xref: /linux/arch/x86/kvm/mmu.h (revision 1a9a71439cc1b270bf127c2f529aac7cf9cb21ab)
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 /*
61  * The number of non-reserved physical address bits irrespective of features
62  * that repurpose legal bits, e.g. MKTME.
63  */
64 extern u8 __read_mostly shadow_phys_bits;
65 
66 static 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 
84 static 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 
103 u8 kvm_mmu_get_max_tdp_level(void);
104 
105 void kvm_mmu_set_mmio_spte_mask(u64 mmio_value, u64 mmio_mask, u64 access_mask);
106 void kvm_mmu_set_me_spte_mask(u64 me_value, u64 me_mask);
107 void kvm_mmu_set_ept_masks(bool has_ad_bits, bool has_exec_only);
108 
109 void kvm_init_mmu(struct kvm_vcpu *vcpu);
110 void kvm_init_shadow_npt_mmu(struct kvm_vcpu *vcpu, unsigned long cr0,
111 			     unsigned long cr4, u64 efer, gpa_t nested_cr3);
112 void kvm_init_shadow_ept_mmu(struct kvm_vcpu *vcpu, bool execonly,
113 			     int huge_page_level, bool accessed_dirty,
114 			     gpa_t new_eptp);
115 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu);
116 int kvm_handle_page_fault(struct kvm_vcpu *vcpu, u64 error_code,
117 				u64 fault_address, char *insn, int insn_len);
118 void __kvm_mmu_refresh_passthrough_bits(struct kvm_vcpu *vcpu,
119 					struct kvm_mmu *mmu);
120 
121 int kvm_mmu_load(struct kvm_vcpu *vcpu);
122 void kvm_mmu_unload(struct kvm_vcpu *vcpu);
123 void kvm_mmu_free_obsolete_roots(struct kvm_vcpu *vcpu);
124 void kvm_mmu_sync_roots(struct kvm_vcpu *vcpu);
125 void kvm_mmu_sync_prev_roots(struct kvm_vcpu *vcpu);
126 void kvm_mmu_track_write(struct kvm_vcpu *vcpu, gpa_t gpa, const u8 *new,
127 			 int bytes);
128 
129 static inline int kvm_mmu_reload(struct kvm_vcpu *vcpu)
130 {
131 	if (likely(vcpu->arch.mmu->root.hpa != INVALID_PAGE))
132 		return 0;
133 
134 	return kvm_mmu_load(vcpu);
135 }
136 
137 static inline unsigned long kvm_get_pcid(struct kvm_vcpu *vcpu, gpa_t cr3)
138 {
139 	BUILD_BUG_ON((X86_CR3_PCID_MASK & PAGE_MASK) != 0);
140 
141 	return kvm_is_cr4_bit_set(vcpu, X86_CR4_PCIDE)
142 	       ? cr3 & X86_CR3_PCID_MASK
143 	       : 0;
144 }
145 
146 static inline unsigned long kvm_get_active_pcid(struct kvm_vcpu *vcpu)
147 {
148 	return kvm_get_pcid(vcpu, kvm_read_cr3(vcpu));
149 }
150 
151 static inline unsigned long kvm_get_active_cr3_lam_bits(struct kvm_vcpu *vcpu)
152 {
153 	if (!guest_can_use(vcpu, X86_FEATURE_LAM))
154 		return 0;
155 
156 	return kvm_read_cr3(vcpu) & (X86_CR3_LAM_U48 | X86_CR3_LAM_U57);
157 }
158 
159 static inline void kvm_mmu_load_pgd(struct kvm_vcpu *vcpu)
160 {
161 	u64 root_hpa = vcpu->arch.mmu->root.hpa;
162 
163 	if (!VALID_PAGE(root_hpa))
164 		return;
165 
166 	static_call(kvm_x86_load_mmu_pgd)(vcpu, root_hpa,
167 					  vcpu->arch.mmu->root_role.level);
168 }
169 
170 static inline void kvm_mmu_refresh_passthrough_bits(struct kvm_vcpu *vcpu,
171 						    struct kvm_mmu *mmu)
172 {
173 	/*
174 	 * When EPT is enabled, KVM may passthrough CR0.WP to the guest, i.e.
175 	 * @mmu's snapshot of CR0.WP and thus all related paging metadata may
176 	 * be stale.  Refresh CR0.WP and the metadata on-demand when checking
177 	 * for permission faults.  Exempt nested MMUs, i.e. MMUs for shadowing
178 	 * nEPT and nNPT, as CR0.WP is ignored in both cases.  Note, KVM does
179 	 * need to refresh nested_mmu, a.k.a. the walker used to translate L2
180 	 * GVAs to GPAs, as that "MMU" needs to honor L2's CR0.WP.
181 	 */
182 	if (!tdp_enabled || mmu == &vcpu->arch.guest_mmu)
183 		return;
184 
185 	__kvm_mmu_refresh_passthrough_bits(vcpu, mmu);
186 }
187 
188 /*
189  * Check if a given access (described through the I/D, W/R and U/S bits of a
190  * page fault error code pfec) causes a permission fault with the given PTE
191  * access rights (in ACC_* format).
192  *
193  * Return zero if the access does not fault; return the page fault error code
194  * if the access faults.
195  */
196 static inline u8 permission_fault(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
197 				  unsigned pte_access, unsigned pte_pkey,
198 				  u64 access)
199 {
200 	/* strip nested paging fault error codes */
201 	unsigned int pfec = access;
202 	unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
203 
204 	/*
205 	 * For explicit supervisor accesses, SMAP is disabled if EFLAGS.AC = 1.
206 	 * For implicit supervisor accesses, SMAP cannot be overridden.
207 	 *
208 	 * SMAP works on supervisor accesses only, and not_smap can
209 	 * be set or not set when user access with neither has any bearing
210 	 * on the result.
211 	 *
212 	 * We put the SMAP checking bit in place of the PFERR_RSVD_MASK bit;
213 	 * this bit will always be zero in pfec, but it will be one in index
214 	 * if SMAP checks are being disabled.
215 	 */
216 	u64 implicit_access = access & PFERR_IMPLICIT_ACCESS;
217 	bool not_smap = ((rflags & X86_EFLAGS_AC) | implicit_access) == X86_EFLAGS_AC;
218 	int index = (pfec | (not_smap ? PFERR_RSVD_MASK : 0)) >> 1;
219 	u32 errcode = PFERR_PRESENT_MASK;
220 	bool fault;
221 
222 	kvm_mmu_refresh_passthrough_bits(vcpu, mmu);
223 
224 	fault = (mmu->permissions[index] >> pte_access) & 1;
225 
226 	WARN_ON(pfec & (PFERR_PK_MASK | PFERR_RSVD_MASK));
227 	if (unlikely(mmu->pkru_mask)) {
228 		u32 pkru_bits, offset;
229 
230 		/*
231 		* PKRU defines 32 bits, there are 16 domains and 2
232 		* attribute bits per domain in pkru.  pte_pkey is the
233 		* index of the protection domain, so pte_pkey * 2 is
234 		* is the index of the first bit for the domain.
235 		*/
236 		pkru_bits = (vcpu->arch.pkru >> (pte_pkey * 2)) & 3;
237 
238 		/* clear present bit, replace PFEC.RSVD with ACC_USER_MASK. */
239 		offset = (pfec & ~1) | ((pte_access & PT_USER_MASK) ? PFERR_RSVD_MASK : 0);
240 
241 		pkru_bits &= mmu->pkru_mask >> offset;
242 		errcode |= -pkru_bits & PFERR_PK_MASK;
243 		fault |= (pkru_bits != 0);
244 	}
245 
246 	return -(u32)fault & errcode;
247 }
248 
249 bool __kvm_mmu_honors_guest_mtrrs(bool vm_has_noncoherent_dma);
250 
251 static inline bool kvm_mmu_honors_guest_mtrrs(struct kvm *kvm)
252 {
253 	return __kvm_mmu_honors_guest_mtrrs(kvm_arch_has_noncoherent_dma(kvm));
254 }
255 
256 void kvm_zap_gfn_range(struct kvm *kvm, gfn_t gfn_start, gfn_t gfn_end);
257 
258 int kvm_arch_write_log_dirty(struct kvm_vcpu *vcpu);
259 
260 int kvm_mmu_post_init_vm(struct kvm *kvm);
261 void kvm_mmu_pre_destroy_vm(struct kvm *kvm);
262 
263 static inline bool kvm_shadow_root_allocated(struct kvm *kvm)
264 {
265 	/*
266 	 * Read shadow_root_allocated before related pointers. Hence, threads
267 	 * reading shadow_root_allocated in any lock context are guaranteed to
268 	 * see the pointers. Pairs with smp_store_release in
269 	 * mmu_first_shadow_root_alloc.
270 	 */
271 	return smp_load_acquire(&kvm->arch.shadow_root_allocated);
272 }
273 
274 #ifdef CONFIG_X86_64
275 extern bool tdp_mmu_enabled;
276 #else
277 #define tdp_mmu_enabled false
278 #endif
279 
280 static inline bool kvm_memslots_have_rmaps(struct kvm *kvm)
281 {
282 	return !tdp_mmu_enabled || kvm_shadow_root_allocated(kvm);
283 }
284 
285 static inline gfn_t gfn_to_index(gfn_t gfn, gfn_t base_gfn, int level)
286 {
287 	/* KVM_HPAGE_GFN_SHIFT(PG_LEVEL_4K) must be 0. */
288 	return (gfn >> KVM_HPAGE_GFN_SHIFT(level)) -
289 		(base_gfn >> KVM_HPAGE_GFN_SHIFT(level));
290 }
291 
292 static inline unsigned long
293 __kvm_mmu_slot_lpages(struct kvm_memory_slot *slot, unsigned long npages,
294 		      int level)
295 {
296 	return gfn_to_index(slot->base_gfn + npages - 1,
297 			    slot->base_gfn, level) + 1;
298 }
299 
300 static inline unsigned long
301 kvm_mmu_slot_lpages(struct kvm_memory_slot *slot, int level)
302 {
303 	return __kvm_mmu_slot_lpages(slot, slot->npages, level);
304 }
305 
306 static inline void kvm_update_page_stats(struct kvm *kvm, int level, int count)
307 {
308 	atomic64_add(count, &kvm->stat.pages[level - 1]);
309 }
310 
311 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u64 access,
312 			   struct x86_exception *exception);
313 
314 static inline gpa_t kvm_translate_gpa(struct kvm_vcpu *vcpu,
315 				      struct kvm_mmu *mmu,
316 				      gpa_t gpa, u64 access,
317 				      struct x86_exception *exception)
318 {
319 	if (mmu != &vcpu->arch.nested_mmu)
320 		return gpa;
321 	return translate_nested_gpa(vcpu, gpa, access, exception);
322 }
323 #endif
324