projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
KVM: MMU: Add some mmu statistics
[deliverable/linux.git] / drivers / kvm / mmu.c
1 /*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * MMU support
8 *
9 * Copyright (C) 2006 Qumranet, Inc.
10 *
11 * Authors:
12 * Yaniv Kamay <yaniv@qumranet.com>
13 * Avi Kivity <avi@qumranet.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
20 #include "vmx.h"
21 #include "kvm.h"
22 #include "x86.h"
23
24 #include <linux/types.h>
25 #include <linux/string.h>
26 #include <linux/mm.h>
27 #include <linux/highmem.h>
28 #include <linux/module.h>
29
30 #include <asm/page.h>
31 #include <asm/cmpxchg.h>
32
33 #undef MMU_DEBUG
34
35 #undef AUDIT
36
37 #ifdef AUDIT
38 static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg);
39 #else
40 static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg) {}
41 #endif
42
43 #ifdef MMU_DEBUG
44
45 #define pgprintk(x...) do { if (dbg) printk(x); } while (0)
46 #define rmap_printk(x...) do { if (dbg) printk(x); } while (0)
47
48 #else
49
50 #define pgprintk(x...) do { } while (0)
51 #define rmap_printk(x...) do { } while (0)
52
53 #endif
54
55 #if defined(MMU_DEBUG) || defined(AUDIT)
56 static int dbg = 1;
57 #endif
58
59 #ifndef MMU_DEBUG
60 #define ASSERT(x) do { } while (0)
61 #else
62 #define ASSERT(x) \
63 if (!(x)) { \
64 printk(KERN_WARNING "assertion failed %s:%d: %s\n", \
65 __FILE__, __LINE__, #x); \
66 }
67 #endif
68
69 #define PT64_PT_BITS 9
70 #define PT64_ENT_PER_PAGE (1 << PT64_PT_BITS)
71 #define PT32_PT_BITS 10
72 #define PT32_ENT_PER_PAGE (1 << PT32_PT_BITS)
73
74 #define PT_WRITABLE_SHIFT 1
75
76 #define PT_PRESENT_MASK (1ULL << 0)
77 #define PT_WRITABLE_MASK (1ULL << PT_WRITABLE_SHIFT)
78 #define PT_USER_MASK (1ULL << 2)
79 #define PT_PWT_MASK (1ULL << 3)
80 #define PT_PCD_MASK (1ULL << 4)
81 #define PT_ACCESSED_MASK (1ULL << 5)
82 #define PT_DIRTY_MASK (1ULL << 6)
83 #define PT_PAGE_SIZE_MASK (1ULL << 7)
84 #define PT_PAT_MASK (1ULL << 7)
85 #define PT_GLOBAL_MASK (1ULL << 8)
86 #define PT64_NX_MASK (1ULL << 63)
87
88 #define PT_PAT_SHIFT 7
89 #define PT_DIR_PAT_SHIFT 12
90 #define PT_DIR_PAT_MASK (1ULL << PT_DIR_PAT_SHIFT)
91
92 #define PT32_DIR_PSE36_SIZE 4
93 #define PT32_DIR_PSE36_SHIFT 13
94 #define PT32_DIR_PSE36_MASK \
95 (((1ULL << PT32_DIR_PSE36_SIZE) - 1) << PT32_DIR_PSE36_SHIFT)
96
97
98 #define PT_FIRST_AVAIL_BITS_SHIFT 9
99 #define PT64_SECOND_AVAIL_BITS_SHIFT 52
100
101 #define PT_SHADOW_IO_MARK (1ULL << PT_FIRST_AVAIL_BITS_SHIFT)
102
103 #define VALID_PAGE(x) ((x) != INVALID_PAGE)
104
105 #define PT64_LEVEL_BITS 9
106
107 #define PT64_LEVEL_SHIFT(level) \
108 (PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS)
109
110 #define PT64_LEVEL_MASK(level) \
111 (((1ULL << PT64_LEVEL_BITS) - 1) << PT64_LEVEL_SHIFT(level))
112
113 #define PT64_INDEX(address, level)\
114 (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1))
115
116
117 #define PT32_LEVEL_BITS 10
118
119 #define PT32_LEVEL_SHIFT(level) \
120 (PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS)
121
122 #define PT32_LEVEL_MASK(level) \
123 (((1ULL << PT32_LEVEL_BITS) - 1) << PT32_LEVEL_SHIFT(level))
124
125 #define PT32_INDEX(address, level)\
126 (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1))
127
128
129 #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1))
130 #define PT64_DIR_BASE_ADDR_MASK \
131 (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1))
132
133 #define PT32_BASE_ADDR_MASK PAGE_MASK
134 #define PT32_DIR_BASE_ADDR_MASK \
135 (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1))
136
137
138 #define PFERR_PRESENT_MASK (1U << 0)
139 #define PFERR_WRITE_MASK (1U << 1)
140 #define PFERR_USER_MASK (1U << 2)
141 #define PFERR_FETCH_MASK (1U << 4)
142
143 #define PT64_ROOT_LEVEL 4
144 #define PT32_ROOT_LEVEL 2
145 #define PT32E_ROOT_LEVEL 3
146
147 #define PT_DIRECTORY_LEVEL 2
148 #define PT_PAGE_TABLE_LEVEL 1
149
150 #define RMAP_EXT 4
151
152 struct kvm_rmap_desc {
153 u64 *shadow_ptes[RMAP_EXT];
154 struct kvm_rmap_desc *more;
155 };
156
157 static struct kmem_cache *pte_chain_cache;
158 static struct kmem_cache *rmap_desc_cache;
159 static struct kmem_cache *mmu_page_header_cache;
160
161 static u64 __read_mostly shadow_trap_nonpresent_pte;
162 static u64 __read_mostly shadow_notrap_nonpresent_pte;
163
164 void kvm_mmu_set_nonpresent_ptes(u64 trap_pte, u64 notrap_pte)
165 {
166 shadow_trap_nonpresent_pte = trap_pte;
167 shadow_notrap_nonpresent_pte = notrap_pte;
168 }
169 EXPORT_SYMBOL_GPL(kvm_mmu_set_nonpresent_ptes);
170
171 static int is_write_protection(struct kvm_vcpu *vcpu)
172 {
173 return vcpu->cr0 & X86_CR0_WP;
174 }
175
176 static int is_cpuid_PSE36(void)
177 {
178 return 1;
179 }
180
181 static int is_nx(struct kvm_vcpu *vcpu)
182 {
183 return vcpu->shadow_efer & EFER_NX;
184 }
185
186 static int is_present_pte(unsigned long pte)
187 {
188 return pte & PT_PRESENT_MASK;
189 }
190
191 static int is_shadow_present_pte(u64 pte)
192 {
193 pte &= ~PT_SHADOW_IO_MARK;
194 return pte != shadow_trap_nonpresent_pte
195 && pte != shadow_notrap_nonpresent_pte;
196 }
197
198 static int is_writeble_pte(unsigned long pte)
199 {
200 return pte & PT_WRITABLE_MASK;
201 }
202
203 static int is_dirty_pte(unsigned long pte)
204 {
205 return pte & PT_DIRTY_MASK;
206 }
207
208 static int is_io_pte(unsigned long pte)
209 {
210 return pte & PT_SHADOW_IO_MARK;
211 }
212
213 static int is_rmap_pte(u64 pte)
214 {
215 return pte != shadow_trap_nonpresent_pte
216 && pte != shadow_notrap_nonpresent_pte;
217 }
218
219 static void set_shadow_pte(u64 *sptep, u64 spte)
220 {
221 #ifdef CONFIG_X86_64
222 set_64bit((unsigned long *)sptep, spte);
223 #else
224 set_64bit((unsigned long long *)sptep, spte);
225 #endif
226 }
227
228 static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache,
229 struct kmem_cache *base_cache, int min)
230 {
231 void *obj;
232
233 if (cache->nobjs >= min)
234 return 0;
235 while (cache->nobjs < ARRAY_SIZE(cache->objects)) {
236 obj = kmem_cache_zalloc(base_cache, GFP_KERNEL);
237 if (!obj)
238 return -ENOMEM;
239 cache->objects[cache->nobjs++] = obj;
240 }
241 return 0;
242 }
243
244 static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
245 {
246 while (mc->nobjs)
247 kfree(mc->objects[--mc->nobjs]);
248 }
249
250 static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache *cache,
251 int min)
252 {
253 struct page *page;
254
255 if (cache->nobjs >= min)
256 return 0;
257 while (cache->nobjs < ARRAY_SIZE(cache->objects)) {
258 page = alloc_page(GFP_KERNEL);
259 if (!page)
260 return -ENOMEM;
261 set_page_private(page, 0);
262 cache->objects[cache->nobjs++] = page_address(page);
263 }
264 return 0;
265 }
266
267 static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache *mc)
268 {
269 while (mc->nobjs)
270 free_page((unsigned long)mc->objects[--mc->nobjs]);
271 }
272
273 static int mmu_topup_memory_caches(struct kvm_vcpu *vcpu)
274 {
275 int r;
276
277 kvm_mmu_free_some_pages(vcpu);
278 r = mmu_topup_memory_cache(&vcpu->mmu_pte_chain_cache,
279 pte_chain_cache, 4);
280 if (r)
281 goto out;
282 r = mmu_topup_memory_cache(&vcpu->mmu_rmap_desc_cache,
283 rmap_desc_cache, 1);
284 if (r)
285 goto out;
286 r = mmu_topup_memory_cache_page(&vcpu->mmu_page_cache, 8);
287 if (r)
288 goto out;
289 r = mmu_topup_memory_cache(&vcpu->mmu_page_header_cache,
290 mmu_page_header_cache, 4);
291 out:
292 return r;
293 }
294
295 static void mmu_free_memory_caches(struct kvm_vcpu *vcpu)
296 {
297 mmu_free_memory_cache(&vcpu->mmu_pte_chain_cache);
298 mmu_free_memory_cache(&vcpu->mmu_rmap_desc_cache);
299 mmu_free_memory_cache_page(&vcpu->mmu_page_cache);
300 mmu_free_memory_cache(&vcpu->mmu_page_header_cache);
301 }
302
303 static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc,
304 size_t size)
305 {
306 void *p;
307
308 BUG_ON(!mc->nobjs);
309 p = mc->objects[--mc->nobjs];
310 memset(p, 0, size);
311 return p;
312 }
313
314 static struct kvm_pte_chain *mmu_alloc_pte_chain(struct kvm_vcpu *vcpu)
315 {
316 return mmu_memory_cache_alloc(&vcpu->mmu_pte_chain_cache,
317 sizeof(struct kvm_pte_chain));
318 }
319
320 static void mmu_free_pte_chain(struct kvm_pte_chain *pc)
321 {
322 kfree(pc);
323 }
324
325 static struct kvm_rmap_desc *mmu_alloc_rmap_desc(struct kvm_vcpu *vcpu)
326 {
327 return mmu_memory_cache_alloc(&vcpu->mmu_rmap_desc_cache,
328 sizeof(struct kvm_rmap_desc));
329 }
330
331 static void mmu_free_rmap_desc(struct kvm_rmap_desc *rd)
332 {
333 kfree(rd);
334 }
335
336 /*
337 * Take gfn and return the reverse mapping to it.
338 * Note: gfn must be unaliased before this function get called
339 */
340
341 static unsigned long *gfn_to_rmap(struct kvm *kvm, gfn_t gfn)
342 {
343 struct kvm_memory_slot *slot;
344
345 slot = gfn_to_memslot(kvm, gfn);
346 return &slot->rmap[gfn - slot->base_gfn];
347 }
348
349 /*
350 * Reverse mapping data structures:
351 *
352 * If rmapp bit zero is zero, then rmapp point to the shadw page table entry
353 * that points to page_address(page).
354 *
355 * If rmapp bit zero is one, (then rmap & ~1) points to a struct kvm_rmap_desc
356 * containing more mappings.
357 */
358 static void rmap_add(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn)
359 {
360 struct kvm_mmu_page *page;
361 struct kvm_rmap_desc *desc;
362 unsigned long *rmapp;
363 int i;
364
365 if (!is_rmap_pte(*spte))
366 return;
367 gfn = unalias_gfn(vcpu->kvm, gfn);
368 page = page_header(__pa(spte));
369 page->gfns[spte - page->spt] = gfn;
370 rmapp = gfn_to_rmap(vcpu->kvm, gfn);
371 if (!*rmapp) {
372 rmap_printk("rmap_add: %p %llx 0->1\n", spte, *spte);
373 *rmapp = (unsigned long)spte;
374 } else if (!(*rmapp & 1)) {
375 rmap_printk("rmap_add: %p %llx 1->many\n", spte, *spte);
376 desc = mmu_alloc_rmap_desc(vcpu);
377 desc->shadow_ptes[0] = (u64 *)*rmapp;
378 desc->shadow_ptes[1] = spte;
379 *rmapp = (unsigned long)desc | 1;
380 } else {
381 rmap_printk("rmap_add: %p %llx many->many\n", spte, *spte);
382 desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
383 while (desc->shadow_ptes[RMAP_EXT-1] && desc->more)
384 desc = desc->more;
385 if (desc->shadow_ptes[RMAP_EXT-1]) {
386 desc->more = mmu_alloc_rmap_desc(vcpu);
387 desc = desc->more;
388 }
389 for (i = 0; desc->shadow_ptes[i]; ++i)
390 ;
391 desc->shadow_ptes[i] = spte;
392 }
393 }
394
395 static void rmap_desc_remove_entry(unsigned long *rmapp,
396 struct kvm_rmap_desc *desc,
397 int i,
398 struct kvm_rmap_desc *prev_desc)
399 {
400 int j;
401
402 for (j = RMAP_EXT - 1; !desc->shadow_ptes[j] && j > i; --j)
403 ;
404 desc->shadow_ptes[i] = desc->shadow_ptes[j];
405 desc->shadow_ptes[j] = NULL;
406 if (j != 0)
407 return;
408 if (!prev_desc && !desc->more)
409 *rmapp = (unsigned long)desc->shadow_ptes[0];
410 else
411 if (prev_desc)
412 prev_desc->more = desc->more;
413 else
414 *rmapp = (unsigned long)desc->more | 1;
415 mmu_free_rmap_desc(desc);
416 }
417
418 static void rmap_remove(struct kvm *kvm, u64 *spte)
419 {
420 struct kvm_rmap_desc *desc;
421 struct kvm_rmap_desc *prev_desc;
422 struct kvm_mmu_page *page;
423 unsigned long *rmapp;
424 int i;
425
426 if (!is_rmap_pte(*spte))
427 return;
428 page = page_header(__pa(spte));
429 kvm_release_page(pfn_to_page((*spte & PT64_BASE_ADDR_MASK) >>
430 PAGE_SHIFT));
431 rmapp = gfn_to_rmap(kvm, page->gfns[spte - page->spt]);
432 if (!*rmapp) {
433 printk(KERN_ERR "rmap_remove: %p %llx 0->BUG\n", spte, *spte);
434 BUG();
435 } else if (!(*rmapp & 1)) {
436 rmap_printk("rmap_remove: %p %llx 1->0\n", spte, *spte);
437 if ((u64 *)*rmapp != spte) {
438 printk(KERN_ERR "rmap_remove: %p %llx 1->BUG\n",
439 spte, *spte);
440 BUG();
441 }
442 *rmapp = 0;
443 } else {
444 rmap_printk("rmap_remove: %p %llx many->many\n", spte, *spte);
445 desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
446 prev_desc = NULL;
447 while (desc) {
448 for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i)
449 if (desc->shadow_ptes[i] == spte) {
450 rmap_desc_remove_entry(rmapp,
451 desc, i,
452 prev_desc);
453 return;
454 }
455 prev_desc = desc;
456 desc = desc->more;
457 }
458 BUG();
459 }
460 }
461
462 static u64 *rmap_next(struct kvm *kvm, unsigned long *rmapp, u64 *spte)
463 {
464 struct kvm_rmap_desc *desc;
465 struct kvm_rmap_desc *prev_desc;
466 u64 *prev_spte;
467 int i;
468
469 if (!*rmapp)
470 return NULL;
471 else if (!(*rmapp & 1)) {
472 if (!spte)
473 return (u64 *)*rmapp;
474 return NULL;
475 }
476 desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
477 prev_desc = NULL;
478 prev_spte = NULL;
479 while (desc) {
480 for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i) {
481 if (prev_spte == spte)
482 return desc->shadow_ptes[i];
483 prev_spte = desc->shadow_ptes[i];
484 }
485 desc = desc->more;
486 }
487 return NULL;
488 }
489
490 static void rmap_write_protect(struct kvm *kvm, u64 gfn)
491 {
492 unsigned long *rmapp;
493 u64 *spte;
494
495 gfn = unalias_gfn(kvm, gfn);
496 rmapp = gfn_to_rmap(kvm, gfn);
497
498 spte = rmap_next(kvm, rmapp, NULL);
499 while (spte) {
500 BUG_ON(!spte);
501 BUG_ON(!(*spte & PT_PRESENT_MASK));
502 rmap_printk("rmap_write_protect: spte %p %llx\n", spte, *spte);
503 if (is_writeble_pte(*spte))
504 set_shadow_pte(spte, *spte & ~PT_WRITABLE_MASK);
505 kvm_flush_remote_tlbs(kvm);
506 spte = rmap_next(kvm, rmapp, spte);
507 }
508 }
509
510 #ifdef MMU_DEBUG
511 static int is_empty_shadow_page(u64 *spt)
512 {
513 u64 *pos;
514 u64 *end;
515
516 for (pos = spt, end = pos + PAGE_SIZE / sizeof(u64); pos != end; pos++)
517 if ((*pos & ~PT_SHADOW_IO_MARK) != shadow_trap_nonpresent_pte) {
518 printk(KERN_ERR "%s: %p %llx\n", __FUNCTION__,
519 pos, *pos);
520 return 0;
521 }
522 return 1;
523 }
524 #endif
525
526 static void kvm_mmu_free_page(struct kvm *kvm,
527 struct kvm_mmu_page *page_head)
528 {
529 ASSERT(is_empty_shadow_page(page_head->spt));
530 list_del(&page_head->link);
531 __free_page(virt_to_page(page_head->spt));
532 __free_page(virt_to_page(page_head->gfns));
533 kfree(page_head);
534 ++kvm->n_free_mmu_pages;
535 }
536
537 static unsigned kvm_page_table_hashfn(gfn_t gfn)
538 {
539 return gfn;
540 }
541
542 static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu,
543 u64 *parent_pte)
544 {
545 struct kvm_mmu_page *page;
546
547 if (!vcpu->kvm->n_free_mmu_pages)
548 return NULL;
549
550 page = mmu_memory_cache_alloc(&vcpu->mmu_page_header_cache,
551 sizeof *page);
552 page->spt = mmu_memory_cache_alloc(&vcpu->mmu_page_cache, PAGE_SIZE);
553 page->gfns = mmu_memory_cache_alloc(&vcpu->mmu_page_cache, PAGE_SIZE);
554 set_page_private(virt_to_page(page->spt), (unsigned long)page);
555 list_add(&page->link, &vcpu->kvm->active_mmu_pages);
556 ASSERT(is_empty_shadow_page(page->spt));
557 page->slot_bitmap = 0;
558 page->multimapped = 0;
559 page->parent_pte = parent_pte;
560 --vcpu->kvm->n_free_mmu_pages;
561 return page;
562 }
563
564 static void mmu_page_add_parent_pte(struct kvm_vcpu *vcpu,
565 struct kvm_mmu_page *page, u64 *parent_pte)
566 {
567 struct kvm_pte_chain *pte_chain;
568 struct hlist_node *node;
569 int i;
570
571 if (!parent_pte)
572 return;
573 if (!page->multimapped) {
574 u64 *old = page->parent_pte;
575
576 if (!old) {
577 page->parent_pte = parent_pte;
578 return;
579 }
580 page->multimapped = 1;
581 pte_chain = mmu_alloc_pte_chain(vcpu);
582 INIT_HLIST_HEAD(&page->parent_ptes);
583 hlist_add_head(&pte_chain->link, &page->parent_ptes);
584 pte_chain->parent_ptes[0] = old;
585 }
586 hlist_for_each_entry(pte_chain, node, &page->parent_ptes, link) {
587 if (pte_chain->parent_ptes[NR_PTE_CHAIN_ENTRIES-1])
588 continue;
589 for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i)
590 if (!pte_chain->parent_ptes[i]) {
591 pte_chain->parent_ptes[i] = parent_pte;
592 return;
593 }
594 }
595 pte_chain = mmu_alloc_pte_chain(vcpu);
596 BUG_ON(!pte_chain);
597 hlist_add_head(&pte_chain->link, &page->parent_ptes);
598 pte_chain->parent_ptes[0] = parent_pte;
599 }
600
601 static void mmu_page_remove_parent_pte(struct kvm_mmu_page *page,
602 u64 *parent_pte)
603 {
604 struct kvm_pte_chain *pte_chain;
605 struct hlist_node *node;
606 int i;
607
608 if (!page->multimapped) {
609 BUG_ON(page->parent_pte != parent_pte);
610 page->parent_pte = NULL;
611 return;
612 }
613 hlist_for_each_entry(pte_chain, node, &page->parent_ptes, link)
614 for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) {
615 if (!pte_chain->parent_ptes[i])
616 break;
617 if (pte_chain->parent_ptes[i] != parent_pte)
618 continue;
619 while (i + 1 < NR_PTE_CHAIN_ENTRIES
620 && pte_chain->parent_ptes[i + 1]) {
621 pte_chain->parent_ptes[i]
622 = pte_chain->parent_ptes[i + 1];
623 ++i;
624 }
625 pte_chain->parent_ptes[i] = NULL;
626 if (i == 0) {
627 hlist_del(&pte_chain->link);
628 mmu_free_pte_chain(pte_chain);
629 if (hlist_empty(&page->parent_ptes)) {
630 page->multimapped = 0;
631 page->parent_pte = NULL;
632 }
633 }
634 return;
635 }
636 BUG();
637 }
638
639 static struct kvm_mmu_page *kvm_mmu_lookup_page(struct kvm *kvm,
640 gfn_t gfn)
641 {
642 unsigned index;
643 struct hlist_head *bucket;
644 struct kvm_mmu_page *page;
645 struct hlist_node *node;
646
647 pgprintk("%s: looking for gfn %lx\n", __FUNCTION__, gfn);
648 index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES;
649 bucket = &kvm->mmu_page_hash[index];
650 hlist_for_each_entry(page, node, bucket, hash_link)
651 if (page->gfn == gfn && !page->role.metaphysical) {
652 pgprintk("%s: found role %x\n",
653 __FUNCTION__, page->role.word);
654 return page;
655 }
656 return NULL;
657 }
658
659 static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
660 gfn_t gfn,
661 gva_t gaddr,
662 unsigned level,
663 int metaphysical,
664 unsigned hugepage_access,
665 u64 *parent_pte)
666 {
667 union kvm_mmu_page_role role;
668 unsigned index;
669 unsigned quadrant;
670 struct hlist_head *bucket;
671 struct kvm_mmu_page *page;
672 struct hlist_node *node;
673
674 role.word = 0;
675 role.glevels = vcpu->mmu.root_level;
676 role.level = level;
677 role.metaphysical = metaphysical;
678 role.hugepage_access = hugepage_access;
679 if (vcpu->mmu.root_level <= PT32_ROOT_LEVEL) {
680 quadrant = gaddr >> (PAGE_SHIFT + (PT64_PT_BITS * level));
681 quadrant &= (1 << ((PT32_PT_BITS - PT64_PT_BITS) * level)) - 1;
682 role.quadrant = quadrant;
683 }
684 pgprintk("%s: looking gfn %lx role %x\n", __FUNCTION__,
685 gfn, role.word);
686 index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES;
687 bucket = &vcpu->kvm->mmu_page_hash[index];
688 hlist_for_each_entry(page, node, bucket, hash_link)
689 if (page->gfn == gfn && page->role.word == role.word) {
690 mmu_page_add_parent_pte(vcpu, page, parent_pte);
691 pgprintk("%s: found\n", __FUNCTION__);
692 return page;
693 }
694 page = kvm_mmu_alloc_page(vcpu, parent_pte);
695 if (!page)
696 return page;
697 pgprintk("%s: adding gfn %lx role %x\n", __FUNCTION__, gfn, role.word);
698 page->gfn = gfn;
699 page->role = role;
700 hlist_add_head(&page->hash_link, bucket);
701 vcpu->mmu.prefetch_page(vcpu, page);
702 if (!metaphysical)
703 rmap_write_protect(vcpu->kvm, gfn);
704 return page;
705 }
706
707 static void kvm_mmu_page_unlink_children(struct kvm *kvm,
708 struct kvm_mmu_page *page)
709 {
710 unsigned i;
711 u64 *pt;
712 u64 ent;
713
714 pt = page->spt;
715
716 if (page->role.level == PT_PAGE_TABLE_LEVEL) {
717 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
718 if (is_shadow_present_pte(pt[i]))
719 rmap_remove(kvm, &pt[i]);
720 pt[i] = shadow_trap_nonpresent_pte;
721 }
722 kvm_flush_remote_tlbs(kvm);
723 return;
724 }
725
726 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
727 ent = pt[i];
728
729 pt[i] = shadow_trap_nonpresent_pte;
730 if (!is_shadow_present_pte(ent))
731 continue;
732 ent &= PT64_BASE_ADDR_MASK;
733 mmu_page_remove_parent_pte(page_header(ent), &pt[i]);
734 }
735 kvm_flush_remote_tlbs(kvm);
736 }
737
738 static void kvm_mmu_put_page(struct kvm_mmu_page *page,
739 u64 *parent_pte)
740 {
741 mmu_page_remove_parent_pte(page, parent_pte);
742 }
743
744 static void kvm_mmu_reset_last_pte_updated(struct kvm *kvm)
745 {
746 int i;
747
748 for (i = 0; i < KVM_MAX_VCPUS; ++i)
749 if (kvm->vcpus[i])
750 kvm->vcpus[i]->last_pte_updated = NULL;
751 }
752
753 static void kvm_mmu_zap_page(struct kvm *kvm,
754 struct kvm_mmu_page *page)
755 {
756 u64 *parent_pte;
757
758 ++kvm->stat.mmu_shadow_zapped;
759 while (page->multimapped || page->parent_pte) {
760 if (!page->multimapped)
761 parent_pte = page->parent_pte;
762 else {
763 struct kvm_pte_chain *chain;
764
765 chain = container_of(page->parent_ptes.first,
766 struct kvm_pte_chain, link);
767 parent_pte = chain->parent_ptes[0];
768 }
769 BUG_ON(!parent_pte);
770 kvm_mmu_put_page(page, parent_pte);
771 set_shadow_pte(parent_pte, shadow_trap_nonpresent_pte);
772 }
773 kvm_mmu_page_unlink_children(kvm, page);
774 if (!page->root_count) {
775 hlist_del(&page->hash_link);
776 kvm_mmu_free_page(kvm, page);
777 } else
778 list_move(&page->link, &kvm->active_mmu_pages);
779 kvm_mmu_reset_last_pte_updated(kvm);
780 }
781
782 /*
783 * Changing the number of mmu pages allocated to the vm
784 * Note: if kvm_nr_mmu_pages is too small, you will get dead lock
785 */
786 void kvm_mmu_change_mmu_pages(struct kvm *kvm, unsigned int kvm_nr_mmu_pages)
787 {
788 /*
789 * If we set the number of mmu pages to be smaller be than the
790 * number of actived pages , we must to free some mmu pages before we
791 * change the value
792 */
793
794 if ((kvm->n_alloc_mmu_pages - kvm->n_free_mmu_pages) >
795 kvm_nr_mmu_pages) {
796 int n_used_mmu_pages = kvm->n_alloc_mmu_pages
797 - kvm->n_free_mmu_pages;
798
799 while (n_used_mmu_pages > kvm_nr_mmu_pages) {
800 struct kvm_mmu_page *page;
801
802 page = container_of(kvm->active_mmu_pages.prev,
803 struct kvm_mmu_page, link);
804 kvm_mmu_zap_page(kvm, page);
805 n_used_mmu_pages--;
806 }
807 kvm->n_free_mmu_pages = 0;
808 }
809 else
810 kvm->n_free_mmu_pages += kvm_nr_mmu_pages
811 - kvm->n_alloc_mmu_pages;
812
813 kvm->n_alloc_mmu_pages = kvm_nr_mmu_pages;
814 }
815
816 static int kvm_mmu_unprotect_page(struct kvm *kvm, gfn_t gfn)
817 {
818 unsigned index;
819 struct hlist_head *bucket;
820 struct kvm_mmu_page *page;
821 struct hlist_node *node, *n;
822 int r;
823
824 pgprintk("%s: looking for gfn %lx\n", __FUNCTION__, gfn);
825 r = 0;
826 index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES;
827 bucket = &kvm->mmu_page_hash[index];
828 hlist_for_each_entry_safe(page, node, n, bucket, hash_link)
829 if (page->gfn == gfn && !page->role.metaphysical) {
830 pgprintk("%s: gfn %lx role %x\n", __FUNCTION__, gfn,
831 page->role.word);
832 kvm_mmu_zap_page(kvm, page);
833 r = 1;
834 }
835 return r;
836 }
837
838 static void mmu_unshadow(struct kvm *kvm, gfn_t gfn)
839 {
840 struct kvm_mmu_page *page;
841
842 while ((page = kvm_mmu_lookup_page(kvm, gfn)) != NULL) {
843 pgprintk("%s: zap %lx %x\n",
844 __FUNCTION__, gfn, page->role.word);
845 kvm_mmu_zap_page(kvm, page);
846 }
847 }
848
849 static void page_header_update_slot(struct kvm *kvm, void *pte, gpa_t gpa)
850 {
851 int slot = memslot_id(kvm, gfn_to_memslot(kvm, gpa >> PAGE_SHIFT));
852 struct kvm_mmu_page *page_head = page_header(__pa(pte));
853
854 __set_bit(slot, &page_head->slot_bitmap);
855 }
856
857 hpa_t gpa_to_hpa(struct kvm *kvm, gpa_t gpa)
858 {
859 struct page *page;
860 hpa_t hpa;
861
862 ASSERT((gpa & HPA_ERR_MASK) == 0);
863 page = gfn_to_page(kvm, gpa >> PAGE_SHIFT);
864 hpa = ((hpa_t)page_to_pfn(page) << PAGE_SHIFT) | (gpa & (PAGE_SIZE-1));
865 if (is_error_page(page))
866 return hpa | HPA_ERR_MASK;
867 return hpa;
868 }
869
870 hpa_t gva_to_hpa(struct kvm_vcpu *vcpu, gva_t gva)
871 {
872 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, gva);
873
874 if (gpa == UNMAPPED_GVA)
875 return UNMAPPED_GVA;
876 return gpa_to_hpa(vcpu->kvm, gpa);
877 }
878
879 struct page *gva_to_page(struct kvm_vcpu *vcpu, gva_t gva)
880 {
881 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, gva);
882
883 if (gpa == UNMAPPED_GVA)
884 return NULL;
885 return pfn_to_page(gpa_to_hpa(vcpu->kvm, gpa) >> PAGE_SHIFT);
886 }
887
888 static void nonpaging_new_cr3(struct kvm_vcpu *vcpu)
889 {
890 }
891
892 static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, hpa_t p)
893 {
894 int level = PT32E_ROOT_LEVEL;
895 hpa_t table_addr = vcpu->mmu.root_hpa;
896
897 for (; ; level--) {
898 u32 index = PT64_INDEX(v, level);
899 u64 *table;
900 u64 pte;
901
902 ASSERT(VALID_PAGE(table_addr));
903 table = __va(table_addr);
904
905 if (level == 1) {
906 int was_rmapped;
907
908 pte = table[index];
909 was_rmapped = is_rmap_pte(pte);
910 if (is_shadow_present_pte(pte) && is_writeble_pte(pte))
911 return 0;
912 mark_page_dirty(vcpu->kvm, v >> PAGE_SHIFT);
913 page_header_update_slot(vcpu->kvm, table, v);
914 table[index] = p | PT_PRESENT_MASK | PT_WRITABLE_MASK |
915 PT_USER_MASK;
916 if (!was_rmapped)
917 rmap_add(vcpu, &table[index], v >> PAGE_SHIFT);
918 else
919 kvm_release_page(pfn_to_page(p >> PAGE_SHIFT));
920 return 0;
921 }
922
923 if (table[index] == shadow_trap_nonpresent_pte) {
924 struct kvm_mmu_page *new_table;
925 gfn_t pseudo_gfn;
926
927 pseudo_gfn = (v & PT64_DIR_BASE_ADDR_MASK)
928 >> PAGE_SHIFT;
929 new_table = kvm_mmu_get_page(vcpu, pseudo_gfn,
930 v, level - 1,
931 1, 3, &table[index]);
932 if (!new_table) {
933 pgprintk("nonpaging_map: ENOMEM\n");
934 kvm_release_page(pfn_to_page(p >> PAGE_SHIFT));
935 return -ENOMEM;
936 }
937
938 table[index] = __pa(new_table->spt) | PT_PRESENT_MASK
939 | PT_WRITABLE_MASK | PT_USER_MASK;
940 }
941 table_addr = table[index] & PT64_BASE_ADDR_MASK;
942 }
943 }
944
945 static void nonpaging_prefetch_page(struct kvm_vcpu *vcpu,
946 struct kvm_mmu_page *sp)
947 {
948 int i;
949
950 for (i = 0; i < PT64_ENT_PER_PAGE; ++i)
951 sp->spt[i] = shadow_trap_nonpresent_pte;
952 }
953
954 static void mmu_free_roots(struct kvm_vcpu *vcpu)
955 {
956 int i;
957 struct kvm_mmu_page *page;
958
959 if (!VALID_PAGE(vcpu->mmu.root_hpa))
960 return;
961 #ifdef CONFIG_X86_64
962 if (vcpu->mmu.shadow_root_level == PT64_ROOT_LEVEL) {
963 hpa_t root = vcpu->mmu.root_hpa;
964
965 page = page_header(root);
966 --page->root_count;
967 vcpu->mmu.root_hpa = INVALID_PAGE;
968 return;
969 }
970 #endif
971 for (i = 0; i < 4; ++i) {
972 hpa_t root = vcpu->mmu.pae_root[i];
973
974 if (root) {
975 root &= PT64_BASE_ADDR_MASK;
976 page = page_header(root);
977 --page->root_count;
978 }
979 vcpu->mmu.pae_root[i] = INVALID_PAGE;
980 }
981 vcpu->mmu.root_hpa = INVALID_PAGE;
982 }
983
984 static void mmu_alloc_roots(struct kvm_vcpu *vcpu)
985 {
986 int i;
987 gfn_t root_gfn;
988 struct kvm_mmu_page *page;
989
990 root_gfn = vcpu->cr3 >> PAGE_SHIFT;
991
992 #ifdef CONFIG_X86_64
993 if (vcpu->mmu.shadow_root_level == PT64_ROOT_LEVEL) {
994 hpa_t root = vcpu->mmu.root_hpa;
995
996 ASSERT(!VALID_PAGE(root));
997 page = kvm_mmu_get_page(vcpu, root_gfn, 0,
998 PT64_ROOT_LEVEL, 0, 0, NULL);
999 root = __pa(page->spt);
1000 ++page->root_count;
1001 vcpu->mmu.root_hpa = root;
1002 return;
1003 }
1004 #endif
1005 for (i = 0; i < 4; ++i) {
1006 hpa_t root = vcpu->mmu.pae_root[i];
1007
1008 ASSERT(!VALID_PAGE(root));
1009 if (vcpu->mmu.root_level == PT32E_ROOT_LEVEL) {
1010 if (!is_present_pte(vcpu->pdptrs[i])) {
1011 vcpu->mmu.pae_root[i] = 0;
1012 continue;
1013 }
1014 root_gfn = vcpu->pdptrs[i] >> PAGE_SHIFT;
1015 } else if (vcpu->mmu.root_level == 0)
1016 root_gfn = 0;
1017 page = kvm_mmu_get_page(vcpu, root_gfn, i << 30,
1018 PT32_ROOT_LEVEL, !is_paging(vcpu),
1019 0, NULL);
1020 root = __pa(page->spt);
1021 ++page->root_count;
1022 vcpu->mmu.pae_root[i] = root | PT_PRESENT_MASK;
1023 }
1024 vcpu->mmu.root_hpa = __pa(vcpu->mmu.pae_root);
1025 }
1026
1027 static gpa_t nonpaging_gva_to_gpa(struct kvm_vcpu *vcpu, gva_t vaddr)
1028 {
1029 return vaddr;
1030 }
1031
1032 static int nonpaging_page_fault(struct kvm_vcpu *vcpu, gva_t gva,
1033 u32 error_code)
1034 {
1035 gpa_t addr = gva;
1036 hpa_t paddr;
1037 int r;
1038
1039 r = mmu_topup_memory_caches(vcpu);
1040 if (r)
1041 return r;
1042
1043 ASSERT(vcpu);
1044 ASSERT(VALID_PAGE(vcpu->mmu.root_hpa));
1045
1046
1047 paddr = gpa_to_hpa(vcpu->kvm, addr & PT64_BASE_ADDR_MASK);
1048
1049 if (is_error_hpa(paddr)) {
1050 kvm_release_page(pfn_to_page((paddr & PT64_BASE_ADDR_MASK)
1051 >> PAGE_SHIFT));
1052 return 1;
1053 }
1054
1055 return nonpaging_map(vcpu, addr & PAGE_MASK, paddr);
1056 }
1057
1058 static void nonpaging_free(struct kvm_vcpu *vcpu)
1059 {
1060 mmu_free_roots(vcpu);
1061 }
1062
1063 static int nonpaging_init_context(struct kvm_vcpu *vcpu)
1064 {
1065 struct kvm_mmu *context = &vcpu->mmu;
1066
1067 context->new_cr3 = nonpaging_new_cr3;
1068 context->page_fault = nonpaging_page_fault;
1069 context->gva_to_gpa = nonpaging_gva_to_gpa;
1070 context->free = nonpaging_free;
1071 context->prefetch_page = nonpaging_prefetch_page;
1072 context->root_level = 0;
1073 context->shadow_root_level = PT32E_ROOT_LEVEL;
1074 context->root_hpa = INVALID_PAGE;
1075 return 0;
1076 }
1077
1078 static void kvm_mmu_flush_tlb(struct kvm_vcpu *vcpu)
1079 {
1080 ++vcpu->stat.tlb_flush;
1081 kvm_x86_ops->tlb_flush(vcpu);
1082 }
1083
1084 static void paging_new_cr3(struct kvm_vcpu *vcpu)
1085 {
1086 pgprintk("%s: cr3 %lx\n", __FUNCTION__, vcpu->cr3);
1087 mmu_free_roots(vcpu);
1088 }
1089
1090 static void inject_page_fault(struct kvm_vcpu *vcpu,
1091 u64 addr,
1092 u32 err_code)
1093 {
1094 kvm_x86_ops->inject_page_fault(vcpu, addr, err_code);
1095 }
1096
1097 static void paging_free(struct kvm_vcpu *vcpu)
1098 {
1099 nonpaging_free(vcpu);
1100 }
1101
1102 #define PTTYPE 64
1103 #include "paging_tmpl.h"
1104 #undef PTTYPE
1105
1106 #define PTTYPE 32
1107 #include "paging_tmpl.h"
1108 #undef PTTYPE
1109
1110 static int paging64_init_context_common(struct kvm_vcpu *vcpu, int level)
1111 {
1112 struct kvm_mmu *context = &vcpu->mmu;
1113
1114 ASSERT(is_pae(vcpu));
1115 context->new_cr3 = paging_new_cr3;
1116 context->page_fault = paging64_page_fault;
1117 context->gva_to_gpa = paging64_gva_to_gpa;
1118 context->prefetch_page = paging64_prefetch_page;
1119 context->free = paging_free;
1120 context->root_level = level;
1121 context->shadow_root_level = level;
1122 context->root_hpa = INVALID_PAGE;
1123 return 0;
1124 }
1125
1126 static int paging64_init_context(struct kvm_vcpu *vcpu)
1127 {
1128 return paging64_init_context_common(vcpu, PT64_ROOT_LEVEL);
1129 }
1130
1131 static int paging32_init_context(struct kvm_vcpu *vcpu)
1132 {
1133 struct kvm_mmu *context = &vcpu->mmu;
1134
1135 context->new_cr3 = paging_new_cr3;
1136 context->page_fault = paging32_page_fault;
1137 context->gva_to_gpa = paging32_gva_to_gpa;
1138 context->free = paging_free;
1139 context->prefetch_page = paging32_prefetch_page;
1140 context->root_level = PT32_ROOT_LEVEL;
1141 context->shadow_root_level = PT32E_ROOT_LEVEL;
1142 context->root_hpa = INVALID_PAGE;
1143 return 0;
1144 }
1145
1146 static int paging32E_init_context(struct kvm_vcpu *vcpu)
1147 {
1148 return paging64_init_context_common(vcpu, PT32E_ROOT_LEVEL);
1149 }
1150
1151 static int init_kvm_mmu(struct kvm_vcpu *vcpu)
1152 {
1153 ASSERT(vcpu);
1154 ASSERT(!VALID_PAGE(vcpu->mmu.root_hpa));
1155
1156 if (!is_paging(vcpu))
1157 return nonpaging_init_context(vcpu);
1158 else if (is_long_mode(vcpu))
1159 return paging64_init_context(vcpu);
1160 else if (is_pae(vcpu))
1161 return paging32E_init_context(vcpu);
1162 else
1163 return paging32_init_context(vcpu);
1164 }
1165
1166 static void destroy_kvm_mmu(struct kvm_vcpu *vcpu)
1167 {
1168 ASSERT(vcpu);
1169 if (VALID_PAGE(vcpu->mmu.root_hpa)) {
1170 vcpu->mmu.free(vcpu);
1171 vcpu->mmu.root_hpa = INVALID_PAGE;
1172 }
1173 }
1174
1175 int kvm_mmu_reset_context(struct kvm_vcpu *vcpu)
1176 {
1177 destroy_kvm_mmu(vcpu);
1178 return init_kvm_mmu(vcpu);
1179 }
1180 EXPORT_SYMBOL_GPL(kvm_mmu_reset_context);
1181
1182 int kvm_mmu_load(struct kvm_vcpu *vcpu)
1183 {
1184 int r;
1185
1186 mutex_lock(&vcpu->kvm->lock);
1187 r = mmu_topup_memory_caches(vcpu);
1188 if (r)
1189 goto out;
1190 mmu_alloc_roots(vcpu);
1191 kvm_x86_ops->set_cr3(vcpu, vcpu->mmu.root_hpa);
1192 kvm_mmu_flush_tlb(vcpu);
1193 out:
1194 mutex_unlock(&vcpu->kvm->lock);
1195 return r;
1196 }
1197 EXPORT_SYMBOL_GPL(kvm_mmu_load);
1198
1199 void kvm_mmu_unload(struct kvm_vcpu *vcpu)
1200 {
1201 mmu_free_roots(vcpu);
1202 }
1203
1204 static void mmu_pte_write_zap_pte(struct kvm_vcpu *vcpu,
1205 struct kvm_mmu_page *page,
1206 u64 *spte)
1207 {
1208 u64 pte;
1209 struct kvm_mmu_page *child;
1210
1211 pte = *spte;
1212 if (is_shadow_present_pte(pte)) {
1213 if (page->role.level == PT_PAGE_TABLE_LEVEL)
1214 rmap_remove(vcpu->kvm, spte);
1215 else {
1216 child = page_header(pte & PT64_BASE_ADDR_MASK);
1217 mmu_page_remove_parent_pte(child, spte);
1218 }
1219 }
1220 set_shadow_pte(spte, shadow_trap_nonpresent_pte);
1221 kvm_flush_remote_tlbs(vcpu->kvm);
1222 }
1223
1224 static void mmu_pte_write_new_pte(struct kvm_vcpu *vcpu,
1225 struct kvm_mmu_page *page,
1226 u64 *spte,
1227 const void *new, int bytes,
1228 int offset_in_pte)
1229 {
1230 if (page->role.level != PT_PAGE_TABLE_LEVEL) {
1231 ++vcpu->kvm->stat.mmu_pde_zapped;
1232 return;
1233 }
1234
1235 ++vcpu->kvm->stat.mmu_pte_updated;
1236 if (page->role.glevels == PT32_ROOT_LEVEL)
1237 paging32_update_pte(vcpu, page, spte, new, bytes,
1238 offset_in_pte);
1239 else
1240 paging64_update_pte(vcpu, page, spte, new, bytes,
1241 offset_in_pte);
1242 }
1243
1244 static bool last_updated_pte_accessed(struct kvm_vcpu *vcpu)
1245 {
1246 u64 *spte = vcpu->last_pte_updated;
1247
1248 return !!(spte && (*spte & PT_ACCESSED_MASK));
1249 }
1250
1251 void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
1252 const u8 *new, int bytes)
1253 {
1254 gfn_t gfn = gpa >> PAGE_SHIFT;
1255 struct kvm_mmu_page *page;
1256 struct hlist_node *node, *n;
1257 struct hlist_head *bucket;
1258 unsigned index;
1259 u64 *spte;
1260 unsigned offset = offset_in_page(gpa);
1261 unsigned pte_size;
1262 unsigned page_offset;
1263 unsigned misaligned;
1264 unsigned quadrant;
1265 int level;
1266 int flooded = 0;
1267 int npte;
1268
1269 pgprintk("%s: gpa %llx bytes %d\n", __FUNCTION__, gpa, bytes);
1270 ++vcpu->kvm->stat.mmu_pte_write;
1271 kvm_mmu_audit(vcpu, "pre pte write");
1272 if (gfn == vcpu->last_pt_write_gfn
1273 && !last_updated_pte_accessed(vcpu)) {
1274 ++vcpu->last_pt_write_count;
1275 if (vcpu->last_pt_write_count >= 3)
1276 flooded = 1;
1277 } else {
1278 vcpu->last_pt_write_gfn = gfn;
1279 vcpu->last_pt_write_count = 1;
1280 vcpu->last_pte_updated = NULL;
1281 }
1282 index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES;
1283 bucket = &vcpu->kvm->mmu_page_hash[index];
1284 hlist_for_each_entry_safe(page, node, n, bucket, hash_link) {
1285 if (page->gfn != gfn || page->role.metaphysical)
1286 continue;
1287 pte_size = page->role.glevels == PT32_ROOT_LEVEL ? 4 : 8;
1288 misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1);
1289 misaligned |= bytes < 4;
1290 if (misaligned || flooded) {
1291 /*
1292 * Misaligned accesses are too much trouble to fix
1293 * up; also, they usually indicate a page is not used
1294 * as a page table.
1295 *
1296 * If we're seeing too many writes to a page,
1297 * it may no longer be a page table, or we may be
1298 * forking, in which case it is better to unmap the
1299 * page.
1300 */
1301 pgprintk("misaligned: gpa %llx bytes %d role %x\n",
1302 gpa, bytes, page->role.word);
1303 kvm_mmu_zap_page(vcpu->kvm, page);
1304 ++vcpu->kvm->stat.mmu_flooded;
1305 continue;
1306 }
1307 page_offset = offset;
1308 level = page->role.level;
1309 npte = 1;
1310 if (page->role.glevels == PT32_ROOT_LEVEL) {
1311 page_offset <<= 1; /* 32->64 */
1312 /*
1313 * A 32-bit pde maps 4MB while the shadow pdes map
1314 * only 2MB. So we need to double the offset again
1315 * and zap two pdes instead of one.
1316 */
1317 if (level == PT32_ROOT_LEVEL) {
1318 page_offset &= ~7; /* kill rounding error */
1319 page_offset <<= 1;
1320 npte = 2;
1321 }
1322 quadrant = page_offset >> PAGE_SHIFT;
1323 page_offset &= ~PAGE_MASK;
1324 if (quadrant != page->role.quadrant)
1325 continue;
1326 }
1327 spte = &page->spt[page_offset / sizeof(*spte)];
1328 while (npte--) {
1329 mmu_pte_write_zap_pte(vcpu, page, spte);
1330 mmu_pte_write_new_pte(vcpu, page, spte, new, bytes,
1331 page_offset & (pte_size - 1));
1332 ++spte;
1333 }
1334 }
1335 kvm_mmu_audit(vcpu, "post pte write");
1336 }
1337
1338 int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva)
1339 {
1340 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, gva);
1341
1342 return kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1343 }
1344
1345 void __kvm_mmu_free_some_pages(struct kvm_vcpu *vcpu)
1346 {
1347 while (vcpu->kvm->n_free_mmu_pages < KVM_REFILL_PAGES) {
1348 struct kvm_mmu_page *page;
1349
1350 page = container_of(vcpu->kvm->active_mmu_pages.prev,
1351 struct kvm_mmu_page, link);
1352 kvm_mmu_zap_page(vcpu->kvm, page);
1353 ++vcpu->kvm->stat.mmu_recycled;
1354 }
1355 }
1356
1357 int kvm_mmu_page_fault(struct kvm_vcpu *vcpu, gva_t cr2, u32 error_code)
1358 {
1359 int r;
1360 enum emulation_result er;
1361
1362 mutex_lock(&vcpu->kvm->lock);
1363 r = vcpu->mmu.page_fault(vcpu, cr2, error_code);
1364 if (r < 0)
1365 goto out;
1366
1367 if (!r) {
1368 r = 1;
1369 goto out;
1370 }
1371
1372 r = mmu_topup_memory_caches(vcpu);
1373 if (r)
1374 goto out;
1375
1376 er = emulate_instruction(vcpu, vcpu->run, cr2, error_code, 0);
1377 mutex_unlock(&vcpu->kvm->lock);
1378
1379 switch (er) {
1380 case EMULATE_DONE:
1381 return 1;
1382 case EMULATE_DO_MMIO:
1383 ++vcpu->stat.mmio_exits;
1384 return 0;
1385 case EMULATE_FAIL:
1386 kvm_report_emulation_failure(vcpu, "pagetable");
1387 return 1;
1388 default:
1389 BUG();
1390 }
1391 out:
1392 mutex_unlock(&vcpu->kvm->lock);
1393 return r;
1394 }
1395 EXPORT_SYMBOL_GPL(kvm_mmu_page_fault);
1396
1397 static void free_mmu_pages(struct kvm_vcpu *vcpu)
1398 {
1399 struct kvm_mmu_page *page;
1400
1401 while (!list_empty(&vcpu->kvm->active_mmu_pages)) {
1402 page = container_of(vcpu->kvm->active_mmu_pages.next,
1403 struct kvm_mmu_page, link);
1404 kvm_mmu_zap_page(vcpu->kvm, page);
1405 }
1406 free_page((unsigned long)vcpu->mmu.pae_root);
1407 }
1408
1409 static int alloc_mmu_pages(struct kvm_vcpu *vcpu)
1410 {
1411 struct page *page;
1412 int i;
1413
1414 ASSERT(vcpu);
1415
1416 if (vcpu->kvm->n_requested_mmu_pages)
1417 vcpu->kvm->n_free_mmu_pages = vcpu->kvm->n_requested_mmu_pages;
1418 else
1419 vcpu->kvm->n_free_mmu_pages = vcpu->kvm->n_alloc_mmu_pages;
1420 /*
1421 * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64.
1422 * Therefore we need to allocate shadow page tables in the first
1423 * 4GB of memory, which happens to fit the DMA32 zone.
1424 */
1425 page = alloc_page(GFP_KERNEL | __GFP_DMA32);
1426 if (!page)
1427 goto error_1;
1428 vcpu->mmu.pae_root = page_address(page);
1429 for (i = 0; i < 4; ++i)
1430 vcpu->mmu.pae_root[i] = INVALID_PAGE;
1431
1432 return 0;
1433
1434 error_1:
1435 free_mmu_pages(vcpu);
1436 return -ENOMEM;
1437 }
1438
1439 int kvm_mmu_create(struct kvm_vcpu *vcpu)
1440 {
1441 ASSERT(vcpu);
1442 ASSERT(!VALID_PAGE(vcpu->mmu.root_hpa));
1443
1444 return alloc_mmu_pages(vcpu);
1445 }
1446
1447 int kvm_mmu_setup(struct kvm_vcpu *vcpu)
1448 {
1449 ASSERT(vcpu);
1450 ASSERT(!VALID_PAGE(vcpu->mmu.root_hpa));
1451
1452 return init_kvm_mmu(vcpu);
1453 }
1454
1455 void kvm_mmu_destroy(struct kvm_vcpu *vcpu)
1456 {
1457 ASSERT(vcpu);
1458
1459 destroy_kvm_mmu(vcpu);
1460 free_mmu_pages(vcpu);
1461 mmu_free_memory_caches(vcpu);
1462 }
1463
1464 void kvm_mmu_slot_remove_write_access(struct kvm *kvm, int slot)
1465 {
1466 struct kvm_mmu_page *page;
1467
1468 list_for_each_entry(page, &kvm->active_mmu_pages, link) {
1469 int i;
1470 u64 *pt;
1471
1472 if (!test_bit(slot, &page->slot_bitmap))
1473 continue;
1474
1475 pt = page->spt;
1476 for (i = 0; i < PT64_ENT_PER_PAGE; ++i)
1477 /* avoid RMW */
1478 if (pt[i] & PT_WRITABLE_MASK)
1479 pt[i] &= ~PT_WRITABLE_MASK;
1480 }
1481 }
1482
1483 void kvm_mmu_zap_all(struct kvm *kvm)
1484 {
1485 struct kvm_mmu_page *page, *node;
1486
1487 list_for_each_entry_safe(page, node, &kvm->active_mmu_pages, link)
1488 kvm_mmu_zap_page(kvm, page);
1489
1490 kvm_flush_remote_tlbs(kvm);
1491 }
1492
1493 void kvm_mmu_module_exit(void)
1494 {
1495 if (pte_chain_cache)
1496 kmem_cache_destroy(pte_chain_cache);
1497 if (rmap_desc_cache)
1498 kmem_cache_destroy(rmap_desc_cache);
1499 if (mmu_page_header_cache)
1500 kmem_cache_destroy(mmu_page_header_cache);
1501 }
1502
1503 int kvm_mmu_module_init(void)
1504 {
1505 pte_chain_cache = kmem_cache_create("kvm_pte_chain",
1506 sizeof(struct kvm_pte_chain),
1507 0, 0, NULL);
1508 if (!pte_chain_cache)
1509 goto nomem;
1510 rmap_desc_cache = kmem_cache_create("kvm_rmap_desc",
1511 sizeof(struct kvm_rmap_desc),
1512 0, 0, NULL);
1513 if (!rmap_desc_cache)
1514 goto nomem;
1515
1516 mmu_page_header_cache = kmem_cache_create("kvm_mmu_page_header",
1517 sizeof(struct kvm_mmu_page),
1518 0, 0, NULL);
1519 if (!mmu_page_header_cache)
1520 goto nomem;
1521
1522 return 0;
1523
1524 nomem:
1525 kvm_mmu_module_exit();
1526 return -ENOMEM;
1527 }
1528
1529 #ifdef AUDIT
1530
1531 static const char *audit_msg;
1532
1533 static gva_t canonicalize(gva_t gva)
1534 {
1535 #ifdef CONFIG_X86_64
1536 gva = (long long)(gva << 16) >> 16;
1537 #endif
1538 return gva;
1539 }
1540
1541 static void audit_mappings_page(struct kvm_vcpu *vcpu, u64 page_pte,
1542 gva_t va, int level)
1543 {
1544 u64 *pt = __va(page_pte & PT64_BASE_ADDR_MASK);
1545 int i;
1546 gva_t va_delta = 1ul << (PAGE_SHIFT + 9 * (level - 1));
1547
1548 for (i = 0; i < PT64_ENT_PER_PAGE; ++i, va += va_delta) {
1549 u64 ent = pt[i];
1550
1551 if (ent == shadow_trap_nonpresent_pte)
1552 continue;
1553
1554 va = canonicalize(va);
1555 if (level > 1) {
1556 if (ent == shadow_notrap_nonpresent_pte)
1557 printk(KERN_ERR "audit: (%s) nontrapping pte"
1558 " in nonleaf level: levels %d gva %lx"
1559 " level %d pte %llx\n", audit_msg,
1560 vcpu->mmu.root_level, va, level, ent);
1561
1562 audit_mappings_page(vcpu, ent, va, level - 1);
1563 } else {
1564 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, va);
1565 hpa_t hpa = gpa_to_hpa(vcpu, gpa);
1566 struct page *page;
1567
1568 if (is_shadow_present_pte(ent)
1569 && (ent & PT64_BASE_ADDR_MASK) != hpa)
1570 printk(KERN_ERR "xx audit error: (%s) levels %d"
1571 " gva %lx gpa %llx hpa %llx ent %llx %d\n",
1572 audit_msg, vcpu->mmu.root_level,
1573 va, gpa, hpa, ent,
1574 is_shadow_present_pte(ent));
1575 else if (ent == shadow_notrap_nonpresent_pte
1576 && !is_error_hpa(hpa))
1577 printk(KERN_ERR "audit: (%s) notrap shadow,"
1578 " valid guest gva %lx\n", audit_msg, va);
1579 page = pfn_to_page((gpa & PT64_BASE_ADDR_MASK)
1580 >> PAGE_SHIFT);
1581 kvm_release_page(page);
1582
1583 }
1584 }
1585 }
1586
1587 static void audit_mappings(struct kvm_vcpu *vcpu)
1588 {
1589 unsigned i;
1590
1591 if (vcpu->mmu.root_level == 4)
1592 audit_mappings_page(vcpu, vcpu->mmu.root_hpa, 0, 4);
1593 else
1594 for (i = 0; i < 4; ++i)
1595 if (vcpu->mmu.pae_root[i] & PT_PRESENT_MASK)
1596 audit_mappings_page(vcpu,
1597 vcpu->mmu.pae_root[i],
1598 i << 30,
1599 2);
1600 }
1601
1602 static int count_rmaps(struct kvm_vcpu *vcpu)
1603 {
1604 int nmaps = 0;
1605 int i, j, k;
1606
1607 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1608 struct kvm_memory_slot *m = &vcpu->kvm->memslots[i];
1609 struct kvm_rmap_desc *d;
1610
1611 for (j = 0; j < m->npages; ++j) {
1612 unsigned long *rmapp = &m->rmap[j];
1613
1614 if (!*rmapp)
1615 continue;
1616 if (!(*rmapp & 1)) {
1617 ++nmaps;
1618 continue;
1619 }
1620 d = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
1621 while (d) {
1622 for (k = 0; k < RMAP_EXT; ++k)
1623 if (d->shadow_ptes[k])
1624 ++nmaps;
1625 else
1626 break;
1627 d = d->more;
1628 }
1629 }
1630 }
1631 return nmaps;
1632 }
1633
1634 static int count_writable_mappings(struct kvm_vcpu *vcpu)
1635 {
1636 int nmaps = 0;
1637 struct kvm_mmu_page *page;
1638 int i;
1639
1640 list_for_each_entry(page, &vcpu->kvm->active_mmu_pages, link) {
1641 u64 *pt = page->spt;
1642
1643 if (page->role.level != PT_PAGE_TABLE_LEVEL)
1644 continue;
1645
1646 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
1647 u64 ent = pt[i];
1648
1649 if (!(ent & PT_PRESENT_MASK))
1650 continue;
1651 if (!(ent & PT_WRITABLE_MASK))
1652 continue;
1653 ++nmaps;
1654 }
1655 }
1656 return nmaps;
1657 }
1658
1659 static void audit_rmap(struct kvm_vcpu *vcpu)
1660 {
1661 int n_rmap = count_rmaps(vcpu);
1662 int n_actual = count_writable_mappings(vcpu);
1663
1664 if (n_rmap != n_actual)
1665 printk(KERN_ERR "%s: (%s) rmap %d actual %d\n",
1666 __FUNCTION__, audit_msg, n_rmap, n_actual);
1667 }
1668
1669 static void audit_write_protection(struct kvm_vcpu *vcpu)
1670 {
1671 struct kvm_mmu_page *page;
1672 struct kvm_memory_slot *slot;
1673 unsigned long *rmapp;
1674 gfn_t gfn;
1675
1676 list_for_each_entry(page, &vcpu->kvm->active_mmu_pages, link) {
1677 if (page->role.metaphysical)
1678 continue;
1679
1680 slot = gfn_to_memslot(vcpu->kvm, page->gfn);
1681 gfn = unalias_gfn(vcpu->kvm, page->gfn);
1682 rmapp = &slot->rmap[gfn - slot->base_gfn];
1683 if (*rmapp)
1684 printk(KERN_ERR "%s: (%s) shadow page has writable"
1685 " mappings: gfn %lx role %x\n",
1686 __FUNCTION__, audit_msg, page->gfn,
1687 page->role.word);
1688 }
1689 }
1690
1691 static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg)
1692 {
1693 int olddbg = dbg;
1694
1695 dbg = 0;
1696 audit_msg = msg;
1697 audit_rmap(vcpu);
1698 audit_write_protection(vcpu);
1699 audit_mappings(vcpu);
1700 dbg = olddbg;
1701 }
1702
1703 #endif
Linux kernel - Deliverables
This page took 0.06638 seconds and 6 git commands to generate.