Merge branch 'for-rmk/samsung3' of git://git.fluff.org/bjdooks/linux into devel-stable
[deliverable/linux.git] / arch / x86 / kvm / paging_tmpl.h
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 /*
21 * We need the mmu code to access both 32-bit and 64-bit guest ptes,
22 * so the code in this file is compiled twice, once per pte size.
23 */
24
25 #if PTTYPE == 64
26 #define pt_element_t u64
27 #define guest_walker guest_walker64
28 #define FNAME(name) paging##64_##name
29 #define PT_BASE_ADDR_MASK PT64_BASE_ADDR_MASK
30 #define PT_LVL_ADDR_MASK(lvl) PT64_LVL_ADDR_MASK(lvl)
31 #define PT_LVL_OFFSET_MASK(lvl) PT64_LVL_OFFSET_MASK(lvl)
32 #define PT_INDEX(addr, level) PT64_INDEX(addr, level)
33 #define PT_LEVEL_MASK(level) PT64_LEVEL_MASK(level)
34 #define PT_LEVEL_BITS PT64_LEVEL_BITS
35 #ifdef CONFIG_X86_64
36 #define PT_MAX_FULL_LEVELS 4
37 #define CMPXCHG cmpxchg
38 #else
39 #define CMPXCHG cmpxchg64
40 #define PT_MAX_FULL_LEVELS 2
41 #endif
42 #elif PTTYPE == 32
43 #define pt_element_t u32
44 #define guest_walker guest_walker32
45 #define FNAME(name) paging##32_##name
46 #define PT_BASE_ADDR_MASK PT32_BASE_ADDR_MASK
47 #define PT_LVL_ADDR_MASK(lvl) PT32_LVL_ADDR_MASK(lvl)
48 #define PT_LVL_OFFSET_MASK(lvl) PT32_LVL_OFFSET_MASK(lvl)
49 #define PT_INDEX(addr, level) PT32_INDEX(addr, level)
50 #define PT_LEVEL_MASK(level) PT32_LEVEL_MASK(level)
51 #define PT_LEVEL_BITS PT32_LEVEL_BITS
52 #define PT_MAX_FULL_LEVELS 2
53 #define CMPXCHG cmpxchg
54 #else
55 #error Invalid PTTYPE value
56 #endif
57
58 #define gpte_to_gfn_lvl FNAME(gpte_to_gfn_lvl)
59 #define gpte_to_gfn(pte) gpte_to_gfn_lvl((pte), PT_PAGE_TABLE_LEVEL)
60
61 /*
62 * The guest_walker structure emulates the behavior of the hardware page
63 * table walker.
64 */
65 struct guest_walker {
66 int level;
67 gfn_t table_gfn[PT_MAX_FULL_LEVELS];
68 pt_element_t ptes[PT_MAX_FULL_LEVELS];
69 gpa_t pte_gpa[PT_MAX_FULL_LEVELS];
70 unsigned pt_access;
71 unsigned pte_access;
72 gfn_t gfn;
73 u32 error_code;
74 };
75
76 static gfn_t gpte_to_gfn_lvl(pt_element_t gpte, int lvl)
77 {
78 return (gpte & PT_LVL_ADDR_MASK(lvl)) >> PAGE_SHIFT;
79 }
80
81 static bool FNAME(cmpxchg_gpte)(struct kvm *kvm,
82 gfn_t table_gfn, unsigned index,
83 pt_element_t orig_pte, pt_element_t new_pte)
84 {
85 pt_element_t ret;
86 pt_element_t *table;
87 struct page *page;
88
89 page = gfn_to_page(kvm, table_gfn);
90
91 table = kmap_atomic(page, KM_USER0);
92 ret = CMPXCHG(&table[index], orig_pte, new_pte);
93 kunmap_atomic(table, KM_USER0);
94
95 kvm_release_page_dirty(page);
96
97 return (ret != orig_pte);
98 }
99
100 static unsigned FNAME(gpte_access)(struct kvm_vcpu *vcpu, pt_element_t gpte)
101 {
102 unsigned access;
103
104 access = (gpte & (PT_WRITABLE_MASK | PT_USER_MASK)) | ACC_EXEC_MASK;
105 #if PTTYPE == 64
106 if (is_nx(vcpu))
107 access &= ~(gpte >> PT64_NX_SHIFT);
108 #endif
109 return access;
110 }
111
112 /*
113 * Fetch a guest pte for a guest virtual address
114 */
115 static int FNAME(walk_addr)(struct guest_walker *walker,
116 struct kvm_vcpu *vcpu, gva_t addr,
117 int write_fault, int user_fault, int fetch_fault)
118 {
119 pt_element_t pte;
120 gfn_t table_gfn;
121 unsigned index, pt_access, pte_access;
122 gpa_t pte_gpa;
123 int rsvd_fault = 0;
124
125 trace_kvm_mmu_pagetable_walk(addr, write_fault, user_fault,
126 fetch_fault);
127 walk:
128 walker->level = vcpu->arch.mmu.root_level;
129 pte = vcpu->arch.cr3;
130 #if PTTYPE == 64
131 if (!is_long_mode(vcpu)) {
132 pte = kvm_pdptr_read(vcpu, (addr >> 30) & 3);
133 trace_kvm_mmu_paging_element(pte, walker->level);
134 if (!is_present_gpte(pte))
135 goto not_present;
136 --walker->level;
137 }
138 #endif
139 ASSERT((!is_long_mode(vcpu) && is_pae(vcpu)) ||
140 (vcpu->arch.cr3 & CR3_NONPAE_RESERVED_BITS) == 0);
141
142 pt_access = ACC_ALL;
143
144 for (;;) {
145 index = PT_INDEX(addr, walker->level);
146
147 table_gfn = gpte_to_gfn(pte);
148 pte_gpa = gfn_to_gpa(table_gfn);
149 pte_gpa += index * sizeof(pt_element_t);
150 walker->table_gfn[walker->level - 1] = table_gfn;
151 walker->pte_gpa[walker->level - 1] = pte_gpa;
152
153 kvm_read_guest(vcpu->kvm, pte_gpa, &pte, sizeof(pte));
154 trace_kvm_mmu_paging_element(pte, walker->level);
155
156 if (!is_present_gpte(pte))
157 goto not_present;
158
159 rsvd_fault = is_rsvd_bits_set(vcpu, pte, walker->level);
160 if (rsvd_fault)
161 goto access_error;
162
163 if (write_fault && !is_writeble_pte(pte))
164 if (user_fault || is_write_protection(vcpu))
165 goto access_error;
166
167 if (user_fault && !(pte & PT_USER_MASK))
168 goto access_error;
169
170 #if PTTYPE == 64
171 if (fetch_fault && is_nx(vcpu) && (pte & PT64_NX_MASK))
172 goto access_error;
173 #endif
174
175 if (!(pte & PT_ACCESSED_MASK)) {
176 trace_kvm_mmu_set_accessed_bit(table_gfn, index,
177 sizeof(pte));
178 mark_page_dirty(vcpu->kvm, table_gfn);
179 if (FNAME(cmpxchg_gpte)(vcpu->kvm, table_gfn,
180 index, pte, pte|PT_ACCESSED_MASK))
181 goto walk;
182 pte |= PT_ACCESSED_MASK;
183 }
184
185 pte_access = pt_access & FNAME(gpte_access)(vcpu, pte);
186
187 walker->ptes[walker->level - 1] = pte;
188
189 if ((walker->level == PT_PAGE_TABLE_LEVEL) ||
190 ((walker->level == PT_DIRECTORY_LEVEL) &&
191 (pte & PT_PAGE_SIZE_MASK) &&
192 (PTTYPE == 64 || is_pse(vcpu))) ||
193 ((walker->level == PT_PDPE_LEVEL) &&
194 (pte & PT_PAGE_SIZE_MASK) &&
195 is_long_mode(vcpu))) {
196 int lvl = walker->level;
197
198 walker->gfn = gpte_to_gfn_lvl(pte, lvl);
199 walker->gfn += (addr & PT_LVL_OFFSET_MASK(lvl))
200 >> PAGE_SHIFT;
201
202 if (PTTYPE == 32 &&
203 walker->level == PT_DIRECTORY_LEVEL &&
204 is_cpuid_PSE36())
205 walker->gfn += pse36_gfn_delta(pte);
206
207 break;
208 }
209
210 pt_access = pte_access;
211 --walker->level;
212 }
213
214 if (write_fault && !is_dirty_gpte(pte)) {
215 bool ret;
216
217 trace_kvm_mmu_set_dirty_bit(table_gfn, index, sizeof(pte));
218 mark_page_dirty(vcpu->kvm, table_gfn);
219 ret = FNAME(cmpxchg_gpte)(vcpu->kvm, table_gfn, index, pte,
220 pte|PT_DIRTY_MASK);
221 if (ret)
222 goto walk;
223 pte |= PT_DIRTY_MASK;
224 walker->ptes[walker->level - 1] = pte;
225 }
226
227 walker->pt_access = pt_access;
228 walker->pte_access = pte_access;
229 pgprintk("%s: pte %llx pte_access %x pt_access %x\n",
230 __func__, (u64)pte, pt_access, pte_access);
231 return 1;
232
233 not_present:
234 walker->error_code = 0;
235 goto err;
236
237 access_error:
238 walker->error_code = PFERR_PRESENT_MASK;
239
240 err:
241 if (write_fault)
242 walker->error_code |= PFERR_WRITE_MASK;
243 if (user_fault)
244 walker->error_code |= PFERR_USER_MASK;
245 if (fetch_fault)
246 walker->error_code |= PFERR_FETCH_MASK;
247 if (rsvd_fault)
248 walker->error_code |= PFERR_RSVD_MASK;
249 trace_kvm_mmu_walker_error(walker->error_code);
250 return 0;
251 }
252
253 static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *page,
254 u64 *spte, const void *pte)
255 {
256 pt_element_t gpte;
257 unsigned pte_access;
258 pfn_t pfn;
259
260 gpte = *(const pt_element_t *)pte;
261 if (~gpte & (PT_PRESENT_MASK | PT_ACCESSED_MASK)) {
262 if (!is_present_gpte(gpte))
263 __set_spte(spte, shadow_notrap_nonpresent_pte);
264 return;
265 }
266 pgprintk("%s: gpte %llx spte %p\n", __func__, (u64)gpte, spte);
267 pte_access = page->role.access & FNAME(gpte_access)(vcpu, gpte);
268 if (gpte_to_gfn(gpte) != vcpu->arch.update_pte.gfn)
269 return;
270 pfn = vcpu->arch.update_pte.pfn;
271 if (is_error_pfn(pfn))
272 return;
273 if (mmu_notifier_retry(vcpu, vcpu->arch.update_pte.mmu_seq))
274 return;
275 kvm_get_pfn(pfn);
276 /*
277 * we call mmu_set_spte() with reset_host_protection = true beacuse that
278 * vcpu->arch.update_pte.pfn was fetched from get_user_pages(write = 1).
279 */
280 mmu_set_spte(vcpu, spte, page->role.access, pte_access, 0, 0,
281 gpte & PT_DIRTY_MASK, NULL, PT_PAGE_TABLE_LEVEL,
282 gpte_to_gfn(gpte), pfn, true, true);
283 }
284
285 /*
286 * Fetch a shadow pte for a specific level in the paging hierarchy.
287 */
288 static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
289 struct guest_walker *gw,
290 int user_fault, int write_fault, int hlevel,
291 int *ptwrite, pfn_t pfn)
292 {
293 unsigned access = gw->pt_access;
294 struct kvm_mmu_page *shadow_page;
295 u64 spte, *sptep = NULL;
296 int direct;
297 gfn_t table_gfn;
298 int r;
299 int level;
300 pt_element_t curr_pte;
301 struct kvm_shadow_walk_iterator iterator;
302
303 if (!is_present_gpte(gw->ptes[gw->level - 1]))
304 return NULL;
305
306 for_each_shadow_entry(vcpu, addr, iterator) {
307 level = iterator.level;
308 sptep = iterator.sptep;
309 if (iterator.level == hlevel) {
310 mmu_set_spte(vcpu, sptep, access,
311 gw->pte_access & access,
312 user_fault, write_fault,
313 gw->ptes[gw->level-1] & PT_DIRTY_MASK,
314 ptwrite, level,
315 gw->gfn, pfn, false, true);
316 break;
317 }
318
319 if (is_shadow_present_pte(*sptep) && !is_large_pte(*sptep))
320 continue;
321
322 if (is_large_pte(*sptep)) {
323 rmap_remove(vcpu->kvm, sptep);
324 __set_spte(sptep, shadow_trap_nonpresent_pte);
325 kvm_flush_remote_tlbs(vcpu->kvm);
326 }
327
328 if (level <= gw->level) {
329 int delta = level - gw->level + 1;
330 direct = 1;
331 if (!is_dirty_gpte(gw->ptes[level - delta]))
332 access &= ~ACC_WRITE_MASK;
333 table_gfn = gpte_to_gfn(gw->ptes[level - delta]);
334 /* advance table_gfn when emulating 1gb pages with 4k */
335 if (delta == 0)
336 table_gfn += PT_INDEX(addr, level);
337 } else {
338 direct = 0;
339 table_gfn = gw->table_gfn[level - 2];
340 }
341 shadow_page = kvm_mmu_get_page(vcpu, table_gfn, addr, level-1,
342 direct, access, sptep);
343 if (!direct) {
344 r = kvm_read_guest_atomic(vcpu->kvm,
345 gw->pte_gpa[level - 2],
346 &curr_pte, sizeof(curr_pte));
347 if (r || curr_pte != gw->ptes[level - 2]) {
348 kvm_mmu_put_page(shadow_page, sptep);
349 kvm_release_pfn_clean(pfn);
350 sptep = NULL;
351 break;
352 }
353 }
354
355 spte = __pa(shadow_page->spt)
356 | PT_PRESENT_MASK | PT_ACCESSED_MASK
357 | PT_WRITABLE_MASK | PT_USER_MASK;
358 *sptep = spte;
359 }
360
361 return sptep;
362 }
363
364 /*
365 * Page fault handler. There are several causes for a page fault:
366 * - there is no shadow pte for the guest pte
367 * - write access through a shadow pte marked read only so that we can set
368 * the dirty bit
369 * - write access to a shadow pte marked read only so we can update the page
370 * dirty bitmap, when userspace requests it
371 * - mmio access; in this case we will never install a present shadow pte
372 * - normal guest page fault due to the guest pte marked not present, not
373 * writable, or not executable
374 *
375 * Returns: 1 if we need to emulate the instruction, 0 otherwise, or
376 * a negative value on error.
377 */
378 static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
379 u32 error_code)
380 {
381 int write_fault = error_code & PFERR_WRITE_MASK;
382 int user_fault = error_code & PFERR_USER_MASK;
383 int fetch_fault = error_code & PFERR_FETCH_MASK;
384 struct guest_walker walker;
385 u64 *sptep;
386 int write_pt = 0;
387 int r;
388 pfn_t pfn;
389 int level = PT_PAGE_TABLE_LEVEL;
390 unsigned long mmu_seq;
391
392 pgprintk("%s: addr %lx err %x\n", __func__, addr, error_code);
393 kvm_mmu_audit(vcpu, "pre page fault");
394
395 r = mmu_topup_memory_caches(vcpu);
396 if (r)
397 return r;
398
399 /*
400 * Look up the guest pte for the faulting address.
401 */
402 r = FNAME(walk_addr)(&walker, vcpu, addr, write_fault, user_fault,
403 fetch_fault);
404
405 /*
406 * The page is not mapped by the guest. Let the guest handle it.
407 */
408 if (!r) {
409 pgprintk("%s: guest page fault\n", __func__);
410 inject_page_fault(vcpu, addr, walker.error_code);
411 vcpu->arch.last_pt_write_count = 0; /* reset fork detector */
412 return 0;
413 }
414
415 if (walker.level >= PT_DIRECTORY_LEVEL) {
416 level = min(walker.level, mapping_level(vcpu, walker.gfn));
417 walker.gfn = walker.gfn & ~(KVM_PAGES_PER_HPAGE(level) - 1);
418 }
419
420 mmu_seq = vcpu->kvm->mmu_notifier_seq;
421 smp_rmb();
422 pfn = gfn_to_pfn(vcpu->kvm, walker.gfn);
423
424 /* mmio */
425 if (is_error_pfn(pfn)) {
426 pgprintk("gfn %lx is mmio\n", walker.gfn);
427 kvm_release_pfn_clean(pfn);
428 return 1;
429 }
430
431 spin_lock(&vcpu->kvm->mmu_lock);
432 if (mmu_notifier_retry(vcpu, mmu_seq))
433 goto out_unlock;
434 kvm_mmu_free_some_pages(vcpu);
435 sptep = FNAME(fetch)(vcpu, addr, &walker, user_fault, write_fault,
436 level, &write_pt, pfn);
437 pgprintk("%s: shadow pte %p %llx ptwrite %d\n", __func__,
438 sptep, *sptep, write_pt);
439
440 if (!write_pt)
441 vcpu->arch.last_pt_write_count = 0; /* reset fork detector */
442
443 ++vcpu->stat.pf_fixed;
444 kvm_mmu_audit(vcpu, "post page fault (fixed)");
445 spin_unlock(&vcpu->kvm->mmu_lock);
446
447 return write_pt;
448
449 out_unlock:
450 spin_unlock(&vcpu->kvm->mmu_lock);
451 kvm_release_pfn_clean(pfn);
452 return 0;
453 }
454
455 static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva)
456 {
457 struct kvm_shadow_walk_iterator iterator;
458 int level;
459 u64 *sptep;
460 int need_flush = 0;
461
462 spin_lock(&vcpu->kvm->mmu_lock);
463
464 for_each_shadow_entry(vcpu, gva, iterator) {
465 level = iterator.level;
466 sptep = iterator.sptep;
467
468 if (level == PT_PAGE_TABLE_LEVEL ||
469 ((level == PT_DIRECTORY_LEVEL && is_large_pte(*sptep))) ||
470 ((level == PT_PDPE_LEVEL && is_large_pte(*sptep)))) {
471
472 if (is_shadow_present_pte(*sptep)) {
473 rmap_remove(vcpu->kvm, sptep);
474 if (is_large_pte(*sptep))
475 --vcpu->kvm->stat.lpages;
476 need_flush = 1;
477 }
478 __set_spte(sptep, shadow_trap_nonpresent_pte);
479 break;
480 }
481
482 if (!is_shadow_present_pte(*sptep))
483 break;
484 }
485
486 if (need_flush)
487 kvm_flush_remote_tlbs(vcpu->kvm);
488 spin_unlock(&vcpu->kvm->mmu_lock);
489 }
490
491 static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr)
492 {
493 struct guest_walker walker;
494 gpa_t gpa = UNMAPPED_GVA;
495 int r;
496
497 r = FNAME(walk_addr)(&walker, vcpu, vaddr, 0, 0, 0);
498
499 if (r) {
500 gpa = gfn_to_gpa(walker.gfn);
501 gpa |= vaddr & ~PAGE_MASK;
502 }
503
504 return gpa;
505 }
506
507 static void FNAME(prefetch_page)(struct kvm_vcpu *vcpu,
508 struct kvm_mmu_page *sp)
509 {
510 int i, j, offset, r;
511 pt_element_t pt[256 / sizeof(pt_element_t)];
512 gpa_t pte_gpa;
513
514 if (sp->role.direct
515 || (PTTYPE == 32 && sp->role.level > PT_PAGE_TABLE_LEVEL)) {
516 nonpaging_prefetch_page(vcpu, sp);
517 return;
518 }
519
520 pte_gpa = gfn_to_gpa(sp->gfn);
521 if (PTTYPE == 32) {
522 offset = sp->role.quadrant << PT64_LEVEL_BITS;
523 pte_gpa += offset * sizeof(pt_element_t);
524 }
525
526 for (i = 0; i < PT64_ENT_PER_PAGE; i += ARRAY_SIZE(pt)) {
527 r = kvm_read_guest_atomic(vcpu->kvm, pte_gpa, pt, sizeof pt);
528 pte_gpa += ARRAY_SIZE(pt) * sizeof(pt_element_t);
529 for (j = 0; j < ARRAY_SIZE(pt); ++j)
530 if (r || is_present_gpte(pt[j]))
531 sp->spt[i+j] = shadow_trap_nonpresent_pte;
532 else
533 sp->spt[i+j] = shadow_notrap_nonpresent_pte;
534 }
535 }
536
537 /*
538 * Using the cached information from sp->gfns is safe because:
539 * - The spte has a reference to the struct page, so the pfn for a given gfn
540 * can't change unless all sptes pointing to it are nuked first.
541 * - Alias changes zap the entire shadow cache.
542 */
543 static int FNAME(sync_page)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
544 {
545 int i, offset, nr_present;
546 bool reset_host_protection;
547
548 offset = nr_present = 0;
549
550 if (PTTYPE == 32)
551 offset = sp->role.quadrant << PT64_LEVEL_BITS;
552
553 for (i = 0; i < PT64_ENT_PER_PAGE; i++) {
554 unsigned pte_access;
555 pt_element_t gpte;
556 gpa_t pte_gpa;
557 gfn_t gfn = sp->gfns[i];
558
559 if (!is_shadow_present_pte(sp->spt[i]))
560 continue;
561
562 pte_gpa = gfn_to_gpa(sp->gfn);
563 pte_gpa += (i+offset) * sizeof(pt_element_t);
564
565 if (kvm_read_guest_atomic(vcpu->kvm, pte_gpa, &gpte,
566 sizeof(pt_element_t)))
567 return -EINVAL;
568
569 if (gpte_to_gfn(gpte) != gfn || !is_present_gpte(gpte) ||
570 !(gpte & PT_ACCESSED_MASK)) {
571 u64 nonpresent;
572
573 rmap_remove(vcpu->kvm, &sp->spt[i]);
574 if (is_present_gpte(gpte))
575 nonpresent = shadow_trap_nonpresent_pte;
576 else
577 nonpresent = shadow_notrap_nonpresent_pte;
578 __set_spte(&sp->spt[i], nonpresent);
579 continue;
580 }
581
582 nr_present++;
583 pte_access = sp->role.access & FNAME(gpte_access)(vcpu, gpte);
584 if (!(sp->spt[i] & SPTE_HOST_WRITEABLE)) {
585 pte_access &= ~ACC_WRITE_MASK;
586 reset_host_protection = 0;
587 } else {
588 reset_host_protection = 1;
589 }
590 set_spte(vcpu, &sp->spt[i], pte_access, 0, 0,
591 is_dirty_gpte(gpte), PT_PAGE_TABLE_LEVEL, gfn,
592 spte_to_pfn(sp->spt[i]), true, false,
593 reset_host_protection);
594 }
595
596 return !nr_present;
597 }
598
599 #undef pt_element_t
600 #undef guest_walker
601 #undef FNAME
602 #undef PT_BASE_ADDR_MASK
603 #undef PT_INDEX
604 #undef PT_LEVEL_MASK
605 #undef PT_LVL_ADDR_MASK
606 #undef PT_LVL_OFFSET_MASK
607 #undef PT_LEVEL_BITS
608 #undef PT_MAX_FULL_LEVELS
609 #undef gpte_to_gfn
610 #undef gpte_to_gfn_lvl
611 #undef CMPXCHG
This page took 0.044777 seconds and 5 git commands to generate.