blob: fd7feb5c789485864f0e6a596eaa209f04819996 [file] [log] [blame]
Thomas Gleixner2874c5f2019-05-27 08:55:01 +02001/* SPDX-License-Identifier: GPL-2.0-or-later */
Uwe Zeisbergerf30c2262006-10-03 23:01:26 +02002/* include/asm-generic/tlb.h
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 *
4 * Generic TLB shootdown code
5 *
6 * Copyright 2001 Red Hat, Inc.
7 * Based on code from mm/memory.c Copyright Linus Torvalds and others.
8 *
Peter Zijlstra90eec102015-11-16 11:08:45 +01009 * Copyright 2011 Red Hat, Inc., Peter Zijlstra
Linus Torvalds1da177e2005-04-16 15:20:36 -070010 */
11#ifndef _ASM_GENERIC__TLB_H
12#define _ASM_GENERIC__TLB_H
13
Nicholas Pigginfd1102f2018-08-23 18:47:09 +100014#include <linux/mmu_notifier.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/swap.h>
Anshuman Khandual03911132020-04-06 20:03:51 -070016#include <linux/hugetlb_inline.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <asm/tlbflush.h>
Peter Zijlstrae7fd28a2018-08-27 13:00:17 +020018#include <asm/cacheflush.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019
Nadav Amit5932c9f2019-04-25 17:11:42 -070020/*
21 * Blindly accessing user memory from NMI context can be dangerous
22 * if we're in the middle of switching the current user task or switching
23 * the loaded mm.
24 */
25#ifndef nmi_uaccess_okay
26# define nmi_uaccess_okay() true
27#endif
28
Will Deaconfaaadaf2018-08-24 13:28:28 +010029#ifdef CONFIG_MMU
30
Peter Zijlstradea24342018-09-04 10:43:14 +020031/*
32 * Generic MMU-gather implementation.
33 *
34 * The mmu_gather data structure is used by the mm code to implement the
35 * correct and efficient ordering of freeing pages and TLB invalidations.
36 *
37 * This correct ordering is:
38 *
39 * 1) unhook page
40 * 2) TLB invalidate page
41 * 3) free page
42 *
43 * That is, we must never free a page before we have ensured there are no live
44 * translations left to it. Otherwise it might be possible to observe (or
45 * worse, change) the page content after it has been reused.
46 *
47 * The mmu_gather API consists of:
48 *
Will Deacond8b45052021-01-27 23:53:44 +000049 * - tlb_gather_mmu() / tlb_gather_mmu_fullmm() / tlb_finish_mmu()
50 *
51 * start and finish a mmu_gather
Peter Zijlstradea24342018-09-04 10:43:14 +020052 *
53 * Finish in particular will issue a (final) TLB invalidate and free
54 * all (remaining) queued pages.
55 *
56 * - tlb_start_vma() / tlb_end_vma(); marks the start / end of a VMA
57 *
58 * Defaults to flushing at tlb_end_vma() to reset the range; helps when
59 * there's large holes between the VMAs.
60 *
Peter Zijlstra0d6e24d2020-02-03 17:37:11 -080061 * - tlb_remove_table()
62 *
63 * tlb_remove_table() is the basic primitive to free page-table directories
64 * (__p*_free_tlb()). In it's most primitive form it is an alias for
65 * tlb_remove_page() below, for when page directories are pages and have no
66 * additional constraints.
67 *
68 * See also MMU_GATHER_TABLE_FREE and MMU_GATHER_RCU_TABLE_FREE.
69 *
Peter Zijlstradea24342018-09-04 10:43:14 +020070 * - tlb_remove_page() / __tlb_remove_page()
71 * - tlb_remove_page_size() / __tlb_remove_page_size()
72 *
73 * __tlb_remove_page_size() is the basic primitive that queues a page for
74 * freeing. __tlb_remove_page() assumes PAGE_SIZE. Both will return a
75 * boolean indicating if the queue is (now) full and a call to
76 * tlb_flush_mmu() is required.
77 *
78 * tlb_remove_page() and tlb_remove_page_size() imply the call to
79 * tlb_flush_mmu() when required and has no return value.
80 *
Peter Zijlstraed6a7932018-08-31 14:46:08 +020081 * - tlb_change_page_size()
Peter Zijlstradea24342018-09-04 10:43:14 +020082 *
83 * call before __tlb_remove_page*() to set the current page-size; implies a
84 * possible tlb_flush_mmu() call.
85 *
Peter Zijlstrafa0aafb2018-09-20 10:54:04 +020086 * - tlb_flush_mmu() / tlb_flush_mmu_tlbonly()
Peter Zijlstradea24342018-09-04 10:43:14 +020087 *
88 * tlb_flush_mmu_tlbonly() - does the TLB invalidate (and resets
89 * related state, like the range)
90 *
Peter Zijlstrafa0aafb2018-09-20 10:54:04 +020091 * tlb_flush_mmu() - in addition to the above TLB invalidate, also frees
92 * whatever pages are still batched.
Peter Zijlstradea24342018-09-04 10:43:14 +020093 *
94 * - mmu_gather::fullmm
95 *
Will Deacond8b45052021-01-27 23:53:44 +000096 * A flag set by tlb_gather_mmu_fullmm() to indicate we're going to free
Peter Zijlstradea24342018-09-04 10:43:14 +020097 * the entire mm; this allows a number of optimizations.
98 *
99 * - We can ignore tlb_{start,end}_vma(); because we don't
100 * care about ranges. Everything will be shot down.
101 *
102 * - (RISC) architectures that use ASIDs can cycle to a new ASID
103 * and delay the invalidation until ASID space runs out.
104 *
105 * - mmu_gather::need_flush_all
106 *
107 * A flag that can be set by the arch code if it wants to force
108 * flush the entire TLB irrespective of the range. For instance
109 * x86-PAE needs this when changing top-level entries.
110 *
Peter Zijlstra5f307be2018-09-04 13:18:15 +0200111 * And allows the architecture to provide and implement tlb_flush():
Peter Zijlstradea24342018-09-04 10:43:14 +0200112 *
113 * tlb_flush() may, in addition to the above mentioned mmu_gather fields, make
114 * use of:
115 *
116 * - mmu_gather::start / mmu_gather::end
117 *
118 * which provides the range that needs to be flushed to cover the pages to
119 * be freed.
120 *
121 * - mmu_gather::freed_tables
122 *
123 * set when we freed page table pages
124 *
125 * - tlb_get_unmap_shift() / tlb_get_unmap_size()
126 *
Peter Zijlstra5f307be2018-09-04 13:18:15 +0200127 * returns the smallest TLB entry size unmapped in this range.
128 *
129 * If an architecture does not provide tlb_flush() a default implementation
Peter Zijlstraa30e32b2018-10-11 16:51:51 +0200130 * based on flush_tlb_range() will be used, unless MMU_GATHER_NO_RANGE is
131 * specified, in which case we'll default to flush_tlb_mm().
Peter Zijlstradea24342018-09-04 10:43:14 +0200132 *
133 * Additionally there are a few opt-in features:
134 *
Peter Zijlstra3af4bd02020-02-03 17:37:05 -0800135 * MMU_GATHER_PAGE_SIZE
Peter Zijlstraed6a7932018-08-31 14:46:08 +0200136 *
137 * This ensures we call tlb_flush() every time tlb_change_page_size() actually
138 * changes the size and provides mmu_gather::page_size to tlb_flush().
139 *
Peter Zijlstra3af4bd02020-02-03 17:37:05 -0800140 * This might be useful if your architecture has size specific TLB
141 * invalidation instructions.
142 *
Peter Zijlstra0d6e24d2020-02-03 17:37:11 -0800143 * MMU_GATHER_TABLE_FREE
Peter Zijlstradea24342018-09-04 10:43:14 +0200144 *
145 * This provides tlb_remove_table(), to be used instead of tlb_remove_page()
Peter Zijlstra0d6e24d2020-02-03 17:37:11 -0800146 * for page directores (__p*_free_tlb()).
147 *
148 * Useful if your architecture has non-page page directories.
Peter Zijlstradea24342018-09-04 10:43:14 +0200149 *
150 * When used, an architecture is expected to provide __tlb_remove_table()
151 * which does the actual freeing of these pages.
152 *
Peter Zijlstra0d6e24d2020-02-03 17:37:11 -0800153 * MMU_GATHER_RCU_TABLE_FREE
154 *
155 * Like MMU_GATHER_TABLE_FREE, and adds semi-RCU semantics to the free (see
156 * comment below).
157 *
158 * Useful if your architecture doesn't use IPIs for remote TLB invalidates
159 * and therefore doesn't naturally serialize with software page-table walkers.
160 *
Peter Zijlstraa30e32b2018-10-11 16:51:51 +0200161 * MMU_GATHER_NO_RANGE
162 *
163 * Use this if your architecture lacks an efficient flush_tlb_range().
Peter Zijlstra580a5862020-02-03 17:37:08 -0800164 *
165 * MMU_GATHER_NO_GATHER
166 *
167 * If the option is set the mmu_gather will not track individual pages for
168 * delayed page free anymore. A platform that enables the option needs to
169 * provide its own implementation of the __tlb_remove_page_size() function to
170 * free pages.
171 *
172 * This is useful if your architecture already flushes TLB entries in the
173 * various ptep_get_and_clear() functions.
Peter Zijlstradea24342018-09-04 10:43:14 +0200174 */
Peter Zijlstradea24342018-09-04 10:43:14 +0200175
Peter Zijlstra0d6e24d2020-02-03 17:37:11 -0800176#ifdef CONFIG_MMU_GATHER_TABLE_FREE
177
Peter Zijlstra26723912011-05-24 17:12:00 -0700178struct mmu_table_batch {
Peter Zijlstra0d6e24d2020-02-03 17:37:11 -0800179#ifdef CONFIG_MMU_GATHER_RCU_TABLE_FREE
Peter Zijlstra26723912011-05-24 17:12:00 -0700180 struct rcu_head rcu;
Peter Zijlstra0d6e24d2020-02-03 17:37:11 -0800181#endif
Peter Zijlstra26723912011-05-24 17:12:00 -0700182 unsigned int nr;
Gustavo A. R. Silva5224f792022-02-14 19:11:44 -0600183 void *tables[];
Peter Zijlstra26723912011-05-24 17:12:00 -0700184};
185
186#define MAX_TABLE_BATCH \
187 ((PAGE_SIZE - sizeof(struct mmu_table_batch)) / sizeof(void *))
188
Peter Zijlstra26723912011-05-24 17:12:00 -0700189extern void tlb_remove_table(struct mmu_gather *tlb, void *table);
190
Peter Zijlstra0d6e24d2020-02-03 17:37:11 -0800191#else /* !CONFIG_MMU_GATHER_HAVE_TABLE_FREE */
192
193/*
194 * Without MMU_GATHER_TABLE_FREE the architecture is assumed to have page based
195 * page directories and we can use the normal page batching to free them.
196 */
197#define tlb_remove_table(tlb, page) tlb_remove_page((tlb), (page))
198
199#endif /* CONFIG_MMU_GATHER_TABLE_FREE */
200
201#ifdef CONFIG_MMU_GATHER_RCU_TABLE_FREE
Peter Zijlstra0ed13252020-02-03 17:36:49 -0800202/*
203 * This allows an architecture that does not use the linux page-tables for
204 * hardware to skip the TLBI when freeing page tables.
205 */
206#ifndef tlb_needs_table_invalidate
207#define tlb_needs_table_invalidate() (true)
Peter Zijlstra26723912011-05-24 17:12:00 -0700208#endif
209
Peter Zijlstra0ed13252020-02-03 17:36:49 -0800210#else
211
212#ifdef tlb_needs_table_invalidate
Peter Zijlstraff2e6d722020-02-03 17:37:02 -0800213#error tlb_needs_table_invalidate() requires MMU_GATHER_RCU_TABLE_FREE
Peter Zijlstra0ed13252020-02-03 17:36:49 -0800214#endif
215
Peter Zijlstraff2e6d722020-02-03 17:37:02 -0800216#endif /* CONFIG_MMU_GATHER_RCU_TABLE_FREE */
Peter Zijlstra0ed13252020-02-03 17:36:49 -0800217
218
Peter Zijlstra580a5862020-02-03 17:37:08 -0800219#ifndef CONFIG_MMU_GATHER_NO_GATHER
Peter Zijlstrad16dfc52011-05-24 17:11:45 -0700220/*
221 * If we can't allocate a page to make a big batch of page pointers
222 * to work on, then just handle a few from the on-stack structure.
223 */
224#define MMU_GATHER_BUNDLE 8
225
Peter Zijlstrae3032972011-05-24 17:12:01 -0700226struct mmu_gather_batch {
227 struct mmu_gather_batch *next;
228 unsigned int nr;
229 unsigned int max;
Gustavo A. R. Silva5224f792022-02-14 19:11:44 -0600230 struct page *pages[];
Peter Zijlstrae3032972011-05-24 17:12:01 -0700231};
232
233#define MAX_GATHER_BATCH \
234 ((PAGE_SIZE - sizeof(struct mmu_gather_batch)) / sizeof(void *))
235
Michal Hocko53a59fc2013-01-04 15:35:12 -0800236/*
237 * Limit the maximum number of mmu_gather batches to reduce a risk of soft
238 * lockups for non-preemptible kernels on huge machines when a lot of memory
239 * is zapped during unmapping.
240 * 10K pages freed at once should be safe even without a preemption point.
241 */
242#define MAX_GATHER_BATCH_COUNT (10000UL/MAX_GATHER_BATCH)
243
Martin Schwidefsky952a31c2018-09-18 14:51:50 +0200244extern bool __tlb_remove_page_size(struct mmu_gather *tlb, struct page *page,
245 int page_size);
246#endif
247
Peter Zijlstradea24342018-09-04 10:43:14 +0200248/*
249 * struct mmu_gather is an opaque type used by the mm code for passing around
Hugh Dickins15a23ff2005-10-29 18:16:01 -0700250 * any data needed by arch specific code for tlb_remove_page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 */
252struct mmu_gather {
253 struct mm_struct *mm;
Peter Zijlstradea24342018-09-04 10:43:14 +0200254
Peter Zijlstra0d6e24d2020-02-03 17:37:11 -0800255#ifdef CONFIG_MMU_GATHER_TABLE_FREE
Peter Zijlstra26723912011-05-24 17:12:00 -0700256 struct mmu_table_batch *batch;
257#endif
Peter Zijlstradea24342018-09-04 10:43:14 +0200258
Alex Shi597e1c32012-06-28 09:02:21 +0800259 unsigned long start;
260 unsigned long end;
Peter Zijlstra22a61c32018-08-23 20:27:25 +0100261 /*
262 * we are in the middle of an operation to clear
263 * a full mm and can make some optimizations
264 */
265 unsigned int fullmm : 1;
266
267 /*
268 * we have performed an operation which
269 * requires a complete flush of the tlb
270 */
271 unsigned int need_flush_all : 1;
272
273 /*
274 * we have removed page directories
275 */
276 unsigned int freed_tables : 1;
Peter Zijlstrae3032972011-05-24 17:12:01 -0700277
Will Deacona6d60242018-08-23 21:01:46 +0100278 /*
279 * at which levels have we cleared entries?
280 */
281 unsigned int cleared_ptes : 1;
282 unsigned int cleared_pmds : 1;
283 unsigned int cleared_puds : 1;
284 unsigned int cleared_p4ds : 1;
285
Peter Zijlstra5f307be2018-09-04 13:18:15 +0200286 /*
287 * tracks VM_EXEC | VM_HUGETLB in tlb_start_vma
288 */
289 unsigned int vma_exec : 1;
290 unsigned int vma_huge : 1;
291
Peter Zijlstraed6a7932018-08-31 14:46:08 +0200292 unsigned int batch_count;
293
Peter Zijlstra580a5862020-02-03 17:37:08 -0800294#ifndef CONFIG_MMU_GATHER_NO_GATHER
Peter Zijlstrae3032972011-05-24 17:12:01 -0700295 struct mmu_gather_batch *active;
296 struct mmu_gather_batch local;
297 struct page *__pages[MMU_GATHER_BUNDLE];
Peter Zijlstraed6a7932018-08-31 14:46:08 +0200298
Peter Zijlstra3af4bd02020-02-03 17:37:05 -0800299#ifdef CONFIG_MMU_GATHER_PAGE_SIZE
Peter Zijlstraed6a7932018-08-31 14:46:08 +0200300 unsigned int page_size;
301#endif
Martin Schwidefsky952a31c2018-09-18 14:51:50 +0200302#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303};
304
Peter Zijlstra9547d012011-05-24 17:12:14 -0700305void tlb_flush_mmu(struct mmu_gather *tlb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306
Will Deaconfb7332a2014-10-29 10:03:09 +0000307static inline void __tlb_adjust_range(struct mmu_gather *tlb,
Aneesh Kumar K.Vb5bc66b2016-12-12 16:42:34 -0800308 unsigned long address,
309 unsigned int range_size)
Will Deaconfb7332a2014-10-29 10:03:09 +0000310{
311 tlb->start = min(tlb->start, address);
Aneesh Kumar K.Vb5bc66b2016-12-12 16:42:34 -0800312 tlb->end = max(tlb->end, address + range_size);
Will Deaconfb7332a2014-10-29 10:03:09 +0000313}
314
315static inline void __tlb_reset_range(struct mmu_gather *tlb)
316{
Will Deacon721c21c2015-01-12 19:10:55 +0000317 if (tlb->fullmm) {
318 tlb->start = tlb->end = ~0;
319 } else {
320 tlb->start = TASK_SIZE;
321 tlb->end = 0;
322 }
Peter Zijlstra22a61c32018-08-23 20:27:25 +0100323 tlb->freed_tables = 0;
Will Deacona6d60242018-08-23 21:01:46 +0100324 tlb->cleared_ptes = 0;
325 tlb->cleared_pmds = 0;
326 tlb->cleared_puds = 0;
327 tlb->cleared_p4ds = 0;
Peter Zijlstra5f307be2018-09-04 13:18:15 +0200328 /*
329 * Do not reset mmu_gather::vma_* fields here, we do not
330 * call into tlb_start_vma() again to set them if there is an
331 * intermediate flush.
332 */
Will Deaconfb7332a2014-10-29 10:03:09 +0000333}
334
Peter Zijlstraa30e32b2018-10-11 16:51:51 +0200335#ifdef CONFIG_MMU_GATHER_NO_RANGE
336
337#if defined(tlb_flush) || defined(tlb_start_vma) || defined(tlb_end_vma)
338#error MMU_GATHER_NO_RANGE relies on default tlb_flush(), tlb_start_vma() and tlb_end_vma()
339#endif
340
341/*
342 * When an architecture does not have efficient means of range flushing TLBs
343 * there is no point in doing intermediate flushes on tlb_end_vma() to keep the
344 * range small. We equally don't have to worry about page granularity or other
345 * things.
346 *
347 * All we need to do is issue a full flush for any !0 range.
348 */
349static inline void tlb_flush(struct mmu_gather *tlb)
350{
351 if (tlb->end)
352 flush_tlb_mm(tlb->mm);
353}
354
355static inline void
356tlb_update_vma_flags(struct mmu_gather *tlb, struct vm_area_struct *vma) { }
357
358#define tlb_end_vma tlb_end_vma
359static inline void tlb_end_vma(struct mmu_gather *tlb, struct vm_area_struct *vma) { }
360
361#else /* CONFIG_MMU_GATHER_NO_RANGE */
362
Peter Zijlstra5f307be2018-09-04 13:18:15 +0200363#ifndef tlb_flush
364
365#if defined(tlb_start_vma) || defined(tlb_end_vma)
366#error Default tlb_flush() relies on default tlb_start_vma() and tlb_end_vma()
367#endif
368
Peter Zijlstraa30e32b2018-10-11 16:51:51 +0200369/*
370 * When an architecture does not provide its own tlb_flush() implementation
371 * but does have a reasonably efficient flush_vma_range() implementation
372 * use that.
373 */
Peter Zijlstra5f307be2018-09-04 13:18:15 +0200374static inline void tlb_flush(struct mmu_gather *tlb)
375{
376 if (tlb->fullmm || tlb->need_flush_all) {
377 flush_tlb_mm(tlb->mm);
378 } else if (tlb->end) {
379 struct vm_area_struct vma = {
380 .vm_mm = tlb->mm,
381 .vm_flags = (tlb->vma_exec ? VM_EXEC : 0) |
382 (tlb->vma_huge ? VM_HUGETLB : 0),
383 };
384
385 flush_tlb_range(&vma, tlb->start, tlb->end);
386 }
387}
388
389static inline void
390tlb_update_vma_flags(struct mmu_gather *tlb, struct vm_area_struct *vma)
391{
392 /*
393 * flush_tlb_range() implementations that look at VM_HUGETLB (tile,
394 * mips-4k) flush only large pages.
395 *
396 * flush_tlb_range() implementations that flush I-TLB also flush D-TLB
397 * (tile, xtensa, arm), so it's ok to just add VM_EXEC to an existing
398 * range.
399 *
400 * We rely on tlb_end_vma() to issue a flush, such that when we reset
401 * these values the batch is empty.
402 */
Anshuman Khandual03911132020-04-06 20:03:51 -0700403 tlb->vma_huge = is_vm_hugetlb_page(vma);
Peter Zijlstra5f307be2018-09-04 13:18:15 +0200404 tlb->vma_exec = !!(vma->vm_flags & VM_EXEC);
405}
406
407#else
408
409static inline void
410tlb_update_vma_flags(struct mmu_gather *tlb, struct vm_area_struct *vma) { }
411
412#endif
413
Peter Zijlstraa30e32b2018-10-11 16:51:51 +0200414#endif /* CONFIG_MMU_GATHER_NO_RANGE */
415
Nicholas Pigginfd1102f2018-08-23 18:47:09 +1000416static inline void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb)
417{
Peter Zijlstra0758cd82020-02-03 17:36:53 -0800418 /*
419 * Anything calling __tlb_adjust_range() also sets at least one of
420 * these bits.
421 */
422 if (!(tlb->freed_tables || tlb->cleared_ptes || tlb->cleared_pmds ||
423 tlb->cleared_puds || tlb->cleared_p4ds))
Nicholas Pigginfd1102f2018-08-23 18:47:09 +1000424 return;
425
426 tlb_flush(tlb);
427 mmu_notifier_invalidate_range(tlb->mm, tlb->start, tlb->end);
428 __tlb_reset_range(tlb);
429}
430
Aneesh Kumar K.Ve77b0852016-07-26 15:24:12 -0700431static inline void tlb_remove_page_size(struct mmu_gather *tlb,
432 struct page *page, int page_size)
433{
Aneesh Kumar K.V692a68c2016-12-12 16:42:43 -0800434 if (__tlb_remove_page_size(tlb, page, page_size))
Aneesh Kumar K.Ve77b0852016-07-26 15:24:12 -0700435 tlb_flush_mmu(tlb);
Aneesh Kumar K.Ve77b0852016-07-26 15:24:12 -0700436}
437
Aneesh Kumar K.V692a68c2016-12-12 16:42:43 -0800438static inline bool __tlb_remove_page(struct mmu_gather *tlb, struct page *page)
Aneesh Kumar K.Ve77b0852016-07-26 15:24:12 -0700439{
440 return __tlb_remove_page_size(tlb, page, PAGE_SIZE);
441}
442
Aneesh Kumar K.Ve9d55e12016-07-26 15:24:09 -0700443/* tlb_remove_page
444 * Similar to __tlb_remove_page but will call tlb_flush_mmu() itself when
445 * required.
446 */
447static inline void tlb_remove_page(struct mmu_gather *tlb, struct page *page)
448{
Aneesh Kumar K.Ve77b0852016-07-26 15:24:12 -0700449 return tlb_remove_page_size(tlb, page, PAGE_SIZE);
Aneesh Kumar K.Ve9d55e12016-07-26 15:24:09 -0700450}
451
Peter Zijlstraed6a7932018-08-31 14:46:08 +0200452static inline void tlb_change_page_size(struct mmu_gather *tlb,
Aneesh Kumar K.V07e32662016-12-12 16:42:40 -0800453 unsigned int page_size)
454{
Peter Zijlstra3af4bd02020-02-03 17:37:05 -0800455#ifdef CONFIG_MMU_GATHER_PAGE_SIZE
Peter Zijlstraed6a7932018-08-31 14:46:08 +0200456 if (tlb->page_size && tlb->page_size != page_size) {
Aneesh Kumar K.V864edb72019-10-24 13:28:01 +0530457 if (!tlb->fullmm && !tlb->need_flush_all)
Peter Zijlstraed6a7932018-08-31 14:46:08 +0200458 tlb_flush_mmu(tlb);
459 }
460
Aneesh Kumar K.V07e32662016-12-12 16:42:40 -0800461 tlb->page_size = page_size;
462#endif
463}
Aneesh Kumar K.V07e32662016-12-12 16:42:40 -0800464
Will Deacona6d60242018-08-23 21:01:46 +0100465static inline unsigned long tlb_get_unmap_shift(struct mmu_gather *tlb)
466{
467 if (tlb->cleared_ptes)
468 return PAGE_SHIFT;
469 if (tlb->cleared_pmds)
470 return PMD_SHIFT;
471 if (tlb->cleared_puds)
472 return PUD_SHIFT;
473 if (tlb->cleared_p4ds)
474 return P4D_SHIFT;
475
476 return PAGE_SHIFT;
477}
478
479static inline unsigned long tlb_get_unmap_size(struct mmu_gather *tlb)
480{
481 return 1UL << tlb_get_unmap_shift(tlb);
482}
483
Will Deaconfb7332a2014-10-29 10:03:09 +0000484/*
485 * In the case of tlb vma handling, we can optimise these away in the
486 * case where we're doing a full MM flush. When we're doing a munmap,
487 * the vmas are adjusted to only cover the region to be torn down.
488 */
489#ifndef tlb_start_vma
Peter Zijlstra5f307be2018-09-04 13:18:15 +0200490static inline void tlb_start_vma(struct mmu_gather *tlb, struct vm_area_struct *vma)
491{
492 if (tlb->fullmm)
493 return;
494
495 tlb_update_vma_flags(tlb, vma);
496 flush_cache_range(vma, vma->vm_start, vma->vm_end);
497}
Will Deaconfb7332a2014-10-29 10:03:09 +0000498#endif
499
Will Deaconfb7332a2014-10-29 10:03:09 +0000500#ifndef tlb_end_vma
Peter Zijlstra5f307be2018-09-04 13:18:15 +0200501static inline void tlb_end_vma(struct mmu_gather *tlb, struct vm_area_struct *vma)
502{
503 if (tlb->fullmm)
504 return;
505
506 /*
507 * Do a TLB flush and reset the range at VMA boundaries; this avoids
508 * the ranges growing with the unused space between consecutive VMAs,
509 * but also the mmu_gather::vma_* flags from tlb_start_vma() rely on
510 * this.
511 */
512 tlb_flush_mmu_tlbonly(tlb);
513}
Will Deaconfb7332a2014-10-29 10:03:09 +0000514#endif
515
Peter Zijlstra (Intel)2631ed02020-06-25 16:03:12 +0800516/*
517 * tlb_flush_{pte|pmd|pud|p4d}_range() adjust the tlb->start and tlb->end,
518 * and set corresponding cleared_*.
519 */
520static inline void tlb_flush_pte_range(struct mmu_gather *tlb,
521 unsigned long address, unsigned long size)
522{
523 __tlb_adjust_range(tlb, address, size);
524 tlb->cleared_ptes = 1;
525}
526
527static inline void tlb_flush_pmd_range(struct mmu_gather *tlb,
528 unsigned long address, unsigned long size)
529{
530 __tlb_adjust_range(tlb, address, size);
531 tlb->cleared_pmds = 1;
532}
533
534static inline void tlb_flush_pud_range(struct mmu_gather *tlb,
535 unsigned long address, unsigned long size)
536{
537 __tlb_adjust_range(tlb, address, size);
538 tlb->cleared_puds = 1;
539}
540
541static inline void tlb_flush_p4d_range(struct mmu_gather *tlb,
542 unsigned long address, unsigned long size)
543{
544 __tlb_adjust_range(tlb, address, size);
545 tlb->cleared_p4ds = 1;
546}
547
Will Deaconfb7332a2014-10-29 10:03:09 +0000548#ifndef __tlb_remove_tlb_entry
549#define __tlb_remove_tlb_entry(tlb, ptep, address) do { } while (0)
550#endif
551
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552/**
553 * tlb_remove_tlb_entry - remember a pte unmapping for later tlb invalidation.
554 *
Will Deaconfb7332a2014-10-29 10:03:09 +0000555 * Record the fact that pte's were really unmapped by updating the range,
556 * so we can later optimise away the tlb invalidate. This helps when
557 * userspace is unmapping already-unmapped pages, which happens quite a lot.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 */
559#define tlb_remove_tlb_entry(tlb, ptep, address) \
560 do { \
Peter Zijlstra (Intel)2631ed02020-06-25 16:03:12 +0800561 tlb_flush_pte_range(tlb, address, PAGE_SIZE); \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 __tlb_remove_tlb_entry(tlb, ptep, address); \
563 } while (0)
564
Will Deacona6d60242018-08-23 21:01:46 +0100565#define tlb_remove_huge_tlb_entry(h, tlb, ptep, address) \
566 do { \
567 unsigned long _sz = huge_page_size(h); \
Will Deacona6d60242018-08-23 21:01:46 +0100568 if (_sz == PMD_SIZE) \
Peter Zijlstra (Intel)2631ed02020-06-25 16:03:12 +0800569 tlb_flush_pmd_range(tlb, address, _sz); \
Will Deacona6d60242018-08-23 21:01:46 +0100570 else if (_sz == PUD_SIZE) \
Peter Zijlstra (Intel)2631ed02020-06-25 16:03:12 +0800571 tlb_flush_pud_range(tlb, address, _sz); \
Will Deacona6d60242018-08-23 21:01:46 +0100572 __tlb_remove_tlb_entry(tlb, ptep, address); \
Aneesh Kumar K.Vb528e4b2016-12-12 16:42:37 -0800573 } while (0)
574
Shaohua Lif21760b2012-01-12 17:19:16 -0800575/**
576 * tlb_remove_pmd_tlb_entry - remember a pmd mapping for later tlb invalidation
577 * This is a nop so far, because only x86 needs it.
578 */
579#ifndef __tlb_remove_pmd_tlb_entry
580#define __tlb_remove_pmd_tlb_entry(tlb, pmdp, address) do {} while (0)
581#endif
582
Aneesh Kumar K.Vb5bc66b2016-12-12 16:42:34 -0800583#define tlb_remove_pmd_tlb_entry(tlb, pmdp, address) \
584 do { \
Peter Zijlstra (Intel)2631ed02020-06-25 16:03:12 +0800585 tlb_flush_pmd_range(tlb, address, HPAGE_PMD_SIZE); \
Aneesh Kumar K.Vb5bc66b2016-12-12 16:42:34 -0800586 __tlb_remove_pmd_tlb_entry(tlb, pmdp, address); \
Shaohua Lif21760b2012-01-12 17:19:16 -0800587 } while (0)
588
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -0800589/**
590 * tlb_remove_pud_tlb_entry - remember a pud mapping for later tlb
591 * invalidation. This is a nop so far, because only x86 needs it.
592 */
593#ifndef __tlb_remove_pud_tlb_entry
594#define __tlb_remove_pud_tlb_entry(tlb, pudp, address) do {} while (0)
595#endif
596
597#define tlb_remove_pud_tlb_entry(tlb, pudp, address) \
598 do { \
Peter Zijlstra (Intel)2631ed02020-06-25 16:03:12 +0800599 tlb_flush_pud_range(tlb, address, HPAGE_PUD_SIZE); \
Matthew Wilcoxa00cc7d2017-02-24 14:57:02 -0800600 __tlb_remove_pud_tlb_entry(tlb, pudp, address); \
601 } while (0)
602
Aneesh Kumar K.Vb5bc66b2016-12-12 16:42:34 -0800603/*
604 * For things like page tables caches (ie caching addresses "inside" the
605 * page tables, like x86 does), for legacy reasons, flushing an
606 * individual page had better flush the page table caches behind it. This
607 * is definitely how x86 works, for example. And if you have an
608 * architected non-legacy page table cache (which I'm not aware of
609 * anybody actually doing), you're going to have some architecturally
610 * explicit flushing for that, likely *separate* from a regular TLB entry
611 * flush, and thus you'd need more than just some range expansion..
612 *
613 * So if we ever find an architecture
614 * that would want something that odd, I think it is up to that
615 * architecture to do its own odd thing, not cause pain for others
616 * http://lkml.kernel.org/r/CA+55aFzBggoXtNXQeng5d_mRoDnaMBE5Y+URs+PHR67nUpMtaw@mail.gmail.com
617 *
618 * For now w.r.t page table cache, mark the range_size as PAGE_SIZE
619 */
620
Nicholas Piggina90744b2018-07-13 16:59:03 -0700621#ifndef pte_free_tlb
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000622#define pte_free_tlb(tlb, ptep, address) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 do { \
Peter Zijlstra (Intel)2631ed02020-06-25 16:03:12 +0800624 tlb_flush_pmd_range(tlb, address, PAGE_SIZE); \
Will Deacona6d60242018-08-23 21:01:46 +0100625 tlb->freed_tables = 1; \
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000626 __pte_free_tlb(tlb, ptep, address); \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 } while (0)
Nicholas Piggina90744b2018-07-13 16:59:03 -0700628#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629
Nicholas Piggina90744b2018-07-13 16:59:03 -0700630#ifndef pmd_free_tlb
Kirill A. Shutemov048456d2017-03-09 17:24:06 +0300631#define pmd_free_tlb(tlb, pmdp, address) \
632 do { \
Peter Zijlstra (Intel)2631ed02020-06-25 16:03:12 +0800633 tlb_flush_pud_range(tlb, address, PAGE_SIZE); \
Will Deacona6d60242018-08-23 21:01:46 +0100634 tlb->freed_tables = 1; \
Kirill A. Shutemov048456d2017-03-09 17:24:06 +0300635 __pmd_free_tlb(tlb, pmdp, address); \
636 } while (0)
Nicholas Piggina90744b2018-07-13 16:59:03 -0700637#endif
Kirill A. Shutemov048456d2017-03-09 17:24:06 +0300638
Nicholas Piggina90744b2018-07-13 16:59:03 -0700639#ifndef pud_free_tlb
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000640#define pud_free_tlb(tlb, pudp, address) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 do { \
Peter Zijlstra (Intel)2631ed02020-06-25 16:03:12 +0800642 tlb_flush_p4d_range(tlb, address, PAGE_SIZE); \
Will Deacona6d60242018-08-23 21:01:46 +0100643 tlb->freed_tables = 1; \
Benjamin Herrenschmidt9e1b32c2009-07-22 15:44:28 +1000644 __pud_free_tlb(tlb, pudp, address); \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 } while (0)
646#endif
647
Nicholas Piggina90744b2018-07-13 16:59:03 -0700648#ifndef p4d_free_tlb
Kirill A. Shutemov048456d2017-03-09 17:24:06 +0300649#define p4d_free_tlb(tlb, pudp, address) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 do { \
Peter Zijlstra22a61c32018-08-23 20:27:25 +0100651 __tlb_adjust_range(tlb, address, PAGE_SIZE); \
Will Deacona6d60242018-08-23 21:01:46 +0100652 tlb->freed_tables = 1; \
Kirill A. Shutemov048456d2017-03-09 17:24:06 +0300653 __p4d_free_tlb(tlb, pudp, address); \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 } while (0)
Kirill A. Shutemov048456d2017-03-09 17:24:06 +0300655#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
Will Deaconfaaadaf2018-08-24 13:28:28 +0100657#endif /* CONFIG_MMU */
658
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659#endif /* _ASM_GENERIC__TLB_H */