xref: /openbmc/linux/arch/x86/kvm/mmu.h (revision d6174299365ddbbf491620c0b8c5ca1a6ef2eea5)
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 void kvm_mmu_set_mmio_spte_mask(u64 mmio_value, u64 mmio_mask, u64 access_mask);
69 void kvm_mmu_set_ept_masks(bool has_ad_bits, bool has_exec_only);
70 
71 void kvm_init_mmu(struct kvm_vcpu *vcpu);
72 void kvm_init_shadow_npt_mmu(struct kvm_vcpu *vcpu, unsigned long cr0,
73 			     unsigned long cr4, u64 efer, gpa_t nested_cr3);
74 void kvm_init_shadow_ept_mmu(struct kvm_vcpu *vcpu, bool execonly,
75 			     int huge_page_level, bool accessed_dirty,
76 			     gpa_t new_eptp);
77 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu);
78 int kvm_handle_page_fault(struct kvm_vcpu *vcpu, u64 error_code,
79 				u64 fault_address, char *insn, int insn_len);
80 
81 int kvm_mmu_load(struct kvm_vcpu *vcpu);
82 void kvm_mmu_unload(struct kvm_vcpu *vcpu);
83 void kvm_mmu_sync_roots(struct kvm_vcpu *vcpu);
84 void kvm_mmu_sync_prev_roots(struct kvm_vcpu *vcpu);
85 
86 static inline int kvm_mmu_reload(struct kvm_vcpu *vcpu)
87 {
88 	if (likely(vcpu->arch.mmu->root_hpa != INVALID_PAGE))
89 		return 0;
90 
91 	return kvm_mmu_load(vcpu);
92 }
93 
94 static inline unsigned long kvm_get_pcid(struct kvm_vcpu *vcpu, gpa_t cr3)
95 {
96 	BUILD_BUG_ON((X86_CR3_PCID_MASK & PAGE_MASK) != 0);
97 
98 	return kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)
99 	       ? cr3 & X86_CR3_PCID_MASK
100 	       : 0;
101 }
102 
103 static inline unsigned long kvm_get_active_pcid(struct kvm_vcpu *vcpu)
104 {
105 	return kvm_get_pcid(vcpu, kvm_read_cr3(vcpu));
106 }
107 
108 static inline void kvm_mmu_load_pgd(struct kvm_vcpu *vcpu)
109 {
110 	u64 root_hpa = vcpu->arch.mmu->root_hpa;
111 
112 	if (!VALID_PAGE(root_hpa))
113 		return;
114 
115 	static_call(kvm_x86_load_mmu_pgd)(vcpu, root_hpa,
116 					  vcpu->arch.mmu->shadow_root_level);
117 }
118 
119 struct kvm_page_fault {
120 	/* arguments to kvm_mmu_do_page_fault.  */
121 	const gpa_t addr;
122 	const u32 error_code;
123 	const bool prefetch;
124 
125 	/* Derived from error_code.  */
126 	const bool exec;
127 	const bool write;
128 	const bool present;
129 	const bool rsvd;
130 	const bool user;
131 
132 	/* Derived from mmu and global state.  */
133 	const bool is_tdp;
134 	const bool nx_huge_page_workaround_enabled;
135 
136 	/*
137 	 * Whether a >4KB mapping can be created or is forbidden due to NX
138 	 * hugepages.
139 	 */
140 	bool huge_page_disallowed;
141 
142 	/*
143 	 * Maximum page size that can be created for this fault; input to
144 	 * FNAME(fetch), __direct_map and kvm_tdp_mmu_map.
145 	 */
146 	u8 max_level;
147 
148 	/*
149 	 * Page size that can be created based on the max_level and the
150 	 * page size used by the host mapping.
151 	 */
152 	u8 req_level;
153 
154 	/*
155 	 * Page size that will be created based on the req_level and
156 	 * huge_page_disallowed.
157 	 */
158 	u8 goal_level;
159 
160 	/* Shifted addr, or result of guest page table walk if addr is a gva.  */
161 	gfn_t gfn;
162 
163 	/* The memslot containing gfn. May be NULL. */
164 	struct kvm_memory_slot *slot;
165 
166 	/* Outputs of kvm_faultin_pfn.  */
167 	kvm_pfn_t pfn;
168 	hva_t hva;
169 	bool map_writable;
170 };
171 
172 int kvm_tdp_page_fault(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault);
173 
174 extern int nx_huge_pages;
175 static inline bool is_nx_huge_page_enabled(void)
176 {
177 	return READ_ONCE(nx_huge_pages);
178 }
179 
180 static inline int kvm_mmu_do_page_fault(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
181 					u32 err, bool prefetch)
182 {
183 	struct kvm_page_fault fault = {
184 		.addr = cr2_or_gpa,
185 		.error_code = err,
186 		.exec = err & PFERR_FETCH_MASK,
187 		.write = err & PFERR_WRITE_MASK,
188 		.present = err & PFERR_PRESENT_MASK,
189 		.rsvd = err & PFERR_RSVD_MASK,
190 		.user = err & PFERR_USER_MASK,
191 		.prefetch = prefetch,
192 		.is_tdp = likely(vcpu->arch.mmu->page_fault == kvm_tdp_page_fault),
193 		.nx_huge_page_workaround_enabled = is_nx_huge_page_enabled(),
194 
195 		.max_level = KVM_MAX_HUGEPAGE_LEVEL,
196 		.req_level = PG_LEVEL_4K,
197 		.goal_level = PG_LEVEL_4K,
198 	};
199 #ifdef CONFIG_RETPOLINE
200 	if (fault.is_tdp)
201 		return kvm_tdp_page_fault(vcpu, &fault);
202 #endif
203 	return vcpu->arch.mmu->page_fault(vcpu, &fault);
204 }
205 
206 /*
207  * Check if a given access (described through the I/D, W/R and U/S bits of a
208  * page fault error code pfec) causes a permission fault with the given PTE
209  * access rights (in ACC_* format).
210  *
211  * Return zero if the access does not fault; return the page fault error code
212  * if the access faults.
213  */
214 static inline u8 permission_fault(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
215 				  unsigned pte_access, unsigned pte_pkey,
216 				  unsigned pfec)
217 {
218 	int cpl = static_call(kvm_x86_get_cpl)(vcpu);
219 	unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
220 
221 	/*
222 	 * If CPL < 3, SMAP prevention are disabled if EFLAGS.AC = 1.
223 	 *
224 	 * If CPL = 3, SMAP applies to all supervisor-mode data accesses
225 	 * (these are implicit supervisor accesses) regardless of the value
226 	 * of EFLAGS.AC.
227 	 *
228 	 * This computes (cpl < 3) && (rflags & X86_EFLAGS_AC), leaving
229 	 * the result in X86_EFLAGS_AC. We then insert it in place of
230 	 * the PFERR_RSVD_MASK bit; this bit will always be zero in pfec,
231 	 * but it will be one in index if SMAP checks are being overridden.
232 	 * It is important to keep this branchless.
233 	 */
234 	unsigned long smap = (cpl - 3) & (rflags & X86_EFLAGS_AC);
235 	int index = (pfec >> 1) +
236 		    (smap >> (X86_EFLAGS_AC_BIT - PFERR_RSVD_BIT + 1));
237 	bool fault = (mmu->permissions[index] >> pte_access) & 1;
238 	u32 errcode = PFERR_PRESENT_MASK;
239 
240 	WARN_ON(pfec & (PFERR_PK_MASK | PFERR_RSVD_MASK));
241 	if (unlikely(mmu->pkru_mask)) {
242 		u32 pkru_bits, offset;
243 
244 		/*
245 		* PKRU defines 32 bits, there are 16 domains and 2
246 		* attribute bits per domain in pkru.  pte_pkey is the
247 		* index of the protection domain, so pte_pkey * 2 is
248 		* is the index of the first bit for the domain.
249 		*/
250 		pkru_bits = (vcpu->arch.pkru >> (pte_pkey * 2)) & 3;
251 
252 		/* clear present bit, replace PFEC.RSVD with ACC_USER_MASK. */
253 		offset = (pfec & ~1) +
254 			((pte_access & PT_USER_MASK) << (PFERR_RSVD_BIT - PT_USER_SHIFT));
255 
256 		pkru_bits &= mmu->pkru_mask >> offset;
257 		errcode |= -pkru_bits & PFERR_PK_MASK;
258 		fault |= (pkru_bits != 0);
259 	}
260 
261 	return -(u32)fault & errcode;
262 }
263 
264 void kvm_zap_gfn_range(struct kvm *kvm, gfn_t gfn_start, gfn_t gfn_end);
265 
266 int kvm_arch_write_log_dirty(struct kvm_vcpu *vcpu);
267 
268 int kvm_mmu_post_init_vm(struct kvm *kvm);
269 void kvm_mmu_pre_destroy_vm(struct kvm *kvm);
270 
271 static inline bool kvm_shadow_root_allocated(struct kvm *kvm)
272 {
273 	/*
274 	 * Read shadow_root_allocated before related pointers. Hence, threads
275 	 * reading shadow_root_allocated in any lock context are guaranteed to
276 	 * see the pointers. Pairs with smp_store_release in
277 	 * mmu_first_shadow_root_alloc.
278 	 */
279 	return smp_load_acquire(&kvm->arch.shadow_root_allocated);
280 }
281 
282 #ifdef CONFIG_X86_64
283 static inline bool is_tdp_mmu_enabled(struct kvm *kvm) { return kvm->arch.tdp_mmu_enabled; }
284 #else
285 static inline bool is_tdp_mmu_enabled(struct kvm *kvm) { return false; }
286 #endif
287 
288 static inline bool kvm_memslots_have_rmaps(struct kvm *kvm)
289 {
290 	return !is_tdp_mmu_enabled(kvm) || kvm_shadow_root_allocated(kvm);
291 }
292 
293 static inline gfn_t gfn_to_index(gfn_t gfn, gfn_t base_gfn, int level)
294 {
295 	/* KVM_HPAGE_GFN_SHIFT(PG_LEVEL_4K) must be 0. */
296 	return (gfn >> KVM_HPAGE_GFN_SHIFT(level)) -
297 		(base_gfn >> KVM_HPAGE_GFN_SHIFT(level));
298 }
299 
300 static inline unsigned long
301 __kvm_mmu_slot_lpages(struct kvm_memory_slot *slot, unsigned long npages,
302 		      int level)
303 {
304 	return gfn_to_index(slot->base_gfn + npages - 1,
305 			    slot->base_gfn, level) + 1;
306 }
307 
308 static inline unsigned long
309 kvm_mmu_slot_lpages(struct kvm_memory_slot *slot, int level)
310 {
311 	return __kvm_mmu_slot_lpages(slot, slot->npages, level);
312 }
313 
314 static inline void kvm_update_page_stats(struct kvm *kvm, int level, int count)
315 {
316 	atomic64_add(count, &kvm->stat.pages[level - 1]);
317 }
318 
319 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
320 			   struct x86_exception *exception);
321 
322 static inline gpa_t kvm_translate_gpa(struct kvm_vcpu *vcpu,
323 				      struct kvm_mmu *mmu,
324 				      gpa_t gpa, u32 access,
325 				      struct x86_exception *exception)
326 {
327 	if (mmu != &vcpu->arch.nested_mmu)
328 		return gpa;
329 	return translate_nested_gpa(vcpu, gpa, access, exception);
330 }
331 #endif
332