock);
2177 return r;
2178 }
2179
2180 long kvm_arch_vm_ioctl(struct file *filp,
2181 unsigned int ioctl, unsigned long arg)
2182 {
2183 struct kvm *kvm = filp->private_data;
2184 void __user *argp = (void __user *)arg;
2185 int r = -EINVAL;
2186 /*
2187 * This union makes it completely explicit to gcc-3.x
2188 * that these two variables' stack usage should be
2189 * combined, not added together.
2190 */
2191 union {
2192 struct kvm_pit_state ps;
2193 struct kvm_pit_state2 ps2;
2194 struct kvm_memory_alias alias;
2195 struct kvm_pit_config pit_config;
2196 } u;
2197
2198 switch (ioctl) {
2199 case KVM_SET_TSS_ADDR:
2200 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
2201 if (r < 0)
2202 goto out;
2203 break;
2204 case KVM_SET_IDENTITY_MAP_ADDR: {
2205 u64 ident_addr;
2206
2207 r = -EFAULT;
2208 if (copy_from_user(&ident_addr, argp, sizeof ident_addr))
2209 goto out;
2210 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
2211 if (r < 0)
2212 goto out;
2213 break;
2214 }
2215 case KVM_SET_MEMORY_REGION: {
2216 struct kvm_memory_region kvm_mem;
2217 struct kvm_userspace_memory_region kvm_userspace_mem;
2218
2219 r = -EFAULT;
2220 if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
2221 goto out;
2222 kvm_userspace_mem.slot = kvm_mem.slot;
2223 kvm_userspace_mem.flags = kvm_mem.flags;
2224 kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
2225 kvm_userspace_mem.memory_size = kvm_mem.memory_size;
2226 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
2227 if (r)
2228 goto out;
2229 break;
2230 }
2231 case KVM_SET_NR_MMU_PAGES:
2232 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
2233 if (r)
2234 goto out;
2235 break;
2236 case KVM_GET_NR_MMU_PAGES:
2237 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
2238 break;
2239 case KVM_SET_MEMORY_ALIAS:
2240 r = -EFAULT;
2241 if (copy_from_user(&u.alias, argp, sizeof(struct kvm_memory_alias)))
2242 goto out;
2243 r = kvm_vm_ioctl_set_memory_alias(kvm, &u.alias);
2244 if (r)
2245 goto out;
2246 break;
2247 case KVM_CREATE_IRQCHIP:
2248 r = -ENOMEM;
2249 kvm->arch.vpic = kvm_create_pic(kvm);
2250 if (kvm->arch.vpic) {
2251 r = kvm_ioapic_init(kvm);
2252 if (r) {
2253 kfree(kvm->arch.vpic);
2254 kvm->arch.vpic = NULL;
2255 goto out;
2256 }
2257 } else
2258 goto out;
2259 r = kvm_setup_default_irq_routing(kvm);
2260 if (r) {
2261 kfree(kvm->arch.vpic);
2262 kfree(kvm->arch.vioapic);
2263 goto out;
2264 }
2265 break;
2266 case KVM_CREATE_PIT:
2267 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
2268 goto create_pit;
2269 case KVM_CREATE_PIT2:
2270 r = -EFAULT;
2271 if (copy_from_user(&u.pit_config, argp,
2272 sizeof(struct kvm_pit_config)))
2273 goto out;
2274 create_pit:
2275 down_write(&kvm->slots_lock);
2276 r = -EEXIST;
2277 if (kvm->arch.vpit)
2278 goto create_pit_unlock;
2279 r = -ENOMEM;
2280 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
2281 if (kvm->arch.vpit)
2282 r = 0;
2283 create_pit_unlock:
2284 up_write(&kvm->slots_lock);
2285 break;
2286 case KVM_IRQ_LINE_STATUS:
2287 case KVM_IRQ_LINE: {
2288 struct kvm_irq_level irq_event;
2289
2290 r = -EFAULT;
2291 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2292 goto out;
2293 if (irqchip_in_kernel(kvm)) {
2294 __s32 status;
2295 mutex_lock(&kvm->irq_lock);
2296 status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
2297 irq_event.irq, irq_event.level);
2298 mutex_unlock(&kvm->irq_lock);
2299 if (ioctl == KVM_IRQ_LINE_STATUS) {
2300 irq_event.status = status;
2301 if (copy_to_user(argp, &irq_event,
2302 sizeof irq_event))
2303 goto out;
2304 }
2305 r = 0;
2306 }
2307 break;
2308 }
2309 case KVM_GET_IRQCHIP: {
2310 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2311 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2312
2313 r = -ENOMEM;
2314 if (!chip)
2315 goto out;
2316 r = -EFAULT;
2317 if (copy_from_user(chip, argp, sizeof *chip))
2318 goto get_irqchip_out;
2319 r = -ENXIO;
2320 if (!irqchip_in_kernel(kvm))
2321 goto get_irqchip_out;
2322 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
2323 if (r)
2324 goto get_irqchip_out;
2325 r = -EFAULT;
2326 if (copy_to_user(argp, chip, sizeof *chip))
2327 goto get_irqchip_out;
2328 r = 0;
2329 get_irqchip_out:
2330 kfree(chip);
2331 if (r)
2332 goto out;
2333 break;
2334 }
2335 case KVM_SET_IRQCHIP: {
2336 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
2337 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL);
2338
2339 r = -ENOMEM;
2340 if (!chip)
2341 goto out;
2342 r = -EFAULT;
2343 if (copy_from_user(chip, argp, sizeof *chip))
2344 goto set_irqchip_out;
2345 r = -ENXIO;
2346 if (!irqchip_in_kernel(kvm))
2347 goto set_irqchip_out;
2348 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
2349 if (r)
2350 goto set_irqchip_out;
2351 r = 0;
2352 set_irqchip_out:
2353 kfree(chip);
2354 if (r)
2355 goto out;
2356 break;
2357 }
2358 case KVM_GET_PIT: {
2359 r = -EFAULT;
2360 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
2361 goto out;
2362 r = -ENXIO;
2363 if (!kvm->arch.vpit)
2364 goto out;
2365 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
2366 if (r)
2367 goto out;
2368 r = -EFAULT;
2369 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
2370 goto out;
2371 r = 0;
2372 break;
2373 }
2374 case KVM_SET_PIT: {
2375 r = -EFAULT;
2376 if (copy_from_user(&u.ps, argp, sizeof u.ps))
2377 goto out;
2378 r = -ENXIO;
2379 if (!kvm->arch.vpit)
2380 goto out;
2381 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
2382 if (r)
2383 goto out;
2384 r = 0;
2385 break;
2386 }
2387 case KVM_GET_PIT2: {
2388 r = -ENXIO;
2389 if (!kvm->arch.vpit)
2390 goto out;
2391 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
2392 if (r)
2393 goto out;
2394 r = -EFAULT;
2395 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
2396 goto out;
2397 r = 0;
2398 break;
2399 }
2400 case KVM_SET_PIT2: {
2401 r = -EFAULT;
2402 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
2403 goto out;
2404 r = -ENXIO;
2405 if (!kvm->arch.vpit)
2406 goto out;
2407 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
2408 if (r)
2409 goto out;
2410 r = 0;
2411 break;
2412 }
2413 case KVM_REINJECT_CONTROL: {
2414 struct kvm_reinject_control control;
2415 r = -EFAULT;
2416 if (copy_from_user(&control, argp, sizeof(control)))
2417 goto out;
2418 r = kvm_vm_ioctl_reinject(kvm, &control);
2419 if (r)
2420 goto out;
2421 r = 0;
2422 break;
2423 }
2424 default:
2425 ;
2426 }
2427 out:
2428 return r;
2429 }
2430
2431 static void kvm_init_msr_list(void)
2432 {
2433 u32 dummy[2];
2434 unsigned i, j;
2435
2436 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
2437 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
2438 continue;
2439 if (j < i)
2440 msrs_to_save[j] = msrs_to_save[i];
2441 j++;
2442 }
2443 num_msrs_to_save = j;
2444 }
2445
2446 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
2447 const void *v)
2448 {
2449 if (vcpu->arch.apic &&
2450 !kvm_iodevice_write(&vcpu->arch.apic->dev, addr, len, v))
2451 return 0;
2452
2453 return kvm_io_bus_write(&vcpu->kvm->mmio_bus, addr, len, v);
2454 }
2455
2456 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
2457 {
2458 if (vcpu->arch.apic &&
2459 !kvm_iodevice_read(&vcpu->arch.apic->dev, addr, len, v))
2460 return 0;
2461
2462 return kvm_io_bus_read(&vcpu->kvm->mmio_bus, addr, len, v);
2463 }
2464
2465 static int kvm_read_guest_virt(gva_t addr, void *val, unsigned int bytes,
2466 struct kvm_vcpu *vcpu)
2467 {
2468 void *data = val;
2469 int r = X86EMUL_CONTINUE;
2470
2471 while (bytes) {
2472 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2473 unsigned offset = addr & (PAGE_SIZE-1);
2474 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
2475 int ret;
2476
2477 if (gpa == UNMAPPED_GVA) {
2478 r = X86EMUL_PROPAGATE_FAULT;
2479 goto out;
2480 }
2481 ret = kvm_read_guest(vcpu->kvm, gpa, data, toread);
2482 if (ret < 0) {
2483 r = X86EMUL_UNHANDLEABLE;
2484 goto out;
2485 }
2486
2487 bytes -= toread;
2488 data += toread;
2489 addr += toread;
2490 }
2491 out:
2492 return r;
2493 }
2494
2495 static int kvm_write_guest_virt(gva_t addr, void *val, unsigned int bytes,
2496 struct kvm_vcpu *vcpu)
2497 {
2498 void *data = val;
2499 int r = X86EMUL_CONTINUE;
2500
2501 while (bytes) {
2502 gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2503 unsigned offset = addr & (PAGE_SIZE-1);
2504 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
2505 int ret;
2506
2507 if (gpa == UNMAPPED_GVA) {
2508 r = X86EMUL_PROPAGATE_FAULT;
2509 goto out;
2510 }
2511 ret = kvm_write_guest(vcpu->kvm, gpa, data, towrite);
2512 if (ret < 0) {
2513 r = X86EMUL_UNHANDLEABLE;
2514 goto out;
2515 }
2516
2517 bytes -= towrite;
2518 data += towrite;
2519 addr += towrite;
2520 }
2521 out:
2522 return r;
2523 }
2524
2525
2526 static int emulator_read_emulated(unsigned long addr,
2527 void *val,
2528 unsigned int bytes,
2529 struct kvm_vcpu *vcpu)
2530 {
2531 gpa_t gpa;
2532
2533 if (vcpu->mmio_read_completed) {
2534 memcpy(val, vcpu->mmio_data, bytes);
2535 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
2536 vcpu->mmio_phys_addr, *(u64 *)val);
2537 vcpu->mmio_read_completed = 0;
2538 return X86EMUL_CONTINUE;
2539 }
2540
2541 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2542
2543 /* For APIC access vmexit */
2544 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2545 goto mmio;
2546
2547 if (kvm_read_guest_virt(addr, val, bytes, vcpu)
2548 == X86EMUL_CONTINUE)
2549 return X86EMUL_CONTINUE;
2550 if (gpa == UNMAPPED_GVA)
2551 return X86EMUL_PROPAGATE_FAULT;
2552
2553 mmio:
2554 /*
2555 * Is this MMIO handled locally?
2556 */
2557 if (!vcpu_mmio_read(vcpu, gpa, bytes, val)) {
2558 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes, gpa, *(u64 *)val);
2559 return X86EMUL_CONTINUE;
2560 }
2561
2562 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, 0);
2563
2564 vcpu->mmio_needed = 1;
2565 vcpu->mmio_phys_addr = gpa;
2566 vcpu->mmio_size = bytes;
2567 vcpu->mmio_is_write = 0;
2568
2569 return X86EMUL_UNHANDLEABLE;
2570 }
2571
2572 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
2573 const void *val, int bytes)
2574 {
2575 int ret;
2576
2577 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
2578 if (ret < 0)
2579 return 0;
2580 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1);
2581 return 1;
2582 }
2583
2584 static int emulator_write_emulated_onepage(unsigned long addr,
2585 const void *val,
2586 unsigned int bytes,
2587 struct kvm_vcpu *vcpu)
2588 {
2589 gpa_t gpa;
2590
2591 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2592
2593 if (gpa == UNMAPPED_GVA) {
2594 kvm_inject_page_fault(vcpu, addr, 2);
2595 return X86EMUL_PROPAGATE_FAULT;
2596 }
2597
2598 /* For APIC access vmexit */
2599 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2600 goto mmio;
2601
2602 if (emulator_write_phys(vcpu, gpa, val, bytes))
2603 return X86EMUL_CONTINUE;
2604
2605 mmio:
2606 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, *(u64 *)val);
2607 /*
2608 * Is this MMIO handled locally?
2609 */
2610 if (!vcpu_mmio_write(vcpu, gpa, bytes, val))
2611 return X86EMUL_CONTINUE;
2612
2613 vcpu->mmio_needed = 1;
2614 vcpu->mmio_phys_addr = gpa;
2615 vcpu->mmio_size = bytes;
2616 vcpu->mmio_is_write = 1;
2617 memcpy(vcpu->mmio_data, val, bytes);
2618
2619 return X86EMUL_CONTINUE;
2620 }
2621
2622 int emulator_write_emulated(unsigned long addr,
2623 const void *val,
2624 unsigned int bytes,
2625 struct kvm_vcpu *vcpu)
2626 {
2627 /* Crossing a page boundary? */
2628 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
2629 int rc, now;
2630
2631 now = -addr & ~PAGE_MASK;
2632 rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
2633 if (rc != X86EMUL_CONTINUE)
2634 return rc;
2635 addr += now;
2636 val += now;
2637 bytes -= now;
2638 }
2639 return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
2640 }
2641 EXPORT_SYMBOL_GPL(emulator_write_emulated);
2642
2643 static int emulator_cmpxchg_emulated(unsigned long addr,
2644 const void *old,
2645 const void *new,
2646 unsigned int bytes,
2647 struct kvm_vcpu *vcpu)
2648 {
2649 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
2650 #ifndef CONFIG_X86_64
2651 /* guests cmpxchg8b have to be emulated atomically */
2652 if (bytes == 8) {
2653 gpa_t gpa;
2654 struct page *page;
2655 char *kaddr;
2656 u64 val;
2657
2658 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
2659
2660 if (gpa == UNMAPPED_GVA ||
2661 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
2662 goto emul_write;
2663
2664 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
2665 goto emul_write;
2666
2667 val = *(u64 *)new;
2668
2669 page = gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
2670
2671 kaddr = kmap_atomic(page, KM_USER0);
2672 set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
2673 kunmap_atomic(kaddr, KM_USER0);
2674 kvm_release_page_dirty(page);
2675 }
2676 emul_write:
2677 #endif
2678
2679 return emulator_write_emulated(addr, new, bytes, vcpu);
2680 }
2681
2682 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
2683 {
2684 return kvm_x86_ops->get_segment_base(vcpu, seg);
2685 }
2686
2687 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
2688 {
2689 kvm_mmu_invlpg(vcpu, address);
2690 return X86EMUL_CONTINUE;
2691 }
2692
2693 int emulate_clts(struct kvm_vcpu *vcpu)
2694 {
2695 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 & ~X86_CR0_TS);
2696 return X86EMUL_CONTINUE;
2697 }
2698
2699 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
2700 {
2701 struct kvm_vcpu *vcpu = ctxt->vcpu;
2702
2703 switch (dr) {
2704 case 0 ... 3:
2705 *dest = kvm_x86_ops->get_dr(vcpu, dr);
2706 return X86EMUL_CONTINUE;
2707 default:
2708 pr_unimpl(vcpu, "%s: unexpected dr %u\n", __func__, dr);
2709 return X86EMUL_UNHANDLEABLE;
2710 }
2711 }
2712
2713 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
2714 {
2715 unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
2716 int exception;
2717
2718 kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
2719 if (exception) {
2720 /* FIXME: better handling */
2721 return X86EMUL_UNHANDLEABLE;
2722 }
2723 return X86EMUL_CONTINUE;
2724 }
2725
2726 void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
2727 {
2728 u8 opcodes[4];
2729 unsigned long rip = kvm_rip_read(vcpu);
2730 unsigned long rip_linear;
2731
2732 if (!printk_ratelimit())
2733 return;
2734
2735 rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
2736
2737 kvm_read_guest_virt(rip_linear, (void *)opcodes, 4, vcpu);
2738
2739 printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
2740 context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
2741 }
2742 EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
2743
2744 static struct x86_emulate_ops emulate_ops = {
2745 .read_std = kvm_read_guest_virt,
2746 .read_emulated = emulator_read_emulated,
2747 .write_emulated = emulator_write_emulated,
2748 .cmpxchg_emulated = emulator_cmpxchg_emulated,
2749 };
2750
2751 static void cache_all_regs(struct kvm_vcpu *vcpu)
2752 {
2753 kvm_register_read(vcpu, VCPU_REGS_RAX);
2754 kvm_register_read(vcpu, VCPU_REGS_RSP);
2755 kvm_register_read(vcpu, VCPU_REGS_RIP);
2756 vcpu->arch.regs_dirty = ~0;
2757 }
2758
2759 int emulate_instruction(struct kvm_vcpu *vcpu,
2760 struct kvm_run *run,
2761 unsigned long cr2,
2762 u16 error_code,
2763 int emulation_type)
2764 {
2765 int r, shadow_mask;
2766 struct decode_cache *c;
2767
2768 kvm_clear_exception_queue(vcpu);
2769 vcpu->arch.mmio_fault_cr2 = cr2;
2770 /*
2771 * TODO: fix emulate.c to use guest_read/write_register
2772 * instead of direct ->regs accesses, can save hundred cycles
2773 * on Intel for instructions that don't read/change RSP, for
2774 * for example.
2775 */
2776 cache_all_regs(vcpu);
2777
2778 vcpu->mmio_is_write = 0;
2779 vcpu->arch.pio.string = 0;
2780
2781 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
2782 int cs_db, cs_l;
2783 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
2784
2785 vcpu->arch.emulate_ctxt.vcpu = vcpu;
2786 vcpu->arch.emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
2787 vcpu->arch.emulate_ctxt.mode =
2788 (vcpu->arch.emulate_ctxt.eflags & X86_EFLAGS_VM)
2789 ? X86EMUL_MODE_REAL : cs_l
2790 ? X86EMUL_MODE_PROT64 : cs_db
2791 ? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
2792
2793 r = x86_decode_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2794
2795 /* Only allow emulation of specific instructions on #UD
2796 * (namely VMMCALL, sysenter, sysexit, syscall)*/
2797 c = &vcpu->arch.emulate_ctxt.decode;
2798 if (emulation_type & EMULTYPE_TRAP_UD) {
2799 if (!c->twobyte)
2800 return EMULATE_FAIL;
2801 switch (c->b) {
2802 case 0x01: /* VMMCALL */
2803 if (c->modrm_mod != 3 || c->modrm_rm != 1)
2804 return EMULATE_FAIL;
2805 break;
2806 case 0x34: /* sysenter */
2807 case 0x35: /* sysexit */
2808 if (c->modrm_mod != 0 || c->modrm_rm != 0)
2809 return EMULATE_FAIL;
2810 break;
2811 case 0x05: /* syscall */
2812 if (c->modrm_mod != 0 || c->modrm_rm != 0)
2813 return EMULATE_FAIL;
2814 break;
2815 default:
2816 return EMULATE_FAIL;
2817 }
2818
2819 if (!(c->modrm_reg == 0 || c->modrm_reg == 3))
2820 return EMULATE_FAIL;
2821 }
2822
2823 ++vcpu->stat.insn_emulation;
2824 if (r) {
2825 ++vcpu->stat.insn_emulation_fail;
2826 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2827 return EMULATE_DONE;
2828 return EMULATE_FAIL;
2829 }
2830 }
2831
2832 if (emulation_type & EMULTYPE_SKIP) {
2833 kvm_rip_write(vcpu, vcpu->arch.emulate_ctxt.decode.eip);
2834 return EMULATE_DONE;
2835 }
2836
2837 r = x86_emulate_insn(&vcpu->arch.emulate_ctxt, &emulate_ops);
2838 shadow_mask = vcpu->arch.emulate_ctxt.interruptibility;
2839
2840 if (r == 0)
2841 kvm_x86_ops->set_interrupt_shadow(vcpu, shadow_mask);
2842
2843 if (vcpu->arch.pio.string)
2844 return EMULATE_DO_MMIO;
2845
2846 if ((r || vcpu->mmio_is_write) && run) {
2847 run->exit_reason = KVM_EXIT_MMIO;
2848 run->mmio.phys_addr = vcpu->mmio_phys_addr;
2849 memcpy(run->mmio.data, vcpu->mmio_data, 8);
2850 run->mmio.len = vcpu->mmio_size;
2851 run->mmio.is_write = vcpu->mmio_is_write;
2852 }
2853
2854 if (r) {
2855 if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
2856 return EMULATE_DONE;
2857 if (!vcpu->mmio_needed) {
2858 kvm_report_emulation_failure(vcpu, "mmio");
2859 return EMULATE_FAIL;
2860 }
2861 return EMULATE_DO_MMIO;
2862 }
2863
2864 kvm_x86_ops->set_rflags(vcpu, vcpu->arch.emulate_ctxt.eflags);
2865
2866 if (vcpu->mmio_is_write) {
2867 vcpu->mmio_needed = 0;
2868 return EMULATE_DO_MMIO;
2869 }
2870
2871 return EMULATE_DONE;
2872 }
2873 EXPORT_SYMBOL_GPL(emulate_instruction);
2874
2875 static int pio_copy_data(struct kvm_vcpu *vcpu)
2876 {
2877 void *p = vcpu->arch.pio_data;
2878 gva_t q = vcpu->arch.pio.guest_gva;
2879 unsigned bytes;
2880 int ret;
2881
2882 bytes = vcpu->arch.pio.size * vcpu->arch.pio.cur_count;
2883 if (vcpu->arch.pio.in)
2884 ret = kvm_write_guest_virt(q, p, bytes, vcpu);
2885 else
2886 ret = kvm_read_guest_virt(q, p, bytes, vcpu);
2887 return ret;
2888 }
2889
2890 int complete_pio(struct kvm_vcpu *vcpu)
2891 {
2892 struct kvm_pio_request *io = &vcpu->arch.pio;
2893 long delta;
2894 int r;
2895 unsigned long val;
2896
2897 if (!io->string) {
2898 if (io->in) {
2899 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2900 memcpy(&val, vcpu->arch.pio_data, io->size);
2901 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
2902 }
2903 } else {
2904 if (io->in) {
2905 r = pio_copy_data(vcpu);
2906 if (r)
2907 return r;
2908 }
2909
2910 delta = 1;
2911 if (io->rep) {
2912 delta *= io->cur_count;
2913 /*
2914 * The size of the register should really depend on
2915 * current address size.
2916 */
2917 val = kvm_register_read(vcpu, VCPU_REGS_RCX);
2918 val -= delta;
2919 kvm_register_write(vcpu, VCPU_REGS_RCX, val);
2920 }
2921 if (io->down)
2922 delta = -delta;
2923 delta *= io->size;
2924 if (io->in) {
2925 val = kvm_register_read(vcpu, VCPU_REGS_RDI);
2926 val += delta;
2927 kvm_register_write(vcpu, VCPU_REGS_RDI, val);
2928 } else {
2929 val = kvm_register_read(vcpu, VCPU_REGS_RSI);
2930 val += delta;
2931 kvm_register_write(vcpu, VCPU_REGS_RSI, val);
2932 }
2933 }
2934
2935 io->count -= io->cur_count;
2936 io->cur_count = 0;
2937
2938 return 0;
2939 }
2940
2941 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
2942 {
2943 /* TODO: String I/O for in kernel device */
2944 int r;
2945
2946 if (vcpu->arch.pio.in)
2947 r = kvm_io_bus_read(&vcpu->kvm->pio_bus, vcpu->arch.pio.port,
2948 vcpu->arch.pio.size, pd);
2949 else
2950 r = kvm_io_bus_write(&vcpu->kvm->pio_bus, vcpu->arch.pio.port,
2951 vcpu->arch.pio.size, pd);
2952 return r;
2953 }
2954
2955 static int pio_string_write(struct kvm_vcpu *vcpu)
2956 {
2957 struct kvm_pio_request *io = &vcpu->arch.pio;
2958 void *pd = vcpu->arch.pio_data;
2959 int i, r = 0;
2960
2961 for (i = 0; i < io->cur_count; i++) {
2962 if (kvm_io_bus_write(&vcpu->kvm->pio_bus,
2963 io->port, io->size, pd)) {
2964 r = -EOPNOTSUPP;
2965 break;
2966 }
2967 pd += io->size;
2968 }
2969 return r;
2970 }
2971
2972 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
2973 int size, unsigned port)
2974 {
2975 unsigned long val;
2976
2977 vcpu->run->exit_reason = KVM_EXIT_IO;
2978 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
2979 vcpu->run->io.size = vcpu->arch.pio.size = size;
2980 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
2981 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = 1;
2982 vcpu->run->io.port = vcpu->arch.pio.port = port;
2983 vcpu->arch.pio.in = in;
2984 vcpu->arch.pio.string = 0;
2985 vcpu->arch.pio.down = 0;
2986 vcpu->arch.pio.rep = 0;
2987
2988 trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
2989 size, 1);
2990
2991 val = kvm_register_read(vcpu, VCPU_REGS_RAX);
2992 memcpy(vcpu->arch.pio_data, &val, 4);
2993
2994 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
2995 complete_pio(vcpu);
2996 return 1;
2997 }
2998 return 0;
2999 }
3000 EXPORT_SYMBOL_GPL(kvm_emulate_pio);
3001
3002 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
3003 int size, unsigned long count, int down,
3004 gva_t address, int rep, unsigned port)
3005 {
3006 unsigned now, in_page;
3007 int ret = 0;
3008
3009 vcpu->run->exit_reason = KVM_EXIT_IO;
3010 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
3011 vcpu->run->io.size = vcpu->arch.pio.size = size;
3012 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
3013 vcpu->run->io.count = vcpu->arch.pio.count = vcpu->arch.pio.cur_count = count;
3014 vcpu->run->io.port = vcpu->arch.pio.port = port;
3015 vcpu->arch.pio.in = in;
3016 vcpu->arch.pio.string = 1;
3017 vcpu->arch.pio.down = down;
3018 vcpu->arch.pio.rep = rep;
3019
3020 trace_kvm_pio(vcpu->run->io.direction == KVM_EXIT_IO_OUT, port,
3021 size, count);
3022
3023 if (!count) {
3024 kvm_x86_ops->skip_emulated_instruction(vcpu);
3025 return 1;
3026 }
3027
3028 if (!down)
3029 in_page = PAGE_SIZE - offset_in_page(address);
3030 else
3031 in_page = offset_in_page(address) + size;
3032 now = min(count, (unsigned long)in_page / size);
3033 if (!now)
3034 now = 1;
3035 if (down) {
3036 /*
3037 * String I/O in reverse. Yuck. Kill the guest, fix later.
3038 */
3039 pr_unimpl(vcpu, "guest string pio down\n");
3040 kvm_inject_gp(vcpu, 0);
3041 return 1;
3042 }
3043 vcpu->run->io.count = now;
3044 vcpu->arch.pio.cur_count = now;
3045
3046 if (vcpu->arch.pio.cur_count == vcpu->arch.pio.count)
3047 kvm_x86_ops->skip_emulated_instruction(vcpu);
3048
3049 vcpu->arch.pio.guest_gva = address;
3050
3051 if (!vcpu->arch.pio.in) {
3052 /* string PIO write */
3053 ret = pio_copy_data(vcpu);
3054 if (ret == X86EMUL_PROPAGATE_FAULT) {
3055 kvm_inject_gp(vcpu, 0);
3056 return 1;
3057 }
3058 if (ret == 0 && !pio_string_write(vcpu)) {
3059 complete_pio(vcpu);
3060 if (vcpu->arch.pio.count == 0)
3061 ret = 1;
3062 }
3063 }
3064 /* no string PIO read support yet */
3065
3066 return ret;
3067 }
3068 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
3069
3070 static void bounce_off(void *info)
3071 {
3072 /* nothing */
3073 }
3074
3075 static unsigned int ref_freq;
3076 static unsigned long tsc_khz_ref;
3077
3078 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
3079 void *data)
3080 {
3081 struct cpufreq_freqs *freq = data;
3082 struct kvm *kvm;
3083 struct kvm_vcpu *vcpu;
3084 int i, send_ipi = 0;
3085
3086 if (!ref_freq)
3087 ref_freq = freq->old;
3088
3089 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
3090 return 0;
3091 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
3092 return 0;
3093 per_cpu(cpu_tsc_khz, freq->cpu) = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
3094
3095 spin_lock(&kvm_lock);
3096 list_for_each_entry(kvm, &vm_list, vm_list) {
3097 kvm_for_each_vcpu(i, vcpu, kvm) {
3098 if (vcpu->cpu != freq->cpu)
3099 continue;
3100 if (!kvm_request_guest_time_update(vcpu))
3101 continue;
3102 if (vcpu->cpu != smp_processor_id())
3103 send_ipi++;
3104 }
3105 }
3106 spin_unlock(&kvm_lock);
3107
3108 if (freq->old < freq->new && send_ipi) {
3109 /*
3110 * We upscale the frequency. Must make the guest
3111 * doesn't see old kvmclock values while running with
3112 * the new frequency, otherwise we risk the guest sees
3113 * time go backwards.
3114 *
3115 * In case we update the frequency for another cpu
3116 * (which might be in guest context) send an interrupt
3117 * to kick the cpu out of guest context. Next time
3118 * guest context is entered kvmclock will be updated,
3119 * so the guest will not see stale values.
3120 */
3121 smp_call_function_single(freq->cpu, bounce_off, NULL, 1);
3122 }
3123 return 0;
3124 }
3125
3126 static struct notifier_block kvmclock_cpufreq_notifier_block = {
3127 .notifier_call = kvmclock_cpufreq_notifier
3128 };
3129
3130 int kvm_arch_init(void *opaque)
3131 {
3132 int r, cpu;
3133 struct kvm_x86_ops *ops = (struct kvm_x86_ops *)opaque;
3134
3135 if (kvm_x86_ops) {
3136 printk(KERN_ERR "kvm: already loaded the other module\n");
3137 r = -EEXIST;
3138 goto out;
3139 }
3140
3141 if (!ops->cpu_has_kvm_support()) {
3142 printk(KERN_ERR "kvm: no hardware support\n");
3143 r = -EOPNOTSUPP;
3144 goto out;
3145 }
3146 if (ops->disabled_by_bios()) {
3147 printk(KERN_ERR "kvm: disabled by bios\n");
3148 r = -EOPNOTSUPP;
3149 goto out;
3150 }
3151
3152 r = kvm_mmu_module_init();
3153 if (r)
3154 goto out;
3155
3156 kvm_init_msr_list();
3157
3158 kvm_x86_ops = ops;
3159 kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
3160 kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
3161 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
3162 PT_DIRTY_MASK, PT64_NX_MASK, 0);
3163
3164 for_each_possible_cpu(cpu)
3165 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
3166 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
3167 tsc_khz_ref = tsc_khz;
3168 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
3169 CPUFREQ_TRANSITION_NOTIFIER);
3170 }
3171
3172 return 0;
3173
3174 out:
3175 return r;
3176 }
3177
3178 void kvm_arch_exit(void)
3179 {
3180 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
3181 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
3182 CPUFREQ_TRANSITION_NOTIFIER);
3183 kvm_x86_ops = NULL;
3184 kvm_mmu_module_exit();
3185 }
3186
3187 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
3188 {
3189 ++vcpu->stat.halt_exits;
3190 if (irqchip_in_kernel(vcpu->kvm)) {
3191 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
3192 return 1;
3193 } else {
3194 vcpu->run->exit_reason = KVM_EXIT_HLT;
3195 return 0;
3196 }
3197 }
3198 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
3199
3200 static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
3201 unsigned long a1)
3202 {
3203 if (is_long_mode(vcpu))
3204 return a0;
3205 else
3206 return a0 | ((gpa_t)a1 << 32);
3207 }
3208
3209 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
3210 {
3211 unsigned long nr, a0, a1, a2, a3, ret;
3212 int r = 1;
3213
3214 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
3215 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
3216 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
3217 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
3218 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
3219
3220 trace_kvm_hypercall(nr, a0, a1, a2, a3);
3221
3222 if (!is_long_mode(vcpu)) {
3223 nr &= 0xFFFFFFFF;
3224 a0 &= 0xFFFFFFFF;
3225 a1 &= 0xFFFFFFFF;
3226 a2 &= 0xFFFFFFFF;
3227 a3 &= 0xFFFFFFFF;
3228 }
3229
3230 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
3231 ret = -KVM_EPERM;
3232 goto out;
3233 }
3234
3235 switch (nr) {
3236 case KVM_HC_VAPIC_POLL_IRQ:
3237 ret = 0;
3238 break;
3239 case KVM_HC_MMU_OP:
3240 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
3241 break;
3242 default:
3243 ret = -KVM_ENOSYS;
3244 break;
3245 }
3246 out:
3247 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
3248 ++vcpu->stat.hypercalls;
3249 return r;
3250 }
3251 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
3252
3253 int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
3254 {
3255 char instruction[3];
3256 int ret = 0;
3257 unsigned long rip = kvm_rip_read(vcpu);
3258
3259
3260 /*
3261 * Blow out the MMU to ensure that no other VCPU has an active mapping
3262 * to ensure that the updated hypercall appears atomically across all
3263 * VCPUs.
3264 */
3265 kvm_mmu_zap_all(vcpu->kvm);
3266
3267 kvm_x86_ops->patch_hypercall(vcpu, instruction);
3268 if (emulator_write_emulated(rip, instruction, 3, vcpu)
3269 != X86EMUL_CONTINUE)
3270 ret = -EFAULT;
3271
3272 return ret;
3273 }
3274
3275 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
3276 {
3277 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
3278 }
3279
3280 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3281 {
3282 struct descriptor_table dt = { limit, base };
3283
3284 kvm_x86_ops->set_gdt(vcpu, &dt);
3285 }
3286
3287 void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
3288 {
3289 struct descriptor_table dt = { limit, base };
3290
3291 kvm_x86_ops->set_idt(vcpu, &dt);
3292 }
3293
3294 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
3295 unsigned long *rflags)
3296 {
3297 kvm_lmsw(vcpu, msw);
3298 *rflags = kvm_x86_ops->get_rflags(vcpu);
3299 }
3300
3301 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
3302 {
3303 unsigned long value;
3304
3305 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3306 switch (cr) {
3307 case 0:
3308 value = vcpu->arch.cr0;
3309 break;
3310 case 2:
3311 value = vcpu->arch.cr2;
3312 break;
3313 case 3:
3314 value = vcpu->arch.cr3;
3315 break;
3316 case 4:
3317 value = vcpu->arch.cr4;
3318 break;
3319 case 8:
3320 value = kvm_get_cr8(vcpu);
3321 break;
3322 default:
3323 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3324 return 0;
3325 }
3326
3327 return value;
3328 }
3329
3330 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
3331 unsigned long *rflags)
3332 {
3333 switch (cr) {
3334 case 0:
3335 kvm_set_cr0(vcpu, mk_cr_64(vcpu->arch.cr0, val));
3336 *rflags = kvm_x86_ops->get_rflags(vcpu);
3337 break;
3338 case 2:
3339 vcpu->arch.cr2 = val;
3340 break;
3341 case 3:
3342 kvm_set_cr3(vcpu, val);
3343 break;
3344 case 4:
3345 kvm_set_cr4(vcpu, mk_cr_64(vcpu->arch.cr4, val));
3346 break;
3347 case 8:
3348 kvm_set_cr8(vcpu, val & 0xfUL);
3349 break;
3350 default:
3351 vcpu_printf(vcpu, "%s: unexpected cr %u\n", __func__, cr);
3352 }
3353 }
3354
3355 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
3356 {
3357 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
3358 int j, nent = vcpu->arch.cpuid_nent;
3359
3360 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
3361 /* when no next entry is found, the current entry[i] is reselected */
3362 for (j = i + 1; ; j = (j + 1) % nent) {
3363 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
3364 if (ej->function == e->function) {
3365 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
3366 return j;
3367 }
3368 }
3369 return 0; /* silence gcc, even though control never reaches here */
3370 }
3371
3372 /* find an entry with matching function, matching index (if needed), and that
3373 * should be read next (if it's stateful) */
3374 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
3375 u32 function, u32 index)
3376 {
3377 if (e->function != function)
3378 return 0;
3379 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
3380 return 0;
3381 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
3382 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
3383 return 0;
3384 return 1;
3385 }
3386
3387 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
3388 u32 function, u32 index)
3389 {
3390 int i;
3391 struct kvm_cpuid_entry2 *best = NULL;
3392
3393 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
3394 struct kvm_cpuid_entry2 *e;
3395
3396 e = &vcpu->arch.cpuid_entries[i];
3397 if (is_matching_cpuid_entry(e, function, index)) {
3398 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
3399 move_to_next_stateful_cpuid_entry(vcpu, i);
3400 best = e;
3401 break;
3402 }
3403 /*
3404 * Both basic or both extended?
3405 */
3406 if (((e->function ^ function) & 0x80000000) == 0)
3407 if (!best || e->function > best->function)
3408 best = e;
3409 }
3410 return best;
3411 }
3412
3413 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
3414 {
3415 struct kvm_cpuid_entry2 *best;
3416
3417 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
3418 if (best)
3419 return best->eax & 0xff;
3420 return 36;
3421 }
3422
3423 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
3424 {
3425 u32 function, index;
3426 struct kvm_cpuid_entry2 *best;
3427
3428 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
3429 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3430 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
3431 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
3432 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
3433 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
3434 best = kvm_find_cpuid_entry(vcpu, function, index);
3435 if (best) {
3436 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
3437 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
3438 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
3439 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
3440 }
3441 kvm_x86_ops->skip_emulated_instruction(vcpu);
3442 trace_kvm_cpuid(function,
3443 kvm_register_read(vcpu, VCPU_REGS_RAX),
3444 kvm_register_read(vcpu, VCPU_REGS_RBX),
3445 kvm_register_read(vcpu, VCPU_REGS_RCX),
3446 kvm_register_read(vcpu, VCPU_REGS_RDX));
3447 }
3448 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
3449
3450 /*
3451 * Check if userspace requested an interrupt window, and that the
3452 * interrupt window is open.
3453 *
3454 * No need to exit to userspace if we already have an interrupt queued.
3455 */
3456 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
3457 struct kvm_run *kvm_run)
3458 {
3459 return (!irqchip_in_kernel(vcpu->kvm) && !kvm_cpu_has_interrupt(vcpu) &&
3460 kvm_run->request_interrupt_window &&
3461 kvm_arch_interrupt_allowed(vcpu));
3462 }
3463
3464 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
3465 struct kvm_run *kvm_run)
3466 {
3467 kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
3468 kvm_run->cr8 = kvm_get_cr8(vcpu);
3469 kvm_run->apic_base = kvm_get_apic_base(vcpu);
3470 if (irqchip_in_kernel(vcpu->kvm))
3471 kvm_run->ready_for_interrupt_injection = 1;
3472 else
3473 kvm_run->ready_for_interrupt_injection =
3474 kvm_arch_interrupt_allowed(vcpu) &&
3475 !kvm_cpu_has_interrupt(vcpu) &&
3476 !kvm_event_needs_reinjection(vcpu);
3477 }
3478
3479 static void vapic_enter(struct kvm_vcpu *vcpu)
3480 {
3481 struct kvm_lapic *apic = vcpu->arch.apic;
3482 struct page *page;
3483
3484 if (!apic || !apic->vapic_addr)
3485 return;
3486
3487 page = gfn_to_page(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3488
3489 vcpu->arch.apic->vapic_page = page;
3490 }
3491
3492 static void vapic_exit(struct kvm_vcpu *vcpu)
3493 {
3494 struct kvm_lapic *apic = vcpu->arch.apic;
3495
3496 if (!apic || !apic->vapic_addr)
3497 return;
3498
3499 down_read(&vcpu->kvm->slots_lock);
3500 kvm_release_page_dirty(apic->vapic_page);
3501 mark_page_dirty(vcpu->kvm, apic->vapic_addr >> PAGE_SHIFT);
3502 up_read(&vcpu->kvm->slots_lock);
3503 }
3504
3505 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
3506 {
3507 int max_irr, tpr;
3508
3509 if (!kvm_x86_ops->update_cr8_intercept)
3510 return;
3511
3512 if (!vcpu->arch.apic)
3513 return;
3514
3515 if (!vcpu->arch.apic->vapic_addr)
3516 max_irr = kvm_lapic_find_highest_irr(vcpu);
3517 else
3518 max_irr = -1;
3519
3520 if (max_irr != -1)
3521 max_irr >>= 4;
3522
3523 tpr = kvm_lapic_get_cr8(vcpu);
3524
3525 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
3526 }
3527
3528 static void inject_pending_event(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3529 {
3530 /* try to reinject previous events if any */
3531 if (vcpu->arch.exception.pending) {
3532 kvm_x86_ops->queue_exception(vcpu, vcpu->arch.exception.nr,
3533 vcpu->arch.exception.has_error_code,
3534 vcpu->arch.exception.error_code);
3535 return;
3536 }
3537
3538 if (vcpu->arch.nmi_injected) {
3539 kvm_x86_ops->set_nmi(vcpu);
3540 return;
3541 }
3542
3543 if (vcpu->arch.interrupt.pending) {
3544 kvm_x86_ops->set_irq(vcpu);
3545 return;
3546 }
3547
3548 /* try to inject new event if pending */
3549 if (vcpu->arch.nmi_pending) {
3550 if (kvm_x86_ops->nmi_allowed(vcpu)) {
3551 vcpu->arch.nmi_pending = false;
3552 vcpu->arch.nmi_injected = true;
3553 kvm_x86_ops->set_nmi(vcpu);
3554 }
3555 } else if (kvm_cpu_has_interrupt(vcpu)) {
3556 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
3557 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
3558 false);
3559 kvm_x86_ops->set_irq(vcpu);
3560 }
3561 }
3562 }
3563
3564 static int vcpu_enter_guest(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3565 {
3566 int r;
3567 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
3568 kvm_run->request_interrupt_window;
3569
3570 if (vcpu->requests)
3571 if (test_and_clear_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
3572 kvm_mmu_unload(vcpu);
3573
3574 r = kvm_mmu_reload(vcpu);
3575 if (unlikely(r))
3576 goto out;
3577
3578 if (vcpu->requests) {
3579 if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
3580 __kvm_migrate_timers(vcpu);
3581 if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE, &vcpu->requests))
3582 kvm_write_guest_time(vcpu);
3583 if (test_and_clear_bit(KVM_REQ_MMU_SYNC, &vcpu->requests))
3584 kvm_mmu_sync_roots(vcpu);
3585 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
3586 kvm_x86_ops->tlb_flush(vcpu);
3587 if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
3588 &vcpu->requests)) {
3589 kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
3590 r = 0;
3591 goto out;
3592 }
3593 if (test_and_clear_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests)) {
3594 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
3595 r = 0;
3596 goto out;
3597 }
3598 }
3599
3600 preempt_disable();
3601
3602 kvm_x86_ops->prepare_guest_switch(vcpu);
3603 kvm_load_guest_fpu(vcpu);
3604
3605 local_irq_disable();
3606
3607 clear_bit(KVM_REQ_KICK, &vcpu->requests);
3608 smp_mb__after_clear_bit();
3609
3610 if (vcpu->requests || need_resched() || signal_pending(current)) {
3611 set_bit(KVM_REQ_KICK, &vcpu->requests);
3612 local_irq_enable();
3613 preempt_enable();
3614 r = 1;
3615 goto out;
3616 }
3617
3618 inject_pending_event(vcpu, kvm_run);
3619
3620 /* enable NMI/IRQ window open exits if needed */
3621 if (vcpu->arch.nmi_pending)
3622 kvm_x86_ops->enable_nmi_window(vcpu);
3623 else if (kvm_cpu_has_interrupt(vcpu) || req_int_win)
3624 kvm_x86_ops->enable_irq_window(vcpu);
3625
3626 if (kvm_lapic_enabled(vcpu)) {
3627 update_cr8_intercept(vcpu);
3628 kvm_lapic_sync_to_vapic(vcpu);
3629 }
3630
3631 up_read(&vcpu->kvm->slots_lock);
3632
3633 kvm_guest_enter();
3634
3635 if (unlikely(vcpu->arch.switch_db_regs)) {
3636 set_debugreg(0, 7);
3637 set_debugreg(vcpu->arch.eff_db[0], 0);
3638 set_debugreg(vcpu->arch.eff_db[1], 1);
3639 set_debugreg(vcpu->arch.eff_db[2], 2);
3640 set_debugreg(vcpu->arch.eff_db[3], 3);
3641 }
3642
3643 trace_kvm_entry(vcpu->vcpu_id);
3644 kvm_x86_ops->run(vcpu, kvm_run);
3645
3646 if (unlikely(vcpu->arch.switch_db_regs || test_thread_flag(TIF_DEBUG))) {
3647 set_debugreg(current->thread.debugreg0, 0);
3648 set_debugreg(current->thread.debugreg1, 1);
3649 set_debugreg(current->thread.debugreg2, 2);
3650 set_debugreg(current->thread.debugreg3, 3);
3651 set_debugreg(current->thread.debugreg6, 6);
3652 set_debugreg(current->thread.debugreg7, 7);
3653 }
3654
3655 set_bit(KVM_REQ_KICK, &vcpu->requests);
3656 local_irq_enable();
3657
3658 ++vcpu->stat.exits;
3659
3660 /*
3661 * We must have an instruction between local_irq_enable() and
3662 * kvm_guest_exit(), so the timer interrupt isn't delayed by
3663 * the interrupt shadow. The stat.exits increment will do nicely.
3664 * But we need to prevent reordering, hence this barrier():
3665 */
3666 barrier();
3667
3668 kvm_guest_exit();
3669
3670 preempt_enable();
3671
3672 down_read(&vcpu->kvm->slots_lock);
3673
3674 /*
3675 * Profile KVM exit RIPs:
3676 */
3677 if (unlikely(prof_on == KVM_PROFILING)) {
3678 unsigned long rip = kvm_rip_read(vcpu);
3679 profile_hit(KVM_PROFILING, (void *)rip);
3680 }
3681
3682
3683 kvm_lapic_sync_from_vapic(vcpu);
3684
3685 r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
3686 out:
3687 return r;
3688 }
3689
3690
3691 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3692 {
3693 int r;
3694
3695 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED)) {
3696 pr_debug("vcpu %d received sipi with vector # %x\n",
3697 vcpu->vcpu_id, vcpu->arch.sipi_vector);
3698 kvm_lapic_reset(vcpu);
3699 r = kvm_arch_vcpu_reset(vcpu);
3700 if (r)
3701 return r;
3702 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
3703 }
3704
3705 down_read(&vcpu->kvm->slots_lock);
3706 vapic_enter(vcpu);
3707
3708 r = 1;
3709 while (r > 0) {
3710 if (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE)
3711 r = vcpu_enter_guest(vcpu, kvm_run);
3712 else {
3713 up_read(&vcpu->kvm->slots_lock);
3714 kvm_vcpu_block(vcpu);
3715 down_read(&vcpu->kvm->slots_lock);
3716 if (test_and_clear_bit(KVM_REQ_UNHALT, &vcpu->requests))
3717 {
3718 switch(vcpu->arch.mp_state) {
3719 case KVM_MP_STATE_HALTED:
3720 vcpu->arch.mp_state =
3721 KVM_MP_STATE_RUNNABLE;
3722 case KVM_MP_STATE_RUNNABLE:
3723 break;
3724 case KVM_MP_STATE_SIPI_RECEIVED:
3725 default:
3726 r = -EINTR;
3727 break;
3728 }
3729 }
3730 }
3731
3732 if (r <= 0)
3733 break;
3734
3735 clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
3736 if (kvm_cpu_has_pending_timer(vcpu))
3737 kvm_inject_pending_timer_irqs(vcpu);
3738
3739 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
3740 r = -EINTR;
3741 kvm_run->exit_reason = KVM_EXIT_INTR;
3742 ++vcpu->stat.request_irq_exits;
3743 }
3744 if (signal_pending(current)) {
3745 r = -EINTR;
3746 kvm_run->exit_reason = KVM_EXIT_INTR;
3747 ++vcpu->stat.signal_exits;
3748 }
3749 if (need_resched()) {
3750 up_read(&vcpu->kvm->slots_lock);
3751 kvm_resched(vcpu);
3752 down_read(&vcpu->kvm->slots_lock);
3753 }
3754 }
3755
3756 up_read(&vcpu->kvm->slots_lock);
3757 post_kvm_run_save(vcpu, kvm_run);
3758
3759 vapic_exit(vcpu);
3760
3761 return r;
3762 }
3763
3764 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3765 {
3766 int r;
3767 sigset_t sigsaved;
3768
3769 vcpu_load(vcpu);
3770
3771 if (vcpu->sigset_active)
3772 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
3773
3774 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
3775 kvm_vcpu_block(vcpu);
3776 clear_bit(KVM_REQ_UNHALT, &vcpu->requests);
3777 r = -EAGAIN;
3778 goto out;
3779 }
3780
3781 /* re-sync apic's tpr */
3782 if (!irqchip_in_kernel(vcpu->kvm))
3783 kvm_set_cr8(vcpu, kvm_run->cr8);
3784
3785 if (vcpu->arch.pio.cur_count) {
3786 r = complete_pio(vcpu);
3787 if (r)
3788 goto out;
3789 }
3790 #if CONFIG_HAS_IOMEM
3791 if (vcpu->mmio_needed) {
3792 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
3793 vcpu->mmio_read_completed = 1;
3794 vcpu->mmio_needed = 0;
3795
3796 down_read(&vcpu->kvm->slots_lock);
3797 r = emulate_instruction(vcpu, kvm_run,
3798 vcpu->arch.mmio_fault_cr2, 0,
3799 EMULTYPE_NO_DECODE);
3800 up_read(&vcpu->kvm->slots_lock);
3801 if (r == EMULATE_DO_MMIO) {
3802 /*
3803 * Read-modify-write. Back to userspace.
3804 */
3805 r = 0;
3806 goto out;
3807 }
3808 }
3809 #endif
3810 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
3811 kvm_register_write(vcpu, VCPU_REGS_RAX,
3812 kvm_run->hypercall.ret);
3813
3814 r = __vcpu_run(vcpu, kvm_run);
3815
3816 out:
3817 if (vcpu->sigset_active)
3818 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
3819
3820 vcpu_put(vcpu);
3821 return r;
3822 }
3823
3824 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3825 {
3826 vcpu_load(vcpu);
3827
3828 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
3829 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
3830 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
3831 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
3832 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
3833 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
3834 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
3835 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
3836 #ifdef CONFIG_X86_64
3837 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
3838 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
3839 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
3840 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
3841 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
3842 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
3843 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
3844 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
3845 #endif
3846
3847 regs->rip = kvm_rip_read(vcpu);
3848 regs->rflags = kvm_x86_ops->get_rflags(vcpu);
3849
3850 /*
3851 * Don't leak debug flags in case they were set for guest debugging
3852 */
3853 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3854 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
3855
3856 vcpu_put(vcpu);
3857
3858 return 0;
3859 }
3860
3861 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
3862 {
3863 vcpu_load(vcpu);
3864
3865 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
3866 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
3867 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
3868 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
3869 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
3870 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
3871 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
3872 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
3873 #ifdef CONFIG_X86_64
3874 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
3875 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
3876 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
3877 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
3878 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
3879 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
3880 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
3881 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
3882
3883 #endif
3884
3885 kvm_rip_write(vcpu, regs->rip);
3886 kvm_x86_ops->set_rflags(vcpu, regs->rflags);
3887
3888
3889 vcpu->arch.exception.pending = false;
3890
3891 vcpu_put(vcpu);
3892
3893 return 0;
3894 }
3895
3896 void kvm_get_segment(struct kvm_vcpu *vcpu,
3897 struct kvm_segment *var, int seg)
3898 {
3899 kvm_x86_ops->get_segment(vcpu, var, seg);
3900 }
3901
3902 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3903 {
3904 struct kvm_segment cs;
3905
3906 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
3907 *db = cs.db;
3908 *l = cs.l;
3909 }
3910 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
3911
3912 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
3913 struct kvm_sregs *sregs)
3914 {
3915 struct descriptor_table dt;
3916
3917 vcpu_load(vcpu);
3918
3919 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
3920 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
3921 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
3922 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
3923 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
3924 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
3925
3926 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
3927 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
3928
3929 kvm_x86_ops->get_idt(vcpu, &dt);
3930 sregs->idt.limit = dt.limit;
3931 sregs->idt.base = dt.base;
3932 kvm_x86_ops->get_gdt(vcpu, &dt);
3933 sregs->gdt.limit = dt.limit;
3934 sregs->gdt.base = dt.base;
3935
3936 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
3937 sregs->cr0 = vcpu->arch.cr0;
3938 sregs->cr2 = vcpu->arch.cr2;
3939 sregs->cr3 = vcpu->arch.cr3;
3940 sregs->cr4 = vcpu->arch.cr4;
3941 sregs->cr8 = kvm_get_cr8(vcpu);
3942 sregs->efer = vcpu->arch.shadow_efer;
3943 sregs->apic_base = kvm_get_apic_base(vcpu);
3944
3945 memset(sregs->interrupt_bitmap, 0, sizeof sregs->interrupt_bitmap);
3946
3947 if (vcpu->arch.interrupt.pending && !vcpu->arch.interrupt.soft)
3948 set_bit(vcpu->arch.interrupt.nr,
3949 (unsigned long *)sregs->interrupt_bitmap);
3950
3951 vcpu_put(vcpu);
3952
3953 return 0;
3954 }
3955
3956 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
3957 struct kvm_mp_state *mp_state)
3958 {
3959 vcpu_load(vcpu);
3960 mp_state->mp_state = vcpu->arch.mp_state;
3961 vcpu_put(vcpu);
3962 return 0;
3963 }
3964
3965 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
3966 struct kvm_mp_state *mp_state)
3967 {
3968 vcpu_load(vcpu);
3969 vcpu->arch.mp_state = mp_state->mp_state;
3970 vcpu_put(vcpu);
3971 return 0;
3972 }
3973
3974 static void kvm_set_segment(struct kvm_vcpu *vcpu,
3975 struct kvm_segment *var, int seg)
3976 {
3977 kvm_x86_ops->set_segment(vcpu, var, seg);
3978 }
3979
3980 static void seg_desct_to_kvm_desct(struct desc_struct *seg_desc, u16 selector,
3981 struct kvm_segment *kvm_desct)
3982 {
3983 kvm_desct->base = get_desc_base(seg_desc);
3984 kvm_desct->limit = get_desc_limit(seg_desc);
3985 if (seg_desc->g) {
3986 kvm_desct->limit <<= 12;
3987 kvm_desct->limit |= 0xfff;
3988 }
3989 kvm_desct->selector = selector;
3990 kvm_desct->type = seg_desc->type;
3991 kvm_desct->present = seg_desc->p;
3992 kvm_desct->dpl = seg_desc->dpl;
3993 kvm_desct->db = seg_desc->d;
3994 kvm_desct->s = seg_desc->s;
3995 kvm_desct->l = seg_desc->l;
3996 kvm_desct->g = seg_desc->g;
3997 kvm_desct->avl = seg_desc->avl;
3998 if (!selector)
3999 kvm_desct->unusable = 1;
4000 else
4001 kvm_desct->unusable = 0;
4002 kvm_desct->padding = 0;
4003 }
4004
4005 static void get_segment_descriptor_dtable(struct kvm_vcpu *vcpu,
4006 u16 selector,
4007 struct descriptor_table *dtable)
4008 {
4009 if (selector & 1 << 2) {
4010 struct kvm_segment kvm_seg;
4011
4012 kvm_get_segment(vcpu, &kvm_seg, VCPU_SREG_LDTR);
4013
4014 if (kvm_seg.unusable)
4015 dtable->limit = 0;
4016 else
4017 dtable->limit = kvm_seg.limit;
4018 dtable->base = kvm_seg.base;
4019 }
4020 else
4021 kvm_x86_ops->get_gdt(vcpu, dtable);
4022 }
4023
4024 /* allowed just for 8 bytes segments */
4025 static int load_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4026 struct desc_struct *seg_desc)
4027 {
4028 struct descriptor_table dtable;
4029 u16 index = selector >> 3;
4030
4031 get_segment_descriptor_dtable(vcpu, selector, &dtable);
4032
4033 if (dtable.limit < index * 8 + 7) {
4034 kvm_queue_exception_e(vcpu, GP_VECTOR, selector & 0xfffc);
4035 return 1;
4036 }
4037 return kvm_read_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4038 }
4039
4040 /* allowed just for 8 bytes segments */
4041 static int save_guest_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4042 struct desc_struct *seg_desc)
4043 {
4044 struct descriptor_table dtable;
4045 u16 index = selector >> 3;
4046
4047 get_segment_descriptor_dtable(vcpu, selector, &dtable);
4048
4049 if (dtable.limit < index * 8 + 7)
4050 return 1;
4051 return kvm_write_guest_virt(dtable.base + index*8, seg_desc, sizeof(*seg_desc), vcpu);
4052 }
4053
4054 static gpa_t get_tss_base_addr(struct kvm_vcpu *vcpu,
4055 struct desc_struct *seg_desc)
4056 {
4057 u32 base_addr = get_desc_base(seg_desc);
4058
4059 return vcpu->arch.mmu.gva_to_gpa(vcpu, base_addr);
4060 }
4061
4062 static u16 get_segment_selector(struct kvm_vcpu *vcpu, int seg)
4063 {
4064 struct kvm_segment kvm_seg;
4065
4066 kvm_get_segment(vcpu, &kvm_seg, seg);
4067 return kvm_seg.selector;
4068 }
4069
4070 static int load_segment_descriptor_to_kvm_desct(struct kvm_vcpu *vcpu,
4071 u16 selector,
4072 struct kvm_segment *kvm_seg)
4073 {
4074 struct desc_struct seg_desc;
4075
4076 if (load_guest_segment_descriptor(vcpu, selector, &seg_desc))
4077 return 1;
4078 seg_desct_to_kvm_desct(&seg_desc, selector, kvm_seg);
4079 return 0;
4080 }
4081
4082 static int kvm_load_realmode_segment(struct kvm_vcpu *vcpu, u16 selector, int seg)
4083 {
4084 struct kvm_segment segvar = {
4085 .base = selector << 4,
4086 .limit = 0xffff,
4087 .selector = selector,
4088 .type = 3,
4089 .present = 1,
4090 .dpl = 3,
4091 .db = 0,
4092 .s = 1,
4093 .l = 0,
4094 .g = 0,
4095 .avl = 0,
4096 .unusable = 0,
4097 };
4098 kvm_x86_ops->set_segment(vcpu, &segvar, seg);
4099 return 0;
4100 }
4101
4102 static int is_vm86_segment(struct kvm_vcpu *vcpu, int seg)
4103 {
4104 return (seg != VCPU_SREG_LDTR) &&
4105 (seg != VCPU_SREG_TR) &&
4106 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_VM);
4107 }
4108
4109 int kvm_load_segment_descriptor(struct kvm_vcpu *vcpu, u16 selector,
4110 int type_bits, int seg)
4111 {
4112 struct kvm_segment kvm_seg;
4113
4114 if (is_vm86_segment(vcpu, seg) || !(vcpu->arch.cr0 & X86_CR0_PE))
4115 return kvm_load_realmode_segment(vcpu, selector, seg);
4116 if (load_segment_descriptor_to_kvm_desct(vcpu, selector, &kvm_seg))
4117 return 1;
4118 kvm_seg.type |= type_bits;
4119
4120 if (seg != VCPU_SREG_SS && seg != VCPU_SREG_CS &&
4121 seg != VCPU_SREG_LDTR)
4122 if (!kvm_seg.s)
4123 kvm_seg.unusable = 1;
4124
4125 kvm_set_segment(vcpu, &kvm_seg, seg);
4126 return 0;
4127 }
4128
4129 static void save_state_to_tss32(struct kvm_vcpu *vcpu,
4130 struct tss_segment_32 *tss)
4131 {
4132 tss->cr3 = vcpu->arch.cr3;
4133 tss->eip = kvm_rip_read(vcpu);
4134 tss->eflags = kvm_x86_ops->get_rflags(vcpu);
4135 tss->eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4136 tss->ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4137 tss->edx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4138 tss->ebx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4139 tss->esp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4140 tss->ebp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4141 tss->esi = kvm_register_read(vcpu, VCPU_REGS_RSI);
4142 tss->edi = kvm_register_read(vcpu, VCPU_REGS_RDI);
4143 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4144 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4145 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4146 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4147 tss->fs = get_segment_selector(vcpu, VCPU_SREG_FS);
4148 tss->gs = get_segment_selector(vcpu, VCPU_SREG_GS);
4149 tss->ldt_selector = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4150 }
4151
4152 static int load_state_from_tss32(struct kvm_vcpu *vcpu,
4153 struct tss_segment_32 *tss)
4154 {
4155 kvm_set_cr3(vcpu, tss->cr3);
4156
4157 kvm_rip_write(vcpu, tss->eip);
4158 kvm_x86_ops->set_rflags(vcpu, tss->eflags | 2);
4159
4160 kvm_register_write(vcpu, VCPU_REGS_RAX, tss->eax);
4161 kvm_register_write(vcpu, VCPU_REGS_RCX, tss->ecx);
4162 kvm_register_write(vcpu, VCPU_REGS_RDX, tss->edx);
4163 kvm_register_write(vcpu, VCPU_REGS_RBX, tss->ebx);
4164 kvm_register_write(vcpu, VCPU_REGS_RSP, tss->esp);
4165 kvm_register_write(vcpu, VCPU_REGS_RBP, tss->ebp);
4166 kvm_register_write(vcpu, VCPU_REGS_RSI, tss->esi);
4167 kvm_register_write(vcpu, VCPU_REGS_RDI, tss->edi);
4168
4169 if (kvm_load_segment_descriptor(vcpu, tss->ldt_selector, 0, VCPU_SREG_LDTR))
4170 return 1;
4171
4172 if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4173 return 1;
4174
4175 if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4176 return 1;
4177
4178 if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4179 return 1;
4180
4181 if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4182 return 1;
4183
4184 if (kvm_load_segment_descriptor(vcpu, tss->fs, 1, VCPU_SREG_FS))
4185 return 1;
4186
4187 if (kvm_load_segment_descriptor(vcpu, tss->gs, 1, VCPU_SREG_GS))
4188 return 1;
4189 return 0;
4190 }
4191
4192 static void save_state_to_tss16(struct kvm_vcpu *vcpu,
4193 struct tss_segment_16 *tss)
4194 {
4195 tss->ip = kvm_rip_read(vcpu);
4196 tss->flag = kvm_x86_ops->get_rflags(vcpu);
4197 tss->ax = kvm_register_read(vcpu, VCPU_REGS_RAX);
4198 tss->cx = kvm_register_read(vcpu, VCPU_REGS_RCX);
4199 tss->dx = kvm_register_read(vcpu, VCPU_REGS_RDX);
4200 tss->bx = kvm_register_read(vcpu, VCPU_REGS_RBX);
4201 tss->sp = kvm_register_read(vcpu, VCPU_REGS_RSP);
4202 tss->bp = kvm_register_read(vcpu, VCPU_REGS_RBP);
4203 tss->si = kvm_register_read(vcpu, VCPU_REGS_RSI);
4204 tss->di = kvm_register_read(vcpu, VCPU_REGS_RDI);
4205
4206 tss->es = get_segment_selector(vcpu, VCPU_SREG_ES);
4207 tss->cs = get_segment_selector(vcpu, VCPU_SREG_CS);
4208 tss->ss = get_segment_selector(vcpu, VCPU_SREG_SS);
4209 tss->ds = get_segment_selector(vcpu, VCPU_SREG_DS);
4210 tss->ldt = get_segment_selector(vcpu, VCPU_SREG_LDTR);
4211 tss->prev_task_link = get_segment_selector(vcpu, VCPU_SREG_TR);
4212 }
4213
4214 static int load_state_from_tss16(struct kvm_vcpu *vcpu,
4215 struct tss_segment_16 *tss)
4216 {
4217 kvm_rip_write(vcpu, tss->ip);
4218 kvm_x86_ops->set_rflags(vcpu, tss->flag | 2);
4219 kvm_register_write(vcpu, VCPU_REGS_RAX, tss->ax);
4220 kvm_register_write(vcpu, VCPU_REGS_RCX, tss->cx);
4221 kvm_register_write(vcpu, VCPU_REGS_RDX, tss->dx);
4222 kvm_register_write(vcpu, VCPU_REGS_RBX, tss->bx);
4223 kvm_register_write(vcpu, VCPU_REGS_RSP, tss->sp);
4224 kvm_register_write(vcpu, VCPU_REGS_RBP, tss->bp);
4225 kvm_register_write(vcpu, VCPU_REGS_RSI, tss->si);
4226 kvm_register_write(vcpu, VCPU_REGS_RDI, tss->di);
4227
4228 if (kvm_load_segment_descriptor(vcpu, tss->ldt, 0, VCPU_SREG_LDTR))
4229 return 1;
4230
4231 if (kvm_load_segment_descriptor(vcpu, tss->es, 1, VCPU_SREG_ES))
4232 return 1;
4233
4234 if (kvm_load_segment_descriptor(vcpu, tss->cs, 9, VCPU_SREG_CS))
4235 return 1;
4236
4237 if (kvm_load_segment_descriptor(vcpu, tss->ss, 1, VCPU_SREG_SS))
4238 return 1;
4239
4240 if (kvm_load_segment_descriptor(vcpu, tss->ds, 1, VCPU_SREG_DS))
4241 return 1;
4242 return 0;
4243 }
4244
4245 static int kvm_task_switch_16(struct kvm_vcpu *vcpu, u16 tss_selector,
4246 u16 old_tss_sel, u32 old_tss_base,
4247 struct desc_struct *nseg_desc)
4248 {
4249 struct tss_segment_16 tss_segment_16;
4250 int ret = 0;
4251
4252 if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4253 sizeof tss_segment_16))
4254 goto out;
4255
4256 save_state_to_tss16(vcpu, &tss_segment_16);
4257
4258 if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_16,
4259 sizeof tss_segment_16))
4260 goto out;
4261
4262 if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4263 &tss_segment_16, sizeof tss_segment_16))
4264 goto out;
4265
4266 if (old_tss_sel != 0xffff) {
4267 tss_segment_16.prev_task_link = old_tss_sel;
4268
4269 if (kvm_write_guest(vcpu->kvm,
4270 get_tss_base_addr(vcpu, nseg_desc),
4271 &tss_segment_16.prev_task_link,
4272 sizeof tss_segment_16.prev_task_link))
4273 goto out;
4274 }
4275
4276 if (load_state_from_tss16(vcpu, &tss_segment_16))
4277 goto out;
4278
4279 ret = 1;
4280 out:
4281 return ret;
4282 }
4283
4284 static int kvm_task_switch_32(struct kvm_vcpu *vcpu, u16 tss_selector,
4285 u16 old_tss_sel, u32 old_tss_base,
4286 struct desc_struct *nseg_desc)
4287 {
4288 struct tss_segment_32 tss_segment_32;
4289 int ret = 0;
4290
4291 if (kvm_read_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4292 sizeof tss_segment_32))
4293 goto out;
4294
4295 save_state_to_tss32(vcpu, &tss_segment_32);
4296
4297 if (kvm_write_guest(vcpu->kvm, old_tss_base, &tss_segment_32,
4298 sizeof tss_segment_32))
4299 goto out;
4300
4301 if (kvm_read_guest(vcpu->kvm, get_tss_base_addr(vcpu, nseg_desc),
4302 &tss_segment_32, sizeof tss_segment_32))
4303 goto out;
4304
4305 if (old_tss_sel != 0xffff) {
4306 tss_segment_32.prev_task_link = old_tss_sel;
4307
4308 if (kvm_write_guest(vcpu->kvm,
4309 get_tss_base_addr(vcpu, nseg_desc),
4310 &tss_segment_32.prev_task_link,
4311 sizeof tss_segment_32.prev_task_link))
4312 goto out;
4313 }
4314
4315 if (load_state_from_tss32(vcpu, &tss_segment_32))
4316 goto out;
4317
4318 ret = 1;
4319 out:
4320 return ret;
4321 }
4322
4323 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
4324 {
4325 struct kvm_segment tr_seg;
4326 struct desc_struct cseg_desc;
4327 struct desc_struct nseg_desc;
4328 int ret = 0;
4329 u32 old_tss_base = get_segment_base(vcpu, VCPU_SREG_TR);
4330 u16 old_tss_sel = get_segment_selector(vcpu, VCPU_SREG_TR);
4331
4332 old_tss_base = vcpu->arch.mmu.gva_to_gpa(vcpu, old_tss_base);
4333
4334 /* FIXME: Handle errors. Failure to read either TSS or their
4335 * descriptors should generate a pagefault.
4336 */
4337 if (load_guest_segment_descriptor(vcpu, tss_selector, &nseg_desc))
4338 goto out;
4339
4340 if (load_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc))
4341 goto out;
4342
4343 if (reason != TASK_SWITCH_IRET) {
4344 int cpl;
4345
4346 cpl = kvm_x86_ops->get_cpl(vcpu);
4347 if ((tss_selector & 3) > nseg_desc.dpl || cpl > nseg_desc.dpl) {
4348 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
4349 return 1;
4350 }
4351 }
4352
4353 if (!nseg_desc.p || get_desc_limit(&nseg_desc) < 0x67) {
4354 kvm_queue_exception_e(vcpu, TS_VECTOR, tss_selector & 0xfffc);
4355 return 1;
4356 }
4357
4358 if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
4359 cseg_desc.type &= ~(1 << 1); //clear the B flag
4360 save_guest_segment_descriptor(vcpu, old_tss_sel, &cseg_desc);
4361 }
4362
4363 if (reason == TASK_SWITCH_IRET) {
4364 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4365 kvm_x86_ops->set_rflags(vcpu, eflags & ~X86_EFLAGS_NT);
4366 }
4367
4368 /* set back link to prev task only if NT bit is set in eflags
4369 note that old_tss_sel is not used afetr this point */
4370 if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4371 old_tss_sel = 0xffff;
4372
4373 /* set back link to prev task only if NT bit is set in eflags
4374 note that old_tss_sel is not used afetr this point */
4375 if (reason != TASK_SWITCH_CALL && reason != TASK_SWITCH_GATE)
4376 old_tss_sel = 0xffff;
4377
4378 if (nseg_desc.type & 8)
4379 ret = kvm_task_switch_32(vcpu, tss_selector, old_tss_sel,
4380 old_tss_base, &nseg_desc);
4381 else
4382 ret = kvm_task_switch_16(vcpu, tss_selector, old_tss_sel,
4383 old_tss_base, &nseg_desc);
4384
4385 if (reason == TASK_SWITCH_CALL || reason == TASK_SWITCH_GATE) {
4386 u32 eflags = kvm_x86_ops->get_rflags(vcpu);
4387 kvm_x86_ops->set_rflags(vcpu, eflags | X86_EFLAGS_NT);
4388 }
4389
4390 if (reason != TASK_SWITCH_IRET) {
4391 nseg_desc.type |= (1 << 1);
4392 save_guest_segment_descriptor(vcpu, tss_selector,
4393 &nseg_desc);
4394 }
4395
4396 kvm_x86_ops->set_cr0(vcpu, vcpu->arch.cr0 | X86_CR0_TS);
4397 seg_desct_to_kvm_desct(&nseg_desc, tss_selector, &tr_seg);
4398 tr_seg.type = 11;
4399 kvm_set_segment(vcpu, &tr_seg, VCPU_SREG_TR);
4400 out:
4401 return ret;
4402 }
4403 EXPORT_SYMBOL_GPL(kvm_task_switch);
4404
4405 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
4406 struct kvm_sregs *sregs)
4407 {
4408 int mmu_reset_needed = 0;
4409 int pending_vec, max_bits;
4410 struct descriptor_table dt;
4411
4412 vcpu_load(vcpu);
4413
4414 dt.limit = sregs->idt.limit;
4415 dt.base = sregs->idt.base;
4416 kvm_x86_ops->set_idt(vcpu, &dt);
4417 dt.limit = sregs->gdt.limit;
4418 dt.base = sregs->gdt.base;
4419 kvm_x86_ops->set_gdt(vcpu, &dt);
4420
4421 vcpu->arch.cr2 = sregs->cr2;
4422 mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
4423 vcpu->arch.cr3 = sregs->cr3;
4424
4425 kvm_set_cr8(vcpu, sregs->cr8);
4426
4427 mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
4428 kvm_x86_ops->set_efer(vcpu, sregs->efer);
4429 kvm_set_apic_base(vcpu, sregs->apic_base);
4430
4431 kvm_x86_ops->decache_cr4_guest_bits(vcpu);
4432
4433 mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
4434 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
4435 vcpu->arch.cr0 = sregs->cr0;
4436
4437 mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
4438 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
4439 if (!is_long_mode(vcpu) && is_pae(vcpu))
4440 load_pdptrs(vcpu, vcpu->arch.cr3);
4441
4442 if (mmu_reset_needed)
4443 kvm_mmu_reset_context(vcpu);
4444
4445 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
4446 pending_vec = find_first_bit(
4447 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
4448 if (pending_vec < max_bits) {
4449 kvm_queue_interrupt(vcpu, pending_vec, false);
4450 pr_debug("Set back pending irq %d\n", pending_vec);
4451 if (irqchip_in_kernel(vcpu->kvm))
4452 kvm_pic_clear_isr_ack(vcpu->kvm);
4453 }
4454
4455 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
4456 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
4457 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
4458 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
4459 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
4460 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
4461
4462 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
4463 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
4464
4465 update_cr8_intercept(vcpu);
4466
4467 /* Older userspace won't unhalt the vcpu on reset. */
4468 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
4469 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
4470 !(vcpu->arch.cr0 & X86_CR0_PE))
4471 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4472
4473 vcpu_put(vcpu);
4474
4475 return 0;
4476 }
4477
4478 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
4479 struct kvm_guest_debug *dbg)
4480 {
4481 int i, r;
4482
4483 vcpu_load(vcpu);
4484
4485 if ((dbg->control & (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) ==
4486 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_HW_BP)) {
4487 for (i = 0; i < KVM_NR_DB_REGS; ++i)
4488 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
4489 vcpu->arch.switch_db_regs =
4490 (dbg->arch.debugreg[7] & DR7_BP_EN_MASK);
4491 } else {
4492 for (i = 0; i < KVM_NR_DB_REGS; i++)
4493 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
4494 vcpu->arch.switch_db_regs = (vcpu->arch.dr7 & DR7_BP_EN_MASK);
4495 }
4496
4497 r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
4498
4499 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
4500 kvm_queue_exception(vcpu, DB_VECTOR);
4501 else if (dbg->control & KVM_GUESTDBG_INJECT_BP)
4502 kvm_queue_exception(vcpu, BP_VECTOR);
4503
4504 vcpu_put(vcpu);
4505
4506 return r;
4507 }
4508
4509 /*
4510 * fxsave fpu state. Taken from x86_64/processor.h. To be killed when
4511 * we have asm/x86/processor.h
4512 */
4513 struct fxsave {
4514 u16 cwd;
4515 u16 swd;
4516 u16 twd;
4517 u16 fop;
4518 u64 rip;
4519 u64 rdp;
4520 u32 mxcsr;
4521 u32 mxcsr_mask;
4522 u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
4523 #ifdef CONFIG_X86_64
4524 u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
4525 #else
4526 u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
4527 #endif
4528 };
4529
4530 /*
4531 * Translate a guest virtual address to a guest physical address.
4532 */
4533 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
4534 struct kvm_translation *tr)
4535 {
4536 unsigned long vaddr = tr->linear_address;
4537 gpa_t gpa;
4538
4539 vcpu_load(vcpu);
4540 down_read(&vcpu->kvm->slots_lock);
4541 gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, vaddr);
4542 up_read(&vcpu->kvm->slots_lock);
4543 tr->physical_address = gpa;
4544 tr->valid = gpa != UNMAPPED_GVA;
4545 tr->writeable = 1;
4546 tr->usermode = 0;
4547 vcpu_put(vcpu);
4548
4549 return 0;
4550 }
4551
4552 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4553 {
4554 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4555
4556 vcpu_load(vcpu);
4557
4558 memcpy(fpu->fpr, fxsave->st_space, 128);
4559 fpu->fcw = fxsave->cwd;
4560 fpu->fsw = fxsave->swd;
4561 fpu->ftwx = fxsave->twd;
4562 fpu->last_opcode = fxsave->fop;
4563 fpu->last_ip = fxsave->rip;
4564 fpu->last_dp = fxsave->rdp;
4565 memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
4566
4567 vcpu_put(vcpu);
4568
4569 return 0;
4570 }
4571
4572 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
4573 {
4574 struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
4575
4576 vcpu_load(vcpu);
4577
4578 memcpy(fxsave->st_space, fpu->fpr, 128);
4579 fxsave->cwd = fpu->fcw;
4580 fxsave->swd = fpu->fsw;
4581 fxsave->twd = fpu->ftwx;
4582 fxsave->fop = fpu->last_opcode;
4583 fxsave->rip = fpu->last_ip;
4584 fxsave->rdp = fpu->last_dp;
4585 memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
4586
4587 vcpu_put(vcpu);
4588
4589 return 0;
4590 }
4591
4592 void fx_init(struct kvm_vcpu *vcpu)
4593 {
4594 unsigned after_mxcsr_mask;
4595
4596 /*
4597 * Touch the fpu the first time in non atomic context as if
4598 * this is the first fpu instruction the exception handler
4599 * will fire before the instruction returns and it'll have to
4600 * allocate ram with GFP_KERNEL.
4601 */
4602 if (!used_math())
4603 kvm_fx_save(&vcpu->arch.host_fx_image);
4604
4605 /* Initialize guest FPU by resetting ours and saving into guest's */
4606 preempt_disable();
4607 kvm_fx_save(&vcpu->arch.host_fx_image);
4608 kvm_fx_finit();
4609 kvm_fx_save(&vcpu->arch.guest_fx_image);
4610 kvm_fx_restore(&vcpu->arch.host_fx_image);
4611 preempt_enable();
4612
4613 vcpu->arch.cr0 |= X86_CR0_ET;
4614 after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
4615 vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
4616 memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
4617 0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
4618 }
4619 EXPORT_SYMBOL_GPL(fx_init);
4620
4621 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
4622 {
4623 if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
4624 return;
4625
4626 vcpu->guest_fpu_loaded = 1;
4627 kvm_fx_save(&vcpu->arch.host_fx_image);
4628 kvm_fx_restore(&vcpu->arch.guest_fx_image);
4629 }
4630 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
4631
4632 void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
4633 {
4634 if (!vcpu->guest_fpu_loaded)
4635 return;
4636
4637 vcpu->guest_fpu_loaded = 0;
4638 kvm_fx_save(&vcpu->arch.guest_fx_image);
4639 kvm_fx_restore(&vcpu->arch.host_fx_image);
4640 ++vcpu->stat.fpu_reload;
4641 }
4642 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
4643
4644 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
4645 {
4646 if (vcpu->arch.time_page) {
4647 kvm_release_page_dirty(vcpu->arch.time_page);
4648 vcpu->arch.time_page = NULL;
4649 }
4650
4651 kvm_x86_ops->vcpu_free(vcpu);
4652 }
4653
4654 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
4655 unsigned int id)
4656 {
4657 return kvm_x86_ops->vcpu_create(kvm, id);
4658 }
4659
4660 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
4661 {
4662 int r;
4663
4664 /* We do fxsave: this must be aligned. */
4665 BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
4666
4667 vcpu->arch.mtrr_state.have_fixed = 1;
4668 vcpu_load(vcpu);
4669 r = kvm_arch_vcpu_reset(vcpu);
4670 if (r == 0)
4671 r = kvm_mmu_setup(vcpu);
4672 vcpu_put(vcpu);
4673 if (r < 0)
4674 goto free_vcpu;
4675
4676 return 0;
4677 free_vcpu:
4678 kvm_x86_ops->vcpu_free(vcpu);
4679 return r;
4680 }
4681
4682 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
4683 {
4684 vcpu_load(vcpu);
4685 kvm_mmu_unload(vcpu);
4686 vcpu_put(vcpu);
4687
4688 kvm_x86_ops->vcpu_free(vcpu);
4689 }
4690
4691 int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
4692 {
4693 vcpu->arch.nmi_pending = false;
4694 vcpu->arch.nmi_injected = false;
4695
4696 vcpu->arch.switch_db_regs = 0;
4697 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
4698 vcpu->arch.dr6 = DR6_FIXED_1;
4699 vcpu->arch.dr7 = DR7_FIXED_1;
4700
4701 return kvm_x86_ops->vcpu_reset(vcpu);
4702 }
4703
4704 void kvm_arch_hardware_enable(void *garbage)
4705 {
4706 kvm_x86_ops->hardware_enable(garbage);
4707 }
4708
4709 void kvm_arch_hardware_disable(void *garbage)
4710 {
4711 kvm_x86_ops->hardware_disable(garbage);
4712 }
4713
4714 int kvm_arch_hardware_setup(void)
4715 {
4716 return kvm_x86_ops->hardware_setup();
4717 }
4718
4719 void kvm_arch_hardware_unsetup(void)
4720 {
4721 kvm_x86_ops->hardware_unsetup();
4722 }
4723
4724 void kvm_arch_check_processor_compat(void *rtn)
4725 {
4726 kvm_x86_ops->check_processor_compatibility(rtn);
4727 }
4728
4729 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
4730 {
4731 struct page *page;
4732 struct kvm *kvm;
4733 int r;
4734
4735 BUG_ON(vcpu->kvm == NULL);
4736 kvm = vcpu->kvm;
4737
4738 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
4739 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
4740 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
4741 else
4742 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
4743
4744 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
4745 if (!page) {
4746 r = -ENOMEM;
4747 goto fail;
4748 }
4749 vcpu->arch.pio_data = page_address(page);
4750
4751 r = kvm_mmu_create(vcpu);
4752 if (r < 0)
4753 goto fail_free_pio_data;
4754
4755 if (irqchip_in_kernel(kvm)) {
4756 r = kvm_create_lapic(vcpu);
4757 if (r < 0)
4758 goto fail_mmu_destroy;
4759 }
4760
4761 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
4762 GFP_KERNEL);
4763 if (!vcpu->arch.mce_banks) {
4764 r = -ENOMEM;
4765 goto fail_mmu_destroy;
4766 }
4767 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
4768
4769 return 0;
4770
4771 fail_mmu_destroy:
4772 kvm_mmu_destroy(vcpu);
4773 fail_free_pio_data:
4774 free_page((unsigned long)vcpu->arch.pio_data);
4775 fail:
4776 return r;
4777 }
4778
4779 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
4780 {
4781 kvm_free_lapic(vcpu);
4782 down_read(&vcpu->kvm->slots_lock);
4783 kvm_mmu_destroy(vcpu);
4784 up_read(&vcpu->kvm->slots_lock);
4785 free_page((unsigned long)vcpu->arch.pio_data);
4786 }
4787
4788 struct kvm *kvm_arch_create_vm(void)
4789 {
4790 struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
4791
4792 if (!kvm)
4793 return ERR_PTR(-ENOMEM);
4794
4795 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
4796 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
4797
4798 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
4799 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
4800
4801 rdtscll(kvm->arch.vm_init_tsc);
4802
4803 return kvm;
4804 }
4805
4806 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
4807 {
4808 vcpu_load(vcpu);
4809 kvm_mmu_unload(vcpu);
4810 vcpu_put(vcpu);
4811 }
4812
4813 static void kvm_free_vcpus(struct kvm *kvm)
4814 {
4815 unsigned int i;
4816 struct kvm_vcpu *vcpu;
4817
4818 /*
4819 * Unpin any mmu pages first.
4820 */
4821 kvm_for_each_vcpu(i, vcpu, kvm)
4822 kvm_unload_vcpu_mmu(vcpu);
4823 kvm_for_each_vcpu(i, vcpu, kvm)
4824 kvm_arch_vcpu_free(vcpu);
4825
4826 mutex_lock(&kvm->lock);
4827 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
4828 kvm->vcpus[i] = NULL;
4829
4830 atomic_set(&kvm->online_vcpus, 0);
4831 mutex_unlock(&kvm->lock);
4832 }
4833
4834 void kvm_arch_sync_events(struct kvm *kvm)
4835 {
4836 kvm_free_all_assigned_devices(kvm);
4837 }
4838
4839 void kvm_arch_destroy_vm(struct kvm *kvm)
4840 {
4841 kvm_iommu_unmap_guest(kvm);
4842 kvm_free_pit(kvm);
4843 kfree(kvm->arch.vpic);
4844 kfree(kvm->arch.vioapic);
4845 kvm_free_vcpus(kvm);
4846 kvm_free_physmem(kvm);
4847 if (kvm->arch.apic_access_page)
4848 put_page(kvm->arch.apic_access_page);
4849 if (kvm->arch.ept_identity_pagetable)
4850 put_page(kvm->arch.ept_identity_pagetable);
4851 kfree(kvm);
4852 }
4853
4854 int kvm_arch_set_memory_region(struct kvm *kvm,
4855 struct kvm_userspace_memory_region *mem,
4856 struct kvm_memory_slot old,
4857 int user_alloc)
4858 {
4859 int npages = mem->memory_size >> PAGE_SHIFT;
4860 struct kvm_memory_slot *memslot = &kvm->memslots[mem->slot];
4861
4862 /*To keep backward compatibility with older userspace,
4863 *x86 needs to hanlde !user_alloc case.
4864 */
4865 if (!user_alloc) {
4866 if (npages && !old.rmap) {
4867 unsigned long userspace_addr;
4868
4869 down_write(&current->mm->mmap_sem);
4870 userspace_addr = do_mmap(NULL, 0,
4871 npages * PAGE_SIZE,
4872 PROT_READ | PROT_WRITE,
4873 MAP_PRIVATE | MAP_ANONYMOUS,
4874 0);
4875 up_write(&current->mm->mmap_sem);
4876
4877 if (IS_ERR((void *)userspace_addr))
4878 return PTR_ERR((void *)userspace_addr);
4879
4880 /* set userspace_addr atomically for kvm_hva_to_rmapp */
4881 spin_lock(&kvm->mmu_lock);
4882 memslot->userspace_addr = userspace_addr;
4883 spin_unlock(&kvm->mmu_lock);
4884 } else {
4885 if (!old.user_alloc && old.rmap) {
4886 int ret;
4887
4888 down_write(&current->mm->mmap_sem);
4889 ret = do_munmap(current->mm, old.userspace_addr,
4890 old.npages * PAGE_SIZE);
4891 up_write(&current->mm->mmap_sem);
4892 if (ret < 0)
4893 printk(KERN_WARNING
4894 "kvm_vm_ioctl_set_memory_region: "
4895 "failed to munmap memory\n");
4896 }
4897 }
4898 }
4899
4900 spin_lock(&kvm->mmu_lock);
4901 if (!kvm->arch.n_requested_mmu_pages) {
4902 unsigned int nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
4903 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
4904 }
4905
4906 kvm_mmu_slot_remove_write_access(kvm, mem->slot);
4907 spin_unlock(&kvm->mmu_lock);
4908
4909 return 0;
4910 }
4911
4912 void kvm_arch_flush_shadow(struct kvm *kvm)
4913 {
4914 kvm_mmu_zap_all(kvm);
4915 kvm_reload_remote_mmus(kvm);
4916 }
4917
4918 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
4919 {
4920 return vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE
4921 || vcpu->arch.mp_state == KVM_MP_STATE_SIPI_RECEIVED
4922 || vcpu->arch.nmi_pending ||
4923 (kvm_arch_interrupt_allowed(vcpu) &&
4924 kvm_cpu_has_interrupt(vcpu));
4925 }
4926
4927 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
4928 {
4929 int me;
4930 int cpu = vcpu->cpu;
4931
4932 if (waitqueue_active(&vcpu->wq)) {
4933 wake_up_interruptible(&vcpu->wq);
4934 ++vcpu->stat.halt_wakeup;
4935 }
4936
4937 me = get_cpu();
4938 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
4939 if (!test_and_set_bit(KVM_REQ_KICK, &vcpu->requests))
4940 smp_send_reschedule(cpu);
4941 put_cpu();
4942 }
4943
4944 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
4945 {
4946 return kvm_x86_ops->interrupt_allowed(vcpu);
4947 }
4948
4949 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
4950 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
4951 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
4952 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
4953 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
Linux with added FPC-III support