1*4882a593Smuzhiyun /* SPDX-License-Identifier: GPL-2.0 */
2*4882a593Smuzhiyun #ifndef __KVM_X86_MMU_INTERNAL_H
3*4882a593Smuzhiyun #define __KVM_X86_MMU_INTERNAL_H
4*4882a593Smuzhiyun
5*4882a593Smuzhiyun #include <linux/types.h>
6*4882a593Smuzhiyun #include <linux/kvm_host.h>
7*4882a593Smuzhiyun #include <asm/kvm_host.h>
8*4882a593Smuzhiyun
9*4882a593Smuzhiyun #undef MMU_DEBUG
10*4882a593Smuzhiyun
11*4882a593Smuzhiyun #ifdef MMU_DEBUG
12*4882a593Smuzhiyun extern bool dbg;
13*4882a593Smuzhiyun
14*4882a593Smuzhiyun #define pgprintk(x...) do { if (dbg) printk(x); } while (0)
15*4882a593Smuzhiyun #define rmap_printk(x...) do { if (dbg) printk(x); } while (0)
16*4882a593Smuzhiyun #define MMU_WARN_ON(x) WARN_ON(x)
17*4882a593Smuzhiyun #else
18*4882a593Smuzhiyun #define pgprintk(x...) do { } while (0)
19*4882a593Smuzhiyun #define rmap_printk(x...) do { } while (0)
20*4882a593Smuzhiyun #define MMU_WARN_ON(x) do { } while (0)
21*4882a593Smuzhiyun #endif
22*4882a593Smuzhiyun
23*4882a593Smuzhiyun struct kvm_mmu_page {
24*4882a593Smuzhiyun struct list_head link;
25*4882a593Smuzhiyun struct hlist_node hash_link;
26*4882a593Smuzhiyun struct list_head lpage_disallowed_link;
27*4882a593Smuzhiyun
28*4882a593Smuzhiyun bool unsync;
29*4882a593Smuzhiyun u8 mmu_valid_gen;
30*4882a593Smuzhiyun bool mmio_cached;
31*4882a593Smuzhiyun bool lpage_disallowed; /* Can't be replaced by an equiv large page */
32*4882a593Smuzhiyun
33*4882a593Smuzhiyun /*
34*4882a593Smuzhiyun * The following two entries are used to key the shadow page in the
35*4882a593Smuzhiyun * hash table.
36*4882a593Smuzhiyun */
37*4882a593Smuzhiyun union kvm_mmu_page_role role;
38*4882a593Smuzhiyun gfn_t gfn;
39*4882a593Smuzhiyun
40*4882a593Smuzhiyun u64 *spt;
41*4882a593Smuzhiyun /* hold the gfn of each spte inside spt */
42*4882a593Smuzhiyun gfn_t *gfns;
43*4882a593Smuzhiyun int root_count; /* Currently serving as active root */
44*4882a593Smuzhiyun unsigned int unsync_children;
45*4882a593Smuzhiyun struct kvm_rmap_head parent_ptes; /* rmap pointers to parent sptes */
46*4882a593Smuzhiyun DECLARE_BITMAP(unsync_child_bitmap, 512);
47*4882a593Smuzhiyun
48*4882a593Smuzhiyun #ifdef CONFIG_X86_32
49*4882a593Smuzhiyun /*
50*4882a593Smuzhiyun * Used out of the mmu-lock to avoid reading spte values while an
51*4882a593Smuzhiyun * update is in progress; see the comments in __get_spte_lockless().
52*4882a593Smuzhiyun */
53*4882a593Smuzhiyun int clear_spte_count;
54*4882a593Smuzhiyun #endif
55*4882a593Smuzhiyun
56*4882a593Smuzhiyun /* Number of writes since the last time traversal visited this page. */
57*4882a593Smuzhiyun atomic_t write_flooding_count;
58*4882a593Smuzhiyun
59*4882a593Smuzhiyun bool tdp_mmu_page;
60*4882a593Smuzhiyun };
61*4882a593Smuzhiyun
62*4882a593Smuzhiyun extern struct kmem_cache *mmu_page_header_cache;
63*4882a593Smuzhiyun
to_shadow_page(hpa_t shadow_page)64*4882a593Smuzhiyun static inline struct kvm_mmu_page *to_shadow_page(hpa_t shadow_page)
65*4882a593Smuzhiyun {
66*4882a593Smuzhiyun struct page *page = pfn_to_page(shadow_page >> PAGE_SHIFT);
67*4882a593Smuzhiyun
68*4882a593Smuzhiyun return (struct kvm_mmu_page *)page_private(page);
69*4882a593Smuzhiyun }
70*4882a593Smuzhiyun
sptep_to_sp(u64 * sptep)71*4882a593Smuzhiyun static inline struct kvm_mmu_page *sptep_to_sp(u64 *sptep)
72*4882a593Smuzhiyun {
73*4882a593Smuzhiyun return to_shadow_page(__pa(sptep));
74*4882a593Smuzhiyun }
75*4882a593Smuzhiyun
kvm_vcpu_ad_need_write_protect(struct kvm_vcpu * vcpu)76*4882a593Smuzhiyun static inline bool kvm_vcpu_ad_need_write_protect(struct kvm_vcpu *vcpu)
77*4882a593Smuzhiyun {
78*4882a593Smuzhiyun /*
79*4882a593Smuzhiyun * When using the EPT page-modification log, the GPAs in the log
80*4882a593Smuzhiyun * would come from L2 rather than L1. Therefore, we need to rely
81*4882a593Smuzhiyun * on write protection to record dirty pages. This also bypasses
82*4882a593Smuzhiyun * PML, since writes now result in a vmexit.
83*4882a593Smuzhiyun */
84*4882a593Smuzhiyun return vcpu->arch.mmu == &vcpu->arch.guest_mmu;
85*4882a593Smuzhiyun }
86*4882a593Smuzhiyun
87*4882a593Smuzhiyun bool is_nx_huge_page_enabled(void);
88*4882a593Smuzhiyun bool mmu_need_write_protect(struct kvm_vcpu *vcpu, gfn_t gfn,
89*4882a593Smuzhiyun bool can_unsync);
90*4882a593Smuzhiyun
91*4882a593Smuzhiyun void kvm_mmu_gfn_disallow_lpage(struct kvm_memory_slot *slot, gfn_t gfn);
92*4882a593Smuzhiyun void kvm_mmu_gfn_allow_lpage(struct kvm_memory_slot *slot, gfn_t gfn);
93*4882a593Smuzhiyun bool kvm_mmu_slot_gfn_write_protect(struct kvm *kvm,
94*4882a593Smuzhiyun struct kvm_memory_slot *slot, u64 gfn);
95*4882a593Smuzhiyun void kvm_flush_remote_tlbs_with_address(struct kvm *kvm,
96*4882a593Smuzhiyun u64 start_gfn, u64 pages);
97*4882a593Smuzhiyun
kvm_mmu_get_root(struct kvm * kvm,struct kvm_mmu_page * sp)98*4882a593Smuzhiyun static inline void kvm_mmu_get_root(struct kvm *kvm, struct kvm_mmu_page *sp)
99*4882a593Smuzhiyun {
100*4882a593Smuzhiyun BUG_ON(!sp->root_count);
101*4882a593Smuzhiyun lockdep_assert_held(&kvm->mmu_lock);
102*4882a593Smuzhiyun
103*4882a593Smuzhiyun ++sp->root_count;
104*4882a593Smuzhiyun }
105*4882a593Smuzhiyun
kvm_mmu_put_root(struct kvm * kvm,struct kvm_mmu_page * sp)106*4882a593Smuzhiyun static inline bool kvm_mmu_put_root(struct kvm *kvm, struct kvm_mmu_page *sp)
107*4882a593Smuzhiyun {
108*4882a593Smuzhiyun lockdep_assert_held(&kvm->mmu_lock);
109*4882a593Smuzhiyun --sp->root_count;
110*4882a593Smuzhiyun
111*4882a593Smuzhiyun return !sp->root_count;
112*4882a593Smuzhiyun }
113*4882a593Smuzhiyun
114*4882a593Smuzhiyun /*
115*4882a593Smuzhiyun * Return values of handle_mmio_page_fault, mmu.page_fault, and fast_page_fault().
116*4882a593Smuzhiyun *
117*4882a593Smuzhiyun * RET_PF_RETRY: let CPU fault again on the address.
118*4882a593Smuzhiyun * RET_PF_EMULATE: mmio page fault, emulate the instruction directly.
119*4882a593Smuzhiyun * RET_PF_INVALID: the spte is invalid, let the real page fault path update it.
120*4882a593Smuzhiyun * RET_PF_FIXED: The faulting entry has been fixed.
121*4882a593Smuzhiyun * RET_PF_SPURIOUS: The faulting entry was already fixed, e.g. by another vCPU.
122*4882a593Smuzhiyun */
123*4882a593Smuzhiyun enum {
124*4882a593Smuzhiyun RET_PF_RETRY = 0,
125*4882a593Smuzhiyun RET_PF_EMULATE,
126*4882a593Smuzhiyun RET_PF_INVALID,
127*4882a593Smuzhiyun RET_PF_FIXED,
128*4882a593Smuzhiyun RET_PF_SPURIOUS,
129*4882a593Smuzhiyun };
130*4882a593Smuzhiyun
131*4882a593Smuzhiyun /* Bits which may be returned by set_spte() */
132*4882a593Smuzhiyun #define SET_SPTE_WRITE_PROTECTED_PT BIT(0)
133*4882a593Smuzhiyun #define SET_SPTE_NEED_REMOTE_TLB_FLUSH BIT(1)
134*4882a593Smuzhiyun #define SET_SPTE_SPURIOUS BIT(2)
135*4882a593Smuzhiyun
136*4882a593Smuzhiyun int kvm_mmu_hugepage_adjust(struct kvm_vcpu *vcpu, gfn_t gfn,
137*4882a593Smuzhiyun int max_level, kvm_pfn_t *pfnp,
138*4882a593Smuzhiyun bool huge_page_disallowed, int *req_level);
139*4882a593Smuzhiyun void disallowed_hugepage_adjust(u64 spte, gfn_t gfn, int cur_level,
140*4882a593Smuzhiyun kvm_pfn_t *pfnp, int *goal_levelp);
141*4882a593Smuzhiyun
142*4882a593Smuzhiyun bool is_nx_huge_page_enabled(void);
143*4882a593Smuzhiyun
144*4882a593Smuzhiyun void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc);
145*4882a593Smuzhiyun
146*4882a593Smuzhiyun void account_huge_nx_page(struct kvm *kvm, struct kvm_mmu_page *sp);
147*4882a593Smuzhiyun void unaccount_huge_nx_page(struct kvm *kvm, struct kvm_mmu_page *sp);
148*4882a593Smuzhiyun
149*4882a593Smuzhiyun #endif /* __KVM_X86_MMU_INTERNAL_H */
150