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#include "cpuid.h" 8 9extern bool __read_mostly enable_mmio_caching; 10 11#define PT_WRITABLE_SHIFT 1 12#define PT_USER_SHIFT 2 13 14#define PT_PRESENT_MASK (1ULL << 0) 15#define PT_WRITABLE_MASK (1ULL << PT_WRITABLE_SHIFT) 16#define PT_USER_MASK (1ULL << PT_USER_SHIFT) 17#define PT_PWT_MASK (1ULL << 3) 18#define PT_PCD_MASK (1ULL << 4) 19#define PT_ACCESSED_SHIFT 5 20#define PT_ACCESSED_MASK (1ULL << PT_ACCESSED_SHIFT) 21#define PT_DIRTY_SHIFT 6 22#define PT_DIRTY_MASK (1ULL << PT_DIRTY_SHIFT) 23#define PT_PAGE_SIZE_SHIFT 7 24#define PT_PAGE_SIZE_MASK (1ULL << PT_PAGE_SIZE_SHIFT) 25#define PT_PAT_MASK (1ULL << 7) 26#define PT_GLOBAL_MASK (1ULL << 8) 27#define PT64_NX_SHIFT 63 28#define PT64_NX_MASK (1ULL << PT64_NX_SHIFT) 29 30#define PT_PAT_SHIFT 7 31#define PT_DIR_PAT_SHIFT 12 32#define PT_DIR_PAT_MASK (1ULL << PT_DIR_PAT_SHIFT) 33 34#define PT64_ROOT_5LEVEL 5 35#define PT64_ROOT_4LEVEL 4 36#define PT32_ROOT_LEVEL 2 37#define PT32E_ROOT_LEVEL 3 38 39#define KVM_MMU_CR4_ROLE_BITS (X86_CR4_PSE | X86_CR4_PAE | X86_CR4_LA57 | \ 40 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE) 41 42#define KVM_MMU_CR0_ROLE_BITS (X86_CR0_PG | X86_CR0_WP) 43#define KVM_MMU_EFER_ROLE_BITS (EFER_LME | EFER_NX) 44 45static __always_inline u64 rsvd_bits(int s, int e) 46{ 47 BUILD_BUG_ON(__builtin_constant_p(e) && __builtin_constant_p(s) && e < s); 48 49 if (__builtin_constant_p(e)) 50 BUILD_BUG_ON(e > 63); 51 else 52 e &= 63; 53 54 if (e < s) 55 return 0; 56 57 return ((2ULL << (e - s)) - 1) << s; 58} 59 60/* 61 * The number of non-reserved physical address bits irrespective of features 62 * that repurpose legal bits, e.g. MKTME. 63 */ 64extern u8 __read_mostly shadow_phys_bits; 65 66static inline gfn_t kvm_mmu_max_gfn(void) 67{ 68 /* 69 * Note that this uses the host MAXPHYADDR, not the guest's. 70 * EPT/NPT cannot support GPAs that would exceed host.MAXPHYADDR; 71 * assuming KVM is running on bare metal, guest accesses beyond 72 * host.MAXPHYADDR will hit a #PF(RSVD) and never cause a vmexit 73 * (either EPT Violation/Misconfig or #NPF), and so KVM will never 74 * install a SPTE for such addresses. If KVM is running as a VM 75 * itself, on the other hand, it might see a MAXPHYADDR that is less 76 * than hardware's real MAXPHYADDR. Using the host MAXPHYADDR 77 * disallows such SPTEs entirely and simplifies the TDP MMU. 78 */ 79 int max_gpa_bits = likely(tdp_enabled) ? shadow_phys_bits : 52; 80 81 return (1ULL << (max_gpa_bits - PAGE_SHIFT)) - 1; 82} 83 84static inline u8 kvm_get_shadow_phys_bits(void) 85{ 86 /* 87 * boot_cpu_data.x86_phys_bits is reduced when MKTME or SME are detected 88 * in CPU detection code, but the processor treats those reduced bits as 89 * 'keyID' thus they are not reserved bits. Therefore KVM needs to look at 90 * the physical address bits reported by CPUID. 91 */ 92 if (likely(boot_cpu_data.extended_cpuid_level >= 0x80000008)) 93 return cpuid_eax(0x80000008) & 0xff; 94 95 /* 96 * Quite weird to have VMX or SVM but not MAXPHYADDR; probably a VM with 97 * custom CPUID. Proceed with whatever the kernel found since these features 98 * aren't virtualizable (SME/SEV also require CPUIDs higher than 0x80000008). 99 */ 100 return boot_cpu_data.x86_phys_bits; 101} 102 103void kvm_mmu_set_mmio_spte_mask(u64 mmio_value, u64 mmio_mask, u64 access_mask); 104void kvm_mmu_set_me_spte_mask(u64 me_value, u64 me_mask); 105void kvm_mmu_set_ept_masks(bool has_ad_bits, bool has_exec_only); 106 107void kvm_init_mmu(struct kvm_vcpu *vcpu); 108void kvm_init_shadow_npt_mmu(struct kvm_vcpu *vcpu, unsigned long cr0, 109 unsigned long cr4, u64 efer, gpa_t nested_cr3); 110void kvm_init_shadow_ept_mmu(struct kvm_vcpu *vcpu, bool execonly, 111 int huge_page_level, bool accessed_dirty, 112 gpa_t new_eptp); 113bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu); 114int kvm_handle_page_fault(struct kvm_vcpu *vcpu, u64 error_code, 115 u64 fault_address, char *insn, int insn_len); 116void __kvm_mmu_refresh_passthrough_bits(struct kvm_vcpu *vcpu, 117 struct kvm_mmu *mmu); 118 119int kvm_mmu_load(struct kvm_vcpu *vcpu); 120void kvm_mmu_unload(struct kvm_vcpu *vcpu); 121void kvm_mmu_free_obsolete_roots(struct kvm_vcpu *vcpu); 122void kvm_mmu_sync_roots(struct kvm_vcpu *vcpu); 123void kvm_mmu_sync_prev_roots(struct kvm_vcpu *vcpu); 124void kvm_mmu_track_write(struct kvm_vcpu *vcpu, gpa_t gpa, const u8 *new, 125 int bytes); 126 127static inline int kvm_mmu_reload(struct kvm_vcpu *vcpu) 128{ 129 if (likely(vcpu->arch.mmu->root.hpa != INVALID_PAGE)) 130 return 0; 131 132 return kvm_mmu_load(vcpu); 133} 134 135static inline unsigned long kvm_get_pcid(struct kvm_vcpu *vcpu, gpa_t cr3) 136{ 137 BUILD_BUG_ON((X86_CR3_PCID_MASK & PAGE_MASK) != 0); 138 139 return kvm_is_cr4_bit_set(vcpu, X86_CR4_PCIDE) 140 ? cr3 & X86_CR3_PCID_MASK 141 : 0; 142} 143 144static inline unsigned long kvm_get_active_pcid(struct kvm_vcpu *vcpu) 145{ 146 return kvm_get_pcid(vcpu, kvm_read_cr3(vcpu)); 147} 148 149static inline void kvm_mmu_load_pgd(struct kvm_vcpu *vcpu) 150{ 151 u64 root_hpa = vcpu->arch.mmu->root.hpa; 152 153 if (!VALID_PAGE(root_hpa)) 154 return; 155 156 static_call(kvm_x86_load_mmu_pgd)(vcpu, root_hpa, 157 vcpu->arch.mmu->root_role.level); 158} 159 160static inline void kvm_mmu_refresh_passthrough_bits(struct kvm_vcpu *vcpu, 161 struct kvm_mmu *mmu) 162{ 163 /* 164 * When EPT is enabled, KVM may passthrough CR0.WP to the guest, i.e. 165 * @mmu's snapshot of CR0.WP and thus all related paging metadata may 166 * be stale. Refresh CR0.WP and the metadata on-demand when checking 167 * for permission faults. Exempt nested MMUs, i.e. MMUs for shadowing 168 * nEPT and nNPT, as CR0.WP is ignored in both cases. Note, KVM does 169 * need to refresh nested_mmu, a.k.a. the walker used to translate L2 170 * GVAs to GPAs, as that "MMU" needs to honor L2's CR0.WP. 171 */ 172 if (!tdp_enabled || mmu == &vcpu->arch.guest_mmu) 173 return; 174 175 __kvm_mmu_refresh_passthrough_bits(vcpu, mmu); 176} 177 178/* 179 * Check if a given access (described through the I/D, W/R and U/S bits of a 180 * page fault error code pfec) causes a permission fault with the given PTE 181 * access rights (in ACC_* format). 182 * 183 * Return zero if the access does not fault; return the page fault error code 184 * if the access faults. 185 */ 186static inline u8 permission_fault(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, 187 unsigned pte_access, unsigned pte_pkey, 188 u64 access) 189{ 190 /* strip nested paging fault error codes */ 191 unsigned int pfec = access; 192 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu); 193 194 /* 195 * For explicit supervisor accesses, SMAP is disabled if EFLAGS.AC = 1. 196 * For implicit supervisor accesses, SMAP cannot be overridden. 197 * 198 * SMAP works on supervisor accesses only, and not_smap can 199 * be set or not set when user access with neither has any bearing 200 * on the result. 201 * 202 * We put the SMAP checking bit in place of the PFERR_RSVD_MASK bit; 203 * this bit will always be zero in pfec, but it will be one in index 204 * if SMAP checks are being disabled. 205 */ 206 u64 implicit_access = access & PFERR_IMPLICIT_ACCESS; 207 bool not_smap = ((rflags & X86_EFLAGS_AC) | implicit_access) == X86_EFLAGS_AC; 208 int index = (pfec + (not_smap << PFERR_RSVD_BIT)) >> 1; 209 u32 errcode = PFERR_PRESENT_MASK; 210 bool fault; 211 212 kvm_mmu_refresh_passthrough_bits(vcpu, mmu); 213 214 fault = (mmu->permissions[index] >> pte_access) & 1; 215 216 WARN_ON(pfec & (PFERR_PK_MASK | PFERR_RSVD_MASK)); 217 if (unlikely(mmu->pkru_mask)) { 218 u32 pkru_bits, offset; 219 220 /* 221 * PKRU defines 32 bits, there are 16 domains and 2 222 * attribute bits per domain in pkru. pte_pkey is the 223 * index of the protection domain, so pte_pkey * 2 is 224 * is the index of the first bit for the domain. 225 */ 226 pkru_bits = (vcpu->arch.pkru >> (pte_pkey * 2)) & 3; 227 228 /* clear present bit, replace PFEC.RSVD with ACC_USER_MASK. */ 229 offset = (pfec & ~1) + 230 ((pte_access & PT_USER_MASK) << (PFERR_RSVD_BIT - PT_USER_SHIFT)); 231 232 pkru_bits &= mmu->pkru_mask >> offset; 233 errcode |= -pkru_bits & PFERR_PK_MASK; 234 fault |= (pkru_bits != 0); 235 } 236 237 return -(u32)fault & errcode; 238} 239 240void kvm_zap_gfn_range(struct kvm *kvm, gfn_t gfn_start, gfn_t gfn_end); 241 242int kvm_arch_write_log_dirty(struct kvm_vcpu *vcpu); 243 244int kvm_mmu_post_init_vm(struct kvm *kvm); 245void kvm_mmu_pre_destroy_vm(struct kvm *kvm); 246 247static inline bool kvm_shadow_root_allocated(struct kvm *kvm) 248{ 249 /* 250 * Read shadow_root_allocated before related pointers. Hence, threads 251 * reading shadow_root_allocated in any lock context are guaranteed to 252 * see the pointers. Pairs with smp_store_release in 253 * mmu_first_shadow_root_alloc. 254 */ 255 return smp_load_acquire(&kvm->arch.shadow_root_allocated); 256} 257 258#ifdef CONFIG_X86_64 259extern bool tdp_mmu_enabled; 260#else 261#define tdp_mmu_enabled false 262#endif 263 264static inline bool kvm_memslots_have_rmaps(struct kvm *kvm) 265{ 266 return !tdp_mmu_enabled || kvm_shadow_root_allocated(kvm); 267} 268 269static inline gfn_t gfn_to_index(gfn_t gfn, gfn_t base_gfn, int level) 270{ 271 /* KVM_HPAGE_GFN_SHIFT(PG_LEVEL_4K) must be 0. */ 272 return (gfn >> KVM_HPAGE_GFN_SHIFT(level)) - 273 (base_gfn >> KVM_HPAGE_GFN_SHIFT(level)); 274} 275 276static inline unsigned long 277__kvm_mmu_slot_lpages(struct kvm_memory_slot *slot, unsigned long npages, 278 int level) 279{ 280 return gfn_to_index(slot->base_gfn + npages - 1, 281 slot->base_gfn, level) + 1; 282} 283 284static inline unsigned long 285kvm_mmu_slot_lpages(struct kvm_memory_slot *slot, int level) 286{ 287 return __kvm_mmu_slot_lpages(slot, slot->npages, level); 288} 289 290static inline void kvm_update_page_stats(struct kvm *kvm, int level, int count) 291{ 292 atomic64_add(count, &kvm->stat.pages[level - 1]); 293} 294 295gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u64 access, 296 struct x86_exception *exception); 297 298static inline gpa_t kvm_translate_gpa(struct kvm_vcpu *vcpu, 299 struct kvm_mmu *mmu, 300 gpa_t gpa, u64 access, 301 struct x86_exception *exception) 302{ 303 if (mmu != &vcpu->arch.nested_mmu) 304 return gpa; 305 return translate_nested_gpa(vcpu, gpa, access, exception); 306} 307#endif 308