]> git.ipfire.org Git - thirdparty/kernel/stable.git/blob - arch/x86/kvm/mtrr.c
9f72cc427158e637b1852e9843c5eddfcf78c43a
[thirdparty/kernel/stable.git] / arch / x86 / kvm / mtrr.c
1 /*
2 * vMTRR implementation
3 *
4 * Copyright (C) 2006 Qumranet, Inc.
5 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
6 * Copyright(C) 2015 Intel Corporation.
7 *
8 * Authors:
9 * Yaniv Kamay <yaniv@qumranet.com>
10 * Avi Kivity <avi@qumranet.com>
11 * Marcelo Tosatti <mtosatti@redhat.com>
12 * Paolo Bonzini <pbonzini@redhat.com>
13 * Xiao Guangrong <guangrong.xiao@linux.intel.com>
14 *
15 * This work is licensed under the terms of the GNU GPL, version 2. See
16 * the COPYING file in the top-level directory.
17 */
18
19 #include <linux/kvm_host.h>
20 #include <asm/mtrr.h>
21
22 #include "cpuid.h"
23 #include "mmu.h"
24
25 #define IA32_MTRR_DEF_TYPE_E (1ULL << 11)
26 #define IA32_MTRR_DEF_TYPE_FE (1ULL << 10)
27 #define IA32_MTRR_DEF_TYPE_TYPE_MASK (0xff)
28
29 static bool msr_mtrr_valid(unsigned msr)
30 {
31 switch (msr) {
32 case 0x200 ... 0x200 + 2 * KVM_NR_VAR_MTRR - 1:
33 case MSR_MTRRfix64K_00000:
34 case MSR_MTRRfix16K_80000:
35 case MSR_MTRRfix16K_A0000:
36 case MSR_MTRRfix4K_C0000:
37 case MSR_MTRRfix4K_C8000:
38 case MSR_MTRRfix4K_D0000:
39 case MSR_MTRRfix4K_D8000:
40 case MSR_MTRRfix4K_E0000:
41 case MSR_MTRRfix4K_E8000:
42 case MSR_MTRRfix4K_F0000:
43 case MSR_MTRRfix4K_F8000:
44 case MSR_MTRRdefType:
45 case MSR_IA32_CR_PAT:
46 return true;
47 }
48 return false;
49 }
50
51 static bool valid_mtrr_type(unsigned t)
52 {
53 return t < 8 && (1 << t) & 0x73; /* 0, 1, 4, 5, 6 */
54 }
55
56 bool kvm_mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data)
57 {
58 int i;
59 u64 mask;
60
61 if (!msr_mtrr_valid(msr))
62 return false;
63
64 if (msr == MSR_IA32_CR_PAT) {
65 return kvm_pat_valid(data);
66 } else if (msr == MSR_MTRRdefType) {
67 if (data & ~0xcff)
68 return false;
69 return valid_mtrr_type(data & 0xff);
70 } else if (msr >= MSR_MTRRfix64K_00000 && msr <= MSR_MTRRfix4K_F8000) {
71 for (i = 0; i < 8 ; i++)
72 if (!valid_mtrr_type((data >> (i * 8)) & 0xff))
73 return false;
74 return true;
75 }
76
77 /* variable MTRRs */
78 WARN_ON(!(msr >= 0x200 && msr < 0x200 + 2 * KVM_NR_VAR_MTRR));
79
80 mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
81 if ((msr & 1) == 0) {
82 /* MTRR base */
83 if (!valid_mtrr_type(data & 0xff))
84 return false;
85 mask |= 0xf00;
86 } else
87 /* MTRR mask */
88 mask |= 0x7ff;
89 if (data & mask) {
90 kvm_inject_gp(vcpu, 0);
91 return false;
92 }
93
94 return true;
95 }
96 EXPORT_SYMBOL_GPL(kvm_mtrr_valid);
97
98 static bool mtrr_is_enabled(struct kvm_mtrr *mtrr_state)
99 {
100 return !!(mtrr_state->deftype & IA32_MTRR_DEF_TYPE_E);
101 }
102
103 static bool fixed_mtrr_is_enabled(struct kvm_mtrr *mtrr_state)
104 {
105 return !!(mtrr_state->deftype & IA32_MTRR_DEF_TYPE_FE);
106 }
107
108 static u8 mtrr_default_type(struct kvm_mtrr *mtrr_state)
109 {
110 return mtrr_state->deftype & IA32_MTRR_DEF_TYPE_TYPE_MASK;
111 }
112
113 static u8 mtrr_disabled_type(struct kvm_vcpu *vcpu)
114 {
115 /*
116 * Intel SDM 11.11.2.2: all MTRRs are disabled when
117 * IA32_MTRR_DEF_TYPE.E bit is cleared, and the UC
118 * memory type is applied to all of physical memory.
119 *
120 * However, virtual machines can be run with CPUID such that
121 * there are no MTRRs. In that case, the firmware will never
122 * enable MTRRs and it is obviously undesirable to run the
123 * guest entirely with UC memory and we use WB.
124 */
125 if (guest_cpuid_has(vcpu, X86_FEATURE_MTRR))
126 return MTRR_TYPE_UNCACHABLE;
127 else
128 return MTRR_TYPE_WRBACK;
129 }
130
131 /*
132 * Three terms are used in the following code:
133 * - segment, it indicates the address segments covered by fixed MTRRs.
134 * - unit, it corresponds to the MSR entry in the segment.
135 * - range, a range is covered in one memory cache type.
136 */
137 struct fixed_mtrr_segment {
138 u64 start;
139 u64 end;
140
141 int range_shift;
142
143 /* the start position in kvm_mtrr.fixed_ranges[]. */
144 int range_start;
145 };
146
147 static struct fixed_mtrr_segment fixed_seg_table[] = {
148 /* MSR_MTRRfix64K_00000, 1 unit. 64K fixed mtrr. */
149 {
150 .start = 0x0,
151 .end = 0x80000,
152 .range_shift = 16, /* 64K */
153 .range_start = 0,
154 },
155
156 /*
157 * MSR_MTRRfix16K_80000 ... MSR_MTRRfix16K_A0000, 2 units,
158 * 16K fixed mtrr.
159 */
160 {
161 .start = 0x80000,
162 .end = 0xc0000,
163 .range_shift = 14, /* 16K */
164 .range_start = 8,
165 },
166
167 /*
168 * MSR_MTRRfix4K_C0000 ... MSR_MTRRfix4K_F8000, 8 units,
169 * 4K fixed mtrr.
170 */
171 {
172 .start = 0xc0000,
173 .end = 0x100000,
174 .range_shift = 12, /* 12K */
175 .range_start = 24,
176 }
177 };
178
179 /*
180 * The size of unit is covered in one MSR, one MSR entry contains
181 * 8 ranges so that unit size is always 8 * 2^range_shift.
182 */
183 static u64 fixed_mtrr_seg_unit_size(int seg)
184 {
185 return 8 << fixed_seg_table[seg].range_shift;
186 }
187
188 static bool fixed_msr_to_seg_unit(u32 msr, int *seg, int *unit)
189 {
190 switch (msr) {
191 case MSR_MTRRfix64K_00000:
192 *seg = 0;
193 *unit = 0;
194 break;
195 case MSR_MTRRfix16K_80000 ... MSR_MTRRfix16K_A0000:
196 *seg = 1;
197 *unit = msr - MSR_MTRRfix16K_80000;
198 break;
199 case MSR_MTRRfix4K_C0000 ... MSR_MTRRfix4K_F8000:
200 *seg = 2;
201 *unit = msr - MSR_MTRRfix4K_C0000;
202 break;
203 default:
204 return false;
205 }
206
207 return true;
208 }
209
210 static void fixed_mtrr_seg_unit_range(int seg, int unit, u64 *start, u64 *end)
211 {
212 struct fixed_mtrr_segment *mtrr_seg = &fixed_seg_table[seg];
213 u64 unit_size = fixed_mtrr_seg_unit_size(seg);
214
215 *start = mtrr_seg->start + unit * unit_size;
216 *end = *start + unit_size;
217 WARN_ON(*end > mtrr_seg->end);
218 }
219
220 static int fixed_mtrr_seg_unit_range_index(int seg, int unit)
221 {
222 struct fixed_mtrr_segment *mtrr_seg = &fixed_seg_table[seg];
223
224 WARN_ON(mtrr_seg->start + unit * fixed_mtrr_seg_unit_size(seg)
225 > mtrr_seg->end);
226
227 /* each unit has 8 ranges. */
228 return mtrr_seg->range_start + 8 * unit;
229 }
230
231 static int fixed_mtrr_seg_end_range_index(int seg)
232 {
233 struct fixed_mtrr_segment *mtrr_seg = &fixed_seg_table[seg];
234 int n;
235
236 n = (mtrr_seg->end - mtrr_seg->start) >> mtrr_seg->range_shift;
237 return mtrr_seg->range_start + n - 1;
238 }
239
240 static bool fixed_msr_to_range(u32 msr, u64 *start, u64 *end)
241 {
242 int seg, unit;
243
244 if (!fixed_msr_to_seg_unit(msr, &seg, &unit))
245 return false;
246
247 fixed_mtrr_seg_unit_range(seg, unit, start, end);
248 return true;
249 }
250
251 static int fixed_msr_to_range_index(u32 msr)
252 {
253 int seg, unit;
254
255 if (!fixed_msr_to_seg_unit(msr, &seg, &unit))
256 return -1;
257
258 return fixed_mtrr_seg_unit_range_index(seg, unit);
259 }
260
261 static int fixed_mtrr_addr_to_seg(u64 addr)
262 {
263 struct fixed_mtrr_segment *mtrr_seg;
264 int seg, seg_num = ARRAY_SIZE(fixed_seg_table);
265
266 for (seg = 0; seg < seg_num; seg++) {
267 mtrr_seg = &fixed_seg_table[seg];
268 if (mtrr_seg->start <= addr && addr < mtrr_seg->end)
269 return seg;
270 }
271
272 return -1;
273 }
274
275 static int fixed_mtrr_addr_seg_to_range_index(u64 addr, int seg)
276 {
277 struct fixed_mtrr_segment *mtrr_seg;
278 int index;
279
280 mtrr_seg = &fixed_seg_table[seg];
281 index = mtrr_seg->range_start;
282 index += (addr - mtrr_seg->start) >> mtrr_seg->range_shift;
283 return index;
284 }
285
286 static u64 fixed_mtrr_range_end_addr(int seg, int index)
287 {
288 struct fixed_mtrr_segment *mtrr_seg = &fixed_seg_table[seg];
289 int pos = index - mtrr_seg->range_start;
290
291 return mtrr_seg->start + ((pos + 1) << mtrr_seg->range_shift);
292 }
293
294 static void var_mtrr_range(struct kvm_mtrr_range *range, u64 *start, u64 *end)
295 {
296 u64 mask;
297
298 *start = range->base & PAGE_MASK;
299
300 mask = range->mask & PAGE_MASK;
301
302 /* This cannot overflow because writing to the reserved bits of
303 * variable MTRRs causes a #GP.
304 */
305 *end = (*start | ~mask) + 1;
306 }
307
308 static void update_mtrr(struct kvm_vcpu *vcpu, u32 msr)
309 {
310 struct kvm_mtrr *mtrr_state = &vcpu->arch.mtrr_state;
311 gfn_t start, end;
312 int index;
313
314 if (msr == MSR_IA32_CR_PAT || !tdp_enabled ||
315 !kvm_arch_has_noncoherent_dma(vcpu->kvm))
316 return;
317
318 if (!mtrr_is_enabled(mtrr_state) && msr != MSR_MTRRdefType)
319 return;
320
321 /* fixed MTRRs. */
322 if (fixed_msr_to_range(msr, &start, &end)) {
323 if (!fixed_mtrr_is_enabled(mtrr_state))
324 return;
325 } else if (msr == MSR_MTRRdefType) {
326 start = 0x0;
327 end = ~0ULL;
328 } else {
329 /* variable range MTRRs. */
330 index = (msr - 0x200) / 2;
331 var_mtrr_range(&mtrr_state->var_ranges[index], &start, &end);
332 }
333
334 kvm_zap_gfn_range(vcpu->kvm, gpa_to_gfn(start), gpa_to_gfn(end));
335 }
336
337 static bool var_mtrr_range_is_valid(struct kvm_mtrr_range *range)
338 {
339 return (range->mask & (1 << 11)) != 0;
340 }
341
342 static void set_var_mtrr_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
343 {
344 struct kvm_mtrr *mtrr_state = &vcpu->arch.mtrr_state;
345 struct kvm_mtrr_range *tmp, *cur;
346 int index, is_mtrr_mask;
347
348 index = (msr - 0x200) / 2;
349 is_mtrr_mask = msr - 0x200 - 2 * index;
350 cur = &mtrr_state->var_ranges[index];
351
352 /* remove the entry if it's in the list. */
353 if (var_mtrr_range_is_valid(cur))
354 list_del(&mtrr_state->var_ranges[index].node);
355
356 /* Extend the mask with all 1 bits to the left, since those
357 * bits must implicitly be 0. The bits are then cleared
358 * when reading them.
359 */
360 if (!is_mtrr_mask)
361 cur->base = data;
362 else
363 cur->mask = data | (-1LL << cpuid_maxphyaddr(vcpu));
364
365 /* add it to the list if it's enabled. */
366 if (var_mtrr_range_is_valid(cur)) {
367 list_for_each_entry(tmp, &mtrr_state->head, node)
368 if (cur->base >= tmp->base)
369 break;
370 list_add_tail(&cur->node, &tmp->node);
371 }
372 }
373
374 int kvm_mtrr_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
375 {
376 int index;
377
378 if (!kvm_mtrr_valid(vcpu, msr, data))
379 return 1;
380
381 index = fixed_msr_to_range_index(msr);
382 if (index >= 0)
383 *(u64 *)&vcpu->arch.mtrr_state.fixed_ranges[index] = data;
384 else if (msr == MSR_MTRRdefType)
385 vcpu->arch.mtrr_state.deftype = data;
386 else if (msr == MSR_IA32_CR_PAT)
387 vcpu->arch.pat = data;
388 else
389 set_var_mtrr_msr(vcpu, msr, data);
390
391 update_mtrr(vcpu, msr);
392 return 0;
393 }
394
395 int kvm_mtrr_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
396 {
397 int index;
398
399 /* MSR_MTRRcap is a readonly MSR. */
400 if (msr == MSR_MTRRcap) {
401 /*
402 * SMRR = 0
403 * WC = 1
404 * FIX = 1
405 * VCNT = KVM_NR_VAR_MTRR
406 */
407 *pdata = 0x500 | KVM_NR_VAR_MTRR;
408 return 0;
409 }
410
411 if (!msr_mtrr_valid(msr))
412 return 1;
413
414 index = fixed_msr_to_range_index(msr);
415 if (index >= 0)
416 *pdata = *(u64 *)&vcpu->arch.mtrr_state.fixed_ranges[index];
417 else if (msr == MSR_MTRRdefType)
418 *pdata = vcpu->arch.mtrr_state.deftype;
419 else if (msr == MSR_IA32_CR_PAT)
420 *pdata = vcpu->arch.pat;
421 else { /* Variable MTRRs */
422 int is_mtrr_mask;
423
424 index = (msr - 0x200) / 2;
425 is_mtrr_mask = msr - 0x200 - 2 * index;
426 if (!is_mtrr_mask)
427 *pdata = vcpu->arch.mtrr_state.var_ranges[index].base;
428 else
429 *pdata = vcpu->arch.mtrr_state.var_ranges[index].mask;
430
431 *pdata &= (1ULL << cpuid_maxphyaddr(vcpu)) - 1;
432 }
433
434 return 0;
435 }
436
437 void kvm_vcpu_mtrr_init(struct kvm_vcpu *vcpu)
438 {
439 INIT_LIST_HEAD(&vcpu->arch.mtrr_state.head);
440 }
441
442 struct mtrr_iter {
443 /* input fields. */
444 struct kvm_mtrr *mtrr_state;
445 u64 start;
446 u64 end;
447
448 /* output fields. */
449 int mem_type;
450 /* mtrr is completely disabled? */
451 bool mtrr_disabled;
452 /* [start, end) is not fully covered in MTRRs? */
453 bool partial_map;
454
455 /* private fields. */
456 union {
457 /* used for fixed MTRRs. */
458 struct {
459 int index;
460 int seg;
461 };
462
463 /* used for var MTRRs. */
464 struct {
465 struct kvm_mtrr_range *range;
466 /* max address has been covered in var MTRRs. */
467 u64 start_max;
468 };
469 };
470
471 bool fixed;
472 };
473
474 static bool mtrr_lookup_fixed_start(struct mtrr_iter *iter)
475 {
476 int seg, index;
477
478 if (!fixed_mtrr_is_enabled(iter->mtrr_state))
479 return false;
480
481 seg = fixed_mtrr_addr_to_seg(iter->start);
482 if (seg < 0)
483 return false;
484
485 iter->fixed = true;
486 index = fixed_mtrr_addr_seg_to_range_index(iter->start, seg);
487 iter->index = index;
488 iter->seg = seg;
489 return true;
490 }
491
492 static bool match_var_range(struct mtrr_iter *iter,
493 struct kvm_mtrr_range *range)
494 {
495 u64 start, end;
496
497 var_mtrr_range(range, &start, &end);
498 if (!(start >= iter->end || end <= iter->start)) {
499 iter->range = range;
500
501 /*
502 * the function is called when we do kvm_mtrr.head walking.
503 * Range has the minimum base address which interleaves
504 * [looker->start_max, looker->end).
505 */
506 iter->partial_map |= iter->start_max < start;
507
508 /* update the max address has been covered. */
509 iter->start_max = max(iter->start_max, end);
510 return true;
511 }
512
513 return false;
514 }
515
516 static void __mtrr_lookup_var_next(struct mtrr_iter *iter)
517 {
518 struct kvm_mtrr *mtrr_state = iter->mtrr_state;
519
520 list_for_each_entry_continue(iter->range, &mtrr_state->head, node)
521 if (match_var_range(iter, iter->range))
522 return;
523
524 iter->range = NULL;
525 iter->partial_map |= iter->start_max < iter->end;
526 }
527
528 static void mtrr_lookup_var_start(struct mtrr_iter *iter)
529 {
530 struct kvm_mtrr *mtrr_state = iter->mtrr_state;
531
532 iter->fixed = false;
533 iter->start_max = iter->start;
534 iter->range = NULL;
535 iter->range = list_prepare_entry(iter->range, &mtrr_state->head, node);
536
537 __mtrr_lookup_var_next(iter);
538 }
539
540 static void mtrr_lookup_fixed_next(struct mtrr_iter *iter)
541 {
542 /* terminate the lookup. */
543 if (fixed_mtrr_range_end_addr(iter->seg, iter->index) >= iter->end) {
544 iter->fixed = false;
545 iter->range = NULL;
546 return;
547 }
548
549 iter->index++;
550
551 /* have looked up for all fixed MTRRs. */
552 if (iter->index >= ARRAY_SIZE(iter->mtrr_state->fixed_ranges))
553 return mtrr_lookup_var_start(iter);
554
555 /* switch to next segment. */
556 if (iter->index > fixed_mtrr_seg_end_range_index(iter->seg))
557 iter->seg++;
558 }
559
560 static void mtrr_lookup_var_next(struct mtrr_iter *iter)
561 {
562 __mtrr_lookup_var_next(iter);
563 }
564
565 static void mtrr_lookup_start(struct mtrr_iter *iter)
566 {
567 if (!mtrr_is_enabled(iter->mtrr_state)) {
568 iter->mtrr_disabled = true;
569 return;
570 }
571
572 if (!mtrr_lookup_fixed_start(iter))
573 mtrr_lookup_var_start(iter);
574 }
575
576 static void mtrr_lookup_init(struct mtrr_iter *iter,
577 struct kvm_mtrr *mtrr_state, u64 start, u64 end)
578 {
579 iter->mtrr_state = mtrr_state;
580 iter->start = start;
581 iter->end = end;
582 iter->mtrr_disabled = false;
583 iter->partial_map = false;
584 iter->fixed = false;
585 iter->range = NULL;
586
587 mtrr_lookup_start(iter);
588 }
589
590 static bool mtrr_lookup_okay(struct mtrr_iter *iter)
591 {
592 if (iter->fixed) {
593 iter->mem_type = iter->mtrr_state->fixed_ranges[iter->index];
594 return true;
595 }
596
597 if (iter->range) {
598 iter->mem_type = iter->range->base & 0xff;
599 return true;
600 }
601
602 return false;
603 }
604
605 static void mtrr_lookup_next(struct mtrr_iter *iter)
606 {
607 if (iter->fixed)
608 mtrr_lookup_fixed_next(iter);
609 else
610 mtrr_lookup_var_next(iter);
611 }
612
613 #define mtrr_for_each_mem_type(_iter_, _mtrr_, _gpa_start_, _gpa_end_) \
614 for (mtrr_lookup_init(_iter_, _mtrr_, _gpa_start_, _gpa_end_); \
615 mtrr_lookup_okay(_iter_); mtrr_lookup_next(_iter_))
616
617 u8 kvm_mtrr_get_guest_memory_type(struct kvm_vcpu *vcpu, gfn_t gfn)
618 {
619 struct kvm_mtrr *mtrr_state = &vcpu->arch.mtrr_state;
620 struct mtrr_iter iter;
621 u64 start, end;
622 int type = -1;
623 const int wt_wb_mask = (1 << MTRR_TYPE_WRBACK)
624 | (1 << MTRR_TYPE_WRTHROUGH);
625
626 start = gfn_to_gpa(gfn);
627 end = start + PAGE_SIZE;
628
629 mtrr_for_each_mem_type(&iter, mtrr_state, start, end) {
630 int curr_type = iter.mem_type;
631
632 /*
633 * Please refer to Intel SDM Volume 3: 11.11.4.1 MTRR
634 * Precedences.
635 */
636
637 if (type == -1) {
638 type = curr_type;
639 continue;
640 }
641
642 /*
643 * If two or more variable memory ranges match and the
644 * memory types are identical, then that memory type is
645 * used.
646 */
647 if (type == curr_type)
648 continue;
649
650 /*
651 * If two or more variable memory ranges match and one of
652 * the memory types is UC, the UC memory type used.
653 */
654 if (curr_type == MTRR_TYPE_UNCACHABLE)
655 return MTRR_TYPE_UNCACHABLE;
656
657 /*
658 * If two or more variable memory ranges match and the
659 * memory types are WT and WB, the WT memory type is used.
660 */
661 if (((1 << type) & wt_wb_mask) &&
662 ((1 << curr_type) & wt_wb_mask)) {
663 type = MTRR_TYPE_WRTHROUGH;
664 continue;
665 }
666
667 /*
668 * For overlaps not defined by the above rules, processor
669 * behavior is undefined.
670 */
671
672 /* We use WB for this undefined behavior. :( */
673 return MTRR_TYPE_WRBACK;
674 }
675
676 if (iter.mtrr_disabled)
677 return mtrr_disabled_type(vcpu);
678
679 /* not contained in any MTRRs. */
680 if (type == -1)
681 return mtrr_default_type(mtrr_state);
682
683 /*
684 * We just check one page, partially covered by MTRRs is
685 * impossible.
686 */
687 WARN_ON(iter.partial_map);
688
689 return type;
690 }
691 EXPORT_SYMBOL_GPL(kvm_mtrr_get_guest_memory_type);
692
693 bool kvm_mtrr_check_gfn_range_consistency(struct kvm_vcpu *vcpu, gfn_t gfn,
694 int page_num)
695 {
696 struct kvm_mtrr *mtrr_state = &vcpu->arch.mtrr_state;
697 struct mtrr_iter iter;
698 u64 start, end;
699 int type = -1;
700
701 start = gfn_to_gpa(gfn);
702 end = gfn_to_gpa(gfn + page_num);
703 mtrr_for_each_mem_type(&iter, mtrr_state, start, end) {
704 if (type == -1) {
705 type = iter.mem_type;
706 continue;
707 }
708
709 if (type != iter.mem_type)
710 return false;
711 }
712
713 if (iter.mtrr_disabled)
714 return true;
715
716 if (!iter.partial_map)
717 return true;
718
719 if (type == -1)
720 return true;
721
722 return type == mtrr_default_type(mtrr_state);
723 }