xref: /linux/include/asm-generic/tlb.h (revision 9c5968db9e625019a0ee5226c7eebef5519d366a)
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_gather_mmu_fullmm() / tlb_finish_mmu()
50  *
51  *    start and finish a mmu_gather
52  *
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  *
61  *  - 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  *
70  *  - tlb_remove_page() / __tlb_remove_page()
71  *  - tlb_remove_page_size() / __tlb_remove_page_size()
72  *  - __tlb_remove_folio_pages()
73  *
74  *    __tlb_remove_page_size() is the basic primitive that queues a page for
75  *    freeing. __tlb_remove_page() assumes PAGE_SIZE. Both will return a
76  *    boolean indicating if the queue is (now) full and a call to
77  *    tlb_flush_mmu() is required.
78  *
79  *    tlb_remove_page() and tlb_remove_page_size() imply the call to
80  *    tlb_flush_mmu() when required and has no return value.
81  *
82  *    __tlb_remove_folio_pages() is similar to __tlb_remove_page(), however,
83  *    instead of removing a single page, remove the given number of consecutive
84  *    pages that are all part of the same (large) folio: just like calling
85  *    __tlb_remove_page() on each page individually.
86  *
87  *  - tlb_change_page_size()
88  *
89  *    call before __tlb_remove_page*() to set the current page-size; implies a
90  *    possible tlb_flush_mmu() call.
91  *
92  *  - tlb_flush_mmu() / tlb_flush_mmu_tlbonly()
93  *
94  *    tlb_flush_mmu_tlbonly() - does the TLB invalidate (and resets
95  *                              related state, like the range)
96  *
97  *    tlb_flush_mmu() - in addition to the above TLB invalidate, also frees
98  *			whatever pages are still batched.
99  *
100  *  - mmu_gather::fullmm
101  *
102  *    A flag set by tlb_gather_mmu_fullmm() to indicate we're going to free
103  *    the entire mm; this allows a number of optimizations.
104  *
105  *    - We can ignore tlb_{start,end}_vma(); because we don't
106  *      care about ranges. Everything will be shot down.
107  *
108  *    - (RISC) architectures that use ASIDs can cycle to a new ASID
109  *      and delay the invalidation until ASID space runs out.
110  *
111  *  - mmu_gather::need_flush_all
112  *
113  *    A flag that can be set by the arch code if it wants to force
114  *    flush the entire TLB irrespective of the range. For instance
115  *    x86-PAE needs this when changing top-level entries.
116  *
117  * And allows the architecture to provide and implement tlb_flush():
118  *
119  * tlb_flush() may, in addition to the above mentioned mmu_gather fields, make
120  * use of:
121  *
122  *  - mmu_gather::start / mmu_gather::end
123  *
124  *    which provides the range that needs to be flushed to cover the pages to
125  *    be freed.
126  *
127  *  - mmu_gather::freed_tables
128  *
129  *    set when we freed page table pages
130  *
131  *  - tlb_get_unmap_shift() / tlb_get_unmap_size()
132  *
133  *    returns the smallest TLB entry size unmapped in this range.
134  *
135  * If an architecture does not provide tlb_flush() a default implementation
136  * based on flush_tlb_range() will be used, unless MMU_GATHER_NO_RANGE is
137  * specified, in which case we'll default to flush_tlb_mm().
138  *
139  * Additionally there are a few opt-in features:
140  *
141  *  MMU_GATHER_PAGE_SIZE
142  *
143  *  This ensures we call tlb_flush() every time tlb_change_page_size() actually
144  *  changes the size and provides mmu_gather::page_size to tlb_flush().
145  *
146  *  This might be useful if your architecture has size specific TLB
147  *  invalidation instructions.
148  *
149  *  MMU_GATHER_TABLE_FREE
150  *
151  *  This provides tlb_remove_table(), to be used instead of tlb_remove_page()
152  *  for page directores (__p*_free_tlb()).
153  *
154  *  Useful if your architecture has non-page page directories.
155  *
156  *  When used, an architecture is expected to provide __tlb_remove_table() or
157  *  use the generic __tlb_remove_table(), which does the actual freeing of these
158  *  pages.
159  *
160  *  MMU_GATHER_RCU_TABLE_FREE
161  *
162  *  Like MMU_GATHER_TABLE_FREE, and adds semi-RCU semantics to the free (see
163  *  comment below).
164  *
165  *  Useful if your architecture doesn't use IPIs for remote TLB invalidates
166  *  and therefore doesn't naturally serialize with software page-table walkers.
167  *
168  *  MMU_GATHER_NO_FLUSH_CACHE
169  *
170  *  Indicates the architecture has flush_cache_range() but it needs *NOT* be called
171  *  before unmapping a VMA.
172  *
173  *  NOTE: strictly speaking we shouldn't have this knob and instead rely on
174  *	  flush_cache_range() being a NOP, except Sparc64 seems to be
175  *	  different here.
176  *
177  *  MMU_GATHER_MERGE_VMAS
178  *
179  *  Indicates the architecture wants to merge ranges over VMAs; typical when
180  *  multiple range invalidates are more expensive than a full invalidate.
181  *
182  *  MMU_GATHER_NO_RANGE
183  *
184  *  Use this if your architecture lacks an efficient flush_tlb_range(). This
185  *  option implies MMU_GATHER_MERGE_VMAS above.
186  *
187  *  MMU_GATHER_NO_GATHER
188  *
189  *  If the option is set the mmu_gather will not track individual pages for
190  *  delayed page free anymore. A platform that enables the option needs to
191  *  provide its own implementation of the __tlb_remove_page_size() function to
192  *  free pages.
193  *
194  *  This is useful if your architecture already flushes TLB entries in the
195  *  various ptep_get_and_clear() functions.
196  */
197 
198 #ifdef CONFIG_MMU_GATHER_TABLE_FREE
199 
200 struct mmu_table_batch {
201 #ifdef CONFIG_MMU_GATHER_RCU_TABLE_FREE
202 	struct rcu_head		rcu;
203 #endif
204 	unsigned int		nr;
205 	void			*tables[];
206 };
207 
208 #define MAX_TABLE_BATCH		\
209 	((PAGE_SIZE - sizeof(struct mmu_table_batch)) / sizeof(void *))
210 
211 #ifndef __HAVE_ARCH_TLB_REMOVE_TABLE
212 static inline void __tlb_remove_table(void *table)
213 {
214 	struct ptdesc *ptdesc = (struct ptdesc *)table;
215 
216 	pagetable_dtor_free(ptdesc);
217 }
218 #endif
219 
220 extern void tlb_remove_table(struct mmu_gather *tlb, void *table);
221 
222 #else /* !CONFIG_MMU_GATHER_TABLE_FREE */
223 
224 static inline void tlb_remove_page(struct mmu_gather *tlb, struct page *page);
225 /*
226  * Without MMU_GATHER_TABLE_FREE the architecture is assumed to have page based
227  * page directories and we can use the normal page batching to free them.
228  */
229 static inline void tlb_remove_table(struct mmu_gather *tlb, void *table)
230 {
231 	struct page *page = (struct page *)table;
232 
233 	pagetable_dtor(page_ptdesc(page));
234 	tlb_remove_page(tlb, page);
235 }
236 #endif /* CONFIG_MMU_GATHER_TABLE_FREE */
237 
238 #ifdef CONFIG_MMU_GATHER_RCU_TABLE_FREE
239 /*
240  * This allows an architecture that does not use the linux page-tables for
241  * hardware to skip the TLBI when freeing page tables.
242  */
243 #ifndef tlb_needs_table_invalidate
244 #define tlb_needs_table_invalidate() (true)
245 #endif
246 
247 void tlb_remove_table_sync_one(void);
248 
249 #else
250 
251 #ifdef tlb_needs_table_invalidate
252 #error tlb_needs_table_invalidate() requires MMU_GATHER_RCU_TABLE_FREE
253 #endif
254 
255 static inline void tlb_remove_table_sync_one(void) { }
256 
257 #endif /* CONFIG_MMU_GATHER_RCU_TABLE_FREE */
258 
259 
260 #ifndef CONFIG_MMU_GATHER_NO_GATHER
261 /*
262  * If we can't allocate a page to make a big batch of page pointers
263  * to work on, then just handle a few from the on-stack structure.
264  */
265 #define MMU_GATHER_BUNDLE	8
266 
267 struct mmu_gather_batch {
268 	struct mmu_gather_batch	*next;
269 	unsigned int		nr;
270 	unsigned int		max;
271 	struct encoded_page	*encoded_pages[];
272 };
273 
274 #define MAX_GATHER_BATCH	\
275 	((PAGE_SIZE - sizeof(struct mmu_gather_batch)) / sizeof(void *))
276 
277 /*
278  * Limit the maximum number of mmu_gather batches to reduce a risk of soft
279  * lockups for non-preemptible kernels on huge machines when a lot of memory
280  * is zapped during unmapping.
281  * 10K pages freed at once should be safe even without a preemption point.
282  */
283 #define MAX_GATHER_BATCH_COUNT	(10000UL/MAX_GATHER_BATCH)
284 
285 extern bool __tlb_remove_page_size(struct mmu_gather *tlb, struct page *page,
286 		bool delay_rmap, int page_size);
287 bool __tlb_remove_folio_pages(struct mmu_gather *tlb, struct page *page,
288 		unsigned int nr_pages, bool delay_rmap);
289 
290 #ifdef CONFIG_SMP
291 /*
292  * This both sets 'delayed_rmap', and returns true. It would be an inline
293  * function, except we define it before the 'struct mmu_gather'.
294  */
295 #define tlb_delay_rmap(tlb) (((tlb)->delayed_rmap = 1), true)
296 extern void tlb_flush_rmaps(struct mmu_gather *tlb, struct vm_area_struct *vma);
297 #endif
298 
299 #endif
300 
301 /*
302  * We have a no-op version of the rmap removal that doesn't
303  * delay anything. That is used on S390, which flushes remote
304  * TLBs synchronously, and on UP, which doesn't have any
305  * remote TLBs to flush and is not preemptible due to this
306  * all happening under the page table lock.
307  */
308 #ifndef tlb_delay_rmap
309 #define tlb_delay_rmap(tlb) (false)
310 static inline void tlb_flush_rmaps(struct mmu_gather *tlb, struct vm_area_struct *vma) { }
311 #endif
312 
313 /*
314  * struct mmu_gather is an opaque type used by the mm code for passing around
315  * any data needed by arch specific code for tlb_remove_page.
316  */
317 struct mmu_gather {
318 	struct mm_struct	*mm;
319 
320 #ifdef CONFIG_MMU_GATHER_TABLE_FREE
321 	struct mmu_table_batch	*batch;
322 #endif
323 
324 	unsigned long		start;
325 	unsigned long		end;
326 	/*
327 	 * we are in the middle of an operation to clear
328 	 * a full mm and can make some optimizations
329 	 */
330 	unsigned int		fullmm : 1;
331 
332 	/*
333 	 * we have performed an operation which
334 	 * requires a complete flush of the tlb
335 	 */
336 	unsigned int		need_flush_all : 1;
337 
338 	/*
339 	 * we have removed page directories
340 	 */
341 	unsigned int		freed_tables : 1;
342 
343 	/*
344 	 * Do we have pending delayed rmap removals?
345 	 */
346 	unsigned int		delayed_rmap : 1;
347 
348 	/*
349 	 * at which levels have we cleared entries?
350 	 */
351 	unsigned int		cleared_ptes : 1;
352 	unsigned int		cleared_pmds : 1;
353 	unsigned int		cleared_puds : 1;
354 	unsigned int		cleared_p4ds : 1;
355 
356 	/*
357 	 * tracks VM_EXEC | VM_HUGETLB in tlb_start_vma
358 	 */
359 	unsigned int		vma_exec : 1;
360 	unsigned int		vma_huge : 1;
361 	unsigned int		vma_pfn  : 1;
362 
363 	unsigned int		batch_count;
364 
365 #ifndef CONFIG_MMU_GATHER_NO_GATHER
366 	struct mmu_gather_batch *active;
367 	struct mmu_gather_batch	local;
368 	struct page		*__pages[MMU_GATHER_BUNDLE];
369 
370 #ifdef CONFIG_MMU_GATHER_PAGE_SIZE
371 	unsigned int page_size;
372 #endif
373 #endif
374 };
375 
376 void tlb_flush_mmu(struct mmu_gather *tlb);
377 
378 static inline void __tlb_adjust_range(struct mmu_gather *tlb,
379 				      unsigned long address,
380 				      unsigned int range_size)
381 {
382 	tlb->start = min(tlb->start, address);
383 	tlb->end = max(tlb->end, address + range_size);
384 }
385 
386 static inline void __tlb_reset_range(struct mmu_gather *tlb)
387 {
388 	if (tlb->fullmm) {
389 		tlb->start = tlb->end = ~0;
390 	} else {
391 		tlb->start = TASK_SIZE;
392 		tlb->end = 0;
393 	}
394 	tlb->freed_tables = 0;
395 	tlb->cleared_ptes = 0;
396 	tlb->cleared_pmds = 0;
397 	tlb->cleared_puds = 0;
398 	tlb->cleared_p4ds = 0;
399 	/*
400 	 * Do not reset mmu_gather::vma_* fields here, we do not
401 	 * call into tlb_start_vma() again to set them if there is an
402 	 * intermediate flush.
403 	 */
404 }
405 
406 #ifdef CONFIG_MMU_GATHER_NO_RANGE
407 
408 #if defined(tlb_flush)
409 #error MMU_GATHER_NO_RANGE relies on default tlb_flush()
410 #endif
411 
412 /*
413  * When an architecture does not have efficient means of range flushing TLBs
414  * there is no point in doing intermediate flushes on tlb_end_vma() to keep the
415  * range small. We equally don't have to worry about page granularity or other
416  * things.
417  *
418  * All we need to do is issue a full flush for any !0 range.
419  */
420 static inline void tlb_flush(struct mmu_gather *tlb)
421 {
422 	if (tlb->end)
423 		flush_tlb_mm(tlb->mm);
424 }
425 
426 #else /* CONFIG_MMU_GATHER_NO_RANGE */
427 
428 #ifndef tlb_flush
429 /*
430  * When an architecture does not provide its own tlb_flush() implementation
431  * but does have a reasonably efficient flush_vma_range() implementation
432  * use that.
433  */
434 static inline void tlb_flush(struct mmu_gather *tlb)
435 {
436 	if (tlb->fullmm || tlb->need_flush_all) {
437 		flush_tlb_mm(tlb->mm);
438 	} else if (tlb->end) {
439 		struct vm_area_struct vma = {
440 			.vm_mm = tlb->mm,
441 			.vm_flags = (tlb->vma_exec ? VM_EXEC    : 0) |
442 				    (tlb->vma_huge ? VM_HUGETLB : 0),
443 		};
444 
445 		flush_tlb_range(&vma, tlb->start, tlb->end);
446 	}
447 }
448 #endif
449 
450 #endif /* CONFIG_MMU_GATHER_NO_RANGE */
451 
452 static inline void
453 tlb_update_vma_flags(struct mmu_gather *tlb, struct vm_area_struct *vma)
454 {
455 	/*
456 	 * flush_tlb_range() implementations that look at VM_HUGETLB (tile,
457 	 * mips-4k) flush only large pages.
458 	 *
459 	 * flush_tlb_range() implementations that flush I-TLB also flush D-TLB
460 	 * (tile, xtensa, arm), so it's ok to just add VM_EXEC to an existing
461 	 * range.
462 	 *
463 	 * We rely on tlb_end_vma() to issue a flush, such that when we reset
464 	 * these values the batch is empty.
465 	 */
466 	tlb->vma_huge = is_vm_hugetlb_page(vma);
467 	tlb->vma_exec = !!(vma->vm_flags & VM_EXEC);
468 	tlb->vma_pfn  = !!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP));
469 }
470 
471 static inline void tlb_flush_mmu_tlbonly(struct mmu_gather *tlb)
472 {
473 	/*
474 	 * Anything calling __tlb_adjust_range() also sets at least one of
475 	 * these bits.
476 	 */
477 	if (!(tlb->freed_tables || tlb->cleared_ptes || tlb->cleared_pmds ||
478 	      tlb->cleared_puds || tlb->cleared_p4ds))
479 		return;
480 
481 	tlb_flush(tlb);
482 	__tlb_reset_range(tlb);
483 }
484 
485 static inline void tlb_remove_page_size(struct mmu_gather *tlb,
486 					struct page *page, int page_size)
487 {
488 	if (__tlb_remove_page_size(tlb, page, false, page_size))
489 		tlb_flush_mmu(tlb);
490 }
491 
492 static __always_inline bool __tlb_remove_page(struct mmu_gather *tlb,
493 		struct page *page, bool delay_rmap)
494 {
495 	return __tlb_remove_page_size(tlb, page, delay_rmap, PAGE_SIZE);
496 }
497 
498 /* tlb_remove_page
499  *	Similar to __tlb_remove_page but will call tlb_flush_mmu() itself when
500  *	required.
501  */
502 static inline void tlb_remove_page(struct mmu_gather *tlb, struct page *page)
503 {
504 	return tlb_remove_page_size(tlb, page, PAGE_SIZE);
505 }
506 
507 static inline void tlb_remove_ptdesc(struct mmu_gather *tlb, void *pt)
508 {
509 	tlb_remove_table(tlb, pt);
510 }
511 
512 /* Like tlb_remove_ptdesc, but for page-like page directories. */
513 static inline void tlb_remove_page_ptdesc(struct mmu_gather *tlb, struct ptdesc *pt)
514 {
515 	tlb_remove_page(tlb, ptdesc_page(pt));
516 }
517 
518 static inline void tlb_change_page_size(struct mmu_gather *tlb,
519 						     unsigned int page_size)
520 {
521 #ifdef CONFIG_MMU_GATHER_PAGE_SIZE
522 	if (tlb->page_size && tlb->page_size != page_size) {
523 		if (!tlb->fullmm && !tlb->need_flush_all)
524 			tlb_flush_mmu(tlb);
525 	}
526 
527 	tlb->page_size = page_size;
528 #endif
529 }
530 
531 static inline unsigned long tlb_get_unmap_shift(struct mmu_gather *tlb)
532 {
533 	if (tlb->cleared_ptes)
534 		return PAGE_SHIFT;
535 	if (tlb->cleared_pmds)
536 		return PMD_SHIFT;
537 	if (tlb->cleared_puds)
538 		return PUD_SHIFT;
539 	if (tlb->cleared_p4ds)
540 		return P4D_SHIFT;
541 
542 	return PAGE_SHIFT;
543 }
544 
545 static inline unsigned long tlb_get_unmap_size(struct mmu_gather *tlb)
546 {
547 	return 1UL << tlb_get_unmap_shift(tlb);
548 }
549 
550 /*
551  * In the case of tlb vma handling, we can optimise these away in the
552  * case where we're doing a full MM flush.  When we're doing a munmap,
553  * the vmas are adjusted to only cover the region to be torn down.
554  */
555 static inline void tlb_start_vma(struct mmu_gather *tlb, struct vm_area_struct *vma)
556 {
557 	if (tlb->fullmm)
558 		return;
559 
560 	tlb_update_vma_flags(tlb, vma);
561 #ifndef CONFIG_MMU_GATHER_NO_FLUSH_CACHE
562 	flush_cache_range(vma, vma->vm_start, vma->vm_end);
563 #endif
564 }
565 
566 static inline void tlb_end_vma(struct mmu_gather *tlb, struct vm_area_struct *vma)
567 {
568 	if (tlb->fullmm)
569 		return;
570 
571 	/*
572 	 * VM_PFNMAP is more fragile because the core mm will not track the
573 	 * page mapcount -- there might not be page-frames for these PFNs after
574 	 * all. Force flush TLBs for such ranges to avoid munmap() vs
575 	 * unmap_mapping_range() races.
576 	 */
577 	if (tlb->vma_pfn || !IS_ENABLED(CONFIG_MMU_GATHER_MERGE_VMAS)) {
578 		/*
579 		 * Do a TLB flush and reset the range at VMA boundaries; this avoids
580 		 * the ranges growing with the unused space between consecutive VMAs.
581 		 */
582 		tlb_flush_mmu_tlbonly(tlb);
583 	}
584 }
585 
586 /*
587  * tlb_flush_{pte|pmd|pud|p4d}_range() adjust the tlb->start and tlb->end,
588  * and set corresponding cleared_*.
589  */
590 static inline void tlb_flush_pte_range(struct mmu_gather *tlb,
591 				     unsigned long address, unsigned long size)
592 {
593 	__tlb_adjust_range(tlb, address, size);
594 	tlb->cleared_ptes = 1;
595 }
596 
597 static inline void tlb_flush_pmd_range(struct mmu_gather *tlb,
598 				     unsigned long address, unsigned long size)
599 {
600 	__tlb_adjust_range(tlb, address, size);
601 	tlb->cleared_pmds = 1;
602 }
603 
604 static inline void tlb_flush_pud_range(struct mmu_gather *tlb,
605 				     unsigned long address, unsigned long size)
606 {
607 	__tlb_adjust_range(tlb, address, size);
608 	tlb->cleared_puds = 1;
609 }
610 
611 static inline void tlb_flush_p4d_range(struct mmu_gather *tlb,
612 				     unsigned long address, unsigned long size)
613 {
614 	__tlb_adjust_range(tlb, address, size);
615 	tlb->cleared_p4ds = 1;
616 }
617 
618 #ifndef __tlb_remove_tlb_entry
619 static inline void __tlb_remove_tlb_entry(struct mmu_gather *tlb, pte_t *ptep, unsigned long address)
620 {
621 }
622 #endif
623 
624 /**
625  * tlb_remove_tlb_entry - remember a pte unmapping for later tlb invalidation.
626  *
627  * Record the fact that pte's were really unmapped by updating the range,
628  * so we can later optimise away the tlb invalidate.   This helps when
629  * userspace is unmapping already-unmapped pages, which happens quite a lot.
630  */
631 #define tlb_remove_tlb_entry(tlb, ptep, address)		\
632 	do {							\
633 		tlb_flush_pte_range(tlb, address, PAGE_SIZE);	\
634 		__tlb_remove_tlb_entry(tlb, ptep, address);	\
635 	} while (0)
636 
637 /**
638  * tlb_remove_tlb_entries - remember unmapping of multiple consecutive ptes for
639  *			    later tlb invalidation.
640  *
641  * Similar to tlb_remove_tlb_entry(), but remember unmapping of multiple
642  * consecutive ptes instead of only a single one.
643  */
644 static inline void tlb_remove_tlb_entries(struct mmu_gather *tlb,
645 		pte_t *ptep, unsigned int nr, unsigned long address)
646 {
647 	tlb_flush_pte_range(tlb, address, PAGE_SIZE * nr);
648 	for (;;) {
649 		__tlb_remove_tlb_entry(tlb, ptep, address);
650 		if (--nr == 0)
651 			break;
652 		ptep++;
653 		address += PAGE_SIZE;
654 	}
655 }
656 
657 #define tlb_remove_huge_tlb_entry(h, tlb, ptep, address)	\
658 	do {							\
659 		unsigned long _sz = huge_page_size(h);		\
660 		if (_sz >= P4D_SIZE)				\
661 			tlb_flush_p4d_range(tlb, address, _sz);	\
662 		else if (_sz >= PUD_SIZE)			\
663 			tlb_flush_pud_range(tlb, address, _sz);	\
664 		else if (_sz >= PMD_SIZE)			\
665 			tlb_flush_pmd_range(tlb, address, _sz);	\
666 		else						\
667 			tlb_flush_pte_range(tlb, address, _sz);	\
668 		__tlb_remove_tlb_entry(tlb, ptep, address);	\
669 	} while (0)
670 
671 /**
672  * tlb_remove_pmd_tlb_entry - remember a pmd mapping for later tlb invalidation
673  * This is a nop so far, because only x86 needs it.
674  */
675 #ifndef __tlb_remove_pmd_tlb_entry
676 #define __tlb_remove_pmd_tlb_entry(tlb, pmdp, address) do {} while (0)
677 #endif
678 
679 #define tlb_remove_pmd_tlb_entry(tlb, pmdp, address)			\
680 	do {								\
681 		tlb_flush_pmd_range(tlb, address, HPAGE_PMD_SIZE);	\
682 		__tlb_remove_pmd_tlb_entry(tlb, pmdp, address);		\
683 	} while (0)
684 
685 /**
686  * tlb_remove_pud_tlb_entry - remember a pud mapping for later tlb
687  * invalidation. This is a nop so far, because only x86 needs it.
688  */
689 #ifndef __tlb_remove_pud_tlb_entry
690 #define __tlb_remove_pud_tlb_entry(tlb, pudp, address) do {} while (0)
691 #endif
692 
693 #define tlb_remove_pud_tlb_entry(tlb, pudp, address)			\
694 	do {								\
695 		tlb_flush_pud_range(tlb, address, HPAGE_PUD_SIZE);	\
696 		__tlb_remove_pud_tlb_entry(tlb, pudp, address);		\
697 	} while (0)
698 
699 /*
700  * For things like page tables caches (ie caching addresses "inside" the
701  * page tables, like x86 does), for legacy reasons, flushing an
702  * individual page had better flush the page table caches behind it. This
703  * is definitely how x86 works, for example. And if you have an
704  * architected non-legacy page table cache (which I'm not aware of
705  * anybody actually doing), you're going to have some architecturally
706  * explicit flushing for that, likely *separate* from a regular TLB entry
707  * flush, and thus you'd need more than just some range expansion..
708  *
709  * So if we ever find an architecture
710  * that would want something that odd, I think it is up to that
711  * architecture to do its own odd thing, not cause pain for others
712  * http://lkml.kernel.org/r/CA+55aFzBggoXtNXQeng5d_mRoDnaMBE5Y+URs+PHR67nUpMtaw@mail.gmail.com
713  *
714  * For now w.r.t page table cache, mark the range_size as PAGE_SIZE
715  */
716 
717 #ifndef pte_free_tlb
718 #define pte_free_tlb(tlb, ptep, address)			\
719 	do {							\
720 		tlb_flush_pmd_range(tlb, address, PAGE_SIZE);	\
721 		tlb->freed_tables = 1;				\
722 		__pte_free_tlb(tlb, ptep, address);		\
723 	} while (0)
724 #endif
725 
726 #ifndef pmd_free_tlb
727 #define pmd_free_tlb(tlb, pmdp, address)			\
728 	do {							\
729 		tlb_flush_pud_range(tlb, address, PAGE_SIZE);	\
730 		tlb->freed_tables = 1;				\
731 		__pmd_free_tlb(tlb, pmdp, address);		\
732 	} while (0)
733 #endif
734 
735 #ifndef pud_free_tlb
736 #define pud_free_tlb(tlb, pudp, address)			\
737 	do {							\
738 		tlb_flush_p4d_range(tlb, address, PAGE_SIZE);	\
739 		tlb->freed_tables = 1;				\
740 		__pud_free_tlb(tlb, pudp, address);		\
741 	} while (0)
742 #endif
743 
744 #ifndef p4d_free_tlb
745 #define p4d_free_tlb(tlb, pudp, address)			\
746 	do {							\
747 		__tlb_adjust_range(tlb, address, PAGE_SIZE);	\
748 		tlb->freed_tables = 1;				\
749 		__p4d_free_tlb(tlb, pudp, address);		\
750 	} while (0)
751 #endif
752 
753 #ifndef pte_needs_flush
754 static inline bool pte_needs_flush(pte_t oldpte, pte_t newpte)
755 {
756 	return true;
757 }
758 #endif
759 
760 #ifndef huge_pmd_needs_flush
761 static inline bool huge_pmd_needs_flush(pmd_t oldpmd, pmd_t newpmd)
762 {
763 	return true;
764 }
765 #endif
766 
767 #endif /* CONFIG_MMU */
768 
769 #endif /* _ASM_GENERIC__TLB_H */
770