xref: /linux/arch/x86/kvm/mmu.h (revision f69e98a91a01fd7c5755dd710e94a17d6e9f583f)
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 void kvm_mmu_set_mmio_spte_mask(u64 mmio_value, u64 mmio_mask, u64 access_mask);
93 void kvm_mmu_set_ept_masks(bool has_ad_bits, bool has_exec_only);
94 
95 void kvm_init_mmu(struct kvm_vcpu *vcpu);
96 void kvm_init_shadow_npt_mmu(struct kvm_vcpu *vcpu, unsigned long cr0,
97 			     unsigned long cr4, u64 efer, gpa_t nested_cr3);
98 void kvm_init_shadow_ept_mmu(struct kvm_vcpu *vcpu, bool execonly,
99 			     int huge_page_level, bool accessed_dirty,
100 			     gpa_t new_eptp);
101 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu);
102 int kvm_handle_page_fault(struct kvm_vcpu *vcpu, u64 error_code,
103 				u64 fault_address, char *insn, int insn_len);
104 
105 int kvm_mmu_load(struct kvm_vcpu *vcpu);
106 void kvm_mmu_unload(struct kvm_vcpu *vcpu);
107 void kvm_mmu_free_obsolete_roots(struct kvm_vcpu *vcpu);
108 void kvm_mmu_sync_roots(struct kvm_vcpu *vcpu);
109 void kvm_mmu_sync_prev_roots(struct kvm_vcpu *vcpu);
110 
111 static inline int kvm_mmu_reload(struct kvm_vcpu *vcpu)
112 {
113 	if (likely(vcpu->arch.mmu->root.hpa != INVALID_PAGE))
114 		return 0;
115 
116 	return kvm_mmu_load(vcpu);
117 }
118 
119 static inline unsigned long kvm_get_pcid(struct kvm_vcpu *vcpu, gpa_t cr3)
120 {
121 	BUILD_BUG_ON((X86_CR3_PCID_MASK & PAGE_MASK) != 0);
122 
123 	return kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)
124 	       ? cr3 & X86_CR3_PCID_MASK
125 	       : 0;
126 }
127 
128 static inline unsigned long kvm_get_active_pcid(struct kvm_vcpu *vcpu)
129 {
130 	return kvm_get_pcid(vcpu, kvm_read_cr3(vcpu));
131 }
132 
133 static inline void kvm_mmu_load_pgd(struct kvm_vcpu *vcpu)
134 {
135 	u64 root_hpa = vcpu->arch.mmu->root.hpa;
136 
137 	if (!VALID_PAGE(root_hpa))
138 		return;
139 
140 	static_call(kvm_x86_load_mmu_pgd)(vcpu, root_hpa,
141 					  vcpu->arch.mmu->shadow_root_level);
142 }
143 
144 struct kvm_page_fault {
145 	/* arguments to kvm_mmu_do_page_fault.  */
146 	const gpa_t addr;
147 	const u32 error_code;
148 	const bool prefetch;
149 
150 	/* Derived from error_code.  */
151 	const bool exec;
152 	const bool write;
153 	const bool present;
154 	const bool rsvd;
155 	const bool user;
156 
157 	/* Derived from mmu and global state.  */
158 	const bool is_tdp;
159 	const bool nx_huge_page_workaround_enabled;
160 
161 	/*
162 	 * Whether a >4KB mapping can be created or is forbidden due to NX
163 	 * hugepages.
164 	 */
165 	bool huge_page_disallowed;
166 
167 	/*
168 	 * Maximum page size that can be created for this fault; input to
169 	 * FNAME(fetch), __direct_map and kvm_tdp_mmu_map.
170 	 */
171 	u8 max_level;
172 
173 	/*
174 	 * Page size that can be created based on the max_level and the
175 	 * page size used by the host mapping.
176 	 */
177 	u8 req_level;
178 
179 	/*
180 	 * Page size that will be created based on the req_level and
181 	 * huge_page_disallowed.
182 	 */
183 	u8 goal_level;
184 
185 	/* Shifted addr, or result of guest page table walk if addr is a gva.  */
186 	gfn_t gfn;
187 
188 	/* The memslot containing gfn. May be NULL. */
189 	struct kvm_memory_slot *slot;
190 
191 	/* Outputs of kvm_faultin_pfn.  */
192 	kvm_pfn_t pfn;
193 	hva_t hva;
194 	bool map_writable;
195 };
196 
197 int kvm_tdp_page_fault(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault);
198 
199 extern int nx_huge_pages;
200 static inline bool is_nx_huge_page_enabled(void)
201 {
202 	return READ_ONCE(nx_huge_pages);
203 }
204 
205 static inline int kvm_mmu_do_page_fault(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
206 					u32 err, bool prefetch)
207 {
208 	struct kvm_page_fault fault = {
209 		.addr = cr2_or_gpa,
210 		.error_code = err,
211 		.exec = err & PFERR_FETCH_MASK,
212 		.write = err & PFERR_WRITE_MASK,
213 		.present = err & PFERR_PRESENT_MASK,
214 		.rsvd = err & PFERR_RSVD_MASK,
215 		.user = err & PFERR_USER_MASK,
216 		.prefetch = prefetch,
217 		.is_tdp = likely(vcpu->arch.mmu->page_fault == kvm_tdp_page_fault),
218 		.nx_huge_page_workaround_enabled = is_nx_huge_page_enabled(),
219 
220 		.max_level = KVM_MAX_HUGEPAGE_LEVEL,
221 		.req_level = PG_LEVEL_4K,
222 		.goal_level = PG_LEVEL_4K,
223 	};
224 #ifdef CONFIG_RETPOLINE
225 	if (fault.is_tdp)
226 		return kvm_tdp_page_fault(vcpu, &fault);
227 #endif
228 	return vcpu->arch.mmu->page_fault(vcpu, &fault);
229 }
230 
231 /*
232  * Check if a given access (described through the I/D, W/R and U/S bits of a
233  * page fault error code pfec) causes a permission fault with the given PTE
234  * access rights (in ACC_* format).
235  *
236  * Return zero if the access does not fault; return the page fault error code
237  * if the access faults.
238  */
239 static inline u8 permission_fault(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
240 				  unsigned pte_access, unsigned pte_pkey,
241 				  u64 access)
242 {
243 	/* strip nested paging fault error codes */
244 	unsigned int pfec = access;
245 	unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
246 
247 	/*
248 	 * For explicit supervisor accesses, SMAP is disabled if EFLAGS.AC = 1.
249 	 * For implicit supervisor accesses, SMAP cannot be overridden.
250 	 *
251 	 * SMAP works on supervisor accesses only, and not_smap can
252 	 * be set or not set when user access with neither has any bearing
253 	 * on the result.
254 	 *
255 	 * We put the SMAP checking bit in place of the PFERR_RSVD_MASK bit;
256 	 * this bit will always be zero in pfec, but it will be one in index
257 	 * if SMAP checks are being disabled.
258 	 */
259 	u64 implicit_access = access & PFERR_IMPLICIT_ACCESS;
260 	bool not_smap = ((rflags & X86_EFLAGS_AC) | implicit_access) == X86_EFLAGS_AC;
261 	int index = (pfec + (not_smap << PFERR_RSVD_BIT)) >> 1;
262 	bool fault = (mmu->permissions[index] >> pte_access) & 1;
263 	u32 errcode = PFERR_PRESENT_MASK;
264 
265 	WARN_ON(pfec & (PFERR_PK_MASK | PFERR_RSVD_MASK));
266 	if (unlikely(mmu->pkru_mask)) {
267 		u32 pkru_bits, offset;
268 
269 		/*
270 		* PKRU defines 32 bits, there are 16 domains and 2
271 		* attribute bits per domain in pkru.  pte_pkey is the
272 		* index of the protection domain, so pte_pkey * 2 is
273 		* is the index of the first bit for the domain.
274 		*/
275 		pkru_bits = (vcpu->arch.pkru >> (pte_pkey * 2)) & 3;
276 
277 		/* clear present bit, replace PFEC.RSVD with ACC_USER_MASK. */
278 		offset = (pfec & ~1) +
279 			((pte_access & PT_USER_MASK) << (PFERR_RSVD_BIT - PT_USER_SHIFT));
280 
281 		pkru_bits &= mmu->pkru_mask >> offset;
282 		errcode |= -pkru_bits & PFERR_PK_MASK;
283 		fault |= (pkru_bits != 0);
284 	}
285 
286 	return -(u32)fault & errcode;
287 }
288 
289 void kvm_zap_gfn_range(struct kvm *kvm, gfn_t gfn_start, gfn_t gfn_end);
290 
291 int kvm_arch_write_log_dirty(struct kvm_vcpu *vcpu);
292 
293 int kvm_mmu_post_init_vm(struct kvm *kvm);
294 void kvm_mmu_pre_destroy_vm(struct kvm *kvm);
295 
296 static inline bool kvm_shadow_root_allocated(struct kvm *kvm)
297 {
298 	/*
299 	 * Read shadow_root_allocated before related pointers. Hence, threads
300 	 * reading shadow_root_allocated in any lock context are guaranteed to
301 	 * see the pointers. Pairs with smp_store_release in
302 	 * mmu_first_shadow_root_alloc.
303 	 */
304 	return smp_load_acquire(&kvm->arch.shadow_root_allocated);
305 }
306 
307 #ifdef CONFIG_X86_64
308 static inline bool is_tdp_mmu_enabled(struct kvm *kvm) { return kvm->arch.tdp_mmu_enabled; }
309 #else
310 static inline bool is_tdp_mmu_enabled(struct kvm *kvm) { return false; }
311 #endif
312 
313 static inline bool kvm_memslots_have_rmaps(struct kvm *kvm)
314 {
315 	return !is_tdp_mmu_enabled(kvm) || kvm_shadow_root_allocated(kvm);
316 }
317 
318 static inline gfn_t gfn_to_index(gfn_t gfn, gfn_t base_gfn, int level)
319 {
320 	/* KVM_HPAGE_GFN_SHIFT(PG_LEVEL_4K) must be 0. */
321 	return (gfn >> KVM_HPAGE_GFN_SHIFT(level)) -
322 		(base_gfn >> KVM_HPAGE_GFN_SHIFT(level));
323 }
324 
325 static inline unsigned long
326 __kvm_mmu_slot_lpages(struct kvm_memory_slot *slot, unsigned long npages,
327 		      int level)
328 {
329 	return gfn_to_index(slot->base_gfn + npages - 1,
330 			    slot->base_gfn, level) + 1;
331 }
332 
333 static inline unsigned long
334 kvm_mmu_slot_lpages(struct kvm_memory_slot *slot, int level)
335 {
336 	return __kvm_mmu_slot_lpages(slot, slot->npages, level);
337 }
338 
339 static inline void kvm_update_page_stats(struct kvm *kvm, int level, int count)
340 {
341 	atomic64_add(count, &kvm->stat.pages[level - 1]);
342 }
343 
344 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u64 access,
345 			   struct x86_exception *exception);
346 
347 static inline gpa_t kvm_translate_gpa(struct kvm_vcpu *vcpu,
348 				      struct kvm_mmu *mmu,
349 				      gpa_t gpa, u64 access,
350 				      struct x86_exception *exception)
351 {
352 	if (mmu != &vcpu->arch.nested_mmu)
353 		return gpa;
354 	return translate_nested_gpa(vcpu, gpa, access, exception);
355 }
356 #endif
357