xref: /openbmc/linux/arch/x86/kvm/mmu/mmu_internal.h (revision 9c840d5f)
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 typedef u64 __rcu *tdp_ptep_t;
34 
35 struct kvm_mmu_page {
36 	/*
37 	 * Note, "link" through "spt" fit in a single 64 byte cache line on
38 	 * 64-bit kernels, keep it that way unless there's a reason not to.
39 	 */
40 	struct list_head link;
41 	struct hlist_node hash_link;
42 
43 	bool tdp_mmu_page;
44 	bool unsync;
45 	u8 mmu_valid_gen;
46 	bool lpage_disallowed; /* Can't be replaced by an equiv large page */
47 
48 	/*
49 	 * The following two entries are used to key the shadow page in the
50 	 * hash table.
51 	 */
52 	union kvm_mmu_page_role role;
53 	gfn_t gfn;
54 
55 	u64 *spt;
56 	/* hold the gfn of each spte inside spt */
57 	gfn_t *gfns;
58 	/* Currently serving as active root */
59 	union {
60 		int root_count;
61 		refcount_t tdp_mmu_root_count;
62 	};
63 	unsigned int unsync_children;
64 	union {
65 		struct kvm_rmap_head parent_ptes; /* rmap pointers to parent sptes */
66 		tdp_ptep_t ptep;
67 	};
68 	union {
69 		DECLARE_BITMAP(unsync_child_bitmap, 512);
70 		struct {
71 			struct work_struct tdp_mmu_async_work;
72 			void *tdp_mmu_async_data;
73 		};
74 	};
75 
76 	struct list_head lpage_disallowed_link;
77 #ifdef CONFIG_X86_32
78 	/*
79 	 * Used out of the mmu-lock to avoid reading spte values while an
80 	 * update is in progress; see the comments in __get_spte_lockless().
81 	 */
82 	int clear_spte_count;
83 #endif
84 
85 	/* Number of writes since the last time traversal visited this page.  */
86 	atomic_t write_flooding_count;
87 
88 #ifdef CONFIG_X86_64
89 	/* Used for freeing the page asynchronously if it is a TDP MMU page. */
90 	struct rcu_head rcu_head;
91 #endif
92 };
93 
94 extern struct kmem_cache *mmu_page_header_cache;
95 
96 static inline struct kvm_mmu_page *to_shadow_page(hpa_t shadow_page)
97 {
98 	struct page *page = pfn_to_page(shadow_page >> PAGE_SHIFT);
99 
100 	return (struct kvm_mmu_page *)page_private(page);
101 }
102 
103 static inline struct kvm_mmu_page *sptep_to_sp(u64 *sptep)
104 {
105 	return to_shadow_page(__pa(sptep));
106 }
107 
108 static inline int kvm_mmu_role_as_id(union kvm_mmu_page_role role)
109 {
110 	return role.smm ? 1 : 0;
111 }
112 
113 static inline int kvm_mmu_page_as_id(struct kvm_mmu_page *sp)
114 {
115 	return kvm_mmu_role_as_id(sp->role);
116 }
117 
118 static inline bool kvm_mmu_page_ad_need_write_protect(struct kvm_mmu_page *sp)
119 {
120 	/*
121 	 * When using the EPT page-modification log, the GPAs in the CPU dirty
122 	 * log would come from L2 rather than L1.  Therefore, we need to rely
123 	 * on write protection to record dirty pages, which bypasses PML, since
124 	 * writes now result in a vmexit.  Note, the check on CPU dirty logging
125 	 * being enabled is mandatory as the bits used to denote WP-only SPTEs
126 	 * are reserved for PAE paging (32-bit KVM).
127 	 */
128 	return kvm_x86_ops.cpu_dirty_log_size && sp->role.guest_mode;
129 }
130 
131 int mmu_try_to_unsync_pages(struct kvm *kvm, const struct kvm_memory_slot *slot,
132 			    gfn_t gfn, bool can_unsync, bool prefetch);
133 
134 void kvm_mmu_gfn_disallow_lpage(const struct kvm_memory_slot *slot, gfn_t gfn);
135 void kvm_mmu_gfn_allow_lpage(const struct kvm_memory_slot *slot, gfn_t gfn);
136 bool kvm_mmu_slot_gfn_write_protect(struct kvm *kvm,
137 				    struct kvm_memory_slot *slot, u64 gfn,
138 				    int min_level);
139 void kvm_flush_remote_tlbs_with_address(struct kvm *kvm,
140 					u64 start_gfn, u64 pages);
141 unsigned int pte_list_count(struct kvm_rmap_head *rmap_head);
142 
143 extern int nx_huge_pages;
144 static inline bool is_nx_huge_page_enabled(void)
145 {
146 	return READ_ONCE(nx_huge_pages);
147 }
148 
149 struct kvm_page_fault {
150 	/* arguments to kvm_mmu_do_page_fault.  */
151 	const gpa_t addr;
152 	const u32 error_code;
153 	const bool prefetch;
154 
155 	/* Derived from error_code.  */
156 	const bool exec;
157 	const bool write;
158 	const bool present;
159 	const bool rsvd;
160 	const bool user;
161 
162 	/* Derived from mmu and global state.  */
163 	const bool is_tdp;
164 	const bool nx_huge_page_workaround_enabled;
165 
166 	/*
167 	 * Whether a >4KB mapping can be created or is forbidden due to NX
168 	 * hugepages.
169 	 */
170 	bool huge_page_disallowed;
171 
172 	/*
173 	 * Maximum page size that can be created for this fault; input to
174 	 * FNAME(fetch), __direct_map and kvm_tdp_mmu_map.
175 	 */
176 	u8 max_level;
177 
178 	/*
179 	 * Page size that can be created based on the max_level and the
180 	 * page size used by the host mapping.
181 	 */
182 	u8 req_level;
183 
184 	/*
185 	 * Page size that will be created based on the req_level and
186 	 * huge_page_disallowed.
187 	 */
188 	u8 goal_level;
189 
190 	/* Shifted addr, or result of guest page table walk if addr is a gva.  */
191 	gfn_t gfn;
192 
193 	/* The memslot containing gfn. May be NULL. */
194 	struct kvm_memory_slot *slot;
195 
196 	/* Outputs of kvm_faultin_pfn.  */
197 	kvm_pfn_t pfn;
198 	hva_t hva;
199 	bool map_writable;
200 };
201 
202 int kvm_tdp_page_fault(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault);
203 
204 /*
205  * Return values of handle_mmio_page_fault(), mmu.page_fault(), fast_page_fault(),
206  * and of course kvm_mmu_do_page_fault().
207  *
208  * RET_PF_CONTINUE: So far, so good, keep handling the page fault.
209  * RET_PF_RETRY: let CPU fault again on the address.
210  * RET_PF_EMULATE: mmio page fault, emulate the instruction directly.
211  * RET_PF_INVALID: the spte is invalid, let the real page fault path update it.
212  * RET_PF_FIXED: The faulting entry has been fixed.
213  * RET_PF_SPURIOUS: The faulting entry was already fixed, e.g. by another vCPU.
214  *
215  * Any names added to this enum should be exported to userspace for use in
216  * tracepoints via TRACE_DEFINE_ENUM() in mmutrace.h
217  *
218  * Note, all values must be greater than or equal to zero so as not to encroach
219  * on -errno return values.  Somewhat arbitrarily use '0' for CONTINUE, which
220  * will allow for efficient machine code when checking for CONTINUE, e.g.
221  * "TEST %rax, %rax, JNZ", as all "stop!" values are non-zero.
222  */
223 enum {
224 	RET_PF_CONTINUE = 0,
225 	RET_PF_RETRY,
226 	RET_PF_EMULATE,
227 	RET_PF_INVALID,
228 	RET_PF_FIXED,
229 	RET_PF_SPURIOUS,
230 };
231 
232 static inline int kvm_mmu_do_page_fault(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
233 					u32 err, bool prefetch)
234 {
235 	struct kvm_page_fault fault = {
236 		.addr = cr2_or_gpa,
237 		.error_code = err,
238 		.exec = err & PFERR_FETCH_MASK,
239 		.write = err & PFERR_WRITE_MASK,
240 		.present = err & PFERR_PRESENT_MASK,
241 		.rsvd = err & PFERR_RSVD_MASK,
242 		.user = err & PFERR_USER_MASK,
243 		.prefetch = prefetch,
244 		.is_tdp = likely(vcpu->arch.mmu->page_fault == kvm_tdp_page_fault),
245 		.nx_huge_page_workaround_enabled = is_nx_huge_page_enabled(),
246 
247 		.max_level = KVM_MAX_HUGEPAGE_LEVEL,
248 		.req_level = PG_LEVEL_4K,
249 		.goal_level = PG_LEVEL_4K,
250 	};
251 	int r;
252 
253 	/*
254 	 * Async #PF "faults", a.k.a. prefetch faults, are not faults from the
255 	 * guest perspective and have already been counted at the time of the
256 	 * original fault.
257 	 */
258 	if (!prefetch)
259 		vcpu->stat.pf_taken++;
260 
261 	if (IS_ENABLED(CONFIG_RETPOLINE) && fault.is_tdp)
262 		r = kvm_tdp_page_fault(vcpu, &fault);
263 	else
264 		r = vcpu->arch.mmu->page_fault(vcpu, &fault);
265 
266 	/*
267 	 * Similar to above, prefetch faults aren't truly spurious, and the
268 	 * async #PF path doesn't do emulation.  Do count faults that are fixed
269 	 * by the async #PF handler though, otherwise they'll never be counted.
270 	 */
271 	if (r == RET_PF_FIXED)
272 		vcpu->stat.pf_fixed++;
273 	else if (prefetch)
274 		;
275 	else if (r == RET_PF_EMULATE)
276 		vcpu->stat.pf_emulate++;
277 	else if (r == RET_PF_SPURIOUS)
278 		vcpu->stat.pf_spurious++;
279 	return r;
280 }
281 
282 int kvm_mmu_max_mapping_level(struct kvm *kvm,
283 			      const struct kvm_memory_slot *slot, gfn_t gfn,
284 			      kvm_pfn_t pfn, int max_level);
285 void kvm_mmu_hugepage_adjust(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault);
286 void disallowed_hugepage_adjust(struct kvm_page_fault *fault, u64 spte, int cur_level);
287 
288 void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc);
289 
290 void account_huge_nx_page(struct kvm *kvm, struct kvm_mmu_page *sp);
291 void unaccount_huge_nx_page(struct kvm *kvm, struct kvm_mmu_page *sp);
292 
293 #endif /* __KVM_X86_MMU_INTERNAL_H */
294