highmem.h

  1/* SPDX-License-Identifier: GPL-2.0 */
  2#ifndef _LINUX_HIGHMEM_H
  3#define _LINUX_HIGHMEM_H
  4
  5#include <linux/fs.h>
  6#include <linux/kernel.h>
  7#include <linux/bug.h>
  8#include <linux/cacheflush.h>
  9#include <linux/kmsan.h>
 10#include <linux/mm.h>
 11#include <linux/uaccess.h>
 12#include <linux/hardirq.h>
 13
 14#include "highmem-internal.h"
 15
 16/**
 17 * kmap - Map a page for long term usage
 18 * @page:	Pointer to the page to be mapped
 19 *
 20 * Returns: The virtual address of the mapping
 21 *
 22 * Can only be invoked from preemptible task context because on 32bit
 23 * systems with CONFIG_HIGHMEM enabled this function might sleep.
 24 *
 25 * For systems with CONFIG_HIGHMEM=n and for pages in the low memory area
 26 * this returns the virtual address of the direct kernel mapping.
 27 *
 28 * The returned virtual address is globally visible and valid up to the
 29 * point where it is unmapped via kunmap(). The pointer can be handed to
 30 * other contexts.
 31 *
 32 * For highmem pages on 32bit systems this can be slow as the mapping space
 33 * is limited and protected by a global lock. In case that there is no
 34 * mapping slot available the function blocks until a slot is released via
 35 * kunmap().
 36 */
 37static inline void *kmap(struct page *page);
 38
 39/**
 40 * kunmap - Unmap the virtual address mapped by kmap()
 41 * @page:	Pointer to the page which was mapped by kmap()
 42 *
 43 * Counterpart to kmap(). A NOOP for CONFIG_HIGHMEM=n and for mappings of
 44 * pages in the low memory area.
 45 */
 46static inline void kunmap(const struct page *page);
 47
 48/**
 49 * kmap_to_page - Get the page for a kmap'ed address
 50 * @addr:	The address to look up
 51 *
 52 * Returns: The page which is mapped to @addr.
 53 */
 54static inline struct page *kmap_to_page(void *addr);
 55
 56/**
 57 * kmap_flush_unused - Flush all unused kmap mappings in order to
 58 *		       remove stray mappings
 59 */
 60static inline void kmap_flush_unused(void);
 61
 62/**
 63 * kmap_local_page - Map a page for temporary usage
 64 * @page: Pointer to the page to be mapped
 65 *
 66 * Returns: The virtual address of the mapping
 67 *
 68 * Can be invoked from any context, including interrupts.
 69 *
 70 * Requires careful handling when nesting multiple mappings because the map
 71 * management is stack based. The unmap has to be in the reverse order of
 72 * the map operation:
 73 *
 74 * addr1 = kmap_local_page(page1);
 75 * addr2 = kmap_local_page(page2);
 76 * ...
 77 * kunmap_local(addr2);
 78 * kunmap_local(addr1);
 79 *
 80 * Unmapping addr1 before addr2 is invalid and causes malfunction.
 81 *
 82 * Contrary to kmap() mappings the mapping is only valid in the context of
 83 * the caller and cannot be handed to other contexts.
 84 *
 85 * On CONFIG_HIGHMEM=n kernels and for low memory pages this returns the
 86 * virtual address of the direct mapping. Only real highmem pages are
 87 * temporarily mapped.
 88 *
 89 * While kmap_local_page() is significantly faster than kmap() for the highmem
 90 * case it comes with restrictions about the pointer validity.
 91 *
 92 * On HIGHMEM enabled systems mapping a highmem page has the side effect of
 93 * disabling migration in order to keep the virtual address stable across
 94 * preemption. No caller of kmap_local_page() can rely on this side effect.
 95 */
 96static inline void *kmap_local_page(const struct page *page);
 97
 98/**
 99 * kmap_local_folio - Map a page in this folio for temporary usage
100 * @folio: The folio containing the page.
101 * @offset: The byte offset within the folio which identifies the page.
102 *
103 * Requires careful handling when nesting multiple mappings because the map
104 * management is stack based. The unmap has to be in the reverse order of
105 * the map operation::
106 *
107 *   addr1 = kmap_local_folio(folio1, offset1);
108 *   addr2 = kmap_local_folio(folio2, offset2);
109 *   ...
110 *   kunmap_local(addr2);
111 *   kunmap_local(addr1);
112 *
113 * Unmapping addr1 before addr2 is invalid and causes malfunction.
114 *
115 * Contrary to kmap() mappings the mapping is only valid in the context of
116 * the caller and cannot be handed to other contexts.
117 *
118 * On CONFIG_HIGHMEM=n kernels and for low memory pages this returns the
119 * virtual address of the direct mapping. Only real highmem pages are
120 * temporarily mapped.
121 *
122 * While it is significantly faster than kmap() for the highmem case it
123 * comes with restrictions about the pointer validity.
124 *
125 * On HIGHMEM enabled systems mapping a highmem page has the side effect of
126 * disabling migration in order to keep the virtual address stable across
127 * preemption. No caller of kmap_local_folio() can rely on this side effect.
128 *
129 * Context: Can be invoked from any context.
130 * Return: The virtual address of @offset.
131 */
132static inline void *kmap_local_folio(const struct folio *folio, size_t offset);
133
134/**
135 * kmap_atomic - Atomically map a page for temporary usage - Deprecated!
136 * @page:	Pointer to the page to be mapped
137 *
138 * Returns: The virtual address of the mapping
139 *
140 * In fact a wrapper around kmap_local_page() which also disables pagefaults
141 * and, depending on PREEMPT_RT configuration, also CPU migration and
142 * preemption. Therefore users should not count on the latter two side effects.
143 *
144 * Mappings should always be released by kunmap_atomic().
145 *
146 * Do not use in new code. Use kmap_local_page() instead.
147 *
148 * It is used in atomic context when code wants to access the contents of a
149 * page that might be allocated from high memory (see __GFP_HIGHMEM), for
150 * example a page in the pagecache.  The API has two functions, and they
151 * can be used in a manner similar to the following::
152 *
153 *   // Find the page of interest.
154 *   struct page *page = find_get_page(mapping, offset);
155 *
156 *   // Gain access to the contents of that page.
157 *   void *vaddr = kmap_atomic(page);
158 *
159 *   // Do something to the contents of that page.
160 *   memset(vaddr, 0, PAGE_SIZE);
161 *
162 *   // Unmap that page.
163 *   kunmap_atomic(vaddr);
164 *
165 * Note that the kunmap_atomic() call takes the result of the kmap_atomic()
166 * call, not the argument.
167 *
168 * If you need to map two pages because you want to copy from one page to
169 * another you need to keep the kmap_atomic calls strictly nested, like:
170 *
171 * vaddr1 = kmap_atomic(page1);
172 * vaddr2 = kmap_atomic(page2);
173 *
174 * memcpy(vaddr1, vaddr2, PAGE_SIZE);
175 *
176 * kunmap_atomic(vaddr2);
177 * kunmap_atomic(vaddr1);
178 */
179static inline void *kmap_atomic(const struct page *page);
180
181/* Highmem related interfaces for management code */
182static inline unsigned long nr_free_highpages(void);
183static inline unsigned long totalhigh_pages(void);
184
185#ifndef ARCH_HAS_FLUSH_ANON_PAGE
186static inline void flush_anon_page(struct vm_area_struct *vma, struct page *page, unsigned long vmaddr)
187{
188}
189#endif
190
191#ifndef ARCH_IMPLEMENTS_FLUSH_KERNEL_VMAP_RANGE
192static inline void flush_kernel_vmap_range(void *vaddr, int size)
193{
194}
195static inline void invalidate_kernel_vmap_range(void *vaddr, int size)
196{
197}
198#endif
199
200/* when CONFIG_HIGHMEM is not set these will be plain clear/copy_page */
201#ifndef clear_user_highpage
202static inline void clear_user_highpage(struct page *page, unsigned long vaddr)
203{
204	void *addr = kmap_local_page(page);
205	clear_user_page(addr, vaddr, page);
206	kunmap_local(addr);
207}
208#endif
209
210#ifndef vma_alloc_zeroed_movable_folio
211/**
212 * vma_alloc_zeroed_movable_folio - Allocate a zeroed page for a VMA.
213 * @vma: The VMA the page is to be allocated for.
214 * @vaddr: The virtual address the page will be inserted into.
215 *
216 * This function will allocate a page suitable for inserting into this
217 * VMA at this virtual address.  It may be allocated from highmem or
218 * the movable zone.  An architecture may provide its own implementation.
219 *
220 * Return: A folio containing one allocated and zeroed page or NULL if
221 * we are out of memory.
222 */
223static inline
224struct folio *vma_alloc_zeroed_movable_folio(struct vm_area_struct *vma,
225				   unsigned long vaddr)
226{
227	struct folio *folio;
228
229	folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0, vma, vaddr);
230	if (folio && user_alloc_needs_zeroing())
231		clear_user_highpage(&folio->page, vaddr);
232
233	return folio;
234}
235#endif
236
237static inline void clear_highpage(struct page *page)
238{
239	void *kaddr = kmap_local_page(page);
240	clear_page(kaddr);
241	kunmap_local(kaddr);
242}
243
244static inline void clear_highpage_kasan_tagged(struct page *page)
245{
246	void *kaddr = kmap_local_page(page);
247
248	clear_page(kasan_reset_tag(kaddr));
249	kunmap_local(kaddr);
250}
251
252#ifndef __HAVE_ARCH_TAG_CLEAR_HIGHPAGES
253
254/* Return false to let people know we did not initialize the pages */
255static inline bool tag_clear_highpages(struct page *page, int numpages)
256{
257	return false;
258}
259
260#endif
261
262/*
263 * If we pass in a base or tail page, we can zero up to PAGE_SIZE.
264 * If we pass in a head page, we can zero up to the size of the compound page.
265 */
266#ifdef CONFIG_HIGHMEM
267void zero_user_segments(struct page *page, unsigned start1, unsigned end1,
268		unsigned start2, unsigned end2);
269#else
270static inline void zero_user_segments(struct page *page,
271		unsigned start1, unsigned end1,
272		unsigned start2, unsigned end2)
273{
274	void *kaddr = kmap_local_page(page);
275	unsigned int i;
276
277	BUG_ON(end1 > page_size(page) || end2 > page_size(page));
278
279	if (end1 > start1)
280		memset(kaddr + start1, 0, end1 - start1);
281
282	if (end2 > start2)
283		memset(kaddr + start2, 0, end2 - start2);
284
285	kunmap_local(kaddr);
286	for (i = 0; i < compound_nr(page); i++)
287		flush_dcache_page(page + i);
288}
289#endif
290
291static inline void zero_user_segment(struct page *page,
292	unsigned start, unsigned end)
293{
294	zero_user_segments(page, start, end, 0, 0);
295}
296
297#ifndef __HAVE_ARCH_COPY_USER_HIGHPAGE
298
299static inline void copy_user_highpage(struct page *to, struct page *from,
300	unsigned long vaddr, struct vm_area_struct *vma)
301{
302	char *vfrom, *vto;
303
304	vfrom = kmap_local_page(from);
305	vto = kmap_local_page(to);
306	copy_user_page(vto, vfrom, vaddr, to);
307	kmsan_unpoison_memory(page_address(to), PAGE_SIZE);
308	kunmap_local(vto);
309	kunmap_local(vfrom);
310}
311
312#endif
313
314#ifndef __HAVE_ARCH_COPY_HIGHPAGE
315
316static inline void copy_highpage(struct page *to, struct page *from)
317{
318	char *vfrom, *vto;
319
320	vfrom = kmap_local_page(from);
321	vto = kmap_local_page(to);
322	copy_page(vto, vfrom);
323	kmsan_copy_page_meta(to, from);
324	kunmap_local(vto);
325	kunmap_local(vfrom);
326}
327
328#endif
329
330#ifdef copy_mc_to_kernel
331/*
332 * If architecture supports machine check exception handling, define the
333 * #MC versions of copy_user_highpage and copy_highpage. They copy a memory
334 * page with #MC in source page (@from) handled, and return the number
335 * of bytes not copied if there was a #MC, otherwise 0 for success.
336 */
337static inline int copy_mc_user_highpage(struct page *to, struct page *from,
338					unsigned long vaddr, struct vm_area_struct *vma)
339{
340	unsigned long ret;
341	char *vfrom, *vto;
342
343	vfrom = kmap_local_page(from);
344	vto = kmap_local_page(to);
345	ret = copy_mc_to_kernel(vto, vfrom, PAGE_SIZE);
346	if (!ret)
347		kmsan_unpoison_memory(page_address(to), PAGE_SIZE);
348	kunmap_local(vto);
349	kunmap_local(vfrom);
350
351	if (ret)
352		memory_failure_queue(page_to_pfn(from), 0);
353
354	return ret;
355}
356
357static inline int copy_mc_highpage(struct page *to, struct page *from)
358{
359	unsigned long ret;
360	char *vfrom, *vto;
361
362	vfrom = kmap_local_page(from);
363	vto = kmap_local_page(to);
364	ret = copy_mc_to_kernel(vto, vfrom, PAGE_SIZE);
365	if (!ret)
366		kmsan_copy_page_meta(to, from);
367	kunmap_local(vto);
368	kunmap_local(vfrom);
369
370	if (ret)
371		memory_failure_queue(page_to_pfn(from), 0);
372
373	return ret;
374}
375#else
376static inline int copy_mc_user_highpage(struct page *to, struct page *from,
377					unsigned long vaddr, struct vm_area_struct *vma)
378{
379	copy_user_highpage(to, from, vaddr, vma);
380	return 0;
381}
382
383static inline int copy_mc_highpage(struct page *to, struct page *from)
384{
385	copy_highpage(to, from);
386	return 0;
387}
388#endif
389
390static inline void memcpy_page(struct page *dst_page, size_t dst_off,
391			       struct page *src_page, size_t src_off,
392			       size_t len)
393{
394	char *dst = kmap_local_page(dst_page);
395	char *src = kmap_local_page(src_page);
396
397	VM_BUG_ON(dst_off + len > PAGE_SIZE || src_off + len > PAGE_SIZE);
398	memcpy(dst + dst_off, src + src_off, len);
399	kunmap_local(src);
400	kunmap_local(dst);
401}
402
403static inline void memcpy_folio(struct folio *dst_folio, size_t dst_off,
404		struct folio *src_folio, size_t src_off, size_t len)
405{
406	VM_BUG_ON(dst_off + len > folio_size(dst_folio));
407	VM_BUG_ON(src_off + len > folio_size(src_folio));
408
409	do {
410		char *dst = kmap_local_folio(dst_folio, dst_off);
411		const char *src = kmap_local_folio(src_folio, src_off);
412		size_t chunk = len;
413
414		if (folio_test_highmem(dst_folio) &&
415		    chunk > PAGE_SIZE - offset_in_page(dst_off))
416			chunk = PAGE_SIZE - offset_in_page(dst_off);
417		if (folio_test_highmem(src_folio) &&
418		    chunk > PAGE_SIZE - offset_in_page(src_off))
419			chunk = PAGE_SIZE - offset_in_page(src_off);
420		memcpy(dst, src, chunk);
421		kunmap_local(src);
422		kunmap_local(dst);
423
424		dst_off += chunk;
425		src_off += chunk;
426		len -= chunk;
427	} while (len > 0);
428}
429
430static inline void memset_page(struct page *page, size_t offset, int val,
431			       size_t len)
432{
433	char *addr = kmap_local_page(page);
434
435	VM_BUG_ON(offset + len > PAGE_SIZE);
436	memset(addr + offset, val, len);
437	kunmap_local(addr);
438}
439
440static inline void memcpy_from_page(char *to, struct page *page,
441				    size_t offset, size_t len)
442{
443	char *from = kmap_local_page(page);
444
445	VM_BUG_ON(offset + len > PAGE_SIZE);
446	memcpy(to, from + offset, len);
447	kunmap_local(from);
448}
449
450static inline void memcpy_to_page(struct page *page, size_t offset,
451				  const char *from, size_t len)
452{
453	char *to = kmap_local_page(page);
454
455	VM_BUG_ON(offset + len > PAGE_SIZE);
456	memcpy(to + offset, from, len);
457	flush_dcache_page(page);
458	kunmap_local(to);
459}
460
461static inline void memzero_page(struct page *page, size_t offset, size_t len)
462{
463	char *addr = kmap_local_page(page);
464
465	VM_BUG_ON(offset + len > PAGE_SIZE);
466	memset(addr + offset, 0, len);
467	flush_dcache_page(page);
468	kunmap_local(addr);
469}
470
471/**
472 * memcpy_from_folio - Copy a range of bytes from a folio.
473 * @to: The memory to copy to.
474 * @folio: The folio to read from.
475 * @offset: The first byte in the folio to read.
476 * @len: The number of bytes to copy.
477 */
478static inline void memcpy_from_folio(char *to, struct folio *folio,
479		size_t offset, size_t len)
480{
481	VM_BUG_ON(offset + len > folio_size(folio));
482
483	do {
484		const char *from = kmap_local_folio(folio, offset);
485		size_t chunk = len;
486
487		if (folio_test_partial_kmap(folio) &&
488		    chunk > PAGE_SIZE - offset_in_page(offset))
489			chunk = PAGE_SIZE - offset_in_page(offset);
490		memcpy(to, from, chunk);
491		kunmap_local(from);
492
493		to += chunk;
494		offset += chunk;
495		len -= chunk;
496	} while (len > 0);
497}
498
499/**
500 * memcpy_to_folio - Copy a range of bytes to a folio.
501 * @folio: The folio to write to.
502 * @offset: The first byte in the folio to store to.
503 * @from: The memory to copy from.
504 * @len: The number of bytes to copy.
505 */
506static inline void memcpy_to_folio(struct folio *folio, size_t offset,
507		const char *from, size_t len)
508{
509	VM_BUG_ON(offset + len > folio_size(folio));
510
511	do {
512		char *to = kmap_local_folio(folio, offset);
513		size_t chunk = len;
514
515		if (folio_test_partial_kmap(folio) &&
516		    chunk > PAGE_SIZE - offset_in_page(offset))
517			chunk = PAGE_SIZE - offset_in_page(offset);
518		memcpy(to, from, chunk);
519		kunmap_local(to);
520
521		from += chunk;
522		offset += chunk;
523		len -= chunk;
524	} while (len > 0);
525
526	flush_dcache_folio(folio);
527}
528
529/**
530 * folio_zero_tail - Zero the tail of a folio.
531 * @folio: The folio to zero.
532 * @offset: The byte offset in the folio to start zeroing at.
533 * @kaddr: The address the folio is currently mapped to.
534 *
535 * If you have already used kmap_local_folio() to map a folio, written
536 * some data to it and now need to zero the end of the folio (and flush
537 * the dcache), you can use this function.  If you do not have the
538 * folio kmapped (eg the folio has been partially populated by DMA),
539 * use folio_zero_range() or folio_zero_segment() instead.
540 *
541 * Return: An address which can be passed to kunmap_local().
542 */
543static inline __must_check void *folio_zero_tail(struct folio *folio,
544		size_t offset, void *kaddr)
545{
546	size_t len = folio_size(folio) - offset;
547
548	if (folio_test_partial_kmap(folio)) {
549		size_t max = PAGE_SIZE - offset_in_page(offset);
550
551		while (len > max) {
552			memset(kaddr, 0, max);
553			kunmap_local(kaddr);
554			len -= max;
555			offset += max;
556			max = PAGE_SIZE;
557			kaddr = kmap_local_folio(folio, offset);
558		}
559	}
560
561	memset(kaddr, 0, len);
562	flush_dcache_folio(folio);
563
564	return kaddr;
565}
566
567/**
568 * folio_fill_tail - Copy some data to a folio and pad with zeroes.
569 * @folio: The destination folio.
570 * @offset: The offset into @folio at which to start copying.
571 * @from: The data to copy.
572 * @len: How many bytes of data to copy.
573 *
574 * This function is most useful for filesystems which support inline data.
575 * When they want to copy data from the inode into the page cache, this
576 * function does everything for them.  It supports large folios even on
577 * HIGHMEM configurations.
578 */
579static inline void folio_fill_tail(struct folio *folio, size_t offset,
580		const char *from, size_t len)
581{
582	char *to = kmap_local_folio(folio, offset);
583
584	VM_BUG_ON(offset + len > folio_size(folio));
585
586	if (folio_test_partial_kmap(folio)) {
587		size_t max = PAGE_SIZE - offset_in_page(offset);
588
589		while (len > max) {
590			memcpy(to, from, max);
591			kunmap_local(to);
592			len -= max;
593			from += max;
594			offset += max;
595			max = PAGE_SIZE;
596			to = kmap_local_folio(folio, offset);
597		}
598	}
599
600	memcpy(to, from, len);
601	to = folio_zero_tail(folio, offset + len, to + len);
602	kunmap_local(to);
603}
604
605/**
606 * memcpy_from_file_folio - Copy some bytes from a file folio.
607 * @to: The destination buffer.
608 * @folio: The folio to copy from.
609 * @pos: The position in the file.
610 * @len: The maximum number of bytes to copy.
611 *
612 * Copy up to @len bytes from this folio.  This may be limited by PAGE_SIZE
613 * if the folio comes from HIGHMEM, and by the size of the folio.
614 *
615 * Return: The number of bytes copied from the folio.
616 */
617static inline size_t memcpy_from_file_folio(char *to, struct folio *folio,
618		loff_t pos, size_t len)
619{
620	size_t offset = offset_in_folio(folio, pos);
621	char *from = kmap_local_folio(folio, offset);
622
623	if (folio_test_partial_kmap(folio)) {
624		offset = offset_in_page(offset);
625		len = min_t(size_t, len, PAGE_SIZE - offset);
626	} else
627		len = min(len, folio_size(folio) - offset);
628
629	memcpy(to, from, len);
630	kunmap_local(from);
631
632	return len;
633}
634
635/**
636 * folio_zero_segments() - Zero two byte ranges in a folio.
637 * @folio: The folio to write to.
638 * @start1: The first byte to zero.
639 * @xend1: One more than the last byte in the first range.
640 * @start2: The first byte to zero in the second range.
641 * @xend2: One more than the last byte in the second range.
642 */
643static inline void folio_zero_segments(struct folio *folio,
644		size_t start1, size_t xend1, size_t start2, size_t xend2)
645{
646	zero_user_segments(&folio->page, start1, xend1, start2, xend2);
647}
648
649/**
650 * folio_zero_segment() - Zero a byte range in a folio.
651 * @folio: The folio to write to.
652 * @start: The first byte to zero.
653 * @xend: One more than the last byte to zero.
654 */
655static inline void folio_zero_segment(struct folio *folio,
656		size_t start, size_t xend)
657{
658	zero_user_segments(&folio->page, start, xend, 0, 0);
659}
660
661/**
662 * folio_zero_range() - Zero a byte range in a folio.
663 * @folio: The folio to write to.
664 * @start: The first byte to zero.
665 * @length: The number of bytes to zero.
666 */
667static inline void folio_zero_range(struct folio *folio,
668		size_t start, size_t length)
669{
670	zero_user_segments(&folio->page, start, start + length, 0, 0);
671}
672
673/**
674 * folio_release_kmap - Unmap a folio and drop a refcount.
675 * @folio: The folio to release.
676 * @addr: The address previously returned by a call to kmap_local_folio().
677 *
678 * It is common, eg in directory handling to kmap a folio.  This function
679 * unmaps the folio and drops the refcount that was being held to keep the
680 * folio alive while we accessed it.
681 */
682static inline void folio_release_kmap(struct folio *folio, void *addr)
683{
684	kunmap_local(addr);
685	folio_put(folio);
686}
687#endif /* _LINUX_HIGHMEM_H */