Linux Audio

Check our new training course

Loading...
v3.1
 
  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 <pzijlstr@redhat.com>
  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/swap.h>
 19#include <asm/pgalloc.h>
 20#include <asm/tlbflush.h>
 
 21
 22#ifdef CONFIG_HAVE_RCU_TABLE_FREE
 23/*
 24 * Semi RCU freeing of the page directories.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 25 *
 26 * This is needed by some architectures to implement software pagetable walkers.
 
 27 *
 28 * gup_fast() and other software pagetable walkers do a lockless page-table
 29 * walk and therefore needs some synchronization with the freeing of the page
 30 * directories. The chosen means to accomplish that is by disabling IRQs over
 31 * the walk.
 32 *
 33 * Architectures that use IPIs to flush TLBs will then automagically DTRT,
 34 * since we unlink the page, flush TLBs, free the page. Since the disabling of
 35 * IRQs delays the completion of the TLB flush we can never observe an already
 36 * freed page.
 37 *
 38 * Architectures that do not have this (PPC) need to delay the freeing by some
 39 * other means, this is that means.
 40 *
 41 * What we do is batch the freed directory pages (tables) and RCU free them.
 42 * We use the sched RCU variant, as that guarantees that IRQ/preempt disabling
 43 * holds off grace periods.
 44 *
 45 * However, in order to batch these pages we need to allocate storage, this
 46 * allocation is deep inside the MM code and can thus easily fail on memory
 47 * pressure. To guarantee progress we fall back to single table freeing, see
 48 * the implementation of tlb_remove_table_one().
 49 *
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 50 */
 
 
 
 51struct mmu_table_batch {
 
 52	struct rcu_head		rcu;
 
 53	unsigned int		nr;
 54	void			*tables[0];
 55};
 56
 57#define MAX_TABLE_BATCH		\
 58	((PAGE_SIZE - sizeof(struct mmu_table_batch)) / sizeof(void *))
 59
 60extern void tlb_table_flush(struct mmu_gather *tlb);
 61extern void tlb_remove_table(struct mmu_gather *tlb, void *table);
 62
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 63#endif
 64
 
 
 
 
 
 
 
 
 
 
 65/*
 66 * If we can't allocate a page to make a big batch of page pointers
 67 * to work on, then just handle a few from the on-stack structure.
 68 */
 69#define MMU_GATHER_BUNDLE	8
 70
 71struct mmu_gather_batch {
 72	struct mmu_gather_batch	*next;
 73	unsigned int		nr;
 74	unsigned int		max;
 75	struct page		*pages[0];
 76};
 77
 78#define MAX_GATHER_BATCH	\
 79	((PAGE_SIZE - sizeof(struct mmu_gather_batch)) / sizeof(void *))
 80
 81/* struct mmu_gather is an opaque type used by the mm code for passing around
 
 
 
 
 
 
 
 
 
 
 
 
 
 82 * any data needed by arch specific code for tlb_remove_page.
 83 */
 84struct mmu_gather {
 85	struct mm_struct	*mm;
 86#ifdef CONFIG_HAVE_RCU_TABLE_FREE
 
 87	struct mmu_table_batch	*batch;
 88#endif
 89	unsigned int		need_flush : 1,	/* Did free PTEs */
 90				fast_mode  : 1; /* No batching   */
 91
 92	unsigned int		fullmm;
 
 
 
 
 
 
 
 
 
 
 
 
 93
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 94	struct mmu_gather_batch *active;
 95	struct mmu_gather_batch	local;
 96	struct page		*__pages[MMU_GATHER_BUNDLE];
 
 
 
 
 
 97};
 98
 99#define HAVE_GENERIC_MMU_GATHER
100
101static inline int tlb_fast_mode(struct mmu_gather *tlb)
 
 
102{
103#ifdef CONFIG_SMP
104	return tlb->fast_mode;
105#else
 
 
 
 
 
 
 
 
 
 
 
 
 
 
106	/*
107	 * For UP we don't need to worry about TLB flush
108	 * and page free order so much..
 
109	 */
110	return 1;
 
 
 
 
 
111#endif
 
 
 
 
 
 
 
 
 
 
 
 
 
112}
113
114void tlb_gather_mmu(struct mmu_gather *tlb, struct mm_struct *mm, bool fullmm);
115void tlb_flush_mmu(struct mmu_gather *tlb);
116void tlb_finish_mmu(struct mmu_gather *tlb, unsigned long start, unsigned long end);
117int __tlb_remove_page(struct mmu_gather *tlb, struct page *page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
118
119/* tlb_remove_page
120 *	Similar to __tlb_remove_page but will call tlb_flush_mmu() itself when
121 *	required.
122 */
123static inline void tlb_remove_page(struct mmu_gather *tlb, struct page *page)
124{
125	if (!__tlb_remove_page(tlb, page))
126		tlb_flush_mmu(tlb);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
127}
128
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
129/**
130 * tlb_remove_tlb_entry - remember a pte unmapping for later tlb invalidation.
131 *
132 * Record the fact that pte's were really umapped in ->need_flush, so we can
133 * later optimise away the tlb invalidate.   This helps when userspace is
134 * unmapping already-unmapped pages, which happens quite a lot.
135 */
136#define tlb_remove_tlb_entry(tlb, ptep, address)		\
137	do {							\
138		tlb->need_flush = 1;				\
 
 
 
 
 
 
 
 
 
 
139		__tlb_remove_tlb_entry(tlb, ptep, address);	\
140	} while (0)
141
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
142#define pte_free_tlb(tlb, ptep, address)			\
143	do {							\
144		tlb->need_flush = 1;				\
 
145		__pte_free_tlb(tlb, ptep, address);		\
146	} while (0)
 
147
148#ifndef __ARCH_HAS_4LEVEL_HACK
 
 
 
 
 
 
 
 
 
149#define pud_free_tlb(tlb, pudp, address)			\
150	do {							\
151		tlb->need_flush = 1;				\
 
152		__pud_free_tlb(tlb, pudp, address);		\
153	} while (0)
154#endif
155
156#define pmd_free_tlb(tlb, pmdp, address)			\
 
157	do {							\
158		tlb->need_flush = 1;				\
159		__pmd_free_tlb(tlb, pmdp, address);		\
 
160	} while (0)
 
161
162#define tlb_migrate_finish(mm) do {} while (0)
163
164#endif /* _ASM_GENERIC__TLB_H */
v5.9
  1/* SPDX-License-Identifier: GPL-2.0-or-later */
  2/* include/asm-generic/tlb.h
  3 *
  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 *
  9 * Copyright 2011 Red Hat, Inc., Peter Zijlstra
 
 
 
 
 
 10 */
 11#ifndef _ASM_GENERIC__TLB_H
 12#define _ASM_GENERIC__TLB_H
 13
 14#include <linux/mmu_notifier.h>
 15#include <linux/swap.h>
 16#include <linux/hugetlb_inline.h>
 17#include <asm/tlbflush.h>
 18#include <asm/cacheflush.h>
 19
 
 20/*
 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
 29#ifdef CONFIG_MMU
 30
 31/*
 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 *
 49 *  - tlb_gather_mmu() / tlb_finish_mmu(); start and finish a mmu_gather
 50 *
 51 *    Finish in particular will issue a (final) TLB invalidate and free
 52 *    all (remaining) queued pages.
 53 *
 54 *  - tlb_start_vma() / tlb_end_vma(); marks the start / end of a VMA
 55 *
 56 *    Defaults to flushing at tlb_end_vma() to reset the range; helps when
 57 *    there's large holes between the VMAs.
 58 *
 59 *  - tlb_remove_table()
 60 *
 61 *    tlb_remove_table() is the basic primitive to free page-table directories
 62 *    (__p*_free_tlb()).  In it's most primitive form it is an alias for
 63 *    tlb_remove_page() below, for when page directories are pages and have no
 64 *    additional constraints.
 65 *
 66 *    See also MMU_GATHER_TABLE_FREE and MMU_GATHER_RCU_TABLE_FREE.
 67 *
 68 *  - tlb_remove_page() / __tlb_remove_page()
 69 *  - tlb_remove_page_size() / __tlb_remove_page_size()
 70 *
 71 *    __tlb_remove_page_size() is the basic primitive that queues a page for
 72 *    freeing. __tlb_remove_page() assumes PAGE_SIZE. Both will return a
 73 *    boolean indicating if the queue is (now) full and a call to
 74 *    tlb_flush_mmu() is required.
 75 *
 76 *    tlb_remove_page() and tlb_remove_page_size() imply the call to
 77 *    tlb_flush_mmu() when required and has no return value.
 78 *
 79 *  - tlb_change_page_size()
 80 *
 81 *    call before __tlb_remove_page*() to set the current page-size; implies a
 82 *    possible tlb_flush_mmu() call.
 83 *
 84 *  - tlb_flush_mmu() / tlb_flush_mmu_tlbonly()
 85 *
 86 *    tlb_flush_mmu_tlbonly() - does the TLB invalidate (and resets
 87 *                              related state, like the range)
 88 *
 89 *    tlb_flush_mmu() - in addition to the above TLB invalidate, also frees
 90 *			whatever pages are still batched.
 91 *
 92 *  - mmu_gather::fullmm
 93 *
 94 *    A flag set by tlb_gather_mmu() to indicate we're going to free
 95 *    the entire mm; this allows a number of optimizations.
 96 *
 97 *    - We can ignore tlb_{start,end}_vma(); because we don't
 98 *      care about ranges. Everything will be shot down.
 
 
 99 *
100 *    - (RISC) architectures that use ASIDs can cycle to a new ASID
101 *      and delay the invalidation until ASID space runs out.
 
 
102 *
103 *  - mmu_gather::need_flush_all
 
104 *
105 *    A flag that can be set by the arch code if it wants to force
106 *    flush the entire TLB irrespective of the range. For instance
107 *    x86-PAE needs this when changing top-level entries.
108 *
109 * And allows the architecture to provide and implement tlb_flush():
 
 
 
110 *
111 * tlb_flush() may, in addition to the above mentioned mmu_gather fields, make
112 * use of:
113 *
114 *  - mmu_gather::start / mmu_gather::end
115 *
116 *    which provides the range that needs to be flushed to cover the pages to
117 *    be freed.
118 *
119 *  - mmu_gather::freed_tables
120 *
121 *    set when we freed page table pages
122 *
123 *  - tlb_get_unmap_shift() / tlb_get_unmap_size()
124 *
125 *    returns the smallest TLB entry size unmapped in this range.
126 *
127 * If an architecture does not provide tlb_flush() a default implementation
128 * based on flush_tlb_range() will be used, unless MMU_GATHER_NO_RANGE is
129 * specified, in which case we'll default to flush_tlb_mm().
130 *
131 * Additionally there are a few opt-in features:
132 *
133 *  MMU_GATHER_PAGE_SIZE
134 *
135 *  This ensures we call tlb_flush() every time tlb_change_page_size() actually
136 *  changes the size and provides mmu_gather::page_size to tlb_flush().
137 *
138 *  This might be useful if your architecture has size specific TLB
139 *  invalidation instructions.
140 *
141 *  MMU_GATHER_TABLE_FREE
142 *
143 *  This provides tlb_remove_table(), to be used instead of tlb_remove_page()
144 *  for page directores (__p*_free_tlb()).
145 *
146 *  Useful if your architecture has non-page page directories.
147 *
148 *  When used, an architecture is expected to provide __tlb_remove_table()
149 *  which does the actual freeing of these pages.
150 *
151 *  MMU_GATHER_RCU_TABLE_FREE
152 *
153 *  Like MMU_GATHER_TABLE_FREE, and adds semi-RCU semantics to the free (see
154 *  comment below).
155 *
156 *  Useful if your architecture doesn't use IPIs for remote TLB invalidates
157 *  and therefore doesn't naturally serialize with software page-table walkers.
158 *
159 *  MMU_GATHER_NO_RANGE
160 *
161 *  Use this if your architecture lacks an efficient flush_tlb_range().
162 *
163 *  MMU_GATHER_NO_GATHER
164 *
165 *  If the option is set the mmu_gather will not track individual pages for
166 *  delayed page free anymore. A platform that enables the option needs to
167 *  provide its own implementation of the __tlb_remove_page_size() function to
168 *  free pages.
169 *
170 *  This is useful if your architecture already flushes TLB entries in the
171 *  various ptep_get_and_clear() functions.
172 */
173
174#ifdef CONFIG_MMU_GATHER_TABLE_FREE
175
176struct mmu_table_batch {
177#ifdef CONFIG_MMU_GATHER_RCU_TABLE_FREE
178	struct rcu_head		rcu;
179#endif
180	unsigned int		nr;
181	void			*tables[0];
182};
183
184#define MAX_TABLE_BATCH		\
185	((PAGE_SIZE - sizeof(struct mmu_table_batch)) / sizeof(void *))
186
 
187extern void tlb_remove_table(struct mmu_gather *tlb, void *table);
188
189#else /* !CONFIG_MMU_GATHER_HAVE_TABLE_FREE */
190
191/*
192 * Without MMU_GATHER_TABLE_FREE the architecture is assumed to have page based
193 * page directories and we can use the normal page batching to free them.
194 */
195#define tlb_remove_table(tlb, page) tlb_remove_page((tlb), (page))
196
197#endif /* CONFIG_MMU_GATHER_TABLE_FREE */
198
199#ifdef CONFIG_MMU_GATHER_RCU_TABLE_FREE
200/*
201 * This allows an architecture that does not use the linux page-tables for
202 * hardware to skip the TLBI when freeing page tables.
203 */
204#ifndef tlb_needs_table_invalidate
205#define tlb_needs_table_invalidate() (true)
206#endif
207
208#else
209
210#ifdef tlb_needs_table_invalidate
211#error tlb_needs_table_invalidate() requires MMU_GATHER_RCU_TABLE_FREE
212#endif
213
214#endif /* CONFIG_MMU_GATHER_RCU_TABLE_FREE */
215
216
217#ifndef CONFIG_MMU_GATHER_NO_GATHER
218/*
219 * If we can't allocate a page to make a big batch of page pointers
220 * to work on, then just handle a few from the on-stack structure.
221 */
222#define MMU_GATHER_BUNDLE	8
223
224struct mmu_gather_batch {
225	struct mmu_gather_batch	*next;
226	unsigned int		nr;
227	unsigned int		max;
228	struct page		*pages[0];
229};
230
231#define MAX_GATHER_BATCH	\
232	((PAGE_SIZE - sizeof(struct mmu_gather_batch)) / sizeof(void *))
233
234/*
235 * Limit the maximum number of mmu_gather batches to reduce a risk of soft
236 * lockups for non-preemptible kernels on huge machines when a lot of memory
237 * is zapped during unmapping.
238 * 10K pages freed at once should be safe even without a preemption point.
239 */
240#define MAX_GATHER_BATCH_COUNT	(10000UL/MAX_GATHER_BATCH)
241
242extern bool __tlb_remove_page_size(struct mmu_gather *tlb, struct page *page,
243				   int page_size);
244#endif
245
246/*
247 * struct mmu_gather is an opaque type used by the mm code for passing around
248 * any data needed by arch specific code for tlb_remove_page.
249 */
250struct mmu_gather {
251	struct mm_struct	*mm;
252
253#ifdef CONFIG_MMU_GATHER_TABLE_FREE
254	struct mmu_table_batch	*batch;
255#endif
 
 
256
257	unsigned long		start;
258	unsigned long		end;
259	/*
260	 * we are in the middle of an operation to clear
261	 * a full mm and can make some optimizations
262	 */
263	unsigned int		fullmm : 1;
264
265	/*
266	 * we have performed an operation which
267	 * requires a complete flush of the tlb
268	 */
269	unsigned int		need_flush_all : 1;
270
271	/*
272	 * we have removed page directories
273	 */
274	unsigned int		freed_tables : 1;
275
276	/*
277	 * at which levels have we cleared entries?
278	 */
279	unsigned int		cleared_ptes : 1;
280	unsigned int		cleared_pmds : 1;
281	unsigned int		cleared_puds : 1;
282	unsigned int		cleared_p4ds : 1;
283
284	/*
285	 * tracks VM_EXEC | VM_HUGETLB in tlb_start_vma
286	 */
287	unsigned int		vma_exec : 1;
288	unsigned int		vma_huge : 1;
289
290	unsigned int		batch_count;
291
292#ifndef CONFIG_MMU_GATHER_NO_GATHER
293	struct mmu_gather_batch *active;
294	struct mmu_gather_batch	local;
295	struct page		*__pages[MMU_GATHER_BUNDLE];
296
297#ifdef CONFIG_MMU_GATHER_PAGE_SIZE
298	unsigned int page_size;
299#endif
300#endif
301};
302
303void tlb_flush_mmu(struct mmu_gather *tlb);
304
305static inline void __tlb_adjust_range(struct mmu_gather *tlb,
306				      unsigned long address,
307				      unsigned int range_size)
308{
309	tlb->start = min(tlb->start, address);
310	tlb->end = max(tlb->end, address + range_size);
311}
312
313static inline void __tlb_reset_range(struct mmu_gather *tlb)
314{
315	if (tlb->fullmm) {
316		tlb->start = tlb->end = ~0;
317	} else {
318		tlb->start = TASK_SIZE;
319		tlb->end = 0;
320	}
321	tlb->freed_tables = 0;
322	tlb->cleared_ptes = 0;
323	tlb->cleared_pmds = 0;
324	tlb->cleared_puds = 0;
325	tlb->cleared_p4ds = 0;
326	/*
327	 * Do not reset mmu_gather::vma_* fields here, we do not
328	 * call into tlb_start_vma() again to set them if there is an
329	 * intermediate flush.
330	 */
331}
332
333#ifdef CONFIG_MMU_GATHER_NO_RANGE
334
335#if defined(tlb_flush) || defined(tlb_start_vma) || defined(tlb_end_vma)
336#error MMU_GATHER_NO_RANGE relies on default tlb_flush(), tlb_start_vma() and tlb_end_vma()
337#endif
338
339/*
340 * When an architecture does not have efficient means of range flushing TLBs
341 * there is no point in doing intermediate flushes on tlb_end_vma() to keep the
342 * range small. We equally don't have to worry about page granularity or other
343 * things.
344 *
345 * All we need to do is issue a full flush for any !0 range.
346 */
347static inline void tlb_flush(struct mmu_gather *tlb)
348{
349	if (tlb->end)
350		flush_tlb_mm(tlb->mm);
351}
352
353static inline void
354tlb_update_vma_flags(struct mmu_gather *tlb, struct vm_area_struct *vma) { }
355
356#define tlb_end_vma tlb_end_vma
357static inline void tlb_end_vma(struct mmu_gather *tlb, struct vm_area_struct *vma) { }
358
359#else /* CONFIG_MMU_GATHER_NO_RANGE */
360
361#ifndef tlb_flush
362
363#if defined(tlb_start_vma) || defined(tlb_end_vma)
364#error Default tlb_flush() relies on default tlb_start_vma() and tlb_end_vma()
365#endif
366
367/*
368 * When an architecture does not provide its own tlb_flush() implementation
369 * but does have a reasonably efficient flush_vma_range() implementation
370 * use that.
371 */
372static inline void tlb_flush(struct mmu_gather *tlb)
373{
374	if (tlb->fullmm || tlb->need_flush_all) {
375		flush_tlb_mm(tlb->mm);
376	} else if (tlb->end) {
377		struct vm_area_struct vma = {
378			.vm_mm = tlb->mm,
379			.vm_flags = (tlb->vma_exec ? VM_EXEC    : 0) |
380				    (tlb->vma_huge ? VM_HUGETLB : 0),
381		};
382
383		flush_tlb_range(&vma, tlb->start, tlb->end);
384	}
385}
386
387static inline void
388tlb_update_vma_flags(struct mmu_gather *tlb, struct vm_area_struct *vma)
389{
390	/*
391	 * flush_tlb_range() implementations that look at VM_HUGETLB (tile,
392	 * mips-4k) flush only large pages.
393	 *
394	 * flush_tlb_range() implementations that flush I-TLB also flush D-TLB
395	 * (tile, xtensa, arm), so it's ok to just add VM_EXEC to an existing
396	 * range.
397	 *
398	 * We rely on tlb_end_vma() to issue a flush, such that when we reset
399	 * these values the batch is empty.
400	 */
401	tlb->vma_huge = is_vm_hugetlb_page(vma);
402	tlb->vma_exec = !!(vma->vm_flags & VM_EXEC);
403}
404
405#else
406
407static inline void
408tlb_update_vma_flags(struct mmu_gather *tlb, struct vm_area_struct *vma) { }
409
410#endif
411
412#endif /* CONFIG_MMU_GATHER_NO_RANGE */
413
414static inline void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb)
415{
416	/*
417	 * Anything calling __tlb_adjust_range() also sets at least one of
418	 * these bits.
419	 */
420	if (!(tlb->freed_tables || tlb->cleared_ptes || tlb->cleared_pmds ||
421	      tlb->cleared_puds || tlb->cleared_p4ds))
422		return;
423
424	tlb_flush(tlb);
425	mmu_notifier_invalidate_range(tlb->mm, tlb->start, tlb->end);
426	__tlb_reset_range(tlb);
427}
428
429static inline void tlb_remove_page_size(struct mmu_gather *tlb,
430					struct page *page, int page_size)
431{
432	if (__tlb_remove_page_size(tlb, page, page_size))
433		tlb_flush_mmu(tlb);
434}
435
436static inline bool __tlb_remove_page(struct mmu_gather *tlb, struct page *page)
437{
438	return __tlb_remove_page_size(tlb, page, PAGE_SIZE);
439}
440
441/* tlb_remove_page
442 *	Similar to __tlb_remove_page but will call tlb_flush_mmu() itself when
443 *	required.
444 */
445static inline void tlb_remove_page(struct mmu_gather *tlb, struct page *page)
446{
447	return tlb_remove_page_size(tlb, page, PAGE_SIZE);
448}
449
450static inline void tlb_change_page_size(struct mmu_gather *tlb,
451						     unsigned int page_size)
452{
453#ifdef CONFIG_MMU_GATHER_PAGE_SIZE
454	if (tlb->page_size && tlb->page_size != page_size) {
455		if (!tlb->fullmm && !tlb->need_flush_all)
456			tlb_flush_mmu(tlb);
457	}
458
459	tlb->page_size = page_size;
460#endif
461}
462
463static inline unsigned long tlb_get_unmap_shift(struct mmu_gather *tlb)
464{
465	if (tlb->cleared_ptes)
466		return PAGE_SHIFT;
467	if (tlb->cleared_pmds)
468		return PMD_SHIFT;
469	if (tlb->cleared_puds)
470		return PUD_SHIFT;
471	if (tlb->cleared_p4ds)
472		return P4D_SHIFT;
473
474	return PAGE_SHIFT;
475}
476
477static inline unsigned long tlb_get_unmap_size(struct mmu_gather *tlb)
478{
479	return 1UL << tlb_get_unmap_shift(tlb);
480}
481
482/*
483 * In the case of tlb vma handling, we can optimise these away in the
484 * case where we're doing a full MM flush.  When we're doing a munmap,
485 * the vmas are adjusted to only cover the region to be torn down.
486 */
487#ifndef tlb_start_vma
488static inline void tlb_start_vma(struct mmu_gather *tlb, struct vm_area_struct *vma)
489{
490	if (tlb->fullmm)
491		return;
492
493	tlb_update_vma_flags(tlb, vma);
494	flush_cache_range(vma, vma->vm_start, vma->vm_end);
495}
496#endif
497
498#ifndef tlb_end_vma
499static inline void tlb_end_vma(struct mmu_gather *tlb, struct vm_area_struct *vma)
500{
501	if (tlb->fullmm)
502		return;
503
504	/*
505	 * Do a TLB flush and reset the range at VMA boundaries; this avoids
506	 * the ranges growing with the unused space between consecutive VMAs,
507	 * but also the mmu_gather::vma_* flags from tlb_start_vma() rely on
508	 * this.
509	 */
510	tlb_flush_mmu_tlbonly(tlb);
511}
512#endif
513
514/*
515 * tlb_flush_{pte|pmd|pud|p4d}_range() adjust the tlb->start and tlb->end,
516 * and set corresponding cleared_*.
517 */
518static inline void tlb_flush_pte_range(struct mmu_gather *tlb,
519				     unsigned long address, unsigned long size)
520{
521	__tlb_adjust_range(tlb, address, size);
522	tlb->cleared_ptes = 1;
523}
524
525static inline void tlb_flush_pmd_range(struct mmu_gather *tlb,
526				     unsigned long address, unsigned long size)
527{
528	__tlb_adjust_range(tlb, address, size);
529	tlb->cleared_pmds = 1;
530}
531
532static inline void tlb_flush_pud_range(struct mmu_gather *tlb,
533				     unsigned long address, unsigned long size)
534{
535	__tlb_adjust_range(tlb, address, size);
536	tlb->cleared_puds = 1;
537}
538
539static inline void tlb_flush_p4d_range(struct mmu_gather *tlb,
540				     unsigned long address, unsigned long size)
541{
542	__tlb_adjust_range(tlb, address, size);
543	tlb->cleared_p4ds = 1;
544}
545
546#ifndef __tlb_remove_tlb_entry
547#define __tlb_remove_tlb_entry(tlb, ptep, address) do { } while (0)
548#endif
549
550/**
551 * tlb_remove_tlb_entry - remember a pte unmapping for later tlb invalidation.
552 *
553 * Record the fact that pte's were really unmapped by updating the range,
554 * so we can later optimise away the tlb invalidate.   This helps when
555 * userspace is unmapping already-unmapped pages, which happens quite a lot.
556 */
557#define tlb_remove_tlb_entry(tlb, ptep, address)		\
558	do {							\
559		tlb_flush_pte_range(tlb, address, PAGE_SIZE);	\
560		__tlb_remove_tlb_entry(tlb, ptep, address);	\
561	} while (0)
562
563#define tlb_remove_huge_tlb_entry(h, tlb, ptep, address)	\
564	do {							\
565		unsigned long _sz = huge_page_size(h);		\
566		if (_sz == PMD_SIZE)				\
567			tlb_flush_pmd_range(tlb, address, _sz);	\
568		else if (_sz == PUD_SIZE)			\
569			tlb_flush_pud_range(tlb, address, _sz);	\
570		__tlb_remove_tlb_entry(tlb, ptep, address);	\
571	} while (0)
572
573/**
574 * tlb_remove_pmd_tlb_entry - remember a pmd mapping for later tlb invalidation
575 * This is a nop so far, because only x86 needs it.
576 */
577#ifndef __tlb_remove_pmd_tlb_entry
578#define __tlb_remove_pmd_tlb_entry(tlb, pmdp, address) do {} while (0)
579#endif
580
581#define tlb_remove_pmd_tlb_entry(tlb, pmdp, address)			\
582	do {								\
583		tlb_flush_pmd_range(tlb, address, HPAGE_PMD_SIZE);	\
584		__tlb_remove_pmd_tlb_entry(tlb, pmdp, address);		\
585	} while (0)
586
587/**
588 * tlb_remove_pud_tlb_entry - remember a pud mapping for later tlb
589 * invalidation. This is a nop so far, because only x86 needs it.
590 */
591#ifndef __tlb_remove_pud_tlb_entry
592#define __tlb_remove_pud_tlb_entry(tlb, pudp, address) do {} while (0)
593#endif
594
595#define tlb_remove_pud_tlb_entry(tlb, pudp, address)			\
596	do {								\
597		tlb_flush_pud_range(tlb, address, HPAGE_PUD_SIZE);	\
598		__tlb_remove_pud_tlb_entry(tlb, pudp, address);		\
599	} while (0)
600
601/*
602 * For things like page tables caches (ie caching addresses "inside" the
603 * page tables, like x86 does), for legacy reasons, flushing an
604 * individual page had better flush the page table caches behind it. This
605 * is definitely how x86 works, for example. And if you have an
606 * architected non-legacy page table cache (which I'm not aware of
607 * anybody actually doing), you're going to have some architecturally
608 * explicit flushing for that, likely *separate* from a regular TLB entry
609 * flush, and thus you'd need more than just some range expansion..
610 *
611 * So if we ever find an architecture
612 * that would want something that odd, I think it is up to that
613 * architecture to do its own odd thing, not cause pain for others
614 * http://lkml.kernel.org/r/CA+55aFzBggoXtNXQeng5d_mRoDnaMBE5Y+URs+PHR67nUpMtaw@mail.gmail.com
615 *
616 * For now w.r.t page table cache, mark the range_size as PAGE_SIZE
617 */
618
619#ifndef pte_free_tlb
620#define pte_free_tlb(tlb, ptep, address)			\
621	do {							\
622		tlb_flush_pmd_range(tlb, address, PAGE_SIZE);	\
623		tlb->freed_tables = 1;				\
624		__pte_free_tlb(tlb, ptep, address);		\
625	} while (0)
626#endif
627
628#ifndef pmd_free_tlb
629#define pmd_free_tlb(tlb, pmdp, address)			\
630	do {							\
631		tlb_flush_pud_range(tlb, address, PAGE_SIZE);	\
632		tlb->freed_tables = 1;				\
633		__pmd_free_tlb(tlb, pmdp, address);		\
634	} while (0)
635#endif
636
637#ifndef pud_free_tlb
638#define pud_free_tlb(tlb, pudp, address)			\
639	do {							\
640		tlb_flush_p4d_range(tlb, address, PAGE_SIZE);	\
641		tlb->freed_tables = 1;				\
642		__pud_free_tlb(tlb, pudp, address);		\
643	} while (0)
644#endif
645
646#ifndef p4d_free_tlb
647#define p4d_free_tlb(tlb, pudp, address)			\
648	do {							\
649		__tlb_adjust_range(tlb, address, PAGE_SIZE);	\
650		tlb->freed_tables = 1;				\
651		__p4d_free_tlb(tlb, pudp, address);		\
652	} while (0)
653#endif
654
655#endif /* CONFIG_MMU */
656
657#endif /* _ASM_GENERIC__TLB_H */