2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
9 * Copyright (C) 2006 Qumranet, Inc.
10 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
13 * Yaniv Kamay <yaniv@qumranet.com>
14 * Avi Kivity <avi@qumranet.com>
16 * This work is licensed under the terms of the GNU GPL, version 2. See
17 * the COPYING file in the top-level directory.
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.
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
36 #define PT_MAX_FULL_LEVELS 4
37 #define CMPXCHG cmpxchg
39 #define CMPXCHG cmpxchg64
40 #define PT_MAX_FULL_LEVELS 2
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
54 #error Invalid PTTYPE value
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)
61 * The guest_walker structure emulates the behavior of the hardware page
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];
73 struct x86_exception fault;
76 static gfn_t gpte_to_gfn_lvl(pt_element_t gpte, int lvl)
78 return (gpte & PT_LVL_ADDR_MASK(lvl)) >> PAGE_SHIFT;
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)
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))
95 table = kmap_atomic(page, KM_USER0);
96 ret = CMPXCHG(&table[index], orig_pte, new_pte);
97 kunmap_atomic(table, KM_USER0);
99 kvm_release_page_dirty(page);
101 return (ret != orig_pte);
104 static unsigned FNAME(gpte_access)(struct kvm_vcpu *vcpu, pt_element_t gpte)
108 access = (gpte & (PT_WRITABLE_MASK | PT_USER_MASK)) | ACC_EXEC_MASK;
110 if (vcpu->arch.mmu.nx)
111 access &= ~(gpte >> PT64_NX_SHIFT);
117 * Fetch a guest pte for a guest virtual address
119 static int FNAME(walk_addr_generic)(struct guest_walker *walker,
120 struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
121 gva_t addr, u32 access)
124 pt_element_t __user *uninitialized_var(ptep_user);
126 unsigned index, pt_access, uninitialized_var(pte_access);
128 bool eperm, present, rsvd_fault;
129 int offset, write_fault, user_fault, fetch_fault;
131 write_fault = access & PFERR_WRITE_MASK;
132 user_fault = access & PFERR_USER_MASK;
133 fetch_fault = access & PFERR_FETCH_MASK;
135 trace_kvm_mmu_pagetable_walk(addr, write_fault, user_fault,
139 eperm = rsvd_fault = false;
140 walker->level = mmu->root_level;
141 pte = mmu->get_cr3(vcpu);
144 if (walker->level == PT32E_ROOT_LEVEL) {
145 pte = kvm_pdptr_read_mmu(vcpu, mmu, (addr >> 30) & 3);
146 trace_kvm_mmu_paging_element(pte, walker->level);
147 if (!is_present_gpte(pte)) {
154 ASSERT((!is_long_mode(vcpu) && is_pae(vcpu)) ||
155 (mmu->get_cr3(vcpu) & CR3_NONPAE_RESERVED_BITS) == 0);
161 unsigned long host_addr;
163 index = PT_INDEX(addr, walker->level);
165 table_gfn = gpte_to_gfn(pte);
166 offset = index * sizeof(pt_element_t);
167 pte_gpa = gfn_to_gpa(table_gfn) + offset;
168 walker->table_gfn[walker->level - 1] = table_gfn;
169 walker->pte_gpa[walker->level - 1] = pte_gpa;
171 real_gfn = mmu->translate_gpa(vcpu, gfn_to_gpa(table_gfn),
172 PFERR_USER_MASK|PFERR_WRITE_MASK);
173 if (unlikely(real_gfn == UNMAPPED_GVA)) {
177 real_gfn = gpa_to_gfn(real_gfn);
179 host_addr = gfn_to_hva(vcpu->kvm, real_gfn);
180 if (unlikely(kvm_is_error_hva(host_addr))) {
185 ptep_user = (pt_element_t __user *)((void *)host_addr + offset);
186 if (unlikely(__copy_from_user(&pte, ptep_user, sizeof(pte)))) {
191 trace_kvm_mmu_paging_element(pte, walker->level);
193 if (unlikely(!is_present_gpte(pte))) {
198 if (unlikely(is_rsvd_bits_set(&vcpu->arch.mmu, pte,
204 if (unlikely(write_fault && !is_writable_pte(pte)
205 && (user_fault || is_write_protection(vcpu))))
208 if (unlikely(user_fault && !(pte & PT_USER_MASK)))
212 if (unlikely(fetch_fault && (pte & PT64_NX_MASK)))
216 if (!eperm && !rsvd_fault
217 && unlikely(!(pte & PT_ACCESSED_MASK))) {
219 trace_kvm_mmu_set_accessed_bit(table_gfn, index,
221 ret = FNAME(cmpxchg_gpte)(vcpu, mmu, ptep_user, index,
222 pte, pte|PT_ACCESSED_MASK);
223 if (unlikely(ret < 0)) {
229 mark_page_dirty(vcpu->kvm, table_gfn);
230 pte |= PT_ACCESSED_MASK;
233 pte_access = pt_access & FNAME(gpte_access)(vcpu, pte);
235 walker->ptes[walker->level - 1] = pte;
237 if ((walker->level == PT_PAGE_TABLE_LEVEL) ||
238 ((walker->level == PT_DIRECTORY_LEVEL) &&
240 (PTTYPE == 64 || is_pse(vcpu))) ||
241 ((walker->level == PT_PDPE_LEVEL) &&
243 mmu->root_level == PT64_ROOT_LEVEL)) {
244 int lvl = walker->level;
249 /* check if the kernel is fetching from user page */
250 if (unlikely(pte_access & PT_USER_MASK) &&
251 kvm_read_cr4_bits(vcpu, X86_CR4_SMEP))
252 if (fetch_fault && !user_fault)
255 gfn = gpte_to_gfn_lvl(pte, lvl);
256 gfn += (addr & PT_LVL_OFFSET_MASK(lvl)) >> PAGE_SHIFT;
259 walker->level == PT_DIRECTORY_LEVEL &&
261 gfn += pse36_gfn_delta(pte);
263 ac = write_fault | fetch_fault | user_fault;
265 real_gpa = mmu->translate_gpa(vcpu, gfn_to_gpa(gfn),
267 if (real_gpa == UNMAPPED_GVA)
270 walker->gfn = real_gpa >> PAGE_SHIFT;
275 pt_access = pte_access;
279 if (unlikely(!present || eperm || rsvd_fault))
282 if (write_fault && unlikely(!is_dirty_gpte(pte))) {
285 trace_kvm_mmu_set_dirty_bit(table_gfn, index, sizeof(pte));
286 ret = FNAME(cmpxchg_gpte)(vcpu, mmu, ptep_user, index,
287 pte, pte|PT_DIRTY_MASK);
288 if (unlikely(ret < 0)) {
294 mark_page_dirty(vcpu->kvm, table_gfn);
295 pte |= PT_DIRTY_MASK;
296 walker->ptes[walker->level - 1] = pte;
299 walker->pt_access = pt_access;
300 walker->pte_access = pte_access;
301 pgprintk("%s: pte %llx pte_access %x pt_access %x\n",
302 __func__, (u64)pte, pte_access, pt_access);
306 walker->fault.vector = PF_VECTOR;
307 walker->fault.error_code_valid = true;
308 walker->fault.error_code = 0;
310 walker->fault.error_code |= PFERR_PRESENT_MASK;
312 walker->fault.error_code |= write_fault | user_fault;
314 if (fetch_fault && (mmu->nx ||
315 kvm_read_cr4_bits(vcpu, X86_CR4_SMEP)))
316 walker->fault.error_code |= PFERR_FETCH_MASK;
318 walker->fault.error_code |= PFERR_RSVD_MASK;
320 walker->fault.address = addr;
321 walker->fault.nested_page_fault = mmu != vcpu->arch.walk_mmu;
323 trace_kvm_mmu_walker_error(walker->fault.error_code);
327 static int FNAME(walk_addr)(struct guest_walker *walker,
328 struct kvm_vcpu *vcpu, gva_t addr, u32 access)
330 return FNAME(walk_addr_generic)(walker, vcpu, &vcpu->arch.mmu, addr,
334 static int FNAME(walk_addr_nested)(struct guest_walker *walker,
335 struct kvm_vcpu *vcpu, gva_t addr,
338 return FNAME(walk_addr_generic)(walker, vcpu, &vcpu->arch.nested_mmu,
342 static bool FNAME(prefetch_invalid_gpte)(struct kvm_vcpu *vcpu,
343 struct kvm_mmu_page *sp, u64 *spte,
346 u64 nonpresent = shadow_trap_nonpresent_pte;
348 if (is_rsvd_bits_set(&vcpu->arch.mmu, gpte, PT_PAGE_TABLE_LEVEL))
351 if (!is_present_gpte(gpte)) {
353 nonpresent = shadow_notrap_nonpresent_pte;
357 if (!(gpte & PT_ACCESSED_MASK))
363 drop_spte(vcpu->kvm, spte, nonpresent);
367 static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
368 u64 *spte, const void *pte)
374 gpte = *(const pt_element_t *)pte;
375 if (FNAME(prefetch_invalid_gpte)(vcpu, sp, spte, gpte))
378 pgprintk("%s: gpte %llx spte %p\n", __func__, (u64)gpte, spte);
379 pte_access = sp->role.access & FNAME(gpte_access)(vcpu, gpte);
380 pfn = gfn_to_pfn_atomic(vcpu->kvm, gpte_to_gfn(gpte));
381 if (is_error_pfn(pfn)) {
382 kvm_release_pfn_clean(pfn);
387 * we call mmu_set_spte() with host_writable = true because that
388 * vcpu->arch.update_pte.pfn was fetched from get_user_pages(write = 1).
390 mmu_set_spte(vcpu, spte, sp->role.access, pte_access, 0, 0,
391 is_dirty_gpte(gpte), NULL, PT_PAGE_TABLE_LEVEL,
392 gpte_to_gfn(gpte), pfn, true, true);
395 static bool FNAME(gpte_changed)(struct kvm_vcpu *vcpu,
396 struct guest_walker *gw, int level)
398 pt_element_t curr_pte;
399 gpa_t base_gpa, pte_gpa = gw->pte_gpa[level - 1];
403 if (level == PT_PAGE_TABLE_LEVEL) {
404 mask = PTE_PREFETCH_NUM * sizeof(pt_element_t) - 1;
405 base_gpa = pte_gpa & ~mask;
406 index = (pte_gpa - base_gpa) / sizeof(pt_element_t);
408 r = kvm_read_guest_atomic(vcpu->kvm, base_gpa,
409 gw->prefetch_ptes, sizeof(gw->prefetch_ptes));
410 curr_pte = gw->prefetch_ptes[index];
412 r = kvm_read_guest_atomic(vcpu->kvm, pte_gpa,
413 &curr_pte, sizeof(curr_pte));
415 return r || curr_pte != gw->ptes[level - 1];
418 static void FNAME(pte_prefetch)(struct kvm_vcpu *vcpu, struct guest_walker *gw,
421 struct kvm_mmu_page *sp;
422 pt_element_t *gptep = gw->prefetch_ptes;
426 sp = page_header(__pa(sptep));
428 if (sp->role.level > PT_PAGE_TABLE_LEVEL)
432 return __direct_pte_prefetch(vcpu, sp, sptep);
434 i = (sptep - sp->spt) & ~(PTE_PREFETCH_NUM - 1);
437 for (i = 0; i < PTE_PREFETCH_NUM; i++, spte++) {
447 if (*spte != shadow_trap_nonpresent_pte)
452 if (FNAME(prefetch_invalid_gpte)(vcpu, sp, spte, gpte))
455 pte_access = sp->role.access & FNAME(gpte_access)(vcpu, gpte);
456 gfn = gpte_to_gfn(gpte);
457 dirty = is_dirty_gpte(gpte);
458 pfn = pte_prefetch_gfn_to_pfn(vcpu, gfn,
459 (pte_access & ACC_WRITE_MASK) && dirty);
460 if (is_error_pfn(pfn)) {
461 kvm_release_pfn_clean(pfn);
465 mmu_set_spte(vcpu, spte, sp->role.access, pte_access, 0, 0,
466 dirty, NULL, PT_PAGE_TABLE_LEVEL, gfn,
472 * Fetch a shadow pte for a specific level in the paging hierarchy.
474 static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
475 struct guest_walker *gw,
476 int user_fault, int write_fault, int hlevel,
477 int *ptwrite, pfn_t pfn, bool map_writable,
480 unsigned access = gw->pt_access;
481 struct kvm_mmu_page *sp = NULL;
482 bool dirty = is_dirty_gpte(gw->ptes[gw->level - 1]);
484 unsigned direct_access;
485 struct kvm_shadow_walk_iterator it;
487 if (!is_present_gpte(gw->ptes[gw->level - 1]))
490 direct_access = gw->pt_access & gw->pte_access;
492 direct_access &= ~ACC_WRITE_MASK;
494 top_level = vcpu->arch.mmu.root_level;
495 if (top_level == PT32E_ROOT_LEVEL)
496 top_level = PT32_ROOT_LEVEL;
498 * Verify that the top-level gpte is still there. Since the page
499 * is a root page, it is either write protected (and cannot be
500 * changed from now on) or it is invalid (in which case, we don't
501 * really care if it changes underneath us after this point).
503 if (FNAME(gpte_changed)(vcpu, gw, top_level))
504 goto out_gpte_changed;
506 for (shadow_walk_init(&it, vcpu, addr);
507 shadow_walk_okay(&it) && it.level > gw->level;
508 shadow_walk_next(&it)) {
511 drop_large_spte(vcpu, it.sptep);
514 if (!is_shadow_present_pte(*it.sptep)) {
515 table_gfn = gw->table_gfn[it.level - 2];
516 sp = kvm_mmu_get_page(vcpu, table_gfn, addr, it.level-1,
517 false, access, it.sptep);
521 * Verify that the gpte in the page we've just write
522 * protected is still there.
524 if (FNAME(gpte_changed)(vcpu, gw, it.level - 1))
525 goto out_gpte_changed;
528 link_shadow_page(it.sptep, sp);
532 shadow_walk_okay(&it) && it.level > hlevel;
533 shadow_walk_next(&it)) {
536 validate_direct_spte(vcpu, it.sptep, direct_access);
538 drop_large_spte(vcpu, it.sptep);
540 if (is_shadow_present_pte(*it.sptep))
543 direct_gfn = gw->gfn & ~(KVM_PAGES_PER_HPAGE(it.level) - 1);
545 sp = kvm_mmu_get_page(vcpu, direct_gfn, addr, it.level-1,
546 true, direct_access, it.sptep);
547 link_shadow_page(it.sptep, sp);
550 mmu_set_spte(vcpu, it.sptep, access, gw->pte_access & access,
551 user_fault, write_fault, dirty, ptwrite, it.level,
552 gw->gfn, pfn, prefault, map_writable);
553 FNAME(pte_prefetch)(vcpu, gw, it.sptep);
559 kvm_mmu_put_page(sp, it.sptep);
560 kvm_release_pfn_clean(pfn);
565 * Page fault handler. There are several causes for a page fault:
566 * - there is no shadow pte for the guest pte
567 * - write access through a shadow pte marked read only so that we can set
569 * - write access to a shadow pte marked read only so we can update the page
570 * dirty bitmap, when userspace requests it
571 * - mmio access; in this case we will never install a present shadow pte
572 * - normal guest page fault due to the guest pte marked not present, not
573 * writable, or not executable
575 * Returns: 1 if we need to emulate the instruction, 0 otherwise, or
576 * a negative value on error.
578 static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr, u32 error_code,
581 int write_fault = error_code & PFERR_WRITE_MASK;
582 int user_fault = error_code & PFERR_USER_MASK;
583 struct guest_walker walker;
588 int level = PT_PAGE_TABLE_LEVEL;
590 unsigned long mmu_seq;
593 pgprintk("%s: addr %lx err %x\n", __func__, addr, error_code);
595 r = mmu_topup_memory_caches(vcpu);
600 * Look up the guest pte for the faulting address.
602 r = FNAME(walk_addr)(&walker, vcpu, addr, error_code);
605 * The page is not mapped by the guest. Let the guest handle it.
608 pgprintk("%s: guest page fault\n", __func__);
610 inject_page_fault(vcpu, &walker.fault);
611 /* reset fork detector */
612 vcpu->arch.last_pt_write_count = 0;
617 if (walker.level >= PT_DIRECTORY_LEVEL)
618 force_pt_level = mapping_level_dirty_bitmap(vcpu, walker.gfn);
621 if (!force_pt_level) {
622 level = min(walker.level, mapping_level(vcpu, walker.gfn));
623 walker.gfn = walker.gfn & ~(KVM_PAGES_PER_HPAGE(level) - 1);
626 mmu_seq = vcpu->kvm->mmu_notifier_seq;
629 if (try_async_pf(vcpu, prefault, walker.gfn, addr, &pfn, write_fault,
634 if (is_error_pfn(pfn))
635 return kvm_handle_bad_page(vcpu->kvm, walker.gfn, pfn);
637 spin_lock(&vcpu->kvm->mmu_lock);
638 if (mmu_notifier_retry(vcpu, mmu_seq))
641 trace_kvm_mmu_audit(vcpu, AUDIT_PRE_PAGE_FAULT);
642 kvm_mmu_free_some_pages(vcpu);
644 transparent_hugepage_adjust(vcpu, &walker.gfn, &pfn, &level);
645 sptep = FNAME(fetch)(vcpu, addr, &walker, user_fault, write_fault,
646 level, &write_pt, pfn, map_writable, prefault);
648 pgprintk("%s: shadow pte %p %llx ptwrite %d\n", __func__,
649 sptep, *sptep, write_pt);
652 vcpu->arch.last_pt_write_count = 0; /* reset fork detector */
654 ++vcpu->stat.pf_fixed;
655 trace_kvm_mmu_audit(vcpu, AUDIT_POST_PAGE_FAULT);
656 spin_unlock(&vcpu->kvm->mmu_lock);
661 spin_unlock(&vcpu->kvm->mmu_lock);
662 kvm_release_pfn_clean(pfn);
666 static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva)
668 struct kvm_shadow_walk_iterator iterator;
669 struct kvm_mmu_page *sp;
675 spin_lock(&vcpu->kvm->mmu_lock);
677 for_each_shadow_entry(vcpu, gva, iterator) {
678 level = iterator.level;
679 sptep = iterator.sptep;
681 sp = page_header(__pa(sptep));
682 if (is_last_spte(*sptep, level)) {
689 (PT_LEVEL_BITS - PT64_LEVEL_BITS) * level;
690 offset = sp->role.quadrant << shift;
692 pte_gpa = (sp->gfn << PAGE_SHIFT) + offset;
693 pte_gpa += (sptep - sp->spt) * sizeof(pt_element_t);
695 if (is_shadow_present_pte(*sptep)) {
696 if (is_large_pte(*sptep))
697 --vcpu->kvm->stat.lpages;
698 drop_spte(vcpu->kvm, sptep,
699 shadow_trap_nonpresent_pte);
702 __set_spte(sptep, shadow_trap_nonpresent_pte);
706 if (!is_shadow_present_pte(*sptep) || !sp->unsync_children)
711 kvm_flush_remote_tlbs(vcpu->kvm);
713 atomic_inc(&vcpu->kvm->arch.invlpg_counter);
715 spin_unlock(&vcpu->kvm->mmu_lock);
720 if (mmu_topup_memory_caches(vcpu))
722 kvm_mmu_pte_write(vcpu, pte_gpa, NULL, sizeof(pt_element_t), 0);
725 static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr, u32 access,
726 struct x86_exception *exception)
728 struct guest_walker walker;
729 gpa_t gpa = UNMAPPED_GVA;
732 r = FNAME(walk_addr)(&walker, vcpu, vaddr, access);
735 gpa = gfn_to_gpa(walker.gfn);
736 gpa |= vaddr & ~PAGE_MASK;
737 } else if (exception)
738 *exception = walker.fault;
743 static gpa_t FNAME(gva_to_gpa_nested)(struct kvm_vcpu *vcpu, gva_t vaddr,
745 struct x86_exception *exception)
747 struct guest_walker walker;
748 gpa_t gpa = UNMAPPED_GVA;
751 r = FNAME(walk_addr_nested)(&walker, vcpu, vaddr, access);
754 gpa = gfn_to_gpa(walker.gfn);
755 gpa |= vaddr & ~PAGE_MASK;
756 } else if (exception)
757 *exception = walker.fault;
762 static void FNAME(prefetch_page)(struct kvm_vcpu *vcpu,
763 struct kvm_mmu_page *sp)
766 pt_element_t pt[256 / sizeof(pt_element_t)];
770 || (PTTYPE == 32 && sp->role.level > PT_PAGE_TABLE_LEVEL)) {
771 nonpaging_prefetch_page(vcpu, sp);
775 pte_gpa = gfn_to_gpa(sp->gfn);
777 offset = sp->role.quadrant << PT64_LEVEL_BITS;
778 pte_gpa += offset * sizeof(pt_element_t);
781 for (i = 0; i < PT64_ENT_PER_PAGE; i += ARRAY_SIZE(pt)) {
782 r = kvm_read_guest_atomic(vcpu->kvm, pte_gpa, pt, sizeof pt);
783 pte_gpa += ARRAY_SIZE(pt) * sizeof(pt_element_t);
784 for (j = 0; j < ARRAY_SIZE(pt); ++j)
785 if (r || is_present_gpte(pt[j]))
786 sp->spt[i+j] = shadow_trap_nonpresent_pte;
788 sp->spt[i+j] = shadow_notrap_nonpresent_pte;
793 * Using the cached information from sp->gfns is safe because:
794 * - The spte has a reference to the struct page, so the pfn for a given gfn
795 * can't change unless all sptes pointing to it are nuked first.
798 * We should flush all tlbs if spte is dropped even though guest is
799 * responsible for it. Since if we don't, kvm_mmu_notifier_invalidate_page
800 * and kvm_mmu_notifier_invalidate_range_start detect the mapping page isn't
801 * used by guest then tlbs are not flushed, so guest is allowed to access the
803 * And we increase kvm->tlbs_dirty to delay tlbs flush in this case.
805 static int FNAME(sync_page)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
807 int i, offset, nr_present;
811 offset = nr_present = 0;
813 /* direct kvm_mmu_page can not be unsync. */
814 BUG_ON(sp->role.direct);
817 offset = sp->role.quadrant << PT64_LEVEL_BITS;
819 first_pte_gpa = gfn_to_gpa(sp->gfn) + offset * sizeof(pt_element_t);
821 for (i = 0; i < PT64_ENT_PER_PAGE; i++) {
827 if (!is_shadow_present_pte(sp->spt[i]))
830 pte_gpa = first_pte_gpa + i * sizeof(pt_element_t);
832 if (kvm_read_guest_atomic(vcpu->kvm, pte_gpa, &gpte,
833 sizeof(pt_element_t)))
836 gfn = gpte_to_gfn(gpte);
838 if (FNAME(prefetch_invalid_gpte)(vcpu, sp, &sp->spt[i], gpte)) {
839 vcpu->kvm->tlbs_dirty++;
843 if (gfn != sp->gfns[i]) {
844 drop_spte(vcpu->kvm, &sp->spt[i],
845 shadow_trap_nonpresent_pte);
846 vcpu->kvm->tlbs_dirty++;
851 pte_access = sp->role.access & FNAME(gpte_access)(vcpu, gpte);
852 host_writable = sp->spt[i] & SPTE_HOST_WRITEABLE;
854 set_spte(vcpu, &sp->spt[i], pte_access, 0, 0,
855 is_dirty_gpte(gpte), PT_PAGE_TABLE_LEVEL, gfn,
856 spte_to_pfn(sp->spt[i]), true, false,
866 #undef PT_BASE_ADDR_MASK
868 #undef PT_LVL_ADDR_MASK
869 #undef PT_LVL_OFFSET_MASK
871 #undef PT_MAX_FULL_LEVELS
873 #undef gpte_to_gfn_lvl