aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/kvm
diff options
context:
space:
mode:
Diffstat (limited to 'arch/x86/kvm')
-rw-r--r--arch/x86/kvm/Kconfig3
-rw-r--r--arch/x86/kvm/Makefile2
-rw-r--r--arch/x86/kvm/cpuid.c670
-rw-r--r--arch/x86/kvm/cpuid.h46
-rw-r--r--arch/x86/kvm/emulate.c436
-rw-r--r--arch/x86/kvm/i8254.c14
-rw-r--r--arch/x86/kvm/i8259.c24
-rw-r--r--arch/x86/kvm/lapic.c3
-rw-r--r--arch/x86/kvm/lapic.h1
-rw-r--r--arch/x86/kvm/mmu.c547
-rw-r--r--arch/x86/kvm/mmu_audit.c29
-rw-r--r--arch/x86/kvm/mmutrace.h19
-rw-r--r--arch/x86/kvm/paging_tmpl.h86
-rw-r--r--arch/x86/kvm/pmu.c533
-rw-r--r--arch/x86/kvm/svm.c15
-rw-r--r--arch/x86/kvm/timer.c26
-rw-r--r--arch/x86/kvm/vmx.c63
-rw-r--r--arch/x86/kvm/x86.c1012
-rw-r--r--arch/x86/kvm/x86.h5
19 files changed, 2083 insertions, 1451 deletions
diff --git a/arch/x86/kvm/Kconfig b/arch/x86/kvm/Kconfig
index ff5790d8e990..1a7fe868f375 100644
--- a/arch/x86/kvm/Kconfig
+++ b/arch/x86/kvm/Kconfig
@@ -35,6 +35,7 @@ config KVM
35 select KVM_MMIO 35 select KVM_MMIO
36 select TASKSTATS 36 select TASKSTATS
37 select TASK_DELAY_ACCT 37 select TASK_DELAY_ACCT
38 select PERF_EVENTS
38 ---help--- 39 ---help---
39 Support hosting fully virtualized guest machines using hardware 40 Support hosting fully virtualized guest machines using hardware
40 virtualization extensions. You will need a fairly recent 41 virtualization extensions. You will need a fairly recent
@@ -52,6 +53,8 @@ config KVM
52config KVM_INTEL 53config KVM_INTEL
53 tristate "KVM for Intel processors support" 54 tristate "KVM for Intel processors support"
54 depends on KVM 55 depends on KVM
56 # for perf_guest_get_msrs():
57 depends on CPU_SUP_INTEL
55 ---help--- 58 ---help---
56 Provides support for KVM on Intel processors equipped with the VT 59 Provides support for KVM on Intel processors equipped with the VT
57 extensions. 60 extensions.
diff --git a/arch/x86/kvm/Makefile b/arch/x86/kvm/Makefile
index f15501f431c8..4f579e8dcacf 100644
--- a/arch/x86/kvm/Makefile
+++ b/arch/x86/kvm/Makefile
@@ -12,7 +12,7 @@ kvm-$(CONFIG_IOMMU_API) += $(addprefix ../../../virt/kvm/, iommu.o)
12kvm-$(CONFIG_KVM_ASYNC_PF) += $(addprefix ../../../virt/kvm/, async_pf.o) 12kvm-$(CONFIG_KVM_ASYNC_PF) += $(addprefix ../../../virt/kvm/, async_pf.o)
13 13
14kvm-y += x86.o mmu.o emulate.o i8259.o irq.o lapic.o \ 14kvm-y += x86.o mmu.o emulate.o i8259.o irq.o lapic.o \
15 i8254.o timer.o 15 i8254.o timer.o cpuid.o pmu.o
16kvm-intel-y += vmx.o 16kvm-intel-y += vmx.o
17kvm-amd-y += svm.o 17kvm-amd-y += svm.o
18 18
diff --git a/arch/x86/kvm/cpuid.c b/arch/x86/kvm/cpuid.c
new file mode 100644
index 000000000000..89b02bfaaca5
--- /dev/null
+++ b/arch/x86/kvm/cpuid.c
@@ -0,0 +1,670 @@
1/*
2 * Kernel-based Virtual Machine driver for Linux
3 * cpuid support routines
4 *
5 * derived from arch/x86/kvm/x86.c
6 *
7 * Copyright 2011 Red Hat, Inc. and/or its affiliates.
8 * Copyright IBM Corporation, 2008
9 *
10 * This work is licensed under the terms of the GNU GPL, version 2. See
11 * the COPYING file in the top-level directory.
12 *
13 */
14
15#include <linux/kvm_host.h>
16#include <linux/module.h>
17#include <linux/vmalloc.h>
18#include <linux/uaccess.h>
19#include <asm/user.h>
20#include <asm/xsave.h>
21#include "cpuid.h"
22#include "lapic.h"
23#include "mmu.h"
24#include "trace.h"
25
26void kvm_update_cpuid(struct kvm_vcpu *vcpu)
27{
28 struct kvm_cpuid_entry2 *best;
29 struct kvm_lapic *apic = vcpu->arch.apic;
30
31 best = kvm_find_cpuid_entry(vcpu, 1, 0);
32 if (!best)
33 return;
34
35 /* Update OSXSAVE bit */
36 if (cpu_has_xsave && best->function == 0x1) {
37 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
38 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
39 best->ecx |= bit(X86_FEATURE_OSXSAVE);
40 }
41
42 if (apic) {
43 if (best->ecx & bit(X86_FEATURE_TSC_DEADLINE_TIMER))
44 apic->lapic_timer.timer_mode_mask = 3 << 17;
45 else
46 apic->lapic_timer.timer_mode_mask = 1 << 17;
47 }
48
49 kvm_pmu_cpuid_update(vcpu);
50}
51
52static int is_efer_nx(void)
53{
54 unsigned long long efer = 0;
55
56 rdmsrl_safe(MSR_EFER, &efer);
57 return efer & EFER_NX;
58}
59
60static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
61{
62 int i;
63 struct kvm_cpuid_entry2 *e, *entry;
64
65 entry = NULL;
66 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
67 e = &vcpu->arch.cpuid_entries[i];
68 if (e->function == 0x80000001) {
69 entry = e;
70 break;
71 }
72 }
73 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
74 entry->edx &= ~(1 << 20);
75 printk(KERN_INFO "kvm: guest NX capability removed\n");
76 }
77}
78
79/* when an old userspace process fills a new kernel module */
80int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
81 struct kvm_cpuid *cpuid,
82 struct kvm_cpuid_entry __user *entries)
83{
84 int r, i;
85 struct kvm_cpuid_entry *cpuid_entries;
86
87 r = -E2BIG;
88 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
89 goto out;
90 r = -ENOMEM;
91 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
92 if (!cpuid_entries)
93 goto out;
94 r = -EFAULT;
95 if (copy_from_user(cpuid_entries, entries,
96 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
97 goto out_free;
98 for (i = 0; i < cpuid->nent; i++) {
99 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
100 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
101 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
102 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
103 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
104 vcpu->arch.cpuid_entries[i].index = 0;
105 vcpu->arch.cpuid_entries[i].flags = 0;
106 vcpu->arch.cpuid_entries[i].padding[0] = 0;
107 vcpu->arch.cpuid_entries[i].padding[1] = 0;
108 vcpu->arch.cpuid_entries[i].padding[2] = 0;
109 }
110 vcpu->arch.cpuid_nent = cpuid->nent;
111 cpuid_fix_nx_cap(vcpu);
112 r = 0;
113 kvm_apic_set_version(vcpu);
114 kvm_x86_ops->cpuid_update(vcpu);
115 kvm_update_cpuid(vcpu);
116
117out_free:
118 vfree(cpuid_entries);
119out:
120 return r;
121}
122
123int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
124 struct kvm_cpuid2 *cpuid,
125 struct kvm_cpuid_entry2 __user *entries)
126{
127 int r;
128
129 r = -E2BIG;
130 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
131 goto out;
132 r = -EFAULT;
133 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
134 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
135 goto out;
136 vcpu->arch.cpuid_nent = cpuid->nent;
137 kvm_apic_set_version(vcpu);
138 kvm_x86_ops->cpuid_update(vcpu);
139 kvm_update_cpuid(vcpu);
140 return 0;
141
142out:
143 return r;
144}
145
146int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
147 struct kvm_cpuid2 *cpuid,
148 struct kvm_cpuid_entry2 __user *entries)
149{
150 int r;
151
152 r = -E2BIG;
153 if (cpuid->nent < vcpu->arch.cpuid_nent)
154 goto out;
155 r = -EFAULT;
156 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
157 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
158 goto out;
159 return 0;
160
161out:
162 cpuid->nent = vcpu->arch.cpuid_nent;
163 return r;
164}
165
166static void cpuid_mask(u32 *word, int wordnum)
167{
168 *word &= boot_cpu_data.x86_capability[wordnum];
169}
170
171static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
172 u32 index)
173{
174 entry->function = function;
175 entry->index = index;
176 cpuid_count(entry->function, entry->index,
177 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
178 entry->flags = 0;
179}
180
181static bool supported_xcr0_bit(unsigned bit)
182{
183 u64 mask = ((u64)1 << bit);
184
185 return mask & (XSTATE_FP | XSTATE_SSE | XSTATE_YMM) & host_xcr0;
186}
187
188#define F(x) bit(X86_FEATURE_##x)
189
190static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
191 u32 index, int *nent, int maxnent)
192{
193 int r;
194 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
195#ifdef CONFIG_X86_64
196 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
197 ? F(GBPAGES) : 0;
198 unsigned f_lm = F(LM);
199#else
200 unsigned f_gbpages = 0;
201 unsigned f_lm = 0;
202#endif
203 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
204
205 /* cpuid 1.edx */
206 const u32 kvm_supported_word0_x86_features =
207 F(FPU) | F(VME) | F(DE) | F(PSE) |
208 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
209 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
210 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
211 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
212 0 /* Reserved, DS, ACPI */ | F(MMX) |
213 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
214 0 /* HTT, TM, Reserved, PBE */;
215 /* cpuid 0x80000001.edx */
216 const u32 kvm_supported_word1_x86_features =
217 F(FPU) | F(VME) | F(DE) | F(PSE) |
218 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
219 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
220 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
221 F(PAT) | F(PSE36) | 0 /* Reserved */ |
222 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
223 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
224 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
225 /* cpuid 1.ecx */
226 const u32 kvm_supported_word4_x86_features =
227 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
228 0 /* DS-CPL, VMX, SMX, EST */ |
229 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
230 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
231 0 /* Reserved, DCA */ | F(XMM4_1) |
232 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
233 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
234 F(F16C) | F(RDRAND);
235 /* cpuid 0x80000001.ecx */
236 const u32 kvm_supported_word6_x86_features =
237 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
238 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
239 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
240 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
241
242 /* cpuid 0xC0000001.edx */
243 const u32 kvm_supported_word5_x86_features =
244 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
245 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
246 F(PMM) | F(PMM_EN);
247
248 /* cpuid 7.0.ebx */
249 const u32 kvm_supported_word9_x86_features =
250 F(FSGSBASE) | F(BMI1) | F(AVX2) | F(SMEP) | F(BMI2) | F(ERMS);
251
252 /* all calls to cpuid_count() should be made on the same cpu */
253 get_cpu();
254
255 r = -E2BIG;
256
257 if (*nent >= maxnent)
258 goto out;
259
260 do_cpuid_1_ent(entry, function, index);
261 ++*nent;
262
263 switch (function) {
264 case 0:
265 entry->eax = min(entry->eax, (u32)0xd);
266 break;
267 case 1:
268 entry->edx &= kvm_supported_word0_x86_features;
269 cpuid_mask(&entry->edx, 0);
270 entry->ecx &= kvm_supported_word4_x86_features;
271 cpuid_mask(&entry->ecx, 4);
272 /* we support x2apic emulation even if host does not support
273 * it since we emulate x2apic in software */
274 entry->ecx |= F(X2APIC);
275 break;
276 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
277 * may return different values. This forces us to get_cpu() before
278 * issuing the first command, and also to emulate this annoying behavior
279 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
280 case 2: {
281 int t, times = entry->eax & 0xff;
282
283 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
284 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
285 for (t = 1; t < times; ++t) {
286 if (*nent >= maxnent)
287 goto out;
288
289 do_cpuid_1_ent(&entry[t], function, 0);
290 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
291 ++*nent;
292 }
293 break;
294 }
295 /* function 4 has additional index. */
296 case 4: {
297 int i, cache_type;
298
299 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
300 /* read more entries until cache_type is zero */
301 for (i = 1; ; ++i) {
302 if (*nent >= maxnent)
303 goto out;
304
305 cache_type = entry[i - 1].eax & 0x1f;
306 if (!cache_type)
307 break;
308 do_cpuid_1_ent(&entry[i], function, i);
309 entry[i].flags |=
310 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
311 ++*nent;
312 }
313 break;
314 }
315 case 7: {
316 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
317 /* Mask ebx against host capbability word 9 */
318 if (index == 0) {
319 entry->ebx &= kvm_supported_word9_x86_features;
320 cpuid_mask(&entry->ebx, 9);
321 } else
322 entry->ebx = 0;
323 entry->eax = 0;
324 entry->ecx = 0;
325 entry->edx = 0;
326 break;
327 }
328 case 9:
329 break;
330 case 0xa: { /* Architectural Performance Monitoring */
331 struct x86_pmu_capability cap;
332 union cpuid10_eax eax;
333 union cpuid10_edx edx;
334
335 perf_get_x86_pmu_capability(&cap);
336
337 /*
338 * Only support guest architectural pmu on a host
339 * with architectural pmu.
340 */
341 if (!cap.version)
342 memset(&cap, 0, sizeof(cap));
343
344 eax.split.version_id = min(cap.version, 2);
345 eax.split.num_counters = cap.num_counters_gp;
346 eax.split.bit_width = cap.bit_width_gp;
347 eax.split.mask_length = cap.events_mask_len;
348
349 edx.split.num_counters_fixed = cap.num_counters_fixed;
350 edx.split.bit_width_fixed = cap.bit_width_fixed;
351 edx.split.reserved = 0;
352
353 entry->eax = eax.full;
354 entry->ebx = cap.events_mask;
355 entry->ecx = 0;
356 entry->edx = edx.full;
357 break;
358 }
359 /* function 0xb has additional index. */
360 case 0xb: {
361 int i, level_type;
362
363 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
364 /* read more entries until level_type is zero */
365 for (i = 1; ; ++i) {
366 if (*nent >= maxnent)
367 goto out;
368
369 level_type = entry[i - 1].ecx & 0xff00;
370 if (!level_type)
371 break;
372 do_cpuid_1_ent(&entry[i], function, i);
373 entry[i].flags |=
374 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
375 ++*nent;
376 }
377 break;
378 }
379 case 0xd: {
380 int idx, i;
381
382 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
383 for (idx = 1, i = 1; idx < 64; ++idx) {
384 if (*nent >= maxnent)
385 goto out;
386
387 do_cpuid_1_ent(&entry[i], function, idx);
388 if (entry[i].eax == 0 || !supported_xcr0_bit(idx))
389 continue;
390 entry[i].flags |=
391 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
392 ++*nent;
393 ++i;
394 }
395 break;
396 }
397 case KVM_CPUID_SIGNATURE: {
398 char signature[12] = "KVMKVMKVM\0\0";
399 u32 *sigptr = (u32 *)signature;
400 entry->eax = 0;
401 entry->ebx = sigptr[0];
402 entry->ecx = sigptr[1];
403 entry->edx = sigptr[2];
404 break;
405 }
406 case KVM_CPUID_FEATURES:
407 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
408 (1 << KVM_FEATURE_NOP_IO_DELAY) |
409 (1 << KVM_FEATURE_CLOCKSOURCE2) |
410 (1 << KVM_FEATURE_ASYNC_PF) |
411 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
412
413 if (sched_info_on())
414 entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
415
416 entry->ebx = 0;
417 entry->ecx = 0;
418 entry->edx = 0;
419 break;
420 case 0x80000000:
421 entry->eax = min(entry->eax, 0x8000001a);
422 break;
423 case 0x80000001:
424 entry->edx &= kvm_supported_word1_x86_features;
425 cpuid_mask(&entry->edx, 1);
426 entry->ecx &= kvm_supported_word6_x86_features;
427 cpuid_mask(&entry->ecx, 6);
428 break;
429 case 0x80000008: {
430 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
431 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
432 unsigned phys_as = entry->eax & 0xff;
433
434 if (!g_phys_as)
435 g_phys_as = phys_as;
436 entry->eax = g_phys_as | (virt_as << 8);
437 entry->ebx = entry->edx = 0;
438 break;
439 }
440 case 0x80000019:
441 entry->ecx = entry->edx = 0;
442 break;
443 case 0x8000001a:
444 break;
445 case 0x8000001d:
446 break;
447 /*Add support for Centaur's CPUID instruction*/
448 case 0xC0000000:
449 /*Just support up to 0xC0000004 now*/
450 entry->eax = min(entry->eax, 0xC0000004);
451 break;
452 case 0xC0000001:
453 entry->edx &= kvm_supported_word5_x86_features;
454 cpuid_mask(&entry->edx, 5);
455 break;
456 case 3: /* Processor serial number */
457 case 5: /* MONITOR/MWAIT */
458 case 6: /* Thermal management */
459 case 0x80000007: /* Advanced power management */
460 case 0xC0000002:
461 case 0xC0000003:
462 case 0xC0000004:
463 default:
464 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
465 break;
466 }
467
468 kvm_x86_ops->set_supported_cpuid(function, entry);
469
470 r = 0;
471
472out:
473 put_cpu();
474
475 return r;
476}
477
478#undef F
479
480struct kvm_cpuid_param {
481 u32 func;
482 u32 idx;
483 bool has_leaf_count;
484 bool (*qualifier)(struct kvm_cpuid_param *param);
485};
486
487static bool is_centaur_cpu(struct kvm_cpuid_param *param)
488{
489 return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
490}
491
492int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
493 struct kvm_cpuid_entry2 __user *entries)
494{
495 struct kvm_cpuid_entry2 *cpuid_entries;
496 int limit, nent = 0, r = -E2BIG, i;
497 u32 func;
498 static struct kvm_cpuid_param param[] = {
499 { .func = 0, .has_leaf_count = true },
500 { .func = 0x80000000, .has_leaf_count = true },
501 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
502 { .func = KVM_CPUID_SIGNATURE },
503 { .func = KVM_CPUID_FEATURES },
504 };
505
506 if (cpuid->nent < 1)
507 goto out;
508 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
509 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
510 r = -ENOMEM;
511 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
512 if (!cpuid_entries)
513 goto out;
514
515 r = 0;
516 for (i = 0; i < ARRAY_SIZE(param); i++) {
517 struct kvm_cpuid_param *ent = &param[i];
518
519 if (ent->qualifier && !ent->qualifier(ent))
520 continue;
521
522 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
523 &nent, cpuid->nent);
524
525 if (r)
526 goto out_free;
527
528 if (!ent->has_leaf_count)
529 continue;
530
531 limit = cpuid_entries[nent - 1].eax;
532 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
533 r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
534 &nent, cpuid->nent);
535
536 if (r)
537 goto out_free;
538 }
539
540 r = -EFAULT;
541 if (copy_to_user(entries, cpuid_entries,
542 nent * sizeof(struct kvm_cpuid_entry2)))
543 goto out_free;
544 cpuid->nent = nent;
545 r = 0;
546
547out_free:
548 vfree(cpuid_entries);
549out:
550 return r;
551}
552
553static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
554{
555 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
556 int j, nent = vcpu->arch.cpuid_nent;
557
558 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
559 /* when no next entry is found, the current entry[i] is reselected */
560 for (j = i + 1; ; j = (j + 1) % nent) {
561 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
562 if (ej->function == e->function) {
563 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
564 return j;
565 }
566 }
567 return 0; /* silence gcc, even though control never reaches here */
568}
569
570/* find an entry with matching function, matching index (if needed), and that
571 * should be read next (if it's stateful) */
572static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
573 u32 function, u32 index)
574{
575 if (e->function != function)
576 return 0;
577 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
578 return 0;
579 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
580 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
581 return 0;
582 return 1;
583}
584
585struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
586 u32 function, u32 index)
587{
588 int i;
589 struct kvm_cpuid_entry2 *best = NULL;
590
591 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
592 struct kvm_cpuid_entry2 *e;
593
594 e = &vcpu->arch.cpuid_entries[i];
595 if (is_matching_cpuid_entry(e, function, index)) {
596 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
597 move_to_next_stateful_cpuid_entry(vcpu, i);
598 best = e;
599 break;
600 }
601 }
602 return best;
603}
604EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
605
606int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
607{
608 struct kvm_cpuid_entry2 *best;
609
610 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
611 if (!best || best->eax < 0x80000008)
612 goto not_found;
613 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
614 if (best)
615 return best->eax & 0xff;
616not_found:
617 return 36;
618}
619
620/*
621 * If no match is found, check whether we exceed the vCPU's limit
622 * and return the content of the highest valid _standard_ leaf instead.
623 * This is to satisfy the CPUID specification.
624 */
625static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
626 u32 function, u32 index)
627{
628 struct kvm_cpuid_entry2 *maxlevel;
629
630 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
631 if (!maxlevel || maxlevel->eax >= function)
632 return NULL;
633 if (function & 0x80000000) {
634 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
635 if (!maxlevel)
636 return NULL;
637 }
638 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
639}
640
641void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
642{
643 u32 function, index;
644 struct kvm_cpuid_entry2 *best;
645
646 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
647 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
648 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
649 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
650 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
651 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
652 best = kvm_find_cpuid_entry(vcpu, function, index);
653
654 if (!best)
655 best = check_cpuid_limit(vcpu, function, index);
656
657 if (best) {
658 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
659 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
660 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
661 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
662 }
663 kvm_x86_ops->skip_emulated_instruction(vcpu);
664 trace_kvm_cpuid(function,
665 kvm_register_read(vcpu, VCPU_REGS_RAX),
666 kvm_register_read(vcpu, VCPU_REGS_RBX),
667 kvm_register_read(vcpu, VCPU_REGS_RCX),
668 kvm_register_read(vcpu, VCPU_REGS_RDX));
669}
670EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
diff --git a/arch/x86/kvm/cpuid.h b/arch/x86/kvm/cpuid.h
new file mode 100644
index 000000000000..5b97e1797a6d
--- /dev/null
+++ b/arch/x86/kvm/cpuid.h
@@ -0,0 +1,46 @@
1#ifndef ARCH_X86_KVM_CPUID_H
2#define ARCH_X86_KVM_CPUID_H
3
4#include "x86.h"
5
6void kvm_update_cpuid(struct kvm_vcpu *vcpu);
7struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
8 u32 function, u32 index);
9int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
10 struct kvm_cpuid_entry2 __user *entries);
11int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
12 struct kvm_cpuid *cpuid,
13 struct kvm_cpuid_entry __user *entries);
14int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
15 struct kvm_cpuid2 *cpuid,
16 struct kvm_cpuid_entry2 __user *entries);
17int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
18 struct kvm_cpuid2 *cpuid,
19 struct kvm_cpuid_entry2 __user *entries);
20
21
22static inline bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
23{
24 struct kvm_cpuid_entry2 *best;
25
26 best = kvm_find_cpuid_entry(vcpu, 1, 0);
27 return best && (best->ecx & bit(X86_FEATURE_XSAVE));
28}
29
30static inline bool guest_cpuid_has_smep(struct kvm_vcpu *vcpu)
31{
32 struct kvm_cpuid_entry2 *best;
33
34 best = kvm_find_cpuid_entry(vcpu, 7, 0);
35 return best && (best->ebx & bit(X86_FEATURE_SMEP));
36}
37
38static inline bool guest_cpuid_has_fsgsbase(struct kvm_vcpu *vcpu)
39{
40 struct kvm_cpuid_entry2 *best;
41
42 best = kvm_find_cpuid_entry(vcpu, 7, 0);
43 return best && (best->ebx & bit(X86_FEATURE_FSGSBASE));
44}
45
46#endif
diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c
index f1e3be18a08f..05a562b85025 100644
--- a/arch/x86/kvm/emulate.c
+++ b/arch/x86/kvm/emulate.c
@@ -125,8 +125,9 @@
125#define Lock (1<<26) /* lock prefix is allowed for the instruction */ 125#define Lock (1<<26) /* lock prefix is allowed for the instruction */
126#define Priv (1<<27) /* instruction generates #GP if current CPL != 0 */ 126#define Priv (1<<27) /* instruction generates #GP if current CPL != 0 */
127#define No64 (1<<28) 127#define No64 (1<<28)
128#define PageTable (1 << 29) /* instruction used to write page table */
128/* Source 2 operand type */ 129/* Source 2 operand type */
129#define Src2Shift (29) 130#define Src2Shift (30)
130#define Src2None (OpNone << Src2Shift) 131#define Src2None (OpNone << Src2Shift)
131#define Src2CL (OpCL << Src2Shift) 132#define Src2CL (OpCL << Src2Shift)
132#define Src2ImmByte (OpImmByte << Src2Shift) 133#define Src2ImmByte (OpImmByte << Src2Shift)
@@ -1674,11 +1675,6 @@ static int em_jmp_far(struct x86_emulate_ctxt *ctxt)
1674 return X86EMUL_CONTINUE; 1675 return X86EMUL_CONTINUE;
1675} 1676}
1676 1677
1677static int em_grp1a(struct x86_emulate_ctxt *ctxt)
1678{
1679 return emulate_pop(ctxt, &ctxt->dst.val, ctxt->dst.bytes);
1680}
1681
1682static int em_grp2(struct x86_emulate_ctxt *ctxt) 1678static int em_grp2(struct x86_emulate_ctxt *ctxt)
1683{ 1679{
1684 switch (ctxt->modrm_reg) { 1680 switch (ctxt->modrm_reg) {
@@ -1788,7 +1784,7 @@ static int em_grp45(struct x86_emulate_ctxt *ctxt)
1788 return rc; 1784 return rc;
1789} 1785}
1790 1786
1791static int em_grp9(struct x86_emulate_ctxt *ctxt) 1787static int em_cmpxchg8b(struct x86_emulate_ctxt *ctxt)
1792{ 1788{
1793 u64 old = ctxt->dst.orig_val64; 1789 u64 old = ctxt->dst.orig_val64;
1794 1790
@@ -1831,6 +1827,24 @@ static int em_ret_far(struct x86_emulate_ctxt *ctxt)
1831 return rc; 1827 return rc;
1832} 1828}
1833 1829
1830static int em_cmpxchg(struct x86_emulate_ctxt *ctxt)
1831{
1832 /* Save real source value, then compare EAX against destination. */
1833 ctxt->src.orig_val = ctxt->src.val;
1834 ctxt->src.val = ctxt->regs[VCPU_REGS_RAX];
1835 emulate_2op_SrcV(ctxt, "cmp");
1836
1837 if (ctxt->eflags & EFLG_ZF) {
1838 /* Success: write back to memory. */
1839 ctxt->dst.val = ctxt->src.orig_val;
1840 } else {
1841 /* Failure: write the value we saw to EAX. */
1842 ctxt->dst.type = OP_REG;
1843 ctxt->dst.addr.reg = (unsigned long *)&ctxt->regs[VCPU_REGS_RAX];
1844 }
1845 return X86EMUL_CONTINUE;
1846}
1847
1834static int em_lseg(struct x86_emulate_ctxt *ctxt) 1848static int em_lseg(struct x86_emulate_ctxt *ctxt)
1835{ 1849{
1836 int seg = ctxt->src2.val; 1850 int seg = ctxt->src2.val;
@@ -2481,6 +2495,15 @@ static int em_das(struct x86_emulate_ctxt *ctxt)
2481 return X86EMUL_CONTINUE; 2495 return X86EMUL_CONTINUE;
2482} 2496}
2483 2497
2498static int em_call(struct x86_emulate_ctxt *ctxt)
2499{
2500 long rel = ctxt->src.val;
2501
2502 ctxt->src.val = (unsigned long)ctxt->_eip;
2503 jmp_rel(ctxt, rel);
2504 return em_push(ctxt);
2505}
2506
2484static int em_call_far(struct x86_emulate_ctxt *ctxt) 2507static int em_call_far(struct x86_emulate_ctxt *ctxt)
2485{ 2508{
2486 u16 sel, old_cs; 2509 u16 sel, old_cs;
@@ -2622,12 +2645,75 @@ static int em_rdtsc(struct x86_emulate_ctxt *ctxt)
2622 return X86EMUL_CONTINUE; 2645 return X86EMUL_CONTINUE;
2623} 2646}
2624 2647
2648static int em_rdpmc(struct x86_emulate_ctxt *ctxt)
2649{
2650 u64 pmc;
2651
2652 if (ctxt->ops->read_pmc(ctxt, ctxt->regs[VCPU_REGS_RCX], &pmc))
2653 return emulate_gp(ctxt, 0);
2654 ctxt->regs[VCPU_REGS_RAX] = (u32)pmc;
2655 ctxt->regs[VCPU_REGS_RDX] = pmc >> 32;
2656 return X86EMUL_CONTINUE;
2657}
2658
2625static int em_mov(struct x86_emulate_ctxt *ctxt) 2659static int em_mov(struct x86_emulate_ctxt *ctxt)
2626{ 2660{
2627 ctxt->dst.val = ctxt->src.val; 2661 ctxt->dst.val = ctxt->src.val;
2628 return X86EMUL_CONTINUE; 2662 return X86EMUL_CONTINUE;
2629} 2663}
2630 2664
2665static int em_cr_write(struct x86_emulate_ctxt *ctxt)
2666{
2667 if (ctxt->ops->set_cr(ctxt, ctxt->modrm_reg, ctxt->src.val))
2668 return emulate_gp(ctxt, 0);
2669
2670 /* Disable writeback. */
2671 ctxt->dst.type = OP_NONE;
2672 return X86EMUL_CONTINUE;
2673}
2674
2675static int em_dr_write(struct x86_emulate_ctxt *ctxt)
2676{
2677 unsigned long val;
2678
2679 if (ctxt->mode == X86EMUL_MODE_PROT64)
2680 val = ctxt->src.val & ~0ULL;
2681 else
2682 val = ctxt->src.val & ~0U;
2683
2684 /* #UD condition is already handled. */
2685 if (ctxt->ops->set_dr(ctxt, ctxt->modrm_reg, val) < 0)
2686 return emulate_gp(ctxt, 0);
2687
2688 /* Disable writeback. */
2689 ctxt->dst.type = OP_NONE;
2690 return X86EMUL_CONTINUE;
2691}
2692
2693static int em_wrmsr(struct x86_emulate_ctxt *ctxt)
2694{
2695 u64 msr_data;
2696
2697 msr_data = (u32)ctxt->regs[VCPU_REGS_RAX]
2698 | ((u64)ctxt->regs[VCPU_REGS_RDX] << 32);
2699 if (ctxt->ops->set_msr(ctxt, ctxt->regs[VCPU_REGS_RCX], msr_data))
2700 return emulate_gp(ctxt, 0);
2701
2702 return X86EMUL_CONTINUE;
2703}
2704
2705static int em_rdmsr(struct x86_emulate_ctxt *ctxt)
2706{
2707 u64 msr_data;
2708
2709 if (ctxt->ops->get_msr(ctxt, ctxt->regs[VCPU_REGS_RCX], &msr_data))
2710 return emulate_gp(ctxt, 0);
2711
2712 ctxt->regs[VCPU_REGS_RAX] = (u32)msr_data;
2713 ctxt->regs[VCPU_REGS_RDX] = msr_data >> 32;
2714 return X86EMUL_CONTINUE;
2715}
2716
2631static int em_mov_rm_sreg(struct x86_emulate_ctxt *ctxt) 2717static int em_mov_rm_sreg(struct x86_emulate_ctxt *ctxt)
2632{ 2718{
2633 if (ctxt->modrm_reg > VCPU_SREG_GS) 2719 if (ctxt->modrm_reg > VCPU_SREG_GS)
@@ -2775,6 +2861,24 @@ static int em_jcxz(struct x86_emulate_ctxt *ctxt)
2775 return X86EMUL_CONTINUE; 2861 return X86EMUL_CONTINUE;
2776} 2862}
2777 2863
2864static int em_in(struct x86_emulate_ctxt *ctxt)
2865{
2866 if (!pio_in_emulated(ctxt, ctxt->dst.bytes, ctxt->src.val,
2867 &ctxt->dst.val))
2868 return X86EMUL_IO_NEEDED;
2869
2870 return X86EMUL_CONTINUE;
2871}
2872
2873static int em_out(struct x86_emulate_ctxt *ctxt)
2874{
2875 ctxt->ops->pio_out_emulated(ctxt, ctxt->src.bytes, ctxt->dst.val,
2876 &ctxt->src.val, 1);
2877 /* Disable writeback. */
2878 ctxt->dst.type = OP_NONE;
2879 return X86EMUL_CONTINUE;
2880}
2881
2778static int em_cli(struct x86_emulate_ctxt *ctxt) 2882static int em_cli(struct x86_emulate_ctxt *ctxt)
2779{ 2883{
2780 if (emulator_bad_iopl(ctxt)) 2884 if (emulator_bad_iopl(ctxt))
@@ -2794,6 +2898,69 @@ static int em_sti(struct x86_emulate_ctxt *ctxt)
2794 return X86EMUL_CONTINUE; 2898 return X86EMUL_CONTINUE;
2795} 2899}
2796 2900
2901static int em_bt(struct x86_emulate_ctxt *ctxt)
2902{
2903 /* Disable writeback. */
2904 ctxt->dst.type = OP_NONE;
2905 /* only subword offset */
2906 ctxt->src.val &= (ctxt->dst.bytes << 3) - 1;
2907
2908 emulate_2op_SrcV_nobyte(ctxt, "bt");
2909 return X86EMUL_CONTINUE;
2910}
2911
2912static int em_bts(struct x86_emulate_ctxt *ctxt)
2913{
2914 emulate_2op_SrcV_nobyte(ctxt, "bts");
2915 return X86EMUL_CONTINUE;
2916}
2917
2918static int em_btr(struct x86_emulate_ctxt *ctxt)
2919{
2920 emulate_2op_SrcV_nobyte(ctxt, "btr");
2921 return X86EMUL_CONTINUE;
2922}
2923
2924static int em_btc(struct x86_emulate_ctxt *ctxt)
2925{
2926 emulate_2op_SrcV_nobyte(ctxt, "btc");
2927 return X86EMUL_CONTINUE;
2928}
2929
2930static int em_bsf(struct x86_emulate_ctxt *ctxt)
2931{
2932 u8 zf;
2933
2934 __asm__ ("bsf %2, %0; setz %1"
2935 : "=r"(ctxt->dst.val), "=q"(zf)
2936 : "r"(ctxt->src.val));
2937
2938 ctxt->eflags &= ~X86_EFLAGS_ZF;
2939 if (zf) {
2940 ctxt->eflags |= X86_EFLAGS_ZF;
2941 /* Disable writeback. */
2942 ctxt->dst.type = OP_NONE;
2943 }
2944 return X86EMUL_CONTINUE;
2945}
2946
2947static int em_bsr(struct x86_emulate_ctxt *ctxt)
2948{
2949 u8 zf;
2950
2951 __asm__ ("bsr %2, %0; setz %1"
2952 : "=r"(ctxt->dst.val), "=q"(zf)
2953 : "r"(ctxt->src.val));
2954
2955 ctxt->eflags &= ~X86_EFLAGS_ZF;
2956 if (zf) {
2957 ctxt->eflags |= X86_EFLAGS_ZF;
2958 /* Disable writeback. */
2959 ctxt->dst.type = OP_NONE;
2960 }
2961 return X86EMUL_CONTINUE;
2962}
2963
2797static bool valid_cr(int nr) 2964static bool valid_cr(int nr)
2798{ 2965{
2799 switch (nr) { 2966 switch (nr) {
@@ -2867,9 +3034,6 @@ static int check_cr_write(struct x86_emulate_ctxt *ctxt)
2867 break; 3034 break;
2868 } 3035 }
2869 case 4: { 3036 case 4: {
2870 u64 cr4;
2871
2872 cr4 = ctxt->ops->get_cr(ctxt, 4);
2873 ctxt->ops->get_msr(ctxt, MSR_EFER, &efer); 3037 ctxt->ops->get_msr(ctxt, MSR_EFER, &efer);
2874 3038
2875 if ((efer & EFER_LMA) && !(new_val & X86_CR4_PAE)) 3039 if ((efer & EFER_LMA) && !(new_val & X86_CR4_PAE))
@@ -3003,6 +3167,8 @@ static int check_perm_out(struct x86_emulate_ctxt *ctxt)
3003#define D2bv(_f) D((_f) | ByteOp), D(_f) 3167#define D2bv(_f) D((_f) | ByteOp), D(_f)
3004#define D2bvIP(_f, _i, _p) DIP((_f) | ByteOp, _i, _p), DIP(_f, _i, _p) 3168#define D2bvIP(_f, _i, _p) DIP((_f) | ByteOp, _i, _p), DIP(_f, _i, _p)
3005#define I2bv(_f, _e) I((_f) | ByteOp, _e), I(_f, _e) 3169#define I2bv(_f, _e) I((_f) | ByteOp, _e), I(_f, _e)
3170#define I2bvIP(_f, _e, _i, _p) \
3171 IIP((_f) | ByteOp, _e, _i, _p), IIP(_f, _e, _i, _p)
3006 3172
3007#define I6ALU(_f, _e) I2bv((_f) | DstMem | SrcReg | ModRM, _e), \ 3173#define I6ALU(_f, _e) I2bv((_f) | DstMem | SrcReg | ModRM, _e), \
3008 I2bv(((_f) | DstReg | SrcMem | ModRM) & ~Lock, _e), \ 3174 I2bv(((_f) | DstReg | SrcMem | ModRM) & ~Lock, _e), \
@@ -3033,17 +3199,17 @@ static struct opcode group7_rm7[] = {
3033 3199
3034static struct opcode group1[] = { 3200static struct opcode group1[] = {
3035 I(Lock, em_add), 3201 I(Lock, em_add),
3036 I(Lock, em_or), 3202 I(Lock | PageTable, em_or),
3037 I(Lock, em_adc), 3203 I(Lock, em_adc),
3038 I(Lock, em_sbb), 3204 I(Lock, em_sbb),
3039 I(Lock, em_and), 3205 I(Lock | PageTable, em_and),
3040 I(Lock, em_sub), 3206 I(Lock, em_sub),
3041 I(Lock, em_xor), 3207 I(Lock, em_xor),
3042 I(0, em_cmp), 3208 I(0, em_cmp),
3043}; 3209};
3044 3210
3045static struct opcode group1A[] = { 3211static struct opcode group1A[] = {
3046 D(DstMem | SrcNone | ModRM | Mov | Stack), N, N, N, N, N, N, N, 3212 I(DstMem | SrcNone | ModRM | Mov | Stack, em_pop), N, N, N, N, N, N, N,
3047}; 3213};
3048 3214
3049static struct opcode group3[] = { 3215static struct opcode group3[] = {
@@ -3058,16 +3224,19 @@ static struct opcode group3[] = {
3058}; 3224};
3059 3225
3060static struct opcode group4[] = { 3226static struct opcode group4[] = {
3061 D(ByteOp | DstMem | SrcNone | ModRM | Lock), D(ByteOp | DstMem | SrcNone | ModRM | Lock), 3227 I(ByteOp | DstMem | SrcNone | ModRM | Lock, em_grp45),
3228 I(ByteOp | DstMem | SrcNone | ModRM | Lock, em_grp45),
3062 N, N, N, N, N, N, 3229 N, N, N, N, N, N,
3063}; 3230};
3064 3231
3065static struct opcode group5[] = { 3232static struct opcode group5[] = {
3066 D(DstMem | SrcNone | ModRM | Lock), D(DstMem | SrcNone | ModRM | Lock), 3233 I(DstMem | SrcNone | ModRM | Lock, em_grp45),
3067 D(SrcMem | ModRM | Stack), 3234 I(DstMem | SrcNone | ModRM | Lock, em_grp45),
3235 I(SrcMem | ModRM | Stack, em_grp45),
3068 I(SrcMemFAddr | ModRM | ImplicitOps | Stack, em_call_far), 3236 I(SrcMemFAddr | ModRM | ImplicitOps | Stack, em_call_far),
3069 D(SrcMem | ModRM | Stack), D(SrcMemFAddr | ModRM | ImplicitOps), 3237 I(SrcMem | ModRM | Stack, em_grp45),
3070 D(SrcMem | ModRM | Stack), N, 3238 I(SrcMemFAddr | ModRM | ImplicitOps, em_grp45),
3239 I(SrcMem | ModRM | Stack, em_grp45), N,
3071}; 3240};
3072 3241
3073static struct opcode group6[] = { 3242static struct opcode group6[] = {
@@ -3096,18 +3265,21 @@ static struct group_dual group7 = { {
3096 3265
3097static struct opcode group8[] = { 3266static struct opcode group8[] = {
3098 N, N, N, N, 3267 N, N, N, N,
3099 D(DstMem | SrcImmByte | ModRM), D(DstMem | SrcImmByte | ModRM | Lock), 3268 I(DstMem | SrcImmByte | ModRM, em_bt),
3100 D(DstMem | SrcImmByte | ModRM | Lock), D(DstMem | SrcImmByte | ModRM | Lock), 3269 I(DstMem | SrcImmByte | ModRM | Lock | PageTable, em_bts),
3270 I(DstMem | SrcImmByte | ModRM | Lock, em_btr),
3271 I(DstMem | SrcImmByte | ModRM | Lock | PageTable, em_btc),
3101}; 3272};
3102 3273
3103static struct group_dual group9 = { { 3274static struct group_dual group9 = { {
3104 N, D(DstMem64 | ModRM | Lock), N, N, N, N, N, N, 3275 N, I(DstMem64 | ModRM | Lock | PageTable, em_cmpxchg8b), N, N, N, N, N, N,
3105}, { 3276}, {
3106 N, N, N, N, N, N, N, N, 3277 N, N, N, N, N, N, N, N,
3107} }; 3278} };
3108 3279
3109static struct opcode group11[] = { 3280static struct opcode group11[] = {
3110 I(DstMem | SrcImm | ModRM | Mov, em_mov), X7(D(Undefined)), 3281 I(DstMem | SrcImm | ModRM | Mov | PageTable, em_mov),
3282 X7(D(Undefined)),
3111}; 3283};
3112 3284
3113static struct gprefix pfx_0f_6f_0f_7f = { 3285static struct gprefix pfx_0f_6f_0f_7f = {
@@ -3120,7 +3292,7 @@ static struct opcode opcode_table[256] = {
3120 I(ImplicitOps | Stack | No64 | Src2ES, em_push_sreg), 3292 I(ImplicitOps | Stack | No64 | Src2ES, em_push_sreg),
3121 I(ImplicitOps | Stack | No64 | Src2ES, em_pop_sreg), 3293 I(ImplicitOps | Stack | No64 | Src2ES, em_pop_sreg),
3122 /* 0x08 - 0x0F */ 3294 /* 0x08 - 0x0F */
3123 I6ALU(Lock, em_or), 3295 I6ALU(Lock | PageTable, em_or),
3124 I(ImplicitOps | Stack | No64 | Src2CS, em_push_sreg), 3296 I(ImplicitOps | Stack | No64 | Src2CS, em_push_sreg),
3125 N, 3297 N,
3126 /* 0x10 - 0x17 */ 3298 /* 0x10 - 0x17 */
@@ -3132,7 +3304,7 @@ static struct opcode opcode_table[256] = {
3132 I(ImplicitOps | Stack | No64 | Src2DS, em_push_sreg), 3304 I(ImplicitOps | Stack | No64 | Src2DS, em_push_sreg),
3133 I(ImplicitOps | Stack | No64 | Src2DS, em_pop_sreg), 3305 I(ImplicitOps | Stack | No64 | Src2DS, em_pop_sreg),
3134 /* 0x20 - 0x27 */ 3306 /* 0x20 - 0x27 */
3135 I6ALU(Lock, em_and), N, N, 3307 I6ALU(Lock | PageTable, em_and), N, N,
3136 /* 0x28 - 0x2F */ 3308 /* 0x28 - 0x2F */
3137 I6ALU(Lock, em_sub), N, I(ByteOp | DstAcc | No64, em_das), 3309 I6ALU(Lock, em_sub), N, I(ByteOp | DstAcc | No64, em_das),
3138 /* 0x30 - 0x37 */ 3310 /* 0x30 - 0x37 */
@@ -3155,8 +3327,8 @@ static struct opcode opcode_table[256] = {
3155 I(DstReg | SrcMem | ModRM | Src2Imm, em_imul_3op), 3327 I(DstReg | SrcMem | ModRM | Src2Imm, em_imul_3op),
3156 I(SrcImmByte | Mov | Stack, em_push), 3328 I(SrcImmByte | Mov | Stack, em_push),
3157 I(DstReg | SrcMem | ModRM | Src2ImmByte, em_imul_3op), 3329 I(DstReg | SrcMem | ModRM | Src2ImmByte, em_imul_3op),
3158 D2bvIP(DstDI | SrcDX | Mov | String, ins, check_perm_in), /* insb, insw/insd */ 3330 I2bvIP(DstDI | SrcDX | Mov | String, em_in, ins, check_perm_in), /* insb, insw/insd */
3159 D2bvIP(SrcSI | DstDX | String, outs, check_perm_out), /* outsb, outsw/outsd */ 3331 I2bvIP(SrcSI | DstDX | String, em_out, outs, check_perm_out), /* outsb, outsw/outsd */
3160 /* 0x70 - 0x7F */ 3332 /* 0x70 - 0x7F */
3161 X16(D(SrcImmByte)), 3333 X16(D(SrcImmByte)),
3162 /* 0x80 - 0x87 */ 3334 /* 0x80 - 0x87 */
@@ -3165,11 +3337,11 @@ static struct opcode opcode_table[256] = {
3165 G(ByteOp | DstMem | SrcImm | ModRM | No64 | Group, group1), 3337 G(ByteOp | DstMem | SrcImm | ModRM | No64 | Group, group1),
3166 G(DstMem | SrcImmByte | ModRM | Group, group1), 3338 G(DstMem | SrcImmByte | ModRM | Group, group1),
3167 I2bv(DstMem | SrcReg | ModRM, em_test), 3339 I2bv(DstMem | SrcReg | ModRM, em_test),
3168 I2bv(DstMem | SrcReg | ModRM | Lock, em_xchg), 3340 I2bv(DstMem | SrcReg | ModRM | Lock | PageTable, em_xchg),
3169 /* 0x88 - 0x8F */ 3341 /* 0x88 - 0x8F */
3170 I2bv(DstMem | SrcReg | ModRM | Mov, em_mov), 3342 I2bv(DstMem | SrcReg | ModRM | Mov | PageTable, em_mov),
3171 I2bv(DstReg | SrcMem | ModRM | Mov, em_mov), 3343 I2bv(DstReg | SrcMem | ModRM | Mov, em_mov),
3172 I(DstMem | SrcNone | ModRM | Mov, em_mov_rm_sreg), 3344 I(DstMem | SrcNone | ModRM | Mov | PageTable, em_mov_rm_sreg),
3173 D(ModRM | SrcMem | NoAccess | DstReg), 3345 D(ModRM | SrcMem | NoAccess | DstReg),
3174 I(ImplicitOps | SrcMem16 | ModRM, em_mov_sreg_rm), 3346 I(ImplicitOps | SrcMem16 | ModRM, em_mov_sreg_rm),
3175 G(0, group1A), 3347 G(0, group1A),
@@ -3182,7 +3354,7 @@ static struct opcode opcode_table[256] = {
3182 II(ImplicitOps | Stack, em_popf, popf), N, N, 3354 II(ImplicitOps | Stack, em_popf, popf), N, N,
3183 /* 0xA0 - 0xA7 */ 3355 /* 0xA0 - 0xA7 */
3184 I2bv(DstAcc | SrcMem | Mov | MemAbs, em_mov), 3356 I2bv(DstAcc | SrcMem | Mov | MemAbs, em_mov),
3185 I2bv(DstMem | SrcAcc | Mov | MemAbs, em_mov), 3357 I2bv(DstMem | SrcAcc | Mov | MemAbs | PageTable, em_mov),
3186 I2bv(SrcSI | DstDI | Mov | String, em_mov), 3358 I2bv(SrcSI | DstDI | Mov | String, em_mov),
3187 I2bv(SrcSI | DstDI | String, em_cmp), 3359 I2bv(SrcSI | DstDI | String, em_cmp),
3188 /* 0xA8 - 0xAF */ 3360 /* 0xA8 - 0xAF */
@@ -3213,13 +3385,13 @@ static struct opcode opcode_table[256] = {
3213 /* 0xE0 - 0xE7 */ 3385 /* 0xE0 - 0xE7 */
3214 X3(I(SrcImmByte, em_loop)), 3386 X3(I(SrcImmByte, em_loop)),
3215 I(SrcImmByte, em_jcxz), 3387 I(SrcImmByte, em_jcxz),
3216 D2bvIP(SrcImmUByte | DstAcc, in, check_perm_in), 3388 I2bvIP(SrcImmUByte | DstAcc, em_in, in, check_perm_in),
3217 D2bvIP(SrcAcc | DstImmUByte, out, check_perm_out), 3389 I2bvIP(SrcAcc | DstImmUByte, em_out, out, check_perm_out),
3218 /* 0xE8 - 0xEF */ 3390 /* 0xE8 - 0xEF */
3219 D(SrcImm | Stack), D(SrcImm | ImplicitOps), 3391 I(SrcImm | Stack, em_call), D(SrcImm | ImplicitOps),
3220 I(SrcImmFAddr | No64, em_jmp_far), D(SrcImmByte | ImplicitOps), 3392 I(SrcImmFAddr | No64, em_jmp_far), D(SrcImmByte | ImplicitOps),
3221 D2bvIP(SrcDX | DstAcc, in, check_perm_in), 3393 I2bvIP(SrcDX | DstAcc, em_in, in, check_perm_in),
3222 D2bvIP(SrcAcc | DstDX, out, check_perm_out), 3394 I2bvIP(SrcAcc | DstDX, em_out, out, check_perm_out),
3223 /* 0xF0 - 0xF7 */ 3395 /* 0xF0 - 0xF7 */
3224 N, DI(ImplicitOps, icebp), N, N, 3396 N, DI(ImplicitOps, icebp), N, N,
3225 DI(ImplicitOps | Priv, hlt), D(ImplicitOps), 3397 DI(ImplicitOps | Priv, hlt), D(ImplicitOps),
@@ -3242,15 +3414,15 @@ static struct opcode twobyte_table[256] = {
3242 /* 0x20 - 0x2F */ 3414 /* 0x20 - 0x2F */
3243 DIP(ModRM | DstMem | Priv | Op3264, cr_read, check_cr_read), 3415 DIP(ModRM | DstMem | Priv | Op3264, cr_read, check_cr_read),
3244 DIP(ModRM | DstMem | Priv | Op3264, dr_read, check_dr_read), 3416 DIP(ModRM | DstMem | Priv | Op3264, dr_read, check_dr_read),
3245 DIP(ModRM | SrcMem | Priv | Op3264, cr_write, check_cr_write), 3417 IIP(ModRM | SrcMem | Priv | Op3264, em_cr_write, cr_write, check_cr_write),
3246 DIP(ModRM | SrcMem | Priv | Op3264, dr_write, check_dr_write), 3418 IIP(ModRM | SrcMem | Priv | Op3264, em_dr_write, dr_write, check_dr_write),
3247 N, N, N, N, 3419 N, N, N, N,
3248 N, N, N, N, N, N, N, N, 3420 N, N, N, N, N, N, N, N,
3249 /* 0x30 - 0x3F */ 3421 /* 0x30 - 0x3F */
3250 DI(ImplicitOps | Priv, wrmsr), 3422 II(ImplicitOps | Priv, em_wrmsr, wrmsr),
3251 IIP(ImplicitOps, em_rdtsc, rdtsc, check_rdtsc), 3423 IIP(ImplicitOps, em_rdtsc, rdtsc, check_rdtsc),
3252 DI(ImplicitOps | Priv, rdmsr), 3424 II(ImplicitOps | Priv, em_rdmsr, rdmsr),
3253 DIP(ImplicitOps | Priv, rdpmc, check_rdpmc), 3425 IIP(ImplicitOps, em_rdpmc, rdpmc, check_rdpmc),
3254 I(ImplicitOps | VendorSpecific, em_sysenter), 3426 I(ImplicitOps | VendorSpecific, em_sysenter),
3255 I(ImplicitOps | Priv | VendorSpecific, em_sysexit), 3427 I(ImplicitOps | Priv | VendorSpecific, em_sysexit),
3256 N, N, 3428 N, N,
@@ -3275,26 +3447,28 @@ static struct opcode twobyte_table[256] = {
3275 X16(D(ByteOp | DstMem | SrcNone | ModRM| Mov)), 3447 X16(D(ByteOp | DstMem | SrcNone | ModRM| Mov)),
3276 /* 0xA0 - 0xA7 */ 3448 /* 0xA0 - 0xA7 */
3277 I(Stack | Src2FS, em_push_sreg), I(Stack | Src2FS, em_pop_sreg), 3449 I(Stack | Src2FS, em_push_sreg), I(Stack | Src2FS, em_pop_sreg),
3278 DI(ImplicitOps, cpuid), D(DstMem | SrcReg | ModRM | BitOp), 3450 DI(ImplicitOps, cpuid), I(DstMem | SrcReg | ModRM | BitOp, em_bt),
3279 D(DstMem | SrcReg | Src2ImmByte | ModRM), 3451 D(DstMem | SrcReg | Src2ImmByte | ModRM),
3280 D(DstMem | SrcReg | Src2CL | ModRM), N, N, 3452 D(DstMem | SrcReg | Src2CL | ModRM), N, N,
3281 /* 0xA8 - 0xAF */ 3453 /* 0xA8 - 0xAF */
3282 I(Stack | Src2GS, em_push_sreg), I(Stack | Src2GS, em_pop_sreg), 3454 I(Stack | Src2GS, em_push_sreg), I(Stack | Src2GS, em_pop_sreg),
3283 DI(ImplicitOps, rsm), D(DstMem | SrcReg | ModRM | BitOp | Lock), 3455 DI(ImplicitOps, rsm),
3456 I(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_bts),
3284 D(DstMem | SrcReg | Src2ImmByte | ModRM), 3457 D(DstMem | SrcReg | Src2ImmByte | ModRM),
3285 D(DstMem | SrcReg | Src2CL | ModRM), 3458 D(DstMem | SrcReg | Src2CL | ModRM),
3286 D(ModRM), I(DstReg | SrcMem | ModRM, em_imul), 3459 D(ModRM), I(DstReg | SrcMem | ModRM, em_imul),
3287 /* 0xB0 - 0xB7 */ 3460 /* 0xB0 - 0xB7 */
3288 D2bv(DstMem | SrcReg | ModRM | Lock), 3461 I2bv(DstMem | SrcReg | ModRM | Lock | PageTable, em_cmpxchg),
3289 I(DstReg | SrcMemFAddr | ModRM | Src2SS, em_lseg), 3462 I(DstReg | SrcMemFAddr | ModRM | Src2SS, em_lseg),
3290 D(DstMem | SrcReg | ModRM | BitOp | Lock), 3463 I(DstMem | SrcReg | ModRM | BitOp | Lock, em_btr),
3291 I(DstReg | SrcMemFAddr | ModRM | Src2FS, em_lseg), 3464 I(DstReg | SrcMemFAddr | ModRM | Src2FS, em_lseg),
3292 I(DstReg | SrcMemFAddr | ModRM | Src2GS, em_lseg), 3465 I(DstReg | SrcMemFAddr | ModRM | Src2GS, em_lseg),
3293 D(ByteOp | DstReg | SrcMem | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov), 3466 D(ByteOp | DstReg | SrcMem | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
3294 /* 0xB8 - 0xBF */ 3467 /* 0xB8 - 0xBF */
3295 N, N, 3468 N, N,
3296 G(BitOp, group8), D(DstMem | SrcReg | ModRM | BitOp | Lock), 3469 G(BitOp, group8),
3297 D(DstReg | SrcMem | ModRM), D(DstReg | SrcMem | ModRM), 3470 I(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_btc),
3471 I(DstReg | SrcMem | ModRM, em_bsf), I(DstReg | SrcMem | ModRM, em_bsr),
3298 D(ByteOp | DstReg | SrcMem | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov), 3472 D(ByteOp | DstReg | SrcMem | ModRM | Mov), D(DstReg | SrcMem16 | ModRM | Mov),
3299 /* 0xC0 - 0xCF */ 3473 /* 0xC0 - 0xCF */
3300 D2bv(DstMem | SrcReg | ModRM | Lock), 3474 D2bv(DstMem | SrcReg | ModRM | Lock),
@@ -3320,6 +3494,7 @@ static struct opcode twobyte_table[256] = {
3320#undef D2bv 3494#undef D2bv
3321#undef D2bvIP 3495#undef D2bvIP
3322#undef I2bv 3496#undef I2bv
3497#undef I2bvIP
3323#undef I6ALU 3498#undef I6ALU
3324 3499
3325static unsigned imm_size(struct x86_emulate_ctxt *ctxt) 3500static unsigned imm_size(struct x86_emulate_ctxt *ctxt)
@@ -3697,6 +3872,11 @@ done:
3697 return (rc != X86EMUL_CONTINUE) ? EMULATION_FAILED : EMULATION_OK; 3872 return (rc != X86EMUL_CONTINUE) ? EMULATION_FAILED : EMULATION_OK;
3698} 3873}
3699 3874
3875bool x86_page_table_writing_insn(struct x86_emulate_ctxt *ctxt)
3876{
3877 return ctxt->d & PageTable;
3878}
3879
3700static bool string_insn_completed(struct x86_emulate_ctxt *ctxt) 3880static bool string_insn_completed(struct x86_emulate_ctxt *ctxt)
3701{ 3881{
3702 /* The second termination condition only applies for REPE 3882 /* The second termination condition only applies for REPE
@@ -3720,7 +3900,6 @@ static bool string_insn_completed(struct x86_emulate_ctxt *ctxt)
3720int x86_emulate_insn(struct x86_emulate_ctxt *ctxt) 3900int x86_emulate_insn(struct x86_emulate_ctxt *ctxt)
3721{ 3901{
3722 struct x86_emulate_ops *ops = ctxt->ops; 3902 struct x86_emulate_ops *ops = ctxt->ops;
3723 u64 msr_data;
3724 int rc = X86EMUL_CONTINUE; 3903 int rc = X86EMUL_CONTINUE;
3725 int saved_dst_type = ctxt->dst.type; 3904 int saved_dst_type = ctxt->dst.type;
3726 3905
@@ -3854,15 +4033,6 @@ special_insn:
3854 goto cannot_emulate; 4033 goto cannot_emulate;
3855 ctxt->dst.val = (s32) ctxt->src.val; 4034 ctxt->dst.val = (s32) ctxt->src.val;
3856 break; 4035 break;
3857 case 0x6c: /* insb */
3858 case 0x6d: /* insw/insd */
3859 ctxt->src.val = ctxt->regs[VCPU_REGS_RDX];
3860 goto do_io_in;
3861 case 0x6e: /* outsb */
3862 case 0x6f: /* outsw/outsd */
3863 ctxt->dst.val = ctxt->regs[VCPU_REGS_RDX];
3864 goto do_io_out;
3865 break;
3866 case 0x70 ... 0x7f: /* jcc (short) */ 4036 case 0x70 ... 0x7f: /* jcc (short) */
3867 if (test_cc(ctxt->b, ctxt->eflags)) 4037 if (test_cc(ctxt->b, ctxt->eflags))
3868 jmp_rel(ctxt, ctxt->src.val); 4038 jmp_rel(ctxt, ctxt->src.val);
@@ -3870,9 +4040,6 @@ special_insn:
3870 case 0x8d: /* lea r16/r32, m */ 4040 case 0x8d: /* lea r16/r32, m */
3871 ctxt->dst.val = ctxt->src.addr.mem.ea; 4041 ctxt->dst.val = ctxt->src.addr.mem.ea;
3872 break; 4042 break;
3873 case 0x8f: /* pop (sole member of Grp1a) */
3874 rc = em_grp1a(ctxt);
3875 break;
3876 case 0x90 ... 0x97: /* nop / xchg reg, rax */ 4043 case 0x90 ... 0x97: /* nop / xchg reg, rax */
3877 if (ctxt->dst.addr.reg == &ctxt->regs[VCPU_REGS_RAX]) 4044 if (ctxt->dst.addr.reg == &ctxt->regs[VCPU_REGS_RAX])
3878 break; 4045 break;
@@ -3905,38 +4072,11 @@ special_insn:
3905 ctxt->src.val = ctxt->regs[VCPU_REGS_RCX]; 4072 ctxt->src.val = ctxt->regs[VCPU_REGS_RCX];
3906 rc = em_grp2(ctxt); 4073 rc = em_grp2(ctxt);
3907 break; 4074 break;
3908 case 0xe4: /* inb */
3909 case 0xe5: /* in */
3910 goto do_io_in;
3911 case 0xe6: /* outb */
3912 case 0xe7: /* out */
3913 goto do_io_out;
3914 case 0xe8: /* call (near) */ {
3915 long int rel = ctxt->src.val;
3916 ctxt->src.val = (unsigned long) ctxt->_eip;
3917 jmp_rel(ctxt, rel);
3918 rc = em_push(ctxt);
3919 break;
3920 }
3921 case 0xe9: /* jmp rel */ 4075 case 0xe9: /* jmp rel */
3922 case 0xeb: /* jmp rel short */ 4076 case 0xeb: /* jmp rel short */
3923 jmp_rel(ctxt, ctxt->src.val); 4077 jmp_rel(ctxt, ctxt->src.val);
3924 ctxt->dst.type = OP_NONE; /* Disable writeback. */ 4078 ctxt->dst.type = OP_NONE; /* Disable writeback. */
3925 break; 4079 break;
3926 case 0xec: /* in al,dx */
3927 case 0xed: /* in (e/r)ax,dx */
3928 do_io_in:
3929 if (!pio_in_emulated(ctxt, ctxt->dst.bytes, ctxt->src.val,
3930 &ctxt->dst.val))
3931 goto done; /* IO is needed */
3932 break;
3933 case 0xee: /* out dx,al */
3934 case 0xef: /* out dx,(e/r)ax */
3935 do_io_out:
3936 ops->pio_out_emulated(ctxt, ctxt->src.bytes, ctxt->dst.val,
3937 &ctxt->src.val, 1);
3938 ctxt->dst.type = OP_NONE; /* Disable writeback. */
3939 break;
3940 case 0xf4: /* hlt */ 4080 case 0xf4: /* hlt */
3941 ctxt->ops->halt(ctxt); 4081 ctxt->ops->halt(ctxt);
3942 break; 4082 break;
@@ -3956,12 +4096,6 @@ special_insn:
3956 case 0xfd: /* std */ 4096 case 0xfd: /* std */
3957 ctxt->eflags |= EFLG_DF; 4097 ctxt->eflags |= EFLG_DF;
3958 break; 4098 break;
3959 case 0xfe: /* Grp4 */
3960 rc = em_grp45(ctxt);
3961 break;
3962 case 0xff: /* Grp5 */
3963 rc = em_grp45(ctxt);
3964 break;
3965 default: 4099 default:
3966 goto cannot_emulate; 4100 goto cannot_emulate;
3967 } 4101 }
@@ -4036,49 +4170,6 @@ twobyte_insn:
4036 case 0x21: /* mov from dr to reg */ 4170 case 0x21: /* mov from dr to reg */
4037 ops->get_dr(ctxt, ctxt->modrm_reg, &ctxt->dst.val); 4171 ops->get_dr(ctxt, ctxt->modrm_reg, &ctxt->dst.val);
4038 break; 4172 break;
4039 case 0x22: /* mov reg, cr */
4040 if (ops->set_cr(ctxt, ctxt->modrm_reg, ctxt->src.val)) {
4041 emulate_gp(ctxt, 0);
4042 rc = X86EMUL_PROPAGATE_FAULT;
4043 goto done;
4044 }
4045 ctxt->dst.type = OP_NONE;
4046 break;
4047 case 0x23: /* mov from reg to dr */
4048 if (ops->set_dr(ctxt, ctxt->modrm_reg, ctxt->src.val &
4049 ((ctxt->mode == X86EMUL_MODE_PROT64) ?
4050 ~0ULL : ~0U)) < 0) {
4051 /* #UD condition is already handled by the code above */
4052 emulate_gp(ctxt, 0);
4053 rc = X86EMUL_PROPAGATE_FAULT;
4054 goto done;
4055 }
4056
4057 ctxt->dst.type = OP_NONE; /* no writeback */
4058 break;
4059 case 0x30:
4060 /* wrmsr */
4061 msr_data = (u32)ctxt->regs[VCPU_REGS_RAX]
4062 | ((u64)ctxt->regs[VCPU_REGS_RDX] << 32);
4063 if (ops->set_msr(ctxt, ctxt->regs[VCPU_REGS_RCX], msr_data)) {
4064 emulate_gp(ctxt, 0);
4065 rc = X86EMUL_PROPAGATE_FAULT;
4066 goto done;
4067 }
4068 rc = X86EMUL_CONTINUE;
4069 break;
4070 case 0x32:
4071 /* rdmsr */
4072 if (ops->get_msr(ctxt, ctxt->regs[VCPU_REGS_RCX], &msr_data)) {
4073 emulate_gp(ctxt, 0);
4074 rc = X86EMUL_PROPAGATE_FAULT;
4075 goto done;
4076 } else {
4077 ctxt->regs[VCPU_REGS_RAX] = (u32)msr_data;
4078 ctxt->regs[VCPU_REGS_RDX] = msr_data >> 32;
4079 }
4080 rc = X86EMUL_CONTINUE;
4081 break;
4082 case 0x40 ... 0x4f: /* cmov */ 4173 case 0x40 ... 0x4f: /* cmov */
4083 ctxt->dst.val = ctxt->dst.orig_val = ctxt->src.val; 4174 ctxt->dst.val = ctxt->dst.orig_val = ctxt->src.val;
4084 if (!test_cc(ctxt->b, ctxt->eflags)) 4175 if (!test_cc(ctxt->b, ctxt->eflags))
@@ -4091,93 +4182,21 @@ twobyte_insn:
4091 case 0x90 ... 0x9f: /* setcc r/m8 */ 4182 case 0x90 ... 0x9f: /* setcc r/m8 */
4092 ctxt->dst.val = test_cc(ctxt->b, ctxt->eflags); 4183 ctxt->dst.val = test_cc(ctxt->b, ctxt->eflags);
4093 break; 4184 break;
4094 case 0xa3:
4095 bt: /* bt */
4096 ctxt->dst.type = OP_NONE;
4097 /* only subword offset */
4098 ctxt->src.val &= (ctxt->dst.bytes << 3) - 1;
4099 emulate_2op_SrcV_nobyte(ctxt, "bt");
4100 break;
4101 case 0xa4: /* shld imm8, r, r/m */ 4185 case 0xa4: /* shld imm8, r, r/m */
4102 case 0xa5: /* shld cl, r, r/m */ 4186 case 0xa5: /* shld cl, r, r/m */
4103 emulate_2op_cl(ctxt, "shld"); 4187 emulate_2op_cl(ctxt, "shld");
4104 break; 4188 break;
4105 case 0xab:
4106 bts: /* bts */
4107 emulate_2op_SrcV_nobyte(ctxt, "bts");
4108 break;
4109 case 0xac: /* shrd imm8, r, r/m */ 4189 case 0xac: /* shrd imm8, r, r/m */
4110 case 0xad: /* shrd cl, r, r/m */ 4190 case 0xad: /* shrd cl, r, r/m */
4111 emulate_2op_cl(ctxt, "shrd"); 4191 emulate_2op_cl(ctxt, "shrd");
4112 break; 4192 break;
4113 case 0xae: /* clflush */ 4193 case 0xae: /* clflush */
4114 break; 4194 break;
4115 case 0xb0 ... 0xb1: /* cmpxchg */
4116 /*
4117 * Save real source value, then compare EAX against
4118 * destination.
4119 */
4120 ctxt->src.orig_val = ctxt->src.val;
4121 ctxt->src.val = ctxt->regs[VCPU_REGS_RAX];
4122 emulate_2op_SrcV(ctxt, "cmp");
4123 if (ctxt->eflags & EFLG_ZF) {
4124 /* Success: write back to memory. */
4125 ctxt->dst.val = ctxt->src.orig_val;
4126 } else {
4127 /* Failure: write the value we saw to EAX. */
4128 ctxt->dst.type = OP_REG;
4129 ctxt->dst.addr.reg = (unsigned long *)&ctxt->regs[VCPU_REGS_RAX];
4130 }
4131 break;
4132 case 0xb3:
4133 btr: /* btr */
4134 emulate_2op_SrcV_nobyte(ctxt, "btr");
4135 break;
4136 case 0xb6 ... 0xb7: /* movzx */ 4195 case 0xb6 ... 0xb7: /* movzx */
4137 ctxt->dst.bytes = ctxt->op_bytes; 4196 ctxt->dst.bytes = ctxt->op_bytes;
4138 ctxt->dst.val = (ctxt->d & ByteOp) ? (u8) ctxt->src.val 4197 ctxt->dst.val = (ctxt->d & ByteOp) ? (u8) ctxt->src.val
4139 : (u16) ctxt->src.val; 4198 : (u16) ctxt->src.val;
4140 break; 4199 break;
4141 case 0xba: /* Grp8 */
4142 switch (ctxt->modrm_reg & 3) {
4143 case 0:
4144 goto bt;
4145 case 1:
4146 goto bts;
4147 case 2:
4148 goto btr;
4149 case 3:
4150 goto btc;
4151 }
4152 break;
4153 case 0xbb:
4154 btc: /* btc */
4155 emulate_2op_SrcV_nobyte(ctxt, "btc");
4156 break;
4157 case 0xbc: { /* bsf */
4158 u8 zf;
4159 __asm__ ("bsf %2, %0; setz %1"
4160 : "=r"(ctxt->dst.val), "=q"(zf)
4161 : "r"(ctxt->src.val));
4162 ctxt->eflags &= ~X86_EFLAGS_ZF;
4163 if (zf) {
4164 ctxt->eflags |= X86_EFLAGS_ZF;
4165 ctxt->dst.type = OP_NONE; /* Disable writeback. */
4166 }
4167 break;
4168 }
4169 case 0xbd: { /* bsr */
4170 u8 zf;
4171 __asm__ ("bsr %2, %0; setz %1"
4172 : "=r"(ctxt->dst.val), "=q"(zf)
4173 : "r"(ctxt->src.val));
4174 ctxt->eflags &= ~X86_EFLAGS_ZF;
4175 if (zf) {
4176 ctxt->eflags |= X86_EFLAGS_ZF;
4177 ctxt->dst.type = OP_NONE; /* Disable writeback. */
4178 }
4179 break;
4180 }
4181 case 0xbe ... 0xbf: /* movsx */ 4200 case 0xbe ... 0xbf: /* movsx */
4182 ctxt->dst.bytes = ctxt->op_bytes; 4201 ctxt->dst.bytes = ctxt->op_bytes;
4183 ctxt->dst.val = (ctxt->d & ByteOp) ? (s8) ctxt->src.val : 4202 ctxt->dst.val = (ctxt->d & ByteOp) ? (s8) ctxt->src.val :
@@ -4194,9 +4213,6 @@ twobyte_insn:
4194 ctxt->dst.val = (ctxt->op_bytes == 4) ? (u32) ctxt->src.val : 4213 ctxt->dst.val = (ctxt->op_bytes == 4) ? (u32) ctxt->src.val :
4195 (u64) ctxt->src.val; 4214 (u64) ctxt->src.val;
4196 break; 4215 break;
4197 case 0xc7: /* Grp9 (cmpxchg8b) */
4198 rc = em_grp9(ctxt);
4199 break;
4200 default: 4216 default:
4201 goto cannot_emulate; 4217 goto cannot_emulate;
4202 } 4218 }
diff --git a/arch/x86/kvm/i8254.c b/arch/x86/kvm/i8254.c
index 76e3f1cd0369..d68f99df690c 100644
--- a/arch/x86/kvm/i8254.c
+++ b/arch/x86/kvm/i8254.c
@@ -338,11 +338,15 @@ static enum hrtimer_restart pit_timer_fn(struct hrtimer *data)
338 return HRTIMER_NORESTART; 338 return HRTIMER_NORESTART;
339} 339}
340 340
341static void create_pit_timer(struct kvm_kpit_state *ps, u32 val, int is_period) 341static void create_pit_timer(struct kvm *kvm, u32 val, int is_period)
342{ 342{
343 struct kvm_kpit_state *ps = &kvm->arch.vpit->pit_state;
343 struct kvm_timer *pt = &ps->pit_timer; 344 struct kvm_timer *pt = &ps->pit_timer;
344 s64 interval; 345 s64 interval;
345 346
347 if (!irqchip_in_kernel(kvm) || ps->flags & KVM_PIT_FLAGS_HPET_LEGACY)
348 return;
349
346 interval = muldiv64(val, NSEC_PER_SEC, KVM_PIT_FREQ); 350 interval = muldiv64(val, NSEC_PER_SEC, KVM_PIT_FREQ);
347 351
348 pr_debug("create pit timer, interval is %llu nsec\n", interval); 352 pr_debug("create pit timer, interval is %llu nsec\n", interval);
@@ -393,15 +397,11 @@ static void pit_load_count(struct kvm *kvm, int channel, u32 val)
393 case 1: 397 case 1:
394 /* FIXME: enhance mode 4 precision */ 398 /* FIXME: enhance mode 4 precision */
395 case 4: 399 case 4:
396 if (!(ps->flags & KVM_PIT_FLAGS_HPET_LEGACY)) { 400 create_pit_timer(kvm, val, 0);
397 create_pit_timer(ps, val, 0);
398 }
399 break; 401 break;
400 case 2: 402 case 2:
401 case 3: 403 case 3:
402 if (!(ps->flags & KVM_PIT_FLAGS_HPET_LEGACY)){ 404 create_pit_timer(kvm, val, 1);
403 create_pit_timer(ps, val, 1);
404 }
405 break; 405 break;
406 default: 406 default:
407 destroy_pit_timer(kvm->arch.vpit); 407 destroy_pit_timer(kvm->arch.vpit);
diff --git a/arch/x86/kvm/i8259.c b/arch/x86/kvm/i8259.c
index cac4746d7ffb..b6a73537e1ef 100644
--- a/arch/x86/kvm/i8259.c
+++ b/arch/x86/kvm/i8259.c
@@ -262,9 +262,10 @@ int kvm_pic_read_irq(struct kvm *kvm)
262 262
263void kvm_pic_reset(struct kvm_kpic_state *s) 263void kvm_pic_reset(struct kvm_kpic_state *s)
264{ 264{
265 int irq; 265 int irq, i;
266 struct kvm_vcpu *vcpu0 = s->pics_state->kvm->bsp_vcpu; 266 struct kvm_vcpu *vcpu;
267 u8 irr = s->irr, isr = s->imr; 267 u8 irr = s->irr, isr = s->imr;
268 bool found = false;
268 269
269 s->last_irr = 0; 270 s->last_irr = 0;
270 s->irr = 0; 271 s->irr = 0;
@@ -281,12 +282,19 @@ void kvm_pic_reset(struct kvm_kpic_state *s)
281 s->special_fully_nested_mode = 0; 282 s->special_fully_nested_mode = 0;
282 s->init4 = 0; 283 s->init4 = 0;
283 284
284 for (irq = 0; irq < PIC_NUM_PINS/2; irq++) { 285 kvm_for_each_vcpu(i, vcpu, s->pics_state->kvm)
285 if (vcpu0 && kvm_apic_accept_pic_intr(vcpu0)) 286 if (kvm_apic_accept_pic_intr(vcpu)) {
286 if (irr & (1 << irq) || isr & (1 << irq)) { 287 found = true;
287 pic_clear_isr(s, irq); 288 break;
288 } 289 }
289 } 290
291
292 if (!found)
293 return;
294
295 for (irq = 0; irq < PIC_NUM_PINS/2; irq++)
296 if (irr & (1 << irq) || isr & (1 << irq))
297 pic_clear_isr(s, irq);
290} 298}
291 299
292static void pic_ioport_write(void *opaque, u32 addr, u32 val) 300static void pic_ioport_write(void *opaque, u32 addr, u32 val)
diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
index 54abb40199d6..cfdc6e0ef002 100644
--- a/arch/x86/kvm/lapic.c
+++ b/arch/x86/kvm/lapic.c
@@ -38,6 +38,7 @@
38#include "irq.h" 38#include "irq.h"
39#include "trace.h" 39#include "trace.h"
40#include "x86.h" 40#include "x86.h"
41#include "cpuid.h"
41 42
42#ifndef CONFIG_X86_64 43#ifndef CONFIG_X86_64
43#define mod_64(x, y) ((x) - (y) * div64_u64(x, y)) 44#define mod_64(x, y) ((x) - (y) * div64_u64(x, y))
@@ -1120,7 +1121,7 @@ int apic_has_pending_timer(struct kvm_vcpu *vcpu)
1120 return 0; 1121 return 0;
1121} 1122}
1122 1123
1123static int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type) 1124int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type)
1124{ 1125{
1125 u32 reg = apic_get_reg(apic, lvt_type); 1126 u32 reg = apic_get_reg(apic, lvt_type);
1126 int vector, mode, trig_mode; 1127 int vector, mode, trig_mode;
diff --git a/arch/x86/kvm/lapic.h b/arch/x86/kvm/lapic.h
index 138e8cc6fea6..6f4ce2575d09 100644
--- a/arch/x86/kvm/lapic.h
+++ b/arch/x86/kvm/lapic.h
@@ -34,6 +34,7 @@ void kvm_apic_set_version(struct kvm_vcpu *vcpu);
34int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest); 34int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest);
35int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda); 35int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda);
36int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq); 36int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq);
37int kvm_apic_local_deliver(struct kvm_lapic *apic, int lvt_type);
37 38
38u64 kvm_get_apic_base(struct kvm_vcpu *vcpu); 39u64 kvm_get_apic_base(struct kvm_vcpu *vcpu);
39void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data); 40void kvm_set_apic_base(struct kvm_vcpu *vcpu, u64 data);
diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index f1b36cf3e3d0..224b02c3cda9 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -59,15 +59,6 @@ enum {
59 AUDIT_POST_SYNC 59 AUDIT_POST_SYNC
60}; 60};
61 61
62char *audit_point_name[] = {
63 "pre page fault",
64 "post page fault",
65 "pre pte write",
66 "post pte write",
67 "pre sync",
68 "post sync"
69};
70
71#undef MMU_DEBUG 62#undef MMU_DEBUG
72 63
73#ifdef MMU_DEBUG 64#ifdef MMU_DEBUG
@@ -83,13 +74,10 @@ char *audit_point_name[] = {
83#endif 74#endif
84 75
85#ifdef MMU_DEBUG 76#ifdef MMU_DEBUG
86static int dbg = 0; 77static bool dbg = 0;
87module_param(dbg, bool, 0644); 78module_param(dbg, bool, 0644);
88#endif 79#endif
89 80
90static int oos_shadow = 1;
91module_param(oos_shadow, bool, 0644);
92
93#ifndef MMU_DEBUG 81#ifndef MMU_DEBUG
94#define ASSERT(x) do { } while (0) 82#define ASSERT(x) do { } while (0)
95#else 83#else
@@ -593,6 +581,11 @@ static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache,
593 return 0; 581 return 0;
594} 582}
595 583
584static int mmu_memory_cache_free_objects(struct kvm_mmu_memory_cache *cache)
585{
586 return cache->nobjs;
587}
588
596static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc, 589static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc,
597 struct kmem_cache *cache) 590 struct kmem_cache *cache)
598{ 591{
@@ -953,21 +946,35 @@ static void pte_list_walk(unsigned long *pte_list, pte_list_walk_fn fn)
953 } 946 }
954} 947}
955 948
949static unsigned long *__gfn_to_rmap(struct kvm *kvm, gfn_t gfn, int level,
950 struct kvm_memory_slot *slot)
951{
952 struct kvm_lpage_info *linfo;
953
954 if (likely(level == PT_PAGE_TABLE_LEVEL))
955 return &slot->rmap[gfn - slot->base_gfn];
956
957 linfo = lpage_info_slot(gfn, slot, level);
958 return &linfo->rmap_pde;
959}
960
956/* 961/*
957 * Take gfn and return the reverse mapping to it. 962 * Take gfn and return the reverse mapping to it.
958 */ 963 */
959static unsigned long *gfn_to_rmap(struct kvm *kvm, gfn_t gfn, int level) 964static unsigned long *gfn_to_rmap(struct kvm *kvm, gfn_t gfn, int level)
960{ 965{
961 struct kvm_memory_slot *slot; 966 struct kvm_memory_slot *slot;
962 struct kvm_lpage_info *linfo;
963 967
964 slot = gfn_to_memslot(kvm, gfn); 968 slot = gfn_to_memslot(kvm, gfn);
965 if (likely(level == PT_PAGE_TABLE_LEVEL)) 969 return __gfn_to_rmap(kvm, gfn, level, slot);
966 return &slot->rmap[gfn - slot->base_gfn]; 970}
967 971
968 linfo = lpage_info_slot(gfn, slot, level); 972static bool rmap_can_add(struct kvm_vcpu *vcpu)
973{
974 struct kvm_mmu_memory_cache *cache;
969 975
970 return &linfo->rmap_pde; 976 cache = &vcpu->arch.mmu_pte_list_desc_cache;
977 return mmu_memory_cache_free_objects(cache);
971} 978}
972 979
973static int rmap_add(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn) 980static int rmap_add(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn)
@@ -1004,17 +1011,16 @@ static void drop_spte(struct kvm *kvm, u64 *sptep)
1004 rmap_remove(kvm, sptep); 1011 rmap_remove(kvm, sptep);
1005} 1012}
1006 1013
1007static int rmap_write_protect(struct kvm *kvm, u64 gfn) 1014int kvm_mmu_rmap_write_protect(struct kvm *kvm, u64 gfn,
1015 struct kvm_memory_slot *slot)
1008{ 1016{
1009 unsigned long *rmapp; 1017 unsigned long *rmapp;
1010 u64 *spte; 1018 u64 *spte;
1011 int i, write_protected = 0; 1019 int i, write_protected = 0;
1012 1020
1013 rmapp = gfn_to_rmap(kvm, gfn, PT_PAGE_TABLE_LEVEL); 1021 rmapp = __gfn_to_rmap(kvm, gfn, PT_PAGE_TABLE_LEVEL, slot);
1014
1015 spte = rmap_next(kvm, rmapp, NULL); 1022 spte = rmap_next(kvm, rmapp, NULL);
1016 while (spte) { 1023 while (spte) {
1017 BUG_ON(!spte);
1018 BUG_ON(!(*spte & PT_PRESENT_MASK)); 1024 BUG_ON(!(*spte & PT_PRESENT_MASK));
1019 rmap_printk("rmap_write_protect: spte %p %llx\n", spte, *spte); 1025 rmap_printk("rmap_write_protect: spte %p %llx\n", spte, *spte);
1020 if (is_writable_pte(*spte)) { 1026 if (is_writable_pte(*spte)) {
@@ -1027,12 +1033,11 @@ static int rmap_write_protect(struct kvm *kvm, u64 gfn)
1027 /* check for huge page mappings */ 1033 /* check for huge page mappings */
1028 for (i = PT_DIRECTORY_LEVEL; 1034 for (i = PT_DIRECTORY_LEVEL;
1029 i < PT_PAGE_TABLE_LEVEL + KVM_NR_PAGE_SIZES; ++i) { 1035 i < PT_PAGE_TABLE_LEVEL + KVM_NR_PAGE_SIZES; ++i) {
1030 rmapp = gfn_to_rmap(kvm, gfn, i); 1036 rmapp = __gfn_to_rmap(kvm, gfn, i, slot);
1031 spte = rmap_next(kvm, rmapp, NULL); 1037 spte = rmap_next(kvm, rmapp, NULL);
1032 while (spte) { 1038 while (spte) {
1033 BUG_ON(!spte);
1034 BUG_ON(!(*spte & PT_PRESENT_MASK)); 1039 BUG_ON(!(*spte & PT_PRESENT_MASK));
1035 BUG_ON((*spte & (PT_PAGE_SIZE_MASK|PT_PRESENT_MASK)) != (PT_PAGE_SIZE_MASK|PT_PRESENT_MASK)); 1040 BUG_ON(!is_large_pte(*spte));
1036 pgprintk("rmap_write_protect(large): spte %p %llx %lld\n", spte, *spte, gfn); 1041 pgprintk("rmap_write_protect(large): spte %p %llx %lld\n", spte, *spte, gfn);
1037 if (is_writable_pte(*spte)) { 1042 if (is_writable_pte(*spte)) {
1038 drop_spte(kvm, spte); 1043 drop_spte(kvm, spte);
@@ -1047,6 +1052,14 @@ static int rmap_write_protect(struct kvm *kvm, u64 gfn)
1047 return write_protected; 1052 return write_protected;
1048} 1053}
1049 1054
1055static int rmap_write_protect(struct kvm *kvm, u64 gfn)
1056{
1057 struct kvm_memory_slot *slot;
1058
1059 slot = gfn_to_memslot(kvm, gfn);
1060 return kvm_mmu_rmap_write_protect(kvm, gfn, slot);
1061}
1062
1050static int kvm_unmap_rmapp(struct kvm *kvm, unsigned long *rmapp, 1063static int kvm_unmap_rmapp(struct kvm *kvm, unsigned long *rmapp,
1051 unsigned long data) 1064 unsigned long data)
1052{ 1065{
@@ -1103,15 +1116,15 @@ static int kvm_handle_hva(struct kvm *kvm, unsigned long hva,
1103 int (*handler)(struct kvm *kvm, unsigned long *rmapp, 1116 int (*handler)(struct kvm *kvm, unsigned long *rmapp,
1104 unsigned long data)) 1117 unsigned long data))
1105{ 1118{
1106 int i, j; 1119 int j;
1107 int ret; 1120 int ret;
1108 int retval = 0; 1121 int retval = 0;
1109 struct kvm_memslots *slots; 1122 struct kvm_memslots *slots;
1123 struct kvm_memory_slot *memslot;
1110 1124
1111 slots = kvm_memslots(kvm); 1125 slots = kvm_memslots(kvm);
1112 1126
1113 for (i = 0; i < slots->nmemslots; i++) { 1127 kvm_for_each_memslot(memslot, slots) {
1114 struct kvm_memory_slot *memslot = &slots->memslots[i];
1115 unsigned long start = memslot->userspace_addr; 1128 unsigned long start = memslot->userspace_addr;
1116 unsigned long end; 1129 unsigned long end;
1117 1130
@@ -1324,7 +1337,7 @@ static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu,
1324 PAGE_SIZE); 1337 PAGE_SIZE);
1325 set_page_private(virt_to_page(sp->spt), (unsigned long)sp); 1338 set_page_private(virt_to_page(sp->spt), (unsigned long)sp);
1326 list_add(&sp->link, &vcpu->kvm->arch.active_mmu_pages); 1339 list_add(&sp->link, &vcpu->kvm->arch.active_mmu_pages);
1327 bitmap_zero(sp->slot_bitmap, KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS); 1340 bitmap_zero(sp->slot_bitmap, KVM_MEM_SLOTS_NUM);
1328 sp->parent_ptes = 0; 1341 sp->parent_ptes = 0;
1329 mmu_page_add_parent_pte(vcpu, sp, parent_pte); 1342 mmu_page_add_parent_pte(vcpu, sp, parent_pte);
1330 kvm_mod_used_mmu_pages(vcpu->kvm, +1); 1343 kvm_mod_used_mmu_pages(vcpu->kvm, +1);
@@ -1511,6 +1524,13 @@ static int kvm_sync_page_transient(struct kvm_vcpu *vcpu,
1511 return ret; 1524 return ret;
1512} 1525}
1513 1526
1527#ifdef CONFIG_KVM_MMU_AUDIT
1528#include "mmu_audit.c"
1529#else
1530static void kvm_mmu_audit(struct kvm_vcpu *vcpu, int point) { }
1531static void mmu_audit_disable(void) { }
1532#endif
1533
1514static int kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp, 1534static int kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
1515 struct list_head *invalid_list) 1535 struct list_head *invalid_list)
1516{ 1536{
@@ -1640,6 +1660,18 @@ static void init_shadow_page_table(struct kvm_mmu_page *sp)
1640 sp->spt[i] = 0ull; 1660 sp->spt[i] = 0ull;
1641} 1661}
1642 1662
1663static void __clear_sp_write_flooding_count(struct kvm_mmu_page *sp)
1664{
1665 sp->write_flooding_count = 0;
1666}
1667
1668static void clear_sp_write_flooding_count(u64 *spte)
1669{
1670 struct kvm_mmu_page *sp = page_header(__pa(spte));
1671
1672 __clear_sp_write_flooding_count(sp);
1673}
1674
1643static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu, 1675static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
1644 gfn_t gfn, 1676 gfn_t gfn,
1645 gva_t gaddr, 1677 gva_t gaddr,
@@ -1683,6 +1715,7 @@ static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
1683 } else if (sp->unsync) 1715 } else if (sp->unsync)
1684 kvm_mmu_mark_parents_unsync(sp); 1716 kvm_mmu_mark_parents_unsync(sp);
1685 1717
1718 __clear_sp_write_flooding_count(sp);
1686 trace_kvm_mmu_get_page(sp, false); 1719 trace_kvm_mmu_get_page(sp, false);
1687 return sp; 1720 return sp;
1688 } 1721 }
@@ -1796,7 +1829,7 @@ static void validate_direct_spte(struct kvm_vcpu *vcpu, u64 *sptep,
1796 } 1829 }
1797} 1830}
1798 1831
1799static void mmu_page_zap_pte(struct kvm *kvm, struct kvm_mmu_page *sp, 1832static bool mmu_page_zap_pte(struct kvm *kvm, struct kvm_mmu_page *sp,
1800 u64 *spte) 1833 u64 *spte)
1801{ 1834{
1802 u64 pte; 1835 u64 pte;
@@ -1804,17 +1837,21 @@ static void mmu_page_zap_pte(struct kvm *kvm, struct kvm_mmu_page *sp,
1804 1837
1805 pte = *spte; 1838 pte = *spte;
1806 if (is_shadow_present_pte(pte)) { 1839 if (is_shadow_present_pte(pte)) {
1807 if (is_last_spte(pte, sp->role.level)) 1840 if (is_last_spte(pte, sp->role.level)) {
1808 drop_spte(kvm, spte); 1841 drop_spte(kvm, spte);
1809 else { 1842 if (is_large_pte(pte))
1843 --kvm->stat.lpages;
1844 } else {
1810 child = page_header(pte & PT64_BASE_ADDR_MASK); 1845 child = page_header(pte & PT64_BASE_ADDR_MASK);
1811 drop_parent_pte(child, spte); 1846 drop_parent_pte(child, spte);
1812 } 1847 }
1813 } else if (is_mmio_spte(pte)) 1848 return true;
1849 }
1850
1851 if (is_mmio_spte(pte))
1814 mmu_spte_clear_no_track(spte); 1852 mmu_spte_clear_no_track(spte);
1815 1853
1816 if (is_large_pte(pte)) 1854 return false;
1817 --kvm->stat.lpages;
1818} 1855}
1819 1856
1820static void kvm_mmu_page_unlink_children(struct kvm *kvm, 1857static void kvm_mmu_page_unlink_children(struct kvm *kvm,
@@ -1831,15 +1868,6 @@ static void kvm_mmu_put_page(struct kvm_mmu_page *sp, u64 *parent_pte)
1831 mmu_page_remove_parent_pte(sp, parent_pte); 1868 mmu_page_remove_parent_pte(sp, parent_pte);
1832} 1869}
1833 1870
1834static void kvm_mmu_reset_last_pte_updated(struct kvm *kvm)
1835{
1836 int i;
1837 struct kvm_vcpu *vcpu;
1838
1839 kvm_for_each_vcpu(i, vcpu, kvm)
1840 vcpu->arch.last_pte_updated = NULL;
1841}
1842
1843static void kvm_mmu_unlink_parents(struct kvm *kvm, struct kvm_mmu_page *sp) 1871static void kvm_mmu_unlink_parents(struct kvm *kvm, struct kvm_mmu_page *sp)
1844{ 1872{
1845 u64 *parent_pte; 1873 u64 *parent_pte;
@@ -1899,7 +1927,6 @@ static int kvm_mmu_prepare_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp,
1899 } 1927 }
1900 1928
1901 sp->role.invalid = 1; 1929 sp->role.invalid = 1;
1902 kvm_mmu_reset_last_pte_updated(kvm);
1903 return ret; 1930 return ret;
1904} 1931}
1905 1932
@@ -1985,7 +2012,7 @@ void kvm_mmu_change_mmu_pages(struct kvm *kvm, unsigned int goal_nr_mmu_pages)
1985 kvm->arch.n_max_mmu_pages = goal_nr_mmu_pages; 2012 kvm->arch.n_max_mmu_pages = goal_nr_mmu_pages;
1986} 2013}
1987 2014
1988static int kvm_mmu_unprotect_page(struct kvm *kvm, gfn_t gfn) 2015int kvm_mmu_unprotect_page(struct kvm *kvm, gfn_t gfn)
1989{ 2016{
1990 struct kvm_mmu_page *sp; 2017 struct kvm_mmu_page *sp;
1991 struct hlist_node *node; 2018 struct hlist_node *node;
@@ -1994,7 +2021,7 @@ static int kvm_mmu_unprotect_page(struct kvm *kvm, gfn_t gfn)
1994 2021
1995 pgprintk("%s: looking for gfn %llx\n", __func__, gfn); 2022 pgprintk("%s: looking for gfn %llx\n", __func__, gfn);
1996 r = 0; 2023 r = 0;
1997 2024 spin_lock(&kvm->mmu_lock);
1998 for_each_gfn_indirect_valid_sp(kvm, sp, gfn, node) { 2025 for_each_gfn_indirect_valid_sp(kvm, sp, gfn, node) {
1999 pgprintk("%s: gfn %llx role %x\n", __func__, gfn, 2026 pgprintk("%s: gfn %llx role %x\n", __func__, gfn,
2000 sp->role.word); 2027 sp->role.word);
@@ -2002,22 +2029,11 @@ static int kvm_mmu_unprotect_page(struct kvm *kvm, gfn_t gfn)
2002 kvm_mmu_prepare_zap_page(kvm, sp, &invalid_list); 2029 kvm_mmu_prepare_zap_page(kvm, sp, &invalid_list);
2003 } 2030 }
2004 kvm_mmu_commit_zap_page(kvm, &invalid_list); 2031 kvm_mmu_commit_zap_page(kvm, &invalid_list);
2005 return r; 2032 spin_unlock(&kvm->mmu_lock);
2006}
2007
2008static void mmu_unshadow(struct kvm *kvm, gfn_t gfn)
2009{
2010 struct kvm_mmu_page *sp;
2011 struct hlist_node *node;
2012 LIST_HEAD(invalid_list);
2013 2033
2014 for_each_gfn_indirect_valid_sp(kvm, sp, gfn, node) { 2034 return r;
2015 pgprintk("%s: zap %llx %x\n",
2016 __func__, gfn, sp->role.word);
2017 kvm_mmu_prepare_zap_page(kvm, sp, &invalid_list);
2018 }
2019 kvm_mmu_commit_zap_page(kvm, &invalid_list);
2020} 2035}
2036EXPORT_SYMBOL_GPL(kvm_mmu_unprotect_page);
2021 2037
2022static void page_header_update_slot(struct kvm *kvm, void *pte, gfn_t gfn) 2038static void page_header_update_slot(struct kvm *kvm, void *pte, gfn_t gfn)
2023{ 2039{
@@ -2169,8 +2185,6 @@ static int mmu_need_write_protect(struct kvm_vcpu *vcpu, gfn_t gfn,
2169 return 1; 2185 return 1;
2170 2186
2171 if (!need_unsync && !s->unsync) { 2187 if (!need_unsync && !s->unsync) {
2172 if (!oos_shadow)
2173 return 1;
2174 need_unsync = true; 2188 need_unsync = true;
2175 } 2189 }
2176 } 2190 }
@@ -2191,11 +2205,6 @@ static int set_spte(struct kvm_vcpu *vcpu, u64 *sptep,
2191 if (set_mmio_spte(sptep, gfn, pfn, pte_access)) 2205 if (set_mmio_spte(sptep, gfn, pfn, pte_access))
2192 return 0; 2206 return 0;
2193 2207
2194 /*
2195 * We don't set the accessed bit, since we sometimes want to see
2196 * whether the guest actually used the pte (in order to detect
2197 * demand paging).
2198 */
2199 spte = PT_PRESENT_MASK; 2208 spte = PT_PRESENT_MASK;
2200 if (!speculative) 2209 if (!speculative)
2201 spte |= shadow_accessed_mask; 2210 spte |= shadow_accessed_mask;
@@ -2346,10 +2355,6 @@ static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *sptep,
2346 } 2355 }
2347 } 2356 }
2348 kvm_release_pfn_clean(pfn); 2357 kvm_release_pfn_clean(pfn);
2349 if (speculative) {
2350 vcpu->arch.last_pte_updated = sptep;
2351 vcpu->arch.last_pte_gfn = gfn;
2352 }
2353} 2358}
2354 2359
2355static void nonpaging_new_cr3(struct kvm_vcpu *vcpu) 2360static void nonpaging_new_cr3(struct kvm_vcpu *vcpu)
@@ -2840,12 +2845,12 @@ static void mmu_sync_roots(struct kvm_vcpu *vcpu)
2840 return; 2845 return;
2841 2846
2842 vcpu_clear_mmio_info(vcpu, ~0ul); 2847 vcpu_clear_mmio_info(vcpu, ~0ul);
2843 trace_kvm_mmu_audit(vcpu, AUDIT_PRE_SYNC); 2848 kvm_mmu_audit(vcpu, AUDIT_PRE_SYNC);
2844 if (vcpu->arch.mmu.root_level == PT64_ROOT_LEVEL) { 2849 if (vcpu->arch.mmu.root_level == PT64_ROOT_LEVEL) {
2845 hpa_t root = vcpu->arch.mmu.root_hpa; 2850 hpa_t root = vcpu->arch.mmu.root_hpa;
2846 sp = page_header(root); 2851 sp = page_header(root);
2847 mmu_sync_children(vcpu, sp); 2852 mmu_sync_children(vcpu, sp);
2848 trace_kvm_mmu_audit(vcpu, AUDIT_POST_SYNC); 2853 kvm_mmu_audit(vcpu, AUDIT_POST_SYNC);
2849 return; 2854 return;
2850 } 2855 }
2851 for (i = 0; i < 4; ++i) { 2856 for (i = 0; i < 4; ++i) {
@@ -2857,7 +2862,7 @@ static void mmu_sync_roots(struct kvm_vcpu *vcpu)
2857 mmu_sync_children(vcpu, sp); 2862 mmu_sync_children(vcpu, sp);
2858 } 2863 }
2859 } 2864 }
2860 trace_kvm_mmu_audit(vcpu, AUDIT_POST_SYNC); 2865 kvm_mmu_audit(vcpu, AUDIT_POST_SYNC);
2861} 2866}
2862 2867
2863void kvm_mmu_sync_roots(struct kvm_vcpu *vcpu) 2868void kvm_mmu_sync_roots(struct kvm_vcpu *vcpu)
@@ -3510,28 +3515,119 @@ static void mmu_pte_write_flush_tlb(struct kvm_vcpu *vcpu, bool zap_page,
3510 kvm_mmu_flush_tlb(vcpu); 3515 kvm_mmu_flush_tlb(vcpu);
3511} 3516}
3512 3517
3513static bool last_updated_pte_accessed(struct kvm_vcpu *vcpu) 3518static u64 mmu_pte_write_fetch_gpte(struct kvm_vcpu *vcpu, gpa_t *gpa,
3519 const u8 *new, int *bytes)
3514{ 3520{
3515 u64 *spte = vcpu->arch.last_pte_updated; 3521 u64 gentry;
3522 int r;
3523
3524 /*
3525 * Assume that the pte write on a page table of the same type
3526 * as the current vcpu paging mode since we update the sptes only
3527 * when they have the same mode.
3528 */
3529 if (is_pae(vcpu) && *bytes == 4) {
3530 /* Handle a 32-bit guest writing two halves of a 64-bit gpte */
3531 *gpa &= ~(gpa_t)7;
3532 *bytes = 8;
3533 r = kvm_read_guest(vcpu->kvm, *gpa, &gentry, min(*bytes, 8));
3534 if (r)
3535 gentry = 0;
3536 new = (const u8 *)&gentry;
3537 }
3516 3538
3517 return !!(spte && (*spte & shadow_accessed_mask)); 3539 switch (*bytes) {
3540 case 4:
3541 gentry = *(const u32 *)new;
3542 break;
3543 case 8:
3544 gentry = *(const u64 *)new;
3545 break;
3546 default:
3547 gentry = 0;
3548 break;
3549 }
3550
3551 return gentry;
3518} 3552}
3519 3553
3520static void kvm_mmu_access_page(struct kvm_vcpu *vcpu, gfn_t gfn) 3554/*
3555 * If we're seeing too many writes to a page, it may no longer be a page table,
3556 * or we may be forking, in which case it is better to unmap the page.
3557 */
3558static bool detect_write_flooding(struct kvm_mmu_page *sp, u64 *spte)
3521{ 3559{
3522 u64 *spte = vcpu->arch.last_pte_updated; 3560 /*
3561 * Skip write-flooding detected for the sp whose level is 1, because
3562 * it can become unsync, then the guest page is not write-protected.
3563 */
3564 if (sp->role.level == 1)
3565 return false;
3523 3566
3524 if (spte 3567 return ++sp->write_flooding_count >= 3;
3525 && vcpu->arch.last_pte_gfn == gfn 3568}
3526 && shadow_accessed_mask 3569
3527 && !(*spte & shadow_accessed_mask) 3570/*
3528 && is_shadow_present_pte(*spte)) 3571 * Misaligned accesses are too much trouble to fix up; also, they usually
3529 set_bit(PT_ACCESSED_SHIFT, (unsigned long *)spte); 3572 * indicate a page is not used as a page table.
3573 */
3574static bool detect_write_misaligned(struct kvm_mmu_page *sp, gpa_t gpa,
3575 int bytes)
3576{
3577 unsigned offset, pte_size, misaligned;
3578
3579 pgprintk("misaligned: gpa %llx bytes %d role %x\n",
3580 gpa, bytes, sp->role.word);
3581
3582 offset = offset_in_page(gpa);
3583 pte_size = sp->role.cr4_pae ? 8 : 4;
3584
3585 /*
3586 * Sometimes, the OS only writes the last one bytes to update status
3587 * bits, for example, in linux, andb instruction is used in clear_bit().
3588 */
3589 if (!(offset & (pte_size - 1)) && bytes == 1)
3590 return false;
3591
3592 misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1);
3593 misaligned |= bytes < 4;
3594
3595 return misaligned;
3596}
3597
3598static u64 *get_written_sptes(struct kvm_mmu_page *sp, gpa_t gpa, int *nspte)
3599{
3600 unsigned page_offset, quadrant;
3601 u64 *spte;
3602 int level;
3603
3604 page_offset = offset_in_page(gpa);
3605 level = sp->role.level;
3606 *nspte = 1;
3607 if (!sp->role.cr4_pae) {
3608 page_offset <<= 1; /* 32->64 */
3609 /*
3610 * A 32-bit pde maps 4MB while the shadow pdes map
3611 * only 2MB. So we need to double the offset again
3612 * and zap two pdes instead of one.
3613 */
3614 if (level == PT32_ROOT_LEVEL) {
3615 page_offset &= ~7; /* kill rounding error */
3616 page_offset <<= 1;
3617 *nspte = 2;
3618 }
3619 quadrant = page_offset >> PAGE_SHIFT;
3620 page_offset &= ~PAGE_MASK;
3621 if (quadrant != sp->role.quadrant)
3622 return NULL;
3623 }
3624
3625 spte = &sp->spt[page_offset / sizeof(*spte)];
3626 return spte;
3530} 3627}
3531 3628
3532void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa, 3629void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
3533 const u8 *new, int bytes, 3630 const u8 *new, int bytes)
3534 bool guest_initiated)
3535{ 3631{
3536 gfn_t gfn = gpa >> PAGE_SHIFT; 3632 gfn_t gfn = gpa >> PAGE_SHIFT;
3537 union kvm_mmu_page_role mask = { .word = 0 }; 3633 union kvm_mmu_page_role mask = { .word = 0 };
@@ -3539,8 +3635,7 @@ void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
3539 struct hlist_node *node; 3635 struct hlist_node *node;
3540 LIST_HEAD(invalid_list); 3636 LIST_HEAD(invalid_list);
3541 u64 entry, gentry, *spte; 3637 u64 entry, gentry, *spte;
3542 unsigned pte_size, page_offset, misaligned, quadrant, offset; 3638 int npte;
3543 int level, npte, invlpg_counter, r, flooded = 0;
3544 bool remote_flush, local_flush, zap_page; 3639 bool remote_flush, local_flush, zap_page;
3545 3640
3546 /* 3641 /*
@@ -3551,112 +3646,45 @@ void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
3551 return; 3646 return;
3552 3647
3553 zap_page = remote_flush = local_flush = false; 3648 zap_page = remote_flush = local_flush = false;
3554 offset = offset_in_page(gpa);
3555 3649
3556 pgprintk("%s: gpa %llx bytes %d\n", __func__, gpa, bytes); 3650 pgprintk("%s: gpa %llx bytes %d\n", __func__, gpa, bytes);
3557 3651
3558 invlpg_counter = atomic_read(&vcpu->kvm->arch.invlpg_counter); 3652 gentry = mmu_pte_write_fetch_gpte(vcpu, &gpa, new, &bytes);
3559 3653
3560 /* 3654 /*
3561 * Assume that the pte write on a page table of the same type 3655 * No need to care whether allocation memory is successful
3562 * as the current vcpu paging mode since we update the sptes only 3656 * or not since pte prefetch is skiped if it does not have
3563 * when they have the same mode. 3657 * enough objects in the cache.
3564 */ 3658 */
3565 if ((is_pae(vcpu) && bytes == 4) || !new) { 3659 mmu_topup_memory_caches(vcpu);
3566 /* Handle a 32-bit guest writing two halves of a 64-bit gpte */
3567 if (is_pae(vcpu)) {
3568 gpa &= ~(gpa_t)7;
3569 bytes = 8;
3570 }
3571 r = kvm_read_guest(vcpu->kvm, gpa, &gentry, min(bytes, 8));
3572 if (r)
3573 gentry = 0;
3574 new = (const u8 *)&gentry;
3575 }
3576
3577 switch (bytes) {
3578 case 4:
3579 gentry = *(const u32 *)new;
3580 break;
3581 case 8:
3582 gentry = *(const u64 *)new;
3583 break;
3584 default:
3585 gentry = 0;
3586 break;
3587 }
3588 3660
3589 spin_lock(&vcpu->kvm->mmu_lock); 3661 spin_lock(&vcpu->kvm->mmu_lock);
3590 if (atomic_read(&vcpu->kvm->arch.invlpg_counter) != invlpg_counter)
3591 gentry = 0;
3592 kvm_mmu_free_some_pages(vcpu);
3593 ++vcpu->kvm->stat.mmu_pte_write; 3662 ++vcpu->kvm->stat.mmu_pte_write;
3594 trace_kvm_mmu_audit(vcpu, AUDIT_PRE_PTE_WRITE); 3663 kvm_mmu_audit(vcpu, AUDIT_PRE_PTE_WRITE);
3595 if (guest_initiated) {
3596 kvm_mmu_access_page(vcpu, gfn);
3597 if (gfn == vcpu->arch.last_pt_write_gfn
3598 && !last_updated_pte_accessed(vcpu)) {
3599 ++vcpu->arch.last_pt_write_count;
3600 if (vcpu->arch.last_pt_write_count >= 3)
3601 flooded = 1;
3602 } else {
3603 vcpu->arch.last_pt_write_gfn = gfn;
3604 vcpu->arch.last_pt_write_count = 1;
3605 vcpu->arch.last_pte_updated = NULL;
3606 }
3607 }
3608 3664
3609 mask.cr0_wp = mask.cr4_pae = mask.nxe = 1; 3665 mask.cr0_wp = mask.cr4_pae = mask.nxe = 1;
3610 for_each_gfn_indirect_valid_sp(vcpu->kvm, sp, gfn, node) { 3666 for_each_gfn_indirect_valid_sp(vcpu->kvm, sp, gfn, node) {
3611 pte_size = sp->role.cr4_pae ? 8 : 4; 3667 spte = get_written_sptes(sp, gpa, &npte);
3612 misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1); 3668
3613 misaligned |= bytes < 4; 3669 if (detect_write_misaligned(sp, gpa, bytes) ||
3614 if (misaligned || flooded) { 3670 detect_write_flooding(sp, spte)) {
3615 /*
3616 * Misaligned accesses are too much trouble to fix
3617 * up; also, they usually indicate a page is not used
3618 * as a page table.
3619 *
3620 * If we're seeing too many writes to a page,
3621 * it may no longer be a page table, or we may be
3622 * forking, in which case it is better to unmap the
3623 * page.
3624 */
3625 pgprintk("misaligned: gpa %llx bytes %d role %x\n",
3626 gpa, bytes, sp->role.word);
3627 zap_page |= !!kvm_mmu_prepare_zap_page(vcpu->kvm, sp, 3671 zap_page |= !!kvm_mmu_prepare_zap_page(vcpu->kvm, sp,
3628 &invalid_list); 3672 &invalid_list);
3629 ++vcpu->kvm->stat.mmu_flooded; 3673 ++vcpu->kvm->stat.mmu_flooded;
3630 continue; 3674 continue;
3631 } 3675 }
3632 page_offset = offset; 3676
3633 level = sp->role.level; 3677 spte = get_written_sptes(sp, gpa, &npte);
3634 npte = 1; 3678 if (!spte)
3635 if (!sp->role.cr4_pae) { 3679 continue;
3636 page_offset <<= 1; /* 32->64 */ 3680
3637 /*
3638 * A 32-bit pde maps 4MB while the shadow pdes map
3639 * only 2MB. So we need to double the offset again
3640 * and zap two pdes instead of one.
3641 */
3642 if (level == PT32_ROOT_LEVEL) {
3643 page_offset &= ~7; /* kill rounding error */
3644 page_offset <<= 1;
3645 npte = 2;
3646 }
3647 quadrant = page_offset >> PAGE_SHIFT;
3648 page_offset &= ~PAGE_MASK;
3649 if (quadrant != sp->role.quadrant)
3650 continue;
3651 }
3652 local_flush = true; 3681 local_flush = true;
3653 spte = &sp->spt[page_offset / sizeof(*spte)];
3654 while (npte--) { 3682 while (npte--) {
3655 entry = *spte; 3683 entry = *spte;
3656 mmu_page_zap_pte(vcpu->kvm, sp, spte); 3684 mmu_page_zap_pte(vcpu->kvm, sp, spte);
3657 if (gentry && 3685 if (gentry &&
3658 !((sp->role.word ^ vcpu->arch.mmu.base_role.word) 3686 !((sp->role.word ^ vcpu->arch.mmu.base_role.word)
3659 & mask.word)) 3687 & mask.word) && rmap_can_add(vcpu))
3660 mmu_pte_write_new_pte(vcpu, sp, spte, &gentry); 3688 mmu_pte_write_new_pte(vcpu, sp, spte, &gentry);
3661 if (!remote_flush && need_remote_flush(entry, *spte)) 3689 if (!remote_flush && need_remote_flush(entry, *spte))
3662 remote_flush = true; 3690 remote_flush = true;
@@ -3665,7 +3693,7 @@ void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
3665 } 3693 }
3666 mmu_pte_write_flush_tlb(vcpu, zap_page, remote_flush, local_flush); 3694 mmu_pte_write_flush_tlb(vcpu, zap_page, remote_flush, local_flush);
3667 kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list); 3695 kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list);
3668 trace_kvm_mmu_audit(vcpu, AUDIT_POST_PTE_WRITE); 3696 kvm_mmu_audit(vcpu, AUDIT_POST_PTE_WRITE);
3669 spin_unlock(&vcpu->kvm->mmu_lock); 3697 spin_unlock(&vcpu->kvm->mmu_lock);
3670} 3698}
3671 3699
@@ -3679,9 +3707,8 @@ int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva)
3679 3707
3680 gpa = kvm_mmu_gva_to_gpa_read(vcpu, gva, NULL); 3708 gpa = kvm_mmu_gva_to_gpa_read(vcpu, gva, NULL);
3681 3709
3682 spin_lock(&vcpu->kvm->mmu_lock);
3683 r = kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT); 3710 r = kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT);
3684 spin_unlock(&vcpu->kvm->mmu_lock); 3711
3685 return r; 3712 return r;
3686} 3713}
3687EXPORT_SYMBOL_GPL(kvm_mmu_unprotect_page_virt); 3714EXPORT_SYMBOL_GPL(kvm_mmu_unprotect_page_virt);
@@ -3702,10 +3729,18 @@ void __kvm_mmu_free_some_pages(struct kvm_vcpu *vcpu)
3702 kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list); 3729 kvm_mmu_commit_zap_page(vcpu->kvm, &invalid_list);
3703} 3730}
3704 3731
3732static bool is_mmio_page_fault(struct kvm_vcpu *vcpu, gva_t addr)
3733{
3734 if (vcpu->arch.mmu.direct_map || mmu_is_nested(vcpu))
3735 return vcpu_match_mmio_gpa(vcpu, addr);
3736
3737 return vcpu_match_mmio_gva(vcpu, addr);
3738}
3739
3705int kvm_mmu_page_fault(struct kvm_vcpu *vcpu, gva_t cr2, u32 error_code, 3740int kvm_mmu_page_fault(struct kvm_vcpu *vcpu, gva_t cr2, u32 error_code,
3706 void *insn, int insn_len) 3741 void *insn, int insn_len)
3707{ 3742{
3708 int r; 3743 int r, emulation_type = EMULTYPE_RETRY;
3709 enum emulation_result er; 3744 enum emulation_result er;
3710 3745
3711 r = vcpu->arch.mmu.page_fault(vcpu, cr2, error_code, false); 3746 r = vcpu->arch.mmu.page_fault(vcpu, cr2, error_code, false);
@@ -3717,11 +3752,10 @@ int kvm_mmu_page_fault(struct kvm_vcpu *vcpu, gva_t cr2, u32 error_code,
3717 goto out; 3752 goto out;
3718 } 3753 }
3719 3754
3720 r = mmu_topup_memory_caches(vcpu); 3755 if (is_mmio_page_fault(vcpu, cr2))
3721 if (r) 3756 emulation_type = 0;
3722 goto out;
3723 3757
3724 er = x86_emulate_instruction(vcpu, cr2, 0, insn, insn_len); 3758 er = x86_emulate_instruction(vcpu, cr2, emulation_type, insn, insn_len);
3725 3759
3726 switch (er) { 3760 switch (er) {
3727 case EMULATE_DONE: 3761 case EMULATE_DONE:
@@ -3792,7 +3826,11 @@ static int alloc_mmu_pages(struct kvm_vcpu *vcpu)
3792int kvm_mmu_create(struct kvm_vcpu *vcpu) 3826int kvm_mmu_create(struct kvm_vcpu *vcpu)
3793{ 3827{
3794 ASSERT(vcpu); 3828 ASSERT(vcpu);
3795 ASSERT(!VALID_PAGE(vcpu->arch.mmu.root_hpa)); 3829
3830 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
3831 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
3832 vcpu->arch.mmu.translate_gpa = translate_gpa;
3833 vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
3796 3834
3797 return alloc_mmu_pages(vcpu); 3835 return alloc_mmu_pages(vcpu);
3798} 3836}
@@ -3852,14 +3890,14 @@ restart:
3852 spin_unlock(&kvm->mmu_lock); 3890 spin_unlock(&kvm->mmu_lock);
3853} 3891}
3854 3892
3855static int kvm_mmu_remove_some_alloc_mmu_pages(struct kvm *kvm, 3893static void kvm_mmu_remove_some_alloc_mmu_pages(struct kvm *kvm,
3856 struct list_head *invalid_list) 3894 struct list_head *invalid_list)
3857{ 3895{
3858 struct kvm_mmu_page *page; 3896 struct kvm_mmu_page *page;
3859 3897
3860 page = container_of(kvm->arch.active_mmu_pages.prev, 3898 page = container_of(kvm->arch.active_mmu_pages.prev,
3861 struct kvm_mmu_page, link); 3899 struct kvm_mmu_page, link);
3862 return kvm_mmu_prepare_zap_page(kvm, page, invalid_list); 3900 kvm_mmu_prepare_zap_page(kvm, page, invalid_list);
3863} 3901}
3864 3902
3865static int mmu_shrink(struct shrinker *shrink, struct shrink_control *sc) 3903static int mmu_shrink(struct shrinker *shrink, struct shrink_control *sc)
@@ -3874,15 +3912,15 @@ static int mmu_shrink(struct shrinker *shrink, struct shrink_control *sc)
3874 raw_spin_lock(&kvm_lock); 3912 raw_spin_lock(&kvm_lock);
3875 3913
3876 list_for_each_entry(kvm, &vm_list, vm_list) { 3914 list_for_each_entry(kvm, &vm_list, vm_list) {
3877 int idx, freed_pages; 3915 int idx;
3878 LIST_HEAD(invalid_list); 3916 LIST_HEAD(invalid_list);
3879 3917
3880 idx = srcu_read_lock(&kvm->srcu); 3918 idx = srcu_read_lock(&kvm->srcu);
3881 spin_lock(&kvm->mmu_lock); 3919 spin_lock(&kvm->mmu_lock);
3882 if (!kvm_freed && nr_to_scan > 0 && 3920 if (!kvm_freed && nr_to_scan > 0 &&
3883 kvm->arch.n_used_mmu_pages > 0) { 3921 kvm->arch.n_used_mmu_pages > 0) {
3884 freed_pages = kvm_mmu_remove_some_alloc_mmu_pages(kvm, 3922 kvm_mmu_remove_some_alloc_mmu_pages(kvm,
3885 &invalid_list); 3923 &invalid_list);
3886 kvm_freed = kvm; 3924 kvm_freed = kvm;
3887 } 3925 }
3888 nr_to_scan--; 3926 nr_to_scan--;
@@ -3944,15 +3982,15 @@ nomem:
3944 */ 3982 */
3945unsigned int kvm_mmu_calculate_mmu_pages(struct kvm *kvm) 3983unsigned int kvm_mmu_calculate_mmu_pages(struct kvm *kvm)
3946{ 3984{
3947 int i;
3948 unsigned int nr_mmu_pages; 3985 unsigned int nr_mmu_pages;
3949 unsigned int nr_pages = 0; 3986 unsigned int nr_pages = 0;
3950 struct kvm_memslots *slots; 3987 struct kvm_memslots *slots;
3988 struct kvm_memory_slot *memslot;
3951 3989
3952 slots = kvm_memslots(kvm); 3990 slots = kvm_memslots(kvm);
3953 3991
3954 for (i = 0; i < slots->nmemslots; i++) 3992 kvm_for_each_memslot(memslot, slots)
3955 nr_pages += slots->memslots[i].npages; 3993 nr_pages += memslot->npages;
3956 3994
3957 nr_mmu_pages = nr_pages * KVM_PERMILLE_MMU_PAGES / 1000; 3995 nr_mmu_pages = nr_pages * KVM_PERMILLE_MMU_PAGES / 1000;
3958 nr_mmu_pages = max(nr_mmu_pages, 3996 nr_mmu_pages = max(nr_mmu_pages,
@@ -3961,127 +3999,6 @@ unsigned int kvm_mmu_calculate_mmu_pages(struct kvm *kvm)
3961 return nr_mmu_pages; 3999 return nr_mmu_pages;
3962} 4000}
3963 4001
3964static void *pv_mmu_peek_buffer(struct kvm_pv_mmu_op_buffer *buffer,
3965 unsigned len)
3966{
3967 if (len > buffer->len)
3968 return NULL;
3969 return buffer->ptr;
3970}
3971
3972static void *pv_mmu_read_buffer(struct kvm_pv_mmu_op_buffer *buffer,
3973 unsigned len)
3974{
3975 void *ret;
3976
3977 ret = pv_mmu_peek_buffer(buffer, len);
3978 if (!ret)
3979 return ret;
3980 buffer->ptr += len;
3981 buffer->len -= len;
3982 buffer->processed += len;
3983 return ret;
3984}
3985
3986static int kvm_pv_mmu_write(struct kvm_vcpu *vcpu,
3987 gpa_t addr, gpa_t value)
3988{
3989 int bytes = 8;
3990 int r;
3991
3992 if (!is_long_mode(vcpu) && !is_pae(vcpu))
3993 bytes = 4;
3994
3995 r = mmu_topup_memory_caches(vcpu);
3996 if (r)
3997 return r;
3998
3999 if (!emulator_write_phys(vcpu, addr, &value, bytes))
4000 return -EFAULT;
4001
4002 return 1;
4003}
4004
4005static int kvm_pv_mmu_flush_tlb(struct kvm_vcpu *vcpu)
4006{
4007 (void)kvm_set_cr3(vcpu, kvm_read_cr3(vcpu));
4008 return 1;
4009}
4010
4011static int kvm_pv_mmu_release_pt(struct kvm_vcpu *vcpu, gpa_t addr)
4012{
4013 spin_lock(&vcpu->kvm->mmu_lock);
4014 mmu_unshadow(vcpu->kvm, addr >> PAGE_SHIFT);
4015 spin_unlock(&vcpu->kvm->mmu_lock);
4016 return 1;
4017}
4018
4019static int kvm_pv_mmu_op_one(struct kvm_vcpu *vcpu,
4020 struct kvm_pv_mmu_op_buffer *buffer)
4021{
4022 struct kvm_mmu_op_header *header;
4023
4024 header = pv_mmu_peek_buffer(buffer, sizeof *header);
4025 if (!header)
4026 return 0;
4027 switch (header->op) {
4028 case KVM_MMU_OP_WRITE_PTE: {
4029 struct kvm_mmu_op_write_pte *wpte;
4030
4031 wpte = pv_mmu_read_buffer(buffer, sizeof *wpte);
4032 if (!wpte)
4033 return 0;
4034 return kvm_pv_mmu_write(vcpu, wpte->pte_phys,
4035 wpte->pte_val);
4036 }
4037 case KVM_MMU_OP_FLUSH_TLB: {
4038 struct kvm_mmu_op_flush_tlb *ftlb;
4039
4040 ftlb = pv_mmu_read_buffer(buffer, sizeof *ftlb);
4041 if (!ftlb)
4042 return 0;
4043 return kvm_pv_mmu_flush_tlb(vcpu);
4044 }
4045 case KVM_MMU_OP_RELEASE_PT: {
4046 struct kvm_mmu_op_release_pt *rpt;
4047
4048 rpt = pv_mmu_read_buffer(buffer, sizeof *rpt);
4049 if (!rpt)
4050 return 0;
4051 return kvm_pv_mmu_release_pt(vcpu, rpt->pt_phys);
4052 }
4053 default: return 0;
4054 }
4055}
4056
4057int kvm_pv_mmu_op(struct kvm_vcpu *vcpu, unsigned long bytes,
4058 gpa_t addr, unsigned long *ret)
4059{
4060 int r;
4061 struct kvm_pv_mmu_op_buffer *buffer = &vcpu->arch.mmu_op_buffer;
4062
4063 buffer->ptr = buffer->buf;
4064 buffer->len = min_t(unsigned long, bytes, sizeof buffer->buf);
4065 buffer->processed = 0;
4066
4067 r = kvm_read_guest(vcpu->kvm, addr, buffer->buf, buffer->len);
4068 if (r)
4069 goto out;
4070
4071 while (buffer->len) {
4072 r = kvm_pv_mmu_op_one(vcpu, buffer);
4073 if (r < 0)
4074 goto out;
4075 if (r == 0)
4076 break;
4077 }
4078
4079 r = 1;
4080out:
4081 *ret = buffer->processed;
4082 return r;
4083}
4084
4085int kvm_mmu_get_spte_hierarchy(struct kvm_vcpu *vcpu, u64 addr, u64 sptes[4]) 4002int kvm_mmu_get_spte_hierarchy(struct kvm_vcpu *vcpu, u64 addr, u64 sptes[4])
4086{ 4003{
4087 struct kvm_shadow_walk_iterator iterator; 4004 struct kvm_shadow_walk_iterator iterator;
@@ -4110,12 +4027,6 @@ void kvm_mmu_destroy(struct kvm_vcpu *vcpu)
4110 mmu_free_memory_caches(vcpu); 4027 mmu_free_memory_caches(vcpu);
4111} 4028}
4112 4029
4113#ifdef CONFIG_KVM_MMU_AUDIT
4114#include "mmu_audit.c"
4115#else
4116static void mmu_audit_disable(void) { }
4117#endif
4118
4119void kvm_mmu_module_exit(void) 4030void kvm_mmu_module_exit(void)
4120{ 4031{
4121 mmu_destroy_caches(); 4032 mmu_destroy_caches();
diff --git a/arch/x86/kvm/mmu_audit.c b/arch/x86/kvm/mmu_audit.c
index 746ec259d024..fe15dcc07a6b 100644
--- a/arch/x86/kvm/mmu_audit.c
+++ b/arch/x86/kvm/mmu_audit.c
@@ -19,6 +19,15 @@
19 19
20#include <linux/ratelimit.h> 20#include <linux/ratelimit.h>
21 21
22char const *audit_point_name[] = {
23 "pre page fault",
24 "post page fault",
25 "pre pte write",
26 "post pte write",
27 "pre sync",
28 "post sync"
29};
30
22#define audit_printk(kvm, fmt, args...) \ 31#define audit_printk(kvm, fmt, args...) \
23 printk(KERN_ERR "audit: (%s) error: " \ 32 printk(KERN_ERR "audit: (%s) error: " \
24 fmt, audit_point_name[kvm->arch.audit_point], ##args) 33 fmt, audit_point_name[kvm->arch.audit_point], ##args)
@@ -224,7 +233,10 @@ static void audit_vcpu_spte(struct kvm_vcpu *vcpu)
224 mmu_spte_walk(vcpu, audit_spte); 233 mmu_spte_walk(vcpu, audit_spte);
225} 234}
226 235
227static void kvm_mmu_audit(void *ignore, struct kvm_vcpu *vcpu, int point) 236static bool mmu_audit;
237static struct jump_label_key mmu_audit_key;
238
239static void __kvm_mmu_audit(struct kvm_vcpu *vcpu, int point)
228{ 240{
229 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10); 241 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
230 242
@@ -236,18 +248,18 @@ static void kvm_mmu_audit(void *ignore, struct kvm_vcpu *vcpu, int point)
236 audit_vcpu_spte(vcpu); 248 audit_vcpu_spte(vcpu);
237} 249}
238 250
239static bool mmu_audit; 251static inline void kvm_mmu_audit(struct kvm_vcpu *vcpu, int point)
252{
253 if (static_branch((&mmu_audit_key)))
254 __kvm_mmu_audit(vcpu, point);
255}
240 256
241static void mmu_audit_enable(void) 257static void mmu_audit_enable(void)
242{ 258{
243 int ret;
244
245 if (mmu_audit) 259 if (mmu_audit)
246 return; 260 return;
247 261
248 ret = register_trace_kvm_mmu_audit(kvm_mmu_audit, NULL); 262 jump_label_inc(&mmu_audit_key);
249 WARN_ON(ret);
250
251 mmu_audit = true; 263 mmu_audit = true;
252} 264}
253 265
@@ -256,8 +268,7 @@ static void mmu_audit_disable(void)
256 if (!mmu_audit) 268 if (!mmu_audit)
257 return; 269 return;
258 270
259 unregister_trace_kvm_mmu_audit(kvm_mmu_audit, NULL); 271 jump_label_dec(&mmu_audit_key);
260 tracepoint_synchronize_unregister();
261 mmu_audit = false; 272 mmu_audit = false;
262} 273}
263 274
diff --git a/arch/x86/kvm/mmutrace.h b/arch/x86/kvm/mmutrace.h
index eed67f34146d..89fb0e81322a 100644
--- a/arch/x86/kvm/mmutrace.h
+++ b/arch/x86/kvm/mmutrace.h
@@ -243,25 +243,6 @@ TRACE_EVENT(
243 TP_printk("addr:%llx gfn %llx access %x", __entry->addr, __entry->gfn, 243 TP_printk("addr:%llx gfn %llx access %x", __entry->addr, __entry->gfn,
244 __entry->access) 244 __entry->access)
245); 245);
246
247TRACE_EVENT(
248 kvm_mmu_audit,
249 TP_PROTO(struct kvm_vcpu *vcpu, int audit_point),
250 TP_ARGS(vcpu, audit_point),
251
252 TP_STRUCT__entry(
253 __field(struct kvm_vcpu *, vcpu)
254 __field(int, audit_point)
255 ),
256
257 TP_fast_assign(
258 __entry->vcpu = vcpu;
259 __entry->audit_point = audit_point;
260 ),
261
262 TP_printk("vcpu:%d %s", __entry->vcpu->cpu,
263 audit_point_name[__entry->audit_point])
264);
265#endif /* _TRACE_KVMMMU_H */ 246#endif /* _TRACE_KVMMMU_H */
266 247
267#undef TRACE_INCLUDE_PATH 248#undef TRACE_INCLUDE_PATH
diff --git a/arch/x86/kvm/paging_tmpl.h b/arch/x86/kvm/paging_tmpl.h
index 92994100638b..15610285ebb6 100644
--- a/arch/x86/kvm/paging_tmpl.h
+++ b/arch/x86/kvm/paging_tmpl.h
@@ -497,6 +497,7 @@ static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
497 shadow_walk_next(&it)) { 497 shadow_walk_next(&it)) {
498 gfn_t table_gfn; 498 gfn_t table_gfn;
499 499
500 clear_sp_write_flooding_count(it.sptep);
500 drop_large_spte(vcpu, it.sptep); 501 drop_large_spte(vcpu, it.sptep);
501 502
502 sp = NULL; 503 sp = NULL;
@@ -522,6 +523,7 @@ static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
522 shadow_walk_next(&it)) { 523 shadow_walk_next(&it)) {
523 gfn_t direct_gfn; 524 gfn_t direct_gfn;
524 525
526 clear_sp_write_flooding_count(it.sptep);
525 validate_direct_spte(vcpu, it.sptep, direct_access); 527 validate_direct_spte(vcpu, it.sptep, direct_access);
526 528
527 drop_large_spte(vcpu, it.sptep); 529 drop_large_spte(vcpu, it.sptep);
@@ -536,6 +538,7 @@ static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
536 link_shadow_page(it.sptep, sp); 538 link_shadow_page(it.sptep, sp);
537 } 539 }
538 540
541 clear_sp_write_flooding_count(it.sptep);
539 mmu_set_spte(vcpu, it.sptep, access, gw->pte_access, 542 mmu_set_spte(vcpu, it.sptep, access, gw->pte_access,
540 user_fault, write_fault, emulate, it.level, 543 user_fault, write_fault, emulate, it.level,
541 gw->gfn, pfn, prefault, map_writable); 544 gw->gfn, pfn, prefault, map_writable);
@@ -599,11 +602,9 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr, u32 error_code,
599 */ 602 */
600 if (!r) { 603 if (!r) {
601 pgprintk("%s: guest page fault\n", __func__); 604 pgprintk("%s: guest page fault\n", __func__);
602 if (!prefault) { 605 if (!prefault)
603 inject_page_fault(vcpu, &walker.fault); 606 inject_page_fault(vcpu, &walker.fault);
604 /* reset fork detector */ 607
605 vcpu->arch.last_pt_write_count = 0;
606 }
607 return 0; 608 return 0;
608 } 609 }
609 610
@@ -631,7 +632,7 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr, u32 error_code,
631 if (mmu_notifier_retry(vcpu, mmu_seq)) 632 if (mmu_notifier_retry(vcpu, mmu_seq))
632 goto out_unlock; 633 goto out_unlock;
633 634
634 trace_kvm_mmu_audit(vcpu, AUDIT_PRE_PAGE_FAULT); 635 kvm_mmu_audit(vcpu, AUDIT_PRE_PAGE_FAULT);
635 kvm_mmu_free_some_pages(vcpu); 636 kvm_mmu_free_some_pages(vcpu);
636 if (!force_pt_level) 637 if (!force_pt_level)
637 transparent_hugepage_adjust(vcpu, &walker.gfn, &pfn, &level); 638 transparent_hugepage_adjust(vcpu, &walker.gfn, &pfn, &level);
@@ -641,11 +642,8 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr, u32 error_code,
641 pgprintk("%s: shadow pte %p %llx emulate %d\n", __func__, 642 pgprintk("%s: shadow pte %p %llx emulate %d\n", __func__,
642 sptep, *sptep, emulate); 643 sptep, *sptep, emulate);
643 644
644 if (!emulate)
645 vcpu->arch.last_pt_write_count = 0; /* reset fork detector */
646
647 ++vcpu->stat.pf_fixed; 645 ++vcpu->stat.pf_fixed;
648 trace_kvm_mmu_audit(vcpu, AUDIT_POST_PAGE_FAULT); 646 kvm_mmu_audit(vcpu, AUDIT_POST_PAGE_FAULT);
649 spin_unlock(&vcpu->kvm->mmu_lock); 647 spin_unlock(&vcpu->kvm->mmu_lock);
650 648
651 return emulate; 649 return emulate;
@@ -656,65 +654,66 @@ out_unlock:
656 return 0; 654 return 0;
657} 655}
658 656
657static gpa_t FNAME(get_level1_sp_gpa)(struct kvm_mmu_page *sp)
658{
659 int offset = 0;
660
661 WARN_ON(sp->role.level != 1);
662
663 if (PTTYPE == 32)
664 offset = sp->role.quadrant << PT64_LEVEL_BITS;
665
666 return gfn_to_gpa(sp->gfn) + offset * sizeof(pt_element_t);
667}
668
659static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva) 669static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva)
660{ 670{
661 struct kvm_shadow_walk_iterator iterator; 671 struct kvm_shadow_walk_iterator iterator;
662 struct kvm_mmu_page *sp; 672 struct kvm_mmu_page *sp;
663 gpa_t pte_gpa = -1;
664 int level; 673 int level;
665 u64 *sptep; 674 u64 *sptep;
666 int need_flush = 0;
667 675
668 vcpu_clear_mmio_info(vcpu, gva); 676 vcpu_clear_mmio_info(vcpu, gva);
669 677
670 spin_lock(&vcpu->kvm->mmu_lock); 678 /*
679 * No need to check return value here, rmap_can_add() can
680 * help us to skip pte prefetch later.
681 */
682 mmu_topup_memory_caches(vcpu);
671 683
684 spin_lock(&vcpu->kvm->mmu_lock);
672 for_each_shadow_entry(vcpu, gva, iterator) { 685 for_each_shadow_entry(vcpu, gva, iterator) {
673 level = iterator.level; 686 level = iterator.level;
674 sptep = iterator.sptep; 687 sptep = iterator.sptep;
675 688
676 sp = page_header(__pa(sptep)); 689 sp = page_header(__pa(sptep));
677 if (is_last_spte(*sptep, level)) { 690 if (is_last_spte(*sptep, level)) {
678 int offset, shift; 691 pt_element_t gpte;
692 gpa_t pte_gpa;
679 693
680 if (!sp->unsync) 694 if (!sp->unsync)
681 break; 695 break;
682 696
683 shift = PAGE_SHIFT - 697 pte_gpa = FNAME(get_level1_sp_gpa)(sp);
684 (PT_LEVEL_BITS - PT64_LEVEL_BITS) * level;
685 offset = sp->role.quadrant << shift;
686
687 pte_gpa = (sp->gfn << PAGE_SHIFT) + offset;
688 pte_gpa += (sptep - sp->spt) * sizeof(pt_element_t); 698 pte_gpa += (sptep - sp->spt) * sizeof(pt_element_t);
689 699
690 if (is_shadow_present_pte(*sptep)) { 700 if (mmu_page_zap_pte(vcpu->kvm, sp, sptep))
691 if (is_large_pte(*sptep)) 701 kvm_flush_remote_tlbs(vcpu->kvm);
692 --vcpu->kvm->stat.lpages;
693 drop_spte(vcpu->kvm, sptep);
694 need_flush = 1;
695 } else if (is_mmio_spte(*sptep))
696 mmu_spte_clear_no_track(sptep);
697 702
698 break; 703 if (!rmap_can_add(vcpu))
704 break;
705
706 if (kvm_read_guest_atomic(vcpu->kvm, pte_gpa, &gpte,
707 sizeof(pt_element_t)))
708 break;
709
710 FNAME(update_pte)(vcpu, sp, sptep, &gpte);
699 } 711 }
700 712
701 if (!is_shadow_present_pte(*sptep) || !sp->unsync_children) 713 if (!is_shadow_present_pte(*sptep) || !sp->unsync_children)
702 break; 714 break;
703 } 715 }
704
705 if (need_flush)
706 kvm_flush_remote_tlbs(vcpu->kvm);
707
708 atomic_inc(&vcpu->kvm->arch.invlpg_counter);
709
710 spin_unlock(&vcpu->kvm->mmu_lock); 716 spin_unlock(&vcpu->kvm->mmu_lock);
711
712 if (pte_gpa == -1)
713 return;
714
715 if (mmu_topup_memory_caches(vcpu))
716 return;
717 kvm_mmu_pte_write(vcpu, pte_gpa, NULL, sizeof(pt_element_t), 0);
718} 717}
719 718
720static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr, u32 access, 719static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr, u32 access,
@@ -769,19 +768,14 @@ static gpa_t FNAME(gva_to_gpa_nested)(struct kvm_vcpu *vcpu, gva_t vaddr,
769 */ 768 */
770static int FNAME(sync_page)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp) 769static int FNAME(sync_page)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
771{ 770{
772 int i, offset, nr_present; 771 int i, nr_present = 0;
773 bool host_writable; 772 bool host_writable;
774 gpa_t first_pte_gpa; 773 gpa_t first_pte_gpa;
775 774
776 offset = nr_present = 0;
777
778 /* direct kvm_mmu_page can not be unsync. */ 775 /* direct kvm_mmu_page can not be unsync. */
779 BUG_ON(sp->role.direct); 776 BUG_ON(sp->role.direct);
780 777
781 if (PTTYPE == 32) 778 first_pte_gpa = FNAME(get_level1_sp_gpa)(sp);
782 offset = sp->role.quadrant << PT64_LEVEL_BITS;
783
784 first_pte_gpa = gfn_to_gpa(sp->gfn) + offset * sizeof(pt_element_t);
785 779
786 for (i = 0; i < PT64_ENT_PER_PAGE; i++) { 780 for (i = 0; i < PT64_ENT_PER_PAGE; i++) {
787 unsigned pte_access; 781 unsigned pte_access;
diff --git a/arch/x86/kvm/pmu.c b/arch/x86/kvm/pmu.c
new file mode 100644
index 000000000000..7aad5446f393
--- /dev/null
+++ b/arch/x86/kvm/pmu.c
@@ -0,0 +1,533 @@
1/*
2 * Kernel-based Virtual Machine -- Performane Monitoring Unit support
3 *
4 * Copyright 2011 Red Hat, Inc. and/or its affiliates.
5 *
6 * Authors:
7 * Avi Kivity <avi@redhat.com>
8 * Gleb Natapov <gleb@redhat.com>
9 *
10 * This work is licensed under the terms of the GNU GPL, version 2. See
11 * the COPYING file in the top-level directory.
12 *
13 */
14
15#include <linux/types.h>
16#include <linux/kvm_host.h>
17#include <linux/perf_event.h>
18#include "x86.h"
19#include "cpuid.h"
20#include "lapic.h"
21
22static struct kvm_arch_event_perf_mapping {
23 u8 eventsel;
24 u8 unit_mask;
25 unsigned event_type;
26 bool inexact;
27} arch_events[] = {
28 /* Index must match CPUID 0x0A.EBX bit vector */
29 [0] = { 0x3c, 0x00, PERF_COUNT_HW_CPU_CYCLES },
30 [1] = { 0xc0, 0x00, PERF_COUNT_HW_INSTRUCTIONS },
31 [2] = { 0x3c, 0x01, PERF_COUNT_HW_BUS_CYCLES },
32 [3] = { 0x2e, 0x4f, PERF_COUNT_HW_CACHE_REFERENCES },
33 [4] = { 0x2e, 0x41, PERF_COUNT_HW_CACHE_MISSES },
34 [5] = { 0xc4, 0x00, PERF_COUNT_HW_BRANCH_INSTRUCTIONS },
35 [6] = { 0xc5, 0x00, PERF_COUNT_HW_BRANCH_MISSES },
36};
37
38/* mapping between fixed pmc index and arch_events array */
39int fixed_pmc_events[] = {1, 0, 2};
40
41static bool pmc_is_gp(struct kvm_pmc *pmc)
42{
43 return pmc->type == KVM_PMC_GP;
44}
45
46static inline u64 pmc_bitmask(struct kvm_pmc *pmc)
47{
48 struct kvm_pmu *pmu = &pmc->vcpu->arch.pmu;
49
50 return pmu->counter_bitmask[pmc->type];
51}
52
53static inline bool pmc_enabled(struct kvm_pmc *pmc)
54{
55 struct kvm_pmu *pmu = &pmc->vcpu->arch.pmu;
56 return test_bit(pmc->idx, (unsigned long *)&pmu->global_ctrl);
57}
58
59static inline struct kvm_pmc *get_gp_pmc(struct kvm_pmu *pmu, u32 msr,
60 u32 base)
61{
62 if (msr >= base && msr < base + pmu->nr_arch_gp_counters)
63 return &pmu->gp_counters[msr - base];
64 return NULL;
65}
66
67static inline struct kvm_pmc *get_fixed_pmc(struct kvm_pmu *pmu, u32 msr)
68{
69 int base = MSR_CORE_PERF_FIXED_CTR0;
70 if (msr >= base && msr < base + pmu->nr_arch_fixed_counters)
71 return &pmu->fixed_counters[msr - base];
72 return NULL;
73}
74
75static inline struct kvm_pmc *get_fixed_pmc_idx(struct kvm_pmu *pmu, int idx)
76{
77 return get_fixed_pmc(pmu, MSR_CORE_PERF_FIXED_CTR0 + idx);
78}
79
80static struct kvm_pmc *global_idx_to_pmc(struct kvm_pmu *pmu, int idx)
81{
82 if (idx < X86_PMC_IDX_FIXED)
83 return get_gp_pmc(pmu, MSR_P6_EVNTSEL0 + idx, MSR_P6_EVNTSEL0);
84 else
85 return get_fixed_pmc_idx(pmu, idx - X86_PMC_IDX_FIXED);
86}
87
88void kvm_deliver_pmi(struct kvm_vcpu *vcpu)
89{
90 if (vcpu->arch.apic)
91 kvm_apic_local_deliver(vcpu->arch.apic, APIC_LVTPC);
92}
93
94static void trigger_pmi(struct irq_work *irq_work)
95{
96 struct kvm_pmu *pmu = container_of(irq_work, struct kvm_pmu,
97 irq_work);
98 struct kvm_vcpu *vcpu = container_of(pmu, struct kvm_vcpu,
99 arch.pmu);
100
101 kvm_deliver_pmi(vcpu);
102}
103
104static void kvm_perf_overflow(struct perf_event *perf_event,
105 struct perf_sample_data *data,
106 struct pt_regs *regs)
107{
108 struct kvm_pmc *pmc = perf_event->overflow_handler_context;
109 struct kvm_pmu *pmu = &pmc->vcpu->arch.pmu;
110 __set_bit(pmc->idx, (unsigned long *)&pmu->global_status);
111}
112
113static void kvm_perf_overflow_intr(struct perf_event *perf_event,
114 struct perf_sample_data *data, struct pt_regs *regs)
115{
116 struct kvm_pmc *pmc = perf_event->overflow_handler_context;
117 struct kvm_pmu *pmu = &pmc->vcpu->arch.pmu;
118 if (!test_and_set_bit(pmc->idx, (unsigned long *)&pmu->reprogram_pmi)) {
119 kvm_perf_overflow(perf_event, data, regs);
120 kvm_make_request(KVM_REQ_PMU, pmc->vcpu);
121 /*
122 * Inject PMI. If vcpu was in a guest mode during NMI PMI
123 * can be ejected on a guest mode re-entry. Otherwise we can't
124 * be sure that vcpu wasn't executing hlt instruction at the
125 * time of vmexit and is not going to re-enter guest mode until,
126 * woken up. So we should wake it, but this is impossible from
127 * NMI context. Do it from irq work instead.
128 */
129 if (!kvm_is_in_guest())
130 irq_work_queue(&pmc->vcpu->arch.pmu.irq_work);
131 else
132 kvm_make_request(KVM_REQ_PMI, pmc->vcpu);
133 }
134}
135
136static u64 read_pmc(struct kvm_pmc *pmc)
137{
138 u64 counter, enabled, running;
139
140 counter = pmc->counter;
141
142 if (pmc->perf_event)
143 counter += perf_event_read_value(pmc->perf_event,
144 &enabled, &running);
145
146 /* FIXME: Scaling needed? */
147
148 return counter & pmc_bitmask(pmc);
149}
150
151static void stop_counter(struct kvm_pmc *pmc)
152{
153 if (pmc->perf_event) {
154 pmc->counter = read_pmc(pmc);
155 perf_event_release_kernel(pmc->perf_event);
156 pmc->perf_event = NULL;
157 }
158}
159
160static void reprogram_counter(struct kvm_pmc *pmc, u32 type,
161 unsigned config, bool exclude_user, bool exclude_kernel,
162 bool intr)
163{
164 struct perf_event *event;
165 struct perf_event_attr attr = {
166 .type = type,
167 .size = sizeof(attr),
168 .pinned = true,
169 .exclude_idle = true,
170 .exclude_host = 1,
171 .exclude_user = exclude_user,
172 .exclude_kernel = exclude_kernel,
173 .config = config,
174 };
175
176 attr.sample_period = (-pmc->counter) & pmc_bitmask(pmc);
177
178 event = perf_event_create_kernel_counter(&attr, -1, current,
179 intr ? kvm_perf_overflow_intr :
180 kvm_perf_overflow, pmc);
181 if (IS_ERR(event)) {
182 printk_once("kvm: pmu event creation failed %ld\n",
183 PTR_ERR(event));
184 return;
185 }
186
187 pmc->perf_event = event;
188 clear_bit(pmc->idx, (unsigned long*)&pmc->vcpu->arch.pmu.reprogram_pmi);
189}
190
191static unsigned find_arch_event(struct kvm_pmu *pmu, u8 event_select,
192 u8 unit_mask)
193{
194 int i;
195
196 for (i = 0; i < ARRAY_SIZE(arch_events); i++)
197 if (arch_events[i].eventsel == event_select
198 && arch_events[i].unit_mask == unit_mask
199 && (pmu->available_event_types & (1 << i)))
200 break;
201
202 if (i == ARRAY_SIZE(arch_events))
203 return PERF_COUNT_HW_MAX;
204
205 return arch_events[i].event_type;
206}
207
208static void reprogram_gp_counter(struct kvm_pmc *pmc, u64 eventsel)
209{
210 unsigned config, type = PERF_TYPE_RAW;
211 u8 event_select, unit_mask;
212
213 pmc->eventsel = eventsel;
214
215 stop_counter(pmc);
216
217 if (!(eventsel & ARCH_PERFMON_EVENTSEL_ENABLE) || !pmc_enabled(pmc))
218 return;
219
220 event_select = eventsel & ARCH_PERFMON_EVENTSEL_EVENT;
221 unit_mask = (eventsel & ARCH_PERFMON_EVENTSEL_UMASK) >> 8;
222
223 if (!(event_select & (ARCH_PERFMON_EVENTSEL_EDGE |
224 ARCH_PERFMON_EVENTSEL_INV |
225 ARCH_PERFMON_EVENTSEL_CMASK))) {
226 config = find_arch_event(&pmc->vcpu->arch.pmu, event_select,
227 unit_mask);
228 if (config != PERF_COUNT_HW_MAX)
229 type = PERF_TYPE_HARDWARE;
230 }
231
232 if (type == PERF_TYPE_RAW)
233 config = eventsel & X86_RAW_EVENT_MASK;
234
235 reprogram_counter(pmc, type, config,
236 !(eventsel & ARCH_PERFMON_EVENTSEL_USR),
237 !(eventsel & ARCH_PERFMON_EVENTSEL_OS),
238 eventsel & ARCH_PERFMON_EVENTSEL_INT);
239}
240
241static void reprogram_fixed_counter(struct kvm_pmc *pmc, u8 en_pmi, int idx)
242{
243 unsigned en = en_pmi & 0x3;
244 bool pmi = en_pmi & 0x8;
245
246 stop_counter(pmc);
247
248 if (!en || !pmc_enabled(pmc))
249 return;
250
251 reprogram_counter(pmc, PERF_TYPE_HARDWARE,
252 arch_events[fixed_pmc_events[idx]].event_type,
253 !(en & 0x2), /* exclude user */
254 !(en & 0x1), /* exclude kernel */
255 pmi);
256}
257
258static inline u8 fixed_en_pmi(u64 ctrl, int idx)
259{
260 return (ctrl >> (idx * 4)) & 0xf;
261}
262
263static void reprogram_fixed_counters(struct kvm_pmu *pmu, u64 data)
264{
265 int i;
266
267 for (i = 0; i < pmu->nr_arch_fixed_counters; i++) {
268 u8 en_pmi = fixed_en_pmi(data, i);
269 struct kvm_pmc *pmc = get_fixed_pmc_idx(pmu, i);
270
271 if (fixed_en_pmi(pmu->fixed_ctr_ctrl, i) == en_pmi)
272 continue;
273
274 reprogram_fixed_counter(pmc, en_pmi, i);
275 }
276
277 pmu->fixed_ctr_ctrl = data;
278}
279
280static void reprogram_idx(struct kvm_pmu *pmu, int idx)
281{
282 struct kvm_pmc *pmc = global_idx_to_pmc(pmu, idx);
283
284 if (!pmc)
285 return;
286
287 if (pmc_is_gp(pmc))
288 reprogram_gp_counter(pmc, pmc->eventsel);
289 else {
290 int fidx = idx - X86_PMC_IDX_FIXED;
291 reprogram_fixed_counter(pmc,
292 fixed_en_pmi(pmu->fixed_ctr_ctrl, fidx), fidx);
293 }
294}
295
296static void global_ctrl_changed(struct kvm_pmu *pmu, u64 data)
297{
298 int bit;
299 u64 diff = pmu->global_ctrl ^ data;
300
301 pmu->global_ctrl = data;
302
303 for_each_set_bit(bit, (unsigned long *)&diff, X86_PMC_IDX_MAX)
304 reprogram_idx(pmu, bit);
305}
306
307bool kvm_pmu_msr(struct kvm_vcpu *vcpu, u32 msr)
308{
309 struct kvm_pmu *pmu = &vcpu->arch.pmu;
310 int ret;
311
312 switch (msr) {
313 case MSR_CORE_PERF_FIXED_CTR_CTRL:
314 case MSR_CORE_PERF_GLOBAL_STATUS:
315 case MSR_CORE_PERF_GLOBAL_CTRL:
316 case MSR_CORE_PERF_GLOBAL_OVF_CTRL:
317 ret = pmu->version > 1;
318 break;
319 default:
320 ret = get_gp_pmc(pmu, msr, MSR_IA32_PERFCTR0)
321 || get_gp_pmc(pmu, msr, MSR_P6_EVNTSEL0)
322 || get_fixed_pmc(pmu, msr);
323 break;
324 }
325 return ret;
326}
327
328int kvm_pmu_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
329{
330 struct kvm_pmu *pmu = &vcpu->arch.pmu;
331 struct kvm_pmc *pmc;
332
333 switch (index) {
334 case MSR_CORE_PERF_FIXED_CTR_CTRL:
335 *data = pmu->fixed_ctr_ctrl;
336 return 0;
337 case MSR_CORE_PERF_GLOBAL_STATUS:
338 *data = pmu->global_status;
339 return 0;
340 case MSR_CORE_PERF_GLOBAL_CTRL:
341 *data = pmu->global_ctrl;
342 return 0;
343 case MSR_CORE_PERF_GLOBAL_OVF_CTRL:
344 *data = pmu->global_ovf_ctrl;
345 return 0;
346 default:
347 if ((pmc = get_gp_pmc(pmu, index, MSR_IA32_PERFCTR0)) ||
348 (pmc = get_fixed_pmc(pmu, index))) {
349 *data = read_pmc(pmc);
350 return 0;
351 } else if ((pmc = get_gp_pmc(pmu, index, MSR_P6_EVNTSEL0))) {
352 *data = pmc->eventsel;
353 return 0;
354 }
355 }
356 return 1;
357}
358
359int kvm_pmu_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
360{
361 struct kvm_pmu *pmu = &vcpu->arch.pmu;
362 struct kvm_pmc *pmc;
363
364 switch (index) {
365 case MSR_CORE_PERF_FIXED_CTR_CTRL:
366 if (pmu->fixed_ctr_ctrl == data)
367 return 0;
368 if (!(data & 0xfffffffffffff444)) {
369 reprogram_fixed_counters(pmu, data);
370 return 0;
371 }
372 break;
373 case MSR_CORE_PERF_GLOBAL_STATUS:
374 break; /* RO MSR */
375 case MSR_CORE_PERF_GLOBAL_CTRL:
376 if (pmu->global_ctrl == data)
377 return 0;
378 if (!(data & pmu->global_ctrl_mask)) {
379 global_ctrl_changed(pmu, data);
380 return 0;
381 }
382 break;
383 case MSR_CORE_PERF_GLOBAL_OVF_CTRL:
384 if (!(data & (pmu->global_ctrl_mask & ~(3ull<<62)))) {
385 pmu->global_status &= ~data;
386 pmu->global_ovf_ctrl = data;
387 return 0;
388 }
389 break;
390 default:
391 if ((pmc = get_gp_pmc(pmu, index, MSR_IA32_PERFCTR0)) ||
392 (pmc = get_fixed_pmc(pmu, index))) {
393 data = (s64)(s32)data;
394 pmc->counter += data - read_pmc(pmc);
395 return 0;
396 } else if ((pmc = get_gp_pmc(pmu, index, MSR_P6_EVNTSEL0))) {
397 if (data == pmc->eventsel)
398 return 0;
399 if (!(data & 0xffffffff00200000ull)) {
400 reprogram_gp_counter(pmc, data);
401 return 0;
402 }
403 }
404 }
405 return 1;
406}
407
408int kvm_pmu_read_pmc(struct kvm_vcpu *vcpu, unsigned pmc, u64 *data)
409{
410 struct kvm_pmu *pmu = &vcpu->arch.pmu;
411 bool fast_mode = pmc & (1u << 31);
412 bool fixed = pmc & (1u << 30);
413 struct kvm_pmc *counters;
414 u64 ctr;
415
416 pmc &= (3u << 30) - 1;
417 if (!fixed && pmc >= pmu->nr_arch_gp_counters)
418 return 1;
419 if (fixed && pmc >= pmu->nr_arch_fixed_counters)
420 return 1;
421 counters = fixed ? pmu->fixed_counters : pmu->gp_counters;
422 ctr = read_pmc(&counters[pmc]);
423 if (fast_mode)
424 ctr = (u32)ctr;
425 *data = ctr;
426
427 return 0;
428}
429
430void kvm_pmu_cpuid_update(struct kvm_vcpu *vcpu)
431{
432 struct kvm_pmu *pmu = &vcpu->arch.pmu;
433 struct kvm_cpuid_entry2 *entry;
434 unsigned bitmap_len;
435
436 pmu->nr_arch_gp_counters = 0;
437 pmu->nr_arch_fixed_counters = 0;
438 pmu->counter_bitmask[KVM_PMC_GP] = 0;
439 pmu->counter_bitmask[KVM_PMC_FIXED] = 0;
440 pmu->version = 0;
441
442 entry = kvm_find_cpuid_entry(vcpu, 0xa, 0);
443 if (!entry)
444 return;
445
446 pmu->version = entry->eax & 0xff;
447 if (!pmu->version)
448 return;
449
450 pmu->nr_arch_gp_counters = min((int)(entry->eax >> 8) & 0xff,
451 X86_PMC_MAX_GENERIC);
452 pmu->counter_bitmask[KVM_PMC_GP] =
453 ((u64)1 << ((entry->eax >> 16) & 0xff)) - 1;
454 bitmap_len = (entry->eax >> 24) & 0xff;
455 pmu->available_event_types = ~entry->ebx & ((1ull << bitmap_len) - 1);
456
457 if (pmu->version == 1) {
458 pmu->global_ctrl = (1 << pmu->nr_arch_gp_counters) - 1;
459 return;
460 }
461
462 pmu->nr_arch_fixed_counters = min((int)(entry->edx & 0x1f),
463 X86_PMC_MAX_FIXED);
464 pmu->counter_bitmask[KVM_PMC_FIXED] =
465 ((u64)1 << ((entry->edx >> 5) & 0xff)) - 1;
466 pmu->global_ctrl_mask = ~(((1 << pmu->nr_arch_gp_counters) - 1)
467 | (((1ull << pmu->nr_arch_fixed_counters) - 1)
468 << X86_PMC_IDX_FIXED));
469}
470
471void kvm_pmu_init(struct kvm_vcpu *vcpu)
472{
473 int i;
474 struct kvm_pmu *pmu = &vcpu->arch.pmu;
475
476 memset(pmu, 0, sizeof(*pmu));
477 for (i = 0; i < X86_PMC_MAX_GENERIC; i++) {
478 pmu->gp_counters[i].type = KVM_PMC_GP;
479 pmu->gp_counters[i].vcpu = vcpu;
480 pmu->gp_counters[i].idx = i;
481 }
482 for (i = 0; i < X86_PMC_MAX_FIXED; i++) {
483 pmu->fixed_counters[i].type = KVM_PMC_FIXED;
484 pmu->fixed_counters[i].vcpu = vcpu;
485 pmu->fixed_counters[i].idx = i + X86_PMC_IDX_FIXED;
486 }
487 init_irq_work(&pmu->irq_work, trigger_pmi);
488 kvm_pmu_cpuid_update(vcpu);
489}
490
491void kvm_pmu_reset(struct kvm_vcpu *vcpu)
492{
493 struct kvm_pmu *pmu = &vcpu->arch.pmu;
494 int i;
495
496 irq_work_sync(&pmu->irq_work);
497 for (i = 0; i < X86_PMC_MAX_GENERIC; i++) {
498 struct kvm_pmc *pmc = &pmu->gp_counters[i];
499 stop_counter(pmc);
500 pmc->counter = pmc->eventsel = 0;
501 }
502
503 for (i = 0; i < X86_PMC_MAX_FIXED; i++)
504 stop_counter(&pmu->fixed_counters[i]);
505
506 pmu->fixed_ctr_ctrl = pmu->global_ctrl = pmu->global_status =
507 pmu->global_ovf_ctrl = 0;
508}
509
510void kvm_pmu_destroy(struct kvm_vcpu *vcpu)
511{
512 kvm_pmu_reset(vcpu);
513}
514
515void kvm_handle_pmu_event(struct kvm_vcpu *vcpu)
516{
517 struct kvm_pmu *pmu = &vcpu->arch.pmu;
518 u64 bitmask;
519 int bit;
520
521 bitmask = pmu->reprogram_pmi;
522
523 for_each_set_bit(bit, (unsigned long *)&bitmask, X86_PMC_IDX_MAX) {
524 struct kvm_pmc *pmc = global_idx_to_pmc(pmu, bit);
525
526 if (unlikely(!pmc || !pmc->perf_event)) {
527 clear_bit(bit, (unsigned long *)&pmu->reprogram_pmi);
528 continue;
529 }
530
531 reprogram_idx(pmu, bit);
532 }
533}
diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index e32243eac2f4..5fa553babe56 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -1014,6 +1014,7 @@ static void init_vmcb(struct vcpu_svm *svm)
1014 set_intercept(svm, INTERCEPT_NMI); 1014 set_intercept(svm, INTERCEPT_NMI);
1015 set_intercept(svm, INTERCEPT_SMI); 1015 set_intercept(svm, INTERCEPT_SMI);
1016 set_intercept(svm, INTERCEPT_SELECTIVE_CR0); 1016 set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1017 set_intercept(svm, INTERCEPT_RDPMC);
1017 set_intercept(svm, INTERCEPT_CPUID); 1018 set_intercept(svm, INTERCEPT_CPUID);
1018 set_intercept(svm, INTERCEPT_INVD); 1019 set_intercept(svm, INTERCEPT_INVD);
1019 set_intercept(svm, INTERCEPT_HLT); 1020 set_intercept(svm, INTERCEPT_HLT);
@@ -2770,6 +2771,19 @@ static int emulate_on_interception(struct vcpu_svm *svm)
2770 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE; 2771 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
2771} 2772}
2772 2773
2774static int rdpmc_interception(struct vcpu_svm *svm)
2775{
2776 int err;
2777
2778 if (!static_cpu_has(X86_FEATURE_NRIPS))
2779 return emulate_on_interception(svm);
2780
2781 err = kvm_rdpmc(&svm->vcpu);
2782 kvm_complete_insn_gp(&svm->vcpu, err);
2783
2784 return 1;
2785}
2786
2773bool check_selective_cr0_intercepted(struct vcpu_svm *svm, unsigned long val) 2787bool check_selective_cr0_intercepted(struct vcpu_svm *svm, unsigned long val)
2774{ 2788{
2775 unsigned long cr0 = svm->vcpu.arch.cr0; 2789 unsigned long cr0 = svm->vcpu.arch.cr0;
@@ -3190,6 +3204,7 @@ static int (*svm_exit_handlers[])(struct vcpu_svm *svm) = {
3190 [SVM_EXIT_SMI] = nop_on_interception, 3204 [SVM_EXIT_SMI] = nop_on_interception,
3191 [SVM_EXIT_INIT] = nop_on_interception, 3205 [SVM_EXIT_INIT] = nop_on_interception,
3192 [SVM_EXIT_VINTR] = interrupt_window_interception, 3206 [SVM_EXIT_VINTR] = interrupt_window_interception,
3207 [SVM_EXIT_RDPMC] = rdpmc_interception,
3193 [SVM_EXIT_CPUID] = cpuid_interception, 3208 [SVM_EXIT_CPUID] = cpuid_interception,
3194 [SVM_EXIT_IRET] = iret_interception, 3209 [SVM_EXIT_IRET] = iret_interception,
3195 [SVM_EXIT_INVD] = emulate_on_interception, 3210 [SVM_EXIT_INVD] = emulate_on_interception,
diff --git a/arch/x86/kvm/timer.c b/arch/x86/kvm/timer.c
index ae432ea1cd83..6b85cc647f34 100644
--- a/arch/x86/kvm/timer.c
+++ b/arch/x86/kvm/timer.c
@@ -18,9 +18,10 @@
18#include <linux/atomic.h> 18#include <linux/atomic.h>
19#include "kvm_timer.h" 19#include "kvm_timer.h"
20 20
21static int __kvm_timer_fn(struct kvm_vcpu *vcpu, struct kvm_timer *ktimer) 21enum hrtimer_restart kvm_timer_fn(struct hrtimer *data)
22{ 22{
23 int restart_timer = 0; 23 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
24 struct kvm_vcpu *vcpu = ktimer->vcpu;
24 wait_queue_head_t *q = &vcpu->wq; 25 wait_queue_head_t *q = &vcpu->wq;
25 26
26 /* 27 /*
@@ -40,26 +41,7 @@ static int __kvm_timer_fn(struct kvm_vcpu *vcpu, struct kvm_timer *ktimer)
40 41
41 if (ktimer->t_ops->is_periodic(ktimer)) { 42 if (ktimer->t_ops->is_periodic(ktimer)) {
42 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period); 43 hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
43 restart_timer = 1;
44 }
45
46 return restart_timer;
47}
48
49enum hrtimer_restart kvm_timer_fn(struct hrtimer *data)
50{
51 int restart_timer;
52 struct kvm_vcpu *vcpu;
53 struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
54
55 vcpu = ktimer->vcpu;
56 if (!vcpu)
57 return HRTIMER_NORESTART;
58
59 restart_timer = __kvm_timer_fn(vcpu, ktimer);
60 if (restart_timer)
61 return HRTIMER_RESTART; 44 return HRTIMER_RESTART;
62 else 45 } else
63 return HRTIMER_NORESTART; 46 return HRTIMER_NORESTART;
64} 47}
65
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index 579a0b51696a..d29216c462b3 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -18,6 +18,7 @@
18 18
19#include "irq.h" 19#include "irq.h"
20#include "mmu.h" 20#include "mmu.h"
21#include "cpuid.h"
21 22
22#include <linux/kvm_host.h> 23#include <linux/kvm_host.h>
23#include <linux/module.h> 24#include <linux/module.h>
@@ -50,29 +51,29 @@
50MODULE_AUTHOR("Qumranet"); 51MODULE_AUTHOR("Qumranet");
51MODULE_LICENSE("GPL"); 52MODULE_LICENSE("GPL");
52 53
53static int __read_mostly enable_vpid = 1; 54static bool __read_mostly enable_vpid = 1;
54module_param_named(vpid, enable_vpid, bool, 0444); 55module_param_named(vpid, enable_vpid, bool, 0444);
55 56
56static int __read_mostly flexpriority_enabled = 1; 57static bool __read_mostly flexpriority_enabled = 1;
57module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO); 58module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
58 59
59static int __read_mostly enable_ept = 1; 60static bool __read_mostly enable_ept = 1;
60module_param_named(ept, enable_ept, bool, S_IRUGO); 61module_param_named(ept, enable_ept, bool, S_IRUGO);
61 62
62static int __read_mostly enable_unrestricted_guest = 1; 63static bool __read_mostly enable_unrestricted_guest = 1;
63module_param_named(unrestricted_guest, 64module_param_named(unrestricted_guest,
64 enable_unrestricted_guest, bool, S_IRUGO); 65 enable_unrestricted_guest, bool, S_IRUGO);
65 66
66static int __read_mostly emulate_invalid_guest_state = 0; 67static bool __read_mostly emulate_invalid_guest_state = 0;
67module_param(emulate_invalid_guest_state, bool, S_IRUGO); 68module_param(emulate_invalid_guest_state, bool, S_IRUGO);
68 69
69static int __read_mostly vmm_exclusive = 1; 70static bool __read_mostly vmm_exclusive = 1;
70module_param(vmm_exclusive, bool, S_IRUGO); 71module_param(vmm_exclusive, bool, S_IRUGO);
71 72
72static int __read_mostly yield_on_hlt = 1; 73static bool __read_mostly yield_on_hlt = 1;
73module_param(yield_on_hlt, bool, S_IRUGO); 74module_param(yield_on_hlt, bool, S_IRUGO);
74 75
75static int __read_mostly fasteoi = 1; 76static bool __read_mostly fasteoi = 1;
76module_param(fasteoi, bool, S_IRUGO); 77module_param(fasteoi, bool, S_IRUGO);
77 78
78/* 79/*
@@ -80,7 +81,7 @@ module_param(fasteoi, bool, S_IRUGO);
80 * VMX and be a hypervisor for its own guests. If nested=0, guests may not 81 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
81 * use VMX instructions. 82 * use VMX instructions.
82 */ 83 */
83static int __read_mostly nested = 0; 84static bool __read_mostly nested = 0;
84module_param(nested, bool, S_IRUGO); 85module_param(nested, bool, S_IRUGO);
85 86
86#define KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST \ 87#define KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST \
@@ -1747,7 +1748,6 @@ static void setup_msrs(struct vcpu_vmx *vmx)
1747 int save_nmsrs, index; 1748 int save_nmsrs, index;
1748 unsigned long *msr_bitmap; 1749 unsigned long *msr_bitmap;
1749 1750
1750 vmx_load_host_state(vmx);
1751 save_nmsrs = 0; 1751 save_nmsrs = 0;
1752#ifdef CONFIG_X86_64 1752#ifdef CONFIG_X86_64
1753 if (is_long_mode(&vmx->vcpu)) { 1753 if (is_long_mode(&vmx->vcpu)) {
@@ -1956,6 +1956,7 @@ static __init void nested_vmx_setup_ctls_msrs(void)
1956#endif 1956#endif
1957 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING | 1957 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
1958 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_EXITING | 1958 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_EXITING |
1959 CPU_BASED_RDPMC_EXITING |
1959 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS; 1960 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1960 /* 1961 /*
1961 * We can allow some features even when not supported by the 1962 * We can allow some features even when not supported by the
@@ -2142,12 +2143,10 @@ static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2142 return 1; 2143 return 1;
2143 /* Otherwise falls through */ 2144 /* Otherwise falls through */
2144 default: 2145 default:
2145 vmx_load_host_state(to_vmx(vcpu));
2146 if (vmx_get_vmx_msr(vcpu, msr_index, pdata)) 2146 if (vmx_get_vmx_msr(vcpu, msr_index, pdata))
2147 return 0; 2147 return 0;
2148 msr = find_msr_entry(to_vmx(vcpu), msr_index); 2148 msr = find_msr_entry(to_vmx(vcpu), msr_index);
2149 if (msr) { 2149 if (msr) {
2150 vmx_load_host_state(to_vmx(vcpu));
2151 data = msr->data; 2150 data = msr->data;
2152 break; 2151 break;
2153 } 2152 }
@@ -2171,7 +2170,6 @@ static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
2171 2170
2172 switch (msr_index) { 2171 switch (msr_index) {
2173 case MSR_EFER: 2172 case MSR_EFER:
2174 vmx_load_host_state(vmx);
2175 ret = kvm_set_msr_common(vcpu, msr_index, data); 2173 ret = kvm_set_msr_common(vcpu, msr_index, data);
2176 break; 2174 break;
2177#ifdef CONFIG_X86_64 2175#ifdef CONFIG_X86_64
@@ -2220,7 +2218,6 @@ static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
2220 break; 2218 break;
2221 msr = find_msr_entry(vmx, msr_index); 2219 msr = find_msr_entry(vmx, msr_index);
2222 if (msr) { 2220 if (msr) {
2223 vmx_load_host_state(vmx);
2224 msr->data = data; 2221 msr->data = data;
2225 break; 2222 break;
2226 } 2223 }
@@ -2414,7 +2411,8 @@ static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
2414 CPU_BASED_USE_TSC_OFFSETING | 2411 CPU_BASED_USE_TSC_OFFSETING |
2415 CPU_BASED_MWAIT_EXITING | 2412 CPU_BASED_MWAIT_EXITING |
2416 CPU_BASED_MONITOR_EXITING | 2413 CPU_BASED_MONITOR_EXITING |
2417 CPU_BASED_INVLPG_EXITING; 2414 CPU_BASED_INVLPG_EXITING |
2415 CPU_BASED_RDPMC_EXITING;
2418 2416
2419 if (yield_on_hlt) 2417 if (yield_on_hlt)
2420 min |= CPU_BASED_HLT_EXITING; 2418 min |= CPU_BASED_HLT_EXITING;
@@ -2716,11 +2714,13 @@ static gva_t rmode_tss_base(struct kvm *kvm)
2716{ 2714{
2717 if (!kvm->arch.tss_addr) { 2715 if (!kvm->arch.tss_addr) {
2718 struct kvm_memslots *slots; 2716 struct kvm_memslots *slots;
2717 struct kvm_memory_slot *slot;
2719 gfn_t base_gfn; 2718 gfn_t base_gfn;
2720 2719
2721 slots = kvm_memslots(kvm); 2720 slots = kvm_memslots(kvm);
2722 base_gfn = slots->memslots[0].base_gfn + 2721 slot = id_to_memslot(slots, 0);
2723 kvm->memslots->memslots[0].npages - 3; 2722 base_gfn = slot->base_gfn + slot->npages - 3;
2723
2724 return base_gfn << PAGE_SHIFT; 2724 return base_gfn << PAGE_SHIFT;
2725 } 2725 }
2726 return kvm->arch.tss_addr; 2726 return kvm->arch.tss_addr;
@@ -3945,12 +3945,15 @@ static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
3945static void enable_irq_window(struct kvm_vcpu *vcpu) 3945static void enable_irq_window(struct kvm_vcpu *vcpu)
3946{ 3946{
3947 u32 cpu_based_vm_exec_control; 3947 u32 cpu_based_vm_exec_control;
3948 if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu)) 3948 if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu)) {
3949 /* We can get here when nested_run_pending caused 3949 /*
3950 * vmx_interrupt_allowed() to return false. In this case, do 3950 * We get here if vmx_interrupt_allowed() said we can't
3951 * nothing - the interrupt will be injected later. 3951 * inject to L1 now because L2 must run. Ask L2 to exit
3952 * right after entry, so we can inject to L1 more promptly.
3952 */ 3953 */
3954 kvm_make_request(KVM_REQ_IMMEDIATE_EXIT, vcpu);
3953 return; 3955 return;
3956 }
3954 3957
3955 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL); 3958 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3956 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING; 3959 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
@@ -4077,11 +4080,12 @@ static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4077static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu) 4080static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4078{ 4081{
4079 if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu)) { 4082 if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu)) {
4080 struct vmcs12 *vmcs12; 4083 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4081 if (to_vmx(vcpu)->nested.nested_run_pending) 4084 if (to_vmx(vcpu)->nested.nested_run_pending ||
4085 (vmcs12->idt_vectoring_info_field &
4086 VECTORING_INFO_VALID_MASK))
4082 return 0; 4087 return 0;
4083 nested_vmx_vmexit(vcpu); 4088 nested_vmx_vmexit(vcpu);
4084 vmcs12 = get_vmcs12(vcpu);
4085 vmcs12->vm_exit_reason = EXIT_REASON_EXTERNAL_INTERRUPT; 4089 vmcs12->vm_exit_reason = EXIT_REASON_EXTERNAL_INTERRUPT;
4086 vmcs12->vm_exit_intr_info = 0; 4090 vmcs12->vm_exit_intr_info = 0;
4087 /* fall through to normal code, but now in L1, not L2 */ 4091 /* fall through to normal code, but now in L1, not L2 */
@@ -4611,6 +4615,16 @@ static int handle_invlpg(struct kvm_vcpu *vcpu)
4611 return 1; 4615 return 1;
4612} 4616}
4613 4617
4618static int handle_rdpmc(struct kvm_vcpu *vcpu)
4619{
4620 int err;
4621
4622 err = kvm_rdpmc(vcpu);
4623 kvm_complete_insn_gp(vcpu, err);
4624
4625 return 1;
4626}
4627
4614static int handle_wbinvd(struct kvm_vcpu *vcpu) 4628static int handle_wbinvd(struct kvm_vcpu *vcpu)
4615{ 4629{
4616 skip_emulated_instruction(vcpu); 4630 skip_emulated_instruction(vcpu);
@@ -5561,6 +5575,7 @@ static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5561 [EXIT_REASON_HLT] = handle_halt, 5575 [EXIT_REASON_HLT] = handle_halt,
5562 [EXIT_REASON_INVD] = handle_invd, 5576 [EXIT_REASON_INVD] = handle_invd,
5563 [EXIT_REASON_INVLPG] = handle_invlpg, 5577 [EXIT_REASON_INVLPG] = handle_invlpg,
5578 [EXIT_REASON_RDPMC] = handle_rdpmc,
5564 [EXIT_REASON_VMCALL] = handle_vmcall, 5579 [EXIT_REASON_VMCALL] = handle_vmcall,
5565 [EXIT_REASON_VMCLEAR] = handle_vmclear, 5580 [EXIT_REASON_VMCLEAR] = handle_vmclear,
5566 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch, 5581 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index c38efd7b792e..14d6cadc4ba6 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -26,6 +26,7 @@
26#include "tss.h" 26#include "tss.h"
27#include "kvm_cache_regs.h" 27#include "kvm_cache_regs.h"
28#include "x86.h" 28#include "x86.h"
29#include "cpuid.h"
29 30
30#include <linux/clocksource.h> 31#include <linux/clocksource.h>
31#include <linux/interrupt.h> 32#include <linux/interrupt.h>
@@ -82,15 +83,13 @@ static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
82#define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU 83#define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
83 84
84static void update_cr8_intercept(struct kvm_vcpu *vcpu); 85static void update_cr8_intercept(struct kvm_vcpu *vcpu);
85static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
86 struct kvm_cpuid_entry2 __user *entries);
87static void process_nmi(struct kvm_vcpu *vcpu); 86static void process_nmi(struct kvm_vcpu *vcpu);
88 87
89struct kvm_x86_ops *kvm_x86_ops; 88struct kvm_x86_ops *kvm_x86_ops;
90EXPORT_SYMBOL_GPL(kvm_x86_ops); 89EXPORT_SYMBOL_GPL(kvm_x86_ops);
91 90
92int ignore_msrs = 0; 91static bool ignore_msrs = 0;
93module_param_named(ignore_msrs, ignore_msrs, bool, S_IRUGO | S_IWUSR); 92module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
94 93
95bool kvm_has_tsc_control; 94bool kvm_has_tsc_control;
96EXPORT_SYMBOL_GPL(kvm_has_tsc_control); 95EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
@@ -574,58 +573,6 @@ int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
574} 573}
575EXPORT_SYMBOL_GPL(kvm_set_xcr); 574EXPORT_SYMBOL_GPL(kvm_set_xcr);
576 575
577static bool guest_cpuid_has_xsave(struct kvm_vcpu *vcpu)
578{
579 struct kvm_cpuid_entry2 *best;
580
581 best = kvm_find_cpuid_entry(vcpu, 1, 0);
582 return best && (best->ecx & bit(X86_FEATURE_XSAVE));
583}
584
585static bool guest_cpuid_has_smep(struct kvm_vcpu *vcpu)
586{
587 struct kvm_cpuid_entry2 *best;
588
589 best = kvm_find_cpuid_entry(vcpu, 7, 0);
590 return best && (best->ebx & bit(X86_FEATURE_SMEP));
591}
592
593static bool guest_cpuid_has_fsgsbase(struct kvm_vcpu *vcpu)
594{
595 struct kvm_cpuid_entry2 *best;
596
597 best = kvm_find_cpuid_entry(vcpu, 7, 0);
598 return best && (best->ebx & bit(X86_FEATURE_FSGSBASE));
599}
600
601static void update_cpuid(struct kvm_vcpu *vcpu)
602{
603 struct kvm_cpuid_entry2 *best;
604 struct kvm_lapic *apic = vcpu->arch.apic;
605 u32 timer_mode_mask;
606
607 best = kvm_find_cpuid_entry(vcpu, 1, 0);
608 if (!best)
609 return;
610
611 /* Update OSXSAVE bit */
612 if (cpu_has_xsave && best->function == 0x1) {
613 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
614 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
615 best->ecx |= bit(X86_FEATURE_OSXSAVE);
616 }
617
618 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
619 best->function == 0x1) {
620 best->ecx |= bit(X86_FEATURE_TSC_DEADLINE_TIMER);
621 timer_mode_mask = 3 << 17;
622 } else
623 timer_mode_mask = 1 << 17;
624
625 if (apic)
626 apic->lapic_timer.timer_mode_mask = timer_mode_mask;
627}
628
629int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4) 576int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
630{ 577{
631 unsigned long old_cr4 = kvm_read_cr4(vcpu); 578 unsigned long old_cr4 = kvm_read_cr4(vcpu);
@@ -659,7 +606,7 @@ int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
659 kvm_mmu_reset_context(vcpu); 606 kvm_mmu_reset_context(vcpu);
660 607
661 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE) 608 if ((cr4 ^ old_cr4) & X86_CR4_OSXSAVE)
662 update_cpuid(vcpu); 609 kvm_update_cpuid(vcpu);
663 610
664 return 0; 611 return 0;
665} 612}
@@ -813,6 +760,21 @@ int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
813} 760}
814EXPORT_SYMBOL_GPL(kvm_get_dr); 761EXPORT_SYMBOL_GPL(kvm_get_dr);
815 762
763bool kvm_rdpmc(struct kvm_vcpu *vcpu)
764{
765 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
766 u64 data;
767 int err;
768
769 err = kvm_pmu_read_pmc(vcpu, ecx, &data);
770 if (err)
771 return err;
772 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
773 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
774 return err;
775}
776EXPORT_SYMBOL_GPL(kvm_rdpmc);
777
816/* 778/*
817 * List of msr numbers which we expose to userspace through KVM_GET_MSRS 779 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
818 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST. 780 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
@@ -1362,12 +1324,11 @@ static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
1362 if (page_num >= blob_size) 1324 if (page_num >= blob_size)
1363 goto out; 1325 goto out;
1364 r = -ENOMEM; 1326 r = -ENOMEM;
1365 page = kzalloc(PAGE_SIZE, GFP_KERNEL); 1327 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
1366 if (!page) 1328 if (IS_ERR(page)) {
1329 r = PTR_ERR(page);
1367 goto out; 1330 goto out;
1368 r = -EFAULT; 1331 }
1369 if (copy_from_user(page, blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE))
1370 goto out_free;
1371 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE)) 1332 if (kvm_write_guest(kvm, page_addr, page, PAGE_SIZE))
1372 goto out_free; 1333 goto out_free;
1373 r = 0; 1334 r = 0;
@@ -1656,8 +1617,6 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1656 * which we perfectly emulate ;-). Any other value should be at least 1617 * which we perfectly emulate ;-). Any other value should be at least
1657 * reported, some guests depend on them. 1618 * reported, some guests depend on them.
1658 */ 1619 */
1659 case MSR_P6_EVNTSEL0:
1660 case MSR_P6_EVNTSEL1:
1661 case MSR_K7_EVNTSEL0: 1620 case MSR_K7_EVNTSEL0:
1662 case MSR_K7_EVNTSEL1: 1621 case MSR_K7_EVNTSEL1:
1663 case MSR_K7_EVNTSEL2: 1622 case MSR_K7_EVNTSEL2:
@@ -1669,8 +1628,6 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1669 /* at least RHEL 4 unconditionally writes to the perfctr registers, 1628 /* at least RHEL 4 unconditionally writes to the perfctr registers,
1670 * so we ignore writes to make it happy. 1629 * so we ignore writes to make it happy.
1671 */ 1630 */
1672 case MSR_P6_PERFCTR0:
1673 case MSR_P6_PERFCTR1:
1674 case MSR_K7_PERFCTR0: 1631 case MSR_K7_PERFCTR0:
1675 case MSR_K7_PERFCTR1: 1632 case MSR_K7_PERFCTR1:
1676 case MSR_K7_PERFCTR2: 1633 case MSR_K7_PERFCTR2:
@@ -1707,6 +1664,8 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
1707 default: 1664 default:
1708 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr)) 1665 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
1709 return xen_hvm_config(vcpu, data); 1666 return xen_hvm_config(vcpu, data);
1667 if (kvm_pmu_msr(vcpu, msr))
1668 return kvm_pmu_set_msr(vcpu, msr, data);
1710 if (!ignore_msrs) { 1669 if (!ignore_msrs) {
1711 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n", 1670 pr_unimpl(vcpu, "unhandled wrmsr: 0x%x data %llx\n",
1712 msr, data); 1671 msr, data);
@@ -1869,10 +1828,6 @@ int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1869 case MSR_K8_SYSCFG: 1828 case MSR_K8_SYSCFG:
1870 case MSR_K7_HWCR: 1829 case MSR_K7_HWCR:
1871 case MSR_VM_HSAVE_PA: 1830 case MSR_VM_HSAVE_PA:
1872 case MSR_P6_PERFCTR0:
1873 case MSR_P6_PERFCTR1:
1874 case MSR_P6_EVNTSEL0:
1875 case MSR_P6_EVNTSEL1:
1876 case MSR_K7_EVNTSEL0: 1831 case MSR_K7_EVNTSEL0:
1877 case MSR_K7_PERFCTR0: 1832 case MSR_K7_PERFCTR0:
1878 case MSR_K8_INT_PENDING_MSG: 1833 case MSR_K8_INT_PENDING_MSG:
@@ -1983,6 +1938,8 @@ int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
1983 data = 0xbe702111; 1938 data = 0xbe702111;
1984 break; 1939 break;
1985 default: 1940 default:
1941 if (kvm_pmu_msr(vcpu, msr))
1942 return kvm_pmu_get_msr(vcpu, msr, pdata);
1986 if (!ignore_msrs) { 1943 if (!ignore_msrs) {
1987 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr); 1944 pr_unimpl(vcpu, "unhandled rdmsr: 0x%x\n", msr);
1988 return 1; 1945 return 1;
@@ -2041,15 +1998,12 @@ static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2041 if (msrs.nmsrs >= MAX_IO_MSRS) 1998 if (msrs.nmsrs >= MAX_IO_MSRS)
2042 goto out; 1999 goto out;
2043 2000
2044 r = -ENOMEM;
2045 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs; 2001 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2046 entries = kmalloc(size, GFP_KERNEL); 2002 entries = memdup_user(user_msrs->entries, size);
2047 if (!entries) 2003 if (IS_ERR(entries)) {
2004 r = PTR_ERR(entries);
2048 goto out; 2005 goto out;
2049 2006 }
2050 r = -EFAULT;
2051 if (copy_from_user(entries, user_msrs->entries, size))
2052 goto out_free;
2053 2007
2054 r = n = __msr_io(vcpu, &msrs, entries, do_msr); 2008 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2055 if (r < 0) 2009 if (r < 0)
@@ -2135,6 +2089,9 @@ int kvm_dev_ioctl_check_extension(long ext)
2135 case KVM_CAP_TSC_CONTROL: 2089 case KVM_CAP_TSC_CONTROL:
2136 r = kvm_has_tsc_control; 2090 r = kvm_has_tsc_control;
2137 break; 2091 break;
2092 case KVM_CAP_TSC_DEADLINE_TIMER:
2093 r = boot_cpu_has(X86_FEATURE_TSC_DEADLINE_TIMER);
2094 break;
2138 default: 2095 default:
2139 r = 0; 2096 r = 0;
2140 break; 2097 break;
@@ -2266,466 +2223,6 @@ void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
2266 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu); 2223 vcpu->arch.last_guest_tsc = kvm_x86_ops->read_l1_tsc(vcpu);
2267} 2224}
2268 2225
2269static int is_efer_nx(void)
2270{
2271 unsigned long long efer = 0;
2272
2273 rdmsrl_safe(MSR_EFER, &efer);
2274 return efer & EFER_NX;
2275}
2276
2277static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
2278{
2279 int i;
2280 struct kvm_cpuid_entry2 *e, *entry;
2281
2282 entry = NULL;
2283 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
2284 e = &vcpu->arch.cpuid_entries[i];
2285 if (e->function == 0x80000001) {
2286 entry = e;
2287 break;
2288 }
2289 }
2290 if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
2291 entry->edx &= ~(1 << 20);
2292 printk(KERN_INFO "kvm: guest NX capability removed\n");
2293 }
2294}
2295
2296/* when an old userspace process fills a new kernel module */
2297static int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
2298 struct kvm_cpuid *cpuid,
2299 struct kvm_cpuid_entry __user *entries)
2300{
2301 int r, i;
2302 struct kvm_cpuid_entry *cpuid_entries;
2303
2304 r = -E2BIG;
2305 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2306 goto out;
2307 r = -ENOMEM;
2308 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
2309 if (!cpuid_entries)
2310 goto out;
2311 r = -EFAULT;
2312 if (copy_from_user(cpuid_entries, entries,
2313 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
2314 goto out_free;
2315 for (i = 0; i < cpuid->nent; i++) {
2316 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
2317 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
2318 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
2319 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
2320 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
2321 vcpu->arch.cpuid_entries[i].index = 0;
2322 vcpu->arch.cpuid_entries[i].flags = 0;
2323 vcpu->arch.cpuid_entries[i].padding[0] = 0;
2324 vcpu->arch.cpuid_entries[i].padding[1] = 0;
2325 vcpu->arch.cpuid_entries[i].padding[2] = 0;
2326 }
2327 vcpu->arch.cpuid_nent = cpuid->nent;
2328 cpuid_fix_nx_cap(vcpu);
2329 r = 0;
2330 kvm_apic_set_version(vcpu);
2331 kvm_x86_ops->cpuid_update(vcpu);
2332 update_cpuid(vcpu);
2333
2334out_free:
2335 vfree(cpuid_entries);
2336out:
2337 return r;
2338}
2339
2340static int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
2341 struct kvm_cpuid2 *cpuid,
2342 struct kvm_cpuid_entry2 __user *entries)
2343{
2344 int r;
2345
2346 r = -E2BIG;
2347 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2348 goto out;
2349 r = -EFAULT;
2350 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
2351 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
2352 goto out;
2353 vcpu->arch.cpuid_nent = cpuid->nent;
2354 kvm_apic_set_version(vcpu);
2355 kvm_x86_ops->cpuid_update(vcpu);
2356 update_cpuid(vcpu);
2357 return 0;
2358
2359out:
2360 return r;
2361}
2362
2363static int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
2364 struct kvm_cpuid2 *cpuid,
2365 struct kvm_cpuid_entry2 __user *entries)
2366{
2367 int r;
2368
2369 r = -E2BIG;
2370 if (cpuid->nent < vcpu->arch.cpuid_nent)
2371 goto out;
2372 r = -EFAULT;
2373 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
2374 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
2375 goto out;
2376 return 0;
2377
2378out:
2379 cpuid->nent = vcpu->arch.cpuid_nent;
2380 return r;
2381}
2382
2383static void cpuid_mask(u32 *word, int wordnum)
2384{
2385 *word &= boot_cpu_data.x86_capability[wordnum];
2386}
2387
2388static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2389 u32 index)
2390{
2391 entry->function = function;
2392 entry->index = index;
2393 cpuid_count(entry->function, entry->index,
2394 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
2395 entry->flags = 0;
2396}
2397
2398static bool supported_xcr0_bit(unsigned bit)
2399{
2400 u64 mask = ((u64)1 << bit);
2401
2402 return mask & (XSTATE_FP | XSTATE_SSE | XSTATE_YMM) & host_xcr0;
2403}
2404
2405#define F(x) bit(X86_FEATURE_##x)
2406
2407static void do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
2408 u32 index, int *nent, int maxnent)
2409{
2410 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
2411#ifdef CONFIG_X86_64
2412 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
2413 ? F(GBPAGES) : 0;
2414 unsigned f_lm = F(LM);
2415#else
2416 unsigned f_gbpages = 0;
2417 unsigned f_lm = 0;
2418#endif
2419 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
2420
2421 /* cpuid 1.edx */
2422 const u32 kvm_supported_word0_x86_features =
2423 F(FPU) | F(VME) | F(DE) | F(PSE) |
2424 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2425 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
2426 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2427 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
2428 0 /* Reserved, DS, ACPI */ | F(MMX) |
2429 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
2430 0 /* HTT, TM, Reserved, PBE */;
2431 /* cpuid 0x80000001.edx */
2432 const u32 kvm_supported_word1_x86_features =
2433 F(FPU) | F(VME) | F(DE) | F(PSE) |
2434 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
2435 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
2436 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
2437 F(PAT) | F(PSE36) | 0 /* Reserved */ |
2438 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
2439 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
2440 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
2441 /* cpuid 1.ecx */
2442 const u32 kvm_supported_word4_x86_features =
2443 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
2444 0 /* DS-CPL, VMX, SMX, EST */ |
2445 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
2446 0 /* Reserved */ | F(CX16) | 0 /* xTPR Update, PDCM */ |
2447 0 /* Reserved, DCA */ | F(XMM4_1) |
2448 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
2449 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
2450 F(F16C) | F(RDRAND);
2451 /* cpuid 0x80000001.ecx */
2452 const u32 kvm_supported_word6_x86_features =
2453 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
2454 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
2455 F(3DNOWPREFETCH) | 0 /* OSVW */ | 0 /* IBS */ | F(XOP) |
2456 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
2457
2458 /* cpuid 0xC0000001.edx */
2459 const u32 kvm_supported_word5_x86_features =
2460 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
2461 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
2462 F(PMM) | F(PMM_EN);
2463
2464 /* cpuid 7.0.ebx */
2465 const u32 kvm_supported_word9_x86_features =
2466 F(SMEP) | F(FSGSBASE) | F(ERMS);
2467
2468 /* all calls to cpuid_count() should be made on the same cpu */
2469 get_cpu();
2470 do_cpuid_1_ent(entry, function, index);
2471 ++*nent;
2472
2473 switch (function) {
2474 case 0:
2475 entry->eax = min(entry->eax, (u32)0xd);
2476 break;
2477 case 1:
2478 entry->edx &= kvm_supported_word0_x86_features;
2479 cpuid_mask(&entry->edx, 0);
2480 entry->ecx &= kvm_supported_word4_x86_features;
2481 cpuid_mask(&entry->ecx, 4);
2482 /* we support x2apic emulation even if host does not support
2483 * it since we emulate x2apic in software */
2484 entry->ecx |= F(X2APIC);
2485 break;
2486 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
2487 * may return different values. This forces us to get_cpu() before
2488 * issuing the first command, and also to emulate this annoying behavior
2489 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
2490 case 2: {
2491 int t, times = entry->eax & 0xff;
2492
2493 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2494 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
2495 for (t = 1; t < times && *nent < maxnent; ++t) {
2496 do_cpuid_1_ent(&entry[t], function, 0);
2497 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
2498 ++*nent;
2499 }
2500 break;
2501 }
2502 /* function 4 has additional index. */
2503 case 4: {
2504 int i, cache_type;
2505
2506 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2507 /* read more entries until cache_type is zero */
2508 for (i = 1; *nent < maxnent; ++i) {
2509 cache_type = entry[i - 1].eax & 0x1f;
2510 if (!cache_type)
2511 break;
2512 do_cpuid_1_ent(&entry[i], function, i);
2513 entry[i].flags |=
2514 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2515 ++*nent;
2516 }
2517 break;
2518 }
2519 case 7: {
2520 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2521 /* Mask ebx against host capbability word 9 */
2522 if (index == 0) {
2523 entry->ebx &= kvm_supported_word9_x86_features;
2524 cpuid_mask(&entry->ebx, 9);
2525 } else
2526 entry->ebx = 0;
2527 entry->eax = 0;
2528 entry->ecx = 0;
2529 entry->edx = 0;
2530 break;
2531 }
2532 case 9:
2533 break;
2534 /* function 0xb has additional index. */
2535 case 0xb: {
2536 int i, level_type;
2537
2538 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2539 /* read more entries until level_type is zero */
2540 for (i = 1; *nent < maxnent; ++i) {
2541 level_type = entry[i - 1].ecx & 0xff00;
2542 if (!level_type)
2543 break;
2544 do_cpuid_1_ent(&entry[i], function, i);
2545 entry[i].flags |=
2546 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2547 ++*nent;
2548 }
2549 break;
2550 }
2551 case 0xd: {
2552 int idx, i;
2553
2554 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2555 for (idx = 1, i = 1; *nent < maxnent && idx < 64; ++idx) {
2556 do_cpuid_1_ent(&entry[i], function, idx);
2557 if (entry[i].eax == 0 || !supported_xcr0_bit(idx))
2558 continue;
2559 entry[i].flags |=
2560 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
2561 ++*nent;
2562 ++i;
2563 }
2564 break;
2565 }
2566 case KVM_CPUID_SIGNATURE: {
2567 char signature[12] = "KVMKVMKVM\0\0";
2568 u32 *sigptr = (u32 *)signature;
2569 entry->eax = 0;
2570 entry->ebx = sigptr[0];
2571 entry->ecx = sigptr[1];
2572 entry->edx = sigptr[2];
2573 break;
2574 }
2575 case KVM_CPUID_FEATURES:
2576 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
2577 (1 << KVM_FEATURE_NOP_IO_DELAY) |
2578 (1 << KVM_FEATURE_CLOCKSOURCE2) |
2579 (1 << KVM_FEATURE_ASYNC_PF) |
2580 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
2581
2582 if (sched_info_on())
2583 entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
2584
2585 entry->ebx = 0;
2586 entry->ecx = 0;
2587 entry->edx = 0;
2588 break;
2589 case 0x80000000:
2590 entry->eax = min(entry->eax, 0x8000001a);
2591 break;
2592 case 0x80000001:
2593 entry->edx &= kvm_supported_word1_x86_features;
2594 cpuid_mask(&entry->edx, 1);
2595 entry->ecx &= kvm_supported_word6_x86_features;
2596 cpuid_mask(&entry->ecx, 6);
2597 break;
2598 case 0x80000008: {
2599 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
2600 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
2601 unsigned phys_as = entry->eax & 0xff;
2602
2603 if (!g_phys_as)
2604 g_phys_as = phys_as;
2605 entry->eax = g_phys_as | (virt_as << 8);
2606 entry->ebx = entry->edx = 0;
2607 break;
2608 }
2609 case 0x80000019:
2610 entry->ecx = entry->edx = 0;
2611 break;
2612 case 0x8000001a:
2613 break;
2614 case 0x8000001d:
2615 break;
2616 /*Add support for Centaur's CPUID instruction*/
2617 case 0xC0000000:
2618 /*Just support up to 0xC0000004 now*/
2619 entry->eax = min(entry->eax, 0xC0000004);
2620 break;
2621 case 0xC0000001:
2622 entry->edx &= kvm_supported_word5_x86_features;
2623 cpuid_mask(&entry->edx, 5);
2624 break;
2625 case 3: /* Processor serial number */
2626 case 5: /* MONITOR/MWAIT */
2627 case 6: /* Thermal management */
2628 case 0xA: /* Architectural Performance Monitoring */
2629 case 0x80000007: /* Advanced power management */
2630 case 0xC0000002:
2631 case 0xC0000003:
2632 case 0xC0000004:
2633 default:
2634 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
2635 break;
2636 }
2637
2638 kvm_x86_ops->set_supported_cpuid(function, entry);
2639
2640 put_cpu();
2641}
2642
2643#undef F
2644
2645static int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
2646 struct kvm_cpuid_entry2 __user *entries)
2647{
2648 struct kvm_cpuid_entry2 *cpuid_entries;
2649 int limit, nent = 0, r = -E2BIG;
2650 u32 func;
2651
2652 if (cpuid->nent < 1)
2653 goto out;
2654 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
2655 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
2656 r = -ENOMEM;
2657 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
2658 if (!cpuid_entries)
2659 goto out;
2660
2661 do_cpuid_ent(&cpuid_entries[0], 0, 0, &nent, cpuid->nent);
2662 limit = cpuid_entries[0].eax;
2663 for (func = 1; func <= limit && nent < cpuid->nent; ++func)
2664 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2665 &nent, cpuid->nent);
2666 r = -E2BIG;
2667 if (nent >= cpuid->nent)
2668 goto out_free;
2669
2670 do_cpuid_ent(&cpuid_entries[nent], 0x80000000, 0, &nent, cpuid->nent);
2671 limit = cpuid_entries[nent - 1].eax;
2672 for (func = 0x80000001; func <= limit && nent < cpuid->nent; ++func)
2673 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2674 &nent, cpuid->nent);
2675
2676
2677
2678 r = -E2BIG;
2679 if (nent >= cpuid->nent)
2680 goto out_free;
2681
2682 /* Add support for Centaur's CPUID instruction. */
2683 if (boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR) {
2684 do_cpuid_ent(&cpuid_entries[nent], 0xC0000000, 0,
2685 &nent, cpuid->nent);
2686
2687 r = -E2BIG;
2688 if (nent >= cpuid->nent)
2689 goto out_free;
2690
2691 limit = cpuid_entries[nent - 1].eax;
2692 for (func = 0xC0000001;
2693 func <= limit && nent < cpuid->nent; ++func)
2694 do_cpuid_ent(&cpuid_entries[nent], func, 0,
2695 &nent, cpuid->nent);
2696
2697 r = -E2BIG;
2698 if (nent >= cpuid->nent)
2699 goto out_free;
2700 }
2701
2702 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_SIGNATURE, 0, &nent,
2703 cpuid->nent);
2704
2705 r = -E2BIG;
2706 if (nent >= cpuid->nent)
2707 goto out_free;
2708
2709 do_cpuid_ent(&cpuid_entries[nent], KVM_CPUID_FEATURES, 0, &nent,
2710 cpuid->nent);
2711
2712 r = -E2BIG;
2713 if (nent >= cpuid->nent)
2714 goto out_free;
2715
2716 r = -EFAULT;
2717 if (copy_to_user(entries, cpuid_entries,
2718 nent * sizeof(struct kvm_cpuid_entry2)))
2719 goto out_free;
2720 cpuid->nent = nent;
2721 r = 0;
2722
2723out_free:
2724 vfree(cpuid_entries);
2725out:
2726 return r;
2727}
2728
2729static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu, 2226static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
2730 struct kvm_lapic_state *s) 2227 struct kvm_lapic_state *s)
2731{ 2228{
@@ -3043,13 +2540,12 @@ long kvm_arch_vcpu_ioctl(struct file *filp,
3043 r = -EINVAL; 2540 r = -EINVAL;
3044 if (!vcpu->arch.apic) 2541 if (!vcpu->arch.apic)
3045 goto out; 2542 goto out;
3046 u.lapic = kmalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL); 2543 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3047 r = -ENOMEM; 2544 if (IS_ERR(u.lapic)) {
3048 if (!u.lapic) 2545 r = PTR_ERR(u.lapic);
3049 goto out;
3050 r = -EFAULT;
3051 if (copy_from_user(u.lapic, argp, sizeof(struct kvm_lapic_state)))
3052 goto out; 2546 goto out;
2547 }
2548
3053 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic); 2549 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3054 if (r) 2550 if (r)
3055 goto out; 2551 goto out;
@@ -3228,14 +2724,11 @@ long kvm_arch_vcpu_ioctl(struct file *filp,
3228 break; 2724 break;
3229 } 2725 }
3230 case KVM_SET_XSAVE: { 2726 case KVM_SET_XSAVE: {
3231 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL); 2727 u.xsave = memdup_user(argp, sizeof(*u.xsave));
3232 r = -ENOMEM; 2728 if (IS_ERR(u.xsave)) {
3233 if (!u.xsave) 2729 r = PTR_ERR(u.xsave);
3234 break; 2730 goto out;
3235 2731 }
3236 r = -EFAULT;
3237 if (copy_from_user(u.xsave, argp, sizeof(struct kvm_xsave)))
3238 break;
3239 2732
3240 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave); 2733 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
3241 break; 2734 break;
@@ -3256,15 +2749,11 @@ long kvm_arch_vcpu_ioctl(struct file *filp,
3256 break; 2749 break;
3257 } 2750 }
3258 case KVM_SET_XCRS: { 2751 case KVM_SET_XCRS: {
3259 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL); 2752 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
3260 r = -ENOMEM; 2753 if (IS_ERR(u.xcrs)) {
3261 if (!u.xcrs) 2754 r = PTR_ERR(u.xcrs);
3262 break; 2755 goto out;
3263 2756 }
3264 r = -EFAULT;
3265 if (copy_from_user(u.xcrs, argp,
3266 sizeof(struct kvm_xcrs)))
3267 break;
3268 2757
3269 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs); 2758 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
3270 break; 2759 break;
@@ -3461,16 +2950,59 @@ static int kvm_vm_ioctl_reinject(struct kvm *kvm,
3461 return 0; 2950 return 0;
3462} 2951}
3463 2952
2953/**
2954 * write_protect_slot - write protect a slot for dirty logging
2955 * @kvm: the kvm instance
2956 * @memslot: the slot we protect
2957 * @dirty_bitmap: the bitmap indicating which pages are dirty
2958 * @nr_dirty_pages: the number of dirty pages
2959 *
2960 * We have two ways to find all sptes to protect:
2961 * 1. Use kvm_mmu_slot_remove_write_access() which walks all shadow pages and
2962 * checks ones that have a spte mapping a page in the slot.
2963 * 2. Use kvm_mmu_rmap_write_protect() for each gfn found in the bitmap.
2964 *
2965 * Generally speaking, if there are not so many dirty pages compared to the
2966 * number of shadow pages, we should use the latter.
2967 *
2968 * Note that letting others write into a page marked dirty in the old bitmap
2969 * by using the remaining tlb entry is not a problem. That page will become
2970 * write protected again when we flush the tlb and then be reported dirty to
2971 * the user space by copying the old bitmap.
2972 */
2973static void write_protect_slot(struct kvm *kvm,
2974 struct kvm_memory_slot *memslot,
2975 unsigned long *dirty_bitmap,
2976 unsigned long nr_dirty_pages)
2977{
2978 /* Not many dirty pages compared to # of shadow pages. */
2979 if (nr_dirty_pages < kvm->arch.n_used_mmu_pages) {
2980 unsigned long gfn_offset;
2981
2982 for_each_set_bit(gfn_offset, dirty_bitmap, memslot->npages) {
2983 unsigned long gfn = memslot->base_gfn + gfn_offset;
2984
2985 spin_lock(&kvm->mmu_lock);
2986 kvm_mmu_rmap_write_protect(kvm, gfn, memslot);
2987 spin_unlock(&kvm->mmu_lock);
2988 }
2989 kvm_flush_remote_tlbs(kvm);
2990 } else {
2991 spin_lock(&kvm->mmu_lock);
2992 kvm_mmu_slot_remove_write_access(kvm, memslot->id);
2993 spin_unlock(&kvm->mmu_lock);
2994 }
2995}
2996
3464/* 2997/*
3465 * Get (and clear) the dirty memory log for a memory slot. 2998 * Get (and clear) the dirty memory log for a memory slot.
3466 */ 2999 */
3467int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, 3000int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3468 struct kvm_dirty_log *log) 3001 struct kvm_dirty_log *log)
3469{ 3002{
3470 int r, i; 3003 int r;
3471 struct kvm_memory_slot *memslot; 3004 struct kvm_memory_slot *memslot;
3472 unsigned long n; 3005 unsigned long n, nr_dirty_pages;
3473 unsigned long is_dirty = 0;
3474 3006
3475 mutex_lock(&kvm->slots_lock); 3007 mutex_lock(&kvm->slots_lock);
3476 3008
@@ -3478,43 +3010,41 @@ int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
3478 if (log->slot >= KVM_MEMORY_SLOTS) 3010 if (log->slot >= KVM_MEMORY_SLOTS)
3479 goto out; 3011 goto out;
3480 3012
3481 memslot = &kvm->memslots->memslots[log->slot]; 3013 memslot = id_to_memslot(kvm->memslots, log->slot);
3482 r = -ENOENT; 3014 r = -ENOENT;
3483 if (!memslot->dirty_bitmap) 3015 if (!memslot->dirty_bitmap)
3484 goto out; 3016 goto out;
3485 3017
3486 n = kvm_dirty_bitmap_bytes(memslot); 3018 n = kvm_dirty_bitmap_bytes(memslot);
3487 3019 nr_dirty_pages = memslot->nr_dirty_pages;
3488 for (i = 0; !is_dirty && i < n/sizeof(long); i++)
3489 is_dirty = memslot->dirty_bitmap[i];
3490 3020
3491 /* If nothing is dirty, don't bother messing with page tables. */ 3021 /* If nothing is dirty, don't bother messing with page tables. */
3492 if (is_dirty) { 3022 if (nr_dirty_pages) {
3493 struct kvm_memslots *slots, *old_slots; 3023 struct kvm_memslots *slots, *old_slots;
3494 unsigned long *dirty_bitmap; 3024 unsigned long *dirty_bitmap, *dirty_bitmap_head;
3495 3025
3496 dirty_bitmap = memslot->dirty_bitmap_head; 3026 dirty_bitmap = memslot->dirty_bitmap;
3497 if (memslot->dirty_bitmap == dirty_bitmap) 3027 dirty_bitmap_head = memslot->dirty_bitmap_head;
3498 dirty_bitmap += n / sizeof(long); 3028 if (dirty_bitmap == dirty_bitmap_head)
3499 memset(dirty_bitmap, 0, n); 3029 dirty_bitmap_head += n / sizeof(long);
3030 memset(dirty_bitmap_head, 0, n);
3500 3031
3501 r = -ENOMEM; 3032 r = -ENOMEM;
3502 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL); 3033 slots = kmemdup(kvm->memslots, sizeof(*kvm->memslots), GFP_KERNEL);
3503 if (!slots) 3034 if (!slots)
3504 goto out; 3035 goto out;
3505 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots)); 3036
3506 slots->memslots[log->slot].dirty_bitmap = dirty_bitmap; 3037 memslot = id_to_memslot(slots, log->slot);
3507 slots->generation++; 3038 memslot->nr_dirty_pages = 0;
3039 memslot->dirty_bitmap = dirty_bitmap_head;
3040 update_memslots(slots, NULL);
3508 3041
3509 old_slots = kvm->memslots; 3042 old_slots = kvm->memslots;
3510 rcu_assign_pointer(kvm->memslots, slots); 3043 rcu_assign_pointer(kvm->memslots, slots);
3511 synchronize_srcu_expedited(&kvm->srcu); 3044 synchronize_srcu_expedited(&kvm->srcu);
3512 dirty_bitmap = old_slots->memslots[log->slot].dirty_bitmap;
3513 kfree(old_slots); 3045 kfree(old_slots);
3514 3046
3515 spin_lock(&kvm->mmu_lock); 3047 write_protect_slot(kvm, memslot, dirty_bitmap, nr_dirty_pages);
3516 kvm_mmu_slot_remove_write_access(kvm, log->slot);
3517 spin_unlock(&kvm->mmu_lock);
3518 3048
3519 r = -EFAULT; 3049 r = -EFAULT;
3520 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n)) 3050 if (copy_to_user(log->dirty_bitmap, dirty_bitmap, n))
@@ -3659,14 +3189,14 @@ long kvm_arch_vm_ioctl(struct file *filp,
3659 } 3189 }
3660 case KVM_GET_IRQCHIP: { 3190 case KVM_GET_IRQCHIP: {
3661 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */ 3191 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3662 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL); 3192 struct kvm_irqchip *chip;
3663 3193
3664 r = -ENOMEM; 3194 chip = memdup_user(argp, sizeof(*chip));
3665 if (!chip) 3195 if (IS_ERR(chip)) {
3196 r = PTR_ERR(chip);
3666 goto out; 3197 goto out;
3667 r = -EFAULT; 3198 }
3668 if (copy_from_user(chip, argp, sizeof *chip)) 3199
3669 goto get_irqchip_out;
3670 r = -ENXIO; 3200 r = -ENXIO;
3671 if (!irqchip_in_kernel(kvm)) 3201 if (!irqchip_in_kernel(kvm))
3672 goto get_irqchip_out; 3202 goto get_irqchip_out;
@@ -3685,14 +3215,14 @@ long kvm_arch_vm_ioctl(struct file *filp,
3685 } 3215 }
3686 case KVM_SET_IRQCHIP: { 3216 case KVM_SET_IRQCHIP: {
3687 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */ 3217 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
3688 struct kvm_irqchip *chip = kmalloc(sizeof(*chip), GFP_KERNEL); 3218 struct kvm_irqchip *chip;
3689 3219
3690 r = -ENOMEM; 3220 chip = memdup_user(argp, sizeof(*chip));
3691 if (!chip) 3221 if (IS_ERR(chip)) {
3222 r = PTR_ERR(chip);
3692 goto out; 3223 goto out;
3693 r = -EFAULT; 3224 }
3694 if (copy_from_user(chip, argp, sizeof *chip)) 3225
3695 goto set_irqchip_out;
3696 r = -ENXIO; 3226 r = -ENXIO;
3697 if (!irqchip_in_kernel(kvm)) 3227 if (!irqchip_in_kernel(kvm))
3698 goto set_irqchip_out; 3228 goto set_irqchip_out;
@@ -3899,12 +3429,7 @@ void kvm_get_segment(struct kvm_vcpu *vcpu,
3899 kvm_x86_ops->get_segment(vcpu, var, seg); 3429 kvm_x86_ops->get_segment(vcpu, var, seg);
3900} 3430}
3901 3431
3902static gpa_t translate_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access) 3432gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3903{
3904 return gpa;
3905}
3906
3907static gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access)
3908{ 3433{
3909 gpa_t t_gpa; 3434 gpa_t t_gpa;
3910 struct x86_exception exception; 3435 struct x86_exception exception;
@@ -4088,7 +3613,7 @@ int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
4088 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes); 3613 ret = kvm_write_guest(vcpu->kvm, gpa, val, bytes);
4089 if (ret < 0) 3614 if (ret < 0)
4090 return 0; 3615 return 0;
4091 kvm_mmu_pte_write(vcpu, gpa, val, bytes, 1); 3616 kvm_mmu_pte_write(vcpu, gpa, val, bytes);
4092 return 1; 3617 return 1;
4093} 3618}
4094 3619
@@ -4325,7 +3850,7 @@ static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
4325 if (!exchanged) 3850 if (!exchanged)
4326 return X86EMUL_CMPXCHG_FAILED; 3851 return X86EMUL_CMPXCHG_FAILED;
4327 3852
4328 kvm_mmu_pte_write(vcpu, gpa, new, bytes, 1); 3853 kvm_mmu_pte_write(vcpu, gpa, new, bytes);
4329 3854
4330 return X86EMUL_CONTINUE; 3855 return X86EMUL_CONTINUE;
4331 3856
@@ -4350,32 +3875,24 @@ static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
4350 return r; 3875 return r;
4351} 3876}
4352 3877
4353 3878static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
4354static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt, 3879 unsigned short port, void *val,
4355 int size, unsigned short port, void *val, 3880 unsigned int count, bool in)
4356 unsigned int count)
4357{ 3881{
4358 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); 3882 trace_kvm_pio(!in, port, size, count);
4359
4360 if (vcpu->arch.pio.count)
4361 goto data_avail;
4362
4363 trace_kvm_pio(0, port, size, count);
4364 3883
4365 vcpu->arch.pio.port = port; 3884 vcpu->arch.pio.port = port;
4366 vcpu->arch.pio.in = 1; 3885 vcpu->arch.pio.in = in;
4367 vcpu->arch.pio.count = count; 3886 vcpu->arch.pio.count = count;
4368 vcpu->arch.pio.size = size; 3887 vcpu->arch.pio.size = size;
4369 3888
4370 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) { 3889 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
4371 data_avail:
4372 memcpy(val, vcpu->arch.pio_data, size * count);
4373 vcpu->arch.pio.count = 0; 3890 vcpu->arch.pio.count = 0;
4374 return 1; 3891 return 1;
4375 } 3892 }
4376 3893
4377 vcpu->run->exit_reason = KVM_EXIT_IO; 3894 vcpu->run->exit_reason = KVM_EXIT_IO;
4378 vcpu->run->io.direction = KVM_EXIT_IO_IN; 3895 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
4379 vcpu->run->io.size = size; 3896 vcpu->run->io.size = size;
4380 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE; 3897 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4381 vcpu->run->io.count = count; 3898 vcpu->run->io.count = count;
@@ -4384,36 +3901,37 @@ static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4384 return 0; 3901 return 0;
4385} 3902}
4386 3903
4387static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt, 3904static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
4388 int size, unsigned short port, 3905 int size, unsigned short port, void *val,
4389 const void *val, unsigned int count) 3906 unsigned int count)
4390{ 3907{
4391 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt); 3908 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3909 int ret;
4392 3910
4393 trace_kvm_pio(1, port, size, count); 3911 if (vcpu->arch.pio.count)
4394 3912 goto data_avail;
4395 vcpu->arch.pio.port = port;
4396 vcpu->arch.pio.in = 0;
4397 vcpu->arch.pio.count = count;
4398 vcpu->arch.pio.size = size;
4399
4400 memcpy(vcpu->arch.pio_data, val, size * count);
4401 3913
4402 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) { 3914 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
3915 if (ret) {
3916data_avail:
3917 memcpy(val, vcpu->arch.pio_data, size * count);
4403 vcpu->arch.pio.count = 0; 3918 vcpu->arch.pio.count = 0;
4404 return 1; 3919 return 1;
4405 } 3920 }
4406 3921
4407 vcpu->run->exit_reason = KVM_EXIT_IO;
4408 vcpu->run->io.direction = KVM_EXIT_IO_OUT;
4409 vcpu->run->io.size = size;
4410 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
4411 vcpu->run->io.count = count;
4412 vcpu->run->io.port = port;
4413
4414 return 0; 3922 return 0;
4415} 3923}
4416 3924
3925static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
3926 int size, unsigned short port,
3927 const void *val, unsigned int count)
3928{
3929 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
3930
3931 memcpy(vcpu->arch.pio_data, val, size * count);
3932 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
3933}
3934
4417static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg) 3935static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
4418{ 3936{
4419 return kvm_x86_ops->get_segment_base(vcpu, seg); 3937 return kvm_x86_ops->get_segment_base(vcpu, seg);
@@ -4628,6 +4146,12 @@ static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
4628 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data); 4146 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
4629} 4147}
4630 4148
4149static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
4150 u32 pmc, u64 *pdata)
4151{
4152 return kvm_pmu_read_pmc(emul_to_vcpu(ctxt), pmc, pdata);
4153}
4154
4631static void emulator_halt(struct x86_emulate_ctxt *ctxt) 4155static void emulator_halt(struct x86_emulate_ctxt *ctxt)
4632{ 4156{
4633 emul_to_vcpu(ctxt)->arch.halt_request = 1; 4157 emul_to_vcpu(ctxt)->arch.halt_request = 1;
@@ -4680,6 +4204,7 @@ static struct x86_emulate_ops emulate_ops = {
4680 .set_dr = emulator_set_dr, 4204 .set_dr = emulator_set_dr,
4681 .set_msr = emulator_set_msr, 4205 .set_msr = emulator_set_msr,
4682 .get_msr = emulator_get_msr, 4206 .get_msr = emulator_get_msr,
4207 .read_pmc = emulator_read_pmc,
4683 .halt = emulator_halt, 4208 .halt = emulator_halt,
4684 .wbinvd = emulator_wbinvd, 4209 .wbinvd = emulator_wbinvd,
4685 .fix_hypercall = emulator_fix_hypercall, 4210 .fix_hypercall = emulator_fix_hypercall,
@@ -4837,6 +4362,50 @@ static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t gva)
4837 return false; 4362 return false;
4838} 4363}
4839 4364
4365static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
4366 unsigned long cr2, int emulation_type)
4367{
4368 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
4369 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
4370
4371 last_retry_eip = vcpu->arch.last_retry_eip;
4372 last_retry_addr = vcpu->arch.last_retry_addr;
4373
4374 /*
4375 * If the emulation is caused by #PF and it is non-page_table
4376 * writing instruction, it means the VM-EXIT is caused by shadow
4377 * page protected, we can zap the shadow page and retry this
4378 * instruction directly.
4379 *
4380 * Note: if the guest uses a non-page-table modifying instruction
4381 * on the PDE that points to the instruction, then we will unmap
4382 * the instruction and go to an infinite loop. So, we cache the
4383 * last retried eip and the last fault address, if we meet the eip
4384 * and the address again, we can break out of the potential infinite
4385 * loop.
4386 */
4387 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
4388
4389 if (!(emulation_type & EMULTYPE_RETRY))
4390 return false;
4391
4392 if (x86_page_table_writing_insn(ctxt))
4393 return false;
4394
4395 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
4396 return false;
4397
4398 vcpu->arch.last_retry_eip = ctxt->eip;
4399 vcpu->arch.last_retry_addr = cr2;
4400
4401 if (!vcpu->arch.mmu.direct_map)
4402 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
4403
4404 kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT);
4405
4406 return true;
4407}
4408
4840int x86_emulate_instruction(struct kvm_vcpu *vcpu, 4409int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4841 unsigned long cr2, 4410 unsigned long cr2,
4842 int emulation_type, 4411 int emulation_type,
@@ -4878,6 +4447,9 @@ int x86_emulate_instruction(struct kvm_vcpu *vcpu,
4878 return EMULATE_DONE; 4447 return EMULATE_DONE;
4879 } 4448 }
4880 4449
4450 if (retry_instruction(ctxt, cr2, emulation_type))
4451 return EMULATE_DONE;
4452
4881 /* this is needed for vmware backdoor interface to work since it 4453 /* this is needed for vmware backdoor interface to work since it
4882 changes registers values during IO operation */ 4454 changes registers values during IO operation */
4883 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) { 4455 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
@@ -5096,17 +4668,17 @@ static void kvm_timer_init(void)
5096 4668
5097static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu); 4669static DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
5098 4670
5099static int kvm_is_in_guest(void) 4671int kvm_is_in_guest(void)
5100{ 4672{
5101 return percpu_read(current_vcpu) != NULL; 4673 return __this_cpu_read(current_vcpu) != NULL;
5102} 4674}
5103 4675
5104static int kvm_is_user_mode(void) 4676static int kvm_is_user_mode(void)
5105{ 4677{
5106 int user_mode = 3; 4678 int user_mode = 3;
5107 4679
5108 if (percpu_read(current_vcpu)) 4680 if (__this_cpu_read(current_vcpu))
5109 user_mode = kvm_x86_ops->get_cpl(percpu_read(current_vcpu)); 4681 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
5110 4682
5111 return user_mode != 0; 4683 return user_mode != 0;
5112} 4684}
@@ -5115,8 +4687,8 @@ static unsigned long kvm_get_guest_ip(void)
5115{ 4687{
5116 unsigned long ip = 0; 4688 unsigned long ip = 0;
5117 4689
5118 if (percpu_read(current_vcpu)) 4690 if (__this_cpu_read(current_vcpu))
5119 ip = kvm_rip_read(percpu_read(current_vcpu)); 4691 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
5120 4692
5121 return ip; 4693 return ip;
5122} 4694}
@@ -5129,13 +4701,13 @@ static struct perf_guest_info_callbacks kvm_guest_cbs = {
5129 4701
5130void kvm_before_handle_nmi(struct kvm_vcpu *vcpu) 4702void kvm_before_handle_nmi(struct kvm_vcpu *vcpu)
5131{ 4703{
5132 percpu_write(current_vcpu, vcpu); 4704 __this_cpu_write(current_vcpu, vcpu);
5133} 4705}
5134EXPORT_SYMBOL_GPL(kvm_before_handle_nmi); 4706EXPORT_SYMBOL_GPL(kvm_before_handle_nmi);
5135 4707
5136void kvm_after_handle_nmi(struct kvm_vcpu *vcpu) 4708void kvm_after_handle_nmi(struct kvm_vcpu *vcpu)
5137{ 4709{
5138 percpu_write(current_vcpu, NULL); 4710 __this_cpu_write(current_vcpu, NULL);
5139} 4711}
5140EXPORT_SYMBOL_GPL(kvm_after_handle_nmi); 4712EXPORT_SYMBOL_GPL(kvm_after_handle_nmi);
5141 4713
@@ -5234,15 +4806,6 @@ int kvm_emulate_halt(struct kvm_vcpu *vcpu)
5234} 4806}
5235EXPORT_SYMBOL_GPL(kvm_emulate_halt); 4807EXPORT_SYMBOL_GPL(kvm_emulate_halt);
5236 4808
5237static inline gpa_t hc_gpa(struct kvm_vcpu *vcpu, unsigned long a0,
5238 unsigned long a1)
5239{
5240 if (is_long_mode(vcpu))
5241 return a0;
5242 else
5243 return a0 | ((gpa_t)a1 << 32);
5244}
5245
5246int kvm_hv_hypercall(struct kvm_vcpu *vcpu) 4809int kvm_hv_hypercall(struct kvm_vcpu *vcpu)
5247{ 4810{
5248 u64 param, ingpa, outgpa, ret; 4811 u64 param, ingpa, outgpa, ret;
@@ -5338,9 +4901,6 @@ int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
5338 case KVM_HC_VAPIC_POLL_IRQ: 4901 case KVM_HC_VAPIC_POLL_IRQ:
5339 ret = 0; 4902 ret = 0;
5340 break; 4903 break;
5341 case KVM_HC_MMU_OP:
5342 r = kvm_pv_mmu_op(vcpu, a0, hc_gpa(vcpu, a1, a2), &ret);
5343 break;
5344 default: 4904 default:
5345 ret = -KVM_ENOSYS; 4905 ret = -KVM_ENOSYS;
5346 break; 4906 break;
@@ -5370,125 +4930,6 @@ int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
5370 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL); 4930 return emulator_write_emulated(ctxt, rip, instruction, 3, NULL);
5371} 4931}
5372 4932
5373static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
5374{
5375 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
5376 int j, nent = vcpu->arch.cpuid_nent;
5377
5378 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
5379 /* when no next entry is found, the current entry[i] is reselected */
5380 for (j = i + 1; ; j = (j + 1) % nent) {
5381 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
5382 if (ej->function == e->function) {
5383 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
5384 return j;
5385 }
5386 }
5387 return 0; /* silence gcc, even though control never reaches here */
5388}
5389
5390/* find an entry with matching function, matching index (if needed), and that
5391 * should be read next (if it's stateful) */
5392static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
5393 u32 function, u32 index)
5394{
5395 if (e->function != function)
5396 return 0;
5397 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
5398 return 0;
5399 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
5400 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
5401 return 0;
5402 return 1;
5403}
5404
5405struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
5406 u32 function, u32 index)
5407{
5408 int i;
5409 struct kvm_cpuid_entry2 *best = NULL;
5410
5411 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
5412 struct kvm_cpuid_entry2 *e;
5413
5414 e = &vcpu->arch.cpuid_entries[i];
5415 if (is_matching_cpuid_entry(e, function, index)) {
5416 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
5417 move_to_next_stateful_cpuid_entry(vcpu, i);
5418 best = e;
5419 break;
5420 }
5421 }
5422 return best;
5423}
5424EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
5425
5426int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
5427{
5428 struct kvm_cpuid_entry2 *best;
5429
5430 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
5431 if (!best || best->eax < 0x80000008)
5432 goto not_found;
5433 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
5434 if (best)
5435 return best->eax & 0xff;
5436not_found:
5437 return 36;
5438}
5439
5440/*
5441 * If no match is found, check whether we exceed the vCPU's limit
5442 * and return the content of the highest valid _standard_ leaf instead.
5443 * This is to satisfy the CPUID specification.
5444 */
5445static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
5446 u32 function, u32 index)
5447{
5448 struct kvm_cpuid_entry2 *maxlevel;
5449
5450 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
5451 if (!maxlevel || maxlevel->eax >= function)
5452 return NULL;
5453 if (function & 0x80000000) {
5454 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
5455 if (!maxlevel)
5456 return NULL;
5457 }
5458 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
5459}
5460
5461void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
5462{
5463 u32 function, index;
5464 struct kvm_cpuid_entry2 *best;
5465
5466 function = kvm_register_read(vcpu, VCPU_REGS_RAX);
5467 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5468 kvm_register_write(vcpu, VCPU_REGS_RAX, 0);
5469 kvm_register_write(vcpu, VCPU_REGS_RBX, 0);
5470 kvm_register_write(vcpu, VCPU_REGS_RCX, 0);
5471 kvm_register_write(vcpu, VCPU_REGS_RDX, 0);
5472 best = kvm_find_cpuid_entry(vcpu, function, index);
5473
5474 if (!best)
5475 best = check_cpuid_limit(vcpu, function, index);
5476
5477 if (best) {
5478 kvm_register_write(vcpu, VCPU_REGS_RAX, best->eax);
5479 kvm_register_write(vcpu, VCPU_REGS_RBX, best->ebx);
5480 kvm_register_write(vcpu, VCPU_REGS_RCX, best->ecx);
5481 kvm_register_write(vcpu, VCPU_REGS_RDX, best->edx);
5482 }
5483 kvm_x86_ops->skip_emulated_instruction(vcpu);
5484 trace_kvm_cpuid(function,
5485 kvm_register_read(vcpu, VCPU_REGS_RAX),
5486 kvm_register_read(vcpu, VCPU_REGS_RBX),
5487 kvm_register_read(vcpu, VCPU_REGS_RCX),
5488 kvm_register_read(vcpu, VCPU_REGS_RDX));
5489}
5490EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
5491
5492/* 4933/*
5493 * Check if userspace requested an interrupt window, and that the 4934 * Check if userspace requested an interrupt window, and that the
5494 * interrupt window is open. 4935 * interrupt window is open.
@@ -5649,6 +5090,7 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5649 int r; 5090 int r;
5650 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) && 5091 bool req_int_win = !irqchip_in_kernel(vcpu->kvm) &&
5651 vcpu->run->request_interrupt_window; 5092 vcpu->run->request_interrupt_window;
5093 bool req_immediate_exit = 0;
5652 5094
5653 if (vcpu->requests) { 5095 if (vcpu->requests) {
5654 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu)) 5096 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
@@ -5688,7 +5130,12 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5688 record_steal_time(vcpu); 5130 record_steal_time(vcpu);
5689 if (kvm_check_request(KVM_REQ_NMI, vcpu)) 5131 if (kvm_check_request(KVM_REQ_NMI, vcpu))
5690 process_nmi(vcpu); 5132 process_nmi(vcpu);
5691 5133 req_immediate_exit =
5134 kvm_check_request(KVM_REQ_IMMEDIATE_EXIT, vcpu);
5135 if (kvm_check_request(KVM_REQ_PMU, vcpu))
5136 kvm_handle_pmu_event(vcpu);
5137 if (kvm_check_request(KVM_REQ_PMI, vcpu))
5138 kvm_deliver_pmi(vcpu);
5692 } 5139 }
5693 5140
5694 r = kvm_mmu_reload(vcpu); 5141 r = kvm_mmu_reload(vcpu);
@@ -5739,6 +5186,9 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
5739 5186
5740 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx); 5187 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
5741 5188
5189 if (req_immediate_exit)
5190 smp_send_reschedule(vcpu->cpu);
5191
5742 kvm_guest_enter(); 5192 kvm_guest_enter();
5743 5193
5744 if (unlikely(vcpu->arch.switch_db_regs)) { 5194 if (unlikely(vcpu->arch.switch_db_regs)) {
@@ -5944,10 +5394,6 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
5944 if (r <= 0) 5394 if (r <= 0)
5945 goto out; 5395 goto out;
5946 5396
5947 if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL)
5948 kvm_register_write(vcpu, VCPU_REGS_RAX,
5949 kvm_run->hypercall.ret);
5950
5951 r = __vcpu_run(vcpu); 5397 r = __vcpu_run(vcpu);
5952 5398
5953out: 5399out:
@@ -6149,7 +5595,7 @@ int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
6149 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4; 5595 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
6150 kvm_x86_ops->set_cr4(vcpu, sregs->cr4); 5596 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
6151 if (sregs->cr4 & X86_CR4_OSXSAVE) 5597 if (sregs->cr4 & X86_CR4_OSXSAVE)
6152 update_cpuid(vcpu); 5598 kvm_update_cpuid(vcpu);
6153 5599
6154 idx = srcu_read_lock(&vcpu->kvm->srcu); 5600 idx = srcu_read_lock(&vcpu->kvm->srcu);
6155 if (!is_long_mode(vcpu) && is_pae(vcpu)) { 5601 if (!is_long_mode(vcpu) && is_pae(vcpu)) {
@@ -6426,6 +5872,8 @@ int kvm_arch_vcpu_reset(struct kvm_vcpu *vcpu)
6426 kvm_async_pf_hash_reset(vcpu); 5872 kvm_async_pf_hash_reset(vcpu);
6427 vcpu->arch.apf.halted = false; 5873 vcpu->arch.apf.halted = false;
6428 5874
5875 kvm_pmu_reset(vcpu);
5876
6429 return kvm_x86_ops->vcpu_reset(vcpu); 5877 return kvm_x86_ops->vcpu_reset(vcpu);
6430} 5878}
6431 5879
@@ -6474,10 +5922,6 @@ int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6474 kvm = vcpu->kvm; 5922 kvm = vcpu->kvm;
6475 5923
6476 vcpu->arch.emulate_ctxt.ops = &emulate_ops; 5924 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
6477 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
6478 vcpu->arch.mmu.root_hpa = INVALID_PAGE;
6479 vcpu->arch.mmu.translate_gpa = translate_gpa;
6480 vcpu->arch.nested_mmu.translate_gpa = translate_nested_gpa;
6481 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu)) 5925 if (!irqchip_in_kernel(kvm) || kvm_vcpu_is_bsp(vcpu))
6482 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE; 5926 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
6483 else 5927 else
@@ -6514,6 +5958,7 @@ int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
6514 goto fail_free_mce_banks; 5958 goto fail_free_mce_banks;
6515 5959
6516 kvm_async_pf_hash_reset(vcpu); 5960 kvm_async_pf_hash_reset(vcpu);
5961 kvm_pmu_init(vcpu);
6517 5962
6518 return 0; 5963 return 0;
6519fail_free_mce_banks: 5964fail_free_mce_banks:
@@ -6532,6 +5977,7 @@ void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
6532{ 5977{
6533 int idx; 5978 int idx;
6534 5979
5980 kvm_pmu_destroy(vcpu);
6535 kfree(vcpu->arch.mce_banks); 5981 kfree(vcpu->arch.mce_banks);
6536 kvm_free_lapic(vcpu); 5982 kvm_free_lapic(vcpu);
6537 idx = srcu_read_lock(&vcpu->kvm->srcu); 5983 idx = srcu_read_lock(&vcpu->kvm->srcu);
diff --git a/arch/x86/kvm/x86.h b/arch/x86/kvm/x86.h
index d36fe237c665..cb80c293cdd8 100644
--- a/arch/x86/kvm/x86.h
+++ b/arch/x86/kvm/x86.h
@@ -33,9 +33,6 @@ static inline bool kvm_exception_is_soft(unsigned int nr)
33 return (nr == BP_VECTOR) || (nr == OF_VECTOR); 33 return (nr == BP_VECTOR) || (nr == OF_VECTOR);
34} 34}
35 35
36struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
37 u32 function, u32 index);
38
39static inline bool is_protmode(struct kvm_vcpu *vcpu) 36static inline bool is_protmode(struct kvm_vcpu *vcpu)
40{ 37{
41 return kvm_read_cr0_bits(vcpu, X86_CR0_PE); 38 return kvm_read_cr0_bits(vcpu, X86_CR0_PE);
@@ -125,4 +122,6 @@ int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
125 gva_t addr, void *val, unsigned int bytes, 122 gva_t addr, void *val, unsigned int bytes,
126 struct x86_exception *exception); 123 struct x86_exception *exception);
127 124
125extern u64 host_xcr0;
126
128#endif 127#endif
generated by cgit v1.2.2 (git 2.25.0) at 2025-02-25 17:51:49 -0500