1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef __KVM_X86_MMU_INTERNAL_H 3 #define __KVM_X86_MMU_INTERNAL_H 4 5 #include <linux/types.h> 6 #include <linux/kvm_host.h> 7 #include <asm/kvm_host.h> 8 9 #undef MMU_DEBUG 10 11 #ifdef MMU_DEBUG 12 extern bool dbg; 13 14 #define pgprintk(x...) do { if (dbg) printk(x); } while (0) 15 #define rmap_printk(fmt, args...) do { if (dbg) printk("%s: " fmt, __func__, ## args); } while (0) 16 #define MMU_WARN_ON(x) WARN_ON(x) 17 #else 18 #define pgprintk(x...) do { } while (0) 19 #define rmap_printk(x...) do { } while (0) 20 #define MMU_WARN_ON(x) do { } while (0) 21 #endif 22 23 /* 24 * Unlike regular MMU roots, PAE "roots", a.k.a. PDPTEs/PDPTRs, have a PRESENT 25 * bit, and thus are guaranteed to be non-zero when valid. And, when a guest 26 * PDPTR is !PRESENT, its corresponding PAE root cannot be set to INVALID_PAGE, 27 * as the CPU would treat that as PRESENT PDPTR with reserved bits set. Use 28 * '0' instead of INVALID_PAGE to indicate an invalid PAE root. 29 */ 30 #define INVALID_PAE_ROOT 0 31 #define IS_VALID_PAE_ROOT(x) (!!(x)) 32 33 typedef u64 __rcu *tdp_ptep_t; 34 35 struct kvm_mmu_page { 36 /* 37 * Note, "link" through "spt" fit in a single 64 byte cache line on 38 * 64-bit kernels, keep it that way unless there's a reason not to. 39 */ 40 struct list_head link; 41 struct hlist_node hash_link; 42 43 bool tdp_mmu_page; 44 bool unsync; 45 u8 mmu_valid_gen; 46 bool lpage_disallowed; /* Can't be replaced by an equiv large page */ 47 48 /* 49 * The following two entries are used to key the shadow page in the 50 * hash table. 51 */ 52 union kvm_mmu_page_role role; 53 gfn_t gfn; 54 55 u64 *spt; 56 /* hold the gfn of each spte inside spt */ 57 gfn_t *gfns; 58 /* Currently serving as active root */ 59 union { 60 int root_count; 61 refcount_t tdp_mmu_root_count; 62 }; 63 unsigned int unsync_children; 64 union { 65 struct kvm_rmap_head parent_ptes; /* rmap pointers to parent sptes */ 66 tdp_ptep_t ptep; 67 }; 68 union { 69 DECLARE_BITMAP(unsync_child_bitmap, 512); 70 struct { 71 struct work_struct tdp_mmu_async_work; 72 void *tdp_mmu_async_data; 73 }; 74 }; 75 76 struct list_head lpage_disallowed_link; 77 #ifdef CONFIG_X86_32 78 /* 79 * Used out of the mmu-lock to avoid reading spte values while an 80 * update is in progress; see the comments in __get_spte_lockless(). 81 */ 82 int clear_spte_count; 83 #endif 84 85 /* Number of writes since the last time traversal visited this page. */ 86 atomic_t write_flooding_count; 87 88 #ifdef CONFIG_X86_64 89 /* Used for freeing the page asynchronously if it is a TDP MMU page. */ 90 struct rcu_head rcu_head; 91 #endif 92 }; 93 94 extern struct kmem_cache *mmu_page_header_cache; 95 96 static inline struct kvm_mmu_page *to_shadow_page(hpa_t shadow_page) 97 { 98 struct page *page = pfn_to_page(shadow_page >> PAGE_SHIFT); 99 100 return (struct kvm_mmu_page *)page_private(page); 101 } 102 103 static inline struct kvm_mmu_page *sptep_to_sp(u64 *sptep) 104 { 105 return to_shadow_page(__pa(sptep)); 106 } 107 108 static inline int kvm_mmu_role_as_id(union kvm_mmu_page_role role) 109 { 110 return role.smm ? 1 : 0; 111 } 112 113 static inline int kvm_mmu_page_as_id(struct kvm_mmu_page *sp) 114 { 115 return kvm_mmu_role_as_id(sp->role); 116 } 117 118 static inline bool kvm_mmu_page_ad_need_write_protect(struct kvm_mmu_page *sp) 119 { 120 /* 121 * When using the EPT page-modification log, the GPAs in the CPU dirty 122 * log would come from L2 rather than L1. Therefore, we need to rely 123 * on write protection to record dirty pages, which bypasses PML, since 124 * writes now result in a vmexit. Note, the check on CPU dirty logging 125 * being enabled is mandatory as the bits used to denote WP-only SPTEs 126 * are reserved for PAE paging (32-bit KVM). 127 */ 128 return kvm_x86_ops.cpu_dirty_log_size && sp->role.guest_mode; 129 } 130 131 int mmu_try_to_unsync_pages(struct kvm *kvm, const struct kvm_memory_slot *slot, 132 gfn_t gfn, bool can_unsync, bool prefetch); 133 134 void kvm_mmu_gfn_disallow_lpage(const struct kvm_memory_slot *slot, gfn_t gfn); 135 void kvm_mmu_gfn_allow_lpage(const struct kvm_memory_slot *slot, gfn_t gfn); 136 bool kvm_mmu_slot_gfn_write_protect(struct kvm *kvm, 137 struct kvm_memory_slot *slot, u64 gfn, 138 int min_level); 139 void kvm_flush_remote_tlbs_with_address(struct kvm *kvm, 140 u64 start_gfn, u64 pages); 141 unsigned int pte_list_count(struct kvm_rmap_head *rmap_head); 142 143 /* 144 * Return values of handle_mmio_page_fault, mmu.page_fault, and fast_page_fault(). 145 * 146 * RET_PF_RETRY: let CPU fault again on the address. 147 * RET_PF_EMULATE: mmio page fault, emulate the instruction directly. 148 * RET_PF_INVALID: the spte is invalid, let the real page fault path update it. 149 * RET_PF_FIXED: The faulting entry has been fixed. 150 * RET_PF_SPURIOUS: The faulting entry was already fixed, e.g. by another vCPU. 151 * 152 * Any names added to this enum should be exported to userspace for use in 153 * tracepoints via TRACE_DEFINE_ENUM() in mmutrace.h 154 */ 155 enum { 156 RET_PF_RETRY = 0, 157 RET_PF_EMULATE, 158 RET_PF_INVALID, 159 RET_PF_FIXED, 160 RET_PF_SPURIOUS, 161 }; 162 163 int kvm_mmu_max_mapping_level(struct kvm *kvm, 164 const struct kvm_memory_slot *slot, gfn_t gfn, 165 kvm_pfn_t pfn, int max_level); 166 void kvm_mmu_hugepage_adjust(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault); 167 void disallowed_hugepage_adjust(struct kvm_page_fault *fault, u64 spte, int cur_level); 168 169 void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc); 170 171 void account_huge_nx_page(struct kvm *kvm, struct kvm_mmu_page *sp); 172 void unaccount_huge_nx_page(struct kvm *kvm, struct kvm_mmu_page *sp); 173 174 #endif /* __KVM_X86_MMU_INTERNAL_H */ 175