1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /* include/asm-generic/tlb.h
4 * Generic TLB shootdown code
6 * Copyright 2001 Red Hat, Inc.
7 * Based on code from mm/memory.c Copyright Linus Torvalds and others.
9 * Copyright 2011 Red Hat, Inc., Peter Zijlstra
11 #ifndef _ASM_GENERIC__TLB_H
12 #define _ASM_GENERIC__TLB_H
14 #include <linux/mmu_notifier.h>
15 #include <linux/swap.h>
16 #include <linux/hugetlb_inline.h>
17 #include <asm/pgalloc.h>
18 #include <asm/tlbflush.h>
19 #include <asm/cacheflush.h>
22 * Blindly accessing user memory from NMI context can be dangerous
23 * if we're in the middle of switching the current user task or switching
26 #ifndef nmi_uaccess_okay
27 # define nmi_uaccess_okay() true
33 * Generic MMU-gather implementation.
35 * The mmu_gather data structure is used by the mm code to implement the
36 * correct and efficient ordering of freeing pages and TLB invalidations.
38 * This correct ordering is:
41 * 2) TLB invalidate page
44 * That is, we must never free a page before we have ensured there are no live
45 * translations left to it. Otherwise it might be possible to observe (or
46 * worse, change) the page content after it has been reused.
48 * The mmu_gather API consists of:
50 * - tlb_gather_mmu() / tlb_finish_mmu(); start and finish a mmu_gather
52 * Finish in particular will issue a (final) TLB invalidate and free
53 * all (remaining) queued pages.
55 * - tlb_start_vma() / tlb_end_vma(); marks the start / end of a VMA
57 * Defaults to flushing at tlb_end_vma() to reset the range; helps when
58 * there's large holes between the VMAs.
60 * - tlb_remove_table()
62 * tlb_remove_table() is the basic primitive to free page-table directories
63 * (__p*_free_tlb()). In it's most primitive form it is an alias for
64 * tlb_remove_page() below, for when page directories are pages and have no
65 * additional constraints.
67 * See also MMU_GATHER_TABLE_FREE and MMU_GATHER_RCU_TABLE_FREE.
69 * - tlb_remove_page() / __tlb_remove_page()
70 * - tlb_remove_page_size() / __tlb_remove_page_size()
72 * __tlb_remove_page_size() is the basic primitive that queues a page for
73 * freeing. __tlb_remove_page() assumes PAGE_SIZE. Both will return a
74 * boolean indicating if the queue is (now) full and a call to
75 * tlb_flush_mmu() is required.
77 * tlb_remove_page() and tlb_remove_page_size() imply the call to
78 * tlb_flush_mmu() when required and has no return value.
80 * - tlb_change_page_size()
82 * call before __tlb_remove_page*() to set the current page-size; implies a
83 * possible tlb_flush_mmu() call.
85 * - tlb_flush_mmu() / tlb_flush_mmu_tlbonly()
87 * tlb_flush_mmu_tlbonly() - does the TLB invalidate (and resets
88 * related state, like the range)
90 * tlb_flush_mmu() - in addition to the above TLB invalidate, also frees
91 * whatever pages are still batched.
93 * - mmu_gather::fullmm
95 * A flag set by tlb_gather_mmu() to indicate we're going to free
96 * the entire mm; this allows a number of optimizations.
98 * - We can ignore tlb_{start,end}_vma(); because we don't
99 * care about ranges. Everything will be shot down.
101 * - (RISC) architectures that use ASIDs can cycle to a new ASID
102 * and delay the invalidation until ASID space runs out.
104 * - mmu_gather::need_flush_all
106 * A flag that can be set by the arch code if it wants to force
107 * flush the entire TLB irrespective of the range. For instance
108 * x86-PAE needs this when changing top-level entries.
110 * And allows the architecture to provide and implement tlb_flush():
112 * tlb_flush() may, in addition to the above mentioned mmu_gather fields, make
115 * - mmu_gather::start / mmu_gather::end
117 * which provides the range that needs to be flushed to cover the pages to
120 * - mmu_gather::freed_tables
122 * set when we freed page table pages
124 * - tlb_get_unmap_shift() / tlb_get_unmap_size()
126 * returns the smallest TLB entry size unmapped in this range.
128 * If an architecture does not provide tlb_flush() a default implementation
129 * based on flush_tlb_range() will be used, unless MMU_GATHER_NO_RANGE is
130 * specified, in which case we'll default to flush_tlb_mm().
132 * Additionally there are a few opt-in features:
134 * MMU_GATHER_PAGE_SIZE
136 * This ensures we call tlb_flush() every time tlb_change_page_size() actually
137 * changes the size and provides mmu_gather::page_size to tlb_flush().
139 * This might be useful if your architecture has size specific TLB
140 * invalidation instructions.
142 * MMU_GATHER_TABLE_FREE
144 * This provides tlb_remove_table(), to be used instead of tlb_remove_page()
145 * for page directores (__p*_free_tlb()).
147 * Useful if your architecture has non-page page directories.
149 * When used, an architecture is expected to provide __tlb_remove_table()
150 * which does the actual freeing of these pages.
152 * MMU_GATHER_RCU_TABLE_FREE
154 * Like MMU_GATHER_TABLE_FREE, and adds semi-RCU semantics to the free (see
157 * Useful if your architecture doesn't use IPIs for remote TLB invalidates
158 * and therefore doesn't naturally serialize with software page-table walkers.
160 * MMU_GATHER_NO_RANGE
162 * Use this if your architecture lacks an efficient flush_tlb_range().
164 * MMU_GATHER_NO_GATHER
166 * If the option is set the mmu_gather will not track individual pages for
167 * delayed page free anymore. A platform that enables the option needs to
168 * provide its own implementation of the __tlb_remove_page_size() function to
171 * This is useful if your architecture already flushes TLB entries in the
172 * various ptep_get_and_clear() functions.
175 #ifdef CONFIG_MMU_GATHER_TABLE_FREE
177 struct mmu_table_batch {
178 #ifdef CONFIG_MMU_GATHER_RCU_TABLE_FREE
185 #define MAX_TABLE_BATCH \
186 ((PAGE_SIZE - sizeof(struct mmu_table_batch)) / sizeof(void *))
188 extern void tlb_remove_table(struct mmu_gather *tlb, void *table);
190 #else /* !CONFIG_MMU_GATHER_HAVE_TABLE_FREE */
193 * Without MMU_GATHER_TABLE_FREE the architecture is assumed to have page based
194 * page directories and we can use the normal page batching to free them.
196 #define tlb_remove_table(tlb, page) tlb_remove_page((tlb), (page))
198 #endif /* CONFIG_MMU_GATHER_TABLE_FREE */
200 #ifdef CONFIG_MMU_GATHER_RCU_TABLE_FREE
202 * This allows an architecture that does not use the linux page-tables for
203 * hardware to skip the TLBI when freeing page tables.
205 #ifndef tlb_needs_table_invalidate
206 #define tlb_needs_table_invalidate() (true)
211 #ifdef tlb_needs_table_invalidate
212 #error tlb_needs_table_invalidate() requires MMU_GATHER_RCU_TABLE_FREE
215 #endif /* CONFIG_MMU_GATHER_RCU_TABLE_FREE */
218 #ifndef CONFIG_MMU_GATHER_NO_GATHER
220 * If we can't allocate a page to make a big batch of page pointers
221 * to work on, then just handle a few from the on-stack structure.
223 #define MMU_GATHER_BUNDLE 8
225 struct mmu_gather_batch {
226 struct mmu_gather_batch *next;
229 struct page *pages[0];
232 #define MAX_GATHER_BATCH \
233 ((PAGE_SIZE - sizeof(struct mmu_gather_batch)) / sizeof(void *))
236 * Limit the maximum number of mmu_gather batches to reduce a risk of soft
237 * lockups for non-preemptible kernels on huge machines when a lot of memory
238 * is zapped during unmapping.
239 * 10K pages freed at once should be safe even without a preemption point.
241 #define MAX_GATHER_BATCH_COUNT (10000UL/MAX_GATHER_BATCH)
243 extern bool __tlb_remove_page_size(struct mmu_gather *tlb, struct page *page,
248 * struct mmu_gather is an opaque type used by the mm code for passing around
249 * any data needed by arch specific code for tlb_remove_page.
252 struct mm_struct *mm;
254 #ifdef CONFIG_MMU_GATHER_TABLE_FREE
255 struct mmu_table_batch *batch;
261 * we are in the middle of an operation to clear
262 * a full mm and can make some optimizations
264 unsigned int fullmm : 1;
267 * we have performed an operation which
268 * requires a complete flush of the tlb
270 unsigned int need_flush_all : 1;
273 * we have removed page directories
275 unsigned int freed_tables : 1;
278 * at which levels have we cleared entries?
280 unsigned int cleared_ptes : 1;
281 unsigned int cleared_pmds : 1;
282 unsigned int cleared_puds : 1;
283 unsigned int cleared_p4ds : 1;
286 * tracks VM_EXEC | VM_HUGETLB in tlb_start_vma
288 unsigned int vma_exec : 1;
289 unsigned int vma_huge : 1;
291 unsigned int batch_count;
293 #ifndef CONFIG_MMU_GATHER_NO_GATHER
294 struct mmu_gather_batch *active;
295 struct mmu_gather_batch local;
296 struct page *__pages[MMU_GATHER_BUNDLE];
298 #ifdef CONFIG_MMU_GATHER_PAGE_SIZE
299 unsigned int page_size;
304 void tlb_flush_mmu(struct mmu_gather *tlb);
306 static inline void __tlb_adjust_range(struct mmu_gather *tlb,
307 unsigned long address,
308 unsigned int range_size)
310 tlb->start = min(tlb->start, address);
311 tlb->end = max(tlb->end, address + range_size);
314 static inline void __tlb_reset_range(struct mmu_gather *tlb)
317 tlb->start = tlb->end = ~0;
319 tlb->start = TASK_SIZE;
322 tlb->freed_tables = 0;
323 tlb->cleared_ptes = 0;
324 tlb->cleared_pmds = 0;
325 tlb->cleared_puds = 0;
326 tlb->cleared_p4ds = 0;
328 * Do not reset mmu_gather::vma_* fields here, we do not
329 * call into tlb_start_vma() again to set them if there is an
330 * intermediate flush.
334 #ifdef CONFIG_MMU_GATHER_NO_RANGE
336 #if defined(tlb_flush) || defined(tlb_start_vma) || defined(tlb_end_vma)
337 #error MMU_GATHER_NO_RANGE relies on default tlb_flush(), tlb_start_vma() and tlb_end_vma()
341 * When an architecture does not have efficient means of range flushing TLBs
342 * there is no point in doing intermediate flushes on tlb_end_vma() to keep the
343 * range small. We equally don't have to worry about page granularity or other
346 * All we need to do is issue a full flush for any !0 range.
348 static inline void tlb_flush(struct mmu_gather *tlb)
351 flush_tlb_mm(tlb->mm);
355 tlb_update_vma_flags(struct mmu_gather *tlb, struct vm_area_struct *vma) { }
357 #define tlb_end_vma tlb_end_vma
358 static inline void tlb_end_vma(struct mmu_gather *tlb, struct vm_area_struct *vma) { }
360 #else /* CONFIG_MMU_GATHER_NO_RANGE */
364 #if defined(tlb_start_vma) || defined(tlb_end_vma)
365 #error Default tlb_flush() relies on default tlb_start_vma() and tlb_end_vma()
369 * When an architecture does not provide its own tlb_flush() implementation
370 * but does have a reasonably efficient flush_vma_range() implementation
373 static inline void tlb_flush(struct mmu_gather *tlb)
375 if (tlb->fullmm || tlb->need_flush_all) {
376 flush_tlb_mm(tlb->mm);
377 } else if (tlb->end) {
378 struct vm_area_struct vma = {
380 .vm_flags = (tlb->vma_exec ? VM_EXEC : 0) |
381 (tlb->vma_huge ? VM_HUGETLB : 0),
384 flush_tlb_range(&vma, tlb->start, tlb->end);
389 tlb_update_vma_flags(struct mmu_gather *tlb, struct vm_area_struct *vma)
392 * flush_tlb_range() implementations that look at VM_HUGETLB (tile,
393 * mips-4k) flush only large pages.
395 * flush_tlb_range() implementations that flush I-TLB also flush D-TLB
396 * (tile, xtensa, arm), so it's ok to just add VM_EXEC to an existing
399 * We rely on tlb_end_vma() to issue a flush, such that when we reset
400 * these values the batch is empty.
402 tlb->vma_huge = is_vm_hugetlb_page(vma);
403 tlb->vma_exec = !!(vma->vm_flags & VM_EXEC);
409 tlb_update_vma_flags(struct mmu_gather *tlb, struct vm_area_struct *vma) { }
413 #endif /* CONFIG_MMU_GATHER_NO_RANGE */
415 static inline void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb)
418 * Anything calling __tlb_adjust_range() also sets at least one of
421 if (!(tlb->freed_tables || tlb->cleared_ptes || tlb->cleared_pmds ||
422 tlb->cleared_puds || tlb->cleared_p4ds))
426 mmu_notifier_invalidate_range(tlb->mm, tlb->start, tlb->end);
427 __tlb_reset_range(tlb);
430 static inline void tlb_remove_page_size(struct mmu_gather *tlb,
431 struct page *page, int page_size)
433 if (__tlb_remove_page_size(tlb, page, page_size))
437 static inline bool __tlb_remove_page(struct mmu_gather *tlb, struct page *page)
439 return __tlb_remove_page_size(tlb, page, PAGE_SIZE);
443 * Similar to __tlb_remove_page but will call tlb_flush_mmu() itself when
446 static inline void tlb_remove_page(struct mmu_gather *tlb, struct page *page)
448 return tlb_remove_page_size(tlb, page, PAGE_SIZE);
451 static inline void tlb_change_page_size(struct mmu_gather *tlb,
452 unsigned int page_size)
454 #ifdef CONFIG_MMU_GATHER_PAGE_SIZE
455 if (tlb->page_size && tlb->page_size != page_size) {
456 if (!tlb->fullmm && !tlb->need_flush_all)
460 tlb->page_size = page_size;
464 static inline unsigned long tlb_get_unmap_shift(struct mmu_gather *tlb)
466 if (tlb->cleared_ptes)
468 if (tlb->cleared_pmds)
470 if (tlb->cleared_puds)
472 if (tlb->cleared_p4ds)
478 static inline unsigned long tlb_get_unmap_size(struct mmu_gather *tlb)
480 return 1UL << tlb_get_unmap_shift(tlb);
484 * In the case of tlb vma handling, we can optimise these away in the
485 * case where we're doing a full MM flush. When we're doing a munmap,
486 * the vmas are adjusted to only cover the region to be torn down.
488 #ifndef tlb_start_vma
489 static inline void tlb_start_vma(struct mmu_gather *tlb, struct vm_area_struct *vma)
494 tlb_update_vma_flags(tlb, vma);
495 flush_cache_range(vma, vma->vm_start, vma->vm_end);
500 static inline void tlb_end_vma(struct mmu_gather *tlb, struct vm_area_struct *vma)
506 * Do a TLB flush and reset the range at VMA boundaries; this avoids
507 * the ranges growing with the unused space between consecutive VMAs,
508 * but also the mmu_gather::vma_* flags from tlb_start_vma() rely on
511 tlb_flush_mmu_tlbonly(tlb);
515 #ifndef __tlb_remove_tlb_entry
516 #define __tlb_remove_tlb_entry(tlb, ptep, address) do { } while (0)
520 * tlb_remove_tlb_entry - remember a pte unmapping for later tlb invalidation.
522 * Record the fact that pte's were really unmapped by updating the range,
523 * so we can later optimise away the tlb invalidate. This helps when
524 * userspace is unmapping already-unmapped pages, which happens quite a lot.
526 #define tlb_remove_tlb_entry(tlb, ptep, address) \
528 __tlb_adjust_range(tlb, address, PAGE_SIZE); \
529 tlb->cleared_ptes = 1; \
530 __tlb_remove_tlb_entry(tlb, ptep, address); \
533 #define tlb_remove_huge_tlb_entry(h, tlb, ptep, address) \
535 unsigned long _sz = huge_page_size(h); \
536 __tlb_adjust_range(tlb, address, _sz); \
537 if (_sz == PMD_SIZE) \
538 tlb->cleared_pmds = 1; \
539 else if (_sz == PUD_SIZE) \
540 tlb->cleared_puds = 1; \
541 __tlb_remove_tlb_entry(tlb, ptep, address); \
545 * tlb_remove_pmd_tlb_entry - remember a pmd mapping for later tlb invalidation
546 * This is a nop so far, because only x86 needs it.
548 #ifndef __tlb_remove_pmd_tlb_entry
549 #define __tlb_remove_pmd_tlb_entry(tlb, pmdp, address) do {} while (0)
552 #define tlb_remove_pmd_tlb_entry(tlb, pmdp, address) \
554 __tlb_adjust_range(tlb, address, HPAGE_PMD_SIZE); \
555 tlb->cleared_pmds = 1; \
556 __tlb_remove_pmd_tlb_entry(tlb, pmdp, address); \
560 * tlb_remove_pud_tlb_entry - remember a pud mapping for later tlb
561 * invalidation. This is a nop so far, because only x86 needs it.
563 #ifndef __tlb_remove_pud_tlb_entry
564 #define __tlb_remove_pud_tlb_entry(tlb, pudp, address) do {} while (0)
567 #define tlb_remove_pud_tlb_entry(tlb, pudp, address) \
569 __tlb_adjust_range(tlb, address, HPAGE_PUD_SIZE); \
570 tlb->cleared_puds = 1; \
571 __tlb_remove_pud_tlb_entry(tlb, pudp, address); \
575 * For things like page tables caches (ie caching addresses "inside" the
576 * page tables, like x86 does), for legacy reasons, flushing an
577 * individual page had better flush the page table caches behind it. This
578 * is definitely how x86 works, for example. And if you have an
579 * architected non-legacy page table cache (which I'm not aware of
580 * anybody actually doing), you're going to have some architecturally
581 * explicit flushing for that, likely *separate* from a regular TLB entry
582 * flush, and thus you'd need more than just some range expansion..
584 * So if we ever find an architecture
585 * that would want something that odd, I think it is up to that
586 * architecture to do its own odd thing, not cause pain for others
587 * http://lkml.kernel.org/r/CA+55aFzBggoXtNXQeng5d_mRoDnaMBE5Y+URs+PHR67nUpMtaw@mail.gmail.com
589 * For now w.r.t page table cache, mark the range_size as PAGE_SIZE
593 #define pte_free_tlb(tlb, ptep, address) \
595 __tlb_adjust_range(tlb, address, PAGE_SIZE); \
596 tlb->freed_tables = 1; \
597 tlb->cleared_pmds = 1; \
598 __pte_free_tlb(tlb, ptep, address); \
603 #define pmd_free_tlb(tlb, pmdp, address) \
605 __tlb_adjust_range(tlb, address, PAGE_SIZE); \
606 tlb->freed_tables = 1; \
607 tlb->cleared_puds = 1; \
608 __pmd_free_tlb(tlb, pmdp, address); \
613 #define pud_free_tlb(tlb, pudp, address) \
615 __tlb_adjust_range(tlb, address, PAGE_SIZE); \
616 tlb->freed_tables = 1; \
617 tlb->cleared_p4ds = 1; \
618 __pud_free_tlb(tlb, pudp, address); \
623 #define p4d_free_tlb(tlb, pudp, address) \
625 __tlb_adjust_range(tlb, address, PAGE_SIZE); \
626 tlb->freed_tables = 1; \
627 __p4d_free_tlb(tlb, pudp, address); \
631 #endif /* CONFIG_MMU */
633 #endif /* _ASM_GENERIC__TLB_H */