1 /* include/asm-generic/tlb.h
2 *
3 * Generic TLB shootdown code
4 *
5 * Copyright 2001 Red Hat, Inc.
6 * Based on code from mm/memory.c Copyright Linus Torvalds and others.
7 *
8 * Copyright 2011 Red Hat, Inc., Peter Zijlstra
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License
12 * as published by the Free Software Foundation; either version
13 * 2 of the License, or (at your option) any later version.
14 */
15 #ifndef _ASM_GENERIC__TLB_H
16 #define _ASM_GENERIC__TLB_H
17
18 #include <linux/mmu_notifier.h>
19 #include <linux/swap.h>
20 #include <asm/pgalloc.h>
21 #include <asm/tlbflush.h>
22
23 #ifdef CONFIG_HAVE_RCU_TABLE_FREE
24 /*
25 * Semi RCU freeing of the page directories.
26 *
27 * This is needed by some architectures to implement software pagetable walkers.
28 *
29 * gup_fast() and other software pagetable walkers do a lockless page-table
30 * walk and therefore needs some synchronization with the freeing of the page
31 * directories. The chosen means to accomplish that is by disabling IRQs over
32 * the walk.
33 *
34 * Architectures that use IPIs to flush TLBs will then automagically DTRT,
35 * since we unlink the page, flush TLBs, free the page. Since the disabling of
36 * IRQs delays the completion of the TLB flush we can never observe an already
37 * freed page.
38 *
39 * Architectures that do not have this (PPC) need to delay the freeing by some
40 * other means, this is that means.
41 *
42 * What we do is batch the freed directory pages (tables) and RCU free them.
43 * We use the sched RCU variant, as that guarantees that IRQ/preempt disabling
44 * holds off grace periods.
45 *
46 * However, in order to batch these pages we need to allocate storage, this
47 * allocation is deep inside the MM code and can thus easily fail on memory
48 * pressure. To guarantee progress we fall back to single table freeing, see
49 * the implementation of tlb_remove_table_one().
50 *
51 */
52 struct mmu_table_batch {
53 struct rcu_head rcu;
54 unsigned int nr;
55 void *tables[0];
56 };
57
58 #define MAX_TABLE_BATCH \
59 ((PAGE_SIZE - sizeof(struct mmu_table_batch)) / sizeof(void *))
60
61 extern void tlb_table_flush(struct mmu_gather *tlb);
62 extern void tlb_remove_table(struct mmu_gather *tlb, void *table);
63
64 void tlb_remove_table_sync_one(void);
65
66 #else
67
tlb_remove_table_sync_one(void)68 static inline void tlb_remove_table_sync_one(void) { }
69
70 #endif
71
72 /*
73 * If we can't allocate a page to make a big batch of page pointers
74 * to work on, then just handle a few from the on-stack structure.
75 */
76 #define MMU_GATHER_BUNDLE 8
77
78 struct mmu_gather_batch {
79 struct mmu_gather_batch *next;
80 unsigned int nr;
81 unsigned int max;
82 struct page *pages[0];
83 };
84
85 #define MAX_GATHER_BATCH \
86 ((PAGE_SIZE - sizeof(struct mmu_gather_batch)) / sizeof(void *))
87
88 /*
89 * Limit the maximum number of mmu_gather batches to reduce a risk of soft
90 * lockups for non-preemptible kernels on huge machines when a lot of memory
91 * is zapped during unmapping.
92 * 10K pages freed at once should be safe even without a preemption point.
93 */
94 #define MAX_GATHER_BATCH_COUNT (10000UL/MAX_GATHER_BATCH)
95
96 /* struct mmu_gather is an opaque type used by the mm code for passing around
97 * any data needed by arch specific code for tlb_remove_page.
98 */
99 struct mmu_gather {
100 struct mm_struct *mm;
101 #ifdef CONFIG_HAVE_RCU_TABLE_FREE
102 struct mmu_table_batch *batch;
103 #endif
104 unsigned long start;
105 unsigned long end;
106 /* we are in the middle of an operation to clear
107 * a full mm and can make some optimizations */
108 unsigned int fullmm : 1,
109 /* we have performed an operation which
110 * requires a complete flush of the tlb */
111 need_flush_all : 1;
112
113 struct mmu_gather_batch *active;
114 struct mmu_gather_batch local;
115 struct page *__pages[MMU_GATHER_BUNDLE];
116 unsigned int batch_count;
117 int page_size;
118 };
119
120 #define HAVE_GENERIC_MMU_GATHER
121
122 void arch_tlb_gather_mmu(struct mmu_gather *tlb,
123 struct mm_struct *mm, unsigned long start, unsigned long end);
124 void tlb_flush_mmu(struct mmu_gather *tlb);
125 void arch_tlb_finish_mmu(struct mmu_gather *tlb,
126 unsigned long start, unsigned long end, bool force);
127 void tlb_flush_pmd_range(struct mmu_gather *tlb, unsigned long address,
128 unsigned long size);
129 extern bool __tlb_remove_page_size(struct mmu_gather *tlb, struct page *page,
130 int page_size);
131
__tlb_adjust_range(struct mmu_gather * tlb,unsigned long address,unsigned int range_size)132 static inline void __tlb_adjust_range(struct mmu_gather *tlb,
133 unsigned long address,
134 unsigned int range_size)
135 {
136 tlb->start = min(tlb->start, address);
137 tlb->end = max(tlb->end, address + range_size);
138 }
139
__tlb_reset_range(struct mmu_gather * tlb)140 static inline void __tlb_reset_range(struct mmu_gather *tlb)
141 {
142 if (tlb->fullmm) {
143 tlb->start = tlb->end = ~0;
144 } else {
145 tlb->start = TASK_SIZE;
146 tlb->end = 0;
147 }
148 }
149
tlb_flush_mmu_tlbonly(struct mmu_gather * tlb)150 static inline void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb)
151 {
152 if (!tlb->end)
153 return;
154
155 tlb_flush(tlb);
156 mmu_notifier_invalidate_range(tlb->mm, tlb->start, tlb->end);
157 __tlb_reset_range(tlb);
158 }
159
tlb_remove_page_size(struct mmu_gather * tlb,struct page * page,int page_size)160 static inline void tlb_remove_page_size(struct mmu_gather *tlb,
161 struct page *page, int page_size)
162 {
163 if (__tlb_remove_page_size(tlb, page, page_size))
164 tlb_flush_mmu(tlb);
165 }
166
__tlb_remove_page(struct mmu_gather * tlb,struct page * page)167 static inline bool __tlb_remove_page(struct mmu_gather *tlb, struct page *page)
168 {
169 return __tlb_remove_page_size(tlb, page, PAGE_SIZE);
170 }
171
172 /* tlb_remove_page
173 * Similar to __tlb_remove_page but will call tlb_flush_mmu() itself when
174 * required.
175 */
tlb_remove_page(struct mmu_gather * tlb,struct page * page)176 static inline void tlb_remove_page(struct mmu_gather *tlb, struct page *page)
177 {
178 return tlb_remove_page_size(tlb, page, PAGE_SIZE);
179 }
180
181 #ifndef tlb_remove_check_page_size_change
182 #define tlb_remove_check_page_size_change tlb_remove_check_page_size_change
tlb_remove_check_page_size_change(struct mmu_gather * tlb,unsigned int page_size)183 static inline void tlb_remove_check_page_size_change(struct mmu_gather *tlb,
184 unsigned int page_size)
185 {
186 /*
187 * We don't care about page size change, just update
188 * mmu_gather page size here so that debug checks
189 * doesn't throw false warning.
190 */
191 #ifdef CONFIG_DEBUG_VM
192 tlb->page_size = page_size;
193 #endif
194 }
195 #endif
196
197 /*
198 * In the case of tlb vma handling, we can optimise these away in the
199 * case where we're doing a full MM flush. When we're doing a munmap,
200 * the vmas are adjusted to only cover the region to be torn down.
201 */
202 #ifndef tlb_start_vma
203 #define tlb_start_vma(tlb, vma) do { } while (0)
204 #endif
205
206 #define __tlb_end_vma(tlb, vma) \
207 do { \
208 if (!tlb->fullmm) \
209 tlb_flush_mmu_tlbonly(tlb); \
210 } while (0)
211
212 #ifndef tlb_end_vma
213 #define tlb_end_vma __tlb_end_vma
214 #endif
215
216 #ifndef __tlb_remove_tlb_entry
217 #define __tlb_remove_tlb_entry(tlb, ptep, address) do { } while (0)
218 #endif
219
220 /**
221 * tlb_remove_tlb_entry - remember a pte unmapping for later tlb invalidation.
222 *
223 * Record the fact that pte's were really unmapped by updating the range,
224 * so we can later optimise away the tlb invalidate. This helps when
225 * userspace is unmapping already-unmapped pages, which happens quite a lot.
226 */
227 #define tlb_remove_tlb_entry(tlb, ptep, address) \
228 do { \
229 __tlb_adjust_range(tlb, address, PAGE_SIZE); \
230 __tlb_remove_tlb_entry(tlb, ptep, address); \
231 } while (0)
232
233 #define tlb_remove_huge_tlb_entry(h, tlb, ptep, address) \
234 do { \
235 __tlb_adjust_range(tlb, address, huge_page_size(h)); \
236 __tlb_remove_tlb_entry(tlb, ptep, address); \
237 } while (0)
238
239 /**
240 * tlb_remove_pmd_tlb_entry - remember a pmd mapping for later tlb invalidation
241 * This is a nop so far, because only x86 needs it.
242 */
243 #ifndef __tlb_remove_pmd_tlb_entry
244 #define __tlb_remove_pmd_tlb_entry(tlb, pmdp, address) do {} while (0)
245 #endif
246
247 #define tlb_remove_pmd_tlb_entry(tlb, pmdp, address) \
248 do { \
249 __tlb_adjust_range(tlb, address, HPAGE_PMD_SIZE); \
250 __tlb_remove_pmd_tlb_entry(tlb, pmdp, address); \
251 } while (0)
252
253 /**
254 * tlb_remove_pud_tlb_entry - remember a pud mapping for later tlb
255 * invalidation. This is a nop so far, because only x86 needs it.
256 */
257 #ifndef __tlb_remove_pud_tlb_entry
258 #define __tlb_remove_pud_tlb_entry(tlb, pudp, address) do {} while (0)
259 #endif
260
261 #define tlb_remove_pud_tlb_entry(tlb, pudp, address) \
262 do { \
263 __tlb_adjust_range(tlb, address, HPAGE_PUD_SIZE); \
264 __tlb_remove_pud_tlb_entry(tlb, pudp, address); \
265 } while (0)
266
267 /*
268 * For things like page tables caches (ie caching addresses "inside" the
269 * page tables, like x86 does), for legacy reasons, flushing an
270 * individual page had better flush the page table caches behind it. This
271 * is definitely how x86 works, for example. And if you have an
272 * architected non-legacy page table cache (which I'm not aware of
273 * anybody actually doing), you're going to have some architecturally
274 * explicit flushing for that, likely *separate* from a regular TLB entry
275 * flush, and thus you'd need more than just some range expansion..
276 *
277 * So if we ever find an architecture
278 * that would want something that odd, I think it is up to that
279 * architecture to do its own odd thing, not cause pain for others
280 * http://lkml.kernel.org/r/CA+55aFzBggoXtNXQeng5d_mRoDnaMBE5Y+URs+PHR67nUpMtaw@mail.gmail.com
281 *
282 * For now w.r.t page table cache, mark the range_size as PAGE_SIZE
283 */
284
285 #ifndef pte_free_tlb
286 #define pte_free_tlb(tlb, ptep, address) \
287 do { \
288 __tlb_adjust_range(tlb, address, PAGE_SIZE); \
289 __pte_free_tlb(tlb, ptep, address); \
290 } while (0)
291 #endif
292
293 #ifndef pmd_free_tlb
294 #define pmd_free_tlb(tlb, pmdp, address) \
295 do { \
296 __tlb_adjust_range(tlb, address, PAGE_SIZE); \
297 __pmd_free_tlb(tlb, pmdp, address); \
298 } while (0)
299 #endif
300
301 #ifndef __ARCH_HAS_4LEVEL_HACK
302 #ifndef pud_free_tlb
303 #define pud_free_tlb(tlb, pudp, address) \
304 do { \
305 __tlb_adjust_range(tlb, address, PAGE_SIZE); \
306 __pud_free_tlb(tlb, pudp, address); \
307 } while (0)
308 #endif
309 #endif
310
311 #ifndef __ARCH_HAS_5LEVEL_HACK
312 #ifndef p4d_free_tlb
313 #define p4d_free_tlb(tlb, pudp, address) \
314 do { \
315 __tlb_adjust_range(tlb, address, PAGE_SIZE); \
316 __p4d_free_tlb(tlb, pudp, address); \
317 } while (0)
318 #endif
319 #endif
320
321 #define tlb_migrate_finish(mm) do {} while (0)
322
323 #endif /* _ASM_GENERIC__TLB_H */
324