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  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11  *
12  * Authors:
13  *   Yaniv Kamay  <yaniv@qumranet.com>
14  *   Avi Kivity   <avi@qumranet.com>
15  *
16  * This work is licensed under the terms of the GNU GPL, version 2.  See
17  * the COPYING file in the top-level directory.
18  *
19  */
20 
21 /*
22  * We need the mmu code to access both 32-bit and 64-bit guest ptes,
23  * so the code in this file is compiled twice, once per pte size.
24  */
25 
26 #if PTTYPE == 64
27 	#define pt_element_t u64
28 	#define guest_walker guest_walker64
29 	#define FNAME(name) paging##64_##name
30 	#define PT_BASE_ADDR_MASK PT64_BASE_ADDR_MASK
31 	#define PT_LVL_ADDR_MASK(lvl) PT64_LVL_ADDR_MASK(lvl)
32 	#define PT_LVL_OFFSET_MASK(lvl) PT64_LVL_OFFSET_MASK(lvl)
33 	#define PT_INDEX(addr, level) PT64_INDEX(addr, 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_BITS PT32_LEVEL_BITS
51 	#define PT_MAX_FULL_LEVELS 2
52 	#define CMPXCHG cmpxchg
53 #else
54 	#error Invalid PTTYPE value
55 #endif
56 
57 #define gpte_to_gfn_lvl FNAME(gpte_to_gfn_lvl)
58 #define gpte_to_gfn(pte) gpte_to_gfn_lvl((pte), PT_PAGE_TABLE_LEVEL)
59 
60 /*
61  * The guest_walker structure emulates the behavior of the hardware page
62  * table walker.
63  */
64 struct guest_walker {
65 	int level;
66 	gfn_t table_gfn[PT_MAX_FULL_LEVELS];
67 	pt_element_t ptes[PT_MAX_FULL_LEVELS];
68 	pt_element_t prefetch_ptes[PTE_PREFETCH_NUM];
69 	gpa_t pte_gpa[PT_MAX_FULL_LEVELS];
70 	unsigned pt_access;
71 	unsigned pte_access;
72 	gfn_t gfn;
73 	struct x86_exception fault;
74 };
75 
gpte_to_gfn_lvl(pt_element_t gpte,int lvl)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 
FNAME(cmpxchg_gpte)81 static int FNAME(cmpxchg_gpte)(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
82 			       pt_element_t __user *ptep_user, unsigned index,
83 			       pt_element_t orig_pte, pt_element_t new_pte)
84 {
85 	int npages;
86 	pt_element_t ret;
87 	pt_element_t *table;
88 	struct page *page;
89 
90 	npages = get_user_pages_fast((unsigned long)ptep_user, 1, 1, &page);
91 	/* Check if the user is doing something meaningless. */
92 	if (unlikely(npages != 1))
93 		return -EFAULT;
94 
95 	table = kmap_atomic(page);
96 	ret = CMPXCHG(&table[index], orig_pte, new_pte);
97 	kunmap_atomic(table);
98 
99 	kvm_release_page_dirty(page);
100 
101 	return (ret != orig_pte);
102 }
103 
FNAME(gpte_access)104 static unsigned FNAME(gpte_access)(struct kvm_vcpu *vcpu, pt_element_t gpte,
105 				   bool last)
106 {
107 	unsigned access;
108 
109 	access = (gpte & (PT_WRITABLE_MASK | PT_USER_MASK)) | ACC_EXEC_MASK;
110 	if (last && !is_dirty_gpte(gpte))
111 		access &= ~ACC_WRITE_MASK;
112 
113 #if PTTYPE == 64
114 	if (vcpu->arch.mmu.nx)
115 		access &= ~(gpte >> PT64_NX_SHIFT);
116 #endif
117 	return access;
118 }
119 
FNAME(is_last_gpte)120 static bool FNAME(is_last_gpte)(struct guest_walker *walker,
121 				struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
122 				pt_element_t gpte)
123 {
124 	if (walker->level == PT_PAGE_TABLE_LEVEL)
125 		return true;
126 
127 	if ((walker->level == PT_DIRECTORY_LEVEL) && is_large_pte(gpte) &&
128 	    (PTTYPE == 64 || is_pse(vcpu)))
129 		return true;
130 
131 	if ((walker->level == PT_PDPE_LEVEL) && is_large_pte(gpte) &&
132 	    (mmu->root_level == PT64_ROOT_LEVEL))
133 		return true;
134 
135 	return false;
136 }
137 
138 /*
139  * Fetch a guest pte for a guest virtual address
140  */
FNAME(walk_addr_generic)141 static int FNAME(walk_addr_generic)(struct guest_walker *walker,
142 				    struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
143 				    gva_t addr, u32 access)
144 {
145 	pt_element_t pte;
146 	pt_element_t __user *uninitialized_var(ptep_user);
147 	gfn_t table_gfn;
148 	unsigned index, pt_access, uninitialized_var(pte_access);
149 	gpa_t pte_gpa;
150 	bool eperm, last_gpte;
151 	int offset;
152 	const int write_fault = access & PFERR_WRITE_MASK;
153 	const int user_fault  = access & PFERR_USER_MASK;
154 	const int fetch_fault = access & PFERR_FETCH_MASK;
155 	u16 errcode = 0;
156 
157 	trace_kvm_mmu_pagetable_walk(addr, write_fault, user_fault,
158 				     fetch_fault);
159 retry_walk:
160 	eperm = false;
161 	walker->level = mmu->root_level;
162 	pte           = mmu->get_cr3(vcpu);
163 
164 #if PTTYPE == 64
165 	if (walker->level == PT32E_ROOT_LEVEL) {
166 		pte = mmu->get_pdptr(vcpu, (addr >> 30) & 3);
167 		trace_kvm_mmu_paging_element(pte, walker->level);
168 		if (!is_present_gpte(pte))
169 			goto error;
170 		--walker->level;
171 	}
172 #endif
173 	ASSERT((!is_long_mode(vcpu) && is_pae(vcpu)) ||
174 	       (mmu->get_cr3(vcpu) & CR3_NONPAE_RESERVED_BITS) == 0);
175 
176 	pt_access = ACC_ALL;
177 
178 	for (;;) {
179 		gfn_t real_gfn;
180 		unsigned long host_addr;
181 
182 		index = PT_INDEX(addr, walker->level);
183 
184 		table_gfn = gpte_to_gfn(pte);
185 		offset    = index * sizeof(pt_element_t);
186 		pte_gpa   = gfn_to_gpa(table_gfn) + offset;
187 		walker->table_gfn[walker->level - 1] = table_gfn;
188 		walker->pte_gpa[walker->level - 1] = pte_gpa;
189 
190 		real_gfn = mmu->translate_gpa(vcpu, gfn_to_gpa(table_gfn),
191 					      PFERR_USER_MASK|PFERR_WRITE_MASK);
192 		if (unlikely(real_gfn == UNMAPPED_GVA))
193 			goto error;
194 		real_gfn = gpa_to_gfn(real_gfn);
195 
196 		host_addr = gfn_to_hva(vcpu->kvm, real_gfn);
197 		if (unlikely(kvm_is_error_hva(host_addr)))
198 			goto error;
199 
200 		ptep_user = (pt_element_t __user *)((void *)host_addr + offset);
201 		if (unlikely(__copy_from_user(&pte, ptep_user, sizeof(pte))))
202 			goto error;
203 
204 		trace_kvm_mmu_paging_element(pte, walker->level);
205 
206 		if (unlikely(!is_present_gpte(pte)))
207 			goto error;
208 
209 		if (unlikely(is_rsvd_bits_set(&vcpu->arch.mmu, pte,
210 					      walker->level))) {
211 			errcode |= PFERR_RSVD_MASK | PFERR_PRESENT_MASK;
212 			goto error;
213 		}
214 
215 		if (!check_write_user_access(vcpu, write_fault, user_fault,
216 					  pte))
217 			eperm = true;
218 
219 #if PTTYPE == 64
220 		if (unlikely(fetch_fault && (pte & PT64_NX_MASK)))
221 			eperm = true;
222 #endif
223 
224 		last_gpte = FNAME(is_last_gpte)(walker, vcpu, mmu, pte);
225 		if (last_gpte) {
226 			pte_access = pt_access &
227 				     FNAME(gpte_access)(vcpu, pte, true);
228 			/* check if the kernel is fetching from user page */
229 			if (unlikely(pte_access & PT_USER_MASK) &&
230 			    kvm_read_cr4_bits(vcpu, X86_CR4_SMEP))
231 				if (fetch_fault && !user_fault)
232 					eperm = true;
233 		}
234 
235 		if (!eperm && unlikely(!(pte & PT_ACCESSED_MASK))) {
236 			int ret;
237 			trace_kvm_mmu_set_accessed_bit(table_gfn, index,
238 						       sizeof(pte));
239 			ret = FNAME(cmpxchg_gpte)(vcpu, mmu, ptep_user, index,
240 						  pte, pte|PT_ACCESSED_MASK);
241 			if (unlikely(ret < 0))
242 				goto error;
243 			else if (ret)
244 				goto retry_walk;
245 
246 			mark_page_dirty(vcpu->kvm, table_gfn);
247 			pte |= PT_ACCESSED_MASK;
248 		}
249 
250 		walker->ptes[walker->level - 1] = pte;
251 
252 		if (last_gpte) {
253 			int lvl = walker->level;
254 			gpa_t real_gpa;
255 			gfn_t gfn;
256 			u32 ac;
257 
258 			gfn = gpte_to_gfn_lvl(pte, lvl);
259 			gfn += (addr & PT_LVL_OFFSET_MASK(lvl)) >> PAGE_SHIFT;
260 
261 			if (PTTYPE == 32 &&
262 			    walker->level == PT_DIRECTORY_LEVEL &&
263 			    is_cpuid_PSE36())
264 				gfn += pse36_gfn_delta(pte);
265 
266 			ac = write_fault | fetch_fault | user_fault;
267 
268 			real_gpa = mmu->translate_gpa(vcpu, gfn_to_gpa(gfn),
269 						      ac);
270 			if (real_gpa == UNMAPPED_GVA)
271 				return 0;
272 
273 			walker->gfn = real_gpa >> PAGE_SHIFT;
274 
275 			break;
276 		}
277 
278 		pt_access &= FNAME(gpte_access)(vcpu, pte, false);
279 		--walker->level;
280 	}
281 
282 	if (unlikely(eperm)) {
283 		errcode |= PFERR_PRESENT_MASK;
284 		goto error;
285 	}
286 
287 	if (write_fault && unlikely(!is_dirty_gpte(pte))) {
288 		int ret;
289 
290 		trace_kvm_mmu_set_dirty_bit(table_gfn, index, sizeof(pte));
291 		ret = FNAME(cmpxchg_gpte)(vcpu, mmu, ptep_user, index,
292 					  pte, pte|PT_DIRTY_MASK);
293 		if (unlikely(ret < 0))
294 			goto error;
295 		else if (ret)
296 			goto retry_walk;
297 
298 		mark_page_dirty(vcpu->kvm, table_gfn);
299 		pte |= PT_DIRTY_MASK;
300 		walker->ptes[walker->level - 1] = pte;
301 	}
302 
303 	walker->pt_access = pt_access;
304 	walker->pte_access = pte_access;
305 	pgprintk("%s: pte %llx pte_access %x pt_access %x\n",
306 		 __func__, (u64)pte, pte_access, pt_access);
307 	return 1;
308 
309 error:
310 	errcode |= write_fault | user_fault;
311 	if (fetch_fault && (mmu->nx ||
312 			    kvm_read_cr4_bits(vcpu, X86_CR4_SMEP)))
313 		errcode |= PFERR_FETCH_MASK;
314 
315 	walker->fault.vector = PF_VECTOR;
316 	walker->fault.error_code_valid = true;
317 	walker->fault.error_code = errcode;
318 	walker->fault.address = addr;
319 	walker->fault.nested_page_fault = mmu != vcpu->arch.walk_mmu;
320 
321 	trace_kvm_mmu_walker_error(walker->fault.error_code);
322 	return 0;
323 }
324 
FNAME(walk_addr)325 static int FNAME(walk_addr)(struct guest_walker *walker,
326 			    struct kvm_vcpu *vcpu, gva_t addr, u32 access)
327 {
328 	return FNAME(walk_addr_generic)(walker, vcpu, &vcpu->arch.mmu, addr,
329 					access);
330 }
331 
FNAME(walk_addr_nested)332 static int FNAME(walk_addr_nested)(struct guest_walker *walker,
333 				   struct kvm_vcpu *vcpu, gva_t addr,
334 				   u32 access)
335 {
336 	return FNAME(walk_addr_generic)(walker, vcpu, &vcpu->arch.nested_mmu,
337 					addr, access);
338 }
339 
FNAME(prefetch_invalid_gpte)340 static bool FNAME(prefetch_invalid_gpte)(struct kvm_vcpu *vcpu,
341 				    struct kvm_mmu_page *sp, u64 *spte,
342 				    pt_element_t gpte)
343 {
344 	if (is_rsvd_bits_set(&vcpu->arch.mmu, gpte, PT_PAGE_TABLE_LEVEL))
345 		goto no_present;
346 
347 	if (!is_present_gpte(gpte))
348 		goto no_present;
349 
350 	if (!(gpte & PT_ACCESSED_MASK))
351 		goto no_present;
352 
353 	return false;
354 
355 no_present:
356 	drop_spte(vcpu->kvm, spte);
357 	return true;
358 }
359 
FNAME(update_pte)360 static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
361 			      u64 *spte, const void *pte)
362 {
363 	pt_element_t gpte;
364 	unsigned pte_access;
365 	pfn_t pfn;
366 
367 	gpte = *(const pt_element_t *)pte;
368 	if (FNAME(prefetch_invalid_gpte)(vcpu, sp, spte, gpte))
369 		return;
370 
371 	pgprintk("%s: gpte %llx spte %p\n", __func__, (u64)gpte, spte);
372 	pte_access = sp->role.access & FNAME(gpte_access)(vcpu, gpte, true);
373 	pfn = gfn_to_pfn_atomic(vcpu->kvm, gpte_to_gfn(gpte));
374 	if (mmu_invalid_pfn(pfn)) {
375 		kvm_release_pfn_clean(pfn);
376 		return;
377 	}
378 
379 	/*
380 	 * we call mmu_set_spte() with host_writable = true because that
381 	 * vcpu->arch.update_pte.pfn was fetched from get_user_pages(write = 1).
382 	 */
383 	mmu_set_spte(vcpu, spte, sp->role.access, pte_access, 0, 0,
384 		     NULL, PT_PAGE_TABLE_LEVEL,
385 		     gpte_to_gfn(gpte), pfn, true, true);
386 }
387 
FNAME(gpte_changed)388 static bool FNAME(gpte_changed)(struct kvm_vcpu *vcpu,
389 				struct guest_walker *gw, int level)
390 {
391 	pt_element_t curr_pte;
392 	gpa_t base_gpa, pte_gpa = gw->pte_gpa[level - 1];
393 	u64 mask;
394 	int r, index;
395 
396 	if (level == PT_PAGE_TABLE_LEVEL) {
397 		mask = PTE_PREFETCH_NUM * sizeof(pt_element_t) - 1;
398 		base_gpa = pte_gpa & ~mask;
399 		index = (pte_gpa - base_gpa) / sizeof(pt_element_t);
400 
401 		r = kvm_read_guest_atomic(vcpu->kvm, base_gpa,
402 				gw->prefetch_ptes, sizeof(gw->prefetch_ptes));
403 		curr_pte = gw->prefetch_ptes[index];
404 	} else
405 		r = kvm_read_guest_atomic(vcpu->kvm, pte_gpa,
406 				  &curr_pte, sizeof(curr_pte));
407 
408 	return r || curr_pte != gw->ptes[level - 1];
409 }
410 
FNAME(pte_prefetch)411 static void FNAME(pte_prefetch)(struct kvm_vcpu *vcpu, struct guest_walker *gw,
412 				u64 *sptep)
413 {
414 	struct kvm_mmu_page *sp;
415 	pt_element_t *gptep = gw->prefetch_ptes;
416 	u64 *spte;
417 	int i;
418 
419 	sp = page_header(__pa(sptep));
420 
421 	if (sp->role.level > PT_PAGE_TABLE_LEVEL)
422 		return;
423 
424 	if (sp->role.direct)
425 		return __direct_pte_prefetch(vcpu, sp, sptep);
426 
427 	i = (sptep - sp->spt) & ~(PTE_PREFETCH_NUM - 1);
428 	spte = sp->spt + i;
429 
430 	for (i = 0; i < PTE_PREFETCH_NUM; i++, spte++) {
431 		pt_element_t gpte;
432 		unsigned pte_access;
433 		gfn_t gfn;
434 		pfn_t pfn;
435 
436 		if (spte == sptep)
437 			continue;
438 
439 		if (is_shadow_present_pte(*spte))
440 			continue;
441 
442 		gpte = gptep[i];
443 
444 		if (FNAME(prefetch_invalid_gpte)(vcpu, sp, spte, gpte))
445 			continue;
446 
447 		pte_access = sp->role.access & FNAME(gpte_access)(vcpu, gpte,
448 								  true);
449 		gfn = gpte_to_gfn(gpte);
450 		pfn = pte_prefetch_gfn_to_pfn(vcpu, gfn,
451 				      pte_access & ACC_WRITE_MASK);
452 		if (mmu_invalid_pfn(pfn)) {
453 			kvm_release_pfn_clean(pfn);
454 			break;
455 		}
456 
457 		mmu_set_spte(vcpu, spte, sp->role.access, pte_access, 0, 0,
458 			     NULL, PT_PAGE_TABLE_LEVEL, gfn,
459 			     pfn, true, true);
460 	}
461 }
462 
463 /*
464  * Fetch a shadow pte for a specific level in the paging hierarchy.
465  */
FNAME(fetch)466 static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
467 			 struct guest_walker *gw,
468 			 int user_fault, int write_fault, int hlevel,
469 			 int *emulate, pfn_t pfn, bool map_writable,
470 			 bool prefault)
471 {
472 	unsigned access = gw->pt_access;
473 	struct kvm_mmu_page *sp = NULL;
474 	int top_level;
475 	unsigned direct_access;
476 	struct kvm_shadow_walk_iterator it;
477 
478 	if (!is_present_gpte(gw->ptes[gw->level - 1]))
479 		return NULL;
480 
481 	direct_access = gw->pte_access;
482 
483 	top_level = vcpu->arch.mmu.root_level;
484 	if (top_level == PT32E_ROOT_LEVEL)
485 		top_level = PT32_ROOT_LEVEL;
486 	/*
487 	 * Verify that the top-level gpte is still there.  Since the page
488 	 * is a root page, it is either write protected (and cannot be
489 	 * changed from now on) or it is invalid (in which case, we don't
490 	 * really care if it changes underneath us after this point).
491 	 */
492 	if (FNAME(gpte_changed)(vcpu, gw, top_level))
493 		goto out_gpte_changed;
494 
495 	for (shadow_walk_init(&it, vcpu, addr);
496 	     shadow_walk_okay(&it) && it.level > gw->level;
497 	     shadow_walk_next(&it)) {
498 		gfn_t table_gfn;
499 
500 		clear_sp_write_flooding_count(it.sptep);
501 		drop_large_spte(vcpu, it.sptep);
502 
503 		sp = NULL;
504 		if (!is_shadow_present_pte(*it.sptep)) {
505 			table_gfn = gw->table_gfn[it.level - 2];
506 			sp = kvm_mmu_get_page(vcpu, table_gfn, addr, it.level-1,
507 					      false, access, it.sptep);
508 		}
509 
510 		/*
511 		 * Verify that the gpte in the page we've just write
512 		 * protected is still there.
513 		 */
514 		if (FNAME(gpte_changed)(vcpu, gw, it.level - 1))
515 			goto out_gpte_changed;
516 
517 		if (sp)
518 			link_shadow_page(it.sptep, sp);
519 	}
520 
521 	for (;
522 	     shadow_walk_okay(&it) && it.level > hlevel;
523 	     shadow_walk_next(&it)) {
524 		gfn_t direct_gfn;
525 
526 		clear_sp_write_flooding_count(it.sptep);
527 		validate_direct_spte(vcpu, it.sptep, direct_access);
528 
529 		drop_large_spte(vcpu, it.sptep);
530 
531 		if (is_shadow_present_pte(*it.sptep))
532 			continue;
533 
534 		direct_gfn = gw->gfn & ~(KVM_PAGES_PER_HPAGE(it.level) - 1);
535 
536 		sp = kvm_mmu_get_page(vcpu, direct_gfn, addr, it.level-1,
537 				      true, direct_access, it.sptep);
538 		link_shadow_page(it.sptep, sp);
539 	}
540 
541 	clear_sp_write_flooding_count(it.sptep);
542 	mmu_set_spte(vcpu, it.sptep, access, gw->pte_access,
543 		     user_fault, write_fault, emulate, it.level,
544 		     gw->gfn, pfn, prefault, map_writable);
545 	FNAME(pte_prefetch)(vcpu, gw, it.sptep);
546 
547 	return it.sptep;
548 
549 out_gpte_changed:
550 	if (sp)
551 		kvm_mmu_put_page(sp, it.sptep);
552 	kvm_release_pfn_clean(pfn);
553 	return NULL;
554 }
555 
556 /*
557  * Page fault handler.  There are several causes for a page fault:
558  *   - there is no shadow pte for the guest pte
559  *   - write access through a shadow pte marked read only so that we can set
560  *     the dirty bit
561  *   - write access to a shadow pte marked read only so we can update the page
562  *     dirty bitmap, when userspace requests it
563  *   - mmio access; in this case we will never install a present shadow pte
564  *   - normal guest page fault due to the guest pte marked not present, not
565  *     writable, or not executable
566  *
567  *  Returns: 1 if we need to emulate the instruction, 0 otherwise, or
568  *           a negative value on error.
569  */
FNAME(page_fault)570 static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr, u32 error_code,
571 			     bool prefault)
572 {
573 	int write_fault = error_code & PFERR_WRITE_MASK;
574 	int user_fault = error_code & PFERR_USER_MASK;
575 	struct guest_walker walker;
576 	u64 *sptep;
577 	int emulate = 0;
578 	int r;
579 	pfn_t pfn;
580 	int level = PT_PAGE_TABLE_LEVEL;
581 	int force_pt_level;
582 	unsigned long mmu_seq;
583 	bool map_writable;
584 
585 	pgprintk("%s: addr %lx err %x\n", __func__, addr, error_code);
586 
587 	if (unlikely(error_code & PFERR_RSVD_MASK))
588 		return handle_mmio_page_fault(vcpu, addr, error_code,
589 					      mmu_is_nested(vcpu));
590 
591 	r = mmu_topup_memory_caches(vcpu);
592 	if (r)
593 		return r;
594 
595 	/*
596 	 * Look up the guest pte for the faulting address.
597 	 */
598 	r = FNAME(walk_addr)(&walker, vcpu, addr, error_code);
599 
600 	/*
601 	 * The page is not mapped by the guest.  Let the guest handle it.
602 	 */
603 	if (!r) {
604 		pgprintk("%s: guest page fault\n", __func__);
605 		if (!prefault)
606 			inject_page_fault(vcpu, &walker.fault);
607 
608 		return 0;
609 	}
610 
611 	if (walker.level >= PT_DIRECTORY_LEVEL)
612 		force_pt_level = mapping_level_dirty_bitmap(vcpu, walker.gfn);
613 	else
614 		force_pt_level = 1;
615 	if (!force_pt_level) {
616 		level = min(walker.level, mapping_level(vcpu, walker.gfn));
617 		walker.gfn = walker.gfn & ~(KVM_PAGES_PER_HPAGE(level) - 1);
618 	}
619 
620 	mmu_seq = vcpu->kvm->mmu_notifier_seq;
621 	smp_rmb();
622 
623 	if (try_async_pf(vcpu, prefault, walker.gfn, addr, &pfn, write_fault,
624 			 &map_writable))
625 		return 0;
626 
627 	if (handle_abnormal_pfn(vcpu, mmu_is_nested(vcpu) ? 0 : addr,
628 				walker.gfn, pfn, walker.pte_access, &r))
629 		return r;
630 
631 	spin_lock(&vcpu->kvm->mmu_lock);
632 	if (mmu_notifier_retry(vcpu, mmu_seq))
633 		goto out_unlock;
634 
635 	kvm_mmu_audit(vcpu, AUDIT_PRE_PAGE_FAULT);
636 	kvm_mmu_free_some_pages(vcpu);
637 	if (!force_pt_level)
638 		transparent_hugepage_adjust(vcpu, &walker.gfn, &pfn, &level);
639 	sptep = FNAME(fetch)(vcpu, addr, &walker, user_fault, write_fault,
640 			     level, &emulate, pfn, map_writable, prefault);
641 	(void)sptep;
642 	pgprintk("%s: shadow pte %p %llx emulate %d\n", __func__,
643 		 sptep, *sptep, emulate);
644 
645 	++vcpu->stat.pf_fixed;
646 	kvm_mmu_audit(vcpu, AUDIT_POST_PAGE_FAULT);
647 	spin_unlock(&vcpu->kvm->mmu_lock);
648 
649 	return emulate;
650 
651 out_unlock:
652 	spin_unlock(&vcpu->kvm->mmu_lock);
653 	kvm_release_pfn_clean(pfn);
654 	return 0;
655 }
656 
FNAME(get_level1_sp_gpa)657 static gpa_t FNAME(get_level1_sp_gpa)(struct kvm_mmu_page *sp)
658 {
659 	int offset = 0;
660 
661 	WARN_ON(sp->role.level != 1);
662 
663 	if (PTTYPE == 32)
664 		offset = sp->role.quadrant << PT64_LEVEL_BITS;
665 
666 	return gfn_to_gpa(sp->gfn) + offset * sizeof(pt_element_t);
667 }
668 
FNAME(invlpg)669 static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva)
670 {
671 	struct kvm_shadow_walk_iterator iterator;
672 	struct kvm_mmu_page *sp;
673 	int level;
674 	u64 *sptep;
675 
676 	vcpu_clear_mmio_info(vcpu, gva);
677 
678 	/*
679 	 * No need to check return value here, rmap_can_add() can
680 	 * help us to skip pte prefetch later.
681 	 */
682 	mmu_topup_memory_caches(vcpu);
683 
684 	spin_lock(&vcpu->kvm->mmu_lock);
685 	for_each_shadow_entry(vcpu, gva, iterator) {
686 		level = iterator.level;
687 		sptep = iterator.sptep;
688 
689 		sp = page_header(__pa(sptep));
690 		if (is_last_spte(*sptep, level)) {
691 			pt_element_t gpte;
692 			gpa_t pte_gpa;
693 
694 			if (!sp->unsync)
695 				break;
696 
697 			pte_gpa = FNAME(get_level1_sp_gpa)(sp);
698 			pte_gpa += (sptep - sp->spt) * sizeof(pt_element_t);
699 
700 			if (mmu_page_zap_pte(vcpu->kvm, sp, sptep))
701 				kvm_flush_remote_tlbs(vcpu->kvm);
702 
703 			if (!rmap_can_add(vcpu))
704 				break;
705 
706 			if (kvm_read_guest_atomic(vcpu->kvm, pte_gpa, &gpte,
707 						  sizeof(pt_element_t)))
708 				break;
709 
710 			FNAME(update_pte)(vcpu, sp, sptep, &gpte);
711 		}
712 
713 		if (!is_shadow_present_pte(*sptep) || !sp->unsync_children)
714 			break;
715 	}
716 	spin_unlock(&vcpu->kvm->mmu_lock);
717 }
718 
FNAME(gva_to_gpa)719 static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr, u32 access,
720 			       struct x86_exception *exception)
721 {
722 	struct guest_walker walker;
723 	gpa_t gpa = UNMAPPED_GVA;
724 	int r;
725 
726 	r = FNAME(walk_addr)(&walker, vcpu, vaddr, access);
727 
728 	if (r) {
729 		gpa = gfn_to_gpa(walker.gfn);
730 		gpa |= vaddr & ~PAGE_MASK;
731 	} else if (exception)
732 		*exception = walker.fault;
733 
734 	return gpa;
735 }
736 
FNAME(gva_to_gpa_nested)737 static gpa_t FNAME(gva_to_gpa_nested)(struct kvm_vcpu *vcpu, gva_t vaddr,
738 				      u32 access,
739 				      struct x86_exception *exception)
740 {
741 	struct guest_walker walker;
742 	gpa_t gpa = UNMAPPED_GVA;
743 	int r;
744 
745 	r = FNAME(walk_addr_nested)(&walker, vcpu, vaddr, access);
746 
747 	if (r) {
748 		gpa = gfn_to_gpa(walker.gfn);
749 		gpa |= vaddr & ~PAGE_MASK;
750 	} else if (exception)
751 		*exception = walker.fault;
752 
753 	return gpa;
754 }
755 
756 /*
757  * Using the cached information from sp->gfns is safe because:
758  * - The spte has a reference to the struct page, so the pfn for a given gfn
759  *   can't change unless all sptes pointing to it are nuked first.
760  *
761  * Note:
762  *   We should flush all tlbs if spte is dropped even though guest is
763  *   responsible for it. Since if we don't, kvm_mmu_notifier_invalidate_page
764  *   and kvm_mmu_notifier_invalidate_range_start detect the mapping page isn't
765  *   used by guest then tlbs are not flushed, so guest is allowed to access the
766  *   freed pages.
767  *   And we increase kvm->tlbs_dirty to delay tlbs flush in this case.
768  */
FNAME(sync_page)769 static int FNAME(sync_page)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
770 {
771 	int i, nr_present = 0;
772 	bool host_writable;
773 	gpa_t first_pte_gpa;
774 
775 	/* direct kvm_mmu_page can not be unsync. */
776 	BUG_ON(sp->role.direct);
777 
778 	first_pte_gpa = FNAME(get_level1_sp_gpa)(sp);
779 
780 	for (i = 0; i < PT64_ENT_PER_PAGE; i++) {
781 		unsigned pte_access;
782 		pt_element_t gpte;
783 		gpa_t pte_gpa;
784 		gfn_t gfn;
785 
786 		if (!sp->spt[i])
787 			continue;
788 
789 		pte_gpa = first_pte_gpa + i * sizeof(pt_element_t);
790 
791 		if (kvm_read_guest_atomic(vcpu->kvm, pte_gpa, &gpte,
792 					  sizeof(pt_element_t)))
793 			return -EINVAL;
794 
795 		if (FNAME(prefetch_invalid_gpte)(vcpu, sp, &sp->spt[i], gpte)) {
796 			vcpu->kvm->tlbs_dirty++;
797 			continue;
798 		}
799 
800 		gfn = gpte_to_gfn(gpte);
801 		pte_access = sp->role.access;
802 		pte_access &= FNAME(gpte_access)(vcpu, gpte, true);
803 
804 		if (sync_mmio_spte(&sp->spt[i], gfn, pte_access, &nr_present))
805 			continue;
806 
807 		if (gfn != sp->gfns[i]) {
808 			drop_spte(vcpu->kvm, &sp->spt[i]);
809 			vcpu->kvm->tlbs_dirty++;
810 			continue;
811 		}
812 
813 		nr_present++;
814 
815 		host_writable = sp->spt[i] & SPTE_HOST_WRITEABLE;
816 
817 		set_spte(vcpu, &sp->spt[i], pte_access, 0, 0,
818 			 PT_PAGE_TABLE_LEVEL, gfn,
819 			 spte_to_pfn(sp->spt[i]), true, false,
820 			 host_writable);
821 	}
822 
823 	return !nr_present;
824 }
825 
826 #undef pt_element_t
827 #undef guest_walker
828 #undef FNAME
829 #undef PT_BASE_ADDR_MASK
830 #undef PT_INDEX
831 #undef PT_LVL_ADDR_MASK
832 #undef PT_LVL_OFFSET_MASK
833 #undef PT_LEVEL_BITS
834 #undef PT_MAX_FULL_LEVELS
835 #undef gpte_to_gfn
836 #undef gpte_to_gfn_lvl
837 #undef CMPXCHG
838