projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
KVM: MMU: Fix inherited permissions for emulated guest pte updates
[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 #include <linux/swap.h>
30
31 #include <asm/page.h>
32 #include <asm/cmpxchg.h>
33 #include <asm/io.h>
34
35 #undef MMU_DEBUG
36
37 #undef AUDIT
38
39 #ifdef AUDIT
40 static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg);
41 #else
42 static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg) {}
43 #endif
44
45 #ifdef MMU_DEBUG
46
47 #define pgprintk(x...) do { if (dbg) printk(x); } while (0)
48 #define rmap_printk(x...) do { if (dbg) printk(x); } while (0)
49
50 #else
51
52 #define pgprintk(x...) do { } while (0)
53 #define rmap_printk(x...) do { } while (0)
54
55 #endif
56
57 #if defined(MMU_DEBUG) || defined(AUDIT)
58 static int dbg = 1;
59 #endif
60
61 #ifndef MMU_DEBUG
62 #define ASSERT(x) do { } while (0)
63 #else
64 #define ASSERT(x) \
65 if (!(x)) { \
66 printk(KERN_WARNING "assertion failed %s:%d: %s\n", \
67 __FILE__, __LINE__, #x); \
68 }
69 #endif
70
71 #define PT64_PT_BITS 9
72 #define PT64_ENT_PER_PAGE (1 << PT64_PT_BITS)
73 #define PT32_PT_BITS 10
74 #define PT32_ENT_PER_PAGE (1 << PT32_PT_BITS)
75
76 #define PT_WRITABLE_SHIFT 1
77
78 #define PT_PRESENT_MASK (1ULL << 0)
79 #define PT_WRITABLE_MASK (1ULL << PT_WRITABLE_SHIFT)
80 #define PT_USER_MASK (1ULL << 2)
81 #define PT_PWT_MASK (1ULL << 3)
82 #define PT_PCD_MASK (1ULL << 4)
83 #define PT_ACCESSED_MASK (1ULL << 5)
84 #define PT_DIRTY_MASK (1ULL << 6)
85 #define PT_PAGE_SIZE_MASK (1ULL << 7)
86 #define PT_PAT_MASK (1ULL << 7)
87 #define PT_GLOBAL_MASK (1ULL << 8)
88 #define PT64_NX_SHIFT 63
89 #define PT64_NX_MASK (1ULL << PT64_NX_SHIFT)
90
91 #define PT_PAT_SHIFT 7
92 #define PT_DIR_PAT_SHIFT 12
93 #define PT_DIR_PAT_MASK (1ULL << PT_DIR_PAT_SHIFT)
94
95 #define PT32_DIR_PSE36_SIZE 4
96 #define PT32_DIR_PSE36_SHIFT 13
97 #define PT32_DIR_PSE36_MASK \
98 (((1ULL << PT32_DIR_PSE36_SIZE) - 1) << PT32_DIR_PSE36_SHIFT)
99
100
101 #define PT_FIRST_AVAIL_BITS_SHIFT 9
102 #define PT64_SECOND_AVAIL_BITS_SHIFT 52
103
104 #define PT_SHADOW_IO_MARK (1ULL << PT_FIRST_AVAIL_BITS_SHIFT)
105
106 #define VALID_PAGE(x) ((x) != INVALID_PAGE)
107
108 #define PT64_LEVEL_BITS 9
109
110 #define PT64_LEVEL_SHIFT(level) \
111 (PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS)
112
113 #define PT64_LEVEL_MASK(level) \
114 (((1ULL << PT64_LEVEL_BITS) - 1) << PT64_LEVEL_SHIFT(level))
115
116 #define PT64_INDEX(address, level)\
117 (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1))
118
119
120 #define PT32_LEVEL_BITS 10
121
122 #define PT32_LEVEL_SHIFT(level) \
123 (PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS)
124
125 #define PT32_LEVEL_MASK(level) \
126 (((1ULL << PT32_LEVEL_BITS) - 1) << PT32_LEVEL_SHIFT(level))
127
128 #define PT32_INDEX(address, level)\
129 (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1))
130
131
132 #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1))
133 #define PT64_DIR_BASE_ADDR_MASK \
134 (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1))
135
136 #define PT32_BASE_ADDR_MASK PAGE_MASK
137 #define PT32_DIR_BASE_ADDR_MASK \
138 (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1))
139
140 #define PT64_PERM_MASK (PT_PRESENT_MASK | PT_WRITABLE_MASK | PT_USER_MASK \
141 | PT64_NX_MASK)
142
143 #define PFERR_PRESENT_MASK (1U << 0)
144 #define PFERR_WRITE_MASK (1U << 1)
145 #define PFERR_USER_MASK (1U << 2)
146 #define PFERR_FETCH_MASK (1U << 4)
147
148 #define PT64_ROOT_LEVEL 4
149 #define PT32_ROOT_LEVEL 2
150 #define PT32E_ROOT_LEVEL 3
151
152 #define PT_DIRECTORY_LEVEL 2
153 #define PT_PAGE_TABLE_LEVEL 1
154
155 #define RMAP_EXT 4
156
157 #define ACC_EXEC_MASK 1
158 #define ACC_WRITE_MASK PT_WRITABLE_MASK
159 #define ACC_USER_MASK PT_USER_MASK
160 #define ACC_ALL (ACC_EXEC_MASK | ACC_WRITE_MASK | ACC_USER_MASK)
161
162 struct kvm_rmap_desc {
163 u64 *shadow_ptes[RMAP_EXT];
164 struct kvm_rmap_desc *more;
165 };
166
167 static struct kmem_cache *pte_chain_cache;
168 static struct kmem_cache *rmap_desc_cache;
169 static struct kmem_cache *mmu_page_header_cache;
170
171 static u64 __read_mostly shadow_trap_nonpresent_pte;
172 static u64 __read_mostly shadow_notrap_nonpresent_pte;
173
174 void kvm_mmu_set_nonpresent_ptes(u64 trap_pte, u64 notrap_pte)
175 {
176 shadow_trap_nonpresent_pte = trap_pte;
177 shadow_notrap_nonpresent_pte = notrap_pte;
178 }
179 EXPORT_SYMBOL_GPL(kvm_mmu_set_nonpresent_ptes);
180
181 static int is_write_protection(struct kvm_vcpu *vcpu)
182 {
183 return vcpu->cr0 & X86_CR0_WP;
184 }
185
186 static int is_cpuid_PSE36(void)
187 {
188 return 1;
189 }
190
191 static int is_nx(struct kvm_vcpu *vcpu)
192 {
193 return vcpu->shadow_efer & EFER_NX;
194 }
195
196 static int is_present_pte(unsigned long pte)
197 {
198 return pte & PT_PRESENT_MASK;
199 }
200
201 static int is_shadow_present_pte(u64 pte)
202 {
203 pte &= ~PT_SHADOW_IO_MARK;
204 return pte != shadow_trap_nonpresent_pte
205 && pte != shadow_notrap_nonpresent_pte;
206 }
207
208 static int is_writeble_pte(unsigned long pte)
209 {
210 return pte & PT_WRITABLE_MASK;
211 }
212
213 static int is_dirty_pte(unsigned long pte)
214 {
215 return pte & PT_DIRTY_MASK;
216 }
217
218 static int is_io_pte(unsigned long pte)
219 {
220 return pte & PT_SHADOW_IO_MARK;
221 }
222
223 static int is_rmap_pte(u64 pte)
224 {
225 return pte != shadow_trap_nonpresent_pte
226 && pte != shadow_notrap_nonpresent_pte;
227 }
228
229 static gfn_t pse36_gfn_delta(u32 gpte)
230 {
231 int shift = 32 - PT32_DIR_PSE36_SHIFT - PAGE_SHIFT;
232
233 return (gpte & PT32_DIR_PSE36_MASK) << shift;
234 }
235
236 static void set_shadow_pte(u64 *sptep, u64 spte)
237 {
238 #ifdef CONFIG_X86_64
239 set_64bit((unsigned long *)sptep, spte);
240 #else
241 set_64bit((unsigned long long *)sptep, spte);
242 #endif
243 }
244
245 static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache,
246 struct kmem_cache *base_cache, int min)
247 {
248 void *obj;
249
250 if (cache->nobjs >= min)
251 return 0;
252 while (cache->nobjs < ARRAY_SIZE(cache->objects)) {
253 obj = kmem_cache_zalloc(base_cache, GFP_KERNEL);
254 if (!obj)
255 return -ENOMEM;
256 cache->objects[cache->nobjs++] = obj;
257 }
258 return 0;
259 }
260
261 static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
262 {
263 while (mc->nobjs)
264 kfree(mc->objects[--mc->nobjs]);
265 }
266
267 static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache *cache,
268 int min)
269 {
270 struct page *page;
271
272 if (cache->nobjs >= min)
273 return 0;
274 while (cache->nobjs < ARRAY_SIZE(cache->objects)) {
275 page = alloc_page(GFP_KERNEL);
276 if (!page)
277 return -ENOMEM;
278 set_page_private(page, 0);
279 cache->objects[cache->nobjs++] = page_address(page);
280 }
281 return 0;
282 }
283
284 static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache *mc)
285 {
286 while (mc->nobjs)
287 free_page((unsigned long)mc->objects[--mc->nobjs]);
288 }
289
290 static int mmu_topup_memory_caches(struct kvm_vcpu *vcpu)
291 {
292 int r;
293
294 kvm_mmu_free_some_pages(vcpu);
295 r = mmu_topup_memory_cache(&vcpu->mmu_pte_chain_cache,
296 pte_chain_cache, 4);
297 if (r)
298 goto out;
299 r = mmu_topup_memory_cache(&vcpu->mmu_rmap_desc_cache,
300 rmap_desc_cache, 1);
301 if (r)
302 goto out;
303 r = mmu_topup_memory_cache_page(&vcpu->mmu_page_cache, 8);
304 if (r)
305 goto out;
306 r = mmu_topup_memory_cache(&vcpu->mmu_page_header_cache,
307 mmu_page_header_cache, 4);
308 out:
309 return r;
310 }
311
312 static void mmu_free_memory_caches(struct kvm_vcpu *vcpu)
313 {
314 mmu_free_memory_cache(&vcpu->mmu_pte_chain_cache);
315 mmu_free_memory_cache(&vcpu->mmu_rmap_desc_cache);
316 mmu_free_memory_cache_page(&vcpu->mmu_page_cache);
317 mmu_free_memory_cache(&vcpu->mmu_page_header_cache);
318 }
319
320 static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc,
321 size_t size)
322 {
323 void *p;
324
325 BUG_ON(!mc->nobjs);
326 p = mc->objects[--mc->nobjs];
327 memset(p, 0, size);
328 return p;
329 }
330
331 static struct kvm_pte_chain *mmu_alloc_pte_chain(struct kvm_vcpu *vcpu)
332 {
333 return mmu_memory_cache_alloc(&vcpu->mmu_pte_chain_cache,
334 sizeof(struct kvm_pte_chain));
335 }
336
337 static void mmu_free_pte_chain(struct kvm_pte_chain *pc)
338 {
339 kfree(pc);
340 }
341
342 static struct kvm_rmap_desc *mmu_alloc_rmap_desc(struct kvm_vcpu *vcpu)
343 {
344 return mmu_memory_cache_alloc(&vcpu->mmu_rmap_desc_cache,
345 sizeof(struct kvm_rmap_desc));
346 }
347
348 static void mmu_free_rmap_desc(struct kvm_rmap_desc *rd)
349 {
350 kfree(rd);
351 }
352
353 /*
354 * Take gfn and return the reverse mapping to it.
355 * Note: gfn must be unaliased before this function get called
356 */
357
358 static unsigned long *gfn_to_rmap(struct kvm *kvm, gfn_t gfn)
359 {
360 struct kvm_memory_slot *slot;
361
362 slot = gfn_to_memslot(kvm, gfn);
363 return &slot->rmap[gfn - slot->base_gfn];
364 }
365
366 /*
367 * Reverse mapping data structures:
368 *
369 * If rmapp bit zero is zero, then rmapp point to the shadw page table entry
370 * that points to page_address(page).
371 *
372 * If rmapp bit zero is one, (then rmap & ~1) points to a struct kvm_rmap_desc
373 * containing more mappings.
374 */
375 static void rmap_add(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn)
376 {
377 struct kvm_mmu_page *sp;
378 struct kvm_rmap_desc *desc;
379 unsigned long *rmapp;
380 int i;
381
382 if (!is_rmap_pte(*spte))
383 return;
384 gfn = unalias_gfn(vcpu->kvm, gfn);
385 sp = page_header(__pa(spte));
386 sp->gfns[spte - sp->spt] = gfn;
387 rmapp = gfn_to_rmap(vcpu->kvm, gfn);
388 if (!*rmapp) {
389 rmap_printk("rmap_add: %p %llx 0->1\n", spte, *spte);
390 *rmapp = (unsigned long)spte;
391 } else if (!(*rmapp & 1)) {
392 rmap_printk("rmap_add: %p %llx 1->many\n", spte, *spte);
393 desc = mmu_alloc_rmap_desc(vcpu);
394 desc->shadow_ptes[0] = (u64 *)*rmapp;
395 desc->shadow_ptes[1] = spte;
396 *rmapp = (unsigned long)desc | 1;
397 } else {
398 rmap_printk("rmap_add: %p %llx many->many\n", spte, *spte);
399 desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
400 while (desc->shadow_ptes[RMAP_EXT-1] && desc->more)
401 desc = desc->more;
402 if (desc->shadow_ptes[RMAP_EXT-1]) {
403 desc->more = mmu_alloc_rmap_desc(vcpu);
404 desc = desc->more;
405 }
406 for (i = 0; desc->shadow_ptes[i]; ++i)
407 ;
408 desc->shadow_ptes[i] = spte;
409 }
410 }
411
412 static void rmap_desc_remove_entry(unsigned long *rmapp,
413 struct kvm_rmap_desc *desc,
414 int i,
415 struct kvm_rmap_desc *prev_desc)
416 {
417 int j;
418
419 for (j = RMAP_EXT - 1; !desc->shadow_ptes[j] && j > i; --j)
420 ;
421 desc->shadow_ptes[i] = desc->shadow_ptes[j];
422 desc->shadow_ptes[j] = NULL;
423 if (j != 0)
424 return;
425 if (!prev_desc && !desc->more)
426 *rmapp = (unsigned long)desc->shadow_ptes[0];
427 else
428 if (prev_desc)
429 prev_desc->more = desc->more;
430 else
431 *rmapp = (unsigned long)desc->more | 1;
432 mmu_free_rmap_desc(desc);
433 }
434
435 static void rmap_remove(struct kvm *kvm, u64 *spte)
436 {
437 struct kvm_rmap_desc *desc;
438 struct kvm_rmap_desc *prev_desc;
439 struct kvm_mmu_page *sp;
440 struct page *page;
441 unsigned long *rmapp;
442 int i;
443
444 if (!is_rmap_pte(*spte))
445 return;
446 sp = page_header(__pa(spte));
447 page = pfn_to_page((*spte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT);
448 mark_page_accessed(page);
449 if (is_writeble_pte(*spte))
450 kvm_release_page_dirty(page);
451 else
452 kvm_release_page_clean(page);
453 rmapp = gfn_to_rmap(kvm, sp->gfns[spte - sp->spt]);
454 if (!*rmapp) {
455 printk(KERN_ERR "rmap_remove: %p %llx 0->BUG\n", spte, *spte);
456 BUG();
457 } else if (!(*rmapp & 1)) {
458 rmap_printk("rmap_remove: %p %llx 1->0\n", spte, *spte);
459 if ((u64 *)*rmapp != spte) {
460 printk(KERN_ERR "rmap_remove: %p %llx 1->BUG\n",
461 spte, *spte);
462 BUG();
463 }
464 *rmapp = 0;
465 } else {
466 rmap_printk("rmap_remove: %p %llx many->many\n", spte, *spte);
467 desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
468 prev_desc = NULL;
469 while (desc) {
470 for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i)
471 if (desc->shadow_ptes[i] == spte) {
472 rmap_desc_remove_entry(rmapp,
473 desc, i,
474 prev_desc);
475 return;
476 }
477 prev_desc = desc;
478 desc = desc->more;
479 }
480 BUG();
481 }
482 }
483
484 static u64 *rmap_next(struct kvm *kvm, unsigned long *rmapp, u64 *spte)
485 {
486 struct kvm_rmap_desc *desc;
487 struct kvm_rmap_desc *prev_desc;
488 u64 *prev_spte;
489 int i;
490
491 if (!*rmapp)
492 return NULL;
493 else if (!(*rmapp & 1)) {
494 if (!spte)
495 return (u64 *)*rmapp;
496 return NULL;
497 }
498 desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
499 prev_desc = NULL;
500 prev_spte = NULL;
501 while (desc) {
502 for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i) {
503 if (prev_spte == spte)
504 return desc->shadow_ptes[i];
505 prev_spte = desc->shadow_ptes[i];
506 }
507 desc = desc->more;
508 }
509 return NULL;
510 }
511
512 static void rmap_write_protect(struct kvm *kvm, u64 gfn)
513 {
514 unsigned long *rmapp;
515 u64 *spte;
516
517 gfn = unalias_gfn(kvm, gfn);
518 rmapp = gfn_to_rmap(kvm, gfn);
519
520 spte = rmap_next(kvm, rmapp, NULL);
521 while (spte) {
522 BUG_ON(!spte);
523 BUG_ON(!(*spte & PT_PRESENT_MASK));
524 rmap_printk("rmap_write_protect: spte %p %llx\n", spte, *spte);
525 if (is_writeble_pte(*spte))
526 set_shadow_pte(spte, *spte & ~PT_WRITABLE_MASK);
527 kvm_flush_remote_tlbs(kvm);
528 spte = rmap_next(kvm, rmapp, spte);
529 }
530 }
531
532 #ifdef MMU_DEBUG
533 static int is_empty_shadow_page(u64 *spt)
534 {
535 u64 *pos;
536 u64 *end;
537
538 for (pos = spt, end = pos + PAGE_SIZE / sizeof(u64); pos != end; pos++)
539 if ((*pos & ~PT_SHADOW_IO_MARK) != shadow_trap_nonpresent_pte) {
540 printk(KERN_ERR "%s: %p %llx\n", __FUNCTION__,
541 pos, *pos);
542 return 0;
543 }
544 return 1;
545 }
546 #endif
547
548 static void kvm_mmu_free_page(struct kvm *kvm, struct kvm_mmu_page *sp)
549 {
550 ASSERT(is_empty_shadow_page(sp->spt));
551 list_del(&sp->link);
552 __free_page(virt_to_page(sp->spt));
553 __free_page(virt_to_page(sp->gfns));
554 kfree(sp);
555 ++kvm->n_free_mmu_pages;
556 }
557
558 static unsigned kvm_page_table_hashfn(gfn_t gfn)
559 {
560 return gfn;
561 }
562
563 static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu,
564 u64 *parent_pte)
565 {
566 struct kvm_mmu_page *sp;
567
568 if (!vcpu->kvm->n_free_mmu_pages)
569 return NULL;
570
571 sp = mmu_memory_cache_alloc(&vcpu->mmu_page_header_cache, sizeof *sp);
572 sp->spt = mmu_memory_cache_alloc(&vcpu->mmu_page_cache, PAGE_SIZE);
573 sp->gfns = mmu_memory_cache_alloc(&vcpu->mmu_page_cache, PAGE_SIZE);
574 set_page_private(virt_to_page(sp->spt), (unsigned long)sp);
575 list_add(&sp->link, &vcpu->kvm->active_mmu_pages);
576 ASSERT(is_empty_shadow_page(sp->spt));
577 sp->slot_bitmap = 0;
578 sp->multimapped = 0;
579 sp->parent_pte = parent_pte;
580 --vcpu->kvm->n_free_mmu_pages;
581 return sp;
582 }
583
584 static void mmu_page_add_parent_pte(struct kvm_vcpu *vcpu,
585 struct kvm_mmu_page *sp, u64 *parent_pte)
586 {
587 struct kvm_pte_chain *pte_chain;
588 struct hlist_node *node;
589 int i;
590
591 if (!parent_pte)
592 return;
593 if (!sp->multimapped) {
594 u64 *old = sp->parent_pte;
595
596 if (!old) {
597 sp->parent_pte = parent_pte;
598 return;
599 }
600 sp->multimapped = 1;
601 pte_chain = mmu_alloc_pte_chain(vcpu);
602 INIT_HLIST_HEAD(&sp->parent_ptes);
603 hlist_add_head(&pte_chain->link, &sp->parent_ptes);
604 pte_chain->parent_ptes[0] = old;
605 }
606 hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link) {
607 if (pte_chain->parent_ptes[NR_PTE_CHAIN_ENTRIES-1])
608 continue;
609 for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i)
610 if (!pte_chain->parent_ptes[i]) {
611 pte_chain->parent_ptes[i] = parent_pte;
612 return;
613 }
614 }
615 pte_chain = mmu_alloc_pte_chain(vcpu);
616 BUG_ON(!pte_chain);
617 hlist_add_head(&pte_chain->link, &sp->parent_ptes);
618 pte_chain->parent_ptes[0] = parent_pte;
619 }
620
621 static void mmu_page_remove_parent_pte(struct kvm_mmu_page *sp,
622 u64 *parent_pte)
623 {
624 struct kvm_pte_chain *pte_chain;
625 struct hlist_node *node;
626 int i;
627
628 if (!sp->multimapped) {
629 BUG_ON(sp->parent_pte != parent_pte);
630 sp->parent_pte = NULL;
631 return;
632 }
633 hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link)
634 for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) {
635 if (!pte_chain->parent_ptes[i])
636 break;
637 if (pte_chain->parent_ptes[i] != parent_pte)
638 continue;
639 while (i + 1 < NR_PTE_CHAIN_ENTRIES
640 && pte_chain->parent_ptes[i + 1]) {
641 pte_chain->parent_ptes[i]
642 = pte_chain->parent_ptes[i + 1];
643 ++i;
644 }
645 pte_chain->parent_ptes[i] = NULL;
646 if (i == 0) {
647 hlist_del(&pte_chain->link);
648 mmu_free_pte_chain(pte_chain);
649 if (hlist_empty(&sp->parent_ptes)) {
650 sp->multimapped = 0;
651 sp->parent_pte = NULL;
652 }
653 }
654 return;
655 }
656 BUG();
657 }
658
659 static struct kvm_mmu_page *kvm_mmu_lookup_page(struct kvm *kvm, gfn_t gfn)
660 {
661 unsigned index;
662 struct hlist_head *bucket;
663 struct kvm_mmu_page *sp;
664 struct hlist_node *node;
665
666 pgprintk("%s: looking for gfn %lx\n", __FUNCTION__, gfn);
667 index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES;
668 bucket = &kvm->mmu_page_hash[index];
669 hlist_for_each_entry(sp, node, bucket, hash_link)
670 if (sp->gfn == gfn && !sp->role.metaphysical) {
671 pgprintk("%s: found role %x\n",
672 __FUNCTION__, sp->role.word);
673 return sp;
674 }
675 return NULL;
676 }
677
678 static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
679 gfn_t gfn,
680 gva_t gaddr,
681 unsigned level,
682 int metaphysical,
683 unsigned access,
684 u64 *parent_pte)
685 {
686 union kvm_mmu_page_role role;
687 unsigned index;
688 unsigned quadrant;
689 struct hlist_head *bucket;
690 struct kvm_mmu_page *sp;
691 struct hlist_node *node;
692
693 role.word = 0;
694 role.glevels = vcpu->mmu.root_level;
695 role.level = level;
696 role.metaphysical = metaphysical;
697 role.access = access;
698 if (vcpu->mmu.root_level <= PT32_ROOT_LEVEL) {
699 quadrant = gaddr >> (PAGE_SHIFT + (PT64_PT_BITS * level));
700 quadrant &= (1 << ((PT32_PT_BITS - PT64_PT_BITS) * level)) - 1;
701 role.quadrant = quadrant;
702 }
703 pgprintk("%s: looking gfn %lx role %x\n", __FUNCTION__,
704 gfn, role.word);
705 index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES;
706 bucket = &vcpu->kvm->mmu_page_hash[index];
707 hlist_for_each_entry(sp, node, bucket, hash_link)
708 if (sp->gfn == gfn && sp->role.word == role.word) {
709 mmu_page_add_parent_pte(vcpu, sp, parent_pte);
710 pgprintk("%s: found\n", __FUNCTION__);
711 return sp;
712 }
713 sp = kvm_mmu_alloc_page(vcpu, parent_pte);
714 if (!sp)
715 return sp;
716 pgprintk("%s: adding gfn %lx role %x\n", __FUNCTION__, gfn, role.word);
717 sp->gfn = gfn;
718 sp->role = role;
719 hlist_add_head(&sp->hash_link, bucket);
720 vcpu->mmu.prefetch_page(vcpu, sp);
721 if (!metaphysical)
722 rmap_write_protect(vcpu->kvm, gfn);
723 return sp;
724 }
725
726 static void kvm_mmu_page_unlink_children(struct kvm *kvm,
727 struct kvm_mmu_page *sp)
728 {
729 unsigned i;
730 u64 *pt;
731 u64 ent;
732
733 pt = sp->spt;
734
735 if (sp->role.level == PT_PAGE_TABLE_LEVEL) {
736 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
737 if (is_shadow_present_pte(pt[i]))
738 rmap_remove(kvm, &pt[i]);
739 pt[i] = shadow_trap_nonpresent_pte;
740 }
741 kvm_flush_remote_tlbs(kvm);
742 return;
743 }
744
745 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
746 ent = pt[i];
747
748 pt[i] = shadow_trap_nonpresent_pte;
749 if (!is_shadow_present_pte(ent))
750 continue;
751 ent &= PT64_BASE_ADDR_MASK;
752 mmu_page_remove_parent_pte(page_header(ent), &pt[i]);
753 }
754 kvm_flush_remote_tlbs(kvm);
755 }
756
757 static void kvm_mmu_put_page(struct kvm_mmu_page *sp, u64 *parent_pte)
758 {
759 mmu_page_remove_parent_pte(sp, parent_pte);
760 }
761
762 static void kvm_mmu_reset_last_pte_updated(struct kvm *kvm)
763 {
764 int i;
765
766 for (i = 0; i < KVM_MAX_VCPUS; ++i)
767 if (kvm->vcpus[i])
768 kvm->vcpus[i]->last_pte_updated = NULL;
769 }
770
771 static void kvm_mmu_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp)
772 {
773 u64 *parent_pte;
774
775 ++kvm->stat.mmu_shadow_zapped;
776 while (sp->multimapped || sp->parent_pte) {
777 if (!sp->multimapped)
778 parent_pte = sp->parent_pte;
779 else {
780 struct kvm_pte_chain *chain;
781
782 chain = container_of(sp->parent_ptes.first,
783 struct kvm_pte_chain, link);
784 parent_pte = chain->parent_ptes[0];
785 }
786 BUG_ON(!parent_pte);
787 kvm_mmu_put_page(sp, parent_pte);
788 set_shadow_pte(parent_pte, shadow_trap_nonpresent_pte);
789 }
790 kvm_mmu_page_unlink_children(kvm, sp);
791 if (!sp->root_count) {
792 hlist_del(&sp->hash_link);
793 kvm_mmu_free_page(kvm, sp);
794 } else
795 list_move(&sp->link, &kvm->active_mmu_pages);
796 kvm_mmu_reset_last_pte_updated(kvm);
797 }
798
799 /*
800 * Changing the number of mmu pages allocated to the vm
801 * Note: if kvm_nr_mmu_pages is too small, you will get dead lock
802 */
803 void kvm_mmu_change_mmu_pages(struct kvm *kvm, unsigned int kvm_nr_mmu_pages)
804 {
805 /*
806 * If we set the number of mmu pages to be smaller be than the
807 * number of actived pages , we must to free some mmu pages before we
808 * change the value
809 */
810
811 if ((kvm->n_alloc_mmu_pages - kvm->n_free_mmu_pages) >
812 kvm_nr_mmu_pages) {
813 int n_used_mmu_pages = kvm->n_alloc_mmu_pages
814 - kvm->n_free_mmu_pages;
815
816 while (n_used_mmu_pages > kvm_nr_mmu_pages) {
817 struct kvm_mmu_page *page;
818
819 page = container_of(kvm->active_mmu_pages.prev,
820 struct kvm_mmu_page, link);
821 kvm_mmu_zap_page(kvm, page);
822 n_used_mmu_pages--;
823 }
824 kvm->n_free_mmu_pages = 0;
825 }
826 else
827 kvm->n_free_mmu_pages += kvm_nr_mmu_pages
828 - kvm->n_alloc_mmu_pages;
829
830 kvm->n_alloc_mmu_pages = kvm_nr_mmu_pages;
831 }
832
833 static int kvm_mmu_unprotect_page(struct kvm *kvm, gfn_t gfn)
834 {
835 unsigned index;
836 struct hlist_head *bucket;
837 struct kvm_mmu_page *sp;
838 struct hlist_node *node, *n;
839 int r;
840
841 pgprintk("%s: looking for gfn %lx\n", __FUNCTION__, gfn);
842 r = 0;
843 index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES;
844 bucket = &kvm->mmu_page_hash[index];
845 hlist_for_each_entry_safe(sp, node, n, bucket, hash_link)
846 if (sp->gfn == gfn && !sp->role.metaphysical) {
847 pgprintk("%s: gfn %lx role %x\n", __FUNCTION__, gfn,
848 sp->role.word);
849 kvm_mmu_zap_page(kvm, sp);
850 r = 1;
851 }
852 return r;
853 }
854
855 static void mmu_unshadow(struct kvm *kvm, gfn_t gfn)
856 {
857 struct kvm_mmu_page *sp;
858
859 while ((sp = kvm_mmu_lookup_page(kvm, gfn)) != NULL) {
860 pgprintk("%s: zap %lx %x\n", __FUNCTION__, gfn, sp->role.word);
861 kvm_mmu_zap_page(kvm, sp);
862 }
863 }
864
865 static void page_header_update_slot(struct kvm *kvm, void *pte, gfn_t gfn)
866 {
867 int slot = memslot_id(kvm, gfn_to_memslot(kvm, gfn));
868 struct kvm_mmu_page *sp = page_header(__pa(pte));
869
870 __set_bit(slot, &sp->slot_bitmap);
871 }
872
873 struct page *gva_to_page(struct kvm_vcpu *vcpu, gva_t gva)
874 {
875 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, gva);
876
877 if (gpa == UNMAPPED_GVA)
878 return NULL;
879 return gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
880 }
881
882 static void nonpaging_new_cr3(struct kvm_vcpu *vcpu)
883 {
884 }
885
886 static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, struct page *page)
887 {
888 int level = PT32E_ROOT_LEVEL;
889 hpa_t table_addr = vcpu->mmu.root_hpa;
890
891 for (; ; level--) {
892 u32 index = PT64_INDEX(v, level);
893 u64 *table;
894 u64 pte;
895
896 ASSERT(VALID_PAGE(table_addr));
897 table = __va(table_addr);
898
899 if (level == 1) {
900 int was_rmapped;
901
902 pte = table[index];
903 was_rmapped = is_rmap_pte(pte);
904 if (is_shadow_present_pte(pte) && is_writeble_pte(pte)) {
905 kvm_release_page_clean(page);
906 return 0;
907 }
908 mark_page_dirty(vcpu->kvm, v >> PAGE_SHIFT);
909 page_header_update_slot(vcpu->kvm, table,
910 v >> PAGE_SHIFT);
911 table[index] = page_to_phys(page)
912 | PT_PRESENT_MASK | PT_WRITABLE_MASK
913 | PT_USER_MASK;
914 if (!was_rmapped)
915 rmap_add(vcpu, &table[index], v >> PAGE_SHIFT);
916 else
917 kvm_release_page_clean(page);
918
919 return 0;
920 }
921
922 if (table[index] == shadow_trap_nonpresent_pte) {
923 struct kvm_mmu_page *new_table;
924 gfn_t pseudo_gfn;
925
926 pseudo_gfn = (v & PT64_DIR_BASE_ADDR_MASK)
927 >> PAGE_SHIFT;
928 new_table = kvm_mmu_get_page(vcpu, pseudo_gfn,
929 v, level - 1,
930 1, ACC_ALL, &table[index]);
931 if (!new_table) {
932 pgprintk("nonpaging_map: ENOMEM\n");
933 kvm_release_page_clean(page);
934 return -ENOMEM;
935 }
936
937 table[index] = __pa(new_table->spt) | PT_PRESENT_MASK
938 | PT_WRITABLE_MASK | PT_USER_MASK;
939 }
940 table_addr = table[index] & PT64_BASE_ADDR_MASK;
941 }
942 }
943
944 static void nonpaging_prefetch_page(struct kvm_vcpu *vcpu,
945 struct kvm_mmu_page *sp)
946 {
947 int i;
948
949 for (i = 0; i < PT64_ENT_PER_PAGE; ++i)
950 sp->spt[i] = shadow_trap_nonpresent_pte;
951 }
952
953 static void mmu_free_roots(struct kvm_vcpu *vcpu)
954 {
955 int i;
956 struct kvm_mmu_page *sp;
957
958 if (!VALID_PAGE(vcpu->mmu.root_hpa))
959 return;
960 #ifdef CONFIG_X86_64
961 if (vcpu->mmu.shadow_root_level == PT64_ROOT_LEVEL) {
962 hpa_t root = vcpu->mmu.root_hpa;
963
964 sp = page_header(root);
965 --sp->root_count;
966 vcpu->mmu.root_hpa = INVALID_PAGE;
967 return;
968 }
969 #endif
970 for (i = 0; i < 4; ++i) {
971 hpa_t root = vcpu->mmu.pae_root[i];
972
973 if (root) {
974 root &= PT64_BASE_ADDR_MASK;
975 sp = page_header(root);
976 --sp->root_count;
977 }
978 vcpu->mmu.pae_root[i] = INVALID_PAGE;
979 }
980 vcpu->mmu.root_hpa = INVALID_PAGE;
981 }
982
983 static void mmu_alloc_roots(struct kvm_vcpu *vcpu)
984 {
985 int i;
986 gfn_t root_gfn;
987 struct kvm_mmu_page *sp;
988
989 root_gfn = vcpu->cr3 >> PAGE_SHIFT;
990
991 #ifdef CONFIG_X86_64
992 if (vcpu->mmu.shadow_root_level == PT64_ROOT_LEVEL) {
993 hpa_t root = vcpu->mmu.root_hpa;
994
995 ASSERT(!VALID_PAGE(root));
996 sp = kvm_mmu_get_page(vcpu, root_gfn, 0,
997 PT64_ROOT_LEVEL, 0, ACC_ALL, NULL);
998 root = __pa(sp->spt);
999 ++sp->root_count;
1000 vcpu->mmu.root_hpa = root;
1001 return;
1002 }
1003 #endif
1004 for (i = 0; i < 4; ++i) {
1005 hpa_t root = vcpu->mmu.pae_root[i];
1006
1007 ASSERT(!VALID_PAGE(root));
1008 if (vcpu->mmu.root_level == PT32E_ROOT_LEVEL) {
1009 if (!is_present_pte(vcpu->pdptrs[i])) {
1010 vcpu->mmu.pae_root[i] = 0;
1011 continue;
1012 }
1013 root_gfn = vcpu->pdptrs[i] >> PAGE_SHIFT;
1014 } else if (vcpu->mmu.root_level == 0)
1015 root_gfn = 0;
1016 sp = kvm_mmu_get_page(vcpu, root_gfn, i << 30,
1017 PT32_ROOT_LEVEL, !is_paging(vcpu),
1018 ACC_ALL, NULL);
1019 root = __pa(sp->spt);
1020 ++sp->root_count;
1021 vcpu->mmu.pae_root[i] = root | PT_PRESENT_MASK;
1022 }
1023 vcpu->mmu.root_hpa = __pa(vcpu->mmu.pae_root);
1024 }
1025
1026 static gpa_t nonpaging_gva_to_gpa(struct kvm_vcpu *vcpu, gva_t vaddr)
1027 {
1028 return vaddr;
1029 }
1030
1031 static int nonpaging_page_fault(struct kvm_vcpu *vcpu, gva_t gva,
1032 u32 error_code)
1033 {
1034 struct page *page;
1035 int r;
1036
1037 r = mmu_topup_memory_caches(vcpu);
1038 if (r)
1039 return r;
1040
1041 ASSERT(vcpu);
1042 ASSERT(VALID_PAGE(vcpu->mmu.root_hpa));
1043
1044 page = gfn_to_page(vcpu->kvm, gva >> PAGE_SHIFT);
1045
1046 if (is_error_page(page)) {
1047 kvm_release_page_clean(page);
1048 return 1;
1049 }
1050
1051 return nonpaging_map(vcpu, gva & PAGE_MASK, page);
1052 }
1053
1054 static void nonpaging_free(struct kvm_vcpu *vcpu)
1055 {
1056 mmu_free_roots(vcpu);
1057 }
1058
1059 static int nonpaging_init_context(struct kvm_vcpu *vcpu)
1060 {
1061 struct kvm_mmu *context = &vcpu->mmu;
1062
1063 context->new_cr3 = nonpaging_new_cr3;
1064 context->page_fault = nonpaging_page_fault;
1065 context->gva_to_gpa = nonpaging_gva_to_gpa;
1066 context->free = nonpaging_free;
1067 context->prefetch_page = nonpaging_prefetch_page;
1068 context->root_level = 0;
1069 context->shadow_root_level = PT32E_ROOT_LEVEL;
1070 context->root_hpa = INVALID_PAGE;
1071 return 0;
1072 }
1073
1074 void kvm_mmu_flush_tlb(struct kvm_vcpu *vcpu)
1075 {
1076 ++vcpu->stat.tlb_flush;
1077 kvm_x86_ops->tlb_flush(vcpu);
1078 }
1079
1080 static void paging_new_cr3(struct kvm_vcpu *vcpu)
1081 {
1082 pgprintk("%s: cr3 %lx\n", __FUNCTION__, vcpu->cr3);
1083 mmu_free_roots(vcpu);
1084 }
1085
1086 static void inject_page_fault(struct kvm_vcpu *vcpu,
1087 u64 addr,
1088 u32 err_code)
1089 {
1090 kvm_inject_page_fault(vcpu, addr, err_code);
1091 }
1092
1093 static void paging_free(struct kvm_vcpu *vcpu)
1094 {
1095 nonpaging_free(vcpu);
1096 }
1097
1098 #define PTTYPE 64
1099 #include "paging_tmpl.h"
1100 #undef PTTYPE
1101
1102 #define PTTYPE 32
1103 #include "paging_tmpl.h"
1104 #undef PTTYPE
1105
1106 static int paging64_init_context_common(struct kvm_vcpu *vcpu, int level)
1107 {
1108 struct kvm_mmu *context = &vcpu->mmu;
1109
1110 ASSERT(is_pae(vcpu));
1111 context->new_cr3 = paging_new_cr3;
1112 context->page_fault = paging64_page_fault;
1113 context->gva_to_gpa = paging64_gva_to_gpa;
1114 context->prefetch_page = paging64_prefetch_page;
1115 context->free = paging_free;
1116 context->root_level = level;
1117 context->shadow_root_level = level;
1118 context->root_hpa = INVALID_PAGE;
1119 return 0;
1120 }
1121
1122 static int paging64_init_context(struct kvm_vcpu *vcpu)
1123 {
1124 return paging64_init_context_common(vcpu, PT64_ROOT_LEVEL);
1125 }
1126
1127 static int paging32_init_context(struct kvm_vcpu *vcpu)
1128 {
1129 struct kvm_mmu *context = &vcpu->mmu;
1130
1131 context->new_cr3 = paging_new_cr3;
1132 context->page_fault = paging32_page_fault;
1133 context->gva_to_gpa = paging32_gva_to_gpa;
1134 context->free = paging_free;
1135 context->prefetch_page = paging32_prefetch_page;
1136 context->root_level = PT32_ROOT_LEVEL;
1137 context->shadow_root_level = PT32E_ROOT_LEVEL;
1138 context->root_hpa = INVALID_PAGE;
1139 return 0;
1140 }
1141
1142 static int paging32E_init_context(struct kvm_vcpu *vcpu)
1143 {
1144 return paging64_init_context_common(vcpu, PT32E_ROOT_LEVEL);
1145 }
1146
1147 static int init_kvm_mmu(struct kvm_vcpu *vcpu)
1148 {
1149 ASSERT(vcpu);
1150 ASSERT(!VALID_PAGE(vcpu->mmu.root_hpa));
1151
1152 if (!is_paging(vcpu))
1153 return nonpaging_init_context(vcpu);
1154 else if (is_long_mode(vcpu))
1155 return paging64_init_context(vcpu);
1156 else if (is_pae(vcpu))
1157 return paging32E_init_context(vcpu);
1158 else
1159 return paging32_init_context(vcpu);
1160 }
1161
1162 static void destroy_kvm_mmu(struct kvm_vcpu *vcpu)
1163 {
1164 ASSERT(vcpu);
1165 if (VALID_PAGE(vcpu->mmu.root_hpa)) {
1166 vcpu->mmu.free(vcpu);
1167 vcpu->mmu.root_hpa = INVALID_PAGE;
1168 }
1169 }
1170
1171 int kvm_mmu_reset_context(struct kvm_vcpu *vcpu)
1172 {
1173 destroy_kvm_mmu(vcpu);
1174 return init_kvm_mmu(vcpu);
1175 }
1176 EXPORT_SYMBOL_GPL(kvm_mmu_reset_context);
1177
1178 int kvm_mmu_load(struct kvm_vcpu *vcpu)
1179 {
1180 int r;
1181
1182 mutex_lock(&vcpu->kvm->lock);
1183 r = mmu_topup_memory_caches(vcpu);
1184 if (r)
1185 goto out;
1186 mmu_alloc_roots(vcpu);
1187 kvm_x86_ops->set_cr3(vcpu, vcpu->mmu.root_hpa);
1188 kvm_mmu_flush_tlb(vcpu);
1189 out:
1190 mutex_unlock(&vcpu->kvm->lock);
1191 return r;
1192 }
1193 EXPORT_SYMBOL_GPL(kvm_mmu_load);
1194
1195 void kvm_mmu_unload(struct kvm_vcpu *vcpu)
1196 {
1197 mmu_free_roots(vcpu);
1198 }
1199
1200 static void mmu_pte_write_zap_pte(struct kvm_vcpu *vcpu,
1201 struct kvm_mmu_page *sp,
1202 u64 *spte)
1203 {
1204 u64 pte;
1205 struct kvm_mmu_page *child;
1206
1207 pte = *spte;
1208 if (is_shadow_present_pte(pte)) {
1209 if (sp->role.level == PT_PAGE_TABLE_LEVEL)
1210 rmap_remove(vcpu->kvm, spte);
1211 else {
1212 child = page_header(pte & PT64_BASE_ADDR_MASK);
1213 mmu_page_remove_parent_pte(child, spte);
1214 }
1215 }
1216 set_shadow_pte(spte, shadow_trap_nonpresent_pte);
1217 }
1218
1219 static void mmu_pte_write_new_pte(struct kvm_vcpu *vcpu,
1220 struct kvm_mmu_page *sp,
1221 u64 *spte,
1222 const void *new, int bytes,
1223 int offset_in_pte)
1224 {
1225 if (sp->role.level != PT_PAGE_TABLE_LEVEL) {
1226 ++vcpu->kvm->stat.mmu_pde_zapped;
1227 return;
1228 }
1229
1230 ++vcpu->kvm->stat.mmu_pte_updated;
1231 if (sp->role.glevels == PT32_ROOT_LEVEL)
1232 paging32_update_pte(vcpu, sp, spte, new, bytes, offset_in_pte);
1233 else
1234 paging64_update_pte(vcpu, sp, spte, new, bytes, offset_in_pte);
1235 }
1236
1237 static bool need_remote_flush(u64 old, u64 new)
1238 {
1239 if (!is_shadow_present_pte(old))
1240 return false;
1241 if (!is_shadow_present_pte(new))
1242 return true;
1243 if ((old ^ new) & PT64_BASE_ADDR_MASK)
1244 return true;
1245 old ^= PT64_NX_MASK;
1246 new ^= PT64_NX_MASK;
1247 return (old & ~new & PT64_PERM_MASK) != 0;
1248 }
1249
1250 static void mmu_pte_write_flush_tlb(struct kvm_vcpu *vcpu, u64 old, u64 new)
1251 {
1252 if (need_remote_flush(old, new))
1253 kvm_flush_remote_tlbs(vcpu->kvm);
1254 else
1255 kvm_mmu_flush_tlb(vcpu);
1256 }
1257
1258 static bool last_updated_pte_accessed(struct kvm_vcpu *vcpu)
1259 {
1260 u64 *spte = vcpu->last_pte_updated;
1261
1262 return !!(spte && (*spte & PT_ACCESSED_MASK));
1263 }
1264
1265 void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
1266 const u8 *new, int bytes)
1267 {
1268 gfn_t gfn = gpa >> PAGE_SHIFT;
1269 struct kvm_mmu_page *sp;
1270 struct hlist_node *node, *n;
1271 struct hlist_head *bucket;
1272 unsigned index;
1273 u64 entry;
1274 u64 *spte;
1275 unsigned offset = offset_in_page(gpa);
1276 unsigned pte_size;
1277 unsigned page_offset;
1278 unsigned misaligned;
1279 unsigned quadrant;
1280 int level;
1281 int flooded = 0;
1282 int npte;
1283
1284 pgprintk("%s: gpa %llx bytes %d\n", __FUNCTION__, gpa, bytes);
1285 ++vcpu->kvm->stat.mmu_pte_write;
1286 kvm_mmu_audit(vcpu, "pre pte write");
1287 if (gfn == vcpu->last_pt_write_gfn
1288 && !last_updated_pte_accessed(vcpu)) {
1289 ++vcpu->last_pt_write_count;
1290 if (vcpu->last_pt_write_count >= 3)
1291 flooded = 1;
1292 } else {
1293 vcpu->last_pt_write_gfn = gfn;
1294 vcpu->last_pt_write_count = 1;
1295 vcpu->last_pte_updated = NULL;
1296 }
1297 index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES;
1298 bucket = &vcpu->kvm->mmu_page_hash[index];
1299 hlist_for_each_entry_safe(sp, node, n, bucket, hash_link) {
1300 if (sp->gfn != gfn || sp->role.metaphysical)
1301 continue;
1302 pte_size = sp->role.glevels == PT32_ROOT_LEVEL ? 4 : 8;
1303 misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1);
1304 misaligned |= bytes < 4;
1305 if (misaligned || flooded) {
1306 /*
1307 * Misaligned accesses are too much trouble to fix
1308 * up; also, they usually indicate a page is not used
1309 * as a page table.
1310 *
1311 * If we're seeing too many writes to a page,
1312 * it may no longer be a page table, or we may be
1313 * forking, in which case it is better to unmap the
1314 * page.
1315 */
1316 pgprintk("misaligned: gpa %llx bytes %d role %x\n",
1317 gpa, bytes, sp->role.word);
1318 kvm_mmu_zap_page(vcpu->kvm, sp);
1319 ++vcpu->kvm->stat.mmu_flooded;
1320 continue;
1321 }
1322 page_offset = offset;
1323 level = sp->role.level;
1324 npte = 1;
1325 if (sp->role.glevels == PT32_ROOT_LEVEL) {
1326 page_offset <<= 1; /* 32->64 */
1327 /*
1328 * A 32-bit pde maps 4MB while the shadow pdes map
1329 * only 2MB. So we need to double the offset again
1330 * and zap two pdes instead of one.
1331 */
1332 if (level == PT32_ROOT_LEVEL) {
1333 page_offset &= ~7; /* kill rounding error */
1334 page_offset <<= 1;
1335 npte = 2;
1336 }
1337 quadrant = page_offset >> PAGE_SHIFT;
1338 page_offset &= ~PAGE_MASK;
1339 if (quadrant != sp->role.quadrant)
1340 continue;
1341 }
1342 spte = &sp->spt[page_offset / sizeof(*spte)];
1343 while (npte--) {
1344 entry = *spte;
1345 mmu_pte_write_zap_pte(vcpu, sp, spte);
1346 mmu_pte_write_new_pte(vcpu, sp, spte, new, bytes,
1347 page_offset & (pte_size - 1));
1348 mmu_pte_write_flush_tlb(vcpu, entry, *spte);
1349 ++spte;
1350 }
1351 }
1352 kvm_mmu_audit(vcpu, "post pte write");
1353 }
1354
1355 int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva)
1356 {
1357 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, gva);
1358
1359 return kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1360 }
1361
1362 void __kvm_mmu_free_some_pages(struct kvm_vcpu *vcpu)
1363 {
1364 while (vcpu->kvm->n_free_mmu_pages < KVM_REFILL_PAGES) {
1365 struct kvm_mmu_page *sp;
1366
1367 sp = container_of(vcpu->kvm->active_mmu_pages.prev,
1368 struct kvm_mmu_page, link);
1369 kvm_mmu_zap_page(vcpu->kvm, sp);
1370 ++vcpu->kvm->stat.mmu_recycled;
1371 }
1372 }
1373
1374 int kvm_mmu_page_fault(struct kvm_vcpu *vcpu, gva_t cr2, u32 error_code)
1375 {
1376 int r;
1377 enum emulation_result er;
1378
1379 mutex_lock(&vcpu->kvm->lock);
1380 r = vcpu->mmu.page_fault(vcpu, cr2, error_code);
1381 if (r < 0)
1382 goto out;
1383
1384 if (!r) {
1385 r = 1;
1386 goto out;
1387 }
1388
1389 r = mmu_topup_memory_caches(vcpu);
1390 if (r)
1391 goto out;
1392
1393 er = emulate_instruction(vcpu, vcpu->run, cr2, error_code, 0);
1394 mutex_unlock(&vcpu->kvm->lock);
1395
1396 switch (er) {
1397 case EMULATE_DONE:
1398 return 1;
1399 case EMULATE_DO_MMIO:
1400 ++vcpu->stat.mmio_exits;
1401 return 0;
1402 case EMULATE_FAIL:
1403 kvm_report_emulation_failure(vcpu, "pagetable");
1404 return 1;
1405 default:
1406 BUG();
1407 }
1408 out:
1409 mutex_unlock(&vcpu->kvm->lock);
1410 return r;
1411 }
1412 EXPORT_SYMBOL_GPL(kvm_mmu_page_fault);
1413
1414 static void free_mmu_pages(struct kvm_vcpu *vcpu)
1415 {
1416 struct kvm_mmu_page *sp;
1417
1418 while (!list_empty(&vcpu->kvm->active_mmu_pages)) {
1419 sp = container_of(vcpu->kvm->active_mmu_pages.next,
1420 struct kvm_mmu_page, link);
1421 kvm_mmu_zap_page(vcpu->kvm, sp);
1422 }
1423 free_page((unsigned long)vcpu->mmu.pae_root);
1424 }
1425
1426 static int alloc_mmu_pages(struct kvm_vcpu *vcpu)
1427 {
1428 struct page *page;
1429 int i;
1430
1431 ASSERT(vcpu);
1432
1433 if (vcpu->kvm->n_requested_mmu_pages)
1434 vcpu->kvm->n_free_mmu_pages = vcpu->kvm->n_requested_mmu_pages;
1435 else
1436 vcpu->kvm->n_free_mmu_pages = vcpu->kvm->n_alloc_mmu_pages;
1437 /*
1438 * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64.
1439 * Therefore we need to allocate shadow page tables in the first
1440 * 4GB of memory, which happens to fit the DMA32 zone.
1441 */
1442 page = alloc_page(GFP_KERNEL | __GFP_DMA32);
1443 if (!page)
1444 goto error_1;
1445 vcpu->mmu.pae_root = page_address(page);
1446 for (i = 0; i < 4; ++i)
1447 vcpu->mmu.pae_root[i] = INVALID_PAGE;
1448
1449 return 0;
1450
1451 error_1:
1452 free_mmu_pages(vcpu);
1453 return -ENOMEM;
1454 }
1455
1456 int kvm_mmu_create(struct kvm_vcpu *vcpu)
1457 {
1458 ASSERT(vcpu);
1459 ASSERT(!VALID_PAGE(vcpu->mmu.root_hpa));
1460
1461 return alloc_mmu_pages(vcpu);
1462 }
1463
1464 int kvm_mmu_setup(struct kvm_vcpu *vcpu)
1465 {
1466 ASSERT(vcpu);
1467 ASSERT(!VALID_PAGE(vcpu->mmu.root_hpa));
1468
1469 return init_kvm_mmu(vcpu);
1470 }
1471
1472 void kvm_mmu_destroy(struct kvm_vcpu *vcpu)
1473 {
1474 ASSERT(vcpu);
1475
1476 destroy_kvm_mmu(vcpu);
1477 free_mmu_pages(vcpu);
1478 mmu_free_memory_caches(vcpu);
1479 }
1480
1481 void kvm_mmu_slot_remove_write_access(struct kvm *kvm, int slot)
1482 {
1483 struct kvm_mmu_page *sp;
1484
1485 list_for_each_entry(sp, &kvm->active_mmu_pages, link) {
1486 int i;
1487 u64 *pt;
1488
1489 if (!test_bit(slot, &sp->slot_bitmap))
1490 continue;
1491
1492 pt = sp->spt;
1493 for (i = 0; i < PT64_ENT_PER_PAGE; ++i)
1494 /* avoid RMW */
1495 if (pt[i] & PT_WRITABLE_MASK)
1496 pt[i] &= ~PT_WRITABLE_MASK;
1497 }
1498 }
1499
1500 void kvm_mmu_zap_all(struct kvm *kvm)
1501 {
1502 struct kvm_mmu_page *sp, *node;
1503
1504 list_for_each_entry_safe(sp, node, &kvm->active_mmu_pages, link)
1505 kvm_mmu_zap_page(kvm, sp);
1506
1507 kvm_flush_remote_tlbs(kvm);
1508 }
1509
1510 void kvm_mmu_module_exit(void)
1511 {
1512 if (pte_chain_cache)
1513 kmem_cache_destroy(pte_chain_cache);
1514 if (rmap_desc_cache)
1515 kmem_cache_destroy(rmap_desc_cache);
1516 if (mmu_page_header_cache)
1517 kmem_cache_destroy(mmu_page_header_cache);
1518 }
1519
1520 int kvm_mmu_module_init(void)
1521 {
1522 pte_chain_cache = kmem_cache_create("kvm_pte_chain",
1523 sizeof(struct kvm_pte_chain),
1524 0, 0, NULL);
1525 if (!pte_chain_cache)
1526 goto nomem;
1527 rmap_desc_cache = kmem_cache_create("kvm_rmap_desc",
1528 sizeof(struct kvm_rmap_desc),
1529 0, 0, NULL);
1530 if (!rmap_desc_cache)
1531 goto nomem;
1532
1533 mmu_page_header_cache = kmem_cache_create("kvm_mmu_page_header",
1534 sizeof(struct kvm_mmu_page),
1535 0, 0, NULL);
1536 if (!mmu_page_header_cache)
1537 goto nomem;
1538
1539 return 0;
1540
1541 nomem:
1542 kvm_mmu_module_exit();
1543 return -ENOMEM;
1544 }
1545
1546 /*
1547 * Caculate mmu pages needed for kvm.
1548 */
1549 unsigned int kvm_mmu_calculate_mmu_pages(struct kvm *kvm)
1550 {
1551 int i;
1552 unsigned int nr_mmu_pages;
1553 unsigned int nr_pages = 0;
1554
1555 for (i = 0; i < kvm->nmemslots; i++)
1556 nr_pages += kvm->memslots[i].npages;
1557
1558 nr_mmu_pages = nr_pages * KVM_PERMILLE_MMU_PAGES / 1000;
1559 nr_mmu_pages = max(nr_mmu_pages,
1560 (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
1561
1562 return nr_mmu_pages;
1563 }
1564
1565 #ifdef AUDIT
1566
1567 static const char *audit_msg;
1568
1569 static gva_t canonicalize(gva_t gva)
1570 {
1571 #ifdef CONFIG_X86_64
1572 gva = (long long)(gva << 16) >> 16;
1573 #endif
1574 return gva;
1575 }
1576
1577 static void audit_mappings_page(struct kvm_vcpu *vcpu, u64 page_pte,
1578 gva_t va, int level)
1579 {
1580 u64 *pt = __va(page_pte & PT64_BASE_ADDR_MASK);
1581 int i;
1582 gva_t va_delta = 1ul << (PAGE_SHIFT + 9 * (level - 1));
1583
1584 for (i = 0; i < PT64_ENT_PER_PAGE; ++i, va += va_delta) {
1585 u64 ent = pt[i];
1586
1587 if (ent == shadow_trap_nonpresent_pte)
1588 continue;
1589
1590 va = canonicalize(va);
1591 if (level > 1) {
1592 if (ent == shadow_notrap_nonpresent_pte)
1593 printk(KERN_ERR "audit: (%s) nontrapping pte"
1594 " in nonleaf level: levels %d gva %lx"
1595 " level %d pte %llx\n", audit_msg,
1596 vcpu->mmu.root_level, va, level, ent);
1597
1598 audit_mappings_page(vcpu, ent, va, level - 1);
1599 } else {
1600 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, va);
1601 struct page *page = gpa_to_page(vcpu, gpa);
1602 hpa_t hpa = page_to_phys(page);
1603
1604 if (is_shadow_present_pte(ent)
1605 && (ent & PT64_BASE_ADDR_MASK) != hpa)
1606 printk(KERN_ERR "xx audit error: (%s) levels %d"
1607 " gva %lx gpa %llx hpa %llx ent %llx %d\n",
1608 audit_msg, vcpu->mmu.root_level,
1609 va, gpa, hpa, ent,
1610 is_shadow_present_pte(ent));
1611 else if (ent == shadow_notrap_nonpresent_pte
1612 && !is_error_hpa(hpa))
1613 printk(KERN_ERR "audit: (%s) notrap shadow,"
1614 " valid guest gva %lx\n", audit_msg, va);
1615 kvm_release_page_clean(page);
1616
1617 }
1618 }
1619 }
1620
1621 static void audit_mappings(struct kvm_vcpu *vcpu)
1622 {
1623 unsigned i;
1624
1625 if (vcpu->mmu.root_level == 4)
1626 audit_mappings_page(vcpu, vcpu->mmu.root_hpa, 0, 4);
1627 else
1628 for (i = 0; i < 4; ++i)
1629 if (vcpu->mmu.pae_root[i] & PT_PRESENT_MASK)
1630 audit_mappings_page(vcpu,
1631 vcpu->mmu.pae_root[i],
1632 i << 30,
1633 2);
1634 }
1635
1636 static int count_rmaps(struct kvm_vcpu *vcpu)
1637 {
1638 int nmaps = 0;
1639 int i, j, k;
1640
1641 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1642 struct kvm_memory_slot *m = &vcpu->kvm->memslots[i];
1643 struct kvm_rmap_desc *d;
1644
1645 for (j = 0; j < m->npages; ++j) {
1646 unsigned long *rmapp = &m->rmap[j];
1647
1648 if (!*rmapp)
1649 continue;
1650 if (!(*rmapp & 1)) {
1651 ++nmaps;
1652 continue;
1653 }
1654 d = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
1655 while (d) {
1656 for (k = 0; k < RMAP_EXT; ++k)
1657 if (d->shadow_ptes[k])
1658 ++nmaps;
1659 else
1660 break;
1661 d = d->more;
1662 }
1663 }
1664 }
1665 return nmaps;
1666 }
1667
1668 static int count_writable_mappings(struct kvm_vcpu *vcpu)
1669 {
1670 int nmaps = 0;
1671 struct kvm_mmu_page *sp;
1672 int i;
1673
1674 list_for_each_entry(sp, &vcpu->kvm->active_mmu_pages, link) {
1675 u64 *pt = sp->spt;
1676
1677 if (sp->role.level != PT_PAGE_TABLE_LEVEL)
1678 continue;
1679
1680 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
1681 u64 ent = pt[i];
1682
1683 if (!(ent & PT_PRESENT_MASK))
1684 continue;
1685 if (!(ent & PT_WRITABLE_MASK))
1686 continue;
1687 ++nmaps;
1688 }
1689 }
1690 return nmaps;
1691 }
1692
1693 static void audit_rmap(struct kvm_vcpu *vcpu)
1694 {
1695 int n_rmap = count_rmaps(vcpu);
1696 int n_actual = count_writable_mappings(vcpu);
1697
1698 if (n_rmap != n_actual)
1699 printk(KERN_ERR "%s: (%s) rmap %d actual %d\n",
1700 __FUNCTION__, audit_msg, n_rmap, n_actual);
1701 }
1702
1703 static void audit_write_protection(struct kvm_vcpu *vcpu)
1704 {
1705 struct kvm_mmu_page *sp;
1706 struct kvm_memory_slot *slot;
1707 unsigned long *rmapp;
1708 gfn_t gfn;
1709
1710 list_for_each_entry(sp, &vcpu->kvm->active_mmu_pages, link) {
1711 if (sp->role.metaphysical)
1712 continue;
1713
1714 slot = gfn_to_memslot(vcpu->kvm, sp->gfn);
1715 gfn = unalias_gfn(vcpu->kvm, sp->gfn);
1716 rmapp = &slot->rmap[gfn - slot->base_gfn];
1717 if (*rmapp)
1718 printk(KERN_ERR "%s: (%s) shadow page has writable"
1719 " mappings: gfn %lx role %x\n",
1720 __FUNCTION__, audit_msg, sp->gfn,
1721 sp->role.word);
1722 }
1723 }
1724
1725 static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg)
1726 {
1727 int olddbg = dbg;
1728
1729 dbg = 0;
1730 audit_msg = msg;
1731 audit_rmap(vcpu);
1732 audit_write_protection(vcpu);
1733 audit_mappings(vcpu);
1734 dbg = olddbg;
1735 }
1736
1737 #endif
Linux kernel - Deliverables
This page took 0.090954 seconds and 6 git commands to generate.