xref: /openbmc/linux/arch/x86/kvm/mmu/mmu_internal.h (revision 2cf1c348)
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 	/*
35 	 * Note, "link" through "spt" fit in a single 64 byte cache line on
36 	 * 64-bit kernels, keep it that way unless there's a reason not to.
37 	 */
38 	struct list_head link;
39 	struct hlist_node hash_link;
40 
41 	bool tdp_mmu_page;
42 	bool unsync;
43 	u8 mmu_valid_gen;
44 	bool lpage_disallowed; /* Can't be replaced by an equiv large page */
45 
46 	/*
47 	 * The following two entries are used to key the shadow page in the
48 	 * hash table.
49 	 */
50 	union kvm_mmu_page_role role;
51 	gfn_t gfn;
52 
53 	u64 *spt;
54 	/* hold the gfn of each spte inside spt */
55 	gfn_t *gfns;
56 	/* Currently serving as active root */
57 	union {
58 		int root_count;
59 		refcount_t tdp_mmu_root_count;
60 	};
61 	unsigned int unsync_children;
62 	struct kvm_rmap_head parent_ptes; /* rmap pointers to parent sptes */
63 	DECLARE_BITMAP(unsync_child_bitmap, 512);
64 
65 	struct list_head lpage_disallowed_link;
66 #ifdef CONFIG_X86_32
67 	/*
68 	 * Used out of the mmu-lock to avoid reading spte values while an
69 	 * update is in progress; see the comments in __get_spte_lockless().
70 	 */
71 	int clear_spte_count;
72 #endif
73 
74 	/* Number of writes since the last time traversal visited this page.  */
75 	atomic_t write_flooding_count;
76 
77 #ifdef CONFIG_X86_64
78 	/* Used for freeing the page asynchronously if it is a TDP MMU page. */
79 	struct rcu_head rcu_head;
80 #endif
81 };
82 
83 extern struct kmem_cache *mmu_page_header_cache;
84 
85 static inline struct kvm_mmu_page *to_shadow_page(hpa_t shadow_page)
86 {
87 	struct page *page = pfn_to_page(shadow_page >> PAGE_SHIFT);
88 
89 	return (struct kvm_mmu_page *)page_private(page);
90 }
91 
92 static inline struct kvm_mmu_page *sptep_to_sp(u64 *sptep)
93 {
94 	return to_shadow_page(__pa(sptep));
95 }
96 
97 static inline int kvm_mmu_role_as_id(union kvm_mmu_page_role role)
98 {
99 	return role.smm ? 1 : 0;
100 }
101 
102 static inline int kvm_mmu_page_as_id(struct kvm_mmu_page *sp)
103 {
104 	return kvm_mmu_role_as_id(sp->role);
105 }
106 
107 static inline bool kvm_mmu_page_ad_need_write_protect(struct kvm_mmu_page *sp)
108 {
109 	/*
110 	 * When using the EPT page-modification log, the GPAs in the CPU dirty
111 	 * log would come from L2 rather than L1.  Therefore, we need to rely
112 	 * on write protection to record dirty pages, which bypasses PML, since
113 	 * writes now result in a vmexit.  Note, the check on CPU dirty logging
114 	 * being enabled is mandatory as the bits used to denote WP-only SPTEs
115 	 * are reserved for PAE paging (32-bit KVM).
116 	 */
117 	return kvm_x86_ops.cpu_dirty_log_size && sp->role.guest_mode;
118 }
119 
120 int mmu_try_to_unsync_pages(struct kvm *kvm, const struct kvm_memory_slot *slot,
121 			    gfn_t gfn, bool can_unsync, bool prefetch);
122 
123 void kvm_mmu_gfn_disallow_lpage(const struct kvm_memory_slot *slot, gfn_t gfn);
124 void kvm_mmu_gfn_allow_lpage(const struct kvm_memory_slot *slot, gfn_t gfn);
125 bool kvm_mmu_slot_gfn_write_protect(struct kvm *kvm,
126 				    struct kvm_memory_slot *slot, u64 gfn,
127 				    int min_level);
128 void kvm_flush_remote_tlbs_with_address(struct kvm *kvm,
129 					u64 start_gfn, u64 pages);
130 unsigned int pte_list_count(struct kvm_rmap_head *rmap_head);
131 
132 /*
133  * Return values of handle_mmio_page_fault, mmu.page_fault, and fast_page_fault().
134  *
135  * RET_PF_RETRY: let CPU fault again on the address.
136  * RET_PF_EMULATE: mmio page fault, emulate the instruction directly.
137  * RET_PF_INVALID: the spte is invalid, let the real page fault path update it.
138  * RET_PF_FIXED: The faulting entry has been fixed.
139  * RET_PF_SPURIOUS: The faulting entry was already fixed, e.g. by another vCPU.
140  *
141  * Any names added to this enum should be exported to userspace for use in
142  * tracepoints via TRACE_DEFINE_ENUM() in mmutrace.h
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 int kvm_mmu_max_mapping_level(struct kvm *kvm,
153 			      const struct kvm_memory_slot *slot, gfn_t gfn,
154 			      kvm_pfn_t pfn, int max_level);
155 void kvm_mmu_hugepage_adjust(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault);
156 void disallowed_hugepage_adjust(struct kvm_page_fault *fault, u64 spte, int cur_level);
157 
158 void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc);
159 
160 void account_huge_nx_page(struct kvm *kvm, struct kvm_mmu_page *sp);
161 void unaccount_huge_nx_page(struct kvm *kvm, struct kvm_mmu_page *sp);
162 
163 #endif /* __KVM_X86_MMU_INTERNAL_H */
164