« back to all changes in this revision
Viewing changes to kernel/x86.c
- browse files at revision 16
- compare with another revision
- download diff
- download tarball
- view history from revision 16
- Committer: Bazaar Package Importer
- Author(s): Soren Hansen
- Date: 2007-11-15 02:21:55 UTC
- mfrom: (1.1.10 upstream)
- Revision ID: james.westby@ubuntu.com-20071115022155-pxoxb8kfcrkn72mi
Tags: 1:52+dfsg-0ubuntu1
* New upstream release.
* 08_default_tdf.patch
- Make -tdf the default and add a -no-tdf option.
* 08_default_tdf.patch
- Make -tdf the default and add a -no-tdf option.
- files added:
- config.mak
- libkvm
- user/test/x86
- files removed:
- patches/kvm-48.patch
- files modified:
- Makefile
added
removed
kernel/x86.c
16
16
17
17
#include "kvm.h"
18
18
#include "x86.h"
19
#include "segment_descriptor.h"
19
20
#include "irq.h"
20
21
21
22
#include <linux/kvm.h>
22
23
#include <linux/fs.h>
23
24
#include <linux/vmalloc.h>
25
#include <linux/module.h>
24
26
25
27
#include <asm/uaccess.h>
26
28
27
29
#define MAX_IO_MSRS 256
30
#define CR0_RESERVED_BITS \
31
(~(unsigned long)(X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS \
32
| X86_CR0_ET | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM \
33
| X86_CR0_NW | X86_CR0_CD | X86_CR0_PG))
34
#define CR4_RESERVED_BITS \
35
(~(unsigned long)(X86_CR4_VME | X86_CR4_PVI | X86_CR4_TSD | X86_CR4_DE\
36
| X86_CR4_PSE | X86_CR4_PAE | X86_CR4_MCE \
37
| X86_CR4_PGE | X86_CR4_PCE | X86_CR4_OSFXSR \
38
| X86_CR4_OSXMMEXCPT | X86_CR4_VMXE))
39
40
#define CR8_RESERVED_BITS (~(unsigned long)X86_CR8_TPR)
41
#define EFER_RESERVED_BITS 0xfffffffffffff2fe
42
43
#define STAT_OFFSET(x) offsetof(struct kvm_vcpu, stat.x)
44
45
struct kvm_stats_debugfs_item debugfs_entries[] = {
46
{ "pf_fixed", STAT_OFFSET(pf_fixed) },
47
{ "pf_guest", STAT_OFFSET(pf_guest) },
48
{ "tlb_flush", STAT_OFFSET(tlb_flush) },
49
{ "invlpg", STAT_OFFSET(invlpg) },
50
{ "exits", STAT_OFFSET(exits) },
51
{ "io_exits", STAT_OFFSET(io_exits) },
52
{ "mmio_exits", STAT_OFFSET(mmio_exits) },
53
{ "signal_exits", STAT_OFFSET(signal_exits) },
54
{ "irq_window", STAT_OFFSET(irq_window_exits) },
55
{ "halt_exits", STAT_OFFSET(halt_exits) },
56
{ "halt_wakeup", STAT_OFFSET(halt_wakeup) },
57
{ "request_irq", STAT_OFFSET(request_irq_exits) },
58
{ "irq_exits", STAT_OFFSET(irq_exits) },
59
{ "light_exits", STAT_OFFSET(light_exits) },
60
{ "efer_reload", STAT_OFFSET(efer_reload) },
61
{ NULL }
62
};
63
64
65
unsigned long segment_base(u16 selector)
66
{
67
struct descriptor_table gdt;
68
struct segment_descriptor *d;
69
unsigned long table_base;
70
unsigned long v;
71
72
if (selector == 0)
73
return 0;
74
75
asm("sgdt %0" : "=m"(gdt));
76
table_base = gdt.base;
77
78
if (selector & 4) { /* from ldt */
79
u16 ldt_selector;
80
81
asm("sldt %0" : "=g"(ldt_selector));
82
table_base = segment_base(ldt_selector);
83
}
84
d = (struct segment_descriptor *)(table_base + (selector & ~7));
85
v = d->base_low | ((unsigned long)d->base_mid << 16) |
86
((unsigned long)d->base_high << 24);
87
#ifdef CONFIG_X86_64
88
if (d->system == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
89
v |= ((unsigned long) \
90
((struct segment_descriptor_64 *)d)->base_higher) << 32;
91
#endif
92
return v;
93
}
94
EXPORT_SYMBOL_GPL(segment_base);
95
96
u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
97
{
98
if (irqchip_in_kernel(vcpu->kvm))
99
return vcpu->apic_base;
100
else
101
return vcpu->apic_base;
102
}
103
EXPORT_SYMBOL_GPL(kvm_get_apic_base);
104
105
void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data)
106
{
107
/* TODO: reserve bits check */
108
if (irqchip_in_kernel(vcpu->kvm))
109
kvm_lapic_set_base(vcpu, data);
110
else
111
vcpu->apic_base = data;
112
}
113
EXPORT_SYMBOL_GPL(kvm_set_apic_base);
114
115
static void inject_gp(struct kvm_vcpu *vcpu)
116
{
117
kvm_x86_ops->inject_gp(vcpu, 0);
118
}
119
120
/*
121
* Load the pae pdptrs. Return true is they are all valid.
122
*/
123
int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
124
{
125
gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
126
unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
127
int i;
128
int ret;
129
u64 pdpte[ARRAY_SIZE(vcpu->pdptrs)];
130
131
mutex_lock(&vcpu->kvm->lock);
132
ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
133
offset * sizeof(u64), sizeof(pdpte));
134
if (ret < 0) {
135
ret = 0;
136
goto out;
137
}
138
for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
139
if ((pdpte[i] & 1) && (pdpte[i] & 0xfffffff0000001e6ull)) {
140
ret = 0;
141
goto out;
142
}
143
}
144
ret = 1;
145
146
memcpy(vcpu->pdptrs, pdpte, sizeof(vcpu->pdptrs));
147
out:
148
mutex_unlock(&vcpu->kvm->lock);
149
150
return ret;
151
}
152
153
void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
154
{
155
if (cr0 & CR0_RESERVED_BITS) {
156
printk(KERN_DEBUG "set_cr0: 0x%lx #GP, reserved bits 0x%lx\n",
157
cr0, vcpu->cr0);
158
inject_gp(vcpu);
159
return;
160
}
161
162
if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD)) {
163
printk(KERN_DEBUG "set_cr0: #GP, CD == 0 && NW == 1\n");
164
inject_gp(vcpu);
165
return;
166
}
167
168
if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE)) {
169
printk(KERN_DEBUG "set_cr0: #GP, set PG flag "
170
"and a clear PE flag\n");
171
inject_gp(vcpu);
172
return;
173
}
174
175
if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
176
#ifdef CONFIG_X86_64
177
if ((vcpu->shadow_efer & EFER_LME)) {
178
int cs_db, cs_l;
179
180
if (!is_pae(vcpu)) {
181
printk(KERN_DEBUG "set_cr0: #GP, start paging "
182
"in long mode while PAE is disabled\n");
183
inject_gp(vcpu);
184
return;
185
}
186
kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
187
if (cs_l) {
188
printk(KERN_DEBUG "set_cr0: #GP, start paging "
189
"in long mode while CS.L == 1\n");
190
inject_gp(vcpu);
191
return;
192
193
}
194
} else
195
#endif
196
if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->cr3)) {
197
printk(KERN_DEBUG "set_cr0: #GP, pdptrs "
198
"reserved bits\n");
199
inject_gp(vcpu);
200
return;
201
}
202
203
}
204
205
kvm_x86_ops->set_cr0(vcpu, cr0);
206
vcpu->cr0 = cr0;
207
208
mutex_lock(&vcpu->kvm->lock);
209
kvm_mmu_reset_context(vcpu);
210
mutex_unlock(&vcpu->kvm->lock);
211
return;
212
}
213
EXPORT_SYMBOL_GPL(set_cr0);
214
215
void lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
216
{
217
set_cr0(vcpu, (vcpu->cr0 & ~0x0ful) | (msw & 0x0f));
218
}
219
EXPORT_SYMBOL_GPL(lmsw);
220
221
void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
222
{
223
if (cr4 & CR4_RESERVED_BITS) {
224
printk(KERN_DEBUG "set_cr4: #GP, reserved bits\n");
225
inject_gp(vcpu);
226
return;
227
}
228
229
if (is_long_mode(vcpu)) {
230
if (!(cr4 & X86_CR4_PAE)) {
231
printk(KERN_DEBUG "set_cr4: #GP, clearing PAE while "
232
"in long mode\n");
233
inject_gp(vcpu);
234
return;
235
}
236
} else if (is_paging(vcpu) && !is_pae(vcpu) && (cr4 & X86_CR4_PAE)
237
&& !load_pdptrs(vcpu, vcpu->cr3)) {
238
printk(KERN_DEBUG "set_cr4: #GP, pdptrs reserved bits\n");
239
inject_gp(vcpu);
240
return;
241
}
242
243
if (cr4 & X86_CR4_VMXE) {
244
printk(KERN_DEBUG "set_cr4: #GP, setting VMXE\n");
245
inject_gp(vcpu);
246
return;
247
}
248
kvm_x86_ops->set_cr4(vcpu, cr4);
249
vcpu->cr4 = cr4;
250
mutex_lock(&vcpu->kvm->lock);
251
kvm_mmu_reset_context(vcpu);
252
mutex_unlock(&vcpu->kvm->lock);
253
}
254
EXPORT_SYMBOL_GPL(set_cr4);
255
256
void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
257
{
258
if (is_long_mode(vcpu)) {
259
if (cr3 & CR3_L_MODE_RESERVED_BITS) {
260
printk(KERN_DEBUG "set_cr3: #GP, reserved bits\n");
261
inject_gp(vcpu);
262
return;
263
}
264
} else {
265
if (is_pae(vcpu)) {
266
if (cr3 & CR3_PAE_RESERVED_BITS) {
267
printk(KERN_DEBUG
268
"set_cr3: #GP, reserved bits\n");
269
inject_gp(vcpu);
270
return;
271
}
272
if (is_paging(vcpu) && !load_pdptrs(vcpu, cr3)) {
273
printk(KERN_DEBUG "set_cr3: #GP, pdptrs "
274
"reserved bits\n");
275
inject_gp(vcpu);
276
return;
277
}
278
}
279
/*
280
* We don't check reserved bits in nonpae mode, because
281
* this isn't enforced, and VMware depends on this.
282
*/
283
}
284
285
mutex_lock(&vcpu->kvm->lock);
286
/*
287
* Does the new cr3 value map to physical memory? (Note, we
288
* catch an invalid cr3 even in real-mode, because it would
289
* cause trouble later on when we turn on paging anyway.)
290
*
291
* A real CPU would silently accept an invalid cr3 and would
292
* attempt to use it - with largely undefined (and often hard
293
* to debug) behavior on the guest side.
294
*/
295
if (unlikely(!gfn_to_memslot(vcpu->kvm, cr3 >> PAGE_SHIFT)))
296
inject_gp(vcpu);
297
else {
298
vcpu->cr3 = cr3;
299
vcpu->mmu.new_cr3(vcpu);
300
}
301
mutex_unlock(&vcpu->kvm->lock);
302
}
303
EXPORT_SYMBOL_GPL(set_cr3);
304
305
void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
306
{
307
if (cr8 & CR8_RESERVED_BITS) {
308
printk(KERN_DEBUG "set_cr8: #GP, reserved bits 0x%lx\n", cr8);
309
inject_gp(vcpu);
310
return;
311
}
312
if (irqchip_in_kernel(vcpu->kvm))
313
kvm_lapic_set_tpr(vcpu, cr8);
314
else
315
vcpu->cr8 = cr8;
316
}
317
EXPORT_SYMBOL_GPL(set_cr8);
318
319
unsigned long get_cr8(struct kvm_vcpu *vcpu)
320
{
321
if (irqchip_in_kernel(vcpu->kvm))
322
return kvm_lapic_get_cr8(vcpu);
323
else
324
return vcpu->cr8;
325
}
326
EXPORT_SYMBOL_GPL(get_cr8);
28
327
29
328
/*
30
329
* List of msr numbers which we expose to userspace through KVM_GET_MSRS
48
347
MSR_IA32_MISC_ENABLE,
49
348
};
50
349
350
#ifdef CONFIG_X86_64
351
352
static void set_efer(struct kvm_vcpu *vcpu, u64 efer)
353
{
354
if (efer & EFER_RESERVED_BITS) {
355
printk(KERN_DEBUG "set_efer: 0x%llx #GP, reserved bits\n",
356
efer);
357
inject_gp(vcpu);
358
return;
359
}
360
361
if (is_paging(vcpu)
362
&& (vcpu->shadow_efer & EFER_LME) != (efer & EFER_LME)) {
363
printk(KERN_DEBUG "set_efer: #GP, change LME while paging\n");
364
inject_gp(vcpu);
365
return;
366
}
367
368
kvm_x86_ops->set_efer(vcpu, efer);
369
370
efer &= ~EFER_LMA;
371
efer |= vcpu->shadow_efer & EFER_LMA;
372
373
vcpu->shadow_efer = efer;
374
}
375
376
#endif
377
378
/*
379
* Writes msr value into into the appropriate "register".
380
* Returns 0 on success, non-0 otherwise.
381
* Assumes vcpu_load() was already called.
382
*/
383
int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
384
{
385
return kvm_x86_ops->set_msr(vcpu, msr_index, data);
386
}
387
51
388
/*
52
389
* Adapt set_msr() to msr_io()'s calling convention
53
390
*/
56
393
return kvm_set_msr(vcpu, index, *data);
57
394
}
58
395
396
397
int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
398
{
399
switch (msr) {
400
#ifdef CONFIG_X86_64
401
case MSR_EFER:
402
set_efer(vcpu, data);
403
break;
404
#endif
405
case MSR_IA32_MC0_STATUS:
406
pr_unimpl(vcpu, "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n",
407
__FUNCTION__, data);
408
break;
409
case MSR_IA32_MCG_STATUS:
410
pr_unimpl(vcpu, "%s: MSR_IA32_MCG_STATUS 0x%llx, nop\n",
411
__FUNCTION__, data);
412
break;
413
case MSR_IA32_UCODE_REV:
414
case MSR_IA32_UCODE_WRITE:
415
case 0x200 ... 0x2ff: /* MTRRs */
416
break;
417
case MSR_IA32_APICBASE:
418
kvm_set_apic_base(vcpu, data);
419
break;
420
case MSR_IA32_MISC_ENABLE:
421
vcpu->ia32_misc_enable_msr = data;
422
break;
423
default:
424
pr_unimpl(vcpu, "unhandled wrmsr: 0x%x\n", msr);
425
return 1;
426
}
427
return 0;
428
}
429
EXPORT_SYMBOL_GPL(kvm_set_msr_common);
430
431
432
/*
433
* Reads an msr value (of 'msr_index') into 'pdata'.
434
* Returns 0 on success, non-0 otherwise.
435
* Assumes vcpu_load() was already called.
436
*/
437
int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
438
{
439
return kvm_x86_ops->get_msr(vcpu, msr_index, pdata);
440
}
441
442
int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
443
{
444
u64 data;
445
446
switch (msr) {
447
case 0xc0010010: /* SYSCFG */
448
case 0xc0010015: /* HWCR */
449
case MSR_IA32_PLATFORM_ID:
450
case MSR_IA32_P5_MC_ADDR:
451
case MSR_IA32_P5_MC_TYPE:
452
case MSR_IA32_MC0_CTL:
453
case MSR_IA32_MCG_STATUS:
454
case MSR_IA32_MCG_CAP:
455
case MSR_IA32_MC0_MISC:
456
case MSR_IA32_MC0_MISC+4:
457
case MSR_IA32_MC0_MISC+8:
458
case MSR_IA32_MC0_MISC+12:
459
case MSR_IA32_MC0_MISC+16:
460
case MSR_IA32_UCODE_REV:
461
case MSR_IA32_PERF_STATUS:
462
case MSR_IA32_EBL_CR_POWERON:
463
/* MTRR registers */
464
case 0xfe:
465
case 0x200 ... 0x2ff:
466
data = 0;
467
break;
468
case 0xcd: /* fsb frequency */
469
data = 3;
470
break;
471
case MSR_IA32_APICBASE:
472
data = kvm_get_apic_base(vcpu);
473
break;
474
case MSR_IA32_MISC_ENABLE:
475
data = vcpu->ia32_misc_enable_msr;
476
break;
477
#ifdef CONFIG_X86_64
478
case MSR_EFER:
479
data = vcpu->shadow_efer;
480
break;
481
#endif
482
default:
483
pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
484
return 1;
485
}
486
*pdata = data;
487
return 0;
488
}
489
EXPORT_SYMBOL_GPL(kvm_get_msr_common);
490
59
491
/*
60
492
* Read or write a bunch of msrs. All parameters are kernel addresses.
61
493
*
251
683
252
684
switch (ioctl) {
253
685
case KVM_GET_LAPIC: {
254
struct kvm_lapic_state lapic;
686
struct kvm_lapic_state l_apic;
255
687
256
memset(&lapic, 0, sizeof lapic);
257
r = kvm_vcpu_ioctl_get_lapic(vcpu, &lapic);
688
memset(&l_apic, 0, sizeof l_apic);
689
r = kvm_vcpu_ioctl_get_lapic(vcpu, &l_apic);
258
690
if (r)
259
691
goto out;
260
692
r = -EFAULT;
261
if (copy_to_user(argp, &lapic, sizeof lapic))
693
if (copy_to_user(argp, &l_apic, sizeof l_apic))
262
694
goto out;
263
695
r = 0;
264
696
break;
265
697
}
266
698
case KVM_SET_LAPIC: {
267
struct kvm_lapic_state lapic;
699
struct kvm_lapic_state l_apic;
268
700
269
701
r = -EFAULT;
270
if (copy_from_user(&lapic, argp, sizeof lapic))
702
if (copy_from_user(&l_apic, argp, sizeof l_apic))
271
703
goto out;
272
r = kvm_vcpu_ioctl_set_lapic(vcpu, &lapic);;
704
r = kvm_vcpu_ioctl_set_lapic(vcpu, &l_apic);;
273
705
if (r)
274
706
goto out;
275
707
r = 0;
300
732
return r;
301
733
}
302
734
735
static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
736
{
737
int ret;
738
739
if (addr > (unsigned int)(-3 * PAGE_SIZE))
740
return -1;
741
ret = kvm_x86_ops->set_tss_addr(kvm, addr);
742
return ret;
743
}
744
745
static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
746
u32 kvm_nr_mmu_pages)
747
{
748
if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
749
return -EINVAL;
750
751
mutex_lock(&kvm->lock);
752
753
kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
754
kvm->n_requested_mmu_pages = kvm_nr_mmu_pages;
755
756
mutex_unlock(&kvm->lock);
757
return 0;
758
}
759
760
static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
761
{
762
return kvm->n_alloc_mmu_pages;
763
}
764
765
/*
766
* Set a new alias region. Aliases map a portion of physical memory into
767
* another portion. This is useful for memory windows, for example the PC
768
* VGA region.
769
*/
770
static int kvm_vm_ioctl_set_memory_alias(struct kvm *kvm,
771
struct kvm_memory_alias *alias)
772
{
773
int r, n;
774
struct kvm_mem_alias *p;
775
776
r = -EINVAL;
777
/* General sanity checks */
778
if (alias->memory_size & (PAGE_SIZE - 1))
779
goto out;
780
if (alias->guest_phys_addr & (PAGE_SIZE - 1))
781
goto out;
782
if (alias->slot >= KVM_ALIAS_SLOTS)
783
goto out;
784
if (alias->guest_phys_addr + alias->memory_size
785
< alias->guest_phys_addr)
786
goto out;
787
if (alias->target_phys_addr + alias->memory_size
788
< alias->target_phys_addr)
789
goto out;
790
791
mutex_lock(&kvm->lock);
792
793
p = &kvm->aliases[alias->slot];
794
p->base_gfn = alias->guest_phys_addr >> PAGE_SHIFT;
795
p->npages = alias->memory_size >> PAGE_SHIFT;
796
p->target_gfn = alias->target_phys_addr >> PAGE_SHIFT;
797
798
for (n = KVM_ALIAS_SLOTS; n > 0; --n)
799
if (kvm->aliases[n - 1].npages)
800
break;
801
kvm->naliases = n;
802
803
kvm_mmu_zap_all(kvm);
804
805
mutex_unlock(&kvm->lock);
806
807
return 0;
808
809
out:
810
return r;
811
}
812
813
static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
814
{
815
int r;
816
817
r = 0;
818
switch (chip->chip_id) {
819
case KVM_IRQCHIP_PIC_MASTER:
820
memcpy(&chip->chip.pic,
821
&pic_irqchip(kvm)->pics[0],
822
sizeof(struct kvm_pic_state));
823
break;
824
case KVM_IRQCHIP_PIC_SLAVE:
825
memcpy(&chip->chip.pic,
826
&pic_irqchip(kvm)->pics[1],
827
sizeof(struct kvm_pic_state));
828
break;
829
case KVM_IRQCHIP_IOAPIC:
830
memcpy(&chip->chip.ioapic,
831
ioapic_irqchip(kvm),
832
sizeof(struct kvm_ioapic_state));
833
break;
834
default:
835
r = -EINVAL;
836
break;
837
}
838
return r;
839
}
840
841
static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
842
{
843
int r;
844
845
r = 0;
846
switch (chip->chip_id) {
847
case KVM_IRQCHIP_PIC_MASTER:
848
memcpy(&pic_irqchip(kvm)->pics[0],
849
&chip->chip.pic,
850
sizeof(struct kvm_pic_state));
851
break;
852
case KVM_IRQCHIP_PIC_SLAVE:
853
memcpy(&pic_irqchip(kvm)->pics[1],
854
&chip->chip.pic,
855
sizeof(struct kvm_pic_state));
856
break;
857
case KVM_IRQCHIP_IOAPIC:
858
memcpy(ioapic_irqchip(kvm),
859
&chip->chip.ioapic,
860
sizeof(struct kvm_ioapic_state));
861
break;
862
default:
863
r = -EINVAL;
864
break;
865
}
866
kvm_pic_update_irq(pic_irqchip(kvm));
867
return r;
868
}
869
870
long kvm_arch_vm_ioctl(struct file *filp,
871
unsigned int ioctl, unsigned long arg)
872
{
873
struct kvm *kvm = filp->private_data;
874
void __user *argp = (void __user *)arg;
875
int r = -EINVAL;
876
877
switch (ioctl) {
878
case KVM_SET_TSS_ADDR:
879
r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
880
if (r < 0)
881
goto out;
882
break;
883
case KVM_SET_MEMORY_REGION: {
884
struct kvm_memory_region kvm_mem;
885
struct kvm_userspace_memory_region kvm_userspace_mem;
886
887
r = -EFAULT;
888
if (copy_from_user(&kvm_mem, argp, sizeof kvm_mem))
889
goto out;
890
kvm_userspace_mem.slot = kvm_mem.slot;
891
kvm_userspace_mem.flags = kvm_mem.flags;
892
kvm_userspace_mem.guest_phys_addr = kvm_mem.guest_phys_addr;
893
kvm_userspace_mem.memory_size = kvm_mem.memory_size;
894
r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 0);
895
if (r)
896
goto out;
897
break;
898
}
899
case KVM_SET_NR_MMU_PAGES:
900
r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
901
if (r)
902
goto out;
903
break;
904
case KVM_GET_NR_MMU_PAGES:
905
r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
906
break;
907
case KVM_SET_MEMORY_ALIAS: {
908
struct kvm_memory_alias alias;
909
910
r = -EFAULT;
911
if (copy_from_user(&alias, argp, sizeof alias))
912
goto out;
913
r = kvm_vm_ioctl_set_memory_alias(kvm, &alias);
914
if (r)
915
goto out;
916
break;
917
}
918
case KVM_CREATE_IRQCHIP:
919
r = -ENOMEM;
920
kvm->vpic = kvm_create_pic(kvm);
921
if (kvm->vpic) {
922
r = kvm_ioapic_init(kvm);
923
if (r) {
924
kfree(kvm->vpic);
925
kvm->vpic = NULL;
926
goto out;
927
}
928
} else
929
goto out;
930
break;
931
case KVM_IRQ_LINE: {
932
struct kvm_irq_level irq_event;
933
934
r = -EFAULT;
935
if (copy_from_user(&irq_event, argp, sizeof irq_event))
936
goto out;
937
if (irqchip_in_kernel(kvm)) {
938
mutex_lock(&kvm->lock);
939
if (irq_event.irq < 16)
940
kvm_pic_set_irq(pic_irqchip(kvm),
941
irq_event.irq,
942
irq_event.level);
943
kvm_ioapic_set_irq(kvm->vioapic,
944
irq_event.irq,
945
irq_event.level);
946
mutex_unlock(&kvm->lock);
947
r = 0;
948
}
949
break;
950
}
951
case KVM_GET_IRQCHIP: {
952
/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
953
struct kvm_irqchip chip;
954
955
r = -EFAULT;
956
if (copy_from_user(&chip, argp, sizeof chip))
957
goto out;
958
r = -ENXIO;
959
if (!irqchip_in_kernel(kvm))
960
goto out;
961
r = kvm_vm_ioctl_get_irqchip(kvm, &chip);
962
if (r)
963
goto out;
964
r = -EFAULT;
965
if (copy_to_user(argp, &chip, sizeof chip))
966
goto out;
967
r = 0;
968
break;
969
}
970
case KVM_SET_IRQCHIP: {
971
/* 0: PIC master, 1: PIC slave, 2: IOAPIC */
972
struct kvm_irqchip chip;
973
974
r = -EFAULT;
975
if (copy_from_user(&chip, argp, sizeof chip))
976
goto out;
977
r = -ENXIO;
978
if (!irqchip_in_kernel(kvm))
979
goto out;
980
r = kvm_vm_ioctl_set_irqchip(kvm, &chip);
981
if (r)
982
goto out;
983
r = 0;
984
break;
985
}
986
default:
987
;
988
}
989
out:
990
return r;
991
}
992
303
993
static __init void kvm_init_msr_list(void)
304
994
{
305
995
u32 dummy[2];
315
1005
num_msrs_to_save = j;
316
1006
}
317
1007
1008
/*
1009
* Only apic need an MMIO device hook, so shortcut now..
1010
*/
1011
static struct kvm_io_device *vcpu_find_pervcpu_dev(struct kvm_vcpu *vcpu,
1012
gpa_t addr)
1013
{
1014
struct kvm_io_device *dev;
1015
1016
if (vcpu->apic) {
1017
dev = &vcpu->apic->dev;
1018
if (dev->in_range(dev, addr))
1019
return dev;
1020
}
1021
return NULL;
1022
}
1023
1024
1025
static struct kvm_io_device *vcpu_find_mmio_dev(struct kvm_vcpu *vcpu,
1026
gpa_t addr)
1027
{
1028
struct kvm_io_device *dev;
1029
1030
dev = vcpu_find_pervcpu_dev(vcpu, addr);
1031
if (dev == NULL)
1032
dev = kvm_io_bus_find_dev(&vcpu->kvm->mmio_bus, addr);
1033
return dev;
1034
}
1035
1036
int emulator_read_std(unsigned long addr,
1037
void *val,
1038
unsigned int bytes,
1039
struct kvm_vcpu *vcpu)
1040
{
1041
void *data = val;
1042
1043
while (bytes) {
1044
gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1045
unsigned offset = addr & (PAGE_SIZE-1);
1046
unsigned tocopy = min(bytes, (unsigned)PAGE_SIZE - offset);
1047
int ret;
1048
1049
if (gpa == UNMAPPED_GVA)
1050
return X86EMUL_PROPAGATE_FAULT;
1051
ret = kvm_read_guest(vcpu->kvm, gpa, data, tocopy);
1052
if (ret < 0)
1053
return X86EMUL_UNHANDLEABLE;
1054
1055
bytes -= tocopy;
1056
data += tocopy;
1057
addr += tocopy;
1058
}
1059
1060
return X86EMUL_CONTINUE;
1061
}
1062
EXPORT_SYMBOL_GPL(emulator_read_std);
1063
1064
static int emulator_write_std(unsigned long addr,
1065
const void *val,
1066
unsigned int bytes,
1067
struct kvm_vcpu *vcpu)
1068
{
1069
pr_unimpl(vcpu, "emulator_write_std: addr %lx n %d\n", addr, bytes);
1070
return X86EMUL_UNHANDLEABLE;
1071
}
1072
1073
static int emulator_read_emulated(unsigned long addr,
1074
void *val,
1075
unsigned int bytes,
1076
struct kvm_vcpu *vcpu)
1077
{
1078
struct kvm_io_device *mmio_dev;
1079
gpa_t gpa;
1080
1081
if (vcpu->mmio_read_completed) {
1082
memcpy(val, vcpu->mmio_data, bytes);
1083
vcpu->mmio_read_completed = 0;
1084
return X86EMUL_CONTINUE;
1085
}
1086
1087
gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1088
1089
/* For APIC access vmexit */
1090
if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1091
goto mmio;
1092
1093
if (emulator_read_std(addr, val, bytes, vcpu)
1094
== X86EMUL_CONTINUE)
1095
return X86EMUL_CONTINUE;
1096
if (gpa == UNMAPPED_GVA)
1097
return X86EMUL_PROPAGATE_FAULT;
1098
1099
mmio:
1100
/*
1101
* Is this MMIO handled locally?
1102
*/
1103
mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1104
if (mmio_dev) {
1105
kvm_iodevice_read(mmio_dev, gpa, bytes, val);
1106
return X86EMUL_CONTINUE;
1107
}
1108
1109
vcpu->mmio_needed = 1;
1110
vcpu->mmio_phys_addr = gpa;
1111
vcpu->mmio_size = bytes;
1112
vcpu->mmio_is_write = 0;
1113
1114
return X86EMUL_UNHANDLEABLE;
1115
}
1116
1117
static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
1118
const void *val, int bytes)
1119
{
1120
int ret;
1121
1122
ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
1123
if (ret < 0)
1124
return 0;
1125
kvm_mmu_pte_write(vcpu, gpa, val, bytes);
1126
return 1;
1127
}
1128
1129
static int emulator_write_emulated_onepage(unsigned long addr,
1130
const void *val,
1131
unsigned int bytes,
1132
struct kvm_vcpu *vcpu)
1133
{
1134
struct kvm_io_device *mmio_dev;
1135
gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, addr);
1136
1137
if (gpa == UNMAPPED_GVA) {
1138
kvm_x86_ops->inject_page_fault(vcpu, addr, 2);
1139
return X86EMUL_PROPAGATE_FAULT;
1140
}
1141
1142
/* For APIC access vmexit */
1143
if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
1144
goto mmio;
1145
1146
if (emulator_write_phys(vcpu, gpa, val, bytes))
1147
return X86EMUL_CONTINUE;
1148
1149
mmio:
1150
/*
1151
* Is this MMIO handled locally?
1152
*/
1153
mmio_dev = vcpu_find_mmio_dev(vcpu, gpa);
1154
if (mmio_dev) {
1155
kvm_iodevice_write(mmio_dev, gpa, bytes, val);
1156
return X86EMUL_CONTINUE;
1157
}
1158
1159
vcpu->mmio_needed = 1;
1160
vcpu->mmio_phys_addr = gpa;
1161
vcpu->mmio_size = bytes;
1162
vcpu->mmio_is_write = 1;
1163
memcpy(vcpu->mmio_data, val, bytes);
1164
1165
return X86EMUL_CONTINUE;
1166
}
1167
1168
int emulator_write_emulated(unsigned long addr,
1169
const void *val,
1170
unsigned int bytes,
1171
struct kvm_vcpu *vcpu)
1172
{
1173
/* Crossing a page boundary? */
1174
if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
1175
int rc, now;
1176
1177
now = -addr & ~PAGE_MASK;
1178
rc = emulator_write_emulated_onepage(addr, val, now, vcpu);
1179
if (rc != X86EMUL_CONTINUE)
1180
return rc;
1181
addr += now;
1182
val += now;
1183
bytes -= now;
1184
}
1185
return emulator_write_emulated_onepage(addr, val, bytes, vcpu);
1186
}
1187
EXPORT_SYMBOL_GPL(emulator_write_emulated);
1188
1189
static int emulator_cmpxchg_emulated(unsigned long addr,
1190
const void *old,
1191
const void *new,
1192
unsigned int bytes,
1193
struct kvm_vcpu *vcpu)
1194
{
1195
static int reported;
1196
1197
if (!reported) {
1198
reported = 1;
1199
printk(KERN_WARNING "kvm: emulating exchange as write\n");
1200
}
1201
return emulator_write_emulated(addr, new, bytes, vcpu);
1202
}
1203
1204
static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
1205
{
1206
return kvm_x86_ops->get_segment_base(vcpu, seg);
1207
}
1208
1209
int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
1210
{
1211
return X86EMUL_CONTINUE;
1212
}
1213
1214
int emulate_clts(struct kvm_vcpu *vcpu)
1215
{
1216
vcpu->cr0 &= ~X86_CR0_TS;
1217
kvm_x86_ops->set_cr0(vcpu, vcpu->cr0);
1218
return X86EMUL_CONTINUE;
1219
}
1220
1221
int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long *dest)
1222
{
1223
struct kvm_vcpu *vcpu = ctxt->vcpu;
1224
1225
switch (dr) {
1226
case 0 ... 3:
1227
*dest = kvm_x86_ops->get_dr(vcpu, dr);
1228
return X86EMUL_CONTINUE;
1229
default:
1230
pr_unimpl(vcpu, "%s: unexpected dr %u\n", __FUNCTION__, dr);
1231
return X86EMUL_UNHANDLEABLE;
1232
}
1233
}
1234
1235
int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
1236
{
1237
unsigned long mask = (ctxt->mode == X86EMUL_MODE_PROT64) ? ~0ULL : ~0U;
1238
int exception;
1239
1240
kvm_x86_ops->set_dr(ctxt->vcpu, dr, value & mask, &exception);
1241
if (exception) {
1242
/* FIXME: better handling */
1243
return X86EMUL_UNHANDLEABLE;
1244
}
1245
return X86EMUL_CONTINUE;
1246
}
1247
1248
void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
1249
{
1250
static int reported;
1251
u8 opcodes[4];
1252
unsigned long rip = vcpu->rip;
1253
unsigned long rip_linear;
1254
1255
rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
1256
1257
if (reported)
1258
return;
1259
1260
emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
1261
1262
printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
1263
context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
1264
reported = 1;
1265
}
1266
EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
1267
1268
struct x86_emulate_ops emulate_ops = {
1269
.read_std = emulator_read_std,
1270
.write_std = emulator_write_std,
1271
.read_emulated = emulator_read_emulated,
1272
.write_emulated = emulator_write_emulated,
1273
.cmpxchg_emulated = emulator_cmpxchg_emulated,
1274
};
1275
1276
int emulate_instruction(struct kvm_vcpu *vcpu,
1277
struct kvm_run *run,
1278
unsigned long cr2,
1279
u16 error_code,
1280
int no_decode)
1281
{
1282
int r;
1283
1284
vcpu->mmio_fault_cr2 = cr2;
1285
kvm_x86_ops->cache_regs(vcpu);
1286
1287
vcpu->mmio_is_write = 0;
1288
vcpu->pio.string = 0;
1289
1290
if (!no_decode) {
1291
int cs_db, cs_l;
1292
kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
1293
1294
vcpu->emulate_ctxt.vcpu = vcpu;
1295
vcpu->emulate_ctxt.eflags = kvm_x86_ops->get_rflags(vcpu);
1296
vcpu->emulate_ctxt.cr2 = cr2;
1297
vcpu->emulate_ctxt.mode =
1298
(vcpu->emulate_ctxt.eflags & X86_EFLAGS_VM)
1299
? X86EMUL_MODE_REAL : cs_l
1300
? X86EMUL_MODE_PROT64 : cs_db
1301
? X86EMUL_MODE_PROT32 : X86EMUL_MODE_PROT16;
1302
1303
if (vcpu->emulate_ctxt.mode == X86EMUL_MODE_PROT64) {
1304
vcpu->emulate_ctxt.cs_base = 0;
1305
vcpu->emulate_ctxt.ds_base = 0;
1306
vcpu->emulate_ctxt.es_base = 0;
1307
vcpu->emulate_ctxt.ss_base = 0;
1308
} else {
1309
vcpu->emulate_ctxt.cs_base =
1310
get_segment_base(vcpu, VCPU_SREG_CS);
1311
vcpu->emulate_ctxt.ds_base =
1312
get_segment_base(vcpu, VCPU_SREG_DS);
1313
vcpu->emulate_ctxt.es_base =
1314
get_segment_base(vcpu, VCPU_SREG_ES);
1315
vcpu->emulate_ctxt.ss_base =
1316
get_segment_base(vcpu, VCPU_SREG_SS);
1317
}
1318
1319
vcpu->emulate_ctxt.gs_base =
1320
get_segment_base(vcpu, VCPU_SREG_GS);
1321
vcpu->emulate_ctxt.fs_base =
1322
get_segment_base(vcpu, VCPU_SREG_FS);
1323
1324
r = x86_decode_insn(&vcpu->emulate_ctxt, &emulate_ops);
1325
if (r) {
1326
if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1327
return EMULATE_DONE;
1328
return EMULATE_FAIL;
1329
}
1330
}
1331
1332
r = x86_emulate_insn(&vcpu->emulate_ctxt, &emulate_ops);
1333
1334
if (vcpu->pio.string)
1335
return EMULATE_DO_MMIO;
1336
1337
if ((r || vcpu->mmio_is_write) && run) {
1338
run->exit_reason = KVM_EXIT_MMIO;
1339
run->mmio.phys_addr = vcpu->mmio_phys_addr;
1340
memcpy(run->mmio.data, vcpu->mmio_data, 8);
1341
run->mmio.len = vcpu->mmio_size;
1342
run->mmio.is_write = vcpu->mmio_is_write;
1343
}
1344
1345
if (r) {
1346
if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
1347
return EMULATE_DONE;
1348
if (!vcpu->mmio_needed) {
1349
kvm_report_emulation_failure(vcpu, "mmio");
1350
return EMULATE_FAIL;
1351
}
1352
return EMULATE_DO_MMIO;
1353
}
1354
1355
kvm_x86_ops->decache_regs(vcpu);
1356
kvm_x86_ops->set_rflags(vcpu, vcpu->emulate_ctxt.eflags);
1357
1358
if (vcpu->mmio_is_write) {
1359
vcpu->mmio_needed = 0;
1360
return EMULATE_DO_MMIO;
1361
}
1362
1363
return EMULATE_DONE;
1364
}
1365
EXPORT_SYMBOL_GPL(emulate_instruction);
1366
1367
static void free_pio_guest_pages(struct kvm_vcpu *vcpu)
1368
{
1369
int i;
1370
1371
for (i = 0; i < ARRAY_SIZE(vcpu->pio.guest_pages); ++i)
1372
if (vcpu->pio.guest_pages[i]) {
1373
kvm_release_page(vcpu->pio.guest_pages[i]);
1374
vcpu->pio.guest_pages[i] = NULL;
1375
}
1376
}
1377
1378
static int pio_copy_data(struct kvm_vcpu *vcpu)
1379
{
1380
void *p = vcpu->pio_data;
1381
void *q;
1382
unsigned bytes;
1383
int nr_pages = vcpu->pio.guest_pages[1] ? 2 : 1;
1384
1385
q = vmap(vcpu->pio.guest_pages, nr_pages, VM_READ|VM_WRITE,
1386
PAGE_KERNEL);
1387
if (!q) {
1388
free_pio_guest_pages(vcpu);
1389
return -ENOMEM;
1390
}
1391
q += vcpu->pio.guest_page_offset;
1392
bytes = vcpu->pio.size * vcpu->pio.cur_count;
1393
if (vcpu->pio.in)
1394
memcpy(q, p, bytes);
1395
else
1396
memcpy(p, q, bytes);
1397
q -= vcpu->pio.guest_page_offset;
1398
vunmap(q);
1399
free_pio_guest_pages(vcpu);
1400
return 0;
1401
}
1402
1403
int complete_pio(struct kvm_vcpu *vcpu)
1404
{
1405
struct kvm_pio_request *io = &vcpu->pio;
1406
long delta;
1407
int r;
1408
1409
kvm_x86_ops->cache_regs(vcpu);
1410
1411
if (!io->string) {
1412
if (io->in)
1413
memcpy(&vcpu->regs[VCPU_REGS_RAX], vcpu->pio_data,
1414
io->size);
1415
} else {
1416
if (io->in) {
1417
r = pio_copy_data(vcpu);
1418
if (r) {
1419
kvm_x86_ops->cache_regs(vcpu);
1420
return r;
1421
}
1422
}
1423
1424
delta = 1;
1425
if (io->rep) {
1426
delta *= io->cur_count;
1427
/*
1428
* The size of the register should really depend on
1429
* current address size.
1430
*/
1431
vcpu->regs[VCPU_REGS_RCX] -= delta;
1432
}
1433
if (io->down)
1434
delta = -delta;
1435
delta *= io->size;
1436
if (io->in)
1437
vcpu->regs[VCPU_REGS_RDI] += delta;
1438
else
1439
vcpu->regs[VCPU_REGS_RSI] += delta;
1440
}
1441
1442
kvm_x86_ops->decache_regs(vcpu);
1443
1444
io->count -= io->cur_count;
1445
io->cur_count = 0;
1446
1447
return 0;
1448
}
1449
1450
static void kernel_pio(struct kvm_io_device *pio_dev,
1451
struct kvm_vcpu *vcpu,
1452
void *pd)
1453
{
1454
/* TODO: String I/O for in kernel device */
1455
1456
mutex_lock(&vcpu->kvm->lock);
1457
if (vcpu->pio.in)
1458
kvm_iodevice_read(pio_dev, vcpu->pio.port,
1459
vcpu->pio.size,
1460
pd);
1461
else
1462
kvm_iodevice_write(pio_dev, vcpu->pio.port,
1463
vcpu->pio.size,
1464
pd);
1465
mutex_unlock(&vcpu->kvm->lock);
1466
}
1467
1468
static void pio_string_write(struct kvm_io_device *pio_dev,
1469
struct kvm_vcpu *vcpu)
1470
{
1471
struct kvm_pio_request *io = &vcpu->pio;
1472
void *pd = vcpu->pio_data;
1473
int i;
1474
1475
mutex_lock(&vcpu->kvm->lock);
1476
for (i = 0; i < io->cur_count; i++) {
1477
kvm_iodevice_write(pio_dev, io->port,
1478
io->size,
1479
pd);
1480
pd += io->size;
1481
}
1482
mutex_unlock(&vcpu->kvm->lock);
1483
}
1484
1485
static struct kvm_io_device *vcpu_find_pio_dev(struct kvm_vcpu *vcpu,
1486
gpa_t addr)
1487
{
1488
return kvm_io_bus_find_dev(&vcpu->kvm->pio_bus, addr);
1489
}
1490
1491
int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1492
int size, unsigned port)
1493
{
1494
struct kvm_io_device *pio_dev;
1495
1496
vcpu->run->exit_reason = KVM_EXIT_IO;
1497
vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1498
vcpu->run->io.size = vcpu->pio.size = size;
1499
vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1500
vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = 1;
1501
vcpu->run->io.port = vcpu->pio.port = port;
1502
vcpu->pio.in = in;
1503
vcpu->pio.string = 0;
1504
vcpu->pio.down = 0;
1505
vcpu->pio.guest_page_offset = 0;
1506
vcpu->pio.rep = 0;
1507
1508
kvm_x86_ops->cache_regs(vcpu);
1509
memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
1510
kvm_x86_ops->decache_regs(vcpu);
1511
1512
kvm_x86_ops->skip_emulated_instruction(vcpu);
1513
1514
pio_dev = vcpu_find_pio_dev(vcpu, port);
1515
if (pio_dev) {
1516
kernel_pio(pio_dev, vcpu, vcpu->pio_data);
1517
complete_pio(vcpu);
1518
return 1;
1519
}
1520
return 0;
1521
}
1522
EXPORT_SYMBOL_GPL(kvm_emulate_pio);
1523
1524
int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
1525
int size, unsigned long count, int down,
1526
gva_t address, int rep, unsigned port)
1527
{
1528
unsigned now, in_page;
1529
int i, ret = 0;
1530
int nr_pages = 1;
1531
struct page *page;
1532
struct kvm_io_device *pio_dev;
1533
1534
vcpu->run->exit_reason = KVM_EXIT_IO;
1535
vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1536
vcpu->run->io.size = vcpu->pio.size = size;
1537
vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
1538
vcpu->run->io.count = vcpu->pio.count = vcpu->pio.cur_count = count;
1539
vcpu->run->io.port = vcpu->pio.port = port;
1540
vcpu->pio.in = in;
1541
vcpu->pio.string = 1;
1542
vcpu->pio.down = down;
1543
vcpu->pio.guest_page_offset = offset_in_page(address);
1544
vcpu->pio.rep = rep;
1545
1546
if (!count) {
1547
kvm_x86_ops->skip_emulated_instruction(vcpu);
1548
return 1;
1549
}
1550
1551
if (!down)
1552
in_page = PAGE_SIZE - offset_in_page(address);
1553
else
1554
in_page = offset_in_page(address) + size;
1555
now = min(count, (unsigned long)in_page / size);
1556
if (!now) {
1557
/*
1558
* String I/O straddles page boundary. Pin two guest pages
1559
* so that we satisfy atomicity constraints. Do just one
1560
* transaction to avoid complexity.
1561
*/
1562
nr_pages = 2;
1563
now = 1;
1564
}
1565
if (down) {
1566
/*
1567
* String I/O in reverse. Yuck. Kill the guest, fix later.
1568
*/
1569
pr_unimpl(vcpu, "guest string pio down\n");
1570
inject_gp(vcpu);
1571
return 1;
1572
}
1573
vcpu->run->io.count = now;
1574
vcpu->pio.cur_count = now;
1575
1576
if (vcpu->pio.cur_count == vcpu->pio.count)
1577
kvm_x86_ops->skip_emulated_instruction(vcpu);
1578
1579
for (i = 0; i < nr_pages; ++i) {
1580
mutex_lock(&vcpu->kvm->lock);
1581
page = gva_to_page(vcpu, address + i * PAGE_SIZE);
1582
vcpu->pio.guest_pages[i] = page;
1583
mutex_unlock(&vcpu->kvm->lock);
1584
if (!page) {
1585
inject_gp(vcpu);
1586
free_pio_guest_pages(vcpu);
1587
return 1;
1588
}
1589
}
1590
1591
pio_dev = vcpu_find_pio_dev(vcpu, port);
1592
if (!vcpu->pio.in) {
1593
/* string PIO write */
1594
ret = pio_copy_data(vcpu);
1595
if (ret >= 0 && pio_dev) {
1596
pio_string_write(pio_dev, vcpu);
1597
complete_pio(vcpu);
1598
if (vcpu->pio.count == 0)
1599
ret = 1;
1600
}
1601
} else if (pio_dev)
1602
pr_unimpl(vcpu, "no string pio read support yet, "
1603
"port %x size %d count %ld\n",
1604
port, size, count);
1605
1606
return ret;
1607
}
1608
EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
1609
318
1610
__init void kvm_arch_init(void)
319
1611
{
320
1612
kvm_init_msr_list();
321
1613
}
1614
1615
int kvm_emulate_halt(struct kvm_vcpu *vcpu)
1616
{
1617
++vcpu->stat.halt_exits;
1618
if (irqchip_in_kernel(vcpu->kvm)) {
1619
vcpu->mp_state = VCPU_MP_STATE_HALTED;
1620
kvm_vcpu_block(vcpu);
1621
if (vcpu->mp_state != VCPU_MP_STATE_RUNNABLE)
1622
return -EINTR;
1623
return 1;
1624
} else {
1625
vcpu->run->exit_reason = KVM_EXIT_HLT;
1626
return 0;
1627
}
1628
}
1629
EXPORT_SYMBOL_GPL(kvm_emulate_halt);
1630
1631
int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
1632
{
1633
unsigned long nr, a0, a1, a2, a3, ret;
1634
1635
kvm_x86_ops->cache_regs(vcpu);
1636
1637
nr = vcpu->regs[VCPU_REGS_RAX];
1638
a0 = vcpu->regs[VCPU_REGS_RBX];
1639
a1 = vcpu->regs[VCPU_REGS_RCX];
1640
a2 = vcpu->regs[VCPU_REGS_RDX];
1641
a3 = vcpu->regs[VCPU_REGS_RSI];
1642
1643
if (!is_long_mode(vcpu)) {
1644
nr &= 0xFFFFFFFF;
1645
a0 &= 0xFFFFFFFF;
1646
a1 &= 0xFFFFFFFF;
1647
a2 &= 0xFFFFFFFF;
1648
a3 &= 0xFFFFFFFF;
1649
}
1650
1651
switch (nr) {
1652
default:
1653
ret = -KVM_ENOSYS;
1654
break;
1655
}
1656
vcpu->regs[VCPU_REGS_RAX] = ret;
1657
kvm_x86_ops->decache_regs(vcpu);
1658
return 0;
1659
}
1660
EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
1661
1662
int kvm_fix_hypercall(struct kvm_vcpu *vcpu)
1663
{
1664
char instruction[3];
1665
int ret = 0;
1666
1667
mutex_lock(&vcpu->kvm->lock);
1668
1669
/*
1670
* Blow out the MMU to ensure that no other VCPU has an active mapping
1671
* to ensure that the updated hypercall appears atomically across all
1672
* VCPUs.
1673
*/
1674
kvm_mmu_zap_all(vcpu->kvm);
1675
1676
kvm_x86_ops->cache_regs(vcpu);
1677
kvm_x86_ops->patch_hypercall(vcpu, instruction);
1678
if (emulator_write_emulated(vcpu->rip, instruction, 3, vcpu)
1679
!= X86EMUL_CONTINUE)
1680
ret = -EFAULT;
1681
1682
mutex_unlock(&vcpu->kvm->lock);
1683
1684
return ret;
1685
}
1686
1687
static u64 mk_cr_64(u64 curr_cr, u32 new_val)
1688
{
1689
return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
1690
}
1691
1692
void realmode_lgdt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1693
{
1694
struct descriptor_table dt = { limit, base };
1695
1696
kvm_x86_ops->set_gdt(vcpu, &dt);
1697
}
1698
1699
void realmode_lidt(struct kvm_vcpu *vcpu, u16 limit, unsigned long base)
1700
{
1701
struct descriptor_table dt = { limit, base };
1702
1703
kvm_x86_ops->set_idt(vcpu, &dt);
1704
}
1705
1706
void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
1707
unsigned long *rflags)
1708
{
1709
lmsw(vcpu, msw);
1710
*rflags = kvm_x86_ops->get_rflags(vcpu);
1711
}
1712
1713
unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr)
1714
{
1715
kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1716
switch (cr) {
1717
case 0:
1718
return vcpu->cr0;
1719
case 2:
1720
return vcpu->cr2;
1721
case 3:
1722
return vcpu->cr3;
1723
case 4:
1724
return vcpu->cr4;
1725
default:
1726
vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1727
return 0;
1728
}
1729
}
1730
1731
void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long val,
1732
unsigned long *rflags)
1733
{
1734
switch (cr) {
1735
case 0:
1736
set_cr0(vcpu, mk_cr_64(vcpu->cr0, val));
1737
*rflags = kvm_x86_ops->get_rflags(vcpu);
1738
break;
1739
case 2:
1740
vcpu->cr2 = val;
1741
break;
1742
case 3:
1743
set_cr3(vcpu, val);
1744
break;
1745
case 4:
1746
set_cr4(vcpu, mk_cr_64(vcpu->cr4, val));
1747
break;
1748
default:
1749
vcpu_printf(vcpu, "%s: unexpected cr %u\n", __FUNCTION__, cr);
1750
}
1751
}
1752
1753
void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1754
{
1755
int i;
1756
u32 function;
1757
struct kvm_cpuid_entry *e, *best;
1758
1759
kvm_x86_ops->cache_regs(vcpu);
1760
function = vcpu->regs[VCPU_REGS_RAX];
1761
vcpu->regs[VCPU_REGS_RAX] = 0;
1762
vcpu->regs[VCPU_REGS_RBX] = 0;
1763
vcpu->regs[VCPU_REGS_RCX] = 0;
1764
vcpu->regs[VCPU_REGS_RDX] = 0;
1765
best = NULL;
1766
for (i = 0; i < vcpu->cpuid_nent; ++i) {
1767
e = &vcpu->cpuid_entries[i];
1768
if (e->function == function) {
1769
best = e;
1770
break;
1771
}
1772
/*
1773
* Both basic or both extended?
1774
*/
1775
if (((e->function ^ function) & 0x80000000) == 0)
1776
if (!best || e->function > best->function)
1777
best = e;
1778
}
1779
if (best) {
1780
vcpu->regs[VCPU_REGS_RAX] = best->eax;
1781
vcpu->regs[VCPU_REGS_RBX] = best->ebx;
1782
vcpu->regs[VCPU_REGS_RCX] = best->ecx;
1783
vcpu->regs[VCPU_REGS_RDX] = best->edx;
1784
}
1785
kvm_x86_ops->decache_regs(vcpu);
1786
kvm_x86_ops->skip_emulated_instruction(vcpu);
1787
}
1788
EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1789
1790
/*
1791
* Check if userspace requested an interrupt window, and that the
1792
* interrupt window is open.
1793
*
1794
* No need to exit to userspace if we already have an interrupt queued.
1795
*/
1796
static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1797
struct kvm_run *kvm_run)
1798
{
1799
return (!vcpu->irq_summary &&
1800
kvm_run->request_interrupt_window &&
1801
vcpu->interrupt_window_open &&
1802
(kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
1803
}
1804
1805
static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1806
struct kvm_run *kvm_run)
1807
{
1808
kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
1809
kvm_run->cr8 = get_cr8(vcpu);
1810
kvm_run->apic_base = kvm_get_apic_base(vcpu);
1811
if (irqchip_in_kernel(vcpu->kvm))
1812
kvm_run->ready_for_interrupt_injection = 1;
1813
else
1814
kvm_run->ready_for_interrupt_injection =
1815
(vcpu->interrupt_window_open &&
1816
vcpu->irq_summary == 0);
1817
}
1818
1819
static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1820
{
1821
int r;
1822
1823
if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
1824
pr_debug("vcpu %d received sipi with vector # %x\n",
1825
vcpu->vcpu_id, vcpu->sipi_vector);
1826
kvm_lapic_reset(vcpu);
1827
r = kvm_x86_ops->vcpu_reset(vcpu);
1828
if (r)
1829
return r;
1830
vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
1831
}
1832
1833
preempted:
1834
if (vcpu->guest_debug.enabled)
1835
kvm_x86_ops->guest_debug_pre(vcpu);
1836
1837
again:
1838
r = kvm_mmu_reload(vcpu);
1839
if (unlikely(r))
1840
goto out;
1841
1842
kvm_inject_pending_timer_irqs(vcpu);
1843
1844
preempt_disable();
1845
in_special_section();
1846
1847
kvm_x86_ops->prepare_guest_switch(vcpu);
1848
kvm_load_guest_fpu(vcpu);
1849
1850
local_irq_disable();
1851
1852
if (signal_pending(current)) {
1853
local_irq_enable();
1854
preempt_enable();
1855
r = -EINTR;
1856
kvm_run->exit_reason = KVM_EXIT_INTR;
1857
++vcpu->stat.signal_exits;
1858
goto out;
1859
}
1860
1861
if (irqchip_in_kernel(vcpu->kvm))
1862
kvm_x86_ops->inject_pending_irq(vcpu);
1863
else if (!vcpu->mmio_read_completed)
1864
kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
1865
1866
vcpu->guest_mode = 1;
1867
kvm_guest_enter();
1868
1869
if (vcpu->requests)
1870
if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
1871
kvm_x86_ops->tlb_flush(vcpu);
1872
1873
kvm_x86_ops->run(vcpu, kvm_run);
1874
special_reload_dr7();
1875
1876
vcpu->guest_mode = 0;
1877
local_irq_enable();
1878
1879
++vcpu->stat.exits;
1880
1881
/*
1882
* We must have an instruction between local_irq_enable() and
1883
* kvm_guest_exit(), so the timer interrupt isn't delayed by
1884
* the interrupt shadow. The stat.exits increment will do nicely.
1885
* But we need to prevent reordering, hence this barrier():
1886
*/
1887
barrier();
1888
1889
kvm_guest_exit();
1890
1891
preempt_enable();
1892
1893
/*
1894
* Profile KVM exit RIPs:
1895
*/
1896
if (unlikely(prof_on == KVM_PROFILING)) {
1897
kvm_x86_ops->cache_regs(vcpu);
1898
profile_hit(KVM_PROFILING, (void *)vcpu->rip);
1899
}
1900
1901
r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
1902
1903
if (r > 0) {
1904
if (dm_request_for_irq_injection(vcpu, kvm_run)) {
1905
r = -EINTR;
1906
kvm_run->exit_reason = KVM_EXIT_INTR;
1907
++vcpu->stat.request_irq_exits;
1908
goto out;
1909
}
1910
if (!need_resched()) {
1911
++vcpu->stat.light_exits;
1912
goto again;
1913
}
1914
}
1915
1916
out:
1917
if (r > 0) {
1918
kvm_resched(vcpu);
1919
goto preempted;
1920
}
1921
1922
post_kvm_run_save(vcpu, kvm_run);
1923
1924
return r;
1925
}
1926
1927
int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1928
{
1929
int r;
1930
sigset_t sigsaved;
1931
1932
vcpu_load(vcpu);
1933
1934
if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
1935
kvm_vcpu_block(vcpu);
1936
vcpu_put(vcpu);
1937
return -EAGAIN;
1938
}
1939
1940
if (vcpu->sigset_active)
1941
sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
1942
1943
/* re-sync apic's tpr */
1944
if (!irqchip_in_kernel(vcpu->kvm))
1945
set_cr8(vcpu, kvm_run->cr8);
1946
1947
if (vcpu->pio.cur_count) {
1948
r = complete_pio(vcpu);
1949
if (r)
1950
goto out;
1951
}
1952
#if CONFIG_HAS_IOMEM
1953
if (vcpu->mmio_needed) {
1954
memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
1955
vcpu->mmio_read_completed = 1;
1956
vcpu->mmio_needed = 0;
1957
r = emulate_instruction(vcpu, kvm_run,
1958
vcpu->mmio_fault_cr2, 0, 1);
1959
if (r == EMULATE_DO_MMIO) {
1960
/*
1961
* Read-modify-write. Back to userspace.
1962
*/
1963
r = 0;
1964
goto out;
1965
}
1966
}
1967
#endif
1968
if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
1969
kvm_x86_ops->cache_regs(vcpu);
1970
vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
1971
kvm_x86_ops->decache_regs(vcpu);
1972
}
1973
1974
r = __vcpu_run(vcpu, kvm_run);
1975
1976
out:
1977
if (vcpu->sigset_active)
1978
sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1979
1980
vcpu_put(vcpu);
1981
return r;
1982
}
1983
1984
int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
1985
{
1986
vcpu_load(vcpu);
1987
1988
kvm_x86_ops->cache_regs(vcpu);
1989
1990
regs->rax = vcpu->regs[VCPU_REGS_RAX];
1991
regs->rbx = vcpu->regs[VCPU_REGS_RBX];
1992
regs->rcx = vcpu->regs[VCPU_REGS_RCX];
1993
regs->rdx = vcpu->regs[VCPU_REGS_RDX];
1994
regs->rsi = vcpu->regs[VCPU_REGS_RSI];
1995
regs->rdi = vcpu->regs[VCPU_REGS_RDI];
1996
regs->rsp = vcpu->regs[VCPU_REGS_RSP];
1997
regs->rbp = vcpu->regs[VCPU_REGS_RBP];
1998
#ifdef CONFIG_X86_64
1999
regs->r8 = vcpu->regs[VCPU_REGS_R8];
2000
regs->r9 = vcpu->regs[VCPU_REGS_R9];
2001
regs->r10 = vcpu->regs[VCPU_REGS_R10];
2002
regs->r11 = vcpu->regs[VCPU_REGS_R11];
2003
regs->r12 = vcpu->regs[VCPU_REGS_R12];
2004
regs->r13 = vcpu->regs[VCPU_REGS_R13];
2005
regs->r14 = vcpu->regs[VCPU_REGS_R14];
2006
regs->r15 = vcpu->regs[VCPU_REGS_R15];
2007
#endif
2008
2009
regs->rip = vcpu->rip;
2010
regs->rflags = kvm_x86_ops->get_rflags(vcpu);
2011
2012
/*
2013
* Don't leak debug flags in case they were set for guest debugging
2014
*/
2015
if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
2016
regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
2017
2018
vcpu_put(vcpu);
2019
2020
return 0;
2021
}
2022
2023
int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
2024
{
2025
vcpu_load(vcpu);
2026
2027
vcpu->regs[VCPU_REGS_RAX] = regs->rax;
2028
vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
2029
vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
2030
vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
2031
vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
2032
vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
2033
vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
2034
vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
2035
#ifdef CONFIG_X86_64
2036
vcpu->regs[VCPU_REGS_R8] = regs->r8;
2037
vcpu->regs[VCPU_REGS_R9] = regs->r9;
2038
vcpu->regs[VCPU_REGS_R10] = regs->r10;
2039
vcpu->regs[VCPU_REGS_R11] = regs->r11;
2040
vcpu->regs[VCPU_REGS_R12] = regs->r12;
2041
vcpu->regs[VCPU_REGS_R13] = regs->r13;
2042
vcpu->regs[VCPU_REGS_R14] = regs->r14;
2043
vcpu->regs[VCPU_REGS_R15] = regs->r15;
2044
#endif
2045
2046
vcpu->rip = regs->rip;
2047
kvm_x86_ops->set_rflags(vcpu, regs->rflags);
2048
2049
kvm_x86_ops->decache_regs(vcpu);
2050
2051
vcpu_put(vcpu);
2052
2053
return 0;
2054
}
2055
2056
static void get_segment(struct kvm_vcpu *vcpu,
2057
struct kvm_segment *var, int seg)
2058
{
2059
return kvm_x86_ops->get_segment(vcpu, var, seg);
2060
}
2061
2062
void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2063
{
2064
struct kvm_segment cs;
2065
2066
get_segment(vcpu, &cs, VCPU_SREG_CS);
2067
*db = cs.db;
2068
*l = cs.l;
2069
}
2070
EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
2071
2072
int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
2073
struct kvm_sregs *sregs)
2074
{
2075
struct descriptor_table dt;
2076
int pending_vec;
2077
2078
vcpu_load(vcpu);
2079
2080
get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2081
get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2082
get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2083
get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2084
get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2085
get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2086
2087
get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2088
get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2089
2090
kvm_x86_ops->get_idt(vcpu, &dt);
2091
sregs->idt.limit = dt.limit;
2092
sregs->idt.base = dt.base;
2093
kvm_x86_ops->get_gdt(vcpu, &dt);
2094
sregs->gdt.limit = dt.limit;
2095
sregs->gdt.base = dt.base;
2096
2097
kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2098
sregs->cr0 = vcpu->cr0;
2099
sregs->cr2 = vcpu->cr2;
2100
sregs->cr3 = vcpu->cr3;
2101
sregs->cr4 = vcpu->cr4;
2102
sregs->cr8 = get_cr8(vcpu);
2103
sregs->efer = vcpu->shadow_efer;
2104
sregs->apic_base = kvm_get_apic_base(vcpu);
2105
2106
if (irqchip_in_kernel(vcpu->kvm)) {
2107
memset(sregs->interrupt_bitmap, 0,
2108
sizeof sregs->interrupt_bitmap);
2109
pending_vec = kvm_x86_ops->get_irq(vcpu);
2110
if (pending_vec >= 0)
2111
set_bit(pending_vec,
2112
(unsigned long *)sregs->interrupt_bitmap);
2113
} else
2114
memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
2115
sizeof sregs->interrupt_bitmap);
2116
2117
vcpu_put(vcpu);
2118
2119
return 0;
2120
}
2121
2122
static void set_segment(struct kvm_vcpu *vcpu,
2123
struct kvm_segment *var, int seg)
2124
{
2125
return kvm_x86_ops->set_segment(vcpu, var, seg);
2126
}
2127
2128
int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
2129
struct kvm_sregs *sregs)
2130
{
2131
int mmu_reset_needed = 0;
2132
int i, pending_vec, max_bits;
2133
struct descriptor_table dt;
2134
2135
vcpu_load(vcpu);
2136
2137
dt.limit = sregs->idt.limit;
2138
dt.base = sregs->idt.base;
2139
kvm_x86_ops->set_idt(vcpu, &dt);
2140
dt.limit = sregs->gdt.limit;
2141
dt.base = sregs->gdt.base;
2142
kvm_x86_ops->set_gdt(vcpu, &dt);
2143
2144
vcpu->cr2 = sregs->cr2;
2145
mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
2146
vcpu->cr3 = sregs->cr3;
2147
2148
set_cr8(vcpu, sregs->cr8);
2149
2150
mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
2151
#ifdef CONFIG_X86_64
2152
kvm_x86_ops->set_efer(vcpu, sregs->efer);
2153
#endif
2154
kvm_set_apic_base(vcpu, sregs->apic_base);
2155
2156
kvm_x86_ops->decache_cr4_guest_bits(vcpu);
2157
2158
mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
2159
vcpu->cr0 = sregs->cr0;
2160
kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
2161
2162
mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
2163
kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
2164
if (!is_long_mode(vcpu) && is_pae(vcpu))
2165
load_pdptrs(vcpu, vcpu->cr3);
2166
2167
if (mmu_reset_needed)
2168
kvm_mmu_reset_context(vcpu);
2169
2170
if (!irqchip_in_kernel(vcpu->kvm)) {
2171
memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
2172
sizeof vcpu->irq_pending);
2173
vcpu->irq_summary = 0;
2174
for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
2175
if (vcpu->irq_pending[i])
2176
__set_bit(i, &vcpu->irq_summary);
2177
} else {
2178
max_bits = (sizeof sregs->interrupt_bitmap) << 3;
2179
pending_vec = find_first_bit(
2180
(const unsigned long *)sregs->interrupt_bitmap,
2181
max_bits);
2182
/* Only pending external irq is handled here */
2183
if (pending_vec < max_bits) {
2184
kvm_x86_ops->set_irq(vcpu, pending_vec);
2185
pr_debug("Set back pending irq %d\n",
2186
pending_vec);
2187
}
2188
}
2189
2190
set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
2191
set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
2192
set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
2193
set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
2194
set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
2195
set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
2196
2197
set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
2198
set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
2199
2200
vcpu_put(vcpu);
2201
2202
return 0;
2203
}
2204
2205
int kvm_arch_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
2206
struct kvm_debug_guest *dbg)
2207
{
2208
int r;
2209
2210
vcpu_load(vcpu);
2211
2212
r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
2213
2214
vcpu_put(vcpu);
2215
2216
return r;
2217
}
2218
2219
/*
2220
* fxsave fpu state. Taken from x86_64/processor.h. To be killed when
2221
* we have asm/x86/processor.h
2222
*/
2223
struct fxsave {
2224
u16 cwd;
2225
u16 swd;
2226
u16 twd;
2227
u16 fop;
2228
u64 rip;
2229
u64 rdp;
2230
u32 mxcsr;
2231
u32 mxcsr_mask;
2232
u32 st_space[32]; /* 8*16 bytes for each FP-reg = 128 bytes */
2233
#ifdef CONFIG_X86_64
2234
u32 xmm_space[64]; /* 16*16 bytes for each XMM-reg = 256 bytes */
2235
#else
2236
u32 xmm_space[32]; /* 8*16 bytes for each XMM-reg = 128 bytes */
2237
#endif
2238
};
2239
2240
int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2241
{
2242
struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2243
2244
vcpu_load(vcpu);
2245
2246
memcpy(fpu->fpr, fxsave->st_space, 128);
2247
fpu->fcw = fxsave->cwd;
2248
fpu->fsw = fxsave->swd;
2249
fpu->ftwx = fxsave->twd;
2250
fpu->last_opcode = fxsave->fop;
2251
fpu->last_ip = fxsave->rip;
2252
fpu->last_dp = fxsave->rdp;
2253
memcpy(fpu->xmm, fxsave->xmm_space, sizeof fxsave->xmm_space);
2254
2255
vcpu_put(vcpu);
2256
2257
return 0;
2258
}
2259
2260
int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
2261
{
2262
struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
2263
2264
vcpu_load(vcpu);
2265
2266
memcpy(fxsave->st_space, fpu->fpr, 128);
2267
fxsave->cwd = fpu->fcw;
2268
fxsave->swd = fpu->fsw;
2269
fxsave->twd = fpu->ftwx;
2270
fxsave->fop = fpu->last_opcode;
2271
fxsave->rip = fpu->last_ip;
2272
fxsave->rdp = fpu->last_dp;
2273
memcpy(fxsave->xmm_space, fpu->xmm, sizeof fxsave->xmm_space);
2274
2275
vcpu_put(vcpu);
2276
2277
return 0;
2278
}
2279
2280
void fx_init(struct kvm_vcpu *vcpu)
2281
{
2282
unsigned after_mxcsr_mask;
2283
2284
/* Initialize guest FPU by resetting ours and saving into guest's */
2285
preempt_disable();
2286
fx_save(&vcpu->host_fx_image);
2287
fpu_init();
2288
fx_save(&vcpu->guest_fx_image);
2289
fx_restore(&vcpu->host_fx_image);
2290
preempt_enable();
2291
2292
vcpu->cr0 |= X86_CR0_ET;
2293
after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
2294
vcpu->guest_fx_image.mxcsr = 0x1f80;
2295
memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
2296
0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
2297
}
2298
EXPORT_SYMBOL_GPL(fx_init);
2299
2300
void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
2301
{
2302
if (!vcpu->fpu_active || vcpu->guest_fpu_loaded)
2303
return;
2304
2305
vcpu->guest_fpu_loaded = 1;
2306
fx_save(&vcpu->host_fx_image);
2307
fx_restore(&vcpu->guest_fx_image);
2308
}
2309
EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
2310
2311
void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
2312
{
2313
if (!vcpu->guest_fpu_loaded)
2314
return;
2315
2316
vcpu->guest_fpu_loaded = 0;
2317
fx_save(&vcpu->guest_fx_image);
2318
fx_restore(&vcpu->host_fx_image);
2319
}
2320
EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
Loggerhead is a web-based interface for Breezy
Version: 2.0.1