xref: /openbmc/linux/arch/x86/kvm/mmu/mmu_internal.h (revision 66c98360)
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 /* Page table builder macros common to shadow (host) PTEs and guest PTEs. */
24 #define __PT_LEVEL_SHIFT(level, bits_per_level)	\
25 	(PAGE_SHIFT + ((level) - 1) * (bits_per_level))
26 #define __PT_INDEX(address, level, bits_per_level) \
27 	(((address) >> __PT_LEVEL_SHIFT(level, bits_per_level)) & ((1 << (bits_per_level)) - 1))
28 
29 #define __PT_LVL_ADDR_MASK(base_addr_mask, level, bits_per_level) \
30 	((base_addr_mask) & ~((1ULL << (PAGE_SHIFT + (((level) - 1) * (bits_per_level)))) - 1))
31 
32 #define __PT_LVL_OFFSET_MASK(base_addr_mask, level, bits_per_level) \
33 	((base_addr_mask) & ((1ULL << (PAGE_SHIFT + (((level) - 1) * (bits_per_level)))) - 1))
34 
35 #define __PT_ENT_PER_PAGE(bits_per_level)  (1 << (bits_per_level))
36 
37 /*
38  * Unlike regular MMU roots, PAE "roots", a.k.a. PDPTEs/PDPTRs, have a PRESENT
39  * bit, and thus are guaranteed to be non-zero when valid.  And, when a guest
40  * PDPTR is !PRESENT, its corresponding PAE root cannot be set to INVALID_PAGE,
41  * as the CPU would treat that as PRESENT PDPTR with reserved bits set.  Use
42  * '0' instead of INVALID_PAGE to indicate an invalid PAE root.
43  */
44 #define INVALID_PAE_ROOT	0
45 #define IS_VALID_PAE_ROOT(x)	(!!(x))
46 
47 typedef u64 __rcu *tdp_ptep_t;
48 
49 struct kvm_mmu_page {
50 	/*
51 	 * Note, "link" through "spt" fit in a single 64 byte cache line on
52 	 * 64-bit kernels, keep it that way unless there's a reason not to.
53 	 */
54 	struct list_head link;
55 	struct hlist_node hash_link;
56 
57 	bool tdp_mmu_page;
58 	bool unsync;
59 	u8 mmu_valid_gen;
60 
61 	 /*
62 	  * The shadow page can't be replaced by an equivalent huge page
63 	  * because it is being used to map an executable page in the guest
64 	  * and the NX huge page mitigation is enabled.
65 	  */
66 	bool nx_huge_page_disallowed;
67 
68 	/*
69 	 * The following two entries are used to key the shadow page in the
70 	 * hash table.
71 	 */
72 	union kvm_mmu_page_role role;
73 	gfn_t gfn;
74 
75 	u64 *spt;
76 
77 	/*
78 	 * Stores the result of the guest translation being shadowed by each
79 	 * SPTE.  KVM shadows two types of guest translations: nGPA -> GPA
80 	 * (shadow EPT/NPT) and GVA -> GPA (traditional shadow paging). In both
81 	 * cases the result of the translation is a GPA and a set of access
82 	 * constraints.
83 	 *
84 	 * The GFN is stored in the upper bits (PAGE_SHIFT) and the shadowed
85 	 * access permissions are stored in the lower bits. Note, for
86 	 * convenience and uniformity across guests, the access permissions are
87 	 * stored in KVM format (e.g.  ACC_EXEC_MASK) not the raw guest format.
88 	 */
89 	u64 *shadowed_translation;
90 
91 	/* Currently serving as active root */
92 	union {
93 		int root_count;
94 		refcount_t tdp_mmu_root_count;
95 	};
96 	unsigned int unsync_children;
97 	union {
98 		struct kvm_rmap_head parent_ptes; /* rmap pointers to parent sptes */
99 		tdp_ptep_t ptep;
100 	};
101 	union {
102 		DECLARE_BITMAP(unsync_child_bitmap, 512);
103 		struct {
104 			struct work_struct tdp_mmu_async_work;
105 			void *tdp_mmu_async_data;
106 		};
107 	};
108 
109 	/*
110 	 * Tracks shadow pages that, if zapped, would allow KVM to create an NX
111 	 * huge page.  A shadow page will have nx_huge_page_disallowed set but
112 	 * not be on the list if a huge page is disallowed for other reasons,
113 	 * e.g. because KVM is shadowing a PTE at the same gfn, the memslot
114 	 * isn't properly aligned, etc...
115 	 */
116 	struct list_head possible_nx_huge_page_link;
117 #ifdef CONFIG_X86_32
118 	/*
119 	 * Used out of the mmu-lock to avoid reading spte values while an
120 	 * update is in progress; see the comments in __get_spte_lockless().
121 	 */
122 	int clear_spte_count;
123 #endif
124 
125 	/* Number of writes since the last time traversal visited this page.  */
126 	atomic_t write_flooding_count;
127 
128 #ifdef CONFIG_X86_64
129 	/* Used for freeing the page asynchronously if it is a TDP MMU page. */
130 	struct rcu_head rcu_head;
131 #endif
132 };
133 
134 extern struct kmem_cache *mmu_page_header_cache;
135 
136 static inline int kvm_mmu_role_as_id(union kvm_mmu_page_role role)
137 {
138 	return role.smm ? 1 : 0;
139 }
140 
141 static inline int kvm_mmu_page_as_id(struct kvm_mmu_page *sp)
142 {
143 	return kvm_mmu_role_as_id(sp->role);
144 }
145 
146 static inline bool kvm_mmu_page_ad_need_write_protect(struct kvm_mmu_page *sp)
147 {
148 	/*
149 	 * When using the EPT page-modification log, the GPAs in the CPU dirty
150 	 * log would come from L2 rather than L1.  Therefore, we need to rely
151 	 * on write protection to record dirty pages, which bypasses PML, since
152 	 * writes now result in a vmexit.  Note, the check on CPU dirty logging
153 	 * being enabled is mandatory as the bits used to denote WP-only SPTEs
154 	 * are reserved for PAE paging (32-bit KVM).
155 	 */
156 	return kvm_x86_ops.cpu_dirty_log_size && sp->role.guest_mode;
157 }
158 
159 static inline gfn_t gfn_round_for_level(gfn_t gfn, int level)
160 {
161 	return gfn & -KVM_PAGES_PER_HPAGE(level);
162 }
163 
164 int mmu_try_to_unsync_pages(struct kvm *kvm, const struct kvm_memory_slot *slot,
165 			    gfn_t gfn, bool can_unsync, bool prefetch);
166 
167 void kvm_mmu_gfn_disallow_lpage(const struct kvm_memory_slot *slot, gfn_t gfn);
168 void kvm_mmu_gfn_allow_lpage(const struct kvm_memory_slot *slot, gfn_t gfn);
169 bool kvm_mmu_slot_gfn_write_protect(struct kvm *kvm,
170 				    struct kvm_memory_slot *slot, u64 gfn,
171 				    int min_level);
172 
173 void kvm_flush_remote_tlbs_range(struct kvm *kvm, gfn_t start_gfn,
174 				 gfn_t nr_pages);
175 
176 /* Flush the given page (huge or not) of guest memory. */
177 static inline void kvm_flush_remote_tlbs_gfn(struct kvm *kvm, gfn_t gfn, int level)
178 {
179 	kvm_flush_remote_tlbs_range(kvm, gfn_round_for_level(gfn, level),
180 				    KVM_PAGES_PER_HPAGE(level));
181 }
182 
183 unsigned int pte_list_count(struct kvm_rmap_head *rmap_head);
184 
185 extern int nx_huge_pages;
186 static inline bool is_nx_huge_page_enabled(struct kvm *kvm)
187 {
188 	return READ_ONCE(nx_huge_pages) && !kvm->arch.disable_nx_huge_pages;
189 }
190 
191 struct kvm_page_fault {
192 	/* arguments to kvm_mmu_do_page_fault.  */
193 	const gpa_t addr;
194 	const u32 error_code;
195 	const bool prefetch;
196 
197 	/* Derived from error_code.  */
198 	const bool exec;
199 	const bool write;
200 	const bool present;
201 	const bool rsvd;
202 	const bool user;
203 
204 	/* Derived from mmu and global state.  */
205 	const bool is_tdp;
206 	const bool nx_huge_page_workaround_enabled;
207 
208 	/*
209 	 * Whether a >4KB mapping can be created or is forbidden due to NX
210 	 * hugepages.
211 	 */
212 	bool huge_page_disallowed;
213 
214 	/*
215 	 * Maximum page size that can be created for this fault; input to
216 	 * FNAME(fetch), direct_map() and kvm_tdp_mmu_map().
217 	 */
218 	u8 max_level;
219 
220 	/*
221 	 * Page size that can be created based on the max_level and the
222 	 * page size used by the host mapping.
223 	 */
224 	u8 req_level;
225 
226 	/*
227 	 * Page size that will be created based on the req_level and
228 	 * huge_page_disallowed.
229 	 */
230 	u8 goal_level;
231 
232 	/* Shifted addr, or result of guest page table walk if addr is a gva.  */
233 	gfn_t gfn;
234 
235 	/* The memslot containing gfn. May be NULL. */
236 	struct kvm_memory_slot *slot;
237 
238 	/* Outputs of kvm_faultin_pfn.  */
239 	unsigned long mmu_seq;
240 	kvm_pfn_t pfn;
241 	hva_t hva;
242 	bool map_writable;
243 
244 	/*
245 	 * Indicates the guest is trying to write a gfn that contains one or
246 	 * more of the PTEs used to translate the write itself, i.e. the access
247 	 * is changing its own translation in the guest page tables.
248 	 */
249 	bool write_fault_to_shadow_pgtable;
250 };
251 
252 int kvm_tdp_page_fault(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault);
253 
254 /*
255  * Return values of handle_mmio_page_fault(), mmu.page_fault(), fast_page_fault(),
256  * and of course kvm_mmu_do_page_fault().
257  *
258  * RET_PF_CONTINUE: So far, so good, keep handling the page fault.
259  * RET_PF_RETRY: let CPU fault again on the address.
260  * RET_PF_EMULATE: mmio page fault, emulate the instruction directly.
261  * RET_PF_INVALID: the spte is invalid, let the real page fault path update it.
262  * RET_PF_FIXED: The faulting entry has been fixed.
263  * RET_PF_SPURIOUS: The faulting entry was already fixed, e.g. by another vCPU.
264  *
265  * Any names added to this enum should be exported to userspace for use in
266  * tracepoints via TRACE_DEFINE_ENUM() in mmutrace.h
267  *
268  * Note, all values must be greater than or equal to zero so as not to encroach
269  * on -errno return values.  Somewhat arbitrarily use '0' for CONTINUE, which
270  * will allow for efficient machine code when checking for CONTINUE, e.g.
271  * "TEST %rax, %rax, JNZ", as all "stop!" values are non-zero.
272  */
273 enum {
274 	RET_PF_CONTINUE = 0,
275 	RET_PF_RETRY,
276 	RET_PF_EMULATE,
277 	RET_PF_INVALID,
278 	RET_PF_FIXED,
279 	RET_PF_SPURIOUS,
280 };
281 
282 static inline int kvm_mmu_do_page_fault(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
283 					u32 err, bool prefetch, int *emulation_type)
284 {
285 	struct kvm_page_fault fault = {
286 		.addr = cr2_or_gpa,
287 		.error_code = err,
288 		.exec = err & PFERR_FETCH_MASK,
289 		.write = err & PFERR_WRITE_MASK,
290 		.present = err & PFERR_PRESENT_MASK,
291 		.rsvd = err & PFERR_RSVD_MASK,
292 		.user = err & PFERR_USER_MASK,
293 		.prefetch = prefetch,
294 		.is_tdp = likely(vcpu->arch.mmu->page_fault == kvm_tdp_page_fault),
295 		.nx_huge_page_workaround_enabled =
296 			is_nx_huge_page_enabled(vcpu->kvm),
297 
298 		.max_level = KVM_MAX_HUGEPAGE_LEVEL,
299 		.req_level = PG_LEVEL_4K,
300 		.goal_level = PG_LEVEL_4K,
301 	};
302 	int r;
303 
304 	if (vcpu->arch.mmu->root_role.direct) {
305 		fault.gfn = fault.addr >> PAGE_SHIFT;
306 		fault.slot = kvm_vcpu_gfn_to_memslot(vcpu, fault.gfn);
307 	}
308 
309 	/*
310 	 * Async #PF "faults", a.k.a. prefetch faults, are not faults from the
311 	 * guest perspective and have already been counted at the time of the
312 	 * original fault.
313 	 */
314 	if (!prefetch)
315 		vcpu->stat.pf_taken++;
316 
317 	if (IS_ENABLED(CONFIG_RETPOLINE) && fault.is_tdp)
318 		r = kvm_tdp_page_fault(vcpu, &fault);
319 	else
320 		r = vcpu->arch.mmu->page_fault(vcpu, &fault);
321 
322 	if (fault.write_fault_to_shadow_pgtable && emulation_type)
323 		*emulation_type |= EMULTYPE_WRITE_PF_TO_SP;
324 
325 	/*
326 	 * Similar to above, prefetch faults aren't truly spurious, and the
327 	 * async #PF path doesn't do emulation.  Do count faults that are fixed
328 	 * by the async #PF handler though, otherwise they'll never be counted.
329 	 */
330 	if (r == RET_PF_FIXED)
331 		vcpu->stat.pf_fixed++;
332 	else if (prefetch)
333 		;
334 	else if (r == RET_PF_EMULATE)
335 		vcpu->stat.pf_emulate++;
336 	else if (r == RET_PF_SPURIOUS)
337 		vcpu->stat.pf_spurious++;
338 	return r;
339 }
340 
341 int kvm_mmu_max_mapping_level(struct kvm *kvm,
342 			      const struct kvm_memory_slot *slot, gfn_t gfn,
343 			      int max_level);
344 void kvm_mmu_hugepage_adjust(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault);
345 void disallowed_hugepage_adjust(struct kvm_page_fault *fault, u64 spte, int cur_level);
346 
347 void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc);
348 
349 void track_possible_nx_huge_page(struct kvm *kvm, struct kvm_mmu_page *sp);
350 void untrack_possible_nx_huge_page(struct kvm *kvm, struct kvm_mmu_page *sp);
351 
352 #endif /* __KVM_X86_MMU_INTERNAL_H */
353