xref: /openbmc/linux/arch/x86/kvm/mmu.h (revision 38857318)
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 #define PT64_PT_BITS 9
10 #define PT64_ENT_PER_PAGE (1 << PT64_PT_BITS)
11 #define PT32_PT_BITS 10
12 #define PT32_ENT_PER_PAGE (1 << PT32_PT_BITS)
13 
14 #define PT_WRITABLE_SHIFT 1
15 #define PT_USER_SHIFT 2
16 
17 #define PT_PRESENT_MASK (1ULL << 0)
18 #define PT_WRITABLE_MASK (1ULL << PT_WRITABLE_SHIFT)
19 #define PT_USER_MASK (1ULL << PT_USER_SHIFT)
20 #define PT_PWT_MASK (1ULL << 3)
21 #define PT_PCD_MASK (1ULL << 4)
22 #define PT_ACCESSED_SHIFT 5
23 #define PT_ACCESSED_MASK (1ULL << PT_ACCESSED_SHIFT)
24 #define PT_DIRTY_SHIFT 6
25 #define PT_DIRTY_MASK (1ULL << PT_DIRTY_SHIFT)
26 #define PT_PAGE_SIZE_SHIFT 7
27 #define PT_PAGE_SIZE_MASK (1ULL << PT_PAGE_SIZE_SHIFT)
28 #define PT_PAT_MASK (1ULL << 7)
29 #define PT_GLOBAL_MASK (1ULL << 8)
30 #define PT64_NX_SHIFT 63
31 #define PT64_NX_MASK (1ULL << PT64_NX_SHIFT)
32 
33 #define PT_PAT_SHIFT 7
34 #define PT_DIR_PAT_SHIFT 12
35 #define PT_DIR_PAT_MASK (1ULL << PT_DIR_PAT_SHIFT)
36 
37 #define PT32_DIR_PSE36_SIZE 4
38 #define PT32_DIR_PSE36_SHIFT 13
39 #define PT32_DIR_PSE36_MASK \
40 	(((1ULL << PT32_DIR_PSE36_SIZE) - 1) << PT32_DIR_PSE36_SHIFT)
41 
42 #define PT64_ROOT_5LEVEL 5
43 #define PT64_ROOT_4LEVEL 4
44 #define PT32_ROOT_LEVEL 2
45 #define PT32E_ROOT_LEVEL 3
46 
47 #define KVM_MMU_CR4_ROLE_BITS (X86_CR4_PSE | X86_CR4_PAE | X86_CR4_LA57 | \
48 			       X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE)
49 
50 #define KVM_MMU_CR0_ROLE_BITS (X86_CR0_PG | X86_CR0_WP)
51 #define KVM_MMU_EFER_ROLE_BITS (EFER_LME | EFER_NX)
52 
53 static __always_inline u64 rsvd_bits(int s, int e)
54 {
55 	BUILD_BUG_ON(__builtin_constant_p(e) && __builtin_constant_p(s) && e < s);
56 
57 	if (__builtin_constant_p(e))
58 		BUILD_BUG_ON(e > 63);
59 	else
60 		e &= 63;
61 
62 	if (e < s)
63 		return 0;
64 
65 	return ((2ULL << (e - s)) - 1) << s;
66 }
67 
68 /*
69  * The number of non-reserved physical address bits irrespective of features
70  * that repurpose legal bits, e.g. MKTME.
71  */
72 extern u8 __read_mostly shadow_phys_bits;
73 
74 static inline gfn_t kvm_mmu_max_gfn(void)
75 {
76 	/*
77 	 * Note that this uses the host MAXPHYADDR, not the guest's.
78 	 * EPT/NPT cannot support GPAs that would exceed host.MAXPHYADDR;
79 	 * assuming KVM is running on bare metal, guest accesses beyond
80 	 * host.MAXPHYADDR will hit a #PF(RSVD) and never cause a vmexit
81 	 * (either EPT Violation/Misconfig or #NPF), and so KVM will never
82 	 * install a SPTE for such addresses.  If KVM is running as a VM
83 	 * itself, on the other hand, it might see a MAXPHYADDR that is less
84 	 * than hardware's real MAXPHYADDR.  Using the host MAXPHYADDR
85 	 * disallows such SPTEs entirely and simplifies the TDP MMU.
86 	 */
87 	int max_gpa_bits = likely(tdp_enabled) ? shadow_phys_bits : 52;
88 
89 	return (1ULL << (max_gpa_bits - PAGE_SHIFT)) - 1;
90 }
91 
92 static inline u8 kvm_get_shadow_phys_bits(void)
93 {
94 	/*
95 	 * boot_cpu_data.x86_phys_bits is reduced when MKTME or SME are detected
96 	 * in CPU detection code, but the processor treats those reduced bits as
97 	 * 'keyID' thus they are not reserved bits. Therefore KVM needs to look at
98 	 * the physical address bits reported by CPUID.
99 	 */
100 	if (likely(boot_cpu_data.extended_cpuid_level >= 0x80000008))
101 		return cpuid_eax(0x80000008) & 0xff;
102 
103 	/*
104 	 * Quite weird to have VMX or SVM but not MAXPHYADDR; probably a VM with
105 	 * custom CPUID.  Proceed with whatever the kernel found since these features
106 	 * aren't virtualizable (SME/SEV also require CPUIDs higher than 0x80000008).
107 	 */
108 	return boot_cpu_data.x86_phys_bits;
109 }
110 
111 void kvm_mmu_set_mmio_spte_mask(u64 mmio_value, u64 mmio_mask, u64 access_mask);
112 void kvm_mmu_set_me_spte_mask(u64 me_value, u64 me_mask);
113 void kvm_mmu_set_ept_masks(bool has_ad_bits, bool has_exec_only);
114 
115 void kvm_init_mmu(struct kvm_vcpu *vcpu);
116 void kvm_init_shadow_npt_mmu(struct kvm_vcpu *vcpu, unsigned long cr0,
117 			     unsigned long cr4, u64 efer, gpa_t nested_cr3);
118 void kvm_init_shadow_ept_mmu(struct kvm_vcpu *vcpu, bool execonly,
119 			     int huge_page_level, bool accessed_dirty,
120 			     gpa_t new_eptp);
121 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu);
122 int kvm_handle_page_fault(struct kvm_vcpu *vcpu, u64 error_code,
123 				u64 fault_address, char *insn, int insn_len);
124 
125 int kvm_mmu_load(struct kvm_vcpu *vcpu);
126 void kvm_mmu_unload(struct kvm_vcpu *vcpu);
127 void kvm_mmu_free_obsolete_roots(struct kvm_vcpu *vcpu);
128 void kvm_mmu_sync_roots(struct kvm_vcpu *vcpu);
129 void kvm_mmu_sync_prev_roots(struct kvm_vcpu *vcpu);
130 
131 static inline int kvm_mmu_reload(struct kvm_vcpu *vcpu)
132 {
133 	if (likely(vcpu->arch.mmu->root.hpa != INVALID_PAGE))
134 		return 0;
135 
136 	return kvm_mmu_load(vcpu);
137 }
138 
139 static inline unsigned long kvm_get_pcid(struct kvm_vcpu *vcpu, gpa_t cr3)
140 {
141 	BUILD_BUG_ON((X86_CR3_PCID_MASK & PAGE_MASK) != 0);
142 
143 	return kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)
144 	       ? cr3 & X86_CR3_PCID_MASK
145 	       : 0;
146 }
147 
148 static inline unsigned long kvm_get_active_pcid(struct kvm_vcpu *vcpu)
149 {
150 	return kvm_get_pcid(vcpu, kvm_read_cr3(vcpu));
151 }
152 
153 static inline void kvm_mmu_load_pgd(struct kvm_vcpu *vcpu)
154 {
155 	u64 root_hpa = vcpu->arch.mmu->root.hpa;
156 
157 	if (!VALID_PAGE(root_hpa))
158 		return;
159 
160 	static_call(kvm_x86_load_mmu_pgd)(vcpu, root_hpa,
161 					  vcpu->arch.mmu->root_role.level);
162 }
163 
164 /*
165  * Check if a given access (described through the I/D, W/R and U/S bits of a
166  * page fault error code pfec) causes a permission fault with the given PTE
167  * access rights (in ACC_* format).
168  *
169  * Return zero if the access does not fault; return the page fault error code
170  * if the access faults.
171  */
172 static inline u8 permission_fault(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
173 				  unsigned pte_access, unsigned pte_pkey,
174 				  u64 access)
175 {
176 	/* strip nested paging fault error codes */
177 	unsigned int pfec = access;
178 	unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
179 
180 	/*
181 	 * For explicit supervisor accesses, SMAP is disabled if EFLAGS.AC = 1.
182 	 * For implicit supervisor accesses, SMAP cannot be overridden.
183 	 *
184 	 * SMAP works on supervisor accesses only, and not_smap can
185 	 * be set or not set when user access with neither has any bearing
186 	 * on the result.
187 	 *
188 	 * We put the SMAP checking bit in place of the PFERR_RSVD_MASK bit;
189 	 * this bit will always be zero in pfec, but it will be one in index
190 	 * if SMAP checks are being disabled.
191 	 */
192 	u64 implicit_access = access & PFERR_IMPLICIT_ACCESS;
193 	bool not_smap = ((rflags & X86_EFLAGS_AC) | implicit_access) == X86_EFLAGS_AC;
194 	int index = (pfec + (not_smap << PFERR_RSVD_BIT)) >> 1;
195 	bool fault = (mmu->permissions[index] >> pte_access) & 1;
196 	u32 errcode = PFERR_PRESENT_MASK;
197 
198 	WARN_ON(pfec & (PFERR_PK_MASK | PFERR_RSVD_MASK));
199 	if (unlikely(mmu->pkru_mask)) {
200 		u32 pkru_bits, offset;
201 
202 		/*
203 		* PKRU defines 32 bits, there are 16 domains and 2
204 		* attribute bits per domain in pkru.  pte_pkey is the
205 		* index of the protection domain, so pte_pkey * 2 is
206 		* is the index of the first bit for the domain.
207 		*/
208 		pkru_bits = (vcpu->arch.pkru >> (pte_pkey * 2)) & 3;
209 
210 		/* clear present bit, replace PFEC.RSVD with ACC_USER_MASK. */
211 		offset = (pfec & ~1) +
212 			((pte_access & PT_USER_MASK) << (PFERR_RSVD_BIT - PT_USER_SHIFT));
213 
214 		pkru_bits &= mmu->pkru_mask >> offset;
215 		errcode |= -pkru_bits & PFERR_PK_MASK;
216 		fault |= (pkru_bits != 0);
217 	}
218 
219 	return -(u32)fault & errcode;
220 }
221 
222 void kvm_zap_gfn_range(struct kvm *kvm, gfn_t gfn_start, gfn_t gfn_end);
223 
224 int kvm_arch_write_log_dirty(struct kvm_vcpu *vcpu);
225 
226 int kvm_mmu_post_init_vm(struct kvm *kvm);
227 void kvm_mmu_pre_destroy_vm(struct kvm *kvm);
228 
229 static inline bool kvm_shadow_root_allocated(struct kvm *kvm)
230 {
231 	/*
232 	 * Read shadow_root_allocated before related pointers. Hence, threads
233 	 * reading shadow_root_allocated in any lock context are guaranteed to
234 	 * see the pointers. Pairs with smp_store_release in
235 	 * mmu_first_shadow_root_alloc.
236 	 */
237 	return smp_load_acquire(&kvm->arch.shadow_root_allocated);
238 }
239 
240 #ifdef CONFIG_X86_64
241 static inline bool is_tdp_mmu_enabled(struct kvm *kvm) { return kvm->arch.tdp_mmu_enabled; }
242 #else
243 static inline bool is_tdp_mmu_enabled(struct kvm *kvm) { return false; }
244 #endif
245 
246 static inline bool kvm_memslots_have_rmaps(struct kvm *kvm)
247 {
248 	return !is_tdp_mmu_enabled(kvm) || kvm_shadow_root_allocated(kvm);
249 }
250 
251 static inline gfn_t gfn_to_index(gfn_t gfn, gfn_t base_gfn, int level)
252 {
253 	/* KVM_HPAGE_GFN_SHIFT(PG_LEVEL_4K) must be 0. */
254 	return (gfn >> KVM_HPAGE_GFN_SHIFT(level)) -
255 		(base_gfn >> KVM_HPAGE_GFN_SHIFT(level));
256 }
257 
258 static inline unsigned long
259 __kvm_mmu_slot_lpages(struct kvm_memory_slot *slot, unsigned long npages,
260 		      int level)
261 {
262 	return gfn_to_index(slot->base_gfn + npages - 1,
263 			    slot->base_gfn, level) + 1;
264 }
265 
266 static inline unsigned long
267 kvm_mmu_slot_lpages(struct kvm_memory_slot *slot, int level)
268 {
269 	return __kvm_mmu_slot_lpages(slot, slot->npages, level);
270 }
271 
272 static inline void kvm_update_page_stats(struct kvm *kvm, int level, int count)
273 {
274 	atomic64_add(count, &kvm->stat.pages[level - 1]);
275 }
276 
277 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u64 access,
278 			   struct x86_exception *exception);
279 
280 static inline gpa_t kvm_translate_gpa(struct kvm_vcpu *vcpu,
281 				      struct kvm_mmu *mmu,
282 				      gpa_t gpa, u64 access,
283 				      struct x86_exception *exception)
284 {
285 	if (mmu != &vcpu->arch.nested_mmu)
286 		return gpa;
287 	return translate_nested_gpa(vcpu, gpa, access, exception);
288 }
289 #endif
290