xref: /linux/arch/x86/kvm/mmu.h (revision 3839a7460721b87501134697b7b90c45dcc7825d)
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 
8 #define PT64_PT_BITS 9
9 #define PT64_ENT_PER_PAGE (1 << PT64_PT_BITS)
10 #define PT32_PT_BITS 10
11 #define PT32_ENT_PER_PAGE (1 << PT32_PT_BITS)
12 
13 #define PT_WRITABLE_SHIFT 1
14 #define PT_USER_SHIFT 2
15 
16 #define PT_PRESENT_MASK (1ULL << 0)
17 #define PT_WRITABLE_MASK (1ULL << PT_WRITABLE_SHIFT)
18 #define PT_USER_MASK (1ULL << PT_USER_SHIFT)
19 #define PT_PWT_MASK (1ULL << 3)
20 #define PT_PCD_MASK (1ULL << 4)
21 #define PT_ACCESSED_SHIFT 5
22 #define PT_ACCESSED_MASK (1ULL << PT_ACCESSED_SHIFT)
23 #define PT_DIRTY_SHIFT 6
24 #define PT_DIRTY_MASK (1ULL << PT_DIRTY_SHIFT)
25 #define PT_PAGE_SIZE_SHIFT 7
26 #define PT_PAGE_SIZE_MASK (1ULL << PT_PAGE_SIZE_SHIFT)
27 #define PT_PAT_MASK (1ULL << 7)
28 #define PT_GLOBAL_MASK (1ULL << 8)
29 #define PT64_NX_SHIFT 63
30 #define PT64_NX_MASK (1ULL << PT64_NX_SHIFT)
31 
32 #define PT_PAT_SHIFT 7
33 #define PT_DIR_PAT_SHIFT 12
34 #define PT_DIR_PAT_MASK (1ULL << PT_DIR_PAT_SHIFT)
35 
36 #define PT32_DIR_PSE36_SIZE 4
37 #define PT32_DIR_PSE36_SHIFT 13
38 #define PT32_DIR_PSE36_MASK \
39 	(((1ULL << PT32_DIR_PSE36_SIZE) - 1) << PT32_DIR_PSE36_SHIFT)
40 
41 #define PT64_ROOT_5LEVEL 5
42 #define PT64_ROOT_4LEVEL 4
43 #define PT32_ROOT_LEVEL 2
44 #define PT32E_ROOT_LEVEL 3
45 
46 static inline u64 rsvd_bits(int s, int e)
47 {
48 	if (e < s)
49 		return 0;
50 
51 	return ((1ULL << (e - s + 1)) - 1) << s;
52 }
53 
54 void kvm_mmu_set_mmio_spte_mask(u64 mmio_value, u64 access_mask);
55 
56 void
57 reset_shadow_zero_bits_mask(struct kvm_vcpu *vcpu, struct kvm_mmu *context);
58 
59 void kvm_init_mmu(struct kvm_vcpu *vcpu, bool reset_roots);
60 void kvm_init_shadow_mmu(struct kvm_vcpu *vcpu, u32 cr0, u32 cr4, u32 efer);
61 void kvm_init_shadow_ept_mmu(struct kvm_vcpu *vcpu, bool execonly,
62 			     bool accessed_dirty, gpa_t new_eptp);
63 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu);
64 int kvm_handle_page_fault(struct kvm_vcpu *vcpu, u64 error_code,
65 				u64 fault_address, char *insn, int insn_len);
66 
67 static inline unsigned long kvm_mmu_available_pages(struct kvm *kvm)
68 {
69 	if (kvm->arch.n_max_mmu_pages > kvm->arch.n_used_mmu_pages)
70 		return kvm->arch.n_max_mmu_pages -
71 			kvm->arch.n_used_mmu_pages;
72 
73 	return 0;
74 }
75 
76 static inline int kvm_mmu_reload(struct kvm_vcpu *vcpu)
77 {
78 	if (likely(vcpu->arch.mmu->root_hpa != INVALID_PAGE))
79 		return 0;
80 
81 	return kvm_mmu_load(vcpu);
82 }
83 
84 static inline unsigned long kvm_get_pcid(struct kvm_vcpu *vcpu, gpa_t cr3)
85 {
86 	BUILD_BUG_ON((X86_CR3_PCID_MASK & PAGE_MASK) != 0);
87 
88 	return kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)
89 	       ? cr3 & X86_CR3_PCID_MASK
90 	       : 0;
91 }
92 
93 static inline unsigned long kvm_get_active_pcid(struct kvm_vcpu *vcpu)
94 {
95 	return kvm_get_pcid(vcpu, kvm_read_cr3(vcpu));
96 }
97 
98 static inline void kvm_mmu_load_pgd(struct kvm_vcpu *vcpu)
99 {
100 	if (VALID_PAGE(vcpu->arch.mmu->root_hpa))
101 		kvm_x86_ops.load_mmu_pgd(vcpu, vcpu->arch.mmu->root_hpa |
102 					       kvm_get_active_pcid(vcpu));
103 }
104 
105 int kvm_tdp_page_fault(struct kvm_vcpu *vcpu, gpa_t gpa, u32 error_code,
106 		       bool prefault);
107 
108 static inline int kvm_mmu_do_page_fault(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
109 					u32 err, bool prefault)
110 {
111 #ifdef CONFIG_RETPOLINE
112 	if (likely(vcpu->arch.mmu->page_fault == kvm_tdp_page_fault))
113 		return kvm_tdp_page_fault(vcpu, cr2_or_gpa, err, prefault);
114 #endif
115 	return vcpu->arch.mmu->page_fault(vcpu, cr2_or_gpa, err, prefault);
116 }
117 
118 /*
119  * Currently, we have two sorts of write-protection, a) the first one
120  * write-protects guest page to sync the guest modification, b) another one is
121  * used to sync dirty bitmap when we do KVM_GET_DIRTY_LOG. The differences
122  * between these two sorts are:
123  * 1) the first case clears SPTE_MMU_WRITEABLE bit.
124  * 2) the first case requires flushing tlb immediately avoiding corrupting
125  *    shadow page table between all vcpus so it should be in the protection of
126  *    mmu-lock. And the another case does not need to flush tlb until returning
127  *    the dirty bitmap to userspace since it only write-protects the page
128  *    logged in the bitmap, that means the page in the dirty bitmap is not
129  *    missed, so it can flush tlb out of mmu-lock.
130  *
131  * So, there is the problem: the first case can meet the corrupted tlb caused
132  * by another case which write-protects pages but without flush tlb
133  * immediately. In order to making the first case be aware this problem we let
134  * it flush tlb if we try to write-protect a spte whose SPTE_MMU_WRITEABLE bit
135  * is set, it works since another case never touches SPTE_MMU_WRITEABLE bit.
136  *
137  * Anyway, whenever a spte is updated (only permission and status bits are
138  * changed) we need to check whether the spte with SPTE_MMU_WRITEABLE becomes
139  * readonly, if that happens, we need to flush tlb. Fortunately,
140  * mmu_spte_update() has already handled it perfectly.
141  *
142  * The rules to use SPTE_MMU_WRITEABLE and PT_WRITABLE_MASK:
143  * - if we want to see if it has writable tlb entry or if the spte can be
144  *   writable on the mmu mapping, check SPTE_MMU_WRITEABLE, this is the most
145  *   case, otherwise
146  * - if we fix page fault on the spte or do write-protection by dirty logging,
147  *   check PT_WRITABLE_MASK.
148  *
149  * TODO: introduce APIs to split these two cases.
150  */
151 static inline int is_writable_pte(unsigned long pte)
152 {
153 	return pte & PT_WRITABLE_MASK;
154 }
155 
156 static inline bool is_write_protection(struct kvm_vcpu *vcpu)
157 {
158 	return kvm_read_cr0_bits(vcpu, X86_CR0_WP);
159 }
160 
161 /*
162  * Check if a given access (described through the I/D, W/R and U/S bits of a
163  * page fault error code pfec) causes a permission fault with the given PTE
164  * access rights (in ACC_* format).
165  *
166  * Return zero if the access does not fault; return the page fault error code
167  * if the access faults.
168  */
169 static inline u8 permission_fault(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
170 				  unsigned pte_access, unsigned pte_pkey,
171 				  unsigned pfec)
172 {
173 	int cpl = kvm_x86_ops.get_cpl(vcpu);
174 	unsigned long rflags = kvm_x86_ops.get_rflags(vcpu);
175 
176 	/*
177 	 * If CPL < 3, SMAP prevention are disabled if EFLAGS.AC = 1.
178 	 *
179 	 * If CPL = 3, SMAP applies to all supervisor-mode data accesses
180 	 * (these are implicit supervisor accesses) regardless of the value
181 	 * of EFLAGS.AC.
182 	 *
183 	 * This computes (cpl < 3) && (rflags & X86_EFLAGS_AC), leaving
184 	 * the result in X86_EFLAGS_AC. We then insert it in place of
185 	 * the PFERR_RSVD_MASK bit; this bit will always be zero in pfec,
186 	 * but it will be one in index if SMAP checks are being overridden.
187 	 * It is important to keep this branchless.
188 	 */
189 	unsigned long smap = (cpl - 3) & (rflags & X86_EFLAGS_AC);
190 	int index = (pfec >> 1) +
191 		    (smap >> (X86_EFLAGS_AC_BIT - PFERR_RSVD_BIT + 1));
192 	bool fault = (mmu->permissions[index] >> pte_access) & 1;
193 	u32 errcode = PFERR_PRESENT_MASK;
194 
195 	WARN_ON(pfec & (PFERR_PK_MASK | PFERR_RSVD_MASK));
196 	if (unlikely(mmu->pkru_mask)) {
197 		u32 pkru_bits, offset;
198 
199 		/*
200 		* PKRU defines 32 bits, there are 16 domains and 2
201 		* attribute bits per domain in pkru.  pte_pkey is the
202 		* index of the protection domain, so pte_pkey * 2 is
203 		* is the index of the first bit for the domain.
204 		*/
205 		pkru_bits = (vcpu->arch.pkru >> (pte_pkey * 2)) & 3;
206 
207 		/* clear present bit, replace PFEC.RSVD with ACC_USER_MASK. */
208 		offset = (pfec & ~1) +
209 			((pte_access & PT_USER_MASK) << (PFERR_RSVD_BIT - PT_USER_SHIFT));
210 
211 		pkru_bits &= mmu->pkru_mask >> offset;
212 		errcode |= -pkru_bits & PFERR_PK_MASK;
213 		fault |= (pkru_bits != 0);
214 	}
215 
216 	return -(u32)fault & errcode;
217 }
218 
219 void kvm_zap_gfn_range(struct kvm *kvm, gfn_t gfn_start, gfn_t gfn_end);
220 
221 void kvm_mmu_gfn_disallow_lpage(struct kvm_memory_slot *slot, gfn_t gfn);
222 void kvm_mmu_gfn_allow_lpage(struct kvm_memory_slot *slot, gfn_t gfn);
223 bool kvm_mmu_slot_gfn_write_protect(struct kvm *kvm,
224 				    struct kvm_memory_slot *slot, u64 gfn);
225 int kvm_arch_write_log_dirty(struct kvm_vcpu *vcpu);
226 
227 int kvm_mmu_post_init_vm(struct kvm *kvm);
228 void kvm_mmu_pre_destroy_vm(struct kvm *kvm);
229 
230 #endif
231