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/tlbflush.h>
18 #include <asm/cacheflush.h>
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
25 #ifndef nmi_uaccess_okay
26 # define nmi_uaccess_okay() true
32 * Generic MMU-gather implementation.
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.
37 * This correct ordering is:
40 * 2) TLB invalidate page
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.
47 * The mmu_gather API consists of:
49 * - tlb_gather_mmu() / tlb_gather_mmu_fullmm() / tlb_finish_mmu()
51 * start and finish a mmu_gather
53 * Finish in particular will issue a (final) TLB invalidate and free
54 * all (remaining) queued pages.
56 * - tlb_start_vma() / tlb_end_vma(); marks the start / end of a VMA
58 * Defaults to flushing at tlb_end_vma() to reset the range; helps when
59 * there's large holes between the VMAs.
61 * - tlb_remove_table()
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.
68 * See also MMU_GATHER_TABLE_FREE and MMU_GATHER_RCU_TABLE_FREE.
70 * - tlb_remove_page() / __tlb_remove_page()
71 * - tlb_remove_page_size() / __tlb_remove_page_size()
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.
78 * tlb_remove_page() and tlb_remove_page_size() imply the call to
79 * tlb_flush_mmu() when required and has no return value.
81 * - tlb_change_page_size()
83 * call before __tlb_remove_page*() to set the current page-size; implies a
84 * possible tlb_flush_mmu() call.
86 * - tlb_flush_mmu() / tlb_flush_mmu_tlbonly()
88 * tlb_flush_mmu_tlbonly() - does the TLB invalidate (and resets
89 * related state, like the range)
91 * tlb_flush_mmu() - in addition to the above TLB invalidate, also frees
92 * whatever pages are still batched.
94 * - mmu_gather::fullmm
96 * A flag set by tlb_gather_mmu_fullmm() to indicate we're going to free
97 * the entire mm; this allows a number of optimizations.
99 * - We can ignore tlb_{start,end}_vma(); because we don't
100 * care about ranges. Everything will be shot down.
102 * - (RISC) architectures that use ASIDs can cycle to a new ASID
103 * and delay the invalidation until ASID space runs out.
105 * - mmu_gather::need_flush_all
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.
111 * And allows the architecture to provide and implement tlb_flush():
113 * tlb_flush() may, in addition to the above mentioned mmu_gather fields, make
116 * - mmu_gather::start / mmu_gather::end
118 * which provides the range that needs to be flushed to cover the pages to
121 * - mmu_gather::freed_tables
123 * set when we freed page table pages
125 * - tlb_get_unmap_shift() / tlb_get_unmap_size()
127 * returns the smallest TLB entry size unmapped in this range.
129 * If an architecture does not provide tlb_flush() a default implementation
130 * 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().
133 * Additionally there are a few opt-in features:
135 * MMU_GATHER_PAGE_SIZE
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().
140 * This might be useful if your architecture has size specific TLB
141 * invalidation instructions.
143 * MMU_GATHER_TABLE_FREE
145 * This provides tlb_remove_table(), to be used instead of tlb_remove_page()
146 * for page directores (__p*_free_tlb()).
148 * Useful if your architecture has non-page page directories.
150 * When used, an architecture is expected to provide __tlb_remove_table()
151 * which does the actual freeing of these pages.
153 * MMU_GATHER_RCU_TABLE_FREE
155 * Like MMU_GATHER_TABLE_FREE, and adds semi-RCU semantics to the free (see
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.
161 * MMU_GATHER_NO_FLUSH_CACHE
163 * Indicates the architecture has flush_cache_range() but it needs *NOT* be called
164 * before unmapping a VMA.
166 * NOTE: strictly speaking we shouldn't have this knob and instead rely on
167 * flush_cache_range() being a NOP, except Sparc64 seems to be
170 * MMU_GATHER_MERGE_VMAS
172 * Indicates the architecture wants to merge ranges over VMAs; typical when
173 * multiple range invalidates are more expensive than a full invalidate.
175 * MMU_GATHER_NO_RANGE
177 * Use this if your architecture lacks an efficient flush_tlb_range(). This
178 * option implies MMU_GATHER_MERGE_VMAS above.
180 * MMU_GATHER_NO_GATHER
182 * If the option is set the mmu_gather will not track individual pages for
183 * delayed page free anymore. A platform that enables the option needs to
184 * provide its own implementation of the __tlb_remove_page_size() function to
187 * This is useful if your architecture already flushes TLB entries in the
188 * various ptep_get_and_clear() functions.
191 #ifdef CONFIG_MMU_GATHER_TABLE_FREE
193 struct mmu_table_batch {
194 #ifdef CONFIG_MMU_GATHER_RCU_TABLE_FREE
201 #define MAX_TABLE_BATCH \
202 ((PAGE_SIZE - sizeof(struct mmu_table_batch)) / sizeof(void *))
204 extern void tlb_remove_table(struct mmu_gather *tlb, void *table);
206 #else /* !CONFIG_MMU_GATHER_HAVE_TABLE_FREE */
209 * Without MMU_GATHER_TABLE_FREE the architecture is assumed to have page based
210 * page directories and we can use the normal page batching to free them.
212 #define tlb_remove_table(tlb, page) tlb_remove_page((tlb), (page))
214 #endif /* CONFIG_MMU_GATHER_TABLE_FREE */
216 #ifdef CONFIG_MMU_GATHER_RCU_TABLE_FREE
218 * This allows an architecture that does not use the linux page-tables for
219 * hardware to skip the TLBI when freeing page tables.
221 #ifndef tlb_needs_table_invalidate
222 #define tlb_needs_table_invalidate() (true)
227 #ifdef tlb_needs_table_invalidate
228 #error tlb_needs_table_invalidate() requires MMU_GATHER_RCU_TABLE_FREE
231 #endif /* CONFIG_MMU_GATHER_RCU_TABLE_FREE */
234 #ifndef CONFIG_MMU_GATHER_NO_GATHER
236 * If we can't allocate a page to make a big batch of page pointers
237 * to work on, then just handle a few from the on-stack structure.
239 #define MMU_GATHER_BUNDLE 8
241 struct mmu_gather_batch {
242 struct mmu_gather_batch *next;
245 struct page *pages[];
248 #define MAX_GATHER_BATCH \
249 ((PAGE_SIZE - sizeof(struct mmu_gather_batch)) / sizeof(void *))
252 * Limit the maximum number of mmu_gather batches to reduce a risk of soft
253 * lockups for non-preemptible kernels on huge machines when a lot of memory
254 * is zapped during unmapping.
255 * 10K pages freed at once should be safe even without a preemption point.
257 #define MAX_GATHER_BATCH_COUNT (10000UL/MAX_GATHER_BATCH)
259 extern bool __tlb_remove_page_size(struct mmu_gather *tlb, struct page *page,
264 * struct mmu_gather is an opaque type used by the mm code for passing around
265 * any data needed by arch specific code for tlb_remove_page.
268 struct mm_struct *mm;
270 #ifdef CONFIG_MMU_GATHER_TABLE_FREE
271 struct mmu_table_batch *batch;
277 * we are in the middle of an operation to clear
278 * a full mm and can make some optimizations
280 unsigned int fullmm : 1;
283 * we have performed an operation which
284 * requires a complete flush of the tlb
286 unsigned int need_flush_all : 1;
289 * we have removed page directories
291 unsigned int freed_tables : 1;
294 * at which levels have we cleared entries?
296 unsigned int cleared_ptes : 1;
297 unsigned int cleared_pmds : 1;
298 unsigned int cleared_puds : 1;
299 unsigned int cleared_p4ds : 1;
302 * tracks VM_EXEC | VM_HUGETLB in tlb_start_vma
304 unsigned int vma_exec : 1;
305 unsigned int vma_huge : 1;
306 unsigned int vma_pfn : 1;
308 unsigned int batch_count;
310 #ifndef CONFIG_MMU_GATHER_NO_GATHER
311 struct mmu_gather_batch *active;
312 struct mmu_gather_batch local;
313 struct page *__pages[MMU_GATHER_BUNDLE];
315 #ifdef CONFIG_MMU_GATHER_PAGE_SIZE
316 unsigned int page_size;
321 void tlb_flush_mmu(struct mmu_gather *tlb);
323 static inline void __tlb_adjust_range(struct mmu_gather *tlb,
324 unsigned long address,
325 unsigned int range_size)
327 tlb->start = min(tlb->start, address);
328 tlb->end = max(tlb->end, address + range_size);
331 static inline void __tlb_reset_range(struct mmu_gather *tlb)
334 tlb->start = tlb->end = ~0;
336 tlb->start = TASK_SIZE;
339 tlb->freed_tables = 0;
340 tlb->cleared_ptes = 0;
341 tlb->cleared_pmds = 0;
342 tlb->cleared_puds = 0;
343 tlb->cleared_p4ds = 0;
345 * Do not reset mmu_gather::vma_* fields here, we do not
346 * call into tlb_start_vma() again to set them if there is an
347 * intermediate flush.
351 #ifdef CONFIG_MMU_GATHER_NO_RANGE
353 #if defined(tlb_flush)
354 #error MMU_GATHER_NO_RANGE relies on default tlb_flush()
358 * When an architecture does not have efficient means of range flushing TLBs
359 * there is no point in doing intermediate flushes on tlb_end_vma() to keep the
360 * range small. We equally don't have to worry about page granularity or other
363 * All we need to do is issue a full flush for any !0 range.
365 static inline void tlb_flush(struct mmu_gather *tlb)
368 flush_tlb_mm(tlb->mm);
371 #else /* CONFIG_MMU_GATHER_NO_RANGE */
375 * When an architecture does not provide its own tlb_flush() implementation
376 * but does have a reasonably efficient flush_vma_range() implementation
379 static inline void tlb_flush(struct mmu_gather *tlb)
381 if (tlb->fullmm || tlb->need_flush_all) {
382 flush_tlb_mm(tlb->mm);
383 } else if (tlb->end) {
384 struct vm_area_struct vma = {
386 .vm_flags = (tlb->vma_exec ? VM_EXEC : 0) |
387 (tlb->vma_huge ? VM_HUGETLB : 0),
390 flush_tlb_range(&vma, tlb->start, tlb->end);
395 #endif /* CONFIG_MMU_GATHER_NO_RANGE */
398 tlb_update_vma_flags(struct mmu_gather *tlb, struct vm_area_struct *vma)
401 * flush_tlb_range() implementations that look at VM_HUGETLB (tile,
402 * mips-4k) flush only large pages.
404 * flush_tlb_range() implementations that flush I-TLB also flush D-TLB
405 * (tile, xtensa, arm), so it's ok to just add VM_EXEC to an existing
408 * We rely on tlb_end_vma() to issue a flush, such that when we reset
409 * these values the batch is empty.
411 tlb->vma_huge = is_vm_hugetlb_page(vma);
412 tlb->vma_exec = !!(vma->vm_flags & VM_EXEC);
413 tlb->vma_pfn = !!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP));
416 static inline void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb)
419 * Anything calling __tlb_adjust_range() also sets at least one of
422 if (!(tlb->freed_tables || tlb->cleared_ptes || tlb->cleared_pmds ||
423 tlb->cleared_puds || tlb->cleared_p4ds))
427 mmu_notifier_invalidate_range(tlb->mm, tlb->start, tlb->end);
428 __tlb_reset_range(tlb);
431 static inline void tlb_remove_page_size(struct mmu_gather *tlb,
432 struct page *page, int page_size)
434 if (__tlb_remove_page_size(tlb, page, page_size))
438 static inline bool __tlb_remove_page(struct mmu_gather *tlb, struct page *page)
440 return __tlb_remove_page_size(tlb, page, PAGE_SIZE);
444 * Similar to __tlb_remove_page but will call tlb_flush_mmu() itself when
447 static inline void tlb_remove_page(struct mmu_gather *tlb, struct page *page)
449 return tlb_remove_page_size(tlb, page, PAGE_SIZE);
452 static inline void tlb_change_page_size(struct mmu_gather *tlb,
453 unsigned int page_size)
455 #ifdef CONFIG_MMU_GATHER_PAGE_SIZE
456 if (tlb->page_size && tlb->page_size != page_size) {
457 if (!tlb->fullmm && !tlb->need_flush_all)
461 tlb->page_size = page_size;
465 static inline unsigned long tlb_get_unmap_shift(struct mmu_gather *tlb)
467 if (tlb->cleared_ptes)
469 if (tlb->cleared_pmds)
471 if (tlb->cleared_puds)
473 if (tlb->cleared_p4ds)
479 static inline unsigned long tlb_get_unmap_size(struct mmu_gather *tlb)
481 return 1UL << tlb_get_unmap_shift(tlb);
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.
489 static inline void tlb_start_vma(struct mmu_gather *tlb, struct vm_area_struct *vma)
494 tlb_update_vma_flags(tlb, vma);
495 #ifndef CONFIG_MMU_GATHER_NO_FLUSH_CACHE
496 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 * VM_PFNMAP is more fragile because the core mm will not track the
507 * page mapcount -- there might not be page-frames for these PFNs after
508 * all. Force flush TLBs for such ranges to avoid munmap() vs
509 * unmap_mapping_range() races.
511 if (tlb->vma_pfn || !IS_ENABLED(CONFIG_MMU_GATHER_MERGE_VMAS)) {
513 * Do a TLB flush and reset the range at VMA boundaries; this avoids
514 * the ranges growing with the unused space between consecutive VMAs.
516 tlb_flush_mmu_tlbonly(tlb);
521 * tlb_flush_{pte|pmd|pud|p4d}_range() adjust the tlb->start and tlb->end,
522 * and set corresponding cleared_*.
524 static inline void tlb_flush_pte_range(struct mmu_gather *tlb,
525 unsigned long address, unsigned long size)
527 __tlb_adjust_range(tlb, address, size);
528 tlb->cleared_ptes = 1;
531 static inline void tlb_flush_pmd_range(struct mmu_gather *tlb,
532 unsigned long address, unsigned long size)
534 __tlb_adjust_range(tlb, address, size);
535 tlb->cleared_pmds = 1;
538 static inline void tlb_flush_pud_range(struct mmu_gather *tlb,
539 unsigned long address, unsigned long size)
541 __tlb_adjust_range(tlb, address, size);
542 tlb->cleared_puds = 1;
545 static inline void tlb_flush_p4d_range(struct mmu_gather *tlb,
546 unsigned long address, unsigned long size)
548 __tlb_adjust_range(tlb, address, size);
549 tlb->cleared_p4ds = 1;
552 #ifndef __tlb_remove_tlb_entry
553 #define __tlb_remove_tlb_entry(tlb, ptep, address) do { } while (0)
557 * tlb_remove_tlb_entry - remember a pte unmapping for later tlb invalidation.
559 * Record the fact that pte's were really unmapped by updating the range,
560 * so we can later optimise away the tlb invalidate. This helps when
561 * userspace is unmapping already-unmapped pages, which happens quite a lot.
563 #define tlb_remove_tlb_entry(tlb, ptep, address) \
565 tlb_flush_pte_range(tlb, address, PAGE_SIZE); \
566 __tlb_remove_tlb_entry(tlb, ptep, address); \
569 #define tlb_remove_huge_tlb_entry(h, tlb, ptep, address) \
571 unsigned long _sz = huge_page_size(h); \
572 if (_sz >= P4D_SIZE) \
573 tlb_flush_p4d_range(tlb, address, _sz); \
574 else if (_sz >= PUD_SIZE) \
575 tlb_flush_pud_range(tlb, address, _sz); \
576 else if (_sz >= PMD_SIZE) \
577 tlb_flush_pmd_range(tlb, address, _sz); \
579 tlb_flush_pte_range(tlb, address, _sz); \
580 __tlb_remove_tlb_entry(tlb, ptep, address); \
584 * tlb_remove_pmd_tlb_entry - remember a pmd mapping for later tlb invalidation
585 * This is a nop so far, because only x86 needs it.
587 #ifndef __tlb_remove_pmd_tlb_entry
588 #define __tlb_remove_pmd_tlb_entry(tlb, pmdp, address) do {} while (0)
591 #define tlb_remove_pmd_tlb_entry(tlb, pmdp, address) \
593 tlb_flush_pmd_range(tlb, address, HPAGE_PMD_SIZE); \
594 __tlb_remove_pmd_tlb_entry(tlb, pmdp, address); \
598 * tlb_remove_pud_tlb_entry - remember a pud mapping for later tlb
599 * invalidation. This is a nop so far, because only x86 needs it.
601 #ifndef __tlb_remove_pud_tlb_entry
602 #define __tlb_remove_pud_tlb_entry(tlb, pudp, address) do {} while (0)
605 #define tlb_remove_pud_tlb_entry(tlb, pudp, address) \
607 tlb_flush_pud_range(tlb, address, HPAGE_PUD_SIZE); \
608 __tlb_remove_pud_tlb_entry(tlb, pudp, address); \
612 * For things like page tables caches (ie caching addresses "inside" the
613 * page tables, like x86 does), for legacy reasons, flushing an
614 * individual page had better flush the page table caches behind it. This
615 * is definitely how x86 works, for example. And if you have an
616 * architected non-legacy page table cache (which I'm not aware of
617 * anybody actually doing), you're going to have some architecturally
618 * explicit flushing for that, likely *separate* from a regular TLB entry
619 * flush, and thus you'd need more than just some range expansion..
621 * So if we ever find an architecture
622 * that would want something that odd, I think it is up to that
623 * architecture to do its own odd thing, not cause pain for others
624 * http://lkml.kernel.org/r/CA+55aFzBggoXtNXQeng5d_mRoDnaMBE5Y+URs+PHR67nUpMtaw@mail.gmail.com
626 * For now w.r.t page table cache, mark the range_size as PAGE_SIZE
630 #define pte_free_tlb(tlb, ptep, address) \
632 tlb_flush_pmd_range(tlb, address, PAGE_SIZE); \
633 tlb->freed_tables = 1; \
634 __pte_free_tlb(tlb, ptep, address); \
639 #define pmd_free_tlb(tlb, pmdp, address) \
641 tlb_flush_pud_range(tlb, address, PAGE_SIZE); \
642 tlb->freed_tables = 1; \
643 __pmd_free_tlb(tlb, pmdp, address); \
648 #define pud_free_tlb(tlb, pudp, address) \
650 tlb_flush_p4d_range(tlb, address, PAGE_SIZE); \
651 tlb->freed_tables = 1; \
652 __pud_free_tlb(tlb, pudp, address); \
657 #define p4d_free_tlb(tlb, pudp, address) \
659 __tlb_adjust_range(tlb, address, PAGE_SIZE); \
660 tlb->freed_tables = 1; \
661 __p4d_free_tlb(tlb, pudp, address); \
665 #ifndef pte_needs_flush
666 static inline bool pte_needs_flush(pte_t oldpte, pte_t newpte)
672 #ifndef huge_pmd_needs_flush
673 static inline bool huge_pmd_needs_flush(pmd_t oldpmd, pmd_t newpmd)
679 #endif /* CONFIG_MMU */
681 #endif /* _ASM_GENERIC__TLB_H */