1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef __KVM_X86_MMU_H 3 #define __KVM_X86_MMU_H 4 5 #include <linux/kvm_host.h> 6 #include "kvm_cache_regs.h" 7 8 #define PT64_PT_BITS 9 9 #define PT64_ENT_PER_PAGE (1 << PT64_PT_BITS) 10 #define PT32_PT_BITS 10 11 #define PT32_ENT_PER_PAGE (1 << PT32_PT_BITS) 12 13 #define PT_WRITABLE_SHIFT 1 14 #define PT_USER_SHIFT 2 15 16 #define PT_PRESENT_MASK (1ULL << 0) 17 #define PT_WRITABLE_MASK (1ULL << PT_WRITABLE_SHIFT) 18 #define PT_USER_MASK (1ULL << PT_USER_SHIFT) 19 #define PT_PWT_MASK (1ULL << 3) 20 #define PT_PCD_MASK (1ULL << 4) 21 #define PT_ACCESSED_SHIFT 5 22 #define PT_ACCESSED_MASK (1ULL << PT_ACCESSED_SHIFT) 23 #define PT_DIRTY_SHIFT 6 24 #define PT_DIRTY_MASK (1ULL << PT_DIRTY_SHIFT) 25 #define PT_PAGE_SIZE_SHIFT 7 26 #define PT_PAGE_SIZE_MASK (1ULL << PT_PAGE_SIZE_SHIFT) 27 #define PT_PAT_MASK (1ULL << 7) 28 #define PT_GLOBAL_MASK (1ULL << 8) 29 #define PT64_NX_SHIFT 63 30 #define PT64_NX_MASK (1ULL << PT64_NX_SHIFT) 31 32 #define PT_PAT_SHIFT 7 33 #define PT_DIR_PAT_SHIFT 12 34 #define PT_DIR_PAT_MASK (1ULL << PT_DIR_PAT_SHIFT) 35 36 #define PT32_DIR_PSE36_SIZE 4 37 #define PT32_DIR_PSE36_SHIFT 13 38 #define PT32_DIR_PSE36_MASK \ 39 (((1ULL << PT32_DIR_PSE36_SIZE) - 1) << PT32_DIR_PSE36_SHIFT) 40 41 #define PT64_ROOT_5LEVEL 5 42 #define PT64_ROOT_4LEVEL 4 43 #define PT32_ROOT_LEVEL 2 44 #define PT32E_ROOT_LEVEL 3 45 46 static inline u64 rsvd_bits(int s, int e) 47 { 48 if (e < s) 49 return 0; 50 51 return ((1ULL << (e - s + 1)) - 1) << s; 52 } 53 54 void kvm_mmu_set_mmio_spte_mask(u64 mmio_mask, u64 mmio_value); 55 56 void 57 reset_shadow_zero_bits_mask(struct kvm_vcpu *vcpu, struct kvm_mmu *context); 58 59 void kvm_init_mmu(struct kvm_vcpu *vcpu, bool reset_roots); 60 void kvm_init_shadow_mmu(struct kvm_vcpu *vcpu); 61 void kvm_init_shadow_ept_mmu(struct kvm_vcpu *vcpu, bool execonly, 62 bool accessed_dirty, gpa_t new_eptp); 63 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu); 64 int kvm_handle_page_fault(struct kvm_vcpu *vcpu, u64 error_code, 65 u64 fault_address, char *insn, int insn_len); 66 67 static inline unsigned long kvm_mmu_available_pages(struct kvm *kvm) 68 { 69 if (kvm->arch.n_max_mmu_pages > kvm->arch.n_used_mmu_pages) 70 return kvm->arch.n_max_mmu_pages - 71 kvm->arch.n_used_mmu_pages; 72 73 return 0; 74 } 75 76 static inline int kvm_mmu_reload(struct kvm_vcpu *vcpu) 77 { 78 if (likely(vcpu->arch.mmu->root_hpa != INVALID_PAGE)) 79 return 0; 80 81 return kvm_mmu_load(vcpu); 82 } 83 84 static inline unsigned long kvm_get_pcid(struct kvm_vcpu *vcpu, gpa_t cr3) 85 { 86 BUILD_BUG_ON((X86_CR3_PCID_MASK & PAGE_MASK) != 0); 87 88 return kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE) 89 ? cr3 & X86_CR3_PCID_MASK 90 : 0; 91 } 92 93 static inline unsigned long kvm_get_active_pcid(struct kvm_vcpu *vcpu) 94 { 95 return kvm_get_pcid(vcpu, kvm_read_cr3(vcpu)); 96 } 97 98 static inline void kvm_mmu_load_cr3(struct kvm_vcpu *vcpu) 99 { 100 if (VALID_PAGE(vcpu->arch.mmu->root_hpa)) 101 vcpu->arch.mmu->set_cr3(vcpu, vcpu->arch.mmu->root_hpa | 102 kvm_get_active_pcid(vcpu)); 103 } 104 105 /* 106 * Currently, we have two sorts of write-protection, a) the first one 107 * write-protects guest page to sync the guest modification, b) another one is 108 * used to sync dirty bitmap when we do KVM_GET_DIRTY_LOG. The differences 109 * between these two sorts are: 110 * 1) the first case clears SPTE_MMU_WRITEABLE bit. 111 * 2) the first case requires flushing tlb immediately avoiding corrupting 112 * shadow page table between all vcpus so it should be in the protection of 113 * mmu-lock. And the another case does not need to flush tlb until returning 114 * the dirty bitmap to userspace since it only write-protects the page 115 * logged in the bitmap, that means the page in the dirty bitmap is not 116 * missed, so it can flush tlb out of mmu-lock. 117 * 118 * So, there is the problem: the first case can meet the corrupted tlb caused 119 * by another case which write-protects pages but without flush tlb 120 * immediately. In order to making the first case be aware this problem we let 121 * it flush tlb if we try to write-protect a spte whose SPTE_MMU_WRITEABLE bit 122 * is set, it works since another case never touches SPTE_MMU_WRITEABLE bit. 123 * 124 * Anyway, whenever a spte is updated (only permission and status bits are 125 * changed) we need to check whether the spte with SPTE_MMU_WRITEABLE becomes 126 * readonly, if that happens, we need to flush tlb. Fortunately, 127 * mmu_spte_update() has already handled it perfectly. 128 * 129 * The rules to use SPTE_MMU_WRITEABLE and PT_WRITABLE_MASK: 130 * - if we want to see if it has writable tlb entry or if the spte can be 131 * writable on the mmu mapping, check SPTE_MMU_WRITEABLE, this is the most 132 * case, otherwise 133 * - if we fix page fault on the spte or do write-protection by dirty logging, 134 * check PT_WRITABLE_MASK. 135 * 136 * TODO: introduce APIs to split these two cases. 137 */ 138 static inline int is_writable_pte(unsigned long pte) 139 { 140 return pte & PT_WRITABLE_MASK; 141 } 142 143 static inline bool is_write_protection(struct kvm_vcpu *vcpu) 144 { 145 return kvm_read_cr0_bits(vcpu, X86_CR0_WP); 146 } 147 148 /* 149 * Check if a given access (described through the I/D, W/R and U/S bits of a 150 * page fault error code pfec) causes a permission fault with the given PTE 151 * access rights (in ACC_* format). 152 * 153 * Return zero if the access does not fault; return the page fault error code 154 * if the access faults. 155 */ 156 static inline u8 permission_fault(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, 157 unsigned pte_access, unsigned pte_pkey, 158 unsigned pfec) 159 { 160 int cpl = kvm_x86_ops->get_cpl(vcpu); 161 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu); 162 163 /* 164 * If CPL < 3, SMAP prevention are disabled if EFLAGS.AC = 1. 165 * 166 * If CPL = 3, SMAP applies to all supervisor-mode data accesses 167 * (these are implicit supervisor accesses) regardless of the value 168 * of EFLAGS.AC. 169 * 170 * This computes (cpl < 3) && (rflags & X86_EFLAGS_AC), leaving 171 * the result in X86_EFLAGS_AC. We then insert it in place of 172 * the PFERR_RSVD_MASK bit; this bit will always be zero in pfec, 173 * but it will be one in index if SMAP checks are being overridden. 174 * It is important to keep this branchless. 175 */ 176 unsigned long smap = (cpl - 3) & (rflags & X86_EFLAGS_AC); 177 int index = (pfec >> 1) + 178 (smap >> (X86_EFLAGS_AC_BIT - PFERR_RSVD_BIT + 1)); 179 bool fault = (mmu->permissions[index] >> pte_access) & 1; 180 u32 errcode = PFERR_PRESENT_MASK; 181 182 WARN_ON(pfec & (PFERR_PK_MASK | PFERR_RSVD_MASK)); 183 if (unlikely(mmu->pkru_mask)) { 184 u32 pkru_bits, offset; 185 186 /* 187 * PKRU defines 32 bits, there are 16 domains and 2 188 * attribute bits per domain in pkru. pte_pkey is the 189 * index of the protection domain, so pte_pkey * 2 is 190 * is the index of the first bit for the domain. 191 */ 192 pkru_bits = (vcpu->arch.pkru >> (pte_pkey * 2)) & 3; 193 194 /* clear present bit, replace PFEC.RSVD with ACC_USER_MASK. */ 195 offset = (pfec & ~1) + 196 ((pte_access & PT_USER_MASK) << (PFERR_RSVD_BIT - PT_USER_SHIFT)); 197 198 pkru_bits &= mmu->pkru_mask >> offset; 199 errcode |= -pkru_bits & PFERR_PK_MASK; 200 fault |= (pkru_bits != 0); 201 } 202 203 return -(u32)fault & errcode; 204 } 205 206 void kvm_zap_gfn_range(struct kvm *kvm, gfn_t gfn_start, gfn_t gfn_end); 207 208 void kvm_mmu_gfn_disallow_lpage(struct kvm_memory_slot *slot, gfn_t gfn); 209 void kvm_mmu_gfn_allow_lpage(struct kvm_memory_slot *slot, gfn_t gfn); 210 bool kvm_mmu_slot_gfn_write_protect(struct kvm *kvm, 211 struct kvm_memory_slot *slot, u64 gfn); 212 int kvm_arch_write_log_dirty(struct kvm_vcpu *vcpu); 213 #endif 214