xref: /openbmc/linux/arch/x86/kvm/mmu/mmu_internal.h (revision aa0dc6a7)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __KVM_X86_MMU_INTERNAL_H
3 #define __KVM_X86_MMU_INTERNAL_H
4 
5 #include <linux/types.h>
6 #include <linux/kvm_host.h>
7 #include <asm/kvm_host.h>
8 
9 #undef MMU_DEBUG
10 
11 #ifdef MMU_DEBUG
12 extern bool dbg;
13 
14 #define pgprintk(x...) do { if (dbg) printk(x); } while (0)
15 #define rmap_printk(fmt, args...) do { if (dbg) printk("%s: " fmt, __func__, ## args); } while (0)
16 #define MMU_WARN_ON(x) WARN_ON(x)
17 #else
18 #define pgprintk(x...) do { } while (0)
19 #define rmap_printk(x...) do { } while (0)
20 #define MMU_WARN_ON(x) do { } while (0)
21 #endif
22 
23 /*
24  * Unlike regular MMU roots, PAE "roots", a.k.a. PDPTEs/PDPTRs, have a PRESENT
25  * bit, and thus are guaranteed to be non-zero when valid.  And, when a guest
26  * PDPTR is !PRESENT, its corresponding PAE root cannot be set to INVALID_PAGE,
27  * as the CPU would treat that as PRESENT PDPTR with reserved bits set.  Use
28  * '0' instead of INVALID_PAGE to indicate an invalid PAE root.
29  */
30 #define INVALID_PAE_ROOT	0
31 #define IS_VALID_PAE_ROOT(x)	(!!(x))
32 
33 struct kvm_mmu_page {
34 	struct list_head link;
35 	struct hlist_node hash_link;
36 	struct list_head lpage_disallowed_link;
37 
38 	bool unsync;
39 	u8 mmu_valid_gen;
40 	bool mmio_cached;
41 	bool lpage_disallowed; /* Can't be replaced by an equiv large page */
42 
43 	/*
44 	 * The following two entries are used to key the shadow page in the
45 	 * hash table.
46 	 */
47 	union kvm_mmu_page_role role;
48 	gfn_t gfn;
49 
50 	u64 *spt;
51 	/* hold the gfn of each spte inside spt */
52 	gfn_t *gfns;
53 	/* Currently serving as active root */
54 	union {
55 		int root_count;
56 		refcount_t tdp_mmu_root_count;
57 	};
58 	unsigned int unsync_children;
59 	struct kvm_rmap_head parent_ptes; /* rmap pointers to parent sptes */
60 	DECLARE_BITMAP(unsync_child_bitmap, 512);
61 
62 #ifdef CONFIG_X86_32
63 	/*
64 	 * Used out of the mmu-lock to avoid reading spte values while an
65 	 * update is in progress; see the comments in __get_spte_lockless().
66 	 */
67 	int clear_spte_count;
68 #endif
69 
70 	/* Number of writes since the last time traversal visited this page.  */
71 	atomic_t write_flooding_count;
72 
73 #ifdef CONFIG_X86_64
74 	bool tdp_mmu_page;
75 
76 	/* Used for freeing the page asynchronously if it is a TDP MMU page. */
77 	struct rcu_head rcu_head;
78 #endif
79 };
80 
81 extern struct kmem_cache *mmu_page_header_cache;
82 
83 static inline struct kvm_mmu_page *to_shadow_page(hpa_t shadow_page)
84 {
85 	struct page *page = pfn_to_page(shadow_page >> PAGE_SHIFT);
86 
87 	return (struct kvm_mmu_page *)page_private(page);
88 }
89 
90 static inline struct kvm_mmu_page *sptep_to_sp(u64 *sptep)
91 {
92 	return to_shadow_page(__pa(sptep));
93 }
94 
95 static inline int kvm_mmu_role_as_id(union kvm_mmu_page_role role)
96 {
97 	return role.smm ? 1 : 0;
98 }
99 
100 static inline int kvm_mmu_page_as_id(struct kvm_mmu_page *sp)
101 {
102 	return kvm_mmu_role_as_id(sp->role);
103 }
104 
105 static inline bool kvm_vcpu_ad_need_write_protect(struct kvm_vcpu *vcpu)
106 {
107 	/*
108 	 * When using the EPT page-modification log, the GPAs in the CPU dirty
109 	 * log would come from L2 rather than L1.  Therefore, we need to rely
110 	 * on write protection to record dirty pages, which bypasses PML, since
111 	 * writes now result in a vmexit.  Note, the check on CPU dirty logging
112 	 * being enabled is mandatory as the bits used to denote WP-only SPTEs
113 	 * are reserved for NPT w/ PAE (32-bit KVM).
114 	 */
115 	return vcpu->arch.mmu == &vcpu->arch.guest_mmu &&
116 	       kvm_x86_ops.cpu_dirty_log_size;
117 }
118 
119 extern int nx_huge_pages;
120 static inline bool is_nx_huge_page_enabled(void)
121 {
122 	return READ_ONCE(nx_huge_pages);
123 }
124 
125 int mmu_try_to_unsync_pages(struct kvm_vcpu *vcpu, gfn_t gfn, bool can_unsync);
126 
127 void kvm_mmu_gfn_disallow_lpage(struct kvm_memory_slot *slot, gfn_t gfn);
128 void kvm_mmu_gfn_allow_lpage(struct kvm_memory_slot *slot, gfn_t gfn);
129 bool kvm_mmu_slot_gfn_write_protect(struct kvm *kvm,
130 				    struct kvm_memory_slot *slot, u64 gfn,
131 				    int min_level);
132 void kvm_flush_remote_tlbs_with_address(struct kvm *kvm,
133 					u64 start_gfn, u64 pages);
134 
135 /*
136  * Return values of handle_mmio_page_fault, mmu.page_fault, and fast_page_fault().
137  *
138  * RET_PF_RETRY: let CPU fault again on the address.
139  * RET_PF_EMULATE: mmio page fault, emulate the instruction directly.
140  * RET_PF_INVALID: the spte is invalid, let the real page fault path update it.
141  * RET_PF_FIXED: The faulting entry has been fixed.
142  * RET_PF_SPURIOUS: The faulting entry was already fixed, e.g. by another vCPU.
143  */
144 enum {
145 	RET_PF_RETRY = 0,
146 	RET_PF_EMULATE,
147 	RET_PF_INVALID,
148 	RET_PF_FIXED,
149 	RET_PF_SPURIOUS,
150 };
151 
152 /* Bits which may be returned by set_spte() */
153 #define SET_SPTE_WRITE_PROTECTED_PT	BIT(0)
154 #define SET_SPTE_NEED_REMOTE_TLB_FLUSH	BIT(1)
155 #define SET_SPTE_SPURIOUS		BIT(2)
156 
157 int kvm_mmu_max_mapping_level(struct kvm *kvm,
158 			      const struct kvm_memory_slot *slot, gfn_t gfn,
159 			      kvm_pfn_t pfn, int max_level);
160 int kvm_mmu_hugepage_adjust(struct kvm_vcpu *vcpu, gfn_t gfn,
161 			    int max_level, kvm_pfn_t *pfnp,
162 			    bool huge_page_disallowed, int *req_level);
163 void disallowed_hugepage_adjust(u64 spte, gfn_t gfn, int cur_level,
164 				kvm_pfn_t *pfnp, int *goal_levelp);
165 
166 void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc);
167 
168 void account_huge_nx_page(struct kvm *kvm, struct kvm_mmu_page *sp);
169 void unaccount_huge_nx_page(struct kvm *kvm, struct kvm_mmu_page *sp);
170 
171 #endif /* __KVM_X86_MMU_INTERNAL_H */
172