xref: /openbmc/linux/arch/x86/kvm/mmu/mmu_internal.h (revision 8957261c)
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 #ifdef CONFIG_KVM_PROVE_MMU
10 #define KVM_MMU_WARN_ON(x) WARN_ON_ONCE(x)
11 #else
12 #define KVM_MMU_WARN_ON(x) BUILD_BUG_ON_INVALID(x)
13 #endif
14 
15 /* Page table builder macros common to shadow (host) PTEs and guest PTEs. */
16 #define __PT_LEVEL_SHIFT(level, bits_per_level)	\
17 	(PAGE_SHIFT + ((level) - 1) * (bits_per_level))
18 #define __PT_INDEX(address, level, bits_per_level) \
19 	(((address) >> __PT_LEVEL_SHIFT(level, bits_per_level)) & ((1 << (bits_per_level)) - 1))
20 
21 #define __PT_LVL_ADDR_MASK(base_addr_mask, level, bits_per_level) \
22 	((base_addr_mask) & ~((1ULL << (PAGE_SHIFT + (((level) - 1) * (bits_per_level)))) - 1))
23 
24 #define __PT_LVL_OFFSET_MASK(base_addr_mask, level, bits_per_level) \
25 	((base_addr_mask) & ((1ULL << (PAGE_SHIFT + (((level) - 1) * (bits_per_level)))) - 1))
26 
27 #define __PT_ENT_PER_PAGE(bits_per_level)  (1 << (bits_per_level))
28 
29 /*
30  * Unlike regular MMU roots, PAE "roots", a.k.a. PDPTEs/PDPTRs, have a PRESENT
31  * bit, and thus are guaranteed to be non-zero when valid.  And, when a guest
32  * PDPTR is !PRESENT, its corresponding PAE root cannot be set to INVALID_PAGE,
33  * as the CPU would treat that as PRESENT PDPTR with reserved bits set.  Use
34  * '0' instead of INVALID_PAGE to indicate an invalid PAE root.
35  */
36 #define INVALID_PAE_ROOT	0
37 #define IS_VALID_PAE_ROOT(x)	(!!(x))
38 
39 static inline hpa_t kvm_mmu_get_dummy_root(void)
40 {
41 	return my_zero_pfn(0) << PAGE_SHIFT;
42 }
43 
44 static inline bool kvm_mmu_is_dummy_root(hpa_t shadow_page)
45 {
46 	return is_zero_pfn(shadow_page >> PAGE_SHIFT);
47 }
48 
49 typedef u64 __rcu *tdp_ptep_t;
50 
51 struct kvm_mmu_page {
52 	/*
53 	 * Note, "link" through "spt" fit in a single 64 byte cache line on
54 	 * 64-bit kernels, keep it that way unless there's a reason not to.
55 	 */
56 	struct list_head link;
57 	struct hlist_node hash_link;
58 
59 	bool tdp_mmu_page;
60 	bool unsync;
61 	u8 mmu_valid_gen;
62 
63 	 /*
64 	  * The shadow page can't be replaced by an equivalent huge page
65 	  * because it is being used to map an executable page in the guest
66 	  * and the NX huge page mitigation is enabled.
67 	  */
68 	bool nx_huge_page_disallowed;
69 
70 	/*
71 	 * The following two entries are used to key the shadow page in the
72 	 * hash table.
73 	 */
74 	union kvm_mmu_page_role role;
75 	gfn_t gfn;
76 
77 	u64 *spt;
78 
79 	/*
80 	 * Stores the result of the guest translation being shadowed by each
81 	 * SPTE.  KVM shadows two types of guest translations: nGPA -> GPA
82 	 * (shadow EPT/NPT) and GVA -> GPA (traditional shadow paging). In both
83 	 * cases the result of the translation is a GPA and a set of access
84 	 * constraints.
85 	 *
86 	 * The GFN is stored in the upper bits (PAGE_SHIFT) and the shadowed
87 	 * access permissions are stored in the lower bits. Note, for
88 	 * convenience and uniformity across guests, the access permissions are
89 	 * stored in KVM format (e.g.  ACC_EXEC_MASK) not the raw guest format.
90 	 */
91 	u64 *shadowed_translation;
92 
93 	/* Currently serving as active root */
94 	union {
95 		int root_count;
96 		refcount_t tdp_mmu_root_count;
97 	};
98 	unsigned int unsync_children;
99 	union {
100 		struct kvm_rmap_head parent_ptes; /* rmap pointers to parent sptes */
101 		tdp_ptep_t ptep;
102 	};
103 	union {
104 		DECLARE_BITMAP(unsync_child_bitmap, 512);
105 		struct {
106 			struct work_struct tdp_mmu_async_work;
107 			void *tdp_mmu_async_data;
108 		};
109 	};
110 
111 	/*
112 	 * Tracks shadow pages that, if zapped, would allow KVM to create an NX
113 	 * huge page.  A shadow page will have nx_huge_page_disallowed set but
114 	 * not be on the list if a huge page is disallowed for other reasons,
115 	 * e.g. because KVM is shadowing a PTE at the same gfn, the memslot
116 	 * isn't properly aligned, etc...
117 	 */
118 	struct list_head possible_nx_huge_page_link;
119 #ifdef CONFIG_X86_32
120 	/*
121 	 * Used out of the mmu-lock to avoid reading spte values while an
122 	 * update is in progress; see the comments in __get_spte_lockless().
123 	 */
124 	int clear_spte_count;
125 #endif
126 
127 	/* Number of writes since the last time traversal visited this page.  */
128 	atomic_t write_flooding_count;
129 
130 #ifdef CONFIG_X86_64
131 	/* Used for freeing the page asynchronously if it is a TDP MMU page. */
132 	struct rcu_head rcu_head;
133 #endif
134 };
135 
136 extern struct kmem_cache *mmu_page_header_cache;
137 
138 static inline int kvm_mmu_role_as_id(union kvm_mmu_page_role role)
139 {
140 	return role.smm ? 1 : 0;
141 }
142 
143 static inline int kvm_mmu_page_as_id(struct kvm_mmu_page *sp)
144 {
145 	return kvm_mmu_role_as_id(sp->role);
146 }
147 
148 static inline bool kvm_mmu_page_ad_need_write_protect(struct kvm_mmu_page *sp)
149 {
150 	/*
151 	 * When using the EPT page-modification log, the GPAs in the CPU dirty
152 	 * log would come from L2 rather than L1.  Therefore, we need to rely
153 	 * on write protection to record dirty pages, which bypasses PML, since
154 	 * writes now result in a vmexit.  Note, the check on CPU dirty logging
155 	 * being enabled is mandatory as the bits used to denote WP-only SPTEs
156 	 * are reserved for PAE paging (32-bit KVM).
157 	 */
158 	return kvm_x86_ops.cpu_dirty_log_size && sp->role.guest_mode;
159 }
160 
161 static inline gfn_t gfn_round_for_level(gfn_t gfn, int level)
162 {
163 	return gfn & -KVM_PAGES_PER_HPAGE(level);
164 }
165 
166 int mmu_try_to_unsync_pages(struct kvm *kvm, const struct kvm_memory_slot *slot,
167 			    gfn_t gfn, bool can_unsync, bool prefetch);
168 
169 void kvm_mmu_gfn_disallow_lpage(const struct kvm_memory_slot *slot, gfn_t gfn);
170 void kvm_mmu_gfn_allow_lpage(const struct kvm_memory_slot *slot, gfn_t gfn);
171 bool kvm_mmu_slot_gfn_write_protect(struct kvm *kvm,
172 				    struct kvm_memory_slot *slot, u64 gfn,
173 				    int min_level);
174 
175 /* Flush the given page (huge or not) of guest memory. */
176 static inline void kvm_flush_remote_tlbs_gfn(struct kvm *kvm, gfn_t gfn, int level)
177 {
178 	kvm_flush_remote_tlbs_range(kvm, gfn_round_for_level(gfn, level),
179 				    KVM_PAGES_PER_HPAGE(level));
180 }
181 
182 unsigned int pte_list_count(struct kvm_rmap_head *rmap_head);
183 
184 extern int nx_huge_pages;
185 static inline bool is_nx_huge_page_enabled(struct kvm *kvm)
186 {
187 	return READ_ONCE(nx_huge_pages) && !kvm->arch.disable_nx_huge_pages;
188 }
189 
190 struct kvm_page_fault {
191 	/* arguments to kvm_mmu_do_page_fault.  */
192 	const gpa_t addr;
193 	const u32 error_code;
194 	const bool prefetch;
195 
196 	/* Derived from error_code.  */
197 	const bool exec;
198 	const bool write;
199 	const bool present;
200 	const bool rsvd;
201 	const bool user;
202 
203 	/* Derived from mmu and global state.  */
204 	const bool is_tdp;
205 	const bool nx_huge_page_workaround_enabled;
206 
207 	/*
208 	 * Whether a >4KB mapping can be created or is forbidden due to NX
209 	 * hugepages.
210 	 */
211 	bool huge_page_disallowed;
212 
213 	/*
214 	 * Maximum page size that can be created for this fault; input to
215 	 * FNAME(fetch), direct_map() and kvm_tdp_mmu_map().
216 	 */
217 	u8 max_level;
218 
219 	/*
220 	 * Page size that can be created based on the max_level and the
221 	 * page size used by the host mapping.
222 	 */
223 	u8 req_level;
224 
225 	/*
226 	 * Page size that will be created based on the req_level and
227 	 * huge_page_disallowed.
228 	 */
229 	u8 goal_level;
230 
231 	/* Shifted addr, or result of guest page table walk if addr is a gva.  */
232 	gfn_t gfn;
233 
234 	/* The memslot containing gfn. May be NULL. */
235 	struct kvm_memory_slot *slot;
236 
237 	/* Outputs of kvm_faultin_pfn.  */
238 	unsigned long mmu_seq;
239 	kvm_pfn_t pfn;
240 	hva_t hva;
241 	bool map_writable;
242 
243 	/*
244 	 * Indicates the guest is trying to write a gfn that contains one or
245 	 * more of the PTEs used to translate the write itself, i.e. the access
246 	 * is changing its own translation in the guest page tables.
247 	 */
248 	bool write_fault_to_shadow_pgtable;
249 };
250 
251 int kvm_tdp_page_fault(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault);
252 
253 /*
254  * Return values of handle_mmio_page_fault(), mmu.page_fault(), fast_page_fault(),
255  * and of course kvm_mmu_do_page_fault().
256  *
257  * RET_PF_CONTINUE: So far, so good, keep handling the page fault.
258  * RET_PF_RETRY: let CPU fault again on the address.
259  * RET_PF_EMULATE: mmio page fault, emulate the instruction directly.
260  * RET_PF_INVALID: the spte is invalid, let the real page fault path update it.
261  * RET_PF_FIXED: The faulting entry has been fixed.
262  * RET_PF_SPURIOUS: The faulting entry was already fixed, e.g. by another vCPU.
263  *
264  * Any names added to this enum should be exported to userspace for use in
265  * tracepoints via TRACE_DEFINE_ENUM() in mmutrace.h
266  *
267  * Note, all values must be greater than or equal to zero so as not to encroach
268  * on -errno return values.  Somewhat arbitrarily use '0' for CONTINUE, which
269  * will allow for efficient machine code when checking for CONTINUE, e.g.
270  * "TEST %rax, %rax, JNZ", as all "stop!" values are non-zero.
271  */
272 enum {
273 	RET_PF_CONTINUE = 0,
274 	RET_PF_RETRY,
275 	RET_PF_EMULATE,
276 	RET_PF_INVALID,
277 	RET_PF_FIXED,
278 	RET_PF_SPURIOUS,
279 };
280 
281 static inline int kvm_mmu_do_page_fault(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
282 					u32 err, bool prefetch, int *emulation_type)
283 {
284 	struct kvm_page_fault fault = {
285 		.addr = cr2_or_gpa,
286 		.error_code = err,
287 		.exec = err & PFERR_FETCH_MASK,
288 		.write = err & PFERR_WRITE_MASK,
289 		.present = err & PFERR_PRESENT_MASK,
290 		.rsvd = err & PFERR_RSVD_MASK,
291 		.user = err & PFERR_USER_MASK,
292 		.prefetch = prefetch,
293 		.is_tdp = likely(vcpu->arch.mmu->page_fault == kvm_tdp_page_fault),
294 		.nx_huge_page_workaround_enabled =
295 			is_nx_huge_page_enabled(vcpu->kvm),
296 
297 		.max_level = KVM_MAX_HUGEPAGE_LEVEL,
298 		.req_level = PG_LEVEL_4K,
299 		.goal_level = PG_LEVEL_4K,
300 	};
301 	int r;
302 
303 	if (vcpu->arch.mmu->root_role.direct) {
304 		fault.gfn = fault.addr >> PAGE_SHIFT;
305 		fault.slot = kvm_vcpu_gfn_to_memslot(vcpu, fault.gfn);
306 	}
307 
308 	/*
309 	 * Async #PF "faults", a.k.a. prefetch faults, are not faults from the
310 	 * guest perspective and have already been counted at the time of the
311 	 * original fault.
312 	 */
313 	if (!prefetch)
314 		vcpu->stat.pf_taken++;
315 
316 	if (IS_ENABLED(CONFIG_RETPOLINE) && fault.is_tdp)
317 		r = kvm_tdp_page_fault(vcpu, &fault);
318 	else
319 		r = vcpu->arch.mmu->page_fault(vcpu, &fault);
320 
321 	if (fault.write_fault_to_shadow_pgtable && emulation_type)
322 		*emulation_type |= EMULTYPE_WRITE_PF_TO_SP;
323 
324 	/*
325 	 * Similar to above, prefetch faults aren't truly spurious, and the
326 	 * async #PF path doesn't do emulation.  Do count faults that are fixed
327 	 * by the async #PF handler though, otherwise they'll never be counted.
328 	 */
329 	if (r == RET_PF_FIXED)
330 		vcpu->stat.pf_fixed++;
331 	else if (prefetch)
332 		;
333 	else if (r == RET_PF_EMULATE)
334 		vcpu->stat.pf_emulate++;
335 	else if (r == RET_PF_SPURIOUS)
336 		vcpu->stat.pf_spurious++;
337 	return r;
338 }
339 
340 int kvm_mmu_max_mapping_level(struct kvm *kvm,
341 			      const struct kvm_memory_slot *slot, gfn_t gfn,
342 			      int max_level);
343 void kvm_mmu_hugepage_adjust(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault);
344 void disallowed_hugepage_adjust(struct kvm_page_fault *fault, u64 spte, int cur_level);
345 
346 void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc);
347 
348 void track_possible_nx_huge_page(struct kvm *kvm, struct kvm_mmu_page *sp);
349 void untrack_possible_nx_huge_page(struct kvm *kvm, struct kvm_mmu_page *sp);
350 
351 #endif /* __KVM_X86_MMU_INTERNAL_H */
352