Lines Matching +full:ext +full:- +full:gen
1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * This module enables machines with Intel VT-x extensions to run virtual
64 int __read_mostly nx_huge_pages = -1;
100 * When setting this variable to true it enables Two-Dimensional-Paging
102 * 1. the guest-virtual to guest-physical
103 * 2. while doing 1. it walks guest-physical to host-physical
204 return !!(regs->reg & flag); \
226 return !!(mmu->cpu_role. base_or_ext . reg##_##name); \
229 BUILD_MMU_ROLE_ACCESSOR(ext, cr4, pse);
230 BUILD_MMU_ROLE_ACCESSOR(ext, cr4, smep);
231 BUILD_MMU_ROLE_ACCESSOR(ext, cr4, smap);
232 BUILD_MMU_ROLE_ACCESSOR(ext, cr4, pke);
233 BUILD_MMU_ROLE_ACCESSOR(ext, cr4, la57);
235 BUILD_MMU_ROLE_ACCESSOR(ext, efer, lma);
239 return mmu->cpu_role.base.level > 0; in is_cr0_pg()
244 return !mmu->cpu_role.base.has_4_byte_gpte; in is_cr4_pae()
252 .efer = vcpu->arch.efer, in vcpu_to_role_regs()
266 if (IS_ENABLED(CONFIG_RETPOLINE) && mmu->get_guest_pgd == get_guest_cr3) in kvm_mmu_get_guest_pgd()
269 return mmu->get_guest_pgd(vcpu); in kvm_mmu_get_guest_pgd()
289 kvm_flush_remote_tlbs_gfn(kvm, gfn, sp->role.level); in kvm_flush_remote_tlbs_sptep()
318 u64 kvm_gen, spte_gen, gen; in check_mmio_spte() local
320 gen = kvm_vcpu_memslots(vcpu)->generation; in check_mmio_spte()
321 if (unlikely(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS)) in check_mmio_spte()
324 kvm_gen = gen & MMIO_SPTE_GEN_MASK; in check_mmio_spte()
374 sp->clear_spte_count++; in count_spte_clear()
384 ssptep->spte_high = sspte.spte_high; in __set_spte()
393 WRITE_ONCE(ssptep->spte_low, sspte.spte_low); in __set_spte()
403 WRITE_ONCE(ssptep->spte_low, sspte.spte_low); in __update_clear_spte_fast()
411 ssptep->spte_high = sspte.spte_high; in __update_clear_spte_fast()
423 orig.spte_low = xchg(&ssptep->spte_low, sspte.spte_low); in __update_clear_spte_slow()
424 orig.spte_high = ssptep->spte_high; in __update_clear_spte_slow()
425 ssptep->spte_high = sspte.spte_high; in __update_clear_spte_slow()
437 * we need to protect against in-progress updates of the spte.
440 * for the high part of the spte. The race is fine for a present->non-present
441 * change (because the high part of the spte is ignored for non-present spte),
442 * but for a present->present change we must reread the spte.
444 * All such changes are done in two steps (present->non-present and
445 * non-present->present), hence it is enough to count the number of
446 * present->non-present updates: if it changed while reading the spte,
456 count = sp->clear_spte_count; in __get_spte_lockless()
459 spte.spte_low = orig->spte_low; in __get_spte_lockless()
462 spte.spte_high = orig->spte_high; in __get_spte_lockless()
465 if (unlikely(spte.spte_low != orig->spte_low || in __get_spte_lockless()
466 count != sp->clear_spte_count)) in __get_spte_lockless()
514 * Whenever an MMU-writable SPTE is overwritten with a read-only SPTE, remote
515 * TLBs must be flushed. Otherwise rmap_write_protect will find a read-only
529 * For the spte updated out of mmu-lock is safe, since in mmu_spte_update()
565 int level = sptep_to_sp(sptep)->role.level; in mmu_spte_clear_track_bits()
577 kvm_update_page_stats(kvm, level, -1); in mmu_spte_clear_track_bits()
623 clear_bit((ffs(shadow_accessed_mask) - 1), in mmu_spte_age()
642 return tdp_mmu_enabled && vcpu->arch.mmu->root_role.direct; in is_tdp_mmu_active()
651 * Prevent page table teardown by making any free-er wait during in walk_shadow_page_lockless_begin()
658 * to vcpu->mode. in walk_shadow_page_lockless_begin()
660 smp_store_mb(vcpu->mode, READING_SHADOW_PAGE_TABLES); in walk_shadow_page_lockless_begin()
670 * Make sure the write to vcpu->mode is not reordered in front of in walk_shadow_page_lockless_end()
674 smp_store_release(&vcpu->mode, OUTSIDE_GUEST_MODE); in walk_shadow_page_lockless_end()
684 r = kvm_mmu_topup_memory_cache(&vcpu->arch.mmu_pte_list_desc_cache, in mmu_topup_memory_caches()
688 r = kvm_mmu_topup_memory_cache(&vcpu->arch.mmu_shadow_page_cache, in mmu_topup_memory_caches()
693 r = kvm_mmu_topup_memory_cache(&vcpu->arch.mmu_shadowed_info_cache, in mmu_topup_memory_caches()
698 return kvm_mmu_topup_memory_cache(&vcpu->arch.mmu_page_header_cache, in mmu_topup_memory_caches()
704 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_pte_list_desc_cache); in mmu_free_memory_caches()
705 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_shadow_page_cache); in mmu_free_memory_caches()
706 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_shadowed_info_cache); in mmu_free_memory_caches()
707 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_header_cache); in mmu_free_memory_caches()
719 if (sp->role.passthrough) in kvm_mmu_page_get_gfn()
720 return sp->gfn; in kvm_mmu_page_get_gfn()
722 if (!sp->role.direct) in kvm_mmu_page_get_gfn()
723 return sp->shadowed_translation[index] >> PAGE_SHIFT; in kvm_mmu_page_get_gfn()
725 return sp->gfn + (index << ((sp->role.level - 1) * SPTE_LEVEL_BITS)); in kvm_mmu_page_get_gfn()
737 return sp->shadowed_translation[index] & ACC_ALL; in kvm_mmu_page_get_access()
740 * For direct MMUs (e.g. TDP or non-paging guests) or passthrough SPs, in kvm_mmu_page_get_access()
749 * In both cases, sp->role.access contains the correct access bits. in kvm_mmu_page_get_access()
751 return sp->role.access; in kvm_mmu_page_get_access()
758 sp->shadowed_translation[index] = (gfn << PAGE_SHIFT) | access; in kvm_mmu_page_set_translation()
764 sp->role.passthrough ? "passthrough" : "direct", in kvm_mmu_page_set_translation()
765 sp->gfn, kvm_mmu_page_get_access(sp, index), access); in kvm_mmu_page_set_translation()
769 sp->role.passthrough ? "passthrough" : "direct", in kvm_mmu_page_set_translation()
770 sp->gfn, kvm_mmu_page_get_gfn(sp, index), gfn); in kvm_mmu_page_set_translation()
790 idx = gfn_to_index(gfn, slot->base_gfn, level); in lpage_info_slot()
791 return &slot->arch.lpage_info[level - 2][idx]; in lpage_info_slot()
811 old = linfo->disallow_lpage; in update_gfn_disallow_lpage_count()
812 linfo->disallow_lpage += count; in update_gfn_disallow_lpage_count()
813 WARN_ON_ONCE((old ^ linfo->disallow_lpage) & KVM_LPAGE_MIXED_FLAG); in update_gfn_disallow_lpage_count()
824 update_gfn_disallow_lpage_count(slot, gfn, -1); in kvm_mmu_gfn_allow_lpage()
833 kvm->arch.indirect_shadow_pages++; in account_shadowed()
834 gfn = sp->gfn; in account_shadowed()
835 slots = kvm_memslots_for_spte_role(kvm, sp->role); in account_shadowed()
838 /* the non-leaf shadow pages are keeping readonly. */ in account_shadowed()
839 if (sp->role.level > PG_LEVEL_4K) in account_shadowed()
858 if (!list_empty(&sp->possible_nx_huge_page_link)) in track_possible_nx_huge_page()
861 ++kvm->stat.nx_lpage_splits; in track_possible_nx_huge_page()
862 list_add_tail(&sp->possible_nx_huge_page_link, in track_possible_nx_huge_page()
863 &kvm->arch.possible_nx_huge_pages); in track_possible_nx_huge_page()
869 sp->nx_huge_page_disallowed = true; in account_nx_huge_page()
881 kvm->arch.indirect_shadow_pages--; in unaccount_shadowed()
882 gfn = sp->gfn; in unaccount_shadowed()
883 slots = kvm_memslots_for_spte_role(kvm, sp->role); in unaccount_shadowed()
885 if (sp->role.level > PG_LEVEL_4K) in unaccount_shadowed()
893 if (list_empty(&sp->possible_nx_huge_page_link)) in untrack_possible_nx_huge_page()
896 --kvm->stat.nx_lpage_splits; in untrack_possible_nx_huge_page()
897 list_del_init(&sp->possible_nx_huge_page_link); in untrack_possible_nx_huge_page()
902 sp->nx_huge_page_disallowed = false; in unaccount_nx_huge_page()
914 if (!slot || slot->flags & KVM_MEMSLOT_INVALID) in gfn_to_memslot_dirty_bitmap()
925 * If the bit zero of rmap_head->val is clear, then it points to the only spte
926 * in this rmap chain. Otherwise, (rmap_head->val & ~1) points to a struct
939 if (!rmap_head->val) { in pte_list_add()
940 rmap_head->val = (unsigned long)spte; in pte_list_add()
941 } else if (!(rmap_head->val & 1)) { in pte_list_add()
943 desc->sptes[0] = (u64 *)rmap_head->val; in pte_list_add()
944 desc->sptes[1] = spte; in pte_list_add()
945 desc->spte_count = 2; in pte_list_add()
946 desc->tail_count = 0; in pte_list_add()
947 rmap_head->val = (unsigned long)desc | 1; in pte_list_add()
950 desc = (struct pte_list_desc *)(rmap_head->val & ~1ul); in pte_list_add()
951 count = desc->tail_count + desc->spte_count; in pte_list_add()
957 if (desc->spte_count == PTE_LIST_EXT) { in pte_list_add()
959 desc->more = (struct pte_list_desc *)(rmap_head->val & ~1ul); in pte_list_add()
960 desc->spte_count = 0; in pte_list_add()
961 desc->tail_count = count; in pte_list_add()
962 rmap_head->val = (unsigned long)desc | 1; in pte_list_add()
964 desc->sptes[desc->spte_count++] = spte; in pte_list_add()
973 struct pte_list_desc *head_desc = (struct pte_list_desc *)(rmap_head->val & ~1ul); in pte_list_desc_remove_entry()
974 int j = head_desc->spte_count - 1; in pte_list_desc_remove_entry()
984 * Replace the to-be-freed SPTE with the last valid entry from the head in pte_list_desc_remove_entry()
988 desc->sptes[i] = head_desc->sptes[j]; in pte_list_desc_remove_entry()
989 head_desc->sptes[j] = NULL; in pte_list_desc_remove_entry()
990 head_desc->spte_count--; in pte_list_desc_remove_entry()
991 if (head_desc->spte_count) in pte_list_desc_remove_entry()
999 if (!head_desc->more) in pte_list_desc_remove_entry()
1000 rmap_head->val = 0; in pte_list_desc_remove_entry()
1002 rmap_head->val = (unsigned long)head_desc->more | 1; in pte_list_desc_remove_entry()
1012 if (KVM_BUG_ON_DATA_CORRUPTION(!rmap_head->val, kvm)) in pte_list_remove()
1015 if (!(rmap_head->val & 1)) { in pte_list_remove()
1016 if (KVM_BUG_ON_DATA_CORRUPTION((u64 *)rmap_head->val != spte, kvm)) in pte_list_remove()
1019 rmap_head->val = 0; in pte_list_remove()
1021 desc = (struct pte_list_desc *)(rmap_head->val & ~1ul); in pte_list_remove()
1023 for (i = 0; i < desc->spte_count; ++i) { in pte_list_remove()
1024 if (desc->sptes[i] == spte) { in pte_list_remove()
1030 desc = desc->more; in pte_list_remove()
1051 if (!rmap_head->val) in kvm_zap_all_rmap_sptes()
1054 if (!(rmap_head->val & 1)) { in kvm_zap_all_rmap_sptes()
1055 mmu_spte_clear_track_bits(kvm, (u64 *)rmap_head->val); in kvm_zap_all_rmap_sptes()
1059 desc = (struct pte_list_desc *)(rmap_head->val & ~1ul); in kvm_zap_all_rmap_sptes()
1062 for (i = 0; i < desc->spte_count; i++) in kvm_zap_all_rmap_sptes()
1063 mmu_spte_clear_track_bits(kvm, desc->sptes[i]); in kvm_zap_all_rmap_sptes()
1064 next = desc->more; in kvm_zap_all_rmap_sptes()
1069 rmap_head->val = 0; in kvm_zap_all_rmap_sptes()
1077 if (!rmap_head->val) in pte_list_count()
1079 else if (!(rmap_head->val & 1)) in pte_list_count()
1082 desc = (struct pte_list_desc *)(rmap_head->val & ~1ul); in pte_list_count()
1083 return desc->tail_count + desc->spte_count; in pte_list_count()
1091 idx = gfn_to_index(gfn, slot->base_gfn, level); in gfn_to_rmap()
1092 return &slot->arch.rmap[level - PG_LEVEL_4K][idx]; in gfn_to_rmap()
1109 * information in sp->role. in rmap_remove()
1111 slots = kvm_memslots_for_spte_role(kvm, sp->role); in rmap_remove()
1114 rmap_head = gfn_to_rmap(gfn, sp->role.level, slot); in rmap_remove()
1141 if (!rmap_head->val) in rmap_get_first()
1144 if (!(rmap_head->val & 1)) { in rmap_get_first()
1145 iter->desc = NULL; in rmap_get_first()
1146 sptep = (u64 *)rmap_head->val; in rmap_get_first()
1150 iter->desc = (struct pte_list_desc *)(rmap_head->val & ~1ul); in rmap_get_first()
1151 iter->pos = 0; in rmap_get_first()
1152 sptep = iter->desc->sptes[iter->pos]; in rmap_get_first()
1167 if (iter->desc) { in rmap_get_next()
1168 if (iter->pos < PTE_LIST_EXT - 1) { in rmap_get_next()
1169 ++iter->pos; in rmap_get_next()
1170 sptep = iter->desc->sptes[iter->pos]; in rmap_get_next()
1175 iter->desc = iter->desc->more; in rmap_get_next()
1177 if (iter->desc) { in rmap_get_next()
1178 iter->pos = 0; in rmap_get_next()
1179 /* desc->sptes[0] cannot be NULL */ in rmap_get_next()
1180 sptep = iter->desc->sptes[iter->pos]; in rmap_get_next()
1208 WARN_ON_ONCE(sp->role.level == PG_LEVEL_4K); in drop_large_spte()
1217 * Write-protect on the specified @sptep, @pt_protect indicates whether
1218 * spte write-protection is caused by protecting shadow page table.
1222 * - for dirty logging, the spte can be set to writable at anytime if
1224 * - for spte protection, the spte can be writable only after unsync-ing
1278 * - D bit on ad-enabled SPTEs, and
1279 * - W bit on ad-disabled SPTEs.
1299 * kvm_mmu_write_protect_pt_masked - write protect selected PT level pages
1315 slot->base_gfn + gfn_offset, mask, true); in kvm_mmu_write_protect_pt_masked()
1321 rmap_head = gfn_to_rmap(slot->base_gfn + gfn_offset + __ffs(mask), in kvm_mmu_write_protect_pt_masked()
1326 mask &= mask - 1; in kvm_mmu_write_protect_pt_masked()
1331 * kvm_mmu_clear_dirty_pt_masked - clear MMU D-bit for PT level pages, or write
1332 * protect the page if the D-bit isn't supported.
1334 * @slot: slot to clear D-bit
1336 * @mask: indicates which pages we should clear D-bit
1338 * Used for PML to re-log the dirty GPAs after userspace querying dirty_bitmap.
1348 slot->base_gfn + gfn_offset, mask, false); in kvm_mmu_clear_dirty_pt_masked()
1354 rmap_head = gfn_to_rmap(slot->base_gfn + gfn_offset + __ffs(mask), in kvm_mmu_clear_dirty_pt_masked()
1359 mask &= mask - 1; in kvm_mmu_clear_dirty_pt_masked()
1364 * kvm_arch_mmu_enable_log_dirty_pt_masked - enable dirty logging for selected
1379 * initially-all-set mode; must write protect them here so that they in kvm_arch_mmu_enable_log_dirty_pt_masked()
1387 gfn_t start = slot->base_gfn + gfn_offset + __ffs(mask); in kvm_arch_mmu_enable_log_dirty_pt_masked()
1388 gfn_t end = slot->base_gfn + gfn_offset + __fls(mask); in kvm_arch_mmu_enable_log_dirty_pt_masked()
1441 return kvm_mmu_slot_gfn_write_protect(vcpu->kvm, slot, gfn, PG_LEVEL_4K); in kvm_vcpu_write_protect_gfn()
1514 iterator->level = level; in rmap_walk_init_level()
1515 iterator->gfn = iterator->start_gfn; in rmap_walk_init_level()
1516 iterator->rmap = gfn_to_rmap(iterator->gfn, level, iterator->slot); in rmap_walk_init_level()
1517 iterator->end_rmap = gfn_to_rmap(iterator->end_gfn, level, iterator->slot); in rmap_walk_init_level()
1525 iterator->slot = slot; in slot_rmap_walk_init()
1526 iterator->start_level = start_level; in slot_rmap_walk_init()
1527 iterator->end_level = end_level; in slot_rmap_walk_init()
1528 iterator->start_gfn = start_gfn; in slot_rmap_walk_init()
1529 iterator->end_gfn = end_gfn; in slot_rmap_walk_init()
1531 rmap_walk_init_level(iterator, iterator->start_level); in slot_rmap_walk_init()
1536 return !!iterator->rmap; in slot_rmap_walk_okay()
1541 while (++iterator->rmap <= iterator->end_rmap) { in slot_rmap_walk_next()
1542 iterator->gfn += (1UL << KVM_HPAGE_GFN_SHIFT(iterator->level)); in slot_rmap_walk_next()
1544 if (iterator->rmap->val) in slot_rmap_walk_next()
1548 if (++iterator->level > iterator->end_level) { in slot_rmap_walk_next()
1549 iterator->rmap = NULL; in slot_rmap_walk_next()
1553 rmap_walk_init_level(iterator, iterator->level); in slot_rmap_walk_next()
1574 for_each_slot_rmap_range(range->slot, PG_LEVEL_4K, KVM_MAX_HUGEPAGE_LEVEL, in kvm_handle_gfn_range()
1575 range->start, range->end - 1, &iterator) in kvm_handle_gfn_range()
1576 ret |= handler(kvm, iterator.rmap, range->slot, iterator.gfn, in kvm_handle_gfn_range()
1577 iterator.level, range->arg.pte); in kvm_handle_gfn_range()
1593 range->slot->id == APIC_ACCESS_PAGE_PRIVATE_MEMSLOT) in kvm_unmap_gfn_range()
1652 kvm_update_page_stats(kvm, sp->role.level, 1); in __rmap_add()
1654 rmap_head = gfn_to_rmap(gfn, sp->role.level, slot); in __rmap_add()
1657 if (rmap_count > kvm->stat.max_mmu_rmap_size) in __rmap_add()
1658 kvm->stat.max_mmu_rmap_size = rmap_count; in __rmap_add()
1661 kvm_flush_remote_tlbs_gfn(kvm, gfn, sp->role.level); in __rmap_add()
1668 struct kvm_mmu_memory_cache *cache = &vcpu->arch.mmu_pte_list_desc_cache; in rmap_add()
1670 __rmap_add(vcpu->kvm, cache, slot, spte, gfn, access); in rmap_add()
1705 if (KVM_MMU_WARN_ON(is_shadow_present_pte(sp->spt[i]))) in kvm_mmu_check_sptes_at_free()
1706 pr_err_ratelimited("SPTE %llx (@ %p) for gfn %llx shadow-present at free", in kvm_mmu_check_sptes_at_free()
1707 sp->spt[i], &sp->spt[i], in kvm_mmu_check_sptes_at_free()
1715 * kvm->arch.n_used_mmu_pages values. We need a global,
1721 kvm->arch.n_used_mmu_pages += nr; in kvm_mod_used_mmu_pages()
1728 kvm_account_pgtable_pages((void *)sp->spt, +1); in kvm_account_mmu_page()
1733 kvm_mod_used_mmu_pages(kvm, -1); in kvm_unaccount_mmu_page()
1734 kvm_account_pgtable_pages((void *)sp->spt, -1); in kvm_unaccount_mmu_page()
1741 hlist_del(&sp->hash_link); in kvm_mmu_free_shadow_page()
1742 list_del(&sp->link); in kvm_mmu_free_shadow_page()
1743 free_page((unsigned long)sp->spt); in kvm_mmu_free_shadow_page()
1744 if (!sp->role.direct) in kvm_mmu_free_shadow_page()
1745 free_page((unsigned long)sp->shadowed_translation); in kvm_mmu_free_shadow_page()
1760 pte_list_add(cache, parent_pte, &sp->parent_ptes); in mmu_page_add_parent_pte()
1766 pte_list_remove(kvm, parent_pte, &sp->parent_ptes); in mmu_page_remove_parent_pte()
1782 for_each_rmap_spte(&sp->parent_ptes, &iter, sptep) { in kvm_mmu_mark_parents_unsync()
1792 if (__test_and_set_bit(spte_index(spte), sp->unsync_child_bitmap)) in mark_unsync()
1794 if (sp->unsync_children++) in mark_unsync()
1814 if (sp->unsync) in mmu_pages_add()
1815 for (i=0; i < pvec->nr; i++) in mmu_pages_add()
1816 if (pvec->page[i].sp == sp) in mmu_pages_add()
1819 pvec->page[pvec->nr].sp = sp; in mmu_pages_add()
1820 pvec->page[pvec->nr].idx = idx; in mmu_pages_add()
1821 pvec->nr++; in mmu_pages_add()
1822 return (pvec->nr == KVM_PAGE_ARRAY_NR); in mmu_pages_add()
1827 --sp->unsync_children; in clear_unsync_child_bit()
1828 WARN_ON_ONCE((int)sp->unsync_children < 0); in clear_unsync_child_bit()
1829 __clear_bit(idx, sp->unsync_child_bitmap); in clear_unsync_child_bit()
1837 for_each_set_bit(i, sp->unsync_child_bitmap, 512) { in __mmu_unsync_walk()
1839 u64 ent = sp->spt[i]; in __mmu_unsync_walk()
1848 if (child->unsync_children) { in __mmu_unsync_walk()
1850 return -ENOSPC; in __mmu_unsync_walk()
1860 } else if (child->unsync) { in __mmu_unsync_walk()
1863 return -ENOSPC; in __mmu_unsync_walk()
1871 #define INVALID_INDEX (-1)
1876 pvec->nr = 0; in mmu_unsync_walk()
1877 if (!sp->unsync_children) in mmu_unsync_walk()
1886 WARN_ON_ONCE(!sp->unsync); in kvm_unlink_unsync_page()
1888 sp->unsync = 0; in kvm_unlink_unsync_page()
1889 --kvm->stat.mmu_unsync; in kvm_unlink_unsync_page()
1899 if (sp->role.direct) in sp_has_gptes()
1902 if (sp->role.passthrough) in sp_has_gptes()
1915 &(_kvm)->arch.mmu_page_hash[kvm_page_table_hashfn(_gfn)]) \
1916 if ((_sp)->gfn != (_gfn) || !sp_has_gptes(_sp)) {} else
1920 union kvm_mmu_page_role root_role = vcpu->arch.mmu->root_role; in kvm_sync_page_check()
1926 * - level: not part of the overall MMU role and will never match as the MMU's in kvm_sync_page_check()
1928 * - access: updated based on the new guest PTE in kvm_sync_page_check()
1929 * - quadrant: not part of the overall MMU role (similar to level) in kvm_sync_page_check()
1941 * differs then the memslot lookup (SMM vs. non-SMM) will be bogus, the in kvm_sync_page_check()
1944 if (WARN_ON_ONCE(sp->role.direct || !vcpu->arch.mmu->sync_spte || in kvm_sync_page_check()
1945 (sp->role.word ^ root_role.word) & ~sync_role_ign.word)) in kvm_sync_page_check()
1953 if (!sp->spt[i]) in kvm_sync_spte()
1956 return vcpu->arch.mmu->sync_spte(vcpu, sp, i); in kvm_sync_spte()
1965 return -1; in __kvm_sync_page()
1970 if (ret < -1) in __kvm_sync_page()
1971 return -1; in __kvm_sync_page()
1993 kvm_mmu_prepare_zap_page(vcpu->kvm, sp, invalid_list); in kvm_sync_page()
2013 if (sp->role.invalid) in is_obsolete_sp()
2018 unlikely(sp->mmu_valid_gen != kvm->arch.mmu_valid_gen); in is_obsolete_sp()
2037 for (n = i+1; n < pvec->nr; n++) { in mmu_pages_next()
2038 struct kvm_mmu_page *sp = pvec->page[n].sp; in mmu_pages_next()
2039 unsigned idx = pvec->page[n].idx; in mmu_pages_next()
2040 int level = sp->role.level; in mmu_pages_next()
2042 parents->idx[level-1] = idx; in mmu_pages_next()
2046 parents->parent[level-2] = sp; in mmu_pages_next()
2058 if (pvec->nr == 0) in mmu_pages_first()
2061 WARN_ON_ONCE(pvec->page[0].idx != INVALID_INDEX); in mmu_pages_first()
2063 sp = pvec->page[0].sp; in mmu_pages_first()
2064 level = sp->role.level; in mmu_pages_first()
2067 parents->parent[level-2] = sp; in mmu_pages_first()
2072 parents->parent[level-1] = NULL; in mmu_pages_first()
2082 unsigned int idx = parents->idx[level]; in mmu_pages_clear_parents()
2083 sp = parents->parent[level]; in mmu_pages_clear_parents()
2090 } while (!sp->unsync_children); in mmu_pages_clear_parents()
2107 protected |= kvm_vcpu_write_protect_gfn(vcpu, sp->gfn); in mmu_sync_children()
2110 kvm_mmu_remote_flush_or_zap(vcpu->kvm, &invalid_list, true); in mmu_sync_children()
2115 kvm_unlink_unsync_page(vcpu->kvm, sp); in mmu_sync_children()
2119 if (need_resched() || rwlock_needbreak(&vcpu->kvm->mmu_lock)) { in mmu_sync_children()
2120 kvm_mmu_remote_flush_or_zap(vcpu->kvm, &invalid_list, flush); in mmu_sync_children()
2123 return -EINTR; in mmu_sync_children()
2126 cond_resched_rwlock_write(&vcpu->kvm->mmu_lock); in mmu_sync_children()
2131 kvm_mmu_remote_flush_or_zap(vcpu->kvm, &invalid_list, flush); in mmu_sync_children()
2137 atomic_set(&sp->write_flooding_count, 0); in __clear_sp_write_flooding_count()
2163 if (sp->gfn != gfn) { in kvm_mmu_find_shadow_page()
2168 if (sp->role.word != role.word) { in kvm_mmu_find_shadow_page()
2170 * If the guest is creating an upper-level page, zap in kvm_mmu_find_shadow_page()
2176 * upper-level page will be write-protected. in kvm_mmu_find_shadow_page()
2178 if (role.level > PG_LEVEL_4K && sp->unsync) in kvm_mmu_find_shadow_page()
2184 /* unsync and write-flooding only apply to indirect SPs. */ in kvm_mmu_find_shadow_page()
2185 if (sp->role.direct) in kvm_mmu_find_shadow_page()
2188 if (sp->unsync) { in kvm_mmu_find_shadow_page()
2195 * it doesn't write-protect the page or mark it synchronized! in kvm_mmu_find_shadow_page()
2219 ++kvm->stat.mmu_cache_miss; in kvm_mmu_find_shadow_page()
2224 if (collisions > kvm->stat.max_mmu_page_hash_collisions) in kvm_mmu_find_shadow_page()
2225 kvm->stat.max_mmu_page_hash_collisions = collisions; in kvm_mmu_find_shadow_page()
2244 sp = kvm_mmu_memory_cache_alloc(caches->page_header_cache); in kvm_mmu_alloc_shadow_page()
2245 sp->spt = kvm_mmu_memory_cache_alloc(caches->shadow_page_cache); in kvm_mmu_alloc_shadow_page()
2247 sp->shadowed_translation = kvm_mmu_memory_cache_alloc(caches->shadowed_info_cache); in kvm_mmu_alloc_shadow_page()
2249 set_page_private(virt_to_page(sp->spt), (unsigned long)sp); in kvm_mmu_alloc_shadow_page()
2251 INIT_LIST_HEAD(&sp->possible_nx_huge_page_link); in kvm_mmu_alloc_shadow_page()
2258 sp->mmu_valid_gen = kvm->arch.mmu_valid_gen; in kvm_mmu_alloc_shadow_page()
2259 list_add(&sp->link, &kvm->arch.active_mmu_pages); in kvm_mmu_alloc_shadow_page()
2262 sp->gfn = gfn; in kvm_mmu_alloc_shadow_page()
2263 sp->role = role; in kvm_mmu_alloc_shadow_page()
2264 hlist_add_head(&sp->hash_link, sp_list); in kvm_mmu_alloc_shadow_page()
2282 sp_list = &kvm->arch.mmu_page_hash[kvm_page_table_hashfn(gfn)]; in __kvm_mmu_get_shadow_page()
2299 .page_header_cache = &vcpu->arch.mmu_page_header_cache, in kvm_mmu_get_shadow_page()
2300 .shadow_page_cache = &vcpu->arch.mmu_shadow_page_cache, in kvm_mmu_get_shadow_page()
2301 .shadowed_info_cache = &vcpu->arch.mmu_shadowed_info_cache, in kvm_mmu_get_shadow_page()
2304 return __kvm_mmu_get_shadow_page(vcpu->kvm, vcpu, &caches, gfn, role); in kvm_mmu_get_shadow_page()
2313 role = parent_sp->role; in kvm_mmu_child_role()
2314 role.level--; in kvm_mmu_child_role()
2320 * If the guest has 4-byte PTEs then that means it's using 32-bit, in kvm_mmu_child_role()
2321 * 2-level, non-PAE paging. KVM shadows such guests with PAE paging in kvm_mmu_child_role()
2322 * (i.e. 8-byte PTEs). The difference in PTE size means that KVM must in kvm_mmu_child_role()
2335 * Concretely, a 4-byte PDE consumes bits 31:22, while an 8-byte PDE in kvm_mmu_child_role()
2337 * PDPTEs; those 4 PAE page directories are pre-allocated and their in kvm_mmu_child_role()
2338 * quadrant is assigned in mmu_alloc_root(). A 4-byte PTE consumes in kvm_mmu_child_role()
2339 * bits 21:12, while an 8-byte PTE consumes bits 20:12. To consume in kvm_mmu_child_role()
2341 * quadrant, i.e. sets quadrant to '0' or '1'. The parent 8-byte PDE in kvm_mmu_child_role()
2360 return ERR_PTR(-EEXIST); in kvm_mmu_get_child_sp()
2370 iterator->addr = addr; in shadow_walk_init_using_root()
2371 iterator->shadow_addr = root; in shadow_walk_init_using_root()
2372 iterator->level = vcpu->arch.mmu->root_role.level; in shadow_walk_init_using_root()
2374 if (iterator->level >= PT64_ROOT_4LEVEL && in shadow_walk_init_using_root()
2375 vcpu->arch.mmu->cpu_role.base.level < PT64_ROOT_4LEVEL && in shadow_walk_init_using_root()
2376 !vcpu->arch.mmu->root_role.direct) in shadow_walk_init_using_root()
2377 iterator->level = PT32E_ROOT_LEVEL; in shadow_walk_init_using_root()
2379 if (iterator->level == PT32E_ROOT_LEVEL) { in shadow_walk_init_using_root()
2381 * prev_root is currently only used for 64-bit hosts. So only in shadow_walk_init_using_root()
2384 BUG_ON(root != vcpu->arch.mmu->root.hpa); in shadow_walk_init_using_root()
2386 iterator->shadow_addr in shadow_walk_init_using_root()
2387 = vcpu->arch.mmu->pae_root[(addr >> 30) & 3]; in shadow_walk_init_using_root()
2388 iterator->shadow_addr &= SPTE_BASE_ADDR_MASK; in shadow_walk_init_using_root()
2389 --iterator->level; in shadow_walk_init_using_root()
2390 if (!iterator->shadow_addr) in shadow_walk_init_using_root()
2391 iterator->level = 0; in shadow_walk_init_using_root()
2398 shadow_walk_init_using_root(iterator, vcpu, vcpu->arch.mmu->root.hpa, in shadow_walk_init()
2404 if (iterator->level < PG_LEVEL_4K) in shadow_walk_okay()
2407 iterator->index = SPTE_INDEX(iterator->addr, iterator->level); in shadow_walk_okay()
2408 iterator->sptep = ((u64 *)__va(iterator->shadow_addr)) + iterator->index; in shadow_walk_okay()
2415 if (!is_shadow_present_pte(spte) || is_last_spte(spte, iterator->level)) { in __shadow_walk_next()
2416 iterator->level = 0; in __shadow_walk_next()
2420 iterator->shadow_addr = spte & SPTE_BASE_ADDR_MASK; in __shadow_walk_next()
2421 --iterator->level; in __shadow_walk_next()
2426 __shadow_walk_next(iterator, *iterator->sptep); in shadow_walk_next()
2445 spte = make_nonleaf_spte(sp->spt, sp_ad_disabled(sp)); in __link_shadow_page()
2452 * The non-direct sub-pagetable must be updated before linking. For in __link_shadow_page()
2454 * kvm_mmu_find_shadow_page() without write-protecting the gfn, in __link_shadow_page()
2455 * so sp->unsync can be true or false. For higher level non-direct in __link_shadow_page()
2457 * FNAME(fetch)(), so sp->unsync_children can only be false. in __link_shadow_page()
2460 if (WARN_ON_ONCE(sp->unsync_children) || sp->unsync) in __link_shadow_page()
2467 __link_shadow_page(vcpu->kvm, &vcpu->arch.mmu_pte_list_desc_cache, sptep, sp, true); in link_shadow_page()
2479 * sp's access: allow writable in the read-only sp, in validate_direct_spte()
2484 if (child->role.access == direct_access) in validate_direct_spte()
2487 drop_parent_pte(vcpu->kvm, child, sptep); in validate_direct_spte()
2488 kvm_flush_remote_tlbs_sptep(vcpu->kvm, sptep); in validate_direct_spte()
2492 /* Returns the number of zapped non-leaf child shadow pages. */
2501 if (is_last_spte(pte, sp->role.level)) { in mmu_page_zap_pte()
2513 child->role.guest_mode && !child->parent_ptes.val) in mmu_page_zap_pte()
2531 zapped += mmu_page_zap_pte(kvm, sp, sp->spt + i, invalid_list); in kvm_mmu_page_unlink_children()
2541 while ((sptep = rmap_get_first(&sp->parent_ptes, &iter))) in kvm_mmu_unlink_parents()
2553 if (parent->role.level == PG_LEVEL_4K) in mmu_zap_unsync_children()
2576 lockdep_assert_held_write(&kvm->mmu_lock); in __kvm_mmu_prepare_zap_page()
2578 ++kvm->stat.mmu_shadow_zapped; in __kvm_mmu_prepare_zap_page()
2586 if (!sp->role.invalid && sp_has_gptes(sp)) in __kvm_mmu_prepare_zap_page()
2589 if (sp->unsync) in __kvm_mmu_prepare_zap_page()
2591 if (!sp->root_count) { in __kvm_mmu_prepare_zap_page()
2598 * !sp->root_count. in __kvm_mmu_prepare_zap_page()
2600 if (sp->role.invalid) in __kvm_mmu_prepare_zap_page()
2601 list_add(&sp->link, invalid_list); in __kvm_mmu_prepare_zap_page()
2603 list_move(&sp->link, invalid_list); in __kvm_mmu_prepare_zap_page()
2610 list_del(&sp->link); in __kvm_mmu_prepare_zap_page()
2620 if (sp->nx_huge_page_disallowed) in __kvm_mmu_prepare_zap_page()
2623 sp->role.invalid = 1; in __kvm_mmu_prepare_zap_page()
2653 * the page tables and see changes to vcpu->mode here. The barrier in kvm_mmu_commit_zap_page()
2663 WARN_ON_ONCE(!sp->role.invalid || sp->root_count); in kvm_mmu_commit_zap_page()
2677 if (list_empty(&kvm->arch.active_mmu_pages)) in kvm_mmu_zap_oldest_mmu_pages()
2681 list_for_each_entry_safe_reverse(sp, tmp, &kvm->arch.active_mmu_pages, link) { in kvm_mmu_zap_oldest_mmu_pages()
2686 if (sp->root_count) in kvm_mmu_zap_oldest_mmu_pages()
2701 kvm->stat.mmu_recycled += total_zapped; in kvm_mmu_zap_oldest_mmu_pages()
2707 if (kvm->arch.n_max_mmu_pages > kvm->arch.n_used_mmu_pages) in kvm_mmu_available_pages()
2708 return kvm->arch.n_max_mmu_pages - in kvm_mmu_available_pages()
2709 kvm->arch.n_used_mmu_pages; in kvm_mmu_available_pages()
2716 unsigned long avail = kvm_mmu_available_pages(vcpu->kvm); in make_mmu_pages_available()
2721 kvm_mmu_zap_oldest_mmu_pages(vcpu->kvm, KVM_REFILL_PAGES - avail); in make_mmu_pages_available()
2726 * four pages, e.g. for PAE roots or for 5-level paging. Temporarily in make_mmu_pages_available()
2732 if (!kvm_mmu_available_pages(vcpu->kvm)) in make_mmu_pages_available()
2733 return -ENOSPC; in make_mmu_pages_available()
2743 write_lock(&kvm->mmu_lock); in kvm_mmu_change_mmu_pages()
2745 if (kvm->arch.n_used_mmu_pages > goal_nr_mmu_pages) { in kvm_mmu_change_mmu_pages()
2746 kvm_mmu_zap_oldest_mmu_pages(kvm, kvm->arch.n_used_mmu_pages - in kvm_mmu_change_mmu_pages()
2749 goal_nr_mmu_pages = kvm->arch.n_used_mmu_pages; in kvm_mmu_change_mmu_pages()
2752 kvm->arch.n_max_mmu_pages = goal_nr_mmu_pages; in kvm_mmu_change_mmu_pages()
2754 write_unlock(&kvm->mmu_lock); in kvm_mmu_change_mmu_pages()
2764 write_lock(&kvm->mmu_lock); in kvm_mmu_unprotect_page()
2770 write_unlock(&kvm->mmu_lock); in kvm_mmu_unprotect_page()
2780 if (vcpu->arch.mmu->root_role.direct) in kvm_mmu_unprotect_page_virt()
2785 r = kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT); in kvm_mmu_unprotect_page_virt()
2793 ++kvm->stat.mmu_unsync; in kvm_unsync_page()
2794 sp->unsync = 1; in kvm_unsync_page()
2802 * were marked unsync (or if there is no shadow page), -EPERM if the SPTE must
2803 * be write-protected.
2812 * Force write-protection if the page is being tracked. Note, the page in mmu_try_to_unsync_pages()
2813 * track machinery is used to write-protect upper-level shadow pages, in mmu_try_to_unsync_pages()
2817 return -EPERM; in mmu_try_to_unsync_pages()
2820 * The page is not write-tracked, mark existing shadow pages unsync in mmu_try_to_unsync_pages()
2827 return -EPERM; in mmu_try_to_unsync_pages()
2829 if (sp->unsync) in mmu_try_to_unsync_pages()
2833 return -EEXIST; in mmu_try_to_unsync_pages()
2844 spin_lock(&kvm->arch.mmu_unsync_pages_lock); in mmu_try_to_unsync_pages()
2850 * possible as clearing sp->unsync _must_ hold mmu_lock in mmu_try_to_unsync_pages()
2851 * for write, i.e. unsync cannot transition from 1->0 in mmu_try_to_unsync_pages()
2854 if (READ_ONCE(sp->unsync)) in mmu_try_to_unsync_pages()
2858 WARN_ON_ONCE(sp->role.level != PG_LEVEL_4K); in mmu_try_to_unsync_pages()
2862 spin_unlock(&kvm->arch.mmu_unsync_pages_lock); in mmu_try_to_unsync_pages()
2869 * before the page had been marked as unsync-ed, something like the in mmu_try_to_unsync_pages()
2873 * --------------------------------------------------------------------- in mmu_try_to_unsync_pages()
2886 * 2.3 Walking of unsync pages sees sp->unsync is in mmu_try_to_unsync_pages()
2895 * (sp->unsync = true) in mmu_try_to_unsync_pages()
2911 int level = sp->role.level; in mmu_set_spte()
2919 bool host_writable = !fault || fault->map_writable; in mmu_set_spte()
2920 bool prefetch = !fault || fault->prefetch; in mmu_set_spte()
2921 bool write_fault = fault && fault->write; in mmu_set_spte()
2924 vcpu->stat.pf_mmio_spte_created++; in mmu_set_spte()
2939 drop_parent_pte(vcpu->kvm, child, sptep); in mmu_set_spte()
2942 drop_spte(vcpu->kvm, sptep); in mmu_set_spte()
2964 kvm_flush_remote_tlbs_gfn(vcpu->kvm, gfn, level); in mmu_set_spte()
2983 unsigned int access = sp->role.access; in direct_pte_prefetch_many()
2990 return -1; in direct_pte_prefetch_many()
2992 ret = gfn_to_page_many_atomic(slot, gfn, pages, end - start); in direct_pte_prefetch_many()
2994 return -1; in direct_pte_prefetch_many()
3011 WARN_ON_ONCE(!sp->role.direct); in __direct_pte_prefetch()
3013 i = spte_index(sptep) & ~(PTE_PREFETCH_NUM - 1); in __direct_pte_prefetch()
3014 spte = sp->spt + i; in __direct_pte_prefetch()
3044 if (sp->role.level > PG_LEVEL_4K) in direct_pte_prefetch()
3051 if (unlikely(vcpu->kvm->mmu_invalidate_in_progress)) in direct_pte_prefetch()
3065 * - Check mmu_invalidate_retry_gfn() after grabbing the mapping level, before
3069 * - Hold mmu_lock AND ensure there is no in-progress MMU notifier invalidation
3073 * - Do not use the result to install new mappings, e.g. use the host mapping
3094 * Note, using the already-retrieved memslot and __gfn_to_hva_memslot() in host_pfn_mapping_level()
3097 * read-only memslots due to gfn_to_hva() assuming writes. Earlier in host_pfn_mapping_level()
3099 * read-only memslot. in host_pfn_mapping_level()
3111 * Read each entry once. As above, a non-leaf entry can be promoted to in host_pfn_mapping_level()
3112 * a huge page _during_ this walk. Re-reading the entry could send the in host_pfn_mapping_level()
3117 pgd = READ_ONCE(*pgd_offset(kvm->mm, hva)); in host_pfn_mapping_level()
3154 for ( ; max_level > PG_LEVEL_4K; max_level--) { in __kvm_mmu_max_mapping_level()
3156 if (!linfo->disallow_lpage) in __kvm_mmu_max_mapping_level()
3182 struct kvm_memory_slot *slot = fault->slot; in kvm_mmu_hugepage_adjust()
3185 fault->huge_page_disallowed = fault->exec && fault->nx_huge_page_workaround_enabled; in kvm_mmu_hugepage_adjust()
3187 if (unlikely(fault->max_level == PG_LEVEL_4K)) in kvm_mmu_hugepage_adjust()
3190 if (is_error_noslot_pfn(fault->pfn)) in kvm_mmu_hugepage_adjust()
3200 fault->req_level = __kvm_mmu_max_mapping_level(vcpu->kvm, slot, in kvm_mmu_hugepage_adjust()
3201 fault->gfn, fault->max_level, in kvm_mmu_hugepage_adjust()
3202 fault->is_private); in kvm_mmu_hugepage_adjust()
3203 if (fault->req_level == PG_LEVEL_4K || fault->huge_page_disallowed) in kvm_mmu_hugepage_adjust()
3210 fault->goal_level = fault->req_level; in kvm_mmu_hugepage_adjust()
3211 mask = KVM_PAGES_PER_HPAGE(fault->goal_level) - 1; in kvm_mmu_hugepage_adjust()
3212 VM_BUG_ON((fault->gfn & mask) != (fault->pfn & mask)); in kvm_mmu_hugepage_adjust()
3213 fault->pfn &= ~mask; in kvm_mmu_hugepage_adjust()
3219 cur_level == fault->goal_level && in disallowed_hugepage_adjust()
3222 spte_to_child_sp(spte)->nx_huge_page_disallowed) { in disallowed_hugepage_adjust()
3230 u64 page_mask = KVM_PAGES_PER_HPAGE(cur_level) - in disallowed_hugepage_adjust()
3231 KVM_PAGES_PER_HPAGE(cur_level - 1); in disallowed_hugepage_adjust()
3232 fault->pfn |= fault->gfn & page_mask; in disallowed_hugepage_adjust()
3233 fault->goal_level--; in disallowed_hugepage_adjust()
3242 gfn_t base_gfn = fault->gfn; in direct_map()
3247 for_each_shadow_entry(vcpu, fault->addr, it) { in direct_map()
3252 if (fault->nx_huge_page_workaround_enabled) in direct_map()
3255 base_gfn = gfn_round_for_level(fault->gfn, it.level); in direct_map()
3256 if (it.level == fault->goal_level) in direct_map()
3260 if (sp == ERR_PTR(-EEXIST)) in direct_map()
3264 if (fault->huge_page_disallowed) in direct_map()
3265 account_nx_huge_page(vcpu->kvm, sp, in direct_map()
3266 fault->req_level >= it.level); in direct_map()
3269 if (WARN_ON_ONCE(it.level != fault->goal_level)) in direct_map()
3270 return -EFAULT; in direct_map()
3272 ret = mmu_set_spte(vcpu, fault->slot, it.sptep, ACC_ALL, in direct_map()
3273 base_gfn, fault->pfn, fault); in direct_map()
3290 if (is_sigpending_pfn(fault->pfn)) { in kvm_handle_error_pfn()
3292 return -EINTR; in kvm_handle_error_pfn()
3300 if (fault->pfn == KVM_PFN_ERR_RO_FAULT) in kvm_handle_error_pfn()
3303 if (fault->pfn == KVM_PFN_ERR_HWPOISON) { in kvm_handle_error_pfn()
3304 kvm_send_hwpoison_signal(fault->slot, fault->gfn); in kvm_handle_error_pfn()
3308 return -EFAULT; in kvm_handle_error_pfn()
3315 gva_t gva = fault->is_tdp ? 0 : fault->addr; in kvm_handle_noslot_fault()
3317 vcpu_cache_mmio_info(vcpu, gva, fault->gfn, in kvm_handle_noslot_fault()
3335 if (unlikely(fault->gfn > kvm_mmu_max_gfn())) in kvm_handle_noslot_fault()
3349 if (fault->rsvd) in page_fault_can_be_fast()
3363 * the fault was caused by a write-protection violation. If the in page_fault_can_be_fast()
3364 * SPTE is MMU-writable (determined later), the fault can be fixed in page_fault_can_be_fast()
3367 if (!fault->present) in page_fault_can_be_fast()
3374 return fault->write; in page_fault_can_be_fast()
3393 * so non-PML cases won't be impacted. in fast_pf_fix_direct_spte()
3401 mark_page_dirty_in_slot(vcpu->kvm, fault->slot, fault->gfn); in fast_pf_fix_direct_spte()
3408 if (fault->exec) in is_access_allowed()
3411 if (fault->write) in is_access_allowed()
3420 * gpa, and sets *spte to the spte value. This spte may be non-preset. If no
3424 * - Must be called between walk_shadow_page_lockless_{begin,end}.
3425 * - The returned sptep must not be used after walk_shadow_page_lockless_end.
3461 sptep = kvm_tdp_mmu_fast_pf_get_last_sptep(vcpu, fault->addr, &spte); in fast_page_fault()
3463 sptep = fast_pf_get_last_sptep(vcpu, fault->addr, &spte); in fast_page_fault()
3477 if (!is_last_spte(spte, sp->role.level)) in fast_page_fault()
3500 * uses A/D bits for non-nested MMUs. Thus, if A/D bits are in fast_page_fault()
3501 * enabled, the SPTE can't be an access-tracked SPTE. in fast_page_fault()
3507 * To keep things simple, only SPTEs that are MMU-writable can in fast_page_fault()
3509 * that were write-protected for dirty-logging or access in fast_page_fault()
3514 * shadow-present, i.e. except for access tracking restoration in fast_page_fault()
3517 if (fault->write && is_mmu_writable_spte(spte)) { in fast_page_fault()
3521 * Do not fix write-permission on the large spte when in fast_page_fault()
3523 * first page into the dirty-bitmap in in fast_page_fault()
3530 if (sp->role.level > PG_LEVEL_4K && in fast_page_fault()
3531 kvm_slot_dirty_track_enabled(fault->slot)) in fast_page_fault()
3561 vcpu->stat.pf_fast++; in fast_page_fault()
3580 else if (!--sp->root_count && sp->role.invalid) in mmu_free_root_page()
3600 && VALID_PAGE(mmu->root.hpa); in kvm_mmu_free_roots()
3605 VALID_PAGE(mmu->prev_roots[i].hpa)) in kvm_mmu_free_roots()
3612 write_lock(&kvm->mmu_lock); in kvm_mmu_free_roots()
3616 mmu_free_root_page(kvm, &mmu->prev_roots[i].hpa, in kvm_mmu_free_roots()
3620 if (kvm_mmu_is_dummy_root(mmu->root.hpa)) { in kvm_mmu_free_roots()
3622 } else if (root_to_sp(mmu->root.hpa)) { in kvm_mmu_free_roots()
3623 mmu_free_root_page(kvm, &mmu->root.hpa, &invalid_list); in kvm_mmu_free_roots()
3624 } else if (mmu->pae_root) { in kvm_mmu_free_roots()
3626 if (!IS_VALID_PAE_ROOT(mmu->pae_root[i])) in kvm_mmu_free_roots()
3629 mmu_free_root_page(kvm, &mmu->pae_root[i], in kvm_mmu_free_roots()
3631 mmu->pae_root[i] = INVALID_PAE_ROOT; in kvm_mmu_free_roots()
3634 mmu->root.hpa = INVALID_PAGE; in kvm_mmu_free_roots()
3635 mmu->root.pgd = 0; in kvm_mmu_free_roots()
3639 write_unlock(&kvm->mmu_lock); in kvm_mmu_free_roots()
3654 WARN_ON_ONCE(mmu->root_role.guest_mode); in kvm_mmu_free_guest_mode_roots()
3657 root_hpa = mmu->prev_roots[i].hpa; in kvm_mmu_free_guest_mode_roots()
3662 if (!sp || sp->role.guest_mode) in kvm_mmu_free_guest_mode_roots()
3673 union kvm_mmu_page_role role = vcpu->arch.mmu->root_role; in mmu_alloc_root()
3683 ++sp->root_count; in mmu_alloc_root()
3685 return __pa(sp->spt); in mmu_alloc_root()
3690 struct kvm_mmu *mmu = vcpu->arch.mmu; in mmu_alloc_direct_roots()
3691 u8 shadow_root_level = mmu->root_role.level; in mmu_alloc_direct_roots()
3696 write_lock(&vcpu->kvm->mmu_lock); in mmu_alloc_direct_roots()
3703 mmu->root.hpa = root; in mmu_alloc_direct_roots()
3706 mmu->root.hpa = root; in mmu_alloc_direct_roots()
3708 if (WARN_ON_ONCE(!mmu->pae_root)) { in mmu_alloc_direct_roots()
3709 r = -EIO; in mmu_alloc_direct_roots()
3714 WARN_ON_ONCE(IS_VALID_PAE_ROOT(mmu->pae_root[i])); in mmu_alloc_direct_roots()
3716 root = mmu_alloc_root(vcpu, i << (30 - PAGE_SHIFT), 0, in mmu_alloc_direct_roots()
3718 mmu->pae_root[i] = root | PT_PRESENT_MASK | in mmu_alloc_direct_roots()
3721 mmu->root.hpa = __pa(mmu->pae_root); in mmu_alloc_direct_roots()
3724 r = -EIO; in mmu_alloc_direct_roots()
3729 mmu->root.pgd = 0; in mmu_alloc_direct_roots()
3731 write_unlock(&vcpu->kvm->mmu_lock); in mmu_alloc_direct_roots()
3748 mutex_lock(&kvm->slots_arch_lock); in mmu_first_shadow_root_alloc()
3766 * Both of these functions are no-ops if the target is in mmu_first_shadow_root_alloc()
3775 r = memslot_rmap_alloc(slot, slot->npages); in mmu_first_shadow_root_alloc()
3789 smp_store_release(&kvm->arch.shadow_root_allocated, true); in mmu_first_shadow_root_alloc()
3792 mutex_unlock(&kvm->slots_arch_lock); in mmu_first_shadow_root_alloc()
3798 struct kvm_mmu *mmu = vcpu->arch.mmu; in mmu_alloc_shadow_roots()
3808 mmu->root.hpa = kvm_mmu_get_dummy_root(); in mmu_alloc_shadow_roots()
3816 if (mmu->cpu_role.base.level == PT32E_ROOT_LEVEL) { in mmu_alloc_shadow_roots()
3818 pdptrs[i] = mmu->get_pdptr(vcpu, i); in mmu_alloc_shadow_roots()
3827 r = mmu_first_shadow_root_alloc(vcpu->kvm); in mmu_alloc_shadow_roots()
3831 write_lock(&vcpu->kvm->mmu_lock); in mmu_alloc_shadow_roots()
3838 * write-protect the guests page table root. in mmu_alloc_shadow_roots()
3840 if (mmu->cpu_role.base.level >= PT64_ROOT_4LEVEL) { in mmu_alloc_shadow_roots()
3842 mmu->root_role.level); in mmu_alloc_shadow_roots()
3843 mmu->root.hpa = root; in mmu_alloc_shadow_roots()
3847 if (WARN_ON_ONCE(!mmu->pae_root)) { in mmu_alloc_shadow_roots()
3848 r = -EIO; in mmu_alloc_shadow_roots()
3853 * We shadow a 32 bit page table. This may be a legacy 2-level in mmu_alloc_shadow_roots()
3854 * or a PAE 3-level page table. In either case we need to be aware that in mmu_alloc_shadow_roots()
3858 if (mmu->root_role.level >= PT64_ROOT_4LEVEL) { in mmu_alloc_shadow_roots()
3861 if (WARN_ON_ONCE(!mmu->pml4_root)) { in mmu_alloc_shadow_roots()
3862 r = -EIO; in mmu_alloc_shadow_roots()
3865 mmu->pml4_root[0] = __pa(mmu->pae_root) | pm_mask; in mmu_alloc_shadow_roots()
3867 if (mmu->root_role.level == PT64_ROOT_5LEVEL) { in mmu_alloc_shadow_roots()
3868 if (WARN_ON_ONCE(!mmu->pml5_root)) { in mmu_alloc_shadow_roots()
3869 r = -EIO; in mmu_alloc_shadow_roots()
3872 mmu->pml5_root[0] = __pa(mmu->pml4_root) | pm_mask; in mmu_alloc_shadow_roots()
3877 WARN_ON_ONCE(IS_VALID_PAE_ROOT(mmu->pae_root[i])); in mmu_alloc_shadow_roots()
3879 if (mmu->cpu_role.base.level == PT32E_ROOT_LEVEL) { in mmu_alloc_shadow_roots()
3881 mmu->pae_root[i] = INVALID_PAE_ROOT; in mmu_alloc_shadow_roots()
3888 * If shadowing 32-bit non-PAE page tables, each PAE page in mmu_alloc_shadow_roots()
3889 * directory maps one quarter of the guest's non-PAE page in mmu_alloc_shadow_roots()
3893 quadrant = (mmu->cpu_role.base.level == PT32_ROOT_LEVEL) ? i : 0; in mmu_alloc_shadow_roots()
3896 mmu->pae_root[i] = root | pm_mask; in mmu_alloc_shadow_roots()
3899 if (mmu->root_role.level == PT64_ROOT_5LEVEL) in mmu_alloc_shadow_roots()
3900 mmu->root.hpa = __pa(mmu->pml5_root); in mmu_alloc_shadow_roots()
3901 else if (mmu->root_role.level == PT64_ROOT_4LEVEL) in mmu_alloc_shadow_roots()
3902 mmu->root.hpa = __pa(mmu->pml4_root); in mmu_alloc_shadow_roots()
3904 mmu->root.hpa = __pa(mmu->pae_root); in mmu_alloc_shadow_roots()
3907 mmu->root.pgd = root_pgd; in mmu_alloc_shadow_roots()
3909 write_unlock(&vcpu->kvm->mmu_lock); in mmu_alloc_shadow_roots()
3916 struct kvm_mmu *mmu = vcpu->arch.mmu; in mmu_alloc_special_roots()
3917 bool need_pml5 = mmu->root_role.level > PT64_ROOT_4LEVEL; in mmu_alloc_special_roots()
3923 * When shadowing 32-bit or PAE NPT with 64-bit NPT, the PML4 and PDP in mmu_alloc_special_roots()
3926 * on demand, as running a 32-bit L1 VMM on 64-bit KVM is very rare. in mmu_alloc_special_roots()
3928 if (mmu->root_role.direct || in mmu_alloc_special_roots()
3929 mmu->cpu_role.base.level >= PT64_ROOT_4LEVEL || in mmu_alloc_special_roots()
3930 mmu->root_role.level < PT64_ROOT_4LEVEL) in mmu_alloc_special_roots()
3935 * of levels for the shadow page tables, e.g. all MMUs are 4-level or in mmu_alloc_special_roots()
3936 * all MMus are 5-level. Thus, this can safely require that pml5_root in mmu_alloc_special_roots()
3940 if (mmu->pae_root && mmu->pml4_root && (!need_pml5 || mmu->pml5_root)) in mmu_alloc_special_roots()
3947 if (WARN_ON_ONCE(!tdp_enabled || mmu->pae_root || mmu->pml4_root || in mmu_alloc_special_roots()
3948 (need_pml5 && mmu->pml5_root))) in mmu_alloc_special_roots()
3949 return -EIO; in mmu_alloc_special_roots()
3952 * Unlike 32-bit NPT, the PDP table doesn't need to be in low mem, and in mmu_alloc_special_roots()
3957 return -ENOMEM; in mmu_alloc_special_roots()
3971 mmu->pae_root = pae_root; in mmu_alloc_special_roots()
3972 mmu->pml4_root = pml4_root; in mmu_alloc_special_roots()
3973 mmu->pml5_root = pml5_root; in mmu_alloc_special_roots()
3982 return -ENOMEM; in mmu_alloc_special_roots()
3995 * walk before the reads of sp->unsync/sp->unsync_children here. in is_unsync_root()
3997 * Even if another CPU was marking the SP as unsync-ed simultaneously, in is_unsync_root()
4015 if (sp->unsync || sp->unsync_children) in is_unsync_root()
4026 if (vcpu->arch.mmu->root_role.direct) in kvm_mmu_sync_roots()
4029 if (!VALID_PAGE(vcpu->arch.mmu->root.hpa)) in kvm_mmu_sync_roots()
4034 if (vcpu->arch.mmu->cpu_role.base.level >= PT64_ROOT_4LEVEL) { in kvm_mmu_sync_roots()
4035 hpa_t root = vcpu->arch.mmu->root.hpa; in kvm_mmu_sync_roots()
4042 write_lock(&vcpu->kvm->mmu_lock); in kvm_mmu_sync_roots()
4044 write_unlock(&vcpu->kvm->mmu_lock); in kvm_mmu_sync_roots()
4048 write_lock(&vcpu->kvm->mmu_lock); in kvm_mmu_sync_roots()
4051 hpa_t root = vcpu->arch.mmu->pae_root[i]; in kvm_mmu_sync_roots()
4059 write_unlock(&vcpu->kvm->mmu_lock); in kvm_mmu_sync_roots()
4068 if (is_unsync_root(vcpu->arch.mmu->prev_roots[i].hpa)) in kvm_mmu_sync_prev_roots()
4072 kvm_mmu_free_roots(vcpu->kvm, vcpu->arch.mmu, roots_to_free); in kvm_mmu_sync_prev_roots()
4080 exception->error_code = 0; in nonpaging_gva_to_gpa()
4101 * That SPTE may be non-present.
4108 int leaf = -1; in get_walk()
4124 /* return true if reserved bit(s) are detected on a valid, non-MMIO SPTE. */
4152 * to detect reserved bits on non-MMIO SPTEs. i.e. buggy SPTEs. in get_mmio_spte()
4157 rsvd_check = &vcpu->arch.mmu->shadow_zero_check; in get_mmio_spte()
4159 for (level = root; level >= leaf; level--) in get_mmio_spte()
4163 pr_err("%s: reserved bits set on MMU-present spte, addr 0x%llx, hierarchy:\n", in get_mmio_spte()
4165 for (level = root; level >= leaf; level--) in get_mmio_spte()
4166 pr_err("------ spte = 0x%llx level = %d, rsvd bits = 0x%llx", in get_mmio_spte()
4184 return -EINVAL; in handle_mmio_page_fault()
4211 if (unlikely(fault->rsvd)) in page_fault_handle_page_track()
4214 if (!fault->present || !fault->write) in page_fault_handle_page_track()
4221 if (kvm_gfn_is_write_tracked(vcpu->kvm, fault->slot, fault->gfn)) in page_fault_handle_page_track()
4241 u32 id = vcpu->arch.apf.id; in alloc_apf_token()
4244 vcpu->arch.apf.id = 1; in alloc_apf_token()
4246 return (vcpu->arch.apf.id++ << 12) | vcpu->vcpu_id; in alloc_apf_token()
4256 arch.direct_map = vcpu->arch.mmu->root_role.direct; in kvm_arch_setup_async_pf()
4257 arch.cr3 = kvm_mmu_get_guest_pgd(vcpu, vcpu->arch.mmu); in kvm_arch_setup_async_pf()
4267 if ((vcpu->arch.mmu->root_role.direct != work->arch.direct_map) || in kvm_arch_async_page_ready()
4268 work->wakeup_all) in kvm_arch_async_page_ready()
4275 if (!vcpu->arch.mmu->root_role.direct && in kvm_arch_async_page_ready()
4276 work->arch.cr3 != kvm_mmu_get_guest_pgd(vcpu, vcpu->arch.mmu)) in kvm_arch_async_page_ready()
4279 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true, NULL); in kvm_arch_async_page_ready()
4302 kvm_prepare_memory_fault_exit(vcpu, fault->gfn << PAGE_SHIFT, in kvm_mmu_prepare_memory_fault_exit()
4303 PAGE_SIZE, fault->write, fault->exec, in kvm_mmu_prepare_memory_fault_exit()
4304 fault->is_private); in kvm_mmu_prepare_memory_fault_exit()
4312 if (!kvm_slot_can_be_private(fault->slot)) { in kvm_faultin_pfn_private()
4314 return -EFAULT; in kvm_faultin_pfn_private()
4317 r = kvm_gmem_get_pfn(vcpu->kvm, fault->slot, fault->gfn, &fault->pfn, in kvm_faultin_pfn_private()
4324 fault->max_level = min(kvm_max_level_for_order(max_order), in kvm_faultin_pfn_private()
4325 fault->max_level); in kvm_faultin_pfn_private()
4326 fault->map_writable = !(fault->slot->flags & KVM_MEM_READONLY); in kvm_faultin_pfn_private()
4333 struct kvm_memory_slot *slot = fault->slot; in __kvm_faultin_pfn()
4341 if (slot && (slot->flags & KVM_MEMSLOT_INVALID)) in __kvm_faultin_pfn()
4347 fault->slot = NULL; in __kvm_faultin_pfn()
4348 fault->pfn = KVM_PFN_NOSLOT; in __kvm_faultin_pfn()
4349 fault->map_writable = false; in __kvm_faultin_pfn()
4356 * when the AVIC is re-enabled. in __kvm_faultin_pfn()
4358 if (slot && slot->id == APIC_ACCESS_PAGE_PRIVATE_MEMSLOT && in __kvm_faultin_pfn()
4359 !kvm_apicv_activated(vcpu->kvm)) in __kvm_faultin_pfn()
4363 if (fault->is_private != kvm_mem_is_private(vcpu->kvm, fault->gfn)) { in __kvm_faultin_pfn()
4365 return -EFAULT; in __kvm_faultin_pfn()
4368 if (fault->is_private) in __kvm_faultin_pfn()
4372 fault->pfn = __gfn_to_pfn_memslot(slot, fault->gfn, false, false, &async, in __kvm_faultin_pfn()
4373 fault->write, &fault->map_writable, in __kvm_faultin_pfn()
4374 &fault->hva); in __kvm_faultin_pfn()
4378 if (!fault->prefetch && kvm_can_do_async_pf(vcpu)) { in __kvm_faultin_pfn()
4379 trace_kvm_try_async_get_page(fault->addr, fault->gfn); in __kvm_faultin_pfn()
4380 if (kvm_find_async_pf_gfn(vcpu, fault->gfn)) { in __kvm_faultin_pfn()
4381 trace_kvm_async_pf_repeated_fault(fault->addr, fault->gfn); in __kvm_faultin_pfn()
4384 } else if (kvm_arch_setup_async_pf(vcpu, fault->addr, fault->gfn)) { in __kvm_faultin_pfn()
4390 * Allow gup to bail on pending non-fatal signals when it's also allowed in __kvm_faultin_pfn()
4394 fault->pfn = __gfn_to_pfn_memslot(slot, fault->gfn, false, true, NULL, in __kvm_faultin_pfn()
4395 fault->write, &fault->map_writable, in __kvm_faultin_pfn()
4396 &fault->hva); in __kvm_faultin_pfn()
4405 fault->mmu_seq = vcpu->kvm->mmu_invalidate_seq; in kvm_faultin_pfn()
4412 * For mmu_lock, if there is an in-progress invalidation and the kernel in kvm_faultin_pfn()
4414 * in response to mmu_lock being contended, which is *very* counter- in kvm_faultin_pfn()
4424 * Do the pre-check even for non-preemtible kernels, i.e. even if KVM in kvm_faultin_pfn()
4429 if (fault->slot && in kvm_faultin_pfn()
4430 mmu_invalidate_retry_gfn_unsafe(vcpu->kvm, fault->mmu_seq, fault->gfn)) in kvm_faultin_pfn()
4437 if (unlikely(is_error_pfn(fault->pfn))) in kvm_faultin_pfn()
4440 if (unlikely(!fault->slot)) in kvm_faultin_pfn()
4450 if (mmu_invalidate_retry_gfn_unsafe(vcpu->kvm, fault->mmu_seq, fault->gfn)) { in kvm_faultin_pfn()
4451 kvm_release_pfn_clean(fault->pfn); in kvm_faultin_pfn()
4465 struct kvm_mmu_page *sp = root_to_sp(vcpu->arch.mmu->root.hpa); in is_page_fault_stale()
4468 if (sp && is_obsolete_sp(vcpu->kvm, sp)) in is_page_fault_stale()
4487 return fault->slot && in is_page_fault_stale()
4488 mmu_invalidate_retry_gfn(vcpu->kvm, fault->mmu_seq, fault->gfn); in is_page_fault_stale()
4496 if (WARN_ON_ONCE(kvm_mmu_is_dummy_root(vcpu->arch.mmu->root.hpa))) in direct_page_fault()
4515 write_lock(&vcpu->kvm->mmu_lock); in direct_page_fault()
4527 write_unlock(&vcpu->kvm->mmu_lock); in direct_page_fault()
4528 kvm_release_pfn_clean(fault->pfn); in direct_page_fault()
4536 fault->max_level = PG_LEVEL_2M; in nonpaging_page_fault()
4544 u32 flags = vcpu->arch.apf.host_apf_flags; in kvm_handle_page_fault()
4547 /* A 64-bit CR2 should be impossible on 32-bit KVM. */ in kvm_handle_page_fault()
4549 return -EFAULT; in kvm_handle_page_fault()
4552 vcpu->arch.l1tf_flush_l1d = true; in kvm_handle_page_fault()
4561 vcpu->arch.apf.host_apf_flags = 0; in kvm_handle_page_fault()
4595 read_lock(&vcpu->kvm->mmu_lock); in kvm_tdp_mmu_page_fault()
4603 read_unlock(&vcpu->kvm->mmu_lock); in kvm_tdp_mmu_page_fault()
4604 kvm_release_pfn_clean(fault->pfn); in kvm_tdp_mmu_page_fault()
4612 * If host MTRRs are ignored (shadow_memtype_mask is non-zero), and the in __kvm_mmu_honors_guest_mtrrs()
4613 * VM has non-coherent DMA (DMA doesn't snoop CPU caches), KVM's ABI is in __kvm_mmu_honors_guest_mtrrs()
4630 if (kvm_mmu_honors_guest_mtrrs(vcpu->kvm)) { in kvm_tdp_page_fault()
4631 for ( ; fault->max_level > PG_LEVEL_4K; --fault->max_level) { in kvm_tdp_page_fault()
4632 int page_num = KVM_PAGES_PER_HPAGE(fault->max_level); in kvm_tdp_page_fault()
4633 gfn_t base = gfn_round_for_level(fault->gfn, in kvm_tdp_page_fault()
4634 fault->max_level); in kvm_tdp_page_fault()
4651 context->page_fault = nonpaging_page_fault; in nonpaging_init_context()
4652 context->gva_to_gpa = nonpaging_gva_to_gpa; in nonpaging_init_context()
4653 context->sync_spte = NULL; in nonpaging_init_context()
4661 if (!VALID_PAGE(root->hpa)) in is_root_usable()
4664 if (!role.direct && pgd != root->pgd) in is_root_usable()
4667 sp = root_to_sp(root->hpa); in is_root_usable()
4671 return role.word == sp->role.word; in is_root_usable()
4677 * If a matching root is found, it is assigned to kvm_mmu->root and
4679 * If no match is found, kvm_mmu->root is left invalid, the LRU root is
4688 if (is_root_usable(&mmu->root, new_pgd, new_role)) in cached_root_find_and_keep_current()
4700 swap(mmu->root, mmu->prev_roots[i]); in cached_root_find_and_keep_current()
4701 if (is_root_usable(&mmu->root, new_pgd, new_role)) in cached_root_find_and_keep_current()
4711 * On entry, mmu->root is invalid.
4712 * If a matching root is found, it is assigned to kvm_mmu->root, the LRU entry
4714 * If no match is found, kvm_mmu->root is left invalid and false is returned.
4723 if (is_root_usable(&mmu->prev_roots[i], new_pgd, new_role)) in cached_root_find_without_current()
4729 swap(mmu->root, mmu->prev_roots[i]); in cached_root_find_without_current()
4731 for (; i < KVM_MMU_NUM_PREV_ROOTS - 1; i++) in cached_root_find_without_current()
4732 mmu->prev_roots[i] = mmu->prev_roots[i + 1]; in cached_root_find_without_current()
4733 mmu->prev_roots[i].hpa = INVALID_PAGE; in cached_root_find_without_current()
4741 * Limit reuse to 64-bit hosts+VMs without "special" roots in order to in fast_pgd_switch()
4744 if (VALID_PAGE(mmu->root.hpa) && !root_to_sp(mmu->root.hpa)) in fast_pgd_switch()
4747 if (VALID_PAGE(mmu->root.hpa)) in fast_pgd_switch()
4755 struct kvm_mmu *mmu = vcpu->arch.mmu; in kvm_mmu_new_pgd()
4756 union kvm_mmu_page_role new_role = mmu->root_role; in kvm_mmu_new_pgd()
4760 * will establish a valid root prior to the next VM-Enter. in kvm_mmu_new_pgd()
4762 if (!fast_pgd_switch(vcpu->kvm, mmu, new_pgd, new_role)) in kvm_mmu_new_pgd()
4780 * switching to a new CR3, that GVA->GPA mapping may no longer be in kvm_mmu_new_pgd()
4791 struct kvm_mmu_page *sp = root_to_sp(vcpu->arch.mmu->root.hpa); in kvm_mmu_new_pgd()
4836 rsvd_check->bad_mt_xwr = 0; in __reset_rsvds_bits_mask()
4851 * Non-leaf PML4Es and PDPEs reserve bit 8 (which would be the G bit for in __reset_rsvds_bits_mask()
4860 rsvd_check->rsvd_bits_mask[0][1] = 0; in __reset_rsvds_bits_mask()
4861 rsvd_check->rsvd_bits_mask[0][0] = 0; in __reset_rsvds_bits_mask()
4862 rsvd_check->rsvd_bits_mask[1][0] = in __reset_rsvds_bits_mask()
4863 rsvd_check->rsvd_bits_mask[0][0]; in __reset_rsvds_bits_mask()
4866 rsvd_check->rsvd_bits_mask[1][1] = 0; in __reset_rsvds_bits_mask()
4872 rsvd_check->rsvd_bits_mask[1][1] = rsvd_bits(17, 21); in __reset_rsvds_bits_mask()
4875 rsvd_check->rsvd_bits_mask[1][1] = rsvd_bits(13, 21); in __reset_rsvds_bits_mask()
4878 rsvd_check->rsvd_bits_mask[0][2] = rsvd_bits(63, 63) | in __reset_rsvds_bits_mask()
4882 rsvd_check->rsvd_bits_mask[0][1] = high_bits_rsvd; /* PDE */ in __reset_rsvds_bits_mask()
4883 rsvd_check->rsvd_bits_mask[0][0] = high_bits_rsvd; /* PTE */ in __reset_rsvds_bits_mask()
4884 rsvd_check->rsvd_bits_mask[1][1] = high_bits_rsvd | in __reset_rsvds_bits_mask()
4886 rsvd_check->rsvd_bits_mask[1][0] = in __reset_rsvds_bits_mask()
4887 rsvd_check->rsvd_bits_mask[0][0]; in __reset_rsvds_bits_mask()
4890 rsvd_check->rsvd_bits_mask[0][4] = high_bits_rsvd | in __reset_rsvds_bits_mask()
4893 rsvd_check->rsvd_bits_mask[1][4] = in __reset_rsvds_bits_mask()
4894 rsvd_check->rsvd_bits_mask[0][4]; in __reset_rsvds_bits_mask()
4897 rsvd_check->rsvd_bits_mask[0][3] = high_bits_rsvd | in __reset_rsvds_bits_mask()
4900 rsvd_check->rsvd_bits_mask[0][2] = high_bits_rsvd | in __reset_rsvds_bits_mask()
4902 rsvd_check->rsvd_bits_mask[0][1] = high_bits_rsvd; in __reset_rsvds_bits_mask()
4903 rsvd_check->rsvd_bits_mask[0][0] = high_bits_rsvd; in __reset_rsvds_bits_mask()
4904 rsvd_check->rsvd_bits_mask[1][3] = in __reset_rsvds_bits_mask()
4905 rsvd_check->rsvd_bits_mask[0][3]; in __reset_rsvds_bits_mask()
4906 rsvd_check->rsvd_bits_mask[1][2] = high_bits_rsvd | in __reset_rsvds_bits_mask()
4909 rsvd_check->rsvd_bits_mask[1][1] = high_bits_rsvd | in __reset_rsvds_bits_mask()
4911 rsvd_check->rsvd_bits_mask[1][0] = in __reset_rsvds_bits_mask()
4912 rsvd_check->rsvd_bits_mask[0][0]; in __reset_rsvds_bits_mask()
4920 __reset_rsvds_bits_mask(&context->guest_rsvd_check, in reset_guest_rsvds_bits_mask()
4921 vcpu->arch.reserved_gpa_bits, in reset_guest_rsvds_bits_mask()
4922 context->cpu_role.base.level, is_efer_nx(context), in reset_guest_rsvds_bits_mask()
4941 rsvd_check->rsvd_bits_mask[0][4] = high_bits_rsvd | rsvd_bits(3, 7); in __reset_rsvds_bits_mask_ept()
4942 rsvd_check->rsvd_bits_mask[0][3] = high_bits_rsvd | rsvd_bits(3, 7); in __reset_rsvds_bits_mask_ept()
4943 rsvd_check->rsvd_bits_mask[0][2] = high_bits_rsvd | rsvd_bits(3, 6) | large_1g_rsvd; in __reset_rsvds_bits_mask_ept()
4944 rsvd_check->rsvd_bits_mask[0][1] = high_bits_rsvd | rsvd_bits(3, 6) | large_2m_rsvd; in __reset_rsvds_bits_mask_ept()
4945 rsvd_check->rsvd_bits_mask[0][0] = high_bits_rsvd; in __reset_rsvds_bits_mask_ept()
4948 rsvd_check->rsvd_bits_mask[1][4] = rsvd_check->rsvd_bits_mask[0][4]; in __reset_rsvds_bits_mask_ept()
4949 rsvd_check->rsvd_bits_mask[1][3] = rsvd_check->rsvd_bits_mask[0][3]; in __reset_rsvds_bits_mask_ept()
4950 rsvd_check->rsvd_bits_mask[1][2] = high_bits_rsvd | rsvd_bits(12, 29) | large_1g_rsvd; in __reset_rsvds_bits_mask_ept()
4951 rsvd_check->rsvd_bits_mask[1][1] = high_bits_rsvd | rsvd_bits(12, 20) | large_2m_rsvd; in __reset_rsvds_bits_mask_ept()
4952 rsvd_check->rsvd_bits_mask[1][0] = rsvd_check->rsvd_bits_mask[0][0]; in __reset_rsvds_bits_mask_ept()
4963 rsvd_check->bad_mt_xwr = bad_mt_xwr; in __reset_rsvds_bits_mask_ept()
4969 __reset_rsvds_bits_mask_ept(&context->guest_rsvd_check, in reset_rsvds_bits_mask_ept()
4970 vcpu->arch.reserved_gpa_bits, execonly, in reset_rsvds_bits_mask_ept()
4989 /* KVM doesn't use 2-level page tables for the shadow MMU. */ in reset_shadow_zero_bits_mask()
4994 WARN_ON_ONCE(context->root_role.level < PT32E_ROOT_LEVEL); in reset_shadow_zero_bits_mask()
4996 shadow_zero_check = &context->shadow_zero_check; in reset_shadow_zero_bits_mask()
4998 context->root_role.level, in reset_shadow_zero_bits_mask()
4999 context->root_role.efer_nx, in reset_shadow_zero_bits_mask()
5006 for (i = context->root_role.level; --i >= 0;) { in reset_shadow_zero_bits_mask()
5013 shadow_zero_check->rsvd_bits_mask[0][i] |= shadow_me_mask; in reset_shadow_zero_bits_mask()
5014 shadow_zero_check->rsvd_bits_mask[1][i] |= shadow_me_mask; in reset_shadow_zero_bits_mask()
5015 shadow_zero_check->rsvd_bits_mask[0][i] &= ~shadow_me_value; in reset_shadow_zero_bits_mask()
5016 shadow_zero_check->rsvd_bits_mask[1][i] &= ~shadow_me_value; in reset_shadow_zero_bits_mask()
5029 * possible, however, kvm currently does not do execution-protection.
5036 shadow_zero_check = &context->shadow_zero_check; in reset_tdp_shadow_zero_bits_mask()
5040 context->root_role.level, true, in reset_tdp_shadow_zero_bits_mask()
5051 for (i = context->root_role.level; --i >= 0;) { in reset_tdp_shadow_zero_bits_mask()
5052 shadow_zero_check->rsvd_bits_mask[0][i] &= ~shadow_me_mask; in reset_tdp_shadow_zero_bits_mask()
5053 shadow_zero_check->rsvd_bits_mask[1][i] &= ~shadow_me_mask; in reset_tdp_shadow_zero_bits_mask()
5064 __reset_rsvds_bits_mask_ept(&context->shadow_zero_check, in reset_ept_shadow_zero_bits_mask()
5092 for (byte = 0; byte < ARRAY_SIZE(mmu->permissions); ++byte) { in update_permission_bitmask()
5100 /* Faults from writes to non-writable pages */ in update_permission_bitmask()
5104 /* Faults from fetches of non-executable pages*/ in update_permission_bitmask()
5128 * SMAP:kernel-mode data accesses from user-mode in update_permission_bitmask()
5132 * - X86_CR4_SMAP is set in CR4 in update_permission_bitmask()
5133 * - A user page is accessed in update_permission_bitmask()
5134 * - The access is not a fetch in update_permission_bitmask()
5135 * - The access is supervisor mode in update_permission_bitmask()
5136 * - If implicit supervisor access or X86_EFLAGS_AC is clear in update_permission_bitmask()
5147 mmu->permissions[byte] = ff | uf | wf | smepf | smapf; in update_permission_bitmask()
5153 * user-mode addresses based on the value in the PKRU register. Protection
5162 * - PK is always zero unless CR4.PKE=1 and EFER.LMA=1
5163 * - PK is always zero if RSVD=1 (reserved bit set) or F=1 (instruction fetch)
5164 * - PK is always zero if U=0 in the page tables
5165 * - PKRU.WD is ignored if CR0.WP=0 and the access is a supervisor access.
5180 mmu->pkru_mask = 0; in update_pkru_bitmask()
5187 for (bit = 0; bit < ARRAY_SIZE(mmu->permissions); ++bit) { in update_pkru_bitmask()
5215 mmu->pkru_mask |= (pkey_bits & 3) << pfec; in update_pkru_bitmask()
5232 context->page_fault = paging64_page_fault; in paging64_init_context()
5233 context->gva_to_gpa = paging64_gva_to_gpa; in paging64_init_context()
5234 context->sync_spte = paging64_sync_spte; in paging64_init_context()
5239 context->page_fault = paging32_page_fault; in paging32_init_context()
5240 context->gva_to_gpa = paging32_gva_to_gpa; in paging32_init_context()
5241 context->sync_spte = paging32_sync_spte; in paging32_init_context()
5252 role.ext.valid = 1; in kvm_calc_cpu_role()
5273 role.ext.cr4_smep = ____is_cr4_smep(regs); in kvm_calc_cpu_role()
5274 role.ext.cr4_smap = ____is_cr4_smap(regs); in kvm_calc_cpu_role()
5275 role.ext.cr4_pse = ____is_cr4_pse(regs); in kvm_calc_cpu_role()
5278 role.ext.cr4_pke = ____is_efer_lma(regs) && ____is_cr4_pke(regs); in kvm_calc_cpu_role()
5279 role.ext.cr4_la57 = ____is_efer_lma(regs) && ____is_cr4_la57(regs); in kvm_calc_cpu_role()
5280 role.ext.efer_lma = ____is_efer_lma(regs); in kvm_calc_cpu_role()
5295 mmu->cpu_role.base.cr0_wp = cr0_wp; in __kvm_mmu_refresh_passthrough_bits()
5305 /* Use 5-level TDP if and only if it's useful/necessary. */ in kvm_mmu_get_tdp_level()
5334 struct kvm_mmu *context = &vcpu->arch.root_mmu; in init_kvm_tdp_mmu()
5337 if (cpu_role.as_u64 == context->cpu_role.as_u64 && in init_kvm_tdp_mmu()
5338 root_role.word == context->root_role.word) in init_kvm_tdp_mmu()
5341 context->cpu_role.as_u64 = cpu_role.as_u64; in init_kvm_tdp_mmu()
5342 context->root_role.word = root_role.word; in init_kvm_tdp_mmu()
5343 context->page_fault = kvm_tdp_page_fault; in init_kvm_tdp_mmu()
5344 context->sync_spte = NULL; in init_kvm_tdp_mmu()
5345 context->get_guest_pgd = get_guest_cr3; in init_kvm_tdp_mmu()
5346 context->get_pdptr = kvm_pdptr_read; in init_kvm_tdp_mmu()
5347 context->inject_page_fault = kvm_inject_page_fault; in init_kvm_tdp_mmu()
5350 context->gva_to_gpa = nonpaging_gva_to_gpa; in init_kvm_tdp_mmu()
5352 context->gva_to_gpa = paging64_gva_to_gpa; in init_kvm_tdp_mmu()
5354 context->gva_to_gpa = paging32_gva_to_gpa; in init_kvm_tdp_mmu()
5364 if (cpu_role.as_u64 == context->cpu_role.as_u64 && in shadow_mmu_init_context()
5365 root_role.word == context->root_role.word) in shadow_mmu_init_context()
5368 context->cpu_role.as_u64 = cpu_role.as_u64; in shadow_mmu_init_context()
5369 context->root_role.word = root_role.word; in shadow_mmu_init_context()
5385 struct kvm_mmu *context = &vcpu->arch.root_mmu; in kvm_init_shadow_mmu()
5390 /* KVM uses PAE paging whenever the guest isn't using 64-bit paging. */ in kvm_init_shadow_mmu()
5396 * notably for huge SPTEs if iTLB multi-hit mitigation is enabled and in kvm_init_shadow_mmu()
5398 * The iTLB multi-hit workaround can be toggled at any time, so assume in kvm_init_shadow_mmu()
5399 * NX can be used by any non-nested shadow MMU to avoid having to reset in kvm_init_shadow_mmu()
5410 struct kvm_mmu *context = &vcpu->arch.guest_mmu; in kvm_init_shadow_npt_mmu()
5451 role.ext.word = 0; in kvm_calc_shadow_ept_root_page_role()
5452 role.ext.execonly = execonly; in kvm_calc_shadow_ept_root_page_role()
5453 role.ext.valid = 1; in kvm_calc_shadow_ept_root_page_role()
5462 struct kvm_mmu *context = &vcpu->arch.guest_mmu; in kvm_init_shadow_ept_mmu()
5468 if (new_mode.as_u64 != context->cpu_role.as_u64) { in kvm_init_shadow_ept_mmu()
5470 context->cpu_role.as_u64 = new_mode.as_u64; in kvm_init_shadow_ept_mmu()
5471 context->root_role.word = new_mode.base.word; in kvm_init_shadow_ept_mmu()
5473 context->page_fault = ept_page_fault; in kvm_init_shadow_ept_mmu()
5474 context->gva_to_gpa = ept_gva_to_gpa; in kvm_init_shadow_ept_mmu()
5475 context->sync_spte = ept_sync_spte; in kvm_init_shadow_ept_mmu()
5478 context->pkru_mask = 0; in kvm_init_shadow_ept_mmu()
5490 struct kvm_mmu *context = &vcpu->arch.root_mmu; in init_kvm_softmmu()
5494 context->get_guest_pgd = get_guest_cr3; in init_kvm_softmmu()
5495 context->get_pdptr = kvm_pdptr_read; in init_kvm_softmmu()
5496 context->inject_page_fault = kvm_inject_page_fault; in init_kvm_softmmu()
5502 struct kvm_mmu *g_context = &vcpu->arch.nested_mmu; in init_kvm_nested_mmu()
5504 if (new_mode.as_u64 == g_context->cpu_role.as_u64) in init_kvm_nested_mmu()
5507 g_context->cpu_role.as_u64 = new_mode.as_u64; in init_kvm_nested_mmu()
5508 g_context->get_guest_pgd = get_guest_cr3; in init_kvm_nested_mmu()
5509 g_context->get_pdptr = kvm_pdptr_read; in init_kvm_nested_mmu()
5510 g_context->inject_page_fault = kvm_inject_page_fault; in init_kvm_nested_mmu()
5516 g_context->sync_spte = NULL; in init_kvm_nested_mmu()
5519 * Note that arch.mmu->gva_to_gpa translates l2_gpa to l1_gpa using in init_kvm_nested_mmu()
5527 g_context->gva_to_gpa = nonpaging_gva_to_gpa; in init_kvm_nested_mmu()
5529 g_context->gva_to_gpa = paging64_gva_to_gpa; in init_kvm_nested_mmu()
5531 g_context->gva_to_gpa = paging64_gva_to_gpa; in init_kvm_nested_mmu()
5533 g_context->gva_to_gpa = paging32_gva_to_gpa; in init_kvm_nested_mmu()
5566 vcpu->arch.root_mmu.root_role.word = 0; in kvm_mmu_after_set_cpuid()
5567 vcpu->arch.guest_mmu.root_role.word = 0; in kvm_mmu_after_set_cpuid()
5568 vcpu->arch.nested_mmu.root_role.word = 0; in kvm_mmu_after_set_cpuid()
5569 vcpu->arch.root_mmu.cpu_role.ext.valid = 0; in kvm_mmu_after_set_cpuid()
5570 vcpu->arch.guest_mmu.cpu_role.ext.valid = 0; in kvm_mmu_after_set_cpuid()
5571 vcpu->arch.nested_mmu.cpu_role.ext.valid = 0; in kvm_mmu_after_set_cpuid()
5578 KVM_BUG_ON(kvm_vcpu_has_run(vcpu), vcpu->kvm); in kvm_mmu_after_set_cpuid()
5592 r = mmu_topup_memory_caches(vcpu, !vcpu->arch.mmu->root_role.direct); in kvm_mmu_load()
5598 if (vcpu->arch.mmu->root_role.direct) in kvm_mmu_load()
5623 struct kvm *kvm = vcpu->kvm; in kvm_mmu_unload()
5625 kvm_mmu_free_roots(kvm, &vcpu->arch.root_mmu, KVM_MMU_ROOTS_ALL); in kvm_mmu_unload()
5626 WARN_ON_ONCE(VALID_PAGE(vcpu->arch.root_mmu.root.hpa)); in kvm_mmu_unload()
5627 kvm_mmu_free_roots(kvm, &vcpu->arch.guest_mmu, KVM_MMU_ROOTS_ALL); in kvm_mmu_unload()
5628 WARN_ON_ONCE(VALID_PAGE(vcpu->arch.guest_mmu.root.hpa)); in kvm_mmu_unload()
5650 * is unlikely to zap an in-use PGD. in is_obsolete_root()
5664 if (is_obsolete_root(kvm, mmu->root.hpa)) in __kvm_mmu_free_obsolete_roots()
5668 if (is_obsolete_root(kvm, mmu->prev_roots[i].hpa)) in __kvm_mmu_free_obsolete_roots()
5678 __kvm_mmu_free_obsolete_roots(vcpu->kvm, &vcpu->arch.root_mmu); in kvm_mmu_free_obsolete_roots()
5679 __kvm_mmu_free_obsolete_roots(vcpu->kvm, &vcpu->arch.guest_mmu); in kvm_mmu_free_obsolete_roots()
5694 /* Handle a 32-bit guest writing two halves of a 64-bit gpte */ in mmu_pte_write_fetch_gpte()
5715 * Skip write-flooding detected for the sp whose level is 1, because in detect_write_flooding()
5716 * it can become unsync, then the guest page is not write-protected. in detect_write_flooding()
5718 if (sp->role.level == PG_LEVEL_4K) in detect_write_flooding()
5721 atomic_inc(&sp->write_flooding_count); in detect_write_flooding()
5722 return atomic_read(&sp->write_flooding_count) >= 3; in detect_write_flooding()
5735 pte_size = sp->role.has_4_byte_gpte ? 4 : 8; in detect_write_misaligned()
5741 if (!(offset & (pte_size - 1)) && bytes == 1) in detect_write_misaligned()
5744 misaligned = (offset ^ (offset + bytes - 1)) & ~(pte_size - 1); in detect_write_misaligned()
5757 level = sp->role.level; in get_written_sptes()
5759 if (sp->role.has_4_byte_gpte) { in get_written_sptes()
5760 page_offset <<= 1; /* 32->64 */ in get_written_sptes()
5762 * A 32-bit pde maps 4MB while the shadow pdes map in get_written_sptes()
5773 if (quadrant != sp->role.quadrant) in get_written_sptes()
5777 spte = &sp->spt[page_offset / sizeof(*spte)]; in get_written_sptes()
5793 * write-protected, so we can exit simply. in kvm_mmu_track_write()
5795 if (!READ_ONCE(vcpu->kvm->arch.indirect_shadow_pages)) in kvm_mmu_track_write()
5798 write_lock(&vcpu->kvm->mmu_lock); in kvm_mmu_track_write()
5802 ++vcpu->kvm->stat.mmu_pte_write; in kvm_mmu_track_write()
5804 for_each_gfn_valid_sp_with_gptes(vcpu->kvm, sp, gfn) { in kvm_mmu_track_write()
5807 kvm_mmu_prepare_zap_page(vcpu->kvm, sp, &invalid_list); in kvm_mmu_track_write()
5808 ++vcpu->kvm->stat.mmu_flooded; in kvm_mmu_track_write()
5816 while (npte--) { in kvm_mmu_track_write()
5818 mmu_page_zap_pte(vcpu->kvm, sp, spte, NULL); in kvm_mmu_track_write()
5819 if (gentry && sp->role.level != PG_LEVEL_4K) in kvm_mmu_track_write()
5820 ++vcpu->kvm->stat.mmu_pde_zapped; in kvm_mmu_track_write()
5826 kvm_mmu_remote_flush_or_zap(vcpu->kvm, &invalid_list, flush); in kvm_mmu_track_write()
5827 write_unlock(&vcpu->kvm->mmu_lock); in kvm_mmu_track_write()
5834 bool direct = vcpu->arch.mmu->root_role.direct; in kvm_mmu_page_fault()
5837 * IMPLICIT_ACCESS is a KVM-defined flag used to correctly perform SMAP in kvm_mmu_page_fault()
5840 * with the KVM-defined value. Clear the flag and continue on, i.e. in kvm_mmu_page_fault()
5847 if (WARN_ON_ONCE(!VALID_PAGE(vcpu->arch.mmu->root.hpa))) in kvm_mmu_page_fault()
5861 if (KVM_BUG_ON(r == RET_PF_INVALID, vcpu->kvm)) in kvm_mmu_page_fault()
5862 return -EIO; in kvm_mmu_page_fault()
5877 if (vcpu->arch.mmu->root_role.direct && in kvm_mmu_page_fault()
5879 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(cr2_or_gpa)); in kvm_mmu_page_fault()
5884 * vcpu->arch.mmu.page_fault returned RET_PF_EMULATE, but we can still in kvm_mmu_page_fault()
5886 * re-execute the instruction that caused the page fault. Do not allow in kvm_mmu_page_fault()
5889 * faulting on the non-existent MMIO address. Retrying an instruction in kvm_mmu_page_fault()
5914 if (WARN_ON_ONCE(mmu != vcpu->arch.mmu)) in __kvm_mmu_invalidate_addr()
5920 write_lock(&vcpu->kvm->mmu_lock); in __kvm_mmu_invalidate_addr()
5924 if (sp->unsync) { in __kvm_mmu_invalidate_addr()
5928 mmu_page_zap_pte(vcpu->kvm, sp, iterator.sptep, NULL); in __kvm_mmu_invalidate_addr()
5930 kvm_flush_remote_tlbs_sptep(vcpu->kvm, iterator.sptep); in __kvm_mmu_invalidate_addr()
5933 if (!sp->unsync_children) in __kvm_mmu_invalidate_addr()
5936 write_unlock(&vcpu->kvm->mmu_lock); in __kvm_mmu_invalidate_addr()
5946 /* It's actually a GPA for vcpu->arch.guest_mmu. */ in kvm_mmu_invalidate_addr()
5947 if (mmu != &vcpu->arch.guest_mmu) { in kvm_mmu_invalidate_addr()
5948 /* INVLPG on a non-canonical address is a NOP according to the SDM. */ in kvm_mmu_invalidate_addr()
5955 if (!mmu->sync_spte) in kvm_mmu_invalidate_addr()
5959 __kvm_mmu_invalidate_addr(vcpu, mmu, addr, mmu->root.hpa); in kvm_mmu_invalidate_addr()
5963 __kvm_mmu_invalidate_addr(vcpu, mmu, addr, mmu->prev_roots[i].hpa); in kvm_mmu_invalidate_addr()
5980 kvm_mmu_invalidate_addr(vcpu, vcpu->arch.walk_mmu, gva, KVM_MMU_ROOTS_ALL); in kvm_mmu_invlpg()
5981 ++vcpu->stat.invlpg; in kvm_mmu_invlpg()
5988 struct kvm_mmu *mmu = vcpu->arch.mmu; in kvm_mmu_invpcid_gva()
5996 if (VALID_PAGE(mmu->prev_roots[i].hpa) && in kvm_mmu_invpcid_gva()
5997 pcid == kvm_get_pcid(vcpu, mmu->prev_roots[i].pgd)) in kvm_mmu_invpcid_gva()
6003 ++vcpu->stat.invlpg; in kvm_mmu_invpcid_gva()
6052 lockdep_assert_held_write(&kvm->mmu_lock); in __walk_slot_rmaps()
6059 if (need_resched() || rwlock_needbreak(&kvm->mmu_lock)) { in __walk_slot_rmaps()
6062 iterator.gfn - start_gfn + 1); in __walk_slot_rmaps()
6065 cond_resched_rwlock_write(&kvm->mmu_lock); in __walk_slot_rmaps()
6079 slot->base_gfn, slot->base_gfn + slot->npages - 1, in walk_slot_rmaps()
6093 if (!tdp_enabled && mmu->pae_root) in free_mmu_pages()
6094 set_memory_encrypted((unsigned long)mmu->pae_root, 1); in free_mmu_pages()
6095 free_page((unsigned long)mmu->pae_root); in free_mmu_pages()
6096 free_page((unsigned long)mmu->pml4_root); in free_mmu_pages()
6097 free_page((unsigned long)mmu->pml5_root); in free_mmu_pages()
6105 mmu->root.hpa = INVALID_PAGE; in __kvm_mmu_create()
6106 mmu->root.pgd = 0; in __kvm_mmu_create()
6108 mmu->prev_roots[i] = KVM_MMU_ROOT_INFO_INVALID; in __kvm_mmu_create()
6110 /* vcpu->arch.guest_mmu isn't used when !tdp_enabled. */ in __kvm_mmu_create()
6111 if (!tdp_enabled && mmu == &vcpu->arch.guest_mmu) in __kvm_mmu_create()
6116 * while the PDP table is a per-vCPU construct that's allocated at MMU in __kvm_mmu_create()
6117 * creation. When emulating 32-bit mode, cr3 is only 32 bits even on in __kvm_mmu_create()
6121 * table. The main exception, handled here, is SVM's 32-bit NPT. The in __kvm_mmu_create()
6122 * other exception is for shadowing L1's 32-bit or PAE NPT on 64-bit in __kvm_mmu_create()
6123 * KVM; that horror is handled on-demand by mmu_alloc_special_roots(). in __kvm_mmu_create()
6130 return -ENOMEM; in __kvm_mmu_create()
6132 mmu->pae_root = page_address(page); in __kvm_mmu_create()
6138 * only necessary when using shadow paging, as 64-bit NPT can get at in __kvm_mmu_create()
6139 * the C-bit even when shadowing 32-bit NPT, and SME isn't supported in __kvm_mmu_create()
6140 * by 32-bit kernels (when KVM itself uses 32-bit NPT). in __kvm_mmu_create()
6143 set_memory_decrypted((unsigned long)mmu->pae_root, 1); in __kvm_mmu_create()
6148 mmu->pae_root[i] = INVALID_PAE_ROOT; in __kvm_mmu_create()
6157 vcpu->arch.mmu_pte_list_desc_cache.kmem_cache = pte_list_desc_cache; in kvm_mmu_create()
6158 vcpu->arch.mmu_pte_list_desc_cache.gfp_zero = __GFP_ZERO; in kvm_mmu_create()
6160 vcpu->arch.mmu_page_header_cache.kmem_cache = mmu_page_header_cache; in kvm_mmu_create()
6161 vcpu->arch.mmu_page_header_cache.gfp_zero = __GFP_ZERO; in kvm_mmu_create()
6163 vcpu->arch.mmu_shadow_page_cache.gfp_zero = __GFP_ZERO; in kvm_mmu_create()
6165 vcpu->arch.mmu = &vcpu->arch.root_mmu; in kvm_mmu_create()
6166 vcpu->arch.walk_mmu = &vcpu->arch.root_mmu; in kvm_mmu_create()
6168 ret = __kvm_mmu_create(vcpu, &vcpu->arch.guest_mmu); in kvm_mmu_create()
6172 ret = __kvm_mmu_create(vcpu, &vcpu->arch.root_mmu); in kvm_mmu_create()
6178 free_mmu_pages(&vcpu->arch.guest_mmu); in kvm_mmu_create()
6191 &kvm->arch.active_mmu_pages, link) { in kvm_zap_obsolete_pages()
6204 if (WARN_ON_ONCE(sp->role.invalid)) in kvm_zap_obsolete_pages()
6214 cond_resched_rwlock_write(&kvm->mmu_lock)) { in kvm_zap_obsolete_pages()
6220 &kvm->arch.zapped_obsolete_pages, &nr_zapped); in kvm_zap_obsolete_pages()
6236 kvm_mmu_commit_zap_page(kvm, &kvm->arch.zapped_obsolete_pages); in kvm_zap_obsolete_pages()
6240 * Fast invalidate all shadow pages and use lock-break technique
6245 * not use any resource of the being-deleted slot or all slots
6250 lockdep_assert_held(&kvm->slots_lock); in kvm_mmu_zap_all_fast()
6252 write_lock(&kvm->mmu_lock); in kvm_mmu_zap_all_fast()
6262 kvm->arch.mmu_valid_gen = kvm->arch.mmu_valid_gen ? 0 : 1; in kvm_mmu_zap_all_fast()
6265 * In order to ensure all vCPUs drop their soon-to-be invalid roots, in kvm_mmu_zap_all_fast()
6285 write_unlock(&kvm->mmu_lock); in kvm_mmu_zap_all_fast()
6293 * lead to use-after-free. in kvm_mmu_zap_all_fast()
6301 return unlikely(!list_empty_careful(&kvm->arch.zapped_obsolete_pages)); in kvm_has_zapped_obsolete_pages()
6306 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages); in kvm_mmu_init_vm()
6307 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages); in kvm_mmu_init_vm()
6308 INIT_LIST_HEAD(&kvm->arch.possible_nx_huge_pages); in kvm_mmu_init_vm()
6309 spin_lock_init(&kvm->arch.mmu_unsync_pages_lock); in kvm_mmu_init_vm()
6314 kvm->arch.split_page_header_cache.kmem_cache = mmu_page_header_cache; in kvm_mmu_init_vm()
6315 kvm->arch.split_page_header_cache.gfp_zero = __GFP_ZERO; in kvm_mmu_init_vm()
6317 kvm->arch.split_shadow_page_cache.gfp_zero = __GFP_ZERO; in kvm_mmu_init_vm()
6319 kvm->arch.split_desc_cache.kmem_cache = pte_list_desc_cache; in kvm_mmu_init_vm()
6320 kvm->arch.split_desc_cache.gfp_zero = __GFP_ZERO; in kvm_mmu_init_vm()
6325 kvm_mmu_free_memory_cache(&kvm->arch.split_desc_cache); in mmu_free_vm_memory_caches()
6326 kvm_mmu_free_memory_cache(&kvm->arch.split_page_header_cache); in mmu_free_vm_memory_caches()
6327 kvm_mmu_free_memory_cache(&kvm->arch.split_shadow_page_cache); in mmu_free_vm_memory_caches()
6355 start = max(gfn_start, memslot->base_gfn); in kvm_rmap_zap_gfn_range()
6356 end = min(gfn_end, memslot->base_gfn + memslot->npages); in kvm_rmap_zap_gfn_range()
6362 start, end - 1, true, flush); in kvm_rmap_zap_gfn_range()
6380 write_lock(&kvm->mmu_lock); in kvm_zap_gfn_range()
6392 kvm_flush_remote_tlbs_range(kvm, gfn_start, gfn_end - gfn_start); in kvm_zap_gfn_range()
6396 write_unlock(&kvm->mmu_lock); in kvm_zap_gfn_range()
6411 write_lock(&kvm->mmu_lock); in kvm_mmu_slot_remove_write_access()
6414 write_unlock(&kvm->mmu_lock); in kvm_mmu_slot_remove_write_access()
6418 read_lock(&kvm->mmu_lock); in kvm_mmu_slot_remove_write_access()
6420 read_unlock(&kvm->mmu_lock); in kvm_mmu_slot_remove_write_access()
6431 if (need_resched() || rwlock_needbreak(&kvm->mmu_lock)) in need_topup_split_caches_or_resched()
6439 return need_topup(&kvm->arch.split_desc_cache, SPLIT_DESC_CACHE_MIN_NR_OBJECTS) || in need_topup_split_caches_or_resched()
6440 need_topup(&kvm->arch.split_page_header_cache, 1) || in need_topup_split_caches_or_resched()
6441 need_topup(&kvm->arch.split_shadow_page_cache, 1); in need_topup_split_caches_or_resched()
6453 * but aliasing rarely occurs post-boot or for many gfns. If there is in topup_split_caches()
6454 * only one rmap entry, rmap->val points directly at that one entry and in topup_split_caches()
6463 lockdep_assert_held(&kvm->slots_lock); in topup_split_caches()
6465 r = __kvm_mmu_topup_memory_cache(&kvm->arch.split_desc_cache, capacity, in topup_split_caches()
6470 r = kvm_mmu_topup_memory_cache(&kvm->arch.split_page_header_cache, 1); in topup_split_caches()
6474 return kvm_mmu_topup_memory_cache(&kvm->arch.split_shadow_page_cache, 1); in topup_split_caches()
6497 caches.page_header_cache = &kvm->arch.split_page_header_cache; in shadow_mmu_get_sp_for_split()
6498 caches.shadow_page_cache = &kvm->arch.split_shadow_page_cache; in shadow_mmu_get_sp_for_split()
6509 struct kvm_mmu_memory_cache *cache = &kvm->arch.split_desc_cache; in shadow_mmu_split_huge_page()
6520 sptep = &sp->spt[index]; in shadow_mmu_split_huge_page()
6527 * gfn-to-pfn translation since the SP is direct, so no need to in shadow_mmu_split_huge_page()
6538 flush |= !is_last_spte(*sptep, sp->role.level); in shadow_mmu_split_huge_page()
6542 spte = make_huge_page_split_spte(kvm, huge_spte, sp->role, index); in shadow_mmu_split_huge_page()
6544 __rmap_add(kvm, cache, slot, sptep, gfn, sp->role.access); in shadow_mmu_split_huge_page()
6561 level = huge_sp->role.level; in shadow_mmu_try_split_huge_page()
6565 r = -ENOSPC; in shadow_mmu_try_split_huge_page()
6570 write_unlock(&kvm->mmu_lock); in shadow_mmu_try_split_huge_page()
6573 * If the topup succeeds, return -EAGAIN to indicate that the in shadow_mmu_try_split_huge_page()
6577 r = topup_split_caches(kvm) ?: -EAGAIN; in shadow_mmu_try_split_huge_page()
6578 write_lock(&kvm->mmu_lock); in shadow_mmu_try_split_huge_page()
6603 if (WARN_ON_ONCE(!sp->role.guest_mode)) in shadow_mmu_try_split_huge_pages()
6606 /* The rmaps should never contain non-leaf SPTEs. */ in shadow_mmu_try_split_huge_pages()
6611 if (WARN_ON_ONCE(sp->unsync)) in shadow_mmu_try_split_huge_pages()
6615 if (sp->role.invalid) in shadow_mmu_try_split_huge_pages()
6625 if (!r || r == -EAGAIN) in shadow_mmu_try_split_huge_pages()
6628 /* The split failed and shouldn't be retried (e.g. -ENOMEM). */ in shadow_mmu_try_split_huge_pages()
6648 for (level = KVM_MAX_HUGEPAGE_LEVEL; level > target_level; level--) in kvm_shadow_mmu_try_split_huge_pages()
6650 level, level, start, end - 1, true, false); in kvm_shadow_mmu_try_split_huge_pages()
6653 /* Must be called with the mmu_lock held in write-mode. */
6677 u64 start = memslot->base_gfn; in kvm_mmu_slot_try_split_huge_pages()
6678 u64 end = start + memslot->npages; in kvm_mmu_slot_try_split_huge_pages()
6684 write_lock(&kvm->mmu_lock); in kvm_mmu_slot_try_split_huge_pages()
6686 write_unlock(&kvm->mmu_lock); in kvm_mmu_slot_try_split_huge_pages()
6689 read_lock(&kvm->mmu_lock); in kvm_mmu_slot_try_split_huge_pages()
6691 read_unlock(&kvm->mmu_lock); in kvm_mmu_slot_try_split_huge_pages()
6695 * write-protecting and/or clearing dirty on the newly split SPTEs to in kvm_mmu_slot_try_split_huge_pages()
6724 if (sp->role.direct && in kvm_mmu_zap_collapsible_spte()
6725 sp->role.level < kvm_mmu_max_mapping_level(kvm, slot, sp->gfn, in kvm_mmu_zap_collapsible_spte()
6745 * Note, use KVM_MAX_HUGEPAGE_LEVEL - 1 since there's no need to zap in kvm_rmap_zap_collapsible_sptes()
6749 PG_LEVEL_4K, KVM_MAX_HUGEPAGE_LEVEL - 1, true)) in kvm_rmap_zap_collapsible_sptes()
6757 write_lock(&kvm->mmu_lock); in kvm_mmu_zap_collapsible_sptes()
6759 write_unlock(&kvm->mmu_lock); in kvm_mmu_zap_collapsible_sptes()
6763 read_lock(&kvm->mmu_lock); in kvm_mmu_zap_collapsible_sptes()
6765 read_unlock(&kvm->mmu_lock); in kvm_mmu_zap_collapsible_sptes()
6773 write_lock(&kvm->mmu_lock); in kvm_mmu_slot_leaf_clear_dirty()
6779 write_unlock(&kvm->mmu_lock); in kvm_mmu_slot_leaf_clear_dirty()
6783 read_lock(&kvm->mmu_lock); in kvm_mmu_slot_leaf_clear_dirty()
6785 read_unlock(&kvm->mmu_lock); in kvm_mmu_slot_leaf_clear_dirty()
6804 write_lock(&kvm->mmu_lock); in kvm_mmu_zap_all()
6806 list_for_each_entry_safe(sp, node, &kvm->arch.active_mmu_pages, link) { in kvm_mmu_zap_all()
6807 if (WARN_ON_ONCE(sp->role.invalid)) in kvm_mmu_zap_all()
6811 if (cond_resched_rwlock_write(&kvm->mmu_lock)) in kvm_mmu_zap_all()
6820 write_unlock(&kvm->mmu_lock); in kvm_mmu_zap_all()
6834 void kvm_mmu_invalidate_mmio_sptes(struct kvm *kvm, u64 gen) in kvm_mmu_invalidate_mmio_sptes() argument
6836 WARN_ON_ONCE(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS); in kvm_mmu_invalidate_mmio_sptes()
6838 gen &= MMIO_SPTE_GEN_MASK; in kvm_mmu_invalidate_mmio_sptes()
6847 gen &= ~((u64)kvm_arch_nr_memslot_as_ids(kvm) - 1); in kvm_mmu_invalidate_mmio_sptes()
6853 if (unlikely(gen == 0)) { in kvm_mmu_invalidate_mmio_sptes()
6863 int nr_to_scan = sc->nr_to_scan; in mmu_shrink_scan()
6873 * Never scan more than sc->nr_to_scan VM instances. in mmu_shrink_scan()
6878 if (!nr_to_scan--) in mmu_shrink_scan()
6881 * n_used_mmu_pages is accessed without holding kvm->mmu_lock in mmu_shrink_scan()
6886 if (!kvm->arch.n_used_mmu_pages && in mmu_shrink_scan()
6890 idx = srcu_read_lock(&kvm->srcu); in mmu_shrink_scan()
6891 write_lock(&kvm->mmu_lock); in mmu_shrink_scan()
6895 &kvm->arch.zapped_obsolete_pages); in mmu_shrink_scan()
6899 freed = kvm_mmu_zap_oldest_mmu_pages(kvm, sc->nr_to_scan); in mmu_shrink_scan()
6902 write_unlock(&kvm->mmu_lock); in mmu_shrink_scan()
6903 srcu_read_unlock(&kvm->srcu, idx); in mmu_shrink_scan()
6907 * per-vm shrinkers cry out in mmu_shrink_scan()
6910 list_move_tail(&kvm->vm_list, &vm_list); in mmu_shrink_scan()
6957 return -EPERM; in set_nx_huge_pages()
6972 return -EBUSY; in set_nx_huge_pages()
6977 return -EINVAL; in set_nx_huge_pages()
6988 mutex_lock(&kvm->slots_lock); in set_nx_huge_pages()
6990 mutex_unlock(&kvm->slots_lock); in set_nx_huge_pages()
6992 wake_up_process(kvm->arch.nx_huge_page_recovery_thread); in set_nx_huge_pages()
7002 * its default value of -1 is technically undefined behavior for a boolean.
7008 if (nx_huge_pages == -1) in kvm_mmu_x86_module_init()
7028 int ret = -ENOMEM; in kvm_mmu_vendor_module_init()
7057 mmu_shrinker = shrinker_alloc(0, "x86-mmu"); in kvm_mmu_vendor_module_init()
7061 mmu_shrinker->count_objects = mmu_shrink_count; in kvm_mmu_vendor_module_init()
7062 mmu_shrinker->scan_objects = mmu_shrink_scan; in kvm_mmu_vendor_module_init()
7063 mmu_shrinker->seeks = DEFAULT_SEEKS * 10; in kvm_mmu_vendor_module_init()
7079 free_mmu_pages(&vcpu->arch.root_mmu); in kvm_mmu_destroy()
7080 free_mmu_pages(&vcpu->arch.guest_mmu); in kvm_mmu_destroy()
7123 return -EPERM; in set_nx_huge_pages_recovery_param()
7140 wake_up_process(kvm->arch.nx_huge_page_recovery_thread); in set_nx_huge_pages_recovery_param()
7150 unsigned long nx_lpage_splits = kvm->stat.nx_lpage_splits; in kvm_recover_nx_huge_pages()
7159 rcu_idx = srcu_read_lock(&kvm->srcu); in kvm_recover_nx_huge_pages()
7160 write_lock(&kvm->mmu_lock); in kvm_recover_nx_huge_pages()
7171 for ( ; to_zap; --to_zap) { in kvm_recover_nx_huge_pages()
7172 if (list_empty(&kvm->arch.possible_nx_huge_pages)) in kvm_recover_nx_huge_pages()
7182 sp = list_first_entry(&kvm->arch.possible_nx_huge_pages, in kvm_recover_nx_huge_pages()
7185 WARN_ON_ONCE(!sp->nx_huge_page_disallowed); in kvm_recover_nx_huge_pages()
7186 WARN_ON_ONCE(!sp->role.direct); in kvm_recover_nx_huge_pages()
7202 * of kvm->nr_memslots_dirty_logging is not a problem: if it is in kvm_recover_nx_huge_pages()
7209 if (atomic_read(&kvm->nr_memslots_dirty_logging)) { in kvm_recover_nx_huge_pages()
7212 slots = kvm_memslots_for_spte_role(kvm, sp->role); in kvm_recover_nx_huge_pages()
7213 slot = __gfn_to_memslot(slots, sp->gfn); in kvm_recover_nx_huge_pages()
7223 WARN_ON_ONCE(sp->nx_huge_page_disallowed); in kvm_recover_nx_huge_pages()
7225 if (need_resched() || rwlock_needbreak(&kvm->mmu_lock)) { in kvm_recover_nx_huge_pages()
7229 cond_resched_rwlock_write(&kvm->mmu_lock); in kvm_recover_nx_huge_pages()
7239 write_unlock(&kvm->mmu_lock); in kvm_recover_nx_huge_pages()
7240 srcu_read_unlock(&kvm->srcu, rcu_idx); in kvm_recover_nx_huge_pages()
7250 return enabled ? start_time + msecs_to_jiffies(period) - get_jiffies_64() in get_nx_huge_page_recovery_timeout()
7287 "kvm-nx-lpage-recovery", in kvm_mmu_post_init_vm()
7288 &kvm->arch.nx_huge_page_recovery_thread); in kvm_mmu_post_init_vm()
7290 kthread_unpark(kvm->arch.nx_huge_page_recovery_thread); in kvm_mmu_post_init_vm()
7297 if (kvm->arch.nx_huge_page_recovery_thread) in kvm_mmu_pre_destroy_vm()
7298 kthread_stop(kvm->arch.nx_huge_page_recovery_thread); in kvm_mmu_pre_destroy_vm()
7325 return lpage_info_slot(gfn, slot, level)->disallow_lpage & KVM_LPAGE_MIXED_FLAG; in hugepage_test_mixed()
7331 lpage_info_slot(gfn, slot, level)->disallow_lpage &= ~KVM_LPAGE_MIXED_FLAG; in hugepage_clear_mixed()
7337 lpage_info_slot(gfn, slot, level)->disallow_lpage |= KVM_LPAGE_MIXED_FLAG; in hugepage_set_mixed()
7349 for (gfn = start; gfn < end; gfn += KVM_PAGES_PER_HPAGE(level - 1)) { in hugepage_has_attrs()
7350 if (hugepage_test_mixed(slot, gfn, level - 1) || in hugepage_has_attrs()
7360 unsigned long attrs = range->arg.attributes; in kvm_arch_post_set_memory_attributes()
7361 struct kvm_memory_slot *slot = range->slot; in kvm_arch_post_set_memory_attributes()
7364 lockdep_assert_held_write(&kvm->mmu_lock); in kvm_arch_post_set_memory_attributes()
7365 lockdep_assert_held(&kvm->slots_lock); in kvm_arch_post_set_memory_attributes()
7382 gfn_t gfn = gfn_round_for_level(range->start, level); in kvm_arch_post_set_memory_attributes()
7385 if (gfn != range->start || gfn + nr_pages > range->end) { in kvm_arch_post_set_memory_attributes()
7391 if (gfn >= slot->base_gfn) { in kvm_arch_post_set_memory_attributes()
7404 for ( ; gfn + nr_pages <= range->end; gfn += nr_pages) in kvm_arch_post_set_memory_attributes()
7412 if (gfn < range->end && in kvm_arch_post_set_memory_attributes()
7413 (gfn + nr_pages) <= (slot->base_gfn + slot->npages)) { in kvm_arch_post_set_memory_attributes()
7437 gfn_t end = gfn_round_for_level(slot->base_gfn + slot->npages, level); in kvm_mmu_init_memslot_memory_attributes()
7438 gfn_t start = gfn_round_for_level(slot->base_gfn, level); in kvm_mmu_init_memslot_memory_attributes()
7442 if (start < slot->base_gfn) in kvm_mmu_init_memslot_memory_attributes()