xref: /openbmc/linux/arch/x86/kvm/mmu.h (revision d2168146)
1 #ifndef __KVM_X86_MMU_H
2 #define __KVM_X86_MMU_H
3 
4 #include <linux/kvm_host.h>
5 #include "kvm_cache_regs.h"
6 
7 #define PT64_PT_BITS 9
8 #define PT64_ENT_PER_PAGE (1 << PT64_PT_BITS)
9 #define PT32_PT_BITS 10
10 #define PT32_ENT_PER_PAGE (1 << PT32_PT_BITS)
11 
12 #define PT_WRITABLE_SHIFT 1
13 
14 #define PT_PRESENT_MASK (1ULL << 0)
15 #define PT_WRITABLE_MASK (1ULL << PT_WRITABLE_SHIFT)
16 #define PT_USER_MASK (1ULL << 2)
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 PT32_DIR_PSE36_SIZE 4
35 #define PT32_DIR_PSE36_SHIFT 13
36 #define PT32_DIR_PSE36_MASK \
37 	(((1ULL << PT32_DIR_PSE36_SIZE) - 1) << PT32_DIR_PSE36_SHIFT)
38 
39 #define PT64_ROOT_LEVEL 4
40 #define PT32_ROOT_LEVEL 2
41 #define PT32E_ROOT_LEVEL 3
42 
43 #define PT_PDPE_LEVEL 3
44 #define PT_DIRECTORY_LEVEL 2
45 #define PT_PAGE_TABLE_LEVEL 1
46 
47 #define PFERR_PRESENT_BIT 0
48 #define PFERR_WRITE_BIT 1
49 #define PFERR_USER_BIT 2
50 #define PFERR_RSVD_BIT 3
51 #define PFERR_FETCH_BIT 4
52 
53 #define PFERR_PRESENT_MASK (1U << PFERR_PRESENT_BIT)
54 #define PFERR_WRITE_MASK (1U << PFERR_WRITE_BIT)
55 #define PFERR_USER_MASK (1U << PFERR_USER_BIT)
56 #define PFERR_RSVD_MASK (1U << PFERR_RSVD_BIT)
57 #define PFERR_FETCH_MASK (1U << PFERR_FETCH_BIT)
58 
59 int kvm_mmu_get_spte_hierarchy(struct kvm_vcpu *vcpu, u64 addr, u64 sptes[4]);
60 void kvm_mmu_set_mmio_spte_mask(u64 mmio_mask);
61 
62 /*
63  * Return values of handle_mmio_page_fault_common:
64  * RET_MMIO_PF_EMULATE: it is a real mmio page fault, emulate the instruction
65  *			directly.
66  * RET_MMIO_PF_INVALID: invalid spte is detected then let the real page
67  *			fault path update the mmio spte.
68  * RET_MMIO_PF_RETRY: let CPU fault again on the address.
69  * RET_MMIO_PF_BUG: bug is detected.
70  */
71 enum {
72 	RET_MMIO_PF_EMULATE = 1,
73 	RET_MMIO_PF_INVALID = 2,
74 	RET_MMIO_PF_RETRY = 0,
75 	RET_MMIO_PF_BUG = -1
76 };
77 
78 int handle_mmio_page_fault_common(struct kvm_vcpu *vcpu, u64 addr, bool direct);
79 void kvm_init_shadow_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *context);
80 void kvm_init_shadow_ept_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *context,
81 		bool execonly);
82 void update_permission_bitmask(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
83 		bool ept);
84 
85 static inline unsigned int kvm_mmu_available_pages(struct kvm *kvm)
86 {
87 	if (kvm->arch.n_max_mmu_pages > kvm->arch.n_used_mmu_pages)
88 		return kvm->arch.n_max_mmu_pages -
89 			kvm->arch.n_used_mmu_pages;
90 
91 	return 0;
92 }
93 
94 static inline int kvm_mmu_reload(struct kvm_vcpu *vcpu)
95 {
96 	if (likely(vcpu->arch.mmu.root_hpa != INVALID_PAGE))
97 		return 0;
98 
99 	return kvm_mmu_load(vcpu);
100 }
101 
102 static inline int is_present_gpte(unsigned long pte)
103 {
104 	return pte & PT_PRESENT_MASK;
105 }
106 
107 /*
108  * Currently, we have two sorts of write-protection, a) the first one
109  * write-protects guest page to sync the guest modification, b) another one is
110  * used to sync dirty bitmap when we do KVM_GET_DIRTY_LOG. The differences
111  * between these two sorts are:
112  * 1) the first case clears SPTE_MMU_WRITEABLE bit.
113  * 2) the first case requires flushing tlb immediately avoiding corrupting
114  *    shadow page table between all vcpus so it should be in the protection of
115  *    mmu-lock. And the another case does not need to flush tlb until returning
116  *    the dirty bitmap to userspace since it only write-protects the page
117  *    logged in the bitmap, that means the page in the dirty bitmap is not
118  *    missed, so it can flush tlb out of mmu-lock.
119  *
120  * So, there is the problem: the first case can meet the corrupted tlb caused
121  * by another case which write-protects pages but without flush tlb
122  * immediately. In order to making the first case be aware this problem we let
123  * it flush tlb if we try to write-protect a spte whose SPTE_MMU_WRITEABLE bit
124  * is set, it works since another case never touches SPTE_MMU_WRITEABLE bit.
125  *
126  * Anyway, whenever a spte is updated (only permission and status bits are
127  * changed) we need to check whether the spte with SPTE_MMU_WRITEABLE becomes
128  * readonly, if that happens, we need to flush tlb. Fortunately,
129  * mmu_spte_update() has already handled it perfectly.
130  *
131  * The rules to use SPTE_MMU_WRITEABLE and PT_WRITABLE_MASK:
132  * - if we want to see if it has writable tlb entry or if the spte can be
133  *   writable on the mmu mapping, check SPTE_MMU_WRITEABLE, this is the most
134  *   case, otherwise
135  * - if we fix page fault on the spte or do write-protection by dirty logging,
136  *   check PT_WRITABLE_MASK.
137  *
138  * TODO: introduce APIs to split these two cases.
139  */
140 static inline int is_writable_pte(unsigned long pte)
141 {
142 	return pte & PT_WRITABLE_MASK;
143 }
144 
145 static inline bool is_write_protection(struct kvm_vcpu *vcpu)
146 {
147 	return kvm_read_cr0_bits(vcpu, X86_CR0_WP);
148 }
149 
150 /*
151  * Will a fault with a given page-fault error code (pfec) cause a permission
152  * fault with the given access (in ACC_* format)?
153  */
154 static inline bool permission_fault(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
155 				    unsigned pte_access, unsigned pfec)
156 {
157 	int cpl = kvm_x86_ops->get_cpl(vcpu);
158 	unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
159 
160 	/*
161 	 * If CPL < 3, SMAP prevention are disabled if EFLAGS.AC = 1.
162 	 *
163 	 * If CPL = 3, SMAP applies to all supervisor-mode data accesses
164 	 * (these are implicit supervisor accesses) regardless of the value
165 	 * of EFLAGS.AC.
166 	 *
167 	 * This computes (cpl < 3) && (rflags & X86_EFLAGS_AC), leaving
168 	 * the result in X86_EFLAGS_AC. We then insert it in place of
169 	 * the PFERR_RSVD_MASK bit; this bit will always be zero in pfec,
170 	 * but it will be one in index if SMAP checks are being overridden.
171 	 * It is important to keep this branchless.
172 	 */
173 	unsigned long smap = (cpl - 3) & (rflags & X86_EFLAGS_AC);
174 	int index = (pfec >> 1) +
175 		    (smap >> (X86_EFLAGS_AC_BIT - PFERR_RSVD_BIT + 1));
176 
177 	return (mmu->permissions[index] >> pte_access) & 1;
178 }
179 
180 void kvm_mmu_invalidate_zap_all_pages(struct kvm *kvm);
181 #endif
182