projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
KVM: Generalize exception injection mechanism
[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_MASK (1ULL << 63)
89
90 #define PT_PAT_SHIFT 7
91 #define PT_DIR_PAT_SHIFT 12
92 #define PT_DIR_PAT_MASK (1ULL << PT_DIR_PAT_SHIFT)
93
94 #define PT32_DIR_PSE36_SIZE 4
95 #define PT32_DIR_PSE36_SHIFT 13
96 #define PT32_DIR_PSE36_MASK \
97 (((1ULL << PT32_DIR_PSE36_SIZE) - 1) << PT32_DIR_PSE36_SHIFT)
98
99
100 #define PT_FIRST_AVAIL_BITS_SHIFT 9
101 #define PT64_SECOND_AVAIL_BITS_SHIFT 52
102
103 #define PT_SHADOW_IO_MARK (1ULL << PT_FIRST_AVAIL_BITS_SHIFT)
104
105 #define VALID_PAGE(x) ((x) != INVALID_PAGE)
106
107 #define PT64_LEVEL_BITS 9
108
109 #define PT64_LEVEL_SHIFT(level) \
110 (PAGE_SHIFT + (level - 1) * PT64_LEVEL_BITS)
111
112 #define PT64_LEVEL_MASK(level) \
113 (((1ULL << PT64_LEVEL_BITS) - 1) << PT64_LEVEL_SHIFT(level))
114
115 #define PT64_INDEX(address, level)\
116 (((address) >> PT64_LEVEL_SHIFT(level)) & ((1 << PT64_LEVEL_BITS) - 1))
117
118
119 #define PT32_LEVEL_BITS 10
120
121 #define PT32_LEVEL_SHIFT(level) \
122 (PAGE_SHIFT + (level - 1) * PT32_LEVEL_BITS)
123
124 #define PT32_LEVEL_MASK(level) \
125 (((1ULL << PT32_LEVEL_BITS) - 1) << PT32_LEVEL_SHIFT(level))
126
127 #define PT32_INDEX(address, level)\
128 (((address) >> PT32_LEVEL_SHIFT(level)) & ((1 << PT32_LEVEL_BITS) - 1))
129
130
131 #define PT64_BASE_ADDR_MASK (((1ULL << 52) - 1) & ~(u64)(PAGE_SIZE-1))
132 #define PT64_DIR_BASE_ADDR_MASK \
133 (PT64_BASE_ADDR_MASK & ~((1ULL << (PAGE_SHIFT + PT64_LEVEL_BITS)) - 1))
134
135 #define PT32_BASE_ADDR_MASK PAGE_MASK
136 #define PT32_DIR_BASE_ADDR_MASK \
137 (PAGE_MASK & ~((1ULL << (PAGE_SHIFT + PT32_LEVEL_BITS)) - 1))
138
139 #define PT64_PERM_MASK (PT_PRESENT_MASK | PT_WRITABLE_MASK | PT_USER_MASK \
140 | PT64_NX_MASK)
141
142 #define PFERR_PRESENT_MASK (1U << 0)
143 #define PFERR_WRITE_MASK (1U << 1)
144 #define PFERR_USER_MASK (1U << 2)
145 #define PFERR_FETCH_MASK (1U << 4)
146
147 #define PT64_ROOT_LEVEL 4
148 #define PT32_ROOT_LEVEL 2
149 #define PT32E_ROOT_LEVEL 3
150
151 #define PT_DIRECTORY_LEVEL 2
152 #define PT_PAGE_TABLE_LEVEL 1
153
154 #define RMAP_EXT 4
155
156 struct kvm_rmap_desc {
157 u64 *shadow_ptes[RMAP_EXT];
158 struct kvm_rmap_desc *more;
159 };
160
161 static struct kmem_cache *pte_chain_cache;
162 static struct kmem_cache *rmap_desc_cache;
163 static struct kmem_cache *mmu_page_header_cache;
164
165 static u64 __read_mostly shadow_trap_nonpresent_pte;
166 static u64 __read_mostly shadow_notrap_nonpresent_pte;
167
168 void kvm_mmu_set_nonpresent_ptes(u64 trap_pte, u64 notrap_pte)
169 {
170 shadow_trap_nonpresent_pte = trap_pte;
171 shadow_notrap_nonpresent_pte = notrap_pte;
172 }
173 EXPORT_SYMBOL_GPL(kvm_mmu_set_nonpresent_ptes);
174
175 static int is_write_protection(struct kvm_vcpu *vcpu)
176 {
177 return vcpu->cr0 & X86_CR0_WP;
178 }
179
180 static int is_cpuid_PSE36(void)
181 {
182 return 1;
183 }
184
185 static int is_nx(struct kvm_vcpu *vcpu)
186 {
187 return vcpu->shadow_efer & EFER_NX;
188 }
189
190 static int is_present_pte(unsigned long pte)
191 {
192 return pte & PT_PRESENT_MASK;
193 }
194
195 static int is_shadow_present_pte(u64 pte)
196 {
197 pte &= ~PT_SHADOW_IO_MARK;
198 return pte != shadow_trap_nonpresent_pte
199 && pte != shadow_notrap_nonpresent_pte;
200 }
201
202 static int is_writeble_pte(unsigned long pte)
203 {
204 return pte & PT_WRITABLE_MASK;
205 }
206
207 static int is_dirty_pte(unsigned long pte)
208 {
209 return pte & PT_DIRTY_MASK;
210 }
211
212 static int is_io_pte(unsigned long pte)
213 {
214 return pte & PT_SHADOW_IO_MARK;
215 }
216
217 static int is_rmap_pte(u64 pte)
218 {
219 return pte != shadow_trap_nonpresent_pte
220 && pte != shadow_notrap_nonpresent_pte;
221 }
222
223 static gfn_t pse36_gfn_delta(u32 gpte)
224 {
225 int shift = 32 - PT32_DIR_PSE36_SHIFT - PAGE_SHIFT;
226
227 return (gpte & PT32_DIR_PSE36_MASK) << shift;
228 }
229
230 static void set_shadow_pte(u64 *sptep, u64 spte)
231 {
232 #ifdef CONFIG_X86_64
233 set_64bit((unsigned long *)sptep, spte);
234 #else
235 set_64bit((unsigned long long *)sptep, spte);
236 #endif
237 }
238
239 static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache,
240 struct kmem_cache *base_cache, int min)
241 {
242 void *obj;
243
244 if (cache->nobjs >= min)
245 return 0;
246 while (cache->nobjs < ARRAY_SIZE(cache->objects)) {
247 obj = kmem_cache_zalloc(base_cache, GFP_KERNEL);
248 if (!obj)
249 return -ENOMEM;
250 cache->objects[cache->nobjs++] = obj;
251 }
252 return 0;
253 }
254
255 static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
256 {
257 while (mc->nobjs)
258 kfree(mc->objects[--mc->nobjs]);
259 }
260
261 static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache *cache,
262 int min)
263 {
264 struct page *page;
265
266 if (cache->nobjs >= min)
267 return 0;
268 while (cache->nobjs < ARRAY_SIZE(cache->objects)) {
269 page = alloc_page(GFP_KERNEL);
270 if (!page)
271 return -ENOMEM;
272 set_page_private(page, 0);
273 cache->objects[cache->nobjs++] = page_address(page);
274 }
275 return 0;
276 }
277
278 static void mmu_free_memory_cache_page(struct kvm_mmu_memory_cache *mc)
279 {
280 while (mc->nobjs)
281 free_page((unsigned long)mc->objects[--mc->nobjs]);
282 }
283
284 static int mmu_topup_memory_caches(struct kvm_vcpu *vcpu)
285 {
286 int r;
287
288 kvm_mmu_free_some_pages(vcpu);
289 r = mmu_topup_memory_cache(&vcpu->mmu_pte_chain_cache,
290 pte_chain_cache, 4);
291 if (r)
292 goto out;
293 r = mmu_topup_memory_cache(&vcpu->mmu_rmap_desc_cache,
294 rmap_desc_cache, 1);
295 if (r)
296 goto out;
297 r = mmu_topup_memory_cache_page(&vcpu->mmu_page_cache, 8);
298 if (r)
299 goto out;
300 r = mmu_topup_memory_cache(&vcpu->mmu_page_header_cache,
301 mmu_page_header_cache, 4);
302 out:
303 return r;
304 }
305
306 static void mmu_free_memory_caches(struct kvm_vcpu *vcpu)
307 {
308 mmu_free_memory_cache(&vcpu->mmu_pte_chain_cache);
309 mmu_free_memory_cache(&vcpu->mmu_rmap_desc_cache);
310 mmu_free_memory_cache_page(&vcpu->mmu_page_cache);
311 mmu_free_memory_cache(&vcpu->mmu_page_header_cache);
312 }
313
314 static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc,
315 size_t size)
316 {
317 void *p;
318
319 BUG_ON(!mc->nobjs);
320 p = mc->objects[--mc->nobjs];
321 memset(p, 0, size);
322 return p;
323 }
324
325 static struct kvm_pte_chain *mmu_alloc_pte_chain(struct kvm_vcpu *vcpu)
326 {
327 return mmu_memory_cache_alloc(&vcpu->mmu_pte_chain_cache,
328 sizeof(struct kvm_pte_chain));
329 }
330
331 static void mmu_free_pte_chain(struct kvm_pte_chain *pc)
332 {
333 kfree(pc);
334 }
335
336 static struct kvm_rmap_desc *mmu_alloc_rmap_desc(struct kvm_vcpu *vcpu)
337 {
338 return mmu_memory_cache_alloc(&vcpu->mmu_rmap_desc_cache,
339 sizeof(struct kvm_rmap_desc));
340 }
341
342 static void mmu_free_rmap_desc(struct kvm_rmap_desc *rd)
343 {
344 kfree(rd);
345 }
346
347 /*
348 * Take gfn and return the reverse mapping to it.
349 * Note: gfn must be unaliased before this function get called
350 */
351
352 static unsigned long *gfn_to_rmap(struct kvm *kvm, gfn_t gfn)
353 {
354 struct kvm_memory_slot *slot;
355
356 slot = gfn_to_memslot(kvm, gfn);
357 return &slot->rmap[gfn - slot->base_gfn];
358 }
359
360 /*
361 * Reverse mapping data structures:
362 *
363 * If rmapp bit zero is zero, then rmapp point to the shadw page table entry
364 * that points to page_address(page).
365 *
366 * If rmapp bit zero is one, (then rmap & ~1) points to a struct kvm_rmap_desc
367 * containing more mappings.
368 */
369 static void rmap_add(struct kvm_vcpu *vcpu, u64 *spte, gfn_t gfn)
370 {
371 struct kvm_mmu_page *sp;
372 struct kvm_rmap_desc *desc;
373 unsigned long *rmapp;
374 int i;
375
376 if (!is_rmap_pte(*spte))
377 return;
378 gfn = unalias_gfn(vcpu->kvm, gfn);
379 sp = page_header(__pa(spte));
380 sp->gfns[spte - sp->spt] = gfn;
381 rmapp = gfn_to_rmap(vcpu->kvm, gfn);
382 if (!*rmapp) {
383 rmap_printk("rmap_add: %p %llx 0->1\n", spte, *spte);
384 *rmapp = (unsigned long)spte;
385 } else if (!(*rmapp & 1)) {
386 rmap_printk("rmap_add: %p %llx 1->many\n", spte, *spte);
387 desc = mmu_alloc_rmap_desc(vcpu);
388 desc->shadow_ptes[0] = (u64 *)*rmapp;
389 desc->shadow_ptes[1] = spte;
390 *rmapp = (unsigned long)desc | 1;
391 } else {
392 rmap_printk("rmap_add: %p %llx many->many\n", spte, *spte);
393 desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
394 while (desc->shadow_ptes[RMAP_EXT-1] && desc->more)
395 desc = desc->more;
396 if (desc->shadow_ptes[RMAP_EXT-1]) {
397 desc->more = mmu_alloc_rmap_desc(vcpu);
398 desc = desc->more;
399 }
400 for (i = 0; desc->shadow_ptes[i]; ++i)
401 ;
402 desc->shadow_ptes[i] = spte;
403 }
404 }
405
406 static void rmap_desc_remove_entry(unsigned long *rmapp,
407 struct kvm_rmap_desc *desc,
408 int i,
409 struct kvm_rmap_desc *prev_desc)
410 {
411 int j;
412
413 for (j = RMAP_EXT - 1; !desc->shadow_ptes[j] && j > i; --j)
414 ;
415 desc->shadow_ptes[i] = desc->shadow_ptes[j];
416 desc->shadow_ptes[j] = NULL;
417 if (j != 0)
418 return;
419 if (!prev_desc && !desc->more)
420 *rmapp = (unsigned long)desc->shadow_ptes[0];
421 else
422 if (prev_desc)
423 prev_desc->more = desc->more;
424 else
425 *rmapp = (unsigned long)desc->more | 1;
426 mmu_free_rmap_desc(desc);
427 }
428
429 static void rmap_remove(struct kvm *kvm, u64 *spte)
430 {
431 struct kvm_rmap_desc *desc;
432 struct kvm_rmap_desc *prev_desc;
433 struct kvm_mmu_page *sp;
434 struct page *page;
435 unsigned long *rmapp;
436 int i;
437
438 if (!is_rmap_pte(*spte))
439 return;
440 sp = page_header(__pa(spte));
441 page = pfn_to_page((*spte & PT64_BASE_ADDR_MASK) >> PAGE_SHIFT);
442 mark_page_accessed(page);
443 if (is_writeble_pte(*spte))
444 kvm_release_page_dirty(page);
445 else
446 kvm_release_page_clean(page);
447 rmapp = gfn_to_rmap(kvm, sp->gfns[spte - sp->spt]);
448 if (!*rmapp) {
449 printk(KERN_ERR "rmap_remove: %p %llx 0->BUG\n", spte, *spte);
450 BUG();
451 } else if (!(*rmapp & 1)) {
452 rmap_printk("rmap_remove: %p %llx 1->0\n", spte, *spte);
453 if ((u64 *)*rmapp != spte) {
454 printk(KERN_ERR "rmap_remove: %p %llx 1->BUG\n",
455 spte, *spte);
456 BUG();
457 }
458 *rmapp = 0;
459 } else {
460 rmap_printk("rmap_remove: %p %llx many->many\n", spte, *spte);
461 desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
462 prev_desc = NULL;
463 while (desc) {
464 for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i)
465 if (desc->shadow_ptes[i] == spte) {
466 rmap_desc_remove_entry(rmapp,
467 desc, i,
468 prev_desc);
469 return;
470 }
471 prev_desc = desc;
472 desc = desc->more;
473 }
474 BUG();
475 }
476 }
477
478 static u64 *rmap_next(struct kvm *kvm, unsigned long *rmapp, u64 *spte)
479 {
480 struct kvm_rmap_desc *desc;
481 struct kvm_rmap_desc *prev_desc;
482 u64 *prev_spte;
483 int i;
484
485 if (!*rmapp)
486 return NULL;
487 else if (!(*rmapp & 1)) {
488 if (!spte)
489 return (u64 *)*rmapp;
490 return NULL;
491 }
492 desc = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
493 prev_desc = NULL;
494 prev_spte = NULL;
495 while (desc) {
496 for (i = 0; i < RMAP_EXT && desc->shadow_ptes[i]; ++i) {
497 if (prev_spte == spte)
498 return desc->shadow_ptes[i];
499 prev_spte = desc->shadow_ptes[i];
500 }
501 desc = desc->more;
502 }
503 return NULL;
504 }
505
506 static void rmap_write_protect(struct kvm *kvm, u64 gfn)
507 {
508 unsigned long *rmapp;
509 u64 *spte;
510
511 gfn = unalias_gfn(kvm, gfn);
512 rmapp = gfn_to_rmap(kvm, gfn);
513
514 spte = rmap_next(kvm, rmapp, NULL);
515 while (spte) {
516 BUG_ON(!spte);
517 BUG_ON(!(*spte & PT_PRESENT_MASK));
518 rmap_printk("rmap_write_protect: spte %p %llx\n", spte, *spte);
519 if (is_writeble_pte(*spte))
520 set_shadow_pte(spte, *spte & ~PT_WRITABLE_MASK);
521 kvm_flush_remote_tlbs(kvm);
522 spte = rmap_next(kvm, rmapp, spte);
523 }
524 }
525
526 #ifdef MMU_DEBUG
527 static int is_empty_shadow_page(u64 *spt)
528 {
529 u64 *pos;
530 u64 *end;
531
532 for (pos = spt, end = pos + PAGE_SIZE / sizeof(u64); pos != end; pos++)
533 if ((*pos & ~PT_SHADOW_IO_MARK) != shadow_trap_nonpresent_pte) {
534 printk(KERN_ERR "%s: %p %llx\n", __FUNCTION__,
535 pos, *pos);
536 return 0;
537 }
538 return 1;
539 }
540 #endif
541
542 static void kvm_mmu_free_page(struct kvm *kvm, struct kvm_mmu_page *sp)
543 {
544 ASSERT(is_empty_shadow_page(sp->spt));
545 list_del(&sp->link);
546 __free_page(virt_to_page(sp->spt));
547 __free_page(virt_to_page(sp->gfns));
548 kfree(sp);
549 ++kvm->n_free_mmu_pages;
550 }
551
552 static unsigned kvm_page_table_hashfn(gfn_t gfn)
553 {
554 return gfn;
555 }
556
557 static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu,
558 u64 *parent_pte)
559 {
560 struct kvm_mmu_page *sp;
561
562 if (!vcpu->kvm->n_free_mmu_pages)
563 return NULL;
564
565 sp = mmu_memory_cache_alloc(&vcpu->mmu_page_header_cache, sizeof *sp);
566 sp->spt = mmu_memory_cache_alloc(&vcpu->mmu_page_cache, PAGE_SIZE);
567 sp->gfns = mmu_memory_cache_alloc(&vcpu->mmu_page_cache, PAGE_SIZE);
568 set_page_private(virt_to_page(sp->spt), (unsigned long)sp);
569 list_add(&sp->link, &vcpu->kvm->active_mmu_pages);
570 ASSERT(is_empty_shadow_page(sp->spt));
571 sp->slot_bitmap = 0;
572 sp->multimapped = 0;
573 sp->parent_pte = parent_pte;
574 --vcpu->kvm->n_free_mmu_pages;
575 return sp;
576 }
577
578 static void mmu_page_add_parent_pte(struct kvm_vcpu *vcpu,
579 struct kvm_mmu_page *sp, u64 *parent_pte)
580 {
581 struct kvm_pte_chain *pte_chain;
582 struct hlist_node *node;
583 int i;
584
585 if (!parent_pte)
586 return;
587 if (!sp->multimapped) {
588 u64 *old = sp->parent_pte;
589
590 if (!old) {
591 sp->parent_pte = parent_pte;
592 return;
593 }
594 sp->multimapped = 1;
595 pte_chain = mmu_alloc_pte_chain(vcpu);
596 INIT_HLIST_HEAD(&sp->parent_ptes);
597 hlist_add_head(&pte_chain->link, &sp->parent_ptes);
598 pte_chain->parent_ptes[0] = old;
599 }
600 hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link) {
601 if (pte_chain->parent_ptes[NR_PTE_CHAIN_ENTRIES-1])
602 continue;
603 for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i)
604 if (!pte_chain->parent_ptes[i]) {
605 pte_chain->parent_ptes[i] = parent_pte;
606 return;
607 }
608 }
609 pte_chain = mmu_alloc_pte_chain(vcpu);
610 BUG_ON(!pte_chain);
611 hlist_add_head(&pte_chain->link, &sp->parent_ptes);
612 pte_chain->parent_ptes[0] = parent_pte;
613 }
614
615 static void mmu_page_remove_parent_pte(struct kvm_mmu_page *sp,
616 u64 *parent_pte)
617 {
618 struct kvm_pte_chain *pte_chain;
619 struct hlist_node *node;
620 int i;
621
622 if (!sp->multimapped) {
623 BUG_ON(sp->parent_pte != parent_pte);
624 sp->parent_pte = NULL;
625 return;
626 }
627 hlist_for_each_entry(pte_chain, node, &sp->parent_ptes, link)
628 for (i = 0; i < NR_PTE_CHAIN_ENTRIES; ++i) {
629 if (!pte_chain->parent_ptes[i])
630 break;
631 if (pte_chain->parent_ptes[i] != parent_pte)
632 continue;
633 while (i + 1 < NR_PTE_CHAIN_ENTRIES
634 && pte_chain->parent_ptes[i + 1]) {
635 pte_chain->parent_ptes[i]
636 = pte_chain->parent_ptes[i + 1];
637 ++i;
638 }
639 pte_chain->parent_ptes[i] = NULL;
640 if (i == 0) {
641 hlist_del(&pte_chain->link);
642 mmu_free_pte_chain(pte_chain);
643 if (hlist_empty(&sp->parent_ptes)) {
644 sp->multimapped = 0;
645 sp->parent_pte = NULL;
646 }
647 }
648 return;
649 }
650 BUG();
651 }
652
653 static struct kvm_mmu_page *kvm_mmu_lookup_page(struct kvm *kvm, gfn_t gfn)
654 {
655 unsigned index;
656 struct hlist_head *bucket;
657 struct kvm_mmu_page *sp;
658 struct hlist_node *node;
659
660 pgprintk("%s: looking for gfn %lx\n", __FUNCTION__, gfn);
661 index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES;
662 bucket = &kvm->mmu_page_hash[index];
663 hlist_for_each_entry(sp, node, bucket, hash_link)
664 if (sp->gfn == gfn && !sp->role.metaphysical) {
665 pgprintk("%s: found role %x\n",
666 __FUNCTION__, sp->role.word);
667 return sp;
668 }
669 return NULL;
670 }
671
672 static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
673 gfn_t gfn,
674 gva_t gaddr,
675 unsigned level,
676 int metaphysical,
677 unsigned hugepage_access,
678 u64 *parent_pte)
679 {
680 union kvm_mmu_page_role role;
681 unsigned index;
682 unsigned quadrant;
683 struct hlist_head *bucket;
684 struct kvm_mmu_page *sp;
685 struct hlist_node *node;
686
687 role.word = 0;
688 role.glevels = vcpu->mmu.root_level;
689 role.level = level;
690 role.metaphysical = metaphysical;
691 role.hugepage_access = hugepage_access;
692 if (vcpu->mmu.root_level <= PT32_ROOT_LEVEL) {
693 quadrant = gaddr >> (PAGE_SHIFT + (PT64_PT_BITS * level));
694 quadrant &= (1 << ((PT32_PT_BITS - PT64_PT_BITS) * level)) - 1;
695 role.quadrant = quadrant;
696 }
697 pgprintk("%s: looking gfn %lx role %x\n", __FUNCTION__,
698 gfn, role.word);
699 index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES;
700 bucket = &vcpu->kvm->mmu_page_hash[index];
701 hlist_for_each_entry(sp, node, bucket, hash_link)
702 if (sp->gfn == gfn && sp->role.word == role.word) {
703 mmu_page_add_parent_pte(vcpu, sp, parent_pte);
704 pgprintk("%s: found\n", __FUNCTION__);
705 return sp;
706 }
707 sp = kvm_mmu_alloc_page(vcpu, parent_pte);
708 if (!sp)
709 return sp;
710 pgprintk("%s: adding gfn %lx role %x\n", __FUNCTION__, gfn, role.word);
711 sp->gfn = gfn;
712 sp->role = role;
713 hlist_add_head(&sp->hash_link, bucket);
714 vcpu->mmu.prefetch_page(vcpu, sp);
715 if (!metaphysical)
716 rmap_write_protect(vcpu->kvm, gfn);
717 return sp;
718 }
719
720 static void kvm_mmu_page_unlink_children(struct kvm *kvm,
721 struct kvm_mmu_page *sp)
722 {
723 unsigned i;
724 u64 *pt;
725 u64 ent;
726
727 pt = sp->spt;
728
729 if (sp->role.level == PT_PAGE_TABLE_LEVEL) {
730 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
731 if (is_shadow_present_pte(pt[i]))
732 rmap_remove(kvm, &pt[i]);
733 pt[i] = shadow_trap_nonpresent_pte;
734 }
735 kvm_flush_remote_tlbs(kvm);
736 return;
737 }
738
739 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
740 ent = pt[i];
741
742 pt[i] = shadow_trap_nonpresent_pte;
743 if (!is_shadow_present_pte(ent))
744 continue;
745 ent &= PT64_BASE_ADDR_MASK;
746 mmu_page_remove_parent_pte(page_header(ent), &pt[i]);
747 }
748 kvm_flush_remote_tlbs(kvm);
749 }
750
751 static void kvm_mmu_put_page(struct kvm_mmu_page *sp, u64 *parent_pte)
752 {
753 mmu_page_remove_parent_pte(sp, parent_pte);
754 }
755
756 static void kvm_mmu_reset_last_pte_updated(struct kvm *kvm)
757 {
758 int i;
759
760 for (i = 0; i < KVM_MAX_VCPUS; ++i)
761 if (kvm->vcpus[i])
762 kvm->vcpus[i]->last_pte_updated = NULL;
763 }
764
765 static void kvm_mmu_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp)
766 {
767 u64 *parent_pte;
768
769 ++kvm->stat.mmu_shadow_zapped;
770 while (sp->multimapped || sp->parent_pte) {
771 if (!sp->multimapped)
772 parent_pte = sp->parent_pte;
773 else {
774 struct kvm_pte_chain *chain;
775
776 chain = container_of(sp->parent_ptes.first,
777 struct kvm_pte_chain, link);
778 parent_pte = chain->parent_ptes[0];
779 }
780 BUG_ON(!parent_pte);
781 kvm_mmu_put_page(sp, parent_pte);
782 set_shadow_pte(parent_pte, shadow_trap_nonpresent_pte);
783 }
784 kvm_mmu_page_unlink_children(kvm, sp);
785 if (!sp->root_count) {
786 hlist_del(&sp->hash_link);
787 kvm_mmu_free_page(kvm, sp);
788 } else
789 list_move(&sp->link, &kvm->active_mmu_pages);
790 kvm_mmu_reset_last_pte_updated(kvm);
791 }
792
793 /*
794 * Changing the number of mmu pages allocated to the vm
795 * Note: if kvm_nr_mmu_pages is too small, you will get dead lock
796 */
797 void kvm_mmu_change_mmu_pages(struct kvm *kvm, unsigned int kvm_nr_mmu_pages)
798 {
799 /*
800 * If we set the number of mmu pages to be smaller be than the
801 * number of actived pages , we must to free some mmu pages before we
802 * change the value
803 */
804
805 if ((kvm->n_alloc_mmu_pages - kvm->n_free_mmu_pages) >
806 kvm_nr_mmu_pages) {
807 int n_used_mmu_pages = kvm->n_alloc_mmu_pages
808 - kvm->n_free_mmu_pages;
809
810 while (n_used_mmu_pages > kvm_nr_mmu_pages) {
811 struct kvm_mmu_page *page;
812
813 page = container_of(kvm->active_mmu_pages.prev,
814 struct kvm_mmu_page, link);
815 kvm_mmu_zap_page(kvm, page);
816 n_used_mmu_pages--;
817 }
818 kvm->n_free_mmu_pages = 0;
819 }
820 else
821 kvm->n_free_mmu_pages += kvm_nr_mmu_pages
822 - kvm->n_alloc_mmu_pages;
823
824 kvm->n_alloc_mmu_pages = kvm_nr_mmu_pages;
825 }
826
827 static int kvm_mmu_unprotect_page(struct kvm *kvm, gfn_t gfn)
828 {
829 unsigned index;
830 struct hlist_head *bucket;
831 struct kvm_mmu_page *sp;
832 struct hlist_node *node, *n;
833 int r;
834
835 pgprintk("%s: looking for gfn %lx\n", __FUNCTION__, gfn);
836 r = 0;
837 index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES;
838 bucket = &kvm->mmu_page_hash[index];
839 hlist_for_each_entry_safe(sp, node, n, bucket, hash_link)
840 if (sp->gfn == gfn && !sp->role.metaphysical) {
841 pgprintk("%s: gfn %lx role %x\n", __FUNCTION__, gfn,
842 sp->role.word);
843 kvm_mmu_zap_page(kvm, sp);
844 r = 1;
845 }
846 return r;
847 }
848
849 static void mmu_unshadow(struct kvm *kvm, gfn_t gfn)
850 {
851 struct kvm_mmu_page *sp;
852
853 while ((sp = kvm_mmu_lookup_page(kvm, gfn)) != NULL) {
854 pgprintk("%s: zap %lx %x\n", __FUNCTION__, gfn, sp->role.word);
855 kvm_mmu_zap_page(kvm, sp);
856 }
857 }
858
859 static void page_header_update_slot(struct kvm *kvm, void *pte, gfn_t gfn)
860 {
861 int slot = memslot_id(kvm, gfn_to_memslot(kvm, gfn));
862 struct kvm_mmu_page *sp = page_header(__pa(pte));
863
864 __set_bit(slot, &sp->slot_bitmap);
865 }
866
867 struct page *gva_to_page(struct kvm_vcpu *vcpu, gva_t gva)
868 {
869 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, gva);
870
871 if (gpa == UNMAPPED_GVA)
872 return NULL;
873 return gfn_to_page(vcpu->kvm, gpa >> PAGE_SHIFT);
874 }
875
876 static void nonpaging_new_cr3(struct kvm_vcpu *vcpu)
877 {
878 }
879
880 static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, struct page *page)
881 {
882 int level = PT32E_ROOT_LEVEL;
883 hpa_t table_addr = vcpu->mmu.root_hpa;
884
885 for (; ; level--) {
886 u32 index = PT64_INDEX(v, level);
887 u64 *table;
888 u64 pte;
889
890 ASSERT(VALID_PAGE(table_addr));
891 table = __va(table_addr);
892
893 if (level == 1) {
894 int was_rmapped;
895
896 pte = table[index];
897 was_rmapped = is_rmap_pte(pte);
898 if (is_shadow_present_pte(pte) && is_writeble_pte(pte)) {
899 kvm_release_page_clean(page);
900 return 0;
901 }
902 mark_page_dirty(vcpu->kvm, v >> PAGE_SHIFT);
903 page_header_update_slot(vcpu->kvm, table,
904 v >> PAGE_SHIFT);
905 table[index] = page_to_phys(page)
906 | PT_PRESENT_MASK | PT_WRITABLE_MASK
907 | PT_USER_MASK;
908 if (!was_rmapped)
909 rmap_add(vcpu, &table[index], v >> PAGE_SHIFT);
910 else
911 kvm_release_page_clean(page);
912
913 return 0;
914 }
915
916 if (table[index] == shadow_trap_nonpresent_pte) {
917 struct kvm_mmu_page *new_table;
918 gfn_t pseudo_gfn;
919
920 pseudo_gfn = (v & PT64_DIR_BASE_ADDR_MASK)
921 >> PAGE_SHIFT;
922 new_table = kvm_mmu_get_page(vcpu, pseudo_gfn,
923 v, level - 1,
924 1, 3, &table[index]);
925 if (!new_table) {
926 pgprintk("nonpaging_map: ENOMEM\n");
927 kvm_release_page_clean(page);
928 return -ENOMEM;
929 }
930
931 table[index] = __pa(new_table->spt) | PT_PRESENT_MASK
932 | PT_WRITABLE_MASK | PT_USER_MASK;
933 }
934 table_addr = table[index] & PT64_BASE_ADDR_MASK;
935 }
936 }
937
938 static void nonpaging_prefetch_page(struct kvm_vcpu *vcpu,
939 struct kvm_mmu_page *sp)
940 {
941 int i;
942
943 for (i = 0; i < PT64_ENT_PER_PAGE; ++i)
944 sp->spt[i] = shadow_trap_nonpresent_pte;
945 }
946
947 static void mmu_free_roots(struct kvm_vcpu *vcpu)
948 {
949 int i;
950 struct kvm_mmu_page *sp;
951
952 if (!VALID_PAGE(vcpu->mmu.root_hpa))
953 return;
954 #ifdef CONFIG_X86_64
955 if (vcpu->mmu.shadow_root_level == PT64_ROOT_LEVEL) {
956 hpa_t root = vcpu->mmu.root_hpa;
957
958 sp = page_header(root);
959 --sp->root_count;
960 vcpu->mmu.root_hpa = INVALID_PAGE;
961 return;
962 }
963 #endif
964 for (i = 0; i < 4; ++i) {
965 hpa_t root = vcpu->mmu.pae_root[i];
966
967 if (root) {
968 root &= PT64_BASE_ADDR_MASK;
969 sp = page_header(root);
970 --sp->root_count;
971 }
972 vcpu->mmu.pae_root[i] = INVALID_PAGE;
973 }
974 vcpu->mmu.root_hpa = INVALID_PAGE;
975 }
976
977 static void mmu_alloc_roots(struct kvm_vcpu *vcpu)
978 {
979 int i;
980 gfn_t root_gfn;
981 struct kvm_mmu_page *sp;
982
983 root_gfn = vcpu->cr3 >> PAGE_SHIFT;
984
985 #ifdef CONFIG_X86_64
986 if (vcpu->mmu.shadow_root_level == PT64_ROOT_LEVEL) {
987 hpa_t root = vcpu->mmu.root_hpa;
988
989 ASSERT(!VALID_PAGE(root));
990 sp = kvm_mmu_get_page(vcpu, root_gfn, 0,
991 PT64_ROOT_LEVEL, 0, 0, NULL);
992 root = __pa(sp->spt);
993 ++sp->root_count;
994 vcpu->mmu.root_hpa = root;
995 return;
996 }
997 #endif
998 for (i = 0; i < 4; ++i) {
999 hpa_t root = vcpu->mmu.pae_root[i];
1000
1001 ASSERT(!VALID_PAGE(root));
1002 if (vcpu->mmu.root_level == PT32E_ROOT_LEVEL) {
1003 if (!is_present_pte(vcpu->pdptrs[i])) {
1004 vcpu->mmu.pae_root[i] = 0;
1005 continue;
1006 }
1007 root_gfn = vcpu->pdptrs[i] >> PAGE_SHIFT;
1008 } else if (vcpu->mmu.root_level == 0)
1009 root_gfn = 0;
1010 sp = kvm_mmu_get_page(vcpu, root_gfn, i << 30,
1011 PT32_ROOT_LEVEL, !is_paging(vcpu),
1012 0, NULL);
1013 root = __pa(sp->spt);
1014 ++sp->root_count;
1015 vcpu->mmu.pae_root[i] = root | PT_PRESENT_MASK;
1016 }
1017 vcpu->mmu.root_hpa = __pa(vcpu->mmu.pae_root);
1018 }
1019
1020 static gpa_t nonpaging_gva_to_gpa(struct kvm_vcpu *vcpu, gva_t vaddr)
1021 {
1022 return vaddr;
1023 }
1024
1025 static int nonpaging_page_fault(struct kvm_vcpu *vcpu, gva_t gva,
1026 u32 error_code)
1027 {
1028 struct page *page;
1029 int r;
1030
1031 r = mmu_topup_memory_caches(vcpu);
1032 if (r)
1033 return r;
1034
1035 ASSERT(vcpu);
1036 ASSERT(VALID_PAGE(vcpu->mmu.root_hpa));
1037
1038 page = gfn_to_page(vcpu->kvm, gva >> PAGE_SHIFT);
1039
1040 if (is_error_page(page)) {
1041 kvm_release_page_clean(page);
1042 return 1;
1043 }
1044
1045 return nonpaging_map(vcpu, gva & PAGE_MASK, page);
1046 }
1047
1048 static void nonpaging_free(struct kvm_vcpu *vcpu)
1049 {
1050 mmu_free_roots(vcpu);
1051 }
1052
1053 static int nonpaging_init_context(struct kvm_vcpu *vcpu)
1054 {
1055 struct kvm_mmu *context = &vcpu->mmu;
1056
1057 context->new_cr3 = nonpaging_new_cr3;
1058 context->page_fault = nonpaging_page_fault;
1059 context->gva_to_gpa = nonpaging_gva_to_gpa;
1060 context->free = nonpaging_free;
1061 context->prefetch_page = nonpaging_prefetch_page;
1062 context->root_level = 0;
1063 context->shadow_root_level = PT32E_ROOT_LEVEL;
1064 context->root_hpa = INVALID_PAGE;
1065 return 0;
1066 }
1067
1068 void kvm_mmu_flush_tlb(struct kvm_vcpu *vcpu)
1069 {
1070 ++vcpu->stat.tlb_flush;
1071 kvm_x86_ops->tlb_flush(vcpu);
1072 }
1073
1074 static void paging_new_cr3(struct kvm_vcpu *vcpu)
1075 {
1076 pgprintk("%s: cr3 %lx\n", __FUNCTION__, vcpu->cr3);
1077 mmu_free_roots(vcpu);
1078 }
1079
1080 static void inject_page_fault(struct kvm_vcpu *vcpu,
1081 u64 addr,
1082 u32 err_code)
1083 {
1084 kvm_x86_ops->inject_page_fault(vcpu, addr, err_code);
1085 }
1086
1087 static void paging_free(struct kvm_vcpu *vcpu)
1088 {
1089 nonpaging_free(vcpu);
1090 }
1091
1092 #define PTTYPE 64
1093 #include "paging_tmpl.h"
1094 #undef PTTYPE
1095
1096 #define PTTYPE 32
1097 #include "paging_tmpl.h"
1098 #undef PTTYPE
1099
1100 static int paging64_init_context_common(struct kvm_vcpu *vcpu, int level)
1101 {
1102 struct kvm_mmu *context = &vcpu->mmu;
1103
1104 ASSERT(is_pae(vcpu));
1105 context->new_cr3 = paging_new_cr3;
1106 context->page_fault = paging64_page_fault;
1107 context->gva_to_gpa = paging64_gva_to_gpa;
1108 context->prefetch_page = paging64_prefetch_page;
1109 context->free = paging_free;
1110 context->root_level = level;
1111 context->shadow_root_level = level;
1112 context->root_hpa = INVALID_PAGE;
1113 return 0;
1114 }
1115
1116 static int paging64_init_context(struct kvm_vcpu *vcpu)
1117 {
1118 return paging64_init_context_common(vcpu, PT64_ROOT_LEVEL);
1119 }
1120
1121 static int paging32_init_context(struct kvm_vcpu *vcpu)
1122 {
1123 struct kvm_mmu *context = &vcpu->mmu;
1124
1125 context->new_cr3 = paging_new_cr3;
1126 context->page_fault = paging32_page_fault;
1127 context->gva_to_gpa = paging32_gva_to_gpa;
1128 context->free = paging_free;
1129 context->prefetch_page = paging32_prefetch_page;
1130 context->root_level = PT32_ROOT_LEVEL;
1131 context->shadow_root_level = PT32E_ROOT_LEVEL;
1132 context->root_hpa = INVALID_PAGE;
1133 return 0;
1134 }
1135
1136 static int paging32E_init_context(struct kvm_vcpu *vcpu)
1137 {
1138 return paging64_init_context_common(vcpu, PT32E_ROOT_LEVEL);
1139 }
1140
1141 static int init_kvm_mmu(struct kvm_vcpu *vcpu)
1142 {
1143 ASSERT(vcpu);
1144 ASSERT(!VALID_PAGE(vcpu->mmu.root_hpa));
1145
1146 if (!is_paging(vcpu))
1147 return nonpaging_init_context(vcpu);
1148 else if (is_long_mode(vcpu))
1149 return paging64_init_context(vcpu);
1150 else if (is_pae(vcpu))
1151 return paging32E_init_context(vcpu);
1152 else
1153 return paging32_init_context(vcpu);
1154 }
1155
1156 static void destroy_kvm_mmu(struct kvm_vcpu *vcpu)
1157 {
1158 ASSERT(vcpu);
1159 if (VALID_PAGE(vcpu->mmu.root_hpa)) {
1160 vcpu->mmu.free(vcpu);
1161 vcpu->mmu.root_hpa = INVALID_PAGE;
1162 }
1163 }
1164
1165 int kvm_mmu_reset_context(struct kvm_vcpu *vcpu)
1166 {
1167 destroy_kvm_mmu(vcpu);
1168 return init_kvm_mmu(vcpu);
1169 }
1170 EXPORT_SYMBOL_GPL(kvm_mmu_reset_context);
1171
1172 int kvm_mmu_load(struct kvm_vcpu *vcpu)
1173 {
1174 int r;
1175
1176 mutex_lock(&vcpu->kvm->lock);
1177 r = mmu_topup_memory_caches(vcpu);
1178 if (r)
1179 goto out;
1180 mmu_alloc_roots(vcpu);
1181 kvm_x86_ops->set_cr3(vcpu, vcpu->mmu.root_hpa);
1182 kvm_mmu_flush_tlb(vcpu);
1183 out:
1184 mutex_unlock(&vcpu->kvm->lock);
1185 return r;
1186 }
1187 EXPORT_SYMBOL_GPL(kvm_mmu_load);
1188
1189 void kvm_mmu_unload(struct kvm_vcpu *vcpu)
1190 {
1191 mmu_free_roots(vcpu);
1192 }
1193
1194 static void mmu_pte_write_zap_pte(struct kvm_vcpu *vcpu,
1195 struct kvm_mmu_page *sp,
1196 u64 *spte)
1197 {
1198 u64 pte;
1199 struct kvm_mmu_page *child;
1200
1201 pte = *spte;
1202 if (is_shadow_present_pte(pte)) {
1203 if (sp->role.level == PT_PAGE_TABLE_LEVEL)
1204 rmap_remove(vcpu->kvm, spte);
1205 else {
1206 child = page_header(pte & PT64_BASE_ADDR_MASK);
1207 mmu_page_remove_parent_pte(child, spte);
1208 }
1209 }
1210 set_shadow_pte(spte, shadow_trap_nonpresent_pte);
1211 }
1212
1213 static void mmu_pte_write_new_pte(struct kvm_vcpu *vcpu,
1214 struct kvm_mmu_page *sp,
1215 u64 *spte,
1216 const void *new, int bytes,
1217 int offset_in_pte)
1218 {
1219 if (sp->role.level != PT_PAGE_TABLE_LEVEL) {
1220 ++vcpu->kvm->stat.mmu_pde_zapped;
1221 return;
1222 }
1223
1224 ++vcpu->kvm->stat.mmu_pte_updated;
1225 if (sp->role.glevels == PT32_ROOT_LEVEL)
1226 paging32_update_pte(vcpu, sp, spte, new, bytes, offset_in_pte);
1227 else
1228 paging64_update_pte(vcpu, sp, spte, new, bytes, offset_in_pte);
1229 }
1230
1231 static bool need_remote_flush(u64 old, u64 new)
1232 {
1233 if (!is_shadow_present_pte(old))
1234 return false;
1235 if (!is_shadow_present_pte(new))
1236 return true;
1237 if ((old ^ new) & PT64_BASE_ADDR_MASK)
1238 return true;
1239 old ^= PT64_NX_MASK;
1240 new ^= PT64_NX_MASK;
1241 return (old & ~new & PT64_PERM_MASK) != 0;
1242 }
1243
1244 static void mmu_pte_write_flush_tlb(struct kvm_vcpu *vcpu, u64 old, u64 new)
1245 {
1246 if (need_remote_flush(old, new))
1247 kvm_flush_remote_tlbs(vcpu->kvm);
1248 else
1249 kvm_mmu_flush_tlb(vcpu);
1250 }
1251
1252 static bool last_updated_pte_accessed(struct kvm_vcpu *vcpu)
1253 {
1254 u64 *spte = vcpu->last_pte_updated;
1255
1256 return !!(spte && (*spte & PT_ACCESSED_MASK));
1257 }
1258
1259 void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
1260 const u8 *new, int bytes)
1261 {
1262 gfn_t gfn = gpa >> PAGE_SHIFT;
1263 struct kvm_mmu_page *sp;
1264 struct hlist_node *node, *n;
1265 struct hlist_head *bucket;
1266 unsigned index;
1267 u64 entry;
1268 u64 *spte;
1269 unsigned offset = offset_in_page(gpa);
1270 unsigned pte_size;
1271 unsigned page_offset;
1272 unsigned misaligned;
1273 unsigned quadrant;
1274 int level;
1275 int flooded = 0;
1276 int npte;
1277
1278 pgprintk("%s: gpa %llx bytes %d\n", __FUNCTION__, gpa, bytes);
1279 ++vcpu->kvm->stat.mmu_pte_write;
1280 kvm_mmu_audit(vcpu, "pre pte write");
1281 if (gfn == vcpu->last_pt_write_gfn
1282 && !last_updated_pte_accessed(vcpu)) {
1283 ++vcpu->last_pt_write_count;
1284 if (vcpu->last_pt_write_count >= 3)
1285 flooded = 1;
1286 } else {
1287 vcpu->last_pt_write_gfn = gfn;
1288 vcpu->last_pt_write_count = 1;
1289 vcpu->last_pte_updated = NULL;
1290 }
1291 index = kvm_page_table_hashfn(gfn) % KVM_NUM_MMU_PAGES;
1292 bucket = &vcpu->kvm->mmu_page_hash[index];
1293 hlist_for_each_entry_safe(sp, node, n, bucket, hash_link) {
1294 if (sp->gfn != gfn || sp->role.metaphysical)
1295 continue;
1296 pte_size = sp->role.glevels == PT32_ROOT_LEVEL ? 4 : 8;
1297 misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1);
1298 misaligned |= bytes < 4;
1299 if (misaligned || flooded) {
1300 /*
1301 * Misaligned accesses are too much trouble to fix
1302 * up; also, they usually indicate a page is not used
1303 * as a page table.
1304 *
1305 * If we're seeing too many writes to a page,
1306 * it may no longer be a page table, or we may be
1307 * forking, in which case it is better to unmap the
1308 * page.
1309 */
1310 pgprintk("misaligned: gpa %llx bytes %d role %x\n",
1311 gpa, bytes, sp->role.word);
1312 kvm_mmu_zap_page(vcpu->kvm, sp);
1313 ++vcpu->kvm->stat.mmu_flooded;
1314 continue;
1315 }
1316 page_offset = offset;
1317 level = sp->role.level;
1318 npte = 1;
1319 if (sp->role.glevels == PT32_ROOT_LEVEL) {
1320 page_offset <<= 1; /* 32->64 */
1321 /*
1322 * A 32-bit pde maps 4MB while the shadow pdes map
1323 * only 2MB. So we need to double the offset again
1324 * and zap two pdes instead of one.
1325 */
1326 if (level == PT32_ROOT_LEVEL) {
1327 page_offset &= ~7; /* kill rounding error */
1328 page_offset <<= 1;
1329 npte = 2;
1330 }
1331 quadrant = page_offset >> PAGE_SHIFT;
1332 page_offset &= ~PAGE_MASK;
1333 if (quadrant != sp->role.quadrant)
1334 continue;
1335 }
1336 spte = &sp->spt[page_offset / sizeof(*spte)];
1337 while (npte--) {
1338 entry = *spte;
1339 mmu_pte_write_zap_pte(vcpu, sp, spte);
1340 mmu_pte_write_new_pte(vcpu, sp, spte, new, bytes,
1341 page_offset & (pte_size - 1));
1342 mmu_pte_write_flush_tlb(vcpu, entry, *spte);
1343 ++spte;
1344 }
1345 }
1346 kvm_mmu_audit(vcpu, "post pte write");
1347 }
1348
1349 int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva)
1350 {
1351 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, gva);
1352
1353 return kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT);
1354 }
1355
1356 void __kvm_mmu_free_some_pages(struct kvm_vcpu *vcpu)
1357 {
1358 while (vcpu->kvm->n_free_mmu_pages < KVM_REFILL_PAGES) {
1359 struct kvm_mmu_page *sp;
1360
1361 sp = container_of(vcpu->kvm->active_mmu_pages.prev,
1362 struct kvm_mmu_page, link);
1363 kvm_mmu_zap_page(vcpu->kvm, sp);
1364 ++vcpu->kvm->stat.mmu_recycled;
1365 }
1366 }
1367
1368 int kvm_mmu_page_fault(struct kvm_vcpu *vcpu, gva_t cr2, u32 error_code)
1369 {
1370 int r;
1371 enum emulation_result er;
1372
1373 mutex_lock(&vcpu->kvm->lock);
1374 r = vcpu->mmu.page_fault(vcpu, cr2, error_code);
1375 if (r < 0)
1376 goto out;
1377
1378 if (!r) {
1379 r = 1;
1380 goto out;
1381 }
1382
1383 r = mmu_topup_memory_caches(vcpu);
1384 if (r)
1385 goto out;
1386
1387 er = emulate_instruction(vcpu, vcpu->run, cr2, error_code, 0);
1388 mutex_unlock(&vcpu->kvm->lock);
1389
1390 switch (er) {
1391 case EMULATE_DONE:
1392 return 1;
1393 case EMULATE_DO_MMIO:
1394 ++vcpu->stat.mmio_exits;
1395 return 0;
1396 case EMULATE_FAIL:
1397 kvm_report_emulation_failure(vcpu, "pagetable");
1398 return 1;
1399 default:
1400 BUG();
1401 }
1402 out:
1403 mutex_unlock(&vcpu->kvm->lock);
1404 return r;
1405 }
1406 EXPORT_SYMBOL_GPL(kvm_mmu_page_fault);
1407
1408 static void free_mmu_pages(struct kvm_vcpu *vcpu)
1409 {
1410 struct kvm_mmu_page *sp;
1411
1412 while (!list_empty(&vcpu->kvm->active_mmu_pages)) {
1413 sp = container_of(vcpu->kvm->active_mmu_pages.next,
1414 struct kvm_mmu_page, link);
1415 kvm_mmu_zap_page(vcpu->kvm, sp);
1416 }
1417 free_page((unsigned long)vcpu->mmu.pae_root);
1418 }
1419
1420 static int alloc_mmu_pages(struct kvm_vcpu *vcpu)
1421 {
1422 struct page *page;
1423 int i;
1424
1425 ASSERT(vcpu);
1426
1427 if (vcpu->kvm->n_requested_mmu_pages)
1428 vcpu->kvm->n_free_mmu_pages = vcpu->kvm->n_requested_mmu_pages;
1429 else
1430 vcpu->kvm->n_free_mmu_pages = vcpu->kvm->n_alloc_mmu_pages;
1431 /*
1432 * When emulating 32-bit mode, cr3 is only 32 bits even on x86_64.
1433 * Therefore we need to allocate shadow page tables in the first
1434 * 4GB of memory, which happens to fit the DMA32 zone.
1435 */
1436 page = alloc_page(GFP_KERNEL | __GFP_DMA32);
1437 if (!page)
1438 goto error_1;
1439 vcpu->mmu.pae_root = page_address(page);
1440 for (i = 0; i < 4; ++i)
1441 vcpu->mmu.pae_root[i] = INVALID_PAGE;
1442
1443 return 0;
1444
1445 error_1:
1446 free_mmu_pages(vcpu);
1447 return -ENOMEM;
1448 }
1449
1450 int kvm_mmu_create(struct kvm_vcpu *vcpu)
1451 {
1452 ASSERT(vcpu);
1453 ASSERT(!VALID_PAGE(vcpu->mmu.root_hpa));
1454
1455 return alloc_mmu_pages(vcpu);
1456 }
1457
1458 int kvm_mmu_setup(struct kvm_vcpu *vcpu)
1459 {
1460 ASSERT(vcpu);
1461 ASSERT(!VALID_PAGE(vcpu->mmu.root_hpa));
1462
1463 return init_kvm_mmu(vcpu);
1464 }
1465
1466 void kvm_mmu_destroy(struct kvm_vcpu *vcpu)
1467 {
1468 ASSERT(vcpu);
1469
1470 destroy_kvm_mmu(vcpu);
1471 free_mmu_pages(vcpu);
1472 mmu_free_memory_caches(vcpu);
1473 }
1474
1475 void kvm_mmu_slot_remove_write_access(struct kvm *kvm, int slot)
1476 {
1477 struct kvm_mmu_page *sp;
1478
1479 list_for_each_entry(sp, &kvm->active_mmu_pages, link) {
1480 int i;
1481 u64 *pt;
1482
1483 if (!test_bit(slot, &sp->slot_bitmap))
1484 continue;
1485
1486 pt = sp->spt;
1487 for (i = 0; i < PT64_ENT_PER_PAGE; ++i)
1488 /* avoid RMW */
1489 if (pt[i] & PT_WRITABLE_MASK)
1490 pt[i] &= ~PT_WRITABLE_MASK;
1491 }
1492 }
1493
1494 void kvm_mmu_zap_all(struct kvm *kvm)
1495 {
1496 struct kvm_mmu_page *sp, *node;
1497
1498 list_for_each_entry_safe(sp, node, &kvm->active_mmu_pages, link)
1499 kvm_mmu_zap_page(kvm, sp);
1500
1501 kvm_flush_remote_tlbs(kvm);
1502 }
1503
1504 void kvm_mmu_module_exit(void)
1505 {
1506 if (pte_chain_cache)
1507 kmem_cache_destroy(pte_chain_cache);
1508 if (rmap_desc_cache)
1509 kmem_cache_destroy(rmap_desc_cache);
1510 if (mmu_page_header_cache)
1511 kmem_cache_destroy(mmu_page_header_cache);
1512 }
1513
1514 int kvm_mmu_module_init(void)
1515 {
1516 pte_chain_cache = kmem_cache_create("kvm_pte_chain",
1517 sizeof(struct kvm_pte_chain),
1518 0, 0, NULL);
1519 if (!pte_chain_cache)
1520 goto nomem;
1521 rmap_desc_cache = kmem_cache_create("kvm_rmap_desc",
1522 sizeof(struct kvm_rmap_desc),
1523 0, 0, NULL);
1524 if (!rmap_desc_cache)
1525 goto nomem;
1526
1527 mmu_page_header_cache = kmem_cache_create("kvm_mmu_page_header",
1528 sizeof(struct kvm_mmu_page),
1529 0, 0, NULL);
1530 if (!mmu_page_header_cache)
1531 goto nomem;
1532
1533 return 0;
1534
1535 nomem:
1536 kvm_mmu_module_exit();
1537 return -ENOMEM;
1538 }
1539
1540 /*
1541 * Caculate mmu pages needed for kvm.
1542 */
1543 unsigned int kvm_mmu_calculate_mmu_pages(struct kvm *kvm)
1544 {
1545 int i;
1546 unsigned int nr_mmu_pages;
1547 unsigned int nr_pages = 0;
1548
1549 for (i = 0; i < kvm->nmemslots; i++)
1550 nr_pages += kvm->memslots[i].npages;
1551
1552 nr_mmu_pages = nr_pages * KVM_PERMILLE_MMU_PAGES / 1000;
1553 nr_mmu_pages = max(nr_mmu_pages,
1554 (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
1555
1556 return nr_mmu_pages;
1557 }
1558
1559 #ifdef AUDIT
1560
1561 static const char *audit_msg;
1562
1563 static gva_t canonicalize(gva_t gva)
1564 {
1565 #ifdef CONFIG_X86_64
1566 gva = (long long)(gva << 16) >> 16;
1567 #endif
1568 return gva;
1569 }
1570
1571 static void audit_mappings_page(struct kvm_vcpu *vcpu, u64 page_pte,
1572 gva_t va, int level)
1573 {
1574 u64 *pt = __va(page_pte & PT64_BASE_ADDR_MASK);
1575 int i;
1576 gva_t va_delta = 1ul << (PAGE_SHIFT + 9 * (level - 1));
1577
1578 for (i = 0; i < PT64_ENT_PER_PAGE; ++i, va += va_delta) {
1579 u64 ent = pt[i];
1580
1581 if (ent == shadow_trap_nonpresent_pte)
1582 continue;
1583
1584 va = canonicalize(va);
1585 if (level > 1) {
1586 if (ent == shadow_notrap_nonpresent_pte)
1587 printk(KERN_ERR "audit: (%s) nontrapping pte"
1588 " in nonleaf level: levels %d gva %lx"
1589 " level %d pte %llx\n", audit_msg,
1590 vcpu->mmu.root_level, va, level, ent);
1591
1592 audit_mappings_page(vcpu, ent, va, level - 1);
1593 } else {
1594 gpa_t gpa = vcpu->mmu.gva_to_gpa(vcpu, va);
1595 struct page *page = gpa_to_page(vcpu, gpa);
1596 hpa_t hpa = page_to_phys(page);
1597
1598 if (is_shadow_present_pte(ent)
1599 && (ent & PT64_BASE_ADDR_MASK) != hpa)
1600 printk(KERN_ERR "xx audit error: (%s) levels %d"
1601 " gva %lx gpa %llx hpa %llx ent %llx %d\n",
1602 audit_msg, vcpu->mmu.root_level,
1603 va, gpa, hpa, ent,
1604 is_shadow_present_pte(ent));
1605 else if (ent == shadow_notrap_nonpresent_pte
1606 && !is_error_hpa(hpa))
1607 printk(KERN_ERR "audit: (%s) notrap shadow,"
1608 " valid guest gva %lx\n", audit_msg, va);
1609 kvm_release_page_clean(page);
1610
1611 }
1612 }
1613 }
1614
1615 static void audit_mappings(struct kvm_vcpu *vcpu)
1616 {
1617 unsigned i;
1618
1619 if (vcpu->mmu.root_level == 4)
1620 audit_mappings_page(vcpu, vcpu->mmu.root_hpa, 0, 4);
1621 else
1622 for (i = 0; i < 4; ++i)
1623 if (vcpu->mmu.pae_root[i] & PT_PRESENT_MASK)
1624 audit_mappings_page(vcpu,
1625 vcpu->mmu.pae_root[i],
1626 i << 30,
1627 2);
1628 }
1629
1630 static int count_rmaps(struct kvm_vcpu *vcpu)
1631 {
1632 int nmaps = 0;
1633 int i, j, k;
1634
1635 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
1636 struct kvm_memory_slot *m = &vcpu->kvm->memslots[i];
1637 struct kvm_rmap_desc *d;
1638
1639 for (j = 0; j < m->npages; ++j) {
1640 unsigned long *rmapp = &m->rmap[j];
1641
1642 if (!*rmapp)
1643 continue;
1644 if (!(*rmapp & 1)) {
1645 ++nmaps;
1646 continue;
1647 }
1648 d = (struct kvm_rmap_desc *)(*rmapp & ~1ul);
1649 while (d) {
1650 for (k = 0; k < RMAP_EXT; ++k)
1651 if (d->shadow_ptes[k])
1652 ++nmaps;
1653 else
1654 break;
1655 d = d->more;
1656 }
1657 }
1658 }
1659 return nmaps;
1660 }
1661
1662 static int count_writable_mappings(struct kvm_vcpu *vcpu)
1663 {
1664 int nmaps = 0;
1665 struct kvm_mmu_page *sp;
1666 int i;
1667
1668 list_for_each_entry(sp, &vcpu->kvm->active_mmu_pages, link) {
1669 u64 *pt = sp->spt;
1670
1671 if (sp->role.level != PT_PAGE_TABLE_LEVEL)
1672 continue;
1673
1674 for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
1675 u64 ent = pt[i];
1676
1677 if (!(ent & PT_PRESENT_MASK))
1678 continue;
1679 if (!(ent & PT_WRITABLE_MASK))
1680 continue;
1681 ++nmaps;
1682 }
1683 }
1684 return nmaps;
1685 }
1686
1687 static void audit_rmap(struct kvm_vcpu *vcpu)
1688 {
1689 int n_rmap = count_rmaps(vcpu);
1690 int n_actual = count_writable_mappings(vcpu);
1691
1692 if (n_rmap != n_actual)
1693 printk(KERN_ERR "%s: (%s) rmap %d actual %d\n",
1694 __FUNCTION__, audit_msg, n_rmap, n_actual);
1695 }
1696
1697 static void audit_write_protection(struct kvm_vcpu *vcpu)
1698 {
1699 struct kvm_mmu_page *sp;
1700 struct kvm_memory_slot *slot;
1701 unsigned long *rmapp;
1702 gfn_t gfn;
1703
1704 list_for_each_entry(sp, &vcpu->kvm->active_mmu_pages, link) {
1705 if (sp->role.metaphysical)
1706 continue;
1707
1708 slot = gfn_to_memslot(vcpu->kvm, sp->gfn);
1709 gfn = unalias_gfn(vcpu->kvm, sp->gfn);
1710 rmapp = &slot->rmap[gfn - slot->base_gfn];
1711 if (*rmapp)
1712 printk(KERN_ERR "%s: (%s) shadow page has writable"
1713 " mappings: gfn %lx role %x\n",
1714 __FUNCTION__, audit_msg, sp->gfn,
1715 sp->role.word);
1716 }
1717 }
1718
1719 static void kvm_mmu_audit(struct kvm_vcpu *vcpu, const char *msg)
1720 {
1721 int olddbg = dbg;
1722
1723 dbg = 0;
1724 audit_msg = msg;
1725 audit_rmap(vcpu);
1726 audit_write_protection(vcpu);
1727 audit_mappings(vcpu);
1728 dbg = olddbg;
1729 }
1730
1731 #endif
Linux kernel - Deliverables
This page took 0.06425 seconds and 6 git commands to generate.