xref: /linux/arch/x86/kvm/mmu.h (revision 4f77c3462308c62ffe7129cc18b9ac937f44b5a5)
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 
9 extern 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 
45 static __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 static inline gfn_t kvm_mmu_max_gfn(void)
61 {
62 	/*
63 	 * Note that this uses the host MAXPHYADDR, not the guest's.
64 	 * EPT/NPT cannot support GPAs that would exceed host.MAXPHYADDR;
65 	 * assuming KVM is running on bare metal, guest accesses beyond
66 	 * host.MAXPHYADDR will hit a #PF(RSVD) and never cause a vmexit
67 	 * (either EPT Violation/Misconfig or #NPF), and so KVM will never
68 	 * install a SPTE for such addresses.  If KVM is running as a VM
69 	 * itself, on the other hand, it might see a MAXPHYADDR that is less
70 	 * than hardware's real MAXPHYADDR.  Using the host MAXPHYADDR
71 	 * disallows such SPTEs entirely and simplifies the TDP MMU.
72 	 */
73 	int max_gpa_bits = likely(tdp_enabled) ? kvm_host.maxphyaddr : 52;
74 
75 	return (1ULL << (max_gpa_bits - PAGE_SHIFT)) - 1;
76 }
77 
78 u8 kvm_mmu_get_max_tdp_level(void);
79 
80 void kvm_mmu_set_mmio_spte_mask(u64 mmio_value, u64 mmio_mask, u64 access_mask);
81 void kvm_mmu_set_me_spte_mask(u64 me_value, u64 me_mask);
82 void kvm_mmu_set_ept_masks(bool has_ad_bits, bool has_exec_only);
83 
84 void kvm_init_mmu(struct kvm_vcpu *vcpu);
85 void kvm_init_shadow_npt_mmu(struct kvm_vcpu *vcpu, unsigned long cr0,
86 			     unsigned long cr4, u64 efer, gpa_t nested_cr3);
87 void kvm_init_shadow_ept_mmu(struct kvm_vcpu *vcpu, bool execonly,
88 			     int huge_page_level, bool accessed_dirty,
89 			     gpa_t new_eptp);
90 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu);
91 int kvm_handle_page_fault(struct kvm_vcpu *vcpu, u64 error_code,
92 				u64 fault_address, char *insn, int insn_len);
93 void __kvm_mmu_refresh_passthrough_bits(struct kvm_vcpu *vcpu,
94 					struct kvm_mmu *mmu);
95 
96 int kvm_mmu_load(struct kvm_vcpu *vcpu);
97 void kvm_mmu_unload(struct kvm_vcpu *vcpu);
98 void kvm_mmu_free_obsolete_roots(struct kvm_vcpu *vcpu);
99 void kvm_mmu_sync_roots(struct kvm_vcpu *vcpu);
100 void kvm_mmu_sync_prev_roots(struct kvm_vcpu *vcpu);
101 void kvm_mmu_track_write(struct kvm_vcpu *vcpu, gpa_t gpa, const u8 *new,
102 			 int bytes);
103 
104 static inline int kvm_mmu_reload(struct kvm_vcpu *vcpu)
105 {
106 	if (likely(vcpu->arch.mmu->root.hpa != INVALID_PAGE))
107 		return 0;
108 
109 	return kvm_mmu_load(vcpu);
110 }
111 
112 static inline unsigned long kvm_get_pcid(struct kvm_vcpu *vcpu, gpa_t cr3)
113 {
114 	BUILD_BUG_ON((X86_CR3_PCID_MASK & PAGE_MASK) != 0);
115 
116 	return kvm_is_cr4_bit_set(vcpu, X86_CR4_PCIDE)
117 	       ? cr3 & X86_CR3_PCID_MASK
118 	       : 0;
119 }
120 
121 static inline unsigned long kvm_get_active_pcid(struct kvm_vcpu *vcpu)
122 {
123 	return kvm_get_pcid(vcpu, kvm_read_cr3(vcpu));
124 }
125 
126 static inline unsigned long kvm_get_active_cr3_lam_bits(struct kvm_vcpu *vcpu)
127 {
128 	if (!guest_can_use(vcpu, X86_FEATURE_LAM))
129 		return 0;
130 
131 	return kvm_read_cr3(vcpu) & (X86_CR3_LAM_U48 | X86_CR3_LAM_U57);
132 }
133 
134 static inline void kvm_mmu_load_pgd(struct kvm_vcpu *vcpu)
135 {
136 	u64 root_hpa = vcpu->arch.mmu->root.hpa;
137 
138 	if (!VALID_PAGE(root_hpa))
139 		return;
140 
141 	kvm_x86_call(load_mmu_pgd)(vcpu, root_hpa,
142 				   vcpu->arch.mmu->root_role.level);
143 }
144 
145 static inline void kvm_mmu_refresh_passthrough_bits(struct kvm_vcpu *vcpu,
146 						    struct kvm_mmu *mmu)
147 {
148 	/*
149 	 * When EPT is enabled, KVM may passthrough CR0.WP to the guest, i.e.
150 	 * @mmu's snapshot of CR0.WP and thus all related paging metadata may
151 	 * be stale.  Refresh CR0.WP and the metadata on-demand when checking
152 	 * for permission faults.  Exempt nested MMUs, i.e. MMUs for shadowing
153 	 * nEPT and nNPT, as CR0.WP is ignored in both cases.  Note, KVM does
154 	 * need to refresh nested_mmu, a.k.a. the walker used to translate L2
155 	 * GVAs to GPAs, as that "MMU" needs to honor L2's CR0.WP.
156 	 */
157 	if (!tdp_enabled || mmu == &vcpu->arch.guest_mmu)
158 		return;
159 
160 	__kvm_mmu_refresh_passthrough_bits(vcpu, mmu);
161 }
162 
163 /*
164  * Check if a given access (described through the I/D, W/R and U/S bits of a
165  * page fault error code pfec) causes a permission fault with the given PTE
166  * access rights (in ACC_* format).
167  *
168  * Return zero if the access does not fault; return the page fault error code
169  * if the access faults.
170  */
171 static inline u8 permission_fault(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
172 				  unsigned pte_access, unsigned pte_pkey,
173 				  u64 access)
174 {
175 	/* strip nested paging fault error codes */
176 	unsigned int pfec = access;
177 	unsigned long rflags = kvm_x86_call(get_rflags)(vcpu);
178 
179 	/*
180 	 * For explicit supervisor accesses, SMAP is disabled if EFLAGS.AC = 1.
181 	 * For implicit supervisor accesses, SMAP cannot be overridden.
182 	 *
183 	 * SMAP works on supervisor accesses only, and not_smap can
184 	 * be set or not set when user access with neither has any bearing
185 	 * on the result.
186 	 *
187 	 * We put the SMAP checking bit in place of the PFERR_RSVD_MASK bit;
188 	 * this bit will always be zero in pfec, but it will be one in index
189 	 * if SMAP checks are being disabled.
190 	 */
191 	u64 implicit_access = access & PFERR_IMPLICIT_ACCESS;
192 	bool not_smap = ((rflags & X86_EFLAGS_AC) | implicit_access) == X86_EFLAGS_AC;
193 	int index = (pfec | (not_smap ? PFERR_RSVD_MASK : 0)) >> 1;
194 	u32 errcode = PFERR_PRESENT_MASK;
195 	bool fault;
196 
197 	kvm_mmu_refresh_passthrough_bits(vcpu, mmu);
198 
199 	fault = (mmu->permissions[index] >> pte_access) & 1;
200 
201 	WARN_ON(pfec & (PFERR_PK_MASK | PFERR_RSVD_MASK));
202 	if (unlikely(mmu->pkru_mask)) {
203 		u32 pkru_bits, offset;
204 
205 		/*
206 		* PKRU defines 32 bits, there are 16 domains and 2
207 		* attribute bits per domain in pkru.  pte_pkey is the
208 		* index of the protection domain, so pte_pkey * 2 is
209 		* is the index of the first bit for the domain.
210 		*/
211 		pkru_bits = (vcpu->arch.pkru >> (pte_pkey * 2)) & 3;
212 
213 		/* clear present bit, replace PFEC.RSVD with ACC_USER_MASK. */
214 		offset = (pfec & ~1) | ((pte_access & PT_USER_MASK) ? PFERR_RSVD_MASK : 0);
215 
216 		pkru_bits &= mmu->pkru_mask >> offset;
217 		errcode |= -pkru_bits & PFERR_PK_MASK;
218 		fault |= (pkru_bits != 0);
219 	}
220 
221 	return -(u32)fault & errcode;
222 }
223 
224 bool kvm_mmu_may_ignore_guest_pat(void);
225 
226 int kvm_arch_write_log_dirty(struct kvm_vcpu *vcpu);
227 
228 int kvm_mmu_post_init_vm(struct kvm *kvm);
229 void kvm_mmu_pre_destroy_vm(struct kvm *kvm);
230 
231 static inline bool kvm_shadow_root_allocated(struct kvm *kvm)
232 {
233 	/*
234 	 * Read shadow_root_allocated before related pointers. Hence, threads
235 	 * reading shadow_root_allocated in any lock context are guaranteed to
236 	 * see the pointers. Pairs with smp_store_release in
237 	 * mmu_first_shadow_root_alloc.
238 	 */
239 	return smp_load_acquire(&kvm->arch.shadow_root_allocated);
240 }
241 
242 #ifdef CONFIG_X86_64
243 extern bool tdp_mmu_enabled;
244 #else
245 #define tdp_mmu_enabled false
246 #endif
247 
248 static inline bool kvm_memslots_have_rmaps(struct kvm *kvm)
249 {
250 	return !tdp_mmu_enabled || kvm_shadow_root_allocated(kvm);
251 }
252 
253 static inline gfn_t gfn_to_index(gfn_t gfn, gfn_t base_gfn, int level)
254 {
255 	/* KVM_HPAGE_GFN_SHIFT(PG_LEVEL_4K) must be 0. */
256 	return (gfn >> KVM_HPAGE_GFN_SHIFT(level)) -
257 		(base_gfn >> KVM_HPAGE_GFN_SHIFT(level));
258 }
259 
260 static inline unsigned long
261 __kvm_mmu_slot_lpages(struct kvm_memory_slot *slot, unsigned long npages,
262 		      int level)
263 {
264 	return gfn_to_index(slot->base_gfn + npages - 1,
265 			    slot->base_gfn, level) + 1;
266 }
267 
268 static inline unsigned long
269 kvm_mmu_slot_lpages(struct kvm_memory_slot *slot, int level)
270 {
271 	return __kvm_mmu_slot_lpages(slot, slot->npages, level);
272 }
273 
274 static inline void kvm_update_page_stats(struct kvm *kvm, int level, int count)
275 {
276 	atomic64_add(count, &kvm->stat.pages[level - 1]);
277 }
278 
279 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u64 access,
280 			   struct x86_exception *exception);
281 
282 static inline gpa_t kvm_translate_gpa(struct kvm_vcpu *vcpu,
283 				      struct kvm_mmu *mmu,
284 				      gpa_t gpa, u64 access,
285 				      struct x86_exception *exception)
286 {
287 	if (mmu != &vcpu->arch.nested_mmu)
288 		return gpa;
289 	return translate_nested_gpa(vcpu, gpa, access, exception);
290 }
291 #endif
292